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, _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, _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;
519 if (!kn || !atomic_dec_and_test(&kn->count))
521 root = kernfs_root(kn);
524 * Moving/renaming is always done while holding reference.
525 * kn->parent won't change beneath us.
529 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
530 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
531 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
533 if (kernfs_type(kn) == KERNFS_LINK)
534 kernfs_put(kn->symlink.target_kn);
536 kfree_const(kn->name);
539 simple_xattrs_free(&kn->iattr->xattrs);
540 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
542 spin_lock(&kernfs_idr_lock);
543 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
544 spin_unlock(&kernfs_idr_lock);
545 kmem_cache_free(kernfs_node_cache, kn);
549 if (atomic_dec_and_test(&kn->count))
552 /* just released the root kn, free @root too */
553 idr_destroy(&root->ino_idr);
557 EXPORT_SYMBOL_GPL(kernfs_put);
559 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
561 struct kernfs_node *kn;
563 if (flags & LOOKUP_RCU)
566 /* Always perform fresh lookup for negatives */
567 if (d_really_is_negative(dentry))
568 goto out_bad_unlocked;
570 kn = kernfs_dentry_node(dentry);
571 mutex_lock(&kernfs_mutex);
573 /* The kernfs node has been deactivated */
574 if (!kernfs_active(kn))
577 /* The kernfs node has been moved? */
578 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
581 /* The kernfs node has been renamed */
582 if (strcmp(dentry->d_name.name, kn->name) != 0)
585 /* The kernfs node has been moved to a different namespace */
586 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
587 kernfs_info(dentry->d_sb)->ns != kn->ns)
590 mutex_unlock(&kernfs_mutex);
593 mutex_unlock(&kernfs_mutex);
598 const struct dentry_operations kernfs_dops = {
599 .d_revalidate = kernfs_dop_revalidate,
603 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
604 * @dentry: the dentry in question
606 * Return the kernfs_node associated with @dentry. If @dentry is not a
607 * kernfs one, %NULL is returned.
609 * While the returned kernfs_node will stay accessible as long as @dentry
610 * is accessible, the returned node can be in any state and the caller is
611 * fully responsible for determining what's accessible.
613 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
615 if (dentry->d_sb->s_op == &kernfs_sops &&
616 !d_really_is_negative(dentry))
617 return kernfs_dentry_node(dentry);
621 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
622 struct kernfs_node *parent,
623 const char *name, umode_t mode,
624 kuid_t uid, kgid_t gid,
627 struct kernfs_node *kn;
631 name = kstrdup_const(name, GFP_KERNEL);
635 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
639 idr_preload(GFP_KERNEL);
640 spin_lock(&kernfs_idr_lock);
641 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
642 if (ret >= 0 && ret < root->last_id_lowbits)
644 id_highbits = root->id_highbits;
645 root->last_id_lowbits = ret;
646 spin_unlock(&kernfs_idr_lock);
651 kn->id = (u64)id_highbits << 32 | ret;
653 atomic_set(&kn->count, 1);
654 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
655 RB_CLEAR_NODE(&kn->rb);
661 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
662 struct iattr iattr = {
663 .ia_valid = ATTR_UID | ATTR_GID,
668 ret = __kernfs_setattr(kn, &iattr);
674 ret = security_kernfs_init_security(parent, kn);
682 spin_lock(&kernfs_idr_lock);
683 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
684 spin_unlock(&kernfs_idr_lock);
686 kmem_cache_free(kernfs_node_cache, kn);
692 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
693 const char *name, umode_t mode,
694 kuid_t uid, kgid_t gid,
697 struct kernfs_node *kn;
699 if (parent->mode & S_ISGID) {
700 /* this code block imitates inode_init_owner() for
705 gid = parent->iattr->ia_gid;
707 if (flags & KERNFS_DIR)
711 kn = __kernfs_new_node(kernfs_root(parent), parent,
712 name, mode, uid, gid, flags);
721 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
722 * @root: the kernfs root
723 * @id: the target node id
725 * @id's lower 32bits encode ino and upper gen. If the gen portion is
726 * zero, all generations are matched.
729 * NULL on failure. Return a kernfs node with reference counter incremented
731 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
734 struct kernfs_node *kn;
735 ino_t ino = kernfs_id_ino(id);
736 u32 gen = kernfs_id_gen(id);
738 spin_lock(&kernfs_idr_lock);
740 kn = idr_find(&root->ino_idr, (u32)ino);
744 if (sizeof(ino_t) >= sizeof(u64)) {
745 /* we looked up with the low 32bits, compare the whole */
746 if (kernfs_ino(kn) != ino)
749 /* 0 matches all generations */
750 if (unlikely(gen && kernfs_gen(kn) != gen))
755 * ACTIVATED is protected with kernfs_mutex but it was clear when
756 * @kn was added to idr and we just wanna see it set. No need to
759 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
760 !atomic_inc_not_zero(&kn->count)))
763 spin_unlock(&kernfs_idr_lock);
766 spin_unlock(&kernfs_idr_lock);
771 * kernfs_add_one - add kernfs_node to parent without warning
772 * @kn: kernfs_node to be added
774 * The caller must already have initialized @kn->parent. This
775 * function increments nlink of the parent's inode if @kn is a
776 * directory and link into the children list of the parent.
779 * 0 on success, -EEXIST if entry with the given name already
782 int kernfs_add_one(struct kernfs_node *kn)
784 struct kernfs_node *parent = kn->parent;
785 struct kernfs_iattrs *ps_iattr;
789 mutex_lock(&kernfs_mutex);
792 has_ns = kernfs_ns_enabled(parent);
793 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
794 has_ns ? "required" : "invalid", parent->name, kn->name))
797 if (kernfs_type(parent) != KERNFS_DIR)
801 if (parent->flags & KERNFS_EMPTY_DIR)
804 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
807 kn->hash = kernfs_name_hash(kn->name, kn->ns);
809 ret = kernfs_link_sibling(kn);
813 /* Update timestamps on the parent */
814 ps_iattr = parent->iattr;
816 ktime_get_real_ts64(&ps_iattr->ia_ctime);
817 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
820 mutex_unlock(&kernfs_mutex);
823 * Activate the new node unless CREATE_DEACTIVATED is requested.
824 * If not activated here, the kernfs user is responsible for
825 * activating the node with kernfs_activate(). A node which hasn't
826 * been activated is not visible to userland and its removal won't
827 * trigger deactivation.
829 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
834 mutex_unlock(&kernfs_mutex);
839 * kernfs_find_ns - find kernfs_node with the given name
840 * @parent: kernfs_node to search under
841 * @name: name to look for
842 * @ns: the namespace tag to use
844 * Look for kernfs_node with name @name under @parent. Returns pointer to
845 * the found kernfs_node on success, %NULL on failure.
847 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
848 const unsigned char *name,
851 struct rb_node *node = parent->dir.children.rb_node;
852 bool has_ns = kernfs_ns_enabled(parent);
855 lockdep_assert_held(&kernfs_mutex);
857 if (has_ns != (bool)ns) {
858 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
859 has_ns ? "required" : "invalid", parent->name, name);
863 hash = kernfs_name_hash(name, ns);
865 struct kernfs_node *kn;
869 result = kernfs_name_compare(hash, name, ns, kn);
871 node = node->rb_left;
873 node = node->rb_right;
880 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
881 const unsigned char *path,
887 lockdep_assert_held(&kernfs_mutex);
889 spin_lock_irq(&kernfs_pr_cont_lock);
891 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
893 if (len >= sizeof(kernfs_pr_cont_buf)) {
894 spin_unlock_irq(&kernfs_pr_cont_lock);
898 p = kernfs_pr_cont_buf;
900 while ((name = strsep(&p, "/")) && parent) {
903 parent = kernfs_find_ns(parent, name, ns);
906 spin_unlock_irq(&kernfs_pr_cont_lock);
912 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
913 * @parent: kernfs_node to search under
914 * @name: name to look for
915 * @ns: the namespace tag to use
917 * Look for kernfs_node with name @name under @parent and get a reference
918 * if found. This function may sleep and returns pointer to the found
919 * kernfs_node on success, %NULL on failure.
921 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
922 const char *name, const void *ns)
924 struct kernfs_node *kn;
926 mutex_lock(&kernfs_mutex);
927 kn = kernfs_find_ns(parent, name, ns);
929 mutex_unlock(&kernfs_mutex);
933 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
936 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
937 * @parent: kernfs_node to search under
938 * @path: path to look for
939 * @ns: the namespace tag to use
941 * Look for kernfs_node with path @path under @parent and get a reference
942 * if found. This function may sleep and returns pointer to the found
943 * kernfs_node on success, %NULL on failure.
945 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
946 const char *path, const void *ns)
948 struct kernfs_node *kn;
950 mutex_lock(&kernfs_mutex);
951 kn = kernfs_walk_ns(parent, path, ns);
953 mutex_unlock(&kernfs_mutex);
959 * kernfs_create_root - create a new kernfs hierarchy
960 * @scops: optional syscall operations for the hierarchy
961 * @flags: KERNFS_ROOT_* flags
962 * @priv: opaque data associated with the new directory
964 * Returns the root of the new hierarchy on success, ERR_PTR() value on
967 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
968 unsigned int flags, void *priv)
970 struct kernfs_root *root;
971 struct kernfs_node *kn;
973 root = kzalloc(sizeof(*root), GFP_KERNEL);
975 return ERR_PTR(-ENOMEM);
977 idr_init(&root->ino_idr);
978 INIT_LIST_HEAD(&root->supers);
981 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
982 * High bits generation. The starting value for both ino and
983 * genenration is 1. Initialize upper 32bit allocation
986 if (sizeof(ino_t) >= sizeof(u64))
987 root->id_highbits = 0;
989 root->id_highbits = 1;
991 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
992 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
995 idr_destroy(&root->ino_idr);
997 return ERR_PTR(-ENOMEM);
1001 kn->dir.root = root;
1003 root->syscall_ops = scops;
1004 root->flags = flags;
1006 init_waitqueue_head(&root->deactivate_waitq);
1008 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
1009 kernfs_activate(kn);
1015 * kernfs_destroy_root - destroy a kernfs hierarchy
1016 * @root: root of the hierarchy to destroy
1018 * Destroy the hierarchy anchored at @root by removing all existing
1019 * directories and destroying @root.
1021 void kernfs_destroy_root(struct kernfs_root *root)
1023 kernfs_remove(root->kn); /* will also free @root */
1027 * kernfs_create_dir_ns - create a directory
1028 * @parent: parent in which to create a new directory
1029 * @name: name of the new directory
1030 * @mode: mode of the new directory
1031 * @uid: uid of the new directory
1032 * @gid: gid of the new directory
1033 * @priv: opaque data associated with the new directory
1034 * @ns: optional namespace tag of the directory
1036 * Returns the created node on success, ERR_PTR() value on failure.
1038 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1039 const char *name, umode_t mode,
1040 kuid_t uid, kgid_t gid,
1041 void *priv, const void *ns)
1043 struct kernfs_node *kn;
1047 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1048 uid, gid, KERNFS_DIR);
1050 return ERR_PTR(-ENOMEM);
1052 kn->dir.root = parent->dir.root;
1057 rc = kernfs_add_one(kn);
1066 * kernfs_create_empty_dir - create an always empty directory
1067 * @parent: parent in which to create a new directory
1068 * @name: name of the new directory
1070 * Returns the created node on success, ERR_PTR() value on failure.
1072 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1075 struct kernfs_node *kn;
1079 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1080 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1082 return ERR_PTR(-ENOMEM);
1084 kn->flags |= KERNFS_EMPTY_DIR;
1085 kn->dir.root = parent->dir.root;
1090 rc = kernfs_add_one(kn);
1098 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1099 struct dentry *dentry,
1103 struct kernfs_node *parent = dir->i_private;
1104 struct kernfs_node *kn;
1105 struct inode *inode;
1106 const void *ns = NULL;
1108 mutex_lock(&kernfs_mutex);
1110 if (kernfs_ns_enabled(parent))
1111 ns = kernfs_info(dir->i_sb)->ns;
1113 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1116 if (!kn || !kernfs_active(kn)) {
1121 /* attach dentry and inode */
1122 inode = kernfs_get_inode(dir->i_sb, kn);
1124 ret = ERR_PTR(-ENOMEM);
1128 /* instantiate and hash dentry */
1129 ret = d_splice_alias(inode, dentry);
1131 mutex_unlock(&kernfs_mutex);
1135 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1138 struct kernfs_node *parent = dir->i_private;
1139 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1142 if (!scops || !scops->mkdir)
1145 if (!kernfs_get_active(parent))
1148 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1150 kernfs_put_active(parent);
1154 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1156 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1157 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1160 if (!scops || !scops->rmdir)
1163 if (!kernfs_get_active(kn))
1166 ret = scops->rmdir(kn);
1168 kernfs_put_active(kn);
1172 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1173 struct inode *new_dir, struct dentry *new_dentry,
1176 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1177 struct kernfs_node *new_parent = new_dir->i_private;
1178 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1184 if (!scops || !scops->rename)
1187 if (!kernfs_get_active(kn))
1190 if (!kernfs_get_active(new_parent)) {
1191 kernfs_put_active(kn);
1195 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1197 kernfs_put_active(new_parent);
1198 kernfs_put_active(kn);
1202 const struct inode_operations kernfs_dir_iops = {
1203 .lookup = kernfs_iop_lookup,
1204 .permission = kernfs_iop_permission,
1205 .setattr = kernfs_iop_setattr,
1206 .getattr = kernfs_iop_getattr,
1207 .listxattr = kernfs_iop_listxattr,
1209 .mkdir = kernfs_iop_mkdir,
1210 .rmdir = kernfs_iop_rmdir,
1211 .rename = kernfs_iop_rename,
1214 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1216 struct kernfs_node *last;
1219 struct rb_node *rbn;
1223 if (kernfs_type(pos) != KERNFS_DIR)
1226 rbn = rb_first(&pos->dir.children);
1230 pos = rb_to_kn(rbn);
1237 * kernfs_next_descendant_post - find the next descendant for post-order walk
1238 * @pos: the current position (%NULL to initiate traversal)
1239 * @root: kernfs_node whose descendants to walk
1241 * Find the next descendant to visit for post-order traversal of @root's
1242 * descendants. @root is included in the iteration and the last node to be
1245 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1246 struct kernfs_node *root)
1248 struct rb_node *rbn;
1250 lockdep_assert_held(&kernfs_mutex);
1252 /* if first iteration, visit leftmost descendant which may be root */
1254 return kernfs_leftmost_descendant(root);
1256 /* if we visited @root, we're done */
1260 /* if there's an unvisited sibling, visit its leftmost descendant */
1261 rbn = rb_next(&pos->rb);
1263 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1265 /* no sibling left, visit parent */
1270 * kernfs_activate - activate a node which started deactivated
1271 * @kn: kernfs_node whose subtree is to be activated
1273 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1274 * needs to be explicitly activated. A node which hasn't been activated
1275 * isn't visible to userland and deactivation is skipped during its
1276 * removal. This is useful to construct atomic init sequences where
1277 * creation of multiple nodes should either succeed or fail atomically.
1279 * The caller is responsible for ensuring that this function is not called
1280 * after kernfs_remove*() is invoked on @kn.
1282 void kernfs_activate(struct kernfs_node *kn)
1284 struct kernfs_node *pos;
1286 mutex_lock(&kernfs_mutex);
1289 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1290 if (pos->flags & KERNFS_ACTIVATED)
1293 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1294 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1296 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1297 pos->flags |= KERNFS_ACTIVATED;
1300 mutex_unlock(&kernfs_mutex);
1303 static void __kernfs_remove(struct kernfs_node *kn)
1305 struct kernfs_node *pos;
1307 lockdep_assert_held(&kernfs_mutex);
1310 * Short-circuit if non-root @kn has already finished removal.
1311 * This is for kernfs_remove_self() which plays with active ref
1314 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1317 pr_debug("kernfs %s: removing\n", kn->name);
1319 /* prevent any new usage under @kn by deactivating all nodes */
1321 while ((pos = kernfs_next_descendant_post(pos, kn)))
1322 if (kernfs_active(pos))
1323 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1325 /* deactivate and unlink the subtree node-by-node */
1327 pos = kernfs_leftmost_descendant(kn);
1330 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1331 * base ref could have been put by someone else by the time
1332 * the function returns. Make sure it doesn't go away
1338 * Drain iff @kn was activated. This avoids draining and
1339 * its lockdep annotations for nodes which have never been
1340 * activated and allows embedding kernfs_remove() in create
1341 * error paths without worrying about draining.
1343 if (kn->flags & KERNFS_ACTIVATED)
1346 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1349 * kernfs_unlink_sibling() succeeds once per node. Use it
1350 * to decide who's responsible for cleanups.
1352 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1353 struct kernfs_iattrs *ps_iattr =
1354 pos->parent ? pos->parent->iattr : NULL;
1356 /* update timestamps on the parent */
1358 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1359 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1366 } while (pos != kn);
1370 * kernfs_remove - remove a kernfs_node recursively
1371 * @kn: the kernfs_node to remove
1373 * Remove @kn along with all its subdirectories and files.
1375 void kernfs_remove(struct kernfs_node *kn)
1377 mutex_lock(&kernfs_mutex);
1378 __kernfs_remove(kn);
1379 mutex_unlock(&kernfs_mutex);
1383 * kernfs_break_active_protection - break out of active protection
1384 * @kn: the self kernfs_node
1386 * The caller must be running off of a kernfs operation which is invoked
1387 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1388 * this function must also be matched with an invocation of
1389 * kernfs_unbreak_active_protection().
1391 * This function releases the active reference of @kn the caller is
1392 * holding. Once this function is called, @kn may be removed at any point
1393 * and the caller is solely responsible for ensuring that the objects it
1394 * dereferences are accessible.
1396 void kernfs_break_active_protection(struct kernfs_node *kn)
1399 * Take out ourself out of the active ref dependency chain. If
1400 * we're called without an active ref, lockdep will complain.
1402 kernfs_put_active(kn);
1406 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1407 * @kn: the self kernfs_node
1409 * If kernfs_break_active_protection() was called, this function must be
1410 * invoked before finishing the kernfs operation. Note that while this
1411 * function restores the active reference, it doesn't and can't actually
1412 * restore the active protection - @kn may already or be in the process of
1413 * being removed. Once kernfs_break_active_protection() is invoked, that
1414 * protection is irreversibly gone for the kernfs operation instance.
1416 * While this function may be called at any point after
1417 * kernfs_break_active_protection() is invoked, its most useful location
1418 * would be right before the enclosing kernfs operation returns.
1420 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1423 * @kn->active could be in any state; however, the increment we do
1424 * here will be undone as soon as the enclosing kernfs operation
1425 * finishes and this temporary bump can't break anything. If @kn
1426 * is alive, nothing changes. If @kn is being deactivated, the
1427 * soon-to-follow put will either finish deactivation or restore
1428 * deactivated state. If @kn is already removed, the temporary
1429 * bump is guaranteed to be gone before @kn is released.
1431 atomic_inc(&kn->active);
1432 if (kernfs_lockdep(kn))
1433 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1437 * kernfs_remove_self - remove a kernfs_node from its own method
1438 * @kn: the self kernfs_node to remove
1440 * The caller must be running off of a kernfs operation which is invoked
1441 * with an active reference - e.g. one of kernfs_ops. This can be used to
1442 * implement a file operation which deletes itself.
1444 * For example, the "delete" file for a sysfs device directory can be
1445 * implemented by invoking kernfs_remove_self() on the "delete" file
1446 * itself. This function breaks the circular dependency of trying to
1447 * deactivate self while holding an active ref itself. It isn't necessary
1448 * to modify the usual removal path to use kernfs_remove_self(). The
1449 * "delete" implementation can simply invoke kernfs_remove_self() on self
1450 * before proceeding with the usual removal path. kernfs will ignore later
1451 * kernfs_remove() on self.
1453 * kernfs_remove_self() can be called multiple times concurrently on the
1454 * same kernfs_node. Only the first one actually performs removal and
1455 * returns %true. All others will wait until the kernfs operation which
1456 * won self-removal finishes and return %false. Note that the losers wait
1457 * for the completion of not only the winning kernfs_remove_self() but also
1458 * the whole kernfs_ops which won the arbitration. This can be used to
1459 * guarantee, for example, all concurrent writes to a "delete" file to
1460 * finish only after the whole operation is complete.
1462 bool kernfs_remove_self(struct kernfs_node *kn)
1466 mutex_lock(&kernfs_mutex);
1467 kernfs_break_active_protection(kn);
1470 * SUICIDAL is used to arbitrate among competing invocations. Only
1471 * the first one will actually perform removal. When the removal
1472 * is complete, SUICIDED is set and the active ref is restored
1473 * while holding kernfs_mutex. The ones which lost arbitration
1474 * waits for SUICDED && drained which can happen only after the
1475 * enclosing kernfs operation which executed the winning instance
1476 * of kernfs_remove_self() finished.
1478 if (!(kn->flags & KERNFS_SUICIDAL)) {
1479 kn->flags |= KERNFS_SUICIDAL;
1480 __kernfs_remove(kn);
1481 kn->flags |= KERNFS_SUICIDED;
1484 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1488 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1490 if ((kn->flags & KERNFS_SUICIDED) &&
1491 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1494 mutex_unlock(&kernfs_mutex);
1496 mutex_lock(&kernfs_mutex);
1498 finish_wait(waitq, &wait);
1499 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1504 * This must be done while holding kernfs_mutex; otherwise, waiting
1505 * for SUICIDED && deactivated could finish prematurely.
1507 kernfs_unbreak_active_protection(kn);
1509 mutex_unlock(&kernfs_mutex);
1514 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1515 * @parent: parent of the target
1516 * @name: name of the kernfs_node to remove
1517 * @ns: namespace tag of the kernfs_node to remove
1519 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1520 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1522 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1525 struct kernfs_node *kn;
1528 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1533 mutex_lock(&kernfs_mutex);
1535 kn = kernfs_find_ns(parent, name, ns);
1538 __kernfs_remove(kn);
1542 mutex_unlock(&kernfs_mutex);
1551 * kernfs_rename_ns - move and rename a kernfs_node
1553 * @new_parent: new parent to put @sd under
1554 * @new_name: new name
1555 * @new_ns: new namespace tag
1557 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1558 const char *new_name, const void *new_ns)
1560 struct kernfs_node *old_parent;
1561 const char *old_name = NULL;
1564 /* can't move or rename root */
1568 mutex_lock(&kernfs_mutex);
1571 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1572 (new_parent->flags & KERNFS_EMPTY_DIR))
1576 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1577 (strcmp(kn->name, new_name) == 0))
1578 goto out; /* nothing to rename */
1581 if (kernfs_find_ns(new_parent, new_name, new_ns))
1584 /* rename kernfs_node */
1585 if (strcmp(kn->name, new_name) != 0) {
1587 new_name = kstrdup_const(new_name, GFP_KERNEL);
1595 * Move to the appropriate place in the appropriate directories rbtree.
1597 kernfs_unlink_sibling(kn);
1598 kernfs_get(new_parent);
1600 /* rename_lock protects ->parent and ->name accessors */
1601 spin_lock_irq(&kernfs_rename_lock);
1603 old_parent = kn->parent;
1604 kn->parent = new_parent;
1608 old_name = kn->name;
1609 kn->name = new_name;
1612 spin_unlock_irq(&kernfs_rename_lock);
1614 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1615 kernfs_link_sibling(kn);
1617 kernfs_put(old_parent);
1618 kfree_const(old_name);
1622 mutex_unlock(&kernfs_mutex);
1626 /* Relationship between s_mode and the DT_xxx types */
1627 static inline unsigned char dt_type(struct kernfs_node *kn)
1629 return (kn->mode >> 12) & 15;
1632 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1634 kernfs_put(filp->private_data);
1638 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1639 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1642 int valid = kernfs_active(pos) &&
1643 pos->parent == parent && hash == pos->hash;
1648 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1649 struct rb_node *node = parent->dir.children.rb_node;
1651 pos = rb_to_kn(node);
1653 if (hash < pos->hash)
1654 node = node->rb_left;
1655 else if (hash > pos->hash)
1656 node = node->rb_right;
1661 /* Skip over entries which are dying/dead or in the wrong namespace */
1662 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1663 struct rb_node *node = rb_next(&pos->rb);
1667 pos = rb_to_kn(node);
1672 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1673 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1675 pos = kernfs_dir_pos(ns, parent, ino, pos);
1678 struct rb_node *node = rb_next(&pos->rb);
1682 pos = rb_to_kn(node);
1683 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1688 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1690 struct dentry *dentry = file->f_path.dentry;
1691 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1692 struct kernfs_node *pos = file->private_data;
1693 const void *ns = NULL;
1695 if (!dir_emit_dots(file, ctx))
1697 mutex_lock(&kernfs_mutex);
1699 if (kernfs_ns_enabled(parent))
1700 ns = kernfs_info(dentry->d_sb)->ns;
1702 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1704 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1705 const char *name = pos->name;
1706 unsigned int type = dt_type(pos);
1707 int len = strlen(name);
1708 ino_t ino = kernfs_ino(pos);
1710 ctx->pos = pos->hash;
1711 file->private_data = pos;
1714 mutex_unlock(&kernfs_mutex);
1715 if (!dir_emit(ctx, name, len, ino, type))
1717 mutex_lock(&kernfs_mutex);
1719 mutex_unlock(&kernfs_mutex);
1720 file->private_data = NULL;
1725 const struct file_operations kernfs_dir_fops = {
1726 .read = generic_read_dir,
1727 .iterate_shared = kernfs_fop_readdir,
1728 .release = kernfs_dir_fop_release,
1729 .llseek = generic_file_llseek,