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 <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user
*filename
, int flags
, int *empty
)
127 struct filename
*result
;
131 result
= audit_reusename(filename
);
135 result
= __getname();
136 if (unlikely(!result
))
137 return ERR_PTR(-ENOMEM
);
140 * First, try to embed the struct filename inside the names_cache
143 kname
= (char *)result
->iname
;
144 result
->name
= kname
;
146 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
147 if (unlikely(len
< 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
159 const size_t size
= offsetof(struct filename
, iname
[1]);
160 kname
= (char *)result
;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result
= kzalloc(size
, GFP_KERNEL
);
168 if (unlikely(!result
)) {
170 return ERR_PTR(-ENOMEM
);
172 result
->name
= kname
;
173 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
174 if (unlikely(len
< 0)) {
179 if (unlikely(len
== PATH_MAX
)) {
182 return ERR_PTR(-ENAMETOOLONG
);
187 /* The empty path is special. */
188 if (unlikely(!len
)) {
191 if (!(flags
& LOOKUP_EMPTY
)) {
193 return ERR_PTR(-ENOENT
);
197 result
->uptr
= filename
;
198 result
->aname
= NULL
;
199 audit_getname(result
);
204 getname(const char __user
* filename
)
206 return getname_flags(filename
, 0, NULL
);
210 getname_kernel(const char * filename
)
212 struct filename
*result
;
213 int len
= strlen(filename
) + 1;
215 result
= __getname();
216 if (unlikely(!result
))
217 return ERR_PTR(-ENOMEM
);
219 if (len
<= EMBEDDED_NAME_MAX
) {
220 result
->name
= (char *)result
->iname
;
221 } else if (len
<= PATH_MAX
) {
222 struct filename
*tmp
;
224 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
225 if (unlikely(!tmp
)) {
227 return ERR_PTR(-ENOMEM
);
229 tmp
->name
= (char *)result
;
233 return ERR_PTR(-ENAMETOOLONG
);
235 memcpy((char *)result
->name
, filename
, len
);
237 result
->aname
= NULL
;
239 audit_getname(result
);
244 void putname(struct filename
*name
)
246 BUG_ON(name
->refcnt
<= 0);
248 if (--name
->refcnt
> 0)
251 if (name
->name
!= name
->iname
) {
252 __putname(name
->name
);
258 static int check_acl(struct inode
*inode
, int mask
)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl
*acl
;
263 if (mask
& MAY_NOT_BLOCK
) {
264 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl
== ACL_NOT_CACHED
)
270 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
273 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
277 int error
= posix_acl_permission(inode
, acl
, mask
);
278 posix_acl_release(acl
);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode
*inode
, int mask
)
291 unsigned int mode
= inode
->i_mode
;
293 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
296 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
297 int error
= check_acl(inode
, mask
);
298 if (error
!= -EAGAIN
)
302 if (in_group_p(inode
->i_gid
))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode
*inode
, int mask
)
333 * Do the basic permission checks.
335 ret
= acl_permission_check(inode
, mask
);
339 if (S_ISDIR(inode
->i_mode
)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
343 if (!(mask
& MAY_WRITE
))
344 if (capable_wrt_inode_uidgid(inode
,
345 CAP_DAC_READ_SEARCH
))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
355 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
359 * Searching includes executable on directories, else just read.
361 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
362 if (mask
== MAY_READ
)
363 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
368 EXPORT_SYMBOL(generic_permission
);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode
*inode
, int mask
)
378 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
379 if (likely(inode
->i_op
->permission
))
380 return inode
->i_op
->permission(inode
, mask
);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode
->i_lock
);
384 inode
->i_opflags
|= IOP_FASTPERM
;
385 spin_unlock(&inode
->i_lock
);
387 return generic_permission(inode
, mask
);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode
*inode
, int mask
)
406 if (unlikely(mask
& MAY_WRITE
)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode
))
414 retval
= do_inode_permission(inode
, mask
);
418 retval
= devcgroup_inode_permission(inode
, mask
);
422 return security_inode_permission(inode
, mask
);
424 EXPORT_SYMBOL(__inode_permission
);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
436 if (unlikely(mask
& MAY_WRITE
)) {
437 umode_t mode
= inode
->i_mode
;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb
->s_flags
& MS_RDONLY
) &&
441 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode
*inode
, int mask
)
462 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
465 return __inode_permission(inode
, mask
);
467 EXPORT_SYMBOL(inode_permission
);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path
*path
)
480 EXPORT_SYMBOL(path_get
);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path
*path
)
493 EXPORT_SYMBOL(path_put
);
502 struct inode
*inode
; /* path.dentry.d_inode */
512 } stack
[MAX_NESTED_LINKS
+ 1];
516 * Path walking has 2 modes, rcu-walk and ref-walk (see
517 * Documentation/filesystems/path-lookup.txt). In situations when we can't
518 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
519 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
520 * mode. Refcounts are grabbed at the last known good point before rcu-walk
521 * got stuck, so ref-walk may continue from there. If this is not successful
522 * (eg. a seqcount has changed), then failure is returned and it's up to caller
523 * to restart the path walk from the beginning in ref-walk mode.
527 * unlazy_walk - try to switch to ref-walk mode.
528 * @nd: nameidata pathwalk data
529 * @dentry: child of nd->path.dentry or NULL
530 * Returns: 0 on success, -ECHILD on failure
532 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
533 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
534 * @nd or NULL. Must be called from rcu-walk context.
536 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
538 struct fs_struct
*fs
= current
->fs
;
539 struct dentry
*parent
= nd
->path
.dentry
;
541 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
544 * After legitimizing the bastards, terminate_walk()
545 * will do the right thing for non-RCU mode, and all our
546 * subsequent exit cases should rcu_read_unlock()
547 * before returning. Do vfsmount first; if dentry
548 * can't be legitimized, just set nd->path.dentry to NULL
549 * and rely on dput(NULL) being a no-op.
551 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
553 nd
->flags
&= ~LOOKUP_RCU
;
555 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
556 nd
->path
.dentry
= NULL
;
561 * For a negative lookup, the lookup sequence point is the parents
562 * sequence point, and it only needs to revalidate the parent dentry.
564 * For a positive lookup, we need to move both the parent and the
565 * dentry from the RCU domain to be properly refcounted. And the
566 * sequence number in the dentry validates *both* dentry counters,
567 * since we checked the sequence number of the parent after we got
568 * the child sequence number. So we know the parent must still
569 * be valid if the child sequence number is still valid.
572 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
574 BUG_ON(nd
->inode
!= parent
->d_inode
);
576 if (!lockref_get_not_dead(&dentry
->d_lockref
))
578 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
583 * Sequence counts matched. Now make sure that the root is
584 * still valid and get it if required.
586 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
587 spin_lock(&fs
->lock
);
588 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
589 goto unlock_and_drop_dentry
;
591 spin_unlock(&fs
->lock
);
597 unlock_and_drop_dentry
:
598 spin_unlock(&fs
->lock
);
606 if (!(nd
->flags
& LOOKUP_ROOT
))
611 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
613 return dentry
->d_op
->d_revalidate(dentry
, flags
);
617 * complete_walk - successful completion of path walk
618 * @nd: pointer nameidata
620 * If we had been in RCU mode, drop out of it and legitimize nd->path.
621 * Revalidate the final result, unless we'd already done that during
622 * the path walk or the filesystem doesn't ask for it. Return 0 on
623 * success, -error on failure. In case of failure caller does not
624 * need to drop nd->path.
626 static int complete_walk(struct nameidata
*nd
)
628 struct dentry
*dentry
= nd
->path
.dentry
;
631 if (nd
->flags
& LOOKUP_RCU
) {
632 nd
->flags
&= ~LOOKUP_RCU
;
633 if (!(nd
->flags
& LOOKUP_ROOT
))
636 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
640 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
642 mntput(nd
->path
.mnt
);
645 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
648 mntput(nd
->path
.mnt
);
654 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
657 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
660 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
671 static __always_inline
void set_root(struct nameidata
*nd
)
673 get_fs_root(current
->fs
, &nd
->root
);
676 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
678 struct fs_struct
*fs
= current
->fs
;
682 seq
= read_seqcount_begin(&fs
->seq
);
684 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
685 } while (read_seqcount_retry(&fs
->seq
, seq
));
689 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
692 if (path
->mnt
!= nd
->path
.mnt
)
696 static inline void path_to_nameidata(const struct path
*path
,
697 struct nameidata
*nd
)
699 if (!(nd
->flags
& LOOKUP_RCU
)) {
700 dput(nd
->path
.dentry
);
701 if (nd
->path
.mnt
!= path
->mnt
)
702 mntput(nd
->path
.mnt
);
704 nd
->path
.mnt
= path
->mnt
;
705 nd
->path
.dentry
= path
->dentry
;
709 * Helper to directly jump to a known parsed path from ->follow_link,
710 * caller must have taken a reference to path beforehand.
712 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
717 nd
->inode
= nd
->path
.dentry
->d_inode
;
718 nd
->flags
|= LOOKUP_JUMPED
;
721 static inline void put_link(struct nameidata
*nd
)
723 struct saved
*last
= nd
->stack
+ nd
->depth
;
724 struct inode
*inode
= last
->link
.dentry
->d_inode
;
725 if (last
->cookie
&& inode
->i_op
->put_link
)
726 inode
->i_op
->put_link(last
->link
.dentry
, last
->cookie
);
727 path_put(&last
->link
);
730 int sysctl_protected_symlinks __read_mostly
= 0;
731 int sysctl_protected_hardlinks __read_mostly
= 0;
734 * may_follow_link - Check symlink following for unsafe situations
735 * @link: The path of the symlink
736 * @nd: nameidata pathwalk data
738 * In the case of the sysctl_protected_symlinks sysctl being enabled,
739 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
740 * in a sticky world-writable directory. This is to protect privileged
741 * processes from failing races against path names that may change out
742 * from under them by way of other users creating malicious symlinks.
743 * It will permit symlinks to be followed only when outside a sticky
744 * world-writable directory, or when the uid of the symlink and follower
745 * match, or when the directory owner matches the symlink's owner.
747 * Returns 0 if following the symlink is allowed, -ve on error.
749 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
751 const struct inode
*inode
;
752 const struct inode
*parent
;
754 if (!sysctl_protected_symlinks
)
757 /* Allowed if owner and follower match. */
758 inode
= link
->dentry
->d_inode
;
759 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
762 /* Allowed if parent directory not sticky and world-writable. */
763 parent
= nd
->path
.dentry
->d_inode
;
764 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
767 /* Allowed if parent directory and link owner match. */
768 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
771 audit_log_link_denied("follow_link", link
);
772 path_put_conditional(link
, nd
);
778 * safe_hardlink_source - Check for safe hardlink conditions
779 * @inode: the source inode to hardlink from
781 * Return false if at least one of the following conditions:
782 * - inode is not a regular file
784 * - inode is setgid and group-exec
785 * - access failure for read and write
787 * Otherwise returns true.
789 static bool safe_hardlink_source(struct inode
*inode
)
791 umode_t mode
= inode
->i_mode
;
793 /* Special files should not get pinned to the filesystem. */
797 /* Setuid files should not get pinned to the filesystem. */
801 /* Executable setgid files should not get pinned to the filesystem. */
802 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
805 /* Hardlinking to unreadable or unwritable sources is dangerous. */
806 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
813 * may_linkat - Check permissions for creating a hardlink
814 * @link: the source to hardlink from
816 * Block hardlink when all of:
817 * - sysctl_protected_hardlinks enabled
818 * - fsuid does not match inode
819 * - hardlink source is unsafe (see safe_hardlink_source() above)
822 * Returns 0 if successful, -ve on error.
824 static int may_linkat(struct path
*link
)
826 const struct cred
*cred
;
829 if (!sysctl_protected_hardlinks
)
832 cred
= current_cred();
833 inode
= link
->dentry
->d_inode
;
835 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
836 * otherwise, it must be a safe source.
838 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
842 audit_log_link_denied("linkat", link
);
846 static __always_inline
847 const char *get_link(struct nameidata
*nd
)
849 struct saved
*last
= nd
->stack
+ nd
->depth
;
850 struct dentry
*dentry
= nd
->link
.dentry
;
851 struct inode
*inode
= dentry
->d_inode
;
855 BUG_ON(nd
->flags
& LOOKUP_RCU
);
857 if (nd
->link
.mnt
== nd
->path
.mnt
)
858 mntget(nd
->link
.mnt
);
860 if (unlikely(current
->total_link_count
>= 40)) {
863 return ERR_PTR(-ELOOP
);
866 last
->link
= nd
->link
;
870 current
->total_link_count
++;
872 touch_atime(&last
->link
);
874 error
= security_inode_follow_link(dentry
);
875 res
= ERR_PTR(error
);
879 nd
->last_type
= LAST_BIND
;
882 res
= inode
->i_op
->follow_link(dentry
, &last
->cookie
, nd
);
886 path_put(&last
->link
);
892 static int follow_up_rcu(struct path
*path
)
894 struct mount
*mnt
= real_mount(path
->mnt
);
895 struct mount
*parent
;
896 struct dentry
*mountpoint
;
898 parent
= mnt
->mnt_parent
;
899 if (&parent
->mnt
== path
->mnt
)
901 mountpoint
= mnt
->mnt_mountpoint
;
902 path
->dentry
= mountpoint
;
903 path
->mnt
= &parent
->mnt
;
908 * follow_up - Find the mountpoint of path's vfsmount
910 * Given a path, find the mountpoint of its source file system.
911 * Replace @path with the path of the mountpoint in the parent mount.
914 * Return 1 if we went up a level and 0 if we were already at the
917 int follow_up(struct path
*path
)
919 struct mount
*mnt
= real_mount(path
->mnt
);
920 struct mount
*parent
;
921 struct dentry
*mountpoint
;
923 read_seqlock_excl(&mount_lock
);
924 parent
= mnt
->mnt_parent
;
926 read_sequnlock_excl(&mount_lock
);
929 mntget(&parent
->mnt
);
930 mountpoint
= dget(mnt
->mnt_mountpoint
);
931 read_sequnlock_excl(&mount_lock
);
933 path
->dentry
= mountpoint
;
935 path
->mnt
= &parent
->mnt
;
938 EXPORT_SYMBOL(follow_up
);
941 * Perform an automount
942 * - return -EISDIR to tell follow_managed() to stop and return the path we
945 static int follow_automount(struct path
*path
, unsigned flags
,
948 struct vfsmount
*mnt
;
951 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
954 /* We don't want to mount if someone's just doing a stat -
955 * unless they're stat'ing a directory and appended a '/' to
958 * We do, however, want to mount if someone wants to open or
959 * create a file of any type under the mountpoint, wants to
960 * traverse through the mountpoint or wants to open the
961 * mounted directory. Also, autofs may mark negative dentries
962 * as being automount points. These will need the attentions
963 * of the daemon to instantiate them before they can be used.
965 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
966 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
967 path
->dentry
->d_inode
)
970 current
->total_link_count
++;
971 if (current
->total_link_count
>= 40)
974 mnt
= path
->dentry
->d_op
->d_automount(path
);
977 * The filesystem is allowed to return -EISDIR here to indicate
978 * it doesn't want to automount. For instance, autofs would do
979 * this so that its userspace daemon can mount on this dentry.
981 * However, we can only permit this if it's a terminal point in
982 * the path being looked up; if it wasn't then the remainder of
983 * the path is inaccessible and we should say so.
985 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
990 if (!mnt
) /* mount collision */
994 /* lock_mount() may release path->mnt on error */
998 err
= finish_automount(mnt
, path
);
1002 /* Someone else made a mount here whilst we were busy */
1007 path
->dentry
= dget(mnt
->mnt_root
);
1016 * Handle a dentry that is managed in some way.
1017 * - Flagged for transit management (autofs)
1018 * - Flagged as mountpoint
1019 * - Flagged as automount point
1021 * This may only be called in refwalk mode.
1023 * Serialization is taken care of in namespace.c
1025 static int follow_managed(struct path
*path
, unsigned flags
)
1027 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1029 bool need_mntput
= false;
1032 /* Given that we're not holding a lock here, we retain the value in a
1033 * local variable for each dentry as we look at it so that we don't see
1034 * the components of that value change under us */
1035 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1036 managed
&= DCACHE_MANAGED_DENTRY
,
1037 unlikely(managed
!= 0)) {
1038 /* Allow the filesystem to manage the transit without i_mutex
1040 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1041 BUG_ON(!path
->dentry
->d_op
);
1042 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1043 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1048 /* Transit to a mounted filesystem. */
1049 if (managed
& DCACHE_MOUNTED
) {
1050 struct vfsmount
*mounted
= lookup_mnt(path
);
1055 path
->mnt
= mounted
;
1056 path
->dentry
= dget(mounted
->mnt_root
);
1061 /* Something is mounted on this dentry in another
1062 * namespace and/or whatever was mounted there in this
1063 * namespace got unmounted before lookup_mnt() could
1067 /* Handle an automount point */
1068 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1069 ret
= follow_automount(path
, flags
, &need_mntput
);
1075 /* We didn't change the current path point */
1079 if (need_mntput
&& path
->mnt
== mnt
)
1083 return ret
< 0 ? ret
: need_mntput
;
1086 int follow_down_one(struct path
*path
)
1088 struct vfsmount
*mounted
;
1090 mounted
= lookup_mnt(path
);
1094 path
->mnt
= mounted
;
1095 path
->dentry
= dget(mounted
->mnt_root
);
1100 EXPORT_SYMBOL(follow_down_one
);
1102 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1104 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1105 dentry
->d_op
->d_manage(dentry
, true) : 0;
1109 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1110 * we meet a managed dentry that would need blocking.
1112 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1113 struct inode
**inode
)
1116 struct mount
*mounted
;
1118 * Don't forget we might have a non-mountpoint managed dentry
1119 * that wants to block transit.
1121 switch (managed_dentry_rcu(path
->dentry
)) {
1131 if (!d_mountpoint(path
->dentry
))
1132 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1134 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1137 path
->mnt
= &mounted
->mnt
;
1138 path
->dentry
= mounted
->mnt
.mnt_root
;
1139 nd
->flags
|= LOOKUP_JUMPED
;
1140 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1142 * Update the inode too. We don't need to re-check the
1143 * dentry sequence number here after this d_inode read,
1144 * because a mount-point is always pinned.
1146 *inode
= path
->dentry
->d_inode
;
1148 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1149 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1152 static int follow_dotdot_rcu(struct nameidata
*nd
)
1154 struct inode
*inode
= nd
->inode
;
1159 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1160 nd
->path
.mnt
== nd
->root
.mnt
) {
1163 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1164 struct dentry
*old
= nd
->path
.dentry
;
1165 struct dentry
*parent
= old
->d_parent
;
1168 inode
= parent
->d_inode
;
1169 seq
= read_seqcount_begin(&parent
->d_seq
);
1170 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1172 nd
->path
.dentry
= parent
;
1176 if (!follow_up_rcu(&nd
->path
))
1178 inode
= nd
->path
.dentry
->d_inode
;
1179 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1181 while (d_mountpoint(nd
->path
.dentry
)) {
1182 struct mount
*mounted
;
1183 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1186 nd
->path
.mnt
= &mounted
->mnt
;
1187 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1188 inode
= nd
->path
.dentry
->d_inode
;
1189 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1190 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1197 nd
->flags
&= ~LOOKUP_RCU
;
1198 if (!(nd
->flags
& LOOKUP_ROOT
))
1199 nd
->root
.mnt
= NULL
;
1205 * Follow down to the covering mount currently visible to userspace. At each
1206 * point, the filesystem owning that dentry may be queried as to whether the
1207 * caller is permitted to proceed or not.
1209 int follow_down(struct path
*path
)
1214 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1215 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1216 /* Allow the filesystem to manage the transit without i_mutex
1219 * We indicate to the filesystem if someone is trying to mount
1220 * something here. This gives autofs the chance to deny anyone
1221 * other than its daemon the right to mount on its
1224 * The filesystem may sleep at this point.
1226 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1227 BUG_ON(!path
->dentry
->d_op
);
1228 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1229 ret
= path
->dentry
->d_op
->d_manage(
1230 path
->dentry
, false);
1232 return ret
== -EISDIR
? 0 : ret
;
1235 /* Transit to a mounted filesystem. */
1236 if (managed
& DCACHE_MOUNTED
) {
1237 struct vfsmount
*mounted
= lookup_mnt(path
);
1242 path
->mnt
= mounted
;
1243 path
->dentry
= dget(mounted
->mnt_root
);
1247 /* Don't handle automount points here */
1252 EXPORT_SYMBOL(follow_down
);
1255 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1257 static void follow_mount(struct path
*path
)
1259 while (d_mountpoint(path
->dentry
)) {
1260 struct vfsmount
*mounted
= lookup_mnt(path
);
1265 path
->mnt
= mounted
;
1266 path
->dentry
= dget(mounted
->mnt_root
);
1270 static void follow_dotdot(struct nameidata
*nd
)
1276 struct dentry
*old
= nd
->path
.dentry
;
1278 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1279 nd
->path
.mnt
== nd
->root
.mnt
) {
1282 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1283 /* rare case of legitimate dget_parent()... */
1284 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1288 if (!follow_up(&nd
->path
))
1291 follow_mount(&nd
->path
);
1292 nd
->inode
= nd
->path
.dentry
->d_inode
;
1296 * This looks up the name in dcache, possibly revalidates the old dentry and
1297 * allocates a new one if not found or not valid. In the need_lookup argument
1298 * returns whether i_op->lookup is necessary.
1300 * dir->d_inode->i_mutex must be held
1302 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1303 unsigned int flags
, bool *need_lookup
)
1305 struct dentry
*dentry
;
1308 *need_lookup
= false;
1309 dentry
= d_lookup(dir
, name
);
1311 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1312 error
= d_revalidate(dentry
, flags
);
1313 if (unlikely(error
<= 0)) {
1316 return ERR_PTR(error
);
1318 d_invalidate(dentry
);
1327 dentry
= d_alloc(dir
, name
);
1328 if (unlikely(!dentry
))
1329 return ERR_PTR(-ENOMEM
);
1331 *need_lookup
= true;
1337 * Call i_op->lookup on the dentry. The dentry must be negative and
1340 * dir->d_inode->i_mutex must be held
1342 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1347 /* Don't create child dentry for a dead directory. */
1348 if (unlikely(IS_DEADDIR(dir
))) {
1350 return ERR_PTR(-ENOENT
);
1353 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1354 if (unlikely(old
)) {
1361 static struct dentry
*__lookup_hash(struct qstr
*name
,
1362 struct dentry
*base
, unsigned int flags
)
1365 struct dentry
*dentry
;
1367 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1371 return lookup_real(base
->d_inode
, dentry
, flags
);
1375 * It's more convoluted than I'd like it to be, but... it's still fairly
1376 * small and for now I'd prefer to have fast path as straight as possible.
1377 * It _is_ time-critical.
1379 static int lookup_fast(struct nameidata
*nd
,
1380 struct path
*path
, struct inode
**inode
)
1382 struct vfsmount
*mnt
= nd
->path
.mnt
;
1383 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1389 * Rename seqlock is not required here because in the off chance
1390 * of a false negative due to a concurrent rename, we're going to
1391 * do the non-racy lookup, below.
1393 if (nd
->flags
& LOOKUP_RCU
) {
1396 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1401 * This sequence count validates that the inode matches
1402 * the dentry name information from lookup.
1404 *inode
= dentry
->d_inode
;
1405 negative
= d_is_negative(dentry
);
1406 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1412 * This sequence count validates that the parent had no
1413 * changes while we did the lookup of the dentry above.
1415 * The memory barrier in read_seqcount_begin of child is
1416 * enough, we can use __read_seqcount_retry here.
1418 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1422 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1423 status
= d_revalidate(dentry
, nd
->flags
);
1424 if (unlikely(status
<= 0)) {
1425 if (status
!= -ECHILD
)
1431 path
->dentry
= dentry
;
1432 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1435 if (unlazy_walk(nd
, dentry
))
1438 dentry
= __d_lookup(parent
, &nd
->last
);
1441 if (unlikely(!dentry
))
1444 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1445 status
= d_revalidate(dentry
, nd
->flags
);
1446 if (unlikely(status
<= 0)) {
1451 d_invalidate(dentry
);
1456 if (unlikely(d_is_negative(dentry
))) {
1461 path
->dentry
= dentry
;
1462 err
= follow_managed(path
, nd
->flags
);
1463 if (unlikely(err
< 0)) {
1464 path_put_conditional(path
, nd
);
1468 nd
->flags
|= LOOKUP_JUMPED
;
1469 *inode
= path
->dentry
->d_inode
;
1476 /* Fast lookup failed, do it the slow way */
1477 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1479 struct dentry
*dentry
, *parent
;
1482 parent
= nd
->path
.dentry
;
1483 BUG_ON(nd
->inode
!= parent
->d_inode
);
1485 mutex_lock(&parent
->d_inode
->i_mutex
);
1486 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1487 mutex_unlock(&parent
->d_inode
->i_mutex
);
1489 return PTR_ERR(dentry
);
1490 path
->mnt
= nd
->path
.mnt
;
1491 path
->dentry
= dentry
;
1492 err
= follow_managed(path
, nd
->flags
);
1493 if (unlikely(err
< 0)) {
1494 path_put_conditional(path
, nd
);
1498 nd
->flags
|= LOOKUP_JUMPED
;
1502 static inline int may_lookup(struct nameidata
*nd
)
1504 if (nd
->flags
& LOOKUP_RCU
) {
1505 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1508 if (unlazy_walk(nd
, NULL
))
1511 return inode_permission(nd
->inode
, MAY_EXEC
);
1514 static inline int handle_dots(struct nameidata
*nd
, int type
)
1516 if (type
== LAST_DOTDOT
) {
1517 if (nd
->flags
& LOOKUP_RCU
) {
1518 if (follow_dotdot_rcu(nd
))
1526 static void terminate_walk(struct nameidata
*nd
)
1528 if (!(nd
->flags
& LOOKUP_RCU
)) {
1529 path_put(&nd
->path
);
1531 nd
->flags
&= ~LOOKUP_RCU
;
1532 if (!(nd
->flags
& LOOKUP_ROOT
))
1533 nd
->root
.mnt
= NULL
;
1539 * Do we need to follow links? We _really_ want to be able
1540 * to do this check without having to look at inode->i_op,
1541 * so we keep a cache of "no, this doesn't need follow_link"
1542 * for the common case.
1544 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1546 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1549 static int walk_component(struct nameidata
*nd
, int follow
)
1552 struct inode
*inode
;
1555 * "." and ".." are special - ".." especially so because it has
1556 * to be able to know about the current root directory and
1557 * parent relationships.
1559 if (unlikely(nd
->last_type
!= LAST_NORM
))
1560 return handle_dots(nd
, nd
->last_type
);
1561 err
= lookup_fast(nd
, &path
, &inode
);
1562 if (unlikely(err
)) {
1566 err
= lookup_slow(nd
, &path
);
1570 inode
= path
.dentry
->d_inode
;
1572 if (d_is_negative(path
.dentry
))
1576 if (should_follow_link(path
.dentry
, follow
)) {
1577 if (nd
->flags
& LOOKUP_RCU
) {
1578 if (unlikely(nd
->path
.mnt
!= path
.mnt
||
1579 unlazy_walk(nd
, path
.dentry
))) {
1584 BUG_ON(inode
!= path
.dentry
->d_inode
);
1588 path_to_nameidata(&path
, nd
);
1593 path_to_nameidata(&path
, nd
);
1600 * We can do the critical dentry name comparison and hashing
1601 * operations one word at a time, but we are limited to:
1603 * - Architectures with fast unaligned word accesses. We could
1604 * do a "get_unaligned()" if this helps and is sufficiently
1607 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1608 * do not trap on the (extremely unlikely) case of a page
1609 * crossing operation.
1611 * - Furthermore, we need an efficient 64-bit compile for the
1612 * 64-bit case in order to generate the "number of bytes in
1613 * the final mask". Again, that could be replaced with a
1614 * efficient population count instruction or similar.
1616 #ifdef CONFIG_DCACHE_WORD_ACCESS
1618 #include <asm/word-at-a-time.h>
1622 static inline unsigned int fold_hash(unsigned long hash
)
1624 return hash_64(hash
, 32);
1627 #else /* 32-bit case */
1629 #define fold_hash(x) (x)
1633 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1635 unsigned long a
, mask
;
1636 unsigned long hash
= 0;
1639 a
= load_unaligned_zeropad(name
);
1640 if (len
< sizeof(unsigned long))
1644 name
+= sizeof(unsigned long);
1645 len
-= sizeof(unsigned long);
1649 mask
= bytemask_from_count(len
);
1652 return fold_hash(hash
);
1654 EXPORT_SYMBOL(full_name_hash
);
1657 * Calculate the length and hash of the path component, and
1658 * return the "hash_len" as the result.
1660 static inline u64
hash_name(const char *name
)
1662 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1663 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1666 len
= -sizeof(unsigned long);
1668 hash
= (hash
+ a
) * 9;
1669 len
+= sizeof(unsigned long);
1670 a
= load_unaligned_zeropad(name
+len
);
1671 b
= a
^ REPEAT_BYTE('/');
1672 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1674 adata
= prep_zero_mask(a
, adata
, &constants
);
1675 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1677 mask
= create_zero_mask(adata
| bdata
);
1679 hash
+= a
& zero_bytemask(mask
);
1680 len
+= find_zero(mask
);
1681 return hashlen_create(fold_hash(hash
), len
);
1686 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1688 unsigned long hash
= init_name_hash();
1690 hash
= partial_name_hash(*name
++, hash
);
1691 return end_name_hash(hash
);
1693 EXPORT_SYMBOL(full_name_hash
);
1696 * We know there's a real path component here of at least
1699 static inline u64
hash_name(const char *name
)
1701 unsigned long hash
= init_name_hash();
1702 unsigned long len
= 0, c
;
1704 c
= (unsigned char)*name
;
1707 hash
= partial_name_hash(c
, hash
);
1708 c
= (unsigned char)name
[len
];
1709 } while (c
&& c
!= '/');
1710 return hashlen_create(end_name_hash(hash
), len
);
1717 * This is the basic name resolution function, turning a pathname into
1718 * the final dentry. We expect 'base' to be positive and a directory.
1720 * Returns 0 and nd will have valid dentry and mnt on success.
1721 * Returns error and drops reference to input namei data on failure.
1723 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1732 /* At this point we know we have a real path component. */
1737 err
= may_lookup(nd
);
1741 hash_len
= hash_name(name
);
1744 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1746 if (name
[1] == '.') {
1748 nd
->flags
|= LOOKUP_JUMPED
;
1754 if (likely(type
== LAST_NORM
)) {
1755 struct dentry
*parent
= nd
->path
.dentry
;
1756 nd
->flags
&= ~LOOKUP_JUMPED
;
1757 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1758 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1759 err
= parent
->d_op
->d_hash(parent
, &this);
1762 hash_len
= this.hash_len
;
1767 nd
->last
.hash_len
= hash_len
;
1768 nd
->last
.name
= name
;
1769 nd
->last_type
= type
;
1771 name
+= hashlen_len(hash_len
);
1775 * If it wasn't NUL, we know it was '/'. Skip that
1776 * slash, and continue until no more slashes.
1780 } while (unlikely(*name
== '/'));
1784 err
= walk_component(nd
, LOOKUP_FOLLOW
);
1792 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1793 path_put_conditional(&nd
->link
, nd
);
1794 path_put(&nd
->path
);
1798 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1801 current
->link_count
++;
1805 if (unlikely(IS_ERR(s
))) {
1807 current
->link_count
--;
1815 current
->link_count
--;
1821 path_put(&nd
->path
);
1822 nd
->path
= nd
->root
;
1823 path_get(&nd
->root
);
1824 nd
->flags
|= LOOKUP_JUMPED
;
1825 while (unlikely(*++s
== '/'))
1828 nd
->inode
= nd
->path
.dentry
->d_inode
;
1829 nd
->stack
[nd
->depth
].name
= name
;
1836 if (!d_can_lookup(nd
->path
.dentry
)) {
1843 while (unlikely(nd
->depth
)) {
1845 current
->link_count
--;
1850 if (unlikely(nd
->depth
)) {
1851 name
= nd
->stack
[nd
->depth
].name
;
1852 err
= walk_component(nd
, LOOKUP_FOLLOW
);
1854 current
->link_count
--;
1861 static int path_init(int dfd
, const struct filename
*name
, unsigned int flags
,
1862 struct nameidata
*nd
)
1865 const char *s
= name
->name
;
1867 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1868 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1871 if (flags
& LOOKUP_ROOT
) {
1872 struct dentry
*root
= nd
->root
.dentry
;
1873 struct inode
*inode
= root
->d_inode
;
1875 if (!d_can_lookup(root
))
1877 retval
= inode_permission(inode
, MAY_EXEC
);
1881 nd
->path
= nd
->root
;
1883 if (flags
& LOOKUP_RCU
) {
1885 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1886 nd
->m_seq
= read_seqbegin(&mount_lock
);
1888 path_get(&nd
->path
);
1893 nd
->root
.mnt
= NULL
;
1895 nd
->m_seq
= read_seqbegin(&mount_lock
);
1897 if (flags
& LOOKUP_RCU
) {
1899 nd
->seq
= set_root_rcu(nd
);
1902 path_get(&nd
->root
);
1904 nd
->path
= nd
->root
;
1905 } else if (dfd
== AT_FDCWD
) {
1906 if (flags
& LOOKUP_RCU
) {
1907 struct fs_struct
*fs
= current
->fs
;
1913 seq
= read_seqcount_begin(&fs
->seq
);
1915 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1916 } while (read_seqcount_retry(&fs
->seq
, seq
));
1918 get_fs_pwd(current
->fs
, &nd
->path
);
1921 /* Caller must check execute permissions on the starting path component */
1922 struct fd f
= fdget_raw(dfd
);
1923 struct dentry
*dentry
;
1928 dentry
= f
.file
->f_path
.dentry
;
1931 if (!d_can_lookup(dentry
)) {
1937 nd
->path
= f
.file
->f_path
;
1938 if (flags
& LOOKUP_RCU
) {
1939 if (f
.flags
& FDPUT_FPUT
)
1941 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1944 path_get(&nd
->path
);
1949 nd
->inode
= nd
->path
.dentry
->d_inode
;
1950 if (!(flags
& LOOKUP_RCU
))
1952 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1954 if (!(nd
->flags
& LOOKUP_ROOT
))
1955 nd
->root
.mnt
= NULL
;
1959 current
->total_link_count
= 0;
1960 return link_path_walk(s
, nd
);
1963 static void path_cleanup(struct nameidata
*nd
)
1965 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1966 path_put(&nd
->root
);
1967 nd
->root
.mnt
= NULL
;
1969 if (unlikely(nd
->base
))
1973 static int trailing_symlink(struct nameidata
*nd
)
1976 int error
= may_follow_link(&nd
->link
, nd
);
1977 if (unlikely(error
))
1979 nd
->flags
|= LOOKUP_PARENT
;
1981 if (unlikely(IS_ERR(s
)))
1988 path_put(&nd
->path
);
1989 nd
->path
= nd
->root
;
1990 path_get(&nd
->root
);
1991 nd
->flags
|= LOOKUP_JUMPED
;
1993 nd
->inode
= nd
->path
.dentry
->d_inode
;
1994 error
= link_path_walk(s
, nd
);
1995 if (unlikely(error
))
2000 static inline int lookup_last(struct nameidata
*nd
)
2002 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2003 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2005 nd
->flags
&= ~LOOKUP_PARENT
;
2006 return walk_component(nd
, nd
->flags
& LOOKUP_FOLLOW
);
2009 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2010 static int path_lookupat(int dfd
, const struct filename
*name
,
2011 unsigned int flags
, struct nameidata
*nd
)
2016 * Path walking is largely split up into 2 different synchronisation
2017 * schemes, rcu-walk and ref-walk (explained in
2018 * Documentation/filesystems/path-lookup.txt). These share much of the
2019 * path walk code, but some things particularly setup, cleanup, and
2020 * following mounts are sufficiently divergent that functions are
2021 * duplicated. Typically there is a function foo(), and its RCU
2022 * analogue, foo_rcu().
2024 * -ECHILD is the error number of choice (just to avoid clashes) that
2025 * is returned if some aspect of an rcu-walk fails. Such an error must
2026 * be handled by restarting a traditional ref-walk (which will always
2027 * be able to complete).
2029 err
= path_init(dfd
, name
, flags
, nd
);
2030 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
2031 err
= lookup_last(nd
);
2033 err
= trailing_symlink(nd
);
2036 err
= lookup_last(nd
);
2042 err
= complete_walk(nd
);
2044 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2045 if (!d_can_lookup(nd
->path
.dentry
)) {
2046 path_put(&nd
->path
);
2055 static int filename_lookup(int dfd
, struct filename
*name
,
2056 unsigned int flags
, struct nameidata
*nd
)
2058 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2059 if (unlikely(retval
== -ECHILD
))
2060 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2061 if (unlikely(retval
== -ESTALE
))
2062 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2064 if (likely(!retval
))
2065 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2069 /* does lookup, returns the object with parent locked */
2070 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2072 struct filename
*filename
= getname_kernel(name
);
2073 struct nameidata nd
;
2077 if (IS_ERR(filename
))
2078 return ERR_CAST(filename
);
2080 err
= filename_lookup(AT_FDCWD
, filename
, LOOKUP_PARENT
, &nd
);
2085 if (nd
.last_type
!= LAST_NORM
) {
2087 d
= ERR_PTR(-EINVAL
);
2090 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2091 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2093 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2103 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2105 struct nameidata nd
;
2106 struct filename
*filename
= getname_kernel(name
);
2107 int res
= PTR_ERR(filename
);
2109 if (!IS_ERR(filename
)) {
2110 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2117 EXPORT_SYMBOL(kern_path
);
2120 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2121 * @dentry: pointer to dentry of the base directory
2122 * @mnt: pointer to vfs mount of the base directory
2123 * @name: pointer to file name
2124 * @flags: lookup flags
2125 * @path: pointer to struct path to fill
2127 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2128 const char *name
, unsigned int flags
,
2131 struct filename
*filename
= getname_kernel(name
);
2132 int err
= PTR_ERR(filename
);
2134 BUG_ON(flags
& LOOKUP_PARENT
);
2136 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2137 if (!IS_ERR(filename
)) {
2138 struct nameidata nd
;
2139 nd
.root
.dentry
= dentry
;
2141 err
= filename_lookup(AT_FDCWD
, filename
,
2142 flags
| LOOKUP_ROOT
, &nd
);
2149 EXPORT_SYMBOL(vfs_path_lookup
);
2152 * lookup_one_len - filesystem helper to lookup single pathname component
2153 * @name: pathname component to lookup
2154 * @base: base directory to lookup from
2155 * @len: maximum length @len should be interpreted to
2157 * Note that this routine is purely a helper for filesystem usage and should
2158 * not be called by generic code.
2160 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2166 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2170 this.hash
= full_name_hash(name
, len
);
2172 return ERR_PTR(-EACCES
);
2174 if (unlikely(name
[0] == '.')) {
2175 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2176 return ERR_PTR(-EACCES
);
2180 c
= *(const unsigned char *)name
++;
2181 if (c
== '/' || c
== '\0')
2182 return ERR_PTR(-EACCES
);
2185 * See if the low-level filesystem might want
2186 * to use its own hash..
2188 if (base
->d_flags
& DCACHE_OP_HASH
) {
2189 int err
= base
->d_op
->d_hash(base
, &this);
2191 return ERR_PTR(err
);
2194 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2196 return ERR_PTR(err
);
2198 return __lookup_hash(&this, base
, 0);
2200 EXPORT_SYMBOL(lookup_one_len
);
2202 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2203 struct path
*path
, int *empty
)
2205 struct nameidata nd
;
2206 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2207 int err
= PTR_ERR(tmp
);
2210 BUG_ON(flags
& LOOKUP_PARENT
);
2212 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2220 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2223 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2225 EXPORT_SYMBOL(user_path_at
);
2228 * NB: most callers don't do anything directly with the reference to the
2229 * to struct filename, but the nd->last pointer points into the name string
2230 * allocated by getname. So we must hold the reference to it until all
2231 * path-walking is complete.
2233 static struct filename
*
2234 user_path_parent(int dfd
, const char __user
*path
,
2235 struct path
*parent
,
2240 struct nameidata nd
;
2241 struct filename
*s
= getname(path
);
2244 /* only LOOKUP_REVAL is allowed in extra flags */
2245 flags
&= LOOKUP_REVAL
;
2250 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, &nd
);
2253 return ERR_PTR(error
);
2257 *type
= nd
.last_type
;
2263 * mountpoint_last - look up last component for umount
2264 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2265 * @path: pointer to container for result
2267 * This is a special lookup_last function just for umount. In this case, we
2268 * need to resolve the path without doing any revalidation.
2270 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2271 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2272 * in almost all cases, this lookup will be served out of the dcache. The only
2273 * cases where it won't are if nd->last refers to a symlink or the path is
2274 * bogus and it doesn't exist.
2277 * -error: if there was an error during lookup. This includes -ENOENT if the
2278 * lookup found a negative dentry. The nd->path reference will also be
2281 * 0: if we successfully resolved nd->path and found it to not to be a
2282 * symlink that needs to be followed. "path" will also be populated.
2283 * The nd->path reference will also be put.
2285 * 1: if we successfully resolved nd->last and found it to be a symlink
2286 * that needs to be followed. "path" will be populated with the path
2287 * to the link, and nd->path will *not* be put.
2290 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2293 struct dentry
*dentry
;
2294 struct dentry
*dir
= nd
->path
.dentry
;
2296 /* If we're in rcuwalk, drop out of it to handle last component */
2297 if (nd
->flags
& LOOKUP_RCU
) {
2298 if (unlazy_walk(nd
, NULL
)) {
2304 nd
->flags
&= ~LOOKUP_PARENT
;
2306 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2307 error
= handle_dots(nd
, nd
->last_type
);
2310 dentry
= dget(nd
->path
.dentry
);
2314 mutex_lock(&dir
->d_inode
->i_mutex
);
2315 dentry
= d_lookup(dir
, &nd
->last
);
2318 * No cached dentry. Mounted dentries are pinned in the cache,
2319 * so that means that this dentry is probably a symlink or the
2320 * path doesn't actually point to a mounted dentry.
2322 dentry
= d_alloc(dir
, &nd
->last
);
2325 mutex_unlock(&dir
->d_inode
->i_mutex
);
2328 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2329 error
= PTR_ERR(dentry
);
2330 if (IS_ERR(dentry
)) {
2331 mutex_unlock(&dir
->d_inode
->i_mutex
);
2335 mutex_unlock(&dir
->d_inode
->i_mutex
);
2338 if (d_is_negative(dentry
)) {
2343 path
->dentry
= dentry
;
2344 path
->mnt
= nd
->path
.mnt
;
2345 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
)) {
2358 * path_mountpoint - look up a path to be umounted
2359 * @dfd: directory file descriptor to start walk from
2360 * @name: full pathname to walk
2361 * @path: pointer to container for result
2362 * @flags: lookup flags
2364 * Look up the given name, but don't attempt to revalidate the last component.
2365 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2368 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2369 struct nameidata
*nd
, unsigned int flags
)
2371 int err
= path_init(dfd
, name
, flags
, nd
);
2375 err
= mountpoint_last(nd
, path
);
2377 err
= trailing_symlink(nd
);
2380 err
= mountpoint_last(nd
, path
);
2389 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2392 struct nameidata nd
;
2395 return PTR_ERR(name
);
2396 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_RCU
);
2397 if (unlikely(error
== -ECHILD
))
2398 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
);
2399 if (unlikely(error
== -ESTALE
))
2400 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_REVAL
);
2402 audit_inode(name
, path
->dentry
, 0);
2408 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2409 * @dfd: directory file descriptor
2410 * @name: pathname from userland
2411 * @flags: lookup flags
2412 * @path: pointer to container to hold result
2414 * A umount is a special case for path walking. We're not actually interested
2415 * in the inode in this situation, and ESTALE errors can be a problem. We
2416 * simply want track down the dentry and vfsmount attached at the mountpoint
2417 * and avoid revalidating the last component.
2419 * Returns 0 and populates "path" on success.
2422 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2425 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2429 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2432 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2434 EXPORT_SYMBOL(kern_path_mountpoint
);
2436 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2438 kuid_t fsuid
= current_fsuid();
2440 if (uid_eq(inode
->i_uid
, fsuid
))
2442 if (uid_eq(dir
->i_uid
, fsuid
))
2444 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2446 EXPORT_SYMBOL(__check_sticky
);
2449 * Check whether we can remove a link victim from directory dir, check
2450 * whether the type of victim is right.
2451 * 1. We can't do it if dir is read-only (done in permission())
2452 * 2. We should have write and exec permissions on dir
2453 * 3. We can't remove anything from append-only dir
2454 * 4. We can't do anything with immutable dir (done in permission())
2455 * 5. If the sticky bit on dir is set we should either
2456 * a. be owner of dir, or
2457 * b. be owner of victim, or
2458 * c. have CAP_FOWNER capability
2459 * 6. If the victim is append-only or immutable we can't do antyhing with
2460 * links pointing to it.
2461 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2462 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2463 * 9. We can't remove a root or mountpoint.
2464 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2465 * nfs_async_unlink().
2467 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2469 struct inode
*inode
= victim
->d_inode
;
2472 if (d_is_negative(victim
))
2476 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2477 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2479 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2485 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2486 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2489 if (!d_is_dir(victim
))
2491 if (IS_ROOT(victim
))
2493 } else if (d_is_dir(victim
))
2495 if (IS_DEADDIR(dir
))
2497 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2502 /* Check whether we can create an object with dentry child in directory
2504 * 1. We can't do it if child already exists (open has special treatment for
2505 * this case, but since we are inlined it's OK)
2506 * 2. We can't do it if dir is read-only (done in permission())
2507 * 3. We should have write and exec permissions on dir
2508 * 4. We can't do it if dir is immutable (done in permission())
2510 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2512 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2515 if (IS_DEADDIR(dir
))
2517 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2521 * p1 and p2 should be directories on the same fs.
2523 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2528 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2532 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2534 p
= d_ancestor(p2
, p1
);
2536 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2537 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2541 p
= d_ancestor(p1
, p2
);
2543 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2544 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2548 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2549 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2552 EXPORT_SYMBOL(lock_rename
);
2554 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2556 mutex_unlock(&p1
->d_inode
->i_mutex
);
2558 mutex_unlock(&p2
->d_inode
->i_mutex
);
2559 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2562 EXPORT_SYMBOL(unlock_rename
);
2564 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2567 int error
= may_create(dir
, dentry
);
2571 if (!dir
->i_op
->create
)
2572 return -EACCES
; /* shouldn't it be ENOSYS? */
2575 error
= security_inode_create(dir
, dentry
, mode
);
2578 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2580 fsnotify_create(dir
, dentry
);
2583 EXPORT_SYMBOL(vfs_create
);
2585 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2587 struct dentry
*dentry
= path
->dentry
;
2588 struct inode
*inode
= dentry
->d_inode
;
2598 switch (inode
->i_mode
& S_IFMT
) {
2602 if (acc_mode
& MAY_WRITE
)
2607 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2616 error
= inode_permission(inode
, acc_mode
);
2621 * An append-only file must be opened in append mode for writing.
2623 if (IS_APPEND(inode
)) {
2624 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2630 /* O_NOATIME can only be set by the owner or superuser */
2631 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2637 static int handle_truncate(struct file
*filp
)
2639 struct path
*path
= &filp
->f_path
;
2640 struct inode
*inode
= path
->dentry
->d_inode
;
2641 int error
= get_write_access(inode
);
2645 * Refuse to truncate files with mandatory locks held on them.
2647 error
= locks_verify_locked(filp
);
2649 error
= security_path_truncate(path
);
2651 error
= do_truncate(path
->dentry
, 0,
2652 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2655 put_write_access(inode
);
2659 static inline int open_to_namei_flags(int flag
)
2661 if ((flag
& O_ACCMODE
) == 3)
2666 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2668 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2672 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2676 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2680 * Attempt to atomically look up, create and open a file from a negative
2683 * Returns 0 if successful. The file will have been created and attached to
2684 * @file by the filesystem calling finish_open().
2686 * Returns 1 if the file was looked up only or didn't need creating. The
2687 * caller will need to perform the open themselves. @path will have been
2688 * updated to point to the new dentry. This may be negative.
2690 * Returns an error code otherwise.
2692 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2693 struct path
*path
, struct file
*file
,
2694 const struct open_flags
*op
,
2695 bool got_write
, bool need_lookup
,
2698 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2699 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2703 int create_error
= 0;
2704 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2707 BUG_ON(dentry
->d_inode
);
2709 /* Don't create child dentry for a dead directory. */
2710 if (unlikely(IS_DEADDIR(dir
))) {
2716 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2717 mode
&= ~current_umask();
2719 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2721 open_flag
&= ~O_TRUNC
;
2724 * Checking write permission is tricky, bacuse we don't know if we are
2725 * going to actually need it: O_CREAT opens should work as long as the
2726 * file exists. But checking existence breaks atomicity. The trick is
2727 * to check access and if not granted clear O_CREAT from the flags.
2729 * Another problem is returing the "right" error value (e.g. for an
2730 * O_EXCL open we want to return EEXIST not EROFS).
2732 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2733 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2734 if (!(open_flag
& O_CREAT
)) {
2736 * No O_CREATE -> atomicity not a requirement -> fall
2737 * back to lookup + open
2740 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2741 /* Fall back and fail with the right error */
2742 create_error
= -EROFS
;
2745 /* No side effects, safe to clear O_CREAT */
2746 create_error
= -EROFS
;
2747 open_flag
&= ~O_CREAT
;
2751 if (open_flag
& O_CREAT
) {
2752 error
= may_o_create(&nd
->path
, dentry
, mode
);
2754 create_error
= error
;
2755 if (open_flag
& O_EXCL
)
2757 open_flag
&= ~O_CREAT
;
2761 if (nd
->flags
& LOOKUP_DIRECTORY
)
2762 open_flag
|= O_DIRECTORY
;
2764 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2765 file
->f_path
.mnt
= nd
->path
.mnt
;
2766 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2769 if (create_error
&& error
== -ENOENT
)
2770 error
= create_error
;
2774 if (error
) { /* returned 1, that is */
2775 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2779 if (file
->f_path
.dentry
) {
2781 dentry
= file
->f_path
.dentry
;
2783 if (*opened
& FILE_CREATED
)
2784 fsnotify_create(dir
, dentry
);
2785 if (!dentry
->d_inode
) {
2786 WARN_ON(*opened
& FILE_CREATED
);
2788 error
= create_error
;
2792 if (excl
&& !(*opened
& FILE_CREATED
)) {
2801 * We didn't have the inode before the open, so check open permission
2804 acc_mode
= op
->acc_mode
;
2805 if (*opened
& FILE_CREATED
) {
2806 WARN_ON(!(open_flag
& O_CREAT
));
2807 fsnotify_create(dir
, dentry
);
2808 acc_mode
= MAY_OPEN
;
2810 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2820 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2822 return PTR_ERR(dentry
);
2825 int open_flag
= op
->open_flag
;
2827 error
= create_error
;
2828 if ((open_flag
& O_EXCL
)) {
2829 if (!dentry
->d_inode
)
2831 } else if (!dentry
->d_inode
) {
2833 } else if ((open_flag
& O_TRUNC
) &&
2837 /* will fail later, go on to get the right error */
2841 path
->dentry
= dentry
;
2842 path
->mnt
= nd
->path
.mnt
;
2847 * Look up and maybe create and open the last component.
2849 * Must be called with i_mutex held on parent.
2851 * Returns 0 if the file was successfully atomically created (if necessary) and
2852 * opened. In this case the file will be returned attached to @file.
2854 * Returns 1 if the file was not completely opened at this time, though lookups
2855 * and creations will have been performed and the dentry returned in @path will
2856 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2857 * specified then a negative dentry may be returned.
2859 * An error code is returned otherwise.
2861 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2862 * cleared otherwise prior to returning.
2864 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2866 const struct open_flags
*op
,
2867 bool got_write
, int *opened
)
2869 struct dentry
*dir
= nd
->path
.dentry
;
2870 struct inode
*dir_inode
= dir
->d_inode
;
2871 struct dentry
*dentry
;
2875 *opened
&= ~FILE_CREATED
;
2876 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2878 return PTR_ERR(dentry
);
2880 /* Cached positive dentry: will open in f_op->open */
2881 if (!need_lookup
&& dentry
->d_inode
)
2884 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2885 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2886 need_lookup
, opened
);
2890 BUG_ON(dentry
->d_inode
);
2892 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2894 return PTR_ERR(dentry
);
2897 /* Negative dentry, just create the file */
2898 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2899 umode_t mode
= op
->mode
;
2900 if (!IS_POSIXACL(dir
->d_inode
))
2901 mode
&= ~current_umask();
2903 * This write is needed to ensure that a
2904 * rw->ro transition does not occur between
2905 * the time when the file is created and when
2906 * a permanent write count is taken through
2907 * the 'struct file' in finish_open().
2913 *opened
|= FILE_CREATED
;
2914 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2917 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2918 nd
->flags
& LOOKUP_EXCL
);
2923 path
->dentry
= dentry
;
2924 path
->mnt
= nd
->path
.mnt
;
2933 * Handle the last step of open()
2935 static int do_last(struct nameidata
*nd
,
2936 struct file
*file
, const struct open_flags
*op
,
2937 int *opened
, struct filename
*name
)
2939 struct dentry
*dir
= nd
->path
.dentry
;
2940 int open_flag
= op
->open_flag
;
2941 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2942 bool got_write
= false;
2943 int acc_mode
= op
->acc_mode
;
2944 struct inode
*inode
;
2945 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2947 bool retried
= false;
2950 nd
->flags
&= ~LOOKUP_PARENT
;
2951 nd
->flags
|= op
->intent
;
2953 if (nd
->last_type
!= LAST_NORM
) {
2954 error
= handle_dots(nd
, nd
->last_type
);
2960 if (!(open_flag
& O_CREAT
)) {
2961 if (nd
->last
.name
[nd
->last
.len
])
2962 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2963 /* we _can_ be in RCU mode here */
2964 error
= lookup_fast(nd
, &path
, &inode
);
2971 BUG_ON(nd
->inode
!= dir
->d_inode
);
2973 /* create side of things */
2975 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2976 * has been cleared when we got to the last component we are
2979 error
= complete_walk(nd
);
2983 audit_inode(name
, dir
, LOOKUP_PARENT
);
2985 /* trailing slashes? */
2986 if (nd
->last
.name
[nd
->last
.len
])
2991 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2992 error
= mnt_want_write(nd
->path
.mnt
);
2996 * do _not_ fail yet - we might not need that or fail with
2997 * a different error; let lookup_open() decide; we'll be
2998 * dropping this one anyway.
3001 mutex_lock(&dir
->d_inode
->i_mutex
);
3002 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3003 mutex_unlock(&dir
->d_inode
->i_mutex
);
3009 if ((*opened
& FILE_CREATED
) ||
3010 !S_ISREG(file_inode(file
)->i_mode
))
3011 will_truncate
= false;
3013 audit_inode(name
, file
->f_path
.dentry
, 0);
3017 if (*opened
& FILE_CREATED
) {
3018 /* Don't check for write permission, don't truncate */
3019 open_flag
&= ~O_TRUNC
;
3020 will_truncate
= false;
3021 acc_mode
= MAY_OPEN
;
3022 path_to_nameidata(&path
, nd
);
3023 goto finish_open_created
;
3027 * create/update audit record if it already exists.
3029 if (d_is_positive(path
.dentry
))
3030 audit_inode(name
, path
.dentry
, 0);
3033 * If atomic_open() acquired write access it is dropped now due to
3034 * possible mount and symlink following (this might be optimized away if
3038 mnt_drop_write(nd
->path
.mnt
);
3043 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3046 error
= follow_managed(&path
, nd
->flags
);
3051 nd
->flags
|= LOOKUP_JUMPED
;
3053 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3054 inode
= path
.dentry
->d_inode
;
3056 if (d_is_negative(path
.dentry
)) {
3057 path_to_nameidata(&path
, nd
);
3061 if (should_follow_link(path
.dentry
, nd
->flags
& LOOKUP_FOLLOW
)) {
3062 if (nd
->flags
& LOOKUP_RCU
) {
3063 if (unlikely(nd
->path
.mnt
!= path
.mnt
||
3064 unlazy_walk(nd
, path
.dentry
))) {
3069 BUG_ON(inode
!= path
.dentry
->d_inode
);
3074 if (unlikely(d_is_symlink(path
.dentry
)) && !(open_flag
& O_PATH
)) {
3075 path_to_nameidata(&path
, nd
);
3080 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3081 path_to_nameidata(&path
, nd
);
3083 save_parent
.dentry
= nd
->path
.dentry
;
3084 save_parent
.mnt
= mntget(path
.mnt
);
3085 nd
->path
.dentry
= path
.dentry
;
3089 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3091 error
= complete_walk(nd
);
3093 path_put(&save_parent
);
3096 audit_inode(name
, nd
->path
.dentry
, 0);
3098 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3101 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3103 if (!d_is_reg(nd
->path
.dentry
))
3104 will_truncate
= false;
3106 if (will_truncate
) {
3107 error
= mnt_want_write(nd
->path
.mnt
);
3112 finish_open_created
:
3113 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3117 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3118 error
= vfs_open(&nd
->path
, file
, current_cred());
3120 *opened
|= FILE_OPENED
;
3122 if (error
== -EOPENSTALE
)
3127 error
= open_check_o_direct(file
);
3130 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3134 if (will_truncate
) {
3135 error
= handle_truncate(file
);
3141 mnt_drop_write(nd
->path
.mnt
);
3142 path_put(&save_parent
);
3147 path_put_conditional(&path
, nd
);
3154 /* If no saved parent or already retried then can't retry */
3155 if (!save_parent
.dentry
|| retried
)
3158 BUG_ON(save_parent
.dentry
!= dir
);
3159 path_put(&nd
->path
);
3160 nd
->path
= save_parent
;
3161 nd
->inode
= dir
->d_inode
;
3162 save_parent
.mnt
= NULL
;
3163 save_parent
.dentry
= NULL
;
3165 mnt_drop_write(nd
->path
.mnt
);
3172 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3173 struct nameidata
*nd
, int flags
,
3174 const struct open_flags
*op
,
3175 struct file
*file
, int *opened
)
3177 static const struct qstr name
= QSTR_INIT("/", 1);
3178 struct dentry
*dentry
, *child
;
3180 int error
= path_lookupat(dfd
, pathname
,
3181 flags
| LOOKUP_DIRECTORY
, nd
);
3182 if (unlikely(error
))
3184 error
= mnt_want_write(nd
->path
.mnt
);
3185 if (unlikely(error
))
3187 /* we want directory to be writable */
3188 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3191 dentry
= nd
->path
.dentry
;
3192 dir
= dentry
->d_inode
;
3193 if (!dir
->i_op
->tmpfile
) {
3194 error
= -EOPNOTSUPP
;
3197 child
= d_alloc(dentry
, &name
);
3198 if (unlikely(!child
)) {
3202 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3203 nd
->flags
|= op
->intent
;
3204 dput(nd
->path
.dentry
);
3205 nd
->path
.dentry
= child
;
3206 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3209 audit_inode(pathname
, nd
->path
.dentry
, 0);
3210 /* Don't check for other permissions, the inode was just created */
3211 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3214 file
->f_path
.mnt
= nd
->path
.mnt
;
3215 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3218 error
= open_check_o_direct(file
);
3221 } else if (!(op
->open_flag
& O_EXCL
)) {
3222 struct inode
*inode
= file_inode(file
);
3223 spin_lock(&inode
->i_lock
);
3224 inode
->i_state
|= I_LINKABLE
;
3225 spin_unlock(&inode
->i_lock
);
3228 mnt_drop_write(nd
->path
.mnt
);
3230 path_put(&nd
->path
);
3234 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3235 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3241 file
= get_empty_filp();
3245 file
->f_flags
= op
->open_flag
;
3247 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3248 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3252 error
= path_init(dfd
, pathname
, flags
, nd
);
3253 if (unlikely(error
))
3256 error
= do_last(nd
, file
, op
, &opened
, pathname
);
3257 while (unlikely(error
> 0)) { /* trailing symlink */
3258 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3259 error
= trailing_symlink(nd
);
3260 if (unlikely(error
))
3262 error
= do_last(nd
, file
, op
, &opened
, pathname
);
3268 if (!(opened
& FILE_OPENED
)) {
3272 if (unlikely(error
)) {
3273 if (error
== -EOPENSTALE
) {
3274 if (flags
& LOOKUP_RCU
)
3279 file
= ERR_PTR(error
);
3284 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3285 const struct open_flags
*op
)
3287 struct nameidata nd
;
3288 int flags
= op
->lookup_flags
;
3291 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3292 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3293 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3294 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3295 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3299 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3300 const char *name
, const struct open_flags
*op
)
3302 struct nameidata nd
;
3304 struct filename
*filename
;
3305 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3308 nd
.root
.dentry
= dentry
;
3310 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3311 return ERR_PTR(-ELOOP
);
3313 filename
= getname_kernel(name
);
3314 if (unlikely(IS_ERR(filename
)))
3315 return ERR_CAST(filename
);
3317 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3318 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3319 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3320 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3321 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3326 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3327 struct path
*path
, unsigned int lookup_flags
)
3329 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3330 struct nameidata nd
;
3333 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3336 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3337 * other flags passed in are ignored!
3339 lookup_flags
&= LOOKUP_REVAL
;
3341 error
= filename_lookup(dfd
, name
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3343 return ERR_PTR(error
);
3346 * Yucky last component or no last component at all?
3347 * (foo/., foo/.., /////)
3349 if (nd
.last_type
!= LAST_NORM
)
3351 nd
.flags
&= ~LOOKUP_PARENT
;
3352 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3354 /* don't fail immediately if it's r/o, at least try to report other errors */
3355 err2
= mnt_want_write(nd
.path
.mnt
);
3357 * Do the final lookup.
3359 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3360 dentry
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, nd
.flags
);
3365 if (d_is_positive(dentry
))
3369 * Special case - lookup gave negative, but... we had foo/bar/
3370 * From the vfs_mknod() POV we just have a negative dentry -
3371 * all is fine. Let's be bastards - you had / on the end, you've
3372 * been asking for (non-existent) directory. -ENOENT for you.
3374 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3378 if (unlikely(err2
)) {
3386 dentry
= ERR_PTR(error
);
3388 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3390 mnt_drop_write(nd
.path
.mnt
);
3396 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3397 struct path
*path
, unsigned int lookup_flags
)
3399 struct filename
*filename
= getname_kernel(pathname
);
3402 if (IS_ERR(filename
))
3403 return ERR_CAST(filename
);
3404 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3408 EXPORT_SYMBOL(kern_path_create
);
3410 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3413 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3414 mnt_drop_write(path
->mnt
);
3417 EXPORT_SYMBOL(done_path_create
);
3419 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3420 struct path
*path
, unsigned int lookup_flags
)
3422 struct filename
*tmp
= getname(pathname
);
3425 return ERR_CAST(tmp
);
3426 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3430 EXPORT_SYMBOL(user_path_create
);
3432 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3434 int error
= may_create(dir
, dentry
);
3439 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3442 if (!dir
->i_op
->mknod
)
3445 error
= devcgroup_inode_mknod(mode
, dev
);
3449 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3453 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3455 fsnotify_create(dir
, dentry
);
3458 EXPORT_SYMBOL(vfs_mknod
);
3460 static int may_mknod(umode_t mode
)
3462 switch (mode
& S_IFMT
) {
3468 case 0: /* zero mode translates to S_IFREG */
3477 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3480 struct dentry
*dentry
;
3483 unsigned int lookup_flags
= 0;
3485 error
= may_mknod(mode
);
3489 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3491 return PTR_ERR(dentry
);
3493 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3494 mode
&= ~current_umask();
3495 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3498 switch (mode
& S_IFMT
) {
3499 case 0: case S_IFREG
:
3500 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3502 case S_IFCHR
: case S_IFBLK
:
3503 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3504 new_decode_dev(dev
));
3506 case S_IFIFO
: case S_IFSOCK
:
3507 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3511 done_path_create(&path
, dentry
);
3512 if (retry_estale(error
, lookup_flags
)) {
3513 lookup_flags
|= LOOKUP_REVAL
;
3519 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3521 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3524 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3526 int error
= may_create(dir
, dentry
);
3527 unsigned max_links
= dir
->i_sb
->s_max_links
;
3532 if (!dir
->i_op
->mkdir
)
3535 mode
&= (S_IRWXUGO
|S_ISVTX
);
3536 error
= security_inode_mkdir(dir
, dentry
, mode
);
3540 if (max_links
&& dir
->i_nlink
>= max_links
)
3543 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3545 fsnotify_mkdir(dir
, dentry
);
3548 EXPORT_SYMBOL(vfs_mkdir
);
3550 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3552 struct dentry
*dentry
;
3555 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3558 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3560 return PTR_ERR(dentry
);
3562 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3563 mode
&= ~current_umask();
3564 error
= security_path_mkdir(&path
, dentry
, mode
);
3566 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3567 done_path_create(&path
, dentry
);
3568 if (retry_estale(error
, lookup_flags
)) {
3569 lookup_flags
|= LOOKUP_REVAL
;
3575 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3577 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3581 * The dentry_unhash() helper will try to drop the dentry early: we
3582 * should have a usage count of 1 if we're the only user of this
3583 * dentry, and if that is true (possibly after pruning the dcache),
3584 * then we drop the dentry now.
3586 * A low-level filesystem can, if it choses, legally
3589 * if (!d_unhashed(dentry))
3592 * if it cannot handle the case of removing a directory
3593 * that is still in use by something else..
3595 void dentry_unhash(struct dentry
*dentry
)
3597 shrink_dcache_parent(dentry
);
3598 spin_lock(&dentry
->d_lock
);
3599 if (dentry
->d_lockref
.count
== 1)
3601 spin_unlock(&dentry
->d_lock
);
3603 EXPORT_SYMBOL(dentry_unhash
);
3605 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3607 int error
= may_delete(dir
, dentry
, 1);
3612 if (!dir
->i_op
->rmdir
)
3616 mutex_lock(&dentry
->d_inode
->i_mutex
);
3619 if (is_local_mountpoint(dentry
))
3622 error
= security_inode_rmdir(dir
, dentry
);
3626 shrink_dcache_parent(dentry
);
3627 error
= dir
->i_op
->rmdir(dir
, dentry
);
3631 dentry
->d_inode
->i_flags
|= S_DEAD
;
3633 detach_mounts(dentry
);
3636 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3642 EXPORT_SYMBOL(vfs_rmdir
);
3644 static long do_rmdir(int dfd
, const char __user
*pathname
)
3647 struct filename
*name
;
3648 struct dentry
*dentry
;
3652 unsigned int lookup_flags
= 0;
3654 name
= user_path_parent(dfd
, pathname
,
3655 &path
, &last
, &type
, lookup_flags
);
3657 return PTR_ERR(name
);
3671 error
= mnt_want_write(path
.mnt
);
3675 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3676 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3677 error
= PTR_ERR(dentry
);
3680 if (!dentry
->d_inode
) {
3684 error
= security_path_rmdir(&path
, dentry
);
3687 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3691 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3692 mnt_drop_write(path
.mnt
);
3696 if (retry_estale(error
, lookup_flags
)) {
3697 lookup_flags
|= LOOKUP_REVAL
;
3703 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3705 return do_rmdir(AT_FDCWD
, pathname
);
3709 * vfs_unlink - unlink a filesystem object
3710 * @dir: parent directory
3712 * @delegated_inode: returns victim inode, if the inode is delegated.
3714 * The caller must hold dir->i_mutex.
3716 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3717 * return a reference to the inode in delegated_inode. The caller
3718 * should then break the delegation on that inode and retry. Because
3719 * breaking a delegation may take a long time, the caller should drop
3720 * dir->i_mutex before doing so.
3722 * Alternatively, a caller may pass NULL for delegated_inode. This may
3723 * be appropriate for callers that expect the underlying filesystem not
3724 * to be NFS exported.
3726 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3728 struct inode
*target
= dentry
->d_inode
;
3729 int error
= may_delete(dir
, dentry
, 0);
3734 if (!dir
->i_op
->unlink
)
3737 mutex_lock(&target
->i_mutex
);
3738 if (is_local_mountpoint(dentry
))
3741 error
= security_inode_unlink(dir
, dentry
);
3743 error
= try_break_deleg(target
, delegated_inode
);
3746 error
= dir
->i_op
->unlink(dir
, dentry
);
3749 detach_mounts(dentry
);
3754 mutex_unlock(&target
->i_mutex
);
3756 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3757 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3758 fsnotify_link_count(target
);
3764 EXPORT_SYMBOL(vfs_unlink
);
3767 * Make sure that the actual truncation of the file will occur outside its
3768 * directory's i_mutex. Truncate can take a long time if there is a lot of
3769 * writeout happening, and we don't want to prevent access to the directory
3770 * while waiting on the I/O.
3772 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3775 struct filename
*name
;
3776 struct dentry
*dentry
;
3780 struct inode
*inode
= NULL
;
3781 struct inode
*delegated_inode
= NULL
;
3782 unsigned int lookup_flags
= 0;
3784 name
= user_path_parent(dfd
, pathname
,
3785 &path
, &last
, &type
, lookup_flags
);
3787 return PTR_ERR(name
);
3790 if (type
!= LAST_NORM
)
3793 error
= mnt_want_write(path
.mnt
);
3797 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3798 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3799 error
= PTR_ERR(dentry
);
3800 if (!IS_ERR(dentry
)) {
3801 /* Why not before? Because we want correct error value */
3802 if (last
.name
[last
.len
])
3804 inode
= dentry
->d_inode
;
3805 if (d_is_negative(dentry
))
3808 error
= security_path_unlink(&path
, dentry
);
3811 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3815 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3817 iput(inode
); /* truncate the inode here */
3819 if (delegated_inode
) {
3820 error
= break_deleg_wait(&delegated_inode
);
3824 mnt_drop_write(path
.mnt
);
3828 if (retry_estale(error
, lookup_flags
)) {
3829 lookup_flags
|= LOOKUP_REVAL
;
3836 if (d_is_negative(dentry
))
3838 else if (d_is_dir(dentry
))
3845 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3847 if ((flag
& ~AT_REMOVEDIR
) != 0)
3850 if (flag
& AT_REMOVEDIR
)
3851 return do_rmdir(dfd
, pathname
);
3853 return do_unlinkat(dfd
, pathname
);
3856 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3858 return do_unlinkat(AT_FDCWD
, pathname
);
3861 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3863 int error
= may_create(dir
, dentry
);
3868 if (!dir
->i_op
->symlink
)
3871 error
= security_inode_symlink(dir
, dentry
, oldname
);
3875 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3877 fsnotify_create(dir
, dentry
);
3880 EXPORT_SYMBOL(vfs_symlink
);
3882 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3883 int, newdfd
, const char __user
*, newname
)
3886 struct filename
*from
;
3887 struct dentry
*dentry
;
3889 unsigned int lookup_flags
= 0;
3891 from
= getname(oldname
);
3893 return PTR_ERR(from
);
3895 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3896 error
= PTR_ERR(dentry
);
3900 error
= security_path_symlink(&path
, dentry
, from
->name
);
3902 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3903 done_path_create(&path
, dentry
);
3904 if (retry_estale(error
, lookup_flags
)) {
3905 lookup_flags
|= LOOKUP_REVAL
;
3913 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3915 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3919 * vfs_link - create a new link
3920 * @old_dentry: object to be linked
3922 * @new_dentry: where to create the new link
3923 * @delegated_inode: returns inode needing a delegation break
3925 * The caller must hold dir->i_mutex
3927 * If vfs_link discovers a delegation on the to-be-linked file in need
3928 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3929 * inode in delegated_inode. The caller should then break the delegation
3930 * and retry. Because breaking a delegation may take a long time, the
3931 * caller should drop the i_mutex before doing so.
3933 * Alternatively, a caller may pass NULL for delegated_inode. This may
3934 * be appropriate for callers that expect the underlying filesystem not
3935 * to be NFS exported.
3937 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3939 struct inode
*inode
= old_dentry
->d_inode
;
3940 unsigned max_links
= dir
->i_sb
->s_max_links
;
3946 error
= may_create(dir
, new_dentry
);
3950 if (dir
->i_sb
!= inode
->i_sb
)
3954 * A link to an append-only or immutable file cannot be created.
3956 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3958 if (!dir
->i_op
->link
)
3960 if (S_ISDIR(inode
->i_mode
))
3963 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3967 mutex_lock(&inode
->i_mutex
);
3968 /* Make sure we don't allow creating hardlink to an unlinked file */
3969 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3971 else if (max_links
&& inode
->i_nlink
>= max_links
)
3974 error
= try_break_deleg(inode
, delegated_inode
);
3976 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3979 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3980 spin_lock(&inode
->i_lock
);
3981 inode
->i_state
&= ~I_LINKABLE
;
3982 spin_unlock(&inode
->i_lock
);
3984 mutex_unlock(&inode
->i_mutex
);
3986 fsnotify_link(dir
, inode
, new_dentry
);
3989 EXPORT_SYMBOL(vfs_link
);
3992 * Hardlinks are often used in delicate situations. We avoid
3993 * security-related surprises by not following symlinks on the
3996 * We don't follow them on the oldname either to be compatible
3997 * with linux 2.0, and to avoid hard-linking to directories
3998 * and other special files. --ADM
4000 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4001 int, newdfd
, const char __user
*, newname
, int, flags
)
4003 struct dentry
*new_dentry
;
4004 struct path old_path
, new_path
;
4005 struct inode
*delegated_inode
= NULL
;
4009 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4012 * To use null names we require CAP_DAC_READ_SEARCH
4013 * This ensures that not everyone will be able to create
4014 * handlink using the passed filedescriptor.
4016 if (flags
& AT_EMPTY_PATH
) {
4017 if (!capable(CAP_DAC_READ_SEARCH
))
4022 if (flags
& AT_SYMLINK_FOLLOW
)
4023 how
|= LOOKUP_FOLLOW
;
4025 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4029 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4030 (how
& LOOKUP_REVAL
));
4031 error
= PTR_ERR(new_dentry
);
4032 if (IS_ERR(new_dentry
))
4036 if (old_path
.mnt
!= new_path
.mnt
)
4038 error
= may_linkat(&old_path
);
4039 if (unlikely(error
))
4041 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4044 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4046 done_path_create(&new_path
, new_dentry
);
4047 if (delegated_inode
) {
4048 error
= break_deleg_wait(&delegated_inode
);
4050 path_put(&old_path
);
4054 if (retry_estale(error
, how
)) {
4055 path_put(&old_path
);
4056 how
|= LOOKUP_REVAL
;
4060 path_put(&old_path
);
4065 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4067 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4071 * vfs_rename - rename a filesystem object
4072 * @old_dir: parent of source
4073 * @old_dentry: source
4074 * @new_dir: parent of destination
4075 * @new_dentry: destination
4076 * @delegated_inode: returns an inode needing a delegation break
4077 * @flags: rename flags
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 * The worst of all namespace operations - renaming directory. "Perverted"
4093 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4095 * a) we can get into loop creation.
4096 * b) race potential - two innocent renames can create a loop together.
4097 * That's where 4.4 screws up. Current fix: serialization on
4098 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4100 * c) we have to lock _four_ objects - parents and victim (if it exists),
4101 * and source (if it is not a directory).
4102 * And that - after we got ->i_mutex on parents (until then we don't know
4103 * whether the target exists). Solution: try to be smart with locking
4104 * order for inodes. We rely on the fact that tree topology may change
4105 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4106 * move will be locked. Thus we can rank directories by the tree
4107 * (ancestors first) and rank all non-directories after them.
4108 * That works since everybody except rename does "lock parent, lookup,
4109 * lock child" and rename is under ->s_vfs_rename_mutex.
4110 * HOWEVER, it relies on the assumption that any object with ->lookup()
4111 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4112 * we'd better make sure that there's no link(2) for them.
4113 * d) conversion from fhandle to dentry may come in the wrong moment - when
4114 * we are removing the target. Solution: we will have to grab ->i_mutex
4115 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4116 * ->i_mutex on parents, which works but leads to some truly excessive
4119 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4120 struct inode
*new_dir
, struct dentry
*new_dentry
,
4121 struct inode
**delegated_inode
, unsigned int flags
)
4124 bool is_dir
= d_is_dir(old_dentry
);
4125 const unsigned char *old_name
;
4126 struct inode
*source
= old_dentry
->d_inode
;
4127 struct inode
*target
= new_dentry
->d_inode
;
4128 bool new_is_dir
= false;
4129 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4131 if (source
== target
)
4134 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4139 error
= may_create(new_dir
, new_dentry
);
4141 new_is_dir
= d_is_dir(new_dentry
);
4143 if (!(flags
& RENAME_EXCHANGE
))
4144 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4146 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4151 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4154 if (flags
&& !old_dir
->i_op
->rename2
)
4158 * If we are going to change the parent - check write permissions,
4159 * we'll need to flip '..'.
4161 if (new_dir
!= old_dir
) {
4163 error
= inode_permission(source
, MAY_WRITE
);
4167 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4168 error
= inode_permission(target
, MAY_WRITE
);
4174 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4179 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4181 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4182 lock_two_nondirectories(source
, target
);
4184 mutex_lock(&target
->i_mutex
);
4187 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4190 if (max_links
&& new_dir
!= old_dir
) {
4192 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4194 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4195 old_dir
->i_nlink
>= max_links
)
4198 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4199 shrink_dcache_parent(new_dentry
);
4201 error
= try_break_deleg(source
, delegated_inode
);
4205 if (target
&& !new_is_dir
) {
4206 error
= try_break_deleg(target
, delegated_inode
);
4210 if (!old_dir
->i_op
->rename2
) {
4211 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4212 new_dir
, new_dentry
);
4214 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4215 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4216 new_dir
, new_dentry
, flags
);
4221 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4223 target
->i_flags
|= S_DEAD
;
4224 dont_mount(new_dentry
);
4225 detach_mounts(new_dentry
);
4227 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4228 if (!(flags
& RENAME_EXCHANGE
))
4229 d_move(old_dentry
, new_dentry
);
4231 d_exchange(old_dentry
, new_dentry
);
4234 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4235 unlock_two_nondirectories(source
, target
);
4237 mutex_unlock(&target
->i_mutex
);
4240 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4241 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4242 if (flags
& RENAME_EXCHANGE
) {
4243 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4244 new_is_dir
, NULL
, new_dentry
);
4247 fsnotify_oldname_free(old_name
);
4251 EXPORT_SYMBOL(vfs_rename
);
4253 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4254 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4256 struct dentry
*old_dentry
, *new_dentry
;
4257 struct dentry
*trap
;
4258 struct path old_path
, new_path
;
4259 struct qstr old_last
, new_last
;
4260 int old_type
, new_type
;
4261 struct inode
*delegated_inode
= NULL
;
4262 struct filename
*from
;
4263 struct filename
*to
;
4264 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4265 bool should_retry
= false;
4268 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4271 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4272 (flags
& RENAME_EXCHANGE
))
4275 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4278 if (flags
& RENAME_EXCHANGE
)
4282 from
= user_path_parent(olddfd
, oldname
,
4283 &old_path
, &old_last
, &old_type
, lookup_flags
);
4285 error
= PTR_ERR(from
);
4289 to
= user_path_parent(newdfd
, newname
,
4290 &new_path
, &new_last
, &new_type
, lookup_flags
);
4292 error
= PTR_ERR(to
);
4297 if (old_path
.mnt
!= new_path
.mnt
)
4301 if (old_type
!= LAST_NORM
)
4304 if (flags
& RENAME_NOREPLACE
)
4306 if (new_type
!= LAST_NORM
)
4309 error
= mnt_want_write(old_path
.mnt
);
4314 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4316 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4317 error
= PTR_ERR(old_dentry
);
4318 if (IS_ERR(old_dentry
))
4320 /* source must exist */
4322 if (d_is_negative(old_dentry
))
4324 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4325 error
= PTR_ERR(new_dentry
);
4326 if (IS_ERR(new_dentry
))
4329 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4331 if (flags
& RENAME_EXCHANGE
) {
4333 if (d_is_negative(new_dentry
))
4336 if (!d_is_dir(new_dentry
)) {
4338 if (new_last
.name
[new_last
.len
])
4342 /* unless the source is a directory trailing slashes give -ENOTDIR */
4343 if (!d_is_dir(old_dentry
)) {
4345 if (old_last
.name
[old_last
.len
])
4347 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4350 /* source should not be ancestor of target */
4352 if (old_dentry
== trap
)
4354 /* target should not be an ancestor of source */
4355 if (!(flags
& RENAME_EXCHANGE
))
4357 if (new_dentry
== trap
)
4360 error
= security_path_rename(&old_path
, old_dentry
,
4361 &new_path
, new_dentry
, flags
);
4364 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4365 new_path
.dentry
->d_inode
, new_dentry
,
4366 &delegated_inode
, flags
);
4372 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4373 if (delegated_inode
) {
4374 error
= break_deleg_wait(&delegated_inode
);
4378 mnt_drop_write(old_path
.mnt
);
4380 if (retry_estale(error
, lookup_flags
))
4381 should_retry
= true;
4382 path_put(&new_path
);
4385 path_put(&old_path
);
4388 should_retry
= false;
4389 lookup_flags
|= LOOKUP_REVAL
;
4396 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4397 int, newdfd
, const char __user
*, newname
)
4399 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4402 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4404 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4407 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4409 int error
= may_create(dir
, dentry
);
4413 if (!dir
->i_op
->mknod
)
4416 return dir
->i_op
->mknod(dir
, dentry
,
4417 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4419 EXPORT_SYMBOL(vfs_whiteout
);
4421 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4423 int len
= PTR_ERR(link
);
4428 if (len
> (unsigned) buflen
)
4430 if (copy_to_user(buffer
, link
, len
))
4435 EXPORT_SYMBOL(readlink_copy
);
4438 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4439 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4440 * using) it for any given inode is up to filesystem.
4442 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4445 const char *link
= dentry
->d_inode
->i_link
;
4449 link
= dentry
->d_inode
->i_op
->follow_link(dentry
, &cookie
, NULL
);
4451 return PTR_ERR(link
);
4453 res
= readlink_copy(buffer
, buflen
, link
);
4454 if (cookie
&& dentry
->d_inode
->i_op
->put_link
)
4455 dentry
->d_inode
->i_op
->put_link(dentry
, cookie
);
4458 EXPORT_SYMBOL(generic_readlink
);
4460 /* get the link contents into pagecache */
4461 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4465 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4466 page
= read_mapping_page(mapping
, 0, NULL
);
4471 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4475 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4477 struct page
*page
= NULL
;
4478 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4481 page_cache_release(page
);
4485 EXPORT_SYMBOL(page_readlink
);
4487 const char *page_follow_link_light(struct dentry
*dentry
, void **cookie
, struct nameidata
*nd
)
4489 struct page
*page
= NULL
;
4490 char *res
= page_getlink(dentry
, &page
);
4495 EXPORT_SYMBOL(page_follow_link_light
);
4497 void page_put_link(struct dentry
*dentry
, void *cookie
)
4499 struct page
*page
= cookie
;
4501 page_cache_release(page
);
4503 EXPORT_SYMBOL(page_put_link
);
4506 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4508 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4510 struct address_space
*mapping
= inode
->i_mapping
;
4515 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4517 flags
|= AOP_FLAG_NOFS
;
4520 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4521 flags
, &page
, &fsdata
);
4525 kaddr
= kmap_atomic(page
);
4526 memcpy(kaddr
, symname
, len
-1);
4527 kunmap_atomic(kaddr
);
4529 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4536 mark_inode_dirty(inode
);
4541 EXPORT_SYMBOL(__page_symlink
);
4543 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4545 return __page_symlink(inode
, symname
, len
,
4546 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4548 EXPORT_SYMBOL(page_symlink
);
4550 const struct inode_operations page_symlink_inode_operations
= {
4551 .readlink
= generic_readlink
,
4552 .follow_link
= page_follow_link_light
,
4553 .put_link
= page_put_link
,
4555 EXPORT_SYMBOL(page_symlink_inode_operations
);