1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user
*filename
, int flags
, int *empty
)
130 struct filename
*result
;
134 result
= audit_reusename(filename
);
138 result
= __getname();
139 if (unlikely(!result
))
140 return ERR_PTR(-ENOMEM
);
143 * First, try to embed the struct filename inside the names_cache
146 kname
= (char *)result
->iname
;
147 result
->name
= kname
;
149 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
150 if (unlikely(len
< 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
162 const size_t size
= offsetof(struct filename
, iname
[1]);
163 kname
= (char *)result
;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result
= kzalloc(size
, GFP_KERNEL
);
171 if (unlikely(!result
)) {
173 return ERR_PTR(-ENOMEM
);
175 result
->name
= kname
;
176 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
177 if (unlikely(len
< 0)) {
182 if (unlikely(len
== PATH_MAX
)) {
185 return ERR_PTR(-ENAMETOOLONG
);
190 /* The empty path is special. */
191 if (unlikely(!len
)) {
194 if (!(flags
& LOOKUP_EMPTY
)) {
196 return ERR_PTR(-ENOENT
);
200 result
->uptr
= filename
;
201 result
->aname
= NULL
;
202 audit_getname(result
);
207 getname(const char __user
* filename
)
209 return getname_flags(filename
, 0, NULL
);
213 getname_kernel(const char * filename
)
215 struct filename
*result
;
216 int len
= strlen(filename
) + 1;
218 result
= __getname();
219 if (unlikely(!result
))
220 return ERR_PTR(-ENOMEM
);
222 if (len
<= EMBEDDED_NAME_MAX
) {
223 result
->name
= (char *)result
->iname
;
224 } else if (len
<= PATH_MAX
) {
225 const size_t size
= offsetof(struct filename
, iname
[1]);
226 struct filename
*tmp
;
228 tmp
= kmalloc(size
, GFP_KERNEL
);
229 if (unlikely(!tmp
)) {
231 return ERR_PTR(-ENOMEM
);
233 tmp
->name
= (char *)result
;
237 return ERR_PTR(-ENAMETOOLONG
);
239 memcpy((char *)result
->name
, filename
, len
);
241 result
->aname
= NULL
;
243 audit_getname(result
);
248 void putname(struct filename
*name
)
250 BUG_ON(name
->refcnt
<= 0);
252 if (--name
->refcnt
> 0)
255 if (name
->name
!= name
->iname
) {
256 __putname(name
->name
);
262 static int check_acl(struct inode
*inode
, int mask
)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl
*acl
;
267 if (mask
& MAY_NOT_BLOCK
) {
268 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl
))
274 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
277 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
281 int error
= posix_acl_permission(inode
, acl
, mask
);
282 posix_acl_release(acl
);
291 * This does the basic permission checking
293 static int acl_permission_check(struct inode
*inode
, int mask
)
295 unsigned int mode
= inode
->i_mode
;
297 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
300 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
301 int error
= check_acl(inode
, mask
);
302 if (error
!= -EAGAIN
)
306 if (in_group_p(inode
->i_gid
))
311 * If the DACs are ok we don't need any capability check.
313 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
332 int generic_permission(struct inode
*inode
, int mask
)
337 * Do the basic permission checks.
339 ret
= acl_permission_check(inode
, mask
);
343 if (S_ISDIR(inode
->i_mode
)) {
344 /* DACs are overridable for directories */
345 if (!(mask
& MAY_WRITE
))
346 if (capable_wrt_inode_uidgid(inode
,
347 CAP_DAC_READ_SEARCH
))
349 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
355 * Searching includes executable on directories, else just read.
357 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
358 if (mask
== MAY_READ
)
359 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
366 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
367 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
372 EXPORT_SYMBOL(generic_permission
);
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
380 static inline int do_inode_permission(struct vfsmount
*mnt
, struct inode
*inode
, int mask
)
382 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
383 if (likely(mnt
&& inode
->i_op
->permission2
))
384 return inode
->i_op
->permission2(mnt
, inode
, mask
);
385 if (likely(inode
->i_op
->permission
))
386 return inode
->i_op
->permission(inode
, mask
);
388 /* This gets set once for the inode lifetime */
389 spin_lock(&inode
->i_lock
);
390 inode
->i_opflags
|= IOP_FASTPERM
;
391 spin_unlock(&inode
->i_lock
);
393 return generic_permission(inode
, mask
);
397 * __inode_permission - Check for access rights to a given inode
398 * @inode: Inode to check permission on
399 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
401 * Check for read/write/execute permissions on an inode.
403 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
405 * This does not check for a read-only file system. You probably want
406 * inode_permission().
408 int __inode_permission2(struct vfsmount
*mnt
, struct inode
*inode
, int mask
)
412 if (unlikely(mask
& MAY_WRITE
)) {
414 * Nobody gets write access to an immutable file.
416 if (IS_IMMUTABLE(inode
))
420 * Updating mtime will likely cause i_uid and i_gid to be
421 * written back improperly if their true value is unknown
424 if (HAS_UNMAPPED_ID(inode
))
428 retval
= do_inode_permission(mnt
, inode
, mask
);
432 retval
= devcgroup_inode_permission(inode
, mask
);
436 retval
= security_inode_permission(inode
, mask
);
439 EXPORT_SYMBOL(__inode_permission2
);
441 int __inode_permission(struct inode
*inode
, int mask
)
443 return __inode_permission2(NULL
, inode
, mask
);
445 EXPORT_SYMBOL(__inode_permission
);
448 * sb_permission - Check superblock-level permissions
449 * @sb: Superblock of inode to check permission on
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
453 * Separate out file-system wide checks from inode-specific permission checks.
455 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
457 if (unlikely(mask
& MAY_WRITE
)) {
458 umode_t mode
= inode
->i_mode
;
460 /* Nobody gets write access to a read-only fs. */
461 if (sb_rdonly(sb
) && (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
468 * inode_permission - Check for access rights to a given inode
469 * @inode: Inode to check permission on
470 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
472 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
473 * this, letting us set arbitrary permissions for filesystem access without
474 * changing the "normal" UIDs which are used for other things.
476 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
478 int inode_permission2(struct vfsmount
*mnt
, struct inode
*inode
, int mask
)
482 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
485 return __inode_permission2(mnt
, inode
, mask
);
487 EXPORT_SYMBOL(inode_permission2
);
489 int inode_permission(struct inode
*inode
, int mask
)
491 return inode_permission2(NULL
, inode
, mask
);
493 EXPORT_SYMBOL(inode_permission
);
496 * path_get - get a reference to a path
497 * @path: path to get the reference to
499 * Given a path increment the reference count to the dentry and the vfsmount.
501 void path_get(const struct path
*path
)
506 EXPORT_SYMBOL(path_get
);
509 * path_put - put a reference to a path
510 * @path: path to put the reference to
512 * Given a path decrement the reference count to the dentry and the vfsmount.
514 void path_put(const struct path
*path
)
519 EXPORT_SYMBOL(path_put
);
521 #define EMBEDDED_LEVELS 2
526 struct inode
*inode
; /* path.dentry.d_inode */
531 int total_link_count
;
534 struct delayed_call done
;
537 } *stack
, internal
[EMBEDDED_LEVELS
];
538 struct filename
*name
;
539 struct nameidata
*saved
;
540 struct inode
*link_inode
;
543 } __randomize_layout
;
545 static void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
547 struct nameidata
*old
= current
->nameidata
;
548 p
->stack
= p
->internal
;
551 p
->total_link_count
= old
? old
->total_link_count
: 0;
553 current
->nameidata
= p
;
556 static void restore_nameidata(void)
558 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
560 current
->nameidata
= old
;
562 old
->total_link_count
= now
->total_link_count
;
563 if (now
->stack
!= now
->internal
)
567 static int __nd_alloc_stack(struct nameidata
*nd
)
571 if (nd
->flags
& LOOKUP_RCU
) {
572 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
577 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
582 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
588 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
589 * @path: nameidate to verify
591 * Rename can sometimes move a file or directory outside of a bind
592 * mount, path_connected allows those cases to be detected.
594 static bool path_connected(const struct path
*path
)
596 struct vfsmount
*mnt
= path
->mnt
;
597 struct super_block
*sb
= mnt
->mnt_sb
;
599 /* Bind mounts and multi-root filesystems can have disconnected paths */
600 if (!(sb
->s_iflags
& SB_I_MULTIROOT
) && (mnt
->mnt_root
== sb
->s_root
))
603 return is_subdir(path
->dentry
, mnt
->mnt_root
);
606 static inline int nd_alloc_stack(struct nameidata
*nd
)
608 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
610 if (likely(nd
->stack
!= nd
->internal
))
612 return __nd_alloc_stack(nd
);
615 static void drop_links(struct nameidata
*nd
)
619 struct saved
*last
= nd
->stack
+ i
;
620 do_delayed_call(&last
->done
);
621 clear_delayed_call(&last
->done
);
625 static void terminate_walk(struct nameidata
*nd
)
628 if (!(nd
->flags
& LOOKUP_RCU
)) {
631 for (i
= 0; i
< nd
->depth
; i
++)
632 path_put(&nd
->stack
[i
].link
);
633 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
638 nd
->flags
&= ~LOOKUP_RCU
;
639 if (!(nd
->flags
& LOOKUP_ROOT
))
646 /* path_put is needed afterwards regardless of success or failure */
647 static bool legitimize_path(struct nameidata
*nd
,
648 struct path
*path
, unsigned seq
)
650 int res
= __legitimize_mnt(path
->mnt
, nd
->m_seq
);
657 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
661 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
664 static bool legitimize_links(struct nameidata
*nd
)
667 for (i
= 0; i
< nd
->depth
; i
++) {
668 struct saved
*last
= nd
->stack
+ i
;
669 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
679 * Path walking has 2 modes, rcu-walk and ref-walk (see
680 * Documentation/filesystems/path-lookup.txt). In situations when we can't
681 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
682 * normal reference counts on dentries and vfsmounts to transition to ref-walk
683 * mode. Refcounts are grabbed at the last known good point before rcu-walk
684 * got stuck, so ref-walk may continue from there. If this is not successful
685 * (eg. a seqcount has changed), then failure is returned and it's up to caller
686 * to restart the path walk from the beginning in ref-walk mode.
690 * unlazy_walk - try to switch to ref-walk mode.
691 * @nd: nameidata pathwalk data
692 * Returns: 0 on success, -ECHILD on failure
694 * unlazy_walk attempts to legitimize the current nd->path and nd->root
696 * Must be called from rcu-walk context.
697 * Nothing should touch nameidata between unlazy_walk() failure and
700 static int unlazy_walk(struct nameidata
*nd
)
702 struct dentry
*parent
= nd
->path
.dentry
;
704 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
706 nd
->flags
&= ~LOOKUP_RCU
;
707 if (unlikely(!legitimize_links(nd
)))
709 if (unlikely(!legitimize_path(nd
, &nd
->path
, nd
->seq
)))
711 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
712 if (unlikely(!legitimize_path(nd
, &nd
->root
, nd
->root_seq
)))
716 BUG_ON(nd
->inode
!= parent
->d_inode
);
721 nd
->path
.dentry
= NULL
;
723 if (!(nd
->flags
& LOOKUP_ROOT
))
731 * unlazy_child - try to switch to ref-walk mode.
732 * @nd: nameidata pathwalk data
733 * @dentry: child of nd->path.dentry
734 * @seq: seq number to check dentry against
735 * Returns: 0 on success, -ECHILD on failure
737 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
738 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
739 * @nd. Must be called from rcu-walk context.
740 * Nothing should touch nameidata between unlazy_child() failure and
743 static int unlazy_child(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
745 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
747 nd
->flags
&= ~LOOKUP_RCU
;
748 if (unlikely(!legitimize_links(nd
)))
750 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
752 if (unlikely(!lockref_get_not_dead(&nd
->path
.dentry
->d_lockref
)))
756 * We need to move both the parent and the dentry from the RCU domain
757 * to be properly refcounted. And the sequence number in the dentry
758 * validates *both* dentry counters, since we checked the sequence
759 * number of the parent after we got the child sequence number. So we
760 * know the parent must still be valid if the child sequence number is
762 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
)))
764 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
))) {
770 * Sequence counts matched. Now make sure that the root is
771 * still valid and get it if required.
773 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
774 if (unlikely(!legitimize_path(nd
, &nd
->root
, nd
->root_seq
))) {
787 nd
->path
.dentry
= NULL
;
791 if (!(nd
->flags
& LOOKUP_ROOT
))
796 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
798 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
799 return dentry
->d_op
->d_revalidate(dentry
, flags
);
805 * complete_walk - successful completion of path walk
806 * @nd: pointer nameidata
808 * If we had been in RCU mode, drop out of it and legitimize nd->path.
809 * Revalidate the final result, unless we'd already done that during
810 * the path walk or the filesystem doesn't ask for it. Return 0 on
811 * success, -error on failure. In case of failure caller does not
812 * need to drop nd->path.
814 static int complete_walk(struct nameidata
*nd
)
816 struct dentry
*dentry
= nd
->path
.dentry
;
819 if (nd
->flags
& LOOKUP_RCU
) {
820 if (!(nd
->flags
& LOOKUP_ROOT
))
822 if (unlikely(unlazy_walk(nd
)))
826 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
829 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
832 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
842 static void set_root(struct nameidata
*nd
)
844 struct fs_struct
*fs
= current
->fs
;
846 if (nd
->flags
& LOOKUP_RCU
) {
850 seq
= read_seqcount_begin(&fs
->seq
);
852 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
853 } while (read_seqcount_retry(&fs
->seq
, seq
));
855 get_fs_root(fs
, &nd
->root
);
859 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
862 if (path
->mnt
!= nd
->path
.mnt
)
866 static inline void path_to_nameidata(const struct path
*path
,
867 struct nameidata
*nd
)
869 if (!(nd
->flags
& LOOKUP_RCU
)) {
870 dput(nd
->path
.dentry
);
871 if (nd
->path
.mnt
!= path
->mnt
)
872 mntput(nd
->path
.mnt
);
874 nd
->path
.mnt
= path
->mnt
;
875 nd
->path
.dentry
= path
->dentry
;
878 static int nd_jump_root(struct nameidata
*nd
)
880 if (nd
->flags
& LOOKUP_RCU
) {
884 nd
->inode
= d
->d_inode
;
885 nd
->seq
= nd
->root_seq
;
886 if (unlikely(read_seqcount_retry(&d
->d_seq
, nd
->seq
)))
892 nd
->inode
= nd
->path
.dentry
->d_inode
;
894 nd
->flags
|= LOOKUP_JUMPED
;
899 * Helper to directly jump to a known parsed path from ->get_link,
900 * caller must have taken a reference to path beforehand.
902 void nd_jump_link(struct path
*path
)
904 struct nameidata
*nd
= current
->nameidata
;
908 nd
->inode
= nd
->path
.dentry
->d_inode
;
909 nd
->flags
|= LOOKUP_JUMPED
;
912 static inline void put_link(struct nameidata
*nd
)
914 struct saved
*last
= nd
->stack
+ --nd
->depth
;
915 do_delayed_call(&last
->done
);
916 if (!(nd
->flags
& LOOKUP_RCU
))
917 path_put(&last
->link
);
920 int sysctl_protected_symlinks __read_mostly
= 0;
921 int sysctl_protected_hardlinks __read_mostly
= 0;
922 int sysctl_protected_fifos __read_mostly
;
923 int sysctl_protected_regular __read_mostly
;
926 * may_follow_link - Check symlink following for unsafe situations
927 * @nd: nameidata pathwalk data
929 * In the case of the sysctl_protected_symlinks sysctl being enabled,
930 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
931 * in a sticky world-writable directory. This is to protect privileged
932 * processes from failing races against path names that may change out
933 * from under them by way of other users creating malicious symlinks.
934 * It will permit symlinks to be followed only when outside a sticky
935 * world-writable directory, or when the uid of the symlink and follower
936 * match, or when the directory owner matches the symlink's owner.
938 * Returns 0 if following the symlink is allowed, -ve on error.
940 static inline int may_follow_link(struct nameidata
*nd
)
942 const struct inode
*inode
;
943 const struct inode
*parent
;
946 if (!sysctl_protected_symlinks
)
949 /* Allowed if owner and follower match. */
950 inode
= nd
->link_inode
;
951 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
954 /* Allowed if parent directory not sticky and world-writable. */
956 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
959 /* Allowed if parent directory and link owner match. */
960 puid
= parent
->i_uid
;
961 if (uid_valid(puid
) && uid_eq(puid
, inode
->i_uid
))
964 if (nd
->flags
& LOOKUP_RCU
)
967 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
972 * safe_hardlink_source - Check for safe hardlink conditions
973 * @inode: the source inode to hardlink from
975 * Return false if at least one of the following conditions:
976 * - inode is not a regular file
978 * - inode is setgid and group-exec
979 * - access failure for read and write
981 * Otherwise returns true.
983 static bool safe_hardlink_source(struct inode
*inode
)
985 umode_t mode
= inode
->i_mode
;
987 /* Special files should not get pinned to the filesystem. */
991 /* Setuid files should not get pinned to the filesystem. */
995 /* Executable setgid files should not get pinned to the filesystem. */
996 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
999 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1000 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
1007 * may_linkat - Check permissions for creating a hardlink
1008 * @link: the source to hardlink from
1010 * Block hardlink when all of:
1011 * - sysctl_protected_hardlinks enabled
1012 * - fsuid does not match inode
1013 * - hardlink source is unsafe (see safe_hardlink_source() above)
1014 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1016 * Returns 0 if successful, -ve on error.
1018 static int may_linkat(struct path
*link
)
1020 struct inode
*inode
;
1022 if (!sysctl_protected_hardlinks
)
1025 inode
= link
->dentry
->d_inode
;
1027 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1028 * otherwise, it must be a safe source.
1030 if (safe_hardlink_source(inode
) || inode_owner_or_capable(inode
))
1033 audit_log_link_denied("linkat", link
);
1038 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1039 * should be allowed, or not, on files that already
1041 * @dir: the sticky parent directory
1042 * @inode: the inode of the file to open
1044 * Block an O_CREAT open of a FIFO (or a regular file) when:
1045 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1046 * - the file already exists
1047 * - we are in a sticky directory
1048 * - we don't own the file
1049 * - the owner of the directory doesn't own the file
1050 * - the directory is world writable
1051 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1052 * the directory doesn't have to be world writable: being group writable will
1055 * Returns 0 if the open is allowed, -ve on error.
1057 static int may_create_in_sticky(struct dentry
* const dir
,
1058 struct inode
* const inode
)
1060 if ((!sysctl_protected_fifos
&& S_ISFIFO(inode
->i_mode
)) ||
1061 (!sysctl_protected_regular
&& S_ISREG(inode
->i_mode
)) ||
1062 likely(!(dir
->d_inode
->i_mode
& S_ISVTX
)) ||
1063 uid_eq(inode
->i_uid
, dir
->d_inode
->i_uid
) ||
1064 uid_eq(current_fsuid(), inode
->i_uid
))
1067 if (likely(dir
->d_inode
->i_mode
& 0002) ||
1068 (dir
->d_inode
->i_mode
& 0020 &&
1069 ((sysctl_protected_fifos
>= 2 && S_ISFIFO(inode
->i_mode
)) ||
1070 (sysctl_protected_regular
>= 2 && S_ISREG(inode
->i_mode
))))) {
1076 static __always_inline
1077 const char *get_link(struct nameidata
*nd
)
1079 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
1080 struct dentry
*dentry
= last
->link
.dentry
;
1081 struct inode
*inode
= nd
->link_inode
;
1085 if (!(nd
->flags
& LOOKUP_RCU
)) {
1086 touch_atime(&last
->link
);
1088 } else if (atime_needs_update_rcu(&last
->link
, inode
)) {
1089 if (unlikely(unlazy_walk(nd
)))
1090 return ERR_PTR(-ECHILD
);
1091 touch_atime(&last
->link
);
1094 error
= security_inode_follow_link(dentry
, inode
,
1095 nd
->flags
& LOOKUP_RCU
);
1096 if (unlikely(error
))
1097 return ERR_PTR(error
);
1099 nd
->last_type
= LAST_BIND
;
1100 res
= inode
->i_link
;
1102 const char * (*get
)(struct dentry
*, struct inode
*,
1103 struct delayed_call
*);
1104 get
= inode
->i_op
->get_link
;
1105 if (nd
->flags
& LOOKUP_RCU
) {
1106 res
= get(NULL
, inode
, &last
->done
);
1107 if (res
== ERR_PTR(-ECHILD
)) {
1108 if (unlikely(unlazy_walk(nd
)))
1109 return ERR_PTR(-ECHILD
);
1110 res
= get(dentry
, inode
, &last
->done
);
1113 res
= get(dentry
, inode
, &last
->done
);
1115 if (IS_ERR_OR_NULL(res
))
1121 if (unlikely(nd_jump_root(nd
)))
1122 return ERR_PTR(-ECHILD
);
1123 while (unlikely(*++res
== '/'))
1132 * follow_up - Find the mountpoint of path's vfsmount
1134 * Given a path, find the mountpoint of its source file system.
1135 * Replace @path with the path of the mountpoint in the parent mount.
1138 * Return 1 if we went up a level and 0 if we were already at the
1141 int follow_up(struct path
*path
)
1143 struct mount
*mnt
= real_mount(path
->mnt
);
1144 struct mount
*parent
;
1145 struct dentry
*mountpoint
;
1147 read_seqlock_excl(&mount_lock
);
1148 parent
= mnt
->mnt_parent
;
1149 if (parent
== mnt
) {
1150 read_sequnlock_excl(&mount_lock
);
1153 mntget(&parent
->mnt
);
1154 mountpoint
= dget(mnt
->mnt_mountpoint
);
1155 read_sequnlock_excl(&mount_lock
);
1157 path
->dentry
= mountpoint
;
1159 path
->mnt
= &parent
->mnt
;
1162 EXPORT_SYMBOL(follow_up
);
1165 * Perform an automount
1166 * - return -EISDIR to tell follow_managed() to stop and return the path we
1169 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
1172 struct vfsmount
*mnt
;
1175 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
1178 /* We don't want to mount if someone's just doing a stat -
1179 * unless they're stat'ing a directory and appended a '/' to
1182 * We do, however, want to mount if someone wants to open or
1183 * create a file of any type under the mountpoint, wants to
1184 * traverse through the mountpoint or wants to open the
1185 * mounted directory. Also, autofs may mark negative dentries
1186 * as being automount points. These will need the attentions
1187 * of the daemon to instantiate them before they can be used.
1189 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1190 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1191 path
->dentry
->d_inode
)
1194 nd
->total_link_count
++;
1195 if (nd
->total_link_count
>= 40)
1198 mnt
= path
->dentry
->d_op
->d_automount(path
);
1201 * The filesystem is allowed to return -EISDIR here to indicate
1202 * it doesn't want to automount. For instance, autofs would do
1203 * this so that its userspace daemon can mount on this dentry.
1205 * However, we can only permit this if it's a terminal point in
1206 * the path being looked up; if it wasn't then the remainder of
1207 * the path is inaccessible and we should say so.
1209 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1211 return PTR_ERR(mnt
);
1214 if (!mnt
) /* mount collision */
1217 if (!*need_mntput
) {
1218 /* lock_mount() may release path->mnt on error */
1220 *need_mntput
= true;
1222 err
= finish_automount(mnt
, path
);
1226 /* Someone else made a mount here whilst we were busy */
1231 path
->dentry
= dget(mnt
->mnt_root
);
1240 * Handle a dentry that is managed in some way.
1241 * - Flagged for transit management (autofs)
1242 * - Flagged as mountpoint
1243 * - Flagged as automount point
1245 * This may only be called in refwalk mode.
1247 * Serialization is taken care of in namespace.c
1249 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1251 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1253 bool need_mntput
= false;
1256 /* Given that we're not holding a lock here, we retain the value in a
1257 * local variable for each dentry as we look at it so that we don't see
1258 * the components of that value change under us */
1259 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1260 managed
&= DCACHE_MANAGED_DENTRY
,
1261 unlikely(managed
!= 0)) {
1262 /* Allow the filesystem to manage the transit without i_mutex
1264 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1265 BUG_ON(!path
->dentry
->d_op
);
1266 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1267 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1272 /* Transit to a mounted filesystem. */
1273 if (managed
& DCACHE_MOUNTED
) {
1274 struct vfsmount
*mounted
= lookup_mnt(path
);
1279 path
->mnt
= mounted
;
1280 path
->dentry
= dget(mounted
->mnt_root
);
1285 /* Something is mounted on this dentry in another
1286 * namespace and/or whatever was mounted there in this
1287 * namespace got unmounted before lookup_mnt() could
1291 /* Handle an automount point */
1292 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1293 ret
= follow_automount(path
, nd
, &need_mntput
);
1299 /* We didn't change the current path point */
1303 if (need_mntput
&& path
->mnt
== mnt
)
1305 if (ret
== -EISDIR
|| !ret
)
1308 nd
->flags
|= LOOKUP_JUMPED
;
1309 if (unlikely(ret
< 0))
1310 path_put_conditional(path
, nd
);
1314 int follow_down_one(struct path
*path
)
1316 struct vfsmount
*mounted
;
1318 mounted
= lookup_mnt(path
);
1322 path
->mnt
= mounted
;
1323 path
->dentry
= dget(mounted
->mnt_root
);
1328 EXPORT_SYMBOL(follow_down_one
);
1330 static inline int managed_dentry_rcu(const struct path
*path
)
1332 return (path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1333 path
->dentry
->d_op
->d_manage(path
, true) : 0;
1337 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1338 * we meet a managed dentry that would need blocking.
1340 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1341 struct inode
**inode
, unsigned *seqp
)
1344 struct mount
*mounted
;
1346 * Don't forget we might have a non-mountpoint managed dentry
1347 * that wants to block transit.
1349 switch (managed_dentry_rcu(path
)) {
1359 if (!d_mountpoint(path
->dentry
))
1360 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1362 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1365 path
->mnt
= &mounted
->mnt
;
1366 path
->dentry
= mounted
->mnt
.mnt_root
;
1367 nd
->flags
|= LOOKUP_JUMPED
;
1368 *seqp
= read_seqcount_begin(&path
->dentry
->d_seq
);
1370 * Update the inode too. We don't need to re-check the
1371 * dentry sequence number here after this d_inode read,
1372 * because a mount-point is always pinned.
1374 *inode
= path
->dentry
->d_inode
;
1376 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1377 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1380 static int follow_dotdot_rcu(struct nameidata
*nd
)
1382 struct inode
*inode
= nd
->inode
;
1385 if (path_equal(&nd
->path
, &nd
->root
))
1387 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1388 struct dentry
*old
= nd
->path
.dentry
;
1389 struct dentry
*parent
= old
->d_parent
;
1392 inode
= parent
->d_inode
;
1393 seq
= read_seqcount_begin(&parent
->d_seq
);
1394 if (unlikely(read_seqcount_retry(&old
->d_seq
, nd
->seq
)))
1396 nd
->path
.dentry
= parent
;
1398 if (unlikely(!path_connected(&nd
->path
)))
1402 struct mount
*mnt
= real_mount(nd
->path
.mnt
);
1403 struct mount
*mparent
= mnt
->mnt_parent
;
1404 struct dentry
*mountpoint
= mnt
->mnt_mountpoint
;
1405 struct inode
*inode2
= mountpoint
->d_inode
;
1406 unsigned seq
= read_seqcount_begin(&mountpoint
->d_seq
);
1407 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1409 if (&mparent
->mnt
== nd
->path
.mnt
)
1411 /* we know that mountpoint was pinned */
1412 nd
->path
.dentry
= mountpoint
;
1413 nd
->path
.mnt
= &mparent
->mnt
;
1418 while (unlikely(d_mountpoint(nd
->path
.dentry
))) {
1419 struct mount
*mounted
;
1420 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1421 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1425 nd
->path
.mnt
= &mounted
->mnt
;
1426 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1427 inode
= nd
->path
.dentry
->d_inode
;
1428 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1435 * Follow down to the covering mount currently visible to userspace. At each
1436 * point, the filesystem owning that dentry may be queried as to whether the
1437 * caller is permitted to proceed or not.
1439 int follow_down(struct path
*path
)
1444 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1445 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1446 /* Allow the filesystem to manage the transit without i_mutex
1449 * We indicate to the filesystem if someone is trying to mount
1450 * something here. This gives autofs the chance to deny anyone
1451 * other than its daemon the right to mount on its
1454 * The filesystem may sleep at this point.
1456 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1457 BUG_ON(!path
->dentry
->d_op
);
1458 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1459 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1461 return ret
== -EISDIR
? 0 : ret
;
1464 /* Transit to a mounted filesystem. */
1465 if (managed
& DCACHE_MOUNTED
) {
1466 struct vfsmount
*mounted
= lookup_mnt(path
);
1471 path
->mnt
= mounted
;
1472 path
->dentry
= dget(mounted
->mnt_root
);
1476 /* Don't handle automount points here */
1481 EXPORT_SYMBOL(follow_down
);
1484 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1486 static void follow_mount(struct path
*path
)
1488 while (d_mountpoint(path
->dentry
)) {
1489 struct vfsmount
*mounted
= lookup_mnt(path
);
1494 path
->mnt
= mounted
;
1495 path
->dentry
= dget(mounted
->mnt_root
);
1499 static int path_parent_directory(struct path
*path
)
1501 struct dentry
*old
= path
->dentry
;
1502 /* rare case of legitimate dget_parent()... */
1503 path
->dentry
= dget_parent(path
->dentry
);
1505 if (unlikely(!path_connected(path
)))
1510 static int follow_dotdot(struct nameidata
*nd
)
1513 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1514 nd
->path
.mnt
== nd
->root
.mnt
) {
1517 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1518 int ret
= path_parent_directory(&nd
->path
);
1523 if (!follow_up(&nd
->path
))
1526 follow_mount(&nd
->path
);
1527 nd
->inode
= nd
->path
.dentry
->d_inode
;
1532 * This looks up the name in dcache and possibly revalidates the found dentry.
1533 * NULL is returned if the dentry does not exist in the cache.
1535 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1539 struct dentry
*dentry
= d_lookup(dir
, name
);
1541 int error
= d_revalidate(dentry
, flags
);
1542 if (unlikely(error
<= 0)) {
1544 d_invalidate(dentry
);
1546 return ERR_PTR(error
);
1553 * Call i_op->lookup on the dentry. The dentry must be negative and
1556 * dir->d_inode->i_mutex must be held
1558 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1563 /* Don't create child dentry for a dead directory. */
1564 if (unlikely(IS_DEADDIR(dir
))) {
1566 return ERR_PTR(-ENOENT
);
1569 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1570 if (unlikely(old
)) {
1577 static struct dentry
*__lookup_hash(const struct qstr
*name
,
1578 struct dentry
*base
, unsigned int flags
)
1580 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1585 dentry
= d_alloc(base
, name
);
1586 if (unlikely(!dentry
))
1587 return ERR_PTR(-ENOMEM
);
1589 return lookup_real(base
->d_inode
, dentry
, flags
);
1592 static int lookup_fast(struct nameidata
*nd
,
1593 struct path
*path
, struct inode
**inode
,
1596 struct vfsmount
*mnt
= nd
->path
.mnt
;
1597 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1602 * Rename seqlock is not required here because in the off chance
1603 * of a false negative due to a concurrent rename, the caller is
1604 * going to fall back to non-racy lookup.
1606 if (nd
->flags
& LOOKUP_RCU
) {
1609 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1610 if (unlikely(!dentry
)) {
1611 if (unlazy_walk(nd
))
1617 * This sequence count validates that the inode matches
1618 * the dentry name information from lookup.
1620 *inode
= d_backing_inode(dentry
);
1621 negative
= d_is_negative(dentry
);
1622 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
1626 * This sequence count validates that the parent had no
1627 * changes while we did the lookup of the dentry above.
1629 * The memory barrier in read_seqcount_begin of child is
1630 * enough, we can use __read_seqcount_retry here.
1632 if (unlikely(__read_seqcount_retry(&parent
->d_seq
, nd
->seq
)))
1636 status
= d_revalidate(dentry
, nd
->flags
);
1637 if (likely(status
> 0)) {
1639 * Note: do negative dentry check after revalidation in
1640 * case that drops it.
1642 if (unlikely(negative
))
1645 path
->dentry
= dentry
;
1646 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1649 if (unlazy_child(nd
, dentry
, seq
))
1651 if (unlikely(status
== -ECHILD
))
1652 /* we'd been told to redo it in non-rcu mode */
1653 status
= d_revalidate(dentry
, nd
->flags
);
1655 dentry
= __d_lookup(parent
, &nd
->last
);
1656 if (unlikely(!dentry
))
1658 status
= d_revalidate(dentry
, nd
->flags
);
1660 if (unlikely(status
<= 0)) {
1662 d_invalidate(dentry
);
1666 if (unlikely(d_is_negative(dentry
))) {
1672 path
->dentry
= dentry
;
1673 err
= follow_managed(path
, nd
);
1674 if (likely(err
> 0))
1675 *inode
= d_backing_inode(path
->dentry
);
1679 /* Fast lookup failed, do it the slow way */
1680 static struct dentry
*lookup_slow(const struct qstr
*name
,
1684 struct dentry
*dentry
= ERR_PTR(-ENOENT
), *old
;
1685 struct inode
*inode
= dir
->d_inode
;
1686 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1688 inode_lock_shared(inode
);
1689 /* Don't go there if it's already dead */
1690 if (unlikely(IS_DEADDIR(inode
)))
1693 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1696 if (unlikely(!d_in_lookup(dentry
))) {
1697 if (!(flags
& LOOKUP_NO_REVAL
)) {
1698 int error
= d_revalidate(dentry
, flags
);
1699 if (unlikely(error
<= 0)) {
1701 d_invalidate(dentry
);
1706 dentry
= ERR_PTR(error
);
1710 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1711 d_lookup_done(dentry
);
1712 if (unlikely(old
)) {
1718 inode_unlock_shared(inode
);
1722 static inline int may_lookup(struct nameidata
*nd
)
1724 if (nd
->flags
& LOOKUP_RCU
) {
1725 int err
= inode_permission2(nd
->path
.mnt
, nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1728 if (unlazy_walk(nd
))
1731 return inode_permission2(nd
->path
.mnt
, nd
->inode
, MAY_EXEC
);
1734 static inline int handle_dots(struct nameidata
*nd
, int type
)
1736 if (type
== LAST_DOTDOT
) {
1739 if (nd
->flags
& LOOKUP_RCU
) {
1740 return follow_dotdot_rcu(nd
);
1742 return follow_dotdot(nd
);
1747 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1748 struct inode
*inode
, unsigned seq
)
1752 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1753 path_to_nameidata(link
, nd
);
1756 if (!(nd
->flags
& LOOKUP_RCU
)) {
1757 if (link
->mnt
== nd
->path
.mnt
)
1760 error
= nd_alloc_stack(nd
);
1761 if (unlikely(error
)) {
1762 if (error
== -ECHILD
) {
1763 if (unlikely(!legitimize_path(nd
, link
, seq
))) {
1766 nd
->flags
&= ~LOOKUP_RCU
;
1767 nd
->path
.mnt
= NULL
;
1768 nd
->path
.dentry
= NULL
;
1769 if (!(nd
->flags
& LOOKUP_ROOT
))
1770 nd
->root
.mnt
= NULL
;
1772 } else if (likely(unlazy_walk(nd
)) == 0)
1773 error
= nd_alloc_stack(nd
);
1781 last
= nd
->stack
+ nd
->depth
++;
1783 clear_delayed_call(&last
->done
);
1784 nd
->link_inode
= inode
;
1789 enum {WALK_FOLLOW
= 1, WALK_MORE
= 2};
1792 * Do we need to follow links? We _really_ want to be able
1793 * to do this check without having to look at inode->i_op,
1794 * so we keep a cache of "no, this doesn't need follow_link"
1795 * for the common case.
1797 static inline int step_into(struct nameidata
*nd
, struct path
*path
,
1798 int flags
, struct inode
*inode
, unsigned seq
)
1800 if (!(flags
& WALK_MORE
) && nd
->depth
)
1802 if (likely(!d_is_symlink(path
->dentry
)) ||
1803 !(flags
& WALK_FOLLOW
|| nd
->flags
& LOOKUP_FOLLOW
)) {
1804 /* not a symlink or should not follow */
1805 path_to_nameidata(path
, nd
);
1810 /* make sure that d_is_symlink above matches inode */
1811 if (nd
->flags
& LOOKUP_RCU
) {
1812 if (read_seqcount_retry(&path
->dentry
->d_seq
, seq
))
1815 return pick_link(nd
, path
, inode
, seq
);
1818 static int walk_component(struct nameidata
*nd
, int flags
)
1821 struct inode
*inode
;
1825 * "." and ".." are special - ".." especially so because it has
1826 * to be able to know about the current root directory and
1827 * parent relationships.
1829 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1830 err
= handle_dots(nd
, nd
->last_type
);
1831 if (!(flags
& WALK_MORE
) && nd
->depth
)
1835 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1836 if (unlikely(err
<= 0)) {
1839 path
.dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
,
1841 if (IS_ERR(path
.dentry
))
1842 return PTR_ERR(path
.dentry
);
1844 path
.mnt
= nd
->path
.mnt
;
1845 err
= follow_managed(&path
, nd
);
1846 if (unlikely(err
< 0))
1849 if (unlikely(d_is_negative(path
.dentry
))) {
1850 path_to_nameidata(&path
, nd
);
1854 seq
= 0; /* we are already out of RCU mode */
1855 inode
= d_backing_inode(path
.dentry
);
1858 return step_into(nd
, &path
, flags
, inode
, seq
);
1862 * We can do the critical dentry name comparison and hashing
1863 * operations one word at a time, but we are limited to:
1865 * - Architectures with fast unaligned word accesses. We could
1866 * do a "get_unaligned()" if this helps and is sufficiently
1869 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1870 * do not trap on the (extremely unlikely) case of a page
1871 * crossing operation.
1873 * - Furthermore, we need an efficient 64-bit compile for the
1874 * 64-bit case in order to generate the "number of bytes in
1875 * the final mask". Again, that could be replaced with a
1876 * efficient population count instruction or similar.
1878 #ifdef CONFIG_DCACHE_WORD_ACCESS
1880 #include <asm/word-at-a-time.h>
1884 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1886 #elif defined(CONFIG_64BIT)
1888 * Register pressure in the mixing function is an issue, particularly
1889 * on 32-bit x86, but almost any function requires one state value and
1890 * one temporary. Instead, use a function designed for two state values
1891 * and no temporaries.
1893 * This function cannot create a collision in only two iterations, so
1894 * we have two iterations to achieve avalanche. In those two iterations,
1895 * we have six layers of mixing, which is enough to spread one bit's
1896 * influence out to 2^6 = 64 state bits.
1898 * Rotate constants are scored by considering either 64 one-bit input
1899 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1900 * probability of that delta causing a change to each of the 128 output
1901 * bits, using a sample of random initial states.
1903 * The Shannon entropy of the computed probabilities is then summed
1904 * to produce a score. Ideally, any input change has a 50% chance of
1905 * toggling any given output bit.
1907 * Mixing scores (in bits) for (12,45):
1908 * Input delta: 1-bit 2-bit
1909 * 1 round: 713.3 42542.6
1910 * 2 rounds: 2753.7 140389.8
1911 * 3 rounds: 5954.1 233458.2
1912 * 4 rounds: 7862.6 256672.2
1913 * Perfect: 8192 258048
1914 * (64*128) (64*63/2 * 128)
1916 #define HASH_MIX(x, y, a) \
1918 y ^= x, x = rol64(x,12),\
1919 x += y, y = rol64(y,45),\
1923 * Fold two longs into one 32-bit hash value. This must be fast, but
1924 * latency isn't quite as critical, as there is a fair bit of additional
1925 * work done before the hash value is used.
1927 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1929 y
^= x
* GOLDEN_RATIO_64
;
1930 y
*= GOLDEN_RATIO_64
;
1934 #else /* 32-bit case */
1937 * Mixing scores (in bits) for (7,20):
1938 * Input delta: 1-bit 2-bit
1939 * 1 round: 330.3 9201.6
1940 * 2 rounds: 1246.4 25475.4
1941 * 3 rounds: 1907.1 31295.1
1942 * 4 rounds: 2042.3 31718.6
1943 * Perfect: 2048 31744
1944 * (32*64) (32*31/2 * 64)
1946 #define HASH_MIX(x, y, a) \
1948 y ^= x, x = rol32(x, 7),\
1949 x += y, y = rol32(y,20),\
1952 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1954 /* Use arch-optimized multiply if one exists */
1955 return __hash_32(y
^ __hash_32(x
));
1961 * Return the hash of a string of known length. This is carfully
1962 * designed to match hash_name(), which is the more critical function.
1963 * In particular, we must end by hashing a final word containing 0..7
1964 * payload bytes, to match the way that hash_name() iterates until it
1965 * finds the delimiter after the name.
1967 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
1969 unsigned long a
, x
= 0, y
= (unsigned long)salt
;
1974 a
= load_unaligned_zeropad(name
);
1975 if (len
< sizeof(unsigned long))
1978 name
+= sizeof(unsigned long);
1979 len
-= sizeof(unsigned long);
1981 x
^= a
& bytemask_from_count(len
);
1983 return fold_hash(x
, y
);
1985 EXPORT_SYMBOL(full_name_hash
);
1987 /* Return the "hash_len" (hash and length) of a null-terminated string */
1988 u64
hashlen_string(const void *salt
, const char *name
)
1990 unsigned long a
= 0, x
= 0, y
= (unsigned long)salt
;
1991 unsigned long adata
, mask
, len
;
1992 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1999 len
+= sizeof(unsigned long);
2001 a
= load_unaligned_zeropad(name
+len
);
2002 } while (!has_zero(a
, &adata
, &constants
));
2004 adata
= prep_zero_mask(a
, adata
, &constants
);
2005 mask
= create_zero_mask(adata
);
2006 x
^= a
& zero_bytemask(mask
);
2008 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2010 EXPORT_SYMBOL(hashlen_string
);
2013 * Calculate the length and hash of the path component, and
2014 * return the "hash_len" as the result.
2016 static inline u64
hash_name(const void *salt
, const char *name
)
2018 unsigned long a
= 0, b
, x
= 0, y
= (unsigned long)salt
;
2019 unsigned long adata
, bdata
, mask
, len
;
2020 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
2027 len
+= sizeof(unsigned long);
2029 a
= load_unaligned_zeropad(name
+len
);
2030 b
= a
^ REPEAT_BYTE('/');
2031 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
2033 adata
= prep_zero_mask(a
, adata
, &constants
);
2034 bdata
= prep_zero_mask(b
, bdata
, &constants
);
2035 mask
= create_zero_mask(adata
| bdata
);
2036 x
^= a
& zero_bytemask(mask
);
2038 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2041 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2043 /* Return the hash of a string of known length */
2044 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
2046 unsigned long hash
= init_name_hash(salt
);
2048 hash
= partial_name_hash((unsigned char)*name
++, hash
);
2049 return end_name_hash(hash
);
2051 EXPORT_SYMBOL(full_name_hash
);
2053 /* Return the "hash_len" (hash and length) of a null-terminated string */
2054 u64
hashlen_string(const void *salt
, const char *name
)
2056 unsigned long hash
= init_name_hash(salt
);
2057 unsigned long len
= 0, c
;
2059 c
= (unsigned char)*name
;
2062 hash
= partial_name_hash(c
, hash
);
2063 c
= (unsigned char)name
[len
];
2065 return hashlen_create(end_name_hash(hash
), len
);
2067 EXPORT_SYMBOL(hashlen_string
);
2070 * We know there's a real path component here of at least
2073 static inline u64
hash_name(const void *salt
, const char *name
)
2075 unsigned long hash
= init_name_hash(salt
);
2076 unsigned long len
= 0, c
;
2078 c
= (unsigned char)*name
;
2081 hash
= partial_name_hash(c
, hash
);
2082 c
= (unsigned char)name
[len
];
2083 } while (c
&& c
!= '/');
2084 return hashlen_create(end_name_hash(hash
), len
);
2091 * This is the basic name resolution function, turning a pathname into
2092 * the final dentry. We expect 'base' to be positive and a directory.
2094 * Returns 0 and nd will have valid dentry and mnt on success.
2095 * Returns error and drops reference to input namei data on failure.
2097 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2106 /* At this point we know we have a real path component. */
2111 err
= may_lookup(nd
);
2115 hash_len
= hash_name(nd
->path
.dentry
, name
);
2118 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2120 if (name
[1] == '.') {
2122 nd
->flags
|= LOOKUP_JUMPED
;
2128 if (likely(type
== LAST_NORM
)) {
2129 struct dentry
*parent
= nd
->path
.dentry
;
2130 nd
->flags
&= ~LOOKUP_JUMPED
;
2131 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2132 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2133 err
= parent
->d_op
->d_hash(parent
, &this);
2136 hash_len
= this.hash_len
;
2141 nd
->last
.hash_len
= hash_len
;
2142 nd
->last
.name
= name
;
2143 nd
->last_type
= type
;
2145 name
+= hashlen_len(hash_len
);
2149 * If it wasn't NUL, we know it was '/'. Skip that
2150 * slash, and continue until no more slashes.
2154 } while (unlikely(*name
== '/'));
2155 if (unlikely(!*name
)) {
2157 /* pathname body, done */
2160 name
= nd
->stack
[nd
->depth
- 1].name
;
2161 /* trailing symlink, done */
2164 /* last component of nested symlink */
2165 err
= walk_component(nd
, WALK_FOLLOW
);
2167 /* not the last component */
2168 err
= walk_component(nd
, WALK_FOLLOW
| WALK_MORE
);
2174 const char *s
= get_link(nd
);
2183 nd
->stack
[nd
->depth
- 1].name
= name
;
2188 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2189 if (nd
->flags
& LOOKUP_RCU
) {
2190 if (unlazy_walk(nd
))
2198 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2200 const char *s
= nd
->name
->name
;
2203 flags
&= ~LOOKUP_RCU
;
2205 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
2206 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
2208 if (flags
& LOOKUP_ROOT
) {
2209 struct dentry
*root
= nd
->root
.dentry
;
2210 struct inode
*inode
= root
->d_inode
;
2211 if (*s
&& unlikely(!d_can_lookup(root
)))
2212 return ERR_PTR(-ENOTDIR
);
2213 nd
->path
= nd
->root
;
2215 if (flags
& LOOKUP_RCU
) {
2217 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2218 nd
->root_seq
= nd
->seq
;
2219 nd
->m_seq
= read_seqbegin(&mount_lock
);
2221 path_get(&nd
->path
);
2226 nd
->root
.mnt
= NULL
;
2227 nd
->path
.mnt
= NULL
;
2228 nd
->path
.dentry
= NULL
;
2230 nd
->m_seq
= read_seqbegin(&mount_lock
);
2232 if (flags
& LOOKUP_RCU
)
2235 if (likely(!nd_jump_root(nd
)))
2237 nd
->root
.mnt
= NULL
;
2239 return ERR_PTR(-ECHILD
);
2240 } else if (nd
->dfd
== AT_FDCWD
) {
2241 if (flags
& LOOKUP_RCU
) {
2242 struct fs_struct
*fs
= current
->fs
;
2248 seq
= read_seqcount_begin(&fs
->seq
);
2250 nd
->inode
= nd
->path
.dentry
->d_inode
;
2251 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2252 } while (read_seqcount_retry(&fs
->seq
, seq
));
2254 get_fs_pwd(current
->fs
, &nd
->path
);
2255 nd
->inode
= nd
->path
.dentry
->d_inode
;
2259 /* Caller must check execute permissions on the starting path component */
2260 struct fd f
= fdget_raw(nd
->dfd
);
2261 struct dentry
*dentry
;
2264 return ERR_PTR(-EBADF
);
2266 dentry
= f
.file
->f_path
.dentry
;
2269 if (!d_can_lookup(dentry
)) {
2271 return ERR_PTR(-ENOTDIR
);
2275 nd
->path
= f
.file
->f_path
;
2276 if (flags
& LOOKUP_RCU
) {
2278 nd
->inode
= nd
->path
.dentry
->d_inode
;
2279 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2281 path_get(&nd
->path
);
2282 nd
->inode
= nd
->path
.dentry
->d_inode
;
2289 static const char *trailing_symlink(struct nameidata
*nd
)
2292 int error
= may_follow_link(nd
);
2293 if (unlikely(error
))
2294 return ERR_PTR(error
);
2295 nd
->flags
|= LOOKUP_PARENT
;
2296 nd
->stack
[0].name
= NULL
;
2301 static inline int lookup_last(struct nameidata
*nd
)
2303 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2304 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2306 nd
->flags
&= ~LOOKUP_PARENT
;
2307 return walk_component(nd
, 0);
2310 static int handle_lookup_down(struct nameidata
*nd
)
2312 struct path path
= nd
->path
;
2313 struct inode
*inode
= nd
->inode
;
2314 unsigned seq
= nd
->seq
;
2317 if (nd
->flags
& LOOKUP_RCU
) {
2319 * don't bother with unlazy_walk on failure - we are
2320 * at the very beginning of walk, so we lose nothing
2321 * if we simply redo everything in non-RCU mode
2323 if (unlikely(!__follow_mount_rcu(nd
, &path
, &inode
, &seq
)))
2327 err
= follow_managed(&path
, nd
);
2328 if (unlikely(err
< 0))
2330 inode
= d_backing_inode(path
.dentry
);
2333 path_to_nameidata(&path
, nd
);
2339 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2340 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2342 const char *s
= path_init(nd
, flags
);
2348 if (unlikely(flags
& LOOKUP_DOWN
)) {
2349 err
= handle_lookup_down(nd
);
2350 if (unlikely(err
< 0)) {
2356 while (!(err
= link_path_walk(s
, nd
))
2357 && ((err
= lookup_last(nd
)) > 0)) {
2358 s
= trailing_symlink(nd
);
2365 err
= complete_walk(nd
);
2367 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2368 if (!d_can_lookup(nd
->path
.dentry
))
2372 nd
->path
.mnt
= NULL
;
2373 nd
->path
.dentry
= NULL
;
2379 static int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2380 struct path
*path
, struct path
*root
)
2383 struct nameidata nd
;
2385 return PTR_ERR(name
);
2386 if (unlikely(root
)) {
2388 flags
|= LOOKUP_ROOT
;
2390 set_nameidata(&nd
, dfd
, name
);
2391 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2392 if (unlikely(retval
== -ECHILD
))
2393 retval
= path_lookupat(&nd
, flags
, path
);
2394 if (unlikely(retval
== -ESTALE
))
2395 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2397 if (likely(!retval
))
2398 audit_inode(name
, path
->dentry
, flags
& LOOKUP_PARENT
);
2399 restore_nameidata();
2404 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2405 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2406 struct path
*parent
)
2408 const char *s
= path_init(nd
, flags
);
2412 err
= link_path_walk(s
, nd
);
2414 err
= complete_walk(nd
);
2417 nd
->path
.mnt
= NULL
;
2418 nd
->path
.dentry
= NULL
;
2424 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2425 unsigned int flags
, struct path
*parent
,
2426 struct qstr
*last
, int *type
)
2429 struct nameidata nd
;
2433 set_nameidata(&nd
, dfd
, name
);
2434 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2435 if (unlikely(retval
== -ECHILD
))
2436 retval
= path_parentat(&nd
, flags
, parent
);
2437 if (unlikely(retval
== -ESTALE
))
2438 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2439 if (likely(!retval
)) {
2441 *type
= nd
.last_type
;
2442 audit_inode(name
, parent
->dentry
, LOOKUP_PARENT
);
2445 name
= ERR_PTR(retval
);
2447 restore_nameidata();
2451 /* does lookup, returns the object with parent locked */
2452 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2454 struct filename
*filename
;
2459 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2461 if (IS_ERR(filename
))
2462 return ERR_CAST(filename
);
2463 if (unlikely(type
!= LAST_NORM
)) {
2466 return ERR_PTR(-EINVAL
);
2468 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2469 d
= __lookup_hash(&last
, path
->dentry
, 0);
2471 inode_unlock(path
->dentry
->d_inode
);
2478 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2480 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2483 EXPORT_SYMBOL(kern_path
);
2486 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2487 * @dentry: pointer to dentry of the base directory
2488 * @mnt: pointer to vfs mount of the base directory
2489 * @name: pointer to file name
2490 * @flags: lookup flags
2491 * @path: pointer to struct path to fill
2493 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2494 const char *name
, unsigned int flags
,
2497 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2498 /* the first argument of filename_lookup() is ignored with root */
2499 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2500 flags
, path
, &root
);
2502 EXPORT_SYMBOL(vfs_path_lookup
);
2505 * lookup_one_len - filesystem helper to lookup single pathname component
2506 * @name: pathname component to lookup
2507 * @mnt: mount we are looking up on
2508 * @base: base directory to lookup from
2509 * @len: maximum length @len should be interpreted to
2511 * Note that this routine is purely a helper for filesystem usage and should
2512 * not be called by generic code.
2514 * The caller must hold base->i_mutex.
2516 struct dentry
*lookup_one_len2(const char *name
, struct vfsmount
*mnt
, struct dentry
*base
, int len
)
2522 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2526 this.hash
= full_name_hash(base
, name
, len
);
2528 return ERR_PTR(-EACCES
);
2530 if (unlikely(name
[0] == '.')) {
2531 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2532 return ERR_PTR(-EACCES
);
2536 c
= *(const unsigned char *)name
++;
2537 if (c
== '/' || c
== '\0')
2538 return ERR_PTR(-EACCES
);
2541 * See if the low-level filesystem might want
2542 * to use its own hash..
2544 if (base
->d_flags
& DCACHE_OP_HASH
) {
2545 int err
= base
->d_op
->d_hash(base
, &this);
2547 return ERR_PTR(err
);
2550 err
= inode_permission2(mnt
, base
->d_inode
, MAY_EXEC
);
2552 return ERR_PTR(err
);
2554 return __lookup_hash(&this, base
, 0);
2556 EXPORT_SYMBOL(lookup_one_len2
);
2558 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2560 return lookup_one_len2(name
, NULL
, base
, len
);
2562 EXPORT_SYMBOL(lookup_one_len
);
2565 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2566 * @name: pathname component to lookup
2567 * @base: base directory to lookup from
2568 * @len: maximum length @len should be interpreted to
2570 * Note that this routine is purely a helper for filesystem usage and should
2571 * not be called by generic code.
2573 * Unlike lookup_one_len, it should be called without the parent
2574 * i_mutex held, and will take the i_mutex itself if necessary.
2576 struct dentry
*lookup_one_len_unlocked(const char *name
,
2577 struct dentry
*base
, int len
)
2586 this.hash
= full_name_hash(base
, name
, len
);
2588 return ERR_PTR(-EACCES
);
2590 if (unlikely(name
[0] == '.')) {
2591 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2592 return ERR_PTR(-EACCES
);
2596 c
= *(const unsigned char *)name
++;
2597 if (c
== '/' || c
== '\0')
2598 return ERR_PTR(-EACCES
);
2601 * See if the low-level filesystem might want
2602 * to use its own hash..
2604 if (base
->d_flags
& DCACHE_OP_HASH
) {
2605 int err
= base
->d_op
->d_hash(base
, &this);
2607 return ERR_PTR(err
);
2610 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2612 return ERR_PTR(err
);
2614 ret
= lookup_dcache(&this, base
, 0);
2616 ret
= lookup_slow(&this, base
, 0);
2619 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2621 #ifdef CONFIG_UNIX98_PTYS
2622 int path_pts(struct path
*path
)
2624 /* Find something mounted on "pts" in the same directory as
2627 struct dentry
*child
, *parent
;
2631 ret
= path_parent_directory(path
);
2635 parent
= path
->dentry
;
2638 child
= d_hash_and_lookup(parent
, &this);
2642 path
->dentry
= child
;
2649 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2650 struct path
*path
, int *empty
)
2652 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2655 EXPORT_SYMBOL(user_path_at_empty
);
2658 * mountpoint_last - look up last component for umount
2659 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2661 * This is a special lookup_last function just for umount. In this case, we
2662 * need to resolve the path without doing any revalidation.
2664 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2665 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2666 * in almost all cases, this lookup will be served out of the dcache. The only
2667 * cases where it won't are if nd->last refers to a symlink or the path is
2668 * bogus and it doesn't exist.
2671 * -error: if there was an error during lookup. This includes -ENOENT if the
2672 * lookup found a negative dentry.
2674 * 0: if we successfully resolved nd->last and found it to not to be a
2675 * symlink that needs to be followed.
2677 * 1: if we successfully resolved nd->last and found it to be a symlink
2678 * that needs to be followed.
2681 mountpoint_last(struct nameidata
*nd
)
2684 struct dentry
*dir
= nd
->path
.dentry
;
2687 /* If we're in rcuwalk, drop out of it to handle last component */
2688 if (nd
->flags
& LOOKUP_RCU
) {
2689 if (unlazy_walk(nd
))
2693 nd
->flags
&= ~LOOKUP_PARENT
;
2695 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2696 error
= handle_dots(nd
, nd
->last_type
);
2699 path
.dentry
= dget(nd
->path
.dentry
);
2701 path
.dentry
= d_lookup(dir
, &nd
->last
);
2704 * No cached dentry. Mounted dentries are pinned in the
2705 * cache, so that means that this dentry is probably
2706 * a symlink or the path doesn't actually point
2707 * to a mounted dentry.
2709 path
.dentry
= lookup_slow(&nd
->last
, dir
,
2710 nd
->flags
| LOOKUP_NO_REVAL
);
2711 if (IS_ERR(path
.dentry
))
2712 return PTR_ERR(path
.dentry
);
2715 if (d_is_negative(path
.dentry
)) {
2719 path
.mnt
= nd
->path
.mnt
;
2720 return step_into(nd
, &path
, 0, d_backing_inode(path
.dentry
), 0);
2724 * path_mountpoint - look up a path to be umounted
2725 * @nd: lookup context
2726 * @flags: lookup flags
2727 * @path: pointer to container for result
2729 * Look up the given name, but don't attempt to revalidate the last component.
2730 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2733 path_mountpoint(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2735 const char *s
= path_init(nd
, flags
);
2739 while (!(err
= link_path_walk(s
, nd
)) &&
2740 (err
= mountpoint_last(nd
)) > 0) {
2741 s
= trailing_symlink(nd
);
2749 nd
->path
.mnt
= NULL
;
2750 nd
->path
.dentry
= NULL
;
2758 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2761 struct nameidata nd
;
2764 return PTR_ERR(name
);
2765 set_nameidata(&nd
, dfd
, name
);
2766 error
= path_mountpoint(&nd
, flags
| LOOKUP_RCU
, path
);
2767 if (unlikely(error
== -ECHILD
))
2768 error
= path_mountpoint(&nd
, flags
, path
);
2769 if (unlikely(error
== -ESTALE
))
2770 error
= path_mountpoint(&nd
, flags
| LOOKUP_REVAL
, path
);
2772 audit_inode(name
, path
->dentry
, 0);
2773 restore_nameidata();
2779 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2780 * @dfd: directory file descriptor
2781 * @name: pathname from userland
2782 * @flags: lookup flags
2783 * @path: pointer to container to hold result
2785 * A umount is a special case for path walking. We're not actually interested
2786 * in the inode in this situation, and ESTALE errors can be a problem. We
2787 * simply want track down the dentry and vfsmount attached at the mountpoint
2788 * and avoid revalidating the last component.
2790 * Returns 0 and populates "path" on success.
2793 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2796 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2800 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2803 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2805 EXPORT_SYMBOL(kern_path_mountpoint
);
2807 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2809 kuid_t fsuid
= current_fsuid();
2811 if (uid_eq(inode
->i_uid
, fsuid
))
2813 if (uid_eq(dir
->i_uid
, fsuid
))
2815 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2817 EXPORT_SYMBOL(__check_sticky
);
2820 * Check whether we can remove a link victim from directory dir, check
2821 * whether the type of victim is right.
2822 * 1. We can't do it if dir is read-only (done in permission())
2823 * 2. We should have write and exec permissions on dir
2824 * 3. We can't remove anything from append-only dir
2825 * 4. We can't do anything with immutable dir (done in permission())
2826 * 5. If the sticky bit on dir is set we should either
2827 * a. be owner of dir, or
2828 * b. be owner of victim, or
2829 * c. have CAP_FOWNER capability
2830 * 6. If the victim is append-only or immutable we can't do antyhing with
2831 * links pointing to it.
2832 * 7. If the victim has an unknown uid or gid we can't change the inode.
2833 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2834 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2835 * 10. We can't remove a root or mountpoint.
2836 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2837 * nfs_async_unlink().
2839 static int may_delete(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2841 struct inode
*inode
= d_backing_inode(victim
);
2844 if (d_is_negative(victim
))
2848 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2849 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2851 error
= inode_permission2(mnt
, dir
, MAY_WRITE
| MAY_EXEC
);
2857 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2858 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
) || HAS_UNMAPPED_ID(inode
))
2861 if (!d_is_dir(victim
))
2863 if (IS_ROOT(victim
))
2865 } else if (d_is_dir(victim
))
2867 if (IS_DEADDIR(dir
))
2869 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2874 /* Check whether we can create an object with dentry child in directory
2876 * 1. We can't do it if child already exists (open has special treatment for
2877 * this case, but since we are inlined it's OK)
2878 * 2. We can't do it if dir is read-only (done in permission())
2879 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2880 * 4. We should have write and exec permissions on dir
2881 * 5. We can't do it if dir is immutable (done in permission())
2883 static inline int may_create(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*child
)
2885 struct user_namespace
*s_user_ns
;
2886 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2889 if (IS_DEADDIR(dir
))
2891 s_user_ns
= dir
->i_sb
->s_user_ns
;
2892 if (!kuid_has_mapping(s_user_ns
, current_fsuid()) ||
2893 !kgid_has_mapping(s_user_ns
, current_fsgid()))
2895 return inode_permission2(mnt
, dir
, MAY_WRITE
| MAY_EXEC
);
2899 * p1 and p2 should be directories on the same fs.
2901 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2906 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2910 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
2912 p
= d_ancestor(p2
, p1
);
2914 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2915 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2919 p
= d_ancestor(p1
, p2
);
2921 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2922 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2926 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2927 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2930 EXPORT_SYMBOL(lock_rename
);
2932 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2934 inode_unlock(p1
->d_inode
);
2936 inode_unlock(p2
->d_inode
);
2937 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
2940 EXPORT_SYMBOL(unlock_rename
);
2942 int vfs_create2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
,
2943 umode_t mode
, bool want_excl
)
2945 int error
= may_create(mnt
, dir
, dentry
);
2949 if (!dir
->i_op
->create
)
2950 return -EACCES
; /* shouldn't it be ENOSYS? */
2953 error
= security_inode_create(dir
, dentry
, mode
);
2956 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2958 fsnotify_create(dir
, dentry
);
2961 EXPORT_SYMBOL(vfs_create2
);
2963 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2966 return vfs_create2(NULL
, dir
, dentry
, mode
, want_excl
);
2968 EXPORT_SYMBOL(vfs_create
);
2970 bool may_open_dev(const struct path
*path
)
2972 return !(path
->mnt
->mnt_flags
& MNT_NODEV
) &&
2973 !(path
->mnt
->mnt_sb
->s_iflags
& SB_I_NODEV
);
2976 static int may_open(const struct path
*path
, int acc_mode
, int flag
)
2978 struct dentry
*dentry
= path
->dentry
;
2979 struct vfsmount
*mnt
= path
->mnt
;
2980 struct inode
*inode
= dentry
->d_inode
;
2986 switch (inode
->i_mode
& S_IFMT
) {
2990 if (acc_mode
& MAY_WRITE
)
2995 if (!may_open_dev(path
))
3004 error
= inode_permission2(mnt
, inode
, MAY_OPEN
| acc_mode
);
3009 * An append-only file must be opened in append mode for writing.
3011 if (IS_APPEND(inode
)) {
3012 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
3018 /* O_NOATIME can only be set by the owner or superuser */
3019 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
3025 static int handle_truncate(struct file
*filp
)
3027 const struct path
*path
= &filp
->f_path
;
3028 struct inode
*inode
= path
->dentry
->d_inode
;
3029 int error
= get_write_access(inode
);
3033 * Refuse to truncate files with mandatory locks held on them.
3035 error
= locks_verify_locked(filp
);
3037 error
= security_path_truncate(path
);
3039 error
= do_truncate2(path
->mnt
, path
->dentry
, 0,
3040 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
3043 put_write_access(inode
);
3047 static inline int open_to_namei_flags(int flag
)
3049 if ((flag
& O_ACCMODE
) == 3)
3054 static int may_o_create(const struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
3056 struct user_namespace
*s_user_ns
;
3057 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
3061 s_user_ns
= dir
->dentry
->d_sb
->s_user_ns
;
3062 if (!kuid_has_mapping(s_user_ns
, current_fsuid()) ||
3063 !kgid_has_mapping(s_user_ns
, current_fsgid()))
3066 error
= inode_permission2(dir
->mnt
, dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
3070 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
3074 * Attempt to atomically look up, create and open a file from a negative
3077 * Returns 0 if successful. The file will have been created and attached to
3078 * @file by the filesystem calling finish_open().
3080 * Returns 1 if the file was looked up only or didn't need creating. The
3081 * caller will need to perform the open themselves. @path will have been
3082 * updated to point to the new dentry. This may be negative.
3084 * Returns an error code otherwise.
3086 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
3087 struct path
*path
, struct file
*file
,
3088 const struct open_flags
*op
,
3089 int open_flag
, umode_t mode
,
3092 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
3093 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
3096 if (!(~open_flag
& (O_EXCL
| O_CREAT
))) /* both O_EXCL and O_CREAT */
3097 open_flag
&= ~O_TRUNC
;
3099 if (nd
->flags
& LOOKUP_DIRECTORY
)
3100 open_flag
|= O_DIRECTORY
;
3102 file
->f_path
.dentry
= DENTRY_NOT_SET
;
3103 file
->f_path
.mnt
= nd
->path
.mnt
;
3104 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
3105 open_to_namei_flags(open_flag
),
3107 d_lookup_done(dentry
);
3110 * We didn't have the inode before the open, so check open
3113 int acc_mode
= op
->acc_mode
;
3114 if (*opened
& FILE_CREATED
) {
3115 WARN_ON(!(open_flag
& O_CREAT
));
3116 fsnotify_create(dir
, dentry
);
3119 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
3120 if (WARN_ON(error
> 0))
3122 } else if (error
> 0) {
3123 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
3126 if (file
->f_path
.dentry
) {
3128 dentry
= file
->f_path
.dentry
;
3130 if (*opened
& FILE_CREATED
)
3131 fsnotify_create(dir
, dentry
);
3132 if (unlikely(d_is_negative(dentry
))) {
3135 path
->dentry
= dentry
;
3136 path
->mnt
= nd
->path
.mnt
;
3146 * Look up and maybe create and open the last component.
3148 * Must be called with i_mutex held on parent.
3150 * Returns 0 if the file was successfully atomically created (if necessary) and
3151 * opened. In this case the file will be returned attached to @file.
3153 * Returns 1 if the file was not completely opened at this time, though lookups
3154 * and creations will have been performed and the dentry returned in @path will
3155 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3156 * specified then a negative dentry may be returned.
3158 * An error code is returned otherwise.
3160 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3161 * cleared otherwise prior to returning.
3163 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
3165 const struct open_flags
*op
,
3166 bool got_write
, int *opened
)
3168 struct dentry
*dir
= nd
->path
.dentry
;
3169 struct inode
*dir_inode
= dir
->d_inode
;
3170 int open_flag
= op
->open_flag
;
3171 struct dentry
*dentry
;
3172 int error
, create_error
= 0;
3173 umode_t mode
= op
->mode
;
3174 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
3176 if (unlikely(IS_DEADDIR(dir_inode
)))
3179 *opened
&= ~FILE_CREATED
;
3180 dentry
= d_lookup(dir
, &nd
->last
);
3183 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
3185 return PTR_ERR(dentry
);
3187 if (d_in_lookup(dentry
))
3190 error
= d_revalidate(dentry
, nd
->flags
);
3191 if (likely(error
> 0))
3195 d_invalidate(dentry
);
3199 if (dentry
->d_inode
) {
3200 /* Cached positive dentry: will open in f_op->open */
3205 * Checking write permission is tricky, bacuse we don't know if we are
3206 * going to actually need it: O_CREAT opens should work as long as the
3207 * file exists. But checking existence breaks atomicity. The trick is
3208 * to check access and if not granted clear O_CREAT from the flags.
3210 * Another problem is returing the "right" error value (e.g. for an
3211 * O_EXCL open we want to return EEXIST not EROFS).
3213 if (open_flag
& O_CREAT
) {
3214 if (!IS_POSIXACL(dir
->d_inode
))
3215 mode
&= ~current_umask();
3216 if (unlikely(!got_write
)) {
3217 create_error
= -EROFS
;
3218 open_flag
&= ~O_CREAT
;
3219 if (open_flag
& (O_EXCL
| O_TRUNC
))
3221 /* No side effects, safe to clear O_CREAT */
3223 create_error
= may_o_create(&nd
->path
, dentry
, mode
);
3225 open_flag
&= ~O_CREAT
;
3226 if (open_flag
& O_EXCL
)
3230 } else if ((open_flag
& (O_TRUNC
|O_WRONLY
|O_RDWR
)) &&
3231 unlikely(!got_write
)) {
3233 * No O_CREATE -> atomicity not a requirement -> fall
3234 * back to lookup + open
3239 if (dir_inode
->i_op
->atomic_open
) {
3240 error
= atomic_open(nd
, dentry
, path
, file
, op
, open_flag
,
3242 if (unlikely(error
== -ENOENT
) && create_error
)
3243 error
= create_error
;
3248 if (d_in_lookup(dentry
)) {
3249 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3251 d_lookup_done(dentry
);
3252 if (unlikely(res
)) {
3254 error
= PTR_ERR(res
);
3262 /* Negative dentry, just create the file */
3263 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3264 *opened
|= FILE_CREATED
;
3265 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3266 if (!dir_inode
->i_op
->create
) {
3270 error
= dir_inode
->i_op
->create(dir_inode
, dentry
, mode
,
3271 open_flag
& O_EXCL
);
3274 fsnotify_create(dir_inode
, dentry
);
3276 if (unlikely(create_error
) && !dentry
->d_inode
) {
3277 error
= create_error
;
3281 path
->dentry
= dentry
;
3282 path
->mnt
= nd
->path
.mnt
;
3291 * Handle the last step of open()
3293 static int do_last(struct nameidata
*nd
,
3294 struct file
*file
, const struct open_flags
*op
,
3297 struct dentry
*dir
= nd
->path
.dentry
;
3298 int open_flag
= op
->open_flag
;
3299 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3300 bool got_write
= false;
3301 int acc_mode
= op
->acc_mode
;
3303 struct inode
*inode
;
3307 nd
->flags
&= ~LOOKUP_PARENT
;
3308 nd
->flags
|= op
->intent
;
3310 if (nd
->last_type
!= LAST_NORM
) {
3311 error
= handle_dots(nd
, nd
->last_type
);
3312 if (unlikely(error
))
3317 if (!(open_flag
& O_CREAT
)) {
3318 if (nd
->last
.name
[nd
->last
.len
])
3319 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3320 /* we _can_ be in RCU mode here */
3321 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3322 if (likely(error
> 0))
3328 BUG_ON(nd
->inode
!= dir
->d_inode
);
3329 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3331 /* create side of things */
3333 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3334 * has been cleared when we got to the last component we are
3337 error
= complete_walk(nd
);
3341 audit_inode(nd
->name
, dir
, LOOKUP_PARENT
);
3342 /* trailing slashes? */
3343 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3347 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3348 error
= mnt_want_write(nd
->path
.mnt
);
3352 * do _not_ fail yet - we might not need that or fail with
3353 * a different error; let lookup_open() decide; we'll be
3354 * dropping this one anyway.
3357 if (open_flag
& O_CREAT
)
3358 inode_lock(dir
->d_inode
);
3360 inode_lock_shared(dir
->d_inode
);
3361 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3362 if (open_flag
& O_CREAT
)
3363 inode_unlock(dir
->d_inode
);
3365 inode_unlock_shared(dir
->d_inode
);
3371 if ((*opened
& FILE_CREATED
) ||
3372 !S_ISREG(file_inode(file
)->i_mode
))
3373 will_truncate
= false;
3375 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3379 if (*opened
& FILE_CREATED
) {
3380 /* Don't check for write permission, don't truncate */
3381 open_flag
&= ~O_TRUNC
;
3382 will_truncate
= false;
3384 path_to_nameidata(&path
, nd
);
3385 goto finish_open_created
;
3389 * If atomic_open() acquired write access it is dropped now due to
3390 * possible mount and symlink following (this might be optimized away if
3394 mnt_drop_write(nd
->path
.mnt
);
3398 error
= follow_managed(&path
, nd
);
3399 if (unlikely(error
< 0))
3402 if (unlikely(d_is_negative(path
.dentry
))) {
3403 path_to_nameidata(&path
, nd
);
3408 * create/update audit record if it already exists.
3410 audit_inode(nd
->name
, path
.dentry
, 0);
3412 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3413 path_to_nameidata(&path
, nd
);
3417 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3418 inode
= d_backing_inode(path
.dentry
);
3420 error
= step_into(nd
, &path
, 0, inode
, seq
);
3421 if (unlikely(error
))
3424 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3425 error
= complete_walk(nd
);
3428 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3429 if (open_flag
& O_CREAT
) {
3431 if (d_is_dir(nd
->path
.dentry
))
3433 error
= may_create_in_sticky(dir
,
3434 d_backing_inode(nd
->path
.dentry
));
3435 if (unlikely(error
))
3439 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3441 if (!d_is_reg(nd
->path
.dentry
))
3442 will_truncate
= false;
3444 if (will_truncate
) {
3445 error
= mnt_want_write(nd
->path
.mnt
);
3450 finish_open_created
:
3451 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3454 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3455 error
= vfs_open(&nd
->path
, file
, current_cred());
3458 *opened
|= FILE_OPENED
;
3460 error
= open_check_o_direct(file
);
3462 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3463 if (!error
&& will_truncate
)
3464 error
= handle_truncate(file
);
3466 if (unlikely(error
) && (*opened
& FILE_OPENED
))
3468 if (unlikely(error
> 0)) {
3473 mnt_drop_write(nd
->path
.mnt
);
3477 struct dentry
*vfs_tmpfile(struct dentry
*dentry
, umode_t mode
, int open_flag
)
3479 struct dentry
*child
= NULL
;
3480 struct inode
*dir
= dentry
->d_inode
;
3481 struct inode
*inode
;
3484 /* we want directory to be writable */
3485 error
= inode_permission2(ERR_PTR(-EOPNOTSUPP
), dir
,
3486 MAY_WRITE
| MAY_EXEC
);
3489 error
= -EOPNOTSUPP
;
3490 if (!dir
->i_op
->tmpfile
)
3493 child
= d_alloc(dentry
, &slash_name
);
3494 if (unlikely(!child
))
3496 error
= dir
->i_op
->tmpfile(dir
, child
, mode
);
3500 inode
= child
->d_inode
;
3501 if (unlikely(!inode
))
3503 if (!(open_flag
& O_EXCL
)) {
3504 spin_lock(&inode
->i_lock
);
3505 inode
->i_state
|= I_LINKABLE
;
3506 spin_unlock(&inode
->i_lock
);
3512 return ERR_PTR(error
);
3514 EXPORT_SYMBOL(vfs_tmpfile
);
3516 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3517 const struct open_flags
*op
,
3518 struct file
*file
, int *opened
)
3520 struct dentry
*child
;
3522 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3523 if (unlikely(error
))
3525 error
= mnt_want_write(path
.mnt
);
3526 if (unlikely(error
))
3528 child
= vfs_tmpfile(path
.dentry
, op
->mode
, op
->open_flag
);
3529 error
= PTR_ERR(child
);
3530 if (unlikely(IS_ERR(child
)))
3533 path
.dentry
= child
;
3534 audit_inode(nd
->name
, child
, 0);
3535 /* Don't check for other permissions, the inode was just created */
3536 error
= may_open(&path
, 0, op
->open_flag
);
3539 file
->f_path
.mnt
= path
.mnt
;
3540 error
= finish_open(file
, child
, NULL
, opened
);
3543 error
= open_check_o_direct(file
);
3547 mnt_drop_write(path
.mnt
);
3553 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3556 int error
= path_lookupat(nd
, flags
, &path
);
3558 audit_inode(nd
->name
, path
.dentry
, 0);
3559 error
= vfs_open(&path
, file
, current_cred());
3565 static struct file
*path_openat(struct nameidata
*nd
,
3566 const struct open_flags
*op
, unsigned flags
)
3573 file
= get_empty_filp();
3577 file
->f_flags
= op
->open_flag
;
3579 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3580 error
= do_tmpfile(nd
, flags
, op
, file
, &opened
);
3584 if (unlikely(file
->f_flags
& O_PATH
)) {
3585 error
= do_o_path(nd
, flags
, file
);
3587 opened
|= FILE_OPENED
;
3591 s
= path_init(nd
, flags
);
3596 while (!(error
= link_path_walk(s
, nd
)) &&
3597 (error
= do_last(nd
, file
, op
, &opened
)) > 0) {
3598 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3599 s
= trailing_symlink(nd
);
3607 if (!(opened
& FILE_OPENED
)) {
3611 if (unlikely(error
)) {
3612 if (error
== -EOPENSTALE
) {
3613 if (flags
& LOOKUP_RCU
)
3618 file
= ERR_PTR(error
);
3623 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3624 const struct open_flags
*op
)
3626 struct nameidata nd
;
3627 int flags
= op
->lookup_flags
;
3630 set_nameidata(&nd
, dfd
, pathname
);
3631 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3632 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3633 filp
= path_openat(&nd
, op
, flags
);
3634 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3635 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3636 restore_nameidata();
3640 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3641 const char *name
, const struct open_flags
*op
)
3643 struct nameidata nd
;
3645 struct filename
*filename
;
3646 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3649 nd
.root
.dentry
= dentry
;
3651 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3652 return ERR_PTR(-ELOOP
);
3654 filename
= getname_kernel(name
);
3655 if (IS_ERR(filename
))
3656 return ERR_CAST(filename
);
3658 set_nameidata(&nd
, -1, filename
);
3659 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3660 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3661 file
= path_openat(&nd
, op
, flags
);
3662 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3663 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3664 restore_nameidata();
3669 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3670 struct path
*path
, unsigned int lookup_flags
)
3672 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3677 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3680 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3681 * other flags passed in are ignored!
3683 lookup_flags
&= LOOKUP_REVAL
;
3685 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3687 return ERR_CAST(name
);
3690 * Yucky last component or no last component at all?
3691 * (foo/., foo/.., /////)
3693 if (unlikely(type
!= LAST_NORM
))
3696 /* don't fail immediately if it's r/o, at least try to report other errors */
3697 err2
= mnt_want_write(path
->mnt
);
3699 * Do the final lookup.
3701 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3702 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3703 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3708 if (d_is_positive(dentry
))
3712 * Special case - lookup gave negative, but... we had foo/bar/
3713 * From the vfs_mknod() POV we just have a negative dentry -
3714 * all is fine. Let's be bastards - you had / on the end, you've
3715 * been asking for (non-existent) directory. -ENOENT for you.
3717 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3721 if (unlikely(err2
)) {
3729 dentry
= ERR_PTR(error
);
3731 inode_unlock(path
->dentry
->d_inode
);
3733 mnt_drop_write(path
->mnt
);
3740 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3741 struct path
*path
, unsigned int lookup_flags
)
3743 return filename_create(dfd
, getname_kernel(pathname
),
3744 path
, lookup_flags
);
3746 EXPORT_SYMBOL(kern_path_create
);
3748 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3751 inode_unlock(path
->dentry
->d_inode
);
3752 mnt_drop_write(path
->mnt
);
3755 EXPORT_SYMBOL(done_path_create
);
3757 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3758 struct path
*path
, unsigned int lookup_flags
)
3760 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3762 EXPORT_SYMBOL(user_path_create
);
3764 int vfs_mknod2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3766 int error
= may_create(mnt
, dir
, dentry
);
3771 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3774 if (!dir
->i_op
->mknod
)
3777 error
= devcgroup_inode_mknod(mode
, dev
);
3781 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3785 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3787 fsnotify_create(dir
, dentry
);
3790 EXPORT_SYMBOL(vfs_mknod2
);
3792 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3794 return vfs_mknod2(NULL
, dir
, dentry
, mode
, dev
);
3796 EXPORT_SYMBOL(vfs_mknod
);
3798 static int may_mknod(umode_t mode
)
3800 switch (mode
& S_IFMT
) {
3806 case 0: /* zero mode translates to S_IFREG */
3815 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3818 struct dentry
*dentry
;
3821 unsigned int lookup_flags
= 0;
3823 error
= may_mknod(mode
);
3827 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3829 return PTR_ERR(dentry
);
3831 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3832 mode
&= ~current_umask();
3833 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3836 switch (mode
& S_IFMT
) {
3837 case 0: case S_IFREG
:
3838 error
= vfs_create2(path
.mnt
, path
.dentry
->d_inode
,dentry
,mode
,true);
3840 ima_post_path_mknod(dentry
);
3842 case S_IFCHR
: case S_IFBLK
:
3843 error
= vfs_mknod2(path
.mnt
, path
.dentry
->d_inode
,dentry
,mode
,
3844 new_decode_dev(dev
));
3846 case S_IFIFO
: case S_IFSOCK
:
3847 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3851 done_path_create(&path
, dentry
);
3852 if (retry_estale(error
, lookup_flags
)) {
3853 lookup_flags
|= LOOKUP_REVAL
;
3859 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3861 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3864 int vfs_mkdir2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3866 int error
= may_create(mnt
, dir
, dentry
);
3867 unsigned max_links
= dir
->i_sb
->s_max_links
;
3872 if (!dir
->i_op
->mkdir
)
3875 mode
&= (S_IRWXUGO
|S_ISVTX
);
3876 error
= security_inode_mkdir(dir
, dentry
, mode
);
3880 if (max_links
&& dir
->i_nlink
>= max_links
)
3883 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3885 fsnotify_mkdir(dir
, dentry
);
3888 EXPORT_SYMBOL(vfs_mkdir2
);
3890 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3892 return vfs_mkdir2(NULL
, dir
, dentry
, mode
);
3894 EXPORT_SYMBOL(vfs_mkdir
);
3896 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3898 struct dentry
*dentry
;
3901 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3904 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3906 return PTR_ERR(dentry
);
3908 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3909 mode
&= ~current_umask();
3910 error
= security_path_mkdir(&path
, dentry
, mode
);
3912 error
= vfs_mkdir2(path
.mnt
, path
.dentry
->d_inode
, dentry
, mode
);
3913 done_path_create(&path
, dentry
);
3914 if (retry_estale(error
, lookup_flags
)) {
3915 lookup_flags
|= LOOKUP_REVAL
;
3921 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3923 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3926 int vfs_rmdir2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
)
3928 int error
= may_delete(mnt
, dir
, dentry
, 1);
3933 if (!dir
->i_op
->rmdir
)
3937 inode_lock(dentry
->d_inode
);
3940 if (is_local_mountpoint(dentry
))
3943 error
= security_inode_rmdir(dir
, dentry
);
3947 shrink_dcache_parent(dentry
);
3948 error
= dir
->i_op
->rmdir(dir
, dentry
);
3952 dentry
->d_inode
->i_flags
|= S_DEAD
;
3954 detach_mounts(dentry
);
3957 inode_unlock(dentry
->d_inode
);
3963 EXPORT_SYMBOL(vfs_rmdir2
);
3965 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3967 return vfs_rmdir2(NULL
, dir
, dentry
);
3969 EXPORT_SYMBOL(vfs_rmdir
);
3971 static long do_rmdir(int dfd
, const char __user
*pathname
)
3974 struct filename
*name
;
3975 struct dentry
*dentry
;
3979 unsigned int lookup_flags
= 0;
3981 name
= filename_parentat(dfd
, getname(pathname
), lookup_flags
,
3982 &path
, &last
, &type
);
3984 return PTR_ERR(name
);
3998 error
= mnt_want_write(path
.mnt
);
4002 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4003 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
4004 error
= PTR_ERR(dentry
);
4007 if (!dentry
->d_inode
) {
4011 error
= security_path_rmdir(&path
, dentry
);
4014 error
= vfs_rmdir2(path
.mnt
, path
.dentry
->d_inode
, dentry
);
4018 inode_unlock(path
.dentry
->d_inode
);
4019 mnt_drop_write(path
.mnt
);
4023 if (retry_estale(error
, lookup_flags
)) {
4024 lookup_flags
|= LOOKUP_REVAL
;
4030 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
4032 return do_rmdir(AT_FDCWD
, pathname
);
4036 * vfs_unlink - unlink a filesystem object
4037 * @dir: parent directory
4039 * @delegated_inode: returns victim inode, if the inode is delegated.
4041 * The caller must hold dir->i_mutex.
4043 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4044 * return a reference to the inode in delegated_inode. The caller
4045 * should then break the delegation on that inode and retry. Because
4046 * breaking a delegation may take a long time, the caller should drop
4047 * dir->i_mutex before doing so.
4049 * Alternatively, a caller may pass NULL for delegated_inode. This may
4050 * be appropriate for callers that expect the underlying filesystem not
4051 * to be NFS exported.
4053 int vfs_unlink2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
4055 struct inode
*target
= dentry
->d_inode
;
4056 int error
= may_delete(mnt
, dir
, dentry
, 0);
4061 if (!dir
->i_op
->unlink
)
4065 if (is_local_mountpoint(dentry
))
4068 error
= security_inode_unlink(dir
, dentry
);
4070 error
= try_break_deleg(target
, delegated_inode
);
4073 error
= dir
->i_op
->unlink(dir
, dentry
);
4076 detach_mounts(dentry
);
4081 inode_unlock(target
);
4083 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4084 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
4085 fsnotify_link_count(target
);
4091 EXPORT_SYMBOL(vfs_unlink2
);
4093 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
4095 return vfs_unlink2(NULL
, dir
, dentry
, delegated_inode
);
4097 EXPORT_SYMBOL(vfs_unlink
);
4100 * Make sure that the actual truncation of the file will occur outside its
4101 * directory's i_mutex. Truncate can take a long time if there is a lot of
4102 * writeout happening, and we don't want to prevent access to the directory
4103 * while waiting on the I/O.
4105 static long do_unlinkat(int dfd
, const char __user
*pathname
)
4108 struct filename
*name
;
4109 struct dentry
*dentry
;
4113 struct inode
*inode
= NULL
;
4114 struct inode
*delegated_inode
= NULL
;
4115 unsigned int lookup_flags
= 0;
4117 name
= filename_parentat(dfd
, getname(pathname
), lookup_flags
,
4118 &path
, &last
, &type
);
4120 return PTR_ERR(name
);
4123 if (type
!= LAST_NORM
)
4126 error
= mnt_want_write(path
.mnt
);
4130 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4131 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
4132 error
= PTR_ERR(dentry
);
4133 if (!IS_ERR(dentry
)) {
4134 /* Why not before? Because we want correct error value */
4135 if (last
.name
[last
.len
])
4137 inode
= dentry
->d_inode
;
4138 if (d_is_negative(dentry
))
4141 error
= security_path_unlink(&path
, dentry
);
4144 error
= vfs_unlink2(path
.mnt
, path
.dentry
->d_inode
, dentry
, &delegated_inode
);
4148 inode_unlock(path
.dentry
->d_inode
);
4150 iput(inode
); /* truncate the inode here */
4152 if (delegated_inode
) {
4153 error
= break_deleg_wait(&delegated_inode
);
4157 mnt_drop_write(path
.mnt
);
4161 if (retry_estale(error
, lookup_flags
)) {
4162 lookup_flags
|= LOOKUP_REVAL
;
4169 if (d_is_negative(dentry
))
4171 else if (d_is_dir(dentry
))
4178 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4180 if ((flag
& ~AT_REMOVEDIR
) != 0)
4183 if (flag
& AT_REMOVEDIR
)
4184 return do_rmdir(dfd
, pathname
);
4186 return do_unlinkat(dfd
, pathname
);
4189 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4191 return do_unlinkat(AT_FDCWD
, pathname
);
4194 int vfs_symlink2(struct vfsmount
*mnt
, struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
4196 int error
= may_create(mnt
, dir
, dentry
);
4201 if (!dir
->i_op
->symlink
)
4204 error
= security_inode_symlink(dir
, dentry
, oldname
);
4208 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
4210 fsnotify_create(dir
, dentry
);
4213 EXPORT_SYMBOL(vfs_symlink2
);
4215 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
4217 return vfs_symlink2(NULL
, dir
, dentry
, oldname
);
4219 EXPORT_SYMBOL(vfs_symlink
);
4221 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4222 int, newdfd
, const char __user
*, newname
)
4225 struct filename
*from
;
4226 struct dentry
*dentry
;
4228 unsigned int lookup_flags
= 0;
4230 from
= getname(oldname
);
4232 return PTR_ERR(from
);
4234 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4235 error
= PTR_ERR(dentry
);
4239 error
= security_path_symlink(&path
, dentry
, from
->name
);
4241 error
= vfs_symlink2(path
.mnt
, path
.dentry
->d_inode
, dentry
, from
->name
);
4242 done_path_create(&path
, dentry
);
4243 if (retry_estale(error
, lookup_flags
)) {
4244 lookup_flags
|= LOOKUP_REVAL
;
4252 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4254 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4258 * vfs_link - create a new link
4259 * @old_dentry: object to be linked
4261 * @new_dentry: where to create the new link
4262 * @delegated_inode: returns inode needing a delegation break
4264 * The caller must hold dir->i_mutex
4266 * If vfs_link discovers a delegation on the to-be-linked file in need
4267 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4268 * inode in delegated_inode. The caller should then break the delegation
4269 * and retry. Because breaking a delegation may take a long time, the
4270 * caller should drop the i_mutex before doing so.
4272 * Alternatively, a caller may pass NULL for delegated_inode. This may
4273 * be appropriate for callers that expect the underlying filesystem not
4274 * to be NFS exported.
4276 int vfs_link2(struct vfsmount
*mnt
, struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4278 struct inode
*inode
= old_dentry
->d_inode
;
4279 unsigned max_links
= dir
->i_sb
->s_max_links
;
4285 error
= may_create(mnt
, dir
, new_dentry
);
4289 if (dir
->i_sb
!= inode
->i_sb
)
4293 * A link to an append-only or immutable file cannot be created.
4295 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4298 * Updating the link count will likely cause i_uid and i_gid to
4299 * be writen back improperly if their true value is unknown to
4302 if (HAS_UNMAPPED_ID(inode
))
4304 if (!dir
->i_op
->link
)
4306 if (S_ISDIR(inode
->i_mode
))
4309 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4314 /* Make sure we don't allow creating hardlink to an unlinked file */
4315 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4317 else if (max_links
&& inode
->i_nlink
>= max_links
)
4320 error
= try_break_deleg(inode
, delegated_inode
);
4322 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4325 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4326 spin_lock(&inode
->i_lock
);
4327 inode
->i_state
&= ~I_LINKABLE
;
4328 spin_unlock(&inode
->i_lock
);
4330 inode_unlock(inode
);
4332 fsnotify_link(dir
, inode
, new_dentry
);
4335 EXPORT_SYMBOL(vfs_link2
);
4337 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4339 return vfs_link2(NULL
, old_dentry
, dir
, new_dentry
, delegated_inode
);
4341 EXPORT_SYMBOL(vfs_link
);
4344 * Hardlinks are often used in delicate situations. We avoid
4345 * security-related surprises by not following symlinks on the
4348 * We don't follow them on the oldname either to be compatible
4349 * with linux 2.0, and to avoid hard-linking to directories
4350 * and other special files. --ADM
4352 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4353 int, newdfd
, const char __user
*, newname
, int, flags
)
4355 struct dentry
*new_dentry
;
4356 struct path old_path
, new_path
;
4357 struct inode
*delegated_inode
= NULL
;
4361 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4364 * To use null names we require CAP_DAC_READ_SEARCH
4365 * This ensures that not everyone will be able to create
4366 * handlink using the passed filedescriptor.
4368 if (flags
& AT_EMPTY_PATH
) {
4369 if (!capable(CAP_DAC_READ_SEARCH
))
4374 if (flags
& AT_SYMLINK_FOLLOW
)
4375 how
|= LOOKUP_FOLLOW
;
4377 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4381 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4382 (how
& LOOKUP_REVAL
));
4383 error
= PTR_ERR(new_dentry
);
4384 if (IS_ERR(new_dentry
))
4388 if (old_path
.mnt
!= new_path
.mnt
)
4390 error
= may_linkat(&old_path
);
4391 if (unlikely(error
))
4393 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4396 error
= vfs_link2(old_path
.mnt
, old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4398 done_path_create(&new_path
, new_dentry
);
4399 if (delegated_inode
) {
4400 error
= break_deleg_wait(&delegated_inode
);
4402 path_put(&old_path
);
4406 if (retry_estale(error
, how
)) {
4407 path_put(&old_path
);
4408 how
|= LOOKUP_REVAL
;
4412 path_put(&old_path
);
4417 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4419 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4423 * vfs_rename - rename a filesystem object
4424 * @old_dir: parent of source
4425 * @old_dentry: source
4426 * @new_dir: parent of destination
4427 * @new_dentry: destination
4428 * @delegated_inode: returns an inode needing a delegation break
4429 * @flags: rename flags
4431 * The caller must hold multiple mutexes--see lock_rename()).
4433 * If vfs_rename discovers a delegation in need of breaking at either
4434 * the source or destination, it will return -EWOULDBLOCK and return a
4435 * reference to the inode in delegated_inode. The caller should then
4436 * break the delegation and retry. Because breaking a delegation may
4437 * take a long time, the caller should drop all locks before doing
4440 * Alternatively, a caller may pass NULL for delegated_inode. This may
4441 * be appropriate for callers that expect the underlying filesystem not
4442 * to be NFS exported.
4444 * The worst of all namespace operations - renaming directory. "Perverted"
4445 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4448 * a) we can get into loop creation.
4449 * b) race potential - two innocent renames can create a loop together.
4450 * That's where 4.4 screws up. Current fix: serialization on
4451 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4453 * c) we have to lock _four_ objects - parents and victim (if it exists),
4454 * and source (if it is not a directory).
4455 * And that - after we got ->i_mutex on parents (until then we don't know
4456 * whether the target exists). Solution: try to be smart with locking
4457 * order for inodes. We rely on the fact that tree topology may change
4458 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4459 * move will be locked. Thus we can rank directories by the tree
4460 * (ancestors first) and rank all non-directories after them.
4461 * That works since everybody except rename does "lock parent, lookup,
4462 * lock child" and rename is under ->s_vfs_rename_mutex.
4463 * HOWEVER, it relies on the assumption that any object with ->lookup()
4464 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4465 * we'd better make sure that there's no link(2) for them.
4466 * d) conversion from fhandle to dentry may come in the wrong moment - when
4467 * we are removing the target. Solution: we will have to grab ->i_mutex
4468 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4469 * ->i_mutex on parents, which works but leads to some truly excessive
4472 int vfs_rename2(struct vfsmount
*mnt
,
4473 struct inode
*old_dir
, struct dentry
*old_dentry
,
4474 struct inode
*new_dir
, struct dentry
*new_dentry
,
4475 struct inode
**delegated_inode
, unsigned int flags
)
4478 bool is_dir
= d_is_dir(old_dentry
);
4479 struct inode
*source
= old_dentry
->d_inode
;
4480 struct inode
*target
= new_dentry
->d_inode
;
4481 bool new_is_dir
= false;
4482 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4483 struct name_snapshot old_name
;
4485 if (source
== target
)
4488 error
= may_delete(mnt
, old_dir
, old_dentry
, is_dir
);
4493 error
= may_create(mnt
, new_dir
, new_dentry
);
4495 new_is_dir
= d_is_dir(new_dentry
);
4497 if (!(flags
& RENAME_EXCHANGE
))
4498 error
= may_delete(mnt
, new_dir
, new_dentry
, is_dir
);
4500 error
= may_delete(mnt
, new_dir
, new_dentry
, new_is_dir
);
4505 if (!old_dir
->i_op
->rename
)
4509 * If we are going to change the parent - check write permissions,
4510 * we'll need to flip '..'.
4512 if (new_dir
!= old_dir
) {
4514 error
= inode_permission2(mnt
, source
, MAY_WRITE
);
4518 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4519 error
= inode_permission2(mnt
, target
, MAY_WRITE
);
4525 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4530 take_dentry_name_snapshot(&old_name
, old_dentry
);
4532 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4533 lock_two_nondirectories(source
, target
);
4538 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4541 if (max_links
&& new_dir
!= old_dir
) {
4543 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4545 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4546 old_dir
->i_nlink
>= max_links
)
4549 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4550 shrink_dcache_parent(new_dentry
);
4552 error
= try_break_deleg(source
, delegated_inode
);
4556 if (target
&& !new_is_dir
) {
4557 error
= try_break_deleg(target
, delegated_inode
);
4561 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4562 new_dir
, new_dentry
, flags
);
4566 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4568 target
->i_flags
|= S_DEAD
;
4569 dont_mount(new_dentry
);
4570 detach_mounts(new_dentry
);
4572 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4573 if (!(flags
& RENAME_EXCHANGE
))
4574 d_move(old_dentry
, new_dentry
);
4576 d_exchange(old_dentry
, new_dentry
);
4579 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4580 unlock_two_nondirectories(source
, target
);
4582 inode_unlock(target
);
4585 fsnotify_move(old_dir
, new_dir
, old_name
.name
, is_dir
,
4586 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4587 if (flags
& RENAME_EXCHANGE
) {
4588 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4589 new_is_dir
, NULL
, new_dentry
);
4592 release_dentry_name_snapshot(&old_name
);
4596 EXPORT_SYMBOL(vfs_rename2
);
4598 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4599 struct inode
*new_dir
, struct dentry
*new_dentry
,
4600 struct inode
**delegated_inode
, unsigned int flags
)
4602 return vfs_rename2(NULL
, old_dir
, old_dentry
, new_dir
, new_dentry
, delegated_inode
, flags
);
4604 EXPORT_SYMBOL(vfs_rename
);
4606 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4607 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4609 struct dentry
*old_dentry
, *new_dentry
;
4610 struct dentry
*trap
;
4611 struct path old_path
, new_path
;
4612 struct qstr old_last
, new_last
;
4613 int old_type
, new_type
;
4614 struct inode
*delegated_inode
= NULL
;
4615 struct filename
*from
;
4616 struct filename
*to
;
4617 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4618 bool should_retry
= false;
4621 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4624 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4625 (flags
& RENAME_EXCHANGE
))
4628 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4631 if (flags
& RENAME_EXCHANGE
)
4635 from
= filename_parentat(olddfd
, getname(oldname
), lookup_flags
,
4636 &old_path
, &old_last
, &old_type
);
4638 error
= PTR_ERR(from
);
4642 to
= filename_parentat(newdfd
, getname(newname
), lookup_flags
,
4643 &new_path
, &new_last
, &new_type
);
4645 error
= PTR_ERR(to
);
4650 if (old_path
.mnt
!= new_path
.mnt
)
4654 if (old_type
!= LAST_NORM
)
4657 if (flags
& RENAME_NOREPLACE
)
4659 if (new_type
!= LAST_NORM
)
4662 error
= mnt_want_write(old_path
.mnt
);
4667 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4669 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4670 error
= PTR_ERR(old_dentry
);
4671 if (IS_ERR(old_dentry
))
4673 /* source must exist */
4675 if (d_is_negative(old_dentry
))
4677 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4678 error
= PTR_ERR(new_dentry
);
4679 if (IS_ERR(new_dentry
))
4682 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4684 if (flags
& RENAME_EXCHANGE
) {
4686 if (d_is_negative(new_dentry
))
4689 if (!d_is_dir(new_dentry
)) {
4691 if (new_last
.name
[new_last
.len
])
4695 /* unless the source is a directory trailing slashes give -ENOTDIR */
4696 if (!d_is_dir(old_dentry
)) {
4698 if (old_last
.name
[old_last
.len
])
4700 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4703 /* source should not be ancestor of target */
4705 if (old_dentry
== trap
)
4707 /* target should not be an ancestor of source */
4708 if (!(flags
& RENAME_EXCHANGE
))
4710 if (new_dentry
== trap
)
4713 error
= security_path_rename(&old_path
, old_dentry
,
4714 &new_path
, new_dentry
, flags
);
4717 error
= vfs_rename2(old_path
.mnt
, old_path
.dentry
->d_inode
, old_dentry
,
4718 new_path
.dentry
->d_inode
, new_dentry
,
4719 &delegated_inode
, flags
);
4725 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4726 if (delegated_inode
) {
4727 error
= break_deleg_wait(&delegated_inode
);
4731 mnt_drop_write(old_path
.mnt
);
4733 if (retry_estale(error
, lookup_flags
))
4734 should_retry
= true;
4735 path_put(&new_path
);
4738 path_put(&old_path
);
4741 should_retry
= false;
4742 lookup_flags
|= LOOKUP_REVAL
;
4749 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4750 int, newdfd
, const char __user
*, newname
)
4752 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4755 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4757 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4760 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4762 int error
= may_create(NULL
, dir
, dentry
);
4766 if (!dir
->i_op
->mknod
)
4769 return dir
->i_op
->mknod(dir
, dentry
,
4770 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4772 EXPORT_SYMBOL(vfs_whiteout
);
4774 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4776 int len
= PTR_ERR(link
);
4781 if (len
> (unsigned) buflen
)
4783 if (copy_to_user(buffer
, link
, len
))
4790 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4791 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4792 * for any given inode is up to filesystem.
4794 static int generic_readlink(struct dentry
*dentry
, char __user
*buffer
,
4797 DEFINE_DELAYED_CALL(done
);
4798 struct inode
*inode
= d_inode(dentry
);
4799 const char *link
= inode
->i_link
;
4803 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4805 return PTR_ERR(link
);
4807 res
= readlink_copy(buffer
, buflen
, link
);
4808 do_delayed_call(&done
);
4813 * vfs_readlink - copy symlink body into userspace buffer
4814 * @dentry: dentry on which to get symbolic link
4815 * @buffer: user memory pointer
4816 * @buflen: size of buffer
4818 * Does not touch atime. That's up to the caller if necessary
4820 * Does not call security hook.
4822 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4824 struct inode
*inode
= d_inode(dentry
);
4826 if (unlikely(!(inode
->i_opflags
& IOP_DEFAULT_READLINK
))) {
4827 if (unlikely(inode
->i_op
->readlink
))
4828 return inode
->i_op
->readlink(dentry
, buffer
, buflen
);
4830 if (!d_is_symlink(dentry
))
4833 spin_lock(&inode
->i_lock
);
4834 inode
->i_opflags
|= IOP_DEFAULT_READLINK
;
4835 spin_unlock(&inode
->i_lock
);
4838 return generic_readlink(dentry
, buffer
, buflen
);
4840 EXPORT_SYMBOL(vfs_readlink
);
4843 * vfs_get_link - get symlink body
4844 * @dentry: dentry on which to get symbolic link
4845 * @done: caller needs to free returned data with this
4847 * Calls security hook and i_op->get_link() on the supplied inode.
4849 * It does not touch atime. That's up to the caller if necessary.
4851 * Does not work on "special" symlinks like /proc/$$/fd/N
4853 const char *vfs_get_link(struct dentry
*dentry
, struct delayed_call
*done
)
4855 const char *res
= ERR_PTR(-EINVAL
);
4856 struct inode
*inode
= d_inode(dentry
);
4858 if (d_is_symlink(dentry
)) {
4859 res
= ERR_PTR(security_inode_readlink(dentry
));
4861 res
= inode
->i_op
->get_link(dentry
, inode
, done
);
4865 EXPORT_SYMBOL(vfs_get_link
);
4867 /* get the link contents into pagecache */
4868 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4869 struct delayed_call
*callback
)
4873 struct address_space
*mapping
= inode
->i_mapping
;
4876 page
= find_get_page(mapping
, 0);
4878 return ERR_PTR(-ECHILD
);
4879 if (!PageUptodate(page
)) {
4881 return ERR_PTR(-ECHILD
);
4884 page
= read_mapping_page(mapping
, 0, NULL
);
4888 set_delayed_call(callback
, page_put_link
, page
);
4889 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4890 kaddr
= page_address(page
);
4891 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4895 EXPORT_SYMBOL(page_get_link
);
4897 void page_put_link(void *arg
)
4901 EXPORT_SYMBOL(page_put_link
);
4903 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4905 DEFINE_DELAYED_CALL(done
);
4906 int res
= readlink_copy(buffer
, buflen
,
4907 page_get_link(dentry
, d_inode(dentry
),
4909 do_delayed_call(&done
);
4912 EXPORT_SYMBOL(page_readlink
);
4915 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4917 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4919 struct address_space
*mapping
= inode
->i_mapping
;
4923 unsigned int flags
= 0;
4925 flags
|= AOP_FLAG_NOFS
;
4928 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4929 flags
, &page
, &fsdata
);
4933 memcpy(page_address(page
), symname
, len
-1);
4935 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4942 mark_inode_dirty(inode
);
4947 EXPORT_SYMBOL(__page_symlink
);
4949 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4951 return __page_symlink(inode
, symname
, len
,
4952 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4954 EXPORT_SYMBOL(page_symlink
);
4956 const struct inode_operations page_symlink_inode_operations
= {
4957 .get_link
= page_get_link
,
4959 EXPORT_SYMBOL(page_symlink_inode_operations
);