2 * fs/kernfs/dir.c - kernfs directory implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
8 * This file is released under the GPLv2.
11 #include <linux/sched.h>
13 #include <linux/namei.h>
14 #include <linux/idr.h>
15 #include <linux/slab.h>
16 #include <linux/security.h>
17 #include <linux/hash.h>
19 #include "kernfs-internal.h"
21 DEFINE_MUTEX(kernfs_mutex);
22 static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
23 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */
24 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
26 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
28 static bool kernfs_active(struct kernfs_node *kn)
30 lockdep_assert_held(&kernfs_mutex);
31 return atomic_read(&kn->active) >= 0;
34 static bool kernfs_lockdep(struct kernfs_node *kn)
36 #ifdef CONFIG_DEBUG_LOCK_ALLOC
37 return kn->flags & KERNFS_LOCKDEP;
43 static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
46 return strlcpy(buf, "(null)", buflen);
48 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
51 /* kernfs_node_depth - compute depth from @from to @to */
52 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
56 while (to->parent && to != from) {
63 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
64 struct kernfs_node *b)
67 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
72 da = kernfs_depth(ra->kn, a);
73 db = kernfs_depth(rb->kn, b);
84 /* worst case b and a will be the same at root */
94 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
95 * where kn_from is treated as root of the path.
96 * @kn_from: kernfs node which should be treated as root for the path
97 * @kn_to: kernfs node to which path is needed
98 * @buf: buffer to copy the path into
99 * @buflen: size of @buf
101 * We need to handle couple of scenarios here:
102 * [1] when @kn_from is an ancestor of @kn_to at some level
104 * kn_to: /n1/n2/n3/n4/n5
107 * [2] when @kn_from is on a different hierarchy and we need to find common
108 * ancestor between @kn_from and @kn_to.
109 * kn_from: /n1/n2/n3/n4
113 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
114 * kn_to: /n1/n2/n3 [depth=3]
117 * [3] when @kn_to is NULL result will be "(null)"
119 * Returns the length of the full path. If the full length is equal to or
120 * greater than @buflen, @buf contains the truncated path with the trailing
121 * '\0'. On error, -errno is returned.
123 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
124 struct kernfs_node *kn_from,
125 char *buf, size_t buflen)
127 struct kernfs_node *kn, *common;
128 const char parent_str[] = "/..";
129 size_t depth_from, depth_to, len = 0;
133 return strlcpy(buf, "(null)", buflen);
136 kn_from = kernfs_root(kn_to)->kn;
138 if (kn_from == kn_to)
139 return strlcpy(buf, "/", buflen);
141 common = kernfs_common_ancestor(kn_from, kn_to);
142 if (WARN_ON(!common))
145 depth_to = kernfs_depth(common, kn_to);
146 depth_from = kernfs_depth(common, kn_from);
151 for (i = 0; i < depth_from; i++)
152 len += strlcpy(buf + len, parent_str,
153 len < buflen ? buflen - len : 0);
155 /* Calculate how many bytes we need for the rest */
156 for (i = depth_to - 1; i >= 0; i--) {
157 for (kn = kn_to, j = 0; j < i; j++)
159 len += strlcpy(buf + len, "/",
160 len < buflen ? buflen - len : 0);
161 len += strlcpy(buf + len, kn->name,
162 len < buflen ? buflen - len : 0);
169 * kernfs_name - obtain the name of a given node
170 * @kn: kernfs_node of interest
171 * @buf: buffer to copy @kn's name into
172 * @buflen: size of @buf
174 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
175 * similar to strlcpy(). It returns the length of @kn's name and if @buf
176 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
178 * Fills buffer with "(null)" if @kn is NULL.
180 * This function can be called from any context.
182 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
187 spin_lock_irqsave(&kernfs_rename_lock, flags);
188 ret = kernfs_name_locked(kn, buf, buflen);
189 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
194 * kernfs_path_from_node - build path of node @to relative to @from.
195 * @from: parent kernfs_node relative to which we need to build the path
196 * @to: kernfs_node of interest
197 * @buf: buffer to copy @to's path into
198 * @buflen: size of @buf
200 * Builds @to's path relative to @from in @buf. @from and @to must
201 * be on the same kernfs-root. If @from is not parent of @to, then a relative
202 * path (which includes '..'s) as needed to reach from @from to @to is
205 * Returns the length of the full path. If the full length is equal to or
206 * greater than @buflen, @buf contains the truncated path with the trailing
207 * '\0'. On error, -errno is returned.
209 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
210 char *buf, size_t buflen)
215 spin_lock_irqsave(&kernfs_rename_lock, flags);
216 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
217 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
220 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
223 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
224 * @kn: kernfs_node of interest
226 * This function can be called from any context.
228 void pr_cont_kernfs_name(struct kernfs_node *kn)
232 spin_lock_irqsave(&kernfs_rename_lock, flags);
234 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
235 pr_cont("%s", kernfs_pr_cont_buf);
237 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
241 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
242 * @kn: kernfs_node of interest
244 * This function can be called from any context.
246 void pr_cont_kernfs_path(struct kernfs_node *kn)
251 spin_lock_irqsave(&kernfs_rename_lock, flags);
253 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
254 sizeof(kernfs_pr_cont_buf));
260 if (sz >= sizeof(kernfs_pr_cont_buf)) {
261 pr_cont("(name too long)");
265 pr_cont("%s", kernfs_pr_cont_buf);
268 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
272 * kernfs_get_parent - determine the parent node and pin it
273 * @kn: kernfs_node of interest
275 * Determines @kn's parent, pins and returns it. This function can be
276 * called from any context.
278 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
280 struct kernfs_node *parent;
283 spin_lock_irqsave(&kernfs_rename_lock, flags);
286 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
293 * @name: Null terminated string to hash
294 * @ns: Namespace tag to hash
296 * Returns 31 bit hash of ns + name (so it fits in an off_t )
298 static unsigned int kernfs_name_hash(const char *name, const void *ns)
300 unsigned long hash = init_name_hash(ns);
301 unsigned int len = strlen(name);
303 hash = partial_name_hash(*name++, hash);
304 hash = end_name_hash(hash);
306 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
314 static int kernfs_name_compare(unsigned int hash, const char *name,
315 const void *ns, const struct kernfs_node *kn)
325 return strcmp(name, kn->name);
328 static int kernfs_sd_compare(const struct kernfs_node *left,
329 const struct kernfs_node *right)
331 return kernfs_name_compare(left->hash, left->name, left->ns, right);
335 * kernfs_link_sibling - link kernfs_node into sibling rbtree
336 * @kn: kernfs_node of interest
338 * Link @kn into its sibling rbtree which starts from
339 * @kn->parent->dir.children.
342 * mutex_lock(kernfs_mutex)
345 * 0 on susccess -EEXIST on failure.
347 static int kernfs_link_sibling(struct kernfs_node *kn)
349 struct rb_node **node = &kn->parent->dir.children.rb_node;
350 struct rb_node *parent = NULL;
353 struct kernfs_node *pos;
356 pos = rb_to_kn(*node);
358 result = kernfs_sd_compare(kn, pos);
360 node = &pos->rb.rb_left;
362 node = &pos->rb.rb_right;
367 /* add new node and rebalance the tree */
368 rb_link_node(&kn->rb, parent, node);
369 rb_insert_color(&kn->rb, &kn->parent->dir.children);
371 /* successfully added, account subdir number */
372 if (kernfs_type(kn) == KERNFS_DIR)
373 kn->parent->dir.subdirs++;
379 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
380 * @kn: kernfs_node of interest
382 * Try to unlink @kn from its sibling rbtree which starts from
383 * kn->parent->dir.children. Returns %true if @kn was actually
384 * removed, %false if @kn wasn't on the rbtree.
387 * mutex_lock(kernfs_mutex)
389 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
391 if (RB_EMPTY_NODE(&kn->rb))
394 if (kernfs_type(kn) == KERNFS_DIR)
395 kn->parent->dir.subdirs--;
397 rb_erase(&kn->rb, &kn->parent->dir.children);
398 RB_CLEAR_NODE(&kn->rb);
403 * kernfs_get_active - get an active reference to kernfs_node
404 * @kn: kernfs_node to get an active reference to
406 * Get an active reference of @kn. This function is noop if @kn
410 * Pointer to @kn on success, NULL on failure.
412 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
417 if (!atomic_inc_unless_negative(&kn->active))
420 if (kernfs_lockdep(kn))
421 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
426 * kernfs_put_active - put an active reference to kernfs_node
427 * @kn: kernfs_node to put an active reference to
429 * Put an active reference to @kn. This function is noop if @kn
432 void kernfs_put_active(struct kernfs_node *kn)
434 struct kernfs_root *root = kernfs_root(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(&root->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;
512 * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
513 * depends on this to filter reused stale node
515 if (!kn || !atomic_dec_and_test(&kn->count))
517 root = kernfs_root(kn);
520 * Moving/renaming is always done while holding reference.
521 * kn->parent won't change beneath us.
525 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
526 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
527 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
529 if (kernfs_type(kn) == KERNFS_LINK)
530 kernfs_put(kn->symlink.target_kn);
532 kfree_const(kn->name);
535 if (kn->iattr->ia_secdata)
536 security_release_secctx(kn->iattr->ia_secdata,
537 kn->iattr->ia_secdata_len);
538 simple_xattrs_free(&kn->iattr->xattrs);
541 spin_lock(&kernfs_idr_lock);
542 idr_remove(&root->ino_idr, kn->ino);
543 spin_unlock(&kernfs_idr_lock);
544 kmem_cache_free(kernfs_node_cache, kn);
548 if (atomic_dec_and_test(&kn->count))
551 /* just released the root kn, free @root too */
552 idr_destroy(&root->ino_idr);
556 EXPORT_SYMBOL_GPL(kernfs_put);
558 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
560 struct kernfs_node *kn;
562 if (flags & LOOKUP_RCU)
565 /* Always perform fresh lookup for negatives */
566 if (d_really_is_negative(dentry))
567 goto out_bad_unlocked;
569 kn = dentry->d_fsdata;
570 mutex_lock(&kernfs_mutex);
572 /* The kernfs node has been deactivated */
573 if (!kernfs_active(kn))
576 /* The kernfs node has been moved? */
577 if (dentry->d_parent->d_fsdata != kn->parent)
580 /* The kernfs node has been renamed */
581 if (strcmp(dentry->d_name.name, kn->name) != 0)
584 /* The kernfs node has been moved to a different namespace */
585 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
586 kernfs_info(dentry->d_sb)->ns != kn->ns)
589 mutex_unlock(&kernfs_mutex);
592 mutex_unlock(&kernfs_mutex);
597 static void kernfs_dop_release(struct dentry *dentry)
599 kernfs_put(dentry->d_fsdata);
602 const struct dentry_operations kernfs_dops = {
603 .d_revalidate = kernfs_dop_revalidate,
604 .d_release = kernfs_dop_release,
608 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
609 * @dentry: the dentry in question
611 * Return the kernfs_node associated with @dentry. If @dentry is not a
612 * kernfs one, %NULL is returned.
614 * While the returned kernfs_node will stay accessible as long as @dentry
615 * is accessible, the returned node can be in any state and the caller is
616 * fully responsible for determining what's accessible.
618 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
620 if (dentry->d_sb->s_op == &kernfs_sops)
621 return dentry->d_fsdata;
625 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
626 const char *name, umode_t mode,
629 struct kernfs_node *kn;
634 name = kstrdup_const(name, GFP_KERNEL);
638 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
642 idr_preload(GFP_KERNEL);
643 spin_lock(&kernfs_idr_lock);
644 cursor = idr_get_cursor(&root->ino_idr);
645 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
646 if (ret >= 0 && ret < cursor)
647 root->next_generation++;
648 gen = root->next_generation;
649 spin_unlock(&kernfs_idr_lock);
654 kn->generation = gen;
657 * set ino first. This barrier is paired with atomic_inc_not_zero in
658 * kernfs_find_and_get_node_by_ino
660 smp_mb__before_atomic();
661 atomic_set(&kn->count, 1);
662 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
663 RB_CLEAR_NODE(&kn->rb);
672 kmem_cache_free(kernfs_node_cache, kn);
678 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
679 const char *name, umode_t mode,
682 struct kernfs_node *kn;
684 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
693 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
694 * @root: the kernfs root
698 * NULL on failure. Return a kernfs node with reference counter incremented
700 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
703 struct kernfs_node *kn;
706 kn = idr_find(&root->ino_idr, ino);
711 * Since kernfs_node is freed in RCU, it's possible an old node for ino
712 * is freed, but reused before RCU grace period. But a freed node (see
713 * kernfs_put) or an incompletedly initialized node (see
714 * __kernfs_new_node) should have 'count' 0. We can use this fact to
715 * filter out such node.
717 if (!atomic_inc_not_zero(&kn->count)) {
723 * The node could be a new node or a reused node. If it's a new node,
724 * we are ok. If it's reused because of RCU (because of
725 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
726 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
727 * hence we can use 'ino' to filter stale node.
741 * kernfs_add_one - add kernfs_node to parent without warning
742 * @kn: kernfs_node to be added
744 * The caller must already have initialized @kn->parent. This
745 * function increments nlink of the parent's inode if @kn is a
746 * directory and link into the children list of the parent.
749 * 0 on success, -EEXIST if entry with the given name already
752 int kernfs_add_one(struct kernfs_node *kn)
754 struct kernfs_node *parent = kn->parent;
755 struct kernfs_iattrs *ps_iattr;
759 mutex_lock(&kernfs_mutex);
762 has_ns = kernfs_ns_enabled(parent);
763 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
764 has_ns ? "required" : "invalid", parent->name, kn->name))
767 if (kernfs_type(parent) != KERNFS_DIR)
771 if (parent->flags & KERNFS_EMPTY_DIR)
774 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
777 kn->hash = kernfs_name_hash(kn->name, kn->ns);
779 ret = kernfs_link_sibling(kn);
783 /* Update timestamps on the parent */
784 ps_iattr = parent->iattr;
786 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
787 ktime_get_real_ts(&ps_iattrs->ia_ctime);
788 ps_iattrs->ia_mtime = ps_iattrs->ia_ctime;
791 mutex_unlock(&kernfs_mutex);
794 * Activate the new node unless CREATE_DEACTIVATED is requested.
795 * If not activated here, the kernfs user is responsible for
796 * activating the node with kernfs_activate(). A node which hasn't
797 * been activated is not visible to userland and its removal won't
798 * trigger deactivation.
800 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
805 mutex_unlock(&kernfs_mutex);
810 * kernfs_find_ns - find kernfs_node with the given name
811 * @parent: kernfs_node to search under
812 * @name: name to look for
813 * @ns: the namespace tag to use
815 * Look for kernfs_node with name @name under @parent. Returns pointer to
816 * the found kernfs_node on success, %NULL on failure.
818 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
819 const unsigned char *name,
822 struct rb_node *node = parent->dir.children.rb_node;
823 bool has_ns = kernfs_ns_enabled(parent);
826 lockdep_assert_held(&kernfs_mutex);
828 if (has_ns != (bool)ns) {
829 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
830 has_ns ? "required" : "invalid", parent->name, name);
834 hash = kernfs_name_hash(name, ns);
836 struct kernfs_node *kn;
840 result = kernfs_name_compare(hash, name, ns, kn);
842 node = node->rb_left;
844 node = node->rb_right;
851 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
852 const unsigned char *path,
858 lockdep_assert_held(&kernfs_mutex);
860 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
861 spin_lock_irq(&kernfs_rename_lock);
863 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
865 if (len >= sizeof(kernfs_pr_cont_buf)) {
866 spin_unlock_irq(&kernfs_rename_lock);
870 p = kernfs_pr_cont_buf;
872 while ((name = strsep(&p, "/")) && parent) {
875 parent = kernfs_find_ns(parent, name, ns);
878 spin_unlock_irq(&kernfs_rename_lock);
884 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
885 * @parent: kernfs_node to search under
886 * @name: name to look for
887 * @ns: the namespace tag to use
889 * Look for kernfs_node with name @name under @parent and get a reference
890 * if found. This function may sleep and returns pointer to the found
891 * kernfs_node on success, %NULL on failure.
893 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
894 const char *name, const void *ns)
896 struct kernfs_node *kn;
898 mutex_lock(&kernfs_mutex);
899 kn = kernfs_find_ns(parent, name, ns);
901 mutex_unlock(&kernfs_mutex);
905 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
908 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
909 * @parent: kernfs_node to search under
910 * @path: path to look for
911 * @ns: the namespace tag to use
913 * Look for kernfs_node with path @path under @parent and get a reference
914 * if found. This function may sleep and returns pointer to the found
915 * kernfs_node on success, %NULL on failure.
917 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
918 const char *path, const void *ns)
920 struct kernfs_node *kn;
922 mutex_lock(&kernfs_mutex);
923 kn = kernfs_walk_ns(parent, path, ns);
925 mutex_unlock(&kernfs_mutex);
931 * kernfs_create_root - create a new kernfs hierarchy
932 * @scops: optional syscall operations for the hierarchy
933 * @flags: KERNFS_ROOT_* flags
934 * @priv: opaque data associated with the new directory
936 * Returns the root of the new hierarchy on success, ERR_PTR() value on
939 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
940 unsigned int flags, void *priv)
942 struct kernfs_root *root;
943 struct kernfs_node *kn;
945 root = kzalloc(sizeof(*root), GFP_KERNEL);
947 return ERR_PTR(-ENOMEM);
949 idr_init(&root->ino_idr);
950 INIT_LIST_HEAD(&root->supers);
951 root->next_generation = 1;
953 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
956 idr_destroy(&root->ino_idr);
958 return ERR_PTR(-ENOMEM);
964 root->syscall_ops = scops;
967 init_waitqueue_head(&root->deactivate_waitq);
969 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
976 * kernfs_destroy_root - destroy a kernfs hierarchy
977 * @root: root of the hierarchy to destroy
979 * Destroy the hierarchy anchored at @root by removing all existing
980 * directories and destroying @root.
982 void kernfs_destroy_root(struct kernfs_root *root)
984 kernfs_remove(root->kn); /* will also free @root */
988 * kernfs_create_dir_ns - create a directory
989 * @parent: parent in which to create a new directory
990 * @name: name of the new directory
991 * @mode: mode of the new directory
992 * @priv: opaque data associated with the new directory
993 * @ns: optional namespace tag of the directory
995 * Returns the created node on success, ERR_PTR() value on failure.
997 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
998 const char *name, umode_t mode,
999 void *priv, const void *ns)
1001 struct kernfs_node *kn;
1005 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
1007 return ERR_PTR(-ENOMEM);
1009 kn->dir.root = parent->dir.root;
1014 rc = kernfs_add_one(kn);
1023 * kernfs_create_empty_dir - create an always empty directory
1024 * @parent: parent in which to create a new directory
1025 * @name: name of the new directory
1027 * Returns the created node on success, ERR_PTR() value on failure.
1029 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1032 struct kernfs_node *kn;
1036 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
1038 return ERR_PTR(-ENOMEM);
1040 kn->flags |= KERNFS_EMPTY_DIR;
1041 kn->dir.root = parent->dir.root;
1046 rc = kernfs_add_one(kn);
1054 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1055 struct dentry *dentry,
1059 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
1060 struct kernfs_node *kn;
1061 struct inode *inode;
1062 const void *ns = NULL;
1064 mutex_lock(&kernfs_mutex);
1066 if (kernfs_ns_enabled(parent))
1067 ns = kernfs_info(dir->i_sb)->ns;
1069 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1072 if (!kn || !kernfs_active(kn)) {
1077 dentry->d_fsdata = kn;
1079 /* attach dentry and inode */
1080 inode = kernfs_get_inode(dir->i_sb, kn);
1082 ret = ERR_PTR(-ENOMEM);
1086 /* instantiate and hash dentry */
1087 ret = d_splice_alias(inode, dentry);
1089 mutex_unlock(&kernfs_mutex);
1093 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1096 struct kernfs_node *parent = dir->i_private;
1097 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1100 if (!scops || !scops->mkdir)
1103 if (!kernfs_get_active(parent))
1106 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1108 kernfs_put_active(parent);
1112 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1114 struct kernfs_node *kn = dentry->d_fsdata;
1115 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1118 if (!scops || !scops->rmdir)
1121 if (!kernfs_get_active(kn))
1124 ret = scops->rmdir(kn);
1126 kernfs_put_active(kn);
1130 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1131 struct inode *new_dir, struct dentry *new_dentry,
1134 struct kernfs_node *kn = old_dentry->d_fsdata;
1135 struct kernfs_node *new_parent = new_dir->i_private;
1136 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1142 if (!scops || !scops->rename)
1145 if (!kernfs_get_active(kn))
1148 if (!kernfs_get_active(new_parent)) {
1149 kernfs_put_active(kn);
1153 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1155 kernfs_put_active(new_parent);
1156 kernfs_put_active(kn);
1160 const struct inode_operations kernfs_dir_iops = {
1161 .lookup = kernfs_iop_lookup,
1162 .permission = kernfs_iop_permission,
1163 .setattr = kernfs_iop_setattr,
1164 .getattr = kernfs_iop_getattr,
1165 .listxattr = kernfs_iop_listxattr,
1167 .mkdir = kernfs_iop_mkdir,
1168 .rmdir = kernfs_iop_rmdir,
1169 .rename = kernfs_iop_rename,
1172 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1174 struct kernfs_node *last;
1177 struct rb_node *rbn;
1181 if (kernfs_type(pos) != KERNFS_DIR)
1184 rbn = rb_first(&pos->dir.children);
1188 pos = rb_to_kn(rbn);
1195 * kernfs_next_descendant_post - find the next descendant for post-order walk
1196 * @pos: the current position (%NULL to initiate traversal)
1197 * @root: kernfs_node whose descendants to walk
1199 * Find the next descendant to visit for post-order traversal of @root's
1200 * descendants. @root is included in the iteration and the last node to be
1203 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1204 struct kernfs_node *root)
1206 struct rb_node *rbn;
1208 lockdep_assert_held(&kernfs_mutex);
1210 /* if first iteration, visit leftmost descendant which may be root */
1212 return kernfs_leftmost_descendant(root);
1214 /* if we visited @root, we're done */
1218 /* if there's an unvisited sibling, visit its leftmost descendant */
1219 rbn = rb_next(&pos->rb);
1221 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1223 /* no sibling left, visit parent */
1228 * kernfs_activate - activate a node which started deactivated
1229 * @kn: kernfs_node whose subtree is to be activated
1231 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1232 * needs to be explicitly activated. A node which hasn't been activated
1233 * isn't visible to userland and deactivation is skipped during its
1234 * removal. This is useful to construct atomic init sequences where
1235 * creation of multiple nodes should either succeed or fail atomically.
1237 * The caller is responsible for ensuring that this function is not called
1238 * after kernfs_remove*() is invoked on @kn.
1240 void kernfs_activate(struct kernfs_node *kn)
1242 struct kernfs_node *pos;
1244 mutex_lock(&kernfs_mutex);
1247 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1248 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1251 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1252 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1254 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1255 pos->flags |= KERNFS_ACTIVATED;
1258 mutex_unlock(&kernfs_mutex);
1261 static void __kernfs_remove(struct kernfs_node *kn)
1263 struct kernfs_node *pos;
1265 lockdep_assert_held(&kernfs_mutex);
1268 * Short-circuit if non-root @kn has already finished removal.
1269 * This is for kernfs_remove_self() which plays with active ref
1272 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1275 pr_debug("kernfs %s: removing\n", kn->name);
1277 /* prevent any new usage under @kn by deactivating all nodes */
1279 while ((pos = kernfs_next_descendant_post(pos, kn)))
1280 if (kernfs_active(pos))
1281 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1283 /* deactivate and unlink the subtree node-by-node */
1285 pos = kernfs_leftmost_descendant(kn);
1288 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1289 * base ref could have been put by someone else by the time
1290 * the function returns. Make sure it doesn't go away
1296 * Drain iff @kn was activated. This avoids draining and
1297 * its lockdep annotations for nodes which have never been
1298 * activated and allows embedding kernfs_remove() in create
1299 * error paths without worrying about draining.
1301 if (kn->flags & KERNFS_ACTIVATED)
1304 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1307 * kernfs_unlink_sibling() succeeds once per node. Use it
1308 * to decide who's responsible for cleanups.
1310 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1311 struct kernfs_iattrs *ps_iattr =
1312 pos->parent ? pos->parent->iattr : NULL;
1314 /* update timestamps on the parent */
1316 ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime);
1317 ps_iattr->ia_iattr.ia_mtime =
1318 ps_iattr->ia_iattr.ia_ctime;
1325 } while (pos != kn);
1329 * kernfs_remove - remove a kernfs_node recursively
1330 * @kn: the kernfs_node to remove
1332 * Remove @kn along with all its subdirectories and files.
1334 void kernfs_remove(struct kernfs_node *kn)
1336 mutex_lock(&kernfs_mutex);
1337 __kernfs_remove(kn);
1338 mutex_unlock(&kernfs_mutex);
1342 * kernfs_break_active_protection - break out of active protection
1343 * @kn: the self kernfs_node
1345 * The caller must be running off of a kernfs operation which is invoked
1346 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1347 * this function must also be matched with an invocation of
1348 * kernfs_unbreak_active_protection().
1350 * This function releases the active reference of @kn the caller is
1351 * holding. Once this function is called, @kn may be removed at any point
1352 * and the caller is solely responsible for ensuring that the objects it
1353 * dereferences are accessible.
1355 void kernfs_break_active_protection(struct kernfs_node *kn)
1358 * Take out ourself out of the active ref dependency chain. If
1359 * we're called without an active ref, lockdep will complain.
1361 kernfs_put_active(kn);
1365 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1366 * @kn: the self kernfs_node
1368 * If kernfs_break_active_protection() was called, this function must be
1369 * invoked before finishing the kernfs operation. Note that while this
1370 * function restores the active reference, it doesn't and can't actually
1371 * restore the active protection - @kn may already or be in the process of
1372 * being removed. Once kernfs_break_active_protection() is invoked, that
1373 * protection is irreversibly gone for the kernfs operation instance.
1375 * While this function may be called at any point after
1376 * kernfs_break_active_protection() is invoked, its most useful location
1377 * would be right before the enclosing kernfs operation returns.
1379 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1382 * @kn->active could be in any state; however, the increment we do
1383 * here will be undone as soon as the enclosing kernfs operation
1384 * finishes and this temporary bump can't break anything. If @kn
1385 * is alive, nothing changes. If @kn is being deactivated, the
1386 * soon-to-follow put will either finish deactivation or restore
1387 * deactivated state. If @kn is already removed, the temporary
1388 * bump is guaranteed to be gone before @kn is released.
1390 atomic_inc(&kn->active);
1391 if (kernfs_lockdep(kn))
1392 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1396 * kernfs_remove_self - remove a kernfs_node from its own method
1397 * @kn: the self kernfs_node to remove
1399 * The caller must be running off of a kernfs operation which is invoked
1400 * with an active reference - e.g. one of kernfs_ops. This can be used to
1401 * implement a file operation which deletes itself.
1403 * For example, the "delete" file for a sysfs device directory can be
1404 * implemented by invoking kernfs_remove_self() on the "delete" file
1405 * itself. This function breaks the circular dependency of trying to
1406 * deactivate self while holding an active ref itself. It isn't necessary
1407 * to modify the usual removal path to use kernfs_remove_self(). The
1408 * "delete" implementation can simply invoke kernfs_remove_self() on self
1409 * before proceeding with the usual removal path. kernfs will ignore later
1410 * kernfs_remove() on self.
1412 * kernfs_remove_self() can be called multiple times concurrently on the
1413 * same kernfs_node. Only the first one actually performs removal and
1414 * returns %true. All others will wait until the kernfs operation which
1415 * won self-removal finishes and return %false. Note that the losers wait
1416 * for the completion of not only the winning kernfs_remove_self() but also
1417 * the whole kernfs_ops which won the arbitration. This can be used to
1418 * guarantee, for example, all concurrent writes to a "delete" file to
1419 * finish only after the whole operation is complete.
1421 bool kernfs_remove_self(struct kernfs_node *kn)
1425 mutex_lock(&kernfs_mutex);
1426 kernfs_break_active_protection(kn);
1429 * SUICIDAL is used to arbitrate among competing invocations. Only
1430 * the first one will actually perform removal. When the removal
1431 * is complete, SUICIDED is set and the active ref is restored
1432 * while holding kernfs_mutex. The ones which lost arbitration
1433 * waits for SUICDED && drained which can happen only after the
1434 * enclosing kernfs operation which executed the winning instance
1435 * of kernfs_remove_self() finished.
1437 if (!(kn->flags & KERNFS_SUICIDAL)) {
1438 kn->flags |= KERNFS_SUICIDAL;
1439 __kernfs_remove(kn);
1440 kn->flags |= KERNFS_SUICIDED;
1443 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1447 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1449 if ((kn->flags & KERNFS_SUICIDED) &&
1450 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1453 mutex_unlock(&kernfs_mutex);
1455 mutex_lock(&kernfs_mutex);
1457 finish_wait(waitq, &wait);
1458 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1463 * This must be done while holding kernfs_mutex; otherwise, waiting
1464 * for SUICIDED && deactivated could finish prematurely.
1466 kernfs_unbreak_active_protection(kn);
1468 mutex_unlock(&kernfs_mutex);
1473 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1474 * @parent: parent of the target
1475 * @name: name of the kernfs_node to remove
1476 * @ns: namespace tag of the kernfs_node to remove
1478 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1479 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1481 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1484 struct kernfs_node *kn;
1487 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1492 mutex_lock(&kernfs_mutex);
1494 kn = kernfs_find_ns(parent, name, ns);
1496 __kernfs_remove(kn);
1498 mutex_unlock(&kernfs_mutex);
1507 * kernfs_rename_ns - move and rename a kernfs_node
1509 * @new_parent: new parent to put @sd under
1510 * @new_name: new name
1511 * @new_ns: new namespace tag
1513 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1514 const char *new_name, const void *new_ns)
1516 struct kernfs_node *old_parent;
1517 const char *old_name = NULL;
1520 /* can't move or rename root */
1524 mutex_lock(&kernfs_mutex);
1527 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1528 (new_parent->flags & KERNFS_EMPTY_DIR))
1532 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1533 (strcmp(kn->name, new_name) == 0))
1534 goto out; /* nothing to rename */
1537 if (kernfs_find_ns(new_parent, new_name, new_ns))
1540 /* rename kernfs_node */
1541 if (strcmp(kn->name, new_name) != 0) {
1543 new_name = kstrdup_const(new_name, GFP_KERNEL);
1551 * Move to the appropriate place in the appropriate directories rbtree.
1553 kernfs_unlink_sibling(kn);
1554 kernfs_get(new_parent);
1556 /* rename_lock protects ->parent and ->name accessors */
1557 spin_lock_irq(&kernfs_rename_lock);
1559 old_parent = kn->parent;
1560 kn->parent = new_parent;
1564 old_name = kn->name;
1565 kn->name = new_name;
1568 spin_unlock_irq(&kernfs_rename_lock);
1570 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1571 kernfs_link_sibling(kn);
1573 kernfs_put(old_parent);
1574 kfree_const(old_name);
1578 mutex_unlock(&kernfs_mutex);
1582 /* Relationship between s_mode and the DT_xxx types */
1583 static inline unsigned char dt_type(struct kernfs_node *kn)
1585 return (kn->mode >> 12) & 15;
1588 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1590 kernfs_put(filp->private_data);
1594 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1595 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1598 int valid = kernfs_active(pos) &&
1599 pos->parent == parent && hash == pos->hash;
1604 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1605 struct rb_node *node = parent->dir.children.rb_node;
1607 pos = rb_to_kn(node);
1609 if (hash < pos->hash)
1610 node = node->rb_left;
1611 else if (hash > pos->hash)
1612 node = node->rb_right;
1617 /* Skip over entries which are dying/dead or in the wrong namespace */
1618 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1619 struct rb_node *node = rb_next(&pos->rb);
1623 pos = rb_to_kn(node);
1628 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1629 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1631 pos = kernfs_dir_pos(ns, parent, ino, pos);
1634 struct rb_node *node = rb_next(&pos->rb);
1638 pos = rb_to_kn(node);
1639 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1644 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1646 struct dentry *dentry = file->f_path.dentry;
1647 struct kernfs_node *parent = dentry->d_fsdata;
1648 struct kernfs_node *pos = file->private_data;
1649 const void *ns = NULL;
1651 if (!dir_emit_dots(file, ctx))
1653 mutex_lock(&kernfs_mutex);
1655 if (kernfs_ns_enabled(parent))
1656 ns = kernfs_info(dentry->d_sb)->ns;
1658 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1660 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1661 const char *name = pos->name;
1662 unsigned int type = dt_type(pos);
1663 int len = strlen(name);
1664 ino_t ino = pos->ino;
1666 ctx->pos = pos->hash;
1667 file->private_data = pos;
1670 mutex_unlock(&kernfs_mutex);
1671 if (!dir_emit(ctx, name, len, ino, type))
1673 mutex_lock(&kernfs_mutex);
1675 mutex_unlock(&kernfs_mutex);
1676 file->private_data = NULL;
1681 const struct file_operations kernfs_dir_fops = {
1682 .read = generic_read_dir,
1683 .iterate_shared = kernfs_fop_readdir,
1684 .release = kernfs_dir_fop_release,
1685 .llseek = generic_file_llseek,