blob: 32bf05e2996d8054b8ced182c41471d7c3e133e6 [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>
27
28#include <asm/unistd.h>
29
30int vfs_statfs(struct super_block *sb, struct kstatfs *buf)
31{
32 int retval = -ENODEV;
33
34 if (sb) {
35 retval = -ENOSYS;
36 if (sb->s_op->statfs) {
37 memset(buf, 0, sizeof(*buf));
38 retval = security_sb_statfs(sb);
39 if (retval)
40 return retval;
41 retval = sb->s_op->statfs(sb, buf);
42 if (retval == 0 && buf->f_frsize == 0)
43 buf->f_frsize = buf->f_bsize;
44 }
45 }
46 return retval;
47}
48
49EXPORT_SYMBOL(vfs_statfs);
50
51static int vfs_statfs_native(struct super_block *sb, struct statfs *buf)
52{
53 struct kstatfs st;
54 int retval;
55
56 retval = vfs_statfs(sb, &st);
57 if (retval)
58 return retval;
59
60 if (sizeof(*buf) == sizeof(st))
61 memcpy(buf, &st, sizeof(st));
62 else {
63 if (sizeof buf->f_blocks == 4) {
64 if ((st.f_blocks | st.f_bfree | st.f_bavail) &
65 0xffffffff00000000ULL)
66 return -EOVERFLOW;
67 /*
68 * f_files and f_ffree may be -1; it's okay to stuff
69 * that into 32 bits
70 */
71 if (st.f_files != -1 &&
72 (st.f_files & 0xffffffff00000000ULL))
73 return -EOVERFLOW;
74 if (st.f_ffree != -1 &&
75 (st.f_ffree & 0xffffffff00000000ULL))
76 return -EOVERFLOW;
77 }
78
79 buf->f_type = st.f_type;
80 buf->f_bsize = st.f_bsize;
81 buf->f_blocks = st.f_blocks;
82 buf->f_bfree = st.f_bfree;
83 buf->f_bavail = st.f_bavail;
84 buf->f_files = st.f_files;
85 buf->f_ffree = st.f_ffree;
86 buf->f_fsid = st.f_fsid;
87 buf->f_namelen = st.f_namelen;
88 buf->f_frsize = st.f_frsize;
89 memset(buf->f_spare, 0, sizeof(buf->f_spare));
90 }
91 return 0;
92}
93
94static int vfs_statfs64(struct super_block *sb, struct statfs64 *buf)
95{
96 struct kstatfs st;
97 int retval;
98
99 retval = vfs_statfs(sb, &st);
100 if (retval)
101 return retval;
102
103 if (sizeof(*buf) == sizeof(st))
104 memcpy(buf, &st, sizeof(st));
105 else {
106 buf->f_type = st.f_type;
107 buf->f_bsize = st.f_bsize;
108 buf->f_blocks = st.f_blocks;
109 buf->f_bfree = st.f_bfree;
110 buf->f_bavail = st.f_bavail;
111 buf->f_files = st.f_files;
112 buf->f_ffree = st.f_ffree;
113 buf->f_fsid = st.f_fsid;
114 buf->f_namelen = st.f_namelen;
115 buf->f_frsize = st.f_frsize;
116 memset(buf->f_spare, 0, sizeof(buf->f_spare));
117 }
118 return 0;
119}
120
121asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
122{
123 struct nameidata nd;
124 int error;
125
126 error = user_path_walk(path, &nd);
127 if (!error) {
128 struct statfs tmp;
129 error = vfs_statfs_native(nd.dentry->d_inode->i_sb, &tmp);
130 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
131 error = -EFAULT;
132 path_release(&nd);
133 }
134 return error;
135}
136
137
138asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
139{
140 struct nameidata nd;
141 long error;
142
143 if (sz != sizeof(*buf))
144 return -EINVAL;
145 error = user_path_walk(path, &nd);
146 if (!error) {
147 struct statfs64 tmp;
148 error = vfs_statfs64(nd.dentry->d_inode->i_sb, &tmp);
149 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
150 error = -EFAULT;
151 path_release(&nd);
152 }
153 return error;
154}
155
156
157asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
158{
159 struct file * file;
160 struct statfs tmp;
161 int error;
162
163 error = -EBADF;
164 file = fget(fd);
165 if (!file)
166 goto out;
167 error = vfs_statfs_native(file->f_dentry->d_inode->i_sb, &tmp);
168 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
169 error = -EFAULT;
170 fput(file);
171out:
172 return error;
173}
174
175asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
176{
177 struct file * file;
178 struct statfs64 tmp;
179 int error;
180
181 if (sz != sizeof(*buf))
182 return -EINVAL;
183
184 error = -EBADF;
185 file = fget(fd);
186 if (!file)
187 goto out;
188 error = vfs_statfs64(file->f_dentry->d_inode->i_sb, &tmp);
189 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
190 error = -EFAULT;
191 fput(file);
192out:
193 return error;
194}
195
196int do_truncate(struct dentry *dentry, loff_t length)
197{
198 int err;
199 struct iattr newattrs;
200
201 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
202 if (length < 0)
203 return -EINVAL;
204
205 newattrs.ia_size = length;
206 newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
207
208 down(&dentry->d_inode->i_sem);
209 err = notify_change(dentry, &newattrs);
210 up(&dentry->d_inode->i_sem);
211 return err;
212}
213
214static inline long do_sys_truncate(const char __user * path, loff_t length)
215{
216 struct nameidata nd;
217 struct inode * inode;
218 int error;
219
220 error = -EINVAL;
221 if (length < 0) /* sorry, but loff_t says... */
222 goto out;
223
224 error = user_path_walk(path, &nd);
225 if (error)
226 goto out;
227 inode = nd.dentry->d_inode;
228
229 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
230 error = -EISDIR;
231 if (S_ISDIR(inode->i_mode))
232 goto dput_and_out;
233
234 error = -EINVAL;
235 if (!S_ISREG(inode->i_mode))
236 goto dput_and_out;
237
238 error = permission(inode,MAY_WRITE,&nd);
239 if (error)
240 goto dput_and_out;
241
242 error = -EROFS;
243 if (IS_RDONLY(inode))
244 goto dput_and_out;
245
246 error = -EPERM;
247 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
248 goto dput_and_out;
249
250 /*
251 * Make sure that there are no leases.
252 */
253 error = break_lease(inode, FMODE_WRITE);
254 if (error)
255 goto dput_and_out;
256
257 error = get_write_access(inode);
258 if (error)
259 goto dput_and_out;
260
261 error = locks_verify_truncate(inode, NULL, length);
262 if (!error) {
263 DQUOT_INIT(inode);
264 error = do_truncate(nd.dentry, length);
265 }
266 put_write_access(inode);
267
268dput_and_out:
269 path_release(&nd);
270out:
271 return error;
272}
273
274asmlinkage long sys_truncate(const char __user * path, unsigned long length)
275{
276 /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
277 return do_sys_truncate(path, (long)length);
278}
279
280static inline long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
281{
282 struct inode * inode;
283 struct dentry *dentry;
284 struct file * file;
285 int error;
286
287 error = -EINVAL;
288 if (length < 0)
289 goto out;
290 error = -EBADF;
291 file = fget(fd);
292 if (!file)
293 goto out;
294
295 /* explicitly opened as large or we are on 64-bit box */
296 if (file->f_flags & O_LARGEFILE)
297 small = 0;
298
299 dentry = file->f_dentry;
300 inode = dentry->d_inode;
301 error = -EINVAL;
302 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
303 goto out_putf;
304
305 error = -EINVAL;
306 /* Cannot ftruncate over 2^31 bytes without large file support */
307 if (small && length > MAX_NON_LFS)
308 goto out_putf;
309
310 error = -EPERM;
311 if (IS_APPEND(inode))
312 goto out_putf;
313
314 error = locks_verify_truncate(inode, file, length);
315 if (!error)
316 error = do_truncate(dentry, length);
317out_putf:
318 fput(file);
319out:
320 return error;
321}
322
323asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
324{
325 return do_sys_ftruncate(fd, length, 1);
326}
327
328/* LFS versions of truncate are only needed on 32 bit machines */
329#if BITS_PER_LONG == 32
330asmlinkage long sys_truncate64(const char __user * path, loff_t length)
331{
332 return do_sys_truncate(path, length);
333}
334
335asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
336{
337 return do_sys_ftruncate(fd, length, 0);
338}
339#endif
340
341#ifdef __ARCH_WANT_SYS_UTIME
342
343/*
344 * sys_utime() can be implemented in user-level using sys_utimes().
345 * Is this for backwards compatibility? If so, why not move it
346 * into the appropriate arch directory (for those architectures that
347 * need it).
348 */
349
350/* If times==NULL, set access and modification to current time,
351 * must be owner or have write permission.
352 * Else, update from *times, must be owner or super user.
353 */
354asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
355{
356 int error;
357 struct nameidata nd;
358 struct inode * inode;
359 struct iattr newattrs;
360
361 error = user_path_walk(filename, &nd);
362 if (error)
363 goto out;
364 inode = nd.dentry->d_inode;
365
366 error = -EROFS;
367 if (IS_RDONLY(inode))
368 goto dput_and_out;
369
370 /* Don't worry, the checks are done in inode_change_ok() */
371 newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
372 if (times) {
373 error = -EPERM;
374 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
375 goto dput_and_out;
376
377 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
378 newattrs.ia_atime.tv_nsec = 0;
379 if (!error)
380 error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
381 newattrs.ia_mtime.tv_nsec = 0;
382 if (error)
383 goto dput_and_out;
384
385 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
386 } else {
387 error = -EACCES;
388 if (IS_IMMUTABLE(inode))
389 goto dput_and_out;
390
391 if (current->fsuid != inode->i_uid &&
392 (error = permission(inode,MAY_WRITE,&nd)) != 0)
393 goto dput_and_out;
394 }
395 down(&inode->i_sem);
396 error = notify_change(nd.dentry, &newattrs);
397 up(&inode->i_sem);
398dput_and_out:
399 path_release(&nd);
400out:
401 return error;
402}
403
404#endif
405
406/* If times==NULL, set access and modification to current time,
407 * must be owner or have write permission.
408 * Else, update from *times, must be owner or super user.
409 */
410long do_utimes(char __user * filename, struct timeval * times)
411{
412 int error;
413 struct nameidata nd;
414 struct inode * inode;
415 struct iattr newattrs;
416
417 error = user_path_walk(filename, &nd);
418
419 if (error)
420 goto out;
421 inode = nd.dentry->d_inode;
422
423 error = -EROFS;
424 if (IS_RDONLY(inode))
425 goto dput_and_out;
426
427 /* Don't worry, the checks are done in inode_change_ok() */
428 newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
429 if (times) {
430 error = -EPERM;
431 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
432 goto dput_and_out;
433
434 newattrs.ia_atime.tv_sec = times[0].tv_sec;
435 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
436 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
437 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
438 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
439 } else {
440 error = -EACCES;
441 if (IS_IMMUTABLE(inode))
442 goto dput_and_out;
443
444 if (current->fsuid != inode->i_uid &&
445 (error = permission(inode,MAY_WRITE,&nd)) != 0)
446 goto dput_and_out;
447 }
448 down(&inode->i_sem);
449 error = notify_change(nd.dentry, &newattrs);
450 up(&inode->i_sem);
451dput_and_out:
452 path_release(&nd);
453out:
454 return error;
455}
456
457asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
458{
459 struct timeval times[2];
460
461 if (utimes && copy_from_user(&times, utimes, sizeof(times)))
462 return -EFAULT;
463 return do_utimes(filename, utimes ? times : NULL);
464}
465
466
467/*
468 * access() needs to use the real uid/gid, not the effective uid/gid.
469 * We do this by temporarily clearing all FS-related capabilities and
470 * switching the fsuid/fsgid around to the real ones.
471 */
472asmlinkage long sys_access(const char __user * filename, int mode)
473{
474 struct nameidata nd;
475 int old_fsuid, old_fsgid;
476 kernel_cap_t old_cap;
477 int res;
478
479 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
480 return -EINVAL;
481
482 old_fsuid = current->fsuid;
483 old_fsgid = current->fsgid;
484 old_cap = current->cap_effective;
485
486 current->fsuid = current->uid;
487 current->fsgid = current->gid;
488
489 /*
490 * Clear the capabilities if we switch to a non-root user
491 *
492 * FIXME: There is a race here against sys_capset. The
493 * capabilities can change yet we will restore the old
494 * value below. We should hold task_capabilities_lock,
495 * but we cannot because user_path_walk can sleep.
496 */
497 if (current->uid)
498 cap_clear(current->cap_effective);
499 else
500 current->cap_effective = current->cap_permitted;
501
502 res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
503 if (!res) {
504 res = permission(nd.dentry->d_inode, mode, &nd);
505 /* SuS v2 requires we report a read only fs too */
506 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
507 && !special_file(nd.dentry->d_inode->i_mode))
508 res = -EROFS;
509 path_release(&nd);
510 }
511
512 current->fsuid = old_fsuid;
513 current->fsgid = old_fsgid;
514 current->cap_effective = old_cap;
515
516 return res;
517}
518
519asmlinkage long sys_chdir(const char __user * filename)
520{
521 struct nameidata nd;
522 int error;
523
524 error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
525 if (error)
526 goto out;
527
528 error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
529 if (error)
530 goto dput_and_out;
531
532 set_fs_pwd(current->fs, nd.mnt, nd.dentry);
533
534dput_and_out:
535 path_release(&nd);
536out:
537 return error;
538}
539
540asmlinkage long sys_fchdir(unsigned int fd)
541{
542 struct file *file;
543 struct dentry *dentry;
544 struct inode *inode;
545 struct vfsmount *mnt;
546 int error;
547
548 error = -EBADF;
549 file = fget(fd);
550 if (!file)
551 goto out;
552
553 dentry = file->f_dentry;
554 mnt = file->f_vfsmnt;
555 inode = dentry->d_inode;
556
557 error = -ENOTDIR;
558 if (!S_ISDIR(inode->i_mode))
559 goto out_putf;
560
561 error = permission(inode, MAY_EXEC, NULL);
562 if (!error)
563 set_fs_pwd(current->fs, mnt, dentry);
564out_putf:
565 fput(file);
566out:
567 return error;
568}
569
570asmlinkage long sys_chroot(const char __user * filename)
571{
572 struct nameidata nd;
573 int error;
574
575 error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
576 if (error)
577 goto out;
578
579 error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
580 if (error)
581 goto dput_and_out;
582
583 error = -EPERM;
584 if (!capable(CAP_SYS_CHROOT))
585 goto dput_and_out;
586
587 set_fs_root(current->fs, nd.mnt, nd.dentry);
588 set_fs_altroot();
589 error = 0;
590dput_and_out:
591 path_release(&nd);
592out:
593 return error;
594}
595
596asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
597{
598 struct inode * inode;
599 struct dentry * dentry;
600 struct file * file;
601 int err = -EBADF;
602 struct iattr newattrs;
603
604 file = fget(fd);
605 if (!file)
606 goto out;
607
608 dentry = file->f_dentry;
609 inode = dentry->d_inode;
610
611 err = -EROFS;
612 if (IS_RDONLY(inode))
613 goto out_putf;
614 err = -EPERM;
615 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
616 goto out_putf;
617 down(&inode->i_sem);
618 if (mode == (mode_t) -1)
619 mode = inode->i_mode;
620 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
621 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
622 err = notify_change(dentry, &newattrs);
623 up(&inode->i_sem);
624
625out_putf:
626 fput(file);
627out:
628 return err;
629}
630
631asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
632{
633 struct nameidata nd;
634 struct inode * inode;
635 int error;
636 struct iattr newattrs;
637
638 error = user_path_walk(filename, &nd);
639 if (error)
640 goto out;
641 inode = nd.dentry->d_inode;
642
643 error = -EROFS;
644 if (IS_RDONLY(inode))
645 goto dput_and_out;
646
647 error = -EPERM;
648 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
649 goto dput_and_out;
650
651 down(&inode->i_sem);
652 if (mode == (mode_t) -1)
653 mode = inode->i_mode;
654 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
655 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
656 error = notify_change(nd.dentry, &newattrs);
657 up(&inode->i_sem);
658
659dput_and_out:
660 path_release(&nd);
661out:
662 return error;
663}
664
665static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
666{
667 struct inode * inode;
668 int error;
669 struct iattr newattrs;
670
671 error = -ENOENT;
672 if (!(inode = dentry->d_inode)) {
673 printk(KERN_ERR "chown_common: NULL inode\n");
674 goto out;
675 }
676 error = -EROFS;
677 if (IS_RDONLY(inode))
678 goto out;
679 error = -EPERM;
680 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
681 goto out;
682 newattrs.ia_valid = ATTR_CTIME;
683 if (user != (uid_t) -1) {
684 newattrs.ia_valid |= ATTR_UID;
685 newattrs.ia_uid = user;
686 }
687 if (group != (gid_t) -1) {
688 newattrs.ia_valid |= ATTR_GID;
689 newattrs.ia_gid = group;
690 }
691 if (!S_ISDIR(inode->i_mode))
692 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
693 down(&inode->i_sem);
694 error = notify_change(dentry, &newattrs);
695 up(&inode->i_sem);
696out:
697 return error;
698}
699
700asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
701{
702 struct nameidata nd;
703 int error;
704
705 error = user_path_walk(filename, &nd);
706 if (!error) {
707 error = chown_common(nd.dentry, user, group);
708 path_release(&nd);
709 }
710 return error;
711}
712
713asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
714{
715 struct nameidata nd;
716 int error;
717
718 error = user_path_walk_link(filename, &nd);
719 if (!error) {
720 error = chown_common(nd.dentry, user, group);
721 path_release(&nd);
722 }
723 return error;
724}
725
726
727asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
728{
729 struct file * file;
730 int error = -EBADF;
731
732 file = fget(fd);
733 if (file) {
734 error = chown_common(file->f_dentry, user, group);
735 fput(file);
736 }
737 return error;
738}
739
740/*
741 * Note that while the flag value (low two bits) for sys_open means:
742 * 00 - read-only
743 * 01 - write-only
744 * 10 - read-write
745 * 11 - special
746 * it is changed into
747 * 00 - no permissions needed
748 * 01 - read-permission
749 * 10 - write-permission
750 * 11 - read-write
751 * for the internal routines (ie open_namei()/follow_link() etc). 00 is
752 * used by symlinks.
753 */
754struct file *filp_open(const char * filename, int flags, int mode)
755{
756 int namei_flags, error;
757 struct nameidata nd;
758
759 namei_flags = flags;
760 if ((namei_flags+1) & O_ACCMODE)
761 namei_flags++;
762 if (namei_flags & O_TRUNC)
763 namei_flags |= 2;
764
765 error = open_namei(filename, namei_flags, mode, &nd);
766 if (!error)
767 return dentry_open(nd.dentry, nd.mnt, flags);
768
769 return ERR_PTR(error);
770}
771
772EXPORT_SYMBOL(filp_open);
773
774struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
775{
776 struct file * f;
777 struct inode *inode;
778 int error;
779
780 error = -ENFILE;
781 f = get_empty_filp();
782 if (!f)
783 goto cleanup_dentry;
784 f->f_flags = flags;
785 f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
786 inode = dentry->d_inode;
787 if (f->f_mode & FMODE_WRITE) {
788 error = get_write_access(inode);
789 if (error)
790 goto cleanup_file;
791 }
792
793 f->f_mapping = inode->i_mapping;
794 f->f_dentry = dentry;
795 f->f_vfsmnt = mnt;
796 f->f_pos = 0;
797 f->f_op = fops_get(inode->i_fop);
798 file_move(f, &inode->i_sb->s_files);
799
800 if (f->f_op && f->f_op->open) {
801 error = f->f_op->open(inode,f);
802 if (error)
803 goto cleanup_all;
804 }
805 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
806
807 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
808
809 /* NB: we're sure to have correct a_ops only after f_op->open */
810 if (f->f_flags & O_DIRECT) {
Carsten Otteceffc072005-06-23 22:05:25 -0700811 if (!f->f_mapping->a_ops ||
812 ((!f->f_mapping->a_ops->direct_IO) &&
813 (!f->f_mapping->a_ops->get_xip_page))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 fput(f);
815 f = ERR_PTR(-EINVAL);
816 }
817 }
818
819 return f;
820
821cleanup_all:
822 fops_put(f->f_op);
823 if (f->f_mode & FMODE_WRITE)
824 put_write_access(inode);
825 file_kill(f);
826 f->f_dentry = NULL;
827 f->f_vfsmnt = NULL;
828cleanup_file:
829 put_filp(f);
830cleanup_dentry:
831 dput(dentry);
832 mntput(mnt);
833 return ERR_PTR(error);
834}
835
836EXPORT_SYMBOL(dentry_open);
837
838/*
839 * Find an empty file descriptor entry, and mark it busy.
840 */
841int get_unused_fd(void)
842{
843 struct files_struct * files = current->files;
844 int fd, error;
845
846 error = -EMFILE;
847 spin_lock(&files->file_lock);
848
849repeat:
850 fd = find_next_zero_bit(files->open_fds->fds_bits,
851 files->max_fdset,
852 files->next_fd);
853
854 /*
855 * N.B. For clone tasks sharing a files structure, this test
856 * will limit the total number of files that can be opened.
857 */
858 if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
859 goto out;
860
861 /* Do we need to expand the fd array or fd set? */
862 error = expand_files(files, fd);
863 if (error < 0)
864 goto out;
865
866 if (error) {
867 /*
868 * If we needed to expand the fs array we
869 * might have blocked - try again.
870 */
871 error = -EMFILE;
872 goto repeat;
873 }
874
875 FD_SET(fd, files->open_fds);
876 FD_CLR(fd, files->close_on_exec);
877 files->next_fd = fd + 1;
878#if 1
879 /* Sanity check */
880 if (files->fd[fd] != NULL) {
881 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
882 files->fd[fd] = NULL;
883 }
884#endif
885 error = fd;
886
887out:
888 spin_unlock(&files->file_lock);
889 return error;
890}
891
892EXPORT_SYMBOL(get_unused_fd);
893
894static inline void __put_unused_fd(struct files_struct *files, unsigned int fd)
895{
896 __FD_CLR(fd, files->open_fds);
897 if (fd < files->next_fd)
898 files->next_fd = fd;
899}
900
901void fastcall put_unused_fd(unsigned int fd)
902{
903 struct files_struct *files = current->files;
904 spin_lock(&files->file_lock);
905 __put_unused_fd(files, fd);
906 spin_unlock(&files->file_lock);
907}
908
909EXPORT_SYMBOL(put_unused_fd);
910
911/*
912 * Install a file pointer in the fd array.
913 *
914 * The VFS is full of places where we drop the files lock between
915 * setting the open_fds bitmap and installing the file in the file
916 * array. At any such point, we are vulnerable to a dup2() race
917 * installing a file in the array before us. We need to detect this and
918 * fput() the struct file we are about to overwrite in this case.
919 *
920 * It should never happen - if we allow dup2() do it, _really_ bad things
921 * will follow.
922 */
923
924void fastcall fd_install(unsigned int fd, struct file * file)
925{
926 struct files_struct *files = current->files;
927 spin_lock(&files->file_lock);
928 if (unlikely(files->fd[fd] != NULL))
929 BUG();
930 files->fd[fd] = file;
931 spin_unlock(&files->file_lock);
932}
933
934EXPORT_SYMBOL(fd_install);
935
936asmlinkage long sys_open(const char __user * filename, int flags, int mode)
937{
938 char * tmp;
Telemaque Ndizihiwefed2fc12005-06-23 00:10:33 -0700939 int fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Yoav Zachef3daed2005-06-23 00:09:58 -0700941 if (force_o_largefile())
942 flags |= O_LARGEFILE;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 tmp = getname(filename);
945 fd = PTR_ERR(tmp);
946 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}
962EXPORT_SYMBOL_GPL(sys_open);
963
964#ifndef __alpha__
965
966/*
967 * For backward compatibility? Maybe this should be moved
968 * into arch/i386 instead?
969 */
970asmlinkage long sys_creat(const char __user * pathname, int mode)
971{
972 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
973}
974
975#endif
976
977/*
978 * "id" is the POSIX thread ID. We use the
979 * files pointer for this..
980 */
981int filp_close(struct file *filp, fl_owner_t id)
982{
Christoph Lameter45778ca2005-06-23 00:10:17 -0700983 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985 if (!file_count(filp)) {
986 printk(KERN_ERR "VFS: Close: file count is 0\n");
Christoph Lameter45778ca2005-06-23 00:10:17 -0700987 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 }
989
Christoph Lameter45778ca2005-06-23 00:10:17 -0700990 if (filp->f_op && filp->f_op->flush)
991 retval = filp->f_op->flush(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 dnotify_flush(filp, id);
994 locks_remove_posix(filp, id);
995 fput(filp);
996 return retval;
997}
998
999EXPORT_SYMBOL(filp_close);
1000
1001/*
1002 * Careful here! We test whether the file pointer is NULL before
1003 * releasing the fd. This ensures that one clone task can't release
1004 * an fd while another clone is opening it.
1005 */
1006asmlinkage long sys_close(unsigned int fd)
1007{
1008 struct file * filp;
1009 struct files_struct *files = current->files;
1010
1011 spin_lock(&files->file_lock);
1012 if (fd >= files->max_fds)
1013 goto out_unlock;
1014 filp = files->fd[fd];
1015 if (!filp)
1016 goto out_unlock;
1017 files->fd[fd] = NULL;
1018 FD_CLR(fd, files->close_on_exec);
1019 __put_unused_fd(files, fd);
1020 spin_unlock(&files->file_lock);
1021 return filp_close(filp, files);
1022
1023out_unlock:
1024 spin_unlock(&files->file_lock);
1025 return -EBADF;
1026}
1027
1028EXPORT_SYMBOL(sys_close);
1029
1030/*
1031 * This routine simulates a hangup on the tty, to arrange that users
1032 * are given clean terminals at login time.
1033 */
1034asmlinkage long sys_vhangup(void)
1035{
1036 if (capable(CAP_SYS_TTY_CONFIG)) {
1037 tty_vhangup(current->signal->tty);
1038 return 0;
1039 }
1040 return -EPERM;
1041}
1042
1043/*
1044 * Called when an inode is about to be open.
1045 * We use this to disallow opening large files on 32bit systems if
1046 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1047 * on this flag in sys_open.
1048 */
1049int generic_file_open(struct inode * inode, struct file * filp)
1050{
1051 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1052 return -EFBIG;
1053 return 0;
1054}
1055
1056EXPORT_SYMBOL(generic_file_open);
1057
1058/*
1059 * This is used by subsystems that don't want seekable
1060 * file descriptors
1061 */
1062int nonseekable_open(struct inode *inode, struct file *filp)
1063{
1064 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1065 return 0;
1066}
1067
1068EXPORT_SYMBOL(nonseekable_open);