2 * fs/kernfs/dir.c - kernfs directory implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
8 * This file is released under the GPLv2.
11 #include <linux/sched.h>
13 #include <linux/namei.h>
14 #include <linux/idr.h>
15 #include <linux/slab.h>
16 #include <linux/security.h>
17 #include <linux/hash.h>
19 #include "kernfs-internal.h"
21 DEFINE_MUTEX(kernfs_mutex
);
22 static DEFINE_SPINLOCK(kernfs_rename_lock
); /* kn->parent and ->name */
23 static char kernfs_pr_cont_buf
[PATH_MAX
]; /* protected by rename_lock */
25 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
27 static bool kernfs_active(struct kernfs_node
*kn
)
29 lockdep_assert_held(&kernfs_mutex
);
30 return atomic_read(&kn
->active
) >= 0;
33 static bool kernfs_lockdep(struct kernfs_node
*kn
)
35 #ifdef CONFIG_DEBUG_LOCK_ALLOC
36 return kn
->flags
& KERNFS_LOCKDEP
;
42 static int kernfs_name_locked(struct kernfs_node
*kn
, char *buf
, size_t buflen
)
44 return strlcpy(buf
, kn
->parent
? kn
->name
: "/", buflen
);
47 static char * __must_check
kernfs_path_locked(struct kernfs_node
*kn
, char *buf
,
50 char *p
= buf
+ buflen
;
56 len
= strlen(kn
->name
);
57 if (p
- buf
< len
+ 1) {
63 memcpy(p
, kn
->name
, len
);
66 } while (kn
&& kn
->parent
);
72 * kernfs_name - obtain the name of a given node
73 * @kn: kernfs_node of interest
74 * @buf: buffer to copy @kn's name into
75 * @buflen: size of @buf
77 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
78 * similar to strlcpy(). It returns the length of @kn's name and if @buf
79 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
81 * This function can be called from any context.
83 int kernfs_name(struct kernfs_node
*kn
, char *buf
, size_t buflen
)
88 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
89 ret
= kernfs_name_locked(kn
, buf
, buflen
);
90 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
95 * kernfs_path_len - determine the length of the full path of a given node
96 * @kn: kernfs_node of interest
98 * The returned length doesn't include the space for the terminating '\0'.
100 size_t kernfs_path_len(struct kernfs_node
*kn
)
105 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
108 len
+= strlen(kn
->name
) + 1;
110 } while (kn
&& kn
->parent
);
112 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
118 * kernfs_path - build full path of a given node
119 * @kn: kernfs_node of interest
120 * @buf: buffer to copy @kn's name into
121 * @buflen: size of @buf
123 * Builds and returns the full path of @kn in @buf of @buflen bytes. The
124 * path is built from the end of @buf so the returned pointer usually
125 * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated
126 * and %NULL is returned.
128 char *kernfs_path(struct kernfs_node
*kn
, char *buf
, size_t buflen
)
133 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
134 p
= kernfs_path_locked(kn
, buf
, buflen
);
135 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
138 EXPORT_SYMBOL_GPL(kernfs_path
);
141 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
142 * @kn: kernfs_node of interest
144 * This function can be called from any context.
146 void pr_cont_kernfs_name(struct kernfs_node
*kn
)
150 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
152 kernfs_name_locked(kn
, kernfs_pr_cont_buf
, sizeof(kernfs_pr_cont_buf
));
153 pr_cont("%s", kernfs_pr_cont_buf
);
155 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
159 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
160 * @kn: kernfs_node of interest
162 * This function can be called from any context.
164 void pr_cont_kernfs_path(struct kernfs_node
*kn
)
169 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
171 p
= kernfs_path_locked(kn
, kernfs_pr_cont_buf
,
172 sizeof(kernfs_pr_cont_buf
));
176 pr_cont("<name too long>");
178 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
182 * kernfs_get_parent - determine the parent node and pin it
183 * @kn: kernfs_node of interest
185 * Determines @kn's parent, pins and returns it. This function can be
186 * called from any context.
188 struct kernfs_node
*kernfs_get_parent(struct kernfs_node
*kn
)
190 struct kernfs_node
*parent
;
193 spin_lock_irqsave(&kernfs_rename_lock
, flags
);
196 spin_unlock_irqrestore(&kernfs_rename_lock
, flags
);
203 * @name: Null terminated string to hash
204 * @ns: Namespace tag to hash
206 * Returns 31 bit hash of ns + name (so it fits in an off_t )
208 static unsigned int kernfs_name_hash(const char *name
, const void *ns
)
210 unsigned long hash
= init_name_hash();
211 unsigned int len
= strlen(name
);
213 hash
= partial_name_hash(*name
++, hash
);
214 hash
= (end_name_hash(hash
) ^ hash_ptr((void *)ns
, 31));
216 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
224 static int kernfs_name_compare(unsigned int hash
, const char *name
,
225 const void *ns
, const struct kernfs_node
*kn
)
235 return strcmp(name
, kn
->name
);
238 static int kernfs_sd_compare(const struct kernfs_node
*left
,
239 const struct kernfs_node
*right
)
241 return kernfs_name_compare(left
->hash
, left
->name
, left
->ns
, right
);
245 * kernfs_link_sibling - link kernfs_node into sibling rbtree
246 * @kn: kernfs_node of interest
248 * Link @kn into its sibling rbtree which starts from
249 * @kn->parent->dir.children.
252 * mutex_lock(kernfs_mutex)
255 * 0 on susccess -EEXIST on failure.
257 static int kernfs_link_sibling(struct kernfs_node
*kn
)
259 struct rb_node
**node
= &kn
->parent
->dir
.children
.rb_node
;
260 struct rb_node
*parent
= NULL
;
263 struct kernfs_node
*pos
;
266 pos
= rb_to_kn(*node
);
268 result
= kernfs_sd_compare(kn
, pos
);
270 node
= &pos
->rb
.rb_left
;
272 node
= &pos
->rb
.rb_right
;
277 /* add new node and rebalance the tree */
278 rb_link_node(&kn
->rb
, parent
, node
);
279 rb_insert_color(&kn
->rb
, &kn
->parent
->dir
.children
);
281 /* successfully added, account subdir number */
282 if (kernfs_type(kn
) == KERNFS_DIR
)
283 kn
->parent
->dir
.subdirs
++;
289 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
290 * @kn: kernfs_node of interest
292 * Try to unlink @kn from its sibling rbtree which starts from
293 * kn->parent->dir.children. Returns %true if @kn was actually
294 * removed, %false if @kn wasn't on the rbtree.
297 * mutex_lock(kernfs_mutex)
299 static bool kernfs_unlink_sibling(struct kernfs_node
*kn
)
301 if (RB_EMPTY_NODE(&kn
->rb
))
304 if (kernfs_type(kn
) == KERNFS_DIR
)
305 kn
->parent
->dir
.subdirs
--;
307 rb_erase(&kn
->rb
, &kn
->parent
->dir
.children
);
308 RB_CLEAR_NODE(&kn
->rb
);
313 * kernfs_get_active - get an active reference to kernfs_node
314 * @kn: kernfs_node to get an active reference to
316 * Get an active reference of @kn. This function is noop if @kn
320 * Pointer to @kn on success, NULL on failure.
322 struct kernfs_node
*kernfs_get_active(struct kernfs_node
*kn
)
327 if (!atomic_inc_unless_negative(&kn
->active
))
330 if (kernfs_lockdep(kn
))
331 rwsem_acquire_read(&kn
->dep_map
, 0, 1, _RET_IP_
);
336 * kernfs_put_active - put an active reference to kernfs_node
337 * @kn: kernfs_node to put an active reference to
339 * Put an active reference to @kn. This function is noop if @kn
342 void kernfs_put_active(struct kernfs_node
*kn
)
344 struct kernfs_root
*root
= kernfs_root(kn
);
350 if (kernfs_lockdep(kn
))
351 rwsem_release(&kn
->dep_map
, 1, _RET_IP_
);
352 v
= atomic_dec_return(&kn
->active
);
353 if (likely(v
!= KN_DEACTIVATED_BIAS
))
356 wake_up_all(&root
->deactivate_waitq
);
360 * kernfs_drain - drain kernfs_node
361 * @kn: kernfs_node to drain
363 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
364 * removers may invoke this function concurrently on @kn and all will
365 * return after draining is complete.
367 static void kernfs_drain(struct kernfs_node
*kn
)
368 __releases(&kernfs_mutex
) __acquires(&kernfs_mutex
)
370 struct kernfs_root
*root
= kernfs_root(kn
);
372 lockdep_assert_held(&kernfs_mutex
);
373 WARN_ON_ONCE(kernfs_active(kn
));
375 mutex_unlock(&kernfs_mutex
);
377 if (kernfs_lockdep(kn
)) {
378 rwsem_acquire(&kn
->dep_map
, 0, 0, _RET_IP_
);
379 if (atomic_read(&kn
->active
) != KN_DEACTIVATED_BIAS
)
380 lock_contended(&kn
->dep_map
, _RET_IP_
);
383 /* but everyone should wait for draining */
384 wait_event(root
->deactivate_waitq
,
385 atomic_read(&kn
->active
) == KN_DEACTIVATED_BIAS
);
387 if (kernfs_lockdep(kn
)) {
388 lock_acquired(&kn
->dep_map
, _RET_IP_
);
389 rwsem_release(&kn
->dep_map
, 1, _RET_IP_
);
392 kernfs_unmap_bin_file(kn
);
394 mutex_lock(&kernfs_mutex
);
398 * kernfs_get - get a reference count on a kernfs_node
399 * @kn: the target kernfs_node
401 void kernfs_get(struct kernfs_node
*kn
)
404 WARN_ON(!atomic_read(&kn
->count
));
405 atomic_inc(&kn
->count
);
408 EXPORT_SYMBOL_GPL(kernfs_get
);
411 * kernfs_put - put a reference count on a kernfs_node
412 * @kn: the target kernfs_node
414 * Put a reference count of @kn and destroy it if it reached zero.
416 void kernfs_put(struct kernfs_node
*kn
)
418 struct kernfs_node
*parent
;
419 struct kernfs_root
*root
;
421 if (!kn
|| !atomic_dec_and_test(&kn
->count
))
423 root
= kernfs_root(kn
);
426 * Moving/renaming is always done while holding reference.
427 * kn->parent won't change beneath us.
431 WARN_ONCE(atomic_read(&kn
->active
) != KN_DEACTIVATED_BIAS
,
432 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
433 parent
? parent
->name
: "", kn
->name
, atomic_read(&kn
->active
));
435 if (kernfs_type(kn
) == KERNFS_LINK
)
436 kernfs_put(kn
->symlink
.target_kn
);
438 kfree_const(kn
->name
);
441 if (kn
->iattr
->ia_secdata
)
442 security_release_secctx(kn
->iattr
->ia_secdata
,
443 kn
->iattr
->ia_secdata_len
);
444 simple_xattrs_free(&kn
->iattr
->xattrs
);
447 ida_simple_remove(&root
->ino_ida
, kn
->ino
);
448 kmem_cache_free(kernfs_node_cache
, kn
);
452 if (atomic_dec_and_test(&kn
->count
))
455 /* just released the root kn, free @root too */
456 ida_destroy(&root
->ino_ida
);
460 EXPORT_SYMBOL_GPL(kernfs_put
);
462 static int kernfs_dop_revalidate(struct dentry
*dentry
, unsigned int flags
)
464 struct kernfs_node
*kn
;
466 if (flags
& LOOKUP_RCU
)
469 /* Always perform fresh lookup for negatives */
470 if (d_really_is_negative(dentry
))
471 goto out_bad_unlocked
;
473 kn
= dentry
->d_fsdata
;
474 mutex_lock(&kernfs_mutex
);
476 /* The kernfs node has been deactivated */
477 if (!kernfs_active(kn
))
480 /* The kernfs node has been moved? */
481 if (dentry
->d_parent
->d_fsdata
!= kn
->parent
)
484 /* The kernfs node has been renamed */
485 if (strcmp(dentry
->d_name
.name
, kn
->name
) != 0)
488 /* The kernfs node has been moved to a different namespace */
489 if (kn
->parent
&& kernfs_ns_enabled(kn
->parent
) &&
490 kernfs_info(dentry
->d_sb
)->ns
!= kn
->ns
)
493 mutex_unlock(&kernfs_mutex
);
496 mutex_unlock(&kernfs_mutex
);
501 static void kernfs_dop_release(struct dentry
*dentry
)
503 kernfs_put(dentry
->d_fsdata
);
506 const struct dentry_operations kernfs_dops
= {
507 .d_revalidate
= kernfs_dop_revalidate
,
508 .d_release
= kernfs_dop_release
,
512 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
513 * @dentry: the dentry in question
515 * Return the kernfs_node associated with @dentry. If @dentry is not a
516 * kernfs one, %NULL is returned.
518 * While the returned kernfs_node will stay accessible as long as @dentry
519 * is accessible, the returned node can be in any state and the caller is
520 * fully responsible for determining what's accessible.
522 struct kernfs_node
*kernfs_node_from_dentry(struct dentry
*dentry
)
524 if (dentry
->d_sb
->s_op
== &kernfs_sops
)
525 return dentry
->d_fsdata
;
529 static struct kernfs_node
*__kernfs_new_node(struct kernfs_root
*root
,
530 const char *name
, umode_t mode
,
533 struct kernfs_node
*kn
;
536 name
= kstrdup_const(name
, GFP_KERNEL
);
540 kn
= kmem_cache_zalloc(kernfs_node_cache
, GFP_KERNEL
);
544 ret
= ida_simple_get(&root
->ino_ida
, 1, 0, GFP_KERNEL
);
549 atomic_set(&kn
->count
, 1);
550 atomic_set(&kn
->active
, KN_DEACTIVATED_BIAS
);
551 RB_CLEAR_NODE(&kn
->rb
);
560 kmem_cache_free(kernfs_node_cache
, kn
);
566 struct kernfs_node
*kernfs_new_node(struct kernfs_node
*parent
,
567 const char *name
, umode_t mode
,
570 struct kernfs_node
*kn
;
572 kn
= __kernfs_new_node(kernfs_root(parent
), name
, mode
, flags
);
581 * kernfs_add_one - add kernfs_node to parent without warning
582 * @kn: kernfs_node to be added
584 * The caller must already have initialized @kn->parent. This
585 * function increments nlink of the parent's inode if @kn is a
586 * directory and link into the children list of the parent.
589 * 0 on success, -EEXIST if entry with the given name already
592 int kernfs_add_one(struct kernfs_node
*kn
)
594 struct kernfs_node
*parent
= kn
->parent
;
595 struct kernfs_iattrs
*ps_iattr
;
599 mutex_lock(&kernfs_mutex
);
602 has_ns
= kernfs_ns_enabled(parent
);
603 if (WARN(has_ns
!= (bool)kn
->ns
, KERN_WARNING
"kernfs: ns %s in '%s' for '%s'\n",
604 has_ns
? "required" : "invalid", parent
->name
, kn
->name
))
607 if (kernfs_type(parent
) != KERNFS_DIR
)
611 if (parent
->flags
& KERNFS_EMPTY_DIR
)
614 if ((parent
->flags
& KERNFS_ACTIVATED
) && !kernfs_active(parent
))
617 kn
->hash
= kernfs_name_hash(kn
->name
, kn
->ns
);
619 ret
= kernfs_link_sibling(kn
);
623 /* Update timestamps on the parent */
624 ps_iattr
= parent
->iattr
;
626 struct iattr
*ps_iattrs
= &ps_iattr
->ia_iattr
;
627 ps_iattrs
->ia_ctime
= ps_iattrs
->ia_mtime
= CURRENT_TIME
;
630 mutex_unlock(&kernfs_mutex
);
633 * Activate the new node unless CREATE_DEACTIVATED is requested.
634 * If not activated here, the kernfs user is responsible for
635 * activating the node with kernfs_activate(). A node which hasn't
636 * been activated is not visible to userland and its removal won't
637 * trigger deactivation.
639 if (!(kernfs_root(kn
)->flags
& KERNFS_ROOT_CREATE_DEACTIVATED
))
644 mutex_unlock(&kernfs_mutex
);
649 * kernfs_find_ns - find kernfs_node with the given name
650 * @parent: kernfs_node to search under
651 * @name: name to look for
652 * @ns: the namespace tag to use
654 * Look for kernfs_node with name @name under @parent. Returns pointer to
655 * the found kernfs_node on success, %NULL on failure.
657 static struct kernfs_node
*kernfs_find_ns(struct kernfs_node
*parent
,
658 const unsigned char *name
,
661 struct rb_node
*node
= parent
->dir
.children
.rb_node
;
662 bool has_ns
= kernfs_ns_enabled(parent
);
665 lockdep_assert_held(&kernfs_mutex
);
667 if (has_ns
!= (bool)ns
) {
668 WARN(1, KERN_WARNING
"kernfs: ns %s in '%s' for '%s'\n",
669 has_ns
? "required" : "invalid", parent
->name
, name
);
673 hash
= kernfs_name_hash(name
, ns
);
675 struct kernfs_node
*kn
;
679 result
= kernfs_name_compare(hash
, name
, ns
, kn
);
681 node
= node
->rb_left
;
683 node
= node
->rb_right
;
690 static struct kernfs_node
*kernfs_walk_ns(struct kernfs_node
*parent
,
691 const unsigned char *path
,
697 lockdep_assert_held(&kernfs_mutex
);
699 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
700 spin_lock_irq(&kernfs_rename_lock
);
702 len
= strlcpy(kernfs_pr_cont_buf
, path
, sizeof(kernfs_pr_cont_buf
));
704 if (len
>= sizeof(kernfs_pr_cont_buf
)) {
705 spin_unlock_irq(&kernfs_rename_lock
);
709 p
= kernfs_pr_cont_buf
;
711 while ((name
= strsep(&p
, "/")) && parent
) {
714 parent
= kernfs_find_ns(parent
, name
, ns
);
717 spin_unlock_irq(&kernfs_rename_lock
);
723 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
724 * @parent: kernfs_node to search under
725 * @name: name to look for
726 * @ns: the namespace tag to use
728 * Look for kernfs_node with name @name under @parent and get a reference
729 * if found. This function may sleep and returns pointer to the found
730 * kernfs_node on success, %NULL on failure.
732 struct kernfs_node
*kernfs_find_and_get_ns(struct kernfs_node
*parent
,
733 const char *name
, const void *ns
)
735 struct kernfs_node
*kn
;
737 mutex_lock(&kernfs_mutex
);
738 kn
= kernfs_find_ns(parent
, name
, ns
);
740 mutex_unlock(&kernfs_mutex
);
744 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns
);
747 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
748 * @parent: kernfs_node to search under
749 * @path: path to look for
750 * @ns: the namespace tag to use
752 * Look for kernfs_node with path @path under @parent and get a reference
753 * if found. This function may sleep and returns pointer to the found
754 * kernfs_node on success, %NULL on failure.
756 struct kernfs_node
*kernfs_walk_and_get_ns(struct kernfs_node
*parent
,
757 const char *path
, const void *ns
)
759 struct kernfs_node
*kn
;
761 mutex_lock(&kernfs_mutex
);
762 kn
= kernfs_walk_ns(parent
, path
, ns
);
764 mutex_unlock(&kernfs_mutex
);
770 * kernfs_create_root - create a new kernfs hierarchy
771 * @scops: optional syscall operations for the hierarchy
772 * @flags: KERNFS_ROOT_* flags
773 * @priv: opaque data associated with the new directory
775 * Returns the root of the new hierarchy on success, ERR_PTR() value on
778 struct kernfs_root
*kernfs_create_root(struct kernfs_syscall_ops
*scops
,
779 unsigned int flags
, void *priv
)
781 struct kernfs_root
*root
;
782 struct kernfs_node
*kn
;
784 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
786 return ERR_PTR(-ENOMEM
);
788 ida_init(&root
->ino_ida
);
789 INIT_LIST_HEAD(&root
->supers
);
791 kn
= __kernfs_new_node(root
, "", S_IFDIR
| S_IRUGO
| S_IXUGO
,
794 ida_destroy(&root
->ino_ida
);
796 return ERR_PTR(-ENOMEM
);
802 root
->syscall_ops
= scops
;
805 init_waitqueue_head(&root
->deactivate_waitq
);
807 if (!(root
->flags
& KERNFS_ROOT_CREATE_DEACTIVATED
))
814 * kernfs_destroy_root - destroy a kernfs hierarchy
815 * @root: root of the hierarchy to destroy
817 * Destroy the hierarchy anchored at @root by removing all existing
818 * directories and destroying @root.
820 void kernfs_destroy_root(struct kernfs_root
*root
)
822 kernfs_remove(root
->kn
); /* will also free @root */
826 * kernfs_create_dir_ns - create a directory
827 * @parent: parent in which to create a new directory
828 * @name: name of the new directory
829 * @mode: mode of the new directory
830 * @priv: opaque data associated with the new directory
831 * @ns: optional namespace tag of the directory
833 * Returns the created node on success, ERR_PTR() value on failure.
835 struct kernfs_node
*kernfs_create_dir_ns(struct kernfs_node
*parent
,
836 const char *name
, umode_t mode
,
837 void *priv
, const void *ns
)
839 struct kernfs_node
*kn
;
843 kn
= kernfs_new_node(parent
, name
, mode
| S_IFDIR
, KERNFS_DIR
);
845 return ERR_PTR(-ENOMEM
);
847 kn
->dir
.root
= parent
->dir
.root
;
852 rc
= kernfs_add_one(kn
);
861 * kernfs_create_empty_dir - create an always empty directory
862 * @parent: parent in which to create a new directory
863 * @name: name of the new directory
865 * Returns the created node on success, ERR_PTR() value on failure.
867 struct kernfs_node
*kernfs_create_empty_dir(struct kernfs_node
*parent
,
870 struct kernfs_node
*kn
;
874 kn
= kernfs_new_node(parent
, name
, S_IRUGO
|S_IXUGO
|S_IFDIR
, KERNFS_DIR
);
876 return ERR_PTR(-ENOMEM
);
878 kn
->flags
|= KERNFS_EMPTY_DIR
;
879 kn
->dir
.root
= parent
->dir
.root
;
884 rc
= kernfs_add_one(kn
);
892 static struct dentry
*kernfs_iop_lookup(struct inode
*dir
,
893 struct dentry
*dentry
,
897 struct kernfs_node
*parent
= dentry
->d_parent
->d_fsdata
;
898 struct kernfs_node
*kn
;
900 const void *ns
= NULL
;
902 mutex_lock(&kernfs_mutex
);
904 if (kernfs_ns_enabled(parent
))
905 ns
= kernfs_info(dir
->i_sb
)->ns
;
907 kn
= kernfs_find_ns(parent
, dentry
->d_name
.name
, ns
);
910 if (!kn
|| !kernfs_active(kn
)) {
915 dentry
->d_fsdata
= kn
;
917 /* attach dentry and inode */
918 inode
= kernfs_get_inode(dir
->i_sb
, kn
);
920 ret
= ERR_PTR(-ENOMEM
);
924 /* instantiate and hash dentry */
925 ret
= d_splice_alias(inode
, dentry
);
927 mutex_unlock(&kernfs_mutex
);
931 static int kernfs_iop_mkdir(struct inode
*dir
, struct dentry
*dentry
,
934 struct kernfs_node
*parent
= dir
->i_private
;
935 struct kernfs_syscall_ops
*scops
= kernfs_root(parent
)->syscall_ops
;
938 if (!scops
|| !scops
->mkdir
)
941 if (!kernfs_get_active(parent
))
944 ret
= scops
->mkdir(parent
, dentry
->d_name
.name
, mode
);
946 kernfs_put_active(parent
);
950 static int kernfs_iop_rmdir(struct inode
*dir
, struct dentry
*dentry
)
952 struct kernfs_node
*kn
= dentry
->d_fsdata
;
953 struct kernfs_syscall_ops
*scops
= kernfs_root(kn
)->syscall_ops
;
956 if (!scops
|| !scops
->rmdir
)
959 if (!kernfs_get_active(kn
))
962 ret
= scops
->rmdir(kn
);
964 kernfs_put_active(kn
);
968 static int kernfs_iop_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
969 struct inode
*new_dir
, struct dentry
*new_dentry
)
971 struct kernfs_node
*kn
= old_dentry
->d_fsdata
;
972 struct kernfs_node
*new_parent
= new_dir
->i_private
;
973 struct kernfs_syscall_ops
*scops
= kernfs_root(kn
)->syscall_ops
;
976 if (!scops
|| !scops
->rename
)
979 if (!kernfs_get_active(kn
))
982 if (!kernfs_get_active(new_parent
)) {
983 kernfs_put_active(kn
);
987 ret
= scops
->rename(kn
, new_parent
, new_dentry
->d_name
.name
);
989 kernfs_put_active(new_parent
);
990 kernfs_put_active(kn
);
994 const struct inode_operations kernfs_dir_iops
= {
995 .lookup
= kernfs_iop_lookup
,
996 .permission
= kernfs_iop_permission
,
997 .setattr
= kernfs_iop_setattr
,
998 .getattr
= kernfs_iop_getattr
,
999 .setxattr
= kernfs_iop_setxattr
,
1000 .removexattr
= kernfs_iop_removexattr
,
1001 .getxattr
= kernfs_iop_getxattr
,
1002 .listxattr
= kernfs_iop_listxattr
,
1004 .mkdir
= kernfs_iop_mkdir
,
1005 .rmdir
= kernfs_iop_rmdir
,
1006 .rename
= kernfs_iop_rename
,
1009 static struct kernfs_node
*kernfs_leftmost_descendant(struct kernfs_node
*pos
)
1011 struct kernfs_node
*last
;
1014 struct rb_node
*rbn
;
1018 if (kernfs_type(pos
) != KERNFS_DIR
)
1021 rbn
= rb_first(&pos
->dir
.children
);
1025 pos
= rb_to_kn(rbn
);
1032 * kernfs_next_descendant_post - find the next descendant for post-order walk
1033 * @pos: the current position (%NULL to initiate traversal)
1034 * @root: kernfs_node whose descendants to walk
1036 * Find the next descendant to visit for post-order traversal of @root's
1037 * descendants. @root is included in the iteration and the last node to be
1040 static struct kernfs_node
*kernfs_next_descendant_post(struct kernfs_node
*pos
,
1041 struct kernfs_node
*root
)
1043 struct rb_node
*rbn
;
1045 lockdep_assert_held(&kernfs_mutex
);
1047 /* if first iteration, visit leftmost descendant which may be root */
1049 return kernfs_leftmost_descendant(root
);
1051 /* if we visited @root, we're done */
1055 /* if there's an unvisited sibling, visit its leftmost descendant */
1056 rbn
= rb_next(&pos
->rb
);
1058 return kernfs_leftmost_descendant(rb_to_kn(rbn
));
1060 /* no sibling left, visit parent */
1065 * kernfs_activate - activate a node which started deactivated
1066 * @kn: kernfs_node whose subtree is to be activated
1068 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1069 * needs to be explicitly activated. A node which hasn't been activated
1070 * isn't visible to userland and deactivation is skipped during its
1071 * removal. This is useful to construct atomic init sequences where
1072 * creation of multiple nodes should either succeed or fail atomically.
1074 * The caller is responsible for ensuring that this function is not called
1075 * after kernfs_remove*() is invoked on @kn.
1077 void kernfs_activate(struct kernfs_node
*kn
)
1079 struct kernfs_node
*pos
;
1081 mutex_lock(&kernfs_mutex
);
1084 while ((pos
= kernfs_next_descendant_post(pos
, kn
))) {
1085 if (!pos
|| (pos
->flags
& KERNFS_ACTIVATED
))
1088 WARN_ON_ONCE(pos
->parent
&& RB_EMPTY_NODE(&pos
->rb
));
1089 WARN_ON_ONCE(atomic_read(&pos
->active
) != KN_DEACTIVATED_BIAS
);
1091 atomic_sub(KN_DEACTIVATED_BIAS
, &pos
->active
);
1092 pos
->flags
|= KERNFS_ACTIVATED
;
1095 mutex_unlock(&kernfs_mutex
);
1098 static void __kernfs_remove(struct kernfs_node
*kn
)
1100 struct kernfs_node
*pos
;
1102 lockdep_assert_held(&kernfs_mutex
);
1105 * Short-circuit if non-root @kn has already finished removal.
1106 * This is for kernfs_remove_self() which plays with active ref
1109 if (!kn
|| (kn
->parent
&& RB_EMPTY_NODE(&kn
->rb
)))
1112 pr_debug("kernfs %s: removing\n", kn
->name
);
1114 /* prevent any new usage under @kn by deactivating all nodes */
1116 while ((pos
= kernfs_next_descendant_post(pos
, kn
)))
1117 if (kernfs_active(pos
))
1118 atomic_add(KN_DEACTIVATED_BIAS
, &pos
->active
);
1120 /* deactivate and unlink the subtree node-by-node */
1122 pos
= kernfs_leftmost_descendant(kn
);
1125 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1126 * base ref could have been put by someone else by the time
1127 * the function returns. Make sure it doesn't go away
1133 * Drain iff @kn was activated. This avoids draining and
1134 * its lockdep annotations for nodes which have never been
1135 * activated and allows embedding kernfs_remove() in create
1136 * error paths without worrying about draining.
1138 if (kn
->flags
& KERNFS_ACTIVATED
)
1141 WARN_ON_ONCE(atomic_read(&kn
->active
) != KN_DEACTIVATED_BIAS
);
1144 * kernfs_unlink_sibling() succeeds once per node. Use it
1145 * to decide who's responsible for cleanups.
1147 if (!pos
->parent
|| kernfs_unlink_sibling(pos
)) {
1148 struct kernfs_iattrs
*ps_iattr
=
1149 pos
->parent
? pos
->parent
->iattr
: NULL
;
1151 /* update timestamps on the parent */
1153 ps_iattr
->ia_iattr
.ia_ctime
= CURRENT_TIME
;
1154 ps_iattr
->ia_iattr
.ia_mtime
= CURRENT_TIME
;
1161 } while (pos
!= kn
);
1165 * kernfs_remove - remove a kernfs_node recursively
1166 * @kn: the kernfs_node to remove
1168 * Remove @kn along with all its subdirectories and files.
1170 void kernfs_remove(struct kernfs_node
*kn
)
1172 mutex_lock(&kernfs_mutex
);
1173 __kernfs_remove(kn
);
1174 mutex_unlock(&kernfs_mutex
);
1178 * kernfs_break_active_protection - break out of active protection
1179 * @kn: the self kernfs_node
1181 * The caller must be running off of a kernfs operation which is invoked
1182 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1183 * this function must also be matched with an invocation of
1184 * kernfs_unbreak_active_protection().
1186 * This function releases the active reference of @kn the caller is
1187 * holding. Once this function is called, @kn may be removed at any point
1188 * and the caller is solely responsible for ensuring that the objects it
1189 * dereferences are accessible.
1191 void kernfs_break_active_protection(struct kernfs_node
*kn
)
1194 * Take out ourself out of the active ref dependency chain. If
1195 * we're called without an active ref, lockdep will complain.
1197 kernfs_put_active(kn
);
1201 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1202 * @kn: the self kernfs_node
1204 * If kernfs_break_active_protection() was called, this function must be
1205 * invoked before finishing the kernfs operation. Note that while this
1206 * function restores the active reference, it doesn't and can't actually
1207 * restore the active protection - @kn may already or be in the process of
1208 * being removed. Once kernfs_break_active_protection() is invoked, that
1209 * protection is irreversibly gone for the kernfs operation instance.
1211 * While this function may be called at any point after
1212 * kernfs_break_active_protection() is invoked, its most useful location
1213 * would be right before the enclosing kernfs operation returns.
1215 void kernfs_unbreak_active_protection(struct kernfs_node
*kn
)
1218 * @kn->active could be in any state; however, the increment we do
1219 * here will be undone as soon as the enclosing kernfs operation
1220 * finishes and this temporary bump can't break anything. If @kn
1221 * is alive, nothing changes. If @kn is being deactivated, the
1222 * soon-to-follow put will either finish deactivation or restore
1223 * deactivated state. If @kn is already removed, the temporary
1224 * bump is guaranteed to be gone before @kn is released.
1226 atomic_inc(&kn
->active
);
1227 if (kernfs_lockdep(kn
))
1228 rwsem_acquire(&kn
->dep_map
, 0, 1, _RET_IP_
);
1232 * kernfs_remove_self - remove a kernfs_node from its own method
1233 * @kn: the self kernfs_node to remove
1235 * The caller must be running off of a kernfs operation which is invoked
1236 * with an active reference - e.g. one of kernfs_ops. This can be used to
1237 * implement a file operation which deletes itself.
1239 * For example, the "delete" file for a sysfs device directory can be
1240 * implemented by invoking kernfs_remove_self() on the "delete" file
1241 * itself. This function breaks the circular dependency of trying to
1242 * deactivate self while holding an active ref itself. It isn't necessary
1243 * to modify the usual removal path to use kernfs_remove_self(). The
1244 * "delete" implementation can simply invoke kernfs_remove_self() on self
1245 * before proceeding with the usual removal path. kernfs will ignore later
1246 * kernfs_remove() on self.
1248 * kernfs_remove_self() can be called multiple times concurrently on the
1249 * same kernfs_node. Only the first one actually performs removal and
1250 * returns %true. All others will wait until the kernfs operation which
1251 * won self-removal finishes and return %false. Note that the losers wait
1252 * for the completion of not only the winning kernfs_remove_self() but also
1253 * the whole kernfs_ops which won the arbitration. This can be used to
1254 * guarantee, for example, all concurrent writes to a "delete" file to
1255 * finish only after the whole operation is complete.
1257 bool kernfs_remove_self(struct kernfs_node
*kn
)
1261 mutex_lock(&kernfs_mutex
);
1262 kernfs_break_active_protection(kn
);
1265 * SUICIDAL is used to arbitrate among competing invocations. Only
1266 * the first one will actually perform removal. When the removal
1267 * is complete, SUICIDED is set and the active ref is restored
1268 * while holding kernfs_mutex. The ones which lost arbitration
1269 * waits for SUICDED && drained which can happen only after the
1270 * enclosing kernfs operation which executed the winning instance
1271 * of kernfs_remove_self() finished.
1273 if (!(kn
->flags
& KERNFS_SUICIDAL
)) {
1274 kn
->flags
|= KERNFS_SUICIDAL
;
1275 __kernfs_remove(kn
);
1276 kn
->flags
|= KERNFS_SUICIDED
;
1279 wait_queue_head_t
*waitq
= &kernfs_root(kn
)->deactivate_waitq
;
1283 prepare_to_wait(waitq
, &wait
, TASK_UNINTERRUPTIBLE
);
1285 if ((kn
->flags
& KERNFS_SUICIDED
) &&
1286 atomic_read(&kn
->active
) == KN_DEACTIVATED_BIAS
)
1289 mutex_unlock(&kernfs_mutex
);
1291 mutex_lock(&kernfs_mutex
);
1293 finish_wait(waitq
, &wait
);
1294 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn
->rb
));
1299 * This must be done while holding kernfs_mutex; otherwise, waiting
1300 * for SUICIDED && deactivated could finish prematurely.
1302 kernfs_unbreak_active_protection(kn
);
1304 mutex_unlock(&kernfs_mutex
);
1309 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1310 * @parent: parent of the target
1311 * @name: name of the kernfs_node to remove
1312 * @ns: namespace tag of the kernfs_node to remove
1314 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1315 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1317 int kernfs_remove_by_name_ns(struct kernfs_node
*parent
, const char *name
,
1320 struct kernfs_node
*kn
;
1323 WARN(1, KERN_WARNING
"kernfs: can not remove '%s', no directory\n",
1328 mutex_lock(&kernfs_mutex
);
1330 kn
= kernfs_find_ns(parent
, name
, ns
);
1332 __kernfs_remove(kn
);
1334 mutex_unlock(&kernfs_mutex
);
1343 * kernfs_rename_ns - move and rename a kernfs_node
1345 * @new_parent: new parent to put @sd under
1346 * @new_name: new name
1347 * @new_ns: new namespace tag
1349 int kernfs_rename_ns(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
1350 const char *new_name
, const void *new_ns
)
1352 struct kernfs_node
*old_parent
;
1353 const char *old_name
= NULL
;
1356 /* can't move or rename root */
1360 mutex_lock(&kernfs_mutex
);
1363 if (!kernfs_active(kn
) || !kernfs_active(new_parent
) ||
1364 (new_parent
->flags
& KERNFS_EMPTY_DIR
))
1368 if ((kn
->parent
== new_parent
) && (kn
->ns
== new_ns
) &&
1369 (strcmp(kn
->name
, new_name
) == 0))
1370 goto out
; /* nothing to rename */
1373 if (kernfs_find_ns(new_parent
, new_name
, new_ns
))
1376 /* rename kernfs_node */
1377 if (strcmp(kn
->name
, new_name
) != 0) {
1379 new_name
= kstrdup_const(new_name
, GFP_KERNEL
);
1387 * Move to the appropriate place in the appropriate directories rbtree.
1389 kernfs_unlink_sibling(kn
);
1390 kernfs_get(new_parent
);
1392 /* rename_lock protects ->parent and ->name accessors */
1393 spin_lock_irq(&kernfs_rename_lock
);
1395 old_parent
= kn
->parent
;
1396 kn
->parent
= new_parent
;
1400 old_name
= kn
->name
;
1401 kn
->name
= new_name
;
1404 spin_unlock_irq(&kernfs_rename_lock
);
1406 kn
->hash
= kernfs_name_hash(kn
->name
, kn
->ns
);
1407 kernfs_link_sibling(kn
);
1409 kernfs_put(old_parent
);
1410 kfree_const(old_name
);
1414 mutex_unlock(&kernfs_mutex
);
1418 /* Relationship between s_mode and the DT_xxx types */
1419 static inline unsigned char dt_type(struct kernfs_node
*kn
)
1421 return (kn
->mode
>> 12) & 15;
1424 static int kernfs_dir_fop_release(struct inode
*inode
, struct file
*filp
)
1426 kernfs_put(filp
->private_data
);
1430 static struct kernfs_node
*kernfs_dir_pos(const void *ns
,
1431 struct kernfs_node
*parent
, loff_t hash
, struct kernfs_node
*pos
)
1434 int valid
= kernfs_active(pos
) &&
1435 pos
->parent
== parent
&& hash
== pos
->hash
;
1440 if (!pos
&& (hash
> 1) && (hash
< INT_MAX
)) {
1441 struct rb_node
*node
= parent
->dir
.children
.rb_node
;
1443 pos
= rb_to_kn(node
);
1445 if (hash
< pos
->hash
)
1446 node
= node
->rb_left
;
1447 else if (hash
> pos
->hash
)
1448 node
= node
->rb_right
;
1453 /* Skip over entries which are dying/dead or in the wrong namespace */
1454 while (pos
&& (!kernfs_active(pos
) || pos
->ns
!= ns
)) {
1455 struct rb_node
*node
= rb_next(&pos
->rb
);
1459 pos
= rb_to_kn(node
);
1464 static struct kernfs_node
*kernfs_dir_next_pos(const void *ns
,
1465 struct kernfs_node
*parent
, ino_t ino
, struct kernfs_node
*pos
)
1467 pos
= kernfs_dir_pos(ns
, parent
, ino
, pos
);
1470 struct rb_node
*node
= rb_next(&pos
->rb
);
1474 pos
= rb_to_kn(node
);
1475 } while (pos
&& (!kernfs_active(pos
) || pos
->ns
!= ns
));
1480 static int kernfs_fop_readdir(struct file
*file
, struct dir_context
*ctx
)
1482 struct dentry
*dentry
= file
->f_path
.dentry
;
1483 struct kernfs_node
*parent
= dentry
->d_fsdata
;
1484 struct kernfs_node
*pos
= file
->private_data
;
1485 const void *ns
= NULL
;
1487 if (!dir_emit_dots(file
, ctx
))
1489 mutex_lock(&kernfs_mutex
);
1491 if (kernfs_ns_enabled(parent
))
1492 ns
= kernfs_info(dentry
->d_sb
)->ns
;
1494 for (pos
= kernfs_dir_pos(ns
, parent
, ctx
->pos
, pos
);
1496 pos
= kernfs_dir_next_pos(ns
, parent
, ctx
->pos
, pos
)) {
1497 const char *name
= pos
->name
;
1498 unsigned int type
= dt_type(pos
);
1499 int len
= strlen(name
);
1500 ino_t ino
= pos
->ino
;
1502 ctx
->pos
= pos
->hash
;
1503 file
->private_data
= pos
;
1506 mutex_unlock(&kernfs_mutex
);
1507 if (!dir_emit(ctx
, name
, len
, ino
, type
))
1509 mutex_lock(&kernfs_mutex
);
1511 mutex_unlock(&kernfs_mutex
);
1512 file
->private_data
= NULL
;
1517 static loff_t
kernfs_dir_fop_llseek(struct file
*file
, loff_t offset
,
1520 struct inode
*inode
= file_inode(file
);
1524 ret
= generic_file_llseek(file
, offset
, whence
);
1525 inode_unlock(inode
);
1530 const struct file_operations kernfs_dir_fops
= {
1531 .read
= generic_read_dir
,
1532 .iterate
= kernfs_fop_readdir
,
1533 .release
= kernfs_dir_fop_release
,
1534 .llseek
= kernfs_dir_fop_llseek
,