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_acl(inode
, ACL_TYPE_ACCESS
);
242 int error
= posix_acl_permission(inode
, acl
, mask
);
243 posix_acl_release(acl
);
252 * This does the basic permission checking
254 static int acl_permission_check(struct inode
*inode
, int mask
)
256 unsigned int mode
= inode
->i_mode
;
258 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
261 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
262 int error
= check_acl(inode
, mask
);
263 if (error
!= -EAGAIN
)
267 if (in_group_p(inode
->i_gid
))
272 * If the DACs are ok we don't need any capability check.
274 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
280 * generic_permission - check for access rights on a Posix-like filesystem
281 * @inode: inode to check access rights for
282 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
284 * Used to check for read/write/execute permissions on a file.
285 * We use "fsuid" for this, letting us set arbitrary permissions
286 * for filesystem access without changing the "normal" uids which
287 * are used for other things.
289 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
290 * request cannot be satisfied (eg. requires blocking or too much complexity).
291 * It would then be called again in ref-walk mode.
293 int generic_permission(struct inode
*inode
, int mask
)
298 * Do the basic permission checks.
300 ret
= acl_permission_check(inode
, mask
);
304 if (S_ISDIR(inode
->i_mode
)) {
305 /* DACs are overridable for directories */
306 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
308 if (!(mask
& MAY_WRITE
))
309 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
314 * Read/write DACs are always overridable.
315 * Executable DACs are overridable when there is
316 * at least one exec bit set.
318 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
319 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
323 * Searching includes executable on directories, else just read.
325 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
326 if (mask
== MAY_READ
)
327 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
334 * We _really_ want to just do "generic_permission()" without
335 * even looking at the inode->i_op values. So we keep a cache
336 * flag in inode->i_opflags, that says "this has not special
337 * permission function, use the fast case".
339 static inline int do_inode_permission(struct inode
*inode
, int mask
)
341 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
342 if (likely(inode
->i_op
->permission
))
343 return inode
->i_op
->permission(inode
, mask
);
345 /* This gets set once for the inode lifetime */
346 spin_lock(&inode
->i_lock
);
347 inode
->i_opflags
|= IOP_FASTPERM
;
348 spin_unlock(&inode
->i_lock
);
350 return generic_permission(inode
, mask
);
354 * __inode_permission - Check for access rights to a given inode
355 * @inode: Inode to check permission on
356 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
358 * Check for read/write/execute permissions on an inode.
360 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
362 * This does not check for a read-only file system. You probably want
363 * inode_permission().
365 int __inode_permission(struct inode
*inode
, int mask
)
369 if (unlikely(mask
& MAY_WRITE
)) {
371 * Nobody gets write access to an immutable file.
373 if (IS_IMMUTABLE(inode
))
377 retval
= do_inode_permission(inode
, mask
);
381 retval
= devcgroup_inode_permission(inode
, mask
);
385 return security_inode_permission(inode
, mask
);
389 * sb_permission - Check superblock-level permissions
390 * @sb: Superblock of inode to check permission on
391 * @inode: Inode to check permission on
392 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
394 * Separate out file-system wide checks from inode-specific permission checks.
396 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
398 if (unlikely(mask
& MAY_WRITE
)) {
399 umode_t mode
= inode
->i_mode
;
401 /* Nobody gets write access to a read-only fs. */
402 if ((sb
->s_flags
& MS_RDONLY
) &&
403 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
410 * inode_permission - Check for access rights to a given inode
411 * @inode: Inode to check permission on
412 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
414 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
415 * this, letting us set arbitrary permissions for filesystem access without
416 * changing the "normal" UIDs which are used for other things.
418 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
420 int inode_permission(struct inode
*inode
, int mask
)
424 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
427 return __inode_permission(inode
, mask
);
431 * path_get - get a reference to a path
432 * @path: path to get the reference to
434 * Given a path increment the reference count to the dentry and the vfsmount.
436 void path_get(const struct path
*path
)
441 EXPORT_SYMBOL(path_get
);
444 * path_put - put a reference to a path
445 * @path: path to put the reference to
447 * Given a path decrement the reference count to the dentry and the vfsmount.
449 void path_put(const struct path
*path
)
454 EXPORT_SYMBOL(path_put
);
457 * Path walking has 2 modes, rcu-walk and ref-walk (see
458 * Documentation/filesystems/path-lookup.txt). In situations when we can't
459 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
460 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
461 * mode. Refcounts are grabbed at the last known good point before rcu-walk
462 * got stuck, so ref-walk may continue from there. If this is not successful
463 * (eg. a seqcount has changed), then failure is returned and it's up to caller
464 * to restart the path walk from the beginning in ref-walk mode.
468 * unlazy_walk - try to switch to ref-walk mode.
469 * @nd: nameidata pathwalk data
470 * @dentry: child of nd->path.dentry or NULL
471 * Returns: 0 on success, -ECHILD on failure
473 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
474 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
475 * @nd or NULL. Must be called from rcu-walk context.
477 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
479 struct fs_struct
*fs
= current
->fs
;
480 struct dentry
*parent
= nd
->path
.dentry
;
482 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
485 * After legitimizing the bastards, terminate_walk()
486 * will do the right thing for non-RCU mode, and all our
487 * subsequent exit cases should rcu_read_unlock()
488 * before returning. Do vfsmount first; if dentry
489 * can't be legitimized, just set nd->path.dentry to NULL
490 * and rely on dput(NULL) being a no-op.
492 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
494 nd
->flags
&= ~LOOKUP_RCU
;
496 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
497 nd
->path
.dentry
= NULL
;
502 * For a negative lookup, the lookup sequence point is the parents
503 * sequence point, and it only needs to revalidate the parent dentry.
505 * For a positive lookup, we need to move both the parent and the
506 * dentry from the RCU domain to be properly refcounted. And the
507 * sequence number in the dentry validates *both* dentry counters,
508 * since we checked the sequence number of the parent after we got
509 * the child sequence number. So we know the parent must still
510 * be valid if the child sequence number is still valid.
513 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
515 BUG_ON(nd
->inode
!= parent
->d_inode
);
517 if (!lockref_get_not_dead(&dentry
->d_lockref
))
519 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
524 * Sequence counts matched. Now make sure that the root is
525 * still valid and get it if required.
527 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
528 spin_lock(&fs
->lock
);
529 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
530 goto unlock_and_drop_dentry
;
532 spin_unlock(&fs
->lock
);
538 unlock_and_drop_dentry
:
539 spin_unlock(&fs
->lock
);
547 if (!(nd
->flags
& LOOKUP_ROOT
))
552 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
554 return dentry
->d_op
->d_revalidate(dentry
, flags
);
558 * complete_walk - successful completion of path walk
559 * @nd: pointer nameidata
561 * If we had been in RCU mode, drop out of it and legitimize nd->path.
562 * Revalidate the final result, unless we'd already done that during
563 * the path walk or the filesystem doesn't ask for it. Return 0 on
564 * success, -error on failure. In case of failure caller does not
565 * need to drop nd->path.
567 static int complete_walk(struct nameidata
*nd
)
569 struct dentry
*dentry
= nd
->path
.dentry
;
572 if (nd
->flags
& LOOKUP_RCU
) {
573 nd
->flags
&= ~LOOKUP_RCU
;
574 if (!(nd
->flags
& LOOKUP_ROOT
))
577 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
581 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
583 mntput(nd
->path
.mnt
);
586 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
589 mntput(nd
->path
.mnt
);
595 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
598 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
601 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
612 static __always_inline
void set_root(struct nameidata
*nd
)
615 get_fs_root(current
->fs
, &nd
->root
);
618 static int link_path_walk(const char *, struct nameidata
*);
620 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
623 struct fs_struct
*fs
= current
->fs
;
627 seq
= read_seqcount_begin(&fs
->seq
);
629 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
630 } while (read_seqcount_retry(&fs
->seq
, seq
));
634 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
637 if (path
->mnt
!= nd
->path
.mnt
)
641 static inline void path_to_nameidata(const struct path
*path
,
642 struct nameidata
*nd
)
644 if (!(nd
->flags
& LOOKUP_RCU
)) {
645 dput(nd
->path
.dentry
);
646 if (nd
->path
.mnt
!= path
->mnt
)
647 mntput(nd
->path
.mnt
);
649 nd
->path
.mnt
= path
->mnt
;
650 nd
->path
.dentry
= path
->dentry
;
654 * Helper to directly jump to a known parsed path from ->follow_link,
655 * caller must have taken a reference to path beforehand.
657 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
662 nd
->inode
= nd
->path
.dentry
->d_inode
;
663 nd
->flags
|= LOOKUP_JUMPED
;
666 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
668 struct inode
*inode
= link
->dentry
->d_inode
;
669 if (inode
->i_op
->put_link
)
670 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
674 int sysctl_protected_symlinks __read_mostly
= 0;
675 int sysctl_protected_hardlinks __read_mostly
= 0;
678 * may_follow_link - Check symlink following for unsafe situations
679 * @link: The path of the symlink
680 * @nd: nameidata pathwalk data
682 * In the case of the sysctl_protected_symlinks sysctl being enabled,
683 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
684 * in a sticky world-writable directory. This is to protect privileged
685 * processes from failing races against path names that may change out
686 * from under them by way of other users creating malicious symlinks.
687 * It will permit symlinks to be followed only when outside a sticky
688 * world-writable directory, or when the uid of the symlink and follower
689 * match, or when the directory owner matches the symlink's owner.
691 * Returns 0 if following the symlink is allowed, -ve on error.
693 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
695 const struct inode
*inode
;
696 const struct inode
*parent
;
698 if (!sysctl_protected_symlinks
)
701 /* Allowed if owner and follower match. */
702 inode
= link
->dentry
->d_inode
;
703 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
706 /* Allowed if parent directory not sticky and world-writable. */
707 parent
= nd
->path
.dentry
->d_inode
;
708 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
711 /* Allowed if parent directory and link owner match. */
712 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
715 audit_log_link_denied("follow_link", link
);
716 path_put_conditional(link
, nd
);
722 * safe_hardlink_source - Check for safe hardlink conditions
723 * @inode: the source inode to hardlink from
725 * Return false if at least one of the following conditions:
726 * - inode is not a regular file
728 * - inode is setgid and group-exec
729 * - access failure for read and write
731 * Otherwise returns true.
733 static bool safe_hardlink_source(struct inode
*inode
)
735 umode_t mode
= inode
->i_mode
;
737 /* Special files should not get pinned to the filesystem. */
741 /* Setuid files should not get pinned to the filesystem. */
745 /* Executable setgid files should not get pinned to the filesystem. */
746 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
749 /* Hardlinking to unreadable or unwritable sources is dangerous. */
750 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
757 * may_linkat - Check permissions for creating a hardlink
758 * @link: the source to hardlink from
760 * Block hardlink when all of:
761 * - sysctl_protected_hardlinks enabled
762 * - fsuid does not match inode
763 * - hardlink source is unsafe (see safe_hardlink_source() above)
766 * Returns 0 if successful, -ve on error.
768 static int may_linkat(struct path
*link
)
770 const struct cred
*cred
;
773 if (!sysctl_protected_hardlinks
)
776 cred
= current_cred();
777 inode
= link
->dentry
->d_inode
;
779 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
780 * otherwise, it must be a safe source.
782 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
786 audit_log_link_denied("linkat", link
);
790 static __always_inline
int
791 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
793 struct dentry
*dentry
= link
->dentry
;
797 BUG_ON(nd
->flags
& LOOKUP_RCU
);
799 if (link
->mnt
== nd
->path
.mnt
)
803 if (unlikely(current
->total_link_count
>= 40))
804 goto out_put_nd_path
;
807 current
->total_link_count
++;
810 nd_set_link(nd
, NULL
);
812 error
= security_inode_follow_link(link
->dentry
, nd
);
814 goto out_put_nd_path
;
816 nd
->last_type
= LAST_BIND
;
817 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
820 goto out_put_nd_path
;
825 if (unlikely(IS_ERR(s
))) {
827 put_link(nd
, link
, *p
);
835 nd
->flags
|= LOOKUP_JUMPED
;
837 nd
->inode
= nd
->path
.dentry
->d_inode
;
838 error
= link_path_walk(s
, nd
);
840 put_link(nd
, link
, *p
);
852 static int follow_up_rcu(struct path
*path
)
854 struct mount
*mnt
= real_mount(path
->mnt
);
855 struct mount
*parent
;
856 struct dentry
*mountpoint
;
858 parent
= mnt
->mnt_parent
;
859 if (&parent
->mnt
== path
->mnt
)
861 mountpoint
= mnt
->mnt_mountpoint
;
862 path
->dentry
= mountpoint
;
863 path
->mnt
= &parent
->mnt
;
868 * follow_up - Find the mountpoint of path's vfsmount
870 * Given a path, find the mountpoint of its source file system.
871 * Replace @path with the path of the mountpoint in the parent mount.
874 * Return 1 if we went up a level and 0 if we were already at the
877 int follow_up(struct path
*path
)
879 struct mount
*mnt
= real_mount(path
->mnt
);
880 struct mount
*parent
;
881 struct dentry
*mountpoint
;
883 read_seqlock_excl(&mount_lock
);
884 parent
= mnt
->mnt_parent
;
886 read_sequnlock_excl(&mount_lock
);
889 mntget(&parent
->mnt
);
890 mountpoint
= dget(mnt
->mnt_mountpoint
);
891 read_sequnlock_excl(&mount_lock
);
893 path
->dentry
= mountpoint
;
895 path
->mnt
= &parent
->mnt
;
900 * Perform an automount
901 * - return -EISDIR to tell follow_managed() to stop and return the path we
904 static int follow_automount(struct path
*path
, unsigned flags
,
907 struct vfsmount
*mnt
;
910 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
913 /* We don't want to mount if someone's just doing a stat -
914 * unless they're stat'ing a directory and appended a '/' to
917 * We do, however, want to mount if someone wants to open or
918 * create a file of any type under the mountpoint, wants to
919 * traverse through the mountpoint or wants to open the
920 * mounted directory. Also, autofs may mark negative dentries
921 * as being automount points. These will need the attentions
922 * of the daemon to instantiate them before they can be used.
924 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
925 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
926 path
->dentry
->d_inode
)
929 current
->total_link_count
++;
930 if (current
->total_link_count
>= 40)
933 mnt
= path
->dentry
->d_op
->d_automount(path
);
936 * The filesystem is allowed to return -EISDIR here to indicate
937 * it doesn't want to automount. For instance, autofs would do
938 * this so that its userspace daemon can mount on this dentry.
940 * However, we can only permit this if it's a terminal point in
941 * the path being looked up; if it wasn't then the remainder of
942 * the path is inaccessible and we should say so.
944 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
949 if (!mnt
) /* mount collision */
953 /* lock_mount() may release path->mnt on error */
957 err
= finish_automount(mnt
, path
);
961 /* Someone else made a mount here whilst we were busy */
966 path
->dentry
= dget(mnt
->mnt_root
);
975 * Handle a dentry that is managed in some way.
976 * - Flagged for transit management (autofs)
977 * - Flagged as mountpoint
978 * - Flagged as automount point
980 * This may only be called in refwalk mode.
982 * Serialization is taken care of in namespace.c
984 static int follow_managed(struct path
*path
, unsigned flags
)
986 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
988 bool need_mntput
= false;
991 /* Given that we're not holding a lock here, we retain the value in a
992 * local variable for each dentry as we look at it so that we don't see
993 * the components of that value change under us */
994 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
995 managed
&= DCACHE_MANAGED_DENTRY
,
996 unlikely(managed
!= 0)) {
997 /* Allow the filesystem to manage the transit without i_mutex
999 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1000 BUG_ON(!path
->dentry
->d_op
);
1001 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1002 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1007 /* Transit to a mounted filesystem. */
1008 if (managed
& DCACHE_MOUNTED
) {
1009 struct vfsmount
*mounted
= lookup_mnt(path
);
1014 path
->mnt
= mounted
;
1015 path
->dentry
= dget(mounted
->mnt_root
);
1020 /* Something is mounted on this dentry in another
1021 * namespace and/or whatever was mounted there in this
1022 * namespace got unmounted before lookup_mnt() could
1026 /* Handle an automount point */
1027 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1028 ret
= follow_automount(path
, flags
, &need_mntput
);
1034 /* We didn't change the current path point */
1038 if (need_mntput
&& path
->mnt
== mnt
)
1042 return ret
< 0 ? ret
: need_mntput
;
1045 int follow_down_one(struct path
*path
)
1047 struct vfsmount
*mounted
;
1049 mounted
= lookup_mnt(path
);
1053 path
->mnt
= mounted
;
1054 path
->dentry
= dget(mounted
->mnt_root
);
1060 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1062 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1063 dentry
->d_op
->d_manage(dentry
, true) < 0);
1067 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1068 * we meet a managed dentry that would need blocking.
1070 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1071 struct inode
**inode
)
1074 struct mount
*mounted
;
1076 * Don't forget we might have a non-mountpoint managed dentry
1077 * that wants to block transit.
1079 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1082 if (!d_mountpoint(path
->dentry
))
1085 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1088 path
->mnt
= &mounted
->mnt
;
1089 path
->dentry
= mounted
->mnt
.mnt_root
;
1090 nd
->flags
|= LOOKUP_JUMPED
;
1091 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1093 * Update the inode too. We don't need to re-check the
1094 * dentry sequence number here after this d_inode read,
1095 * because a mount-point is always pinned.
1097 *inode
= path
->dentry
->d_inode
;
1102 static void follow_mount_rcu(struct nameidata
*nd
)
1104 while (d_mountpoint(nd
->path
.dentry
)) {
1105 struct mount
*mounted
;
1106 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1109 nd
->path
.mnt
= &mounted
->mnt
;
1110 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1111 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1115 static int follow_dotdot_rcu(struct nameidata
*nd
)
1120 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1121 nd
->path
.mnt
== nd
->root
.mnt
) {
1124 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1125 struct dentry
*old
= nd
->path
.dentry
;
1126 struct dentry
*parent
= old
->d_parent
;
1129 seq
= read_seqcount_begin(&parent
->d_seq
);
1130 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1132 nd
->path
.dentry
= parent
;
1136 if (!follow_up_rcu(&nd
->path
))
1138 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1140 follow_mount_rcu(nd
);
1141 nd
->inode
= nd
->path
.dentry
->d_inode
;
1145 nd
->flags
&= ~LOOKUP_RCU
;
1146 if (!(nd
->flags
& LOOKUP_ROOT
))
1147 nd
->root
.mnt
= NULL
;
1153 * Follow down to the covering mount currently visible to userspace. At each
1154 * point, the filesystem owning that dentry may be queried as to whether the
1155 * caller is permitted to proceed or not.
1157 int follow_down(struct path
*path
)
1162 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1163 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1164 /* Allow the filesystem to manage the transit without i_mutex
1167 * We indicate to the filesystem if someone is trying to mount
1168 * something here. This gives autofs the chance to deny anyone
1169 * other than its daemon the right to mount on its
1172 * The filesystem may sleep at this point.
1174 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1175 BUG_ON(!path
->dentry
->d_op
);
1176 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1177 ret
= path
->dentry
->d_op
->d_manage(
1178 path
->dentry
, false);
1180 return ret
== -EISDIR
? 0 : ret
;
1183 /* Transit to a mounted filesystem. */
1184 if (managed
& DCACHE_MOUNTED
) {
1185 struct vfsmount
*mounted
= lookup_mnt(path
);
1190 path
->mnt
= mounted
;
1191 path
->dentry
= dget(mounted
->mnt_root
);
1195 /* Don't handle automount points here */
1202 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1204 static void follow_mount(struct path
*path
)
1206 while (d_mountpoint(path
->dentry
)) {
1207 struct vfsmount
*mounted
= lookup_mnt(path
);
1212 path
->mnt
= mounted
;
1213 path
->dentry
= dget(mounted
->mnt_root
);
1217 static void follow_dotdot(struct nameidata
*nd
)
1222 struct dentry
*old
= nd
->path
.dentry
;
1224 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1225 nd
->path
.mnt
== nd
->root
.mnt
) {
1228 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1229 /* rare case of legitimate dget_parent()... */
1230 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1234 if (!follow_up(&nd
->path
))
1237 follow_mount(&nd
->path
);
1238 nd
->inode
= nd
->path
.dentry
->d_inode
;
1242 * This looks up the name in dcache, possibly revalidates the old dentry and
1243 * allocates a new one if not found or not valid. In the need_lookup argument
1244 * returns whether i_op->lookup is necessary.
1246 * dir->d_inode->i_mutex must be held
1248 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1249 unsigned int flags
, bool *need_lookup
)
1251 struct dentry
*dentry
;
1254 *need_lookup
= false;
1255 dentry
= d_lookup(dir
, name
);
1257 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1258 error
= d_revalidate(dentry
, flags
);
1259 if (unlikely(error
<= 0)) {
1262 return ERR_PTR(error
);
1263 } else if (!d_invalidate(dentry
)) {
1272 dentry
= d_alloc(dir
, name
);
1273 if (unlikely(!dentry
))
1274 return ERR_PTR(-ENOMEM
);
1276 *need_lookup
= true;
1282 * Call i_op->lookup on the dentry. The dentry must be negative and
1285 * dir->d_inode->i_mutex must be held
1287 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1292 /* Don't create child dentry for a dead directory. */
1293 if (unlikely(IS_DEADDIR(dir
))) {
1295 return ERR_PTR(-ENOENT
);
1298 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1299 if (unlikely(old
)) {
1306 static struct dentry
*__lookup_hash(struct qstr
*name
,
1307 struct dentry
*base
, unsigned int flags
)
1310 struct dentry
*dentry
;
1312 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1316 return lookup_real(base
->d_inode
, dentry
, flags
);
1320 * It's more convoluted than I'd like it to be, but... it's still fairly
1321 * small and for now I'd prefer to have fast path as straight as possible.
1322 * It _is_ time-critical.
1324 static int lookup_fast(struct nameidata
*nd
,
1325 struct path
*path
, struct inode
**inode
)
1327 struct vfsmount
*mnt
= nd
->path
.mnt
;
1328 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1334 * Rename seqlock is not required here because in the off chance
1335 * of a false negative due to a concurrent rename, we're going to
1336 * do the non-racy lookup, below.
1338 if (nd
->flags
& LOOKUP_RCU
) {
1340 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1345 * This sequence count validates that the inode matches
1346 * the dentry name information from lookup.
1348 *inode
= dentry
->d_inode
;
1349 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1353 * This sequence count validates that the parent had no
1354 * changes while we did the lookup of the dentry above.
1356 * The memory barrier in read_seqcount_begin of child is
1357 * enough, we can use __read_seqcount_retry here.
1359 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1363 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1364 status
= d_revalidate(dentry
, nd
->flags
);
1365 if (unlikely(status
<= 0)) {
1366 if (status
!= -ECHILD
)
1372 path
->dentry
= dentry
;
1373 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1375 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1379 if (unlazy_walk(nd
, dentry
))
1382 dentry
= __d_lookup(parent
, &nd
->last
);
1385 if (unlikely(!dentry
))
1388 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1389 status
= d_revalidate(dentry
, nd
->flags
);
1390 if (unlikely(status
<= 0)) {
1395 if (!d_invalidate(dentry
)) {
1402 path
->dentry
= dentry
;
1403 err
= follow_managed(path
, nd
->flags
);
1404 if (unlikely(err
< 0)) {
1405 path_put_conditional(path
, nd
);
1409 nd
->flags
|= LOOKUP_JUMPED
;
1410 *inode
= path
->dentry
->d_inode
;
1417 /* Fast lookup failed, do it the slow way */
1418 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1420 struct dentry
*dentry
, *parent
;
1423 parent
= nd
->path
.dentry
;
1424 BUG_ON(nd
->inode
!= parent
->d_inode
);
1426 mutex_lock(&parent
->d_inode
->i_mutex
);
1427 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1428 mutex_unlock(&parent
->d_inode
->i_mutex
);
1430 return PTR_ERR(dentry
);
1431 path
->mnt
= nd
->path
.mnt
;
1432 path
->dentry
= dentry
;
1433 err
= follow_managed(path
, nd
->flags
);
1434 if (unlikely(err
< 0)) {
1435 path_put_conditional(path
, nd
);
1439 nd
->flags
|= LOOKUP_JUMPED
;
1443 static inline int may_lookup(struct nameidata
*nd
)
1445 if (nd
->flags
& LOOKUP_RCU
) {
1446 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1449 if (unlazy_walk(nd
, NULL
))
1452 return inode_permission(nd
->inode
, MAY_EXEC
);
1455 static inline int handle_dots(struct nameidata
*nd
, int type
)
1457 if (type
== LAST_DOTDOT
) {
1458 if (nd
->flags
& LOOKUP_RCU
) {
1459 if (follow_dotdot_rcu(nd
))
1467 static void terminate_walk(struct nameidata
*nd
)
1469 if (!(nd
->flags
& LOOKUP_RCU
)) {
1470 path_put(&nd
->path
);
1472 nd
->flags
&= ~LOOKUP_RCU
;
1473 if (!(nd
->flags
& LOOKUP_ROOT
))
1474 nd
->root
.mnt
= NULL
;
1480 * Do we need to follow links? We _really_ want to be able
1481 * to do this check without having to look at inode->i_op,
1482 * so we keep a cache of "no, this doesn't need follow_link"
1483 * for the common case.
1485 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1487 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1490 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1493 struct inode
*inode
;
1496 * "." and ".." are special - ".." especially so because it has
1497 * to be able to know about the current root directory and
1498 * parent relationships.
1500 if (unlikely(nd
->last_type
!= LAST_NORM
))
1501 return handle_dots(nd
, nd
->last_type
);
1502 err
= lookup_fast(nd
, path
, &inode
);
1503 if (unlikely(err
)) {
1507 err
= lookup_slow(nd
, path
);
1511 inode
= path
->dentry
->d_inode
;
1517 if (should_follow_link(path
->dentry
, follow
)) {
1518 if (nd
->flags
& LOOKUP_RCU
) {
1519 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1524 BUG_ON(inode
!= path
->dentry
->d_inode
);
1527 path_to_nameidata(path
, nd
);
1532 path_to_nameidata(path
, nd
);
1539 * This limits recursive symlink follows to 8, while
1540 * limiting consecutive symlinks to 40.
1542 * Without that kind of total limit, nasty chains of consecutive
1543 * symlinks can cause almost arbitrarily long lookups.
1545 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1549 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1550 path_put_conditional(path
, nd
);
1551 path_put(&nd
->path
);
1554 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1557 current
->link_count
++;
1560 struct path link
= *path
;
1563 res
= follow_link(&link
, nd
, &cookie
);
1566 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1567 put_link(nd
, &link
, cookie
);
1570 current
->link_count
--;
1576 * We can do the critical dentry name comparison and hashing
1577 * operations one word at a time, but we are limited to:
1579 * - Architectures with fast unaligned word accesses. We could
1580 * do a "get_unaligned()" if this helps and is sufficiently
1583 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1584 * do not trap on the (extremely unlikely) case of a page
1585 * crossing operation.
1587 * - Furthermore, we need an efficient 64-bit compile for the
1588 * 64-bit case in order to generate the "number of bytes in
1589 * the final mask". Again, that could be replaced with a
1590 * efficient population count instruction or similar.
1592 #ifdef CONFIG_DCACHE_WORD_ACCESS
1594 #include <asm/word-at-a-time.h>
1598 static inline unsigned int fold_hash(unsigned long hash
)
1600 hash
+= hash
>> (8*sizeof(int));
1604 #else /* 32-bit case */
1606 #define fold_hash(x) (x)
1610 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1612 unsigned long a
, mask
;
1613 unsigned long hash
= 0;
1616 a
= load_unaligned_zeropad(name
);
1617 if (len
< sizeof(unsigned long))
1621 name
+= sizeof(unsigned long);
1622 len
-= sizeof(unsigned long);
1626 mask
= bytemask_from_count(len
);
1629 return fold_hash(hash
);
1631 EXPORT_SYMBOL(full_name_hash
);
1634 * Calculate the length and hash of the path component, and
1635 * return the length of the component;
1637 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1639 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1640 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1643 len
= -sizeof(unsigned long);
1645 hash
= (hash
+ a
) * 9;
1646 len
+= sizeof(unsigned long);
1647 a
= load_unaligned_zeropad(name
+len
);
1648 b
= a
^ REPEAT_BYTE('/');
1649 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1651 adata
= prep_zero_mask(a
, adata
, &constants
);
1652 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1654 mask
= create_zero_mask(adata
| bdata
);
1656 hash
+= a
& zero_bytemask(mask
);
1657 *hashp
= fold_hash(hash
);
1659 return len
+ find_zero(mask
);
1664 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1666 unsigned long hash
= init_name_hash();
1668 hash
= partial_name_hash(*name
++, hash
);
1669 return end_name_hash(hash
);
1671 EXPORT_SYMBOL(full_name_hash
);
1674 * We know there's a real path component here of at least
1677 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1679 unsigned long hash
= init_name_hash();
1680 unsigned long len
= 0, c
;
1682 c
= (unsigned char)*name
;
1685 hash
= partial_name_hash(c
, hash
);
1686 c
= (unsigned char)name
[len
];
1687 } while (c
&& c
!= '/');
1688 *hashp
= end_name_hash(hash
);
1696 * This is the basic name resolution function, turning a pathname into
1697 * the final dentry. We expect 'base' to be positive and a directory.
1699 * Returns 0 and nd will have valid dentry and mnt on success.
1700 * Returns error and drops reference to input namei data on failure.
1702 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1712 /* At this point we know we have a real path component. */
1718 err
= may_lookup(nd
);
1722 len
= hash_name(name
, &this.hash
);
1727 if (name
[0] == '.') switch (len
) {
1729 if (name
[1] == '.') {
1731 nd
->flags
|= LOOKUP_JUMPED
;
1737 if (likely(type
== LAST_NORM
)) {
1738 struct dentry
*parent
= nd
->path
.dentry
;
1739 nd
->flags
&= ~LOOKUP_JUMPED
;
1740 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1741 err
= parent
->d_op
->d_hash(parent
, &this);
1748 nd
->last_type
= type
;
1753 * If it wasn't NUL, we know it was '/'. Skip that
1754 * slash, and continue until no more slashes.
1758 } while (unlikely(name
[len
] == '/'));
1764 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1769 err
= nested_symlink(&next
, nd
);
1773 if (!d_is_directory(nd
->path
.dentry
)) {
1782 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1783 struct nameidata
*nd
, struct file
**fp
)
1787 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1788 nd
->flags
= flags
| LOOKUP_JUMPED
;
1790 if (flags
& LOOKUP_ROOT
) {
1791 struct dentry
*root
= nd
->root
.dentry
;
1792 struct inode
*inode
= root
->d_inode
;
1794 if (!d_is_directory(root
))
1796 retval
= inode_permission(inode
, MAY_EXEC
);
1800 nd
->path
= nd
->root
;
1802 if (flags
& LOOKUP_RCU
) {
1804 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1805 nd
->m_seq
= read_seqbegin(&mount_lock
);
1807 path_get(&nd
->path
);
1812 nd
->root
.mnt
= NULL
;
1814 nd
->m_seq
= read_seqbegin(&mount_lock
);
1816 if (flags
& LOOKUP_RCU
) {
1821 path_get(&nd
->root
);
1823 nd
->path
= nd
->root
;
1824 } else if (dfd
== AT_FDCWD
) {
1825 if (flags
& LOOKUP_RCU
) {
1826 struct fs_struct
*fs
= current
->fs
;
1832 seq
= read_seqcount_begin(&fs
->seq
);
1834 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1835 } while (read_seqcount_retry(&fs
->seq
, seq
));
1837 get_fs_pwd(current
->fs
, &nd
->path
);
1840 /* Caller must check execute permissions on the starting path component */
1841 struct fd f
= fdget_raw(dfd
);
1842 struct dentry
*dentry
;
1847 dentry
= f
.file
->f_path
.dentry
;
1850 if (!d_is_directory(dentry
)) {
1856 nd
->path
= f
.file
->f_path
;
1857 if (flags
& LOOKUP_RCU
) {
1860 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1863 path_get(&nd
->path
);
1868 nd
->inode
= nd
->path
.dentry
->d_inode
;
1872 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1874 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1875 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1877 nd
->flags
&= ~LOOKUP_PARENT
;
1878 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1881 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1882 static int path_lookupat(int dfd
, const char *name
,
1883 unsigned int flags
, struct nameidata
*nd
)
1885 struct file
*base
= NULL
;
1890 * Path walking is largely split up into 2 different synchronisation
1891 * schemes, rcu-walk and ref-walk (explained in
1892 * Documentation/filesystems/path-lookup.txt). These share much of the
1893 * path walk code, but some things particularly setup, cleanup, and
1894 * following mounts are sufficiently divergent that functions are
1895 * duplicated. Typically there is a function foo(), and its RCU
1896 * analogue, foo_rcu().
1898 * -ECHILD is the error number of choice (just to avoid clashes) that
1899 * is returned if some aspect of an rcu-walk fails. Such an error must
1900 * be handled by restarting a traditional ref-walk (which will always
1901 * be able to complete).
1903 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1908 current
->total_link_count
= 0;
1909 err
= link_path_walk(name
, nd
);
1911 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1912 err
= lookup_last(nd
, &path
);
1915 struct path link
= path
;
1916 err
= may_follow_link(&link
, nd
);
1919 nd
->flags
|= LOOKUP_PARENT
;
1920 err
= follow_link(&link
, nd
, &cookie
);
1923 err
= lookup_last(nd
, &path
);
1924 put_link(nd
, &link
, cookie
);
1929 err
= complete_walk(nd
);
1931 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1932 if (!d_is_directory(nd
->path
.dentry
)) {
1933 path_put(&nd
->path
);
1941 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1942 path_put(&nd
->root
);
1943 nd
->root
.mnt
= NULL
;
1948 static int filename_lookup(int dfd
, struct filename
*name
,
1949 unsigned int flags
, struct nameidata
*nd
)
1951 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
1952 if (unlikely(retval
== -ECHILD
))
1953 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
1954 if (unlikely(retval
== -ESTALE
))
1955 retval
= path_lookupat(dfd
, name
->name
,
1956 flags
| LOOKUP_REVAL
, nd
);
1958 if (likely(!retval
))
1959 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
1963 static int do_path_lookup(int dfd
, const char *name
,
1964 unsigned int flags
, struct nameidata
*nd
)
1966 struct filename filename
= { .name
= name
};
1968 return filename_lookup(dfd
, &filename
, flags
, nd
);
1971 /* does lookup, returns the object with parent locked */
1972 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
1974 struct nameidata nd
;
1976 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
1978 return ERR_PTR(err
);
1979 if (nd
.last_type
!= LAST_NORM
) {
1981 return ERR_PTR(-EINVAL
);
1983 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1984 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
1986 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
1994 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1996 struct nameidata nd
;
1997 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2004 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2005 * @dentry: pointer to dentry of the base directory
2006 * @mnt: pointer to vfs mount of the base directory
2007 * @name: pointer to file name
2008 * @flags: lookup flags
2009 * @path: pointer to struct path to fill
2011 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2012 const char *name
, unsigned int flags
,
2015 struct nameidata nd
;
2017 nd
.root
.dentry
= dentry
;
2019 BUG_ON(flags
& LOOKUP_PARENT
);
2020 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2021 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2028 * Restricted form of lookup. Doesn't follow links, single-component only,
2029 * needs parent already locked. Doesn't follow mounts.
2032 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2034 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2038 * lookup_one_len - filesystem helper to lookup single pathname component
2039 * @name: pathname component to lookup
2040 * @base: base directory to lookup from
2041 * @len: maximum length @len should be interpreted to
2043 * Note that this routine is purely a helper for filesystem usage and should
2044 * not be called by generic code. Also note that by using this function the
2045 * nameidata argument is passed to the filesystem methods and a filesystem
2046 * using this helper needs to be prepared for that.
2048 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2054 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2058 this.hash
= full_name_hash(name
, len
);
2060 return ERR_PTR(-EACCES
);
2062 if (unlikely(name
[0] == '.')) {
2063 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2064 return ERR_PTR(-EACCES
);
2068 c
= *(const unsigned char *)name
++;
2069 if (c
== '/' || c
== '\0')
2070 return ERR_PTR(-EACCES
);
2073 * See if the low-level filesystem might want
2074 * to use its own hash..
2076 if (base
->d_flags
& DCACHE_OP_HASH
) {
2077 int err
= base
->d_op
->d_hash(base
, &this);
2079 return ERR_PTR(err
);
2082 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2084 return ERR_PTR(err
);
2086 return __lookup_hash(&this, base
, 0);
2089 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2090 struct path
*path
, int *empty
)
2092 struct nameidata nd
;
2093 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2094 int err
= PTR_ERR(tmp
);
2097 BUG_ON(flags
& LOOKUP_PARENT
);
2099 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2107 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2110 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2114 * NB: most callers don't do anything directly with the reference to the
2115 * to struct filename, but the nd->last pointer points into the name string
2116 * allocated by getname. So we must hold the reference to it until all
2117 * path-walking is complete.
2119 static struct filename
*
2120 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2123 struct filename
*s
= getname(path
);
2126 /* only LOOKUP_REVAL is allowed in extra flags */
2127 flags
&= LOOKUP_REVAL
;
2132 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2135 return ERR_PTR(error
);
2142 * mountpoint_last - look up last component for umount
2143 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2144 * @path: pointer to container for result
2146 * This is a special lookup_last function just for umount. In this case, we
2147 * need to resolve the path without doing any revalidation.
2149 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2150 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2151 * in almost all cases, this lookup will be served out of the dcache. The only
2152 * cases where it won't are if nd->last refers to a symlink or the path is
2153 * bogus and it doesn't exist.
2156 * -error: if there was an error during lookup. This includes -ENOENT if the
2157 * lookup found a negative dentry. The nd->path reference will also be
2160 * 0: if we successfully resolved nd->path and found it to not to be a
2161 * symlink that needs to be followed. "path" will also be populated.
2162 * The nd->path reference will also be put.
2164 * 1: if we successfully resolved nd->last and found it to be a symlink
2165 * that needs to be followed. "path" will be populated with the path
2166 * to the link, and nd->path will *not* be put.
2169 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2172 struct dentry
*dentry
;
2173 struct dentry
*dir
= nd
->path
.dentry
;
2175 /* If we're in rcuwalk, drop out of it to handle last component */
2176 if (nd
->flags
& LOOKUP_RCU
) {
2177 if (unlazy_walk(nd
, NULL
)) {
2183 nd
->flags
&= ~LOOKUP_PARENT
;
2185 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2186 error
= handle_dots(nd
, nd
->last_type
);
2189 dentry
= dget(nd
->path
.dentry
);
2193 mutex_lock(&dir
->d_inode
->i_mutex
);
2194 dentry
= d_lookup(dir
, &nd
->last
);
2197 * No cached dentry. Mounted dentries are pinned in the cache,
2198 * so that means that this dentry is probably a symlink or the
2199 * path doesn't actually point to a mounted dentry.
2201 dentry
= d_alloc(dir
, &nd
->last
);
2204 mutex_unlock(&dir
->d_inode
->i_mutex
);
2207 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2208 error
= PTR_ERR(dentry
);
2209 if (IS_ERR(dentry
)) {
2210 mutex_unlock(&dir
->d_inode
->i_mutex
);
2214 mutex_unlock(&dir
->d_inode
->i_mutex
);
2217 if (!dentry
->d_inode
) {
2222 path
->dentry
= dentry
;
2223 path
->mnt
= mntget(nd
->path
.mnt
);
2224 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
))
2234 * path_mountpoint - look up a path to be umounted
2235 * @dfd: directory file descriptor to start walk from
2236 * @name: full pathname to walk
2237 * @path: pointer to container for result
2238 * @flags: lookup flags
2240 * Look up the given name, but don't attempt to revalidate the last component.
2241 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2244 path_mountpoint(int dfd
, const char *name
, struct path
*path
, unsigned int flags
)
2246 struct file
*base
= NULL
;
2247 struct nameidata nd
;
2250 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2254 current
->total_link_count
= 0;
2255 err
= link_path_walk(name
, &nd
);
2259 err
= mountpoint_last(&nd
, path
);
2262 struct path link
= *path
;
2263 err
= may_follow_link(&link
, &nd
);
2266 nd
.flags
|= LOOKUP_PARENT
;
2267 err
= follow_link(&link
, &nd
, &cookie
);
2270 err
= mountpoint_last(&nd
, path
);
2271 put_link(&nd
, &link
, cookie
);
2277 if (nd
.root
.mnt
&& !(nd
.flags
& LOOKUP_ROOT
))
2284 filename_mountpoint(int dfd
, struct filename
*s
, struct path
*path
,
2287 int error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_RCU
);
2288 if (unlikely(error
== -ECHILD
))
2289 error
= path_mountpoint(dfd
, s
->name
, path
, flags
);
2290 if (unlikely(error
== -ESTALE
))
2291 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_REVAL
);
2293 audit_inode(s
, path
->dentry
, 0);
2298 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2299 * @dfd: directory file descriptor
2300 * @name: pathname from userland
2301 * @flags: lookup flags
2302 * @path: pointer to container to hold result
2304 * A umount is a special case for path walking. We're not actually interested
2305 * in the inode in this situation, and ESTALE errors can be a problem. We
2306 * simply want track down the dentry and vfsmount attached at the mountpoint
2307 * and avoid revalidating the last component.
2309 * Returns 0 and populates "path" on success.
2312 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2315 struct filename
*s
= getname(name
);
2319 error
= filename_mountpoint(dfd
, s
, path
, flags
);
2325 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2328 struct filename s
= {.name
= name
};
2329 return filename_mountpoint(dfd
, &s
, path
, flags
);
2331 EXPORT_SYMBOL(kern_path_mountpoint
);
2334 * It's inline, so penalty for filesystems that don't use sticky bit is
2337 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2339 kuid_t fsuid
= current_fsuid();
2341 if (!(dir
->i_mode
& S_ISVTX
))
2343 if (uid_eq(inode
->i_uid
, fsuid
))
2345 if (uid_eq(dir
->i_uid
, fsuid
))
2347 return !inode_capable(inode
, CAP_FOWNER
);
2351 * Check whether we can remove a link victim from directory dir, check
2352 * whether the type of victim is right.
2353 * 1. We can't do it if dir is read-only (done in permission())
2354 * 2. We should have write and exec permissions on dir
2355 * 3. We can't remove anything from append-only dir
2356 * 4. We can't do anything with immutable dir (done in permission())
2357 * 5. If the sticky bit on dir is set we should either
2358 * a. be owner of dir, or
2359 * b. be owner of victim, or
2360 * c. have CAP_FOWNER capability
2361 * 6. If the victim is append-only or immutable we can't do antyhing with
2362 * links pointing to it.
2363 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2364 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2365 * 9. We can't remove a root or mountpoint.
2366 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2367 * nfs_async_unlink().
2369 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2371 struct inode
*inode
= victim
->d_inode
;
2374 if (d_is_negative(victim
))
2378 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2379 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2381 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2387 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2388 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2391 if (!d_is_directory(victim
) && !d_is_autodir(victim
))
2393 if (IS_ROOT(victim
))
2395 } else if (d_is_directory(victim
) || d_is_autodir(victim
))
2397 if (IS_DEADDIR(dir
))
2399 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2404 /* Check whether we can create an object with dentry child in directory
2406 * 1. We can't do it if child already exists (open has special treatment for
2407 * this case, but since we are inlined it's OK)
2408 * 2. We can't do it if dir is read-only (done in permission())
2409 * 3. We should have write and exec permissions on dir
2410 * 4. We can't do it if dir is immutable (done in permission())
2412 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2414 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2417 if (IS_DEADDIR(dir
))
2419 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2423 * p1 and p2 should be directories on the same fs.
2425 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2430 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2434 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2436 p
= d_ancestor(p2
, p1
);
2438 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2439 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2443 p
= d_ancestor(p1
, p2
);
2445 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2446 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2450 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2451 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2455 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2457 mutex_unlock(&p1
->d_inode
->i_mutex
);
2459 mutex_unlock(&p2
->d_inode
->i_mutex
);
2460 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2464 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2467 int error
= may_create(dir
, dentry
);
2471 if (!dir
->i_op
->create
)
2472 return -EACCES
; /* shouldn't it be ENOSYS? */
2475 error
= security_inode_create(dir
, dentry
, mode
);
2478 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2480 fsnotify_create(dir
, dentry
);
2484 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2486 struct dentry
*dentry
= path
->dentry
;
2487 struct inode
*inode
= dentry
->d_inode
;
2497 switch (inode
->i_mode
& S_IFMT
) {
2501 if (acc_mode
& MAY_WRITE
)
2506 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2515 error
= inode_permission(inode
, acc_mode
);
2520 * An append-only file must be opened in append mode for writing.
2522 if (IS_APPEND(inode
)) {
2523 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2529 /* O_NOATIME can only be set by the owner or superuser */
2530 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2536 static int handle_truncate(struct file
*filp
)
2538 struct path
*path
= &filp
->f_path
;
2539 struct inode
*inode
= path
->dentry
->d_inode
;
2540 int error
= get_write_access(inode
);
2544 * Refuse to truncate files with mandatory locks held on them.
2546 error
= locks_verify_locked(inode
);
2548 error
= security_path_truncate(path
);
2550 error
= do_truncate(path
->dentry
, 0,
2551 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2554 put_write_access(inode
);
2558 static inline int open_to_namei_flags(int flag
)
2560 if ((flag
& O_ACCMODE
) == 3)
2565 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2567 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2571 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2575 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2579 * Attempt to atomically look up, create and open a file from a negative
2582 * Returns 0 if successful. The file will have been created and attached to
2583 * @file by the filesystem calling finish_open().
2585 * Returns 1 if the file was looked up only or didn't need creating. The
2586 * caller will need to perform the open themselves. @path will have been
2587 * updated to point to the new dentry. This may be negative.
2589 * Returns an error code otherwise.
2591 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2592 struct path
*path
, struct file
*file
,
2593 const struct open_flags
*op
,
2594 bool got_write
, bool need_lookup
,
2597 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2598 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2602 int create_error
= 0;
2603 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2606 BUG_ON(dentry
->d_inode
);
2608 /* Don't create child dentry for a dead directory. */
2609 if (unlikely(IS_DEADDIR(dir
))) {
2615 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2616 mode
&= ~current_umask();
2618 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2620 open_flag
&= ~O_TRUNC
;
2623 * Checking write permission is tricky, bacuse we don't know if we are
2624 * going to actually need it: O_CREAT opens should work as long as the
2625 * file exists. But checking existence breaks atomicity. The trick is
2626 * to check access and if not granted clear O_CREAT from the flags.
2628 * Another problem is returing the "right" error value (e.g. for an
2629 * O_EXCL open we want to return EEXIST not EROFS).
2631 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2632 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2633 if (!(open_flag
& O_CREAT
)) {
2635 * No O_CREATE -> atomicity not a requirement -> fall
2636 * back to lookup + open
2639 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2640 /* Fall back and fail with the right error */
2641 create_error
= -EROFS
;
2644 /* No side effects, safe to clear O_CREAT */
2645 create_error
= -EROFS
;
2646 open_flag
&= ~O_CREAT
;
2650 if (open_flag
& O_CREAT
) {
2651 error
= may_o_create(&nd
->path
, dentry
, mode
);
2653 create_error
= error
;
2654 if (open_flag
& O_EXCL
)
2656 open_flag
&= ~O_CREAT
;
2660 if (nd
->flags
& LOOKUP_DIRECTORY
)
2661 open_flag
|= O_DIRECTORY
;
2663 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2664 file
->f_path
.mnt
= nd
->path
.mnt
;
2665 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2668 if (create_error
&& error
== -ENOENT
)
2669 error
= create_error
;
2673 if (error
) { /* returned 1, that is */
2674 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2678 if (file
->f_path
.dentry
) {
2680 dentry
= file
->f_path
.dentry
;
2682 if (*opened
& FILE_CREATED
)
2683 fsnotify_create(dir
, dentry
);
2684 if (!dentry
->d_inode
) {
2685 WARN_ON(*opened
& FILE_CREATED
);
2687 error
= create_error
;
2691 if (excl
&& !(*opened
& FILE_CREATED
)) {
2700 * We didn't have the inode before the open, so check open permission
2703 acc_mode
= op
->acc_mode
;
2704 if (*opened
& FILE_CREATED
) {
2705 WARN_ON(!(open_flag
& O_CREAT
));
2706 fsnotify_create(dir
, dentry
);
2707 acc_mode
= MAY_OPEN
;
2709 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2719 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2721 return PTR_ERR(dentry
);
2724 int open_flag
= op
->open_flag
;
2726 error
= create_error
;
2727 if ((open_flag
& O_EXCL
)) {
2728 if (!dentry
->d_inode
)
2730 } else if (!dentry
->d_inode
) {
2732 } else if ((open_flag
& O_TRUNC
) &&
2733 S_ISREG(dentry
->d_inode
->i_mode
)) {
2736 /* will fail later, go on to get the right error */
2740 path
->dentry
= dentry
;
2741 path
->mnt
= nd
->path
.mnt
;
2746 * Look up and maybe create and open the last component.
2748 * Must be called with i_mutex held on parent.
2750 * Returns 0 if the file was successfully atomically created (if necessary) and
2751 * opened. In this case the file will be returned attached to @file.
2753 * Returns 1 if the file was not completely opened at this time, though lookups
2754 * and creations will have been performed and the dentry returned in @path will
2755 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2756 * specified then a negative dentry may be returned.
2758 * An error code is returned otherwise.
2760 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2761 * cleared otherwise prior to returning.
2763 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2765 const struct open_flags
*op
,
2766 bool got_write
, int *opened
)
2768 struct dentry
*dir
= nd
->path
.dentry
;
2769 struct inode
*dir_inode
= dir
->d_inode
;
2770 struct dentry
*dentry
;
2774 *opened
&= ~FILE_CREATED
;
2775 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2777 return PTR_ERR(dentry
);
2779 /* Cached positive dentry: will open in f_op->open */
2780 if (!need_lookup
&& dentry
->d_inode
)
2783 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2784 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2785 need_lookup
, opened
);
2789 BUG_ON(dentry
->d_inode
);
2791 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2793 return PTR_ERR(dentry
);
2796 /* Negative dentry, just create the file */
2797 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2798 umode_t mode
= op
->mode
;
2799 if (!IS_POSIXACL(dir
->d_inode
))
2800 mode
&= ~current_umask();
2802 * This write is needed to ensure that a
2803 * rw->ro transition does not occur between
2804 * the time when the file is created and when
2805 * a permanent write count is taken through
2806 * the 'struct file' in finish_open().
2812 *opened
|= FILE_CREATED
;
2813 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2816 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2817 nd
->flags
& LOOKUP_EXCL
);
2822 path
->dentry
= dentry
;
2823 path
->mnt
= nd
->path
.mnt
;
2832 * Handle the last step of open()
2834 static int do_last(struct nameidata
*nd
, struct path
*path
,
2835 struct file
*file
, const struct open_flags
*op
,
2836 int *opened
, struct filename
*name
)
2838 struct dentry
*dir
= nd
->path
.dentry
;
2839 int open_flag
= op
->open_flag
;
2840 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2841 bool got_write
= false;
2842 int acc_mode
= op
->acc_mode
;
2843 struct inode
*inode
;
2844 bool symlink_ok
= false;
2845 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2846 bool retried
= false;
2849 nd
->flags
&= ~LOOKUP_PARENT
;
2850 nd
->flags
|= op
->intent
;
2852 if (nd
->last_type
!= LAST_NORM
) {
2853 error
= handle_dots(nd
, nd
->last_type
);
2859 if (!(open_flag
& O_CREAT
)) {
2860 if (nd
->last
.name
[nd
->last
.len
])
2861 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2862 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2864 /* we _can_ be in RCU mode here */
2865 error
= lookup_fast(nd
, path
, &inode
);
2872 BUG_ON(nd
->inode
!= dir
->d_inode
);
2874 /* create side of things */
2876 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2877 * has been cleared when we got to the last component we are
2880 error
= complete_walk(nd
);
2884 audit_inode(name
, dir
, LOOKUP_PARENT
);
2886 /* trailing slashes? */
2887 if (nd
->last
.name
[nd
->last
.len
])
2892 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2893 error
= mnt_want_write(nd
->path
.mnt
);
2897 * do _not_ fail yet - we might not need that or fail with
2898 * a different error; let lookup_open() decide; we'll be
2899 * dropping this one anyway.
2902 mutex_lock(&dir
->d_inode
->i_mutex
);
2903 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2904 mutex_unlock(&dir
->d_inode
->i_mutex
);
2910 if ((*opened
& FILE_CREATED
) ||
2911 !S_ISREG(file_inode(file
)->i_mode
))
2912 will_truncate
= false;
2914 audit_inode(name
, file
->f_path
.dentry
, 0);
2918 if (*opened
& FILE_CREATED
) {
2919 /* Don't check for write permission, don't truncate */
2920 open_flag
&= ~O_TRUNC
;
2921 will_truncate
= false;
2922 acc_mode
= MAY_OPEN
;
2923 path_to_nameidata(path
, nd
);
2924 goto finish_open_created
;
2928 * create/update audit record if it already exists.
2930 if (d_is_positive(path
->dentry
))
2931 audit_inode(name
, path
->dentry
, 0);
2934 * If atomic_open() acquired write access it is dropped now due to
2935 * possible mount and symlink following (this might be optimized away if
2939 mnt_drop_write(nd
->path
.mnt
);
2944 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2947 error
= follow_managed(path
, nd
->flags
);
2952 nd
->flags
|= LOOKUP_JUMPED
;
2954 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2955 inode
= path
->dentry
->d_inode
;
2957 /* we _can_ be in RCU mode here */
2959 if (d_is_negative(path
->dentry
)) {
2960 path_to_nameidata(path
, nd
);
2964 if (should_follow_link(path
->dentry
, !symlink_ok
)) {
2965 if (nd
->flags
& LOOKUP_RCU
) {
2966 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2971 BUG_ON(inode
!= path
->dentry
->d_inode
);
2975 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2976 path_to_nameidata(path
, nd
);
2978 save_parent
.dentry
= nd
->path
.dentry
;
2979 save_parent
.mnt
= mntget(path
->mnt
);
2980 nd
->path
.dentry
= path
->dentry
;
2984 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2986 error
= complete_walk(nd
);
2988 path_put(&save_parent
);
2991 audit_inode(name
, nd
->path
.dentry
, 0);
2993 if ((open_flag
& O_CREAT
) &&
2994 (d_is_directory(nd
->path
.dentry
) || d_is_autodir(nd
->path
.dentry
)))
2997 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_is_directory(nd
->path
.dentry
))
2999 if (!S_ISREG(nd
->inode
->i_mode
))
3000 will_truncate
= false;
3002 if (will_truncate
) {
3003 error
= mnt_want_write(nd
->path
.mnt
);
3008 finish_open_created
:
3009 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3012 file
->f_path
.mnt
= nd
->path
.mnt
;
3013 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3015 if (error
== -EOPENSTALE
)
3020 error
= open_check_o_direct(file
);
3023 error
= ima_file_check(file
, op
->acc_mode
);
3027 if (will_truncate
) {
3028 error
= handle_truncate(file
);
3034 mnt_drop_write(nd
->path
.mnt
);
3035 path_put(&save_parent
);
3040 path_put_conditional(path
, nd
);
3047 /* If no saved parent or already retried then can't retry */
3048 if (!save_parent
.dentry
|| retried
)
3051 BUG_ON(save_parent
.dentry
!= dir
);
3052 path_put(&nd
->path
);
3053 nd
->path
= save_parent
;
3054 nd
->inode
= dir
->d_inode
;
3055 save_parent
.mnt
= NULL
;
3056 save_parent
.dentry
= NULL
;
3058 mnt_drop_write(nd
->path
.mnt
);
3065 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3066 struct nameidata
*nd
, int flags
,
3067 const struct open_flags
*op
,
3068 struct file
*file
, int *opened
)
3070 static const struct qstr name
= QSTR_INIT("/", 1);
3071 struct dentry
*dentry
, *child
;
3073 int error
= path_lookupat(dfd
, pathname
->name
,
3074 flags
| LOOKUP_DIRECTORY
, nd
);
3075 if (unlikely(error
))
3077 error
= mnt_want_write(nd
->path
.mnt
);
3078 if (unlikely(error
))
3080 /* we want directory to be writable */
3081 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3084 dentry
= nd
->path
.dentry
;
3085 dir
= dentry
->d_inode
;
3086 if (!dir
->i_op
->tmpfile
) {
3087 error
= -EOPNOTSUPP
;
3090 child
= d_alloc(dentry
, &name
);
3091 if (unlikely(!child
)) {
3095 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3096 nd
->flags
|= op
->intent
;
3097 dput(nd
->path
.dentry
);
3098 nd
->path
.dentry
= child
;
3099 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3102 audit_inode(pathname
, nd
->path
.dentry
, 0);
3103 error
= may_open(&nd
->path
, op
->acc_mode
, op
->open_flag
);
3106 file
->f_path
.mnt
= nd
->path
.mnt
;
3107 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3110 error
= open_check_o_direct(file
);
3113 } else if (!(op
->open_flag
& O_EXCL
)) {
3114 struct inode
*inode
= file_inode(file
);
3115 spin_lock(&inode
->i_lock
);
3116 inode
->i_state
|= I_LINKABLE
;
3117 spin_unlock(&inode
->i_lock
);
3120 mnt_drop_write(nd
->path
.mnt
);
3122 path_put(&nd
->path
);
3126 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3127 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3129 struct file
*base
= NULL
;
3135 file
= get_empty_filp();
3139 file
->f_flags
= op
->open_flag
;
3141 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3142 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3146 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
3147 if (unlikely(error
))
3150 current
->total_link_count
= 0;
3151 error
= link_path_walk(pathname
->name
, nd
);
3152 if (unlikely(error
))
3155 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3156 while (unlikely(error
> 0)) { /* trailing symlink */
3157 struct path link
= path
;
3159 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3160 path_put_conditional(&path
, nd
);
3161 path_put(&nd
->path
);
3165 error
= may_follow_link(&link
, nd
);
3166 if (unlikely(error
))
3168 nd
->flags
|= LOOKUP_PARENT
;
3169 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3170 error
= follow_link(&link
, nd
, &cookie
);
3171 if (unlikely(error
))
3173 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3174 put_link(nd
, &link
, cookie
);
3177 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
3178 path_put(&nd
->root
);
3181 if (!(opened
& FILE_OPENED
)) {
3185 if (unlikely(error
)) {
3186 if (error
== -EOPENSTALE
) {
3187 if (flags
& LOOKUP_RCU
)
3192 file
= ERR_PTR(error
);
3197 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3198 const struct open_flags
*op
)
3200 struct nameidata nd
;
3201 int flags
= op
->lookup_flags
;
3204 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3205 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3206 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3207 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3208 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3212 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3213 const char *name
, const struct open_flags
*op
)
3215 struct nameidata nd
;
3217 struct filename filename
= { .name
= name
};
3218 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3221 nd
.root
.dentry
= dentry
;
3223 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3224 return ERR_PTR(-ELOOP
);
3226 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3227 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3228 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3229 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3230 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3234 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3235 struct path
*path
, unsigned int lookup_flags
)
3237 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3238 struct nameidata nd
;
3241 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3244 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3245 * other flags passed in are ignored!
3247 lookup_flags
&= LOOKUP_REVAL
;
3249 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3251 return ERR_PTR(error
);
3254 * Yucky last component or no last component at all?
3255 * (foo/., foo/.., /////)
3257 if (nd
.last_type
!= LAST_NORM
)
3259 nd
.flags
&= ~LOOKUP_PARENT
;
3260 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3262 /* don't fail immediately if it's r/o, at least try to report other errors */
3263 err2
= mnt_want_write(nd
.path
.mnt
);
3265 * Do the final lookup.
3267 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3268 dentry
= lookup_hash(&nd
);
3273 if (d_is_positive(dentry
))
3277 * Special case - lookup gave negative, but... we had foo/bar/
3278 * From the vfs_mknod() POV we just have a negative dentry -
3279 * all is fine. Let's be bastards - you had / on the end, you've
3280 * been asking for (non-existent) directory. -ENOENT for you.
3282 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3286 if (unlikely(err2
)) {
3294 dentry
= ERR_PTR(error
);
3296 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3298 mnt_drop_write(nd
.path
.mnt
);
3303 EXPORT_SYMBOL(kern_path_create
);
3305 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3308 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3309 mnt_drop_write(path
->mnt
);
3312 EXPORT_SYMBOL(done_path_create
);
3314 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3315 struct path
*path
, unsigned int lookup_flags
)
3317 struct filename
*tmp
= getname(pathname
);
3320 return ERR_CAST(tmp
);
3321 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3325 EXPORT_SYMBOL(user_path_create
);
3327 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3329 int error
= may_create(dir
, dentry
);
3334 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3337 if (!dir
->i_op
->mknod
)
3340 error
= devcgroup_inode_mknod(mode
, dev
);
3344 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3348 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3350 fsnotify_create(dir
, dentry
);
3354 static int may_mknod(umode_t mode
)
3356 switch (mode
& S_IFMT
) {
3362 case 0: /* zero mode translates to S_IFREG */
3371 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3374 struct dentry
*dentry
;
3377 unsigned int lookup_flags
= 0;
3379 error
= may_mknod(mode
);
3383 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3385 return PTR_ERR(dentry
);
3387 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3388 mode
&= ~current_umask();
3389 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3392 switch (mode
& S_IFMT
) {
3393 case 0: case S_IFREG
:
3394 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3396 case S_IFCHR
: case S_IFBLK
:
3397 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3398 new_decode_dev(dev
));
3400 case S_IFIFO
: case S_IFSOCK
:
3401 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3405 done_path_create(&path
, dentry
);
3406 if (retry_estale(error
, lookup_flags
)) {
3407 lookup_flags
|= LOOKUP_REVAL
;
3413 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3415 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3418 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3420 int error
= may_create(dir
, dentry
);
3421 unsigned max_links
= dir
->i_sb
->s_max_links
;
3426 if (!dir
->i_op
->mkdir
)
3429 mode
&= (S_IRWXUGO
|S_ISVTX
);
3430 error
= security_inode_mkdir(dir
, dentry
, mode
);
3434 if (max_links
&& dir
->i_nlink
>= max_links
)
3437 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3439 fsnotify_mkdir(dir
, dentry
);
3443 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3445 struct dentry
*dentry
;
3448 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3451 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3453 return PTR_ERR(dentry
);
3455 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3456 mode
&= ~current_umask();
3457 error
= security_path_mkdir(&path
, dentry
, mode
);
3459 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3460 done_path_create(&path
, dentry
);
3461 if (retry_estale(error
, lookup_flags
)) {
3462 lookup_flags
|= LOOKUP_REVAL
;
3468 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3470 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3474 * The dentry_unhash() helper will try to drop the dentry early: we
3475 * should have a usage count of 1 if we're the only user of this
3476 * dentry, and if that is true (possibly after pruning the dcache),
3477 * then we drop the dentry now.
3479 * A low-level filesystem can, if it choses, legally
3482 * if (!d_unhashed(dentry))
3485 * if it cannot handle the case of removing a directory
3486 * that is still in use by something else..
3488 void dentry_unhash(struct dentry
*dentry
)
3490 shrink_dcache_parent(dentry
);
3491 spin_lock(&dentry
->d_lock
);
3492 if (dentry
->d_lockref
.count
== 1)
3494 spin_unlock(&dentry
->d_lock
);
3497 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3499 int error
= may_delete(dir
, dentry
, 1);
3504 if (!dir
->i_op
->rmdir
)
3508 mutex_lock(&dentry
->d_inode
->i_mutex
);
3511 if (d_mountpoint(dentry
))
3514 error
= security_inode_rmdir(dir
, dentry
);
3518 shrink_dcache_parent(dentry
);
3519 error
= dir
->i_op
->rmdir(dir
, dentry
);
3523 dentry
->d_inode
->i_flags
|= S_DEAD
;
3527 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3534 static long do_rmdir(int dfd
, const char __user
*pathname
)
3537 struct filename
*name
;
3538 struct dentry
*dentry
;
3539 struct nameidata nd
;
3540 unsigned int lookup_flags
= 0;
3542 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3544 return PTR_ERR(name
);
3546 switch(nd
.last_type
) {
3558 nd
.flags
&= ~LOOKUP_PARENT
;
3559 error
= mnt_want_write(nd
.path
.mnt
);
3563 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3564 dentry
= lookup_hash(&nd
);
3565 error
= PTR_ERR(dentry
);
3568 if (!dentry
->d_inode
) {
3572 error
= security_path_rmdir(&nd
.path
, dentry
);
3575 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3579 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3580 mnt_drop_write(nd
.path
.mnt
);
3584 if (retry_estale(error
, lookup_flags
)) {
3585 lookup_flags
|= LOOKUP_REVAL
;
3591 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3593 return do_rmdir(AT_FDCWD
, pathname
);
3597 * vfs_unlink - unlink a filesystem object
3598 * @dir: parent directory
3600 * @delegated_inode: returns victim inode, if the inode is delegated.
3602 * The caller must hold dir->i_mutex.
3604 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3605 * return a reference to the inode in delegated_inode. The caller
3606 * should then break the delegation on that inode and retry. Because
3607 * breaking a delegation may take a long time, the caller should drop
3608 * dir->i_mutex before doing so.
3610 * Alternatively, a caller may pass NULL for delegated_inode. This may
3611 * be appropriate for callers that expect the underlying filesystem not
3612 * to be NFS exported.
3614 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3616 struct inode
*target
= dentry
->d_inode
;
3617 int error
= may_delete(dir
, dentry
, 0);
3622 if (!dir
->i_op
->unlink
)
3625 mutex_lock(&target
->i_mutex
);
3626 if (d_mountpoint(dentry
))
3629 error
= security_inode_unlink(dir
, dentry
);
3631 error
= try_break_deleg(target
, delegated_inode
);
3634 error
= dir
->i_op
->unlink(dir
, dentry
);
3640 mutex_unlock(&target
->i_mutex
);
3642 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3643 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3644 fsnotify_link_count(target
);
3652 * Make sure that the actual truncation of the file will occur outside its
3653 * directory's i_mutex. Truncate can take a long time if there is a lot of
3654 * writeout happening, and we don't want to prevent access to the directory
3655 * while waiting on the I/O.
3657 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3660 struct filename
*name
;
3661 struct dentry
*dentry
;
3662 struct nameidata nd
;
3663 struct inode
*inode
= NULL
;
3664 struct inode
*delegated_inode
= NULL
;
3665 unsigned int lookup_flags
= 0;
3667 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3669 return PTR_ERR(name
);
3672 if (nd
.last_type
!= LAST_NORM
)
3675 nd
.flags
&= ~LOOKUP_PARENT
;
3676 error
= mnt_want_write(nd
.path
.mnt
);
3680 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3681 dentry
= lookup_hash(&nd
);
3682 error
= PTR_ERR(dentry
);
3683 if (!IS_ERR(dentry
)) {
3684 /* Why not before? Because we want correct error value */
3685 if (nd
.last
.name
[nd
.last
.len
])
3687 inode
= dentry
->d_inode
;
3688 if (d_is_negative(dentry
))
3691 error
= security_path_unlink(&nd
.path
, dentry
);
3694 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3698 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3700 iput(inode
); /* truncate the inode here */
3702 if (delegated_inode
) {
3703 error
= break_deleg_wait(&delegated_inode
);
3707 mnt_drop_write(nd
.path
.mnt
);
3711 if (retry_estale(error
, lookup_flags
)) {
3712 lookup_flags
|= LOOKUP_REVAL
;
3719 if (d_is_negative(dentry
))
3721 else if (d_is_directory(dentry
) || d_is_autodir(dentry
))
3728 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3730 if ((flag
& ~AT_REMOVEDIR
) != 0)
3733 if (flag
& AT_REMOVEDIR
)
3734 return do_rmdir(dfd
, pathname
);
3736 return do_unlinkat(dfd
, pathname
);
3739 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3741 return do_unlinkat(AT_FDCWD
, pathname
);
3744 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3746 int error
= may_create(dir
, dentry
);
3751 if (!dir
->i_op
->symlink
)
3754 error
= security_inode_symlink(dir
, dentry
, oldname
);
3758 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3760 fsnotify_create(dir
, dentry
);
3764 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3765 int, newdfd
, const char __user
*, newname
)
3768 struct filename
*from
;
3769 struct dentry
*dentry
;
3771 unsigned int lookup_flags
= 0;
3773 from
= getname(oldname
);
3775 return PTR_ERR(from
);
3777 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3778 error
= PTR_ERR(dentry
);
3782 error
= security_path_symlink(&path
, dentry
, from
->name
);
3784 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3785 done_path_create(&path
, dentry
);
3786 if (retry_estale(error
, lookup_flags
)) {
3787 lookup_flags
|= LOOKUP_REVAL
;
3795 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3797 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3801 * vfs_link - create a new link
3802 * @old_dentry: object to be linked
3804 * @new_dentry: where to create the new link
3805 * @delegated_inode: returns inode needing a delegation break
3807 * The caller must hold dir->i_mutex
3809 * If vfs_link discovers a delegation on the to-be-linked file in need
3810 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3811 * inode in delegated_inode. The caller should then break the delegation
3812 * and retry. Because breaking a delegation may take a long time, the
3813 * caller should drop the i_mutex before doing so.
3815 * Alternatively, a caller may pass NULL for delegated_inode. This may
3816 * be appropriate for callers that expect the underlying filesystem not
3817 * to be NFS exported.
3819 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3821 struct inode
*inode
= old_dentry
->d_inode
;
3822 unsigned max_links
= dir
->i_sb
->s_max_links
;
3828 error
= may_create(dir
, new_dentry
);
3832 if (dir
->i_sb
!= inode
->i_sb
)
3836 * A link to an append-only or immutable file cannot be created.
3838 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3840 if (!dir
->i_op
->link
)
3842 if (S_ISDIR(inode
->i_mode
))
3845 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3849 mutex_lock(&inode
->i_mutex
);
3850 /* Make sure we don't allow creating hardlink to an unlinked file */
3851 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3853 else if (max_links
&& inode
->i_nlink
>= max_links
)
3856 error
= try_break_deleg(inode
, delegated_inode
);
3858 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3861 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3862 spin_lock(&inode
->i_lock
);
3863 inode
->i_state
&= ~I_LINKABLE
;
3864 spin_unlock(&inode
->i_lock
);
3866 mutex_unlock(&inode
->i_mutex
);
3868 fsnotify_link(dir
, inode
, new_dentry
);
3873 * Hardlinks are often used in delicate situations. We avoid
3874 * security-related surprises by not following symlinks on the
3877 * We don't follow them on the oldname either to be compatible
3878 * with linux 2.0, and to avoid hard-linking to directories
3879 * and other special files. --ADM
3881 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3882 int, newdfd
, const char __user
*, newname
, int, flags
)
3884 struct dentry
*new_dentry
;
3885 struct path old_path
, new_path
;
3886 struct inode
*delegated_inode
= NULL
;
3890 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3893 * To use null names we require CAP_DAC_READ_SEARCH
3894 * This ensures that not everyone will be able to create
3895 * handlink using the passed filedescriptor.
3897 if (flags
& AT_EMPTY_PATH
) {
3898 if (!capable(CAP_DAC_READ_SEARCH
))
3903 if (flags
& AT_SYMLINK_FOLLOW
)
3904 how
|= LOOKUP_FOLLOW
;
3906 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3910 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3911 (how
& LOOKUP_REVAL
));
3912 error
= PTR_ERR(new_dentry
);
3913 if (IS_ERR(new_dentry
))
3917 if (old_path
.mnt
!= new_path
.mnt
)
3919 error
= may_linkat(&old_path
);
3920 if (unlikely(error
))
3922 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3925 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
3927 done_path_create(&new_path
, new_dentry
);
3928 if (delegated_inode
) {
3929 error
= break_deleg_wait(&delegated_inode
);
3933 if (retry_estale(error
, how
)) {
3934 how
|= LOOKUP_REVAL
;
3938 path_put(&old_path
);
3943 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3945 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3949 * The worst of all namespace operations - renaming directory. "Perverted"
3950 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3952 * a) we can get into loop creation. Check is done in is_subdir().
3953 * b) race potential - two innocent renames can create a loop together.
3954 * That's where 4.4 screws up. Current fix: serialization on
3955 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3957 * c) we have to lock _four_ objects - parents and victim (if it exists),
3958 * and source (if it is not a directory).
3959 * And that - after we got ->i_mutex on parents (until then we don't know
3960 * whether the target exists). Solution: try to be smart with locking
3961 * order for inodes. We rely on the fact that tree topology may change
3962 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3963 * move will be locked. Thus we can rank directories by the tree
3964 * (ancestors first) and rank all non-directories after them.
3965 * That works since everybody except rename does "lock parent, lookup,
3966 * lock child" and rename is under ->s_vfs_rename_mutex.
3967 * HOWEVER, it relies on the assumption that any object with ->lookup()
3968 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3969 * we'd better make sure that there's no link(2) for them.
3970 * d) conversion from fhandle to dentry may come in the wrong moment - when
3971 * we are removing the target. Solution: we will have to grab ->i_mutex
3972 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3973 * ->i_mutex on parents, which works but leads to some truly excessive
3976 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3977 struct inode
*new_dir
, struct dentry
*new_dentry
)
3980 struct inode
*target
= new_dentry
->d_inode
;
3981 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3984 * If we are going to change the parent - check write permissions,
3985 * we'll need to flip '..'.
3987 if (new_dir
!= old_dir
) {
3988 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3993 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3999 mutex_lock(&target
->i_mutex
);
4002 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
4006 if (max_links
&& !target
&& new_dir
!= old_dir
&&
4007 new_dir
->i_nlink
>= max_links
)
4011 shrink_dcache_parent(new_dentry
);
4012 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4017 target
->i_flags
|= S_DEAD
;
4018 dont_mount(new_dentry
);
4022 mutex_unlock(&target
->i_mutex
);
4025 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4026 d_move(old_dentry
,new_dentry
);
4030 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
4031 struct inode
*new_dir
, struct dentry
*new_dentry
,
4032 struct inode
**delegated_inode
)
4034 struct inode
*target
= new_dentry
->d_inode
;
4035 struct inode
*source
= old_dentry
->d_inode
;
4038 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4043 lock_two_nondirectories(source
, target
);
4046 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
4049 error
= try_break_deleg(source
, delegated_inode
);
4053 error
= try_break_deleg(target
, delegated_inode
);
4057 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4062 dont_mount(new_dentry
);
4063 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4064 d_move(old_dentry
, new_dentry
);
4066 unlock_two_nondirectories(source
, target
);
4072 * vfs_rename - rename a filesystem object
4073 * @old_dir: parent of source
4074 * @old_dentry: source
4075 * @new_dir: parent of destination
4076 * @new_dentry: destination
4077 * @delegated_inode: returns an inode needing a delegation break
4079 * The caller must hold multiple mutexes--see lock_rename()).
4081 * If vfs_rename discovers a delegation in need of breaking at either
4082 * the source or destination, it will return -EWOULDBLOCK and return a
4083 * reference to the inode in delegated_inode. The caller should then
4084 * break the delegation and retry. Because breaking a delegation may
4085 * take a long time, the caller should drop all locks before doing
4088 * Alternatively, a caller may pass NULL for delegated_inode. This may
4089 * be appropriate for callers that expect the underlying filesystem not
4090 * to be NFS exported.
4092 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4093 struct inode
*new_dir
, struct dentry
*new_dentry
,
4094 struct inode
**delegated_inode
)
4097 int is_dir
= d_is_directory(old_dentry
) || d_is_autodir(old_dentry
);
4098 const unsigned char *old_name
;
4100 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
4103 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4107 if (!new_dentry
->d_inode
)
4108 error
= may_create(new_dir
, new_dentry
);
4110 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4114 if (!old_dir
->i_op
->rename
)
4117 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4120 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
4122 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
,delegated_inode
);
4124 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4125 new_dentry
->d_inode
, old_dentry
);
4126 fsnotify_oldname_free(old_name
);
4131 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4132 int, newdfd
, const char __user
*, newname
)
4134 struct dentry
*old_dir
, *new_dir
;
4135 struct dentry
*old_dentry
, *new_dentry
;
4136 struct dentry
*trap
;
4137 struct nameidata oldnd
, newnd
;
4138 struct inode
*delegated_inode
= NULL
;
4139 struct filename
*from
;
4140 struct filename
*to
;
4141 unsigned int lookup_flags
= 0;
4142 bool should_retry
= false;
4145 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4147 error
= PTR_ERR(from
);
4151 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4153 error
= PTR_ERR(to
);
4158 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4161 old_dir
= oldnd
.path
.dentry
;
4163 if (oldnd
.last_type
!= LAST_NORM
)
4166 new_dir
= newnd
.path
.dentry
;
4167 if (newnd
.last_type
!= LAST_NORM
)
4170 error
= mnt_want_write(oldnd
.path
.mnt
);
4174 oldnd
.flags
&= ~LOOKUP_PARENT
;
4175 newnd
.flags
&= ~LOOKUP_PARENT
;
4176 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4179 trap
= lock_rename(new_dir
, old_dir
);
4181 old_dentry
= lookup_hash(&oldnd
);
4182 error
= PTR_ERR(old_dentry
);
4183 if (IS_ERR(old_dentry
))
4185 /* source must exist */
4187 if (d_is_negative(old_dentry
))
4189 /* unless the source is a directory trailing slashes give -ENOTDIR */
4190 if (!d_is_directory(old_dentry
) && !d_is_autodir(old_dentry
)) {
4192 if (oldnd
.last
.name
[oldnd
.last
.len
])
4194 if (newnd
.last
.name
[newnd
.last
.len
])
4197 /* source should not be ancestor of target */
4199 if (old_dentry
== trap
)
4201 new_dentry
= lookup_hash(&newnd
);
4202 error
= PTR_ERR(new_dentry
);
4203 if (IS_ERR(new_dentry
))
4205 /* target should not be an ancestor of source */
4207 if (new_dentry
== trap
)
4210 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4211 &newnd
.path
, new_dentry
);
4214 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4215 new_dir
->d_inode
, new_dentry
,
4222 unlock_rename(new_dir
, old_dir
);
4223 if (delegated_inode
) {
4224 error
= break_deleg_wait(&delegated_inode
);
4228 mnt_drop_write(oldnd
.path
.mnt
);
4230 if (retry_estale(error
, lookup_flags
))
4231 should_retry
= true;
4232 path_put(&newnd
.path
);
4235 path_put(&oldnd
.path
);
4238 should_retry
= false;
4239 lookup_flags
|= LOOKUP_REVAL
;
4246 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4248 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
4251 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
4255 len
= PTR_ERR(link
);
4260 if (len
> (unsigned) buflen
)
4262 if (copy_to_user(buffer
, link
, len
))
4269 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4270 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4271 * using) it for any given inode is up to filesystem.
4273 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4275 struct nameidata nd
;
4280 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4282 return PTR_ERR(cookie
);
4284 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
4285 if (dentry
->d_inode
->i_op
->put_link
)
4286 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4290 /* get the link contents into pagecache */
4291 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4295 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4296 page
= read_mapping_page(mapping
, 0, NULL
);
4301 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4305 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4307 struct page
*page
= NULL
;
4308 char *s
= page_getlink(dentry
, &page
);
4309 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4312 page_cache_release(page
);
4317 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4319 struct page
*page
= NULL
;
4320 nd_set_link(nd
, page_getlink(dentry
, &page
));
4324 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4326 struct page
*page
= cookie
;
4330 page_cache_release(page
);
4335 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4337 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4339 struct address_space
*mapping
= inode
->i_mapping
;
4344 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4346 flags
|= AOP_FLAG_NOFS
;
4349 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4350 flags
, &page
, &fsdata
);
4354 kaddr
= kmap_atomic(page
);
4355 memcpy(kaddr
, symname
, len
-1);
4356 kunmap_atomic(kaddr
);
4358 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4365 mark_inode_dirty(inode
);
4371 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4373 return __page_symlink(inode
, symname
, len
,
4374 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4377 const struct inode_operations page_symlink_inode_operations
= {
4378 .readlink
= generic_readlink
,
4379 .follow_link
= page_follow_link_light
,
4380 .put_link
= page_put_link
,
4383 EXPORT_SYMBOL(user_path_at
);
4384 EXPORT_SYMBOL(follow_down_one
);
4385 EXPORT_SYMBOL(follow_down
);
4386 EXPORT_SYMBOL(follow_up
);
4387 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4388 EXPORT_SYMBOL(lock_rename
);
4389 EXPORT_SYMBOL(lookup_one_len
);
4390 EXPORT_SYMBOL(page_follow_link_light
);
4391 EXPORT_SYMBOL(page_put_link
);
4392 EXPORT_SYMBOL(page_readlink
);
4393 EXPORT_SYMBOL(__page_symlink
);
4394 EXPORT_SYMBOL(page_symlink
);
4395 EXPORT_SYMBOL(page_symlink_inode_operations
);
4396 EXPORT_SYMBOL(kern_path
);
4397 EXPORT_SYMBOL(vfs_path_lookup
);
4398 EXPORT_SYMBOL(inode_permission
);
4399 EXPORT_SYMBOL(unlock_rename
);
4400 EXPORT_SYMBOL(vfs_create
);
4401 EXPORT_SYMBOL(vfs_link
);
4402 EXPORT_SYMBOL(vfs_mkdir
);
4403 EXPORT_SYMBOL(vfs_mknod
);
4404 EXPORT_SYMBOL(generic_permission
);
4405 EXPORT_SYMBOL(vfs_readlink
);
4406 EXPORT_SYMBOL(vfs_rename
);
4407 EXPORT_SYMBOL(vfs_rmdir
);
4408 EXPORT_SYMBOL(vfs_symlink
);
4409 EXPORT_SYMBOL(vfs_unlink
);
4410 EXPORT_SYMBOL(dentry_unhash
);
4411 EXPORT_SYMBOL(generic_readlink
);