2 * fs/kernfs/dir.c - kernfs directory implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
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 - build full path of a given node
96 * @kn: kernfs_node of interest
97 * @buf: buffer to copy @kn's name into
98 * @buflen: size of @buf
100 * Builds and returns the full path of @kn in @buf of @buflen bytes. The
101 * path is built from the end of @buf so the returned pointer usually
102 * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated
103 * and %NULL is returned.
105 char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen)
110 spin_lock_irqsave(&kernfs_rename_lock, flags);
111 p = kernfs_path_locked(kn, buf, buflen);
112 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
115 EXPORT_SYMBOL_GPL(kernfs_path);
118 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
119 * @kn: kernfs_node of interest
121 * This function can be called from any context.
123 void pr_cont_kernfs_name(struct kernfs_node *kn)
127 spin_lock_irqsave(&kernfs_rename_lock, flags);
129 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
130 pr_cont("%s", kernfs_pr_cont_buf);
132 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
136 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
137 * @kn: kernfs_node of interest
139 * This function can be called from any context.
141 void pr_cont_kernfs_path(struct kernfs_node *kn)
146 spin_lock_irqsave(&kernfs_rename_lock, flags);
148 p = kernfs_path_locked(kn, kernfs_pr_cont_buf,
149 sizeof(kernfs_pr_cont_buf));
153 pr_cont("<name too long>");
155 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
159 * kernfs_get_parent - determine the parent node and pin it
160 * @kn: kernfs_node of interest
162 * Determines @kn's parent, pins and returns it. This function can be
163 * called from any context.
165 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
167 struct kernfs_node *parent;
170 spin_lock_irqsave(&kernfs_rename_lock, flags);
173 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
180 * @name: Null terminated string to hash
181 * @ns: Namespace tag to hash
183 * Returns 31 bit hash of ns + name (so it fits in an off_t )
185 static unsigned int kernfs_name_hash(const char *name, const void *ns)
187 unsigned long hash = init_name_hash();
188 unsigned int len = strlen(name);
190 hash = partial_name_hash(*name++, hash);
191 hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31));
193 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
201 static int kernfs_name_compare(unsigned int hash, const char *name,
202 const void *ns, const struct kernfs_node *kn)
212 return strcmp(name, kn->name);
215 static int kernfs_sd_compare(const struct kernfs_node *left,
216 const struct kernfs_node *right)
218 return kernfs_name_compare(left->hash, left->name, left->ns, right);
222 * kernfs_link_sibling - link kernfs_node into sibling rbtree
223 * @kn: kernfs_node of interest
225 * Link @kn into its sibling rbtree which starts from
226 * @kn->parent->dir.children.
229 * mutex_lock(kernfs_mutex)
232 * 0 on susccess -EEXIST on failure.
234 static int kernfs_link_sibling(struct kernfs_node *kn)
236 struct rb_node **node = &kn->parent->dir.children.rb_node;
237 struct rb_node *parent = NULL;
240 struct kernfs_node *pos;
243 pos = rb_to_kn(*node);
245 result = kernfs_sd_compare(kn, pos);
247 node = &pos->rb.rb_left;
249 node = &pos->rb.rb_right;
254 /* add new node and rebalance the tree */
255 rb_link_node(&kn->rb, parent, node);
256 rb_insert_color(&kn->rb, &kn->parent->dir.children);
258 /* successfully added, account subdir number */
259 if (kernfs_type(kn) == KERNFS_DIR)
260 kn->parent->dir.subdirs++;
266 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
267 * @kn: kernfs_node of interest
269 * Try to unlink @kn from its sibling rbtree which starts from
270 * kn->parent->dir.children. Returns %true if @kn was actually
271 * removed, %false if @kn wasn't on the rbtree.
274 * mutex_lock(kernfs_mutex)
276 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
278 if (RB_EMPTY_NODE(&kn->rb))
281 if (kernfs_type(kn) == KERNFS_DIR)
282 kn->parent->dir.subdirs--;
284 rb_erase(&kn->rb, &kn->parent->dir.children);
285 RB_CLEAR_NODE(&kn->rb);
290 * kernfs_get_active - get an active reference to kernfs_node
291 * @kn: kernfs_node to get an active reference to
293 * Get an active reference of @kn. This function is noop if @kn
297 * Pointer to @kn on success, NULL on failure.
299 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
304 if (!atomic_inc_unless_negative(&kn->active))
307 if (kernfs_lockdep(kn))
308 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
313 * kernfs_put_active - put an active reference to kernfs_node
314 * @kn: kernfs_node to put an active reference to
316 * Put an active reference to @kn. This function is noop if @kn
319 void kernfs_put_active(struct kernfs_node *kn)
321 struct kernfs_root *root = kernfs_root(kn);
327 if (kernfs_lockdep(kn))
328 rwsem_release(&kn->dep_map, 1, _RET_IP_);
329 v = atomic_dec_return(&kn->active);
330 if (likely(v != KN_DEACTIVATED_BIAS))
333 wake_up_all(&root->deactivate_waitq);
337 * kernfs_drain - drain kernfs_node
338 * @kn: kernfs_node to drain
340 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
341 * removers may invoke this function concurrently on @kn and all will
342 * return after draining is complete.
344 static void kernfs_drain(struct kernfs_node *kn)
345 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
347 struct kernfs_root *root = kernfs_root(kn);
349 lockdep_assert_held(&kernfs_mutex);
350 WARN_ON_ONCE(kernfs_active(kn));
352 mutex_unlock(&kernfs_mutex);
354 if (kernfs_lockdep(kn)) {
355 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
356 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
357 lock_contended(&kn->dep_map, _RET_IP_);
360 /* but everyone should wait for draining */
361 wait_event(root->deactivate_waitq,
362 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
364 if (kernfs_lockdep(kn)) {
365 lock_acquired(&kn->dep_map, _RET_IP_);
366 rwsem_release(&kn->dep_map, 1, _RET_IP_);
369 kernfs_unmap_bin_file(kn);
371 mutex_lock(&kernfs_mutex);
375 * kernfs_get - get a reference count on a kernfs_node
376 * @kn: the target kernfs_node
378 void kernfs_get(struct kernfs_node *kn)
381 WARN_ON(!atomic_read(&kn->count));
382 atomic_inc(&kn->count);
385 EXPORT_SYMBOL_GPL(kernfs_get);
388 * kernfs_put - put a reference count on a kernfs_node
389 * @kn: the target kernfs_node
391 * Put a reference count of @kn and destroy it if it reached zero.
393 void kernfs_put(struct kernfs_node *kn)
395 struct kernfs_node *parent;
396 struct kernfs_root *root;
398 if (!kn || !atomic_dec_and_test(&kn->count))
400 root = kernfs_root(kn);
403 * Moving/renaming is always done while holding reference.
404 * kn->parent won't change beneath us.
408 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
409 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
410 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
412 if (kernfs_type(kn) == KERNFS_LINK)
413 kernfs_put(kn->symlink.target_kn);
415 kfree_const(kn->name);
418 if (kn->iattr->ia_secdata)
419 security_release_secctx(kn->iattr->ia_secdata,
420 kn->iattr->ia_secdata_len);
421 simple_xattrs_free(&kn->iattr->xattrs);
424 ida_simple_remove(&root->ino_ida, kn->ino);
425 kmem_cache_free(kernfs_node_cache, kn);
429 if (atomic_dec_and_test(&kn->count))
432 /* just released the root kn, free @root too */
433 ida_destroy(&root->ino_ida);
437 EXPORT_SYMBOL_GPL(kernfs_put);
439 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
441 struct kernfs_node *kn;
443 if (flags & LOOKUP_RCU)
446 /* Always perform fresh lookup for negatives */
447 if (d_really_is_negative(dentry))
448 goto out_bad_unlocked;
450 kn = dentry->d_fsdata;
451 mutex_lock(&kernfs_mutex);
453 /* The kernfs node has been deactivated */
454 if (!kernfs_active(kn))
457 /* The kernfs node has been moved? */
458 if (dentry->d_parent->d_fsdata != kn->parent)
461 /* The kernfs node has been renamed */
462 if (strcmp(dentry->d_name.name, kn->name) != 0)
465 /* The kernfs node has been moved to a different namespace */
466 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
467 kernfs_info(dentry->d_sb)->ns != kn->ns)
470 mutex_unlock(&kernfs_mutex);
473 mutex_unlock(&kernfs_mutex);
478 static void kernfs_dop_release(struct dentry *dentry)
480 kernfs_put(dentry->d_fsdata);
483 const struct dentry_operations kernfs_dops = {
484 .d_revalidate = kernfs_dop_revalidate,
485 .d_release = kernfs_dop_release,
489 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
490 * @dentry: the dentry in question
492 * Return the kernfs_node associated with @dentry. If @dentry is not a
493 * kernfs one, %NULL is returned.
495 * While the returned kernfs_node will stay accessible as long as @dentry
496 * is accessible, the returned node can be in any state and the caller is
497 * fully responsible for determining what's accessible.
499 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
501 if (dentry->d_sb->s_op == &kernfs_sops)
502 return dentry->d_fsdata;
506 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
507 const char *name, umode_t mode,
510 struct kernfs_node *kn;
513 name = kstrdup_const(name, GFP_KERNEL);
517 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
522 * If the ino of the sysfs entry created for a kmem cache gets
523 * allocated from an ida layer, which is accounted to the memcg that
524 * owns the cache, the memcg will get pinned forever. So do not account
525 * ino ida allocations.
527 ret = ida_simple_get(&root->ino_ida, 1, 0,
528 GFP_KERNEL | __GFP_NOACCOUNT);
533 atomic_set(&kn->count, 1);
534 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
535 RB_CLEAR_NODE(&kn->rb);
544 kmem_cache_free(kernfs_node_cache, kn);
550 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
551 const char *name, umode_t mode,
554 struct kernfs_node *kn;
556 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
565 * kernfs_add_one - add kernfs_node to parent without warning
566 * @kn: kernfs_node to be added
568 * The caller must already have initialized @kn->parent. This
569 * function increments nlink of the parent's inode if @kn is a
570 * directory and link into the children list of the parent.
573 * 0 on success, -EEXIST if entry with the given name already
576 int kernfs_add_one(struct kernfs_node *kn)
578 struct kernfs_node *parent = kn->parent;
579 struct kernfs_iattrs *ps_iattr;
583 mutex_lock(&kernfs_mutex);
586 has_ns = kernfs_ns_enabled(parent);
587 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
588 has_ns ? "required" : "invalid", parent->name, kn->name))
591 if (kernfs_type(parent) != KERNFS_DIR)
595 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
598 kn->hash = kernfs_name_hash(kn->name, kn->ns);
600 ret = kernfs_link_sibling(kn);
604 /* Update timestamps on the parent */
605 ps_iattr = parent->iattr;
607 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
608 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
611 mutex_unlock(&kernfs_mutex);
614 * Activate the new node unless CREATE_DEACTIVATED is requested.
615 * If not activated here, the kernfs user is responsible for
616 * activating the node with kernfs_activate(). A node which hasn't
617 * been activated is not visible to userland and its removal won't
618 * trigger deactivation.
620 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
625 mutex_unlock(&kernfs_mutex);
630 * kernfs_find_ns - find kernfs_node with the given name
631 * @parent: kernfs_node to search under
632 * @name: name to look for
633 * @ns: the namespace tag to use
635 * Look for kernfs_node with name @name under @parent. Returns pointer to
636 * the found kernfs_node on success, %NULL on failure.
638 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
639 const unsigned char *name,
642 struct rb_node *node = parent->dir.children.rb_node;
643 bool has_ns = kernfs_ns_enabled(parent);
646 lockdep_assert_held(&kernfs_mutex);
648 if (has_ns != (bool)ns) {
649 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
650 has_ns ? "required" : "invalid", parent->name, name);
654 hash = kernfs_name_hash(name, ns);
656 struct kernfs_node *kn;
660 result = kernfs_name_compare(hash, name, ns, kn);
662 node = node->rb_left;
664 node = node->rb_right;
672 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
673 * @parent: kernfs_node to search under
674 * @name: name to look for
675 * @ns: the namespace tag to use
677 * Look for kernfs_node with name @name under @parent and get a reference
678 * if found. This function may sleep and returns pointer to the found
679 * kernfs_node on success, %NULL on failure.
681 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
682 const char *name, const void *ns)
684 struct kernfs_node *kn;
686 mutex_lock(&kernfs_mutex);
687 kn = kernfs_find_ns(parent, name, ns);
689 mutex_unlock(&kernfs_mutex);
693 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
696 * kernfs_create_root - create a new kernfs hierarchy
697 * @scops: optional syscall operations for the hierarchy
698 * @flags: KERNFS_ROOT_* flags
699 * @priv: opaque data associated with the new directory
701 * Returns the root of the new hierarchy on success, ERR_PTR() value on
704 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
705 unsigned int flags, void *priv)
707 struct kernfs_root *root;
708 struct kernfs_node *kn;
710 root = kzalloc(sizeof(*root), GFP_KERNEL);
712 return ERR_PTR(-ENOMEM);
714 ida_init(&root->ino_ida);
715 INIT_LIST_HEAD(&root->supers);
717 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
720 ida_destroy(&root->ino_ida);
722 return ERR_PTR(-ENOMEM);
728 root->syscall_ops = scops;
731 init_waitqueue_head(&root->deactivate_waitq);
733 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
740 * kernfs_destroy_root - destroy a kernfs hierarchy
741 * @root: root of the hierarchy to destroy
743 * Destroy the hierarchy anchored at @root by removing all existing
744 * directories and destroying @root.
746 void kernfs_destroy_root(struct kernfs_root *root)
748 kernfs_remove(root->kn); /* will also free @root */
752 * kernfs_create_dir_ns - create a directory
753 * @parent: parent in which to create a new directory
754 * @name: name of the new directory
755 * @mode: mode of the new directory
756 * @priv: opaque data associated with the new directory
757 * @ns: optional namespace tag of the directory
759 * Returns the created node on success, ERR_PTR() value on failure.
761 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
762 const char *name, umode_t mode,
763 void *priv, const void *ns)
765 struct kernfs_node *kn;
769 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
771 return ERR_PTR(-ENOMEM);
773 kn->dir.root = parent->dir.root;
778 rc = kernfs_add_one(kn);
786 static struct dentry *kernfs_iop_lookup(struct inode *dir,
787 struct dentry *dentry,
791 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
792 struct kernfs_node *kn;
794 const void *ns = NULL;
796 mutex_lock(&kernfs_mutex);
798 if (kernfs_ns_enabled(parent))
799 ns = kernfs_info(dir->i_sb)->ns;
801 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
804 if (!kn || !kernfs_active(kn)) {
809 dentry->d_fsdata = kn;
811 /* attach dentry and inode */
812 inode = kernfs_get_inode(dir->i_sb, kn);
814 ret = ERR_PTR(-ENOMEM);
818 /* instantiate and hash dentry */
819 ret = d_splice_alias(inode, dentry);
821 mutex_unlock(&kernfs_mutex);
825 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
828 struct kernfs_node *parent = dir->i_private;
829 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
832 if (!scops || !scops->mkdir)
835 if (!kernfs_get_active(parent))
838 ret = scops->mkdir(parent, dentry->d_name.name, mode);
840 kernfs_put_active(parent);
844 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
846 struct kernfs_node *kn = dentry->d_fsdata;
847 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
850 if (!scops || !scops->rmdir)
853 if (!kernfs_get_active(kn))
856 ret = scops->rmdir(kn);
858 kernfs_put_active(kn);
862 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
863 struct inode *new_dir, struct dentry *new_dentry)
865 struct kernfs_node *kn = old_dentry->d_fsdata;
866 struct kernfs_node *new_parent = new_dir->i_private;
867 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
870 if (!scops || !scops->rename)
873 if (!kernfs_get_active(kn))
876 if (!kernfs_get_active(new_parent)) {
877 kernfs_put_active(kn);
881 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
883 kernfs_put_active(new_parent);
884 kernfs_put_active(kn);
888 const struct inode_operations kernfs_dir_iops = {
889 .lookup = kernfs_iop_lookup,
890 .permission = kernfs_iop_permission,
891 .setattr = kernfs_iop_setattr,
892 .getattr = kernfs_iop_getattr,
893 .setxattr = kernfs_iop_setxattr,
894 .removexattr = kernfs_iop_removexattr,
895 .getxattr = kernfs_iop_getxattr,
896 .listxattr = kernfs_iop_listxattr,
898 .mkdir = kernfs_iop_mkdir,
899 .rmdir = kernfs_iop_rmdir,
900 .rename = kernfs_iop_rename,
903 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
905 struct kernfs_node *last;
912 if (kernfs_type(pos) != KERNFS_DIR)
915 rbn = rb_first(&pos->dir.children);
926 * kernfs_next_descendant_post - find the next descendant for post-order walk
927 * @pos: the current position (%NULL to initiate traversal)
928 * @root: kernfs_node whose descendants to walk
930 * Find the next descendant to visit for post-order traversal of @root's
931 * descendants. @root is included in the iteration and the last node to be
934 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
935 struct kernfs_node *root)
939 lockdep_assert_held(&kernfs_mutex);
941 /* if first iteration, visit leftmost descendant which may be root */
943 return kernfs_leftmost_descendant(root);
945 /* if we visited @root, we're done */
949 /* if there's an unvisited sibling, visit its leftmost descendant */
950 rbn = rb_next(&pos->rb);
952 return kernfs_leftmost_descendant(rb_to_kn(rbn));
954 /* no sibling left, visit parent */
959 * kernfs_activate - activate a node which started deactivated
960 * @kn: kernfs_node whose subtree is to be activated
962 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
963 * needs to be explicitly activated. A node which hasn't been activated
964 * isn't visible to userland and deactivation is skipped during its
965 * removal. This is useful to construct atomic init sequences where
966 * creation of multiple nodes should either succeed or fail atomically.
968 * The caller is responsible for ensuring that this function is not called
969 * after kernfs_remove*() is invoked on @kn.
971 void kernfs_activate(struct kernfs_node *kn)
973 struct kernfs_node *pos;
975 mutex_lock(&kernfs_mutex);
978 while ((pos = kernfs_next_descendant_post(pos, kn))) {
979 if (!pos || (pos->flags & KERNFS_ACTIVATED))
982 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
983 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
985 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
986 pos->flags |= KERNFS_ACTIVATED;
989 mutex_unlock(&kernfs_mutex);
992 static void __kernfs_remove(struct kernfs_node *kn)
994 struct kernfs_node *pos;
996 lockdep_assert_held(&kernfs_mutex);
999 * Short-circuit if non-root @kn has already finished removal.
1000 * This is for kernfs_remove_self() which plays with active ref
1003 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1006 pr_debug("kernfs %s: removing\n", kn->name);
1008 /* prevent any new usage under @kn by deactivating all nodes */
1010 while ((pos = kernfs_next_descendant_post(pos, kn)))
1011 if (kernfs_active(pos))
1012 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1014 /* deactivate and unlink the subtree node-by-node */
1016 pos = kernfs_leftmost_descendant(kn);
1019 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1020 * base ref could have been put by someone else by the time
1021 * the function returns. Make sure it doesn't go away
1027 * Drain iff @kn was activated. This avoids draining and
1028 * its lockdep annotations for nodes which have never been
1029 * activated and allows embedding kernfs_remove() in create
1030 * error paths without worrying about draining.
1032 if (kn->flags & KERNFS_ACTIVATED)
1035 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1038 * kernfs_unlink_sibling() succeeds once per node. Use it
1039 * to decide who's responsible for cleanups.
1041 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1042 struct kernfs_iattrs *ps_iattr =
1043 pos->parent ? pos->parent->iattr : NULL;
1045 /* update timestamps on the parent */
1047 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1048 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1055 } while (pos != kn);
1059 * kernfs_remove - remove a kernfs_node recursively
1060 * @kn: the kernfs_node to remove
1062 * Remove @kn along with all its subdirectories and files.
1064 void kernfs_remove(struct kernfs_node *kn)
1066 mutex_lock(&kernfs_mutex);
1067 __kernfs_remove(kn);
1068 mutex_unlock(&kernfs_mutex);
1072 * kernfs_break_active_protection - break out of active protection
1073 * @kn: the self kernfs_node
1075 * The caller must be running off of a kernfs operation which is invoked
1076 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1077 * this function must also be matched with an invocation of
1078 * kernfs_unbreak_active_protection().
1080 * This function releases the active reference of @kn the caller is
1081 * holding. Once this function is called, @kn may be removed at any point
1082 * and the caller is solely responsible for ensuring that the objects it
1083 * dereferences are accessible.
1085 void kernfs_break_active_protection(struct kernfs_node *kn)
1088 * Take out ourself out of the active ref dependency chain. If
1089 * we're called without an active ref, lockdep will complain.
1091 kernfs_put_active(kn);
1095 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1096 * @kn: the self kernfs_node
1098 * If kernfs_break_active_protection() was called, this function must be
1099 * invoked before finishing the kernfs operation. Note that while this
1100 * function restores the active reference, it doesn't and can't actually
1101 * restore the active protection - @kn may already or be in the process of
1102 * being removed. Once kernfs_break_active_protection() is invoked, that
1103 * protection is irreversibly gone for the kernfs operation instance.
1105 * While this function may be called at any point after
1106 * kernfs_break_active_protection() is invoked, its most useful location
1107 * would be right before the enclosing kernfs operation returns.
1109 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1112 * @kn->active could be in any state; however, the increment we do
1113 * here will be undone as soon as the enclosing kernfs operation
1114 * finishes and this temporary bump can't break anything. If @kn
1115 * is alive, nothing changes. If @kn is being deactivated, the
1116 * soon-to-follow put will either finish deactivation or restore
1117 * deactivated state. If @kn is already removed, the temporary
1118 * bump is guaranteed to be gone before @kn is released.
1120 atomic_inc(&kn->active);
1121 if (kernfs_lockdep(kn))
1122 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1126 * kernfs_remove_self - remove a kernfs_node from its own method
1127 * @kn: the self kernfs_node to remove
1129 * The caller must be running off of a kernfs operation which is invoked
1130 * with an active reference - e.g. one of kernfs_ops. This can be used to
1131 * implement a file operation which deletes itself.
1133 * For example, the "delete" file for a sysfs device directory can be
1134 * implemented by invoking kernfs_remove_self() on the "delete" file
1135 * itself. This function breaks the circular dependency of trying to
1136 * deactivate self while holding an active ref itself. It isn't necessary
1137 * to modify the usual removal path to use kernfs_remove_self(). The
1138 * "delete" implementation can simply invoke kernfs_remove_self() on self
1139 * before proceeding with the usual removal path. kernfs will ignore later
1140 * kernfs_remove() on self.
1142 * kernfs_remove_self() can be called multiple times concurrently on the
1143 * same kernfs_node. Only the first one actually performs removal and
1144 * returns %true. All others will wait until the kernfs operation which
1145 * won self-removal finishes and return %false. Note that the losers wait
1146 * for the completion of not only the winning kernfs_remove_self() but also
1147 * the whole kernfs_ops which won the arbitration. This can be used to
1148 * guarantee, for example, all concurrent writes to a "delete" file to
1149 * finish only after the whole operation is complete.
1151 bool kernfs_remove_self(struct kernfs_node *kn)
1155 mutex_lock(&kernfs_mutex);
1156 kernfs_break_active_protection(kn);
1159 * SUICIDAL is used to arbitrate among competing invocations. Only
1160 * the first one will actually perform removal. When the removal
1161 * is complete, SUICIDED is set and the active ref is restored
1162 * while holding kernfs_mutex. The ones which lost arbitration
1163 * waits for SUICDED && drained which can happen only after the
1164 * enclosing kernfs operation which executed the winning instance
1165 * of kernfs_remove_self() finished.
1167 if (!(kn->flags & KERNFS_SUICIDAL)) {
1168 kn->flags |= KERNFS_SUICIDAL;
1169 __kernfs_remove(kn);
1170 kn->flags |= KERNFS_SUICIDED;
1173 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1177 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1179 if ((kn->flags & KERNFS_SUICIDED) &&
1180 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1183 mutex_unlock(&kernfs_mutex);
1185 mutex_lock(&kernfs_mutex);
1187 finish_wait(waitq, &wait);
1188 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1193 * This must be done while holding kernfs_mutex; otherwise, waiting
1194 * for SUICIDED && deactivated could finish prematurely.
1196 kernfs_unbreak_active_protection(kn);
1198 mutex_unlock(&kernfs_mutex);
1203 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1204 * @parent: parent of the target
1205 * @name: name of the kernfs_node to remove
1206 * @ns: namespace tag of the kernfs_node to remove
1208 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1209 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1211 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1214 struct kernfs_node *kn;
1217 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1222 mutex_lock(&kernfs_mutex);
1224 kn = kernfs_find_ns(parent, name, ns);
1226 __kernfs_remove(kn);
1228 mutex_unlock(&kernfs_mutex);
1237 * kernfs_rename_ns - move and rename a kernfs_node
1239 * @new_parent: new parent to put @sd under
1240 * @new_name: new name
1241 * @new_ns: new namespace tag
1243 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1244 const char *new_name, const void *new_ns)
1246 struct kernfs_node *old_parent;
1247 const char *old_name = NULL;
1250 /* can't move or rename root */
1254 mutex_lock(&kernfs_mutex);
1257 if (!kernfs_active(kn) || !kernfs_active(new_parent))
1261 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1262 (strcmp(kn->name, new_name) == 0))
1263 goto out; /* nothing to rename */
1266 if (kernfs_find_ns(new_parent, new_name, new_ns))
1269 /* rename kernfs_node */
1270 if (strcmp(kn->name, new_name) != 0) {
1272 new_name = kstrdup_const(new_name, GFP_KERNEL);
1280 * Move to the appropriate place in the appropriate directories rbtree.
1282 kernfs_unlink_sibling(kn);
1283 kernfs_get(new_parent);
1285 /* rename_lock protects ->parent and ->name accessors */
1286 spin_lock_irq(&kernfs_rename_lock);
1288 old_parent = kn->parent;
1289 kn->parent = new_parent;
1293 old_name = kn->name;
1294 kn->name = new_name;
1297 spin_unlock_irq(&kernfs_rename_lock);
1299 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1300 kernfs_link_sibling(kn);
1302 kernfs_put(old_parent);
1303 kfree_const(old_name);
1307 mutex_unlock(&kernfs_mutex);
1311 /* Relationship between s_mode and the DT_xxx types */
1312 static inline unsigned char dt_type(struct kernfs_node *kn)
1314 return (kn->mode >> 12) & 15;
1317 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1319 kernfs_put(filp->private_data);
1323 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1324 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1327 int valid = kernfs_active(pos) &&
1328 pos->parent == parent && hash == pos->hash;
1333 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1334 struct rb_node *node = parent->dir.children.rb_node;
1336 pos = rb_to_kn(node);
1338 if (hash < pos->hash)
1339 node = node->rb_left;
1340 else if (hash > pos->hash)
1341 node = node->rb_right;
1346 /* Skip over entries which are dying/dead or in the wrong namespace */
1347 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1348 struct rb_node *node = rb_next(&pos->rb);
1352 pos = rb_to_kn(node);
1357 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1358 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1360 pos = kernfs_dir_pos(ns, parent, ino, pos);
1363 struct rb_node *node = rb_next(&pos->rb);
1367 pos = rb_to_kn(node);
1368 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1373 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1375 struct dentry *dentry = file->f_path.dentry;
1376 struct kernfs_node *parent = dentry->d_fsdata;
1377 struct kernfs_node *pos = file->private_data;
1378 const void *ns = NULL;
1380 if (!dir_emit_dots(file, ctx))
1382 mutex_lock(&kernfs_mutex);
1384 if (kernfs_ns_enabled(parent))
1385 ns = kernfs_info(dentry->d_sb)->ns;
1387 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1389 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1390 const char *name = pos->name;
1391 unsigned int type = dt_type(pos);
1392 int len = strlen(name);
1393 ino_t ino = pos->ino;
1395 ctx->pos = pos->hash;
1396 file->private_data = pos;
1399 mutex_unlock(&kernfs_mutex);
1400 if (!dir_emit(ctx, name, len, ino, type))
1402 mutex_lock(&kernfs_mutex);
1404 mutex_unlock(&kernfs_mutex);
1405 file->private_data = NULL;
1410 static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1413 struct inode *inode = file_inode(file);
1416 mutex_lock(&inode->i_mutex);
1417 ret = generic_file_llseek(file, offset, whence);
1418 mutex_unlock(&inode->i_mutex);
1423 const struct file_operations kernfs_dir_fops = {
1424 .read = generic_read_dir,
1425 .iterate = kernfs_fop_readdir,
1426 .release = kernfs_dir_fop_release,
1427 .llseek = kernfs_dir_fop_llseek,