blob: 2cd44990d8d3daa6ce0495422549ea4745a56e1b [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>
7 * Ross Biro, <bir7@leland.Stanford.Edu>
8 * 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>
84
85#ifdef CONFIG_NET_RADIO
86#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
87#endif /* CONFIG_NET_RADIO */
88
89#include <asm/uaccess.h>
90#include <asm/unistd.h>
91
92#include <net/compat.h>
93
94#include <net/sock.h>
95#include <linux/netfilter.h>
96
97static int sock_no_open(struct inode *irrelevant, struct file *dontcare);
98static ssize_t sock_aio_read(struct kiocb *iocb, char __user *buf,
99 size_t size, loff_t pos);
100static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *buf,
101 size_t size, loff_t pos);
102static int sock_mmap(struct file *file, struct vm_area_struct * vma);
103
104static int sock_close(struct inode *inode, struct file *file);
105static unsigned int sock_poll(struct file *file,
106 struct poll_table_struct *wait);
107static long sock_ioctl(struct file *file,
108 unsigned int cmd, unsigned long arg);
109static int sock_fasync(int fd, struct file *filp, int on);
110static ssize_t sock_readv(struct file *file, const struct iovec *vector,
111 unsigned long count, loff_t *ppos);
112static ssize_t sock_writev(struct file *file, const struct iovec *vector,
113 unsigned long count, loff_t *ppos);
114static ssize_t sock_sendpage(struct file *file, struct page *page,
115 int offset, size_t size, loff_t *ppos, int more);
116
117
118/*
119 * Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
120 * in the operation structures but are done directly via the socketcall() multiplexor.
121 */
122
123static struct file_operations socket_file_ops = {
124 .owner = THIS_MODULE,
125 .llseek = no_llseek,
126 .aio_read = sock_aio_read,
127 .aio_write = sock_aio_write,
128 .poll = sock_poll,
129 .unlocked_ioctl = sock_ioctl,
130 .mmap = sock_mmap,
131 .open = sock_no_open, /* special open code to disallow open via /proc */
132 .release = sock_close,
133 .fasync = sock_fasync,
134 .readv = sock_readv,
135 .writev = sock_writev,
136 .sendpage = sock_sendpage
137};
138
139/*
140 * The protocol list. Each protocol is registered in here.
141 */
142
143static struct net_proto_family *net_families[NPROTO];
144
145#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
146static atomic_t net_family_lockct = ATOMIC_INIT(0);
147static DEFINE_SPINLOCK(net_family_lock);
148
149/* The strategy is: modifications net_family vector are short, do not
150 sleep and veeery rare, but read access should be free of any exclusive
151 locks.
152 */
153
154static void net_family_write_lock(void)
155{
156 spin_lock(&net_family_lock);
157 while (atomic_read(&net_family_lockct) != 0) {
158 spin_unlock(&net_family_lock);
159
160 yield();
161
162 spin_lock(&net_family_lock);
163 }
164}
165
166static __inline__ void net_family_write_unlock(void)
167{
168 spin_unlock(&net_family_lock);
169}
170
171static __inline__ void net_family_read_lock(void)
172{
173 atomic_inc(&net_family_lockct);
174 spin_unlock_wait(&net_family_lock);
175}
176
177static __inline__ void net_family_read_unlock(void)
178{
179 atomic_dec(&net_family_lockct);
180}
181
182#else
183#define net_family_write_lock() do { } while(0)
184#define net_family_write_unlock() do { } while(0)
185#define net_family_read_lock() do { } while(0)
186#define net_family_read_unlock() do { } while(0)
187#endif
188
189
190/*
191 * Statistics counters of the socket lists
192 */
193
194static DEFINE_PER_CPU(int, sockets_in_use) = 0;
195
196/*
197 * Support routines. Move socket addresses back and forth across the kernel/user
198 * divide and look after the messy bits.
199 */
200
201#define MAX_SOCK_ADDR 128 /* 108 for Unix domain -
202 16 for IP, 16 for IPX,
203 24 for IPv6,
204 about 80 for AX.25
205 must be at least one bigger than
206 the AF_UNIX size (see net/unix/af_unix.c
207 :unix_mkname()).
208 */
209
210/**
211 * move_addr_to_kernel - copy a socket address into kernel space
212 * @uaddr: Address in user space
213 * @kaddr: Address in kernel space
214 * @ulen: Length in user space
215 *
216 * The address is copied into kernel space. If the provided address is
217 * too long an error code of -EINVAL is returned. If the copy gives
218 * invalid addresses -EFAULT is returned. On a success 0 is returned.
219 */
220
221int move_addr_to_kernel(void __user *uaddr, int ulen, void *kaddr)
222{
223 if(ulen<0||ulen>MAX_SOCK_ADDR)
224 return -EINVAL;
225 if(ulen==0)
226 return 0;
227 if(copy_from_user(kaddr,uaddr,ulen))
228 return -EFAULT;
229 return 0;
230}
231
232/**
233 * move_addr_to_user - copy an address to user space
234 * @kaddr: kernel space address
235 * @klen: length of address in kernel
236 * @uaddr: user space address
237 * @ulen: pointer to user length field
238 *
239 * The value pointed to by ulen on entry is the buffer length available.
240 * This is overwritten with the buffer space used. -EINVAL is returned
241 * if an overlong buffer is specified or a negative buffer size. -EFAULT
242 * is returned if either the buffer or the length field are not
243 * accessible.
244 * After copying the data up to the limit the user specifies, the true
245 * length of the data is written over the length limit the user
246 * specified. Zero is returned for a success.
247 */
248
249int move_addr_to_user(void *kaddr, int klen, void __user *uaddr, int __user *ulen)
250{
251 int err;
252 int len;
253
254 if((err=get_user(len, ulen)))
255 return err;
256 if(len>klen)
257 len=klen;
258 if(len<0 || len> MAX_SOCK_ADDR)
259 return -EINVAL;
260 if(len)
261 {
262 if(copy_to_user(uaddr,kaddr,len))
263 return -EFAULT;
264 }
265 /*
266 * "fromlen shall refer to the value before truncation.."
267 * 1003.1g
268 */
269 return __put_user(klen, ulen);
270}
271
272#define SOCKFS_MAGIC 0x534F434B
273
274static kmem_cache_t * sock_inode_cachep;
275
276static struct inode *sock_alloc_inode(struct super_block *sb)
277{
278 struct socket_alloc *ei;
279 ei = (struct socket_alloc *)kmem_cache_alloc(sock_inode_cachep, SLAB_KERNEL);
280 if (!ei)
281 return NULL;
282 init_waitqueue_head(&ei->socket.wait);
283
284 ei->socket.fasync_list = NULL;
285 ei->socket.state = SS_UNCONNECTED;
286 ei->socket.flags = 0;
287 ei->socket.ops = NULL;
288 ei->socket.sk = NULL;
289 ei->socket.file = NULL;
290 ei->socket.flags = 0;
291
292 return &ei->vfs_inode;
293}
294
295static void sock_destroy_inode(struct inode *inode)
296{
297 kmem_cache_free(sock_inode_cachep,
298 container_of(inode, struct socket_alloc, vfs_inode));
299}
300
301static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
302{
303 struct socket_alloc *ei = (struct socket_alloc *) foo;
304
305 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
306 SLAB_CTOR_CONSTRUCTOR)
307 inode_init_once(&ei->vfs_inode);
308}
309
310static int init_inodecache(void)
311{
312 sock_inode_cachep = kmem_cache_create("sock_inode_cache",
313 sizeof(struct socket_alloc),
314 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
315 init_once, NULL);
316 if (sock_inode_cachep == NULL)
317 return -ENOMEM;
318 return 0;
319}
320
321static struct super_operations sockfs_ops = {
322 .alloc_inode = sock_alloc_inode,
323 .destroy_inode =sock_destroy_inode,
324 .statfs = simple_statfs,
325};
326
327static struct super_block *sockfs_get_sb(struct file_system_type *fs_type,
328 int flags, const char *dev_name, void *data)
329{
330 return get_sb_pseudo(fs_type, "socket:", &sockfs_ops, SOCKFS_MAGIC);
331}
332
333static struct vfsmount *sock_mnt;
334
335static struct file_system_type sock_fs_type = {
336 .name = "sockfs",
337 .get_sb = sockfs_get_sb,
338 .kill_sb = kill_anon_super,
339};
340static int sockfs_delete_dentry(struct dentry *dentry)
341{
342 return 1;
343}
344static struct dentry_operations sockfs_dentry_operations = {
345 .d_delete = sockfs_delete_dentry,
346};
347
348/*
349 * Obtains the first available file descriptor and sets it up for use.
350 *
351 * This function creates file structure and maps it to fd space
352 * of current process. On success it returns file descriptor
353 * and file struct implicitly stored in sock->file.
354 * Note that another thread may close file descriptor before we return
355 * from this function. We use the fact that now we do not refer
356 * to socket after mapping. If one day we will need it, this
357 * function will increment ref. count on file by 1.
358 *
359 * In any case returned fd MAY BE not valid!
360 * This race condition is unavoidable
361 * with shared fd spaces, we cannot solve it inside kernel,
362 * but we take care of internal coherence yet.
363 */
364
365int sock_map_fd(struct socket *sock)
366{
367 int fd;
368 struct qstr this;
369 char name[32];
370
371 /*
372 * Find a file descriptor suitable for return to the user.
373 */
374
375 fd = get_unused_fd();
376 if (fd >= 0) {
377 struct file *file = get_empty_filp();
378
379 if (!file) {
380 put_unused_fd(fd);
381 fd = -ENFILE;
382 goto out;
383 }
384
385 sprintf(name, "[%lu]", SOCK_INODE(sock)->i_ino);
386 this.name = name;
387 this.len = strlen(name);
388 this.hash = SOCK_INODE(sock)->i_ino;
389
390 file->f_dentry = d_alloc(sock_mnt->mnt_sb->s_root, &this);
391 if (!file->f_dentry) {
392 put_filp(file);
393 put_unused_fd(fd);
394 fd = -ENOMEM;
395 goto out;
396 }
397 file->f_dentry->d_op = &sockfs_dentry_operations;
398 d_add(file->f_dentry, SOCK_INODE(sock));
399 file->f_vfsmnt = mntget(sock_mnt);
400 file->f_mapping = file->f_dentry->d_inode->i_mapping;
401
402 sock->file = file;
403 file->f_op = SOCK_INODE(sock)->i_fop = &socket_file_ops;
404 file->f_mode = FMODE_READ | FMODE_WRITE;
405 file->f_flags = O_RDWR;
406 file->f_pos = 0;
407 fd_install(fd, file);
408 }
409
410out:
411 return fd;
412}
413
414/**
415 * sockfd_lookup - Go from a file number to its socket slot
416 * @fd: file handle
417 * @err: pointer to an error code return
418 *
419 * The file handle passed in is locked and the socket it is bound
420 * too is returned. If an error occurs the err pointer is overwritten
421 * with a negative errno code and NULL is returned. The function checks
422 * for both invalid handles and passing a handle which is not a socket.
423 *
424 * On a success the socket object pointer is returned.
425 */
426
427struct socket *sockfd_lookup(int fd, int *err)
428{
429 struct file *file;
430 struct inode *inode;
431 struct socket *sock;
432
433 if (!(file = fget(fd)))
434 {
435 *err = -EBADF;
436 return NULL;
437 }
438
439 inode = file->f_dentry->d_inode;
440 if (!S_ISSOCK(inode->i_mode)) {
441 *err = -ENOTSOCK;
442 fput(file);
443 return NULL;
444 }
445
446 sock = SOCKET_I(inode);
447 if (sock->file != file) {
448 printk(KERN_ERR "socki_lookup: socket file changed!\n");
449 sock->file = file;
450 }
451 return sock;
452}
453
454/**
455 * sock_alloc - allocate a socket
456 *
457 * Allocate a new inode and socket object. The two are bound together
458 * and initialised. The socket is then returned. If we are out of inodes
459 * NULL is returned.
460 */
461
462static struct socket *sock_alloc(void)
463{
464 struct inode * inode;
465 struct socket * sock;
466
467 inode = new_inode(sock_mnt->mnt_sb);
468 if (!inode)
469 return NULL;
470
471 sock = SOCKET_I(inode);
472
473 inode->i_mode = S_IFSOCK|S_IRWXUGO;
474 inode->i_uid = current->fsuid;
475 inode->i_gid = current->fsgid;
476
477 get_cpu_var(sockets_in_use)++;
478 put_cpu_var(sockets_in_use);
479 return sock;
480}
481
482/*
483 * In theory you can't get an open on this inode, but /proc provides
484 * a back door. Remember to keep it shut otherwise you'll let the
485 * creepy crawlies in.
486 */
487
488static int sock_no_open(struct inode *irrelevant, struct file *dontcare)
489{
490 return -ENXIO;
491}
492
493struct file_operations bad_sock_fops = {
494 .owner = THIS_MODULE,
495 .open = sock_no_open,
496};
497
498/**
499 * sock_release - close a socket
500 * @sock: socket to close
501 *
502 * The socket is released from the protocol stack if it has a release
503 * callback, and the inode is then released if the socket is bound to
504 * an inode not a file.
505 */
506
507void sock_release(struct socket *sock)
508{
509 if (sock->ops) {
510 struct module *owner = sock->ops->owner;
511
512 sock->ops->release(sock);
513 sock->ops = NULL;
514 module_put(owner);
515 }
516
517 if (sock->fasync_list)
518 printk(KERN_ERR "sock_release: fasync list not empty!\n");
519
520 get_cpu_var(sockets_in_use)--;
521 put_cpu_var(sockets_in_use);
522 if (!sock->file) {
523 iput(SOCK_INODE(sock));
524 return;
525 }
526 sock->file=NULL;
527}
528
529static inline int __sock_sendmsg(struct kiocb *iocb, struct socket *sock,
530 struct msghdr *msg, size_t size)
531{
532 struct sock_iocb *si = kiocb_to_siocb(iocb);
533 int err;
534
535 si->sock = sock;
536 si->scm = NULL;
537 si->msg = msg;
538 si->size = size;
539
540 err = security_socket_sendmsg(sock, msg, size);
541 if (err)
542 return err;
543
544 return sock->ops->sendmsg(iocb, sock, msg, size);
545}
546
547int sock_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
548{
549 struct kiocb iocb;
550 struct sock_iocb siocb;
551 int ret;
552
553 init_sync_kiocb(&iocb, NULL);
554 iocb.private = &siocb;
555 ret = __sock_sendmsg(&iocb, sock, msg, size);
556 if (-EIOCBQUEUED == ret)
557 ret = wait_on_sync_kiocb(&iocb);
558 return ret;
559}
560
561int kernel_sendmsg(struct socket *sock, struct msghdr *msg,
562 struct kvec *vec, size_t num, size_t size)
563{
564 mm_segment_t oldfs = get_fs();
565 int result;
566
567 set_fs(KERNEL_DS);
568 /*
569 * the following is safe, since for compiler definitions of kvec and
570 * iovec are identical, yielding the same in-core layout and alignment
571 */
572 msg->msg_iov = (struct iovec *)vec,
573 msg->msg_iovlen = num;
574 result = sock_sendmsg(sock, msg, size);
575 set_fs(oldfs);
576 return result;
577}
578
579static inline int __sock_recvmsg(struct kiocb *iocb, struct socket *sock,
580 struct msghdr *msg, size_t size, int flags)
581{
582 int err;
583 struct sock_iocb *si = kiocb_to_siocb(iocb);
584
585 si->sock = sock;
586 si->scm = NULL;
587 si->msg = msg;
588 si->size = size;
589 si->flags = flags;
590
591 err = security_socket_recvmsg(sock, msg, size, flags);
592 if (err)
593 return err;
594
595 return sock->ops->recvmsg(iocb, sock, msg, size, flags);
596}
597
598int sock_recvmsg(struct socket *sock, struct msghdr *msg,
599 size_t size, int flags)
600{
601 struct kiocb iocb;
602 struct sock_iocb siocb;
603 int ret;
604
605 init_sync_kiocb(&iocb, NULL);
606 iocb.private = &siocb;
607 ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
608 if (-EIOCBQUEUED == ret)
609 ret = wait_on_sync_kiocb(&iocb);
610 return ret;
611}
612
613int kernel_recvmsg(struct socket *sock, struct msghdr *msg,
614 struct kvec *vec, size_t num,
615 size_t size, int flags)
616{
617 mm_segment_t oldfs = get_fs();
618 int result;
619
620 set_fs(KERNEL_DS);
621 /*
622 * the following is safe, since for compiler definitions of kvec and
623 * iovec are identical, yielding the same in-core layout and alignment
624 */
625 msg->msg_iov = (struct iovec *)vec,
626 msg->msg_iovlen = num;
627 result = sock_recvmsg(sock, msg, size, flags);
628 set_fs(oldfs);
629 return result;
630}
631
632static void sock_aio_dtor(struct kiocb *iocb)
633{
634 kfree(iocb->private);
635}
636
637/*
638 * Read data from a socket. ubuf is a user mode pointer. We make sure the user
639 * area ubuf...ubuf+size-1 is writable before asking the protocol.
640 */
641
642static ssize_t sock_aio_read(struct kiocb *iocb, char __user *ubuf,
643 size_t size, loff_t pos)
644{
645 struct sock_iocb *x, siocb;
646 struct socket *sock;
647 int flags;
648
649 if (pos != 0)
650 return -ESPIPE;
651 if (size==0) /* Match SYS5 behaviour */
652 return 0;
653
654 if (is_sync_kiocb(iocb))
655 x = &siocb;
656 else {
657 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
658 if (!x)
659 return -ENOMEM;
660 iocb->ki_dtor = sock_aio_dtor;
661 }
662 iocb->private = x;
663 x->kiocb = iocb;
664 sock = SOCKET_I(iocb->ki_filp->f_dentry->d_inode);
665
666 x->async_msg.msg_name = NULL;
667 x->async_msg.msg_namelen = 0;
668 x->async_msg.msg_iov = &x->async_iov;
669 x->async_msg.msg_iovlen = 1;
670 x->async_msg.msg_control = NULL;
671 x->async_msg.msg_controllen = 0;
672 x->async_iov.iov_base = ubuf;
673 x->async_iov.iov_len = size;
674 flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
675
676 return __sock_recvmsg(iocb, sock, &x->async_msg, size, flags);
677}
678
679
680/*
681 * Write data to a socket. We verify that the user area ubuf..ubuf+size-1
682 * is readable by the user process.
683 */
684
685static ssize_t sock_aio_write(struct kiocb *iocb, const char __user *ubuf,
686 size_t size, loff_t pos)
687{
688 struct sock_iocb *x, siocb;
689 struct socket *sock;
690
691 if (pos != 0)
692 return -ESPIPE;
693 if(size==0) /* Match SYS5 behaviour */
694 return 0;
695
696 if (is_sync_kiocb(iocb))
697 x = &siocb;
698 else {
699 x = kmalloc(sizeof(struct sock_iocb), GFP_KERNEL);
700 if (!x)
701 return -ENOMEM;
702 iocb->ki_dtor = sock_aio_dtor;
703 }
704 iocb->private = x;
705 x->kiocb = iocb;
706 sock = SOCKET_I(iocb->ki_filp->f_dentry->d_inode);
707
708 x->async_msg.msg_name = NULL;
709 x->async_msg.msg_namelen = 0;
710 x->async_msg.msg_iov = &x->async_iov;
711 x->async_msg.msg_iovlen = 1;
712 x->async_msg.msg_control = NULL;
713 x->async_msg.msg_controllen = 0;
714 x->async_msg.msg_flags = !(iocb->ki_filp->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
715 if (sock->type == SOCK_SEQPACKET)
716 x->async_msg.msg_flags |= MSG_EOR;
717 x->async_iov.iov_base = (void __user *)ubuf;
718 x->async_iov.iov_len = size;
719
720 return __sock_sendmsg(iocb, sock, &x->async_msg, size);
721}
722
723ssize_t sock_sendpage(struct file *file, struct page *page,
724 int offset, size_t size, loff_t *ppos, int more)
725{
726 struct socket *sock;
727 int flags;
728
729 sock = SOCKET_I(file->f_dentry->d_inode);
730
731 flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT;
732 if (more)
733 flags |= MSG_MORE;
734
735 return sock->ops->sendpage(sock, page, offset, size, flags);
736}
737
738static int sock_readv_writev(int type, struct inode * inode,
739 struct file * file, const struct iovec * iov,
740 long count, size_t size)
741{
742 struct msghdr msg;
743 struct socket *sock;
744
745 sock = SOCKET_I(inode);
746
747 msg.msg_name = NULL;
748 msg.msg_namelen = 0;
749 msg.msg_control = NULL;
750 msg.msg_controllen = 0;
751 msg.msg_iov = (struct iovec *) iov;
752 msg.msg_iovlen = count;
753 msg.msg_flags = (file->f_flags & O_NONBLOCK) ? MSG_DONTWAIT : 0;
754
755 /* read() does a VERIFY_WRITE */
756 if (type == VERIFY_WRITE)
757 return sock_recvmsg(sock, &msg, size, msg.msg_flags);
758
759 if (sock->type == SOCK_SEQPACKET)
760 msg.msg_flags |= MSG_EOR;
761
762 return sock_sendmsg(sock, &msg, size);
763}
764
765static ssize_t sock_readv(struct file *file, const struct iovec *vector,
766 unsigned long count, loff_t *ppos)
767{
768 size_t tot_len = 0;
769 int i;
770 for (i = 0 ; i < count ; i++)
771 tot_len += vector[i].iov_len;
772 return sock_readv_writev(VERIFY_WRITE, file->f_dentry->d_inode,
773 file, vector, count, tot_len);
774}
775
776static ssize_t sock_writev(struct file *file, const struct iovec *vector,
777 unsigned long count, loff_t *ppos)
778{
779 size_t tot_len = 0;
780 int i;
781 for (i = 0 ; i < count ; i++)
782 tot_len += vector[i].iov_len;
783 return sock_readv_writev(VERIFY_READ, file->f_dentry->d_inode,
784 file, vector, count, tot_len);
785}
786
787
788/*
789 * Atomic setting of ioctl hooks to avoid race
790 * with module unload.
791 */
792
793static DECLARE_MUTEX(br_ioctl_mutex);
794static int (*br_ioctl_hook)(unsigned int cmd, void __user *arg) = NULL;
795
796void brioctl_set(int (*hook)(unsigned int, void __user *))
797{
798 down(&br_ioctl_mutex);
799 br_ioctl_hook = hook;
800 up(&br_ioctl_mutex);
801}
802EXPORT_SYMBOL(brioctl_set);
803
804static DECLARE_MUTEX(vlan_ioctl_mutex);
805static int (*vlan_ioctl_hook)(void __user *arg);
806
807void vlan_ioctl_set(int (*hook)(void __user *))
808{
809 down(&vlan_ioctl_mutex);
810 vlan_ioctl_hook = hook;
811 up(&vlan_ioctl_mutex);
812}
813EXPORT_SYMBOL(vlan_ioctl_set);
814
815static DECLARE_MUTEX(dlci_ioctl_mutex);
816static int (*dlci_ioctl_hook)(unsigned int, void __user *);
817
818void dlci_ioctl_set(int (*hook)(unsigned int, void __user *))
819{
820 down(&dlci_ioctl_mutex);
821 dlci_ioctl_hook = hook;
822 up(&dlci_ioctl_mutex);
823}
824EXPORT_SYMBOL(dlci_ioctl_set);
825
826/*
827 * With an ioctl, arg may well be a user mode pointer, but we don't know
828 * what to do with it - that's up to the protocol still.
829 */
830
831static long sock_ioctl(struct file *file, unsigned cmd, unsigned long arg)
832{
833 struct socket *sock;
834 void __user *argp = (void __user *)arg;
835 int pid, err;
836
837 sock = SOCKET_I(file->f_dentry->d_inode);
838 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
839 err = dev_ioctl(cmd, argp);
840 } else
841#ifdef WIRELESS_EXT
842 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
843 err = dev_ioctl(cmd, argp);
844 } else
845#endif /* WIRELESS_EXT */
846 switch (cmd) {
847 case FIOSETOWN:
848 case SIOCSPGRP:
849 err = -EFAULT;
850 if (get_user(pid, (int __user *)argp))
851 break;
852 err = f_setown(sock->file, pid, 1);
853 break;
854 case FIOGETOWN:
855 case SIOCGPGRP:
856 err = put_user(sock->file->f_owner.pid, (int __user *)argp);
857 break;
858 case SIOCGIFBR:
859 case SIOCSIFBR:
860 case SIOCBRADDBR:
861 case SIOCBRDELBR:
862 err = -ENOPKG;
863 if (!br_ioctl_hook)
864 request_module("bridge");
865
866 down(&br_ioctl_mutex);
867 if (br_ioctl_hook)
868 err = br_ioctl_hook(cmd, argp);
869 up(&br_ioctl_mutex);
870 break;
871 case SIOCGIFVLAN:
872 case SIOCSIFVLAN:
873 err = -ENOPKG;
874 if (!vlan_ioctl_hook)
875 request_module("8021q");
876
877 down(&vlan_ioctl_mutex);
878 if (vlan_ioctl_hook)
879 err = vlan_ioctl_hook(argp);
880 up(&vlan_ioctl_mutex);
881 break;
882 case SIOCGIFDIVERT:
883 case SIOCSIFDIVERT:
884 /* Convert this to call through a hook */
885 err = divert_ioctl(cmd, argp);
886 break;
887 case SIOCADDDLCI:
888 case SIOCDELDLCI:
889 err = -ENOPKG;
890 if (!dlci_ioctl_hook)
891 request_module("dlci");
892
893 if (dlci_ioctl_hook) {
894 down(&dlci_ioctl_mutex);
895 err = dlci_ioctl_hook(cmd, argp);
896 up(&dlci_ioctl_mutex);
897 }
898 break;
899 default:
900 err = sock->ops->ioctl(sock, cmd, arg);
901 break;
902 }
903 return err;
904}
905
906int sock_create_lite(int family, int type, int protocol, struct socket **res)
907{
908 int err;
909 struct socket *sock = NULL;
910
911 err = security_socket_create(family, type, protocol, 1);
912 if (err)
913 goto out;
914
915 sock = sock_alloc();
916 if (!sock) {
917 err = -ENOMEM;
918 goto out;
919 }
920
921 security_socket_post_create(sock, family, type, protocol, 1);
922 sock->type = type;
923out:
924 *res = sock;
925 return err;
926}
927
928/* No kernel lock held - perfect */
929static unsigned int sock_poll(struct file *file, poll_table * wait)
930{
931 struct socket *sock;
932
933 /*
934 * We can't return errors to poll, so it's either yes or no.
935 */
936 sock = SOCKET_I(file->f_dentry->d_inode);
937 return sock->ops->poll(file, sock, wait);
938}
939
940static int sock_mmap(struct file * file, struct vm_area_struct * vma)
941{
942 struct socket *sock = SOCKET_I(file->f_dentry->d_inode);
943
944 return sock->ops->mmap(file, sock, vma);
945}
946
947int sock_close(struct inode *inode, struct file *filp)
948{
949 /*
950 * It was possible the inode is NULL we were
951 * closing an unfinished socket.
952 */
953
954 if (!inode)
955 {
956 printk(KERN_DEBUG "sock_close: NULL inode\n");
957 return 0;
958 }
959 sock_fasync(-1, filp, 0);
960 sock_release(SOCKET_I(inode));
961 return 0;
962}
963
964/*
965 * Update the socket async list
966 *
967 * Fasync_list locking strategy.
968 *
969 * 1. fasync_list is modified only under process context socket lock
970 * i.e. under semaphore.
971 * 2. fasync_list is used under read_lock(&sk->sk_callback_lock)
972 * or under socket lock.
973 * 3. fasync_list can be used from softirq context, so that
974 * modification under socket lock have to be enhanced with
975 * write_lock_bh(&sk->sk_callback_lock).
976 * --ANK (990710)
977 */
978
979static int sock_fasync(int fd, struct file *filp, int on)
980{
981 struct fasync_struct *fa, *fna=NULL, **prev;
982 struct socket *sock;
983 struct sock *sk;
984
985 if (on)
986 {
987 fna=(struct fasync_struct *)kmalloc(sizeof(struct fasync_struct), GFP_KERNEL);
988 if(fna==NULL)
989 return -ENOMEM;
990 }
991
992 sock = SOCKET_I(filp->f_dentry->d_inode);
993
994 if ((sk=sock->sk) == NULL) {
995 kfree(fna);
996 return -EINVAL;
997 }
998
999 lock_sock(sk);
1000
1001 prev=&(sock->fasync_list);
1002
1003 for (fa=*prev; fa!=NULL; prev=&fa->fa_next,fa=*prev)
1004 if (fa->fa_file==filp)
1005 break;
1006
1007 if(on)
1008 {
1009 if(fa!=NULL)
1010 {
1011 write_lock_bh(&sk->sk_callback_lock);
1012 fa->fa_fd=fd;
1013 write_unlock_bh(&sk->sk_callback_lock);
1014
1015 kfree(fna);
1016 goto out;
1017 }
1018 fna->fa_file=filp;
1019 fna->fa_fd=fd;
1020 fna->magic=FASYNC_MAGIC;
1021 fna->fa_next=sock->fasync_list;
1022 write_lock_bh(&sk->sk_callback_lock);
1023 sock->fasync_list=fna;
1024 write_unlock_bh(&sk->sk_callback_lock);
1025 }
1026 else
1027 {
1028 if (fa!=NULL)
1029 {
1030 write_lock_bh(&sk->sk_callback_lock);
1031 *prev=fa->fa_next;
1032 write_unlock_bh(&sk->sk_callback_lock);
1033 kfree(fa);
1034 }
1035 }
1036
1037out:
1038 release_sock(sock->sk);
1039 return 0;
1040}
1041
1042/* This function may be called only under socket lock or callback_lock */
1043
1044int sock_wake_async(struct socket *sock, int how, int band)
1045{
1046 if (!sock || !sock->fasync_list)
1047 return -1;
1048 switch (how)
1049 {
1050 case 1:
1051
1052 if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags))
1053 break;
1054 goto call_kill;
1055 case 2:
1056 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags))
1057 break;
1058 /* fall through */
1059 case 0:
1060 call_kill:
1061 __kill_fasync(sock->fasync_list, SIGIO, band);
1062 break;
1063 case 3:
1064 __kill_fasync(sock->fasync_list, SIGURG, band);
1065 }
1066 return 0;
1067}
1068
1069static int __sock_create(int family, int type, int protocol, struct socket **res, int kern)
1070{
1071 int err;
1072 struct socket *sock;
1073
1074 /*
1075 * Check protocol is in range
1076 */
1077 if (family < 0 || family >= NPROTO)
1078 return -EAFNOSUPPORT;
1079 if (type < 0 || type >= SOCK_MAX)
1080 return -EINVAL;
1081
1082 /* Compatibility.
1083
1084 This uglymoron is moved from INET layer to here to avoid
1085 deadlock in module load.
1086 */
1087 if (family == PF_INET && type == SOCK_PACKET) {
1088 static int warned;
1089 if (!warned) {
1090 warned = 1;
1091 printk(KERN_INFO "%s uses obsolete (PF_INET,SOCK_PACKET)\n", current->comm);
1092 }
1093 family = PF_PACKET;
1094 }
1095
1096 err = security_socket_create(family, type, protocol, kern);
1097 if (err)
1098 return err;
1099
1100#if defined(CONFIG_KMOD)
1101 /* Attempt to load a protocol module if the find failed.
1102 *
1103 * 12/09/1996 Marcin: But! this makes REALLY only sense, if the user
1104 * requested real, full-featured networking support upon configuration.
1105 * Otherwise module support will break!
1106 */
1107 if (net_families[family]==NULL)
1108 {
1109 request_module("net-pf-%d",family);
1110 }
1111#endif
1112
1113 net_family_read_lock();
1114 if (net_families[family] == NULL) {
1115 err = -EAFNOSUPPORT;
1116 goto out;
1117 }
1118
1119/*
1120 * Allocate the socket and allow the family to set things up. if
1121 * the protocol is 0, the family is instructed to select an appropriate
1122 * default.
1123 */
1124
1125 if (!(sock = sock_alloc())) {
1126 printk(KERN_WARNING "socket: no more sockets\n");
1127 err = -ENFILE; /* Not exactly a match, but its the
1128 closest posix thing */
1129 goto out;
1130 }
1131
1132 sock->type = type;
1133
1134 /*
1135 * We will call the ->create function, that possibly is in a loadable
1136 * module, so we have to bump that loadable module refcnt first.
1137 */
1138 err = -EAFNOSUPPORT;
1139 if (!try_module_get(net_families[family]->owner))
1140 goto out_release;
1141
1142 if ((err = net_families[family]->create(sock, protocol)) < 0)
1143 goto out_module_put;
1144 /*
1145 * Now to bump the refcnt of the [loadable] module that owns this
1146 * socket at sock_release time we decrement its refcnt.
1147 */
1148 if (!try_module_get(sock->ops->owner)) {
1149 sock->ops = NULL;
1150 goto out_module_put;
1151 }
1152 /*
1153 * Now that we're done with the ->create function, the [loadable]
1154 * module can have its refcnt decremented
1155 */
1156 module_put(net_families[family]->owner);
1157 *res = sock;
1158 security_socket_post_create(sock, family, type, protocol, kern);
1159
1160out:
1161 net_family_read_unlock();
1162 return err;
1163out_module_put:
1164 module_put(net_families[family]->owner);
1165out_release:
1166 sock_release(sock);
1167 goto out;
1168}
1169
1170int sock_create(int family, int type, int protocol, struct socket **res)
1171{
1172 return __sock_create(family, type, protocol, res, 0);
1173}
1174
1175int sock_create_kern(int family, int type, int protocol, struct socket **res)
1176{
1177 return __sock_create(family, type, protocol, res, 1);
1178}
1179
1180asmlinkage long sys_socket(int family, int type, int protocol)
1181{
1182 int retval;
1183 struct socket *sock;
1184
1185 retval = sock_create(family, type, protocol, &sock);
1186 if (retval < 0)
1187 goto out;
1188
1189 retval = sock_map_fd(sock);
1190 if (retval < 0)
1191 goto out_release;
1192
1193out:
1194 /* It may be already another descriptor 8) Not kernel problem. */
1195 return retval;
1196
1197out_release:
1198 sock_release(sock);
1199 return retval;
1200}
1201
1202/*
1203 * Create a pair of connected sockets.
1204 */
1205
1206asmlinkage long sys_socketpair(int family, int type, int protocol, int __user *usockvec)
1207{
1208 struct socket *sock1, *sock2;
1209 int fd1, fd2, err;
1210
1211 /*
1212 * Obtain the first socket and check if the underlying protocol
1213 * supports the socketpair call.
1214 */
1215
1216 err = sock_create(family, type, protocol, &sock1);
1217 if (err < 0)
1218 goto out;
1219
1220 err = sock_create(family, type, protocol, &sock2);
1221 if (err < 0)
1222 goto out_release_1;
1223
1224 err = sock1->ops->socketpair(sock1, sock2);
1225 if (err < 0)
1226 goto out_release_both;
1227
1228 fd1 = fd2 = -1;
1229
1230 err = sock_map_fd(sock1);
1231 if (err < 0)
1232 goto out_release_both;
1233 fd1 = err;
1234
1235 err = sock_map_fd(sock2);
1236 if (err < 0)
1237 goto out_close_1;
1238 fd2 = err;
1239
1240 /* fd1 and fd2 may be already another descriptors.
1241 * Not kernel problem.
1242 */
1243
1244 err = put_user(fd1, &usockvec[0]);
1245 if (!err)
1246 err = put_user(fd2, &usockvec[1]);
1247 if (!err)
1248 return 0;
1249
1250 sys_close(fd2);
1251 sys_close(fd1);
1252 return err;
1253
1254out_close_1:
1255 sock_release(sock2);
1256 sys_close(fd1);
1257 return err;
1258
1259out_release_both:
1260 sock_release(sock2);
1261out_release_1:
1262 sock_release(sock1);
1263out:
1264 return err;
1265}
1266
1267
1268/*
1269 * Bind a name to a socket. Nothing much to do here since it's
1270 * the protocol's responsibility to handle the local address.
1271 *
1272 * We move the socket address to kernel space before we call
1273 * the protocol layer (having also checked the address is ok).
1274 */
1275
1276asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)
1277{
1278 struct socket *sock;
1279 char address[MAX_SOCK_ADDR];
1280 int err;
1281
1282 if((sock = sockfd_lookup(fd,&err))!=NULL)
1283 {
1284 if((err=move_addr_to_kernel(umyaddr,addrlen,address))>=0) {
1285 err = security_socket_bind(sock, (struct sockaddr *)address, addrlen);
1286 if (err) {
1287 sockfd_put(sock);
1288 return err;
1289 }
1290 err = sock->ops->bind(sock, (struct sockaddr *)address, addrlen);
1291 }
1292 sockfd_put(sock);
1293 }
1294 return err;
1295}
1296
1297
1298/*
1299 * Perform a listen. Basically, we allow the protocol to do anything
1300 * necessary for a listen, and if that works, we mark the socket as
1301 * ready for listening.
1302 */
1303
1304int sysctl_somaxconn = SOMAXCONN;
1305
1306asmlinkage long sys_listen(int fd, int backlog)
1307{
1308 struct socket *sock;
1309 int err;
1310
1311 if ((sock = sockfd_lookup(fd, &err)) != NULL) {
1312 if ((unsigned) backlog > sysctl_somaxconn)
1313 backlog = sysctl_somaxconn;
1314
1315 err = security_socket_listen(sock, backlog);
1316 if (err) {
1317 sockfd_put(sock);
1318 return err;
1319 }
1320
1321 err=sock->ops->listen(sock, backlog);
1322 sockfd_put(sock);
1323 }
1324 return err;
1325}
1326
1327
1328/*
1329 * For accept, we attempt to create a new socket, set up the link
1330 * with the client, wake up the client, then return the new
1331 * connected fd. We collect the address of the connector in kernel
1332 * space and move it to user at the very end. This is unclean because
1333 * we open the socket then return an error.
1334 *
1335 * 1003.1g adds the ability to recvmsg() to query connection pending
1336 * status to recvmsg. We need to add that support in a way thats
1337 * clean when we restucture accept also.
1338 */
1339
1340asmlinkage long sys_accept(int fd, struct sockaddr __user *upeer_sockaddr, int __user *upeer_addrlen)
1341{
1342 struct socket *sock, *newsock;
1343 int err, len;
1344 char address[MAX_SOCK_ADDR];
1345
1346 sock = sockfd_lookup(fd, &err);
1347 if (!sock)
1348 goto out;
1349
1350 err = -ENFILE;
1351 if (!(newsock = sock_alloc()))
1352 goto out_put;
1353
1354 newsock->type = sock->type;
1355 newsock->ops = sock->ops;
1356
1357 err = security_socket_accept(sock, newsock);
1358 if (err)
1359 goto out_release;
1360
1361 /*
1362 * We don't need try_module_get here, as the listening socket (sock)
1363 * has the protocol module (sock->ops->owner) held.
1364 */
1365 __module_get(newsock->ops->owner);
1366
1367 err = sock->ops->accept(sock, newsock, sock->file->f_flags);
1368 if (err < 0)
1369 goto out_release;
1370
1371 if (upeer_sockaddr) {
1372 if(newsock->ops->getname(newsock, (struct sockaddr *)address, &len, 2)<0) {
1373 err = -ECONNABORTED;
1374 goto out_release;
1375 }
1376 err = move_addr_to_user(address, len, upeer_sockaddr, upeer_addrlen);
1377 if (err < 0)
1378 goto out_release;
1379 }
1380
1381 /* File flags are not inherited via accept() unlike another OSes. */
1382
1383 if ((err = sock_map_fd(newsock)) < 0)
1384 goto out_release;
1385
1386 security_socket_post_accept(sock, newsock);
1387
1388out_put:
1389 sockfd_put(sock);
1390out:
1391 return err;
1392out_release:
1393 sock_release(newsock);
1394 goto out_put;
1395}
1396
1397
1398/*
1399 * Attempt to connect to a socket with the server address. The address
1400 * is in user space so we verify it is OK and move it to kernel space.
1401 *
1402 * For 1003.1g we need to add clean support for a bind to AF_UNSPEC to
1403 * break bindings
1404 *
1405 * NOTE: 1003.1g draft 6.3 is broken with respect to AX.25/NetROM and
1406 * other SEQPACKET protocols that take time to connect() as it doesn't
1407 * include the -EINPROGRESS status for such sockets.
1408 */
1409
1410asmlinkage long sys_connect(int fd, struct sockaddr __user *uservaddr, int addrlen)
1411{
1412 struct socket *sock;
1413 char address[MAX_SOCK_ADDR];
1414 int err;
1415
1416 sock = sockfd_lookup(fd, &err);
1417 if (!sock)
1418 goto out;
1419 err = move_addr_to_kernel(uservaddr, addrlen, address);
1420 if (err < 0)
1421 goto out_put;
1422
1423 err = security_socket_connect(sock, (struct sockaddr *)address, addrlen);
1424 if (err)
1425 goto out_put;
1426
1427 err = sock->ops->connect(sock, (struct sockaddr *) address, addrlen,
1428 sock->file->f_flags);
1429out_put:
1430 sockfd_put(sock);
1431out:
1432 return err;
1433}
1434
1435/*
1436 * Get the local address ('name') of a socket object. Move the obtained
1437 * name to user space.
1438 */
1439
1440asmlinkage long sys_getsockname(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1441{
1442 struct socket *sock;
1443 char address[MAX_SOCK_ADDR];
1444 int len, err;
1445
1446 sock = sockfd_lookup(fd, &err);
1447 if (!sock)
1448 goto out;
1449
1450 err = security_socket_getsockname(sock);
1451 if (err)
1452 goto out_put;
1453
1454 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 0);
1455 if (err)
1456 goto out_put;
1457 err = move_addr_to_user(address, len, usockaddr, usockaddr_len);
1458
1459out_put:
1460 sockfd_put(sock);
1461out:
1462 return err;
1463}
1464
1465/*
1466 * Get the remote address ('name') of a socket object. Move the obtained
1467 * name to user space.
1468 */
1469
1470asmlinkage long sys_getpeername(int fd, struct sockaddr __user *usockaddr, int __user *usockaddr_len)
1471{
1472 struct socket *sock;
1473 char address[MAX_SOCK_ADDR];
1474 int len, err;
1475
1476 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1477 {
1478 err = security_socket_getpeername(sock);
1479 if (err) {
1480 sockfd_put(sock);
1481 return err;
1482 }
1483
1484 err = sock->ops->getname(sock, (struct sockaddr *)address, &len, 1);
1485 if (!err)
1486 err=move_addr_to_user(address,len, usockaddr, usockaddr_len);
1487 sockfd_put(sock);
1488 }
1489 return err;
1490}
1491
1492/*
1493 * Send a datagram to a given address. We move the address into kernel
1494 * space and check the user space data area is readable before invoking
1495 * the protocol.
1496 */
1497
1498asmlinkage long sys_sendto(int fd, void __user * buff, size_t len, unsigned flags,
1499 struct sockaddr __user *addr, int addr_len)
1500{
1501 struct socket *sock;
1502 char address[MAX_SOCK_ADDR];
1503 int err;
1504 struct msghdr msg;
1505 struct iovec iov;
1506
1507 sock = sockfd_lookup(fd, &err);
1508 if (!sock)
1509 goto out;
1510 iov.iov_base=buff;
1511 iov.iov_len=len;
1512 msg.msg_name=NULL;
1513 msg.msg_iov=&iov;
1514 msg.msg_iovlen=1;
1515 msg.msg_control=NULL;
1516 msg.msg_controllen=0;
1517 msg.msg_namelen=0;
1518 if(addr)
1519 {
1520 err = move_addr_to_kernel(addr, addr_len, address);
1521 if (err < 0)
1522 goto out_put;
1523 msg.msg_name=address;
1524 msg.msg_namelen=addr_len;
1525 }
1526 if (sock->file->f_flags & O_NONBLOCK)
1527 flags |= MSG_DONTWAIT;
1528 msg.msg_flags = flags;
1529 err = sock_sendmsg(sock, &msg, len);
1530
1531out_put:
1532 sockfd_put(sock);
1533out:
1534 return err;
1535}
1536
1537/*
1538 * Send a datagram down a socket.
1539 */
1540
1541asmlinkage long sys_send(int fd, void __user * buff, size_t len, unsigned flags)
1542{
1543 return sys_sendto(fd, buff, len, flags, NULL, 0);
1544}
1545
1546/*
1547 * Receive a frame from the socket and optionally record the address of the
1548 * sender. We verify the buffers are writable and if needed move the
1549 * sender address from kernel to user space.
1550 */
1551
1552asmlinkage long sys_recvfrom(int fd, void __user * ubuf, size_t size, unsigned flags,
1553 struct sockaddr __user *addr, int __user *addr_len)
1554{
1555 struct socket *sock;
1556 struct iovec iov;
1557 struct msghdr msg;
1558 char address[MAX_SOCK_ADDR];
1559 int err,err2;
1560
1561 sock = sockfd_lookup(fd, &err);
1562 if (!sock)
1563 goto out;
1564
1565 msg.msg_control=NULL;
1566 msg.msg_controllen=0;
1567 msg.msg_iovlen=1;
1568 msg.msg_iov=&iov;
1569 iov.iov_len=size;
1570 iov.iov_base=ubuf;
1571 msg.msg_name=address;
1572 msg.msg_namelen=MAX_SOCK_ADDR;
1573 if (sock->file->f_flags & O_NONBLOCK)
1574 flags |= MSG_DONTWAIT;
1575 err=sock_recvmsg(sock, &msg, size, flags);
1576
1577 if(err >= 0 && addr != NULL)
1578 {
1579 err2=move_addr_to_user(address, msg.msg_namelen, addr, addr_len);
1580 if(err2<0)
1581 err=err2;
1582 }
1583 sockfd_put(sock);
1584out:
1585 return err;
1586}
1587
1588/*
1589 * Receive a datagram from a socket.
1590 */
1591
1592asmlinkage long sys_recv(int fd, void __user * ubuf, size_t size, unsigned flags)
1593{
1594 return sys_recvfrom(fd, ubuf, size, flags, NULL, NULL);
1595}
1596
1597/*
1598 * Set a socket option. Because we don't know the option lengths we have
1599 * to pass the user mode parameter for the protocols to sort out.
1600 */
1601
1602asmlinkage long sys_setsockopt(int fd, int level, int optname, char __user *optval, int optlen)
1603{
1604 int err;
1605 struct socket *sock;
1606
1607 if (optlen < 0)
1608 return -EINVAL;
1609
1610 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1611 {
1612 err = security_socket_setsockopt(sock,level,optname);
1613 if (err) {
1614 sockfd_put(sock);
1615 return err;
1616 }
1617
1618 if (level == SOL_SOCKET)
1619 err=sock_setsockopt(sock,level,optname,optval,optlen);
1620 else
1621 err=sock->ops->setsockopt(sock, level, optname, optval, optlen);
1622 sockfd_put(sock);
1623 }
1624 return err;
1625}
1626
1627/*
1628 * Get a socket option. Because we don't know the option lengths we have
1629 * to pass a user mode parameter for the protocols to sort out.
1630 */
1631
1632asmlinkage long sys_getsockopt(int fd, int level, int optname, char __user *optval, int __user *optlen)
1633{
1634 int err;
1635 struct socket *sock;
1636
1637 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1638 {
1639 err = security_socket_getsockopt(sock, level,
1640 optname);
1641 if (err) {
1642 sockfd_put(sock);
1643 return err;
1644 }
1645
1646 if (level == SOL_SOCKET)
1647 err=sock_getsockopt(sock,level,optname,optval,optlen);
1648 else
1649 err=sock->ops->getsockopt(sock, level, optname, optval, optlen);
1650 sockfd_put(sock);
1651 }
1652 return err;
1653}
1654
1655
1656/*
1657 * Shutdown a socket.
1658 */
1659
1660asmlinkage long sys_shutdown(int fd, int how)
1661{
1662 int err;
1663 struct socket *sock;
1664
1665 if ((sock = sockfd_lookup(fd, &err))!=NULL)
1666 {
1667 err = security_socket_shutdown(sock, how);
1668 if (err) {
1669 sockfd_put(sock);
1670 return err;
1671 }
1672
1673 err=sock->ops->shutdown(sock, how);
1674 sockfd_put(sock);
1675 }
1676 return err;
1677}
1678
1679/* A couple of helpful macros for getting the address of the 32/64 bit
1680 * fields which are the same type (int / unsigned) on our platforms.
1681 */
1682#define COMPAT_MSG(msg, member) ((MSG_CMSG_COMPAT & flags) ? &msg##_compat->member : &msg->member)
1683#define COMPAT_NAMELEN(msg) COMPAT_MSG(msg, msg_namelen)
1684#define COMPAT_FLAGS(msg) COMPAT_MSG(msg, msg_flags)
1685
1686
1687/*
1688 * BSD sendmsg interface
1689 */
1690
1691asmlinkage long sys_sendmsg(int fd, struct msghdr __user *msg, unsigned flags)
1692{
1693 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1694 struct socket *sock;
1695 char address[MAX_SOCK_ADDR];
1696 struct iovec iovstack[UIO_FASTIOV], *iov = iovstack;
1697 unsigned char ctl[sizeof(struct cmsghdr) + 20]; /* 20 is size of ipv6_pktinfo */
1698 unsigned char *ctl_buf = ctl;
1699 struct msghdr msg_sys;
1700 int err, ctl_len, iov_size, total_len;
1701
1702 err = -EFAULT;
1703 if (MSG_CMSG_COMPAT & flags) {
1704 if (get_compat_msghdr(&msg_sys, msg_compat))
1705 return -EFAULT;
1706 } else if (copy_from_user(&msg_sys, msg, sizeof(struct msghdr)))
1707 return -EFAULT;
1708
1709 sock = sockfd_lookup(fd, &err);
1710 if (!sock)
1711 goto out;
1712
1713 /* do not move before msg_sys is valid */
1714 err = -EMSGSIZE;
1715 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1716 goto out_put;
1717
1718 /* Check whether to allocate the iovec area*/
1719 err = -ENOMEM;
1720 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1721 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1722 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1723 if (!iov)
1724 goto out_put;
1725 }
1726
1727 /* This will also move the address data into kernel space */
1728 if (MSG_CMSG_COMPAT & flags) {
1729 err = verify_compat_iovec(&msg_sys, iov, address, VERIFY_READ);
1730 } else
1731 err = verify_iovec(&msg_sys, iov, address, VERIFY_READ);
1732 if (err < 0)
1733 goto out_freeiov;
1734 total_len = err;
1735
1736 err = -ENOBUFS;
1737
1738 if (msg_sys.msg_controllen > INT_MAX)
1739 goto out_freeiov;
1740 ctl_len = msg_sys.msg_controllen;
1741 if ((MSG_CMSG_COMPAT & flags) && ctl_len) {
1742 err = cmsghdr_from_user_compat_to_kern(&msg_sys, ctl, sizeof(ctl));
1743 if (err)
1744 goto out_freeiov;
1745 ctl_buf = msg_sys.msg_control;
1746 } else if (ctl_len) {
1747 if (ctl_len > sizeof(ctl))
1748 {
1749 ctl_buf = sock_kmalloc(sock->sk, ctl_len, GFP_KERNEL);
1750 if (ctl_buf == NULL)
1751 goto out_freeiov;
1752 }
1753 err = -EFAULT;
1754 /*
1755 * Careful! Before this, msg_sys.msg_control contains a user pointer.
1756 * Afterwards, it will be a kernel pointer. Thus the compiler-assisted
1757 * checking falls down on this.
1758 */
1759 if (copy_from_user(ctl_buf, (void __user *) msg_sys.msg_control, ctl_len))
1760 goto out_freectl;
1761 msg_sys.msg_control = ctl_buf;
1762 }
1763 msg_sys.msg_flags = flags;
1764
1765 if (sock->file->f_flags & O_NONBLOCK)
1766 msg_sys.msg_flags |= MSG_DONTWAIT;
1767 err = sock_sendmsg(sock, &msg_sys, total_len);
1768
1769out_freectl:
1770 if (ctl_buf != ctl)
1771 sock_kfree_s(sock->sk, ctl_buf, ctl_len);
1772out_freeiov:
1773 if (iov != iovstack)
1774 sock_kfree_s(sock->sk, iov, iov_size);
1775out_put:
1776 sockfd_put(sock);
1777out:
1778 return err;
1779}
1780
1781/*
1782 * BSD recvmsg interface
1783 */
1784
1785asmlinkage long sys_recvmsg(int fd, struct msghdr __user *msg, unsigned int flags)
1786{
1787 struct compat_msghdr __user *msg_compat = (struct compat_msghdr __user *)msg;
1788 struct socket *sock;
1789 struct iovec iovstack[UIO_FASTIOV];
1790 struct iovec *iov=iovstack;
1791 struct msghdr msg_sys;
1792 unsigned long cmsg_ptr;
1793 int err, iov_size, total_len, len;
1794
1795 /* kernel mode address */
1796 char addr[MAX_SOCK_ADDR];
1797
1798 /* user mode address pointers */
1799 struct sockaddr __user *uaddr;
1800 int __user *uaddr_len;
1801
1802 if (MSG_CMSG_COMPAT & flags) {
1803 if (get_compat_msghdr(&msg_sys, msg_compat))
1804 return -EFAULT;
1805 } else
1806 if (copy_from_user(&msg_sys,msg,sizeof(struct msghdr)))
1807 return -EFAULT;
1808
1809 sock = sockfd_lookup(fd, &err);
1810 if (!sock)
1811 goto out;
1812
1813 err = -EMSGSIZE;
1814 if (msg_sys.msg_iovlen > UIO_MAXIOV)
1815 goto out_put;
1816
1817 /* Check whether to allocate the iovec area*/
1818 err = -ENOMEM;
1819 iov_size = msg_sys.msg_iovlen * sizeof(struct iovec);
1820 if (msg_sys.msg_iovlen > UIO_FASTIOV) {
1821 iov = sock_kmalloc(sock->sk, iov_size, GFP_KERNEL);
1822 if (!iov)
1823 goto out_put;
1824 }
1825
1826 /*
1827 * Save the user-mode address (verify_iovec will change the
1828 * kernel msghdr to use the kernel address space)
1829 */
1830
1831 uaddr = (void __user *) msg_sys.msg_name;
1832 uaddr_len = COMPAT_NAMELEN(msg);
1833 if (MSG_CMSG_COMPAT & flags) {
1834 err = verify_compat_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1835 } else
1836 err = verify_iovec(&msg_sys, iov, addr, VERIFY_WRITE);
1837 if (err < 0)
1838 goto out_freeiov;
1839 total_len=err;
1840
1841 cmsg_ptr = (unsigned long)msg_sys.msg_control;
1842 msg_sys.msg_flags = 0;
1843 if (MSG_CMSG_COMPAT & flags)
1844 msg_sys.msg_flags = MSG_CMSG_COMPAT;
1845
1846 if (sock->file->f_flags & O_NONBLOCK)
1847 flags |= MSG_DONTWAIT;
1848 err = sock_recvmsg(sock, &msg_sys, total_len, flags);
1849 if (err < 0)
1850 goto out_freeiov;
1851 len = err;
1852
1853 if (uaddr != NULL) {
1854 err = move_addr_to_user(addr, msg_sys.msg_namelen, uaddr, uaddr_len);
1855 if (err < 0)
1856 goto out_freeiov;
1857 }
1858 err = __put_user(msg_sys.msg_flags, COMPAT_FLAGS(msg));
1859 if (err)
1860 goto out_freeiov;
1861 if (MSG_CMSG_COMPAT & flags)
1862 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1863 &msg_compat->msg_controllen);
1864 else
1865 err = __put_user((unsigned long)msg_sys.msg_control-cmsg_ptr,
1866 &msg->msg_controllen);
1867 if (err)
1868 goto out_freeiov;
1869 err = len;
1870
1871out_freeiov:
1872 if (iov != iovstack)
1873 sock_kfree_s(sock->sk, iov, iov_size);
1874out_put:
1875 sockfd_put(sock);
1876out:
1877 return err;
1878}
1879
1880#ifdef __ARCH_WANT_SYS_SOCKETCALL
1881
1882/* Argument list sizes for sys_socketcall */
1883#define AL(x) ((x) * sizeof(unsigned long))
1884static unsigned char nargs[18]={AL(0),AL(3),AL(3),AL(3),AL(2),AL(3),
1885 AL(3),AL(3),AL(4),AL(4),AL(4),AL(6),
1886 AL(6),AL(2),AL(5),AL(5),AL(3),AL(3)};
1887#undef AL
1888
1889/*
1890 * System call vectors.
1891 *
1892 * Argument checking cleaned up. Saved 20% in size.
1893 * This function doesn't need to set the kernel lock because
1894 * it is set by the callees.
1895 */
1896
1897asmlinkage long sys_socketcall(int call, unsigned long __user *args)
1898{
1899 unsigned long a[6];
1900 unsigned long a0,a1;
1901 int err;
1902
1903 if(call<1||call>SYS_RECVMSG)
1904 return -EINVAL;
1905
1906 /* copy_from_user should be SMP safe. */
1907 if (copy_from_user(a, args, nargs[call]))
1908 return -EFAULT;
1909
1910 a0=a[0];
1911 a1=a[1];
1912
1913 switch(call)
1914 {
1915 case SYS_SOCKET:
1916 err = sys_socket(a0,a1,a[2]);
1917 break;
1918 case SYS_BIND:
1919 err = sys_bind(a0,(struct sockaddr __user *)a1, a[2]);
1920 break;
1921 case SYS_CONNECT:
1922 err = sys_connect(a0, (struct sockaddr __user *)a1, a[2]);
1923 break;
1924 case SYS_LISTEN:
1925 err = sys_listen(a0,a1);
1926 break;
1927 case SYS_ACCEPT:
1928 err = sys_accept(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1929 break;
1930 case SYS_GETSOCKNAME:
1931 err = sys_getsockname(a0,(struct sockaddr __user *)a1, (int __user *)a[2]);
1932 break;
1933 case SYS_GETPEERNAME:
1934 err = sys_getpeername(a0, (struct sockaddr __user *)a1, (int __user *)a[2]);
1935 break;
1936 case SYS_SOCKETPAIR:
1937 err = sys_socketpair(a0,a1, a[2], (int __user *)a[3]);
1938 break;
1939 case SYS_SEND:
1940 err = sys_send(a0, (void __user *)a1, a[2], a[3]);
1941 break;
1942 case SYS_SENDTO:
1943 err = sys_sendto(a0,(void __user *)a1, a[2], a[3],
1944 (struct sockaddr __user *)a[4], a[5]);
1945 break;
1946 case SYS_RECV:
1947 err = sys_recv(a0, (void __user *)a1, a[2], a[3]);
1948 break;
1949 case SYS_RECVFROM:
1950 err = sys_recvfrom(a0, (void __user *)a1, a[2], a[3],
1951 (struct sockaddr __user *)a[4], (int __user *)a[5]);
1952 break;
1953 case SYS_SHUTDOWN:
1954 err = sys_shutdown(a0,a1);
1955 break;
1956 case SYS_SETSOCKOPT:
1957 err = sys_setsockopt(a0, a1, a[2], (char __user *)a[3], a[4]);
1958 break;
1959 case SYS_GETSOCKOPT:
1960 err = sys_getsockopt(a0, a1, a[2], (char __user *)a[3], (int __user *)a[4]);
1961 break;
1962 case SYS_SENDMSG:
1963 err = sys_sendmsg(a0, (struct msghdr __user *) a1, a[2]);
1964 break;
1965 case SYS_RECVMSG:
1966 err = sys_recvmsg(a0, (struct msghdr __user *) a1, a[2]);
1967 break;
1968 default:
1969 err = -EINVAL;
1970 break;
1971 }
1972 return err;
1973}
1974
1975#endif /* __ARCH_WANT_SYS_SOCKETCALL */
1976
1977/*
1978 * This function is called by a protocol handler that wants to
1979 * advertise its address family, and have it linked into the
1980 * SOCKET module.
1981 */
1982
1983int sock_register(struct net_proto_family *ops)
1984{
1985 int err;
1986
1987 if (ops->family >= NPROTO) {
1988 printk(KERN_CRIT "protocol %d >= NPROTO(%d)\n", ops->family, NPROTO);
1989 return -ENOBUFS;
1990 }
1991 net_family_write_lock();
1992 err = -EEXIST;
1993 if (net_families[ops->family] == NULL) {
1994 net_families[ops->family]=ops;
1995 err = 0;
1996 }
1997 net_family_write_unlock();
1998 printk(KERN_INFO "NET: Registered protocol family %d\n",
1999 ops->family);
2000 return err;
2001}
2002
2003/*
2004 * This function is called by a protocol handler that wants to
2005 * remove its address family, and have it unlinked from the
2006 * SOCKET module.
2007 */
2008
2009int sock_unregister(int family)
2010{
2011 if (family < 0 || family >= NPROTO)
2012 return -1;
2013
2014 net_family_write_lock();
2015 net_families[family]=NULL;
2016 net_family_write_unlock();
2017 printk(KERN_INFO "NET: Unregistered protocol family %d\n",
2018 family);
2019 return 0;
2020}
2021
2022
2023extern void sk_init(void);
2024
2025void __init sock_init(void)
2026{
2027 /*
2028 * Initialize sock SLAB cache.
2029 */
2030
2031 sk_init();
2032
2033#ifdef SLAB_SKB
2034 /*
2035 * Initialize skbuff SLAB cache
2036 */
2037 skb_init();
2038#endif
2039
2040 /*
2041 * Initialize the protocols module.
2042 */
2043
2044 init_inodecache();
2045 register_filesystem(&sock_fs_type);
2046 sock_mnt = kern_mount(&sock_fs_type);
2047 /* The real protocol initialization is performed when
2048 * do_initcalls is run.
2049 */
2050
2051#ifdef CONFIG_NETFILTER
2052 netfilter_init();
2053#endif
2054}
2055
2056#ifdef CONFIG_PROC_FS
2057void socket_seq_show(struct seq_file *seq)
2058{
2059 int cpu;
2060 int counter = 0;
2061
2062 for (cpu = 0; cpu < NR_CPUS; cpu++)
2063 counter += per_cpu(sockets_in_use, cpu);
2064
2065 /* It can be negative, by the way. 8) */
2066 if (counter < 0)
2067 counter = 0;
2068
2069 seq_printf(seq, "sockets: used %d\n", counter);
2070}
2071#endif /* CONFIG_PROC_FS */
2072
2073/* ABI emulation layers need these two */
2074EXPORT_SYMBOL(move_addr_to_kernel);
2075EXPORT_SYMBOL(move_addr_to_user);
2076EXPORT_SYMBOL(sock_create);
2077EXPORT_SYMBOL(sock_create_kern);
2078EXPORT_SYMBOL(sock_create_lite);
2079EXPORT_SYMBOL(sock_map_fd);
2080EXPORT_SYMBOL(sock_recvmsg);
2081EXPORT_SYMBOL(sock_register);
2082EXPORT_SYMBOL(sock_release);
2083EXPORT_SYMBOL(sock_sendmsg);
2084EXPORT_SYMBOL(sock_unregister);
2085EXPORT_SYMBOL(sock_wake_async);
2086EXPORT_SYMBOL(sockfd_lookup);
2087EXPORT_SYMBOL(kernel_sendmsg);
2088EXPORT_SYMBOL(kernel_recvmsg);