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->id.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 = kernfs_dentry_node(dentry);
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 (kernfs_dentry_node(dentry->d_parent) != 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 const struct dentry_operations kernfs_dops = {
598 .d_revalidate = kernfs_dop_revalidate,
602 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
603 * @dentry: the dentry in question
605 * Return the kernfs_node associated with @dentry. If @dentry is not a
606 * kernfs one, %NULL is returned.
608 * While the returned kernfs_node will stay accessible as long as @dentry
609 * is accessible, the returned node can be in any state and the caller is
610 * fully responsible for determining what's accessible.
612 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
614 if (dentry->d_sb->s_op == &kernfs_sops &&
615 !d_really_is_negative(dentry))
616 return kernfs_dentry_node(dentry);
620 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
621 const char *name, umode_t mode,
624 struct kernfs_node *kn;
628 name = kstrdup_const(name, GFP_KERNEL);
632 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
636 idr_preload(GFP_KERNEL);
637 spin_lock(&kernfs_idr_lock);
638 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
639 if (ret >= 0 && ret < root->last_ino)
640 root->next_generation++;
641 gen = root->next_generation;
642 root->last_ino = ret;
643 spin_unlock(&kernfs_idr_lock);
648 kn->id.generation = gen;
651 * set ino first. This RELEASE is paired with atomic_inc_not_zero in
652 * kernfs_find_and_get_node_by_ino
654 atomic_set_release(&kn->count, 1);
655 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
656 RB_CLEAR_NODE(&kn->rb);
665 kmem_cache_free(kernfs_node_cache, kn);
671 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
672 const char *name, umode_t mode,
675 struct kernfs_node *kn;
677 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
686 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
687 * @root: the kernfs root
691 * NULL on failure. Return a kernfs node with reference counter incremented
693 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
696 struct kernfs_node *kn;
699 kn = idr_find(&root->ino_idr, ino);
704 * Since kernfs_node is freed in RCU, it's possible an old node for ino
705 * is freed, but reused before RCU grace period. But a freed node (see
706 * kernfs_put) or an incompletedly initialized node (see
707 * __kernfs_new_node) should have 'count' 0. We can use this fact to
708 * filter out such node.
710 if (!atomic_inc_not_zero(&kn->count)) {
716 * The node could be a new node or a reused node. If it's a new node,
717 * we are ok. If it's reused because of RCU (because of
718 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
719 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
720 * hence we can use 'ino' to filter stale node.
722 if (kn->id.ino != ino)
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 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
780 ktime_get_real_ts(&ps_iattrs->ia_ctime);
781 ps_iattrs->ia_mtime = ps_iattrs->ia_ctime;
784 mutex_unlock(&kernfs_mutex);
787 * Activate the new node unless CREATE_DEACTIVATED is requested.
788 * If not activated here, the kernfs user is responsible for
789 * activating the node with kernfs_activate(). A node which hasn't
790 * been activated is not visible to userland and its removal won't
791 * trigger deactivation.
793 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
798 mutex_unlock(&kernfs_mutex);
803 * kernfs_find_ns - find kernfs_node with the given name
804 * @parent: kernfs_node to search under
805 * @name: name to look for
806 * @ns: the namespace tag to use
808 * Look for kernfs_node with name @name under @parent. Returns pointer to
809 * the found kernfs_node on success, %NULL on failure.
811 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
812 const unsigned char *name,
815 struct rb_node *node = parent->dir.children.rb_node;
816 bool has_ns = kernfs_ns_enabled(parent);
819 lockdep_assert_held(&kernfs_mutex);
821 if (has_ns != (bool)ns) {
822 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
823 has_ns ? "required" : "invalid", parent->name, name);
827 hash = kernfs_name_hash(name, ns);
829 struct kernfs_node *kn;
833 result = kernfs_name_compare(hash, name, ns, kn);
835 node = node->rb_left;
837 node = node->rb_right;
844 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
845 const unsigned char *path,
851 lockdep_assert_held(&kernfs_mutex);
853 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
854 spin_lock_irq(&kernfs_rename_lock);
856 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
858 if (len >= sizeof(kernfs_pr_cont_buf)) {
859 spin_unlock_irq(&kernfs_rename_lock);
863 p = kernfs_pr_cont_buf;
865 while ((name = strsep(&p, "/")) && parent) {
868 parent = kernfs_find_ns(parent, name, ns);
871 spin_unlock_irq(&kernfs_rename_lock);
877 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
878 * @parent: kernfs_node to search under
879 * @name: name to look for
880 * @ns: the namespace tag to use
882 * Look for kernfs_node with name @name under @parent and get a reference
883 * if found. This function may sleep and returns pointer to the found
884 * kernfs_node on success, %NULL on failure.
886 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
887 const char *name, const void *ns)
889 struct kernfs_node *kn;
891 mutex_lock(&kernfs_mutex);
892 kn = kernfs_find_ns(parent, name, ns);
894 mutex_unlock(&kernfs_mutex);
898 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
901 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
902 * @parent: kernfs_node to search under
903 * @path: path to look for
904 * @ns: the namespace tag to use
906 * Look for kernfs_node with path @path under @parent and get a reference
907 * if found. This function may sleep and returns pointer to the found
908 * kernfs_node on success, %NULL on failure.
910 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
911 const char *path, const void *ns)
913 struct kernfs_node *kn;
915 mutex_lock(&kernfs_mutex);
916 kn = kernfs_walk_ns(parent, path, ns);
918 mutex_unlock(&kernfs_mutex);
924 * kernfs_create_root - create a new kernfs hierarchy
925 * @scops: optional syscall operations for the hierarchy
926 * @flags: KERNFS_ROOT_* flags
927 * @priv: opaque data associated with the new directory
929 * Returns the root of the new hierarchy on success, ERR_PTR() value on
932 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
933 unsigned int flags, void *priv)
935 struct kernfs_root *root;
936 struct kernfs_node *kn;
938 root = kzalloc(sizeof(*root), GFP_KERNEL);
940 return ERR_PTR(-ENOMEM);
942 idr_init(&root->ino_idr);
943 INIT_LIST_HEAD(&root->supers);
944 root->next_generation = 1;
946 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
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 * @priv: opaque data associated with the new directory
986 * @ns: optional namespace tag of the directory
988 * Returns the created node on success, ERR_PTR() value on failure.
990 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
991 const char *name, umode_t mode,
992 void *priv, const void *ns)
994 struct kernfs_node *kn;
998 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
1000 return ERR_PTR(-ENOMEM);
1002 kn->dir.root = parent->dir.root;
1007 rc = kernfs_add_one(kn);
1016 * kernfs_create_empty_dir - create an always empty directory
1017 * @parent: parent in which to create a new directory
1018 * @name: name of the new directory
1020 * Returns the created node on success, ERR_PTR() value on failure.
1022 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1025 struct kernfs_node *kn;
1029 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
1031 return ERR_PTR(-ENOMEM);
1033 kn->flags |= KERNFS_EMPTY_DIR;
1034 kn->dir.root = parent->dir.root;
1039 rc = kernfs_add_one(kn);
1047 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1048 struct dentry *dentry,
1052 struct kernfs_node *parent = dir->i_private;
1053 struct kernfs_node *kn;
1054 struct inode *inode;
1055 const void *ns = NULL;
1057 mutex_lock(&kernfs_mutex);
1059 if (kernfs_ns_enabled(parent))
1060 ns = kernfs_info(dir->i_sb)->ns;
1062 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1065 if (!kn || !kernfs_active(kn)) {
1070 /* attach dentry and inode */
1071 inode = kernfs_get_inode(dir->i_sb, kn);
1073 ret = ERR_PTR(-ENOMEM);
1077 /* instantiate and hash dentry */
1078 ret = d_splice_alias(inode, dentry);
1080 mutex_unlock(&kernfs_mutex);
1084 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1087 struct kernfs_node *parent = dir->i_private;
1088 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1091 if (!scops || !scops->mkdir)
1094 if (!kernfs_get_active(parent))
1097 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1099 kernfs_put_active(parent);
1103 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1105 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1106 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1109 if (!scops || !scops->rmdir)
1112 if (!kernfs_get_active(kn))
1115 ret = scops->rmdir(kn);
1117 kernfs_put_active(kn);
1121 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1122 struct inode *new_dir, struct dentry *new_dentry,
1125 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1126 struct kernfs_node *new_parent = new_dir->i_private;
1127 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1133 if (!scops || !scops->rename)
1136 if (!kernfs_get_active(kn))
1139 if (!kernfs_get_active(new_parent)) {
1140 kernfs_put_active(kn);
1144 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1146 kernfs_put_active(new_parent);
1147 kernfs_put_active(kn);
1151 const struct inode_operations kernfs_dir_iops = {
1152 .lookup = kernfs_iop_lookup,
1153 .permission = kernfs_iop_permission,
1154 .setattr = kernfs_iop_setattr,
1155 .getattr = kernfs_iop_getattr,
1156 .listxattr = kernfs_iop_listxattr,
1158 .mkdir = kernfs_iop_mkdir,
1159 .rmdir = kernfs_iop_rmdir,
1160 .rename = kernfs_iop_rename,
1163 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1165 struct kernfs_node *last;
1168 struct rb_node *rbn;
1172 if (kernfs_type(pos) != KERNFS_DIR)
1175 rbn = rb_first(&pos->dir.children);
1179 pos = rb_to_kn(rbn);
1186 * kernfs_next_descendant_post - find the next descendant for post-order walk
1187 * @pos: the current position (%NULL to initiate traversal)
1188 * @root: kernfs_node whose descendants to walk
1190 * Find the next descendant to visit for post-order traversal of @root's
1191 * descendants. @root is included in the iteration and the last node to be
1194 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1195 struct kernfs_node *root)
1197 struct rb_node *rbn;
1199 lockdep_assert_held(&kernfs_mutex);
1201 /* if first iteration, visit leftmost descendant which may be root */
1203 return kernfs_leftmost_descendant(root);
1205 /* if we visited @root, we're done */
1209 /* if there's an unvisited sibling, visit its leftmost descendant */
1210 rbn = rb_next(&pos->rb);
1212 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1214 /* no sibling left, visit parent */
1219 * kernfs_activate - activate a node which started deactivated
1220 * @kn: kernfs_node whose subtree is to be activated
1222 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1223 * needs to be explicitly activated. A node which hasn't been activated
1224 * isn't visible to userland and deactivation is skipped during its
1225 * removal. This is useful to construct atomic init sequences where
1226 * creation of multiple nodes should either succeed or fail atomically.
1228 * The caller is responsible for ensuring that this function is not called
1229 * after kernfs_remove*() is invoked on @kn.
1231 void kernfs_activate(struct kernfs_node *kn)
1233 struct kernfs_node *pos;
1235 mutex_lock(&kernfs_mutex);
1238 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1239 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1242 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1243 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1245 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1246 pos->flags |= KERNFS_ACTIVATED;
1249 mutex_unlock(&kernfs_mutex);
1252 static void __kernfs_remove(struct kernfs_node *kn)
1254 struct kernfs_node *pos;
1256 lockdep_assert_held(&kernfs_mutex);
1259 * Short-circuit if non-root @kn has already finished removal.
1260 * This is for kernfs_remove_self() which plays with active ref
1263 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1266 pr_debug("kernfs %s: removing\n", kn->name);
1268 /* prevent any new usage under @kn by deactivating all nodes */
1270 while ((pos = kernfs_next_descendant_post(pos, kn)))
1271 if (kernfs_active(pos))
1272 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1274 /* deactivate and unlink the subtree node-by-node */
1276 pos = kernfs_leftmost_descendant(kn);
1279 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1280 * base ref could have been put by someone else by the time
1281 * the function returns. Make sure it doesn't go away
1287 * Drain iff @kn was activated. This avoids draining and
1288 * its lockdep annotations for nodes which have never been
1289 * activated and allows embedding kernfs_remove() in create
1290 * error paths without worrying about draining.
1292 if (kn->flags & KERNFS_ACTIVATED)
1295 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1298 * kernfs_unlink_sibling() succeeds once per node. Use it
1299 * to decide who's responsible for cleanups.
1301 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1302 struct kernfs_iattrs *ps_iattr =
1303 pos->parent ? pos->parent->iattr : NULL;
1305 /* update timestamps on the parent */
1307 ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime);
1308 ps_iattr->ia_iattr.ia_mtime =
1309 ps_iattr->ia_iattr.ia_ctime;
1316 } while (pos != kn);
1320 * kernfs_remove - remove a kernfs_node recursively
1321 * @kn: the kernfs_node to remove
1323 * Remove @kn along with all its subdirectories and files.
1325 void kernfs_remove(struct kernfs_node *kn)
1327 mutex_lock(&kernfs_mutex);
1328 __kernfs_remove(kn);
1329 mutex_unlock(&kernfs_mutex);
1333 * kernfs_break_active_protection - break out of active protection
1334 * @kn: the self kernfs_node
1336 * The caller must be running off of a kernfs operation which is invoked
1337 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1338 * this function must also be matched with an invocation of
1339 * kernfs_unbreak_active_protection().
1341 * This function releases the active reference of @kn the caller is
1342 * holding. Once this function is called, @kn may be removed at any point
1343 * and the caller is solely responsible for ensuring that the objects it
1344 * dereferences are accessible.
1346 void kernfs_break_active_protection(struct kernfs_node *kn)
1349 * Take out ourself out of the active ref dependency chain. If
1350 * we're called without an active ref, lockdep will complain.
1352 kernfs_put_active(kn);
1356 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1357 * @kn: the self kernfs_node
1359 * If kernfs_break_active_protection() was called, this function must be
1360 * invoked before finishing the kernfs operation. Note that while this
1361 * function restores the active reference, it doesn't and can't actually
1362 * restore the active protection - @kn may already or be in the process of
1363 * being removed. Once kernfs_break_active_protection() is invoked, that
1364 * protection is irreversibly gone for the kernfs operation instance.
1366 * While this function may be called at any point after
1367 * kernfs_break_active_protection() is invoked, its most useful location
1368 * would be right before the enclosing kernfs operation returns.
1370 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1373 * @kn->active could be in any state; however, the increment we do
1374 * here will be undone as soon as the enclosing kernfs operation
1375 * finishes and this temporary bump can't break anything. If @kn
1376 * is alive, nothing changes. If @kn is being deactivated, the
1377 * soon-to-follow put will either finish deactivation or restore
1378 * deactivated state. If @kn is already removed, the temporary
1379 * bump is guaranteed to be gone before @kn is released.
1381 atomic_inc(&kn->active);
1382 if (kernfs_lockdep(kn))
1383 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1387 * kernfs_remove_self - remove a kernfs_node from its own method
1388 * @kn: the self kernfs_node to remove
1390 * The caller must be running off of a kernfs operation which is invoked
1391 * with an active reference - e.g. one of kernfs_ops. This can be used to
1392 * implement a file operation which deletes itself.
1394 * For example, the "delete" file for a sysfs device directory can be
1395 * implemented by invoking kernfs_remove_self() on the "delete" file
1396 * itself. This function breaks the circular dependency of trying to
1397 * deactivate self while holding an active ref itself. It isn't necessary
1398 * to modify the usual removal path to use kernfs_remove_self(). The
1399 * "delete" implementation can simply invoke kernfs_remove_self() on self
1400 * before proceeding with the usual removal path. kernfs will ignore later
1401 * kernfs_remove() on self.
1403 * kernfs_remove_self() can be called multiple times concurrently on the
1404 * same kernfs_node. Only the first one actually performs removal and
1405 * returns %true. All others will wait until the kernfs operation which
1406 * won self-removal finishes and return %false. Note that the losers wait
1407 * for the completion of not only the winning kernfs_remove_self() but also
1408 * the whole kernfs_ops which won the arbitration. This can be used to
1409 * guarantee, for example, all concurrent writes to a "delete" file to
1410 * finish only after the whole operation is complete.
1412 bool kernfs_remove_self(struct kernfs_node *kn)
1416 mutex_lock(&kernfs_mutex);
1417 kernfs_break_active_protection(kn);
1420 * SUICIDAL is used to arbitrate among competing invocations. Only
1421 * the first one will actually perform removal. When the removal
1422 * is complete, SUICIDED is set and the active ref is restored
1423 * while holding kernfs_mutex. The ones which lost arbitration
1424 * waits for SUICDED && drained which can happen only after the
1425 * enclosing kernfs operation which executed the winning instance
1426 * of kernfs_remove_self() finished.
1428 if (!(kn->flags & KERNFS_SUICIDAL)) {
1429 kn->flags |= KERNFS_SUICIDAL;
1430 __kernfs_remove(kn);
1431 kn->flags |= KERNFS_SUICIDED;
1434 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1438 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1440 if ((kn->flags & KERNFS_SUICIDED) &&
1441 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1444 mutex_unlock(&kernfs_mutex);
1446 mutex_lock(&kernfs_mutex);
1448 finish_wait(waitq, &wait);
1449 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1454 * This must be done while holding kernfs_mutex; otherwise, waiting
1455 * for SUICIDED && deactivated could finish prematurely.
1457 kernfs_unbreak_active_protection(kn);
1459 mutex_unlock(&kernfs_mutex);
1464 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1465 * @parent: parent of the target
1466 * @name: name of the kernfs_node to remove
1467 * @ns: namespace tag of the kernfs_node to remove
1469 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1470 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1472 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1475 struct kernfs_node *kn;
1478 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1483 mutex_lock(&kernfs_mutex);
1485 kn = kernfs_find_ns(parent, name, ns);
1487 __kernfs_remove(kn);
1489 mutex_unlock(&kernfs_mutex);
1498 * kernfs_rename_ns - move and rename a kernfs_node
1500 * @new_parent: new parent to put @sd under
1501 * @new_name: new name
1502 * @new_ns: new namespace tag
1504 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1505 const char *new_name, const void *new_ns)
1507 struct kernfs_node *old_parent;
1508 const char *old_name = NULL;
1511 /* can't move or rename root */
1515 mutex_lock(&kernfs_mutex);
1518 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1519 (new_parent->flags & KERNFS_EMPTY_DIR))
1523 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1524 (strcmp(kn->name, new_name) == 0))
1525 goto out; /* nothing to rename */
1528 if (kernfs_find_ns(new_parent, new_name, new_ns))
1531 /* rename kernfs_node */
1532 if (strcmp(kn->name, new_name) != 0) {
1534 new_name = kstrdup_const(new_name, GFP_KERNEL);
1542 * Move to the appropriate place in the appropriate directories rbtree.
1544 kernfs_unlink_sibling(kn);
1545 kernfs_get(new_parent);
1547 /* rename_lock protects ->parent and ->name accessors */
1548 spin_lock_irq(&kernfs_rename_lock);
1550 old_parent = kn->parent;
1551 kn->parent = new_parent;
1555 old_name = kn->name;
1556 kn->name = new_name;
1559 spin_unlock_irq(&kernfs_rename_lock);
1561 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1562 kernfs_link_sibling(kn);
1564 kernfs_put(old_parent);
1565 kfree_const(old_name);
1569 mutex_unlock(&kernfs_mutex);
1573 /* Relationship between s_mode and the DT_xxx types */
1574 static inline unsigned char dt_type(struct kernfs_node *kn)
1576 return (kn->mode >> 12) & 15;
1579 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1581 kernfs_put(filp->private_data);
1585 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1586 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1589 int valid = kernfs_active(pos) &&
1590 pos->parent == parent && hash == pos->hash;
1595 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1596 struct rb_node *node = parent->dir.children.rb_node;
1598 pos = rb_to_kn(node);
1600 if (hash < pos->hash)
1601 node = node->rb_left;
1602 else if (hash > pos->hash)
1603 node = node->rb_right;
1608 /* Skip over entries which are dying/dead or in the wrong namespace */
1609 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1610 struct rb_node *node = rb_next(&pos->rb);
1614 pos = rb_to_kn(node);
1619 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1620 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1622 pos = kernfs_dir_pos(ns, parent, ino, pos);
1625 struct rb_node *node = rb_next(&pos->rb);
1629 pos = rb_to_kn(node);
1630 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1635 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1637 struct dentry *dentry = file->f_path.dentry;
1638 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1639 struct kernfs_node *pos = file->private_data;
1640 const void *ns = NULL;
1642 if (!dir_emit_dots(file, ctx))
1644 mutex_lock(&kernfs_mutex);
1646 if (kernfs_ns_enabled(parent))
1647 ns = kernfs_info(dentry->d_sb)->ns;
1649 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1651 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1652 const char *name = pos->name;
1653 unsigned int type = dt_type(pos);
1654 int len = strlen(name);
1655 ino_t ino = pos->id.ino;
1657 ctx->pos = pos->hash;
1658 file->private_data = pos;
1661 mutex_unlock(&kernfs_mutex);
1662 if (!dir_emit(ctx, name, len, ino, type))
1664 mutex_lock(&kernfs_mutex);
1666 mutex_unlock(&kernfs_mutex);
1667 file->private_data = NULL;
1672 const struct file_operations kernfs_dir_fops = {
1673 .read = generic_read_dir,
1674 .iterate_shared = kernfs_fop_readdir,
1675 .release = kernfs_dir_fop_release,
1676 .llseek = generic_file_llseek,