4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user
*filename
, int flags
, int *empty
)
127 struct filename
*result
;
131 result
= audit_reusename(filename
);
135 result
= __getname();
136 if (unlikely(!result
))
137 return ERR_PTR(-ENOMEM
);
140 * First, try to embed the struct filename inside the names_cache
143 kname
= (char *)result
->iname
;
144 result
->name
= kname
;
146 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
147 if (unlikely(len
< 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
159 const size_t size
= offsetof(struct filename
, iname
[1]);
160 kname
= (char *)result
;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result
= kzalloc(size
, GFP_KERNEL
);
168 if (unlikely(!result
)) {
170 return ERR_PTR(-ENOMEM
);
172 result
->name
= kname
;
173 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
174 if (unlikely(len
< 0)) {
179 if (unlikely(len
== PATH_MAX
)) {
182 return ERR_PTR(-ENAMETOOLONG
);
187 /* The empty path is special. */
188 if (unlikely(!len
)) {
191 if (!(flags
& LOOKUP_EMPTY
)) {
193 return ERR_PTR(-ENOENT
);
197 result
->uptr
= filename
;
198 result
->aname
= NULL
;
199 audit_getname(result
);
204 getname(const char __user
* filename
)
206 return getname_flags(filename
, 0, NULL
);
210 getname_kernel(const char * filename
)
212 struct filename
*result
;
213 int len
= strlen(filename
) + 1;
215 result
= __getname();
216 if (unlikely(!result
))
217 return ERR_PTR(-ENOMEM
);
219 if (len
<= EMBEDDED_NAME_MAX
) {
220 result
->name
= (char *)result
->iname
;
221 } else if (len
<= PATH_MAX
) {
222 struct filename
*tmp
;
224 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
225 if (unlikely(!tmp
)) {
227 return ERR_PTR(-ENOMEM
);
229 tmp
->name
= (char *)result
;
233 return ERR_PTR(-ENAMETOOLONG
);
235 memcpy((char *)result
->name
, filename
, len
);
237 result
->aname
= NULL
;
239 audit_getname(result
);
244 void putname(struct filename
*name
)
246 BUG_ON(name
->refcnt
<= 0);
248 if (--name
->refcnt
> 0)
251 if (name
->name
!= name
->iname
) {
252 __putname(name
->name
);
258 static int check_acl(struct inode
*inode
, int mask
)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl
*acl
;
263 if (mask
& MAY_NOT_BLOCK
) {
264 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl
== ACL_NOT_CACHED
)
270 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
273 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
277 int error
= posix_acl_permission(inode
, acl
, mask
);
278 posix_acl_release(acl
);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode
*inode
, int mask
)
291 unsigned int mode
= inode
->i_mode
;
293 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
296 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
297 int error
= check_acl(inode
, mask
);
298 if (error
!= -EAGAIN
)
302 if (in_group_p(inode
->i_gid
))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode
*inode
, int mask
)
333 * Do the basic permission checks.
335 ret
= acl_permission_check(inode
, mask
);
339 if (S_ISDIR(inode
->i_mode
)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
343 if (!(mask
& MAY_WRITE
))
344 if (capable_wrt_inode_uidgid(inode
,
345 CAP_DAC_READ_SEARCH
))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
355 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
359 * Searching includes executable on directories, else just read.
361 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
362 if (mask
== MAY_READ
)
363 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
368 EXPORT_SYMBOL(generic_permission
);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode
*inode
, int mask
)
378 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
379 if (likely(inode
->i_op
->permission
))
380 return inode
->i_op
->permission(inode
, mask
);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode
->i_lock
);
384 inode
->i_opflags
|= IOP_FASTPERM
;
385 spin_unlock(&inode
->i_lock
);
387 return generic_permission(inode
, mask
);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode
*inode
, int mask
)
406 if (unlikely(mask
& MAY_WRITE
)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode
))
414 retval
= do_inode_permission(inode
, mask
);
418 retval
= devcgroup_inode_permission(inode
, mask
);
422 return security_inode_permission(inode
, mask
);
424 EXPORT_SYMBOL(__inode_permission
);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
436 if (unlikely(mask
& MAY_WRITE
)) {
437 umode_t mode
= inode
->i_mode
;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb
->s_flags
& MS_RDONLY
) &&
441 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode
*inode
, int mask
)
462 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
465 return __inode_permission(inode
, mask
);
467 EXPORT_SYMBOL(inode_permission
);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path
*path
)
480 EXPORT_SYMBOL(path_get
);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path
*path
)
493 EXPORT_SYMBOL(path_put
);
495 #define EMBEDDED_LEVELS 2
503 struct inode
*inode
; /* path.dentry.d_inode */
513 } *stack
, internal
[EMBEDDED_LEVELS
];
516 static void set_nameidata(struct nameidata
*nd
)
518 nd
->stack
= nd
->internal
;
521 static void restore_nameidata(struct nameidata
*nd
)
523 if (nd
->stack
!= nd
->internal
) {
525 nd
->stack
= nd
->internal
;
529 static int __nd_alloc_stack(struct nameidata
*nd
)
531 struct saved
*p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
535 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
540 static inline int nd_alloc_stack(struct nameidata
*nd
)
542 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
544 if (likely(nd
->stack
!= nd
->internal
))
546 return __nd_alloc_stack(nd
);
550 * Path walking has 2 modes, rcu-walk and ref-walk (see
551 * Documentation/filesystems/path-lookup.txt). In situations when we can't
552 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
553 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
554 * mode. Refcounts are grabbed at the last known good point before rcu-walk
555 * got stuck, so ref-walk may continue from there. If this is not successful
556 * (eg. a seqcount has changed), then failure is returned and it's up to caller
557 * to restart the path walk from the beginning in ref-walk mode.
561 * unlazy_walk - try to switch to ref-walk mode.
562 * @nd: nameidata pathwalk data
563 * @dentry: child of nd->path.dentry or NULL
564 * Returns: 0 on success, -ECHILD on failure
566 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
567 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
568 * @nd or NULL. Must be called from rcu-walk context.
570 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
572 struct fs_struct
*fs
= current
->fs
;
573 struct dentry
*parent
= nd
->path
.dentry
;
575 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
578 * After legitimizing the bastards, terminate_walk()
579 * will do the right thing for non-RCU mode, and all our
580 * subsequent exit cases should rcu_read_unlock()
581 * before returning. Do vfsmount first; if dentry
582 * can't be legitimized, just set nd->path.dentry to NULL
583 * and rely on dput(NULL) being a no-op.
585 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
587 nd
->flags
&= ~LOOKUP_RCU
;
589 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
590 nd
->path
.dentry
= NULL
;
595 * For a negative lookup, the lookup sequence point is the parents
596 * sequence point, and it only needs to revalidate the parent dentry.
598 * For a positive lookup, we need to move both the parent and the
599 * dentry from the RCU domain to be properly refcounted. And the
600 * sequence number in the dentry validates *both* dentry counters,
601 * since we checked the sequence number of the parent after we got
602 * the child sequence number. So we know the parent must still
603 * be valid if the child sequence number is still valid.
606 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
608 BUG_ON(nd
->inode
!= parent
->d_inode
);
610 if (!lockref_get_not_dead(&dentry
->d_lockref
))
612 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
617 * Sequence counts matched. Now make sure that the root is
618 * still valid and get it if required.
620 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
621 spin_lock(&fs
->lock
);
622 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
623 goto unlock_and_drop_dentry
;
625 spin_unlock(&fs
->lock
);
631 unlock_and_drop_dentry
:
632 spin_unlock(&fs
->lock
);
640 if (!(nd
->flags
& LOOKUP_ROOT
))
645 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
647 return dentry
->d_op
->d_revalidate(dentry
, flags
);
651 * complete_walk - successful completion of path walk
652 * @nd: pointer nameidata
654 * If we had been in RCU mode, drop out of it and legitimize nd->path.
655 * Revalidate the final result, unless we'd already done that during
656 * the path walk or the filesystem doesn't ask for it. Return 0 on
657 * success, -error on failure. In case of failure caller does not
658 * need to drop nd->path.
660 static int complete_walk(struct nameidata
*nd
)
662 struct dentry
*dentry
= nd
->path
.dentry
;
665 if (nd
->flags
& LOOKUP_RCU
) {
666 nd
->flags
&= ~LOOKUP_RCU
;
667 if (!(nd
->flags
& LOOKUP_ROOT
))
670 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
674 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
676 mntput(nd
->path
.mnt
);
679 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
682 mntput(nd
->path
.mnt
);
688 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
691 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
694 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
705 static __always_inline
void set_root(struct nameidata
*nd
)
707 get_fs_root(current
->fs
, &nd
->root
);
710 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
712 struct fs_struct
*fs
= current
->fs
;
716 seq
= read_seqcount_begin(&fs
->seq
);
718 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
719 } while (read_seqcount_retry(&fs
->seq
, seq
));
723 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
726 if (path
->mnt
!= nd
->path
.mnt
)
730 static inline void path_to_nameidata(const struct path
*path
,
731 struct nameidata
*nd
)
733 if (!(nd
->flags
& LOOKUP_RCU
)) {
734 dput(nd
->path
.dentry
);
735 if (nd
->path
.mnt
!= path
->mnt
)
736 mntput(nd
->path
.mnt
);
738 nd
->path
.mnt
= path
->mnt
;
739 nd
->path
.dentry
= path
->dentry
;
743 * Helper to directly jump to a known parsed path from ->follow_link,
744 * caller must have taken a reference to path beforehand.
746 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
751 nd
->inode
= nd
->path
.dentry
->d_inode
;
752 nd
->flags
|= LOOKUP_JUMPED
;
755 static inline void put_link(struct nameidata
*nd
)
757 struct saved
*last
= nd
->stack
+ --nd
->depth
;
758 struct inode
*inode
= last
->link
.dentry
->d_inode
;
759 if (last
->cookie
&& inode
->i_op
->put_link
)
760 inode
->i_op
->put_link(last
->link
.dentry
, last
->cookie
);
761 path_put(&last
->link
);
764 int sysctl_protected_symlinks __read_mostly
= 0;
765 int sysctl_protected_hardlinks __read_mostly
= 0;
768 * may_follow_link - Check symlink following for unsafe situations
769 * @link: The path of the symlink
770 * @nd: nameidata pathwalk data
772 * In the case of the sysctl_protected_symlinks sysctl being enabled,
773 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
774 * in a sticky world-writable directory. This is to protect privileged
775 * processes from failing races against path names that may change out
776 * from under them by way of other users creating malicious symlinks.
777 * It will permit symlinks to be followed only when outside a sticky
778 * world-writable directory, or when the uid of the symlink and follower
779 * match, or when the directory owner matches the symlink's owner.
781 * Returns 0 if following the symlink is allowed, -ve on error.
783 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
785 const struct inode
*inode
;
786 const struct inode
*parent
;
788 if (!sysctl_protected_symlinks
)
791 /* Allowed if owner and follower match. */
792 inode
= link
->dentry
->d_inode
;
793 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
796 /* Allowed if parent directory not sticky and world-writable. */
797 parent
= nd
->path
.dentry
->d_inode
;
798 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
801 /* Allowed if parent directory and link owner match. */
802 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
805 audit_log_link_denied("follow_link", link
);
806 path_put_conditional(link
, nd
);
812 * safe_hardlink_source - Check for safe hardlink conditions
813 * @inode: the source inode to hardlink from
815 * Return false if at least one of the following conditions:
816 * - inode is not a regular file
818 * - inode is setgid and group-exec
819 * - access failure for read and write
821 * Otherwise returns true.
823 static bool safe_hardlink_source(struct inode
*inode
)
825 umode_t mode
= inode
->i_mode
;
827 /* Special files should not get pinned to the filesystem. */
831 /* Setuid files should not get pinned to the filesystem. */
835 /* Executable setgid files should not get pinned to the filesystem. */
836 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
839 /* Hardlinking to unreadable or unwritable sources is dangerous. */
840 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
847 * may_linkat - Check permissions for creating a hardlink
848 * @link: the source to hardlink from
850 * Block hardlink when all of:
851 * - sysctl_protected_hardlinks enabled
852 * - fsuid does not match inode
853 * - hardlink source is unsafe (see safe_hardlink_source() above)
856 * Returns 0 if successful, -ve on error.
858 static int may_linkat(struct path
*link
)
860 const struct cred
*cred
;
863 if (!sysctl_protected_hardlinks
)
866 cred
= current_cred();
867 inode
= link
->dentry
->d_inode
;
869 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
870 * otherwise, it must be a safe source.
872 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
876 audit_log_link_denied("linkat", link
);
880 static __always_inline
881 const char *get_link(struct nameidata
*nd
)
883 struct saved
*last
= nd
->stack
+ nd
->depth
;
884 struct dentry
*dentry
= nd
->link
.dentry
;
885 struct inode
*inode
= dentry
->d_inode
;
889 BUG_ON(nd
->flags
& LOOKUP_RCU
);
891 if (nd
->link
.mnt
== nd
->path
.mnt
)
892 mntget(nd
->link
.mnt
);
894 if (unlikely(current
->total_link_count
>= MAXSYMLINKS
)) {
897 return ERR_PTR(-ELOOP
);
900 last
->link
= nd
->link
;
904 current
->total_link_count
++;
906 touch_atime(&last
->link
);
908 error
= security_inode_follow_link(dentry
);
909 res
= ERR_PTR(error
);
913 nd
->last_type
= LAST_BIND
;
916 res
= inode
->i_op
->follow_link(dentry
, &last
->cookie
, nd
);
920 path_put(&last
->link
);
928 static int follow_up_rcu(struct path
*path
)
930 struct mount
*mnt
= real_mount(path
->mnt
);
931 struct mount
*parent
;
932 struct dentry
*mountpoint
;
934 parent
= mnt
->mnt_parent
;
935 if (&parent
->mnt
== path
->mnt
)
937 mountpoint
= mnt
->mnt_mountpoint
;
938 path
->dentry
= mountpoint
;
939 path
->mnt
= &parent
->mnt
;
944 * follow_up - Find the mountpoint of path's vfsmount
946 * Given a path, find the mountpoint of its source file system.
947 * Replace @path with the path of the mountpoint in the parent mount.
950 * Return 1 if we went up a level and 0 if we were already at the
953 int follow_up(struct path
*path
)
955 struct mount
*mnt
= real_mount(path
->mnt
);
956 struct mount
*parent
;
957 struct dentry
*mountpoint
;
959 read_seqlock_excl(&mount_lock
);
960 parent
= mnt
->mnt_parent
;
962 read_sequnlock_excl(&mount_lock
);
965 mntget(&parent
->mnt
);
966 mountpoint
= dget(mnt
->mnt_mountpoint
);
967 read_sequnlock_excl(&mount_lock
);
969 path
->dentry
= mountpoint
;
971 path
->mnt
= &parent
->mnt
;
974 EXPORT_SYMBOL(follow_up
);
977 * Perform an automount
978 * - return -EISDIR to tell follow_managed() to stop and return the path we
981 static int follow_automount(struct path
*path
, unsigned flags
,
984 struct vfsmount
*mnt
;
987 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
990 /* We don't want to mount if someone's just doing a stat -
991 * unless they're stat'ing a directory and appended a '/' to
994 * We do, however, want to mount if someone wants to open or
995 * create a file of any type under the mountpoint, wants to
996 * traverse through the mountpoint or wants to open the
997 * mounted directory. Also, autofs may mark negative dentries
998 * as being automount points. These will need the attentions
999 * of the daemon to instantiate them before they can be used.
1001 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1002 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1003 path
->dentry
->d_inode
)
1006 current
->total_link_count
++;
1007 if (current
->total_link_count
>= 40)
1010 mnt
= path
->dentry
->d_op
->d_automount(path
);
1013 * The filesystem is allowed to return -EISDIR here to indicate
1014 * it doesn't want to automount. For instance, autofs would do
1015 * this so that its userspace daemon can mount on this dentry.
1017 * However, we can only permit this if it's a terminal point in
1018 * the path being looked up; if it wasn't then the remainder of
1019 * the path is inaccessible and we should say so.
1021 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1023 return PTR_ERR(mnt
);
1026 if (!mnt
) /* mount collision */
1029 if (!*need_mntput
) {
1030 /* lock_mount() may release path->mnt on error */
1032 *need_mntput
= true;
1034 err
= finish_automount(mnt
, path
);
1038 /* Someone else made a mount here whilst we were busy */
1043 path
->dentry
= dget(mnt
->mnt_root
);
1052 * Handle a dentry that is managed in some way.
1053 * - Flagged for transit management (autofs)
1054 * - Flagged as mountpoint
1055 * - Flagged as automount point
1057 * This may only be called in refwalk mode.
1059 * Serialization is taken care of in namespace.c
1061 static int follow_managed(struct path
*path
, unsigned flags
)
1063 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1065 bool need_mntput
= false;
1068 /* Given that we're not holding a lock here, we retain the value in a
1069 * local variable for each dentry as we look at it so that we don't see
1070 * the components of that value change under us */
1071 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1072 managed
&= DCACHE_MANAGED_DENTRY
,
1073 unlikely(managed
!= 0)) {
1074 /* Allow the filesystem to manage the transit without i_mutex
1076 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1077 BUG_ON(!path
->dentry
->d_op
);
1078 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1079 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1084 /* Transit to a mounted filesystem. */
1085 if (managed
& DCACHE_MOUNTED
) {
1086 struct vfsmount
*mounted
= lookup_mnt(path
);
1091 path
->mnt
= mounted
;
1092 path
->dentry
= dget(mounted
->mnt_root
);
1097 /* Something is mounted on this dentry in another
1098 * namespace and/or whatever was mounted there in this
1099 * namespace got unmounted before lookup_mnt() could
1103 /* Handle an automount point */
1104 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1105 ret
= follow_automount(path
, flags
, &need_mntput
);
1111 /* We didn't change the current path point */
1115 if (need_mntput
&& path
->mnt
== mnt
)
1119 return ret
< 0 ? ret
: need_mntput
;
1122 int follow_down_one(struct path
*path
)
1124 struct vfsmount
*mounted
;
1126 mounted
= lookup_mnt(path
);
1130 path
->mnt
= mounted
;
1131 path
->dentry
= dget(mounted
->mnt_root
);
1136 EXPORT_SYMBOL(follow_down_one
);
1138 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1140 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1141 dentry
->d_op
->d_manage(dentry
, true) : 0;
1145 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1146 * we meet a managed dentry that would need blocking.
1148 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1149 struct inode
**inode
)
1152 struct mount
*mounted
;
1154 * Don't forget we might have a non-mountpoint managed dentry
1155 * that wants to block transit.
1157 switch (managed_dentry_rcu(path
->dentry
)) {
1167 if (!d_mountpoint(path
->dentry
))
1168 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1170 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1173 path
->mnt
= &mounted
->mnt
;
1174 path
->dentry
= mounted
->mnt
.mnt_root
;
1175 nd
->flags
|= LOOKUP_JUMPED
;
1176 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1178 * Update the inode too. We don't need to re-check the
1179 * dentry sequence number here after this d_inode read,
1180 * because a mount-point is always pinned.
1182 *inode
= path
->dentry
->d_inode
;
1184 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1185 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1188 static int follow_dotdot_rcu(struct nameidata
*nd
)
1190 struct inode
*inode
= nd
->inode
;
1195 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1196 nd
->path
.mnt
== nd
->root
.mnt
) {
1199 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1200 struct dentry
*old
= nd
->path
.dentry
;
1201 struct dentry
*parent
= old
->d_parent
;
1204 inode
= parent
->d_inode
;
1205 seq
= read_seqcount_begin(&parent
->d_seq
);
1206 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1208 nd
->path
.dentry
= parent
;
1212 if (!follow_up_rcu(&nd
->path
))
1214 inode
= nd
->path
.dentry
->d_inode
;
1215 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1217 while (d_mountpoint(nd
->path
.dentry
)) {
1218 struct mount
*mounted
;
1219 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1222 nd
->path
.mnt
= &mounted
->mnt
;
1223 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1224 inode
= nd
->path
.dentry
->d_inode
;
1225 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1226 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1237 * Follow down to the covering mount currently visible to userspace. At each
1238 * point, the filesystem owning that dentry may be queried as to whether the
1239 * caller is permitted to proceed or not.
1241 int follow_down(struct path
*path
)
1246 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1247 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1248 /* Allow the filesystem to manage the transit without i_mutex
1251 * We indicate to the filesystem if someone is trying to mount
1252 * something here. This gives autofs the chance to deny anyone
1253 * other than its daemon the right to mount on its
1256 * The filesystem may sleep at this point.
1258 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1259 BUG_ON(!path
->dentry
->d_op
);
1260 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1261 ret
= path
->dentry
->d_op
->d_manage(
1262 path
->dentry
, false);
1264 return ret
== -EISDIR
? 0 : ret
;
1267 /* Transit to a mounted filesystem. */
1268 if (managed
& DCACHE_MOUNTED
) {
1269 struct vfsmount
*mounted
= lookup_mnt(path
);
1274 path
->mnt
= mounted
;
1275 path
->dentry
= dget(mounted
->mnt_root
);
1279 /* Don't handle automount points here */
1284 EXPORT_SYMBOL(follow_down
);
1287 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1289 static void follow_mount(struct path
*path
)
1291 while (d_mountpoint(path
->dentry
)) {
1292 struct vfsmount
*mounted
= lookup_mnt(path
);
1297 path
->mnt
= mounted
;
1298 path
->dentry
= dget(mounted
->mnt_root
);
1302 static void follow_dotdot(struct nameidata
*nd
)
1308 struct dentry
*old
= nd
->path
.dentry
;
1310 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1311 nd
->path
.mnt
== nd
->root
.mnt
) {
1314 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1315 /* rare case of legitimate dget_parent()... */
1316 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1320 if (!follow_up(&nd
->path
))
1323 follow_mount(&nd
->path
);
1324 nd
->inode
= nd
->path
.dentry
->d_inode
;
1328 * This looks up the name in dcache, possibly revalidates the old dentry and
1329 * allocates a new one if not found or not valid. In the need_lookup argument
1330 * returns whether i_op->lookup is necessary.
1332 * dir->d_inode->i_mutex must be held
1334 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1335 unsigned int flags
, bool *need_lookup
)
1337 struct dentry
*dentry
;
1340 *need_lookup
= false;
1341 dentry
= d_lookup(dir
, name
);
1343 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1344 error
= d_revalidate(dentry
, flags
);
1345 if (unlikely(error
<= 0)) {
1348 return ERR_PTR(error
);
1350 d_invalidate(dentry
);
1359 dentry
= d_alloc(dir
, name
);
1360 if (unlikely(!dentry
))
1361 return ERR_PTR(-ENOMEM
);
1363 *need_lookup
= true;
1369 * Call i_op->lookup on the dentry. The dentry must be negative and
1372 * dir->d_inode->i_mutex must be held
1374 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1379 /* Don't create child dentry for a dead directory. */
1380 if (unlikely(IS_DEADDIR(dir
))) {
1382 return ERR_PTR(-ENOENT
);
1385 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1386 if (unlikely(old
)) {
1393 static struct dentry
*__lookup_hash(struct qstr
*name
,
1394 struct dentry
*base
, unsigned int flags
)
1397 struct dentry
*dentry
;
1399 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1403 return lookup_real(base
->d_inode
, dentry
, flags
);
1407 * It's more convoluted than I'd like it to be, but... it's still fairly
1408 * small and for now I'd prefer to have fast path as straight as possible.
1409 * It _is_ time-critical.
1411 static int lookup_fast(struct nameidata
*nd
,
1412 struct path
*path
, struct inode
**inode
)
1414 struct vfsmount
*mnt
= nd
->path
.mnt
;
1415 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1421 * Rename seqlock is not required here because in the off chance
1422 * of a false negative due to a concurrent rename, we're going to
1423 * do the non-racy lookup, below.
1425 if (nd
->flags
& LOOKUP_RCU
) {
1428 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1433 * This sequence count validates that the inode matches
1434 * the dentry name information from lookup.
1436 *inode
= dentry
->d_inode
;
1437 negative
= d_is_negative(dentry
);
1438 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1444 * This sequence count validates that the parent had no
1445 * changes while we did the lookup of the dentry above.
1447 * The memory barrier in read_seqcount_begin of child is
1448 * enough, we can use __read_seqcount_retry here.
1450 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1454 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1455 status
= d_revalidate(dentry
, nd
->flags
);
1456 if (unlikely(status
<= 0)) {
1457 if (status
!= -ECHILD
)
1463 path
->dentry
= dentry
;
1464 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1467 if (unlazy_walk(nd
, dentry
))
1470 dentry
= __d_lookup(parent
, &nd
->last
);
1473 if (unlikely(!dentry
))
1476 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1477 status
= d_revalidate(dentry
, nd
->flags
);
1478 if (unlikely(status
<= 0)) {
1483 d_invalidate(dentry
);
1488 if (unlikely(d_is_negative(dentry
))) {
1493 path
->dentry
= dentry
;
1494 err
= follow_managed(path
, nd
->flags
);
1495 if (unlikely(err
< 0)) {
1496 path_put_conditional(path
, nd
);
1500 nd
->flags
|= LOOKUP_JUMPED
;
1501 *inode
= path
->dentry
->d_inode
;
1508 /* Fast lookup failed, do it the slow way */
1509 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1511 struct dentry
*dentry
, *parent
;
1514 parent
= nd
->path
.dentry
;
1515 BUG_ON(nd
->inode
!= parent
->d_inode
);
1517 mutex_lock(&parent
->d_inode
->i_mutex
);
1518 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1519 mutex_unlock(&parent
->d_inode
->i_mutex
);
1521 return PTR_ERR(dentry
);
1522 path
->mnt
= nd
->path
.mnt
;
1523 path
->dentry
= dentry
;
1524 err
= follow_managed(path
, nd
->flags
);
1525 if (unlikely(err
< 0)) {
1526 path_put_conditional(path
, nd
);
1530 nd
->flags
|= LOOKUP_JUMPED
;
1534 static inline int may_lookup(struct nameidata
*nd
)
1536 if (nd
->flags
& LOOKUP_RCU
) {
1537 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1540 if (unlazy_walk(nd
, NULL
))
1543 return inode_permission(nd
->inode
, MAY_EXEC
);
1546 static inline int handle_dots(struct nameidata
*nd
, int type
)
1548 if (type
== LAST_DOTDOT
) {
1549 if (nd
->flags
& LOOKUP_RCU
) {
1550 return follow_dotdot_rcu(nd
);
1557 static void terminate_walk(struct nameidata
*nd
)
1559 if (!(nd
->flags
& LOOKUP_RCU
)) {
1560 path_put(&nd
->path
);
1562 nd
->flags
&= ~LOOKUP_RCU
;
1563 if (!(nd
->flags
& LOOKUP_ROOT
))
1564 nd
->root
.mnt
= NULL
;
1570 * Do we need to follow links? We _really_ want to be able
1571 * to do this check without having to look at inode->i_op,
1572 * so we keep a cache of "no, this doesn't need follow_link"
1573 * for the common case.
1575 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1577 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1580 static int walk_component(struct nameidata
*nd
, int follow
)
1583 struct inode
*inode
;
1586 * "." and ".." are special - ".." especially so because it has
1587 * to be able to know about the current root directory and
1588 * parent relationships.
1590 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1591 err
= handle_dots(nd
, nd
->last_type
);
1596 err
= lookup_fast(nd
, &path
, &inode
);
1597 if (unlikely(err
)) {
1601 err
= lookup_slow(nd
, &path
);
1605 inode
= path
.dentry
->d_inode
;
1607 if (d_is_negative(path
.dentry
))
1611 if (should_follow_link(path
.dentry
, follow
)) {
1612 if (nd
->flags
& LOOKUP_RCU
) {
1613 if (unlikely(nd
->path
.mnt
!= path
.mnt
||
1614 unlazy_walk(nd
, path
.dentry
))) {
1619 BUG_ON(inode
!= path
.dentry
->d_inode
);
1623 path_to_nameidata(&path
, nd
);
1628 path_to_nameidata(&path
, nd
);
1635 * We can do the critical dentry name comparison and hashing
1636 * operations one word at a time, but we are limited to:
1638 * - Architectures with fast unaligned word accesses. We could
1639 * do a "get_unaligned()" if this helps and is sufficiently
1642 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1643 * do not trap on the (extremely unlikely) case of a page
1644 * crossing operation.
1646 * - Furthermore, we need an efficient 64-bit compile for the
1647 * 64-bit case in order to generate the "number of bytes in
1648 * the final mask". Again, that could be replaced with a
1649 * efficient population count instruction or similar.
1651 #ifdef CONFIG_DCACHE_WORD_ACCESS
1653 #include <asm/word-at-a-time.h>
1657 static inline unsigned int fold_hash(unsigned long hash
)
1659 return hash_64(hash
, 32);
1662 #else /* 32-bit case */
1664 #define fold_hash(x) (x)
1668 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1670 unsigned long a
, mask
;
1671 unsigned long hash
= 0;
1674 a
= load_unaligned_zeropad(name
);
1675 if (len
< sizeof(unsigned long))
1679 name
+= sizeof(unsigned long);
1680 len
-= sizeof(unsigned long);
1684 mask
= bytemask_from_count(len
);
1687 return fold_hash(hash
);
1689 EXPORT_SYMBOL(full_name_hash
);
1692 * Calculate the length and hash of the path component, and
1693 * return the "hash_len" as the result.
1695 static inline u64
hash_name(const char *name
)
1697 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1698 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1701 len
= -sizeof(unsigned long);
1703 hash
= (hash
+ a
) * 9;
1704 len
+= sizeof(unsigned long);
1705 a
= load_unaligned_zeropad(name
+len
);
1706 b
= a
^ REPEAT_BYTE('/');
1707 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1709 adata
= prep_zero_mask(a
, adata
, &constants
);
1710 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1712 mask
= create_zero_mask(adata
| bdata
);
1714 hash
+= a
& zero_bytemask(mask
);
1715 len
+= find_zero(mask
);
1716 return hashlen_create(fold_hash(hash
), len
);
1721 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1723 unsigned long hash
= init_name_hash();
1725 hash
= partial_name_hash(*name
++, hash
);
1726 return end_name_hash(hash
);
1728 EXPORT_SYMBOL(full_name_hash
);
1731 * We know there's a real path component here of at least
1734 static inline u64
hash_name(const char *name
)
1736 unsigned long hash
= init_name_hash();
1737 unsigned long len
= 0, c
;
1739 c
= (unsigned char)*name
;
1742 hash
= partial_name_hash(c
, hash
);
1743 c
= (unsigned char)name
[len
];
1744 } while (c
&& c
!= '/');
1745 return hashlen_create(end_name_hash(hash
), len
);
1752 * This is the basic name resolution function, turning a pathname into
1753 * the final dentry. We expect 'base' to be positive and a directory.
1755 * Returns 0 and nd will have valid dentry and mnt on success.
1756 * Returns error and drops reference to input namei data on failure.
1758 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1767 /* At this point we know we have a real path component. */
1772 err
= may_lookup(nd
);
1776 hash_len
= hash_name(name
);
1779 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1781 if (name
[1] == '.') {
1783 nd
->flags
|= LOOKUP_JUMPED
;
1789 if (likely(type
== LAST_NORM
)) {
1790 struct dentry
*parent
= nd
->path
.dentry
;
1791 nd
->flags
&= ~LOOKUP_JUMPED
;
1792 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1793 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1794 err
= parent
->d_op
->d_hash(parent
, &this);
1797 hash_len
= this.hash_len
;
1802 nd
->last
.hash_len
= hash_len
;
1803 nd
->last
.name
= name
;
1804 nd
->last_type
= type
;
1806 name
+= hashlen_len(hash_len
);
1810 * If it wasn't NUL, we know it was '/'. Skip that
1811 * slash, and continue until no more slashes.
1815 } while (unlikely(*name
== '/'));
1819 err
= walk_component(nd
, LOOKUP_FOLLOW
);
1827 err
= nd_alloc_stack(nd
);
1828 if (unlikely(err
)) {
1829 path_to_nameidata(&nd
->link
, nd
);
1835 if (unlikely(IS_ERR(s
))) {
1847 path_put(&nd
->path
);
1848 nd
->path
= nd
->root
;
1849 path_get(&nd
->root
);
1850 nd
->flags
|= LOOKUP_JUMPED
;
1851 while (unlikely(*++s
== '/'))
1854 nd
->inode
= nd
->path
.dentry
->d_inode
;
1855 nd
->stack
[nd
->depth
- 1].name
= name
;
1862 if (!d_can_lookup(nd
->path
.dentry
)) {
1869 while (unlikely(nd
->depth
))
1873 if (!nd
->depth
) /* called from path_init(), done */
1875 name
= nd
->stack
[nd
->depth
- 1].name
;
1876 if (!name
) /* called from trailing_symlink(), done */
1879 err
= walk_component(nd
, LOOKUP_FOLLOW
);
1884 static int path_init(int dfd
, const struct filename
*name
, unsigned int flags
,
1885 struct nameidata
*nd
)
1888 const char *s
= name
->name
;
1890 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1891 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1894 if (flags
& LOOKUP_ROOT
) {
1895 struct dentry
*root
= nd
->root
.dentry
;
1896 struct inode
*inode
= root
->d_inode
;
1898 if (!d_can_lookup(root
))
1900 retval
= inode_permission(inode
, MAY_EXEC
);
1904 nd
->path
= nd
->root
;
1906 if (flags
& LOOKUP_RCU
) {
1908 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1909 nd
->m_seq
= read_seqbegin(&mount_lock
);
1911 path_get(&nd
->path
);
1916 nd
->root
.mnt
= NULL
;
1918 nd
->m_seq
= read_seqbegin(&mount_lock
);
1920 if (flags
& LOOKUP_RCU
) {
1922 nd
->seq
= set_root_rcu(nd
);
1925 path_get(&nd
->root
);
1927 nd
->path
= nd
->root
;
1928 } else if (dfd
== AT_FDCWD
) {
1929 if (flags
& LOOKUP_RCU
) {
1930 struct fs_struct
*fs
= current
->fs
;
1936 seq
= read_seqcount_begin(&fs
->seq
);
1938 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1939 } while (read_seqcount_retry(&fs
->seq
, seq
));
1941 get_fs_pwd(current
->fs
, &nd
->path
);
1944 /* Caller must check execute permissions on the starting path component */
1945 struct fd f
= fdget_raw(dfd
);
1946 struct dentry
*dentry
;
1951 dentry
= f
.file
->f_path
.dentry
;
1954 if (!d_can_lookup(dentry
)) {
1960 nd
->path
= f
.file
->f_path
;
1961 if (flags
& LOOKUP_RCU
) {
1962 if (f
.flags
& FDPUT_FPUT
)
1964 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1967 path_get(&nd
->path
);
1972 nd
->inode
= nd
->path
.dentry
->d_inode
;
1973 if (!(flags
& LOOKUP_RCU
))
1975 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1977 if (!(nd
->flags
& LOOKUP_ROOT
))
1978 nd
->root
.mnt
= NULL
;
1982 current
->total_link_count
= 0;
1983 return link_path_walk(s
, nd
);
1986 static void path_cleanup(struct nameidata
*nd
)
1988 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1989 path_put(&nd
->root
);
1990 nd
->root
.mnt
= NULL
;
1992 if (unlikely(nd
->base
))
1996 static int trailing_symlink(struct nameidata
*nd
)
1999 int error
= may_follow_link(&nd
->link
, nd
);
2000 if (unlikely(error
))
2002 nd
->flags
|= LOOKUP_PARENT
;
2004 if (unlikely(IS_ERR(s
)))
2011 path_put(&nd
->path
);
2012 nd
->path
= nd
->root
;
2013 path_get(&nd
->root
);
2014 nd
->flags
|= LOOKUP_JUMPED
;
2016 nd
->inode
= nd
->path
.dentry
->d_inode
;
2017 nd
->stack
[0].name
= NULL
;
2018 return link_path_walk(s
, nd
);
2021 static inline int lookup_last(struct nameidata
*nd
)
2023 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2024 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2026 nd
->flags
&= ~LOOKUP_PARENT
;
2027 return walk_component(nd
, nd
->flags
& LOOKUP_FOLLOW
);
2030 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2031 static int path_lookupat(int dfd
, const struct filename
*name
,
2032 unsigned int flags
, struct nameidata
*nd
)
2037 * Path walking is largely split up into 2 different synchronisation
2038 * schemes, rcu-walk and ref-walk (explained in
2039 * Documentation/filesystems/path-lookup.txt). These share much of the
2040 * path walk code, but some things particularly setup, cleanup, and
2041 * following mounts are sufficiently divergent that functions are
2042 * duplicated. Typically there is a function foo(), and its RCU
2043 * analogue, foo_rcu().
2045 * -ECHILD is the error number of choice (just to avoid clashes) that
2046 * is returned if some aspect of an rcu-walk fails. Such an error must
2047 * be handled by restarting a traditional ref-walk (which will always
2048 * be able to complete).
2050 err
= path_init(dfd
, name
, flags
, nd
);
2051 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
2052 err
= lookup_last(nd
);
2054 err
= trailing_symlink(nd
);
2057 err
= lookup_last(nd
);
2063 err
= complete_walk(nd
);
2065 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2066 if (!d_can_lookup(nd
->path
.dentry
)) {
2067 path_put(&nd
->path
);
2076 static int filename_lookup(int dfd
, struct filename
*name
,
2077 unsigned int flags
, struct nameidata
*nd
)
2082 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2084 if (unlikely(retval
== -ECHILD
))
2085 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2086 if (unlikely(retval
== -ESTALE
))
2087 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2089 if (likely(!retval
))
2090 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2091 restore_nameidata(nd
);
2095 /* does lookup, returns the object with parent locked */
2096 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2098 struct filename
*filename
= getname_kernel(name
);
2099 struct nameidata nd
;
2103 if (IS_ERR(filename
))
2104 return ERR_CAST(filename
);
2106 err
= filename_lookup(AT_FDCWD
, filename
, LOOKUP_PARENT
, &nd
);
2111 if (nd
.last_type
!= LAST_NORM
) {
2113 d
= ERR_PTR(-EINVAL
);
2116 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2117 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2119 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2129 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2131 struct nameidata nd
;
2132 struct filename
*filename
= getname_kernel(name
);
2133 int res
= PTR_ERR(filename
);
2135 if (!IS_ERR(filename
)) {
2136 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2143 EXPORT_SYMBOL(kern_path
);
2146 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2147 * @dentry: pointer to dentry of the base directory
2148 * @mnt: pointer to vfs mount of the base directory
2149 * @name: pointer to file name
2150 * @flags: lookup flags
2151 * @path: pointer to struct path to fill
2153 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2154 const char *name
, unsigned int flags
,
2157 struct filename
*filename
= getname_kernel(name
);
2158 int err
= PTR_ERR(filename
);
2160 BUG_ON(flags
& LOOKUP_PARENT
);
2162 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2163 if (!IS_ERR(filename
)) {
2164 struct nameidata nd
;
2165 nd
.root
.dentry
= dentry
;
2167 err
= filename_lookup(AT_FDCWD
, filename
,
2168 flags
| LOOKUP_ROOT
, &nd
);
2175 EXPORT_SYMBOL(vfs_path_lookup
);
2178 * lookup_one_len - filesystem helper to lookup single pathname component
2179 * @name: pathname component to lookup
2180 * @base: base directory to lookup from
2181 * @len: maximum length @len should be interpreted to
2183 * Note that this routine is purely a helper for filesystem usage and should
2184 * not be called by generic code.
2186 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2192 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2196 this.hash
= full_name_hash(name
, len
);
2198 return ERR_PTR(-EACCES
);
2200 if (unlikely(name
[0] == '.')) {
2201 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2202 return ERR_PTR(-EACCES
);
2206 c
= *(const unsigned char *)name
++;
2207 if (c
== '/' || c
== '\0')
2208 return ERR_PTR(-EACCES
);
2211 * See if the low-level filesystem might want
2212 * to use its own hash..
2214 if (base
->d_flags
& DCACHE_OP_HASH
) {
2215 int err
= base
->d_op
->d_hash(base
, &this);
2217 return ERR_PTR(err
);
2220 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2222 return ERR_PTR(err
);
2224 return __lookup_hash(&this, base
, 0);
2226 EXPORT_SYMBOL(lookup_one_len
);
2228 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2229 struct path
*path
, int *empty
)
2231 struct nameidata nd
;
2232 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2233 int err
= PTR_ERR(tmp
);
2236 BUG_ON(flags
& LOOKUP_PARENT
);
2238 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2246 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2249 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2251 EXPORT_SYMBOL(user_path_at
);
2254 * NB: most callers don't do anything directly with the reference to the
2255 * to struct filename, but the nd->last pointer points into the name string
2256 * allocated by getname. So we must hold the reference to it until all
2257 * path-walking is complete.
2259 static struct filename
*
2260 user_path_parent(int dfd
, const char __user
*path
,
2261 struct path
*parent
,
2266 struct nameidata nd
;
2267 struct filename
*s
= getname(path
);
2270 /* only LOOKUP_REVAL is allowed in extra flags */
2271 flags
&= LOOKUP_REVAL
;
2276 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, &nd
);
2279 return ERR_PTR(error
);
2283 *type
= nd
.last_type
;
2289 * mountpoint_last - look up last component for umount
2290 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2291 * @path: pointer to container for result
2293 * This is a special lookup_last function just for umount. In this case, we
2294 * need to resolve the path without doing any revalidation.
2296 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2297 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2298 * in almost all cases, this lookup will be served out of the dcache. The only
2299 * cases where it won't are if nd->last refers to a symlink or the path is
2300 * bogus and it doesn't exist.
2303 * -error: if there was an error during lookup. This includes -ENOENT if the
2304 * lookup found a negative dentry. The nd->path reference will also be
2307 * 0: if we successfully resolved nd->path and found it to not to be a
2308 * symlink that needs to be followed. "path" will also be populated.
2309 * The nd->path reference will also be put.
2311 * 1: if we successfully resolved nd->last and found it to be a symlink
2312 * that needs to be followed. "path" will be populated with the path
2313 * to the link, and nd->path will *not* be put.
2316 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2319 struct dentry
*dentry
;
2320 struct dentry
*dir
= nd
->path
.dentry
;
2322 /* If we're in rcuwalk, drop out of it to handle last component */
2323 if (nd
->flags
& LOOKUP_RCU
) {
2324 if (unlazy_walk(nd
, NULL
)) {
2330 nd
->flags
&= ~LOOKUP_PARENT
;
2332 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2333 error
= handle_dots(nd
, nd
->last_type
);
2336 dentry
= dget(nd
->path
.dentry
);
2340 mutex_lock(&dir
->d_inode
->i_mutex
);
2341 dentry
= d_lookup(dir
, &nd
->last
);
2344 * No cached dentry. Mounted dentries are pinned in the cache,
2345 * so that means that this dentry is probably a symlink or the
2346 * path doesn't actually point to a mounted dentry.
2348 dentry
= d_alloc(dir
, &nd
->last
);
2351 mutex_unlock(&dir
->d_inode
->i_mutex
);
2354 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2355 error
= PTR_ERR(dentry
);
2356 if (IS_ERR(dentry
)) {
2357 mutex_unlock(&dir
->d_inode
->i_mutex
);
2361 mutex_unlock(&dir
->d_inode
->i_mutex
);
2364 if (d_is_negative(dentry
)) {
2369 path
->dentry
= dentry
;
2370 path
->mnt
= nd
->path
.mnt
;
2371 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
)) {
2384 * path_mountpoint - look up a path to be umounted
2385 * @dfd: directory file descriptor to start walk from
2386 * @name: full pathname to walk
2387 * @path: pointer to container for result
2388 * @flags: lookup flags
2390 * Look up the given name, but don't attempt to revalidate the last component.
2391 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2394 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2395 struct nameidata
*nd
, unsigned int flags
)
2397 int err
= path_init(dfd
, name
, flags
, nd
);
2401 err
= mountpoint_last(nd
, path
);
2403 err
= trailing_symlink(nd
);
2406 err
= mountpoint_last(nd
, path
);
2415 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2418 struct nameidata nd
;
2421 return PTR_ERR(name
);
2423 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_RCU
);
2424 if (unlikely(error
== -ECHILD
))
2425 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
);
2426 if (unlikely(error
== -ESTALE
))
2427 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_REVAL
);
2429 audit_inode(name
, path
->dentry
, 0);
2430 restore_nameidata(&nd
);
2436 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2437 * @dfd: directory file descriptor
2438 * @name: pathname from userland
2439 * @flags: lookup flags
2440 * @path: pointer to container to hold result
2442 * A umount is a special case for path walking. We're not actually interested
2443 * in the inode in this situation, and ESTALE errors can be a problem. We
2444 * simply want track down the dentry and vfsmount attached at the mountpoint
2445 * and avoid revalidating the last component.
2447 * Returns 0 and populates "path" on success.
2450 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2453 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2457 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2460 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2462 EXPORT_SYMBOL(kern_path_mountpoint
);
2464 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2466 kuid_t fsuid
= current_fsuid();
2468 if (uid_eq(inode
->i_uid
, fsuid
))
2470 if (uid_eq(dir
->i_uid
, fsuid
))
2472 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2474 EXPORT_SYMBOL(__check_sticky
);
2477 * Check whether we can remove a link victim from directory dir, check
2478 * whether the type of victim is right.
2479 * 1. We can't do it if dir is read-only (done in permission())
2480 * 2. We should have write and exec permissions on dir
2481 * 3. We can't remove anything from append-only dir
2482 * 4. We can't do anything with immutable dir (done in permission())
2483 * 5. If the sticky bit on dir is set we should either
2484 * a. be owner of dir, or
2485 * b. be owner of victim, or
2486 * c. have CAP_FOWNER capability
2487 * 6. If the victim is append-only or immutable we can't do antyhing with
2488 * links pointing to it.
2489 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2490 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2491 * 9. We can't remove a root or mountpoint.
2492 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2493 * nfs_async_unlink().
2495 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2497 struct inode
*inode
= victim
->d_inode
;
2500 if (d_is_negative(victim
))
2504 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2505 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2507 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2513 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2514 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2517 if (!d_is_dir(victim
))
2519 if (IS_ROOT(victim
))
2521 } else if (d_is_dir(victim
))
2523 if (IS_DEADDIR(dir
))
2525 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2530 /* Check whether we can create an object with dentry child in directory
2532 * 1. We can't do it if child already exists (open has special treatment for
2533 * this case, but since we are inlined it's OK)
2534 * 2. We can't do it if dir is read-only (done in permission())
2535 * 3. We should have write and exec permissions on dir
2536 * 4. We can't do it if dir is immutable (done in permission())
2538 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2540 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2543 if (IS_DEADDIR(dir
))
2545 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2549 * p1 and p2 should be directories on the same fs.
2551 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2556 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2560 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2562 p
= d_ancestor(p2
, p1
);
2564 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2565 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2569 p
= d_ancestor(p1
, p2
);
2571 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2572 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2576 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2577 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2580 EXPORT_SYMBOL(lock_rename
);
2582 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2584 mutex_unlock(&p1
->d_inode
->i_mutex
);
2586 mutex_unlock(&p2
->d_inode
->i_mutex
);
2587 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2590 EXPORT_SYMBOL(unlock_rename
);
2592 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2595 int error
= may_create(dir
, dentry
);
2599 if (!dir
->i_op
->create
)
2600 return -EACCES
; /* shouldn't it be ENOSYS? */
2603 error
= security_inode_create(dir
, dentry
, mode
);
2606 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2608 fsnotify_create(dir
, dentry
);
2611 EXPORT_SYMBOL(vfs_create
);
2613 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2615 struct dentry
*dentry
= path
->dentry
;
2616 struct inode
*inode
= dentry
->d_inode
;
2626 switch (inode
->i_mode
& S_IFMT
) {
2630 if (acc_mode
& MAY_WRITE
)
2635 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2644 error
= inode_permission(inode
, acc_mode
);
2649 * An append-only file must be opened in append mode for writing.
2651 if (IS_APPEND(inode
)) {
2652 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2658 /* O_NOATIME can only be set by the owner or superuser */
2659 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2665 static int handle_truncate(struct file
*filp
)
2667 struct path
*path
= &filp
->f_path
;
2668 struct inode
*inode
= path
->dentry
->d_inode
;
2669 int error
= get_write_access(inode
);
2673 * Refuse to truncate files with mandatory locks held on them.
2675 error
= locks_verify_locked(filp
);
2677 error
= security_path_truncate(path
);
2679 error
= do_truncate(path
->dentry
, 0,
2680 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2683 put_write_access(inode
);
2687 static inline int open_to_namei_flags(int flag
)
2689 if ((flag
& O_ACCMODE
) == 3)
2694 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2696 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2700 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2704 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2708 * Attempt to atomically look up, create and open a file from a negative
2711 * Returns 0 if successful. The file will have been created and attached to
2712 * @file by the filesystem calling finish_open().
2714 * Returns 1 if the file was looked up only or didn't need creating. The
2715 * caller will need to perform the open themselves. @path will have been
2716 * updated to point to the new dentry. This may be negative.
2718 * Returns an error code otherwise.
2720 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2721 struct path
*path
, struct file
*file
,
2722 const struct open_flags
*op
,
2723 bool got_write
, bool need_lookup
,
2726 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2727 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2731 int create_error
= 0;
2732 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2735 BUG_ON(dentry
->d_inode
);
2737 /* Don't create child dentry for a dead directory. */
2738 if (unlikely(IS_DEADDIR(dir
))) {
2744 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2745 mode
&= ~current_umask();
2747 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2749 open_flag
&= ~O_TRUNC
;
2752 * Checking write permission is tricky, bacuse we don't know if we are
2753 * going to actually need it: O_CREAT opens should work as long as the
2754 * file exists. But checking existence breaks atomicity. The trick is
2755 * to check access and if not granted clear O_CREAT from the flags.
2757 * Another problem is returing the "right" error value (e.g. for an
2758 * O_EXCL open we want to return EEXIST not EROFS).
2760 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2761 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2762 if (!(open_flag
& O_CREAT
)) {
2764 * No O_CREATE -> atomicity not a requirement -> fall
2765 * back to lookup + open
2768 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2769 /* Fall back and fail with the right error */
2770 create_error
= -EROFS
;
2773 /* No side effects, safe to clear O_CREAT */
2774 create_error
= -EROFS
;
2775 open_flag
&= ~O_CREAT
;
2779 if (open_flag
& O_CREAT
) {
2780 error
= may_o_create(&nd
->path
, dentry
, mode
);
2782 create_error
= error
;
2783 if (open_flag
& O_EXCL
)
2785 open_flag
&= ~O_CREAT
;
2789 if (nd
->flags
& LOOKUP_DIRECTORY
)
2790 open_flag
|= O_DIRECTORY
;
2792 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2793 file
->f_path
.mnt
= nd
->path
.mnt
;
2794 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2797 if (create_error
&& error
== -ENOENT
)
2798 error
= create_error
;
2802 if (error
) { /* returned 1, that is */
2803 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2807 if (file
->f_path
.dentry
) {
2809 dentry
= file
->f_path
.dentry
;
2811 if (*opened
& FILE_CREATED
)
2812 fsnotify_create(dir
, dentry
);
2813 if (!dentry
->d_inode
) {
2814 WARN_ON(*opened
& FILE_CREATED
);
2816 error
= create_error
;
2820 if (excl
&& !(*opened
& FILE_CREATED
)) {
2829 * We didn't have the inode before the open, so check open permission
2832 acc_mode
= op
->acc_mode
;
2833 if (*opened
& FILE_CREATED
) {
2834 WARN_ON(!(open_flag
& O_CREAT
));
2835 fsnotify_create(dir
, dentry
);
2836 acc_mode
= MAY_OPEN
;
2838 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2848 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2850 return PTR_ERR(dentry
);
2853 int open_flag
= op
->open_flag
;
2855 error
= create_error
;
2856 if ((open_flag
& O_EXCL
)) {
2857 if (!dentry
->d_inode
)
2859 } else if (!dentry
->d_inode
) {
2861 } else if ((open_flag
& O_TRUNC
) &&
2865 /* will fail later, go on to get the right error */
2869 path
->dentry
= dentry
;
2870 path
->mnt
= nd
->path
.mnt
;
2875 * Look up and maybe create and open the last component.
2877 * Must be called with i_mutex held on parent.
2879 * Returns 0 if the file was successfully atomically created (if necessary) and
2880 * opened. In this case the file will be returned attached to @file.
2882 * Returns 1 if the file was not completely opened at this time, though lookups
2883 * and creations will have been performed and the dentry returned in @path will
2884 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2885 * specified then a negative dentry may be returned.
2887 * An error code is returned otherwise.
2889 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2890 * cleared otherwise prior to returning.
2892 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2894 const struct open_flags
*op
,
2895 bool got_write
, int *opened
)
2897 struct dentry
*dir
= nd
->path
.dentry
;
2898 struct inode
*dir_inode
= dir
->d_inode
;
2899 struct dentry
*dentry
;
2903 *opened
&= ~FILE_CREATED
;
2904 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2906 return PTR_ERR(dentry
);
2908 /* Cached positive dentry: will open in f_op->open */
2909 if (!need_lookup
&& dentry
->d_inode
)
2912 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2913 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2914 need_lookup
, opened
);
2918 BUG_ON(dentry
->d_inode
);
2920 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2922 return PTR_ERR(dentry
);
2925 /* Negative dentry, just create the file */
2926 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2927 umode_t mode
= op
->mode
;
2928 if (!IS_POSIXACL(dir
->d_inode
))
2929 mode
&= ~current_umask();
2931 * This write is needed to ensure that a
2932 * rw->ro transition does not occur between
2933 * the time when the file is created and when
2934 * a permanent write count is taken through
2935 * the 'struct file' in finish_open().
2941 *opened
|= FILE_CREATED
;
2942 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2945 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2946 nd
->flags
& LOOKUP_EXCL
);
2951 path
->dentry
= dentry
;
2952 path
->mnt
= nd
->path
.mnt
;
2961 * Handle the last step of open()
2963 static int do_last(struct nameidata
*nd
,
2964 struct file
*file
, const struct open_flags
*op
,
2965 int *opened
, struct filename
*name
)
2967 struct dentry
*dir
= nd
->path
.dentry
;
2968 int open_flag
= op
->open_flag
;
2969 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2970 bool got_write
= false;
2971 int acc_mode
= op
->acc_mode
;
2972 struct inode
*inode
;
2973 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2975 bool retried
= false;
2978 nd
->flags
&= ~LOOKUP_PARENT
;
2979 nd
->flags
|= op
->intent
;
2981 if (nd
->last_type
!= LAST_NORM
) {
2982 error
= handle_dots(nd
, nd
->last_type
);
2983 if (unlikely(error
)) {
2990 if (!(open_flag
& O_CREAT
)) {
2991 if (nd
->last
.name
[nd
->last
.len
])
2992 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2993 /* we _can_ be in RCU mode here */
2994 error
= lookup_fast(nd
, &path
, &inode
);
3001 BUG_ON(nd
->inode
!= dir
->d_inode
);
3003 /* create side of things */
3005 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3006 * has been cleared when we got to the last component we are
3009 error
= complete_walk(nd
);
3013 audit_inode(name
, dir
, LOOKUP_PARENT
);
3015 /* trailing slashes? */
3016 if (nd
->last
.name
[nd
->last
.len
])
3021 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3022 error
= mnt_want_write(nd
->path
.mnt
);
3026 * do _not_ fail yet - we might not need that or fail with
3027 * a different error; let lookup_open() decide; we'll be
3028 * dropping this one anyway.
3031 mutex_lock(&dir
->d_inode
->i_mutex
);
3032 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3033 mutex_unlock(&dir
->d_inode
->i_mutex
);
3039 if ((*opened
& FILE_CREATED
) ||
3040 !S_ISREG(file_inode(file
)->i_mode
))
3041 will_truncate
= false;
3043 audit_inode(name
, file
->f_path
.dentry
, 0);
3047 if (*opened
& FILE_CREATED
) {
3048 /* Don't check for write permission, don't truncate */
3049 open_flag
&= ~O_TRUNC
;
3050 will_truncate
= false;
3051 acc_mode
= MAY_OPEN
;
3052 path_to_nameidata(&path
, nd
);
3053 goto finish_open_created
;
3057 * create/update audit record if it already exists.
3059 if (d_is_positive(path
.dentry
))
3060 audit_inode(name
, path
.dentry
, 0);
3063 * If atomic_open() acquired write access it is dropped now due to
3064 * possible mount and symlink following (this might be optimized away if
3068 mnt_drop_write(nd
->path
.mnt
);
3073 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3076 error
= follow_managed(&path
, nd
->flags
);
3081 nd
->flags
|= LOOKUP_JUMPED
;
3083 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3084 inode
= path
.dentry
->d_inode
;
3086 if (d_is_negative(path
.dentry
)) {
3087 path_to_nameidata(&path
, nd
);
3091 if (should_follow_link(path
.dentry
, nd
->flags
& LOOKUP_FOLLOW
)) {
3092 if (nd
->flags
& LOOKUP_RCU
) {
3093 if (unlikely(nd
->path
.mnt
!= path
.mnt
||
3094 unlazy_walk(nd
, path
.dentry
))) {
3099 BUG_ON(inode
!= path
.dentry
->d_inode
);
3104 if (unlikely(d_is_symlink(path
.dentry
)) && !(open_flag
& O_PATH
)) {
3105 path_to_nameidata(&path
, nd
);
3110 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3111 path_to_nameidata(&path
, nd
);
3113 save_parent
.dentry
= nd
->path
.dentry
;
3114 save_parent
.mnt
= mntget(path
.mnt
);
3115 nd
->path
.dentry
= path
.dentry
;
3119 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3121 error
= complete_walk(nd
);
3123 path_put(&save_parent
);
3126 audit_inode(name
, nd
->path
.dentry
, 0);
3128 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3131 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3133 if (!d_is_reg(nd
->path
.dentry
))
3134 will_truncate
= false;
3136 if (will_truncate
) {
3137 error
= mnt_want_write(nd
->path
.mnt
);
3142 finish_open_created
:
3143 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3147 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3148 error
= vfs_open(&nd
->path
, file
, current_cred());
3150 *opened
|= FILE_OPENED
;
3152 if (error
== -EOPENSTALE
)
3157 error
= open_check_o_direct(file
);
3160 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3164 if (will_truncate
) {
3165 error
= handle_truncate(file
);
3171 mnt_drop_write(nd
->path
.mnt
);
3172 path_put(&save_parent
);
3177 path_put_conditional(&path
, nd
);
3184 /* If no saved parent or already retried then can't retry */
3185 if (!save_parent
.dentry
|| retried
)
3188 BUG_ON(save_parent
.dentry
!= dir
);
3189 path_put(&nd
->path
);
3190 nd
->path
= save_parent
;
3191 nd
->inode
= dir
->d_inode
;
3192 save_parent
.mnt
= NULL
;
3193 save_parent
.dentry
= NULL
;
3195 mnt_drop_write(nd
->path
.mnt
);
3202 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3203 struct nameidata
*nd
, int flags
,
3204 const struct open_flags
*op
,
3205 struct file
*file
, int *opened
)
3207 static const struct qstr name
= QSTR_INIT("/", 1);
3208 struct dentry
*dentry
, *child
;
3210 int error
= path_lookupat(dfd
, pathname
,
3211 flags
| LOOKUP_DIRECTORY
, nd
);
3212 if (unlikely(error
))
3214 error
= mnt_want_write(nd
->path
.mnt
);
3215 if (unlikely(error
))
3217 /* we want directory to be writable */
3218 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3221 dentry
= nd
->path
.dentry
;
3222 dir
= dentry
->d_inode
;
3223 if (!dir
->i_op
->tmpfile
) {
3224 error
= -EOPNOTSUPP
;
3227 child
= d_alloc(dentry
, &name
);
3228 if (unlikely(!child
)) {
3232 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3233 nd
->flags
|= op
->intent
;
3234 dput(nd
->path
.dentry
);
3235 nd
->path
.dentry
= child
;
3236 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3239 audit_inode(pathname
, nd
->path
.dentry
, 0);
3240 /* Don't check for other permissions, the inode was just created */
3241 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3244 file
->f_path
.mnt
= nd
->path
.mnt
;
3245 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3248 error
= open_check_o_direct(file
);
3251 } else if (!(op
->open_flag
& O_EXCL
)) {
3252 struct inode
*inode
= file_inode(file
);
3253 spin_lock(&inode
->i_lock
);
3254 inode
->i_state
|= I_LINKABLE
;
3255 spin_unlock(&inode
->i_lock
);
3258 mnt_drop_write(nd
->path
.mnt
);
3260 path_put(&nd
->path
);
3264 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3265 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3271 file
= get_empty_filp();
3275 file
->f_flags
= op
->open_flag
;
3277 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3278 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3282 error
= path_init(dfd
, pathname
, flags
, nd
);
3283 if (unlikely(error
))
3286 error
= do_last(nd
, file
, op
, &opened
, pathname
);
3287 while (unlikely(error
> 0)) { /* trailing symlink */
3288 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3289 error
= trailing_symlink(nd
);
3290 if (unlikely(error
))
3292 error
= do_last(nd
, file
, op
, &opened
, pathname
);
3298 if (!(opened
& FILE_OPENED
)) {
3302 if (unlikely(error
)) {
3303 if (error
== -EOPENSTALE
) {
3304 if (flags
& LOOKUP_RCU
)
3309 file
= ERR_PTR(error
);
3314 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3315 const struct open_flags
*op
)
3317 struct nameidata nd
;
3318 int flags
= op
->lookup_flags
;
3322 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3323 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3324 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3325 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3326 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3327 restore_nameidata(&nd
);
3331 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3332 const char *name
, const struct open_flags
*op
)
3334 struct nameidata nd
;
3336 struct filename
*filename
;
3337 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3340 nd
.root
.dentry
= dentry
;
3343 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3344 return ERR_PTR(-ELOOP
);
3346 filename
= getname_kernel(name
);
3347 if (unlikely(IS_ERR(filename
)))
3348 return ERR_CAST(filename
);
3350 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3351 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3352 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3353 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3354 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3355 restore_nameidata(&nd
);
3360 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3361 struct path
*path
, unsigned int lookup_flags
)
3363 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3364 struct nameidata nd
;
3367 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3370 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3371 * other flags passed in are ignored!
3373 lookup_flags
&= LOOKUP_REVAL
;
3375 error
= filename_lookup(dfd
, name
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3377 return ERR_PTR(error
);
3380 * Yucky last component or no last component at all?
3381 * (foo/., foo/.., /////)
3383 if (nd
.last_type
!= LAST_NORM
)
3385 nd
.flags
&= ~LOOKUP_PARENT
;
3386 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3388 /* don't fail immediately if it's r/o, at least try to report other errors */
3389 err2
= mnt_want_write(nd
.path
.mnt
);
3391 * Do the final lookup.
3393 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3394 dentry
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, nd
.flags
);
3399 if (d_is_positive(dentry
))
3403 * Special case - lookup gave negative, but... we had foo/bar/
3404 * From the vfs_mknod() POV we just have a negative dentry -
3405 * all is fine. Let's be bastards - you had / on the end, you've
3406 * been asking for (non-existent) directory. -ENOENT for you.
3408 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3412 if (unlikely(err2
)) {
3420 dentry
= ERR_PTR(error
);
3422 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3424 mnt_drop_write(nd
.path
.mnt
);
3430 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3431 struct path
*path
, unsigned int lookup_flags
)
3433 struct filename
*filename
= getname_kernel(pathname
);
3436 if (IS_ERR(filename
))
3437 return ERR_CAST(filename
);
3438 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3442 EXPORT_SYMBOL(kern_path_create
);
3444 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3447 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3448 mnt_drop_write(path
->mnt
);
3451 EXPORT_SYMBOL(done_path_create
);
3453 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3454 struct path
*path
, unsigned int lookup_flags
)
3456 struct filename
*tmp
= getname(pathname
);
3459 return ERR_CAST(tmp
);
3460 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3464 EXPORT_SYMBOL(user_path_create
);
3466 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3468 int error
= may_create(dir
, dentry
);
3473 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3476 if (!dir
->i_op
->mknod
)
3479 error
= devcgroup_inode_mknod(mode
, dev
);
3483 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3487 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3489 fsnotify_create(dir
, dentry
);
3492 EXPORT_SYMBOL(vfs_mknod
);
3494 static int may_mknod(umode_t mode
)
3496 switch (mode
& S_IFMT
) {
3502 case 0: /* zero mode translates to S_IFREG */
3511 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3514 struct dentry
*dentry
;
3517 unsigned int lookup_flags
= 0;
3519 error
= may_mknod(mode
);
3523 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3525 return PTR_ERR(dentry
);
3527 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3528 mode
&= ~current_umask();
3529 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3532 switch (mode
& S_IFMT
) {
3533 case 0: case S_IFREG
:
3534 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3536 case S_IFCHR
: case S_IFBLK
:
3537 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3538 new_decode_dev(dev
));
3540 case S_IFIFO
: case S_IFSOCK
:
3541 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3545 done_path_create(&path
, dentry
);
3546 if (retry_estale(error
, lookup_flags
)) {
3547 lookup_flags
|= LOOKUP_REVAL
;
3553 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3555 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3558 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3560 int error
= may_create(dir
, dentry
);
3561 unsigned max_links
= dir
->i_sb
->s_max_links
;
3566 if (!dir
->i_op
->mkdir
)
3569 mode
&= (S_IRWXUGO
|S_ISVTX
);
3570 error
= security_inode_mkdir(dir
, dentry
, mode
);
3574 if (max_links
&& dir
->i_nlink
>= max_links
)
3577 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3579 fsnotify_mkdir(dir
, dentry
);
3582 EXPORT_SYMBOL(vfs_mkdir
);
3584 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3586 struct dentry
*dentry
;
3589 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3592 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3594 return PTR_ERR(dentry
);
3596 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3597 mode
&= ~current_umask();
3598 error
= security_path_mkdir(&path
, dentry
, mode
);
3600 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3601 done_path_create(&path
, dentry
);
3602 if (retry_estale(error
, lookup_flags
)) {
3603 lookup_flags
|= LOOKUP_REVAL
;
3609 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3611 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3615 * The dentry_unhash() helper will try to drop the dentry early: we
3616 * should have a usage count of 1 if we're the only user of this
3617 * dentry, and if that is true (possibly after pruning the dcache),
3618 * then we drop the dentry now.
3620 * A low-level filesystem can, if it choses, legally
3623 * if (!d_unhashed(dentry))
3626 * if it cannot handle the case of removing a directory
3627 * that is still in use by something else..
3629 void dentry_unhash(struct dentry
*dentry
)
3631 shrink_dcache_parent(dentry
);
3632 spin_lock(&dentry
->d_lock
);
3633 if (dentry
->d_lockref
.count
== 1)
3635 spin_unlock(&dentry
->d_lock
);
3637 EXPORT_SYMBOL(dentry_unhash
);
3639 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3641 int error
= may_delete(dir
, dentry
, 1);
3646 if (!dir
->i_op
->rmdir
)
3650 mutex_lock(&dentry
->d_inode
->i_mutex
);
3653 if (is_local_mountpoint(dentry
))
3656 error
= security_inode_rmdir(dir
, dentry
);
3660 shrink_dcache_parent(dentry
);
3661 error
= dir
->i_op
->rmdir(dir
, dentry
);
3665 dentry
->d_inode
->i_flags
|= S_DEAD
;
3667 detach_mounts(dentry
);
3670 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3676 EXPORT_SYMBOL(vfs_rmdir
);
3678 static long do_rmdir(int dfd
, const char __user
*pathname
)
3681 struct filename
*name
;
3682 struct dentry
*dentry
;
3686 unsigned int lookup_flags
= 0;
3688 name
= user_path_parent(dfd
, pathname
,
3689 &path
, &last
, &type
, lookup_flags
);
3691 return PTR_ERR(name
);
3705 error
= mnt_want_write(path
.mnt
);
3709 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3710 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3711 error
= PTR_ERR(dentry
);
3714 if (!dentry
->d_inode
) {
3718 error
= security_path_rmdir(&path
, dentry
);
3721 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3725 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3726 mnt_drop_write(path
.mnt
);
3730 if (retry_estale(error
, lookup_flags
)) {
3731 lookup_flags
|= LOOKUP_REVAL
;
3737 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3739 return do_rmdir(AT_FDCWD
, pathname
);
3743 * vfs_unlink - unlink a filesystem object
3744 * @dir: parent directory
3746 * @delegated_inode: returns victim inode, if the inode is delegated.
3748 * The caller must hold dir->i_mutex.
3750 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3751 * return a reference to the inode in delegated_inode. The caller
3752 * should then break the delegation on that inode and retry. Because
3753 * breaking a delegation may take a long time, the caller should drop
3754 * dir->i_mutex before doing so.
3756 * Alternatively, a caller may pass NULL for delegated_inode. This may
3757 * be appropriate for callers that expect the underlying filesystem not
3758 * to be NFS exported.
3760 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3762 struct inode
*target
= dentry
->d_inode
;
3763 int error
= may_delete(dir
, dentry
, 0);
3768 if (!dir
->i_op
->unlink
)
3771 mutex_lock(&target
->i_mutex
);
3772 if (is_local_mountpoint(dentry
))
3775 error
= security_inode_unlink(dir
, dentry
);
3777 error
= try_break_deleg(target
, delegated_inode
);
3780 error
= dir
->i_op
->unlink(dir
, dentry
);
3783 detach_mounts(dentry
);
3788 mutex_unlock(&target
->i_mutex
);
3790 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3791 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3792 fsnotify_link_count(target
);
3798 EXPORT_SYMBOL(vfs_unlink
);
3801 * Make sure that the actual truncation of the file will occur outside its
3802 * directory's i_mutex. Truncate can take a long time if there is a lot of
3803 * writeout happening, and we don't want to prevent access to the directory
3804 * while waiting on the I/O.
3806 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3809 struct filename
*name
;
3810 struct dentry
*dentry
;
3814 struct inode
*inode
= NULL
;
3815 struct inode
*delegated_inode
= NULL
;
3816 unsigned int lookup_flags
= 0;
3818 name
= user_path_parent(dfd
, pathname
,
3819 &path
, &last
, &type
, lookup_flags
);
3821 return PTR_ERR(name
);
3824 if (type
!= LAST_NORM
)
3827 error
= mnt_want_write(path
.mnt
);
3831 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3832 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3833 error
= PTR_ERR(dentry
);
3834 if (!IS_ERR(dentry
)) {
3835 /* Why not before? Because we want correct error value */
3836 if (last
.name
[last
.len
])
3838 inode
= dentry
->d_inode
;
3839 if (d_is_negative(dentry
))
3842 error
= security_path_unlink(&path
, dentry
);
3845 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3849 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3851 iput(inode
); /* truncate the inode here */
3853 if (delegated_inode
) {
3854 error
= break_deleg_wait(&delegated_inode
);
3858 mnt_drop_write(path
.mnt
);
3862 if (retry_estale(error
, lookup_flags
)) {
3863 lookup_flags
|= LOOKUP_REVAL
;
3870 if (d_is_negative(dentry
))
3872 else if (d_is_dir(dentry
))
3879 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3881 if ((flag
& ~AT_REMOVEDIR
) != 0)
3884 if (flag
& AT_REMOVEDIR
)
3885 return do_rmdir(dfd
, pathname
);
3887 return do_unlinkat(dfd
, pathname
);
3890 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3892 return do_unlinkat(AT_FDCWD
, pathname
);
3895 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3897 int error
= may_create(dir
, dentry
);
3902 if (!dir
->i_op
->symlink
)
3905 error
= security_inode_symlink(dir
, dentry
, oldname
);
3909 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3911 fsnotify_create(dir
, dentry
);
3914 EXPORT_SYMBOL(vfs_symlink
);
3916 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3917 int, newdfd
, const char __user
*, newname
)
3920 struct filename
*from
;
3921 struct dentry
*dentry
;
3923 unsigned int lookup_flags
= 0;
3925 from
= getname(oldname
);
3927 return PTR_ERR(from
);
3929 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3930 error
= PTR_ERR(dentry
);
3934 error
= security_path_symlink(&path
, dentry
, from
->name
);
3936 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3937 done_path_create(&path
, dentry
);
3938 if (retry_estale(error
, lookup_flags
)) {
3939 lookup_flags
|= LOOKUP_REVAL
;
3947 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3949 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3953 * vfs_link - create a new link
3954 * @old_dentry: object to be linked
3956 * @new_dentry: where to create the new link
3957 * @delegated_inode: returns inode needing a delegation break
3959 * The caller must hold dir->i_mutex
3961 * If vfs_link discovers a delegation on the to-be-linked file in need
3962 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3963 * inode in delegated_inode. The caller should then break the delegation
3964 * and retry. Because breaking a delegation may take a long time, the
3965 * caller should drop the i_mutex before doing so.
3967 * Alternatively, a caller may pass NULL for delegated_inode. This may
3968 * be appropriate for callers that expect the underlying filesystem not
3969 * to be NFS exported.
3971 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3973 struct inode
*inode
= old_dentry
->d_inode
;
3974 unsigned max_links
= dir
->i_sb
->s_max_links
;
3980 error
= may_create(dir
, new_dentry
);
3984 if (dir
->i_sb
!= inode
->i_sb
)
3988 * A link to an append-only or immutable file cannot be created.
3990 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3992 if (!dir
->i_op
->link
)
3994 if (S_ISDIR(inode
->i_mode
))
3997 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4001 mutex_lock(&inode
->i_mutex
);
4002 /* Make sure we don't allow creating hardlink to an unlinked file */
4003 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4005 else if (max_links
&& inode
->i_nlink
>= max_links
)
4008 error
= try_break_deleg(inode
, delegated_inode
);
4010 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4013 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4014 spin_lock(&inode
->i_lock
);
4015 inode
->i_state
&= ~I_LINKABLE
;
4016 spin_unlock(&inode
->i_lock
);
4018 mutex_unlock(&inode
->i_mutex
);
4020 fsnotify_link(dir
, inode
, new_dentry
);
4023 EXPORT_SYMBOL(vfs_link
);
4026 * Hardlinks are often used in delicate situations. We avoid
4027 * security-related surprises by not following symlinks on the
4030 * We don't follow them on the oldname either to be compatible
4031 * with linux 2.0, and to avoid hard-linking to directories
4032 * and other special files. --ADM
4034 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4035 int, newdfd
, const char __user
*, newname
, int, flags
)
4037 struct dentry
*new_dentry
;
4038 struct path old_path
, new_path
;
4039 struct inode
*delegated_inode
= NULL
;
4043 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4046 * To use null names we require CAP_DAC_READ_SEARCH
4047 * This ensures that not everyone will be able to create
4048 * handlink using the passed filedescriptor.
4050 if (flags
& AT_EMPTY_PATH
) {
4051 if (!capable(CAP_DAC_READ_SEARCH
))
4056 if (flags
& AT_SYMLINK_FOLLOW
)
4057 how
|= LOOKUP_FOLLOW
;
4059 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4063 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4064 (how
& LOOKUP_REVAL
));
4065 error
= PTR_ERR(new_dentry
);
4066 if (IS_ERR(new_dentry
))
4070 if (old_path
.mnt
!= new_path
.mnt
)
4072 error
= may_linkat(&old_path
);
4073 if (unlikely(error
))
4075 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4078 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4080 done_path_create(&new_path
, new_dentry
);
4081 if (delegated_inode
) {
4082 error
= break_deleg_wait(&delegated_inode
);
4084 path_put(&old_path
);
4088 if (retry_estale(error
, how
)) {
4089 path_put(&old_path
);
4090 how
|= LOOKUP_REVAL
;
4094 path_put(&old_path
);
4099 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4101 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4105 * vfs_rename - rename a filesystem object
4106 * @old_dir: parent of source
4107 * @old_dentry: source
4108 * @new_dir: parent of destination
4109 * @new_dentry: destination
4110 * @delegated_inode: returns an inode needing a delegation break
4111 * @flags: rename flags
4113 * The caller must hold multiple mutexes--see lock_rename()).
4115 * If vfs_rename discovers a delegation in need of breaking at either
4116 * the source or destination, it will return -EWOULDBLOCK and return a
4117 * reference to the inode in delegated_inode. The caller should then
4118 * break the delegation and retry. Because breaking a delegation may
4119 * take a long time, the caller should drop all locks before doing
4122 * Alternatively, a caller may pass NULL for delegated_inode. This may
4123 * be appropriate for callers that expect the underlying filesystem not
4124 * to be NFS exported.
4126 * The worst of all namespace operations - renaming directory. "Perverted"
4127 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4129 * a) we can get into loop creation.
4130 * b) race potential - two innocent renames can create a loop together.
4131 * That's where 4.4 screws up. Current fix: serialization on
4132 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4134 * c) we have to lock _four_ objects - parents and victim (if it exists),
4135 * and source (if it is not a directory).
4136 * And that - after we got ->i_mutex on parents (until then we don't know
4137 * whether the target exists). Solution: try to be smart with locking
4138 * order for inodes. We rely on the fact that tree topology may change
4139 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4140 * move will be locked. Thus we can rank directories by the tree
4141 * (ancestors first) and rank all non-directories after them.
4142 * That works since everybody except rename does "lock parent, lookup,
4143 * lock child" and rename is under ->s_vfs_rename_mutex.
4144 * HOWEVER, it relies on the assumption that any object with ->lookup()
4145 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4146 * we'd better make sure that there's no link(2) for them.
4147 * d) conversion from fhandle to dentry may come in the wrong moment - when
4148 * we are removing the target. Solution: we will have to grab ->i_mutex
4149 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4150 * ->i_mutex on parents, which works but leads to some truly excessive
4153 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4154 struct inode
*new_dir
, struct dentry
*new_dentry
,
4155 struct inode
**delegated_inode
, unsigned int flags
)
4158 bool is_dir
= d_is_dir(old_dentry
);
4159 const unsigned char *old_name
;
4160 struct inode
*source
= old_dentry
->d_inode
;
4161 struct inode
*target
= new_dentry
->d_inode
;
4162 bool new_is_dir
= false;
4163 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4165 if (source
== target
)
4168 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4173 error
= may_create(new_dir
, new_dentry
);
4175 new_is_dir
= d_is_dir(new_dentry
);
4177 if (!(flags
& RENAME_EXCHANGE
))
4178 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4180 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4185 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4188 if (flags
&& !old_dir
->i_op
->rename2
)
4192 * If we are going to change the parent - check write permissions,
4193 * we'll need to flip '..'.
4195 if (new_dir
!= old_dir
) {
4197 error
= inode_permission(source
, MAY_WRITE
);
4201 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4202 error
= inode_permission(target
, MAY_WRITE
);
4208 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4213 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4215 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4216 lock_two_nondirectories(source
, target
);
4218 mutex_lock(&target
->i_mutex
);
4221 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4224 if (max_links
&& new_dir
!= old_dir
) {
4226 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4228 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4229 old_dir
->i_nlink
>= max_links
)
4232 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4233 shrink_dcache_parent(new_dentry
);
4235 error
= try_break_deleg(source
, delegated_inode
);
4239 if (target
&& !new_is_dir
) {
4240 error
= try_break_deleg(target
, delegated_inode
);
4244 if (!old_dir
->i_op
->rename2
) {
4245 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4246 new_dir
, new_dentry
);
4248 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4249 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4250 new_dir
, new_dentry
, flags
);
4255 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4257 target
->i_flags
|= S_DEAD
;
4258 dont_mount(new_dentry
);
4259 detach_mounts(new_dentry
);
4261 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4262 if (!(flags
& RENAME_EXCHANGE
))
4263 d_move(old_dentry
, new_dentry
);
4265 d_exchange(old_dentry
, new_dentry
);
4268 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4269 unlock_two_nondirectories(source
, target
);
4271 mutex_unlock(&target
->i_mutex
);
4274 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4275 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4276 if (flags
& RENAME_EXCHANGE
) {
4277 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4278 new_is_dir
, NULL
, new_dentry
);
4281 fsnotify_oldname_free(old_name
);
4285 EXPORT_SYMBOL(vfs_rename
);
4287 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4288 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4290 struct dentry
*old_dentry
, *new_dentry
;
4291 struct dentry
*trap
;
4292 struct path old_path
, new_path
;
4293 struct qstr old_last
, new_last
;
4294 int old_type
, new_type
;
4295 struct inode
*delegated_inode
= NULL
;
4296 struct filename
*from
;
4297 struct filename
*to
;
4298 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4299 bool should_retry
= false;
4302 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4305 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4306 (flags
& RENAME_EXCHANGE
))
4309 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4312 if (flags
& RENAME_EXCHANGE
)
4316 from
= user_path_parent(olddfd
, oldname
,
4317 &old_path
, &old_last
, &old_type
, lookup_flags
);
4319 error
= PTR_ERR(from
);
4323 to
= user_path_parent(newdfd
, newname
,
4324 &new_path
, &new_last
, &new_type
, lookup_flags
);
4326 error
= PTR_ERR(to
);
4331 if (old_path
.mnt
!= new_path
.mnt
)
4335 if (old_type
!= LAST_NORM
)
4338 if (flags
& RENAME_NOREPLACE
)
4340 if (new_type
!= LAST_NORM
)
4343 error
= mnt_want_write(old_path
.mnt
);
4348 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4350 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4351 error
= PTR_ERR(old_dentry
);
4352 if (IS_ERR(old_dentry
))
4354 /* source must exist */
4356 if (d_is_negative(old_dentry
))
4358 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4359 error
= PTR_ERR(new_dentry
);
4360 if (IS_ERR(new_dentry
))
4363 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4365 if (flags
& RENAME_EXCHANGE
) {
4367 if (d_is_negative(new_dentry
))
4370 if (!d_is_dir(new_dentry
)) {
4372 if (new_last
.name
[new_last
.len
])
4376 /* unless the source is a directory trailing slashes give -ENOTDIR */
4377 if (!d_is_dir(old_dentry
)) {
4379 if (old_last
.name
[old_last
.len
])
4381 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4384 /* source should not be ancestor of target */
4386 if (old_dentry
== trap
)
4388 /* target should not be an ancestor of source */
4389 if (!(flags
& RENAME_EXCHANGE
))
4391 if (new_dentry
== trap
)
4394 error
= security_path_rename(&old_path
, old_dentry
,
4395 &new_path
, new_dentry
, flags
);
4398 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4399 new_path
.dentry
->d_inode
, new_dentry
,
4400 &delegated_inode
, flags
);
4406 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4407 if (delegated_inode
) {
4408 error
= break_deleg_wait(&delegated_inode
);
4412 mnt_drop_write(old_path
.mnt
);
4414 if (retry_estale(error
, lookup_flags
))
4415 should_retry
= true;
4416 path_put(&new_path
);
4419 path_put(&old_path
);
4422 should_retry
= false;
4423 lookup_flags
|= LOOKUP_REVAL
;
4430 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4431 int, newdfd
, const char __user
*, newname
)
4433 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4436 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4438 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4441 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4443 int error
= may_create(dir
, dentry
);
4447 if (!dir
->i_op
->mknod
)
4450 return dir
->i_op
->mknod(dir
, dentry
,
4451 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4453 EXPORT_SYMBOL(vfs_whiteout
);
4455 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4457 int len
= PTR_ERR(link
);
4462 if (len
> (unsigned) buflen
)
4464 if (copy_to_user(buffer
, link
, len
))
4469 EXPORT_SYMBOL(readlink_copy
);
4472 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4473 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4474 * using) it for any given inode is up to filesystem.
4476 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4479 const char *link
= dentry
->d_inode
->i_link
;
4483 link
= dentry
->d_inode
->i_op
->follow_link(dentry
, &cookie
, NULL
);
4485 return PTR_ERR(link
);
4487 res
= readlink_copy(buffer
, buflen
, link
);
4488 if (cookie
&& dentry
->d_inode
->i_op
->put_link
)
4489 dentry
->d_inode
->i_op
->put_link(dentry
, cookie
);
4492 EXPORT_SYMBOL(generic_readlink
);
4494 /* get the link contents into pagecache */
4495 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4499 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4500 page
= read_mapping_page(mapping
, 0, NULL
);
4505 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4509 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4511 struct page
*page
= NULL
;
4512 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4515 page_cache_release(page
);
4519 EXPORT_SYMBOL(page_readlink
);
4521 const char *page_follow_link_light(struct dentry
*dentry
, void **cookie
, struct nameidata
*nd
)
4523 struct page
*page
= NULL
;
4524 char *res
= page_getlink(dentry
, &page
);
4529 EXPORT_SYMBOL(page_follow_link_light
);
4531 void page_put_link(struct dentry
*dentry
, void *cookie
)
4533 struct page
*page
= cookie
;
4535 page_cache_release(page
);
4537 EXPORT_SYMBOL(page_put_link
);
4540 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4542 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4544 struct address_space
*mapping
= inode
->i_mapping
;
4549 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4551 flags
|= AOP_FLAG_NOFS
;
4554 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4555 flags
, &page
, &fsdata
);
4559 kaddr
= kmap_atomic(page
);
4560 memcpy(kaddr
, symname
, len
-1);
4561 kunmap_atomic(kaddr
);
4563 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4570 mark_inode_dirty(inode
);
4575 EXPORT_SYMBOL(__page_symlink
);
4577 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4579 return __page_symlink(inode
, symname
, len
,
4580 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4582 EXPORT_SYMBOL(page_symlink
);
4584 const struct inode_operations page_symlink_inode_operations
= {
4585 .readlink
= generic_readlink
,
4586 .follow_link
= page_follow_link_light
,
4587 .put_link
= page_put_link
,
4589 EXPORT_SYMBOL(page_symlink_inode_operations
);