blob: 2fac58c51910cf250607a8f7161edc608d8c18cc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/utime.h>
10#include <linux/file.h>
11#include <linux/smp_lock.h>
12#include <linux/quotaops.h>
Robert Love0eeca282005-07-12 17:06:03 -040013#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/module.h>
15#include <linux/slab.h>
16#include <linux/tty.h>
17#include <linux/namei.h>
18#include <linux/backing-dev.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/vfs.h>
22#include <asm/uaccess.h>
23#include <linux/fs.h>
Yoav Zachef3daed2005-06-23 00:09:58 -070024#include <linux/personality.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/pagemap.h>
26#include <linux/syscalls.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070027#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#include <asm/unistd.h>
30
31int vfs_statfs(struct super_block *sb, struct kstatfs *buf)
32{
33 int retval = -ENODEV;
34
35 if (sb) {
36 retval = -ENOSYS;
37 if (sb->s_op->statfs) {
38 memset(buf, 0, sizeof(*buf));
39 retval = security_sb_statfs(sb);
40 if (retval)
41 return retval;
42 retval = sb->s_op->statfs(sb, buf);
43 if (retval == 0 && buf->f_frsize == 0)
44 buf->f_frsize = buf->f_bsize;
45 }
46 }
47 return retval;
48}
49
50EXPORT_SYMBOL(vfs_statfs);
51
52static int vfs_statfs_native(struct super_block *sb, struct statfs *buf)
53{
54 struct kstatfs st;
55 int retval;
56
57 retval = vfs_statfs(sb, &st);
58 if (retval)
59 return retval;
60
61 if (sizeof(*buf) == sizeof(st))
62 memcpy(buf, &st, sizeof(st));
63 else {
64 if (sizeof buf->f_blocks == 4) {
65 if ((st.f_blocks | st.f_bfree | st.f_bavail) &
66 0xffffffff00000000ULL)
67 return -EOVERFLOW;
68 /*
69 * f_files and f_ffree may be -1; it's okay to stuff
70 * that into 32 bits
71 */
72 if (st.f_files != -1 &&
73 (st.f_files & 0xffffffff00000000ULL))
74 return -EOVERFLOW;
75 if (st.f_ffree != -1 &&
76 (st.f_ffree & 0xffffffff00000000ULL))
77 return -EOVERFLOW;
78 }
79
80 buf->f_type = st.f_type;
81 buf->f_bsize = st.f_bsize;
82 buf->f_blocks = st.f_blocks;
83 buf->f_bfree = st.f_bfree;
84 buf->f_bavail = st.f_bavail;
85 buf->f_files = st.f_files;
86 buf->f_ffree = st.f_ffree;
87 buf->f_fsid = st.f_fsid;
88 buf->f_namelen = st.f_namelen;
89 buf->f_frsize = st.f_frsize;
90 memset(buf->f_spare, 0, sizeof(buf->f_spare));
91 }
92 return 0;
93}
94
95static int vfs_statfs64(struct super_block *sb, struct statfs64 *buf)
96{
97 struct kstatfs st;
98 int retval;
99
100 retval = vfs_statfs(sb, &st);
101 if (retval)
102 return retval;
103
104 if (sizeof(*buf) == sizeof(st))
105 memcpy(buf, &st, sizeof(st));
106 else {
107 buf->f_type = st.f_type;
108 buf->f_bsize = st.f_bsize;
109 buf->f_blocks = st.f_blocks;
110 buf->f_bfree = st.f_bfree;
111 buf->f_bavail = st.f_bavail;
112 buf->f_files = st.f_files;
113 buf->f_ffree = st.f_ffree;
114 buf->f_fsid = st.f_fsid;
115 buf->f_namelen = st.f_namelen;
116 buf->f_frsize = st.f_frsize;
117 memset(buf->f_spare, 0, sizeof(buf->f_spare));
118 }
119 return 0;
120}
121
122asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
123{
124 struct nameidata nd;
125 int error;
126
127 error = user_path_walk(path, &nd);
128 if (!error) {
129 struct statfs tmp;
130 error = vfs_statfs_native(nd.dentry->d_inode->i_sb, &tmp);
131 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
132 error = -EFAULT;
133 path_release(&nd);
134 }
135 return error;
136}
137
138
139asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
140{
141 struct nameidata nd;
142 long error;
143
144 if (sz != sizeof(*buf))
145 return -EINVAL;
146 error = user_path_walk(path, &nd);
147 if (!error) {
148 struct statfs64 tmp;
149 error = vfs_statfs64(nd.dentry->d_inode->i_sb, &tmp);
150 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
151 error = -EFAULT;
152 path_release(&nd);
153 }
154 return error;
155}
156
157
158asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
159{
160 struct file * file;
161 struct statfs tmp;
162 int error;
163
164 error = -EBADF;
165 file = fget(fd);
166 if (!file)
167 goto out;
168 error = vfs_statfs_native(file->f_dentry->d_inode->i_sb, &tmp);
169 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
170 error = -EFAULT;
171 fput(file);
172out:
173 return error;
174}
175
176asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
177{
178 struct file * file;
179 struct statfs64 tmp;
180 int error;
181
182 if (sz != sizeof(*buf))
183 return -EINVAL;
184
185 error = -EBADF;
186 file = fget(fd);
187 if (!file)
188 goto out;
189 error = vfs_statfs64(file->f_dentry->d_inode->i_sb, &tmp);
190 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
191 error = -EFAULT;
192 fput(file);
193out:
194 return error;
195}
196
197int do_truncate(struct dentry *dentry, loff_t length)
198{
199 int err;
200 struct iattr newattrs;
201
202 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
203 if (length < 0)
204 return -EINVAL;
205
206 newattrs.ia_size = length;
207 newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
208
209 down(&dentry->d_inode->i_sem);
210 err = notify_change(dentry, &newattrs);
211 up(&dentry->d_inode->i_sem);
212 return err;
213}
214
215static inline long do_sys_truncate(const char __user * path, loff_t length)
216{
217 struct nameidata nd;
218 struct inode * inode;
219 int error;
220
221 error = -EINVAL;
222 if (length < 0) /* sorry, but loff_t says... */
223 goto out;
224
225 error = user_path_walk(path, &nd);
226 if (error)
227 goto out;
228 inode = nd.dentry->d_inode;
229
230 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
231 error = -EISDIR;
232 if (S_ISDIR(inode->i_mode))
233 goto dput_and_out;
234
235 error = -EINVAL;
236 if (!S_ISREG(inode->i_mode))
237 goto dput_and_out;
238
239 error = permission(inode,MAY_WRITE,&nd);
240 if (error)
241 goto dput_and_out;
242
243 error = -EROFS;
244 if (IS_RDONLY(inode))
245 goto dput_and_out;
246
247 error = -EPERM;
248 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
249 goto dput_and_out;
250
251 /*
252 * Make sure that there are no leases.
253 */
254 error = break_lease(inode, FMODE_WRITE);
255 if (error)
256 goto dput_and_out;
257
258 error = get_write_access(inode);
259 if (error)
260 goto dput_and_out;
261
262 error = locks_verify_truncate(inode, NULL, length);
263 if (!error) {
264 DQUOT_INIT(inode);
265 error = do_truncate(nd.dentry, length);
266 }
267 put_write_access(inode);
268
269dput_and_out:
270 path_release(&nd);
271out:
272 return error;
273}
274
275asmlinkage long sys_truncate(const char __user * path, unsigned long length)
276{
277 /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
278 return do_sys_truncate(path, (long)length);
279}
280
281static inline long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
282{
283 struct inode * inode;
284 struct dentry *dentry;
285 struct file * file;
286 int error;
287
288 error = -EINVAL;
289 if (length < 0)
290 goto out;
291 error = -EBADF;
292 file = fget(fd);
293 if (!file)
294 goto out;
295
296 /* explicitly opened as large or we are on 64-bit box */
297 if (file->f_flags & O_LARGEFILE)
298 small = 0;
299
300 dentry = file->f_dentry;
301 inode = dentry->d_inode;
302 error = -EINVAL;
303 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
304 goto out_putf;
305
306 error = -EINVAL;
307 /* Cannot ftruncate over 2^31 bytes without large file support */
308 if (small && length > MAX_NON_LFS)
309 goto out_putf;
310
311 error = -EPERM;
312 if (IS_APPEND(inode))
313 goto out_putf;
314
315 error = locks_verify_truncate(inode, file, length);
316 if (!error)
317 error = do_truncate(dentry, length);
318out_putf:
319 fput(file);
320out:
321 return error;
322}
323
324asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
325{
326 return do_sys_ftruncate(fd, length, 1);
327}
328
329/* LFS versions of truncate are only needed on 32 bit machines */
330#if BITS_PER_LONG == 32
331asmlinkage long sys_truncate64(const char __user * path, loff_t length)
332{
333 return do_sys_truncate(path, length);
334}
335
336asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
337{
338 return do_sys_ftruncate(fd, length, 0);
339}
340#endif
341
342#ifdef __ARCH_WANT_SYS_UTIME
343
344/*
345 * sys_utime() can be implemented in user-level using sys_utimes().
346 * Is this for backwards compatibility? If so, why not move it
347 * into the appropriate arch directory (for those architectures that
348 * need it).
349 */
350
351/* If times==NULL, set access and modification to current time,
352 * must be owner or have write permission.
353 * Else, update from *times, must be owner or super user.
354 */
355asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
356{
357 int error;
358 struct nameidata nd;
359 struct inode * inode;
360 struct iattr newattrs;
361
362 error = user_path_walk(filename, &nd);
363 if (error)
364 goto out;
365 inode = nd.dentry->d_inode;
366
367 error = -EROFS;
368 if (IS_RDONLY(inode))
369 goto dput_and_out;
370
371 /* Don't worry, the checks are done in inode_change_ok() */
372 newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
373 if (times) {
374 error = -EPERM;
375 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
376 goto dput_and_out;
377
378 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
379 newattrs.ia_atime.tv_nsec = 0;
380 if (!error)
381 error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
382 newattrs.ia_mtime.tv_nsec = 0;
383 if (error)
384 goto dput_and_out;
385
386 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
387 } else {
388 error = -EACCES;
389 if (IS_IMMUTABLE(inode))
390 goto dput_and_out;
391
392 if (current->fsuid != inode->i_uid &&
393 (error = permission(inode,MAY_WRITE,&nd)) != 0)
394 goto dput_and_out;
395 }
396 down(&inode->i_sem);
397 error = notify_change(nd.dentry, &newattrs);
398 up(&inode->i_sem);
399dput_and_out:
400 path_release(&nd);
401out:
402 return error;
403}
404
405#endif
406
407/* If times==NULL, set access and modification to current time,
408 * must be owner or have write permission.
409 * Else, update from *times, must be owner or super user.
410 */
411long do_utimes(char __user * filename, struct timeval * times)
412{
413 int error;
414 struct nameidata nd;
415 struct inode * inode;
416 struct iattr newattrs;
417
418 error = user_path_walk(filename, &nd);
419
420 if (error)
421 goto out;
422 inode = nd.dentry->d_inode;
423
424 error = -EROFS;
425 if (IS_RDONLY(inode))
426 goto dput_and_out;
427
428 /* Don't worry, the checks are done in inode_change_ok() */
429 newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
430 if (times) {
431 error = -EPERM;
432 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
433 goto dput_and_out;
434
435 newattrs.ia_atime.tv_sec = times[0].tv_sec;
436 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
437 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
438 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
439 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
440 } else {
441 error = -EACCES;
442 if (IS_IMMUTABLE(inode))
443 goto dput_and_out;
444
445 if (current->fsuid != inode->i_uid &&
446 (error = permission(inode,MAY_WRITE,&nd)) != 0)
447 goto dput_and_out;
448 }
449 down(&inode->i_sem);
450 error = notify_change(nd.dentry, &newattrs);
451 up(&inode->i_sem);
452dput_and_out:
453 path_release(&nd);
454out:
455 return error;
456}
457
458asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
459{
460 struct timeval times[2];
461
462 if (utimes && copy_from_user(&times, utimes, sizeof(times)))
463 return -EFAULT;
464 return do_utimes(filename, utimes ? times : NULL);
465}
466
467
468/*
469 * access() needs to use the real uid/gid, not the effective uid/gid.
470 * We do this by temporarily clearing all FS-related capabilities and
471 * switching the fsuid/fsgid around to the real ones.
472 */
473asmlinkage long sys_access(const char __user * filename, int mode)
474{
475 struct nameidata nd;
476 int old_fsuid, old_fsgid;
477 kernel_cap_t old_cap;
478 int res;
479
480 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
481 return -EINVAL;
482
483 old_fsuid = current->fsuid;
484 old_fsgid = current->fsgid;
485 old_cap = current->cap_effective;
486
487 current->fsuid = current->uid;
488 current->fsgid = current->gid;
489
490 /*
491 * Clear the capabilities if we switch to a non-root user
492 *
493 * FIXME: There is a race here against sys_capset. The
494 * capabilities can change yet we will restore the old
495 * value below. We should hold task_capabilities_lock,
496 * but we cannot because user_path_walk can sleep.
497 */
498 if (current->uid)
499 cap_clear(current->cap_effective);
500 else
501 current->cap_effective = current->cap_permitted;
502
503 res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
504 if (!res) {
505 res = permission(nd.dentry->d_inode, mode, &nd);
506 /* SuS v2 requires we report a read only fs too */
507 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
508 && !special_file(nd.dentry->d_inode->i_mode))
509 res = -EROFS;
510 path_release(&nd);
511 }
512
513 current->fsuid = old_fsuid;
514 current->fsgid = old_fsgid;
515 current->cap_effective = old_cap;
516
517 return res;
518}
519
520asmlinkage long sys_chdir(const char __user * filename)
521{
522 struct nameidata nd;
523 int error;
524
525 error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
526 if (error)
527 goto out;
528
529 error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
530 if (error)
531 goto dput_and_out;
532
533 set_fs_pwd(current->fs, nd.mnt, nd.dentry);
534
535dput_and_out:
536 path_release(&nd);
537out:
538 return error;
539}
540
541asmlinkage long sys_fchdir(unsigned int fd)
542{
543 struct file *file;
544 struct dentry *dentry;
545 struct inode *inode;
546 struct vfsmount *mnt;
547 int error;
548
549 error = -EBADF;
550 file = fget(fd);
551 if (!file)
552 goto out;
553
554 dentry = file->f_dentry;
555 mnt = file->f_vfsmnt;
556 inode = dentry->d_inode;
557
558 error = -ENOTDIR;
559 if (!S_ISDIR(inode->i_mode))
560 goto out_putf;
561
562 error = permission(inode, MAY_EXEC, NULL);
563 if (!error)
564 set_fs_pwd(current->fs, mnt, dentry);
565out_putf:
566 fput(file);
567out:
568 return error;
569}
570
571asmlinkage long sys_chroot(const char __user * filename)
572{
573 struct nameidata nd;
574 int error;
575
576 error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
577 if (error)
578 goto out;
579
580 error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
581 if (error)
582 goto dput_and_out;
583
584 error = -EPERM;
585 if (!capable(CAP_SYS_CHROOT))
586 goto dput_and_out;
587
588 set_fs_root(current->fs, nd.mnt, nd.dentry);
589 set_fs_altroot();
590 error = 0;
591dput_and_out:
592 path_release(&nd);
593out:
594 return error;
595}
596
597asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
598{
599 struct inode * inode;
600 struct dentry * dentry;
601 struct file * file;
602 int err = -EBADF;
603 struct iattr newattrs;
604
605 file = fget(fd);
606 if (!file)
607 goto out;
608
609 dentry = file->f_dentry;
610 inode = dentry->d_inode;
611
612 err = -EROFS;
613 if (IS_RDONLY(inode))
614 goto out_putf;
615 err = -EPERM;
616 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
617 goto out_putf;
618 down(&inode->i_sem);
619 if (mode == (mode_t) -1)
620 mode = inode->i_mode;
621 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
622 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
623 err = notify_change(dentry, &newattrs);
624 up(&inode->i_sem);
625
626out_putf:
627 fput(file);
628out:
629 return err;
630}
631
632asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
633{
634 struct nameidata nd;
635 struct inode * inode;
636 int error;
637 struct iattr newattrs;
638
639 error = user_path_walk(filename, &nd);
640 if (error)
641 goto out;
642 inode = nd.dentry->d_inode;
643
644 error = -EROFS;
645 if (IS_RDONLY(inode))
646 goto dput_and_out;
647
648 error = -EPERM;
649 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
650 goto dput_and_out;
651
652 down(&inode->i_sem);
653 if (mode == (mode_t) -1)
654 mode = inode->i_mode;
655 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
656 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
657 error = notify_change(nd.dentry, &newattrs);
658 up(&inode->i_sem);
659
660dput_and_out:
661 path_release(&nd);
662out:
663 return error;
664}
665
666static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
667{
668 struct inode * inode;
669 int error;
670 struct iattr newattrs;
671
672 error = -ENOENT;
673 if (!(inode = dentry->d_inode)) {
674 printk(KERN_ERR "chown_common: NULL inode\n");
675 goto out;
676 }
677 error = -EROFS;
678 if (IS_RDONLY(inode))
679 goto out;
680 error = -EPERM;
681 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
682 goto out;
683 newattrs.ia_valid = ATTR_CTIME;
684 if (user != (uid_t) -1) {
685 newattrs.ia_valid |= ATTR_UID;
686 newattrs.ia_uid = user;
687 }
688 if (group != (gid_t) -1) {
689 newattrs.ia_valid |= ATTR_GID;
690 newattrs.ia_gid = group;
691 }
692 if (!S_ISDIR(inode->i_mode))
693 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
694 down(&inode->i_sem);
695 error = notify_change(dentry, &newattrs);
696 up(&inode->i_sem);
697out:
698 return error;
699}
700
701asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
702{
703 struct nameidata nd;
704 int error;
705
706 error = user_path_walk(filename, &nd);
707 if (!error) {
708 error = chown_common(nd.dentry, user, group);
709 path_release(&nd);
710 }
711 return error;
712}
713
714asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
715{
716 struct nameidata nd;
717 int error;
718
719 error = user_path_walk_link(filename, &nd);
720 if (!error) {
721 error = chown_common(nd.dentry, user, group);
722 path_release(&nd);
723 }
724 return error;
725}
726
727
728asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
729{
730 struct file * file;
731 int error = -EBADF;
732
733 file = fget(fd);
734 if (file) {
735 error = chown_common(file->f_dentry, user, group);
736 fput(file);
737 }
738 return error;
739}
740
741/*
742 * Note that while the flag value (low two bits) for sys_open means:
743 * 00 - read-only
744 * 01 - write-only
745 * 10 - read-write
746 * 11 - special
747 * it is changed into
748 * 00 - no permissions needed
749 * 01 - read-permission
750 * 10 - write-permission
751 * 11 - read-write
752 * for the internal routines (ie open_namei()/follow_link() etc). 00 is
753 * used by symlinks.
754 */
755struct file *filp_open(const char * filename, int flags, int mode)
756{
757 int namei_flags, error;
758 struct nameidata nd;
759
760 namei_flags = flags;
761 if ((namei_flags+1) & O_ACCMODE)
762 namei_flags++;
763 if (namei_flags & O_TRUNC)
764 namei_flags |= 2;
765
766 error = open_namei(filename, namei_flags, mode, &nd);
767 if (!error)
768 return dentry_open(nd.dentry, nd.mnt, flags);
769
770 return ERR_PTR(error);
771}
772
773EXPORT_SYMBOL(filp_open);
774
775struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
776{
777 struct file * f;
778 struct inode *inode;
779 int error;
780
781 error = -ENFILE;
782 f = get_empty_filp();
783 if (!f)
784 goto cleanup_dentry;
785 f->f_flags = flags;
786 f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
787 inode = dentry->d_inode;
788 if (f->f_mode & FMODE_WRITE) {
789 error = get_write_access(inode);
790 if (error)
791 goto cleanup_file;
792 }
793
794 f->f_mapping = inode->i_mapping;
795 f->f_dentry = dentry;
796 f->f_vfsmnt = mnt;
797 f->f_pos = 0;
798 f->f_op = fops_get(inode->i_fop);
799 file_move(f, &inode->i_sb->s_files);
800
801 if (f->f_op && f->f_op->open) {
802 error = f->f_op->open(inode,f);
803 if (error)
804 goto cleanup_all;
805 }
806 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
807
808 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
809
810 /* NB: we're sure to have correct a_ops only after f_op->open */
811 if (f->f_flags & O_DIRECT) {
Carsten Otteceffc072005-06-23 22:05:25 -0700812 if (!f->f_mapping->a_ops ||
813 ((!f->f_mapping->a_ops->direct_IO) &&
814 (!f->f_mapping->a_ops->get_xip_page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 fput(f);
816 f = ERR_PTR(-EINVAL);
817 }
818 }
819
820 return f;
821
822cleanup_all:
823 fops_put(f->f_op);
824 if (f->f_mode & FMODE_WRITE)
825 put_write_access(inode);
826 file_kill(f);
827 f->f_dentry = NULL;
828 f->f_vfsmnt = NULL;
829cleanup_file:
830 put_filp(f);
831cleanup_dentry:
832 dput(dentry);
833 mntput(mnt);
834 return ERR_PTR(error);
835}
836
837EXPORT_SYMBOL(dentry_open);
838
839/*
840 * Find an empty file descriptor entry, and mark it busy.
841 */
842int get_unused_fd(void)
843{
844 struct files_struct * files = current->files;
845 int fd, error;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700846 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 error = -EMFILE;
849 spin_lock(&files->file_lock);
850
851repeat:
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700852 fdt = files_fdtable(files);
853 fd = find_next_zero_bit(fdt->open_fds->fds_bits,
854 fdt->max_fdset,
855 fdt->next_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 /*
858 * N.B. For clone tasks sharing a files structure, this test
859 * will limit the total number of files that can be opened.
860 */
861 if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
862 goto out;
863
864 /* Do we need to expand the fd array or fd set? */
865 error = expand_files(files, fd);
866 if (error < 0)
867 goto out;
868
869 if (error) {
870 /*
871 * If we needed to expand the fs array we
872 * might have blocked - try again.
873 */
874 error = -EMFILE;
875 goto repeat;
876 }
877
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700878 FD_SET(fd, fdt->open_fds);
879 FD_CLR(fd, fdt->close_on_exec);
880 fdt->next_fd = fd + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881#if 1
882 /* Sanity check */
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700883 if (fdt->fd[fd] != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700885 fdt->fd[fd] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887#endif
888 error = fd;
889
890out:
891 spin_unlock(&files->file_lock);
892 return error;
893}
894
895EXPORT_SYMBOL(get_unused_fd);
896
897static inline void __put_unused_fd(struct files_struct *files, unsigned int fd)
898{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700899 struct fdtable *fdt = files_fdtable(files);
900 __FD_CLR(fd, fdt->open_fds);
901 if (fd < fdt->next_fd)
902 fdt->next_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903}
904
905void fastcall put_unused_fd(unsigned int fd)
906{
907 struct files_struct *files = current->files;
908 spin_lock(&files->file_lock);
909 __put_unused_fd(files, fd);
910 spin_unlock(&files->file_lock);
911}
912
913EXPORT_SYMBOL(put_unused_fd);
914
915/*
916 * Install a file pointer in the fd array.
917 *
918 * The VFS is full of places where we drop the files lock between
919 * setting the open_fds bitmap and installing the file in the file
920 * array. At any such point, we are vulnerable to a dup2() race
921 * installing a file in the array before us. We need to detect this and
922 * fput() the struct file we are about to overwrite in this case.
923 *
924 * It should never happen - if we allow dup2() do it, _really_ bad things
925 * will follow.
926 */
927
928void fastcall fd_install(unsigned int fd, struct file * file)
929{
930 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700931 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700933 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700934 BUG_ON(fdt->fd[fd] != NULL);
935 rcu_assign_pointer(fdt->fd[fd], file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 spin_unlock(&files->file_lock);
937}
938
939EXPORT_SYMBOL(fd_install);
940
Miklos Szeredie922efc2005-09-06 15:18:25 -0700941long do_sys_open(const char __user *filename, int flags, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Miklos Szeredie922efc2005-09-06 15:18:25 -0700943 char *tmp = getname(filename);
944 int fd = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 if (!IS_ERR(tmp)) {
947 fd = get_unused_fd();
948 if (fd >= 0) {
949 struct file *f = filp_open(tmp, flags, mode);
Telemaque Ndizihiwefed2fc12005-06-23 00:10:33 -0700950 if (IS_ERR(f)) {
951 put_unused_fd(fd);
952 fd = PTR_ERR(f);
953 } else {
Robert Love0eeca282005-07-12 17:06:03 -0400954 fsnotify_open(f->f_dentry);
Telemaque Ndizihiwefed2fc12005-06-23 00:10:33 -0700955 fd_install(fd, f);
956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 putname(tmp);
959 }
960 return fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
Miklos Szeredie922efc2005-09-06 15:18:25 -0700962
963asmlinkage long sys_open(const char __user *filename, int flags, int mode)
964{
965 if (force_o_largefile())
966 flags |= O_LARGEFILE;
967
968 return do_sys_open(filename, flags, mode);
969}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970EXPORT_SYMBOL_GPL(sys_open);
971
972#ifndef __alpha__
973
974/*
975 * For backward compatibility? Maybe this should be moved
976 * into arch/i386 instead?
977 */
978asmlinkage long sys_creat(const char __user * pathname, int mode)
979{
980 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
981}
982
983#endif
984
985/*
986 * "id" is the POSIX thread ID. We use the
987 * files pointer for this..
988 */
989int filp_close(struct file *filp, fl_owner_t id)
990{
Christoph Lameter45778ca2005-06-23 00:10:17 -0700991 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 if (!file_count(filp)) {
994 printk(KERN_ERR "VFS: Close: file count is 0\n");
Christoph Lameter45778ca2005-06-23 00:10:17 -0700995 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 }
997
Christoph Lameter45778ca2005-06-23 00:10:17 -0700998 if (filp->f_op && filp->f_op->flush)
999 retval = filp->f_op->flush(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 dnotify_flush(filp, id);
1002 locks_remove_posix(filp, id);
1003 fput(filp);
1004 return retval;
1005}
1006
1007EXPORT_SYMBOL(filp_close);
1008
1009/*
1010 * Careful here! We test whether the file pointer is NULL before
1011 * releasing the fd. This ensures that one clone task can't release
1012 * an fd while another clone is opening it.
1013 */
1014asmlinkage long sys_close(unsigned int fd)
1015{
1016 struct file * filp;
1017 struct files_struct *files = current->files;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001018 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001021 fdt = files_fdtable(files);
1022 if (fd >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 goto out_unlock;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001024 filp = fdt->fd[fd];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (!filp)
1026 goto out_unlock;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -07001027 rcu_assign_pointer(fdt->fd[fd], NULL);
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001028 FD_CLR(fd, fdt->close_on_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 __put_unused_fd(files, fd);
1030 spin_unlock(&files->file_lock);
1031 return filp_close(filp, files);
1032
1033out_unlock:
1034 spin_unlock(&files->file_lock);
1035 return -EBADF;
1036}
1037
1038EXPORT_SYMBOL(sys_close);
1039
1040/*
1041 * This routine simulates a hangup on the tty, to arrange that users
1042 * are given clean terminals at login time.
1043 */
1044asmlinkage long sys_vhangup(void)
1045{
1046 if (capable(CAP_SYS_TTY_CONFIG)) {
1047 tty_vhangup(current->signal->tty);
1048 return 0;
1049 }
1050 return -EPERM;
1051}
1052
1053/*
1054 * Called when an inode is about to be open.
1055 * We use this to disallow opening large files on 32bit systems if
1056 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1057 * on this flag in sys_open.
1058 */
1059int generic_file_open(struct inode * inode, struct file * filp)
1060{
1061 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1062 return -EFBIG;
1063 return 0;
1064}
1065
1066EXPORT_SYMBOL(generic_file_open);
1067
1068/*
1069 * This is used by subsystems that don't want seekable
1070 * file descriptors
1071 */
1072int nonseekable_open(struct inode *inode, struct file *filp)
1073{
1074 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1075 return 0;
1076}
1077
1078EXPORT_SYMBOL(nonseekable_open);