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
);
499 struct inode
*inode
; /* path.dentry.d_inode */
508 * Path walking has 2 modes, rcu-walk and ref-walk (see
509 * Documentation/filesystems/path-lookup.txt). In situations when we can't
510 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
511 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
512 * mode. Refcounts are grabbed at the last known good point before rcu-walk
513 * got stuck, so ref-walk may continue from there. If this is not successful
514 * (eg. a seqcount has changed), then failure is returned and it's up to caller
515 * to restart the path walk from the beginning in ref-walk mode.
519 * unlazy_walk - try to switch to ref-walk mode.
520 * @nd: nameidata pathwalk data
521 * @dentry: child of nd->path.dentry or NULL
522 * Returns: 0 on success, -ECHILD on failure
524 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
525 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
526 * @nd or NULL. Must be called from rcu-walk context.
528 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
530 struct fs_struct
*fs
= current
->fs
;
531 struct dentry
*parent
= nd
->path
.dentry
;
533 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
536 * After legitimizing the bastards, terminate_walk()
537 * will do the right thing for non-RCU mode, and all our
538 * subsequent exit cases should rcu_read_unlock()
539 * before returning. Do vfsmount first; if dentry
540 * can't be legitimized, just set nd->path.dentry to NULL
541 * and rely on dput(NULL) being a no-op.
543 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
545 nd
->flags
&= ~LOOKUP_RCU
;
547 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
548 nd
->path
.dentry
= NULL
;
553 * For a negative lookup, the lookup sequence point is the parents
554 * sequence point, and it only needs to revalidate the parent dentry.
556 * For a positive lookup, we need to move both the parent and the
557 * dentry from the RCU domain to be properly refcounted. And the
558 * sequence number in the dentry validates *both* dentry counters,
559 * since we checked the sequence number of the parent after we got
560 * the child sequence number. So we know the parent must still
561 * be valid if the child sequence number is still valid.
564 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
566 BUG_ON(nd
->inode
!= parent
->d_inode
);
568 if (!lockref_get_not_dead(&dentry
->d_lockref
))
570 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
575 * Sequence counts matched. Now make sure that the root is
576 * still valid and get it if required.
578 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
579 spin_lock(&fs
->lock
);
580 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
581 goto unlock_and_drop_dentry
;
583 spin_unlock(&fs
->lock
);
589 unlock_and_drop_dentry
:
590 spin_unlock(&fs
->lock
);
598 if (!(nd
->flags
& LOOKUP_ROOT
))
603 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
605 return dentry
->d_op
->d_revalidate(dentry
, flags
);
609 * complete_walk - successful completion of path walk
610 * @nd: pointer nameidata
612 * If we had been in RCU mode, drop out of it and legitimize nd->path.
613 * Revalidate the final result, unless we'd already done that during
614 * the path walk or the filesystem doesn't ask for it. Return 0 on
615 * success, -error on failure. In case of failure caller does not
616 * need to drop nd->path.
618 static int complete_walk(struct nameidata
*nd
)
620 struct dentry
*dentry
= nd
->path
.dentry
;
623 if (nd
->flags
& LOOKUP_RCU
) {
624 nd
->flags
&= ~LOOKUP_RCU
;
625 if (!(nd
->flags
& LOOKUP_ROOT
))
628 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
632 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
634 mntput(nd
->path
.mnt
);
637 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
640 mntput(nd
->path
.mnt
);
646 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
649 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
652 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
663 static __always_inline
void set_root(struct nameidata
*nd
)
665 get_fs_root(current
->fs
, &nd
->root
);
668 static int link_path_walk(const char *, struct nameidata
*);
670 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
672 struct fs_struct
*fs
= current
->fs
;
676 seq
= read_seqcount_begin(&fs
->seq
);
678 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
679 } while (read_seqcount_retry(&fs
->seq
, seq
));
683 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
686 if (path
->mnt
!= nd
->path
.mnt
)
690 static inline void path_to_nameidata(const struct path
*path
,
691 struct nameidata
*nd
)
693 if (!(nd
->flags
& LOOKUP_RCU
)) {
694 dput(nd
->path
.dentry
);
695 if (nd
->path
.mnt
!= path
->mnt
)
696 mntput(nd
->path
.mnt
);
698 nd
->path
.mnt
= path
->mnt
;
699 nd
->path
.dentry
= path
->dentry
;
703 * Helper to directly jump to a known parsed path from ->follow_link,
704 * caller must have taken a reference to path beforehand.
706 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
711 nd
->inode
= nd
->path
.dentry
->d_inode
;
712 nd
->flags
|= LOOKUP_JUMPED
;
715 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
717 struct inode
*inode
= link
->dentry
->d_inode
;
718 if (cookie
&& inode
->i_op
->put_link
)
719 inode
->i_op
->put_link(link
->dentry
, cookie
);
723 int sysctl_protected_symlinks __read_mostly
= 0;
724 int sysctl_protected_hardlinks __read_mostly
= 0;
727 * may_follow_link - Check symlink following for unsafe situations
728 * @link: The path of the symlink
729 * @nd: nameidata pathwalk data
731 * In the case of the sysctl_protected_symlinks sysctl being enabled,
732 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
733 * in a sticky world-writable directory. This is to protect privileged
734 * processes from failing races against path names that may change out
735 * from under them by way of other users creating malicious symlinks.
736 * It will permit symlinks to be followed only when outside a sticky
737 * world-writable directory, or when the uid of the symlink and follower
738 * match, or when the directory owner matches the symlink's owner.
740 * Returns 0 if following the symlink is allowed, -ve on error.
742 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
744 const struct inode
*inode
;
745 const struct inode
*parent
;
747 if (!sysctl_protected_symlinks
)
750 /* Allowed if owner and follower match. */
751 inode
= link
->dentry
->d_inode
;
752 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
755 /* Allowed if parent directory not sticky and world-writable. */
756 parent
= nd
->path
.dentry
->d_inode
;
757 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
760 /* Allowed if parent directory and link owner match. */
761 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
764 audit_log_link_denied("follow_link", link
);
765 path_put_conditional(link
, nd
);
771 * safe_hardlink_source - Check for safe hardlink conditions
772 * @inode: the source inode to hardlink from
774 * Return false if at least one of the following conditions:
775 * - inode is not a regular file
777 * - inode is setgid and group-exec
778 * - access failure for read and write
780 * Otherwise returns true.
782 static bool safe_hardlink_source(struct inode
*inode
)
784 umode_t mode
= inode
->i_mode
;
786 /* Special files should not get pinned to the filesystem. */
790 /* Setuid files should not get pinned to the filesystem. */
794 /* Executable setgid files should not get pinned to the filesystem. */
795 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
798 /* Hardlinking to unreadable or unwritable sources is dangerous. */
799 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
806 * may_linkat - Check permissions for creating a hardlink
807 * @link: the source to hardlink from
809 * Block hardlink when all of:
810 * - sysctl_protected_hardlinks enabled
811 * - fsuid does not match inode
812 * - hardlink source is unsafe (see safe_hardlink_source() above)
815 * Returns 0 if successful, -ve on error.
817 static int may_linkat(struct path
*link
)
819 const struct cred
*cred
;
822 if (!sysctl_protected_hardlinks
)
825 cred
= current_cred();
826 inode
= link
->dentry
->d_inode
;
828 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
829 * otherwise, it must be a safe source.
831 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
835 audit_log_link_denied("linkat", link
);
839 static __always_inline
const char *
840 get_link(struct path
*link
, struct nameidata
*nd
, void **p
)
842 struct dentry
*dentry
= link
->dentry
;
843 struct inode
*inode
= dentry
->d_inode
;
847 BUG_ON(nd
->flags
& LOOKUP_RCU
);
849 if (link
->mnt
== nd
->path
.mnt
)
852 res
= ERR_PTR(-ELOOP
);
853 if (unlikely(current
->total_link_count
>= 40))
857 current
->total_link_count
++;
861 error
= security_inode_follow_link(dentry
);
862 res
= ERR_PTR(error
);
866 nd
->last_type
= LAST_BIND
;
868 res
= inode
->i_op
->follow_link(dentry
, p
, nd
);
877 static __always_inline
int
878 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
880 const char *s
= get_link(link
, nd
, p
);
883 if (unlikely(IS_ERR(s
)))
893 nd
->flags
|= LOOKUP_JUMPED
;
895 nd
->inode
= nd
->path
.dentry
->d_inode
;
896 error
= link_path_walk(s
, nd
);
898 put_link(nd
, link
, *p
);
902 static int follow_up_rcu(struct path
*path
)
904 struct mount
*mnt
= real_mount(path
->mnt
);
905 struct mount
*parent
;
906 struct dentry
*mountpoint
;
908 parent
= mnt
->mnt_parent
;
909 if (&parent
->mnt
== path
->mnt
)
911 mountpoint
= mnt
->mnt_mountpoint
;
912 path
->dentry
= mountpoint
;
913 path
->mnt
= &parent
->mnt
;
918 * follow_up - Find the mountpoint of path's vfsmount
920 * Given a path, find the mountpoint of its source file system.
921 * Replace @path with the path of the mountpoint in the parent mount.
924 * Return 1 if we went up a level and 0 if we were already at the
927 int follow_up(struct path
*path
)
929 struct mount
*mnt
= real_mount(path
->mnt
);
930 struct mount
*parent
;
931 struct dentry
*mountpoint
;
933 read_seqlock_excl(&mount_lock
);
934 parent
= mnt
->mnt_parent
;
936 read_sequnlock_excl(&mount_lock
);
939 mntget(&parent
->mnt
);
940 mountpoint
= dget(mnt
->mnt_mountpoint
);
941 read_sequnlock_excl(&mount_lock
);
943 path
->dentry
= mountpoint
;
945 path
->mnt
= &parent
->mnt
;
948 EXPORT_SYMBOL(follow_up
);
951 * Perform an automount
952 * - return -EISDIR to tell follow_managed() to stop and return the path we
955 static int follow_automount(struct path
*path
, unsigned flags
,
958 struct vfsmount
*mnt
;
961 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
964 /* We don't want to mount if someone's just doing a stat -
965 * unless they're stat'ing a directory and appended a '/' to
968 * We do, however, want to mount if someone wants to open or
969 * create a file of any type under the mountpoint, wants to
970 * traverse through the mountpoint or wants to open the
971 * mounted directory. Also, autofs may mark negative dentries
972 * as being automount points. These will need the attentions
973 * of the daemon to instantiate them before they can be used.
975 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
976 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
977 path
->dentry
->d_inode
)
980 current
->total_link_count
++;
981 if (current
->total_link_count
>= 40)
984 mnt
= path
->dentry
->d_op
->d_automount(path
);
987 * The filesystem is allowed to return -EISDIR here to indicate
988 * it doesn't want to automount. For instance, autofs would do
989 * this so that its userspace daemon can mount on this dentry.
991 * However, we can only permit this if it's a terminal point in
992 * the path being looked up; if it wasn't then the remainder of
993 * the path is inaccessible and we should say so.
995 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1000 if (!mnt
) /* mount collision */
1003 if (!*need_mntput
) {
1004 /* lock_mount() may release path->mnt on error */
1006 *need_mntput
= true;
1008 err
= finish_automount(mnt
, path
);
1012 /* Someone else made a mount here whilst we were busy */
1017 path
->dentry
= dget(mnt
->mnt_root
);
1026 * Handle a dentry that is managed in some way.
1027 * - Flagged for transit management (autofs)
1028 * - Flagged as mountpoint
1029 * - Flagged as automount point
1031 * This may only be called in refwalk mode.
1033 * Serialization is taken care of in namespace.c
1035 static int follow_managed(struct path
*path
, unsigned flags
)
1037 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1039 bool need_mntput
= false;
1042 /* Given that we're not holding a lock here, we retain the value in a
1043 * local variable for each dentry as we look at it so that we don't see
1044 * the components of that value change under us */
1045 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1046 managed
&= DCACHE_MANAGED_DENTRY
,
1047 unlikely(managed
!= 0)) {
1048 /* Allow the filesystem to manage the transit without i_mutex
1050 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1051 BUG_ON(!path
->dentry
->d_op
);
1052 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1053 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1058 /* Transit to a mounted filesystem. */
1059 if (managed
& DCACHE_MOUNTED
) {
1060 struct vfsmount
*mounted
= lookup_mnt(path
);
1065 path
->mnt
= mounted
;
1066 path
->dentry
= dget(mounted
->mnt_root
);
1071 /* Something is mounted on this dentry in another
1072 * namespace and/or whatever was mounted there in this
1073 * namespace got unmounted before lookup_mnt() could
1077 /* Handle an automount point */
1078 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1079 ret
= follow_automount(path
, flags
, &need_mntput
);
1085 /* We didn't change the current path point */
1089 if (need_mntput
&& path
->mnt
== mnt
)
1093 return ret
< 0 ? ret
: need_mntput
;
1096 int follow_down_one(struct path
*path
)
1098 struct vfsmount
*mounted
;
1100 mounted
= lookup_mnt(path
);
1104 path
->mnt
= mounted
;
1105 path
->dentry
= dget(mounted
->mnt_root
);
1110 EXPORT_SYMBOL(follow_down_one
);
1112 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1114 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1115 dentry
->d_op
->d_manage(dentry
, true) : 0;
1119 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1120 * we meet a managed dentry that would need blocking.
1122 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1123 struct inode
**inode
)
1126 struct mount
*mounted
;
1128 * Don't forget we might have a non-mountpoint managed dentry
1129 * that wants to block transit.
1131 switch (managed_dentry_rcu(path
->dentry
)) {
1141 if (!d_mountpoint(path
->dentry
))
1142 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1144 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1147 path
->mnt
= &mounted
->mnt
;
1148 path
->dentry
= mounted
->mnt
.mnt_root
;
1149 nd
->flags
|= LOOKUP_JUMPED
;
1150 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1152 * Update the inode too. We don't need to re-check the
1153 * dentry sequence number here after this d_inode read,
1154 * because a mount-point is always pinned.
1156 *inode
= path
->dentry
->d_inode
;
1158 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1159 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1162 static int follow_dotdot_rcu(struct nameidata
*nd
)
1164 struct inode
*inode
= nd
->inode
;
1169 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1170 nd
->path
.mnt
== nd
->root
.mnt
) {
1173 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1174 struct dentry
*old
= nd
->path
.dentry
;
1175 struct dentry
*parent
= old
->d_parent
;
1178 inode
= parent
->d_inode
;
1179 seq
= read_seqcount_begin(&parent
->d_seq
);
1180 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1182 nd
->path
.dentry
= parent
;
1186 if (!follow_up_rcu(&nd
->path
))
1188 inode
= nd
->path
.dentry
->d_inode
;
1189 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1191 while (d_mountpoint(nd
->path
.dentry
)) {
1192 struct mount
*mounted
;
1193 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1196 nd
->path
.mnt
= &mounted
->mnt
;
1197 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1198 inode
= nd
->path
.dentry
->d_inode
;
1199 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1200 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1207 nd
->flags
&= ~LOOKUP_RCU
;
1208 if (!(nd
->flags
& LOOKUP_ROOT
))
1209 nd
->root
.mnt
= NULL
;
1215 * Follow down to the covering mount currently visible to userspace. At each
1216 * point, the filesystem owning that dentry may be queried as to whether the
1217 * caller is permitted to proceed or not.
1219 int follow_down(struct path
*path
)
1224 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1225 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1226 /* Allow the filesystem to manage the transit without i_mutex
1229 * We indicate to the filesystem if someone is trying to mount
1230 * something here. This gives autofs the chance to deny anyone
1231 * other than its daemon the right to mount on its
1234 * The filesystem may sleep at this point.
1236 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1237 BUG_ON(!path
->dentry
->d_op
);
1238 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1239 ret
= path
->dentry
->d_op
->d_manage(
1240 path
->dentry
, false);
1242 return ret
== -EISDIR
? 0 : ret
;
1245 /* Transit to a mounted filesystem. */
1246 if (managed
& DCACHE_MOUNTED
) {
1247 struct vfsmount
*mounted
= lookup_mnt(path
);
1252 path
->mnt
= mounted
;
1253 path
->dentry
= dget(mounted
->mnt_root
);
1257 /* Don't handle automount points here */
1262 EXPORT_SYMBOL(follow_down
);
1265 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1267 static void follow_mount(struct path
*path
)
1269 while (d_mountpoint(path
->dentry
)) {
1270 struct vfsmount
*mounted
= lookup_mnt(path
);
1275 path
->mnt
= mounted
;
1276 path
->dentry
= dget(mounted
->mnt_root
);
1280 static void follow_dotdot(struct nameidata
*nd
)
1286 struct dentry
*old
= nd
->path
.dentry
;
1288 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1289 nd
->path
.mnt
== nd
->root
.mnt
) {
1292 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1293 /* rare case of legitimate dget_parent()... */
1294 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1298 if (!follow_up(&nd
->path
))
1301 follow_mount(&nd
->path
);
1302 nd
->inode
= nd
->path
.dentry
->d_inode
;
1306 * This looks up the name in dcache, possibly revalidates the old dentry and
1307 * allocates a new one if not found or not valid. In the need_lookup argument
1308 * returns whether i_op->lookup is necessary.
1310 * dir->d_inode->i_mutex must be held
1312 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1313 unsigned int flags
, bool *need_lookup
)
1315 struct dentry
*dentry
;
1318 *need_lookup
= false;
1319 dentry
= d_lookup(dir
, name
);
1321 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1322 error
= d_revalidate(dentry
, flags
);
1323 if (unlikely(error
<= 0)) {
1326 return ERR_PTR(error
);
1328 d_invalidate(dentry
);
1337 dentry
= d_alloc(dir
, name
);
1338 if (unlikely(!dentry
))
1339 return ERR_PTR(-ENOMEM
);
1341 *need_lookup
= true;
1347 * Call i_op->lookup on the dentry. The dentry must be negative and
1350 * dir->d_inode->i_mutex must be held
1352 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1357 /* Don't create child dentry for a dead directory. */
1358 if (unlikely(IS_DEADDIR(dir
))) {
1360 return ERR_PTR(-ENOENT
);
1363 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1364 if (unlikely(old
)) {
1371 static struct dentry
*__lookup_hash(struct qstr
*name
,
1372 struct dentry
*base
, unsigned int flags
)
1375 struct dentry
*dentry
;
1377 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1381 return lookup_real(base
->d_inode
, dentry
, flags
);
1385 * It's more convoluted than I'd like it to be, but... it's still fairly
1386 * small and for now I'd prefer to have fast path as straight as possible.
1387 * It _is_ time-critical.
1389 static int lookup_fast(struct nameidata
*nd
,
1390 struct path
*path
, struct inode
**inode
)
1392 struct vfsmount
*mnt
= nd
->path
.mnt
;
1393 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1399 * Rename seqlock is not required here because in the off chance
1400 * of a false negative due to a concurrent rename, we're going to
1401 * do the non-racy lookup, below.
1403 if (nd
->flags
& LOOKUP_RCU
) {
1406 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1411 * This sequence count validates that the inode matches
1412 * the dentry name information from lookup.
1414 *inode
= dentry
->d_inode
;
1415 negative
= d_is_negative(dentry
);
1416 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1422 * This sequence count validates that the parent had no
1423 * changes while we did the lookup of the dentry above.
1425 * The memory barrier in read_seqcount_begin of child is
1426 * enough, we can use __read_seqcount_retry here.
1428 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1432 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1433 status
= d_revalidate(dentry
, nd
->flags
);
1434 if (unlikely(status
<= 0)) {
1435 if (status
!= -ECHILD
)
1441 path
->dentry
= dentry
;
1442 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1445 if (unlazy_walk(nd
, dentry
))
1448 dentry
= __d_lookup(parent
, &nd
->last
);
1451 if (unlikely(!dentry
))
1454 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1455 status
= d_revalidate(dentry
, nd
->flags
);
1456 if (unlikely(status
<= 0)) {
1461 d_invalidate(dentry
);
1466 if (unlikely(d_is_negative(dentry
))) {
1471 path
->dentry
= dentry
;
1472 err
= follow_managed(path
, nd
->flags
);
1473 if (unlikely(err
< 0)) {
1474 path_put_conditional(path
, nd
);
1478 nd
->flags
|= LOOKUP_JUMPED
;
1479 *inode
= path
->dentry
->d_inode
;
1486 /* Fast lookup failed, do it the slow way */
1487 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1489 struct dentry
*dentry
, *parent
;
1492 parent
= nd
->path
.dentry
;
1493 BUG_ON(nd
->inode
!= parent
->d_inode
);
1495 mutex_lock(&parent
->d_inode
->i_mutex
);
1496 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1497 mutex_unlock(&parent
->d_inode
->i_mutex
);
1499 return PTR_ERR(dentry
);
1500 path
->mnt
= nd
->path
.mnt
;
1501 path
->dentry
= dentry
;
1502 err
= follow_managed(path
, nd
->flags
);
1503 if (unlikely(err
< 0)) {
1504 path_put_conditional(path
, nd
);
1508 nd
->flags
|= LOOKUP_JUMPED
;
1512 static inline int may_lookup(struct nameidata
*nd
)
1514 if (nd
->flags
& LOOKUP_RCU
) {
1515 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1518 if (unlazy_walk(nd
, NULL
))
1521 return inode_permission(nd
->inode
, MAY_EXEC
);
1524 static inline int handle_dots(struct nameidata
*nd
, int type
)
1526 if (type
== LAST_DOTDOT
) {
1527 if (nd
->flags
& LOOKUP_RCU
) {
1528 if (follow_dotdot_rcu(nd
))
1536 static void terminate_walk(struct nameidata
*nd
)
1538 if (!(nd
->flags
& LOOKUP_RCU
)) {
1539 path_put(&nd
->path
);
1541 nd
->flags
&= ~LOOKUP_RCU
;
1542 if (!(nd
->flags
& LOOKUP_ROOT
))
1543 nd
->root
.mnt
= NULL
;
1549 * Do we need to follow links? We _really_ want to be able
1550 * to do this check without having to look at inode->i_op,
1551 * so we keep a cache of "no, this doesn't need follow_link"
1552 * for the common case.
1554 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1556 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1559 static int walk_component(struct nameidata
*nd
, struct path
*path
, int follow
)
1561 struct inode
*inode
;
1564 * "." and ".." are special - ".." especially so because it has
1565 * to be able to know about the current root directory and
1566 * parent relationships.
1568 if (unlikely(nd
->last_type
!= LAST_NORM
))
1569 return handle_dots(nd
, nd
->last_type
);
1570 err
= lookup_fast(nd
, path
, &inode
);
1571 if (unlikely(err
)) {
1575 err
= lookup_slow(nd
, path
);
1579 inode
= path
->dentry
->d_inode
;
1581 if (d_is_negative(path
->dentry
))
1585 if (should_follow_link(path
->dentry
, follow
)) {
1586 if (nd
->flags
& LOOKUP_RCU
) {
1587 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
1588 unlazy_walk(nd
, path
->dentry
))) {
1593 BUG_ON(inode
!= path
->dentry
->d_inode
);
1596 path_to_nameidata(path
, nd
);
1601 path_to_nameidata(path
, nd
);
1608 * This limits recursive symlink follows to 8, while
1609 * limiting consecutive symlinks to 40.
1611 * Without that kind of total limit, nasty chains of consecutive
1612 * symlinks can cause almost arbitrarily long lookups.
1614 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1618 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1619 path_put_conditional(path
, nd
);
1620 path_put(&nd
->path
);
1623 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1626 current
->link_count
++;
1629 struct path link
= *path
;
1632 res
= follow_link(&link
, nd
, &cookie
);
1635 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1636 put_link(nd
, &link
, cookie
);
1639 current
->link_count
--;
1645 * We can do the critical dentry name comparison and hashing
1646 * operations one word at a time, but we are limited to:
1648 * - Architectures with fast unaligned word accesses. We could
1649 * do a "get_unaligned()" if this helps and is sufficiently
1652 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1653 * do not trap on the (extremely unlikely) case of a page
1654 * crossing operation.
1656 * - Furthermore, we need an efficient 64-bit compile for the
1657 * 64-bit case in order to generate the "number of bytes in
1658 * the final mask". Again, that could be replaced with a
1659 * efficient population count instruction or similar.
1661 #ifdef CONFIG_DCACHE_WORD_ACCESS
1663 #include <asm/word-at-a-time.h>
1667 static inline unsigned int fold_hash(unsigned long hash
)
1669 return hash_64(hash
, 32);
1672 #else /* 32-bit case */
1674 #define fold_hash(x) (x)
1678 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1680 unsigned long a
, mask
;
1681 unsigned long hash
= 0;
1684 a
= load_unaligned_zeropad(name
);
1685 if (len
< sizeof(unsigned long))
1689 name
+= sizeof(unsigned long);
1690 len
-= sizeof(unsigned long);
1694 mask
= bytemask_from_count(len
);
1697 return fold_hash(hash
);
1699 EXPORT_SYMBOL(full_name_hash
);
1702 * Calculate the length and hash of the path component, and
1703 * return the "hash_len" as the result.
1705 static inline u64
hash_name(const char *name
)
1707 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1708 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1711 len
= -sizeof(unsigned long);
1713 hash
= (hash
+ a
) * 9;
1714 len
+= sizeof(unsigned long);
1715 a
= load_unaligned_zeropad(name
+len
);
1716 b
= a
^ REPEAT_BYTE('/');
1717 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1719 adata
= prep_zero_mask(a
, adata
, &constants
);
1720 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1722 mask
= create_zero_mask(adata
| bdata
);
1724 hash
+= a
& zero_bytemask(mask
);
1725 len
+= find_zero(mask
);
1726 return hashlen_create(fold_hash(hash
), len
);
1731 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1733 unsigned long hash
= init_name_hash();
1735 hash
= partial_name_hash(*name
++, hash
);
1736 return end_name_hash(hash
);
1738 EXPORT_SYMBOL(full_name_hash
);
1741 * We know there's a real path component here of at least
1744 static inline u64
hash_name(const char *name
)
1746 unsigned long hash
= init_name_hash();
1747 unsigned long len
= 0, c
;
1749 c
= (unsigned char)*name
;
1752 hash
= partial_name_hash(c
, hash
);
1753 c
= (unsigned char)name
[len
];
1754 } while (c
&& c
!= '/');
1755 return hashlen_create(end_name_hash(hash
), len
);
1762 * This is the basic name resolution function, turning a pathname into
1763 * the final dentry. We expect 'base' to be positive and a directory.
1765 * Returns 0 and nd will have valid dentry and mnt on success.
1766 * Returns error and drops reference to input namei data on failure.
1768 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1778 /* At this point we know we have a real path component. */
1783 err
= may_lookup(nd
);
1787 hash_len
= hash_name(name
);
1790 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1792 if (name
[1] == '.') {
1794 nd
->flags
|= LOOKUP_JUMPED
;
1800 if (likely(type
== LAST_NORM
)) {
1801 struct dentry
*parent
= nd
->path
.dentry
;
1802 nd
->flags
&= ~LOOKUP_JUMPED
;
1803 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1804 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1805 err
= parent
->d_op
->d_hash(parent
, &this);
1808 hash_len
= this.hash_len
;
1813 nd
->last
.hash_len
= hash_len
;
1814 nd
->last
.name
= name
;
1815 nd
->last_type
= type
;
1817 name
+= hashlen_len(hash_len
);
1821 * If it wasn't NUL, we know it was '/'. Skip that
1822 * slash, and continue until no more slashes.
1826 } while (unlikely(*name
== '/'));
1830 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1835 err
= nested_symlink(&next
, nd
);
1839 if (!d_can_lookup(nd
->path
.dentry
)) {
1848 static int path_init(int dfd
, const struct filename
*name
, unsigned int flags
,
1849 struct nameidata
*nd
)
1852 const char *s
= name
->name
;
1854 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1855 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1858 if (flags
& LOOKUP_ROOT
) {
1859 struct dentry
*root
= nd
->root
.dentry
;
1860 struct inode
*inode
= root
->d_inode
;
1862 if (!d_can_lookup(root
))
1864 retval
= inode_permission(inode
, MAY_EXEC
);
1868 nd
->path
= nd
->root
;
1870 if (flags
& LOOKUP_RCU
) {
1872 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1873 nd
->m_seq
= read_seqbegin(&mount_lock
);
1875 path_get(&nd
->path
);
1880 nd
->root
.mnt
= NULL
;
1882 nd
->m_seq
= read_seqbegin(&mount_lock
);
1884 if (flags
& LOOKUP_RCU
) {
1886 nd
->seq
= set_root_rcu(nd
);
1889 path_get(&nd
->root
);
1891 nd
->path
= nd
->root
;
1892 } else if (dfd
== AT_FDCWD
) {
1893 if (flags
& LOOKUP_RCU
) {
1894 struct fs_struct
*fs
= current
->fs
;
1900 seq
= read_seqcount_begin(&fs
->seq
);
1902 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1903 } while (read_seqcount_retry(&fs
->seq
, seq
));
1905 get_fs_pwd(current
->fs
, &nd
->path
);
1908 /* Caller must check execute permissions on the starting path component */
1909 struct fd f
= fdget_raw(dfd
);
1910 struct dentry
*dentry
;
1915 dentry
= f
.file
->f_path
.dentry
;
1918 if (!d_can_lookup(dentry
)) {
1924 nd
->path
= f
.file
->f_path
;
1925 if (flags
& LOOKUP_RCU
) {
1926 if (f
.flags
& FDPUT_FPUT
)
1928 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1931 path_get(&nd
->path
);
1936 nd
->inode
= nd
->path
.dentry
->d_inode
;
1937 if (!(flags
& LOOKUP_RCU
))
1939 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1941 if (!(nd
->flags
& LOOKUP_ROOT
))
1942 nd
->root
.mnt
= NULL
;
1946 current
->total_link_count
= 0;
1947 return link_path_walk(s
, nd
);
1950 static void path_cleanup(struct nameidata
*nd
)
1952 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1953 path_put(&nd
->root
);
1954 nd
->root
.mnt
= NULL
;
1956 if (unlikely(nd
->base
))
1960 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1962 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1963 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1965 nd
->flags
&= ~LOOKUP_PARENT
;
1966 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1969 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1970 static int path_lookupat(int dfd
, const struct filename
*name
,
1971 unsigned int flags
, struct nameidata
*nd
)
1977 * Path walking is largely split up into 2 different synchronisation
1978 * schemes, rcu-walk and ref-walk (explained in
1979 * Documentation/filesystems/path-lookup.txt). These share much of the
1980 * path walk code, but some things particularly setup, cleanup, and
1981 * following mounts are sufficiently divergent that functions are
1982 * duplicated. Typically there is a function foo(), and its RCU
1983 * analogue, foo_rcu().
1985 * -ECHILD is the error number of choice (just to avoid clashes) that
1986 * is returned if some aspect of an rcu-walk fails. Such an error must
1987 * be handled by restarting a traditional ref-walk (which will always
1988 * be able to complete).
1990 err
= path_init(dfd
, name
, flags
, nd
);
1991 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1992 err
= lookup_last(nd
, &path
);
1995 struct path link
= path
;
1996 err
= may_follow_link(&link
, nd
);
1999 nd
->flags
|= LOOKUP_PARENT
;
2000 err
= follow_link(&link
, nd
, &cookie
);
2003 err
= lookup_last(nd
, &path
);
2004 put_link(nd
, &link
, cookie
);
2009 err
= complete_walk(nd
);
2011 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2012 if (!d_can_lookup(nd
->path
.dentry
)) {
2013 path_put(&nd
->path
);
2022 static int filename_lookup(int dfd
, struct filename
*name
,
2023 unsigned int flags
, struct nameidata
*nd
)
2025 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2026 if (unlikely(retval
== -ECHILD
))
2027 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2028 if (unlikely(retval
== -ESTALE
))
2029 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2031 if (likely(!retval
))
2032 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2036 /* does lookup, returns the object with parent locked */
2037 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2039 struct filename
*filename
= getname_kernel(name
);
2040 struct nameidata nd
;
2044 if (IS_ERR(filename
))
2045 return ERR_CAST(filename
);
2047 err
= filename_lookup(AT_FDCWD
, filename
, LOOKUP_PARENT
, &nd
);
2052 if (nd
.last_type
!= LAST_NORM
) {
2054 d
= ERR_PTR(-EINVAL
);
2057 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2058 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2060 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2070 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2072 struct nameidata nd
;
2073 struct filename
*filename
= getname_kernel(name
);
2074 int res
= PTR_ERR(filename
);
2076 if (!IS_ERR(filename
)) {
2077 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2084 EXPORT_SYMBOL(kern_path
);
2087 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2088 * @dentry: pointer to dentry of the base directory
2089 * @mnt: pointer to vfs mount of the base directory
2090 * @name: pointer to file name
2091 * @flags: lookup flags
2092 * @path: pointer to struct path to fill
2094 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2095 const char *name
, unsigned int flags
,
2098 struct filename
*filename
= getname_kernel(name
);
2099 int err
= PTR_ERR(filename
);
2101 BUG_ON(flags
& LOOKUP_PARENT
);
2103 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2104 if (!IS_ERR(filename
)) {
2105 struct nameidata nd
;
2106 nd
.root
.dentry
= dentry
;
2108 err
= filename_lookup(AT_FDCWD
, filename
,
2109 flags
| LOOKUP_ROOT
, &nd
);
2116 EXPORT_SYMBOL(vfs_path_lookup
);
2119 * lookup_one_len - filesystem helper to lookup single pathname component
2120 * @name: pathname component to lookup
2121 * @base: base directory to lookup from
2122 * @len: maximum length @len should be interpreted to
2124 * Note that this routine is purely a helper for filesystem usage and should
2125 * not be called by generic code.
2127 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2133 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2137 this.hash
= full_name_hash(name
, len
);
2139 return ERR_PTR(-EACCES
);
2141 if (unlikely(name
[0] == '.')) {
2142 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2143 return ERR_PTR(-EACCES
);
2147 c
= *(const unsigned char *)name
++;
2148 if (c
== '/' || c
== '\0')
2149 return ERR_PTR(-EACCES
);
2152 * See if the low-level filesystem might want
2153 * to use its own hash..
2155 if (base
->d_flags
& DCACHE_OP_HASH
) {
2156 int err
= base
->d_op
->d_hash(base
, &this);
2158 return ERR_PTR(err
);
2161 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2163 return ERR_PTR(err
);
2165 return __lookup_hash(&this, base
, 0);
2167 EXPORT_SYMBOL(lookup_one_len
);
2169 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2170 struct path
*path
, int *empty
)
2172 struct nameidata nd
;
2173 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2174 int err
= PTR_ERR(tmp
);
2177 BUG_ON(flags
& LOOKUP_PARENT
);
2179 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2187 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2190 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2192 EXPORT_SYMBOL(user_path_at
);
2195 * NB: most callers don't do anything directly with the reference to the
2196 * to struct filename, but the nd->last pointer points into the name string
2197 * allocated by getname. So we must hold the reference to it until all
2198 * path-walking is complete.
2200 static struct filename
*
2201 user_path_parent(int dfd
, const char __user
*path
,
2202 struct path
*parent
,
2207 struct nameidata nd
;
2208 struct filename
*s
= getname(path
);
2211 /* only LOOKUP_REVAL is allowed in extra flags */
2212 flags
&= LOOKUP_REVAL
;
2217 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, &nd
);
2220 return ERR_PTR(error
);
2224 *type
= nd
.last_type
;
2230 * mountpoint_last - look up last component for umount
2231 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2232 * @path: pointer to container for result
2234 * This is a special lookup_last function just for umount. In this case, we
2235 * need to resolve the path without doing any revalidation.
2237 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2238 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2239 * in almost all cases, this lookup will be served out of the dcache. The only
2240 * cases where it won't are if nd->last refers to a symlink or the path is
2241 * bogus and it doesn't exist.
2244 * -error: if there was an error during lookup. This includes -ENOENT if the
2245 * lookup found a negative dentry. The nd->path reference will also be
2248 * 0: if we successfully resolved nd->path and found it to not to be a
2249 * symlink that needs to be followed. "path" will also be populated.
2250 * The nd->path reference will also be put.
2252 * 1: if we successfully resolved nd->last and found it to be a symlink
2253 * that needs to be followed. "path" will be populated with the path
2254 * to the link, and nd->path will *not* be put.
2257 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2260 struct dentry
*dentry
;
2261 struct dentry
*dir
= nd
->path
.dentry
;
2263 /* If we're in rcuwalk, drop out of it to handle last component */
2264 if (nd
->flags
& LOOKUP_RCU
) {
2265 if (unlazy_walk(nd
, NULL
)) {
2271 nd
->flags
&= ~LOOKUP_PARENT
;
2273 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2274 error
= handle_dots(nd
, nd
->last_type
);
2277 dentry
= dget(nd
->path
.dentry
);
2281 mutex_lock(&dir
->d_inode
->i_mutex
);
2282 dentry
= d_lookup(dir
, &nd
->last
);
2285 * No cached dentry. Mounted dentries are pinned in the cache,
2286 * so that means that this dentry is probably a symlink or the
2287 * path doesn't actually point to a mounted dentry.
2289 dentry
= d_alloc(dir
, &nd
->last
);
2292 mutex_unlock(&dir
->d_inode
->i_mutex
);
2295 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2296 error
= PTR_ERR(dentry
);
2297 if (IS_ERR(dentry
)) {
2298 mutex_unlock(&dir
->d_inode
->i_mutex
);
2302 mutex_unlock(&dir
->d_inode
->i_mutex
);
2305 if (d_is_negative(dentry
)) {
2310 path
->dentry
= dentry
;
2311 path
->mnt
= nd
->path
.mnt
;
2312 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
))
2323 * path_mountpoint - look up a path to be umounted
2324 * @dfd: directory file descriptor to start walk from
2325 * @name: full pathname to walk
2326 * @path: pointer to container for result
2327 * @flags: lookup flags
2329 * Look up the given name, but don't attempt to revalidate the last component.
2330 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2333 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2334 struct nameidata
*nd
, unsigned int flags
)
2336 int err
= path_init(dfd
, name
, flags
, nd
);
2340 err
= mountpoint_last(nd
, path
);
2343 struct path link
= *path
;
2344 err
= may_follow_link(&link
, nd
);
2347 nd
->flags
|= LOOKUP_PARENT
;
2348 err
= follow_link(&link
, nd
, &cookie
);
2351 err
= mountpoint_last(nd
, path
);
2352 put_link(nd
, &link
, cookie
);
2360 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2363 struct nameidata nd
;
2366 return PTR_ERR(name
);
2367 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_RCU
);
2368 if (unlikely(error
== -ECHILD
))
2369 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
);
2370 if (unlikely(error
== -ESTALE
))
2371 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_REVAL
);
2373 audit_inode(name
, path
->dentry
, 0);
2379 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2380 * @dfd: directory file descriptor
2381 * @name: pathname from userland
2382 * @flags: lookup flags
2383 * @path: pointer to container to hold result
2385 * A umount is a special case for path walking. We're not actually interested
2386 * in the inode in this situation, and ESTALE errors can be a problem. We
2387 * simply want track down the dentry and vfsmount attached at the mountpoint
2388 * and avoid revalidating the last component.
2390 * Returns 0 and populates "path" on success.
2393 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2396 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2400 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2403 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2405 EXPORT_SYMBOL(kern_path_mountpoint
);
2407 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2409 kuid_t fsuid
= current_fsuid();
2411 if (uid_eq(inode
->i_uid
, fsuid
))
2413 if (uid_eq(dir
->i_uid
, fsuid
))
2415 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2417 EXPORT_SYMBOL(__check_sticky
);
2420 * Check whether we can remove a link victim from directory dir, check
2421 * whether the type of victim is right.
2422 * 1. We can't do it if dir is read-only (done in permission())
2423 * 2. We should have write and exec permissions on dir
2424 * 3. We can't remove anything from append-only dir
2425 * 4. We can't do anything with immutable dir (done in permission())
2426 * 5. If the sticky bit on dir is set we should either
2427 * a. be owner of dir, or
2428 * b. be owner of victim, or
2429 * c. have CAP_FOWNER capability
2430 * 6. If the victim is append-only or immutable we can't do antyhing with
2431 * links pointing to it.
2432 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2433 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2434 * 9. We can't remove a root or mountpoint.
2435 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2436 * nfs_async_unlink().
2438 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2440 struct inode
*inode
= victim
->d_inode
;
2443 if (d_is_negative(victim
))
2447 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2448 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2450 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2456 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2457 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2460 if (!d_is_dir(victim
))
2462 if (IS_ROOT(victim
))
2464 } else if (d_is_dir(victim
))
2466 if (IS_DEADDIR(dir
))
2468 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2473 /* Check whether we can create an object with dentry child in directory
2475 * 1. We can't do it if child already exists (open has special treatment for
2476 * this case, but since we are inlined it's OK)
2477 * 2. We can't do it if dir is read-only (done in permission())
2478 * 3. We should have write and exec permissions on dir
2479 * 4. We can't do it if dir is immutable (done in permission())
2481 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2483 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2486 if (IS_DEADDIR(dir
))
2488 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2492 * p1 and p2 should be directories on the same fs.
2494 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2499 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2503 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2505 p
= d_ancestor(p2
, p1
);
2507 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2508 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2512 p
= d_ancestor(p1
, p2
);
2514 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2515 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2519 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2520 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2523 EXPORT_SYMBOL(lock_rename
);
2525 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2527 mutex_unlock(&p1
->d_inode
->i_mutex
);
2529 mutex_unlock(&p2
->d_inode
->i_mutex
);
2530 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2533 EXPORT_SYMBOL(unlock_rename
);
2535 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2538 int error
= may_create(dir
, dentry
);
2542 if (!dir
->i_op
->create
)
2543 return -EACCES
; /* shouldn't it be ENOSYS? */
2546 error
= security_inode_create(dir
, dentry
, mode
);
2549 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2551 fsnotify_create(dir
, dentry
);
2554 EXPORT_SYMBOL(vfs_create
);
2556 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2558 struct dentry
*dentry
= path
->dentry
;
2559 struct inode
*inode
= dentry
->d_inode
;
2569 switch (inode
->i_mode
& S_IFMT
) {
2573 if (acc_mode
& MAY_WRITE
)
2578 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2587 error
= inode_permission(inode
, acc_mode
);
2592 * An append-only file must be opened in append mode for writing.
2594 if (IS_APPEND(inode
)) {
2595 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2601 /* O_NOATIME can only be set by the owner or superuser */
2602 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2608 static int handle_truncate(struct file
*filp
)
2610 struct path
*path
= &filp
->f_path
;
2611 struct inode
*inode
= path
->dentry
->d_inode
;
2612 int error
= get_write_access(inode
);
2616 * Refuse to truncate files with mandatory locks held on them.
2618 error
= locks_verify_locked(filp
);
2620 error
= security_path_truncate(path
);
2622 error
= do_truncate(path
->dentry
, 0,
2623 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2626 put_write_access(inode
);
2630 static inline int open_to_namei_flags(int flag
)
2632 if ((flag
& O_ACCMODE
) == 3)
2637 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2639 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2643 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2647 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2651 * Attempt to atomically look up, create and open a file from a negative
2654 * Returns 0 if successful. The file will have been created and attached to
2655 * @file by the filesystem calling finish_open().
2657 * Returns 1 if the file was looked up only or didn't need creating. The
2658 * caller will need to perform the open themselves. @path will have been
2659 * updated to point to the new dentry. This may be negative.
2661 * Returns an error code otherwise.
2663 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2664 struct path
*path
, struct file
*file
,
2665 const struct open_flags
*op
,
2666 bool got_write
, bool need_lookup
,
2669 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2670 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2674 int create_error
= 0;
2675 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2678 BUG_ON(dentry
->d_inode
);
2680 /* Don't create child dentry for a dead directory. */
2681 if (unlikely(IS_DEADDIR(dir
))) {
2687 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2688 mode
&= ~current_umask();
2690 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2692 open_flag
&= ~O_TRUNC
;
2695 * Checking write permission is tricky, bacuse we don't know if we are
2696 * going to actually need it: O_CREAT opens should work as long as the
2697 * file exists. But checking existence breaks atomicity. The trick is
2698 * to check access and if not granted clear O_CREAT from the flags.
2700 * Another problem is returing the "right" error value (e.g. for an
2701 * O_EXCL open we want to return EEXIST not EROFS).
2703 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2704 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2705 if (!(open_flag
& O_CREAT
)) {
2707 * No O_CREATE -> atomicity not a requirement -> fall
2708 * back to lookup + open
2711 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2712 /* Fall back and fail with the right error */
2713 create_error
= -EROFS
;
2716 /* No side effects, safe to clear O_CREAT */
2717 create_error
= -EROFS
;
2718 open_flag
&= ~O_CREAT
;
2722 if (open_flag
& O_CREAT
) {
2723 error
= may_o_create(&nd
->path
, dentry
, mode
);
2725 create_error
= error
;
2726 if (open_flag
& O_EXCL
)
2728 open_flag
&= ~O_CREAT
;
2732 if (nd
->flags
& LOOKUP_DIRECTORY
)
2733 open_flag
|= O_DIRECTORY
;
2735 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2736 file
->f_path
.mnt
= nd
->path
.mnt
;
2737 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2740 if (create_error
&& error
== -ENOENT
)
2741 error
= create_error
;
2745 if (error
) { /* returned 1, that is */
2746 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2750 if (file
->f_path
.dentry
) {
2752 dentry
= file
->f_path
.dentry
;
2754 if (*opened
& FILE_CREATED
)
2755 fsnotify_create(dir
, dentry
);
2756 if (!dentry
->d_inode
) {
2757 WARN_ON(*opened
& FILE_CREATED
);
2759 error
= create_error
;
2763 if (excl
&& !(*opened
& FILE_CREATED
)) {
2772 * We didn't have the inode before the open, so check open permission
2775 acc_mode
= op
->acc_mode
;
2776 if (*opened
& FILE_CREATED
) {
2777 WARN_ON(!(open_flag
& O_CREAT
));
2778 fsnotify_create(dir
, dentry
);
2779 acc_mode
= MAY_OPEN
;
2781 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2791 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2793 return PTR_ERR(dentry
);
2796 int open_flag
= op
->open_flag
;
2798 error
= create_error
;
2799 if ((open_flag
& O_EXCL
)) {
2800 if (!dentry
->d_inode
)
2802 } else if (!dentry
->d_inode
) {
2804 } else if ((open_flag
& O_TRUNC
) &&
2808 /* will fail later, go on to get the right error */
2812 path
->dentry
= dentry
;
2813 path
->mnt
= nd
->path
.mnt
;
2818 * Look up and maybe create and open the last component.
2820 * Must be called with i_mutex held on parent.
2822 * Returns 0 if the file was successfully atomically created (if necessary) and
2823 * opened. In this case the file will be returned attached to @file.
2825 * Returns 1 if the file was not completely opened at this time, though lookups
2826 * and creations will have been performed and the dentry returned in @path will
2827 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2828 * specified then a negative dentry may be returned.
2830 * An error code is returned otherwise.
2832 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2833 * cleared otherwise prior to returning.
2835 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2837 const struct open_flags
*op
,
2838 bool got_write
, int *opened
)
2840 struct dentry
*dir
= nd
->path
.dentry
;
2841 struct inode
*dir_inode
= dir
->d_inode
;
2842 struct dentry
*dentry
;
2846 *opened
&= ~FILE_CREATED
;
2847 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2849 return PTR_ERR(dentry
);
2851 /* Cached positive dentry: will open in f_op->open */
2852 if (!need_lookup
&& dentry
->d_inode
)
2855 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2856 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2857 need_lookup
, opened
);
2861 BUG_ON(dentry
->d_inode
);
2863 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2865 return PTR_ERR(dentry
);
2868 /* Negative dentry, just create the file */
2869 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2870 umode_t mode
= op
->mode
;
2871 if (!IS_POSIXACL(dir
->d_inode
))
2872 mode
&= ~current_umask();
2874 * This write is needed to ensure that a
2875 * rw->ro transition does not occur between
2876 * the time when the file is created and when
2877 * a permanent write count is taken through
2878 * the 'struct file' in finish_open().
2884 *opened
|= FILE_CREATED
;
2885 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2888 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2889 nd
->flags
& LOOKUP_EXCL
);
2894 path
->dentry
= dentry
;
2895 path
->mnt
= nd
->path
.mnt
;
2904 * Handle the last step of open()
2906 static int do_last(struct nameidata
*nd
, struct path
*path
,
2907 struct file
*file
, const struct open_flags
*op
,
2908 int *opened
, struct filename
*name
)
2910 struct dentry
*dir
= nd
->path
.dentry
;
2911 int open_flag
= op
->open_flag
;
2912 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2913 bool got_write
= false;
2914 int acc_mode
= op
->acc_mode
;
2915 struct inode
*inode
;
2916 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2917 bool retried
= false;
2920 nd
->flags
&= ~LOOKUP_PARENT
;
2921 nd
->flags
|= op
->intent
;
2923 if (nd
->last_type
!= LAST_NORM
) {
2924 error
= handle_dots(nd
, nd
->last_type
);
2930 if (!(open_flag
& O_CREAT
)) {
2931 if (nd
->last
.name
[nd
->last
.len
])
2932 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2933 /* we _can_ be in RCU mode here */
2934 error
= lookup_fast(nd
, path
, &inode
);
2941 BUG_ON(nd
->inode
!= dir
->d_inode
);
2943 /* create side of things */
2945 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2946 * has been cleared when we got to the last component we are
2949 error
= complete_walk(nd
);
2953 audit_inode(name
, dir
, LOOKUP_PARENT
);
2955 /* trailing slashes? */
2956 if (nd
->last
.name
[nd
->last
.len
])
2961 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2962 error
= mnt_want_write(nd
->path
.mnt
);
2966 * do _not_ fail yet - we might not need that or fail with
2967 * a different error; let lookup_open() decide; we'll be
2968 * dropping this one anyway.
2971 mutex_lock(&dir
->d_inode
->i_mutex
);
2972 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2973 mutex_unlock(&dir
->d_inode
->i_mutex
);
2979 if ((*opened
& FILE_CREATED
) ||
2980 !S_ISREG(file_inode(file
)->i_mode
))
2981 will_truncate
= false;
2983 audit_inode(name
, file
->f_path
.dentry
, 0);
2987 if (*opened
& FILE_CREATED
) {
2988 /* Don't check for write permission, don't truncate */
2989 open_flag
&= ~O_TRUNC
;
2990 will_truncate
= false;
2991 acc_mode
= MAY_OPEN
;
2992 path_to_nameidata(path
, nd
);
2993 goto finish_open_created
;
2997 * create/update audit record if it already exists.
2999 if (d_is_positive(path
->dentry
))
3000 audit_inode(name
, path
->dentry
, 0);
3003 * If atomic_open() acquired write access it is dropped now due to
3004 * possible mount and symlink following (this might be optimized away if
3008 mnt_drop_write(nd
->path
.mnt
);
3013 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3016 error
= follow_managed(path
, nd
->flags
);
3021 nd
->flags
|= LOOKUP_JUMPED
;
3023 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3024 inode
= path
->dentry
->d_inode
;
3026 if (d_is_negative(path
->dentry
)) {
3027 path_to_nameidata(path
, nd
);
3031 if (should_follow_link(path
->dentry
, nd
->flags
& LOOKUP_FOLLOW
)) {
3032 if (nd
->flags
& LOOKUP_RCU
) {
3033 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
3034 unlazy_walk(nd
, path
->dentry
))) {
3039 BUG_ON(inode
!= path
->dentry
->d_inode
);
3043 if (unlikely(d_is_symlink(path
->dentry
)) && !(open_flag
& O_PATH
)) {
3044 path_to_nameidata(path
, nd
);
3049 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3050 path_to_nameidata(path
, nd
);
3052 save_parent
.dentry
= nd
->path
.dentry
;
3053 save_parent
.mnt
= mntget(path
->mnt
);
3054 nd
->path
.dentry
= path
->dentry
;
3058 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3060 error
= complete_walk(nd
);
3062 path_put(&save_parent
);
3065 audit_inode(name
, nd
->path
.dentry
, 0);
3067 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3070 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3072 if (!d_is_reg(nd
->path
.dentry
))
3073 will_truncate
= false;
3075 if (will_truncate
) {
3076 error
= mnt_want_write(nd
->path
.mnt
);
3081 finish_open_created
:
3082 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3086 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3087 error
= vfs_open(&nd
->path
, file
, current_cred());
3089 *opened
|= FILE_OPENED
;
3091 if (error
== -EOPENSTALE
)
3096 error
= open_check_o_direct(file
);
3099 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3103 if (will_truncate
) {
3104 error
= handle_truncate(file
);
3110 mnt_drop_write(nd
->path
.mnt
);
3111 path_put(&save_parent
);
3116 path_put_conditional(path
, nd
);
3123 /* If no saved parent or already retried then can't retry */
3124 if (!save_parent
.dentry
|| retried
)
3127 BUG_ON(save_parent
.dentry
!= dir
);
3128 path_put(&nd
->path
);
3129 nd
->path
= save_parent
;
3130 nd
->inode
= dir
->d_inode
;
3131 save_parent
.mnt
= NULL
;
3132 save_parent
.dentry
= NULL
;
3134 mnt_drop_write(nd
->path
.mnt
);
3141 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3142 struct nameidata
*nd
, int flags
,
3143 const struct open_flags
*op
,
3144 struct file
*file
, int *opened
)
3146 static const struct qstr name
= QSTR_INIT("/", 1);
3147 struct dentry
*dentry
, *child
;
3149 int error
= path_lookupat(dfd
, pathname
,
3150 flags
| LOOKUP_DIRECTORY
, nd
);
3151 if (unlikely(error
))
3153 error
= mnt_want_write(nd
->path
.mnt
);
3154 if (unlikely(error
))
3156 /* we want directory to be writable */
3157 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3160 dentry
= nd
->path
.dentry
;
3161 dir
= dentry
->d_inode
;
3162 if (!dir
->i_op
->tmpfile
) {
3163 error
= -EOPNOTSUPP
;
3166 child
= d_alloc(dentry
, &name
);
3167 if (unlikely(!child
)) {
3171 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3172 nd
->flags
|= op
->intent
;
3173 dput(nd
->path
.dentry
);
3174 nd
->path
.dentry
= child
;
3175 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3178 audit_inode(pathname
, nd
->path
.dentry
, 0);
3179 /* Don't check for other permissions, the inode was just created */
3180 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3183 file
->f_path
.mnt
= nd
->path
.mnt
;
3184 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3187 error
= open_check_o_direct(file
);
3190 } else if (!(op
->open_flag
& O_EXCL
)) {
3191 struct inode
*inode
= file_inode(file
);
3192 spin_lock(&inode
->i_lock
);
3193 inode
->i_state
|= I_LINKABLE
;
3194 spin_unlock(&inode
->i_lock
);
3197 mnt_drop_write(nd
->path
.mnt
);
3199 path_put(&nd
->path
);
3203 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3204 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3211 file
= get_empty_filp();
3215 file
->f_flags
= op
->open_flag
;
3217 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3218 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3222 error
= path_init(dfd
, pathname
, flags
, nd
);
3223 if (unlikely(error
))
3226 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3227 while (unlikely(error
> 0)) { /* trailing symlink */
3228 struct path link
= path
;
3230 error
= may_follow_link(&link
, nd
);
3231 if (unlikely(error
))
3233 nd
->flags
|= LOOKUP_PARENT
;
3234 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3235 error
= follow_link(&link
, nd
, &cookie
);
3236 if (unlikely(error
))
3238 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3239 put_link(nd
, &link
, cookie
);
3244 if (!(opened
& FILE_OPENED
)) {
3248 if (unlikely(error
)) {
3249 if (error
== -EOPENSTALE
) {
3250 if (flags
& LOOKUP_RCU
)
3255 file
= ERR_PTR(error
);
3260 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3261 const struct open_flags
*op
)
3263 struct nameidata nd
;
3264 int flags
= op
->lookup_flags
;
3267 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3268 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3269 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3270 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3271 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3275 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3276 const char *name
, const struct open_flags
*op
)
3278 struct nameidata nd
;
3280 struct filename
*filename
;
3281 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3284 nd
.root
.dentry
= dentry
;
3286 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3287 return ERR_PTR(-ELOOP
);
3289 filename
= getname_kernel(name
);
3290 if (unlikely(IS_ERR(filename
)))
3291 return ERR_CAST(filename
);
3293 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3294 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3295 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3296 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3297 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3302 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3303 struct path
*path
, unsigned int lookup_flags
)
3305 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3306 struct nameidata nd
;
3309 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3312 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3313 * other flags passed in are ignored!
3315 lookup_flags
&= LOOKUP_REVAL
;
3317 error
= filename_lookup(dfd
, name
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3319 return ERR_PTR(error
);
3322 * Yucky last component or no last component at all?
3323 * (foo/., foo/.., /////)
3325 if (nd
.last_type
!= LAST_NORM
)
3327 nd
.flags
&= ~LOOKUP_PARENT
;
3328 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3330 /* don't fail immediately if it's r/o, at least try to report other errors */
3331 err2
= mnt_want_write(nd
.path
.mnt
);
3333 * Do the final lookup.
3335 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3336 dentry
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, nd
.flags
);
3341 if (d_is_positive(dentry
))
3345 * Special case - lookup gave negative, but... we had foo/bar/
3346 * From the vfs_mknod() POV we just have a negative dentry -
3347 * all is fine. Let's be bastards - you had / on the end, you've
3348 * been asking for (non-existent) directory. -ENOENT for you.
3350 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3354 if (unlikely(err2
)) {
3362 dentry
= ERR_PTR(error
);
3364 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3366 mnt_drop_write(nd
.path
.mnt
);
3372 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3373 struct path
*path
, unsigned int lookup_flags
)
3375 struct filename
*filename
= getname_kernel(pathname
);
3378 if (IS_ERR(filename
))
3379 return ERR_CAST(filename
);
3380 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3384 EXPORT_SYMBOL(kern_path_create
);
3386 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3389 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3390 mnt_drop_write(path
->mnt
);
3393 EXPORT_SYMBOL(done_path_create
);
3395 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3396 struct path
*path
, unsigned int lookup_flags
)
3398 struct filename
*tmp
= getname(pathname
);
3401 return ERR_CAST(tmp
);
3402 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3406 EXPORT_SYMBOL(user_path_create
);
3408 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3410 int error
= may_create(dir
, dentry
);
3415 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3418 if (!dir
->i_op
->mknod
)
3421 error
= devcgroup_inode_mknod(mode
, dev
);
3425 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3429 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3431 fsnotify_create(dir
, dentry
);
3434 EXPORT_SYMBOL(vfs_mknod
);
3436 static int may_mknod(umode_t mode
)
3438 switch (mode
& S_IFMT
) {
3444 case 0: /* zero mode translates to S_IFREG */
3453 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3456 struct dentry
*dentry
;
3459 unsigned int lookup_flags
= 0;
3461 error
= may_mknod(mode
);
3465 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3467 return PTR_ERR(dentry
);
3469 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3470 mode
&= ~current_umask();
3471 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3474 switch (mode
& S_IFMT
) {
3475 case 0: case S_IFREG
:
3476 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3478 case S_IFCHR
: case S_IFBLK
:
3479 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3480 new_decode_dev(dev
));
3482 case S_IFIFO
: case S_IFSOCK
:
3483 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3487 done_path_create(&path
, dentry
);
3488 if (retry_estale(error
, lookup_flags
)) {
3489 lookup_flags
|= LOOKUP_REVAL
;
3495 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3497 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3500 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3502 int error
= may_create(dir
, dentry
);
3503 unsigned max_links
= dir
->i_sb
->s_max_links
;
3508 if (!dir
->i_op
->mkdir
)
3511 mode
&= (S_IRWXUGO
|S_ISVTX
);
3512 error
= security_inode_mkdir(dir
, dentry
, mode
);
3516 if (max_links
&& dir
->i_nlink
>= max_links
)
3519 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3521 fsnotify_mkdir(dir
, dentry
);
3524 EXPORT_SYMBOL(vfs_mkdir
);
3526 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3528 struct dentry
*dentry
;
3531 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3534 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3536 return PTR_ERR(dentry
);
3538 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3539 mode
&= ~current_umask();
3540 error
= security_path_mkdir(&path
, dentry
, mode
);
3542 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3543 done_path_create(&path
, dentry
);
3544 if (retry_estale(error
, lookup_flags
)) {
3545 lookup_flags
|= LOOKUP_REVAL
;
3551 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3553 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3557 * The dentry_unhash() helper will try to drop the dentry early: we
3558 * should have a usage count of 1 if we're the only user of this
3559 * dentry, and if that is true (possibly after pruning the dcache),
3560 * then we drop the dentry now.
3562 * A low-level filesystem can, if it choses, legally
3565 * if (!d_unhashed(dentry))
3568 * if it cannot handle the case of removing a directory
3569 * that is still in use by something else..
3571 void dentry_unhash(struct dentry
*dentry
)
3573 shrink_dcache_parent(dentry
);
3574 spin_lock(&dentry
->d_lock
);
3575 if (dentry
->d_lockref
.count
== 1)
3577 spin_unlock(&dentry
->d_lock
);
3579 EXPORT_SYMBOL(dentry_unhash
);
3581 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3583 int error
= may_delete(dir
, dentry
, 1);
3588 if (!dir
->i_op
->rmdir
)
3592 mutex_lock(&dentry
->d_inode
->i_mutex
);
3595 if (is_local_mountpoint(dentry
))
3598 error
= security_inode_rmdir(dir
, dentry
);
3602 shrink_dcache_parent(dentry
);
3603 error
= dir
->i_op
->rmdir(dir
, dentry
);
3607 dentry
->d_inode
->i_flags
|= S_DEAD
;
3609 detach_mounts(dentry
);
3612 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3618 EXPORT_SYMBOL(vfs_rmdir
);
3620 static long do_rmdir(int dfd
, const char __user
*pathname
)
3623 struct filename
*name
;
3624 struct dentry
*dentry
;
3628 unsigned int lookup_flags
= 0;
3630 name
= user_path_parent(dfd
, pathname
,
3631 &path
, &last
, &type
, lookup_flags
);
3633 return PTR_ERR(name
);
3647 error
= mnt_want_write(path
.mnt
);
3651 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3652 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3653 error
= PTR_ERR(dentry
);
3656 if (!dentry
->d_inode
) {
3660 error
= security_path_rmdir(&path
, dentry
);
3663 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3667 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3668 mnt_drop_write(path
.mnt
);
3672 if (retry_estale(error
, lookup_flags
)) {
3673 lookup_flags
|= LOOKUP_REVAL
;
3679 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3681 return do_rmdir(AT_FDCWD
, pathname
);
3685 * vfs_unlink - unlink a filesystem object
3686 * @dir: parent directory
3688 * @delegated_inode: returns victim inode, if the inode is delegated.
3690 * The caller must hold dir->i_mutex.
3692 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3693 * return a reference to the inode in delegated_inode. The caller
3694 * should then break the delegation on that inode and retry. Because
3695 * breaking a delegation may take a long time, the caller should drop
3696 * dir->i_mutex before doing so.
3698 * Alternatively, a caller may pass NULL for delegated_inode. This may
3699 * be appropriate for callers that expect the underlying filesystem not
3700 * to be NFS exported.
3702 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3704 struct inode
*target
= dentry
->d_inode
;
3705 int error
= may_delete(dir
, dentry
, 0);
3710 if (!dir
->i_op
->unlink
)
3713 mutex_lock(&target
->i_mutex
);
3714 if (is_local_mountpoint(dentry
))
3717 error
= security_inode_unlink(dir
, dentry
);
3719 error
= try_break_deleg(target
, delegated_inode
);
3722 error
= dir
->i_op
->unlink(dir
, dentry
);
3725 detach_mounts(dentry
);
3730 mutex_unlock(&target
->i_mutex
);
3732 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3733 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3734 fsnotify_link_count(target
);
3740 EXPORT_SYMBOL(vfs_unlink
);
3743 * Make sure that the actual truncation of the file will occur outside its
3744 * directory's i_mutex. Truncate can take a long time if there is a lot of
3745 * writeout happening, and we don't want to prevent access to the directory
3746 * while waiting on the I/O.
3748 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3751 struct filename
*name
;
3752 struct dentry
*dentry
;
3756 struct inode
*inode
= NULL
;
3757 struct inode
*delegated_inode
= NULL
;
3758 unsigned int lookup_flags
= 0;
3760 name
= user_path_parent(dfd
, pathname
,
3761 &path
, &last
, &type
, lookup_flags
);
3763 return PTR_ERR(name
);
3766 if (type
!= LAST_NORM
)
3769 error
= mnt_want_write(path
.mnt
);
3773 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3774 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3775 error
= PTR_ERR(dentry
);
3776 if (!IS_ERR(dentry
)) {
3777 /* Why not before? Because we want correct error value */
3778 if (last
.name
[last
.len
])
3780 inode
= dentry
->d_inode
;
3781 if (d_is_negative(dentry
))
3784 error
= security_path_unlink(&path
, dentry
);
3787 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3791 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3793 iput(inode
); /* truncate the inode here */
3795 if (delegated_inode
) {
3796 error
= break_deleg_wait(&delegated_inode
);
3800 mnt_drop_write(path
.mnt
);
3804 if (retry_estale(error
, lookup_flags
)) {
3805 lookup_flags
|= LOOKUP_REVAL
;
3812 if (d_is_negative(dentry
))
3814 else if (d_is_dir(dentry
))
3821 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3823 if ((flag
& ~AT_REMOVEDIR
) != 0)
3826 if (flag
& AT_REMOVEDIR
)
3827 return do_rmdir(dfd
, pathname
);
3829 return do_unlinkat(dfd
, pathname
);
3832 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3834 return do_unlinkat(AT_FDCWD
, pathname
);
3837 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3839 int error
= may_create(dir
, dentry
);
3844 if (!dir
->i_op
->symlink
)
3847 error
= security_inode_symlink(dir
, dentry
, oldname
);
3851 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3853 fsnotify_create(dir
, dentry
);
3856 EXPORT_SYMBOL(vfs_symlink
);
3858 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3859 int, newdfd
, const char __user
*, newname
)
3862 struct filename
*from
;
3863 struct dentry
*dentry
;
3865 unsigned int lookup_flags
= 0;
3867 from
= getname(oldname
);
3869 return PTR_ERR(from
);
3871 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3872 error
= PTR_ERR(dentry
);
3876 error
= security_path_symlink(&path
, dentry
, from
->name
);
3878 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3879 done_path_create(&path
, dentry
);
3880 if (retry_estale(error
, lookup_flags
)) {
3881 lookup_flags
|= LOOKUP_REVAL
;
3889 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3891 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3895 * vfs_link - create a new link
3896 * @old_dentry: object to be linked
3898 * @new_dentry: where to create the new link
3899 * @delegated_inode: returns inode needing a delegation break
3901 * The caller must hold dir->i_mutex
3903 * If vfs_link discovers a delegation on the to-be-linked file in need
3904 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3905 * inode in delegated_inode. The caller should then break the delegation
3906 * and retry. Because breaking a delegation may take a long time, the
3907 * caller should drop the i_mutex before doing so.
3909 * Alternatively, a caller may pass NULL for delegated_inode. This may
3910 * be appropriate for callers that expect the underlying filesystem not
3911 * to be NFS exported.
3913 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3915 struct inode
*inode
= old_dentry
->d_inode
;
3916 unsigned max_links
= dir
->i_sb
->s_max_links
;
3922 error
= may_create(dir
, new_dentry
);
3926 if (dir
->i_sb
!= inode
->i_sb
)
3930 * A link to an append-only or immutable file cannot be created.
3932 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3934 if (!dir
->i_op
->link
)
3936 if (S_ISDIR(inode
->i_mode
))
3939 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3943 mutex_lock(&inode
->i_mutex
);
3944 /* Make sure we don't allow creating hardlink to an unlinked file */
3945 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3947 else if (max_links
&& inode
->i_nlink
>= max_links
)
3950 error
= try_break_deleg(inode
, delegated_inode
);
3952 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3955 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3956 spin_lock(&inode
->i_lock
);
3957 inode
->i_state
&= ~I_LINKABLE
;
3958 spin_unlock(&inode
->i_lock
);
3960 mutex_unlock(&inode
->i_mutex
);
3962 fsnotify_link(dir
, inode
, new_dentry
);
3965 EXPORT_SYMBOL(vfs_link
);
3968 * Hardlinks are often used in delicate situations. We avoid
3969 * security-related surprises by not following symlinks on the
3972 * We don't follow them on the oldname either to be compatible
3973 * with linux 2.0, and to avoid hard-linking to directories
3974 * and other special files. --ADM
3976 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3977 int, newdfd
, const char __user
*, newname
, int, flags
)
3979 struct dentry
*new_dentry
;
3980 struct path old_path
, new_path
;
3981 struct inode
*delegated_inode
= NULL
;
3985 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3988 * To use null names we require CAP_DAC_READ_SEARCH
3989 * This ensures that not everyone will be able to create
3990 * handlink using the passed filedescriptor.
3992 if (flags
& AT_EMPTY_PATH
) {
3993 if (!capable(CAP_DAC_READ_SEARCH
))
3998 if (flags
& AT_SYMLINK_FOLLOW
)
3999 how
|= LOOKUP_FOLLOW
;
4001 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4005 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4006 (how
& LOOKUP_REVAL
));
4007 error
= PTR_ERR(new_dentry
);
4008 if (IS_ERR(new_dentry
))
4012 if (old_path
.mnt
!= new_path
.mnt
)
4014 error
= may_linkat(&old_path
);
4015 if (unlikely(error
))
4017 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4020 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4022 done_path_create(&new_path
, new_dentry
);
4023 if (delegated_inode
) {
4024 error
= break_deleg_wait(&delegated_inode
);
4026 path_put(&old_path
);
4030 if (retry_estale(error
, how
)) {
4031 path_put(&old_path
);
4032 how
|= LOOKUP_REVAL
;
4036 path_put(&old_path
);
4041 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4043 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4047 * vfs_rename - rename a filesystem object
4048 * @old_dir: parent of source
4049 * @old_dentry: source
4050 * @new_dir: parent of destination
4051 * @new_dentry: destination
4052 * @delegated_inode: returns an inode needing a delegation break
4053 * @flags: rename flags
4055 * The caller must hold multiple mutexes--see lock_rename()).
4057 * If vfs_rename discovers a delegation in need of breaking at either
4058 * the source or destination, it will return -EWOULDBLOCK and return a
4059 * reference to the inode in delegated_inode. The caller should then
4060 * break the delegation and retry. Because breaking a delegation may
4061 * take a long time, the caller should drop all locks before doing
4064 * Alternatively, a caller may pass NULL for delegated_inode. This may
4065 * be appropriate for callers that expect the underlying filesystem not
4066 * to be NFS exported.
4068 * The worst of all namespace operations - renaming directory. "Perverted"
4069 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4071 * a) we can get into loop creation.
4072 * b) race potential - two innocent renames can create a loop together.
4073 * That's where 4.4 screws up. Current fix: serialization on
4074 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4076 * c) we have to lock _four_ objects - parents and victim (if it exists),
4077 * and source (if it is not a directory).
4078 * And that - after we got ->i_mutex on parents (until then we don't know
4079 * whether the target exists). Solution: try to be smart with locking
4080 * order for inodes. We rely on the fact that tree topology may change
4081 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4082 * move will be locked. Thus we can rank directories by the tree
4083 * (ancestors first) and rank all non-directories after them.
4084 * That works since everybody except rename does "lock parent, lookup,
4085 * lock child" and rename is under ->s_vfs_rename_mutex.
4086 * HOWEVER, it relies on the assumption that any object with ->lookup()
4087 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4088 * we'd better make sure that there's no link(2) for them.
4089 * d) conversion from fhandle to dentry may come in the wrong moment - when
4090 * we are removing the target. Solution: we will have to grab ->i_mutex
4091 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4092 * ->i_mutex on parents, which works but leads to some truly excessive
4095 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4096 struct inode
*new_dir
, struct dentry
*new_dentry
,
4097 struct inode
**delegated_inode
, unsigned int flags
)
4100 bool is_dir
= d_is_dir(old_dentry
);
4101 const unsigned char *old_name
;
4102 struct inode
*source
= old_dentry
->d_inode
;
4103 struct inode
*target
= new_dentry
->d_inode
;
4104 bool new_is_dir
= false;
4105 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4107 if (source
== target
)
4110 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4115 error
= may_create(new_dir
, new_dentry
);
4117 new_is_dir
= d_is_dir(new_dentry
);
4119 if (!(flags
& RENAME_EXCHANGE
))
4120 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4122 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4127 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4130 if (flags
&& !old_dir
->i_op
->rename2
)
4134 * If we are going to change the parent - check write permissions,
4135 * we'll need to flip '..'.
4137 if (new_dir
!= old_dir
) {
4139 error
= inode_permission(source
, MAY_WRITE
);
4143 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4144 error
= inode_permission(target
, MAY_WRITE
);
4150 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4155 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4157 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4158 lock_two_nondirectories(source
, target
);
4160 mutex_lock(&target
->i_mutex
);
4163 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4166 if (max_links
&& new_dir
!= old_dir
) {
4168 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4170 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4171 old_dir
->i_nlink
>= max_links
)
4174 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4175 shrink_dcache_parent(new_dentry
);
4177 error
= try_break_deleg(source
, delegated_inode
);
4181 if (target
&& !new_is_dir
) {
4182 error
= try_break_deleg(target
, delegated_inode
);
4186 if (!old_dir
->i_op
->rename2
) {
4187 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4188 new_dir
, new_dentry
);
4190 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4191 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4192 new_dir
, new_dentry
, flags
);
4197 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4199 target
->i_flags
|= S_DEAD
;
4200 dont_mount(new_dentry
);
4201 detach_mounts(new_dentry
);
4203 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4204 if (!(flags
& RENAME_EXCHANGE
))
4205 d_move(old_dentry
, new_dentry
);
4207 d_exchange(old_dentry
, new_dentry
);
4210 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4211 unlock_two_nondirectories(source
, target
);
4213 mutex_unlock(&target
->i_mutex
);
4216 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4217 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4218 if (flags
& RENAME_EXCHANGE
) {
4219 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4220 new_is_dir
, NULL
, new_dentry
);
4223 fsnotify_oldname_free(old_name
);
4227 EXPORT_SYMBOL(vfs_rename
);
4229 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4230 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4232 struct dentry
*old_dentry
, *new_dentry
;
4233 struct dentry
*trap
;
4234 struct path old_path
, new_path
;
4235 struct qstr old_last
, new_last
;
4236 int old_type
, new_type
;
4237 struct inode
*delegated_inode
= NULL
;
4238 struct filename
*from
;
4239 struct filename
*to
;
4240 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4241 bool should_retry
= false;
4244 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4247 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4248 (flags
& RENAME_EXCHANGE
))
4251 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4254 if (flags
& RENAME_EXCHANGE
)
4258 from
= user_path_parent(olddfd
, oldname
,
4259 &old_path
, &old_last
, &old_type
, lookup_flags
);
4261 error
= PTR_ERR(from
);
4265 to
= user_path_parent(newdfd
, newname
,
4266 &new_path
, &new_last
, &new_type
, lookup_flags
);
4268 error
= PTR_ERR(to
);
4273 if (old_path
.mnt
!= new_path
.mnt
)
4277 if (old_type
!= LAST_NORM
)
4280 if (flags
& RENAME_NOREPLACE
)
4282 if (new_type
!= LAST_NORM
)
4285 error
= mnt_want_write(old_path
.mnt
);
4290 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4292 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4293 error
= PTR_ERR(old_dentry
);
4294 if (IS_ERR(old_dentry
))
4296 /* source must exist */
4298 if (d_is_negative(old_dentry
))
4300 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4301 error
= PTR_ERR(new_dentry
);
4302 if (IS_ERR(new_dentry
))
4305 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4307 if (flags
& RENAME_EXCHANGE
) {
4309 if (d_is_negative(new_dentry
))
4312 if (!d_is_dir(new_dentry
)) {
4314 if (new_last
.name
[new_last
.len
])
4318 /* unless the source is a directory trailing slashes give -ENOTDIR */
4319 if (!d_is_dir(old_dentry
)) {
4321 if (old_last
.name
[old_last
.len
])
4323 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4326 /* source should not be ancestor of target */
4328 if (old_dentry
== trap
)
4330 /* target should not be an ancestor of source */
4331 if (!(flags
& RENAME_EXCHANGE
))
4333 if (new_dentry
== trap
)
4336 error
= security_path_rename(&old_path
, old_dentry
,
4337 &new_path
, new_dentry
, flags
);
4340 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4341 new_path
.dentry
->d_inode
, new_dentry
,
4342 &delegated_inode
, flags
);
4348 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4349 if (delegated_inode
) {
4350 error
= break_deleg_wait(&delegated_inode
);
4354 mnt_drop_write(old_path
.mnt
);
4356 if (retry_estale(error
, lookup_flags
))
4357 should_retry
= true;
4358 path_put(&new_path
);
4361 path_put(&old_path
);
4364 should_retry
= false;
4365 lookup_flags
|= LOOKUP_REVAL
;
4372 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4373 int, newdfd
, const char __user
*, newname
)
4375 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4378 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4380 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4383 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4385 int error
= may_create(dir
, dentry
);
4389 if (!dir
->i_op
->mknod
)
4392 return dir
->i_op
->mknod(dir
, dentry
,
4393 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4395 EXPORT_SYMBOL(vfs_whiteout
);
4397 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4399 int len
= PTR_ERR(link
);
4404 if (len
> (unsigned) buflen
)
4406 if (copy_to_user(buffer
, link
, len
))
4411 EXPORT_SYMBOL(readlink_copy
);
4414 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4415 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4416 * using) it for any given inode is up to filesystem.
4418 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4421 const char *link
= dentry
->d_inode
->i_op
->follow_link(dentry
, &cookie
, NULL
);
4425 return PTR_ERR(link
);
4426 res
= readlink_copy(buffer
, buflen
, link
);
4427 if (cookie
&& dentry
->d_inode
->i_op
->put_link
)
4428 dentry
->d_inode
->i_op
->put_link(dentry
, cookie
);
4431 EXPORT_SYMBOL(generic_readlink
);
4433 /* get the link contents into pagecache */
4434 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4438 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4439 page
= read_mapping_page(mapping
, 0, NULL
);
4444 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4448 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4450 struct page
*page
= NULL
;
4451 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4454 page_cache_release(page
);
4458 EXPORT_SYMBOL(page_readlink
);
4460 const char *page_follow_link_light(struct dentry
*dentry
, void **cookie
, struct nameidata
*nd
)
4462 struct page
*page
= NULL
;
4463 char *res
= page_getlink(dentry
, &page
);
4468 EXPORT_SYMBOL(page_follow_link_light
);
4470 void page_put_link(struct dentry
*dentry
, void *cookie
)
4472 struct page
*page
= cookie
;
4474 page_cache_release(page
);
4476 EXPORT_SYMBOL(page_put_link
);
4479 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4481 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4483 struct address_space
*mapping
= inode
->i_mapping
;
4488 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4490 flags
|= AOP_FLAG_NOFS
;
4493 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4494 flags
, &page
, &fsdata
);
4498 kaddr
= kmap_atomic(page
);
4499 memcpy(kaddr
, symname
, len
-1);
4500 kunmap_atomic(kaddr
);
4502 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4509 mark_inode_dirty(inode
);
4514 EXPORT_SYMBOL(__page_symlink
);
4516 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4518 return __page_symlink(inode
, symname
, len
,
4519 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4521 EXPORT_SYMBOL(page_symlink
);
4523 const struct inode_operations page_symlink_inode_operations
= {
4524 .readlink
= generic_readlink
,
4525 .follow_link
= page_follow_link_light
,
4526 .put_link
= page_put_link
,
4528 EXPORT_SYMBOL(page_symlink_inode_operations
);