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 - sizeof(struct filename))
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
+ sizeof(*result
);
144 result
->name
= kname
;
145 result
->separate
= false;
147 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
148 if (unlikely(len
< 0)) {
154 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155 * separate struct filename so we can dedicate the entire
156 * names_cache allocation for the pathname, and re-do the copy from
159 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
160 kname
= (char *)result
;
162 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
163 if (unlikely(!result
)) {
165 return ERR_PTR(-ENOMEM
);
167 result
->name
= kname
;
168 result
->separate
= true;
169 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
170 if (unlikely(len
< 0)) {
175 if (unlikely(len
== PATH_MAX
)) {
178 return ERR_PTR(-ENAMETOOLONG
);
183 /* The empty path is special. */
184 if (unlikely(!len
)) {
187 if (!(flags
& LOOKUP_EMPTY
)) {
189 return ERR_PTR(-ENOENT
);
193 result
->uptr
= filename
;
194 result
->aname
= NULL
;
195 audit_getname(result
);
200 getname(const char __user
* filename
)
202 return getname_flags(filename
, 0, NULL
);
206 getname_kernel(const char * filename
)
208 struct filename
*result
;
209 int len
= strlen(filename
) + 1;
211 result
= __getname();
212 if (unlikely(!result
))
213 return ERR_PTR(-ENOMEM
);
215 if (len
<= EMBEDDED_NAME_MAX
) {
216 result
->name
= (char *)(result
) + sizeof(*result
);
217 result
->separate
= false;
218 } else if (len
<= PATH_MAX
) {
219 struct filename
*tmp
;
221 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
222 if (unlikely(!tmp
)) {
224 return ERR_PTR(-ENOMEM
);
226 tmp
->name
= (char *)result
;
227 tmp
->separate
= true;
231 return ERR_PTR(-ENAMETOOLONG
);
233 memcpy((char *)result
->name
, filename
, len
);
235 result
->aname
= NULL
;
237 audit_getname(result
);
242 void putname(struct filename
*name
)
244 BUG_ON(name
->refcnt
<= 0);
246 if (--name
->refcnt
> 0)
249 if (name
->separate
) {
250 __putname(name
->name
);
256 static int check_acl(struct inode
*inode
, int mask
)
258 #ifdef CONFIG_FS_POSIX_ACL
259 struct posix_acl
*acl
;
261 if (mask
& MAY_NOT_BLOCK
) {
262 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
265 /* no ->get_acl() calls in RCU mode... */
266 if (acl
== ACL_NOT_CACHED
)
268 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
271 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
275 int error
= posix_acl_permission(inode
, acl
, mask
);
276 posix_acl_release(acl
);
285 * This does the basic permission checking
287 static int acl_permission_check(struct inode
*inode
, int mask
)
289 unsigned int mode
= inode
->i_mode
;
291 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
294 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
295 int error
= check_acl(inode
, mask
);
296 if (error
!= -EAGAIN
)
300 if (in_group_p(inode
->i_gid
))
305 * If the DACs are ok we don't need any capability check.
307 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
313 * generic_permission - check for access rights on a Posix-like filesystem
314 * @inode: inode to check access rights for
315 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
317 * Used to check for read/write/execute permissions on a file.
318 * We use "fsuid" for this, letting us set arbitrary permissions
319 * for filesystem access without changing the "normal" uids which
320 * are used for other things.
322 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
323 * request cannot be satisfied (eg. requires blocking or too much complexity).
324 * It would then be called again in ref-walk mode.
326 int generic_permission(struct inode
*inode
, int mask
)
331 * Do the basic permission checks.
333 ret
= acl_permission_check(inode
, mask
);
337 if (S_ISDIR(inode
->i_mode
)) {
338 /* DACs are overridable for directories */
339 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
341 if (!(mask
& MAY_WRITE
))
342 if (capable_wrt_inode_uidgid(inode
,
343 CAP_DAC_READ_SEARCH
))
348 * Read/write DACs are always overridable.
349 * Executable DACs are overridable when there is
350 * at least one exec bit set.
352 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
353 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
357 * Searching includes executable on directories, else just read.
359 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
360 if (mask
== MAY_READ
)
361 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
366 EXPORT_SYMBOL(generic_permission
);
369 * We _really_ want to just do "generic_permission()" without
370 * even looking at the inode->i_op values. So we keep a cache
371 * flag in inode->i_opflags, that says "this has not special
372 * permission function, use the fast case".
374 static inline int do_inode_permission(struct inode
*inode
, int mask
)
376 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
377 if (likely(inode
->i_op
->permission
))
378 return inode
->i_op
->permission(inode
, mask
);
380 /* This gets set once for the inode lifetime */
381 spin_lock(&inode
->i_lock
);
382 inode
->i_opflags
|= IOP_FASTPERM
;
383 spin_unlock(&inode
->i_lock
);
385 return generic_permission(inode
, mask
);
389 * __inode_permission - Check for access rights to a given inode
390 * @inode: Inode to check permission on
391 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
393 * Check for read/write/execute permissions on an inode.
395 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
397 * This does not check for a read-only file system. You probably want
398 * inode_permission().
400 int __inode_permission(struct inode
*inode
, int mask
)
404 if (unlikely(mask
& MAY_WRITE
)) {
406 * Nobody gets write access to an immutable file.
408 if (IS_IMMUTABLE(inode
))
412 retval
= do_inode_permission(inode
, mask
);
416 retval
= devcgroup_inode_permission(inode
, mask
);
420 return security_inode_permission(inode
, mask
);
422 EXPORT_SYMBOL(__inode_permission
);
425 * sb_permission - Check superblock-level permissions
426 * @sb: Superblock of inode to check permission on
427 * @inode: Inode to check permission on
428 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
430 * Separate out file-system wide checks from inode-specific permission checks.
432 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
434 if (unlikely(mask
& MAY_WRITE
)) {
435 umode_t mode
= inode
->i_mode
;
437 /* Nobody gets write access to a read-only fs. */
438 if ((sb
->s_flags
& MS_RDONLY
) &&
439 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
446 * inode_permission - Check for access rights to a given inode
447 * @inode: Inode to check permission on
448 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
450 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
451 * this, letting us set arbitrary permissions for filesystem access without
452 * changing the "normal" UIDs which are used for other things.
454 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
456 int inode_permission(struct inode
*inode
, int mask
)
460 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
463 return __inode_permission(inode
, mask
);
465 EXPORT_SYMBOL(inode_permission
);
468 * path_get - get a reference to a path
469 * @path: path to get the reference to
471 * Given a path increment the reference count to the dentry and the vfsmount.
473 void path_get(const struct path
*path
)
478 EXPORT_SYMBOL(path_get
);
481 * path_put - put a reference to a path
482 * @path: path to put the reference to
484 * Given a path decrement the reference count to the dentry and the vfsmount.
486 void path_put(const struct path
*path
)
491 EXPORT_SYMBOL(path_put
);
497 struct inode
*inode
; /* path.dentry.d_inode */
503 char *saved_names
[MAX_NESTED_LINKS
+ 1];
507 * Path walking has 2 modes, rcu-walk and ref-walk (see
508 * Documentation/filesystems/path-lookup.txt). In situations when we can't
509 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
510 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
511 * mode. Refcounts are grabbed at the last known good point before rcu-walk
512 * got stuck, so ref-walk may continue from there. If this is not successful
513 * (eg. a seqcount has changed), then failure is returned and it's up to caller
514 * to restart the path walk from the beginning in ref-walk mode.
518 * unlazy_walk - try to switch to ref-walk mode.
519 * @nd: nameidata pathwalk data
520 * @dentry: child of nd->path.dentry or NULL
521 * Returns: 0 on success, -ECHILD on failure
523 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
524 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
525 * @nd or NULL. Must be called from rcu-walk context.
527 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
529 struct fs_struct
*fs
= current
->fs
;
530 struct dentry
*parent
= nd
->path
.dentry
;
532 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
535 * After legitimizing the bastards, terminate_walk()
536 * will do the right thing for non-RCU mode, and all our
537 * subsequent exit cases should rcu_read_unlock()
538 * before returning. Do vfsmount first; if dentry
539 * can't be legitimized, just set nd->path.dentry to NULL
540 * and rely on dput(NULL) being a no-op.
542 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
544 nd
->flags
&= ~LOOKUP_RCU
;
546 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
547 nd
->path
.dentry
= NULL
;
552 * For a negative lookup, the lookup sequence point is the parents
553 * sequence point, and it only needs to revalidate the parent dentry.
555 * For a positive lookup, we need to move both the parent and the
556 * dentry from the RCU domain to be properly refcounted. And the
557 * sequence number in the dentry validates *both* dentry counters,
558 * since we checked the sequence number of the parent after we got
559 * the child sequence number. So we know the parent must still
560 * be valid if the child sequence number is still valid.
563 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
565 BUG_ON(nd
->inode
!= parent
->d_inode
);
567 if (!lockref_get_not_dead(&dentry
->d_lockref
))
569 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
574 * Sequence counts matched. Now make sure that the root is
575 * still valid and get it if required.
577 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
578 spin_lock(&fs
->lock
);
579 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
580 goto unlock_and_drop_dentry
;
582 spin_unlock(&fs
->lock
);
588 unlock_and_drop_dentry
:
589 spin_unlock(&fs
->lock
);
597 if (!(nd
->flags
& LOOKUP_ROOT
))
602 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
604 return dentry
->d_op
->d_revalidate(dentry
, flags
);
608 * complete_walk - successful completion of path walk
609 * @nd: pointer nameidata
611 * If we had been in RCU mode, drop out of it and legitimize nd->path.
612 * Revalidate the final result, unless we'd already done that during
613 * the path walk or the filesystem doesn't ask for it. Return 0 on
614 * success, -error on failure. In case of failure caller does not
615 * need to drop nd->path.
617 static int complete_walk(struct nameidata
*nd
)
619 struct dentry
*dentry
= nd
->path
.dentry
;
622 if (nd
->flags
& LOOKUP_RCU
) {
623 nd
->flags
&= ~LOOKUP_RCU
;
624 if (!(nd
->flags
& LOOKUP_ROOT
))
627 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
631 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
633 mntput(nd
->path
.mnt
);
636 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
639 mntput(nd
->path
.mnt
);
645 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
648 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
651 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
662 static __always_inline
void set_root(struct nameidata
*nd
)
664 get_fs_root(current
->fs
, &nd
->root
);
667 static int link_path_walk(const char *, struct nameidata
*);
669 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
671 struct fs_struct
*fs
= current
->fs
;
675 seq
= read_seqcount_begin(&fs
->seq
);
677 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
678 } while (read_seqcount_retry(&fs
->seq
, seq
));
682 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
685 if (path
->mnt
!= nd
->path
.mnt
)
689 static inline void path_to_nameidata(const struct path
*path
,
690 struct nameidata
*nd
)
692 if (!(nd
->flags
& LOOKUP_RCU
)) {
693 dput(nd
->path
.dentry
);
694 if (nd
->path
.mnt
!= path
->mnt
)
695 mntput(nd
->path
.mnt
);
697 nd
->path
.mnt
= path
->mnt
;
698 nd
->path
.dentry
= path
->dentry
;
702 * Helper to directly jump to a known parsed path from ->follow_link,
703 * caller must have taken a reference to path beforehand.
705 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
710 nd
->inode
= nd
->path
.dentry
->d_inode
;
711 nd
->flags
|= LOOKUP_JUMPED
;
714 void nd_set_link(struct nameidata
*nd
, char *path
)
716 nd
->saved_names
[nd
->depth
] = path
;
718 EXPORT_SYMBOL(nd_set_link
);
720 char *nd_get_link(struct nameidata
*nd
)
722 return nd
->saved_names
[nd
->depth
];
724 EXPORT_SYMBOL(nd_get_link
);
726 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
728 struct inode
*inode
= link
->dentry
->d_inode
;
729 if (inode
->i_op
->put_link
)
730 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
734 int sysctl_protected_symlinks __read_mostly
= 0;
735 int sysctl_protected_hardlinks __read_mostly
= 0;
738 * may_follow_link - Check symlink following for unsafe situations
739 * @link: The path of the symlink
740 * @nd: nameidata pathwalk data
742 * In the case of the sysctl_protected_symlinks sysctl being enabled,
743 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
744 * in a sticky world-writable directory. This is to protect privileged
745 * processes from failing races against path names that may change out
746 * from under them by way of other users creating malicious symlinks.
747 * It will permit symlinks to be followed only when outside a sticky
748 * world-writable directory, or when the uid of the symlink and follower
749 * match, or when the directory owner matches the symlink's owner.
751 * Returns 0 if following the symlink is allowed, -ve on error.
753 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
755 const struct inode
*inode
;
756 const struct inode
*parent
;
758 if (!sysctl_protected_symlinks
)
761 /* Allowed if owner and follower match. */
762 inode
= link
->dentry
->d_inode
;
763 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
766 /* Allowed if parent directory not sticky and world-writable. */
767 parent
= nd
->path
.dentry
->d_inode
;
768 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
771 /* Allowed if parent directory and link owner match. */
772 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
775 audit_log_link_denied("follow_link", link
);
776 path_put_conditional(link
, nd
);
782 * safe_hardlink_source - Check for safe hardlink conditions
783 * @inode: the source inode to hardlink from
785 * Return false if at least one of the following conditions:
786 * - inode is not a regular file
788 * - inode is setgid and group-exec
789 * - access failure for read and write
791 * Otherwise returns true.
793 static bool safe_hardlink_source(struct inode
*inode
)
795 umode_t mode
= inode
->i_mode
;
797 /* Special files should not get pinned to the filesystem. */
801 /* Setuid files should not get pinned to the filesystem. */
805 /* Executable setgid files should not get pinned to the filesystem. */
806 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
809 /* Hardlinking to unreadable or unwritable sources is dangerous. */
810 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
817 * may_linkat - Check permissions for creating a hardlink
818 * @link: the source to hardlink from
820 * Block hardlink when all of:
821 * - sysctl_protected_hardlinks enabled
822 * - fsuid does not match inode
823 * - hardlink source is unsafe (see safe_hardlink_source() above)
826 * Returns 0 if successful, -ve on error.
828 static int may_linkat(struct path
*link
)
830 const struct cred
*cred
;
833 if (!sysctl_protected_hardlinks
)
836 cred
= current_cred();
837 inode
= link
->dentry
->d_inode
;
839 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
840 * otherwise, it must be a safe source.
842 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
846 audit_log_link_denied("linkat", link
);
850 static __always_inline
int
851 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
853 struct dentry
*dentry
= link
->dentry
;
857 BUG_ON(nd
->flags
& LOOKUP_RCU
);
859 if (link
->mnt
== nd
->path
.mnt
)
863 if (unlikely(current
->total_link_count
>= 40))
864 goto out_put_nd_path
;
867 current
->total_link_count
++;
870 nd_set_link(nd
, NULL
);
872 error
= security_inode_follow_link(link
->dentry
, nd
);
874 goto out_put_nd_path
;
876 nd
->last_type
= LAST_BIND
;
877 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
880 goto out_put_nd_path
;
885 if (unlikely(IS_ERR(s
))) {
887 put_link(nd
, link
, *p
);
896 nd
->flags
|= LOOKUP_JUMPED
;
898 nd
->inode
= nd
->path
.dentry
->d_inode
;
899 error
= link_path_walk(s
, nd
);
901 put_link(nd
, link
, *p
);
913 static int follow_up_rcu(struct path
*path
)
915 struct mount
*mnt
= real_mount(path
->mnt
);
916 struct mount
*parent
;
917 struct dentry
*mountpoint
;
919 parent
= mnt
->mnt_parent
;
920 if (&parent
->mnt
== path
->mnt
)
922 mountpoint
= mnt
->mnt_mountpoint
;
923 path
->dentry
= mountpoint
;
924 path
->mnt
= &parent
->mnt
;
929 * follow_up - Find the mountpoint of path's vfsmount
931 * Given a path, find the mountpoint of its source file system.
932 * Replace @path with the path of the mountpoint in the parent mount.
935 * Return 1 if we went up a level and 0 if we were already at the
938 int follow_up(struct path
*path
)
940 struct mount
*mnt
= real_mount(path
->mnt
);
941 struct mount
*parent
;
942 struct dentry
*mountpoint
;
944 read_seqlock_excl(&mount_lock
);
945 parent
= mnt
->mnt_parent
;
947 read_sequnlock_excl(&mount_lock
);
950 mntget(&parent
->mnt
);
951 mountpoint
= dget(mnt
->mnt_mountpoint
);
952 read_sequnlock_excl(&mount_lock
);
954 path
->dentry
= mountpoint
;
956 path
->mnt
= &parent
->mnt
;
959 EXPORT_SYMBOL(follow_up
);
962 * Perform an automount
963 * - return -EISDIR to tell follow_managed() to stop and return the path we
966 static int follow_automount(struct path
*path
, unsigned flags
,
969 struct vfsmount
*mnt
;
972 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
975 /* We don't want to mount if someone's just doing a stat -
976 * unless they're stat'ing a directory and appended a '/' to
979 * We do, however, want to mount if someone wants to open or
980 * create a file of any type under the mountpoint, wants to
981 * traverse through the mountpoint or wants to open the
982 * mounted directory. Also, autofs may mark negative dentries
983 * as being automount points. These will need the attentions
984 * of the daemon to instantiate them before they can be used.
986 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
987 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
988 path
->dentry
->d_inode
)
991 current
->total_link_count
++;
992 if (current
->total_link_count
>= 40)
995 mnt
= path
->dentry
->d_op
->d_automount(path
);
998 * The filesystem is allowed to return -EISDIR here to indicate
999 * it doesn't want to automount. For instance, autofs would do
1000 * this so that its userspace daemon can mount on this dentry.
1002 * However, we can only permit this if it's a terminal point in
1003 * the path being looked up; if it wasn't then the remainder of
1004 * the path is inaccessible and we should say so.
1006 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1008 return PTR_ERR(mnt
);
1011 if (!mnt
) /* mount collision */
1014 if (!*need_mntput
) {
1015 /* lock_mount() may release path->mnt on error */
1017 *need_mntput
= true;
1019 err
= finish_automount(mnt
, path
);
1023 /* Someone else made a mount here whilst we were busy */
1028 path
->dentry
= dget(mnt
->mnt_root
);
1037 * Handle a dentry that is managed in some way.
1038 * - Flagged for transit management (autofs)
1039 * - Flagged as mountpoint
1040 * - Flagged as automount point
1042 * This may only be called in refwalk mode.
1044 * Serialization is taken care of in namespace.c
1046 static int follow_managed(struct path
*path
, unsigned flags
)
1048 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1050 bool need_mntput
= false;
1053 /* Given that we're not holding a lock here, we retain the value in a
1054 * local variable for each dentry as we look at it so that we don't see
1055 * the components of that value change under us */
1056 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1057 managed
&= DCACHE_MANAGED_DENTRY
,
1058 unlikely(managed
!= 0)) {
1059 /* Allow the filesystem to manage the transit without i_mutex
1061 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1062 BUG_ON(!path
->dentry
->d_op
);
1063 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1064 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1069 /* Transit to a mounted filesystem. */
1070 if (managed
& DCACHE_MOUNTED
) {
1071 struct vfsmount
*mounted
= lookup_mnt(path
);
1076 path
->mnt
= mounted
;
1077 path
->dentry
= dget(mounted
->mnt_root
);
1082 /* Something is mounted on this dentry in another
1083 * namespace and/or whatever was mounted there in this
1084 * namespace got unmounted before lookup_mnt() could
1088 /* Handle an automount point */
1089 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1090 ret
= follow_automount(path
, flags
, &need_mntput
);
1096 /* We didn't change the current path point */
1100 if (need_mntput
&& path
->mnt
== mnt
)
1104 return ret
< 0 ? ret
: need_mntput
;
1107 int follow_down_one(struct path
*path
)
1109 struct vfsmount
*mounted
;
1111 mounted
= lookup_mnt(path
);
1115 path
->mnt
= mounted
;
1116 path
->dentry
= dget(mounted
->mnt_root
);
1121 EXPORT_SYMBOL(follow_down_one
);
1123 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1125 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1126 dentry
->d_op
->d_manage(dentry
, true) : 0;
1130 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1131 * we meet a managed dentry that would need blocking.
1133 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1134 struct inode
**inode
)
1137 struct mount
*mounted
;
1139 * Don't forget we might have a non-mountpoint managed dentry
1140 * that wants to block transit.
1142 switch (managed_dentry_rcu(path
->dentry
)) {
1152 if (!d_mountpoint(path
->dentry
))
1153 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1155 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1158 path
->mnt
= &mounted
->mnt
;
1159 path
->dentry
= mounted
->mnt
.mnt_root
;
1160 nd
->flags
|= LOOKUP_JUMPED
;
1161 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1163 * Update the inode too. We don't need to re-check the
1164 * dentry sequence number here after this d_inode read,
1165 * because a mount-point is always pinned.
1167 *inode
= path
->dentry
->d_inode
;
1169 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1170 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1173 static int follow_dotdot_rcu(struct nameidata
*nd
)
1175 struct inode
*inode
= nd
->inode
;
1180 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1181 nd
->path
.mnt
== nd
->root
.mnt
) {
1184 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1185 struct dentry
*old
= nd
->path
.dentry
;
1186 struct dentry
*parent
= old
->d_parent
;
1189 inode
= parent
->d_inode
;
1190 seq
= read_seqcount_begin(&parent
->d_seq
);
1191 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1193 nd
->path
.dentry
= parent
;
1197 if (!follow_up_rcu(&nd
->path
))
1199 inode
= nd
->path
.dentry
->d_inode
;
1200 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1202 while (d_mountpoint(nd
->path
.dentry
)) {
1203 struct mount
*mounted
;
1204 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1207 nd
->path
.mnt
= &mounted
->mnt
;
1208 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1209 inode
= nd
->path
.dentry
->d_inode
;
1210 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1211 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1218 nd
->flags
&= ~LOOKUP_RCU
;
1219 if (!(nd
->flags
& LOOKUP_ROOT
))
1220 nd
->root
.mnt
= NULL
;
1226 * Follow down to the covering mount currently visible to userspace. At each
1227 * point, the filesystem owning that dentry may be queried as to whether the
1228 * caller is permitted to proceed or not.
1230 int follow_down(struct path
*path
)
1235 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1236 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1237 /* Allow the filesystem to manage the transit without i_mutex
1240 * We indicate to the filesystem if someone is trying to mount
1241 * something here. This gives autofs the chance to deny anyone
1242 * other than its daemon the right to mount on its
1245 * The filesystem may sleep at this point.
1247 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1248 BUG_ON(!path
->dentry
->d_op
);
1249 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1250 ret
= path
->dentry
->d_op
->d_manage(
1251 path
->dentry
, false);
1253 return ret
== -EISDIR
? 0 : ret
;
1256 /* Transit to a mounted filesystem. */
1257 if (managed
& DCACHE_MOUNTED
) {
1258 struct vfsmount
*mounted
= lookup_mnt(path
);
1263 path
->mnt
= mounted
;
1264 path
->dentry
= dget(mounted
->mnt_root
);
1268 /* Don't handle automount points here */
1273 EXPORT_SYMBOL(follow_down
);
1276 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1278 static void follow_mount(struct path
*path
)
1280 while (d_mountpoint(path
->dentry
)) {
1281 struct vfsmount
*mounted
= lookup_mnt(path
);
1286 path
->mnt
= mounted
;
1287 path
->dentry
= dget(mounted
->mnt_root
);
1291 static void follow_dotdot(struct nameidata
*nd
)
1297 struct dentry
*old
= nd
->path
.dentry
;
1299 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1300 nd
->path
.mnt
== nd
->root
.mnt
) {
1303 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1304 /* rare case of legitimate dget_parent()... */
1305 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1309 if (!follow_up(&nd
->path
))
1312 follow_mount(&nd
->path
);
1313 nd
->inode
= nd
->path
.dentry
->d_inode
;
1317 * This looks up the name in dcache, possibly revalidates the old dentry and
1318 * allocates a new one if not found or not valid. In the need_lookup argument
1319 * returns whether i_op->lookup is necessary.
1321 * dir->d_inode->i_mutex must be held
1323 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1324 unsigned int flags
, bool *need_lookup
)
1326 struct dentry
*dentry
;
1329 *need_lookup
= false;
1330 dentry
= d_lookup(dir
, name
);
1332 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1333 error
= d_revalidate(dentry
, flags
);
1334 if (unlikely(error
<= 0)) {
1337 return ERR_PTR(error
);
1339 d_invalidate(dentry
);
1348 dentry
= d_alloc(dir
, name
);
1349 if (unlikely(!dentry
))
1350 return ERR_PTR(-ENOMEM
);
1352 *need_lookup
= true;
1358 * Call i_op->lookup on the dentry. The dentry must be negative and
1361 * dir->d_inode->i_mutex must be held
1363 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1368 /* Don't create child dentry for a dead directory. */
1369 if (unlikely(IS_DEADDIR(dir
))) {
1371 return ERR_PTR(-ENOENT
);
1374 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1375 if (unlikely(old
)) {
1382 static struct dentry
*__lookup_hash(struct qstr
*name
,
1383 struct dentry
*base
, unsigned int flags
)
1386 struct dentry
*dentry
;
1388 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1392 return lookup_real(base
->d_inode
, dentry
, flags
);
1396 * It's more convoluted than I'd like it to be, but... it's still fairly
1397 * small and for now I'd prefer to have fast path as straight as possible.
1398 * It _is_ time-critical.
1400 static int lookup_fast(struct nameidata
*nd
,
1401 struct path
*path
, struct inode
**inode
)
1403 struct vfsmount
*mnt
= nd
->path
.mnt
;
1404 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1410 * Rename seqlock is not required here because in the off chance
1411 * of a false negative due to a concurrent rename, we're going to
1412 * do the non-racy lookup, below.
1414 if (nd
->flags
& LOOKUP_RCU
) {
1416 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1421 * This sequence count validates that the inode matches
1422 * the dentry name information from lookup.
1424 *inode
= dentry
->d_inode
;
1425 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1429 * This sequence count validates that the parent had no
1430 * changes while we did the lookup of the dentry above.
1432 * The memory barrier in read_seqcount_begin of child is
1433 * enough, we can use __read_seqcount_retry here.
1435 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1439 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1440 status
= d_revalidate(dentry
, nd
->flags
);
1441 if (unlikely(status
<= 0)) {
1442 if (status
!= -ECHILD
)
1448 path
->dentry
= dentry
;
1449 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1452 if (unlazy_walk(nd
, dentry
))
1455 dentry
= __d_lookup(parent
, &nd
->last
);
1458 if (unlikely(!dentry
))
1461 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1462 status
= d_revalidate(dentry
, nd
->flags
);
1463 if (unlikely(status
<= 0)) {
1468 d_invalidate(dentry
);
1474 path
->dentry
= dentry
;
1475 err
= follow_managed(path
, nd
->flags
);
1476 if (unlikely(err
< 0)) {
1477 path_put_conditional(path
, nd
);
1481 nd
->flags
|= LOOKUP_JUMPED
;
1482 *inode
= path
->dentry
->d_inode
;
1489 /* Fast lookup failed, do it the slow way */
1490 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1492 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 err
= follow_managed(path
, nd
->flags
);
1506 if (unlikely(err
< 0)) {
1507 path_put_conditional(path
, nd
);
1511 nd
->flags
|= LOOKUP_JUMPED
;
1515 static inline int may_lookup(struct nameidata
*nd
)
1517 if (nd
->flags
& LOOKUP_RCU
) {
1518 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1521 if (unlazy_walk(nd
, NULL
))
1524 return inode_permission(nd
->inode
, MAY_EXEC
);
1527 static inline int handle_dots(struct nameidata
*nd
, int type
)
1529 if (type
== LAST_DOTDOT
) {
1530 if (nd
->flags
& LOOKUP_RCU
) {
1531 if (follow_dotdot_rcu(nd
))
1539 static void terminate_walk(struct nameidata
*nd
)
1541 if (!(nd
->flags
& LOOKUP_RCU
)) {
1542 path_put(&nd
->path
);
1544 nd
->flags
&= ~LOOKUP_RCU
;
1545 if (!(nd
->flags
& LOOKUP_ROOT
))
1546 nd
->root
.mnt
= NULL
;
1552 * Do we need to follow links? We _really_ want to be able
1553 * to do this check without having to look at inode->i_op,
1554 * so we keep a cache of "no, this doesn't need follow_link"
1555 * for the common case.
1557 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1559 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1562 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1565 struct inode
*inode
;
1568 * "." and ".." are special - ".." especially so because it has
1569 * to be able to know about the current root directory and
1570 * parent relationships.
1572 if (unlikely(nd
->last_type
!= LAST_NORM
))
1573 return handle_dots(nd
, nd
->last_type
);
1574 err
= lookup_fast(nd
, path
, &inode
);
1575 if (unlikely(err
)) {
1579 err
= lookup_slow(nd
, path
);
1583 inode
= path
->dentry
->d_inode
;
1586 if (!inode
|| d_is_negative(path
->dentry
))
1589 if (should_follow_link(path
->dentry
, follow
)) {
1590 if (nd
->flags
& LOOKUP_RCU
) {
1591 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1596 BUG_ON(inode
!= path
->dentry
->d_inode
);
1599 path_to_nameidata(path
, nd
);
1604 path_to_nameidata(path
, nd
);
1611 * This limits recursive symlink follows to 8, while
1612 * limiting consecutive symlinks to 40.
1614 * Without that kind of total limit, nasty chains of consecutive
1615 * symlinks can cause almost arbitrarily long lookups.
1617 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1621 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1622 path_put_conditional(path
, nd
);
1623 path_put(&nd
->path
);
1626 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1629 current
->link_count
++;
1632 struct path link
= *path
;
1635 res
= follow_link(&link
, nd
, &cookie
);
1638 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1639 put_link(nd
, &link
, cookie
);
1642 current
->link_count
--;
1648 * We can do the critical dentry name comparison and hashing
1649 * operations one word at a time, but we are limited to:
1651 * - Architectures with fast unaligned word accesses. We could
1652 * do a "get_unaligned()" if this helps and is sufficiently
1655 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1656 * do not trap on the (extremely unlikely) case of a page
1657 * crossing operation.
1659 * - Furthermore, we need an efficient 64-bit compile for the
1660 * 64-bit case in order to generate the "number of bytes in
1661 * the final mask". Again, that could be replaced with a
1662 * efficient population count instruction or similar.
1664 #ifdef CONFIG_DCACHE_WORD_ACCESS
1666 #include <asm/word-at-a-time.h>
1670 static inline unsigned int fold_hash(unsigned long hash
)
1672 return hash_64(hash
, 32);
1675 #else /* 32-bit case */
1677 #define fold_hash(x) (x)
1681 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1683 unsigned long a
, mask
;
1684 unsigned long hash
= 0;
1687 a
= load_unaligned_zeropad(name
);
1688 if (len
< sizeof(unsigned long))
1692 name
+= sizeof(unsigned long);
1693 len
-= sizeof(unsigned long);
1697 mask
= bytemask_from_count(len
);
1700 return fold_hash(hash
);
1702 EXPORT_SYMBOL(full_name_hash
);
1705 * Calculate the length and hash of the path component, and
1706 * return the "hash_len" as the result.
1708 static inline u64
hash_name(const char *name
)
1710 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1711 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1714 len
= -sizeof(unsigned long);
1716 hash
= (hash
+ a
) * 9;
1717 len
+= sizeof(unsigned long);
1718 a
= load_unaligned_zeropad(name
+len
);
1719 b
= a
^ REPEAT_BYTE('/');
1720 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1722 adata
= prep_zero_mask(a
, adata
, &constants
);
1723 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1725 mask
= create_zero_mask(adata
| bdata
);
1727 hash
+= a
& zero_bytemask(mask
);
1728 len
+= find_zero(mask
);
1729 return hashlen_create(fold_hash(hash
), len
);
1734 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1736 unsigned long hash
= init_name_hash();
1738 hash
= partial_name_hash(*name
++, hash
);
1739 return end_name_hash(hash
);
1741 EXPORT_SYMBOL(full_name_hash
);
1744 * We know there's a real path component here of at least
1747 static inline u64
hash_name(const char *name
)
1749 unsigned long hash
= init_name_hash();
1750 unsigned long len
= 0, c
;
1752 c
= (unsigned char)*name
;
1755 hash
= partial_name_hash(c
, hash
);
1756 c
= (unsigned char)name
[len
];
1757 } while (c
&& c
!= '/');
1758 return hashlen_create(end_name_hash(hash
), len
);
1765 * This is the basic name resolution function, turning a pathname into
1766 * the final dentry. We expect 'base' to be positive and a directory.
1768 * Returns 0 and nd will have valid dentry and mnt on success.
1769 * Returns error and drops reference to input namei data on failure.
1771 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1781 /* At this point we know we have a real path component. */
1786 err
= may_lookup(nd
);
1790 hash_len
= hash_name(name
);
1793 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1795 if (name
[1] == '.') {
1797 nd
->flags
|= LOOKUP_JUMPED
;
1803 if (likely(type
== LAST_NORM
)) {
1804 struct dentry
*parent
= nd
->path
.dentry
;
1805 nd
->flags
&= ~LOOKUP_JUMPED
;
1806 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1807 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1808 err
= parent
->d_op
->d_hash(parent
, &this);
1811 hash_len
= this.hash_len
;
1816 nd
->last
.hash_len
= hash_len
;
1817 nd
->last
.name
= name
;
1818 nd
->last_type
= type
;
1820 name
+= hashlen_len(hash_len
);
1824 * If it wasn't NUL, we know it was '/'. Skip that
1825 * slash, and continue until no more slashes.
1829 } while (unlikely(*name
== '/'));
1833 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1838 err
= nested_symlink(&next
, nd
);
1842 if (!d_can_lookup(nd
->path
.dentry
)) {
1851 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1852 struct nameidata
*nd
)
1856 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1857 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
1860 if (flags
& LOOKUP_ROOT
) {
1861 struct dentry
*root
= nd
->root
.dentry
;
1862 struct inode
*inode
= root
->d_inode
;
1864 if (!d_can_lookup(root
))
1866 retval
= inode_permission(inode
, MAY_EXEC
);
1870 nd
->path
= nd
->root
;
1872 if (flags
& LOOKUP_RCU
) {
1874 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1875 nd
->m_seq
= read_seqbegin(&mount_lock
);
1877 path_get(&nd
->path
);
1882 nd
->root
.mnt
= NULL
;
1884 nd
->m_seq
= read_seqbegin(&mount_lock
);
1886 if (flags
& LOOKUP_RCU
) {
1888 nd
->seq
= set_root_rcu(nd
);
1891 path_get(&nd
->root
);
1893 nd
->path
= nd
->root
;
1894 } else if (dfd
== AT_FDCWD
) {
1895 if (flags
& LOOKUP_RCU
) {
1896 struct fs_struct
*fs
= current
->fs
;
1902 seq
= read_seqcount_begin(&fs
->seq
);
1904 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1905 } while (read_seqcount_retry(&fs
->seq
, seq
));
1907 get_fs_pwd(current
->fs
, &nd
->path
);
1910 /* Caller must check execute permissions on the starting path component */
1911 struct fd f
= fdget_raw(dfd
);
1912 struct dentry
*dentry
;
1917 dentry
= f
.file
->f_path
.dentry
;
1920 if (!d_can_lookup(dentry
)) {
1926 nd
->path
= f
.file
->f_path
;
1927 if (flags
& LOOKUP_RCU
) {
1928 if (f
.flags
& FDPUT_FPUT
)
1930 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1933 path_get(&nd
->path
);
1938 nd
->inode
= nd
->path
.dentry
->d_inode
;
1939 if (!(flags
& LOOKUP_RCU
))
1941 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1943 if (!(nd
->flags
& LOOKUP_ROOT
))
1944 nd
->root
.mnt
= NULL
;
1948 current
->total_link_count
= 0;
1949 return link_path_walk(name
, nd
);
1952 static void path_cleanup(struct nameidata
*nd
)
1954 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
1955 path_put(&nd
->root
);
1956 nd
->root
.mnt
= NULL
;
1958 if (unlikely(nd
->base
))
1962 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1964 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1965 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1967 nd
->flags
&= ~LOOKUP_PARENT
;
1968 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1971 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1972 static int path_lookupat(int dfd
, const struct filename
*name
,
1973 unsigned int flags
, struct nameidata
*nd
)
1979 * Path walking is largely split up into 2 different synchronisation
1980 * schemes, rcu-walk and ref-walk (explained in
1981 * Documentation/filesystems/path-lookup.txt). These share much of the
1982 * path walk code, but some things particularly setup, cleanup, and
1983 * following mounts are sufficiently divergent that functions are
1984 * duplicated. Typically there is a function foo(), and its RCU
1985 * analogue, foo_rcu().
1987 * -ECHILD is the error number of choice (just to avoid clashes) that
1988 * is returned if some aspect of an rcu-walk fails. Such an error must
1989 * be handled by restarting a traditional ref-walk (which will always
1990 * be able to complete).
1992 err
= path_init(dfd
, name
->name
, flags
, nd
);
1993 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1994 err
= lookup_last(nd
, &path
);
1997 struct path link
= path
;
1998 err
= may_follow_link(&link
, nd
);
2001 nd
->flags
|= LOOKUP_PARENT
;
2002 err
= follow_link(&link
, nd
, &cookie
);
2005 err
= lookup_last(nd
, &path
);
2006 put_link(nd
, &link
, cookie
);
2011 err
= complete_walk(nd
);
2013 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2014 if (!d_can_lookup(nd
->path
.dentry
)) {
2015 path_put(&nd
->path
);
2024 static int filename_lookup(int dfd
, struct filename
*name
,
2025 unsigned int flags
, struct nameidata
*nd
)
2027 int retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_RCU
, nd
);
2028 if (unlikely(retval
== -ECHILD
))
2029 retval
= path_lookupat(dfd
, name
, flags
, nd
);
2030 if (unlikely(retval
== -ESTALE
))
2031 retval
= path_lookupat(dfd
, name
, flags
| LOOKUP_REVAL
, nd
);
2033 if (likely(!retval
))
2034 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2038 static int do_path_lookup(int dfd
, const char *name
,
2039 unsigned int flags
, struct nameidata
*nd
)
2041 struct filename
*filename
= getname_kernel(name
);
2042 int retval
= PTR_ERR(filename
);
2044 if (!IS_ERR(filename
)) {
2045 retval
= filename_lookup(dfd
, filename
, flags
, nd
);
2051 /* does lookup, returns the object with parent locked */
2052 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2054 struct filename
*filename
= getname_kernel(name
);
2055 struct nameidata nd
;
2059 if (IS_ERR(filename
))
2060 return ERR_CAST(filename
);
2062 err
= filename_lookup(AT_FDCWD
, filename
, LOOKUP_PARENT
, &nd
);
2067 if (nd
.last_type
!= LAST_NORM
) {
2069 d
= ERR_PTR(-EINVAL
);
2072 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2073 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2075 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2085 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2087 struct nameidata nd
;
2088 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2093 EXPORT_SYMBOL(kern_path
);
2096 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2097 * @dentry: pointer to dentry of the base directory
2098 * @mnt: pointer to vfs mount of the base directory
2099 * @name: pointer to file name
2100 * @flags: lookup flags
2101 * @path: pointer to struct path to fill
2103 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2104 const char *name
, unsigned int flags
,
2107 struct nameidata nd
;
2109 nd
.root
.dentry
= dentry
;
2111 BUG_ON(flags
& LOOKUP_PARENT
);
2112 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2113 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2118 EXPORT_SYMBOL(vfs_path_lookup
);
2121 * Restricted form of lookup. Doesn't follow links, single-component only,
2122 * needs parent already locked. Doesn't follow mounts.
2125 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2127 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2131 * lookup_one_len - filesystem helper to lookup single pathname component
2132 * @name: pathname component to lookup
2133 * @base: base directory to lookup from
2134 * @len: maximum length @len should be interpreted to
2136 * Note that this routine is purely a helper for filesystem usage and should
2137 * not be called by generic code. Also note that by using this function the
2138 * nameidata argument is passed to the filesystem methods and a filesystem
2139 * using this helper needs to be prepared for that.
2141 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2147 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2151 this.hash
= full_name_hash(name
, len
);
2153 return ERR_PTR(-EACCES
);
2155 if (unlikely(name
[0] == '.')) {
2156 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2157 return ERR_PTR(-EACCES
);
2161 c
= *(const unsigned char *)name
++;
2162 if (c
== '/' || c
== '\0')
2163 return ERR_PTR(-EACCES
);
2166 * See if the low-level filesystem might want
2167 * to use its own hash..
2169 if (base
->d_flags
& DCACHE_OP_HASH
) {
2170 int err
= base
->d_op
->d_hash(base
, &this);
2172 return ERR_PTR(err
);
2175 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2177 return ERR_PTR(err
);
2179 return __lookup_hash(&this, base
, 0);
2181 EXPORT_SYMBOL(lookup_one_len
);
2183 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2184 struct path
*path
, int *empty
)
2186 struct nameidata nd
;
2187 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2188 int err
= PTR_ERR(tmp
);
2191 BUG_ON(flags
& LOOKUP_PARENT
);
2193 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2201 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2204 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2206 EXPORT_SYMBOL(user_path_at
);
2209 * NB: most callers don't do anything directly with the reference to the
2210 * to struct filename, but the nd->last pointer points into the name string
2211 * allocated by getname. So we must hold the reference to it until all
2212 * path-walking is complete.
2214 static struct filename
*
2215 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2218 struct filename
*s
= getname(path
);
2221 /* only LOOKUP_REVAL is allowed in extra flags */
2222 flags
&= LOOKUP_REVAL
;
2227 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2230 return ERR_PTR(error
);
2237 * mountpoint_last - look up last component for umount
2238 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2239 * @path: pointer to container for result
2241 * This is a special lookup_last function just for umount. In this case, we
2242 * need to resolve the path without doing any revalidation.
2244 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2245 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2246 * in almost all cases, this lookup will be served out of the dcache. The only
2247 * cases where it won't are if nd->last refers to a symlink or the path is
2248 * bogus and it doesn't exist.
2251 * -error: if there was an error during lookup. This includes -ENOENT if the
2252 * lookup found a negative dentry. The nd->path reference will also be
2255 * 0: if we successfully resolved nd->path and found it to not to be a
2256 * symlink that needs to be followed. "path" will also be populated.
2257 * The nd->path reference will also be put.
2259 * 1: if we successfully resolved nd->last and found it to be a symlink
2260 * that needs to be followed. "path" will be populated with the path
2261 * to the link, and nd->path will *not* be put.
2264 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2267 struct dentry
*dentry
;
2268 struct dentry
*dir
= nd
->path
.dentry
;
2270 /* If we're in rcuwalk, drop out of it to handle last component */
2271 if (nd
->flags
& LOOKUP_RCU
) {
2272 if (unlazy_walk(nd
, NULL
)) {
2278 nd
->flags
&= ~LOOKUP_PARENT
;
2280 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2281 error
= handle_dots(nd
, nd
->last_type
);
2284 dentry
= dget(nd
->path
.dentry
);
2288 mutex_lock(&dir
->d_inode
->i_mutex
);
2289 dentry
= d_lookup(dir
, &nd
->last
);
2292 * No cached dentry. Mounted dentries are pinned in the cache,
2293 * so that means that this dentry is probably a symlink or the
2294 * path doesn't actually point to a mounted dentry.
2296 dentry
= d_alloc(dir
, &nd
->last
);
2299 mutex_unlock(&dir
->d_inode
->i_mutex
);
2302 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2303 error
= PTR_ERR(dentry
);
2304 if (IS_ERR(dentry
)) {
2305 mutex_unlock(&dir
->d_inode
->i_mutex
);
2309 mutex_unlock(&dir
->d_inode
->i_mutex
);
2312 if (!dentry
->d_inode
|| d_is_negative(dentry
)) {
2317 path
->dentry
= dentry
;
2318 path
->mnt
= nd
->path
.mnt
;
2319 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
))
2330 * path_mountpoint - look up a path to be umounted
2331 * @dfd: directory file descriptor to start walk from
2332 * @name: full pathname to walk
2333 * @path: pointer to container for result
2334 * @flags: lookup flags
2336 * Look up the given name, but don't attempt to revalidate the last component.
2337 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2340 path_mountpoint(int dfd
, const char *name
, struct path
*path
, unsigned int flags
)
2342 struct nameidata nd
;
2345 err
= path_init(dfd
, name
, flags
, &nd
);
2349 err
= mountpoint_last(&nd
, path
);
2352 struct path link
= *path
;
2353 err
= may_follow_link(&link
, &nd
);
2356 nd
.flags
|= LOOKUP_PARENT
;
2357 err
= follow_link(&link
, &nd
, &cookie
);
2360 err
= mountpoint_last(&nd
, path
);
2361 put_link(&nd
, &link
, cookie
);
2369 filename_mountpoint(int dfd
, struct filename
*s
, struct path
*path
,
2375 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_RCU
);
2376 if (unlikely(error
== -ECHILD
))
2377 error
= path_mountpoint(dfd
, s
->name
, path
, flags
);
2378 if (unlikely(error
== -ESTALE
))
2379 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_REVAL
);
2381 audit_inode(s
, path
->dentry
, 0);
2387 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2388 * @dfd: directory file descriptor
2389 * @name: pathname from userland
2390 * @flags: lookup flags
2391 * @path: pointer to container to hold result
2393 * A umount is a special case for path walking. We're not actually interested
2394 * in the inode in this situation, and ESTALE errors can be a problem. We
2395 * simply want track down the dentry and vfsmount attached at the mountpoint
2396 * and avoid revalidating the last component.
2398 * Returns 0 and populates "path" on success.
2401 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2404 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2408 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2411 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2413 EXPORT_SYMBOL(kern_path_mountpoint
);
2415 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2417 kuid_t fsuid
= current_fsuid();
2419 if (uid_eq(inode
->i_uid
, fsuid
))
2421 if (uid_eq(dir
->i_uid
, fsuid
))
2423 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2425 EXPORT_SYMBOL(__check_sticky
);
2428 * Check whether we can remove a link victim from directory dir, check
2429 * whether the type of victim is right.
2430 * 1. We can't do it if dir is read-only (done in permission())
2431 * 2. We should have write and exec permissions on dir
2432 * 3. We can't remove anything from append-only dir
2433 * 4. We can't do anything with immutable dir (done in permission())
2434 * 5. If the sticky bit on dir is set we should either
2435 * a. be owner of dir, or
2436 * b. be owner of victim, or
2437 * c. have CAP_FOWNER capability
2438 * 6. If the victim is append-only or immutable we can't do antyhing with
2439 * links pointing to it.
2440 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2441 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2442 * 9. We can't remove a root or mountpoint.
2443 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2444 * nfs_async_unlink().
2446 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2448 struct inode
*inode
= victim
->d_inode
;
2451 if (d_is_negative(victim
))
2455 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2456 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2458 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2464 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2465 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2468 if (!d_is_dir(victim
))
2470 if (IS_ROOT(victim
))
2472 } else if (d_is_dir(victim
))
2474 if (IS_DEADDIR(dir
))
2476 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2481 /* Check whether we can create an object with dentry child in directory
2483 * 1. We can't do it if child already exists (open has special treatment for
2484 * this case, but since we are inlined it's OK)
2485 * 2. We can't do it if dir is read-only (done in permission())
2486 * 3. We should have write and exec permissions on dir
2487 * 4. We can't do it if dir is immutable (done in permission())
2489 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2491 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2494 if (IS_DEADDIR(dir
))
2496 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2500 * p1 and p2 should be directories on the same fs.
2502 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2507 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2511 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2513 p
= d_ancestor(p2
, p1
);
2515 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2516 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2520 p
= d_ancestor(p1
, p2
);
2522 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2523 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2527 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2528 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2531 EXPORT_SYMBOL(lock_rename
);
2533 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2535 mutex_unlock(&p1
->d_inode
->i_mutex
);
2537 mutex_unlock(&p2
->d_inode
->i_mutex
);
2538 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2541 EXPORT_SYMBOL(unlock_rename
);
2543 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2546 int error
= may_create(dir
, dentry
);
2550 if (!dir
->i_op
->create
)
2551 return -EACCES
; /* shouldn't it be ENOSYS? */
2554 error
= security_inode_create(dir
, dentry
, mode
);
2557 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2559 fsnotify_create(dir
, dentry
);
2562 EXPORT_SYMBOL(vfs_create
);
2564 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2566 struct dentry
*dentry
= path
->dentry
;
2567 struct inode
*inode
= dentry
->d_inode
;
2577 switch (inode
->i_mode
& S_IFMT
) {
2581 if (acc_mode
& MAY_WRITE
)
2586 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2595 error
= inode_permission(inode
, acc_mode
);
2600 * An append-only file must be opened in append mode for writing.
2602 if (IS_APPEND(inode
)) {
2603 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2609 /* O_NOATIME can only be set by the owner or superuser */
2610 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2616 static int handle_truncate(struct file
*filp
)
2618 struct path
*path
= &filp
->f_path
;
2619 struct inode
*inode
= path
->dentry
->d_inode
;
2620 int error
= get_write_access(inode
);
2624 * Refuse to truncate files with mandatory locks held on them.
2626 error
= locks_verify_locked(filp
);
2628 error
= security_path_truncate(path
);
2630 error
= do_truncate(path
->dentry
, 0,
2631 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2634 put_write_access(inode
);
2638 static inline int open_to_namei_flags(int flag
)
2640 if ((flag
& O_ACCMODE
) == 3)
2645 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2647 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2651 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2655 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2659 * Attempt to atomically look up, create and open a file from a negative
2662 * Returns 0 if successful. The file will have been created and attached to
2663 * @file by the filesystem calling finish_open().
2665 * Returns 1 if the file was looked up only or didn't need creating. The
2666 * caller will need to perform the open themselves. @path will have been
2667 * updated to point to the new dentry. This may be negative.
2669 * Returns an error code otherwise.
2671 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2672 struct path
*path
, struct file
*file
,
2673 const struct open_flags
*op
,
2674 bool got_write
, bool need_lookup
,
2677 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2678 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2682 int create_error
= 0;
2683 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2686 BUG_ON(dentry
->d_inode
);
2688 /* Don't create child dentry for a dead directory. */
2689 if (unlikely(IS_DEADDIR(dir
))) {
2695 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2696 mode
&= ~current_umask();
2698 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2700 open_flag
&= ~O_TRUNC
;
2703 * Checking write permission is tricky, bacuse we don't know if we are
2704 * going to actually need it: O_CREAT opens should work as long as the
2705 * file exists. But checking existence breaks atomicity. The trick is
2706 * to check access and if not granted clear O_CREAT from the flags.
2708 * Another problem is returing the "right" error value (e.g. for an
2709 * O_EXCL open we want to return EEXIST not EROFS).
2711 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2712 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2713 if (!(open_flag
& O_CREAT
)) {
2715 * No O_CREATE -> atomicity not a requirement -> fall
2716 * back to lookup + open
2719 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2720 /* Fall back and fail with the right error */
2721 create_error
= -EROFS
;
2724 /* No side effects, safe to clear O_CREAT */
2725 create_error
= -EROFS
;
2726 open_flag
&= ~O_CREAT
;
2730 if (open_flag
& O_CREAT
) {
2731 error
= may_o_create(&nd
->path
, dentry
, mode
);
2733 create_error
= error
;
2734 if (open_flag
& O_EXCL
)
2736 open_flag
&= ~O_CREAT
;
2740 if (nd
->flags
& LOOKUP_DIRECTORY
)
2741 open_flag
|= O_DIRECTORY
;
2743 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2744 file
->f_path
.mnt
= nd
->path
.mnt
;
2745 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2748 if (create_error
&& error
== -ENOENT
)
2749 error
= create_error
;
2753 if (error
) { /* returned 1, that is */
2754 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2758 if (file
->f_path
.dentry
) {
2760 dentry
= file
->f_path
.dentry
;
2762 if (*opened
& FILE_CREATED
)
2763 fsnotify_create(dir
, dentry
);
2764 if (!dentry
->d_inode
) {
2765 WARN_ON(*opened
& FILE_CREATED
);
2767 error
= create_error
;
2771 if (excl
&& !(*opened
& FILE_CREATED
)) {
2780 * We didn't have the inode before the open, so check open permission
2783 acc_mode
= op
->acc_mode
;
2784 if (*opened
& FILE_CREATED
) {
2785 WARN_ON(!(open_flag
& O_CREAT
));
2786 fsnotify_create(dir
, dentry
);
2787 acc_mode
= MAY_OPEN
;
2789 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2799 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2801 return PTR_ERR(dentry
);
2804 int open_flag
= op
->open_flag
;
2806 error
= create_error
;
2807 if ((open_flag
& O_EXCL
)) {
2808 if (!dentry
->d_inode
)
2810 } else if (!dentry
->d_inode
) {
2812 } else if ((open_flag
& O_TRUNC
) &&
2816 /* will fail later, go on to get the right error */
2820 path
->dentry
= dentry
;
2821 path
->mnt
= nd
->path
.mnt
;
2826 * Look up and maybe create and open the last component.
2828 * Must be called with i_mutex held on parent.
2830 * Returns 0 if the file was successfully atomically created (if necessary) and
2831 * opened. In this case the file will be returned attached to @file.
2833 * Returns 1 if the file was not completely opened at this time, though lookups
2834 * and creations will have been performed and the dentry returned in @path will
2835 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2836 * specified then a negative dentry may be returned.
2838 * An error code is returned otherwise.
2840 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2841 * cleared otherwise prior to returning.
2843 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2845 const struct open_flags
*op
,
2846 bool got_write
, int *opened
)
2848 struct dentry
*dir
= nd
->path
.dentry
;
2849 struct inode
*dir_inode
= dir
->d_inode
;
2850 struct dentry
*dentry
;
2854 *opened
&= ~FILE_CREATED
;
2855 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2857 return PTR_ERR(dentry
);
2859 /* Cached positive dentry: will open in f_op->open */
2860 if (!need_lookup
&& dentry
->d_inode
)
2863 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2864 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2865 need_lookup
, opened
);
2869 BUG_ON(dentry
->d_inode
);
2871 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2873 return PTR_ERR(dentry
);
2876 /* Negative dentry, just create the file */
2877 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2878 umode_t mode
= op
->mode
;
2879 if (!IS_POSIXACL(dir
->d_inode
))
2880 mode
&= ~current_umask();
2882 * This write is needed to ensure that a
2883 * rw->ro transition does not occur between
2884 * the time when the file is created and when
2885 * a permanent write count is taken through
2886 * the 'struct file' in finish_open().
2892 *opened
|= FILE_CREATED
;
2893 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2896 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2897 nd
->flags
& LOOKUP_EXCL
);
2902 path
->dentry
= dentry
;
2903 path
->mnt
= nd
->path
.mnt
;
2912 * Handle the last step of open()
2914 static int do_last(struct nameidata
*nd
, struct path
*path
,
2915 struct file
*file
, const struct open_flags
*op
,
2916 int *opened
, struct filename
*name
)
2918 struct dentry
*dir
= nd
->path
.dentry
;
2919 int open_flag
= op
->open_flag
;
2920 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2921 bool got_write
= false;
2922 int acc_mode
= op
->acc_mode
;
2923 struct inode
*inode
;
2924 bool symlink_ok
= false;
2925 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2926 bool retried
= false;
2929 nd
->flags
&= ~LOOKUP_PARENT
;
2930 nd
->flags
|= op
->intent
;
2932 if (nd
->last_type
!= LAST_NORM
) {
2933 error
= handle_dots(nd
, nd
->last_type
);
2939 if (!(open_flag
& O_CREAT
)) {
2940 if (nd
->last
.name
[nd
->last
.len
])
2941 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2942 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2944 /* we _can_ be in RCU mode here */
2945 error
= lookup_fast(nd
, path
, &inode
);
2952 BUG_ON(nd
->inode
!= dir
->d_inode
);
2954 /* create side of things */
2956 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2957 * has been cleared when we got to the last component we are
2960 error
= complete_walk(nd
);
2964 audit_inode(name
, dir
, LOOKUP_PARENT
);
2966 /* trailing slashes? */
2967 if (nd
->last
.name
[nd
->last
.len
])
2972 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2973 error
= mnt_want_write(nd
->path
.mnt
);
2977 * do _not_ fail yet - we might not need that or fail with
2978 * a different error; let lookup_open() decide; we'll be
2979 * dropping this one anyway.
2982 mutex_lock(&dir
->d_inode
->i_mutex
);
2983 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2984 mutex_unlock(&dir
->d_inode
->i_mutex
);
2990 if ((*opened
& FILE_CREATED
) ||
2991 !S_ISREG(file_inode(file
)->i_mode
))
2992 will_truncate
= false;
2994 audit_inode(name
, file
->f_path
.dentry
, 0);
2998 if (*opened
& FILE_CREATED
) {
2999 /* Don't check for write permission, don't truncate */
3000 open_flag
&= ~O_TRUNC
;
3001 will_truncate
= false;
3002 acc_mode
= MAY_OPEN
;
3003 path_to_nameidata(path
, nd
);
3004 goto finish_open_created
;
3008 * create/update audit record if it already exists.
3010 if (d_is_positive(path
->dentry
))
3011 audit_inode(name
, path
->dentry
, 0);
3014 * If atomic_open() acquired write access it is dropped now due to
3015 * possible mount and symlink following (this might be optimized away if
3019 mnt_drop_write(nd
->path
.mnt
);
3024 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3027 error
= follow_managed(path
, nd
->flags
);
3032 nd
->flags
|= LOOKUP_JUMPED
;
3034 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3035 inode
= path
->dentry
->d_inode
;
3037 /* we _can_ be in RCU mode here */
3039 if (!inode
|| d_is_negative(path
->dentry
)) {
3040 path_to_nameidata(path
, nd
);
3044 if (should_follow_link(path
->dentry
, !symlink_ok
)) {
3045 if (nd
->flags
& LOOKUP_RCU
) {
3046 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
3051 BUG_ON(inode
!= path
->dentry
->d_inode
);
3055 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3056 path_to_nameidata(path
, nd
);
3058 save_parent
.dentry
= nd
->path
.dentry
;
3059 save_parent
.mnt
= mntget(path
->mnt
);
3060 nd
->path
.dentry
= path
->dentry
;
3064 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3066 error
= complete_walk(nd
);
3068 path_put(&save_parent
);
3071 audit_inode(name
, nd
->path
.dentry
, 0);
3073 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3076 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3078 if (!S_ISREG(nd
->inode
->i_mode
))
3079 will_truncate
= false;
3081 if (will_truncate
) {
3082 error
= mnt_want_write(nd
->path
.mnt
);
3087 finish_open_created
:
3088 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3092 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3093 error
= vfs_open(&nd
->path
, file
, current_cred());
3095 *opened
|= FILE_OPENED
;
3097 if (error
== -EOPENSTALE
)
3102 error
= open_check_o_direct(file
);
3105 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3109 if (will_truncate
) {
3110 error
= handle_truncate(file
);
3116 mnt_drop_write(nd
->path
.mnt
);
3117 path_put(&save_parent
);
3122 path_put_conditional(path
, nd
);
3129 /* If no saved parent or already retried then can't retry */
3130 if (!save_parent
.dentry
|| retried
)
3133 BUG_ON(save_parent
.dentry
!= dir
);
3134 path_put(&nd
->path
);
3135 nd
->path
= save_parent
;
3136 nd
->inode
= dir
->d_inode
;
3137 save_parent
.mnt
= NULL
;
3138 save_parent
.dentry
= NULL
;
3140 mnt_drop_write(nd
->path
.mnt
);
3147 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3148 struct nameidata
*nd
, int flags
,
3149 const struct open_flags
*op
,
3150 struct file
*file
, int *opened
)
3152 static const struct qstr name
= QSTR_INIT("/", 1);
3153 struct dentry
*dentry
, *child
;
3155 int error
= path_lookupat(dfd
, pathname
,
3156 flags
| LOOKUP_DIRECTORY
, nd
);
3157 if (unlikely(error
))
3159 error
= mnt_want_write(nd
->path
.mnt
);
3160 if (unlikely(error
))
3162 /* we want directory to be writable */
3163 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3166 dentry
= nd
->path
.dentry
;
3167 dir
= dentry
->d_inode
;
3168 if (!dir
->i_op
->tmpfile
) {
3169 error
= -EOPNOTSUPP
;
3172 child
= d_alloc(dentry
, &name
);
3173 if (unlikely(!child
)) {
3177 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3178 nd
->flags
|= op
->intent
;
3179 dput(nd
->path
.dentry
);
3180 nd
->path
.dentry
= child
;
3181 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3184 audit_inode(pathname
, nd
->path
.dentry
, 0);
3185 /* Don't check for other permissions, the inode was just created */
3186 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3189 file
->f_path
.mnt
= nd
->path
.mnt
;
3190 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3193 error
= open_check_o_direct(file
);
3196 } else if (!(op
->open_flag
& O_EXCL
)) {
3197 struct inode
*inode
= file_inode(file
);
3198 spin_lock(&inode
->i_lock
);
3199 inode
->i_state
|= I_LINKABLE
;
3200 spin_unlock(&inode
->i_lock
);
3203 mnt_drop_write(nd
->path
.mnt
);
3205 path_put(&nd
->path
);
3209 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3210 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3217 file
= get_empty_filp();
3221 file
->f_flags
= op
->open_flag
;
3223 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3224 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3228 error
= path_init(dfd
, pathname
->name
, flags
, nd
);
3229 if (unlikely(error
))
3232 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3233 while (unlikely(error
> 0)) { /* trailing symlink */
3234 struct path link
= path
;
3236 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3237 path_put_conditional(&path
, nd
);
3238 path_put(&nd
->path
);
3242 error
= may_follow_link(&link
, nd
);
3243 if (unlikely(error
))
3245 nd
->flags
|= LOOKUP_PARENT
;
3246 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3247 error
= follow_link(&link
, nd
, &cookie
);
3248 if (unlikely(error
))
3250 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3251 put_link(nd
, &link
, cookie
);
3255 if (!(opened
& FILE_OPENED
)) {
3259 if (unlikely(error
)) {
3260 if (error
== -EOPENSTALE
) {
3261 if (flags
& LOOKUP_RCU
)
3266 file
= ERR_PTR(error
);
3271 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3272 const struct open_flags
*op
)
3274 struct nameidata nd
;
3275 int flags
= op
->lookup_flags
;
3278 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3279 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3280 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3281 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3282 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3286 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3287 const char *name
, const struct open_flags
*op
)
3289 struct nameidata nd
;
3291 struct filename
*filename
;
3292 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3295 nd
.root
.dentry
= dentry
;
3297 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3298 return ERR_PTR(-ELOOP
);
3300 filename
= getname_kernel(name
);
3301 if (unlikely(IS_ERR(filename
)))
3302 return ERR_CAST(filename
);
3304 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3305 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3306 file
= path_openat(-1, filename
, &nd
, op
, flags
);
3307 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3308 file
= path_openat(-1, filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3313 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3314 struct path
*path
, unsigned int lookup_flags
)
3316 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3317 struct nameidata nd
;
3320 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3323 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3324 * other flags passed in are ignored!
3326 lookup_flags
&= LOOKUP_REVAL
;
3328 error
= filename_lookup(dfd
, name
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3330 return ERR_PTR(error
);
3333 * Yucky last component or no last component at all?
3334 * (foo/., foo/.., /////)
3336 if (nd
.last_type
!= LAST_NORM
)
3338 nd
.flags
&= ~LOOKUP_PARENT
;
3339 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3341 /* don't fail immediately if it's r/o, at least try to report other errors */
3342 err2
= mnt_want_write(nd
.path
.mnt
);
3344 * Do the final lookup.
3346 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3347 dentry
= lookup_hash(&nd
);
3352 if (d_is_positive(dentry
))
3356 * Special case - lookup gave negative, but... we had foo/bar/
3357 * From the vfs_mknod() POV we just have a negative dentry -
3358 * all is fine. Let's be bastards - you had / on the end, you've
3359 * been asking for (non-existent) directory. -ENOENT for you.
3361 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3365 if (unlikely(err2
)) {
3373 dentry
= ERR_PTR(error
);
3375 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3377 mnt_drop_write(nd
.path
.mnt
);
3383 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3384 struct path
*path
, unsigned int lookup_flags
)
3386 struct filename
*filename
= getname_kernel(pathname
);
3389 if (IS_ERR(filename
))
3390 return ERR_CAST(filename
);
3391 res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3395 EXPORT_SYMBOL(kern_path_create
);
3397 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3400 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3401 mnt_drop_write(path
->mnt
);
3404 EXPORT_SYMBOL(done_path_create
);
3406 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3407 struct path
*path
, unsigned int lookup_flags
)
3409 struct filename
*tmp
= getname(pathname
);
3412 return ERR_CAST(tmp
);
3413 res
= filename_create(dfd
, tmp
, path
, lookup_flags
);
3417 EXPORT_SYMBOL(user_path_create
);
3419 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3421 int error
= may_create(dir
, dentry
);
3426 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3429 if (!dir
->i_op
->mknod
)
3432 error
= devcgroup_inode_mknod(mode
, dev
);
3436 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3440 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3442 fsnotify_create(dir
, dentry
);
3445 EXPORT_SYMBOL(vfs_mknod
);
3447 static int may_mknod(umode_t mode
)
3449 switch (mode
& S_IFMT
) {
3455 case 0: /* zero mode translates to S_IFREG */
3464 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3467 struct dentry
*dentry
;
3470 unsigned int lookup_flags
= 0;
3472 error
= may_mknod(mode
);
3476 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3478 return PTR_ERR(dentry
);
3480 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3481 mode
&= ~current_umask();
3482 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3485 switch (mode
& S_IFMT
) {
3486 case 0: case S_IFREG
:
3487 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3489 case S_IFCHR
: case S_IFBLK
:
3490 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3491 new_decode_dev(dev
));
3493 case S_IFIFO
: case S_IFSOCK
:
3494 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3498 done_path_create(&path
, dentry
);
3499 if (retry_estale(error
, lookup_flags
)) {
3500 lookup_flags
|= LOOKUP_REVAL
;
3506 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3508 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3511 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3513 int error
= may_create(dir
, dentry
);
3514 unsigned max_links
= dir
->i_sb
->s_max_links
;
3519 if (!dir
->i_op
->mkdir
)
3522 mode
&= (S_IRWXUGO
|S_ISVTX
);
3523 error
= security_inode_mkdir(dir
, dentry
, mode
);
3527 if (max_links
&& dir
->i_nlink
>= max_links
)
3530 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3532 fsnotify_mkdir(dir
, dentry
);
3535 EXPORT_SYMBOL(vfs_mkdir
);
3537 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3539 struct dentry
*dentry
;
3542 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3545 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3547 return PTR_ERR(dentry
);
3549 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3550 mode
&= ~current_umask();
3551 error
= security_path_mkdir(&path
, dentry
, mode
);
3553 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3554 done_path_create(&path
, dentry
);
3555 if (retry_estale(error
, lookup_flags
)) {
3556 lookup_flags
|= LOOKUP_REVAL
;
3562 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3564 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3568 * The dentry_unhash() helper will try to drop the dentry early: we
3569 * should have a usage count of 1 if we're the only user of this
3570 * dentry, and if that is true (possibly after pruning the dcache),
3571 * then we drop the dentry now.
3573 * A low-level filesystem can, if it choses, legally
3576 * if (!d_unhashed(dentry))
3579 * if it cannot handle the case of removing a directory
3580 * that is still in use by something else..
3582 void dentry_unhash(struct dentry
*dentry
)
3584 shrink_dcache_parent(dentry
);
3585 spin_lock(&dentry
->d_lock
);
3586 if (dentry
->d_lockref
.count
== 1)
3588 spin_unlock(&dentry
->d_lock
);
3590 EXPORT_SYMBOL(dentry_unhash
);
3592 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3594 int error
= may_delete(dir
, dentry
, 1);
3599 if (!dir
->i_op
->rmdir
)
3603 mutex_lock(&dentry
->d_inode
->i_mutex
);
3606 if (is_local_mountpoint(dentry
))
3609 error
= security_inode_rmdir(dir
, dentry
);
3613 shrink_dcache_parent(dentry
);
3614 error
= dir
->i_op
->rmdir(dir
, dentry
);
3618 dentry
->d_inode
->i_flags
|= S_DEAD
;
3620 detach_mounts(dentry
);
3623 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3629 EXPORT_SYMBOL(vfs_rmdir
);
3631 static long do_rmdir(int dfd
, const char __user
*pathname
)
3634 struct filename
*name
;
3635 struct dentry
*dentry
;
3636 struct nameidata nd
;
3637 unsigned int lookup_flags
= 0;
3639 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3641 return PTR_ERR(name
);
3643 switch(nd
.last_type
) {
3655 nd
.flags
&= ~LOOKUP_PARENT
;
3656 error
= mnt_want_write(nd
.path
.mnt
);
3660 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3661 dentry
= lookup_hash(&nd
);
3662 error
= PTR_ERR(dentry
);
3665 if (!dentry
->d_inode
) {
3669 error
= security_path_rmdir(&nd
.path
, dentry
);
3672 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3676 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3677 mnt_drop_write(nd
.path
.mnt
);
3681 if (retry_estale(error
, lookup_flags
)) {
3682 lookup_flags
|= LOOKUP_REVAL
;
3688 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3690 return do_rmdir(AT_FDCWD
, pathname
);
3694 * vfs_unlink - unlink a filesystem object
3695 * @dir: parent directory
3697 * @delegated_inode: returns victim inode, if the inode is delegated.
3699 * The caller must hold dir->i_mutex.
3701 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3702 * return a reference to the inode in delegated_inode. The caller
3703 * should then break the delegation on that inode and retry. Because
3704 * breaking a delegation may take a long time, the caller should drop
3705 * dir->i_mutex before doing so.
3707 * Alternatively, a caller may pass NULL for delegated_inode. This may
3708 * be appropriate for callers that expect the underlying filesystem not
3709 * to be NFS exported.
3711 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3713 struct inode
*target
= dentry
->d_inode
;
3714 int error
= may_delete(dir
, dentry
, 0);
3719 if (!dir
->i_op
->unlink
)
3722 mutex_lock(&target
->i_mutex
);
3723 if (is_local_mountpoint(dentry
))
3726 error
= security_inode_unlink(dir
, dentry
);
3728 error
= try_break_deleg(target
, delegated_inode
);
3731 error
= dir
->i_op
->unlink(dir
, dentry
);
3734 detach_mounts(dentry
);
3739 mutex_unlock(&target
->i_mutex
);
3741 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3742 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3743 fsnotify_link_count(target
);
3749 EXPORT_SYMBOL(vfs_unlink
);
3752 * Make sure that the actual truncation of the file will occur outside its
3753 * directory's i_mutex. Truncate can take a long time if there is a lot of
3754 * writeout happening, and we don't want to prevent access to the directory
3755 * while waiting on the I/O.
3757 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3760 struct filename
*name
;
3761 struct dentry
*dentry
;
3762 struct nameidata nd
;
3763 struct inode
*inode
= NULL
;
3764 struct inode
*delegated_inode
= NULL
;
3765 unsigned int lookup_flags
= 0;
3767 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3769 return PTR_ERR(name
);
3772 if (nd
.last_type
!= LAST_NORM
)
3775 nd
.flags
&= ~LOOKUP_PARENT
;
3776 error
= mnt_want_write(nd
.path
.mnt
);
3780 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3781 dentry
= lookup_hash(&nd
);
3782 error
= PTR_ERR(dentry
);
3783 if (!IS_ERR(dentry
)) {
3784 /* Why not before? Because we want correct error value */
3785 if (nd
.last
.name
[nd
.last
.len
])
3787 inode
= dentry
->d_inode
;
3788 if (d_is_negative(dentry
))
3791 error
= security_path_unlink(&nd
.path
, dentry
);
3794 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3798 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3800 iput(inode
); /* truncate the inode here */
3802 if (delegated_inode
) {
3803 error
= break_deleg_wait(&delegated_inode
);
3807 mnt_drop_write(nd
.path
.mnt
);
3811 if (retry_estale(error
, lookup_flags
)) {
3812 lookup_flags
|= LOOKUP_REVAL
;
3819 if (d_is_negative(dentry
))
3821 else if (d_is_dir(dentry
))
3828 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3830 if ((flag
& ~AT_REMOVEDIR
) != 0)
3833 if (flag
& AT_REMOVEDIR
)
3834 return do_rmdir(dfd
, pathname
);
3836 return do_unlinkat(dfd
, pathname
);
3839 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3841 return do_unlinkat(AT_FDCWD
, pathname
);
3844 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3846 int error
= may_create(dir
, dentry
);
3851 if (!dir
->i_op
->symlink
)
3854 error
= security_inode_symlink(dir
, dentry
, oldname
);
3858 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3860 fsnotify_create(dir
, dentry
);
3863 EXPORT_SYMBOL(vfs_symlink
);
3865 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3866 int, newdfd
, const char __user
*, newname
)
3869 struct filename
*from
;
3870 struct dentry
*dentry
;
3872 unsigned int lookup_flags
= 0;
3874 from
= getname(oldname
);
3876 return PTR_ERR(from
);
3878 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3879 error
= PTR_ERR(dentry
);
3883 error
= security_path_symlink(&path
, dentry
, from
->name
);
3885 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3886 done_path_create(&path
, dentry
);
3887 if (retry_estale(error
, lookup_flags
)) {
3888 lookup_flags
|= LOOKUP_REVAL
;
3896 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3898 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3902 * vfs_link - create a new link
3903 * @old_dentry: object to be linked
3905 * @new_dentry: where to create the new link
3906 * @delegated_inode: returns inode needing a delegation break
3908 * The caller must hold dir->i_mutex
3910 * If vfs_link discovers a delegation on the to-be-linked file in need
3911 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3912 * inode in delegated_inode. The caller should then break the delegation
3913 * and retry. Because breaking a delegation may take a long time, the
3914 * caller should drop the i_mutex before doing so.
3916 * Alternatively, a caller may pass NULL for delegated_inode. This may
3917 * be appropriate for callers that expect the underlying filesystem not
3918 * to be NFS exported.
3920 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3922 struct inode
*inode
= old_dentry
->d_inode
;
3923 unsigned max_links
= dir
->i_sb
->s_max_links
;
3929 error
= may_create(dir
, new_dentry
);
3933 if (dir
->i_sb
!= inode
->i_sb
)
3937 * A link to an append-only or immutable file cannot be created.
3939 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3941 if (!dir
->i_op
->link
)
3943 if (S_ISDIR(inode
->i_mode
))
3946 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3950 mutex_lock(&inode
->i_mutex
);
3951 /* Make sure we don't allow creating hardlink to an unlinked file */
3952 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3954 else if (max_links
&& inode
->i_nlink
>= max_links
)
3957 error
= try_break_deleg(inode
, delegated_inode
);
3959 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3962 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3963 spin_lock(&inode
->i_lock
);
3964 inode
->i_state
&= ~I_LINKABLE
;
3965 spin_unlock(&inode
->i_lock
);
3967 mutex_unlock(&inode
->i_mutex
);
3969 fsnotify_link(dir
, inode
, new_dentry
);
3972 EXPORT_SYMBOL(vfs_link
);
3975 * Hardlinks are often used in delicate situations. We avoid
3976 * security-related surprises by not following symlinks on the
3979 * We don't follow them on the oldname either to be compatible
3980 * with linux 2.0, and to avoid hard-linking to directories
3981 * and other special files. --ADM
3983 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3984 int, newdfd
, const char __user
*, newname
, int, flags
)
3986 struct dentry
*new_dentry
;
3987 struct path old_path
, new_path
;
3988 struct inode
*delegated_inode
= NULL
;
3992 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3995 * To use null names we require CAP_DAC_READ_SEARCH
3996 * This ensures that not everyone will be able to create
3997 * handlink using the passed filedescriptor.
3999 if (flags
& AT_EMPTY_PATH
) {
4000 if (!capable(CAP_DAC_READ_SEARCH
))
4005 if (flags
& AT_SYMLINK_FOLLOW
)
4006 how
|= LOOKUP_FOLLOW
;
4008 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4012 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4013 (how
& LOOKUP_REVAL
));
4014 error
= PTR_ERR(new_dentry
);
4015 if (IS_ERR(new_dentry
))
4019 if (old_path
.mnt
!= new_path
.mnt
)
4021 error
= may_linkat(&old_path
);
4022 if (unlikely(error
))
4024 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4027 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4029 done_path_create(&new_path
, new_dentry
);
4030 if (delegated_inode
) {
4031 error
= break_deleg_wait(&delegated_inode
);
4033 path_put(&old_path
);
4037 if (retry_estale(error
, how
)) {
4038 path_put(&old_path
);
4039 how
|= LOOKUP_REVAL
;
4043 path_put(&old_path
);
4048 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4050 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4054 * vfs_rename - rename a filesystem object
4055 * @old_dir: parent of source
4056 * @old_dentry: source
4057 * @new_dir: parent of destination
4058 * @new_dentry: destination
4059 * @delegated_inode: returns an inode needing a delegation break
4060 * @flags: rename flags
4062 * The caller must hold multiple mutexes--see lock_rename()).
4064 * If vfs_rename discovers a delegation in need of breaking at either
4065 * the source or destination, it will return -EWOULDBLOCK and return a
4066 * reference to the inode in delegated_inode. The caller should then
4067 * break the delegation and retry. Because breaking a delegation may
4068 * take a long time, the caller should drop all locks before doing
4071 * Alternatively, a caller may pass NULL for delegated_inode. This may
4072 * be appropriate for callers that expect the underlying filesystem not
4073 * to be NFS exported.
4075 * The worst of all namespace operations - renaming directory. "Perverted"
4076 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4078 * a) we can get into loop creation.
4079 * b) race potential - two innocent renames can create a loop together.
4080 * That's where 4.4 screws up. Current fix: serialization on
4081 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4083 * c) we have to lock _four_ objects - parents and victim (if it exists),
4084 * and source (if it is not a directory).
4085 * And that - after we got ->i_mutex on parents (until then we don't know
4086 * whether the target exists). Solution: try to be smart with locking
4087 * order for inodes. We rely on the fact that tree topology may change
4088 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4089 * move will be locked. Thus we can rank directories by the tree
4090 * (ancestors first) and rank all non-directories after them.
4091 * That works since everybody except rename does "lock parent, lookup,
4092 * lock child" and rename is under ->s_vfs_rename_mutex.
4093 * HOWEVER, it relies on the assumption that any object with ->lookup()
4094 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4095 * we'd better make sure that there's no link(2) for them.
4096 * d) conversion from fhandle to dentry may come in the wrong moment - when
4097 * we are removing the target. Solution: we will have to grab ->i_mutex
4098 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4099 * ->i_mutex on parents, which works but leads to some truly excessive
4102 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4103 struct inode
*new_dir
, struct dentry
*new_dentry
,
4104 struct inode
**delegated_inode
, unsigned int flags
)
4107 bool is_dir
= d_is_dir(old_dentry
);
4108 const unsigned char *old_name
;
4109 struct inode
*source
= old_dentry
->d_inode
;
4110 struct inode
*target
= new_dentry
->d_inode
;
4111 bool new_is_dir
= false;
4112 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4114 if (source
== target
)
4117 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4122 error
= may_create(new_dir
, new_dentry
);
4124 new_is_dir
= d_is_dir(new_dentry
);
4126 if (!(flags
& RENAME_EXCHANGE
))
4127 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4129 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4134 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4137 if (flags
&& !old_dir
->i_op
->rename2
)
4141 * If we are going to change the parent - check write permissions,
4142 * we'll need to flip '..'.
4144 if (new_dir
!= old_dir
) {
4146 error
= inode_permission(source
, MAY_WRITE
);
4150 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4151 error
= inode_permission(target
, MAY_WRITE
);
4157 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4162 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4164 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4165 lock_two_nondirectories(source
, target
);
4167 mutex_lock(&target
->i_mutex
);
4170 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4173 if (max_links
&& new_dir
!= old_dir
) {
4175 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4177 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4178 old_dir
->i_nlink
>= max_links
)
4181 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4182 shrink_dcache_parent(new_dentry
);
4184 error
= try_break_deleg(source
, delegated_inode
);
4188 if (target
&& !new_is_dir
) {
4189 error
= try_break_deleg(target
, delegated_inode
);
4193 if (!old_dir
->i_op
->rename2
) {
4194 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4195 new_dir
, new_dentry
);
4197 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4198 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4199 new_dir
, new_dentry
, flags
);
4204 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4206 target
->i_flags
|= S_DEAD
;
4207 dont_mount(new_dentry
);
4208 detach_mounts(new_dentry
);
4210 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4211 if (!(flags
& RENAME_EXCHANGE
))
4212 d_move(old_dentry
, new_dentry
);
4214 d_exchange(old_dentry
, new_dentry
);
4217 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4218 unlock_two_nondirectories(source
, target
);
4220 mutex_unlock(&target
->i_mutex
);
4223 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4224 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4225 if (flags
& RENAME_EXCHANGE
) {
4226 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4227 new_is_dir
, NULL
, new_dentry
);
4230 fsnotify_oldname_free(old_name
);
4234 EXPORT_SYMBOL(vfs_rename
);
4236 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4237 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4239 struct dentry
*old_dir
, *new_dir
;
4240 struct dentry
*old_dentry
, *new_dentry
;
4241 struct dentry
*trap
;
4242 struct nameidata oldnd
, newnd
;
4243 struct inode
*delegated_inode
= NULL
;
4244 struct filename
*from
;
4245 struct filename
*to
;
4246 unsigned int lookup_flags
= 0;
4247 bool should_retry
= false;
4250 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4253 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4254 (flags
& RENAME_EXCHANGE
))
4257 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4261 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4263 error
= PTR_ERR(from
);
4267 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4269 error
= PTR_ERR(to
);
4274 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4277 old_dir
= oldnd
.path
.dentry
;
4279 if (oldnd
.last_type
!= LAST_NORM
)
4282 new_dir
= newnd
.path
.dentry
;
4283 if (flags
& RENAME_NOREPLACE
)
4285 if (newnd
.last_type
!= LAST_NORM
)
4288 error
= mnt_want_write(oldnd
.path
.mnt
);
4292 oldnd
.flags
&= ~LOOKUP_PARENT
;
4293 newnd
.flags
&= ~LOOKUP_PARENT
;
4294 if (!(flags
& RENAME_EXCHANGE
))
4295 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4298 trap
= lock_rename(new_dir
, old_dir
);
4300 old_dentry
= lookup_hash(&oldnd
);
4301 error
= PTR_ERR(old_dentry
);
4302 if (IS_ERR(old_dentry
))
4304 /* source must exist */
4306 if (d_is_negative(old_dentry
))
4308 new_dentry
= lookup_hash(&newnd
);
4309 error
= PTR_ERR(new_dentry
);
4310 if (IS_ERR(new_dentry
))
4313 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4315 if (flags
& RENAME_EXCHANGE
) {
4317 if (d_is_negative(new_dentry
))
4320 if (!d_is_dir(new_dentry
)) {
4322 if (newnd
.last
.name
[newnd
.last
.len
])
4326 /* unless the source is a directory trailing slashes give -ENOTDIR */
4327 if (!d_is_dir(old_dentry
)) {
4329 if (oldnd
.last
.name
[oldnd
.last
.len
])
4331 if (!(flags
& RENAME_EXCHANGE
) && newnd
.last
.name
[newnd
.last
.len
])
4334 /* source should not be ancestor of target */
4336 if (old_dentry
== trap
)
4338 /* target should not be an ancestor of source */
4339 if (!(flags
& RENAME_EXCHANGE
))
4341 if (new_dentry
== trap
)
4344 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4345 &newnd
.path
, new_dentry
, flags
);
4348 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4349 new_dir
->d_inode
, new_dentry
,
4350 &delegated_inode
, flags
);
4356 unlock_rename(new_dir
, old_dir
);
4357 if (delegated_inode
) {
4358 error
= break_deleg_wait(&delegated_inode
);
4362 mnt_drop_write(oldnd
.path
.mnt
);
4364 if (retry_estale(error
, lookup_flags
))
4365 should_retry
= true;
4366 path_put(&newnd
.path
);
4369 path_put(&oldnd
.path
);
4372 should_retry
= false;
4373 lookup_flags
|= LOOKUP_REVAL
;
4380 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4381 int, newdfd
, const char __user
*, newname
)
4383 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4386 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4388 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4391 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4393 int error
= may_create(dir
, dentry
);
4397 if (!dir
->i_op
->mknod
)
4400 return dir
->i_op
->mknod(dir
, dentry
,
4401 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4403 EXPORT_SYMBOL(vfs_whiteout
);
4405 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4407 int len
= PTR_ERR(link
);
4412 if (len
> (unsigned) buflen
)
4414 if (copy_to_user(buffer
, link
, len
))
4419 EXPORT_SYMBOL(readlink_copy
);
4422 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4423 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4424 * using) it for any given inode is up to filesystem.
4426 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4428 struct nameidata nd
;
4433 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4435 return PTR_ERR(cookie
);
4437 res
= readlink_copy(buffer
, buflen
, nd_get_link(&nd
));
4438 if (dentry
->d_inode
->i_op
->put_link
)
4439 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4442 EXPORT_SYMBOL(generic_readlink
);
4444 /* get the link contents into pagecache */
4445 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4449 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4450 page
= read_mapping_page(mapping
, 0, NULL
);
4455 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4459 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4461 struct page
*page
= NULL
;
4462 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4465 page_cache_release(page
);
4469 EXPORT_SYMBOL(page_readlink
);
4471 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4473 struct page
*page
= NULL
;
4474 nd_set_link(nd
, page_getlink(dentry
, &page
));
4477 EXPORT_SYMBOL(page_follow_link_light
);
4479 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4481 struct page
*page
= cookie
;
4485 page_cache_release(page
);
4488 EXPORT_SYMBOL(page_put_link
);
4491 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4493 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4495 struct address_space
*mapping
= inode
->i_mapping
;
4500 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4502 flags
|= AOP_FLAG_NOFS
;
4505 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4506 flags
, &page
, &fsdata
);
4510 kaddr
= kmap_atomic(page
);
4511 memcpy(kaddr
, symname
, len
-1);
4512 kunmap_atomic(kaddr
);
4514 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4521 mark_inode_dirty(inode
);
4526 EXPORT_SYMBOL(__page_symlink
);
4528 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4530 return __page_symlink(inode
, symname
, len
,
4531 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4533 EXPORT_SYMBOL(page_symlink
);
4535 const struct inode_operations page_symlink_inode_operations
= {
4536 .readlink
= generic_readlink
,
4537 .follow_link
= page_follow_link_light
,
4538 .put_link
= page_put_link
,
4540 EXPORT_SYMBOL(page_symlink_inode_operations
);