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);
414 if (!(kn->flags & KERNFS_STATIC_NAME))
417 if (kn->iattr->ia_secdata)
418 security_release_secctx(kn->iattr->ia_secdata,
419 kn->iattr->ia_secdata_len);
420 simple_xattrs_free(&kn->iattr->xattrs);
423 ida_simple_remove(&root->ino_ida, kn->ino);
424 kmem_cache_free(kernfs_node_cache, kn);
428 if (atomic_dec_and_test(&kn->count))
431 /* just released the root kn, free @root too */
432 ida_destroy(&root->ino_ida);
436 EXPORT_SYMBOL_GPL(kernfs_put);
438 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
440 struct kernfs_node *kn;
442 if (flags & LOOKUP_RCU)
445 /* Always perform fresh lookup for negatives */
446 if (!dentry->d_inode)
447 goto out_bad_unlocked;
449 kn = dentry->d_fsdata;
450 mutex_lock(&kernfs_mutex);
452 /* The kernfs node has been deactivated */
453 if (!kernfs_active(kn))
456 /* The kernfs node has been moved? */
457 if (dentry->d_parent->d_fsdata != kn->parent)
460 /* The kernfs node has been renamed */
461 if (strcmp(dentry->d_name.name, kn->name) != 0)
464 /* The kernfs node has been moved to a different namespace */
465 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
466 kernfs_info(dentry->d_sb)->ns != kn->ns)
469 mutex_unlock(&kernfs_mutex);
472 mutex_unlock(&kernfs_mutex);
477 static void kernfs_dop_release(struct dentry *dentry)
479 kernfs_put(dentry->d_fsdata);
482 const struct dentry_operations kernfs_dops = {
483 .d_revalidate = kernfs_dop_revalidate,
484 .d_release = kernfs_dop_release,
488 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
489 * @dentry: the dentry in question
491 * Return the kernfs_node associated with @dentry. If @dentry is not a
492 * kernfs one, %NULL is returned.
494 * While the returned kernfs_node will stay accessible as long as @dentry
495 * is accessible, the returned node can be in any state and the caller is
496 * fully responsible for determining what's accessible.
498 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
500 if (dentry->d_sb->s_op == &kernfs_sops)
501 return dentry->d_fsdata;
505 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
506 const char *name, umode_t mode,
509 char *dup_name = NULL;
510 struct kernfs_node *kn;
513 if (!(flags & KERNFS_STATIC_NAME)) {
514 name = dup_name = kstrdup(name, GFP_KERNEL);
519 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
523 ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
528 atomic_set(&kn->count, 1);
529 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
530 RB_CLEAR_NODE(&kn->rb);
539 kmem_cache_free(kernfs_node_cache, kn);
545 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
546 const char *name, umode_t mode,
549 struct kernfs_node *kn;
551 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
560 * kernfs_add_one - add kernfs_node to parent without warning
561 * @kn: kernfs_node to be added
563 * The caller must already have initialized @kn->parent. This
564 * function increments nlink of the parent's inode if @kn is a
565 * directory and link into the children list of the parent.
568 * 0 on success, -EEXIST if entry with the given name already
571 int kernfs_add_one(struct kernfs_node *kn)
573 struct kernfs_node *parent = kn->parent;
574 struct kernfs_iattrs *ps_iattr;
578 mutex_lock(&kernfs_mutex);
581 has_ns = kernfs_ns_enabled(parent);
582 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
583 has_ns ? "required" : "invalid", parent->name, kn->name))
586 if (kernfs_type(parent) != KERNFS_DIR)
590 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
593 kn->hash = kernfs_name_hash(kn->name, kn->ns);
595 ret = kernfs_link_sibling(kn);
599 /* Update timestamps on the parent */
600 ps_iattr = parent->iattr;
602 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
603 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
606 mutex_unlock(&kernfs_mutex);
609 * Activate the new node unless CREATE_DEACTIVATED is requested.
610 * If not activated here, the kernfs user is responsible for
611 * activating the node with kernfs_activate(). A node which hasn't
612 * been activated is not visible to userland and its removal won't
613 * trigger deactivation.
615 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
620 mutex_unlock(&kernfs_mutex);
625 * kernfs_find_ns - find kernfs_node with the given name
626 * @parent: kernfs_node to search under
627 * @name: name to look for
628 * @ns: the namespace tag to use
630 * Look for kernfs_node with name @name under @parent. Returns pointer to
631 * the found kernfs_node on success, %NULL on failure.
633 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
634 const unsigned char *name,
637 struct rb_node *node = parent->dir.children.rb_node;
638 bool has_ns = kernfs_ns_enabled(parent);
641 lockdep_assert_held(&kernfs_mutex);
643 if (has_ns != (bool)ns) {
644 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
645 has_ns ? "required" : "invalid", parent->name, name);
649 hash = kernfs_name_hash(name, ns);
651 struct kernfs_node *kn;
655 result = kernfs_name_compare(hash, name, ns, kn);
657 node = node->rb_left;
659 node = node->rb_right;
667 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
668 * @parent: kernfs_node to search under
669 * @name: name to look for
670 * @ns: the namespace tag to use
672 * Look for kernfs_node with name @name under @parent and get a reference
673 * if found. This function may sleep and returns pointer to the found
674 * kernfs_node on success, %NULL on failure.
676 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
677 const char *name, const void *ns)
679 struct kernfs_node *kn;
681 mutex_lock(&kernfs_mutex);
682 kn = kernfs_find_ns(parent, name, ns);
684 mutex_unlock(&kernfs_mutex);
688 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
691 * kernfs_create_root - create a new kernfs hierarchy
692 * @scops: optional syscall operations for the hierarchy
693 * @flags: KERNFS_ROOT_* flags
694 * @priv: opaque data associated with the new directory
696 * Returns the root of the new hierarchy on success, ERR_PTR() value on
699 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
700 unsigned int flags, void *priv)
702 struct kernfs_root *root;
703 struct kernfs_node *kn;
705 root = kzalloc(sizeof(*root), GFP_KERNEL);
707 return ERR_PTR(-ENOMEM);
709 ida_init(&root->ino_ida);
710 INIT_LIST_HEAD(&root->supers);
712 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
715 ida_destroy(&root->ino_ida);
717 return ERR_PTR(-ENOMEM);
723 root->syscall_ops = scops;
726 init_waitqueue_head(&root->deactivate_waitq);
728 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
735 * kernfs_destroy_root - destroy a kernfs hierarchy
736 * @root: root of the hierarchy to destroy
738 * Destroy the hierarchy anchored at @root by removing all existing
739 * directories and destroying @root.
741 void kernfs_destroy_root(struct kernfs_root *root)
743 kernfs_remove(root->kn); /* will also free @root */
747 * kernfs_create_dir_ns - create a directory
748 * @parent: parent in which to create a new directory
749 * @name: name of the new directory
750 * @mode: mode of the new directory
751 * @priv: opaque data associated with the new directory
752 * @ns: optional namespace tag of the directory
754 * Returns the created node on success, ERR_PTR() value on failure.
756 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
757 const char *name, umode_t mode,
758 void *priv, const void *ns)
760 struct kernfs_node *kn;
764 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
766 return ERR_PTR(-ENOMEM);
768 kn->dir.root = parent->dir.root;
773 rc = kernfs_add_one(kn);
781 static struct dentry *kernfs_iop_lookup(struct inode *dir,
782 struct dentry *dentry,
786 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
787 struct kernfs_node *kn;
789 const void *ns = NULL;
791 mutex_lock(&kernfs_mutex);
793 if (kernfs_ns_enabled(parent))
794 ns = kernfs_info(dir->i_sb)->ns;
796 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
799 if (!kn || !kernfs_active(kn)) {
804 dentry->d_fsdata = kn;
806 /* attach dentry and inode */
807 inode = kernfs_get_inode(dir->i_sb, kn);
809 ret = ERR_PTR(-ENOMEM);
813 /* instantiate and hash dentry */
814 ret = d_splice_alias(inode, dentry);
816 mutex_unlock(&kernfs_mutex);
820 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
823 struct kernfs_node *parent = dir->i_private;
824 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
827 if (!scops || !scops->mkdir)
830 if (!kernfs_get_active(parent))
833 ret = scops->mkdir(parent, dentry->d_name.name, mode);
835 kernfs_put_active(parent);
839 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
841 struct kernfs_node *kn = dentry->d_fsdata;
842 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
845 if (!scops || !scops->rmdir)
848 if (!kernfs_get_active(kn))
851 ret = scops->rmdir(kn);
853 kernfs_put_active(kn);
857 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
858 struct inode *new_dir, struct dentry *new_dentry)
860 struct kernfs_node *kn = old_dentry->d_fsdata;
861 struct kernfs_node *new_parent = new_dir->i_private;
862 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
865 if (!scops || !scops->rename)
868 if (!kernfs_get_active(kn))
871 if (!kernfs_get_active(new_parent)) {
872 kernfs_put_active(kn);
876 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
878 kernfs_put_active(new_parent);
879 kernfs_put_active(kn);
883 const struct inode_operations kernfs_dir_iops = {
884 .lookup = kernfs_iop_lookup,
885 .permission = kernfs_iop_permission,
886 .setattr = kernfs_iop_setattr,
887 .getattr = kernfs_iop_getattr,
888 .setxattr = kernfs_iop_setxattr,
889 .removexattr = kernfs_iop_removexattr,
890 .getxattr = kernfs_iop_getxattr,
891 .listxattr = kernfs_iop_listxattr,
893 .mkdir = kernfs_iop_mkdir,
894 .rmdir = kernfs_iop_rmdir,
895 .rename = kernfs_iop_rename,
898 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
900 struct kernfs_node *last;
907 if (kernfs_type(pos) != KERNFS_DIR)
910 rbn = rb_first(&pos->dir.children);
921 * kernfs_next_descendant_post - find the next descendant for post-order walk
922 * @pos: the current position (%NULL to initiate traversal)
923 * @root: kernfs_node whose descendants to walk
925 * Find the next descendant to visit for post-order traversal of @root's
926 * descendants. @root is included in the iteration and the last node to be
929 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
930 struct kernfs_node *root)
934 lockdep_assert_held(&kernfs_mutex);
936 /* if first iteration, visit leftmost descendant which may be root */
938 return kernfs_leftmost_descendant(root);
940 /* if we visited @root, we're done */
944 /* if there's an unvisited sibling, visit its leftmost descendant */
945 rbn = rb_next(&pos->rb);
947 return kernfs_leftmost_descendant(rb_to_kn(rbn));
949 /* no sibling left, visit parent */
954 * kernfs_activate - activate a node which started deactivated
955 * @kn: kernfs_node whose subtree is to be activated
957 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
958 * needs to be explicitly activated. A node which hasn't been activated
959 * isn't visible to userland and deactivation is skipped during its
960 * removal. This is useful to construct atomic init sequences where
961 * creation of multiple nodes should either succeed or fail atomically.
963 * The caller is responsible for ensuring that this function is not called
964 * after kernfs_remove*() is invoked on @kn.
966 void kernfs_activate(struct kernfs_node *kn)
968 struct kernfs_node *pos;
970 mutex_lock(&kernfs_mutex);
973 while ((pos = kernfs_next_descendant_post(pos, kn))) {
974 if (!pos || (pos->flags & KERNFS_ACTIVATED))
977 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
978 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
980 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
981 pos->flags |= KERNFS_ACTIVATED;
984 mutex_unlock(&kernfs_mutex);
987 static void __kernfs_remove(struct kernfs_node *kn)
989 struct kernfs_node *pos;
991 lockdep_assert_held(&kernfs_mutex);
994 * Short-circuit if non-root @kn has already finished removal.
995 * This is for kernfs_remove_self() which plays with active ref
998 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1001 pr_debug("kernfs %s: removing\n", kn->name);
1003 /* prevent any new usage under @kn by deactivating all nodes */
1005 while ((pos = kernfs_next_descendant_post(pos, kn)))
1006 if (kernfs_active(pos))
1007 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1009 /* deactivate and unlink the subtree node-by-node */
1011 pos = kernfs_leftmost_descendant(kn);
1014 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1015 * base ref could have been put by someone else by the time
1016 * the function returns. Make sure it doesn't go away
1022 * Drain iff @kn was activated. This avoids draining and
1023 * its lockdep annotations for nodes which have never been
1024 * activated and allows embedding kernfs_remove() in create
1025 * error paths without worrying about draining.
1027 if (kn->flags & KERNFS_ACTIVATED)
1030 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1033 * kernfs_unlink_sibling() succeeds once per node. Use it
1034 * to decide who's responsible for cleanups.
1036 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1037 struct kernfs_iattrs *ps_iattr =
1038 pos->parent ? pos->parent->iattr : NULL;
1040 /* update timestamps on the parent */
1042 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1043 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1050 } while (pos != kn);
1054 * kernfs_remove - remove a kernfs_node recursively
1055 * @kn: the kernfs_node to remove
1057 * Remove @kn along with all its subdirectories and files.
1059 void kernfs_remove(struct kernfs_node *kn)
1061 mutex_lock(&kernfs_mutex);
1062 __kernfs_remove(kn);
1063 mutex_unlock(&kernfs_mutex);
1067 * kernfs_break_active_protection - break out of active protection
1068 * @kn: the self kernfs_node
1070 * The caller must be running off of a kernfs operation which is invoked
1071 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1072 * this function must also be matched with an invocation of
1073 * kernfs_unbreak_active_protection().
1075 * This function releases the active reference of @kn the caller is
1076 * holding. Once this function is called, @kn may be removed at any point
1077 * and the caller is solely responsible for ensuring that the objects it
1078 * dereferences are accessible.
1080 void kernfs_break_active_protection(struct kernfs_node *kn)
1083 * Take out ourself out of the active ref dependency chain. If
1084 * we're called without an active ref, lockdep will complain.
1086 kernfs_put_active(kn);
1090 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1091 * @kn: the self kernfs_node
1093 * If kernfs_break_active_protection() was called, this function must be
1094 * invoked before finishing the kernfs operation. Note that while this
1095 * function restores the active reference, it doesn't and can't actually
1096 * restore the active protection - @kn may already or be in the process of
1097 * being removed. Once kernfs_break_active_protection() is invoked, that
1098 * protection is irreversibly gone for the kernfs operation instance.
1100 * While this function may be called at any point after
1101 * kernfs_break_active_protection() is invoked, its most useful location
1102 * would be right before the enclosing kernfs operation returns.
1104 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1107 * @kn->active could be in any state; however, the increment we do
1108 * here will be undone as soon as the enclosing kernfs operation
1109 * finishes and this temporary bump can't break anything. If @kn
1110 * is alive, nothing changes. If @kn is being deactivated, the
1111 * soon-to-follow put will either finish deactivation or restore
1112 * deactivated state. If @kn is already removed, the temporary
1113 * bump is guaranteed to be gone before @kn is released.
1115 atomic_inc(&kn->active);
1116 if (kernfs_lockdep(kn))
1117 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1121 * kernfs_remove_self - remove a kernfs_node from its own method
1122 * @kn: the self kernfs_node to remove
1124 * The caller must be running off of a kernfs operation which is invoked
1125 * with an active reference - e.g. one of kernfs_ops. This can be used to
1126 * implement a file operation which deletes itself.
1128 * For example, the "delete" file for a sysfs device directory can be
1129 * implemented by invoking kernfs_remove_self() on the "delete" file
1130 * itself. This function breaks the circular dependency of trying to
1131 * deactivate self while holding an active ref itself. It isn't necessary
1132 * to modify the usual removal path to use kernfs_remove_self(). The
1133 * "delete" implementation can simply invoke kernfs_remove_self() on self
1134 * before proceeding with the usual removal path. kernfs will ignore later
1135 * kernfs_remove() on self.
1137 * kernfs_remove_self() can be called multiple times concurrently on the
1138 * same kernfs_node. Only the first one actually performs removal and
1139 * returns %true. All others will wait until the kernfs operation which
1140 * won self-removal finishes and return %false. Note that the losers wait
1141 * for the completion of not only the winning kernfs_remove_self() but also
1142 * the whole kernfs_ops which won the arbitration. This can be used to
1143 * guarantee, for example, all concurrent writes to a "delete" file to
1144 * finish only after the whole operation is complete.
1146 bool kernfs_remove_self(struct kernfs_node *kn)
1150 mutex_lock(&kernfs_mutex);
1151 kernfs_break_active_protection(kn);
1154 * SUICIDAL is used to arbitrate among competing invocations. Only
1155 * the first one will actually perform removal. When the removal
1156 * is complete, SUICIDED is set and the active ref is restored
1157 * while holding kernfs_mutex. The ones which lost arbitration
1158 * waits for SUICDED && drained which can happen only after the
1159 * enclosing kernfs operation which executed the winning instance
1160 * of kernfs_remove_self() finished.
1162 if (!(kn->flags & KERNFS_SUICIDAL)) {
1163 kn->flags |= KERNFS_SUICIDAL;
1164 __kernfs_remove(kn);
1165 kn->flags |= KERNFS_SUICIDED;
1168 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1172 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1174 if ((kn->flags & KERNFS_SUICIDED) &&
1175 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1178 mutex_unlock(&kernfs_mutex);
1180 mutex_lock(&kernfs_mutex);
1182 finish_wait(waitq, &wait);
1183 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1188 * This must be done while holding kernfs_mutex; otherwise, waiting
1189 * for SUICIDED && deactivated could finish prematurely.
1191 kernfs_unbreak_active_protection(kn);
1193 mutex_unlock(&kernfs_mutex);
1198 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1199 * @parent: parent of the target
1200 * @name: name of the kernfs_node to remove
1201 * @ns: namespace tag of the kernfs_node to remove
1203 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1204 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1206 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1209 struct kernfs_node *kn;
1212 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1217 mutex_lock(&kernfs_mutex);
1219 kn = kernfs_find_ns(parent, name, ns);
1221 __kernfs_remove(kn);
1223 mutex_unlock(&kernfs_mutex);
1232 * kernfs_rename_ns - move and rename a kernfs_node
1234 * @new_parent: new parent to put @sd under
1235 * @new_name: new name
1236 * @new_ns: new namespace tag
1238 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1239 const char *new_name, const void *new_ns)
1241 struct kernfs_node *old_parent;
1242 const char *old_name = NULL;
1245 /* can't move or rename root */
1249 mutex_lock(&kernfs_mutex);
1252 if (!kernfs_active(kn) || !kernfs_active(new_parent))
1256 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1257 (strcmp(kn->name, new_name) == 0))
1258 goto out; /* nothing to rename */
1261 if (kernfs_find_ns(new_parent, new_name, new_ns))
1264 /* rename kernfs_node */
1265 if (strcmp(kn->name, new_name) != 0) {
1267 new_name = kstrdup(new_name, GFP_KERNEL);
1275 * Move to the appropriate place in the appropriate directories rbtree.
1277 kernfs_unlink_sibling(kn);
1278 kernfs_get(new_parent);
1280 /* rename_lock protects ->parent and ->name accessors */
1281 spin_lock_irq(&kernfs_rename_lock);
1283 old_parent = kn->parent;
1284 kn->parent = new_parent;
1288 if (!(kn->flags & KERNFS_STATIC_NAME))
1289 old_name = kn->name;
1290 kn->flags &= ~KERNFS_STATIC_NAME;
1291 kn->name = new_name;
1294 spin_unlock_irq(&kernfs_rename_lock);
1296 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1297 kernfs_link_sibling(kn);
1299 kernfs_put(old_parent);
1304 mutex_unlock(&kernfs_mutex);
1308 /* Relationship between s_mode and the DT_xxx types */
1309 static inline unsigned char dt_type(struct kernfs_node *kn)
1311 return (kn->mode >> 12) & 15;
1314 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1316 kernfs_put(filp->private_data);
1320 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1321 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1324 int valid = kernfs_active(pos) &&
1325 pos->parent == parent && hash == pos->hash;
1330 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1331 struct rb_node *node = parent->dir.children.rb_node;
1333 pos = rb_to_kn(node);
1335 if (hash < pos->hash)
1336 node = node->rb_left;
1337 else if (hash > pos->hash)
1338 node = node->rb_right;
1343 /* Skip over entries which are dying/dead or in the wrong namespace */
1344 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1345 struct rb_node *node = rb_next(&pos->rb);
1349 pos = rb_to_kn(node);
1354 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1355 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1357 pos = kernfs_dir_pos(ns, parent, ino, pos);
1360 struct rb_node *node = rb_next(&pos->rb);
1364 pos = rb_to_kn(node);
1365 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1370 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1372 struct dentry *dentry = file->f_path.dentry;
1373 struct kernfs_node *parent = dentry->d_fsdata;
1374 struct kernfs_node *pos = file->private_data;
1375 const void *ns = NULL;
1377 if (!dir_emit_dots(file, ctx))
1379 mutex_lock(&kernfs_mutex);
1381 if (kernfs_ns_enabled(parent))
1382 ns = kernfs_info(dentry->d_sb)->ns;
1384 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1386 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1387 const char *name = pos->name;
1388 unsigned int type = dt_type(pos);
1389 int len = strlen(name);
1390 ino_t ino = pos->ino;
1392 ctx->pos = pos->hash;
1393 file->private_data = pos;
1396 mutex_unlock(&kernfs_mutex);
1397 if (!dir_emit(ctx, name, len, ino, type))
1399 mutex_lock(&kernfs_mutex);
1401 mutex_unlock(&kernfs_mutex);
1402 file->private_data = NULL;
1407 static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1410 struct inode *inode = file_inode(file);
1413 mutex_lock(&inode->i_mutex);
1414 ret = generic_file_llseek(file, offset, whence);
1415 mutex_unlock(&inode->i_mutex);
1420 const struct file_operations kernfs_dir_fops = {
1421 .read = generic_read_dir,
1422 .iterate = kernfs_fop_readdir,
1423 .release = kernfs_dir_fop_release,
1424 .llseek = kernfs_dir_fop_llseek,