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);
143 common = kernfs_common_ancestor(kn_from, kn_to);
144 if (WARN_ON(!common))
147 depth_to = kernfs_depth(common, kn_to);
148 depth_from = kernfs_depth(common, kn_from);
152 for (i = 0; i < depth_from; i++)
153 len += strlcpy(buf + len, parent_str,
154 len < buflen ? buflen - len : 0);
156 /* Calculate how many bytes we need for the rest */
157 for (i = depth_to - 1; i >= 0; i--) {
158 for (kn = kn_to, j = 0; j < i; j++)
160 len += strlcpy(buf + len, "/",
161 len < buflen ? buflen - len : 0);
162 len += strlcpy(buf + len, kn->name,
163 len < buflen ? buflen - len : 0);
170 * kernfs_name - obtain the name of a given node
171 * @kn: kernfs_node of interest
172 * @buf: buffer to copy @kn's name into
173 * @buflen: size of @buf
175 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
176 * similar to strlcpy(). It returns the length of @kn's name and if @buf
177 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
179 * Fills buffer with "(null)" if @kn is NULL.
181 * This function can be called from any context.
183 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
188 spin_lock_irqsave(&kernfs_rename_lock, flags);
189 ret = kernfs_name_locked(kn, buf, buflen);
190 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
195 * kernfs_path_from_node - build path of node @to relative to @from.
196 * @from: parent kernfs_node relative to which we need to build the path
197 * @to: kernfs_node of interest
198 * @buf: buffer to copy @to's path into
199 * @buflen: size of @buf
201 * Builds @to's path relative to @from in @buf. @from and @to must
202 * be on the same kernfs-root. If @from is not parent of @to, then a relative
203 * path (which includes '..'s) as needed to reach from @from to @to is
206 * Returns the length of the full path. If the full length is equal to or
207 * greater than @buflen, @buf contains the truncated path with the trailing
208 * '\0'. On error, -errno is returned.
210 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
211 char *buf, size_t buflen)
216 spin_lock_irqsave(&kernfs_rename_lock, flags);
217 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
218 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
221 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
224 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
225 * @kn: kernfs_node of interest
227 * This function can be called from any context.
229 void pr_cont_kernfs_name(struct kernfs_node *kn)
233 spin_lock_irqsave(&kernfs_rename_lock, flags);
235 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
236 pr_cont("%s", kernfs_pr_cont_buf);
238 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
242 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
243 * @kn: kernfs_node of interest
245 * This function can be called from any context.
247 void pr_cont_kernfs_path(struct kernfs_node *kn)
252 spin_lock_irqsave(&kernfs_rename_lock, flags);
254 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
255 sizeof(kernfs_pr_cont_buf));
261 if (sz >= sizeof(kernfs_pr_cont_buf)) {
262 pr_cont("(name too long)");
266 pr_cont("%s", kernfs_pr_cont_buf);
269 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
273 * kernfs_get_parent - determine the parent node and pin it
274 * @kn: kernfs_node of interest
276 * Determines @kn's parent, pins and returns it. This function can be
277 * called from any context.
279 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
281 struct kernfs_node *parent;
284 spin_lock_irqsave(&kernfs_rename_lock, flags);
287 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
294 * @name: Null terminated string to hash
295 * @ns: Namespace tag to hash
297 * Returns 31 bit hash of ns + name (so it fits in an off_t )
299 static unsigned int kernfs_name_hash(const char *name, const void *ns)
301 unsigned long hash = init_name_hash(ns);
302 unsigned int len = strlen(name);
304 hash = partial_name_hash(*name++, hash);
305 hash = end_name_hash(hash);
307 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
315 static int kernfs_name_compare(unsigned int hash, const char *name,
316 const void *ns, const struct kernfs_node *kn)
326 return strcmp(name, kn->name);
329 static int kernfs_sd_compare(const struct kernfs_node *left,
330 const struct kernfs_node *right)
332 return kernfs_name_compare(left->hash, left->name, left->ns, right);
336 * kernfs_link_sibling - link kernfs_node into sibling rbtree
337 * @kn: kernfs_node of interest
339 * Link @kn into its sibling rbtree which starts from
340 * @kn->parent->dir.children.
343 * mutex_lock(kernfs_mutex)
346 * 0 on susccess -EEXIST on failure.
348 static int kernfs_link_sibling(struct kernfs_node *kn)
350 struct rb_node **node = &kn->parent->dir.children.rb_node;
351 struct rb_node *parent = NULL;
354 struct kernfs_node *pos;
357 pos = rb_to_kn(*node);
359 result = kernfs_sd_compare(kn, pos);
361 node = &pos->rb.rb_left;
363 node = &pos->rb.rb_right;
368 /* add new node and rebalance the tree */
369 rb_link_node(&kn->rb, parent, node);
370 rb_insert_color(&kn->rb, &kn->parent->dir.children);
372 /* successfully added, account subdir number */
373 if (kernfs_type(kn) == KERNFS_DIR)
374 kn->parent->dir.subdirs++;
380 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
381 * @kn: kernfs_node of interest
383 * Try to unlink @kn from its sibling rbtree which starts from
384 * kn->parent->dir.children. Returns %true if @kn was actually
385 * removed, %false if @kn wasn't on the rbtree.
388 * mutex_lock(kernfs_mutex)
390 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
392 if (RB_EMPTY_NODE(&kn->rb))
395 if (kernfs_type(kn) == KERNFS_DIR)
396 kn->parent->dir.subdirs--;
398 rb_erase(&kn->rb, &kn->parent->dir.children);
399 RB_CLEAR_NODE(&kn->rb);
404 * kernfs_get_active - get an active reference to kernfs_node
405 * @kn: kernfs_node to get an active reference to
407 * Get an active reference of @kn. This function is noop if @kn
411 * Pointer to @kn on success, NULL on failure.
413 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
418 if (!atomic_inc_unless_negative(&kn->active))
421 if (kernfs_lockdep(kn))
422 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
427 * kernfs_put_active - put an active reference to kernfs_node
428 * @kn: kernfs_node to put an active reference to
430 * Put an active reference to @kn. This function is noop if @kn
433 void kernfs_put_active(struct kernfs_node *kn)
440 if (kernfs_lockdep(kn))
441 rwsem_release(&kn->dep_map, 1, _RET_IP_);
442 v = atomic_dec_return(&kn->active);
443 if (likely(v != KN_DEACTIVATED_BIAS))
446 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
450 * kernfs_drain - drain kernfs_node
451 * @kn: kernfs_node to drain
453 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
454 * removers may invoke this function concurrently on @kn and all will
455 * return after draining is complete.
457 static void kernfs_drain(struct kernfs_node *kn)
458 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
460 struct kernfs_root *root = kernfs_root(kn);
462 lockdep_assert_held(&kernfs_mutex);
463 WARN_ON_ONCE(kernfs_active(kn));
465 mutex_unlock(&kernfs_mutex);
467 if (kernfs_lockdep(kn)) {
468 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
469 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
470 lock_contended(&kn->dep_map, _RET_IP_);
473 /* but everyone should wait for draining */
474 wait_event(root->deactivate_waitq,
475 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
477 if (kernfs_lockdep(kn)) {
478 lock_acquired(&kn->dep_map, _RET_IP_);
479 rwsem_release(&kn->dep_map, 1, _RET_IP_);
482 kernfs_drain_open_files(kn);
484 mutex_lock(&kernfs_mutex);
488 * kernfs_get - get a reference count on a kernfs_node
489 * @kn: the target kernfs_node
491 void kernfs_get(struct kernfs_node *kn)
494 WARN_ON(!atomic_read(&kn->count));
495 atomic_inc(&kn->count);
498 EXPORT_SYMBOL_GPL(kernfs_get);
501 * kernfs_put - put a reference count on a kernfs_node
502 * @kn: the target kernfs_node
504 * Put a reference count of @kn and destroy it if it reached zero.
506 void kernfs_put(struct kernfs_node *kn)
508 struct kernfs_node *parent;
509 struct kernfs_root *root;
511 if (!kn || !atomic_dec_and_test(&kn->count))
513 root = kernfs_root(kn);
516 * Moving/renaming is always done while holding reference.
517 * kn->parent won't change beneath us.
521 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
522 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
523 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
525 if (kernfs_type(kn) == KERNFS_LINK)
526 kernfs_put(kn->symlink.target_kn);
528 kfree_const(kn->name);
531 simple_xattrs_free(&kn->iattr->xattrs);
532 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
534 spin_lock(&kernfs_idr_lock);
535 idr_remove(&root->ino_idr, kn->id.ino);
536 spin_unlock(&kernfs_idr_lock);
537 kmem_cache_free(kernfs_node_cache, kn);
541 if (atomic_dec_and_test(&kn->count))
544 /* just released the root kn, free @root too */
545 idr_destroy(&root->ino_idr);
549 EXPORT_SYMBOL_GPL(kernfs_put);
551 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
553 struct kernfs_node *kn;
555 if (flags & LOOKUP_RCU)
558 /* Always perform fresh lookup for negatives */
559 if (d_really_is_negative(dentry))
560 goto out_bad_unlocked;
562 kn = kernfs_dentry_node(dentry);
563 mutex_lock(&kernfs_mutex);
565 /* The kernfs node has been deactivated */
566 if (!kernfs_active(kn))
569 /* The kernfs node has been moved? */
570 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
573 /* The kernfs node has been renamed */
574 if (strcmp(dentry->d_name.name, kn->name) != 0)
577 /* The kernfs node has been moved to a different namespace */
578 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
579 kernfs_info(dentry->d_sb)->ns != kn->ns)
582 mutex_unlock(&kernfs_mutex);
585 mutex_unlock(&kernfs_mutex);
590 const struct dentry_operations kernfs_dops = {
591 .d_revalidate = kernfs_dop_revalidate,
595 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
596 * @dentry: the dentry in question
598 * Return the kernfs_node associated with @dentry. If @dentry is not a
599 * kernfs one, %NULL is returned.
601 * While the returned kernfs_node will stay accessible as long as @dentry
602 * is accessible, the returned node can be in any state and the caller is
603 * fully responsible for determining what's accessible.
605 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
607 if (dentry->d_sb->s_op == &kernfs_sops &&
608 !d_really_is_negative(dentry))
609 return kernfs_dentry_node(dentry);
613 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
614 struct kernfs_node *parent,
615 const char *name, umode_t mode,
616 kuid_t uid, kgid_t gid,
619 struct kernfs_node *kn;
623 name = kstrdup_const(name, GFP_KERNEL);
627 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
631 idr_preload(GFP_KERNEL);
632 spin_lock(&kernfs_idr_lock);
633 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
634 if (ret >= 0 && ret < root->last_ino)
635 root->next_generation++;
636 gen = root->next_generation;
637 root->last_ino = ret;
638 spin_unlock(&kernfs_idr_lock);
643 kn->id.generation = gen;
645 atomic_set(&kn->count, 1);
646 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
647 RB_CLEAR_NODE(&kn->rb);
653 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
654 struct iattr iattr = {
655 .ia_valid = ATTR_UID | ATTR_GID,
660 ret = __kernfs_setattr(kn, &iattr);
666 ret = security_kernfs_init_security(parent, kn);
674 idr_remove(&root->ino_idr, kn->id.ino);
676 kmem_cache_free(kernfs_node_cache, kn);
682 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
683 const char *name, umode_t mode,
684 kuid_t uid, kgid_t gid,
687 struct kernfs_node *kn;
689 kn = __kernfs_new_node(kernfs_root(parent), parent,
690 name, mode, uid, gid, flags);
699 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
700 * @root: the kernfs root
704 * NULL on failure. Return a kernfs node with reference counter incremented
706 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
709 struct kernfs_node *kn;
711 spin_lock(&kernfs_idr_lock);
713 kn = idr_find(&root->ino_idr, ino);
718 * ACTIVATED is protected with kernfs_mutex but it was clear when
719 * @kn was added to idr and we just wanna see it set. No need to
722 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
723 !atomic_inc_not_zero(&kn->count)))
726 spin_unlock(&kernfs_idr_lock);
729 spin_unlock(&kernfs_idr_lock);
734 * kernfs_add_one - add kernfs_node to parent without warning
735 * @kn: kernfs_node to be added
737 * The caller must already have initialized @kn->parent. This
738 * function increments nlink of the parent's inode if @kn is a
739 * directory and link into the children list of the parent.
742 * 0 on success, -EEXIST if entry with the given name already
745 int kernfs_add_one(struct kernfs_node *kn)
747 struct kernfs_node *parent = kn->parent;
748 struct kernfs_iattrs *ps_iattr;
752 mutex_lock(&kernfs_mutex);
755 has_ns = kernfs_ns_enabled(parent);
756 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
757 has_ns ? "required" : "invalid", parent->name, kn->name))
760 if (kernfs_type(parent) != KERNFS_DIR)
764 if (parent->flags & KERNFS_EMPTY_DIR)
767 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
770 kn->hash = kernfs_name_hash(kn->name, kn->ns);
772 ret = kernfs_link_sibling(kn);
776 /* Update timestamps on the parent */
777 ps_iattr = parent->iattr;
779 ktime_get_real_ts64(&ps_iattr->ia_ctime);
780 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
783 mutex_unlock(&kernfs_mutex);
786 * Activate the new node unless CREATE_DEACTIVATED is requested.
787 * If not activated here, the kernfs user is responsible for
788 * activating the node with kernfs_activate(). A node which hasn't
789 * been activated is not visible to userland and its removal won't
790 * trigger deactivation.
792 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
797 mutex_unlock(&kernfs_mutex);
802 * kernfs_find_ns - find kernfs_node with the given name
803 * @parent: kernfs_node to search under
804 * @name: name to look for
805 * @ns: the namespace tag to use
807 * Look for kernfs_node with name @name under @parent. Returns pointer to
808 * the found kernfs_node on success, %NULL on failure.
810 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
811 const unsigned char *name,
814 struct rb_node *node = parent->dir.children.rb_node;
815 bool has_ns = kernfs_ns_enabled(parent);
818 lockdep_assert_held(&kernfs_mutex);
820 if (has_ns != (bool)ns) {
821 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
822 has_ns ? "required" : "invalid", parent->name, name);
826 hash = kernfs_name_hash(name, ns);
828 struct kernfs_node *kn;
832 result = kernfs_name_compare(hash, name, ns, kn);
834 node = node->rb_left;
836 node = node->rb_right;
843 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
844 const unsigned char *path,
850 lockdep_assert_held(&kernfs_mutex);
852 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
853 spin_lock_irq(&kernfs_rename_lock);
855 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
857 if (len >= sizeof(kernfs_pr_cont_buf)) {
858 spin_unlock_irq(&kernfs_rename_lock);
862 p = kernfs_pr_cont_buf;
864 while ((name = strsep(&p, "/")) && parent) {
867 parent = kernfs_find_ns(parent, name, ns);
870 spin_unlock_irq(&kernfs_rename_lock);
876 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
877 * @parent: kernfs_node to search under
878 * @name: name to look for
879 * @ns: the namespace tag to use
881 * Look for kernfs_node with name @name under @parent and get a reference
882 * if found. This function may sleep and returns pointer to the found
883 * kernfs_node on success, %NULL on failure.
885 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
886 const char *name, const void *ns)
888 struct kernfs_node *kn;
890 mutex_lock(&kernfs_mutex);
891 kn = kernfs_find_ns(parent, name, ns);
893 mutex_unlock(&kernfs_mutex);
897 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
900 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
901 * @parent: kernfs_node to search under
902 * @path: path to look for
903 * @ns: the namespace tag to use
905 * Look for kernfs_node with path @path under @parent and get a reference
906 * if found. This function may sleep and returns pointer to the found
907 * kernfs_node on success, %NULL on failure.
909 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
910 const char *path, const void *ns)
912 struct kernfs_node *kn;
914 mutex_lock(&kernfs_mutex);
915 kn = kernfs_walk_ns(parent, path, ns);
917 mutex_unlock(&kernfs_mutex);
923 * kernfs_create_root - create a new kernfs hierarchy
924 * @scops: optional syscall operations for the hierarchy
925 * @flags: KERNFS_ROOT_* flags
926 * @priv: opaque data associated with the new directory
928 * Returns the root of the new hierarchy on success, ERR_PTR() value on
931 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
932 unsigned int flags, void *priv)
934 struct kernfs_root *root;
935 struct kernfs_node *kn;
937 root = kzalloc(sizeof(*root), GFP_KERNEL);
939 return ERR_PTR(-ENOMEM);
941 idr_init(&root->ino_idr);
942 INIT_LIST_HEAD(&root->supers);
943 root->next_generation = 1;
945 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
946 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
949 idr_destroy(&root->ino_idr);
951 return ERR_PTR(-ENOMEM);
957 root->syscall_ops = scops;
960 init_waitqueue_head(&root->deactivate_waitq);
962 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
969 * kernfs_destroy_root - destroy a kernfs hierarchy
970 * @root: root of the hierarchy to destroy
972 * Destroy the hierarchy anchored at @root by removing all existing
973 * directories and destroying @root.
975 void kernfs_destroy_root(struct kernfs_root *root)
977 kernfs_remove(root->kn); /* will also free @root */
981 * kernfs_create_dir_ns - create a directory
982 * @parent: parent in which to create a new directory
983 * @name: name of the new directory
984 * @mode: mode of the new directory
985 * @uid: uid of the new directory
986 * @gid: gid of the new directory
987 * @priv: opaque data associated with the new directory
988 * @ns: optional namespace tag of the directory
990 * Returns the created node on success, ERR_PTR() value on failure.
992 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
993 const char *name, umode_t mode,
994 kuid_t uid, kgid_t gid,
995 void *priv, const void *ns)
997 struct kernfs_node *kn;
1001 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1002 uid, gid, KERNFS_DIR);
1004 return ERR_PTR(-ENOMEM);
1006 kn->dir.root = parent->dir.root;
1011 rc = kernfs_add_one(kn);
1020 * kernfs_create_empty_dir - create an always empty directory
1021 * @parent: parent in which to create a new directory
1022 * @name: name of the new directory
1024 * Returns the created node on success, ERR_PTR() value on failure.
1026 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1029 struct kernfs_node *kn;
1033 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1034 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1036 return ERR_PTR(-ENOMEM);
1038 kn->flags |= KERNFS_EMPTY_DIR;
1039 kn->dir.root = parent->dir.root;
1044 rc = kernfs_add_one(kn);
1052 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1053 struct dentry *dentry,
1057 struct kernfs_node *parent = dir->i_private;
1058 struct kernfs_node *kn;
1059 struct inode *inode;
1060 const void *ns = NULL;
1062 mutex_lock(&kernfs_mutex);
1064 if (kernfs_ns_enabled(parent))
1065 ns = kernfs_info(dir->i_sb)->ns;
1067 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1070 if (!kn || !kernfs_active(kn)) {
1075 /* attach dentry and inode */
1076 inode = kernfs_get_inode(dir->i_sb, kn);
1078 ret = ERR_PTR(-ENOMEM);
1082 /* instantiate and hash dentry */
1083 ret = d_splice_alias(inode, dentry);
1085 mutex_unlock(&kernfs_mutex);
1089 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1092 struct kernfs_node *parent = dir->i_private;
1093 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1096 if (!scops || !scops->mkdir)
1099 if (!kernfs_get_active(parent))
1102 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1104 kernfs_put_active(parent);
1108 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1110 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1111 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1114 if (!scops || !scops->rmdir)
1117 if (!kernfs_get_active(kn))
1120 ret = scops->rmdir(kn);
1122 kernfs_put_active(kn);
1126 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1127 struct inode *new_dir, struct dentry *new_dentry,
1130 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1131 struct kernfs_node *new_parent = new_dir->i_private;
1132 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1138 if (!scops || !scops->rename)
1141 if (!kernfs_get_active(kn))
1144 if (!kernfs_get_active(new_parent)) {
1145 kernfs_put_active(kn);
1149 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1151 kernfs_put_active(new_parent);
1152 kernfs_put_active(kn);
1156 const struct inode_operations kernfs_dir_iops = {
1157 .lookup = kernfs_iop_lookup,
1158 .permission = kernfs_iop_permission,
1159 .setattr = kernfs_iop_setattr,
1160 .getattr = kernfs_iop_getattr,
1161 .listxattr = kernfs_iop_listxattr,
1163 .mkdir = kernfs_iop_mkdir,
1164 .rmdir = kernfs_iop_rmdir,
1165 .rename = kernfs_iop_rename,
1168 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1170 struct kernfs_node *last;
1173 struct rb_node *rbn;
1177 if (kernfs_type(pos) != KERNFS_DIR)
1180 rbn = rb_first(&pos->dir.children);
1184 pos = rb_to_kn(rbn);
1191 * kernfs_next_descendant_post - find the next descendant for post-order walk
1192 * @pos: the current position (%NULL to initiate traversal)
1193 * @root: kernfs_node whose descendants to walk
1195 * Find the next descendant to visit for post-order traversal of @root's
1196 * descendants. @root is included in the iteration and the last node to be
1199 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1200 struct kernfs_node *root)
1202 struct rb_node *rbn;
1204 lockdep_assert_held(&kernfs_mutex);
1206 /* if first iteration, visit leftmost descendant which may be root */
1208 return kernfs_leftmost_descendant(root);
1210 /* if we visited @root, we're done */
1214 /* if there's an unvisited sibling, visit its leftmost descendant */
1215 rbn = rb_next(&pos->rb);
1217 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1219 /* no sibling left, visit parent */
1224 * kernfs_activate - activate a node which started deactivated
1225 * @kn: kernfs_node whose subtree is to be activated
1227 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1228 * needs to be explicitly activated. A node which hasn't been activated
1229 * isn't visible to userland and deactivation is skipped during its
1230 * removal. This is useful to construct atomic init sequences where
1231 * creation of multiple nodes should either succeed or fail atomically.
1233 * The caller is responsible for ensuring that this function is not called
1234 * after kernfs_remove*() is invoked on @kn.
1236 void kernfs_activate(struct kernfs_node *kn)
1238 struct kernfs_node *pos;
1240 mutex_lock(&kernfs_mutex);
1243 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1244 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1247 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1248 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1250 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1251 pos->flags |= KERNFS_ACTIVATED;
1254 mutex_unlock(&kernfs_mutex);
1257 static void __kernfs_remove(struct kernfs_node *kn)
1259 struct kernfs_node *pos;
1261 lockdep_assert_held(&kernfs_mutex);
1264 * Short-circuit if non-root @kn has already finished removal.
1265 * This is for kernfs_remove_self() which plays with active ref
1268 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1271 pr_debug("kernfs %s: removing\n", kn->name);
1273 /* prevent any new usage under @kn by deactivating all nodes */
1275 while ((pos = kernfs_next_descendant_post(pos, kn)))
1276 if (kernfs_active(pos))
1277 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1279 /* deactivate and unlink the subtree node-by-node */
1281 pos = kernfs_leftmost_descendant(kn);
1284 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1285 * base ref could have been put by someone else by the time
1286 * the function returns. Make sure it doesn't go away
1292 * Drain iff @kn was activated. This avoids draining and
1293 * its lockdep annotations for nodes which have never been
1294 * activated and allows embedding kernfs_remove() in create
1295 * error paths without worrying about draining.
1297 if (kn->flags & KERNFS_ACTIVATED)
1300 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1303 * kernfs_unlink_sibling() succeeds once per node. Use it
1304 * to decide who's responsible for cleanups.
1306 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1307 struct kernfs_iattrs *ps_iattr =
1308 pos->parent ? pos->parent->iattr : NULL;
1310 /* update timestamps on the parent */
1312 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1313 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1320 } while (pos != kn);
1324 * kernfs_remove - remove a kernfs_node recursively
1325 * @kn: the kernfs_node to remove
1327 * Remove @kn along with all its subdirectories and files.
1329 void kernfs_remove(struct kernfs_node *kn)
1331 mutex_lock(&kernfs_mutex);
1332 __kernfs_remove(kn);
1333 mutex_unlock(&kernfs_mutex);
1337 * kernfs_break_active_protection - break out of active protection
1338 * @kn: the self kernfs_node
1340 * The caller must be running off of a kernfs operation which is invoked
1341 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1342 * this function must also be matched with an invocation of
1343 * kernfs_unbreak_active_protection().
1345 * This function releases the active reference of @kn the caller is
1346 * holding. Once this function is called, @kn may be removed at any point
1347 * and the caller is solely responsible for ensuring that the objects it
1348 * dereferences are accessible.
1350 void kernfs_break_active_protection(struct kernfs_node *kn)
1353 * Take out ourself out of the active ref dependency chain. If
1354 * we're called without an active ref, lockdep will complain.
1356 kernfs_put_active(kn);
1360 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1361 * @kn: the self kernfs_node
1363 * If kernfs_break_active_protection() was called, this function must be
1364 * invoked before finishing the kernfs operation. Note that while this
1365 * function restores the active reference, it doesn't and can't actually
1366 * restore the active protection - @kn may already or be in the process of
1367 * being removed. Once kernfs_break_active_protection() is invoked, that
1368 * protection is irreversibly gone for the kernfs operation instance.
1370 * While this function may be called at any point after
1371 * kernfs_break_active_protection() is invoked, its most useful location
1372 * would be right before the enclosing kernfs operation returns.
1374 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1377 * @kn->active could be in any state; however, the increment we do
1378 * here will be undone as soon as the enclosing kernfs operation
1379 * finishes and this temporary bump can't break anything. If @kn
1380 * is alive, nothing changes. If @kn is being deactivated, the
1381 * soon-to-follow put will either finish deactivation or restore
1382 * deactivated state. If @kn is already removed, the temporary
1383 * bump is guaranteed to be gone before @kn is released.
1385 atomic_inc(&kn->active);
1386 if (kernfs_lockdep(kn))
1387 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1391 * kernfs_remove_self - remove a kernfs_node from its own method
1392 * @kn: the self kernfs_node to remove
1394 * The caller must be running off of a kernfs operation which is invoked
1395 * with an active reference - e.g. one of kernfs_ops. This can be used to
1396 * implement a file operation which deletes itself.
1398 * For example, the "delete" file for a sysfs device directory can be
1399 * implemented by invoking kernfs_remove_self() on the "delete" file
1400 * itself. This function breaks the circular dependency of trying to
1401 * deactivate self while holding an active ref itself. It isn't necessary
1402 * to modify the usual removal path to use kernfs_remove_self(). The
1403 * "delete" implementation can simply invoke kernfs_remove_self() on self
1404 * before proceeding with the usual removal path. kernfs will ignore later
1405 * kernfs_remove() on self.
1407 * kernfs_remove_self() can be called multiple times concurrently on the
1408 * same kernfs_node. Only the first one actually performs removal and
1409 * returns %true. All others will wait until the kernfs operation which
1410 * won self-removal finishes and return %false. Note that the losers wait
1411 * for the completion of not only the winning kernfs_remove_self() but also
1412 * the whole kernfs_ops which won the arbitration. This can be used to
1413 * guarantee, for example, all concurrent writes to a "delete" file to
1414 * finish only after the whole operation is complete.
1416 bool kernfs_remove_self(struct kernfs_node *kn)
1420 mutex_lock(&kernfs_mutex);
1421 kernfs_break_active_protection(kn);
1424 * SUICIDAL is used to arbitrate among competing invocations. Only
1425 * the first one will actually perform removal. When the removal
1426 * is complete, SUICIDED is set and the active ref is restored
1427 * while holding kernfs_mutex. The ones which lost arbitration
1428 * waits for SUICDED && drained which can happen only after the
1429 * enclosing kernfs operation which executed the winning instance
1430 * of kernfs_remove_self() finished.
1432 if (!(kn->flags & KERNFS_SUICIDAL)) {
1433 kn->flags |= KERNFS_SUICIDAL;
1434 __kernfs_remove(kn);
1435 kn->flags |= KERNFS_SUICIDED;
1438 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1442 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1444 if ((kn->flags & KERNFS_SUICIDED) &&
1445 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1448 mutex_unlock(&kernfs_mutex);
1450 mutex_lock(&kernfs_mutex);
1452 finish_wait(waitq, &wait);
1453 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1458 * This must be done while holding kernfs_mutex; otherwise, waiting
1459 * for SUICIDED && deactivated could finish prematurely.
1461 kernfs_unbreak_active_protection(kn);
1463 mutex_unlock(&kernfs_mutex);
1468 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1469 * @parent: parent of the target
1470 * @name: name of the kernfs_node to remove
1471 * @ns: namespace tag of the kernfs_node to remove
1473 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1474 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1476 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1479 struct kernfs_node *kn;
1482 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1487 mutex_lock(&kernfs_mutex);
1489 kn = kernfs_find_ns(parent, name, ns);
1491 __kernfs_remove(kn);
1493 mutex_unlock(&kernfs_mutex);
1502 * kernfs_rename_ns - move and rename a kernfs_node
1504 * @new_parent: new parent to put @sd under
1505 * @new_name: new name
1506 * @new_ns: new namespace tag
1508 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1509 const char *new_name, const void *new_ns)
1511 struct kernfs_node *old_parent;
1512 const char *old_name = NULL;
1515 /* can't move or rename root */
1519 mutex_lock(&kernfs_mutex);
1522 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1523 (new_parent->flags & KERNFS_EMPTY_DIR))
1527 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1528 (strcmp(kn->name, new_name) == 0))
1529 goto out; /* nothing to rename */
1532 if (kernfs_find_ns(new_parent, new_name, new_ns))
1535 /* rename kernfs_node */
1536 if (strcmp(kn->name, new_name) != 0) {
1538 new_name = kstrdup_const(new_name, GFP_KERNEL);
1546 * Move to the appropriate place in the appropriate directories rbtree.
1548 kernfs_unlink_sibling(kn);
1549 kernfs_get(new_parent);
1551 /* rename_lock protects ->parent and ->name accessors */
1552 spin_lock_irq(&kernfs_rename_lock);
1554 old_parent = kn->parent;
1555 kn->parent = new_parent;
1559 old_name = kn->name;
1560 kn->name = new_name;
1563 spin_unlock_irq(&kernfs_rename_lock);
1565 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1566 kernfs_link_sibling(kn);
1568 kernfs_put(old_parent);
1569 kfree_const(old_name);
1573 mutex_unlock(&kernfs_mutex);
1577 /* Relationship between s_mode and the DT_xxx types */
1578 static inline unsigned char dt_type(struct kernfs_node *kn)
1580 return (kn->mode >> 12) & 15;
1583 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1585 kernfs_put(filp->private_data);
1589 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1590 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1593 int valid = kernfs_active(pos) &&
1594 pos->parent == parent && hash == pos->hash;
1599 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1600 struct rb_node *node = parent->dir.children.rb_node;
1602 pos = rb_to_kn(node);
1604 if (hash < pos->hash)
1605 node = node->rb_left;
1606 else if (hash > pos->hash)
1607 node = node->rb_right;
1612 /* Skip over entries which are dying/dead or in the wrong namespace */
1613 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1614 struct rb_node *node = rb_next(&pos->rb);
1618 pos = rb_to_kn(node);
1623 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1624 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1626 pos = kernfs_dir_pos(ns, parent, ino, pos);
1629 struct rb_node *node = rb_next(&pos->rb);
1633 pos = rb_to_kn(node);
1634 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1639 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1641 struct dentry *dentry = file->f_path.dentry;
1642 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1643 struct kernfs_node *pos = file->private_data;
1644 const void *ns = NULL;
1646 if (!dir_emit_dots(file, ctx))
1648 mutex_lock(&kernfs_mutex);
1650 if (kernfs_ns_enabled(parent))
1651 ns = kernfs_info(dentry->d_sb)->ns;
1653 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1655 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1656 const char *name = pos->name;
1657 unsigned int type = dt_type(pos);
1658 int len = strlen(name);
1659 ino_t ino = pos->id.ino;
1661 ctx->pos = pos->hash;
1662 file->private_data = pos;
1665 mutex_unlock(&kernfs_mutex);
1666 if (!dir_emit(ctx, name, len, ino, type))
1668 mutex_lock(&kernfs_mutex);
1670 mutex_unlock(&kernfs_mutex);
1671 file->private_data = NULL;
1676 const struct file_operations kernfs_dir_fops = {
1677 .read = generic_read_dir,
1678 .iterate_shared = kernfs_fop_readdir,
1679 .release = kernfs_dir_fop_release,
1680 .llseek = generic_file_llseek,