4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user
*filename
, int flags
, int *empty
)
127 struct filename
*result
;
131 result
= audit_reusename(filename
);
135 result
= __getname();
136 if (unlikely(!result
))
137 return ERR_PTR(-ENOMEM
);
140 * First, try to embed the struct filename inside the names_cache
143 kname
= (char *)result
->iname
;
144 result
->name
= kname
;
146 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
147 if (unlikely(len
< 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
159 const size_t size
= offsetof(struct filename
, iname
[1]);
160 kname
= (char *)result
;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result
= kzalloc(size
, GFP_KERNEL
);
168 if (unlikely(!result
)) {
170 return ERR_PTR(-ENOMEM
);
172 result
->name
= kname
;
173 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
174 if (unlikely(len
< 0)) {
179 if (unlikely(len
== PATH_MAX
)) {
182 return ERR_PTR(-ENAMETOOLONG
);
187 /* The empty path is special. */
188 if (unlikely(!len
)) {
191 if (!(flags
& LOOKUP_EMPTY
)) {
193 return ERR_PTR(-ENOENT
);
197 result
->uptr
= filename
;
198 result
->aname
= NULL
;
199 audit_getname(result
);
204 getname(const char __user
* filename
)
206 return getname_flags(filename
, 0, NULL
);
210 getname_kernel(const char * filename
)
212 struct filename
*result
;
213 int len
= strlen(filename
) + 1;
215 result
= __getname();
216 if (unlikely(!result
))
217 return ERR_PTR(-ENOMEM
);
219 if (len
<= EMBEDDED_NAME_MAX
) {
220 result
->name
= (char *)result
->iname
;
221 } else if (len
<= PATH_MAX
) {
222 struct filename
*tmp
;
224 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
225 if (unlikely(!tmp
)) {
227 return ERR_PTR(-ENOMEM
);
229 tmp
->name
= (char *)result
;
233 return ERR_PTR(-ENAMETOOLONG
);
235 memcpy((char *)result
->name
, filename
, len
);
237 result
->aname
= NULL
;
239 audit_getname(result
);
244 void putname(struct filename
*name
)
246 BUG_ON(name
->refcnt
<= 0);
248 if (--name
->refcnt
> 0)
251 if (name
->name
!= name
->iname
) {
252 __putname(name
->name
);
258 static int check_acl(struct inode
*inode
, int mask
)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl
*acl
;
263 if (mask
& MAY_NOT_BLOCK
) {
264 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl
== ACL_NOT_CACHED
)
270 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
273 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
277 int error
= posix_acl_permission(inode
, acl
, mask
);
278 posix_acl_release(acl
);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode
*inode
, int mask
)
291 unsigned int mode
= inode
->i_mode
;
293 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
296 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
297 int error
= check_acl(inode
, mask
);
298 if (error
!= -EAGAIN
)
302 if (in_group_p(inode
->i_gid
))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode
*inode
, int mask
)
333 * Do the basic permission checks.
335 ret
= acl_permission_check(inode
, mask
);
339 if (S_ISDIR(inode
->i_mode
)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
343 if (!(mask
& MAY_WRITE
))
344 if (capable_wrt_inode_uidgid(inode
,
345 CAP_DAC_READ_SEARCH
))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
355 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
359 * Searching includes executable on directories, else just read.
361 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
362 if (mask
== MAY_READ
)
363 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
368 EXPORT_SYMBOL(generic_permission
);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode
*inode
, int mask
)
378 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
379 if (likely(inode
->i_op
->permission
))
380 return inode
->i_op
->permission(inode
, mask
);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode
->i_lock
);
384 inode
->i_opflags
|= IOP_FASTPERM
;
385 spin_unlock(&inode
->i_lock
);
387 return generic_permission(inode
, mask
);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode
*inode
, int mask
)
406 if (unlikely(mask
& MAY_WRITE
)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode
))
414 retval
= do_inode_permission(inode
, mask
);
418 retval
= devcgroup_inode_permission(inode
, mask
);
422 return security_inode_permission(inode
, mask
);
424 EXPORT_SYMBOL(__inode_permission
);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
436 if (unlikely(mask
& MAY_WRITE
)) {
437 umode_t mode
= inode
->i_mode
;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb
->s_flags
& MS_RDONLY
) &&
441 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode
*inode
, int mask
)
462 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
465 return __inode_permission(inode
, mask
);
467 EXPORT_SYMBOL(inode_permission
);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path
*path
)
480 EXPORT_SYMBOL(path_get
);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path
*path
)
493 EXPORT_SYMBOL(path_put
);
495 #define EMBEDDED_LEVELS 2
500 struct inode
*inode
; /* path.dentry.d_inode */
505 int total_link_count
;
510 } *stack
, internal
[EMBEDDED_LEVELS
];
513 static struct nameidata
*set_nameidata(struct nameidata
*p
)
515 struct nameidata
*old
= current
->nameidata
;
516 p
->stack
= p
->internal
;
517 p
->total_link_count
= old
? old
->total_link_count
: 0;
518 current
->nameidata
= p
;
522 static void restore_nameidata(struct nameidata
*old
)
524 struct nameidata
*now
= current
->nameidata
;
526 current
->nameidata
= old
;
528 old
->total_link_count
= now
->total_link_count
;
529 if (now
->stack
!= now
->internal
) {
531 now
->stack
= now
->internal
;
535 static int __nd_alloc_stack(struct nameidata
*nd
)
537 struct saved
*p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
541 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
546 static inline int nd_alloc_stack(struct nameidata
*nd
)
548 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
550 if (likely(nd
->stack
!= nd
->internal
))
552 return __nd_alloc_stack(nd
);
556 * Path walking has 2 modes, rcu-walk and ref-walk (see
557 * Documentation/filesystems/path-lookup.txt). In situations when we can't
558 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
559 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
560 * mode. Refcounts are grabbed at the last known good point before rcu-walk
561 * got stuck, so ref-walk may continue from there. If this is not successful
562 * (eg. a seqcount has changed), then failure is returned and it's up to caller
563 * to restart the path walk from the beginning in ref-walk mode.
567 * unlazy_walk - try to switch to ref-walk mode.
568 * @nd: nameidata pathwalk data
569 * @dentry: child of nd->path.dentry or NULL
570 * Returns: 0 on success, -ECHILD on failure
572 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
573 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
574 * @nd or NULL. Must be called from rcu-walk context.
576 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
578 struct fs_struct
*fs
= current
->fs
;
579 struct dentry
*parent
= nd
->path
.dentry
;
581 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
584 * After legitimizing the bastards, terminate_walk()
585 * will do the right thing for non-RCU mode, and all our
586 * subsequent exit cases should rcu_read_unlock()
587 * before returning. Do vfsmount first; if dentry
588 * can't be legitimized, just set nd->path.dentry to NULL
589 * and rely on dput(NULL) being a no-op.
591 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
593 nd
->flags
&= ~LOOKUP_RCU
;
595 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
596 nd
->path
.dentry
= NULL
;
601 * For a negative lookup, the lookup sequence point is the parents
602 * sequence point, and it only needs to revalidate the parent dentry.
604 * For a positive lookup, we need to move both the parent and the
605 * dentry from the RCU domain to be properly refcounted. And the
606 * sequence number in the dentry validates *both* dentry counters,
607 * since we checked the sequence number of the parent after we got
608 * the child sequence number. So we know the parent must still
609 * be valid if the child sequence number is still valid.
612 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
614 BUG_ON(nd
->inode
!= parent
->d_inode
);
616 if (!lockref_get_not_dead(&dentry
->d_lockref
))
618 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
623 * Sequence counts matched. Now make sure that the root is
624 * still valid and get it if required.
626 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
627 spin_lock(&fs
->lock
);
628 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
629 goto unlock_and_drop_dentry
;
631 spin_unlock(&fs
->lock
);
637 unlock_and_drop_dentry
:
638 spin_unlock(&fs
->lock
);
646 if (!(nd
->flags
& LOOKUP_ROOT
))
651 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
653 return dentry
->d_op
->d_revalidate(dentry
, flags
);
657 * complete_walk - successful completion of path walk
658 * @nd: pointer nameidata
660 * If we had been in RCU mode, drop out of it and legitimize nd->path.
661 * Revalidate the final result, unless we'd already done that during
662 * the path walk or the filesystem doesn't ask for it. Return 0 on
663 * success, -error on failure. In case of failure caller does not
664 * need to drop nd->path.
666 static int complete_walk(struct nameidata
*nd
)
668 struct dentry
*dentry
= nd
->path
.dentry
;
671 if (nd
->flags
& LOOKUP_RCU
) {
672 if (!(nd
->flags
& LOOKUP_ROOT
))
674 if (unlikely(unlazy_walk(nd
, NULL
)))
678 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
681 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
684 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
694 static __always_inline
void set_root(struct nameidata
*nd
)
696 get_fs_root(current
->fs
, &nd
->root
);
699 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
701 struct fs_struct
*fs
= current
->fs
;
705 seq
= read_seqcount_begin(&fs
->seq
);
707 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
708 } while (read_seqcount_retry(&fs
->seq
, seq
));
712 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
715 if (path
->mnt
!= nd
->path
.mnt
)
719 static inline void path_to_nameidata(const struct path
*path
,
720 struct nameidata
*nd
)
722 if (!(nd
->flags
& LOOKUP_RCU
)) {
723 dput(nd
->path
.dentry
);
724 if (nd
->path
.mnt
!= path
->mnt
)
725 mntput(nd
->path
.mnt
);
727 nd
->path
.mnt
= path
->mnt
;
728 nd
->path
.dentry
= path
->dentry
;
732 * Helper to directly jump to a known parsed path from ->follow_link,
733 * caller must have taken a reference to path beforehand.
735 void nd_jump_link(struct path
*path
)
737 struct nameidata
*nd
= current
->nameidata
;
741 nd
->inode
= nd
->path
.dentry
->d_inode
;
742 nd
->flags
|= LOOKUP_JUMPED
;
745 static inline void put_link(struct nameidata
*nd
)
747 struct saved
*last
= nd
->stack
+ --nd
->depth
;
748 struct inode
*inode
= last
->link
.dentry
->d_inode
;
749 if (last
->cookie
&& inode
->i_op
->put_link
)
750 inode
->i_op
->put_link(last
->link
.dentry
, last
->cookie
);
751 path_put(&last
->link
);
754 int sysctl_protected_symlinks __read_mostly
= 0;
755 int sysctl_protected_hardlinks __read_mostly
= 0;
758 * may_follow_link - Check symlink following for unsafe situations
759 * @nd: nameidata pathwalk data
761 * In the case of the sysctl_protected_symlinks sysctl being enabled,
762 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
763 * in a sticky world-writable directory. This is to protect privileged
764 * processes from failing races against path names that may change out
765 * from under them by way of other users creating malicious symlinks.
766 * It will permit symlinks to be followed only when outside a sticky
767 * world-writable directory, or when the uid of the symlink and follower
768 * match, or when the directory owner matches the symlink's owner.
770 * Returns 0 if following the symlink is allowed, -ve on error.
772 static inline int may_follow_link(struct nameidata
*nd
)
774 const struct inode
*inode
;
775 const struct inode
*parent
;
777 if (!sysctl_protected_symlinks
)
780 /* Allowed if owner and follower match. */
781 inode
= nd
->stack
[0].link
.dentry
->d_inode
;
782 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
785 /* Allowed if parent directory not sticky and world-writable. */
786 parent
= nd
->path
.dentry
->d_inode
;
787 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
790 /* Allowed if parent directory and link owner match. */
791 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
794 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
799 * safe_hardlink_source - Check for safe hardlink conditions
800 * @inode: the source inode to hardlink from
802 * Return false if at least one of the following conditions:
803 * - inode is not a regular file
805 * - inode is setgid and group-exec
806 * - access failure for read and write
808 * Otherwise returns true.
810 static bool safe_hardlink_source(struct inode
*inode
)
812 umode_t mode
= inode
->i_mode
;
814 /* Special files should not get pinned to the filesystem. */
818 /* Setuid files should not get pinned to the filesystem. */
822 /* Executable setgid files should not get pinned to the filesystem. */
823 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
826 /* Hardlinking to unreadable or unwritable sources is dangerous. */
827 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
834 * may_linkat - Check permissions for creating a hardlink
835 * @link: the source to hardlink from
837 * Block hardlink when all of:
838 * - sysctl_protected_hardlinks enabled
839 * - fsuid does not match inode
840 * - hardlink source is unsafe (see safe_hardlink_source() above)
843 * Returns 0 if successful, -ve on error.
845 static int may_linkat(struct path
*link
)
847 const struct cred
*cred
;
850 if (!sysctl_protected_hardlinks
)
853 cred
= current_cred();
854 inode
= link
->dentry
->d_inode
;
856 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
857 * otherwise, it must be a safe source.
859 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
863 audit_log_link_denied("linkat", link
);
867 static __always_inline
868 const char *get_link(struct nameidata
*nd
)
870 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
871 struct dentry
*dentry
= last
->link
.dentry
;
872 struct inode
*inode
= dentry
->d_inode
;
876 BUG_ON(nd
->flags
& LOOKUP_RCU
);
880 touch_atime(&last
->link
);
882 error
= security_inode_follow_link(dentry
);
884 return ERR_PTR(error
);
886 nd
->last_type
= LAST_BIND
;
889 res
= inode
->i_op
->follow_link(dentry
, &last
->cookie
);
890 if (IS_ERR_OR_NULL(res
)) {
901 nd
->inode
= nd
->path
.dentry
->d_inode
;
902 nd
->flags
|= LOOKUP_JUMPED
;
903 while (unlikely(*++res
== '/'))
911 static int follow_up_rcu(struct path
*path
)
913 struct mount
*mnt
= real_mount(path
->mnt
);
914 struct mount
*parent
;
915 struct dentry
*mountpoint
;
917 parent
= mnt
->mnt_parent
;
918 if (&parent
->mnt
== path
->mnt
)
920 mountpoint
= mnt
->mnt_mountpoint
;
921 path
->dentry
= mountpoint
;
922 path
->mnt
= &parent
->mnt
;
927 * follow_up - Find the mountpoint of path's vfsmount
929 * Given a path, find the mountpoint of its source file system.
930 * Replace @path with the path of the mountpoint in the parent mount.
933 * Return 1 if we went up a level and 0 if we were already at the
936 int follow_up(struct path
*path
)
938 struct mount
*mnt
= real_mount(path
->mnt
);
939 struct mount
*parent
;
940 struct dentry
*mountpoint
;
942 read_seqlock_excl(&mount_lock
);
943 parent
= mnt
->mnt_parent
;
945 read_sequnlock_excl(&mount_lock
);
948 mntget(&parent
->mnt
);
949 mountpoint
= dget(mnt
->mnt_mountpoint
);
950 read_sequnlock_excl(&mount_lock
);
952 path
->dentry
= mountpoint
;
954 path
->mnt
= &parent
->mnt
;
957 EXPORT_SYMBOL(follow_up
);
960 * Perform an automount
961 * - return -EISDIR to tell follow_managed() to stop and return the path we
964 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
967 struct vfsmount
*mnt
;
970 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
973 /* We don't want to mount if someone's just doing a stat -
974 * unless they're stat'ing a directory and appended a '/' to
977 * We do, however, want to mount if someone wants to open or
978 * create a file of any type under the mountpoint, wants to
979 * traverse through the mountpoint or wants to open the
980 * mounted directory. Also, autofs may mark negative dentries
981 * as being automount points. These will need the attentions
982 * of the daemon to instantiate them before they can be used.
984 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
985 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
986 path
->dentry
->d_inode
)
989 nd
->total_link_count
++;
990 if (nd
->total_link_count
>= 40)
993 mnt
= path
->dentry
->d_op
->d_automount(path
);
996 * The filesystem is allowed to return -EISDIR here to indicate
997 * it doesn't want to automount. For instance, autofs would do
998 * this so that its userspace daemon can mount on this dentry.
1000 * However, we can only permit this if it's a terminal point in
1001 * the path being looked up; if it wasn't then the remainder of
1002 * the path is inaccessible and we should say so.
1004 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1006 return PTR_ERR(mnt
);
1009 if (!mnt
) /* mount collision */
1012 if (!*need_mntput
) {
1013 /* lock_mount() may release path->mnt on error */
1015 *need_mntput
= true;
1017 err
= finish_automount(mnt
, path
);
1021 /* Someone else made a mount here whilst we were busy */
1026 path
->dentry
= dget(mnt
->mnt_root
);
1035 * Handle a dentry that is managed in some way.
1036 * - Flagged for transit management (autofs)
1037 * - Flagged as mountpoint
1038 * - Flagged as automount point
1040 * This may only be called in refwalk mode.
1042 * Serialization is taken care of in namespace.c
1044 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1046 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1048 bool need_mntput
= false;
1051 /* Given that we're not holding a lock here, we retain the value in a
1052 * local variable for each dentry as we look at it so that we don't see
1053 * the components of that value change under us */
1054 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1055 managed
&= DCACHE_MANAGED_DENTRY
,
1056 unlikely(managed
!= 0)) {
1057 /* Allow the filesystem to manage the transit without i_mutex
1059 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1060 BUG_ON(!path
->dentry
->d_op
);
1061 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1062 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1067 /* Transit to a mounted filesystem. */
1068 if (managed
& DCACHE_MOUNTED
) {
1069 struct vfsmount
*mounted
= lookup_mnt(path
);
1074 path
->mnt
= mounted
;
1075 path
->dentry
= dget(mounted
->mnt_root
);
1080 /* Something is mounted on this dentry in another
1081 * namespace and/or whatever was mounted there in this
1082 * namespace got unmounted before lookup_mnt() could
1086 /* Handle an automount point */
1087 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1088 ret
= follow_automount(path
, nd
, &need_mntput
);
1094 /* We didn't change the current path point */
1098 if (need_mntput
&& path
->mnt
== mnt
)
1103 nd
->flags
|= LOOKUP_JUMPED
;
1104 if (unlikely(ret
< 0))
1105 path_put_conditional(path
, nd
);
1109 int follow_down_one(struct path
*path
)
1111 struct vfsmount
*mounted
;
1113 mounted
= lookup_mnt(path
);
1117 path
->mnt
= mounted
;
1118 path
->dentry
= dget(mounted
->mnt_root
);
1123 EXPORT_SYMBOL(follow_down_one
);
1125 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1127 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1128 dentry
->d_op
->d_manage(dentry
, true) : 0;
1132 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1133 * we meet a managed dentry that would need blocking.
1135 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1136 struct inode
**inode
)
1139 struct mount
*mounted
;
1141 * Don't forget we might have a non-mountpoint managed dentry
1142 * that wants to block transit.
1144 switch (managed_dentry_rcu(path
->dentry
)) {
1154 if (!d_mountpoint(path
->dentry
))
1155 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1157 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1160 path
->mnt
= &mounted
->mnt
;
1161 path
->dentry
= mounted
->mnt
.mnt_root
;
1162 nd
->flags
|= LOOKUP_JUMPED
;
1163 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1165 * Update the inode too. We don't need to re-check the
1166 * dentry sequence number here after this d_inode read,
1167 * because a mount-point is always pinned.
1169 *inode
= path
->dentry
->d_inode
;
1171 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1172 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1175 static int follow_dotdot_rcu(struct nameidata
*nd
)
1177 struct inode
*inode
= nd
->inode
;
1182 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1183 nd
->path
.mnt
== nd
->root
.mnt
) {
1186 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1187 struct dentry
*old
= nd
->path
.dentry
;
1188 struct dentry
*parent
= old
->d_parent
;
1191 inode
= parent
->d_inode
;
1192 seq
= read_seqcount_begin(&parent
->d_seq
);
1193 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1195 nd
->path
.dentry
= parent
;
1199 if (!follow_up_rcu(&nd
->path
))
1201 inode
= nd
->path
.dentry
->d_inode
;
1202 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1204 while (d_mountpoint(nd
->path
.dentry
)) {
1205 struct mount
*mounted
;
1206 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1209 nd
->path
.mnt
= &mounted
->mnt
;
1210 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1211 inode
= nd
->path
.dentry
->d_inode
;
1212 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1213 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1224 * Follow down to the covering mount currently visible to userspace. At each
1225 * point, the filesystem owning that dentry may be queried as to whether the
1226 * caller is permitted to proceed or not.
1228 int follow_down(struct path
*path
)
1233 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1234 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1235 /* Allow the filesystem to manage the transit without i_mutex
1238 * We indicate to the filesystem if someone is trying to mount
1239 * something here. This gives autofs the chance to deny anyone
1240 * other than its daemon the right to mount on its
1243 * The filesystem may sleep at this point.
1245 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1246 BUG_ON(!path
->dentry
->d_op
);
1247 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1248 ret
= path
->dentry
->d_op
->d_manage(
1249 path
->dentry
, false);
1251 return ret
== -EISDIR
? 0 : ret
;
1254 /* Transit to a mounted filesystem. */
1255 if (managed
& DCACHE_MOUNTED
) {
1256 struct vfsmount
*mounted
= lookup_mnt(path
);
1261 path
->mnt
= mounted
;
1262 path
->dentry
= dget(mounted
->mnt_root
);
1266 /* Don't handle automount points here */
1271 EXPORT_SYMBOL(follow_down
);
1274 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1276 static void follow_mount(struct path
*path
)
1278 while (d_mountpoint(path
->dentry
)) {
1279 struct vfsmount
*mounted
= lookup_mnt(path
);
1284 path
->mnt
= mounted
;
1285 path
->dentry
= dget(mounted
->mnt_root
);
1289 static void follow_dotdot(struct nameidata
*nd
)
1295 struct dentry
*old
= nd
->path
.dentry
;
1297 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1298 nd
->path
.mnt
== nd
->root
.mnt
) {
1301 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1302 /* rare case of legitimate dget_parent()... */
1303 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1307 if (!follow_up(&nd
->path
))
1310 follow_mount(&nd
->path
);
1311 nd
->inode
= nd
->path
.dentry
->d_inode
;
1315 * This looks up the name in dcache, possibly revalidates the old dentry and
1316 * allocates a new one if not found or not valid. In the need_lookup argument
1317 * returns whether i_op->lookup is necessary.
1319 * dir->d_inode->i_mutex must be held
1321 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1322 unsigned int flags
, bool *need_lookup
)
1324 struct dentry
*dentry
;
1327 *need_lookup
= false;
1328 dentry
= d_lookup(dir
, name
);
1330 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1331 error
= d_revalidate(dentry
, flags
);
1332 if (unlikely(error
<= 0)) {
1335 return ERR_PTR(error
);
1337 d_invalidate(dentry
);
1346 dentry
= d_alloc(dir
, name
);
1347 if (unlikely(!dentry
))
1348 return ERR_PTR(-ENOMEM
);
1350 *need_lookup
= true;
1356 * Call i_op->lookup on the dentry. The dentry must be negative and
1359 * dir->d_inode->i_mutex must be held
1361 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1366 /* Don't create child dentry for a dead directory. */
1367 if (unlikely(IS_DEADDIR(dir
))) {
1369 return ERR_PTR(-ENOENT
);
1372 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1373 if (unlikely(old
)) {
1380 static struct dentry
*__lookup_hash(struct qstr
*name
,
1381 struct dentry
*base
, unsigned int flags
)
1384 struct dentry
*dentry
;
1386 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1390 return lookup_real(base
->d_inode
, dentry
, flags
);
1394 * It's more convoluted than I'd like it to be, but... it's still fairly
1395 * small and for now I'd prefer to have fast path as straight as possible.
1396 * It _is_ time-critical.
1398 static int lookup_fast(struct nameidata
*nd
,
1399 struct path
*path
, struct inode
**inode
)
1401 struct vfsmount
*mnt
= nd
->path
.mnt
;
1402 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1408 * Rename seqlock is not required here because in the off chance
1409 * of a false negative due to a concurrent rename, we're going to
1410 * do the non-racy lookup, below.
1412 if (nd
->flags
& LOOKUP_RCU
) {
1415 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1420 * This sequence count validates that the inode matches
1421 * the dentry name information from lookup.
1423 *inode
= dentry
->d_inode
;
1424 negative
= d_is_negative(dentry
);
1425 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1431 * This sequence count validates that the parent had no
1432 * changes while we did the lookup of the dentry above.
1434 * The memory barrier in read_seqcount_begin of child is
1435 * enough, we can use __read_seqcount_retry here.
1437 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1441 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1442 status
= d_revalidate(dentry
, nd
->flags
);
1443 if (unlikely(status
<= 0)) {
1444 if (status
!= -ECHILD
)
1450 path
->dentry
= dentry
;
1451 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1454 if (unlazy_walk(nd
, dentry
))
1457 dentry
= __d_lookup(parent
, &nd
->last
);
1460 if (unlikely(!dentry
))
1463 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1464 status
= d_revalidate(dentry
, nd
->flags
);
1465 if (unlikely(status
<= 0)) {
1470 d_invalidate(dentry
);
1475 if (unlikely(d_is_negative(dentry
))) {
1480 path
->dentry
= dentry
;
1481 err
= follow_managed(path
, nd
);
1483 *inode
= path
->dentry
->d_inode
;
1490 /* Fast lookup failed, do it the slow way */
1491 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1493 struct dentry
*dentry
, *parent
;
1495 parent
= nd
->path
.dentry
;
1496 BUG_ON(nd
->inode
!= parent
->d_inode
);
1498 mutex_lock(&parent
->d_inode
->i_mutex
);
1499 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1500 mutex_unlock(&parent
->d_inode
->i_mutex
);
1502 return PTR_ERR(dentry
);
1503 path
->mnt
= nd
->path
.mnt
;
1504 path
->dentry
= dentry
;
1505 return follow_managed(path
, nd
);
1508 static inline int may_lookup(struct nameidata
*nd
)
1510 if (nd
->flags
& LOOKUP_RCU
) {
1511 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1514 if (unlazy_walk(nd
, NULL
))
1517 return inode_permission(nd
->inode
, MAY_EXEC
);
1520 static inline int handle_dots(struct nameidata
*nd
, int type
)
1522 if (type
== LAST_DOTDOT
) {
1523 if (nd
->flags
& LOOKUP_RCU
) {
1524 return follow_dotdot_rcu(nd
);
1531 static void terminate_walk(struct nameidata
*nd
)
1533 if (!(nd
->flags
& LOOKUP_RCU
)) {
1534 path_put(&nd
->path
);
1536 nd
->flags
&= ~LOOKUP_RCU
;
1537 if (!(nd
->flags
& LOOKUP_ROOT
))
1538 nd
->root
.mnt
= NULL
;
1541 while (unlikely(nd
->depth
))
1545 static int pick_link(struct nameidata
*nd
, struct path
*link
)
1549 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1550 path_to_nameidata(link
, nd
);
1553 if (nd
->flags
& LOOKUP_RCU
) {
1554 if (unlikely(nd
->path
.mnt
!= link
->mnt
||
1555 unlazy_walk(nd
, link
->dentry
))) {
1559 if (link
->mnt
== nd
->path
.mnt
)
1561 error
= nd_alloc_stack(nd
);
1562 if (unlikely(error
)) {
1567 last
= nd
->stack
+ nd
->depth
++;
1569 last
->cookie
= NULL
;
1574 * Do we need to follow links? We _really_ want to be able
1575 * to do this check without having to look at inode->i_op,
1576 * so we keep a cache of "no, this doesn't need follow_link"
1577 * for the common case.
1579 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
, int follow
)
1581 if (likely(!d_is_symlink(link
->dentry
)))
1585 return pick_link(nd
, link
);
1588 enum {WALK_GET
= 1, WALK_PUT
= 2};
1590 static int walk_component(struct nameidata
*nd
, int flags
)
1593 struct inode
*inode
;
1596 * "." and ".." are special - ".." especially so because it has
1597 * to be able to know about the current root directory and
1598 * parent relationships.
1600 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1601 err
= handle_dots(nd
, nd
->last_type
);
1602 if (flags
& WALK_PUT
)
1606 err
= lookup_fast(nd
, &path
, &inode
);
1607 if (unlikely(err
)) {
1611 err
= lookup_slow(nd
, &path
);
1615 inode
= path
.dentry
->d_inode
;
1617 if (d_is_negative(path
.dentry
))
1621 if (flags
& WALK_PUT
)
1623 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
);
1626 path_to_nameidata(&path
, nd
);
1631 path_to_nameidata(&path
, nd
);
1636 * We can do the critical dentry name comparison and hashing
1637 * operations one word at a time, but we are limited to:
1639 * - Architectures with fast unaligned word accesses. We could
1640 * do a "get_unaligned()" if this helps and is sufficiently
1643 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1644 * do not trap on the (extremely unlikely) case of a page
1645 * crossing operation.
1647 * - Furthermore, we need an efficient 64-bit compile for the
1648 * 64-bit case in order to generate the "number of bytes in
1649 * the final mask". Again, that could be replaced with a
1650 * efficient population count instruction or similar.
1652 #ifdef CONFIG_DCACHE_WORD_ACCESS
1654 #include <asm/word-at-a-time.h>
1658 static inline unsigned int fold_hash(unsigned long hash
)
1660 return hash_64(hash
, 32);
1663 #else /* 32-bit case */
1665 #define fold_hash(x) (x)
1669 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1671 unsigned long a
, mask
;
1672 unsigned long hash
= 0;
1675 a
= load_unaligned_zeropad(name
);
1676 if (len
< sizeof(unsigned long))
1680 name
+= sizeof(unsigned long);
1681 len
-= sizeof(unsigned long);
1685 mask
= bytemask_from_count(len
);
1688 return fold_hash(hash
);
1690 EXPORT_SYMBOL(full_name_hash
);
1693 * Calculate the length and hash of the path component, and
1694 * return the "hash_len" as the result.
1696 static inline u64
hash_name(const char *name
)
1698 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1699 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1702 len
= -sizeof(unsigned long);
1704 hash
= (hash
+ a
) * 9;
1705 len
+= sizeof(unsigned long);
1706 a
= load_unaligned_zeropad(name
+len
);
1707 b
= a
^ REPEAT_BYTE('/');
1708 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1710 adata
= prep_zero_mask(a
, adata
, &constants
);
1711 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1713 mask
= create_zero_mask(adata
| bdata
);
1715 hash
+= a
& zero_bytemask(mask
);
1716 len
+= find_zero(mask
);
1717 return hashlen_create(fold_hash(hash
), len
);
1722 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1724 unsigned long hash
= init_name_hash();
1726 hash
= partial_name_hash(*name
++, hash
);
1727 return end_name_hash(hash
);
1729 EXPORT_SYMBOL(full_name_hash
);
1732 * We know there's a real path component here of at least
1735 static inline u64
hash_name(const char *name
)
1737 unsigned long hash
= init_name_hash();
1738 unsigned long len
= 0, c
;
1740 c
= (unsigned char)*name
;
1743 hash
= partial_name_hash(c
, hash
);
1744 c
= (unsigned char)name
[len
];
1745 } while (c
&& c
!= '/');
1746 return hashlen_create(end_name_hash(hash
), len
);
1753 * This is the basic name resolution function, turning a pathname into
1754 * the final dentry. We expect 'base' to be positive and a directory.
1756 * Returns 0 and nd will have valid dentry and mnt on success.
1757 * Returns error and drops reference to input namei data on failure.
1759 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1768 /* At this point we know we have a real path component. */
1773 err
= may_lookup(nd
);
1777 hash_len
= hash_name(name
);
1780 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1782 if (name
[1] == '.') {
1784 nd
->flags
|= LOOKUP_JUMPED
;
1790 if (likely(type
== LAST_NORM
)) {
1791 struct dentry
*parent
= nd
->path
.dentry
;
1792 nd
->flags
&= ~LOOKUP_JUMPED
;
1793 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1794 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1795 err
= parent
->d_op
->d_hash(parent
, &this);
1798 hash_len
= this.hash_len
;
1803 nd
->last
.hash_len
= hash_len
;
1804 nd
->last
.name
= name
;
1805 nd
->last_type
= type
;
1807 name
+= hashlen_len(hash_len
);
1811 * If it wasn't NUL, we know it was '/'. Skip that
1812 * slash, and continue until no more slashes.
1816 } while (unlikely(*name
== '/'));
1817 if (unlikely(!*name
)) {
1819 /* pathname body, done */
1822 name
= nd
->stack
[nd
->depth
- 1].name
;
1823 /* trailing symlink, done */
1826 /* last component of nested symlink */
1827 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
1829 err
= walk_component(nd
, WALK_GET
);
1835 const char *s
= get_link(nd
);
1837 if (unlikely(IS_ERR(s
))) {
1846 nd
->stack
[nd
->depth
- 1].name
= name
;
1851 if (!d_can_lookup(nd
->path
.dentry
)) {
1859 static const char *path_init(int dfd
, const struct filename
*name
,
1860 unsigned int flags
, struct nameidata
*nd
)
1863 const char *s
= name
->name
;
1865 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1866 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1868 nd
->total_link_count
= 0;
1869 if (flags
& LOOKUP_ROOT
) {
1870 struct dentry
*root
= nd
->root
.dentry
;
1871 struct inode
*inode
= root
->d_inode
;
1873 if (!d_can_lookup(root
))
1874 return ERR_PTR(-ENOTDIR
);
1875 retval
= inode_permission(inode
, MAY_EXEC
);
1877 return ERR_PTR(retval
);
1879 nd
->path
= nd
->root
;
1881 if (flags
& LOOKUP_RCU
) {
1883 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1884 nd
->m_seq
= read_seqbegin(&mount_lock
);
1886 path_get(&nd
->path
);
1891 nd
->root
.mnt
= NULL
;
1893 nd
->m_seq
= read_seqbegin(&mount_lock
);
1895 if (flags
& LOOKUP_RCU
) {
1897 nd
->seq
= set_root_rcu(nd
);
1900 path_get(&nd
->root
);
1902 nd
->path
= nd
->root
;
1903 } else if (dfd
== AT_FDCWD
) {
1904 if (flags
& LOOKUP_RCU
) {
1905 struct fs_struct
*fs
= current
->fs
;
1911 seq
= read_seqcount_begin(&fs
->seq
);
1913 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1914 } while (read_seqcount_retry(&fs
->seq
, seq
));
1916 get_fs_pwd(current
->fs
, &nd
->path
);
1919 /* Caller must check execute permissions on the starting path component */
1920 struct fd f
= fdget_raw(dfd
);
1921 struct dentry
*dentry
;
1924 return ERR_PTR(-EBADF
);
1926 dentry
= f
.file
->f_path
.dentry
;
1929 if (!d_can_lookup(dentry
)) {
1931 return ERR_PTR(-ENOTDIR
);
1935 nd
->path
= f
.file
->f_path
;
1936 if (flags
& LOOKUP_RCU
) {
1938 nd
->inode
= nd
->path
.dentry
->d_inode
;
1939 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1941 path_get(&nd
->path
);
1942 nd
->inode
= nd
->path
.dentry
->d_inode
;
1948 nd
->inode
= nd
->path
.dentry
->d_inode
;
1949 if (!(flags
& LOOKUP_RCU
))
1951 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1953 if (!(nd
->flags
& LOOKUP_ROOT
))
1954 nd
->root
.mnt
= NULL
;
1956 return ERR_PTR(-ECHILD
);
1959 static void path_cleanup(struct nameidata
*nd
)
1961 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1962 path_put(&nd
->root
);
1963 nd
->root
.mnt
= NULL
;
1967 static const char *trailing_symlink(struct nameidata
*nd
)
1970 int error
= may_follow_link(nd
);
1971 if (unlikely(error
))
1972 return ERR_PTR(error
);
1973 nd
->flags
|= LOOKUP_PARENT
;
1974 nd
->stack
[0].name
= NULL
;
1979 static inline int lookup_last(struct nameidata
*nd
)
1981 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1982 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1984 nd
->flags
&= ~LOOKUP_PARENT
;
1985 return walk_component(nd
,
1986 nd
->flags
& LOOKUP_FOLLOW
1988 ? WALK_PUT
| WALK_GET
1993 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1994 static int path_lookupat(int dfd
, const struct filename
*name
,
1995 unsigned int flags
, struct nameidata
*nd
)
1997 const char *s
= path_init(dfd
, name
, flags
, nd
);
2002 while (!(err
= link_path_walk(s
, nd
))
2003 && ((err
= lookup_last(nd
)) > 0)) {
2004 s
= trailing_symlink(nd
);
2011 err
= complete_walk(nd
);
2013 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2014 if (!d_can_lookup(nd
->path
.dentry
))
2023 static int filename_lookup(int dfd
, struct filename
*name
,
2024 unsigned int flags
, struct nameidata
*nd
)
2027 struct nameidata
*saved_nd
= set_nameidata(nd
);
2029 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2030 if (unlikely(retval
== -ECHILD
))
2031 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2032 if (unlikely(retval
== -ESTALE
))
2033 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2035 if (likely(!retval
))
2036 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2037 restore_nameidata(saved_nd
);
2041 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2042 static int path_parentat(int dfd
, const struct filename
*name
,
2043 unsigned int flags
, struct nameidata
*nd
)
2045 const char *s
= path_init(dfd
, name
, flags
, nd
);
2049 err
= link_path_walk(s
, nd
);
2051 err
= complete_walk(nd
);
2058 static int filename_parentat(int dfd
, struct filename
*name
,
2059 unsigned int flags
, struct nameidata
*nd
)
2062 struct nameidata
*saved_nd
= set_nameidata(nd
);
2064 retval
= path_parentat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2065 if (unlikely(retval
== -ECHILD
))
2066 retval
= path_parentat(dfd
, name
, flags
, nd
);
2067 if (unlikely(retval
== -ESTALE
))
2068 retval
= path_parentat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2070 if (likely(!retval
))
2071 audit_inode(name
, nd
->path
.dentry
, LOOKUP_PARENT
);
2072 restore_nameidata(saved_nd
);
2076 /* does lookup, returns the object with parent locked */
2077 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2079 struct filename
*filename
= getname_kernel(name
);
2080 struct nameidata nd
;
2084 if (IS_ERR(filename
))
2085 return ERR_CAST(filename
);
2087 err
= filename_parentat(AT_FDCWD
, filename
, 0, &nd
);
2092 if (nd
.last_type
!= LAST_NORM
) {
2094 d
= ERR_PTR(-EINVAL
);
2097 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2098 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2100 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2110 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2112 struct nameidata nd
;
2113 struct filename
*filename
= getname_kernel(name
);
2114 int res
= PTR_ERR(filename
);
2116 if (!IS_ERR(filename
)) {
2117 res
= filename_lookup(AT_FDCWD
, filename
, flags
, &nd
);
2124 EXPORT_SYMBOL(kern_path
);
2127 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2128 * @dentry: pointer to dentry of the base directory
2129 * @mnt: pointer to vfs mount of the base directory
2130 * @name: pointer to file name
2131 * @flags: lookup flags
2132 * @path: pointer to struct path to fill
2134 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2135 const char *name
, unsigned int flags
,
2138 struct filename
*filename
= getname_kernel(name
);
2139 int err
= PTR_ERR(filename
);
2141 BUG_ON(flags
& LOOKUP_PARENT
);
2143 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2144 if (!IS_ERR(filename
)) {
2145 struct nameidata nd
;
2146 nd
.root
.dentry
= dentry
;
2148 err
= filename_lookup(AT_FDCWD
, filename
,
2149 flags
| LOOKUP_ROOT
, &nd
);
2156 EXPORT_SYMBOL(vfs_path_lookup
);
2159 * lookup_one_len - filesystem helper to lookup single pathname component
2160 * @name: pathname component to lookup
2161 * @base: base directory to lookup from
2162 * @len: maximum length @len should be interpreted to
2164 * Note that this routine is purely a helper for filesystem usage and should
2165 * not be called by generic code.
2167 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2173 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2177 this.hash
= full_name_hash(name
, len
);
2179 return ERR_PTR(-EACCES
);
2181 if (unlikely(name
[0] == '.')) {
2182 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2183 return ERR_PTR(-EACCES
);
2187 c
= *(const unsigned char *)name
++;
2188 if (c
== '/' || c
== '\0')
2189 return ERR_PTR(-EACCES
);
2192 * See if the low-level filesystem might want
2193 * to use its own hash..
2195 if (base
->d_flags
& DCACHE_OP_HASH
) {
2196 int err
= base
->d_op
->d_hash(base
, &this);
2198 return ERR_PTR(err
);
2201 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2203 return ERR_PTR(err
);
2205 return __lookup_hash(&this, base
, 0);
2207 EXPORT_SYMBOL(lookup_one_len
);
2209 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2210 struct path
*path
, int *empty
)
2212 struct nameidata nd
;
2213 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2214 int err
= PTR_ERR(tmp
);
2217 BUG_ON(flags
& LOOKUP_PARENT
);
2219 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2227 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2230 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2232 EXPORT_SYMBOL(user_path_at
);
2235 * NB: most callers don't do anything directly with the reference to the
2236 * to struct filename, but the nd->last pointer points into the name string
2237 * allocated by getname. So we must hold the reference to it until all
2238 * path-walking is complete.
2240 static struct filename
*
2241 user_path_parent(int dfd
, const char __user
*path
,
2242 struct path
*parent
,
2247 struct nameidata nd
;
2248 struct filename
*s
= getname(path
);
2251 /* only LOOKUP_REVAL is allowed in extra flags */
2252 flags
&= LOOKUP_REVAL
;
2257 error
= filename_parentat(dfd
, s
, flags
, &nd
);
2260 return ERR_PTR(error
);
2264 *type
= nd
.last_type
;
2270 * mountpoint_last - look up last component for umount
2271 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2272 * @path: pointer to container for result
2274 * This is a special lookup_last function just for umount. In this case, we
2275 * need to resolve the path without doing any revalidation.
2277 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2278 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2279 * in almost all cases, this lookup will be served out of the dcache. The only
2280 * cases where it won't are if nd->last refers to a symlink or the path is
2281 * bogus and it doesn't exist.
2284 * -error: if there was an error during lookup. This includes -ENOENT if the
2285 * lookup found a negative dentry. The nd->path reference will also be
2288 * 0: if we successfully resolved nd->path and found it to not to be a
2289 * symlink that needs to be followed. "path" will also be populated.
2290 * The nd->path reference will also be put.
2292 * 1: if we successfully resolved nd->last and found it to be a symlink
2293 * that needs to be followed. "path" will be populated with the path
2294 * to the link, and nd->path will *not* be put.
2297 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2300 struct dentry
*dentry
;
2301 struct dentry
*dir
= nd
->path
.dentry
;
2303 /* If we're in rcuwalk, drop out of it to handle last component */
2304 if (nd
->flags
& LOOKUP_RCU
) {
2305 if (unlazy_walk(nd
, NULL
))
2309 nd
->flags
&= ~LOOKUP_PARENT
;
2311 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2312 error
= handle_dots(nd
, nd
->last_type
);
2315 dentry
= dget(nd
->path
.dentry
);
2319 mutex_lock(&dir
->d_inode
->i_mutex
);
2320 dentry
= d_lookup(dir
, &nd
->last
);
2323 * No cached dentry. Mounted dentries are pinned in the cache,
2324 * so that means that this dentry is probably a symlink or the
2325 * path doesn't actually point to a mounted dentry.
2327 dentry
= d_alloc(dir
, &nd
->last
);
2329 mutex_unlock(&dir
->d_inode
->i_mutex
);
2332 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2333 if (IS_ERR(dentry
)) {
2334 mutex_unlock(&dir
->d_inode
->i_mutex
);
2335 return PTR_ERR(dentry
);
2338 mutex_unlock(&dir
->d_inode
->i_mutex
);
2341 if (d_is_negative(dentry
)) {
2347 path
->dentry
= dentry
;
2348 path
->mnt
= nd
->path
.mnt
;
2349 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
2350 if (unlikely(error
))
2358 * path_mountpoint - look up a path to be umounted
2359 * @dfd: directory file descriptor to start walk from
2360 * @name: full pathname to walk
2361 * @path: pointer to container for result
2362 * @flags: lookup flags
2364 * Look up the given name, but don't attempt to revalidate the last component.
2365 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2368 path_mountpoint(int dfd
, const struct filename
*name
, struct path
*path
,
2369 struct nameidata
*nd
, unsigned int flags
)
2371 const char *s
= path_init(dfd
, name
, flags
, nd
);
2375 while (!(err
= link_path_walk(s
, nd
)) &&
2376 (err
= mountpoint_last(nd
, path
)) > 0) {
2377 s
= trailing_symlink(nd
);
2389 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2392 struct nameidata nd
, *saved
;
2395 return PTR_ERR(name
);
2396 saved
= set_nameidata(&nd
);
2397 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_RCU
);
2398 if (unlikely(error
== -ECHILD
))
2399 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
);
2400 if (unlikely(error
== -ESTALE
))
2401 error
= path_mountpoint(dfd
, name
, path
, &nd
, flags
| LOOKUP_REVAL
);
2403 audit_inode(name
, path
->dentry
, 0);
2404 restore_nameidata(saved
);
2410 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2411 * @dfd: directory file descriptor
2412 * @name: pathname from userland
2413 * @flags: lookup flags
2414 * @path: pointer to container to hold result
2416 * A umount is a special case for path walking. We're not actually interested
2417 * in the inode in this situation, and ESTALE errors can be a problem. We
2418 * simply want track down the dentry and vfsmount attached at the mountpoint
2419 * and avoid revalidating the last component.
2421 * Returns 0 and populates "path" on success.
2424 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2427 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2431 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2434 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2436 EXPORT_SYMBOL(kern_path_mountpoint
);
2438 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2440 kuid_t fsuid
= current_fsuid();
2442 if (uid_eq(inode
->i_uid
, fsuid
))
2444 if (uid_eq(dir
->i_uid
, fsuid
))
2446 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2448 EXPORT_SYMBOL(__check_sticky
);
2451 * Check whether we can remove a link victim from directory dir, check
2452 * whether the type of victim is right.
2453 * 1. We can't do it if dir is read-only (done in permission())
2454 * 2. We should have write and exec permissions on dir
2455 * 3. We can't remove anything from append-only dir
2456 * 4. We can't do anything with immutable dir (done in permission())
2457 * 5. If the sticky bit on dir is set we should either
2458 * a. be owner of dir, or
2459 * b. be owner of victim, or
2460 * c. have CAP_FOWNER capability
2461 * 6. If the victim is append-only or immutable we can't do antyhing with
2462 * links pointing to it.
2463 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2464 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2465 * 9. We can't remove a root or mountpoint.
2466 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2467 * nfs_async_unlink().
2469 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2471 struct inode
*inode
= victim
->d_inode
;
2474 if (d_is_negative(victim
))
2478 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2479 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2481 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2487 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2488 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2491 if (!d_is_dir(victim
))
2493 if (IS_ROOT(victim
))
2495 } else if (d_is_dir(victim
))
2497 if (IS_DEADDIR(dir
))
2499 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2504 /* Check whether we can create an object with dentry child in directory
2506 * 1. We can't do it if child already exists (open has special treatment for
2507 * this case, but since we are inlined it's OK)
2508 * 2. We can't do it if dir is read-only (done in permission())
2509 * 3. We should have write and exec permissions on dir
2510 * 4. We can't do it if dir is immutable (done in permission())
2512 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2514 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2517 if (IS_DEADDIR(dir
))
2519 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2523 * p1 and p2 should be directories on the same fs.
2525 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2530 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2534 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2536 p
= d_ancestor(p2
, p1
);
2538 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2539 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2543 p
= d_ancestor(p1
, p2
);
2545 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2546 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2550 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2551 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2554 EXPORT_SYMBOL(lock_rename
);
2556 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2558 mutex_unlock(&p1
->d_inode
->i_mutex
);
2560 mutex_unlock(&p2
->d_inode
->i_mutex
);
2561 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2564 EXPORT_SYMBOL(unlock_rename
);
2566 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2569 int error
= may_create(dir
, dentry
);
2573 if (!dir
->i_op
->create
)
2574 return -EACCES
; /* shouldn't it be ENOSYS? */
2577 error
= security_inode_create(dir
, dentry
, mode
);
2580 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2582 fsnotify_create(dir
, dentry
);
2585 EXPORT_SYMBOL(vfs_create
);
2587 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2589 struct dentry
*dentry
= path
->dentry
;
2590 struct inode
*inode
= dentry
->d_inode
;
2600 switch (inode
->i_mode
& S_IFMT
) {
2604 if (acc_mode
& MAY_WRITE
)
2609 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2618 error
= inode_permission(inode
, acc_mode
);
2623 * An append-only file must be opened in append mode for writing.
2625 if (IS_APPEND(inode
)) {
2626 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2632 /* O_NOATIME can only be set by the owner or superuser */
2633 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2639 static int handle_truncate(struct file
*filp
)
2641 struct path
*path
= &filp
->f_path
;
2642 struct inode
*inode
= path
->dentry
->d_inode
;
2643 int error
= get_write_access(inode
);
2647 * Refuse to truncate files with mandatory locks held on them.
2649 error
= locks_verify_locked(filp
);
2651 error
= security_path_truncate(path
);
2653 error
= do_truncate(path
->dentry
, 0,
2654 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2657 put_write_access(inode
);
2661 static inline int open_to_namei_flags(int flag
)
2663 if ((flag
& O_ACCMODE
) == 3)
2668 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2670 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2674 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2678 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2682 * Attempt to atomically look up, create and open a file from a negative
2685 * Returns 0 if successful. The file will have been created and attached to
2686 * @file by the filesystem calling finish_open().
2688 * Returns 1 if the file was looked up only or didn't need creating. The
2689 * caller will need to perform the open themselves. @path will have been
2690 * updated to point to the new dentry. This may be negative.
2692 * Returns an error code otherwise.
2694 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2695 struct path
*path
, struct file
*file
,
2696 const struct open_flags
*op
,
2697 bool got_write
, bool need_lookup
,
2700 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2701 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2705 int create_error
= 0;
2706 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2709 BUG_ON(dentry
->d_inode
);
2711 /* Don't create child dentry for a dead directory. */
2712 if (unlikely(IS_DEADDIR(dir
))) {
2718 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2719 mode
&= ~current_umask();
2721 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2723 open_flag
&= ~O_TRUNC
;
2726 * Checking write permission is tricky, bacuse we don't know if we are
2727 * going to actually need it: O_CREAT opens should work as long as the
2728 * file exists. But checking existence breaks atomicity. The trick is
2729 * to check access and if not granted clear O_CREAT from the flags.
2731 * Another problem is returing the "right" error value (e.g. for an
2732 * O_EXCL open we want to return EEXIST not EROFS).
2734 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2735 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2736 if (!(open_flag
& O_CREAT
)) {
2738 * No O_CREATE -> atomicity not a requirement -> fall
2739 * back to lookup + open
2742 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2743 /* Fall back and fail with the right error */
2744 create_error
= -EROFS
;
2747 /* No side effects, safe to clear O_CREAT */
2748 create_error
= -EROFS
;
2749 open_flag
&= ~O_CREAT
;
2753 if (open_flag
& O_CREAT
) {
2754 error
= may_o_create(&nd
->path
, dentry
, mode
);
2756 create_error
= error
;
2757 if (open_flag
& O_EXCL
)
2759 open_flag
&= ~O_CREAT
;
2763 if (nd
->flags
& LOOKUP_DIRECTORY
)
2764 open_flag
|= O_DIRECTORY
;
2766 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2767 file
->f_path
.mnt
= nd
->path
.mnt
;
2768 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2771 if (create_error
&& error
== -ENOENT
)
2772 error
= create_error
;
2776 if (error
) { /* returned 1, that is */
2777 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2781 if (file
->f_path
.dentry
) {
2783 dentry
= file
->f_path
.dentry
;
2785 if (*opened
& FILE_CREATED
)
2786 fsnotify_create(dir
, dentry
);
2787 if (!dentry
->d_inode
) {
2788 WARN_ON(*opened
& FILE_CREATED
);
2790 error
= create_error
;
2794 if (excl
&& !(*opened
& FILE_CREATED
)) {
2803 * We didn't have the inode before the open, so check open permission
2806 acc_mode
= op
->acc_mode
;
2807 if (*opened
& FILE_CREATED
) {
2808 WARN_ON(!(open_flag
& O_CREAT
));
2809 fsnotify_create(dir
, dentry
);
2810 acc_mode
= MAY_OPEN
;
2812 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2822 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2824 return PTR_ERR(dentry
);
2827 int open_flag
= op
->open_flag
;
2829 error
= create_error
;
2830 if ((open_flag
& O_EXCL
)) {
2831 if (!dentry
->d_inode
)
2833 } else if (!dentry
->d_inode
) {
2835 } else if ((open_flag
& O_TRUNC
) &&
2839 /* will fail later, go on to get the right error */
2843 path
->dentry
= dentry
;
2844 path
->mnt
= nd
->path
.mnt
;
2849 * Look up and maybe create and open the last component.
2851 * Must be called with i_mutex held on parent.
2853 * Returns 0 if the file was successfully atomically created (if necessary) and
2854 * opened. In this case the file will be returned attached to @file.
2856 * Returns 1 if the file was not completely opened at this time, though lookups
2857 * and creations will have been performed and the dentry returned in @path will
2858 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2859 * specified then a negative dentry may be returned.
2861 * An error code is returned otherwise.
2863 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2864 * cleared otherwise prior to returning.
2866 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2868 const struct open_flags
*op
,
2869 bool got_write
, int *opened
)
2871 struct dentry
*dir
= nd
->path
.dentry
;
2872 struct inode
*dir_inode
= dir
->d_inode
;
2873 struct dentry
*dentry
;
2877 *opened
&= ~FILE_CREATED
;
2878 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2880 return PTR_ERR(dentry
);
2882 /* Cached positive dentry: will open in f_op->open */
2883 if (!need_lookup
&& dentry
->d_inode
)
2886 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2887 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2888 need_lookup
, opened
);
2892 BUG_ON(dentry
->d_inode
);
2894 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2896 return PTR_ERR(dentry
);
2899 /* Negative dentry, just create the file */
2900 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2901 umode_t mode
= op
->mode
;
2902 if (!IS_POSIXACL(dir
->d_inode
))
2903 mode
&= ~current_umask();
2905 * This write is needed to ensure that a
2906 * rw->ro transition does not occur between
2907 * the time when the file is created and when
2908 * a permanent write count is taken through
2909 * the 'struct file' in finish_open().
2915 *opened
|= FILE_CREATED
;
2916 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2919 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2920 nd
->flags
& LOOKUP_EXCL
);
2925 path
->dentry
= dentry
;
2926 path
->mnt
= nd
->path
.mnt
;
2935 * Handle the last step of open()
2937 static int do_last(struct nameidata
*nd
,
2938 struct file
*file
, const struct open_flags
*op
,
2939 int *opened
, struct filename
*name
)
2941 struct dentry
*dir
= nd
->path
.dentry
;
2942 int open_flag
= op
->open_flag
;
2943 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2944 bool got_write
= false;
2945 int acc_mode
= op
->acc_mode
;
2946 struct inode
*inode
;
2947 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2949 bool retried
= false;
2952 nd
->flags
&= ~LOOKUP_PARENT
;
2953 nd
->flags
|= op
->intent
;
2955 if (nd
->last_type
!= LAST_NORM
) {
2956 error
= handle_dots(nd
, nd
->last_type
);
2957 if (unlikely(error
))
2962 if (!(open_flag
& O_CREAT
)) {
2963 if (nd
->last
.name
[nd
->last
.len
])
2964 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2965 /* we _can_ be in RCU mode here */
2966 error
= lookup_fast(nd
, &path
, &inode
);
2973 BUG_ON(nd
->inode
!= dir
->d_inode
);
2975 /* create side of things */
2977 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2978 * has been cleared when we got to the last component we are
2981 error
= complete_walk(nd
);
2985 audit_inode(name
, dir
, LOOKUP_PARENT
);
2986 /* trailing slashes? */
2987 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
2992 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2993 error
= mnt_want_write(nd
->path
.mnt
);
2997 * do _not_ fail yet - we might not need that or fail with
2998 * a different error; let lookup_open() decide; we'll be
2999 * dropping this one anyway.
3002 mutex_lock(&dir
->d_inode
->i_mutex
);
3003 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3004 mutex_unlock(&dir
->d_inode
->i_mutex
);
3010 if ((*opened
& FILE_CREATED
) ||
3011 !S_ISREG(file_inode(file
)->i_mode
))
3012 will_truncate
= false;
3014 audit_inode(name
, file
->f_path
.dentry
, 0);
3018 if (*opened
& FILE_CREATED
) {
3019 /* Don't check for write permission, don't truncate */
3020 open_flag
&= ~O_TRUNC
;
3021 will_truncate
= false;
3022 acc_mode
= MAY_OPEN
;
3023 path_to_nameidata(&path
, nd
);
3024 goto finish_open_created
;
3028 * create/update audit record if it already exists.
3030 if (d_is_positive(path
.dentry
))
3031 audit_inode(name
, path
.dentry
, 0);
3034 * If atomic_open() acquired write access it is dropped now due to
3035 * possible mount and symlink following (this might be optimized away if
3039 mnt_drop_write(nd
->path
.mnt
);
3043 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3044 path_to_nameidata(&path
, nd
);
3048 error
= follow_managed(&path
, nd
);
3049 if (unlikely(error
< 0))
3052 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3053 inode
= path
.dentry
->d_inode
;
3054 if (unlikely(d_is_negative(path
.dentry
))) {
3055 path_to_nameidata(&path
, nd
);
3061 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
);
3062 if (unlikely(error
))
3065 if (unlikely(d_is_symlink(path
.dentry
)) && !(open_flag
& O_PATH
)) {
3066 path_to_nameidata(&path
, nd
);
3070 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3071 path_to_nameidata(&path
, nd
);
3073 save_parent
.dentry
= nd
->path
.dentry
;
3074 save_parent
.mnt
= mntget(path
.mnt
);
3075 nd
->path
.dentry
= path
.dentry
;
3079 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3081 error
= complete_walk(nd
);
3083 path_put(&save_parent
);
3086 audit_inode(name
, nd
->path
.dentry
, 0);
3088 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3091 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3093 if (!d_is_reg(nd
->path
.dentry
))
3094 will_truncate
= false;
3096 if (will_truncate
) {
3097 error
= mnt_want_write(nd
->path
.mnt
);
3102 finish_open_created
:
3103 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3107 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3108 error
= vfs_open(&nd
->path
, file
, current_cred());
3110 *opened
|= FILE_OPENED
;
3112 if (error
== -EOPENSTALE
)
3117 error
= open_check_o_direct(file
);
3120 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3124 if (will_truncate
) {
3125 error
= handle_truncate(file
);
3131 mnt_drop_write(nd
->path
.mnt
);
3132 path_put(&save_parent
);
3140 /* If no saved parent or already retried then can't retry */
3141 if (!save_parent
.dentry
|| retried
)
3144 BUG_ON(save_parent
.dentry
!= dir
);
3145 path_put(&nd
->path
);
3146 nd
->path
= save_parent
;
3147 nd
->inode
= dir
->d_inode
;
3148 save_parent
.mnt
= NULL
;
3149 save_parent
.dentry
= NULL
;
3151 mnt_drop_write(nd
->path
.mnt
);
3158 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3159 struct nameidata
*nd
, int flags
,
3160 const struct open_flags
*op
,
3161 struct file
*file
, int *opened
)
3163 static const struct qstr name
= QSTR_INIT("/", 1);
3164 struct dentry
*dentry
, *child
;
3166 int error
= path_lookupat(dfd
, pathname
,
3167 flags
| LOOKUP_DIRECTORY
, nd
);
3168 if (unlikely(error
))
3170 error
= mnt_want_write(nd
->path
.mnt
);
3171 if (unlikely(error
))
3173 /* we want directory to be writable */
3174 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3177 dentry
= nd
->path
.dentry
;
3178 dir
= dentry
->d_inode
;
3179 if (!dir
->i_op
->tmpfile
) {
3180 error
= -EOPNOTSUPP
;
3183 child
= d_alloc(dentry
, &name
);
3184 if (unlikely(!child
)) {
3188 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3189 nd
->flags
|= op
->intent
;
3190 dput(nd
->path
.dentry
);
3191 nd
->path
.dentry
= child
;
3192 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3195 audit_inode(pathname
, nd
->path
.dentry
, 0);
3196 /* Don't check for other permissions, the inode was just created */
3197 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3200 file
->f_path
.mnt
= nd
->path
.mnt
;
3201 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3204 error
= open_check_o_direct(file
);
3207 } else if (!(op
->open_flag
& O_EXCL
)) {
3208 struct inode
*inode
= file_inode(file
);
3209 spin_lock(&inode
->i_lock
);
3210 inode
->i_state
|= I_LINKABLE
;
3211 spin_unlock(&inode
->i_lock
);
3214 mnt_drop_write(nd
->path
.mnt
);
3216 path_put(&nd
->path
);
3220 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3221 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3228 file
= get_empty_filp();
3232 file
->f_flags
= op
->open_flag
;
3234 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3235 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3239 s
= path_init(dfd
, pathname
, flags
, nd
);
3244 while (!(error
= link_path_walk(s
, nd
)) &&
3245 (error
= do_last(nd
, file
, op
, &opened
, pathname
)) > 0) {
3246 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3247 s
= trailing_symlink(nd
);
3256 if (!(opened
& FILE_OPENED
)) {
3260 if (unlikely(error
)) {
3261 if (error
== -EOPENSTALE
) {
3262 if (flags
& LOOKUP_RCU
)
3267 file
= ERR_PTR(error
);
3272 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3273 const struct open_flags
*op
)
3275 struct nameidata nd
, *saved_nd
= set_nameidata(&nd
);
3276 int flags
= op
->lookup_flags
;
3279 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3280 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3281 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3282 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3283 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3284 restore_nameidata(saved_nd
);
3288 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3289 const char *name
, const struct open_flags
*op
)
3291 struct nameidata nd
, *saved_nd
;
3293 struct filename
*filename
;
3294 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3297 nd
.root
.dentry
= dentry
;
3299 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3300 return ERR_PTR(-ELOOP
);
3302 filename
= getname_kernel(name
);
3303 if (unlikely(IS_ERR(filename
)))
3304 return ERR_CAST(filename
);
3306 saved_nd
= set_nameidata(&nd
);
3307 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3308 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3309 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3310 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3311 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3312 restore_nameidata(saved_nd
);
3317 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3318 struct path
*path
, unsigned int lookup_flags
)
3320 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3321 struct nameidata nd
;
3324 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3327 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3328 * other flags passed in are ignored!
3330 lookup_flags
&= LOOKUP_REVAL
;
3332 error
= filename_parentat(dfd
, name
, lookup_flags
, &nd
);
3334 return ERR_PTR(error
);
3337 * Yucky last component or no last component at all?
3338 * (foo/., foo/.., /////)
3340 if (nd
.last_type
!= LAST_NORM
)
3342 nd
.flags
&= ~LOOKUP_PARENT
;
3343 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3345 /* don't fail immediately if it's r/o, at least try to report other errors */
3346 err2
= mnt_want_write(nd
.path
.mnt
);
3348 * Do the final lookup.
3350 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3351 dentry
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, nd
.flags
);
3356 if (d_is_positive(dentry
))
3360 * Special case - lookup gave negative, but... we had foo/bar/
3361 * From the vfs_mknod() POV we just have a negative dentry -
3362 * all is fine. Let's be bastards - you had / on the end, you've
3363 * been asking for (non-existent) directory. -ENOENT for you.
3365 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3369 if (unlikely(err2
)) {
3377 dentry
= ERR_PTR(error
);
3379 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3381 mnt_drop_write(nd
.path
.mnt
);
3387 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3388 struct path
*path
, unsigned int lookup_flags
)
3390 struct filename
*filename
= getname_kernel(pathname
);
3393 if (IS_ERR(filename
))
3394 return ERR_CAST(filename
);
3395 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3399 EXPORT_SYMBOL(kern_path_create
);
3401 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3404 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3405 mnt_drop_write(path
->mnt
);
3408 EXPORT_SYMBOL(done_path_create
);
3410 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3411 struct path
*path
, unsigned int lookup_flags
)
3413 struct filename
*tmp
= getname(pathname
);
3416 return ERR_CAST(tmp
);
3417 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3421 EXPORT_SYMBOL(user_path_create
);
3423 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3425 int error
= may_create(dir
, dentry
);
3430 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3433 if (!dir
->i_op
->mknod
)
3436 error
= devcgroup_inode_mknod(mode
, dev
);
3440 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3444 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3446 fsnotify_create(dir
, dentry
);
3449 EXPORT_SYMBOL(vfs_mknod
);
3451 static int may_mknod(umode_t mode
)
3453 switch (mode
& S_IFMT
) {
3459 case 0: /* zero mode translates to S_IFREG */
3468 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3471 struct dentry
*dentry
;
3474 unsigned int lookup_flags
= 0;
3476 error
= may_mknod(mode
);
3480 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3482 return PTR_ERR(dentry
);
3484 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3485 mode
&= ~current_umask();
3486 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3489 switch (mode
& S_IFMT
) {
3490 case 0: case S_IFREG
:
3491 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3493 case S_IFCHR
: case S_IFBLK
:
3494 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3495 new_decode_dev(dev
));
3497 case S_IFIFO
: case S_IFSOCK
:
3498 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3502 done_path_create(&path
, dentry
);
3503 if (retry_estale(error
, lookup_flags
)) {
3504 lookup_flags
|= LOOKUP_REVAL
;
3510 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3512 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3515 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3517 int error
= may_create(dir
, dentry
);
3518 unsigned max_links
= dir
->i_sb
->s_max_links
;
3523 if (!dir
->i_op
->mkdir
)
3526 mode
&= (S_IRWXUGO
|S_ISVTX
);
3527 error
= security_inode_mkdir(dir
, dentry
, mode
);
3531 if (max_links
&& dir
->i_nlink
>= max_links
)
3534 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3536 fsnotify_mkdir(dir
, dentry
);
3539 EXPORT_SYMBOL(vfs_mkdir
);
3541 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3543 struct dentry
*dentry
;
3546 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3549 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3551 return PTR_ERR(dentry
);
3553 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3554 mode
&= ~current_umask();
3555 error
= security_path_mkdir(&path
, dentry
, mode
);
3557 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3558 done_path_create(&path
, dentry
);
3559 if (retry_estale(error
, lookup_flags
)) {
3560 lookup_flags
|= LOOKUP_REVAL
;
3566 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3568 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3572 * The dentry_unhash() helper will try to drop the dentry early: we
3573 * should have a usage count of 1 if we're the only user of this
3574 * dentry, and if that is true (possibly after pruning the dcache),
3575 * then we drop the dentry now.
3577 * A low-level filesystem can, if it choses, legally
3580 * if (!d_unhashed(dentry))
3583 * if it cannot handle the case of removing a directory
3584 * that is still in use by something else..
3586 void dentry_unhash(struct dentry
*dentry
)
3588 shrink_dcache_parent(dentry
);
3589 spin_lock(&dentry
->d_lock
);
3590 if (dentry
->d_lockref
.count
== 1)
3592 spin_unlock(&dentry
->d_lock
);
3594 EXPORT_SYMBOL(dentry_unhash
);
3596 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3598 int error
= may_delete(dir
, dentry
, 1);
3603 if (!dir
->i_op
->rmdir
)
3607 mutex_lock(&dentry
->d_inode
->i_mutex
);
3610 if (is_local_mountpoint(dentry
))
3613 error
= security_inode_rmdir(dir
, dentry
);
3617 shrink_dcache_parent(dentry
);
3618 error
= dir
->i_op
->rmdir(dir
, dentry
);
3622 dentry
->d_inode
->i_flags
|= S_DEAD
;
3624 detach_mounts(dentry
);
3627 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3633 EXPORT_SYMBOL(vfs_rmdir
);
3635 static long do_rmdir(int dfd
, const char __user
*pathname
)
3638 struct filename
*name
;
3639 struct dentry
*dentry
;
3643 unsigned int lookup_flags
= 0;
3645 name
= user_path_parent(dfd
, pathname
,
3646 &path
, &last
, &type
, lookup_flags
);
3648 return PTR_ERR(name
);
3662 error
= mnt_want_write(path
.mnt
);
3666 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3667 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3668 error
= PTR_ERR(dentry
);
3671 if (!dentry
->d_inode
) {
3675 error
= security_path_rmdir(&path
, dentry
);
3678 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3682 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3683 mnt_drop_write(path
.mnt
);
3687 if (retry_estale(error
, lookup_flags
)) {
3688 lookup_flags
|= LOOKUP_REVAL
;
3694 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3696 return do_rmdir(AT_FDCWD
, pathname
);
3700 * vfs_unlink - unlink a filesystem object
3701 * @dir: parent directory
3703 * @delegated_inode: returns victim inode, if the inode is delegated.
3705 * The caller must hold dir->i_mutex.
3707 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3708 * return a reference to the inode in delegated_inode. The caller
3709 * should then break the delegation on that inode and retry. Because
3710 * breaking a delegation may take a long time, the caller should drop
3711 * dir->i_mutex before doing so.
3713 * Alternatively, a caller may pass NULL for delegated_inode. This may
3714 * be appropriate for callers that expect the underlying filesystem not
3715 * to be NFS exported.
3717 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3719 struct inode
*target
= dentry
->d_inode
;
3720 int error
= may_delete(dir
, dentry
, 0);
3725 if (!dir
->i_op
->unlink
)
3728 mutex_lock(&target
->i_mutex
);
3729 if (is_local_mountpoint(dentry
))
3732 error
= security_inode_unlink(dir
, dentry
);
3734 error
= try_break_deleg(target
, delegated_inode
);
3737 error
= dir
->i_op
->unlink(dir
, dentry
);
3740 detach_mounts(dentry
);
3745 mutex_unlock(&target
->i_mutex
);
3747 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3748 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3749 fsnotify_link_count(target
);
3755 EXPORT_SYMBOL(vfs_unlink
);
3758 * Make sure that the actual truncation of the file will occur outside its
3759 * directory's i_mutex. Truncate can take a long time if there is a lot of
3760 * writeout happening, and we don't want to prevent access to the directory
3761 * while waiting on the I/O.
3763 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3766 struct filename
*name
;
3767 struct dentry
*dentry
;
3771 struct inode
*inode
= NULL
;
3772 struct inode
*delegated_inode
= NULL
;
3773 unsigned int lookup_flags
= 0;
3775 name
= user_path_parent(dfd
, pathname
,
3776 &path
, &last
, &type
, lookup_flags
);
3778 return PTR_ERR(name
);
3781 if (type
!= LAST_NORM
)
3784 error
= mnt_want_write(path
.mnt
);
3788 mutex_lock_nested(&path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3789 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3790 error
= PTR_ERR(dentry
);
3791 if (!IS_ERR(dentry
)) {
3792 /* Why not before? Because we want correct error value */
3793 if (last
.name
[last
.len
])
3795 inode
= dentry
->d_inode
;
3796 if (d_is_negative(dentry
))
3799 error
= security_path_unlink(&path
, dentry
);
3802 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3806 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
3808 iput(inode
); /* truncate the inode here */
3810 if (delegated_inode
) {
3811 error
= break_deleg_wait(&delegated_inode
);
3815 mnt_drop_write(path
.mnt
);
3819 if (retry_estale(error
, lookup_flags
)) {
3820 lookup_flags
|= LOOKUP_REVAL
;
3827 if (d_is_negative(dentry
))
3829 else if (d_is_dir(dentry
))
3836 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3838 if ((flag
& ~AT_REMOVEDIR
) != 0)
3841 if (flag
& AT_REMOVEDIR
)
3842 return do_rmdir(dfd
, pathname
);
3844 return do_unlinkat(dfd
, pathname
);
3847 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3849 return do_unlinkat(AT_FDCWD
, pathname
);
3852 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3854 int error
= may_create(dir
, dentry
);
3859 if (!dir
->i_op
->symlink
)
3862 error
= security_inode_symlink(dir
, dentry
, oldname
);
3866 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3868 fsnotify_create(dir
, dentry
);
3871 EXPORT_SYMBOL(vfs_symlink
);
3873 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3874 int, newdfd
, const char __user
*, newname
)
3877 struct filename
*from
;
3878 struct dentry
*dentry
;
3880 unsigned int lookup_flags
= 0;
3882 from
= getname(oldname
);
3884 return PTR_ERR(from
);
3886 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3887 error
= PTR_ERR(dentry
);
3891 error
= security_path_symlink(&path
, dentry
, from
->name
);
3893 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3894 done_path_create(&path
, dentry
);
3895 if (retry_estale(error
, lookup_flags
)) {
3896 lookup_flags
|= LOOKUP_REVAL
;
3904 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3906 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3910 * vfs_link - create a new link
3911 * @old_dentry: object to be linked
3913 * @new_dentry: where to create the new link
3914 * @delegated_inode: returns inode needing a delegation break
3916 * The caller must hold dir->i_mutex
3918 * If vfs_link discovers a delegation on the to-be-linked file in need
3919 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3920 * inode in delegated_inode. The caller should then break the delegation
3921 * and retry. Because breaking a delegation may take a long time, the
3922 * caller should drop the i_mutex before doing so.
3924 * Alternatively, a caller may pass NULL for delegated_inode. This may
3925 * be appropriate for callers that expect the underlying filesystem not
3926 * to be NFS exported.
3928 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3930 struct inode
*inode
= old_dentry
->d_inode
;
3931 unsigned max_links
= dir
->i_sb
->s_max_links
;
3937 error
= may_create(dir
, new_dentry
);
3941 if (dir
->i_sb
!= inode
->i_sb
)
3945 * A link to an append-only or immutable file cannot be created.
3947 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3949 if (!dir
->i_op
->link
)
3951 if (S_ISDIR(inode
->i_mode
))
3954 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3958 mutex_lock(&inode
->i_mutex
);
3959 /* Make sure we don't allow creating hardlink to an unlinked file */
3960 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3962 else if (max_links
&& inode
->i_nlink
>= max_links
)
3965 error
= try_break_deleg(inode
, delegated_inode
);
3967 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3970 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3971 spin_lock(&inode
->i_lock
);
3972 inode
->i_state
&= ~I_LINKABLE
;
3973 spin_unlock(&inode
->i_lock
);
3975 mutex_unlock(&inode
->i_mutex
);
3977 fsnotify_link(dir
, inode
, new_dentry
);
3980 EXPORT_SYMBOL(vfs_link
);
3983 * Hardlinks are often used in delicate situations. We avoid
3984 * security-related surprises by not following symlinks on the
3987 * We don't follow them on the oldname either to be compatible
3988 * with linux 2.0, and to avoid hard-linking to directories
3989 * and other special files. --ADM
3991 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3992 int, newdfd
, const char __user
*, newname
, int, flags
)
3994 struct dentry
*new_dentry
;
3995 struct path old_path
, new_path
;
3996 struct inode
*delegated_inode
= NULL
;
4000 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4003 * To use null names we require CAP_DAC_READ_SEARCH
4004 * This ensures that not everyone will be able to create
4005 * handlink using the passed filedescriptor.
4007 if (flags
& AT_EMPTY_PATH
) {
4008 if (!capable(CAP_DAC_READ_SEARCH
))
4013 if (flags
& AT_SYMLINK_FOLLOW
)
4014 how
|= LOOKUP_FOLLOW
;
4016 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4020 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4021 (how
& LOOKUP_REVAL
));
4022 error
= PTR_ERR(new_dentry
);
4023 if (IS_ERR(new_dentry
))
4027 if (old_path
.mnt
!= new_path
.mnt
)
4029 error
= may_linkat(&old_path
);
4030 if (unlikely(error
))
4032 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4035 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4037 done_path_create(&new_path
, new_dentry
);
4038 if (delegated_inode
) {
4039 error
= break_deleg_wait(&delegated_inode
);
4041 path_put(&old_path
);
4045 if (retry_estale(error
, how
)) {
4046 path_put(&old_path
);
4047 how
|= LOOKUP_REVAL
;
4051 path_put(&old_path
);
4056 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4058 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4062 * vfs_rename - rename a filesystem object
4063 * @old_dir: parent of source
4064 * @old_dentry: source
4065 * @new_dir: parent of destination
4066 * @new_dentry: destination
4067 * @delegated_inode: returns an inode needing a delegation break
4068 * @flags: rename flags
4070 * The caller must hold multiple mutexes--see lock_rename()).
4072 * If vfs_rename discovers a delegation in need of breaking at either
4073 * the source or destination, it will return -EWOULDBLOCK and return a
4074 * reference to the inode in delegated_inode. The caller should then
4075 * break the delegation and retry. Because breaking a delegation may
4076 * take a long time, the caller should drop all locks before doing
4079 * Alternatively, a caller may pass NULL for delegated_inode. This may
4080 * be appropriate for callers that expect the underlying filesystem not
4081 * to be NFS exported.
4083 * The worst of all namespace operations - renaming directory. "Perverted"
4084 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4086 * a) we can get into loop creation.
4087 * b) race potential - two innocent renames can create a loop together.
4088 * That's where 4.4 screws up. Current fix: serialization on
4089 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4091 * c) we have to lock _four_ objects - parents and victim (if it exists),
4092 * and source (if it is not a directory).
4093 * And that - after we got ->i_mutex on parents (until then we don't know
4094 * whether the target exists). Solution: try to be smart with locking
4095 * order for inodes. We rely on the fact that tree topology may change
4096 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4097 * move will be locked. Thus we can rank directories by the tree
4098 * (ancestors first) and rank all non-directories after them.
4099 * That works since everybody except rename does "lock parent, lookup,
4100 * lock child" and rename is under ->s_vfs_rename_mutex.
4101 * HOWEVER, it relies on the assumption that any object with ->lookup()
4102 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4103 * we'd better make sure that there's no link(2) for them.
4104 * d) conversion from fhandle to dentry may come in the wrong moment - when
4105 * we are removing the target. Solution: we will have to grab ->i_mutex
4106 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4107 * ->i_mutex on parents, which works but leads to some truly excessive
4110 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4111 struct inode
*new_dir
, struct dentry
*new_dentry
,
4112 struct inode
**delegated_inode
, unsigned int flags
)
4115 bool is_dir
= d_is_dir(old_dentry
);
4116 const unsigned char *old_name
;
4117 struct inode
*source
= old_dentry
->d_inode
;
4118 struct inode
*target
= new_dentry
->d_inode
;
4119 bool new_is_dir
= false;
4120 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4122 if (source
== target
)
4125 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4130 error
= may_create(new_dir
, new_dentry
);
4132 new_is_dir
= d_is_dir(new_dentry
);
4134 if (!(flags
& RENAME_EXCHANGE
))
4135 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4137 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4142 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4145 if (flags
&& !old_dir
->i_op
->rename2
)
4149 * If we are going to change the parent - check write permissions,
4150 * we'll need to flip '..'.
4152 if (new_dir
!= old_dir
) {
4154 error
= inode_permission(source
, MAY_WRITE
);
4158 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4159 error
= inode_permission(target
, MAY_WRITE
);
4165 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4170 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4172 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4173 lock_two_nondirectories(source
, target
);
4175 mutex_lock(&target
->i_mutex
);
4178 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4181 if (max_links
&& new_dir
!= old_dir
) {
4183 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4185 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4186 old_dir
->i_nlink
>= max_links
)
4189 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4190 shrink_dcache_parent(new_dentry
);
4192 error
= try_break_deleg(source
, delegated_inode
);
4196 if (target
&& !new_is_dir
) {
4197 error
= try_break_deleg(target
, delegated_inode
);
4201 if (!old_dir
->i_op
->rename2
) {
4202 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4203 new_dir
, new_dentry
);
4205 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4206 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4207 new_dir
, new_dentry
, flags
);
4212 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4214 target
->i_flags
|= S_DEAD
;
4215 dont_mount(new_dentry
);
4216 detach_mounts(new_dentry
);
4218 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4219 if (!(flags
& RENAME_EXCHANGE
))
4220 d_move(old_dentry
, new_dentry
);
4222 d_exchange(old_dentry
, new_dentry
);
4225 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4226 unlock_two_nondirectories(source
, target
);
4228 mutex_unlock(&target
->i_mutex
);
4231 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4232 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4233 if (flags
& RENAME_EXCHANGE
) {
4234 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4235 new_is_dir
, NULL
, new_dentry
);
4238 fsnotify_oldname_free(old_name
);
4242 EXPORT_SYMBOL(vfs_rename
);
4244 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4245 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4247 struct dentry
*old_dentry
, *new_dentry
;
4248 struct dentry
*trap
;
4249 struct path old_path
, new_path
;
4250 struct qstr old_last
, new_last
;
4251 int old_type
, new_type
;
4252 struct inode
*delegated_inode
= NULL
;
4253 struct filename
*from
;
4254 struct filename
*to
;
4255 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4256 bool should_retry
= false;
4259 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4262 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4263 (flags
& RENAME_EXCHANGE
))
4266 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4269 if (flags
& RENAME_EXCHANGE
)
4273 from
= user_path_parent(olddfd
, oldname
,
4274 &old_path
, &old_last
, &old_type
, lookup_flags
);
4276 error
= PTR_ERR(from
);
4280 to
= user_path_parent(newdfd
, newname
,
4281 &new_path
, &new_last
, &new_type
, lookup_flags
);
4283 error
= PTR_ERR(to
);
4288 if (old_path
.mnt
!= new_path
.mnt
)
4292 if (old_type
!= LAST_NORM
)
4295 if (flags
& RENAME_NOREPLACE
)
4297 if (new_type
!= LAST_NORM
)
4300 error
= mnt_want_write(old_path
.mnt
);
4305 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4307 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4308 error
= PTR_ERR(old_dentry
);
4309 if (IS_ERR(old_dentry
))
4311 /* source must exist */
4313 if (d_is_negative(old_dentry
))
4315 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4316 error
= PTR_ERR(new_dentry
);
4317 if (IS_ERR(new_dentry
))
4320 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4322 if (flags
& RENAME_EXCHANGE
) {
4324 if (d_is_negative(new_dentry
))
4327 if (!d_is_dir(new_dentry
)) {
4329 if (new_last
.name
[new_last
.len
])
4333 /* unless the source is a directory trailing slashes give -ENOTDIR */
4334 if (!d_is_dir(old_dentry
)) {
4336 if (old_last
.name
[old_last
.len
])
4338 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4341 /* source should not be ancestor of target */
4343 if (old_dentry
== trap
)
4345 /* target should not be an ancestor of source */
4346 if (!(flags
& RENAME_EXCHANGE
))
4348 if (new_dentry
== trap
)
4351 error
= security_path_rename(&old_path
, old_dentry
,
4352 &new_path
, new_dentry
, flags
);
4355 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4356 new_path
.dentry
->d_inode
, new_dentry
,
4357 &delegated_inode
, flags
);
4363 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4364 if (delegated_inode
) {
4365 error
= break_deleg_wait(&delegated_inode
);
4369 mnt_drop_write(old_path
.mnt
);
4371 if (retry_estale(error
, lookup_flags
))
4372 should_retry
= true;
4373 path_put(&new_path
);
4376 path_put(&old_path
);
4379 should_retry
= false;
4380 lookup_flags
|= LOOKUP_REVAL
;
4387 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4388 int, newdfd
, const char __user
*, newname
)
4390 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4393 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4395 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4398 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4400 int error
= may_create(dir
, dentry
);
4404 if (!dir
->i_op
->mknod
)
4407 return dir
->i_op
->mknod(dir
, dentry
,
4408 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4410 EXPORT_SYMBOL(vfs_whiteout
);
4412 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4414 int len
= PTR_ERR(link
);
4419 if (len
> (unsigned) buflen
)
4421 if (copy_to_user(buffer
, link
, len
))
4426 EXPORT_SYMBOL(readlink_copy
);
4429 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4430 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4431 * using) it for any given inode is up to filesystem.
4433 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4436 const char *link
= dentry
->d_inode
->i_link
;
4440 link
= dentry
->d_inode
->i_op
->follow_link(dentry
, &cookie
);
4442 return PTR_ERR(link
);
4444 res
= readlink_copy(buffer
, buflen
, link
);
4445 if (dentry
->d_inode
->i_op
->put_link
)
4446 dentry
->d_inode
->i_op
->put_link(dentry
, cookie
);
4449 EXPORT_SYMBOL(generic_readlink
);
4451 /* get the link contents into pagecache */
4452 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4456 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4457 page
= read_mapping_page(mapping
, 0, NULL
);
4462 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4466 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4468 struct page
*page
= NULL
;
4469 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4472 page_cache_release(page
);
4476 EXPORT_SYMBOL(page_readlink
);
4478 const char *page_follow_link_light(struct dentry
*dentry
, void **cookie
)
4480 struct page
*page
= NULL
;
4481 char *res
= page_getlink(dentry
, &page
);
4486 EXPORT_SYMBOL(page_follow_link_light
);
4488 void page_put_link(struct dentry
*dentry
, void *cookie
)
4490 struct page
*page
= cookie
;
4492 page_cache_release(page
);
4494 EXPORT_SYMBOL(page_put_link
);
4497 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4499 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4501 struct address_space
*mapping
= inode
->i_mapping
;
4506 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4508 flags
|= AOP_FLAG_NOFS
;
4511 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4512 flags
, &page
, &fsdata
);
4516 kaddr
= kmap_atomic(page
);
4517 memcpy(kaddr
, symname
, len
-1);
4518 kunmap_atomic(kaddr
);
4520 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4527 mark_inode_dirty(inode
);
4532 EXPORT_SYMBOL(__page_symlink
);
4534 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4536 return __page_symlink(inode
, symname
, len
,
4537 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4539 EXPORT_SYMBOL(page_symlink
);
4541 const struct inode_operations page_symlink_inode_operations
= {
4542 .readlink
= generic_readlink
,
4543 .follow_link
= page_follow_link_light
,
4544 .put_link
= page_put_link
,
4546 EXPORT_SYMBOL(page_symlink_inode_operations
);