1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/filelock.h>
24 #include <linux/namei.h>
25 #include <linux/pagemap.h>
26 #include <linux/sched/mm.h>
27 #include <linux/fsnotify.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/ima.h>
31 #include <linux/syscalls.h>
32 #include <linux/mount.h>
33 #include <linux/audit.h>
34 #include <linux/capability.h>
35 #include <linux/file.h>
36 #include <linux/fcntl.h>
37 #include <linux/device_cgroup.h>
38 #include <linux/fs_struct.h>
39 #include <linux/posix_acl.h>
40 #include <linux/hash.h>
41 #include <linux/bitops.h>
42 #include <linux/init_task.h>
43 #include <linux/uaccess.h>
48 /* [Feb-1997 T. Schoebel-Theuer]
49 * Fundamental changes in the pathname lookup mechanisms (namei)
50 * were necessary because of omirr. The reason is that omirr needs
51 * to know the _real_ pathname, not the user-supplied one, in case
52 * of symlinks (and also when transname replacements occur).
54 * The new code replaces the old recursive symlink resolution with
55 * an iterative one (in case of non-nested symlink chains). It does
56 * this with calls to <fs>_follow_link().
57 * As a side effect, dir_namei(), _namei() and follow_link() are now
58 * replaced with a single function lookup_dentry() that can handle all
59 * the special cases of the former code.
61 * With the new dcache, the pathname is stored at each inode, at least as
62 * long as the refcount of the inode is positive. As a side effect, the
63 * size of the dcache depends on the inode cache and thus is dynamic.
65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 * resolution to correspond with current state of the code.
68 * Note that the symlink resolution is not *completely* iterative.
69 * There is still a significant amount of tail- and mid- recursion in
70 * the algorithm. Also, note that <fs>_readlink() is not used in
71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 * may return different results than <fs>_follow_link(). Many virtual
73 * filesystems (including /proc) exhibit this behavior.
76 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 * and the name already exists in form of a symlink, try to create the new
79 * name indicated by the symlink. The old code always complained that the
80 * name already exists, due to not following the symlink even if its target
81 * is nonexistent. The new semantics affects also mknod() and link() when
82 * the name is a symlink pointing to a non-existent name.
84 * I don't know which semantics is the right one, since I have no access
85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 * "old" one. Personally, I think the new semantics is much more logical.
88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 * file does succeed in both HP-UX and SunOs, but not in Solaris
90 * and in the old Linux semantics.
93 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 * semantics. See the comments in "open_namei" and "do_link" below.
96 * [10-Sep-98 Alan Modra] Another symlink change.
99 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 * inside the path - always follow.
101 * in the last component in creation/removal/renaming - never follow.
102 * if LOOKUP_FOLLOW passed - follow.
103 * if the pathname has trailing slashes - follow.
104 * otherwise - don't follow.
105 * (applied in that order).
107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 * During the 2.4 we need to fix the userland stuff depending on it -
110 * hopefully we will be able to get rid of that wart in 2.5. So far only
111 * XEmacs seems to be relying on it...
114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 * any extra contention...
119 /* In order to reduce some races, while at the same time doing additional
120 * checking and hopefully speeding things up, we copy filenames to the
121 * kernel data space before using them..
123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 * PATH_MAX includes the nul terminator --RR.
127 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
130 getname_flags(const char __user
*filename
, int flags
, int *empty
)
132 struct filename
*result
;
136 result
= audit_reusename(filename
);
140 result
= __getname();
141 if (unlikely(!result
))
142 return ERR_PTR(-ENOMEM
);
145 * First, try to embed the struct filename inside the names_cache
148 kname
= (char *)result
->iname
;
149 result
->name
= kname
;
151 len
= strncpy_from_user(kname
, filename
, EMBEDDED_NAME_MAX
);
152 if (unlikely(len
< 0)) {
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
163 if (unlikely(len
== EMBEDDED_NAME_MAX
)) {
164 const size_t size
= offsetof(struct filename
, iname
[1]);
165 kname
= (char *)result
;
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
172 result
= kzalloc(size
, GFP_KERNEL
);
173 if (unlikely(!result
)) {
175 return ERR_PTR(-ENOMEM
);
177 result
->name
= kname
;
178 len
= strncpy_from_user(kname
, filename
, PATH_MAX
);
179 if (unlikely(len
< 0)) {
184 if (unlikely(len
== PATH_MAX
)) {
187 return ERR_PTR(-ENAMETOOLONG
);
191 atomic_set(&result
->refcnt
, 1);
192 /* The empty path is special. */
193 if (unlikely(!len
)) {
196 if (!(flags
& LOOKUP_EMPTY
)) {
198 return ERR_PTR(-ENOENT
);
202 result
->uptr
= filename
;
203 result
->aname
= NULL
;
204 audit_getname(result
);
209 getname_uflags(const char __user
*filename
, int uflags
)
211 int flags
= (uflags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
213 return getname_flags(filename
, flags
, NULL
);
217 getname(const char __user
* filename
)
219 return getname_flags(filename
, 0, NULL
);
223 getname_kernel(const char * filename
)
225 struct filename
*result
;
226 int len
= strlen(filename
) + 1;
228 result
= __getname();
229 if (unlikely(!result
))
230 return ERR_PTR(-ENOMEM
);
232 if (len
<= EMBEDDED_NAME_MAX
) {
233 result
->name
= (char *)result
->iname
;
234 } else if (len
<= PATH_MAX
) {
235 const size_t size
= offsetof(struct filename
, iname
[1]);
236 struct filename
*tmp
;
238 tmp
= kmalloc(size
, GFP_KERNEL
);
239 if (unlikely(!tmp
)) {
241 return ERR_PTR(-ENOMEM
);
243 tmp
->name
= (char *)result
;
247 return ERR_PTR(-ENAMETOOLONG
);
249 memcpy((char *)result
->name
, filename
, len
);
251 result
->aname
= NULL
;
252 atomic_set(&result
->refcnt
, 1);
253 audit_getname(result
);
257 EXPORT_SYMBOL(getname_kernel
);
259 void putname(struct filename
*name
)
264 if (WARN_ON_ONCE(!atomic_read(&name
->refcnt
)))
267 if (!atomic_dec_and_test(&name
->refcnt
))
270 if (name
->name
!= name
->iname
) {
271 __putname(name
->name
);
276 EXPORT_SYMBOL(putname
);
279 * check_acl - perform ACL permission checking
280 * @idmap: idmap of the mount the inode was found from
281 * @inode: inode to check permissions on
282 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
284 * This function performs the ACL permission checking. Since this function
285 * retrieve POSIX acls it needs to know whether it is called from a blocking or
286 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
288 * If the inode has been found through an idmapped mount the idmap of
289 * the vfsmount must be passed through @idmap. This function will then take
290 * care to map the inode according to @idmap before checking permissions.
291 * On non-idmapped mounts or if permission checking is to be performed on the
292 * raw inode simply passs @nop_mnt_idmap.
294 static int check_acl(struct mnt_idmap
*idmap
,
295 struct inode
*inode
, int mask
)
297 #ifdef CONFIG_FS_POSIX_ACL
298 struct posix_acl
*acl
;
300 if (mask
& MAY_NOT_BLOCK
) {
301 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
304 /* no ->get_inode_acl() calls in RCU mode... */
305 if (is_uncached_acl(acl
))
307 return posix_acl_permission(idmap
, inode
, acl
, mask
);
310 acl
= get_inode_acl(inode
, ACL_TYPE_ACCESS
);
314 int error
= posix_acl_permission(idmap
, inode
, acl
, mask
);
315 posix_acl_release(acl
);
324 * acl_permission_check - perform basic UNIX permission checking
325 * @idmap: idmap of the mount the inode was found from
326 * @inode: inode to check permissions on
327 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
329 * This function performs the basic UNIX permission checking. Since this
330 * function may retrieve POSIX acls it needs to know whether it is called from a
331 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
333 * If the inode has been found through an idmapped mount the idmap of
334 * the vfsmount must be passed through @idmap. This function will then take
335 * care to map the inode according to @idmap before checking permissions.
336 * On non-idmapped mounts or if permission checking is to be performed on the
337 * raw inode simply passs @nop_mnt_idmap.
339 static int acl_permission_check(struct mnt_idmap
*idmap
,
340 struct inode
*inode
, int mask
)
342 unsigned int mode
= inode
->i_mode
;
345 /* Are we the owner? If so, ACL's don't matter */
346 vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
347 if (likely(vfsuid_eq_kuid(vfsuid
, current_fsuid()))) {
350 return (mask
& ~mode
) ? -EACCES
: 0;
353 /* Do we have ACL's? */
354 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
355 int error
= check_acl(idmap
, inode
, mask
);
356 if (error
!= -EAGAIN
)
360 /* Only RWX matters for group/other mode bits */
364 * Are the group permissions different from
365 * the other permissions in the bits we care
366 * about? Need to check group ownership if so.
368 if (mask
& (mode
^ (mode
>> 3))) {
369 vfsgid_t vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
370 if (vfsgid_in_group_p(vfsgid
))
374 /* Bits in 'mode' clear that we require? */
375 return (mask
& ~mode
) ? -EACCES
: 0;
379 * generic_permission - check for access rights on a Posix-like filesystem
380 * @idmap: idmap of the mount the inode was found from
381 * @inode: inode to check access rights for
382 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
383 * %MAY_NOT_BLOCK ...)
385 * Used to check for read/write/execute permissions on a file.
386 * We use "fsuid" for this, letting us set arbitrary permissions
387 * for filesystem access without changing the "normal" uids which
388 * are used for other things.
390 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
391 * request cannot be satisfied (eg. requires blocking or too much complexity).
392 * It would then be called again in ref-walk mode.
394 * If the inode has been found through an idmapped mount the idmap of
395 * the vfsmount must be passed through @idmap. This function will then take
396 * care to map the inode according to @idmap before checking permissions.
397 * On non-idmapped mounts or if permission checking is to be performed on the
398 * raw inode simply passs @nop_mnt_idmap.
400 int generic_permission(struct mnt_idmap
*idmap
, struct inode
*inode
,
406 * Do the basic permission checks.
408 ret
= acl_permission_check(idmap
, inode
, mask
);
412 if (S_ISDIR(inode
->i_mode
)) {
413 /* DACs are overridable for directories */
414 if (!(mask
& MAY_WRITE
))
415 if (capable_wrt_inode_uidgid(idmap
, inode
,
416 CAP_DAC_READ_SEARCH
))
418 if (capable_wrt_inode_uidgid(idmap
, inode
,
425 * Searching includes executable on directories, else just read.
427 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
428 if (mask
== MAY_READ
)
429 if (capable_wrt_inode_uidgid(idmap
, inode
,
430 CAP_DAC_READ_SEARCH
))
433 * Read/write DACs are always overridable.
434 * Executable DACs are overridable when there is
435 * at least one exec bit set.
437 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
438 if (capable_wrt_inode_uidgid(idmap
, inode
,
444 EXPORT_SYMBOL(generic_permission
);
447 * do_inode_permission - UNIX permission checking
448 * @idmap: idmap of the mount the inode was found from
449 * @inode: inode to check permissions on
450 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
452 * We _really_ want to just do "generic_permission()" without
453 * even looking at the inode->i_op values. So we keep a cache
454 * flag in inode->i_opflags, that says "this has not special
455 * permission function, use the fast case".
457 static inline int do_inode_permission(struct mnt_idmap
*idmap
,
458 struct inode
*inode
, int mask
)
460 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
461 if (likely(inode
->i_op
->permission
))
462 return inode
->i_op
->permission(idmap
, inode
, mask
);
464 /* This gets set once for the inode lifetime */
465 spin_lock(&inode
->i_lock
);
466 inode
->i_opflags
|= IOP_FASTPERM
;
467 spin_unlock(&inode
->i_lock
);
469 return generic_permission(idmap
, inode
, mask
);
473 * sb_permission - Check superblock-level permissions
474 * @sb: Superblock of inode to check permission on
475 * @inode: Inode to check permission on
476 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
478 * Separate out file-system wide checks from inode-specific permission checks.
480 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
482 if (unlikely(mask
& MAY_WRITE
)) {
483 umode_t mode
= inode
->i_mode
;
485 /* Nobody gets write access to a read-only fs. */
486 if (sb_rdonly(sb
) && (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
493 * inode_permission - Check for access rights to a given inode
494 * @idmap: idmap of the mount the inode was found from
495 * @inode: Inode to check permission on
496 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
498 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
499 * this, letting us set arbitrary permissions for filesystem access without
500 * changing the "normal" UIDs which are used for other things.
502 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
504 int inode_permission(struct mnt_idmap
*idmap
,
505 struct inode
*inode
, int mask
)
509 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
513 if (unlikely(mask
& MAY_WRITE
)) {
515 * Nobody gets write access to an immutable file.
517 if (IS_IMMUTABLE(inode
))
521 * Updating mtime will likely cause i_uid and i_gid to be
522 * written back improperly if their true value is unknown
525 if (HAS_UNMAPPED_ID(idmap
, inode
))
529 retval
= do_inode_permission(idmap
, inode
, mask
);
533 retval
= devcgroup_inode_permission(inode
, mask
);
537 return security_inode_permission(inode
, mask
);
539 EXPORT_SYMBOL(inode_permission
);
542 * path_get - get a reference to a path
543 * @path: path to get the reference to
545 * Given a path increment the reference count to the dentry and the vfsmount.
547 void path_get(const struct path
*path
)
552 EXPORT_SYMBOL(path_get
);
555 * path_put - put a reference to a path
556 * @path: path to put the reference to
558 * Given a path decrement the reference count to the dentry and the vfsmount.
560 void path_put(const struct path
*path
)
565 EXPORT_SYMBOL(path_put
);
567 #define EMBEDDED_LEVELS 2
572 struct inode
*inode
; /* path.dentry.d_inode */
573 unsigned int flags
, state
;
574 unsigned seq
, next_seq
, m_seq
, r_seq
;
577 int total_link_count
;
580 struct delayed_call done
;
583 } *stack
, internal
[EMBEDDED_LEVELS
];
584 struct filename
*name
;
585 struct nameidata
*saved
;
590 } __randomize_layout
;
592 #define ND_ROOT_PRESET 1
593 #define ND_ROOT_GRABBED 2
596 static void __set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
)
598 struct nameidata
*old
= current
->nameidata
;
599 p
->stack
= p
->internal
;
604 p
->path
.dentry
= NULL
;
605 p
->total_link_count
= old
? old
->total_link_count
: 0;
607 current
->nameidata
= p
;
610 static inline void set_nameidata(struct nameidata
*p
, int dfd
, struct filename
*name
,
611 const struct path
*root
)
613 __set_nameidata(p
, dfd
, name
);
615 if (unlikely(root
)) {
616 p
->state
= ND_ROOT_PRESET
;
621 static void restore_nameidata(void)
623 struct nameidata
*now
= current
->nameidata
, *old
= now
->saved
;
625 current
->nameidata
= old
;
627 old
->total_link_count
= now
->total_link_count
;
628 if (now
->stack
!= now
->internal
)
632 static bool nd_alloc_stack(struct nameidata
*nd
)
636 p
= kmalloc_array(MAXSYMLINKS
, sizeof(struct saved
),
637 nd
->flags
& LOOKUP_RCU
? GFP_ATOMIC
: GFP_KERNEL
);
640 memcpy(p
, nd
->internal
, sizeof(nd
->internal
));
646 * path_connected - Verify that a dentry is below mnt.mnt_root
647 * @mnt: The mountpoint to check.
648 * @dentry: The dentry to check.
650 * Rename can sometimes move a file or directory outside of a bind
651 * mount, path_connected allows those cases to be detected.
653 static bool path_connected(struct vfsmount
*mnt
, struct dentry
*dentry
)
655 struct super_block
*sb
= mnt
->mnt_sb
;
657 /* Bind mounts can have disconnected paths */
658 if (mnt
->mnt_root
== sb
->s_root
)
661 return is_subdir(dentry
, mnt
->mnt_root
);
664 static void drop_links(struct nameidata
*nd
)
668 struct saved
*last
= nd
->stack
+ i
;
669 do_delayed_call(&last
->done
);
670 clear_delayed_call(&last
->done
);
674 static void leave_rcu(struct nameidata
*nd
)
676 nd
->flags
&= ~LOOKUP_RCU
;
677 nd
->seq
= nd
->next_seq
= 0;
681 static void terminate_walk(struct nameidata
*nd
)
684 if (!(nd
->flags
& LOOKUP_RCU
)) {
687 for (i
= 0; i
< nd
->depth
; i
++)
688 path_put(&nd
->stack
[i
].link
);
689 if (nd
->state
& ND_ROOT_GRABBED
) {
691 nd
->state
&= ~ND_ROOT_GRABBED
;
698 nd
->path
.dentry
= NULL
;
701 /* path_put is needed afterwards regardless of success or failure */
702 static bool __legitimize_path(struct path
*path
, unsigned seq
, unsigned mseq
)
704 int res
= __legitimize_mnt(path
->mnt
, mseq
);
711 if (unlikely(!lockref_get_not_dead(&path
->dentry
->d_lockref
))) {
715 return !read_seqcount_retry(&path
->dentry
->d_seq
, seq
);
718 static inline bool legitimize_path(struct nameidata
*nd
,
719 struct path
*path
, unsigned seq
)
721 return __legitimize_path(path
, seq
, nd
->m_seq
);
724 static bool legitimize_links(struct nameidata
*nd
)
727 if (unlikely(nd
->flags
& LOOKUP_CACHED
)) {
732 for (i
= 0; i
< nd
->depth
; i
++) {
733 struct saved
*last
= nd
->stack
+ i
;
734 if (unlikely(!legitimize_path(nd
, &last
->link
, last
->seq
))) {
743 static bool legitimize_root(struct nameidata
*nd
)
745 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
746 if (!nd
->root
.mnt
|| (nd
->state
& ND_ROOT_PRESET
))
748 nd
->state
|= ND_ROOT_GRABBED
;
749 return legitimize_path(nd
, &nd
->root
, nd
->root_seq
);
753 * Path walking has 2 modes, rcu-walk and ref-walk (see
754 * Documentation/filesystems/path-lookup.txt). In situations when we can't
755 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
756 * normal reference counts on dentries and vfsmounts to transition to ref-walk
757 * mode. Refcounts are grabbed at the last known good point before rcu-walk
758 * got stuck, so ref-walk may continue from there. If this is not successful
759 * (eg. a seqcount has changed), then failure is returned and it's up to caller
760 * to restart the path walk from the beginning in ref-walk mode.
764 * try_to_unlazy - try to switch to ref-walk mode.
765 * @nd: nameidata pathwalk data
766 * Returns: true on success, false on failure
768 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
770 * Must be called from rcu-walk context.
771 * Nothing should touch nameidata between try_to_unlazy() failure and
774 static bool try_to_unlazy(struct nameidata
*nd
)
776 struct dentry
*parent
= nd
->path
.dentry
;
778 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
780 if (unlikely(!legitimize_links(nd
)))
782 if (unlikely(!legitimize_path(nd
, &nd
->path
, nd
->seq
)))
784 if (unlikely(!legitimize_root(nd
)))
787 BUG_ON(nd
->inode
!= parent
->d_inode
);
792 nd
->path
.dentry
= NULL
;
799 * try_to_unlazy_next - try to switch to ref-walk mode.
800 * @nd: nameidata pathwalk data
801 * @dentry: next dentry to step into
802 * Returns: true on success, false on failure
804 * Similar to try_to_unlazy(), but here we have the next dentry already
805 * picked by rcu-walk and want to legitimize that in addition to the current
806 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
807 * Nothing should touch nameidata between try_to_unlazy_next() failure and
810 static bool try_to_unlazy_next(struct nameidata
*nd
, struct dentry
*dentry
)
813 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
815 if (unlikely(!legitimize_links(nd
)))
817 res
= __legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
);
823 if (unlikely(!lockref_get_not_dead(&nd
->path
.dentry
->d_lockref
)))
827 * We need to move both the parent and the dentry from the RCU domain
828 * to be properly refcounted. And the sequence number in the dentry
829 * validates *both* dentry counters, since we checked the sequence
830 * number of the parent after we got the child sequence number. So we
831 * know the parent must still be valid if the child sequence number is
833 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
)))
835 if (read_seqcount_retry(&dentry
->d_seq
, nd
->next_seq
))
838 * Sequence counts matched. Now make sure that the root is
839 * still valid and get it if required.
841 if (unlikely(!legitimize_root(nd
)))
849 nd
->path
.dentry
= NULL
;
859 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
861 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
862 return dentry
->d_op
->d_revalidate(dentry
, flags
);
868 * complete_walk - successful completion of path walk
869 * @nd: pointer nameidata
871 * If we had been in RCU mode, drop out of it and legitimize nd->path.
872 * Revalidate the final result, unless we'd already done that during
873 * the path walk or the filesystem doesn't ask for it. Return 0 on
874 * success, -error on failure. In case of failure caller does not
875 * need to drop nd->path.
877 static int complete_walk(struct nameidata
*nd
)
879 struct dentry
*dentry
= nd
->path
.dentry
;
882 if (nd
->flags
& LOOKUP_RCU
) {
884 * We don't want to zero nd->root for scoped-lookups or
885 * externally-managed nd->root.
887 if (!(nd
->state
& ND_ROOT_PRESET
))
888 if (!(nd
->flags
& LOOKUP_IS_SCOPED
))
890 nd
->flags
&= ~LOOKUP_CACHED
;
891 if (!try_to_unlazy(nd
))
895 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
)) {
897 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
898 * ever step outside the root during lookup" and should already
899 * be guaranteed by the rest of namei, we want to avoid a namei
900 * BUG resulting in userspace being given a path that was not
901 * scoped within the root at some point during the lookup.
903 * So, do a final sanity-check to make sure that in the
904 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
905 * we won't silently return an fd completely outside of the
906 * requested root to userspace.
908 * Userspace could move the path outside the root after this
909 * check, but as discussed elsewhere this is not a concern (the
910 * resolved file was inside the root at some point).
912 if (!path_is_under(&nd
->path
, &nd
->root
))
916 if (likely(!(nd
->state
& ND_JUMPED
)))
919 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
922 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
932 static int set_root(struct nameidata
*nd
)
934 struct fs_struct
*fs
= current
->fs
;
937 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
938 * still have to ensure it doesn't happen because it will cause a breakout
941 if (WARN_ON(nd
->flags
& LOOKUP_IS_SCOPED
))
942 return -ENOTRECOVERABLE
;
944 if (nd
->flags
& LOOKUP_RCU
) {
948 seq
= read_seqcount_begin(&fs
->seq
);
950 nd
->root_seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
951 } while (read_seqcount_retry(&fs
->seq
, seq
));
953 get_fs_root(fs
, &nd
->root
);
954 nd
->state
|= ND_ROOT_GRABBED
;
959 static int nd_jump_root(struct nameidata
*nd
)
961 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
963 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
)) {
964 /* Absolute path arguments to path_init() are allowed. */
965 if (nd
->path
.mnt
!= NULL
&& nd
->path
.mnt
!= nd
->root
.mnt
)
969 int error
= set_root(nd
);
973 if (nd
->flags
& LOOKUP_RCU
) {
977 nd
->inode
= d
->d_inode
;
978 nd
->seq
= nd
->root_seq
;
979 if (read_seqcount_retry(&d
->d_seq
, nd
->seq
))
985 nd
->inode
= nd
->path
.dentry
->d_inode
;
987 nd
->state
|= ND_JUMPED
;
992 * Helper to directly jump to a known parsed path from ->get_link,
993 * caller must have taken a reference to path beforehand.
995 int nd_jump_link(const struct path
*path
)
998 struct nameidata
*nd
= current
->nameidata
;
1000 if (unlikely(nd
->flags
& LOOKUP_NO_MAGICLINKS
))
1004 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
)) {
1005 if (nd
->path
.mnt
!= path
->mnt
)
1008 /* Not currently safe for scoped-lookups. */
1009 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
))
1012 path_put(&nd
->path
);
1014 nd
->inode
= nd
->path
.dentry
->d_inode
;
1015 nd
->state
|= ND_JUMPED
;
1023 static inline void put_link(struct nameidata
*nd
)
1025 struct saved
*last
= nd
->stack
+ --nd
->depth
;
1026 do_delayed_call(&last
->done
);
1027 if (!(nd
->flags
& LOOKUP_RCU
))
1028 path_put(&last
->link
);
1031 static int sysctl_protected_symlinks __read_mostly
;
1032 static int sysctl_protected_hardlinks __read_mostly
;
1033 static int sysctl_protected_fifos __read_mostly
;
1034 static int sysctl_protected_regular __read_mostly
;
1036 #ifdef CONFIG_SYSCTL
1037 static struct ctl_table namei_sysctls
[] = {
1039 .procname
= "protected_symlinks",
1040 .data
= &sysctl_protected_symlinks
,
1041 .maxlen
= sizeof(int),
1043 .proc_handler
= proc_dointvec_minmax
,
1044 .extra1
= SYSCTL_ZERO
,
1045 .extra2
= SYSCTL_ONE
,
1048 .procname
= "protected_hardlinks",
1049 .data
= &sysctl_protected_hardlinks
,
1050 .maxlen
= sizeof(int),
1052 .proc_handler
= proc_dointvec_minmax
,
1053 .extra1
= SYSCTL_ZERO
,
1054 .extra2
= SYSCTL_ONE
,
1057 .procname
= "protected_fifos",
1058 .data
= &sysctl_protected_fifos
,
1059 .maxlen
= sizeof(int),
1061 .proc_handler
= proc_dointvec_minmax
,
1062 .extra1
= SYSCTL_ZERO
,
1063 .extra2
= SYSCTL_TWO
,
1066 .procname
= "protected_regular",
1067 .data
= &sysctl_protected_regular
,
1068 .maxlen
= sizeof(int),
1070 .proc_handler
= proc_dointvec_minmax
,
1071 .extra1
= SYSCTL_ZERO
,
1072 .extra2
= SYSCTL_TWO
,
1077 static int __init
init_fs_namei_sysctls(void)
1079 register_sysctl_init("fs", namei_sysctls
);
1082 fs_initcall(init_fs_namei_sysctls
);
1084 #endif /* CONFIG_SYSCTL */
1087 * may_follow_link - Check symlink following for unsafe situations
1088 * @nd: nameidata pathwalk data
1089 * @inode: Used for idmapping.
1091 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1092 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1093 * in a sticky world-writable directory. This is to protect privileged
1094 * processes from failing races against path names that may change out
1095 * from under them by way of other users creating malicious symlinks.
1096 * It will permit symlinks to be followed only when outside a sticky
1097 * world-writable directory, or when the uid of the symlink and follower
1098 * match, or when the directory owner matches the symlink's owner.
1100 * Returns 0 if following the symlink is allowed, -ve on error.
1102 static inline int may_follow_link(struct nameidata
*nd
, const struct inode
*inode
)
1104 struct mnt_idmap
*idmap
;
1107 if (!sysctl_protected_symlinks
)
1110 idmap
= mnt_idmap(nd
->path
.mnt
);
1111 vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
1112 /* Allowed if owner and follower match. */
1113 if (vfsuid_eq_kuid(vfsuid
, current_fsuid()))
1116 /* Allowed if parent directory not sticky and world-writable. */
1117 if ((nd
->dir_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
1120 /* Allowed if parent directory and link owner match. */
1121 if (vfsuid_valid(nd
->dir_vfsuid
) && vfsuid_eq(nd
->dir_vfsuid
, vfsuid
))
1124 if (nd
->flags
& LOOKUP_RCU
)
1127 audit_inode(nd
->name
, nd
->stack
[0].link
.dentry
, 0);
1128 audit_log_path_denied(AUDIT_ANOM_LINK
, "follow_link");
1133 * safe_hardlink_source - Check for safe hardlink conditions
1134 * @idmap: idmap of the mount the inode was found from
1135 * @inode: the source inode to hardlink from
1137 * Return false if at least one of the following conditions:
1138 * - inode is not a regular file
1140 * - inode is setgid and group-exec
1141 * - access failure for read and write
1143 * Otherwise returns true.
1145 static bool safe_hardlink_source(struct mnt_idmap
*idmap
,
1146 struct inode
*inode
)
1148 umode_t mode
= inode
->i_mode
;
1150 /* Special files should not get pinned to the filesystem. */
1154 /* Setuid files should not get pinned to the filesystem. */
1158 /* Executable setgid files should not get pinned to the filesystem. */
1159 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
1162 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1163 if (inode_permission(idmap
, inode
, MAY_READ
| MAY_WRITE
))
1170 * may_linkat - Check permissions for creating a hardlink
1171 * @idmap: idmap of the mount the inode was found from
1172 * @link: the source to hardlink from
1174 * Block hardlink when all of:
1175 * - sysctl_protected_hardlinks enabled
1176 * - fsuid does not match inode
1177 * - hardlink source is unsafe (see safe_hardlink_source() above)
1178 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1180 * If the inode has been found through an idmapped mount the idmap of
1181 * the vfsmount must be passed through @idmap. This function will then take
1182 * care to map the inode according to @idmap before checking permissions.
1183 * On non-idmapped mounts or if permission checking is to be performed on the
1184 * raw inode simply pass @nop_mnt_idmap.
1186 * Returns 0 if successful, -ve on error.
1188 int may_linkat(struct mnt_idmap
*idmap
, const struct path
*link
)
1190 struct inode
*inode
= link
->dentry
->d_inode
;
1192 /* Inode writeback is not safe when the uid or gid are invalid. */
1193 if (!vfsuid_valid(i_uid_into_vfsuid(idmap
, inode
)) ||
1194 !vfsgid_valid(i_gid_into_vfsgid(idmap
, inode
)))
1197 if (!sysctl_protected_hardlinks
)
1200 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1201 * otherwise, it must be a safe source.
1203 if (safe_hardlink_source(idmap
, inode
) ||
1204 inode_owner_or_capable(idmap
, inode
))
1207 audit_log_path_denied(AUDIT_ANOM_LINK
, "linkat");
1212 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1213 * should be allowed, or not, on files that already
1215 * @idmap: idmap of the mount the inode was found from
1216 * @nd: nameidata pathwalk data
1217 * @inode: the inode of the file to open
1219 * Block an O_CREAT open of a FIFO (or a regular file) when:
1220 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1221 * - the file already exists
1222 * - we are in a sticky directory
1223 * - we don't own the file
1224 * - the owner of the directory doesn't own the file
1225 * - the directory is world writable
1226 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1227 * the directory doesn't have to be world writable: being group writable will
1230 * If the inode has been found through an idmapped mount the idmap of
1231 * the vfsmount must be passed through @idmap. This function will then take
1232 * care to map the inode according to @idmap before checking permissions.
1233 * On non-idmapped mounts or if permission checking is to be performed on the
1234 * raw inode simply pass @nop_mnt_idmap.
1236 * Returns 0 if the open is allowed, -ve on error.
1238 static int may_create_in_sticky(struct mnt_idmap
*idmap
,
1239 struct nameidata
*nd
, struct inode
*const inode
)
1241 umode_t dir_mode
= nd
->dir_mode
;
1242 vfsuid_t dir_vfsuid
= nd
->dir_vfsuid
;
1244 if ((!sysctl_protected_fifos
&& S_ISFIFO(inode
->i_mode
)) ||
1245 (!sysctl_protected_regular
&& S_ISREG(inode
->i_mode
)) ||
1246 likely(!(dir_mode
& S_ISVTX
)) ||
1247 vfsuid_eq(i_uid_into_vfsuid(idmap
, inode
), dir_vfsuid
) ||
1248 vfsuid_eq_kuid(i_uid_into_vfsuid(idmap
, inode
), current_fsuid()))
1251 if (likely(dir_mode
& 0002) ||
1253 ((sysctl_protected_fifos
>= 2 && S_ISFIFO(inode
->i_mode
)) ||
1254 (sysctl_protected_regular
>= 2 && S_ISREG(inode
->i_mode
))))) {
1255 const char *operation
= S_ISFIFO(inode
->i_mode
) ?
1256 "sticky_create_fifo" :
1257 "sticky_create_regular";
1258 audit_log_path_denied(AUDIT_ANOM_CREAT
, operation
);
1265 * follow_up - Find the mountpoint of path's vfsmount
1267 * Given a path, find the mountpoint of its source file system.
1268 * Replace @path with the path of the mountpoint in the parent mount.
1271 * Return 1 if we went up a level and 0 if we were already at the
1274 int follow_up(struct path
*path
)
1276 struct mount
*mnt
= real_mount(path
->mnt
);
1277 struct mount
*parent
;
1278 struct dentry
*mountpoint
;
1280 read_seqlock_excl(&mount_lock
);
1281 parent
= mnt
->mnt_parent
;
1282 if (parent
== mnt
) {
1283 read_sequnlock_excl(&mount_lock
);
1286 mntget(&parent
->mnt
);
1287 mountpoint
= dget(mnt
->mnt_mountpoint
);
1288 read_sequnlock_excl(&mount_lock
);
1290 path
->dentry
= mountpoint
;
1292 path
->mnt
= &parent
->mnt
;
1295 EXPORT_SYMBOL(follow_up
);
1297 static bool choose_mountpoint_rcu(struct mount
*m
, const struct path
*root
,
1298 struct path
*path
, unsigned *seqp
)
1300 while (mnt_has_parent(m
)) {
1301 struct dentry
*mountpoint
= m
->mnt_mountpoint
;
1304 if (unlikely(root
->dentry
== mountpoint
&&
1305 root
->mnt
== &m
->mnt
))
1307 if (mountpoint
!= m
->mnt
.mnt_root
) {
1308 path
->mnt
= &m
->mnt
;
1309 path
->dentry
= mountpoint
;
1310 *seqp
= read_seqcount_begin(&mountpoint
->d_seq
);
1317 static bool choose_mountpoint(struct mount
*m
, const struct path
*root
,
1324 unsigned seq
, mseq
= read_seqbegin(&mount_lock
);
1326 found
= choose_mountpoint_rcu(m
, root
, path
, &seq
);
1327 if (unlikely(!found
)) {
1328 if (!read_seqretry(&mount_lock
, mseq
))
1331 if (likely(__legitimize_path(path
, seq
, mseq
)))
1343 * Perform an automount
1344 * - return -EISDIR to tell follow_managed() to stop and return the path we
1347 static int follow_automount(struct path
*path
, int *count
, unsigned lookup_flags
)
1349 struct dentry
*dentry
= path
->dentry
;
1351 /* We don't want to mount if someone's just doing a stat -
1352 * unless they're stat'ing a directory and appended a '/' to
1355 * We do, however, want to mount if someone wants to open or
1356 * create a file of any type under the mountpoint, wants to
1357 * traverse through the mountpoint or wants to open the
1358 * mounted directory. Also, autofs may mark negative dentries
1359 * as being automount points. These will need the attentions
1360 * of the daemon to instantiate them before they can be used.
1362 if (!(lookup_flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
1363 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
1367 if (count
&& (*count
)++ >= MAXSYMLINKS
)
1370 return finish_automount(dentry
->d_op
->d_automount(path
), path
);
1374 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1375 * dentries are pinned but not locked here, so negative dentry can go
1376 * positive right under us. Use of smp_load_acquire() provides a barrier
1377 * sufficient for ->d_inode and ->d_flags consistency.
1379 static int __traverse_mounts(struct path
*path
, unsigned flags
, bool *jumped
,
1380 int *count
, unsigned lookup_flags
)
1382 struct vfsmount
*mnt
= path
->mnt
;
1383 bool need_mntput
= false;
1386 while (flags
& DCACHE_MANAGED_DENTRY
) {
1387 /* Allow the filesystem to manage the transit without i_mutex
1389 if (flags
& DCACHE_MANAGE_TRANSIT
) {
1390 ret
= path
->dentry
->d_op
->d_manage(path
, false);
1391 flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1396 if (flags
& DCACHE_MOUNTED
) { // something's mounted on it..
1397 struct vfsmount
*mounted
= lookup_mnt(path
);
1398 if (mounted
) { // ... in our namespace
1402 path
->mnt
= mounted
;
1403 path
->dentry
= dget(mounted
->mnt_root
);
1404 // here we know it's positive
1405 flags
= path
->dentry
->d_flags
;
1411 if (!(flags
& DCACHE_NEED_AUTOMOUNT
))
1414 // uncovered automount point
1415 ret
= follow_automount(path
, count
, lookup_flags
);
1416 flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1423 // possible if you race with several mount --move
1424 if (need_mntput
&& path
->mnt
== mnt
)
1426 if (!ret
&& unlikely(d_flags_negative(flags
)))
1428 *jumped
= need_mntput
;
1432 static inline int traverse_mounts(struct path
*path
, bool *jumped
,
1433 int *count
, unsigned lookup_flags
)
1435 unsigned flags
= smp_load_acquire(&path
->dentry
->d_flags
);
1438 if (likely(!(flags
& DCACHE_MANAGED_DENTRY
))) {
1440 if (unlikely(d_flags_negative(flags
)))
1444 return __traverse_mounts(path
, flags
, jumped
, count
, lookup_flags
);
1447 int follow_down_one(struct path
*path
)
1449 struct vfsmount
*mounted
;
1451 mounted
= lookup_mnt(path
);
1455 path
->mnt
= mounted
;
1456 path
->dentry
= dget(mounted
->mnt_root
);
1461 EXPORT_SYMBOL(follow_down_one
);
1464 * Follow down to the covering mount currently visible to userspace. At each
1465 * point, the filesystem owning that dentry may be queried as to whether the
1466 * caller is permitted to proceed or not.
1468 int follow_down(struct path
*path
, unsigned int flags
)
1470 struct vfsmount
*mnt
= path
->mnt
;
1472 int ret
= traverse_mounts(path
, &jumped
, NULL
, flags
);
1474 if (path
->mnt
!= mnt
)
1478 EXPORT_SYMBOL(follow_down
);
1481 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1482 * we meet a managed dentry that would need blocking.
1484 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
)
1486 struct dentry
*dentry
= path
->dentry
;
1487 unsigned int flags
= dentry
->d_flags
;
1489 if (likely(!(flags
& DCACHE_MANAGED_DENTRY
)))
1492 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1497 * Don't forget we might have a non-mountpoint managed dentry
1498 * that wants to block transit.
1500 if (unlikely(flags
& DCACHE_MANAGE_TRANSIT
)) {
1501 int res
= dentry
->d_op
->d_manage(path
, true);
1503 return res
== -EISDIR
;
1504 flags
= dentry
->d_flags
;
1507 if (flags
& DCACHE_MOUNTED
) {
1508 struct mount
*mounted
= __lookup_mnt(path
->mnt
, dentry
);
1510 path
->mnt
= &mounted
->mnt
;
1511 dentry
= path
->dentry
= mounted
->mnt
.mnt_root
;
1512 nd
->state
|= ND_JUMPED
;
1513 nd
->next_seq
= read_seqcount_begin(&dentry
->d_seq
);
1514 flags
= dentry
->d_flags
;
1515 // makes sure that non-RCU pathwalk could reach
1517 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1521 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1524 return !(flags
& DCACHE_NEED_AUTOMOUNT
);
1528 static inline int handle_mounts(struct nameidata
*nd
, struct dentry
*dentry
,
1534 path
->mnt
= nd
->path
.mnt
;
1535 path
->dentry
= dentry
;
1536 if (nd
->flags
& LOOKUP_RCU
) {
1537 unsigned int seq
= nd
->next_seq
;
1538 if (likely(__follow_mount_rcu(nd
, path
)))
1540 // *path and nd->next_seq might've been clobbered
1541 path
->mnt
= nd
->path
.mnt
;
1542 path
->dentry
= dentry
;
1544 if (!try_to_unlazy_next(nd
, dentry
))
1547 ret
= traverse_mounts(path
, &jumped
, &nd
->total_link_count
, nd
->flags
);
1549 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1552 nd
->state
|= ND_JUMPED
;
1554 if (unlikely(ret
)) {
1556 if (path
->mnt
!= nd
->path
.mnt
)
1563 * This looks up the name in dcache and possibly revalidates the found dentry.
1564 * NULL is returned if the dentry does not exist in the cache.
1566 static struct dentry
*lookup_dcache(const struct qstr
*name
,
1570 struct dentry
*dentry
= d_lookup(dir
, name
);
1572 int error
= d_revalidate(dentry
, flags
);
1573 if (unlikely(error
<= 0)) {
1575 d_invalidate(dentry
);
1577 return ERR_PTR(error
);
1584 * Parent directory has inode locked exclusive. This is one
1585 * and only case when ->lookup() gets called on non in-lookup
1586 * dentries - as the matter of fact, this only gets called
1587 * when directory is guaranteed to have no in-lookup children
1590 struct dentry
*lookup_one_qstr_excl(const struct qstr
*name
,
1591 struct dentry
*base
,
1594 struct dentry
*dentry
= lookup_dcache(name
, base
, flags
);
1596 struct inode
*dir
= base
->d_inode
;
1601 /* Don't create child dentry for a dead directory. */
1602 if (unlikely(IS_DEADDIR(dir
)))
1603 return ERR_PTR(-ENOENT
);
1605 dentry
= d_alloc(base
, name
);
1606 if (unlikely(!dentry
))
1607 return ERR_PTR(-ENOMEM
);
1609 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1610 if (unlikely(old
)) {
1616 EXPORT_SYMBOL(lookup_one_qstr_excl
);
1618 static struct dentry
*lookup_fast(struct nameidata
*nd
)
1620 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1624 * Rename seqlock is not required here because in the off chance
1625 * of a false negative due to a concurrent rename, the caller is
1626 * going to fall back to non-racy lookup.
1628 if (nd
->flags
& LOOKUP_RCU
) {
1629 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &nd
->next_seq
);
1630 if (unlikely(!dentry
)) {
1631 if (!try_to_unlazy(nd
))
1632 return ERR_PTR(-ECHILD
);
1637 * This sequence count validates that the parent had no
1638 * changes while we did the lookup of the dentry above.
1640 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1641 return ERR_PTR(-ECHILD
);
1643 status
= d_revalidate(dentry
, nd
->flags
);
1644 if (likely(status
> 0))
1646 if (!try_to_unlazy_next(nd
, dentry
))
1647 return ERR_PTR(-ECHILD
);
1648 if (status
== -ECHILD
)
1649 /* we'd been told to redo it in non-rcu mode */
1650 status
= d_revalidate(dentry
, nd
->flags
);
1652 dentry
= __d_lookup(parent
, &nd
->last
);
1653 if (unlikely(!dentry
))
1655 status
= d_revalidate(dentry
, nd
->flags
);
1657 if (unlikely(status
<= 0)) {
1659 d_invalidate(dentry
);
1661 return ERR_PTR(status
);
1666 /* Fast lookup failed, do it the slow way */
1667 static struct dentry
*__lookup_slow(const struct qstr
*name
,
1671 struct dentry
*dentry
, *old
;
1672 struct inode
*inode
= dir
->d_inode
;
1673 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
1675 /* Don't go there if it's already dead */
1676 if (unlikely(IS_DEADDIR(inode
)))
1677 return ERR_PTR(-ENOENT
);
1679 dentry
= d_alloc_parallel(dir
, name
, &wq
);
1682 if (unlikely(!d_in_lookup(dentry
))) {
1683 int error
= d_revalidate(dentry
, flags
);
1684 if (unlikely(error
<= 0)) {
1686 d_invalidate(dentry
);
1691 dentry
= ERR_PTR(error
);
1694 old
= inode
->i_op
->lookup(inode
, dentry
, flags
);
1695 d_lookup_done(dentry
);
1696 if (unlikely(old
)) {
1704 static struct dentry
*lookup_slow(const struct qstr
*name
,
1708 struct inode
*inode
= dir
->d_inode
;
1710 inode_lock_shared(inode
);
1711 res
= __lookup_slow(name
, dir
, flags
);
1712 inode_unlock_shared(inode
);
1716 static inline int may_lookup(struct mnt_idmap
*idmap
,
1717 struct nameidata
*nd
)
1719 if (nd
->flags
& LOOKUP_RCU
) {
1720 int err
= inode_permission(idmap
, nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1721 if (err
!= -ECHILD
|| !try_to_unlazy(nd
))
1724 return inode_permission(idmap
, nd
->inode
, MAY_EXEC
);
1727 static int reserve_stack(struct nameidata
*nd
, struct path
*link
)
1729 if (unlikely(nd
->total_link_count
++ >= MAXSYMLINKS
))
1732 if (likely(nd
->depth
!= EMBEDDED_LEVELS
))
1734 if (likely(nd
->stack
!= nd
->internal
))
1736 if (likely(nd_alloc_stack(nd
)))
1739 if (nd
->flags
& LOOKUP_RCU
) {
1740 // we need to grab link before we do unlazy. And we can't skip
1741 // unlazy even if we fail to grab the link - cleanup needs it
1742 bool grabbed_link
= legitimize_path(nd
, link
, nd
->next_seq
);
1744 if (!try_to_unlazy(nd
) || !grabbed_link
)
1747 if (nd_alloc_stack(nd
))
1753 enum {WALK_TRAILING
= 1, WALK_MORE
= 2, WALK_NOFOLLOW
= 4};
1755 static const char *pick_link(struct nameidata
*nd
, struct path
*link
,
1756 struct inode
*inode
, int flags
)
1760 int error
= reserve_stack(nd
, link
);
1762 if (unlikely(error
)) {
1763 if (!(nd
->flags
& LOOKUP_RCU
))
1765 return ERR_PTR(error
);
1767 last
= nd
->stack
+ nd
->depth
++;
1769 clear_delayed_call(&last
->done
);
1770 last
->seq
= nd
->next_seq
;
1772 if (flags
& WALK_TRAILING
) {
1773 error
= may_follow_link(nd
, inode
);
1774 if (unlikely(error
))
1775 return ERR_PTR(error
);
1778 if (unlikely(nd
->flags
& LOOKUP_NO_SYMLINKS
) ||
1779 unlikely(link
->mnt
->mnt_flags
& MNT_NOSYMFOLLOW
))
1780 return ERR_PTR(-ELOOP
);
1782 if (!(nd
->flags
& LOOKUP_RCU
)) {
1783 touch_atime(&last
->link
);
1785 } else if (atime_needs_update(&last
->link
, inode
)) {
1786 if (!try_to_unlazy(nd
))
1787 return ERR_PTR(-ECHILD
);
1788 touch_atime(&last
->link
);
1791 error
= security_inode_follow_link(link
->dentry
, inode
,
1792 nd
->flags
& LOOKUP_RCU
);
1793 if (unlikely(error
))
1794 return ERR_PTR(error
);
1796 res
= READ_ONCE(inode
->i_link
);
1798 const char * (*get
)(struct dentry
*, struct inode
*,
1799 struct delayed_call
*);
1800 get
= inode
->i_op
->get_link
;
1801 if (nd
->flags
& LOOKUP_RCU
) {
1802 res
= get(NULL
, inode
, &last
->done
);
1803 if (res
== ERR_PTR(-ECHILD
) && try_to_unlazy(nd
))
1804 res
= get(link
->dentry
, inode
, &last
->done
);
1806 res
= get(link
->dentry
, inode
, &last
->done
);
1814 error
= nd_jump_root(nd
);
1815 if (unlikely(error
))
1816 return ERR_PTR(error
);
1817 while (unlikely(*++res
== '/'))
1822 all_done
: // pure jump
1828 * Do we need to follow links? We _really_ want to be able
1829 * to do this check without having to look at inode->i_op,
1830 * so we keep a cache of "no, this doesn't need follow_link"
1831 * for the common case.
1833 * NOTE: dentry must be what nd->next_seq had been sampled from.
1835 static const char *step_into(struct nameidata
*nd
, int flags
,
1836 struct dentry
*dentry
)
1839 struct inode
*inode
;
1840 int err
= handle_mounts(nd
, dentry
, &path
);
1843 return ERR_PTR(err
);
1844 inode
= path
.dentry
->d_inode
;
1845 if (likely(!d_is_symlink(path
.dentry
)) ||
1846 ((flags
& WALK_TRAILING
) && !(nd
->flags
& LOOKUP_FOLLOW
)) ||
1847 (flags
& WALK_NOFOLLOW
)) {
1848 /* not a symlink or should not follow */
1849 if (nd
->flags
& LOOKUP_RCU
) {
1850 if (read_seqcount_retry(&path
.dentry
->d_seq
, nd
->next_seq
))
1851 return ERR_PTR(-ECHILD
);
1852 if (unlikely(!inode
))
1853 return ERR_PTR(-ENOENT
);
1855 dput(nd
->path
.dentry
);
1856 if (nd
->path
.mnt
!= path
.mnt
)
1857 mntput(nd
->path
.mnt
);
1861 nd
->seq
= nd
->next_seq
;
1864 if (nd
->flags
& LOOKUP_RCU
) {
1865 /* make sure that d_is_symlink above matches inode */
1866 if (read_seqcount_retry(&path
.dentry
->d_seq
, nd
->next_seq
))
1867 return ERR_PTR(-ECHILD
);
1869 if (path
.mnt
== nd
->path
.mnt
)
1872 return pick_link(nd
, &path
, inode
, flags
);
1875 static struct dentry
*follow_dotdot_rcu(struct nameidata
*nd
)
1877 struct dentry
*parent
, *old
;
1879 if (path_equal(&nd
->path
, &nd
->root
))
1881 if (unlikely(nd
->path
.dentry
== nd
->path
.mnt
->mnt_root
)) {
1884 if (!choose_mountpoint_rcu(real_mount(nd
->path
.mnt
),
1885 &nd
->root
, &path
, &seq
))
1887 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1888 return ERR_PTR(-ECHILD
);
1890 nd
->inode
= path
.dentry
->d_inode
;
1892 // makes sure that non-RCU pathwalk could reach this state
1893 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1894 return ERR_PTR(-ECHILD
);
1895 /* we know that mountpoint was pinned */
1897 old
= nd
->path
.dentry
;
1898 parent
= old
->d_parent
;
1899 nd
->next_seq
= read_seqcount_begin(&parent
->d_seq
);
1900 // makes sure that non-RCU pathwalk could reach this state
1901 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1902 return ERR_PTR(-ECHILD
);
1903 if (unlikely(!path_connected(nd
->path
.mnt
, parent
)))
1904 return ERR_PTR(-ECHILD
);
1907 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1908 return ERR_PTR(-ECHILD
);
1909 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
1910 return ERR_PTR(-ECHILD
);
1911 nd
->next_seq
= nd
->seq
;
1912 return nd
->path
.dentry
;
1915 static struct dentry
*follow_dotdot(struct nameidata
*nd
)
1917 struct dentry
*parent
;
1919 if (path_equal(&nd
->path
, &nd
->root
))
1921 if (unlikely(nd
->path
.dentry
== nd
->path
.mnt
->mnt_root
)) {
1924 if (!choose_mountpoint(real_mount(nd
->path
.mnt
),
1927 path_put(&nd
->path
);
1929 nd
->inode
= path
.dentry
->d_inode
;
1930 if (unlikely(nd
->flags
& LOOKUP_NO_XDEV
))
1931 return ERR_PTR(-EXDEV
);
1933 /* rare case of legitimate dget_parent()... */
1934 parent
= dget_parent(nd
->path
.dentry
);
1935 if (unlikely(!path_connected(nd
->path
.mnt
, parent
))) {
1937 return ERR_PTR(-ENOENT
);
1942 if (unlikely(nd
->flags
& LOOKUP_BENEATH
))
1943 return ERR_PTR(-EXDEV
);
1944 return dget(nd
->path
.dentry
);
1947 static const char *handle_dots(struct nameidata
*nd
, int type
)
1949 if (type
== LAST_DOTDOT
) {
1950 const char *error
= NULL
;
1951 struct dentry
*parent
;
1953 if (!nd
->root
.mnt
) {
1954 error
= ERR_PTR(set_root(nd
));
1958 if (nd
->flags
& LOOKUP_RCU
)
1959 parent
= follow_dotdot_rcu(nd
);
1961 parent
= follow_dotdot(nd
);
1963 return ERR_CAST(parent
);
1964 error
= step_into(nd
, WALK_NOFOLLOW
, parent
);
1965 if (unlikely(error
))
1968 if (unlikely(nd
->flags
& LOOKUP_IS_SCOPED
)) {
1970 * If there was a racing rename or mount along our
1971 * path, then we can't be sure that ".." hasn't jumped
1972 * above nd->root (and so userspace should retry or use
1976 if (__read_seqcount_retry(&mount_lock
.seqcount
, nd
->m_seq
))
1977 return ERR_PTR(-EAGAIN
);
1978 if (__read_seqcount_retry(&rename_lock
.seqcount
, nd
->r_seq
))
1979 return ERR_PTR(-EAGAIN
);
1985 static const char *walk_component(struct nameidata
*nd
, int flags
)
1987 struct dentry
*dentry
;
1989 * "." and ".." are special - ".." especially so because it has
1990 * to be able to know about the current root directory and
1991 * parent relationships.
1993 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
1994 if (!(flags
& WALK_MORE
) && nd
->depth
)
1996 return handle_dots(nd
, nd
->last_type
);
1998 dentry
= lookup_fast(nd
);
2000 return ERR_CAST(dentry
);
2001 if (unlikely(!dentry
)) {
2002 dentry
= lookup_slow(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2004 return ERR_CAST(dentry
);
2006 if (!(flags
& WALK_MORE
) && nd
->depth
)
2008 return step_into(nd
, flags
, dentry
);
2012 * We can do the critical dentry name comparison and hashing
2013 * operations one word at a time, but we are limited to:
2015 * - Architectures with fast unaligned word accesses. We could
2016 * do a "get_unaligned()" if this helps and is sufficiently
2019 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2020 * do not trap on the (extremely unlikely) case of a page
2021 * crossing operation.
2023 * - Furthermore, we need an efficient 64-bit compile for the
2024 * 64-bit case in order to generate the "number of bytes in
2025 * the final mask". Again, that could be replaced with a
2026 * efficient population count instruction or similar.
2028 #ifdef CONFIG_DCACHE_WORD_ACCESS
2030 #include <asm/word-at-a-time.h>
2034 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2036 #elif defined(CONFIG_64BIT)
2038 * Register pressure in the mixing function is an issue, particularly
2039 * on 32-bit x86, but almost any function requires one state value and
2040 * one temporary. Instead, use a function designed for two state values
2041 * and no temporaries.
2043 * This function cannot create a collision in only two iterations, so
2044 * we have two iterations to achieve avalanche. In those two iterations,
2045 * we have six layers of mixing, which is enough to spread one bit's
2046 * influence out to 2^6 = 64 state bits.
2048 * Rotate constants are scored by considering either 64 one-bit input
2049 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2050 * probability of that delta causing a change to each of the 128 output
2051 * bits, using a sample of random initial states.
2053 * The Shannon entropy of the computed probabilities is then summed
2054 * to produce a score. Ideally, any input change has a 50% chance of
2055 * toggling any given output bit.
2057 * Mixing scores (in bits) for (12,45):
2058 * Input delta: 1-bit 2-bit
2059 * 1 round: 713.3 42542.6
2060 * 2 rounds: 2753.7 140389.8
2061 * 3 rounds: 5954.1 233458.2
2062 * 4 rounds: 7862.6 256672.2
2063 * Perfect: 8192 258048
2064 * (64*128) (64*63/2 * 128)
2066 #define HASH_MIX(x, y, a) \
2068 y ^= x, x = rol64(x,12),\
2069 x += y, y = rol64(y,45),\
2073 * Fold two longs into one 32-bit hash value. This must be fast, but
2074 * latency isn't quite as critical, as there is a fair bit of additional
2075 * work done before the hash value is used.
2077 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
2079 y
^= x
* GOLDEN_RATIO_64
;
2080 y
*= GOLDEN_RATIO_64
;
2084 #else /* 32-bit case */
2087 * Mixing scores (in bits) for (7,20):
2088 * Input delta: 1-bit 2-bit
2089 * 1 round: 330.3 9201.6
2090 * 2 rounds: 1246.4 25475.4
2091 * 3 rounds: 1907.1 31295.1
2092 * 4 rounds: 2042.3 31718.6
2093 * Perfect: 2048 31744
2094 * (32*64) (32*31/2 * 64)
2096 #define HASH_MIX(x, y, a) \
2098 y ^= x, x = rol32(x, 7),\
2099 x += y, y = rol32(y,20),\
2102 static inline unsigned int fold_hash(unsigned long x
, unsigned long y
)
2104 /* Use arch-optimized multiply if one exists */
2105 return __hash_32(y
^ __hash_32(x
));
2111 * Return the hash of a string of known length. This is carfully
2112 * designed to match hash_name(), which is the more critical function.
2113 * In particular, we must end by hashing a final word containing 0..7
2114 * payload bytes, to match the way that hash_name() iterates until it
2115 * finds the delimiter after the name.
2117 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
2119 unsigned long a
, x
= 0, y
= (unsigned long)salt
;
2124 a
= load_unaligned_zeropad(name
);
2125 if (len
< sizeof(unsigned long))
2128 name
+= sizeof(unsigned long);
2129 len
-= sizeof(unsigned long);
2131 x
^= a
& bytemask_from_count(len
);
2133 return fold_hash(x
, y
);
2135 EXPORT_SYMBOL(full_name_hash
);
2137 /* Return the "hash_len" (hash and length) of a null-terminated string */
2138 u64
hashlen_string(const void *salt
, const char *name
)
2140 unsigned long a
= 0, x
= 0, y
= (unsigned long)salt
;
2141 unsigned long adata
, mask
, len
;
2142 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
2149 len
+= sizeof(unsigned long);
2151 a
= load_unaligned_zeropad(name
+len
);
2152 } while (!has_zero(a
, &adata
, &constants
));
2154 adata
= prep_zero_mask(a
, adata
, &constants
);
2155 mask
= create_zero_mask(adata
);
2156 x
^= a
& zero_bytemask(mask
);
2158 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2160 EXPORT_SYMBOL(hashlen_string
);
2163 * Calculate the length and hash of the path component, and
2164 * return the "hash_len" as the result.
2166 static inline u64
hash_name(const void *salt
, const char *name
)
2168 unsigned long a
= 0, b
, x
= 0, y
= (unsigned long)salt
;
2169 unsigned long adata
, bdata
, mask
, len
;
2170 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
2177 len
+= sizeof(unsigned long);
2179 a
= load_unaligned_zeropad(name
+len
);
2180 b
= a
^ REPEAT_BYTE('/');
2181 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
2183 adata
= prep_zero_mask(a
, adata
, &constants
);
2184 bdata
= prep_zero_mask(b
, bdata
, &constants
);
2185 mask
= create_zero_mask(adata
| bdata
);
2186 x
^= a
& zero_bytemask(mask
);
2188 return hashlen_create(fold_hash(x
, y
), len
+ find_zero(mask
));
2191 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2193 /* Return the hash of a string of known length */
2194 unsigned int full_name_hash(const void *salt
, const char *name
, unsigned int len
)
2196 unsigned long hash
= init_name_hash(salt
);
2198 hash
= partial_name_hash((unsigned char)*name
++, hash
);
2199 return end_name_hash(hash
);
2201 EXPORT_SYMBOL(full_name_hash
);
2203 /* Return the "hash_len" (hash and length) of a null-terminated string */
2204 u64
hashlen_string(const void *salt
, const char *name
)
2206 unsigned long hash
= init_name_hash(salt
);
2207 unsigned long len
= 0, c
;
2209 c
= (unsigned char)*name
;
2212 hash
= partial_name_hash(c
, hash
);
2213 c
= (unsigned char)name
[len
];
2215 return hashlen_create(end_name_hash(hash
), len
);
2217 EXPORT_SYMBOL(hashlen_string
);
2220 * We know there's a real path component here of at least
2223 static inline u64
hash_name(const void *salt
, const char *name
)
2225 unsigned long hash
= init_name_hash(salt
);
2226 unsigned long len
= 0, c
;
2228 c
= (unsigned char)*name
;
2231 hash
= partial_name_hash(c
, hash
);
2232 c
= (unsigned char)name
[len
];
2233 } while (c
&& c
!= '/');
2234 return hashlen_create(end_name_hash(hash
), len
);
2241 * This is the basic name resolution function, turning a pathname into
2242 * the final dentry. We expect 'base' to be positive and a directory.
2244 * Returns 0 and nd will have valid dentry and mnt on success.
2245 * Returns error and drops reference to input namei data on failure.
2247 static int link_path_walk(const char *name
, struct nameidata
*nd
)
2249 int depth
= 0; // depth <= nd->depth
2252 nd
->last_type
= LAST_ROOT
;
2253 nd
->flags
|= LOOKUP_PARENT
;
2255 return PTR_ERR(name
);
2259 nd
->dir_mode
= 0; // short-circuit the 'hardening' idiocy
2263 /* At this point we know we have a real path component. */
2265 struct mnt_idmap
*idmap
;
2270 idmap
= mnt_idmap(nd
->path
.mnt
);
2271 err
= may_lookup(idmap
, nd
);
2275 hash_len
= hash_name(nd
->path
.dentry
, name
);
2278 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
2280 if (name
[1] == '.') {
2282 nd
->state
|= ND_JUMPED
;
2288 if (likely(type
== LAST_NORM
)) {
2289 struct dentry
*parent
= nd
->path
.dentry
;
2290 nd
->state
&= ~ND_JUMPED
;
2291 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
2292 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
2293 err
= parent
->d_op
->d_hash(parent
, &this);
2296 hash_len
= this.hash_len
;
2301 nd
->last
.hash_len
= hash_len
;
2302 nd
->last
.name
= name
;
2303 nd
->last_type
= type
;
2305 name
+= hashlen_len(hash_len
);
2309 * If it wasn't NUL, we know it was '/'. Skip that
2310 * slash, and continue until no more slashes.
2314 } while (unlikely(*name
== '/'));
2315 if (unlikely(!*name
)) {
2317 /* pathname or trailing symlink, done */
2319 nd
->dir_vfsuid
= i_uid_into_vfsuid(idmap
, nd
->inode
);
2320 nd
->dir_mode
= nd
->inode
->i_mode
;
2321 nd
->flags
&= ~LOOKUP_PARENT
;
2324 /* last component of nested symlink */
2325 name
= nd
->stack
[--depth
].name
;
2326 link
= walk_component(nd
, 0);
2328 /* not the last component */
2329 link
= walk_component(nd
, WALK_MORE
);
2331 if (unlikely(link
)) {
2333 return PTR_ERR(link
);
2334 /* a symlink to follow */
2335 nd
->stack
[depth
++].name
= name
;
2339 if (unlikely(!d_can_lookup(nd
->path
.dentry
))) {
2340 if (nd
->flags
& LOOKUP_RCU
) {
2341 if (!try_to_unlazy(nd
))
2349 /* must be paired with terminate_walk() */
2350 static const char *path_init(struct nameidata
*nd
, unsigned flags
)
2353 const char *s
= nd
->name
->name
;
2355 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2356 if ((flags
& (LOOKUP_RCU
| LOOKUP_CACHED
)) == LOOKUP_CACHED
)
2357 return ERR_PTR(-EAGAIN
);
2360 flags
&= ~LOOKUP_RCU
;
2361 if (flags
& LOOKUP_RCU
)
2364 nd
->seq
= nd
->next_seq
= 0;
2367 nd
->state
|= ND_JUMPED
;
2369 nd
->m_seq
= __read_seqcount_begin(&mount_lock
.seqcount
);
2370 nd
->r_seq
= __read_seqcount_begin(&rename_lock
.seqcount
);
2373 if (nd
->state
& ND_ROOT_PRESET
) {
2374 struct dentry
*root
= nd
->root
.dentry
;
2375 struct inode
*inode
= root
->d_inode
;
2376 if (*s
&& unlikely(!d_can_lookup(root
)))
2377 return ERR_PTR(-ENOTDIR
);
2378 nd
->path
= nd
->root
;
2380 if (flags
& LOOKUP_RCU
) {
2381 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2382 nd
->root_seq
= nd
->seq
;
2384 path_get(&nd
->path
);
2389 nd
->root
.mnt
= NULL
;
2391 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2392 if (*s
== '/' && !(flags
& LOOKUP_IN_ROOT
)) {
2393 error
= nd_jump_root(nd
);
2394 if (unlikely(error
))
2395 return ERR_PTR(error
);
2399 /* Relative pathname -- get the starting-point it is relative to. */
2400 if (nd
->dfd
== AT_FDCWD
) {
2401 if (flags
& LOOKUP_RCU
) {
2402 struct fs_struct
*fs
= current
->fs
;
2406 seq
= read_seqcount_begin(&fs
->seq
);
2408 nd
->inode
= nd
->path
.dentry
->d_inode
;
2409 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2410 } while (read_seqcount_retry(&fs
->seq
, seq
));
2412 get_fs_pwd(current
->fs
, &nd
->path
);
2413 nd
->inode
= nd
->path
.dentry
->d_inode
;
2416 /* Caller must check execute permissions on the starting path component */
2417 struct fd f
= fdget_raw(nd
->dfd
);
2418 struct dentry
*dentry
;
2421 return ERR_PTR(-EBADF
);
2423 dentry
= f
.file
->f_path
.dentry
;
2425 if (*s
&& unlikely(!d_can_lookup(dentry
))) {
2427 return ERR_PTR(-ENOTDIR
);
2430 nd
->path
= f
.file
->f_path
;
2431 if (flags
& LOOKUP_RCU
) {
2432 nd
->inode
= nd
->path
.dentry
->d_inode
;
2433 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
2435 path_get(&nd
->path
);
2436 nd
->inode
= nd
->path
.dentry
->d_inode
;
2441 /* For scoped-lookups we need to set the root to the dirfd as well. */
2442 if (flags
& LOOKUP_IS_SCOPED
) {
2443 nd
->root
= nd
->path
;
2444 if (flags
& LOOKUP_RCU
) {
2445 nd
->root_seq
= nd
->seq
;
2447 path_get(&nd
->root
);
2448 nd
->state
|= ND_ROOT_GRABBED
;
2454 static inline const char *lookup_last(struct nameidata
*nd
)
2456 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
2457 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2459 return walk_component(nd
, WALK_TRAILING
);
2462 static int handle_lookup_down(struct nameidata
*nd
)
2464 if (!(nd
->flags
& LOOKUP_RCU
))
2465 dget(nd
->path
.dentry
);
2466 nd
->next_seq
= nd
->seq
;
2467 return PTR_ERR(step_into(nd
, WALK_NOFOLLOW
, nd
->path
.dentry
));
2470 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2471 static int path_lookupat(struct nameidata
*nd
, unsigned flags
, struct path
*path
)
2473 const char *s
= path_init(nd
, flags
);
2476 if (unlikely(flags
& LOOKUP_DOWN
) && !IS_ERR(s
)) {
2477 err
= handle_lookup_down(nd
);
2478 if (unlikely(err
< 0))
2482 while (!(err
= link_path_walk(s
, nd
)) &&
2483 (s
= lookup_last(nd
)) != NULL
)
2485 if (!err
&& unlikely(nd
->flags
& LOOKUP_MOUNTPOINT
)) {
2486 err
= handle_lookup_down(nd
);
2487 nd
->state
&= ~ND_JUMPED
; // no d_weak_revalidate(), please...
2490 err
= complete_walk(nd
);
2492 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
)
2493 if (!d_can_lookup(nd
->path
.dentry
))
2497 nd
->path
.mnt
= NULL
;
2498 nd
->path
.dentry
= NULL
;
2504 int filename_lookup(int dfd
, struct filename
*name
, unsigned flags
,
2505 struct path
*path
, struct path
*root
)
2508 struct nameidata nd
;
2510 return PTR_ERR(name
);
2511 set_nameidata(&nd
, dfd
, name
, root
);
2512 retval
= path_lookupat(&nd
, flags
| LOOKUP_RCU
, path
);
2513 if (unlikely(retval
== -ECHILD
))
2514 retval
= path_lookupat(&nd
, flags
, path
);
2515 if (unlikely(retval
== -ESTALE
))
2516 retval
= path_lookupat(&nd
, flags
| LOOKUP_REVAL
, path
);
2518 if (likely(!retval
))
2519 audit_inode(name
, path
->dentry
,
2520 flags
& LOOKUP_MOUNTPOINT
? AUDIT_INODE_NOEVAL
: 0);
2521 restore_nameidata();
2525 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2526 static int path_parentat(struct nameidata
*nd
, unsigned flags
,
2527 struct path
*parent
)
2529 const char *s
= path_init(nd
, flags
);
2530 int err
= link_path_walk(s
, nd
);
2532 err
= complete_walk(nd
);
2535 nd
->path
.mnt
= NULL
;
2536 nd
->path
.dentry
= NULL
;
2542 /* Note: this does not consume "name" */
2543 static int __filename_parentat(int dfd
, struct filename
*name
,
2544 unsigned int flags
, struct path
*parent
,
2545 struct qstr
*last
, int *type
,
2546 const struct path
*root
)
2549 struct nameidata nd
;
2552 return PTR_ERR(name
);
2553 set_nameidata(&nd
, dfd
, name
, root
);
2554 retval
= path_parentat(&nd
, flags
| LOOKUP_RCU
, parent
);
2555 if (unlikely(retval
== -ECHILD
))
2556 retval
= path_parentat(&nd
, flags
, parent
);
2557 if (unlikely(retval
== -ESTALE
))
2558 retval
= path_parentat(&nd
, flags
| LOOKUP_REVAL
, parent
);
2559 if (likely(!retval
)) {
2561 *type
= nd
.last_type
;
2562 audit_inode(name
, parent
->dentry
, AUDIT_INODE_PARENT
);
2564 restore_nameidata();
2568 static int filename_parentat(int dfd
, struct filename
*name
,
2569 unsigned int flags
, struct path
*parent
,
2570 struct qstr
*last
, int *type
)
2572 return __filename_parentat(dfd
, name
, flags
, parent
, last
, type
, NULL
);
2575 /* does lookup, returns the object with parent locked */
2576 static struct dentry
*__kern_path_locked(struct filename
*name
, struct path
*path
)
2582 error
= filename_parentat(AT_FDCWD
, name
, 0, path
, &last
, &type
);
2584 return ERR_PTR(error
);
2585 if (unlikely(type
!= LAST_NORM
)) {
2587 return ERR_PTR(-EINVAL
);
2589 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
2590 d
= lookup_one_qstr_excl(&last
, path
->dentry
, 0);
2592 inode_unlock(path
->dentry
->d_inode
);
2598 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2600 struct filename
*filename
= getname_kernel(name
);
2601 struct dentry
*res
= __kern_path_locked(filename
, path
);
2607 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2609 struct filename
*filename
= getname_kernel(name
);
2610 int ret
= filename_lookup(AT_FDCWD
, filename
, flags
, path
, NULL
);
2616 EXPORT_SYMBOL(kern_path
);
2619 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2620 * @filename: filename structure
2621 * @flags: lookup flags
2622 * @parent: pointer to struct path to fill
2623 * @last: last component
2624 * @type: type of the last component
2625 * @root: pointer to struct path of the base directory
2627 int vfs_path_parent_lookup(struct filename
*filename
, unsigned int flags
,
2628 struct path
*parent
, struct qstr
*last
, int *type
,
2629 const struct path
*root
)
2631 return __filename_parentat(AT_FDCWD
, filename
, flags
, parent
, last
,
2634 EXPORT_SYMBOL(vfs_path_parent_lookup
);
2637 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2638 * @dentry: pointer to dentry of the base directory
2639 * @mnt: pointer to vfs mount of the base directory
2640 * @name: pointer to file name
2641 * @flags: lookup flags
2642 * @path: pointer to struct path to fill
2644 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2645 const char *name
, unsigned int flags
,
2648 struct filename
*filename
;
2649 struct path root
= {.mnt
= mnt
, .dentry
= dentry
};
2652 filename
= getname_kernel(name
);
2653 /* the first argument of filename_lookup() is ignored with root */
2654 ret
= filename_lookup(AT_FDCWD
, filename
, flags
, path
, &root
);
2658 EXPORT_SYMBOL(vfs_path_lookup
);
2660 static int lookup_one_common(struct mnt_idmap
*idmap
,
2661 const char *name
, struct dentry
*base
, int len
,
2666 this->hash
= full_name_hash(base
, name
, len
);
2670 if (unlikely(name
[0] == '.')) {
2671 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2676 unsigned int c
= *(const unsigned char *)name
++;
2677 if (c
== '/' || c
== '\0')
2681 * See if the low-level filesystem might want
2682 * to use its own hash..
2684 if (base
->d_flags
& DCACHE_OP_HASH
) {
2685 int err
= base
->d_op
->d_hash(base
, this);
2690 return inode_permission(idmap
, base
->d_inode
, MAY_EXEC
);
2694 * try_lookup_one_len - filesystem helper to lookup single pathname component
2695 * @name: pathname component to lookup
2696 * @base: base directory to lookup from
2697 * @len: maximum length @len should be interpreted to
2699 * Look up a dentry by name in the dcache, returning NULL if it does not
2700 * currently exist. The function does not try to create a dentry.
2702 * Note that this routine is purely a helper for filesystem usage and should
2703 * not be called by generic code.
2705 * The caller must hold base->i_mutex.
2707 struct dentry
*try_lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2712 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2714 err
= lookup_one_common(&nop_mnt_idmap
, name
, base
, len
, &this);
2716 return ERR_PTR(err
);
2718 return lookup_dcache(&this, base
, 0);
2720 EXPORT_SYMBOL(try_lookup_one_len
);
2723 * lookup_one_len - filesystem helper to lookup single pathname component
2724 * @name: pathname component to lookup
2725 * @base: base directory to lookup from
2726 * @len: maximum length @len should be interpreted to
2728 * Note that this routine is purely a helper for filesystem usage and should
2729 * not be called by generic code.
2731 * The caller must hold base->i_mutex.
2733 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2735 struct dentry
*dentry
;
2739 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2741 err
= lookup_one_common(&nop_mnt_idmap
, name
, base
, len
, &this);
2743 return ERR_PTR(err
);
2745 dentry
= lookup_dcache(&this, base
, 0);
2746 return dentry
? dentry
: __lookup_slow(&this, base
, 0);
2748 EXPORT_SYMBOL(lookup_one_len
);
2751 * lookup_one - filesystem helper to lookup single pathname component
2752 * @idmap: idmap of the mount the lookup is performed from
2753 * @name: pathname component to lookup
2754 * @base: base directory to lookup from
2755 * @len: maximum length @len should be interpreted to
2757 * Note that this routine is purely a helper for filesystem usage and should
2758 * not be called by generic code.
2760 * The caller must hold base->i_mutex.
2762 struct dentry
*lookup_one(struct mnt_idmap
*idmap
, const char *name
,
2763 struct dentry
*base
, int len
)
2765 struct dentry
*dentry
;
2769 WARN_ON_ONCE(!inode_is_locked(base
->d_inode
));
2771 err
= lookup_one_common(idmap
, name
, base
, len
, &this);
2773 return ERR_PTR(err
);
2775 dentry
= lookup_dcache(&this, base
, 0);
2776 return dentry
? dentry
: __lookup_slow(&this, base
, 0);
2778 EXPORT_SYMBOL(lookup_one
);
2781 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2782 * @idmap: idmap of the mount the lookup is performed from
2783 * @name: pathname component to lookup
2784 * @base: base directory to lookup from
2785 * @len: maximum length @len should be interpreted to
2787 * Note that this routine is purely a helper for filesystem usage and should
2788 * not be called by generic code.
2790 * Unlike lookup_one_len, it should be called without the parent
2791 * i_mutex held, and will take the i_mutex itself if necessary.
2793 struct dentry
*lookup_one_unlocked(struct mnt_idmap
*idmap
,
2794 const char *name
, struct dentry
*base
,
2801 err
= lookup_one_common(idmap
, name
, base
, len
, &this);
2803 return ERR_PTR(err
);
2805 ret
= lookup_dcache(&this, base
, 0);
2807 ret
= lookup_slow(&this, base
, 0);
2810 EXPORT_SYMBOL(lookup_one_unlocked
);
2813 * lookup_one_positive_unlocked - filesystem helper to lookup single
2814 * pathname component
2815 * @idmap: idmap of the mount the lookup is performed from
2816 * @name: pathname component to lookup
2817 * @base: base directory to lookup from
2818 * @len: maximum length @len should be interpreted to
2820 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2821 * known positive or ERR_PTR(). This is what most of the users want.
2823 * Note that pinned negative with unlocked parent _can_ become positive at any
2824 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2825 * positives have >d_inode stable, so this one avoids such problems.
2827 * Note that this routine is purely a helper for filesystem usage and should
2828 * not be called by generic code.
2830 * The helper should be called without i_mutex held.
2832 struct dentry
*lookup_one_positive_unlocked(struct mnt_idmap
*idmap
,
2834 struct dentry
*base
, int len
)
2836 struct dentry
*ret
= lookup_one_unlocked(idmap
, name
, base
, len
);
2838 if (!IS_ERR(ret
) && d_flags_negative(smp_load_acquire(&ret
->d_flags
))) {
2840 ret
= ERR_PTR(-ENOENT
);
2844 EXPORT_SYMBOL(lookup_one_positive_unlocked
);
2847 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2848 * @name: pathname component to lookup
2849 * @base: base directory to lookup from
2850 * @len: maximum length @len should be interpreted to
2852 * Note that this routine is purely a helper for filesystem usage and should
2853 * not be called by generic code.
2855 * Unlike lookup_one_len, it should be called without the parent
2856 * i_mutex held, and will take the i_mutex itself if necessary.
2858 struct dentry
*lookup_one_len_unlocked(const char *name
,
2859 struct dentry
*base
, int len
)
2861 return lookup_one_unlocked(&nop_mnt_idmap
, name
, base
, len
);
2863 EXPORT_SYMBOL(lookup_one_len_unlocked
);
2866 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2867 * on negatives. Returns known positive or ERR_PTR(); that's what
2868 * most of the users want. Note that pinned negative with unlocked parent
2869 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2870 * need to be very careful; pinned positives have ->d_inode stable, so
2871 * this one avoids such problems.
2873 struct dentry
*lookup_positive_unlocked(const char *name
,
2874 struct dentry
*base
, int len
)
2876 return lookup_one_positive_unlocked(&nop_mnt_idmap
, name
, base
, len
);
2878 EXPORT_SYMBOL(lookup_positive_unlocked
);
2880 #ifdef CONFIG_UNIX98_PTYS
2881 int path_pts(struct path
*path
)
2883 /* Find something mounted on "pts" in the same directory as
2886 struct dentry
*parent
= dget_parent(path
->dentry
);
2887 struct dentry
*child
;
2888 struct qstr
this = QSTR_INIT("pts", 3);
2890 if (unlikely(!path_connected(path
->mnt
, parent
))) {
2895 path
->dentry
= parent
;
2896 child
= d_hash_and_lookup(parent
, &this);
2897 if (IS_ERR_OR_NULL(child
))
2900 path
->dentry
= child
;
2902 follow_down(path
, 0);
2907 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2908 struct path
*path
, int *empty
)
2910 struct filename
*filename
= getname_flags(name
, flags
, empty
);
2911 int ret
= filename_lookup(dfd
, filename
, flags
, path
, NULL
);
2916 EXPORT_SYMBOL(user_path_at_empty
);
2918 int __check_sticky(struct mnt_idmap
*idmap
, struct inode
*dir
,
2919 struct inode
*inode
)
2921 kuid_t fsuid
= current_fsuid();
2923 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap
, inode
), fsuid
))
2925 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap
, dir
), fsuid
))
2927 return !capable_wrt_inode_uidgid(idmap
, inode
, CAP_FOWNER
);
2929 EXPORT_SYMBOL(__check_sticky
);
2932 * Check whether we can remove a link victim from directory dir, check
2933 * whether the type of victim is right.
2934 * 1. We can't do it if dir is read-only (done in permission())
2935 * 2. We should have write and exec permissions on dir
2936 * 3. We can't remove anything from append-only dir
2937 * 4. We can't do anything with immutable dir (done in permission())
2938 * 5. If the sticky bit on dir is set we should either
2939 * a. be owner of dir, or
2940 * b. be owner of victim, or
2941 * c. have CAP_FOWNER capability
2942 * 6. If the victim is append-only or immutable we can't do antyhing with
2943 * links pointing to it.
2944 * 7. If the victim has an unknown uid or gid we can't change the inode.
2945 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2946 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2947 * 10. We can't remove a root or mountpoint.
2948 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2949 * nfs_async_unlink().
2951 static int may_delete(struct mnt_idmap
*idmap
, struct inode
*dir
,
2952 struct dentry
*victim
, bool isdir
)
2954 struct inode
*inode
= d_backing_inode(victim
);
2957 if (d_is_negative(victim
))
2961 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2963 /* Inode writeback is not safe when the uid or gid are invalid. */
2964 if (!vfsuid_valid(i_uid_into_vfsuid(idmap
, inode
)) ||
2965 !vfsgid_valid(i_gid_into_vfsgid(idmap
, inode
)))
2968 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2970 error
= inode_permission(idmap
, dir
, MAY_WRITE
| MAY_EXEC
);
2976 if (check_sticky(idmap
, dir
, inode
) || IS_APPEND(inode
) ||
2977 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
) ||
2978 HAS_UNMAPPED_ID(idmap
, inode
))
2981 if (!d_is_dir(victim
))
2983 if (IS_ROOT(victim
))
2985 } else if (d_is_dir(victim
))
2987 if (IS_DEADDIR(dir
))
2989 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2994 /* Check whether we can create an object with dentry child in directory
2996 * 1. We can't do it if child already exists (open has special treatment for
2997 * this case, but since we are inlined it's OK)
2998 * 2. We can't do it if dir is read-only (done in permission())
2999 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3000 * 4. We should have write and exec permissions on dir
3001 * 5. We can't do it if dir is immutable (done in permission())
3003 static inline int may_create(struct mnt_idmap
*idmap
,
3004 struct inode
*dir
, struct dentry
*child
)
3006 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
3009 if (IS_DEADDIR(dir
))
3011 if (!fsuidgid_has_mapping(dir
->i_sb
, idmap
))
3014 return inode_permission(idmap
, dir
, MAY_WRITE
| MAY_EXEC
);
3017 static struct dentry
*lock_two_directories(struct dentry
*p1
, struct dentry
*p2
)
3021 p
= d_ancestor(p2
, p1
);
3023 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
3024 inode_lock_nested(p1
->d_inode
, I_MUTEX_CHILD
);
3028 p
= d_ancestor(p1
, p2
);
3030 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
3031 inode_lock_nested(p2
->d_inode
, I_MUTEX_CHILD
);
3035 lock_two_inodes(p1
->d_inode
, p2
->d_inode
,
3036 I_MUTEX_PARENT
, I_MUTEX_PARENT2
);
3041 * p1 and p2 should be directories on the same fs.
3043 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
3046 inode_lock_nested(p1
->d_inode
, I_MUTEX_PARENT
);
3050 mutex_lock(&p1
->d_sb
->s_vfs_rename_mutex
);
3051 return lock_two_directories(p1
, p2
);
3053 EXPORT_SYMBOL(lock_rename
);
3056 * c1 and p2 should be on the same fs.
3058 struct dentry
*lock_rename_child(struct dentry
*c1
, struct dentry
*p2
)
3060 if (READ_ONCE(c1
->d_parent
) == p2
) {
3062 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3064 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
3066 * now that p2 is locked, nobody can move in or out of it,
3067 * so the test below is safe.
3069 if (likely(c1
->d_parent
== p2
))
3073 * c1 got moved out of p2 while we'd been taking locks;
3074 * unlock and fall back to slow case.
3076 inode_unlock(p2
->d_inode
);
3079 mutex_lock(&c1
->d_sb
->s_vfs_rename_mutex
);
3081 * nobody can move out of any directories on this fs.
3083 if (likely(c1
->d_parent
!= p2
))
3084 return lock_two_directories(c1
->d_parent
, p2
);
3087 * c1 got moved into p2 while we were taking locks;
3088 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3089 * for consistency with lock_rename().
3091 inode_lock_nested(p2
->d_inode
, I_MUTEX_PARENT
);
3092 mutex_unlock(&c1
->d_sb
->s_vfs_rename_mutex
);
3095 EXPORT_SYMBOL(lock_rename_child
);
3097 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
3099 inode_unlock(p1
->d_inode
);
3101 inode_unlock(p2
->d_inode
);
3102 mutex_unlock(&p1
->d_sb
->s_vfs_rename_mutex
);
3105 EXPORT_SYMBOL(unlock_rename
);
3108 * mode_strip_umask - handle vfs umask stripping
3109 * @dir: parent directory of the new inode
3110 * @mode: mode of the new inode to be created in @dir
3112 * Umask stripping depends on whether or not the filesystem supports POSIX
3113 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3114 * in here. If the filesystem does support POSIX ACLs umask stripping is
3115 * deferred until the filesystem calls posix_acl_create().
3119 static inline umode_t
mode_strip_umask(const struct inode
*dir
, umode_t mode
)
3121 if (!IS_POSIXACL(dir
))
3122 mode
&= ~current_umask();
3127 * vfs_prepare_mode - prepare the mode to be used for a new inode
3128 * @idmap: idmap of the mount the inode was found from
3129 * @dir: parent directory of the new inode
3130 * @mode: mode of the new inode
3131 * @mask_perms: allowed permission by the vfs
3132 * @type: type of file to be created
3134 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3135 * object to be created.
3137 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3138 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3139 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3140 * POSIX ACL supporting filesystems.
3142 * Note that it's currently valid for @type to be 0 if a directory is created.
3143 * Filesystems raise that flag individually and we need to check whether each
3144 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3147 * Returns: mode to be passed to the filesystem
3149 static inline umode_t
vfs_prepare_mode(struct mnt_idmap
*idmap
,
3150 const struct inode
*dir
, umode_t mode
,
3151 umode_t mask_perms
, umode_t type
)
3153 mode
= mode_strip_sgid(idmap
, dir
, mode
);
3154 mode
= mode_strip_umask(dir
, mode
);
3157 * Apply the vfs mandated allowed permission mask and set the type of
3158 * file to be created before we call into the filesystem.
3160 mode
&= (mask_perms
& ~S_IFMT
);
3161 mode
|= (type
& S_IFMT
);
3167 * vfs_create - create new file
3168 * @idmap: idmap of the mount the inode was found from
3169 * @dir: inode of @dentry
3170 * @dentry: pointer to dentry of the base directory
3171 * @mode: mode of the new file
3172 * @want_excl: whether the file must not yet exist
3174 * Create a new file.
3176 * If the inode has been found through an idmapped mount the idmap of
3177 * the vfsmount must be passed through @idmap. This function will then take
3178 * care to map the inode according to @idmap before checking permissions.
3179 * On non-idmapped mounts or if permission checking is to be performed on the
3180 * raw inode simply passs @nop_mnt_idmap.
3182 int vfs_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
3183 struct dentry
*dentry
, umode_t mode
, bool want_excl
)
3187 error
= may_create(idmap
, dir
, dentry
);
3191 if (!dir
->i_op
->create
)
3192 return -EACCES
; /* shouldn't it be ENOSYS? */
3194 mode
= vfs_prepare_mode(idmap
, dir
, mode
, S_IALLUGO
, S_IFREG
);
3195 error
= security_inode_create(dir
, dentry
, mode
);
3198 error
= dir
->i_op
->create(idmap
, dir
, dentry
, mode
, want_excl
);
3200 fsnotify_create(dir
, dentry
);
3203 EXPORT_SYMBOL(vfs_create
);
3205 int vfs_mkobj(struct dentry
*dentry
, umode_t mode
,
3206 int (*f
)(struct dentry
*, umode_t
, void *),
3209 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3210 int error
= may_create(&nop_mnt_idmap
, dir
, dentry
);
3216 error
= security_inode_create(dir
, dentry
, mode
);
3219 error
= f(dentry
, mode
, arg
);
3221 fsnotify_create(dir
, dentry
);
3224 EXPORT_SYMBOL(vfs_mkobj
);
3226 bool may_open_dev(const struct path
*path
)
3228 return !(path
->mnt
->mnt_flags
& MNT_NODEV
) &&
3229 !(path
->mnt
->mnt_sb
->s_iflags
& SB_I_NODEV
);
3232 static int may_open(struct mnt_idmap
*idmap
, const struct path
*path
,
3233 int acc_mode
, int flag
)
3235 struct dentry
*dentry
= path
->dentry
;
3236 struct inode
*inode
= dentry
->d_inode
;
3242 switch (inode
->i_mode
& S_IFMT
) {
3246 if (acc_mode
& MAY_WRITE
)
3248 if (acc_mode
& MAY_EXEC
)
3253 if (!may_open_dev(path
))
3258 if (acc_mode
& MAY_EXEC
)
3263 if ((acc_mode
& MAY_EXEC
) && path_noexec(path
))
3268 error
= inode_permission(idmap
, inode
, MAY_OPEN
| acc_mode
);
3273 * An append-only file must be opened in append mode for writing.
3275 if (IS_APPEND(inode
)) {
3276 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
3282 /* O_NOATIME can only be set by the owner or superuser */
3283 if (flag
& O_NOATIME
&& !inode_owner_or_capable(idmap
, inode
))
3289 static int handle_truncate(struct mnt_idmap
*idmap
, struct file
*filp
)
3291 const struct path
*path
= &filp
->f_path
;
3292 struct inode
*inode
= path
->dentry
->d_inode
;
3293 int error
= get_write_access(inode
);
3297 error
= security_file_truncate(filp
);
3299 error
= do_truncate(idmap
, path
->dentry
, 0,
3300 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
3303 put_write_access(inode
);
3307 static inline int open_to_namei_flags(int flag
)
3309 if ((flag
& O_ACCMODE
) == 3)
3314 static int may_o_create(struct mnt_idmap
*idmap
,
3315 const struct path
*dir
, struct dentry
*dentry
,
3318 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
3322 if (!fsuidgid_has_mapping(dir
->dentry
->d_sb
, idmap
))
3325 error
= inode_permission(idmap
, dir
->dentry
->d_inode
,
3326 MAY_WRITE
| MAY_EXEC
);
3330 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
3334 * Attempt to atomically look up, create and open a file from a negative
3337 * Returns 0 if successful. The file will have been created and attached to
3338 * @file by the filesystem calling finish_open().
3340 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3341 * be set. The caller will need to perform the open themselves. @path will
3342 * have been updated to point to the new dentry. This may be negative.
3344 * Returns an error code otherwise.
3346 static struct dentry
*atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
3348 int open_flag
, umode_t mode
)
3350 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
3351 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
3354 if (nd
->flags
& LOOKUP_DIRECTORY
)
3355 open_flag
|= O_DIRECTORY
;
3357 file
->f_path
.dentry
= DENTRY_NOT_SET
;
3358 file
->f_path
.mnt
= nd
->path
.mnt
;
3359 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
,
3360 open_to_namei_flags(open_flag
), mode
);
3361 d_lookup_done(dentry
);
3363 if (file
->f_mode
& FMODE_OPENED
) {
3364 if (unlikely(dentry
!= file
->f_path
.dentry
)) {
3366 dentry
= dget(file
->f_path
.dentry
);
3368 } else if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
3371 if (file
->f_path
.dentry
) {
3373 dentry
= file
->f_path
.dentry
;
3375 if (unlikely(d_is_negative(dentry
)))
3381 dentry
= ERR_PTR(error
);
3387 * Look up and maybe create and open the last component.
3389 * Must be called with parent locked (exclusive in O_CREAT case).
3391 * Returns 0 on success, that is, if
3392 * the file was successfully atomically created (if necessary) and opened, or
3393 * the file was not completely opened at this time, though lookups and
3394 * creations were performed.
3395 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3396 * In the latter case dentry returned in @path might be negative if O_CREAT
3397 * hadn't been specified.
3399 * An error code is returned on failure.
3401 static struct dentry
*lookup_open(struct nameidata
*nd
, struct file
*file
,
3402 const struct open_flags
*op
,
3405 struct mnt_idmap
*idmap
;
3406 struct dentry
*dir
= nd
->path
.dentry
;
3407 struct inode
*dir_inode
= dir
->d_inode
;
3408 int open_flag
= op
->open_flag
;
3409 struct dentry
*dentry
;
3410 int error
, create_error
= 0;
3411 umode_t mode
= op
->mode
;
3412 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq
);
3414 if (unlikely(IS_DEADDIR(dir_inode
)))
3415 return ERR_PTR(-ENOENT
);
3417 file
->f_mode
&= ~FMODE_CREATED
;
3418 dentry
= d_lookup(dir
, &nd
->last
);
3421 dentry
= d_alloc_parallel(dir
, &nd
->last
, &wq
);
3425 if (d_in_lookup(dentry
))
3428 error
= d_revalidate(dentry
, nd
->flags
);
3429 if (likely(error
> 0))
3433 d_invalidate(dentry
);
3437 if (dentry
->d_inode
) {
3438 /* Cached positive dentry: will open in f_op->open */
3443 * Checking write permission is tricky, bacuse we don't know if we are
3444 * going to actually need it: O_CREAT opens should work as long as the
3445 * file exists. But checking existence breaks atomicity. The trick is
3446 * to check access and if not granted clear O_CREAT from the flags.
3448 * Another problem is returing the "right" error value (e.g. for an
3449 * O_EXCL open we want to return EEXIST not EROFS).
3451 if (unlikely(!got_write
))
3452 open_flag
&= ~O_TRUNC
;
3453 idmap
= mnt_idmap(nd
->path
.mnt
);
3454 if (open_flag
& O_CREAT
) {
3455 if (open_flag
& O_EXCL
)
3456 open_flag
&= ~O_TRUNC
;
3457 mode
= vfs_prepare_mode(idmap
, dir
->d_inode
, mode
, mode
, mode
);
3458 if (likely(got_write
))
3459 create_error
= may_o_create(idmap
, &nd
->path
,
3462 create_error
= -EROFS
;
3465 open_flag
&= ~O_CREAT
;
3466 if (dir_inode
->i_op
->atomic_open
) {
3467 dentry
= atomic_open(nd
, dentry
, file
, open_flag
, mode
);
3468 if (unlikely(create_error
) && dentry
== ERR_PTR(-ENOENT
))
3469 dentry
= ERR_PTR(create_error
);
3473 if (d_in_lookup(dentry
)) {
3474 struct dentry
*res
= dir_inode
->i_op
->lookup(dir_inode
, dentry
,
3476 d_lookup_done(dentry
);
3477 if (unlikely(res
)) {
3479 error
= PTR_ERR(res
);
3487 /* Negative dentry, just create the file */
3488 if (!dentry
->d_inode
&& (open_flag
& O_CREAT
)) {
3489 file
->f_mode
|= FMODE_CREATED
;
3490 audit_inode_child(dir_inode
, dentry
, AUDIT_TYPE_CHILD_CREATE
);
3491 if (!dir_inode
->i_op
->create
) {
3496 error
= dir_inode
->i_op
->create(idmap
, dir_inode
, dentry
,
3497 mode
, open_flag
& O_EXCL
);
3501 if (unlikely(create_error
) && !dentry
->d_inode
) {
3502 error
= create_error
;
3509 return ERR_PTR(error
);
3512 static const char *open_last_lookups(struct nameidata
*nd
,
3513 struct file
*file
, const struct open_flags
*op
)
3515 struct dentry
*dir
= nd
->path
.dentry
;
3516 int open_flag
= op
->open_flag
;
3517 bool got_write
= false;
3518 struct dentry
*dentry
;
3521 nd
->flags
|= op
->intent
;
3523 if (nd
->last_type
!= LAST_NORM
) {
3526 return handle_dots(nd
, nd
->last_type
);
3529 if (!(open_flag
& O_CREAT
)) {
3530 if (nd
->last
.name
[nd
->last
.len
])
3531 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
3532 /* we _can_ be in RCU mode here */
3533 dentry
= lookup_fast(nd
);
3535 return ERR_CAST(dentry
);
3539 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3541 /* create side of things */
3542 if (nd
->flags
& LOOKUP_RCU
) {
3543 if (!try_to_unlazy(nd
))
3544 return ERR_PTR(-ECHILD
);
3546 audit_inode(nd
->name
, dir
, AUDIT_INODE_PARENT
);
3547 /* trailing slashes? */
3548 if (unlikely(nd
->last
.name
[nd
->last
.len
]))
3549 return ERR_PTR(-EISDIR
);
3552 if (open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
3553 got_write
= !mnt_want_write(nd
->path
.mnt
);
3555 * do _not_ fail yet - we might not need that or fail with
3556 * a different error; let lookup_open() decide; we'll be
3557 * dropping this one anyway.
3560 if (open_flag
& O_CREAT
)
3561 inode_lock(dir
->d_inode
);
3563 inode_lock_shared(dir
->d_inode
);
3564 dentry
= lookup_open(nd
, file
, op
, got_write
);
3565 if (!IS_ERR(dentry
) && (file
->f_mode
& FMODE_CREATED
))
3566 fsnotify_create(dir
->d_inode
, dentry
);
3567 if (open_flag
& O_CREAT
)
3568 inode_unlock(dir
->d_inode
);
3570 inode_unlock_shared(dir
->d_inode
);
3573 mnt_drop_write(nd
->path
.mnt
);
3576 return ERR_CAST(dentry
);
3578 if (file
->f_mode
& (FMODE_OPENED
| FMODE_CREATED
)) {
3579 dput(nd
->path
.dentry
);
3580 nd
->path
.dentry
= dentry
;
3587 res
= step_into(nd
, WALK_TRAILING
, dentry
);
3589 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3594 * Handle the last step of open()
3596 static int do_open(struct nameidata
*nd
,
3597 struct file
*file
, const struct open_flags
*op
)
3599 struct mnt_idmap
*idmap
;
3600 int open_flag
= op
->open_flag
;
3605 if (!(file
->f_mode
& (FMODE_OPENED
| FMODE_CREATED
))) {
3606 error
= complete_walk(nd
);
3610 if (!(file
->f_mode
& FMODE_CREATED
))
3611 audit_inode(nd
->name
, nd
->path
.dentry
, 0);
3612 idmap
= mnt_idmap(nd
->path
.mnt
);
3613 if (open_flag
& O_CREAT
) {
3614 if ((open_flag
& O_EXCL
) && !(file
->f_mode
& FMODE_CREATED
))
3616 if (d_is_dir(nd
->path
.dentry
))
3618 error
= may_create_in_sticky(idmap
, nd
,
3619 d_backing_inode(nd
->path
.dentry
));
3620 if (unlikely(error
))
3623 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3626 do_truncate
= false;
3627 acc_mode
= op
->acc_mode
;
3628 if (file
->f_mode
& FMODE_CREATED
) {
3629 /* Don't check for write permission, don't truncate */
3630 open_flag
&= ~O_TRUNC
;
3632 } else if (d_is_reg(nd
->path
.dentry
) && open_flag
& O_TRUNC
) {
3633 error
= mnt_want_write(nd
->path
.mnt
);
3638 error
= may_open(idmap
, &nd
->path
, acc_mode
, open_flag
);
3639 if (!error
&& !(file
->f_mode
& FMODE_OPENED
))
3640 error
= vfs_open(&nd
->path
, file
);
3642 error
= ima_file_check(file
, op
->acc_mode
);
3643 if (!error
&& do_truncate
)
3644 error
= handle_truncate(idmap
, file
);
3645 if (unlikely(error
> 0)) {
3650 mnt_drop_write(nd
->path
.mnt
);
3655 * vfs_tmpfile - create tmpfile
3656 * @idmap: idmap of the mount the inode was found from
3657 * @parentpath: pointer to the path of the base directory
3658 * @file: file descriptor of the new tmpfile
3659 * @mode: mode of the new tmpfile
3661 * Create a temporary file.
3663 * If the inode has been found through an idmapped mount the idmap of
3664 * the vfsmount must be passed through @idmap. This function will then take
3665 * care to map the inode according to @idmap before checking permissions.
3666 * On non-idmapped mounts or if permission checking is to be performed on the
3667 * raw inode simply passs @nop_mnt_idmap.
3669 static int vfs_tmpfile(struct mnt_idmap
*idmap
,
3670 const struct path
*parentpath
,
3671 struct file
*file
, umode_t mode
)
3673 struct dentry
*child
;
3674 struct inode
*dir
= d_inode(parentpath
->dentry
);
3675 struct inode
*inode
;
3677 int open_flag
= file
->f_flags
;
3679 /* we want directory to be writable */
3680 error
= inode_permission(idmap
, dir
, MAY_WRITE
| MAY_EXEC
);
3683 if (!dir
->i_op
->tmpfile
)
3685 child
= d_alloc(parentpath
->dentry
, &slash_name
);
3686 if (unlikely(!child
))
3688 file
->f_path
.mnt
= parentpath
->mnt
;
3689 file
->f_path
.dentry
= child
;
3690 mode
= vfs_prepare_mode(idmap
, dir
, mode
, mode
, mode
);
3691 error
= dir
->i_op
->tmpfile(idmap
, dir
, file
, mode
);
3695 /* Don't check for other permissions, the inode was just created */
3696 error
= may_open(idmap
, &file
->f_path
, 0, file
->f_flags
);
3699 inode
= file_inode(file
);
3700 if (!(open_flag
& O_EXCL
)) {
3701 spin_lock(&inode
->i_lock
);
3702 inode
->i_state
|= I_LINKABLE
;
3703 spin_unlock(&inode
->i_lock
);
3705 ima_post_create_tmpfile(idmap
, inode
);
3710 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3711 * @idmap: idmap of the mount the inode was found from
3712 * @parentpath: path of the base directory
3713 * @mode: mode of the new tmpfile
3715 * @cred: credentials for open
3717 * Create and open a temporary file. The file is not accounted in nr_files,
3718 * hence this is only for kernel internal use, and must not be installed into
3719 * file tables or such.
3721 struct file
*kernel_tmpfile_open(struct mnt_idmap
*idmap
,
3722 const struct path
*parentpath
,
3723 umode_t mode
, int open_flag
,
3724 const struct cred
*cred
)
3729 file
= alloc_empty_file_noaccount(open_flag
, cred
);
3733 error
= vfs_tmpfile(idmap
, parentpath
, file
, mode
);
3736 file
= ERR_PTR(error
);
3740 EXPORT_SYMBOL(kernel_tmpfile_open
);
3742 static int do_tmpfile(struct nameidata
*nd
, unsigned flags
,
3743 const struct open_flags
*op
,
3747 int error
= path_lookupat(nd
, flags
| LOOKUP_DIRECTORY
, &path
);
3749 if (unlikely(error
))
3751 error
= mnt_want_write(path
.mnt
);
3752 if (unlikely(error
))
3754 error
= vfs_tmpfile(mnt_idmap(path
.mnt
), &path
, file
, op
->mode
);
3757 audit_inode(nd
->name
, file
->f_path
.dentry
, 0);
3759 mnt_drop_write(path
.mnt
);
3765 static int do_o_path(struct nameidata
*nd
, unsigned flags
, struct file
*file
)
3768 int error
= path_lookupat(nd
, flags
, &path
);
3770 audit_inode(nd
->name
, path
.dentry
, 0);
3771 error
= vfs_open(&path
, file
);
3777 static struct file
*path_openat(struct nameidata
*nd
,
3778 const struct open_flags
*op
, unsigned flags
)
3783 file
= alloc_empty_file(op
->open_flag
, current_cred());
3787 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3788 error
= do_tmpfile(nd
, flags
, op
, file
);
3789 } else if (unlikely(file
->f_flags
& O_PATH
)) {
3790 error
= do_o_path(nd
, flags
, file
);
3792 const char *s
= path_init(nd
, flags
);
3793 while (!(error
= link_path_walk(s
, nd
)) &&
3794 (s
= open_last_lookups(nd
, file
, op
)) != NULL
)
3797 error
= do_open(nd
, file
, op
);
3800 if (likely(!error
)) {
3801 if (likely(file
->f_mode
& FMODE_OPENED
))
3807 if (error
== -EOPENSTALE
) {
3808 if (flags
& LOOKUP_RCU
)
3813 return ERR_PTR(error
);
3816 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3817 const struct open_flags
*op
)
3819 struct nameidata nd
;
3820 int flags
= op
->lookup_flags
;
3823 set_nameidata(&nd
, dfd
, pathname
, NULL
);
3824 filp
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3825 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3826 filp
= path_openat(&nd
, op
, flags
);
3827 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3828 filp
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3829 restore_nameidata();
3833 struct file
*do_file_open_root(const struct path
*root
,
3834 const char *name
, const struct open_flags
*op
)
3836 struct nameidata nd
;
3838 struct filename
*filename
;
3839 int flags
= op
->lookup_flags
;
3841 if (d_is_symlink(root
->dentry
) && op
->intent
& LOOKUP_OPEN
)
3842 return ERR_PTR(-ELOOP
);
3844 filename
= getname_kernel(name
);
3845 if (IS_ERR(filename
))
3846 return ERR_CAST(filename
);
3848 set_nameidata(&nd
, -1, filename
, root
);
3849 file
= path_openat(&nd
, op
, flags
| LOOKUP_RCU
);
3850 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3851 file
= path_openat(&nd
, op
, flags
);
3852 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3853 file
= path_openat(&nd
, op
, flags
| LOOKUP_REVAL
);
3854 restore_nameidata();
3859 static struct dentry
*filename_create(int dfd
, struct filename
*name
,
3860 struct path
*path
, unsigned int lookup_flags
)
3862 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3864 bool want_dir
= lookup_flags
& LOOKUP_DIRECTORY
;
3865 unsigned int reval_flag
= lookup_flags
& LOOKUP_REVAL
;
3866 unsigned int create_flags
= LOOKUP_CREATE
| LOOKUP_EXCL
;
3871 error
= filename_parentat(dfd
, name
, reval_flag
, path
, &last
, &type
);
3873 return ERR_PTR(error
);
3876 * Yucky last component or no last component at all?
3877 * (foo/., foo/.., /////)
3879 if (unlikely(type
!= LAST_NORM
))
3882 /* don't fail immediately if it's r/o, at least try to report other errors */
3883 err2
= mnt_want_write(path
->mnt
);
3885 * Do the final lookup. Suppress 'create' if there is a trailing
3886 * '/', and a directory wasn't requested.
3888 if (last
.name
[last
.len
] && !want_dir
)
3890 inode_lock_nested(path
->dentry
->d_inode
, I_MUTEX_PARENT
);
3891 dentry
= lookup_one_qstr_excl(&last
, path
->dentry
,
3892 reval_flag
| create_flags
);
3897 if (d_is_positive(dentry
))
3901 * Special case - lookup gave negative, but... we had foo/bar/
3902 * From the vfs_mknod() POV we just have a negative dentry -
3903 * all is fine. Let's be bastards - you had / on the end, you've
3904 * been asking for (non-existent) directory. -ENOENT for you.
3906 if (unlikely(!create_flags
)) {
3910 if (unlikely(err2
)) {
3917 dentry
= ERR_PTR(error
);
3919 inode_unlock(path
->dentry
->d_inode
);
3921 mnt_drop_write(path
->mnt
);
3927 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3928 struct path
*path
, unsigned int lookup_flags
)
3930 struct filename
*filename
= getname_kernel(pathname
);
3931 struct dentry
*res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3936 EXPORT_SYMBOL(kern_path_create
);
3938 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3941 inode_unlock(path
->dentry
->d_inode
);
3942 mnt_drop_write(path
->mnt
);
3945 EXPORT_SYMBOL(done_path_create
);
3947 inline struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3948 struct path
*path
, unsigned int lookup_flags
)
3950 struct filename
*filename
= getname(pathname
);
3951 struct dentry
*res
= filename_create(dfd
, filename
, path
, lookup_flags
);
3956 EXPORT_SYMBOL(user_path_create
);
3959 * vfs_mknod - create device node or file
3960 * @idmap: idmap of the mount the inode was found from
3961 * @dir: inode of @dentry
3962 * @dentry: pointer to dentry of the base directory
3963 * @mode: mode of the new device node or file
3964 * @dev: device number of device to create
3966 * Create a device node or file.
3968 * If the inode has been found through an idmapped mount the idmap of
3969 * the vfsmount must be passed through @idmap. This function will then take
3970 * care to map the inode according to @idmap before checking permissions.
3971 * On non-idmapped mounts or if permission checking is to be performed on the
3972 * raw inode simply passs @nop_mnt_idmap.
3974 int vfs_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
3975 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3977 bool is_whiteout
= S_ISCHR(mode
) && dev
== WHITEOUT_DEV
;
3978 int error
= may_create(idmap
, dir
, dentry
);
3983 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !is_whiteout
&&
3984 !capable(CAP_MKNOD
))
3987 if (!dir
->i_op
->mknod
)
3990 mode
= vfs_prepare_mode(idmap
, dir
, mode
, mode
, mode
);
3991 error
= devcgroup_inode_mknod(mode
, dev
);
3995 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3999 error
= dir
->i_op
->mknod(idmap
, dir
, dentry
, mode
, dev
);
4001 fsnotify_create(dir
, dentry
);
4004 EXPORT_SYMBOL(vfs_mknod
);
4006 static int may_mknod(umode_t mode
)
4008 switch (mode
& S_IFMT
) {
4014 case 0: /* zero mode translates to S_IFREG */
4023 static int do_mknodat(int dfd
, struct filename
*name
, umode_t mode
,
4026 struct mnt_idmap
*idmap
;
4027 struct dentry
*dentry
;
4030 unsigned int lookup_flags
= 0;
4032 error
= may_mknod(mode
);
4036 dentry
= filename_create(dfd
, name
, &path
, lookup_flags
);
4037 error
= PTR_ERR(dentry
);
4041 error
= security_path_mknod(&path
, dentry
,
4042 mode_strip_umask(path
.dentry
->d_inode
, mode
), dev
);
4046 idmap
= mnt_idmap(path
.mnt
);
4047 switch (mode
& S_IFMT
) {
4048 case 0: case S_IFREG
:
4049 error
= vfs_create(idmap
, path
.dentry
->d_inode
,
4050 dentry
, mode
, true);
4052 ima_post_path_mknod(idmap
, dentry
);
4054 case S_IFCHR
: case S_IFBLK
:
4055 error
= vfs_mknod(idmap
, path
.dentry
->d_inode
,
4056 dentry
, mode
, new_decode_dev(dev
));
4058 case S_IFIFO
: case S_IFSOCK
:
4059 error
= vfs_mknod(idmap
, path
.dentry
->d_inode
,
4064 done_path_create(&path
, dentry
);
4065 if (retry_estale(error
, lookup_flags
)) {
4066 lookup_flags
|= LOOKUP_REVAL
;
4074 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
4077 return do_mknodat(dfd
, getname(filename
), mode
, dev
);
4080 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
4082 return do_mknodat(AT_FDCWD
, getname(filename
), mode
, dev
);
4086 * vfs_mkdir - create directory
4087 * @idmap: idmap of the mount the inode was found from
4088 * @dir: inode of @dentry
4089 * @dentry: pointer to dentry of the base directory
4090 * @mode: mode of the new directory
4092 * Create a directory.
4094 * If the inode has been found through an idmapped mount the idmap of
4095 * the vfsmount must be passed through @idmap. This function will then take
4096 * care to map the inode according to @idmap before checking permissions.
4097 * On non-idmapped mounts or if permission checking is to be performed on the
4098 * raw inode simply passs @nop_mnt_idmap.
4100 int vfs_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
4101 struct dentry
*dentry
, umode_t mode
)
4104 unsigned max_links
= dir
->i_sb
->s_max_links
;
4106 error
= may_create(idmap
, dir
, dentry
);
4110 if (!dir
->i_op
->mkdir
)
4113 mode
= vfs_prepare_mode(idmap
, dir
, mode
, S_IRWXUGO
| S_ISVTX
, 0);
4114 error
= security_inode_mkdir(dir
, dentry
, mode
);
4118 if (max_links
&& dir
->i_nlink
>= max_links
)
4121 error
= dir
->i_op
->mkdir(idmap
, dir
, dentry
, mode
);
4123 fsnotify_mkdir(dir
, dentry
);
4126 EXPORT_SYMBOL(vfs_mkdir
);
4128 int do_mkdirat(int dfd
, struct filename
*name
, umode_t mode
)
4130 struct dentry
*dentry
;
4133 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
4136 dentry
= filename_create(dfd
, name
, &path
, lookup_flags
);
4137 error
= PTR_ERR(dentry
);
4141 error
= security_path_mkdir(&path
, dentry
,
4142 mode_strip_umask(path
.dentry
->d_inode
, mode
));
4144 error
= vfs_mkdir(mnt_idmap(path
.mnt
), path
.dentry
->d_inode
,
4147 done_path_create(&path
, dentry
);
4148 if (retry_estale(error
, lookup_flags
)) {
4149 lookup_flags
|= LOOKUP_REVAL
;
4157 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
4159 return do_mkdirat(dfd
, getname(pathname
), mode
);
4162 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
4164 return do_mkdirat(AT_FDCWD
, getname(pathname
), mode
);
4168 * vfs_rmdir - remove directory
4169 * @idmap: idmap of the mount the inode was found from
4170 * @dir: inode of @dentry
4171 * @dentry: pointer to dentry of the base directory
4173 * Remove a directory.
4175 * If the inode has been found through an idmapped mount the idmap of
4176 * the vfsmount must be passed through @idmap. This function will then take
4177 * care to map the inode according to @idmap before checking permissions.
4178 * On non-idmapped mounts or if permission checking is to be performed on the
4179 * raw inode simply passs @nop_mnt_idmap.
4181 int vfs_rmdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
4182 struct dentry
*dentry
)
4184 int error
= may_delete(idmap
, dir
, dentry
, 1);
4189 if (!dir
->i_op
->rmdir
)
4193 inode_lock(dentry
->d_inode
);
4196 if (is_local_mountpoint(dentry
) ||
4197 (dentry
->d_inode
->i_flags
& S_KERNEL_FILE
))
4200 error
= security_inode_rmdir(dir
, dentry
);
4204 error
= dir
->i_op
->rmdir(dir
, dentry
);
4208 shrink_dcache_parent(dentry
);
4209 dentry
->d_inode
->i_flags
|= S_DEAD
;
4211 detach_mounts(dentry
);
4214 inode_unlock(dentry
->d_inode
);
4217 d_delete_notify(dir
, dentry
);
4220 EXPORT_SYMBOL(vfs_rmdir
);
4222 int do_rmdir(int dfd
, struct filename
*name
)
4225 struct dentry
*dentry
;
4229 unsigned int lookup_flags
= 0;
4231 error
= filename_parentat(dfd
, name
, lookup_flags
, &path
, &last
, &type
);
4247 error
= mnt_want_write(path
.mnt
);
4251 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4252 dentry
= lookup_one_qstr_excl(&last
, path
.dentry
, lookup_flags
);
4253 error
= PTR_ERR(dentry
);
4256 if (!dentry
->d_inode
) {
4260 error
= security_path_rmdir(&path
, dentry
);
4263 error
= vfs_rmdir(mnt_idmap(path
.mnt
), path
.dentry
->d_inode
, dentry
);
4267 inode_unlock(path
.dentry
->d_inode
);
4268 mnt_drop_write(path
.mnt
);
4271 if (retry_estale(error
, lookup_flags
)) {
4272 lookup_flags
|= LOOKUP_REVAL
;
4280 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
4282 return do_rmdir(AT_FDCWD
, getname(pathname
));
4286 * vfs_unlink - unlink a filesystem object
4287 * @idmap: idmap of the mount the inode was found from
4288 * @dir: parent directory
4290 * @delegated_inode: returns victim inode, if the inode is delegated.
4292 * The caller must hold dir->i_mutex.
4294 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4295 * return a reference to the inode in delegated_inode. The caller
4296 * should then break the delegation on that inode and retry. Because
4297 * breaking a delegation may take a long time, the caller should drop
4298 * dir->i_mutex before doing so.
4300 * Alternatively, a caller may pass NULL for delegated_inode. This may
4301 * be appropriate for callers that expect the underlying filesystem not
4302 * to be NFS exported.
4304 * If the inode has been found through an idmapped mount the idmap of
4305 * the vfsmount must be passed through @idmap. This function will then take
4306 * care to map the inode according to @idmap before checking permissions.
4307 * On non-idmapped mounts or if permission checking is to be performed on the
4308 * raw inode simply passs @nop_mnt_idmap.
4310 int vfs_unlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
4311 struct dentry
*dentry
, struct inode
**delegated_inode
)
4313 struct inode
*target
= dentry
->d_inode
;
4314 int error
= may_delete(idmap
, dir
, dentry
, 0);
4319 if (!dir
->i_op
->unlink
)
4323 if (IS_SWAPFILE(target
))
4325 else if (is_local_mountpoint(dentry
))
4328 error
= security_inode_unlink(dir
, dentry
);
4330 error
= try_break_deleg(target
, delegated_inode
);
4333 error
= dir
->i_op
->unlink(dir
, dentry
);
4336 detach_mounts(dentry
);
4341 inode_unlock(target
);
4343 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4344 if (!error
&& dentry
->d_flags
& DCACHE_NFSFS_RENAMED
) {
4345 fsnotify_unlink(dir
, dentry
);
4346 } else if (!error
) {
4347 fsnotify_link_count(target
);
4348 d_delete_notify(dir
, dentry
);
4353 EXPORT_SYMBOL(vfs_unlink
);
4356 * Make sure that the actual truncation of the file will occur outside its
4357 * directory's i_mutex. Truncate can take a long time if there is a lot of
4358 * writeout happening, and we don't want to prevent access to the directory
4359 * while waiting on the I/O.
4361 int do_unlinkat(int dfd
, struct filename
*name
)
4364 struct dentry
*dentry
;
4368 struct inode
*inode
= NULL
;
4369 struct inode
*delegated_inode
= NULL
;
4370 unsigned int lookup_flags
= 0;
4372 error
= filename_parentat(dfd
, name
, lookup_flags
, &path
, &last
, &type
);
4377 if (type
!= LAST_NORM
)
4380 error
= mnt_want_write(path
.mnt
);
4384 inode_lock_nested(path
.dentry
->d_inode
, I_MUTEX_PARENT
);
4385 dentry
= lookup_one_qstr_excl(&last
, path
.dentry
, lookup_flags
);
4386 error
= PTR_ERR(dentry
);
4387 if (!IS_ERR(dentry
)) {
4389 /* Why not before? Because we want correct error value */
4390 if (last
.name
[last
.len
])
4392 inode
= dentry
->d_inode
;
4393 if (d_is_negative(dentry
))
4396 error
= security_path_unlink(&path
, dentry
);
4399 error
= vfs_unlink(mnt_idmap(path
.mnt
), path
.dentry
->d_inode
,
4400 dentry
, &delegated_inode
);
4404 inode_unlock(path
.dentry
->d_inode
);
4406 iput(inode
); /* truncate the inode here */
4408 if (delegated_inode
) {
4409 error
= break_deleg_wait(&delegated_inode
);
4413 mnt_drop_write(path
.mnt
);
4416 if (retry_estale(error
, lookup_flags
)) {
4417 lookup_flags
|= LOOKUP_REVAL
;
4426 if (d_is_negative(dentry
))
4428 else if (d_is_dir(dentry
))
4435 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
4437 if ((flag
& ~AT_REMOVEDIR
) != 0)
4440 if (flag
& AT_REMOVEDIR
)
4441 return do_rmdir(dfd
, getname(pathname
));
4442 return do_unlinkat(dfd
, getname(pathname
));
4445 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
4447 return do_unlinkat(AT_FDCWD
, getname(pathname
));
4451 * vfs_symlink - create symlink
4452 * @idmap: idmap of the mount the inode was found from
4453 * @dir: inode of @dentry
4454 * @dentry: pointer to dentry of the base directory
4455 * @oldname: name of the file to link to
4459 * If the inode has been found through an idmapped mount the idmap of
4460 * the vfsmount must be passed through @idmap. This function will then take
4461 * care to map the inode according to @idmap before checking permissions.
4462 * On non-idmapped mounts or if permission checking is to be performed on the
4463 * raw inode simply passs @nop_mnt_idmap.
4465 int vfs_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
4466 struct dentry
*dentry
, const char *oldname
)
4470 error
= may_create(idmap
, dir
, dentry
);
4474 if (!dir
->i_op
->symlink
)
4477 error
= security_inode_symlink(dir
, dentry
, oldname
);
4481 error
= dir
->i_op
->symlink(idmap
, dir
, dentry
, oldname
);
4483 fsnotify_create(dir
, dentry
);
4486 EXPORT_SYMBOL(vfs_symlink
);
4488 int do_symlinkat(struct filename
*from
, int newdfd
, struct filename
*to
)
4491 struct dentry
*dentry
;
4493 unsigned int lookup_flags
= 0;
4496 error
= PTR_ERR(from
);
4500 dentry
= filename_create(newdfd
, to
, &path
, lookup_flags
);
4501 error
= PTR_ERR(dentry
);
4505 error
= security_path_symlink(&path
, dentry
, from
->name
);
4507 error
= vfs_symlink(mnt_idmap(path
.mnt
), path
.dentry
->d_inode
,
4508 dentry
, from
->name
);
4509 done_path_create(&path
, dentry
);
4510 if (retry_estale(error
, lookup_flags
)) {
4511 lookup_flags
|= LOOKUP_REVAL
;
4520 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
4521 int, newdfd
, const char __user
*, newname
)
4523 return do_symlinkat(getname(oldname
), newdfd
, getname(newname
));
4526 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
4528 return do_symlinkat(getname(oldname
), AT_FDCWD
, getname(newname
));
4532 * vfs_link - create a new link
4533 * @old_dentry: object to be linked
4534 * @idmap: idmap of the mount
4536 * @new_dentry: where to create the new link
4537 * @delegated_inode: returns inode needing a delegation break
4539 * The caller must hold dir->i_mutex
4541 * If vfs_link discovers a delegation on the to-be-linked file in need
4542 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4543 * inode in delegated_inode. The caller should then break the delegation
4544 * and retry. Because breaking a delegation may take a long time, the
4545 * caller should drop the i_mutex before doing so.
4547 * Alternatively, a caller may pass NULL for delegated_inode. This may
4548 * be appropriate for callers that expect the underlying filesystem not
4549 * to be NFS exported.
4551 * If the inode has been found through an idmapped mount the idmap of
4552 * the vfsmount must be passed through @idmap. This function will then take
4553 * care to map the inode according to @idmap before checking permissions.
4554 * On non-idmapped mounts or if permission checking is to be performed on the
4555 * raw inode simply passs @nop_mnt_idmap.
4557 int vfs_link(struct dentry
*old_dentry
, struct mnt_idmap
*idmap
,
4558 struct inode
*dir
, struct dentry
*new_dentry
,
4559 struct inode
**delegated_inode
)
4561 struct inode
*inode
= old_dentry
->d_inode
;
4562 unsigned max_links
= dir
->i_sb
->s_max_links
;
4568 error
= may_create(idmap
, dir
, new_dentry
);
4572 if (dir
->i_sb
!= inode
->i_sb
)
4576 * A link to an append-only or immutable file cannot be created.
4578 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4581 * Updating the link count will likely cause i_uid and i_gid to
4582 * be writen back improperly if their true value is unknown to
4585 if (HAS_UNMAPPED_ID(idmap
, inode
))
4587 if (!dir
->i_op
->link
)
4589 if (S_ISDIR(inode
->i_mode
))
4592 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
4597 /* Make sure we don't allow creating hardlink to an unlinked file */
4598 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
4600 else if (max_links
&& inode
->i_nlink
>= max_links
)
4603 error
= try_break_deleg(inode
, delegated_inode
);
4605 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
4608 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
4609 spin_lock(&inode
->i_lock
);
4610 inode
->i_state
&= ~I_LINKABLE
;
4611 spin_unlock(&inode
->i_lock
);
4613 inode_unlock(inode
);
4615 fsnotify_link(dir
, inode
, new_dentry
);
4618 EXPORT_SYMBOL(vfs_link
);
4621 * Hardlinks are often used in delicate situations. We avoid
4622 * security-related surprises by not following symlinks on the
4625 * We don't follow them on the oldname either to be compatible
4626 * with linux 2.0, and to avoid hard-linking to directories
4627 * and other special files. --ADM
4629 int do_linkat(int olddfd
, struct filename
*old
, int newdfd
,
4630 struct filename
*new, int flags
)
4632 struct mnt_idmap
*idmap
;
4633 struct dentry
*new_dentry
;
4634 struct path old_path
, new_path
;
4635 struct inode
*delegated_inode
= NULL
;
4639 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0) {
4644 * To use null names we require CAP_DAC_READ_SEARCH
4645 * This ensures that not everyone will be able to create
4646 * handlink using the passed filedescriptor.
4648 if (flags
& AT_EMPTY_PATH
&& !capable(CAP_DAC_READ_SEARCH
)) {
4653 if (flags
& AT_SYMLINK_FOLLOW
)
4654 how
|= LOOKUP_FOLLOW
;
4656 error
= filename_lookup(olddfd
, old
, how
, &old_path
, NULL
);
4660 new_dentry
= filename_create(newdfd
, new, &new_path
,
4661 (how
& LOOKUP_REVAL
));
4662 error
= PTR_ERR(new_dentry
);
4663 if (IS_ERR(new_dentry
))
4667 if (old_path
.mnt
!= new_path
.mnt
)
4669 idmap
= mnt_idmap(new_path
.mnt
);
4670 error
= may_linkat(idmap
, &old_path
);
4671 if (unlikely(error
))
4673 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4676 error
= vfs_link(old_path
.dentry
, idmap
, new_path
.dentry
->d_inode
,
4677 new_dentry
, &delegated_inode
);
4679 done_path_create(&new_path
, new_dentry
);
4680 if (delegated_inode
) {
4681 error
= break_deleg_wait(&delegated_inode
);
4683 path_put(&old_path
);
4687 if (retry_estale(error
, how
)) {
4688 path_put(&old_path
);
4689 how
|= LOOKUP_REVAL
;
4693 path_put(&old_path
);
4701 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
4702 int, newdfd
, const char __user
*, newname
, int, flags
)
4704 return do_linkat(olddfd
, getname_uflags(oldname
, flags
),
4705 newdfd
, getname(newname
), flags
);
4708 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4710 return do_linkat(AT_FDCWD
, getname(oldname
), AT_FDCWD
, getname(newname
), 0);
4714 * vfs_rename - rename a filesystem object
4715 * @rd: pointer to &struct renamedata info
4717 * The caller must hold multiple mutexes--see lock_rename()).
4719 * If vfs_rename discovers a delegation in need of breaking at either
4720 * the source or destination, it will return -EWOULDBLOCK and return a
4721 * reference to the inode in delegated_inode. The caller should then
4722 * break the delegation and retry. Because breaking a delegation may
4723 * take a long time, the caller should drop all locks before doing
4726 * Alternatively, a caller may pass NULL for delegated_inode. This may
4727 * be appropriate for callers that expect the underlying filesystem not
4728 * to be NFS exported.
4730 * The worst of all namespace operations - renaming directory. "Perverted"
4731 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4734 * a) we can get into loop creation.
4735 * b) race potential - two innocent renames can create a loop together.
4736 * That's where 4.4 screws up. Current fix: serialization on
4737 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4739 * c) we have to lock _four_ objects - parents and victim (if it exists),
4741 * And that - after we got ->i_mutex on parents (until then we don't know
4742 * whether the target exists). Solution: try to be smart with locking
4743 * order for inodes. We rely on the fact that tree topology may change
4744 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4745 * move will be locked. Thus we can rank directories by the tree
4746 * (ancestors first) and rank all non-directories after them.
4747 * That works since everybody except rename does "lock parent, lookup,
4748 * lock child" and rename is under ->s_vfs_rename_mutex.
4749 * HOWEVER, it relies on the assumption that any object with ->lookup()
4750 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4751 * we'd better make sure that there's no link(2) for them.
4752 * d) conversion from fhandle to dentry may come in the wrong moment - when
4753 * we are removing the target. Solution: we will have to grab ->i_mutex
4754 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4755 * ->i_mutex on parents, which works but leads to some truly excessive
4758 int vfs_rename(struct renamedata
*rd
)
4761 struct inode
*old_dir
= rd
->old_dir
, *new_dir
= rd
->new_dir
;
4762 struct dentry
*old_dentry
= rd
->old_dentry
;
4763 struct dentry
*new_dentry
= rd
->new_dentry
;
4764 struct inode
**delegated_inode
= rd
->delegated_inode
;
4765 unsigned int flags
= rd
->flags
;
4766 bool is_dir
= d_is_dir(old_dentry
);
4767 struct inode
*source
= old_dentry
->d_inode
;
4768 struct inode
*target
= new_dentry
->d_inode
;
4769 bool new_is_dir
= false;
4770 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4771 struct name_snapshot old_name
;
4773 if (source
== target
)
4776 error
= may_delete(rd
->old_mnt_idmap
, old_dir
, old_dentry
, is_dir
);
4781 error
= may_create(rd
->new_mnt_idmap
, new_dir
, new_dentry
);
4783 new_is_dir
= d_is_dir(new_dentry
);
4785 if (!(flags
& RENAME_EXCHANGE
))
4786 error
= may_delete(rd
->new_mnt_idmap
, new_dir
,
4787 new_dentry
, is_dir
);
4789 error
= may_delete(rd
->new_mnt_idmap
, new_dir
,
4790 new_dentry
, new_is_dir
);
4795 if (!old_dir
->i_op
->rename
)
4799 * If we are going to change the parent - check write permissions,
4800 * we'll need to flip '..'.
4802 if (new_dir
!= old_dir
) {
4804 error
= inode_permission(rd
->old_mnt_idmap
, source
,
4809 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4810 error
= inode_permission(rd
->new_mnt_idmap
, target
,
4817 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4822 take_dentry_name_snapshot(&old_name
, old_dentry
);
4825 * Lock all moved children. Moved directories may need to change parent
4826 * pointer so they need the lock to prevent against concurrent
4827 * directory changes moving parent pointer. For regular files we've
4828 * historically always done this. The lockdep locking subclasses are
4829 * somewhat arbitrary but RENAME_EXCHANGE in particular can swap
4830 * regular files and directories so it's difficult to tell which
4831 * subclasses to use.
4833 lock_two_inodes(source
, target
, I_MUTEX_NORMAL
, I_MUTEX_NONDIR2
);
4836 if (IS_SWAPFILE(source
) || (target
&& IS_SWAPFILE(target
)))
4840 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4843 if (max_links
&& new_dir
!= old_dir
) {
4845 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4847 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4848 old_dir
->i_nlink
>= max_links
)
4852 error
= try_break_deleg(source
, delegated_inode
);
4856 if (target
&& !new_is_dir
) {
4857 error
= try_break_deleg(target
, delegated_inode
);
4861 error
= old_dir
->i_op
->rename(rd
->new_mnt_idmap
, old_dir
, old_dentry
,
4862 new_dir
, new_dentry
, flags
);
4866 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4868 shrink_dcache_parent(new_dentry
);
4869 target
->i_flags
|= S_DEAD
;
4871 dont_mount(new_dentry
);
4872 detach_mounts(new_dentry
);
4874 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4875 if (!(flags
& RENAME_EXCHANGE
))
4876 d_move(old_dentry
, new_dentry
);
4878 d_exchange(old_dentry
, new_dentry
);
4881 inode_unlock(source
);
4883 inode_unlock(target
);
4886 fsnotify_move(old_dir
, new_dir
, &old_name
.name
, is_dir
,
4887 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4888 if (flags
& RENAME_EXCHANGE
) {
4889 fsnotify_move(new_dir
, old_dir
, &old_dentry
->d_name
,
4890 new_is_dir
, NULL
, new_dentry
);
4893 release_dentry_name_snapshot(&old_name
);
4897 EXPORT_SYMBOL(vfs_rename
);
4899 int do_renameat2(int olddfd
, struct filename
*from
, int newdfd
,
4900 struct filename
*to
, unsigned int flags
)
4902 struct renamedata rd
;
4903 struct dentry
*old_dentry
, *new_dentry
;
4904 struct dentry
*trap
;
4905 struct path old_path
, new_path
;
4906 struct qstr old_last
, new_last
;
4907 int old_type
, new_type
;
4908 struct inode
*delegated_inode
= NULL
;
4909 unsigned int lookup_flags
= 0, target_flags
= LOOKUP_RENAME_TARGET
;
4910 bool should_retry
= false;
4911 int error
= -EINVAL
;
4913 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4916 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4917 (flags
& RENAME_EXCHANGE
))
4920 if (flags
& RENAME_EXCHANGE
)
4924 error
= filename_parentat(olddfd
, from
, lookup_flags
, &old_path
,
4925 &old_last
, &old_type
);
4929 error
= filename_parentat(newdfd
, to
, lookup_flags
, &new_path
, &new_last
,
4935 if (old_path
.mnt
!= new_path
.mnt
)
4939 if (old_type
!= LAST_NORM
)
4942 if (flags
& RENAME_NOREPLACE
)
4944 if (new_type
!= LAST_NORM
)
4947 error
= mnt_want_write(old_path
.mnt
);
4952 trap
= lock_rename(new_path
.dentry
, old_path
.dentry
);
4954 old_dentry
= lookup_one_qstr_excl(&old_last
, old_path
.dentry
,
4956 error
= PTR_ERR(old_dentry
);
4957 if (IS_ERR(old_dentry
))
4959 /* source must exist */
4961 if (d_is_negative(old_dentry
))
4963 new_dentry
= lookup_one_qstr_excl(&new_last
, new_path
.dentry
,
4964 lookup_flags
| target_flags
);
4965 error
= PTR_ERR(new_dentry
);
4966 if (IS_ERR(new_dentry
))
4969 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4971 if (flags
& RENAME_EXCHANGE
) {
4973 if (d_is_negative(new_dentry
))
4976 if (!d_is_dir(new_dentry
)) {
4978 if (new_last
.name
[new_last
.len
])
4982 /* unless the source is a directory trailing slashes give -ENOTDIR */
4983 if (!d_is_dir(old_dentry
)) {
4985 if (old_last
.name
[old_last
.len
])
4987 if (!(flags
& RENAME_EXCHANGE
) && new_last
.name
[new_last
.len
])
4990 /* source should not be ancestor of target */
4992 if (old_dentry
== trap
)
4994 /* target should not be an ancestor of source */
4995 if (!(flags
& RENAME_EXCHANGE
))
4997 if (new_dentry
== trap
)
5000 error
= security_path_rename(&old_path
, old_dentry
,
5001 &new_path
, new_dentry
, flags
);
5005 rd
.old_dir
= old_path
.dentry
->d_inode
;
5006 rd
.old_dentry
= old_dentry
;
5007 rd
.old_mnt_idmap
= mnt_idmap(old_path
.mnt
);
5008 rd
.new_dir
= new_path
.dentry
->d_inode
;
5009 rd
.new_dentry
= new_dentry
;
5010 rd
.new_mnt_idmap
= mnt_idmap(new_path
.mnt
);
5011 rd
.delegated_inode
= &delegated_inode
;
5013 error
= vfs_rename(&rd
);
5019 unlock_rename(new_path
.dentry
, old_path
.dentry
);
5020 if (delegated_inode
) {
5021 error
= break_deleg_wait(&delegated_inode
);
5025 mnt_drop_write(old_path
.mnt
);
5027 if (retry_estale(error
, lookup_flags
))
5028 should_retry
= true;
5029 path_put(&new_path
);
5031 path_put(&old_path
);
5033 should_retry
= false;
5034 lookup_flags
|= LOOKUP_REVAL
;
5043 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
5044 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
5046 return do_renameat2(olddfd
, getname(oldname
), newdfd
, getname(newname
),
5050 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
5051 int, newdfd
, const char __user
*, newname
)
5053 return do_renameat2(olddfd
, getname(oldname
), newdfd
, getname(newname
),
5057 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
5059 return do_renameat2(AT_FDCWD
, getname(oldname
), AT_FDCWD
,
5060 getname(newname
), 0);
5063 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
5065 int len
= PTR_ERR(link
);
5070 if (len
> (unsigned) buflen
)
5072 if (copy_to_user(buffer
, link
, len
))
5079 * vfs_readlink - copy symlink body into userspace buffer
5080 * @dentry: dentry on which to get symbolic link
5081 * @buffer: user memory pointer
5082 * @buflen: size of buffer
5084 * Does not touch atime. That's up to the caller if necessary
5086 * Does not call security hook.
5088 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
5090 struct inode
*inode
= d_inode(dentry
);
5091 DEFINE_DELAYED_CALL(done
);
5095 if (unlikely(!(inode
->i_opflags
& IOP_DEFAULT_READLINK
))) {
5096 if (unlikely(inode
->i_op
->readlink
))
5097 return inode
->i_op
->readlink(dentry
, buffer
, buflen
);
5099 if (!d_is_symlink(dentry
))
5102 spin_lock(&inode
->i_lock
);
5103 inode
->i_opflags
|= IOP_DEFAULT_READLINK
;
5104 spin_unlock(&inode
->i_lock
);
5107 link
= READ_ONCE(inode
->i_link
);
5109 link
= inode
->i_op
->get_link(dentry
, inode
, &done
);
5111 return PTR_ERR(link
);
5113 res
= readlink_copy(buffer
, buflen
, link
);
5114 do_delayed_call(&done
);
5117 EXPORT_SYMBOL(vfs_readlink
);
5120 * vfs_get_link - get symlink body
5121 * @dentry: dentry on which to get symbolic link
5122 * @done: caller needs to free returned data with this
5124 * Calls security hook and i_op->get_link() on the supplied inode.
5126 * It does not touch atime. That's up to the caller if necessary.
5128 * Does not work on "special" symlinks like /proc/$$/fd/N
5130 const char *vfs_get_link(struct dentry
*dentry
, struct delayed_call
*done
)
5132 const char *res
= ERR_PTR(-EINVAL
);
5133 struct inode
*inode
= d_inode(dentry
);
5135 if (d_is_symlink(dentry
)) {
5136 res
= ERR_PTR(security_inode_readlink(dentry
));
5138 res
= inode
->i_op
->get_link(dentry
, inode
, done
);
5142 EXPORT_SYMBOL(vfs_get_link
);
5144 /* get the link contents into pagecache */
5145 const char *page_get_link(struct dentry
*dentry
, struct inode
*inode
,
5146 struct delayed_call
*callback
)
5150 struct address_space
*mapping
= inode
->i_mapping
;
5153 page
= find_get_page(mapping
, 0);
5155 return ERR_PTR(-ECHILD
);
5156 if (!PageUptodate(page
)) {
5158 return ERR_PTR(-ECHILD
);
5161 page
= read_mapping_page(mapping
, 0, NULL
);
5165 set_delayed_call(callback
, page_put_link
, page
);
5166 BUG_ON(mapping_gfp_mask(mapping
) & __GFP_HIGHMEM
);
5167 kaddr
= page_address(page
);
5168 nd_terminate_link(kaddr
, inode
->i_size
, PAGE_SIZE
- 1);
5172 EXPORT_SYMBOL(page_get_link
);
5174 void page_put_link(void *arg
)
5178 EXPORT_SYMBOL(page_put_link
);
5180 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
5182 DEFINE_DELAYED_CALL(done
);
5183 int res
= readlink_copy(buffer
, buflen
,
5184 page_get_link(dentry
, d_inode(dentry
),
5186 do_delayed_call(&done
);
5189 EXPORT_SYMBOL(page_readlink
);
5191 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
5193 struct address_space
*mapping
= inode
->i_mapping
;
5194 const struct address_space_operations
*aops
= mapping
->a_ops
;
5195 bool nofs
= !mapping_gfp_constraint(mapping
, __GFP_FS
);
5197 void *fsdata
= NULL
;
5203 flags
= memalloc_nofs_save();
5204 err
= aops
->write_begin(NULL
, mapping
, 0, len
-1, &page
, &fsdata
);
5206 memalloc_nofs_restore(flags
);
5210 memcpy(page_address(page
), symname
, len
-1);
5212 err
= aops
->write_end(NULL
, mapping
, 0, len
-1, len
-1,
5219 mark_inode_dirty(inode
);
5224 EXPORT_SYMBOL(page_symlink
);
5226 const struct inode_operations page_symlink_inode_operations
= {
5227 .get_link
= page_get_link
,
5229 EXPORT_SYMBOL(page_symlink_inode_operations
);