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 */
505 char *saved_names
[MAX_NESTED_LINKS
+ 1];
509 * Path walking has 2 modes, rcu-walk and ref-walk (see
510 * Documentation/filesystems/path-lookup.txt). In situations when we can't
511 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
512 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
513 * mode. Refcounts are grabbed at the last known good point before rcu-walk
514 * got stuck, so ref-walk may continue from there. If this is not successful
515 * (eg. a seqcount has changed), then failure is returned and it's up to caller
516 * to restart the path walk from the beginning in ref-walk mode.
520 * unlazy_walk - try to switch to ref-walk mode.
521 * @nd: nameidata pathwalk data
522 * @dentry: child of nd->path.dentry or NULL
523 * Returns: 0 on success, -ECHILD on failure
525 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
526 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
527 * @nd or NULL. Must be called from rcu-walk context.
529 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
531 struct fs_struct
*fs
= current
->fs
;
532 struct dentry
*parent
= nd
->path
.dentry
;
534 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
537 * After legitimizing the bastards, terminate_walk()
538 * will do the right thing for non-RCU mode, and all our
539 * subsequent exit cases should rcu_read_unlock()
540 * before returning. Do vfsmount first; if dentry
541 * can't be legitimized, just set nd->path.dentry to NULL
542 * and rely on dput(NULL) being a no-op.
544 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
546 nd
->flags
&= ~LOOKUP_RCU
;
548 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
549 nd
->path
.dentry
= NULL
;
554 * For a negative lookup, the lookup sequence point is the parents
555 * sequence point, and it only needs to revalidate the parent dentry.
557 * For a positive lookup, we need to move both the parent and the
558 * dentry from the RCU domain to be properly refcounted. And the
559 * sequence number in the dentry validates *both* dentry counters,
560 * since we checked the sequence number of the parent after we got
561 * the child sequence number. So we know the parent must still
562 * be valid if the child sequence number is still valid.
565 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
567 BUG_ON(nd
->inode
!= parent
->d_inode
);
569 if (!lockref_get_not_dead(&dentry
->d_lockref
))
571 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
576 * Sequence counts matched. Now make sure that the root is
577 * still valid and get it if required.
579 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
580 spin_lock(&fs
->lock
);
581 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
582 goto unlock_and_drop_dentry
;
584 spin_unlock(&fs
->lock
);
590 unlock_and_drop_dentry
:
591 spin_unlock(&fs
->lock
);
599 if (!(nd
->flags
& LOOKUP_ROOT
))
604 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
606 return dentry
->d_op
->d_revalidate(dentry
, flags
);
610 * complete_walk - successful completion of path walk
611 * @nd: pointer nameidata
613 * If we had been in RCU mode, drop out of it and legitimize nd->path.
614 * Revalidate the final result, unless we'd already done that during
615 * the path walk or the filesystem doesn't ask for it. Return 0 on
616 * success, -error on failure. In case of failure caller does not
617 * need to drop nd->path.
619 static int complete_walk(struct nameidata
*nd
)
621 struct dentry
*dentry
= nd
->path
.dentry
;
624 if (nd
->flags
& LOOKUP_RCU
) {
625 nd
->flags
&= ~LOOKUP_RCU
;
626 if (!(nd
->flags
& LOOKUP_ROOT
))
629 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
633 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
635 mntput(nd
->path
.mnt
);
638 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
641 mntput(nd
->path
.mnt
);
647 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
650 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
653 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
664 static __always_inline
void set_root(struct nameidata
*nd
)
666 get_fs_root(current
->fs
, &nd
->root
);
669 static int link_path_walk(const char *, struct nameidata
*);
671 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
673 struct fs_struct
*fs
= current
->fs
;
677 seq
= read_seqcount_begin(&fs
->seq
);
679 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
680 } while (read_seqcount_retry(&fs
->seq
, seq
));
684 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
687 if (path
->mnt
!= nd
->path
.mnt
)
691 static inline void path_to_nameidata(const struct path
*path
,
692 struct nameidata
*nd
)
694 if (!(nd
->flags
& LOOKUP_RCU
)) {
695 dput(nd
->path
.dentry
);
696 if (nd
->path
.mnt
!= path
->mnt
)
697 mntput(nd
->path
.mnt
);
699 nd
->path
.mnt
= path
->mnt
;
700 nd
->path
.dentry
= path
->dentry
;
704 * Helper to directly jump to a known parsed path from ->follow_link,
705 * caller must have taken a reference to path beforehand.
707 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
712 nd
->inode
= nd
->path
.dentry
->d_inode
;
713 nd
->flags
|= LOOKUP_JUMPED
;
716 void nd_set_link(struct nameidata
*nd
, char *path
)
718 nd
->saved_names
[nd
->depth
] = path
;
720 EXPORT_SYMBOL(nd_set_link
);
722 char *nd_get_link(struct nameidata
*nd
)
724 return nd
->saved_names
[nd
->depth
];
726 EXPORT_SYMBOL(nd_get_link
);
728 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
730 struct inode
*inode
= link
->dentry
->d_inode
;
731 if (inode
->i_op
->put_link
)
732 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
736 int sysctl_protected_symlinks __read_mostly
= 0;
737 int sysctl_protected_hardlinks __read_mostly
= 0;
740 * may_follow_link - Check symlink following for unsafe situations
741 * @link: The path of the symlink
742 * @nd: nameidata pathwalk data
744 * In the case of the sysctl_protected_symlinks sysctl being enabled,
745 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
746 * in a sticky world-writable directory. This is to protect privileged
747 * processes from failing races against path names that may change out
748 * from under them by way of other users creating malicious symlinks.
749 * It will permit symlinks to be followed only when outside a sticky
750 * world-writable directory, or when the uid of the symlink and follower
751 * match, or when the directory owner matches the symlink's owner.
753 * Returns 0 if following the symlink is allowed, -ve on error.
755 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
757 const struct inode
*inode
;
758 const struct inode
*parent
;
760 if (!sysctl_protected_symlinks
)
763 /* Allowed if owner and follower match. */
764 inode
= link
->dentry
->d_inode
;
765 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
768 /* Allowed if parent directory not sticky and world-writable. */
769 parent
= nd
->path
.dentry
->d_inode
;
770 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
773 /* Allowed if parent directory and link owner match. */
774 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
777 audit_log_link_denied("follow_link", link
);
778 path_put_conditional(link
, nd
);
784 * safe_hardlink_source - Check for safe hardlink conditions
785 * @inode: the source inode to hardlink from
787 * Return false if at least one of the following conditions:
788 * - inode is not a regular file
790 * - inode is setgid and group-exec
791 * - access failure for read and write
793 * Otherwise returns true.
795 static bool safe_hardlink_source(struct inode
*inode
)
797 umode_t mode
= inode
->i_mode
;
799 /* Special files should not get pinned to the filesystem. */
803 /* Setuid files should not get pinned to the filesystem. */
807 /* Executable setgid files should not get pinned to the filesystem. */
808 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
811 /* Hardlinking to unreadable or unwritable sources is dangerous. */
812 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
819 * may_linkat - Check permissions for creating a hardlink
820 * @link: the source to hardlink from
822 * Block hardlink when all of:
823 * - sysctl_protected_hardlinks enabled
824 * - fsuid does not match inode
825 * - hardlink source is unsafe (see safe_hardlink_source() above)
828 * Returns 0 if successful, -ve on error.
830 static int may_linkat(struct path
*link
)
832 const struct cred
*cred
;
835 if (!sysctl_protected_hardlinks
)
838 cred
= current_cred();
839 inode
= link
->dentry
->d_inode
;
841 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
842 * otherwise, it must be a safe source.
844 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
848 audit_log_link_denied("linkat", link
);
852 static __always_inline
int
853 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
855 struct dentry
*dentry
= link
->dentry
;
859 BUG_ON(nd
->flags
& LOOKUP_RCU
);
861 if (link
->mnt
== nd
->path
.mnt
)
865 if (unlikely(current
->total_link_count
>= 40))
866 goto out_put_nd_path
;
869 current
->total_link_count
++;
872 nd_set_link(nd
, NULL
);
874 error
= security_inode_follow_link(link
->dentry
, nd
);
876 goto out_put_nd_path
;
878 nd
->last_type
= LAST_BIND
;
879 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
882 goto out_put_nd_path
;
887 if (unlikely(IS_ERR(s
))) {
889 put_link(nd
, link
, *p
);
898 nd
->flags
|= LOOKUP_JUMPED
;
900 nd
->inode
= nd
->path
.dentry
->d_inode
;
901 error
= link_path_walk(s
, nd
);
903 put_link(nd
, link
, *p
);
915 static int follow_up_rcu(struct path
*path
)
917 struct mount
*mnt
= real_mount(path
->mnt
);
918 struct mount
*parent
;
919 struct dentry
*mountpoint
;
921 parent
= mnt
->mnt_parent
;
922 if (&parent
->mnt
== path
->mnt
)
924 mountpoint
= mnt
->mnt_mountpoint
;
925 path
->dentry
= mountpoint
;
926 path
->mnt
= &parent
->mnt
;
931 * follow_up - Find the mountpoint of path's vfsmount
933 * Given a path, find the mountpoint of its source file system.
934 * Replace @path with the path of the mountpoint in the parent mount.
937 * Return 1 if we went up a level and 0 if we were already at the
940 int follow_up(struct path
*path
)
942 struct mount
*mnt
= real_mount(path
->mnt
);
943 struct mount
*parent
;
944 struct dentry
*mountpoint
;
946 read_seqlock_excl(&mount_lock
);
947 parent
= mnt
->mnt_parent
;
949 read_sequnlock_excl(&mount_lock
);
952 mntget(&parent
->mnt
);
953 mountpoint
= dget(mnt
->mnt_mountpoint
);
954 read_sequnlock_excl(&mount_lock
);
956 path
->dentry
= mountpoint
;
958 path
->mnt
= &parent
->mnt
;
961 EXPORT_SYMBOL(follow_up
);
964 * Perform an automount
965 * - return -EISDIR to tell follow_managed() to stop and return the path we
968 static int follow_automount(struct path
*path
, unsigned flags
,
971 struct vfsmount
*mnt
;
974 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
977 /* We don't want to mount if someone's just doing a stat -
978 * unless they're stat'ing a directory and appended a '/' to
981 * We do, however, want to mount if someone wants to open or
982 * create a file of any type under the mountpoint, wants to
983 * traverse through the mountpoint or wants to open the
984 * mounted directory. Also, autofs may mark negative dentries
985 * as being automount points. These will need the attentions
986 * of the daemon to instantiate them before they can be used.
988 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
989 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
990 path
->dentry
->d_inode
)
993 current
->total_link_count
++;
994 if (current
->total_link_count
>= 40)
997 mnt
= path
->dentry
->d_op
->d_automount(path
);
1000 * The filesystem is allowed to return -EISDIR here to indicate
1001 * it doesn't want to automount. For instance, autofs would do
1002 * this so that its userspace daemon can mount on this dentry.
1004 * However, we can only permit this if it's a terminal point in
1005 * the path being looked up; if it wasn't then the remainder of
1006 * the path is inaccessible and we should say so.
1008 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1010 return PTR_ERR(mnt
);
1013 if (!mnt
) /* mount collision */
1016 if (!*need_mntput
) {
1017 /* lock_mount() may release path->mnt on error */
1019 *need_mntput
= true;
1021 err
= finish_automount(mnt
, path
);
1025 /* Someone else made a mount here whilst we were busy */
1030 path
->dentry
= dget(mnt
->mnt_root
);
1039 * Handle a dentry that is managed in some way.
1040 * - Flagged for transit management (autofs)
1041 * - Flagged as mountpoint
1042 * - Flagged as automount point
1044 * This may only be called in refwalk mode.
1046 * Serialization is taken care of in namespace.c
1048 static int follow_managed(struct path
*path
, unsigned flags
)
1050 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1052 bool need_mntput
= false;
1055 /* Given that we're not holding a lock here, we retain the value in a
1056 * local variable for each dentry as we look at it so that we don't see
1057 * the components of that value change under us */
1058 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1059 managed
&= DCACHE_MANAGED_DENTRY
,
1060 unlikely(managed
!= 0)) {
1061 /* Allow the filesystem to manage the transit without i_mutex
1063 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1064 BUG_ON(!path
->dentry
->d_op
);
1065 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1066 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1071 /* Transit to a mounted filesystem. */
1072 if (managed
& DCACHE_MOUNTED
) {
1073 struct vfsmount
*mounted
= lookup_mnt(path
);
1078 path
->mnt
= mounted
;
1079 path
->dentry
= dget(mounted
->mnt_root
);
1084 /* Something is mounted on this dentry in another
1085 * namespace and/or whatever was mounted there in this
1086 * namespace got unmounted before lookup_mnt() could
1090 /* Handle an automount point */
1091 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1092 ret
= follow_automount(path
, flags
, &need_mntput
);
1098 /* We didn't change the current path point */
1102 if (need_mntput
&& path
->mnt
== mnt
)
1106 return ret
< 0 ? ret
: need_mntput
;
1109 int follow_down_one(struct path
*path
)
1111 struct vfsmount
*mounted
;
1113 mounted
= lookup_mnt(path
);
1117 path
->mnt
= mounted
;
1118 path
->dentry
= dget(mounted
->mnt_root
);
1123 EXPORT_SYMBOL(follow_down_one
);
1125 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1127 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1128 dentry
->d_op
->d_manage(dentry
, true) : 0;
1132 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1133 * we meet a managed dentry that would need blocking.
1135 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1136 struct inode
**inode
)
1139 struct mount
*mounted
;
1141 * Don't forget we might have a non-mountpoint managed dentry
1142 * that wants to block transit.
1144 switch (managed_dentry_rcu(path
->dentry
)) {
1154 if (!d_mountpoint(path
->dentry
))
1155 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1157 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1160 path
->mnt
= &mounted
->mnt
;
1161 path
->dentry
= mounted
->mnt
.mnt_root
;
1162 nd
->flags
|= LOOKUP_JUMPED
;
1163 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1165 * Update the inode too. We don't need to re-check the
1166 * dentry sequence number here after this d_inode read,
1167 * because a mount-point is always pinned.
1169 *inode
= path
->dentry
->d_inode
;
1171 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1172 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1175 static int follow_dotdot_rcu(struct nameidata
*nd
)
1177 struct inode
*inode
= nd
->inode
;
1182 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1183 nd
->path
.mnt
== nd
->root
.mnt
) {
1186 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1187 struct dentry
*old
= nd
->path
.dentry
;
1188 struct dentry
*parent
= old
->d_parent
;
1191 inode
= parent
->d_inode
;
1192 seq
= read_seqcount_begin(&parent
->d_seq
);
1193 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1195 nd
->path
.dentry
= parent
;
1199 if (!follow_up_rcu(&nd
->path
))
1201 inode
= nd
->path
.dentry
->d_inode
;
1202 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1204 while (d_mountpoint(nd
->path
.dentry
)) {
1205 struct mount
*mounted
;
1206 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1209 nd
->path
.mnt
= &mounted
->mnt
;
1210 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1211 inode
= nd
->path
.dentry
->d_inode
;
1212 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1213 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1220 nd
->flags
&= ~LOOKUP_RCU
;
1221 if (!(nd
->flags
& LOOKUP_ROOT
))
1222 nd
->root
.mnt
= NULL
;
1228 * Follow down to the covering mount currently visible to userspace. At each
1229 * point, the filesystem owning that dentry may be queried as to whether the
1230 * caller is permitted to proceed or not.
1232 int follow_down(struct path
*path
)
1237 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1238 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1239 /* Allow the filesystem to manage the transit without i_mutex
1242 * We indicate to the filesystem if someone is trying to mount
1243 * something here. This gives autofs the chance to deny anyone
1244 * other than its daemon the right to mount on its
1247 * The filesystem may sleep at this point.
1249 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1250 BUG_ON(!path
->dentry
->d_op
);
1251 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1252 ret
= path
->dentry
->d_op
->d_manage(
1253 path
->dentry
, false);
1255 return ret
== -EISDIR
? 0 : ret
;
1258 /* Transit to a mounted filesystem. */
1259 if (managed
& DCACHE_MOUNTED
) {
1260 struct vfsmount
*mounted
= lookup_mnt(path
);
1265 path
->mnt
= mounted
;
1266 path
->dentry
= dget(mounted
->mnt_root
);
1270 /* Don't handle automount points here */
1275 EXPORT_SYMBOL(follow_down
);
1278 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1280 static void follow_mount(struct path
*path
)
1282 while (d_mountpoint(path
->dentry
)) {
1283 struct vfsmount
*mounted
= lookup_mnt(path
);
1288 path
->mnt
= mounted
;
1289 path
->dentry
= dget(mounted
->mnt_root
);
1293 static void follow_dotdot(struct nameidata
*nd
)
1299 struct dentry
*old
= nd
->path
.dentry
;
1301 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1302 nd
->path
.mnt
== nd
->root
.mnt
) {
1305 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1306 /* rare case of legitimate dget_parent()... */
1307 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1311 if (!follow_up(&nd
->path
))
1314 follow_mount(&nd
->path
);
1315 nd
->inode
= nd
->path
.dentry
->d_inode
;
1319 * This looks up the name in dcache, possibly revalidates the old dentry and
1320 * allocates a new one if not found or not valid. In the need_lookup argument
1321 * returns whether i_op->lookup is necessary.
1323 * dir->d_inode->i_mutex must be held
1325 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1326 unsigned int flags
, bool *need_lookup
)
1328 struct dentry
*dentry
;
1331 *need_lookup
= false;
1332 dentry
= d_lookup(dir
, name
);
1334 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1335 error
= d_revalidate(dentry
, flags
);
1336 if (unlikely(error
<= 0)) {
1339 return ERR_PTR(error
);
1341 d_invalidate(dentry
);
1350 dentry
= d_alloc(dir
, name
);
1351 if (unlikely(!dentry
))
1352 return ERR_PTR(-ENOMEM
);
1354 *need_lookup
= true;
1360 * Call i_op->lookup on the dentry. The dentry must be negative and
1363 * dir->d_inode->i_mutex must be held
1365 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1370 /* Don't create child dentry for a dead directory. */
1371 if (unlikely(IS_DEADDIR(dir
))) {
1373 return ERR_PTR(-ENOENT
);
1376 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1377 if (unlikely(old
)) {
1384 static struct dentry
*__lookup_hash(struct qstr
*name
,
1385 struct dentry
*base
, unsigned int flags
)
1388 struct dentry
*dentry
;
1390 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1394 return lookup_real(base
->d_inode
, dentry
, flags
);
1398 * It's more convoluted than I'd like it to be, but... it's still fairly
1399 * small and for now I'd prefer to have fast path as straight as possible.
1400 * It _is_ time-critical.
1402 static int lookup_fast(struct nameidata
*nd
,
1403 struct path
*path
, struct inode
**inode
)
1405 struct vfsmount
*mnt
= nd
->path
.mnt
;
1406 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1412 * Rename seqlock is not required here because in the off chance
1413 * of a false negative due to a concurrent rename, we're going to
1414 * do the non-racy lookup, below.
1416 if (nd
->flags
& LOOKUP_RCU
) {
1418 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1423 * This sequence count validates that the inode matches
1424 * the dentry name information from lookup.
1426 *inode
= dentry
->d_inode
;
1427 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1431 * This sequence count validates that the parent had no
1432 * changes while we did the lookup of the dentry above.
1434 * The memory barrier in read_seqcount_begin of child is
1435 * enough, we can use __read_seqcount_retry here.
1437 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1441 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1442 status
= d_revalidate(dentry
, nd
->flags
);
1443 if (unlikely(status
<= 0)) {
1444 if (status
!= -ECHILD
)
1450 path
->dentry
= dentry
;
1451 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1454 if (unlazy_walk(nd
, dentry
))
1457 dentry
= __d_lookup(parent
, &nd
->last
);
1460 if (unlikely(!dentry
))
1463 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1464 status
= d_revalidate(dentry
, nd
->flags
);
1465 if (unlikely(status
<= 0)) {
1470 d_invalidate(dentry
);
1476 path
->dentry
= dentry
;
1477 err
= follow_managed(path
, nd
->flags
);
1478 if (unlikely(err
< 0)) {
1479 path_put_conditional(path
, nd
);
1483 nd
->flags
|= LOOKUP_JUMPED
;
1484 *inode
= path
->dentry
->d_inode
;
1491 /* Fast lookup failed, do it the slow way */
1492 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1494 struct dentry
*dentry
, *parent
;
1497 parent
= nd
->path
.dentry
;
1498 BUG_ON(nd
->inode
!= parent
->d_inode
);
1500 mutex_lock(&parent
->d_inode
->i_mutex
);
1501 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1502 mutex_unlock(&parent
->d_inode
->i_mutex
);
1504 return PTR_ERR(dentry
);
1505 path
->mnt
= nd
->path
.mnt
;
1506 path
->dentry
= dentry
;
1507 err
= follow_managed(path
, nd
->flags
);
1508 if (unlikely(err
< 0)) {
1509 path_put_conditional(path
, nd
);
1513 nd
->flags
|= LOOKUP_JUMPED
;
1517 static inline int may_lookup(struct nameidata
*nd
)
1519 if (nd
->flags
& LOOKUP_RCU
) {
1520 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1523 if (unlazy_walk(nd
, NULL
))
1526 return inode_permission(nd
->inode
, MAY_EXEC
);
1529 static inline int handle_dots(struct nameidata
*nd
, int type
)
1531 if (type
== LAST_DOTDOT
) {
1532 if (nd
->flags
& LOOKUP_RCU
) {
1533 if (follow_dotdot_rcu(nd
))
1541 static void terminate_walk(struct nameidata
*nd
)
1543 if (!(nd
->flags
& LOOKUP_RCU
)) {
1544 path_put(&nd
->path
);
1546 nd
->flags
&= ~LOOKUP_RCU
;
1547 if (!(nd
->flags
& LOOKUP_ROOT
))
1548 nd
->root
.mnt
= NULL
;
1554 * Do we need to follow links? We _really_ want to be able
1555 * to do this check without having to look at inode->i_op,
1556 * so we keep a cache of "no, this doesn't need follow_link"
1557 * for the common case.
1559 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1561 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1564 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1567 struct inode
*inode
;
1570 * "." and ".." are special - ".." especially so because it has
1571 * to be able to know about the current root directory and
1572 * parent relationships.
1574 if (unlikely(nd
->last_type
!= LAST_NORM
))
1575 return handle_dots(nd
, nd
->last_type
);
1576 err
= lookup_fast(nd
, path
, &inode
);
1577 if (unlikely(err
)) {
1581 err
= lookup_slow(nd
, path
);
1585 inode
= path
->dentry
->d_inode
;
1588 if (d_is_negative(path
->dentry
))
1591 if (should_follow_link(path
->dentry
, follow
)) {
1592 if (nd
->flags
& LOOKUP_RCU
) {
1593 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
1594 unlazy_walk(nd
, path
->dentry
))) {
1599 BUG_ON(inode
!= path
->dentry
->d_inode
);
1602 path_to_nameidata(path
, nd
);
1607 path_to_nameidata(path
, nd
);
1614 * This limits recursive symlink follows to 8, while
1615 * limiting consecutive symlinks to 40.
1617 * Without that kind of total limit, nasty chains of consecutive
1618 * symlinks can cause almost arbitrarily long lookups.
1620 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1624 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1625 path_put_conditional(path
, nd
);
1626 path_put(&nd
->path
);
1629 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1632 current
->link_count
++;
1635 struct path link
= *path
;
1638 res
= follow_link(&link
, nd
, &cookie
);
1641 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1642 put_link(nd
, &link
, cookie
);
1645 current
->link_count
--;
1651 * We can do the critical dentry name comparison and hashing
1652 * operations one word at a time, but we are limited to:
1654 * - Architectures with fast unaligned word accesses. We could
1655 * do a "get_unaligned()" if this helps and is sufficiently
1658 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1659 * do not trap on the (extremely unlikely) case of a page
1660 * crossing operation.
1662 * - Furthermore, we need an efficient 64-bit compile for the
1663 * 64-bit case in order to generate the "number of bytes in
1664 * the final mask". Again, that could be replaced with a
1665 * efficient population count instruction or similar.
1667 #ifdef CONFIG_DCACHE_WORD_ACCESS
1669 #include <asm/word-at-a-time.h>
1673 static inline unsigned int fold_hash(unsigned long hash
)
1675 return hash_64(hash
, 32);
1678 #else /* 32-bit case */
1680 #define fold_hash(x) (x)
1684 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1686 unsigned long a
, mask
;
1687 unsigned long hash
= 0;
1690 a
= load_unaligned_zeropad(name
);
1691 if (len
< sizeof(unsigned long))
1695 name
+= sizeof(unsigned long);
1696 len
-= sizeof(unsigned long);
1700 mask
= bytemask_from_count(len
);
1703 return fold_hash(hash
);
1705 EXPORT_SYMBOL(full_name_hash
);
1708 * Calculate the length and hash of the path component, and
1709 * return the "hash_len" as the result.
1711 static inline u64
hash_name(const char *name
)
1713 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1714 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1717 len
= -sizeof(unsigned long);
1719 hash
= (hash
+ a
) * 9;
1720 len
+= sizeof(unsigned long);
1721 a
= load_unaligned_zeropad(name
+len
);
1722 b
= a
^ REPEAT_BYTE('/');
1723 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1725 adata
= prep_zero_mask(a
, adata
, &constants
);
1726 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1728 mask
= create_zero_mask(adata
| bdata
);
1730 hash
+= a
& zero_bytemask(mask
);
1731 len
+= find_zero(mask
);
1732 return hashlen_create(fold_hash(hash
), len
);
1737 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1739 unsigned long hash
= init_name_hash();
1741 hash
= partial_name_hash(*name
++, hash
);
1742 return end_name_hash(hash
);
1744 EXPORT_SYMBOL(full_name_hash
);
1747 * We know there's a real path component here of at least
1750 static inline u64
hash_name(const char *name
)
1752 unsigned long hash
= init_name_hash();
1753 unsigned long len
= 0, c
;
1755 c
= (unsigned char)*name
;
1758 hash
= partial_name_hash(c
, hash
);
1759 c
= (unsigned char)name
[len
];
1760 } while (c
&& c
!= '/');
1761 return hashlen_create(end_name_hash(hash
), len
);
1768 * This is the basic name resolution function, turning a pathname into
1769 * the final dentry. We expect 'base' to be positive and a directory.
1771 * Returns 0 and nd will have valid dentry and mnt on success.
1772 * Returns error and drops reference to input namei data on failure.
1774 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1784 /* At this point we know we have a real path component. */
1789 err
= may_lookup(nd
);
1793 hash_len
= hash_name(name
);
1796 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1798 if (name
[1] == '.') {
1800 nd
->flags
|= LOOKUP_JUMPED
;
1806 if (likely(type
== LAST_NORM
)) {
1807 struct dentry
*parent
= nd
->path
.dentry
;
1808 nd
->flags
&= ~LOOKUP_JUMPED
;
1809 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1810 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1811 err
= parent
->d_op
->d_hash(parent
, &this);
1814 hash_len
= this.hash_len
;
1819 nd
->last
.hash_len
= hash_len
;
1820 nd
->last
.name
= name
;
1821 nd
->last_type
= type
;
1823 name
+= hashlen_len(hash_len
);
1827 * If it wasn't NUL, we know it was '/'. Skip that
1828 * slash, and continue until no more slashes.
1832 } while (unlikely(*name
== '/'));
1836 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1841 err
= nested_symlink(&next
, nd
);
1845 if (!d_can_lookup(nd
->path
.dentry
)) {
1854 static int path_init(int dfd
, const struct filename
*name
, unsigned int flags
,
1855 struct nameidata
*nd
)
1858 const char *s
= name
->name
;
1860 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1861 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1864 if (flags
& LOOKUP_ROOT
) {
1865 struct dentry
*root
= nd
->root
.dentry
;
1866 struct inode
*inode
= root
->d_inode
;
1868 if (!d_can_lookup(root
))
1870 retval
= inode_permission(inode
, MAY_EXEC
);
1874 nd
->path
= nd
->root
;
1876 if (flags
& LOOKUP_RCU
) {
1878 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1879 nd
->m_seq
= read_seqbegin(&mount_lock
);
1881 path_get(&nd
->path
);
1886 nd
->root
.mnt
= NULL
;
1888 nd
->m_seq
= read_seqbegin(&mount_lock
);
1890 if (flags
& LOOKUP_RCU
) {
1892 nd
->seq
= set_root_rcu(nd
);
1895 path_get(&nd
->root
);
1897 nd
->path
= nd
->root
;
1898 } else if (dfd
== AT_FDCWD
) {
1899 if (flags
& LOOKUP_RCU
) {
1900 struct fs_struct
*fs
= current
->fs
;
1906 seq
= read_seqcount_begin(&fs
->seq
);
1908 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1909 } while (read_seqcount_retry(&fs
->seq
, seq
));
1911 get_fs_pwd(current
->fs
, &nd
->path
);
1914 /* Caller must check execute permissions on the starting path component */
1915 struct fd f
= fdget_raw(dfd
);
1916 struct dentry
*dentry
;
1921 dentry
= f
.file
->f_path
.dentry
;
1924 if (!d_can_lookup(dentry
)) {
1930 nd
->path
= f
.file
->f_path
;
1931 if (flags
& LOOKUP_RCU
) {
1932 if (f
.flags
& FDPUT_FPUT
)
1934 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1937 path_get(&nd
->path
);
1942 nd
->inode
= nd
->path
.dentry
->d_inode
;
1943 if (!(flags
& LOOKUP_RCU
))
1945 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1947 if (!(nd
->flags
& LOOKUP_ROOT
))
1948 nd
->root
.mnt
= NULL
;
1952 current
->total_link_count
= 0;
1953 return link_path_walk(s
, nd
);
1956 static void path_cleanup(struct nameidata
*nd
)
1958 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1959 path_put(&nd
->root
);
1960 nd
->root
.mnt
= NULL
;
1962 if (unlikely(nd
->base
))
1966 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1968 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1969 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1971 nd
->flags
&= ~LOOKUP_PARENT
;
1972 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1975 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1976 static int path_lookupat(int dfd
, const struct filename
*name
,
1977 unsigned int flags
, struct nameidata
*nd
)
1983 * Path walking is largely split up into 2 different synchronisation
1984 * schemes, rcu-walk and ref-walk (explained in
1985 * Documentation/filesystems/path-lookup.txt). These share much of the
1986 * path walk code, but some things particularly setup, cleanup, and
1987 * following mounts are sufficiently divergent that functions are
1988 * duplicated. Typically there is a function foo(), and its RCU
1989 * analogue, foo_rcu().
1991 * -ECHILD is the error number of choice (just to avoid clashes) that
1992 * is returned if some aspect of an rcu-walk fails. Such an error must
1993 * be handled by restarting a traditional ref-walk (which will always
1994 * be able to complete).
1996 err
= path_init(dfd
, name
, flags
, nd
);
1997 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1998 err
= lookup_last(nd
, &path
);
2001 struct path link
= path
;
2002 err
= may_follow_link(&link
, nd
);
2005 nd
->flags
|= LOOKUP_PARENT
;
2006 err
= follow_link(&link
, nd
, &cookie
);
2009 err
= lookup_last(nd
, &path
);
2010 put_link(nd
, &link
, cookie
);
2015 err
= complete_walk(nd
);
2017 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2018 if (!d_can_lookup(nd
->path
.dentry
)) {
2019 path_put(&nd
->path
);
2028 static int filename_lookup(int dfd
, struct filename
*name
,
2029 unsigned int flags
, struct nameidata
*nd
)
2031 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2032 if (unlikely(retval
== -ECHILD
))
2033 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2034 if (unlikely(retval
== -ESTALE
))
2035 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2037 if (likely(!retval
))
2038 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2042 /* does lookup, returns the object with parent locked */
2043 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2045 struct filename
*filename
= getname_kernel(name
);
2046 struct nameidata nd
;
2050 if (IS_ERR(filename
))
2051 return ERR_CAST(filename
);
2053 err
= filename_lookup(AT_FDCWD
, filename
, LOOKUP_PARENT
, &nd
);
2058 if (nd
.last_type
!= LAST_NORM
) {
2060 d
= ERR_PTR(-EINVAL
);
2063 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2064 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2066 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2076 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2078 struct nameidata nd
;
2079 struct filename
*filename
= getname_kernel(name
);
2080 int res
= PTR_ERR(filename
);
2082 if (!IS_ERR(filename
)) {
2083 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2090 EXPORT_SYMBOL(kern_path
);
2093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2094 * @dentry: pointer to dentry of the base directory
2095 * @mnt: pointer to vfs mount of the base directory
2096 * @name: pointer to file name
2097 * @flags: lookup flags
2098 * @path: pointer to struct path to fill
2100 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2101 const char *name
, unsigned int flags
,
2104 struct filename
*filename
= getname_kernel(name
);
2105 int err
= PTR_ERR(filename
);
2107 BUG_ON(flags
& LOOKUP_PARENT
);
2109 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2110 if (!IS_ERR(filename
)) {
2111 struct nameidata nd
;
2112 nd
.root
.dentry
= dentry
;
2114 err
= filename_lookup(AT_FDCWD
, filename
,
2115 flags
| LOOKUP_ROOT
, &nd
);
2122 EXPORT_SYMBOL(vfs_path_lookup
);
2125 * Restricted form of lookup. Doesn't follow links, single-component only,
2126 * needs parent already locked. Doesn't follow mounts.
2129 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2131 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2135 * lookup_one_len - filesystem helper to lookup single pathname component
2136 * @name: pathname component to lookup
2137 * @base: base directory to lookup from
2138 * @len: maximum length @len should be interpreted to
2140 * Note that this routine is purely a helper for filesystem usage and should
2141 * not be called by generic code.
2143 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2149 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2153 this.hash
= full_name_hash(name
, len
);
2155 return ERR_PTR(-EACCES
);
2157 if (unlikely(name
[0] == '.')) {
2158 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2159 return ERR_PTR(-EACCES
);
2163 c
= *(const unsigned char *)name
++;
2164 if (c
== '/' || c
== '\0')
2165 return ERR_PTR(-EACCES
);
2168 * See if the low-level filesystem might want
2169 * to use its own hash..
2171 if (base
->d_flags
& DCACHE_OP_HASH
) {
2172 int err
= base
->d_op
->d_hash(base
, &this);
2174 return ERR_PTR(err
);
2177 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2179 return ERR_PTR(err
);
2181 return __lookup_hash(&this, base
, 0);
2183 EXPORT_SYMBOL(lookup_one_len
);
2185 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2186 struct path
*path
, int *empty
)
2188 struct nameidata nd
;
2189 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2190 int err
= PTR_ERR(tmp
);
2193 BUG_ON(flags
& LOOKUP_PARENT
);
2195 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2203 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2206 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2208 EXPORT_SYMBOL(user_path_at
);
2211 * NB: most callers don't do anything directly with the reference to the
2212 * to struct filename, but the nd->last pointer points into the name string
2213 * allocated by getname. So we must hold the reference to it until all
2214 * path-walking is complete.
2216 static struct filename
*
2217 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2220 struct filename
*s
= getname(path
);
2223 /* only LOOKUP_REVAL is allowed in extra flags */
2224 flags
&= LOOKUP_REVAL
;
2229 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2232 return ERR_PTR(error
);
2239 * mountpoint_last - look up last component for umount
2240 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2241 * @path: pointer to container for result
2243 * This is a special lookup_last function just for umount. In this case, we
2244 * need to resolve the path without doing any revalidation.
2246 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2247 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2248 * in almost all cases, this lookup will be served out of the dcache. The only
2249 * cases where it won't are if nd->last refers to a symlink or the path is
2250 * bogus and it doesn't exist.
2253 * -error: if there was an error during lookup. This includes -ENOENT if the
2254 * lookup found a negative dentry. The nd->path reference will also be
2257 * 0: if we successfully resolved nd->path and found it to not to be a
2258 * symlink that needs to be followed. "path" will also be populated.
2259 * The nd->path reference will also be put.
2261 * 1: if we successfully resolved nd->last and found it to be a symlink
2262 * that needs to be followed. "path" will be populated with the path
2263 * to the link, and nd->path will *not* be put.
2266 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2269 struct dentry
*dentry
;
2270 struct dentry
*dir
= nd
->path
.dentry
;
2272 /* If we're in rcuwalk, drop out of it to handle last component */
2273 if (nd
->flags
& LOOKUP_RCU
) {
2274 if (unlazy_walk(nd
, NULL
)) {
2280 nd
->flags
&= ~LOOKUP_PARENT
;
2282 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2283 error
= handle_dots(nd
, nd
->last_type
);
2286 dentry
= dget(nd
->path
.dentry
);
2290 mutex_lock(&dir
->d_inode
->i_mutex
);
2291 dentry
= d_lookup(dir
, &nd
->last
);
2294 * No cached dentry. Mounted dentries are pinned in the cache,
2295 * so that means that this dentry is probably a symlink or the
2296 * path doesn't actually point to a mounted dentry.
2298 dentry
= d_alloc(dir
, &nd
->last
);
2301 mutex_unlock(&dir
->d_inode
->i_mutex
);
2304 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2305 error
= PTR_ERR(dentry
);
2306 if (IS_ERR(dentry
)) {
2307 mutex_unlock(&dir
->d_inode
->i_mutex
);
2311 mutex_unlock(&dir
->d_inode
->i_mutex
);
2314 if (d_is_negative(dentry
)) {
2319 path
->dentry
= dentry
;
2320 path
->mnt
= nd
->path
.mnt
;
2321 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
))
2332 * path_mountpoint - look up a path to be umounted
2333 * @dfd: directory file descriptor to start walk from
2334 * @name: full pathname to walk
2335 * @path: pointer to container for result
2336 * @flags: lookup flags
2338 * Look up the given name, but don't attempt to revalidate the last component.
2339 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2342 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2345 struct nameidata nd
;
2348 err
= path_init(dfd
, name
, flags
, &nd
);
2352 err
= mountpoint_last(&nd
, path
);
2355 struct path link
= *path
;
2356 err
= may_follow_link(&link
, &nd
);
2359 nd
.flags
|= LOOKUP_PARENT
;
2360 err
= follow_link(&link
, &nd
, &cookie
);
2363 err
= mountpoint_last(&nd
, path
);
2364 put_link(&nd
, &link
, cookie
);
2372 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2377 return PTR_ERR(name
);
2378 error
= path_mountpoint(dfd
, name
, path
, flags
| LOOKUP_RCU
);
2379 if (unlikely(error
== -ECHILD
))
2380 error
= path_mountpoint(dfd
, name
, path
, flags
);
2381 if (unlikely(error
== -ESTALE
))
2382 error
= path_mountpoint(dfd
, name
, path
, flags
| LOOKUP_REVAL
);
2384 audit_inode(name
, path
->dentry
, 0);
2390 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2391 * @dfd: directory file descriptor
2392 * @name: pathname from userland
2393 * @flags: lookup flags
2394 * @path: pointer to container to hold result
2396 * A umount is a special case for path walking. We're not actually interested
2397 * in the inode in this situation, and ESTALE errors can be a problem. We
2398 * simply want track down the dentry and vfsmount attached at the mountpoint
2399 * and avoid revalidating the last component.
2401 * Returns 0 and populates "path" on success.
2404 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2407 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2411 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2414 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2416 EXPORT_SYMBOL(kern_path_mountpoint
);
2418 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2420 kuid_t fsuid
= current_fsuid();
2422 if (uid_eq(inode
->i_uid
, fsuid
))
2424 if (uid_eq(dir
->i_uid
, fsuid
))
2426 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2428 EXPORT_SYMBOL(__check_sticky
);
2431 * Check whether we can remove a link victim from directory dir, check
2432 * whether the type of victim is right.
2433 * 1. We can't do it if dir is read-only (done in permission())
2434 * 2. We should have write and exec permissions on dir
2435 * 3. We can't remove anything from append-only dir
2436 * 4. We can't do anything with immutable dir (done in permission())
2437 * 5. If the sticky bit on dir is set we should either
2438 * a. be owner of dir, or
2439 * b. be owner of victim, or
2440 * c. have CAP_FOWNER capability
2441 * 6. If the victim is append-only or immutable we can't do antyhing with
2442 * links pointing to it.
2443 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2444 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2445 * 9. We can't remove a root or mountpoint.
2446 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2447 * nfs_async_unlink().
2449 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2451 struct inode
*inode
= victim
->d_inode
;
2454 if (d_is_negative(victim
))
2458 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2459 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2461 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2467 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2468 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2471 if (!d_is_dir(victim
))
2473 if (IS_ROOT(victim
))
2475 } else if (d_is_dir(victim
))
2477 if (IS_DEADDIR(dir
))
2479 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2484 /* Check whether we can create an object with dentry child in directory
2486 * 1. We can't do it if child already exists (open has special treatment for
2487 * this case, but since we are inlined it's OK)
2488 * 2. We can't do it if dir is read-only (done in permission())
2489 * 3. We should have write and exec permissions on dir
2490 * 4. We can't do it if dir is immutable (done in permission())
2492 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2494 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2497 if (IS_DEADDIR(dir
))
2499 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2503 * p1 and p2 should be directories on the same fs.
2505 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2510 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2514 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2516 p
= d_ancestor(p2
, p1
);
2518 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2519 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2523 p
= d_ancestor(p1
, p2
);
2525 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2526 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2530 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2531 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2534 EXPORT_SYMBOL(lock_rename
);
2536 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2538 mutex_unlock(&p1
->d_inode
->i_mutex
);
2540 mutex_unlock(&p2
->d_inode
->i_mutex
);
2541 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2544 EXPORT_SYMBOL(unlock_rename
);
2546 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2549 int error
= may_create(dir
, dentry
);
2553 if (!dir
->i_op
->create
)
2554 return -EACCES
; /* shouldn't it be ENOSYS? */
2557 error
= security_inode_create(dir
, dentry
, mode
);
2560 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2562 fsnotify_create(dir
, dentry
);
2565 EXPORT_SYMBOL(vfs_create
);
2567 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2569 struct dentry
*dentry
= path
->dentry
;
2570 struct inode
*inode
= dentry
->d_inode
;
2580 switch (inode
->i_mode
& S_IFMT
) {
2584 if (acc_mode
& MAY_WRITE
)
2589 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2598 error
= inode_permission(inode
, acc_mode
);
2603 * An append-only file must be opened in append mode for writing.
2605 if (IS_APPEND(inode
)) {
2606 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2612 /* O_NOATIME can only be set by the owner or superuser */
2613 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2619 static int handle_truncate(struct file
*filp
)
2621 struct path
*path
= &filp
->f_path
;
2622 struct inode
*inode
= path
->dentry
->d_inode
;
2623 int error
= get_write_access(inode
);
2627 * Refuse to truncate files with mandatory locks held on them.
2629 error
= locks_verify_locked(filp
);
2631 error
= security_path_truncate(path
);
2633 error
= do_truncate(path
->dentry
, 0,
2634 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2637 put_write_access(inode
);
2641 static inline int open_to_namei_flags(int flag
)
2643 if ((flag
& O_ACCMODE
) == 3)
2648 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2650 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2654 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2658 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2662 * Attempt to atomically look up, create and open a file from a negative
2665 * Returns 0 if successful. The file will have been created and attached to
2666 * @file by the filesystem calling finish_open().
2668 * Returns 1 if the file was looked up only or didn't need creating. The
2669 * caller will need to perform the open themselves. @path will have been
2670 * updated to point to the new dentry. This may be negative.
2672 * Returns an error code otherwise.
2674 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2675 struct path
*path
, struct file
*file
,
2676 const struct open_flags
*op
,
2677 bool got_write
, bool need_lookup
,
2680 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2681 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2685 int create_error
= 0;
2686 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2689 BUG_ON(dentry
->d_inode
);
2691 /* Don't create child dentry for a dead directory. */
2692 if (unlikely(IS_DEADDIR(dir
))) {
2698 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2699 mode
&= ~current_umask();
2701 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2703 open_flag
&= ~O_TRUNC
;
2706 * Checking write permission is tricky, bacuse we don't know if we are
2707 * going to actually need it: O_CREAT opens should work as long as the
2708 * file exists. But checking existence breaks atomicity. The trick is
2709 * to check access and if not granted clear O_CREAT from the flags.
2711 * Another problem is returing the "right" error value (e.g. for an
2712 * O_EXCL open we want to return EEXIST not EROFS).
2714 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2715 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2716 if (!(open_flag
& O_CREAT
)) {
2718 * No O_CREATE -> atomicity not a requirement -> fall
2719 * back to lookup + open
2722 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2723 /* Fall back and fail with the right error */
2724 create_error
= -EROFS
;
2727 /* No side effects, safe to clear O_CREAT */
2728 create_error
= -EROFS
;
2729 open_flag
&= ~O_CREAT
;
2733 if (open_flag
& O_CREAT
) {
2734 error
= may_o_create(&nd
->path
, dentry
, mode
);
2736 create_error
= error
;
2737 if (open_flag
& O_EXCL
)
2739 open_flag
&= ~O_CREAT
;
2743 if (nd
->flags
& LOOKUP_DIRECTORY
)
2744 open_flag
|= O_DIRECTORY
;
2746 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2747 file
->f_path
.mnt
= nd
->path
.mnt
;
2748 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2751 if (create_error
&& error
== -ENOENT
)
2752 error
= create_error
;
2756 if (error
) { /* returned 1, that is */
2757 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2761 if (file
->f_path
.dentry
) {
2763 dentry
= file
->f_path
.dentry
;
2765 if (*opened
& FILE_CREATED
)
2766 fsnotify_create(dir
, dentry
);
2767 if (!dentry
->d_inode
) {
2768 WARN_ON(*opened
& FILE_CREATED
);
2770 error
= create_error
;
2774 if (excl
&& !(*opened
& FILE_CREATED
)) {
2783 * We didn't have the inode before the open, so check open permission
2786 acc_mode
= op
->acc_mode
;
2787 if (*opened
& FILE_CREATED
) {
2788 WARN_ON(!(open_flag
& O_CREAT
));
2789 fsnotify_create(dir
, dentry
);
2790 acc_mode
= MAY_OPEN
;
2792 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2802 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2804 return PTR_ERR(dentry
);
2807 int open_flag
= op
->open_flag
;
2809 error
= create_error
;
2810 if ((open_flag
& O_EXCL
)) {
2811 if (!dentry
->d_inode
)
2813 } else if (!dentry
->d_inode
) {
2815 } else if ((open_flag
& O_TRUNC
) &&
2819 /* will fail later, go on to get the right error */
2823 path
->dentry
= dentry
;
2824 path
->mnt
= nd
->path
.mnt
;
2829 * Look up and maybe create and open the last component.
2831 * Must be called with i_mutex held on parent.
2833 * Returns 0 if the file was successfully atomically created (if necessary) and
2834 * opened. In this case the file will be returned attached to @file.
2836 * Returns 1 if the file was not completely opened at this time, though lookups
2837 * and creations will have been performed and the dentry returned in @path will
2838 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2839 * specified then a negative dentry may be returned.
2841 * An error code is returned otherwise.
2843 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2844 * cleared otherwise prior to returning.
2846 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2848 const struct open_flags
*op
,
2849 bool got_write
, int *opened
)
2851 struct dentry
*dir
= nd
->path
.dentry
;
2852 struct inode
*dir_inode
= dir
->d_inode
;
2853 struct dentry
*dentry
;
2857 *opened
&= ~FILE_CREATED
;
2858 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2860 return PTR_ERR(dentry
);
2862 /* Cached positive dentry: will open in f_op->open */
2863 if (!need_lookup
&& dentry
->d_inode
)
2866 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2867 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2868 need_lookup
, opened
);
2872 BUG_ON(dentry
->d_inode
);
2874 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2876 return PTR_ERR(dentry
);
2879 /* Negative dentry, just create the file */
2880 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2881 umode_t mode
= op
->mode
;
2882 if (!IS_POSIXACL(dir
->d_inode
))
2883 mode
&= ~current_umask();
2885 * This write is needed to ensure that a
2886 * rw->ro transition does not occur between
2887 * the time when the file is created and when
2888 * a permanent write count is taken through
2889 * the 'struct file' in finish_open().
2895 *opened
|= FILE_CREATED
;
2896 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2899 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2900 nd
->flags
& LOOKUP_EXCL
);
2905 path
->dentry
= dentry
;
2906 path
->mnt
= nd
->path
.mnt
;
2915 * Handle the last step of open()
2917 static int do_last(struct nameidata
*nd
, struct path
*path
,
2918 struct file
*file
, const struct open_flags
*op
,
2919 int *opened
, struct filename
*name
)
2921 struct dentry
*dir
= nd
->path
.dentry
;
2922 int open_flag
= op
->open_flag
;
2923 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2924 bool got_write
= false;
2925 int acc_mode
= op
->acc_mode
;
2926 struct inode
*inode
;
2927 bool symlink_ok
= false;
2928 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2929 bool retried
= false;
2932 nd
->flags
&= ~LOOKUP_PARENT
;
2933 nd
->flags
|= op
->intent
;
2935 if (nd
->last_type
!= LAST_NORM
) {
2936 error
= handle_dots(nd
, nd
->last_type
);
2942 if (!(open_flag
& O_CREAT
)) {
2943 if (nd
->last
.name
[nd
->last
.len
])
2944 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2945 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2947 /* we _can_ be in RCU mode here */
2948 error
= lookup_fast(nd
, path
, &inode
);
2955 BUG_ON(nd
->inode
!= dir
->d_inode
);
2957 /* create side of things */
2959 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2960 * has been cleared when we got to the last component we are
2963 error
= complete_walk(nd
);
2967 audit_inode(name
, dir
, LOOKUP_PARENT
);
2969 /* trailing slashes? */
2970 if (nd
->last
.name
[nd
->last
.len
])
2975 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2976 error
= mnt_want_write(nd
->path
.mnt
);
2980 * do _not_ fail yet - we might not need that or fail with
2981 * a different error; let lookup_open() decide; we'll be
2982 * dropping this one anyway.
2985 mutex_lock(&dir
->d_inode
->i_mutex
);
2986 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2987 mutex_unlock(&dir
->d_inode
->i_mutex
);
2993 if ((*opened
& FILE_CREATED
) ||
2994 !S_ISREG(file_inode(file
)->i_mode
))
2995 will_truncate
= false;
2997 audit_inode(name
, file
->f_path
.dentry
, 0);
3001 if (*opened
& FILE_CREATED
) {
3002 /* Don't check for write permission, don't truncate */
3003 open_flag
&= ~O_TRUNC
;
3004 will_truncate
= false;
3005 acc_mode
= MAY_OPEN
;
3006 path_to_nameidata(path
, nd
);
3007 goto finish_open_created
;
3011 * create/update audit record if it already exists.
3013 if (d_is_positive(path
->dentry
))
3014 audit_inode(name
, path
->dentry
, 0);
3017 * If atomic_open() acquired write access it is dropped now due to
3018 * possible mount and symlink following (this might be optimized away if
3022 mnt_drop_write(nd
->path
.mnt
);
3027 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3030 error
= follow_managed(path
, nd
->flags
);
3035 nd
->flags
|= LOOKUP_JUMPED
;
3037 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3038 inode
= path
->dentry
->d_inode
;
3040 /* we _can_ be in RCU mode here */
3042 if (d_is_negative(path
->dentry
)) {
3043 path_to_nameidata(path
, nd
);
3047 if (should_follow_link(path
->dentry
, !symlink_ok
)) {
3048 if (nd
->flags
& LOOKUP_RCU
) {
3049 if (unlikely(nd
->path
.mnt
!= path
->mnt
||
3050 unlazy_walk(nd
, path
->dentry
))) {
3055 BUG_ON(inode
!= path
->dentry
->d_inode
);
3059 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3060 path_to_nameidata(path
, nd
);
3062 save_parent
.dentry
= nd
->path
.dentry
;
3063 save_parent
.mnt
= mntget(path
->mnt
);
3064 nd
->path
.dentry
= path
->dentry
;
3068 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3070 error
= complete_walk(nd
);
3072 path_put(&save_parent
);
3075 audit_inode(name
, nd
->path
.dentry
, 0);
3077 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3080 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3082 if (!d_is_reg(nd
->path
.dentry
))
3083 will_truncate
= false;
3085 if (will_truncate
) {
3086 error
= mnt_want_write(nd
->path
.mnt
);
3091 finish_open_created
:
3092 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3096 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3097 error
= vfs_open(&nd
->path
, file
, current_cred());
3099 *opened
|= FILE_OPENED
;
3101 if (error
== -EOPENSTALE
)
3106 error
= open_check_o_direct(file
);
3109 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3113 if (will_truncate
) {
3114 error
= handle_truncate(file
);
3120 mnt_drop_write(nd
->path
.mnt
);
3121 path_put(&save_parent
);
3126 path_put_conditional(path
, nd
);
3133 /* If no saved parent or already retried then can't retry */
3134 if (!save_parent
.dentry
|| retried
)
3137 BUG_ON(save_parent
.dentry
!= dir
);
3138 path_put(&nd
->path
);
3139 nd
->path
= save_parent
;
3140 nd
->inode
= dir
->d_inode
;
3141 save_parent
.mnt
= NULL
;
3142 save_parent
.dentry
= NULL
;
3144 mnt_drop_write(nd
->path
.mnt
);
3151 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3152 struct nameidata
*nd
, int flags
,
3153 const struct open_flags
*op
,
3154 struct file
*file
, int *opened
)
3156 static const struct qstr name
= QSTR_INIT("/", 1);
3157 struct dentry
*dentry
, *child
;
3159 int error
= path_lookupat(dfd
, pathname
,
3160 flags
| LOOKUP_DIRECTORY
, nd
);
3161 if (unlikely(error
))
3163 error
= mnt_want_write(nd
->path
.mnt
);
3164 if (unlikely(error
))
3166 /* we want directory to be writable */
3167 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3170 dentry
= nd
->path
.dentry
;
3171 dir
= dentry
->d_inode
;
3172 if (!dir
->i_op
->tmpfile
) {
3173 error
= -EOPNOTSUPP
;
3176 child
= d_alloc(dentry
, &name
);
3177 if (unlikely(!child
)) {
3181 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3182 nd
->flags
|= op
->intent
;
3183 dput(nd
->path
.dentry
);
3184 nd
->path
.dentry
= child
;
3185 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3188 audit_inode(pathname
, nd
->path
.dentry
, 0);
3189 /* Don't check for other permissions, the inode was just created */
3190 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3193 file
->f_path
.mnt
= nd
->path
.mnt
;
3194 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3197 error
= open_check_o_direct(file
);
3200 } else if (!(op
->open_flag
& O_EXCL
)) {
3201 struct inode
*inode
= file_inode(file
);
3202 spin_lock(&inode
->i_lock
);
3203 inode
->i_state
|= I_LINKABLE
;
3204 spin_unlock(&inode
->i_lock
);
3207 mnt_drop_write(nd
->path
.mnt
);
3209 path_put(&nd
->path
);
3213 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3214 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3221 file
= get_empty_filp();
3225 file
->f_flags
= op
->open_flag
;
3227 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3228 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3232 error
= path_init(dfd
, pathname
, flags
, nd
);
3233 if (unlikely(error
))
3236 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3237 while (unlikely(error
> 0)) { /* trailing symlink */
3238 struct path link
= path
;
3240 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3241 path_put_conditional(&path
, nd
);
3242 path_put(&nd
->path
);
3246 error
= may_follow_link(&link
, nd
);
3247 if (unlikely(error
))
3249 nd
->flags
|= LOOKUP_PARENT
;
3250 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3251 error
= follow_link(&link
, nd
, &cookie
);
3252 if (unlikely(error
))
3254 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3255 put_link(nd
, &link
, cookie
);
3259 if (!(opened
& FILE_OPENED
)) {
3263 if (unlikely(error
)) {
3264 if (error
== -EOPENSTALE
) {
3265 if (flags
& LOOKUP_RCU
)
3270 file
= ERR_PTR(error
);
3275 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3276 const struct open_flags
*op
)
3278 struct nameidata nd
;
3279 int flags
= op
->lookup_flags
;
3282 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3283 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3284 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3285 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3286 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3290 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3291 const char *name
, const struct open_flags
*op
)
3293 struct nameidata nd
;
3295 struct filename
*filename
;
3296 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3299 nd
.root
.dentry
= dentry
;
3301 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3302 return ERR_PTR(-ELOOP
);
3304 filename
= getname_kernel(name
);
3305 if (unlikely(IS_ERR(filename
)))
3306 return ERR_CAST(filename
);
3308 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3309 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3310 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3311 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3312 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3317 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3318 struct path
*path
, unsigned int lookup_flags
)
3320 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3321 struct nameidata nd
;
3324 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3327 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3328 * other flags passed in are ignored!
3330 lookup_flags
&= LOOKUP_REVAL
;
3332 error
= filename_lookup(dfd
, name
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3334 return ERR_PTR(error
);
3337 * Yucky last component or no last component at all?
3338 * (foo/., foo/.., /////)
3340 if (nd
.last_type
!= LAST_NORM
)
3342 nd
.flags
&= ~LOOKUP_PARENT
;
3343 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3345 /* don't fail immediately if it's r/o, at least try to report other errors */
3346 err2
= mnt_want_write(nd
.path
.mnt
);
3348 * Do the final lookup.
3350 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3351 dentry
= lookup_hash(&nd
);
3356 if (d_is_positive(dentry
))
3360 * Special case - lookup gave negative, but... we had foo/bar/
3361 * From the vfs_mknod() POV we just have a negative dentry -
3362 * all is fine. Let's be bastards - you had / on the end, you've
3363 * been asking for (non-existent) directory. -ENOENT for you.
3365 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3369 if (unlikely(err2
)) {
3377 dentry
= ERR_PTR(error
);
3379 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3381 mnt_drop_write(nd
.path
.mnt
);
3387 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3388 struct path
*path
, unsigned int lookup_flags
)
3390 struct filename
*filename
= getname_kernel(pathname
);
3393 if (IS_ERR(filename
))
3394 return ERR_CAST(filename
);
3395 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3399 EXPORT_SYMBOL(kern_path_create
);
3401 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3404 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3405 mnt_drop_write(path
->mnt
);
3408 EXPORT_SYMBOL(done_path_create
);
3410 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3411 struct path
*path
, unsigned int lookup_flags
)
3413 struct filename
*tmp
= getname(pathname
);
3416 return ERR_CAST(tmp
);
3417 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3421 EXPORT_SYMBOL(user_path_create
);
3423 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3425 int error
= may_create(dir
, dentry
);
3430 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3433 if (!dir
->i_op
->mknod
)
3436 error
= devcgroup_inode_mknod(mode
, dev
);
3440 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3444 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3446 fsnotify_create(dir
, dentry
);
3449 EXPORT_SYMBOL(vfs_mknod
);
3451 static int may_mknod(umode_t mode
)
3453 switch (mode
& S_IFMT
) {
3459 case 0: /* zero mode translates to S_IFREG */
3468 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3471 struct dentry
*dentry
;
3474 unsigned int lookup_flags
= 0;
3476 error
= may_mknod(mode
);
3480 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3482 return PTR_ERR(dentry
);
3484 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3485 mode
&= ~current_umask();
3486 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3489 switch (mode
& S_IFMT
) {
3490 case 0: case S_IFREG
:
3491 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3493 case S_IFCHR
: case S_IFBLK
:
3494 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3495 new_decode_dev(dev
));
3497 case S_IFIFO
: case S_IFSOCK
:
3498 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3502 done_path_create(&path
, dentry
);
3503 if (retry_estale(error
, lookup_flags
)) {
3504 lookup_flags
|= LOOKUP_REVAL
;
3510 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3512 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3515 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3517 int error
= may_create(dir
, dentry
);
3518 unsigned max_links
= dir
->i_sb
->s_max_links
;
3523 if (!dir
->i_op
->mkdir
)
3526 mode
&= (S_IRWXUGO
|S_ISVTX
);
3527 error
= security_inode_mkdir(dir
, dentry
, mode
);
3531 if (max_links
&& dir
->i_nlink
>= max_links
)
3534 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3536 fsnotify_mkdir(dir
, dentry
);
3539 EXPORT_SYMBOL(vfs_mkdir
);
3541 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3543 struct dentry
*dentry
;
3546 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3549 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3551 return PTR_ERR(dentry
);
3553 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3554 mode
&= ~current_umask();
3555 error
= security_path_mkdir(&path
, dentry
, mode
);
3557 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3558 done_path_create(&path
, dentry
);
3559 if (retry_estale(error
, lookup_flags
)) {
3560 lookup_flags
|= LOOKUP_REVAL
;
3566 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3568 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3572 * The dentry_unhash() helper will try to drop the dentry early: we
3573 * should have a usage count of 1 if we're the only user of this
3574 * dentry, and if that is true (possibly after pruning the dcache),
3575 * then we drop the dentry now.
3577 * A low-level filesystem can, if it choses, legally
3580 * if (!d_unhashed(dentry))
3583 * if it cannot handle the case of removing a directory
3584 * that is still in use by something else..
3586 void dentry_unhash(struct dentry
*dentry
)
3588 shrink_dcache_parent(dentry
);
3589 spin_lock(&dentry
->d_lock
);
3590 if (dentry
->d_lockref
.count
== 1)
3592 spin_unlock(&dentry
->d_lock
);
3594 EXPORT_SYMBOL(dentry_unhash
);
3596 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3598 int error
= may_delete(dir
, dentry
, 1);
3603 if (!dir
->i_op
->rmdir
)
3607 mutex_lock(&dentry
->d_inode
->i_mutex
);
3610 if (is_local_mountpoint(dentry
))
3613 error
= security_inode_rmdir(dir
, dentry
);
3617 shrink_dcache_parent(dentry
);
3618 error
= dir
->i_op
->rmdir(dir
, dentry
);
3622 dentry
->d_inode
->i_flags
|= S_DEAD
;
3624 detach_mounts(dentry
);
3627 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3633 EXPORT_SYMBOL(vfs_rmdir
);
3635 static long do_rmdir(int dfd
, const char __user
*pathname
)
3638 struct filename
*name
;
3639 struct dentry
*dentry
;
3640 struct nameidata nd
;
3641 unsigned int lookup_flags
= 0;
3643 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3645 return PTR_ERR(name
);
3647 switch(nd
.last_type
) {
3659 nd
.flags
&= ~LOOKUP_PARENT
;
3660 error
= mnt_want_write(nd
.path
.mnt
);
3664 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3665 dentry
= lookup_hash(&nd
);
3666 error
= PTR_ERR(dentry
);
3669 if (!dentry
->d_inode
) {
3673 error
= security_path_rmdir(&nd
.path
, dentry
);
3676 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3680 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3681 mnt_drop_write(nd
.path
.mnt
);
3685 if (retry_estale(error
, lookup_flags
)) {
3686 lookup_flags
|= LOOKUP_REVAL
;
3692 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3694 return do_rmdir(AT_FDCWD
, pathname
);
3698 * vfs_unlink - unlink a filesystem object
3699 * @dir: parent directory
3701 * @delegated_inode: returns victim inode, if the inode is delegated.
3703 * The caller must hold dir->i_mutex.
3705 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3706 * return a reference to the inode in delegated_inode. The caller
3707 * should then break the delegation on that inode and retry. Because
3708 * breaking a delegation may take a long time, the caller should drop
3709 * dir->i_mutex before doing so.
3711 * Alternatively, a caller may pass NULL for delegated_inode. This may
3712 * be appropriate for callers that expect the underlying filesystem not
3713 * to be NFS exported.
3715 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3717 struct inode
*target
= dentry
->d_inode
;
3718 int error
= may_delete(dir
, dentry
, 0);
3723 if (!dir
->i_op
->unlink
)
3726 mutex_lock(&target
->i_mutex
);
3727 if (is_local_mountpoint(dentry
))
3730 error
= security_inode_unlink(dir
, dentry
);
3732 error
= try_break_deleg(target
, delegated_inode
);
3735 error
= dir
->i_op
->unlink(dir
, dentry
);
3738 detach_mounts(dentry
);
3743 mutex_unlock(&target
->i_mutex
);
3745 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3746 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3747 fsnotify_link_count(target
);
3753 EXPORT_SYMBOL(vfs_unlink
);
3756 * Make sure that the actual truncation of the file will occur outside its
3757 * directory's i_mutex. Truncate can take a long time if there is a lot of
3758 * writeout happening, and we don't want to prevent access to the directory
3759 * while waiting on the I/O.
3761 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3764 struct filename
*name
;
3765 struct dentry
*dentry
;
3766 struct nameidata nd
;
3767 struct inode
*inode
= NULL
;
3768 struct inode
*delegated_inode
= NULL
;
3769 unsigned int lookup_flags
= 0;
3771 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3773 return PTR_ERR(name
);
3776 if (nd
.last_type
!= LAST_NORM
)
3779 nd
.flags
&= ~LOOKUP_PARENT
;
3780 error
= mnt_want_write(nd
.path
.mnt
);
3784 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3785 dentry
= lookup_hash(&nd
);
3786 error
= PTR_ERR(dentry
);
3787 if (!IS_ERR(dentry
)) {
3788 /* Why not before? Because we want correct error value */
3789 if (nd
.last
.name
[nd
.last
.len
])
3791 inode
= dentry
->d_inode
;
3792 if (d_is_negative(dentry
))
3795 error
= security_path_unlink(&nd
.path
, dentry
);
3798 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3802 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3804 iput(inode
); /* truncate the inode here */
3806 if (delegated_inode
) {
3807 error
= break_deleg_wait(&delegated_inode
);
3811 mnt_drop_write(nd
.path
.mnt
);
3815 if (retry_estale(error
, lookup_flags
)) {
3816 lookup_flags
|= LOOKUP_REVAL
;
3823 if (d_is_negative(dentry
))
3825 else if (d_is_dir(dentry
))
3832 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3834 if ((flag
& ~AT_REMOVEDIR
) != 0)
3837 if (flag
& AT_REMOVEDIR
)
3838 return do_rmdir(dfd
, pathname
);
3840 return do_unlinkat(dfd
, pathname
);
3843 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3845 return do_unlinkat(AT_FDCWD
, pathname
);
3848 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3850 int error
= may_create(dir
, dentry
);
3855 if (!dir
->i_op
->symlink
)
3858 error
= security_inode_symlink(dir
, dentry
, oldname
);
3862 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3864 fsnotify_create(dir
, dentry
);
3867 EXPORT_SYMBOL(vfs_symlink
);
3869 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3870 int, newdfd
, const char __user
*, newname
)
3873 struct filename
*from
;
3874 struct dentry
*dentry
;
3876 unsigned int lookup_flags
= 0;
3878 from
= getname(oldname
);
3880 return PTR_ERR(from
);
3882 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3883 error
= PTR_ERR(dentry
);
3887 error
= security_path_symlink(&path
, dentry
, from
->name
);
3889 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3890 done_path_create(&path
, dentry
);
3891 if (retry_estale(error
, lookup_flags
)) {
3892 lookup_flags
|= LOOKUP_REVAL
;
3900 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3902 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3906 * vfs_link - create a new link
3907 * @old_dentry: object to be linked
3909 * @new_dentry: where to create the new link
3910 * @delegated_inode: returns inode needing a delegation break
3912 * The caller must hold dir->i_mutex
3914 * If vfs_link discovers a delegation on the to-be-linked file in need
3915 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3916 * inode in delegated_inode. The caller should then break the delegation
3917 * and retry. Because breaking a delegation may take a long time, the
3918 * caller should drop the i_mutex before doing so.
3920 * Alternatively, a caller may pass NULL for delegated_inode. This may
3921 * be appropriate for callers that expect the underlying filesystem not
3922 * to be NFS exported.
3924 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3926 struct inode
*inode
= old_dentry
->d_inode
;
3927 unsigned max_links
= dir
->i_sb
->s_max_links
;
3933 error
= may_create(dir
, new_dentry
);
3937 if (dir
->i_sb
!= inode
->i_sb
)
3941 * A link to an append-only or immutable file cannot be created.
3943 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3945 if (!dir
->i_op
->link
)
3947 if (S_ISDIR(inode
->i_mode
))
3950 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3954 mutex_lock(&inode
->i_mutex
);
3955 /* Make sure we don't allow creating hardlink to an unlinked file */
3956 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3958 else if (max_links
&& inode
->i_nlink
>= max_links
)
3961 error
= try_break_deleg(inode
, delegated_inode
);
3963 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3966 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3967 spin_lock(&inode
->i_lock
);
3968 inode
->i_state
&= ~I_LINKABLE
;
3969 spin_unlock(&inode
->i_lock
);
3971 mutex_unlock(&inode
->i_mutex
);
3973 fsnotify_link(dir
, inode
, new_dentry
);
3976 EXPORT_SYMBOL(vfs_link
);
3979 * Hardlinks are often used in delicate situations. We avoid
3980 * security-related surprises by not following symlinks on the
3983 * We don't follow them on the oldname either to be compatible
3984 * with linux 2.0, and to avoid hard-linking to directories
3985 * and other special files. --ADM
3987 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3988 int, newdfd
, const char __user
*, newname
, int, flags
)
3990 struct dentry
*new_dentry
;
3991 struct path old_path
, new_path
;
3992 struct inode
*delegated_inode
= NULL
;
3996 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3999 * To use null names we require CAP_DAC_READ_SEARCH
4000 * This ensures that not everyone will be able to create
4001 * handlink using the passed filedescriptor.
4003 if (flags
& AT_EMPTY_PATH
) {
4004 if (!capable(CAP_DAC_READ_SEARCH
))
4009 if (flags
& AT_SYMLINK_FOLLOW
)
4010 how
|= LOOKUP_FOLLOW
;
4012 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4016 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4017 (how
& LOOKUP_REVAL
));
4018 error
= PTR_ERR(new_dentry
);
4019 if (IS_ERR(new_dentry
))
4023 if (old_path
.mnt
!= new_path
.mnt
)
4025 error
= may_linkat(&old_path
);
4026 if (unlikely(error
))
4028 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4031 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4033 done_path_create(&new_path
, new_dentry
);
4034 if (delegated_inode
) {
4035 error
= break_deleg_wait(&delegated_inode
);
4037 path_put(&old_path
);
4041 if (retry_estale(error
, how
)) {
4042 path_put(&old_path
);
4043 how
|= LOOKUP_REVAL
;
4047 path_put(&old_path
);
4052 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4054 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4058 * vfs_rename - rename a filesystem object
4059 * @old_dir: parent of source
4060 * @old_dentry: source
4061 * @new_dir: parent of destination
4062 * @new_dentry: destination
4063 * @delegated_inode: returns an inode needing a delegation break
4064 * @flags: rename flags
4066 * The caller must hold multiple mutexes--see lock_rename()).
4068 * If vfs_rename discovers a delegation in need of breaking at either
4069 * the source or destination, it will return -EWOULDBLOCK and return a
4070 * reference to the inode in delegated_inode. The caller should then
4071 * break the delegation and retry. Because breaking a delegation may
4072 * take a long time, the caller should drop all locks before doing
4075 * Alternatively, a caller may pass NULL for delegated_inode. This may
4076 * be appropriate for callers that expect the underlying filesystem not
4077 * to be NFS exported.
4079 * The worst of all namespace operations - renaming directory. "Perverted"
4080 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4082 * a) we can get into loop creation.
4083 * b) race potential - two innocent renames can create a loop together.
4084 * That's where 4.4 screws up. Current fix: serialization on
4085 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4087 * c) we have to lock _four_ objects - parents and victim (if it exists),
4088 * and source (if it is not a directory).
4089 * And that - after we got ->i_mutex on parents (until then we don't know
4090 * whether the target exists). Solution: try to be smart with locking
4091 * order for inodes. We rely on the fact that tree topology may change
4092 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4093 * move will be locked. Thus we can rank directories by the tree
4094 * (ancestors first) and rank all non-directories after them.
4095 * That works since everybody except rename does "lock parent, lookup,
4096 * lock child" and rename is under ->s_vfs_rename_mutex.
4097 * HOWEVER, it relies on the assumption that any object with ->lookup()
4098 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4099 * we'd better make sure that there's no link(2) for them.
4100 * d) conversion from fhandle to dentry may come in the wrong moment - when
4101 * we are removing the target. Solution: we will have to grab ->i_mutex
4102 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4103 * ->i_mutex on parents, which works but leads to some truly excessive
4106 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4107 struct inode
*new_dir
, struct dentry
*new_dentry
,
4108 struct inode
**delegated_inode
, unsigned int flags
)
4111 bool is_dir
= d_is_dir(old_dentry
);
4112 const unsigned char *old_name
;
4113 struct inode
*source
= old_dentry
->d_inode
;
4114 struct inode
*target
= new_dentry
->d_inode
;
4115 bool new_is_dir
= false;
4116 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4118 if (source
== target
)
4121 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4126 error
= may_create(new_dir
, new_dentry
);
4128 new_is_dir
= d_is_dir(new_dentry
);
4130 if (!(flags
& RENAME_EXCHANGE
))
4131 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4133 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4138 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4141 if (flags
&& !old_dir
->i_op
->rename2
)
4145 * If we are going to change the parent - check write permissions,
4146 * we'll need to flip '..'.
4148 if (new_dir
!= old_dir
) {
4150 error
= inode_permission(source
, MAY_WRITE
);
4154 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4155 error
= inode_permission(target
, MAY_WRITE
);
4161 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4166 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4168 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4169 lock_two_nondirectories(source
, target
);
4171 mutex_lock(&target
->i_mutex
);
4174 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4177 if (max_links
&& new_dir
!= old_dir
) {
4179 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4181 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4182 old_dir
->i_nlink
>= max_links
)
4185 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4186 shrink_dcache_parent(new_dentry
);
4188 error
= try_break_deleg(source
, delegated_inode
);
4192 if (target
&& !new_is_dir
) {
4193 error
= try_break_deleg(target
, delegated_inode
);
4197 if (!old_dir
->i_op
->rename2
) {
4198 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4199 new_dir
, new_dentry
);
4201 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4202 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4203 new_dir
, new_dentry
, flags
);
4208 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4210 target
->i_flags
|= S_DEAD
;
4211 dont_mount(new_dentry
);
4212 detach_mounts(new_dentry
);
4214 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4215 if (!(flags
& RENAME_EXCHANGE
))
4216 d_move(old_dentry
, new_dentry
);
4218 d_exchange(old_dentry
, new_dentry
);
4221 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4222 unlock_two_nondirectories(source
, target
);
4224 mutex_unlock(&target
->i_mutex
);
4227 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4228 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4229 if (flags
& RENAME_EXCHANGE
) {
4230 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4231 new_is_dir
, NULL
, new_dentry
);
4234 fsnotify_oldname_free(old_name
);
4238 EXPORT_SYMBOL(vfs_rename
);
4240 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4241 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4243 struct dentry
*old_dir
, *new_dir
;
4244 struct dentry
*old_dentry
, *new_dentry
;
4245 struct dentry
*trap
;
4246 struct nameidata oldnd
, newnd
;
4247 struct inode
*delegated_inode
= NULL
;
4248 struct filename
*from
;
4249 struct filename
*to
;
4250 unsigned int lookup_flags
= 0;
4251 bool should_retry
= false;
4254 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4257 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4258 (flags
& RENAME_EXCHANGE
))
4261 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4265 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4267 error
= PTR_ERR(from
);
4271 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4273 error
= PTR_ERR(to
);
4278 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4281 old_dir
= oldnd
.path
.dentry
;
4283 if (oldnd
.last_type
!= LAST_NORM
)
4286 new_dir
= newnd
.path
.dentry
;
4287 if (flags
& RENAME_NOREPLACE
)
4289 if (newnd
.last_type
!= LAST_NORM
)
4292 error
= mnt_want_write(oldnd
.path
.mnt
);
4296 oldnd
.flags
&= ~LOOKUP_PARENT
;
4297 newnd
.flags
&= ~LOOKUP_PARENT
;
4298 if (!(flags
& RENAME_EXCHANGE
))
4299 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4302 trap
= lock_rename(new_dir
, old_dir
);
4304 old_dentry
= lookup_hash(&oldnd
);
4305 error
= PTR_ERR(old_dentry
);
4306 if (IS_ERR(old_dentry
))
4308 /* source must exist */
4310 if (d_is_negative(old_dentry
))
4312 new_dentry
= lookup_hash(&newnd
);
4313 error
= PTR_ERR(new_dentry
);
4314 if (IS_ERR(new_dentry
))
4317 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4319 if (flags
& RENAME_EXCHANGE
) {
4321 if (d_is_negative(new_dentry
))
4324 if (!d_is_dir(new_dentry
)) {
4326 if (newnd
.last
.name
[newnd
.last
.len
])
4330 /* unless the source is a directory trailing slashes give -ENOTDIR */
4331 if (!d_is_dir(old_dentry
)) {
4333 if (oldnd
.last
.name
[oldnd
.last
.len
])
4335 if (!(flags
& RENAME_EXCHANGE
) && newnd
.last
.name
[newnd
.last
.len
])
4338 /* source should not be ancestor of target */
4340 if (old_dentry
== trap
)
4342 /* target should not be an ancestor of source */
4343 if (!(flags
& RENAME_EXCHANGE
))
4345 if (new_dentry
== trap
)
4348 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4349 &newnd
.path
, new_dentry
, flags
);
4352 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4353 new_dir
->d_inode
, new_dentry
,
4354 &delegated_inode
, flags
);
4360 unlock_rename(new_dir
, old_dir
);
4361 if (delegated_inode
) {
4362 error
= break_deleg_wait(&delegated_inode
);
4366 mnt_drop_write(oldnd
.path
.mnt
);
4368 if (retry_estale(error
, lookup_flags
))
4369 should_retry
= true;
4370 path_put(&newnd
.path
);
4373 path_put(&oldnd
.path
);
4376 should_retry
= false;
4377 lookup_flags
|= LOOKUP_REVAL
;
4384 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4385 int, newdfd
, const char __user
*, newname
)
4387 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4390 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4392 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4395 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4397 int error
= may_create(dir
, dentry
);
4401 if (!dir
->i_op
->mknod
)
4404 return dir
->i_op
->mknod(dir
, dentry
,
4405 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4407 EXPORT_SYMBOL(vfs_whiteout
);
4409 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4411 int len
= PTR_ERR(link
);
4416 if (len
> (unsigned) buflen
)
4418 if (copy_to_user(buffer
, link
, len
))
4423 EXPORT_SYMBOL(readlink_copy
);
4426 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4427 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4428 * using) it for any given inode is up to filesystem.
4430 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4432 struct nameidata nd
;
4437 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4439 return PTR_ERR(cookie
);
4441 res
= readlink_copy(buffer
, buflen
, nd_get_link(&nd
));
4442 if (dentry
->d_inode
->i_op
->put_link
)
4443 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4446 EXPORT_SYMBOL(generic_readlink
);
4448 /* get the link contents into pagecache */
4449 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4453 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4454 page
= read_mapping_page(mapping
, 0, NULL
);
4459 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4463 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4465 struct page
*page
= NULL
;
4466 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4469 page_cache_release(page
);
4473 EXPORT_SYMBOL(page_readlink
);
4475 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4477 struct page
*page
= NULL
;
4478 nd_set_link(nd
, page_getlink(dentry
, &page
));
4481 EXPORT_SYMBOL(page_follow_link_light
);
4483 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4485 struct page
*page
= cookie
;
4489 page_cache_release(page
);
4492 EXPORT_SYMBOL(page_put_link
);
4495 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4497 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4499 struct address_space
*mapping
= inode
->i_mapping
;
4504 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4506 flags
|= AOP_FLAG_NOFS
;
4509 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4510 flags
, &page
, &fsdata
);
4514 kaddr
= kmap_atomic(page
);
4515 memcpy(kaddr
, symname
, len
-1);
4516 kunmap_atomic(kaddr
);
4518 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4525 mark_inode_dirty(inode
);
4530 EXPORT_SYMBOL(__page_symlink
);
4532 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4534 return __page_symlink(inode
, symname
, len
,
4535 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4537 EXPORT_SYMBOL(page_symlink
);
4539 const struct inode_operations page_symlink_inode_operations
= {
4540 .readlink
= generic_readlink
,
4541 .follow_link
= page_follow_link_light
,
4542 .put_link
= page_put_link
,
4544 EXPORT_SYMBOL(page_symlink_inode_operations
);