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);
117 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
118 * @kn: kernfs_node of interest
120 * This function can be called from any context.
122 void pr_cont_kernfs_name(struct kernfs_node *kn)
126 spin_lock_irqsave(&kernfs_rename_lock, flags);
128 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
129 pr_cont("%s", kernfs_pr_cont_buf);
131 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
135 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
136 * @kn: kernfs_node of interest
138 * This function can be called from any context.
140 void pr_cont_kernfs_path(struct kernfs_node *kn)
145 spin_lock_irqsave(&kernfs_rename_lock, flags);
147 p = kernfs_path_locked(kn, kernfs_pr_cont_buf,
148 sizeof(kernfs_pr_cont_buf));
152 pr_cont("<name too long>");
154 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
158 * kernfs_get_parent - determine the parent node and pin it
159 * @kn: kernfs_node of interest
161 * Determines @kn's parent, pins and returns it. This function can be
162 * called from any context.
164 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
166 struct kernfs_node *parent;
169 spin_lock_irqsave(&kernfs_rename_lock, flags);
172 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
179 * @name: Null terminated string to hash
180 * @ns: Namespace tag to hash
182 * Returns 31 bit hash of ns + name (so it fits in an off_t )
184 static unsigned int kernfs_name_hash(const char *name, const void *ns)
186 unsigned long hash = init_name_hash();
187 unsigned int len = strlen(name);
189 hash = partial_name_hash(*name++, hash);
190 hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31));
192 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
200 static int kernfs_name_compare(unsigned int hash, const char *name,
201 const void *ns, const struct kernfs_node *kn)
203 if (hash != kn->hash)
204 return hash - kn->hash;
207 return strcmp(name, kn->name);
210 static int kernfs_sd_compare(const struct kernfs_node *left,
211 const struct kernfs_node *right)
213 return kernfs_name_compare(left->hash, left->name, left->ns, right);
217 * kernfs_link_sibling - link kernfs_node into sibling rbtree
218 * @kn: kernfs_node of interest
220 * Link @kn into its sibling rbtree which starts from
221 * @kn->parent->dir.children.
224 * mutex_lock(kernfs_mutex)
227 * 0 on susccess -EEXIST on failure.
229 static int kernfs_link_sibling(struct kernfs_node *kn)
231 struct rb_node **node = &kn->parent->dir.children.rb_node;
232 struct rb_node *parent = NULL;
234 if (kernfs_type(kn) == KERNFS_DIR)
235 kn->parent->dir.subdirs++;
238 struct kernfs_node *pos;
241 pos = rb_to_kn(*node);
243 result = kernfs_sd_compare(kn, pos);
245 node = &pos->rb.rb_left;
247 node = &pos->rb.rb_right;
251 /* add new node and rebalance the tree */
252 rb_link_node(&kn->rb, parent, node);
253 rb_insert_color(&kn->rb, &kn->parent->dir.children);
258 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
259 * @kn: kernfs_node of interest
261 * Try to unlink @kn from its sibling rbtree which starts from
262 * kn->parent->dir.children. Returns %true if @kn was actually
263 * removed, %false if @kn wasn't on the rbtree.
266 * mutex_lock(kernfs_mutex)
268 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
270 if (RB_EMPTY_NODE(&kn->rb))
273 if (kernfs_type(kn) == KERNFS_DIR)
274 kn->parent->dir.subdirs--;
276 rb_erase(&kn->rb, &kn->parent->dir.children);
277 RB_CLEAR_NODE(&kn->rb);
282 * kernfs_get_active - get an active reference to kernfs_node
283 * @kn: kernfs_node to get an active reference to
285 * Get an active reference of @kn. This function is noop if @kn
289 * Pointer to @kn on success, NULL on failure.
291 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
296 if (!atomic_inc_unless_negative(&kn->active))
299 if (kernfs_lockdep(kn))
300 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
305 * kernfs_put_active - put an active reference to kernfs_node
306 * @kn: kernfs_node to put an active reference to
308 * Put an active reference to @kn. This function is noop if @kn
311 void kernfs_put_active(struct kernfs_node *kn)
313 struct kernfs_root *root = kernfs_root(kn);
319 if (kernfs_lockdep(kn))
320 rwsem_release(&kn->dep_map, 1, _RET_IP_);
321 v = atomic_dec_return(&kn->active);
322 if (likely(v != KN_DEACTIVATED_BIAS))
325 wake_up_all(&root->deactivate_waitq);
329 * kernfs_drain - drain kernfs_node
330 * @kn: kernfs_node to drain
332 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
333 * removers may invoke this function concurrently on @kn and all will
334 * return after draining is complete.
336 static void kernfs_drain(struct kernfs_node *kn)
337 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
339 struct kernfs_root *root = kernfs_root(kn);
341 lockdep_assert_held(&kernfs_mutex);
342 WARN_ON_ONCE(kernfs_active(kn));
344 mutex_unlock(&kernfs_mutex);
346 if (kernfs_lockdep(kn)) {
347 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
348 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
349 lock_contended(&kn->dep_map, _RET_IP_);
352 /* but everyone should wait for draining */
353 wait_event(root->deactivate_waitq,
354 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
356 if (kernfs_lockdep(kn)) {
357 lock_acquired(&kn->dep_map, _RET_IP_);
358 rwsem_release(&kn->dep_map, 1, _RET_IP_);
361 kernfs_unmap_bin_file(kn);
363 mutex_lock(&kernfs_mutex);
367 * kernfs_get - get a reference count on a kernfs_node
368 * @kn: the target kernfs_node
370 void kernfs_get(struct kernfs_node *kn)
373 WARN_ON(!atomic_read(&kn->count));
374 atomic_inc(&kn->count);
377 EXPORT_SYMBOL_GPL(kernfs_get);
380 * kernfs_put - put a reference count on a kernfs_node
381 * @kn: the target kernfs_node
383 * Put a reference count of @kn and destroy it if it reached zero.
385 void kernfs_put(struct kernfs_node *kn)
387 struct kernfs_node *parent;
388 struct kernfs_root *root;
390 if (!kn || !atomic_dec_and_test(&kn->count))
392 root = kernfs_root(kn);
395 * Moving/renaming is always done while holding reference.
396 * kn->parent won't change beneath us.
400 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
401 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
402 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
404 if (kernfs_type(kn) == KERNFS_LINK)
405 kernfs_put(kn->symlink.target_kn);
406 if (!(kn->flags & KERNFS_STATIC_NAME))
409 if (kn->iattr->ia_secdata)
410 security_release_secctx(kn->iattr->ia_secdata,
411 kn->iattr->ia_secdata_len);
412 simple_xattrs_free(&kn->iattr->xattrs);
415 ida_simple_remove(&root->ino_ida, kn->ino);
416 kmem_cache_free(kernfs_node_cache, kn);
420 if (atomic_dec_and_test(&kn->count))
423 /* just released the root kn, free @root too */
424 ida_destroy(&root->ino_ida);
428 EXPORT_SYMBOL_GPL(kernfs_put);
430 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
432 struct kernfs_node *kn;
434 if (flags & LOOKUP_RCU)
437 /* Always perform fresh lookup for negatives */
438 if (!dentry->d_inode)
439 goto out_bad_unlocked;
441 kn = dentry->d_fsdata;
442 mutex_lock(&kernfs_mutex);
444 /* The kernfs node has been deactivated */
445 if (!kernfs_active(kn))
448 /* The kernfs node has been moved? */
449 if (dentry->d_parent->d_fsdata != kn->parent)
452 /* The kernfs node has been renamed */
453 if (strcmp(dentry->d_name.name, kn->name) != 0)
456 /* The kernfs node has been moved to a different namespace */
457 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
458 kernfs_info(dentry->d_sb)->ns != kn->ns)
461 mutex_unlock(&kernfs_mutex);
465 mutex_unlock(&kernfs_mutex);
468 * @dentry doesn't match the underlying kernfs node, drop the
469 * dentry and force lookup. If we have submounts we must allow the
470 * vfs caches to lie about the state of the filesystem to prevent
471 * leaks and other nasty things, so use check_submounts_and_drop()
472 * instead of d_drop().
474 if (check_submounts_and_drop(dentry) != 0)
480 static void kernfs_dop_release(struct dentry *dentry)
482 kernfs_put(dentry->d_fsdata);
485 const struct dentry_operations kernfs_dops = {
486 .d_revalidate = kernfs_dop_revalidate,
487 .d_release = kernfs_dop_release,
491 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
492 * @dentry: the dentry in question
494 * Return the kernfs_node associated with @dentry. If @dentry is not a
495 * kernfs one, %NULL is returned.
497 * While the returned kernfs_node will stay accessible as long as @dentry
498 * is accessible, the returned node can be in any state and the caller is
499 * fully responsible for determining what's accessible.
501 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
503 if (dentry->d_sb->s_op == &kernfs_sops)
504 return dentry->d_fsdata;
508 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
509 const char *name, umode_t mode,
512 char *dup_name = NULL;
513 struct kernfs_node *kn;
516 if (!(flags & KERNFS_STATIC_NAME)) {
517 name = dup_name = kstrdup(name, GFP_KERNEL);
522 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
526 ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
531 atomic_set(&kn->count, 1);
532 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
533 RB_CLEAR_NODE(&kn->rb);
542 kmem_cache_free(kernfs_node_cache, kn);
548 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
549 const char *name, umode_t mode,
552 struct kernfs_node *kn;
554 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
563 * kernfs_add_one - add kernfs_node to parent without warning
564 * @kn: kernfs_node to be added
566 * The caller must already have initialized @kn->parent. This
567 * function increments nlink of the parent's inode if @kn is a
568 * directory and link into the children list of the parent.
571 * 0 on success, -EEXIST if entry with the given name already
574 int kernfs_add_one(struct kernfs_node *kn)
576 struct kernfs_node *parent = kn->parent;
577 struct kernfs_iattrs *ps_iattr;
581 mutex_lock(&kernfs_mutex);
584 has_ns = kernfs_ns_enabled(parent);
585 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
586 has_ns ? "required" : "invalid", parent->name, kn->name))
589 if (kernfs_type(parent) != KERNFS_DIR)
593 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
596 kn->hash = kernfs_name_hash(kn->name, kn->ns);
598 ret = kernfs_link_sibling(kn);
602 /* Update timestamps on the parent */
603 ps_iattr = parent->iattr;
605 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
606 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
609 mutex_unlock(&kernfs_mutex);
612 * Activate the new node unless CREATE_DEACTIVATED is requested.
613 * If not activated here, the kernfs user is responsible for
614 * activating the node with kernfs_activate(). A node which hasn't
615 * been activated is not visible to userland and its removal won't
616 * trigger deactivation.
618 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
623 mutex_unlock(&kernfs_mutex);
628 * kernfs_find_ns - find kernfs_node with the given name
629 * @parent: kernfs_node to search under
630 * @name: name to look for
631 * @ns: the namespace tag to use
633 * Look for kernfs_node with name @name under @parent. Returns pointer to
634 * the found kernfs_node on success, %NULL on failure.
636 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
637 const unsigned char *name,
640 struct rb_node *node = parent->dir.children.rb_node;
641 bool has_ns = kernfs_ns_enabled(parent);
644 lockdep_assert_held(&kernfs_mutex);
646 if (has_ns != (bool)ns) {
647 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
648 has_ns ? "required" : "invalid", parent->name, name);
652 hash = kernfs_name_hash(name, ns);
654 struct kernfs_node *kn;
658 result = kernfs_name_compare(hash, name, ns, kn);
660 node = node->rb_left;
662 node = node->rb_right;
670 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
671 * @parent: kernfs_node to search under
672 * @name: name to look for
673 * @ns: the namespace tag to use
675 * Look for kernfs_node with name @name under @parent and get a reference
676 * if found. This function may sleep and returns pointer to the found
677 * kernfs_node on success, %NULL on failure.
679 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
680 const char *name, const void *ns)
682 struct kernfs_node *kn;
684 mutex_lock(&kernfs_mutex);
685 kn = kernfs_find_ns(parent, name, ns);
687 mutex_unlock(&kernfs_mutex);
691 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
694 * kernfs_create_root - create a new kernfs hierarchy
695 * @scops: optional syscall operations for the hierarchy
696 * @flags: KERNFS_ROOT_* flags
697 * @priv: opaque data associated with the new directory
699 * Returns the root of the new hierarchy on success, ERR_PTR() value on
702 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
703 unsigned int flags, void *priv)
705 struct kernfs_root *root;
706 struct kernfs_node *kn;
708 root = kzalloc(sizeof(*root), GFP_KERNEL);
710 return ERR_PTR(-ENOMEM);
712 ida_init(&root->ino_ida);
714 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
717 ida_destroy(&root->ino_ida);
719 return ERR_PTR(-ENOMEM);
725 root->syscall_ops = scops;
728 init_waitqueue_head(&root->deactivate_waitq);
730 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
737 * kernfs_destroy_root - destroy a kernfs hierarchy
738 * @root: root of the hierarchy to destroy
740 * Destroy the hierarchy anchored at @root by removing all existing
741 * directories and destroying @root.
743 void kernfs_destroy_root(struct kernfs_root *root)
745 kernfs_remove(root->kn); /* will also free @root */
749 * kernfs_create_dir_ns - create a directory
750 * @parent: parent in which to create a new directory
751 * @name: name of the new directory
752 * @mode: mode of the new directory
753 * @priv: opaque data associated with the new directory
754 * @ns: optional namespace tag of the directory
756 * Returns the created node on success, ERR_PTR() value on failure.
758 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
759 const char *name, umode_t mode,
760 void *priv, const void *ns)
762 struct kernfs_node *kn;
766 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
768 return ERR_PTR(-ENOMEM);
770 kn->dir.root = parent->dir.root;
775 rc = kernfs_add_one(kn);
783 static struct dentry *kernfs_iop_lookup(struct inode *dir,
784 struct dentry *dentry,
788 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
789 struct kernfs_node *kn;
791 const void *ns = NULL;
793 mutex_lock(&kernfs_mutex);
795 if (kernfs_ns_enabled(parent))
796 ns = kernfs_info(dir->i_sb)->ns;
798 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
801 if (!kn || !kernfs_active(kn)) {
806 dentry->d_fsdata = kn;
808 /* attach dentry and inode */
809 inode = kernfs_get_inode(dir->i_sb, kn);
811 ret = ERR_PTR(-ENOMEM);
815 /* instantiate and hash dentry */
816 ret = d_materialise_unique(dentry, inode);
818 mutex_unlock(&kernfs_mutex);
822 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
825 struct kernfs_node *parent = dir->i_private;
826 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
829 if (!scops || !scops->mkdir)
832 if (!kernfs_get_active(parent))
835 ret = scops->mkdir(parent, dentry->d_name.name, mode);
837 kernfs_put_active(parent);
841 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
843 struct kernfs_node *kn = dentry->d_fsdata;
844 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
847 if (!scops || !scops->rmdir)
850 if (!kernfs_get_active(kn))
853 ret = scops->rmdir(kn);
855 kernfs_put_active(kn);
859 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
860 struct inode *new_dir, struct dentry *new_dentry)
862 struct kernfs_node *kn = old_dentry->d_fsdata;
863 struct kernfs_node *new_parent = new_dir->i_private;
864 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
867 if (!scops || !scops->rename)
870 if (!kernfs_get_active(kn))
873 if (!kernfs_get_active(new_parent)) {
874 kernfs_put_active(kn);
878 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
880 kernfs_put_active(new_parent);
881 kernfs_put_active(kn);
885 const struct inode_operations kernfs_dir_iops = {
886 .lookup = kernfs_iop_lookup,
887 .permission = kernfs_iop_permission,
888 .setattr = kernfs_iop_setattr,
889 .getattr = kernfs_iop_getattr,
890 .setxattr = kernfs_iop_setxattr,
891 .removexattr = kernfs_iop_removexattr,
892 .getxattr = kernfs_iop_getxattr,
893 .listxattr = kernfs_iop_listxattr,
895 .mkdir = kernfs_iop_mkdir,
896 .rmdir = kernfs_iop_rmdir,
897 .rename = kernfs_iop_rename,
900 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
902 struct kernfs_node *last;
909 if (kernfs_type(pos) != KERNFS_DIR)
912 rbn = rb_first(&pos->dir.children);
923 * kernfs_next_descendant_post - find the next descendant for post-order walk
924 * @pos: the current position (%NULL to initiate traversal)
925 * @root: kernfs_node whose descendants to walk
927 * Find the next descendant to visit for post-order traversal of @root's
928 * descendants. @root is included in the iteration and the last node to be
931 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
932 struct kernfs_node *root)
936 lockdep_assert_held(&kernfs_mutex);
938 /* if first iteration, visit leftmost descendant which may be root */
940 return kernfs_leftmost_descendant(root);
942 /* if we visited @root, we're done */
946 /* if there's an unvisited sibling, visit its leftmost descendant */
947 rbn = rb_next(&pos->rb);
949 return kernfs_leftmost_descendant(rb_to_kn(rbn));
951 /* no sibling left, visit parent */
956 * kernfs_activate - activate a node which started deactivated
957 * @kn: kernfs_node whose subtree is to be activated
959 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
960 * needs to be explicitly activated. A node which hasn't been activated
961 * isn't visible to userland and deactivation is skipped during its
962 * removal. This is useful to construct atomic init sequences where
963 * creation of multiple nodes should either succeed or fail atomically.
965 * The caller is responsible for ensuring that this function is not called
966 * after kernfs_remove*() is invoked on @kn.
968 void kernfs_activate(struct kernfs_node *kn)
970 struct kernfs_node *pos;
972 mutex_lock(&kernfs_mutex);
975 while ((pos = kernfs_next_descendant_post(pos, kn))) {
976 if (!pos || (pos->flags & KERNFS_ACTIVATED))
979 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
980 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
982 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
983 pos->flags |= KERNFS_ACTIVATED;
986 mutex_unlock(&kernfs_mutex);
989 static void __kernfs_remove(struct kernfs_node *kn)
991 struct kernfs_node *pos;
993 lockdep_assert_held(&kernfs_mutex);
996 * Short-circuit if non-root @kn has already finished removal.
997 * This is for kernfs_remove_self() which plays with active ref
1000 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1003 pr_debug("kernfs %s: removing\n", kn->name);
1005 /* prevent any new usage under @kn by deactivating all nodes */
1007 while ((pos = kernfs_next_descendant_post(pos, kn)))
1008 if (kernfs_active(pos))
1009 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1011 /* deactivate and unlink the subtree node-by-node */
1013 pos = kernfs_leftmost_descendant(kn);
1016 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1017 * base ref could have been put by someone else by the time
1018 * the function returns. Make sure it doesn't go away
1024 * Drain iff @kn was activated. This avoids draining and
1025 * its lockdep annotations for nodes which have never been
1026 * activated and allows embedding kernfs_remove() in create
1027 * error paths without worrying about draining.
1029 if (kn->flags & KERNFS_ACTIVATED)
1032 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1035 * kernfs_unlink_sibling() succeeds once per node. Use it
1036 * to decide who's responsible for cleanups.
1038 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1039 struct kernfs_iattrs *ps_iattr =
1040 pos->parent ? pos->parent->iattr : NULL;
1042 /* update timestamps on the parent */
1044 ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME;
1045 ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME;
1052 } while (pos != kn);
1056 * kernfs_remove - remove a kernfs_node recursively
1057 * @kn: the kernfs_node to remove
1059 * Remove @kn along with all its subdirectories and files.
1061 void kernfs_remove(struct kernfs_node *kn)
1063 mutex_lock(&kernfs_mutex);
1064 __kernfs_remove(kn);
1065 mutex_unlock(&kernfs_mutex);
1069 * kernfs_break_active_protection - break out of active protection
1070 * @kn: the self kernfs_node
1072 * The caller must be running off of a kernfs operation which is invoked
1073 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1074 * this function must also be matched with an invocation of
1075 * kernfs_unbreak_active_protection().
1077 * This function releases the active reference of @kn the caller is
1078 * holding. Once this function is called, @kn may be removed at any point
1079 * and the caller is solely responsible for ensuring that the objects it
1080 * dereferences are accessible.
1082 void kernfs_break_active_protection(struct kernfs_node *kn)
1085 * Take out ourself out of the active ref dependency chain. If
1086 * we're called without an active ref, lockdep will complain.
1088 kernfs_put_active(kn);
1092 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1093 * @kn: the self kernfs_node
1095 * If kernfs_break_active_protection() was called, this function must be
1096 * invoked before finishing the kernfs operation. Note that while this
1097 * function restores the active reference, it doesn't and can't actually
1098 * restore the active protection - @kn may already or be in the process of
1099 * being removed. Once kernfs_break_active_protection() is invoked, that
1100 * protection is irreversibly gone for the kernfs operation instance.
1102 * While this function may be called at any point after
1103 * kernfs_break_active_protection() is invoked, its most useful location
1104 * would be right before the enclosing kernfs operation returns.
1106 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1109 * @kn->active could be in any state; however, the increment we do
1110 * here will be undone as soon as the enclosing kernfs operation
1111 * finishes and this temporary bump can't break anything. If @kn
1112 * is alive, nothing changes. If @kn is being deactivated, the
1113 * soon-to-follow put will either finish deactivation or restore
1114 * deactivated state. If @kn is already removed, the temporary
1115 * bump is guaranteed to be gone before @kn is released.
1117 atomic_inc(&kn->active);
1118 if (kernfs_lockdep(kn))
1119 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1123 * kernfs_remove_self - remove a kernfs_node from its own method
1124 * @kn: the self kernfs_node to remove
1126 * The caller must be running off of a kernfs operation which is invoked
1127 * with an active reference - e.g. one of kernfs_ops. This can be used to
1128 * implement a file operation which deletes itself.
1130 * For example, the "delete" file for a sysfs device directory can be
1131 * implemented by invoking kernfs_remove_self() on the "delete" file
1132 * itself. This function breaks the circular dependency of trying to
1133 * deactivate self while holding an active ref itself. It isn't necessary
1134 * to modify the usual removal path to use kernfs_remove_self(). The
1135 * "delete" implementation can simply invoke kernfs_remove_self() on self
1136 * before proceeding with the usual removal path. kernfs will ignore later
1137 * kernfs_remove() on self.
1139 * kernfs_remove_self() can be called multiple times concurrently on the
1140 * same kernfs_node. Only the first one actually performs removal and
1141 * returns %true. All others will wait until the kernfs operation which
1142 * won self-removal finishes and return %false. Note that the losers wait
1143 * for the completion of not only the winning kernfs_remove_self() but also
1144 * the whole kernfs_ops which won the arbitration. This can be used to
1145 * guarantee, for example, all concurrent writes to a "delete" file to
1146 * finish only after the whole operation is complete.
1148 bool kernfs_remove_self(struct kernfs_node *kn)
1152 mutex_lock(&kernfs_mutex);
1153 kernfs_break_active_protection(kn);
1156 * SUICIDAL is used to arbitrate among competing invocations. Only
1157 * the first one will actually perform removal. When the removal
1158 * is complete, SUICIDED is set and the active ref is restored
1159 * while holding kernfs_mutex. The ones which lost arbitration
1160 * waits for SUICDED && drained which can happen only after the
1161 * enclosing kernfs operation which executed the winning instance
1162 * of kernfs_remove_self() finished.
1164 if (!(kn->flags & KERNFS_SUICIDAL)) {
1165 kn->flags |= KERNFS_SUICIDAL;
1166 __kernfs_remove(kn);
1167 kn->flags |= KERNFS_SUICIDED;
1170 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1174 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1176 if ((kn->flags & KERNFS_SUICIDED) &&
1177 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1180 mutex_unlock(&kernfs_mutex);
1182 mutex_lock(&kernfs_mutex);
1184 finish_wait(waitq, &wait);
1185 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1190 * This must be done while holding kernfs_mutex; otherwise, waiting
1191 * for SUICIDED && deactivated could finish prematurely.
1193 kernfs_unbreak_active_protection(kn);
1195 mutex_unlock(&kernfs_mutex);
1200 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1201 * @parent: parent of the target
1202 * @name: name of the kernfs_node to remove
1203 * @ns: namespace tag of the kernfs_node to remove
1205 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1206 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1208 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1211 struct kernfs_node *kn;
1214 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1219 mutex_lock(&kernfs_mutex);
1221 kn = kernfs_find_ns(parent, name, ns);
1223 __kernfs_remove(kn);
1225 mutex_unlock(&kernfs_mutex);
1234 * kernfs_rename_ns - move and rename a kernfs_node
1236 * @new_parent: new parent to put @sd under
1237 * @new_name: new name
1238 * @new_ns: new namespace tag
1240 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1241 const char *new_name, const void *new_ns)
1243 struct kernfs_node *old_parent;
1244 const char *old_name = NULL;
1247 /* can't move or rename root */
1251 mutex_lock(&kernfs_mutex);
1254 if (!kernfs_active(kn) || !kernfs_active(new_parent))
1258 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1259 (strcmp(kn->name, new_name) == 0))
1260 goto out; /* nothing to rename */
1263 if (kernfs_find_ns(new_parent, new_name, new_ns))
1266 /* rename kernfs_node */
1267 if (strcmp(kn->name, new_name) != 0) {
1269 new_name = kstrdup(new_name, GFP_KERNEL);
1277 * Move to the appropriate place in the appropriate directories rbtree.
1279 kernfs_unlink_sibling(kn);
1280 kernfs_get(new_parent);
1282 /* rename_lock protects ->parent and ->name accessors */
1283 spin_lock_irq(&kernfs_rename_lock);
1285 old_parent = kn->parent;
1286 kn->parent = new_parent;
1290 if (!(kn->flags & KERNFS_STATIC_NAME))
1291 old_name = kn->name;
1292 kn->flags &= ~KERNFS_STATIC_NAME;
1293 kn->name = new_name;
1296 spin_unlock_irq(&kernfs_rename_lock);
1298 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1299 kernfs_link_sibling(kn);
1301 kernfs_put(old_parent);
1306 mutex_unlock(&kernfs_mutex);
1310 /* Relationship between s_mode and the DT_xxx types */
1311 static inline unsigned char dt_type(struct kernfs_node *kn)
1313 return (kn->mode >> 12) & 15;
1316 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1318 kernfs_put(filp->private_data);
1322 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1323 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1326 int valid = kernfs_active(pos) &&
1327 pos->parent == parent && hash == pos->hash;
1332 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1333 struct rb_node *node = parent->dir.children.rb_node;
1335 pos = rb_to_kn(node);
1337 if (hash < pos->hash)
1338 node = node->rb_left;
1339 else if (hash > pos->hash)
1340 node = node->rb_right;
1345 /* Skip over entries which are dying/dead or in the wrong namespace */
1346 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1347 struct rb_node *node = rb_next(&pos->rb);
1351 pos = rb_to_kn(node);
1356 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1357 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1359 pos = kernfs_dir_pos(ns, parent, ino, pos);
1362 struct rb_node *node = rb_next(&pos->rb);
1366 pos = rb_to_kn(node);
1367 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1372 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1374 struct dentry *dentry = file->f_path.dentry;
1375 struct kernfs_node *parent = dentry->d_fsdata;
1376 struct kernfs_node *pos = file->private_data;
1377 const void *ns = NULL;
1379 if (!dir_emit_dots(file, ctx))
1381 mutex_lock(&kernfs_mutex);
1383 if (kernfs_ns_enabled(parent))
1384 ns = kernfs_info(dentry->d_sb)->ns;
1386 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1388 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1389 const char *name = pos->name;
1390 unsigned int type = dt_type(pos);
1391 int len = strlen(name);
1392 ino_t ino = pos->ino;
1394 ctx->pos = pos->hash;
1395 file->private_data = pos;
1398 mutex_unlock(&kernfs_mutex);
1399 if (!dir_emit(ctx, name, len, ino, type))
1401 mutex_lock(&kernfs_mutex);
1403 mutex_unlock(&kernfs_mutex);
1404 file->private_data = NULL;
1409 static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset,
1412 struct inode *inode = file_inode(file);
1415 mutex_lock(&inode->i_mutex);
1416 ret = generic_file_llseek(file, offset, whence);
1417 mutex_unlock(&inode->i_mutex);
1422 const struct file_operations kernfs_dir_fops = {
1423 .read = generic_read_dir,
1424 .iterate = kernfs_fop_readdir,
1425 .release = kernfs_dir_fop_release,
1426 .llseek = kernfs_dir_fop_llseek,