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
);
263 * check_acl - perform ACL permission checking
264 * @mnt_userns: user namespace of the mount the inode was found from
265 * @inode: inode to check permissions on
266 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
268 * This function performs the ACL permission checking. Since this function
269 * retrieve POSIX acls it needs to know whether it is called from a blocking or
270 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
272 * If the inode has been found through an idmapped mount the user namespace of
273 * the vfsmount must be passed through @mnt_userns. This function will then take
274 * care to map the inode according to @mnt_userns before checking permissions.
275 * On non-idmapped mounts or if permission checking is to be performed on the
276 * raw inode simply passs init_user_ns.
278 static int check_acl(struct user_namespace
*mnt_userns
,
279 struct inode
*inode
, int mask
)
281 #ifdef CONFIG_FS_POSIX_ACL
282 struct posix_acl
*acl
;
284 if (mask
& MAY_NOT_BLOCK
) {
285 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
288 /* no ->get_acl() calls in RCU mode... */
289 if (is_uncached_acl(acl
))
291 return posix_acl_permission(mnt_userns
, inode
, acl
, mask
);
294 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
298 int error
= posix_acl_permission(mnt_userns
, inode
, acl
, mask
);
299 posix_acl_release(acl
);
308 * acl_permission_check - perform basic UNIX permission checking
309 * @mnt_userns: user namespace of the mount the inode was found from
310 * @inode: inode to check permissions on
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
313 * This function performs the basic UNIX permission checking. Since this
314 * function may retrieve POSIX acls it needs to know whether it is called from a
315 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
317 * If the inode has been found through an idmapped mount the user namespace of
318 * the vfsmount must be passed through @mnt_userns. This function will then take
319 * care to map the inode according to @mnt_userns before checking permissions.
320 * On non-idmapped mounts or if permission checking is to be performed on the
321 * raw inode simply passs init_user_ns.
323 static int acl_permission_check(struct user_namespace
*mnt_userns
,
324 struct inode
*inode
, int mask
)
326 unsigned int mode
= inode
->i_mode
;
329 /* Are we the owner? If so, ACL's don't matter */
330 i_uid
= i_uid_into_mnt(mnt_userns
, inode
);
331 if (likely(uid_eq(current_fsuid(), i_uid
))) {
334 return (mask
& ~mode
) ? -EACCES
: 0;
337 /* Do we have ACL's? */
338 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
339 int error
= check_acl(mnt_userns
, inode
, mask
);
340 if (error
!= -EAGAIN
)
344 /* Only RWX matters for group/other mode bits */
348 * Are the group permissions different from
349 * the other permissions in the bits we care
350 * about? Need to check group ownership if so.
352 if (mask
& (mode
^ (mode
>> 3))) {
353 kgid_t kgid
= i_gid_into_mnt(mnt_userns
, inode
);
354 if (in_group_p(kgid
))
358 /* Bits in 'mode' clear that we require? */
359 return (mask
& ~mode
) ? -EACCES
: 0;
363 * generic_permission - check for access rights on a Posix-like filesystem
364 * @mnt_userns: user namespace of the mount the inode was found from
365 * @inode: inode to check access rights for
366 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
367 * %MAY_NOT_BLOCK ...)
369 * Used to check for read/write/execute permissions on a file.
370 * We use "fsuid" for this, letting us set arbitrary permissions
371 * for filesystem access without changing the "normal" uids which
372 * are used for other things.
374 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
375 * request cannot be satisfied (eg. requires blocking or too much complexity).
376 * It would then be called again in ref-walk mode.
378 * If the inode has been found through an idmapped mount the user namespace of
379 * the vfsmount must be passed through @mnt_userns. This function will then take
380 * care to map the inode according to @mnt_userns before checking permissions.
381 * On non-idmapped mounts or if permission checking is to be performed on the
382 * raw inode simply passs init_user_ns.
384 int generic_permission(struct user_namespace
*mnt_userns
, struct inode
*inode
,
390 * Do the basic permission checks.
392 ret
= acl_permission_check(mnt_userns
, inode
, mask
);
396 if (S_ISDIR(inode
->i_mode
)) {
397 /* DACs are overridable for directories */
398 if (!(mask
& MAY_WRITE
))
399 if (capable_wrt_inode_uidgid(mnt_userns
, inode
,
400 CAP_DAC_READ_SEARCH
))
402 if (capable_wrt_inode_uidgid(mnt_userns
, inode
,
409 * Searching includes executable on directories, else just read.
411 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
412 if (mask
== MAY_READ
)
413 if (capable_wrt_inode_uidgid(mnt_userns
, inode
,
414 CAP_DAC_READ_SEARCH
))
417 * Read/write DACs are always overridable.
418 * Executable DACs are overridable when there is
419 * at least one exec bit set.
421 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
422 if (capable_wrt_inode_uidgid(mnt_userns
, inode
,
428 EXPORT_SYMBOL(generic_permission
);
431 * do_inode_permission - UNIX permission checking
432 * @mnt_userns: user namespace of the mount the inode was found from
433 * @inode: inode to check permissions on
434 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
436 * We _really_ want to just do "generic_permission()" without
437 * even looking at the inode->i_op values. So we keep a cache
438 * flag in inode->i_opflags, that says "this has not special
439 * permission function, use the fast case".
441 static inline int do_inode_permission(struct user_namespace
*mnt_userns
,
442 struct inode
*inode
, int mask
)
444 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
445 if (likely(inode
->i_op
->permission
))
446 return inode
->i_op
->permission(mnt_userns
, inode
, mask
);
448 /* This gets set once for the inode lifetime */
449 spin_lock(&inode
->i_lock
);
450 inode
->i_opflags
|= IOP_FASTPERM
;
451 spin_unlock(&inode
->i_lock
);
453 return generic_permission(mnt_userns
, inode
, mask
);
457 * sb_permission - Check superblock-level permissions
458 * @sb: Superblock of inode to check permission on
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Separate out file-system wide checks from inode-specific permission checks.
464 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
466 if (unlikely(mask
& MAY_WRITE
)) {
467 umode_t mode
= inode
->i_mode
;
469 /* Nobody gets write access to a read-only fs. */
470 if (sb_rdonly(sb
) && (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
477 * inode_permission - Check for access rights to a given inode
478 * @mnt_userns: User namespace of the mount the inode was found from
479 * @inode: Inode to check permission on
480 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
482 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
483 * this, letting us set arbitrary permissions for filesystem access without
484 * changing the "normal" UIDs which are used for other things.
486 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
488 int inode_permission(struct user_namespace
*mnt_userns
,
489 struct inode
*inode
, int mask
)
493 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
497 if (unlikely(mask
& MAY_WRITE
)) {
499 * Nobody gets write access to an immutable file.
501 if (IS_IMMUTABLE(inode
))
505 * Updating mtime will likely cause i_uid and i_gid to be
506 * written back improperly if their true value is unknown
509 if (HAS_UNMAPPED_ID(mnt_userns
, inode
))
513 retval
= do_inode_permission(mnt_userns
, inode
, mask
);
517 retval
= devcgroup_inode_permission(inode
, mask
);
521 return security_inode_permission(inode
, mask
);
523 EXPORT_SYMBOL(inode_permission
);
526 * path_get - get a reference to a path
527 * @path: path to get the reference to
529 * Given a path increment the reference count to the dentry and the vfsmount.
531 void path_get(const struct path
*path
)
536 EXPORT_SYMBOL(path_get
);
539 * path_put - put a reference to a path
540 * @path: path to put the reference to
542 * Given a path decrement the reference count to the dentry and the vfsmount.
544 void path_put(const struct path
*path
)
549 EXPORT_SYMBOL(path_put
);
551 #define EMBEDDED_LEVELS 2
556 struct inode
*inode
; /* path.dentry.d_inode */
558 unsigned seq
, m_seq
, r_seq
;
561 int total_link_count
;
564 struct delayed_call done
;
567 } *stack
, internal
[EMBEDDED_LEVELS
];
568 struct filename
*name
;
569 struct nameidata
*saved
;
574 } __randomize_layout
;
576 static void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
578 struct nameidata
*old
= current
->nameidata
;
579 p
->stack
= p
->internal
;
582 p
->total_link_count
= old
? old
->total_link_count
: 0;
584 current
->nameidata
= p
;
587 static void restore_nameidata(void)
589 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
591 current
->nameidata
= old
;
593 old
->total_link_count
= now
->total_link_count
;
594 if (now
->stack
!= now
->internal
)
598 static bool nd_alloc_stack(struct nameidata
*nd
)
602 p
= kmalloc_array(MAXSYMLINKS
, sizeof(struct saved
),
603 nd
->flags
& LOOKUP_RCU
? GFP_ATOMIC
: GFP_KERNEL
);
606 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
612 * path_connected - Verify that a dentry is below mnt.mnt_root
614 * Rename can sometimes move a file or directory outside of a bind
615 * mount, path_connected allows those cases to be detected.
617 static bool path_connected(struct vfsmount
*mnt
, struct dentry
*dentry
)
619 struct super_block
*sb
= mnt
->mnt_sb
;
621 /* Bind mounts can have disconnected paths */
622 if (mnt
->mnt_root
== sb
->s_root
)
625 return is_subdir(dentry
, mnt
->mnt_root
);
628 static void drop_links(struct nameidata
*nd
)
632 struct saved
*last
= nd
->stack
+ i
;
633 do_delayed_call(&last
->done
);
634 clear_delayed_call(&last
->done
);
638 static void terminate_walk(struct nameidata
*nd
)
641 if (!(nd
->flags
& LOOKUP_RCU
)) {
644 for (i
= 0; i
< nd
->depth
; i
++)
645 path_put(&nd
->stack
[i
].link
);
646 if (nd
->flags
& LOOKUP_ROOT_GRABBED
) {
648 nd
->flags
&= ~LOOKUP_ROOT_GRABBED
;
651 nd
->flags
&= ~LOOKUP_RCU
;
657 /* path_put is needed afterwards regardless of success or failure */
658 static bool __legitimize_path(struct path
*path
, unsigned seq
, unsigned mseq
)
660 int res
= __legitimize_mnt(path
->mnt
, mseq
);
667 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
671 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
674 static inline bool legitimize_path(struct nameidata
*nd
,
675 struct path
*path
, unsigned seq
)
677 return __legitimize_path(path
, seq
, nd
->m_seq
);
680 static bool legitimize_links(struct nameidata
*nd
)
683 if (unlikely(nd
->flags
& LOOKUP_CACHED
)) {
688 for (i
= 0; i
< nd
->depth
; i
++) {
689 struct saved
*last
= nd
->stack
+ i
;
690 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
699 static bool legitimize_root(struct nameidata
*nd
)
702 * For scoped-lookups (where nd->root has been zeroed), we need to
703 * restart the whole lookup from scratch -- because set_root() is wrong
704 * for these lookups (nd->dfd is the root, not the filesystem root).
706 if (!nd
->root
.mnt
&& (nd
->flags
& LOOKUP_IS_SCOPED
))
708 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
709 if (!nd
->root
.mnt
|| (nd
->flags
& LOOKUP_ROOT
))
711 nd
->flags
|= LOOKUP_ROOT_GRABBED
;
712 return legitimize_path(nd
, &nd
->root
, nd
->root_seq
);
716 * Path walking has 2 modes, rcu-walk and ref-walk (see
717 * Documentation/filesystems/path-lookup.txt). In situations when we can't
718 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
719 * normal reference counts on dentries and vfsmounts to transition to ref-walk
720 * mode. Refcounts are grabbed at the last known good point before rcu-walk
721 * got stuck, so ref-walk may continue from there. If this is not successful
722 * (eg. a seqcount has changed), then failure is returned and it's up to caller
723 * to restart the path walk from the beginning in ref-walk mode.
727 * try_to_unlazy - try to switch to ref-walk mode.
728 * @nd: nameidata pathwalk data
729 * Returns: true on success, false on failure
731 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
733 * Must be called from rcu-walk context.
734 * Nothing should touch nameidata between try_to_unlazy() failure and
737 static bool try_to_unlazy(struct nameidata
*nd
)
739 struct dentry
*parent
= nd
->path
.dentry
;
741 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
743 nd
->flags
&= ~LOOKUP_RCU
;
744 if (unlikely(!legitimize_links(nd
)))
746 if (unlikely(!legitimize_path(nd
, &nd
->path
, nd
->seq
)))
748 if (unlikely(!legitimize_root(nd
)))
751 BUG_ON(nd
->inode
!= parent
->d_inode
);
756 nd
->path
.dentry
= NULL
;
763 * try_to_unlazy_next - try to switch to ref-walk mode.
764 * @nd: nameidata pathwalk data
765 * @dentry: next dentry to step into
766 * @seq: seq number to check @dentry against
767 * Returns: true on success, false on failure
769 * Similar to to try_to_unlazy(), but here we have the next dentry already
770 * picked by rcu-walk and want to legitimize that in addition to the current
771 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
772 * Nothing should touch nameidata between try_to_unlazy_next() failure and
775 static bool try_to_unlazy_next(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
777 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
779 nd
->flags
&= ~LOOKUP_RCU
;
780 if (unlikely(!legitimize_links(nd
)))
782 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
784 if (unlikely(!lockref_get_not_dead(&nd
->path
.dentry
->d_lockref
)))
788 * We need to move both the parent and the dentry from the RCU domain
789 * to be properly refcounted. And the sequence number in the dentry
790 * validates *both* dentry counters, since we checked the sequence
791 * number of the parent after we got the child sequence number. So we
792 * know the parent must still be valid if the child sequence number is
794 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
)))
796 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
799 * Sequence counts matched. Now make sure that the root is
800 * still valid and get it if required.
802 if (unlikely(!legitimize_root(nd
)))
810 nd
->path
.dentry
= NULL
;
820 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
822 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
823 return dentry
->d_op
->d_revalidate(dentry
, flags
);
829 * complete_walk - successful completion of path walk
830 * @nd: pointer nameidata
832 * If we had been in RCU mode, drop out of it and legitimize nd->path.
833 * Revalidate the final result, unless we'd already done that during
834 * the path walk or the filesystem doesn't ask for it. Return 0 on
835 * success, -error on failure. In case of failure caller does not
836 * need to drop nd->path.
838 static int complete_walk(struct nameidata
*nd
)
840 struct dentry
*dentry
= nd
->path
.dentry
;
843 if (nd
->flags
& LOOKUP_RCU
) {
845 * We don't want to zero nd->root for scoped-lookups or
846 * externally-managed nd->root.
848 if (!(nd
->flags
& (LOOKUP_ROOT
| LOOKUP_IS_SCOPED
)))
850 nd
->flags
&= ~LOOKUP_CACHED
;
851 if (!try_to_unlazy(nd
))
855 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
)) {
857 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
858 * ever step outside the root during lookup" and should already
859 * be guaranteed by the rest of namei, we want to avoid a namei
860 * BUG resulting in userspace being given a path that was not
861 * scoped within the root at some point during the lookup.
863 * So, do a final sanity-check to make sure that in the
864 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
865 * we won't silently return an fd completely outside of the
866 * requested root to userspace.
868 * Userspace could move the path outside the root after this
869 * check, but as discussed elsewhere this is not a concern (the
870 * resolved file was inside the root at some point).
872 if (!path_is_under(&nd
->path
, &nd
->root
))
876 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
879 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
882 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
892 static int set_root(struct nameidata
*nd
)
894 struct fs_struct
*fs
= current
->fs
;
897 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
898 * still have to ensure it doesn't happen because it will cause a breakout
901 if (WARN_ON(nd
->flags
& LOOKUP_IS_SCOPED
))
902 return -ENOTRECOVERABLE
;
904 if (nd
->flags
& LOOKUP_RCU
) {
908 seq
= read_seqcount_begin(&fs
->seq
);
910 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
911 } while (read_seqcount_retry(&fs
->seq
, seq
));
913 get_fs_root(fs
, &nd
->root
);
914 nd
->flags
|= LOOKUP_ROOT_GRABBED
;
919 static int nd_jump_root(struct nameidata
*nd
)
921 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
923 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
)) {
924 /* Absolute path arguments to path_init() are allowed. */
925 if (nd
->path
.mnt
!= NULL
&& nd
->path
.mnt
!= nd
->root
.mnt
)
929 int error
= set_root(nd
);
933 if (nd
->flags
& LOOKUP_RCU
) {
937 nd
->inode
= d
->d_inode
;
938 nd
->seq
= nd
->root_seq
;
939 if (unlikely(read_seqcount_retry(&d
->d_seq
, nd
->seq
)))
945 nd
->inode
= nd
->path
.dentry
->d_inode
;
947 nd
->flags
|= LOOKUP_JUMPED
;
952 * Helper to directly jump to a known parsed path from ->get_link,
953 * caller must have taken a reference to path beforehand.
955 int nd_jump_link(struct path
*path
)
958 struct nameidata
*nd
= current
->nameidata
;
960 if (unlikely(nd
->flags
& LOOKUP_NO_MAGICLINKS
))
964 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
)) {
965 if (nd
->path
.mnt
!= path
->mnt
)
968 /* Not currently safe for scoped-lookups. */
969 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
))
974 nd
->inode
= nd
->path
.dentry
->d_inode
;
975 nd
->flags
|= LOOKUP_JUMPED
;
983 static inline void put_link(struct nameidata
*nd
)
985 struct saved
*last
= nd
->stack
+ --nd
->depth
;
986 do_delayed_call(&last
->done
);
987 if (!(nd
->flags
& LOOKUP_RCU
))
988 path_put(&last
->link
);
991 int sysctl_protected_symlinks __read_mostly
= 0;
992 int sysctl_protected_hardlinks __read_mostly
= 0;
993 int sysctl_protected_fifos __read_mostly
;
994 int sysctl_protected_regular __read_mostly
;
997 * may_follow_link - Check symlink following for unsafe situations
998 * @nd: nameidata pathwalk data
1000 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1001 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1002 * in a sticky world-writable directory. This is to protect privileged
1003 * processes from failing races against path names that may change out
1004 * from under them by way of other users creating malicious symlinks.
1005 * It will permit symlinks to be followed only when outside a sticky
1006 * world-writable directory, or when the uid of the symlink and follower
1007 * match, or when the directory owner matches the symlink's owner.
1009 * Returns 0 if following the symlink is allowed, -ve on error.
1011 static inline int may_follow_link(struct nameidata
*nd
, const struct inode
*inode
)
1013 struct user_namespace
*mnt_userns
;
1016 if (!sysctl_protected_symlinks
)
1019 mnt_userns
= mnt_user_ns(nd
->path
.mnt
);
1020 i_uid
= i_uid_into_mnt(mnt_userns
, inode
);
1021 /* Allowed if owner and follower match. */
1022 if (uid_eq(current_cred()->fsuid
, i_uid
))
1025 /* Allowed if parent directory not sticky and world-writable. */
1026 if ((nd
->dir_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
1029 /* Allowed if parent directory and link owner match. */
1030 if (uid_valid(nd
->dir_uid
) && uid_eq(nd
->dir_uid
, i_uid
))
1033 if (nd
->flags
& LOOKUP_RCU
)
1036 audit_inode(nd
->name
, nd
->stack
[0].link
.dentry
, 0);
1037 audit_log_path_denied(AUDIT_ANOM_LINK
, "follow_link");
1042 * safe_hardlink_source - Check for safe hardlink conditions
1043 * @mnt_userns: user namespace of the mount the inode was found from
1044 * @inode: the source inode to hardlink from
1046 * Return false if at least one of the following conditions:
1047 * - inode is not a regular file
1049 * - inode is setgid and group-exec
1050 * - access failure for read and write
1052 * Otherwise returns true.
1054 static bool safe_hardlink_source(struct user_namespace
*mnt_userns
,
1055 struct inode
*inode
)
1057 umode_t mode
= inode
->i_mode
;
1059 /* Special files should not get pinned to the filesystem. */
1063 /* Setuid files should not get pinned to the filesystem. */
1067 /* Executable setgid files should not get pinned to the filesystem. */
1068 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
1071 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1072 if (inode_permission(mnt_userns
, inode
, MAY_READ
| MAY_WRITE
))
1079 * may_linkat - Check permissions for creating a hardlink
1080 * @mnt_userns: user namespace of the mount the inode was found from
1081 * @link: the source to hardlink from
1083 * Block hardlink when all of:
1084 * - sysctl_protected_hardlinks enabled
1085 * - fsuid does not match inode
1086 * - hardlink source is unsafe (see safe_hardlink_source() above)
1087 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1089 * If the inode has been found through an idmapped mount the user namespace of
1090 * the vfsmount must be passed through @mnt_userns. This function will then take
1091 * care to map the inode according to @mnt_userns before checking permissions.
1092 * On non-idmapped mounts or if permission checking is to be performed on the
1093 * raw inode simply passs init_user_ns.
1095 * Returns 0 if successful, -ve on error.
1097 int may_linkat(struct user_namespace
*mnt_userns
, struct path
*link
)
1099 struct inode
*inode
= link
->dentry
->d_inode
;
1101 /* Inode writeback is not safe when the uid or gid are invalid. */
1102 if (!uid_valid(i_uid_into_mnt(mnt_userns
, inode
)) ||
1103 !gid_valid(i_gid_into_mnt(mnt_userns
, inode
)))
1106 if (!sysctl_protected_hardlinks
)
1109 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1110 * otherwise, it must be a safe source.
1112 if (safe_hardlink_source(mnt_userns
, inode
) ||
1113 inode_owner_or_capable(mnt_userns
, inode
))
1116 audit_log_path_denied(AUDIT_ANOM_LINK
, "linkat");
1121 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1122 * should be allowed, or not, on files that already
1124 * @mnt_userns: user namespace of the mount the inode was found from
1125 * @dir_mode: mode bits of directory
1126 * @dir_uid: owner of directory
1127 * @inode: the inode of the file to open
1129 * Block an O_CREAT open of a FIFO (or a regular file) when:
1130 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1131 * - the file already exists
1132 * - we are in a sticky directory
1133 * - we don't own the file
1134 * - the owner of the directory doesn't own the file
1135 * - the directory is world writable
1136 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1137 * the directory doesn't have to be world writable: being group writable will
1140 * If the inode has been found through an idmapped mount the user namespace of
1141 * the vfsmount must be passed through @mnt_userns. This function will then take
1142 * care to map the inode according to @mnt_userns before checking permissions.
1143 * On non-idmapped mounts or if permission checking is to be performed on the
1144 * raw inode simply passs init_user_ns.
1146 * Returns 0 if the open is allowed, -ve on error.
1148 static int may_create_in_sticky(struct user_namespace
*mnt_userns
,
1149 struct nameidata
*nd
, struct inode
*const inode
)
1151 umode_t dir_mode
= nd
->dir_mode
;
1152 kuid_t dir_uid
= nd
->dir_uid
;
1154 if ((!sysctl_protected_fifos
&& S_ISFIFO(inode
->i_mode
)) ||
1155 (!sysctl_protected_regular
&& S_ISREG(inode
->i_mode
)) ||
1156 likely(!(dir_mode
& S_ISVTX
)) ||
1157 uid_eq(i_uid_into_mnt(mnt_userns
, inode
), dir_uid
) ||
1158 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns
, inode
)))
1161 if (likely(dir_mode
& 0002) ||
1163 ((sysctl_protected_fifos
>= 2 && S_ISFIFO(inode
->i_mode
)) ||
1164 (sysctl_protected_regular
>= 2 && S_ISREG(inode
->i_mode
))))) {
1165 const char *operation
= S_ISFIFO(inode
->i_mode
) ?
1166 "sticky_create_fifo" :
1167 "sticky_create_regular";
1168 audit_log_path_denied(AUDIT_ANOM_CREAT
, operation
);
1175 * follow_up - Find the mountpoint of path's vfsmount
1177 * Given a path, find the mountpoint of its source file system.
1178 * Replace @path with the path of the mountpoint in the parent mount.
1181 * Return 1 if we went up a level and 0 if we were already at the
1184 int follow_up(struct path
*path
)
1186 struct mount
*mnt
= real_mount(path
->mnt
);
1187 struct mount
*parent
;
1188 struct dentry
*mountpoint
;
1190 read_seqlock_excl(&mount_lock
);
1191 parent
= mnt
->mnt_parent
;
1192 if (parent
== mnt
) {
1193 read_sequnlock_excl(&mount_lock
);
1196 mntget(&parent
->mnt
);
1197 mountpoint
= dget(mnt
->mnt_mountpoint
);
1198 read_sequnlock_excl(&mount_lock
);
1200 path
->dentry
= mountpoint
;
1202 path
->mnt
= &parent
->mnt
;
1205 EXPORT_SYMBOL(follow_up
);
1207 static bool choose_mountpoint_rcu(struct mount
*m
, const struct path
*root
,
1208 struct path
*path
, unsigned *seqp
)
1210 while (mnt_has_parent(m
)) {
1211 struct dentry
*mountpoint
= m
->mnt_mountpoint
;
1214 if (unlikely(root
->dentry
== mountpoint
&&
1215 root
->mnt
== &m
->mnt
))
1217 if (mountpoint
!= m
->mnt
.mnt_root
) {
1218 path
->mnt
= &m
->mnt
;
1219 path
->dentry
= mountpoint
;
1220 *seqp
= read_seqcount_begin(&mountpoint
->d_seq
);
1227 static bool choose_mountpoint(struct mount
*m
, const struct path
*root
,
1234 unsigned seq
, mseq
= read_seqbegin(&mount_lock
);
1236 found
= choose_mountpoint_rcu(m
, root
, path
, &seq
);
1237 if (unlikely(!found
)) {
1238 if (!read_seqretry(&mount_lock
, mseq
))
1241 if (likely(__legitimize_path(path
, seq
, mseq
)))
1253 * Perform an automount
1254 * - return -EISDIR to tell follow_managed() to stop and return the path we
1257 static int follow_automount(struct path
*path
, int *count
, unsigned lookup_flags
)
1259 struct dentry
*dentry
= path
->dentry
;
1261 /* We don't want to mount if someone's just doing a stat -
1262 * unless they're stat'ing a directory and appended a '/' to
1265 * We do, however, want to mount if someone wants to open or
1266 * create a file of any type under the mountpoint, wants to
1267 * traverse through the mountpoint or wants to open the
1268 * mounted directory. Also, autofs may mark negative dentries
1269 * as being automount points. These will need the attentions
1270 * of the daemon to instantiate them before they can be used.
1272 if (!(lookup_flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1273 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1277 if (count
&& (*count
)++ >= MAXSYMLINKS
)
1280 return finish_automount(dentry
->d_op
->d_automount(path
), path
);
1284 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1285 * dentries are pinned but not locked here, so negative dentry can go
1286 * positive right under us. Use of smp_load_acquire() provides a barrier
1287 * sufficient for ->d_inode and ->d_flags consistency.
1289 static int __traverse_mounts(struct path
*path
, unsigned flags
, bool *jumped
,
1290 int *count
, unsigned lookup_flags
)
1292 struct vfsmount
*mnt
= path
->mnt
;
1293 bool need_mntput
= false;
1296 while (flags
& DCACHE_MANAGED_DENTRY
) {
1297 /* Allow the filesystem to manage the transit without i_mutex
1299 if (flags
& DCACHE_MANAGE_TRANSIT
) {
1300 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1301 flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1306 if (flags
& DCACHE_MOUNTED
) { // something's mounted on it..
1307 struct vfsmount
*mounted
= lookup_mnt(path
);
1308 if (mounted
) { // ... in our namespace
1312 path
->mnt
= mounted
;
1313 path
->dentry
= dget(mounted
->mnt_root
);
1314 // here we know it's positive
1315 flags
= path
->dentry
->d_flags
;
1321 if (!(flags
& DCACHE_NEED_AUTOMOUNT
))
1324 // uncovered automount point
1325 ret
= follow_automount(path
, count
, lookup_flags
);
1326 flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1333 // possible if you race with several mount --move
1334 if (need_mntput
&& path
->mnt
== mnt
)
1336 if (!ret
&& unlikely(d_flags_negative(flags
)))
1338 *jumped
= need_mntput
;
1342 static inline int traverse_mounts(struct path
*path
, bool *jumped
,
1343 int *count
, unsigned lookup_flags
)
1345 unsigned flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1348 if (likely(!(flags
& DCACHE_MANAGED_DENTRY
))) {
1350 if (unlikely(d_flags_negative(flags
)))
1354 return __traverse_mounts(path
, flags
, jumped
, count
, lookup_flags
);
1357 int follow_down_one(struct path
*path
)
1359 struct vfsmount
*mounted
;
1361 mounted
= lookup_mnt(path
);
1365 path
->mnt
= mounted
;
1366 path
->dentry
= dget(mounted
->mnt_root
);
1371 EXPORT_SYMBOL(follow_down_one
);
1374 * Follow down to the covering mount currently visible to userspace. At each
1375 * point, the filesystem owning that dentry may be queried as to whether the
1376 * caller is permitted to proceed or not.
1378 int follow_down(struct path
*path
)
1380 struct vfsmount
*mnt
= path
->mnt
;
1382 int ret
= traverse_mounts(path
, &jumped
, NULL
, 0);
1384 if (path
->mnt
!= mnt
)
1388 EXPORT_SYMBOL(follow_down
);
1391 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1392 * we meet a managed dentry that would need blocking.
1394 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1395 struct inode
**inode
, unsigned *seqp
)
1397 struct dentry
*dentry
= path
->dentry
;
1398 unsigned int flags
= dentry
->d_flags
;
1400 if (likely(!(flags
& DCACHE_MANAGED_DENTRY
)))
1403 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1408 * Don't forget we might have a non-mountpoint managed dentry
1409 * that wants to block transit.
1411 if (unlikely(flags
& DCACHE_MANAGE_TRANSIT
)) {
1412 int res
= dentry
->d_op
->d_manage(path
, true);
1414 return res
== -EISDIR
;
1415 flags
= dentry
->d_flags
;
1418 if (flags
& DCACHE_MOUNTED
) {
1419 struct mount
*mounted
= __lookup_mnt(path
->mnt
, dentry
);
1421 path
->mnt
= &mounted
->mnt
;
1422 dentry
= path
->dentry
= mounted
->mnt
.mnt_root
;
1423 nd
->flags
|= LOOKUP_JUMPED
;
1424 *seqp
= read_seqcount_begin(&dentry
->d_seq
);
1425 *inode
= dentry
->d_inode
;
1427 * We don't need to re-check ->d_seq after this
1428 * ->d_inode read - there will be an RCU delay
1429 * between mount hash removal and ->mnt_root
1430 * becoming unpinned.
1432 flags
= dentry
->d_flags
;
1435 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1438 return !(flags
& DCACHE_NEED_AUTOMOUNT
);
1442 static inline int handle_mounts(struct nameidata
*nd
, struct dentry
*dentry
,
1443 struct path
*path
, struct inode
**inode
,
1449 path
->mnt
= nd
->path
.mnt
;
1450 path
->dentry
= dentry
;
1451 if (nd
->flags
& LOOKUP_RCU
) {
1452 unsigned int seq
= *seqp
;
1453 if (unlikely(!*inode
))
1455 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1457 if (!try_to_unlazy_next(nd
, dentry
, seq
))
1459 // *path might've been clobbered by __follow_mount_rcu()
1460 path
->mnt
= nd
->path
.mnt
;
1461 path
->dentry
= dentry
;
1463 ret
= traverse_mounts(path
, &jumped
, &nd
->total_link_count
, nd
->flags
);
1465 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1468 nd
->flags
|= LOOKUP_JUMPED
;
1470 if (unlikely(ret
)) {
1472 if (path
->mnt
!= nd
->path
.mnt
)
1475 *inode
= d_backing_inode(path
->dentry
);
1476 *seqp
= 0; /* out of RCU mode, so the value doesn't matter */
1482 * This looks up the name in dcache and possibly revalidates the found dentry.
1483 * NULL is returned if the dentry does not exist in the cache.
1485 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1489 struct dentry
*dentry
= d_lookup(dir
, name
);
1491 int error
= d_revalidate(dentry
, flags
);
1492 if (unlikely(error
<= 0)) {
1494 d_invalidate(dentry
);
1496 return ERR_PTR(error
);
1503 * Parent directory has inode locked exclusive. This is one
1504 * and only case when ->lookup() gets called on non in-lookup
1505 * dentries - as the matter of fact, this only gets called
1506 * when directory is guaranteed to have no in-lookup children
1509 static struct dentry
*__lookup_hash(const struct qstr
*name
,
1510 struct dentry
*base
, unsigned int flags
)
1512 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1514 struct inode
*dir
= base
->d_inode
;
1519 /* Don't create child dentry for a dead directory. */
1520 if (unlikely(IS_DEADDIR(dir
)))
1521 return ERR_PTR(-ENOENT
);
1523 dentry
= d_alloc(base
, name
);
1524 if (unlikely(!dentry
))
1525 return ERR_PTR(-ENOMEM
);
1527 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1528 if (unlikely(old
)) {
1535 static struct dentry
*lookup_fast(struct nameidata
*nd
,
1536 struct inode
**inode
,
1539 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1543 * Rename seqlock is not required here because in the off chance
1544 * of a false negative due to a concurrent rename, the caller is
1545 * going to fall back to non-racy lookup.
1547 if (nd
->flags
& LOOKUP_RCU
) {
1549 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1550 if (unlikely(!dentry
)) {
1551 if (!try_to_unlazy(nd
))
1552 return ERR_PTR(-ECHILD
);
1557 * This sequence count validates that the inode matches
1558 * the dentry name information from lookup.
1560 *inode
= d_backing_inode(dentry
);
1561 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
1562 return ERR_PTR(-ECHILD
);
1565 * This sequence count validates that the parent had no
1566 * changes while we did the lookup of the dentry above.
1568 * The memory barrier in read_seqcount_begin of child is
1569 * enough, we can use __read_seqcount_retry here.
1571 if (unlikely(__read_seqcount_retry(&parent
->d_seq
, nd
->seq
)))
1572 return ERR_PTR(-ECHILD
);
1575 status
= d_revalidate(dentry
, nd
->flags
);
1576 if (likely(status
> 0))
1578 if (!try_to_unlazy_next(nd
, dentry
, seq
))
1579 return ERR_PTR(-ECHILD
);
1580 if (status
== -ECHILD
)
1581 /* we'd been told to redo it in non-rcu mode */
1582 status
= d_revalidate(dentry
, nd
->flags
);
1584 dentry
= __d_lookup(parent
, &nd
->last
);
1585 if (unlikely(!dentry
))
1587 status
= d_revalidate(dentry
, nd
->flags
);
1589 if (unlikely(status
<= 0)) {
1591 d_invalidate(dentry
);
1593 return ERR_PTR(status
);
1598 /* Fast lookup failed, do it the slow way */
1599 static struct dentry
*__lookup_slow(const struct qstr
*name
,
1603 struct dentry
*dentry
, *old
;
1604 struct inode
*inode
= dir
->d_inode
;
1605 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1607 /* Don't go there if it's already dead */
1608 if (unlikely(IS_DEADDIR(inode
)))
1609 return ERR_PTR(-ENOENT
);
1611 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1614 if (unlikely(!d_in_lookup(dentry
))) {
1615 int error
= d_revalidate(dentry
, flags
);
1616 if (unlikely(error
<= 0)) {
1618 d_invalidate(dentry
);
1623 dentry
= ERR_PTR(error
);
1626 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1627 d_lookup_done(dentry
);
1628 if (unlikely(old
)) {
1636 static struct dentry
*lookup_slow(const struct qstr
*name
,
1640 struct inode
*inode
= dir
->d_inode
;
1642 inode_lock_shared(inode
);
1643 res
= __lookup_slow(name
, dir
, flags
);
1644 inode_unlock_shared(inode
);
1648 static inline int may_lookup(struct user_namespace
*mnt_userns
,
1649 struct nameidata
*nd
)
1651 if (nd
->flags
& LOOKUP_RCU
) {
1652 int err
= inode_permission(mnt_userns
, nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1653 if (err
!= -ECHILD
|| !try_to_unlazy(nd
))
1656 return inode_permission(mnt_userns
, nd
->inode
, MAY_EXEC
);
1659 static int reserve_stack(struct nameidata
*nd
, struct path
*link
, unsigned seq
)
1661 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
))
1664 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
1666 if (likely(nd
->stack
!= nd
->internal
))
1668 if (likely(nd_alloc_stack(nd
)))
1671 if (nd
->flags
& LOOKUP_RCU
) {
1672 // we need to grab link before we do unlazy. And we can't skip
1673 // unlazy even if we fail to grab the link - cleanup needs it
1674 bool grabbed_link
= legitimize_path(nd
, link
, seq
);
1676 if (!try_to_unlazy(nd
) != 0 || !grabbed_link
)
1679 if (nd_alloc_stack(nd
))
1685 enum {WALK_TRAILING
= 1, WALK_MORE
= 2, WALK_NOFOLLOW
= 4};
1687 static const char *pick_link(struct nameidata
*nd
, struct path
*link
,
1688 struct inode
*inode
, unsigned seq
, int flags
)
1692 int error
= reserve_stack(nd
, link
, seq
);
1694 if (unlikely(error
)) {
1695 if (!(nd
->flags
& LOOKUP_RCU
))
1697 return ERR_PTR(error
);
1699 last
= nd
->stack
+ nd
->depth
++;
1701 clear_delayed_call(&last
->done
);
1704 if (flags
& WALK_TRAILING
) {
1705 error
= may_follow_link(nd
, inode
);
1706 if (unlikely(error
))
1707 return ERR_PTR(error
);
1710 if (unlikely(nd
->flags
& LOOKUP_NO_SYMLINKS
) ||
1711 unlikely(link
->mnt
->mnt_flags
& MNT_NOSYMFOLLOW
))
1712 return ERR_PTR(-ELOOP
);
1714 if (!(nd
->flags
& LOOKUP_RCU
)) {
1715 touch_atime(&last
->link
);
1717 } else if (atime_needs_update(&last
->link
, inode
)) {
1718 if (!try_to_unlazy(nd
))
1719 return ERR_PTR(-ECHILD
);
1720 touch_atime(&last
->link
);
1723 error
= security_inode_follow_link(link
->dentry
, inode
,
1724 nd
->flags
& LOOKUP_RCU
);
1725 if (unlikely(error
))
1726 return ERR_PTR(error
);
1728 res
= READ_ONCE(inode
->i_link
);
1730 const char * (*get
)(struct dentry
*, struct inode
*,
1731 struct delayed_call
*);
1732 get
= inode
->i_op
->get_link
;
1733 if (nd
->flags
& LOOKUP_RCU
) {
1734 res
= get(NULL
, inode
, &last
->done
);
1735 if (res
== ERR_PTR(-ECHILD
) && try_to_unlazy(nd
))
1736 res
= get(link
->dentry
, inode
, &last
->done
);
1738 res
= get(link
->dentry
, inode
, &last
->done
);
1746 error
= nd_jump_root(nd
);
1747 if (unlikely(error
))
1748 return ERR_PTR(error
);
1749 while (unlikely(*++res
== '/'))
1754 all_done
: // pure jump
1760 * Do we need to follow links? We _really_ want to be able
1761 * to do this check without having to look at inode->i_op,
1762 * so we keep a cache of "no, this doesn't need follow_link"
1763 * for the common case.
1765 static const char *step_into(struct nameidata
*nd
, int flags
,
1766 struct dentry
*dentry
, struct inode
*inode
, unsigned seq
)
1769 int err
= handle_mounts(nd
, dentry
, &path
, &inode
, &seq
);
1772 return ERR_PTR(err
);
1773 if (likely(!d_is_symlink(path
.dentry
)) ||
1774 ((flags
& WALK_TRAILING
) && !(nd
->flags
& LOOKUP_FOLLOW
)) ||
1775 (flags
& WALK_NOFOLLOW
)) {
1776 /* not a symlink or should not follow */
1777 if (!(nd
->flags
& LOOKUP_RCU
)) {
1778 dput(nd
->path
.dentry
);
1779 if (nd
->path
.mnt
!= path
.mnt
)
1780 mntput(nd
->path
.mnt
);
1787 if (nd
->flags
& LOOKUP_RCU
) {
1788 /* make sure that d_is_symlink above matches inode */
1789 if (read_seqcount_retry(&path
.dentry
->d_seq
, seq
))
1790 return ERR_PTR(-ECHILD
);
1792 if (path
.mnt
== nd
->path
.mnt
)
1795 return pick_link(nd
, &path
, inode
, seq
, flags
);
1798 static struct dentry
*follow_dotdot_rcu(struct nameidata
*nd
,
1799 struct inode
**inodep
,
1802 struct dentry
*parent
, *old
;
1804 if (path_equal(&nd
->path
, &nd
->root
))
1806 if (unlikely(nd
->path
.dentry
== nd
->path
.mnt
->mnt_root
)) {
1809 if (!choose_mountpoint_rcu(real_mount(nd
->path
.mnt
),
1810 &nd
->root
, &path
, &seq
))
1812 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1813 return ERR_PTR(-ECHILD
);
1815 nd
->inode
= path
.dentry
->d_inode
;
1817 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1818 return ERR_PTR(-ECHILD
);
1819 /* we know that mountpoint was pinned */
1821 old
= nd
->path
.dentry
;
1822 parent
= old
->d_parent
;
1823 *inodep
= parent
->d_inode
;
1824 *seqp
= read_seqcount_begin(&parent
->d_seq
);
1825 if (unlikely(read_seqcount_retry(&old
->d_seq
, nd
->seq
)))
1826 return ERR_PTR(-ECHILD
);
1827 if (unlikely(!path_connected(nd
->path
.mnt
, parent
)))
1828 return ERR_PTR(-ECHILD
);
1831 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1832 return ERR_PTR(-ECHILD
);
1833 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
1834 return ERR_PTR(-ECHILD
);
1838 static struct dentry
*follow_dotdot(struct nameidata
*nd
,
1839 struct inode
**inodep
,
1842 struct dentry
*parent
;
1844 if (path_equal(&nd
->path
, &nd
->root
))
1846 if (unlikely(nd
->path
.dentry
== nd
->path
.mnt
->mnt_root
)) {
1849 if (!choose_mountpoint(real_mount(nd
->path
.mnt
),
1852 path_put(&nd
->path
);
1854 nd
->inode
= path
.dentry
->d_inode
;
1855 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1856 return ERR_PTR(-EXDEV
);
1858 /* rare case of legitimate dget_parent()... */
1859 parent
= dget_parent(nd
->path
.dentry
);
1860 if (unlikely(!path_connected(nd
->path
.mnt
, parent
))) {
1862 return ERR_PTR(-ENOENT
);
1865 *inodep
= parent
->d_inode
;
1869 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
1870 return ERR_PTR(-EXDEV
);
1871 dget(nd
->path
.dentry
);
1875 static const char *handle_dots(struct nameidata
*nd
, int type
)
1877 if (type
== LAST_DOTDOT
) {
1878 const char *error
= NULL
;
1879 struct dentry
*parent
;
1880 struct inode
*inode
;
1883 if (!nd
->root
.mnt
) {
1884 error
= ERR_PTR(set_root(nd
));
1888 if (nd
->flags
& LOOKUP_RCU
)
1889 parent
= follow_dotdot_rcu(nd
, &inode
, &seq
);
1891 parent
= follow_dotdot(nd
, &inode
, &seq
);
1893 return ERR_CAST(parent
);
1894 if (unlikely(!parent
))
1895 error
= step_into(nd
, WALK_NOFOLLOW
,
1896 nd
->path
.dentry
, nd
->inode
, nd
->seq
);
1898 error
= step_into(nd
, WALK_NOFOLLOW
,
1899 parent
, inode
, seq
);
1900 if (unlikely(error
))
1903 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
)) {
1905 * If there was a racing rename or mount along our
1906 * path, then we can't be sure that ".." hasn't jumped
1907 * above nd->root (and so userspace should retry or use
1911 if (unlikely(__read_seqcount_retry(&mount_lock
.seqcount
, nd
->m_seq
)))
1912 return ERR_PTR(-EAGAIN
);
1913 if (unlikely(__read_seqcount_retry(&rename_lock
.seqcount
, nd
->r_seq
)))
1914 return ERR_PTR(-EAGAIN
);
1920 static const char *walk_component(struct nameidata
*nd
, int flags
)
1922 struct dentry
*dentry
;
1923 struct inode
*inode
;
1926 * "." and ".." are special - ".." especially so because it has
1927 * to be able to know about the current root directory and
1928 * parent relationships.
1930 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1931 if (!(flags
& WALK_MORE
) && nd
->depth
)
1933 return handle_dots(nd
, nd
->last_type
);
1935 dentry
= lookup_fast(nd
, &inode
, &seq
);
1937 return ERR_CAST(dentry
);
1938 if (unlikely(!dentry
)) {
1939 dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
1941 return ERR_CAST(dentry
);
1943 if (!(flags
& WALK_MORE
) && nd
->depth
)
1945 return step_into(nd
, flags
, dentry
, inode
, seq
);
1949 * We can do the critical dentry name comparison and hashing
1950 * operations one word at a time, but we are limited to:
1952 * - Architectures with fast unaligned word accesses. We could
1953 * do a "get_unaligned()" if this helps and is sufficiently
1956 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1957 * do not trap on the (extremely unlikely) case of a page
1958 * crossing operation.
1960 * - Furthermore, we need an efficient 64-bit compile for the
1961 * 64-bit case in order to generate the "number of bytes in
1962 * the final mask". Again, that could be replaced with a
1963 * efficient population count instruction or similar.
1965 #ifdef CONFIG_DCACHE_WORD_ACCESS
1967 #include <asm/word-at-a-time.h>
1971 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1973 #elif defined(CONFIG_64BIT)
1975 * Register pressure in the mixing function is an issue, particularly
1976 * on 32-bit x86, but almost any function requires one state value and
1977 * one temporary. Instead, use a function designed for two state values
1978 * and no temporaries.
1980 * This function cannot create a collision in only two iterations, so
1981 * we have two iterations to achieve avalanche. In those two iterations,
1982 * we have six layers of mixing, which is enough to spread one bit's
1983 * influence out to 2^6 = 64 state bits.
1985 * Rotate constants are scored by considering either 64 one-bit input
1986 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1987 * probability of that delta causing a change to each of the 128 output
1988 * bits, using a sample of random initial states.
1990 * The Shannon entropy of the computed probabilities is then summed
1991 * to produce a score. Ideally, any input change has a 50% chance of
1992 * toggling any given output bit.
1994 * Mixing scores (in bits) for (12,45):
1995 * Input delta: 1-bit 2-bit
1996 * 1 round: 713.3 42542.6
1997 * 2 rounds: 2753.7 140389.8
1998 * 3 rounds: 5954.1 233458.2
1999 * 4 rounds: 7862.6 256672.2
2000 * Perfect: 8192 258048
2001 * (64*128) (64*63/2 * 128)
2003 #define HASH_MIX(x, y, a) \
2005 y ^= x, x = rol64(x,12),\
2006 x += y, y = rol64(y,45),\
2010 * Fold two longs into one 32-bit hash value. This must be fast, but
2011 * latency isn't quite as critical, as there is a fair bit of additional
2012 * work done before the hash value is used.
2014 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
2016 y
^= x
* GOLDEN_RATIO_64
;
2017 y
*= GOLDEN_RATIO_64
;
2021 #else /* 32-bit case */
2024 * Mixing scores (in bits) for (7,20):
2025 * Input delta: 1-bit 2-bit
2026 * 1 round: 330.3 9201.6
2027 * 2 rounds: 1246.4 25475.4
2028 * 3 rounds: 1907.1 31295.1
2029 * 4 rounds: 2042.3 31718.6
2030 * Perfect: 2048 31744
2031 * (32*64) (32*31/2 * 64)
2033 #define HASH_MIX(x, y, a) \
2035 y ^= x, x = rol32(x, 7),\
2036 x += y, y = rol32(y,20),\
2039 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
2041 /* Use arch-optimized multiply if one exists */
2042 return __hash_32(y
^ __hash_32(x
));
2048 * Return the hash of a string of known length. This is carfully
2049 * designed to match hash_name(), which is the more critical function.
2050 * In particular, we must end by hashing a final word containing 0..7
2051 * payload bytes, to match the way that hash_name() iterates until it
2052 * finds the delimiter after the name.
2054 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
2056 unsigned long a
, x
= 0, y
= (unsigned long)salt
;
2061 a
= load_unaligned_zeropad(name
);
2062 if (len
< sizeof(unsigned long))
2065 name
+= sizeof(unsigned long);
2066 len
-= sizeof(unsigned long);
2068 x
^= a
& bytemask_from_count(len
);
2070 return fold_hash(x
, y
);
2072 EXPORT_SYMBOL(full_name_hash
);
2074 /* Return the "hash_len" (hash and length) of a null-terminated string */
2075 u64
hashlen_string(const void *salt
, const char *name
)
2077 unsigned long a
= 0, x
= 0, y
= (unsigned long)salt
;
2078 unsigned long adata
, mask
, len
;
2079 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
2086 len
+= sizeof(unsigned long);
2088 a
= load_unaligned_zeropad(name
+len
);
2089 } while (!has_zero(a
, &adata
, &constants
));
2091 adata
= prep_zero_mask(a
, adata
, &constants
);
2092 mask
= create_zero_mask(adata
);
2093 x
^= a
& zero_bytemask(mask
);
2095 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2097 EXPORT_SYMBOL(hashlen_string
);
2100 * Calculate the length and hash of the path component, and
2101 * return the "hash_len" as the result.
2103 static inline u64
hash_name(const void *salt
, const char *name
)
2105 unsigned long a
= 0, b
, x
= 0, y
= (unsigned long)salt
;
2106 unsigned long adata
, bdata
, mask
, len
;
2107 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
2114 len
+= sizeof(unsigned long);
2116 a
= load_unaligned_zeropad(name
+len
);
2117 b
= a
^ REPEAT_BYTE('/');
2118 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
2120 adata
= prep_zero_mask(a
, adata
, &constants
);
2121 bdata
= prep_zero_mask(b
, bdata
, &constants
);
2122 mask
= create_zero_mask(adata
| bdata
);
2123 x
^= a
& zero_bytemask(mask
);
2125 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2128 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2130 /* Return the hash of a string of known length */
2131 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
2133 unsigned long hash
= init_name_hash(salt
);
2135 hash
= partial_name_hash((unsigned char)*name
++, hash
);
2136 return end_name_hash(hash
);
2138 EXPORT_SYMBOL(full_name_hash
);
2140 /* Return the "hash_len" (hash and length) of a null-terminated string */
2141 u64
hashlen_string(const void *salt
, const char *name
)
2143 unsigned long hash
= init_name_hash(salt
);
2144 unsigned long len
= 0, c
;
2146 c
= (unsigned char)*name
;
2149 hash
= partial_name_hash(c
, hash
);
2150 c
= (unsigned char)name
[len
];
2152 return hashlen_create(end_name_hash(hash
), len
);
2154 EXPORT_SYMBOL(hashlen_string
);
2157 * We know there's a real path component here of at least
2160 static inline u64
hash_name(const void *salt
, const char *name
)
2162 unsigned long hash
= init_name_hash(salt
);
2163 unsigned long len
= 0, c
;
2165 c
= (unsigned char)*name
;
2168 hash
= partial_name_hash(c
, hash
);
2169 c
= (unsigned char)name
[len
];
2170 } while (c
&& c
!= '/');
2171 return hashlen_create(end_name_hash(hash
), len
);
2178 * This is the basic name resolution function, turning a pathname into
2179 * the final dentry. We expect 'base' to be positive and a directory.
2181 * Returns 0 and nd will have valid dentry and mnt on success.
2182 * Returns error and drops reference to input namei data on failure.
2184 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2186 int depth
= 0; // depth <= nd->depth
2189 nd
->last_type
= LAST_ROOT
;
2190 nd
->flags
|= LOOKUP_PARENT
;
2192 return PTR_ERR(name
);
2196 nd
->dir_mode
= 0; // short-circuit the 'hardening' idiocy
2200 /* At this point we know we have a real path component. */
2202 struct user_namespace
*mnt_userns
;
2207 mnt_userns
= mnt_user_ns(nd
->path
.mnt
);
2208 err
= may_lookup(mnt_userns
, nd
);
2212 hash_len
= hash_name(nd
->path
.dentry
, name
);
2215 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2217 if (name
[1] == '.') {
2219 nd
->flags
|= LOOKUP_JUMPED
;
2225 if (likely(type
== LAST_NORM
)) {
2226 struct dentry
*parent
= nd
->path
.dentry
;
2227 nd
->flags
&= ~LOOKUP_JUMPED
;
2228 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2229 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2230 err
= parent
->d_op
->d_hash(parent
, &this);
2233 hash_len
= this.hash_len
;
2238 nd
->last
.hash_len
= hash_len
;
2239 nd
->last
.name
= name
;
2240 nd
->last_type
= type
;
2242 name
+= hashlen_len(hash_len
);
2246 * If it wasn't NUL, we know it was '/'. Skip that
2247 * slash, and continue until no more slashes.
2251 } while (unlikely(*name
== '/'));
2252 if (unlikely(!*name
)) {
2254 /* pathname or trailing symlink, done */
2256 nd
->dir_uid
= i_uid_into_mnt(mnt_userns
, nd
->inode
);
2257 nd
->dir_mode
= nd
->inode
->i_mode
;
2258 nd
->flags
&= ~LOOKUP_PARENT
;
2261 /* last component of nested symlink */
2262 name
= nd
->stack
[--depth
].name
;
2263 link
= walk_component(nd
, 0);
2265 /* not the last component */
2266 link
= walk_component(nd
, WALK_MORE
);
2268 if (unlikely(link
)) {
2270 return PTR_ERR(link
);
2271 /* a symlink to follow */
2272 nd
->stack
[depth
++].name
= name
;
2276 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2277 if (nd
->flags
& LOOKUP_RCU
) {
2278 if (!try_to_unlazy(nd
))
2286 /* must be paired with terminate_walk() */
2287 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2290 const char *s
= nd
->name
->name
;
2292 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2293 if ((flags
& (LOOKUP_RCU
| LOOKUP_CACHED
)) == LOOKUP_CACHED
)
2294 return ERR_PTR(-EAGAIN
);
2297 flags
&= ~LOOKUP_RCU
;
2298 if (flags
& LOOKUP_RCU
)
2301 nd
->flags
= flags
| LOOKUP_JUMPED
;
2304 nd
->m_seq
= __read_seqcount_begin(&mount_lock
.seqcount
);
2305 nd
->r_seq
= __read_seqcount_begin(&rename_lock
.seqcount
);
2308 if (flags
& LOOKUP_ROOT
) {
2309 struct dentry
*root
= nd
->root
.dentry
;
2310 struct inode
*inode
= root
->d_inode
;
2311 if (*s
&& unlikely(!d_can_lookup(root
)))
2312 return ERR_PTR(-ENOTDIR
);
2313 nd
->path
= nd
->root
;
2315 if (flags
& LOOKUP_RCU
) {
2316 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2317 nd
->root_seq
= nd
->seq
;
2319 path_get(&nd
->path
);
2324 nd
->root
.mnt
= NULL
;
2325 nd
->path
.mnt
= NULL
;
2326 nd
->path
.dentry
= NULL
;
2328 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2329 if (*s
== '/' && !(flags
& LOOKUP_IN_ROOT
)) {
2330 error
= nd_jump_root(nd
);
2331 if (unlikely(error
))
2332 return ERR_PTR(error
);
2336 /* Relative pathname -- get the starting-point it is relative to. */
2337 if (nd
->dfd
== AT_FDCWD
) {
2338 if (flags
& LOOKUP_RCU
) {
2339 struct fs_struct
*fs
= current
->fs
;
2343 seq
= read_seqcount_begin(&fs
->seq
);
2345 nd
->inode
= nd
->path
.dentry
->d_inode
;
2346 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2347 } while (read_seqcount_retry(&fs
->seq
, seq
));
2349 get_fs_pwd(current
->fs
, &nd
->path
);
2350 nd
->inode
= nd
->path
.dentry
->d_inode
;
2353 /* Caller must check execute permissions on the starting path component */
2354 struct fd f
= fdget_raw(nd
->dfd
);
2355 struct dentry
*dentry
;
2358 return ERR_PTR(-EBADF
);
2360 dentry
= f
.file
->f_path
.dentry
;
2362 if (*s
&& unlikely(!d_can_lookup(dentry
))) {
2364 return ERR_PTR(-ENOTDIR
);
2367 nd
->path
= f
.file
->f_path
;
2368 if (flags
& LOOKUP_RCU
) {
2369 nd
->inode
= nd
->path
.dentry
->d_inode
;
2370 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2372 path_get(&nd
->path
);
2373 nd
->inode
= nd
->path
.dentry
->d_inode
;
2378 /* For scoped-lookups we need to set the root to the dirfd as well. */
2379 if (flags
& LOOKUP_IS_SCOPED
) {
2380 nd
->root
= nd
->path
;
2381 if (flags
& LOOKUP_RCU
) {
2382 nd
->root_seq
= nd
->seq
;
2384 path_get(&nd
->root
);
2385 nd
->flags
|= LOOKUP_ROOT_GRABBED
;
2391 static inline const char *lookup_last(struct nameidata
*nd
)
2393 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2394 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2396 return walk_component(nd
, WALK_TRAILING
);
2399 static int handle_lookup_down(struct nameidata
*nd
)
2401 if (!(nd
->flags
& LOOKUP_RCU
))
2402 dget(nd
->path
.dentry
);
2403 return PTR_ERR(step_into(nd
, WALK_NOFOLLOW
,
2404 nd
->path
.dentry
, nd
->inode
, nd
->seq
));
2407 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2408 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2410 const char *s
= path_init(nd
, flags
);
2413 if (unlikely(flags
& LOOKUP_DOWN
) && !IS_ERR(s
)) {
2414 err
= handle_lookup_down(nd
);
2415 if (unlikely(err
< 0))
2419 while (!(err
= link_path_walk(s
, nd
)) &&
2420 (s
= lookup_last(nd
)) != NULL
)
2423 err
= complete_walk(nd
);
2425 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2426 if (!d_can_lookup(nd
->path
.dentry
))
2428 if (!err
&& unlikely(nd
->flags
& LOOKUP_MOUNTPOINT
)) {
2429 err
= handle_lookup_down(nd
);
2430 nd
->flags
&= ~LOOKUP_JUMPED
; // no d_weak_revalidate(), please...
2434 nd
->path
.mnt
= NULL
;
2435 nd
->path
.dentry
= NULL
;
2441 int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2442 struct path
*path
, struct path
*root
)
2445 struct nameidata nd
;
2447 return PTR_ERR(name
);
2448 if (unlikely(root
)) {
2450 flags
|= LOOKUP_ROOT
;
2452 set_nameidata(&nd
, dfd
, name
);
2453 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2454 if (unlikely(retval
== -ECHILD
))
2455 retval
= path_lookupat(&nd
, flags
, path
);
2456 if (unlikely(retval
== -ESTALE
))
2457 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2459 if (likely(!retval
))
2460 audit_inode(name
, path
->dentry
,
2461 flags
& LOOKUP_MOUNTPOINT
? AUDIT_INODE_NOEVAL
: 0);
2462 restore_nameidata();
2467 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2468 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2469 struct path
*parent
)
2471 const char *s
= path_init(nd
, flags
);
2472 int err
= link_path_walk(s
, nd
);
2474 err
= complete_walk(nd
);
2477 nd
->path
.mnt
= NULL
;
2478 nd
->path
.dentry
= NULL
;
2484 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2485 unsigned int flags
, struct path
*parent
,
2486 struct qstr
*last
, int *type
)
2489 struct nameidata nd
;
2493 set_nameidata(&nd
, dfd
, name
);
2494 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2495 if (unlikely(retval
== -ECHILD
))
2496 retval
= path_parentat(&nd
, flags
, parent
);
2497 if (unlikely(retval
== -ESTALE
))
2498 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2499 if (likely(!retval
)) {
2501 *type
= nd
.last_type
;
2502 audit_inode(name
, parent
->dentry
, AUDIT_INODE_PARENT
);
2505 name
= ERR_PTR(retval
);
2507 restore_nameidata();
2511 /* does lookup, returns the object with parent locked */
2512 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2514 struct filename
*filename
;
2519 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2521 if (IS_ERR(filename
))
2522 return ERR_CAST(filename
);
2523 if (unlikely(type
!= LAST_NORM
)) {
2526 return ERR_PTR(-EINVAL
);
2528 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2529 d
= __lookup_hash(&last
, path
->dentry
, 0);
2531 inode_unlock(path
->dentry
->d_inode
);
2538 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2540 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2543 EXPORT_SYMBOL(kern_path
);
2546 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2547 * @dentry: pointer to dentry of the base directory
2548 * @mnt: pointer to vfs mount of the base directory
2549 * @name: pointer to file name
2550 * @flags: lookup flags
2551 * @path: pointer to struct path to fill
2553 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2554 const char *name
, unsigned int flags
,
2557 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2558 /* the first argument of filename_lookup() is ignored with root */
2559 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2560 flags
, path
, &root
);
2562 EXPORT_SYMBOL(vfs_path_lookup
);
2564 static int lookup_one_len_common(const char *name
, struct dentry
*base
,
2565 int len
, struct qstr
*this)
2569 this->hash
= full_name_hash(base
, name
, len
);
2573 if (unlikely(name
[0] == '.')) {
2574 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2579 unsigned int c
= *(const unsigned char *)name
++;
2580 if (c
== '/' || c
== '\0')
2584 * See if the low-level filesystem might want
2585 * to use its own hash..
2587 if (base
->d_flags
& DCACHE_OP_HASH
) {
2588 int err
= base
->d_op
->d_hash(base
, this);
2593 return inode_permission(&init_user_ns
, base
->d_inode
, MAY_EXEC
);
2597 * try_lookup_one_len - filesystem helper to lookup single pathname component
2598 * @name: pathname component to lookup
2599 * @base: base directory to lookup from
2600 * @len: maximum length @len should be interpreted to
2602 * Look up a dentry by name in the dcache, returning NULL if it does not
2603 * currently exist. The function does not try to create a dentry.
2605 * Note that this routine is purely a helper for filesystem usage and should
2606 * not be called by generic code.
2608 * The caller must hold base->i_mutex.
2610 struct dentry
*try_lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2615 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2617 err
= lookup_one_len_common(name
, base
, len
, &this);
2619 return ERR_PTR(err
);
2621 return lookup_dcache(&this, base
, 0);
2623 EXPORT_SYMBOL(try_lookup_one_len
);
2626 * lookup_one_len - filesystem helper to lookup single pathname component
2627 * @name: pathname component to lookup
2628 * @base: base directory to lookup from
2629 * @len: maximum length @len should be interpreted to
2631 * Note that this routine is purely a helper for filesystem usage and should
2632 * not be called by generic code.
2634 * The caller must hold base->i_mutex.
2636 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2638 struct dentry
*dentry
;
2642 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2644 err
= lookup_one_len_common(name
, base
, len
, &this);
2646 return ERR_PTR(err
);
2648 dentry
= lookup_dcache(&this, base
, 0);
2649 return dentry
? dentry
: __lookup_slow(&this, base
, 0);
2651 EXPORT_SYMBOL(lookup_one_len
);
2654 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2655 * @name: pathname component to lookup
2656 * @base: base directory to lookup from
2657 * @len: maximum length @len should be interpreted to
2659 * Note that this routine is purely a helper for filesystem usage and should
2660 * not be called by generic code.
2662 * Unlike lookup_one_len, it should be called without the parent
2663 * i_mutex held, and will take the i_mutex itself if necessary.
2665 struct dentry
*lookup_one_len_unlocked(const char *name
,
2666 struct dentry
*base
, int len
)
2672 err
= lookup_one_len_common(name
, base
, len
, &this);
2674 return ERR_PTR(err
);
2676 ret
= lookup_dcache(&this, base
, 0);
2678 ret
= lookup_slow(&this, base
, 0);
2681 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2684 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2685 * on negatives. Returns known positive or ERR_PTR(); that's what
2686 * most of the users want. Note that pinned negative with unlocked parent
2687 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2688 * need to be very careful; pinned positives have ->d_inode stable, so
2689 * this one avoids such problems.
2691 struct dentry
*lookup_positive_unlocked(const char *name
,
2692 struct dentry
*base
, int len
)
2694 struct dentry
*ret
= lookup_one_len_unlocked(name
, base
, len
);
2695 if (!IS_ERR(ret
) && d_flags_negative(smp_load_acquire(&ret
->d_flags
))) {
2697 ret
= ERR_PTR(-ENOENT
);
2701 EXPORT_SYMBOL(lookup_positive_unlocked
);
2703 #ifdef CONFIG_UNIX98_PTYS
2704 int path_pts(struct path
*path
)
2706 /* Find something mounted on "pts" in the same directory as
2709 struct dentry
*parent
= dget_parent(path
->dentry
);
2710 struct dentry
*child
;
2711 struct qstr
this = QSTR_INIT("pts", 3);
2713 if (unlikely(!path_connected(path
->mnt
, parent
))) {
2718 path
->dentry
= parent
;
2719 child
= d_hash_and_lookup(parent
, &this);
2723 path
->dentry
= child
;
2730 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2731 struct path
*path
, int *empty
)
2733 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2736 EXPORT_SYMBOL(user_path_at_empty
);
2738 int __check_sticky(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2739 struct inode
*inode
)
2741 kuid_t fsuid
= current_fsuid();
2743 if (uid_eq(i_uid_into_mnt(mnt_userns
, inode
), fsuid
))
2745 if (uid_eq(i_uid_into_mnt(mnt_userns
, dir
), fsuid
))
2747 return !capable_wrt_inode_uidgid(mnt_userns
, inode
, CAP_FOWNER
);
2749 EXPORT_SYMBOL(__check_sticky
);
2752 * Check whether we can remove a link victim from directory dir, check
2753 * whether the type of victim is right.
2754 * 1. We can't do it if dir is read-only (done in permission())
2755 * 2. We should have write and exec permissions on dir
2756 * 3. We can't remove anything from append-only dir
2757 * 4. We can't do anything with immutable dir (done in permission())
2758 * 5. If the sticky bit on dir is set we should either
2759 * a. be owner of dir, or
2760 * b. be owner of victim, or
2761 * c. have CAP_FOWNER capability
2762 * 6. If the victim is append-only or immutable we can't do antyhing with
2763 * links pointing to it.
2764 * 7. If the victim has an unknown uid or gid we can't change the inode.
2765 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2766 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2767 * 10. We can't remove a root or mountpoint.
2768 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2769 * nfs_async_unlink().
2771 static int may_delete(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2772 struct dentry
*victim
, bool isdir
)
2774 struct inode
*inode
= d_backing_inode(victim
);
2777 if (d_is_negative(victim
))
2781 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2783 /* Inode writeback is not safe when the uid or gid are invalid. */
2784 if (!uid_valid(i_uid_into_mnt(mnt_userns
, inode
)) ||
2785 !gid_valid(i_gid_into_mnt(mnt_userns
, inode
)))
2788 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2790 error
= inode_permission(mnt_userns
, dir
, MAY_WRITE
| MAY_EXEC
);
2796 if (check_sticky(mnt_userns
, dir
, inode
) || IS_APPEND(inode
) ||
2797 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
) ||
2798 HAS_UNMAPPED_ID(mnt_userns
, inode
))
2801 if (!d_is_dir(victim
))
2803 if (IS_ROOT(victim
))
2805 } else if (d_is_dir(victim
))
2807 if (IS_DEADDIR(dir
))
2809 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2814 /* Check whether we can create an object with dentry child in directory
2816 * 1. We can't do it if child already exists (open has special treatment for
2817 * this case, but since we are inlined it's OK)
2818 * 2. We can't do it if dir is read-only (done in permission())
2819 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2820 * 4. We should have write and exec permissions on dir
2821 * 5. We can't do it if dir is immutable (done in permission())
2823 static inline int may_create(struct user_namespace
*mnt_userns
,
2824 struct inode
*dir
, struct dentry
*child
)
2826 struct user_namespace
*s_user_ns
;
2827 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2830 if (IS_DEADDIR(dir
))
2832 s_user_ns
= dir
->i_sb
->s_user_ns
;
2833 if (!kuid_has_mapping(s_user_ns
, fsuid_into_mnt(mnt_userns
)) ||
2834 !kgid_has_mapping(s_user_ns
, fsgid_into_mnt(mnt_userns
)))
2836 return inode_permission(mnt_userns
, dir
, MAY_WRITE
| MAY_EXEC
);
2840 * p1 and p2 should be directories on the same fs.
2842 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2847 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2851 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
2853 p
= d_ancestor(p2
, p1
);
2855 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2856 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2860 p
= d_ancestor(p1
, p2
);
2862 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2863 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2867 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2868 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2871 EXPORT_SYMBOL(lock_rename
);
2873 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2875 inode_unlock(p1
->d_inode
);
2877 inode_unlock(p2
->d_inode
);
2878 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
2881 EXPORT_SYMBOL(unlock_rename
);
2884 * vfs_create - create new file
2885 * @mnt_userns: user namespace of the mount the inode was found from
2886 * @dir: inode of @dentry
2887 * @dentry: pointer to dentry of the base directory
2888 * @mode: mode of the new file
2889 * @want_excl: whether the file must not yet exist
2891 * Create a new file.
2893 * If the inode has been found through an idmapped mount the user namespace of
2894 * the vfsmount must be passed through @mnt_userns. This function will then take
2895 * care to map the inode according to @mnt_userns before checking permissions.
2896 * On non-idmapped mounts or if permission checking is to be performed on the
2897 * raw inode simply passs init_user_ns.
2899 int vfs_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2900 struct dentry
*dentry
, umode_t mode
, bool want_excl
)
2902 int error
= may_create(mnt_userns
, dir
, dentry
);
2906 if (!dir
->i_op
->create
)
2907 return -EACCES
; /* shouldn't it be ENOSYS? */
2910 error
= security_inode_create(dir
, dentry
, mode
);
2913 error
= dir
->i_op
->create(mnt_userns
, dir
, dentry
, mode
, want_excl
);
2915 fsnotify_create(dir
, dentry
);
2918 EXPORT_SYMBOL(vfs_create
);
2920 int vfs_mkobj(struct dentry
*dentry
, umode_t mode
,
2921 int (*f
)(struct dentry
*, umode_t
, void *),
2924 struct inode
*dir
= dentry
->d_parent
->d_inode
;
2925 int error
= may_create(&init_user_ns
, dir
, dentry
);
2931 error
= security_inode_create(dir
, dentry
, mode
);
2934 error
= f(dentry
, mode
, arg
);
2936 fsnotify_create(dir
, dentry
);
2939 EXPORT_SYMBOL(vfs_mkobj
);
2941 bool may_open_dev(const struct path
*path
)
2943 return !(path
->mnt
->mnt_flags
& MNT_NODEV
) &&
2944 !(path
->mnt
->mnt_sb
->s_iflags
& SB_I_NODEV
);
2947 static int may_open(struct user_namespace
*mnt_userns
, const struct path
*path
,
2948 int acc_mode
, int flag
)
2950 struct dentry
*dentry
= path
->dentry
;
2951 struct inode
*inode
= dentry
->d_inode
;
2957 switch (inode
->i_mode
& S_IFMT
) {
2961 if (acc_mode
& MAY_WRITE
)
2963 if (acc_mode
& MAY_EXEC
)
2968 if (!may_open_dev(path
))
2973 if (acc_mode
& MAY_EXEC
)
2978 if ((acc_mode
& MAY_EXEC
) && path_noexec(path
))
2983 error
= inode_permission(mnt_userns
, inode
, MAY_OPEN
| acc_mode
);
2988 * An append-only file must be opened in append mode for writing.
2990 if (IS_APPEND(inode
)) {
2991 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2997 /* O_NOATIME can only be set by the owner or superuser */
2998 if (flag
& O_NOATIME
&& !inode_owner_or_capable(mnt_userns
, inode
))
3004 static int handle_truncate(struct user_namespace
*mnt_userns
, struct file
*filp
)
3006 const struct path
*path
= &filp
->f_path
;
3007 struct inode
*inode
= path
->dentry
->d_inode
;
3008 int error
= get_write_access(inode
);
3012 * Refuse to truncate files with mandatory locks held on them.
3014 error
= locks_verify_locked(filp
);
3016 error
= security_path_truncate(path
);
3018 error
= do_truncate(mnt_userns
, path
->dentry
, 0,
3019 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
3022 put_write_access(inode
);
3026 static inline int open_to_namei_flags(int flag
)
3028 if ((flag
& O_ACCMODE
) == 3)
3033 static int may_o_create(struct user_namespace
*mnt_userns
,
3034 const struct path
*dir
, struct dentry
*dentry
,
3037 struct user_namespace
*s_user_ns
;
3038 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
3042 s_user_ns
= dir
->dentry
->d_sb
->s_user_ns
;
3043 if (!kuid_has_mapping(s_user_ns
, fsuid_into_mnt(mnt_userns
)) ||
3044 !kgid_has_mapping(s_user_ns
, fsgid_into_mnt(mnt_userns
)))
3047 error
= inode_permission(mnt_userns
, dir
->dentry
->d_inode
,
3048 MAY_WRITE
| MAY_EXEC
);
3052 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
3056 * Attempt to atomically look up, create and open a file from a negative
3059 * Returns 0 if successful. The file will have been created and attached to
3060 * @file by the filesystem calling finish_open().
3062 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3063 * be set. The caller will need to perform the open themselves. @path will
3064 * have been updated to point to the new dentry. This may be negative.
3066 * Returns an error code otherwise.
3068 static struct dentry
*atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
3070 int open_flag
, umode_t mode
)
3072 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
3073 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
3076 if (nd
->flags
& LOOKUP_DIRECTORY
)
3077 open_flag
|= O_DIRECTORY
;
3079 file
->f_path
.dentry
= DENTRY_NOT_SET
;
3080 file
->f_path
.mnt
= nd
->path
.mnt
;
3081 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
3082 open_to_namei_flags(open_flag
), mode
);
3083 d_lookup_done(dentry
);
3085 if (file
->f_mode
& FMODE_OPENED
) {
3086 if (unlikely(dentry
!= file
->f_path
.dentry
)) {
3088 dentry
= dget(file
->f_path
.dentry
);
3090 } else if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
3093 if (file
->f_path
.dentry
) {
3095 dentry
= file
->f_path
.dentry
;
3097 if (unlikely(d_is_negative(dentry
)))
3103 dentry
= ERR_PTR(error
);
3109 * Look up and maybe create and open the last component.
3111 * Must be called with parent locked (exclusive in O_CREAT case).
3113 * Returns 0 on success, that is, if
3114 * the file was successfully atomically created (if necessary) and opened, or
3115 * the file was not completely opened at this time, though lookups and
3116 * creations were performed.
3117 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3118 * In the latter case dentry returned in @path might be negative if O_CREAT
3119 * hadn't been specified.
3121 * An error code is returned on failure.
3123 static struct dentry
*lookup_open(struct nameidata
*nd
, struct file
*file
,
3124 const struct open_flags
*op
,
3127 struct user_namespace
*mnt_userns
;
3128 struct dentry
*dir
= nd
->path
.dentry
;
3129 struct inode
*dir_inode
= dir
->d_inode
;
3130 int open_flag
= op
->open_flag
;
3131 struct dentry
*dentry
;
3132 int error
, create_error
= 0;
3133 umode_t mode
= op
->mode
;
3134 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
3136 if (unlikely(IS_DEADDIR(dir_inode
)))
3137 return ERR_PTR(-ENOENT
);
3139 file
->f_mode
&= ~FMODE_CREATED
;
3140 dentry
= d_lookup(dir
, &nd
->last
);
3143 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
3147 if (d_in_lookup(dentry
))
3150 error
= d_revalidate(dentry
, nd
->flags
);
3151 if (likely(error
> 0))
3155 d_invalidate(dentry
);
3159 if (dentry
->d_inode
) {
3160 /* Cached positive dentry: will open in f_op->open */
3165 * Checking write permission is tricky, bacuse we don't know if we are
3166 * going to actually need it: O_CREAT opens should work as long as the
3167 * file exists. But checking existence breaks atomicity. The trick is
3168 * to check access and if not granted clear O_CREAT from the flags.
3170 * Another problem is returing the "right" error value (e.g. for an
3171 * O_EXCL open we want to return EEXIST not EROFS).
3173 if (unlikely(!got_write
))
3174 open_flag
&= ~O_TRUNC
;
3175 mnt_userns
= mnt_user_ns(nd
->path
.mnt
);
3176 if (open_flag
& O_CREAT
) {
3177 if (open_flag
& O_EXCL
)
3178 open_flag
&= ~O_TRUNC
;
3179 if (!IS_POSIXACL(dir
->d_inode
))
3180 mode
&= ~current_umask();
3181 if (likely(got_write
))
3182 create_error
= may_o_create(mnt_userns
, &nd
->path
,
3185 create_error
= -EROFS
;
3188 open_flag
&= ~O_CREAT
;
3189 if (dir_inode
->i_op
->atomic_open
) {
3190 dentry
= atomic_open(nd
, dentry
, file
, open_flag
, mode
);
3191 if (unlikely(create_error
) && dentry
== ERR_PTR(-ENOENT
))
3192 dentry
= ERR_PTR(create_error
);
3196 if (d_in_lookup(dentry
)) {
3197 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3199 d_lookup_done(dentry
);
3200 if (unlikely(res
)) {
3202 error
= PTR_ERR(res
);
3210 /* Negative dentry, just create the file */
3211 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3212 file
->f_mode
|= FMODE_CREATED
;
3213 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3214 if (!dir_inode
->i_op
->create
) {
3219 error
= dir_inode
->i_op
->create(mnt_userns
, dir_inode
, dentry
,
3220 mode
, open_flag
& O_EXCL
);
3224 if (unlikely(create_error
) && !dentry
->d_inode
) {
3225 error
= create_error
;
3232 return ERR_PTR(error
);
3235 static const char *open_last_lookups(struct nameidata
*nd
,
3236 struct file
*file
, const struct open_flags
*op
)
3238 struct dentry
*dir
= nd
->path
.dentry
;
3239 int open_flag
= op
->open_flag
;
3240 bool got_write
= false;
3242 struct inode
*inode
;
3243 struct dentry
*dentry
;
3246 nd
->flags
|= op
->intent
;
3248 if (nd
->last_type
!= LAST_NORM
) {
3251 return handle_dots(nd
, nd
->last_type
);
3254 if (!(open_flag
& O_CREAT
)) {
3255 if (nd
->last
.name
[nd
->last
.len
])
3256 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3257 /* we _can_ be in RCU mode here */
3258 dentry
= lookup_fast(nd
, &inode
, &seq
);
3260 return ERR_CAST(dentry
);
3264 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3266 /* create side of things */
3267 if (nd
->flags
& LOOKUP_RCU
) {
3268 if (!try_to_unlazy(nd
))
3269 return ERR_PTR(-ECHILD
);
3271 audit_inode(nd
->name
, dir
, AUDIT_INODE_PARENT
);
3272 /* trailing slashes? */
3273 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3274 return ERR_PTR(-EISDIR
);
3277 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3278 got_write
= !mnt_want_write(nd
->path
.mnt
);
3280 * do _not_ fail yet - we might not need that or fail with
3281 * a different error; let lookup_open() decide; we'll be
3282 * dropping this one anyway.
3285 if (open_flag
& O_CREAT
)
3286 inode_lock(dir
->d_inode
);
3288 inode_lock_shared(dir
->d_inode
);
3289 dentry
= lookup_open(nd
, file
, op
, got_write
);
3290 if (!IS_ERR(dentry
) && (file
->f_mode
& FMODE_CREATED
))
3291 fsnotify_create(dir
->d_inode
, dentry
);
3292 if (open_flag
& O_CREAT
)
3293 inode_unlock(dir
->d_inode
);
3295 inode_unlock_shared(dir
->d_inode
);
3298 mnt_drop_write(nd
->path
.mnt
);
3301 return ERR_CAST(dentry
);
3303 if (file
->f_mode
& (FMODE_OPENED
| FMODE_CREATED
)) {
3304 dput(nd
->path
.dentry
);
3305 nd
->path
.dentry
= dentry
;
3312 res
= step_into(nd
, WALK_TRAILING
, dentry
, inode
, seq
);
3314 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3319 * Handle the last step of open()
3321 static int do_open(struct nameidata
*nd
,
3322 struct file
*file
, const struct open_flags
*op
)
3324 struct user_namespace
*mnt_userns
;
3325 int open_flag
= op
->open_flag
;
3330 if (!(file
->f_mode
& (FMODE_OPENED
| FMODE_CREATED
))) {
3331 error
= complete_walk(nd
);
3335 if (!(file
->f_mode
& FMODE_CREATED
))
3336 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3337 mnt_userns
= mnt_user_ns(nd
->path
.mnt
);
3338 if (open_flag
& O_CREAT
) {
3339 if ((open_flag
& O_EXCL
) && !(file
->f_mode
& FMODE_CREATED
))
3341 if (d_is_dir(nd
->path
.dentry
))
3343 error
= may_create_in_sticky(mnt_userns
, nd
,
3344 d_backing_inode(nd
->path
.dentry
));
3345 if (unlikely(error
))
3348 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3351 do_truncate
= false;
3352 acc_mode
= op
->acc_mode
;
3353 if (file
->f_mode
& FMODE_CREATED
) {
3354 /* Don't check for write permission, don't truncate */
3355 open_flag
&= ~O_TRUNC
;
3357 } else if (d_is_reg(nd
->path
.dentry
) && open_flag
& O_TRUNC
) {
3358 error
= mnt_want_write(nd
->path
.mnt
);
3363 error
= may_open(mnt_userns
, &nd
->path
, acc_mode
, open_flag
);
3364 if (!error
&& !(file
->f_mode
& FMODE_OPENED
))
3365 error
= vfs_open(&nd
->path
, file
);
3367 error
= ima_file_check(file
, op
->acc_mode
);
3368 if (!error
&& do_truncate
)
3369 error
= handle_truncate(mnt_userns
, file
);
3370 if (unlikely(error
> 0)) {
3375 mnt_drop_write(nd
->path
.mnt
);
3380 * vfs_tmpfile - create tmpfile
3381 * @mnt_userns: user namespace of the mount the inode was found from
3382 * @dentry: pointer to dentry of the base directory
3383 * @mode: mode of the new tmpfile
3384 * @open_flags: flags
3386 * Create a temporary file.
3388 * If the inode has been found through an idmapped mount the user namespace of
3389 * the vfsmount must be passed through @mnt_userns. This function will then take
3390 * care to map the inode according to @mnt_userns before checking permissions.
3391 * On non-idmapped mounts or if permission checking is to be performed on the
3392 * raw inode simply passs init_user_ns.
3394 struct dentry
*vfs_tmpfile(struct user_namespace
*mnt_userns
,
3395 struct dentry
*dentry
, umode_t mode
, int open_flag
)
3397 struct dentry
*child
= NULL
;
3398 struct inode
*dir
= dentry
->d_inode
;
3399 struct inode
*inode
;
3402 /* we want directory to be writable */
3403 error
= inode_permission(mnt_userns
, dir
, MAY_WRITE
| MAY_EXEC
);
3406 error
= -EOPNOTSUPP
;
3407 if (!dir
->i_op
->tmpfile
)
3410 child
= d_alloc(dentry
, &slash_name
);
3411 if (unlikely(!child
))
3413 error
= dir
->i_op
->tmpfile(mnt_userns
, dir
, child
, mode
);
3417 inode
= child
->d_inode
;
3418 if (unlikely(!inode
))
3420 if (!(open_flag
& O_EXCL
)) {
3421 spin_lock(&inode
->i_lock
);
3422 inode
->i_state
|= I_LINKABLE
;
3423 spin_unlock(&inode
->i_lock
);
3425 ima_post_create_tmpfile(mnt_userns
, inode
);
3430 return ERR_PTR(error
);
3432 EXPORT_SYMBOL(vfs_tmpfile
);
3434 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3435 const struct open_flags
*op
,
3438 struct user_namespace
*mnt_userns
;
3439 struct dentry
*child
;
3441 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3442 if (unlikely(error
))
3444 error
= mnt_want_write(path
.mnt
);
3445 if (unlikely(error
))
3447 mnt_userns
= mnt_user_ns(path
.mnt
);
3448 child
= vfs_tmpfile(mnt_userns
, path
.dentry
, op
->mode
, op
->open_flag
);
3449 error
= PTR_ERR(child
);
3453 path
.dentry
= child
;
3454 audit_inode(nd
->name
, child
, 0);
3455 /* Don't check for other permissions, the inode was just created */
3456 error
= may_open(mnt_userns
, &path
, 0, op
->open_flag
);
3458 error
= vfs_open(&path
, file
);
3460 mnt_drop_write(path
.mnt
);
3466 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3469 int error
= path_lookupat(nd
, flags
, &path
);
3471 audit_inode(nd
->name
, path
.dentry
, 0);
3472 error
= vfs_open(&path
, file
);
3478 static struct file
*path_openat(struct nameidata
*nd
,
3479 const struct open_flags
*op
, unsigned flags
)
3484 file
= alloc_empty_file(op
->open_flag
, current_cred());
3488 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3489 error
= do_tmpfile(nd
, flags
, op
, file
);
3490 } else if (unlikely(file
->f_flags
& O_PATH
)) {
3491 error
= do_o_path(nd
, flags
, file
);
3493 const char *s
= path_init(nd
, flags
);
3494 while (!(error
= link_path_walk(s
, nd
)) &&
3495 (s
= open_last_lookups(nd
, file
, op
)) != NULL
)
3498 error
= do_open(nd
, file
, op
);
3501 if (likely(!error
)) {
3502 if (likely(file
->f_mode
& FMODE_OPENED
))
3508 if (error
== -EOPENSTALE
) {
3509 if (flags
& LOOKUP_RCU
)
3514 return ERR_PTR(error
);
3517 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3518 const struct open_flags
*op
)
3520 struct nameidata nd
;
3521 int flags
= op
->lookup_flags
;
3524 set_nameidata(&nd
, dfd
, pathname
);
3525 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3526 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3527 filp
= path_openat(&nd
, op
, flags
);
3528 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3529 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3530 restore_nameidata();
3534 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3535 const char *name
, const struct open_flags
*op
)
3537 struct nameidata nd
;
3539 struct filename
*filename
;
3540 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3543 nd
.root
.dentry
= dentry
;
3545 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3546 return ERR_PTR(-ELOOP
);
3548 filename
= getname_kernel(name
);
3549 if (IS_ERR(filename
))
3550 return ERR_CAST(filename
);
3552 set_nameidata(&nd
, -1, filename
);
3553 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3554 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3555 file
= path_openat(&nd
, op
, flags
);
3556 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3557 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3558 restore_nameidata();
3563 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3564 struct path
*path
, unsigned int lookup_flags
)
3566 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3571 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3574 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3575 * other flags passed in are ignored!
3577 lookup_flags
&= LOOKUP_REVAL
;
3579 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3581 return ERR_CAST(name
);
3584 * Yucky last component or no last component at all?
3585 * (foo/., foo/.., /////)
3587 if (unlikely(type
!= LAST_NORM
))
3590 /* don't fail immediately if it's r/o, at least try to report other errors */
3591 err2
= mnt_want_write(path
->mnt
);
3593 * Do the final lookup.
3595 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3596 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3597 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3602 if (d_is_positive(dentry
))
3606 * Special case - lookup gave negative, but... we had foo/bar/
3607 * From the vfs_mknod() POV we just have a negative dentry -
3608 * all is fine. Let's be bastards - you had / on the end, you've
3609 * been asking for (non-existent) directory. -ENOENT for you.
3611 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3615 if (unlikely(err2
)) {
3623 dentry
= ERR_PTR(error
);
3625 inode_unlock(path
->dentry
->d_inode
);
3627 mnt_drop_write(path
->mnt
);
3634 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3635 struct path
*path
, unsigned int lookup_flags
)
3637 return filename_create(dfd
, getname_kernel(pathname
),
3638 path
, lookup_flags
);
3640 EXPORT_SYMBOL(kern_path_create
);
3642 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3645 inode_unlock(path
->dentry
->d_inode
);
3646 mnt_drop_write(path
->mnt
);
3649 EXPORT_SYMBOL(done_path_create
);
3651 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3652 struct path
*path
, unsigned int lookup_flags
)
3654 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3656 EXPORT_SYMBOL(user_path_create
);
3659 * vfs_mknod - create device node or file
3660 * @mnt_userns: user namespace of the mount the inode was found from
3661 * @dir: inode of @dentry
3662 * @dentry: pointer to dentry of the base directory
3663 * @mode: mode of the new device node or file
3664 * @dev: device number of device to create
3666 * Create a device node or file.
3668 * If the inode has been found through an idmapped mount the user namespace of
3669 * the vfsmount must be passed through @mnt_userns. This function will then take
3670 * care to map the inode according to @mnt_userns before checking permissions.
3671 * On non-idmapped mounts or if permission checking is to be performed on the
3672 * raw inode simply passs init_user_ns.
3674 int vfs_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3675 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3677 bool is_whiteout
= S_ISCHR(mode
) && dev
== WHITEOUT_DEV
;
3678 int error
= may_create(mnt_userns
, dir
, dentry
);
3683 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !is_whiteout
&&
3684 !capable(CAP_MKNOD
))
3687 if (!dir
->i_op
->mknod
)
3690 error
= devcgroup_inode_mknod(mode
, dev
);
3694 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3698 error
= dir
->i_op
->mknod(mnt_userns
, dir
, dentry
, mode
, dev
);
3700 fsnotify_create(dir
, dentry
);
3703 EXPORT_SYMBOL(vfs_mknod
);
3705 static int may_mknod(umode_t mode
)
3707 switch (mode
& S_IFMT
) {
3713 case 0: /* zero mode translates to S_IFREG */
3722 static long do_mknodat(int dfd
, const char __user
*filename
, umode_t mode
,
3725 struct user_namespace
*mnt_userns
;
3726 struct dentry
*dentry
;
3729 unsigned int lookup_flags
= 0;
3731 error
= may_mknod(mode
);
3735 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3737 return PTR_ERR(dentry
);
3739 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3740 mode
&= ~current_umask();
3741 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3745 mnt_userns
= mnt_user_ns(path
.mnt
);
3746 switch (mode
& S_IFMT
) {
3747 case 0: case S_IFREG
:
3748 error
= vfs_create(mnt_userns
, path
.dentry
->d_inode
,
3749 dentry
, mode
, true);
3751 ima_post_path_mknod(mnt_userns
, dentry
);
3753 case S_IFCHR
: case S_IFBLK
:
3754 error
= vfs_mknod(mnt_userns
, path
.dentry
->d_inode
,
3755 dentry
, mode
, new_decode_dev(dev
));
3757 case S_IFIFO
: case S_IFSOCK
:
3758 error
= vfs_mknod(mnt_userns
, path
.dentry
->d_inode
,
3763 done_path_create(&path
, dentry
);
3764 if (retry_estale(error
, lookup_flags
)) {
3765 lookup_flags
|= LOOKUP_REVAL
;
3771 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3774 return do_mknodat(dfd
, filename
, mode
, dev
);
3777 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3779 return do_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3783 * vfs_mkdir - create directory
3784 * @mnt_userns: user namespace of the mount the inode was found from
3785 * @dir: inode of @dentry
3786 * @dentry: pointer to dentry of the base directory
3787 * @mode: mode of the new directory
3789 * Create a directory.
3791 * If the inode has been found through an idmapped mount the user namespace of
3792 * the vfsmount must be passed through @mnt_userns. This function will then take
3793 * care to map the inode according to @mnt_userns before checking permissions.
3794 * On non-idmapped mounts or if permission checking is to be performed on the
3795 * raw inode simply passs init_user_ns.
3797 int vfs_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3798 struct dentry
*dentry
, umode_t mode
)
3800 int error
= may_create(mnt_userns
, dir
, dentry
);
3801 unsigned max_links
= dir
->i_sb
->s_max_links
;
3806 if (!dir
->i_op
->mkdir
)
3809 mode
&= (S_IRWXUGO
|S_ISVTX
);
3810 error
= security_inode_mkdir(dir
, dentry
, mode
);
3814 if (max_links
&& dir
->i_nlink
>= max_links
)
3817 error
= dir
->i_op
->mkdir(mnt_userns
, dir
, dentry
, mode
);
3819 fsnotify_mkdir(dir
, dentry
);
3822 EXPORT_SYMBOL(vfs_mkdir
);
3824 static long do_mkdirat(int dfd
, const char __user
*pathname
, umode_t mode
)
3826 struct dentry
*dentry
;
3829 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3832 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3834 return PTR_ERR(dentry
);
3836 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3837 mode
&= ~current_umask();
3838 error
= security_path_mkdir(&path
, dentry
, mode
);
3840 struct user_namespace
*mnt_userns
;
3841 mnt_userns
= mnt_user_ns(path
.mnt
);
3842 error
= vfs_mkdir(mnt_userns
, path
.dentry
->d_inode
, dentry
,
3845 done_path_create(&path
, dentry
);
3846 if (retry_estale(error
, lookup_flags
)) {
3847 lookup_flags
|= LOOKUP_REVAL
;
3853 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3855 return do_mkdirat(dfd
, pathname
, mode
);
3858 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3860 return do_mkdirat(AT_FDCWD
, pathname
, mode
);
3864 * vfs_rmdir - remove directory
3865 * @mnt_userns: user namespace of the mount the inode was found from
3866 * @dir: inode of @dentry
3867 * @dentry: pointer to dentry of the base directory
3869 * Remove a directory.
3871 * If the inode has been found through an idmapped mount the user namespace of
3872 * the vfsmount must be passed through @mnt_userns. This function will then take
3873 * care to map the inode according to @mnt_userns before checking permissions.
3874 * On non-idmapped mounts or if permission checking is to be performed on the
3875 * raw inode simply passs init_user_ns.
3877 int vfs_rmdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3878 struct dentry
*dentry
)
3880 int error
= may_delete(mnt_userns
, dir
, dentry
, 1);
3885 if (!dir
->i_op
->rmdir
)
3889 inode_lock(dentry
->d_inode
);
3892 if (is_local_mountpoint(dentry
))
3895 error
= security_inode_rmdir(dir
, dentry
);
3899 error
= dir
->i_op
->rmdir(dir
, dentry
);
3903 shrink_dcache_parent(dentry
);
3904 dentry
->d_inode
->i_flags
|= S_DEAD
;
3906 detach_mounts(dentry
);
3907 fsnotify_rmdir(dir
, dentry
);
3910 inode_unlock(dentry
->d_inode
);
3916 EXPORT_SYMBOL(vfs_rmdir
);
3918 long do_rmdir(int dfd
, struct filename
*name
)
3920 struct user_namespace
*mnt_userns
;
3922 struct dentry
*dentry
;
3926 unsigned int lookup_flags
= 0;
3928 name
= filename_parentat(dfd
, name
, lookup_flags
,
3929 &path
, &last
, &type
);
3931 return PTR_ERR(name
);
3945 error
= mnt_want_write(path
.mnt
);
3949 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3950 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3951 error
= PTR_ERR(dentry
);
3954 if (!dentry
->d_inode
) {
3958 error
= security_path_rmdir(&path
, dentry
);
3961 mnt_userns
= mnt_user_ns(path
.mnt
);
3962 error
= vfs_rmdir(mnt_userns
, path
.dentry
->d_inode
, dentry
);
3966 inode_unlock(path
.dentry
->d_inode
);
3967 mnt_drop_write(path
.mnt
);
3970 if (retry_estale(error
, lookup_flags
)) {
3971 lookup_flags
|= LOOKUP_REVAL
;
3978 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3980 return do_rmdir(AT_FDCWD
, getname(pathname
));
3984 * vfs_unlink - unlink a filesystem object
3985 * @mnt_userns: user namespace of the mount the inode was found from
3986 * @dir: parent directory
3988 * @delegated_inode: returns victim inode, if the inode is delegated.
3990 * The caller must hold dir->i_mutex.
3992 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3993 * return a reference to the inode in delegated_inode. The caller
3994 * should then break the delegation on that inode and retry. Because
3995 * breaking a delegation may take a long time, the caller should drop
3996 * dir->i_mutex before doing so.
3998 * Alternatively, a caller may pass NULL for delegated_inode. This may
3999 * be appropriate for callers that expect the underlying filesystem not
4000 * to be NFS exported.
4002 * If the inode has been found through an idmapped mount the user namespace of
4003 * the vfsmount must be passed through @mnt_userns. This function will then take
4004 * care to map the inode according to @mnt_userns before checking permissions.
4005 * On non-idmapped mounts or if permission checking is to be performed on the
4006 * raw inode simply passs init_user_ns.
4008 int vfs_unlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
4009 struct dentry
*dentry
, struct inode
**delegated_inode
)
4011 struct inode
*target
= dentry
->d_inode
;
4012 int error
= may_delete(mnt_userns
, dir
, dentry
, 0);
4017 if (!dir
->i_op
->unlink
)
4021 if (is_local_mountpoint(dentry
))
4024 error
= security_inode_unlink(dir
, dentry
);
4026 error
= try_break_deleg(target
, delegated_inode
);
4029 error
= dir
->i_op
->unlink(dir
, dentry
);
4032 detach_mounts(dentry
);
4033 fsnotify_unlink(dir
, dentry
);
4038 inode_unlock(target
);
4040 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4041 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
4042 fsnotify_link_count(target
);
4048 EXPORT_SYMBOL(vfs_unlink
);
4051 * Make sure that the actual truncation of the file will occur outside its
4052 * directory's i_mutex. Truncate can take a long time if there is a lot of
4053 * writeout happening, and we don't want to prevent access to the directory
4054 * while waiting on the I/O.
4056 long do_unlinkat(int dfd
, struct filename
*name
)
4059 struct dentry
*dentry
;
4063 struct inode
*inode
= NULL
;
4064 struct inode
*delegated_inode
= NULL
;
4065 unsigned int lookup_flags
= 0;
4067 name
= filename_parentat(dfd
, name
, lookup_flags
, &path
, &last
, &type
);
4069 return PTR_ERR(name
);
4072 if (type
!= LAST_NORM
)
4075 error
= mnt_want_write(path
.mnt
);
4079 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4080 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
4081 error
= PTR_ERR(dentry
);
4082 if (!IS_ERR(dentry
)) {
4083 struct user_namespace
*mnt_userns
;
4085 /* Why not before? Because we want correct error value */
4086 if (last
.name
[last
.len
])
4088 inode
= dentry
->d_inode
;
4089 if (d_is_negative(dentry
))
4092 error
= security_path_unlink(&path
, dentry
);
4095 mnt_userns
= mnt_user_ns(path
.mnt
);
4096 error
= vfs_unlink(mnt_userns
, path
.dentry
->d_inode
, dentry
,
4101 inode_unlock(path
.dentry
->d_inode
);
4103 iput(inode
); /* truncate the inode here */
4105 if (delegated_inode
) {
4106 error
= break_deleg_wait(&delegated_inode
);
4110 mnt_drop_write(path
.mnt
);
4113 if (retry_estale(error
, lookup_flags
)) {
4114 lookup_flags
|= LOOKUP_REVAL
;
4122 if (d_is_negative(dentry
))
4124 else if (d_is_dir(dentry
))
4131 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4133 if ((flag
& ~AT_REMOVEDIR
) != 0)
4136 if (flag
& AT_REMOVEDIR
)
4137 return do_rmdir(dfd
, getname(pathname
));
4138 return do_unlinkat(dfd
, getname(pathname
));
4141 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4143 return do_unlinkat(AT_FDCWD
, getname(pathname
));
4147 * vfs_symlink - create symlink
4148 * @mnt_userns: user namespace of the mount the inode was found from
4149 * @dir: inode of @dentry
4150 * @dentry: pointer to dentry of the base directory
4151 * @oldname: name of the file to link to
4155 * If the inode has been found through an idmapped mount the user namespace of
4156 * the vfsmount must be passed through @mnt_userns. This function will then take
4157 * care to map the inode according to @mnt_userns before checking permissions.
4158 * On non-idmapped mounts or if permission checking is to be performed on the
4159 * raw inode simply passs init_user_ns.
4161 int vfs_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
4162 struct dentry
*dentry
, const char *oldname
)
4164 int error
= may_create(mnt_userns
, dir
, dentry
);
4169 if (!dir
->i_op
->symlink
)
4172 error
= security_inode_symlink(dir
, dentry
, oldname
);
4176 error
= dir
->i_op
->symlink(mnt_userns
, dir
, dentry
, oldname
);
4178 fsnotify_create(dir
, dentry
);
4181 EXPORT_SYMBOL(vfs_symlink
);
4183 static long do_symlinkat(const char __user
*oldname
, int newdfd
,
4184 const char __user
*newname
)
4187 struct filename
*from
;
4188 struct dentry
*dentry
;
4190 unsigned int lookup_flags
= 0;
4192 from
= getname(oldname
);
4194 return PTR_ERR(from
);
4196 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4197 error
= PTR_ERR(dentry
);
4201 error
= security_path_symlink(&path
, dentry
, from
->name
);
4203 struct user_namespace
*mnt_userns
;
4205 mnt_userns
= mnt_user_ns(path
.mnt
);
4206 error
= vfs_symlink(mnt_userns
, path
.dentry
->d_inode
, dentry
,
4209 done_path_create(&path
, dentry
);
4210 if (retry_estale(error
, lookup_flags
)) {
4211 lookup_flags
|= LOOKUP_REVAL
;
4219 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4220 int, newdfd
, const char __user
*, newname
)
4222 return do_symlinkat(oldname
, newdfd
, newname
);
4225 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4227 return do_symlinkat(oldname
, AT_FDCWD
, newname
);
4231 * vfs_link - create a new link
4232 * @old_dentry: object to be linked
4233 * @mnt_userns: the user namespace of the mount
4235 * @new_dentry: where to create the new link
4236 * @delegated_inode: returns inode needing a delegation break
4238 * The caller must hold dir->i_mutex
4240 * If vfs_link discovers a delegation on the to-be-linked file in need
4241 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4242 * inode in delegated_inode. The caller should then break the delegation
4243 * and retry. Because breaking a delegation may take a long time, the
4244 * caller should drop the i_mutex before doing so.
4246 * Alternatively, a caller may pass NULL for delegated_inode. This may
4247 * be appropriate for callers that expect the underlying filesystem not
4248 * to be NFS exported.
4250 * If the inode has been found through an idmapped mount the user namespace of
4251 * the vfsmount must be passed through @mnt_userns. This function will then take
4252 * care to map the inode according to @mnt_userns before checking permissions.
4253 * On non-idmapped mounts or if permission checking is to be performed on the
4254 * raw inode simply passs init_user_ns.
4256 int vfs_link(struct dentry
*old_dentry
, struct user_namespace
*mnt_userns
,
4257 struct inode
*dir
, struct dentry
*new_dentry
,
4258 struct inode
**delegated_inode
)
4260 struct inode
*inode
= old_dentry
->d_inode
;
4261 unsigned max_links
= dir
->i_sb
->s_max_links
;
4267 error
= may_create(mnt_userns
, dir
, new_dentry
);
4271 if (dir
->i_sb
!= inode
->i_sb
)
4275 * A link to an append-only or immutable file cannot be created.
4277 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4280 * Updating the link count will likely cause i_uid and i_gid to
4281 * be writen back improperly if their true value is unknown to
4284 if (HAS_UNMAPPED_ID(mnt_userns
, inode
))
4286 if (!dir
->i_op
->link
)
4288 if (S_ISDIR(inode
->i_mode
))
4291 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4296 /* Make sure we don't allow creating hardlink to an unlinked file */
4297 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4299 else if (max_links
&& inode
->i_nlink
>= max_links
)
4302 error
= try_break_deleg(inode
, delegated_inode
);
4304 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4307 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4308 spin_lock(&inode
->i_lock
);
4309 inode
->i_state
&= ~I_LINKABLE
;
4310 spin_unlock(&inode
->i_lock
);
4312 inode_unlock(inode
);
4314 fsnotify_link(dir
, inode
, new_dentry
);
4317 EXPORT_SYMBOL(vfs_link
);
4320 * Hardlinks are often used in delicate situations. We avoid
4321 * security-related surprises by not following symlinks on the
4324 * We don't follow them on the oldname either to be compatible
4325 * with linux 2.0, and to avoid hard-linking to directories
4326 * and other special files. --ADM
4328 static int do_linkat(int olddfd
, const char __user
*oldname
, int newdfd
,
4329 const char __user
*newname
, int flags
)
4331 struct user_namespace
*mnt_userns
;
4332 struct dentry
*new_dentry
;
4333 struct path old_path
, new_path
;
4334 struct inode
*delegated_inode
= NULL
;
4338 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4341 * To use null names we require CAP_DAC_READ_SEARCH
4342 * This ensures that not everyone will be able to create
4343 * handlink using the passed filedescriptor.
4345 if (flags
& AT_EMPTY_PATH
) {
4346 if (!capable(CAP_DAC_READ_SEARCH
))
4351 if (flags
& AT_SYMLINK_FOLLOW
)
4352 how
|= LOOKUP_FOLLOW
;
4354 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4358 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4359 (how
& LOOKUP_REVAL
));
4360 error
= PTR_ERR(new_dentry
);
4361 if (IS_ERR(new_dentry
))
4365 if (old_path
.mnt
!= new_path
.mnt
)
4367 mnt_userns
= mnt_user_ns(new_path
.mnt
);
4368 error
= may_linkat(mnt_userns
, &old_path
);
4369 if (unlikely(error
))
4371 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4374 error
= vfs_link(old_path
.dentry
, mnt_userns
, new_path
.dentry
->d_inode
,
4375 new_dentry
, &delegated_inode
);
4377 done_path_create(&new_path
, new_dentry
);
4378 if (delegated_inode
) {
4379 error
= break_deleg_wait(&delegated_inode
);
4381 path_put(&old_path
);
4385 if (retry_estale(error
, how
)) {
4386 path_put(&old_path
);
4387 how
|= LOOKUP_REVAL
;
4391 path_put(&old_path
);
4396 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4397 int, newdfd
, const char __user
*, newname
, int, flags
)
4399 return do_linkat(olddfd
, oldname
, newdfd
, newname
, flags
);
4402 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4404 return do_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4408 * vfs_rename - rename a filesystem object
4409 * @old_mnt_userns: old user namespace of the mount the inode was found from
4410 * @old_dir: parent of source
4411 * @old_dentry: source
4412 * @new_mnt_userns: new user namespace of the mount the inode was found from
4413 * @new_dir: parent of destination
4414 * @new_dentry: destination
4415 * @delegated_inode: returns an inode needing a delegation break
4416 * @flags: rename flags
4418 * The caller must hold multiple mutexes--see lock_rename()).
4420 * If vfs_rename discovers a delegation in need of breaking at either
4421 * the source or destination, it will return -EWOULDBLOCK and return a
4422 * reference to the inode in delegated_inode. The caller should then
4423 * break the delegation and retry. Because breaking a delegation may
4424 * take a long time, the caller should drop all locks before doing
4427 * Alternatively, a caller may pass NULL for delegated_inode. This may
4428 * be appropriate for callers that expect the underlying filesystem not
4429 * to be NFS exported.
4431 * The worst of all namespace operations - renaming directory. "Perverted"
4432 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4435 * a) we can get into loop creation.
4436 * b) race potential - two innocent renames can create a loop together.
4437 * That's where 4.4 screws up. Current fix: serialization on
4438 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4440 * c) we have to lock _four_ objects - parents and victim (if it exists),
4441 * and source (if it is not a directory).
4442 * And that - after we got ->i_mutex on parents (until then we don't know
4443 * whether the target exists). Solution: try to be smart with locking
4444 * order for inodes. We rely on the fact that tree topology may change
4445 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4446 * move will be locked. Thus we can rank directories by the tree
4447 * (ancestors first) and rank all non-directories after them.
4448 * That works since everybody except rename does "lock parent, lookup,
4449 * lock child" and rename is under ->s_vfs_rename_mutex.
4450 * HOWEVER, it relies on the assumption that any object with ->lookup()
4451 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4452 * we'd better make sure that there's no link(2) for them.
4453 * d) conversion from fhandle to dentry may come in the wrong moment - when
4454 * we are removing the target. Solution: we will have to grab ->i_mutex
4455 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4456 * ->i_mutex on parents, which works but leads to some truly excessive
4459 int vfs_rename(struct renamedata
*rd
)
4462 struct inode
*old_dir
= rd
->old_dir
, *new_dir
= rd
->new_dir
;
4463 struct dentry
*old_dentry
= rd
->old_dentry
;
4464 struct dentry
*new_dentry
= rd
->new_dentry
;
4465 struct inode
**delegated_inode
= rd
->delegated_inode
;
4466 unsigned int flags
= rd
->flags
;
4467 bool is_dir
= d_is_dir(old_dentry
);
4468 struct inode
*source
= old_dentry
->d_inode
;
4469 struct inode
*target
= new_dentry
->d_inode
;
4470 bool new_is_dir
= false;
4471 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4472 struct name_snapshot old_name
;
4474 if (source
== target
)
4477 error
= may_delete(rd
->old_mnt_userns
, old_dir
, old_dentry
, is_dir
);
4482 error
= may_create(rd
->new_mnt_userns
, new_dir
, new_dentry
);
4484 new_is_dir
= d_is_dir(new_dentry
);
4486 if (!(flags
& RENAME_EXCHANGE
))
4487 error
= may_delete(rd
->new_mnt_userns
, new_dir
,
4488 new_dentry
, is_dir
);
4490 error
= may_delete(rd
->new_mnt_userns
, new_dir
,
4491 new_dentry
, new_is_dir
);
4496 if (!old_dir
->i_op
->rename
)
4500 * If we are going to change the parent - check write permissions,
4501 * we'll need to flip '..'.
4503 if (new_dir
!= old_dir
) {
4505 error
= inode_permission(rd
->old_mnt_userns
, source
,
4510 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4511 error
= inode_permission(rd
->new_mnt_userns
, target
,
4518 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4523 take_dentry_name_snapshot(&old_name
, old_dentry
);
4525 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4526 lock_two_nondirectories(source
, target
);
4531 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4534 if (max_links
&& new_dir
!= old_dir
) {
4536 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4538 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4539 old_dir
->i_nlink
>= max_links
)
4543 error
= try_break_deleg(source
, delegated_inode
);
4547 if (target
&& !new_is_dir
) {
4548 error
= try_break_deleg(target
, delegated_inode
);
4552 error
= old_dir
->i_op
->rename(rd
->new_mnt_userns
, old_dir
, old_dentry
,
4553 new_dir
, new_dentry
, flags
);
4557 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4559 shrink_dcache_parent(new_dentry
);
4560 target
->i_flags
|= S_DEAD
;
4562 dont_mount(new_dentry
);
4563 detach_mounts(new_dentry
);
4565 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4566 if (!(flags
& RENAME_EXCHANGE
))
4567 d_move(old_dentry
, new_dentry
);
4569 d_exchange(old_dentry
, new_dentry
);
4572 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4573 unlock_two_nondirectories(source
, target
);
4575 inode_unlock(target
);
4578 fsnotify_move(old_dir
, new_dir
, &old_name
.name
, is_dir
,
4579 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4580 if (flags
& RENAME_EXCHANGE
) {
4581 fsnotify_move(new_dir
, old_dir
, &old_dentry
->d_name
,
4582 new_is_dir
, NULL
, new_dentry
);
4585 release_dentry_name_snapshot(&old_name
);
4589 EXPORT_SYMBOL(vfs_rename
);
4591 int do_renameat2(int olddfd
, struct filename
*from
, int newdfd
,
4592 struct filename
*to
, unsigned int flags
)
4594 struct renamedata rd
;
4595 struct dentry
*old_dentry
, *new_dentry
;
4596 struct dentry
*trap
;
4597 struct path old_path
, new_path
;
4598 struct qstr old_last
, new_last
;
4599 int old_type
, new_type
;
4600 struct inode
*delegated_inode
= NULL
;
4601 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4602 bool should_retry
= false;
4603 int error
= -EINVAL
;
4605 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4608 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4609 (flags
& RENAME_EXCHANGE
))
4612 if (flags
& RENAME_EXCHANGE
)
4616 from
= filename_parentat(olddfd
, from
, lookup_flags
, &old_path
,
4617 &old_last
, &old_type
);
4619 error
= PTR_ERR(from
);
4623 to
= filename_parentat(newdfd
, to
, lookup_flags
, &new_path
, &new_last
,
4626 error
= PTR_ERR(to
);
4631 if (old_path
.mnt
!= new_path
.mnt
)
4635 if (old_type
!= LAST_NORM
)
4638 if (flags
& RENAME_NOREPLACE
)
4640 if (new_type
!= LAST_NORM
)
4643 error
= mnt_want_write(old_path
.mnt
);
4648 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4650 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4651 error
= PTR_ERR(old_dentry
);
4652 if (IS_ERR(old_dentry
))
4654 /* source must exist */
4656 if (d_is_negative(old_dentry
))
4658 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4659 error
= PTR_ERR(new_dentry
);
4660 if (IS_ERR(new_dentry
))
4663 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4665 if (flags
& RENAME_EXCHANGE
) {
4667 if (d_is_negative(new_dentry
))
4670 if (!d_is_dir(new_dentry
)) {
4672 if (new_last
.name
[new_last
.len
])
4676 /* unless the source is a directory trailing slashes give -ENOTDIR */
4677 if (!d_is_dir(old_dentry
)) {
4679 if (old_last
.name
[old_last
.len
])
4681 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4684 /* source should not be ancestor of target */
4686 if (old_dentry
== trap
)
4688 /* target should not be an ancestor of source */
4689 if (!(flags
& RENAME_EXCHANGE
))
4691 if (new_dentry
== trap
)
4694 error
= security_path_rename(&old_path
, old_dentry
,
4695 &new_path
, new_dentry
, flags
);
4699 rd
.old_dir
= old_path
.dentry
->d_inode
;
4700 rd
.old_dentry
= old_dentry
;
4701 rd
.old_mnt_userns
= mnt_user_ns(old_path
.mnt
);
4702 rd
.new_dir
= new_path
.dentry
->d_inode
;
4703 rd
.new_dentry
= new_dentry
;
4704 rd
.new_mnt_userns
= mnt_user_ns(new_path
.mnt
);
4705 rd
.delegated_inode
= &delegated_inode
;
4707 error
= vfs_rename(&rd
);
4713 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4714 if (delegated_inode
) {
4715 error
= break_deleg_wait(&delegated_inode
);
4719 mnt_drop_write(old_path
.mnt
);
4721 if (retry_estale(error
, lookup_flags
))
4722 should_retry
= true;
4723 path_put(&new_path
);
4725 path_put(&old_path
);
4727 should_retry
= false;
4728 lookup_flags
|= LOOKUP_REVAL
;
4740 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4741 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4743 return do_renameat2(olddfd
, getname(oldname
), newdfd
, getname(newname
),
4747 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4748 int, newdfd
, const char __user
*, newname
)
4750 return do_renameat2(olddfd
, getname(oldname
), newdfd
, getname(newname
),
4754 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4756 return do_renameat2(AT_FDCWD
, getname(oldname
), AT_FDCWD
,
4757 getname(newname
), 0);
4760 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4762 int len
= PTR_ERR(link
);
4767 if (len
> (unsigned) buflen
)
4769 if (copy_to_user(buffer
, link
, len
))
4776 * vfs_readlink - copy symlink body into userspace buffer
4777 * @dentry: dentry on which to get symbolic link
4778 * @buffer: user memory pointer
4779 * @buflen: size of buffer
4781 * Does not touch atime. That's up to the caller if necessary
4783 * Does not call security hook.
4785 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4787 struct inode
*inode
= d_inode(dentry
);
4788 DEFINE_DELAYED_CALL(done
);
4792 if (unlikely(!(inode
->i_opflags
& IOP_DEFAULT_READLINK
))) {
4793 if (unlikely(inode
->i_op
->readlink
))
4794 return inode
->i_op
->readlink(dentry
, buffer
, buflen
);
4796 if (!d_is_symlink(dentry
))
4799 spin_lock(&inode
->i_lock
);
4800 inode
->i_opflags
|= IOP_DEFAULT_READLINK
;
4801 spin_unlock(&inode
->i_lock
);
4804 link
= READ_ONCE(inode
->i_link
);
4806 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4808 return PTR_ERR(link
);
4810 res
= readlink_copy(buffer
, buflen
, link
);
4811 do_delayed_call(&done
);
4814 EXPORT_SYMBOL(vfs_readlink
);
4817 * vfs_get_link - get symlink body
4818 * @dentry: dentry on which to get symbolic link
4819 * @done: caller needs to free returned data with this
4821 * Calls security hook and i_op->get_link() on the supplied inode.
4823 * It does not touch atime. That's up to the caller if necessary.
4825 * Does not work on "special" symlinks like /proc/$$/fd/N
4827 const char *vfs_get_link(struct dentry
*dentry
, struct delayed_call
*done
)
4829 const char *res
= ERR_PTR(-EINVAL
);
4830 struct inode
*inode
= d_inode(dentry
);
4832 if (d_is_symlink(dentry
)) {
4833 res
= ERR_PTR(security_inode_readlink(dentry
));
4835 res
= inode
->i_op
->get_link(dentry
, inode
, done
);
4839 EXPORT_SYMBOL(vfs_get_link
);
4841 /* get the link contents into pagecache */
4842 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4843 struct delayed_call
*callback
)
4847 struct address_space
*mapping
= inode
->i_mapping
;
4850 page
= find_get_page(mapping
, 0);
4852 return ERR_PTR(-ECHILD
);
4853 if (!PageUptodate(page
)) {
4855 return ERR_PTR(-ECHILD
);
4858 page
= read_mapping_page(mapping
, 0, NULL
);
4862 set_delayed_call(callback
, page_put_link
, page
);
4863 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4864 kaddr
= page_address(page
);
4865 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4869 EXPORT_SYMBOL(page_get_link
);
4871 void page_put_link(void *arg
)
4875 EXPORT_SYMBOL(page_put_link
);
4877 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4879 DEFINE_DELAYED_CALL(done
);
4880 int res
= readlink_copy(buffer
, buflen
,
4881 page_get_link(dentry
, d_inode(dentry
),
4883 do_delayed_call(&done
);
4886 EXPORT_SYMBOL(page_readlink
);
4889 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4891 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4893 struct address_space
*mapping
= inode
->i_mapping
;
4897 unsigned int flags
= 0;
4899 flags
|= AOP_FLAG_NOFS
;
4902 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4903 flags
, &page
, &fsdata
);
4907 memcpy(page_address(page
), symname
, len
-1);
4909 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4916 mark_inode_dirty(inode
);
4921 EXPORT_SYMBOL(__page_symlink
);
4923 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4925 return __page_symlink(inode
, symname
, len
,
4926 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4928 EXPORT_SYMBOL(page_symlink
);
4930 const struct inode_operations page_symlink_inode_operations
= {
4931 .get_link
= page_get_link
,
4933 EXPORT_SYMBOL(page_symlink_inode_operations
);