4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
173 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
183 int error
= check_acl(inode
, mask
, flags
);
184 if (error
!= -EAGAIN
)
188 if (in_group_p(inode
->i_gid
))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask
& ~mode
) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
217 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
222 * Do the basic POSIX ACL permission checks.
224 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
233 if (capable(CAP_DAC_OVERRIDE
))
237 * Searching includes executable on directories, else just read.
239 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
240 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
241 if (capable(CAP_DAC_READ_SEARCH
))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode
*inode
, int mask
)
261 if (mask
& MAY_WRITE
) {
262 umode_t mode
= inode
->i_mode
;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode
) &&
268 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode
))
278 if (inode
->i_op
->permission
)
279 retval
= inode
->i_op
->permission(inode
, mask
, 0);
281 retval
= generic_permission(inode
, mask
, 0,
282 inode
->i_op
->check_acl
);
287 retval
= devcgroup_inode_permission(inode
, mask
);
291 return security_inode_permission(inode
, mask
);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file
*file
, int mask
)
308 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode
* inode
)
331 spin_lock(&inode
->i_lock
);
332 if (atomic_read(&inode
->i_writecount
) < 0) {
333 spin_unlock(&inode
->i_lock
);
336 atomic_inc(&inode
->i_writecount
);
337 spin_unlock(&inode
->i_lock
);
342 int deny_write_access(struct file
* file
)
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
346 spin_lock(&inode
->i_lock
);
347 if (atomic_read(&inode
->i_writecount
) > 0) {
348 spin_unlock(&inode
->i_lock
);
351 atomic_dec(&inode
->i_writecount
);
352 spin_unlock(&inode
->i_lock
);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path
*path
)
368 EXPORT_SYMBOL(path_get
);
371 * path_get_long - get a long reference to a path
372 * @path: path to get the reference to
374 * Given a path increment the reference count to the dentry and the vfsmount.
376 void path_get_long(struct path
*path
)
378 mntget_long(path
->mnt
);
383 * path_put - put a reference to a path
384 * @path: path to put the reference to
386 * Given a path decrement the reference count to the dentry and the vfsmount.
388 void path_put(struct path
*path
)
393 EXPORT_SYMBOL(path_put
);
396 * path_put_long - put a long reference to a path
397 * @path: path to put the reference to
399 * Given a path decrement the reference count to the dentry and the vfsmount.
401 void path_put_long(struct path
*path
)
404 mntput_long(path
->mnt
);
408 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
409 * @nd: nameidata pathwalk data to drop
410 * Returns: 0 on success, -ECHILD on failure
412 * Path walking has 2 modes, rcu-walk and ref-walk (see
413 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
414 * to drop out of rcu-walk mode and take normal reference counts on dentries
415 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
416 * refcounts at the last known good point before rcu-walk got stuck, so
417 * ref-walk may continue from there. If this is not successful (eg. a seqcount
418 * has changed), then failure is returned and path walk restarts from the
419 * beginning in ref-walk mode.
421 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
422 * ref-walk. Must be called from rcu-walk context.
424 static int nameidata_drop_rcu(struct nameidata
*nd
)
426 struct fs_struct
*fs
= current
->fs
;
427 struct dentry
*dentry
= nd
->path
.dentry
;
429 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
431 spin_lock(&fs
->lock
);
432 if (nd
->root
.mnt
!= fs
->root
.mnt
||
433 nd
->root
.dentry
!= fs
->root
.dentry
)
436 spin_lock(&dentry
->d_lock
);
437 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
439 BUG_ON(nd
->inode
!= dentry
->d_inode
);
440 spin_unlock(&dentry
->d_lock
);
443 spin_unlock(&fs
->lock
);
445 mntget(nd
->path
.mnt
);
448 br_read_unlock(vfsmount_lock
);
449 nd
->flags
&= ~LOOKUP_RCU
;
452 spin_unlock(&dentry
->d_lock
);
455 spin_unlock(&fs
->lock
);
459 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
460 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
462 if (nd
->flags
& LOOKUP_RCU
)
463 return nameidata_drop_rcu(nd
);
468 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
469 * @nd: nameidata pathwalk data to drop
470 * @dentry: dentry to drop
471 * Returns: 0 on success, -ECHILD on failure
473 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
474 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
475 * @nd. Must be called from rcu-walk context.
477 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
479 struct fs_struct
*fs
= current
->fs
;
480 struct dentry
*parent
= nd
->path
.dentry
;
483 * It can be possible to revalidate the dentry that we started
484 * the path walk with. force_reval_path may also revalidate the
485 * dentry already committed to the nameidata.
487 if (unlikely(parent
== dentry
))
488 return nameidata_drop_rcu(nd
);
490 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
492 spin_lock(&fs
->lock
);
493 if (nd
->root
.mnt
!= fs
->root
.mnt
||
494 nd
->root
.dentry
!= fs
->root
.dentry
)
497 spin_lock(&parent
->d_lock
);
498 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
499 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
502 * If the sequence check on the child dentry passed, then the child has
503 * not been removed from its parent. This means the parent dentry must
504 * be valid and able to take a reference at this point.
506 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
507 BUG_ON(!parent
->d_count
);
509 spin_unlock(&dentry
->d_lock
);
510 spin_unlock(&parent
->d_lock
);
513 spin_unlock(&fs
->lock
);
515 mntget(nd
->path
.mnt
);
518 br_read_unlock(vfsmount_lock
);
519 nd
->flags
&= ~LOOKUP_RCU
;
522 spin_unlock(&dentry
->d_lock
);
523 spin_unlock(&parent
->d_lock
);
526 spin_unlock(&fs
->lock
);
530 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
531 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
533 if (nd
->flags
& LOOKUP_RCU
)
534 return nameidata_dentry_drop_rcu(nd
, dentry
);
539 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
540 * @nd: nameidata pathwalk data to drop
541 * Returns: 0 on success, -ECHILD on failure
543 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
544 * nd->path should be the final element of the lookup, so nd->root is discarded.
545 * Must be called from rcu-walk context.
547 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
549 struct dentry
*dentry
= nd
->path
.dentry
;
551 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
552 nd
->flags
&= ~LOOKUP_RCU
;
554 spin_lock(&dentry
->d_lock
);
555 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
557 BUG_ON(nd
->inode
!= dentry
->d_inode
);
558 spin_unlock(&dentry
->d_lock
);
560 mntget(nd
->path
.mnt
);
563 br_read_unlock(vfsmount_lock
);
568 spin_unlock(&dentry
->d_lock
);
570 br_read_unlock(vfsmount_lock
);
574 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
575 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
577 if (likely(nd
->flags
& LOOKUP_RCU
))
578 return nameidata_drop_rcu_last(nd
);
583 * release_open_intent - free up open intent resources
584 * @nd: pointer to nameidata
586 void release_open_intent(struct nameidata
*nd
)
588 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
589 put_filp(nd
->intent
.open
.file
);
591 fput(nd
->intent
.open
.file
);
595 * Call d_revalidate and handle filesystems that request rcu-walk
596 * to be dropped. This may be called and return in rcu-walk mode,
597 * regardless of success or error. If -ECHILD is returned, the caller
598 * must return -ECHILD back up the path walk stack so path walk may
599 * be restarted in ref-walk mode.
601 static int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
605 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
606 if (status
== -ECHILD
) {
607 if (nameidata_dentry_drop_rcu(nd
, dentry
))
609 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
615 static inline struct dentry
*
616 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
620 status
= d_revalidate(dentry
, nd
);
621 if (unlikely(status
<= 0)) {
623 * The dentry failed validation.
624 * If d_revalidate returned 0 attempt to invalidate
625 * the dentry otherwise d_revalidate is asking us
626 * to return a fail status.
629 /* If we're in rcu-walk, we don't have a ref */
630 if (!(nd
->flags
& LOOKUP_RCU
))
632 dentry
= ERR_PTR(status
);
635 /* Don't d_invalidate in rcu-walk mode */
636 if (nameidata_dentry_drop_rcu_maybe(nd
, dentry
))
637 return ERR_PTR(-ECHILD
);
638 if (!d_invalidate(dentry
)) {
647 static inline int need_reval_dot(struct dentry
*dentry
)
649 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
652 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
659 * force_reval_path - force revalidation of a dentry
661 * In some situations the path walking code will trust dentries without
662 * revalidating them. This causes problems for filesystems that depend on
663 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
664 * (which indicates that it's possible for the dentry to go stale), force
665 * a d_revalidate call before proceeding.
667 * Returns 0 if the revalidation was successful. If the revalidation fails,
668 * either return the error returned by d_revalidate or -ESTALE if the
669 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
670 * invalidate the dentry. It's up to the caller to handle putting references
671 * to the path if necessary.
674 force_reval_path(struct path
*path
, struct nameidata
*nd
)
677 struct dentry
*dentry
= path
->dentry
;
680 * only check on filesystems where it's possible for the dentry to
683 if (!need_reval_dot(dentry
))
686 status
= d_revalidate(dentry
, nd
);
691 /* Don't d_invalidate in rcu-walk mode */
692 if (nameidata_drop_rcu(nd
))
694 d_invalidate(dentry
);
701 * Short-cut version of permission(), for calling on directories
702 * during pathname resolution. Combines parts of permission()
703 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
705 * If appropriate, check DAC only. If not appropriate, or
706 * short-cut DAC fails, then call ->permission() to do more
707 * complete permission check.
709 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
713 if (inode
->i_op
->permission
) {
714 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
716 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
717 inode
->i_op
->check_acl
);
724 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
729 return security_inode_exec_permission(inode
, flags
);
732 static __always_inline
void set_root(struct nameidata
*nd
)
735 get_fs_root(current
->fs
, &nd
->root
);
738 static int link_path_walk(const char *, struct nameidata
*);
740 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
743 struct fs_struct
*fs
= current
->fs
;
747 seq
= read_seqcount_begin(&fs
->seq
);
749 } while (read_seqcount_retry(&fs
->seq
, seq
));
753 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
766 nd
->inode
= nd
->path
.dentry
->d_inode
;
768 ret
= link_path_walk(link
, nd
);
772 return PTR_ERR(link
);
775 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
778 if (path
->mnt
!= nd
->path
.mnt
)
782 static inline void path_to_nameidata(const struct path
*path
,
783 struct nameidata
*nd
)
785 if (!(nd
->flags
& LOOKUP_RCU
)) {
786 dput(nd
->path
.dentry
);
787 if (nd
->path
.mnt
!= path
->mnt
)
788 mntput(nd
->path
.mnt
);
790 nd
->path
.mnt
= path
->mnt
;
791 nd
->path
.dentry
= path
->dentry
;
794 static __always_inline
int
795 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
798 struct dentry
*dentry
= link
->dentry
;
800 touch_atime(link
->mnt
, dentry
);
801 nd_set_link(nd
, NULL
);
803 if (link
->mnt
!= nd
->path
.mnt
) {
804 path_to_nameidata(link
, nd
);
805 nd
->inode
= nd
->path
.dentry
->d_inode
;
810 nd
->last_type
= LAST_BIND
;
811 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
814 char *s
= nd_get_link(nd
);
817 error
= __vfs_follow_link(nd
, s
);
818 else if (nd
->last_type
== LAST_BIND
) {
819 error
= force_reval_path(&nd
->path
, nd
);
828 * This limits recursive symlink follows to 8, while
829 * limiting consecutive symlinks to 40.
831 * Without that kind of total limit, nasty chains of consecutive
832 * symlinks can cause almost arbitrarily long lookups.
834 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
838 if (current
->link_count
>= MAX_NESTED_LINKS
)
840 if (current
->total_link_count
>= 40)
842 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
844 err
= security_inode_follow_link(path
->dentry
, nd
);
847 current
->link_count
++;
848 current
->total_link_count
++;
850 err
= __do_follow_link(path
, nd
, &cookie
);
851 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
852 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
854 current
->link_count
--;
858 path_put_conditional(path
, nd
);
863 static int follow_up_rcu(struct path
*path
)
865 struct vfsmount
*parent
;
866 struct dentry
*mountpoint
;
868 parent
= path
->mnt
->mnt_parent
;
869 if (parent
== path
->mnt
)
871 mountpoint
= path
->mnt
->mnt_mountpoint
;
872 path
->dentry
= mountpoint
;
877 int follow_up(struct path
*path
)
879 struct vfsmount
*parent
;
880 struct dentry
*mountpoint
;
882 br_read_lock(vfsmount_lock
);
883 parent
= path
->mnt
->mnt_parent
;
884 if (parent
== path
->mnt
) {
885 br_read_unlock(vfsmount_lock
);
889 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
890 br_read_unlock(vfsmount_lock
);
892 path
->dentry
= mountpoint
;
899 * serialization is taken care of in namespace.c
901 static void __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
902 struct inode
**inode
)
904 while (d_mountpoint(path
->dentry
)) {
905 struct vfsmount
*mounted
;
906 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
910 path
->dentry
= mounted
->mnt_root
;
911 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
912 *inode
= path
->dentry
->d_inode
;
916 static int __follow_mount(struct path
*path
)
919 while (d_mountpoint(path
->dentry
)) {
920 struct vfsmount
*mounted
= lookup_mnt(path
);
927 path
->dentry
= dget(mounted
->mnt_root
);
933 static void follow_mount(struct path
*path
)
935 while (d_mountpoint(path
->dentry
)) {
936 struct vfsmount
*mounted
= lookup_mnt(path
);
942 path
->dentry
= dget(mounted
->mnt_root
);
946 int follow_down(struct path
*path
)
948 struct vfsmount
*mounted
;
950 mounted
= lookup_mnt(path
);
955 path
->dentry
= dget(mounted
->mnt_root
);
961 static int follow_dotdot_rcu(struct nameidata
*nd
)
963 struct inode
*inode
= nd
->inode
;
968 if (nd
->path
.dentry
== nd
->root
.dentry
&&
969 nd
->path
.mnt
== nd
->root
.mnt
) {
972 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
973 struct dentry
*old
= nd
->path
.dentry
;
974 struct dentry
*parent
= old
->d_parent
;
977 seq
= read_seqcount_begin(&parent
->d_seq
);
978 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
980 inode
= parent
->d_inode
;
981 nd
->path
.dentry
= parent
;
985 if (!follow_up_rcu(&nd
->path
))
987 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
988 inode
= nd
->path
.dentry
->d_inode
;
990 __follow_mount_rcu(nd
, &nd
->path
, &inode
);
996 static void follow_dotdot(struct nameidata
*nd
)
1001 struct dentry
*old
= nd
->path
.dentry
;
1003 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1004 nd
->path
.mnt
== nd
->root
.mnt
) {
1007 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1008 /* rare case of legitimate dget_parent()... */
1009 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1013 if (!follow_up(&nd
->path
))
1016 follow_mount(&nd
->path
);
1017 nd
->inode
= nd
->path
.dentry
->d_inode
;
1021 * Allocate a dentry with name and parent, and perform a parent
1022 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1023 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1024 * have verified that no child exists while under i_mutex.
1026 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1027 struct qstr
*name
, struct nameidata
*nd
)
1029 struct inode
*inode
= parent
->d_inode
;
1030 struct dentry
*dentry
;
1033 /* Don't create child dentry for a dead directory. */
1034 if (unlikely(IS_DEADDIR(inode
)))
1035 return ERR_PTR(-ENOENT
);
1037 dentry
= d_alloc(parent
, name
);
1038 if (unlikely(!dentry
))
1039 return ERR_PTR(-ENOMEM
);
1041 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1042 if (unlikely(old
)) {
1050 * It's more convoluted than I'd like it to be, but... it's still fairly
1051 * small and for now I'd prefer to have fast path as straight as possible.
1052 * It _is_ time-critical.
1054 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1055 struct path
*path
, struct inode
**inode
)
1057 struct vfsmount
*mnt
= nd
->path
.mnt
;
1058 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1061 * See if the low-level filesystem might want
1062 * to use its own hash..
1064 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1065 int err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1071 * Rename seqlock is not required here because in the off chance
1072 * of a false negative due to a concurrent rename, we're going to
1073 * do the non-racy lookup, below.
1075 if (nd
->flags
& LOOKUP_RCU
) {
1079 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1081 if (nameidata_drop_rcu(nd
))
1085 /* Memory barrier in read_seqcount_begin of child is enough */
1086 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1090 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1091 goto need_revalidate
;
1093 path
->dentry
= dentry
;
1094 __follow_mount_rcu(nd
, path
, inode
);
1096 dentry
= __d_lookup(parent
, name
);
1100 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1101 goto need_revalidate
;
1104 path
->dentry
= dentry
;
1105 __follow_mount(path
);
1106 *inode
= path
->dentry
->d_inode
;
1111 dir
= parent
->d_inode
;
1112 BUG_ON(nd
->inode
!= dir
);
1114 mutex_lock(&dir
->i_mutex
);
1116 * First re-do the cached lookup just in case it was created
1117 * while we waited for the directory semaphore, or the first
1118 * lookup failed due to an unrelated rename.
1120 * This could use version numbering or similar to avoid unnecessary
1121 * cache lookups, but then we'd have to do the first lookup in the
1122 * non-racy way. However in the common case here, everything should
1123 * be hot in cache, so would it be a big win?
1125 dentry
= d_lookup(parent
, name
);
1126 if (likely(!dentry
)) {
1127 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1128 mutex_unlock(&dir
->i_mutex
);
1134 * Uhhuh! Nasty case: the cache was re-populated while
1135 * we waited on the semaphore. Need to revalidate.
1137 mutex_unlock(&dir
->i_mutex
);
1141 dentry
= do_revalidate(dentry
, nd
);
1149 return PTR_ERR(dentry
);
1153 * This is a temporary kludge to deal with "automount" symlinks; proper
1154 * solution is to trigger them on follow_mount(), so that do_lookup()
1155 * would DTRT. To be killed before 2.6.34-final.
1157 static inline int follow_on_final(struct inode
*inode
, unsigned lookup_flags
)
1159 return inode
&& unlikely(inode
->i_op
->follow_link
) &&
1160 ((lookup_flags
& LOOKUP_FOLLOW
) || S_ISDIR(inode
->i_mode
));
1165 * This is the basic name resolution function, turning a pathname into
1166 * the final dentry. We expect 'base' to be positive and a directory.
1168 * Returns 0 and nd will have valid dentry and mnt on success.
1169 * Returns error and drops reference to input namei data on failure.
1171 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1175 unsigned int lookup_flags
= nd
->flags
;
1183 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1185 /* At this point we know we have a real path component. */
1187 struct inode
*inode
;
1192 nd
->flags
|= LOOKUP_CONTINUE
;
1193 if (nd
->flags
& LOOKUP_RCU
) {
1194 err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1195 if (err
== -ECHILD
) {
1196 if (nameidata_drop_rcu(nd
))
1202 err
= exec_permission(nd
->inode
, 0);
1208 c
= *(const unsigned char *)name
;
1210 hash
= init_name_hash();
1213 hash
= partial_name_hash(c
, hash
);
1214 c
= *(const unsigned char *)name
;
1215 } while (c
&& (c
!= '/'));
1216 this.len
= name
- (const char *) this.name
;
1217 this.hash
= end_name_hash(hash
);
1219 /* remove trailing slashes? */
1221 goto last_component
;
1222 while (*++name
== '/');
1224 goto last_with_slashes
;
1227 * "." and ".." are special - ".." especially so because it has
1228 * to be able to know about the current root directory and
1229 * parent relationships.
1231 if (this.name
[0] == '.') switch (this.len
) {
1235 if (this.name
[1] != '.')
1237 if (nd
->flags
& LOOKUP_RCU
) {
1238 if (follow_dotdot_rcu(nd
))
1246 /* This does the actual lookups.. */
1247 err
= do_lookup(nd
, &this, &next
, &inode
);
1254 if (inode
->i_op
->follow_link
) {
1255 /* We commonly drop rcu-walk here */
1256 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1258 BUG_ON(inode
!= next
.dentry
->d_inode
);
1259 err
= do_follow_link(&next
, nd
);
1262 nd
->inode
= nd
->path
.dentry
->d_inode
;
1267 path_to_nameidata(&next
, nd
);
1271 if (!nd
->inode
->i_op
->lookup
)
1274 /* here ends the main loop */
1277 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1279 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1280 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1281 if (lookup_flags
& LOOKUP_PARENT
)
1283 if (this.name
[0] == '.') switch (this.len
) {
1287 if (this.name
[1] != '.')
1289 if (nd
->flags
& LOOKUP_RCU
) {
1290 if (follow_dotdot_rcu(nd
))
1298 err
= do_lookup(nd
, &this, &next
, &inode
);
1301 if (follow_on_final(inode
, lookup_flags
)) {
1302 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1304 BUG_ON(inode
!= next
.dentry
->d_inode
);
1305 err
= do_follow_link(&next
, nd
);
1308 nd
->inode
= nd
->path
.dentry
->d_inode
;
1310 path_to_nameidata(&next
, nd
);
1316 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1318 if (!nd
->inode
->i_op
->lookup
)
1324 nd
->last_type
= LAST_NORM
;
1325 if (this.name
[0] != '.')
1328 nd
->last_type
= LAST_DOT
;
1329 else if (this.len
== 2 && this.name
[1] == '.')
1330 nd
->last_type
= LAST_DOTDOT
;
1335 * We bypassed the ordinary revalidation routines.
1336 * We may need to check the cached dentry for staleness.
1338 if (need_reval_dot(nd
->path
.dentry
)) {
1339 /* Note: we do not d_invalidate() */
1340 err
= d_revalidate(nd
->path
.dentry
, nd
);
1347 if (nameidata_drop_rcu_last_maybe(nd
))
1351 if (!(nd
->flags
& LOOKUP_RCU
))
1352 path_put_conditional(&next
, nd
);
1355 if (!(nd
->flags
& LOOKUP_RCU
))
1356 path_put(&nd
->path
);
1361 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1363 current
->total_link_count
= 0;
1365 return link_path_walk(name
, nd
);
1368 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1370 current
->total_link_count
= 0;
1372 return link_path_walk(name
, nd
);
1375 static int path_walk(const char *name
, struct nameidata
*nd
)
1377 struct path save
= nd
->path
;
1380 current
->total_link_count
= 0;
1382 /* make sure the stuff we saved doesn't go away */
1385 result
= link_path_walk(name
, nd
);
1386 if (result
== -ESTALE
) {
1387 /* nd->path had been dropped */
1388 current
->total_link_count
= 0;
1390 path_get(&nd
->path
);
1391 nd
->flags
|= LOOKUP_REVAL
;
1392 result
= link_path_walk(name
, nd
);
1400 static void path_finish_rcu(struct nameidata
*nd
)
1402 if (nd
->flags
& LOOKUP_RCU
) {
1403 /* RCU dangling. Cancel it. */
1404 nd
->flags
&= ~LOOKUP_RCU
;
1405 nd
->root
.mnt
= NULL
;
1407 br_read_unlock(vfsmount_lock
);
1413 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1419 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1420 nd
->flags
= flags
| LOOKUP_RCU
;
1422 nd
->root
.mnt
= NULL
;
1426 struct fs_struct
*fs
= current
->fs
;
1429 br_read_lock(vfsmount_lock
);
1433 seq
= read_seqcount_begin(&fs
->seq
);
1434 nd
->root
= fs
->root
;
1435 nd
->path
= nd
->root
;
1436 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1437 } while (read_seqcount_retry(&fs
->seq
, seq
));
1439 } else if (dfd
== AT_FDCWD
) {
1440 struct fs_struct
*fs
= current
->fs
;
1443 br_read_lock(vfsmount_lock
);
1447 seq
= read_seqcount_begin(&fs
->seq
);
1449 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1450 } while (read_seqcount_retry(&fs
->seq
, seq
));
1453 struct dentry
*dentry
;
1455 file
= fget_light(dfd
, &fput_needed
);
1460 dentry
= file
->f_path
.dentry
;
1463 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1466 retval
= file_permission(file
, MAY_EXEC
);
1470 nd
->path
= file
->f_path
;
1474 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1475 br_read_lock(vfsmount_lock
);
1478 nd
->inode
= nd
->path
.dentry
->d_inode
;
1482 fput_light(file
, fput_needed
);
1487 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1493 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1496 nd
->root
.mnt
= NULL
;
1500 nd
->path
= nd
->root
;
1501 path_get(&nd
->root
);
1502 } else if (dfd
== AT_FDCWD
) {
1503 get_fs_pwd(current
->fs
, &nd
->path
);
1505 struct dentry
*dentry
;
1507 file
= fget_light(dfd
, &fput_needed
);
1512 dentry
= file
->f_path
.dentry
;
1515 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1518 retval
= file_permission(file
, MAY_EXEC
);
1522 nd
->path
= file
->f_path
;
1523 path_get(&file
->f_path
);
1525 fput_light(file
, fput_needed
);
1527 nd
->inode
= nd
->path
.dentry
->d_inode
;
1531 fput_light(file
, fput_needed
);
1536 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1537 static int do_path_lookup(int dfd
, const char *name
,
1538 unsigned int flags
, struct nameidata
*nd
)
1543 * Path walking is largely split up into 2 different synchronisation
1544 * schemes, rcu-walk and ref-walk (explained in
1545 * Documentation/filesystems/path-lookup.txt). These share much of the
1546 * path walk code, but some things particularly setup, cleanup, and
1547 * following mounts are sufficiently divergent that functions are
1548 * duplicated. Typically there is a function foo(), and its RCU
1549 * analogue, foo_rcu().
1551 * -ECHILD is the error number of choice (just to avoid clashes) that
1552 * is returned if some aspect of an rcu-walk fails. Such an error must
1553 * be handled by restarting a traditional ref-walk (which will always
1554 * be able to complete).
1556 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1557 if (unlikely(retval
))
1559 retval
= path_walk_rcu(name
, nd
);
1560 path_finish_rcu(nd
);
1562 path_put(&nd
->root
);
1563 nd
->root
.mnt
= NULL
;
1566 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1567 /* slower, locked walk */
1568 if (retval
== -ESTALE
)
1569 flags
|= LOOKUP_REVAL
;
1570 retval
= path_init(dfd
, name
, flags
, nd
);
1571 if (unlikely(retval
))
1573 retval
= path_walk(name
, nd
);
1575 path_put(&nd
->root
);
1576 nd
->root
.mnt
= NULL
;
1580 if (likely(!retval
)) {
1581 if (unlikely(!audit_dummy_context())) {
1582 if (nd
->path
.dentry
&& nd
->inode
)
1583 audit_inode(name
, nd
->path
.dentry
);
1590 int path_lookup(const char *name
, unsigned int flags
,
1591 struct nameidata
*nd
)
1593 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1596 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1598 struct nameidata nd
;
1599 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1606 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1607 * @dentry: pointer to dentry of the base directory
1608 * @mnt: pointer to vfs mount of the base directory
1609 * @name: pointer to file name
1610 * @flags: lookup flags
1611 * @nd: pointer to nameidata
1613 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1614 const char *name
, unsigned int flags
,
1615 struct nameidata
*nd
)
1619 /* same as do_path_lookup */
1620 nd
->last_type
= LAST_ROOT
;
1624 nd
->path
.dentry
= dentry
;
1626 path_get(&nd
->path
);
1627 nd
->root
= nd
->path
;
1628 path_get(&nd
->root
);
1629 nd
->inode
= nd
->path
.dentry
->d_inode
;
1631 retval
= path_walk(name
, nd
);
1632 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1634 audit_inode(name
, nd
->path
.dentry
);
1636 path_put(&nd
->root
);
1637 nd
->root
.mnt
= NULL
;
1642 static struct dentry
*__lookup_hash(struct qstr
*name
,
1643 struct dentry
*base
, struct nameidata
*nd
)
1645 struct inode
*inode
= base
->d_inode
;
1646 struct dentry
*dentry
;
1649 err
= exec_permission(inode
, 0);
1651 return ERR_PTR(err
);
1654 * See if the low-level filesystem might want
1655 * to use its own hash..
1657 if (base
->d_flags
& DCACHE_OP_HASH
) {
1658 err
= base
->d_op
->d_hash(base
, inode
, name
);
1659 dentry
= ERR_PTR(err
);
1665 * Don't bother with __d_lookup: callers are for creat as
1666 * well as unlink, so a lot of the time it would cost
1669 dentry
= d_lookup(base
, name
);
1671 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1672 dentry
= do_revalidate(dentry
, nd
);
1675 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1681 * Restricted form of lookup. Doesn't follow links, single-component only,
1682 * needs parent already locked. Doesn't follow mounts.
1685 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1687 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1690 static int __lookup_one_len(const char *name
, struct qstr
*this,
1691 struct dentry
*base
, int len
)
1701 hash
= init_name_hash();
1703 c
= *(const unsigned char *)name
++;
1704 if (c
== '/' || c
== '\0')
1706 hash
= partial_name_hash(c
, hash
);
1708 this->hash
= end_name_hash(hash
);
1713 * lookup_one_len - filesystem helper to lookup single pathname component
1714 * @name: pathname component to lookup
1715 * @base: base directory to lookup from
1716 * @len: maximum length @len should be interpreted to
1718 * Note that this routine is purely a helper for filesystem usage and should
1719 * not be called by generic code. Also note that by using this function the
1720 * nameidata argument is passed to the filesystem methods and a filesystem
1721 * using this helper needs to be prepared for that.
1723 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1728 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1730 err
= __lookup_one_len(name
, &this, base
, len
);
1732 return ERR_PTR(err
);
1734 return __lookup_hash(&this, base
, NULL
);
1737 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1740 struct nameidata nd
;
1741 char *tmp
= getname(name
);
1742 int err
= PTR_ERR(tmp
);
1745 BUG_ON(flags
& LOOKUP_PARENT
);
1747 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1755 static int user_path_parent(int dfd
, const char __user
*path
,
1756 struct nameidata
*nd
, char **name
)
1758 char *s
= getname(path
);
1764 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1774 * It's inline, so penalty for filesystems that don't use sticky bit is
1777 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1779 uid_t fsuid
= current_fsuid();
1781 if (!(dir
->i_mode
& S_ISVTX
))
1783 if (inode
->i_uid
== fsuid
)
1785 if (dir
->i_uid
== fsuid
)
1787 return !capable(CAP_FOWNER
);
1791 * Check whether we can remove a link victim from directory dir, check
1792 * whether the type of victim is right.
1793 * 1. We can't do it if dir is read-only (done in permission())
1794 * 2. We should have write and exec permissions on dir
1795 * 3. We can't remove anything from append-only dir
1796 * 4. We can't do anything with immutable dir (done in permission())
1797 * 5. If the sticky bit on dir is set we should either
1798 * a. be owner of dir, or
1799 * b. be owner of victim, or
1800 * c. have CAP_FOWNER capability
1801 * 6. If the victim is append-only or immutable we can't do antyhing with
1802 * links pointing to it.
1803 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1804 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1805 * 9. We can't remove a root or mountpoint.
1806 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1807 * nfs_async_unlink().
1809 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1813 if (!victim
->d_inode
)
1816 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1817 audit_inode_child(victim
, dir
);
1819 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1824 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1825 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1828 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1830 if (IS_ROOT(victim
))
1832 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1834 if (IS_DEADDIR(dir
))
1836 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1841 /* Check whether we can create an object with dentry child in directory
1843 * 1. We can't do it if child already exists (open has special treatment for
1844 * this case, but since we are inlined it's OK)
1845 * 2. We can't do it if dir is read-only (done in permission())
1846 * 3. We should have write and exec permissions on dir
1847 * 4. We can't do it if dir is immutable (done in permission())
1849 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1853 if (IS_DEADDIR(dir
))
1855 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1859 * p1 and p2 should be directories on the same fs.
1861 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1866 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1870 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1872 p
= d_ancestor(p2
, p1
);
1874 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1875 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1879 p
= d_ancestor(p1
, p2
);
1881 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1882 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1886 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1887 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1891 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1893 mutex_unlock(&p1
->d_inode
->i_mutex
);
1895 mutex_unlock(&p2
->d_inode
->i_mutex
);
1896 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1900 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1901 struct nameidata
*nd
)
1903 int error
= may_create(dir
, dentry
);
1908 if (!dir
->i_op
->create
)
1909 return -EACCES
; /* shouldn't it be ENOSYS? */
1912 error
= security_inode_create(dir
, dentry
, mode
);
1915 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1917 fsnotify_create(dir
, dentry
);
1921 int may_open(struct path
*path
, int acc_mode
, int flag
)
1923 struct dentry
*dentry
= path
->dentry
;
1924 struct inode
*inode
= dentry
->d_inode
;
1930 switch (inode
->i_mode
& S_IFMT
) {
1934 if (acc_mode
& MAY_WRITE
)
1939 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1948 error
= inode_permission(inode
, acc_mode
);
1953 * An append-only file must be opened in append mode for writing.
1955 if (IS_APPEND(inode
)) {
1956 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1962 /* O_NOATIME can only be set by the owner or superuser */
1963 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
1967 * Ensure there are no outstanding leases on the file.
1969 return break_lease(inode
, flag
);
1972 static int handle_truncate(struct file
*filp
)
1974 struct path
*path
= &filp
->f_path
;
1975 struct inode
*inode
= path
->dentry
->d_inode
;
1976 int error
= get_write_access(inode
);
1980 * Refuse to truncate files with mandatory locks held on them.
1982 error
= locks_verify_locked(inode
);
1984 error
= security_path_truncate(path
);
1986 error
= do_truncate(path
->dentry
, 0,
1987 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1990 put_write_access(inode
);
1995 * Be careful about ever adding any more callers of this
1996 * function. Its flags must be in the namei format, not
1997 * what get passed to sys_open().
1999 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2000 int open_flag
, int mode
)
2003 struct dentry
*dir
= nd
->path
.dentry
;
2005 if (!IS_POSIXACL(dir
->d_inode
))
2006 mode
&= ~current_umask();
2007 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2010 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2012 mutex_unlock(&dir
->d_inode
->i_mutex
);
2013 dput(nd
->path
.dentry
);
2014 nd
->path
.dentry
= path
->dentry
;
2018 /* Don't check for write permission, don't truncate */
2019 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2023 * Note that while the flag value (low two bits) for sys_open means:
2028 * it is changed into
2029 * 00 - no permissions needed
2030 * 01 - read-permission
2031 * 10 - write-permission
2033 * for the internal routines (ie open_namei()/follow_link() etc)
2034 * This is more logical, and also allows the 00 "no perm needed"
2035 * to be used for symlinks (where the permissions are checked
2039 static inline int open_to_namei_flags(int flag
)
2041 if ((flag
+1) & O_ACCMODE
)
2046 static int open_will_truncate(int flag
, struct inode
*inode
)
2049 * We'll never write to the fs underlying
2052 if (special_file(inode
->i_mode
))
2054 return (flag
& O_TRUNC
);
2057 static struct file
*finish_open(struct nameidata
*nd
,
2058 int open_flag
, int acc_mode
)
2064 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2065 if (will_truncate
) {
2066 error
= mnt_want_write(nd
->path
.mnt
);
2070 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2073 mnt_drop_write(nd
->path
.mnt
);
2076 filp
= nameidata_to_filp(nd
);
2077 if (!IS_ERR(filp
)) {
2078 error
= ima_file_check(filp
, acc_mode
);
2081 filp
= ERR_PTR(error
);
2084 if (!IS_ERR(filp
)) {
2085 if (will_truncate
) {
2086 error
= handle_truncate(filp
);
2089 filp
= ERR_PTR(error
);
2094 * It is now safe to drop the mnt write
2095 * because the filp has had a write taken
2099 mnt_drop_write(nd
->path
.mnt
);
2100 path_put(&nd
->path
);
2104 if (!IS_ERR(nd
->intent
.open
.file
))
2105 release_open_intent(nd
);
2106 path_put(&nd
->path
);
2107 return ERR_PTR(error
);
2111 * Handle O_CREAT case for do_filp_open
2113 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2114 int open_flag
, int acc_mode
,
2115 int mode
, const char *pathname
)
2117 struct dentry
*dir
= nd
->path
.dentry
;
2119 int error
= -EISDIR
;
2121 switch (nd
->last_type
) {
2124 dir
= nd
->path
.dentry
;
2126 if (need_reval_dot(dir
)) {
2127 int status
= d_revalidate(nd
->path
.dentry
, nd
);
2139 audit_inode(pathname
, dir
);
2143 /* trailing slashes? */
2144 if (nd
->last
.name
[nd
->last
.len
])
2147 mutex_lock(&dir
->d_inode
->i_mutex
);
2149 path
->dentry
= lookup_hash(nd
);
2150 path
->mnt
= nd
->path
.mnt
;
2152 error
= PTR_ERR(path
->dentry
);
2153 if (IS_ERR(path
->dentry
)) {
2154 mutex_unlock(&dir
->d_inode
->i_mutex
);
2158 if (IS_ERR(nd
->intent
.open
.file
)) {
2159 error
= PTR_ERR(nd
->intent
.open
.file
);
2160 goto exit_mutex_unlock
;
2163 /* Negative dentry, just create the file */
2164 if (!path
->dentry
->d_inode
) {
2166 * This write is needed to ensure that a
2167 * ro->rw transition does not occur between
2168 * the time when the file is created and when
2169 * a permanent write count is taken through
2170 * the 'struct file' in nameidata_to_filp().
2172 error
= mnt_want_write(nd
->path
.mnt
);
2174 goto exit_mutex_unlock
;
2175 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2177 mnt_drop_write(nd
->path
.mnt
);
2180 filp
= nameidata_to_filp(nd
);
2181 mnt_drop_write(nd
->path
.mnt
);
2182 path_put(&nd
->path
);
2183 if (!IS_ERR(filp
)) {
2184 error
= ima_file_check(filp
, acc_mode
);
2187 filp
= ERR_PTR(error
);
2194 * It already exists.
2196 mutex_unlock(&dir
->d_inode
->i_mutex
);
2197 audit_inode(pathname
, path
->dentry
);
2200 if (open_flag
& O_EXCL
)
2203 if (__follow_mount(path
)) {
2205 if (open_flag
& O_NOFOLLOW
)
2210 if (!path
->dentry
->d_inode
)
2213 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2216 path_to_nameidata(path
, nd
);
2217 nd
->inode
= path
->dentry
->d_inode
;
2219 if (S_ISDIR(nd
->inode
->i_mode
))
2222 filp
= finish_open(nd
, open_flag
, acc_mode
);
2226 mutex_unlock(&dir
->d_inode
->i_mutex
);
2228 path_put_conditional(path
, nd
);
2230 if (!IS_ERR(nd
->intent
.open
.file
))
2231 release_open_intent(nd
);
2232 path_put(&nd
->path
);
2233 return ERR_PTR(error
);
2237 * Note that the low bits of the passed in "open_flag"
2238 * are not the same as in the local variable "flag". See
2239 * open_to_namei_flags() for more details.
2241 struct file
*do_filp_open(int dfd
, const char *pathname
,
2242 int open_flag
, int mode
, int acc_mode
)
2245 struct nameidata nd
;
2249 int flag
= open_to_namei_flags(open_flag
);
2252 if (!(open_flag
& O_CREAT
))
2255 /* Must never be set by userspace */
2256 open_flag
&= ~FMODE_NONOTIFY
;
2259 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2260 * check for O_DSYNC if the need any syncing at all we enforce it's
2261 * always set instead of having to deal with possibly weird behaviour
2262 * for malicious applications setting only __O_SYNC.
2264 if (open_flag
& __O_SYNC
)
2265 open_flag
|= O_DSYNC
;
2268 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2270 /* O_TRUNC implies we need access checks for write permissions */
2271 if (open_flag
& O_TRUNC
)
2272 acc_mode
|= MAY_WRITE
;
2274 /* Allow the LSM permission hook to distinguish append
2275 access from general write access. */
2276 if (open_flag
& O_APPEND
)
2277 acc_mode
|= MAY_APPEND
;
2279 flags
= LOOKUP_OPEN
;
2280 if (open_flag
& O_CREAT
) {
2281 flags
|= LOOKUP_CREATE
;
2282 if (open_flag
& O_EXCL
)
2283 flags
|= LOOKUP_EXCL
;
2285 if (open_flag
& O_DIRECTORY
)
2286 flags
|= LOOKUP_DIRECTORY
;
2287 if (!(open_flag
& O_NOFOLLOW
))
2288 flags
|= LOOKUP_FOLLOW
;
2290 filp
= get_empty_filp();
2292 return ERR_PTR(-ENFILE
);
2294 filp
->f_flags
= open_flag
;
2295 nd
.intent
.open
.file
= filp
;
2296 nd
.intent
.open
.flags
= flag
;
2297 nd
.intent
.open
.create_mode
= mode
;
2299 if (open_flag
& O_CREAT
)
2302 /* !O_CREAT, simple open */
2303 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2304 if (unlikely(error
))
2307 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2308 if (nd
.inode
->i_op
->follow_link
)
2312 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2313 if (!nd
.inode
->i_op
->lookup
)
2316 audit_inode(pathname
, nd
.path
.dentry
);
2317 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2321 /* OK, have to create the file. Find the parent. */
2322 error
= path_init_rcu(dfd
, pathname
,
2323 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2326 error
= path_walk_rcu(pathname
, &nd
);
2327 path_finish_rcu(&nd
);
2328 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2329 /* slower, locked walk */
2330 if (error
== -ESTALE
) {
2332 flags
|= LOOKUP_REVAL
;
2334 error
= path_init(dfd
, pathname
,
2335 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2339 error
= path_walk_simple(pathname
, &nd
);
2341 if (unlikely(error
))
2343 if (unlikely(!audit_dummy_context()))
2344 audit_inode(pathname
, nd
.path
.dentry
);
2347 * We have the parent and last component.
2350 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2351 while (unlikely(!filp
)) { /* trailing symlink */
2352 struct path link
= path
;
2353 struct inode
*linki
= link
.dentry
->d_inode
;
2356 /* S_ISDIR part is a temporary automount kludge */
2357 if (!(nd
.flags
& LOOKUP_FOLLOW
) && !S_ISDIR(linki
->i_mode
))
2362 * This is subtle. Instead of calling do_follow_link() we do
2363 * the thing by hands. The reason is that this way we have zero
2364 * link_count and path_walk() (called from ->follow_link)
2365 * honoring LOOKUP_PARENT. After that we have the parent and
2366 * last component, i.e. we are in the same situation as after
2367 * the first path_walk(). Well, almost - if the last component
2368 * is normal we get its copy stored in nd->last.name and we will
2369 * have to putname() it when we are done. Procfs-like symlinks
2370 * just set LAST_BIND.
2372 nd
.flags
|= LOOKUP_PARENT
;
2373 error
= security_inode_follow_link(link
.dentry
, &nd
);
2376 error
= __do_follow_link(&link
, &nd
, &cookie
);
2377 if (unlikely(error
)) {
2378 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2379 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2380 /* nd.path had been dropped */
2384 nd
.flags
&= ~LOOKUP_PARENT
;
2385 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2386 if (linki
->i_op
->put_link
)
2387 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2393 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2398 path_put_conditional(&path
, &nd
);
2402 if (!IS_ERR(nd
.intent
.open
.file
))
2403 release_open_intent(&nd
);
2404 filp
= ERR_PTR(error
);
2409 * filp_open - open file and return file pointer
2411 * @filename: path to open
2412 * @flags: open flags as per the open(2) second argument
2413 * @mode: mode for the new file if O_CREAT is set, else ignored
2415 * This is the helper to open a file from kernelspace if you really
2416 * have to. But in generally you should not do this, so please move
2417 * along, nothing to see here..
2419 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2421 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2423 EXPORT_SYMBOL(filp_open
);
2426 * lookup_create - lookup a dentry, creating it if it doesn't exist
2427 * @nd: nameidata info
2428 * @is_dir: directory flag
2430 * Simple function to lookup and return a dentry and create it
2431 * if it doesn't exist. Is SMP-safe.
2433 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2435 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2437 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2439 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2441 * Yucky last component or no last component at all?
2442 * (foo/., foo/.., /////)
2444 if (nd
->last_type
!= LAST_NORM
)
2446 nd
->flags
&= ~LOOKUP_PARENT
;
2447 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2448 nd
->intent
.open
.flags
= O_EXCL
;
2451 * Do the final lookup.
2453 dentry
= lookup_hash(nd
);
2457 if (dentry
->d_inode
)
2460 * Special case - lookup gave negative, but... we had foo/bar/
2461 * From the vfs_mknod() POV we just have a negative dentry -
2462 * all is fine. Let's be bastards - you had / on the end, you've
2463 * been asking for (non-existent) directory. -ENOENT for you.
2465 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2467 dentry
= ERR_PTR(-ENOENT
);
2472 dentry
= ERR_PTR(-EEXIST
);
2476 EXPORT_SYMBOL_GPL(lookup_create
);
2478 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2480 int error
= may_create(dir
, dentry
);
2485 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2488 if (!dir
->i_op
->mknod
)
2491 error
= devcgroup_inode_mknod(mode
, dev
);
2495 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2499 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2501 fsnotify_create(dir
, dentry
);
2505 static int may_mknod(mode_t mode
)
2507 switch (mode
& S_IFMT
) {
2513 case 0: /* zero mode translates to S_IFREG */
2522 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2527 struct dentry
*dentry
;
2528 struct nameidata nd
;
2533 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2537 dentry
= lookup_create(&nd
, 0);
2538 if (IS_ERR(dentry
)) {
2539 error
= PTR_ERR(dentry
);
2542 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2543 mode
&= ~current_umask();
2544 error
= may_mknod(mode
);
2547 error
= mnt_want_write(nd
.path
.mnt
);
2550 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2552 goto out_drop_write
;
2553 switch (mode
& S_IFMT
) {
2554 case 0: case S_IFREG
:
2555 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2557 case S_IFCHR
: case S_IFBLK
:
2558 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2559 new_decode_dev(dev
));
2561 case S_IFIFO
: case S_IFSOCK
:
2562 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2566 mnt_drop_write(nd
.path
.mnt
);
2570 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2577 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2579 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2582 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2584 int error
= may_create(dir
, dentry
);
2589 if (!dir
->i_op
->mkdir
)
2592 mode
&= (S_IRWXUGO
|S_ISVTX
);
2593 error
= security_inode_mkdir(dir
, dentry
, mode
);
2597 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2599 fsnotify_mkdir(dir
, dentry
);
2603 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2607 struct dentry
*dentry
;
2608 struct nameidata nd
;
2610 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2614 dentry
= lookup_create(&nd
, 1);
2615 error
= PTR_ERR(dentry
);
2619 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2620 mode
&= ~current_umask();
2621 error
= mnt_want_write(nd
.path
.mnt
);
2624 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2626 goto out_drop_write
;
2627 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2629 mnt_drop_write(nd
.path
.mnt
);
2633 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2640 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2642 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2646 * We try to drop the dentry early: we should have
2647 * a usage count of 2 if we're the only user of this
2648 * dentry, and if that is true (possibly after pruning
2649 * the dcache), then we drop the dentry now.
2651 * A low-level filesystem can, if it choses, legally
2654 * if (!d_unhashed(dentry))
2657 * if it cannot handle the case of removing a directory
2658 * that is still in use by something else..
2660 void dentry_unhash(struct dentry
*dentry
)
2663 shrink_dcache_parent(dentry
);
2664 spin_lock(&dentry
->d_lock
);
2665 if (dentry
->d_count
== 2)
2667 spin_unlock(&dentry
->d_lock
);
2670 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2672 int error
= may_delete(dir
, dentry
, 1);
2677 if (!dir
->i_op
->rmdir
)
2680 mutex_lock(&dentry
->d_inode
->i_mutex
);
2681 dentry_unhash(dentry
);
2682 if (d_mountpoint(dentry
))
2685 error
= security_inode_rmdir(dir
, dentry
);
2687 error
= dir
->i_op
->rmdir(dir
, dentry
);
2689 dentry
->d_inode
->i_flags
|= S_DEAD
;
2694 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2703 static long do_rmdir(int dfd
, const char __user
*pathname
)
2707 struct dentry
*dentry
;
2708 struct nameidata nd
;
2710 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2714 switch(nd
.last_type
) {
2726 nd
.flags
&= ~LOOKUP_PARENT
;
2728 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2729 dentry
= lookup_hash(&nd
);
2730 error
= PTR_ERR(dentry
);
2733 error
= mnt_want_write(nd
.path
.mnt
);
2736 error
= security_path_rmdir(&nd
.path
, dentry
);
2739 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2741 mnt_drop_write(nd
.path
.mnt
);
2745 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2752 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2754 return do_rmdir(AT_FDCWD
, pathname
);
2757 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2759 int error
= may_delete(dir
, dentry
, 0);
2764 if (!dir
->i_op
->unlink
)
2767 mutex_lock(&dentry
->d_inode
->i_mutex
);
2768 if (d_mountpoint(dentry
))
2771 error
= security_inode_unlink(dir
, dentry
);
2773 error
= dir
->i_op
->unlink(dir
, dentry
);
2778 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2780 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2781 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2782 fsnotify_link_count(dentry
->d_inode
);
2790 * Make sure that the actual truncation of the file will occur outside its
2791 * directory's i_mutex. Truncate can take a long time if there is a lot of
2792 * writeout happening, and we don't want to prevent access to the directory
2793 * while waiting on the I/O.
2795 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2799 struct dentry
*dentry
;
2800 struct nameidata nd
;
2801 struct inode
*inode
= NULL
;
2803 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2808 if (nd
.last_type
!= LAST_NORM
)
2811 nd
.flags
&= ~LOOKUP_PARENT
;
2813 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2814 dentry
= lookup_hash(&nd
);
2815 error
= PTR_ERR(dentry
);
2816 if (!IS_ERR(dentry
)) {
2817 /* Why not before? Because we want correct error value */
2818 if (nd
.last
.name
[nd
.last
.len
])
2820 inode
= dentry
->d_inode
;
2823 error
= mnt_want_write(nd
.path
.mnt
);
2826 error
= security_path_unlink(&nd
.path
, dentry
);
2829 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2831 mnt_drop_write(nd
.path
.mnt
);
2835 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2837 iput(inode
); /* truncate the inode here */
2844 error
= !dentry
->d_inode
? -ENOENT
:
2845 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2849 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2851 if ((flag
& ~AT_REMOVEDIR
) != 0)
2854 if (flag
& AT_REMOVEDIR
)
2855 return do_rmdir(dfd
, pathname
);
2857 return do_unlinkat(dfd
, pathname
);
2860 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2862 return do_unlinkat(AT_FDCWD
, pathname
);
2865 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2867 int error
= may_create(dir
, dentry
);
2872 if (!dir
->i_op
->symlink
)
2875 error
= security_inode_symlink(dir
, dentry
, oldname
);
2879 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2881 fsnotify_create(dir
, dentry
);
2885 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2886 int, newdfd
, const char __user
*, newname
)
2891 struct dentry
*dentry
;
2892 struct nameidata nd
;
2894 from
= getname(oldname
);
2896 return PTR_ERR(from
);
2898 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2902 dentry
= lookup_create(&nd
, 0);
2903 error
= PTR_ERR(dentry
);
2907 error
= mnt_want_write(nd
.path
.mnt
);
2910 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2912 goto out_drop_write
;
2913 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2915 mnt_drop_write(nd
.path
.mnt
);
2919 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2927 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2929 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2932 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2934 struct inode
*inode
= old_dentry
->d_inode
;
2940 error
= may_create(dir
, new_dentry
);
2944 if (dir
->i_sb
!= inode
->i_sb
)
2948 * A link to an append-only or immutable file cannot be created.
2950 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2952 if (!dir
->i_op
->link
)
2954 if (S_ISDIR(inode
->i_mode
))
2957 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2961 mutex_lock(&inode
->i_mutex
);
2962 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2963 mutex_unlock(&inode
->i_mutex
);
2965 fsnotify_link(dir
, inode
, new_dentry
);
2970 * Hardlinks are often used in delicate situations. We avoid
2971 * security-related surprises by not following symlinks on the
2974 * We don't follow them on the oldname either to be compatible
2975 * with linux 2.0, and to avoid hard-linking to directories
2976 * and other special files. --ADM
2978 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2979 int, newdfd
, const char __user
*, newname
, int, flags
)
2981 struct dentry
*new_dentry
;
2982 struct nameidata nd
;
2983 struct path old_path
;
2987 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2990 error
= user_path_at(olddfd
, oldname
,
2991 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2996 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3000 if (old_path
.mnt
!= nd
.path
.mnt
)
3002 new_dentry
= lookup_create(&nd
, 0);
3003 error
= PTR_ERR(new_dentry
);
3004 if (IS_ERR(new_dentry
))
3006 error
= mnt_want_write(nd
.path
.mnt
);
3009 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3011 goto out_drop_write
;
3012 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3014 mnt_drop_write(nd
.path
.mnt
);
3018 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3023 path_put(&old_path
);
3028 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3030 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3034 * The worst of all namespace operations - renaming directory. "Perverted"
3035 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3037 * a) we can get into loop creation. Check is done in is_subdir().
3038 * b) race potential - two innocent renames can create a loop together.
3039 * That's where 4.4 screws up. Current fix: serialization on
3040 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3042 * c) we have to lock _three_ objects - parents and victim (if it exists).
3043 * And that - after we got ->i_mutex on parents (until then we don't know
3044 * whether the target exists). Solution: try to be smart with locking
3045 * order for inodes. We rely on the fact that tree topology may change
3046 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3047 * move will be locked. Thus we can rank directories by the tree
3048 * (ancestors first) and rank all non-directories after them.
3049 * That works since everybody except rename does "lock parent, lookup,
3050 * lock child" and rename is under ->s_vfs_rename_mutex.
3051 * HOWEVER, it relies on the assumption that any object with ->lookup()
3052 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3053 * we'd better make sure that there's no link(2) for them.
3054 * d) some filesystems don't support opened-but-unlinked directories,
3055 * either because of layout or because they are not ready to deal with
3056 * all cases correctly. The latter will be fixed (taking this sort of
3057 * stuff into VFS), but the former is not going away. Solution: the same
3058 * trick as in rmdir().
3059 * e) conversion from fhandle to dentry may come in the wrong moment - when
3060 * we are removing the target. Solution: we will have to grab ->i_mutex
3061 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3062 * ->i_mutex on parents, which works but leads to some truly excessive
3065 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3066 struct inode
*new_dir
, struct dentry
*new_dentry
)
3069 struct inode
*target
;
3072 * If we are going to change the parent - check write permissions,
3073 * we'll need to flip '..'.
3075 if (new_dir
!= old_dir
) {
3076 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3081 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3085 target
= new_dentry
->d_inode
;
3087 mutex_lock(&target
->i_mutex
);
3088 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3092 dentry_unhash(new_dentry
);
3093 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3097 target
->i_flags
|= S_DEAD
;
3098 dont_mount(new_dentry
);
3100 mutex_unlock(&target
->i_mutex
);
3101 if (d_unhashed(new_dentry
))
3102 d_rehash(new_dentry
);
3106 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3107 d_move(old_dentry
,new_dentry
);
3111 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3112 struct inode
*new_dir
, struct dentry
*new_dentry
)
3114 struct inode
*target
;
3117 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3122 target
= new_dentry
->d_inode
;
3124 mutex_lock(&target
->i_mutex
);
3125 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3128 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3131 dont_mount(new_dentry
);
3132 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3133 d_move(old_dentry
, new_dentry
);
3136 mutex_unlock(&target
->i_mutex
);
3141 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3142 struct inode
*new_dir
, struct dentry
*new_dentry
)
3145 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3146 const unsigned char *old_name
;
3148 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3151 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3155 if (!new_dentry
->d_inode
)
3156 error
= may_create(new_dir
, new_dentry
);
3158 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3162 if (!old_dir
->i_op
->rename
)
3165 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3168 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3170 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3172 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3173 new_dentry
->d_inode
, old_dentry
);
3174 fsnotify_oldname_free(old_name
);
3179 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3180 int, newdfd
, const char __user
*, newname
)
3182 struct dentry
*old_dir
, *new_dir
;
3183 struct dentry
*old_dentry
, *new_dentry
;
3184 struct dentry
*trap
;
3185 struct nameidata oldnd
, newnd
;
3190 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3194 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3199 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3202 old_dir
= oldnd
.path
.dentry
;
3204 if (oldnd
.last_type
!= LAST_NORM
)
3207 new_dir
= newnd
.path
.dentry
;
3208 if (newnd
.last_type
!= LAST_NORM
)
3211 oldnd
.flags
&= ~LOOKUP_PARENT
;
3212 newnd
.flags
&= ~LOOKUP_PARENT
;
3213 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3215 trap
= lock_rename(new_dir
, old_dir
);
3217 old_dentry
= lookup_hash(&oldnd
);
3218 error
= PTR_ERR(old_dentry
);
3219 if (IS_ERR(old_dentry
))
3221 /* source must exist */
3223 if (!old_dentry
->d_inode
)
3225 /* unless the source is a directory trailing slashes give -ENOTDIR */
3226 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3228 if (oldnd
.last
.name
[oldnd
.last
.len
])
3230 if (newnd
.last
.name
[newnd
.last
.len
])
3233 /* source should not be ancestor of target */
3235 if (old_dentry
== trap
)
3237 new_dentry
= lookup_hash(&newnd
);
3238 error
= PTR_ERR(new_dentry
);
3239 if (IS_ERR(new_dentry
))
3241 /* target should not be an ancestor of source */
3243 if (new_dentry
== trap
)
3246 error
= mnt_want_write(oldnd
.path
.mnt
);
3249 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3250 &newnd
.path
, new_dentry
);
3253 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3254 new_dir
->d_inode
, new_dentry
);
3256 mnt_drop_write(oldnd
.path
.mnt
);
3262 unlock_rename(new_dir
, old_dir
);
3264 path_put(&newnd
.path
);
3267 path_put(&oldnd
.path
);
3273 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3275 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3278 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3282 len
= PTR_ERR(link
);
3287 if (len
> (unsigned) buflen
)
3289 if (copy_to_user(buffer
, link
, len
))
3296 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3297 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3298 * using) it for any given inode is up to filesystem.
3300 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3302 struct nameidata nd
;
3307 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3309 return PTR_ERR(cookie
);
3311 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3312 if (dentry
->d_inode
->i_op
->put_link
)
3313 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3317 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3319 return __vfs_follow_link(nd
, link
);
3322 /* get the link contents into pagecache */
3323 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3327 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3328 page
= read_mapping_page(mapping
, 0, NULL
);
3333 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3337 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3339 struct page
*page
= NULL
;
3340 char *s
= page_getlink(dentry
, &page
);
3341 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3344 page_cache_release(page
);
3349 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3351 struct page
*page
= NULL
;
3352 nd_set_link(nd
, page_getlink(dentry
, &page
));
3356 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3358 struct page
*page
= cookie
;
3362 page_cache_release(page
);
3367 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3369 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3371 struct address_space
*mapping
= inode
->i_mapping
;
3376 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3378 flags
|= AOP_FLAG_NOFS
;
3381 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3382 flags
, &page
, &fsdata
);
3386 kaddr
= kmap_atomic(page
, KM_USER0
);
3387 memcpy(kaddr
, symname
, len
-1);
3388 kunmap_atomic(kaddr
, KM_USER0
);
3390 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3397 mark_inode_dirty(inode
);
3403 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3405 return __page_symlink(inode
, symname
, len
,
3406 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3409 const struct inode_operations page_symlink_inode_operations
= {
3410 .readlink
= generic_readlink
,
3411 .follow_link
= page_follow_link_light
,
3412 .put_link
= page_put_link
,
3415 EXPORT_SYMBOL(user_path_at
);
3416 EXPORT_SYMBOL(follow_down
);
3417 EXPORT_SYMBOL(follow_up
);
3418 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3419 EXPORT_SYMBOL(getname
);
3420 EXPORT_SYMBOL(lock_rename
);
3421 EXPORT_SYMBOL(lookup_one_len
);
3422 EXPORT_SYMBOL(page_follow_link_light
);
3423 EXPORT_SYMBOL(page_put_link
);
3424 EXPORT_SYMBOL(page_readlink
);
3425 EXPORT_SYMBOL(__page_symlink
);
3426 EXPORT_SYMBOL(page_symlink
);
3427 EXPORT_SYMBOL(page_symlink_inode_operations
);
3428 EXPORT_SYMBOL(path_lookup
);
3429 EXPORT_SYMBOL(kern_path
);
3430 EXPORT_SYMBOL(vfs_path_lookup
);
3431 EXPORT_SYMBOL(inode_permission
);
3432 EXPORT_SYMBOL(file_permission
);
3433 EXPORT_SYMBOL(unlock_rename
);
3434 EXPORT_SYMBOL(vfs_create
);
3435 EXPORT_SYMBOL(vfs_follow_link
);
3436 EXPORT_SYMBOL(vfs_link
);
3437 EXPORT_SYMBOL(vfs_mkdir
);
3438 EXPORT_SYMBOL(vfs_mknod
);
3439 EXPORT_SYMBOL(generic_permission
);
3440 EXPORT_SYMBOL(vfs_readlink
);
3441 EXPORT_SYMBOL(vfs_rename
);
3442 EXPORT_SYMBOL(vfs_rmdir
);
3443 EXPORT_SYMBOL(vfs_symlink
);
3444 EXPORT_SYMBOL(vfs_unlink
);
3445 EXPORT_SYMBOL(dentry_unhash
);
3446 EXPORT_SYMBOL(generic_readlink
);