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 (is_uncached_acl(acl
))
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
;
508 struct delayed_call done
;
511 } *stack
, internal
[EMBEDDED_LEVELS
];
512 struct filename
*name
;
513 struct nameidata
*saved
;
514 struct inode
*link_inode
;
519 static void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
521 struct nameidata
*old
= current
->nameidata
;
522 p
->stack
= p
->internal
;
525 p
->total_link_count
= old
? old
->total_link_count
: 0;
527 current
->nameidata
= p
;
530 static void restore_nameidata(void)
532 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
534 current
->nameidata
= old
;
536 old
->total_link_count
= now
->total_link_count
;
537 if (now
->stack
!= now
->internal
)
541 static int __nd_alloc_stack(struct nameidata
*nd
)
545 if (nd
->flags
& LOOKUP_RCU
) {
546 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
551 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
556 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
562 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
563 * @path: nameidate to verify
565 * Rename can sometimes move a file or directory outside of a bind
566 * mount, path_connected allows those cases to be detected.
568 static bool path_connected(const struct path
*path
)
570 struct vfsmount
*mnt
= path
->mnt
;
572 /* Only bind mounts can have disconnected paths */
573 if (mnt
->mnt_root
== mnt
->mnt_sb
->s_root
)
576 return is_subdir(path
->dentry
, mnt
->mnt_root
);
579 static inline int nd_alloc_stack(struct nameidata
*nd
)
581 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
583 if (likely(nd
->stack
!= nd
->internal
))
585 return __nd_alloc_stack(nd
);
588 static void drop_links(struct nameidata
*nd
)
592 struct saved
*last
= nd
->stack
+ i
;
593 do_delayed_call(&last
->done
);
594 clear_delayed_call(&last
->done
);
598 static void terminate_walk(struct nameidata
*nd
)
601 if (!(nd
->flags
& LOOKUP_RCU
)) {
604 for (i
= 0; i
< nd
->depth
; i
++)
605 path_put(&nd
->stack
[i
].link
);
606 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
611 nd
->flags
&= ~LOOKUP_RCU
;
612 if (!(nd
->flags
& LOOKUP_ROOT
))
619 /* path_put is needed afterwards regardless of success or failure */
620 static bool legitimize_path(struct nameidata
*nd
,
621 struct path
*path
, unsigned seq
)
623 int res
= __legitimize_mnt(path
->mnt
, nd
->m_seq
);
630 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
634 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
637 static bool legitimize_links(struct nameidata
*nd
)
640 for (i
= 0; i
< nd
->depth
; i
++) {
641 struct saved
*last
= nd
->stack
+ i
;
642 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
652 * Path walking has 2 modes, rcu-walk and ref-walk (see
653 * Documentation/filesystems/path-lookup.txt). In situations when we can't
654 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
655 * normal reference counts on dentries and vfsmounts to transition to ref-walk
656 * mode. Refcounts are grabbed at the last known good point before rcu-walk
657 * got stuck, so ref-walk may continue from there. If this is not successful
658 * (eg. a seqcount has changed), then failure is returned and it's up to caller
659 * to restart the path walk from the beginning in ref-walk mode.
663 * unlazy_walk - try to switch to ref-walk mode.
664 * @nd: nameidata pathwalk data
665 * @dentry: child of nd->path.dentry or NULL
666 * @seq: seq number to check dentry against
667 * Returns: 0 on success, -ECHILD on failure
669 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
670 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
671 * @nd or NULL. Must be called from rcu-walk context.
672 * Nothing should touch nameidata between unlazy_walk() failure and
675 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
677 struct dentry
*parent
= nd
->path
.dentry
;
679 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
681 nd
->flags
&= ~LOOKUP_RCU
;
682 if (unlikely(!legitimize_links(nd
)))
684 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
686 if (unlikely(!lockref_get_not_dead(&parent
->d_lockref
)))
690 * For a negative lookup, the lookup sequence point is the parents
691 * sequence point, and it only needs to revalidate the parent dentry.
693 * For a positive lookup, we need to move both the parent and the
694 * dentry from the RCU domain to be properly refcounted. And the
695 * sequence number in the dentry validates *both* dentry counters,
696 * since we checked the sequence number of the parent after we got
697 * the child sequence number. So we know the parent must still
698 * be valid if the child sequence number is still valid.
701 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
703 BUG_ON(nd
->inode
!= parent
->d_inode
);
705 if (!lockref_get_not_dead(&dentry
->d_lockref
))
707 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
712 * Sequence counts matched. Now make sure that the root is
713 * still valid and get it if required.
715 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
716 if (unlikely(!legitimize_path(nd
, &nd
->root
, nd
->root_seq
))) {
733 nd
->path
.dentry
= NULL
;
737 if (!(nd
->flags
& LOOKUP_ROOT
))
742 static int unlazy_link(struct nameidata
*nd
, struct path
*link
, unsigned seq
)
744 if (unlikely(!legitimize_path(nd
, link
, seq
))) {
747 nd
->flags
&= ~LOOKUP_RCU
;
749 nd
->path
.dentry
= NULL
;
750 if (!(nd
->flags
& LOOKUP_ROOT
))
753 } else if (likely(unlazy_walk(nd
, NULL
, 0)) == 0) {
760 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
762 return dentry
->d_op
->d_revalidate(dentry
, flags
);
766 * complete_walk - successful completion of path walk
767 * @nd: pointer nameidata
769 * If we had been in RCU mode, drop out of it and legitimize nd->path.
770 * Revalidate the final result, unless we'd already done that during
771 * the path walk or the filesystem doesn't ask for it. Return 0 on
772 * success, -error on failure. In case of failure caller does not
773 * need to drop nd->path.
775 static int complete_walk(struct nameidata
*nd
)
777 struct dentry
*dentry
= nd
->path
.dentry
;
780 if (nd
->flags
& LOOKUP_RCU
) {
781 if (!(nd
->flags
& LOOKUP_ROOT
))
783 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
787 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
790 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
793 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
803 static void set_root(struct nameidata
*nd
)
805 struct fs_struct
*fs
= current
->fs
;
807 if (nd
->flags
& LOOKUP_RCU
) {
811 seq
= read_seqcount_begin(&fs
->seq
);
813 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
814 } while (read_seqcount_retry(&fs
->seq
, seq
));
816 get_fs_root(fs
, &nd
->root
);
820 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
823 if (path
->mnt
!= nd
->path
.mnt
)
827 static inline void path_to_nameidata(const struct path
*path
,
828 struct nameidata
*nd
)
830 if (!(nd
->flags
& LOOKUP_RCU
)) {
831 dput(nd
->path
.dentry
);
832 if (nd
->path
.mnt
!= path
->mnt
)
833 mntput(nd
->path
.mnt
);
835 nd
->path
.mnt
= path
->mnt
;
836 nd
->path
.dentry
= path
->dentry
;
839 static int nd_jump_root(struct nameidata
*nd
)
841 if (nd
->flags
& LOOKUP_RCU
) {
845 nd
->inode
= d
->d_inode
;
846 nd
->seq
= nd
->root_seq
;
847 if (unlikely(read_seqcount_retry(&d
->d_seq
, nd
->seq
)))
853 nd
->inode
= nd
->path
.dentry
->d_inode
;
855 nd
->flags
|= LOOKUP_JUMPED
;
860 * Helper to directly jump to a known parsed path from ->get_link,
861 * caller must have taken a reference to path beforehand.
863 void nd_jump_link(struct path
*path
)
865 struct nameidata
*nd
= current
->nameidata
;
869 nd
->inode
= nd
->path
.dentry
->d_inode
;
870 nd
->flags
|= LOOKUP_JUMPED
;
873 static inline void put_link(struct nameidata
*nd
)
875 struct saved
*last
= nd
->stack
+ --nd
->depth
;
876 do_delayed_call(&last
->done
);
877 if (!(nd
->flags
& LOOKUP_RCU
))
878 path_put(&last
->link
);
881 int sysctl_protected_symlinks __read_mostly
= 0;
882 int sysctl_protected_hardlinks __read_mostly
= 0;
885 * may_follow_link - Check symlink following for unsafe situations
886 * @nd: nameidata pathwalk data
888 * In the case of the sysctl_protected_symlinks sysctl being enabled,
889 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
890 * in a sticky world-writable directory. This is to protect privileged
891 * processes from failing races against path names that may change out
892 * from under them by way of other users creating malicious symlinks.
893 * It will permit symlinks to be followed only when outside a sticky
894 * world-writable directory, or when the uid of the symlink and follower
895 * match, or when the directory owner matches the symlink's owner.
897 * Returns 0 if following the symlink is allowed, -ve on error.
899 static inline int may_follow_link(struct nameidata
*nd
)
901 const struct inode
*inode
;
902 const struct inode
*parent
;
904 if (!sysctl_protected_symlinks
)
907 /* Allowed if owner and follower match. */
908 inode
= nd
->link_inode
;
909 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
912 /* Allowed if parent directory not sticky and world-writable. */
914 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
917 /* Allowed if parent directory and link owner match. */
918 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
921 if (nd
->flags
& LOOKUP_RCU
)
924 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
929 * safe_hardlink_source - Check for safe hardlink conditions
930 * @inode: the source inode to hardlink from
932 * Return false if at least one of the following conditions:
933 * - inode is not a regular file
935 * - inode is setgid and group-exec
936 * - access failure for read and write
938 * Otherwise returns true.
940 static bool safe_hardlink_source(struct inode
*inode
)
942 umode_t mode
= inode
->i_mode
;
944 /* Special files should not get pinned to the filesystem. */
948 /* Setuid files should not get pinned to the filesystem. */
952 /* Executable setgid files should not get pinned to the filesystem. */
953 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
956 /* Hardlinking to unreadable or unwritable sources is dangerous. */
957 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
964 * may_linkat - Check permissions for creating a hardlink
965 * @link: the source to hardlink from
967 * Block hardlink when all of:
968 * - sysctl_protected_hardlinks enabled
969 * - fsuid does not match inode
970 * - hardlink source is unsafe (see safe_hardlink_source() above)
971 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
973 * Returns 0 if successful, -ve on error.
975 static int may_linkat(struct path
*link
)
979 if (!sysctl_protected_hardlinks
)
982 inode
= link
->dentry
->d_inode
;
984 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
985 * otherwise, it must be a safe source.
987 if (inode_owner_or_capable(inode
) || safe_hardlink_source(inode
))
990 audit_log_link_denied("linkat", link
);
994 static __always_inline
995 const char *get_link(struct nameidata
*nd
)
997 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
998 struct dentry
*dentry
= last
->link
.dentry
;
999 struct inode
*inode
= nd
->link_inode
;
1003 if (!(nd
->flags
& LOOKUP_RCU
)) {
1004 touch_atime(&last
->link
);
1006 } else if (atime_needs_update(&last
->link
, inode
)) {
1007 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1008 return ERR_PTR(-ECHILD
);
1009 touch_atime(&last
->link
);
1012 error
= security_inode_follow_link(dentry
, inode
,
1013 nd
->flags
& LOOKUP_RCU
);
1014 if (unlikely(error
))
1015 return ERR_PTR(error
);
1017 nd
->last_type
= LAST_BIND
;
1018 res
= inode
->i_link
;
1020 const char * (*get
)(struct dentry
*, struct inode
*,
1021 struct delayed_call
*);
1022 get
= inode
->i_op
->get_link
;
1023 if (nd
->flags
& LOOKUP_RCU
) {
1024 res
= get(NULL
, inode
, &last
->done
);
1025 if (res
== ERR_PTR(-ECHILD
)) {
1026 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1027 return ERR_PTR(-ECHILD
);
1028 res
= get(dentry
, inode
, &last
->done
);
1031 res
= get(dentry
, inode
, &last
->done
);
1033 if (IS_ERR_OR_NULL(res
))
1039 if (unlikely(nd_jump_root(nd
)))
1040 return ERR_PTR(-ECHILD
);
1041 while (unlikely(*++res
== '/'))
1050 * follow_up - Find the mountpoint of path's vfsmount
1052 * Given a path, find the mountpoint of its source file system.
1053 * Replace @path with the path of the mountpoint in the parent mount.
1056 * Return 1 if we went up a level and 0 if we were already at the
1059 int follow_up(struct path
*path
)
1061 struct mount
*mnt
= real_mount(path
->mnt
);
1062 struct mount
*parent
;
1063 struct dentry
*mountpoint
;
1065 read_seqlock_excl(&mount_lock
);
1066 parent
= mnt
->mnt_parent
;
1067 if (parent
== mnt
) {
1068 read_sequnlock_excl(&mount_lock
);
1071 mntget(&parent
->mnt
);
1072 mountpoint
= dget(mnt
->mnt_mountpoint
);
1073 read_sequnlock_excl(&mount_lock
);
1075 path
->dentry
= mountpoint
;
1077 path
->mnt
= &parent
->mnt
;
1080 EXPORT_SYMBOL(follow_up
);
1083 * Perform an automount
1084 * - return -EISDIR to tell follow_managed() to stop and return the path we
1087 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
1090 struct vfsmount
*mnt
;
1093 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
1096 /* We don't want to mount if someone's just doing a stat -
1097 * unless they're stat'ing a directory and appended a '/' to
1100 * We do, however, want to mount if someone wants to open or
1101 * create a file of any type under the mountpoint, wants to
1102 * traverse through the mountpoint or wants to open the
1103 * mounted directory. Also, autofs may mark negative dentries
1104 * as being automount points. These will need the attentions
1105 * of the daemon to instantiate them before they can be used.
1107 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1108 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1109 path
->dentry
->d_inode
)
1112 nd
->total_link_count
++;
1113 if (nd
->total_link_count
>= 40)
1116 mnt
= path
->dentry
->d_op
->d_automount(path
);
1119 * The filesystem is allowed to return -EISDIR here to indicate
1120 * it doesn't want to automount. For instance, autofs would do
1121 * this so that its userspace daemon can mount on this dentry.
1123 * However, we can only permit this if it's a terminal point in
1124 * the path being looked up; if it wasn't then the remainder of
1125 * the path is inaccessible and we should say so.
1127 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1129 return PTR_ERR(mnt
);
1132 if (!mnt
) /* mount collision */
1135 if (!*need_mntput
) {
1136 /* lock_mount() may release path->mnt on error */
1138 *need_mntput
= true;
1140 err
= finish_automount(mnt
, path
);
1144 /* Someone else made a mount here whilst we were busy */
1149 path
->dentry
= dget(mnt
->mnt_root
);
1158 * Handle a dentry that is managed in some way.
1159 * - Flagged for transit management (autofs)
1160 * - Flagged as mountpoint
1161 * - Flagged as automount point
1163 * This may only be called in refwalk mode.
1165 * Serialization is taken care of in namespace.c
1167 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1169 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1171 bool need_mntput
= false;
1174 /* Given that we're not holding a lock here, we retain the value in a
1175 * local variable for each dentry as we look at it so that we don't see
1176 * the components of that value change under us */
1177 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1178 managed
&= DCACHE_MANAGED_DENTRY
,
1179 unlikely(managed
!= 0)) {
1180 /* Allow the filesystem to manage the transit without i_mutex
1182 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1183 BUG_ON(!path
->dentry
->d_op
);
1184 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1185 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1190 /* Transit to a mounted filesystem. */
1191 if (managed
& DCACHE_MOUNTED
) {
1192 struct vfsmount
*mounted
= lookup_mnt(path
);
1197 path
->mnt
= mounted
;
1198 path
->dentry
= dget(mounted
->mnt_root
);
1203 /* Something is mounted on this dentry in another
1204 * namespace and/or whatever was mounted there in this
1205 * namespace got unmounted before lookup_mnt() could
1209 /* Handle an automount point */
1210 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1211 ret
= follow_automount(path
, nd
, &need_mntput
);
1217 /* We didn't change the current path point */
1221 if (need_mntput
&& path
->mnt
== mnt
)
1223 if (ret
== -EISDIR
|| !ret
)
1226 nd
->flags
|= LOOKUP_JUMPED
;
1227 if (unlikely(ret
< 0))
1228 path_put_conditional(path
, nd
);
1232 int follow_down_one(struct path
*path
)
1234 struct vfsmount
*mounted
;
1236 mounted
= lookup_mnt(path
);
1240 path
->mnt
= mounted
;
1241 path
->dentry
= dget(mounted
->mnt_root
);
1246 EXPORT_SYMBOL(follow_down_one
);
1248 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1250 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1251 dentry
->d_op
->d_manage(dentry
, true) : 0;
1255 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1256 * we meet a managed dentry that would need blocking.
1258 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1259 struct inode
**inode
, unsigned *seqp
)
1262 struct mount
*mounted
;
1264 * Don't forget we might have a non-mountpoint managed dentry
1265 * that wants to block transit.
1267 switch (managed_dentry_rcu(path
->dentry
)) {
1277 if (!d_mountpoint(path
->dentry
))
1278 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1280 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1283 path
->mnt
= &mounted
->mnt
;
1284 path
->dentry
= mounted
->mnt
.mnt_root
;
1285 nd
->flags
|= LOOKUP_JUMPED
;
1286 *seqp
= read_seqcount_begin(&path
->dentry
->d_seq
);
1288 * Update the inode too. We don't need to re-check the
1289 * dentry sequence number here after this d_inode read,
1290 * because a mount-point is always pinned.
1292 *inode
= path
->dentry
->d_inode
;
1294 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1295 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1298 static int follow_dotdot_rcu(struct nameidata
*nd
)
1300 struct inode
*inode
= nd
->inode
;
1303 if (path_equal(&nd
->path
, &nd
->root
))
1305 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1306 struct dentry
*old
= nd
->path
.dentry
;
1307 struct dentry
*parent
= old
->d_parent
;
1310 inode
= parent
->d_inode
;
1311 seq
= read_seqcount_begin(&parent
->d_seq
);
1312 if (unlikely(read_seqcount_retry(&old
->d_seq
, nd
->seq
)))
1314 nd
->path
.dentry
= parent
;
1316 if (unlikely(!path_connected(&nd
->path
)))
1320 struct mount
*mnt
= real_mount(nd
->path
.mnt
);
1321 struct mount
*mparent
= mnt
->mnt_parent
;
1322 struct dentry
*mountpoint
= mnt
->mnt_mountpoint
;
1323 struct inode
*inode2
= mountpoint
->d_inode
;
1324 unsigned seq
= read_seqcount_begin(&mountpoint
->d_seq
);
1325 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1327 if (&mparent
->mnt
== nd
->path
.mnt
)
1329 /* we know that mountpoint was pinned */
1330 nd
->path
.dentry
= mountpoint
;
1331 nd
->path
.mnt
= &mparent
->mnt
;
1336 while (unlikely(d_mountpoint(nd
->path
.dentry
))) {
1337 struct mount
*mounted
;
1338 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1339 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1343 nd
->path
.mnt
= &mounted
->mnt
;
1344 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1345 inode
= nd
->path
.dentry
->d_inode
;
1346 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1353 * Follow down to the covering mount currently visible to userspace. At each
1354 * point, the filesystem owning that dentry may be queried as to whether the
1355 * caller is permitted to proceed or not.
1357 int follow_down(struct path
*path
)
1362 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1363 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1364 /* Allow the filesystem to manage the transit without i_mutex
1367 * We indicate to the filesystem if someone is trying to mount
1368 * something here. This gives autofs the chance to deny anyone
1369 * other than its daemon the right to mount on its
1372 * The filesystem may sleep at this point.
1374 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1375 BUG_ON(!path
->dentry
->d_op
);
1376 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1377 ret
= path
->dentry
->d_op
->d_manage(
1378 path
->dentry
, false);
1380 return ret
== -EISDIR
? 0 : ret
;
1383 /* Transit to a mounted filesystem. */
1384 if (managed
& DCACHE_MOUNTED
) {
1385 struct vfsmount
*mounted
= lookup_mnt(path
);
1390 path
->mnt
= mounted
;
1391 path
->dentry
= dget(mounted
->mnt_root
);
1395 /* Don't handle automount points here */
1400 EXPORT_SYMBOL(follow_down
);
1403 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1405 static void follow_mount(struct path
*path
)
1407 while (d_mountpoint(path
->dentry
)) {
1408 struct vfsmount
*mounted
= lookup_mnt(path
);
1413 path
->mnt
= mounted
;
1414 path
->dentry
= dget(mounted
->mnt_root
);
1418 static int follow_dotdot(struct nameidata
*nd
)
1421 struct dentry
*old
= nd
->path
.dentry
;
1423 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1424 nd
->path
.mnt
== nd
->root
.mnt
) {
1427 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1428 /* rare case of legitimate dget_parent()... */
1429 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1431 if (unlikely(!path_connected(&nd
->path
)))
1435 if (!follow_up(&nd
->path
))
1438 follow_mount(&nd
->path
);
1439 nd
->inode
= nd
->path
.dentry
->d_inode
;
1444 * This looks up the name in dcache, possibly revalidates the old dentry and
1445 * allocates a new one if not found or not valid. In the need_lookup argument
1446 * returns whether i_op->lookup is necessary.
1448 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1452 struct dentry
*dentry
;
1455 dentry
= d_lookup(dir
, name
);
1457 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1458 error
= d_revalidate(dentry
, flags
);
1459 if (unlikely(error
<= 0)) {
1461 d_invalidate(dentry
);
1463 return ERR_PTR(error
);
1471 * Call i_op->lookup on the dentry. The dentry must be negative and
1474 * dir->d_inode->i_mutex must be held
1476 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1481 /* Don't create child dentry for a dead directory. */
1482 if (unlikely(IS_DEADDIR(dir
))) {
1484 return ERR_PTR(-ENOENT
);
1487 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1488 if (unlikely(old
)) {
1495 static struct dentry
*__lookup_hash(const struct qstr
*name
,
1496 struct dentry
*base
, unsigned int flags
)
1498 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1503 dentry
= d_alloc(base
, name
);
1504 if (unlikely(!dentry
))
1505 return ERR_PTR(-ENOMEM
);
1507 return lookup_real(base
->d_inode
, dentry
, flags
);
1510 static int lookup_fast(struct nameidata
*nd
,
1511 struct path
*path
, struct inode
**inode
,
1514 struct vfsmount
*mnt
= nd
->path
.mnt
;
1515 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1520 * Rename seqlock is not required here because in the off chance
1521 * of a false negative due to a concurrent rename, the caller is
1522 * going to fall back to non-racy lookup.
1524 if (nd
->flags
& LOOKUP_RCU
) {
1527 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1528 if (unlikely(!dentry
)) {
1529 if (unlazy_walk(nd
, NULL
, 0))
1535 * This sequence count validates that the inode matches
1536 * the dentry name information from lookup.
1538 *inode
= d_backing_inode(dentry
);
1539 negative
= d_is_negative(dentry
);
1540 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
1544 * This sequence count validates that the parent had no
1545 * changes while we did the lookup of the dentry above.
1547 * The memory barrier in read_seqcount_begin of child is
1548 * enough, we can use __read_seqcount_retry here.
1550 if (unlikely(__read_seqcount_retry(&parent
->d_seq
, nd
->seq
)))
1554 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1555 status
= d_revalidate(dentry
, nd
->flags
);
1556 if (unlikely(status
<= 0)) {
1557 if (unlazy_walk(nd
, dentry
, seq
))
1559 if (status
== -ECHILD
)
1560 status
= d_revalidate(dentry
, nd
->flags
);
1563 * Note: do negative dentry check after revalidation in
1564 * case that drops it.
1566 if (unlikely(negative
))
1569 path
->dentry
= dentry
;
1570 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1572 if (unlazy_walk(nd
, dentry
, seq
))
1576 dentry
= __d_lookup(parent
, &nd
->last
);
1577 if (unlikely(!dentry
))
1579 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1580 status
= d_revalidate(dentry
, nd
->flags
);
1582 if (unlikely(status
<= 0)) {
1584 d_invalidate(dentry
);
1588 if (unlikely(d_is_negative(dentry
))) {
1594 path
->dentry
= dentry
;
1595 err
= follow_managed(path
, nd
);
1596 if (likely(err
> 0))
1597 *inode
= d_backing_inode(path
->dentry
);
1601 /* Fast lookup failed, do it the slow way */
1602 static struct dentry
*lookup_slow(const struct qstr
*name
,
1606 struct dentry
*dentry
= ERR_PTR(-ENOENT
), *old
;
1607 struct inode
*inode
= dir
->d_inode
;
1608 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1610 inode_lock_shared(inode
);
1611 /* Don't go there if it's already dead */
1612 if (unlikely(IS_DEADDIR(inode
)))
1615 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1618 if (unlikely(!d_in_lookup(dentry
))) {
1619 if ((dentry
->d_flags
& DCACHE_OP_REVALIDATE
) &&
1620 !(flags
& LOOKUP_NO_REVAL
)) {
1621 int error
= d_revalidate(dentry
, flags
);
1622 if (unlikely(error
<= 0)) {
1624 d_invalidate(dentry
);
1629 dentry
= ERR_PTR(error
);
1633 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1634 d_lookup_done(dentry
);
1635 if (unlikely(old
)) {
1641 inode_unlock_shared(inode
);
1645 static inline int may_lookup(struct nameidata
*nd
)
1647 if (nd
->flags
& LOOKUP_RCU
) {
1648 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1651 if (unlazy_walk(nd
, NULL
, 0))
1654 return inode_permission(nd
->inode
, MAY_EXEC
);
1657 static inline int handle_dots(struct nameidata
*nd
, int type
)
1659 if (type
== LAST_DOTDOT
) {
1662 if (nd
->flags
& LOOKUP_RCU
) {
1663 return follow_dotdot_rcu(nd
);
1665 return follow_dotdot(nd
);
1670 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1671 struct inode
*inode
, unsigned seq
)
1675 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1676 path_to_nameidata(link
, nd
);
1679 if (!(nd
->flags
& LOOKUP_RCU
)) {
1680 if (link
->mnt
== nd
->path
.mnt
)
1683 error
= nd_alloc_stack(nd
);
1684 if (unlikely(error
)) {
1685 if (error
== -ECHILD
) {
1686 if (unlikely(unlazy_link(nd
, link
, seq
)))
1688 error
= nd_alloc_stack(nd
);
1696 last
= nd
->stack
+ nd
->depth
++;
1698 clear_delayed_call(&last
->done
);
1699 nd
->link_inode
= inode
;
1705 * Do we need to follow links? We _really_ want to be able
1706 * to do this check without having to look at inode->i_op,
1707 * so we keep a cache of "no, this doesn't need follow_link"
1708 * for the common case.
1710 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
,
1712 struct inode
*inode
, unsigned seq
)
1714 if (likely(!d_is_symlink(link
->dentry
)))
1718 /* make sure that d_is_symlink above matches inode */
1719 if (nd
->flags
& LOOKUP_RCU
) {
1720 if (read_seqcount_retry(&link
->dentry
->d_seq
, seq
))
1723 return pick_link(nd
, link
, inode
, seq
);
1726 enum {WALK_GET
= 1, WALK_PUT
= 2};
1728 static int walk_component(struct nameidata
*nd
, int flags
)
1731 struct inode
*inode
;
1735 * "." and ".." are special - ".." especially so because it has
1736 * to be able to know about the current root directory and
1737 * parent relationships.
1739 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1740 err
= handle_dots(nd
, nd
->last_type
);
1741 if (flags
& WALK_PUT
)
1745 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1746 if (unlikely(err
<= 0)) {
1749 path
.dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
,
1751 if (IS_ERR(path
.dentry
))
1752 return PTR_ERR(path
.dentry
);
1754 path
.mnt
= nd
->path
.mnt
;
1755 err
= follow_managed(&path
, nd
);
1756 if (unlikely(err
< 0))
1759 if (unlikely(d_is_negative(path
.dentry
))) {
1760 path_to_nameidata(&path
, nd
);
1764 seq
= 0; /* we are already out of RCU mode */
1765 inode
= d_backing_inode(path
.dentry
);
1768 if (flags
& WALK_PUT
)
1770 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
, inode
, seq
);
1773 path_to_nameidata(&path
, nd
);
1780 * We can do the critical dentry name comparison and hashing
1781 * operations one word at a time, but we are limited to:
1783 * - Architectures with fast unaligned word accesses. We could
1784 * do a "get_unaligned()" if this helps and is sufficiently
1787 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1788 * do not trap on the (extremely unlikely) case of a page
1789 * crossing operation.
1791 * - Furthermore, we need an efficient 64-bit compile for the
1792 * 64-bit case in order to generate the "number of bytes in
1793 * the final mask". Again, that could be replaced with a
1794 * efficient population count instruction or similar.
1796 #ifdef CONFIG_DCACHE_WORD_ACCESS
1798 #include <asm/word-at-a-time.h>
1802 static inline unsigned int fold_hash(unsigned long hash
)
1804 return hash_64(hash
, 32);
1808 * This is George Marsaglia's XORSHIFT generator.
1809 * It implements a maximum-period LFSR in only a few
1810 * instructions. It also has the property (required
1811 * by hash_name()) that mix_hash(0) = 0.
1813 static inline unsigned long mix_hash(unsigned long hash
)
1821 #else /* 32-bit case */
1823 #define fold_hash(x) (x)
1825 static inline unsigned long mix_hash(unsigned long hash
)
1835 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1837 unsigned long a
, hash
= 0;
1840 a
= load_unaligned_zeropad(name
);
1841 if (len
< sizeof(unsigned long))
1843 hash
= mix_hash(hash
+ a
);
1844 name
+= sizeof(unsigned long);
1845 len
-= sizeof(unsigned long);
1849 hash
+= a
& bytemask_from_count(len
);
1851 return fold_hash(hash
);
1853 EXPORT_SYMBOL(full_name_hash
);
1856 * Calculate the length and hash of the path component, and
1857 * return the "hash_len" as the result.
1859 static inline u64
hash_name(const char *name
)
1861 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1862 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1865 len
= -sizeof(unsigned long);
1867 hash
= mix_hash(hash
+ a
);
1868 len
+= sizeof(unsigned long);
1869 a
= load_unaligned_zeropad(name
+len
);
1870 b
= a
^ REPEAT_BYTE('/');
1871 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1873 adata
= prep_zero_mask(a
, adata
, &constants
);
1874 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1876 mask
= create_zero_mask(adata
| bdata
);
1878 hash
+= a
& zero_bytemask(mask
);
1879 len
+= find_zero(mask
);
1880 return hashlen_create(fold_hash(hash
), len
);
1885 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1887 unsigned long hash
= init_name_hash();
1889 hash
= partial_name_hash(*name
++, hash
);
1890 return end_name_hash(hash
);
1892 EXPORT_SYMBOL(full_name_hash
);
1895 * We know there's a real path component here of at least
1898 static inline u64
hash_name(const char *name
)
1900 unsigned long hash
= init_name_hash();
1901 unsigned long len
= 0, c
;
1903 c
= (unsigned char)*name
;
1906 hash
= partial_name_hash(c
, hash
);
1907 c
= (unsigned char)name
[len
];
1908 } while (c
&& c
!= '/');
1909 return hashlen_create(end_name_hash(hash
), len
);
1916 * This is the basic name resolution function, turning a pathname into
1917 * the final dentry. We expect 'base' to be positive and a directory.
1919 * Returns 0 and nd will have valid dentry and mnt on success.
1920 * Returns error and drops reference to input namei data on failure.
1922 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1931 /* At this point we know we have a real path component. */
1936 err
= may_lookup(nd
);
1940 hash_len
= hash_name(name
);
1943 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1945 if (name
[1] == '.') {
1947 nd
->flags
|= LOOKUP_JUMPED
;
1953 if (likely(type
== LAST_NORM
)) {
1954 struct dentry
*parent
= nd
->path
.dentry
;
1955 nd
->flags
&= ~LOOKUP_JUMPED
;
1956 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1957 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1958 err
= parent
->d_op
->d_hash(parent
, &this);
1961 hash_len
= this.hash_len
;
1966 nd
->last
.hash_len
= hash_len
;
1967 nd
->last
.name
= name
;
1968 nd
->last_type
= type
;
1970 name
+= hashlen_len(hash_len
);
1974 * If it wasn't NUL, we know it was '/'. Skip that
1975 * slash, and continue until no more slashes.
1979 } while (unlikely(*name
== '/'));
1980 if (unlikely(!*name
)) {
1982 /* pathname body, done */
1985 name
= nd
->stack
[nd
->depth
- 1].name
;
1986 /* trailing symlink, done */
1989 /* last component of nested symlink */
1990 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
1992 err
= walk_component(nd
, WALK_GET
);
1998 const char *s
= get_link(nd
);
2007 nd
->stack
[nd
->depth
- 1].name
= name
;
2012 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2013 if (nd
->flags
& LOOKUP_RCU
) {
2014 if (unlazy_walk(nd
, NULL
, 0))
2022 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2025 const char *s
= nd
->name
->name
;
2027 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
2028 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
2030 if (flags
& LOOKUP_ROOT
) {
2031 struct dentry
*root
= nd
->root
.dentry
;
2032 struct inode
*inode
= root
->d_inode
;
2034 if (!d_can_lookup(root
))
2035 return ERR_PTR(-ENOTDIR
);
2036 retval
= inode_permission(inode
, MAY_EXEC
);
2038 return ERR_PTR(retval
);
2040 nd
->path
= nd
->root
;
2042 if (flags
& LOOKUP_RCU
) {
2044 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2045 nd
->root_seq
= nd
->seq
;
2046 nd
->m_seq
= read_seqbegin(&mount_lock
);
2048 path_get(&nd
->path
);
2053 nd
->root
.mnt
= NULL
;
2054 nd
->path
.mnt
= NULL
;
2055 nd
->path
.dentry
= NULL
;
2057 nd
->m_seq
= read_seqbegin(&mount_lock
);
2059 if (flags
& LOOKUP_RCU
)
2062 if (likely(!nd_jump_root(nd
)))
2064 nd
->root
.mnt
= NULL
;
2066 return ERR_PTR(-ECHILD
);
2067 } else if (nd
->dfd
== AT_FDCWD
) {
2068 if (flags
& LOOKUP_RCU
) {
2069 struct fs_struct
*fs
= current
->fs
;
2075 seq
= read_seqcount_begin(&fs
->seq
);
2077 nd
->inode
= nd
->path
.dentry
->d_inode
;
2078 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2079 } while (read_seqcount_retry(&fs
->seq
, seq
));
2081 get_fs_pwd(current
->fs
, &nd
->path
);
2082 nd
->inode
= nd
->path
.dentry
->d_inode
;
2086 /* Caller must check execute permissions on the starting path component */
2087 struct fd f
= fdget_raw(nd
->dfd
);
2088 struct dentry
*dentry
;
2091 return ERR_PTR(-EBADF
);
2093 dentry
= f
.file
->f_path
.dentry
;
2096 if (!d_can_lookup(dentry
)) {
2098 return ERR_PTR(-ENOTDIR
);
2102 nd
->path
= f
.file
->f_path
;
2103 if (flags
& LOOKUP_RCU
) {
2105 nd
->inode
= nd
->path
.dentry
->d_inode
;
2106 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2108 path_get(&nd
->path
);
2109 nd
->inode
= nd
->path
.dentry
->d_inode
;
2116 static const char *trailing_symlink(struct nameidata
*nd
)
2119 int error
= may_follow_link(nd
);
2120 if (unlikely(error
))
2121 return ERR_PTR(error
);
2122 nd
->flags
|= LOOKUP_PARENT
;
2123 nd
->stack
[0].name
= NULL
;
2128 static inline int lookup_last(struct nameidata
*nd
)
2130 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2131 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2133 nd
->flags
&= ~LOOKUP_PARENT
;
2134 return walk_component(nd
,
2135 nd
->flags
& LOOKUP_FOLLOW
2137 ? WALK_PUT
| WALK_GET
2142 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2143 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2145 const char *s
= path_init(nd
, flags
);
2150 while (!(err
= link_path_walk(s
, nd
))
2151 && ((err
= lookup_last(nd
)) > 0)) {
2152 s
= trailing_symlink(nd
);
2159 err
= complete_walk(nd
);
2161 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2162 if (!d_can_lookup(nd
->path
.dentry
))
2166 nd
->path
.mnt
= NULL
;
2167 nd
->path
.dentry
= NULL
;
2173 static int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2174 struct path
*path
, struct path
*root
)
2177 struct nameidata nd
;
2179 return PTR_ERR(name
);
2180 if (unlikely(root
)) {
2182 flags
|= LOOKUP_ROOT
;
2184 set_nameidata(&nd
, dfd
, name
);
2185 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2186 if (unlikely(retval
== -ECHILD
))
2187 retval
= path_lookupat(&nd
, flags
, path
);
2188 if (unlikely(retval
== -ESTALE
))
2189 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2191 if (likely(!retval
))
2192 audit_inode(name
, path
->dentry
, flags
& LOOKUP_PARENT
);
2193 restore_nameidata();
2198 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2199 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2200 struct path
*parent
)
2202 const char *s
= path_init(nd
, flags
);
2206 err
= link_path_walk(s
, nd
);
2208 err
= complete_walk(nd
);
2211 nd
->path
.mnt
= NULL
;
2212 nd
->path
.dentry
= NULL
;
2218 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2219 unsigned int flags
, struct path
*parent
,
2220 struct qstr
*last
, int *type
)
2223 struct nameidata nd
;
2227 set_nameidata(&nd
, dfd
, name
);
2228 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2229 if (unlikely(retval
== -ECHILD
))
2230 retval
= path_parentat(&nd
, flags
, parent
);
2231 if (unlikely(retval
== -ESTALE
))
2232 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2233 if (likely(!retval
)) {
2235 *type
= nd
.last_type
;
2236 audit_inode(name
, parent
->dentry
, LOOKUP_PARENT
);
2239 name
= ERR_PTR(retval
);
2241 restore_nameidata();
2245 /* does lookup, returns the object with parent locked */
2246 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2248 struct filename
*filename
;
2253 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2255 if (IS_ERR(filename
))
2256 return ERR_CAST(filename
);
2257 if (unlikely(type
!= LAST_NORM
)) {
2260 return ERR_PTR(-EINVAL
);
2262 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2263 d
= __lookup_hash(&last
, path
->dentry
, 0);
2265 inode_unlock(path
->dentry
->d_inode
);
2272 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2274 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2277 EXPORT_SYMBOL(kern_path
);
2280 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2281 * @dentry: pointer to dentry of the base directory
2282 * @mnt: pointer to vfs mount of the base directory
2283 * @name: pointer to file name
2284 * @flags: lookup flags
2285 * @path: pointer to struct path to fill
2287 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2288 const char *name
, unsigned int flags
,
2291 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2292 /* the first argument of filename_lookup() is ignored with root */
2293 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2294 flags
, path
, &root
);
2296 EXPORT_SYMBOL(vfs_path_lookup
);
2299 * lookup_hash - lookup single pathname component on already hashed name
2300 * @name: name and hash to lookup
2301 * @base: base directory to lookup from
2303 * The name must have been verified and hashed (see lookup_one_len()). Using
2304 * this after just full_name_hash() is unsafe.
2306 * This function also doesn't check for search permission on base directory.
2308 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2311 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2313 struct dentry
*lookup_hash(const struct qstr
*name
, struct dentry
*base
)
2317 ret
= lookup_dcache(name
, base
, 0);
2319 ret
= lookup_slow(name
, base
, 0);
2323 EXPORT_SYMBOL(lookup_hash
);
2326 * lookup_one_len - filesystem helper to lookup single pathname component
2327 * @name: pathname component to lookup
2328 * @base: base directory to lookup from
2329 * @len: maximum length @len should be interpreted to
2331 * Note that this routine is purely a helper for filesystem usage and should
2332 * not be called by generic code.
2334 * The caller must hold base->i_mutex.
2336 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2342 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2346 this.hash
= full_name_hash(name
, len
);
2348 return ERR_PTR(-EACCES
);
2350 if (unlikely(name
[0] == '.')) {
2351 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2352 return ERR_PTR(-EACCES
);
2356 c
= *(const unsigned char *)name
++;
2357 if (c
== '/' || c
== '\0')
2358 return ERR_PTR(-EACCES
);
2361 * See if the low-level filesystem might want
2362 * to use its own hash..
2364 if (base
->d_flags
& DCACHE_OP_HASH
) {
2365 int err
= base
->d_op
->d_hash(base
, &this);
2367 return ERR_PTR(err
);
2370 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2372 return ERR_PTR(err
);
2374 return __lookup_hash(&this, base
, 0);
2376 EXPORT_SYMBOL(lookup_one_len
);
2379 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2380 * @name: pathname component to lookup
2381 * @base: base directory to lookup from
2382 * @len: maximum length @len should be interpreted to
2384 * Note that this routine is purely a helper for filesystem usage and should
2385 * not be called by generic code.
2387 * Unlike lookup_one_len, it should be called without the parent
2388 * i_mutex held, and will take the i_mutex itself if necessary.
2390 struct dentry
*lookup_one_len_unlocked(const char *name
,
2391 struct dentry
*base
, int len
)
2399 this.hash
= full_name_hash(name
, len
);
2401 return ERR_PTR(-EACCES
);
2403 if (unlikely(name
[0] == '.')) {
2404 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2405 return ERR_PTR(-EACCES
);
2409 c
= *(const unsigned char *)name
++;
2410 if (c
== '/' || c
== '\0')
2411 return ERR_PTR(-EACCES
);
2414 * See if the low-level filesystem might want
2415 * to use its own hash..
2417 if (base
->d_flags
& DCACHE_OP_HASH
) {
2418 int err
= base
->d_op
->d_hash(base
, &this);
2420 return ERR_PTR(err
);
2423 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2425 return ERR_PTR(err
);
2427 return lookup_hash(&this, base
);
2429 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2431 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2432 struct path
*path
, int *empty
)
2434 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2437 EXPORT_SYMBOL(user_path_at_empty
);
2440 * NB: most callers don't do anything directly with the reference to the
2441 * to struct filename, but the nd->last pointer points into the name string
2442 * allocated by getname. So we must hold the reference to it until all
2443 * path-walking is complete.
2445 static inline struct filename
*
2446 user_path_parent(int dfd
, const char __user
*path
,
2447 struct path
*parent
,
2452 /* only LOOKUP_REVAL is allowed in extra flags */
2453 return filename_parentat(dfd
, getname(path
), flags
& LOOKUP_REVAL
,
2454 parent
, last
, type
);
2458 * mountpoint_last - look up last component for umount
2459 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2460 * @path: pointer to container for result
2462 * This is a special lookup_last function just for umount. In this case, we
2463 * need to resolve the path without doing any revalidation.
2465 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2466 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2467 * in almost all cases, this lookup will be served out of the dcache. The only
2468 * cases where it won't are if nd->last refers to a symlink or the path is
2469 * bogus and it doesn't exist.
2472 * -error: if there was an error during lookup. This includes -ENOENT if the
2473 * lookup found a negative dentry. The nd->path reference will also be
2476 * 0: if we successfully resolved nd->path and found it to not to be a
2477 * symlink that needs to be followed. "path" will also be populated.
2478 * The nd->path reference will also be put.
2480 * 1: if we successfully resolved nd->last and found it to be a symlink
2481 * that needs to be followed. "path" will be populated with the path
2482 * to the link, and nd->path will *not* be put.
2485 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2488 struct dentry
*dentry
;
2489 struct dentry
*dir
= nd
->path
.dentry
;
2491 /* If we're in rcuwalk, drop out of it to handle last component */
2492 if (nd
->flags
& LOOKUP_RCU
) {
2493 if (unlazy_walk(nd
, NULL
, 0))
2497 nd
->flags
&= ~LOOKUP_PARENT
;
2499 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2500 error
= handle_dots(nd
, nd
->last_type
);
2503 dentry
= dget(nd
->path
.dentry
);
2505 dentry
= d_lookup(dir
, &nd
->last
);
2508 * No cached dentry. Mounted dentries are pinned in the
2509 * cache, so that means that this dentry is probably
2510 * a symlink or the path doesn't actually point
2511 * to a mounted dentry.
2513 dentry
= lookup_slow(&nd
->last
, dir
,
2514 nd
->flags
| LOOKUP_NO_REVAL
);
2516 return PTR_ERR(dentry
);
2519 if (d_is_negative(dentry
)) {
2525 path
->dentry
= dentry
;
2526 path
->mnt
= nd
->path
.mnt
;
2527 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
,
2528 d_backing_inode(dentry
), 0);
2529 if (unlikely(error
))
2537 * path_mountpoint - look up a path to be umounted
2538 * @nd: lookup context
2539 * @flags: lookup flags
2540 * @path: pointer to container for result
2542 * Look up the given name, but don't attempt to revalidate the last component.
2543 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2546 path_mountpoint(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2548 const char *s
= path_init(nd
, flags
);
2552 while (!(err
= link_path_walk(s
, nd
)) &&
2553 (err
= mountpoint_last(nd
, path
)) > 0) {
2554 s
= trailing_symlink(nd
);
2565 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2568 struct nameidata nd
;
2571 return PTR_ERR(name
);
2572 set_nameidata(&nd
, dfd
, name
);
2573 error
= path_mountpoint(&nd
, flags
| LOOKUP_RCU
, path
);
2574 if (unlikely(error
== -ECHILD
))
2575 error
= path_mountpoint(&nd
, flags
, path
);
2576 if (unlikely(error
== -ESTALE
))
2577 error
= path_mountpoint(&nd
, flags
| LOOKUP_REVAL
, path
);
2579 audit_inode(name
, path
->dentry
, 0);
2580 restore_nameidata();
2586 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2587 * @dfd: directory file descriptor
2588 * @name: pathname from userland
2589 * @flags: lookup flags
2590 * @path: pointer to container to hold result
2592 * A umount is a special case for path walking. We're not actually interested
2593 * in the inode in this situation, and ESTALE errors can be a problem. We
2594 * simply want track down the dentry and vfsmount attached at the mountpoint
2595 * and avoid revalidating the last component.
2597 * Returns 0 and populates "path" on success.
2600 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2603 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2607 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2610 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2612 EXPORT_SYMBOL(kern_path_mountpoint
);
2614 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2616 kuid_t fsuid
= current_fsuid();
2618 if (uid_eq(inode
->i_uid
, fsuid
))
2620 if (uid_eq(dir
->i_uid
, fsuid
))
2622 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2624 EXPORT_SYMBOL(__check_sticky
);
2627 * Check whether we can remove a link victim from directory dir, check
2628 * whether the type of victim is right.
2629 * 1. We can't do it if dir is read-only (done in permission())
2630 * 2. We should have write and exec permissions on dir
2631 * 3. We can't remove anything from append-only dir
2632 * 4. We can't do anything with immutable dir (done in permission())
2633 * 5. If the sticky bit on dir is set we should either
2634 * a. be owner of dir, or
2635 * b. be owner of victim, or
2636 * c. have CAP_FOWNER capability
2637 * 6. If the victim is append-only or immutable we can't do antyhing with
2638 * links pointing to it.
2639 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2640 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2641 * 9. We can't remove a root or mountpoint.
2642 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2643 * nfs_async_unlink().
2645 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2647 struct inode
*inode
= d_backing_inode(victim
);
2650 if (d_is_negative(victim
))
2654 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2655 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2657 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2663 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2664 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2667 if (!d_is_dir(victim
))
2669 if (IS_ROOT(victim
))
2671 } else if (d_is_dir(victim
))
2673 if (IS_DEADDIR(dir
))
2675 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2680 /* Check whether we can create an object with dentry child in directory
2682 * 1. We can't do it if child already exists (open has special treatment for
2683 * this case, but since we are inlined it's OK)
2684 * 2. We can't do it if dir is read-only (done in permission())
2685 * 3. We should have write and exec permissions on dir
2686 * 4. We can't do it if dir is immutable (done in permission())
2688 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2690 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2693 if (IS_DEADDIR(dir
))
2695 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2699 * p1 and p2 should be directories on the same fs.
2701 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2706 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2710 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
2712 p
= d_ancestor(p2
, p1
);
2714 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2715 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2719 p
= d_ancestor(p1
, p2
);
2721 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2722 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2726 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2727 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2730 EXPORT_SYMBOL(lock_rename
);
2732 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2734 inode_unlock(p1
->d_inode
);
2736 inode_unlock(p2
->d_inode
);
2737 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
2740 EXPORT_SYMBOL(unlock_rename
);
2742 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2745 int error
= may_create(dir
, dentry
);
2749 if (!dir
->i_op
->create
)
2750 return -EACCES
; /* shouldn't it be ENOSYS? */
2753 error
= security_inode_create(dir
, dentry
, mode
);
2756 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2758 fsnotify_create(dir
, dentry
);
2761 EXPORT_SYMBOL(vfs_create
);
2763 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2765 struct dentry
*dentry
= path
->dentry
;
2766 struct inode
*inode
= dentry
->d_inode
;
2772 switch (inode
->i_mode
& S_IFMT
) {
2776 if (acc_mode
& MAY_WRITE
)
2781 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2790 error
= inode_permission(inode
, MAY_OPEN
| acc_mode
);
2795 * An append-only file must be opened in append mode for writing.
2797 if (IS_APPEND(inode
)) {
2798 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2804 /* O_NOATIME can only be set by the owner or superuser */
2805 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2811 static int handle_truncate(struct file
*filp
)
2813 struct path
*path
= &filp
->f_path
;
2814 struct inode
*inode
= path
->dentry
->d_inode
;
2815 int error
= get_write_access(inode
);
2819 * Refuse to truncate files with mandatory locks held on them.
2821 error
= locks_verify_locked(filp
);
2823 error
= security_path_truncate(path
);
2825 error
= do_truncate(path
->dentry
, 0,
2826 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2829 put_write_access(inode
);
2833 static inline int open_to_namei_flags(int flag
)
2835 if ((flag
& O_ACCMODE
) == 3)
2840 static int may_o_create(const struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2842 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2846 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2850 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2854 * Attempt to atomically look up, create and open a file from a negative
2857 * Returns 0 if successful. The file will have been created and attached to
2858 * @file by the filesystem calling finish_open().
2860 * Returns 1 if the file was looked up only or didn't need creating. The
2861 * caller will need to perform the open themselves. @path will have been
2862 * updated to point to the new dentry. This may be negative.
2864 * Returns an error code otherwise.
2866 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2867 struct path
*path
, struct file
*file
,
2868 const struct open_flags
*op
,
2869 int open_flag
, umode_t mode
,
2872 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2873 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2876 if (!(~open_flag
& (O_EXCL
| O_CREAT
))) /* both O_EXCL and O_CREAT */
2877 open_flag
&= ~O_TRUNC
;
2879 if (nd
->flags
& LOOKUP_DIRECTORY
)
2880 open_flag
|= O_DIRECTORY
;
2882 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2883 file
->f_path
.mnt
= nd
->path
.mnt
;
2884 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
2885 open_to_namei_flags(open_flag
),
2887 d_lookup_done(dentry
);
2890 * We didn't have the inode before the open, so check open
2893 int acc_mode
= op
->acc_mode
;
2894 if (*opened
& FILE_CREATED
) {
2895 WARN_ON(!(open_flag
& O_CREAT
));
2896 fsnotify_create(dir
, dentry
);
2899 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2900 if (WARN_ON(error
> 0))
2902 } else if (error
> 0) {
2903 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2906 if (file
->f_path
.dentry
) {
2908 dentry
= file
->f_path
.dentry
;
2910 if (*opened
& FILE_CREATED
)
2911 fsnotify_create(dir
, dentry
);
2912 path
->dentry
= dentry
;
2913 path
->mnt
= nd
->path
.mnt
;
2922 * Look up and maybe create and open the last component.
2924 * Must be called with i_mutex held on parent.
2926 * Returns 0 if the file was successfully atomically created (if necessary) and
2927 * opened. In this case the file will be returned attached to @file.
2929 * Returns 1 if the file was not completely opened at this time, though lookups
2930 * and creations will have been performed and the dentry returned in @path will
2931 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2932 * specified then a negative dentry may be returned.
2934 * An error code is returned otherwise.
2936 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2937 * cleared otherwise prior to returning.
2939 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2941 const struct open_flags
*op
,
2942 bool got_write
, int *opened
)
2944 struct dentry
*dir
= nd
->path
.dentry
;
2945 struct inode
*dir_inode
= dir
->d_inode
;
2946 int open_flag
= op
->open_flag
;
2947 struct dentry
*dentry
;
2948 int error
, create_error
= 0;
2949 umode_t mode
= op
->mode
;
2950 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
2952 if (unlikely(IS_DEADDIR(dir_inode
)))
2955 *opened
&= ~FILE_CREATED
;
2956 dentry
= d_lookup(dir
, &nd
->last
);
2959 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
2961 return PTR_ERR(dentry
);
2963 if (d_in_lookup(dentry
))
2966 if (!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
2969 error
= d_revalidate(dentry
, nd
->flags
);
2970 if (likely(error
> 0))
2974 d_invalidate(dentry
);
2978 if (dentry
->d_inode
) {
2979 /* Cached positive dentry: will open in f_op->open */
2984 * Checking write permission is tricky, bacuse we don't know if we are
2985 * going to actually need it: O_CREAT opens should work as long as the
2986 * file exists. But checking existence breaks atomicity. The trick is
2987 * to check access and if not granted clear O_CREAT from the flags.
2989 * Another problem is returing the "right" error value (e.g. for an
2990 * O_EXCL open we want to return EEXIST not EROFS).
2992 if (open_flag
& O_CREAT
) {
2993 if (!IS_POSIXACL(dir
->d_inode
))
2994 mode
&= ~current_umask();
2995 if (unlikely(!got_write
)) {
2996 create_error
= -EROFS
;
2997 open_flag
&= ~O_CREAT
;
2998 if (open_flag
& (O_EXCL
| O_TRUNC
))
3000 /* No side effects, safe to clear O_CREAT */
3002 create_error
= may_o_create(&nd
->path
, dentry
, mode
);
3004 open_flag
&= ~O_CREAT
;
3005 if (open_flag
& O_EXCL
)
3009 } else if ((open_flag
& (O_TRUNC
|O_WRONLY
|O_RDWR
)) &&
3010 unlikely(!got_write
)) {
3012 * No O_CREATE -> atomicity not a requirement -> fall
3013 * back to lookup + open
3018 if (dir_inode
->i_op
->atomic_open
) {
3019 error
= atomic_open(nd
, dentry
, path
, file
, op
, open_flag
,
3021 if (unlikely(error
== -ENOENT
) && create_error
)
3022 error
= create_error
;
3027 if (d_in_lookup(dentry
)) {
3028 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3030 d_lookup_done(dentry
);
3031 if (unlikely(res
)) {
3033 error
= PTR_ERR(res
);
3041 /* Negative dentry, just create the file */
3042 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3043 *opened
|= FILE_CREATED
;
3044 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3045 if (!dir_inode
->i_op
->create
) {
3049 error
= dir_inode
->i_op
->create(dir_inode
, dentry
, mode
,
3050 open_flag
& O_EXCL
);
3053 fsnotify_create(dir_inode
, dentry
);
3055 if (unlikely(create_error
) && !dentry
->d_inode
) {
3056 error
= create_error
;
3060 path
->dentry
= dentry
;
3061 path
->mnt
= nd
->path
.mnt
;
3070 * Handle the last step of open()
3072 static int do_last(struct nameidata
*nd
,
3073 struct file
*file
, const struct open_flags
*op
,
3076 struct dentry
*dir
= nd
->path
.dentry
;
3077 int open_flag
= op
->open_flag
;
3078 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3079 bool got_write
= false;
3080 int acc_mode
= op
->acc_mode
;
3082 struct inode
*inode
;
3083 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
3085 bool retried
= false;
3088 nd
->flags
&= ~LOOKUP_PARENT
;
3089 nd
->flags
|= op
->intent
;
3091 if (nd
->last_type
!= LAST_NORM
) {
3092 error
= handle_dots(nd
, nd
->last_type
);
3093 if (unlikely(error
))
3098 if (!(open_flag
& O_CREAT
)) {
3099 if (nd
->last
.name
[nd
->last
.len
])
3100 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3101 /* we _can_ be in RCU mode here */
3102 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3103 if (likely(error
> 0))
3109 BUG_ON(nd
->inode
!= dir
->d_inode
);
3110 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3112 /* create side of things */
3114 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3115 * has been cleared when we got to the last component we are
3118 error
= complete_walk(nd
);
3122 audit_inode(nd
->name
, dir
, LOOKUP_PARENT
);
3123 /* trailing slashes? */
3124 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3129 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3130 error
= mnt_want_write(nd
->path
.mnt
);
3134 * do _not_ fail yet - we might not need that or fail with
3135 * a different error; let lookup_open() decide; we'll be
3136 * dropping this one anyway.
3139 if (open_flag
& O_CREAT
)
3140 inode_lock(dir
->d_inode
);
3142 inode_lock_shared(dir
->d_inode
);
3143 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3144 if (open_flag
& O_CREAT
)
3145 inode_unlock(dir
->d_inode
);
3147 inode_unlock_shared(dir
->d_inode
);
3153 if ((*opened
& FILE_CREATED
) ||
3154 !S_ISREG(file_inode(file
)->i_mode
))
3155 will_truncate
= false;
3157 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3161 if (*opened
& FILE_CREATED
) {
3162 /* Don't check for write permission, don't truncate */
3163 open_flag
&= ~O_TRUNC
;
3164 will_truncate
= false;
3166 path_to_nameidata(&path
, nd
);
3167 goto finish_open_created
;
3171 * If atomic_open() acquired write access it is dropped now due to
3172 * possible mount and symlink following (this might be optimized away if
3176 mnt_drop_write(nd
->path
.mnt
);
3180 if (unlikely(d_is_negative(path
.dentry
))) {
3181 path_to_nameidata(&path
, nd
);
3186 * create/update audit record if it already exists.
3188 audit_inode(nd
->name
, path
.dentry
, 0);
3190 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3191 path_to_nameidata(&path
, nd
);
3195 error
= follow_managed(&path
, nd
);
3196 if (unlikely(error
< 0))
3199 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3200 inode
= d_backing_inode(path
.dentry
);
3204 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
,
3206 if (unlikely(error
))
3209 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3210 path_to_nameidata(&path
, nd
);
3212 save_parent
.dentry
= nd
->path
.dentry
;
3213 save_parent
.mnt
= mntget(path
.mnt
);
3214 nd
->path
.dentry
= path
.dentry
;
3219 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3221 error
= complete_walk(nd
);
3223 path_put(&save_parent
);
3226 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3228 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3231 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3233 if (!d_is_reg(nd
->path
.dentry
))
3234 will_truncate
= false;
3236 if (will_truncate
) {
3237 error
= mnt_want_write(nd
->path
.mnt
);
3242 finish_open_created
:
3243 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3246 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3247 error
= vfs_open(&nd
->path
, file
, current_cred());
3249 *opened
|= FILE_OPENED
;
3251 if (error
== -EOPENSTALE
)
3256 error
= open_check_o_direct(file
);
3258 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3259 if (!error
&& will_truncate
)
3260 error
= handle_truncate(file
);
3262 if (unlikely(error
) && (*opened
& FILE_OPENED
))
3264 if (unlikely(error
> 0)) {
3269 mnt_drop_write(nd
->path
.mnt
);
3270 path_put(&save_parent
);
3274 /* If no saved parent or already retried then can't retry */
3275 if (!save_parent
.dentry
|| retried
)
3278 BUG_ON(save_parent
.dentry
!= dir
);
3279 path_put(&nd
->path
);
3280 nd
->path
= save_parent
;
3281 nd
->inode
= dir
->d_inode
;
3282 save_parent
.mnt
= NULL
;
3283 save_parent
.dentry
= NULL
;
3285 mnt_drop_write(nd
->path
.mnt
);
3292 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3293 const struct open_flags
*op
,
3294 struct file
*file
, int *opened
)
3296 static const struct qstr name
= QSTR_INIT("/", 1);
3297 struct dentry
*child
;
3300 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3301 if (unlikely(error
))
3303 error
= mnt_want_write(path
.mnt
);
3304 if (unlikely(error
))
3306 dir
= path
.dentry
->d_inode
;
3307 /* we want directory to be writable */
3308 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
3311 if (!dir
->i_op
->tmpfile
) {
3312 error
= -EOPNOTSUPP
;
3315 child
= d_alloc(path
.dentry
, &name
);
3316 if (unlikely(!child
)) {
3321 path
.dentry
= child
;
3322 error
= dir
->i_op
->tmpfile(dir
, child
, op
->mode
);
3325 audit_inode(nd
->name
, child
, 0);
3326 /* Don't check for other permissions, the inode was just created */
3327 error
= may_open(&path
, 0, op
->open_flag
);
3330 file
->f_path
.mnt
= path
.mnt
;
3331 error
= finish_open(file
, child
, NULL
, opened
);
3334 error
= open_check_o_direct(file
);
3337 } else if (!(op
->open_flag
& O_EXCL
)) {
3338 struct inode
*inode
= file_inode(file
);
3339 spin_lock(&inode
->i_lock
);
3340 inode
->i_state
|= I_LINKABLE
;
3341 spin_unlock(&inode
->i_lock
);
3344 mnt_drop_write(path
.mnt
);
3350 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3353 int error
= path_lookupat(nd
, flags
, &path
);
3355 audit_inode(nd
->name
, path
.dentry
, 0);
3356 error
= vfs_open(&path
, file
, current_cred());
3362 static struct file
*path_openat(struct nameidata
*nd
,
3363 const struct open_flags
*op
, unsigned flags
)
3370 file
= get_empty_filp();
3374 file
->f_flags
= op
->open_flag
;
3376 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3377 error
= do_tmpfile(nd
, flags
, op
, file
, &opened
);
3381 if (unlikely(file
->f_flags
& O_PATH
)) {
3382 error
= do_o_path(nd
, flags
, file
);
3384 opened
|= FILE_OPENED
;
3388 s
= path_init(nd
, flags
);
3393 while (!(error
= link_path_walk(s
, nd
)) &&
3394 (error
= do_last(nd
, file
, op
, &opened
)) > 0) {
3395 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3396 s
= trailing_symlink(nd
);
3404 if (!(opened
& FILE_OPENED
)) {
3408 if (unlikely(error
)) {
3409 if (error
== -EOPENSTALE
) {
3410 if (flags
& LOOKUP_RCU
)
3415 file
= ERR_PTR(error
);
3420 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3421 const struct open_flags
*op
)
3423 struct nameidata nd
;
3424 int flags
= op
->lookup_flags
;
3427 set_nameidata(&nd
, dfd
, pathname
);
3428 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3429 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3430 filp
= path_openat(&nd
, op
, flags
);
3431 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3432 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3433 restore_nameidata();
3437 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3438 const char *name
, const struct open_flags
*op
)
3440 struct nameidata nd
;
3442 struct filename
*filename
;
3443 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3446 nd
.root
.dentry
= dentry
;
3448 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3449 return ERR_PTR(-ELOOP
);
3451 filename
= getname_kernel(name
);
3452 if (IS_ERR(filename
))
3453 return ERR_CAST(filename
);
3455 set_nameidata(&nd
, -1, filename
);
3456 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3457 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3458 file
= path_openat(&nd
, op
, flags
);
3459 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3460 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3461 restore_nameidata();
3466 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3467 struct path
*path
, unsigned int lookup_flags
)
3469 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3474 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3477 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3478 * other flags passed in are ignored!
3480 lookup_flags
&= LOOKUP_REVAL
;
3482 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3484 return ERR_CAST(name
);
3487 * Yucky last component or no last component at all?
3488 * (foo/., foo/.., /////)
3490 if (unlikely(type
!= LAST_NORM
))
3493 /* don't fail immediately if it's r/o, at least try to report other errors */
3494 err2
= mnt_want_write(path
->mnt
);
3496 * Do the final lookup.
3498 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3499 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3500 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3505 if (d_is_positive(dentry
))
3509 * Special case - lookup gave negative, but... we had foo/bar/
3510 * From the vfs_mknod() POV we just have a negative dentry -
3511 * all is fine. Let's be bastards - you had / on the end, you've
3512 * been asking for (non-existent) directory. -ENOENT for you.
3514 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3518 if (unlikely(err2
)) {
3526 dentry
= ERR_PTR(error
);
3528 inode_unlock(path
->dentry
->d_inode
);
3530 mnt_drop_write(path
->mnt
);
3537 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3538 struct path
*path
, unsigned int lookup_flags
)
3540 return filename_create(dfd
, getname_kernel(pathname
),
3541 path
, lookup_flags
);
3543 EXPORT_SYMBOL(kern_path_create
);
3545 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3548 inode_unlock(path
->dentry
->d_inode
);
3549 mnt_drop_write(path
->mnt
);
3552 EXPORT_SYMBOL(done_path_create
);
3554 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3555 struct path
*path
, unsigned int lookup_flags
)
3557 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3559 EXPORT_SYMBOL(user_path_create
);
3561 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3563 int error
= may_create(dir
, dentry
);
3568 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3571 if (!dir
->i_op
->mknod
)
3574 error
= devcgroup_inode_mknod(mode
, dev
);
3578 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3582 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3584 fsnotify_create(dir
, dentry
);
3587 EXPORT_SYMBOL(vfs_mknod
);
3589 static int may_mknod(umode_t mode
)
3591 switch (mode
& S_IFMT
) {
3597 case 0: /* zero mode translates to S_IFREG */
3606 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3609 struct dentry
*dentry
;
3612 unsigned int lookup_flags
= 0;
3614 error
= may_mknod(mode
);
3618 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3620 return PTR_ERR(dentry
);
3622 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3623 mode
&= ~current_umask();
3624 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3627 switch (mode
& S_IFMT
) {
3628 case 0: case S_IFREG
:
3629 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3631 case S_IFCHR
: case S_IFBLK
:
3632 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3633 new_decode_dev(dev
));
3635 case S_IFIFO
: case S_IFSOCK
:
3636 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3640 done_path_create(&path
, dentry
);
3641 if (retry_estale(error
, lookup_flags
)) {
3642 lookup_flags
|= LOOKUP_REVAL
;
3648 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3650 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3653 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3655 int error
= may_create(dir
, dentry
);
3656 unsigned max_links
= dir
->i_sb
->s_max_links
;
3661 if (!dir
->i_op
->mkdir
)
3664 mode
&= (S_IRWXUGO
|S_ISVTX
);
3665 error
= security_inode_mkdir(dir
, dentry
, mode
);
3669 if (max_links
&& dir
->i_nlink
>= max_links
)
3672 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3674 fsnotify_mkdir(dir
, dentry
);
3677 EXPORT_SYMBOL(vfs_mkdir
);
3679 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3681 struct dentry
*dentry
;
3684 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3687 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3689 return PTR_ERR(dentry
);
3691 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3692 mode
&= ~current_umask();
3693 error
= security_path_mkdir(&path
, dentry
, mode
);
3695 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3696 done_path_create(&path
, dentry
);
3697 if (retry_estale(error
, lookup_flags
)) {
3698 lookup_flags
|= LOOKUP_REVAL
;
3704 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3706 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3709 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3711 int error
= may_delete(dir
, dentry
, 1);
3716 if (!dir
->i_op
->rmdir
)
3720 inode_lock(dentry
->d_inode
);
3723 if (is_local_mountpoint(dentry
))
3726 error
= security_inode_rmdir(dir
, dentry
);
3730 shrink_dcache_parent(dentry
);
3731 error
= dir
->i_op
->rmdir(dir
, dentry
);
3735 dentry
->d_inode
->i_flags
|= S_DEAD
;
3737 detach_mounts(dentry
);
3740 inode_unlock(dentry
->d_inode
);
3746 EXPORT_SYMBOL(vfs_rmdir
);
3748 static long do_rmdir(int dfd
, const char __user
*pathname
)
3751 struct filename
*name
;
3752 struct dentry
*dentry
;
3756 unsigned int lookup_flags
= 0;
3758 name
= user_path_parent(dfd
, pathname
,
3759 &path
, &last
, &type
, lookup_flags
);
3761 return PTR_ERR(name
);
3775 error
= mnt_want_write(path
.mnt
);
3779 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3780 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3781 error
= PTR_ERR(dentry
);
3784 if (!dentry
->d_inode
) {
3788 error
= security_path_rmdir(&path
, dentry
);
3791 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3795 inode_unlock(path
.dentry
->d_inode
);
3796 mnt_drop_write(path
.mnt
);
3800 if (retry_estale(error
, lookup_flags
)) {
3801 lookup_flags
|= LOOKUP_REVAL
;
3807 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3809 return do_rmdir(AT_FDCWD
, pathname
);
3813 * vfs_unlink - unlink a filesystem object
3814 * @dir: parent directory
3816 * @delegated_inode: returns victim inode, if the inode is delegated.
3818 * The caller must hold dir->i_mutex.
3820 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3821 * return a reference to the inode in delegated_inode. The caller
3822 * should then break the delegation on that inode and retry. Because
3823 * breaking a delegation may take a long time, the caller should drop
3824 * dir->i_mutex before doing so.
3826 * Alternatively, a caller may pass NULL for delegated_inode. This may
3827 * be appropriate for callers that expect the underlying filesystem not
3828 * to be NFS exported.
3830 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3832 struct inode
*target
= dentry
->d_inode
;
3833 int error
= may_delete(dir
, dentry
, 0);
3838 if (!dir
->i_op
->unlink
)
3842 if (is_local_mountpoint(dentry
))
3845 error
= security_inode_unlink(dir
, dentry
);
3847 error
= try_break_deleg(target
, delegated_inode
);
3850 error
= dir
->i_op
->unlink(dir
, dentry
);
3853 detach_mounts(dentry
);
3858 inode_unlock(target
);
3860 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3861 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3862 fsnotify_link_count(target
);
3868 EXPORT_SYMBOL(vfs_unlink
);
3871 * Make sure that the actual truncation of the file will occur outside its
3872 * directory's i_mutex. Truncate can take a long time if there is a lot of
3873 * writeout happening, and we don't want to prevent access to the directory
3874 * while waiting on the I/O.
3876 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3879 struct filename
*name
;
3880 struct dentry
*dentry
;
3884 struct inode
*inode
= NULL
;
3885 struct inode
*delegated_inode
= NULL
;
3886 unsigned int lookup_flags
= 0;
3888 name
= user_path_parent(dfd
, pathname
,
3889 &path
, &last
, &type
, lookup_flags
);
3891 return PTR_ERR(name
);
3894 if (type
!= LAST_NORM
)
3897 error
= mnt_want_write(path
.mnt
);
3901 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3902 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3903 error
= PTR_ERR(dentry
);
3904 if (!IS_ERR(dentry
)) {
3905 /* Why not before? Because we want correct error value */
3906 if (last
.name
[last
.len
])
3908 inode
= dentry
->d_inode
;
3909 if (d_is_negative(dentry
))
3912 error
= security_path_unlink(&path
, dentry
);
3915 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3919 inode_unlock(path
.dentry
->d_inode
);
3921 iput(inode
); /* truncate the inode here */
3923 if (delegated_inode
) {
3924 error
= break_deleg_wait(&delegated_inode
);
3928 mnt_drop_write(path
.mnt
);
3932 if (retry_estale(error
, lookup_flags
)) {
3933 lookup_flags
|= LOOKUP_REVAL
;
3940 if (d_is_negative(dentry
))
3942 else if (d_is_dir(dentry
))
3949 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3951 if ((flag
& ~AT_REMOVEDIR
) != 0)
3954 if (flag
& AT_REMOVEDIR
)
3955 return do_rmdir(dfd
, pathname
);
3957 return do_unlinkat(dfd
, pathname
);
3960 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3962 return do_unlinkat(AT_FDCWD
, pathname
);
3965 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3967 int error
= may_create(dir
, dentry
);
3972 if (!dir
->i_op
->symlink
)
3975 error
= security_inode_symlink(dir
, dentry
, oldname
);
3979 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3981 fsnotify_create(dir
, dentry
);
3984 EXPORT_SYMBOL(vfs_symlink
);
3986 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3987 int, newdfd
, const char __user
*, newname
)
3990 struct filename
*from
;
3991 struct dentry
*dentry
;
3993 unsigned int lookup_flags
= 0;
3995 from
= getname(oldname
);
3997 return PTR_ERR(from
);
3999 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4000 error
= PTR_ERR(dentry
);
4004 error
= security_path_symlink(&path
, dentry
, from
->name
);
4006 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
4007 done_path_create(&path
, dentry
);
4008 if (retry_estale(error
, lookup_flags
)) {
4009 lookup_flags
|= LOOKUP_REVAL
;
4017 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4019 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4023 * vfs_link - create a new link
4024 * @old_dentry: object to be linked
4026 * @new_dentry: where to create the new link
4027 * @delegated_inode: returns inode needing a delegation break
4029 * The caller must hold dir->i_mutex
4031 * If vfs_link discovers a delegation on the to-be-linked file in need
4032 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4033 * inode in delegated_inode. The caller should then break the delegation
4034 * and retry. Because breaking a delegation may take a long time, the
4035 * caller should drop the i_mutex before doing so.
4037 * Alternatively, a caller may pass NULL for delegated_inode. This may
4038 * be appropriate for callers that expect the underlying filesystem not
4039 * to be NFS exported.
4041 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4043 struct inode
*inode
= old_dentry
->d_inode
;
4044 unsigned max_links
= dir
->i_sb
->s_max_links
;
4050 error
= may_create(dir
, new_dentry
);
4054 if (dir
->i_sb
!= inode
->i_sb
)
4058 * A link to an append-only or immutable file cannot be created.
4060 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4062 if (!dir
->i_op
->link
)
4064 if (S_ISDIR(inode
->i_mode
))
4067 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4072 /* Make sure we don't allow creating hardlink to an unlinked file */
4073 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4075 else if (max_links
&& inode
->i_nlink
>= max_links
)
4078 error
= try_break_deleg(inode
, delegated_inode
);
4080 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4083 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4084 spin_lock(&inode
->i_lock
);
4085 inode
->i_state
&= ~I_LINKABLE
;
4086 spin_unlock(&inode
->i_lock
);
4088 inode_unlock(inode
);
4090 fsnotify_link(dir
, inode
, new_dentry
);
4093 EXPORT_SYMBOL(vfs_link
);
4096 * Hardlinks are often used in delicate situations. We avoid
4097 * security-related surprises by not following symlinks on the
4100 * We don't follow them on the oldname either to be compatible
4101 * with linux 2.0, and to avoid hard-linking to directories
4102 * and other special files. --ADM
4104 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4105 int, newdfd
, const char __user
*, newname
, int, flags
)
4107 struct dentry
*new_dentry
;
4108 struct path old_path
, new_path
;
4109 struct inode
*delegated_inode
= NULL
;
4113 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4116 * To use null names we require CAP_DAC_READ_SEARCH
4117 * This ensures that not everyone will be able to create
4118 * handlink using the passed filedescriptor.
4120 if (flags
& AT_EMPTY_PATH
) {
4121 if (!capable(CAP_DAC_READ_SEARCH
))
4126 if (flags
& AT_SYMLINK_FOLLOW
)
4127 how
|= LOOKUP_FOLLOW
;
4129 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4133 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4134 (how
& LOOKUP_REVAL
));
4135 error
= PTR_ERR(new_dentry
);
4136 if (IS_ERR(new_dentry
))
4140 if (old_path
.mnt
!= new_path
.mnt
)
4142 error
= may_linkat(&old_path
);
4143 if (unlikely(error
))
4145 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4148 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4150 done_path_create(&new_path
, new_dentry
);
4151 if (delegated_inode
) {
4152 error
= break_deleg_wait(&delegated_inode
);
4154 path_put(&old_path
);
4158 if (retry_estale(error
, how
)) {
4159 path_put(&old_path
);
4160 how
|= LOOKUP_REVAL
;
4164 path_put(&old_path
);
4169 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4171 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4175 * vfs_rename - rename a filesystem object
4176 * @old_dir: parent of source
4177 * @old_dentry: source
4178 * @new_dir: parent of destination
4179 * @new_dentry: destination
4180 * @delegated_inode: returns an inode needing a delegation break
4181 * @flags: rename flags
4183 * The caller must hold multiple mutexes--see lock_rename()).
4185 * If vfs_rename discovers a delegation in need of breaking at either
4186 * the source or destination, it will return -EWOULDBLOCK and return a
4187 * reference to the inode in delegated_inode. The caller should then
4188 * break the delegation and retry. Because breaking a delegation may
4189 * take a long time, the caller should drop all locks before doing
4192 * Alternatively, a caller may pass NULL for delegated_inode. This may
4193 * be appropriate for callers that expect the underlying filesystem not
4194 * to be NFS exported.
4196 * The worst of all namespace operations - renaming directory. "Perverted"
4197 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4199 * a) we can get into loop creation.
4200 * b) race potential - two innocent renames can create a loop together.
4201 * That's where 4.4 screws up. Current fix: serialization on
4202 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4204 * c) we have to lock _four_ objects - parents and victim (if it exists),
4205 * and source (if it is not a directory).
4206 * And that - after we got ->i_mutex on parents (until then we don't know
4207 * whether the target exists). Solution: try to be smart with locking
4208 * order for inodes. We rely on the fact that tree topology may change
4209 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4210 * move will be locked. Thus we can rank directories by the tree
4211 * (ancestors first) and rank all non-directories after them.
4212 * That works since everybody except rename does "lock parent, lookup,
4213 * lock child" and rename is under ->s_vfs_rename_mutex.
4214 * HOWEVER, it relies on the assumption that any object with ->lookup()
4215 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4216 * we'd better make sure that there's no link(2) for them.
4217 * d) conversion from fhandle to dentry may come in the wrong moment - when
4218 * we are removing the target. Solution: we will have to grab ->i_mutex
4219 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4220 * ->i_mutex on parents, which works but leads to some truly excessive
4223 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4224 struct inode
*new_dir
, struct dentry
*new_dentry
,
4225 struct inode
**delegated_inode
, unsigned int flags
)
4228 bool is_dir
= d_is_dir(old_dentry
);
4229 const unsigned char *old_name
;
4230 struct inode
*source
= old_dentry
->d_inode
;
4231 struct inode
*target
= new_dentry
->d_inode
;
4232 bool new_is_dir
= false;
4233 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4236 * Check source == target.
4237 * On overlayfs need to look at underlying inodes.
4239 if (vfs_select_inode(old_dentry
, 0) == vfs_select_inode(new_dentry
, 0))
4242 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4247 error
= may_create(new_dir
, new_dentry
);
4249 new_is_dir
= d_is_dir(new_dentry
);
4251 if (!(flags
& RENAME_EXCHANGE
))
4252 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4254 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4259 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4262 if (flags
&& !old_dir
->i_op
->rename2
)
4266 * If we are going to change the parent - check write permissions,
4267 * we'll need to flip '..'.
4269 if (new_dir
!= old_dir
) {
4271 error
= inode_permission(source
, MAY_WRITE
);
4275 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4276 error
= inode_permission(target
, MAY_WRITE
);
4282 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4287 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4289 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4290 lock_two_nondirectories(source
, target
);
4295 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4298 if (max_links
&& new_dir
!= old_dir
) {
4300 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4302 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4303 old_dir
->i_nlink
>= max_links
)
4306 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4307 shrink_dcache_parent(new_dentry
);
4309 error
= try_break_deleg(source
, delegated_inode
);
4313 if (target
&& !new_is_dir
) {
4314 error
= try_break_deleg(target
, delegated_inode
);
4318 if (!old_dir
->i_op
->rename2
) {
4319 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4320 new_dir
, new_dentry
);
4322 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4323 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4324 new_dir
, new_dentry
, flags
);
4329 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4331 target
->i_flags
|= S_DEAD
;
4332 dont_mount(new_dentry
);
4333 detach_mounts(new_dentry
);
4335 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4336 if (!(flags
& RENAME_EXCHANGE
))
4337 d_move(old_dentry
, new_dentry
);
4339 d_exchange(old_dentry
, new_dentry
);
4342 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4343 unlock_two_nondirectories(source
, target
);
4345 inode_unlock(target
);
4348 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4349 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4350 if (flags
& RENAME_EXCHANGE
) {
4351 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4352 new_is_dir
, NULL
, new_dentry
);
4355 fsnotify_oldname_free(old_name
);
4359 EXPORT_SYMBOL(vfs_rename
);
4361 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4362 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4364 struct dentry
*old_dentry
, *new_dentry
;
4365 struct dentry
*trap
;
4366 struct path old_path
, new_path
;
4367 struct qstr old_last
, new_last
;
4368 int old_type
, new_type
;
4369 struct inode
*delegated_inode
= NULL
;
4370 struct filename
*from
;
4371 struct filename
*to
;
4372 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4373 bool should_retry
= false;
4376 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4379 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4380 (flags
& RENAME_EXCHANGE
))
4383 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4386 if (flags
& RENAME_EXCHANGE
)
4390 from
= user_path_parent(olddfd
, oldname
,
4391 &old_path
, &old_last
, &old_type
, lookup_flags
);
4393 error
= PTR_ERR(from
);
4397 to
= user_path_parent(newdfd
, newname
,
4398 &new_path
, &new_last
, &new_type
, lookup_flags
);
4400 error
= PTR_ERR(to
);
4405 if (old_path
.mnt
!= new_path
.mnt
)
4409 if (old_type
!= LAST_NORM
)
4412 if (flags
& RENAME_NOREPLACE
)
4414 if (new_type
!= LAST_NORM
)
4417 error
= mnt_want_write(old_path
.mnt
);
4422 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4424 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4425 error
= PTR_ERR(old_dentry
);
4426 if (IS_ERR(old_dentry
))
4428 /* source must exist */
4430 if (d_is_negative(old_dentry
))
4432 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4433 error
= PTR_ERR(new_dentry
);
4434 if (IS_ERR(new_dentry
))
4437 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4439 if (flags
& RENAME_EXCHANGE
) {
4441 if (d_is_negative(new_dentry
))
4444 if (!d_is_dir(new_dentry
)) {
4446 if (new_last
.name
[new_last
.len
])
4450 /* unless the source is a directory trailing slashes give -ENOTDIR */
4451 if (!d_is_dir(old_dentry
)) {
4453 if (old_last
.name
[old_last
.len
])
4455 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4458 /* source should not be ancestor of target */
4460 if (old_dentry
== trap
)
4462 /* target should not be an ancestor of source */
4463 if (!(flags
& RENAME_EXCHANGE
))
4465 if (new_dentry
== trap
)
4468 error
= security_path_rename(&old_path
, old_dentry
,
4469 &new_path
, new_dentry
, flags
);
4472 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4473 new_path
.dentry
->d_inode
, new_dentry
,
4474 &delegated_inode
, flags
);
4480 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4481 if (delegated_inode
) {
4482 error
= break_deleg_wait(&delegated_inode
);
4486 mnt_drop_write(old_path
.mnt
);
4488 if (retry_estale(error
, lookup_flags
))
4489 should_retry
= true;
4490 path_put(&new_path
);
4493 path_put(&old_path
);
4496 should_retry
= false;
4497 lookup_flags
|= LOOKUP_REVAL
;
4504 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4505 int, newdfd
, const char __user
*, newname
)
4507 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4510 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4512 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4515 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4517 int error
= may_create(dir
, dentry
);
4521 if (!dir
->i_op
->mknod
)
4524 return dir
->i_op
->mknod(dir
, dentry
,
4525 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4527 EXPORT_SYMBOL(vfs_whiteout
);
4529 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4531 int len
= PTR_ERR(link
);
4536 if (len
> (unsigned) buflen
)
4538 if (copy_to_user(buffer
, link
, len
))
4543 EXPORT_SYMBOL(readlink_copy
);
4546 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4547 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4548 * for any given inode is up to filesystem.
4550 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4552 DEFINE_DELAYED_CALL(done
);
4553 struct inode
*inode
= d_inode(dentry
);
4554 const char *link
= inode
->i_link
;
4558 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4560 return PTR_ERR(link
);
4562 res
= readlink_copy(buffer
, buflen
, link
);
4563 do_delayed_call(&done
);
4566 EXPORT_SYMBOL(generic_readlink
);
4568 /* get the link contents into pagecache */
4569 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4570 struct delayed_call
*callback
)
4574 struct address_space
*mapping
= inode
->i_mapping
;
4577 page
= find_get_page(mapping
, 0);
4579 return ERR_PTR(-ECHILD
);
4580 if (!PageUptodate(page
)) {
4582 return ERR_PTR(-ECHILD
);
4585 page
= read_mapping_page(mapping
, 0, NULL
);
4589 set_delayed_call(callback
, page_put_link
, page
);
4590 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4591 kaddr
= page_address(page
);
4592 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4596 EXPORT_SYMBOL(page_get_link
);
4598 void page_put_link(void *arg
)
4602 EXPORT_SYMBOL(page_put_link
);
4604 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4606 DEFINE_DELAYED_CALL(done
);
4607 int res
= readlink_copy(buffer
, buflen
,
4608 page_get_link(dentry
, d_inode(dentry
),
4610 do_delayed_call(&done
);
4613 EXPORT_SYMBOL(page_readlink
);
4616 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4618 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4620 struct address_space
*mapping
= inode
->i_mapping
;
4624 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4626 flags
|= AOP_FLAG_NOFS
;
4629 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4630 flags
, &page
, &fsdata
);
4634 memcpy(page_address(page
), symname
, len
-1);
4636 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4643 mark_inode_dirty(inode
);
4648 EXPORT_SYMBOL(__page_symlink
);
4650 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4652 return __page_symlink(inode
, symname
, len
,
4653 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4655 EXPORT_SYMBOL(page_symlink
);
4657 const struct inode_operations page_symlink_inode_operations
= {
4658 .readlink
= generic_readlink
,
4659 .get_link
= page_get_link
,
4661 EXPORT_SYMBOL(page_symlink_inode_operations
);