blob: c699e93c33d72141332fdfa0c16afa87ed4a971e [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>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030073#include <linux/if_frad.h>
74#include <linux/if_vlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <linux/init.h>
76#include <linux/poll.h>
77#include <linux/cache.h>
78#include <linux/module.h>
79#include <linux/highmem.h>
80#include <linux/divert.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>
David Woodhouse3ec3b2f2005-05-17 12:08:48 +010086#include <linux/audit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88#ifdef CONFIG_NET_RADIO
89#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
90#endif /* CONFIG_NET_RADIO */
91
92#include <asm/uaccess.h>
93#include <asm/unistd.h>
94
95#include <net/compat.h>
96
97#include <net/sock.h>
98#include <linux/netfilter.h>
99
100static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
101static ssize_t sock_aio_read(struct kiocb *iocb, char __user *buf,
102 size_t size, loff_t pos);
103static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *buf,
104 size_t size, loff_t pos);
105static int sock_mmap(struct file *file, struct vm_area_struct * vma);
106
107static int sock_close(struct inode *inode, struct file *file);
108static unsigned int sock_poll(struct file *file,
109 struct poll_table_struct *wait);
110static long sock_ioctl(struct file *file,
111 unsigned int cmd, unsigned long arg);
112static int sock_fasync(int fd, struct file *filp, int on);
113static ssize_t sock_readv(struct file *file, const struct iovec *vector,
114 unsigned long count, loff_t *ppos);
115static ssize_t sock_writev(struct file *file, const struct iovec *vector,
116 unsigned long count, loff_t *ppos);
117static ssize_t sock_sendpage(struct file *file, struct page *page,
118 int offset, size_t size, loff_t *ppos, int more);
119
120
121/*
122 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
123 * in the operation structures but are done directly via the socketcall() multiplexor.
124 */
125
126static struct file_operations socket_file_ops = {
127 .owner = THIS_MODULE,
128 .llseek = no_llseek,
129 .aio_read = sock_aio_read,
130 .aio_write = sock_aio_write,
131 .poll = sock_poll,
132 .unlocked_ioctl = sock_ioctl,
133 .mmap = sock_mmap,
134 .open = sock_no_open, /* special open code to disallow open via /proc */
135 .release = sock_close,
136 .fasync = sock_fasync,
137 .readv = sock_readv,
138 .writev = sock_writev,
139 .sendpage = sock_sendpage
140};
141
142/*
143 * The protocol list. Each protocol is registered in here.
144 */
145
146static struct net_proto_family *net_families[NPROTO];
147
148#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
149static atomic_t net_family_lockct = ATOMIC_INIT(0);
150static DEFINE_SPINLOCK(net_family_lock);
151
152/* The strategy is: modifications net_family vector are short, do not
153 sleep and veeery rare, but read access should be free of any exclusive
154 locks.
155 */
156
157static void net_family_write_lock(void)
158{
159 spin_lock(&net_family_lock);
160 while (atomic_read(&net_family_lockct) != 0) {
161 spin_unlock(&net_family_lock);
162
163 yield();
164
165 spin_lock(&net_family_lock);
166 }
167}
168
169static __inline__ void net_family_write_unlock(void)
170{
171 spin_unlock(&net_family_lock);
172}
173
174static __inline__ void net_family_read_lock(void)
175{
176 atomic_inc(&net_family_lockct);
177 spin_unlock_wait(&net_family_lock);
178}
179
180static __inline__ void net_family_read_unlock(void)
181{
182 atomic_dec(&net_family_lockct);
183}
184
185#else
186#define net_family_write_lock() do { } while(0)
187#define net_family_write_unlock() do { } while(0)
188#define net_family_read_lock() do { } while(0)
189#define net_family_read_unlock() do { } while(0)
190#endif
191
192
193/*
194 * Statistics counters of the socket lists
195 */
196
197static DEFINE_PER_CPU(int, sockets_in_use) = 0;
198
199/*
200 * Support routines. Move socket addresses back and forth across the kernel/user
201 * divide and look after the messy bits.
202 */
203
204#define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
205 16 for IP, 16 for IPX,
206 24 for IPv6,
207 about 80 for AX.25
208 must be at least one bigger than
209 the AF_UNIX size (see net/unix/af_unix.c
210 :unix_mkname()).
211 */
212
213/**
214 * move_addr_to_kernel - copy a socket address into kernel space
215 * @uaddr: Address in user space
216 * @kaddr: Address in kernel space
217 * @ulen: Length in user space
218 *
219 * The address is copied into kernel space. If the provided address is
220 * too long an error code of -EINVAL is returned. If the copy gives
221 * invalid addresses -EFAULT is returned. On a success 0 is returned.
222 */
223
224int move_addr_to_kernel(void __user *uaddr, int ulen, void *kaddr)
225{
226 if(ulen<0||ulen>MAX_SOCK_ADDR)
227 return -EINVAL;
228 if(ulen==0)
229 return 0;
230 if(copy_from_user(kaddr,uaddr,ulen))
231 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +0100232 return audit_sockaddr(ulen, kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235/**
236 * move_addr_to_user - copy an address to user space
237 * @kaddr: kernel space address
238 * @klen: length of address in kernel
239 * @uaddr: user space address
240 * @ulen: pointer to user length field
241 *
242 * The value pointed to by ulen on entry is the buffer length available.
243 * This is overwritten with the buffer space used. -EINVAL is returned
244 * if an overlong buffer is specified or a negative buffer size. -EFAULT
245 * is returned if either the buffer or the length field are not
246 * accessible.
247 * After copying the data up to the limit the user specifies, the true
248 * length of the data is written over the length limit the user
249 * specified. Zero is returned for a success.
250 */
251
252int move_addr_to_user(void *kaddr, int klen, void __user *uaddr, int __user *ulen)
253{
254 int err;
255 int len;
256
257 if((err=get_user(len, ulen)))
258 return err;
259 if(len>klen)
260 len=klen;
261 if(len<0 || len> MAX_SOCK_ADDR)
262 return -EINVAL;
263 if(len)
264 {
265 if(copy_to_user(uaddr,kaddr,len))
266 return -EFAULT;
267 }
268 /*
269 * "fromlen shall refer to the value before truncation.."
270 * 1003.1g
271 */
272 return __put_user(klen, ulen);
273}
274
275#define SOCKFS_MAGIC 0x534F434B
276
Eric Dumazetba899662005-08-26 12:05:31 -0700277static kmem_cache_t * sock_inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279static struct inode *sock_alloc_inode(struct super_block *sb)
280{
281 struct socket_alloc *ei;
282 ei = (struct socket_alloc *)kmem_cache_alloc(sock_inode_cachep, SLAB_KERNEL);
283 if (!ei)
284 return NULL;
285 init_waitqueue_head(&ei->socket.wait);
286
287 ei->socket.fasync_list = NULL;
288 ei->socket.state = SS_UNCONNECTED;
289 ei->socket.flags = 0;
290 ei->socket.ops = NULL;
291 ei->socket.sk = NULL;
292 ei->socket.file = NULL;
293 ei->socket.flags = 0;
294
295 return &ei->vfs_inode;
296}
297
298static void sock_destroy_inode(struct inode *inode)
299{
300 kmem_cache_free(sock_inode_cachep,
301 container_of(inode, struct socket_alloc, vfs_inode));
302}
303
304static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
305{
306 struct socket_alloc *ei = (struct socket_alloc *) foo;
307
308 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
309 SLAB_CTOR_CONSTRUCTOR)
310 inode_init_once(&ei->vfs_inode);
311}
312
313static int init_inodecache(void)
314{
315 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
316 sizeof(struct socket_alloc),
317 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
318 init_once, NULL);
319 if (sock_inode_cachep == NULL)
320 return -ENOMEM;
321 return 0;
322}
323
324static struct super_operations sockfs_ops = {
325 .alloc_inode = sock_alloc_inode,
326 .destroy_inode =sock_destroy_inode,
327 .statfs = simple_statfs,
328};
329
330static struct super_block *sockfs_get_sb(struct file_system_type *fs_type,
331 int flags, const char *dev_name, void *data)
332{
333 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC);
334}
335
Eric Dumazetba899662005-08-26 12:05:31 -0700336static struct vfsmount *sock_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338static struct file_system_type sock_fs_type = {
339 .name = "sockfs",
340 .get_sb = sockfs_get_sb,
341 .kill_sb = kill_anon_super,
342};
343static int sockfs_delete_dentry(struct dentry *dentry)
344{
345 return 1;
346}
347static struct dentry_operations sockfs_dentry_operations = {
348 .d_delete = sockfs_delete_dentry,
349};
350
351/*
352 * Obtains the first available file descriptor and sets it up for use.
353 *
354 * This function creates file structure and maps it to fd space
355 * of current process. On success it returns file descriptor
356 * and file struct implicitly stored in sock->file.
357 * Note that another thread may close file descriptor before we return
358 * from this function. We use the fact that now we do not refer
359 * to socket after mapping. If one day we will need it, this
360 * function will increment ref. count on file by 1.
361 *
362 * In any case returned fd MAY BE not valid!
363 * This race condition is unavoidable
364 * with shared fd spaces, we cannot solve it inside kernel,
365 * but we take care of internal coherence yet.
366 */
367
368int sock_map_fd(struct socket *sock)
369{
370 int fd;
371 struct qstr this;
372 char name[32];
373
374 /*
375 * Find a file descriptor suitable for return to the user.
376 */
377
378 fd = get_unused_fd();
379 if (fd >= 0) {
380 struct file *file = get_empty_filp();
381
382 if (!file) {
383 put_unused_fd(fd);
384 fd = -ENFILE;
385 goto out;
386 }
387
Eric Dumazetf31f5f02005-06-22 14:32:51 -0700388 this.len = sprintf(name, "[%lu]", SOCK_INODE(sock)->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 this.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 this.hash = SOCK_INODE(sock)->i_ino;
391
392 file->f_dentry = d_alloc(sock_mnt->mnt_sb->s_root, &this);
393 if (!file->f_dentry) {
394 put_filp(file);
395 put_unused_fd(fd);
396 fd = -ENOMEM;
397 goto out;
398 }
399 file->f_dentry->d_op = &sockfs_dentry_operations;
400 d_add(file->f_dentry, SOCK_INODE(sock));
401 file->f_vfsmnt = mntget(sock_mnt);
402 file->f_mapping = file->f_dentry->d_inode->i_mapping;
403
404 sock->file = file;
405 file->f_op = SOCK_INODE(sock)->i_fop = &socket_file_ops;
406 file->f_mode = FMODE_READ | FMODE_WRITE;
407 file->f_flags = O_RDWR;
408 file->f_pos = 0;
Benjamin LaHaise07dc3f02005-08-10 14:16:04 -0700409 file->private_data = sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 fd_install(fd, file);
411 }
412
413out:
414 return fd;
415}
416
417/**
418 * sockfd_lookup - Go from a file number to its socket slot
419 * @fd: file handle
420 * @err: pointer to an error code return
421 *
422 * The file handle passed in is locked and the socket it is bound
423 * too is returned. If an error occurs the err pointer is overwritten
424 * with a negative errno code and NULL is returned. The function checks
425 * for both invalid handles and passing a handle which is not a socket.
426 *
427 * On a success the socket object pointer is returned.
428 */
429
430struct socket *sockfd_lookup(int fd, int *err)
431{
432 struct file *file;
433 struct inode *inode;
434 struct socket *sock;
435
436 if (!(file = fget(fd)))
437 {
438 *err = -EBADF;
439 return NULL;
440 }
441
Benjamin LaHaise07dc3f02005-08-10 14:16:04 -0700442 if (file->f_op == &socket_file_ops)
443 return file->private_data; /* set in sock_map_fd */
444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 inode = file->f_dentry->d_inode;
446 if (!S_ISSOCK(inode->i_mode)) {
447 *err = -ENOTSOCK;
448 fput(file);
449 return NULL;
450 }
451
452 sock = SOCKET_I(inode);
453 if (sock->file != file) {
454 printk(KERN_ERR "socki_lookup: socket file changed!\n");
455 sock->file = file;
456 }
457 return sock;
458}
459
460/**
461 * sock_alloc - allocate a socket
462 *
463 * Allocate a new inode and socket object. The two are bound together
464 * and initialised. The socket is then returned. If we are out of inodes
465 * NULL is returned.
466 */
467
468static struct socket *sock_alloc(void)
469{
470 struct inode * inode;
471 struct socket * sock;
472
473 inode = new_inode(sock_mnt->mnt_sb);
474 if (!inode)
475 return NULL;
476
477 sock = SOCKET_I(inode);
478
479 inode->i_mode = S_IFSOCK|S_IRWXUGO;
480 inode->i_uid = current->fsuid;
481 inode->i_gid = current->fsgid;
482
483 get_cpu_var(sockets_in_use)++;
484 put_cpu_var(sockets_in_use);
485 return sock;
486}
487
488/*
489 * In theory you can't get an open on this inode, but /proc provides
490 * a back door. Remember to keep it shut otherwise you'll let the
491 * creepy crawlies in.
492 */
493
494static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
495{
496 return -ENXIO;
497}
498
499struct file_operations bad_sock_fops = {
500 .owner = THIS_MODULE,
501 .open = sock_no_open,
502};
503
504/**
505 * sock_release - close a socket
506 * @sock: socket to close
507 *
508 * The socket is released from the protocol stack if it has a release
509 * callback, and the inode is then released if the socket is bound to
510 * an inode not a file.
511 */
512
513void sock_release(struct socket *sock)
514{
515 if (sock->ops) {
516 struct module *owner = sock->ops->owner;
517
518 sock->ops->release(sock);
519 sock->ops = NULL;
520 module_put(owner);
521 }
522
523 if (sock->fasync_list)
524 printk(KERN_ERR "sock_release: fasync list not empty!\n");
525
526 get_cpu_var(sockets_in_use)--;
527 put_cpu_var(sockets_in_use);
528 if (!sock->file) {
529 iput(SOCK_INODE(sock));
530 return;
531 }
532 sock->file=NULL;
533}
534
535static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
536 struct msghdr *msg, size_t size)
537{
538 struct sock_iocb *si = kiocb_to_siocb(iocb);
539 int err;
540
541 si->sock = sock;
542 si->scm = NULL;
543 si->msg = msg;
544 si->size = size;
545
546 err = security_socket_sendmsg(sock, msg, size);
547 if (err)
548 return err;
549
550 return sock->ops->sendmsg(iocb, sock, msg, size);
551}
552
553int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
554{
555 struct kiocb iocb;
556 struct sock_iocb siocb;
557 int ret;
558
559 init_sync_kiocb(&iocb, NULL);
560 iocb.private = &siocb;
561 ret = __sock_sendmsg(&iocb, sock, msg, size);
562 if (-EIOCBQUEUED == ret)
563 ret = wait_on_sync_kiocb(&iocb);
564 return ret;
565}
566
567int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
568 struct kvec *vec, size_t num, size_t size)
569{
570 mm_segment_t oldfs = get_fs();
571 int result;
572
573 set_fs(KERNEL_DS);
574 /*
575 * the following is safe, since for compiler definitions of kvec and
576 * iovec are identical, yielding the same in-core layout and alignment
577 */
578 msg->msg_iov = (struct iovec *)vec,
579 msg->msg_iovlen = num;
580 result = sock_sendmsg(sock, msg, size);
581 set_fs(oldfs);
582 return result;
583}
584
585static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
586 struct msghdr *msg, size_t size, int flags)
587{
588 int err;
589 struct sock_iocb *si = kiocb_to_siocb(iocb);
590
591 si->sock = sock;
592 si->scm = NULL;
593 si->msg = msg;
594 si->size = size;
595 si->flags = flags;
596
597 err = security_socket_recvmsg(sock, msg, size, flags);
598 if (err)
599 return err;
600
601 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
602}
603
604int sock_recvmsg(struct socket *sock, struct msghdr *msg,
605 size_t size, int flags)
606{
607 struct kiocb iocb;
608 struct sock_iocb siocb;
609 int ret;
610
611 init_sync_kiocb(&iocb, NULL);
612 iocb.private = &siocb;
613 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
614 if (-EIOCBQUEUED == ret)
615 ret = wait_on_sync_kiocb(&iocb);
616 return ret;
617}
618
619int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
620 struct kvec *vec, size_t num,
621 size_t size, int flags)
622{
623 mm_segment_t oldfs = get_fs();
624 int result;
625
626 set_fs(KERNEL_DS);
627 /*
628 * the following is safe, since for compiler definitions of kvec and
629 * iovec are identical, yielding the same in-core layout and alignment
630 */
631 msg->msg_iov = (struct iovec *)vec,
632 msg->msg_iovlen = num;
633 result = sock_recvmsg(sock, msg, size, flags);
634 set_fs(oldfs);
635 return result;
636}
637
638static void sock_aio_dtor(struct kiocb *iocb)
639{
640 kfree(iocb->private);
641}
642
643/*
644 * Read data from a socket. ubuf is a user mode pointer. We make sure the user
645 * area ubuf...ubuf+size-1 is writable before asking the protocol.
646 */
647
648static ssize_t sock_aio_read(struct kiocb *iocb, char __user *ubuf,
649 size_t size, loff_t pos)
650{
651 struct sock_iocb *x, siocb;
652 struct socket *sock;
653 int flags;
654
655 if (pos != 0)
656 return -ESPIPE;
657 if (size==0) /* Match SYS5 behaviour */
658 return 0;
659
660 if (is_sync_kiocb(iocb))
661 x = &siocb;
662 else {
663 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
664 if (!x)
665 return -ENOMEM;
666 iocb->ki_dtor = sock_aio_dtor;
667 }
668 iocb->private = x;
669 x->kiocb = iocb;
Eric Dumazetb69aee02005-09-06 14:42:45 -0700670 sock = iocb->ki_filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
672 x->async_msg.msg_name = NULL;
673 x->async_msg.msg_namelen = 0;
674 x->async_msg.msg_iov = &x->async_iov;
675 x->async_msg.msg_iovlen = 1;
676 x->async_msg.msg_control = NULL;
677 x->async_msg.msg_controllen = 0;
678 x->async_iov.iov_base = ubuf;
679 x->async_iov.iov_len = size;
680 flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
681
682 return __sock_recvmsg(iocb, sock, &x->async_msg, size, flags);
683}
684
685
686/*
687 * Write data to a socket. We verify that the user area ubuf..ubuf+size-1
688 * is readable by the user process.
689 */
690
691static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *ubuf,
692 size_t size, loff_t pos)
693{
694 struct sock_iocb *x, siocb;
695 struct socket *sock;
696
697 if (pos != 0)
698 return -ESPIPE;
699 if(size==0) /* Match SYS5 behaviour */
700 return 0;
701
702 if (is_sync_kiocb(iocb))
703 x = &siocb;
704 else {
705 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
706 if (!x)
707 return -ENOMEM;
708 iocb->ki_dtor = sock_aio_dtor;
709 }
710 iocb->private = x;
711 x->kiocb = iocb;
Eric Dumazetb69aee02005-09-06 14:42:45 -0700712 sock = iocb->ki_filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 x->async_msg.msg_name = NULL;
715 x->async_msg.msg_namelen = 0;
716 x->async_msg.msg_iov = &x->async_iov;
717 x->async_msg.msg_iovlen = 1;
718 x->async_msg.msg_control = NULL;
719 x->async_msg.msg_controllen = 0;
720 x->async_msg.msg_flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
721 if (sock->type == SOCK_SEQPACKET)
722 x->async_msg.msg_flags |= MSG_EOR;
723 x->async_iov.iov_base = (void __user *)ubuf;
724 x->async_iov.iov_len = size;
725
726 return __sock_sendmsg(iocb, sock, &x->async_msg, size);
727}
728
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300729static ssize_t sock_sendpage(struct file *file, struct page *page,
730 int offset, size_t size, loff_t *ppos, int more)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 struct socket *sock;
733 int flags;
734
Eric Dumazetb69aee02005-09-06 14:42:45 -0700735 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
738 if (more)
739 flags |= MSG_MORE;
740
741 return sock->ops->sendpage(sock, page, offset, size, flags);
742}
743
Eric Dumazetb69aee02005-09-06 14:42:45 -0700744static int sock_readv_writev(int type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 struct file * file, const struct iovec * iov,
746 long count, size_t size)
747{
748 struct msghdr msg;
749 struct socket *sock;
750
Eric Dumazetb69aee02005-09-06 14:42:45 -0700751 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 msg.msg_name = NULL;
754 msg.msg_namelen = 0;
755 msg.msg_control = NULL;
756 msg.msg_controllen = 0;
757 msg.msg_iov = (struct iovec *) iov;
758 msg.msg_iovlen = count;
759 msg.msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
760
761 /* read() does a VERIFY_WRITE */
762 if (type == VERIFY_WRITE)
763 return sock_recvmsg(sock, &msg, size, msg.msg_flags);
764
765 if (sock->type == SOCK_SEQPACKET)
766 msg.msg_flags |= MSG_EOR;
767
768 return sock_sendmsg(sock, &msg, size);
769}
770
771static ssize_t sock_readv(struct file *file, const struct iovec *vector,
772 unsigned long count, loff_t *ppos)
773{
774 size_t tot_len = 0;
775 int i;
776 for (i = 0 ; i < count ; i++)
777 tot_len += vector[i].iov_len;
Eric Dumazetb69aee02005-09-06 14:42:45 -0700778 return sock_readv_writev(VERIFY_WRITE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 file, vector, count, tot_len);
780}
781
782static ssize_t sock_writev(struct file *file, const struct iovec *vector,
783 unsigned long count, loff_t *ppos)
784{
785 size_t tot_len = 0;
786 int i;
787 for (i = 0 ; i < count ; i++)
788 tot_len += vector[i].iov_len;
Eric Dumazetb69aee02005-09-06 14:42:45 -0700789 return sock_readv_writev(VERIFY_READ,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 file, vector, count, tot_len);
791}
792
793
794/*
795 * Atomic setting of ioctl hooks to avoid race
796 * with module unload.
797 */
798
799static DECLARE_MUTEX(br_ioctl_mutex);
800static int (*br_ioctl_hook)(unsigned int cmd, void __user *arg) = NULL;
801
802void brioctl_set(int (*hook)(unsigned int, void __user *))
803{
804 down(&br_ioctl_mutex);
805 br_ioctl_hook = hook;
806 up(&br_ioctl_mutex);
807}
808EXPORT_SYMBOL(brioctl_set);
809
810static DECLARE_MUTEX(vlan_ioctl_mutex);
811static int (*vlan_ioctl_hook)(void __user *arg);
812
813void vlan_ioctl_set(int (*hook)(void __user *))
814{
815 down(&vlan_ioctl_mutex);
816 vlan_ioctl_hook = hook;
817 up(&vlan_ioctl_mutex);
818}
819EXPORT_SYMBOL(vlan_ioctl_set);
820
821static DECLARE_MUTEX(dlci_ioctl_mutex);
822static int (*dlci_ioctl_hook)(unsigned int, void __user *);
823
824void dlci_ioctl_set(int (*hook)(unsigned int, void __user *))
825{
826 down(&dlci_ioctl_mutex);
827 dlci_ioctl_hook = hook;
828 up(&dlci_ioctl_mutex);
829}
830EXPORT_SYMBOL(dlci_ioctl_set);
831
832/*
833 * With an ioctl, arg may well be a user mode pointer, but we don't know
834 * what to do with it - that's up to the protocol still.
835 */
836
837static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
838{
839 struct socket *sock;
840 void __user *argp = (void __user *)arg;
841 int pid, err;
842
Eric Dumazetb69aee02005-09-06 14:42:45 -0700843 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
845 err = dev_ioctl(cmd, argp);
846 } else
847#ifdef WIRELESS_EXT
848 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
849 err = dev_ioctl(cmd, argp);
850 } else
851#endif /* WIRELESS_EXT */
852 switch (cmd) {
853 case FIOSETOWN:
854 case SIOCSPGRP:
855 err = -EFAULT;
856 if (get_user(pid, (int __user *)argp))
857 break;
858 err = f_setown(sock->file, pid, 1);
859 break;
860 case FIOGETOWN:
861 case SIOCGPGRP:
862 err = put_user(sock->file->f_owner.pid, (int __user *)argp);
863 break;
864 case SIOCGIFBR:
865 case SIOCSIFBR:
866 case SIOCBRADDBR:
867 case SIOCBRDELBR:
868 err = -ENOPKG;
869 if (!br_ioctl_hook)
870 request_module("bridge");
871
872 down(&br_ioctl_mutex);
873 if (br_ioctl_hook)
874 err = br_ioctl_hook(cmd, argp);
875 up(&br_ioctl_mutex);
876 break;
877 case SIOCGIFVLAN:
878 case SIOCSIFVLAN:
879 err = -ENOPKG;
880 if (!vlan_ioctl_hook)
881 request_module("8021q");
882
883 down(&vlan_ioctl_mutex);
884 if (vlan_ioctl_hook)
885 err = vlan_ioctl_hook(argp);
886 up(&vlan_ioctl_mutex);
887 break;
888 case SIOCGIFDIVERT:
889 case SIOCSIFDIVERT:
890 /* Convert this to call through a hook */
891 err = divert_ioctl(cmd, argp);
892 break;
893 case SIOCADDDLCI:
894 case SIOCDELDLCI:
895 err = -ENOPKG;
896 if (!dlci_ioctl_hook)
897 request_module("dlci");
898
899 if (dlci_ioctl_hook) {
900 down(&dlci_ioctl_mutex);
901 err = dlci_ioctl_hook(cmd, argp);
902 up(&dlci_ioctl_mutex);
903 }
904 break;
905 default:
906 err = sock->ops->ioctl(sock, cmd, arg);
907 break;
908 }
909 return err;
910}
911
912int sock_create_lite(int family, int type, int protocol, struct socket **res)
913{
914 int err;
915 struct socket *sock = NULL;
916
917 err = security_socket_create(family, type, protocol, 1);
918 if (err)
919 goto out;
920
921 sock = sock_alloc();
922 if (!sock) {
923 err = -ENOMEM;
924 goto out;
925 }
926
927 security_socket_post_create(sock, family, type, protocol, 1);
928 sock->type = type;
929out:
930 *res = sock;
931 return err;
932}
933
934/* No kernel lock held - perfect */
935static unsigned int sock_poll(struct file *file, poll_table * wait)
936{
937 struct socket *sock;
938
939 /*
940 * We can't return errors to poll, so it's either yes or no.
941 */
Eric Dumazetb69aee02005-09-06 14:42:45 -0700942 sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 return sock->ops->poll(file, sock, wait);
944}
945
946static int sock_mmap(struct file * file, struct vm_area_struct * vma)
947{
Eric Dumazetb69aee02005-09-06 14:42:45 -0700948 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 return sock->ops->mmap(file, sock, vma);
951}
952
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300953static int sock_close(struct inode *inode, struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
955 /*
956 * It was possible the inode is NULL we were
957 * closing an unfinished socket.
958 */
959
960 if (!inode)
961 {
962 printk(KERN_DEBUG "sock_close: NULL inode\n");
963 return 0;
964 }
965 sock_fasync(-1, filp, 0);
966 sock_release(SOCKET_I(inode));
967 return 0;
968}
969
970/*
971 * Update the socket async list
972 *
973 * Fasync_list locking strategy.
974 *
975 * 1. fasync_list is modified only under process context socket lock
976 * i.e. under semaphore.
977 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
978 * or under socket lock.
979 * 3. fasync_list can be used from softirq context, so that
980 * modification under socket lock have to be enhanced with
981 * write_lock_bh(&sk->sk_callback_lock).
982 * --ANK (990710)
983 */
984
985static int sock_fasync(int fd, struct file *filp, int on)
986{
987 struct fasync_struct *fa, *fna=NULL, **prev;
988 struct socket *sock;
989 struct sock *sk;
990
991 if (on)
992 {
993 fna=(struct fasync_struct *)kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
994 if(fna==NULL)
995 return -ENOMEM;
996 }
997
Eric Dumazetb69aee02005-09-06 14:42:45 -0700998 sock = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 if ((sk=sock->sk) == NULL) {
1001 kfree(fna);
1002 return -EINVAL;
1003 }
1004
1005 lock_sock(sk);
1006
1007 prev=&(sock->fasync_list);
1008
1009 for (fa=*prev; fa!=NULL; prev=&fa->fa_next,fa=*prev)
1010 if (fa->fa_file==filp)
1011 break;
1012
1013 if(on)
1014 {
1015 if(fa!=NULL)
1016 {
1017 write_lock_bh(&sk->sk_callback_lock);
1018 fa->fa_fd=fd;
1019 write_unlock_bh(&sk->sk_callback_lock);
1020
1021 kfree(fna);
1022 goto out;
1023 }
1024 fna->fa_file=filp;
1025 fna->fa_fd=fd;
1026 fna->magic=FASYNC_MAGIC;
1027 fna->fa_next=sock->fasync_list;
1028 write_lock_bh(&sk->sk_callback_lock);
1029 sock->fasync_list=fna;
1030 write_unlock_bh(&sk->sk_callback_lock);
1031 }
1032 else
1033 {
1034 if (fa!=NULL)
1035 {
1036 write_lock_bh(&sk->sk_callback_lock);
1037 *prev=fa->fa_next;
1038 write_unlock_bh(&sk->sk_callback_lock);
1039 kfree(fa);
1040 }
1041 }
1042
1043out:
1044 release_sock(sock->sk);
1045 return 0;
1046}
1047
1048/* This function may be called only under socket lock or callback_lock */
1049
1050int sock_wake_async(struct socket *sock, int how, int band)
1051{
1052 if (!sock || !sock->fasync_list)
1053 return -1;
1054 switch (how)
1055 {
1056 case 1:
1057
1058 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1059 break;
1060 goto call_kill;
1061 case 2:
1062 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1063 break;
1064 /* fall through */
1065 case 0:
1066 call_kill:
1067 __kill_fasync(sock->fasync_list, SIGIO, band);
1068 break;
1069 case 3:
1070 __kill_fasync(sock->fasync_list, SIGURG, band);
1071 }
1072 return 0;
1073}
1074
1075static int __sock_create(int family, int type, int protocol, struct socket **res, int kern)
1076{
1077 int err;
1078 struct socket *sock;
1079
1080 /*
1081 * Check protocol is in range
1082 */
1083 if (family < 0 || family >= NPROTO)
1084 return -EAFNOSUPPORT;
1085 if (type < 0 || type >= SOCK_MAX)
1086 return -EINVAL;
1087
1088 /* Compatibility.
1089
1090 This uglymoron is moved from INET layer to here to avoid
1091 deadlock in module load.
1092 */
1093 if (family == PF_INET && type == SOCK_PACKET) {
1094 static int warned;
1095 if (!warned) {
1096 warned = 1;
1097 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n", current->comm);
1098 }
1099 family = PF_PACKET;
1100 }
1101
1102 err = security_socket_create(family, type, protocol, kern);
1103 if (err)
1104 return err;
1105
1106#if defined(CONFIG_KMOD)
1107 /* Attempt to load a protocol module if the find failed.
1108 *
1109 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1110 * requested real, full-featured networking support upon configuration.
1111 * Otherwise module support will break!
1112 */
1113 if (net_families[family]==NULL)
1114 {
1115 request_module("net-pf-%d",family);
1116 }
1117#endif
1118
1119 net_family_read_lock();
1120 if (net_families[family] == NULL) {
1121 err = -EAFNOSUPPORT;
1122 goto out;
1123 }
1124
1125/*
1126 * Allocate the socket and allow the family to set things up. if
1127 * the protocol is 0, the family is instructed to select an appropriate
1128 * default.
1129 */
1130
1131 if (!(sock = sock_alloc())) {
1132 printk(KERN_WARNING "socket: no more sockets\n");
1133 err = -ENFILE; /* Not exactly a match, but its the
1134 closest posix thing */
1135 goto out;
1136 }
1137
1138 sock->type = type;
1139
1140 /*
1141 * We will call the ->create function, that possibly is in a loadable
1142 * module, so we have to bump that loadable module refcnt first.
1143 */
1144 err = -EAFNOSUPPORT;
1145 if (!try_module_get(net_families[family]->owner))
1146 goto out_release;
1147
1148 if ((err = net_families[family]->create(sock, protocol)) < 0)
1149 goto out_module_put;
1150 /*
1151 * Now to bump the refcnt of the [loadable] module that owns this
1152 * socket at sock_release time we decrement its refcnt.
1153 */
1154 if (!try_module_get(sock->ops->owner)) {
1155 sock->ops = NULL;
1156 goto out_module_put;
1157 }
1158 /*
1159 * Now that we're done with the ->create function, the [loadable]
1160 * module can have its refcnt decremented
1161 */
1162 module_put(net_families[family]->owner);
1163 *res = sock;
1164 security_socket_post_create(sock, family, type, protocol, kern);
1165
1166out:
1167 net_family_read_unlock();
1168 return err;
1169out_module_put:
1170 module_put(net_families[family]->owner);
1171out_release:
1172 sock_release(sock);
1173 goto out;
1174}
1175
1176int sock_create(int family, int type, int protocol, struct socket **res)
1177{
1178 return __sock_create(family, type, protocol, res, 0);
1179}
1180
1181int sock_create_kern(int family, int type, int protocol, struct socket **res)
1182{
1183 return __sock_create(family, type, protocol, res, 1);
1184}
1185
1186asmlinkage long sys_socket(int family, int type, int protocol)
1187{
1188 int retval;
1189 struct socket *sock;
1190
1191 retval = sock_create(family, type, protocol, &sock);
1192 if (retval < 0)
1193 goto out;
1194
1195 retval = sock_map_fd(sock);
1196 if (retval < 0)
1197 goto out_release;
1198
1199out:
1200 /* It may be already another descriptor 8) Not kernel problem. */
1201 return retval;
1202
1203out_release:
1204 sock_release(sock);
1205 return retval;
1206}
1207
1208/*
1209 * Create a pair of connected sockets.
1210 */
1211
1212asmlinkage long sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1213{
1214 struct socket *sock1, *sock2;
1215 int fd1, fd2, err;
1216
1217 /*
1218 * Obtain the first socket and check if the underlying protocol
1219 * supports the socketpair call.
1220 */
1221
1222 err = sock_create(family, type, protocol, &sock1);
1223 if (err < 0)
1224 goto out;
1225
1226 err = sock_create(family, type, protocol, &sock2);
1227 if (err < 0)
1228 goto out_release_1;
1229
1230 err = sock1->ops->socketpair(sock1, sock2);
1231 if (err < 0)
1232 goto out_release_both;
1233
1234 fd1 = fd2 = -1;
1235
1236 err = sock_map_fd(sock1);
1237 if (err < 0)
1238 goto out_release_both;
1239 fd1 = err;
1240
1241 err = sock_map_fd(sock2);
1242 if (err < 0)
1243 goto out_close_1;
1244 fd2 = err;
1245
1246 /* fd1 and fd2 may be already another descriptors.
1247 * Not kernel problem.
1248 */
1249
1250 err = put_user(fd1, &usockvec[0]);
1251 if (!err)
1252 err = put_user(fd2, &usockvec[1]);
1253 if (!err)
1254 return 0;
1255
1256 sys_close(fd2);
1257 sys_close(fd1);
1258 return err;
1259
1260out_close_1:
1261 sock_release(sock2);
1262 sys_close(fd1);
1263 return err;
1264
1265out_release_both:
1266 sock_release(sock2);
1267out_release_1:
1268 sock_release(sock1);
1269out:
1270 return err;
1271}
1272
1273
1274/*
1275 * Bind a name to a socket. Nothing much to do here since it's
1276 * the protocol's responsibility to handle the local address.
1277 *
1278 * We move the socket address to kernel space before we call
1279 * the protocol layer (having also checked the address is ok).
1280 */
1281
1282asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1283{
1284 struct socket *sock;
1285 char address[MAX_SOCK_ADDR];
1286 int err;
1287
1288 if((sock = sockfd_lookup(fd,&err))!=NULL)
1289 {
1290 if((err=move_addr_to_kernel(umyaddr,addrlen,address))>=0) {
1291 err = security_socket_bind(sock, (struct sockaddr *)address, addrlen);
1292 if (err) {
1293 sockfd_put(sock);
1294 return err;
1295 }
1296 err = sock->ops->bind(sock, (struct sockaddr *)address, addrlen);
1297 }
1298 sockfd_put(sock);
1299 }
1300 return err;
1301}
1302
1303
1304/*
1305 * Perform a listen. Basically, we allow the protocol to do anything
1306 * necessary for a listen, and if that works, we mark the socket as
1307 * ready for listening.
1308 */
1309
1310int sysctl_somaxconn = SOMAXCONN;
1311
1312asmlinkage long sys_listen(int fd, int backlog)
1313{
1314 struct socket *sock;
1315 int err;
1316
1317 if ((sock = sockfd_lookup(fd, &err)) != NULL) {
1318 if ((unsigned) backlog > sysctl_somaxconn)
1319 backlog = sysctl_somaxconn;
1320
1321 err = security_socket_listen(sock, backlog);
1322 if (err) {
1323 sockfd_put(sock);
1324 return err;
1325 }
1326
1327 err=sock->ops->listen(sock, backlog);
1328 sockfd_put(sock);
1329 }
1330 return err;
1331}
1332
1333
1334/*
1335 * For accept, we attempt to create a new socket, set up the link
1336 * with the client, wake up the client, then return the new
1337 * connected fd. We collect the address of the connector in kernel
1338 * space and move it to user at the very end. This is unclean because
1339 * we open the socket then return an error.
1340 *
1341 * 1003.1g adds the ability to recvmsg() to query connection pending
1342 * status to recvmsg. We need to add that support in a way thats
1343 * clean when we restucture accept also.
1344 */
1345
1346asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr, int __user *upeer_addrlen)
1347{
1348 struct socket *sock, *newsock;
1349 int err, len;
1350 char address[MAX_SOCK_ADDR];
1351
1352 sock = sockfd_lookup(fd, &err);
1353 if (!sock)
1354 goto out;
1355
1356 err = -ENFILE;
1357 if (!(newsock = sock_alloc()))
1358 goto out_put;
1359
1360 newsock->type = sock->type;
1361 newsock->ops = sock->ops;
1362
1363 err = security_socket_accept(sock, newsock);
1364 if (err)
1365 goto out_release;
1366
1367 /*
1368 * We don't need try_module_get here, as the listening socket (sock)
1369 * has the protocol module (sock->ops->owner) held.
1370 */
1371 __module_get(newsock->ops->owner);
1372
1373 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1374 if (err < 0)
1375 goto out_release;
1376
1377 if (upeer_sockaddr) {
1378 if(newsock->ops->getname(newsock, (struct sockaddr *)address, &len, 2)<0) {
1379 err = -ECONNABORTED;
1380 goto out_release;
1381 }
1382 err = move_addr_to_user(address, len, upeer_sockaddr, upeer_addrlen);
1383 if (err < 0)
1384 goto out_release;
1385 }
1386
1387 /* File flags are not inherited via accept() unlike another OSes. */
1388
1389 if ((err = sock_map_fd(newsock)) < 0)
1390 goto out_release;
1391
1392 security_socket_post_accept(sock, newsock);
1393
1394out_put:
1395 sockfd_put(sock);
1396out:
1397 return err;
1398out_release:
1399 sock_release(newsock);
1400 goto out_put;
1401}
1402
1403
1404/*
1405 * Attempt to connect to a socket with the server address. The address
1406 * is in user space so we verify it is OK and move it to kernel space.
1407 *
1408 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1409 * break bindings
1410 *
1411 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1412 * other SEQPACKET protocols that take time to connect() as it doesn't
1413 * include the -EINPROGRESS status for such sockets.
1414 */
1415
1416asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1417{
1418 struct socket *sock;
1419 char address[MAX_SOCK_ADDR];
1420 int err;
1421
1422 sock = sockfd_lookup(fd, &err);
1423 if (!sock)
1424 goto out;
1425 err = move_addr_to_kernel(uservaddr, addrlen, address);
1426 if (err < 0)
1427 goto out_put;
1428
1429 err = security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1430 if (err)
1431 goto out_put;
1432
1433 err = sock->ops->connect(sock, (struct sockaddr *) address, addrlen,
1434 sock->file->f_flags);
1435out_put:
1436 sockfd_put(sock);
1437out:
1438 return err;
1439}
1440
1441/*
1442 * Get the local address ('name') of a socket object. Move the obtained
1443 * name to user space.
1444 */
1445
1446asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1447{
1448 struct socket *sock;
1449 char address[MAX_SOCK_ADDR];
1450 int len, err;
1451
1452 sock = sockfd_lookup(fd, &err);
1453 if (!sock)
1454 goto out;
1455
1456 err = security_socket_getsockname(sock);
1457 if (err)
1458 goto out_put;
1459
1460 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 0);
1461 if (err)
1462 goto out_put;
1463 err = move_addr_to_user(address, len, usockaddr, usockaddr_len);
1464
1465out_put:
1466 sockfd_put(sock);
1467out:
1468 return err;
1469}
1470
1471/*
1472 * Get the remote address ('name') of a socket object. Move the obtained
1473 * name to user space.
1474 */
1475
1476asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1477{
1478 struct socket *sock;
1479 char address[MAX_SOCK_ADDR];
1480 int len, err;
1481
1482 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1483 {
1484 err = security_socket_getpeername(sock);
1485 if (err) {
1486 sockfd_put(sock);
1487 return err;
1488 }
1489
1490 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 1);
1491 if (!err)
1492 err=move_addr_to_user(address,len, usockaddr, usockaddr_len);
1493 sockfd_put(sock);
1494 }
1495 return err;
1496}
1497
1498/*
1499 * Send a datagram to a given address. We move the address into kernel
1500 * space and check the user space data area is readable before invoking
1501 * the protocol.
1502 */
1503
1504asmlinkage long sys_sendto(int fd, void __user * buff, size_t len, unsigned flags,
1505 struct sockaddr __user *addr, int addr_len)
1506{
1507 struct socket *sock;
1508 char address[MAX_SOCK_ADDR];
1509 int err;
1510 struct msghdr msg;
1511 struct iovec iov;
1512
1513 sock = sockfd_lookup(fd, &err);
1514 if (!sock)
1515 goto out;
1516 iov.iov_base=buff;
1517 iov.iov_len=len;
1518 msg.msg_name=NULL;
1519 msg.msg_iov=&iov;
1520 msg.msg_iovlen=1;
1521 msg.msg_control=NULL;
1522 msg.msg_controllen=0;
1523 msg.msg_namelen=0;
1524 if(addr)
1525 {
1526 err = move_addr_to_kernel(addr, addr_len, address);
1527 if (err < 0)
1528 goto out_put;
1529 msg.msg_name=address;
1530 msg.msg_namelen=addr_len;
1531 }
1532 if (sock->file->f_flags & O_NONBLOCK)
1533 flags |= MSG_DONTWAIT;
1534 msg.msg_flags = flags;
1535 err = sock_sendmsg(sock, &msg, len);
1536
1537out_put:
1538 sockfd_put(sock);
1539out:
1540 return err;
1541}
1542
1543/*
1544 * Send a datagram down a socket.
1545 */
1546
1547asmlinkage long sys_send(int fd, void __user * buff, size_t len, unsigned flags)
1548{
1549 return sys_sendto(fd, buff, len, flags, NULL, 0);
1550}
1551
1552/*
1553 * Receive a frame from the socket and optionally record the address of the
1554 * sender. We verify the buffers are writable and if needed move the
1555 * sender address from kernel to user space.
1556 */
1557
1558asmlinkage long sys_recvfrom(int fd, void __user * ubuf, size_t size, unsigned flags,
1559 struct sockaddr __user *addr, int __user *addr_len)
1560{
1561 struct socket *sock;
1562 struct iovec iov;
1563 struct msghdr msg;
1564 char address[MAX_SOCK_ADDR];
1565 int err,err2;
1566
1567 sock = sockfd_lookup(fd, &err);
1568 if (!sock)
1569 goto out;
1570
1571 msg.msg_control=NULL;
1572 msg.msg_controllen=0;
1573 msg.msg_iovlen=1;
1574 msg.msg_iov=&iov;
1575 iov.iov_len=size;
1576 iov.iov_base=ubuf;
1577 msg.msg_name=address;
1578 msg.msg_namelen=MAX_SOCK_ADDR;
1579 if (sock->file->f_flags & O_NONBLOCK)
1580 flags |= MSG_DONTWAIT;
1581 err=sock_recvmsg(sock, &msg, size, flags);
1582
1583 if(err >= 0 && addr != NULL)
1584 {
1585 err2=move_addr_to_user(address, msg.msg_namelen, addr, addr_len);
1586 if(err2<0)
1587 err=err2;
1588 }
1589 sockfd_put(sock);
1590out:
1591 return err;
1592}
1593
1594/*
1595 * Receive a datagram from a socket.
1596 */
1597
1598asmlinkage long sys_recv(int fd, void __user * ubuf, size_t size, unsigned flags)
1599{
1600 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1601}
1602
1603/*
1604 * Set a socket option. Because we don't know the option lengths we have
1605 * to pass the user mode parameter for the protocols to sort out.
1606 */
1607
1608asmlinkage long sys_setsockopt(int fd, int level, int optname, char __user *optval, int optlen)
1609{
1610 int err;
1611 struct socket *sock;
1612
1613 if (optlen < 0)
1614 return -EINVAL;
1615
1616 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1617 {
1618 err = security_socket_setsockopt(sock,level,optname);
1619 if (err) {
1620 sockfd_put(sock);
1621 return err;
1622 }
1623
1624 if (level == SOL_SOCKET)
1625 err=sock_setsockopt(sock,level,optname,optval,optlen);
1626 else
1627 err=sock->ops->setsockopt(sock, level, optname, optval, optlen);
1628 sockfd_put(sock);
1629 }
1630 return err;
1631}
1632
1633/*
1634 * Get a socket option. Because we don't know the option lengths we have
1635 * to pass a user mode parameter for the protocols to sort out.
1636 */
1637
1638asmlinkage long sys_getsockopt(int fd, int level, int optname, char __user *optval, int __user *optlen)
1639{
1640 int err;
1641 struct socket *sock;
1642
1643 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1644 {
1645 err = security_socket_getsockopt(sock, level,
1646 optname);
1647 if (err) {
1648 sockfd_put(sock);
1649 return err;
1650 }
1651
1652 if (level == SOL_SOCKET)
1653 err=sock_getsockopt(sock,level,optname,optval,optlen);
1654 else
1655 err=sock->ops->getsockopt(sock, level, optname, optval, optlen);
1656 sockfd_put(sock);
1657 }
1658 return err;
1659}
1660
1661
1662/*
1663 * Shutdown a socket.
1664 */
1665
1666asmlinkage long sys_shutdown(int fd, int how)
1667{
1668 int err;
1669 struct socket *sock;
1670
1671 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1672 {
1673 err = security_socket_shutdown(sock, how);
1674 if (err) {
1675 sockfd_put(sock);
1676 return err;
1677 }
1678
1679 err=sock->ops->shutdown(sock, how);
1680 sockfd_put(sock);
1681 }
1682 return err;
1683}
1684
1685/* A couple of helpful macros for getting the address of the 32/64 bit
1686 * fields which are the same type (int / unsigned) on our platforms.
1687 */
1688#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1689#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1690#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1691
1692
1693/*
1694 * BSD sendmsg interface
1695 */
1696
1697asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1698{
1699 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1700 struct socket *sock;
1701 char address[MAX_SOCK_ADDR];
1702 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1703 unsigned char ctl[sizeof(struct cmsghdr) + 20]; /* 20 is size of ipv6_pktinfo */
1704 unsigned char *ctl_buf = ctl;
1705 struct msghdr msg_sys;
1706 int err, ctl_len, iov_size, total_len;
1707
1708 err = -EFAULT;
1709 if (MSG_CMSG_COMPAT & flags) {
1710 if (get_compat_msghdr(&msg_sys, msg_compat))
1711 return -EFAULT;
1712 } else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1713 return -EFAULT;
1714
1715 sock = sockfd_lookup(fd, &err);
1716 if (!sock)
1717 goto out;
1718
1719 /* do not move before msg_sys is valid */
1720 err = -EMSGSIZE;
1721 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1722 goto out_put;
1723
1724 /* Check whether to allocate the iovec area*/
1725 err = -ENOMEM;
1726 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1727 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1728 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1729 if (!iov)
1730 goto out_put;
1731 }
1732
1733 /* This will also move the address data into kernel space */
1734 if (MSG_CMSG_COMPAT & flags) {
1735 err = verify_compat_iovec(&msg_sys, iov, address, VERIFY_READ);
1736 } else
1737 err = verify_iovec(&msg_sys, iov, address, VERIFY_READ);
1738 if (err < 0)
1739 goto out_freeiov;
1740 total_len = err;
1741
1742 err = -ENOBUFS;
1743
1744 if (msg_sys.msg_controllen > INT_MAX)
1745 goto out_freeiov;
1746 ctl_len = msg_sys.msg_controllen;
1747 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
Al Viro8920e8f2005-09-07 18:28:51 -07001748 err = cmsghdr_from_user_compat_to_kern(&msg_sys, sock->sk, ctl, sizeof(ctl));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 if (err)
1750 goto out_freeiov;
1751 ctl_buf = msg_sys.msg_control;
Al Viro8920e8f2005-09-07 18:28:51 -07001752 ctl_len = msg_sys.msg_controllen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 } else if (ctl_len) {
1754 if (ctl_len > sizeof(ctl))
1755 {
1756 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1757 if (ctl_buf == NULL)
1758 goto out_freeiov;
1759 }
1760 err = -EFAULT;
1761 /*
1762 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1763 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1764 * checking falls down on this.
1765 */
1766 if (copy_from_user(ctl_buf, (void __user *) msg_sys.msg_control, ctl_len))
1767 goto out_freectl;
1768 msg_sys.msg_control = ctl_buf;
1769 }
1770 msg_sys.msg_flags = flags;
1771
1772 if (sock->file->f_flags & O_NONBLOCK)
1773 msg_sys.msg_flags |= MSG_DONTWAIT;
1774 err = sock_sendmsg(sock, &msg_sys, total_len);
1775
1776out_freectl:
1777 if (ctl_buf != ctl)
1778 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1779out_freeiov:
1780 if (iov != iovstack)
1781 sock_kfree_s(sock->sk, iov, iov_size);
1782out_put:
1783 sockfd_put(sock);
1784out:
1785 return err;
1786}
1787
1788/*
1789 * BSD recvmsg interface
1790 */
1791
1792asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg, unsigned int flags)
1793{
1794 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1795 struct socket *sock;
1796 struct iovec iovstack[UIO_FASTIOV];
1797 struct iovec *iov=iovstack;
1798 struct msghdr msg_sys;
1799 unsigned long cmsg_ptr;
1800 int err, iov_size, total_len, len;
1801
1802 /* kernel mode address */
1803 char addr[MAX_SOCK_ADDR];
1804
1805 /* user mode address pointers */
1806 struct sockaddr __user *uaddr;
1807 int __user *uaddr_len;
1808
1809 if (MSG_CMSG_COMPAT & flags) {
1810 if (get_compat_msghdr(&msg_sys, msg_compat))
1811 return -EFAULT;
1812 } else
1813 if (copy_from_user(&msg_sys,msg,sizeof(struct msghdr)))
1814 return -EFAULT;
1815
1816 sock = sockfd_lookup(fd, &err);
1817 if (!sock)
1818 goto out;
1819
1820 err = -EMSGSIZE;
1821 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1822 goto out_put;
1823
1824 /* Check whether to allocate the iovec area*/
1825 err = -ENOMEM;
1826 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1827 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1828 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1829 if (!iov)
1830 goto out_put;
1831 }
1832
1833 /*
1834 * Save the user-mode address (verify_iovec will change the
1835 * kernel msghdr to use the kernel address space)
1836 */
1837
1838 uaddr = (void __user *) msg_sys.msg_name;
1839 uaddr_len = COMPAT_NAMELEN(msg);
1840 if (MSG_CMSG_COMPAT & flags) {
1841 err = verify_compat_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1842 } else
1843 err = verify_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1844 if (err < 0)
1845 goto out_freeiov;
1846 total_len=err;
1847
1848 cmsg_ptr = (unsigned long)msg_sys.msg_control;
1849 msg_sys.msg_flags = 0;
1850 if (MSG_CMSG_COMPAT & flags)
1851 msg_sys.msg_flags = MSG_CMSG_COMPAT;
1852
1853 if (sock->file->f_flags & O_NONBLOCK)
1854 flags |= MSG_DONTWAIT;
1855 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1856 if (err < 0)
1857 goto out_freeiov;
1858 len = err;
1859
1860 if (uaddr != NULL) {
1861 err = move_addr_to_user(addr, msg_sys.msg_namelen, uaddr, uaddr_len);
1862 if (err < 0)
1863 goto out_freeiov;
1864 }
1865 err = __put_user(msg_sys.msg_flags, COMPAT_FLAGS(msg));
1866 if (err)
1867 goto out_freeiov;
1868 if (MSG_CMSG_COMPAT & flags)
1869 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1870 &msg_compat->msg_controllen);
1871 else
1872 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1873 &msg->msg_controllen);
1874 if (err)
1875 goto out_freeiov;
1876 err = len;
1877
1878out_freeiov:
1879 if (iov != iovstack)
1880 sock_kfree_s(sock->sk, iov, iov_size);
1881out_put:
1882 sockfd_put(sock);
1883out:
1884 return err;
1885}
1886
1887#ifdef __ARCH_WANT_SYS_SOCKETCALL
1888
1889/* Argument list sizes for sys_socketcall */
1890#define AL(x) ((x) * sizeof(unsigned long))
1891static unsigned char nargs[18]={AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
1892 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
1893 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3)};
1894#undef AL
1895
1896/*
1897 * System call vectors.
1898 *
1899 * Argument checking cleaned up. Saved 20% in size.
1900 * This function doesn't need to set the kernel lock because
1901 * it is set by the callees.
1902 */
1903
1904asmlinkage long sys_socketcall(int call, unsigned long __user *args)
1905{
1906 unsigned long a[6];
1907 unsigned long a0,a1;
1908 int err;
1909
1910 if(call<1||call>SYS_RECVMSG)
1911 return -EINVAL;
1912
1913 /* copy_from_user should be SMP safe. */
1914 if (copy_from_user(a, args, nargs[call]))
1915 return -EFAULT;
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01001916
David Woodhouse4bcff1b2005-06-02 12:13:21 +01001917 err = audit_socketcall(nargs[call]/sizeof(unsigned long), a);
David Woodhouse3ec3b2f2005-05-17 12:08:48 +01001918 if (err)
1919 return err;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 a0=a[0];
1922 a1=a[1];
1923
1924 switch(call)
1925 {
1926 case SYS_SOCKET:
1927 err = sys_socket(a0,a1,a[2]);
1928 break;
1929 case SYS_BIND:
1930 err = sys_bind(a0,(struct sockaddr __user *)a1, a[2]);
1931 break;
1932 case SYS_CONNECT:
1933 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
1934 break;
1935 case SYS_LISTEN:
1936 err = sys_listen(a0,a1);
1937 break;
1938 case SYS_ACCEPT:
1939 err = sys_accept(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1940 break;
1941 case SYS_GETSOCKNAME:
1942 err = sys_getsockname(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1943 break;
1944 case SYS_GETPEERNAME:
1945 err = sys_getpeername(a0, (struct sockaddr __user *)a1, (int __user *)a[2]);
1946 break;
1947 case SYS_SOCKETPAIR:
1948 err = sys_socketpair(a0,a1, a[2], (int __user *)a[3]);
1949 break;
1950 case SYS_SEND:
1951 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
1952 break;
1953 case SYS_SENDTO:
1954 err = sys_sendto(a0,(void __user *)a1, a[2], a[3],
1955 (struct sockaddr __user *)a[4], a[5]);
1956 break;
1957 case SYS_RECV:
1958 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
1959 break;
1960 case SYS_RECVFROM:
1961 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
1962 (struct sockaddr __user *)a[4], (int __user *)a[5]);
1963 break;
1964 case SYS_SHUTDOWN:
1965 err = sys_shutdown(a0,a1);
1966 break;
1967 case SYS_SETSOCKOPT:
1968 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
1969 break;
1970 case SYS_GETSOCKOPT:
1971 err = sys_getsockopt(a0, a1, a[2], (char __user *)a[3], (int __user *)a[4]);
1972 break;
1973 case SYS_SENDMSG:
1974 err = sys_sendmsg(a0, (struct msghdr __user *) a1, a[2]);
1975 break;
1976 case SYS_RECVMSG:
1977 err = sys_recvmsg(a0, (struct msghdr __user *) a1, a[2]);
1978 break;
1979 default:
1980 err = -EINVAL;
1981 break;
1982 }
1983 return err;
1984}
1985
1986#endif /* __ARCH_WANT_SYS_SOCKETCALL */
1987
1988/*
1989 * This function is called by a protocol handler that wants to
1990 * advertise its address family, and have it linked into the
1991 * SOCKET module.
1992 */
1993
1994int sock_register(struct net_proto_family *ops)
1995{
1996 int err;
1997
1998 if (ops->family >= NPROTO) {
1999 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
2000 return -ENOBUFS;
2001 }
2002 net_family_write_lock();
2003 err = -EEXIST;
2004 if (net_families[ops->family] == NULL) {
2005 net_families[ops->family]=ops;
2006 err = 0;
2007 }
2008 net_family_write_unlock();
2009 printk(KERN_INFO "NET: Registered protocol family %d\n",
2010 ops->family);
2011 return err;
2012}
2013
2014/*
2015 * This function is called by a protocol handler that wants to
2016 * remove its address family, and have it unlinked from the
2017 * SOCKET module.
2018 */
2019
2020int sock_unregister(int family)
2021{
2022 if (family < 0 || family >= NPROTO)
2023 return -1;
2024
2025 net_family_write_lock();
2026 net_families[family]=NULL;
2027 net_family_write_unlock();
2028 printk(KERN_INFO "NET: Unregistered protocol family %d\n",
2029 family);
2030 return 0;
2031}
2032
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033void __init sock_init(void)
2034{
2035 /*
2036 * Initialize sock SLAB cache.
2037 */
2038
2039 sk_init();
2040
2041#ifdef SLAB_SKB
2042 /*
2043 * Initialize skbuff SLAB cache
2044 */
2045 skb_init();
2046#endif
2047
2048 /*
2049 * Initialize the protocols module.
2050 */
2051
2052 init_inodecache();
2053 register_filesystem(&sock_fs_type);
2054 sock_mnt = kern_mount(&sock_fs_type);
2055 /* The real protocol initialization is performed when
2056 * do_initcalls is run.
2057 */
2058
2059#ifdef CONFIG_NETFILTER
2060 netfilter_init();
2061#endif
2062}
2063
2064#ifdef CONFIG_PROC_FS
2065void socket_seq_show(struct seq_file *seq)
2066{
2067 int cpu;
2068 int counter = 0;
2069
2070 for (cpu = 0; cpu < NR_CPUS; cpu++)
2071 counter += per_cpu(sockets_in_use, cpu);
2072
2073 /* It can be negative, by the way. 8) */
2074 if (counter < 0)
2075 counter = 0;
2076
2077 seq_printf(seq, "sockets: used %d\n", counter);
2078}
2079#endif /* CONFIG_PROC_FS */
2080
2081/* ABI emulation layers need these two */
2082EXPORT_SYMBOL(move_addr_to_kernel);
2083EXPORT_SYMBOL(move_addr_to_user);
2084EXPORT_SYMBOL(sock_create);
2085EXPORT_SYMBOL(sock_create_kern);
2086EXPORT_SYMBOL(sock_create_lite);
2087EXPORT_SYMBOL(sock_map_fd);
2088EXPORT_SYMBOL(sock_recvmsg);
2089EXPORT_SYMBOL(sock_register);
2090EXPORT_SYMBOL(sock_release);
2091EXPORT_SYMBOL(sock_sendmsg);
2092EXPORT_SYMBOL(sock_unregister);
2093EXPORT_SYMBOL(sock_wake_async);
2094EXPORT_SYMBOL(sockfd_lookup);
2095EXPORT_SYMBOL(kernel_sendmsg);
2096EXPORT_SYMBOL(kernel_recvmsg);