4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <linux/bitops.h>
39 #include <asm/uaccess.h>
44 /* [Feb-1997 T. Schoebel-Theuer]
45 * Fundamental changes in the pathname lookup mechanisms (namei)
46 * were necessary because of omirr. The reason is that omirr needs
47 * to know the _real_ pathname, not the user-supplied one, in case
48 * of symlinks (and also when transname replacements occur).
50 * The new code replaces the old recursive symlink resolution with
51 * an iterative one (in case of non-nested symlink chains). It does
52 * this with calls to <fs>_follow_link().
53 * As a side effect, dir_namei(), _namei() and follow_link() are now
54 * replaced with a single function lookup_dentry() that can handle all
55 * the special cases of the former code.
57 * With the new dcache, the pathname is stored at each inode, at least as
58 * long as the refcount of the inode is positive. As a side effect, the
59 * size of the dcache depends on the inode cache and thus is dynamic.
61 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
62 * resolution to correspond with current state of the code.
64 * Note that the symlink resolution is not *completely* iterative.
65 * There is still a significant amount of tail- and mid- recursion in
66 * the algorithm. Also, note that <fs>_readlink() is not used in
67 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
68 * may return different results than <fs>_follow_link(). Many virtual
69 * filesystems (including /proc) exhibit this behavior.
72 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
73 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
74 * and the name already exists in form of a symlink, try to create the new
75 * name indicated by the symlink. The old code always complained that the
76 * name already exists, due to not following the symlink even if its target
77 * is nonexistent. The new semantics affects also mknod() and link() when
78 * the name is a symlink pointing to a non-existent name.
80 * I don't know which semantics is the right one, since I have no access
81 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
82 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
83 * "old" one. Personally, I think the new semantics is much more logical.
84 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
85 * file does succeed in both HP-UX and SunOs, but not in Solaris
86 * and in the old Linux semantics.
89 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
90 * semantics. See the comments in "open_namei" and "do_link" below.
92 * [10-Sep-98 Alan Modra] Another symlink change.
95 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
96 * inside the path - always follow.
97 * in the last component in creation/removal/renaming - never follow.
98 * if LOOKUP_FOLLOW passed - follow.
99 * if the pathname has trailing slashes - follow.
100 * otherwise - don't follow.
101 * (applied in that order).
103 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
104 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
105 * During the 2.4 we need to fix the userland stuff depending on it -
106 * hopefully we will be able to get rid of that wart in 2.5. So far only
107 * XEmacs seems to be relying on it...
110 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
111 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
112 * any extra contention...
115 /* In order to reduce some races, while at the same time doing additional
116 * checking and hopefully speeding things up, we copy filenames to the
117 * kernel data space before using them..
119 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
120 * PATH_MAX includes the nul terminator --RR.
123 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126 getname_flags(const char __user
*filename
, int flags
, int *empty
)
128 struct filename
*result
;
132 result
= audit_reusename(filename
);
136 result
= __getname();
137 if (unlikely(!result
))
138 return ERR_PTR(-ENOMEM
);
141 * First, try to embed the struct filename inside the names_cache
144 kname
= (char *)result
->iname
;
145 result
->name
= kname
;
147 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
148 if (unlikely(len
< 0)) {
154 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155 * separate struct filename so we can dedicate the entire
156 * names_cache allocation for the pathname, and re-do the copy from
159 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
160 const size_t size
= offsetof(struct filename
, iname
[1]);
161 kname
= (char *)result
;
164 * size is chosen that way we to guarantee that
165 * result->iname[0] is within the same object and that
166 * kname can't be equal to result->iname, no matter what.
168 result
= kzalloc(size
, GFP_KERNEL
);
169 if (unlikely(!result
)) {
171 return ERR_PTR(-ENOMEM
);
173 result
->name
= kname
;
174 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
175 if (unlikely(len
< 0)) {
180 if (unlikely(len
== PATH_MAX
)) {
183 return ERR_PTR(-ENAMETOOLONG
);
188 /* The empty path is special. */
189 if (unlikely(!len
)) {
192 if (!(flags
& LOOKUP_EMPTY
)) {
194 return ERR_PTR(-ENOENT
);
198 result
->uptr
= filename
;
199 result
->aname
= NULL
;
200 audit_getname(result
);
205 getname(const char __user
* filename
)
207 return getname_flags(filename
, 0, NULL
);
211 getname_kernel(const char * filename
)
213 struct filename
*result
;
214 int len
= strlen(filename
) + 1;
216 result
= __getname();
217 if (unlikely(!result
))
218 return ERR_PTR(-ENOMEM
);
220 if (len
<= EMBEDDED_NAME_MAX
) {
221 result
->name
= (char *)result
->iname
;
222 } else if (len
<= PATH_MAX
) {
223 struct filename
*tmp
;
225 tmp
= kmalloc(sizeof(*tmp
), GFP_KERNEL
);
226 if (unlikely(!tmp
)) {
228 return ERR_PTR(-ENOMEM
);
230 tmp
->name
= (char *)result
;
234 return ERR_PTR(-ENAMETOOLONG
);
236 memcpy((char *)result
->name
, filename
, len
);
238 result
->aname
= NULL
;
240 audit_getname(result
);
245 void putname(struct filename
*name
)
247 BUG_ON(name
->refcnt
<= 0);
249 if (--name
->refcnt
> 0)
252 if (name
->name
!= name
->iname
) {
253 __putname(name
->name
);
259 static int check_acl(struct inode
*inode
, int mask
)
261 #ifdef CONFIG_FS_POSIX_ACL
262 struct posix_acl
*acl
;
264 if (mask
& MAY_NOT_BLOCK
) {
265 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
268 /* no ->get_acl() calls in RCU mode... */
269 if (is_uncached_acl(acl
))
271 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
274 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
278 int error
= posix_acl_permission(inode
, acl
, mask
);
279 posix_acl_release(acl
);
288 * This does the basic permission checking
290 static int acl_permission_check(struct inode
*inode
, int mask
)
292 unsigned int mode
= inode
->i_mode
;
294 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
297 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
298 int error
= check_acl(inode
, mask
);
299 if (error
!= -EAGAIN
)
303 if (in_group_p(inode
->i_gid
))
308 * If the DACs are ok we don't need any capability check.
310 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
316 * generic_permission - check for access rights on a Posix-like filesystem
317 * @inode: inode to check access rights for
318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 * Used to check for read/write/execute permissions on a file.
321 * We use "fsuid" for this, letting us set arbitrary permissions
322 * for filesystem access without changing the "normal" uids which
323 * are used for other things.
325 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326 * request cannot be satisfied (eg. requires blocking or too much complexity).
327 * It would then be called again in ref-walk mode.
329 int generic_permission(struct inode
*inode
, int mask
)
334 * Do the basic permission checks.
336 ret
= acl_permission_check(inode
, mask
);
340 if (S_ISDIR(inode
->i_mode
)) {
341 /* DACs are overridable for directories */
342 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
344 if (!(mask
& MAY_WRITE
))
345 if (capable_wrt_inode_uidgid(inode
,
346 CAP_DAC_READ_SEARCH
))
351 * Read/write DACs are always overridable.
352 * Executable DACs are overridable when there is
353 * at least one exec bit set.
355 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
356 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
360 * Searching includes executable on directories, else just read.
362 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
363 if (mask
== MAY_READ
)
364 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
369 EXPORT_SYMBOL(generic_permission
);
372 * We _really_ want to just do "generic_permission()" without
373 * even looking at the inode->i_op values. So we keep a cache
374 * flag in inode->i_opflags, that says "this has not special
375 * permission function, use the fast case".
377 static inline int do_inode_permission(struct inode
*inode
, int mask
)
379 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
380 if (likely(inode
->i_op
->permission
))
381 return inode
->i_op
->permission(inode
, mask
);
383 /* This gets set once for the inode lifetime */
384 spin_lock(&inode
->i_lock
);
385 inode
->i_opflags
|= IOP_FASTPERM
;
386 spin_unlock(&inode
->i_lock
);
388 return generic_permission(inode
, mask
);
392 * __inode_permission - Check for access rights to a given inode
393 * @inode: Inode to check permission on
394 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 * Check for read/write/execute permissions on an inode.
398 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 * This does not check for a read-only file system. You probably want
401 * inode_permission().
403 int __inode_permission(struct inode
*inode
, int mask
)
407 if (unlikely(mask
& MAY_WRITE
)) {
409 * Nobody gets write access to an immutable file.
411 if (IS_IMMUTABLE(inode
))
415 retval
= do_inode_permission(inode
, mask
);
419 retval
= devcgroup_inode_permission(inode
, mask
);
423 return security_inode_permission(inode
, mask
);
425 EXPORT_SYMBOL(__inode_permission
);
428 * sb_permission - Check superblock-level permissions
429 * @sb: Superblock of inode to check permission on
430 * @inode: Inode to check permission on
431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
433 * Separate out file-system wide checks from inode-specific permission checks.
435 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
437 if (unlikely(mask
& MAY_WRITE
)) {
438 umode_t mode
= inode
->i_mode
;
440 /* Nobody gets write access to a read-only fs. */
441 if ((sb
->s_flags
& MS_RDONLY
) &&
442 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
449 * inode_permission - Check for access rights to a given inode
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
453 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
454 * this, letting us set arbitrary permissions for filesystem access without
455 * changing the "normal" UIDs which are used for other things.
457 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
459 int inode_permission(struct inode
*inode
, int mask
)
463 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
466 return __inode_permission(inode
, mask
);
468 EXPORT_SYMBOL(inode_permission
);
471 * path_get - get a reference to a path
472 * @path: path to get the reference to
474 * Given a path increment the reference count to the dentry and the vfsmount.
476 void path_get(const struct path
*path
)
481 EXPORT_SYMBOL(path_get
);
484 * path_put - put a reference to a path
485 * @path: path to put the reference to
487 * Given a path decrement the reference count to the dentry and the vfsmount.
489 void path_put(const struct path
*path
)
494 EXPORT_SYMBOL(path_put
);
496 #define EMBEDDED_LEVELS 2
501 struct inode
*inode
; /* path.dentry.d_inode */
506 int total_link_count
;
509 struct delayed_call done
;
512 } *stack
, internal
[EMBEDDED_LEVELS
];
513 struct filename
*name
;
514 struct nameidata
*saved
;
515 struct inode
*link_inode
;
520 static void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
522 struct nameidata
*old
= current
->nameidata
;
523 p
->stack
= p
->internal
;
526 p
->total_link_count
= old
? old
->total_link_count
: 0;
528 current
->nameidata
= p
;
531 static void restore_nameidata(void)
533 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
535 current
->nameidata
= old
;
537 old
->total_link_count
= now
->total_link_count
;
538 if (now
->stack
!= now
->internal
)
542 static int __nd_alloc_stack(struct nameidata
*nd
)
546 if (nd
->flags
& LOOKUP_RCU
) {
547 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
552 p
= kmalloc(MAXSYMLINKS
* sizeof(struct saved
),
557 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
563 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
564 * @path: nameidate to verify
566 * Rename can sometimes move a file or directory outside of a bind
567 * mount, path_connected allows those cases to be detected.
569 static bool path_connected(const struct path
*path
)
571 struct vfsmount
*mnt
= path
->mnt
;
573 /* Only bind mounts can have disconnected paths */
574 if (mnt
->mnt_root
== mnt
->mnt_sb
->s_root
)
577 return is_subdir(path
->dentry
, mnt
->mnt_root
);
580 static inline int nd_alloc_stack(struct nameidata
*nd
)
582 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
584 if (likely(nd
->stack
!= nd
->internal
))
586 return __nd_alloc_stack(nd
);
589 static void drop_links(struct nameidata
*nd
)
593 struct saved
*last
= nd
->stack
+ i
;
594 do_delayed_call(&last
->done
);
595 clear_delayed_call(&last
->done
);
599 static void terminate_walk(struct nameidata
*nd
)
602 if (!(nd
->flags
& LOOKUP_RCU
)) {
605 for (i
= 0; i
< nd
->depth
; i
++)
606 path_put(&nd
->stack
[i
].link
);
607 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
612 nd
->flags
&= ~LOOKUP_RCU
;
613 if (!(nd
->flags
& LOOKUP_ROOT
))
620 /* path_put is needed afterwards regardless of success or failure */
621 static bool legitimize_path(struct nameidata
*nd
,
622 struct path
*path
, unsigned seq
)
624 int res
= __legitimize_mnt(path
->mnt
, nd
->m_seq
);
631 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
635 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
638 static bool legitimize_links(struct nameidata
*nd
)
641 for (i
= 0; i
< nd
->depth
; i
++) {
642 struct saved
*last
= nd
->stack
+ i
;
643 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
653 * Path walking has 2 modes, rcu-walk and ref-walk (see
654 * Documentation/filesystems/path-lookup.txt). In situations when we can't
655 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
656 * normal reference counts on dentries and vfsmounts to transition to ref-walk
657 * mode. Refcounts are grabbed at the last known good point before rcu-walk
658 * got stuck, so ref-walk may continue from there. If this is not successful
659 * (eg. a seqcount has changed), then failure is returned and it's up to caller
660 * to restart the path walk from the beginning in ref-walk mode.
664 * unlazy_walk - try to switch to ref-walk mode.
665 * @nd: nameidata pathwalk data
666 * @dentry: child of nd->path.dentry or NULL
667 * @seq: seq number to check dentry against
668 * Returns: 0 on success, -ECHILD on failure
670 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
671 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
672 * @nd or NULL. Must be called from rcu-walk context.
673 * Nothing should touch nameidata between unlazy_walk() failure and
676 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
, unsigned seq
)
678 struct dentry
*parent
= nd
->path
.dentry
;
680 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
682 nd
->flags
&= ~LOOKUP_RCU
;
683 if (unlikely(!legitimize_links(nd
)))
685 if (unlikely(!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)))
687 if (unlikely(!lockref_get_not_dead(&parent
->d_lockref
)))
691 * For a negative lookup, the lookup sequence point is the parents
692 * sequence point, and it only needs to revalidate the parent dentry.
694 * For a positive lookup, we need to move both the parent and the
695 * dentry from the RCU domain to be properly refcounted. And the
696 * sequence number in the dentry validates *both* dentry counters,
697 * since we checked the sequence number of the parent after we got
698 * the child sequence number. So we know the parent must still
699 * be valid if the child sequence number is still valid.
702 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
704 BUG_ON(nd
->inode
!= parent
->d_inode
);
706 if (!lockref_get_not_dead(&dentry
->d_lockref
))
708 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
713 * Sequence counts matched. Now make sure that the root is
714 * still valid and get it if required.
716 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
717 if (unlikely(!legitimize_path(nd
, &nd
->root
, nd
->root_seq
))) {
734 nd
->path
.dentry
= NULL
;
738 if (!(nd
->flags
& LOOKUP_ROOT
))
743 static int unlazy_link(struct nameidata
*nd
, struct path
*link
, unsigned seq
)
745 if (unlikely(!legitimize_path(nd
, link
, seq
))) {
748 nd
->flags
&= ~LOOKUP_RCU
;
750 nd
->path
.dentry
= NULL
;
751 if (!(nd
->flags
& LOOKUP_ROOT
))
754 } else if (likely(unlazy_walk(nd
, NULL
, 0)) == 0) {
761 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
763 return dentry
->d_op
->d_revalidate(dentry
, flags
);
767 * complete_walk - successful completion of path walk
768 * @nd: pointer nameidata
770 * If we had been in RCU mode, drop out of it and legitimize nd->path.
771 * Revalidate the final result, unless we'd already done that during
772 * the path walk or the filesystem doesn't ask for it. Return 0 on
773 * success, -error on failure. In case of failure caller does not
774 * need to drop nd->path.
776 static int complete_walk(struct nameidata
*nd
)
778 struct dentry
*dentry
= nd
->path
.dentry
;
781 if (nd
->flags
& LOOKUP_RCU
) {
782 if (!(nd
->flags
& LOOKUP_ROOT
))
784 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
788 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
791 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
794 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
804 static void set_root(struct nameidata
*nd
)
806 struct fs_struct
*fs
= current
->fs
;
808 if (nd
->flags
& LOOKUP_RCU
) {
812 seq
= read_seqcount_begin(&fs
->seq
);
814 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
815 } while (read_seqcount_retry(&fs
->seq
, seq
));
817 get_fs_root(fs
, &nd
->root
);
821 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
824 if (path
->mnt
!= nd
->path
.mnt
)
828 static inline void path_to_nameidata(const struct path
*path
,
829 struct nameidata
*nd
)
831 if (!(nd
->flags
& LOOKUP_RCU
)) {
832 dput(nd
->path
.dentry
);
833 if (nd
->path
.mnt
!= path
->mnt
)
834 mntput(nd
->path
.mnt
);
836 nd
->path
.mnt
= path
->mnt
;
837 nd
->path
.dentry
= path
->dentry
;
840 static int nd_jump_root(struct nameidata
*nd
)
842 if (nd
->flags
& LOOKUP_RCU
) {
846 nd
->inode
= d
->d_inode
;
847 nd
->seq
= nd
->root_seq
;
848 if (unlikely(read_seqcount_retry(&d
->d_seq
, nd
->seq
)))
854 nd
->inode
= nd
->path
.dentry
->d_inode
;
856 nd
->flags
|= LOOKUP_JUMPED
;
861 * Helper to directly jump to a known parsed path from ->get_link,
862 * caller must have taken a reference to path beforehand.
864 void nd_jump_link(struct path
*path
)
866 struct nameidata
*nd
= current
->nameidata
;
870 nd
->inode
= nd
->path
.dentry
->d_inode
;
871 nd
->flags
|= LOOKUP_JUMPED
;
874 static inline void put_link(struct nameidata
*nd
)
876 struct saved
*last
= nd
->stack
+ --nd
->depth
;
877 do_delayed_call(&last
->done
);
878 if (!(nd
->flags
& LOOKUP_RCU
))
879 path_put(&last
->link
);
882 int sysctl_protected_symlinks __read_mostly
= 0;
883 int sysctl_protected_hardlinks __read_mostly
= 0;
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
898 * Returns 0 if following the symlink is allowed, -ve on error.
900 static inline int may_follow_link(struct nameidata
*nd
)
902 const struct inode
*inode
;
903 const struct inode
*parent
;
905 if (!sysctl_protected_symlinks
)
908 /* Allowed if owner and follower match. */
909 inode
= nd
->link_inode
;
910 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
913 /* Allowed if parent directory not sticky and world-writable. */
915 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
918 /* Allowed if parent directory and link owner match. */
919 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
922 if (nd
->flags
& LOOKUP_RCU
)
925 audit_log_link_denied("follow_link", &nd
->stack
[0].link
);
930 * safe_hardlink_source - Check for safe hardlink conditions
931 * @inode: the source inode to hardlink from
933 * Return false if at least one of the following conditions:
934 * - inode is not a regular file
936 * - inode is setgid and group-exec
937 * - access failure for read and write
939 * Otherwise returns true.
941 static bool safe_hardlink_source(struct inode
*inode
)
943 umode_t mode
= inode
->i_mode
;
945 /* Special files should not get pinned to the filesystem. */
949 /* Setuid files should not get pinned to the filesystem. */
953 /* Executable setgid files should not get pinned to the filesystem. */
954 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
957 /* Hardlinking to unreadable or unwritable sources is dangerous. */
958 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
965 * may_linkat - Check permissions for creating a hardlink
966 * @link: the source to hardlink from
968 * Block hardlink when all of:
969 * - sysctl_protected_hardlinks enabled
970 * - fsuid does not match inode
971 * - hardlink source is unsafe (see safe_hardlink_source() above)
972 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
974 * Returns 0 if successful, -ve on error.
976 static int may_linkat(struct path
*link
)
980 if (!sysctl_protected_hardlinks
)
983 inode
= link
->dentry
->d_inode
;
985 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
986 * otherwise, it must be a safe source.
988 if (inode_owner_or_capable(inode
) || safe_hardlink_source(inode
))
991 audit_log_link_denied("linkat", link
);
995 static __always_inline
996 const char *get_link(struct nameidata
*nd
)
998 struct saved
*last
= nd
->stack
+ nd
->depth
- 1;
999 struct dentry
*dentry
= last
->link
.dentry
;
1000 struct inode
*inode
= nd
->link_inode
;
1004 if (!(nd
->flags
& LOOKUP_RCU
)) {
1005 touch_atime(&last
->link
);
1007 } else if (atime_needs_update(&last
->link
, inode
)) {
1008 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1009 return ERR_PTR(-ECHILD
);
1010 touch_atime(&last
->link
);
1013 error
= security_inode_follow_link(dentry
, inode
,
1014 nd
->flags
& LOOKUP_RCU
);
1015 if (unlikely(error
))
1016 return ERR_PTR(error
);
1018 nd
->last_type
= LAST_BIND
;
1019 res
= inode
->i_link
;
1021 const char * (*get
)(struct dentry
*, struct inode
*,
1022 struct delayed_call
*);
1023 get
= inode
->i_op
->get_link
;
1024 if (nd
->flags
& LOOKUP_RCU
) {
1025 res
= get(NULL
, inode
, &last
->done
);
1026 if (res
== ERR_PTR(-ECHILD
)) {
1027 if (unlikely(unlazy_walk(nd
, NULL
, 0)))
1028 return ERR_PTR(-ECHILD
);
1029 res
= get(dentry
, inode
, &last
->done
);
1032 res
= get(dentry
, inode
, &last
->done
);
1034 if (IS_ERR_OR_NULL(res
))
1040 if (unlikely(nd_jump_root(nd
)))
1041 return ERR_PTR(-ECHILD
);
1042 while (unlikely(*++res
== '/'))
1051 * follow_up - Find the mountpoint of path's vfsmount
1053 * Given a path, find the mountpoint of its source file system.
1054 * Replace @path with the path of the mountpoint in the parent mount.
1057 * Return 1 if we went up a level and 0 if we were already at the
1060 int follow_up(struct path
*path
)
1062 struct mount
*mnt
= real_mount(path
->mnt
);
1063 struct mount
*parent
;
1064 struct dentry
*mountpoint
;
1066 read_seqlock_excl(&mount_lock
);
1067 parent
= mnt
->mnt_parent
;
1068 if (parent
== mnt
) {
1069 read_sequnlock_excl(&mount_lock
);
1072 mntget(&parent
->mnt
);
1073 mountpoint
= dget(mnt
->mnt_mountpoint
);
1074 read_sequnlock_excl(&mount_lock
);
1076 path
->dentry
= mountpoint
;
1078 path
->mnt
= &parent
->mnt
;
1081 EXPORT_SYMBOL(follow_up
);
1084 * Perform an automount
1085 * - return -EISDIR to tell follow_managed() to stop and return the path we
1088 static int follow_automount(struct path
*path
, struct nameidata
*nd
,
1091 struct vfsmount
*mnt
;
1094 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
1097 /* We don't want to mount if someone's just doing a stat -
1098 * unless they're stat'ing a directory and appended a '/' to
1101 * We do, however, want to mount if someone wants to open or
1102 * create a file of any type under the mountpoint, wants to
1103 * traverse through the mountpoint or wants to open the
1104 * mounted directory. Also, autofs may mark negative dentries
1105 * as being automount points. These will need the attentions
1106 * of the daemon to instantiate them before they can be used.
1108 if (!(nd
->flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1109 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1110 path
->dentry
->d_inode
)
1113 nd
->total_link_count
++;
1114 if (nd
->total_link_count
>= 40)
1117 mnt
= path
->dentry
->d_op
->d_automount(path
);
1120 * The filesystem is allowed to return -EISDIR here to indicate
1121 * it doesn't want to automount. For instance, autofs would do
1122 * this so that its userspace daemon can mount on this dentry.
1124 * However, we can only permit this if it's a terminal point in
1125 * the path being looked up; if it wasn't then the remainder of
1126 * the path is inaccessible and we should say so.
1128 if (PTR_ERR(mnt
) == -EISDIR
&& (nd
->flags
& LOOKUP_PARENT
))
1130 return PTR_ERR(mnt
);
1133 if (!mnt
) /* mount collision */
1136 if (!*need_mntput
) {
1137 /* lock_mount() may release path->mnt on error */
1139 *need_mntput
= true;
1141 err
= finish_automount(mnt
, path
);
1145 /* Someone else made a mount here whilst we were busy */
1150 path
->dentry
= dget(mnt
->mnt_root
);
1159 * Handle a dentry that is managed in some way.
1160 * - Flagged for transit management (autofs)
1161 * - Flagged as mountpoint
1162 * - Flagged as automount point
1164 * This may only be called in refwalk mode.
1166 * Serialization is taken care of in namespace.c
1168 static int follow_managed(struct path
*path
, struct nameidata
*nd
)
1170 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1172 bool need_mntput
= false;
1175 /* Given that we're not holding a lock here, we retain the value in a
1176 * local variable for each dentry as we look at it so that we don't see
1177 * the components of that value change under us */
1178 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1179 managed
&= DCACHE_MANAGED_DENTRY
,
1180 unlikely(managed
!= 0)) {
1181 /* Allow the filesystem to manage the transit without i_mutex
1183 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1184 BUG_ON(!path
->dentry
->d_op
);
1185 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1186 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1191 /* Transit to a mounted filesystem. */
1192 if (managed
& DCACHE_MOUNTED
) {
1193 struct vfsmount
*mounted
= lookup_mnt(path
);
1198 path
->mnt
= mounted
;
1199 path
->dentry
= dget(mounted
->mnt_root
);
1204 /* Something is mounted on this dentry in another
1205 * namespace and/or whatever was mounted there in this
1206 * namespace got unmounted before lookup_mnt() could
1210 /* Handle an automount point */
1211 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1212 ret
= follow_automount(path
, nd
, &need_mntput
);
1218 /* We didn't change the current path point */
1222 if (need_mntput
&& path
->mnt
== mnt
)
1224 if (ret
== -EISDIR
|| !ret
)
1227 nd
->flags
|= LOOKUP_JUMPED
;
1228 if (unlikely(ret
< 0))
1229 path_put_conditional(path
, nd
);
1233 int follow_down_one(struct path
*path
)
1235 struct vfsmount
*mounted
;
1237 mounted
= lookup_mnt(path
);
1241 path
->mnt
= mounted
;
1242 path
->dentry
= dget(mounted
->mnt_root
);
1247 EXPORT_SYMBOL(follow_down_one
);
1249 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1251 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1252 dentry
->d_op
->d_manage(dentry
, true) : 0;
1256 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1257 * we meet a managed dentry that would need blocking.
1259 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1260 struct inode
**inode
, unsigned *seqp
)
1263 struct mount
*mounted
;
1265 * Don't forget we might have a non-mountpoint managed dentry
1266 * that wants to block transit.
1268 switch (managed_dentry_rcu(path
->dentry
)) {
1278 if (!d_mountpoint(path
->dentry
))
1279 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1281 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1284 path
->mnt
= &mounted
->mnt
;
1285 path
->dentry
= mounted
->mnt
.mnt_root
;
1286 nd
->flags
|= LOOKUP_JUMPED
;
1287 *seqp
= read_seqcount_begin(&path
->dentry
->d_seq
);
1289 * Update the inode too. We don't need to re-check the
1290 * dentry sequence number here after this d_inode read,
1291 * because a mount-point is always pinned.
1293 *inode
= path
->dentry
->d_inode
;
1295 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1296 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1299 static int follow_dotdot_rcu(struct nameidata
*nd
)
1301 struct inode
*inode
= nd
->inode
;
1304 if (path_equal(&nd
->path
, &nd
->root
))
1306 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1307 struct dentry
*old
= nd
->path
.dentry
;
1308 struct dentry
*parent
= old
->d_parent
;
1311 inode
= parent
->d_inode
;
1312 seq
= read_seqcount_begin(&parent
->d_seq
);
1313 if (unlikely(read_seqcount_retry(&old
->d_seq
, nd
->seq
)))
1315 nd
->path
.dentry
= parent
;
1317 if (unlikely(!path_connected(&nd
->path
)))
1321 struct mount
*mnt
= real_mount(nd
->path
.mnt
);
1322 struct mount
*mparent
= mnt
->mnt_parent
;
1323 struct dentry
*mountpoint
= mnt
->mnt_mountpoint
;
1324 struct inode
*inode2
= mountpoint
->d_inode
;
1325 unsigned seq
= read_seqcount_begin(&mountpoint
->d_seq
);
1326 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1328 if (&mparent
->mnt
== nd
->path
.mnt
)
1330 /* we know that mountpoint was pinned */
1331 nd
->path
.dentry
= mountpoint
;
1332 nd
->path
.mnt
= &mparent
->mnt
;
1337 while (unlikely(d_mountpoint(nd
->path
.dentry
))) {
1338 struct mount
*mounted
;
1339 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1340 if (unlikely(read_seqretry(&mount_lock
, nd
->m_seq
)))
1344 nd
->path
.mnt
= &mounted
->mnt
;
1345 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1346 inode
= nd
->path
.dentry
->d_inode
;
1347 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1354 * Follow down to the covering mount currently visible to userspace. At each
1355 * point, the filesystem owning that dentry may be queried as to whether the
1356 * caller is permitted to proceed or not.
1358 int follow_down(struct path
*path
)
1363 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1364 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1365 /* Allow the filesystem to manage the transit without i_mutex
1368 * We indicate to the filesystem if someone is trying to mount
1369 * something here. This gives autofs the chance to deny anyone
1370 * other than its daemon the right to mount on its
1373 * The filesystem may sleep at this point.
1375 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1376 BUG_ON(!path
->dentry
->d_op
);
1377 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1378 ret
= path
->dentry
->d_op
->d_manage(
1379 path
->dentry
, false);
1381 return ret
== -EISDIR
? 0 : ret
;
1384 /* Transit to a mounted filesystem. */
1385 if (managed
& DCACHE_MOUNTED
) {
1386 struct vfsmount
*mounted
= lookup_mnt(path
);
1391 path
->mnt
= mounted
;
1392 path
->dentry
= dget(mounted
->mnt_root
);
1396 /* Don't handle automount points here */
1401 EXPORT_SYMBOL(follow_down
);
1404 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1406 static void follow_mount(struct path
*path
)
1408 while (d_mountpoint(path
->dentry
)) {
1409 struct vfsmount
*mounted
= lookup_mnt(path
);
1414 path
->mnt
= mounted
;
1415 path
->dentry
= dget(mounted
->mnt_root
);
1419 static int follow_dotdot(struct nameidata
*nd
)
1422 struct dentry
*old
= nd
->path
.dentry
;
1424 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1425 nd
->path
.mnt
== nd
->root
.mnt
) {
1428 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1429 /* rare case of legitimate dget_parent()... */
1430 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1432 if (unlikely(!path_connected(&nd
->path
)))
1436 if (!follow_up(&nd
->path
))
1439 follow_mount(&nd
->path
);
1440 nd
->inode
= nd
->path
.dentry
->d_inode
;
1445 * This looks up the name in dcache, possibly revalidates the old dentry and
1446 * allocates a new one if not found or not valid. In the need_lookup argument
1447 * returns whether i_op->lookup is necessary.
1449 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1453 struct dentry
*dentry
;
1456 dentry
= d_lookup(dir
, name
);
1458 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1459 error
= d_revalidate(dentry
, flags
);
1460 if (unlikely(error
<= 0)) {
1462 d_invalidate(dentry
);
1464 return ERR_PTR(error
);
1472 * Call i_op->lookup on the dentry. The dentry must be negative and
1475 * dir->d_inode->i_mutex must be held
1477 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1482 /* Don't create child dentry for a dead directory. */
1483 if (unlikely(IS_DEADDIR(dir
))) {
1485 return ERR_PTR(-ENOENT
);
1488 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1489 if (unlikely(old
)) {
1496 static struct dentry
*__lookup_hash(const struct qstr
*name
,
1497 struct dentry
*base
, unsigned int flags
)
1499 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1504 dentry
= d_alloc(base
, name
);
1505 if (unlikely(!dentry
))
1506 return ERR_PTR(-ENOMEM
);
1508 return lookup_real(base
->d_inode
, dentry
, flags
);
1511 static int lookup_fast(struct nameidata
*nd
,
1512 struct path
*path
, struct inode
**inode
,
1515 struct vfsmount
*mnt
= nd
->path
.mnt
;
1516 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1521 * Rename seqlock is not required here because in the off chance
1522 * of a false negative due to a concurrent rename, the caller is
1523 * going to fall back to non-racy lookup.
1525 if (nd
->flags
& LOOKUP_RCU
) {
1528 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1529 if (unlikely(!dentry
)) {
1530 if (unlazy_walk(nd
, NULL
, 0))
1536 * This sequence count validates that the inode matches
1537 * the dentry name information from lookup.
1539 *inode
= d_backing_inode(dentry
);
1540 negative
= d_is_negative(dentry
);
1541 if (unlikely(read_seqcount_retry(&dentry
->d_seq
, seq
)))
1545 * This sequence count validates that the parent had no
1546 * changes while we did the lookup of the dentry above.
1548 * The memory barrier in read_seqcount_begin of child is
1549 * enough, we can use __read_seqcount_retry here.
1551 if (unlikely(__read_seqcount_retry(&parent
->d_seq
, nd
->seq
)))
1555 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1556 status
= d_revalidate(dentry
, nd
->flags
);
1557 if (unlikely(status
<= 0)) {
1558 if (unlazy_walk(nd
, dentry
, seq
))
1560 if (status
== -ECHILD
)
1561 status
= d_revalidate(dentry
, nd
->flags
);
1564 * Note: do negative dentry check after revalidation in
1565 * case that drops it.
1567 if (unlikely(negative
))
1570 path
->dentry
= dentry
;
1571 if (likely(__follow_mount_rcu(nd
, path
, inode
, seqp
)))
1573 if (unlazy_walk(nd
, dentry
, seq
))
1577 dentry
= __d_lookup(parent
, &nd
->last
);
1578 if (unlikely(!dentry
))
1580 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1581 status
= d_revalidate(dentry
, nd
->flags
);
1583 if (unlikely(status
<= 0)) {
1585 d_invalidate(dentry
);
1589 if (unlikely(d_is_negative(dentry
))) {
1595 path
->dentry
= dentry
;
1596 err
= follow_managed(path
, nd
);
1597 if (likely(err
> 0))
1598 *inode
= d_backing_inode(path
->dentry
);
1602 /* Fast lookup failed, do it the slow way */
1603 static struct dentry
*lookup_slow(const struct qstr
*name
,
1607 struct dentry
*dentry
= ERR_PTR(-ENOENT
), *old
;
1608 struct inode
*inode
= dir
->d_inode
;
1609 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1611 inode_lock_shared(inode
);
1612 /* Don't go there if it's already dead */
1613 if (unlikely(IS_DEADDIR(inode
)))
1616 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1619 if (unlikely(!d_in_lookup(dentry
))) {
1620 if ((dentry
->d_flags
& DCACHE_OP_REVALIDATE
) &&
1621 !(flags
& LOOKUP_NO_REVAL
)) {
1622 int error
= d_revalidate(dentry
, flags
);
1623 if (unlikely(error
<= 0)) {
1625 d_invalidate(dentry
);
1630 dentry
= ERR_PTR(error
);
1634 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1635 d_lookup_done(dentry
);
1636 if (unlikely(old
)) {
1642 inode_unlock_shared(inode
);
1646 static inline int may_lookup(struct nameidata
*nd
)
1648 if (nd
->flags
& LOOKUP_RCU
) {
1649 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1652 if (unlazy_walk(nd
, NULL
, 0))
1655 return inode_permission(nd
->inode
, MAY_EXEC
);
1658 static inline int handle_dots(struct nameidata
*nd
, int type
)
1660 if (type
== LAST_DOTDOT
) {
1663 if (nd
->flags
& LOOKUP_RCU
) {
1664 return follow_dotdot_rcu(nd
);
1666 return follow_dotdot(nd
);
1671 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1672 struct inode
*inode
, unsigned seq
)
1676 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1677 path_to_nameidata(link
, nd
);
1680 if (!(nd
->flags
& LOOKUP_RCU
)) {
1681 if (link
->mnt
== nd
->path
.mnt
)
1684 error
= nd_alloc_stack(nd
);
1685 if (unlikely(error
)) {
1686 if (error
== -ECHILD
) {
1687 if (unlikely(unlazy_link(nd
, link
, seq
)))
1689 error
= nd_alloc_stack(nd
);
1697 last
= nd
->stack
+ nd
->depth
++;
1699 clear_delayed_call(&last
->done
);
1700 nd
->link_inode
= inode
;
1706 * Do we need to follow links? We _really_ want to be able
1707 * to do this check without having to look at inode->i_op,
1708 * so we keep a cache of "no, this doesn't need follow_link"
1709 * for the common case.
1711 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
,
1713 struct inode
*inode
, unsigned seq
)
1715 if (likely(!d_is_symlink(link
->dentry
)))
1719 /* make sure that d_is_symlink above matches inode */
1720 if (nd
->flags
& LOOKUP_RCU
) {
1721 if (read_seqcount_retry(&link
->dentry
->d_seq
, seq
))
1724 return pick_link(nd
, link
, inode
, seq
);
1727 enum {WALK_GET
= 1, WALK_PUT
= 2};
1729 static int walk_component(struct nameidata
*nd
, int flags
)
1732 struct inode
*inode
;
1736 * "." and ".." are special - ".." especially so because it has
1737 * to be able to know about the current root directory and
1738 * parent relationships.
1740 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1741 err
= handle_dots(nd
, nd
->last_type
);
1742 if (flags
& WALK_PUT
)
1746 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1747 if (unlikely(err
<= 0)) {
1750 path
.dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
,
1752 if (IS_ERR(path
.dentry
))
1753 return PTR_ERR(path
.dentry
);
1755 path
.mnt
= nd
->path
.mnt
;
1756 err
= follow_managed(&path
, nd
);
1757 if (unlikely(err
< 0))
1760 if (unlikely(d_is_negative(path
.dentry
))) {
1761 path_to_nameidata(&path
, nd
);
1765 seq
= 0; /* we are already out of RCU mode */
1766 inode
= d_backing_inode(path
.dentry
);
1769 if (flags
& WALK_PUT
)
1771 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
, inode
, seq
);
1774 path_to_nameidata(&path
, nd
);
1781 * We can do the critical dentry name comparison and hashing
1782 * operations one word at a time, but we are limited to:
1784 * - Architectures with fast unaligned word accesses. We could
1785 * do a "get_unaligned()" if this helps and is sufficiently
1788 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1789 * do not trap on the (extremely unlikely) case of a page
1790 * crossing operation.
1792 * - Furthermore, we need an efficient 64-bit compile for the
1793 * 64-bit case in order to generate the "number of bytes in
1794 * the final mask". Again, that could be replaced with a
1795 * efficient population count instruction or similar.
1797 #ifdef CONFIG_DCACHE_WORD_ACCESS
1799 #include <asm/word-at-a-time.h>
1803 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1805 #elif defined(CONFIG_64BIT)
1807 * Register pressure in the mixing function is an issue, particularly
1808 * on 32-bit x86, but almost any function requires one state value and
1809 * one temporary. Instead, use a function designed for two state values
1810 * and no temporaries.
1812 * This function cannot create a collision in only two iterations, so
1813 * we have two iterations to achieve avalanche. In those two iterations,
1814 * we have six layers of mixing, which is enough to spread one bit's
1815 * influence out to 2^6 = 64 state bits.
1817 * Rotate constants are scored by considering either 64 one-bit input
1818 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1819 * probability of that delta causing a change to each of the 128 output
1820 * bits, using a sample of random initial states.
1822 * The Shannon entropy of the computed probabilities is then summed
1823 * to produce a score. Ideally, any input change has a 50% chance of
1824 * toggling any given output bit.
1826 * Mixing scores (in bits) for (12,45):
1827 * Input delta: 1-bit 2-bit
1828 * 1 round: 713.3 42542.6
1829 * 2 rounds: 2753.7 140389.8
1830 * 3 rounds: 5954.1 233458.2
1831 * 4 rounds: 7862.6 256672.2
1832 * Perfect: 8192 258048
1833 * (64*128) (64*63/2 * 128)
1835 #define HASH_MIX(x, y, a) \
1837 y ^= x, x = rol64(x,12),\
1838 x += y, y = rol64(y,45),\
1842 * Fold two longs into one 32-bit hash value. This must be fast, but
1843 * latency isn't quite as critical, as there is a fair bit of additional
1844 * work done before the hash value is used.
1846 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1848 y
^= x
* GOLDEN_RATIO_64
;
1849 y
*= GOLDEN_RATIO_64
;
1853 #else /* 32-bit case */
1856 * Mixing scores (in bits) for (7,20):
1857 * Input delta: 1-bit 2-bit
1858 * 1 round: 330.3 9201.6
1859 * 2 rounds: 1246.4 25475.4
1860 * 3 rounds: 1907.1 31295.1
1861 * 4 rounds: 2042.3 31718.6
1862 * Perfect: 2048 31744
1863 * (32*64) (32*31/2 * 64)
1865 #define HASH_MIX(x, y, a) \
1867 y ^= x, x = rol32(x, 7),\
1868 x += y, y = rol32(y,20),\
1871 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1873 /* Use arch-optimized multiply if one exists */
1874 return __hash_32(y
^ __hash_32(x
));
1880 * Return the hash of a string of known length. This is carfully
1881 * designed to match hash_name(), which is the more critical function.
1882 * In particular, we must end by hashing a final word containing 0..7
1883 * payload bytes, to match the way that hash_name() iterates until it
1884 * finds the delimiter after the name.
1886 unsigned int full_name_hash(const char *name
, unsigned int len
)
1888 unsigned long a
, x
= 0, y
= 0;
1893 a
= load_unaligned_zeropad(name
);
1894 if (len
< sizeof(unsigned long))
1897 name
+= sizeof(unsigned long);
1898 len
-= sizeof(unsigned long);
1900 x
^= a
& bytemask_from_count(len
);
1902 return fold_hash(x
, y
);
1904 EXPORT_SYMBOL(full_name_hash
);
1906 /* Return the "hash_len" (hash and length) of a null-terminated string */
1907 u64
hashlen_string(const char *name
)
1909 unsigned long a
= 0, x
= 0, y
= 0, adata
, mask
, len
;
1910 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1912 len
= -sizeof(unsigned long);
1915 len
+= sizeof(unsigned long);
1916 a
= load_unaligned_zeropad(name
+len
);
1917 } while (!has_zero(a
, &adata
, &constants
));
1919 adata
= prep_zero_mask(a
, adata
, &constants
);
1920 mask
= create_zero_mask(adata
);
1921 x
^= a
& zero_bytemask(mask
);
1923 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1925 EXPORT_SYMBOL(hashlen_string
);
1928 * Calculate the length and hash of the path component, and
1929 * return the "hash_len" as the result.
1931 static inline u64
hash_name(const char *name
)
1933 unsigned long a
= 0, b
, x
= 0, y
= 0, adata
, bdata
, mask
, len
;
1934 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1936 len
= -sizeof(unsigned long);
1939 len
+= sizeof(unsigned long);
1940 a
= load_unaligned_zeropad(name
+len
);
1941 b
= a
^ REPEAT_BYTE('/');
1942 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1944 adata
= prep_zero_mask(a
, adata
, &constants
);
1945 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1946 mask
= create_zero_mask(adata
| bdata
);
1947 x
^= a
& zero_bytemask(mask
);
1949 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1952 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1954 /* Return the hash of a string of known length */
1955 unsigned int full_name_hash(const char *name
, unsigned int len
)
1957 unsigned long hash
= init_name_hash();
1959 hash
= partial_name_hash((unsigned char)*name
++, hash
);
1960 return end_name_hash(hash
);
1962 EXPORT_SYMBOL(full_name_hash
);
1964 /* Return the "hash_len" (hash and length) of a null-terminated string */
1965 u64
hashlen_string(const char *name
)
1967 unsigned long hash
= init_name_hash();
1968 unsigned long len
= 0, c
;
1970 c
= (unsigned char)*name
;
1973 hash
= partial_name_hash(c
, hash
);
1974 c
= (unsigned char)name
[len
];
1976 return hashlen_create(end_name_hash(hash
), len
);
1978 EXPORT_SYMBOL(hashlen_string
);
1981 * We know there's a real path component here of at least
1984 static inline u64
hash_name(const char *name
)
1986 unsigned long hash
= init_name_hash();
1987 unsigned long len
= 0, c
;
1989 c
= (unsigned char)*name
;
1992 hash
= partial_name_hash(c
, hash
);
1993 c
= (unsigned char)name
[len
];
1994 } while (c
&& c
!= '/');
1995 return hashlen_create(end_name_hash(hash
), len
);
2002 * This is the basic name resolution function, turning a pathname into
2003 * the final dentry. We expect 'base' to be positive and a directory.
2005 * Returns 0 and nd will have valid dentry and mnt on success.
2006 * Returns error and drops reference to input namei data on failure.
2008 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2017 /* At this point we know we have a real path component. */
2022 err
= may_lookup(nd
);
2026 hash_len
= hash_name(name
);
2029 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2031 if (name
[1] == '.') {
2033 nd
->flags
|= LOOKUP_JUMPED
;
2039 if (likely(type
== LAST_NORM
)) {
2040 struct dentry
*parent
= nd
->path
.dentry
;
2041 nd
->flags
&= ~LOOKUP_JUMPED
;
2042 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2043 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2044 err
= parent
->d_op
->d_hash(parent
, &this);
2047 hash_len
= this.hash_len
;
2052 nd
->last
.hash_len
= hash_len
;
2053 nd
->last
.name
= name
;
2054 nd
->last_type
= type
;
2056 name
+= hashlen_len(hash_len
);
2060 * If it wasn't NUL, we know it was '/'. Skip that
2061 * slash, and continue until no more slashes.
2065 } while (unlikely(*name
== '/'));
2066 if (unlikely(!*name
)) {
2068 /* pathname body, done */
2071 name
= nd
->stack
[nd
->depth
- 1].name
;
2072 /* trailing symlink, done */
2075 /* last component of nested symlink */
2076 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
2078 err
= walk_component(nd
, WALK_GET
);
2084 const char *s
= get_link(nd
);
2093 nd
->stack
[nd
->depth
- 1].name
= name
;
2098 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2099 if (nd
->flags
& LOOKUP_RCU
) {
2100 if (unlazy_walk(nd
, NULL
, 0))
2108 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2111 const char *s
= nd
->name
->name
;
2113 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
2114 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
2116 if (flags
& LOOKUP_ROOT
) {
2117 struct dentry
*root
= nd
->root
.dentry
;
2118 struct inode
*inode
= root
->d_inode
;
2120 if (!d_can_lookup(root
))
2121 return ERR_PTR(-ENOTDIR
);
2122 retval
= inode_permission(inode
, MAY_EXEC
);
2124 return ERR_PTR(retval
);
2126 nd
->path
= nd
->root
;
2128 if (flags
& LOOKUP_RCU
) {
2130 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2131 nd
->root_seq
= nd
->seq
;
2132 nd
->m_seq
= read_seqbegin(&mount_lock
);
2134 path_get(&nd
->path
);
2139 nd
->root
.mnt
= NULL
;
2140 nd
->path
.mnt
= NULL
;
2141 nd
->path
.dentry
= NULL
;
2143 nd
->m_seq
= read_seqbegin(&mount_lock
);
2145 if (flags
& LOOKUP_RCU
)
2148 if (likely(!nd_jump_root(nd
)))
2150 nd
->root
.mnt
= NULL
;
2152 return ERR_PTR(-ECHILD
);
2153 } else if (nd
->dfd
== AT_FDCWD
) {
2154 if (flags
& LOOKUP_RCU
) {
2155 struct fs_struct
*fs
= current
->fs
;
2161 seq
= read_seqcount_begin(&fs
->seq
);
2163 nd
->inode
= nd
->path
.dentry
->d_inode
;
2164 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2165 } while (read_seqcount_retry(&fs
->seq
, seq
));
2167 get_fs_pwd(current
->fs
, &nd
->path
);
2168 nd
->inode
= nd
->path
.dentry
->d_inode
;
2172 /* Caller must check execute permissions on the starting path component */
2173 struct fd f
= fdget_raw(nd
->dfd
);
2174 struct dentry
*dentry
;
2177 return ERR_PTR(-EBADF
);
2179 dentry
= f
.file
->f_path
.dentry
;
2182 if (!d_can_lookup(dentry
)) {
2184 return ERR_PTR(-ENOTDIR
);
2188 nd
->path
= f
.file
->f_path
;
2189 if (flags
& LOOKUP_RCU
) {
2191 nd
->inode
= nd
->path
.dentry
->d_inode
;
2192 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2194 path_get(&nd
->path
);
2195 nd
->inode
= nd
->path
.dentry
->d_inode
;
2202 static const char *trailing_symlink(struct nameidata
*nd
)
2205 int error
= may_follow_link(nd
);
2206 if (unlikely(error
))
2207 return ERR_PTR(error
);
2208 nd
->flags
|= LOOKUP_PARENT
;
2209 nd
->stack
[0].name
= NULL
;
2214 static inline int lookup_last(struct nameidata
*nd
)
2216 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2217 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2219 nd
->flags
&= ~LOOKUP_PARENT
;
2220 return walk_component(nd
,
2221 nd
->flags
& LOOKUP_FOLLOW
2223 ? WALK_PUT
| WALK_GET
2228 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2229 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2231 const char *s
= path_init(nd
, flags
);
2236 while (!(err
= link_path_walk(s
, nd
))
2237 && ((err
= lookup_last(nd
)) > 0)) {
2238 s
= trailing_symlink(nd
);
2245 err
= complete_walk(nd
);
2247 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2248 if (!d_can_lookup(nd
->path
.dentry
))
2252 nd
->path
.mnt
= NULL
;
2253 nd
->path
.dentry
= NULL
;
2259 static int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2260 struct path
*path
, struct path
*root
)
2263 struct nameidata nd
;
2265 return PTR_ERR(name
);
2266 if (unlikely(root
)) {
2268 flags
|= LOOKUP_ROOT
;
2270 set_nameidata(&nd
, dfd
, name
);
2271 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2272 if (unlikely(retval
== -ECHILD
))
2273 retval
= path_lookupat(&nd
, flags
, path
);
2274 if (unlikely(retval
== -ESTALE
))
2275 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2277 if (likely(!retval
))
2278 audit_inode(name
, path
->dentry
, flags
& LOOKUP_PARENT
);
2279 restore_nameidata();
2284 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2285 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2286 struct path
*parent
)
2288 const char *s
= path_init(nd
, flags
);
2292 err
= link_path_walk(s
, nd
);
2294 err
= complete_walk(nd
);
2297 nd
->path
.mnt
= NULL
;
2298 nd
->path
.dentry
= NULL
;
2304 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2305 unsigned int flags
, struct path
*parent
,
2306 struct qstr
*last
, int *type
)
2309 struct nameidata nd
;
2313 set_nameidata(&nd
, dfd
, name
);
2314 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2315 if (unlikely(retval
== -ECHILD
))
2316 retval
= path_parentat(&nd
, flags
, parent
);
2317 if (unlikely(retval
== -ESTALE
))
2318 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2319 if (likely(!retval
)) {
2321 *type
= nd
.last_type
;
2322 audit_inode(name
, parent
->dentry
, LOOKUP_PARENT
);
2325 name
= ERR_PTR(retval
);
2327 restore_nameidata();
2331 /* does lookup, returns the object with parent locked */
2332 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2334 struct filename
*filename
;
2339 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2341 if (IS_ERR(filename
))
2342 return ERR_CAST(filename
);
2343 if (unlikely(type
!= LAST_NORM
)) {
2346 return ERR_PTR(-EINVAL
);
2348 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2349 d
= __lookup_hash(&last
, path
->dentry
, 0);
2351 inode_unlock(path
->dentry
->d_inode
);
2358 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2360 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2363 EXPORT_SYMBOL(kern_path
);
2366 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2367 * @dentry: pointer to dentry of the base directory
2368 * @mnt: pointer to vfs mount of the base directory
2369 * @name: pointer to file name
2370 * @flags: lookup flags
2371 * @path: pointer to struct path to fill
2373 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2374 const char *name
, unsigned int flags
,
2377 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2378 /* the first argument of filename_lookup() is ignored with root */
2379 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2380 flags
, path
, &root
);
2382 EXPORT_SYMBOL(vfs_path_lookup
);
2385 * lookup_hash - lookup single pathname component on already hashed name
2386 * @name: name and hash to lookup
2387 * @base: base directory to lookup from
2389 * The name must have been verified and hashed (see lookup_one_len()). Using
2390 * this after just full_name_hash() is unsafe.
2392 * This function also doesn't check for search permission on base directory.
2394 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2397 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2399 struct dentry
*lookup_hash(const struct qstr
*name
, struct dentry
*base
)
2403 ret
= lookup_dcache(name
, base
, 0);
2405 ret
= lookup_slow(name
, base
, 0);
2409 EXPORT_SYMBOL(lookup_hash
);
2412 * lookup_one_len - filesystem helper to lookup single pathname component
2413 * @name: pathname component to lookup
2414 * @base: base directory to lookup from
2415 * @len: maximum length @len should be interpreted to
2417 * Note that this routine is purely a helper for filesystem usage and should
2418 * not be called by generic code.
2420 * The caller must hold base->i_mutex.
2422 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2428 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2432 this.hash
= full_name_hash(name
, len
);
2434 return ERR_PTR(-EACCES
);
2436 if (unlikely(name
[0] == '.')) {
2437 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2438 return ERR_PTR(-EACCES
);
2442 c
= *(const unsigned char *)name
++;
2443 if (c
== '/' || c
== '\0')
2444 return ERR_PTR(-EACCES
);
2447 * See if the low-level filesystem might want
2448 * to use its own hash..
2450 if (base
->d_flags
& DCACHE_OP_HASH
) {
2451 int err
= base
->d_op
->d_hash(base
, &this);
2453 return ERR_PTR(err
);
2456 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2458 return ERR_PTR(err
);
2460 return __lookup_hash(&this, base
, 0);
2462 EXPORT_SYMBOL(lookup_one_len
);
2465 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2466 * @name: pathname component to lookup
2467 * @base: base directory to lookup from
2468 * @len: maximum length @len should be interpreted to
2470 * Note that this routine is purely a helper for filesystem usage and should
2471 * not be called by generic code.
2473 * Unlike lookup_one_len, it should be called without the parent
2474 * i_mutex held, and will take the i_mutex itself if necessary.
2476 struct dentry
*lookup_one_len_unlocked(const char *name
,
2477 struct dentry
*base
, int len
)
2485 this.hash
= full_name_hash(name
, len
);
2487 return ERR_PTR(-EACCES
);
2489 if (unlikely(name
[0] == '.')) {
2490 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2491 return ERR_PTR(-EACCES
);
2495 c
= *(const unsigned char *)name
++;
2496 if (c
== '/' || c
== '\0')
2497 return ERR_PTR(-EACCES
);
2500 * See if the low-level filesystem might want
2501 * to use its own hash..
2503 if (base
->d_flags
& DCACHE_OP_HASH
) {
2504 int err
= base
->d_op
->d_hash(base
, &this);
2506 return ERR_PTR(err
);
2509 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2511 return ERR_PTR(err
);
2513 return lookup_hash(&this, base
);
2515 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2517 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2518 struct path
*path
, int *empty
)
2520 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2523 EXPORT_SYMBOL(user_path_at_empty
);
2526 * NB: most callers don't do anything directly with the reference to the
2527 * to struct filename, but the nd->last pointer points into the name string
2528 * allocated by getname. So we must hold the reference to it until all
2529 * path-walking is complete.
2531 static inline struct filename
*
2532 user_path_parent(int dfd
, const char __user
*path
,
2533 struct path
*parent
,
2538 /* only LOOKUP_REVAL is allowed in extra flags */
2539 return filename_parentat(dfd
, getname(path
), flags
& LOOKUP_REVAL
,
2540 parent
, last
, type
);
2544 * mountpoint_last - look up last component for umount
2545 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2546 * @path: pointer to container for result
2548 * This is a special lookup_last function just for umount. In this case, we
2549 * need to resolve the path without doing any revalidation.
2551 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2552 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2553 * in almost all cases, this lookup will be served out of the dcache. The only
2554 * cases where it won't are if nd->last refers to a symlink or the path is
2555 * bogus and it doesn't exist.
2558 * -error: if there was an error during lookup. This includes -ENOENT if the
2559 * lookup found a negative dentry. The nd->path reference will also be
2562 * 0: if we successfully resolved nd->path and found it to not to be a
2563 * symlink that needs to be followed. "path" will also be populated.
2564 * The nd->path reference will also be put.
2566 * 1: if we successfully resolved nd->last and found it to be a symlink
2567 * that needs to be followed. "path" will be populated with the path
2568 * to the link, and nd->path will *not* be put.
2571 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2574 struct dentry
*dentry
;
2575 struct dentry
*dir
= nd
->path
.dentry
;
2577 /* If we're in rcuwalk, drop out of it to handle last component */
2578 if (nd
->flags
& LOOKUP_RCU
) {
2579 if (unlazy_walk(nd
, NULL
, 0))
2583 nd
->flags
&= ~LOOKUP_PARENT
;
2585 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2586 error
= handle_dots(nd
, nd
->last_type
);
2589 dentry
= dget(nd
->path
.dentry
);
2591 dentry
= d_lookup(dir
, &nd
->last
);
2594 * No cached dentry. Mounted dentries are pinned in the
2595 * cache, so that means that this dentry is probably
2596 * a symlink or the path doesn't actually point
2597 * to a mounted dentry.
2599 dentry
= lookup_slow(&nd
->last
, dir
,
2600 nd
->flags
| LOOKUP_NO_REVAL
);
2602 return PTR_ERR(dentry
);
2605 if (d_is_negative(dentry
)) {
2611 path
->dentry
= dentry
;
2612 path
->mnt
= nd
->path
.mnt
;
2613 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
,
2614 d_backing_inode(dentry
), 0);
2615 if (unlikely(error
))
2623 * path_mountpoint - look up a path to be umounted
2624 * @nd: lookup context
2625 * @flags: lookup flags
2626 * @path: pointer to container for result
2628 * Look up the given name, but don't attempt to revalidate the last component.
2629 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2632 path_mountpoint(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2634 const char *s
= path_init(nd
, flags
);
2638 while (!(err
= link_path_walk(s
, nd
)) &&
2639 (err
= mountpoint_last(nd
, path
)) > 0) {
2640 s
= trailing_symlink(nd
);
2651 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2654 struct nameidata nd
;
2657 return PTR_ERR(name
);
2658 set_nameidata(&nd
, dfd
, name
);
2659 error
= path_mountpoint(&nd
, flags
| LOOKUP_RCU
, path
);
2660 if (unlikely(error
== -ECHILD
))
2661 error
= path_mountpoint(&nd
, flags
, path
);
2662 if (unlikely(error
== -ESTALE
))
2663 error
= path_mountpoint(&nd
, flags
| LOOKUP_REVAL
, path
);
2665 audit_inode(name
, path
->dentry
, 0);
2666 restore_nameidata();
2672 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2673 * @dfd: directory file descriptor
2674 * @name: pathname from userland
2675 * @flags: lookup flags
2676 * @path: pointer to container to hold result
2678 * A umount is a special case for path walking. We're not actually interested
2679 * in the inode in this situation, and ESTALE errors can be a problem. We
2680 * simply want track down the dentry and vfsmount attached at the mountpoint
2681 * and avoid revalidating the last component.
2683 * Returns 0 and populates "path" on success.
2686 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2689 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2693 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2696 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2698 EXPORT_SYMBOL(kern_path_mountpoint
);
2700 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2702 kuid_t fsuid
= current_fsuid();
2704 if (uid_eq(inode
->i_uid
, fsuid
))
2706 if (uid_eq(dir
->i_uid
, fsuid
))
2708 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2710 EXPORT_SYMBOL(__check_sticky
);
2713 * Check whether we can remove a link victim from directory dir, check
2714 * whether the type of victim is right.
2715 * 1. We can't do it if dir is read-only (done in permission())
2716 * 2. We should have write and exec permissions on dir
2717 * 3. We can't remove anything from append-only dir
2718 * 4. We can't do anything with immutable dir (done in permission())
2719 * 5. If the sticky bit on dir is set we should either
2720 * a. be owner of dir, or
2721 * b. be owner of victim, or
2722 * c. have CAP_FOWNER capability
2723 * 6. If the victim is append-only or immutable we can't do antyhing with
2724 * links pointing to it.
2725 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2726 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2727 * 9. We can't remove a root or mountpoint.
2728 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2729 * nfs_async_unlink().
2731 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2733 struct inode
*inode
= d_backing_inode(victim
);
2736 if (d_is_negative(victim
))
2740 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2741 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2743 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2749 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2750 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2753 if (!d_is_dir(victim
))
2755 if (IS_ROOT(victim
))
2757 } else if (d_is_dir(victim
))
2759 if (IS_DEADDIR(dir
))
2761 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2766 /* Check whether we can create an object with dentry child in directory
2768 * 1. We can't do it if child already exists (open has special treatment for
2769 * this case, but since we are inlined it's OK)
2770 * 2. We can't do it if dir is read-only (done in permission())
2771 * 3. We should have write and exec permissions on dir
2772 * 4. We can't do it if dir is immutable (done in permission())
2774 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2776 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2779 if (IS_DEADDIR(dir
))
2781 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2785 * p1 and p2 should be directories on the same fs.
2787 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2792 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2796 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
2798 p
= d_ancestor(p2
, p1
);
2800 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2801 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2805 p
= d_ancestor(p1
, p2
);
2807 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2808 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2812 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2813 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2816 EXPORT_SYMBOL(lock_rename
);
2818 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2820 inode_unlock(p1
->d_inode
);
2822 inode_unlock(p2
->d_inode
);
2823 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
2826 EXPORT_SYMBOL(unlock_rename
);
2828 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2831 int error
= may_create(dir
, dentry
);
2835 if (!dir
->i_op
->create
)
2836 return -EACCES
; /* shouldn't it be ENOSYS? */
2839 error
= security_inode_create(dir
, dentry
, mode
);
2842 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2844 fsnotify_create(dir
, dentry
);
2847 EXPORT_SYMBOL(vfs_create
);
2849 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2851 struct dentry
*dentry
= path
->dentry
;
2852 struct inode
*inode
= dentry
->d_inode
;
2858 switch (inode
->i_mode
& S_IFMT
) {
2862 if (acc_mode
& MAY_WRITE
)
2867 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2876 error
= inode_permission(inode
, MAY_OPEN
| acc_mode
);
2881 * An append-only file must be opened in append mode for writing.
2883 if (IS_APPEND(inode
)) {
2884 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2890 /* O_NOATIME can only be set by the owner or superuser */
2891 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2897 static int handle_truncate(struct file
*filp
)
2899 struct path
*path
= &filp
->f_path
;
2900 struct inode
*inode
= path
->dentry
->d_inode
;
2901 int error
= get_write_access(inode
);
2905 * Refuse to truncate files with mandatory locks held on them.
2907 error
= locks_verify_locked(filp
);
2909 error
= security_path_truncate(path
);
2911 error
= do_truncate(path
->dentry
, 0,
2912 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2915 put_write_access(inode
);
2919 static inline int open_to_namei_flags(int flag
)
2921 if ((flag
& O_ACCMODE
) == 3)
2926 static int may_o_create(const struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2928 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2932 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2936 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2940 * Attempt to atomically look up, create and open a file from a negative
2943 * Returns 0 if successful. The file will have been created and attached to
2944 * @file by the filesystem calling finish_open().
2946 * Returns 1 if the file was looked up only or didn't need creating. The
2947 * caller will need to perform the open themselves. @path will have been
2948 * updated to point to the new dentry. This may be negative.
2950 * Returns an error code otherwise.
2952 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2953 struct path
*path
, struct file
*file
,
2954 const struct open_flags
*op
,
2955 int open_flag
, umode_t mode
,
2958 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2959 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2962 if (!(~open_flag
& (O_EXCL
| O_CREAT
))) /* both O_EXCL and O_CREAT */
2963 open_flag
&= ~O_TRUNC
;
2965 if (nd
->flags
& LOOKUP_DIRECTORY
)
2966 open_flag
|= O_DIRECTORY
;
2968 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2969 file
->f_path
.mnt
= nd
->path
.mnt
;
2970 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
2971 open_to_namei_flags(open_flag
),
2973 d_lookup_done(dentry
);
2976 * We didn't have the inode before the open, so check open
2979 int acc_mode
= op
->acc_mode
;
2980 if (*opened
& FILE_CREATED
) {
2981 WARN_ON(!(open_flag
& O_CREAT
));
2982 fsnotify_create(dir
, dentry
);
2985 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2986 if (WARN_ON(error
> 0))
2988 } else if (error
> 0) {
2989 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2992 if (file
->f_path
.dentry
) {
2994 dentry
= file
->f_path
.dentry
;
2996 if (*opened
& FILE_CREATED
)
2997 fsnotify_create(dir
, dentry
);
2998 path
->dentry
= dentry
;
2999 path
->mnt
= nd
->path
.mnt
;
3008 * Look up and maybe create and open the last component.
3010 * Must be called with i_mutex held on parent.
3012 * Returns 0 if the file was successfully atomically created (if necessary) and
3013 * opened. In this case the file will be returned attached to @file.
3015 * Returns 1 if the file was not completely opened at this time, though lookups
3016 * and creations will have been performed and the dentry returned in @path will
3017 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3018 * specified then a negative dentry may be returned.
3020 * An error code is returned otherwise.
3022 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3023 * cleared otherwise prior to returning.
3025 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
3027 const struct open_flags
*op
,
3028 bool got_write
, int *opened
)
3030 struct dentry
*dir
= nd
->path
.dentry
;
3031 struct inode
*dir_inode
= dir
->d_inode
;
3032 int open_flag
= op
->open_flag
;
3033 struct dentry
*dentry
;
3034 int error
, create_error
= 0;
3035 umode_t mode
= op
->mode
;
3036 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
3038 if (unlikely(IS_DEADDIR(dir_inode
)))
3041 *opened
&= ~FILE_CREATED
;
3042 dentry
= d_lookup(dir
, &nd
->last
);
3045 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
3047 return PTR_ERR(dentry
);
3049 if (d_in_lookup(dentry
))
3052 if (!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
3055 error
= d_revalidate(dentry
, nd
->flags
);
3056 if (likely(error
> 0))
3060 d_invalidate(dentry
);
3064 if (dentry
->d_inode
) {
3065 /* Cached positive dentry: will open in f_op->open */
3070 * Checking write permission is tricky, bacuse we don't know if we are
3071 * going to actually need it: O_CREAT opens should work as long as the
3072 * file exists. But checking existence breaks atomicity. The trick is
3073 * to check access and if not granted clear O_CREAT from the flags.
3075 * Another problem is returing the "right" error value (e.g. for an
3076 * O_EXCL open we want to return EEXIST not EROFS).
3078 if (open_flag
& O_CREAT
) {
3079 if (!IS_POSIXACL(dir
->d_inode
))
3080 mode
&= ~current_umask();
3081 if (unlikely(!got_write
)) {
3082 create_error
= -EROFS
;
3083 open_flag
&= ~O_CREAT
;
3084 if (open_flag
& (O_EXCL
| O_TRUNC
))
3086 /* No side effects, safe to clear O_CREAT */
3088 create_error
= may_o_create(&nd
->path
, dentry
, mode
);
3090 open_flag
&= ~O_CREAT
;
3091 if (open_flag
& O_EXCL
)
3095 } else if ((open_flag
& (O_TRUNC
|O_WRONLY
|O_RDWR
)) &&
3096 unlikely(!got_write
)) {
3098 * No O_CREATE -> atomicity not a requirement -> fall
3099 * back to lookup + open
3104 if (dir_inode
->i_op
->atomic_open
) {
3105 error
= atomic_open(nd
, dentry
, path
, file
, op
, open_flag
,
3107 if (unlikely(error
== -ENOENT
) && create_error
)
3108 error
= create_error
;
3113 if (d_in_lookup(dentry
)) {
3114 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3116 d_lookup_done(dentry
);
3117 if (unlikely(res
)) {
3119 error
= PTR_ERR(res
);
3127 /* Negative dentry, just create the file */
3128 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3129 *opened
|= FILE_CREATED
;
3130 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3131 if (!dir_inode
->i_op
->create
) {
3135 error
= dir_inode
->i_op
->create(dir_inode
, dentry
, mode
,
3136 open_flag
& O_EXCL
);
3139 fsnotify_create(dir_inode
, dentry
);
3141 if (unlikely(create_error
) && !dentry
->d_inode
) {
3142 error
= create_error
;
3146 path
->dentry
= dentry
;
3147 path
->mnt
= nd
->path
.mnt
;
3156 * Handle the last step of open()
3158 static int do_last(struct nameidata
*nd
,
3159 struct file
*file
, const struct open_flags
*op
,
3162 struct dentry
*dir
= nd
->path
.dentry
;
3163 int open_flag
= op
->open_flag
;
3164 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3165 bool got_write
= false;
3166 int acc_mode
= op
->acc_mode
;
3168 struct inode
*inode
;
3169 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
3171 bool retried
= false;
3174 nd
->flags
&= ~LOOKUP_PARENT
;
3175 nd
->flags
|= op
->intent
;
3177 if (nd
->last_type
!= LAST_NORM
) {
3178 error
= handle_dots(nd
, nd
->last_type
);
3179 if (unlikely(error
))
3184 if (!(open_flag
& O_CREAT
)) {
3185 if (nd
->last
.name
[nd
->last
.len
])
3186 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3187 /* we _can_ be in RCU mode here */
3188 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3189 if (likely(error
> 0))
3195 BUG_ON(nd
->inode
!= dir
->d_inode
);
3196 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3198 /* create side of things */
3200 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3201 * has been cleared when we got to the last component we are
3204 error
= complete_walk(nd
);
3208 audit_inode(nd
->name
, dir
, LOOKUP_PARENT
);
3209 /* trailing slashes? */
3210 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3215 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3216 error
= mnt_want_write(nd
->path
.mnt
);
3220 * do _not_ fail yet - we might not need that or fail with
3221 * a different error; let lookup_open() decide; we'll be
3222 * dropping this one anyway.
3225 if (open_flag
& O_CREAT
)
3226 inode_lock(dir
->d_inode
);
3228 inode_lock_shared(dir
->d_inode
);
3229 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3230 if (open_flag
& O_CREAT
)
3231 inode_unlock(dir
->d_inode
);
3233 inode_unlock_shared(dir
->d_inode
);
3239 if ((*opened
& FILE_CREATED
) ||
3240 !S_ISREG(file_inode(file
)->i_mode
))
3241 will_truncate
= false;
3243 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3247 if (*opened
& FILE_CREATED
) {
3248 /* Don't check for write permission, don't truncate */
3249 open_flag
&= ~O_TRUNC
;
3250 will_truncate
= false;
3252 path_to_nameidata(&path
, nd
);
3253 goto finish_open_created
;
3257 * If atomic_open() acquired write access it is dropped now due to
3258 * possible mount and symlink following (this might be optimized away if
3262 mnt_drop_write(nd
->path
.mnt
);
3266 if (unlikely(d_is_negative(path
.dentry
))) {
3267 path_to_nameidata(&path
, nd
);
3272 * create/update audit record if it already exists.
3274 audit_inode(nd
->name
, path
.dentry
, 0);
3276 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3277 path_to_nameidata(&path
, nd
);
3281 error
= follow_managed(&path
, nd
);
3282 if (unlikely(error
< 0))
3285 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3286 inode
= d_backing_inode(path
.dentry
);
3290 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
,
3292 if (unlikely(error
))
3295 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3296 path_to_nameidata(&path
, nd
);
3298 save_parent
.dentry
= nd
->path
.dentry
;
3299 save_parent
.mnt
= mntget(path
.mnt
);
3300 nd
->path
.dentry
= path
.dentry
;
3305 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3307 error
= complete_walk(nd
);
3309 path_put(&save_parent
);
3312 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3314 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3317 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3319 if (!d_is_reg(nd
->path
.dentry
))
3320 will_truncate
= false;
3322 if (will_truncate
) {
3323 error
= mnt_want_write(nd
->path
.mnt
);
3328 finish_open_created
:
3329 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3332 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3333 error
= vfs_open(&nd
->path
, file
, current_cred());
3335 *opened
|= FILE_OPENED
;
3337 if (error
== -EOPENSTALE
)
3342 error
= open_check_o_direct(file
);
3344 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3345 if (!error
&& will_truncate
)
3346 error
= handle_truncate(file
);
3348 if (unlikely(error
) && (*opened
& FILE_OPENED
))
3350 if (unlikely(error
> 0)) {
3355 mnt_drop_write(nd
->path
.mnt
);
3356 path_put(&save_parent
);
3360 /* If no saved parent or already retried then can't retry */
3361 if (!save_parent
.dentry
|| retried
)
3364 BUG_ON(save_parent
.dentry
!= dir
);
3365 path_put(&nd
->path
);
3366 nd
->path
= save_parent
;
3367 nd
->inode
= dir
->d_inode
;
3368 save_parent
.mnt
= NULL
;
3369 save_parent
.dentry
= NULL
;
3371 mnt_drop_write(nd
->path
.mnt
);
3378 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3379 const struct open_flags
*op
,
3380 struct file
*file
, int *opened
)
3382 static const struct qstr name
= QSTR_INIT("/", 1);
3383 struct dentry
*child
;
3386 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3387 if (unlikely(error
))
3389 error
= mnt_want_write(path
.mnt
);
3390 if (unlikely(error
))
3392 dir
= path
.dentry
->d_inode
;
3393 /* we want directory to be writable */
3394 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
3397 if (!dir
->i_op
->tmpfile
) {
3398 error
= -EOPNOTSUPP
;
3401 child
= d_alloc(path
.dentry
, &name
);
3402 if (unlikely(!child
)) {
3407 path
.dentry
= child
;
3408 error
= dir
->i_op
->tmpfile(dir
, child
, op
->mode
);
3411 audit_inode(nd
->name
, child
, 0);
3412 /* Don't check for other permissions, the inode was just created */
3413 error
= may_open(&path
, 0, op
->open_flag
);
3416 file
->f_path
.mnt
= path
.mnt
;
3417 error
= finish_open(file
, child
, NULL
, opened
);
3420 error
= open_check_o_direct(file
);
3423 } else if (!(op
->open_flag
& O_EXCL
)) {
3424 struct inode
*inode
= file_inode(file
);
3425 spin_lock(&inode
->i_lock
);
3426 inode
->i_state
|= I_LINKABLE
;
3427 spin_unlock(&inode
->i_lock
);
3430 mnt_drop_write(path
.mnt
);
3436 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3439 int error
= path_lookupat(nd
, flags
, &path
);
3441 audit_inode(nd
->name
, path
.dentry
, 0);
3442 error
= vfs_open(&path
, file
, current_cred());
3448 static struct file
*path_openat(struct nameidata
*nd
,
3449 const struct open_flags
*op
, unsigned flags
)
3456 file
= get_empty_filp();
3460 file
->f_flags
= op
->open_flag
;
3462 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3463 error
= do_tmpfile(nd
, flags
, op
, file
, &opened
);
3467 if (unlikely(file
->f_flags
& O_PATH
)) {
3468 error
= do_o_path(nd
, flags
, file
);
3470 opened
|= FILE_OPENED
;
3474 s
= path_init(nd
, flags
);
3479 while (!(error
= link_path_walk(s
, nd
)) &&
3480 (error
= do_last(nd
, file
, op
, &opened
)) > 0) {
3481 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3482 s
= trailing_symlink(nd
);
3490 if (!(opened
& FILE_OPENED
)) {
3494 if (unlikely(error
)) {
3495 if (error
== -EOPENSTALE
) {
3496 if (flags
& LOOKUP_RCU
)
3501 file
= ERR_PTR(error
);
3506 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3507 const struct open_flags
*op
)
3509 struct nameidata nd
;
3510 int flags
= op
->lookup_flags
;
3513 set_nameidata(&nd
, dfd
, pathname
);
3514 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3515 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3516 filp
= path_openat(&nd
, op
, flags
);
3517 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3518 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3519 restore_nameidata();
3523 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3524 const char *name
, const struct open_flags
*op
)
3526 struct nameidata nd
;
3528 struct filename
*filename
;
3529 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3532 nd
.root
.dentry
= dentry
;
3534 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3535 return ERR_PTR(-ELOOP
);
3537 filename
= getname_kernel(name
);
3538 if (IS_ERR(filename
))
3539 return ERR_CAST(filename
);
3541 set_nameidata(&nd
, -1, filename
);
3542 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3543 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3544 file
= path_openat(&nd
, op
, flags
);
3545 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3546 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3547 restore_nameidata();
3552 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3553 struct path
*path
, unsigned int lookup_flags
)
3555 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3560 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3563 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3564 * other flags passed in are ignored!
3566 lookup_flags
&= LOOKUP_REVAL
;
3568 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3570 return ERR_CAST(name
);
3573 * Yucky last component or no last component at all?
3574 * (foo/., foo/.., /////)
3576 if (unlikely(type
!= LAST_NORM
))
3579 /* don't fail immediately if it's r/o, at least try to report other errors */
3580 err2
= mnt_want_write(path
->mnt
);
3582 * Do the final lookup.
3584 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3585 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3586 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3591 if (d_is_positive(dentry
))
3595 * Special case - lookup gave negative, but... we had foo/bar/
3596 * From the vfs_mknod() POV we just have a negative dentry -
3597 * all is fine. Let's be bastards - you had / on the end, you've
3598 * been asking for (non-existent) directory. -ENOENT for you.
3600 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3604 if (unlikely(err2
)) {
3612 dentry
= ERR_PTR(error
);
3614 inode_unlock(path
->dentry
->d_inode
);
3616 mnt_drop_write(path
->mnt
);
3623 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3624 struct path
*path
, unsigned int lookup_flags
)
3626 return filename_create(dfd
, getname_kernel(pathname
),
3627 path
, lookup_flags
);
3629 EXPORT_SYMBOL(kern_path_create
);
3631 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3634 inode_unlock(path
->dentry
->d_inode
);
3635 mnt_drop_write(path
->mnt
);
3638 EXPORT_SYMBOL(done_path_create
);
3640 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3641 struct path
*path
, unsigned int lookup_flags
)
3643 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3645 EXPORT_SYMBOL(user_path_create
);
3647 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3649 int error
= may_create(dir
, dentry
);
3654 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3657 if (!dir
->i_op
->mknod
)
3660 error
= devcgroup_inode_mknod(mode
, dev
);
3664 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3668 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3670 fsnotify_create(dir
, dentry
);
3673 EXPORT_SYMBOL(vfs_mknod
);
3675 static int may_mknod(umode_t mode
)
3677 switch (mode
& S_IFMT
) {
3683 case 0: /* zero mode translates to S_IFREG */
3692 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3695 struct dentry
*dentry
;
3698 unsigned int lookup_flags
= 0;
3700 error
= may_mknod(mode
);
3704 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3706 return PTR_ERR(dentry
);
3708 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3709 mode
&= ~current_umask();
3710 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3713 switch (mode
& S_IFMT
) {
3714 case 0: case S_IFREG
:
3715 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3717 ima_post_path_mknod(dentry
);
3719 case S_IFCHR
: case S_IFBLK
:
3720 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3721 new_decode_dev(dev
));
3723 case S_IFIFO
: case S_IFSOCK
:
3724 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3728 done_path_create(&path
, dentry
);
3729 if (retry_estale(error
, lookup_flags
)) {
3730 lookup_flags
|= LOOKUP_REVAL
;
3736 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3738 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3741 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3743 int error
= may_create(dir
, dentry
);
3744 unsigned max_links
= dir
->i_sb
->s_max_links
;
3749 if (!dir
->i_op
->mkdir
)
3752 mode
&= (S_IRWXUGO
|S_ISVTX
);
3753 error
= security_inode_mkdir(dir
, dentry
, mode
);
3757 if (max_links
&& dir
->i_nlink
>= max_links
)
3760 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3762 fsnotify_mkdir(dir
, dentry
);
3765 EXPORT_SYMBOL(vfs_mkdir
);
3767 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3769 struct dentry
*dentry
;
3772 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3775 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3777 return PTR_ERR(dentry
);
3779 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3780 mode
&= ~current_umask();
3781 error
= security_path_mkdir(&path
, dentry
, mode
);
3783 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3784 done_path_create(&path
, dentry
);
3785 if (retry_estale(error
, lookup_flags
)) {
3786 lookup_flags
|= LOOKUP_REVAL
;
3792 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3794 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3797 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3799 int error
= may_delete(dir
, dentry
, 1);
3804 if (!dir
->i_op
->rmdir
)
3808 inode_lock(dentry
->d_inode
);
3811 if (is_local_mountpoint(dentry
))
3814 error
= security_inode_rmdir(dir
, dentry
);
3818 shrink_dcache_parent(dentry
);
3819 error
= dir
->i_op
->rmdir(dir
, dentry
);
3823 dentry
->d_inode
->i_flags
|= S_DEAD
;
3825 detach_mounts(dentry
);
3828 inode_unlock(dentry
->d_inode
);
3834 EXPORT_SYMBOL(vfs_rmdir
);
3836 static long do_rmdir(int dfd
, const char __user
*pathname
)
3839 struct filename
*name
;
3840 struct dentry
*dentry
;
3844 unsigned int lookup_flags
= 0;
3846 name
= user_path_parent(dfd
, pathname
,
3847 &path
, &last
, &type
, lookup_flags
);
3849 return PTR_ERR(name
);
3863 error
= mnt_want_write(path
.mnt
);
3867 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3868 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3869 error
= PTR_ERR(dentry
);
3872 if (!dentry
->d_inode
) {
3876 error
= security_path_rmdir(&path
, dentry
);
3879 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3883 inode_unlock(path
.dentry
->d_inode
);
3884 mnt_drop_write(path
.mnt
);
3888 if (retry_estale(error
, lookup_flags
)) {
3889 lookup_flags
|= LOOKUP_REVAL
;
3895 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3897 return do_rmdir(AT_FDCWD
, pathname
);
3901 * vfs_unlink - unlink a filesystem object
3902 * @dir: parent directory
3904 * @delegated_inode: returns victim inode, if the inode is delegated.
3906 * The caller must hold dir->i_mutex.
3908 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3909 * return a reference to the inode in delegated_inode. The caller
3910 * should then break the delegation on that inode and retry. Because
3911 * breaking a delegation may take a long time, the caller should drop
3912 * dir->i_mutex before doing so.
3914 * Alternatively, a caller may pass NULL for delegated_inode. This may
3915 * be appropriate for callers that expect the underlying filesystem not
3916 * to be NFS exported.
3918 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3920 struct inode
*target
= dentry
->d_inode
;
3921 int error
= may_delete(dir
, dentry
, 0);
3926 if (!dir
->i_op
->unlink
)
3930 if (is_local_mountpoint(dentry
))
3933 error
= security_inode_unlink(dir
, dentry
);
3935 error
= try_break_deleg(target
, delegated_inode
);
3938 error
= dir
->i_op
->unlink(dir
, dentry
);
3941 detach_mounts(dentry
);
3946 inode_unlock(target
);
3948 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3949 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3950 fsnotify_link_count(target
);
3956 EXPORT_SYMBOL(vfs_unlink
);
3959 * Make sure that the actual truncation of the file will occur outside its
3960 * directory's i_mutex. Truncate can take a long time if there is a lot of
3961 * writeout happening, and we don't want to prevent access to the directory
3962 * while waiting on the I/O.
3964 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3967 struct filename
*name
;
3968 struct dentry
*dentry
;
3972 struct inode
*inode
= NULL
;
3973 struct inode
*delegated_inode
= NULL
;
3974 unsigned int lookup_flags
= 0;
3976 name
= user_path_parent(dfd
, pathname
,
3977 &path
, &last
, &type
, lookup_flags
);
3979 return PTR_ERR(name
);
3982 if (type
!= LAST_NORM
)
3985 error
= mnt_want_write(path
.mnt
);
3989 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3990 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3991 error
= PTR_ERR(dentry
);
3992 if (!IS_ERR(dentry
)) {
3993 /* Why not before? Because we want correct error value */
3994 if (last
.name
[last
.len
])
3996 inode
= dentry
->d_inode
;
3997 if (d_is_negative(dentry
))
4000 error
= security_path_unlink(&path
, dentry
);
4003 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
4007 inode_unlock(path
.dentry
->d_inode
);
4009 iput(inode
); /* truncate the inode here */
4011 if (delegated_inode
) {
4012 error
= break_deleg_wait(&delegated_inode
);
4016 mnt_drop_write(path
.mnt
);
4020 if (retry_estale(error
, lookup_flags
)) {
4021 lookup_flags
|= LOOKUP_REVAL
;
4028 if (d_is_negative(dentry
))
4030 else if (d_is_dir(dentry
))
4037 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4039 if ((flag
& ~AT_REMOVEDIR
) != 0)
4042 if (flag
& AT_REMOVEDIR
)
4043 return do_rmdir(dfd
, pathname
);
4045 return do_unlinkat(dfd
, pathname
);
4048 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4050 return do_unlinkat(AT_FDCWD
, pathname
);
4053 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
4055 int error
= may_create(dir
, dentry
);
4060 if (!dir
->i_op
->symlink
)
4063 error
= security_inode_symlink(dir
, dentry
, oldname
);
4067 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
4069 fsnotify_create(dir
, dentry
);
4072 EXPORT_SYMBOL(vfs_symlink
);
4074 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4075 int, newdfd
, const char __user
*, newname
)
4078 struct filename
*from
;
4079 struct dentry
*dentry
;
4081 unsigned int lookup_flags
= 0;
4083 from
= getname(oldname
);
4085 return PTR_ERR(from
);
4087 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4088 error
= PTR_ERR(dentry
);
4092 error
= security_path_symlink(&path
, dentry
, from
->name
);
4094 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
4095 done_path_create(&path
, dentry
);
4096 if (retry_estale(error
, lookup_flags
)) {
4097 lookup_flags
|= LOOKUP_REVAL
;
4105 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4107 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4111 * vfs_link - create a new link
4112 * @old_dentry: object to be linked
4114 * @new_dentry: where to create the new link
4115 * @delegated_inode: returns inode needing a delegation break
4117 * The caller must hold dir->i_mutex
4119 * If vfs_link discovers a delegation on the to-be-linked file in need
4120 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4121 * inode in delegated_inode. The caller should then break the delegation
4122 * and retry. Because breaking a delegation may take a long time, the
4123 * caller should drop the i_mutex before doing so.
4125 * Alternatively, a caller may pass NULL for delegated_inode. This may
4126 * be appropriate for callers that expect the underlying filesystem not
4127 * to be NFS exported.
4129 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4131 struct inode
*inode
= old_dentry
->d_inode
;
4132 unsigned max_links
= dir
->i_sb
->s_max_links
;
4138 error
= may_create(dir
, new_dentry
);
4142 if (dir
->i_sb
!= inode
->i_sb
)
4146 * A link to an append-only or immutable file cannot be created.
4148 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4150 if (!dir
->i_op
->link
)
4152 if (S_ISDIR(inode
->i_mode
))
4155 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4160 /* Make sure we don't allow creating hardlink to an unlinked file */
4161 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4163 else if (max_links
&& inode
->i_nlink
>= max_links
)
4166 error
= try_break_deleg(inode
, delegated_inode
);
4168 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4171 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4172 spin_lock(&inode
->i_lock
);
4173 inode
->i_state
&= ~I_LINKABLE
;
4174 spin_unlock(&inode
->i_lock
);
4176 inode_unlock(inode
);
4178 fsnotify_link(dir
, inode
, new_dentry
);
4181 EXPORT_SYMBOL(vfs_link
);
4184 * Hardlinks are often used in delicate situations. We avoid
4185 * security-related surprises by not following symlinks on the
4188 * We don't follow them on the oldname either to be compatible
4189 * with linux 2.0, and to avoid hard-linking to directories
4190 * and other special files. --ADM
4192 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4193 int, newdfd
, const char __user
*, newname
, int, flags
)
4195 struct dentry
*new_dentry
;
4196 struct path old_path
, new_path
;
4197 struct inode
*delegated_inode
= NULL
;
4201 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4204 * To use null names we require CAP_DAC_READ_SEARCH
4205 * This ensures that not everyone will be able to create
4206 * handlink using the passed filedescriptor.
4208 if (flags
& AT_EMPTY_PATH
) {
4209 if (!capable(CAP_DAC_READ_SEARCH
))
4214 if (flags
& AT_SYMLINK_FOLLOW
)
4215 how
|= LOOKUP_FOLLOW
;
4217 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4221 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4222 (how
& LOOKUP_REVAL
));
4223 error
= PTR_ERR(new_dentry
);
4224 if (IS_ERR(new_dentry
))
4228 if (old_path
.mnt
!= new_path
.mnt
)
4230 error
= may_linkat(&old_path
);
4231 if (unlikely(error
))
4233 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4236 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4238 done_path_create(&new_path
, new_dentry
);
4239 if (delegated_inode
) {
4240 error
= break_deleg_wait(&delegated_inode
);
4242 path_put(&old_path
);
4246 if (retry_estale(error
, how
)) {
4247 path_put(&old_path
);
4248 how
|= LOOKUP_REVAL
;
4252 path_put(&old_path
);
4257 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4259 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4263 * vfs_rename - rename a filesystem object
4264 * @old_dir: parent of source
4265 * @old_dentry: source
4266 * @new_dir: parent of destination
4267 * @new_dentry: destination
4268 * @delegated_inode: returns an inode needing a delegation break
4269 * @flags: rename flags
4271 * The caller must hold multiple mutexes--see lock_rename()).
4273 * If vfs_rename discovers a delegation in need of breaking at either
4274 * the source or destination, it will return -EWOULDBLOCK and return a
4275 * reference to the inode in delegated_inode. The caller should then
4276 * break the delegation and retry. Because breaking a delegation may
4277 * take a long time, the caller should drop all locks before doing
4280 * Alternatively, a caller may pass NULL for delegated_inode. This may
4281 * be appropriate for callers that expect the underlying filesystem not
4282 * to be NFS exported.
4284 * The worst of all namespace operations - renaming directory. "Perverted"
4285 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4287 * a) we can get into loop creation.
4288 * b) race potential - two innocent renames can create a loop together.
4289 * That's where 4.4 screws up. Current fix: serialization on
4290 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4292 * c) we have to lock _four_ objects - parents and victim (if it exists),
4293 * and source (if it is not a directory).
4294 * And that - after we got ->i_mutex on parents (until then we don't know
4295 * whether the target exists). Solution: try to be smart with locking
4296 * order for inodes. We rely on the fact that tree topology may change
4297 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4298 * move will be locked. Thus we can rank directories by the tree
4299 * (ancestors first) and rank all non-directories after them.
4300 * That works since everybody except rename does "lock parent, lookup,
4301 * lock child" and rename is under ->s_vfs_rename_mutex.
4302 * HOWEVER, it relies on the assumption that any object with ->lookup()
4303 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4304 * we'd better make sure that there's no link(2) for them.
4305 * d) conversion from fhandle to dentry may come in the wrong moment - when
4306 * we are removing the target. Solution: we will have to grab ->i_mutex
4307 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4308 * ->i_mutex on parents, which works but leads to some truly excessive
4311 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4312 struct inode
*new_dir
, struct dentry
*new_dentry
,
4313 struct inode
**delegated_inode
, unsigned int flags
)
4316 bool is_dir
= d_is_dir(old_dentry
);
4317 const unsigned char *old_name
;
4318 struct inode
*source
= old_dentry
->d_inode
;
4319 struct inode
*target
= new_dentry
->d_inode
;
4320 bool new_is_dir
= false;
4321 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4324 * Check source == target.
4325 * On overlayfs need to look at underlying inodes.
4327 if (vfs_select_inode(old_dentry
, 0) == vfs_select_inode(new_dentry
, 0))
4330 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4335 error
= may_create(new_dir
, new_dentry
);
4337 new_is_dir
= d_is_dir(new_dentry
);
4339 if (!(flags
& RENAME_EXCHANGE
))
4340 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4342 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4347 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4350 if (flags
&& !old_dir
->i_op
->rename2
)
4354 * If we are going to change the parent - check write permissions,
4355 * we'll need to flip '..'.
4357 if (new_dir
!= old_dir
) {
4359 error
= inode_permission(source
, MAY_WRITE
);
4363 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4364 error
= inode_permission(target
, MAY_WRITE
);
4370 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4375 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4377 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4378 lock_two_nondirectories(source
, target
);
4383 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4386 if (max_links
&& new_dir
!= old_dir
) {
4388 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4390 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4391 old_dir
->i_nlink
>= max_links
)
4394 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4395 shrink_dcache_parent(new_dentry
);
4397 error
= try_break_deleg(source
, delegated_inode
);
4401 if (target
&& !new_is_dir
) {
4402 error
= try_break_deleg(target
, delegated_inode
);
4406 if (!old_dir
->i_op
->rename2
) {
4407 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4408 new_dir
, new_dentry
);
4410 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4411 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4412 new_dir
, new_dentry
, flags
);
4417 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4419 target
->i_flags
|= S_DEAD
;
4420 dont_mount(new_dentry
);
4421 detach_mounts(new_dentry
);
4423 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4424 if (!(flags
& RENAME_EXCHANGE
))
4425 d_move(old_dentry
, new_dentry
);
4427 d_exchange(old_dentry
, new_dentry
);
4430 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4431 unlock_two_nondirectories(source
, target
);
4433 inode_unlock(target
);
4436 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4437 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4438 if (flags
& RENAME_EXCHANGE
) {
4439 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4440 new_is_dir
, NULL
, new_dentry
);
4443 fsnotify_oldname_free(old_name
);
4447 EXPORT_SYMBOL(vfs_rename
);
4449 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4450 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4452 struct dentry
*old_dentry
, *new_dentry
;
4453 struct dentry
*trap
;
4454 struct path old_path
, new_path
;
4455 struct qstr old_last
, new_last
;
4456 int old_type
, new_type
;
4457 struct inode
*delegated_inode
= NULL
;
4458 struct filename
*from
;
4459 struct filename
*to
;
4460 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4461 bool should_retry
= false;
4464 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4467 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4468 (flags
& RENAME_EXCHANGE
))
4471 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4474 if (flags
& RENAME_EXCHANGE
)
4478 from
= user_path_parent(olddfd
, oldname
,
4479 &old_path
, &old_last
, &old_type
, lookup_flags
);
4481 error
= PTR_ERR(from
);
4485 to
= user_path_parent(newdfd
, newname
,
4486 &new_path
, &new_last
, &new_type
, lookup_flags
);
4488 error
= PTR_ERR(to
);
4493 if (old_path
.mnt
!= new_path
.mnt
)
4497 if (old_type
!= LAST_NORM
)
4500 if (flags
& RENAME_NOREPLACE
)
4502 if (new_type
!= LAST_NORM
)
4505 error
= mnt_want_write(old_path
.mnt
);
4510 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4512 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4513 error
= PTR_ERR(old_dentry
);
4514 if (IS_ERR(old_dentry
))
4516 /* source must exist */
4518 if (d_is_negative(old_dentry
))
4520 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4521 error
= PTR_ERR(new_dentry
);
4522 if (IS_ERR(new_dentry
))
4525 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4527 if (flags
& RENAME_EXCHANGE
) {
4529 if (d_is_negative(new_dentry
))
4532 if (!d_is_dir(new_dentry
)) {
4534 if (new_last
.name
[new_last
.len
])
4538 /* unless the source is a directory trailing slashes give -ENOTDIR */
4539 if (!d_is_dir(old_dentry
)) {
4541 if (old_last
.name
[old_last
.len
])
4543 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4546 /* source should not be ancestor of target */
4548 if (old_dentry
== trap
)
4550 /* target should not be an ancestor of source */
4551 if (!(flags
& RENAME_EXCHANGE
))
4553 if (new_dentry
== trap
)
4556 error
= security_path_rename(&old_path
, old_dentry
,
4557 &new_path
, new_dentry
, flags
);
4560 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4561 new_path
.dentry
->d_inode
, new_dentry
,
4562 &delegated_inode
, flags
);
4568 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4569 if (delegated_inode
) {
4570 error
= break_deleg_wait(&delegated_inode
);
4574 mnt_drop_write(old_path
.mnt
);
4576 if (retry_estale(error
, lookup_flags
))
4577 should_retry
= true;
4578 path_put(&new_path
);
4581 path_put(&old_path
);
4584 should_retry
= false;
4585 lookup_flags
|= LOOKUP_REVAL
;
4592 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4593 int, newdfd
, const char __user
*, newname
)
4595 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4598 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4600 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4603 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4605 int error
= may_create(dir
, dentry
);
4609 if (!dir
->i_op
->mknod
)
4612 return dir
->i_op
->mknod(dir
, dentry
,
4613 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4615 EXPORT_SYMBOL(vfs_whiteout
);
4617 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4619 int len
= PTR_ERR(link
);
4624 if (len
> (unsigned) buflen
)
4626 if (copy_to_user(buffer
, link
, len
))
4633 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4634 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4635 * for any given inode is up to filesystem.
4637 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4639 DEFINE_DELAYED_CALL(done
);
4640 struct inode
*inode
= d_inode(dentry
);
4641 const char *link
= inode
->i_link
;
4645 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4647 return PTR_ERR(link
);
4649 res
= readlink_copy(buffer
, buflen
, link
);
4650 do_delayed_call(&done
);
4653 EXPORT_SYMBOL(generic_readlink
);
4655 /* get the link contents into pagecache */
4656 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4657 struct delayed_call
*callback
)
4661 struct address_space
*mapping
= inode
->i_mapping
;
4664 page
= find_get_page(mapping
, 0);
4666 return ERR_PTR(-ECHILD
);
4667 if (!PageUptodate(page
)) {
4669 return ERR_PTR(-ECHILD
);
4672 page
= read_mapping_page(mapping
, 0, NULL
);
4676 set_delayed_call(callback
, page_put_link
, page
);
4677 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4678 kaddr
= page_address(page
);
4679 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4683 EXPORT_SYMBOL(page_get_link
);
4685 void page_put_link(void *arg
)
4689 EXPORT_SYMBOL(page_put_link
);
4691 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4693 DEFINE_DELAYED_CALL(done
);
4694 int res
= readlink_copy(buffer
, buflen
,
4695 page_get_link(dentry
, d_inode(dentry
),
4697 do_delayed_call(&done
);
4700 EXPORT_SYMBOL(page_readlink
);
4703 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4705 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4707 struct address_space
*mapping
= inode
->i_mapping
;
4711 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4713 flags
|= AOP_FLAG_NOFS
;
4716 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4717 flags
, &page
, &fsdata
);
4721 memcpy(page_address(page
), symname
, len
-1);
4723 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4730 mark_inode_dirty(inode
);
4735 EXPORT_SYMBOL(__page_symlink
);
4737 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4739 return __page_symlink(inode
, symname
, len
,
4740 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4742 EXPORT_SYMBOL(page_symlink
);
4744 const struct inode_operations page_symlink_inode_operations
= {
4745 .readlink
= generic_readlink
,
4746 .get_link
= page_get_link
,
4748 EXPORT_SYMBOL(page_symlink_inode_operations
);