blob: 6b7c3b51a7c12ab6ecc160cc08cda83dbb2a9a9c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NET An implementation of the SOCKET network access protocol.
3 *
4 * Version: @(#)socket.c 1.1.93 18/02/95
5 *
6 * Authors: Orest Zborowski, <obz@Kodak.COM>
Jesper Juhl02c30a82005-05-05 16:16:16 -07007 * Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
9 *
10 * Fixes:
11 * Anonymous : NOTSOCK/BADF cleanup. Error fix in
12 * shutdown()
13 * Alan Cox : verify_area() fixes
14 * Alan Cox : Removed DDI
15 * Jonathan Kamens : SOCK_DGRAM reconnect bug
16 * Alan Cox : Moved a load of checks to the very
17 * top level.
18 * Alan Cox : Move address structures to/from user
19 * mode above the protocol layers.
20 * Rob Janssen : Allow 0 length sends.
21 * Alan Cox : Asynchronous I/O support (cribbed from the
22 * tty drivers).
23 * Niibe Yutaka : Asynchronous I/O for writes (4.4BSD style)
24 * Jeff Uphoff : Made max number of sockets command-line
25 * configurable.
26 * Matti Aarnio : Made the number of sockets dynamic,
27 * to be allocated when needed, and mr.
28 * Uphoff's max is used as max to be
29 * allowed to allocate.
30 * Linus : Argh. removed all the socket allocation
31 * altogether: it's in the inode now.
32 * Alan Cox : Made sock_alloc()/sock_release() public
33 * for NetROM and future kernel nfsd type
34 * stuff.
35 * Alan Cox : sendmsg/recvmsg basics.
36 * Tom Dyas : Export net symbols.
37 * Marcin Dalecki : Fixed problems with CONFIG_NET="n".
38 * Alan Cox : Added thread locking to sys_* calls
39 * for sockets. May have errors at the
40 * moment.
41 * Kevin Buhr : Fixed the dumb errors in the above.
42 * Andi Kleen : Some small cleanups, optimizations,
43 * and fixed a copy_from_user() bug.
44 * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0)
45 * Tigran Aivazian : Made listen(2) backlog sanity checks
46 * protocol-independent
47 *
48 *
49 * This program is free software; you can redistribute it and/or
50 * modify it under the terms of the GNU General Public License
51 * as published by the Free Software Foundation; either version
52 * 2 of the License, or (at your option) any later version.
53 *
54 *
55 * This module is effectively the top level interface to the BSD socket
56 * paradigm.
57 *
58 * Based upon Swansea University Computer Society NET3.039
59 */
60
61#include <linux/config.h>
62#include <linux/mm.h>
63#include <linux/smp_lock.h>
64#include <linux/socket.h>
65#include <linux/file.h>
66#include <linux/net.h>
67#include <linux/interrupt.h>
68#include <linux/netdevice.h>
69#include <linux/proc_fs.h>
70#include <linux/seq_file.h>
71#include <linux/wanrouter.h>
72#include <linux/if_bridge.h>
73#include <linux/init.h>
74#include <linux/poll.h>
75#include <linux/cache.h>
76#include <linux/module.h>
77#include <linux/highmem.h>
78#include <linux/divert.h>
79#include <linux/mount.h>
80#include <linux/security.h>
81#include <linux/syscalls.h>
82#include <linux/compat.h>
83#include <linux/kmod.h>
David Woodhouse3ec3b2f2005-05-17 12:08:48 +010084#include <linux/audit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86#ifdef CONFIG_NET_RADIO
87#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
88#endif /* CONFIG_NET_RADIO */
89
90#include <asm/uaccess.h>
91#include <asm/unistd.h>
92
93#include <net/compat.h>
94
95#include <net/sock.h>
96#include <linux/netfilter.h>
97
98static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
99static ssize_t sock_aio_read(struct kiocb *iocb, char __user *buf,
100 size_t size, loff_t pos);
101static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *buf,
102 size_t size, loff_t pos);
103static int sock_mmap(struct file *file, struct vm_area_struct * vma);
104
105static int sock_close(struct inode *inode, struct file *file);
106static unsigned int sock_poll(struct file *file,
107 struct poll_table_struct *wait);
108static long sock_ioctl(struct file *file,
109 unsigned int cmd, unsigned long arg);
110static int sock_fasync(int fd, struct file *filp, int on);
111static ssize_t sock_readv(struct file *file, const struct iovec *vector,
112 unsigned long count, loff_t *ppos);
113static ssize_t sock_writev(struct file *file, const struct iovec *vector,
114 unsigned long count, loff_t *ppos);
115static ssize_t sock_sendpage(struct file *file, struct page *page,
116 int offset, size_t size, loff_t *ppos, int more);
117
118
119/*
120 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
121 * in the operation structures but are done directly via the socketcall() multiplexor.
122 */
123
124static struct file_operations socket_file_ops = {
125 .owner = THIS_MODULE,
126 .llseek = no_llseek,
127 .aio_read = sock_aio_read,
128 .aio_write = sock_aio_write,
129 .poll = sock_poll,
130 .unlocked_ioctl = sock_ioctl,
131 .mmap = sock_mmap,
132 .open = sock_no_open, /* special open code to disallow open via /proc */
133 .release = sock_close,
134 .fasync = sock_fasync,
135 .readv = sock_readv,
136 .writev = sock_writev,
137 .sendpage = sock_sendpage
138};
139
140/*
141 * The protocol list. Each protocol is registered in here.
142 */
143
144static struct net_proto_family *net_families[NPROTO];
145
146#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
147static atomic_t net_family_lockct = ATOMIC_INIT(0);
148static DEFINE_SPINLOCK(net_family_lock);
149
150/* The strategy is: modifications net_family vector are short, do not
151 sleep and veeery rare, but read access should be free of any exclusive
152 locks.
153 */
154
155static void net_family_write_lock(void)
156{
157 spin_lock(&net_family_lock);
158 while (atomic_read(&net_family_lockct) != 0) {
159 spin_unlock(&net_family_lock);
160
161 yield();
162
163 spin_lock(&net_family_lock);
164 }
165}
166
167static __inline__ void net_family_write_unlock(void)
168{
169 spin_unlock(&net_family_lock);
170}
171
172static __inline__ void net_family_read_lock(void)
173{
174 atomic_inc(&net_family_lockct);
175 spin_unlock_wait(&net_family_lock);
176}
177
178static __inline__ void net_family_read_unlock(void)
179{
180 atomic_dec(&net_family_lockct);
181}
182
183#else
184#define net_family_write_lock() do { } while(0)
185#define net_family_write_unlock() do { } while(0)
186#define net_family_read_lock() do { } while(0)
187#define net_family_read_unlock() do { } while(0)
188#endif
189
190
191/*
192 * Statistics counters of the socket lists
193 */
194
195static DEFINE_PER_CPU(int, sockets_in_use) = 0;
196
197/*
198 * Support routines. Move socket addresses back and forth across the kernel/user
199 * divide and look after the messy bits.
200 */
201
202#define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
203 16 for IP, 16 for IPX,
204 24 for IPv6,
205 about 80 for AX.25
206 must be at least one bigger than
207 the AF_UNIX size (see net/unix/af_unix.c
208 :unix_mkname()).
209 */
210
211/**
212 * move_addr_to_kernel - copy a socket address into kernel space
213 * @uaddr: Address in user space
214 * @kaddr: Address in kernel space
215 * @ulen: Length in user space
216 *
217 * The address is copied into kernel space. If the provided address is
218 * too long an error code of -EINVAL is returned. If the copy gives
219 * invalid addresses -EFAULT is returned. On a success 0 is returned.
220 */
221
222int move_addr_to_kernel(void __user *uaddr, int ulen, void *kaddr)
223{
224 if(ulen<0||ulen>MAX_SOCK_ADDR)
225 return -EINVAL;
226 if(ulen==0)
227 return 0;
228 if(copy_from_user(kaddr,uaddr,ulen))
229 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +0100230 return audit_sockaddr(ulen, kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
233/**
234 * move_addr_to_user - copy an address to user space
235 * @kaddr: kernel space address
236 * @klen: length of address in kernel
237 * @uaddr: user space address
238 * @ulen: pointer to user length field
239 *
240 * The value pointed to by ulen on entry is the buffer length available.
241 * This is overwritten with the buffer space used. -EINVAL is returned
242 * if an overlong buffer is specified or a negative buffer size. -EFAULT
243 * is returned if either the buffer or the length field are not
244 * accessible.
245 * After copying the data up to the limit the user specifies, the true
246 * length of the data is written over the length limit the user
247 * specified. Zero is returned for a success.
248 */
249
250int move_addr_to_user(void *kaddr, int klen, void __user *uaddr, int __user *ulen)
251{
252 int err;
253 int len;
254
255 if((err=get_user(len, ulen)))
256 return err;
257 if(len>klen)
258 len=klen;
259 if(len<0 || len> MAX_SOCK_ADDR)
260 return -EINVAL;
261 if(len)
262 {
263 if(copy_to_user(uaddr,kaddr,len))
264 return -EFAULT;
265 }
266 /*
267 * "fromlen shall refer to the value before truncation.."
268 * 1003.1g
269 */
270 return __put_user(klen, ulen);
271}
272
273#define SOCKFS_MAGIC 0x534F434B
274
275static kmem_cache_t * sock_inode_cachep;
276
277static struct inode *sock_alloc_inode(struct super_block *sb)
278{
279 struct socket_alloc *ei;
280 ei = (struct socket_alloc *)kmem_cache_alloc(sock_inode_cachep, SLAB_KERNEL);
281 if (!ei)
282 return NULL;
283 init_waitqueue_head(&ei->socket.wait);
284
285 ei->socket.fasync_list = NULL;
286 ei->socket.state = SS_UNCONNECTED;
287 ei->socket.flags = 0;
288 ei->socket.ops = NULL;
289 ei->socket.sk = NULL;
290 ei->socket.file = NULL;
291 ei->socket.flags = 0;
292
293 return &ei->vfs_inode;
294}
295
296static void sock_destroy_inode(struct inode *inode)
297{
298 kmem_cache_free(sock_inode_cachep,
299 container_of(inode, struct socket_alloc, vfs_inode));
300}
301
302static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
303{
304 struct socket_alloc *ei = (struct socket_alloc *) foo;
305
306 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
307 SLAB_CTOR_CONSTRUCTOR)
308 inode_init_once(&ei->vfs_inode);
309}
310
311static int init_inodecache(void)
312{
313 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
314 sizeof(struct socket_alloc),
315 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
316 init_once, NULL);
317 if (sock_inode_cachep == NULL)
318 return -ENOMEM;
319 return 0;
320}
321
322static struct super_operations sockfs_ops = {
323 .alloc_inode = sock_alloc_inode,
324 .destroy_inode =sock_destroy_inode,
325 .statfs = simple_statfs,
326};
327
328static struct super_block *sockfs_get_sb(struct file_system_type *fs_type,
329 int flags, const char *dev_name, void *data)
330{
331 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC);
332}
333
334static struct vfsmount *sock_mnt;
335
336static struct file_system_type sock_fs_type = {
337 .name = "sockfs",
338 .get_sb = sockfs_get_sb,
339 .kill_sb = kill_anon_super,
340};
341static int sockfs_delete_dentry(struct dentry *dentry)
342{
343 return 1;
344}
345static struct dentry_operations sockfs_dentry_operations = {
346 .d_delete = sockfs_delete_dentry,
347};
348
349/*
350 * Obtains the first available file descriptor and sets it up for use.
351 *
352 * This function creates file structure and maps it to fd space
353 * of current process. On success it returns file descriptor
354 * and file struct implicitly stored in sock->file.
355 * Note that another thread may close file descriptor before we return
356 * from this function. We use the fact that now we do not refer
357 * to socket after mapping. If one day we will need it, this
358 * function will increment ref. count on file by 1.
359 *
360 * In any case returned fd MAY BE not valid!
361 * This race condition is unavoidable
362 * with shared fd spaces, we cannot solve it inside kernel,
363 * but we take care of internal coherence yet.
364 */
365
366int sock_map_fd(struct socket *sock)
367{
368 int fd;
369 struct qstr this;
370 char name[32];
371
372 /*
373 * Find a file descriptor suitable for return to the user.
374 */
375
376 fd = get_unused_fd();
377 if (fd >= 0) {
378 struct file *file = get_empty_filp();
379
380 if (!file) {
381 put_unused_fd(fd);
382 fd = -ENFILE;
383 goto out;
384 }
385
386 sprintf(name, "[%lu]", SOCK_INODE(sock)->i_ino);
387 this.name = name;
388 this.len = strlen(name);
389 this.hash = SOCK_INODE(sock)->i_ino;
390
391 file->f_dentry = d_alloc(sock_mnt->mnt_sb->s_root, &this);
392 if (!file->f_dentry) {
393 put_filp(file);
394 put_unused_fd(fd);
395 fd = -ENOMEM;
396 goto out;
397 }
398 file->f_dentry->d_op = &sockfs_dentry_operations;
399 d_add(file->f_dentry, SOCK_INODE(sock));
400 file->f_vfsmnt = mntget(sock_mnt);
401 file->f_mapping = file->f_dentry->d_inode->i_mapping;
402
403 sock->file = file;
404 file->f_op = SOCK_INODE(sock)->i_fop = &socket_file_ops;
405 file->f_mode = FMODE_READ | FMODE_WRITE;
406 file->f_flags = O_RDWR;
407 file->f_pos = 0;
408 fd_install(fd, file);
409 }
410
411out:
412 return fd;
413}
414
415/**
416 * sockfd_lookup - Go from a file number to its socket slot
417 * @fd: file handle
418 * @err: pointer to an error code return
419 *
420 * The file handle passed in is locked and the socket it is bound
421 * too is returned. If an error occurs the err pointer is overwritten
422 * with a negative errno code and NULL is returned. The function checks
423 * for both invalid handles and passing a handle which is not a socket.
424 *
425 * On a success the socket object pointer is returned.
426 */
427
428struct socket *sockfd_lookup(int fd, int *err)
429{
430 struct file *file;
431 struct inode *inode;
432 struct socket *sock;
433
434 if (!(file = fget(fd)))
435 {
436 *err = -EBADF;
437 return NULL;
438 }
439
440 inode = file->f_dentry->d_inode;
441 if (!S_ISSOCK(inode->i_mode)) {
442 *err = -ENOTSOCK;
443 fput(file);
444 return NULL;
445 }
446
447 sock = SOCKET_I(inode);
448 if (sock->file != file) {
449 printk(KERN_ERR "socki_lookup: socket file changed!\n");
450 sock->file = file;
451 }
452 return sock;
453}
454
455/**
456 * sock_alloc - allocate a socket
457 *
458 * Allocate a new inode and socket object. The two are bound together
459 * and initialised. The socket is then returned. If we are out of inodes
460 * NULL is returned.
461 */
462
463static struct socket *sock_alloc(void)
464{
465 struct inode * inode;
466 struct socket * sock;
467
468 inode = new_inode(sock_mnt->mnt_sb);
469 if (!inode)
470 return NULL;
471
472 sock = SOCKET_I(inode);
473
474 inode->i_mode = S_IFSOCK|S_IRWXUGO;
475 inode->i_uid = current->fsuid;
476 inode->i_gid = current->fsgid;
477
478 get_cpu_var(sockets_in_use)++;
479 put_cpu_var(sockets_in_use);
480 return sock;
481}
482
483/*
484 * In theory you can't get an open on this inode, but /proc provides
485 * a back door. Remember to keep it shut otherwise you'll let the
486 * creepy crawlies in.
487 */
488
489static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
490{
491 return -ENXIO;
492}
493
494struct file_operations bad_sock_fops = {
495 .owner = THIS_MODULE,
496 .open = sock_no_open,
497};
498
499/**
500 * sock_release - close a socket
501 * @sock: socket to close
502 *
503 * The socket is released from the protocol stack if it has a release
504 * callback, and the inode is then released if the socket is bound to
505 * an inode not a file.
506 */
507
508void sock_release(struct socket *sock)
509{
510 if (sock->ops) {
511 struct module *owner = sock->ops->owner;
512
513 sock->ops->release(sock);
514 sock->ops = NULL;
515 module_put(owner);
516 }
517
518 if (sock->fasync_list)
519 printk(KERN_ERR "sock_release: fasync list not empty!\n");
520
521 get_cpu_var(sockets_in_use)--;
522 put_cpu_var(sockets_in_use);
523 if (!sock->file) {
524 iput(SOCK_INODE(sock));
525 return;
526 }
527 sock->file=NULL;
528}
529
530static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
531 struct msghdr *msg, size_t size)
532{
533 struct sock_iocb *si = kiocb_to_siocb(iocb);
534 int err;
535
536 si->sock = sock;
537 si->scm = NULL;
538 si->msg = msg;
539 si->size = size;
540
541 err = security_socket_sendmsg(sock, msg, size);
542 if (err)
543 return err;
544
545 return sock->ops->sendmsg(iocb, sock, msg, size);
546}
547
548int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
549{
550 struct kiocb iocb;
551 struct sock_iocb siocb;
552 int ret;
553
554 init_sync_kiocb(&iocb, NULL);
555 iocb.private = &siocb;
556 ret = __sock_sendmsg(&iocb, sock, msg, size);
557 if (-EIOCBQUEUED == ret)
558 ret = wait_on_sync_kiocb(&iocb);
559 return ret;
560}
561
562int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
563 struct kvec *vec, size_t num, size_t size)
564{
565 mm_segment_t oldfs = get_fs();
566 int result;
567
568 set_fs(KERNEL_DS);
569 /*
570 * the following is safe, since for compiler definitions of kvec and
571 * iovec are identical, yielding the same in-core layout and alignment
572 */
573 msg->msg_iov = (struct iovec *)vec,
574 msg->msg_iovlen = num;
575 result = sock_sendmsg(sock, msg, size);
576 set_fs(oldfs);
577 return result;
578}
579
580static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
581 struct msghdr *msg, size_t size, int flags)
582{
583 int err;
584 struct sock_iocb *si = kiocb_to_siocb(iocb);
585
586 si->sock = sock;
587 si->scm = NULL;
588 si->msg = msg;
589 si->size = size;
590 si->flags = flags;
591
592 err = security_socket_recvmsg(sock, msg, size, flags);
593 if (err)
594 return err;
595
596 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
597}
598
599int sock_recvmsg(struct socket *sock, struct msghdr *msg,
600 size_t size, int flags)
601{
602 struct kiocb iocb;
603 struct sock_iocb siocb;
604 int ret;
605
606 init_sync_kiocb(&iocb, NULL);
607 iocb.private = &siocb;
608 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
609 if (-EIOCBQUEUED == ret)
610 ret = wait_on_sync_kiocb(&iocb);
611 return ret;
612}
613
614int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
615 struct kvec *vec, size_t num,
616 size_t size, int flags)
617{
618 mm_segment_t oldfs = get_fs();
619 int result;
620
621 set_fs(KERNEL_DS);
622 /*
623 * the following is safe, since for compiler definitions of kvec and
624 * iovec are identical, yielding the same in-core layout and alignment
625 */
626 msg->msg_iov = (struct iovec *)vec,
627 msg->msg_iovlen = num;
628 result = sock_recvmsg(sock, msg, size, flags);
629 set_fs(oldfs);
630 return result;
631}
632
633static void sock_aio_dtor(struct kiocb *iocb)
634{
635 kfree(iocb->private);
636}
637
638/*
639 * Read data from a socket. ubuf is a user mode pointer. We make sure the user
640 * area ubuf...ubuf+size-1 is writable before asking the protocol.
641 */
642
643static ssize_t sock_aio_read(struct kiocb *iocb, char __user *ubuf,
644 size_t size, loff_t pos)
645{
646 struct sock_iocb *x, siocb;
647 struct socket *sock;
648 int flags;
649
650 if (pos != 0)
651 return -ESPIPE;
652 if (size==0) /* Match SYS5 behaviour */
653 return 0;
654
655 if (is_sync_kiocb(iocb))
656 x = &siocb;
657 else {
658 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
659 if (!x)
660 return -ENOMEM;
661 iocb->ki_dtor = sock_aio_dtor;
662 }
663 iocb->private = x;
664 x->kiocb = iocb;
665 sock = SOCKET_I(iocb->ki_filp->f_dentry->d_inode);
666
667 x->async_msg.msg_name = NULL;
668 x->async_msg.msg_namelen = 0;
669 x->async_msg.msg_iov = &x->async_iov;
670 x->async_msg.msg_iovlen = 1;
671 x->async_msg.msg_control = NULL;
672 x->async_msg.msg_controllen = 0;
673 x->async_iov.iov_base = ubuf;
674 x->async_iov.iov_len = size;
675 flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
676
677 return __sock_recvmsg(iocb, sock, &x->async_msg, size, flags);
678}
679
680
681/*
682 * Write data to a socket. We verify that the user area ubuf..ubuf+size-1
683 * is readable by the user process.
684 */
685
686static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *ubuf,
687 size_t size, loff_t pos)
688{
689 struct sock_iocb *x, siocb;
690 struct socket *sock;
691
692 if (pos != 0)
693 return -ESPIPE;
694 if(size==0) /* Match SYS5 behaviour */
695 return 0;
696
697 if (is_sync_kiocb(iocb))
698 x = &siocb;
699 else {
700 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
701 if (!x)
702 return -ENOMEM;
703 iocb->ki_dtor = sock_aio_dtor;
704 }
705 iocb->private = x;
706 x->kiocb = iocb;
707 sock = SOCKET_I(iocb->ki_filp->f_dentry->d_inode);
708
709 x->async_msg.msg_name = NULL;
710 x->async_msg.msg_namelen = 0;
711 x->async_msg.msg_iov = &x->async_iov;
712 x->async_msg.msg_iovlen = 1;
713 x->async_msg.msg_control = NULL;
714 x->async_msg.msg_controllen = 0;
715 x->async_msg.msg_flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
716 if (sock->type == SOCK_SEQPACKET)
717 x->async_msg.msg_flags |= MSG_EOR;
718 x->async_iov.iov_base = (void __user *)ubuf;
719 x->async_iov.iov_len = size;
720
721 return __sock_sendmsg(iocb, sock, &x->async_msg, size);
722}
723
724ssize_t sock_sendpage(struct file *file, struct page *page,
725 int offset, size_t size, loff_t *ppos, int more)
726{
727 struct socket *sock;
728 int flags;
729
730 sock = SOCKET_I(file->f_dentry->d_inode);
731
732 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
733 if (more)
734 flags |= MSG_MORE;
735
736 return sock->ops->sendpage(sock, page, offset, size, flags);
737}
738
739static int sock_readv_writev(int type, struct inode * inode,
740 struct file * file, const struct iovec * iov,
741 long count, size_t size)
742{
743 struct msghdr msg;
744 struct socket *sock;
745
746 sock = SOCKET_I(inode);
747
748 msg.msg_name = NULL;
749 msg.msg_namelen = 0;
750 msg.msg_control = NULL;
751 msg.msg_controllen = 0;
752 msg.msg_iov = (struct iovec *) iov;
753 msg.msg_iovlen = count;
754 msg.msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
755
756 /* read() does a VERIFY_WRITE */
757 if (type == VERIFY_WRITE)
758 return sock_recvmsg(sock, &msg, size, msg.msg_flags);
759
760 if (sock->type == SOCK_SEQPACKET)
761 msg.msg_flags |= MSG_EOR;
762
763 return sock_sendmsg(sock, &msg, size);
764}
765
766static ssize_t sock_readv(struct file *file, const struct iovec *vector,
767 unsigned long count, loff_t *ppos)
768{
769 size_t tot_len = 0;
770 int i;
771 for (i = 0 ; i < count ; i++)
772 tot_len += vector[i].iov_len;
773 return sock_readv_writev(VERIFY_WRITE, file->f_dentry->d_inode,
774 file, vector, count, tot_len);
775}
776
777static ssize_t sock_writev(struct file *file, const struct iovec *vector,
778 unsigned long count, loff_t *ppos)
779{
780 size_t tot_len = 0;
781 int i;
782 for (i = 0 ; i < count ; i++)
783 tot_len += vector[i].iov_len;
784 return sock_readv_writev(VERIFY_READ, file->f_dentry->d_inode,
785 file, vector, count, tot_len);
786}
787
788
789/*
790 * Atomic setting of ioctl hooks to avoid race
791 * with module unload.
792 */
793
794static DECLARE_MUTEX(br_ioctl_mutex);
795static int (*br_ioctl_hook)(unsigned int cmd, void __user *arg) = NULL;
796
797void brioctl_set(int (*hook)(unsigned int, void __user *))
798{
799 down(&br_ioctl_mutex);
800 br_ioctl_hook = hook;
801 up(&br_ioctl_mutex);
802}
803EXPORT_SYMBOL(brioctl_set);
804
805static DECLARE_MUTEX(vlan_ioctl_mutex);
806static int (*vlan_ioctl_hook)(void __user *arg);
807
808void vlan_ioctl_set(int (*hook)(void __user *))
809{
810 down(&vlan_ioctl_mutex);
811 vlan_ioctl_hook = hook;
812 up(&vlan_ioctl_mutex);
813}
814EXPORT_SYMBOL(vlan_ioctl_set);
815
816static DECLARE_MUTEX(dlci_ioctl_mutex);
817static int (*dlci_ioctl_hook)(unsigned int, void __user *);
818
819void dlci_ioctl_set(int (*hook)(unsigned int, void __user *))
820{
821 down(&dlci_ioctl_mutex);
822 dlci_ioctl_hook = hook;
823 up(&dlci_ioctl_mutex);
824}
825EXPORT_SYMBOL(dlci_ioctl_set);
826
827/*
828 * With an ioctl, arg may well be a user mode pointer, but we don't know
829 * what to do with it - that's up to the protocol still.
830 */
831
832static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
833{
834 struct socket *sock;
835 void __user *argp = (void __user *)arg;
836 int pid, err;
837
838 sock = SOCKET_I(file->f_dentry->d_inode);
839 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
840 err = dev_ioctl(cmd, argp);
841 } else
842#ifdef WIRELESS_EXT
843 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
844 err = dev_ioctl(cmd, argp);
845 } else
846#endif /* WIRELESS_EXT */
847 switch (cmd) {
848 case FIOSETOWN:
849 case SIOCSPGRP:
850 err = -EFAULT;
851 if (get_user(pid, (int __user *)argp))
852 break;
853 err = f_setown(sock->file, pid, 1);
854 break;
855 case FIOGETOWN:
856 case SIOCGPGRP:
857 err = put_user(sock->file->f_owner.pid, (int __user *)argp);
858 break;
859 case SIOCGIFBR:
860 case SIOCSIFBR:
861 case SIOCBRADDBR:
862 case SIOCBRDELBR:
863 err = -ENOPKG;
864 if (!br_ioctl_hook)
865 request_module("bridge");
866
867 down(&br_ioctl_mutex);
868 if (br_ioctl_hook)
869 err = br_ioctl_hook(cmd, argp);
870 up(&br_ioctl_mutex);
871 break;
872 case SIOCGIFVLAN:
873 case SIOCSIFVLAN:
874 err = -ENOPKG;
875 if (!vlan_ioctl_hook)
876 request_module("8021q");
877
878 down(&vlan_ioctl_mutex);
879 if (vlan_ioctl_hook)
880 err = vlan_ioctl_hook(argp);
881 up(&vlan_ioctl_mutex);
882 break;
883 case SIOCGIFDIVERT:
884 case SIOCSIFDIVERT:
885 /* Convert this to call through a hook */
886 err = divert_ioctl(cmd, argp);
887 break;
888 case SIOCADDDLCI:
889 case SIOCDELDLCI:
890 err = -ENOPKG;
891 if (!dlci_ioctl_hook)
892 request_module("dlci");
893
894 if (dlci_ioctl_hook) {
895 down(&dlci_ioctl_mutex);
896 err = dlci_ioctl_hook(cmd, argp);
897 up(&dlci_ioctl_mutex);
898 }
899 break;
900 default:
901 err = sock->ops->ioctl(sock, cmd, arg);
902 break;
903 }
904 return err;
905}
906
907int sock_create_lite(int family, int type, int protocol, struct socket **res)
908{
909 int err;
910 struct socket *sock = NULL;
911
912 err = security_socket_create(family, type, protocol, 1);
913 if (err)
914 goto out;
915
916 sock = sock_alloc();
917 if (!sock) {
918 err = -ENOMEM;
919 goto out;
920 }
921
922 security_socket_post_create(sock, family, type, protocol, 1);
923 sock->type = type;
924out:
925 *res = sock;
926 return err;
927}
928
929/* No kernel lock held - perfect */
930static unsigned int sock_poll(struct file *file, poll_table * wait)
931{
932 struct socket *sock;
933
934 /*
935 * We can't return errors to poll, so it's either yes or no.
936 */
937 sock = SOCKET_I(file->f_dentry->d_inode);
938 return sock->ops->poll(file, sock, wait);
939}
940
941static int sock_mmap(struct file * file, struct vm_area_struct * vma)
942{
943 struct socket *sock = SOCKET_I(file->f_dentry->d_inode);
944
945 return sock->ops->mmap(file, sock, vma);
946}
947
948int sock_close(struct inode *inode, struct file *filp)
949{
950 /*
951 * It was possible the inode is NULL we were
952 * closing an unfinished socket.
953 */
954
955 if (!inode)
956 {
957 printk(KERN_DEBUG "sock_close: NULL inode\n");
958 return 0;
959 }
960 sock_fasync(-1, filp, 0);
961 sock_release(SOCKET_I(inode));
962 return 0;
963}
964
965/*
966 * Update the socket async list
967 *
968 * Fasync_list locking strategy.
969 *
970 * 1. fasync_list is modified only under process context socket lock
971 * i.e. under semaphore.
972 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
973 * or under socket lock.
974 * 3. fasync_list can be used from softirq context, so that
975 * modification under socket lock have to be enhanced with
976 * write_lock_bh(&sk->sk_callback_lock).
977 * --ANK (990710)
978 */
979
980static int sock_fasync(int fd, struct file *filp, int on)
981{
982 struct fasync_struct *fa, *fna=NULL, **prev;
983 struct socket *sock;
984 struct sock *sk;
985
986 if (on)
987 {
988 fna=(struct fasync_struct *)kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
989 if(fna==NULL)
990 return -ENOMEM;
991 }
992
993 sock = SOCKET_I(filp->f_dentry->d_inode);
994
995 if ((sk=sock->sk) == NULL) {
996 kfree(fna);
997 return -EINVAL;
998 }
999
1000 lock_sock(sk);
1001
1002 prev=&(sock->fasync_list);
1003
1004 for (fa=*prev; fa!=NULL; prev=&fa->fa_next,fa=*prev)
1005 if (fa->fa_file==filp)
1006 break;
1007
1008 if(on)
1009 {
1010 if(fa!=NULL)
1011 {
1012 write_lock_bh(&sk->sk_callback_lock);
1013 fa->fa_fd=fd;
1014 write_unlock_bh(&sk->sk_callback_lock);
1015
1016 kfree(fna);
1017 goto out;
1018 }
1019 fna->fa_file=filp;
1020 fna->fa_fd=fd;
1021 fna->magic=FASYNC_MAGIC;
1022 fna->fa_next=sock->fasync_list;
1023 write_lock_bh(&sk->sk_callback_lock);
1024 sock->fasync_list=fna;
1025 write_unlock_bh(&sk->sk_callback_lock);
1026 }
1027 else
1028 {
1029 if (fa!=NULL)
1030 {
1031 write_lock_bh(&sk->sk_callback_lock);
1032 *prev=fa->fa_next;
1033 write_unlock_bh(&sk->sk_callback_lock);
1034 kfree(fa);
1035 }
1036 }
1037
1038out:
1039 release_sock(sock->sk);
1040 return 0;
1041}
1042
1043/* This function may be called only under socket lock or callback_lock */
1044
1045int sock_wake_async(struct socket *sock, int how, int band)
1046{
1047 if (!sock || !sock->fasync_list)
1048 return -1;
1049 switch (how)
1050 {
1051 case 1:
1052
1053 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1054 break;
1055 goto call_kill;
1056 case 2:
1057 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1058 break;
1059 /* fall through */
1060 case 0:
1061 call_kill:
1062 __kill_fasync(sock->fasync_list, SIGIO, band);
1063 break;
1064 case 3:
1065 __kill_fasync(sock->fasync_list, SIGURG, band);
1066 }
1067 return 0;
1068}
1069
1070static int __sock_create(int family, int type, int protocol, struct socket **res, int kern)
1071{
1072 int err;
1073 struct socket *sock;
1074
1075 /*
1076 * Check protocol is in range
1077 */
1078 if (family < 0 || family >= NPROTO)
1079 return -EAFNOSUPPORT;
1080 if (type < 0 || type >= SOCK_MAX)
1081 return -EINVAL;
1082
1083 /* Compatibility.
1084
1085 This uglymoron is moved from INET layer to here to avoid
1086 deadlock in module load.
1087 */
1088 if (family == PF_INET && type == SOCK_PACKET) {
1089 static int warned;
1090 if (!warned) {
1091 warned = 1;
1092 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n", current->comm);
1093 }
1094 family = PF_PACKET;
1095 }
1096
1097 err = security_socket_create(family, type, protocol, kern);
1098 if (err)
1099 return err;
1100
1101#if defined(CONFIG_KMOD)
1102 /* Attempt to load a protocol module if the find failed.
1103 *
1104 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1105 * requested real, full-featured networking support upon configuration.
1106 * Otherwise module support will break!
1107 */
1108 if (net_families[family]==NULL)
1109 {
1110 request_module("net-pf-%d",family);
1111 }
1112#endif
1113
1114 net_family_read_lock();
1115 if (net_families[family] == NULL) {
1116 err = -EAFNOSUPPORT;
1117 goto out;
1118 }
1119
1120/*
1121 * Allocate the socket and allow the family to set things up. if
1122 * the protocol is 0, the family is instructed to select an appropriate
1123 * default.
1124 */
1125
1126 if (!(sock = sock_alloc())) {
1127 printk(KERN_WARNING "socket: no more sockets\n");
1128 err = -ENFILE; /* Not exactly a match, but its the
1129 closest posix thing */
1130 goto out;
1131 }
1132
1133 sock->type = type;
1134
1135 /*
1136 * We will call the ->create function, that possibly is in a loadable
1137 * module, so we have to bump that loadable module refcnt first.
1138 */
1139 err = -EAFNOSUPPORT;
1140 if (!try_module_get(net_families[family]->owner))
1141 goto out_release;
1142
1143 if ((err = net_families[family]->create(sock, protocol)) < 0)
1144 goto out_module_put;
1145 /*
1146 * Now to bump the refcnt of the [loadable] module that owns this
1147 * socket at sock_release time we decrement its refcnt.
1148 */
1149 if (!try_module_get(sock->ops->owner)) {
1150 sock->ops = NULL;
1151 goto out_module_put;
1152 }
1153 /*
1154 * Now that we're done with the ->create function, the [loadable]
1155 * module can have its refcnt decremented
1156 */
1157 module_put(net_families[family]->owner);
1158 *res = sock;
1159 security_socket_post_create(sock, family, type, protocol, kern);
1160
1161out:
1162 net_family_read_unlock();
1163 return err;
1164out_module_put:
1165 module_put(net_families[family]->owner);
1166out_release:
1167 sock_release(sock);
1168 goto out;
1169}
1170
1171int sock_create(int family, int type, int protocol, struct socket **res)
1172{
1173 return __sock_create(family, type, protocol, res, 0);
1174}
1175
1176int sock_create_kern(int family, int type, int protocol, struct socket **res)
1177{
1178 return __sock_create(family, type, protocol, res, 1);
1179}
1180
1181asmlinkage long sys_socket(int family, int type, int protocol)
1182{
1183 int retval;
1184 struct socket *sock;
1185
1186 retval = sock_create(family, type, protocol, &sock);
1187 if (retval < 0)
1188 goto out;
1189
1190 retval = sock_map_fd(sock);
1191 if (retval < 0)
1192 goto out_release;
1193
1194out:
1195 /* It may be already another descriptor 8) Not kernel problem. */
1196 return retval;
1197
1198out_release:
1199 sock_release(sock);
1200 return retval;
1201}
1202
1203/*
1204 * Create a pair of connected sockets.
1205 */
1206
1207asmlinkage long sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1208{
1209 struct socket *sock1, *sock2;
1210 int fd1, fd2, err;
1211
1212 /*
1213 * Obtain the first socket and check if the underlying protocol
1214 * supports the socketpair call.
1215 */
1216
1217 err = sock_create(family, type, protocol, &sock1);
1218 if (err < 0)
1219 goto out;
1220
1221 err = sock_create(family, type, protocol, &sock2);
1222 if (err < 0)
1223 goto out_release_1;
1224
1225 err = sock1->ops->socketpair(sock1, sock2);
1226 if (err < 0)
1227 goto out_release_both;
1228
1229 fd1 = fd2 = -1;
1230
1231 err = sock_map_fd(sock1);
1232 if (err < 0)
1233 goto out_release_both;
1234 fd1 = err;
1235
1236 err = sock_map_fd(sock2);
1237 if (err < 0)
1238 goto out_close_1;
1239 fd2 = err;
1240
1241 /* fd1 and fd2 may be already another descriptors.
1242 * Not kernel problem.
1243 */
1244
1245 err = put_user(fd1, &usockvec[0]);
1246 if (!err)
1247 err = put_user(fd2, &usockvec[1]);
1248 if (!err)
1249 return 0;
1250
1251 sys_close(fd2);
1252 sys_close(fd1);
1253 return err;
1254
1255out_close_1:
1256 sock_release(sock2);
1257 sys_close(fd1);
1258 return err;
1259
1260out_release_both:
1261 sock_release(sock2);
1262out_release_1:
1263 sock_release(sock1);
1264out:
1265 return err;
1266}
1267
1268
1269/*
1270 * Bind a name to a socket. Nothing much to do here since it's
1271 * the protocol's responsibility to handle the local address.
1272 *
1273 * We move the socket address to kernel space before we call
1274 * the protocol layer (having also checked the address is ok).
1275 */
1276
1277asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1278{
1279 struct socket *sock;
1280 char address[MAX_SOCK_ADDR];
1281 int err;
1282
1283 if((sock = sockfd_lookup(fd,&err))!=NULL)
1284 {
1285 if((err=move_addr_to_kernel(umyaddr,addrlen,address))>=0) {
1286 err = security_socket_bind(sock, (struct sockaddr *)address, addrlen);
1287 if (err) {
1288 sockfd_put(sock);
1289 return err;
1290 }
1291 err = sock->ops->bind(sock, (struct sockaddr *)address, addrlen);
1292 }
1293 sockfd_put(sock);
1294 }
1295 return err;
1296}
1297
1298
1299/*
1300 * Perform a listen. Basically, we allow the protocol to do anything
1301 * necessary for a listen, and if that works, we mark the socket as
1302 * ready for listening.
1303 */
1304
1305int sysctl_somaxconn = SOMAXCONN;
1306
1307asmlinkage long sys_listen(int fd, int backlog)
1308{
1309 struct socket *sock;
1310 int err;
1311
1312 if ((sock = sockfd_lookup(fd, &err)) != NULL) {
1313 if ((unsigned) backlog > sysctl_somaxconn)
1314 backlog = sysctl_somaxconn;
1315
1316 err = security_socket_listen(sock, backlog);
1317 if (err) {
1318 sockfd_put(sock);
1319 return err;
1320 }
1321
1322 err=sock->ops->listen(sock, backlog);
1323 sockfd_put(sock);
1324 }
1325 return err;
1326}
1327
1328
1329/*
1330 * For accept, we attempt to create a new socket, set up the link
1331 * with the client, wake up the client, then return the new
1332 * connected fd. We collect the address of the connector in kernel
1333 * space and move it to user at the very end. This is unclean because
1334 * we open the socket then return an error.
1335 *
1336 * 1003.1g adds the ability to recvmsg() to query connection pending
1337 * status to recvmsg. We need to add that support in a way thats
1338 * clean when we restucture accept also.
1339 */
1340
1341asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr, int __user *upeer_addrlen)
1342{
1343 struct socket *sock, *newsock;
1344 int err, len;
1345 char address[MAX_SOCK_ADDR];
1346
1347 sock = sockfd_lookup(fd, &err);
1348 if (!sock)
1349 goto out;
1350
1351 err = -ENFILE;
1352 if (!(newsock = sock_alloc()))
1353 goto out_put;
1354
1355 newsock->type = sock->type;
1356 newsock->ops = sock->ops;
1357
1358 err = security_socket_accept(sock, newsock);
1359 if (err)
1360 goto out_release;
1361
1362 /*
1363 * We don't need try_module_get here, as the listening socket (sock)
1364 * has the protocol module (sock->ops->owner) held.
1365 */
1366 __module_get(newsock->ops->owner);
1367
1368 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1369 if (err < 0)
1370 goto out_release;
1371
1372 if (upeer_sockaddr) {
1373 if(newsock->ops->getname(newsock, (struct sockaddr *)address, &len, 2)<0) {
1374 err = -ECONNABORTED;
1375 goto out_release;
1376 }
1377 err = move_addr_to_user(address, len, upeer_sockaddr, upeer_addrlen);
1378 if (err < 0)
1379 goto out_release;
1380 }
1381
1382 /* File flags are not inherited via accept() unlike another OSes. */
1383
1384 if ((err = sock_map_fd(newsock)) < 0)
1385 goto out_release;
1386
1387 security_socket_post_accept(sock, newsock);
1388
1389out_put:
1390 sockfd_put(sock);
1391out:
1392 return err;
1393out_release:
1394 sock_release(newsock);
1395 goto out_put;
1396}
1397
1398
1399/*
1400 * Attempt to connect to a socket with the server address. The address
1401 * is in user space so we verify it is OK and move it to kernel space.
1402 *
1403 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1404 * break bindings
1405 *
1406 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1407 * other SEQPACKET protocols that take time to connect() as it doesn't
1408 * include the -EINPROGRESS status for such sockets.
1409 */
1410
1411asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1412{
1413 struct socket *sock;
1414 char address[MAX_SOCK_ADDR];
1415 int err;
1416
1417 sock = sockfd_lookup(fd, &err);
1418 if (!sock)
1419 goto out;
1420 err = move_addr_to_kernel(uservaddr, addrlen, address);
1421 if (err < 0)
1422 goto out_put;
1423
1424 err = security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1425 if (err)
1426 goto out_put;
1427
1428 err = sock->ops->connect(sock, (struct sockaddr *) address, addrlen,
1429 sock->file->f_flags);
1430out_put:
1431 sockfd_put(sock);
1432out:
1433 return err;
1434}
1435
1436/*
1437 * Get the local address ('name') of a socket object. Move the obtained
1438 * name to user space.
1439 */
1440
1441asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1442{
1443 struct socket *sock;
1444 char address[MAX_SOCK_ADDR];
1445 int len, err;
1446
1447 sock = sockfd_lookup(fd, &err);
1448 if (!sock)
1449 goto out;
1450
1451 err = security_socket_getsockname(sock);
1452 if (err)
1453 goto out_put;
1454
1455 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 0);
1456 if (err)
1457 goto out_put;
1458 err = move_addr_to_user(address, len, usockaddr, usockaddr_len);
1459
1460out_put:
1461 sockfd_put(sock);
1462out:
1463 return err;
1464}
1465
1466/*
1467 * Get the remote address ('name') of a socket object. Move the obtained
1468 * name to user space.
1469 */
1470
1471asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1472{
1473 struct socket *sock;
1474 char address[MAX_SOCK_ADDR];
1475 int len, err;
1476
1477 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1478 {
1479 err = security_socket_getpeername(sock);
1480 if (err) {
1481 sockfd_put(sock);
1482 return err;
1483 }
1484
1485 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 1);
1486 if (!err)
1487 err=move_addr_to_user(address,len, usockaddr, usockaddr_len);
1488 sockfd_put(sock);
1489 }
1490 return err;
1491}
1492
1493/*
1494 * Send a datagram to a given address. We move the address into kernel
1495 * space and check the user space data area is readable before invoking
1496 * the protocol.
1497 */
1498
1499asmlinkage long sys_sendto(int fd, void __user * buff, size_t len, unsigned flags,
1500 struct sockaddr __user *addr, int addr_len)
1501{
1502 struct socket *sock;
1503 char address[MAX_SOCK_ADDR];
1504 int err;
1505 struct msghdr msg;
1506 struct iovec iov;
1507
1508 sock = sockfd_lookup(fd, &err);
1509 if (!sock)
1510 goto out;
1511 iov.iov_base=buff;
1512 iov.iov_len=len;
1513 msg.msg_name=NULL;
1514 msg.msg_iov=&iov;
1515 msg.msg_iovlen=1;
1516 msg.msg_control=NULL;
1517 msg.msg_controllen=0;
1518 msg.msg_namelen=0;
1519 if(addr)
1520 {
1521 err = move_addr_to_kernel(addr, addr_len, address);
1522 if (err < 0)
1523 goto out_put;
1524 msg.msg_name=address;
1525 msg.msg_namelen=addr_len;
1526 }
1527 if (sock->file->f_flags & O_NONBLOCK)
1528 flags |= MSG_DONTWAIT;
1529 msg.msg_flags = flags;
1530 err = sock_sendmsg(sock, &msg, len);
1531
1532out_put:
1533 sockfd_put(sock);
1534out:
1535 return err;
1536}
1537
1538/*
1539 * Send a datagram down a socket.
1540 */
1541
1542asmlinkage long sys_send(int fd, void __user * buff, size_t len, unsigned flags)
1543{
1544 return sys_sendto(fd, buff, len, flags, NULL, 0);
1545}
1546
1547/*
1548 * Receive a frame from the socket and optionally record the address of the
1549 * sender. We verify the buffers are writable and if needed move the
1550 * sender address from kernel to user space.
1551 */
1552
1553asmlinkage long sys_recvfrom(int fd, void __user * ubuf, size_t size, unsigned flags,
1554 struct sockaddr __user *addr, int __user *addr_len)
1555{
1556 struct socket *sock;
1557 struct iovec iov;
1558 struct msghdr msg;
1559 char address[MAX_SOCK_ADDR];
1560 int err,err2;
1561
1562 sock = sockfd_lookup(fd, &err);
1563 if (!sock)
1564 goto out;
1565
1566 msg.msg_control=NULL;
1567 msg.msg_controllen=0;
1568 msg.msg_iovlen=1;
1569 msg.msg_iov=&iov;
1570 iov.iov_len=size;
1571 iov.iov_base=ubuf;
1572 msg.msg_name=address;
1573 msg.msg_namelen=MAX_SOCK_ADDR;
1574 if (sock->file->f_flags & O_NONBLOCK)
1575 flags |= MSG_DONTWAIT;
1576 err=sock_recvmsg(sock, &msg, size, flags);
1577
1578 if(err >= 0 && addr != NULL)
1579 {
1580 err2=move_addr_to_user(address, msg.msg_namelen, addr, addr_len);
1581 if(err2<0)
1582 err=err2;
1583 }
1584 sockfd_put(sock);
1585out:
1586 return err;
1587}
1588
1589/*
1590 * Receive a datagram from a socket.
1591 */
1592
1593asmlinkage long sys_recv(int fd, void __user * ubuf, size_t size, unsigned flags)
1594{
1595 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1596}
1597
1598/*
1599 * Set a socket option. Because we don't know the option lengths we have
1600 * to pass the user mode parameter for the protocols to sort out.
1601 */
1602
1603asmlinkage long sys_setsockopt(int fd, int level, int optname, char __user *optval, int optlen)
1604{
1605 int err;
1606 struct socket *sock;
1607
1608 if (optlen < 0)
1609 return -EINVAL;
1610
1611 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1612 {
1613 err = security_socket_setsockopt(sock,level,optname);
1614 if (err) {
1615 sockfd_put(sock);
1616 return err;
1617 }
1618
1619 if (level == SOL_SOCKET)
1620 err=sock_setsockopt(sock,level,optname,optval,optlen);
1621 else
1622 err=sock->ops->setsockopt(sock, level, optname, optval, optlen);
1623 sockfd_put(sock);
1624 }
1625 return err;
1626}
1627
1628/*
1629 * Get a socket option. Because we don't know the option lengths we have
1630 * to pass a user mode parameter for the protocols to sort out.
1631 */
1632
1633asmlinkage long sys_getsockopt(int fd, int level, int optname, char __user *optval, int __user *optlen)
1634{
1635 int err;
1636 struct socket *sock;
1637
1638 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1639 {
1640 err = security_socket_getsockopt(sock, level,
1641 optname);
1642 if (err) {
1643 sockfd_put(sock);
1644 return err;
1645 }
1646
1647 if (level == SOL_SOCKET)
1648 err=sock_getsockopt(sock,level,optname,optval,optlen);
1649 else
1650 err=sock->ops->getsockopt(sock, level, optname, optval, optlen);
1651 sockfd_put(sock);
1652 }
1653 return err;
1654}
1655
1656
1657/*
1658 * Shutdown a socket.
1659 */
1660
1661asmlinkage long sys_shutdown(int fd, int how)
1662{
1663 int err;
1664 struct socket *sock;
1665
1666 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1667 {
1668 err = security_socket_shutdown(sock, how);
1669 if (err) {
1670 sockfd_put(sock);
1671 return err;
1672 }
1673
1674 err=sock->ops->shutdown(sock, how);
1675 sockfd_put(sock);
1676 }
1677 return err;
1678}
1679
1680/* A couple of helpful macros for getting the address of the 32/64 bit
1681 * fields which are the same type (int / unsigned) on our platforms.
1682 */
1683#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1684#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1685#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1686
1687
1688/*
1689 * BSD sendmsg interface
1690 */
1691
1692asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1693{
1694 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1695 struct socket *sock;
1696 char address[MAX_SOCK_ADDR];
1697 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1698 unsigned char ctl[sizeof(struct cmsghdr) + 20]; /* 20 is size of ipv6_pktinfo */
1699 unsigned char *ctl_buf = ctl;
1700 struct msghdr msg_sys;
1701 int err, ctl_len, iov_size, total_len;
1702
1703 err = -EFAULT;
1704 if (MSG_CMSG_COMPAT & flags) {
1705 if (get_compat_msghdr(&msg_sys, msg_compat))
1706 return -EFAULT;
1707 } else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1708 return -EFAULT;
1709
1710 sock = sockfd_lookup(fd, &err);
1711 if (!sock)
1712 goto out;
1713
1714 /* do not move before msg_sys is valid */
1715 err = -EMSGSIZE;
1716 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1717 goto out_put;
1718
1719 /* Check whether to allocate the iovec area*/
1720 err = -ENOMEM;
1721 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1722 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1723 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1724 if (!iov)
1725 goto out_put;
1726 }
1727
1728 /* This will also move the address data into kernel space */
1729 if (MSG_CMSG_COMPAT & flags) {
1730 err = verify_compat_iovec(&msg_sys, iov, address, VERIFY_READ);
1731 } else
1732 err = verify_iovec(&msg_sys, iov, address, VERIFY_READ);
1733 if (err < 0)
1734 goto out_freeiov;
1735 total_len = err;
1736
1737 err = -ENOBUFS;
1738
1739 if (msg_sys.msg_controllen > INT_MAX)
1740 goto out_freeiov;
1741 ctl_len = msg_sys.msg_controllen;
1742 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1743 err = cmsghdr_from_user_compat_to_kern(&msg_sys, ctl, sizeof(ctl));
1744 if (err)
1745 goto out_freeiov;
1746 ctl_buf = msg_sys.msg_control;
1747 } else if (ctl_len) {
1748 if (ctl_len > sizeof(ctl))
1749 {
1750 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1751 if (ctl_buf == NULL)
1752 goto out_freeiov;
1753 }
1754 err = -EFAULT;
1755 /*
1756 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1757 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1758 * checking falls down on this.
1759 */
1760 if (copy_from_user(ctl_buf, (void __user *) msg_sys.msg_control, ctl_len))
1761 goto out_freectl;
1762 msg_sys.msg_control = ctl_buf;
1763 }
1764 msg_sys.msg_flags = flags;
1765
1766 if (sock->file->f_flags & O_NONBLOCK)
1767 msg_sys.msg_flags |= MSG_DONTWAIT;
1768 err = sock_sendmsg(sock, &msg_sys, total_len);
1769
1770out_freectl:
1771 if (ctl_buf != ctl)
1772 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1773out_freeiov:
1774 if (iov != iovstack)
1775 sock_kfree_s(sock->sk, iov, iov_size);
1776out_put:
1777 sockfd_put(sock);
1778out:
1779 return err;
1780}
1781
1782/*
1783 * BSD recvmsg interface
1784 */
1785
1786asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg, unsigned int flags)
1787{
1788 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1789 struct socket *sock;
1790 struct iovec iovstack[UIO_FASTIOV];
1791 struct iovec *iov=iovstack;
1792 struct msghdr msg_sys;
1793 unsigned long cmsg_ptr;
1794 int err, iov_size, total_len, len;
1795
1796 /* kernel mode address */
1797 char addr[MAX_SOCK_ADDR];
1798
1799 /* user mode address pointers */
1800 struct sockaddr __user *uaddr;
1801 int __user *uaddr_len;
1802
1803 if (MSG_CMSG_COMPAT & flags) {
1804 if (get_compat_msghdr(&msg_sys, msg_compat))
1805 return -EFAULT;
1806 } else
1807 if (copy_from_user(&msg_sys,msg,sizeof(struct msghdr)))
1808 return -EFAULT;
1809
1810 sock = sockfd_lookup(fd, &err);
1811 if (!sock)
1812 goto out;
1813
1814 err = -EMSGSIZE;
1815 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1816 goto out_put;
1817
1818 /* Check whether to allocate the iovec area*/
1819 err = -ENOMEM;
1820 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1821 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1822 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1823 if (!iov)
1824 goto out_put;
1825 }
1826
1827 /*
1828 * Save the user-mode address (verify_iovec will change the
1829 * kernel msghdr to use the kernel address space)
1830 */
1831
1832 uaddr = (void __user *) msg_sys.msg_name;
1833 uaddr_len = COMPAT_NAMELEN(msg);
1834 if (MSG_CMSG_COMPAT & flags) {
1835 err = verify_compat_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1836 } else
1837 err = verify_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1838 if (err < 0)
1839 goto out_freeiov;
1840 total_len=err;
1841
1842 cmsg_ptr = (unsigned long)msg_sys.msg_control;
1843 msg_sys.msg_flags = 0;
1844 if (MSG_CMSG_COMPAT & flags)
1845 msg_sys.msg_flags = MSG_CMSG_COMPAT;
1846
1847 if (sock->file->f_flags & O_NONBLOCK)
1848 flags |= MSG_DONTWAIT;
1849 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1850 if (err < 0)
1851 goto out_freeiov;
1852 len = err;
1853
1854 if (uaddr != NULL) {
1855 err = move_addr_to_user(addr, msg_sys.msg_namelen, uaddr, uaddr_len);
1856 if (err < 0)
1857 goto out_freeiov;
1858 }
1859 err = __put_user(msg_sys.msg_flags, COMPAT_FLAGS(msg));
1860 if (err)
1861 goto out_freeiov;
1862 if (MSG_CMSG_COMPAT & flags)
1863 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1864 &msg_compat->msg_controllen);
1865 else
1866 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1867 &msg->msg_controllen);
1868 if (err)
1869 goto out_freeiov;
1870 err = len;
1871
1872out_freeiov:
1873 if (iov != iovstack)
1874 sock_kfree_s(sock->sk, iov, iov_size);
1875out_put:
1876 sockfd_put(sock);
1877out:
1878 return err;
1879}
1880
1881#ifdef __ARCH_WANT_SYS_SOCKETCALL
1882
1883/* Argument list sizes for sys_socketcall */
1884#define AL(x) ((x) * sizeof(unsigned long))
1885static unsigned char nargs[18]={AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
1886 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
1887 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3)};
1888#undef AL
1889
1890/*
1891 * System call vectors.
1892 *
1893 * Argument checking cleaned up. Saved 20% in size.
1894 * This function doesn't need to set the kernel lock because
1895 * it is set by the callees.
1896 */
1897
1898asmlinkage long sys_socketcall(int call, unsigned long __user *args)
1899{
1900 unsigned long a[6];
1901 unsigned long a0,a1;
1902 int err;
1903
1904 if(call<1||call>SYS_RECVMSG)
1905 return -EINVAL;
1906
1907 /* copy_from_user should be SMP safe. */
1908 if (copy_from_user(a, args, nargs[call]))
1909 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01001910
1911 err = audit_socketcall(nargs[call]/sizeof(unsigned long), args);
1912 if (err)
1913 return err;
1914
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 a0=a[0];
1916 a1=a[1];
1917
1918 switch(call)
1919 {
1920 case SYS_SOCKET:
1921 err = sys_socket(a0,a1,a[2]);
1922 break;
1923 case SYS_BIND:
1924 err = sys_bind(a0,(struct sockaddr __user *)a1, a[2]);
1925 break;
1926 case SYS_CONNECT:
1927 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
1928 break;
1929 case SYS_LISTEN:
1930 err = sys_listen(a0,a1);
1931 break;
1932 case SYS_ACCEPT:
1933 err = sys_accept(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1934 break;
1935 case SYS_GETSOCKNAME:
1936 err = sys_getsockname(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1937 break;
1938 case SYS_GETPEERNAME:
1939 err = sys_getpeername(a0, (struct sockaddr __user *)a1, (int __user *)a[2]);
1940 break;
1941 case SYS_SOCKETPAIR:
1942 err = sys_socketpair(a0,a1, a[2], (int __user *)a[3]);
1943 break;
1944 case SYS_SEND:
1945 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
1946 break;
1947 case SYS_SENDTO:
1948 err = sys_sendto(a0,(void __user *)a1, a[2], a[3],
1949 (struct sockaddr __user *)a[4], a[5]);
1950 break;
1951 case SYS_RECV:
1952 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
1953 break;
1954 case SYS_RECVFROM:
1955 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
1956 (struct sockaddr __user *)a[4], (int __user *)a[5]);
1957 break;
1958 case SYS_SHUTDOWN:
1959 err = sys_shutdown(a0,a1);
1960 break;
1961 case SYS_SETSOCKOPT:
1962 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
1963 break;
1964 case SYS_GETSOCKOPT:
1965 err = sys_getsockopt(a0, a1, a[2], (char __user *)a[3], (int __user *)a[4]);
1966 break;
1967 case SYS_SENDMSG:
1968 err = sys_sendmsg(a0, (struct msghdr __user *) a1, a[2]);
1969 break;
1970 case SYS_RECVMSG:
1971 err = sys_recvmsg(a0, (struct msghdr __user *) a1, a[2]);
1972 break;
1973 default:
1974 err = -EINVAL;
1975 break;
1976 }
1977 return err;
1978}
1979
1980#endif /* __ARCH_WANT_SYS_SOCKETCALL */
1981
1982/*
1983 * This function is called by a protocol handler that wants to
1984 * advertise its address family, and have it linked into the
1985 * SOCKET module.
1986 */
1987
1988int sock_register(struct net_proto_family *ops)
1989{
1990 int err;
1991
1992 if (ops->family >= NPROTO) {
1993 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
1994 return -ENOBUFS;
1995 }
1996 net_family_write_lock();
1997 err = -EEXIST;
1998 if (net_families[ops->family] == NULL) {
1999 net_families[ops->family]=ops;
2000 err = 0;
2001 }
2002 net_family_write_unlock();
2003 printk(KERN_INFO "NET: Registered protocol family %d\n",
2004 ops->family);
2005 return err;
2006}
2007
2008/*
2009 * This function is called by a protocol handler that wants to
2010 * remove its address family, and have it unlinked from the
2011 * SOCKET module.
2012 */
2013
2014int sock_unregister(int family)
2015{
2016 if (family < 0 || family >= NPROTO)
2017 return -1;
2018
2019 net_family_write_lock();
2020 net_families[family]=NULL;
2021 net_family_write_unlock();
2022 printk(KERN_INFO "NET: Unregistered protocol family %d\n",
2023 family);
2024 return 0;
2025}
2026
2027
2028extern void sk_init(void);
2029
2030void __init sock_init(void)
2031{
2032 /*
2033 * Initialize sock SLAB cache.
2034 */
2035
2036 sk_init();
2037
2038#ifdef SLAB_SKB
2039 /*
2040 * Initialize skbuff SLAB cache
2041 */
2042 skb_init();
2043#endif
2044
2045 /*
2046 * Initialize the protocols module.
2047 */
2048
2049 init_inodecache();
2050 register_filesystem(&sock_fs_type);
2051 sock_mnt = kern_mount(&sock_fs_type);
2052 /* The real protocol initialization is performed when
2053 * do_initcalls is run.
2054 */
2055
2056#ifdef CONFIG_NETFILTER
2057 netfilter_init();
2058#endif
2059}
2060
2061#ifdef CONFIG_PROC_FS
2062void socket_seq_show(struct seq_file *seq)
2063{
2064 int cpu;
2065 int counter = 0;
2066
2067 for (cpu = 0; cpu < NR_CPUS; cpu++)
2068 counter += per_cpu(sockets_in_use, cpu);
2069
2070 /* It can be negative, by the way. 8) */
2071 if (counter < 0)
2072 counter = 0;
2073
2074 seq_printf(seq, "sockets: used %d\n", counter);
2075}
2076#endif /* CONFIG_PROC_FS */
2077
2078/* ABI emulation layers need these two */
2079EXPORT_SYMBOL(move_addr_to_kernel);
2080EXPORT_SYMBOL(move_addr_to_user);
2081EXPORT_SYMBOL(sock_create);
2082EXPORT_SYMBOL(sock_create_kern);
2083EXPORT_SYMBOL(sock_create_lite);
2084EXPORT_SYMBOL(sock_map_fd);
2085EXPORT_SYMBOL(sock_recvmsg);
2086EXPORT_SYMBOL(sock_register);
2087EXPORT_SYMBOL(sock_release);
2088EXPORT_SYMBOL(sock_sendmsg);
2089EXPORT_SYMBOL(sock_unregister);
2090EXPORT_SYMBOL(sock_wake_async);
2091EXPORT_SYMBOL(sockfd_lookup);
2092EXPORT_SYMBOL(kernel_sendmsg);
2093EXPORT_SYMBOL(kernel_recvmsg);