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(const 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(const 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
;
512 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
515 * Get a reference to the parent first: we're
516 * going to make "path_put(nd->path)" valid in
517 * non-RCU context for "terminate_walk()".
519 * If this doesn't work, return immediately with
520 * RCU walking still active (and then we will do
521 * the RCU walk cleanup in terminate_walk()).
523 if (!lockref_get_not_dead(&parent
->d_lockref
))
527 * After the mntget(), we terminate_walk() will do
528 * the right thing for non-RCU mode, and all our
529 * subsequent exit cases should unlock_rcu_walk()
532 mntget(nd
->path
.mnt
);
533 nd
->flags
&= ~LOOKUP_RCU
;
536 * For a negative lookup, the lookup sequence point is the parents
537 * sequence point, and it only needs to revalidate the parent dentry.
539 * For a positive lookup, we need to move both the parent and the
540 * dentry from the RCU domain to be properly refcounted. And the
541 * sequence number in the dentry validates *both* dentry counters,
542 * since we checked the sequence number of the parent after we got
543 * the child sequence number. So we know the parent must still
544 * be valid if the child sequence number is still valid.
547 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
549 BUG_ON(nd
->inode
!= parent
->d_inode
);
551 if (!lockref_get_not_dead(&dentry
->d_lockref
))
553 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
558 * Sequence counts matched. Now make sure that the root is
559 * still valid and get it if required.
561 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
562 spin_lock(&fs
->lock
);
563 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
564 goto unlock_and_drop_dentry
;
566 spin_unlock(&fs
->lock
);
572 unlock_and_drop_dentry
:
573 spin_unlock(&fs
->lock
);
581 if (!(nd
->flags
& LOOKUP_ROOT
))
586 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
588 return dentry
->d_op
->d_revalidate(dentry
, flags
);
592 * complete_walk - successful completion of path walk
593 * @nd: pointer nameidata
595 * If we had been in RCU mode, drop out of it and legitimize nd->path.
596 * Revalidate the final result, unless we'd already done that during
597 * the path walk or the filesystem doesn't ask for it. Return 0 on
598 * success, -error on failure. In case of failure caller does not
599 * need to drop nd->path.
601 static int complete_walk(struct nameidata
*nd
)
603 struct dentry
*dentry
= nd
->path
.dentry
;
606 if (nd
->flags
& LOOKUP_RCU
) {
607 nd
->flags
&= ~LOOKUP_RCU
;
608 if (!(nd
->flags
& LOOKUP_ROOT
))
611 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
615 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
620 mntget(nd
->path
.mnt
);
624 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
627 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
630 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
641 static __always_inline
void set_root(struct nameidata
*nd
)
644 get_fs_root(current
->fs
, &nd
->root
);
647 static int link_path_walk(const char *, struct nameidata
*);
649 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
652 struct fs_struct
*fs
= current
->fs
;
656 seq
= read_seqcount_begin(&fs
->seq
);
658 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
659 } while (read_seqcount_retry(&fs
->seq
, seq
));
663 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
675 nd
->flags
|= LOOKUP_JUMPED
;
677 nd
->inode
= nd
->path
.dentry
->d_inode
;
679 ret
= link_path_walk(link
, nd
);
683 return PTR_ERR(link
);
686 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
689 if (path
->mnt
!= nd
->path
.mnt
)
693 static inline void path_to_nameidata(const struct path
*path
,
694 struct nameidata
*nd
)
696 if (!(nd
->flags
& LOOKUP_RCU
)) {
697 dput(nd
->path
.dentry
);
698 if (nd
->path
.mnt
!= path
->mnt
)
699 mntput(nd
->path
.mnt
);
701 nd
->path
.mnt
= path
->mnt
;
702 nd
->path
.dentry
= path
->dentry
;
706 * Helper to directly jump to a known parsed path from ->follow_link,
707 * caller must have taken a reference to path beforehand.
709 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
714 nd
->inode
= nd
->path
.dentry
->d_inode
;
715 nd
->flags
|= LOOKUP_JUMPED
;
718 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
720 struct inode
*inode
= link
->dentry
->d_inode
;
721 if (inode
->i_op
->put_link
)
722 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
726 int sysctl_protected_symlinks __read_mostly
= 0;
727 int sysctl_protected_hardlinks __read_mostly
= 0;
730 * may_follow_link - Check symlink following for unsafe situations
731 * @link: The path of the symlink
732 * @nd: nameidata pathwalk data
734 * In the case of the sysctl_protected_symlinks sysctl being enabled,
735 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
736 * in a sticky world-writable directory. This is to protect privileged
737 * processes from failing races against path names that may change out
738 * from under them by way of other users creating malicious symlinks.
739 * It will permit symlinks to be followed only when outside a sticky
740 * world-writable directory, or when the uid of the symlink and follower
741 * match, or when the directory owner matches the symlink's owner.
743 * Returns 0 if following the symlink is allowed, -ve on error.
745 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
747 const struct inode
*inode
;
748 const struct inode
*parent
;
750 if (!sysctl_protected_symlinks
)
753 /* Allowed if owner and follower match. */
754 inode
= link
->dentry
->d_inode
;
755 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
758 /* Allowed if parent directory not sticky and world-writable. */
759 parent
= nd
->path
.dentry
->d_inode
;
760 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
763 /* Allowed if parent directory and link owner match. */
764 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
767 audit_log_link_denied("follow_link", link
);
768 path_put_conditional(link
, nd
);
774 * safe_hardlink_source - Check for safe hardlink conditions
775 * @inode: the source inode to hardlink from
777 * Return false if at least one of the following conditions:
778 * - inode is not a regular file
780 * - inode is setgid and group-exec
781 * - access failure for read and write
783 * Otherwise returns true.
785 static bool safe_hardlink_source(struct inode
*inode
)
787 umode_t mode
= inode
->i_mode
;
789 /* Special files should not get pinned to the filesystem. */
793 /* Setuid files should not get pinned to the filesystem. */
797 /* Executable setgid files should not get pinned to the filesystem. */
798 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
801 /* Hardlinking to unreadable or unwritable sources is dangerous. */
802 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
809 * may_linkat - Check permissions for creating a hardlink
810 * @link: the source to hardlink from
812 * Block hardlink when all of:
813 * - sysctl_protected_hardlinks enabled
814 * - fsuid does not match inode
815 * - hardlink source is unsafe (see safe_hardlink_source() above)
818 * Returns 0 if successful, -ve on error.
820 static int may_linkat(struct path
*link
)
822 const struct cred
*cred
;
825 if (!sysctl_protected_hardlinks
)
828 cred
= current_cred();
829 inode
= link
->dentry
->d_inode
;
831 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
832 * otherwise, it must be a safe source.
834 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
838 audit_log_link_denied("linkat", link
);
842 static __always_inline
int
843 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
845 struct dentry
*dentry
= link
->dentry
;
849 BUG_ON(nd
->flags
& LOOKUP_RCU
);
851 if (link
->mnt
== nd
->path
.mnt
)
855 if (unlikely(current
->total_link_count
>= 40))
856 goto out_put_nd_path
;
859 current
->total_link_count
++;
862 nd_set_link(nd
, NULL
);
864 error
= security_inode_follow_link(link
->dentry
, nd
);
866 goto out_put_nd_path
;
868 nd
->last_type
= LAST_BIND
;
869 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
872 goto out_put_nd_path
;
877 error
= __vfs_follow_link(nd
, s
);
879 put_link(nd
, link
, *p
);
891 static int follow_up_rcu(struct path
*path
)
893 struct mount
*mnt
= real_mount(path
->mnt
);
894 struct mount
*parent
;
895 struct dentry
*mountpoint
;
897 parent
= mnt
->mnt_parent
;
898 if (&parent
->mnt
== path
->mnt
)
900 mountpoint
= mnt
->mnt_mountpoint
;
901 path
->dentry
= mountpoint
;
902 path
->mnt
= &parent
->mnt
;
907 * follow_up - Find the mountpoint of path's vfsmount
909 * Given a path, find the mountpoint of its source file system.
910 * Replace @path with the path of the mountpoint in the parent mount.
913 * Return 1 if we went up a level and 0 if we were already at the
916 int follow_up(struct path
*path
)
918 struct mount
*mnt
= real_mount(path
->mnt
);
919 struct mount
*parent
;
920 struct dentry
*mountpoint
;
922 br_read_lock(&vfsmount_lock
);
923 parent
= mnt
->mnt_parent
;
925 br_read_unlock(&vfsmount_lock
);
928 mntget(&parent
->mnt
);
929 mountpoint
= dget(mnt
->mnt_mountpoint
);
930 br_read_unlock(&vfsmount_lock
);
932 path
->dentry
= mountpoint
;
934 path
->mnt
= &parent
->mnt
;
939 * Perform an automount
940 * - return -EISDIR to tell follow_managed() to stop and return the path we
943 static int follow_automount(struct path
*path
, unsigned flags
,
946 struct vfsmount
*mnt
;
949 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
952 /* We don't want to mount if someone's just doing a stat -
953 * unless they're stat'ing a directory and appended a '/' to
956 * We do, however, want to mount if someone wants to open or
957 * create a file of any type under the mountpoint, wants to
958 * traverse through the mountpoint or wants to open the
959 * mounted directory. Also, autofs may mark negative dentries
960 * as being automount points. These will need the attentions
961 * of the daemon to instantiate them before they can be used.
963 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
964 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
965 path
->dentry
->d_inode
)
968 current
->total_link_count
++;
969 if (current
->total_link_count
>= 40)
972 mnt
= path
->dentry
->d_op
->d_automount(path
);
975 * The filesystem is allowed to return -EISDIR here to indicate
976 * it doesn't want to automount. For instance, autofs would do
977 * this so that its userspace daemon can mount on this dentry.
979 * However, we can only permit this if it's a terminal point in
980 * the path being looked up; if it wasn't then the remainder of
981 * the path is inaccessible and we should say so.
983 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
988 if (!mnt
) /* mount collision */
992 /* lock_mount() may release path->mnt on error */
996 err
= finish_automount(mnt
, path
);
1000 /* Someone else made a mount here whilst we were busy */
1005 path
->dentry
= dget(mnt
->mnt_root
);
1014 * Handle a dentry that is managed in some way.
1015 * - Flagged for transit management (autofs)
1016 * - Flagged as mountpoint
1017 * - Flagged as automount point
1019 * This may only be called in refwalk mode.
1021 * Serialization is taken care of in namespace.c
1023 static int follow_managed(struct path
*path
, unsigned flags
)
1025 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1027 bool need_mntput
= false;
1030 /* Given that we're not holding a lock here, we retain the value in a
1031 * local variable for each dentry as we look at it so that we don't see
1032 * the components of that value change under us */
1033 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1034 managed
&= DCACHE_MANAGED_DENTRY
,
1035 unlikely(managed
!= 0)) {
1036 /* Allow the filesystem to manage the transit without i_mutex
1038 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1039 BUG_ON(!path
->dentry
->d_op
);
1040 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1041 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1046 /* Transit to a mounted filesystem. */
1047 if (managed
& DCACHE_MOUNTED
) {
1048 struct vfsmount
*mounted
= lookup_mnt(path
);
1053 path
->mnt
= mounted
;
1054 path
->dentry
= dget(mounted
->mnt_root
);
1059 /* Something is mounted on this dentry in another
1060 * namespace and/or whatever was mounted there in this
1061 * namespace got unmounted before we managed to get the
1065 /* Handle an automount point */
1066 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1067 ret
= follow_automount(path
, flags
, &need_mntput
);
1073 /* We didn't change the current path point */
1077 if (need_mntput
&& path
->mnt
== mnt
)
1081 return ret
< 0 ? ret
: need_mntput
;
1084 int follow_down_one(struct path
*path
)
1086 struct vfsmount
*mounted
;
1088 mounted
= lookup_mnt(path
);
1092 path
->mnt
= mounted
;
1093 path
->dentry
= dget(mounted
->mnt_root
);
1099 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1101 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1102 dentry
->d_op
->d_manage(dentry
, true) < 0);
1106 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1107 * we meet a managed dentry that would need blocking.
1109 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1110 struct inode
**inode
)
1113 struct mount
*mounted
;
1115 * Don't forget we might have a non-mountpoint managed dentry
1116 * that wants to block transit.
1118 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1121 if (!d_mountpoint(path
->dentry
))
1124 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1127 path
->mnt
= &mounted
->mnt
;
1128 path
->dentry
= mounted
->mnt
.mnt_root
;
1129 nd
->flags
|= LOOKUP_JUMPED
;
1130 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1132 * Update the inode too. We don't need to re-check the
1133 * dentry sequence number here after this d_inode read,
1134 * because a mount-point is always pinned.
1136 *inode
= path
->dentry
->d_inode
;
1141 static void follow_mount_rcu(struct nameidata
*nd
)
1143 while (d_mountpoint(nd
->path
.dentry
)) {
1144 struct mount
*mounted
;
1145 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1148 nd
->path
.mnt
= &mounted
->mnt
;
1149 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1150 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1154 static int follow_dotdot_rcu(struct nameidata
*nd
)
1159 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1160 nd
->path
.mnt
== nd
->root
.mnt
) {
1163 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1164 struct dentry
*old
= nd
->path
.dentry
;
1165 struct dentry
*parent
= old
->d_parent
;
1168 seq
= read_seqcount_begin(&parent
->d_seq
);
1169 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1171 nd
->path
.dentry
= parent
;
1175 if (!follow_up_rcu(&nd
->path
))
1177 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1179 follow_mount_rcu(nd
);
1180 nd
->inode
= nd
->path
.dentry
->d_inode
;
1184 nd
->flags
&= ~LOOKUP_RCU
;
1185 if (!(nd
->flags
& LOOKUP_ROOT
))
1186 nd
->root
.mnt
= NULL
;
1192 * Follow down to the covering mount currently visible to userspace. At each
1193 * point, the filesystem owning that dentry may be queried as to whether the
1194 * caller is permitted to proceed or not.
1196 int follow_down(struct path
*path
)
1201 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1202 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1203 /* Allow the filesystem to manage the transit without i_mutex
1206 * We indicate to the filesystem if someone is trying to mount
1207 * something here. This gives autofs the chance to deny anyone
1208 * other than its daemon the right to mount on its
1211 * The filesystem may sleep at this point.
1213 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1214 BUG_ON(!path
->dentry
->d_op
);
1215 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1216 ret
= path
->dentry
->d_op
->d_manage(
1217 path
->dentry
, false);
1219 return ret
== -EISDIR
? 0 : ret
;
1222 /* Transit to a mounted filesystem. */
1223 if (managed
& DCACHE_MOUNTED
) {
1224 struct vfsmount
*mounted
= lookup_mnt(path
);
1229 path
->mnt
= mounted
;
1230 path
->dentry
= dget(mounted
->mnt_root
);
1234 /* Don't handle automount points here */
1241 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1243 static void follow_mount(struct path
*path
)
1245 while (d_mountpoint(path
->dentry
)) {
1246 struct vfsmount
*mounted
= lookup_mnt(path
);
1251 path
->mnt
= mounted
;
1252 path
->dentry
= dget(mounted
->mnt_root
);
1256 static void follow_dotdot(struct nameidata
*nd
)
1261 struct dentry
*old
= nd
->path
.dentry
;
1263 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1264 nd
->path
.mnt
== nd
->root
.mnt
) {
1267 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1268 /* rare case of legitimate dget_parent()... */
1269 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1273 if (!follow_up(&nd
->path
))
1276 follow_mount(&nd
->path
);
1277 nd
->inode
= nd
->path
.dentry
->d_inode
;
1281 * This looks up the name in dcache, possibly revalidates the old dentry and
1282 * allocates a new one if not found or not valid. In the need_lookup argument
1283 * returns whether i_op->lookup is necessary.
1285 * dir->d_inode->i_mutex must be held
1287 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1288 unsigned int flags
, bool *need_lookup
)
1290 struct dentry
*dentry
;
1293 *need_lookup
= false;
1294 dentry
= d_lookup(dir
, name
);
1296 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1297 error
= d_revalidate(dentry
, flags
);
1298 if (unlikely(error
<= 0)) {
1301 return ERR_PTR(error
);
1302 } else if (!d_invalidate(dentry
)) {
1311 dentry
= d_alloc(dir
, name
);
1312 if (unlikely(!dentry
))
1313 return ERR_PTR(-ENOMEM
);
1315 *need_lookup
= true;
1321 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1322 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1324 * dir->d_inode->i_mutex must be held
1326 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1331 /* Don't create child dentry for a dead directory. */
1332 if (unlikely(IS_DEADDIR(dir
))) {
1334 return ERR_PTR(-ENOENT
);
1337 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1338 if (unlikely(old
)) {
1345 static struct dentry
*__lookup_hash(struct qstr
*name
,
1346 struct dentry
*base
, unsigned int flags
)
1349 struct dentry
*dentry
;
1351 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1355 return lookup_real(base
->d_inode
, dentry
, flags
);
1359 * It's more convoluted than I'd like it to be, but... it's still fairly
1360 * small and for now I'd prefer to have fast path as straight as possible.
1361 * It _is_ time-critical.
1363 static int lookup_fast(struct nameidata
*nd
,
1364 struct path
*path
, struct inode
**inode
)
1366 struct vfsmount
*mnt
= nd
->path
.mnt
;
1367 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1373 * Rename seqlock is not required here because in the off chance
1374 * of a false negative due to a concurrent rename, we're going to
1375 * do the non-racy lookup, below.
1377 if (nd
->flags
& LOOKUP_RCU
) {
1379 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1384 * This sequence count validates that the inode matches
1385 * the dentry name information from lookup.
1387 *inode
= dentry
->d_inode
;
1388 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1392 * This sequence count validates that the parent had no
1393 * changes while we did the lookup of the dentry above.
1395 * The memory barrier in read_seqcount_begin of child is
1396 * enough, we can use __read_seqcount_retry here.
1398 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1402 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1403 status
= d_revalidate(dentry
, nd
->flags
);
1404 if (unlikely(status
<= 0)) {
1405 if (status
!= -ECHILD
)
1411 path
->dentry
= dentry
;
1412 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1414 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1418 if (unlazy_walk(nd
, dentry
))
1421 dentry
= __d_lookup(parent
, &nd
->last
);
1424 if (unlikely(!dentry
))
1427 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1428 status
= d_revalidate(dentry
, nd
->flags
);
1429 if (unlikely(status
<= 0)) {
1434 if (!d_invalidate(dentry
)) {
1441 path
->dentry
= dentry
;
1442 err
= follow_managed(path
, nd
->flags
);
1443 if (unlikely(err
< 0)) {
1444 path_put_conditional(path
, nd
);
1448 nd
->flags
|= LOOKUP_JUMPED
;
1449 *inode
= path
->dentry
->d_inode
;
1456 /* Fast lookup failed, do it the slow way */
1457 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1459 struct dentry
*dentry
, *parent
;
1462 parent
= nd
->path
.dentry
;
1463 BUG_ON(nd
->inode
!= parent
->d_inode
);
1465 mutex_lock(&parent
->d_inode
->i_mutex
);
1466 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1467 mutex_unlock(&parent
->d_inode
->i_mutex
);
1469 return PTR_ERR(dentry
);
1470 path
->mnt
= nd
->path
.mnt
;
1471 path
->dentry
= dentry
;
1472 err
= follow_managed(path
, nd
->flags
);
1473 if (unlikely(err
< 0)) {
1474 path_put_conditional(path
, nd
);
1478 nd
->flags
|= LOOKUP_JUMPED
;
1482 static inline int may_lookup(struct nameidata
*nd
)
1484 if (nd
->flags
& LOOKUP_RCU
) {
1485 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1488 if (unlazy_walk(nd
, NULL
))
1491 return inode_permission(nd
->inode
, MAY_EXEC
);
1494 static inline int handle_dots(struct nameidata
*nd
, int type
)
1496 if (type
== LAST_DOTDOT
) {
1497 if (nd
->flags
& LOOKUP_RCU
) {
1498 if (follow_dotdot_rcu(nd
))
1506 static void terminate_walk(struct nameidata
*nd
)
1508 if (!(nd
->flags
& LOOKUP_RCU
)) {
1509 path_put(&nd
->path
);
1511 nd
->flags
&= ~LOOKUP_RCU
;
1512 if (!(nd
->flags
& LOOKUP_ROOT
))
1513 nd
->root
.mnt
= NULL
;
1519 * Do we need to follow links? We _really_ want to be able
1520 * to do this check without having to look at inode->i_op,
1521 * so we keep a cache of "no, this doesn't need follow_link"
1522 * for the common case.
1524 static inline int should_follow_link(struct inode
*inode
, int follow
)
1526 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1527 if (likely(inode
->i_op
->follow_link
))
1530 /* This gets set once for the inode lifetime */
1531 spin_lock(&inode
->i_lock
);
1532 inode
->i_opflags
|= IOP_NOFOLLOW
;
1533 spin_unlock(&inode
->i_lock
);
1538 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1541 struct inode
*inode
;
1544 * "." and ".." are special - ".." especially so because it has
1545 * to be able to know about the current root directory and
1546 * parent relationships.
1548 if (unlikely(nd
->last_type
!= LAST_NORM
))
1549 return handle_dots(nd
, nd
->last_type
);
1550 err
= lookup_fast(nd
, path
, &inode
);
1551 if (unlikely(err
)) {
1555 err
= lookup_slow(nd
, path
);
1559 inode
= path
->dentry
->d_inode
;
1565 if (should_follow_link(inode
, follow
)) {
1566 if (nd
->flags
& LOOKUP_RCU
) {
1567 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1572 BUG_ON(inode
!= path
->dentry
->d_inode
);
1575 path_to_nameidata(path
, nd
);
1580 path_to_nameidata(path
, nd
);
1587 * This limits recursive symlink follows to 8, while
1588 * limiting consecutive symlinks to 40.
1590 * Without that kind of total limit, nasty chains of consecutive
1591 * symlinks can cause almost arbitrarily long lookups.
1593 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1597 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1598 path_put_conditional(path
, nd
);
1599 path_put(&nd
->path
);
1602 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1605 current
->link_count
++;
1608 struct path link
= *path
;
1611 res
= follow_link(&link
, nd
, &cookie
);
1614 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1615 put_link(nd
, &link
, cookie
);
1618 current
->link_count
--;
1624 * We really don't want to look at inode->i_op->lookup
1625 * when we don't have to. So we keep a cache bit in
1626 * the inode ->i_opflags field that says "yes, we can
1627 * do lookup on this inode".
1629 static inline int can_lookup(struct inode
*inode
)
1631 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1633 if (likely(!inode
->i_op
->lookup
))
1636 /* We do this once for the lifetime of the inode */
1637 spin_lock(&inode
->i_lock
);
1638 inode
->i_opflags
|= IOP_LOOKUP
;
1639 spin_unlock(&inode
->i_lock
);
1644 * We can do the critical dentry name comparison and hashing
1645 * operations one word at a time, but we are limited to:
1647 * - Architectures with fast unaligned word accesses. We could
1648 * do a "get_unaligned()" if this helps and is sufficiently
1651 * - Little-endian machines (so that we can generate the mask
1652 * of low bytes efficiently). Again, we *could* do a byte
1653 * swapping load on big-endian architectures if that is not
1654 * expensive enough to make the optimization worthless.
1656 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1657 * do not trap on the (extremely unlikely) case of a page
1658 * crossing operation.
1660 * - Furthermore, we need an efficient 64-bit compile for the
1661 * 64-bit case in order to generate the "number of bytes in
1662 * the final mask". Again, that could be replaced with a
1663 * efficient population count instruction or similar.
1665 #ifdef CONFIG_DCACHE_WORD_ACCESS
1667 #include <asm/word-at-a-time.h>
1671 static inline unsigned int fold_hash(unsigned long hash
)
1673 hash
+= hash
>> (8*sizeof(int));
1677 #else /* 32-bit case */
1679 #define fold_hash(x) (x)
1683 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1685 unsigned long a
, mask
;
1686 unsigned long hash
= 0;
1689 a
= load_unaligned_zeropad(name
);
1690 if (len
< sizeof(unsigned long))
1694 name
+= sizeof(unsigned long);
1695 len
-= sizeof(unsigned long);
1699 mask
= ~(~0ul << len
*8);
1702 return fold_hash(hash
);
1704 EXPORT_SYMBOL(full_name_hash
);
1707 * Calculate the length and hash of the path component, and
1708 * return the length of the component;
1710 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1712 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1713 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1716 len
= -sizeof(unsigned long);
1718 hash
= (hash
+ a
) * 9;
1719 len
+= sizeof(unsigned long);
1720 a
= load_unaligned_zeropad(name
+len
);
1721 b
= a
^ REPEAT_BYTE('/');
1722 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1724 adata
= prep_zero_mask(a
, adata
, &constants
);
1725 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1727 mask
= create_zero_mask(adata
| bdata
);
1729 hash
+= a
& zero_bytemask(mask
);
1730 *hashp
= fold_hash(hash
);
1732 return len
+ find_zero(mask
);
1737 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1739 unsigned long hash
= init_name_hash();
1741 hash
= partial_name_hash(*name
++, hash
);
1742 return end_name_hash(hash
);
1744 EXPORT_SYMBOL(full_name_hash
);
1747 * We know there's a real path component here of at least
1750 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1752 unsigned long hash
= init_name_hash();
1753 unsigned long len
= 0, c
;
1755 c
= (unsigned char)*name
;
1758 hash
= partial_name_hash(c
, hash
);
1759 c
= (unsigned char)name
[len
];
1760 } while (c
&& c
!= '/');
1761 *hashp
= end_name_hash(hash
);
1769 * This is the basic name resolution function, turning a pathname into
1770 * the final dentry. We expect 'base' to be positive and a directory.
1772 * Returns 0 and nd will have valid dentry and mnt on success.
1773 * Returns error and drops reference to input namei data on failure.
1775 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1785 /* At this point we know we have a real path component. */
1791 err
= may_lookup(nd
);
1795 len
= hash_name(name
, &this.hash
);
1800 if (name
[0] == '.') switch (len
) {
1802 if (name
[1] == '.') {
1804 nd
->flags
|= LOOKUP_JUMPED
;
1810 if (likely(type
== LAST_NORM
)) {
1811 struct dentry
*parent
= nd
->path
.dentry
;
1812 nd
->flags
&= ~LOOKUP_JUMPED
;
1813 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1814 err
= parent
->d_op
->d_hash(parent
, &this);
1821 nd
->last_type
= type
;
1826 * If it wasn't NUL, we know it was '/'. Skip that
1827 * slash, and continue until no more slashes.
1831 } while (unlikely(name
[len
] == '/'));
1837 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1842 err
= nested_symlink(&next
, nd
);
1846 if (!can_lookup(nd
->inode
)) {
1855 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1856 struct nameidata
*nd
, struct file
**fp
)
1860 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1861 nd
->flags
= flags
| LOOKUP_JUMPED
;
1863 if (flags
& LOOKUP_ROOT
) {
1864 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1866 if (!can_lookup(inode
))
1868 retval
= inode_permission(inode
, MAY_EXEC
);
1872 nd
->path
= nd
->root
;
1874 if (flags
& LOOKUP_RCU
) {
1876 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1878 path_get(&nd
->path
);
1883 nd
->root
.mnt
= NULL
;
1886 if (flags
& LOOKUP_RCU
) {
1891 path_get(&nd
->root
);
1893 nd
->path
= nd
->root
;
1894 } else if (dfd
== AT_FDCWD
) {
1895 if (flags
& LOOKUP_RCU
) {
1896 struct fs_struct
*fs
= current
->fs
;
1902 seq
= read_seqcount_begin(&fs
->seq
);
1904 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1905 } while (read_seqcount_retry(&fs
->seq
, seq
));
1907 get_fs_pwd(current
->fs
, &nd
->path
);
1910 /* Caller must check execute permissions on the starting path component */
1911 struct fd f
= fdget_raw(dfd
);
1912 struct dentry
*dentry
;
1917 dentry
= f
.file
->f_path
.dentry
;
1920 if (!can_lookup(dentry
->d_inode
)) {
1926 nd
->path
= f
.file
->f_path
;
1927 if (flags
& LOOKUP_RCU
) {
1930 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1933 path_get(&nd
->path
);
1938 nd
->inode
= nd
->path
.dentry
->d_inode
;
1942 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1944 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1945 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1947 nd
->flags
&= ~LOOKUP_PARENT
;
1948 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1951 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1952 static int path_lookupat(int dfd
, const char *name
,
1953 unsigned int flags
, struct nameidata
*nd
)
1955 struct file
*base
= NULL
;
1960 * Path walking is largely split up into 2 different synchronisation
1961 * schemes, rcu-walk and ref-walk (explained in
1962 * Documentation/filesystems/path-lookup.txt). These share much of the
1963 * path walk code, but some things particularly setup, cleanup, and
1964 * following mounts are sufficiently divergent that functions are
1965 * duplicated. Typically there is a function foo(), and its RCU
1966 * analogue, foo_rcu().
1968 * -ECHILD is the error number of choice (just to avoid clashes) that
1969 * is returned if some aspect of an rcu-walk fails. Such an error must
1970 * be handled by restarting a traditional ref-walk (which will always
1971 * be able to complete).
1973 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1978 current
->total_link_count
= 0;
1979 err
= link_path_walk(name
, nd
);
1981 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1982 err
= lookup_last(nd
, &path
);
1985 struct path link
= path
;
1986 err
= may_follow_link(&link
, nd
);
1989 nd
->flags
|= LOOKUP_PARENT
;
1990 err
= follow_link(&link
, nd
, &cookie
);
1993 err
= lookup_last(nd
, &path
);
1994 put_link(nd
, &link
, cookie
);
1999 err
= complete_walk(nd
);
2001 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2002 if (!can_lookup(nd
->inode
)) {
2003 path_put(&nd
->path
);
2011 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2012 path_put(&nd
->root
);
2013 nd
->root
.mnt
= NULL
;
2018 static int filename_lookup(int dfd
, struct filename
*name
,
2019 unsigned int flags
, struct nameidata
*nd
)
2021 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2022 if (unlikely(retval
== -ECHILD
))
2023 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2024 if (unlikely(retval
== -ESTALE
))
2025 retval
= path_lookupat(dfd
, name
->name
,
2026 flags
| LOOKUP_REVAL
, nd
);
2028 if (likely(!retval
))
2029 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2033 static int do_path_lookup(int dfd
, const char *name
,
2034 unsigned int flags
, struct nameidata
*nd
)
2036 struct filename filename
= { .name
= name
};
2038 return filename_lookup(dfd
, &filename
, flags
, nd
);
2041 /* does lookup, returns the object with parent locked */
2042 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2044 struct nameidata nd
;
2046 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2048 return ERR_PTR(err
);
2049 if (nd
.last_type
!= LAST_NORM
) {
2051 return ERR_PTR(-EINVAL
);
2053 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2054 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2056 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2064 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2066 struct nameidata nd
;
2067 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2074 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2075 * @dentry: pointer to dentry of the base directory
2076 * @mnt: pointer to vfs mount of the base directory
2077 * @name: pointer to file name
2078 * @flags: lookup flags
2079 * @path: pointer to struct path to fill
2081 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2082 const char *name
, unsigned int flags
,
2085 struct nameidata nd
;
2087 nd
.root
.dentry
= dentry
;
2089 BUG_ON(flags
& LOOKUP_PARENT
);
2090 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2091 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2098 * Restricted form of lookup. Doesn't follow links, single-component only,
2099 * needs parent already locked. Doesn't follow mounts.
2102 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2104 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2108 * lookup_one_len - filesystem helper to lookup single pathname component
2109 * @name: pathname component to lookup
2110 * @base: base directory to lookup from
2111 * @len: maximum length @len should be interpreted to
2113 * Note that this routine is purely a helper for filesystem usage and should
2114 * not be called by generic code. Also note that by using this function the
2115 * nameidata argument is passed to the filesystem methods and a filesystem
2116 * using this helper needs to be prepared for that.
2118 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2124 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2128 this.hash
= full_name_hash(name
, len
);
2130 return ERR_PTR(-EACCES
);
2132 if (unlikely(name
[0] == '.')) {
2133 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2134 return ERR_PTR(-EACCES
);
2138 c
= *(const unsigned char *)name
++;
2139 if (c
== '/' || c
== '\0')
2140 return ERR_PTR(-EACCES
);
2143 * See if the low-level filesystem might want
2144 * to use its own hash..
2146 if (base
->d_flags
& DCACHE_OP_HASH
) {
2147 int err
= base
->d_op
->d_hash(base
, &this);
2149 return ERR_PTR(err
);
2152 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2154 return ERR_PTR(err
);
2156 return __lookup_hash(&this, base
, 0);
2159 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2160 struct path
*path
, int *empty
)
2162 struct nameidata nd
;
2163 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2164 int err
= PTR_ERR(tmp
);
2167 BUG_ON(flags
& LOOKUP_PARENT
);
2169 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2177 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2180 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2184 * NB: most callers don't do anything directly with the reference to the
2185 * to struct filename, but the nd->last pointer points into the name string
2186 * allocated by getname. So we must hold the reference to it until all
2187 * path-walking is complete.
2189 static struct filename
*
2190 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2193 struct filename
*s
= getname(path
);
2196 /* only LOOKUP_REVAL is allowed in extra flags */
2197 flags
&= LOOKUP_REVAL
;
2202 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2205 return ERR_PTR(error
);
2212 * mountpoint_last - look up last component for umount
2213 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2214 * @path: pointer to container for result
2216 * This is a special lookup_last function just for umount. In this case, we
2217 * need to resolve the path without doing any revalidation.
2219 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2220 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2221 * in almost all cases, this lookup will be served out of the dcache. The only
2222 * cases where it won't are if nd->last refers to a symlink or the path is
2223 * bogus and it doesn't exist.
2226 * -error: if there was an error during lookup. This includes -ENOENT if the
2227 * lookup found a negative dentry. The nd->path reference will also be
2230 * 0: if we successfully resolved nd->path and found it to not to be a
2231 * symlink that needs to be followed. "path" will also be populated.
2232 * The nd->path reference will also be put.
2234 * 1: if we successfully resolved nd->last and found it to be a symlink
2235 * that needs to be followed. "path" will be populated with the path
2236 * to the link, and nd->path will *not* be put.
2239 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2242 struct dentry
*dentry
;
2243 struct dentry
*dir
= nd
->path
.dentry
;
2245 /* If we're in rcuwalk, drop out of it to handle last component */
2246 if (nd
->flags
& LOOKUP_RCU
) {
2247 if (unlazy_walk(nd
, NULL
)) {
2253 nd
->flags
&= ~LOOKUP_PARENT
;
2255 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2256 error
= handle_dots(nd
, nd
->last_type
);
2259 dentry
= dget(nd
->path
.dentry
);
2263 mutex_lock(&dir
->d_inode
->i_mutex
);
2264 dentry
= d_lookup(dir
, &nd
->last
);
2267 * No cached dentry. Mounted dentries are pinned in the cache,
2268 * so that means that this dentry is probably a symlink or the
2269 * path doesn't actually point to a mounted dentry.
2271 dentry
= d_alloc(dir
, &nd
->last
);
2276 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2277 error
= PTR_ERR(dentry
);
2281 mutex_unlock(&dir
->d_inode
->i_mutex
);
2284 if (!dentry
->d_inode
) {
2289 path
->dentry
= dentry
;
2290 path
->mnt
= mntget(nd
->path
.mnt
);
2291 if (should_follow_link(dentry
->d_inode
, nd
->flags
& LOOKUP_FOLLOW
))
2301 * path_mountpoint - look up a path to be umounted
2302 * @dfd: directory file descriptor to start walk from
2303 * @name: full pathname to walk
2304 * @flags: lookup flags
2306 * Look up the given name, but don't attempt to revalidate the last component.
2307 * Returns 0 and "path" will be valid on success; Retuns error otherwise.
2310 path_mountpoint(int dfd
, const char *name
, struct path
*path
, unsigned int flags
)
2312 struct file
*base
= NULL
;
2313 struct nameidata nd
;
2316 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2320 current
->total_link_count
= 0;
2321 err
= link_path_walk(name
, &nd
);
2325 err
= mountpoint_last(&nd
, path
);
2328 struct path link
= *path
;
2329 err
= may_follow_link(&link
, &nd
);
2332 nd
.flags
|= LOOKUP_PARENT
;
2333 err
= follow_link(&link
, &nd
, &cookie
);
2336 err
= mountpoint_last(&nd
, path
);
2337 put_link(&nd
, &link
, cookie
);
2343 if (nd
.root
.mnt
&& !(nd
.flags
& LOOKUP_ROOT
))
2350 filename_mountpoint(int dfd
, struct filename
*s
, struct path
*path
,
2353 int error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_RCU
);
2354 if (unlikely(error
== -ECHILD
))
2355 error
= path_mountpoint(dfd
, s
->name
, path
, flags
);
2356 if (unlikely(error
== -ESTALE
))
2357 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_REVAL
);
2359 audit_inode(s
, path
->dentry
, 0);
2364 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2365 * @dfd: directory file descriptor
2366 * @name: pathname from userland
2367 * @flags: lookup flags
2368 * @path: pointer to container to hold result
2370 * A umount is a special case for path walking. We're not actually interested
2371 * in the inode in this situation, and ESTALE errors can be a problem. We
2372 * simply want track down the dentry and vfsmount attached at the mountpoint
2373 * and avoid revalidating the last component.
2375 * Returns 0 and populates "path" on success.
2378 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2381 struct filename
*s
= getname(name
);
2385 error
= filename_mountpoint(dfd
, s
, path
, flags
);
2391 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2394 struct filename s
= {.name
= name
};
2395 return filename_mountpoint(dfd
, &s
, path
, flags
);
2397 EXPORT_SYMBOL(kern_path_mountpoint
);
2400 * It's inline, so penalty for filesystems that don't use sticky bit is
2403 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2405 kuid_t fsuid
= current_fsuid();
2407 if (!(dir
->i_mode
& S_ISVTX
))
2409 if (uid_eq(inode
->i_uid
, fsuid
))
2411 if (uid_eq(dir
->i_uid
, fsuid
))
2413 return !inode_capable(inode
, CAP_FOWNER
);
2417 * Check whether we can remove a link victim from directory dir, check
2418 * whether the type of victim is right.
2419 * 1. We can't do it if dir is read-only (done in permission())
2420 * 2. We should have write and exec permissions on dir
2421 * 3. We can't remove anything from append-only dir
2422 * 4. We can't do anything with immutable dir (done in permission())
2423 * 5. If the sticky bit on dir is set we should either
2424 * a. be owner of dir, or
2425 * b. be owner of victim, or
2426 * c. have CAP_FOWNER capability
2427 * 6. If the victim is append-only or immutable we can't do antyhing with
2428 * links pointing to it.
2429 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2430 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2431 * 9. We can't remove a root or mountpoint.
2432 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2433 * nfs_async_unlink().
2435 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2439 if (!victim
->d_inode
)
2442 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2443 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2445 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2450 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2451 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2454 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2456 if (IS_ROOT(victim
))
2458 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2460 if (IS_DEADDIR(dir
))
2462 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2467 /* Check whether we can create an object with dentry child in directory
2469 * 1. We can't do it if child already exists (open has special treatment for
2470 * this case, but since we are inlined it's OK)
2471 * 2. We can't do it if dir is read-only (done in permission())
2472 * 3. We should have write and exec permissions on dir
2473 * 4. We can't do it if dir is immutable (done in permission())
2475 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2479 if (IS_DEADDIR(dir
))
2481 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2485 * p1 and p2 should be directories on the same fs.
2487 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2492 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2496 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2498 p
= d_ancestor(p2
, p1
);
2500 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2501 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2505 p
= d_ancestor(p1
, p2
);
2507 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2508 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2512 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2513 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2517 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2519 mutex_unlock(&p1
->d_inode
->i_mutex
);
2521 mutex_unlock(&p2
->d_inode
->i_mutex
);
2522 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2526 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2529 int error
= may_create(dir
, dentry
);
2533 if (!dir
->i_op
->create
)
2534 return -EACCES
; /* shouldn't it be ENOSYS? */
2537 error
= security_inode_create(dir
, dentry
, mode
);
2540 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2542 fsnotify_create(dir
, dentry
);
2546 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2548 struct dentry
*dentry
= path
->dentry
;
2549 struct inode
*inode
= dentry
->d_inode
;
2559 switch (inode
->i_mode
& S_IFMT
) {
2563 if (acc_mode
& MAY_WRITE
)
2568 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2577 error
= inode_permission(inode
, acc_mode
);
2582 * An append-only file must be opened in append mode for writing.
2584 if (IS_APPEND(inode
)) {
2585 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2591 /* O_NOATIME can only be set by the owner or superuser */
2592 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2598 static int handle_truncate(struct file
*filp
)
2600 struct path
*path
= &filp
->f_path
;
2601 struct inode
*inode
= path
->dentry
->d_inode
;
2602 int error
= get_write_access(inode
);
2606 * Refuse to truncate files with mandatory locks held on them.
2608 error
= locks_verify_locked(inode
);
2610 error
= security_path_truncate(path
);
2612 error
= do_truncate(path
->dentry
, 0,
2613 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2616 put_write_access(inode
);
2620 static inline int open_to_namei_flags(int flag
)
2622 if ((flag
& O_ACCMODE
) == 3)
2627 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2629 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2633 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2637 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2641 * Attempt to atomically look up, create and open a file from a negative
2644 * Returns 0 if successful. The file will have been created and attached to
2645 * @file by the filesystem calling finish_open().
2647 * Returns 1 if the file was looked up only or didn't need creating. The
2648 * caller will need to perform the open themselves. @path will have been
2649 * updated to point to the new dentry. This may be negative.
2651 * Returns an error code otherwise.
2653 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2654 struct path
*path
, struct file
*file
,
2655 const struct open_flags
*op
,
2656 bool got_write
, bool need_lookup
,
2659 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2660 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2664 int create_error
= 0;
2665 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2667 BUG_ON(dentry
->d_inode
);
2669 /* Don't create child dentry for a dead directory. */
2670 if (unlikely(IS_DEADDIR(dir
))) {
2676 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2677 mode
&= ~current_umask();
2679 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2680 open_flag
&= ~O_TRUNC
;
2681 *opened
|= FILE_CREATED
;
2685 * Checking write permission is tricky, bacuse we don't know if we are
2686 * going to actually need it: O_CREAT opens should work as long as the
2687 * file exists. But checking existence breaks atomicity. The trick is
2688 * to check access and if not granted clear O_CREAT from the flags.
2690 * Another problem is returing the "right" error value (e.g. for an
2691 * O_EXCL open we want to return EEXIST not EROFS).
2693 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2694 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2695 if (!(open_flag
& O_CREAT
)) {
2697 * No O_CREATE -> atomicity not a requirement -> fall
2698 * back to lookup + open
2701 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2702 /* Fall back and fail with the right error */
2703 create_error
= -EROFS
;
2706 /* No side effects, safe to clear O_CREAT */
2707 create_error
= -EROFS
;
2708 open_flag
&= ~O_CREAT
;
2712 if (open_flag
& O_CREAT
) {
2713 error
= may_o_create(&nd
->path
, dentry
, mode
);
2715 create_error
= error
;
2716 if (open_flag
& O_EXCL
)
2718 open_flag
&= ~O_CREAT
;
2722 if (nd
->flags
& LOOKUP_DIRECTORY
)
2723 open_flag
|= O_DIRECTORY
;
2725 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2726 file
->f_path
.mnt
= nd
->path
.mnt
;
2727 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2730 if (create_error
&& error
== -ENOENT
)
2731 error
= create_error
;
2735 acc_mode
= op
->acc_mode
;
2736 if (*opened
& FILE_CREATED
) {
2737 fsnotify_create(dir
, dentry
);
2738 acc_mode
= MAY_OPEN
;
2741 if (error
) { /* returned 1, that is */
2742 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2746 if (file
->f_path
.dentry
) {
2748 dentry
= file
->f_path
.dentry
;
2750 if (create_error
&& dentry
->d_inode
== NULL
) {
2751 error
= create_error
;
2758 * We didn't have the inode before the open, so check open permission
2761 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2771 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2773 return PTR_ERR(dentry
);
2776 int open_flag
= op
->open_flag
;
2778 error
= create_error
;
2779 if ((open_flag
& O_EXCL
)) {
2780 if (!dentry
->d_inode
)
2782 } else if (!dentry
->d_inode
) {
2784 } else if ((open_flag
& O_TRUNC
) &&
2785 S_ISREG(dentry
->d_inode
->i_mode
)) {
2788 /* will fail later, go on to get the right error */
2792 path
->dentry
= dentry
;
2793 path
->mnt
= nd
->path
.mnt
;
2798 * Look up and maybe create and open the last component.
2800 * Must be called with i_mutex held on parent.
2802 * Returns 0 if the file was successfully atomically created (if necessary) and
2803 * opened. In this case the file will be returned attached to @file.
2805 * Returns 1 if the file was not completely opened at this time, though lookups
2806 * and creations will have been performed and the dentry returned in @path will
2807 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2808 * specified then a negative dentry may be returned.
2810 * An error code is returned otherwise.
2812 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2813 * cleared otherwise prior to returning.
2815 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2817 const struct open_flags
*op
,
2818 bool got_write
, int *opened
)
2820 struct dentry
*dir
= nd
->path
.dentry
;
2821 struct inode
*dir_inode
= dir
->d_inode
;
2822 struct dentry
*dentry
;
2826 *opened
&= ~FILE_CREATED
;
2827 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2829 return PTR_ERR(dentry
);
2831 /* Cached positive dentry: will open in f_op->open */
2832 if (!need_lookup
&& dentry
->d_inode
)
2835 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2836 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2837 need_lookup
, opened
);
2841 BUG_ON(dentry
->d_inode
);
2843 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2845 return PTR_ERR(dentry
);
2848 /* Negative dentry, just create the file */
2849 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2850 umode_t mode
= op
->mode
;
2851 if (!IS_POSIXACL(dir
->d_inode
))
2852 mode
&= ~current_umask();
2854 * This write is needed to ensure that a
2855 * rw->ro transition does not occur between
2856 * the time when the file is created and when
2857 * a permanent write count is taken through
2858 * the 'struct file' in finish_open().
2864 *opened
|= FILE_CREATED
;
2865 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2868 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2869 nd
->flags
& LOOKUP_EXCL
);
2874 path
->dentry
= dentry
;
2875 path
->mnt
= nd
->path
.mnt
;
2884 * Handle the last step of open()
2886 static int do_last(struct nameidata
*nd
, struct path
*path
,
2887 struct file
*file
, const struct open_flags
*op
,
2888 int *opened
, struct filename
*name
)
2890 struct dentry
*dir
= nd
->path
.dentry
;
2891 int open_flag
= op
->open_flag
;
2892 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2893 bool got_write
= false;
2894 int acc_mode
= op
->acc_mode
;
2895 struct inode
*inode
;
2896 bool symlink_ok
= false;
2897 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2898 bool retried
= false;
2901 nd
->flags
&= ~LOOKUP_PARENT
;
2902 nd
->flags
|= op
->intent
;
2904 if (nd
->last_type
!= LAST_NORM
) {
2905 error
= handle_dots(nd
, nd
->last_type
);
2911 if (!(open_flag
& O_CREAT
)) {
2912 if (nd
->last
.name
[nd
->last
.len
])
2913 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2914 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2916 /* we _can_ be in RCU mode here */
2917 error
= lookup_fast(nd
, path
, &inode
);
2924 BUG_ON(nd
->inode
!= dir
->d_inode
);
2926 /* create side of things */
2928 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2929 * has been cleared when we got to the last component we are
2932 error
= complete_walk(nd
);
2936 audit_inode(name
, dir
, LOOKUP_PARENT
);
2938 /* trailing slashes? */
2939 if (nd
->last
.name
[nd
->last
.len
])
2944 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2945 error
= mnt_want_write(nd
->path
.mnt
);
2949 * do _not_ fail yet - we might not need that or fail with
2950 * a different error; let lookup_open() decide; we'll be
2951 * dropping this one anyway.
2954 mutex_lock(&dir
->d_inode
->i_mutex
);
2955 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2956 mutex_unlock(&dir
->d_inode
->i_mutex
);
2962 if ((*opened
& FILE_CREATED
) ||
2963 !S_ISREG(file_inode(file
)->i_mode
))
2964 will_truncate
= false;
2966 audit_inode(name
, file
->f_path
.dentry
, 0);
2970 if (*opened
& FILE_CREATED
) {
2971 /* Don't check for write permission, don't truncate */
2972 open_flag
&= ~O_TRUNC
;
2973 will_truncate
= false;
2974 acc_mode
= MAY_OPEN
;
2975 path_to_nameidata(path
, nd
);
2976 goto finish_open_created
;
2980 * create/update audit record if it already exists.
2982 if (path
->dentry
->d_inode
)
2983 audit_inode(name
, path
->dentry
, 0);
2986 * If atomic_open() acquired write access it is dropped now due to
2987 * possible mount and symlink following (this might be optimized away if
2991 mnt_drop_write(nd
->path
.mnt
);
2996 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2999 error
= follow_managed(path
, nd
->flags
);
3004 nd
->flags
|= LOOKUP_JUMPED
;
3006 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3007 inode
= path
->dentry
->d_inode
;
3009 /* we _can_ be in RCU mode here */
3012 path_to_nameidata(path
, nd
);
3016 if (should_follow_link(inode
, !symlink_ok
)) {
3017 if (nd
->flags
& LOOKUP_RCU
) {
3018 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
3023 BUG_ON(inode
!= path
->dentry
->d_inode
);
3027 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3028 path_to_nameidata(path
, nd
);
3030 save_parent
.dentry
= nd
->path
.dentry
;
3031 save_parent
.mnt
= mntget(path
->mnt
);
3032 nd
->path
.dentry
= path
->dentry
;
3036 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3038 error
= complete_walk(nd
);
3040 path_put(&save_parent
);
3043 audit_inode(name
, nd
->path
.dentry
, 0);
3045 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
3048 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !can_lookup(nd
->inode
))
3050 if (!S_ISREG(nd
->inode
->i_mode
))
3051 will_truncate
= false;
3053 if (will_truncate
) {
3054 error
= mnt_want_write(nd
->path
.mnt
);
3059 finish_open_created
:
3060 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3063 file
->f_path
.mnt
= nd
->path
.mnt
;
3064 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3066 if (error
== -EOPENSTALE
)
3071 error
= open_check_o_direct(file
);
3074 error
= ima_file_check(file
, op
->acc_mode
);
3078 if (will_truncate
) {
3079 error
= handle_truncate(file
);
3085 mnt_drop_write(nd
->path
.mnt
);
3086 path_put(&save_parent
);
3091 path_put_conditional(path
, nd
);
3098 /* If no saved parent or already retried then can't retry */
3099 if (!save_parent
.dentry
|| retried
)
3102 BUG_ON(save_parent
.dentry
!= dir
);
3103 path_put(&nd
->path
);
3104 nd
->path
= save_parent
;
3105 nd
->inode
= dir
->d_inode
;
3106 save_parent
.mnt
= NULL
;
3107 save_parent
.dentry
= NULL
;
3109 mnt_drop_write(nd
->path
.mnt
);
3116 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3117 struct nameidata
*nd
, int flags
,
3118 const struct open_flags
*op
,
3119 struct file
*file
, int *opened
)
3121 static const struct qstr name
= QSTR_INIT("/", 1);
3122 struct dentry
*dentry
, *child
;
3124 int error
= path_lookupat(dfd
, pathname
->name
,
3125 flags
| LOOKUP_DIRECTORY
, nd
);
3126 if (unlikely(error
))
3128 error
= mnt_want_write(nd
->path
.mnt
);
3129 if (unlikely(error
))
3131 /* we want directory to be writable */
3132 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3135 dentry
= nd
->path
.dentry
;
3136 dir
= dentry
->d_inode
;
3137 if (!dir
->i_op
->tmpfile
) {
3138 error
= -EOPNOTSUPP
;
3141 child
= d_alloc(dentry
, &name
);
3142 if (unlikely(!child
)) {
3146 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3147 nd
->flags
|= op
->intent
;
3148 dput(nd
->path
.dentry
);
3149 nd
->path
.dentry
= child
;
3150 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3153 audit_inode(pathname
, nd
->path
.dentry
, 0);
3154 error
= may_open(&nd
->path
, op
->acc_mode
, op
->open_flag
);
3157 file
->f_path
.mnt
= nd
->path
.mnt
;
3158 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3161 error
= open_check_o_direct(file
);
3164 } else if (!(op
->open_flag
& O_EXCL
)) {
3165 struct inode
*inode
= file_inode(file
);
3166 spin_lock(&inode
->i_lock
);
3167 inode
->i_state
|= I_LINKABLE
;
3168 spin_unlock(&inode
->i_lock
);
3171 mnt_drop_write(nd
->path
.mnt
);
3173 path_put(&nd
->path
);
3177 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3178 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3180 struct file
*base
= NULL
;
3186 file
= get_empty_filp();
3190 file
->f_flags
= op
->open_flag
;
3192 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3193 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3197 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
3198 if (unlikely(error
))
3201 current
->total_link_count
= 0;
3202 error
= link_path_walk(pathname
->name
, nd
);
3203 if (unlikely(error
))
3206 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3207 while (unlikely(error
> 0)) { /* trailing symlink */
3208 struct path link
= path
;
3210 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3211 path_put_conditional(&path
, nd
);
3212 path_put(&nd
->path
);
3216 error
= may_follow_link(&link
, nd
);
3217 if (unlikely(error
))
3219 nd
->flags
|= LOOKUP_PARENT
;
3220 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3221 error
= follow_link(&link
, nd
, &cookie
);
3222 if (unlikely(error
))
3224 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3225 put_link(nd
, &link
, cookie
);
3228 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
3229 path_put(&nd
->root
);
3232 if (!(opened
& FILE_OPENED
)) {
3236 if (unlikely(error
)) {
3237 if (error
== -EOPENSTALE
) {
3238 if (flags
& LOOKUP_RCU
)
3243 file
= ERR_PTR(error
);
3248 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3249 const struct open_flags
*op
)
3251 struct nameidata nd
;
3252 int flags
= op
->lookup_flags
;
3255 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3256 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3257 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3258 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3259 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3263 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3264 const char *name
, const struct open_flags
*op
)
3266 struct nameidata nd
;
3268 struct filename filename
= { .name
= name
};
3269 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3272 nd
.root
.dentry
= dentry
;
3274 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3275 return ERR_PTR(-ELOOP
);
3277 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3278 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3279 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3280 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3281 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3285 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3286 struct path
*path
, unsigned int lookup_flags
)
3288 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3289 struct nameidata nd
;
3292 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3295 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3296 * other flags passed in are ignored!
3298 lookup_flags
&= LOOKUP_REVAL
;
3300 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3302 return ERR_PTR(error
);
3305 * Yucky last component or no last component at all?
3306 * (foo/., foo/.., /////)
3308 if (nd
.last_type
!= LAST_NORM
)
3310 nd
.flags
&= ~LOOKUP_PARENT
;
3311 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3313 /* don't fail immediately if it's r/o, at least try to report other errors */
3314 err2
= mnt_want_write(nd
.path
.mnt
);
3316 * Do the final lookup.
3318 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3319 dentry
= lookup_hash(&nd
);
3324 if (dentry
->d_inode
)
3327 * Special case - lookup gave negative, but... we had foo/bar/
3328 * From the vfs_mknod() POV we just have a negative dentry -
3329 * all is fine. Let's be bastards - you had / on the end, you've
3330 * been asking for (non-existent) directory. -ENOENT for you.
3332 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3336 if (unlikely(err2
)) {
3344 dentry
= ERR_PTR(error
);
3346 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3348 mnt_drop_write(nd
.path
.mnt
);
3353 EXPORT_SYMBOL(kern_path_create
);
3355 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3358 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3359 mnt_drop_write(path
->mnt
);
3362 EXPORT_SYMBOL(done_path_create
);
3364 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3365 struct path
*path
, unsigned int lookup_flags
)
3367 struct filename
*tmp
= getname(pathname
);
3370 return ERR_CAST(tmp
);
3371 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3375 EXPORT_SYMBOL(user_path_create
);
3377 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3379 int error
= may_create(dir
, dentry
);
3384 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3387 if (!dir
->i_op
->mknod
)
3390 error
= devcgroup_inode_mknod(mode
, dev
);
3394 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3398 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3400 fsnotify_create(dir
, dentry
);
3404 static int may_mknod(umode_t mode
)
3406 switch (mode
& S_IFMT
) {
3412 case 0: /* zero mode translates to S_IFREG */
3421 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3424 struct dentry
*dentry
;
3427 unsigned int lookup_flags
= 0;
3429 error
= may_mknod(mode
);
3433 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3435 return PTR_ERR(dentry
);
3437 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3438 mode
&= ~current_umask();
3439 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3442 switch (mode
& S_IFMT
) {
3443 case 0: case S_IFREG
:
3444 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3446 case S_IFCHR
: case S_IFBLK
:
3447 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3448 new_decode_dev(dev
));
3450 case S_IFIFO
: case S_IFSOCK
:
3451 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3455 done_path_create(&path
, dentry
);
3456 if (retry_estale(error
, lookup_flags
)) {
3457 lookup_flags
|= LOOKUP_REVAL
;
3463 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3465 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3468 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3470 int error
= may_create(dir
, dentry
);
3471 unsigned max_links
= dir
->i_sb
->s_max_links
;
3476 if (!dir
->i_op
->mkdir
)
3479 mode
&= (S_IRWXUGO
|S_ISVTX
);
3480 error
= security_inode_mkdir(dir
, dentry
, mode
);
3484 if (max_links
&& dir
->i_nlink
>= max_links
)
3487 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3489 fsnotify_mkdir(dir
, dentry
);
3493 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3495 struct dentry
*dentry
;
3498 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3501 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3503 return PTR_ERR(dentry
);
3505 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3506 mode
&= ~current_umask();
3507 error
= security_path_mkdir(&path
, dentry
, mode
);
3509 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3510 done_path_create(&path
, dentry
);
3511 if (retry_estale(error
, lookup_flags
)) {
3512 lookup_flags
|= LOOKUP_REVAL
;
3518 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3520 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3524 * The dentry_unhash() helper will try to drop the dentry early: we
3525 * should have a usage count of 1 if we're the only user of this
3526 * dentry, and if that is true (possibly after pruning the dcache),
3527 * then we drop the dentry now.
3529 * A low-level filesystem can, if it choses, legally
3532 * if (!d_unhashed(dentry))
3535 * if it cannot handle the case of removing a directory
3536 * that is still in use by something else..
3538 void dentry_unhash(struct dentry
*dentry
)
3540 shrink_dcache_parent(dentry
);
3541 spin_lock(&dentry
->d_lock
);
3542 if (dentry
->d_lockref
.count
== 1)
3544 spin_unlock(&dentry
->d_lock
);
3547 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3549 int error
= may_delete(dir
, dentry
, 1);
3554 if (!dir
->i_op
->rmdir
)
3558 mutex_lock(&dentry
->d_inode
->i_mutex
);
3561 if (d_mountpoint(dentry
))
3564 error
= security_inode_rmdir(dir
, dentry
);
3568 shrink_dcache_parent(dentry
);
3569 error
= dir
->i_op
->rmdir(dir
, dentry
);
3573 dentry
->d_inode
->i_flags
|= S_DEAD
;
3577 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3584 static long do_rmdir(int dfd
, const char __user
*pathname
)
3587 struct filename
*name
;
3588 struct dentry
*dentry
;
3589 struct nameidata nd
;
3590 unsigned int lookup_flags
= 0;
3592 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3594 return PTR_ERR(name
);
3596 switch(nd
.last_type
) {
3608 nd
.flags
&= ~LOOKUP_PARENT
;
3609 error
= mnt_want_write(nd
.path
.mnt
);
3613 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3614 dentry
= lookup_hash(&nd
);
3615 error
= PTR_ERR(dentry
);
3618 if (!dentry
->d_inode
) {
3622 error
= security_path_rmdir(&nd
.path
, dentry
);
3625 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3629 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3630 mnt_drop_write(nd
.path
.mnt
);
3634 if (retry_estale(error
, lookup_flags
)) {
3635 lookup_flags
|= LOOKUP_REVAL
;
3641 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3643 return do_rmdir(AT_FDCWD
, pathname
);
3646 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3648 int error
= may_delete(dir
, dentry
, 0);
3653 if (!dir
->i_op
->unlink
)
3656 mutex_lock(&dentry
->d_inode
->i_mutex
);
3657 if (d_mountpoint(dentry
))
3660 error
= security_inode_unlink(dir
, dentry
);
3662 error
= dir
->i_op
->unlink(dir
, dentry
);
3667 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3669 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3670 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3671 fsnotify_link_count(dentry
->d_inode
);
3679 * Make sure that the actual truncation of the file will occur outside its
3680 * directory's i_mutex. Truncate can take a long time if there is a lot of
3681 * writeout happening, and we don't want to prevent access to the directory
3682 * while waiting on the I/O.
3684 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3687 struct filename
*name
;
3688 struct dentry
*dentry
;
3689 struct nameidata nd
;
3690 struct inode
*inode
= NULL
;
3691 unsigned int lookup_flags
= 0;
3693 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3695 return PTR_ERR(name
);
3698 if (nd
.last_type
!= LAST_NORM
)
3701 nd
.flags
&= ~LOOKUP_PARENT
;
3702 error
= mnt_want_write(nd
.path
.mnt
);
3706 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3707 dentry
= lookup_hash(&nd
);
3708 error
= PTR_ERR(dentry
);
3709 if (!IS_ERR(dentry
)) {
3710 /* Why not before? Because we want correct error value */
3711 if (nd
.last
.name
[nd
.last
.len
])
3713 inode
= dentry
->d_inode
;
3717 error
= security_path_unlink(&nd
.path
, dentry
);
3720 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3724 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3726 iput(inode
); /* truncate the inode here */
3727 mnt_drop_write(nd
.path
.mnt
);
3731 if (retry_estale(error
, lookup_flags
)) {
3732 lookup_flags
|= LOOKUP_REVAL
;
3739 error
= !dentry
->d_inode
? -ENOENT
:
3740 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3744 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3746 if ((flag
& ~AT_REMOVEDIR
) != 0)
3749 if (flag
& AT_REMOVEDIR
)
3750 return do_rmdir(dfd
, pathname
);
3752 return do_unlinkat(dfd
, pathname
);
3755 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3757 return do_unlinkat(AT_FDCWD
, pathname
);
3760 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3762 int error
= may_create(dir
, dentry
);
3767 if (!dir
->i_op
->symlink
)
3770 error
= security_inode_symlink(dir
, dentry
, oldname
);
3774 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3776 fsnotify_create(dir
, dentry
);
3780 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3781 int, newdfd
, const char __user
*, newname
)
3784 struct filename
*from
;
3785 struct dentry
*dentry
;
3787 unsigned int lookup_flags
= 0;
3789 from
= getname(oldname
);
3791 return PTR_ERR(from
);
3793 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3794 error
= PTR_ERR(dentry
);
3798 error
= security_path_symlink(&path
, dentry
, from
->name
);
3800 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3801 done_path_create(&path
, dentry
);
3802 if (retry_estale(error
, lookup_flags
)) {
3803 lookup_flags
|= LOOKUP_REVAL
;
3811 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3813 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3816 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3818 struct inode
*inode
= old_dentry
->d_inode
;
3819 unsigned max_links
= dir
->i_sb
->s_max_links
;
3825 error
= may_create(dir
, new_dentry
);
3829 if (dir
->i_sb
!= inode
->i_sb
)
3833 * A link to an append-only or immutable file cannot be created.
3835 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3837 if (!dir
->i_op
->link
)
3839 if (S_ISDIR(inode
->i_mode
))
3842 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3846 mutex_lock(&inode
->i_mutex
);
3847 /* Make sure we don't allow creating hardlink to an unlinked file */
3848 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3850 else if (max_links
&& inode
->i_nlink
>= max_links
)
3853 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3855 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3856 spin_lock(&inode
->i_lock
);
3857 inode
->i_state
&= ~I_LINKABLE
;
3858 spin_unlock(&inode
->i_lock
);
3860 mutex_unlock(&inode
->i_mutex
);
3862 fsnotify_link(dir
, inode
, new_dentry
);
3867 * Hardlinks are often used in delicate situations. We avoid
3868 * security-related surprises by not following symlinks on the
3871 * We don't follow them on the oldname either to be compatible
3872 * with linux 2.0, and to avoid hard-linking to directories
3873 * and other special files. --ADM
3875 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3876 int, newdfd
, const char __user
*, newname
, int, flags
)
3878 struct dentry
*new_dentry
;
3879 struct path old_path
, new_path
;
3883 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3886 * To use null names we require CAP_DAC_READ_SEARCH
3887 * This ensures that not everyone will be able to create
3888 * handlink using the passed filedescriptor.
3890 if (flags
& AT_EMPTY_PATH
) {
3891 if (!capable(CAP_DAC_READ_SEARCH
))
3896 if (flags
& AT_SYMLINK_FOLLOW
)
3897 how
|= LOOKUP_FOLLOW
;
3899 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3903 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3904 (how
& LOOKUP_REVAL
));
3905 error
= PTR_ERR(new_dentry
);
3906 if (IS_ERR(new_dentry
))
3910 if (old_path
.mnt
!= new_path
.mnt
)
3912 error
= may_linkat(&old_path
);
3913 if (unlikely(error
))
3915 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3918 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3920 done_path_create(&new_path
, new_dentry
);
3921 if (retry_estale(error
, how
)) {
3922 how
|= LOOKUP_REVAL
;
3926 path_put(&old_path
);
3931 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3933 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3937 * The worst of all namespace operations - renaming directory. "Perverted"
3938 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3940 * a) we can get into loop creation. Check is done in is_subdir().
3941 * b) race potential - two innocent renames can create a loop together.
3942 * That's where 4.4 screws up. Current fix: serialization on
3943 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3945 * c) we have to lock _three_ objects - parents and victim (if it exists).
3946 * And that - after we got ->i_mutex on parents (until then we don't know
3947 * whether the target exists). Solution: try to be smart with locking
3948 * order for inodes. We rely on the fact that tree topology may change
3949 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3950 * move will be locked. Thus we can rank directories by the tree
3951 * (ancestors first) and rank all non-directories after them.
3952 * That works since everybody except rename does "lock parent, lookup,
3953 * lock child" and rename is under ->s_vfs_rename_mutex.
3954 * HOWEVER, it relies on the assumption that any object with ->lookup()
3955 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3956 * we'd better make sure that there's no link(2) for them.
3957 * d) conversion from fhandle to dentry may come in the wrong moment - when
3958 * we are removing the target. Solution: we will have to grab ->i_mutex
3959 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3960 * ->i_mutex on parents, which works but leads to some truly excessive
3963 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3964 struct inode
*new_dir
, struct dentry
*new_dentry
)
3967 struct inode
*target
= new_dentry
->d_inode
;
3968 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3971 * If we are going to change the parent - check write permissions,
3972 * we'll need to flip '..'.
3974 if (new_dir
!= old_dir
) {
3975 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3980 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3986 mutex_lock(&target
->i_mutex
);
3989 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3993 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3994 new_dir
->i_nlink
>= max_links
)
3998 shrink_dcache_parent(new_dentry
);
3999 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4004 target
->i_flags
|= S_DEAD
;
4005 dont_mount(new_dentry
);
4009 mutex_unlock(&target
->i_mutex
);
4012 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4013 d_move(old_dentry
,new_dentry
);
4017 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
4018 struct inode
*new_dir
, struct dentry
*new_dentry
)
4020 struct inode
*target
= new_dentry
->d_inode
;
4023 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4029 mutex_lock(&target
->i_mutex
);
4032 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
4035 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4040 dont_mount(new_dentry
);
4041 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4042 d_move(old_dentry
, new_dentry
);
4045 mutex_unlock(&target
->i_mutex
);
4050 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4051 struct inode
*new_dir
, struct dentry
*new_dentry
)
4054 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
4055 const unsigned char *old_name
;
4057 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
4060 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4064 if (!new_dentry
->d_inode
)
4065 error
= may_create(new_dir
, new_dentry
);
4067 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4071 if (!old_dir
->i_op
->rename
)
4074 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4077 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
4079 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
4081 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4082 new_dentry
->d_inode
, old_dentry
);
4083 fsnotify_oldname_free(old_name
);
4088 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4089 int, newdfd
, const char __user
*, newname
)
4091 struct dentry
*old_dir
, *new_dir
;
4092 struct dentry
*old_dentry
, *new_dentry
;
4093 struct dentry
*trap
;
4094 struct nameidata oldnd
, newnd
;
4095 struct filename
*from
;
4096 struct filename
*to
;
4097 unsigned int lookup_flags
= 0;
4098 bool should_retry
= false;
4101 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4103 error
= PTR_ERR(from
);
4107 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4109 error
= PTR_ERR(to
);
4114 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4117 old_dir
= oldnd
.path
.dentry
;
4119 if (oldnd
.last_type
!= LAST_NORM
)
4122 new_dir
= newnd
.path
.dentry
;
4123 if (newnd
.last_type
!= LAST_NORM
)
4126 error
= mnt_want_write(oldnd
.path
.mnt
);
4130 oldnd
.flags
&= ~LOOKUP_PARENT
;
4131 newnd
.flags
&= ~LOOKUP_PARENT
;
4132 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4134 trap
= lock_rename(new_dir
, old_dir
);
4136 old_dentry
= lookup_hash(&oldnd
);
4137 error
= PTR_ERR(old_dentry
);
4138 if (IS_ERR(old_dentry
))
4140 /* source must exist */
4142 if (!old_dentry
->d_inode
)
4144 /* unless the source is a directory trailing slashes give -ENOTDIR */
4145 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
4147 if (oldnd
.last
.name
[oldnd
.last
.len
])
4149 if (newnd
.last
.name
[newnd
.last
.len
])
4152 /* source should not be ancestor of target */
4154 if (old_dentry
== trap
)
4156 new_dentry
= lookup_hash(&newnd
);
4157 error
= PTR_ERR(new_dentry
);
4158 if (IS_ERR(new_dentry
))
4160 /* target should not be an ancestor of source */
4162 if (new_dentry
== trap
)
4165 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4166 &newnd
.path
, new_dentry
);
4169 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4170 new_dir
->d_inode
, new_dentry
);
4176 unlock_rename(new_dir
, old_dir
);
4177 mnt_drop_write(oldnd
.path
.mnt
);
4179 if (retry_estale(error
, lookup_flags
))
4180 should_retry
= true;
4181 path_put(&newnd
.path
);
4184 path_put(&oldnd
.path
);
4187 should_retry
= false;
4188 lookup_flags
|= LOOKUP_REVAL
;
4195 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4197 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
4200 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
4204 len
= PTR_ERR(link
);
4209 if (len
> (unsigned) buflen
)
4211 if (copy_to_user(buffer
, link
, len
))
4218 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4219 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4220 * using) it for any given inode is up to filesystem.
4222 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4224 struct nameidata nd
;
4229 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4231 return PTR_ERR(cookie
);
4233 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
4234 if (dentry
->d_inode
->i_op
->put_link
)
4235 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4239 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
4241 return __vfs_follow_link(nd
, link
);
4244 /* get the link contents into pagecache */
4245 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4249 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4250 page
= read_mapping_page(mapping
, 0, NULL
);
4255 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4259 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4261 struct page
*page
= NULL
;
4262 char *s
= page_getlink(dentry
, &page
);
4263 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4266 page_cache_release(page
);
4271 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4273 struct page
*page
= NULL
;
4274 nd_set_link(nd
, page_getlink(dentry
, &page
));
4278 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4280 struct page
*page
= cookie
;
4284 page_cache_release(page
);
4289 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4291 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4293 struct address_space
*mapping
= inode
->i_mapping
;
4298 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4300 flags
|= AOP_FLAG_NOFS
;
4303 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4304 flags
, &page
, &fsdata
);
4308 kaddr
= kmap_atomic(page
);
4309 memcpy(kaddr
, symname
, len
-1);
4310 kunmap_atomic(kaddr
);
4312 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4319 mark_inode_dirty(inode
);
4325 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4327 return __page_symlink(inode
, symname
, len
,
4328 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4331 const struct inode_operations page_symlink_inode_operations
= {
4332 .readlink
= generic_readlink
,
4333 .follow_link
= page_follow_link_light
,
4334 .put_link
= page_put_link
,
4337 EXPORT_SYMBOL(user_path_at
);
4338 EXPORT_SYMBOL(follow_down_one
);
4339 EXPORT_SYMBOL(follow_down
);
4340 EXPORT_SYMBOL(follow_up
);
4341 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4342 EXPORT_SYMBOL(lock_rename
);
4343 EXPORT_SYMBOL(lookup_one_len
);
4344 EXPORT_SYMBOL(page_follow_link_light
);
4345 EXPORT_SYMBOL(page_put_link
);
4346 EXPORT_SYMBOL(page_readlink
);
4347 EXPORT_SYMBOL(__page_symlink
);
4348 EXPORT_SYMBOL(page_symlink
);
4349 EXPORT_SYMBOL(page_symlink_inode_operations
);
4350 EXPORT_SYMBOL(kern_path
);
4351 EXPORT_SYMBOL(vfs_path_lookup
);
4352 EXPORT_SYMBOL(inode_permission
);
4353 EXPORT_SYMBOL(unlock_rename
);
4354 EXPORT_SYMBOL(vfs_create
);
4355 EXPORT_SYMBOL(vfs_follow_link
);
4356 EXPORT_SYMBOL(vfs_link
);
4357 EXPORT_SYMBOL(vfs_mkdir
);
4358 EXPORT_SYMBOL(vfs_mknod
);
4359 EXPORT_SYMBOL(generic_permission
);
4360 EXPORT_SYMBOL(vfs_readlink
);
4361 EXPORT_SYMBOL(vfs_rename
);
4362 EXPORT_SYMBOL(vfs_rmdir
);
4363 EXPORT_SYMBOL(vfs_symlink
);
4364 EXPORT_SYMBOL(vfs_unlink
);
4365 EXPORT_SYMBOL(dentry_unhash
);
4366 EXPORT_SYMBOL(generic_readlink
);