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 <linux/bitops.h>
39 #include <linux/init_task.h>
40 #include <asm/uaccess.h>
45 /* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
73 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
79 * the name is a symlink pointing to a non-existent name.
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
90 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
93 * [10-Sep-98 Alan Modra] Another symlink change.
96 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
113 * any extra contention...
116 /* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
124 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127 getname_flags(const char __user
*filename
, int flags
, int *empty
)
129 struct filename
*result
;
133 result
= audit_reusename(filename
);
137 result
= __getname();
138 if (unlikely(!result
))
139 return ERR_PTR(-ENOMEM
);
142 * First, try to embed the struct filename inside the names_cache
145 kname
= (char *)result
->iname
;
146 result
->name
= kname
;
148 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
149 if (unlikely(len
< 0)) {
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
160 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
161 const size_t size
= offsetof(struct filename
, iname
[1]);
162 kname
= (char *)result
;
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
169 result
= kzalloc(size
, GFP_KERNEL
);
170 if (unlikely(!result
)) {
172 return ERR_PTR(-ENOMEM
);
174 result
->name
= kname
;
175 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
176 if (unlikely(len
< 0)) {
181 if (unlikely(len
== PATH_MAX
)) {
184 return ERR_PTR(-ENAMETOOLONG
);
189 /* The empty path is special. */
190 if (unlikely(!len
)) {
193 if (!(flags
& LOOKUP_EMPTY
)) {
195 return ERR_PTR(-ENOENT
);
199 result
->uptr
= filename
;
200 result
->aname
= NULL
;
201 audit_getname(result
);
206 getname(const char __user
* filename
)
208 return getname_flags(filename
, 0, NULL
);
212 getname_kernel(const char * filename
)
214 struct filename
*result
;
215 int len
= strlen(filename
) + 1;
217 result
= __getname();
218 if (unlikely(!result
))
219 return ERR_PTR(-ENOMEM
);
221 if (len
<= EMBEDDED_NAME_MAX
) {
222 result
->name
= (char *)result
->iname
;
223 } else if (len
<= PATH_MAX
) {
224 struct filename
*tmp
;
226 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
227 if (unlikely(!tmp
)) {
229 return ERR_PTR(-ENOMEM
);
231 tmp
->name
= (char *)result
;
235 return ERR_PTR(-ENAMETOOLONG
);
237 memcpy((char *)result
->name
, filename
, len
);
239 result
->aname
= NULL
;
241 audit_getname(result
);
246 void putname(struct filename
*name
)
248 BUG_ON(name
->refcnt
<= 0);
250 if (--name
->refcnt
> 0)
253 if (name
->name
!= name
->iname
) {
254 __putname(name
->name
);
260 static int check_acl(struct inode
*inode
, int mask
)
262 #ifdef CONFIG_FS_POSIX_ACL
263 struct posix_acl
*acl
;
265 if (mask
& MAY_NOT_BLOCK
) {
266 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
269 /* no ->get_acl() calls in RCU mode... */
270 if (is_uncached_acl(acl
))
272 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
275 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
279 int error
= posix_acl_permission(inode
, acl
, mask
);
280 posix_acl_release(acl
);
289 * This does the basic permission checking
291 static int acl_permission_check(struct inode
*inode
, int mask
)
293 unsigned int mode
= inode
->i_mode
;
295 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
298 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
299 int error
= check_acl(inode
, mask
);
300 if (error
!= -EAGAIN
)
304 if (in_group_p(inode
->i_gid
))
309 * If the DACs are ok we don't need any capability check.
311 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
317 * generic_permission - check for access rights on a Posix-like filesystem
318 * @inode: inode to check access rights for
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
330 int generic_permission(struct inode
*inode
, int mask
)
335 * Do the basic permission checks.
337 ret
= acl_permission_check(inode
, mask
);
341 if (S_ISDIR(inode
->i_mode
)) {
342 /* DACs are overridable for directories */
343 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
345 if (!(mask
& MAY_WRITE
))
346 if (capable_wrt_inode_uidgid(inode
,
347 CAP_DAC_READ_SEARCH
))
352 * Read/write DACs are always overridable.
353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
356 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
357 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
361 * Searching includes executable on directories, else just read.
363 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
364 if (mask
== MAY_READ
)
365 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
370 EXPORT_SYMBOL(generic_permission
);
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
378 static inline int do_inode_permission(struct inode
*inode
, int mask
)
380 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
381 if (likely(inode
->i_op
->permission
))
382 return inode
->i_op
->permission(inode
, mask
);
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode
->i_lock
);
386 inode
->i_opflags
|= IOP_FASTPERM
;
387 spin_unlock(&inode
->i_lock
);
389 return generic_permission(inode
, mask
);
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
397 * Check for read/write/execute permissions on an inode.
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
404 int __inode_permission(struct inode
*inode
, int mask
)
408 if (unlikely(mask
& MAY_WRITE
)) {
410 * Nobody gets write access to an immutable file.
412 if (IS_IMMUTABLE(inode
))
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
420 if (HAS_UNMAPPED_ID(inode
))
424 retval
= do_inode_permission(inode
, mask
);
428 retval
= devcgroup_inode_permission(inode
, mask
);
432 return security_inode_permission(inode
, mask
);
434 EXPORT_SYMBOL(__inode_permission
);
437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
439 * @inode: Inode to check permission on
440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442 * Separate out file-system wide checks from inode-specific permission checks.
444 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
446 if (unlikely(mask
& MAY_WRITE
)) {
447 umode_t mode
= inode
->i_mode
;
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb
->s_flags
& MS_RDONLY
) &&
451 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
468 int inode_permission(struct inode
*inode
, int mask
)
472 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
475 return __inode_permission(inode
, mask
);
477 EXPORT_SYMBOL(inode_permission
);
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
483 * Given a path increment the reference count to the dentry and the vfsmount.
485 void path_get(const struct path
*path
)
490 EXPORT_SYMBOL(path_get
);
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
496 * Given a path decrement the reference count to the dentry and the vfsmount.
498 void path_put(const struct path
*path
)
503 EXPORT_SYMBOL(path_put
);
505 #define EMBEDDED_LEVELS 2
510 struct inode
*inode
; /* path.dentry.d_inode */
515 int total_link_count
;
518 struct delayed_call done
;
521 } *stack
, internal
[EMBEDDED_LEVELS
];
522 struct filename
*name
;
523 struct nameidata
*saved
;
524 struct inode
*link_inode
;
529 static void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
531 struct nameidata
*old
= current
->nameidata
;
532 p
->stack
= p
->internal
;
535 p
->total_link_count
= old
? old
->total_link_count
: 0;
537 current
->nameidata
= p
;
540 static void restore_nameidata(void)
542 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
544 current
->nameidata
= old
;
546 old
->total_link_count
= now
->total_link_count
;
547 if (now
->stack
!= now
->internal
)
551 static int __nd_alloc_stack(struct nameidata
*nd
)
555 if (nd
->flags
& LOOKUP_RCU
) {
556 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
561 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
566 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
578 static bool path_connected(const struct path
*path
)
580 struct vfsmount
*mnt
= path
->mnt
;
582 /* Only bind mounts can have disconnected paths */
583 if (mnt
->mnt_root
== mnt
->mnt_sb
->s_root
)
586 return is_subdir(path
->dentry
, mnt
->mnt_root
);
589 static inline int nd_alloc_stack(struct nameidata
*nd
)
591 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
593 if (likely(nd
->stack
!= nd
->internal
))
595 return __nd_alloc_stack(nd
);
598 static void drop_links(struct nameidata
*nd
)
602 struct saved
*last
= nd
->stack
+ i
;
603 do_delayed_call(&last
->done
);
604 clear_delayed_call(&last
->done
);
608 static void terminate_walk(struct nameidata
*nd
)
611 if (!(nd
->flags
& LOOKUP_RCU
)) {
614 for (i
= 0; i
< nd
->depth
; i
++)
615 path_put(&nd
->stack
[i
].link
);
616 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
621 nd
->flags
&= ~LOOKUP_RCU
;
622 if (!(nd
->flags
& LOOKUP_ROOT
))
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata
*nd
,
631 struct path
*path
, unsigned seq
)
633 int res
= __legitimize_mnt(path
->mnt
, nd
->m_seq
);
640 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
644 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
647 static bool legitimize_links(struct nameidata
*nd
)
650 for (i
= 0; i
< nd
->depth
; i
++) {
651 struct saved
*last
= nd
->stack
+ i
;
652 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
676 * @seq: seq number to check dentry against
677 * Returns: 0 on success, -ECHILD on failure
679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
681 * @nd or NULL. Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
685 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
687 struct dentry
*parent
= nd
->path
.dentry
;
689 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
691 nd
->flags
&= ~LOOKUP_RCU
;
692 if (unlikely(!legitimize_links(nd
)))
694 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
696 if (unlikely(!lockref_get_not_dead(&parent
->d_lockref
)))
700 * For a negative lookup, the lookup sequence point is the parents
701 * sequence point, and it only needs to revalidate the parent dentry.
703 * For a positive lookup, we need to move both the parent and the
704 * dentry from the RCU domain to be properly refcounted. And the
705 * sequence number in the dentry validates *both* dentry counters,
706 * since we checked the sequence number of the parent after we got
707 * the child sequence number. So we know the parent must still
708 * be valid if the child sequence number is still valid.
711 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
713 BUG_ON(nd
->inode
!= parent
->d_inode
);
715 if (!lockref_get_not_dead(&dentry
->d_lockref
))
717 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
722 * Sequence counts matched. Now make sure that the root is
723 * still valid and get it if required.
725 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
726 if (unlikely(!legitimize_path(nd
, &nd
->root
, nd
->root_seq
))) {
743 nd
->path
.dentry
= NULL
;
747 if (!(nd
->flags
& LOOKUP_ROOT
))
752 static int unlazy_link(struct nameidata
*nd
, struct path
*link
, unsigned seq
)
754 if (unlikely(!legitimize_path(nd
, link
, seq
))) {
757 nd
->flags
&= ~LOOKUP_RCU
;
759 nd
->path
.dentry
= NULL
;
760 if (!(nd
->flags
& LOOKUP_ROOT
))
763 } else if (likely(unlazy_walk(nd
, NULL
, 0)) == 0) {
770 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
772 return dentry
->d_op
->d_revalidate(dentry
, flags
);
776 * complete_walk - successful completion of path walk
777 * @nd: pointer nameidata
779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it. Return 0 on
782 * success, -error on failure. In case of failure caller does not
783 * need to drop nd->path.
785 static int complete_walk(struct nameidata
*nd
)
787 struct dentry
*dentry
= nd
->path
.dentry
;
790 if (nd
->flags
& LOOKUP_RCU
) {
791 if (!(nd
->flags
& LOOKUP_ROOT
))
793 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
797 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
800 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
803 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
813 static void set_root(struct nameidata
*nd
)
815 struct fs_struct
*fs
= current
->fs
;
817 if (nd
->flags
& LOOKUP_RCU
) {
821 seq
= read_seqcount_begin(&fs
->seq
);
823 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
824 } while (read_seqcount_retry(&fs
->seq
, seq
));
826 get_fs_root(fs
, &nd
->root
);
830 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
833 if (path
->mnt
!= nd
->path
.mnt
)
837 static inline void path_to_nameidata(const struct path
*path
,
838 struct nameidata
*nd
)
840 if (!(nd
->flags
& LOOKUP_RCU
)) {
841 dput(nd
->path
.dentry
);
842 if (nd
->path
.mnt
!= path
->mnt
)
843 mntput(nd
->path
.mnt
);
845 nd
->path
.mnt
= path
->mnt
;
846 nd
->path
.dentry
= path
->dentry
;
849 static int nd_jump_root(struct nameidata
*nd
)
851 if (nd
->flags
& LOOKUP_RCU
) {
855 nd
->inode
= d
->d_inode
;
856 nd
->seq
= nd
->root_seq
;
857 if (unlikely(read_seqcount_retry(&d
->d_seq
, nd
->seq
)))
863 nd
->inode
= nd
->path
.dentry
->d_inode
;
865 nd
->flags
|= LOOKUP_JUMPED
;
870 * Helper to directly jump to a known parsed path from ->get_link,
871 * caller must have taken a reference to path beforehand.
873 void nd_jump_link(struct path
*path
)
875 struct nameidata
*nd
= current
->nameidata
;
879 nd
->inode
= nd
->path
.dentry
->d_inode
;
880 nd
->flags
|= LOOKUP_JUMPED
;
883 static inline void put_link(struct nameidata
*nd
)
885 struct saved
*last
= nd
->stack
+ --nd
->depth
;
886 do_delayed_call(&last
->done
);
887 if (!(nd
->flags
& LOOKUP_RCU
))
888 path_put(&last
->link
);
891 int sysctl_protected_symlinks __read_mostly
= 0;
892 int sysctl_protected_hardlinks __read_mostly
= 0;
895 * may_follow_link - Check symlink following for unsafe situations
896 * @nd: nameidata pathwalk data
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
907 * Returns 0 if following the symlink is allowed, -ve on error.
909 static inline int may_follow_link(struct nameidata
*nd
)
911 const struct inode
*inode
;
912 const struct inode
*parent
;
915 if (!sysctl_protected_symlinks
)
918 /* Allowed if owner and follower match. */
919 inode
= nd
->link_inode
;
920 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
923 /* Allowed if parent directory not sticky and world-writable. */
925 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
928 /* Allowed if parent directory and link owner match. */
929 puid
= parent
->i_uid
;
930 if (uid_valid(puid
) && uid_eq(puid
, inode
->i_uid
))
933 if (nd
->flags
& LOOKUP_RCU
)
936 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
947 * - inode is setgid and group-exec
948 * - access failure for read and write
950 * Otherwise returns true.
952 static bool safe_hardlink_source(struct inode
*inode
)
954 umode_t mode
= inode
->i_mode
;
956 /* Special files should not get pinned to the filesystem. */
960 /* Setuid files should not get pinned to the filesystem. */
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
985 * Returns 0 if successful, -ve on error.
987 static int may_linkat(struct path
*link
)
991 if (!sysctl_protected_hardlinks
)
994 inode
= link
->dentry
->d_inode
;
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
999 if (inode_owner_or_capable(inode
) || safe_hardlink_source(inode
))
1002 audit_log_link_denied("linkat", link
);
1006 static __always_inline
1007 const char *get_link(struct nameidata
*nd
)
1009 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
1010 struct dentry
*dentry
= last
->link
.dentry
;
1011 struct inode
*inode
= nd
->link_inode
;
1015 if (!(nd
->flags
& LOOKUP_RCU
)) {
1016 touch_atime(&last
->link
);
1018 } else if (atime_needs_update_rcu(&last
->link
, inode
)) {
1019 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1020 return ERR_PTR(-ECHILD
);
1021 touch_atime(&last
->link
);
1024 error
= security_inode_follow_link(dentry
, inode
,
1025 nd
->flags
& LOOKUP_RCU
);
1026 if (unlikely(error
))
1027 return ERR_PTR(error
);
1029 nd
->last_type
= LAST_BIND
;
1030 res
= inode
->i_link
;
1032 const char * (*get
)(struct dentry
*, struct inode
*,
1033 struct delayed_call
*);
1034 get
= inode
->i_op
->get_link
;
1035 if (nd
->flags
& LOOKUP_RCU
) {
1036 res
= get(NULL
, inode
, &last
->done
);
1037 if (res
== ERR_PTR(-ECHILD
)) {
1038 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1039 return ERR_PTR(-ECHILD
);
1040 res
= get(dentry
, inode
, &last
->done
);
1043 res
= get(dentry
, inode
, &last
->done
);
1045 if (IS_ERR_OR_NULL(res
))
1051 if (unlikely(nd_jump_root(nd
)))
1052 return ERR_PTR(-ECHILD
);
1053 while (unlikely(*++res
== '/'))
1062 * follow_up - Find the mountpoint of path's vfsmount
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1068 * Return 1 if we went up a level and 0 if we were already at the
1071 int follow_up(struct path
*path
)
1073 struct mount
*mnt
= real_mount(path
->mnt
);
1074 struct mount
*parent
;
1075 struct dentry
*mountpoint
;
1077 read_seqlock_excl(&mount_lock
);
1078 parent
= mnt
->mnt_parent
;
1079 if (parent
== mnt
) {
1080 read_sequnlock_excl(&mount_lock
);
1083 mntget(&parent
->mnt
);
1084 mountpoint
= dget(mnt
->mnt_mountpoint
);
1085 read_sequnlock_excl(&mount_lock
);
1087 path
->dentry
= mountpoint
;
1089 path
->mnt
= &parent
->mnt
;
1092 EXPORT_SYMBOL(follow_up
);
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1099 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
1102 struct vfsmount
*mnt
;
1103 const struct cred
*old_cred
;
1106 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
1109 /* We don't want to mount if someone's just doing a stat -
1110 * unless they're stat'ing a directory and appended a '/' to
1113 * We do, however, want to mount if someone wants to open or
1114 * create a file of any type under the mountpoint, wants to
1115 * traverse through the mountpoint or wants to open the
1116 * mounted directory. Also, autofs may mark negative dentries
1117 * as being automount points. These will need the attentions
1118 * of the daemon to instantiate them before they can be used.
1120 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1121 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1122 path
->dentry
->d_inode
)
1125 if (path
->dentry
->d_sb
->s_user_ns
!= &init_user_ns
)
1128 nd
->total_link_count
++;
1129 if (nd
->total_link_count
>= 40)
1132 old_cred
= override_creds(&init_cred
);
1133 mnt
= path
->dentry
->d_op
->d_automount(path
);
1134 revert_creds(old_cred
);
1137 * The filesystem is allowed to return -EISDIR here to indicate
1138 * it doesn't want to automount. For instance, autofs would do
1139 * this so that its userspace daemon can mount on this dentry.
1141 * However, we can only permit this if it's a terminal point in
1142 * the path being looked up; if it wasn't then the remainder of
1143 * the path is inaccessible and we should say so.
1145 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1147 return PTR_ERR(mnt
);
1150 if (!mnt
) /* mount collision */
1153 if (!*need_mntput
) {
1154 /* lock_mount() may release path->mnt on error */
1156 *need_mntput
= true;
1158 err
= finish_automount(mnt
, path
);
1162 /* Someone else made a mount here whilst we were busy */
1167 path
->dentry
= dget(mnt
->mnt_root
);
1176 * Handle a dentry that is managed in some way.
1177 * - Flagged for transit management (autofs)
1178 * - Flagged as mountpoint
1179 * - Flagged as automount point
1181 * This may only be called in refwalk mode.
1183 * Serialization is taken care of in namespace.c
1185 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1187 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1189 bool need_mntput
= false;
1192 /* Given that we're not holding a lock here, we retain the value in a
1193 * local variable for each dentry as we look at it so that we don't see
1194 * the components of that value change under us */
1195 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1196 managed
&= DCACHE_MANAGED_DENTRY
,
1197 unlikely(managed
!= 0)) {
1198 /* Allow the filesystem to manage the transit without i_mutex
1200 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1201 BUG_ON(!path
->dentry
->d_op
);
1202 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1203 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1208 /* Transit to a mounted filesystem. */
1209 if (managed
& DCACHE_MOUNTED
) {
1210 struct vfsmount
*mounted
= lookup_mnt(path
);
1215 path
->mnt
= mounted
;
1216 path
->dentry
= dget(mounted
->mnt_root
);
1221 /* Something is mounted on this dentry in another
1222 * namespace and/or whatever was mounted there in this
1223 * namespace got unmounted before lookup_mnt() could
1227 /* Handle an automount point */
1228 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1229 ret
= follow_automount(path
, nd
, &need_mntput
);
1235 /* We didn't change the current path point */
1239 if (need_mntput
&& path
->mnt
== mnt
)
1241 if (ret
== -EISDIR
|| !ret
)
1244 nd
->flags
|= LOOKUP_JUMPED
;
1245 if (unlikely(ret
< 0))
1246 path_put_conditional(path
, nd
);
1250 int follow_down_one(struct path
*path
)
1252 struct vfsmount
*mounted
;
1254 mounted
= lookup_mnt(path
);
1258 path
->mnt
= mounted
;
1259 path
->dentry
= dget(mounted
->mnt_root
);
1264 EXPORT_SYMBOL(follow_down_one
);
1266 static inline int managed_dentry_rcu(const struct path
*path
)
1268 return (path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1269 path
->dentry
->d_op
->d_manage(path
, true) : 0;
1273 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1274 * we meet a managed dentry that would need blocking.
1276 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1277 struct inode
**inode
, unsigned *seqp
)
1280 struct mount
*mounted
;
1282 * Don't forget we might have a non-mountpoint managed dentry
1283 * that wants to block transit.
1285 switch (managed_dentry_rcu(path
)) {
1295 if (!d_mountpoint(path
->dentry
))
1296 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1298 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1301 path
->mnt
= &mounted
->mnt
;
1302 path
->dentry
= mounted
->mnt
.mnt_root
;
1303 nd
->flags
|= LOOKUP_JUMPED
;
1304 *seqp
= read_seqcount_begin(&path
->dentry
->d_seq
);
1306 * Update the inode too. We don't need to re-check the
1307 * dentry sequence number here after this d_inode read,
1308 * because a mount-point is always pinned.
1310 *inode
= path
->dentry
->d_inode
;
1312 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1313 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1316 static int follow_dotdot_rcu(struct nameidata
*nd
)
1318 struct inode
*inode
= nd
->inode
;
1321 if (path_equal(&nd
->path
, &nd
->root
))
1323 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1324 struct dentry
*old
= nd
->path
.dentry
;
1325 struct dentry
*parent
= old
->d_parent
;
1328 inode
= parent
->d_inode
;
1329 seq
= read_seqcount_begin(&parent
->d_seq
);
1330 if (unlikely(read_seqcount_retry(&old
->d_seq
, nd
->seq
)))
1332 nd
->path
.dentry
= parent
;
1334 if (unlikely(!path_connected(&nd
->path
)))
1338 struct mount
*mnt
= real_mount(nd
->path
.mnt
);
1339 struct mount
*mparent
= mnt
->mnt_parent
;
1340 struct dentry
*mountpoint
= mnt
->mnt_mountpoint
;
1341 struct inode
*inode2
= mountpoint
->d_inode
;
1342 unsigned seq
= read_seqcount_begin(&mountpoint
->d_seq
);
1343 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1345 if (&mparent
->mnt
== nd
->path
.mnt
)
1347 /* we know that mountpoint was pinned */
1348 nd
->path
.dentry
= mountpoint
;
1349 nd
->path
.mnt
= &mparent
->mnt
;
1354 while (unlikely(d_mountpoint(nd
->path
.dentry
))) {
1355 struct mount
*mounted
;
1356 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1357 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1361 nd
->path
.mnt
= &mounted
->mnt
;
1362 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1363 inode
= nd
->path
.dentry
->d_inode
;
1364 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1371 * Follow down to the covering mount currently visible to userspace. At each
1372 * point, the filesystem owning that dentry may be queried as to whether the
1373 * caller is permitted to proceed or not.
1375 int follow_down(struct path
*path
)
1380 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1381 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1382 /* Allow the filesystem to manage the transit without i_mutex
1385 * We indicate to the filesystem if someone is trying to mount
1386 * something here. This gives autofs the chance to deny anyone
1387 * other than its daemon the right to mount on its
1390 * The filesystem may sleep at this point.
1392 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1393 BUG_ON(!path
->dentry
->d_op
);
1394 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1395 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1397 return ret
== -EISDIR
? 0 : ret
;
1400 /* Transit to a mounted filesystem. */
1401 if (managed
& DCACHE_MOUNTED
) {
1402 struct vfsmount
*mounted
= lookup_mnt(path
);
1407 path
->mnt
= mounted
;
1408 path
->dentry
= dget(mounted
->mnt_root
);
1412 /* Don't handle automount points here */
1417 EXPORT_SYMBOL(follow_down
);
1420 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1422 static void follow_mount(struct path
*path
)
1424 while (d_mountpoint(path
->dentry
)) {
1425 struct vfsmount
*mounted
= lookup_mnt(path
);
1430 path
->mnt
= mounted
;
1431 path
->dentry
= dget(mounted
->mnt_root
);
1435 static int path_parent_directory(struct path
*path
)
1437 struct dentry
*old
= path
->dentry
;
1438 /* rare case of legitimate dget_parent()... */
1439 path
->dentry
= dget_parent(path
->dentry
);
1441 if (unlikely(!path_connected(path
)))
1446 static int follow_dotdot(struct nameidata
*nd
)
1449 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1450 nd
->path
.mnt
== nd
->root
.mnt
) {
1453 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1454 int ret
= path_parent_directory(&nd
->path
);
1459 if (!follow_up(&nd
->path
))
1462 follow_mount(&nd
->path
);
1463 nd
->inode
= nd
->path
.dentry
->d_inode
;
1468 * This looks up the name in dcache and possibly revalidates the found dentry.
1469 * NULL is returned if the dentry does not exist in the cache.
1471 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1475 struct dentry
*dentry
;
1478 dentry
= d_lookup(dir
, name
);
1480 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1481 error
= d_revalidate(dentry
, flags
);
1482 if (unlikely(error
<= 0)) {
1484 d_invalidate(dentry
);
1486 return ERR_PTR(error
);
1494 * Call i_op->lookup on the dentry. The dentry must be negative and
1497 * dir->d_inode->i_mutex must be held
1499 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1504 /* Don't create child dentry for a dead directory. */
1505 if (unlikely(IS_DEADDIR(dir
))) {
1507 return ERR_PTR(-ENOENT
);
1510 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1511 if (unlikely(old
)) {
1518 static struct dentry
*__lookup_hash(const struct qstr
*name
,
1519 struct dentry
*base
, unsigned int flags
)
1521 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1526 dentry
= d_alloc(base
, name
);
1527 if (unlikely(!dentry
))
1528 return ERR_PTR(-ENOMEM
);
1530 return lookup_real(base
->d_inode
, dentry
, flags
);
1533 static int lookup_fast(struct nameidata
*nd
,
1534 struct path
*path
, struct inode
**inode
,
1537 struct vfsmount
*mnt
= nd
->path
.mnt
;
1538 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1543 * Rename seqlock is not required here because in the off chance
1544 * of a false negative due to a concurrent rename, the caller is
1545 * going to fall back to non-racy lookup.
1547 if (nd
->flags
& LOOKUP_RCU
) {
1550 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1551 if (unlikely(!dentry
)) {
1552 if (unlazy_walk(nd
, NULL
, 0))
1558 * This sequence count validates that the inode matches
1559 * the dentry name information from lookup.
1561 *inode
= d_backing_inode(dentry
);
1562 negative
= d_is_negative(dentry
);
1563 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
1567 * This sequence count validates that the parent had no
1568 * changes while we did the lookup of the dentry above.
1570 * The memory barrier in read_seqcount_begin of child is
1571 * enough, we can use __read_seqcount_retry here.
1573 if (unlikely(__read_seqcount_retry(&parent
->d_seq
, nd
->seq
)))
1577 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1578 status
= d_revalidate(dentry
, nd
->flags
);
1579 if (unlikely(status
<= 0)) {
1580 if (unlazy_walk(nd
, dentry
, seq
))
1582 if (status
== -ECHILD
)
1583 status
= d_revalidate(dentry
, nd
->flags
);
1586 * Note: do negative dentry check after revalidation in
1587 * case that drops it.
1589 if (unlikely(negative
))
1592 path
->dentry
= dentry
;
1593 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1595 if (unlazy_walk(nd
, dentry
, seq
))
1599 dentry
= __d_lookup(parent
, &nd
->last
);
1600 if (unlikely(!dentry
))
1602 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1603 status
= d_revalidate(dentry
, nd
->flags
);
1605 if (unlikely(status
<= 0)) {
1607 d_invalidate(dentry
);
1611 if (unlikely(d_is_negative(dentry
))) {
1617 path
->dentry
= dentry
;
1618 err
= follow_managed(path
, nd
);
1619 if (likely(err
> 0))
1620 *inode
= d_backing_inode(path
->dentry
);
1624 /* Fast lookup failed, do it the slow way */
1625 static struct dentry
*lookup_slow(const struct qstr
*name
,
1629 struct dentry
*dentry
= ERR_PTR(-ENOENT
), *old
;
1630 struct inode
*inode
= dir
->d_inode
;
1631 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1633 inode_lock_shared(inode
);
1634 /* Don't go there if it's already dead */
1635 if (unlikely(IS_DEADDIR(inode
)))
1638 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1641 if (unlikely(!d_in_lookup(dentry
))) {
1642 if ((dentry
->d_flags
& DCACHE_OP_REVALIDATE
) &&
1643 !(flags
& LOOKUP_NO_REVAL
)) {
1644 int error
= d_revalidate(dentry
, flags
);
1645 if (unlikely(error
<= 0)) {
1647 d_invalidate(dentry
);
1652 dentry
= ERR_PTR(error
);
1656 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1657 d_lookup_done(dentry
);
1658 if (unlikely(old
)) {
1664 inode_unlock_shared(inode
);
1668 static inline int may_lookup(struct nameidata
*nd
)
1670 if (nd
->flags
& LOOKUP_RCU
) {
1671 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1674 if (unlazy_walk(nd
, NULL
, 0))
1677 return inode_permission(nd
->inode
, MAY_EXEC
);
1680 static inline int handle_dots(struct nameidata
*nd
, int type
)
1682 if (type
== LAST_DOTDOT
) {
1685 if (nd
->flags
& LOOKUP_RCU
) {
1686 return follow_dotdot_rcu(nd
);
1688 return follow_dotdot(nd
);
1693 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1694 struct inode
*inode
, unsigned seq
)
1698 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1699 path_to_nameidata(link
, nd
);
1702 if (!(nd
->flags
& LOOKUP_RCU
)) {
1703 if (link
->mnt
== nd
->path
.mnt
)
1706 error
= nd_alloc_stack(nd
);
1707 if (unlikely(error
)) {
1708 if (error
== -ECHILD
) {
1709 if (unlikely(unlazy_link(nd
, link
, seq
)))
1711 error
= nd_alloc_stack(nd
);
1719 last
= nd
->stack
+ nd
->depth
++;
1721 clear_delayed_call(&last
->done
);
1722 nd
->link_inode
= inode
;
1728 * Do we need to follow links? We _really_ want to be able
1729 * to do this check without having to look at inode->i_op,
1730 * so we keep a cache of "no, this doesn't need follow_link"
1731 * for the common case.
1733 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
,
1735 struct inode
*inode
, unsigned seq
)
1737 if (likely(!d_is_symlink(link
->dentry
)))
1741 /* make sure that d_is_symlink above matches inode */
1742 if (nd
->flags
& LOOKUP_RCU
) {
1743 if (read_seqcount_retry(&link
->dentry
->d_seq
, seq
))
1746 return pick_link(nd
, link
, inode
, seq
);
1749 enum {WALK_GET
= 1, WALK_PUT
= 2};
1751 static int walk_component(struct nameidata
*nd
, int flags
)
1754 struct inode
*inode
;
1758 * "." and ".." are special - ".." especially so because it has
1759 * to be able to know about the current root directory and
1760 * parent relationships.
1762 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1763 err
= handle_dots(nd
, nd
->last_type
);
1764 if (flags
& WALK_PUT
)
1768 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1769 if (unlikely(err
<= 0)) {
1772 path
.dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
,
1774 if (IS_ERR(path
.dentry
))
1775 return PTR_ERR(path
.dentry
);
1777 path
.mnt
= nd
->path
.mnt
;
1778 err
= follow_managed(&path
, nd
);
1779 if (unlikely(err
< 0))
1782 if (unlikely(d_is_negative(path
.dentry
))) {
1783 path_to_nameidata(&path
, nd
);
1787 seq
= 0; /* we are already out of RCU mode */
1788 inode
= d_backing_inode(path
.dentry
);
1791 if (flags
& WALK_PUT
)
1793 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
, inode
, seq
);
1796 path_to_nameidata(&path
, nd
);
1803 * We can do the critical dentry name comparison and hashing
1804 * operations one word at a time, but we are limited to:
1806 * - Architectures with fast unaligned word accesses. We could
1807 * do a "get_unaligned()" if this helps and is sufficiently
1810 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1811 * do not trap on the (extremely unlikely) case of a page
1812 * crossing operation.
1814 * - Furthermore, we need an efficient 64-bit compile for the
1815 * 64-bit case in order to generate the "number of bytes in
1816 * the final mask". Again, that could be replaced with a
1817 * efficient population count instruction or similar.
1819 #ifdef CONFIG_DCACHE_WORD_ACCESS
1821 #include <asm/word-at-a-time.h>
1825 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1827 #elif defined(CONFIG_64BIT)
1829 * Register pressure in the mixing function is an issue, particularly
1830 * on 32-bit x86, but almost any function requires one state value and
1831 * one temporary. Instead, use a function designed for two state values
1832 * and no temporaries.
1834 * This function cannot create a collision in only two iterations, so
1835 * we have two iterations to achieve avalanche. In those two iterations,
1836 * we have six layers of mixing, which is enough to spread one bit's
1837 * influence out to 2^6 = 64 state bits.
1839 * Rotate constants are scored by considering either 64 one-bit input
1840 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1841 * probability of that delta causing a change to each of the 128 output
1842 * bits, using a sample of random initial states.
1844 * The Shannon entropy of the computed probabilities is then summed
1845 * to produce a score. Ideally, any input change has a 50% chance of
1846 * toggling any given output bit.
1848 * Mixing scores (in bits) for (12,45):
1849 * Input delta: 1-bit 2-bit
1850 * 1 round: 713.3 42542.6
1851 * 2 rounds: 2753.7 140389.8
1852 * 3 rounds: 5954.1 233458.2
1853 * 4 rounds: 7862.6 256672.2
1854 * Perfect: 8192 258048
1855 * (64*128) (64*63/2 * 128)
1857 #define HASH_MIX(x, y, a) \
1859 y ^= x, x = rol64(x,12),\
1860 x += y, y = rol64(y,45),\
1864 * Fold two longs into one 32-bit hash value. This must be fast, but
1865 * latency isn't quite as critical, as there is a fair bit of additional
1866 * work done before the hash value is used.
1868 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1870 y
^= x
* GOLDEN_RATIO_64
;
1871 y
*= GOLDEN_RATIO_64
;
1875 #else /* 32-bit case */
1878 * Mixing scores (in bits) for (7,20):
1879 * Input delta: 1-bit 2-bit
1880 * 1 round: 330.3 9201.6
1881 * 2 rounds: 1246.4 25475.4
1882 * 3 rounds: 1907.1 31295.1
1883 * 4 rounds: 2042.3 31718.6
1884 * Perfect: 2048 31744
1885 * (32*64) (32*31/2 * 64)
1887 #define HASH_MIX(x, y, a) \
1889 y ^= x, x = rol32(x, 7),\
1890 x += y, y = rol32(y,20),\
1893 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1895 /* Use arch-optimized multiply if one exists */
1896 return __hash_32(y
^ __hash_32(x
));
1902 * Return the hash of a string of known length. This is carfully
1903 * designed to match hash_name(), which is the more critical function.
1904 * In particular, we must end by hashing a final word containing 0..7
1905 * payload bytes, to match the way that hash_name() iterates until it
1906 * finds the delimiter after the name.
1908 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
1910 unsigned long a
, x
= 0, y
= (unsigned long)salt
;
1915 a
= load_unaligned_zeropad(name
);
1916 if (len
< sizeof(unsigned long))
1919 name
+= sizeof(unsigned long);
1920 len
-= sizeof(unsigned long);
1922 x
^= a
& bytemask_from_count(len
);
1924 return fold_hash(x
, y
);
1926 EXPORT_SYMBOL(full_name_hash
);
1928 /* Return the "hash_len" (hash and length) of a null-terminated string */
1929 u64
hashlen_string(const void *salt
, const char *name
)
1931 unsigned long a
= 0, x
= 0, y
= (unsigned long)salt
;
1932 unsigned long adata
, mask
, len
;
1933 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1940 len
+= sizeof(unsigned long);
1942 a
= load_unaligned_zeropad(name
+len
);
1943 } while (!has_zero(a
, &adata
, &constants
));
1945 adata
= prep_zero_mask(a
, adata
, &constants
);
1946 mask
= create_zero_mask(adata
);
1947 x
^= a
& zero_bytemask(mask
);
1949 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1951 EXPORT_SYMBOL(hashlen_string
);
1954 * Calculate the length and hash of the path component, and
1955 * return the "hash_len" as the result.
1957 static inline u64
hash_name(const void *salt
, const char *name
)
1959 unsigned long a
= 0, b
, x
= 0, y
= (unsigned long)salt
;
1960 unsigned long adata
, bdata
, mask
, len
;
1961 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1968 len
+= sizeof(unsigned long);
1970 a
= load_unaligned_zeropad(name
+len
);
1971 b
= a
^ REPEAT_BYTE('/');
1972 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1974 adata
= prep_zero_mask(a
, adata
, &constants
);
1975 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1976 mask
= create_zero_mask(adata
| bdata
);
1977 x
^= a
& zero_bytemask(mask
);
1979 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1982 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1984 /* Return the hash of a string of known length */
1985 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
1987 unsigned long hash
= init_name_hash(salt
);
1989 hash
= partial_name_hash((unsigned char)*name
++, hash
);
1990 return end_name_hash(hash
);
1992 EXPORT_SYMBOL(full_name_hash
);
1994 /* Return the "hash_len" (hash and length) of a null-terminated string */
1995 u64
hashlen_string(const void *salt
, const char *name
)
1997 unsigned long hash
= init_name_hash(salt
);
1998 unsigned long len
= 0, c
;
2000 c
= (unsigned char)*name
;
2003 hash
= partial_name_hash(c
, hash
);
2004 c
= (unsigned char)name
[len
];
2006 return hashlen_create(end_name_hash(hash
), len
);
2008 EXPORT_SYMBOL(hashlen_string
);
2011 * We know there's a real path component here of at least
2014 static inline u64
hash_name(const void *salt
, const char *name
)
2016 unsigned long hash
= init_name_hash(salt
);
2017 unsigned long len
= 0, c
;
2019 c
= (unsigned char)*name
;
2022 hash
= partial_name_hash(c
, hash
);
2023 c
= (unsigned char)name
[len
];
2024 } while (c
&& c
!= '/');
2025 return hashlen_create(end_name_hash(hash
), len
);
2032 * This is the basic name resolution function, turning a pathname into
2033 * the final dentry. We expect 'base' to be positive and a directory.
2035 * Returns 0 and nd will have valid dentry and mnt on success.
2036 * Returns error and drops reference to input namei data on failure.
2038 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2047 /* At this point we know we have a real path component. */
2052 err
= may_lookup(nd
);
2056 hash_len
= hash_name(nd
->path
.dentry
, name
);
2059 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2061 if (name
[1] == '.') {
2063 nd
->flags
|= LOOKUP_JUMPED
;
2069 if (likely(type
== LAST_NORM
)) {
2070 struct dentry
*parent
= nd
->path
.dentry
;
2071 nd
->flags
&= ~LOOKUP_JUMPED
;
2072 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2073 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2074 err
= parent
->d_op
->d_hash(parent
, &this);
2077 hash_len
= this.hash_len
;
2082 nd
->last
.hash_len
= hash_len
;
2083 nd
->last
.name
= name
;
2084 nd
->last_type
= type
;
2086 name
+= hashlen_len(hash_len
);
2090 * If it wasn't NUL, we know it was '/'. Skip that
2091 * slash, and continue until no more slashes.
2095 } while (unlikely(*name
== '/'));
2096 if (unlikely(!*name
)) {
2098 /* pathname body, done */
2101 name
= nd
->stack
[nd
->depth
- 1].name
;
2102 /* trailing symlink, done */
2105 /* last component of nested symlink */
2106 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
2108 err
= walk_component(nd
, WALK_GET
);
2114 const char *s
= get_link(nd
);
2123 nd
->stack
[nd
->depth
- 1].name
= name
;
2128 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2129 if (nd
->flags
& LOOKUP_RCU
) {
2130 if (unlazy_walk(nd
, NULL
, 0))
2138 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2141 const char *s
= nd
->name
->name
;
2143 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
2144 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
2146 if (flags
& LOOKUP_ROOT
) {
2147 struct dentry
*root
= nd
->root
.dentry
;
2148 struct inode
*inode
= root
->d_inode
;
2150 if (!d_can_lookup(root
))
2151 return ERR_PTR(-ENOTDIR
);
2152 retval
= inode_permission(inode
, MAY_EXEC
);
2154 return ERR_PTR(retval
);
2156 nd
->path
= nd
->root
;
2158 if (flags
& LOOKUP_RCU
) {
2160 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2161 nd
->root_seq
= nd
->seq
;
2162 nd
->m_seq
= read_seqbegin(&mount_lock
);
2164 path_get(&nd
->path
);
2169 nd
->root
.mnt
= NULL
;
2170 nd
->path
.mnt
= NULL
;
2171 nd
->path
.dentry
= NULL
;
2173 nd
->m_seq
= read_seqbegin(&mount_lock
);
2175 if (flags
& LOOKUP_RCU
)
2178 if (likely(!nd_jump_root(nd
)))
2180 nd
->root
.mnt
= NULL
;
2182 return ERR_PTR(-ECHILD
);
2183 } else if (nd
->dfd
== AT_FDCWD
) {
2184 if (flags
& LOOKUP_RCU
) {
2185 struct fs_struct
*fs
= current
->fs
;
2191 seq
= read_seqcount_begin(&fs
->seq
);
2193 nd
->inode
= nd
->path
.dentry
->d_inode
;
2194 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2195 } while (read_seqcount_retry(&fs
->seq
, seq
));
2197 get_fs_pwd(current
->fs
, &nd
->path
);
2198 nd
->inode
= nd
->path
.dentry
->d_inode
;
2202 /* Caller must check execute permissions on the starting path component */
2203 struct fd f
= fdget_raw(nd
->dfd
);
2204 struct dentry
*dentry
;
2207 return ERR_PTR(-EBADF
);
2209 dentry
= f
.file
->f_path
.dentry
;
2212 if (!d_can_lookup(dentry
)) {
2214 return ERR_PTR(-ENOTDIR
);
2218 nd
->path
= f
.file
->f_path
;
2219 if (flags
& LOOKUP_RCU
) {
2221 nd
->inode
= nd
->path
.dentry
->d_inode
;
2222 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2224 path_get(&nd
->path
);
2225 nd
->inode
= nd
->path
.dentry
->d_inode
;
2232 static const char *trailing_symlink(struct nameidata
*nd
)
2235 int error
= may_follow_link(nd
);
2236 if (unlikely(error
))
2237 return ERR_PTR(error
);
2238 nd
->flags
|= LOOKUP_PARENT
;
2239 nd
->stack
[0].name
= NULL
;
2244 static inline int lookup_last(struct nameidata
*nd
)
2246 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2247 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2249 nd
->flags
&= ~LOOKUP_PARENT
;
2250 return walk_component(nd
,
2251 nd
->flags
& LOOKUP_FOLLOW
2253 ? WALK_PUT
| WALK_GET
2258 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2259 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2261 const char *s
= path_init(nd
, flags
);
2266 while (!(err
= link_path_walk(s
, nd
))
2267 && ((err
= lookup_last(nd
)) > 0)) {
2268 s
= trailing_symlink(nd
);
2275 err
= complete_walk(nd
);
2277 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2278 if (!d_can_lookup(nd
->path
.dentry
))
2282 nd
->path
.mnt
= NULL
;
2283 nd
->path
.dentry
= NULL
;
2289 static int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2290 struct path
*path
, struct path
*root
)
2293 struct nameidata nd
;
2295 return PTR_ERR(name
);
2296 if (unlikely(root
)) {
2298 flags
|= LOOKUP_ROOT
;
2300 set_nameidata(&nd
, dfd
, name
);
2301 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2302 if (unlikely(retval
== -ECHILD
))
2303 retval
= path_lookupat(&nd
, flags
, path
);
2304 if (unlikely(retval
== -ESTALE
))
2305 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2307 if (likely(!retval
))
2308 audit_inode(name
, path
->dentry
, flags
& LOOKUP_PARENT
);
2309 restore_nameidata();
2314 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2315 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2316 struct path
*parent
)
2318 const char *s
= path_init(nd
, flags
);
2322 err
= link_path_walk(s
, nd
);
2324 err
= complete_walk(nd
);
2327 nd
->path
.mnt
= NULL
;
2328 nd
->path
.dentry
= NULL
;
2334 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2335 unsigned int flags
, struct path
*parent
,
2336 struct qstr
*last
, int *type
)
2339 struct nameidata nd
;
2343 set_nameidata(&nd
, dfd
, name
);
2344 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2345 if (unlikely(retval
== -ECHILD
))
2346 retval
= path_parentat(&nd
, flags
, parent
);
2347 if (unlikely(retval
== -ESTALE
))
2348 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2349 if (likely(!retval
)) {
2351 *type
= nd
.last_type
;
2352 audit_inode(name
, parent
->dentry
, LOOKUP_PARENT
);
2355 name
= ERR_PTR(retval
);
2357 restore_nameidata();
2361 /* does lookup, returns the object with parent locked */
2362 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2364 struct filename
*filename
;
2369 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2371 if (IS_ERR(filename
))
2372 return ERR_CAST(filename
);
2373 if (unlikely(type
!= LAST_NORM
)) {
2376 return ERR_PTR(-EINVAL
);
2378 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2379 d
= __lookup_hash(&last
, path
->dentry
, 0);
2381 inode_unlock(path
->dentry
->d_inode
);
2388 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2390 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2393 EXPORT_SYMBOL(kern_path
);
2396 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2397 * @dentry: pointer to dentry of the base directory
2398 * @mnt: pointer to vfs mount of the base directory
2399 * @name: pointer to file name
2400 * @flags: lookup flags
2401 * @path: pointer to struct path to fill
2403 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2404 const char *name
, unsigned int flags
,
2407 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2408 /* the first argument of filename_lookup() is ignored with root */
2409 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2410 flags
, path
, &root
);
2412 EXPORT_SYMBOL(vfs_path_lookup
);
2415 * lookup_one_len - filesystem helper to lookup single pathname component
2416 * @name: pathname component to lookup
2417 * @base: base directory to lookup from
2418 * @len: maximum length @len should be interpreted to
2420 * Note that this routine is purely a helper for filesystem usage and should
2421 * not be called by generic code.
2423 * The caller must hold base->i_mutex.
2425 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2431 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2435 this.hash
= full_name_hash(base
, name
, len
);
2437 return ERR_PTR(-EACCES
);
2439 if (unlikely(name
[0] == '.')) {
2440 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2441 return ERR_PTR(-EACCES
);
2445 c
= *(const unsigned char *)name
++;
2446 if (c
== '/' || c
== '\0')
2447 return ERR_PTR(-EACCES
);
2450 * See if the low-level filesystem might want
2451 * to use its own hash..
2453 if (base
->d_flags
& DCACHE_OP_HASH
) {
2454 int err
= base
->d_op
->d_hash(base
, &this);
2456 return ERR_PTR(err
);
2459 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2461 return ERR_PTR(err
);
2463 return __lookup_hash(&this, base
, 0);
2465 EXPORT_SYMBOL(lookup_one_len
);
2468 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2469 * @name: pathname component to lookup
2470 * @base: base directory to lookup from
2471 * @len: maximum length @len should be interpreted to
2473 * Note that this routine is purely a helper for filesystem usage and should
2474 * not be called by generic code.
2476 * Unlike lookup_one_len, it should be called without the parent
2477 * i_mutex held, and will take the i_mutex itself if necessary.
2479 struct dentry
*lookup_one_len_unlocked(const char *name
,
2480 struct dentry
*base
, int len
)
2489 this.hash
= full_name_hash(base
, name
, len
);
2491 return ERR_PTR(-EACCES
);
2493 if (unlikely(name
[0] == '.')) {
2494 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2495 return ERR_PTR(-EACCES
);
2499 c
= *(const unsigned char *)name
++;
2500 if (c
== '/' || c
== '\0')
2501 return ERR_PTR(-EACCES
);
2504 * See if the low-level filesystem might want
2505 * to use its own hash..
2507 if (base
->d_flags
& DCACHE_OP_HASH
) {
2508 int err
= base
->d_op
->d_hash(base
, &this);
2510 return ERR_PTR(err
);
2513 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2515 return ERR_PTR(err
);
2517 ret
= lookup_dcache(&this, base
, 0);
2519 ret
= lookup_slow(&this, base
, 0);
2522 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2524 #ifdef CONFIG_UNIX98_PTYS
2525 int path_pts(struct path
*path
)
2527 /* Find something mounted on "pts" in the same directory as
2530 struct dentry
*child
, *parent
;
2534 ret
= path_parent_directory(path
);
2538 parent
= path
->dentry
;
2541 child
= d_hash_and_lookup(parent
, &this);
2545 path
->dentry
= child
;
2552 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2553 struct path
*path
, int *empty
)
2555 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2558 EXPORT_SYMBOL(user_path_at_empty
);
2561 * NB: most callers don't do anything directly with the reference to the
2562 * to struct filename, but the nd->last pointer points into the name string
2563 * allocated by getname. So we must hold the reference to it until all
2564 * path-walking is complete.
2566 static inline struct filename
*
2567 user_path_parent(int dfd
, const char __user
*path
,
2568 struct path
*parent
,
2573 /* only LOOKUP_REVAL is allowed in extra flags */
2574 return filename_parentat(dfd
, getname(path
), flags
& LOOKUP_REVAL
,
2575 parent
, last
, type
);
2579 * mountpoint_last - look up last component for umount
2580 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2581 * @path: pointer to container for result
2583 * This is a special lookup_last function just for umount. In this case, we
2584 * need to resolve the path without doing any revalidation.
2586 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2587 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2588 * in almost all cases, this lookup will be served out of the dcache. The only
2589 * cases where it won't are if nd->last refers to a symlink or the path is
2590 * bogus and it doesn't exist.
2593 * -error: if there was an error during lookup. This includes -ENOENT if the
2594 * lookup found a negative dentry. The nd->path reference will also be
2597 * 0: if we successfully resolved nd->path and found it to not to be a
2598 * symlink that needs to be followed. "path" will also be populated.
2599 * The nd->path reference will also be put.
2601 * 1: if we successfully resolved nd->last and found it to be a symlink
2602 * that needs to be followed. "path" will be populated with the path
2603 * to the link, and nd->path will *not* be put.
2606 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2609 struct dentry
*dentry
;
2610 struct dentry
*dir
= nd
->path
.dentry
;
2612 /* If we're in rcuwalk, drop out of it to handle last component */
2613 if (nd
->flags
& LOOKUP_RCU
) {
2614 if (unlazy_walk(nd
, NULL
, 0))
2618 nd
->flags
&= ~LOOKUP_PARENT
;
2620 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2621 error
= handle_dots(nd
, nd
->last_type
);
2624 dentry
= dget(nd
->path
.dentry
);
2626 dentry
= d_lookup(dir
, &nd
->last
);
2629 * No cached dentry. Mounted dentries are pinned in the
2630 * cache, so that means that this dentry is probably
2631 * a symlink or the path doesn't actually point
2632 * to a mounted dentry.
2634 dentry
= lookup_slow(&nd
->last
, dir
,
2635 nd
->flags
| LOOKUP_NO_REVAL
);
2637 return PTR_ERR(dentry
);
2640 if (d_is_negative(dentry
)) {
2646 path
->dentry
= dentry
;
2647 path
->mnt
= nd
->path
.mnt
;
2648 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
,
2649 d_backing_inode(dentry
), 0);
2650 if (unlikely(error
))
2658 * path_mountpoint - look up a path to be umounted
2659 * @nd: lookup context
2660 * @flags: lookup flags
2661 * @path: pointer to container for result
2663 * Look up the given name, but don't attempt to revalidate the last component.
2664 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2667 path_mountpoint(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2669 const char *s
= path_init(nd
, flags
);
2673 while (!(err
= link_path_walk(s
, nd
)) &&
2674 (err
= mountpoint_last(nd
, path
)) > 0) {
2675 s
= trailing_symlink(nd
);
2686 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2689 struct nameidata nd
;
2692 return PTR_ERR(name
);
2693 set_nameidata(&nd
, dfd
, name
);
2694 error
= path_mountpoint(&nd
, flags
| LOOKUP_RCU
, path
);
2695 if (unlikely(error
== -ECHILD
))
2696 error
= path_mountpoint(&nd
, flags
, path
);
2697 if (unlikely(error
== -ESTALE
))
2698 error
= path_mountpoint(&nd
, flags
| LOOKUP_REVAL
, path
);
2700 audit_inode(name
, path
->dentry
, 0);
2701 restore_nameidata();
2707 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2708 * @dfd: directory file descriptor
2709 * @name: pathname from userland
2710 * @flags: lookup flags
2711 * @path: pointer to container to hold result
2713 * A umount is a special case for path walking. We're not actually interested
2714 * in the inode in this situation, and ESTALE errors can be a problem. We
2715 * simply want track down the dentry and vfsmount attached at the mountpoint
2716 * and avoid revalidating the last component.
2718 * Returns 0 and populates "path" on success.
2721 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2724 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2728 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2731 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2733 EXPORT_SYMBOL(kern_path_mountpoint
);
2735 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2737 kuid_t fsuid
= current_fsuid();
2739 if (uid_eq(inode
->i_uid
, fsuid
))
2741 if (uid_eq(dir
->i_uid
, fsuid
))
2743 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2745 EXPORT_SYMBOL(__check_sticky
);
2748 * Check whether we can remove a link victim from directory dir, check
2749 * whether the type of victim is right.
2750 * 1. We can't do it if dir is read-only (done in permission())
2751 * 2. We should have write and exec permissions on dir
2752 * 3. We can't remove anything from append-only dir
2753 * 4. We can't do anything with immutable dir (done in permission())
2754 * 5. If the sticky bit on dir is set we should either
2755 * a. be owner of dir, or
2756 * b. be owner of victim, or
2757 * c. have CAP_FOWNER capability
2758 * 6. If the victim is append-only or immutable we can't do antyhing with
2759 * links pointing to it.
2760 * 7. If the victim has an unknown uid or gid we can't change the inode.
2761 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2762 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2763 * 10. We can't remove a root or mountpoint.
2764 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2765 * nfs_async_unlink().
2767 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2769 struct inode
*inode
= d_backing_inode(victim
);
2772 if (d_is_negative(victim
))
2776 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2777 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2779 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2785 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2786 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
) || HAS_UNMAPPED_ID(inode
))
2789 if (!d_is_dir(victim
))
2791 if (IS_ROOT(victim
))
2793 } else if (d_is_dir(victim
))
2795 if (IS_DEADDIR(dir
))
2797 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2802 /* Check whether we can create an object with dentry child in directory
2804 * 1. We can't do it if child already exists (open has special treatment for
2805 * this case, but since we are inlined it's OK)
2806 * 2. We can't do it if dir is read-only (done in permission())
2807 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2808 * 4. We should have write and exec permissions on dir
2809 * 5. We can't do it if dir is immutable (done in permission())
2811 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2813 struct user_namespace
*s_user_ns
;
2814 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2817 if (IS_DEADDIR(dir
))
2819 s_user_ns
= dir
->i_sb
->s_user_ns
;
2820 if (!kuid_has_mapping(s_user_ns
, current_fsuid()) ||
2821 !kgid_has_mapping(s_user_ns
, current_fsgid()))
2823 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2827 * p1 and p2 should be directories on the same fs.
2829 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2834 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2838 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
2840 p
= d_ancestor(p2
, p1
);
2842 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2843 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2847 p
= d_ancestor(p1
, p2
);
2849 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2850 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2854 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2855 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2858 EXPORT_SYMBOL(lock_rename
);
2860 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2862 inode_unlock(p1
->d_inode
);
2864 inode_unlock(p2
->d_inode
);
2865 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
2868 EXPORT_SYMBOL(unlock_rename
);
2870 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2873 int error
= may_create(dir
, dentry
);
2877 if (!dir
->i_op
->create
)
2878 return -EACCES
; /* shouldn't it be ENOSYS? */
2881 error
= security_inode_create(dir
, dentry
, mode
);
2884 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2886 fsnotify_create(dir
, dentry
);
2889 EXPORT_SYMBOL(vfs_create
);
2891 bool may_open_dev(const struct path
*path
)
2893 return !(path
->mnt
->mnt_flags
& MNT_NODEV
) &&
2894 !(path
->mnt
->mnt_sb
->s_iflags
& SB_I_NODEV
);
2897 static int may_open(const struct path
*path
, int acc_mode
, int flag
)
2899 struct dentry
*dentry
= path
->dentry
;
2900 struct inode
*inode
= dentry
->d_inode
;
2906 switch (inode
->i_mode
& S_IFMT
) {
2910 if (acc_mode
& MAY_WRITE
)
2915 if (!may_open_dev(path
))
2924 error
= inode_permission(inode
, MAY_OPEN
| acc_mode
);
2929 * An append-only file must be opened in append mode for writing.
2931 if (IS_APPEND(inode
)) {
2932 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2938 /* O_NOATIME can only be set by the owner or superuser */
2939 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2945 static int handle_truncate(struct file
*filp
)
2947 const struct path
*path
= &filp
->f_path
;
2948 struct inode
*inode
= path
->dentry
->d_inode
;
2949 int error
= get_write_access(inode
);
2953 * Refuse to truncate files with mandatory locks held on them.
2955 error
= locks_verify_locked(filp
);
2957 error
= security_path_truncate(path
);
2959 error
= do_truncate(path
->dentry
, 0,
2960 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2963 put_write_access(inode
);
2967 static inline int open_to_namei_flags(int flag
)
2969 if ((flag
& O_ACCMODE
) == 3)
2974 static int may_o_create(const struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2976 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2980 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2984 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2988 * Attempt to atomically look up, create and open a file from a negative
2991 * Returns 0 if successful. The file will have been created and attached to
2992 * @file by the filesystem calling finish_open().
2994 * Returns 1 if the file was looked up only or didn't need creating. The
2995 * caller will need to perform the open themselves. @path will have been
2996 * updated to point to the new dentry. This may be negative.
2998 * Returns an error code otherwise.
3000 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
3001 struct path
*path
, struct file
*file
,
3002 const struct open_flags
*op
,
3003 int open_flag
, umode_t mode
,
3006 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
3007 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
3010 if (!(~open_flag
& (O_EXCL
| O_CREAT
))) /* both O_EXCL and O_CREAT */
3011 open_flag
&= ~O_TRUNC
;
3013 if (nd
->flags
& LOOKUP_DIRECTORY
)
3014 open_flag
|= O_DIRECTORY
;
3016 file
->f_path
.dentry
= DENTRY_NOT_SET
;
3017 file
->f_path
.mnt
= nd
->path
.mnt
;
3018 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
3019 open_to_namei_flags(open_flag
),
3021 d_lookup_done(dentry
);
3024 * We didn't have the inode before the open, so check open
3027 int acc_mode
= op
->acc_mode
;
3028 if (*opened
& FILE_CREATED
) {
3029 WARN_ON(!(open_flag
& O_CREAT
));
3030 fsnotify_create(dir
, dentry
);
3033 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
3034 if (WARN_ON(error
> 0))
3036 } else if (error
> 0) {
3037 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
3040 if (file
->f_path
.dentry
) {
3042 dentry
= file
->f_path
.dentry
;
3044 if (*opened
& FILE_CREATED
)
3045 fsnotify_create(dir
, dentry
);
3046 if (unlikely(d_is_negative(dentry
))) {
3049 path
->dentry
= dentry
;
3050 path
->mnt
= nd
->path
.mnt
;
3060 * Look up and maybe create and open the last component.
3062 * Must be called with i_mutex held on parent.
3064 * Returns 0 if the file was successfully atomically created (if necessary) and
3065 * opened. In this case the file will be returned attached to @file.
3067 * Returns 1 if the file was not completely opened at this time, though lookups
3068 * and creations will have been performed and the dentry returned in @path will
3069 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3070 * specified then a negative dentry may be returned.
3072 * An error code is returned otherwise.
3074 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3075 * cleared otherwise prior to returning.
3077 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
3079 const struct open_flags
*op
,
3080 bool got_write
, int *opened
)
3082 struct dentry
*dir
= nd
->path
.dentry
;
3083 struct inode
*dir_inode
= dir
->d_inode
;
3084 int open_flag
= op
->open_flag
;
3085 struct dentry
*dentry
;
3086 int error
, create_error
= 0;
3087 umode_t mode
= op
->mode
;
3088 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
3090 if (unlikely(IS_DEADDIR(dir_inode
)))
3093 *opened
&= ~FILE_CREATED
;
3094 dentry
= d_lookup(dir
, &nd
->last
);
3097 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
3099 return PTR_ERR(dentry
);
3101 if (d_in_lookup(dentry
))
3104 if (!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
3107 error
= d_revalidate(dentry
, nd
->flags
);
3108 if (likely(error
> 0))
3112 d_invalidate(dentry
);
3116 if (dentry
->d_inode
) {
3117 /* Cached positive dentry: will open in f_op->open */
3122 * Checking write permission is tricky, bacuse we don't know if we are
3123 * going to actually need it: O_CREAT opens should work as long as the
3124 * file exists. But checking existence breaks atomicity. The trick is
3125 * to check access and if not granted clear O_CREAT from the flags.
3127 * Another problem is returing the "right" error value (e.g. for an
3128 * O_EXCL open we want to return EEXIST not EROFS).
3130 if (open_flag
& O_CREAT
) {
3131 if (!IS_POSIXACL(dir
->d_inode
))
3132 mode
&= ~current_umask();
3133 if (unlikely(!got_write
)) {
3134 create_error
= -EROFS
;
3135 open_flag
&= ~O_CREAT
;
3136 if (open_flag
& (O_EXCL
| O_TRUNC
))
3138 /* No side effects, safe to clear O_CREAT */
3140 create_error
= may_o_create(&nd
->path
, dentry
, mode
);
3142 open_flag
&= ~O_CREAT
;
3143 if (open_flag
& O_EXCL
)
3147 } else if ((open_flag
& (O_TRUNC
|O_WRONLY
|O_RDWR
)) &&
3148 unlikely(!got_write
)) {
3150 * No O_CREATE -> atomicity not a requirement -> fall
3151 * back to lookup + open
3156 if (dir_inode
->i_op
->atomic_open
) {
3157 error
= atomic_open(nd
, dentry
, path
, file
, op
, open_flag
,
3159 if (unlikely(error
== -ENOENT
) && create_error
)
3160 error
= create_error
;
3165 if (d_in_lookup(dentry
)) {
3166 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3168 d_lookup_done(dentry
);
3169 if (unlikely(res
)) {
3171 error
= PTR_ERR(res
);
3179 /* Negative dentry, just create the file */
3180 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3181 *opened
|= FILE_CREATED
;
3182 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3183 if (!dir_inode
->i_op
->create
) {
3187 error
= dir_inode
->i_op
->create(dir_inode
, dentry
, mode
,
3188 open_flag
& O_EXCL
);
3191 fsnotify_create(dir_inode
, dentry
);
3193 if (unlikely(create_error
) && !dentry
->d_inode
) {
3194 error
= create_error
;
3198 path
->dentry
= dentry
;
3199 path
->mnt
= nd
->path
.mnt
;
3208 * Handle the last step of open()
3210 static int do_last(struct nameidata
*nd
,
3211 struct file
*file
, const struct open_flags
*op
,
3214 struct dentry
*dir
= nd
->path
.dentry
;
3215 int open_flag
= op
->open_flag
;
3216 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3217 bool got_write
= false;
3218 int acc_mode
= op
->acc_mode
;
3220 struct inode
*inode
;
3224 nd
->flags
&= ~LOOKUP_PARENT
;
3225 nd
->flags
|= op
->intent
;
3227 if (nd
->last_type
!= LAST_NORM
) {
3228 error
= handle_dots(nd
, nd
->last_type
);
3229 if (unlikely(error
))
3234 if (!(open_flag
& O_CREAT
)) {
3235 if (nd
->last
.name
[nd
->last
.len
])
3236 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3237 /* we _can_ be in RCU mode here */
3238 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3239 if (likely(error
> 0))
3245 BUG_ON(nd
->inode
!= dir
->d_inode
);
3246 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3248 /* create side of things */
3250 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3251 * has been cleared when we got to the last component we are
3254 error
= complete_walk(nd
);
3258 audit_inode(nd
->name
, dir
, LOOKUP_PARENT
);
3259 /* trailing slashes? */
3260 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3264 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3265 error
= mnt_want_write(nd
->path
.mnt
);
3269 * do _not_ fail yet - we might not need that or fail with
3270 * a different error; let lookup_open() decide; we'll be
3271 * dropping this one anyway.
3274 if (open_flag
& O_CREAT
)
3275 inode_lock(dir
->d_inode
);
3277 inode_lock_shared(dir
->d_inode
);
3278 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3279 if (open_flag
& O_CREAT
)
3280 inode_unlock(dir
->d_inode
);
3282 inode_unlock_shared(dir
->d_inode
);
3288 if ((*opened
& FILE_CREATED
) ||
3289 !S_ISREG(file_inode(file
)->i_mode
))
3290 will_truncate
= false;
3292 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3296 if (*opened
& FILE_CREATED
) {
3297 /* Don't check for write permission, don't truncate */
3298 open_flag
&= ~O_TRUNC
;
3299 will_truncate
= false;
3301 path_to_nameidata(&path
, nd
);
3302 goto finish_open_created
;
3306 * If atomic_open() acquired write access it is dropped now due to
3307 * possible mount and symlink following (this might be optimized away if
3311 mnt_drop_write(nd
->path
.mnt
);
3315 error
= follow_managed(&path
, nd
);
3316 if (unlikely(error
< 0))
3319 if (unlikely(d_is_negative(path
.dentry
))) {
3320 path_to_nameidata(&path
, nd
);
3325 * create/update audit record if it already exists.
3327 audit_inode(nd
->name
, path
.dentry
, 0);
3329 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3330 path_to_nameidata(&path
, nd
);
3334 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3335 inode
= d_backing_inode(path
.dentry
);
3339 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
,
3341 if (unlikely(error
))
3344 path_to_nameidata(&path
, nd
);
3347 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3349 error
= complete_walk(nd
);
3352 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3354 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3357 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3359 if (!d_is_reg(nd
->path
.dentry
))
3360 will_truncate
= false;
3362 if (will_truncate
) {
3363 error
= mnt_want_write(nd
->path
.mnt
);
3368 finish_open_created
:
3369 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3372 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3373 error
= vfs_open(&nd
->path
, file
, current_cred());
3376 *opened
|= FILE_OPENED
;
3378 error
= open_check_o_direct(file
);
3380 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3381 if (!error
&& will_truncate
)
3382 error
= handle_truncate(file
);
3384 if (unlikely(error
) && (*opened
& FILE_OPENED
))
3386 if (unlikely(error
> 0)) {
3391 mnt_drop_write(nd
->path
.mnt
);
3395 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3396 const struct open_flags
*op
,
3397 struct file
*file
, int *opened
)
3399 static const struct qstr name
= QSTR_INIT("/", 1);
3400 struct dentry
*child
;
3403 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3404 if (unlikely(error
))
3406 error
= mnt_want_write(path
.mnt
);
3407 if (unlikely(error
))
3409 dir
= path
.dentry
->d_inode
;
3410 /* we want directory to be writable */
3411 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
3414 if (!dir
->i_op
->tmpfile
) {
3415 error
= -EOPNOTSUPP
;
3418 child
= d_alloc(path
.dentry
, &name
);
3419 if (unlikely(!child
)) {
3424 path
.dentry
= child
;
3425 error
= dir
->i_op
->tmpfile(dir
, child
, op
->mode
);
3428 audit_inode(nd
->name
, child
, 0);
3429 /* Don't check for other permissions, the inode was just created */
3430 error
= may_open(&path
, 0, op
->open_flag
);
3433 file
->f_path
.mnt
= path
.mnt
;
3434 error
= finish_open(file
, child
, NULL
, opened
);
3437 error
= open_check_o_direct(file
);
3440 } else if (!(op
->open_flag
& O_EXCL
)) {
3441 struct inode
*inode
= file_inode(file
);
3442 spin_lock(&inode
->i_lock
);
3443 inode
->i_state
|= I_LINKABLE
;
3444 spin_unlock(&inode
->i_lock
);
3447 mnt_drop_write(path
.mnt
);
3453 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3456 int error
= path_lookupat(nd
, flags
, &path
);
3458 audit_inode(nd
->name
, path
.dentry
, 0);
3459 error
= vfs_open(&path
, file
, current_cred());
3465 static struct file
*path_openat(struct nameidata
*nd
,
3466 const struct open_flags
*op
, unsigned flags
)
3473 file
= get_empty_filp();
3477 file
->f_flags
= op
->open_flag
;
3479 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3480 error
= do_tmpfile(nd
, flags
, op
, file
, &opened
);
3484 if (unlikely(file
->f_flags
& O_PATH
)) {
3485 error
= do_o_path(nd
, flags
, file
);
3487 opened
|= FILE_OPENED
;
3491 s
= path_init(nd
, flags
);
3496 while (!(error
= link_path_walk(s
, nd
)) &&
3497 (error
= do_last(nd
, file
, op
, &opened
)) > 0) {
3498 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3499 s
= trailing_symlink(nd
);
3507 if (!(opened
& FILE_OPENED
)) {
3511 if (unlikely(error
)) {
3512 if (error
== -EOPENSTALE
) {
3513 if (flags
& LOOKUP_RCU
)
3518 file
= ERR_PTR(error
);
3523 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3524 const struct open_flags
*op
)
3526 struct nameidata nd
;
3527 int flags
= op
->lookup_flags
;
3530 set_nameidata(&nd
, dfd
, pathname
);
3531 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3532 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3533 filp
= path_openat(&nd
, op
, flags
);
3534 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3535 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3536 restore_nameidata();
3540 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3541 const char *name
, const struct open_flags
*op
)
3543 struct nameidata nd
;
3545 struct filename
*filename
;
3546 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3549 nd
.root
.dentry
= dentry
;
3551 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3552 return ERR_PTR(-ELOOP
);
3554 filename
= getname_kernel(name
);
3555 if (IS_ERR(filename
))
3556 return ERR_CAST(filename
);
3558 set_nameidata(&nd
, -1, filename
);
3559 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3560 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3561 file
= path_openat(&nd
, op
, flags
);
3562 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3563 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3564 restore_nameidata();
3569 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3570 struct path
*path
, unsigned int lookup_flags
)
3572 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3577 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3580 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3581 * other flags passed in are ignored!
3583 lookup_flags
&= LOOKUP_REVAL
;
3585 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3587 return ERR_CAST(name
);
3590 * Yucky last component or no last component at all?
3591 * (foo/., foo/.., /////)
3593 if (unlikely(type
!= LAST_NORM
))
3596 /* don't fail immediately if it's r/o, at least try to report other errors */
3597 err2
= mnt_want_write(path
->mnt
);
3599 * Do the final lookup.
3601 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3602 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3603 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3608 if (d_is_positive(dentry
))
3612 * Special case - lookup gave negative, but... we had foo/bar/
3613 * From the vfs_mknod() POV we just have a negative dentry -
3614 * all is fine. Let's be bastards - you had / on the end, you've
3615 * been asking for (non-existent) directory. -ENOENT for you.
3617 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3621 if (unlikely(err2
)) {
3629 dentry
= ERR_PTR(error
);
3631 inode_unlock(path
->dentry
->d_inode
);
3633 mnt_drop_write(path
->mnt
);
3640 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3641 struct path
*path
, unsigned int lookup_flags
)
3643 return filename_create(dfd
, getname_kernel(pathname
),
3644 path
, lookup_flags
);
3646 EXPORT_SYMBOL(kern_path_create
);
3648 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3651 inode_unlock(path
->dentry
->d_inode
);
3652 mnt_drop_write(path
->mnt
);
3655 EXPORT_SYMBOL(done_path_create
);
3657 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3658 struct path
*path
, unsigned int lookup_flags
)
3660 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3662 EXPORT_SYMBOL(user_path_create
);
3664 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3666 int error
= may_create(dir
, dentry
);
3671 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3674 if (!dir
->i_op
->mknod
)
3677 error
= devcgroup_inode_mknod(mode
, dev
);
3681 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3685 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3687 fsnotify_create(dir
, dentry
);
3690 EXPORT_SYMBOL(vfs_mknod
);
3692 static int may_mknod(umode_t mode
)
3694 switch (mode
& S_IFMT
) {
3700 case 0: /* zero mode translates to S_IFREG */
3709 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3712 struct dentry
*dentry
;
3715 unsigned int lookup_flags
= 0;
3717 error
= may_mknod(mode
);
3721 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3723 return PTR_ERR(dentry
);
3725 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3726 mode
&= ~current_umask();
3727 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3730 switch (mode
& S_IFMT
) {
3731 case 0: case S_IFREG
:
3732 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3734 ima_post_path_mknod(dentry
);
3736 case S_IFCHR
: case S_IFBLK
:
3737 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3738 new_decode_dev(dev
));
3740 case S_IFIFO
: case S_IFSOCK
:
3741 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3745 done_path_create(&path
, dentry
);
3746 if (retry_estale(error
, lookup_flags
)) {
3747 lookup_flags
|= LOOKUP_REVAL
;
3753 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3755 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3758 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3760 int error
= may_create(dir
, dentry
);
3761 unsigned max_links
= dir
->i_sb
->s_max_links
;
3766 if (!dir
->i_op
->mkdir
)
3769 mode
&= (S_IRWXUGO
|S_ISVTX
);
3770 error
= security_inode_mkdir(dir
, dentry
, mode
);
3774 if (max_links
&& dir
->i_nlink
>= max_links
)
3777 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3779 fsnotify_mkdir(dir
, dentry
);
3782 EXPORT_SYMBOL(vfs_mkdir
);
3784 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3786 struct dentry
*dentry
;
3789 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3792 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3794 return PTR_ERR(dentry
);
3796 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3797 mode
&= ~current_umask();
3798 error
= security_path_mkdir(&path
, dentry
, mode
);
3800 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3801 done_path_create(&path
, dentry
);
3802 if (retry_estale(error
, lookup_flags
)) {
3803 lookup_flags
|= LOOKUP_REVAL
;
3809 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3811 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3814 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3816 int error
= may_delete(dir
, dentry
, 1);
3821 if (!dir
->i_op
->rmdir
)
3825 inode_lock(dentry
->d_inode
);
3828 if (is_local_mountpoint(dentry
))
3831 error
= security_inode_rmdir(dir
, dentry
);
3835 shrink_dcache_parent(dentry
);
3836 error
= dir
->i_op
->rmdir(dir
, dentry
);
3840 dentry
->d_inode
->i_flags
|= S_DEAD
;
3842 detach_mounts(dentry
);
3845 inode_unlock(dentry
->d_inode
);
3851 EXPORT_SYMBOL(vfs_rmdir
);
3853 static long do_rmdir(int dfd
, const char __user
*pathname
)
3856 struct filename
*name
;
3857 struct dentry
*dentry
;
3861 unsigned int lookup_flags
= 0;
3863 name
= user_path_parent(dfd
, pathname
,
3864 &path
, &last
, &type
, lookup_flags
);
3866 return PTR_ERR(name
);
3880 error
= mnt_want_write(path
.mnt
);
3884 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3885 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3886 error
= PTR_ERR(dentry
);
3889 if (!dentry
->d_inode
) {
3893 error
= security_path_rmdir(&path
, dentry
);
3896 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3900 inode_unlock(path
.dentry
->d_inode
);
3901 mnt_drop_write(path
.mnt
);
3905 if (retry_estale(error
, lookup_flags
)) {
3906 lookup_flags
|= LOOKUP_REVAL
;
3912 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3914 return do_rmdir(AT_FDCWD
, pathname
);
3918 * vfs_unlink - unlink a filesystem object
3919 * @dir: parent directory
3921 * @delegated_inode: returns victim inode, if the inode is delegated.
3923 * The caller must hold dir->i_mutex.
3925 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3926 * return a reference to the inode in delegated_inode. The caller
3927 * should then break the delegation on that inode and retry. Because
3928 * breaking a delegation may take a long time, the caller should drop
3929 * dir->i_mutex before doing so.
3931 * Alternatively, a caller may pass NULL for delegated_inode. This may
3932 * be appropriate for callers that expect the underlying filesystem not
3933 * to be NFS exported.
3935 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3937 struct inode
*target
= dentry
->d_inode
;
3938 int error
= may_delete(dir
, dentry
, 0);
3943 if (!dir
->i_op
->unlink
)
3947 if (is_local_mountpoint(dentry
))
3950 error
= security_inode_unlink(dir
, dentry
);
3952 error
= try_break_deleg(target
, delegated_inode
);
3955 error
= dir
->i_op
->unlink(dir
, dentry
);
3958 detach_mounts(dentry
);
3963 inode_unlock(target
);
3965 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3966 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3967 fsnotify_link_count(target
);
3973 EXPORT_SYMBOL(vfs_unlink
);
3976 * Make sure that the actual truncation of the file will occur outside its
3977 * directory's i_mutex. Truncate can take a long time if there is a lot of
3978 * writeout happening, and we don't want to prevent access to the directory
3979 * while waiting on the I/O.
3981 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3984 struct filename
*name
;
3985 struct dentry
*dentry
;
3989 struct inode
*inode
= NULL
;
3990 struct inode
*delegated_inode
= NULL
;
3991 unsigned int lookup_flags
= 0;
3993 name
= user_path_parent(dfd
, pathname
,
3994 &path
, &last
, &type
, lookup_flags
);
3996 return PTR_ERR(name
);
3999 if (type
!= LAST_NORM
)
4002 error
= mnt_want_write(path
.mnt
);
4006 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4007 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
4008 error
= PTR_ERR(dentry
);
4009 if (!IS_ERR(dentry
)) {
4010 /* Why not before? Because we want correct error value */
4011 if (last
.name
[last
.len
])
4013 inode
= dentry
->d_inode
;
4014 if (d_is_negative(dentry
))
4017 error
= security_path_unlink(&path
, dentry
);
4020 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
4024 inode_unlock(path
.dentry
->d_inode
);
4026 iput(inode
); /* truncate the inode here */
4028 if (delegated_inode
) {
4029 error
= break_deleg_wait(&delegated_inode
);
4033 mnt_drop_write(path
.mnt
);
4037 if (retry_estale(error
, lookup_flags
)) {
4038 lookup_flags
|= LOOKUP_REVAL
;
4045 if (d_is_negative(dentry
))
4047 else if (d_is_dir(dentry
))
4054 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4056 if ((flag
& ~AT_REMOVEDIR
) != 0)
4059 if (flag
& AT_REMOVEDIR
)
4060 return do_rmdir(dfd
, pathname
);
4062 return do_unlinkat(dfd
, pathname
);
4065 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4067 return do_unlinkat(AT_FDCWD
, pathname
);
4070 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
4072 int error
= may_create(dir
, dentry
);
4077 if (!dir
->i_op
->symlink
)
4080 error
= security_inode_symlink(dir
, dentry
, oldname
);
4084 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
4086 fsnotify_create(dir
, dentry
);
4089 EXPORT_SYMBOL(vfs_symlink
);
4091 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4092 int, newdfd
, const char __user
*, newname
)
4095 struct filename
*from
;
4096 struct dentry
*dentry
;
4098 unsigned int lookup_flags
= 0;
4100 from
= getname(oldname
);
4102 return PTR_ERR(from
);
4104 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4105 error
= PTR_ERR(dentry
);
4109 error
= security_path_symlink(&path
, dentry
, from
->name
);
4111 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
4112 done_path_create(&path
, dentry
);
4113 if (retry_estale(error
, lookup_flags
)) {
4114 lookup_flags
|= LOOKUP_REVAL
;
4122 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4124 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4128 * vfs_link - create a new link
4129 * @old_dentry: object to be linked
4131 * @new_dentry: where to create the new link
4132 * @delegated_inode: returns inode needing a delegation break
4134 * The caller must hold dir->i_mutex
4136 * If vfs_link discovers a delegation on the to-be-linked file in need
4137 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4138 * inode in delegated_inode. The caller should then break the delegation
4139 * and retry. Because breaking a delegation may take a long time, the
4140 * caller should drop the i_mutex before doing so.
4142 * Alternatively, a caller may pass NULL for delegated_inode. This may
4143 * be appropriate for callers that expect the underlying filesystem not
4144 * to be NFS exported.
4146 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4148 struct inode
*inode
= old_dentry
->d_inode
;
4149 unsigned max_links
= dir
->i_sb
->s_max_links
;
4155 error
= may_create(dir
, new_dentry
);
4159 if (dir
->i_sb
!= inode
->i_sb
)
4163 * A link to an append-only or immutable file cannot be created.
4165 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4168 * Updating the link count will likely cause i_uid and i_gid to
4169 * be writen back improperly if their true value is unknown to
4172 if (HAS_UNMAPPED_ID(inode
))
4174 if (!dir
->i_op
->link
)
4176 if (S_ISDIR(inode
->i_mode
))
4179 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4184 /* Make sure we don't allow creating hardlink to an unlinked file */
4185 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4187 else if (max_links
&& inode
->i_nlink
>= max_links
)
4190 error
= try_break_deleg(inode
, delegated_inode
);
4192 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4195 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4196 spin_lock(&inode
->i_lock
);
4197 inode
->i_state
&= ~I_LINKABLE
;
4198 spin_unlock(&inode
->i_lock
);
4200 inode_unlock(inode
);
4202 fsnotify_link(dir
, inode
, new_dentry
);
4205 EXPORT_SYMBOL(vfs_link
);
4208 * Hardlinks are often used in delicate situations. We avoid
4209 * security-related surprises by not following symlinks on the
4212 * We don't follow them on the oldname either to be compatible
4213 * with linux 2.0, and to avoid hard-linking to directories
4214 * and other special files. --ADM
4216 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4217 int, newdfd
, const char __user
*, newname
, int, flags
)
4219 struct dentry
*new_dentry
;
4220 struct path old_path
, new_path
;
4221 struct inode
*delegated_inode
= NULL
;
4225 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4228 * To use null names we require CAP_DAC_READ_SEARCH
4229 * This ensures that not everyone will be able to create
4230 * handlink using the passed filedescriptor.
4232 if (flags
& AT_EMPTY_PATH
) {
4233 if (!capable(CAP_DAC_READ_SEARCH
))
4238 if (flags
& AT_SYMLINK_FOLLOW
)
4239 how
|= LOOKUP_FOLLOW
;
4241 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4245 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4246 (how
& LOOKUP_REVAL
));
4247 error
= PTR_ERR(new_dentry
);
4248 if (IS_ERR(new_dentry
))
4252 if (old_path
.mnt
!= new_path
.mnt
)
4254 error
= may_linkat(&old_path
);
4255 if (unlikely(error
))
4257 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4260 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4262 done_path_create(&new_path
, new_dentry
);
4263 if (delegated_inode
) {
4264 error
= break_deleg_wait(&delegated_inode
);
4266 path_put(&old_path
);
4270 if (retry_estale(error
, how
)) {
4271 path_put(&old_path
);
4272 how
|= LOOKUP_REVAL
;
4276 path_put(&old_path
);
4281 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4283 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4287 * vfs_rename - rename a filesystem object
4288 * @old_dir: parent of source
4289 * @old_dentry: source
4290 * @new_dir: parent of destination
4291 * @new_dentry: destination
4292 * @delegated_inode: returns an inode needing a delegation break
4293 * @flags: rename flags
4295 * The caller must hold multiple mutexes--see lock_rename()).
4297 * If vfs_rename discovers a delegation in need of breaking at either
4298 * the source or destination, it will return -EWOULDBLOCK and return a
4299 * reference to the inode in delegated_inode. The caller should then
4300 * break the delegation and retry. Because breaking a delegation may
4301 * take a long time, the caller should drop all locks before doing
4304 * Alternatively, a caller may pass NULL for delegated_inode. This may
4305 * be appropriate for callers that expect the underlying filesystem not
4306 * to be NFS exported.
4308 * The worst of all namespace operations - renaming directory. "Perverted"
4309 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4311 * a) we can get into loop creation.
4312 * b) race potential - two innocent renames can create a loop together.
4313 * That's where 4.4 screws up. Current fix: serialization on
4314 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4316 * c) we have to lock _four_ objects - parents and victim (if it exists),
4317 * and source (if it is not a directory).
4318 * And that - after we got ->i_mutex on parents (until then we don't know
4319 * whether the target exists). Solution: try to be smart with locking
4320 * order for inodes. We rely on the fact that tree topology may change
4321 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4322 * move will be locked. Thus we can rank directories by the tree
4323 * (ancestors first) and rank all non-directories after them.
4324 * That works since everybody except rename does "lock parent, lookup,
4325 * lock child" and rename is under ->s_vfs_rename_mutex.
4326 * HOWEVER, it relies on the assumption that any object with ->lookup()
4327 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4328 * we'd better make sure that there's no link(2) for them.
4329 * d) conversion from fhandle to dentry may come in the wrong moment - when
4330 * we are removing the target. Solution: we will have to grab ->i_mutex
4331 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4332 * ->i_mutex on parents, which works but leads to some truly excessive
4335 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4336 struct inode
*new_dir
, struct dentry
*new_dentry
,
4337 struct inode
**delegated_inode
, unsigned int flags
)
4340 bool is_dir
= d_is_dir(old_dentry
);
4341 const unsigned char *old_name
;
4342 struct inode
*source
= old_dentry
->d_inode
;
4343 struct inode
*target
= new_dentry
->d_inode
;
4344 bool new_is_dir
= false;
4345 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4348 * Check source == target.
4349 * On overlayfs need to look at underlying inodes.
4351 if (d_real_inode(old_dentry
) == d_real_inode(new_dentry
))
4354 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4359 error
= may_create(new_dir
, new_dentry
);
4361 new_is_dir
= d_is_dir(new_dentry
);
4363 if (!(flags
& RENAME_EXCHANGE
))
4364 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4366 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4371 if (!old_dir
->i_op
->rename
)
4375 * If we are going to change the parent - check write permissions,
4376 * we'll need to flip '..'.
4378 if (new_dir
!= old_dir
) {
4380 error
= inode_permission(source
, MAY_WRITE
);
4384 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4385 error
= inode_permission(target
, MAY_WRITE
);
4391 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4396 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4398 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4399 lock_two_nondirectories(source
, target
);
4404 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4407 if (max_links
&& new_dir
!= old_dir
) {
4409 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4411 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4412 old_dir
->i_nlink
>= max_links
)
4415 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4416 shrink_dcache_parent(new_dentry
);
4418 error
= try_break_deleg(source
, delegated_inode
);
4422 if (target
&& !new_is_dir
) {
4423 error
= try_break_deleg(target
, delegated_inode
);
4427 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4428 new_dir
, new_dentry
, flags
);
4432 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4434 target
->i_flags
|= S_DEAD
;
4435 dont_mount(new_dentry
);
4436 detach_mounts(new_dentry
);
4438 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4439 if (!(flags
& RENAME_EXCHANGE
))
4440 d_move(old_dentry
, new_dentry
);
4442 d_exchange(old_dentry
, new_dentry
);
4445 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4446 unlock_two_nondirectories(source
, target
);
4448 inode_unlock(target
);
4451 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4452 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4453 if (flags
& RENAME_EXCHANGE
) {
4454 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4455 new_is_dir
, NULL
, new_dentry
);
4458 fsnotify_oldname_free(old_name
);
4462 EXPORT_SYMBOL(vfs_rename
);
4464 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4465 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4467 struct dentry
*old_dentry
, *new_dentry
;
4468 struct dentry
*trap
;
4469 struct path old_path
, new_path
;
4470 struct qstr old_last
, new_last
;
4471 int old_type
, new_type
;
4472 struct inode
*delegated_inode
= NULL
;
4473 struct filename
*from
;
4474 struct filename
*to
;
4475 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4476 bool should_retry
= false;
4479 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4482 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4483 (flags
& RENAME_EXCHANGE
))
4486 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4489 if (flags
& RENAME_EXCHANGE
)
4493 from
= user_path_parent(olddfd
, oldname
,
4494 &old_path
, &old_last
, &old_type
, lookup_flags
);
4496 error
= PTR_ERR(from
);
4500 to
= user_path_parent(newdfd
, newname
,
4501 &new_path
, &new_last
, &new_type
, lookup_flags
);
4503 error
= PTR_ERR(to
);
4508 if (old_path
.mnt
!= new_path
.mnt
)
4512 if (old_type
!= LAST_NORM
)
4515 if (flags
& RENAME_NOREPLACE
)
4517 if (new_type
!= LAST_NORM
)
4520 error
= mnt_want_write(old_path
.mnt
);
4525 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4527 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4528 error
= PTR_ERR(old_dentry
);
4529 if (IS_ERR(old_dentry
))
4531 /* source must exist */
4533 if (d_is_negative(old_dentry
))
4535 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4536 error
= PTR_ERR(new_dentry
);
4537 if (IS_ERR(new_dentry
))
4540 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4542 if (flags
& RENAME_EXCHANGE
) {
4544 if (d_is_negative(new_dentry
))
4547 if (!d_is_dir(new_dentry
)) {
4549 if (new_last
.name
[new_last
.len
])
4553 /* unless the source is a directory trailing slashes give -ENOTDIR */
4554 if (!d_is_dir(old_dentry
)) {
4556 if (old_last
.name
[old_last
.len
])
4558 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4561 /* source should not be ancestor of target */
4563 if (old_dentry
== trap
)
4565 /* target should not be an ancestor of source */
4566 if (!(flags
& RENAME_EXCHANGE
))
4568 if (new_dentry
== trap
)
4571 error
= security_path_rename(&old_path
, old_dentry
,
4572 &new_path
, new_dentry
, flags
);
4575 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4576 new_path
.dentry
->d_inode
, new_dentry
,
4577 &delegated_inode
, flags
);
4583 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4584 if (delegated_inode
) {
4585 error
= break_deleg_wait(&delegated_inode
);
4589 mnt_drop_write(old_path
.mnt
);
4591 if (retry_estale(error
, lookup_flags
))
4592 should_retry
= true;
4593 path_put(&new_path
);
4596 path_put(&old_path
);
4599 should_retry
= false;
4600 lookup_flags
|= LOOKUP_REVAL
;
4607 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4608 int, newdfd
, const char __user
*, newname
)
4610 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4613 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4615 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4618 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4620 int error
= may_create(dir
, dentry
);
4624 if (!dir
->i_op
->mknod
)
4627 return dir
->i_op
->mknod(dir
, dentry
,
4628 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4630 EXPORT_SYMBOL(vfs_whiteout
);
4632 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4634 int len
= PTR_ERR(link
);
4639 if (len
> (unsigned) buflen
)
4641 if (copy_to_user(buffer
, link
, len
))
4648 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4649 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4650 * for any given inode is up to filesystem.
4652 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4654 DEFINE_DELAYED_CALL(done
);
4655 struct inode
*inode
= d_inode(dentry
);
4656 const char *link
= inode
->i_link
;
4660 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4662 return PTR_ERR(link
);
4664 res
= readlink_copy(buffer
, buflen
, link
);
4665 do_delayed_call(&done
);
4668 EXPORT_SYMBOL(generic_readlink
);
4671 * vfs_get_link - get symlink body
4672 * @dentry: dentry on which to get symbolic link
4673 * @done: caller needs to free returned data with this
4675 * Calls security hook and i_op->get_link() on the supplied inode.
4677 * It does not touch atime. That's up to the caller if necessary.
4679 * Does not work on "special" symlinks like /proc/$$/fd/N
4681 const char *vfs_get_link(struct dentry
*dentry
, struct delayed_call
*done
)
4683 const char *res
= ERR_PTR(-EINVAL
);
4684 struct inode
*inode
= d_inode(dentry
);
4686 if (d_is_symlink(dentry
)) {
4687 res
= ERR_PTR(security_inode_readlink(dentry
));
4689 res
= inode
->i_op
->get_link(dentry
, inode
, done
);
4693 EXPORT_SYMBOL(vfs_get_link
);
4695 /* get the link contents into pagecache */
4696 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4697 struct delayed_call
*callback
)
4701 struct address_space
*mapping
= inode
->i_mapping
;
4704 page
= find_get_page(mapping
, 0);
4706 return ERR_PTR(-ECHILD
);
4707 if (!PageUptodate(page
)) {
4709 return ERR_PTR(-ECHILD
);
4712 page
= read_mapping_page(mapping
, 0, NULL
);
4716 set_delayed_call(callback
, page_put_link
, page
);
4717 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4718 kaddr
= page_address(page
);
4719 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4723 EXPORT_SYMBOL(page_get_link
);
4725 void page_put_link(void *arg
)
4729 EXPORT_SYMBOL(page_put_link
);
4731 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4733 DEFINE_DELAYED_CALL(done
);
4734 int res
= readlink_copy(buffer
, buflen
,
4735 page_get_link(dentry
, d_inode(dentry
),
4737 do_delayed_call(&done
);
4740 EXPORT_SYMBOL(page_readlink
);
4743 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4745 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4747 struct address_space
*mapping
= inode
->i_mapping
;
4751 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4753 flags
|= AOP_FLAG_NOFS
;
4756 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4757 flags
, &page
, &fsdata
);
4761 memcpy(page_address(page
), symname
, len
-1);
4763 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4770 mark_inode_dirty(inode
);
4775 EXPORT_SYMBOL(__page_symlink
);
4777 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4779 return __page_symlink(inode
, symname
, len
,
4780 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4782 EXPORT_SYMBOL(page_symlink
);
4784 const struct inode_operations page_symlink_inode_operations
= {
4785 .readlink
= generic_readlink
,
4786 .get_link
= page_get_link
,
4788 EXPORT_SYMBOL(page_symlink_inode_operations
);