4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static int acl_permission_check(struct inode
*inode
, int mask
, unsigned int flags
,
173 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
183 int error
= check_acl(inode
, mask
, flags
);
184 if (error
!= -EAGAIN
)
188 if (in_group_p(inode
->i_gid
))
193 * If the DACs are ok we don't need any capability check.
195 if ((mask
& ~mode
) == 0)
201 * generic_permission - check for access rights on a Posix-like filesystem
202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
205 * @flags: IPERM_FLAG_ flags.
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
210 * are used for other things.
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
216 int generic_permission(struct inode
*inode
, int mask
, unsigned int flags
,
217 int (*check_acl
)(struct inode
*inode
, int mask
, unsigned int flags
))
222 * Do the basic POSIX ACL permission checks.
224 ret
= acl_permission_check(inode
, mask
, flags
, check_acl
);
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
232 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
233 if (capable(CAP_DAC_OVERRIDE
))
237 * Searching includes executable on directories, else just read.
239 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
240 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
241 if (capable(CAP_DAC_READ_SEARCH
))
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
257 int inode_permission(struct inode
*inode
, int mask
)
261 if (mask
& MAY_WRITE
) {
262 umode_t mode
= inode
->i_mode
;
265 * Nobody gets write access to a read-only fs.
267 if (IS_RDONLY(inode
) &&
268 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
272 * Nobody gets write access to an immutable file.
274 if (IS_IMMUTABLE(inode
))
278 if (inode
->i_op
->permission
)
279 retval
= inode
->i_op
->permission(inode
, mask
, 0);
281 retval
= generic_permission(inode
, mask
, 0,
282 inode
->i_op
->check_acl
);
287 retval
= devcgroup_inode_permission(inode
, mask
);
291 return security_inode_permission(inode
, mask
);
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
299 * Used to check for read/write/execute permissions on an already opened
303 * Do not use this function in new code. All access checks should
304 * be done using inode_permission().
306 int file_permission(struct file
*file
, int mask
)
308 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
329 int get_write_access(struct inode
* inode
)
331 spin_lock(&inode
->i_lock
);
332 if (atomic_read(&inode
->i_writecount
) < 0) {
333 spin_unlock(&inode
->i_lock
);
336 atomic_inc(&inode
->i_writecount
);
337 spin_unlock(&inode
->i_lock
);
342 int deny_write_access(struct file
* file
)
344 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
346 spin_lock(&inode
->i_lock
);
347 if (atomic_read(&inode
->i_writecount
) > 0) {
348 spin_unlock(&inode
->i_lock
);
351 atomic_dec(&inode
->i_writecount
);
352 spin_unlock(&inode
->i_lock
);
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
361 * Given a path increment the reference count to the dentry and the vfsmount.
363 void path_get(struct path
*path
)
368 EXPORT_SYMBOL(path_get
);
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
374 * Given a path decrement the reference count to the dentry and the vfsmount.
376 void path_put(struct path
*path
)
381 EXPORT_SYMBOL(path_put
);
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
386 * Returns: 0 on success, -ECHILD on failure
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
400 static int nameidata_drop_rcu(struct nameidata
*nd
)
402 struct fs_struct
*fs
= current
->fs
;
403 struct dentry
*dentry
= nd
->path
.dentry
;
405 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
407 spin_lock(&fs
->lock
);
408 if (nd
->root
.mnt
!= fs
->root
.mnt
||
409 nd
->root
.dentry
!= fs
->root
.dentry
)
412 spin_lock(&dentry
->d_lock
);
413 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
415 BUG_ON(nd
->inode
!= dentry
->d_inode
);
416 spin_unlock(&dentry
->d_lock
);
419 spin_unlock(&fs
->lock
);
421 mntget(nd
->path
.mnt
);
424 br_read_unlock(vfsmount_lock
);
425 nd
->flags
&= ~LOOKUP_RCU
;
428 spin_unlock(&dentry
->d_lock
);
431 spin_unlock(&fs
->lock
);
435 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
438 if (nd
->flags
& LOOKUP_RCU
)
439 return nameidata_drop_rcu(nd
);
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
447 * Returns: 0 on success, -ECHILD on failure
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
453 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
455 struct fs_struct
*fs
= current
->fs
;
456 struct dentry
*parent
= nd
->path
.dentry
;
458 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
460 spin_lock(&fs
->lock
);
461 if (nd
->root
.mnt
!= fs
->root
.mnt
||
462 nd
->root
.dentry
!= fs
->root
.dentry
)
465 spin_lock(&parent
->d_lock
);
466 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
467 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
474 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
475 BUG_ON(!parent
->d_count
);
477 spin_unlock(&dentry
->d_lock
);
478 spin_unlock(&parent
->d_lock
);
481 spin_unlock(&fs
->lock
);
483 mntget(nd
->path
.mnt
);
486 br_read_unlock(vfsmount_lock
);
487 nd
->flags
&= ~LOOKUP_RCU
;
490 spin_unlock(&dentry
->d_lock
);
491 spin_unlock(&parent
->d_lock
);
494 spin_unlock(&fs
->lock
);
498 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
501 if (nd
->flags
& LOOKUP_RCU
)
502 return nameidata_dentry_drop_rcu(nd
, dentry
);
507 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
508 * @nd: nameidata pathwalk data to drop
509 * Returns: 0 on success, -ECHILD on failure
511 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
512 * nd->path should be the final element of the lookup, so nd->root is discarded.
513 * Must be called from rcu-walk context.
515 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
517 struct dentry
*dentry
= nd
->path
.dentry
;
519 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
520 nd
->flags
&= ~LOOKUP_RCU
;
522 spin_lock(&dentry
->d_lock
);
523 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
525 BUG_ON(nd
->inode
!= dentry
->d_inode
);
526 spin_unlock(&dentry
->d_lock
);
528 mntget(nd
->path
.mnt
);
531 br_read_unlock(vfsmount_lock
);
536 spin_unlock(&dentry
->d_lock
);
538 br_read_unlock(vfsmount_lock
);
542 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
543 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
545 if (likely(nd
->flags
& LOOKUP_RCU
))
546 return nameidata_drop_rcu_last(nd
);
551 * release_open_intent - free up open intent resources
552 * @nd: pointer to nameidata
554 void release_open_intent(struct nameidata
*nd
)
556 struct file
*file
= nd
->intent
.open
.file
;
558 if (file
&& !IS_ERR(file
)) {
559 if (file
->f_path
.dentry
== NULL
)
566 static inline int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
568 return dentry
->d_op
->d_revalidate(dentry
, nd
);
571 static struct dentry
*
572 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
574 int status
= d_revalidate(dentry
, nd
);
575 if (unlikely(status
<= 0)) {
577 * The dentry failed validation.
578 * If d_revalidate returned 0 attempt to invalidate
579 * the dentry otherwise d_revalidate is asking us
580 * to return a fail status.
584 dentry
= ERR_PTR(status
);
585 } else if (!d_invalidate(dentry
)) {
593 static inline struct dentry
*
594 do_revalidate_rcu(struct dentry
*dentry
, struct nameidata
*nd
)
596 int status
= d_revalidate(dentry
, nd
);
597 if (likely(status
> 0))
599 if (status
== -ECHILD
) {
600 if (nameidata_dentry_drop_rcu(nd
, dentry
))
601 return ERR_PTR(-ECHILD
);
602 return do_revalidate(dentry
, nd
);
605 return ERR_PTR(status
);
606 /* Don't d_invalidate in rcu-walk mode */
607 if (nameidata_dentry_drop_rcu(nd
, dentry
))
608 return ERR_PTR(-ECHILD
);
609 if (!d_invalidate(dentry
)) {
616 static inline int need_reval_dot(struct dentry
*dentry
)
618 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
621 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
628 * force_reval_path - force revalidation of a dentry
630 * In some situations the path walking code will trust dentries without
631 * revalidating them. This causes problems for filesystems that depend on
632 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
633 * (which indicates that it's possible for the dentry to go stale), force
634 * a d_revalidate call before proceeding.
636 * Returns 0 if the revalidation was successful. If the revalidation fails,
637 * either return the error returned by d_revalidate or -ESTALE if the
638 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
639 * invalidate the dentry. It's up to the caller to handle putting references
640 * to the path if necessary.
643 force_reval_path(struct path
*path
, struct nameidata
*nd
)
646 struct dentry
*dentry
= path
->dentry
;
649 * only check on filesystems where it's possible for the dentry to
652 if (!need_reval_dot(dentry
))
655 status
= d_revalidate(dentry
, nd
);
660 d_invalidate(dentry
);
667 * Short-cut version of permission(), for calling on directories
668 * during pathname resolution. Combines parts of permission()
669 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
671 * If appropriate, check DAC only. If not appropriate, or
672 * short-cut DAC fails, then call ->permission() to do more
673 * complete permission check.
675 static inline int exec_permission(struct inode
*inode
, unsigned int flags
)
679 if (inode
->i_op
->permission
) {
680 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
, flags
);
682 ret
= acl_permission_check(inode
, MAY_EXEC
, flags
,
683 inode
->i_op
->check_acl
);
690 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
695 return security_inode_exec_permission(inode
, flags
);
698 static __always_inline
void set_root(struct nameidata
*nd
)
701 get_fs_root(current
->fs
, &nd
->root
);
704 static int link_path_walk(const char *, struct nameidata
*);
706 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
709 struct fs_struct
*fs
= current
->fs
;
713 seq
= read_seqcount_begin(&fs
->seq
);
715 } while (read_seqcount_retry(&fs
->seq
, seq
));
719 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
732 nd
->inode
= nd
->path
.dentry
->d_inode
;
734 ret
= link_path_walk(link
, nd
);
738 return PTR_ERR(link
);
741 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
744 if (path
->mnt
!= nd
->path
.mnt
)
748 static inline void path_to_nameidata(const struct path
*path
,
749 struct nameidata
*nd
)
751 if (!(nd
->flags
& LOOKUP_RCU
)) {
752 dput(nd
->path
.dentry
);
753 if (nd
->path
.mnt
!= path
->mnt
)
754 mntput(nd
->path
.mnt
);
756 nd
->path
.mnt
= path
->mnt
;
757 nd
->path
.dentry
= path
->dentry
;
760 static __always_inline
int
761 __do_follow_link(const struct path
*link
, struct nameidata
*nd
, void **p
)
764 struct dentry
*dentry
= link
->dentry
;
766 BUG_ON(nd
->flags
& LOOKUP_RCU
);
768 touch_atime(link
->mnt
, dentry
);
769 nd_set_link(nd
, NULL
);
771 if (link
->mnt
== nd
->path
.mnt
)
774 nd
->last_type
= LAST_BIND
;
775 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
778 char *s
= nd_get_link(nd
);
781 error
= __vfs_follow_link(nd
, s
);
782 else if (nd
->last_type
== LAST_BIND
) {
783 error
= force_reval_path(&nd
->path
, nd
);
792 * This limits recursive symlink follows to 8, while
793 * limiting consecutive symlinks to 40.
795 * Without that kind of total limit, nasty chains of consecutive
796 * symlinks can cause almost arbitrarily long lookups.
798 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
803 /* We drop rcu-walk here */
804 if (nameidata_dentry_drop_rcu_maybe(nd
, path
->dentry
))
807 if (current
->link_count
>= MAX_NESTED_LINKS
)
809 if (current
->total_link_count
>= 40)
811 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
813 err
= security_inode_follow_link(path
->dentry
, nd
);
816 current
->link_count
++;
817 current
->total_link_count
++;
819 err
= __do_follow_link(path
, nd
, &cookie
);
820 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
821 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
823 current
->link_count
--;
827 path_put_conditional(path
, nd
);
832 static int follow_up_rcu(struct path
*path
)
834 struct vfsmount
*parent
;
835 struct dentry
*mountpoint
;
837 parent
= path
->mnt
->mnt_parent
;
838 if (parent
== path
->mnt
)
840 mountpoint
= path
->mnt
->mnt_mountpoint
;
841 path
->dentry
= mountpoint
;
846 int follow_up(struct path
*path
)
848 struct vfsmount
*parent
;
849 struct dentry
*mountpoint
;
851 br_read_lock(vfsmount_lock
);
852 parent
= path
->mnt
->mnt_parent
;
853 if (parent
== path
->mnt
) {
854 br_read_unlock(vfsmount_lock
);
858 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
859 br_read_unlock(vfsmount_lock
);
861 path
->dentry
= mountpoint
;
868 * Perform an automount
869 * - return -EISDIR to tell follow_managed() to stop and return the path we
872 static int follow_automount(struct path
*path
, unsigned flags
,
875 struct vfsmount
*mnt
;
878 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
881 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
882 * and this is the terminal part of the path.
884 if ((flags
& LOOKUP_NO_AUTOMOUNT
) && !(flags
& LOOKUP_CONTINUE
))
885 return -EISDIR
; /* we actually want to stop here */
887 /* We want to mount if someone is trying to open/create a file of any
888 * type under the mountpoint, wants to traverse through the mountpoint
889 * or wants to open the mounted directory.
891 * We don't want to mount if someone's just doing a stat and they've
892 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
893 * appended a '/' to the name.
895 if (!(flags
& LOOKUP_FOLLOW
) &&
896 !(flags
& (LOOKUP_CONTINUE
| LOOKUP_DIRECTORY
|
897 LOOKUP_OPEN
| LOOKUP_CREATE
)))
900 current
->total_link_count
++;
901 if (current
->total_link_count
>= 40)
904 mnt
= path
->dentry
->d_op
->d_automount(path
);
907 * The filesystem is allowed to return -EISDIR here to indicate
908 * it doesn't want to automount. For instance, autofs would do
909 * this so that its userspace daemon can mount on this dentry.
911 * However, we can only permit this if it's a terminal point in
912 * the path being looked up; if it wasn't then the remainder of
913 * the path is inaccessible and we should say so.
915 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_CONTINUE
))
920 if (!mnt
) /* mount collision */
923 err
= finish_automount(mnt
, path
);
927 /* Someone else made a mount here whilst we were busy */
934 path
->dentry
= dget(mnt
->mnt_root
);
944 * Handle a dentry that is managed in some way.
945 * - Flagged for transit management (autofs)
946 * - Flagged as mountpoint
947 * - Flagged as automount point
949 * This may only be called in refwalk mode.
951 * Serialization is taken care of in namespace.c
953 static int follow_managed(struct path
*path
, unsigned flags
)
956 bool need_mntput
= false;
959 /* Given that we're not holding a lock here, we retain the value in a
960 * local variable for each dentry as we look at it so that we don't see
961 * the components of that value change under us */
962 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
963 managed
&= DCACHE_MANAGED_DENTRY
,
964 unlikely(managed
!= 0)) {
965 /* Allow the filesystem to manage the transit without i_mutex
967 if (managed
& DCACHE_MANAGE_TRANSIT
) {
968 BUG_ON(!path
->dentry
->d_op
);
969 BUG_ON(!path
->dentry
->d_op
->d_manage
);
970 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
,
973 return ret
== -EISDIR
? 0 : ret
;
976 /* Transit to a mounted filesystem. */
977 if (managed
& DCACHE_MOUNTED
) {
978 struct vfsmount
*mounted
= lookup_mnt(path
);
984 path
->dentry
= dget(mounted
->mnt_root
);
989 /* Something is mounted on this dentry in another
990 * namespace and/or whatever was mounted there in this
991 * namespace got unmounted before we managed to get the
995 /* Handle an automount point */
996 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
997 ret
= follow_automount(path
, flags
, &need_mntput
);
999 return ret
== -EISDIR
? 0 : ret
;
1003 /* We didn't change the current path point */
1009 int follow_down_one(struct path
*path
)
1011 struct vfsmount
*mounted
;
1013 mounted
= lookup_mnt(path
);
1017 path
->mnt
= mounted
;
1018 path
->dentry
= dget(mounted
->mnt_root
);
1025 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
1026 * meet a managed dentry and we're not walking to "..". True is returned to
1027 * continue, false to abort.
1029 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1030 struct inode
**inode
, bool reverse_transit
)
1032 while (d_mountpoint(path
->dentry
)) {
1033 struct vfsmount
*mounted
;
1034 if (unlikely(path
->dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) &&
1036 path
->dentry
->d_op
->d_manage(path
->dentry
, false, true) < 0)
1038 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1041 path
->mnt
= mounted
;
1042 path
->dentry
= mounted
->mnt_root
;
1043 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1044 *inode
= path
->dentry
->d_inode
;
1047 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1048 return reverse_transit
;
1052 static int follow_dotdot_rcu(struct nameidata
*nd
)
1054 struct inode
*inode
= nd
->inode
;
1059 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1060 nd
->path
.mnt
== nd
->root
.mnt
) {
1063 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1064 struct dentry
*old
= nd
->path
.dentry
;
1065 struct dentry
*parent
= old
->d_parent
;
1068 seq
= read_seqcount_begin(&parent
->d_seq
);
1069 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1071 inode
= parent
->d_inode
;
1072 nd
->path
.dentry
= parent
;
1076 if (!follow_up_rcu(&nd
->path
))
1078 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1079 inode
= nd
->path
.dentry
->d_inode
;
1081 __follow_mount_rcu(nd
, &nd
->path
, &inode
, true);
1088 * Follow down to the covering mount currently visible to userspace. At each
1089 * point, the filesystem owning that dentry may be queried as to whether the
1090 * caller is permitted to proceed or not.
1092 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1095 int follow_down(struct path
*path
, bool mounting_here
)
1100 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1101 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1102 /* Allow the filesystem to manage the transit without i_mutex
1105 * We indicate to the filesystem if someone is trying to mount
1106 * something here. This gives autofs the chance to deny anyone
1107 * other than its daemon the right to mount on its
1110 * The filesystem may sleep at this point.
1112 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1113 BUG_ON(!path
->dentry
->d_op
);
1114 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1115 ret
= path
->dentry
->d_op
->d_manage(
1116 path
->dentry
, mounting_here
, false);
1118 return ret
== -EISDIR
? 0 : ret
;
1121 /* Transit to a mounted filesystem. */
1122 if (managed
& DCACHE_MOUNTED
) {
1123 struct vfsmount
*mounted
= lookup_mnt(path
);
1128 path
->mnt
= mounted
;
1129 path
->dentry
= dget(mounted
->mnt_root
);
1133 /* Don't handle automount points here */
1140 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1142 static void follow_mount(struct path
*path
)
1144 while (d_mountpoint(path
->dentry
)) {
1145 struct vfsmount
*mounted
= lookup_mnt(path
);
1150 path
->mnt
= mounted
;
1151 path
->dentry
= dget(mounted
->mnt_root
);
1155 static void follow_dotdot(struct nameidata
*nd
)
1160 struct dentry
*old
= nd
->path
.dentry
;
1162 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1163 nd
->path
.mnt
== nd
->root
.mnt
) {
1166 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1167 /* rare case of legitimate dget_parent()... */
1168 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1172 if (!follow_up(&nd
->path
))
1175 follow_mount(&nd
->path
);
1176 nd
->inode
= nd
->path
.dentry
->d_inode
;
1180 * Allocate a dentry with name and parent, and perform a parent
1181 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1182 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1183 * have verified that no child exists while under i_mutex.
1185 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1186 struct qstr
*name
, struct nameidata
*nd
)
1188 struct inode
*inode
= parent
->d_inode
;
1189 struct dentry
*dentry
;
1192 /* Don't create child dentry for a dead directory. */
1193 if (unlikely(IS_DEADDIR(inode
)))
1194 return ERR_PTR(-ENOENT
);
1196 dentry
= d_alloc(parent
, name
);
1197 if (unlikely(!dentry
))
1198 return ERR_PTR(-ENOMEM
);
1200 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1201 if (unlikely(old
)) {
1209 * It's more convoluted than I'd like it to be, but... it's still fairly
1210 * small and for now I'd prefer to have fast path as straight as possible.
1211 * It _is_ time-critical.
1213 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1214 struct path
*path
, struct inode
**inode
)
1216 struct vfsmount
*mnt
= nd
->path
.mnt
;
1217 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1222 * See if the low-level filesystem might want
1223 * to use its own hash..
1225 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1226 err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1232 * Rename seqlock is not required here because in the off chance
1233 * of a false negative due to a concurrent rename, we're going to
1234 * do the non-racy lookup, below.
1236 if (nd
->flags
& LOOKUP_RCU
) {
1240 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1242 if (nameidata_drop_rcu(nd
))
1246 /* Memory barrier in read_seqcount_begin of child is enough */
1247 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1251 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1252 dentry
= do_revalidate_rcu(dentry
, nd
);
1257 if (!(nd
->flags
& LOOKUP_RCU
))
1261 path
->dentry
= dentry
;
1262 if (likely(__follow_mount_rcu(nd
, path
, inode
, false)))
1264 if (nameidata_drop_rcu(nd
))
1268 dentry
= __d_lookup(parent
, name
);
1272 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1273 dentry
= do_revalidate(dentry
, nd
);
1281 path
->dentry
= dentry
;
1282 err
= follow_managed(path
, nd
->flags
);
1283 if (unlikely(err
< 0)) {
1284 path_put_conditional(path
, nd
);
1287 *inode
= path
->dentry
->d_inode
;
1291 dir
= parent
->d_inode
;
1292 BUG_ON(nd
->inode
!= dir
);
1294 mutex_lock(&dir
->i_mutex
);
1296 * First re-do the cached lookup just in case it was created
1297 * while we waited for the directory semaphore, or the first
1298 * lookup failed due to an unrelated rename.
1300 * This could use version numbering or similar to avoid unnecessary
1301 * cache lookups, but then we'd have to do the first lookup in the
1302 * non-racy way. However in the common case here, everything should
1303 * be hot in cache, so would it be a big win?
1305 dentry
= d_lookup(parent
, name
);
1306 if (likely(!dentry
)) {
1307 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1308 mutex_unlock(&dir
->i_mutex
);
1314 * Uhhuh! Nasty case: the cache was re-populated while
1315 * we waited on the semaphore. Need to revalidate.
1317 mutex_unlock(&dir
->i_mutex
);
1321 return PTR_ERR(dentry
);
1326 * This is the basic name resolution function, turning a pathname into
1327 * the final dentry. We expect 'base' to be positive and a directory.
1329 * Returns 0 and nd will have valid dentry and mnt on success.
1330 * Returns error and drops reference to input namei data on failure.
1332 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1336 unsigned int lookup_flags
= nd
->flags
;
1344 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1346 /* At this point we know we have a real path component. */
1348 struct inode
*inode
;
1353 nd
->flags
|= LOOKUP_CONTINUE
;
1354 if (nd
->flags
& LOOKUP_RCU
) {
1355 err
= exec_permission(nd
->inode
, IPERM_FLAG_RCU
);
1356 if (err
== -ECHILD
) {
1357 if (nameidata_drop_rcu(nd
))
1363 err
= exec_permission(nd
->inode
, 0);
1369 c
= *(const unsigned char *)name
;
1371 hash
= init_name_hash();
1374 hash
= partial_name_hash(c
, hash
);
1375 c
= *(const unsigned char *)name
;
1376 } while (c
&& (c
!= '/'));
1377 this.len
= name
- (const char *) this.name
;
1378 this.hash
= end_name_hash(hash
);
1380 /* remove trailing slashes? */
1382 goto last_component
;
1383 while (*++name
== '/');
1385 goto last_with_slashes
;
1388 * "." and ".." are special - ".." especially so because it has
1389 * to be able to know about the current root directory and
1390 * parent relationships.
1392 if (this.name
[0] == '.') switch (this.len
) {
1396 if (this.name
[1] != '.')
1398 if (nd
->flags
& LOOKUP_RCU
) {
1399 if (follow_dotdot_rcu(nd
))
1407 /* This does the actual lookups.. */
1408 err
= do_lookup(nd
, &this, &next
, &inode
);
1415 if (inode
->i_op
->follow_link
) {
1416 BUG_ON(inode
!= next
.dentry
->d_inode
);
1417 err
= do_follow_link(&next
, nd
);
1420 nd
->inode
= nd
->path
.dentry
->d_inode
;
1425 path_to_nameidata(&next
, nd
);
1429 if (!nd
->inode
->i_op
->lookup
)
1432 /* here ends the main loop */
1435 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1437 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1438 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1439 if (lookup_flags
& LOOKUP_PARENT
)
1441 if (this.name
[0] == '.') switch (this.len
) {
1445 if (this.name
[1] != '.')
1447 if (nd
->flags
& LOOKUP_RCU
) {
1448 if (follow_dotdot_rcu(nd
))
1456 err
= do_lookup(nd
, &this, &next
, &inode
);
1459 if (inode
&& unlikely(inode
->i_op
->follow_link
) &&
1460 (lookup_flags
& LOOKUP_FOLLOW
)) {
1461 BUG_ON(inode
!= next
.dentry
->d_inode
);
1462 err
= do_follow_link(&next
, nd
);
1465 nd
->inode
= nd
->path
.dentry
->d_inode
;
1467 path_to_nameidata(&next
, nd
);
1473 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1475 if (!nd
->inode
->i_op
->lookup
)
1481 nd
->last_type
= LAST_NORM
;
1482 if (this.name
[0] != '.')
1485 nd
->last_type
= LAST_DOT
;
1486 else if (this.len
== 2 && this.name
[1] == '.')
1487 nd
->last_type
= LAST_DOTDOT
;
1492 * We bypassed the ordinary revalidation routines.
1493 * We may need to check the cached dentry for staleness.
1495 if (need_reval_dot(nd
->path
.dentry
)) {
1496 if (nameidata_drop_rcu_last_maybe(nd
))
1498 /* Note: we do not d_invalidate() */
1499 err
= d_revalidate(nd
->path
.dentry
, nd
);
1507 if (nameidata_drop_rcu_last_maybe(nd
))
1511 if (!(nd
->flags
& LOOKUP_RCU
))
1512 path_put_conditional(&next
, nd
);
1515 if (!(nd
->flags
& LOOKUP_RCU
))
1516 path_put(&nd
->path
);
1521 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1523 current
->total_link_count
= 0;
1525 return link_path_walk(name
, nd
);
1528 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1530 current
->total_link_count
= 0;
1532 return link_path_walk(name
, nd
);
1535 static int path_walk(const char *name
, struct nameidata
*nd
)
1537 struct path save
= nd
->path
;
1540 current
->total_link_count
= 0;
1542 /* make sure the stuff we saved doesn't go away */
1545 result
= link_path_walk(name
, nd
);
1546 if (result
== -ESTALE
) {
1547 /* nd->path had been dropped */
1548 current
->total_link_count
= 0;
1550 path_get(&nd
->path
);
1551 nd
->flags
|= LOOKUP_REVAL
;
1552 result
= link_path_walk(name
, nd
);
1560 static void path_finish_rcu(struct nameidata
*nd
)
1562 if (nd
->flags
& LOOKUP_RCU
) {
1563 /* RCU dangling. Cancel it. */
1564 nd
->flags
&= ~LOOKUP_RCU
;
1565 nd
->root
.mnt
= NULL
;
1567 br_read_unlock(vfsmount_lock
);
1573 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1579 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1580 nd
->flags
= flags
| LOOKUP_RCU
;
1582 nd
->root
.mnt
= NULL
;
1586 struct fs_struct
*fs
= current
->fs
;
1589 br_read_lock(vfsmount_lock
);
1593 seq
= read_seqcount_begin(&fs
->seq
);
1594 nd
->root
= fs
->root
;
1595 nd
->path
= nd
->root
;
1596 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1597 } while (read_seqcount_retry(&fs
->seq
, seq
));
1599 } else if (dfd
== AT_FDCWD
) {
1600 struct fs_struct
*fs
= current
->fs
;
1603 br_read_lock(vfsmount_lock
);
1607 seq
= read_seqcount_begin(&fs
->seq
);
1609 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1610 } while (read_seqcount_retry(&fs
->seq
, seq
));
1613 struct dentry
*dentry
;
1615 file
= fget_light(dfd
, &fput_needed
);
1620 dentry
= file
->f_path
.dentry
;
1623 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1626 retval
= file_permission(file
, MAY_EXEC
);
1630 nd
->path
= file
->f_path
;
1634 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1635 br_read_lock(vfsmount_lock
);
1638 nd
->inode
= nd
->path
.dentry
->d_inode
;
1642 fput_light(file
, fput_needed
);
1647 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1653 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1656 nd
->root
.mnt
= NULL
;
1660 nd
->path
= nd
->root
;
1661 path_get(&nd
->root
);
1662 } else if (dfd
== AT_FDCWD
) {
1663 get_fs_pwd(current
->fs
, &nd
->path
);
1665 struct dentry
*dentry
;
1667 file
= fget_light(dfd
, &fput_needed
);
1672 dentry
= file
->f_path
.dentry
;
1675 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1678 retval
= file_permission(file
, MAY_EXEC
);
1682 nd
->path
= file
->f_path
;
1683 path_get(&file
->f_path
);
1685 fput_light(file
, fput_needed
);
1687 nd
->inode
= nd
->path
.dentry
->d_inode
;
1691 fput_light(file
, fput_needed
);
1696 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1697 static int do_path_lookup(int dfd
, const char *name
,
1698 unsigned int flags
, struct nameidata
*nd
)
1703 * Path walking is largely split up into 2 different synchronisation
1704 * schemes, rcu-walk and ref-walk (explained in
1705 * Documentation/filesystems/path-lookup.txt). These share much of the
1706 * path walk code, but some things particularly setup, cleanup, and
1707 * following mounts are sufficiently divergent that functions are
1708 * duplicated. Typically there is a function foo(), and its RCU
1709 * analogue, foo_rcu().
1711 * -ECHILD is the error number of choice (just to avoid clashes) that
1712 * is returned if some aspect of an rcu-walk fails. Such an error must
1713 * be handled by restarting a traditional ref-walk (which will always
1714 * be able to complete).
1716 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1717 if (unlikely(retval
))
1719 retval
= path_walk_rcu(name
, nd
);
1720 path_finish_rcu(nd
);
1722 path_put(&nd
->root
);
1723 nd
->root
.mnt
= NULL
;
1726 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1727 /* slower, locked walk */
1728 if (retval
== -ESTALE
)
1729 flags
|= LOOKUP_REVAL
;
1730 retval
= path_init(dfd
, name
, flags
, nd
);
1731 if (unlikely(retval
))
1733 retval
= path_walk(name
, nd
);
1735 path_put(&nd
->root
);
1736 nd
->root
.mnt
= NULL
;
1740 if (likely(!retval
)) {
1741 if (unlikely(!audit_dummy_context())) {
1742 if (nd
->path
.dentry
&& nd
->inode
)
1743 audit_inode(name
, nd
->path
.dentry
);
1750 int path_lookup(const char *name
, unsigned int flags
,
1751 struct nameidata
*nd
)
1753 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1756 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1758 struct nameidata nd
;
1759 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1766 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1767 * @dentry: pointer to dentry of the base directory
1768 * @mnt: pointer to vfs mount of the base directory
1769 * @name: pointer to file name
1770 * @flags: lookup flags
1771 * @nd: pointer to nameidata
1773 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1774 const char *name
, unsigned int flags
,
1775 struct nameidata
*nd
)
1779 /* same as do_path_lookup */
1780 nd
->last_type
= LAST_ROOT
;
1784 nd
->path
.dentry
= dentry
;
1786 path_get(&nd
->path
);
1787 nd
->root
= nd
->path
;
1788 path_get(&nd
->root
);
1789 nd
->inode
= nd
->path
.dentry
->d_inode
;
1791 retval
= path_walk(name
, nd
);
1792 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1794 audit_inode(name
, nd
->path
.dentry
);
1796 path_put(&nd
->root
);
1797 nd
->root
.mnt
= NULL
;
1802 static struct dentry
*__lookup_hash(struct qstr
*name
,
1803 struct dentry
*base
, struct nameidata
*nd
)
1805 struct inode
*inode
= base
->d_inode
;
1806 struct dentry
*dentry
;
1809 err
= exec_permission(inode
, 0);
1811 return ERR_PTR(err
);
1814 * See if the low-level filesystem might want
1815 * to use its own hash..
1817 if (base
->d_flags
& DCACHE_OP_HASH
) {
1818 err
= base
->d_op
->d_hash(base
, inode
, name
);
1819 dentry
= ERR_PTR(err
);
1825 * Don't bother with __d_lookup: callers are for creat as
1826 * well as unlink, so a lot of the time it would cost
1829 dentry
= d_lookup(base
, name
);
1831 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1832 dentry
= do_revalidate(dentry
, nd
);
1835 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1841 * Restricted form of lookup. Doesn't follow links, single-component only,
1842 * needs parent already locked. Doesn't follow mounts.
1845 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1847 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1850 static int __lookup_one_len(const char *name
, struct qstr
*this,
1851 struct dentry
*base
, int len
)
1861 hash
= init_name_hash();
1863 c
= *(const unsigned char *)name
++;
1864 if (c
== '/' || c
== '\0')
1866 hash
= partial_name_hash(c
, hash
);
1868 this->hash
= end_name_hash(hash
);
1873 * lookup_one_len - filesystem helper to lookup single pathname component
1874 * @name: pathname component to lookup
1875 * @base: base directory to lookup from
1876 * @len: maximum length @len should be interpreted to
1878 * Note that this routine is purely a helper for filesystem usage and should
1879 * not be called by generic code. Also note that by using this function the
1880 * nameidata argument is passed to the filesystem methods and a filesystem
1881 * using this helper needs to be prepared for that.
1883 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1888 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1890 err
= __lookup_one_len(name
, &this, base
, len
);
1892 return ERR_PTR(err
);
1894 return __lookup_hash(&this, base
, NULL
);
1897 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1900 struct nameidata nd
;
1901 char *tmp
= getname(name
);
1902 int err
= PTR_ERR(tmp
);
1905 BUG_ON(flags
& LOOKUP_PARENT
);
1907 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1915 static int user_path_parent(int dfd
, const char __user
*path
,
1916 struct nameidata
*nd
, char **name
)
1918 char *s
= getname(path
);
1924 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1934 * It's inline, so penalty for filesystems that don't use sticky bit is
1937 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1939 uid_t fsuid
= current_fsuid();
1941 if (!(dir
->i_mode
& S_ISVTX
))
1943 if (inode
->i_uid
== fsuid
)
1945 if (dir
->i_uid
== fsuid
)
1947 return !capable(CAP_FOWNER
);
1951 * Check whether we can remove a link victim from directory dir, check
1952 * whether the type of victim is right.
1953 * 1. We can't do it if dir is read-only (done in permission())
1954 * 2. We should have write and exec permissions on dir
1955 * 3. We can't remove anything from append-only dir
1956 * 4. We can't do anything with immutable dir (done in permission())
1957 * 5. If the sticky bit on dir is set we should either
1958 * a. be owner of dir, or
1959 * b. be owner of victim, or
1960 * c. have CAP_FOWNER capability
1961 * 6. If the victim is append-only or immutable we can't do antyhing with
1962 * links pointing to it.
1963 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1964 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1965 * 9. We can't remove a root or mountpoint.
1966 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1967 * nfs_async_unlink().
1969 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1973 if (!victim
->d_inode
)
1976 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1977 audit_inode_child(victim
, dir
);
1979 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1984 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1985 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1988 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1990 if (IS_ROOT(victim
))
1992 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1994 if (IS_DEADDIR(dir
))
1996 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2001 /* Check whether we can create an object with dentry child in directory
2003 * 1. We can't do it if child already exists (open has special treatment for
2004 * this case, but since we are inlined it's OK)
2005 * 2. We can't do it if dir is read-only (done in permission())
2006 * 3. We should have write and exec permissions on dir
2007 * 4. We can't do it if dir is immutable (done in permission())
2009 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2013 if (IS_DEADDIR(dir
))
2015 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2019 * p1 and p2 should be directories on the same fs.
2021 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2026 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2030 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2032 p
= d_ancestor(p2
, p1
);
2034 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2035 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2039 p
= d_ancestor(p1
, p2
);
2041 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2042 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2046 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2047 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2051 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2053 mutex_unlock(&p1
->d_inode
->i_mutex
);
2055 mutex_unlock(&p2
->d_inode
->i_mutex
);
2056 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2060 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
2061 struct nameidata
*nd
)
2063 int error
= may_create(dir
, dentry
);
2068 if (!dir
->i_op
->create
)
2069 return -EACCES
; /* shouldn't it be ENOSYS? */
2072 error
= security_inode_create(dir
, dentry
, mode
);
2075 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
2077 fsnotify_create(dir
, dentry
);
2081 int may_open(struct path
*path
, int acc_mode
, int flag
)
2083 struct dentry
*dentry
= path
->dentry
;
2084 struct inode
*inode
= dentry
->d_inode
;
2090 switch (inode
->i_mode
& S_IFMT
) {
2094 if (acc_mode
& MAY_WRITE
)
2099 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2108 error
= inode_permission(inode
, acc_mode
);
2113 * An append-only file must be opened in append mode for writing.
2115 if (IS_APPEND(inode
)) {
2116 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2122 /* O_NOATIME can only be set by the owner or superuser */
2123 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
2127 * Ensure there are no outstanding leases on the file.
2129 return break_lease(inode
, flag
);
2132 static int handle_truncate(struct file
*filp
)
2134 struct path
*path
= &filp
->f_path
;
2135 struct inode
*inode
= path
->dentry
->d_inode
;
2136 int error
= get_write_access(inode
);
2140 * Refuse to truncate files with mandatory locks held on them.
2142 error
= locks_verify_locked(inode
);
2144 error
= security_path_truncate(path
);
2146 error
= do_truncate(path
->dentry
, 0,
2147 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2150 put_write_access(inode
);
2155 * Be careful about ever adding any more callers of this
2156 * function. Its flags must be in the namei format, not
2157 * what get passed to sys_open().
2159 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
2160 int open_flag
, int mode
)
2163 struct dentry
*dir
= nd
->path
.dentry
;
2165 if (!IS_POSIXACL(dir
->d_inode
))
2166 mode
&= ~current_umask();
2167 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
2170 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
2172 mutex_unlock(&dir
->d_inode
->i_mutex
);
2173 dput(nd
->path
.dentry
);
2174 nd
->path
.dentry
= path
->dentry
;
2178 /* Don't check for write permission, don't truncate */
2179 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
2183 * Note that while the flag value (low two bits) for sys_open means:
2188 * it is changed into
2189 * 00 - no permissions needed
2190 * 01 - read-permission
2191 * 10 - write-permission
2193 * for the internal routines (ie open_namei()/follow_link() etc)
2194 * This is more logical, and also allows the 00 "no perm needed"
2195 * to be used for symlinks (where the permissions are checked
2199 static inline int open_to_namei_flags(int flag
)
2201 if ((flag
+1) & O_ACCMODE
)
2206 static int open_will_truncate(int flag
, struct inode
*inode
)
2209 * We'll never write to the fs underlying
2212 if (special_file(inode
->i_mode
))
2214 return (flag
& O_TRUNC
);
2217 static struct file
*finish_open(struct nameidata
*nd
,
2218 int open_flag
, int acc_mode
)
2224 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2225 if (will_truncate
) {
2226 error
= mnt_want_write(nd
->path
.mnt
);
2230 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2233 mnt_drop_write(nd
->path
.mnt
);
2236 filp
= nameidata_to_filp(nd
);
2237 if (!IS_ERR(filp
)) {
2238 error
= ima_file_check(filp
, acc_mode
);
2241 filp
= ERR_PTR(error
);
2244 if (!IS_ERR(filp
)) {
2245 if (will_truncate
) {
2246 error
= handle_truncate(filp
);
2249 filp
= ERR_PTR(error
);
2254 * It is now safe to drop the mnt write
2255 * because the filp has had a write taken
2259 mnt_drop_write(nd
->path
.mnt
);
2260 path_put(&nd
->path
);
2264 path_put(&nd
->path
);
2265 return ERR_PTR(error
);
2269 * Handle O_CREAT case for do_filp_open
2271 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2272 int open_flag
, int acc_mode
,
2273 int mode
, const char *pathname
)
2275 struct dentry
*dir
= nd
->path
.dentry
;
2277 int error
= -EISDIR
;
2279 switch (nd
->last_type
) {
2282 dir
= nd
->path
.dentry
;
2284 if (need_reval_dot(dir
)) {
2285 int status
= d_revalidate(nd
->path
.dentry
, nd
);
2297 audit_inode(pathname
, dir
);
2301 /* trailing slashes? */
2302 if (nd
->last
.name
[nd
->last
.len
])
2305 mutex_lock(&dir
->d_inode
->i_mutex
);
2307 path
->dentry
= lookup_hash(nd
);
2308 path
->mnt
= nd
->path
.mnt
;
2310 error
= PTR_ERR(path
->dentry
);
2311 if (IS_ERR(path
->dentry
)) {
2312 mutex_unlock(&dir
->d_inode
->i_mutex
);
2316 if (IS_ERR(nd
->intent
.open
.file
)) {
2317 error
= PTR_ERR(nd
->intent
.open
.file
);
2318 goto exit_mutex_unlock
;
2321 /* Negative dentry, just create the file */
2322 if (!path
->dentry
->d_inode
) {
2324 * This write is needed to ensure that a
2325 * ro->rw transition does not occur between
2326 * the time when the file is created and when
2327 * a permanent write count is taken through
2328 * the 'struct file' in nameidata_to_filp().
2330 error
= mnt_want_write(nd
->path
.mnt
);
2332 goto exit_mutex_unlock
;
2333 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2335 mnt_drop_write(nd
->path
.mnt
);
2338 filp
= nameidata_to_filp(nd
);
2339 mnt_drop_write(nd
->path
.mnt
);
2340 path_put(&nd
->path
);
2341 if (!IS_ERR(filp
)) {
2342 error
= ima_file_check(filp
, acc_mode
);
2345 filp
= ERR_PTR(error
);
2352 * It already exists.
2354 mutex_unlock(&dir
->d_inode
->i_mutex
);
2355 audit_inode(pathname
, path
->dentry
);
2358 if (open_flag
& O_EXCL
)
2361 error
= follow_managed(path
, nd
->flags
);
2366 if (!path
->dentry
->d_inode
)
2369 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2372 path_to_nameidata(path
, nd
);
2373 nd
->inode
= path
->dentry
->d_inode
;
2375 if (S_ISDIR(nd
->inode
->i_mode
))
2378 filp
= finish_open(nd
, open_flag
, acc_mode
);
2382 mutex_unlock(&dir
->d_inode
->i_mutex
);
2384 path_put_conditional(path
, nd
);
2386 path_put(&nd
->path
);
2387 return ERR_PTR(error
);
2391 * Note that the low bits of the passed in "open_flag"
2392 * are not the same as in the local variable "flag". See
2393 * open_to_namei_flags() for more details.
2395 struct file
*do_filp_open(int dfd
, const char *pathname
,
2396 int open_flag
, int mode
, int acc_mode
)
2399 struct nameidata nd
;
2403 int flag
= open_to_namei_flags(open_flag
);
2406 if (!(open_flag
& O_CREAT
))
2409 /* Must never be set by userspace */
2410 open_flag
&= ~FMODE_NONOTIFY
;
2413 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2414 * check for O_DSYNC if the need any syncing at all we enforce it's
2415 * always set instead of having to deal with possibly weird behaviour
2416 * for malicious applications setting only __O_SYNC.
2418 if (open_flag
& __O_SYNC
)
2419 open_flag
|= O_DSYNC
;
2422 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2424 /* O_TRUNC implies we need access checks for write permissions */
2425 if (open_flag
& O_TRUNC
)
2426 acc_mode
|= MAY_WRITE
;
2428 /* Allow the LSM permission hook to distinguish append
2429 access from general write access. */
2430 if (open_flag
& O_APPEND
)
2431 acc_mode
|= MAY_APPEND
;
2433 flags
= LOOKUP_OPEN
;
2434 if (open_flag
& O_CREAT
) {
2435 flags
|= LOOKUP_CREATE
;
2436 if (open_flag
& O_EXCL
)
2437 flags
|= LOOKUP_EXCL
;
2439 if (open_flag
& O_DIRECTORY
)
2440 flags
|= LOOKUP_DIRECTORY
;
2441 if (!(open_flag
& O_NOFOLLOW
))
2442 flags
|= LOOKUP_FOLLOW
;
2444 filp
= get_empty_filp();
2446 return ERR_PTR(-ENFILE
);
2448 filp
->f_flags
= open_flag
;
2449 nd
.intent
.open
.file
= filp
;
2450 nd
.intent
.open
.flags
= flag
;
2451 nd
.intent
.open
.create_mode
= mode
;
2453 if (open_flag
& O_CREAT
)
2456 /* !O_CREAT, simple open */
2457 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2458 if (unlikely(error
))
2461 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2462 if (nd
.inode
->i_op
->follow_link
)
2466 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2467 if (!nd
.inode
->i_op
->lookup
)
2470 audit_inode(pathname
, nd
.path
.dentry
);
2471 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2472 release_open_intent(&nd
);
2476 /* OK, have to create the file. Find the parent. */
2477 error
= path_init_rcu(dfd
, pathname
,
2478 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2481 error
= path_walk_rcu(pathname
, &nd
);
2482 path_finish_rcu(&nd
);
2483 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2484 /* slower, locked walk */
2485 if (error
== -ESTALE
) {
2487 flags
|= LOOKUP_REVAL
;
2489 error
= path_init(dfd
, pathname
,
2490 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2494 error
= path_walk_simple(pathname
, &nd
);
2496 if (unlikely(error
))
2498 if (unlikely(!audit_dummy_context()))
2499 audit_inode(pathname
, nd
.path
.dentry
);
2502 * We have the parent and last component.
2505 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2506 while (unlikely(!filp
)) { /* trailing symlink */
2507 struct path link
= path
;
2508 struct inode
*linki
= link
.dentry
->d_inode
;
2511 if (!(nd
.flags
& LOOKUP_FOLLOW
))
2516 * This is subtle. Instead of calling do_follow_link() we do
2517 * the thing by hands. The reason is that this way we have zero
2518 * link_count and path_walk() (called from ->follow_link)
2519 * honoring LOOKUP_PARENT. After that we have the parent and
2520 * last component, i.e. we are in the same situation as after
2521 * the first path_walk(). Well, almost - if the last component
2522 * is normal we get its copy stored in nd->last.name and we will
2523 * have to putname() it when we are done. Procfs-like symlinks
2524 * just set LAST_BIND.
2526 nd
.flags
|= LOOKUP_PARENT
;
2527 error
= security_inode_follow_link(link
.dentry
, &nd
);
2530 error
= __do_follow_link(&link
, &nd
, &cookie
);
2531 if (unlikely(error
)) {
2532 if (!IS_ERR(cookie
) && linki
->i_op
->put_link
)
2533 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2534 /* nd.path had been dropped */
2538 nd
.flags
&= ~LOOKUP_PARENT
;
2539 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2540 if (linki
->i_op
->put_link
)
2541 linki
->i_op
->put_link(link
.dentry
, &nd
, cookie
);
2547 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2549 release_open_intent(&nd
);
2553 path_put_conditional(&path
, &nd
);
2557 filp
= ERR_PTR(error
);
2562 * filp_open - open file and return file pointer
2564 * @filename: path to open
2565 * @flags: open flags as per the open(2) second argument
2566 * @mode: mode for the new file if O_CREAT is set, else ignored
2568 * This is the helper to open a file from kernelspace if you really
2569 * have to. But in generally you should not do this, so please move
2570 * along, nothing to see here..
2572 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2574 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2576 EXPORT_SYMBOL(filp_open
);
2579 * lookup_create - lookup a dentry, creating it if it doesn't exist
2580 * @nd: nameidata info
2581 * @is_dir: directory flag
2583 * Simple function to lookup and return a dentry and create it
2584 * if it doesn't exist. Is SMP-safe.
2586 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2588 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2590 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2592 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2594 * Yucky last component or no last component at all?
2595 * (foo/., foo/.., /////)
2597 if (nd
->last_type
!= LAST_NORM
)
2599 nd
->flags
&= ~LOOKUP_PARENT
;
2600 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2601 nd
->intent
.open
.flags
= O_EXCL
;
2604 * Do the final lookup.
2606 dentry
= lookup_hash(nd
);
2610 if (dentry
->d_inode
)
2613 * Special case - lookup gave negative, but... we had foo/bar/
2614 * From the vfs_mknod() POV we just have a negative dentry -
2615 * all is fine. Let's be bastards - you had / on the end, you've
2616 * been asking for (non-existent) directory. -ENOENT for you.
2618 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2620 dentry
= ERR_PTR(-ENOENT
);
2625 dentry
= ERR_PTR(-EEXIST
);
2629 EXPORT_SYMBOL_GPL(lookup_create
);
2631 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2633 int error
= may_create(dir
, dentry
);
2638 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2641 if (!dir
->i_op
->mknod
)
2644 error
= devcgroup_inode_mknod(mode
, dev
);
2648 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2652 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2654 fsnotify_create(dir
, dentry
);
2658 static int may_mknod(mode_t mode
)
2660 switch (mode
& S_IFMT
) {
2666 case 0: /* zero mode translates to S_IFREG */
2675 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2680 struct dentry
*dentry
;
2681 struct nameidata nd
;
2686 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2690 dentry
= lookup_create(&nd
, 0);
2691 if (IS_ERR(dentry
)) {
2692 error
= PTR_ERR(dentry
);
2695 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2696 mode
&= ~current_umask();
2697 error
= may_mknod(mode
);
2700 error
= mnt_want_write(nd
.path
.mnt
);
2703 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2705 goto out_drop_write
;
2706 switch (mode
& S_IFMT
) {
2707 case 0: case S_IFREG
:
2708 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2710 case S_IFCHR
: case S_IFBLK
:
2711 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2712 new_decode_dev(dev
));
2714 case S_IFIFO
: case S_IFSOCK
:
2715 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2719 mnt_drop_write(nd
.path
.mnt
);
2723 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2730 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2732 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2735 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2737 int error
= may_create(dir
, dentry
);
2742 if (!dir
->i_op
->mkdir
)
2745 mode
&= (S_IRWXUGO
|S_ISVTX
);
2746 error
= security_inode_mkdir(dir
, dentry
, mode
);
2750 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2752 fsnotify_mkdir(dir
, dentry
);
2756 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2760 struct dentry
*dentry
;
2761 struct nameidata nd
;
2763 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2767 dentry
= lookup_create(&nd
, 1);
2768 error
= PTR_ERR(dentry
);
2772 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2773 mode
&= ~current_umask();
2774 error
= mnt_want_write(nd
.path
.mnt
);
2777 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2779 goto out_drop_write
;
2780 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2782 mnt_drop_write(nd
.path
.mnt
);
2786 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2793 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2795 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2799 * We try to drop the dentry early: we should have
2800 * a usage count of 2 if we're the only user of this
2801 * dentry, and if that is true (possibly after pruning
2802 * the dcache), then we drop the dentry now.
2804 * A low-level filesystem can, if it choses, legally
2807 * if (!d_unhashed(dentry))
2810 * if it cannot handle the case of removing a directory
2811 * that is still in use by something else..
2813 void dentry_unhash(struct dentry
*dentry
)
2816 shrink_dcache_parent(dentry
);
2817 spin_lock(&dentry
->d_lock
);
2818 if (dentry
->d_count
== 2)
2820 spin_unlock(&dentry
->d_lock
);
2823 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2825 int error
= may_delete(dir
, dentry
, 1);
2830 if (!dir
->i_op
->rmdir
)
2833 mutex_lock(&dentry
->d_inode
->i_mutex
);
2834 dentry_unhash(dentry
);
2835 if (d_mountpoint(dentry
))
2838 error
= security_inode_rmdir(dir
, dentry
);
2840 error
= dir
->i_op
->rmdir(dir
, dentry
);
2842 dentry
->d_inode
->i_flags
|= S_DEAD
;
2847 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2856 static long do_rmdir(int dfd
, const char __user
*pathname
)
2860 struct dentry
*dentry
;
2861 struct nameidata nd
;
2863 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2867 switch(nd
.last_type
) {
2879 nd
.flags
&= ~LOOKUP_PARENT
;
2881 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2882 dentry
= lookup_hash(&nd
);
2883 error
= PTR_ERR(dentry
);
2886 error
= mnt_want_write(nd
.path
.mnt
);
2889 error
= security_path_rmdir(&nd
.path
, dentry
);
2892 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2894 mnt_drop_write(nd
.path
.mnt
);
2898 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2905 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2907 return do_rmdir(AT_FDCWD
, pathname
);
2910 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2912 int error
= may_delete(dir
, dentry
, 0);
2917 if (!dir
->i_op
->unlink
)
2920 mutex_lock(&dentry
->d_inode
->i_mutex
);
2921 if (d_mountpoint(dentry
))
2924 error
= security_inode_unlink(dir
, dentry
);
2926 error
= dir
->i_op
->unlink(dir
, dentry
);
2931 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2933 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2934 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2935 fsnotify_link_count(dentry
->d_inode
);
2943 * Make sure that the actual truncation of the file will occur outside its
2944 * directory's i_mutex. Truncate can take a long time if there is a lot of
2945 * writeout happening, and we don't want to prevent access to the directory
2946 * while waiting on the I/O.
2948 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2952 struct dentry
*dentry
;
2953 struct nameidata nd
;
2954 struct inode
*inode
= NULL
;
2956 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2961 if (nd
.last_type
!= LAST_NORM
)
2964 nd
.flags
&= ~LOOKUP_PARENT
;
2966 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2967 dentry
= lookup_hash(&nd
);
2968 error
= PTR_ERR(dentry
);
2969 if (!IS_ERR(dentry
)) {
2970 /* Why not before? Because we want correct error value */
2971 if (nd
.last
.name
[nd
.last
.len
])
2973 inode
= dentry
->d_inode
;
2976 error
= mnt_want_write(nd
.path
.mnt
);
2979 error
= security_path_unlink(&nd
.path
, dentry
);
2982 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2984 mnt_drop_write(nd
.path
.mnt
);
2988 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2990 iput(inode
); /* truncate the inode here */
2997 error
= !dentry
->d_inode
? -ENOENT
:
2998 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3002 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3004 if ((flag
& ~AT_REMOVEDIR
) != 0)
3007 if (flag
& AT_REMOVEDIR
)
3008 return do_rmdir(dfd
, pathname
);
3010 return do_unlinkat(dfd
, pathname
);
3013 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3015 return do_unlinkat(AT_FDCWD
, pathname
);
3018 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3020 int error
= may_create(dir
, dentry
);
3025 if (!dir
->i_op
->symlink
)
3028 error
= security_inode_symlink(dir
, dentry
, oldname
);
3032 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3034 fsnotify_create(dir
, dentry
);
3038 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3039 int, newdfd
, const char __user
*, newname
)
3044 struct dentry
*dentry
;
3045 struct nameidata nd
;
3047 from
= getname(oldname
);
3049 return PTR_ERR(from
);
3051 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3055 dentry
= lookup_create(&nd
, 0);
3056 error
= PTR_ERR(dentry
);
3060 error
= mnt_want_write(nd
.path
.mnt
);
3063 error
= security_path_symlink(&nd
.path
, dentry
, from
);
3065 goto out_drop_write
;
3066 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
3068 mnt_drop_write(nd
.path
.mnt
);
3072 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3080 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3082 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3085 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3087 struct inode
*inode
= old_dentry
->d_inode
;
3093 error
= may_create(dir
, new_dentry
);
3097 if (dir
->i_sb
!= inode
->i_sb
)
3101 * A link to an append-only or immutable file cannot be created.
3103 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3105 if (!dir
->i_op
->link
)
3107 if (S_ISDIR(inode
->i_mode
))
3110 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3114 mutex_lock(&inode
->i_mutex
);
3115 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3116 mutex_unlock(&inode
->i_mutex
);
3118 fsnotify_link(dir
, inode
, new_dentry
);
3123 * Hardlinks are often used in delicate situations. We avoid
3124 * security-related surprises by not following symlinks on the
3127 * We don't follow them on the oldname either to be compatible
3128 * with linux 2.0, and to avoid hard-linking to directories
3129 * and other special files. --ADM
3131 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3132 int, newdfd
, const char __user
*, newname
, int, flags
)
3134 struct dentry
*new_dentry
;
3135 struct nameidata nd
;
3136 struct path old_path
;
3140 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
3143 error
= user_path_at(olddfd
, oldname
,
3144 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
3149 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
3153 if (old_path
.mnt
!= nd
.path
.mnt
)
3155 new_dentry
= lookup_create(&nd
, 0);
3156 error
= PTR_ERR(new_dentry
);
3157 if (IS_ERR(new_dentry
))
3159 error
= mnt_want_write(nd
.path
.mnt
);
3162 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
3164 goto out_drop_write
;
3165 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
3167 mnt_drop_write(nd
.path
.mnt
);
3171 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3176 path_put(&old_path
);
3181 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3183 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3187 * The worst of all namespace operations - renaming directory. "Perverted"
3188 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3190 * a) we can get into loop creation. Check is done in is_subdir().
3191 * b) race potential - two innocent renames can create a loop together.
3192 * That's where 4.4 screws up. Current fix: serialization on
3193 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3195 * c) we have to lock _three_ objects - parents and victim (if it exists).
3196 * And that - after we got ->i_mutex on parents (until then we don't know
3197 * whether the target exists). Solution: try to be smart with locking
3198 * order for inodes. We rely on the fact that tree topology may change
3199 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3200 * move will be locked. Thus we can rank directories by the tree
3201 * (ancestors first) and rank all non-directories after them.
3202 * That works since everybody except rename does "lock parent, lookup,
3203 * lock child" and rename is under ->s_vfs_rename_mutex.
3204 * HOWEVER, it relies on the assumption that any object with ->lookup()
3205 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3206 * we'd better make sure that there's no link(2) for them.
3207 * d) some filesystems don't support opened-but-unlinked directories,
3208 * either because of layout or because they are not ready to deal with
3209 * all cases correctly. The latter will be fixed (taking this sort of
3210 * stuff into VFS), but the former is not going away. Solution: the same
3211 * trick as in rmdir().
3212 * e) conversion from fhandle to dentry may come in the wrong moment - when
3213 * we are removing the target. Solution: we will have to grab ->i_mutex
3214 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3215 * ->i_mutex on parents, which works but leads to some truly excessive
3218 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3219 struct inode
*new_dir
, struct dentry
*new_dentry
)
3222 struct inode
*target
;
3225 * If we are going to change the parent - check write permissions,
3226 * we'll need to flip '..'.
3228 if (new_dir
!= old_dir
) {
3229 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3234 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3238 target
= new_dentry
->d_inode
;
3240 mutex_lock(&target
->i_mutex
);
3241 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3245 dentry_unhash(new_dentry
);
3246 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3250 target
->i_flags
|= S_DEAD
;
3251 dont_mount(new_dentry
);
3253 mutex_unlock(&target
->i_mutex
);
3254 if (d_unhashed(new_dentry
))
3255 d_rehash(new_dentry
);
3259 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3260 d_move(old_dentry
,new_dentry
);
3264 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3265 struct inode
*new_dir
, struct dentry
*new_dentry
)
3267 struct inode
*target
;
3270 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3275 target
= new_dentry
->d_inode
;
3277 mutex_lock(&target
->i_mutex
);
3278 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3281 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3284 dont_mount(new_dentry
);
3285 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3286 d_move(old_dentry
, new_dentry
);
3289 mutex_unlock(&target
->i_mutex
);
3294 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3295 struct inode
*new_dir
, struct dentry
*new_dentry
)
3298 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3299 const unsigned char *old_name
;
3301 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3304 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3308 if (!new_dentry
->d_inode
)
3309 error
= may_create(new_dir
, new_dentry
);
3311 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3315 if (!old_dir
->i_op
->rename
)
3318 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3321 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3323 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3325 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3326 new_dentry
->d_inode
, old_dentry
);
3327 fsnotify_oldname_free(old_name
);
3332 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3333 int, newdfd
, const char __user
*, newname
)
3335 struct dentry
*old_dir
, *new_dir
;
3336 struct dentry
*old_dentry
, *new_dentry
;
3337 struct dentry
*trap
;
3338 struct nameidata oldnd
, newnd
;
3343 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3347 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3352 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3355 old_dir
= oldnd
.path
.dentry
;
3357 if (oldnd
.last_type
!= LAST_NORM
)
3360 new_dir
= newnd
.path
.dentry
;
3361 if (newnd
.last_type
!= LAST_NORM
)
3364 oldnd
.flags
&= ~LOOKUP_PARENT
;
3365 newnd
.flags
&= ~LOOKUP_PARENT
;
3366 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3368 trap
= lock_rename(new_dir
, old_dir
);
3370 old_dentry
= lookup_hash(&oldnd
);
3371 error
= PTR_ERR(old_dentry
);
3372 if (IS_ERR(old_dentry
))
3374 /* source must exist */
3376 if (!old_dentry
->d_inode
)
3378 /* unless the source is a directory trailing slashes give -ENOTDIR */
3379 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3381 if (oldnd
.last
.name
[oldnd
.last
.len
])
3383 if (newnd
.last
.name
[newnd
.last
.len
])
3386 /* source should not be ancestor of target */
3388 if (old_dentry
== trap
)
3390 new_dentry
= lookup_hash(&newnd
);
3391 error
= PTR_ERR(new_dentry
);
3392 if (IS_ERR(new_dentry
))
3394 /* target should not be an ancestor of source */
3396 if (new_dentry
== trap
)
3399 error
= mnt_want_write(oldnd
.path
.mnt
);
3402 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3403 &newnd
.path
, new_dentry
);
3406 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3407 new_dir
->d_inode
, new_dentry
);
3409 mnt_drop_write(oldnd
.path
.mnt
);
3415 unlock_rename(new_dir
, old_dir
);
3417 path_put(&newnd
.path
);
3420 path_put(&oldnd
.path
);
3426 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3428 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3431 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3435 len
= PTR_ERR(link
);
3440 if (len
> (unsigned) buflen
)
3442 if (copy_to_user(buffer
, link
, len
))
3449 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3450 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3451 * using) it for any given inode is up to filesystem.
3453 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3455 struct nameidata nd
;
3460 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3462 return PTR_ERR(cookie
);
3464 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3465 if (dentry
->d_inode
->i_op
->put_link
)
3466 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3470 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3472 return __vfs_follow_link(nd
, link
);
3475 /* get the link contents into pagecache */
3476 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3480 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3481 page
= read_mapping_page(mapping
, 0, NULL
);
3486 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3490 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3492 struct page
*page
= NULL
;
3493 char *s
= page_getlink(dentry
, &page
);
3494 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3497 page_cache_release(page
);
3502 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3504 struct page
*page
= NULL
;
3505 nd_set_link(nd
, page_getlink(dentry
, &page
));
3509 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3511 struct page
*page
= cookie
;
3515 page_cache_release(page
);
3520 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3522 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3524 struct address_space
*mapping
= inode
->i_mapping
;
3529 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3531 flags
|= AOP_FLAG_NOFS
;
3534 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3535 flags
, &page
, &fsdata
);
3539 kaddr
= kmap_atomic(page
, KM_USER0
);
3540 memcpy(kaddr
, symname
, len
-1);
3541 kunmap_atomic(kaddr
, KM_USER0
);
3543 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3550 mark_inode_dirty(inode
);
3556 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3558 return __page_symlink(inode
, symname
, len
,
3559 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3562 const struct inode_operations page_symlink_inode_operations
= {
3563 .readlink
= generic_readlink
,
3564 .follow_link
= page_follow_link_light
,
3565 .put_link
= page_put_link
,
3568 EXPORT_SYMBOL(user_path_at
);
3569 EXPORT_SYMBOL(follow_down_one
);
3570 EXPORT_SYMBOL(follow_down
);
3571 EXPORT_SYMBOL(follow_up
);
3572 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3573 EXPORT_SYMBOL(getname
);
3574 EXPORT_SYMBOL(lock_rename
);
3575 EXPORT_SYMBOL(lookup_one_len
);
3576 EXPORT_SYMBOL(page_follow_link_light
);
3577 EXPORT_SYMBOL(page_put_link
);
3578 EXPORT_SYMBOL(page_readlink
);
3579 EXPORT_SYMBOL(__page_symlink
);
3580 EXPORT_SYMBOL(page_symlink
);
3581 EXPORT_SYMBOL(page_symlink_inode_operations
);
3582 EXPORT_SYMBOL(path_lookup
);
3583 EXPORT_SYMBOL(kern_path
);
3584 EXPORT_SYMBOL(vfs_path_lookup
);
3585 EXPORT_SYMBOL(inode_permission
);
3586 EXPORT_SYMBOL(file_permission
);
3587 EXPORT_SYMBOL(unlock_rename
);
3588 EXPORT_SYMBOL(vfs_create
);
3589 EXPORT_SYMBOL(vfs_follow_link
);
3590 EXPORT_SYMBOL(vfs_link
);
3591 EXPORT_SYMBOL(vfs_mkdir
);
3592 EXPORT_SYMBOL(vfs_mknod
);
3593 EXPORT_SYMBOL(generic_permission
);
3594 EXPORT_SYMBOL(vfs_readlink
);
3595 EXPORT_SYMBOL(vfs_rename
);
3596 EXPORT_SYMBOL(vfs_rmdir
);
3597 EXPORT_SYMBOL(vfs_symlink
);
3598 EXPORT_SYMBOL(vfs_unlink
);
3599 EXPORT_SYMBOL(dentry_unhash
);
3600 EXPORT_SYMBOL(generic_readlink
);