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 (acl
== ACL_NOT_CACHED
)
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
;
1608 inode_lock(dir
->d_inode
);
1609 dentry
= d_lookup(dir
, name
);
1610 if (unlikely(dentry
)) {
1611 if ((dentry
->d_flags
& DCACHE_OP_REVALIDATE
) &&
1612 !(flags
& LOOKUP_NO_REVAL
)) {
1613 int error
= d_revalidate(dentry
, flags
);
1614 if (unlikely(error
<= 0)) {
1616 d_invalidate(dentry
);
1618 dentry
= ERR_PTR(error
);
1622 inode_unlock(dir
->d_inode
);
1626 dentry
= d_alloc(dir
, name
);
1627 if (unlikely(!dentry
)) {
1628 inode_unlock(dir
->d_inode
);
1629 return ERR_PTR(-ENOMEM
);
1631 dentry
= lookup_real(dir
->d_inode
, dentry
, flags
);
1632 inode_unlock(dir
->d_inode
);
1636 static inline int may_lookup(struct nameidata
*nd
)
1638 if (nd
->flags
& LOOKUP_RCU
) {
1639 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1642 if (unlazy_walk(nd
, NULL
, 0))
1645 return inode_permission(nd
->inode
, MAY_EXEC
);
1648 static inline int handle_dots(struct nameidata
*nd
, int type
)
1650 if (type
== LAST_DOTDOT
) {
1653 if (nd
->flags
& LOOKUP_RCU
) {
1654 return follow_dotdot_rcu(nd
);
1656 return follow_dotdot(nd
);
1661 static int pick_link(struct nameidata
*nd
, struct path
*link
,
1662 struct inode
*inode
, unsigned seq
)
1666 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
)) {
1667 path_to_nameidata(link
, nd
);
1670 if (!(nd
->flags
& LOOKUP_RCU
)) {
1671 if (link
->mnt
== nd
->path
.mnt
)
1674 error
= nd_alloc_stack(nd
);
1675 if (unlikely(error
)) {
1676 if (error
== -ECHILD
) {
1677 if (unlikely(unlazy_link(nd
, link
, seq
)))
1679 error
= nd_alloc_stack(nd
);
1687 last
= nd
->stack
+ nd
->depth
++;
1689 clear_delayed_call(&last
->done
);
1690 nd
->link_inode
= inode
;
1696 * Do we need to follow links? We _really_ want to be able
1697 * to do this check without having to look at inode->i_op,
1698 * so we keep a cache of "no, this doesn't need follow_link"
1699 * for the common case.
1701 static inline int should_follow_link(struct nameidata
*nd
, struct path
*link
,
1703 struct inode
*inode
, unsigned seq
)
1705 if (likely(!d_is_symlink(link
->dentry
)))
1709 /* make sure that d_is_symlink above matches inode */
1710 if (nd
->flags
& LOOKUP_RCU
) {
1711 if (read_seqcount_retry(&link
->dentry
->d_seq
, seq
))
1714 return pick_link(nd
, link
, inode
, seq
);
1717 enum {WALK_GET
= 1, WALK_PUT
= 2};
1719 static int walk_component(struct nameidata
*nd
, int flags
)
1722 struct inode
*inode
;
1726 * "." and ".." are special - ".." especially so because it has
1727 * to be able to know about the current root directory and
1728 * parent relationships.
1730 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1731 err
= handle_dots(nd
, nd
->last_type
);
1732 if (flags
& WALK_PUT
)
1736 err
= lookup_fast(nd
, &path
, &inode
, &seq
);
1737 if (unlikely(err
<= 0)) {
1740 path
.dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
,
1742 if (IS_ERR(path
.dentry
))
1743 return PTR_ERR(path
.dentry
);
1745 path
.mnt
= nd
->path
.mnt
;
1746 err
= follow_managed(&path
, nd
);
1747 if (unlikely(err
< 0))
1750 if (unlikely(d_is_negative(path
.dentry
))) {
1751 path_to_nameidata(&path
, nd
);
1755 seq
= 0; /* we are already out of RCU mode */
1756 inode
= d_backing_inode(path
.dentry
);
1759 if (flags
& WALK_PUT
)
1761 err
= should_follow_link(nd
, &path
, flags
& WALK_GET
, inode
, seq
);
1764 path_to_nameidata(&path
, nd
);
1771 * We can do the critical dentry name comparison and hashing
1772 * operations one word at a time, but we are limited to:
1774 * - Architectures with fast unaligned word accesses. We could
1775 * do a "get_unaligned()" if this helps and is sufficiently
1778 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1779 * do not trap on the (extremely unlikely) case of a page
1780 * crossing operation.
1782 * - Furthermore, we need an efficient 64-bit compile for the
1783 * 64-bit case in order to generate the "number of bytes in
1784 * the final mask". Again, that could be replaced with a
1785 * efficient population count instruction or similar.
1787 #ifdef CONFIG_DCACHE_WORD_ACCESS
1789 #include <asm/word-at-a-time.h>
1793 * Register pressure in the mixing function is an issue, particularly
1794 * on 32-bit x86, but almost any function requires one state value and
1795 * one temporary. Instead, use a function designed for two state values
1796 * and no temporaries.
1798 * This function cannot create a collision in only two iterations, so
1799 * we have two iterations to achieve avalanche. In those two iterations,
1800 * we have six layers of mixing, which is enough to spread one bit's
1801 * influence out to 2^6 = 64 state bits.
1803 * Rotate constants are scored by considering either 64 one-bit input
1804 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1805 * probability of that delta causing a change to each of the 128 output
1806 * bits, using a sample of random initial states.
1808 * The Shannon entropy of the computed probabilities is then summed
1809 * to produce a score. Ideally, any input change has a 50% chance of
1810 * toggling any given output bit.
1812 * Mixing scores (in bits) for (12,45):
1813 * Input delta: 1-bit 2-bit
1814 * 1 round: 713.3 42542.6
1815 * 2 rounds: 2753.7 140389.8
1816 * 3 rounds: 5954.1 233458.2
1817 * 4 rounds: 7862.6 256672.2
1818 * Perfect: 8192 258048
1819 * (64*128) (64*63/2 * 128)
1821 #define HASH_MIX(x, y, a) \
1823 y ^= x, x = rol64(x,12),\
1824 x += y, y = rol64(y,45),\
1828 * Fold two longs into one 32-bit hash value. This must be fast, but
1829 * latency isn't quite as critical, as there is a fair bit of additional
1830 * work done before the hash value is used.
1832 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1834 y
^= x
* GOLDEN_RATIO_64
;
1835 y
*= GOLDEN_RATIO_64
;
1839 #else /* 32-bit case */
1842 * Mixing scores (in bits) for (7,20):
1843 * Input delta: 1-bit 2-bit
1844 * 1 round: 330.3 9201.6
1845 * 2 rounds: 1246.4 25475.4
1846 * 3 rounds: 1907.1 31295.1
1847 * 4 rounds: 2042.3 31718.6
1848 * Perfect: 2048 31744
1849 * (32*64) (32*31/2 * 64)
1851 #define HASH_MIX(x, y, a) \
1853 y ^= x, x = rol32(x, 7),\
1854 x += y, y = rol32(y,20),\
1857 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
1859 /* Use arch-optimized multiply if one exists */
1860 return __hash_32(y
^ __hash_32(x
));
1866 * Return the hash of a string of known length. This is carfully
1867 * designed to match hash_name(), which is the more critical function.
1868 * In particular, we must end by hashing a final word containing 0..7
1869 * payload bytes, to match the way that hash_name() iterates until it
1870 * finds the delimiter after the name.
1872 unsigned int full_name_hash(const char *name
, unsigned int len
)
1874 unsigned long a
, x
= 0, y
= 0;
1879 a
= load_unaligned_zeropad(name
);
1880 if (len
< sizeof(unsigned long))
1883 name
+= sizeof(unsigned long);
1884 len
-= sizeof(unsigned long);
1886 x
^= a
& bytemask_from_count(len
);
1888 return fold_hash(x
, y
);
1890 EXPORT_SYMBOL(full_name_hash
);
1892 /* Return the "hash_len" (hash and length) of a null-terminated string */
1893 u64
hashlen_string(const char *name
)
1895 unsigned long a
= 0, x
= 0, y
= 0, adata
, mask
, len
;
1896 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1898 len
= -sizeof(unsigned long);
1901 len
+= sizeof(unsigned long);
1902 a
= load_unaligned_zeropad(name
+len
);
1903 } while (!has_zero(a
, &adata
, &constants
));
1905 adata
= prep_zero_mask(a
, adata
, &constants
);
1906 mask
= create_zero_mask(adata
);
1907 x
^= a
& zero_bytemask(mask
);
1909 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1911 EXPORT_SYMBOL(hashlen_string
);
1914 * Calculate the length and hash of the path component, and
1915 * return the "hash_len" as the result.
1917 static inline u64
hash_name(const char *name
)
1919 unsigned long a
= 0, b
, x
= 0, y
= 0, adata
, bdata
, mask
, len
;
1920 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1922 len
= -sizeof(unsigned long);
1925 len
+= sizeof(unsigned long);
1926 a
= load_unaligned_zeropad(name
+len
);
1927 b
= a
^ REPEAT_BYTE('/');
1928 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1930 adata
= prep_zero_mask(a
, adata
, &constants
);
1931 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1932 mask
= create_zero_mask(adata
| bdata
);
1933 x
^= a
& zero_bytemask(mask
);
1935 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
1938 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1940 /* Return the hash of a string of known length */
1941 unsigned int full_name_hash(const char *name
, unsigned int len
)
1943 unsigned long hash
= init_name_hash();
1945 hash
= partial_name_hash((unsigned char)*name
++, hash
);
1946 return end_name_hash(hash
);
1948 EXPORT_SYMBOL(full_name_hash
);
1950 /* Return the "hash_len" (hash and length) of a null-terminated string */
1951 u64
hash_string(const char *name
)
1953 unsigned long hash
= init_name_hash();
1954 unsigned long len
= 0, c
;
1956 c
= (unsigned char)*name
;
1959 hash
= partial_name_hash(c
, hash
);
1960 c
= (unsigned char)name
[len
];
1962 return hashlen_create(end_name_hash(hash
), len
);
1964 EXPORT_SYMBOL(hash_string
);
1967 * We know there's a real path component here of at least
1970 static inline u64
hash_name(const char *name
)
1972 unsigned long hash
= init_name_hash();
1973 unsigned long len
= 0, c
;
1975 c
= (unsigned char)*name
;
1978 hash
= partial_name_hash(c
, hash
);
1979 c
= (unsigned char)name
[len
];
1980 } while (c
&& c
!= '/');
1981 return hashlen_create(end_name_hash(hash
), len
);
1988 * This is the basic name resolution function, turning a pathname into
1989 * the final dentry. We expect 'base' to be positive and a directory.
1991 * Returns 0 and nd will have valid dentry and mnt on success.
1992 * Returns error and drops reference to input namei data on failure.
1994 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2003 /* At this point we know we have a real path component. */
2008 err
= may_lookup(nd
);
2012 hash_len
= hash_name(name
);
2015 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2017 if (name
[1] == '.') {
2019 nd
->flags
|= LOOKUP_JUMPED
;
2025 if (likely(type
== LAST_NORM
)) {
2026 struct dentry
*parent
= nd
->path
.dentry
;
2027 nd
->flags
&= ~LOOKUP_JUMPED
;
2028 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2029 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2030 err
= parent
->d_op
->d_hash(parent
, &this);
2033 hash_len
= this.hash_len
;
2038 nd
->last
.hash_len
= hash_len
;
2039 nd
->last
.name
= name
;
2040 nd
->last_type
= type
;
2042 name
+= hashlen_len(hash_len
);
2046 * If it wasn't NUL, we know it was '/'. Skip that
2047 * slash, and continue until no more slashes.
2051 } while (unlikely(*name
== '/'));
2052 if (unlikely(!*name
)) {
2054 /* pathname body, done */
2057 name
= nd
->stack
[nd
->depth
- 1].name
;
2058 /* trailing symlink, done */
2061 /* last component of nested symlink */
2062 err
= walk_component(nd
, WALK_GET
| WALK_PUT
);
2064 err
= walk_component(nd
, WALK_GET
);
2070 const char *s
= get_link(nd
);
2079 nd
->stack
[nd
->depth
- 1].name
= name
;
2084 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2085 if (nd
->flags
& LOOKUP_RCU
) {
2086 if (unlazy_walk(nd
, NULL
, 0))
2094 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2097 const char *s
= nd
->name
->name
;
2099 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
2100 nd
->flags
= flags
| LOOKUP_JUMPED
| LOOKUP_PARENT
;
2102 if (flags
& LOOKUP_ROOT
) {
2103 struct dentry
*root
= nd
->root
.dentry
;
2104 struct inode
*inode
= root
->d_inode
;
2106 if (!d_can_lookup(root
))
2107 return ERR_PTR(-ENOTDIR
);
2108 retval
= inode_permission(inode
, MAY_EXEC
);
2110 return ERR_PTR(retval
);
2112 nd
->path
= nd
->root
;
2114 if (flags
& LOOKUP_RCU
) {
2116 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2117 nd
->root_seq
= nd
->seq
;
2118 nd
->m_seq
= read_seqbegin(&mount_lock
);
2120 path_get(&nd
->path
);
2125 nd
->root
.mnt
= NULL
;
2126 nd
->path
.mnt
= NULL
;
2127 nd
->path
.dentry
= NULL
;
2129 nd
->m_seq
= read_seqbegin(&mount_lock
);
2131 if (flags
& LOOKUP_RCU
)
2134 if (likely(!nd_jump_root(nd
)))
2136 nd
->root
.mnt
= NULL
;
2138 return ERR_PTR(-ECHILD
);
2139 } else if (nd
->dfd
== AT_FDCWD
) {
2140 if (flags
& LOOKUP_RCU
) {
2141 struct fs_struct
*fs
= current
->fs
;
2147 seq
= read_seqcount_begin(&fs
->seq
);
2149 nd
->inode
= nd
->path
.dentry
->d_inode
;
2150 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2151 } while (read_seqcount_retry(&fs
->seq
, seq
));
2153 get_fs_pwd(current
->fs
, &nd
->path
);
2154 nd
->inode
= nd
->path
.dentry
->d_inode
;
2158 /* Caller must check execute permissions on the starting path component */
2159 struct fd f
= fdget_raw(nd
->dfd
);
2160 struct dentry
*dentry
;
2163 return ERR_PTR(-EBADF
);
2165 dentry
= f
.file
->f_path
.dentry
;
2168 if (!d_can_lookup(dentry
)) {
2170 return ERR_PTR(-ENOTDIR
);
2174 nd
->path
= f
.file
->f_path
;
2175 if (flags
& LOOKUP_RCU
) {
2177 nd
->inode
= nd
->path
.dentry
->d_inode
;
2178 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2180 path_get(&nd
->path
);
2181 nd
->inode
= nd
->path
.dentry
->d_inode
;
2188 static const char *trailing_symlink(struct nameidata
*nd
)
2191 int error
= may_follow_link(nd
);
2192 if (unlikely(error
))
2193 return ERR_PTR(error
);
2194 nd
->flags
|= LOOKUP_PARENT
;
2195 nd
->stack
[0].name
= NULL
;
2200 static inline int lookup_last(struct nameidata
*nd
)
2202 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2203 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2205 nd
->flags
&= ~LOOKUP_PARENT
;
2206 return walk_component(nd
,
2207 nd
->flags
& LOOKUP_FOLLOW
2209 ? WALK_PUT
| WALK_GET
2214 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2215 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2217 const char *s
= path_init(nd
, flags
);
2222 while (!(err
= link_path_walk(s
, nd
))
2223 && ((err
= lookup_last(nd
)) > 0)) {
2224 s
= trailing_symlink(nd
);
2231 err
= complete_walk(nd
);
2233 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2234 if (!d_can_lookup(nd
->path
.dentry
))
2238 nd
->path
.mnt
= NULL
;
2239 nd
->path
.dentry
= NULL
;
2245 static int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2246 struct path
*path
, struct path
*root
)
2249 struct nameidata nd
;
2251 return PTR_ERR(name
);
2252 if (unlikely(root
)) {
2254 flags
|= LOOKUP_ROOT
;
2256 set_nameidata(&nd
, dfd
, name
);
2257 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2258 if (unlikely(retval
== -ECHILD
))
2259 retval
= path_lookupat(&nd
, flags
, path
);
2260 if (unlikely(retval
== -ESTALE
))
2261 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2263 if (likely(!retval
))
2264 audit_inode(name
, path
->dentry
, flags
& LOOKUP_PARENT
);
2265 restore_nameidata();
2270 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2271 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2272 struct path
*parent
)
2274 const char *s
= path_init(nd
, flags
);
2278 err
= link_path_walk(s
, nd
);
2280 err
= complete_walk(nd
);
2283 nd
->path
.mnt
= NULL
;
2284 nd
->path
.dentry
= NULL
;
2290 static struct filename
*filename_parentat(int dfd
, struct filename
*name
,
2291 unsigned int flags
, struct path
*parent
,
2292 struct qstr
*last
, int *type
)
2295 struct nameidata nd
;
2299 set_nameidata(&nd
, dfd
, name
);
2300 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2301 if (unlikely(retval
== -ECHILD
))
2302 retval
= path_parentat(&nd
, flags
, parent
);
2303 if (unlikely(retval
== -ESTALE
))
2304 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2305 if (likely(!retval
)) {
2307 *type
= nd
.last_type
;
2308 audit_inode(name
, parent
->dentry
, LOOKUP_PARENT
);
2311 name
= ERR_PTR(retval
);
2313 restore_nameidata();
2317 /* does lookup, returns the object with parent locked */
2318 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2320 struct filename
*filename
;
2325 filename
= filename_parentat(AT_FDCWD
, getname_kernel(name
), 0, path
,
2327 if (IS_ERR(filename
))
2328 return ERR_CAST(filename
);
2329 if (unlikely(type
!= LAST_NORM
)) {
2332 return ERR_PTR(-EINVAL
);
2334 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2335 d
= __lookup_hash(&last
, path
->dentry
, 0);
2337 inode_unlock(path
->dentry
->d_inode
);
2344 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2346 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2349 EXPORT_SYMBOL(kern_path
);
2352 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2353 * @dentry: pointer to dentry of the base directory
2354 * @mnt: pointer to vfs mount of the base directory
2355 * @name: pointer to file name
2356 * @flags: lookup flags
2357 * @path: pointer to struct path to fill
2359 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2360 const char *name
, unsigned int flags
,
2363 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2364 /* the first argument of filename_lookup() is ignored with root */
2365 return filename_lookup(AT_FDCWD
, getname_kernel(name
),
2366 flags
, path
, &root
);
2368 EXPORT_SYMBOL(vfs_path_lookup
);
2371 * lookup_hash - lookup single pathname component on already hashed name
2372 * @name: name and hash to lookup
2373 * @base: base directory to lookup from
2375 * The name must have been verified and hashed (see lookup_one_len()). Using
2376 * this after just full_name_hash() is unsafe.
2378 * This function also doesn't check for search permission on base directory.
2380 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2383 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2385 struct dentry
*lookup_hash(const struct qstr
*name
, struct dentry
*base
)
2389 ret
= lookup_dcache(name
, base
, 0);
2391 ret
= lookup_slow(name
, base
, 0);
2395 EXPORT_SYMBOL(lookup_hash
);
2398 * lookup_one_len - filesystem helper to lookup single pathname component
2399 * @name: pathname component to lookup
2400 * @base: base directory to lookup from
2401 * @len: maximum length @len should be interpreted to
2403 * Note that this routine is purely a helper for filesystem usage and should
2404 * not be called by generic code.
2406 * The caller must hold base->i_mutex.
2408 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2414 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2418 this.hash
= full_name_hash(name
, len
);
2420 return ERR_PTR(-EACCES
);
2422 if (unlikely(name
[0] == '.')) {
2423 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2424 return ERR_PTR(-EACCES
);
2428 c
= *(const unsigned char *)name
++;
2429 if (c
== '/' || c
== '\0')
2430 return ERR_PTR(-EACCES
);
2433 * See if the low-level filesystem might want
2434 * to use its own hash..
2436 if (base
->d_flags
& DCACHE_OP_HASH
) {
2437 int err
= base
->d_op
->d_hash(base
, &this);
2439 return ERR_PTR(err
);
2442 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2444 return ERR_PTR(err
);
2446 return __lookup_hash(&this, base
, 0);
2448 EXPORT_SYMBOL(lookup_one_len
);
2451 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2452 * @name: pathname component to lookup
2453 * @base: base directory to lookup from
2454 * @len: maximum length @len should be interpreted to
2456 * Note that this routine is purely a helper for filesystem usage and should
2457 * not be called by generic code.
2459 * Unlike lookup_one_len, it should be called without the parent
2460 * i_mutex held, and will take the i_mutex itself if necessary.
2462 struct dentry
*lookup_one_len_unlocked(const char *name
,
2463 struct dentry
*base
, int len
)
2471 this.hash
= full_name_hash(name
, len
);
2473 return ERR_PTR(-EACCES
);
2475 if (unlikely(name
[0] == '.')) {
2476 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2477 return ERR_PTR(-EACCES
);
2481 c
= *(const unsigned char *)name
++;
2482 if (c
== '/' || c
== '\0')
2483 return ERR_PTR(-EACCES
);
2486 * See if the low-level filesystem might want
2487 * to use its own hash..
2489 if (base
->d_flags
& DCACHE_OP_HASH
) {
2490 int err
= base
->d_op
->d_hash(base
, &this);
2492 return ERR_PTR(err
);
2495 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2497 return ERR_PTR(err
);
2499 return lookup_hash(&this, base
);
2501 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2503 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2504 struct path
*path
, int *empty
)
2506 return filename_lookup(dfd
, getname_flags(name
, flags
, empty
),
2509 EXPORT_SYMBOL(user_path_at_empty
);
2512 * NB: most callers don't do anything directly with the reference to the
2513 * to struct filename, but the nd->last pointer points into the name string
2514 * allocated by getname. So we must hold the reference to it until all
2515 * path-walking is complete.
2517 static inline struct filename
*
2518 user_path_parent(int dfd
, const char __user
*path
,
2519 struct path
*parent
,
2524 /* only LOOKUP_REVAL is allowed in extra flags */
2525 return filename_parentat(dfd
, getname(path
), flags
& LOOKUP_REVAL
,
2526 parent
, last
, type
);
2530 * mountpoint_last - look up last component for umount
2531 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2532 * @path: pointer to container for result
2534 * This is a special lookup_last function just for umount. In this case, we
2535 * need to resolve the path without doing any revalidation.
2537 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2538 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2539 * in almost all cases, this lookup will be served out of the dcache. The only
2540 * cases where it won't are if nd->last refers to a symlink or the path is
2541 * bogus and it doesn't exist.
2544 * -error: if there was an error during lookup. This includes -ENOENT if the
2545 * lookup found a negative dentry. The nd->path reference will also be
2548 * 0: if we successfully resolved nd->path and found it to not to be a
2549 * symlink that needs to be followed. "path" will also be populated.
2550 * The nd->path reference will also be put.
2552 * 1: if we successfully resolved nd->last and found it to be a symlink
2553 * that needs to be followed. "path" will be populated with the path
2554 * to the link, and nd->path will *not* be put.
2557 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2560 struct dentry
*dentry
;
2561 struct dentry
*dir
= nd
->path
.dentry
;
2563 /* If we're in rcuwalk, drop out of it to handle last component */
2564 if (nd
->flags
& LOOKUP_RCU
) {
2565 if (unlazy_walk(nd
, NULL
, 0))
2569 nd
->flags
&= ~LOOKUP_PARENT
;
2571 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2572 error
= handle_dots(nd
, nd
->last_type
);
2575 dentry
= dget(nd
->path
.dentry
);
2577 dentry
= d_lookup(dir
, &nd
->last
);
2580 * No cached dentry. Mounted dentries are pinned in the
2581 * cache, so that means that this dentry is probably
2582 * a symlink or the path doesn't actually point
2583 * to a mounted dentry.
2585 dentry
= lookup_slow(&nd
->last
, dir
,
2586 nd
->flags
| LOOKUP_NO_REVAL
);
2588 return PTR_ERR(dentry
);
2591 if (d_is_negative(dentry
)) {
2597 path
->dentry
= dentry
;
2598 path
->mnt
= nd
->path
.mnt
;
2599 error
= should_follow_link(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
,
2600 d_backing_inode(dentry
), 0);
2601 if (unlikely(error
))
2609 * path_mountpoint - look up a path to be umounted
2610 * @nd: lookup context
2611 * @flags: lookup flags
2612 * @path: pointer to container for result
2614 * Look up the given name, but don't attempt to revalidate the last component.
2615 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2618 path_mountpoint(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2620 const char *s
= path_init(nd
, flags
);
2624 while (!(err
= link_path_walk(s
, nd
)) &&
2625 (err
= mountpoint_last(nd
, path
)) > 0) {
2626 s
= trailing_symlink(nd
);
2637 filename_mountpoint(int dfd
, struct filename
*name
, struct path
*path
,
2640 struct nameidata nd
;
2643 return PTR_ERR(name
);
2644 set_nameidata(&nd
, dfd
, name
);
2645 error
= path_mountpoint(&nd
, flags
| LOOKUP_RCU
, path
);
2646 if (unlikely(error
== -ECHILD
))
2647 error
= path_mountpoint(&nd
, flags
, path
);
2648 if (unlikely(error
== -ESTALE
))
2649 error
= path_mountpoint(&nd
, flags
| LOOKUP_REVAL
, path
);
2651 audit_inode(name
, path
->dentry
, 0);
2652 restore_nameidata();
2658 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2659 * @dfd: directory file descriptor
2660 * @name: pathname from userland
2661 * @flags: lookup flags
2662 * @path: pointer to container to hold result
2664 * A umount is a special case for path walking. We're not actually interested
2665 * in the inode in this situation, and ESTALE errors can be a problem. We
2666 * simply want track down the dentry and vfsmount attached at the mountpoint
2667 * and avoid revalidating the last component.
2669 * Returns 0 and populates "path" on success.
2672 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2675 return filename_mountpoint(dfd
, getname(name
), path
, flags
);
2679 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2682 return filename_mountpoint(dfd
, getname_kernel(name
), path
, flags
);
2684 EXPORT_SYMBOL(kern_path_mountpoint
);
2686 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2688 kuid_t fsuid
= current_fsuid();
2690 if (uid_eq(inode
->i_uid
, fsuid
))
2692 if (uid_eq(dir
->i_uid
, fsuid
))
2694 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2696 EXPORT_SYMBOL(__check_sticky
);
2699 * Check whether we can remove a link victim from directory dir, check
2700 * whether the type of victim is right.
2701 * 1. We can't do it if dir is read-only (done in permission())
2702 * 2. We should have write and exec permissions on dir
2703 * 3. We can't remove anything from append-only dir
2704 * 4. We can't do anything with immutable dir (done in permission())
2705 * 5. If the sticky bit on dir is set we should either
2706 * a. be owner of dir, or
2707 * b. be owner of victim, or
2708 * c. have CAP_FOWNER capability
2709 * 6. If the victim is append-only or immutable we can't do antyhing with
2710 * links pointing to it.
2711 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2712 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2713 * 9. We can't remove a root or mountpoint.
2714 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2715 * nfs_async_unlink().
2717 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2719 struct inode
*inode
= d_backing_inode(victim
);
2722 if (d_is_negative(victim
))
2726 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2727 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2729 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2735 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2736 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2739 if (!d_is_dir(victim
))
2741 if (IS_ROOT(victim
))
2743 } else if (d_is_dir(victim
))
2745 if (IS_DEADDIR(dir
))
2747 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2752 /* Check whether we can create an object with dentry child in directory
2754 * 1. We can't do it if child already exists (open has special treatment for
2755 * this case, but since we are inlined it's OK)
2756 * 2. We can't do it if dir is read-only (done in permission())
2757 * 3. We should have write and exec permissions on dir
2758 * 4. We can't do it if dir is immutable (done in permission())
2760 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2762 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2765 if (IS_DEADDIR(dir
))
2767 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2771 * p1 and p2 should be directories on the same fs.
2773 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2778 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2782 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2784 p
= d_ancestor(p2
, p1
);
2786 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
2787 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
2791 p
= d_ancestor(p1
, p2
);
2793 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2794 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
2798 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
2799 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT2
);
2802 EXPORT_SYMBOL(lock_rename
);
2804 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2806 inode_unlock(p1
->d_inode
);
2808 inode_unlock(p2
->d_inode
);
2809 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2812 EXPORT_SYMBOL(unlock_rename
);
2814 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2817 int error
= may_create(dir
, dentry
);
2821 if (!dir
->i_op
->create
)
2822 return -EACCES
; /* shouldn't it be ENOSYS? */
2825 error
= security_inode_create(dir
, dentry
, mode
);
2828 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2830 fsnotify_create(dir
, dentry
);
2833 EXPORT_SYMBOL(vfs_create
);
2835 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2837 struct dentry
*dentry
= path
->dentry
;
2838 struct inode
*inode
= dentry
->d_inode
;
2844 switch (inode
->i_mode
& S_IFMT
) {
2848 if (acc_mode
& MAY_WRITE
)
2853 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2862 error
= inode_permission(inode
, MAY_OPEN
| acc_mode
);
2867 * An append-only file must be opened in append mode for writing.
2869 if (IS_APPEND(inode
)) {
2870 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2876 /* O_NOATIME can only be set by the owner or superuser */
2877 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2883 static int handle_truncate(struct file
*filp
)
2885 struct path
*path
= &filp
->f_path
;
2886 struct inode
*inode
= path
->dentry
->d_inode
;
2887 int error
= get_write_access(inode
);
2891 * Refuse to truncate files with mandatory locks held on them.
2893 error
= locks_verify_locked(filp
);
2895 error
= security_path_truncate(path
);
2897 error
= do_truncate(path
->dentry
, 0,
2898 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2901 put_write_access(inode
);
2905 static inline int open_to_namei_flags(int flag
)
2907 if ((flag
& O_ACCMODE
) == 3)
2912 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2914 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2918 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2922 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2926 * Attempt to atomically look up, create and open a file from a negative
2929 * Returns 0 if successful. The file will have been created and attached to
2930 * @file by the filesystem calling finish_open().
2932 * Returns 1 if the file was looked up only or didn't need creating. The
2933 * caller will need to perform the open themselves. @path will have been
2934 * updated to point to the new dentry. This may be negative.
2936 * Returns an error code otherwise.
2938 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2939 struct path
*path
, struct file
*file
,
2940 const struct open_flags
*op
,
2941 bool got_write
, bool need_lookup
,
2944 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2945 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2949 int create_error
= 0;
2950 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2953 BUG_ON(dentry
->d_inode
);
2955 /* Don't create child dentry for a dead directory. */
2956 if (unlikely(IS_DEADDIR(dir
))) {
2962 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2963 mode
&= ~current_umask();
2965 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2967 open_flag
&= ~O_TRUNC
;
2970 * Checking write permission is tricky, bacuse we don't know if we are
2971 * going to actually need it: O_CREAT opens should work as long as the
2972 * file exists. But checking existence breaks atomicity. The trick is
2973 * to check access and if not granted clear O_CREAT from the flags.
2975 * Another problem is returing the "right" error value (e.g. for an
2976 * O_EXCL open we want to return EEXIST not EROFS).
2978 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2979 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2980 if (!(open_flag
& O_CREAT
)) {
2982 * No O_CREATE -> atomicity not a requirement -> fall
2983 * back to lookup + open
2986 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2987 /* Fall back and fail with the right error */
2988 create_error
= -EROFS
;
2991 /* No side effects, safe to clear O_CREAT */
2992 create_error
= -EROFS
;
2993 open_flag
&= ~O_CREAT
;
2997 if (open_flag
& O_CREAT
) {
2998 error
= may_o_create(&nd
->path
, dentry
, mode
);
3000 create_error
= error
;
3001 if (open_flag
& O_EXCL
)
3003 open_flag
&= ~O_CREAT
;
3007 if (nd
->flags
& LOOKUP_DIRECTORY
)
3008 open_flag
|= O_DIRECTORY
;
3010 file
->f_path
.dentry
= DENTRY_NOT_SET
;
3011 file
->f_path
.mnt
= nd
->path
.mnt
;
3012 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
3015 if (create_error
&& error
== -ENOENT
)
3016 error
= create_error
;
3020 if (error
) { /* returned 1, that is */
3021 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
3025 if (file
->f_path
.dentry
) {
3027 dentry
= file
->f_path
.dentry
;
3029 if (*opened
& FILE_CREATED
)
3030 fsnotify_create(dir
, dentry
);
3031 if (!dentry
->d_inode
) {
3032 WARN_ON(*opened
& FILE_CREATED
);
3034 error
= create_error
;
3038 if (excl
&& !(*opened
& FILE_CREATED
)) {
3047 * We didn't have the inode before the open, so check open permission
3050 acc_mode
= op
->acc_mode
;
3051 if (*opened
& FILE_CREATED
) {
3052 WARN_ON(!(open_flag
& O_CREAT
));
3053 fsnotify_create(dir
, dentry
);
3056 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
3066 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
3068 return PTR_ERR(dentry
);
3070 if (create_error
&& !dentry
->d_inode
) {
3071 error
= create_error
;
3075 path
->dentry
= dentry
;
3076 path
->mnt
= nd
->path
.mnt
;
3081 * Look up and maybe create and open the last component.
3083 * Must be called with i_mutex held on parent.
3085 * Returns 0 if the file was successfully atomically created (if necessary) and
3086 * opened. In this case the file will be returned attached to @file.
3088 * Returns 1 if the file was not completely opened at this time, though lookups
3089 * and creations will have been performed and the dentry returned in @path will
3090 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3091 * specified then a negative dentry may be returned.
3093 * An error code is returned otherwise.
3095 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3096 * cleared otherwise prior to returning.
3098 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
3100 const struct open_flags
*op
,
3101 bool got_write
, int *opened
)
3103 struct dentry
*dir
= nd
->path
.dentry
;
3104 struct inode
*dir_inode
= dir
->d_inode
;
3105 struct dentry
*dentry
;
3107 bool need_lookup
= false;
3109 *opened
&= ~FILE_CREATED
;
3110 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
);
3112 return PTR_ERR(dentry
);
3115 dentry
= d_alloc(dir
, &nd
->last
);
3116 if (unlikely(!dentry
))
3119 } else if (dentry
->d_inode
) {
3120 /* Cached positive dentry: will open in f_op->open */
3124 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
3125 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
3126 need_lookup
, opened
);
3130 BUG_ON(dentry
->d_inode
);
3132 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
3134 return PTR_ERR(dentry
);
3137 /* Negative dentry, just create the file */
3138 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
3139 umode_t mode
= op
->mode
;
3140 if (!IS_POSIXACL(dir
->d_inode
))
3141 mode
&= ~current_umask();
3143 * This write is needed to ensure that a
3144 * rw->ro transition does not occur between
3145 * the time when the file is created and when
3146 * a permanent write count is taken through
3147 * the 'struct file' in finish_open().
3153 *opened
|= FILE_CREATED
;
3154 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
3157 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
3158 nd
->flags
& LOOKUP_EXCL
);
3163 path
->dentry
= dentry
;
3164 path
->mnt
= nd
->path
.mnt
;
3173 * Handle the last step of open()
3175 static int do_last(struct nameidata
*nd
,
3176 struct file
*file
, const struct open_flags
*op
,
3179 struct dentry
*dir
= nd
->path
.dentry
;
3180 int open_flag
= op
->open_flag
;
3181 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
3182 bool got_write
= false;
3183 int acc_mode
= op
->acc_mode
;
3185 struct inode
*inode
;
3186 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
3188 bool retried
= false;
3191 nd
->flags
&= ~LOOKUP_PARENT
;
3192 nd
->flags
|= op
->intent
;
3194 if (nd
->last_type
!= LAST_NORM
) {
3195 error
= handle_dots(nd
, nd
->last_type
);
3196 if (unlikely(error
))
3201 if (!(open_flag
& O_CREAT
)) {
3202 if (nd
->last
.name
[nd
->last
.len
])
3203 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3204 /* we _can_ be in RCU mode here */
3205 error
= lookup_fast(nd
, &path
, &inode
, &seq
);
3206 if (likely(error
> 0))
3212 BUG_ON(nd
->inode
!= dir
->d_inode
);
3213 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3215 /* create side of things */
3217 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3218 * has been cleared when we got to the last component we are
3221 error
= complete_walk(nd
);
3225 audit_inode(nd
->name
, dir
, LOOKUP_PARENT
);
3226 /* trailing slashes? */
3227 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3232 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3233 error
= mnt_want_write(nd
->path
.mnt
);
3237 * do _not_ fail yet - we might not need that or fail with
3238 * a different error; let lookup_open() decide; we'll be
3239 * dropping this one anyway.
3242 inode_lock(dir
->d_inode
);
3243 error
= lookup_open(nd
, &path
, file
, op
, got_write
, opened
);
3244 inode_unlock(dir
->d_inode
);
3250 if ((*opened
& FILE_CREATED
) ||
3251 !S_ISREG(file_inode(file
)->i_mode
))
3252 will_truncate
= false;
3254 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3258 if (*opened
& FILE_CREATED
) {
3259 /* Don't check for write permission, don't truncate */
3260 open_flag
&= ~O_TRUNC
;
3261 will_truncate
= false;
3263 path_to_nameidata(&path
, nd
);
3264 goto finish_open_created
;
3268 * If atomic_open() acquired write access it is dropped now due to
3269 * possible mount and symlink following (this might be optimized away if
3273 mnt_drop_write(nd
->path
.mnt
);
3277 if (unlikely(d_is_negative(path
.dentry
))) {
3278 path_to_nameidata(&path
, nd
);
3283 * create/update audit record if it already exists.
3285 audit_inode(nd
->name
, path
.dentry
, 0);
3287 if (unlikely((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))) {
3288 path_to_nameidata(&path
, nd
);
3292 error
= follow_managed(&path
, nd
);
3293 if (unlikely(error
< 0))
3296 seq
= 0; /* out of RCU mode, so the value doesn't matter */
3297 inode
= d_backing_inode(path
.dentry
);
3301 error
= should_follow_link(nd
, &path
, nd
->flags
& LOOKUP_FOLLOW
,
3303 if (unlikely(error
))
3306 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
.mnt
) {
3307 path_to_nameidata(&path
, nd
);
3309 save_parent
.dentry
= nd
->path
.dentry
;
3310 save_parent
.mnt
= mntget(path
.mnt
);
3311 nd
->path
.dentry
= path
.dentry
;
3316 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3318 error
= complete_walk(nd
);
3320 path_put(&save_parent
);
3323 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3324 if (unlikely(d_is_symlink(nd
->path
.dentry
)) && !(open_flag
& O_PATH
)) {
3329 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3332 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3334 if (!d_is_reg(nd
->path
.dentry
))
3335 will_truncate
= false;
3337 if (will_truncate
) {
3338 error
= mnt_want_write(nd
->path
.mnt
);
3343 finish_open_created
:
3344 if (likely(!(open_flag
& O_PATH
))) {
3345 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3349 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3350 error
= vfs_open(&nd
->path
, file
, current_cred());
3352 *opened
|= FILE_OPENED
;
3354 if (error
== -EOPENSTALE
)
3359 error
= open_check_o_direct(file
);
3362 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3366 if (will_truncate
) {
3367 error
= handle_truncate(file
);
3372 if (unlikely(error
> 0)) {
3377 mnt_drop_write(nd
->path
.mnt
);
3378 path_put(&save_parent
);
3386 /* If no saved parent or already retried then can't retry */
3387 if (!save_parent
.dentry
|| retried
)
3390 BUG_ON(save_parent
.dentry
!= dir
);
3391 path_put(&nd
->path
);
3392 nd
->path
= save_parent
;
3393 nd
->inode
= dir
->d_inode
;
3394 save_parent
.mnt
= NULL
;
3395 save_parent
.dentry
= NULL
;
3397 mnt_drop_write(nd
->path
.mnt
);
3404 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3405 const struct open_flags
*op
,
3406 struct file
*file
, int *opened
)
3408 static const struct qstr name
= QSTR_INIT("/", 1);
3409 struct dentry
*child
;
3412 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3413 if (unlikely(error
))
3415 error
= mnt_want_write(path
.mnt
);
3416 if (unlikely(error
))
3418 dir
= path
.dentry
->d_inode
;
3419 /* we want directory to be writable */
3420 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
3423 if (!dir
->i_op
->tmpfile
) {
3424 error
= -EOPNOTSUPP
;
3427 child
= d_alloc(path
.dentry
, &name
);
3428 if (unlikely(!child
)) {
3433 path
.dentry
= child
;
3434 error
= dir
->i_op
->tmpfile(dir
, child
, op
->mode
);
3437 audit_inode(nd
->name
, child
, 0);
3438 /* Don't check for other permissions, the inode was just created */
3439 error
= may_open(&path
, 0, op
->open_flag
);
3442 file
->f_path
.mnt
= path
.mnt
;
3443 error
= finish_open(file
, child
, NULL
, opened
);
3446 error
= open_check_o_direct(file
);
3449 } else if (!(op
->open_flag
& O_EXCL
)) {
3450 struct inode
*inode
= file_inode(file
);
3451 spin_lock(&inode
->i_lock
);
3452 inode
->i_state
|= I_LINKABLE
;
3453 spin_unlock(&inode
->i_lock
);
3456 mnt_drop_write(path
.mnt
);
3462 static struct file
*path_openat(struct nameidata
*nd
,
3463 const struct open_flags
*op
, unsigned flags
)
3470 file
= get_empty_filp();
3474 file
->f_flags
= op
->open_flag
;
3476 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3477 error
= do_tmpfile(nd
, flags
, op
, file
, &opened
);
3481 s
= path_init(nd
, flags
);
3486 while (!(error
= link_path_walk(s
, nd
)) &&
3487 (error
= do_last(nd
, file
, op
, &opened
)) > 0) {
3488 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3489 s
= trailing_symlink(nd
);
3497 if (!(opened
& FILE_OPENED
)) {
3501 if (unlikely(error
)) {
3502 if (error
== -EOPENSTALE
) {
3503 if (flags
& LOOKUP_RCU
)
3508 file
= ERR_PTR(error
);
3513 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3514 const struct open_flags
*op
)
3516 struct nameidata nd
;
3517 int flags
= op
->lookup_flags
;
3520 set_nameidata(&nd
, dfd
, pathname
);
3521 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3522 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3523 filp
= path_openat(&nd
, op
, flags
);
3524 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3525 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3526 restore_nameidata();
3530 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3531 const char *name
, const struct open_flags
*op
)
3533 struct nameidata nd
;
3535 struct filename
*filename
;
3536 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3539 nd
.root
.dentry
= dentry
;
3541 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3542 return ERR_PTR(-ELOOP
);
3544 filename
= getname_kernel(name
);
3545 if (IS_ERR(filename
))
3546 return ERR_CAST(filename
);
3548 set_nameidata(&nd
, -1, filename
);
3549 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3550 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3551 file
= path_openat(&nd
, op
, flags
);
3552 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3553 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3554 restore_nameidata();
3559 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3560 struct path
*path
, unsigned int lookup_flags
)
3562 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3567 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3570 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3571 * other flags passed in are ignored!
3573 lookup_flags
&= LOOKUP_REVAL
;
3575 name
= filename_parentat(dfd
, name
, lookup_flags
, path
, &last
, &type
);
3577 return ERR_CAST(name
);
3580 * Yucky last component or no last component at all?
3581 * (foo/., foo/.., /////)
3583 if (unlikely(type
!= LAST_NORM
))
3586 /* don't fail immediately if it's r/o, at least try to report other errors */
3587 err2
= mnt_want_write(path
->mnt
);
3589 * Do the final lookup.
3591 lookup_flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3592 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3593 dentry
= __lookup_hash(&last
, path
->dentry
, lookup_flags
);
3598 if (d_is_positive(dentry
))
3602 * Special case - lookup gave negative, but... we had foo/bar/
3603 * From the vfs_mknod() POV we just have a negative dentry -
3604 * all is fine. Let's be bastards - you had / on the end, you've
3605 * been asking for (non-existent) directory. -ENOENT for you.
3607 if (unlikely(!is_dir
&& last
.name
[last
.len
])) {
3611 if (unlikely(err2
)) {
3619 dentry
= ERR_PTR(error
);
3621 inode_unlock(path
->dentry
->d_inode
);
3623 mnt_drop_write(path
->mnt
);
3630 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3631 struct path
*path
, unsigned int lookup_flags
)
3633 return filename_create(dfd
, getname_kernel(pathname
),
3634 path
, lookup_flags
);
3636 EXPORT_SYMBOL(kern_path_create
);
3638 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3641 inode_unlock(path
->dentry
->d_inode
);
3642 mnt_drop_write(path
->mnt
);
3645 EXPORT_SYMBOL(done_path_create
);
3647 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3648 struct path
*path
, unsigned int lookup_flags
)
3650 return filename_create(dfd
, getname(pathname
), path
, lookup_flags
);
3652 EXPORT_SYMBOL(user_path_create
);
3654 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3656 int error
= may_create(dir
, dentry
);
3661 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3664 if (!dir
->i_op
->mknod
)
3667 error
= devcgroup_inode_mknod(mode
, dev
);
3671 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3675 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3677 fsnotify_create(dir
, dentry
);
3680 EXPORT_SYMBOL(vfs_mknod
);
3682 static int may_mknod(umode_t mode
)
3684 switch (mode
& S_IFMT
) {
3690 case 0: /* zero mode translates to S_IFREG */
3699 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3702 struct dentry
*dentry
;
3705 unsigned int lookup_flags
= 0;
3707 error
= may_mknod(mode
);
3711 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3713 return PTR_ERR(dentry
);
3715 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3716 mode
&= ~current_umask();
3717 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3720 switch (mode
& S_IFMT
) {
3721 case 0: case S_IFREG
:
3722 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3724 case S_IFCHR
: case S_IFBLK
:
3725 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3726 new_decode_dev(dev
));
3728 case S_IFIFO
: case S_IFSOCK
:
3729 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3733 done_path_create(&path
, dentry
);
3734 if (retry_estale(error
, lookup_flags
)) {
3735 lookup_flags
|= LOOKUP_REVAL
;
3741 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3743 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3746 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3748 int error
= may_create(dir
, dentry
);
3749 unsigned max_links
= dir
->i_sb
->s_max_links
;
3754 if (!dir
->i_op
->mkdir
)
3757 mode
&= (S_IRWXUGO
|S_ISVTX
);
3758 error
= security_inode_mkdir(dir
, dentry
, mode
);
3762 if (max_links
&& dir
->i_nlink
>= max_links
)
3765 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3767 fsnotify_mkdir(dir
, dentry
);
3770 EXPORT_SYMBOL(vfs_mkdir
);
3772 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3774 struct dentry
*dentry
;
3777 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3780 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3782 return PTR_ERR(dentry
);
3784 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3785 mode
&= ~current_umask();
3786 error
= security_path_mkdir(&path
, dentry
, mode
);
3788 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3789 done_path_create(&path
, dentry
);
3790 if (retry_estale(error
, lookup_flags
)) {
3791 lookup_flags
|= LOOKUP_REVAL
;
3797 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3799 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3802 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3804 int error
= may_delete(dir
, dentry
, 1);
3809 if (!dir
->i_op
->rmdir
)
3813 inode_lock(dentry
->d_inode
);
3816 if (is_local_mountpoint(dentry
))
3819 error
= security_inode_rmdir(dir
, dentry
);
3823 shrink_dcache_parent(dentry
);
3824 error
= dir
->i_op
->rmdir(dir
, dentry
);
3828 dentry
->d_inode
->i_flags
|= S_DEAD
;
3830 detach_mounts(dentry
);
3833 inode_unlock(dentry
->d_inode
);
3839 EXPORT_SYMBOL(vfs_rmdir
);
3841 static long do_rmdir(int dfd
, const char __user
*pathname
)
3844 struct filename
*name
;
3845 struct dentry
*dentry
;
3849 unsigned int lookup_flags
= 0;
3851 name
= user_path_parent(dfd
, pathname
,
3852 &path
, &last
, &type
, lookup_flags
);
3854 return PTR_ERR(name
);
3868 error
= mnt_want_write(path
.mnt
);
3872 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3873 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3874 error
= PTR_ERR(dentry
);
3877 if (!dentry
->d_inode
) {
3881 error
= security_path_rmdir(&path
, dentry
);
3884 error
= vfs_rmdir(path
.dentry
->d_inode
, dentry
);
3888 inode_unlock(path
.dentry
->d_inode
);
3889 mnt_drop_write(path
.mnt
);
3893 if (retry_estale(error
, lookup_flags
)) {
3894 lookup_flags
|= LOOKUP_REVAL
;
3900 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3902 return do_rmdir(AT_FDCWD
, pathname
);
3906 * vfs_unlink - unlink a filesystem object
3907 * @dir: parent directory
3909 * @delegated_inode: returns victim inode, if the inode is delegated.
3911 * The caller must hold dir->i_mutex.
3913 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3914 * return a reference to the inode in delegated_inode. The caller
3915 * should then break the delegation on that inode and retry. Because
3916 * breaking a delegation may take a long time, the caller should drop
3917 * dir->i_mutex before doing so.
3919 * Alternatively, a caller may pass NULL for delegated_inode. This may
3920 * be appropriate for callers that expect the underlying filesystem not
3921 * to be NFS exported.
3923 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3925 struct inode
*target
= dentry
->d_inode
;
3926 int error
= may_delete(dir
, dentry
, 0);
3931 if (!dir
->i_op
->unlink
)
3935 if (is_local_mountpoint(dentry
))
3938 error
= security_inode_unlink(dir
, dentry
);
3940 error
= try_break_deleg(target
, delegated_inode
);
3943 error
= dir
->i_op
->unlink(dir
, dentry
);
3946 detach_mounts(dentry
);
3951 inode_unlock(target
);
3953 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3954 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3955 fsnotify_link_count(target
);
3961 EXPORT_SYMBOL(vfs_unlink
);
3964 * Make sure that the actual truncation of the file will occur outside its
3965 * directory's i_mutex. Truncate can take a long time if there is a lot of
3966 * writeout happening, and we don't want to prevent access to the directory
3967 * while waiting on the I/O.
3969 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3972 struct filename
*name
;
3973 struct dentry
*dentry
;
3977 struct inode
*inode
= NULL
;
3978 struct inode
*delegated_inode
= NULL
;
3979 unsigned int lookup_flags
= 0;
3981 name
= user_path_parent(dfd
, pathname
,
3982 &path
, &last
, &type
, lookup_flags
);
3984 return PTR_ERR(name
);
3987 if (type
!= LAST_NORM
)
3990 error
= mnt_want_write(path
.mnt
);
3994 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
3995 dentry
= __lookup_hash(&last
, path
.dentry
, lookup_flags
);
3996 error
= PTR_ERR(dentry
);
3997 if (!IS_ERR(dentry
)) {
3998 /* Why not before? Because we want correct error value */
3999 if (last
.name
[last
.len
])
4001 inode
= dentry
->d_inode
;
4002 if (d_is_negative(dentry
))
4005 error
= security_path_unlink(&path
, dentry
);
4008 error
= vfs_unlink(path
.dentry
->d_inode
, dentry
, &delegated_inode
);
4012 inode_unlock(path
.dentry
->d_inode
);
4014 iput(inode
); /* truncate the inode here */
4016 if (delegated_inode
) {
4017 error
= break_deleg_wait(&delegated_inode
);
4021 mnt_drop_write(path
.mnt
);
4025 if (retry_estale(error
, lookup_flags
)) {
4026 lookup_flags
|= LOOKUP_REVAL
;
4033 if (d_is_negative(dentry
))
4035 else if (d_is_dir(dentry
))
4042 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4044 if ((flag
& ~AT_REMOVEDIR
) != 0)
4047 if (flag
& AT_REMOVEDIR
)
4048 return do_rmdir(dfd
, pathname
);
4050 return do_unlinkat(dfd
, pathname
);
4053 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4055 return do_unlinkat(AT_FDCWD
, pathname
);
4058 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
4060 int error
= may_create(dir
, dentry
);
4065 if (!dir
->i_op
->symlink
)
4068 error
= security_inode_symlink(dir
, dentry
, oldname
);
4072 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
4074 fsnotify_create(dir
, dentry
);
4077 EXPORT_SYMBOL(vfs_symlink
);
4079 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4080 int, newdfd
, const char __user
*, newname
)
4083 struct filename
*from
;
4084 struct dentry
*dentry
;
4086 unsigned int lookup_flags
= 0;
4088 from
= getname(oldname
);
4090 return PTR_ERR(from
);
4092 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
4093 error
= PTR_ERR(dentry
);
4097 error
= security_path_symlink(&path
, dentry
, from
->name
);
4099 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
4100 done_path_create(&path
, dentry
);
4101 if (retry_estale(error
, lookup_flags
)) {
4102 lookup_flags
|= LOOKUP_REVAL
;
4110 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4112 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
4116 * vfs_link - create a new link
4117 * @old_dentry: object to be linked
4119 * @new_dentry: where to create the new link
4120 * @delegated_inode: returns inode needing a delegation break
4122 * The caller must hold dir->i_mutex
4124 * If vfs_link discovers a delegation on the to-be-linked file in need
4125 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4126 * inode in delegated_inode. The caller should then break the delegation
4127 * and retry. Because breaking a delegation may take a long time, the
4128 * caller should drop the i_mutex before doing so.
4130 * Alternatively, a caller may pass NULL for delegated_inode. This may
4131 * be appropriate for callers that expect the underlying filesystem not
4132 * to be NFS exported.
4134 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
4136 struct inode
*inode
= old_dentry
->d_inode
;
4137 unsigned max_links
= dir
->i_sb
->s_max_links
;
4143 error
= may_create(dir
, new_dentry
);
4147 if (dir
->i_sb
!= inode
->i_sb
)
4151 * A link to an append-only or immutable file cannot be created.
4153 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4155 if (!dir
->i_op
->link
)
4157 if (S_ISDIR(inode
->i_mode
))
4160 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4165 /* Make sure we don't allow creating hardlink to an unlinked file */
4166 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4168 else if (max_links
&& inode
->i_nlink
>= max_links
)
4171 error
= try_break_deleg(inode
, delegated_inode
);
4173 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4176 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4177 spin_lock(&inode
->i_lock
);
4178 inode
->i_state
&= ~I_LINKABLE
;
4179 spin_unlock(&inode
->i_lock
);
4181 inode_unlock(inode
);
4183 fsnotify_link(dir
, inode
, new_dentry
);
4186 EXPORT_SYMBOL(vfs_link
);
4189 * Hardlinks are often used in delicate situations. We avoid
4190 * security-related surprises by not following symlinks on the
4193 * We don't follow them on the oldname either to be compatible
4194 * with linux 2.0, and to avoid hard-linking to directories
4195 * and other special files. --ADM
4197 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4198 int, newdfd
, const char __user
*, newname
, int, flags
)
4200 struct dentry
*new_dentry
;
4201 struct path old_path
, new_path
;
4202 struct inode
*delegated_inode
= NULL
;
4206 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
4209 * To use null names we require CAP_DAC_READ_SEARCH
4210 * This ensures that not everyone will be able to create
4211 * handlink using the passed filedescriptor.
4213 if (flags
& AT_EMPTY_PATH
) {
4214 if (!capable(CAP_DAC_READ_SEARCH
))
4219 if (flags
& AT_SYMLINK_FOLLOW
)
4220 how
|= LOOKUP_FOLLOW
;
4222 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
4226 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4227 (how
& LOOKUP_REVAL
));
4228 error
= PTR_ERR(new_dentry
);
4229 if (IS_ERR(new_dentry
))
4233 if (old_path
.mnt
!= new_path
.mnt
)
4235 error
= may_linkat(&old_path
);
4236 if (unlikely(error
))
4238 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4241 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4243 done_path_create(&new_path
, new_dentry
);
4244 if (delegated_inode
) {
4245 error
= break_deleg_wait(&delegated_inode
);
4247 path_put(&old_path
);
4251 if (retry_estale(error
, how
)) {
4252 path_put(&old_path
);
4253 how
|= LOOKUP_REVAL
;
4257 path_put(&old_path
);
4262 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4264 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4268 * vfs_rename - rename a filesystem object
4269 * @old_dir: parent of source
4270 * @old_dentry: source
4271 * @new_dir: parent of destination
4272 * @new_dentry: destination
4273 * @delegated_inode: returns an inode needing a delegation break
4274 * @flags: rename flags
4276 * The caller must hold multiple mutexes--see lock_rename()).
4278 * If vfs_rename discovers a delegation in need of breaking at either
4279 * the source or destination, it will return -EWOULDBLOCK and return a
4280 * reference to the inode in delegated_inode. The caller should then
4281 * break the delegation and retry. Because breaking a delegation may
4282 * take a long time, the caller should drop all locks before doing
4285 * Alternatively, a caller may pass NULL for delegated_inode. This may
4286 * be appropriate for callers that expect the underlying filesystem not
4287 * to be NFS exported.
4289 * The worst of all namespace operations - renaming directory. "Perverted"
4290 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4292 * a) we can get into loop creation.
4293 * b) race potential - two innocent renames can create a loop together.
4294 * That's where 4.4 screws up. Current fix: serialization on
4295 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4297 * c) we have to lock _four_ objects - parents and victim (if it exists),
4298 * and source (if it is not a directory).
4299 * And that - after we got ->i_mutex on parents (until then we don't know
4300 * whether the target exists). Solution: try to be smart with locking
4301 * order for inodes. We rely on the fact that tree topology may change
4302 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4303 * move will be locked. Thus we can rank directories by the tree
4304 * (ancestors first) and rank all non-directories after them.
4305 * That works since everybody except rename does "lock parent, lookup,
4306 * lock child" and rename is under ->s_vfs_rename_mutex.
4307 * HOWEVER, it relies on the assumption that any object with ->lookup()
4308 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4309 * we'd better make sure that there's no link(2) for them.
4310 * d) conversion from fhandle to dentry may come in the wrong moment - when
4311 * we are removing the target. Solution: we will have to grab ->i_mutex
4312 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4313 * ->i_mutex on parents, which works but leads to some truly excessive
4316 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4317 struct inode
*new_dir
, struct dentry
*new_dentry
,
4318 struct inode
**delegated_inode
, unsigned int flags
)
4321 bool is_dir
= d_is_dir(old_dentry
);
4322 const unsigned char *old_name
;
4323 struct inode
*source
= old_dentry
->d_inode
;
4324 struct inode
*target
= new_dentry
->d_inode
;
4325 bool new_is_dir
= false;
4326 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4329 * Check source == target.
4330 * On overlayfs need to look at underlying inodes.
4332 if (vfs_select_inode(old_dentry
, 0) == vfs_select_inode(new_dentry
, 0))
4335 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4340 error
= may_create(new_dir
, new_dentry
);
4342 new_is_dir
= d_is_dir(new_dentry
);
4344 if (!(flags
& RENAME_EXCHANGE
))
4345 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4347 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4352 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4355 if (flags
&& !old_dir
->i_op
->rename2
)
4359 * If we are going to change the parent - check write permissions,
4360 * we'll need to flip '..'.
4362 if (new_dir
!= old_dir
) {
4364 error
= inode_permission(source
, MAY_WRITE
);
4368 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4369 error
= inode_permission(target
, MAY_WRITE
);
4375 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4380 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4382 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4383 lock_two_nondirectories(source
, target
);
4388 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4391 if (max_links
&& new_dir
!= old_dir
) {
4393 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4395 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4396 old_dir
->i_nlink
>= max_links
)
4399 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4400 shrink_dcache_parent(new_dentry
);
4402 error
= try_break_deleg(source
, delegated_inode
);
4406 if (target
&& !new_is_dir
) {
4407 error
= try_break_deleg(target
, delegated_inode
);
4411 if (!old_dir
->i_op
->rename2
) {
4412 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4413 new_dir
, new_dentry
);
4415 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4416 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4417 new_dir
, new_dentry
, flags
);
4422 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4424 target
->i_flags
|= S_DEAD
;
4425 dont_mount(new_dentry
);
4426 detach_mounts(new_dentry
);
4428 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4429 if (!(flags
& RENAME_EXCHANGE
))
4430 d_move(old_dentry
, new_dentry
);
4432 d_exchange(old_dentry
, new_dentry
);
4435 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4436 unlock_two_nondirectories(source
, target
);
4438 inode_unlock(target
);
4441 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4442 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4443 if (flags
& RENAME_EXCHANGE
) {
4444 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4445 new_is_dir
, NULL
, new_dentry
);
4448 fsnotify_oldname_free(old_name
);
4452 EXPORT_SYMBOL(vfs_rename
);
4454 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4455 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4457 struct dentry
*old_dentry
, *new_dentry
;
4458 struct dentry
*trap
;
4459 struct path old_path
, new_path
;
4460 struct qstr old_last
, new_last
;
4461 int old_type
, new_type
;
4462 struct inode
*delegated_inode
= NULL
;
4463 struct filename
*from
;
4464 struct filename
*to
;
4465 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4466 bool should_retry
= false;
4469 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4472 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4473 (flags
& RENAME_EXCHANGE
))
4476 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4479 if (flags
& RENAME_EXCHANGE
)
4483 from
= user_path_parent(olddfd
, oldname
,
4484 &old_path
, &old_last
, &old_type
, lookup_flags
);
4486 error
= PTR_ERR(from
);
4490 to
= user_path_parent(newdfd
, newname
,
4491 &new_path
, &new_last
, &new_type
, lookup_flags
);
4493 error
= PTR_ERR(to
);
4498 if (old_path
.mnt
!= new_path
.mnt
)
4502 if (old_type
!= LAST_NORM
)
4505 if (flags
& RENAME_NOREPLACE
)
4507 if (new_type
!= LAST_NORM
)
4510 error
= mnt_want_write(old_path
.mnt
);
4515 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4517 old_dentry
= __lookup_hash(&old_last
, old_path
.dentry
, lookup_flags
);
4518 error
= PTR_ERR(old_dentry
);
4519 if (IS_ERR(old_dentry
))
4521 /* source must exist */
4523 if (d_is_negative(old_dentry
))
4525 new_dentry
= __lookup_hash(&new_last
, new_path
.dentry
, lookup_flags
| target_flags
);
4526 error
= PTR_ERR(new_dentry
);
4527 if (IS_ERR(new_dentry
))
4530 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4532 if (flags
& RENAME_EXCHANGE
) {
4534 if (d_is_negative(new_dentry
))
4537 if (!d_is_dir(new_dentry
)) {
4539 if (new_last
.name
[new_last
.len
])
4543 /* unless the source is a directory trailing slashes give -ENOTDIR */
4544 if (!d_is_dir(old_dentry
)) {
4546 if (old_last
.name
[old_last
.len
])
4548 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4551 /* source should not be ancestor of target */
4553 if (old_dentry
== trap
)
4555 /* target should not be an ancestor of source */
4556 if (!(flags
& RENAME_EXCHANGE
))
4558 if (new_dentry
== trap
)
4561 error
= security_path_rename(&old_path
, old_dentry
,
4562 &new_path
, new_dentry
, flags
);
4565 error
= vfs_rename(old_path
.dentry
->d_inode
, old_dentry
,
4566 new_path
.dentry
->d_inode
, new_dentry
,
4567 &delegated_inode
, flags
);
4573 unlock_rename(new_path
.dentry
, old_path
.dentry
);
4574 if (delegated_inode
) {
4575 error
= break_deleg_wait(&delegated_inode
);
4579 mnt_drop_write(old_path
.mnt
);
4581 if (retry_estale(error
, lookup_flags
))
4582 should_retry
= true;
4583 path_put(&new_path
);
4586 path_put(&old_path
);
4589 should_retry
= false;
4590 lookup_flags
|= LOOKUP_REVAL
;
4597 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4598 int, newdfd
, const char __user
*, newname
)
4600 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4603 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4605 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4608 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4610 int error
= may_create(dir
, dentry
);
4614 if (!dir
->i_op
->mknod
)
4617 return dir
->i_op
->mknod(dir
, dentry
,
4618 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4620 EXPORT_SYMBOL(vfs_whiteout
);
4622 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4624 int len
= PTR_ERR(link
);
4629 if (len
> (unsigned) buflen
)
4631 if (copy_to_user(buffer
, link
, len
))
4636 EXPORT_SYMBOL(readlink_copy
);
4639 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4640 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4641 * for any given inode is up to filesystem.
4643 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4645 DEFINE_DELAYED_CALL(done
);
4646 struct inode
*inode
= d_inode(dentry
);
4647 const char *link
= inode
->i_link
;
4651 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
4653 return PTR_ERR(link
);
4655 res
= readlink_copy(buffer
, buflen
, link
);
4656 do_delayed_call(&done
);
4659 EXPORT_SYMBOL(generic_readlink
);
4661 /* get the link contents into pagecache */
4662 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
4663 struct delayed_call
*callback
)
4667 struct address_space
*mapping
= inode
->i_mapping
;
4670 page
= find_get_page(mapping
, 0);
4672 return ERR_PTR(-ECHILD
);
4673 if (!PageUptodate(page
)) {
4675 return ERR_PTR(-ECHILD
);
4678 page
= read_mapping_page(mapping
, 0, NULL
);
4682 set_delayed_call(callback
, page_put_link
, page
);
4683 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
4684 kaddr
= page_address(page
);
4685 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
4689 EXPORT_SYMBOL(page_get_link
);
4691 void page_put_link(void *arg
)
4695 EXPORT_SYMBOL(page_put_link
);
4697 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4699 DEFINE_DELAYED_CALL(done
);
4700 int res
= readlink_copy(buffer
, buflen
,
4701 page_get_link(dentry
, d_inode(dentry
),
4703 do_delayed_call(&done
);
4706 EXPORT_SYMBOL(page_readlink
);
4709 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4711 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4713 struct address_space
*mapping
= inode
->i_mapping
;
4717 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4719 flags
|= AOP_FLAG_NOFS
;
4722 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4723 flags
, &page
, &fsdata
);
4727 memcpy(page_address(page
), symname
, len
-1);
4729 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4736 mark_inode_dirty(inode
);
4741 EXPORT_SYMBOL(__page_symlink
);
4743 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4745 return __page_symlink(inode
, symname
, len
,
4746 !mapping_gfp_constraint(inode
->i_mapping
, __GFP_FS
));
4748 EXPORT_SYMBOL(page_symlink
);
4750 const struct inode_operations page_symlink_inode_operations
= {
4751 .readlink
= generic_readlink
,
4752 .get_link
= page_get_link
,
4754 EXPORT_SYMBOL(page_symlink_inode_operations
);