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);
545 * If the ino of the sysfs entry created for a kmem cache gets
546 * allocated from an ida layer, which is accounted to the memcg that
547 * owns the cache, the memcg will get pinned forever. So do not account
548 * ino ida allocations.
550 ret = ida_simple_get(&root->ino_ida, 1, 0,
551 GFP_KERNEL | __GFP_NOACCOUNT);
556 atomic_set(&kn->count, 1);
557 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
558 RB_CLEAR_NODE(&kn->rb);
567 kmem_cache_free(kernfs_node_cache, kn);
573 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
574 const char *name, umode_t mode,
577 struct kernfs_node *kn;
579 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
588 * kernfs_add_one - add kernfs_node to parent without warning
589 * @kn: kernfs_node to be added
591 * The caller must already have initialized @kn->parent. This
592 * function increments nlink of the parent's inode if @kn is a
593 * directory and link into the children list of the parent.
596 * 0 on success, -EEXIST if entry with the given name already
599 int kernfs_add_one(struct kernfs_node *kn)
601 struct kernfs_node *parent = kn->parent;
602 struct kernfs_iattrs *ps_iattr;
606 mutex_lock(&kernfs_mutex);
609 has_ns = kernfs_ns_enabled(parent);
610 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
611 has_ns ? "required" : "invalid", parent->name, kn->name))
614 if (kernfs_type(parent) != KERNFS_DIR)
618 if (parent->flags & KERNFS_EMPTY_DIR)
621 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
624 kn->hash = kernfs_name_hash(kn->name, kn->ns);
626 ret = kernfs_link_sibling(kn);
630 /* Update timestamps on the parent */
631 ps_iattr = parent->iattr;
633 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
634 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
637 mutex_unlock(&kernfs_mutex);
640 * Activate the new node unless CREATE_DEACTIVATED is requested.
641 * If not activated here, the kernfs user is responsible for
642 * activating the node with kernfs_activate(). A node which hasn't
643 * been activated is not visible to userland and its removal won't
644 * trigger deactivation.
646 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
651 mutex_unlock(&kernfs_mutex);
656 * kernfs_find_ns - find kernfs_node with the given name
657 * @parent: kernfs_node to search under
658 * @name: name to look for
659 * @ns: the namespace tag to use
661 * Look for kernfs_node with name @name under @parent. Returns pointer to
662 * the found kernfs_node on success, %NULL on failure.
664 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
665 const unsigned char *name,
668 struct rb_node *node = parent->dir.children.rb_node;
669 bool has_ns = kernfs_ns_enabled(parent);
672 lockdep_assert_held(&kernfs_mutex);
674 if (has_ns != (bool)ns) {
675 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
676 has_ns ? "required" : "invalid", parent->name, name);
680 hash = kernfs_name_hash(name, ns);
682 struct kernfs_node *kn;
686 result = kernfs_name_compare(hash, name, ns, kn);
688 node = node->rb_left;
690 node = node->rb_right;
697 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
698 const unsigned char *path,
701 static char path_buf[PATH_MAX]; /* protected by kernfs_mutex */
702 size_t len = strlcpy(path_buf, path, PATH_MAX);
706 lockdep_assert_held(&kernfs_mutex);
711 while ((name = strsep(&p, "/")) && parent) {
714 parent = kernfs_find_ns(parent, name, ns);
721 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
722 * @parent: kernfs_node to search under
723 * @name: name to look for
724 * @ns: the namespace tag to use
726 * Look for kernfs_node with name @name under @parent and get a reference
727 * if found. This function may sleep and returns pointer to the found
728 * kernfs_node on success, %NULL on failure.
730 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
731 const char *name, const void *ns)
733 struct kernfs_node *kn;
735 mutex_lock(&kernfs_mutex);
736 kn = kernfs_find_ns(parent, name, ns);
738 mutex_unlock(&kernfs_mutex);
742 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
745 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
746 * @parent: kernfs_node to search under
747 * @path: path to look for
748 * @ns: the namespace tag to use
750 * Look for kernfs_node with path @path under @parent and get a reference
751 * if found. This function may sleep and returns pointer to the found
752 * kernfs_node on success, %NULL on failure.
754 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
755 const char *path, const void *ns)
757 struct kernfs_node *kn;
759 mutex_lock(&kernfs_mutex);
760 kn = kernfs_walk_ns(parent, path, ns);
762 mutex_unlock(&kernfs_mutex);
768 * kernfs_create_root - create a new kernfs hierarchy
769 * @scops: optional syscall operations for the hierarchy
770 * @flags: KERNFS_ROOT_* flags
771 * @priv: opaque data associated with the new directory
773 * Returns the root of the new hierarchy on success, ERR_PTR() value on
776 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
777 unsigned int flags, void *priv)
779 struct kernfs_root *root;
780 struct kernfs_node *kn;
782 root = kzalloc(sizeof(*root), GFP_KERNEL);
784 return ERR_PTR(-ENOMEM);
786 ida_init(&root->ino_ida);
787 INIT_LIST_HEAD(&root->supers);
789 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
792 ida_destroy(&root->ino_ida);
794 return ERR_PTR(-ENOMEM);
800 root->syscall_ops = scops;
803 init_waitqueue_head(&root->deactivate_waitq);
805 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
812 * kernfs_destroy_root - destroy a kernfs hierarchy
813 * @root: root of the hierarchy to destroy
815 * Destroy the hierarchy anchored at @root by removing all existing
816 * directories and destroying @root.
818 void kernfs_destroy_root(struct kernfs_root *root)
820 kernfs_remove(root->kn); /* will also free @root */
824 * kernfs_create_dir_ns - create a directory
825 * @parent: parent in which to create a new directory
826 * @name: name of the new directory
827 * @mode: mode of the new directory
828 * @priv: opaque data associated with the new directory
829 * @ns: optional namespace tag of the directory
831 * Returns the created node on success, ERR_PTR() value on failure.
833 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
834 const char *name, umode_t mode,
835 void *priv, const void *ns)
837 struct kernfs_node *kn;
841 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
843 return ERR_PTR(-ENOMEM);
845 kn->dir.root = parent->dir.root;
850 rc = kernfs_add_one(kn);
859 * kernfs_create_empty_dir - create an always empty directory
860 * @parent: parent in which to create a new directory
861 * @name: name of the new directory
863 * Returns the created node on success, ERR_PTR() value on failure.
865 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
868 struct kernfs_node *kn;
872 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
874 return ERR_PTR(-ENOMEM);
876 kn->flags |= KERNFS_EMPTY_DIR;
877 kn->dir.root = parent->dir.root;
882 rc = kernfs_add_one(kn);
890 static struct dentry *kernfs_iop_lookup(struct inode *dir,
891 struct dentry *dentry,
895 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
896 struct kernfs_node *kn;
898 const void *ns = NULL;
900 mutex_lock(&kernfs_mutex);
902 if (kernfs_ns_enabled(parent))
903 ns = kernfs_info(dir->i_sb)->ns;
905 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
908 if (!kn || !kernfs_active(kn)) {
913 dentry->d_fsdata = kn;
915 /* attach dentry and inode */
916 inode = kernfs_get_inode(dir->i_sb, kn);
918 ret = ERR_PTR(-ENOMEM);
922 /* instantiate and hash dentry */
923 ret = d_splice_alias(inode, dentry);
925 mutex_unlock(&kernfs_mutex);
929 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
932 struct kernfs_node *parent = dir->i_private;
933 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
936 if (!scops || !scops->mkdir)
939 if (!kernfs_get_active(parent))
942 ret = scops->mkdir(parent, dentry->d_name.name, mode);
944 kernfs_put_active(parent);
948 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
950 struct kernfs_node *kn = dentry->d_fsdata;
951 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
954 if (!scops || !scops->rmdir)
957 if (!kernfs_get_active(kn))
960 ret = scops->rmdir(kn);
962 kernfs_put_active(kn);
966 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
967 struct inode *new_dir, struct dentry *new_dentry)
969 struct kernfs_node *kn = old_dentry->d_fsdata;
970 struct kernfs_node *new_parent = new_dir->i_private;
971 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
974 if (!scops || !scops->rename)
977 if (!kernfs_get_active(kn))
980 if (!kernfs_get_active(new_parent)) {
981 kernfs_put_active(kn);
985 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
987 kernfs_put_active(new_parent);
988 kernfs_put_active(kn);
992 const struct inode_operations kernfs_dir_iops = {
993 .lookup = kernfs_iop_lookup,
994 .permission = kernfs_iop_permission,
995 .setattr = kernfs_iop_setattr,
996 .getattr = kernfs_iop_getattr,
997 .setxattr = kernfs_iop_setxattr,
998 .removexattr = kernfs_iop_removexattr,
999 .getxattr = kernfs_iop_getxattr,
1000 .listxattr = kernfs_iop_listxattr,
1002 .mkdir = kernfs_iop_mkdir,
1003 .rmdir = kernfs_iop_rmdir,
1004 .rename = kernfs_iop_rename,
1007 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1009 struct kernfs_node *last;
1012 struct rb_node *rbn;
1016 if (kernfs_type(pos) != KERNFS_DIR)
1019 rbn = rb_first(&pos->dir.children);
1023 pos = rb_to_kn(rbn);
1030 * kernfs_next_descendant_post - find the next descendant for post-order walk
1031 * @pos: the current position (%NULL to initiate traversal)
1032 * @root: kernfs_node whose descendants to walk
1034 * Find the next descendant to visit for post-order traversal of @root's
1035 * descendants. @root is included in the iteration and the last node to be
1038 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1039 struct kernfs_node *root)
1041 struct rb_node *rbn;
1043 lockdep_assert_held(&kernfs_mutex);
1045 /* if first iteration, visit leftmost descendant which may be root */
1047 return kernfs_leftmost_descendant(root);
1049 /* if we visited @root, we're done */
1053 /* if there's an unvisited sibling, visit its leftmost descendant */
1054 rbn = rb_next(&pos->rb);
1056 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1058 /* no sibling left, visit parent */
1063 * kernfs_activate - activate a node which started deactivated
1064 * @kn: kernfs_node whose subtree is to be activated
1066 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1067 * needs to be explicitly activated. A node which hasn't been activated
1068 * isn't visible to userland and deactivation is skipped during its
1069 * removal. This is useful to construct atomic init sequences where
1070 * creation of multiple nodes should either succeed or fail atomically.
1072 * The caller is responsible for ensuring that this function is not called
1073 * after kernfs_remove*() is invoked on @kn.
1075 void kernfs_activate(struct kernfs_node *kn)
1077 struct kernfs_node *pos;
1079 mutex_lock(&kernfs_mutex);
1082 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1083 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1086 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1087 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1089 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1090 pos->flags |= KERNFS_ACTIVATED;
1093 mutex_unlock(&kernfs_mutex);
1096 static void __kernfs_remove(struct kernfs_node *kn)
1098 struct kernfs_node *pos;
1100 lockdep_assert_held(&kernfs_mutex);
1103 * Short-circuit if non-root @kn has already finished removal.
1104 * This is for kernfs_remove_self() which plays with active ref
1107 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1110 pr_debug("kernfs %s: removing\n", kn->name);
1112 /* prevent any new usage under @kn by deactivating all nodes */
1114 while ((pos = kernfs_next_descendant_post(pos, kn)))
1115 if (kernfs_active(pos))
1116 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1118 /* deactivate and unlink the subtree node-by-node */
1120 pos = kernfs_leftmost_descendant(kn);
1123 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1124 * base ref could have been put by someone else by the time
1125 * the function returns. Make sure it doesn't go away
1131 * Drain iff @kn was activated. This avoids draining and
1132 * its lockdep annotations for nodes which have never been
1133 * activated and allows embedding kernfs_remove() in create
1134 * error paths without worrying about draining.
1136 if (kn->flags & KERNFS_ACTIVATED)
1139 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1142 * kernfs_unlink_sibling() succeeds once per node. Use it
1143 * to decide who's responsible for cleanups.
1145 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1146 struct kernfs_iattrs *ps_iattr =
1147 pos->parent ? pos->parent->iattr : NULL;
1149 /* update timestamps on the parent */
1151 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1152 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1159 } while (pos != kn);
1163 * kernfs_remove - remove a kernfs_node recursively
1164 * @kn: the kernfs_node to remove
1166 * Remove @kn along with all its subdirectories and files.
1168 void kernfs_remove(struct kernfs_node *kn)
1170 mutex_lock(&kernfs_mutex);
1171 __kernfs_remove(kn);
1172 mutex_unlock(&kernfs_mutex);
1176 * kernfs_break_active_protection - break out of active protection
1177 * @kn: the self kernfs_node
1179 * The caller must be running off of a kernfs operation which is invoked
1180 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1181 * this function must also be matched with an invocation of
1182 * kernfs_unbreak_active_protection().
1184 * This function releases the active reference of @kn the caller is
1185 * holding. Once this function is called, @kn may be removed at any point
1186 * and the caller is solely responsible for ensuring that the objects it
1187 * dereferences are accessible.
1189 void kernfs_break_active_protection(struct kernfs_node *kn)
1192 * Take out ourself out of the active ref dependency chain. If
1193 * we're called without an active ref, lockdep will complain.
1195 kernfs_put_active(kn);
1199 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1200 * @kn: the self kernfs_node
1202 * If kernfs_break_active_protection() was called, this function must be
1203 * invoked before finishing the kernfs operation. Note that while this
1204 * function restores the active reference, it doesn't and can't actually
1205 * restore the active protection - @kn may already or be in the process of
1206 * being removed. Once kernfs_break_active_protection() is invoked, that
1207 * protection is irreversibly gone for the kernfs operation instance.
1209 * While this function may be called at any point after
1210 * kernfs_break_active_protection() is invoked, its most useful location
1211 * would be right before the enclosing kernfs operation returns.
1213 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1216 * @kn->active could be in any state; however, the increment we do
1217 * here will be undone as soon as the enclosing kernfs operation
1218 * finishes and this temporary bump can't break anything. If @kn
1219 * is alive, nothing changes. If @kn is being deactivated, the
1220 * soon-to-follow put will either finish deactivation or restore
1221 * deactivated state. If @kn is already removed, the temporary
1222 * bump is guaranteed to be gone before @kn is released.
1224 atomic_inc(&kn->active);
1225 if (kernfs_lockdep(kn))
1226 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1230 * kernfs_remove_self - remove a kernfs_node from its own method
1231 * @kn: the self kernfs_node to remove
1233 * The caller must be running off of a kernfs operation which is invoked
1234 * with an active reference - e.g. one of kernfs_ops. This can be used to
1235 * implement a file operation which deletes itself.
1237 * For example, the "delete" file for a sysfs device directory can be
1238 * implemented by invoking kernfs_remove_self() on the "delete" file
1239 * itself. This function breaks the circular dependency of trying to
1240 * deactivate self while holding an active ref itself. It isn't necessary
1241 * to modify the usual removal path to use kernfs_remove_self(). The
1242 * "delete" implementation can simply invoke kernfs_remove_self() on self
1243 * before proceeding with the usual removal path. kernfs will ignore later
1244 * kernfs_remove() on self.
1246 * kernfs_remove_self() can be called multiple times concurrently on the
1247 * same kernfs_node. Only the first one actually performs removal and
1248 * returns %true. All others will wait until the kernfs operation which
1249 * won self-removal finishes and return %false. Note that the losers wait
1250 * for the completion of not only the winning kernfs_remove_self() but also
1251 * the whole kernfs_ops which won the arbitration. This can be used to
1252 * guarantee, for example, all concurrent writes to a "delete" file to
1253 * finish only after the whole operation is complete.
1255 bool kernfs_remove_self(struct kernfs_node *kn)
1259 mutex_lock(&kernfs_mutex);
1260 kernfs_break_active_protection(kn);
1263 * SUICIDAL is used to arbitrate among competing invocations. Only
1264 * the first one will actually perform removal. When the removal
1265 * is complete, SUICIDED is set and the active ref is restored
1266 * while holding kernfs_mutex. The ones which lost arbitration
1267 * waits for SUICDED && drained which can happen only after the
1268 * enclosing kernfs operation which executed the winning instance
1269 * of kernfs_remove_self() finished.
1271 if (!(kn->flags & KERNFS_SUICIDAL)) {
1272 kn->flags |= KERNFS_SUICIDAL;
1273 __kernfs_remove(kn);
1274 kn->flags |= KERNFS_SUICIDED;
1277 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1281 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1283 if ((kn->flags & KERNFS_SUICIDED) &&
1284 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1287 mutex_unlock(&kernfs_mutex);
1289 mutex_lock(&kernfs_mutex);
1291 finish_wait(waitq, &wait);
1292 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1297 * This must be done while holding kernfs_mutex; otherwise, waiting
1298 * for SUICIDED && deactivated could finish prematurely.
1300 kernfs_unbreak_active_protection(kn);
1302 mutex_unlock(&kernfs_mutex);
1307 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1308 * @parent: parent of the target
1309 * @name: name of the kernfs_node to remove
1310 * @ns: namespace tag of the kernfs_node to remove
1312 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1313 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1315 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1318 struct kernfs_node *kn;
1321 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1326 mutex_lock(&kernfs_mutex);
1328 kn = kernfs_find_ns(parent, name, ns);
1330 __kernfs_remove(kn);
1332 mutex_unlock(&kernfs_mutex);
1341 * kernfs_rename_ns - move and rename a kernfs_node
1343 * @new_parent: new parent to put @sd under
1344 * @new_name: new name
1345 * @new_ns: new namespace tag
1347 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1348 const char *new_name, const void *new_ns)
1350 struct kernfs_node *old_parent;
1351 const char *old_name = NULL;
1354 /* can't move or rename root */
1358 mutex_lock(&kernfs_mutex);
1361 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1362 (new_parent->flags & KERNFS_EMPTY_DIR))
1366 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1367 (strcmp(kn->name, new_name) == 0))
1368 goto out; /* nothing to rename */
1371 if (kernfs_find_ns(new_parent, new_name, new_ns))
1374 /* rename kernfs_node */
1375 if (strcmp(kn->name, new_name) != 0) {
1377 new_name = kstrdup_const(new_name, GFP_KERNEL);
1385 * Move to the appropriate place in the appropriate directories rbtree.
1387 kernfs_unlink_sibling(kn);
1388 kernfs_get(new_parent);
1390 /* rename_lock protects ->parent and ->name accessors */
1391 spin_lock_irq(&kernfs_rename_lock);
1393 old_parent = kn->parent;
1394 kn->parent = new_parent;
1398 old_name = kn->name;
1399 kn->name = new_name;
1402 spin_unlock_irq(&kernfs_rename_lock);
1404 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1405 kernfs_link_sibling(kn);
1407 kernfs_put(old_parent);
1408 kfree_const(old_name);
1412 mutex_unlock(&kernfs_mutex);
1416 /* Relationship between s_mode and the DT_xxx types */
1417 static inline unsigned char dt_type(struct kernfs_node *kn)
1419 return (kn->mode >> 12) & 15;
1422 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1424 kernfs_put(filp->private_data);
1428 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1429 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1432 int valid = kernfs_active(pos) &&
1433 pos->parent == parent && hash == pos->hash;
1438 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1439 struct rb_node *node = parent->dir.children.rb_node;
1441 pos = rb_to_kn(node);
1443 if (hash < pos->hash)
1444 node = node->rb_left;
1445 else if (hash > pos->hash)
1446 node = node->rb_right;
1451 /* Skip over entries which are dying/dead or in the wrong namespace */
1452 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1453 struct rb_node *node = rb_next(&pos->rb);
1457 pos = rb_to_kn(node);
1462 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1463 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1465 pos = kernfs_dir_pos(ns, parent, ino, pos);
1468 struct rb_node *node = rb_next(&pos->rb);
1472 pos = rb_to_kn(node);
1473 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1478 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1480 struct dentry *dentry = file->f_path.dentry;
1481 struct kernfs_node *parent = dentry->d_fsdata;
1482 struct kernfs_node *pos = file->private_data;
1483 const void *ns = NULL;
1485 if (!dir_emit_dots(file, ctx))
1487 mutex_lock(&kernfs_mutex);
1489 if (kernfs_ns_enabled(parent))
1490 ns = kernfs_info(dentry->d_sb)->ns;
1492 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1494 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1495 const char *name = pos->name;
1496 unsigned int type = dt_type(pos);
1497 int len = strlen(name);
1498 ino_t ino = pos->ino;
1500 ctx->pos = pos->hash;
1501 file->private_data = pos;
1504 mutex_unlock(&kernfs_mutex);
1505 if (!dir_emit(ctx, name, len, ino, type))
1507 mutex_lock(&kernfs_mutex);
1509 mutex_unlock(&kernfs_mutex);
1510 file->private_data = NULL;
1515 static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1518 struct inode *inode = file_inode(file);
1521 mutex_lock(&inode->i_mutex);
1522 ret = generic_file_llseek(file, offset, whence);
1523 mutex_unlock(&inode->i_mutex);
1528 const struct file_operations kernfs_dir_fops = {
1529 .read = generic_read_dir,
1530 .iterate = kernfs_fop_readdir,
1531 .release = kernfs_dir_fop_release,
1532 .llseek = kernfs_dir_fop_llseek,