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 <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user
*filename
, int flags
, int *empty
)
122 char *result
= __getname(), *err
;
125 if (unlikely(!result
))
126 return ERR_PTR(-ENOMEM
);
128 len
= strncpy_from_user(result
, filename
, PATH_MAX
);
130 if (unlikely(len
< 0))
133 /* The empty path is special. */
134 if (unlikely(!len
)) {
137 err
= ERR_PTR(-ENOENT
);
138 if (!(flags
& LOOKUP_EMPTY
))
142 err
= ERR_PTR(-ENAMETOOLONG
);
143 if (likely(len
< PATH_MAX
)) {
144 audit_getname(result
);
153 char *getname(const char __user
* filename
)
155 return getname_flags(filename
, 0, NULL
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
169 static int check_acl(struct inode
*inode
, int mask
)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl
*acl
;
174 if (mask
& MAY_NOT_BLOCK
) {
175 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl
== ACL_NOT_CACHED
)
181 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
184 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl
== ACL_NOT_CACHED
) {
195 if (inode
->i_op
->get_acl
) {
196 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
200 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
206 int error
= posix_acl_permission(inode
, acl
, mask
);
207 posix_acl_release(acl
);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode
*inode
, int mask
)
220 unsigned int mode
= inode
->i_mode
;
222 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
225 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
226 int error
= check_acl(inode
, mask
);
227 if (error
!= -EAGAIN
)
231 if (in_group_p(inode
->i_gid
))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode
*inode
, int mask
)
262 * Do the basic permission checks.
264 ret
= acl_permission_check(inode
, mask
);
268 if (S_ISDIR(inode
->i_mode
)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
272 if (!(mask
& MAY_WRITE
))
273 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
283 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
287 * Searching includes executable on directories, else just read.
289 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
290 if (mask
== MAY_READ
)
291 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode
*inode
, int mask
)
305 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
306 if (likely(inode
->i_op
->permission
))
307 return inode
->i_op
->permission(inode
, mask
);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode
->i_lock
);
311 inode
->i_opflags
|= IOP_FASTPERM
;
312 spin_unlock(&inode
->i_lock
);
314 return generic_permission(inode
, mask
);
318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
322 * Check for read/write/execute permissions on an inode.
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
329 int __inode_permission(struct inode
*inode
, int mask
)
333 if (unlikely(mask
& MAY_WRITE
)) {
335 * Nobody gets write access to an immutable file.
337 if (IS_IMMUTABLE(inode
))
341 retval
= do_inode_permission(inode
, mask
);
345 retval
= devcgroup_inode_permission(inode
, mask
);
349 return security_inode_permission(inode
, mask
);
353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @inode: Inode to check permission on
356 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
358 * Separate out file-system wide checks from inode-specific permission checks.
360 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
362 if (unlikely(mask
& MAY_WRITE
)) {
363 umode_t mode
= inode
->i_mode
;
365 /* Nobody gets write access to a read-only fs. */
366 if ((sb
->s_flags
& MS_RDONLY
) &&
367 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
374 * inode_permission - Check for access rights to a given inode
375 * @inode: Inode to check permission on
376 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
378 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
379 * this, letting us set arbitrary permissions for filesystem access without
380 * changing the "normal" UIDs which are used for other things.
382 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
384 int inode_permission(struct inode
*inode
, int mask
)
388 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
391 return __inode_permission(inode
, mask
);
395 * path_get - get a reference to a path
396 * @path: path to get the reference to
398 * Given a path increment the reference count to the dentry and the vfsmount.
400 void path_get(struct path
*path
)
405 EXPORT_SYMBOL(path_get
);
408 * path_put - put a reference to a path
409 * @path: path to put the reference to
411 * Given a path decrement the reference count to the dentry and the vfsmount.
413 void path_put(struct path
*path
)
418 EXPORT_SYMBOL(path_put
);
421 * Path walking has 2 modes, rcu-walk and ref-walk (see
422 * Documentation/filesystems/path-lookup.txt). In situations when we can't
423 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
424 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
425 * mode. Refcounts are grabbed at the last known good point before rcu-walk
426 * got stuck, so ref-walk may continue from there. If this is not successful
427 * (eg. a seqcount has changed), then failure is returned and it's up to caller
428 * to restart the path walk from the beginning in ref-walk mode.
431 static inline void lock_rcu_walk(void)
433 br_read_lock(&vfsmount_lock
);
437 static inline void unlock_rcu_walk(void)
440 br_read_unlock(&vfsmount_lock
);
444 * unlazy_walk - try to switch to ref-walk mode.
445 * @nd: nameidata pathwalk data
446 * @dentry: child of nd->path.dentry or NULL
447 * Returns: 0 on success, -ECHILD on failure
449 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
450 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
451 * @nd or NULL. Must be called from rcu-walk context.
453 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
459 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
460 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
462 spin_lock(&fs
->lock
);
463 if (nd
->root
.mnt
!= fs
->root
.mnt
||
464 nd
->root
.dentry
!= fs
->root
.dentry
)
467 spin_lock(&parent
->d_lock
);
469 if (!__d_rcu_to_refcount(parent
, nd
->seq
))
471 BUG_ON(nd
->inode
!= parent
->d_inode
);
473 if (dentry
->d_parent
!= parent
)
475 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
476 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
479 * If the sequence check on the child dentry passed, then
480 * the child has not been removed from its parent. This
481 * means the parent dentry must be valid and able to take
482 * a reference at this point.
484 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
485 BUG_ON(!parent
->d_count
);
487 spin_unlock(&dentry
->d_lock
);
489 spin_unlock(&parent
->d_lock
);
492 spin_unlock(&fs
->lock
);
494 mntget(nd
->path
.mnt
);
497 nd
->flags
&= ~LOOKUP_RCU
;
501 spin_unlock(&dentry
->d_lock
);
503 spin_unlock(&parent
->d_lock
);
506 spin_unlock(&fs
->lock
);
510 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
512 return dentry
->d_op
->d_revalidate(dentry
, flags
);
516 * complete_walk - successful completion of path walk
517 * @nd: pointer nameidata
519 * If we had been in RCU mode, drop out of it and legitimize nd->path.
520 * Revalidate the final result, unless we'd already done that during
521 * the path walk or the filesystem doesn't ask for it. Return 0 on
522 * success, -error on failure. In case of failure caller does not
523 * need to drop nd->path.
525 static int complete_walk(struct nameidata
*nd
)
527 struct dentry
*dentry
= nd
->path
.dentry
;
530 if (nd
->flags
& LOOKUP_RCU
) {
531 nd
->flags
&= ~LOOKUP_RCU
;
532 if (!(nd
->flags
& LOOKUP_ROOT
))
534 spin_lock(&dentry
->d_lock
);
535 if (unlikely(!__d_rcu_to_refcount(dentry
, nd
->seq
))) {
536 spin_unlock(&dentry
->d_lock
);
540 BUG_ON(nd
->inode
!= dentry
->d_inode
);
541 spin_unlock(&dentry
->d_lock
);
542 mntget(nd
->path
.mnt
);
546 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
549 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
552 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
555 /* Note: we do not d_invalidate() */
556 status
= d_revalidate(dentry
, nd
->flags
);
567 static __always_inline
void set_root(struct nameidata
*nd
)
570 get_fs_root(current
->fs
, &nd
->root
);
573 static int link_path_walk(const char *, struct nameidata
*);
575 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
578 struct fs_struct
*fs
= current
->fs
;
582 seq
= read_seqcount_begin(&fs
->seq
);
584 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
585 } while (read_seqcount_retry(&fs
->seq
, seq
));
589 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
601 nd
->flags
|= LOOKUP_JUMPED
;
603 nd
->inode
= nd
->path
.dentry
->d_inode
;
605 ret
= link_path_walk(link
, nd
);
609 return PTR_ERR(link
);
612 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
615 if (path
->mnt
!= nd
->path
.mnt
)
619 static inline void path_to_nameidata(const struct path
*path
,
620 struct nameidata
*nd
)
622 if (!(nd
->flags
& LOOKUP_RCU
)) {
623 dput(nd
->path
.dentry
);
624 if (nd
->path
.mnt
!= path
->mnt
)
625 mntput(nd
->path
.mnt
);
627 nd
->path
.mnt
= path
->mnt
;
628 nd
->path
.dentry
= path
->dentry
;
632 * Helper to directly jump to a known parsed path from ->follow_link,
633 * caller must have taken a reference to path beforehand.
635 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
640 nd
->inode
= nd
->path
.dentry
->d_inode
;
641 nd
->flags
|= LOOKUP_JUMPED
;
643 BUG_ON(nd
->inode
->i_op
->follow_link
);
646 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
648 struct inode
*inode
= link
->dentry
->d_inode
;
649 if (inode
->i_op
->put_link
)
650 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
654 int sysctl_protected_symlinks __read_mostly
= 1;
655 int sysctl_protected_hardlinks __read_mostly
= 1;
658 * may_follow_link - Check symlink following for unsafe situations
659 * @link: The path of the symlink
660 * @nd: nameidata pathwalk data
662 * In the case of the sysctl_protected_symlinks sysctl being enabled,
663 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
664 * in a sticky world-writable directory. This is to protect privileged
665 * processes from failing races against path names that may change out
666 * from under them by way of other users creating malicious symlinks.
667 * It will permit symlinks to be followed only when outside a sticky
668 * world-writable directory, or when the uid of the symlink and follower
669 * match, or when the directory owner matches the symlink's owner.
671 * Returns 0 if following the symlink is allowed, -ve on error.
673 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
675 const struct inode
*inode
;
676 const struct inode
*parent
;
678 if (!sysctl_protected_symlinks
)
681 /* Allowed if owner and follower match. */
682 inode
= link
->dentry
->d_inode
;
683 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
686 /* Allowed if parent directory not sticky and world-writable. */
687 parent
= nd
->path
.dentry
->d_inode
;
688 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
691 /* Allowed if parent directory and link owner match. */
692 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
695 audit_log_link_denied("follow_link", link
);
696 path_put_conditional(link
, nd
);
702 * safe_hardlink_source - Check for safe hardlink conditions
703 * @inode: the source inode to hardlink from
705 * Return false if at least one of the following conditions:
706 * - inode is not a regular file
708 * - inode is setgid and group-exec
709 * - access failure for read and write
711 * Otherwise returns true.
713 static bool safe_hardlink_source(struct inode
*inode
)
715 umode_t mode
= inode
->i_mode
;
717 /* Special files should not get pinned to the filesystem. */
721 /* Setuid files should not get pinned to the filesystem. */
725 /* Executable setgid files should not get pinned to the filesystem. */
726 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
729 /* Hardlinking to unreadable or unwritable sources is dangerous. */
730 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
737 * may_linkat - Check permissions for creating a hardlink
738 * @link: the source to hardlink from
740 * Block hardlink when all of:
741 * - sysctl_protected_hardlinks enabled
742 * - fsuid does not match inode
743 * - hardlink source is unsafe (see safe_hardlink_source() above)
746 * Returns 0 if successful, -ve on error.
748 static int may_linkat(struct path
*link
)
750 const struct cred
*cred
;
753 if (!sysctl_protected_hardlinks
)
756 cred
= current_cred();
757 inode
= link
->dentry
->d_inode
;
759 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
760 * otherwise, it must be a safe source.
762 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
766 audit_log_link_denied("linkat", link
);
770 static __always_inline
int
771 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
773 struct dentry
*dentry
= link
->dentry
;
777 BUG_ON(nd
->flags
& LOOKUP_RCU
);
779 if (link
->mnt
== nd
->path
.mnt
)
783 if (unlikely(current
->total_link_count
>= 40))
784 goto out_put_nd_path
;
787 current
->total_link_count
++;
790 nd_set_link(nd
, NULL
);
792 error
= security_inode_follow_link(link
->dentry
, nd
);
794 goto out_put_nd_path
;
796 nd
->last_type
= LAST_BIND
;
797 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
800 goto out_put_nd_path
;
805 error
= __vfs_follow_link(nd
, s
);
807 put_link(nd
, link
, *p
);
819 static int follow_up_rcu(struct path
*path
)
821 struct mount
*mnt
= real_mount(path
->mnt
);
822 struct mount
*parent
;
823 struct dentry
*mountpoint
;
825 parent
= mnt
->mnt_parent
;
826 if (&parent
->mnt
== path
->mnt
)
828 mountpoint
= mnt
->mnt_mountpoint
;
829 path
->dentry
= mountpoint
;
830 path
->mnt
= &parent
->mnt
;
835 * follow_up - Find the mountpoint of path's vfsmount
837 * Given a path, find the mountpoint of its source file system.
838 * Replace @path with the path of the mountpoint in the parent mount.
841 * Return 1 if we went up a level and 0 if we were already at the
844 int follow_up(struct path
*path
)
846 struct mount
*mnt
= real_mount(path
->mnt
);
847 struct mount
*parent
;
848 struct dentry
*mountpoint
;
850 br_read_lock(&vfsmount_lock
);
851 parent
= mnt
->mnt_parent
;
853 br_read_unlock(&vfsmount_lock
);
856 mntget(&parent
->mnt
);
857 mountpoint
= dget(mnt
->mnt_mountpoint
);
858 br_read_unlock(&vfsmount_lock
);
860 path
->dentry
= mountpoint
;
862 path
->mnt
= &parent
->mnt
;
867 * Perform an automount
868 * - return -EISDIR to tell follow_managed() to stop and return the path we
871 static int follow_automount(struct path
*path
, unsigned flags
,
874 struct vfsmount
*mnt
;
877 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
880 /* We don't want to mount if someone's just doing a stat -
881 * unless they're stat'ing a directory and appended a '/' to
884 * We do, however, want to mount if someone wants to open or
885 * create a file of any type under the mountpoint, wants to
886 * traverse through the mountpoint or wants to open the
887 * mounted directory. Also, autofs may mark negative dentries
888 * as being automount points. These will need the attentions
889 * of the daemon to instantiate them before they can be used.
891 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
892 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
893 path
->dentry
->d_inode
)
896 current
->total_link_count
++;
897 if (current
->total_link_count
>= 40)
900 mnt
= path
->dentry
->d_op
->d_automount(path
);
903 * The filesystem is allowed to return -EISDIR here to indicate
904 * it doesn't want to automount. For instance, autofs would do
905 * this so that its userspace daemon can mount on this dentry.
907 * However, we can only permit this if it's a terminal point in
908 * the path being looked up; if it wasn't then the remainder of
909 * the path is inaccessible and we should say so.
911 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
916 if (!mnt
) /* mount collision */
920 /* lock_mount() may release path->mnt on error */
924 err
= finish_automount(mnt
, path
);
928 /* Someone else made a mount here whilst we were busy */
933 path
->dentry
= dget(mnt
->mnt_root
);
942 * Handle a dentry that is managed in some way.
943 * - Flagged for transit management (autofs)
944 * - Flagged as mountpoint
945 * - Flagged as automount point
947 * This may only be called in refwalk mode.
949 * Serialization is taken care of in namespace.c
951 static int follow_managed(struct path
*path
, unsigned flags
)
953 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
955 bool need_mntput
= false;
958 /* Given that we're not holding a lock here, we retain the value in a
959 * local variable for each dentry as we look at it so that we don't see
960 * the components of that value change under us */
961 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
962 managed
&= DCACHE_MANAGED_DENTRY
,
963 unlikely(managed
!= 0)) {
964 /* Allow the filesystem to manage the transit without i_mutex
966 if (managed
& DCACHE_MANAGE_TRANSIT
) {
967 BUG_ON(!path
->dentry
->d_op
);
968 BUG_ON(!path
->dentry
->d_op
->d_manage
);
969 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
974 /* Transit to a mounted filesystem. */
975 if (managed
& DCACHE_MOUNTED
) {
976 struct vfsmount
*mounted
= lookup_mnt(path
);
982 path
->dentry
= dget(mounted
->mnt_root
);
987 /* Something is mounted on this dentry in another
988 * namespace and/or whatever was mounted there in this
989 * namespace got unmounted before we managed to get the
993 /* Handle an automount point */
994 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
995 ret
= follow_automount(path
, flags
, &need_mntput
);
1001 /* We didn't change the current path point */
1005 if (need_mntput
&& path
->mnt
== mnt
)
1009 return ret
< 0 ? ret
: need_mntput
;
1012 int follow_down_one(struct path
*path
)
1014 struct vfsmount
*mounted
;
1016 mounted
= lookup_mnt(path
);
1020 path
->mnt
= mounted
;
1021 path
->dentry
= dget(mounted
->mnt_root
);
1027 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1029 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1030 dentry
->d_op
->d_manage(dentry
, true) < 0);
1034 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1035 * we meet a managed dentry that would need blocking.
1037 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1038 struct inode
**inode
)
1041 struct mount
*mounted
;
1043 * Don't forget we might have a non-mountpoint managed dentry
1044 * that wants to block transit.
1046 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1049 if (!d_mountpoint(path
->dentry
))
1052 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1055 path
->mnt
= &mounted
->mnt
;
1056 path
->dentry
= mounted
->mnt
.mnt_root
;
1057 nd
->flags
|= LOOKUP_JUMPED
;
1058 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1060 * Update the inode too. We don't need to re-check the
1061 * dentry sequence number here after this d_inode read,
1062 * because a mount-point is always pinned.
1064 *inode
= path
->dentry
->d_inode
;
1069 static void follow_mount_rcu(struct nameidata
*nd
)
1071 while (d_mountpoint(nd
->path
.dentry
)) {
1072 struct mount
*mounted
;
1073 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1076 nd
->path
.mnt
= &mounted
->mnt
;
1077 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1078 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1082 static int follow_dotdot_rcu(struct nameidata
*nd
)
1087 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1088 nd
->path
.mnt
== nd
->root
.mnt
) {
1091 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1092 struct dentry
*old
= nd
->path
.dentry
;
1093 struct dentry
*parent
= old
->d_parent
;
1096 seq
= read_seqcount_begin(&parent
->d_seq
);
1097 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1099 nd
->path
.dentry
= parent
;
1103 if (!follow_up_rcu(&nd
->path
))
1105 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1107 follow_mount_rcu(nd
);
1108 nd
->inode
= nd
->path
.dentry
->d_inode
;
1112 nd
->flags
&= ~LOOKUP_RCU
;
1113 if (!(nd
->flags
& LOOKUP_ROOT
))
1114 nd
->root
.mnt
= NULL
;
1120 * Follow down to the covering mount currently visible to userspace. At each
1121 * point, the filesystem owning that dentry may be queried as to whether the
1122 * caller is permitted to proceed or not.
1124 int follow_down(struct path
*path
)
1129 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1130 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1131 /* Allow the filesystem to manage the transit without i_mutex
1134 * We indicate to the filesystem if someone is trying to mount
1135 * something here. This gives autofs the chance to deny anyone
1136 * other than its daemon the right to mount on its
1139 * The filesystem may sleep at this point.
1141 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1142 BUG_ON(!path
->dentry
->d_op
);
1143 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1144 ret
= path
->dentry
->d_op
->d_manage(
1145 path
->dentry
, false);
1147 return ret
== -EISDIR
? 0 : ret
;
1150 /* Transit to a mounted filesystem. */
1151 if (managed
& DCACHE_MOUNTED
) {
1152 struct vfsmount
*mounted
= lookup_mnt(path
);
1157 path
->mnt
= mounted
;
1158 path
->dentry
= dget(mounted
->mnt_root
);
1162 /* Don't handle automount points here */
1169 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1171 static void follow_mount(struct path
*path
)
1173 while (d_mountpoint(path
->dentry
)) {
1174 struct vfsmount
*mounted
= lookup_mnt(path
);
1179 path
->mnt
= mounted
;
1180 path
->dentry
= dget(mounted
->mnt_root
);
1184 static void follow_dotdot(struct nameidata
*nd
)
1189 struct dentry
*old
= nd
->path
.dentry
;
1191 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1192 nd
->path
.mnt
== nd
->root
.mnt
) {
1195 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1196 /* rare case of legitimate dget_parent()... */
1197 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1201 if (!follow_up(&nd
->path
))
1204 follow_mount(&nd
->path
);
1205 nd
->inode
= nd
->path
.dentry
->d_inode
;
1209 * This looks up the name in dcache, possibly revalidates the old dentry and
1210 * allocates a new one if not found or not valid. In the need_lookup argument
1211 * returns whether i_op->lookup is necessary.
1213 * dir->d_inode->i_mutex must be held
1215 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1216 unsigned int flags
, bool *need_lookup
)
1218 struct dentry
*dentry
;
1221 *need_lookup
= false;
1222 dentry
= d_lookup(dir
, name
);
1224 if (d_need_lookup(dentry
)) {
1225 *need_lookup
= true;
1226 } else if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1227 error
= d_revalidate(dentry
, flags
);
1228 if (unlikely(error
<= 0)) {
1231 return ERR_PTR(error
);
1232 } else if (!d_invalidate(dentry
)) {
1241 dentry
= d_alloc(dir
, name
);
1242 if (unlikely(!dentry
))
1243 return ERR_PTR(-ENOMEM
);
1245 *need_lookup
= true;
1251 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1252 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1254 * dir->d_inode->i_mutex must be held
1256 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1261 /* Don't create child dentry for a dead directory. */
1262 if (unlikely(IS_DEADDIR(dir
))) {
1264 return ERR_PTR(-ENOENT
);
1267 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1268 if (unlikely(old
)) {
1275 static struct dentry
*__lookup_hash(struct qstr
*name
,
1276 struct dentry
*base
, unsigned int flags
)
1279 struct dentry
*dentry
;
1281 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1285 return lookup_real(base
->d_inode
, dentry
, flags
);
1289 * It's more convoluted than I'd like it to be, but... it's still fairly
1290 * small and for now I'd prefer to have fast path as straight as possible.
1291 * It _is_ time-critical.
1293 static int lookup_fast(struct nameidata
*nd
, struct qstr
*name
,
1294 struct path
*path
, struct inode
**inode
)
1296 struct vfsmount
*mnt
= nd
->path
.mnt
;
1297 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1303 * Rename seqlock is not required here because in the off chance
1304 * of a false negative due to a concurrent rename, we're going to
1305 * do the non-racy lookup, below.
1307 if (nd
->flags
& LOOKUP_RCU
) {
1309 dentry
= __d_lookup_rcu(parent
, name
, &seq
, nd
->inode
);
1314 * This sequence count validates that the inode matches
1315 * the dentry name information from lookup.
1317 *inode
= dentry
->d_inode
;
1318 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1322 * This sequence count validates that the parent had no
1323 * changes while we did the lookup of the dentry above.
1325 * The memory barrier in read_seqcount_begin of child is
1326 * enough, we can use __read_seqcount_retry here.
1328 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1332 if (unlikely(d_need_lookup(dentry
)))
1334 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1335 status
= d_revalidate(dentry
, nd
->flags
);
1336 if (unlikely(status
<= 0)) {
1337 if (status
!= -ECHILD
)
1343 path
->dentry
= dentry
;
1344 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1346 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1350 if (unlazy_walk(nd
, dentry
))
1353 dentry
= __d_lookup(parent
, name
);
1356 if (unlikely(!dentry
))
1359 if (unlikely(d_need_lookup(dentry
))) {
1364 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1365 status
= d_revalidate(dentry
, nd
->flags
);
1366 if (unlikely(status
<= 0)) {
1371 if (!d_invalidate(dentry
)) {
1378 path
->dentry
= dentry
;
1379 err
= follow_managed(path
, nd
->flags
);
1380 if (unlikely(err
< 0)) {
1381 path_put_conditional(path
, nd
);
1385 nd
->flags
|= LOOKUP_JUMPED
;
1386 *inode
= path
->dentry
->d_inode
;
1393 /* Fast lookup failed, do it the slow way */
1394 static int lookup_slow(struct nameidata
*nd
, struct qstr
*name
,
1397 struct dentry
*dentry
, *parent
;
1400 parent
= nd
->path
.dentry
;
1401 BUG_ON(nd
->inode
!= parent
->d_inode
);
1403 mutex_lock(&parent
->d_inode
->i_mutex
);
1404 dentry
= __lookup_hash(name
, parent
, nd
->flags
);
1405 mutex_unlock(&parent
->d_inode
->i_mutex
);
1407 return PTR_ERR(dentry
);
1408 path
->mnt
= nd
->path
.mnt
;
1409 path
->dentry
= dentry
;
1410 err
= follow_managed(path
, nd
->flags
);
1411 if (unlikely(err
< 0)) {
1412 path_put_conditional(path
, nd
);
1416 nd
->flags
|= LOOKUP_JUMPED
;
1420 static inline int may_lookup(struct nameidata
*nd
)
1422 if (nd
->flags
& LOOKUP_RCU
) {
1423 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1426 if (unlazy_walk(nd
, NULL
))
1429 return inode_permission(nd
->inode
, MAY_EXEC
);
1432 static inline int handle_dots(struct nameidata
*nd
, int type
)
1434 if (type
== LAST_DOTDOT
) {
1435 if (nd
->flags
& LOOKUP_RCU
) {
1436 if (follow_dotdot_rcu(nd
))
1444 static void terminate_walk(struct nameidata
*nd
)
1446 if (!(nd
->flags
& LOOKUP_RCU
)) {
1447 path_put(&nd
->path
);
1449 nd
->flags
&= ~LOOKUP_RCU
;
1450 if (!(nd
->flags
& LOOKUP_ROOT
))
1451 nd
->root
.mnt
= NULL
;
1457 * Do we need to follow links? We _really_ want to be able
1458 * to do this check without having to look at inode->i_op,
1459 * so we keep a cache of "no, this doesn't need follow_link"
1460 * for the common case.
1462 static inline int should_follow_link(struct inode
*inode
, int follow
)
1464 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1465 if (likely(inode
->i_op
->follow_link
))
1468 /* This gets set once for the inode lifetime */
1469 spin_lock(&inode
->i_lock
);
1470 inode
->i_opflags
|= IOP_NOFOLLOW
;
1471 spin_unlock(&inode
->i_lock
);
1476 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1477 struct qstr
*name
, int type
, int follow
)
1479 struct inode
*inode
;
1482 * "." and ".." are special - ".." especially so because it has
1483 * to be able to know about the current root directory and
1484 * parent relationships.
1486 if (unlikely(type
!= LAST_NORM
))
1487 return handle_dots(nd
, type
);
1488 err
= lookup_fast(nd
, name
, path
, &inode
);
1489 if (unlikely(err
)) {
1493 err
= lookup_slow(nd
, name
, path
);
1497 inode
= path
->dentry
->d_inode
;
1503 if (should_follow_link(inode
, follow
)) {
1504 if (nd
->flags
& LOOKUP_RCU
) {
1505 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1510 BUG_ON(inode
!= path
->dentry
->d_inode
);
1513 path_to_nameidata(path
, nd
);
1518 path_to_nameidata(path
, nd
);
1525 * This limits recursive symlink follows to 8, while
1526 * limiting consecutive symlinks to 40.
1528 * Without that kind of total limit, nasty chains of consecutive
1529 * symlinks can cause almost arbitrarily long lookups.
1531 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1535 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1536 path_put_conditional(path
, nd
);
1537 path_put(&nd
->path
);
1540 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1543 current
->link_count
++;
1546 struct path link
= *path
;
1549 res
= follow_link(&link
, nd
, &cookie
);
1552 res
= walk_component(nd
, path
, &nd
->last
,
1553 nd
->last_type
, LOOKUP_FOLLOW
);
1554 put_link(nd
, &link
, cookie
);
1557 current
->link_count
--;
1563 * We really don't want to look at inode->i_op->lookup
1564 * when we don't have to. So we keep a cache bit in
1565 * the inode ->i_opflags field that says "yes, we can
1566 * do lookup on this inode".
1568 static inline int can_lookup(struct inode
*inode
)
1570 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1572 if (likely(!inode
->i_op
->lookup
))
1575 /* We do this once for the lifetime of the inode */
1576 spin_lock(&inode
->i_lock
);
1577 inode
->i_opflags
|= IOP_LOOKUP
;
1578 spin_unlock(&inode
->i_lock
);
1583 * We can do the critical dentry name comparison and hashing
1584 * operations one word at a time, but we are limited to:
1586 * - Architectures with fast unaligned word accesses. We could
1587 * do a "get_unaligned()" if this helps and is sufficiently
1590 * - Little-endian machines (so that we can generate the mask
1591 * of low bytes efficiently). Again, we *could* do a byte
1592 * swapping load on big-endian architectures if that is not
1593 * expensive enough to make the optimization worthless.
1595 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1596 * do not trap on the (extremely unlikely) case of a page
1597 * crossing operation.
1599 * - Furthermore, we need an efficient 64-bit compile for the
1600 * 64-bit case in order to generate the "number of bytes in
1601 * the final mask". Again, that could be replaced with a
1602 * efficient population count instruction or similar.
1604 #ifdef CONFIG_DCACHE_WORD_ACCESS
1606 #include <asm/word-at-a-time.h>
1610 static inline unsigned int fold_hash(unsigned long hash
)
1612 hash
+= hash
>> (8*sizeof(int));
1616 #else /* 32-bit case */
1618 #define fold_hash(x) (x)
1622 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1624 unsigned long a
, mask
;
1625 unsigned long hash
= 0;
1628 a
= load_unaligned_zeropad(name
);
1629 if (len
< sizeof(unsigned long))
1633 name
+= sizeof(unsigned long);
1634 len
-= sizeof(unsigned long);
1638 mask
= ~(~0ul << len
*8);
1641 return fold_hash(hash
);
1643 EXPORT_SYMBOL(full_name_hash
);
1646 * Calculate the length and hash of the path component, and
1647 * return the length of the component;
1649 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1651 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1652 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1655 len
= -sizeof(unsigned long);
1657 hash
= (hash
+ a
) * 9;
1658 len
+= sizeof(unsigned long);
1659 a
= load_unaligned_zeropad(name
+len
);
1660 b
= a
^ REPEAT_BYTE('/');
1661 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1663 adata
= prep_zero_mask(a
, adata
, &constants
);
1664 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1666 mask
= create_zero_mask(adata
| bdata
);
1668 hash
+= a
& zero_bytemask(mask
);
1669 *hashp
= fold_hash(hash
);
1671 return len
+ find_zero(mask
);
1676 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1678 unsigned long hash
= init_name_hash();
1680 hash
= partial_name_hash(*name
++, hash
);
1681 return end_name_hash(hash
);
1683 EXPORT_SYMBOL(full_name_hash
);
1686 * We know there's a real path component here of at least
1689 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1691 unsigned long hash
= init_name_hash();
1692 unsigned long len
= 0, c
;
1694 c
= (unsigned char)*name
;
1697 hash
= partial_name_hash(c
, hash
);
1698 c
= (unsigned char)name
[len
];
1699 } while (c
&& c
!= '/');
1700 *hashp
= end_name_hash(hash
);
1708 * This is the basic name resolution function, turning a pathname into
1709 * the final dentry. We expect 'base' to be positive and a directory.
1711 * Returns 0 and nd will have valid dentry and mnt on success.
1712 * Returns error and drops reference to input namei data on failure.
1714 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1724 /* At this point we know we have a real path component. */
1730 err
= may_lookup(nd
);
1734 len
= hash_name(name
, &this.hash
);
1739 if (name
[0] == '.') switch (len
) {
1741 if (name
[1] == '.') {
1743 nd
->flags
|= LOOKUP_JUMPED
;
1749 if (likely(type
== LAST_NORM
)) {
1750 struct dentry
*parent
= nd
->path
.dentry
;
1751 nd
->flags
&= ~LOOKUP_JUMPED
;
1752 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1753 err
= parent
->d_op
->d_hash(parent
, nd
->inode
,
1761 goto last_component
;
1763 * If it wasn't NUL, we know it was '/'. Skip that
1764 * slash, and continue until no more slashes.
1768 } while (unlikely(name
[len
] == '/'));
1770 goto last_component
;
1773 err
= walk_component(nd
, &next
, &this, type
, LOOKUP_FOLLOW
);
1778 err
= nested_symlink(&next
, nd
);
1782 if (can_lookup(nd
->inode
))
1786 /* here ends the main loop */
1790 nd
->last_type
= type
;
1797 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1798 struct nameidata
*nd
, struct file
**fp
)
1802 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1803 nd
->flags
= flags
| LOOKUP_JUMPED
;
1805 if (flags
& LOOKUP_ROOT
) {
1806 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1808 if (!inode
->i_op
->lookup
)
1810 retval
= inode_permission(inode
, MAY_EXEC
);
1814 nd
->path
= nd
->root
;
1816 if (flags
& LOOKUP_RCU
) {
1818 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1820 path_get(&nd
->path
);
1825 nd
->root
.mnt
= NULL
;
1828 if (flags
& LOOKUP_RCU
) {
1833 path_get(&nd
->root
);
1835 nd
->path
= nd
->root
;
1836 } else if (dfd
== AT_FDCWD
) {
1837 if (flags
& LOOKUP_RCU
) {
1838 struct fs_struct
*fs
= current
->fs
;
1844 seq
= read_seqcount_begin(&fs
->seq
);
1846 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1847 } while (read_seqcount_retry(&fs
->seq
, seq
));
1849 get_fs_pwd(current
->fs
, &nd
->path
);
1852 struct fd f
= fdget_raw(dfd
);
1853 struct dentry
*dentry
;
1858 dentry
= f
.file
->f_path
.dentry
;
1861 if (!S_ISDIR(dentry
->d_inode
->i_mode
)) {
1866 retval
= inode_permission(dentry
->d_inode
, MAY_EXEC
);
1873 nd
->path
= f
.file
->f_path
;
1874 if (flags
& LOOKUP_RCU
) {
1877 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1880 path_get(&nd
->path
);
1885 nd
->inode
= nd
->path
.dentry
->d_inode
;
1889 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1891 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1892 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1894 nd
->flags
&= ~LOOKUP_PARENT
;
1895 return walk_component(nd
, path
, &nd
->last
, nd
->last_type
,
1896 nd
->flags
& LOOKUP_FOLLOW
);
1899 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1900 static int path_lookupat(int dfd
, const char *name
,
1901 unsigned int flags
, struct nameidata
*nd
)
1903 struct file
*base
= NULL
;
1908 * Path walking is largely split up into 2 different synchronisation
1909 * schemes, rcu-walk and ref-walk (explained in
1910 * Documentation/filesystems/path-lookup.txt). These share much of the
1911 * path walk code, but some things particularly setup, cleanup, and
1912 * following mounts are sufficiently divergent that functions are
1913 * duplicated. Typically there is a function foo(), and its RCU
1914 * analogue, foo_rcu().
1916 * -ECHILD is the error number of choice (just to avoid clashes) that
1917 * is returned if some aspect of an rcu-walk fails. Such an error must
1918 * be handled by restarting a traditional ref-walk (which will always
1919 * be able to complete).
1921 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1926 current
->total_link_count
= 0;
1927 err
= link_path_walk(name
, nd
);
1929 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1930 err
= lookup_last(nd
, &path
);
1933 struct path link
= path
;
1934 err
= may_follow_link(&link
, nd
);
1937 nd
->flags
|= LOOKUP_PARENT
;
1938 err
= follow_link(&link
, nd
, &cookie
);
1941 err
= lookup_last(nd
, &path
);
1942 put_link(nd
, &link
, cookie
);
1947 err
= complete_walk(nd
);
1949 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
1950 if (!nd
->inode
->i_op
->lookup
) {
1951 path_put(&nd
->path
);
1959 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1960 path_put(&nd
->root
);
1961 nd
->root
.mnt
= NULL
;
1966 static int do_path_lookup(int dfd
, const char *name
,
1967 unsigned int flags
, struct nameidata
*nd
)
1969 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
1970 if (unlikely(retval
== -ECHILD
))
1971 retval
= path_lookupat(dfd
, name
, flags
, nd
);
1972 if (unlikely(retval
== -ESTALE
))
1973 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
1975 if (likely(!retval
)) {
1976 if (unlikely(!audit_dummy_context())) {
1977 if (nd
->path
.dentry
&& nd
->inode
)
1978 audit_inode(name
, nd
->path
.dentry
);
1984 /* does lookup, returns the object with parent locked */
1985 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
1987 struct nameidata nd
;
1989 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
1991 return ERR_PTR(err
);
1992 if (nd
.last_type
!= LAST_NORM
) {
1994 return ERR_PTR(-EINVAL
);
1996 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1997 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
1999 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2007 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2009 struct nameidata nd
;
2010 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2017 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2018 * @dentry: pointer to dentry of the base directory
2019 * @mnt: pointer to vfs mount of the base directory
2020 * @name: pointer to file name
2021 * @flags: lookup flags
2022 * @path: pointer to struct path to fill
2024 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2025 const char *name
, unsigned int flags
,
2028 struct nameidata nd
;
2030 nd
.root
.dentry
= dentry
;
2032 BUG_ON(flags
& LOOKUP_PARENT
);
2033 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2034 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2041 * Restricted form of lookup. Doesn't follow links, single-component only,
2042 * needs parent already locked. Doesn't follow mounts.
2045 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2047 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2051 * lookup_one_len - filesystem helper to lookup single pathname component
2052 * @name: pathname component to lookup
2053 * @base: base directory to lookup from
2054 * @len: maximum length @len should be interpreted to
2056 * Note that this routine is purely a helper for filesystem usage and should
2057 * not be called by generic code. Also note that by using this function the
2058 * nameidata argument is passed to the filesystem methods and a filesystem
2059 * using this helper needs to be prepared for that.
2061 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2067 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2071 this.hash
= full_name_hash(name
, len
);
2073 return ERR_PTR(-EACCES
);
2076 c
= *(const unsigned char *)name
++;
2077 if (c
== '/' || c
== '\0')
2078 return ERR_PTR(-EACCES
);
2081 * See if the low-level filesystem might want
2082 * to use its own hash..
2084 if (base
->d_flags
& DCACHE_OP_HASH
) {
2085 int err
= base
->d_op
->d_hash(base
, base
->d_inode
, &this);
2087 return ERR_PTR(err
);
2090 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2092 return ERR_PTR(err
);
2094 return __lookup_hash(&this, base
, 0);
2097 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2098 struct path
*path
, int *empty
)
2100 struct nameidata nd
;
2101 char *tmp
= getname_flags(name
, flags
, empty
);
2102 int err
= PTR_ERR(tmp
);
2105 BUG_ON(flags
& LOOKUP_PARENT
);
2107 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
2115 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2118 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2121 static int user_path_parent(int dfd
, const char __user
*path
,
2122 struct nameidata
*nd
, char **name
)
2124 char *s
= getname(path
);
2130 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
2140 * It's inline, so penalty for filesystems that don't use sticky bit is
2143 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2145 kuid_t fsuid
= current_fsuid();
2147 if (!(dir
->i_mode
& S_ISVTX
))
2149 if (uid_eq(inode
->i_uid
, fsuid
))
2151 if (uid_eq(dir
->i_uid
, fsuid
))
2153 return !inode_capable(inode
, CAP_FOWNER
);
2157 * Check whether we can remove a link victim from directory dir, check
2158 * whether the type of victim is right.
2159 * 1. We can't do it if dir is read-only (done in permission())
2160 * 2. We should have write and exec permissions on dir
2161 * 3. We can't remove anything from append-only dir
2162 * 4. We can't do anything with immutable dir (done in permission())
2163 * 5. If the sticky bit on dir is set we should either
2164 * a. be owner of dir, or
2165 * b. be owner of victim, or
2166 * c. have CAP_FOWNER capability
2167 * 6. If the victim is append-only or immutable we can't do antyhing with
2168 * links pointing to it.
2169 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2170 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2171 * 9. We can't remove a root or mountpoint.
2172 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2173 * nfs_async_unlink().
2175 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2179 if (!victim
->d_inode
)
2182 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2183 audit_inode_child(victim
, dir
);
2185 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2190 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2191 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2194 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2196 if (IS_ROOT(victim
))
2198 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2200 if (IS_DEADDIR(dir
))
2202 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2207 /* Check whether we can create an object with dentry child in directory
2209 * 1. We can't do it if child already exists (open has special treatment for
2210 * this case, but since we are inlined it's OK)
2211 * 2. We can't do it if dir is read-only (done in permission())
2212 * 3. We should have write and exec permissions on dir
2213 * 4. We can't do it if dir is immutable (done in permission())
2215 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2219 if (IS_DEADDIR(dir
))
2221 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2225 * p1 and p2 should be directories on the same fs.
2227 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2232 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2236 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2238 p
= d_ancestor(p2
, p1
);
2240 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2241 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2245 p
= d_ancestor(p1
, p2
);
2247 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2248 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2252 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2253 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2257 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2259 mutex_unlock(&p1
->d_inode
->i_mutex
);
2261 mutex_unlock(&p2
->d_inode
->i_mutex
);
2262 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2266 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2269 int error
= may_create(dir
, dentry
);
2273 if (!dir
->i_op
->create
)
2274 return -EACCES
; /* shouldn't it be ENOSYS? */
2277 error
= security_inode_create(dir
, dentry
, mode
);
2280 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2282 fsnotify_create(dir
, dentry
);
2286 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2288 struct dentry
*dentry
= path
->dentry
;
2289 struct inode
*inode
= dentry
->d_inode
;
2299 switch (inode
->i_mode
& S_IFMT
) {
2303 if (acc_mode
& MAY_WRITE
)
2308 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2317 error
= inode_permission(inode
, acc_mode
);
2322 * An append-only file must be opened in append mode for writing.
2324 if (IS_APPEND(inode
)) {
2325 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2331 /* O_NOATIME can only be set by the owner or superuser */
2332 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2338 static int handle_truncate(struct file
*filp
)
2340 struct path
*path
= &filp
->f_path
;
2341 struct inode
*inode
= path
->dentry
->d_inode
;
2342 int error
= get_write_access(inode
);
2346 * Refuse to truncate files with mandatory locks held on them.
2348 error
= locks_verify_locked(inode
);
2350 error
= security_path_truncate(path
);
2352 error
= do_truncate(path
->dentry
, 0,
2353 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2356 put_write_access(inode
);
2360 static inline int open_to_namei_flags(int flag
)
2362 if ((flag
& O_ACCMODE
) == 3)
2367 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2369 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2373 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2377 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2381 * Attempt to atomically look up, create and open a file from a negative
2384 * Returns 0 if successful. The file will have been created and attached to
2385 * @file by the filesystem calling finish_open().
2387 * Returns 1 if the file was looked up only or didn't need creating. The
2388 * caller will need to perform the open themselves. @path will have been
2389 * updated to point to the new dentry. This may be negative.
2391 * Returns an error code otherwise.
2393 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2394 struct path
*path
, struct file
*file
,
2395 const struct open_flags
*op
,
2396 bool got_write
, bool need_lookup
,
2399 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2400 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2404 int create_error
= 0;
2405 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2407 BUG_ON(dentry
->d_inode
);
2409 /* Don't create child dentry for a dead directory. */
2410 if (unlikely(IS_DEADDIR(dir
))) {
2416 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2417 mode
&= ~current_umask();
2419 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2420 open_flag
&= ~O_TRUNC
;
2421 *opened
|= FILE_CREATED
;
2425 * Checking write permission is tricky, bacuse we don't know if we are
2426 * going to actually need it: O_CREAT opens should work as long as the
2427 * file exists. But checking existence breaks atomicity. The trick is
2428 * to check access and if not granted clear O_CREAT from the flags.
2430 * Another problem is returing the "right" error value (e.g. for an
2431 * O_EXCL open we want to return EEXIST not EROFS).
2433 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2434 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2435 if (!(open_flag
& O_CREAT
)) {
2437 * No O_CREATE -> atomicity not a requirement -> fall
2438 * back to lookup + open
2441 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2442 /* Fall back and fail with the right error */
2443 create_error
= -EROFS
;
2446 /* No side effects, safe to clear O_CREAT */
2447 create_error
= -EROFS
;
2448 open_flag
&= ~O_CREAT
;
2452 if (open_flag
& O_CREAT
) {
2453 error
= may_o_create(&nd
->path
, dentry
, mode
);
2455 create_error
= error
;
2456 if (open_flag
& O_EXCL
)
2458 open_flag
&= ~O_CREAT
;
2462 if (nd
->flags
& LOOKUP_DIRECTORY
)
2463 open_flag
|= O_DIRECTORY
;
2465 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2466 file
->f_path
.mnt
= nd
->path
.mnt
;
2467 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2470 if (create_error
&& error
== -ENOENT
)
2471 error
= create_error
;
2475 acc_mode
= op
->acc_mode
;
2476 if (*opened
& FILE_CREATED
) {
2477 fsnotify_create(dir
, dentry
);
2478 acc_mode
= MAY_OPEN
;
2481 if (error
) { /* returned 1, that is */
2482 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2486 if (file
->f_path
.dentry
) {
2488 dentry
= file
->f_path
.dentry
;
2490 if (create_error
&& dentry
->d_inode
== NULL
) {
2491 error
= create_error
;
2498 * We didn't have the inode before the open, so check open permission
2501 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2511 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2513 return PTR_ERR(dentry
);
2516 int open_flag
= op
->open_flag
;
2518 error
= create_error
;
2519 if ((open_flag
& O_EXCL
)) {
2520 if (!dentry
->d_inode
)
2522 } else if (!dentry
->d_inode
) {
2524 } else if ((open_flag
& O_TRUNC
) &&
2525 S_ISREG(dentry
->d_inode
->i_mode
)) {
2528 /* will fail later, go on to get the right error */
2532 path
->dentry
= dentry
;
2533 path
->mnt
= nd
->path
.mnt
;
2538 * Look up and maybe create and open the last component.
2540 * Must be called with i_mutex held on parent.
2542 * Returns 0 if the file was successfully atomically created (if necessary) and
2543 * opened. In this case the file will be returned attached to @file.
2545 * Returns 1 if the file was not completely opened at this time, though lookups
2546 * and creations will have been performed and the dentry returned in @path will
2547 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2548 * specified then a negative dentry may be returned.
2550 * An error code is returned otherwise.
2552 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2553 * cleared otherwise prior to returning.
2555 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2557 const struct open_flags
*op
,
2558 bool got_write
, int *opened
)
2560 struct dentry
*dir
= nd
->path
.dentry
;
2561 struct inode
*dir_inode
= dir
->d_inode
;
2562 struct dentry
*dentry
;
2566 *opened
&= ~FILE_CREATED
;
2567 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2569 return PTR_ERR(dentry
);
2571 /* Cached positive dentry: will open in f_op->open */
2572 if (!need_lookup
&& dentry
->d_inode
)
2575 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2576 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2577 need_lookup
, opened
);
2581 BUG_ON(dentry
->d_inode
);
2583 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2585 return PTR_ERR(dentry
);
2588 /* Negative dentry, just create the file */
2589 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2590 umode_t mode
= op
->mode
;
2591 if (!IS_POSIXACL(dir
->d_inode
))
2592 mode
&= ~current_umask();
2594 * This write is needed to ensure that a
2595 * rw->ro transition does not occur between
2596 * the time when the file is created and when
2597 * a permanent write count is taken through
2598 * the 'struct file' in finish_open().
2604 *opened
|= FILE_CREATED
;
2605 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2608 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2609 nd
->flags
& LOOKUP_EXCL
);
2614 path
->dentry
= dentry
;
2615 path
->mnt
= nd
->path
.mnt
;
2624 * Handle the last step of open()
2626 static int do_last(struct nameidata
*nd
, struct path
*path
,
2627 struct file
*file
, const struct open_flags
*op
,
2628 int *opened
, const char *pathname
)
2630 struct dentry
*dir
= nd
->path
.dentry
;
2631 int open_flag
= op
->open_flag
;
2632 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2633 bool got_write
= false;
2634 int acc_mode
= op
->acc_mode
;
2635 struct inode
*inode
;
2636 bool symlink_ok
= false;
2637 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2638 bool retried
= false;
2641 nd
->flags
&= ~LOOKUP_PARENT
;
2642 nd
->flags
|= op
->intent
;
2644 switch (nd
->last_type
) {
2647 error
= handle_dots(nd
, nd
->last_type
);
2652 error
= complete_walk(nd
);
2655 audit_inode(pathname
, nd
->path
.dentry
);
2656 if (open_flag
& O_CREAT
) {
2662 error
= complete_walk(nd
);
2665 audit_inode(pathname
, dir
);
2669 if (!(open_flag
& O_CREAT
)) {
2670 if (nd
->last
.name
[nd
->last
.len
])
2671 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2672 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2674 /* we _can_ be in RCU mode here */
2675 error
= lookup_fast(nd
, &nd
->last
, path
, &inode
);
2682 BUG_ON(nd
->inode
!= dir
->d_inode
);
2684 /* create side of things */
2686 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2687 * has been cleared when we got to the last component we are
2690 error
= complete_walk(nd
);
2694 audit_inode(pathname
, dir
);
2696 /* trailing slashes? */
2697 if (nd
->last
.name
[nd
->last
.len
])
2702 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2703 error
= mnt_want_write(nd
->path
.mnt
);
2707 * do _not_ fail yet - we might not need that or fail with
2708 * a different error; let lookup_open() decide; we'll be
2709 * dropping this one anyway.
2712 mutex_lock(&dir
->d_inode
->i_mutex
);
2713 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2714 mutex_unlock(&dir
->d_inode
->i_mutex
);
2720 if ((*opened
& FILE_CREATED
) ||
2721 !S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
2722 will_truncate
= false;
2724 audit_inode(pathname
, file
->f_path
.dentry
);
2728 if (*opened
& FILE_CREATED
) {
2729 /* Don't check for write permission, don't truncate */
2730 open_flag
&= ~O_TRUNC
;
2731 will_truncate
= false;
2732 acc_mode
= MAY_OPEN
;
2733 path_to_nameidata(path
, nd
);
2734 goto finish_open_created
;
2738 * create/update audit record if it already exists.
2740 if (path
->dentry
->d_inode
)
2741 audit_inode(pathname
, path
->dentry
);
2744 * If atomic_open() acquired write access it is dropped now due to
2745 * possible mount and symlink following (this might be optimized away if
2749 mnt_drop_write(nd
->path
.mnt
);
2754 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
2757 error
= follow_managed(path
, nd
->flags
);
2762 nd
->flags
|= LOOKUP_JUMPED
;
2764 BUG_ON(nd
->flags
& LOOKUP_RCU
);
2765 inode
= path
->dentry
->d_inode
;
2767 /* we _can_ be in RCU mode here */
2770 path_to_nameidata(path
, nd
);
2774 if (should_follow_link(inode
, !symlink_ok
)) {
2775 if (nd
->flags
& LOOKUP_RCU
) {
2776 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
2781 BUG_ON(inode
!= path
->dentry
->d_inode
);
2785 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
2786 path_to_nameidata(path
, nd
);
2788 save_parent
.dentry
= nd
->path
.dentry
;
2789 save_parent
.mnt
= mntget(path
->mnt
);
2790 nd
->path
.dentry
= path
->dentry
;
2794 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2795 error
= complete_walk(nd
);
2797 path_put(&save_parent
);
2801 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
2804 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !nd
->inode
->i_op
->lookup
)
2806 audit_inode(pathname
, nd
->path
.dentry
);
2808 if (!S_ISREG(nd
->inode
->i_mode
))
2809 will_truncate
= false;
2811 if (will_truncate
) {
2812 error
= mnt_want_write(nd
->path
.mnt
);
2817 finish_open_created
:
2818 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2821 file
->f_path
.mnt
= nd
->path
.mnt
;
2822 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
2824 if (error
== -EOPENSTALE
)
2829 error
= open_check_o_direct(file
);
2832 error
= ima_file_check(file
, op
->acc_mode
);
2836 if (will_truncate
) {
2837 error
= handle_truncate(file
);
2843 mnt_drop_write(nd
->path
.mnt
);
2844 path_put(&save_parent
);
2849 path_put_conditional(path
, nd
);
2856 /* If no saved parent or already retried then can't retry */
2857 if (!save_parent
.dentry
|| retried
)
2860 BUG_ON(save_parent
.dentry
!= dir
);
2861 path_put(&nd
->path
);
2862 nd
->path
= save_parent
;
2863 nd
->inode
= dir
->d_inode
;
2864 save_parent
.mnt
= NULL
;
2865 save_parent
.dentry
= NULL
;
2867 mnt_drop_write(nd
->path
.mnt
);
2874 static struct file
*path_openat(int dfd
, const char *pathname
,
2875 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
2877 struct file
*base
= NULL
;
2883 file
= get_empty_filp();
2885 return ERR_PTR(-ENFILE
);
2887 file
->f_flags
= op
->open_flag
;
2889 error
= path_init(dfd
, pathname
, flags
| LOOKUP_PARENT
, nd
, &base
);
2890 if (unlikely(error
))
2893 current
->total_link_count
= 0;
2894 error
= link_path_walk(pathname
, nd
);
2895 if (unlikely(error
))
2898 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2899 while (unlikely(error
> 0)) { /* trailing symlink */
2900 struct path link
= path
;
2902 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
2903 path_put_conditional(&path
, nd
);
2904 path_put(&nd
->path
);
2908 error
= may_follow_link(&link
, nd
);
2909 if (unlikely(error
))
2911 nd
->flags
|= LOOKUP_PARENT
;
2912 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
2913 error
= follow_link(&link
, nd
, &cookie
);
2914 if (unlikely(error
))
2916 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
2917 put_link(nd
, &link
, cookie
);
2920 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
2921 path_put(&nd
->root
);
2924 if (!(opened
& FILE_OPENED
)) {
2928 if (unlikely(error
)) {
2929 if (error
== -EOPENSTALE
) {
2930 if (flags
& LOOKUP_RCU
)
2935 file
= ERR_PTR(error
);
2940 struct file
*do_filp_open(int dfd
, const char *pathname
,
2941 const struct open_flags
*op
, int flags
)
2943 struct nameidata nd
;
2946 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
2947 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
2948 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
2949 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
2950 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
2954 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
2955 const char *name
, const struct open_flags
*op
, int flags
)
2957 struct nameidata nd
;
2961 nd
.root
.dentry
= dentry
;
2963 flags
|= LOOKUP_ROOT
;
2965 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
2966 return ERR_PTR(-ELOOP
);
2968 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_RCU
);
2969 if (unlikely(file
== ERR_PTR(-ECHILD
)))
2970 file
= path_openat(-1, name
, &nd
, op
, flags
);
2971 if (unlikely(file
== ERR_PTR(-ESTALE
)))
2972 file
= path_openat(-1, name
, &nd
, op
, flags
| LOOKUP_REVAL
);
2976 struct dentry
*kern_path_create(int dfd
, const char *pathname
, struct path
*path
, int is_dir
)
2978 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2979 struct nameidata nd
;
2981 int error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
, &nd
);
2983 return ERR_PTR(error
);
2986 * Yucky last component or no last component at all?
2987 * (foo/., foo/.., /////)
2989 if (nd
.last_type
!= LAST_NORM
)
2991 nd
.flags
&= ~LOOKUP_PARENT
;
2992 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2994 /* don't fail immediately if it's r/o, at least try to report other errors */
2995 err2
= mnt_want_write(nd
.path
.mnt
);
2997 * Do the final lookup.
2999 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3000 dentry
= lookup_hash(&nd
);
3005 if (dentry
->d_inode
)
3008 * Special case - lookup gave negative, but... we had foo/bar/
3009 * From the vfs_mknod() POV we just have a negative dentry -
3010 * all is fine. Let's be bastards - you had / on the end, you've
3011 * been asking for (non-existent) directory. -ENOENT for you.
3013 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3017 if (unlikely(err2
)) {
3025 dentry
= ERR_PTR(error
);
3027 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3029 mnt_drop_write(nd
.path
.mnt
);
3034 EXPORT_SYMBOL(kern_path_create
);
3036 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3039 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3040 mnt_drop_write(path
->mnt
);
3043 EXPORT_SYMBOL(done_path_create
);
3045 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
, struct path
*path
, int is_dir
)
3047 char *tmp
= getname(pathname
);
3050 return ERR_CAST(tmp
);
3051 res
= kern_path_create(dfd
, tmp
, path
, is_dir
);
3055 EXPORT_SYMBOL(user_path_create
);
3057 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3059 int error
= may_create(dir
, dentry
);
3064 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3067 if (!dir
->i_op
->mknod
)
3070 error
= devcgroup_inode_mknod(mode
, dev
);
3074 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3078 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3080 fsnotify_create(dir
, dentry
);
3084 static int may_mknod(umode_t mode
)
3086 switch (mode
& S_IFMT
) {
3092 case 0: /* zero mode translates to S_IFREG */
3101 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3104 struct dentry
*dentry
;
3108 error
= may_mknod(mode
);
3112 dentry
= user_path_create(dfd
, filename
, &path
, 0);
3114 return PTR_ERR(dentry
);
3116 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3117 mode
&= ~current_umask();
3118 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3121 switch (mode
& S_IFMT
) {
3122 case 0: case S_IFREG
:
3123 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3125 case S_IFCHR
: case S_IFBLK
:
3126 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3127 new_decode_dev(dev
));
3129 case S_IFIFO
: case S_IFSOCK
:
3130 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3134 done_path_create(&path
, dentry
);
3138 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3140 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3143 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3145 int error
= may_create(dir
, dentry
);
3146 unsigned max_links
= dir
->i_sb
->s_max_links
;
3151 if (!dir
->i_op
->mkdir
)
3154 mode
&= (S_IRWXUGO
|S_ISVTX
);
3155 error
= security_inode_mkdir(dir
, dentry
, mode
);
3159 if (max_links
&& dir
->i_nlink
>= max_links
)
3162 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3164 fsnotify_mkdir(dir
, dentry
);
3168 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3170 struct dentry
*dentry
;
3174 dentry
= user_path_create(dfd
, pathname
, &path
, 1);
3176 return PTR_ERR(dentry
);
3178 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3179 mode
&= ~current_umask();
3180 error
= security_path_mkdir(&path
, dentry
, mode
);
3182 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3183 done_path_create(&path
, dentry
);
3187 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3189 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3193 * The dentry_unhash() helper will try to drop the dentry early: we
3194 * should have a usage count of 1 if we're the only user of this
3195 * dentry, and if that is true (possibly after pruning the dcache),
3196 * then we drop the dentry now.
3198 * A low-level filesystem can, if it choses, legally
3201 * if (!d_unhashed(dentry))
3204 * if it cannot handle the case of removing a directory
3205 * that is still in use by something else..
3207 void dentry_unhash(struct dentry
*dentry
)
3209 shrink_dcache_parent(dentry
);
3210 spin_lock(&dentry
->d_lock
);
3211 if (dentry
->d_count
== 1)
3213 spin_unlock(&dentry
->d_lock
);
3216 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3218 int error
= may_delete(dir
, dentry
, 1);
3223 if (!dir
->i_op
->rmdir
)
3227 mutex_lock(&dentry
->d_inode
->i_mutex
);
3230 if (d_mountpoint(dentry
))
3233 error
= security_inode_rmdir(dir
, dentry
);
3237 shrink_dcache_parent(dentry
);
3238 error
= dir
->i_op
->rmdir(dir
, dentry
);
3242 dentry
->d_inode
->i_flags
|= S_DEAD
;
3246 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3253 static long do_rmdir(int dfd
, const char __user
*pathname
)
3257 struct dentry
*dentry
;
3258 struct nameidata nd
;
3260 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3264 switch(nd
.last_type
) {
3276 nd
.flags
&= ~LOOKUP_PARENT
;
3277 error
= mnt_want_write(nd
.path
.mnt
);
3281 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3282 dentry
= lookup_hash(&nd
);
3283 error
= PTR_ERR(dentry
);
3286 if (!dentry
->d_inode
) {
3290 error
= security_path_rmdir(&nd
.path
, dentry
);
3293 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3297 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3298 mnt_drop_write(nd
.path
.mnt
);
3305 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3307 return do_rmdir(AT_FDCWD
, pathname
);
3310 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3312 int error
= may_delete(dir
, dentry
, 0);
3317 if (!dir
->i_op
->unlink
)
3320 mutex_lock(&dentry
->d_inode
->i_mutex
);
3321 if (d_mountpoint(dentry
))
3324 error
= security_inode_unlink(dir
, dentry
);
3326 error
= dir
->i_op
->unlink(dir
, dentry
);
3331 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3333 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3334 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3335 fsnotify_link_count(dentry
->d_inode
);
3343 * Make sure that the actual truncation of the file will occur outside its
3344 * directory's i_mutex. Truncate can take a long time if there is a lot of
3345 * writeout happening, and we don't want to prevent access to the directory
3346 * while waiting on the I/O.
3348 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3352 struct dentry
*dentry
;
3353 struct nameidata nd
;
3354 struct inode
*inode
= NULL
;
3356 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
3361 if (nd
.last_type
!= LAST_NORM
)
3364 nd
.flags
&= ~LOOKUP_PARENT
;
3365 error
= mnt_want_write(nd
.path
.mnt
);
3369 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3370 dentry
= lookup_hash(&nd
);
3371 error
= PTR_ERR(dentry
);
3372 if (!IS_ERR(dentry
)) {
3373 /* Why not before? Because we want correct error value */
3374 if (nd
.last
.name
[nd
.last
.len
])
3376 inode
= dentry
->d_inode
;
3380 error
= security_path_unlink(&nd
.path
, dentry
);
3383 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3387 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3389 iput(inode
); /* truncate the inode here */
3390 mnt_drop_write(nd
.path
.mnt
);
3397 error
= !dentry
->d_inode
? -ENOENT
:
3398 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3402 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3404 if ((flag
& ~AT_REMOVEDIR
) != 0)
3407 if (flag
& AT_REMOVEDIR
)
3408 return do_rmdir(dfd
, pathname
);
3410 return do_unlinkat(dfd
, pathname
);
3413 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3415 return do_unlinkat(AT_FDCWD
, pathname
);
3418 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3420 int error
= may_create(dir
, dentry
);
3425 if (!dir
->i_op
->symlink
)
3428 error
= security_inode_symlink(dir
, dentry
, oldname
);
3432 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3434 fsnotify_create(dir
, dentry
);
3438 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3439 int, newdfd
, const char __user
*, newname
)
3443 struct dentry
*dentry
;
3446 from
= getname(oldname
);
3448 return PTR_ERR(from
);
3450 dentry
= user_path_create(newdfd
, newname
, &path
, 0);
3451 error
= PTR_ERR(dentry
);
3455 error
= security_path_symlink(&path
, dentry
, from
);
3457 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
);
3458 done_path_create(&path
, dentry
);
3464 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3466 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3469 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3471 struct inode
*inode
= old_dentry
->d_inode
;
3472 unsigned max_links
= dir
->i_sb
->s_max_links
;
3478 error
= may_create(dir
, new_dentry
);
3482 if (dir
->i_sb
!= inode
->i_sb
)
3486 * A link to an append-only or immutable file cannot be created.
3488 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3490 if (!dir
->i_op
->link
)
3492 if (S_ISDIR(inode
->i_mode
))
3495 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3499 mutex_lock(&inode
->i_mutex
);
3500 /* Make sure we don't allow creating hardlink to an unlinked file */
3501 if (inode
->i_nlink
== 0)
3503 else if (max_links
&& inode
->i_nlink
>= max_links
)
3506 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3507 mutex_unlock(&inode
->i_mutex
);
3509 fsnotify_link(dir
, inode
, new_dentry
);
3514 * Hardlinks are often used in delicate situations. We avoid
3515 * security-related surprises by not following symlinks on the
3518 * We don't follow them on the oldname either to be compatible
3519 * with linux 2.0, and to avoid hard-linking to directories
3520 * and other special files. --ADM
3522 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3523 int, newdfd
, const char __user
*, newname
, int, flags
)
3525 struct dentry
*new_dentry
;
3526 struct path old_path
, new_path
;
3530 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3533 * To use null names we require CAP_DAC_READ_SEARCH
3534 * This ensures that not everyone will be able to create
3535 * handlink using the passed filedescriptor.
3537 if (flags
& AT_EMPTY_PATH
) {
3538 if (!capable(CAP_DAC_READ_SEARCH
))
3543 if (flags
& AT_SYMLINK_FOLLOW
)
3544 how
|= LOOKUP_FOLLOW
;
3546 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3550 new_dentry
= user_path_create(newdfd
, newname
, &new_path
, 0);
3551 error
= PTR_ERR(new_dentry
);
3552 if (IS_ERR(new_dentry
))
3556 if (old_path
.mnt
!= new_path
.mnt
)
3558 error
= may_linkat(&old_path
);
3559 if (unlikely(error
))
3561 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3564 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3566 done_path_create(&new_path
, new_dentry
);
3568 path_put(&old_path
);
3573 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3575 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3579 * The worst of all namespace operations - renaming directory. "Perverted"
3580 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3582 * a) we can get into loop creation. Check is done in is_subdir().
3583 * b) race potential - two innocent renames can create a loop together.
3584 * That's where 4.4 screws up. Current fix: serialization on
3585 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3587 * c) we have to lock _three_ objects - parents and victim (if it exists).
3588 * And that - after we got ->i_mutex on parents (until then we don't know
3589 * whether the target exists). Solution: try to be smart with locking
3590 * order for inodes. We rely on the fact that tree topology may change
3591 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3592 * move will be locked. Thus we can rank directories by the tree
3593 * (ancestors first) and rank all non-directories after them.
3594 * That works since everybody except rename does "lock parent, lookup,
3595 * lock child" and rename is under ->s_vfs_rename_mutex.
3596 * HOWEVER, it relies on the assumption that any object with ->lookup()
3597 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3598 * we'd better make sure that there's no link(2) for them.
3599 * d) conversion from fhandle to dentry may come in the wrong moment - when
3600 * we are removing the target. Solution: we will have to grab ->i_mutex
3601 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3602 * ->i_mutex on parents, which works but leads to some truly excessive
3605 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3606 struct inode
*new_dir
, struct dentry
*new_dentry
)
3609 struct inode
*target
= new_dentry
->d_inode
;
3610 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3613 * If we are going to change the parent - check write permissions,
3614 * we'll need to flip '..'.
3616 if (new_dir
!= old_dir
) {
3617 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3622 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3628 mutex_lock(&target
->i_mutex
);
3631 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
3635 if (max_links
&& !target
&& new_dir
!= old_dir
&&
3636 new_dir
->i_nlink
>= max_links
)
3640 shrink_dcache_parent(new_dentry
);
3641 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3646 target
->i_flags
|= S_DEAD
;
3647 dont_mount(new_dentry
);
3651 mutex_unlock(&target
->i_mutex
);
3654 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3655 d_move(old_dentry
,new_dentry
);
3659 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3660 struct inode
*new_dir
, struct dentry
*new_dentry
)
3662 struct inode
*target
= new_dentry
->d_inode
;
3665 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3671 mutex_lock(&target
->i_mutex
);
3674 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3677 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3682 dont_mount(new_dentry
);
3683 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3684 d_move(old_dentry
, new_dentry
);
3687 mutex_unlock(&target
->i_mutex
);
3692 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3693 struct inode
*new_dir
, struct dentry
*new_dentry
)
3696 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3697 const unsigned char *old_name
;
3699 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3702 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3706 if (!new_dentry
->d_inode
)
3707 error
= may_create(new_dir
, new_dentry
);
3709 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3713 if (!old_dir
->i_op
->rename
)
3716 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3719 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3721 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3723 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3724 new_dentry
->d_inode
, old_dentry
);
3725 fsnotify_oldname_free(old_name
);
3730 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3731 int, newdfd
, const char __user
*, newname
)
3733 struct dentry
*old_dir
, *new_dir
;
3734 struct dentry
*old_dentry
, *new_dentry
;
3735 struct dentry
*trap
;
3736 struct nameidata oldnd
, newnd
;
3741 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3745 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3750 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3753 old_dir
= oldnd
.path
.dentry
;
3755 if (oldnd
.last_type
!= LAST_NORM
)
3758 new_dir
= newnd
.path
.dentry
;
3759 if (newnd
.last_type
!= LAST_NORM
)
3762 error
= mnt_want_write(oldnd
.path
.mnt
);
3766 oldnd
.flags
&= ~LOOKUP_PARENT
;
3767 newnd
.flags
&= ~LOOKUP_PARENT
;
3768 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3770 trap
= lock_rename(new_dir
, old_dir
);
3772 old_dentry
= lookup_hash(&oldnd
);
3773 error
= PTR_ERR(old_dentry
);
3774 if (IS_ERR(old_dentry
))
3776 /* source must exist */
3778 if (!old_dentry
->d_inode
)
3780 /* unless the source is a directory trailing slashes give -ENOTDIR */
3781 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3783 if (oldnd
.last
.name
[oldnd
.last
.len
])
3785 if (newnd
.last
.name
[newnd
.last
.len
])
3788 /* source should not be ancestor of target */
3790 if (old_dentry
== trap
)
3792 new_dentry
= lookup_hash(&newnd
);
3793 error
= PTR_ERR(new_dentry
);
3794 if (IS_ERR(new_dentry
))
3796 /* target should not be an ancestor of source */
3798 if (new_dentry
== trap
)
3801 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3802 &newnd
.path
, new_dentry
);
3805 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3806 new_dir
->d_inode
, new_dentry
);
3812 unlock_rename(new_dir
, old_dir
);
3813 mnt_drop_write(oldnd
.path
.mnt
);
3815 path_put(&newnd
.path
);
3818 path_put(&oldnd
.path
);
3824 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3826 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3829 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3833 len
= PTR_ERR(link
);
3838 if (len
> (unsigned) buflen
)
3840 if (copy_to_user(buffer
, link
, len
))
3847 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3848 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3849 * using) it for any given inode is up to filesystem.
3851 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3853 struct nameidata nd
;
3858 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3860 return PTR_ERR(cookie
);
3862 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3863 if (dentry
->d_inode
->i_op
->put_link
)
3864 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3868 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3870 return __vfs_follow_link(nd
, link
);
3873 /* get the link contents into pagecache */
3874 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3878 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3879 page
= read_mapping_page(mapping
, 0, NULL
);
3884 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3888 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3890 struct page
*page
= NULL
;
3891 char *s
= page_getlink(dentry
, &page
);
3892 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3895 page_cache_release(page
);
3900 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3902 struct page
*page
= NULL
;
3903 nd_set_link(nd
, page_getlink(dentry
, &page
));
3907 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3909 struct page
*page
= cookie
;
3913 page_cache_release(page
);
3918 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3920 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3922 struct address_space
*mapping
= inode
->i_mapping
;
3927 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3929 flags
|= AOP_FLAG_NOFS
;
3932 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3933 flags
, &page
, &fsdata
);
3937 kaddr
= kmap_atomic(page
);
3938 memcpy(kaddr
, symname
, len
-1);
3939 kunmap_atomic(kaddr
);
3941 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3948 mark_inode_dirty(inode
);
3954 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3956 return __page_symlink(inode
, symname
, len
,
3957 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3960 const struct inode_operations page_symlink_inode_operations
= {
3961 .readlink
= generic_readlink
,
3962 .follow_link
= page_follow_link_light
,
3963 .put_link
= page_put_link
,
3966 EXPORT_SYMBOL(user_path_at
);
3967 EXPORT_SYMBOL(follow_down_one
);
3968 EXPORT_SYMBOL(follow_down
);
3969 EXPORT_SYMBOL(follow_up
);
3970 EXPORT_SYMBOL(get_write_access
); /* nfsd */
3971 EXPORT_SYMBOL(getname
);
3972 EXPORT_SYMBOL(lock_rename
);
3973 EXPORT_SYMBOL(lookup_one_len
);
3974 EXPORT_SYMBOL(page_follow_link_light
);
3975 EXPORT_SYMBOL(page_put_link
);
3976 EXPORT_SYMBOL(page_readlink
);
3977 EXPORT_SYMBOL(__page_symlink
);
3978 EXPORT_SYMBOL(page_symlink
);
3979 EXPORT_SYMBOL(page_symlink_inode_operations
);
3980 EXPORT_SYMBOL(kern_path
);
3981 EXPORT_SYMBOL(vfs_path_lookup
);
3982 EXPORT_SYMBOL(inode_permission
);
3983 EXPORT_SYMBOL(unlock_rename
);
3984 EXPORT_SYMBOL(vfs_create
);
3985 EXPORT_SYMBOL(vfs_follow_link
);
3986 EXPORT_SYMBOL(vfs_link
);
3987 EXPORT_SYMBOL(vfs_mkdir
);
3988 EXPORT_SYMBOL(vfs_mknod
);
3989 EXPORT_SYMBOL(generic_permission
);
3990 EXPORT_SYMBOL(vfs_readlink
);
3991 EXPORT_SYMBOL(vfs_rename
);
3992 EXPORT_SYMBOL(vfs_rmdir
);
3993 EXPORT_SYMBOL(vfs_symlink
);
3994 EXPORT_SYMBOL(vfs_unlink
);
3995 EXPORT_SYMBOL(dentry_unhash
);
3996 EXPORT_SYMBOL(generic_readlink
);