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/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 void final_putname(struct filename
*name
)
122 if (name
->separate
) {
123 __putname(name
->name
);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename
*
133 getname_flags(const char __user
*filename
, int flags
, int *empty
)
135 struct filename
*result
, *err
;
140 result
= audit_reusename(filename
);
144 result
= __getname();
145 if (unlikely(!result
))
146 return ERR_PTR(-ENOMEM
);
149 * First, try to embed the struct filename inside the names_cache
152 kname
= (char *)result
+ sizeof(*result
);
153 result
->name
= kname
;
154 result
->separate
= false;
155 max
= EMBEDDED_NAME_MAX
;
158 len
= strncpy_from_user(kname
, filename
, max
);
159 if (unlikely(len
< 0)) {
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
170 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
171 kname
= (char *)result
;
173 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
175 err
= ERR_PTR(-ENOMEM
);
176 result
= (struct filename
*)kname
;
179 result
->name
= kname
;
180 result
->separate
= true;
185 /* The empty path is special. */
186 if (unlikely(!len
)) {
189 err
= ERR_PTR(-ENOENT
);
190 if (!(flags
& LOOKUP_EMPTY
))
194 err
= ERR_PTR(-ENAMETOOLONG
);
195 if (unlikely(len
>= PATH_MAX
))
198 result
->uptr
= filename
;
199 audit_getname(result
);
203 final_putname(result
);
208 getname(const char __user
* filename
)
210 return getname_flags(filename
, 0, NULL
);
212 EXPORT_SYMBOL(getname
);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename
*name
)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name
);
223 static int check_acl(struct inode
*inode
, int mask
)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl
*acl
;
228 if (mask
& MAY_NOT_BLOCK
) {
229 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl
== ACL_NOT_CACHED
)
235 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
238 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl
== ACL_NOT_CACHED
) {
249 if (inode
->i_op
->get_acl
) {
250 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
254 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
260 int error
= posix_acl_permission(inode
, acl
, mask
);
261 posix_acl_release(acl
);
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode
*inode
, int mask
)
274 unsigned int mode
= inode
->i_mode
;
276 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
279 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
280 int error
= check_acl(inode
, mask
);
281 if (error
!= -EAGAIN
)
285 if (in_group_p(inode
->i_gid
))
290 * If the DACs are ok we don't need any capability check.
292 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
298 * generic_permission - check for access rights on a Posix-like filesystem
299 * @inode: inode to check access rights for
300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
305 * are used for other things.
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
311 int generic_permission(struct inode
*inode
, int mask
)
316 * Do the basic permission checks.
318 ret
= acl_permission_check(inode
, mask
);
322 if (S_ISDIR(inode
->i_mode
)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
326 if (!(mask
& MAY_WRITE
))
327 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
337 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
341 * Searching includes executable on directories, else just read.
343 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
344 if (mask
== MAY_READ
)
345 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
357 static inline int do_inode_permission(struct inode
*inode
, int mask
)
359 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
360 if (likely(inode
->i_op
->permission
))
361 return inode
->i_op
->permission(inode
, mask
);
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode
->i_lock
);
365 inode
->i_opflags
|= IOP_FASTPERM
;
366 spin_unlock(&inode
->i_lock
);
368 return generic_permission(inode
, mask
);
372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode
*inode
, int mask
)
387 if (unlikely(mask
& MAY_WRITE
)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode
))
395 retval
= do_inode_permission(inode
, mask
);
399 retval
= devcgroup_inode_permission(inode
, mask
);
403 return security_inode_permission(inode
, mask
);
407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
409 * @inode: Inode to check permission on
410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
416 if (unlikely(mask
& MAY_WRITE
)) {
417 umode_t mode
= inode
->i_mode
;
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb
->s_flags
& MS_RDONLY
) &&
421 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode
*inode
, int mask
)
442 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
445 return __inode_permission(inode
, mask
);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(struct path
*path
)
459 EXPORT_SYMBOL(path_get
);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(struct path
*path
)
472 EXPORT_SYMBOL(path_put
);
475 * Path walking has 2 modes, rcu-walk and ref-walk (see
476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
485 static inline void lock_rcu_walk(void)
487 br_read_lock(&vfsmount_lock
);
491 static inline void unlock_rcu_walk(void)
494 br_read_unlock(&vfsmount_lock
);
498 * unlazy_walk - try to switch to ref-walk mode.
499 * @nd: nameidata pathwalk data
500 * @dentry: child of nd->path.dentry or NULL
501 * Returns: 0 on success, -ECHILD on failure
503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
504 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
505 * @nd or NULL. Must be called from rcu-walk context.
507 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
509 struct fs_struct
*fs
= current
->fs
;
510 struct dentry
*parent
= nd
->path
.dentry
;
513 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
514 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
516 spin_lock(&fs
->lock
);
517 if (nd
->root
.mnt
!= fs
->root
.mnt
||
518 nd
->root
.dentry
!= fs
->root
.dentry
)
521 spin_lock(&parent
->d_lock
);
523 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
525 BUG_ON(nd
->inode
!= parent
->d_inode
);
527 if (dentry
->d_parent
!= parent
)
529 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
530 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
533 * If the sequence check on the child dentry passed, then
534 * the child has not been removed from its parent. This
535 * means the parent dentry must be valid and able to take
536 * a reference at this point.
538 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
539 BUG_ON(!parent
->d_count
);
541 spin_unlock(&dentry
->d_lock
);
543 spin_unlock(&parent
->d_lock
);
546 spin_unlock(&fs
->lock
);
548 mntget(nd
->path
.mnt
);
551 nd
->flags
&= ~LOOKUP_RCU
;
555 spin_unlock(&dentry
->d_lock
);
557 spin_unlock(&parent
->d_lock
);
560 spin_unlock(&fs
->lock
);
564 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
566 return dentry
->d_op
->d_revalidate(dentry
, flags
);
570 * complete_walk - successful completion of path walk
571 * @nd: pointer nameidata
573 * If we had been in RCU mode, drop out of it and legitimize nd->path.
574 * Revalidate the final result, unless we'd already done that during
575 * the path walk or the filesystem doesn't ask for it. Return 0 on
576 * success, -error on failure. In case of failure caller does not
577 * need to drop nd->path.
579 static int complete_walk(struct nameidata
*nd
)
581 struct dentry
*dentry
= nd
->path
.dentry
;
584 if (nd
->flags
& LOOKUP_RCU
) {
585 nd
->flags
&= ~LOOKUP_RCU
;
586 if (!(nd
->flags
& LOOKUP_ROOT
))
588 spin_lock(&dentry
->d_lock
);
589 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
590 spin_unlock(&dentry
->d_lock
);
594 BUG_ON(nd
->inode
!= dentry
->d_inode
);
595 spin_unlock(&dentry
->d_lock
);
596 mntget(nd
->path
.mnt
);
600 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
603 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
606 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
609 /* Note: we do not d_invalidate() */
610 status
= d_revalidate(dentry
, nd
->flags
);
621 static __always_inline
void set_root(struct nameidata
*nd
)
624 get_fs_root(current
->fs
, &nd
->root
);
627 static int link_path_walk(const char *, struct nameidata
*);
629 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
632 struct fs_struct
*fs
= current
->fs
;
636 seq
= read_seqcount_begin(&fs
->seq
);
638 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
639 } while (read_seqcount_retry(&fs
->seq
, seq
));
643 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
655 nd
->flags
|= LOOKUP_JUMPED
;
657 nd
->inode
= nd
->path
.dentry
->d_inode
;
659 ret
= link_path_walk(link
, nd
);
663 return PTR_ERR(link
);
666 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
669 if (path
->mnt
!= nd
->path
.mnt
)
673 static inline void path_to_nameidata(const struct path
*path
,
674 struct nameidata
*nd
)
676 if (!(nd
->flags
& LOOKUP_RCU
)) {
677 dput(nd
->path
.dentry
);
678 if (nd
->path
.mnt
!= path
->mnt
)
679 mntput(nd
->path
.mnt
);
681 nd
->path
.mnt
= path
->mnt
;
682 nd
->path
.dentry
= path
->dentry
;
686 * Helper to directly jump to a known parsed path from ->follow_link,
687 * caller must have taken a reference to path beforehand.
689 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
694 nd
->inode
= nd
->path
.dentry
->d_inode
;
695 nd
->flags
|= LOOKUP_JUMPED
;
697 BUG_ON(nd
->inode
->i_op
->follow_link
);
700 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
702 struct inode
*inode
= link
->dentry
->d_inode
;
703 if (inode
->i_op
->put_link
)
704 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
708 int sysctl_protected_symlinks __read_mostly
= 1;
709 int sysctl_protected_hardlinks __read_mostly
= 1;
712 * may_follow_link - Check symlink following for unsafe situations
713 * @link: The path of the symlink
714 * @nd: nameidata pathwalk data
716 * In the case of the sysctl_protected_symlinks sysctl being enabled,
717 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
718 * in a sticky world-writable directory. This is to protect privileged
719 * processes from failing races against path names that may change out
720 * from under them by way of other users creating malicious symlinks.
721 * It will permit symlinks to be followed only when outside a sticky
722 * world-writable directory, or when the uid of the symlink and follower
723 * match, or when the directory owner matches the symlink's owner.
725 * Returns 0 if following the symlink is allowed, -ve on error.
727 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
729 const struct inode
*inode
;
730 const struct inode
*parent
;
732 if (!sysctl_protected_symlinks
)
735 /* Allowed if owner and follower match. */
736 inode
= link
->dentry
->d_inode
;
737 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
740 /* Allowed if parent directory not sticky and world-writable. */
741 parent
= nd
->path
.dentry
->d_inode
;
742 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
745 /* Allowed if parent directory and link owner match. */
746 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
749 audit_log_link_denied("follow_link", link
);
750 path_put_conditional(link
, nd
);
756 * safe_hardlink_source - Check for safe hardlink conditions
757 * @inode: the source inode to hardlink from
759 * Return false if at least one of the following conditions:
760 * - inode is not a regular file
762 * - inode is setgid and group-exec
763 * - access failure for read and write
765 * Otherwise returns true.
767 static bool safe_hardlink_source(struct inode
*inode
)
769 umode_t mode
= inode
->i_mode
;
771 /* Special files should not get pinned to the filesystem. */
775 /* Setuid files should not get pinned to the filesystem. */
779 /* Executable setgid files should not get pinned to the filesystem. */
780 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
783 /* Hardlinking to unreadable or unwritable sources is dangerous. */
784 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
791 * may_linkat - Check permissions for creating a hardlink
792 * @link: the source to hardlink from
794 * Block hardlink when all of:
795 * - sysctl_protected_hardlinks enabled
796 * - fsuid does not match inode
797 * - hardlink source is unsafe (see safe_hardlink_source() above)
800 * Returns 0 if successful, -ve on error.
802 static int may_linkat(struct path
*link
)
804 const struct cred
*cred
;
807 if (!sysctl_protected_hardlinks
)
810 cred
= current_cred();
811 inode
= link
->dentry
->d_inode
;
813 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
814 * otherwise, it must be a safe source.
816 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
820 audit_log_link_denied("linkat", link
);
824 static __always_inline
int
825 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
827 struct dentry
*dentry
= link
->dentry
;
831 BUG_ON(nd
->flags
& LOOKUP_RCU
);
833 if (link
->mnt
== nd
->path
.mnt
)
837 if (unlikely(current
->total_link_count
>= 40))
838 goto out_put_nd_path
;
841 current
->total_link_count
++;
844 nd_set_link(nd
, NULL
);
846 error
= security_inode_follow_link(link
->dentry
, nd
);
848 goto out_put_nd_path
;
850 nd
->last_type
= LAST_BIND
;
851 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
854 goto out_put_nd_path
;
859 error
= __vfs_follow_link(nd
, s
);
861 put_link(nd
, link
, *p
);
873 static int follow_up_rcu(struct path
*path
)
875 struct mount
*mnt
= real_mount(path
->mnt
);
876 struct mount
*parent
;
877 struct dentry
*mountpoint
;
879 parent
= mnt
->mnt_parent
;
880 if (&parent
->mnt
== path
->mnt
)
882 mountpoint
= mnt
->mnt_mountpoint
;
883 path
->dentry
= mountpoint
;
884 path
->mnt
= &parent
->mnt
;
889 * follow_up - Find the mountpoint of path's vfsmount
891 * Given a path, find the mountpoint of its source file system.
892 * Replace @path with the path of the mountpoint in the parent mount.
895 * Return 1 if we went up a level and 0 if we were already at the
898 int follow_up(struct path
*path
)
900 struct mount
*mnt
= real_mount(path
->mnt
);
901 struct mount
*parent
;
902 struct dentry
*mountpoint
;
904 br_read_lock(&vfsmount_lock
);
905 parent
= mnt
->mnt_parent
;
907 br_read_unlock(&vfsmount_lock
);
910 mntget(&parent
->mnt
);
911 mountpoint
= dget(mnt
->mnt_mountpoint
);
912 br_read_unlock(&vfsmount_lock
);
914 path
->dentry
= mountpoint
;
916 path
->mnt
= &parent
->mnt
;
921 * Perform an automount
922 * - return -EISDIR to tell follow_managed() to stop and return the path we
925 static int follow_automount(struct path
*path
, unsigned flags
,
928 struct vfsmount
*mnt
;
931 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
934 /* We don't want to mount if someone's just doing a stat -
935 * unless they're stat'ing a directory and appended a '/' to
938 * We do, however, want to mount if someone wants to open or
939 * create a file of any type under the mountpoint, wants to
940 * traverse through the mountpoint or wants to open the
941 * mounted directory. Also, autofs may mark negative dentries
942 * as being automount points. These will need the attentions
943 * of the daemon to instantiate them before they can be used.
945 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
946 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
947 path
->dentry
->d_inode
)
950 current
->total_link_count
++;
951 if (current
->total_link_count
>= 40)
954 mnt
= path
->dentry
->d_op
->d_automount(path
);
957 * The filesystem is allowed to return -EISDIR here to indicate
958 * it doesn't want to automount. For instance, autofs would do
959 * this so that its userspace daemon can mount on this dentry.
961 * However, we can only permit this if it's a terminal point in
962 * the path being looked up; if it wasn't then the remainder of
963 * the path is inaccessible and we should say so.
965 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
970 if (!mnt
) /* mount collision */
974 /* lock_mount() may release path->mnt on error */
978 err
= finish_automount(mnt
, path
);
982 /* Someone else made a mount here whilst we were busy */
987 path
->dentry
= dget(mnt
->mnt_root
);
996 * Handle a dentry that is managed in some way.
997 * - Flagged for transit management (autofs)
998 * - Flagged as mountpoint
999 * - Flagged as automount point
1001 * This may only be called in refwalk mode.
1003 * Serialization is taken care of in namespace.c
1005 static int follow_managed(struct path
*path
, unsigned flags
)
1007 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1009 bool need_mntput
= false;
1012 /* Given that we're not holding a lock here, we retain the value in a
1013 * local variable for each dentry as we look at it so that we don't see
1014 * the components of that value change under us */
1015 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1016 managed
&= DCACHE_MANAGED_DENTRY
,
1017 unlikely(managed
!= 0)) {
1018 /* Allow the filesystem to manage the transit without i_mutex
1020 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1021 BUG_ON(!path
->dentry
->d_op
);
1022 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1023 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1028 /* Transit to a mounted filesystem. */
1029 if (managed
& DCACHE_MOUNTED
) {
1030 struct vfsmount
*mounted
= lookup_mnt(path
);
1035 path
->mnt
= mounted
;
1036 path
->dentry
= dget(mounted
->mnt_root
);
1041 /* Something is mounted on this dentry in another
1042 * namespace and/or whatever was mounted there in this
1043 * namespace got unmounted before we managed to get the
1047 /* Handle an automount point */
1048 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1049 ret
= follow_automount(path
, flags
, &need_mntput
);
1055 /* We didn't change the current path point */
1059 if (need_mntput
&& path
->mnt
== mnt
)
1063 return ret
< 0 ? ret
: need_mntput
;
1066 int follow_down_one(struct path
*path
)
1068 struct vfsmount
*mounted
;
1070 mounted
= lookup_mnt(path
);
1074 path
->mnt
= mounted
;
1075 path
->dentry
= dget(mounted
->mnt_root
);
1081 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1083 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1084 dentry
->d_op
->d_manage(dentry
, true) < 0);
1088 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1089 * we meet a managed dentry that would need blocking.
1091 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1092 struct inode
**inode
)
1095 struct mount
*mounted
;
1097 * Don't forget we might have a non-mountpoint managed dentry
1098 * that wants to block transit.
1100 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1103 if (!d_mountpoint(path
->dentry
))
1106 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1109 path
->mnt
= &mounted
->mnt
;
1110 path
->dentry
= mounted
->mnt
.mnt_root
;
1111 nd
->flags
|= LOOKUP_JUMPED
;
1112 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1114 * Update the inode too. We don't need to re-check the
1115 * dentry sequence number here after this d_inode read,
1116 * because a mount-point is always pinned.
1118 *inode
= path
->dentry
->d_inode
;
1123 static void follow_mount_rcu(struct nameidata
*nd
)
1125 while (d_mountpoint(nd
->path
.dentry
)) {
1126 struct mount
*mounted
;
1127 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1130 nd
->path
.mnt
= &mounted
->mnt
;
1131 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1132 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1136 static int follow_dotdot_rcu(struct nameidata
*nd
)
1141 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1142 nd
->path
.mnt
== nd
->root
.mnt
) {
1145 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1146 struct dentry
*old
= nd
->path
.dentry
;
1147 struct dentry
*parent
= old
->d_parent
;
1150 seq
= read_seqcount_begin(&parent
->d_seq
);
1151 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1153 nd
->path
.dentry
= parent
;
1157 if (!follow_up_rcu(&nd
->path
))
1159 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1161 follow_mount_rcu(nd
);
1162 nd
->inode
= nd
->path
.dentry
->d_inode
;
1166 nd
->flags
&= ~LOOKUP_RCU
;
1167 if (!(nd
->flags
& LOOKUP_ROOT
))
1168 nd
->root
.mnt
= NULL
;
1174 * Follow down to the covering mount currently visible to userspace. At each
1175 * point, the filesystem owning that dentry may be queried as to whether the
1176 * caller is permitted to proceed or not.
1178 int follow_down(struct path
*path
)
1183 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1184 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1185 /* Allow the filesystem to manage the transit without i_mutex
1188 * We indicate to the filesystem if someone is trying to mount
1189 * something here. This gives autofs the chance to deny anyone
1190 * other than its daemon the right to mount on its
1193 * The filesystem may sleep at this point.
1195 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1196 BUG_ON(!path
->dentry
->d_op
);
1197 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1198 ret
= path
->dentry
->d_op
->d_manage(
1199 path
->dentry
, false);
1201 return ret
== -EISDIR
? 0 : ret
;
1204 /* Transit to a mounted filesystem. */
1205 if (managed
& DCACHE_MOUNTED
) {
1206 struct vfsmount
*mounted
= lookup_mnt(path
);
1211 path
->mnt
= mounted
;
1212 path
->dentry
= dget(mounted
->mnt_root
);
1216 /* Don't handle automount points here */
1223 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1225 static void follow_mount(struct path
*path
)
1227 while (d_mountpoint(path
->dentry
)) {
1228 struct vfsmount
*mounted
= lookup_mnt(path
);
1233 path
->mnt
= mounted
;
1234 path
->dentry
= dget(mounted
->mnt_root
);
1238 static void follow_dotdot(struct nameidata
*nd
)
1243 struct dentry
*old
= nd
->path
.dentry
;
1245 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1246 nd
->path
.mnt
== nd
->root
.mnt
) {
1249 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1250 /* rare case of legitimate dget_parent()... */
1251 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1255 if (!follow_up(&nd
->path
))
1258 follow_mount(&nd
->path
);
1259 nd
->inode
= nd
->path
.dentry
->d_inode
;
1263 * This looks up the name in dcache, possibly revalidates the old dentry and
1264 * allocates a new one if not found or not valid. In the need_lookup argument
1265 * returns whether i_op->lookup is necessary.
1267 * dir->d_inode->i_mutex must be held
1269 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1270 unsigned int flags
, bool *need_lookup
)
1272 struct dentry
*dentry
;
1275 *need_lookup
= false;
1276 dentry
= d_lookup(dir
, name
);
1278 if (d_need_lookup(dentry
)) {
1279 *need_lookup
= true;
1280 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1281 error
= d_revalidate(dentry
, flags
);
1282 if (unlikely(error
<= 0)) {
1285 return ERR_PTR(error
);
1286 } else if (!d_invalidate(dentry
)) {
1295 dentry
= d_alloc(dir
, name
);
1296 if (unlikely(!dentry
))
1297 return ERR_PTR(-ENOMEM
);
1299 *need_lookup
= true;
1305 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1306 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1308 * dir->d_inode->i_mutex must be held
1310 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1315 /* Don't create child dentry for a dead directory. */
1316 if (unlikely(IS_DEADDIR(dir
))) {
1318 return ERR_PTR(-ENOENT
);
1321 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1322 if (unlikely(old
)) {
1329 static struct dentry
*__lookup_hash(struct qstr
*name
,
1330 struct dentry
*base
, unsigned int flags
)
1333 struct dentry
*dentry
;
1335 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1339 return lookup_real(base
->d_inode
, dentry
, flags
);
1343 * It's more convoluted than I'd like it to be, but... it's still fairly
1344 * small and for now I'd prefer to have fast path as straight as possible.
1345 * It _is_ time-critical.
1347 static int lookup_fast(struct nameidata
*nd
, struct qstr
*name
,
1348 struct path
*path
, struct inode
**inode
)
1350 struct vfsmount
*mnt
= nd
->path
.mnt
;
1351 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1357 * Rename seqlock is not required here because in the off chance
1358 * of a false negative due to a concurrent rename, we're going to
1359 * do the non-racy lookup, below.
1361 if (nd
->flags
& LOOKUP_RCU
) {
1363 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1368 * This sequence count validates that the inode matches
1369 * the dentry name information from lookup.
1371 *inode
= dentry
->d_inode
;
1372 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1376 * This sequence count validates that the parent had no
1377 * changes while we did the lookup of the dentry above.
1379 * The memory barrier in read_seqcount_begin of child is
1380 * enough, we can use __read_seqcount_retry here.
1382 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1386 if (unlikely(d_need_lookup(dentry
)))
1388 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1389 status
= d_revalidate(dentry
, nd
->flags
);
1390 if (unlikely(status
<= 0)) {
1391 if (status
!= -ECHILD
)
1397 path
->dentry
= dentry
;
1398 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1400 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1404 if (unlazy_walk(nd
, dentry
))
1407 dentry
= __d_lookup(parent
, name
);
1410 if (unlikely(!dentry
))
1413 if (unlikely(d_need_lookup(dentry
))) {
1418 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1419 status
= d_revalidate(dentry
, nd
->flags
);
1420 if (unlikely(status
<= 0)) {
1425 if (!d_invalidate(dentry
)) {
1432 path
->dentry
= dentry
;
1433 err
= follow_managed(path
, nd
->flags
);
1434 if (unlikely(err
< 0)) {
1435 path_put_conditional(path
, nd
);
1439 nd
->flags
|= LOOKUP_JUMPED
;
1440 *inode
= path
->dentry
->d_inode
;
1447 /* Fast lookup failed, do it the slow way */
1448 static int lookup_slow(struct nameidata
*nd
, struct qstr
*name
,
1451 struct dentry
*dentry
, *parent
;
1454 parent
= nd
->path
.dentry
;
1455 BUG_ON(nd
->inode
!= parent
->d_inode
);
1457 mutex_lock(&parent
->d_inode
->i_mutex
);
1458 dentry
= __lookup_hash(name
, parent
, nd
->flags
);
1459 mutex_unlock(&parent
->d_inode
->i_mutex
);
1461 return PTR_ERR(dentry
);
1462 path
->mnt
= nd
->path
.mnt
;
1463 path
->dentry
= dentry
;
1464 err
= follow_managed(path
, nd
->flags
);
1465 if (unlikely(err
< 0)) {
1466 path_put_conditional(path
, nd
);
1470 nd
->flags
|= LOOKUP_JUMPED
;
1474 static inline int may_lookup(struct nameidata
*nd
)
1476 if (nd
->flags
& LOOKUP_RCU
) {
1477 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1480 if (unlazy_walk(nd
, NULL
))
1483 return inode_permission(nd
->inode
, MAY_EXEC
);
1486 static inline int handle_dots(struct nameidata
*nd
, int type
)
1488 if (type
== LAST_DOTDOT
) {
1489 if (nd
->flags
& LOOKUP_RCU
) {
1490 if (follow_dotdot_rcu(nd
))
1498 static void terminate_walk(struct nameidata
*nd
)
1500 if (!(nd
->flags
& LOOKUP_RCU
)) {
1501 path_put(&nd
->path
);
1503 nd
->flags
&= ~LOOKUP_RCU
;
1504 if (!(nd
->flags
& LOOKUP_ROOT
))
1505 nd
->root
.mnt
= NULL
;
1511 * Do we need to follow links? We _really_ want to be able
1512 * to do this check without having to look at inode->i_op,
1513 * so we keep a cache of "no, this doesn't need follow_link"
1514 * for the common case.
1516 static inline int should_follow_link(struct inode
*inode
, int follow
)
1518 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1519 if (likely(inode
->i_op
->follow_link
))
1522 /* This gets set once for the inode lifetime */
1523 spin_lock(&inode
->i_lock
);
1524 inode
->i_opflags
|= IOP_NOFOLLOW
;
1525 spin_unlock(&inode
->i_lock
);
1530 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1531 struct qstr
*name
, int type
, int follow
)
1533 struct inode
*inode
;
1536 * "." and ".." are special - ".." especially so because it has
1537 * to be able to know about the current root directory and
1538 * parent relationships.
1540 if (unlikely(type
!= LAST_NORM
))
1541 return handle_dots(nd
, type
);
1542 err
= lookup_fast(nd
, name
, path
, &inode
);
1543 if (unlikely(err
)) {
1547 err
= lookup_slow(nd
, name
, path
);
1551 inode
= path
->dentry
->d_inode
;
1557 if (should_follow_link(inode
, follow
)) {
1558 if (nd
->flags
& LOOKUP_RCU
) {
1559 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1564 BUG_ON(inode
!= path
->dentry
->d_inode
);
1567 path_to_nameidata(path
, nd
);
1572 path_to_nameidata(path
, nd
);
1579 * This limits recursive symlink follows to 8, while
1580 * limiting consecutive symlinks to 40.
1582 * Without that kind of total limit, nasty chains of consecutive
1583 * symlinks can cause almost arbitrarily long lookups.
1585 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1589 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1590 path_put_conditional(path
, nd
);
1591 path_put(&nd
->path
);
1594 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1597 current
->link_count
++;
1600 struct path link
= *path
;
1603 res
= follow_link(&link
, nd
, &cookie
);
1606 res
= walk_component(nd
, path
, &nd
->last
,
1607 nd
->last_type
, LOOKUP_FOLLOW
);
1608 put_link(nd
, &link
, cookie
);
1611 current
->link_count
--;
1617 * We really don't want to look at inode->i_op->lookup
1618 * when we don't have to. So we keep a cache bit in
1619 * the inode ->i_opflags field that says "yes, we can
1620 * do lookup on this inode".
1622 static inline int can_lookup(struct inode
*inode
)
1624 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1626 if (likely(!inode
->i_op
->lookup
))
1629 /* We do this once for the lifetime of the inode */
1630 spin_lock(&inode
->i_lock
);
1631 inode
->i_opflags
|= IOP_LOOKUP
;
1632 spin_unlock(&inode
->i_lock
);
1637 * We can do the critical dentry name comparison and hashing
1638 * operations one word at a time, but we are limited to:
1640 * - Architectures with fast unaligned word accesses. We could
1641 * do a "get_unaligned()" if this helps and is sufficiently
1644 * - Little-endian machines (so that we can generate the mask
1645 * of low bytes efficiently). Again, we *could* do a byte
1646 * swapping load on big-endian architectures if that is not
1647 * expensive enough to make the optimization worthless.
1649 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1650 * do not trap on the (extremely unlikely) case of a page
1651 * crossing operation.
1653 * - Furthermore, we need an efficient 64-bit compile for the
1654 * 64-bit case in order to generate the "number of bytes in
1655 * the final mask". Again, that could be replaced with a
1656 * efficient population count instruction or similar.
1658 #ifdef CONFIG_DCACHE_WORD_ACCESS
1660 #include <asm/word-at-a-time.h>
1664 static inline unsigned int fold_hash(unsigned long hash
)
1666 hash
+= hash
>> (8*sizeof(int));
1670 #else /* 32-bit case */
1672 #define fold_hash(x) (x)
1676 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1678 unsigned long a
, mask
;
1679 unsigned long hash
= 0;
1682 a
= load_unaligned_zeropad(name
);
1683 if (len
< sizeof(unsigned long))
1687 name
+= sizeof(unsigned long);
1688 len
-= sizeof(unsigned long);
1692 mask
= ~(~0ul << len
*8);
1695 return fold_hash(hash
);
1697 EXPORT_SYMBOL(full_name_hash
);
1700 * Calculate the length and hash of the path component, and
1701 * return the length of the component;
1703 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1705 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1706 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1709 len
= -sizeof(unsigned long);
1711 hash
= (hash
+ a
) * 9;
1712 len
+= sizeof(unsigned long);
1713 a
= load_unaligned_zeropad(name
+len
);
1714 b
= a
^ REPEAT_BYTE('/');
1715 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1717 adata
= prep_zero_mask(a
, adata
, &constants
);
1718 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1720 mask
= create_zero_mask(adata
| bdata
);
1722 hash
+= a
& zero_bytemask(mask
);
1723 *hashp
= fold_hash(hash
);
1725 return len
+ find_zero(mask
);
1730 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1732 unsigned long hash
= init_name_hash();
1734 hash
= partial_name_hash(*name
++, hash
);
1735 return end_name_hash(hash
);
1737 EXPORT_SYMBOL(full_name_hash
);
1740 * We know there's a real path component here of at least
1743 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1745 unsigned long hash
= init_name_hash();
1746 unsigned long len
= 0, c
;
1748 c
= (unsigned char)*name
;
1751 hash
= partial_name_hash(c
, hash
);
1752 c
= (unsigned char)name
[len
];
1753 } while (c
&& c
!= '/');
1754 *hashp
= end_name_hash(hash
);
1762 * This is the basic name resolution function, turning a pathname into
1763 * the final dentry. We expect 'base' to be positive and a directory.
1765 * Returns 0 and nd will have valid dentry and mnt on success.
1766 * Returns error and drops reference to input namei data on failure.
1768 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1778 /* At this point we know we have a real path component. */
1784 err
= may_lookup(nd
);
1788 len
= hash_name(name
, &this.hash
);
1793 if (name
[0] == '.') switch (len
) {
1795 if (name
[1] == '.') {
1797 nd
->flags
|= LOOKUP_JUMPED
;
1803 if (likely(type
== LAST_NORM
)) {
1804 struct dentry
*parent
= nd
->path
.dentry
;
1805 nd
->flags
&= ~LOOKUP_JUMPED
;
1806 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1807 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1815 goto last_component
;
1817 * If it wasn't NUL, we know it was '/'. Skip that
1818 * slash, and continue until no more slashes.
1822 } while (unlikely(name
[len
] == '/'));
1824 goto last_component
;
1827 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1832 err
= nested_symlink(&next
, nd
);
1836 if (can_lookup(nd
->inode
))
1840 /* here ends the main loop */
1844 nd
->last_type
= type
;
1851 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1852 struct nameidata
*nd
, struct file
**fp
)
1856 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1857 nd
->flags
= flags
| LOOKUP_JUMPED
;
1859 if (flags
& LOOKUP_ROOT
) {
1860 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1862 if (!inode
->i_op
->lookup
)
1864 retval
= inode_permission(inode
, MAY_EXEC
);
1868 nd
->path
= nd
->root
;
1870 if (flags
& LOOKUP_RCU
) {
1872 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1874 path_get(&nd
->path
);
1879 nd
->root
.mnt
= NULL
;
1882 if (flags
& LOOKUP_RCU
) {
1887 path_get(&nd
->root
);
1889 nd
->path
= nd
->root
;
1890 } else if (dfd
== AT_FDCWD
) {
1891 if (flags
& LOOKUP_RCU
) {
1892 struct fs_struct
*fs
= current
->fs
;
1898 seq
= read_seqcount_begin(&fs
->seq
);
1900 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1901 } while (read_seqcount_retry(&fs
->seq
, seq
));
1903 get_fs_pwd(current
->fs
, &nd
->path
);
1906 struct fd f
= fdget_raw(dfd
);
1907 struct dentry
*dentry
;
1912 dentry
= f
.file
->f_path
.dentry
;
1915 if (!S_ISDIR(dentry
->d_inode
->i_mode
)) {
1920 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1927 nd
->path
= f
.file
->f_path
;
1928 if (flags
& LOOKUP_RCU
) {
1931 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1934 path_get(&nd
->path
);
1939 nd
->inode
= nd
->path
.dentry
->d_inode
;
1943 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1945 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1946 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1948 nd
->flags
&= ~LOOKUP_PARENT
;
1949 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1950 nd
->flags
& LOOKUP_FOLLOW
);
1953 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1954 static int path_lookupat(int dfd
, const char *name
,
1955 unsigned int flags
, struct nameidata
*nd
)
1957 struct file
*base
= NULL
;
1962 * Path walking is largely split up into 2 different synchronisation
1963 * schemes, rcu-walk and ref-walk (explained in
1964 * Documentation/filesystems/path-lookup.txt). These share much of the
1965 * path walk code, but some things particularly setup, cleanup, and
1966 * following mounts are sufficiently divergent that functions are
1967 * duplicated. Typically there is a function foo(), and its RCU
1968 * analogue, foo_rcu().
1970 * -ECHILD is the error number of choice (just to avoid clashes) that
1971 * is returned if some aspect of an rcu-walk fails. Such an error must
1972 * be handled by restarting a traditional ref-walk (which will always
1973 * be able to complete).
1975 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1980 current
->total_link_count
= 0;
1981 err
= link_path_walk(name
, nd
);
1983 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1984 err
= lookup_last(nd
, &path
);
1987 struct path link
= path
;
1988 err
= may_follow_link(&link
, nd
);
1991 nd
->flags
|= LOOKUP_PARENT
;
1992 err
= follow_link(&link
, nd
, &cookie
);
1995 err
= lookup_last(nd
, &path
);
1996 put_link(nd
, &link
, cookie
);
2001 err
= complete_walk(nd
);
2003 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2004 if (!nd
->inode
->i_op
->lookup
) {
2005 path_put(&nd
->path
);
2013 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2014 path_put(&nd
->root
);
2015 nd
->root
.mnt
= NULL
;
2020 static int filename_lookup(int dfd
, struct filename
*name
,
2021 unsigned int flags
, struct nameidata
*nd
)
2023 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2024 if (unlikely(retval
== -ECHILD
))
2025 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2026 if (unlikely(retval
== -ESTALE
))
2027 retval
= path_lookupat(dfd
, name
->name
,
2028 flags
| LOOKUP_REVAL
, nd
);
2030 if (likely(!retval
))
2031 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2035 static int do_path_lookup(int dfd
, const char *name
,
2036 unsigned int flags
, struct nameidata
*nd
)
2038 struct filename filename
= { .name
= name
};
2040 return filename_lookup(dfd
, &filename
, flags
, nd
);
2043 /* does lookup, returns the object with parent locked */
2044 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2046 struct nameidata nd
;
2048 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2050 return ERR_PTR(err
);
2051 if (nd
.last_type
!= LAST_NORM
) {
2053 return ERR_PTR(-EINVAL
);
2055 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2056 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2058 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2066 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2068 struct nameidata nd
;
2069 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2076 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2077 * @dentry: pointer to dentry of the base directory
2078 * @mnt: pointer to vfs mount of the base directory
2079 * @name: pointer to file name
2080 * @flags: lookup flags
2081 * @path: pointer to struct path to fill
2083 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2084 const char *name
, unsigned int flags
,
2087 struct nameidata nd
;
2089 nd
.root
.dentry
= dentry
;
2091 BUG_ON(flags
& LOOKUP_PARENT
);
2092 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2093 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2100 * Restricted form of lookup. Doesn't follow links, single-component only,
2101 * needs parent already locked. Doesn't follow mounts.
2104 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2106 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2110 * lookup_one_len - filesystem helper to lookup single pathname component
2111 * @name: pathname component to lookup
2112 * @base: base directory to lookup from
2113 * @len: maximum length @len should be interpreted to
2115 * Note that this routine is purely a helper for filesystem usage and should
2116 * not be called by generic code. Also note that by using this function the
2117 * nameidata argument is passed to the filesystem methods and a filesystem
2118 * using this helper needs to be prepared for that.
2120 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2126 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2130 this.hash
= full_name_hash(name
, len
);
2132 return ERR_PTR(-EACCES
);
2135 c
= *(const unsigned char *)name
++;
2136 if (c
== '/' || c
== '\0')
2137 return ERR_PTR(-EACCES
);
2140 * See if the low-level filesystem might want
2141 * to use its own hash..
2143 if (base
->d_flags
& DCACHE_OP_HASH
) {
2144 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2146 return ERR_PTR(err
);
2149 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2151 return ERR_PTR(err
);
2153 return __lookup_hash(&this, base
, 0);
2156 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2157 struct path
*path
, int *empty
)
2159 struct nameidata nd
;
2160 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2161 int err
= PTR_ERR(tmp
);
2164 BUG_ON(flags
& LOOKUP_PARENT
);
2166 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2174 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2177 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2181 * NB: most callers don't do anything directly with the reference to the
2182 * to struct filename, but the nd->last pointer points into the name string
2183 * allocated by getname. So we must hold the reference to it until all
2184 * path-walking is complete.
2186 static struct filename
*
2187 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
)
2189 struct filename
*s
= getname(path
);
2195 error
= filename_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
2198 return ERR_PTR(error
);
2205 * It's inline, so penalty for filesystems that don't use sticky bit is
2208 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2210 kuid_t fsuid
= current_fsuid();
2212 if (!(dir
->i_mode
& S_ISVTX
))
2214 if (uid_eq(inode
->i_uid
, fsuid
))
2216 if (uid_eq(dir
->i_uid
, fsuid
))
2218 return !inode_capable(inode
, CAP_FOWNER
);
2222 * Check whether we can remove a link victim from directory dir, check
2223 * whether the type of victim is right.
2224 * 1. We can't do it if dir is read-only (done in permission())
2225 * 2. We should have write and exec permissions on dir
2226 * 3. We can't remove anything from append-only dir
2227 * 4. We can't do anything with immutable dir (done in permission())
2228 * 5. If the sticky bit on dir is set we should either
2229 * a. be owner of dir, or
2230 * b. be owner of victim, or
2231 * c. have CAP_FOWNER capability
2232 * 6. If the victim is append-only or immutable we can't do antyhing with
2233 * links pointing to it.
2234 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2235 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2236 * 9. We can't remove a root or mountpoint.
2237 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2238 * nfs_async_unlink().
2240 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2244 if (!victim
->d_inode
)
2247 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2248 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2250 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2255 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2256 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2259 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2261 if (IS_ROOT(victim
))
2263 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2265 if (IS_DEADDIR(dir
))
2267 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2272 /* Check whether we can create an object with dentry child in directory
2274 * 1. We can't do it if child already exists (open has special treatment for
2275 * this case, but since we are inlined it's OK)
2276 * 2. We can't do it if dir is read-only (done in permission())
2277 * 3. We should have write and exec permissions on dir
2278 * 4. We can't do it if dir is immutable (done in permission())
2280 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2284 if (IS_DEADDIR(dir
))
2286 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2290 * p1 and p2 should be directories on the same fs.
2292 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2297 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2301 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2303 p
= d_ancestor(p2
, p1
);
2305 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2306 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2310 p
= d_ancestor(p1
, p2
);
2312 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2313 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2317 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2318 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2322 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2324 mutex_unlock(&p1
->d_inode
->i_mutex
);
2326 mutex_unlock(&p2
->d_inode
->i_mutex
);
2327 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2331 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2334 int error
= may_create(dir
, dentry
);
2338 if (!dir
->i_op
->create
)
2339 return -EACCES
; /* shouldn't it be ENOSYS? */
2342 error
= security_inode_create(dir
, dentry
, mode
);
2345 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2347 fsnotify_create(dir
, dentry
);
2351 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2353 struct dentry
*dentry
= path
->dentry
;
2354 struct inode
*inode
= dentry
->d_inode
;
2364 switch (inode
->i_mode
& S_IFMT
) {
2368 if (acc_mode
& MAY_WRITE
)
2373 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2382 error
= inode_permission(inode
, acc_mode
);
2387 * An append-only file must be opened in append mode for writing.
2389 if (IS_APPEND(inode
)) {
2390 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2396 /* O_NOATIME can only be set by the owner or superuser */
2397 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2403 static int handle_truncate(struct file
*filp
)
2405 struct path
*path
= &filp
->f_path
;
2406 struct inode
*inode
= path
->dentry
->d_inode
;
2407 int error
= get_write_access(inode
);
2411 * Refuse to truncate files with mandatory locks held on them.
2413 error
= locks_verify_locked(inode
);
2415 error
= security_path_truncate(path
);
2417 error
= do_truncate(path
->dentry
, 0,
2418 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2421 put_write_access(inode
);
2425 static inline int open_to_namei_flags(int flag
)
2427 if ((flag
& O_ACCMODE
) == 3)
2432 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2434 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2438 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2442 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2446 * Attempt to atomically look up, create and open a file from a negative
2449 * Returns 0 if successful. The file will have been created and attached to
2450 * @file by the filesystem calling finish_open().
2452 * Returns 1 if the file was looked up only or didn't need creating. The
2453 * caller will need to perform the open themselves. @path will have been
2454 * updated to point to the new dentry. This may be negative.
2456 * Returns an error code otherwise.
2458 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2459 struct path
*path
, struct file
*file
,
2460 const struct open_flags
*op
,
2461 bool got_write
, bool need_lookup
,
2464 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2465 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2469 int create_error
= 0;
2470 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2472 BUG_ON(dentry
->d_inode
);
2474 /* Don't create child dentry for a dead directory. */
2475 if (unlikely(IS_DEADDIR(dir
))) {
2481 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2482 mode
&= ~current_umask();
2484 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2485 open_flag
&= ~O_TRUNC
;
2486 *opened
|= FILE_CREATED
;
2490 * Checking write permission is tricky, bacuse we don't know if we are
2491 * going to actually need it: O_CREAT opens should work as long as the
2492 * file exists. But checking existence breaks atomicity. The trick is
2493 * to check access and if not granted clear O_CREAT from the flags.
2495 * Another problem is returing the "right" error value (e.g. for an
2496 * O_EXCL open we want to return EEXIST not EROFS).
2498 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2499 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2500 if (!(open_flag
& O_CREAT
)) {
2502 * No O_CREATE -> atomicity not a requirement -> fall
2503 * back to lookup + open
2506 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2507 /* Fall back and fail with the right error */
2508 create_error
= -EROFS
;
2511 /* No side effects, safe to clear O_CREAT */
2512 create_error
= -EROFS
;
2513 open_flag
&= ~O_CREAT
;
2517 if (open_flag
& O_CREAT
) {
2518 error
= may_o_create(&nd
->path
, dentry
, mode
);
2520 create_error
= error
;
2521 if (open_flag
& O_EXCL
)
2523 open_flag
&= ~O_CREAT
;
2527 if (nd
->flags
& LOOKUP_DIRECTORY
)
2528 open_flag
|= O_DIRECTORY
;
2530 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2531 file
->f_path
.mnt
= nd
->path
.mnt
;
2532 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2535 if (create_error
&& error
== -ENOENT
)
2536 error
= create_error
;
2540 acc_mode
= op
->acc_mode
;
2541 if (*opened
& FILE_CREATED
) {
2542 fsnotify_create(dir
, dentry
);
2543 acc_mode
= MAY_OPEN
;
2546 if (error
) { /* returned 1, that is */
2547 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2551 if (file
->f_path
.dentry
) {
2553 dentry
= file
->f_path
.dentry
;
2555 if (create_error
&& dentry
->d_inode
== NULL
) {
2556 error
= create_error
;
2563 * We didn't have the inode before the open, so check open permission
2566 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2576 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2578 return PTR_ERR(dentry
);
2581 int open_flag
= op
->open_flag
;
2583 error
= create_error
;
2584 if ((open_flag
& O_EXCL
)) {
2585 if (!dentry
->d_inode
)
2587 } else if (!dentry
->d_inode
) {
2589 } else if ((open_flag
& O_TRUNC
) &&
2590 S_ISREG(dentry
->d_inode
->i_mode
)) {
2593 /* will fail later, go on to get the right error */
2597 path
->dentry
= dentry
;
2598 path
->mnt
= nd
->path
.mnt
;
2603 * Look up and maybe create and open the last component.
2605 * Must be called with i_mutex held on parent.
2607 * Returns 0 if the file was successfully atomically created (if necessary) and
2608 * opened. In this case the file will be returned attached to @file.
2610 * Returns 1 if the file was not completely opened at this time, though lookups
2611 * and creations will have been performed and the dentry returned in @path will
2612 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2613 * specified then a negative dentry may be returned.
2615 * An error code is returned otherwise.
2617 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2618 * cleared otherwise prior to returning.
2620 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2622 const struct open_flags
*op
,
2623 bool got_write
, int *opened
)
2625 struct dentry
*dir
= nd
->path
.dentry
;
2626 struct inode
*dir_inode
= dir
->d_inode
;
2627 struct dentry
*dentry
;
2631 *opened
&= ~FILE_CREATED
;
2632 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2634 return PTR_ERR(dentry
);
2636 /* Cached positive dentry: will open in f_op->open */
2637 if (!need_lookup
&& dentry
->d_inode
)
2640 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2641 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2642 need_lookup
, opened
);
2646 BUG_ON(dentry
->d_inode
);
2648 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2650 return PTR_ERR(dentry
);
2653 /* Negative dentry, just create the file */
2654 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2655 umode_t mode
= op
->mode
;
2656 if (!IS_POSIXACL(dir
->d_inode
))
2657 mode
&= ~current_umask();
2659 * This write is needed to ensure that a
2660 * rw->ro transition does not occur between
2661 * the time when the file is created and when
2662 * a permanent write count is taken through
2663 * the 'struct file' in finish_open().
2669 *opened
|= FILE_CREATED
;
2670 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2673 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2674 nd
->flags
& LOOKUP_EXCL
);
2679 path
->dentry
= dentry
;
2680 path
->mnt
= nd
->path
.mnt
;
2689 * Handle the last step of open()
2691 static int do_last(struct nameidata
*nd
, struct path
*path
,
2692 struct file
*file
, const struct open_flags
*op
,
2693 int *opened
, struct filename
*name
)
2695 struct dentry
*dir
= nd
->path
.dentry
;
2696 int open_flag
= op
->open_flag
;
2697 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2698 bool got_write
= false;
2699 int acc_mode
= op
->acc_mode
;
2700 struct inode
*inode
;
2701 bool symlink_ok
= false;
2702 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2703 bool retried
= false;
2706 nd
->flags
&= ~LOOKUP_PARENT
;
2707 nd
->flags
|= op
->intent
;
2709 switch (nd
->last_type
) {
2712 error
= handle_dots(nd
, nd
->last_type
);
2717 error
= complete_walk(nd
);
2720 audit_inode(name
, nd
->path
.dentry
, 0);
2721 if (open_flag
& O_CREAT
) {
2727 error
= complete_walk(nd
);
2730 audit_inode(name
, dir
, 0);
2734 if (!(open_flag
& O_CREAT
)) {
2735 if (nd
->last
.name
[nd
->last
.len
])
2736 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2737 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2739 /* we _can_ be in RCU mode here */
2740 error
= lookup_fast(nd
, &nd
->last
, path
, &inode
);
2747 BUG_ON(nd
->inode
!= dir
->d_inode
);
2749 /* create side of things */
2751 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2752 * has been cleared when we got to the last component we are
2755 error
= complete_walk(nd
);
2759 audit_inode(name
, dir
, 0);
2761 /* trailing slashes? */
2762 if (nd
->last
.name
[nd
->last
.len
])
2767 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2768 error
= mnt_want_write(nd
->path
.mnt
);
2772 * do _not_ fail yet - we might not need that or fail with
2773 * a different error; let lookup_open() decide; we'll be
2774 * dropping this one anyway.
2777 mutex_lock(&dir
->d_inode
->i_mutex
);
2778 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2779 mutex_unlock(&dir
->d_inode
->i_mutex
);
2785 if ((*opened
& FILE_CREATED
) ||
2786 !S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
2787 will_truncate
= false;
2789 audit_inode(name
, file
->f_path
.dentry
, 0);
2793 if (*opened
& FILE_CREATED
) {
2794 /* Don't check for write permission, don't truncate */
2795 open_flag
&= ~O_TRUNC
;
2796 will_truncate
= false;
2797 acc_mode
= MAY_OPEN
;
2798 path_to_nameidata(path
, nd
);
2799 goto finish_open_created
;
2803 * create/update audit record if it already exists.
2805 if (path
->dentry
->d_inode
)
2806 audit_inode(name
, path
->dentry
, 0);
2809 * If atomic_open() acquired write access it is dropped now due to
2810 * possible mount and symlink following (this might be optimized away if
2814 mnt_drop_write(nd
->path
.mnt
);
2819 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2822 error
= follow_managed(path
, nd
->flags
);
2827 nd
->flags
|= LOOKUP_JUMPED
;
2829 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2830 inode
= path
->dentry
->d_inode
;
2832 /* we _can_ be in RCU mode here */
2835 path_to_nameidata(path
, nd
);
2839 if (should_follow_link(inode
, !symlink_ok
)) {
2840 if (nd
->flags
& LOOKUP_RCU
) {
2841 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2846 BUG_ON(inode
!= path
->dentry
->d_inode
);
2850 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2851 path_to_nameidata(path
, nd
);
2853 save_parent
.dentry
= nd
->path
.dentry
;
2854 save_parent
.mnt
= mntget(path
->mnt
);
2855 nd
->path
.dentry
= path
->dentry
;
2859 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2860 error
= complete_walk(nd
);
2862 path_put(&save_parent
);
2866 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2869 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !nd
->inode
->i_op
->lookup
)
2871 audit_inode(name
, nd
->path
.dentry
, 0);
2873 if (!S_ISREG(nd
->inode
->i_mode
))
2874 will_truncate
= false;
2876 if (will_truncate
) {
2877 error
= mnt_want_write(nd
->path
.mnt
);
2882 finish_open_created
:
2883 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2886 file
->f_path
.mnt
= nd
->path
.mnt
;
2887 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2889 if (error
== -EOPENSTALE
)
2894 error
= open_check_o_direct(file
);
2897 error
= ima_file_check(file
, op
->acc_mode
);
2901 if (will_truncate
) {
2902 error
= handle_truncate(file
);
2908 mnt_drop_write(nd
->path
.mnt
);
2909 path_put(&save_parent
);
2914 path_put_conditional(path
, nd
);
2921 /* If no saved parent or already retried then can't retry */
2922 if (!save_parent
.dentry
|| retried
)
2925 BUG_ON(save_parent
.dentry
!= dir
);
2926 path_put(&nd
->path
);
2927 nd
->path
= save_parent
;
2928 nd
->inode
= dir
->d_inode
;
2929 save_parent
.mnt
= NULL
;
2930 save_parent
.dentry
= NULL
;
2932 mnt_drop_write(nd
->path
.mnt
);
2939 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
2940 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2942 struct file
*base
= NULL
;
2948 file
= get_empty_filp();
2950 return ERR_PTR(-ENFILE
);
2952 file
->f_flags
= op
->open_flag
;
2954 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
2955 if (unlikely(error
))
2958 current
->total_link_count
= 0;
2959 error
= link_path_walk(pathname
->name
, nd
);
2960 if (unlikely(error
))
2963 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2964 while (unlikely(error
> 0)) { /* trailing symlink */
2965 struct path link
= path
;
2967 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2968 path_put_conditional(&path
, nd
);
2969 path_put(&nd
->path
);
2973 error
= may_follow_link(&link
, nd
);
2974 if (unlikely(error
))
2976 nd
->flags
|= LOOKUP_PARENT
;
2977 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2978 error
= follow_link(&link
, nd
, &cookie
);
2979 if (unlikely(error
))
2981 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2982 put_link(nd
, &link
, cookie
);
2985 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2986 path_put(&nd
->root
);
2989 if (!(opened
& FILE_OPENED
)) {
2993 if (unlikely(error
)) {
2994 if (error
== -EOPENSTALE
) {
2995 if (flags
& LOOKUP_RCU
)
3000 file
= ERR_PTR(error
);
3005 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3006 const struct open_flags
*op
, int flags
)
3008 struct nameidata nd
;
3011 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3012 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3013 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3014 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3015 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3019 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3020 const char *name
, const struct open_flags
*op
, int flags
)
3022 struct nameidata nd
;
3024 struct filename filename
= { .name
= name
};
3027 nd
.root
.dentry
= dentry
;
3029 flags
|= LOOKUP_ROOT
;
3031 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3032 return ERR_PTR(-ELOOP
);
3034 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3035 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3036 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3037 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3038 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3042 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
3044 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3045 struct nameidata nd
;
3047 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
3049 return ERR_PTR(error
);
3052 * Yucky last component or no last component at all?
3053 * (foo/., foo/.., /////)
3055 if (nd
.last_type
!= LAST_NORM
)
3057 nd
.flags
&= ~LOOKUP_PARENT
;
3058 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3060 /* don't fail immediately if it's r/o, at least try to report other errors */
3061 err2
= mnt_want_write(nd
.path
.mnt
);
3063 * Do the final lookup.
3065 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3066 dentry
= lookup_hash(&nd
);
3071 if (dentry
->d_inode
)
3074 * Special case - lookup gave negative, but... we had foo/bar/
3075 * From the vfs_mknod() POV we just have a negative dentry -
3076 * all is fine. Let's be bastards - you had / on the end, you've
3077 * been asking for (non-existent) directory. -ENOENT for you.
3079 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3083 if (unlikely(err2
)) {
3091 dentry
= ERR_PTR(error
);
3093 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3095 mnt_drop_write(nd
.path
.mnt
);
3100 EXPORT_SYMBOL(kern_path_create
);
3102 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3105 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3106 mnt_drop_write(path
->mnt
);
3109 EXPORT_SYMBOL(done_path_create
);
3111 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
3113 struct filename
*tmp
= getname(pathname
);
3116 return ERR_CAST(tmp
);
3117 res
= kern_path_create(dfd
, tmp
->name
, path
, is_dir
);
3121 EXPORT_SYMBOL(user_path_create
);
3123 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3125 int error
= may_create(dir
, dentry
);
3130 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3133 if (!dir
->i_op
->mknod
)
3136 error
= devcgroup_inode_mknod(mode
, dev
);
3140 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3144 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3146 fsnotify_create(dir
, dentry
);
3150 static int may_mknod(umode_t mode
)
3152 switch (mode
& S_IFMT
) {
3158 case 0: /* zero mode translates to S_IFREG */
3167 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3170 struct dentry
*dentry
;
3174 error
= may_mknod(mode
);
3178 dentry
= user_path_create(dfd
, filename
, &path
, 0);
3180 return PTR_ERR(dentry
);
3182 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3183 mode
&= ~current_umask();
3184 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3187 switch (mode
& S_IFMT
) {
3188 case 0: case S_IFREG
:
3189 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3191 case S_IFCHR
: case S_IFBLK
:
3192 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3193 new_decode_dev(dev
));
3195 case S_IFIFO
: case S_IFSOCK
:
3196 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3200 done_path_create(&path
, dentry
);
3204 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3206 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3209 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3211 int error
= may_create(dir
, dentry
);
3212 unsigned max_links
= dir
->i_sb
->s_max_links
;
3217 if (!dir
->i_op
->mkdir
)
3220 mode
&= (S_IRWXUGO
|S_ISVTX
);
3221 error
= security_inode_mkdir(dir
, dentry
, mode
);
3225 if (max_links
&& dir
->i_nlink
>= max_links
)
3228 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3230 fsnotify_mkdir(dir
, dentry
);
3234 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3236 struct dentry
*dentry
;
3240 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
3242 return PTR_ERR(dentry
);
3244 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3245 mode
&= ~current_umask();
3246 error
= security_path_mkdir(&path
, dentry
, mode
);
3248 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3249 done_path_create(&path
, dentry
);
3253 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3255 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3259 * The dentry_unhash() helper will try to drop the dentry early: we
3260 * should have a usage count of 1 if we're the only user of this
3261 * dentry, and if that is true (possibly after pruning the dcache),
3262 * then we drop the dentry now.
3264 * A low-level filesystem can, if it choses, legally
3267 * if (!d_unhashed(dentry))
3270 * if it cannot handle the case of removing a directory
3271 * that is still in use by something else..
3273 void dentry_unhash(struct dentry
*dentry
)
3275 shrink_dcache_parent(dentry
);
3276 spin_lock(&dentry
->d_lock
);
3277 if (dentry
->d_count
== 1)
3279 spin_unlock(&dentry
->d_lock
);
3282 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3284 int error
= may_delete(dir
, dentry
, 1);
3289 if (!dir
->i_op
->rmdir
)
3293 mutex_lock(&dentry
->d_inode
->i_mutex
);
3296 if (d_mountpoint(dentry
))
3299 error
= security_inode_rmdir(dir
, dentry
);
3303 shrink_dcache_parent(dentry
);
3304 error
= dir
->i_op
->rmdir(dir
, dentry
);
3308 dentry
->d_inode
->i_flags
|= S_DEAD
;
3312 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3319 static long do_rmdir(int dfd
, const char __user
*pathname
)
3322 struct filename
*name
;
3323 struct dentry
*dentry
;
3324 struct nameidata nd
;
3326 name
= user_path_parent(dfd
, pathname
, &nd
);
3328 return PTR_ERR(name
);
3330 switch(nd
.last_type
) {
3342 nd
.flags
&= ~LOOKUP_PARENT
;
3343 error
= mnt_want_write(nd
.path
.mnt
);
3347 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3348 dentry
= lookup_hash(&nd
);
3349 error
= PTR_ERR(dentry
);
3352 if (!dentry
->d_inode
) {
3356 error
= security_path_rmdir(&nd
.path
, dentry
);
3359 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3363 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3364 mnt_drop_write(nd
.path
.mnt
);
3371 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3373 return do_rmdir(AT_FDCWD
, pathname
);
3376 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3378 int error
= may_delete(dir
, dentry
, 0);
3383 if (!dir
->i_op
->unlink
)
3386 mutex_lock(&dentry
->d_inode
->i_mutex
);
3387 if (d_mountpoint(dentry
))
3390 error
= security_inode_unlink(dir
, dentry
);
3392 error
= dir
->i_op
->unlink(dir
, dentry
);
3397 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3399 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3400 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3401 fsnotify_link_count(dentry
->d_inode
);
3409 * Make sure that the actual truncation of the file will occur outside its
3410 * directory's i_mutex. Truncate can take a long time if there is a lot of
3411 * writeout happening, and we don't want to prevent access to the directory
3412 * while waiting on the I/O.
3414 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3417 struct filename
*name
;
3418 struct dentry
*dentry
;
3419 struct nameidata nd
;
3420 struct inode
*inode
= NULL
;
3422 name
= user_path_parent(dfd
, pathname
, &nd
);
3424 return PTR_ERR(name
);
3427 if (nd
.last_type
!= LAST_NORM
)
3430 nd
.flags
&= ~LOOKUP_PARENT
;
3431 error
= mnt_want_write(nd
.path
.mnt
);
3435 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3436 dentry
= lookup_hash(&nd
);
3437 error
= PTR_ERR(dentry
);
3438 if (!IS_ERR(dentry
)) {
3439 /* Why not before? Because we want correct error value */
3440 if (nd
.last
.name
[nd
.last
.len
])
3442 inode
= dentry
->d_inode
;
3446 error
= security_path_unlink(&nd
.path
, dentry
);
3449 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3453 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3455 iput(inode
); /* truncate the inode here */
3456 mnt_drop_write(nd
.path
.mnt
);
3463 error
= !dentry
->d_inode
? -ENOENT
:
3464 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3468 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3470 if ((flag
& ~AT_REMOVEDIR
) != 0)
3473 if (flag
& AT_REMOVEDIR
)
3474 return do_rmdir(dfd
, pathname
);
3476 return do_unlinkat(dfd
, pathname
);
3479 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3481 return do_unlinkat(AT_FDCWD
, pathname
);
3484 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3486 int error
= may_create(dir
, dentry
);
3491 if (!dir
->i_op
->symlink
)
3494 error
= security_inode_symlink(dir
, dentry
, oldname
);
3498 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3500 fsnotify_create(dir
, dentry
);
3504 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3505 int, newdfd
, const char __user
*, newname
)
3508 struct filename
*from
;
3509 struct dentry
*dentry
;
3512 from
= getname(oldname
);
3514 return PTR_ERR(from
);
3516 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
3517 error
= PTR_ERR(dentry
);
3521 error
= security_path_symlink(&path
, dentry
, from
->name
);
3523 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3524 done_path_create(&path
, dentry
);
3530 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3532 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3535 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3537 struct inode
*inode
= old_dentry
->d_inode
;
3538 unsigned max_links
= dir
->i_sb
->s_max_links
;
3544 error
= may_create(dir
, new_dentry
);
3548 if (dir
->i_sb
!= inode
->i_sb
)
3552 * A link to an append-only or immutable file cannot be created.
3554 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3556 if (!dir
->i_op
->link
)
3558 if (S_ISDIR(inode
->i_mode
))
3561 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3565 mutex_lock(&inode
->i_mutex
);
3566 /* Make sure we don't allow creating hardlink to an unlinked file */
3567 if (inode
->i_nlink
== 0)
3569 else if (max_links
&& inode
->i_nlink
>= max_links
)
3572 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3573 mutex_unlock(&inode
->i_mutex
);
3575 fsnotify_link(dir
, inode
, new_dentry
);
3580 * Hardlinks are often used in delicate situations. We avoid
3581 * security-related surprises by not following symlinks on the
3584 * We don't follow them on the oldname either to be compatible
3585 * with linux 2.0, and to avoid hard-linking to directories
3586 * and other special files. --ADM
3588 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3589 int, newdfd
, const char __user
*, newname
, int, flags
)
3591 struct dentry
*new_dentry
;
3592 struct path old_path
, new_path
;
3596 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3599 * To use null names we require CAP_DAC_READ_SEARCH
3600 * This ensures that not everyone will be able to create
3601 * handlink using the passed filedescriptor.
3603 if (flags
& AT_EMPTY_PATH
) {
3604 if (!capable(CAP_DAC_READ_SEARCH
))
3609 if (flags
& AT_SYMLINK_FOLLOW
)
3610 how
|= LOOKUP_FOLLOW
;
3612 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3616 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3617 error
= PTR_ERR(new_dentry
);
3618 if (IS_ERR(new_dentry
))
3622 if (old_path
.mnt
!= new_path
.mnt
)
3624 error
= may_linkat(&old_path
);
3625 if (unlikely(error
))
3627 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3630 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3632 done_path_create(&new_path
, new_dentry
);
3634 path_put(&old_path
);
3639 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3641 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3645 * The worst of all namespace operations - renaming directory. "Perverted"
3646 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3648 * a) we can get into loop creation. Check is done in is_subdir().
3649 * b) race potential - two innocent renames can create a loop together.
3650 * That's where 4.4 screws up. Current fix: serialization on
3651 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3653 * c) we have to lock _three_ objects - parents and victim (if it exists).
3654 * And that - after we got ->i_mutex on parents (until then we don't know
3655 * whether the target exists). Solution: try to be smart with locking
3656 * order for inodes. We rely on the fact that tree topology may change
3657 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3658 * move will be locked. Thus we can rank directories by the tree
3659 * (ancestors first) and rank all non-directories after them.
3660 * That works since everybody except rename does "lock parent, lookup,
3661 * lock child" and rename is under ->s_vfs_rename_mutex.
3662 * HOWEVER, it relies on the assumption that any object with ->lookup()
3663 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3664 * we'd better make sure that there's no link(2) for them.
3665 * d) conversion from fhandle to dentry may come in the wrong moment - when
3666 * we are removing the target. Solution: we will have to grab ->i_mutex
3667 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3668 * ->i_mutex on parents, which works but leads to some truly excessive
3671 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3672 struct inode
*new_dir
, struct dentry
*new_dentry
)
3675 struct inode
*target
= new_dentry
->d_inode
;
3676 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3679 * If we are going to change the parent - check write permissions,
3680 * we'll need to flip '..'.
3682 if (new_dir
!= old_dir
) {
3683 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3688 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3694 mutex_lock(&target
->i_mutex
);
3697 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3701 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3702 new_dir
->i_nlink
>= max_links
)
3706 shrink_dcache_parent(new_dentry
);
3707 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3712 target
->i_flags
|= S_DEAD
;
3713 dont_mount(new_dentry
);
3717 mutex_unlock(&target
->i_mutex
);
3720 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3721 d_move(old_dentry
,new_dentry
);
3725 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3726 struct inode
*new_dir
, struct dentry
*new_dentry
)
3728 struct inode
*target
= new_dentry
->d_inode
;
3731 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3737 mutex_lock(&target
->i_mutex
);
3740 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3743 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3748 dont_mount(new_dentry
);
3749 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3750 d_move(old_dentry
, new_dentry
);
3753 mutex_unlock(&target
->i_mutex
);
3758 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3759 struct inode
*new_dir
, struct dentry
*new_dentry
)
3762 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3763 const unsigned char *old_name
;
3765 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3768 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3772 if (!new_dentry
->d_inode
)
3773 error
= may_create(new_dir
, new_dentry
);
3775 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3779 if (!old_dir
->i_op
->rename
)
3782 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3785 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3787 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3789 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3790 new_dentry
->d_inode
, old_dentry
);
3791 fsnotify_oldname_free(old_name
);
3796 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3797 int, newdfd
, const char __user
*, newname
)
3799 struct dentry
*old_dir
, *new_dir
;
3800 struct dentry
*old_dentry
, *new_dentry
;
3801 struct dentry
*trap
;
3802 struct nameidata oldnd
, newnd
;
3803 struct filename
*from
;
3804 struct filename
*to
;
3807 from
= user_path_parent(olddfd
, oldname
, &oldnd
);
3809 error
= PTR_ERR(from
);
3813 to
= user_path_parent(newdfd
, newname
, &newnd
);
3815 error
= PTR_ERR(to
);
3820 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3823 old_dir
= oldnd
.path
.dentry
;
3825 if (oldnd
.last_type
!= LAST_NORM
)
3828 new_dir
= newnd
.path
.dentry
;
3829 if (newnd
.last_type
!= LAST_NORM
)
3832 error
= mnt_want_write(oldnd
.path
.mnt
);
3836 oldnd
.flags
&= ~LOOKUP_PARENT
;
3837 newnd
.flags
&= ~LOOKUP_PARENT
;
3838 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3840 trap
= lock_rename(new_dir
, old_dir
);
3842 old_dentry
= lookup_hash(&oldnd
);
3843 error
= PTR_ERR(old_dentry
);
3844 if (IS_ERR(old_dentry
))
3846 /* source must exist */
3848 if (!old_dentry
->d_inode
)
3850 /* unless the source is a directory trailing slashes give -ENOTDIR */
3851 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3853 if (oldnd
.last
.name
[oldnd
.last
.len
])
3855 if (newnd
.last
.name
[newnd
.last
.len
])
3858 /* source should not be ancestor of target */
3860 if (old_dentry
== trap
)
3862 new_dentry
= lookup_hash(&newnd
);
3863 error
= PTR_ERR(new_dentry
);
3864 if (IS_ERR(new_dentry
))
3866 /* target should not be an ancestor of source */
3868 if (new_dentry
== trap
)
3871 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3872 &newnd
.path
, new_dentry
);
3875 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3876 new_dir
->d_inode
, new_dentry
);
3882 unlock_rename(new_dir
, old_dir
);
3883 mnt_drop_write(oldnd
.path
.mnt
);
3885 path_put(&newnd
.path
);
3888 path_put(&oldnd
.path
);
3894 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3896 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3899 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3903 len
= PTR_ERR(link
);
3908 if (len
> (unsigned) buflen
)
3910 if (copy_to_user(buffer
, link
, len
))
3917 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3918 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3919 * using) it for any given inode is up to filesystem.
3921 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3923 struct nameidata nd
;
3928 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3930 return PTR_ERR(cookie
);
3932 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3933 if (dentry
->d_inode
->i_op
->put_link
)
3934 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3938 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3940 return __vfs_follow_link(nd
, link
);
3943 /* get the link contents into pagecache */
3944 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3948 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3949 page
= read_mapping_page(mapping
, 0, NULL
);
3954 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3958 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3960 struct page
*page
= NULL
;
3961 char *s
= page_getlink(dentry
, &page
);
3962 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3965 page_cache_release(page
);
3970 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3972 struct page
*page
= NULL
;
3973 nd_set_link(nd
, page_getlink(dentry
, &page
));
3977 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3979 struct page
*page
= cookie
;
3983 page_cache_release(page
);
3988 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3990 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3992 struct address_space
*mapping
= inode
->i_mapping
;
3997 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3999 flags
|= AOP_FLAG_NOFS
;
4002 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4003 flags
, &page
, &fsdata
);
4007 kaddr
= kmap_atomic(page
);
4008 memcpy(kaddr
, symname
, len
-1);
4009 kunmap_atomic(kaddr
);
4011 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4018 mark_inode_dirty(inode
);
4024 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4026 return __page_symlink(inode
, symname
, len
,
4027 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4030 const struct inode_operations page_symlink_inode_operations
= {
4031 .readlink
= generic_readlink
,
4032 .follow_link
= page_follow_link_light
,
4033 .put_link
= page_put_link
,
4036 EXPORT_SYMBOL(user_path_at
);
4037 EXPORT_SYMBOL(follow_down_one
);
4038 EXPORT_SYMBOL(follow_down
);
4039 EXPORT_SYMBOL(follow_up
);
4040 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4041 EXPORT_SYMBOL(lock_rename
);
4042 EXPORT_SYMBOL(lookup_one_len
);
4043 EXPORT_SYMBOL(page_follow_link_light
);
4044 EXPORT_SYMBOL(page_put_link
);
4045 EXPORT_SYMBOL(page_readlink
);
4046 EXPORT_SYMBOL(__page_symlink
);
4047 EXPORT_SYMBOL(page_symlink
);
4048 EXPORT_SYMBOL(page_symlink_inode_operations
);
4049 EXPORT_SYMBOL(kern_path
);
4050 EXPORT_SYMBOL(vfs_path_lookup
);
4051 EXPORT_SYMBOL(inode_permission
);
4052 EXPORT_SYMBOL(unlock_rename
);
4053 EXPORT_SYMBOL(vfs_create
);
4054 EXPORT_SYMBOL(vfs_follow_link
);
4055 EXPORT_SYMBOL(vfs_link
);
4056 EXPORT_SYMBOL(vfs_mkdir
);
4057 EXPORT_SYMBOL(vfs_mknod
);
4058 EXPORT_SYMBOL(generic_permission
);
4059 EXPORT_SYMBOL(vfs_readlink
);
4060 EXPORT_SYMBOL(vfs_rename
);
4061 EXPORT_SYMBOL(vfs_rmdir
);
4062 EXPORT_SYMBOL(vfs_symlink
);
4063 EXPORT_SYMBOL(vfs_unlink
);
4064 EXPORT_SYMBOL(dentry_unhash
);
4065 EXPORT_SYMBOL(generic_readlink
);