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 inline void sock_recv_drops(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
673 if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
674 put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
675 sizeof(__u32), &skb->dropcount);
678 void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
681 sock_recv_timestamp(msg, sk, skb);
682 sock_recv_drops(msg, sk, skb);
684 EXPORT_SYMBOL_GPL(sock_recv_ts_and_drops);
686 static inline int __sock_recvmsg_nosec(struct kiocb *iocb, struct socket *sock,
687 struct msghdr *msg, size_t size, int flags)
689 struct sock_iocb *si = kiocb_to_siocb(iocb);
697 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
700 static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
701 struct msghdr *msg, size_t size, int flags)
703 int err = security_socket_recvmsg(sock, msg, size, flags);
705 return err ?: __sock_recvmsg_nosec(iocb, sock, msg, size, flags);
708 int sock_recvmsg(struct socket *sock, struct msghdr *msg,
709 size_t size, int flags)
712 struct sock_iocb siocb;
715 init_sync_kiocb(&iocb, NULL);
716 iocb.private = &siocb;
717 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
718 if (-EIOCBQUEUED == ret)
719 ret = wait_on_sync_kiocb(&iocb);
723 static int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
724 size_t size, int flags)
727 struct sock_iocb siocb;
730 init_sync_kiocb(&iocb, NULL);
731 iocb.private = &siocb;
732 ret = __sock_recvmsg_nosec(&iocb, sock, msg, size, flags);
733 if (-EIOCBQUEUED == ret)
734 ret = wait_on_sync_kiocb(&iocb);
738 int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
739 struct kvec *vec, size_t num, size_t size, int flags)
741 mm_segment_t oldfs = get_fs();
746 * the following is safe, since for compiler definitions of kvec and
747 * iovec are identical, yielding the same in-core layout and alignment
749 msg->msg_iov = (struct iovec *)vec, msg->msg_iovlen = num;
750 result = sock_recvmsg(sock, msg, size, flags);
755 static void sock_aio_dtor(struct kiocb *iocb)
757 kfree(iocb->private);
760 static ssize_t sock_sendpage(struct file *file, struct page *page,
761 int offset, size_t size, loff_t *ppos, int more)
766 sock = file->private_data;
768 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
772 return kernel_sendpage(sock, page, offset, size, flags);
775 static ssize_t sock_splice_read(struct file *file, loff_t *ppos,
776 struct pipe_inode_info *pipe, size_t len,
779 struct socket *sock = file->private_data;
781 if (unlikely(!sock->ops->splice_read))
784 return sock->ops->splice_read(sock, ppos, pipe, len, flags);
787 static struct sock_iocb *alloc_sock_iocb(struct kiocb *iocb,
788 struct sock_iocb *siocb)
790 if (!is_sync_kiocb(iocb)) {
791 siocb = kmalloc(sizeof(*siocb), GFP_KERNEL);
794 iocb->ki_dtor = sock_aio_dtor;
798 iocb->private = siocb;
802 static ssize_t do_sock_read(struct msghdr *msg, struct kiocb *iocb,
803 struct file *file, const struct iovec *iov,
804 unsigned long nr_segs)
806 struct socket *sock = file->private_data;
810 for (i = 0; i < nr_segs; i++)
811 size += iov[i].iov_len;
813 msg->msg_name = NULL;
814 msg->msg_namelen = 0;
815 msg->msg_control = NULL;
816 msg->msg_controllen = 0;
817 msg->msg_iov = (struct iovec *)iov;
818 msg->msg_iovlen = nr_segs;
819 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
821 return __sock_recvmsg(iocb, sock, msg, size, msg->msg_flags);
824 static ssize_t sock_aio_read(struct kiocb *iocb, const struct iovec *iov,
825 unsigned long nr_segs, loff_t pos)
827 struct sock_iocb siocb, *x;
832 if (iocb->ki_left == 0) /* Match SYS5 behaviour */
836 x = alloc_sock_iocb(iocb, &siocb);
839 return do_sock_read(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
842 static ssize_t do_sock_write(struct msghdr *msg, struct kiocb *iocb,
843 struct file *file, const struct iovec *iov,
844 unsigned long nr_segs)
846 struct socket *sock = file->private_data;
850 for (i = 0; i < nr_segs; i++)
851 size += iov[i].iov_len;
853 msg->msg_name = NULL;
854 msg->msg_namelen = 0;
855 msg->msg_control = NULL;
856 msg->msg_controllen = 0;
857 msg->msg_iov = (struct iovec *)iov;
858 msg->msg_iovlen = nr_segs;
859 msg->msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
860 if (sock->type == SOCK_SEQPACKET)
861 msg->msg_flags |= MSG_EOR;
863 return __sock_sendmsg(iocb, sock, msg, size);
866 static ssize_t sock_aio_write(struct kiocb *iocb, const struct iovec *iov,
867 unsigned long nr_segs, loff_t pos)
869 struct sock_iocb siocb, *x;
874 x = alloc_sock_iocb(iocb, &siocb);
878 return do_sock_write(&x->async_msg, iocb, iocb->ki_filp, iov, nr_segs);
882 * Atomic setting of ioctl hooks to avoid race
883 * with module unload.
886 static DEFINE_MUTEX(br_ioctl_mutex);
887 static int (*br_ioctl_hook) (struct net *, unsigned int cmd, void __user *arg) = NULL;
889 void brioctl_set(int (*hook) (struct net *, unsigned int, void __user *))
891 mutex_lock(&br_ioctl_mutex);
892 br_ioctl_hook = hook;
893 mutex_unlock(&br_ioctl_mutex);
896 EXPORT_SYMBOL(brioctl_set);
898 static DEFINE_MUTEX(vlan_ioctl_mutex);
899 static int (*vlan_ioctl_hook) (struct net *, void __user *arg);
901 void vlan_ioctl_set(int (*hook) (struct net *, void __user *))
903 mutex_lock(&vlan_ioctl_mutex);
904 vlan_ioctl_hook = hook;
905 mutex_unlock(&vlan_ioctl_mutex);
908 EXPORT_SYMBOL(vlan_ioctl_set);
910 static DEFINE_MUTEX(dlci_ioctl_mutex);
911 static int (*dlci_ioctl_hook) (unsigned int, void __user *);
913 void dlci_ioctl_set(int (*hook) (unsigned int, void __user *))
915 mutex_lock(&dlci_ioctl_mutex);
916 dlci_ioctl_hook = hook;
917 mutex_unlock(&dlci_ioctl_mutex);
920 EXPORT_SYMBOL(dlci_ioctl_set);
923 * With an ioctl, arg may well be a user mode pointer, but we don't know
924 * what to do with it - that's up to the protocol still.
927 static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
931 void __user *argp = (void __user *)arg;
935 sock = file->private_data;
938 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
939 err = dev_ioctl(net, cmd, argp);
941 #ifdef CONFIG_WEXT_CORE
942 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
943 err = dev_ioctl(net, cmd, argp);
950 if (get_user(pid, (int __user *)argp))
952 err = f_setown(sock->file, pid, 1);
956 err = put_user(f_getown(sock->file),
965 request_module("bridge");
967 mutex_lock(&br_ioctl_mutex);
969 err = br_ioctl_hook(net, cmd, argp);
970 mutex_unlock(&br_ioctl_mutex);
975 if (!vlan_ioctl_hook)
976 request_module("8021q");
978 mutex_lock(&vlan_ioctl_mutex);
980 err = vlan_ioctl_hook(net, argp);
981 mutex_unlock(&vlan_ioctl_mutex);
986 if (!dlci_ioctl_hook)
987 request_module("dlci");
989 mutex_lock(&dlci_ioctl_mutex);
991 err = dlci_ioctl_hook(cmd, argp);
992 mutex_unlock(&dlci_ioctl_mutex);
995 err = sock->ops->ioctl(sock, cmd, arg);
998 * If this ioctl is unknown try to hand it down
1001 if (err == -ENOIOCTLCMD)
1002 err = dev_ioctl(net, cmd, argp);
1008 int sock_create_lite(int family, int type, int protocol, struct socket **res)
1011 struct socket *sock = NULL;
1013 err = security_socket_create(family, type, protocol, 1);
1017 sock = sock_alloc();
1024 err = security_socket_post_create(sock, family, type, protocol, 1);
1037 /* No kernel lock held - perfect */
1038 static unsigned int sock_poll(struct file *file, poll_table *wait)
1040 struct socket *sock;
1043 * We can't return errors to poll, so it's either yes or no.
1045 sock = file->private_data;
1046 return sock->ops->poll(file, sock, wait);
1049 static int sock_mmap(struct file *file, struct vm_area_struct *vma)
1051 struct socket *sock = file->private_data;
1053 return sock->ops->mmap(file, sock, vma);
1056 static int sock_close(struct inode *inode, struct file *filp)
1059 * It was possible the inode is NULL we were
1060 * closing an unfinished socket.
1064 printk(KERN_DEBUG "sock_close: NULL inode\n");
1067 sock_release(SOCKET_I(inode));
1072 * Update the socket async list
1074 * Fasync_list locking strategy.
1076 * 1. fasync_list is modified only under process context socket lock
1077 * i.e. under semaphore.
1078 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
1079 * or under socket lock.
1080 * 3. fasync_list can be used from softirq context, so that
1081 * modification under socket lock have to be enhanced with
1082 * write_lock_bh(&sk->sk_callback_lock).
1086 static int sock_fasync(int fd, struct file *filp, int on)
1088 struct fasync_struct *fa, *fna = NULL, **prev;
1089 struct socket *sock;
1093 fna = kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
1098 sock = filp->private_data;
1108 spin_lock(&filp->f_lock);
1110 filp->f_flags |= FASYNC;
1112 filp->f_flags &= ~FASYNC;
1113 spin_unlock(&filp->f_lock);
1115 prev = &(sock->fasync_list);
1117 for (fa = *prev; fa != NULL; prev = &fa->fa_next, fa = *prev)
1118 if (fa->fa_file == filp)
1123 write_lock_bh(&sk->sk_callback_lock);
1125 write_unlock_bh(&sk->sk_callback_lock);
1130 fna->fa_file = filp;
1132 fna->magic = FASYNC_MAGIC;
1133 fna->fa_next = sock->fasync_list;
1134 write_lock_bh(&sk->sk_callback_lock);
1135 sock->fasync_list = fna;
1136 sock_set_flag(sk, SOCK_FASYNC);
1137 write_unlock_bh(&sk->sk_callback_lock);
1140 write_lock_bh(&sk->sk_callback_lock);
1141 *prev = fa->fa_next;
1142 if (!sock->fasync_list)
1143 sock_reset_flag(sk, SOCK_FASYNC);
1144 write_unlock_bh(&sk->sk_callback_lock);
1150 release_sock(sock->sk);
1154 /* This function may be called only under socket lock or callback_lock */
1156 int sock_wake_async(struct socket *sock, int how, int band)
1158 if (!sock || !sock->fasync_list)
1161 case SOCK_WAKE_WAITD:
1162 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1165 case SOCK_WAKE_SPACE:
1166 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1171 __kill_fasync(sock->fasync_list, SIGIO, band);
1174 __kill_fasync(sock->fasync_list, SIGURG, band);
1179 static int __sock_create(struct net *net, int family, int type, int protocol,
1180 struct socket **res, int kern)
1183 struct socket *sock;
1184 const struct net_proto_family *pf;
1187 * Check protocol is in range
1189 if (family < 0 || family >= NPROTO)
1190 return -EAFNOSUPPORT;
1191 if (type < 0 || type >= SOCK_MAX)
1196 This uglymoron is moved from INET layer to here to avoid
1197 deadlock in module load.
1199 if (family == PF_INET && type == SOCK_PACKET) {
1203 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n",
1209 err = security_socket_create(family, type, protocol, kern);
1214 * Allocate the socket and allow the family to set things up. if
1215 * the protocol is 0, the family is instructed to select an appropriate
1218 sock = sock_alloc();
1220 if (net_ratelimit())
1221 printk(KERN_WARNING "socket: no more sockets\n");
1222 return -ENFILE; /* Not exactly a match, but its the
1223 closest posix thing */
1228 #ifdef CONFIG_MODULES
1229 /* Attempt to load a protocol module if the find failed.
1231 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1232 * requested real, full-featured networking support upon configuration.
1233 * Otherwise module support will break!
1235 if (net_families[family] == NULL)
1236 request_module("net-pf-%d", family);
1240 pf = rcu_dereference(net_families[family]);
1241 err = -EAFNOSUPPORT;
1246 * We will call the ->create function, that possibly is in a loadable
1247 * module, so we have to bump that loadable module refcnt first.
1249 if (!try_module_get(pf->owner))
1252 /* Now protected by module ref count */
1255 err = pf->create(net, sock, protocol);
1257 goto out_module_put;
1260 * Now to bump the refcnt of the [loadable] module that owns this
1261 * socket at sock_release time we decrement its refcnt.
1263 if (!try_module_get(sock->ops->owner))
1264 goto out_module_busy;
1267 * Now that we're done with the ->create function, the [loadable]
1268 * module can have its refcnt decremented
1270 module_put(pf->owner);
1271 err = security_socket_post_create(sock, family, type, protocol, kern);
1273 goto out_sock_release;
1279 err = -EAFNOSUPPORT;
1282 module_put(pf->owner);
1289 goto out_sock_release;
1292 int sock_create(int family, int type, int protocol, struct socket **res)
1294 return __sock_create(current->nsproxy->net_ns, family, type, protocol, res, 0);
1297 int sock_create_kern(int family, int type, int protocol, struct socket **res)
1299 return __sock_create(&init_net, family, type, protocol, res, 1);
1302 SYSCALL_DEFINE3(socket, int, family, int, type, int, protocol)
1305 struct socket *sock;
1308 /* Check the SOCK_* constants for consistency. */
1309 BUILD_BUG_ON(SOCK_CLOEXEC != O_CLOEXEC);
1310 BUILD_BUG_ON((SOCK_MAX | SOCK_TYPE_MASK) != SOCK_TYPE_MASK);
1311 BUILD_BUG_ON(SOCK_CLOEXEC & SOCK_TYPE_MASK);
1312 BUILD_BUG_ON(SOCK_NONBLOCK & SOCK_TYPE_MASK);
1314 flags = type & ~SOCK_TYPE_MASK;
1315 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1317 type &= SOCK_TYPE_MASK;
1319 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1320 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1322 retval = sock_create(family, type, protocol, &sock);
1326 retval = sock_map_fd(sock, flags & (O_CLOEXEC | O_NONBLOCK));
1331 /* It may be already another descriptor 8) Not kernel problem. */
1340 * Create a pair of connected sockets.
1343 SYSCALL_DEFINE4(socketpair, int, family, int, type, int, protocol,
1344 int __user *, usockvec)
1346 struct socket *sock1, *sock2;
1348 struct file *newfile1, *newfile2;
1351 flags = type & ~SOCK_TYPE_MASK;
1352 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1354 type &= SOCK_TYPE_MASK;
1356 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1357 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1360 * Obtain the first socket and check if the underlying protocol
1361 * supports the socketpair call.
1364 err = sock_create(family, type, protocol, &sock1);
1368 err = sock_create(family, type, protocol, &sock2);
1372 err = sock1->ops->socketpair(sock1, sock2);
1374 goto out_release_both;
1376 fd1 = sock_alloc_fd(&newfile1, flags & O_CLOEXEC);
1377 if (unlikely(fd1 < 0)) {
1379 goto out_release_both;
1382 fd2 = sock_alloc_fd(&newfile2, flags & O_CLOEXEC);
1383 if (unlikely(fd2 < 0)) {
1387 goto out_release_both;
1390 err = sock_attach_fd(sock1, newfile1, flags & O_NONBLOCK);
1391 if (unlikely(err < 0)) {
1395 err = sock_attach_fd(sock2, newfile2, flags & O_NONBLOCK);
1396 if (unlikely(err < 0)) {
1401 audit_fd_pair(fd1, fd2);
1402 fd_install(fd1, newfile1);
1403 fd_install(fd2, newfile2);
1404 /* fd1 and fd2 may be already another descriptors.
1405 * Not kernel problem.
1408 err = put_user(fd1, &usockvec[0]);
1410 err = put_user(fd2, &usockvec[1]);
1419 sock_release(sock2);
1421 sock_release(sock1);
1427 sock_release(sock1);
1430 sock_release(sock2);
1437 * Bind a name to a socket. Nothing much to do here since it's
1438 * the protocol's responsibility to handle the local address.
1440 * We move the socket address to kernel space before we call
1441 * the protocol layer (having also checked the address is ok).
1444 SYSCALL_DEFINE3(bind, int, fd, struct sockaddr __user *, umyaddr, int, addrlen)
1446 struct socket *sock;
1447 struct sockaddr_storage address;
1448 int err, fput_needed;
1450 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1452 err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);
1454 err = security_socket_bind(sock,
1455 (struct sockaddr *)&address,
1458 err = sock->ops->bind(sock,
1462 fput_light(sock->file, fput_needed);
1468 * Perform a listen. Basically, we allow the protocol to do anything
1469 * necessary for a listen, and if that works, we mark the socket as
1470 * ready for listening.
1473 SYSCALL_DEFINE2(listen, int, fd, int, backlog)
1475 struct socket *sock;
1476 int err, fput_needed;
1479 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1481 somaxconn = sock_net(sock->sk)->core.sysctl_somaxconn;
1482 if ((unsigned)backlog > somaxconn)
1483 backlog = somaxconn;
1485 err = security_socket_listen(sock, backlog);
1487 err = sock->ops->listen(sock, backlog);
1489 fput_light(sock->file, fput_needed);
1495 * For accept, we attempt to create a new socket, set up the link
1496 * with the client, wake up the client, then return the new
1497 * connected fd. We collect the address of the connector in kernel
1498 * space and move it to user at the very end. This is unclean because
1499 * we open the socket then return an error.
1501 * 1003.1g adds the ability to recvmsg() to query connection pending
1502 * status to recvmsg. We need to add that support in a way thats
1503 * clean when we restucture accept also.
1506 SYSCALL_DEFINE4(accept4, int, fd, struct sockaddr __user *, upeer_sockaddr,
1507 int __user *, upeer_addrlen, int, flags)
1509 struct socket *sock, *newsock;
1510 struct file *newfile;
1511 int err, len, newfd, fput_needed;
1512 struct sockaddr_storage address;
1514 if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1517 if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
1518 flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1520 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1525 if (!(newsock = sock_alloc()))
1528 newsock->type = sock->type;
1529 newsock->ops = sock->ops;
1532 * We don't need try_module_get here, as the listening socket (sock)
1533 * has the protocol module (sock->ops->owner) held.
1535 __module_get(newsock->ops->owner);
1537 newfd = sock_alloc_fd(&newfile, flags & O_CLOEXEC);
1538 if (unlikely(newfd < 0)) {
1540 sock_release(newsock);
1544 err = sock_attach_fd(newsock, newfile, flags & O_NONBLOCK);
1548 err = security_socket_accept(sock, newsock);
1552 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1556 if (upeer_sockaddr) {
1557 if (newsock->ops->getname(newsock, (struct sockaddr *)&address,
1559 err = -ECONNABORTED;
1562 err = move_addr_to_user((struct sockaddr *)&address,
1563 len, upeer_sockaddr, upeer_addrlen);
1568 /* File flags are not inherited via accept() unlike another OSes. */
1570 fd_install(newfd, newfile);
1574 fput_light(sock->file, fput_needed);
1578 sock_release(newsock);
1580 put_unused_fd(newfd);
1584 put_unused_fd(newfd);
1588 SYSCALL_DEFINE3(accept, int, fd, struct sockaddr __user *, upeer_sockaddr,
1589 int __user *, upeer_addrlen)
1591 return sys_accept4(fd, upeer_sockaddr, upeer_addrlen, 0);
1595 * Attempt to connect to a socket with the server address. The address
1596 * is in user space so we verify it is OK and move it to kernel space.
1598 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1601 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1602 * other SEQPACKET protocols that take time to connect() as it doesn't
1603 * include the -EINPROGRESS status for such sockets.
1606 SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
1609 struct socket *sock;
1610 struct sockaddr_storage address;
1611 int err, fput_needed;
1613 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1616 err = move_addr_to_kernel(uservaddr, addrlen, (struct sockaddr *)&address);
1621 security_socket_connect(sock, (struct sockaddr *)&address, addrlen);
1625 err = sock->ops->connect(sock, (struct sockaddr *)&address, addrlen,
1626 sock->file->f_flags);
1628 fput_light(sock->file, fput_needed);
1634 * Get the local address ('name') of a socket object. Move the obtained
1635 * name to user space.
1638 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
1639 int __user *, usockaddr_len)
1641 struct socket *sock;
1642 struct sockaddr_storage address;
1643 int len, err, fput_needed;
1645 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1649 err = security_socket_getsockname(sock);
1653 err = sock->ops->getname(sock, (struct sockaddr *)&address, &len, 0);
1656 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr, usockaddr_len);
1659 fput_light(sock->file, fput_needed);
1665 * Get the remote address ('name') of a socket object. Move the obtained
1666 * name to user space.
1669 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
1670 int __user *, usockaddr_len)
1672 struct socket *sock;
1673 struct sockaddr_storage address;
1674 int len, err, fput_needed;
1676 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1678 err = security_socket_getpeername(sock);
1680 fput_light(sock->file, fput_needed);
1685 sock->ops->getname(sock, (struct sockaddr *)&address, &len,
1688 err = move_addr_to_user((struct sockaddr *)&address, len, usockaddr,
1690 fput_light(sock->file, fput_needed);
1696 * Send a datagram to a given address. We move the address into kernel
1697 * space and check the user space data area is readable before invoking
1701 SYSCALL_DEFINE6(sendto, int, fd, void __user *, buff, size_t, len,
1702 unsigned, flags, struct sockaddr __user *, addr,
1705 struct socket *sock;
1706 struct sockaddr_storage address;
1712 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1716 iov.iov_base = buff;
1718 msg.msg_name = NULL;
1721 msg.msg_control = NULL;
1722 msg.msg_controllen = 0;
1723 msg.msg_namelen = 0;
1725 err = move_addr_to_kernel(addr, addr_len, (struct sockaddr *)&address);
1728 msg.msg_name = (struct sockaddr *)&address;
1729 msg.msg_namelen = addr_len;
1731 if (sock->file->f_flags & O_NONBLOCK)
1732 flags |= MSG_DONTWAIT;
1733 msg.msg_flags = flags;
1734 err = sock_sendmsg(sock, &msg, len);
1737 fput_light(sock->file, fput_needed);
1743 * Send a datagram down a socket.
1746 SYSCALL_DEFINE4(send, int, fd, void __user *, buff, size_t, len,
1749 return sys_sendto(fd, buff, len, flags, NULL, 0);
1753 * Receive a frame from the socket and optionally record the address of the
1754 * sender. We verify the buffers are writable and if needed move the
1755 * sender address from kernel to user space.
1758 SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
1759 unsigned, flags, struct sockaddr __user *, addr,
1760 int __user *, addr_len)
1762 struct socket *sock;
1765 struct sockaddr_storage address;
1769 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1773 msg.msg_control = NULL;
1774 msg.msg_controllen = 0;
1778 iov.iov_base = ubuf;
1779 msg.msg_name = (struct sockaddr *)&address;
1780 msg.msg_namelen = sizeof(address);
1781 if (sock->file->f_flags & O_NONBLOCK)
1782 flags |= MSG_DONTWAIT;
1783 err = sock_recvmsg(sock, &msg, size, flags);
1785 if (err >= 0 && addr != NULL) {
1786 err2 = move_addr_to_user((struct sockaddr *)&address,
1787 msg.msg_namelen, addr, addr_len);
1792 fput_light(sock->file, fput_needed);
1798 * Receive a datagram from a socket.
1801 asmlinkage long sys_recv(int fd, void __user *ubuf, size_t size,
1804 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1808 * Set a socket option. Because we don't know the option lengths we have
1809 * to pass the user mode parameter for the protocols to sort out.
1812 SYSCALL_DEFINE5(setsockopt, int, fd, int, level, int, optname,
1813 char __user *, optval, int, optlen)
1815 int err, fput_needed;
1816 struct socket *sock;
1821 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1823 err = security_socket_setsockopt(sock, level, optname);
1827 if (level == SOL_SOCKET)
1829 sock_setsockopt(sock, level, optname, optval,
1833 sock->ops->setsockopt(sock, level, optname, optval,
1836 fput_light(sock->file, fput_needed);
1842 * Get a socket option. Because we don't know the option lengths we have
1843 * to pass a user mode parameter for the protocols to sort out.
1846 SYSCALL_DEFINE5(getsockopt, int, fd, int, level, int, optname,
1847 char __user *, optval, int __user *, optlen)
1849 int err, fput_needed;
1850 struct socket *sock;
1852 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1854 err = security_socket_getsockopt(sock, level, optname);
1858 if (level == SOL_SOCKET)
1860 sock_getsockopt(sock, level, optname, optval,
1864 sock->ops->getsockopt(sock, level, optname, optval,
1867 fput_light(sock->file, fput_needed);
1873 * Shutdown a socket.
1876 SYSCALL_DEFINE2(shutdown, int, fd, int, how)
1878 int err, fput_needed;
1879 struct socket *sock;
1881 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1883 err = security_socket_shutdown(sock, how);
1885 err = sock->ops->shutdown(sock, how);
1886 fput_light(sock->file, fput_needed);
1891 /* A couple of helpful macros for getting the address of the 32/64 bit
1892 * fields which are the same type (int / unsigned) on our platforms.
1894 #define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1895 #define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1896 #define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1899 * BSD sendmsg interface
1902 SYSCALL_DEFINE3(sendmsg, int, fd, struct msghdr __user *, msg, unsigned, flags)
1904 struct compat_msghdr __user *msg_compat =
1905 (struct compat_msghdr __user *)msg;
1906 struct socket *sock;
1907 struct sockaddr_storage address;
1908 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1909 unsigned char ctl[sizeof(struct cmsghdr) + 20]
1910 __attribute__ ((aligned(sizeof(__kernel_size_t))));
1911 /* 20 is size of ipv6_pktinfo */
1912 unsigned char *ctl_buf = ctl;
1913 struct msghdr msg_sys;
1914 int err, ctl_len, iov_size, total_len;
1918 if (MSG_CMSG_COMPAT & flags) {
1919 if (get_compat_msghdr(&msg_sys, msg_compat))
1922 else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1925 sock = sockfd_lookup_light(fd, &err, &fput_needed);
1929 /* do not move before msg_sys is valid */
1931 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1934 /* Check whether to allocate the iovec area */
1936 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1937 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1938 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1943 /* This will also move the address data into kernel space */
1944 if (MSG_CMSG_COMPAT & flags) {
1945 err = verify_compat_iovec(&msg_sys, iov,
1946 (struct sockaddr *)&address,
1949 err = verify_iovec(&msg_sys, iov,
1950 (struct sockaddr *)&address,
1958 if (msg_sys.msg_controllen > INT_MAX)
1960 ctl_len = msg_sys.msg_controllen;
1961 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1963 cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl,
1967 ctl_buf = msg_sys.msg_control;
1968 ctl_len = msg_sys.msg_controllen;
1969 } else if (ctl_len) {
1970 if (ctl_len > sizeof(ctl)) {
1971 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1972 if (ctl_buf == NULL)
1977 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1978 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1979 * checking falls down on this.
1981 if (copy_from_user(ctl_buf, (void __user *)msg_sys.msg_control,
1984 msg_sys.msg_control = ctl_buf;
1986 msg_sys.msg_flags = flags;
1988 if (sock->file->f_flags & O_NONBLOCK)
1989 msg_sys.msg_flags |= MSG_DONTWAIT;
1990 err = sock_sendmsg(sock, &msg_sys, total_len);
1994 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1996 if (iov != iovstack)
1997 sock_kfree_s(sock->sk, iov, iov_size);
1999 fput_light(sock->file, fput_needed);
2004 static int __sys_recvmsg(struct socket *sock, struct msghdr __user *msg,
2005 struct msghdr *msg_sys, unsigned flags, int nosec)
2007 struct compat_msghdr __user *msg_compat =
2008 (struct compat_msghdr __user *)msg;
2009 struct iovec iovstack[UIO_FASTIOV];
2010 struct iovec *iov = iovstack;
2011 unsigned long cmsg_ptr;
2012 int err, iov_size, total_len, len;
2014 /* kernel mode address */
2015 struct sockaddr_storage addr;
2017 /* user mode address pointers */
2018 struct sockaddr __user *uaddr;
2019 int __user *uaddr_len;
2021 if (MSG_CMSG_COMPAT & flags) {
2022 if (get_compat_msghdr(msg_sys, msg_compat))
2025 else if (copy_from_user(msg_sys, msg, sizeof(struct msghdr)))
2029 if (msg_sys->msg_iovlen > UIO_MAXIOV)
2032 /* Check whether to allocate the iovec area */
2034 iov_size = msg_sys->msg_iovlen * sizeof(struct iovec);
2035 if (msg_sys->msg_iovlen > UIO_FASTIOV) {
2036 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
2042 * Save the user-mode address (verify_iovec will change the
2043 * kernel msghdr to use the kernel address space)
2046 uaddr = (__force void __user *)msg_sys->msg_name;
2047 uaddr_len = COMPAT_NAMELEN(msg);
2048 if (MSG_CMSG_COMPAT & flags) {
2049 err = verify_compat_iovec(msg_sys, iov,
2050 (struct sockaddr *)&addr,
2053 err = verify_iovec(msg_sys, iov,
2054 (struct sockaddr *)&addr,
2060 cmsg_ptr = (unsigned long)msg_sys->msg_control;
2061 msg_sys->msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
2063 if (sock->file->f_flags & O_NONBLOCK)
2064 flags |= MSG_DONTWAIT;
2065 err = (nosec ? sock_recvmsg_nosec : sock_recvmsg)(sock, msg_sys,
2071 if (uaddr != NULL) {
2072 err = move_addr_to_user((struct sockaddr *)&addr,
2073 msg_sys->msg_namelen, uaddr,
2078 err = __put_user((msg_sys->msg_flags & ~MSG_CMSG_COMPAT),
2082 if (MSG_CMSG_COMPAT & flags)
2083 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2084 &msg_compat->msg_controllen);
2086 err = __put_user((unsigned long)msg_sys->msg_control - cmsg_ptr,
2087 &msg->msg_controllen);
2093 if (iov != iovstack)
2094 sock_kfree_s(sock->sk, iov, iov_size);
2100 * BSD recvmsg interface
2103 SYSCALL_DEFINE3(recvmsg, int, fd, struct msghdr __user *, msg,
2104 unsigned int, flags)
2106 int fput_needed, err;
2107 struct msghdr msg_sys;
2108 struct socket *sock = sockfd_lookup_light(fd, &err, &fput_needed);
2113 err = __sys_recvmsg(sock, msg, &msg_sys, flags, 0);
2115 fput_light(sock->file, fput_needed);
2121 * Linux recvmmsg interface
2124 int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen,
2125 unsigned int flags, struct timespec *timeout)
2127 int fput_needed, err, datagrams;
2128 struct socket *sock;
2129 struct mmsghdr __user *entry;
2130 struct msghdr msg_sys;
2131 struct timespec end_time;
2134 poll_select_set_timeout(&end_time, timeout->tv_sec,
2140 sock = sockfd_lookup_light(fd, &err, &fput_needed);
2144 err = sock_error(sock->sk);
2150 while (datagrams < vlen) {
2152 * No need to ask LSM for more than the first datagram.
2154 err = __sys_recvmsg(sock, (struct msghdr __user *)entry,
2155 &msg_sys, flags, datagrams);
2158 err = put_user(err, &entry->msg_len);
2165 ktime_get_ts(timeout);
2166 *timeout = timespec_sub(end_time, *timeout);
2167 if (timeout->tv_sec < 0) {
2168 timeout->tv_sec = timeout->tv_nsec = 0;
2172 /* Timeout, return less than vlen datagrams */
2173 if (timeout->tv_nsec == 0 && timeout->tv_sec == 0)
2177 /* Out of band data, return right away */
2178 if (msg_sys.msg_flags & MSG_OOB)
2183 fput_light(sock->file, fput_needed);
2188 if (datagrams != 0) {
2190 * We may return less entries than requested (vlen) if the
2191 * sock is non block and there aren't enough datagrams...
2193 if (err != -EAGAIN) {
2195 * ... or if recvmsg returns an error after we
2196 * received some datagrams, where we record the
2197 * error to return on the next call or if the
2198 * app asks about it using getsockopt(SO_ERROR).
2200 sock->sk->sk_err = -err;
2209 SYSCALL_DEFINE5(recvmmsg, int, fd, struct mmsghdr __user *, mmsg,
2210 unsigned int, vlen, unsigned int, flags,
2211 struct timespec __user *, timeout)
2214 struct timespec timeout_sys;
2217 return __sys_recvmmsg(fd, mmsg, vlen, flags, NULL);
2219 if (copy_from_user(&timeout_sys, timeout, sizeof(timeout_sys)))
2222 datagrams = __sys_recvmmsg(fd, mmsg, vlen, flags, &timeout_sys);
2224 if (datagrams > 0 &&
2225 copy_to_user(timeout, &timeout_sys, sizeof(timeout_sys)))
2226 datagrams = -EFAULT;
2231 #ifdef __ARCH_WANT_SYS_SOCKETCALL
2232 /* Argument list sizes for sys_socketcall */
2233 #define AL(x) ((x) * sizeof(unsigned long))
2234 static const unsigned char nargs[20] = {
2235 AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
2236 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
2237 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3),
2244 * System call vectors.
2246 * Argument checking cleaned up. Saved 20% in size.
2247 * This function doesn't need to set the kernel lock because
2248 * it is set by the callees.
2251 SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
2254 unsigned long a0, a1;
2258 if (call < 1 || call > SYS_RECVMMSG)
2262 if (len > sizeof(a))
2265 /* copy_from_user should be SMP safe. */
2266 if (copy_from_user(a, args, len))
2269 audit_socketcall(nargs[call] / sizeof(unsigned long), a);
2276 err = sys_socket(a0, a1, a[2]);
2279 err = sys_bind(a0, (struct sockaddr __user *)a1, a[2]);
2282 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
2285 err = sys_listen(a0, a1);
2288 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2289 (int __user *)a[2], 0);
2291 case SYS_GETSOCKNAME:
2293 sys_getsockname(a0, (struct sockaddr __user *)a1,
2294 (int __user *)a[2]);
2296 case SYS_GETPEERNAME:
2298 sys_getpeername(a0, (struct sockaddr __user *)a1,
2299 (int __user *)a[2]);
2301 case SYS_SOCKETPAIR:
2302 err = sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
2305 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
2308 err = sys_sendto(a0, (void __user *)a1, a[2], a[3],
2309 (struct sockaddr __user *)a[4], a[5]);
2312 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
2315 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
2316 (struct sockaddr __user *)a[4],
2317 (int __user *)a[5]);
2320 err = sys_shutdown(a0, a1);
2322 case SYS_SETSOCKOPT:
2323 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
2325 case SYS_GETSOCKOPT:
2327 sys_getsockopt(a0, a1, a[2], (char __user *)a[3],
2328 (int __user *)a[4]);
2331 err = sys_sendmsg(a0, (struct msghdr __user *)a1, a[2]);
2334 err = sys_recvmsg(a0, (struct msghdr __user *)a1, a[2]);
2337 err = sys_recvmmsg(a0, (struct mmsghdr __user *)a1, a[2], a[3],
2338 (struct timespec __user *)a[4]);
2341 err = sys_accept4(a0, (struct sockaddr __user *)a1,
2342 (int __user *)a[2], a[3]);
2351 #endif /* __ARCH_WANT_SYS_SOCKETCALL */
2354 * sock_register - add a socket protocol handler
2355 * @ops: description of protocol
2357 * This function is called by a protocol handler that wants to
2358 * advertise its address family, and have it linked into the
2359 * socket interface. The value ops->family coresponds to the
2360 * socket system call protocol family.
2362 int sock_register(const struct net_proto_family *ops)
2366 if (ops->family >= NPROTO) {
2367 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family,
2372 spin_lock(&net_family_lock);
2373 if (net_families[ops->family])
2376 net_families[ops->family] = ops;
2379 spin_unlock(&net_family_lock);
2381 printk(KERN_INFO "NET: Registered protocol family %d\n", ops->family);
2386 * sock_unregister - remove a protocol handler
2387 * @family: protocol family to remove
2389 * This function is called by a protocol handler that wants to
2390 * remove its address family, and have it unlinked from the
2391 * new socket creation.
2393 * If protocol handler is a module, then it can use module reference
2394 * counts to protect against new references. If protocol handler is not
2395 * a module then it needs to provide its own protection in
2396 * the ops->create routine.
2398 void sock_unregister(int family)
2400 BUG_ON(family < 0 || family >= NPROTO);
2402 spin_lock(&net_family_lock);
2403 net_families[family] = NULL;
2404 spin_unlock(&net_family_lock);
2408 printk(KERN_INFO "NET: Unregistered protocol family %d\n", family);
2411 static int __init sock_init(void)
2414 * Initialize sock SLAB cache.
2420 * Initialize skbuff SLAB cache
2425 * Initialize the protocols module.
2429 register_filesystem(&sock_fs_type);
2430 sock_mnt = kern_mount(&sock_fs_type);
2432 /* The real protocol initialization is performed in later initcalls.
2435 #ifdef CONFIG_NETFILTER
2442 core_initcall(sock_init); /* early initcall */
2444 #ifdef CONFIG_PROC_FS
2445 void socket_seq_show(struct seq_file *seq)
2450 for_each_possible_cpu(cpu)
2451 counter += per_cpu(sockets_in_use, cpu);
2453 /* It can be negative, by the way. 8) */
2457 seq_printf(seq, "sockets: used %d\n", counter);
2459 #endif /* CONFIG_PROC_FS */
2461 #ifdef CONFIG_COMPAT
2462 static long compat_sock_ioctl(struct file *file, unsigned cmd,
2465 struct socket *sock = file->private_data;
2466 int ret = -ENOIOCTLCMD;
2473 if (sock->ops->compat_ioctl)
2474 ret = sock->ops->compat_ioctl(sock, cmd, arg);
2476 if (ret == -ENOIOCTLCMD &&
2477 (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST))
2478 ret = compat_wext_handle_ioctl(net, cmd, arg);
2484 int kernel_bind(struct socket *sock, struct sockaddr *addr, int addrlen)
2486 return sock->ops->bind(sock, addr, addrlen);
2489 int kernel_listen(struct socket *sock, int backlog)
2491 return sock->ops->listen(sock, backlog);
2494 int kernel_accept(struct socket *sock, struct socket **newsock, int flags)
2496 struct sock *sk = sock->sk;
2499 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
2504 err = sock->ops->accept(sock, *newsock, flags);
2506 sock_release(*newsock);
2511 (*newsock)->ops = sock->ops;
2512 __module_get((*newsock)->ops->owner);
2518 int kernel_connect(struct socket *sock, struct sockaddr *addr, int addrlen,
2521 return sock->ops->connect(sock, addr, addrlen, flags);
2524 int kernel_getsockname(struct socket *sock, struct sockaddr *addr,
2527 return sock->ops->getname(sock, addr, addrlen, 0);
2530 int kernel_getpeername(struct socket *sock, struct sockaddr *addr,
2533 return sock->ops->getname(sock, addr, addrlen, 1);
2536 int kernel_getsockopt(struct socket *sock, int level, int optname,
2537 char *optval, int *optlen)
2539 mm_segment_t oldfs = get_fs();
2543 if (level == SOL_SOCKET)
2544 err = sock_getsockopt(sock, level, optname, optval, optlen);
2546 err = sock->ops->getsockopt(sock, level, optname, optval,
2552 int kernel_setsockopt(struct socket *sock, int level, int optname,
2553 char *optval, unsigned int optlen)
2555 mm_segment_t oldfs = get_fs();
2559 if (level == SOL_SOCKET)
2560 err = sock_setsockopt(sock, level, optname, optval, optlen);
2562 err = sock->ops->setsockopt(sock, level, optname, optval,
2568 int kernel_sendpage(struct socket *sock, struct page *page, int offset,
2569 size_t size, int flags)
2571 if (sock->ops->sendpage)
2572 return sock->ops->sendpage(sock, page, offset, size, flags);
2574 return sock_no_sendpage(sock, page, offset, size, flags);
2577 int kernel_sock_ioctl(struct socket *sock, int cmd, unsigned long arg)
2579 mm_segment_t oldfs = get_fs();
2583 err = sock->ops->ioctl(sock, cmd, arg);
2589 int kernel_sock_shutdown(struct socket *sock, enum sock_shutdown_cmd how)
2591 return sock->ops->shutdown(sock, how);
2594 EXPORT_SYMBOL(sock_create);
2595 EXPORT_SYMBOL(sock_create_kern);
2596 EXPORT_SYMBOL(sock_create_lite);
2597 EXPORT_SYMBOL(sock_map_fd);
2598 EXPORT_SYMBOL(sock_recvmsg);
2599 EXPORT_SYMBOL(sock_register);
2600 EXPORT_SYMBOL(sock_release);
2601 EXPORT_SYMBOL(sock_sendmsg);
2602 EXPORT_SYMBOL(sock_unregister);
2603 EXPORT_SYMBOL(sock_wake_async);
2604 EXPORT_SYMBOL(sockfd_lookup);
2605 EXPORT_SYMBOL(kernel_sendmsg);
2606 EXPORT_SYMBOL(kernel_recvmsg);
2607 EXPORT_SYMBOL(kernel_bind);
2608 EXPORT_SYMBOL(kernel_listen);
2609 EXPORT_SYMBOL(kernel_accept);
2610 EXPORT_SYMBOL(kernel_connect);
2611 EXPORT_SYMBOL(kernel_getsockname);
2612 EXPORT_SYMBOL(kernel_getpeername);
2613 EXPORT_SYMBOL(kernel_getsockopt);
2614 EXPORT_SYMBOL(kernel_setsockopt);
2615 EXPORT_SYMBOL(kernel_sendpage);
2616 EXPORT_SYMBOL(kernel_sock_ioctl);
2617 EXPORT_SYMBOL(kernel_sock_shutdown);