1 // SPDX-License-Identifier: GPL-2.0-only
3 * fs/kernfs/dir.c - kernfs directory implementation
5 * Copyright (c) 2001-3 Patrick Mochel
6 * Copyright (c) 2007 SUSE Linux Products GmbH
7 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
10 #include <linux/sched.h>
12 #include <linux/namei.h>
13 #include <linux/idr.h>
14 #include <linux/slab.h>
15 #include <linux/security.h>
16 #include <linux/hash.h>
18 #include "kernfs-internal.h"
20 DEFINE_MUTEX(kernfs_mutex);
21 static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
22 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */
23 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
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)
45 return strlcpy(buf, "(null)", buflen);
47 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
50 /* kernfs_node_depth - compute depth from @from to @to */
51 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
55 while (to->parent && to != from) {
62 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
63 struct kernfs_node *b)
66 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
71 da = kernfs_depth(ra->kn, a);
72 db = kernfs_depth(rb->kn, b);
83 /* worst case b and a will be the same at root */
93 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
94 * where kn_from is treated as root of the path.
95 * @kn_from: kernfs node which should be treated as root for the path
96 * @kn_to: kernfs node to which path is needed
97 * @buf: buffer to copy the path into
98 * @buflen: size of @buf
100 * We need to handle couple of scenarios here:
101 * [1] when @kn_from is an ancestor of @kn_to at some level
103 * kn_to: /n1/n2/n3/n4/n5
106 * [2] when @kn_from is on a different hierarchy and we need to find common
107 * ancestor between @kn_from and @kn_to.
108 * kn_from: /n1/n2/n3/n4
112 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
113 * kn_to: /n1/n2/n3 [depth=3]
116 * [3] when @kn_to is NULL result will be "(null)"
118 * Returns the length of the full path. If the full length is equal to or
119 * greater than @buflen, @buf contains the truncated path with the trailing
120 * '\0'. On error, -errno is returned.
122 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
123 struct kernfs_node *kn_from,
124 char *buf, size_t buflen)
126 struct kernfs_node *kn, *common;
127 const char parent_str[] = "/..";
128 size_t depth_from, depth_to, len = 0;
132 return strlcpy(buf, "(null)", buflen);
135 kn_from = kernfs_root(kn_to)->kn;
137 if (kn_from == kn_to)
138 return strlcpy(buf, "/", buflen);
140 common = kernfs_common_ancestor(kn_from, kn_to);
141 if (WARN_ON(!common))
144 depth_to = kernfs_depth(common, kn_to);
145 depth_from = kernfs_depth(common, kn_from);
150 for (i = 0; i < depth_from; i++)
151 len += strlcpy(buf + len, parent_str,
152 len < buflen ? buflen - len : 0);
154 /* Calculate how many bytes we need for the rest */
155 for (i = depth_to - 1; i >= 0; i--) {
156 for (kn = kn_to, j = 0; j < i; j++)
158 len += strlcpy(buf + len, "/",
159 len < buflen ? buflen - len : 0);
160 len += strlcpy(buf + len, kn->name,
161 len < buflen ? buflen - len : 0);
168 * kernfs_name - obtain the name of a given node
169 * @kn: kernfs_node of interest
170 * @buf: buffer to copy @kn's name into
171 * @buflen: size of @buf
173 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
174 * similar to strlcpy(). It returns the length of @kn's name and if @buf
175 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
177 * Fills buffer with "(null)" if @kn is NULL.
179 * This function can be called from any context.
181 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
186 spin_lock_irqsave(&kernfs_rename_lock, flags);
187 ret = kernfs_name_locked(kn, buf, buflen);
188 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
193 * kernfs_path_from_node - build path of node @to relative to @from.
194 * @from: parent kernfs_node relative to which we need to build the path
195 * @to: kernfs_node of interest
196 * @buf: buffer to copy @to's path into
197 * @buflen: size of @buf
199 * Builds @to's path relative to @from in @buf. @from and @to must
200 * be on the same kernfs-root. If @from is not parent of @to, then a relative
201 * path (which includes '..'s) as needed to reach from @from to @to is
204 * Returns the length of the full path. If the full length is equal to or
205 * greater than @buflen, @buf contains the truncated path with the trailing
206 * '\0'. On error, -errno is returned.
208 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
209 char *buf, size_t buflen)
214 spin_lock_irqsave(&kernfs_rename_lock, flags);
215 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
216 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
219 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
222 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
223 * @kn: kernfs_node of interest
225 * This function can be called from any context.
227 void pr_cont_kernfs_name(struct kernfs_node *kn)
231 spin_lock_irqsave(&kernfs_rename_lock, flags);
233 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
234 pr_cont("%s", kernfs_pr_cont_buf);
236 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
240 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
241 * @kn: kernfs_node of interest
243 * This function can be called from any context.
245 void pr_cont_kernfs_path(struct kernfs_node *kn)
250 spin_lock_irqsave(&kernfs_rename_lock, flags);
252 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
253 sizeof(kernfs_pr_cont_buf));
259 if (sz >= sizeof(kernfs_pr_cont_buf)) {
260 pr_cont("(name too long)");
264 pr_cont("%s", kernfs_pr_cont_buf);
267 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
271 * kernfs_get_parent - determine the parent node and pin it
272 * @kn: kernfs_node of interest
274 * Determines @kn's parent, pins and returns it. This function can be
275 * called from any context.
277 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
279 struct kernfs_node *parent;
282 spin_lock_irqsave(&kernfs_rename_lock, flags);
285 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
292 * @name: Null terminated string to hash
293 * @ns: Namespace tag to hash
295 * Returns 31 bit hash of ns + name (so it fits in an off_t )
297 static unsigned int kernfs_name_hash(const char *name, const void *ns)
299 unsigned long hash = init_name_hash(ns);
300 unsigned int len = strlen(name);
302 hash = partial_name_hash(*name++, hash);
303 hash = end_name_hash(hash);
305 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
313 static int kernfs_name_compare(unsigned int hash, const char *name,
314 const void *ns, const struct kernfs_node *kn)
324 return strcmp(name, kn->name);
327 static int kernfs_sd_compare(const struct kernfs_node *left,
328 const struct kernfs_node *right)
330 return kernfs_name_compare(left->hash, left->name, left->ns, right);
334 * kernfs_link_sibling - link kernfs_node into sibling rbtree
335 * @kn: kernfs_node of interest
337 * Link @kn into its sibling rbtree which starts from
338 * @kn->parent->dir.children.
341 * mutex_lock(kernfs_mutex)
344 * 0 on susccess -EEXIST on failure.
346 static int kernfs_link_sibling(struct kernfs_node *kn)
348 struct rb_node **node = &kn->parent->dir.children.rb_node;
349 struct rb_node *parent = NULL;
352 struct kernfs_node *pos;
355 pos = rb_to_kn(*node);
357 result = kernfs_sd_compare(kn, pos);
359 node = &pos->rb.rb_left;
361 node = &pos->rb.rb_right;
366 /* add new node and rebalance the tree */
367 rb_link_node(&kn->rb, parent, node);
368 rb_insert_color(&kn->rb, &kn->parent->dir.children);
370 /* successfully added, account subdir number */
371 if (kernfs_type(kn) == KERNFS_DIR)
372 kn->parent->dir.subdirs++;
378 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
379 * @kn: kernfs_node of interest
381 * Try to unlink @kn from its sibling rbtree which starts from
382 * kn->parent->dir.children. Returns %true if @kn was actually
383 * removed, %false if @kn wasn't on the rbtree.
386 * mutex_lock(kernfs_mutex)
388 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
390 if (RB_EMPTY_NODE(&kn->rb))
393 if (kernfs_type(kn) == KERNFS_DIR)
394 kn->parent->dir.subdirs--;
396 rb_erase(&kn->rb, &kn->parent->dir.children);
397 RB_CLEAR_NODE(&kn->rb);
402 * kernfs_get_active - get an active reference to kernfs_node
403 * @kn: kernfs_node to get an active reference to
405 * Get an active reference of @kn. This function is noop if @kn
409 * Pointer to @kn on success, NULL on failure.
411 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
416 if (!atomic_inc_unless_negative(&kn->active))
419 if (kernfs_lockdep(kn))
420 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
425 * kernfs_put_active - put an active reference to kernfs_node
426 * @kn: kernfs_node to put an active reference to
428 * Put an active reference to @kn. This function is noop if @kn
431 void kernfs_put_active(struct kernfs_node *kn)
438 if (kernfs_lockdep(kn))
439 rwsem_release(&kn->dep_map, 1, _RET_IP_);
440 v = atomic_dec_return(&kn->active);
441 if (likely(v != KN_DEACTIVATED_BIAS))
444 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
448 * kernfs_drain - drain kernfs_node
449 * @kn: kernfs_node to drain
451 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
452 * removers may invoke this function concurrently on @kn and all will
453 * return after draining is complete.
455 static void kernfs_drain(struct kernfs_node *kn)
456 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
458 struct kernfs_root *root = kernfs_root(kn);
460 lockdep_assert_held(&kernfs_mutex);
461 WARN_ON_ONCE(kernfs_active(kn));
463 mutex_unlock(&kernfs_mutex);
465 if (kernfs_lockdep(kn)) {
466 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
467 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
468 lock_contended(&kn->dep_map, _RET_IP_);
471 /* but everyone should wait for draining */
472 wait_event(root->deactivate_waitq,
473 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
475 if (kernfs_lockdep(kn)) {
476 lock_acquired(&kn->dep_map, _RET_IP_);
477 rwsem_release(&kn->dep_map, 1, _RET_IP_);
480 kernfs_drain_open_files(kn);
482 mutex_lock(&kernfs_mutex);
486 * kernfs_get - get a reference count on a kernfs_node
487 * @kn: the target kernfs_node
489 void kernfs_get(struct kernfs_node *kn)
492 WARN_ON(!atomic_read(&kn->count));
493 atomic_inc(&kn->count);
496 EXPORT_SYMBOL_GPL(kernfs_get);
499 * kernfs_put - put a reference count on a kernfs_node
500 * @kn: the target kernfs_node
502 * Put a reference count of @kn and destroy it if it reached zero.
504 void kernfs_put(struct kernfs_node *kn)
506 struct kernfs_node *parent;
507 struct kernfs_root *root;
510 * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
511 * depends on this to filter reused stale node
513 if (!kn || !atomic_dec_and_test(&kn->count))
515 root = kernfs_root(kn);
518 * Moving/renaming is always done while holding reference.
519 * kn->parent won't change beneath us.
523 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
524 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
525 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
527 if (kernfs_type(kn) == KERNFS_LINK)
528 kernfs_put(kn->symlink.target_kn);
530 kfree_const(kn->name);
533 simple_xattrs_free(&kn->iattr->xattrs);
534 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
536 spin_lock(&kernfs_idr_lock);
537 idr_remove(&root->ino_idr, kn->id.ino);
538 spin_unlock(&kernfs_idr_lock);
539 kmem_cache_free(kernfs_node_cache, kn);
543 if (atomic_dec_and_test(&kn->count))
546 /* just released the root kn, free @root too */
547 idr_destroy(&root->ino_idr);
551 EXPORT_SYMBOL_GPL(kernfs_put);
553 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
555 struct kernfs_node *kn;
557 if (flags & LOOKUP_RCU)
560 /* Always perform fresh lookup for negatives */
561 if (d_really_is_negative(dentry))
562 goto out_bad_unlocked;
564 kn = kernfs_dentry_node(dentry);
565 mutex_lock(&kernfs_mutex);
567 /* The kernfs node has been deactivated */
568 if (!kernfs_active(kn))
571 /* The kernfs node has been moved? */
572 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
575 /* The kernfs node has been renamed */
576 if (strcmp(dentry->d_name.name, kn->name) != 0)
579 /* The kernfs node has been moved to a different namespace */
580 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
581 kernfs_info(dentry->d_sb)->ns != kn->ns)
584 mutex_unlock(&kernfs_mutex);
587 mutex_unlock(&kernfs_mutex);
592 const struct dentry_operations kernfs_dops = {
593 .d_revalidate = kernfs_dop_revalidate,
597 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
598 * @dentry: the dentry in question
600 * Return the kernfs_node associated with @dentry. If @dentry is not a
601 * kernfs one, %NULL is returned.
603 * While the returned kernfs_node will stay accessible as long as @dentry
604 * is accessible, the returned node can be in any state and the caller is
605 * fully responsible for determining what's accessible.
607 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
609 if (dentry->d_sb->s_op == &kernfs_sops &&
610 !d_really_is_negative(dentry))
611 return kernfs_dentry_node(dentry);
615 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
616 struct kernfs_node *parent,
617 const char *name, umode_t mode,
618 kuid_t uid, kgid_t gid,
621 struct kernfs_node *kn;
626 name = kstrdup_const(name, GFP_KERNEL);
630 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
634 idr_preload(GFP_KERNEL);
635 spin_lock(&kernfs_idr_lock);
636 cursor = idr_get_cursor(&root->ino_idr);
637 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
638 if (ret >= 0 && ret < cursor)
639 root->next_generation++;
640 gen = root->next_generation;
641 spin_unlock(&kernfs_idr_lock);
646 kn->id.generation = gen;
649 * set ino first. This RELEASE is paired with atomic_inc_not_zero in
650 * kernfs_find_and_get_node_by_ino
652 atomic_set_release(&kn->count, 1);
653 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
654 RB_CLEAR_NODE(&kn->rb);
660 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
661 struct iattr iattr = {
662 .ia_valid = ATTR_UID | ATTR_GID,
667 ret = __kernfs_setattr(kn, &iattr);
673 ret = security_kernfs_init_security(parent, kn);
681 idr_remove(&root->ino_idr, kn->id.ino);
683 kmem_cache_free(kernfs_node_cache, kn);
689 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
690 const char *name, umode_t mode,
691 kuid_t uid, kgid_t gid,
694 struct kernfs_node *kn;
696 kn = __kernfs_new_node(kernfs_root(parent), parent,
697 name, mode, uid, gid, flags);
706 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
707 * @root: the kernfs root
711 * NULL on failure. Return a kernfs node with reference counter incremented
713 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
716 struct kernfs_node *kn;
719 kn = idr_find(&root->ino_idr, ino);
724 * Since kernfs_node is freed in RCU, it's possible an old node for ino
725 * is freed, but reused before RCU grace period. But a freed node (see
726 * kernfs_put) or an incompletedly initialized node (see
727 * __kernfs_new_node) should have 'count' 0. We can use this fact to
728 * filter out such node.
730 if (!atomic_inc_not_zero(&kn->count)) {
736 * The node could be a new node or a reused node. If it's a new node,
737 * we are ok. If it's reused because of RCU (because of
738 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
739 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
740 * hence we can use 'ino' to filter stale node.
742 if (kn->id.ino != ino)
754 * kernfs_add_one - add kernfs_node to parent without warning
755 * @kn: kernfs_node to be added
757 * The caller must already have initialized @kn->parent. This
758 * function increments nlink of the parent's inode if @kn is a
759 * directory and link into the children list of the parent.
762 * 0 on success, -EEXIST if entry with the given name already
765 int kernfs_add_one(struct kernfs_node *kn)
767 struct kernfs_node *parent = kn->parent;
768 struct kernfs_iattrs *ps_iattr;
772 mutex_lock(&kernfs_mutex);
775 has_ns = kernfs_ns_enabled(parent);
776 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
777 has_ns ? "required" : "invalid", parent->name, kn->name))
780 if (kernfs_type(parent) != KERNFS_DIR)
784 if (parent->flags & KERNFS_EMPTY_DIR)
787 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
790 kn->hash = kernfs_name_hash(kn->name, kn->ns);
792 ret = kernfs_link_sibling(kn);
796 /* Update timestamps on the parent */
797 ps_iattr = parent->iattr;
799 ktime_get_real_ts64(&ps_iattr->ia_ctime);
800 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
803 mutex_unlock(&kernfs_mutex);
806 * Activate the new node unless CREATE_DEACTIVATED is requested.
807 * If not activated here, the kernfs user is responsible for
808 * activating the node with kernfs_activate(). A node which hasn't
809 * been activated is not visible to userland and its removal won't
810 * trigger deactivation.
812 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
817 mutex_unlock(&kernfs_mutex);
822 * kernfs_find_ns - find kernfs_node with the given name
823 * @parent: kernfs_node to search under
824 * @name: name to look for
825 * @ns: the namespace tag to use
827 * Look for kernfs_node with name @name under @parent. Returns pointer to
828 * the found kernfs_node on success, %NULL on failure.
830 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
831 const unsigned char *name,
834 struct rb_node *node = parent->dir.children.rb_node;
835 bool has_ns = kernfs_ns_enabled(parent);
838 lockdep_assert_held(&kernfs_mutex);
840 if (has_ns != (bool)ns) {
841 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
842 has_ns ? "required" : "invalid", parent->name, name);
846 hash = kernfs_name_hash(name, ns);
848 struct kernfs_node *kn;
852 result = kernfs_name_compare(hash, name, ns, kn);
854 node = node->rb_left;
856 node = node->rb_right;
863 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
864 const unsigned char *path,
870 lockdep_assert_held(&kernfs_mutex);
872 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
873 spin_lock_irq(&kernfs_rename_lock);
875 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
877 if (len >= sizeof(kernfs_pr_cont_buf)) {
878 spin_unlock_irq(&kernfs_rename_lock);
882 p = kernfs_pr_cont_buf;
884 while ((name = strsep(&p, "/")) && parent) {
887 parent = kernfs_find_ns(parent, name, ns);
890 spin_unlock_irq(&kernfs_rename_lock);
896 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
897 * @parent: kernfs_node to search under
898 * @name: name to look for
899 * @ns: the namespace tag to use
901 * Look for kernfs_node with name @name under @parent and get a reference
902 * if found. This function may sleep and returns pointer to the found
903 * kernfs_node on success, %NULL on failure.
905 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
906 const char *name, const void *ns)
908 struct kernfs_node *kn;
910 mutex_lock(&kernfs_mutex);
911 kn = kernfs_find_ns(parent, name, ns);
913 mutex_unlock(&kernfs_mutex);
917 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
920 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
921 * @parent: kernfs_node to search under
922 * @path: path to look for
923 * @ns: the namespace tag to use
925 * Look for kernfs_node with path @path under @parent and get a reference
926 * if found. This function may sleep and returns pointer to the found
927 * kernfs_node on success, %NULL on failure.
929 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
930 const char *path, const void *ns)
932 struct kernfs_node *kn;
934 mutex_lock(&kernfs_mutex);
935 kn = kernfs_walk_ns(parent, path, ns);
937 mutex_unlock(&kernfs_mutex);
943 * kernfs_create_root - create a new kernfs hierarchy
944 * @scops: optional syscall operations for the hierarchy
945 * @flags: KERNFS_ROOT_* flags
946 * @priv: opaque data associated with the new directory
948 * Returns the root of the new hierarchy on success, ERR_PTR() value on
951 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
952 unsigned int flags, void *priv)
954 struct kernfs_root *root;
955 struct kernfs_node *kn;
957 root = kzalloc(sizeof(*root), GFP_KERNEL);
959 return ERR_PTR(-ENOMEM);
961 idr_init(&root->ino_idr);
962 INIT_LIST_HEAD(&root->supers);
963 root->next_generation = 1;
965 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
966 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
969 idr_destroy(&root->ino_idr);
971 return ERR_PTR(-ENOMEM);
977 root->syscall_ops = scops;
980 init_waitqueue_head(&root->deactivate_waitq);
982 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
989 * kernfs_destroy_root - destroy a kernfs hierarchy
990 * @root: root of the hierarchy to destroy
992 * Destroy the hierarchy anchored at @root by removing all existing
993 * directories and destroying @root.
995 void kernfs_destroy_root(struct kernfs_root *root)
997 kernfs_remove(root->kn); /* will also free @root */
1001 * kernfs_create_dir_ns - create a directory
1002 * @parent: parent in which to create a new directory
1003 * @name: name of the new directory
1004 * @mode: mode of the new directory
1005 * @uid: uid of the new directory
1006 * @gid: gid of the new directory
1007 * @priv: opaque data associated with the new directory
1008 * @ns: optional namespace tag of the directory
1010 * Returns the created node on success, ERR_PTR() value on failure.
1012 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1013 const char *name, umode_t mode,
1014 kuid_t uid, kgid_t gid,
1015 void *priv, const void *ns)
1017 struct kernfs_node *kn;
1021 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1022 uid, gid, KERNFS_DIR);
1024 return ERR_PTR(-ENOMEM);
1026 kn->dir.root = parent->dir.root;
1031 rc = kernfs_add_one(kn);
1040 * kernfs_create_empty_dir - create an always empty directory
1041 * @parent: parent in which to create a new directory
1042 * @name: name of the new directory
1044 * Returns the created node on success, ERR_PTR() value on failure.
1046 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1049 struct kernfs_node *kn;
1053 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1054 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1056 return ERR_PTR(-ENOMEM);
1058 kn->flags |= KERNFS_EMPTY_DIR;
1059 kn->dir.root = parent->dir.root;
1064 rc = kernfs_add_one(kn);
1072 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1073 struct dentry *dentry,
1077 struct kernfs_node *parent = dir->i_private;
1078 struct kernfs_node *kn;
1079 struct inode *inode;
1080 const void *ns = NULL;
1082 mutex_lock(&kernfs_mutex);
1084 if (kernfs_ns_enabled(parent))
1085 ns = kernfs_info(dir->i_sb)->ns;
1087 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1090 if (!kn || !kernfs_active(kn)) {
1095 /* attach dentry and inode */
1096 inode = kernfs_get_inode(dir->i_sb, kn);
1098 ret = ERR_PTR(-ENOMEM);
1102 /* instantiate and hash dentry */
1103 ret = d_splice_alias(inode, dentry);
1105 mutex_unlock(&kernfs_mutex);
1109 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1112 struct kernfs_node *parent = dir->i_private;
1113 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1116 if (!scops || !scops->mkdir)
1119 if (!kernfs_get_active(parent))
1122 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1124 kernfs_put_active(parent);
1128 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1130 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1131 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1134 if (!scops || !scops->rmdir)
1137 if (!kernfs_get_active(kn))
1140 ret = scops->rmdir(kn);
1142 kernfs_put_active(kn);
1146 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1147 struct inode *new_dir, struct dentry *new_dentry,
1150 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1151 struct kernfs_node *new_parent = new_dir->i_private;
1152 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1158 if (!scops || !scops->rename)
1161 if (!kernfs_get_active(kn))
1164 if (!kernfs_get_active(new_parent)) {
1165 kernfs_put_active(kn);
1169 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1171 kernfs_put_active(new_parent);
1172 kernfs_put_active(kn);
1176 const struct inode_operations kernfs_dir_iops = {
1177 .lookup = kernfs_iop_lookup,
1178 .permission = kernfs_iop_permission,
1179 .setattr = kernfs_iop_setattr,
1180 .getattr = kernfs_iop_getattr,
1181 .listxattr = kernfs_iop_listxattr,
1183 .mkdir = kernfs_iop_mkdir,
1184 .rmdir = kernfs_iop_rmdir,
1185 .rename = kernfs_iop_rename,
1188 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1190 struct kernfs_node *last;
1193 struct rb_node *rbn;
1197 if (kernfs_type(pos) != KERNFS_DIR)
1200 rbn = rb_first(&pos->dir.children);
1204 pos = rb_to_kn(rbn);
1211 * kernfs_next_descendant_post - find the next descendant for post-order walk
1212 * @pos: the current position (%NULL to initiate traversal)
1213 * @root: kernfs_node whose descendants to walk
1215 * Find the next descendant to visit for post-order traversal of @root's
1216 * descendants. @root is included in the iteration and the last node to be
1219 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1220 struct kernfs_node *root)
1222 struct rb_node *rbn;
1224 lockdep_assert_held(&kernfs_mutex);
1226 /* if first iteration, visit leftmost descendant which may be root */
1228 return kernfs_leftmost_descendant(root);
1230 /* if we visited @root, we're done */
1234 /* if there's an unvisited sibling, visit its leftmost descendant */
1235 rbn = rb_next(&pos->rb);
1237 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1239 /* no sibling left, visit parent */
1244 * kernfs_activate - activate a node which started deactivated
1245 * @kn: kernfs_node whose subtree is to be activated
1247 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1248 * needs to be explicitly activated. A node which hasn't been activated
1249 * isn't visible to userland and deactivation is skipped during its
1250 * removal. This is useful to construct atomic init sequences where
1251 * creation of multiple nodes should either succeed or fail atomically.
1253 * The caller is responsible for ensuring that this function is not called
1254 * after kernfs_remove*() is invoked on @kn.
1256 void kernfs_activate(struct kernfs_node *kn)
1258 struct kernfs_node *pos;
1260 mutex_lock(&kernfs_mutex);
1263 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1264 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1267 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1268 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1270 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1271 pos->flags |= KERNFS_ACTIVATED;
1274 mutex_unlock(&kernfs_mutex);
1277 static void __kernfs_remove(struct kernfs_node *kn)
1279 struct kernfs_node *pos;
1281 lockdep_assert_held(&kernfs_mutex);
1284 * Short-circuit if non-root @kn has already finished removal.
1285 * This is for kernfs_remove_self() which plays with active ref
1288 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1291 pr_debug("kernfs %s: removing\n", kn->name);
1293 /* prevent any new usage under @kn by deactivating all nodes */
1295 while ((pos = kernfs_next_descendant_post(pos, kn)))
1296 if (kernfs_active(pos))
1297 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1299 /* deactivate and unlink the subtree node-by-node */
1301 pos = kernfs_leftmost_descendant(kn);
1304 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1305 * base ref could have been put by someone else by the time
1306 * the function returns. Make sure it doesn't go away
1312 * Drain iff @kn was activated. This avoids draining and
1313 * its lockdep annotations for nodes which have never been
1314 * activated and allows embedding kernfs_remove() in create
1315 * error paths without worrying about draining.
1317 if (kn->flags & KERNFS_ACTIVATED)
1320 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1323 * kernfs_unlink_sibling() succeeds once per node. Use it
1324 * to decide who's responsible for cleanups.
1326 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1327 struct kernfs_iattrs *ps_iattr =
1328 pos->parent ? pos->parent->iattr : NULL;
1330 /* update timestamps on the parent */
1332 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1333 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1340 } while (pos != kn);
1344 * kernfs_remove - remove a kernfs_node recursively
1345 * @kn: the kernfs_node to remove
1347 * Remove @kn along with all its subdirectories and files.
1349 void kernfs_remove(struct kernfs_node *kn)
1351 mutex_lock(&kernfs_mutex);
1352 __kernfs_remove(kn);
1353 mutex_unlock(&kernfs_mutex);
1357 * kernfs_break_active_protection - break out of active protection
1358 * @kn: the self kernfs_node
1360 * The caller must be running off of a kernfs operation which is invoked
1361 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1362 * this function must also be matched with an invocation of
1363 * kernfs_unbreak_active_protection().
1365 * This function releases the active reference of @kn the caller is
1366 * holding. Once this function is called, @kn may be removed at any point
1367 * and the caller is solely responsible for ensuring that the objects it
1368 * dereferences are accessible.
1370 void kernfs_break_active_protection(struct kernfs_node *kn)
1373 * Take out ourself out of the active ref dependency chain. If
1374 * we're called without an active ref, lockdep will complain.
1376 kernfs_put_active(kn);
1380 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1381 * @kn: the self kernfs_node
1383 * If kernfs_break_active_protection() was called, this function must be
1384 * invoked before finishing the kernfs operation. Note that while this
1385 * function restores the active reference, it doesn't and can't actually
1386 * restore the active protection - @kn may already or be in the process of
1387 * being removed. Once kernfs_break_active_protection() is invoked, that
1388 * protection is irreversibly gone for the kernfs operation instance.
1390 * While this function may be called at any point after
1391 * kernfs_break_active_protection() is invoked, its most useful location
1392 * would be right before the enclosing kernfs operation returns.
1394 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1397 * @kn->active could be in any state; however, the increment we do
1398 * here will be undone as soon as the enclosing kernfs operation
1399 * finishes and this temporary bump can't break anything. If @kn
1400 * is alive, nothing changes. If @kn is being deactivated, the
1401 * soon-to-follow put will either finish deactivation or restore
1402 * deactivated state. If @kn is already removed, the temporary
1403 * bump is guaranteed to be gone before @kn is released.
1405 atomic_inc(&kn->active);
1406 if (kernfs_lockdep(kn))
1407 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1411 * kernfs_remove_self - remove a kernfs_node from its own method
1412 * @kn: the self kernfs_node to remove
1414 * The caller must be running off of a kernfs operation which is invoked
1415 * with an active reference - e.g. one of kernfs_ops. This can be used to
1416 * implement a file operation which deletes itself.
1418 * For example, the "delete" file for a sysfs device directory can be
1419 * implemented by invoking kernfs_remove_self() on the "delete" file
1420 * itself. This function breaks the circular dependency of trying to
1421 * deactivate self while holding an active ref itself. It isn't necessary
1422 * to modify the usual removal path to use kernfs_remove_self(). The
1423 * "delete" implementation can simply invoke kernfs_remove_self() on self
1424 * before proceeding with the usual removal path. kernfs will ignore later
1425 * kernfs_remove() on self.
1427 * kernfs_remove_self() can be called multiple times concurrently on the
1428 * same kernfs_node. Only the first one actually performs removal and
1429 * returns %true. All others will wait until the kernfs operation which
1430 * won self-removal finishes and return %false. Note that the losers wait
1431 * for the completion of not only the winning kernfs_remove_self() but also
1432 * the whole kernfs_ops which won the arbitration. This can be used to
1433 * guarantee, for example, all concurrent writes to a "delete" file to
1434 * finish only after the whole operation is complete.
1436 bool kernfs_remove_self(struct kernfs_node *kn)
1440 mutex_lock(&kernfs_mutex);
1441 kernfs_break_active_protection(kn);
1444 * SUICIDAL is used to arbitrate among competing invocations. Only
1445 * the first one will actually perform removal. When the removal
1446 * is complete, SUICIDED is set and the active ref is restored
1447 * while holding kernfs_mutex. The ones which lost arbitration
1448 * waits for SUICDED && drained which can happen only after the
1449 * enclosing kernfs operation which executed the winning instance
1450 * of kernfs_remove_self() finished.
1452 if (!(kn->flags & KERNFS_SUICIDAL)) {
1453 kn->flags |= KERNFS_SUICIDAL;
1454 __kernfs_remove(kn);
1455 kn->flags |= KERNFS_SUICIDED;
1458 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1462 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1464 if ((kn->flags & KERNFS_SUICIDED) &&
1465 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1468 mutex_unlock(&kernfs_mutex);
1470 mutex_lock(&kernfs_mutex);
1472 finish_wait(waitq, &wait);
1473 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1478 * This must be done while holding kernfs_mutex; otherwise, waiting
1479 * for SUICIDED && deactivated could finish prematurely.
1481 kernfs_unbreak_active_protection(kn);
1483 mutex_unlock(&kernfs_mutex);
1488 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1489 * @parent: parent of the target
1490 * @name: name of the kernfs_node to remove
1491 * @ns: namespace tag of the kernfs_node to remove
1493 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1494 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1496 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1499 struct kernfs_node *kn;
1502 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1507 mutex_lock(&kernfs_mutex);
1509 kn = kernfs_find_ns(parent, name, ns);
1511 __kernfs_remove(kn);
1513 mutex_unlock(&kernfs_mutex);
1522 * kernfs_rename_ns - move and rename a kernfs_node
1524 * @new_parent: new parent to put @sd under
1525 * @new_name: new name
1526 * @new_ns: new namespace tag
1528 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1529 const char *new_name, const void *new_ns)
1531 struct kernfs_node *old_parent;
1532 const char *old_name = NULL;
1535 /* can't move or rename root */
1539 mutex_lock(&kernfs_mutex);
1542 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1543 (new_parent->flags & KERNFS_EMPTY_DIR))
1547 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1548 (strcmp(kn->name, new_name) == 0))
1549 goto out; /* nothing to rename */
1552 if (kernfs_find_ns(new_parent, new_name, new_ns))
1555 /* rename kernfs_node */
1556 if (strcmp(kn->name, new_name) != 0) {
1558 new_name = kstrdup_const(new_name, GFP_KERNEL);
1566 * Move to the appropriate place in the appropriate directories rbtree.
1568 kernfs_unlink_sibling(kn);
1569 kernfs_get(new_parent);
1571 /* rename_lock protects ->parent and ->name accessors */
1572 spin_lock_irq(&kernfs_rename_lock);
1574 old_parent = kn->parent;
1575 kn->parent = new_parent;
1579 old_name = kn->name;
1580 kn->name = new_name;
1583 spin_unlock_irq(&kernfs_rename_lock);
1585 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1586 kernfs_link_sibling(kn);
1588 kernfs_put(old_parent);
1589 kfree_const(old_name);
1593 mutex_unlock(&kernfs_mutex);
1597 /* Relationship between s_mode and the DT_xxx types */
1598 static inline unsigned char dt_type(struct kernfs_node *kn)
1600 return (kn->mode >> 12) & 15;
1603 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1605 kernfs_put(filp->private_data);
1609 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1610 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1613 int valid = kernfs_active(pos) &&
1614 pos->parent == parent && hash == pos->hash;
1619 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1620 struct rb_node *node = parent->dir.children.rb_node;
1622 pos = rb_to_kn(node);
1624 if (hash < pos->hash)
1625 node = node->rb_left;
1626 else if (hash > pos->hash)
1627 node = node->rb_right;
1632 /* Skip over entries which are dying/dead or in the wrong namespace */
1633 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1634 struct rb_node *node = rb_next(&pos->rb);
1638 pos = rb_to_kn(node);
1643 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1644 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1646 pos = kernfs_dir_pos(ns, parent, ino, pos);
1649 struct rb_node *node = rb_next(&pos->rb);
1653 pos = rb_to_kn(node);
1654 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1659 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1661 struct dentry *dentry = file->f_path.dentry;
1662 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1663 struct kernfs_node *pos = file->private_data;
1664 const void *ns = NULL;
1666 if (!dir_emit_dots(file, ctx))
1668 mutex_lock(&kernfs_mutex);
1670 if (kernfs_ns_enabled(parent))
1671 ns = kernfs_info(dentry->d_sb)->ns;
1673 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1675 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1676 const char *name = pos->name;
1677 unsigned int type = dt_type(pos);
1678 int len = strlen(name);
1679 ino_t ino = pos->id.ino;
1681 ctx->pos = pos->hash;
1682 file->private_data = pos;
1685 mutex_unlock(&kernfs_mutex);
1686 if (!dir_emit(ctx, name, len, ino, type))
1688 mutex_lock(&kernfs_mutex);
1690 mutex_unlock(&kernfs_mutex);
1691 file->private_data = NULL;
1696 const struct file_operations kernfs_dir_fops = {
1697 .read = generic_read_dir,
1698 .iterate_shared = kernfs_fop_readdir,
1699 .release = kernfs_dir_fop_release,
1700 .llseek = generic_file_llseek,