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 DECLARE_RWSEM(kernfs_rwsem);
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_rwsem);
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 * kernfs_rwsem held exclusive
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++;
383 kernfs_inc_rev(kn->parent);
389 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
390 * @kn: kernfs_node of interest
392 * Try to unlink @kn from its sibling rbtree which starts from
393 * kn->parent->dir.children. Returns %true if @kn was actually
394 * removed, %false if @kn wasn't on the rbtree.
397 * kernfs_rwsem held exclusive
399 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
401 if (RB_EMPTY_NODE(&kn->rb))
404 if (kernfs_type(kn) == KERNFS_DIR)
405 kn->parent->dir.subdirs--;
406 kernfs_inc_rev(kn->parent);
408 rb_erase(&kn->rb, &kn->parent->dir.children);
409 RB_CLEAR_NODE(&kn->rb);
414 * kernfs_get_active - get an active reference to kernfs_node
415 * @kn: kernfs_node to get an active reference to
417 * Get an active reference of @kn. This function is noop if @kn
421 * Pointer to @kn on success, NULL on failure.
423 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
428 if (!atomic_inc_unless_negative(&kn->active))
431 if (kernfs_lockdep(kn))
432 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
437 * kernfs_put_active - put an active reference to kernfs_node
438 * @kn: kernfs_node to put an active reference to
440 * Put an active reference to @kn. This function is noop if @kn
443 void kernfs_put_active(struct kernfs_node *kn)
450 if (kernfs_lockdep(kn))
451 rwsem_release(&kn->dep_map, _RET_IP_);
452 v = atomic_dec_return(&kn->active);
453 if (likely(v != KN_DEACTIVATED_BIAS))
456 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
460 * kernfs_drain - drain kernfs_node
461 * @kn: kernfs_node to drain
463 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
464 * removers may invoke this function concurrently on @kn and all will
465 * return after draining is complete.
467 static void kernfs_drain(struct kernfs_node *kn)
468 __releases(&kernfs_rwsem) __acquires(&kernfs_rwsem)
470 struct kernfs_root *root = kernfs_root(kn);
472 lockdep_assert_held_write(&kernfs_rwsem);
473 WARN_ON_ONCE(kernfs_active(kn));
475 up_write(&kernfs_rwsem);
477 if (kernfs_lockdep(kn)) {
478 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
479 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
480 lock_contended(&kn->dep_map, _RET_IP_);
483 /* but everyone should wait for draining */
484 wait_event(root->deactivate_waitq,
485 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
487 if (kernfs_lockdep(kn)) {
488 lock_acquired(&kn->dep_map, _RET_IP_);
489 rwsem_release(&kn->dep_map, _RET_IP_);
492 kernfs_drain_open_files(kn);
494 down_write(&kernfs_rwsem);
498 * kernfs_get - get a reference count on a kernfs_node
499 * @kn: the target kernfs_node
501 void kernfs_get(struct kernfs_node *kn)
504 WARN_ON(!atomic_read(&kn->count));
505 atomic_inc(&kn->count);
508 EXPORT_SYMBOL_GPL(kernfs_get);
511 * kernfs_put - put a reference count on a kernfs_node
512 * @kn: the target kernfs_node
514 * Put a reference count of @kn and destroy it if it reached zero.
516 void kernfs_put(struct kernfs_node *kn)
518 struct kernfs_node *parent;
519 struct kernfs_root *root;
521 if (!kn || !atomic_dec_and_test(&kn->count))
523 root = kernfs_root(kn);
526 * Moving/renaming is always done while holding reference.
527 * kn->parent won't change beneath us.
531 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
532 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
533 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
535 if (kernfs_type(kn) == KERNFS_LINK)
536 kernfs_put(kn->symlink.target_kn);
538 kfree_const(kn->name);
541 simple_xattrs_free(&kn->iattr->xattrs);
542 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
544 spin_lock(&kernfs_idr_lock);
545 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
546 spin_unlock(&kernfs_idr_lock);
547 kmem_cache_free(kernfs_node_cache, kn);
551 if (atomic_dec_and_test(&kn->count))
554 /* just released the root kn, free @root too */
555 idr_destroy(&root->ino_idr);
559 EXPORT_SYMBOL_GPL(kernfs_put);
562 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
563 * @dentry: the dentry in question
565 * Return the kernfs_node associated with @dentry. If @dentry is not a
566 * kernfs one, %NULL is returned.
568 * While the returned kernfs_node will stay accessible as long as @dentry
569 * is accessible, the returned node can be in any state and the caller is
570 * fully responsible for determining what's accessible.
572 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
574 if (dentry->d_sb->s_op == &kernfs_sops)
575 return kernfs_dentry_node(dentry);
579 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
580 struct kernfs_node *parent,
581 const char *name, umode_t mode,
582 kuid_t uid, kgid_t gid,
585 struct kernfs_node *kn;
589 name = kstrdup_const(name, GFP_KERNEL);
593 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
597 idr_preload(GFP_KERNEL);
598 spin_lock(&kernfs_idr_lock);
599 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
600 if (ret >= 0 && ret < root->last_id_lowbits)
602 id_highbits = root->id_highbits;
603 root->last_id_lowbits = ret;
604 spin_unlock(&kernfs_idr_lock);
609 kn->id = (u64)id_highbits << 32 | ret;
611 atomic_set(&kn->count, 1);
612 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
613 RB_CLEAR_NODE(&kn->rb);
619 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
620 struct iattr iattr = {
621 .ia_valid = ATTR_UID | ATTR_GID,
626 ret = __kernfs_setattr(kn, &iattr);
632 ret = security_kernfs_init_security(parent, kn);
640 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
642 kmem_cache_free(kernfs_node_cache, kn);
648 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
649 const char *name, umode_t mode,
650 kuid_t uid, kgid_t gid,
653 struct kernfs_node *kn;
655 kn = __kernfs_new_node(kernfs_root(parent), parent,
656 name, mode, uid, gid, flags);
665 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
666 * @root: the kernfs root
667 * @id: the target node id
669 * @id's lower 32bits encode ino and upper gen. If the gen portion is
670 * zero, all generations are matched.
673 * NULL on failure. Return a kernfs node with reference counter incremented
675 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
678 struct kernfs_node *kn;
679 ino_t ino = kernfs_id_ino(id);
680 u32 gen = kernfs_id_gen(id);
682 spin_lock(&kernfs_idr_lock);
684 kn = idr_find(&root->ino_idr, (u32)ino);
688 if (sizeof(ino_t) >= sizeof(u64)) {
689 /* we looked up with the low 32bits, compare the whole */
690 if (kernfs_ino(kn) != ino)
693 /* 0 matches all generations */
694 if (unlikely(gen && kernfs_gen(kn) != gen))
699 * ACTIVATED is protected with kernfs_mutex but it was clear when
700 * @kn was added to idr and we just wanna see it set. No need to
703 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
704 !atomic_inc_not_zero(&kn->count)))
707 spin_unlock(&kernfs_idr_lock);
710 spin_unlock(&kernfs_idr_lock);
715 * kernfs_add_one - add kernfs_node to parent without warning
716 * @kn: kernfs_node to be added
718 * The caller must already have initialized @kn->parent. This
719 * function increments nlink of the parent's inode if @kn is a
720 * directory and link into the children list of the parent.
723 * 0 on success, -EEXIST if entry with the given name already
726 int kernfs_add_one(struct kernfs_node *kn)
728 struct kernfs_node *parent = kn->parent;
729 struct kernfs_iattrs *ps_iattr;
733 down_write(&kernfs_rwsem);
736 has_ns = kernfs_ns_enabled(parent);
737 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
738 has_ns ? "required" : "invalid", parent->name, kn->name))
741 if (kernfs_type(parent) != KERNFS_DIR)
745 if (parent->flags & KERNFS_EMPTY_DIR)
748 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
751 kn->hash = kernfs_name_hash(kn->name, kn->ns);
753 ret = kernfs_link_sibling(kn);
757 /* Update timestamps on the parent */
758 ps_iattr = parent->iattr;
760 ktime_get_real_ts64(&ps_iattr->ia_ctime);
761 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
764 up_write(&kernfs_rwsem);
767 * Activate the new node unless CREATE_DEACTIVATED is requested.
768 * If not activated here, the kernfs user is responsible for
769 * activating the node with kernfs_activate(). A node which hasn't
770 * been activated is not visible to userland and its removal won't
771 * trigger deactivation.
773 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
778 up_write(&kernfs_rwsem);
783 * kernfs_find_ns - find kernfs_node with the given name
784 * @parent: kernfs_node to search under
785 * @name: name to look for
786 * @ns: the namespace tag to use
788 * Look for kernfs_node with name @name under @parent. Returns pointer to
789 * the found kernfs_node on success, %NULL on failure.
791 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
792 const unsigned char *name,
795 struct rb_node *node = parent->dir.children.rb_node;
796 bool has_ns = kernfs_ns_enabled(parent);
799 lockdep_assert_held(&kernfs_rwsem);
801 if (has_ns != (bool)ns) {
802 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
803 has_ns ? "required" : "invalid", parent->name, name);
807 hash = kernfs_name_hash(name, ns);
809 struct kernfs_node *kn;
813 result = kernfs_name_compare(hash, name, ns, kn);
815 node = node->rb_left;
817 node = node->rb_right;
824 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
825 const unsigned char *path,
831 lockdep_assert_held_read(&kernfs_rwsem);
833 spin_lock_irq(&kernfs_pr_cont_lock);
835 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
837 if (len >= sizeof(kernfs_pr_cont_buf)) {
838 spin_unlock_irq(&kernfs_pr_cont_lock);
842 p = kernfs_pr_cont_buf;
844 while ((name = strsep(&p, "/")) && parent) {
847 parent = kernfs_find_ns(parent, name, ns);
850 spin_unlock_irq(&kernfs_pr_cont_lock);
856 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
857 * @parent: kernfs_node to search under
858 * @name: name to look for
859 * @ns: the namespace tag to use
861 * Look for kernfs_node with name @name under @parent and get a reference
862 * if found. This function may sleep and returns pointer to the found
863 * kernfs_node on success, %NULL on failure.
865 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
866 const char *name, const void *ns)
868 struct kernfs_node *kn;
870 down_read(&kernfs_rwsem);
871 kn = kernfs_find_ns(parent, name, ns);
873 up_read(&kernfs_rwsem);
877 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
880 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
881 * @parent: kernfs_node to search under
882 * @path: path to look for
883 * @ns: the namespace tag to use
885 * Look for kernfs_node with path @path under @parent and get a reference
886 * if found. This function may sleep and returns pointer to the found
887 * kernfs_node on success, %NULL on failure.
889 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
890 const char *path, const void *ns)
892 struct kernfs_node *kn;
894 down_read(&kernfs_rwsem);
895 kn = kernfs_walk_ns(parent, path, ns);
897 up_read(&kernfs_rwsem);
903 * kernfs_create_root - create a new kernfs hierarchy
904 * @scops: optional syscall operations for the hierarchy
905 * @flags: KERNFS_ROOT_* flags
906 * @priv: opaque data associated with the new directory
908 * Returns the root of the new hierarchy on success, ERR_PTR() value on
911 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
912 unsigned int flags, void *priv)
914 struct kernfs_root *root;
915 struct kernfs_node *kn;
917 root = kzalloc(sizeof(*root), GFP_KERNEL);
919 return ERR_PTR(-ENOMEM);
921 idr_init(&root->ino_idr);
922 INIT_LIST_HEAD(&root->supers);
925 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
926 * High bits generation. The starting value for both ino and
927 * genenration is 1. Initialize upper 32bit allocation
930 if (sizeof(ino_t) >= sizeof(u64))
931 root->id_highbits = 0;
933 root->id_highbits = 1;
935 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
936 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
939 idr_destroy(&root->ino_idr);
941 return ERR_PTR(-ENOMEM);
947 root->syscall_ops = scops;
950 init_waitqueue_head(&root->deactivate_waitq);
952 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
959 * kernfs_destroy_root - destroy a kernfs hierarchy
960 * @root: root of the hierarchy to destroy
962 * Destroy the hierarchy anchored at @root by removing all existing
963 * directories and destroying @root.
965 void kernfs_destroy_root(struct kernfs_root *root)
967 kernfs_remove(root->kn); /* will also free @root */
971 * kernfs_create_dir_ns - create a directory
972 * @parent: parent in which to create a new directory
973 * @name: name of the new directory
974 * @mode: mode of the new directory
975 * @uid: uid of the new directory
976 * @gid: gid of the new directory
977 * @priv: opaque data associated with the new directory
978 * @ns: optional namespace tag of the directory
980 * Returns the created node on success, ERR_PTR() value on failure.
982 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
983 const char *name, umode_t mode,
984 kuid_t uid, kgid_t gid,
985 void *priv, const void *ns)
987 struct kernfs_node *kn;
991 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
992 uid, gid, KERNFS_DIR);
994 return ERR_PTR(-ENOMEM);
996 kn->dir.root = parent->dir.root;
1001 rc = kernfs_add_one(kn);
1010 * kernfs_create_empty_dir - create an always empty directory
1011 * @parent: parent in which to create a new directory
1012 * @name: name of the new directory
1014 * Returns the created node on success, ERR_PTR() value on failure.
1016 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1019 struct kernfs_node *kn;
1023 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1024 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1026 return ERR_PTR(-ENOMEM);
1028 kn->flags |= KERNFS_EMPTY_DIR;
1029 kn->dir.root = parent->dir.root;
1034 rc = kernfs_add_one(kn);
1042 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1044 struct kernfs_node *kn;
1046 if (flags & LOOKUP_RCU)
1049 /* Negative hashed dentry? */
1050 if (d_really_is_negative(dentry)) {
1051 struct kernfs_node *parent;
1053 /* If the kernfs parent node has changed discard and
1054 * proceed to ->lookup.
1056 down_read(&kernfs_rwsem);
1057 spin_lock(&dentry->d_lock);
1058 parent = kernfs_dentry_node(dentry->d_parent);
1060 if (kernfs_dir_changed(parent, dentry)) {
1061 spin_unlock(&dentry->d_lock);
1062 up_read(&kernfs_rwsem);
1066 spin_unlock(&dentry->d_lock);
1067 up_read(&kernfs_rwsem);
1069 /* The kernfs parent node hasn't changed, leave the
1070 * dentry negative and return success.
1075 kn = kernfs_dentry_node(dentry);
1076 down_read(&kernfs_rwsem);
1078 /* The kernfs node has been deactivated */
1079 if (!kernfs_active(kn))
1082 /* The kernfs node has been moved? */
1083 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1086 /* The kernfs node has been renamed */
1087 if (strcmp(dentry->d_name.name, kn->name) != 0)
1090 /* The kernfs node has been moved to a different namespace */
1091 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1092 kernfs_info(dentry->d_sb)->ns != kn->ns)
1095 up_read(&kernfs_rwsem);
1098 up_read(&kernfs_rwsem);
1102 const struct dentry_operations kernfs_dops = {
1103 .d_revalidate = kernfs_dop_revalidate,
1106 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1107 struct dentry *dentry,
1110 struct kernfs_node *parent = dir->i_private;
1111 struct kernfs_node *kn;
1112 struct inode *inode = NULL;
1113 const void *ns = NULL;
1115 down_read(&kernfs_rwsem);
1116 if (kernfs_ns_enabled(parent))
1117 ns = kernfs_info(dir->i_sb)->ns;
1119 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1120 /* attach dentry and inode */
1122 /* Inactive nodes are invisible to the VFS so don't
1123 * create a negative.
1125 if (!kernfs_active(kn)) {
1126 up_read(&kernfs_rwsem);
1129 inode = kernfs_get_inode(dir->i_sb, kn);
1131 inode = ERR_PTR(-ENOMEM);
1134 * Needed for negative dentry validation.
1135 * The negative dentry can be created in kernfs_iop_lookup()
1136 * or transforms from positive dentry in dentry_unlink_inode()
1137 * called from vfs_rmdir().
1140 kernfs_set_rev(parent, dentry);
1141 up_read(&kernfs_rwsem);
1143 /* instantiate and hash (possibly negative) dentry */
1144 return d_splice_alias(inode, dentry);
1147 static int kernfs_iop_mkdir(struct user_namespace *mnt_userns,
1148 struct inode *dir, struct dentry *dentry,
1151 struct kernfs_node *parent = dir->i_private;
1152 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1155 if (!scops || !scops->mkdir)
1158 if (!kernfs_get_active(parent))
1161 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1163 kernfs_put_active(parent);
1167 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1169 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1170 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1173 if (!scops || !scops->rmdir)
1176 if (!kernfs_get_active(kn))
1179 ret = scops->rmdir(kn);
1181 kernfs_put_active(kn);
1185 static int kernfs_iop_rename(struct user_namespace *mnt_userns,
1186 struct inode *old_dir, struct dentry *old_dentry,
1187 struct inode *new_dir, struct dentry *new_dentry,
1190 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1191 struct kernfs_node *new_parent = new_dir->i_private;
1192 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1198 if (!scops || !scops->rename)
1201 if (!kernfs_get_active(kn))
1204 if (!kernfs_get_active(new_parent)) {
1205 kernfs_put_active(kn);
1209 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1211 kernfs_put_active(new_parent);
1212 kernfs_put_active(kn);
1216 const struct inode_operations kernfs_dir_iops = {
1217 .lookup = kernfs_iop_lookup,
1218 .permission = kernfs_iop_permission,
1219 .setattr = kernfs_iop_setattr,
1220 .getattr = kernfs_iop_getattr,
1221 .listxattr = kernfs_iop_listxattr,
1223 .mkdir = kernfs_iop_mkdir,
1224 .rmdir = kernfs_iop_rmdir,
1225 .rename = kernfs_iop_rename,
1228 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1230 struct kernfs_node *last;
1233 struct rb_node *rbn;
1237 if (kernfs_type(pos) != KERNFS_DIR)
1240 rbn = rb_first(&pos->dir.children);
1244 pos = rb_to_kn(rbn);
1251 * kernfs_next_descendant_post - find the next descendant for post-order walk
1252 * @pos: the current position (%NULL to initiate traversal)
1253 * @root: kernfs_node whose descendants to walk
1255 * Find the next descendant to visit for post-order traversal of @root's
1256 * descendants. @root is included in the iteration and the last node to be
1259 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1260 struct kernfs_node *root)
1262 struct rb_node *rbn;
1264 lockdep_assert_held_write(&kernfs_rwsem);
1266 /* if first iteration, visit leftmost descendant which may be root */
1268 return kernfs_leftmost_descendant(root);
1270 /* if we visited @root, we're done */
1274 /* if there's an unvisited sibling, visit its leftmost descendant */
1275 rbn = rb_next(&pos->rb);
1277 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1279 /* no sibling left, visit parent */
1284 * kernfs_activate - activate a node which started deactivated
1285 * @kn: kernfs_node whose subtree is to be activated
1287 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1288 * needs to be explicitly activated. A node which hasn't been activated
1289 * isn't visible to userland and deactivation is skipped during its
1290 * removal. This is useful to construct atomic init sequences where
1291 * creation of multiple nodes should either succeed or fail atomically.
1293 * The caller is responsible for ensuring that this function is not called
1294 * after kernfs_remove*() is invoked on @kn.
1296 void kernfs_activate(struct kernfs_node *kn)
1298 struct kernfs_node *pos;
1300 down_write(&kernfs_rwsem);
1303 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1304 if (pos->flags & KERNFS_ACTIVATED)
1307 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1308 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1310 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1311 pos->flags |= KERNFS_ACTIVATED;
1314 up_write(&kernfs_rwsem);
1317 static void __kernfs_remove(struct kernfs_node *kn)
1319 struct kernfs_node *pos;
1321 lockdep_assert_held_write(&kernfs_rwsem);
1324 * Short-circuit if non-root @kn has already finished removal.
1325 * This is for kernfs_remove_self() which plays with active ref
1328 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1331 pr_debug("kernfs %s: removing\n", kn->name);
1333 /* prevent any new usage under @kn by deactivating all nodes */
1335 while ((pos = kernfs_next_descendant_post(pos, kn)))
1336 if (kernfs_active(pos))
1337 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1339 /* deactivate and unlink the subtree node-by-node */
1341 pos = kernfs_leftmost_descendant(kn);
1344 * kernfs_drain() drops kernfs_rwsem temporarily and @pos's
1345 * base ref could have been put by someone else by the time
1346 * the function returns. Make sure it doesn't go away
1352 * Drain iff @kn was activated. This avoids draining and
1353 * its lockdep annotations for nodes which have never been
1354 * activated and allows embedding kernfs_remove() in create
1355 * error paths without worrying about draining.
1357 if (kn->flags & KERNFS_ACTIVATED)
1360 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1363 * kernfs_unlink_sibling() succeeds once per node. Use it
1364 * to decide who's responsible for cleanups.
1366 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1367 struct kernfs_iattrs *ps_iattr =
1368 pos->parent ? pos->parent->iattr : NULL;
1370 /* update timestamps on the parent */
1372 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1373 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1380 } while (pos != kn);
1384 * kernfs_remove - remove a kernfs_node recursively
1385 * @kn: the kernfs_node to remove
1387 * Remove @kn along with all its subdirectories and files.
1389 void kernfs_remove(struct kernfs_node *kn)
1391 down_write(&kernfs_rwsem);
1392 __kernfs_remove(kn);
1393 up_write(&kernfs_rwsem);
1397 * kernfs_break_active_protection - break out of active protection
1398 * @kn: the self kernfs_node
1400 * The caller must be running off of a kernfs operation which is invoked
1401 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1402 * this function must also be matched with an invocation of
1403 * kernfs_unbreak_active_protection().
1405 * This function releases the active reference of @kn the caller is
1406 * holding. Once this function is called, @kn may be removed at any point
1407 * and the caller is solely responsible for ensuring that the objects it
1408 * dereferences are accessible.
1410 void kernfs_break_active_protection(struct kernfs_node *kn)
1413 * Take out ourself out of the active ref dependency chain. If
1414 * we're called without an active ref, lockdep will complain.
1416 kernfs_put_active(kn);
1420 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1421 * @kn: the self kernfs_node
1423 * If kernfs_break_active_protection() was called, this function must be
1424 * invoked before finishing the kernfs operation. Note that while this
1425 * function restores the active reference, it doesn't and can't actually
1426 * restore the active protection - @kn may already or be in the process of
1427 * being removed. Once kernfs_break_active_protection() is invoked, that
1428 * protection is irreversibly gone for the kernfs operation instance.
1430 * While this function may be called at any point after
1431 * kernfs_break_active_protection() is invoked, its most useful location
1432 * would be right before the enclosing kernfs operation returns.
1434 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1437 * @kn->active could be in any state; however, the increment we do
1438 * here will be undone as soon as the enclosing kernfs operation
1439 * finishes and this temporary bump can't break anything. If @kn
1440 * is alive, nothing changes. If @kn is being deactivated, the
1441 * soon-to-follow put will either finish deactivation or restore
1442 * deactivated state. If @kn is already removed, the temporary
1443 * bump is guaranteed to be gone before @kn is released.
1445 atomic_inc(&kn->active);
1446 if (kernfs_lockdep(kn))
1447 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1451 * kernfs_remove_self - remove a kernfs_node from its own method
1452 * @kn: the self kernfs_node to remove
1454 * The caller must be running off of a kernfs operation which is invoked
1455 * with an active reference - e.g. one of kernfs_ops. This can be used to
1456 * implement a file operation which deletes itself.
1458 * For example, the "delete" file for a sysfs device directory can be
1459 * implemented by invoking kernfs_remove_self() on the "delete" file
1460 * itself. This function breaks the circular dependency of trying to
1461 * deactivate self while holding an active ref itself. It isn't necessary
1462 * to modify the usual removal path to use kernfs_remove_self(). The
1463 * "delete" implementation can simply invoke kernfs_remove_self() on self
1464 * before proceeding with the usual removal path. kernfs will ignore later
1465 * kernfs_remove() on self.
1467 * kernfs_remove_self() can be called multiple times concurrently on the
1468 * same kernfs_node. Only the first one actually performs removal and
1469 * returns %true. All others will wait until the kernfs operation which
1470 * won self-removal finishes and return %false. Note that the losers wait
1471 * for the completion of not only the winning kernfs_remove_self() but also
1472 * the whole kernfs_ops which won the arbitration. This can be used to
1473 * guarantee, for example, all concurrent writes to a "delete" file to
1474 * finish only after the whole operation is complete.
1476 bool kernfs_remove_self(struct kernfs_node *kn)
1480 down_write(&kernfs_rwsem);
1481 kernfs_break_active_protection(kn);
1484 * SUICIDAL is used to arbitrate among competing invocations. Only
1485 * the first one will actually perform removal. When the removal
1486 * is complete, SUICIDED is set and the active ref is restored
1487 * while kernfs_rwsem for held exclusive. The ones which lost
1488 * arbitration waits for SUICIDED && drained which can happen only
1489 * after the enclosing kernfs operation which executed the winning
1490 * instance of kernfs_remove_self() finished.
1492 if (!(kn->flags & KERNFS_SUICIDAL)) {
1493 kn->flags |= KERNFS_SUICIDAL;
1494 __kernfs_remove(kn);
1495 kn->flags |= KERNFS_SUICIDED;
1498 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1502 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1504 if ((kn->flags & KERNFS_SUICIDED) &&
1505 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1508 up_write(&kernfs_rwsem);
1510 down_write(&kernfs_rwsem);
1512 finish_wait(waitq, &wait);
1513 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1518 * This must be done while kernfs_rwsem held exclusive; otherwise,
1519 * waiting for SUICIDED && deactivated could finish prematurely.
1521 kernfs_unbreak_active_protection(kn);
1523 up_write(&kernfs_rwsem);
1528 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1529 * @parent: parent of the target
1530 * @name: name of the kernfs_node to remove
1531 * @ns: namespace tag of the kernfs_node to remove
1533 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1534 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1536 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1539 struct kernfs_node *kn;
1542 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1547 down_write(&kernfs_rwsem);
1549 kn = kernfs_find_ns(parent, name, ns);
1551 __kernfs_remove(kn);
1553 up_write(&kernfs_rwsem);
1562 * kernfs_rename_ns - move and rename a kernfs_node
1564 * @new_parent: new parent to put @sd under
1565 * @new_name: new name
1566 * @new_ns: new namespace tag
1568 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1569 const char *new_name, const void *new_ns)
1571 struct kernfs_node *old_parent;
1572 const char *old_name = NULL;
1575 /* can't move or rename root */
1579 down_write(&kernfs_rwsem);
1582 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1583 (new_parent->flags & KERNFS_EMPTY_DIR))
1587 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1588 (strcmp(kn->name, new_name) == 0))
1589 goto out; /* nothing to rename */
1592 if (kernfs_find_ns(new_parent, new_name, new_ns))
1595 /* rename kernfs_node */
1596 if (strcmp(kn->name, new_name) != 0) {
1598 new_name = kstrdup_const(new_name, GFP_KERNEL);
1606 * Move to the appropriate place in the appropriate directories rbtree.
1608 kernfs_unlink_sibling(kn);
1609 kernfs_get(new_parent);
1611 /* rename_lock protects ->parent and ->name accessors */
1612 spin_lock_irq(&kernfs_rename_lock);
1614 old_parent = kn->parent;
1615 kn->parent = new_parent;
1619 old_name = kn->name;
1620 kn->name = new_name;
1623 spin_unlock_irq(&kernfs_rename_lock);
1625 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1626 kernfs_link_sibling(kn);
1628 kernfs_put(old_parent);
1629 kfree_const(old_name);
1633 up_write(&kernfs_rwsem);
1637 /* Relationship between mode and the DT_xxx types */
1638 static inline unsigned char dt_type(struct kernfs_node *kn)
1640 return (kn->mode >> 12) & 15;
1643 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1645 kernfs_put(filp->private_data);
1649 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1650 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1653 int valid = kernfs_active(pos) &&
1654 pos->parent == parent && hash == pos->hash;
1659 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1660 struct rb_node *node = parent->dir.children.rb_node;
1662 pos = rb_to_kn(node);
1664 if (hash < pos->hash)
1665 node = node->rb_left;
1666 else if (hash > pos->hash)
1667 node = node->rb_right;
1672 /* Skip over entries which are dying/dead or in the wrong namespace */
1673 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1674 struct rb_node *node = rb_next(&pos->rb);
1678 pos = rb_to_kn(node);
1683 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1684 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1686 pos = kernfs_dir_pos(ns, parent, ino, pos);
1689 struct rb_node *node = rb_next(&pos->rb);
1693 pos = rb_to_kn(node);
1694 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1699 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1701 struct dentry *dentry = file->f_path.dentry;
1702 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1703 struct kernfs_node *pos = file->private_data;
1704 const void *ns = NULL;
1706 if (!dir_emit_dots(file, ctx))
1708 down_read(&kernfs_rwsem);
1710 if (kernfs_ns_enabled(parent))
1711 ns = kernfs_info(dentry->d_sb)->ns;
1713 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1715 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1716 const char *name = pos->name;
1717 unsigned int type = dt_type(pos);
1718 int len = strlen(name);
1719 ino_t ino = kernfs_ino(pos);
1721 ctx->pos = pos->hash;
1722 file->private_data = pos;
1725 up_read(&kernfs_rwsem);
1726 if (!dir_emit(ctx, name, len, ino, type))
1728 down_read(&kernfs_rwsem);
1730 up_read(&kernfs_rwsem);
1731 file->private_data = NULL;
1736 const struct file_operations kernfs_dir_fops = {
1737 .read = generic_read_dir,
1738 .iterate_shared = kernfs_fop_readdir,
1739 .release = kernfs_dir_fop_release,
1740 .llseek = generic_file_llseek,