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 */
23 * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
24 * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
25 * will perform wakeups when releasing console_sem. Holding rename_lock
26 * will introduce deadlock if the scheduler reads the kernfs_name in the
29 static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
30 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by pr_cont_lock */
31 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
33 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
35 static bool kernfs_active(struct kernfs_node *kn)
37 lockdep_assert_held(&kernfs_mutex);
38 return atomic_read(&kn->active) >= 0;
41 static bool kernfs_lockdep(struct kernfs_node *kn)
43 #ifdef CONFIG_DEBUG_LOCK_ALLOC
44 return kn->flags & KERNFS_LOCKDEP;
50 static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
53 return strlcpy(buf, "(null)", buflen);
55 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
58 /* kernfs_node_depth - compute depth from @from to @to */
59 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
63 while (to->parent && to != from) {
70 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
71 struct kernfs_node *b)
74 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
79 da = kernfs_depth(ra->kn, a);
80 db = kernfs_depth(rb->kn, b);
91 /* worst case b and a will be the same at root */
101 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
102 * where kn_from is treated as root of the path.
103 * @kn_from: kernfs node which should be treated as root for the path
104 * @kn_to: kernfs node to which path is needed
105 * @buf: buffer to copy the path into
106 * @buflen: size of @buf
108 * We need to handle couple of scenarios here:
109 * [1] when @kn_from is an ancestor of @kn_to at some level
111 * kn_to: /n1/n2/n3/n4/n5
114 * [2] when @kn_from is on a different hierarchy and we need to find common
115 * ancestor between @kn_from and @kn_to.
116 * kn_from: /n1/n2/n3/n4
120 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
121 * kn_to: /n1/n2/n3 [depth=3]
124 * [3] when @kn_to is NULL result will be "(null)"
126 * Returns the length of the full path. If the full length is equal to or
127 * greater than @buflen, @buf contains the truncated path with the trailing
128 * '\0'. On error, -errno is returned.
130 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
131 struct kernfs_node *kn_from,
132 char *buf, size_t buflen)
134 struct kernfs_node *kn, *common;
135 const char parent_str[] = "/..";
136 size_t depth_from, depth_to, len = 0;
140 return strlcpy(buf, "(null)", buflen);
143 kn_from = kernfs_root(kn_to)->kn;
145 if (kn_from == kn_to)
146 return strlcpy(buf, "/", buflen);
151 common = kernfs_common_ancestor(kn_from, kn_to);
152 if (WARN_ON(!common))
155 depth_to = kernfs_depth(common, kn_to);
156 depth_from = kernfs_depth(common, kn_from);
160 for (i = 0; i < depth_from; i++)
161 len += strlcpy(buf + len, parent_str,
162 len < buflen ? buflen - len : 0);
164 /* Calculate how many bytes we need for the rest */
165 for (i = depth_to - 1; i >= 0; i--) {
166 for (kn = kn_to, j = 0; j < i; j++)
168 len += strlcpy(buf + len, "/",
169 len < buflen ? buflen - len : 0);
170 len += strlcpy(buf + len, kn->name,
171 len < buflen ? buflen - len : 0);
178 * kernfs_name - obtain the name of a given node
179 * @kn: kernfs_node of interest
180 * @buf: buffer to copy @kn's name into
181 * @buflen: size of @buf
183 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
184 * similar to strlcpy(). It returns the length of @kn's name and if @buf
185 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
187 * Fills buffer with "(null)" if @kn is NULL.
189 * This function can be called from any context.
191 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
196 spin_lock_irqsave(&kernfs_rename_lock, flags);
197 ret = kernfs_name_locked(kn, buf, buflen);
198 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
203 * kernfs_path_from_node - build path of node @to relative to @from.
204 * @from: parent kernfs_node relative to which we need to build the path
205 * @to: kernfs_node of interest
206 * @buf: buffer to copy @to's path into
207 * @buflen: size of @buf
209 * Builds @to's path relative to @from in @buf. @from and @to must
210 * be on the same kernfs-root. If @from is not parent of @to, then a relative
211 * path (which includes '..'s) as needed to reach from @from to @to is
214 * Returns the length of the full path. If the full length is equal to or
215 * greater than @buflen, @buf contains the truncated path with the trailing
216 * '\0'. On error, -errno is returned.
218 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
219 char *buf, size_t buflen)
224 spin_lock_irqsave(&kernfs_rename_lock, flags);
225 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
226 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
229 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
232 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
233 * @kn: kernfs_node of interest
235 * This function can be called from any context.
237 void pr_cont_kernfs_name(struct kernfs_node *kn)
241 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
243 kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
244 pr_cont("%s", kernfs_pr_cont_buf);
246 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
250 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
251 * @kn: kernfs_node of interest
253 * This function can be called from any context.
255 void pr_cont_kernfs_path(struct kernfs_node *kn)
260 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
262 sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
263 sizeof(kernfs_pr_cont_buf));
269 if (sz >= sizeof(kernfs_pr_cont_buf)) {
270 pr_cont("(name too long)");
274 pr_cont("%s", kernfs_pr_cont_buf);
277 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
281 * kernfs_get_parent - determine the parent node and pin it
282 * @kn: kernfs_node of interest
284 * Determines @kn's parent, pins and returns it. This function can be
285 * called from any context.
287 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
289 struct kernfs_node *parent;
292 spin_lock_irqsave(&kernfs_rename_lock, flags);
295 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
302 * @name: Null terminated string to hash
303 * @ns: Namespace tag to hash
305 * Returns 31 bit hash of ns + name (so it fits in an off_t )
307 static unsigned int kernfs_name_hash(const char *name, const void *ns)
309 unsigned long hash = init_name_hash(ns);
310 unsigned int len = strlen(name);
312 hash = partial_name_hash(*name++, hash);
313 hash = end_name_hash(hash);
315 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
323 static int kernfs_name_compare(unsigned int hash, const char *name,
324 const void *ns, const struct kernfs_node *kn)
334 return strcmp(name, kn->name);
337 static int kernfs_sd_compare(const struct kernfs_node *left,
338 const struct kernfs_node *right)
340 return kernfs_name_compare(left->hash, left->name, left->ns, right);
344 * kernfs_link_sibling - link kernfs_node into sibling rbtree
345 * @kn: kernfs_node of interest
347 * Link @kn into its sibling rbtree which starts from
348 * @kn->parent->dir.children.
351 * mutex_lock(kernfs_mutex)
354 * 0 on susccess -EEXIST on failure.
356 static int kernfs_link_sibling(struct kernfs_node *kn)
358 struct rb_node **node = &kn->parent->dir.children.rb_node;
359 struct rb_node *parent = NULL;
362 struct kernfs_node *pos;
365 pos = rb_to_kn(*node);
367 result = kernfs_sd_compare(kn, pos);
369 node = &pos->rb.rb_left;
371 node = &pos->rb.rb_right;
376 /* add new node and rebalance the tree */
377 rb_link_node(&kn->rb, parent, node);
378 rb_insert_color(&kn->rb, &kn->parent->dir.children);
380 /* successfully added, account subdir number */
381 if (kernfs_type(kn) == KERNFS_DIR)
382 kn->parent->dir.subdirs++;
388 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
389 * @kn: kernfs_node of interest
391 * Try to unlink @kn from its sibling rbtree which starts from
392 * kn->parent->dir.children. Returns %true if @kn was actually
393 * removed, %false if @kn wasn't on the rbtree.
396 * mutex_lock(kernfs_mutex)
398 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
400 if (RB_EMPTY_NODE(&kn->rb))
403 if (kernfs_type(kn) == KERNFS_DIR)
404 kn->parent->dir.subdirs--;
406 rb_erase(&kn->rb, &kn->parent->dir.children);
407 RB_CLEAR_NODE(&kn->rb);
412 * kernfs_get_active - get an active reference to kernfs_node
413 * @kn: kernfs_node to get an active reference to
415 * Get an active reference of @kn. This function is noop if @kn
419 * Pointer to @kn on success, NULL on failure.
421 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
426 if (!atomic_inc_unless_negative(&kn->active))
429 if (kernfs_lockdep(kn))
430 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
435 * kernfs_put_active - put an active reference to kernfs_node
436 * @kn: kernfs_node to put an active reference to
438 * Put an active reference to @kn. This function is noop if @kn
441 void kernfs_put_active(struct kernfs_node *kn)
448 if (kernfs_lockdep(kn))
449 rwsem_release(&kn->dep_map, 1, _RET_IP_);
450 v = atomic_dec_return(&kn->active);
451 if (likely(v != KN_DEACTIVATED_BIAS))
454 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
458 * kernfs_drain - drain kernfs_node
459 * @kn: kernfs_node to drain
461 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
462 * removers may invoke this function concurrently on @kn and all will
463 * return after draining is complete.
465 static void kernfs_drain(struct kernfs_node *kn)
466 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
468 struct kernfs_root *root = kernfs_root(kn);
470 lockdep_assert_held(&kernfs_mutex);
471 WARN_ON_ONCE(kernfs_active(kn));
473 mutex_unlock(&kernfs_mutex);
475 if (kernfs_lockdep(kn)) {
476 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
477 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
478 lock_contended(&kn->dep_map, _RET_IP_);
481 /* but everyone should wait for draining */
482 wait_event(root->deactivate_waitq,
483 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
485 if (kernfs_lockdep(kn)) {
486 lock_acquired(&kn->dep_map, _RET_IP_);
487 rwsem_release(&kn->dep_map, 1, _RET_IP_);
490 kernfs_drain_open_files(kn);
492 mutex_lock(&kernfs_mutex);
496 * kernfs_get - get a reference count on a kernfs_node
497 * @kn: the target kernfs_node
499 void kernfs_get(struct kernfs_node *kn)
502 WARN_ON(!atomic_read(&kn->count));
503 atomic_inc(&kn->count);
506 EXPORT_SYMBOL_GPL(kernfs_get);
509 * kernfs_put - put a reference count on a kernfs_node
510 * @kn: the target kernfs_node
512 * Put a reference count of @kn and destroy it if it reached zero.
514 void kernfs_put(struct kernfs_node *kn)
516 struct kernfs_node *parent;
517 struct kernfs_root *root;
520 * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
521 * depends on this to filter reused stale node
523 if (!kn || !atomic_dec_and_test(&kn->count))
525 root = kernfs_root(kn);
528 * Moving/renaming is always done while holding reference.
529 * kn->parent won't change beneath us.
533 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
534 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
535 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
537 if (kernfs_type(kn) == KERNFS_LINK)
538 kernfs_put(kn->symlink.target_kn);
540 kfree_const(kn->name);
543 simple_xattrs_free(&kn->iattr->xattrs);
544 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
546 spin_lock(&kernfs_idr_lock);
547 idr_remove(&root->ino_idr, kn->id.ino);
548 spin_unlock(&kernfs_idr_lock);
549 kmem_cache_free(kernfs_node_cache, kn);
553 if (atomic_dec_and_test(&kn->count))
556 /* just released the root kn, free @root too */
557 idr_destroy(&root->ino_idr);
561 EXPORT_SYMBOL_GPL(kernfs_put);
563 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
565 struct kernfs_node *kn;
567 if (flags & LOOKUP_RCU)
570 /* Always perform fresh lookup for negatives */
571 if (d_really_is_negative(dentry))
572 goto out_bad_unlocked;
574 kn = kernfs_dentry_node(dentry);
575 mutex_lock(&kernfs_mutex);
577 /* The kernfs node has been deactivated */
578 if (!kernfs_active(kn))
581 /* The kernfs node has been moved? */
582 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
585 /* The kernfs node has been renamed */
586 if (strcmp(dentry->d_name.name, kn->name) != 0)
589 /* The kernfs node has been moved to a different namespace */
590 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
591 kernfs_info(dentry->d_sb)->ns != kn->ns)
594 mutex_unlock(&kernfs_mutex);
597 mutex_unlock(&kernfs_mutex);
602 const struct dentry_operations kernfs_dops = {
603 .d_revalidate = kernfs_dop_revalidate,
607 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
608 * @dentry: the dentry in question
610 * Return the kernfs_node associated with @dentry. If @dentry is not a
611 * kernfs one, %NULL is returned.
613 * While the returned kernfs_node will stay accessible as long as @dentry
614 * is accessible, the returned node can be in any state and the caller is
615 * fully responsible for determining what's accessible.
617 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
619 if (dentry->d_sb->s_op == &kernfs_sops &&
620 !d_really_is_negative(dentry))
621 return kernfs_dentry_node(dentry);
625 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
626 struct kernfs_node *parent,
627 const char *name, umode_t mode,
628 kuid_t uid, kgid_t gid,
631 struct kernfs_node *kn;
635 name = kstrdup_const(name, GFP_KERNEL);
639 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
643 idr_preload(GFP_KERNEL);
644 spin_lock(&kernfs_idr_lock);
645 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
646 if (ret >= 0 && ret < root->last_ino)
647 root->next_generation++;
648 gen = root->next_generation;
649 root->last_ino = ret;
650 spin_unlock(&kernfs_idr_lock);
655 kn->id.generation = gen;
658 * set ino first. This RELEASE is paired with atomic_inc_not_zero in
659 * kernfs_find_and_get_node_by_ino
661 atomic_set_release(&kn->count, 1);
662 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
663 RB_CLEAR_NODE(&kn->rb);
669 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
670 struct iattr iattr = {
671 .ia_valid = ATTR_UID | ATTR_GID,
676 ret = __kernfs_setattr(kn, &iattr);
682 ret = security_kernfs_init_security(parent, kn);
690 idr_remove(&root->ino_idr, kn->id.ino);
692 kmem_cache_free(kernfs_node_cache, kn);
698 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
699 const char *name, umode_t mode,
700 kuid_t uid, kgid_t gid,
703 struct kernfs_node *kn;
705 kn = __kernfs_new_node(kernfs_root(parent), parent,
706 name, mode, uid, gid, flags);
715 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
716 * @root: the kernfs root
720 * NULL on failure. Return a kernfs node with reference counter incremented
722 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
725 struct kernfs_node *kn;
728 kn = idr_find(&root->ino_idr, ino);
733 * Since kernfs_node is freed in RCU, it's possible an old node for ino
734 * is freed, but reused before RCU grace period. But a freed node (see
735 * kernfs_put) or an incompletedly initialized node (see
736 * __kernfs_new_node) should have 'count' 0. We can use this fact to
737 * filter out such node.
739 if (!atomic_inc_not_zero(&kn->count)) {
745 * The node could be a new node or a reused node. If it's a new node,
746 * we are ok. If it's reused because of RCU (because of
747 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
748 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
749 * hence we can use 'ino' to filter stale node.
751 if (kn->id.ino != ino)
763 * kernfs_add_one - add kernfs_node to parent without warning
764 * @kn: kernfs_node to be added
766 * The caller must already have initialized @kn->parent. This
767 * function increments nlink of the parent's inode if @kn is a
768 * directory and link into the children list of the parent.
771 * 0 on success, -EEXIST if entry with the given name already
774 int kernfs_add_one(struct kernfs_node *kn)
776 struct kernfs_node *parent = kn->parent;
777 struct kernfs_iattrs *ps_iattr;
781 mutex_lock(&kernfs_mutex);
784 has_ns = kernfs_ns_enabled(parent);
785 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
786 has_ns ? "required" : "invalid", parent->name, kn->name))
789 if (kernfs_type(parent) != KERNFS_DIR)
793 if (parent->flags & KERNFS_EMPTY_DIR)
796 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
799 kn->hash = kernfs_name_hash(kn->name, kn->ns);
801 ret = kernfs_link_sibling(kn);
805 /* Update timestamps on the parent */
806 ps_iattr = parent->iattr;
808 ktime_get_real_ts64(&ps_iattr->ia_ctime);
809 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
812 mutex_unlock(&kernfs_mutex);
815 * Activate the new node unless CREATE_DEACTIVATED is requested.
816 * If not activated here, the kernfs user is responsible for
817 * activating the node with kernfs_activate(). A node which hasn't
818 * been activated is not visible to userland and its removal won't
819 * trigger deactivation.
821 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
826 mutex_unlock(&kernfs_mutex);
831 * kernfs_find_ns - find kernfs_node with the given name
832 * @parent: kernfs_node to search under
833 * @name: name to look for
834 * @ns: the namespace tag to use
836 * Look for kernfs_node with name @name under @parent. Returns pointer to
837 * the found kernfs_node on success, %NULL on failure.
839 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
840 const unsigned char *name,
843 struct rb_node *node = parent->dir.children.rb_node;
844 bool has_ns = kernfs_ns_enabled(parent);
847 lockdep_assert_held(&kernfs_mutex);
849 if (has_ns != (bool)ns) {
850 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
851 has_ns ? "required" : "invalid", parent->name, name);
855 hash = kernfs_name_hash(name, ns);
857 struct kernfs_node *kn;
861 result = kernfs_name_compare(hash, name, ns, kn);
863 node = node->rb_left;
865 node = node->rb_right;
872 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
873 const unsigned char *path,
879 lockdep_assert_held(&kernfs_mutex);
881 spin_lock_irq(&kernfs_pr_cont_lock);
883 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
885 if (len >= sizeof(kernfs_pr_cont_buf)) {
886 spin_unlock_irq(&kernfs_pr_cont_lock);
890 p = kernfs_pr_cont_buf;
892 while ((name = strsep(&p, "/")) && parent) {
895 parent = kernfs_find_ns(parent, name, ns);
898 spin_unlock_irq(&kernfs_pr_cont_lock);
904 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
905 * @parent: kernfs_node to search under
906 * @name: name to look for
907 * @ns: the namespace tag to use
909 * Look for kernfs_node with name @name under @parent and get a reference
910 * if found. This function may sleep and returns pointer to the found
911 * kernfs_node on success, %NULL on failure.
913 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
914 const char *name, const void *ns)
916 struct kernfs_node *kn;
918 mutex_lock(&kernfs_mutex);
919 kn = kernfs_find_ns(parent, name, ns);
921 mutex_unlock(&kernfs_mutex);
925 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
928 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
929 * @parent: kernfs_node to search under
930 * @path: path to look for
931 * @ns: the namespace tag to use
933 * Look for kernfs_node with path @path under @parent and get a reference
934 * if found. This function may sleep and returns pointer to the found
935 * kernfs_node on success, %NULL on failure.
937 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
938 const char *path, const void *ns)
940 struct kernfs_node *kn;
942 mutex_lock(&kernfs_mutex);
943 kn = kernfs_walk_ns(parent, path, ns);
945 mutex_unlock(&kernfs_mutex);
951 * kernfs_create_root - create a new kernfs hierarchy
952 * @scops: optional syscall operations for the hierarchy
953 * @flags: KERNFS_ROOT_* flags
954 * @priv: opaque data associated with the new directory
956 * Returns the root of the new hierarchy on success, ERR_PTR() value on
959 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
960 unsigned int flags, void *priv)
962 struct kernfs_root *root;
963 struct kernfs_node *kn;
965 root = kzalloc(sizeof(*root), GFP_KERNEL);
967 return ERR_PTR(-ENOMEM);
969 idr_init(&root->ino_idr);
970 INIT_LIST_HEAD(&root->supers);
971 root->next_generation = 1;
973 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
974 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
977 idr_destroy(&root->ino_idr);
979 return ERR_PTR(-ENOMEM);
985 root->syscall_ops = scops;
988 init_waitqueue_head(&root->deactivate_waitq);
990 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
997 * kernfs_destroy_root - destroy a kernfs hierarchy
998 * @root: root of the hierarchy to destroy
1000 * Destroy the hierarchy anchored at @root by removing all existing
1001 * directories and destroying @root.
1003 void kernfs_destroy_root(struct kernfs_root *root)
1005 kernfs_remove(root->kn); /* will also free @root */
1009 * kernfs_create_dir_ns - create a directory
1010 * @parent: parent in which to create a new directory
1011 * @name: name of the new directory
1012 * @mode: mode of the new directory
1013 * @uid: uid of the new directory
1014 * @gid: gid of the new directory
1015 * @priv: opaque data associated with the new directory
1016 * @ns: optional namespace tag of the directory
1018 * Returns the created node on success, ERR_PTR() value on failure.
1020 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1021 const char *name, umode_t mode,
1022 kuid_t uid, kgid_t gid,
1023 void *priv, const void *ns)
1025 struct kernfs_node *kn;
1029 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1030 uid, gid, KERNFS_DIR);
1032 return ERR_PTR(-ENOMEM);
1034 kn->dir.root = parent->dir.root;
1039 rc = kernfs_add_one(kn);
1048 * kernfs_create_empty_dir - create an always empty directory
1049 * @parent: parent in which to create a new directory
1050 * @name: name of the new directory
1052 * Returns the created node on success, ERR_PTR() value on failure.
1054 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1057 struct kernfs_node *kn;
1061 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1062 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1064 return ERR_PTR(-ENOMEM);
1066 kn->flags |= KERNFS_EMPTY_DIR;
1067 kn->dir.root = parent->dir.root;
1072 rc = kernfs_add_one(kn);
1080 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1081 struct dentry *dentry,
1085 struct kernfs_node *parent = dir->i_private;
1086 struct kernfs_node *kn;
1087 struct inode *inode;
1088 const void *ns = NULL;
1090 mutex_lock(&kernfs_mutex);
1092 if (kernfs_ns_enabled(parent))
1093 ns = kernfs_info(dir->i_sb)->ns;
1095 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1098 if (!kn || !kernfs_active(kn)) {
1103 /* attach dentry and inode */
1104 inode = kernfs_get_inode(dir->i_sb, kn);
1106 ret = ERR_PTR(-ENOMEM);
1110 /* instantiate and hash dentry */
1111 ret = d_splice_alias(inode, dentry);
1113 mutex_unlock(&kernfs_mutex);
1117 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1120 struct kernfs_node *parent = dir->i_private;
1121 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1124 if (!scops || !scops->mkdir)
1127 if (!kernfs_get_active(parent))
1130 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1132 kernfs_put_active(parent);
1136 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1138 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1139 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1142 if (!scops || !scops->rmdir)
1145 if (!kernfs_get_active(kn))
1148 ret = scops->rmdir(kn);
1150 kernfs_put_active(kn);
1154 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1155 struct inode *new_dir, struct dentry *new_dentry,
1158 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1159 struct kernfs_node *new_parent = new_dir->i_private;
1160 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1166 if (!scops || !scops->rename)
1169 if (!kernfs_get_active(kn))
1172 if (!kernfs_get_active(new_parent)) {
1173 kernfs_put_active(kn);
1177 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1179 kernfs_put_active(new_parent);
1180 kernfs_put_active(kn);
1184 const struct inode_operations kernfs_dir_iops = {
1185 .lookup = kernfs_iop_lookup,
1186 .permission = kernfs_iop_permission,
1187 .setattr = kernfs_iop_setattr,
1188 .getattr = kernfs_iop_getattr,
1189 .listxattr = kernfs_iop_listxattr,
1191 .mkdir = kernfs_iop_mkdir,
1192 .rmdir = kernfs_iop_rmdir,
1193 .rename = kernfs_iop_rename,
1196 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1198 struct kernfs_node *last;
1201 struct rb_node *rbn;
1205 if (kernfs_type(pos) != KERNFS_DIR)
1208 rbn = rb_first(&pos->dir.children);
1212 pos = rb_to_kn(rbn);
1219 * kernfs_next_descendant_post - find the next descendant for post-order walk
1220 * @pos: the current position (%NULL to initiate traversal)
1221 * @root: kernfs_node whose descendants to walk
1223 * Find the next descendant to visit for post-order traversal of @root's
1224 * descendants. @root is included in the iteration and the last node to be
1227 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1228 struct kernfs_node *root)
1230 struct rb_node *rbn;
1232 lockdep_assert_held(&kernfs_mutex);
1234 /* if first iteration, visit leftmost descendant which may be root */
1236 return kernfs_leftmost_descendant(root);
1238 /* if we visited @root, we're done */
1242 /* if there's an unvisited sibling, visit its leftmost descendant */
1243 rbn = rb_next(&pos->rb);
1245 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1247 /* no sibling left, visit parent */
1252 * kernfs_activate - activate a node which started deactivated
1253 * @kn: kernfs_node whose subtree is to be activated
1255 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1256 * needs to be explicitly activated. A node which hasn't been activated
1257 * isn't visible to userland and deactivation is skipped during its
1258 * removal. This is useful to construct atomic init sequences where
1259 * creation of multiple nodes should either succeed or fail atomically.
1261 * The caller is responsible for ensuring that this function is not called
1262 * after kernfs_remove*() is invoked on @kn.
1264 void kernfs_activate(struct kernfs_node *kn)
1266 struct kernfs_node *pos;
1268 mutex_lock(&kernfs_mutex);
1271 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1272 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1275 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1276 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1278 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1279 pos->flags |= KERNFS_ACTIVATED;
1282 mutex_unlock(&kernfs_mutex);
1285 static void __kernfs_remove(struct kernfs_node *kn)
1287 struct kernfs_node *pos;
1289 lockdep_assert_held(&kernfs_mutex);
1292 * Short-circuit if non-root @kn has already finished removal.
1293 * This is for kernfs_remove_self() which plays with active ref
1296 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1299 pr_debug("kernfs %s: removing\n", kn->name);
1301 /* prevent any new usage under @kn by deactivating all nodes */
1303 while ((pos = kernfs_next_descendant_post(pos, kn)))
1304 if (kernfs_active(pos))
1305 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1307 /* deactivate and unlink the subtree node-by-node */
1309 pos = kernfs_leftmost_descendant(kn);
1312 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1313 * base ref could have been put by someone else by the time
1314 * the function returns. Make sure it doesn't go away
1320 * Drain iff @kn was activated. This avoids draining and
1321 * its lockdep annotations for nodes which have never been
1322 * activated and allows embedding kernfs_remove() in create
1323 * error paths without worrying about draining.
1325 if (kn->flags & KERNFS_ACTIVATED)
1328 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1331 * kernfs_unlink_sibling() succeeds once per node. Use it
1332 * to decide who's responsible for cleanups.
1334 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1335 struct kernfs_iattrs *ps_iattr =
1336 pos->parent ? pos->parent->iattr : NULL;
1338 /* update timestamps on the parent */
1340 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1341 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1348 } while (pos != kn);
1352 * kernfs_remove - remove a kernfs_node recursively
1353 * @kn: the kernfs_node to remove
1355 * Remove @kn along with all its subdirectories and files.
1357 void kernfs_remove(struct kernfs_node *kn)
1359 mutex_lock(&kernfs_mutex);
1360 __kernfs_remove(kn);
1361 mutex_unlock(&kernfs_mutex);
1365 * kernfs_break_active_protection - break out of active protection
1366 * @kn: the self kernfs_node
1368 * The caller must be running off of a kernfs operation which is invoked
1369 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1370 * this function must also be matched with an invocation of
1371 * kernfs_unbreak_active_protection().
1373 * This function releases the active reference of @kn the caller is
1374 * holding. Once this function is called, @kn may be removed at any point
1375 * and the caller is solely responsible for ensuring that the objects it
1376 * dereferences are accessible.
1378 void kernfs_break_active_protection(struct kernfs_node *kn)
1381 * Take out ourself out of the active ref dependency chain. If
1382 * we're called without an active ref, lockdep will complain.
1384 kernfs_put_active(kn);
1388 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1389 * @kn: the self kernfs_node
1391 * If kernfs_break_active_protection() was called, this function must be
1392 * invoked before finishing the kernfs operation. Note that while this
1393 * function restores the active reference, it doesn't and can't actually
1394 * restore the active protection - @kn may already or be in the process of
1395 * being removed. Once kernfs_break_active_protection() is invoked, that
1396 * protection is irreversibly gone for the kernfs operation instance.
1398 * While this function may be called at any point after
1399 * kernfs_break_active_protection() is invoked, its most useful location
1400 * would be right before the enclosing kernfs operation returns.
1402 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1405 * @kn->active could be in any state; however, the increment we do
1406 * here will be undone as soon as the enclosing kernfs operation
1407 * finishes and this temporary bump can't break anything. If @kn
1408 * is alive, nothing changes. If @kn is being deactivated, the
1409 * soon-to-follow put will either finish deactivation or restore
1410 * deactivated state. If @kn is already removed, the temporary
1411 * bump is guaranteed to be gone before @kn is released.
1413 atomic_inc(&kn->active);
1414 if (kernfs_lockdep(kn))
1415 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1419 * kernfs_remove_self - remove a kernfs_node from its own method
1420 * @kn: the self kernfs_node to remove
1422 * The caller must be running off of a kernfs operation which is invoked
1423 * with an active reference - e.g. one of kernfs_ops. This can be used to
1424 * implement a file operation which deletes itself.
1426 * For example, the "delete" file for a sysfs device directory can be
1427 * implemented by invoking kernfs_remove_self() on the "delete" file
1428 * itself. This function breaks the circular dependency of trying to
1429 * deactivate self while holding an active ref itself. It isn't necessary
1430 * to modify the usual removal path to use kernfs_remove_self(). The
1431 * "delete" implementation can simply invoke kernfs_remove_self() on self
1432 * before proceeding with the usual removal path. kernfs will ignore later
1433 * kernfs_remove() on self.
1435 * kernfs_remove_self() can be called multiple times concurrently on the
1436 * same kernfs_node. Only the first one actually performs removal and
1437 * returns %true. All others will wait until the kernfs operation which
1438 * won self-removal finishes and return %false. Note that the losers wait
1439 * for the completion of not only the winning kernfs_remove_self() but also
1440 * the whole kernfs_ops which won the arbitration. This can be used to
1441 * guarantee, for example, all concurrent writes to a "delete" file to
1442 * finish only after the whole operation is complete.
1444 bool kernfs_remove_self(struct kernfs_node *kn)
1448 mutex_lock(&kernfs_mutex);
1449 kernfs_break_active_protection(kn);
1452 * SUICIDAL is used to arbitrate among competing invocations. Only
1453 * the first one will actually perform removal. When the removal
1454 * is complete, SUICIDED is set and the active ref is restored
1455 * while holding kernfs_mutex. The ones which lost arbitration
1456 * waits for SUICDED && drained which can happen only after the
1457 * enclosing kernfs operation which executed the winning instance
1458 * of kernfs_remove_self() finished.
1460 if (!(kn->flags & KERNFS_SUICIDAL)) {
1461 kn->flags |= KERNFS_SUICIDAL;
1462 __kernfs_remove(kn);
1463 kn->flags |= KERNFS_SUICIDED;
1466 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1470 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1472 if ((kn->flags & KERNFS_SUICIDED) &&
1473 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1476 mutex_unlock(&kernfs_mutex);
1478 mutex_lock(&kernfs_mutex);
1480 finish_wait(waitq, &wait);
1481 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1486 * This must be done while holding kernfs_mutex; otherwise, waiting
1487 * for SUICIDED && deactivated could finish prematurely.
1489 kernfs_unbreak_active_protection(kn);
1491 mutex_unlock(&kernfs_mutex);
1496 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1497 * @parent: parent of the target
1498 * @name: name of the kernfs_node to remove
1499 * @ns: namespace tag of the kernfs_node to remove
1501 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1502 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1504 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1507 struct kernfs_node *kn;
1510 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1515 mutex_lock(&kernfs_mutex);
1517 kn = kernfs_find_ns(parent, name, ns);
1520 __kernfs_remove(kn);
1524 mutex_unlock(&kernfs_mutex);
1533 * kernfs_rename_ns - move and rename a kernfs_node
1535 * @new_parent: new parent to put @sd under
1536 * @new_name: new name
1537 * @new_ns: new namespace tag
1539 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1540 const char *new_name, const void *new_ns)
1542 struct kernfs_node *old_parent;
1543 const char *old_name = NULL;
1546 /* can't move or rename root */
1550 mutex_lock(&kernfs_mutex);
1553 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1554 (new_parent->flags & KERNFS_EMPTY_DIR))
1558 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1559 (strcmp(kn->name, new_name) == 0))
1560 goto out; /* nothing to rename */
1563 if (kernfs_find_ns(new_parent, new_name, new_ns))
1566 /* rename kernfs_node */
1567 if (strcmp(kn->name, new_name) != 0) {
1569 new_name = kstrdup_const(new_name, GFP_KERNEL);
1577 * Move to the appropriate place in the appropriate directories rbtree.
1579 kernfs_unlink_sibling(kn);
1580 kernfs_get(new_parent);
1582 /* rename_lock protects ->parent and ->name accessors */
1583 spin_lock_irq(&kernfs_rename_lock);
1585 old_parent = kn->parent;
1586 kn->parent = new_parent;
1590 old_name = kn->name;
1591 kn->name = new_name;
1594 spin_unlock_irq(&kernfs_rename_lock);
1596 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1597 kernfs_link_sibling(kn);
1599 kernfs_put(old_parent);
1600 kfree_const(old_name);
1604 mutex_unlock(&kernfs_mutex);
1608 /* Relationship between s_mode and the DT_xxx types */
1609 static inline unsigned char dt_type(struct kernfs_node *kn)
1611 return (kn->mode >> 12) & 15;
1614 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1616 kernfs_put(filp->private_data);
1620 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1621 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1624 int valid = kernfs_active(pos) &&
1625 pos->parent == parent && hash == pos->hash;
1630 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1631 struct rb_node *node = parent->dir.children.rb_node;
1633 pos = rb_to_kn(node);
1635 if (hash < pos->hash)
1636 node = node->rb_left;
1637 else if (hash > pos->hash)
1638 node = node->rb_right;
1643 /* Skip over entries which are dying/dead or in the wrong namespace */
1644 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1645 struct rb_node *node = rb_next(&pos->rb);
1649 pos = rb_to_kn(node);
1654 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1655 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1657 pos = kernfs_dir_pos(ns, parent, ino, pos);
1660 struct rb_node *node = rb_next(&pos->rb);
1664 pos = rb_to_kn(node);
1665 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1670 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1672 struct dentry *dentry = file->f_path.dentry;
1673 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1674 struct kernfs_node *pos = file->private_data;
1675 const void *ns = NULL;
1677 if (!dir_emit_dots(file, ctx))
1679 mutex_lock(&kernfs_mutex);
1681 if (kernfs_ns_enabled(parent))
1682 ns = kernfs_info(dentry->d_sb)->ns;
1684 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1686 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1687 const char *name = pos->name;
1688 unsigned int type = dt_type(pos);
1689 int len = strlen(name);
1690 ino_t ino = pos->id.ino;
1692 ctx->pos = pos->hash;
1693 file->private_data = pos;
1696 mutex_unlock(&kernfs_mutex);
1697 if (!dir_emit(ctx, name, len, ino, type))
1699 mutex_lock(&kernfs_mutex);
1701 mutex_unlock(&kernfs_mutex);
1702 file->private_data = NULL;
1707 const struct file_operations kernfs_dir_fops = {
1708 .read = generic_read_dir,
1709 .iterate_shared = kernfs_fop_readdir,
1710 .release = kernfs_dir_fop_release,
1711 .llseek = generic_file_llseek,