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
);
495 #define EMBEDDED_LEVELS 2
500 struct inode
*inode
; /* path.dentry.d_inode */
505 int total_link_count
;
512 } *stack
, internal
[EMBEDDED_LEVELS
];
515 static struct nameidata
*set_nameidata(struct nameidata
*p
)
517 struct nameidata
*old
= current
->nameidata
;
518 p
->stack
= p
->internal
;
519 p
->total_link_count
= old
? old
->total_link_count
: 0;
520 current
->nameidata
= p
;
524 static void restore_nameidata(struct nameidata
*old
)
526 struct nameidata
*now
= current
->nameidata
;
528 current
->nameidata
= old
;
530 old
->total_link_count
= now
->total_link_count
;
531 if (now
->stack
!= now
->internal
) {
533 now
->stack
= now
->internal
;
537 static int __nd_alloc_stack(struct nameidata
*nd
)
541 if (nd
->flags
& LOOKUP_RCU
) {
542 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
547 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
552 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
557 static inline int nd_alloc_stack(struct nameidata
*nd
)
559 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
561 if (likely(nd
->stack
!= nd
->internal
))
563 return __nd_alloc_stack(nd
);
566 static void drop_links(struct nameidata
*nd
)
570 struct saved
*last
= nd
->stack
+ i
;
571 struct inode
*inode
= last
->inode
;
572 if (last
->cookie
&& inode
->i_op
->put_link
) {
573 inode
->i_op
->put_link(inode
, last
->cookie
);
579 static void terminate_walk(struct nameidata
*nd
)
582 if (!(nd
->flags
& LOOKUP_RCU
)) {
585 for (i
= 0; i
< nd
->depth
; i
++)
586 path_put(&nd
->stack
[i
].link
);
588 nd
->flags
&= ~LOOKUP_RCU
;
589 if (!(nd
->flags
& LOOKUP_ROOT
))
596 /* path_put is needed afterwards regardless of success or failure */
597 static bool legitimize_path(struct nameidata
*nd
,
598 struct path
*path
, unsigned seq
)
600 int res
= __legitimize_mnt(path
->mnt
, nd
->m_seq
);
607 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
611 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
614 static bool legitimize_links(struct nameidata
*nd
)
617 for (i
= 0; i
< nd
->depth
; i
++) {
618 struct saved
*last
= nd
->stack
+ i
;
619 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
629 * Path walking has 2 modes, rcu-walk and ref-walk (see
630 * Documentation/filesystems/path-lookup.txt). In situations when we can't
631 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
632 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
633 * mode. Refcounts are grabbed at the last known good point before rcu-walk
634 * got stuck, so ref-walk may continue from there. If this is not successful
635 * (eg. a seqcount has changed), then failure is returned and it's up to caller
636 * to restart the path walk from the beginning in ref-walk mode.
640 * unlazy_walk - try to switch to ref-walk mode.
641 * @nd: nameidata pathwalk data
642 * @dentry: child of nd->path.dentry or NULL
643 * @seq: seq number to check dentry against
644 * Returns: 0 on success, -ECHILD on failure
646 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
647 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
648 * @nd or NULL. Must be called from rcu-walk context.
649 * Nothing should touch nameidata between unlazy_walk() failure and
652 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
654 struct fs_struct
*fs
= current
->fs
;
655 struct dentry
*parent
= nd
->path
.dentry
;
657 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
659 nd
->flags
&= ~LOOKUP_RCU
;
660 if (unlikely(!legitimize_links(nd
)))
662 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
664 if (unlikely(!lockref_get_not_dead(&parent
->d_lockref
)))
668 * For a negative lookup, the lookup sequence point is the parents
669 * sequence point, and it only needs to revalidate the parent dentry.
671 * For a positive lookup, we need to move both the parent and the
672 * dentry from the RCU domain to be properly refcounted. And the
673 * sequence number in the dentry validates *both* dentry counters,
674 * since we checked the sequence number of the parent after we got
675 * the child sequence number. So we know the parent must still
676 * be valid if the child sequence number is still valid.
679 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
681 BUG_ON(nd
->inode
!= parent
->d_inode
);
683 if (!lockref_get_not_dead(&dentry
->d_lockref
))
685 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
690 * Sequence counts matched. Now make sure that the root is
691 * still valid and get it if required.
693 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
694 spin_lock(&fs
->lock
);
695 if (unlikely(!path_equal(&nd
->root
, &fs
->root
))) {
696 spin_unlock(&fs
->lock
);
700 spin_unlock(&fs
->lock
);
713 nd
->path
.dentry
= NULL
;
717 if (!(nd
->flags
& LOOKUP_ROOT
))
722 static int unlazy_link(struct nameidata
*nd
, struct path
*link
, unsigned seq
)
724 if (unlikely(!legitimize_path(nd
, link
, seq
))) {
727 nd
->flags
&= ~LOOKUP_RCU
;
729 nd
->path
.dentry
= NULL
;
730 if (!(nd
->flags
& LOOKUP_ROOT
))
733 } else if (likely(unlazy_walk(nd
, NULL
, 0)) == 0) {
740 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
742 return dentry
->d_op
->d_revalidate(dentry
, flags
);
746 * complete_walk - successful completion of path walk
747 * @nd: pointer nameidata
749 * If we had been in RCU mode, drop out of it and legitimize nd->path.
750 * Revalidate the final result, unless we'd already done that during
751 * the path walk or the filesystem doesn't ask for it. Return 0 on
752 * success, -error on failure. In case of failure caller does not
753 * need to drop nd->path.
755 static int complete_walk(struct nameidata
*nd
)
757 struct dentry
*dentry
= nd
->path
.dentry
;
760 if (nd
->flags
& LOOKUP_RCU
) {
761 if (!(nd
->flags
& LOOKUP_ROOT
))
763 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
767 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
770 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
773 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
783 static __always_inline
void set_root(struct nameidata
*nd
)
785 get_fs_root(current
->fs
, &nd
->root
);
788 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
790 struct fs_struct
*fs
= current
->fs
;
794 seq
= read_seqcount_begin(&fs
->seq
);
796 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
797 } while (read_seqcount_retry(&fs
->seq
, seq
));
801 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
804 if (path
->mnt
!= nd
->path
.mnt
)
808 static inline void path_to_nameidata(const struct path
*path
,
809 struct nameidata
*nd
)
811 if (!(nd
->flags
& LOOKUP_RCU
)) {
812 dput(nd
->path
.dentry
);
813 if (nd
->path
.mnt
!= path
->mnt
)
814 mntput(nd
->path
.mnt
);
816 nd
->path
.mnt
= path
->mnt
;
817 nd
->path
.dentry
= path
->dentry
;
821 * Helper to directly jump to a known parsed path from ->follow_link,
822 * caller must have taken a reference to path beforehand.
824 void nd_jump_link(struct path
*path
)
826 struct nameidata
*nd
= current
->nameidata
;
830 nd
->inode
= nd
->path
.dentry
->d_inode
;
831 nd
->flags
|= LOOKUP_JUMPED
;
834 static inline void put_link(struct nameidata
*nd
)
836 struct saved
*last
= nd
->stack
+ --nd
->depth
;
837 struct inode
*inode
= last
->inode
;
838 if (last
->cookie
&& inode
->i_op
->put_link
)
839 inode
->i_op
->put_link(inode
, last
->cookie
);
840 if (!(nd
->flags
& LOOKUP_RCU
))
841 path_put(&last
->link
);
844 int sysctl_protected_symlinks __read_mostly
= 0;
845 int sysctl_protected_hardlinks __read_mostly
= 0;
848 * may_follow_link - Check symlink following for unsafe situations
849 * @nd: nameidata pathwalk data
851 * In the case of the sysctl_protected_symlinks sysctl being enabled,
852 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
853 * in a sticky world-writable directory. This is to protect privileged
854 * processes from failing races against path names that may change out
855 * from under them by way of other users creating malicious symlinks.
856 * It will permit symlinks to be followed only when outside a sticky
857 * world-writable directory, or when the uid of the symlink and follower
858 * match, or when the directory owner matches the symlink's owner.
860 * Returns 0 if following the symlink is allowed, -ve on error.
862 static inline int may_follow_link(struct nameidata
*nd
)
864 const struct inode
*inode
;
865 const struct inode
*parent
;
867 if (!sysctl_protected_symlinks
)
870 /* Allowed if owner and follower match. */
871 inode
= nd
->stack
[0].inode
;
872 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
875 /* Allowed if parent directory not sticky and world-writable. */
876 parent
= nd
->path
.dentry
->d_inode
;
877 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
880 /* Allowed if parent directory and link owner match. */
881 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
884 if (nd
->flags
& LOOKUP_RCU
)
887 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
892 * safe_hardlink_source - Check for safe hardlink conditions
893 * @inode: the source inode to hardlink from
895 * Return false if at least one of the following conditions:
896 * - inode is not a regular file
898 * - inode is setgid and group-exec
899 * - access failure for read and write
901 * Otherwise returns true.
903 static bool safe_hardlink_source(struct inode
*inode
)
905 umode_t mode
= inode
->i_mode
;
907 /* Special files should not get pinned to the filesystem. */
911 /* Setuid files should not get pinned to the filesystem. */
915 /* Executable setgid files should not get pinned to the filesystem. */
916 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
919 /* Hardlinking to unreadable or unwritable sources is dangerous. */
920 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
927 * may_linkat - Check permissions for creating a hardlink
928 * @link: the source to hardlink from
930 * Block hardlink when all of:
931 * - sysctl_protected_hardlinks enabled
932 * - fsuid does not match inode
933 * - hardlink source is unsafe (see safe_hardlink_source() above)
936 * Returns 0 if successful, -ve on error.
938 static int may_linkat(struct path
*link
)
940 const struct cred
*cred
;
943 if (!sysctl_protected_hardlinks
)
946 cred
= current_cred();
947 inode
= link
->dentry
->d_inode
;
949 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
950 * otherwise, it must be a safe source.
952 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
956 audit_log_link_denied("linkat", link
);
960 static __always_inline
961 const char *get_link(struct nameidata
*nd
)
963 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
964 struct dentry
*dentry
= last
->link
.dentry
;
965 struct inode
*inode
= last
->inode
;
969 if (!(nd
->flags
& LOOKUP_RCU
)) {
970 touch_atime(&last
->link
);
972 } else if (atime_needs_update(&last
->link
, inode
)) {
973 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
974 return ERR_PTR(-ECHILD
);
975 touch_atime(&last
->link
);
978 error
= security_inode_follow_link(dentry
, inode
,
979 nd
->flags
& LOOKUP_RCU
);
981 return ERR_PTR(error
);
983 nd
->last_type
= LAST_BIND
;
986 if (nd
->flags
& LOOKUP_RCU
) {
987 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
988 return ERR_PTR(-ECHILD
);
990 res
= inode
->i_op
->follow_link(dentry
, &last
->cookie
);
991 if (IS_ERR_OR_NULL(res
)) {
997 if (nd
->flags
& LOOKUP_RCU
) {
998 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
999 return ERR_PTR(-ECHILD
);
1003 path_put(&nd
->path
);
1004 nd
->path
= nd
->root
;
1005 path_get(&nd
->root
);
1006 nd
->inode
= nd
->path
.dentry
->d_inode
;
1007 nd
->flags
|= LOOKUP_JUMPED
;
1008 while (unlikely(*++res
== '/'))
1016 static int follow_up_rcu(struct path
*path
)
1018 struct mount
*mnt
= real_mount(path
->mnt
);
1019 struct mount
*parent
;
1020 struct dentry
*mountpoint
;
1022 parent
= mnt
->mnt_parent
;
1023 if (&parent
->mnt
== path
->mnt
)
1025 mountpoint
= mnt
->mnt_mountpoint
;
1026 path
->dentry
= mountpoint
;
1027 path
->mnt
= &parent
->mnt
;
1032 * follow_up - Find the mountpoint of path's vfsmount
1034 * Given a path, find the mountpoint of its source file system.
1035 * Replace @path with the path of the mountpoint in the parent mount.
1038 * Return 1 if we went up a level and 0 if we were already at the
1041 int follow_up(struct path
*path
)
1043 struct mount
*mnt
= real_mount(path
->mnt
);
1044 struct mount
*parent
;
1045 struct dentry
*mountpoint
;
1047 read_seqlock_excl(&mount_lock
);
1048 parent
= mnt
->mnt_parent
;
1049 if (parent
== mnt
) {
1050 read_sequnlock_excl(&mount_lock
);
1053 mntget(&parent
->mnt
);
1054 mountpoint
= dget(mnt
->mnt_mountpoint
);
1055 read_sequnlock_excl(&mount_lock
);
1057 path
->dentry
= mountpoint
;
1059 path
->mnt
= &parent
->mnt
;
1062 EXPORT_SYMBOL(follow_up
);
1065 * Perform an automount
1066 * - return -EISDIR to tell follow_managed() to stop and return the path we
1069 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
1072 struct vfsmount
*mnt
;
1075 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
1078 /* We don't want to mount if someone's just doing a stat -
1079 * unless they're stat'ing a directory and appended a '/' to
1082 * We do, however, want to mount if someone wants to open or
1083 * create a file of any type under the mountpoint, wants to
1084 * traverse through the mountpoint or wants to open the
1085 * mounted directory. Also, autofs may mark negative dentries
1086 * as being automount points. These will need the attentions
1087 * of the daemon to instantiate them before they can be used.
1089 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1090 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1091 path
->dentry
->d_inode
)
1094 nd
->total_link_count
++;
1095 if (nd
->total_link_count
>= 40)
1098 mnt
= path
->dentry
->d_op
->d_automount(path
);
1101 * The filesystem is allowed to return -EISDIR here to indicate
1102 * it doesn't want to automount. For instance, autofs would do
1103 * this so that its userspace daemon can mount on this dentry.
1105 * However, we can only permit this if it's a terminal point in
1106 * the path being looked up; if it wasn't then the remainder of
1107 * the path is inaccessible and we should say so.
1109 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1111 return PTR_ERR(mnt
);
1114 if (!mnt
) /* mount collision */
1117 if (!*need_mntput
) {
1118 /* lock_mount() may release path->mnt on error */
1120 *need_mntput
= true;
1122 err
= finish_automount(mnt
, path
);
1126 /* Someone else made a mount here whilst we were busy */
1131 path
->dentry
= dget(mnt
->mnt_root
);
1140 * Handle a dentry that is managed in some way.
1141 * - Flagged for transit management (autofs)
1142 * - Flagged as mountpoint
1143 * - Flagged as automount point
1145 * This may only be called in refwalk mode.
1147 * Serialization is taken care of in namespace.c
1149 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1151 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1153 bool need_mntput
= false;
1156 /* Given that we're not holding a lock here, we retain the value in a
1157 * local variable for each dentry as we look at it so that we don't see
1158 * the components of that value change under us */
1159 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1160 managed
&= DCACHE_MANAGED_DENTRY
,
1161 unlikely(managed
!= 0)) {
1162 /* Allow the filesystem to manage the transit without i_mutex
1164 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1165 BUG_ON(!path
->dentry
->d_op
);
1166 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1167 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1172 /* Transit to a mounted filesystem. */
1173 if (managed
& DCACHE_MOUNTED
) {
1174 struct vfsmount
*mounted
= lookup_mnt(path
);
1179 path
->mnt
= mounted
;
1180 path
->dentry
= dget(mounted
->mnt_root
);
1185 /* Something is mounted on this dentry in another
1186 * namespace and/or whatever was mounted there in this
1187 * namespace got unmounted before lookup_mnt() could
1191 /* Handle an automount point */
1192 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1193 ret
= follow_automount(path
, nd
, &need_mntput
);
1199 /* We didn't change the current path point */
1203 if (need_mntput
&& path
->mnt
== mnt
)
1208 nd
->flags
|= LOOKUP_JUMPED
;
1209 if (unlikely(ret
< 0))
1210 path_put_conditional(path
, nd
);
1214 int follow_down_one(struct path
*path
)
1216 struct vfsmount
*mounted
;
1218 mounted
= lookup_mnt(path
);
1222 path
->mnt
= mounted
;
1223 path
->dentry
= dget(mounted
->mnt_root
);
1228 EXPORT_SYMBOL(follow_down_one
);
1230 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1232 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1233 dentry
->d_op
->d_manage(dentry
, true) : 0;
1237 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1238 * we meet a managed dentry that would need blocking.
1240 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1241 struct inode
**inode
, unsigned *seqp
)
1244 struct mount
*mounted
;
1246 * Don't forget we might have a non-mountpoint managed dentry
1247 * that wants to block transit.
1249 switch (managed_dentry_rcu(path
->dentry
)) {
1259 if (!d_mountpoint(path
->dentry
))
1260 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1262 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1265 path
->mnt
= &mounted
->mnt
;
1266 path
->dentry
= mounted
->mnt
.mnt_root
;
1267 nd
->flags
|= LOOKUP_JUMPED
;
1268 *seqp
= read_seqcount_begin(&path
->dentry
->d_seq
);
1270 * Update the inode too. We don't need to re-check the
1271 * dentry sequence number here after this d_inode read,
1272 * because a mount-point is always pinned.
1274 *inode
= path
->dentry
->d_inode
;
1276 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1277 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1280 static int follow_dotdot_rcu(struct nameidata
*nd
)
1282 struct inode
*inode
= nd
->inode
;
1287 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1288 nd
->path
.mnt
== nd
->root
.mnt
) {
1291 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1292 struct dentry
*old
= nd
->path
.dentry
;
1293 struct dentry
*parent
= old
->d_parent
;
1296 inode
= parent
->d_inode
;
1297 seq
= read_seqcount_begin(&parent
->d_seq
);
1298 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1300 nd
->path
.dentry
= parent
;
1304 if (!follow_up_rcu(&nd
->path
))
1306 inode
= nd
->path
.dentry
->d_inode
;
1307 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1309 while (d_mountpoint(nd
->path
.dentry
)) {
1310 struct mount
*mounted
;
1311 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1314 nd
->path
.mnt
= &mounted
->mnt
;
1315 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1316 inode
= nd
->path
.dentry
->d_inode
;
1317 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1318 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1329 * Follow down to the covering mount currently visible to userspace. At each
1330 * point, the filesystem owning that dentry may be queried as to whether the
1331 * caller is permitted to proceed or not.
1333 int follow_down(struct path
*path
)
1338 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1339 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1340 /* Allow the filesystem to manage the transit without i_mutex
1343 * We indicate to the filesystem if someone is trying to mount
1344 * something here. This gives autofs the chance to deny anyone
1345 * other than its daemon the right to mount on its
1348 * The filesystem may sleep at this point.
1350 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1351 BUG_ON(!path
->dentry
->d_op
);
1352 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1353 ret
= path
->dentry
->d_op
->d_manage(
1354 path
->dentry
, false);
1356 return ret
== -EISDIR
? 0 : ret
;
1359 /* Transit to a mounted filesystem. */
1360 if (managed
& DCACHE_MOUNTED
) {
1361 struct vfsmount
*mounted
= lookup_mnt(path
);
1366 path
->mnt
= mounted
;
1367 path
->dentry
= dget(mounted
->mnt_root
);
1371 /* Don't handle automount points here */
1376 EXPORT_SYMBOL(follow_down
);
1379 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1381 static void follow_mount(struct path
*path
)
1383 while (d_mountpoint(path
->dentry
)) {
1384 struct vfsmount
*mounted
= lookup_mnt(path
);
1389 path
->mnt
= mounted
;
1390 path
->dentry
= dget(mounted
->mnt_root
);
1394 static void follow_dotdot(struct nameidata
*nd
)
1400 struct dentry
*old
= nd
->path
.dentry
;
1402 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1403 nd
->path
.mnt
== nd
->root
.mnt
) {
1406 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1407 /* rare case of legitimate dget_parent()... */
1408 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1412 if (!follow_up(&nd
->path
))
1415 follow_mount(&nd
->path
);
1416 nd
->inode
= nd
->path
.dentry
->d_inode
;
1420 * This looks up the name in dcache, possibly revalidates the old dentry and
1421 * allocates a new one if not found or not valid. In the need_lookup argument
1422 * returns whether i_op->lookup is necessary.
1424 * dir->d_inode->i_mutex must be held
1426 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1427 unsigned int flags
, bool *need_lookup
)
1429 struct dentry
*dentry
;
1432 *need_lookup
= false;
1433 dentry
= d_lookup(dir
, name
);
1435 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1436 error
= d_revalidate(dentry
, flags
);
1437 if (unlikely(error
<= 0)) {
1440 return ERR_PTR(error
);
1442 d_invalidate(dentry
);
1451 dentry
= d_alloc(dir
, name
);
1452 if (unlikely(!dentry
))
1453 return ERR_PTR(-ENOMEM
);
1455 *need_lookup
= true;
1461 * Call i_op->lookup on the dentry. The dentry must be negative and
1464 * dir->d_inode->i_mutex must be held
1466 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1471 /* Don't create child dentry for a dead directory. */
1472 if (unlikely(IS_DEADDIR(dir
))) {
1474 return ERR_PTR(-ENOENT
);
1477 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1478 if (unlikely(old
)) {
1485 static struct dentry
*__lookup_hash(struct qstr
*name
,
1486 struct dentry
*base
, unsigned int flags
)
1489 struct dentry
*dentry
;
1491 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1495 return lookup_real(base
->d_inode
, dentry
, flags
);
1499 * It's more convoluted than I'd like it to be, but... it's still fairly
1500 * small and for now I'd prefer to have fast path as straight as possible.
1501 * It _is_ time-critical.
1503 static int lookup_fast(struct nameidata
*nd
,
1504 struct path
*path
, struct inode
**inode
,
1507 struct vfsmount
*mnt
= nd
->path
.mnt
;
1508 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1514 * Rename seqlock is not required here because in the off chance
1515 * of a false negative due to a concurrent rename, we're going to
1516 * do the non-racy lookup, below.
1518 if (nd
->flags
& LOOKUP_RCU
) {
1521 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1526 * This sequence count validates that the inode matches
1527 * the dentry name information from lookup.
1529 *inode
= d_backing_inode(dentry
);
1530 negative
= d_is_negative(dentry
);
1531 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1537 * This sequence count validates that the parent had no
1538 * changes while we did the lookup of the dentry above.
1540 * The memory barrier in read_seqcount_begin of child is
1541 * enough, we can use __read_seqcount_retry here.
1543 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1547 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1548 status
= d_revalidate(dentry
, nd
->flags
);
1549 if (unlikely(status
<= 0)) {
1550 if (status
!= -ECHILD
)
1556 path
->dentry
= dentry
;
1557 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1560 if (unlazy_walk(nd
, dentry
, seq
))
1563 dentry
= __d_lookup(parent
, &nd
->last
);
1566 if (unlikely(!dentry
))
1569 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1570 status
= d_revalidate(dentry
, nd
->flags
);
1571 if (unlikely(status
<= 0)) {
1576 d_invalidate(dentry
);
1581 if (unlikely(d_is_negative(dentry
))) {
1586 path
->dentry
= dentry
;
1587 err
= follow_managed(path
, nd
);
1589 *inode
= d_backing_inode(path
->dentry
);
1596 /* Fast lookup failed, do it the slow way */
1597 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1599 struct dentry
*dentry
, *parent
;
1601 parent
= nd
->path
.dentry
;
1602 BUG_ON(nd
->inode
!= parent
->d_inode
);
1604 mutex_lock(&parent
->d_inode
->i_mutex
);
1605 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1606 mutex_unlock(&parent
->d_inode
->i_mutex
);
1608 return PTR_ERR(dentry
);
1609 path
->mnt
= nd
->path
.mnt
;
1610 path
->dentry
= dentry
;
1611 return follow_managed(path
, nd
);
1614 static inline int may_lookup(struct nameidata
*nd
)
1616 if (nd
->flags
& LOOKUP_RCU
) {
1617 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1620 if (unlazy_walk(nd
, NULL
, 0))
1623 return inode_permission(nd
->inode
, MAY_EXEC
);
1626 static inline int handle_dots(struct nameidata
*nd
, int type
)
1628 if (type
== LAST_DOTDOT
) {
1629 if (nd
->flags
& LOOKUP_RCU
) {
1630 return follow_dotdot_rcu(nd
);
1637 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1638 struct inode
*inode
, unsigned seq
)
1642 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1643 path_to_nameidata(link
, nd
);
1646 if (!(nd
->flags
& LOOKUP_RCU
)) {
1647 if (link
->mnt
== nd
->path
.mnt
)
1650 error
= nd_alloc_stack(nd
);
1651 if (unlikely(error
)) {
1652 if (error
== -ECHILD
) {
1653 if (unlikely(unlazy_link(nd
, link
, seq
)))
1655 error
= nd_alloc_stack(nd
);
1663 last
= nd
->stack
+ nd
->depth
++;
1665 last
->cookie
= NULL
;
1666 last
->inode
= inode
;
1672 * Do we need to follow links? We _really_ want to be able
1673 * to do this check without having to look at inode->i_op,
1674 * so we keep a cache of "no, this doesn't need follow_link"
1675 * for the common case.
1677 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
,
1679 struct inode
*inode
, unsigned seq
)
1681 if (likely(!d_is_symlink(link
->dentry
)))
1685 return pick_link(nd
, link
, inode
, seq
);
1688 enum {WALK_GET
= 1, WALK_PUT
= 2};
1690 static int walk_component(struct nameidata
*nd
, int flags
)
1693 struct inode
*inode
;
1697 * "." and ".." are special - ".." especially so because it has
1698 * to be able to know about the current root directory and
1699 * parent relationships.
1701 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1702 err
= handle_dots(nd
, nd
->last_type
);
1703 if (flags
& WALK_PUT
)
1707 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1708 if (unlikely(err
)) {
1712 err
= lookup_slow(nd
, &path
);
1716 inode
= d_backing_inode(path
.dentry
);
1717 seq
= 0; /* we are already out of RCU mode */
1719 if (d_is_negative(path
.dentry
))
1723 if (flags
& WALK_PUT
)
1725 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
, inode
, seq
);
1728 path_to_nameidata(&path
, nd
);
1734 path_to_nameidata(&path
, nd
);
1739 * We can do the critical dentry name comparison and hashing
1740 * operations one word at a time, but we are limited to:
1742 * - Architectures with fast unaligned word accesses. We could
1743 * do a "get_unaligned()" if this helps and is sufficiently
1746 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1747 * do not trap on the (extremely unlikely) case of a page
1748 * crossing operation.
1750 * - Furthermore, we need an efficient 64-bit compile for the
1751 * 64-bit case in order to generate the "number of bytes in
1752 * the final mask". Again, that could be replaced with a
1753 * efficient population count instruction or similar.
1755 #ifdef CONFIG_DCACHE_WORD_ACCESS
1757 #include <asm/word-at-a-time.h>
1761 static inline unsigned int fold_hash(unsigned long hash
)
1763 return hash_64(hash
, 32);
1766 #else /* 32-bit case */
1768 #define fold_hash(x) (x)
1772 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1774 unsigned long a
, mask
;
1775 unsigned long hash
= 0;
1778 a
= load_unaligned_zeropad(name
);
1779 if (len
< sizeof(unsigned long))
1783 name
+= sizeof(unsigned long);
1784 len
-= sizeof(unsigned long);
1788 mask
= bytemask_from_count(len
);
1791 return fold_hash(hash
);
1793 EXPORT_SYMBOL(full_name_hash
);
1796 * Calculate the length and hash of the path component, and
1797 * return the "hash_len" as the result.
1799 static inline u64
hash_name(const char *name
)
1801 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1802 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1805 len
= -sizeof(unsigned long);
1807 hash
= (hash
+ a
) * 9;
1808 len
+= sizeof(unsigned long);
1809 a
= load_unaligned_zeropad(name
+len
);
1810 b
= a
^ REPEAT_BYTE('/');
1811 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1813 adata
= prep_zero_mask(a
, adata
, &constants
);
1814 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1816 mask
= create_zero_mask(adata
| bdata
);
1818 hash
+= a
& zero_bytemask(mask
);
1819 len
+= find_zero(mask
);
1820 return hashlen_create(fold_hash(hash
), len
);
1825 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1827 unsigned long hash
= init_name_hash();
1829 hash
= partial_name_hash(*name
++, hash
);
1830 return end_name_hash(hash
);
1832 EXPORT_SYMBOL(full_name_hash
);
1835 * We know there's a real path component here of at least
1838 static inline u64
hash_name(const char *name
)
1840 unsigned long hash
= init_name_hash();
1841 unsigned long len
= 0, c
;
1843 c
= (unsigned char)*name
;
1846 hash
= partial_name_hash(c
, hash
);
1847 c
= (unsigned char)name
[len
];
1848 } while (c
&& c
!= '/');
1849 return hashlen_create(end_name_hash(hash
), len
);
1856 * This is the basic name resolution function, turning a pathname into
1857 * the final dentry. We expect 'base' to be positive and a directory.
1859 * Returns 0 and nd will have valid dentry and mnt on success.
1860 * Returns error and drops reference to input namei data on failure.
1862 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1871 /* At this point we know we have a real path component. */
1876 err
= may_lookup(nd
);
1880 hash_len
= hash_name(name
);
1883 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1885 if (name
[1] == '.') {
1887 nd
->flags
|= LOOKUP_JUMPED
;
1893 if (likely(type
== LAST_NORM
)) {
1894 struct dentry
*parent
= nd
->path
.dentry
;
1895 nd
->flags
&= ~LOOKUP_JUMPED
;
1896 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1897 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1898 err
= parent
->d_op
->d_hash(parent
, &this);
1901 hash_len
= this.hash_len
;
1906 nd
->last
.hash_len
= hash_len
;
1907 nd
->last
.name
= name
;
1908 nd
->last_type
= type
;
1910 name
+= hashlen_len(hash_len
);
1914 * If it wasn't NUL, we know it was '/'. Skip that
1915 * slash, and continue until no more slashes.
1919 } while (unlikely(*name
== '/'));
1920 if (unlikely(!*name
)) {
1922 /* pathname body, done */
1925 name
= nd
->stack
[nd
->depth
- 1].name
;
1926 /* trailing symlink, done */
1929 /* last component of nested symlink */
1930 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
1932 err
= walk_component(nd
, WALK_GET
);
1938 const char *s
= get_link(nd
);
1940 if (unlikely(IS_ERR(s
)))
1947 nd
->stack
[nd
->depth
- 1].name
= name
;
1952 if (unlikely(!d_can_lookup(nd
->path
.dentry
)))
1957 static const char *path_init(int dfd
, const struct filename
*name
,
1958 unsigned int flags
, struct nameidata
*nd
)
1961 const char *s
= name
->name
;
1963 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1964 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1966 nd
->total_link_count
= 0;
1967 if (flags
& LOOKUP_ROOT
) {
1968 struct dentry
*root
= nd
->root
.dentry
;
1969 struct inode
*inode
= root
->d_inode
;
1971 if (!d_can_lookup(root
))
1972 return ERR_PTR(-ENOTDIR
);
1973 retval
= inode_permission(inode
, MAY_EXEC
);
1975 return ERR_PTR(retval
);
1977 nd
->path
= nd
->root
;
1979 if (flags
& LOOKUP_RCU
) {
1981 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1982 nd
->m_seq
= read_seqbegin(&mount_lock
);
1984 path_get(&nd
->path
);
1989 nd
->root
.mnt
= NULL
;
1991 nd
->m_seq
= read_seqbegin(&mount_lock
);
1993 if (flags
& LOOKUP_RCU
) {
1995 nd
->seq
= set_root_rcu(nd
);
1998 path_get(&nd
->root
);
2000 nd
->path
= nd
->root
;
2001 } else if (dfd
== AT_FDCWD
) {
2002 if (flags
& LOOKUP_RCU
) {
2003 struct fs_struct
*fs
= current
->fs
;
2009 seq
= read_seqcount_begin(&fs
->seq
);
2011 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2012 } while (read_seqcount_retry(&fs
->seq
, seq
));
2014 get_fs_pwd(current
->fs
, &nd
->path
);
2017 /* Caller must check execute permissions on the starting path component */
2018 struct fd f
= fdget_raw(dfd
);
2019 struct dentry
*dentry
;
2022 return ERR_PTR(-EBADF
);
2024 dentry
= f
.file
->f_path
.dentry
;
2027 if (!d_can_lookup(dentry
)) {
2029 return ERR_PTR(-ENOTDIR
);
2033 nd
->path
= f
.file
->f_path
;
2034 if (flags
& LOOKUP_RCU
) {
2036 nd
->inode
= nd
->path
.dentry
->d_inode
;
2037 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2039 path_get(&nd
->path
);
2040 nd
->inode
= nd
->path
.dentry
->d_inode
;
2046 nd
->inode
= nd
->path
.dentry
->d_inode
;
2047 if (!(flags
& LOOKUP_RCU
))
2049 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
2051 if (!(nd
->flags
& LOOKUP_ROOT
))
2052 nd
->root
.mnt
= NULL
;
2054 return ERR_PTR(-ECHILD
);
2057 static void path_cleanup(struct nameidata
*nd
)
2059 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2060 path_put(&nd
->root
);
2061 nd
->root
.mnt
= NULL
;
2065 static const char *trailing_symlink(struct nameidata
*nd
)
2068 int error
= may_follow_link(nd
);
2069 if (unlikely(error
))
2070 return ERR_PTR(error
);
2071 nd
->flags
|= LOOKUP_PARENT
;
2072 nd
->stack
[0].name
= NULL
;
2077 static inline int lookup_last(struct nameidata
*nd
)
2079 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2080 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2082 nd
->flags
&= ~LOOKUP_PARENT
;
2083 return walk_component(nd
,
2084 nd
->flags
& LOOKUP_FOLLOW
2086 ? WALK_PUT
| WALK_GET
2091 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2092 static int path_lookupat(int dfd
, const struct filename
*name
,
2093 unsigned int flags
, struct nameidata
*nd
)
2095 const char *s
= path_init(dfd
, name
, flags
, nd
);
2100 while (!(err
= link_path_walk(s
, nd
))
2101 && ((err
= lookup_last(nd
)) > 0)) {
2102 s
= trailing_symlink(nd
);
2109 err
= complete_walk(nd
);
2111 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2112 if (!d_can_lookup(nd
->path
.dentry
))
2121 static int filename_lookup(int dfd
, struct filename
*name
,
2122 unsigned int flags
, struct nameidata
*nd
)
2125 struct nameidata
*saved_nd
= set_nameidata(nd
);
2127 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2128 if (unlikely(retval
== -ECHILD
))
2129 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2130 if (unlikely(retval
== -ESTALE
))
2131 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2133 if (likely(!retval
))
2134 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2135 restore_nameidata(saved_nd
);
2139 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2140 static int path_parentat(int dfd
, const struct filename
*name
,
2141 unsigned int flags
, struct nameidata
*nd
)
2143 const char *s
= path_init(dfd
, name
, flags
, nd
);
2147 err
= link_path_walk(s
, nd
);
2149 err
= complete_walk(nd
);
2156 static int filename_parentat(int dfd
, struct filename
*name
,
2157 unsigned int flags
, struct nameidata
*nd
)
2160 struct nameidata
*saved_nd
= set_nameidata(nd
);
2162 retval
= path_parentat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2163 if (unlikely(retval
== -ECHILD
))
2164 retval
= path_parentat(dfd
, name
, flags
, nd
);
2165 if (unlikely(retval
== -ESTALE
))
2166 retval
= path_parentat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2168 if (likely(!retval
))
2169 audit_inode(name
, nd
->path
.dentry
, LOOKUP_PARENT
);
2170 restore_nameidata(saved_nd
);
2174 /* does lookup, returns the object with parent locked */
2175 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2177 struct filename
*filename
= getname_kernel(name
);
2178 struct nameidata nd
;
2182 if (IS_ERR(filename
))
2183 return ERR_CAST(filename
);
2185 err
= filename_parentat(AT_FDCWD
, filename
, 0, &nd
);
2190 if (nd
.last_type
!= LAST_NORM
) {
2192 d
= ERR_PTR(-EINVAL
);
2195 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2196 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2198 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2208 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2210 struct nameidata nd
;
2211 struct filename
*filename
= getname_kernel(name
);
2212 int res
= PTR_ERR(filename
);
2214 if (!IS_ERR(filename
)) {
2215 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2222 EXPORT_SYMBOL(kern_path
);
2225 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2226 * @dentry: pointer to dentry of the base directory
2227 * @mnt: pointer to vfs mount of the base directory
2228 * @name: pointer to file name
2229 * @flags: lookup flags
2230 * @path: pointer to struct path to fill
2232 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2233 const char *name
, unsigned int flags
,
2236 struct filename
*filename
= getname_kernel(name
);
2237 int err
= PTR_ERR(filename
);
2239 BUG_ON(flags
& LOOKUP_PARENT
);
2241 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2242 if (!IS_ERR(filename
)) {
2243 struct nameidata nd
;
2244 nd
.root
.dentry
= dentry
;
2246 err
= filename_lookup(AT_FDCWD
, filename
,
2247 flags
| LOOKUP_ROOT
, &nd
);
2254 EXPORT_SYMBOL(vfs_path_lookup
);
2257 * lookup_one_len - filesystem helper to lookup single pathname component
2258 * @name: pathname component to lookup
2259 * @base: base directory to lookup from
2260 * @len: maximum length @len should be interpreted to
2262 * Note that this routine is purely a helper for filesystem usage and should
2263 * not be called by generic code.
2265 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2271 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2275 this.hash
= full_name_hash(name
, len
);
2277 return ERR_PTR(-EACCES
);
2279 if (unlikely(name
[0] == '.')) {
2280 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2281 return ERR_PTR(-EACCES
);
2285 c
= *(const unsigned char *)name
++;
2286 if (c
== '/' || c
== '\0')
2287 return ERR_PTR(-EACCES
);
2290 * See if the low-level filesystem might want
2291 * to use its own hash..
2293 if (base
->d_flags
& DCACHE_OP_HASH
) {
2294 int err
= base
->d_op
->d_hash(base
, &this);
2296 return ERR_PTR(err
);
2299 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2301 return ERR_PTR(err
);
2303 return __lookup_hash(&this, base
, 0);
2305 EXPORT_SYMBOL(lookup_one_len
);
2307 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2308 struct path
*path
, int *empty
)
2310 struct nameidata nd
;
2311 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2312 int err
= PTR_ERR(tmp
);
2315 BUG_ON(flags
& LOOKUP_PARENT
);
2317 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2325 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2328 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2330 EXPORT_SYMBOL(user_path_at
);
2333 * NB: most callers don't do anything directly with the reference to the
2334 * to struct filename, but the nd->last pointer points into the name string
2335 * allocated by getname. So we must hold the reference to it until all
2336 * path-walking is complete.
2338 static struct filename
*
2339 user_path_parent(int dfd
, const char __user
*path
,
2340 struct path
*parent
,
2345 struct nameidata nd
;
2346 struct filename
*s
= getname(path
);
2349 /* only LOOKUP_REVAL is allowed in extra flags */
2350 flags
&= LOOKUP_REVAL
;
2355 error
= filename_parentat(dfd
, s
, flags
, &nd
);
2358 return ERR_PTR(error
);
2362 *type
= nd
.last_type
;
2368 * mountpoint_last - look up last component for umount
2369 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2370 * @path: pointer to container for result
2372 * This is a special lookup_last function just for umount. In this case, we
2373 * need to resolve the path without doing any revalidation.
2375 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2376 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2377 * in almost all cases, this lookup will be served out of the dcache. The only
2378 * cases where it won't are if nd->last refers to a symlink or the path is
2379 * bogus and it doesn't exist.
2382 * -error: if there was an error during lookup. This includes -ENOENT if the
2383 * lookup found a negative dentry. The nd->path reference will also be
2386 * 0: if we successfully resolved nd->path and found it to not to be a
2387 * symlink that needs to be followed. "path" will also be populated.
2388 * The nd->path reference will also be put.
2390 * 1: if we successfully resolved nd->last and found it to be a symlink
2391 * that needs to be followed. "path" will be populated with the path
2392 * to the link, and nd->path will *not* be put.
2395 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2398 struct dentry
*dentry
;
2399 struct dentry
*dir
= nd
->path
.dentry
;
2401 /* If we're in rcuwalk, drop out of it to handle last component */
2402 if (nd
->flags
& LOOKUP_RCU
) {
2403 if (unlazy_walk(nd
, NULL
, 0))
2407 nd
->flags
&= ~LOOKUP_PARENT
;
2409 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2410 error
= handle_dots(nd
, nd
->last_type
);
2413 dentry
= dget(nd
->path
.dentry
);
2417 mutex_lock(&dir
->d_inode
->i_mutex
);
2418 dentry
= d_lookup(dir
, &nd
->last
);
2421 * No cached dentry. Mounted dentries are pinned in the cache,
2422 * so that means that this dentry is probably a symlink or the
2423 * path doesn't actually point to a mounted dentry.
2425 dentry
= d_alloc(dir
, &nd
->last
);
2427 mutex_unlock(&dir
->d_inode
->i_mutex
);
2430 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2431 if (IS_ERR(dentry
)) {
2432 mutex_unlock(&dir
->d_inode
->i_mutex
);
2433 return PTR_ERR(dentry
);
2436 mutex_unlock(&dir
->d_inode
->i_mutex
);
2439 if (d_is_negative(dentry
)) {
2445 path
->dentry
= dentry
;
2446 path
->mnt
= nd
->path
.mnt
;
2447 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
,
2448 d_backing_inode(dentry
), 0);
2449 if (unlikely(error
))
2457 * path_mountpoint - look up a path to be umounted
2458 * @dfd: directory file descriptor to start walk from
2459 * @name: full pathname to walk
2460 * @path: pointer to container for result
2461 * @flags: lookup flags
2463 * Look up the given name, but don't attempt to revalidate the last component.
2464 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2467 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2468 struct nameidata
*nd
, unsigned int flags
)
2470 const char *s
= path_init(dfd
, name
, flags
, nd
);
2474 while (!(err
= link_path_walk(s
, nd
)) &&
2475 (err
= mountpoint_last(nd
, path
)) > 0) {
2476 s
= trailing_symlink(nd
);
2488 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2491 struct nameidata nd
, *saved
;
2494 return PTR_ERR(name
);
2495 saved
= set_nameidata(&nd
);
2496 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_RCU
);
2497 if (unlikely(error
== -ECHILD
))
2498 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
);
2499 if (unlikely(error
== -ESTALE
))
2500 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_REVAL
);
2502 audit_inode(name
, path
->dentry
, 0);
2503 restore_nameidata(saved
);
2509 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2510 * @dfd: directory file descriptor
2511 * @name: pathname from userland
2512 * @flags: lookup flags
2513 * @path: pointer to container to hold result
2515 * A umount is a special case for path walking. We're not actually interested
2516 * in the inode in this situation, and ESTALE errors can be a problem. We
2517 * simply want track down the dentry and vfsmount attached at the mountpoint
2518 * and avoid revalidating the last component.
2520 * Returns 0 and populates "path" on success.
2523 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2526 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2530 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2533 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2535 EXPORT_SYMBOL(kern_path_mountpoint
);
2537 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2539 kuid_t fsuid
= current_fsuid();
2541 if (uid_eq(inode
->i_uid
, fsuid
))
2543 if (uid_eq(dir
->i_uid
, fsuid
))
2545 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2547 EXPORT_SYMBOL(__check_sticky
);
2550 * Check whether we can remove a link victim from directory dir, check
2551 * whether the type of victim is right.
2552 * 1. We can't do it if dir is read-only (done in permission())
2553 * 2. We should have write and exec permissions on dir
2554 * 3. We can't remove anything from append-only dir
2555 * 4. We can't do anything with immutable dir (done in permission())
2556 * 5. If the sticky bit on dir is set we should either
2557 * a. be owner of dir, or
2558 * b. be owner of victim, or
2559 * c. have CAP_FOWNER capability
2560 * 6. If the victim is append-only or immutable we can't do antyhing with
2561 * links pointing to it.
2562 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2563 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2564 * 9. We can't remove a root or mountpoint.
2565 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2566 * nfs_async_unlink().
2568 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2570 struct inode
*inode
= d_backing_inode(victim
);
2573 if (d_is_negative(victim
))
2577 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2578 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2580 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2586 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2587 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2590 if (!d_is_dir(victim
))
2592 if (IS_ROOT(victim
))
2594 } else if (d_is_dir(victim
))
2596 if (IS_DEADDIR(dir
))
2598 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2603 /* Check whether we can create an object with dentry child in directory
2605 * 1. We can't do it if child already exists (open has special treatment for
2606 * this case, but since we are inlined it's OK)
2607 * 2. We can't do it if dir is read-only (done in permission())
2608 * 3. We should have write and exec permissions on dir
2609 * 4. We can't do it if dir is immutable (done in permission())
2611 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2613 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2616 if (IS_DEADDIR(dir
))
2618 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2622 * p1 and p2 should be directories on the same fs.
2624 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2629 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2633 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2635 p
= d_ancestor(p2
, p1
);
2637 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2638 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2642 p
= d_ancestor(p1
, p2
);
2644 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2645 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2649 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2650 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2653 EXPORT_SYMBOL(lock_rename
);
2655 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2657 mutex_unlock(&p1
->d_inode
->i_mutex
);
2659 mutex_unlock(&p2
->d_inode
->i_mutex
);
2660 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2663 EXPORT_SYMBOL(unlock_rename
);
2665 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2668 int error
= may_create(dir
, dentry
);
2672 if (!dir
->i_op
->create
)
2673 return -EACCES
; /* shouldn't it be ENOSYS? */
2676 error
= security_inode_create(dir
, dentry
, mode
);
2679 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2681 fsnotify_create(dir
, dentry
);
2684 EXPORT_SYMBOL(vfs_create
);
2686 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2688 struct dentry
*dentry
= path
->dentry
;
2689 struct inode
*inode
= dentry
->d_inode
;
2699 switch (inode
->i_mode
& S_IFMT
) {
2703 if (acc_mode
& MAY_WRITE
)
2708 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2717 error
= inode_permission(inode
, acc_mode
);
2722 * An append-only file must be opened in append mode for writing.
2724 if (IS_APPEND(inode
)) {
2725 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2731 /* O_NOATIME can only be set by the owner or superuser */
2732 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2738 static int handle_truncate(struct file
*filp
)
2740 struct path
*path
= &filp
->f_path
;
2741 struct inode
*inode
= path
->dentry
->d_inode
;
2742 int error
= get_write_access(inode
);
2746 * Refuse to truncate files with mandatory locks held on them.
2748 error
= locks_verify_locked(filp
);
2750 error
= security_path_truncate(path
);
2752 error
= do_truncate(path
->dentry
, 0,
2753 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2756 put_write_access(inode
);
2760 static inline int open_to_namei_flags(int flag
)
2762 if ((flag
& O_ACCMODE
) == 3)
2767 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2769 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2773 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2777 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2781 * Attempt to atomically look up, create and open a file from a negative
2784 * Returns 0 if successful. The file will have been created and attached to
2785 * @file by the filesystem calling finish_open().
2787 * Returns 1 if the file was looked up only or didn't need creating. The
2788 * caller will need to perform the open themselves. @path will have been
2789 * updated to point to the new dentry. This may be negative.
2791 * Returns an error code otherwise.
2793 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2794 struct path
*path
, struct file
*file
,
2795 const struct open_flags
*op
,
2796 bool got_write
, bool need_lookup
,
2799 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2800 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2804 int create_error
= 0;
2805 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2808 BUG_ON(dentry
->d_inode
);
2810 /* Don't create child dentry for a dead directory. */
2811 if (unlikely(IS_DEADDIR(dir
))) {
2817 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2818 mode
&= ~current_umask();
2820 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2822 open_flag
&= ~O_TRUNC
;
2825 * Checking write permission is tricky, bacuse we don't know if we are
2826 * going to actually need it: O_CREAT opens should work as long as the
2827 * file exists. But checking existence breaks atomicity. The trick is
2828 * to check access and if not granted clear O_CREAT from the flags.
2830 * Another problem is returing the "right" error value (e.g. for an
2831 * O_EXCL open we want to return EEXIST not EROFS).
2833 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2834 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2835 if (!(open_flag
& O_CREAT
)) {
2837 * No O_CREATE -> atomicity not a requirement -> fall
2838 * back to lookup + open
2841 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2842 /* Fall back and fail with the right error */
2843 create_error
= -EROFS
;
2846 /* No side effects, safe to clear O_CREAT */
2847 create_error
= -EROFS
;
2848 open_flag
&= ~O_CREAT
;
2852 if (open_flag
& O_CREAT
) {
2853 error
= may_o_create(&nd
->path
, dentry
, mode
);
2855 create_error
= error
;
2856 if (open_flag
& O_EXCL
)
2858 open_flag
&= ~O_CREAT
;
2862 if (nd
->flags
& LOOKUP_DIRECTORY
)
2863 open_flag
|= O_DIRECTORY
;
2865 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2866 file
->f_path
.mnt
= nd
->path
.mnt
;
2867 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2870 if (create_error
&& error
== -ENOENT
)
2871 error
= create_error
;
2875 if (error
) { /* returned 1, that is */
2876 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2880 if (file
->f_path
.dentry
) {
2882 dentry
= file
->f_path
.dentry
;
2884 if (*opened
& FILE_CREATED
)
2885 fsnotify_create(dir
, dentry
);
2886 if (!dentry
->d_inode
) {
2887 WARN_ON(*opened
& FILE_CREATED
);
2889 error
= create_error
;
2893 if (excl
&& !(*opened
& FILE_CREATED
)) {
2902 * We didn't have the inode before the open, so check open permission
2905 acc_mode
= op
->acc_mode
;
2906 if (*opened
& FILE_CREATED
) {
2907 WARN_ON(!(open_flag
& O_CREAT
));
2908 fsnotify_create(dir
, dentry
);
2909 acc_mode
= MAY_OPEN
;
2911 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2921 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2923 return PTR_ERR(dentry
);
2926 int open_flag
= op
->open_flag
;
2928 error
= create_error
;
2929 if ((open_flag
& O_EXCL
)) {
2930 if (!dentry
->d_inode
)
2932 } else if (!dentry
->d_inode
) {
2934 } else if ((open_flag
& O_TRUNC
) &&
2938 /* will fail later, go on to get the right error */
2942 path
->dentry
= dentry
;
2943 path
->mnt
= nd
->path
.mnt
;
2948 * Look up and maybe create and open the last component.
2950 * Must be called with i_mutex held on parent.
2952 * Returns 0 if the file was successfully atomically created (if necessary) and
2953 * opened. In this case the file will be returned attached to @file.
2955 * Returns 1 if the file was not completely opened at this time, though lookups
2956 * and creations will have been performed and the dentry returned in @path will
2957 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2958 * specified then a negative dentry may be returned.
2960 * An error code is returned otherwise.
2962 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2963 * cleared otherwise prior to returning.
2965 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2967 const struct open_flags
*op
,
2968 bool got_write
, int *opened
)
2970 struct dentry
*dir
= nd
->path
.dentry
;
2971 struct inode
*dir_inode
= dir
->d_inode
;
2972 struct dentry
*dentry
;
2976 *opened
&= ~FILE_CREATED
;
2977 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2979 return PTR_ERR(dentry
);
2981 /* Cached positive dentry: will open in f_op->open */
2982 if (!need_lookup
&& dentry
->d_inode
)
2985 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2986 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2987 need_lookup
, opened
);
2991 BUG_ON(dentry
->d_inode
);
2993 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2995 return PTR_ERR(dentry
);
2998 /* Negative dentry, just create the file */
2999 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
3000 umode_t mode
= op
->mode
;
3001 if (!IS_POSIXACL(dir
->d_inode
))
3002 mode
&= ~current_umask();
3004 * This write is needed to ensure that a
3005 * rw->ro transition does not occur between
3006 * the time when the file is created and when
3007 * a permanent write count is taken through
3008 * the 'struct file' in finish_open().
3014 *opened
|= FILE_CREATED
;
3015 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
3018 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
3019 nd
->flags
& LOOKUP_EXCL
);
3024 path
->dentry
= dentry
;
3025 path
->mnt
= nd
->path
.mnt
;
3034 * Handle the last step of open()
3036 static int do_last(struct nameidata
*nd
,
3037 struct file
*file
, const struct open_flags
*op
,
3038 int *opened
, struct filename
*name
)
3040 struct dentry
*dir
= nd
->path
.dentry
;
3041 int open_flag
= op
->open_flag
;
3042 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3043 bool got_write
= false;
3044 int acc_mode
= op
->acc_mode
;
3046 struct inode
*inode
;
3047 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
3049 bool retried
= false;
3052 nd
->flags
&= ~LOOKUP_PARENT
;
3053 nd
->flags
|= op
->intent
;
3055 if (nd
->last_type
!= LAST_NORM
) {
3056 error
= handle_dots(nd
, nd
->last_type
);
3057 if (unlikely(error
))
3062 if (!(open_flag
& O_CREAT
)) {
3063 if (nd
->last
.name
[nd
->last
.len
])
3064 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3065 /* we _can_ be in RCU mode here */
3066 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3073 BUG_ON(nd
->inode
!= dir
->d_inode
);
3075 /* create side of things */
3077 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3078 * has been cleared when we got to the last component we are
3081 error
= complete_walk(nd
);
3085 audit_inode(name
, dir
, LOOKUP_PARENT
);
3086 /* trailing slashes? */
3087 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3092 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3093 error
= mnt_want_write(nd
->path
.mnt
);
3097 * do _not_ fail yet - we might not need that or fail with
3098 * a different error; let lookup_open() decide; we'll be
3099 * dropping this one anyway.
3102 mutex_lock(&dir
->d_inode
->i_mutex
);
3103 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3104 mutex_unlock(&dir
->d_inode
->i_mutex
);
3110 if ((*opened
& FILE_CREATED
) ||
3111 !S_ISREG(file_inode(file
)->i_mode
))
3112 will_truncate
= false;
3114 audit_inode(name
, file
->f_path
.dentry
, 0);
3118 if (*opened
& FILE_CREATED
) {
3119 /* Don't check for write permission, don't truncate */
3120 open_flag
&= ~O_TRUNC
;
3121 will_truncate
= false;
3122 acc_mode
= MAY_OPEN
;
3123 path_to_nameidata(&path
, nd
);
3124 goto finish_open_created
;
3128 * create/update audit record if it already exists.
3130 if (d_is_positive(path
.dentry
))
3131 audit_inode(name
, path
.dentry
, 0);
3134 * If atomic_open() acquired write access it is dropped now due to
3135 * possible mount and symlink following (this might be optimized away if
3139 mnt_drop_write(nd
->path
.mnt
);
3143 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3144 path_to_nameidata(&path
, nd
);
3148 error
= follow_managed(&path
, nd
);
3149 if (unlikely(error
< 0))
3152 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3153 inode
= d_backing_inode(path
.dentry
);
3154 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3155 if (unlikely(d_is_negative(path
.dentry
))) {
3156 path_to_nameidata(&path
, nd
);
3162 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
,
3164 if (unlikely(error
))
3167 if (unlikely(d_is_symlink(path
.dentry
)) && !(open_flag
& O_PATH
)) {
3168 path_to_nameidata(&path
, nd
);
3172 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3173 path_to_nameidata(&path
, nd
);
3175 save_parent
.dentry
= nd
->path
.dentry
;
3176 save_parent
.mnt
= mntget(path
.mnt
);
3177 nd
->path
.dentry
= path
.dentry
;
3182 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3184 error
= complete_walk(nd
);
3186 path_put(&save_parent
);
3189 audit_inode(name
, nd
->path
.dentry
, 0);
3191 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3194 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3196 if (!d_is_reg(nd
->path
.dentry
))
3197 will_truncate
= false;
3199 if (will_truncate
) {
3200 error
= mnt_want_write(nd
->path
.mnt
);
3205 finish_open_created
:
3206 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3210 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3211 error
= vfs_open(&nd
->path
, file
, current_cred());
3213 *opened
|= FILE_OPENED
;
3215 if (error
== -EOPENSTALE
)
3220 error
= open_check_o_direct(file
);
3223 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3227 if (will_truncate
) {
3228 error
= handle_truncate(file
);
3234 mnt_drop_write(nd
->path
.mnt
);
3235 path_put(&save_parent
);
3243 /* If no saved parent or already retried then can't retry */
3244 if (!save_parent
.dentry
|| retried
)
3247 BUG_ON(save_parent
.dentry
!= dir
);
3248 path_put(&nd
->path
);
3249 nd
->path
= save_parent
;
3250 nd
->inode
= dir
->d_inode
;
3251 save_parent
.mnt
= NULL
;
3252 save_parent
.dentry
= NULL
;
3254 mnt_drop_write(nd
->path
.mnt
);
3261 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3262 struct nameidata
*nd
, int flags
,
3263 const struct open_flags
*op
,
3264 struct file
*file
, int *opened
)
3266 static const struct qstr name
= QSTR_INIT("/", 1);
3267 struct dentry
*dentry
, *child
;
3269 int error
= path_lookupat(dfd
, pathname
,
3270 flags
| LOOKUP_DIRECTORY
, nd
);
3271 if (unlikely(error
))
3273 error
= mnt_want_write(nd
->path
.mnt
);
3274 if (unlikely(error
))
3276 /* we want directory to be writable */
3277 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3280 dentry
= nd
->path
.dentry
;
3281 dir
= dentry
->d_inode
;
3282 if (!dir
->i_op
->tmpfile
) {
3283 error
= -EOPNOTSUPP
;
3286 child
= d_alloc(dentry
, &name
);
3287 if (unlikely(!child
)) {
3291 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3292 nd
->flags
|= op
->intent
;
3293 dput(nd
->path
.dentry
);
3294 nd
->path
.dentry
= child
;
3295 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3298 audit_inode(pathname
, nd
->path
.dentry
, 0);
3299 /* Don't check for other permissions, the inode was just created */
3300 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3303 file
->f_path
.mnt
= nd
->path
.mnt
;
3304 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3307 error
= open_check_o_direct(file
);
3310 } else if (!(op
->open_flag
& O_EXCL
)) {
3311 struct inode
*inode
= file_inode(file
);
3312 spin_lock(&inode
->i_lock
);
3313 inode
->i_state
|= I_LINKABLE
;
3314 spin_unlock(&inode
->i_lock
);
3317 mnt_drop_write(nd
->path
.mnt
);
3319 path_put(&nd
->path
);
3323 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3324 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3331 file
= get_empty_filp();
3335 file
->f_flags
= op
->open_flag
;
3337 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3338 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3342 s
= path_init(dfd
, pathname
, flags
, nd
);
3347 while (!(error
= link_path_walk(s
, nd
)) &&
3348 (error
= do_last(nd
, file
, op
, &opened
, pathname
)) > 0) {
3349 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3350 s
= trailing_symlink(nd
);
3359 if (!(opened
& FILE_OPENED
)) {
3363 if (unlikely(error
)) {
3364 if (error
== -EOPENSTALE
) {
3365 if (flags
& LOOKUP_RCU
)
3370 file
= ERR_PTR(error
);
3375 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3376 const struct open_flags
*op
)
3378 struct nameidata nd
, *saved_nd
= set_nameidata(&nd
);
3379 int flags
= op
->lookup_flags
;
3382 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3383 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3384 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3385 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3386 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3387 restore_nameidata(saved_nd
);
3391 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3392 const char *name
, const struct open_flags
*op
)
3394 struct nameidata nd
, *saved_nd
;
3396 struct filename
*filename
;
3397 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3400 nd
.root
.dentry
= dentry
;
3402 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3403 return ERR_PTR(-ELOOP
);
3405 filename
= getname_kernel(name
);
3406 if (unlikely(IS_ERR(filename
)))
3407 return ERR_CAST(filename
);
3409 saved_nd
= set_nameidata(&nd
);
3410 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3411 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3412 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3413 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3414 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3415 restore_nameidata(saved_nd
);
3420 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3421 struct path
*path
, unsigned int lookup_flags
)
3423 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3424 struct nameidata nd
;
3427 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3430 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3431 * other flags passed in are ignored!
3433 lookup_flags
&= LOOKUP_REVAL
;
3435 error
= filename_parentat(dfd
, name
, lookup_flags
, &nd
);
3437 return ERR_PTR(error
);
3440 * Yucky last component or no last component at all?
3441 * (foo/., foo/.., /////)
3443 if (nd
.last_type
!= LAST_NORM
)
3445 nd
.flags
&= ~LOOKUP_PARENT
;
3446 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3448 /* don't fail immediately if it's r/o, at least try to report other errors */
3449 err2
= mnt_want_write(nd
.path
.mnt
);
3451 * Do the final lookup.
3453 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3454 dentry
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, nd
.flags
);
3459 if (d_is_positive(dentry
))
3463 * Special case - lookup gave negative, but... we had foo/bar/
3464 * From the vfs_mknod() POV we just have a negative dentry -
3465 * all is fine. Let's be bastards - you had / on the end, you've
3466 * been asking for (non-existent) directory. -ENOENT for you.
3468 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3472 if (unlikely(err2
)) {
3480 dentry
= ERR_PTR(error
);
3482 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3484 mnt_drop_write(nd
.path
.mnt
);
3490 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3491 struct path
*path
, unsigned int lookup_flags
)
3493 struct filename
*filename
= getname_kernel(pathname
);
3496 if (IS_ERR(filename
))
3497 return ERR_CAST(filename
);
3498 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3502 EXPORT_SYMBOL(kern_path_create
);
3504 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3507 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3508 mnt_drop_write(path
->mnt
);
3511 EXPORT_SYMBOL(done_path_create
);
3513 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3514 struct path
*path
, unsigned int lookup_flags
)
3516 struct filename
*tmp
= getname(pathname
);
3519 return ERR_CAST(tmp
);
3520 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3524 EXPORT_SYMBOL(user_path_create
);
3526 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3528 int error
= may_create(dir
, dentry
);
3533 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3536 if (!dir
->i_op
->mknod
)
3539 error
= devcgroup_inode_mknod(mode
, dev
);
3543 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3547 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3549 fsnotify_create(dir
, dentry
);
3552 EXPORT_SYMBOL(vfs_mknod
);
3554 static int may_mknod(umode_t mode
)
3556 switch (mode
& S_IFMT
) {
3562 case 0: /* zero mode translates to S_IFREG */
3571 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3574 struct dentry
*dentry
;
3577 unsigned int lookup_flags
= 0;
3579 error
= may_mknod(mode
);
3583 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3585 return PTR_ERR(dentry
);
3587 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3588 mode
&= ~current_umask();
3589 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3592 switch (mode
& S_IFMT
) {
3593 case 0: case S_IFREG
:
3594 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3596 case S_IFCHR
: case S_IFBLK
:
3597 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3598 new_decode_dev(dev
));
3600 case S_IFIFO
: case S_IFSOCK
:
3601 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3605 done_path_create(&path
, dentry
);
3606 if (retry_estale(error
, lookup_flags
)) {
3607 lookup_flags
|= LOOKUP_REVAL
;
3613 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3615 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3618 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3620 int error
= may_create(dir
, dentry
);
3621 unsigned max_links
= dir
->i_sb
->s_max_links
;
3626 if (!dir
->i_op
->mkdir
)
3629 mode
&= (S_IRWXUGO
|S_ISVTX
);
3630 error
= security_inode_mkdir(dir
, dentry
, mode
);
3634 if (max_links
&& dir
->i_nlink
>= max_links
)
3637 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3639 fsnotify_mkdir(dir
, dentry
);
3642 EXPORT_SYMBOL(vfs_mkdir
);
3644 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3646 struct dentry
*dentry
;
3649 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3652 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3654 return PTR_ERR(dentry
);
3656 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3657 mode
&= ~current_umask();
3658 error
= security_path_mkdir(&path
, dentry
, mode
);
3660 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3661 done_path_create(&path
, dentry
);
3662 if (retry_estale(error
, lookup_flags
)) {
3663 lookup_flags
|= LOOKUP_REVAL
;
3669 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3671 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3675 * The dentry_unhash() helper will try to drop the dentry early: we
3676 * should have a usage count of 1 if we're the only user of this
3677 * dentry, and if that is true (possibly after pruning the dcache),
3678 * then we drop the dentry now.
3680 * A low-level filesystem can, if it choses, legally
3683 * if (!d_unhashed(dentry))
3686 * if it cannot handle the case of removing a directory
3687 * that is still in use by something else..
3689 void dentry_unhash(struct dentry
*dentry
)
3691 shrink_dcache_parent(dentry
);
3692 spin_lock(&dentry
->d_lock
);
3693 if (dentry
->d_lockref
.count
== 1)
3695 spin_unlock(&dentry
->d_lock
);
3697 EXPORT_SYMBOL(dentry_unhash
);
3699 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3701 int error
= may_delete(dir
, dentry
, 1);
3706 if (!dir
->i_op
->rmdir
)
3710 mutex_lock(&dentry
->d_inode
->i_mutex
);
3713 if (is_local_mountpoint(dentry
))
3716 error
= security_inode_rmdir(dir
, dentry
);
3720 shrink_dcache_parent(dentry
);
3721 error
= dir
->i_op
->rmdir(dir
, dentry
);
3725 dentry
->d_inode
->i_flags
|= S_DEAD
;
3727 detach_mounts(dentry
);
3730 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3736 EXPORT_SYMBOL(vfs_rmdir
);
3738 static long do_rmdir(int dfd
, const char __user
*pathname
)
3741 struct filename
*name
;
3742 struct dentry
*dentry
;
3746 unsigned int lookup_flags
= 0;
3748 name
= user_path_parent(dfd
, pathname
,
3749 &path
, &last
, &type
, lookup_flags
);
3751 return PTR_ERR(name
);
3765 error
= mnt_want_write(path
.mnt
);
3769 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3770 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3771 error
= PTR_ERR(dentry
);
3774 if (!dentry
->d_inode
) {
3778 error
= security_path_rmdir(&path
, dentry
);
3781 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3785 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3786 mnt_drop_write(path
.mnt
);
3790 if (retry_estale(error
, lookup_flags
)) {
3791 lookup_flags
|= LOOKUP_REVAL
;
3797 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3799 return do_rmdir(AT_FDCWD
, pathname
);
3803 * vfs_unlink - unlink a filesystem object
3804 * @dir: parent directory
3806 * @delegated_inode: returns victim inode, if the inode is delegated.
3808 * The caller must hold dir->i_mutex.
3810 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3811 * return a reference to the inode in delegated_inode. The caller
3812 * should then break the delegation on that inode and retry. Because
3813 * breaking a delegation may take a long time, the caller should drop
3814 * dir->i_mutex before doing so.
3816 * Alternatively, a caller may pass NULL for delegated_inode. This may
3817 * be appropriate for callers that expect the underlying filesystem not
3818 * to be NFS exported.
3820 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3822 struct inode
*target
= dentry
->d_inode
;
3823 int error
= may_delete(dir
, dentry
, 0);
3828 if (!dir
->i_op
->unlink
)
3831 mutex_lock(&target
->i_mutex
);
3832 if (is_local_mountpoint(dentry
))
3835 error
= security_inode_unlink(dir
, dentry
);
3837 error
= try_break_deleg(target
, delegated_inode
);
3840 error
= dir
->i_op
->unlink(dir
, dentry
);
3843 detach_mounts(dentry
);
3848 mutex_unlock(&target
->i_mutex
);
3850 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3851 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3852 fsnotify_link_count(target
);
3858 EXPORT_SYMBOL(vfs_unlink
);
3861 * Make sure that the actual truncation of the file will occur outside its
3862 * directory's i_mutex. Truncate can take a long time if there is a lot of
3863 * writeout happening, and we don't want to prevent access to the directory
3864 * while waiting on the I/O.
3866 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3869 struct filename
*name
;
3870 struct dentry
*dentry
;
3874 struct inode
*inode
= NULL
;
3875 struct inode
*delegated_inode
= NULL
;
3876 unsigned int lookup_flags
= 0;
3878 name
= user_path_parent(dfd
, pathname
,
3879 &path
, &last
, &type
, lookup_flags
);
3881 return PTR_ERR(name
);
3884 if (type
!= LAST_NORM
)
3887 error
= mnt_want_write(path
.mnt
);
3891 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3892 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3893 error
= PTR_ERR(dentry
);
3894 if (!IS_ERR(dentry
)) {
3895 /* Why not before? Because we want correct error value */
3896 if (last
.name
[last
.len
])
3898 inode
= dentry
->d_inode
;
3899 if (d_is_negative(dentry
))
3902 error
= security_path_unlink(&path
, dentry
);
3905 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3909 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3911 iput(inode
); /* truncate the inode here */
3913 if (delegated_inode
) {
3914 error
= break_deleg_wait(&delegated_inode
);
3918 mnt_drop_write(path
.mnt
);
3922 if (retry_estale(error
, lookup_flags
)) {
3923 lookup_flags
|= LOOKUP_REVAL
;
3930 if (d_is_negative(dentry
))
3932 else if (d_is_dir(dentry
))
3939 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3941 if ((flag
& ~AT_REMOVEDIR
) != 0)
3944 if (flag
& AT_REMOVEDIR
)
3945 return do_rmdir(dfd
, pathname
);
3947 return do_unlinkat(dfd
, pathname
);
3950 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3952 return do_unlinkat(AT_FDCWD
, pathname
);
3955 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3957 int error
= may_create(dir
, dentry
);
3962 if (!dir
->i_op
->symlink
)
3965 error
= security_inode_symlink(dir
, dentry
, oldname
);
3969 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3971 fsnotify_create(dir
, dentry
);
3974 EXPORT_SYMBOL(vfs_symlink
);
3976 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3977 int, newdfd
, const char __user
*, newname
)
3980 struct filename
*from
;
3981 struct dentry
*dentry
;
3983 unsigned int lookup_flags
= 0;
3985 from
= getname(oldname
);
3987 return PTR_ERR(from
);
3989 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3990 error
= PTR_ERR(dentry
);
3994 error
= security_path_symlink(&path
, dentry
, from
->name
);
3996 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3997 done_path_create(&path
, dentry
);
3998 if (retry_estale(error
, lookup_flags
)) {
3999 lookup_flags
|= LOOKUP_REVAL
;
4007 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4009 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4013 * vfs_link - create a new link
4014 * @old_dentry: object to be linked
4016 * @new_dentry: where to create the new link
4017 * @delegated_inode: returns inode needing a delegation break
4019 * The caller must hold dir->i_mutex
4021 * If vfs_link discovers a delegation on the to-be-linked file in need
4022 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4023 * inode in delegated_inode. The caller should then break the delegation
4024 * and retry. Because breaking a delegation may take a long time, the
4025 * caller should drop the i_mutex before doing so.
4027 * Alternatively, a caller may pass NULL for delegated_inode. This may
4028 * be appropriate for callers that expect the underlying filesystem not
4029 * to be NFS exported.
4031 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4033 struct inode
*inode
= old_dentry
->d_inode
;
4034 unsigned max_links
= dir
->i_sb
->s_max_links
;
4040 error
= may_create(dir
, new_dentry
);
4044 if (dir
->i_sb
!= inode
->i_sb
)
4048 * A link to an append-only or immutable file cannot be created.
4050 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4052 if (!dir
->i_op
->link
)
4054 if (S_ISDIR(inode
->i_mode
))
4057 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4061 mutex_lock(&inode
->i_mutex
);
4062 /* Make sure we don't allow creating hardlink to an unlinked file */
4063 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4065 else if (max_links
&& inode
->i_nlink
>= max_links
)
4068 error
= try_break_deleg(inode
, delegated_inode
);
4070 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4073 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4074 spin_lock(&inode
->i_lock
);
4075 inode
->i_state
&= ~I_LINKABLE
;
4076 spin_unlock(&inode
->i_lock
);
4078 mutex_unlock(&inode
->i_mutex
);
4080 fsnotify_link(dir
, inode
, new_dentry
);
4083 EXPORT_SYMBOL(vfs_link
);
4086 * Hardlinks are often used in delicate situations. We avoid
4087 * security-related surprises by not following symlinks on the
4090 * We don't follow them on the oldname either to be compatible
4091 * with linux 2.0, and to avoid hard-linking to directories
4092 * and other special files. --ADM
4094 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4095 int, newdfd
, const char __user
*, newname
, int, flags
)
4097 struct dentry
*new_dentry
;
4098 struct path old_path
, new_path
;
4099 struct inode
*delegated_inode
= NULL
;
4103 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4106 * To use null names we require CAP_DAC_READ_SEARCH
4107 * This ensures that not everyone will be able to create
4108 * handlink using the passed filedescriptor.
4110 if (flags
& AT_EMPTY_PATH
) {
4111 if (!capable(CAP_DAC_READ_SEARCH
))
4116 if (flags
& AT_SYMLINK_FOLLOW
)
4117 how
|= LOOKUP_FOLLOW
;
4119 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4123 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4124 (how
& LOOKUP_REVAL
));
4125 error
= PTR_ERR(new_dentry
);
4126 if (IS_ERR(new_dentry
))
4130 if (old_path
.mnt
!= new_path
.mnt
)
4132 error
= may_linkat(&old_path
);
4133 if (unlikely(error
))
4135 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4138 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4140 done_path_create(&new_path
, new_dentry
);
4141 if (delegated_inode
) {
4142 error
= break_deleg_wait(&delegated_inode
);
4144 path_put(&old_path
);
4148 if (retry_estale(error
, how
)) {
4149 path_put(&old_path
);
4150 how
|= LOOKUP_REVAL
;
4154 path_put(&old_path
);
4159 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4161 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4165 * vfs_rename - rename a filesystem object
4166 * @old_dir: parent of source
4167 * @old_dentry: source
4168 * @new_dir: parent of destination
4169 * @new_dentry: destination
4170 * @delegated_inode: returns an inode needing a delegation break
4171 * @flags: rename flags
4173 * The caller must hold multiple mutexes--see lock_rename()).
4175 * If vfs_rename discovers a delegation in need of breaking at either
4176 * the source or destination, it will return -EWOULDBLOCK and return a
4177 * reference to the inode in delegated_inode. The caller should then
4178 * break the delegation and retry. Because breaking a delegation may
4179 * take a long time, the caller should drop all locks before doing
4182 * Alternatively, a caller may pass NULL for delegated_inode. This may
4183 * be appropriate for callers that expect the underlying filesystem not
4184 * to be NFS exported.
4186 * The worst of all namespace operations - renaming directory. "Perverted"
4187 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4189 * a) we can get into loop creation.
4190 * b) race potential - two innocent renames can create a loop together.
4191 * That's where 4.4 screws up. Current fix: serialization on
4192 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4194 * c) we have to lock _four_ objects - parents and victim (if it exists),
4195 * and source (if it is not a directory).
4196 * And that - after we got ->i_mutex on parents (until then we don't know
4197 * whether the target exists). Solution: try to be smart with locking
4198 * order for inodes. We rely on the fact that tree topology may change
4199 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4200 * move will be locked. Thus we can rank directories by the tree
4201 * (ancestors first) and rank all non-directories after them.
4202 * That works since everybody except rename does "lock parent, lookup,
4203 * lock child" and rename is under ->s_vfs_rename_mutex.
4204 * HOWEVER, it relies on the assumption that any object with ->lookup()
4205 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4206 * we'd better make sure that there's no link(2) for them.
4207 * d) conversion from fhandle to dentry may come in the wrong moment - when
4208 * we are removing the target. Solution: we will have to grab ->i_mutex
4209 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4210 * ->i_mutex on parents, which works but leads to some truly excessive
4213 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4214 struct inode
*new_dir
, struct dentry
*new_dentry
,
4215 struct inode
**delegated_inode
, unsigned int flags
)
4218 bool is_dir
= d_is_dir(old_dentry
);
4219 const unsigned char *old_name
;
4220 struct inode
*source
= old_dentry
->d_inode
;
4221 struct inode
*target
= new_dentry
->d_inode
;
4222 bool new_is_dir
= false;
4223 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4225 if (source
== target
)
4228 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4233 error
= may_create(new_dir
, new_dentry
);
4235 new_is_dir
= d_is_dir(new_dentry
);
4237 if (!(flags
& RENAME_EXCHANGE
))
4238 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4240 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4245 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4248 if (flags
&& !old_dir
->i_op
->rename2
)
4252 * If we are going to change the parent - check write permissions,
4253 * we'll need to flip '..'.
4255 if (new_dir
!= old_dir
) {
4257 error
= inode_permission(source
, MAY_WRITE
);
4261 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4262 error
= inode_permission(target
, MAY_WRITE
);
4268 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4273 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4275 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4276 lock_two_nondirectories(source
, target
);
4278 mutex_lock(&target
->i_mutex
);
4281 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4284 if (max_links
&& new_dir
!= old_dir
) {
4286 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4288 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4289 old_dir
->i_nlink
>= max_links
)
4292 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4293 shrink_dcache_parent(new_dentry
);
4295 error
= try_break_deleg(source
, delegated_inode
);
4299 if (target
&& !new_is_dir
) {
4300 error
= try_break_deleg(target
, delegated_inode
);
4304 if (!old_dir
->i_op
->rename2
) {
4305 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4306 new_dir
, new_dentry
);
4308 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4309 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4310 new_dir
, new_dentry
, flags
);
4315 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4317 target
->i_flags
|= S_DEAD
;
4318 dont_mount(new_dentry
);
4319 detach_mounts(new_dentry
);
4321 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4322 if (!(flags
& RENAME_EXCHANGE
))
4323 d_move(old_dentry
, new_dentry
);
4325 d_exchange(old_dentry
, new_dentry
);
4328 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4329 unlock_two_nondirectories(source
, target
);
4331 mutex_unlock(&target
->i_mutex
);
4334 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4335 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4336 if (flags
& RENAME_EXCHANGE
) {
4337 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4338 new_is_dir
, NULL
, new_dentry
);
4341 fsnotify_oldname_free(old_name
);
4345 EXPORT_SYMBOL(vfs_rename
);
4347 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4348 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4350 struct dentry
*old_dentry
, *new_dentry
;
4351 struct dentry
*trap
;
4352 struct path old_path
, new_path
;
4353 struct qstr old_last
, new_last
;
4354 int old_type
, new_type
;
4355 struct inode
*delegated_inode
= NULL
;
4356 struct filename
*from
;
4357 struct filename
*to
;
4358 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4359 bool should_retry
= false;
4362 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4365 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4366 (flags
& RENAME_EXCHANGE
))
4369 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4372 if (flags
& RENAME_EXCHANGE
)
4376 from
= user_path_parent(olddfd
, oldname
,
4377 &old_path
, &old_last
, &old_type
, lookup_flags
);
4379 error
= PTR_ERR(from
);
4383 to
= user_path_parent(newdfd
, newname
,
4384 &new_path
, &new_last
, &new_type
, lookup_flags
);
4386 error
= PTR_ERR(to
);
4391 if (old_path
.mnt
!= new_path
.mnt
)
4395 if (old_type
!= LAST_NORM
)
4398 if (flags
& RENAME_NOREPLACE
)
4400 if (new_type
!= LAST_NORM
)
4403 error
= mnt_want_write(old_path
.mnt
);
4408 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4410 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4411 error
= PTR_ERR(old_dentry
);
4412 if (IS_ERR(old_dentry
))
4414 /* source must exist */
4416 if (d_is_negative(old_dentry
))
4418 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4419 error
= PTR_ERR(new_dentry
);
4420 if (IS_ERR(new_dentry
))
4423 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4425 if (flags
& RENAME_EXCHANGE
) {
4427 if (d_is_negative(new_dentry
))
4430 if (!d_is_dir(new_dentry
)) {
4432 if (new_last
.name
[new_last
.len
])
4436 /* unless the source is a directory trailing slashes give -ENOTDIR */
4437 if (!d_is_dir(old_dentry
)) {
4439 if (old_last
.name
[old_last
.len
])
4441 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4444 /* source should not be ancestor of target */
4446 if (old_dentry
== trap
)
4448 /* target should not be an ancestor of source */
4449 if (!(flags
& RENAME_EXCHANGE
))
4451 if (new_dentry
== trap
)
4454 error
= security_path_rename(&old_path
, old_dentry
,
4455 &new_path
, new_dentry
, flags
);
4458 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4459 new_path
.dentry
->d_inode
, new_dentry
,
4460 &delegated_inode
, flags
);
4466 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4467 if (delegated_inode
) {
4468 error
= break_deleg_wait(&delegated_inode
);
4472 mnt_drop_write(old_path
.mnt
);
4474 if (retry_estale(error
, lookup_flags
))
4475 should_retry
= true;
4476 path_put(&new_path
);
4479 path_put(&old_path
);
4482 should_retry
= false;
4483 lookup_flags
|= LOOKUP_REVAL
;
4490 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4491 int, newdfd
, const char __user
*, newname
)
4493 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4496 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4498 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4501 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4503 int error
= may_create(dir
, dentry
);
4507 if (!dir
->i_op
->mknod
)
4510 return dir
->i_op
->mknod(dir
, dentry
,
4511 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4513 EXPORT_SYMBOL(vfs_whiteout
);
4515 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4517 int len
= PTR_ERR(link
);
4522 if (len
> (unsigned) buflen
)
4524 if (copy_to_user(buffer
, link
, len
))
4529 EXPORT_SYMBOL(readlink_copy
);
4532 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4533 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4534 * using) it for any given inode is up to filesystem.
4536 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4539 struct inode
*inode
= d_inode(dentry
);
4540 const char *link
= inode
->i_link
;
4544 link
= inode
->i_op
->follow_link(dentry
, &cookie
);
4546 return PTR_ERR(link
);
4548 res
= readlink_copy(buffer
, buflen
, link
);
4549 if (inode
->i_op
->put_link
)
4550 inode
->i_op
->put_link(inode
, cookie
);
4553 EXPORT_SYMBOL(generic_readlink
);
4555 /* get the link contents into pagecache */
4556 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4560 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4561 page
= read_mapping_page(mapping
, 0, NULL
);
4566 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4570 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4572 struct page
*page
= NULL
;
4573 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4576 page_cache_release(page
);
4580 EXPORT_SYMBOL(page_readlink
);
4582 const char *page_follow_link_light(struct dentry
*dentry
, void **cookie
)
4584 struct page
*page
= NULL
;
4585 char *res
= page_getlink(dentry
, &page
);
4590 EXPORT_SYMBOL(page_follow_link_light
);
4592 void page_put_link(struct inode
*unused
, void *cookie
)
4594 struct page
*page
= cookie
;
4596 page_cache_release(page
);
4598 EXPORT_SYMBOL(page_put_link
);
4601 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4603 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4605 struct address_space
*mapping
= inode
->i_mapping
;
4610 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4612 flags
|= AOP_FLAG_NOFS
;
4615 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4616 flags
, &page
, &fsdata
);
4620 kaddr
= kmap_atomic(page
);
4621 memcpy(kaddr
, symname
, len
-1);
4622 kunmap_atomic(kaddr
);
4624 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4631 mark_inode_dirty(inode
);
4636 EXPORT_SYMBOL(__page_symlink
);
4638 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4640 return __page_symlink(inode
, symname
, len
,
4641 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4643 EXPORT_SYMBOL(page_symlink
);
4645 const struct inode_operations page_symlink_inode_operations
= {
4646 .readlink
= generic_readlink
,
4647 .follow_link
= page_follow_link_light
,
4648 .put_link
= page_put_link
,
4650 EXPORT_SYMBOL(page_symlink_inode_operations
);