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 spin_lock(&kernfs_idr_lock);
641 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
642 spin_unlock(&kernfs_idr_lock);
644 kmem_cache_free(kernfs_node_cache, kn);
650 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
651 const char *name, umode_t mode,
652 kuid_t uid, kgid_t gid,
655 struct kernfs_node *kn;
657 kn = __kernfs_new_node(kernfs_root(parent), parent,
658 name, mode, uid, gid, flags);
667 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
668 * @root: the kernfs root
669 * @id: the target node id
671 * @id's lower 32bits encode ino and upper gen. If the gen portion is
672 * zero, all generations are matched.
675 * NULL on failure. Return a kernfs node with reference counter incremented
677 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
680 struct kernfs_node *kn;
681 ino_t ino = kernfs_id_ino(id);
682 u32 gen = kernfs_id_gen(id);
684 spin_lock(&kernfs_idr_lock);
686 kn = idr_find(&root->ino_idr, (u32)ino);
690 if (sizeof(ino_t) >= sizeof(u64)) {
691 /* we looked up with the low 32bits, compare the whole */
692 if (kernfs_ino(kn) != ino)
695 /* 0 matches all generations */
696 if (unlikely(gen && kernfs_gen(kn) != gen))
701 * ACTIVATED is protected with kernfs_mutex but it was clear when
702 * @kn was added to idr and we just wanna see it set. No need to
705 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
706 !atomic_inc_not_zero(&kn->count)))
709 spin_unlock(&kernfs_idr_lock);
712 spin_unlock(&kernfs_idr_lock);
717 * kernfs_add_one - add kernfs_node to parent without warning
718 * @kn: kernfs_node to be added
720 * The caller must already have initialized @kn->parent. This
721 * function increments nlink of the parent's inode if @kn is a
722 * directory and link into the children list of the parent.
725 * 0 on success, -EEXIST if entry with the given name already
728 int kernfs_add_one(struct kernfs_node *kn)
730 struct kernfs_node *parent = kn->parent;
731 struct kernfs_iattrs *ps_iattr;
735 down_write(&kernfs_rwsem);
738 has_ns = kernfs_ns_enabled(parent);
739 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
740 has_ns ? "required" : "invalid", parent->name, kn->name))
743 if (kernfs_type(parent) != KERNFS_DIR)
747 if (parent->flags & KERNFS_EMPTY_DIR)
750 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
753 kn->hash = kernfs_name_hash(kn->name, kn->ns);
755 ret = kernfs_link_sibling(kn);
759 /* Update timestamps on the parent */
760 ps_iattr = parent->iattr;
762 ktime_get_real_ts64(&ps_iattr->ia_ctime);
763 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
766 up_write(&kernfs_rwsem);
769 * Activate the new node unless CREATE_DEACTIVATED is requested.
770 * If not activated here, the kernfs user is responsible for
771 * activating the node with kernfs_activate(). A node which hasn't
772 * been activated is not visible to userland and its removal won't
773 * trigger deactivation.
775 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
780 up_write(&kernfs_rwsem);
785 * kernfs_find_ns - find kernfs_node with the given name
786 * @parent: kernfs_node to search under
787 * @name: name to look for
788 * @ns: the namespace tag to use
790 * Look for kernfs_node with name @name under @parent. Returns pointer to
791 * the found kernfs_node on success, %NULL on failure.
793 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
794 const unsigned char *name,
797 struct rb_node *node = parent->dir.children.rb_node;
798 bool has_ns = kernfs_ns_enabled(parent);
801 lockdep_assert_held(&kernfs_rwsem);
803 if (has_ns != (bool)ns) {
804 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
805 has_ns ? "required" : "invalid", parent->name, name);
809 hash = kernfs_name_hash(name, ns);
811 struct kernfs_node *kn;
815 result = kernfs_name_compare(hash, name, ns, kn);
817 node = node->rb_left;
819 node = node->rb_right;
826 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
827 const unsigned char *path,
833 lockdep_assert_held_read(&kernfs_rwsem);
835 spin_lock_irq(&kernfs_pr_cont_lock);
837 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
839 if (len >= sizeof(kernfs_pr_cont_buf)) {
840 spin_unlock_irq(&kernfs_pr_cont_lock);
844 p = kernfs_pr_cont_buf;
846 while ((name = strsep(&p, "/")) && parent) {
849 parent = kernfs_find_ns(parent, name, ns);
852 spin_unlock_irq(&kernfs_pr_cont_lock);
858 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
859 * @parent: kernfs_node to search under
860 * @name: name to look for
861 * @ns: the namespace tag to use
863 * Look for kernfs_node with name @name under @parent and get a reference
864 * if found. This function may sleep and returns pointer to the found
865 * kernfs_node on success, %NULL on failure.
867 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
868 const char *name, const void *ns)
870 struct kernfs_node *kn;
872 down_read(&kernfs_rwsem);
873 kn = kernfs_find_ns(parent, name, ns);
875 up_read(&kernfs_rwsem);
879 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
882 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
883 * @parent: kernfs_node to search under
884 * @path: path to look for
885 * @ns: the namespace tag to use
887 * Look for kernfs_node with path @path under @parent and get a reference
888 * if found. This function may sleep and returns pointer to the found
889 * kernfs_node on success, %NULL on failure.
891 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
892 const char *path, const void *ns)
894 struct kernfs_node *kn;
896 down_read(&kernfs_rwsem);
897 kn = kernfs_walk_ns(parent, path, ns);
899 up_read(&kernfs_rwsem);
905 * kernfs_create_root - create a new kernfs hierarchy
906 * @scops: optional syscall operations for the hierarchy
907 * @flags: KERNFS_ROOT_* flags
908 * @priv: opaque data associated with the new directory
910 * Returns the root of the new hierarchy on success, ERR_PTR() value on
913 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
914 unsigned int flags, void *priv)
916 struct kernfs_root *root;
917 struct kernfs_node *kn;
919 root = kzalloc(sizeof(*root), GFP_KERNEL);
921 return ERR_PTR(-ENOMEM);
923 idr_init(&root->ino_idr);
924 INIT_LIST_HEAD(&root->supers);
927 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
928 * High bits generation. The starting value for both ino and
929 * genenration is 1. Initialize upper 32bit allocation
932 if (sizeof(ino_t) >= sizeof(u64))
933 root->id_highbits = 0;
935 root->id_highbits = 1;
937 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
938 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
941 idr_destroy(&root->ino_idr);
943 return ERR_PTR(-ENOMEM);
949 root->syscall_ops = scops;
952 init_waitqueue_head(&root->deactivate_waitq);
954 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
961 * kernfs_destroy_root - destroy a kernfs hierarchy
962 * @root: root of the hierarchy to destroy
964 * Destroy the hierarchy anchored at @root by removing all existing
965 * directories and destroying @root.
967 void kernfs_destroy_root(struct kernfs_root *root)
969 kernfs_remove(root->kn); /* will also free @root */
973 * kernfs_create_dir_ns - create a directory
974 * @parent: parent in which to create a new directory
975 * @name: name of the new directory
976 * @mode: mode of the new directory
977 * @uid: uid of the new directory
978 * @gid: gid of the new directory
979 * @priv: opaque data associated with the new directory
980 * @ns: optional namespace tag of the directory
982 * Returns the created node on success, ERR_PTR() value on failure.
984 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
985 const char *name, umode_t mode,
986 kuid_t uid, kgid_t gid,
987 void *priv, const void *ns)
989 struct kernfs_node *kn;
993 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
994 uid, gid, KERNFS_DIR);
996 return ERR_PTR(-ENOMEM);
998 kn->dir.root = parent->dir.root;
1003 rc = kernfs_add_one(kn);
1012 * kernfs_create_empty_dir - create an always empty directory
1013 * @parent: parent in which to create a new directory
1014 * @name: name of the new directory
1016 * Returns the created node on success, ERR_PTR() value on failure.
1018 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1021 struct kernfs_node *kn;
1025 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1026 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1028 return ERR_PTR(-ENOMEM);
1030 kn->flags |= KERNFS_EMPTY_DIR;
1031 kn->dir.root = parent->dir.root;
1036 rc = kernfs_add_one(kn);
1044 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1046 struct kernfs_node *kn;
1048 if (flags & LOOKUP_RCU)
1051 /* Negative hashed dentry? */
1052 if (d_really_is_negative(dentry)) {
1053 struct kernfs_node *parent;
1055 /* If the kernfs parent node has changed discard and
1056 * proceed to ->lookup.
1058 down_read(&kernfs_rwsem);
1059 spin_lock(&dentry->d_lock);
1060 parent = kernfs_dentry_node(dentry->d_parent);
1062 if (kernfs_dir_changed(parent, dentry)) {
1063 spin_unlock(&dentry->d_lock);
1064 up_read(&kernfs_rwsem);
1068 spin_unlock(&dentry->d_lock);
1069 up_read(&kernfs_rwsem);
1071 /* The kernfs parent node hasn't changed, leave the
1072 * dentry negative and return success.
1077 kn = kernfs_dentry_node(dentry);
1078 down_read(&kernfs_rwsem);
1080 /* The kernfs node has been deactivated */
1081 if (!kernfs_active(kn))
1084 /* The kernfs node has been moved? */
1085 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1088 /* The kernfs node has been renamed */
1089 if (strcmp(dentry->d_name.name, kn->name) != 0)
1092 /* The kernfs node has been moved to a different namespace */
1093 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1094 kernfs_info(dentry->d_sb)->ns != kn->ns)
1097 up_read(&kernfs_rwsem);
1100 up_read(&kernfs_rwsem);
1104 const struct dentry_operations kernfs_dops = {
1105 .d_revalidate = kernfs_dop_revalidate,
1108 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1109 struct dentry *dentry,
1112 struct kernfs_node *parent = dir->i_private;
1113 struct kernfs_node *kn;
1114 struct inode *inode = NULL;
1115 const void *ns = NULL;
1117 down_read(&kernfs_rwsem);
1118 if (kernfs_ns_enabled(parent))
1119 ns = kernfs_info(dir->i_sb)->ns;
1121 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1122 /* attach dentry and inode */
1124 /* Inactive nodes are invisible to the VFS so don't
1125 * create a negative.
1127 if (!kernfs_active(kn)) {
1128 up_read(&kernfs_rwsem);
1131 inode = kernfs_get_inode(dir->i_sb, kn);
1133 inode = ERR_PTR(-ENOMEM);
1136 * Needed for negative dentry validation.
1137 * The negative dentry can be created in kernfs_iop_lookup()
1138 * or transforms from positive dentry in dentry_unlink_inode()
1139 * called from vfs_rmdir().
1142 kernfs_set_rev(parent, dentry);
1143 up_read(&kernfs_rwsem);
1145 /* instantiate and hash (possibly negative) dentry */
1146 return d_splice_alias(inode, dentry);
1149 static int kernfs_iop_mkdir(struct user_namespace *mnt_userns,
1150 struct inode *dir, struct dentry *dentry,
1153 struct kernfs_node *parent = dir->i_private;
1154 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1157 if (!scops || !scops->mkdir)
1160 if (!kernfs_get_active(parent))
1163 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1165 kernfs_put_active(parent);
1169 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1171 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1172 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1175 if (!scops || !scops->rmdir)
1178 if (!kernfs_get_active(kn))
1181 ret = scops->rmdir(kn);
1183 kernfs_put_active(kn);
1187 static int kernfs_iop_rename(struct user_namespace *mnt_userns,
1188 struct inode *old_dir, struct dentry *old_dentry,
1189 struct inode *new_dir, struct dentry *new_dentry,
1192 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1193 struct kernfs_node *new_parent = new_dir->i_private;
1194 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1200 if (!scops || !scops->rename)
1203 if (!kernfs_get_active(kn))
1206 if (!kernfs_get_active(new_parent)) {
1207 kernfs_put_active(kn);
1211 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1213 kernfs_put_active(new_parent);
1214 kernfs_put_active(kn);
1218 const struct inode_operations kernfs_dir_iops = {
1219 .lookup = kernfs_iop_lookup,
1220 .permission = kernfs_iop_permission,
1221 .setattr = kernfs_iop_setattr,
1222 .getattr = kernfs_iop_getattr,
1223 .listxattr = kernfs_iop_listxattr,
1225 .mkdir = kernfs_iop_mkdir,
1226 .rmdir = kernfs_iop_rmdir,
1227 .rename = kernfs_iop_rename,
1230 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1232 struct kernfs_node *last;
1235 struct rb_node *rbn;
1239 if (kernfs_type(pos) != KERNFS_DIR)
1242 rbn = rb_first(&pos->dir.children);
1246 pos = rb_to_kn(rbn);
1253 * kernfs_next_descendant_post - find the next descendant for post-order walk
1254 * @pos: the current position (%NULL to initiate traversal)
1255 * @root: kernfs_node whose descendants to walk
1257 * Find the next descendant to visit for post-order traversal of @root's
1258 * descendants. @root is included in the iteration and the last node to be
1261 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1262 struct kernfs_node *root)
1264 struct rb_node *rbn;
1266 lockdep_assert_held_write(&kernfs_rwsem);
1268 /* if first iteration, visit leftmost descendant which may be root */
1270 return kernfs_leftmost_descendant(root);
1272 /* if we visited @root, we're done */
1276 /* if there's an unvisited sibling, visit its leftmost descendant */
1277 rbn = rb_next(&pos->rb);
1279 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1281 /* no sibling left, visit parent */
1286 * kernfs_activate - activate a node which started deactivated
1287 * @kn: kernfs_node whose subtree is to be activated
1289 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1290 * needs to be explicitly activated. A node which hasn't been activated
1291 * isn't visible to userland and deactivation is skipped during its
1292 * removal. This is useful to construct atomic init sequences where
1293 * creation of multiple nodes should either succeed or fail atomically.
1295 * The caller is responsible for ensuring that this function is not called
1296 * after kernfs_remove*() is invoked on @kn.
1298 void kernfs_activate(struct kernfs_node *kn)
1300 struct kernfs_node *pos;
1302 down_write(&kernfs_rwsem);
1305 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1306 if (pos->flags & KERNFS_ACTIVATED)
1309 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1310 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1312 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1313 pos->flags |= KERNFS_ACTIVATED;
1316 up_write(&kernfs_rwsem);
1319 static void __kernfs_remove(struct kernfs_node *kn)
1321 struct kernfs_node *pos;
1323 lockdep_assert_held_write(&kernfs_rwsem);
1326 * Short-circuit if non-root @kn has already finished removal.
1327 * This is for kernfs_remove_self() which plays with active ref
1330 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1333 pr_debug("kernfs %s: removing\n", kn->name);
1335 /* prevent any new usage under @kn by deactivating all nodes */
1337 while ((pos = kernfs_next_descendant_post(pos, kn)))
1338 if (kernfs_active(pos))
1339 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1341 /* deactivate and unlink the subtree node-by-node */
1343 pos = kernfs_leftmost_descendant(kn);
1346 * kernfs_drain() drops kernfs_rwsem temporarily and @pos's
1347 * base ref could have been put by someone else by the time
1348 * the function returns. Make sure it doesn't go away
1354 * Drain iff @kn was activated. This avoids draining and
1355 * its lockdep annotations for nodes which have never been
1356 * activated and allows embedding kernfs_remove() in create
1357 * error paths without worrying about draining.
1359 if (kn->flags & KERNFS_ACTIVATED)
1362 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1365 * kernfs_unlink_sibling() succeeds once per node. Use it
1366 * to decide who's responsible for cleanups.
1368 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1369 struct kernfs_iattrs *ps_iattr =
1370 pos->parent ? pos->parent->iattr : NULL;
1372 /* update timestamps on the parent */
1374 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1375 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1382 } while (pos != kn);
1386 * kernfs_remove - remove a kernfs_node recursively
1387 * @kn: the kernfs_node to remove
1389 * Remove @kn along with all its subdirectories and files.
1391 void kernfs_remove(struct kernfs_node *kn)
1393 down_write(&kernfs_rwsem);
1394 __kernfs_remove(kn);
1395 up_write(&kernfs_rwsem);
1399 * kernfs_break_active_protection - break out of active protection
1400 * @kn: the self kernfs_node
1402 * The caller must be running off of a kernfs operation which is invoked
1403 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1404 * this function must also be matched with an invocation of
1405 * kernfs_unbreak_active_protection().
1407 * This function releases the active reference of @kn the caller is
1408 * holding. Once this function is called, @kn may be removed at any point
1409 * and the caller is solely responsible for ensuring that the objects it
1410 * dereferences are accessible.
1412 void kernfs_break_active_protection(struct kernfs_node *kn)
1415 * Take out ourself out of the active ref dependency chain. If
1416 * we're called without an active ref, lockdep will complain.
1418 kernfs_put_active(kn);
1422 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1423 * @kn: the self kernfs_node
1425 * If kernfs_break_active_protection() was called, this function must be
1426 * invoked before finishing the kernfs operation. Note that while this
1427 * function restores the active reference, it doesn't and can't actually
1428 * restore the active protection - @kn may already or be in the process of
1429 * being removed. Once kernfs_break_active_protection() is invoked, that
1430 * protection is irreversibly gone for the kernfs operation instance.
1432 * While this function may be called at any point after
1433 * kernfs_break_active_protection() is invoked, its most useful location
1434 * would be right before the enclosing kernfs operation returns.
1436 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1439 * @kn->active could be in any state; however, the increment we do
1440 * here will be undone as soon as the enclosing kernfs operation
1441 * finishes and this temporary bump can't break anything. If @kn
1442 * is alive, nothing changes. If @kn is being deactivated, the
1443 * soon-to-follow put will either finish deactivation or restore
1444 * deactivated state. If @kn is already removed, the temporary
1445 * bump is guaranteed to be gone before @kn is released.
1447 atomic_inc(&kn->active);
1448 if (kernfs_lockdep(kn))
1449 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1453 * kernfs_remove_self - remove a kernfs_node from its own method
1454 * @kn: the self kernfs_node to remove
1456 * The caller must be running off of a kernfs operation which is invoked
1457 * with an active reference - e.g. one of kernfs_ops. This can be used to
1458 * implement a file operation which deletes itself.
1460 * For example, the "delete" file for a sysfs device directory can be
1461 * implemented by invoking kernfs_remove_self() on the "delete" file
1462 * itself. This function breaks the circular dependency of trying to
1463 * deactivate self while holding an active ref itself. It isn't necessary
1464 * to modify the usual removal path to use kernfs_remove_self(). The
1465 * "delete" implementation can simply invoke kernfs_remove_self() on self
1466 * before proceeding with the usual removal path. kernfs will ignore later
1467 * kernfs_remove() on self.
1469 * kernfs_remove_self() can be called multiple times concurrently on the
1470 * same kernfs_node. Only the first one actually performs removal and
1471 * returns %true. All others will wait until the kernfs operation which
1472 * won self-removal finishes and return %false. Note that the losers wait
1473 * for the completion of not only the winning kernfs_remove_self() but also
1474 * the whole kernfs_ops which won the arbitration. This can be used to
1475 * guarantee, for example, all concurrent writes to a "delete" file to
1476 * finish only after the whole operation is complete.
1478 bool kernfs_remove_self(struct kernfs_node *kn)
1482 down_write(&kernfs_rwsem);
1483 kernfs_break_active_protection(kn);
1486 * SUICIDAL is used to arbitrate among competing invocations. Only
1487 * the first one will actually perform removal. When the removal
1488 * is complete, SUICIDED is set and the active ref is restored
1489 * while kernfs_rwsem for held exclusive. The ones which lost
1490 * arbitration waits for SUICIDED && drained which can happen only
1491 * after the enclosing kernfs operation which executed the winning
1492 * instance of kernfs_remove_self() finished.
1494 if (!(kn->flags & KERNFS_SUICIDAL)) {
1495 kn->flags |= KERNFS_SUICIDAL;
1496 __kernfs_remove(kn);
1497 kn->flags |= KERNFS_SUICIDED;
1500 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1504 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1506 if ((kn->flags & KERNFS_SUICIDED) &&
1507 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1510 up_write(&kernfs_rwsem);
1512 down_write(&kernfs_rwsem);
1514 finish_wait(waitq, &wait);
1515 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1520 * This must be done while kernfs_rwsem held exclusive; otherwise,
1521 * waiting for SUICIDED && deactivated could finish prematurely.
1523 kernfs_unbreak_active_protection(kn);
1525 up_write(&kernfs_rwsem);
1530 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1531 * @parent: parent of the target
1532 * @name: name of the kernfs_node to remove
1533 * @ns: namespace tag of the kernfs_node to remove
1535 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1536 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1538 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1541 struct kernfs_node *kn;
1544 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1549 down_write(&kernfs_rwsem);
1551 kn = kernfs_find_ns(parent, name, ns);
1554 __kernfs_remove(kn);
1558 up_write(&kernfs_rwsem);
1567 * kernfs_rename_ns - move and rename a kernfs_node
1569 * @new_parent: new parent to put @sd under
1570 * @new_name: new name
1571 * @new_ns: new namespace tag
1573 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1574 const char *new_name, const void *new_ns)
1576 struct kernfs_node *old_parent;
1577 const char *old_name = NULL;
1580 /* can't move or rename root */
1584 down_write(&kernfs_rwsem);
1587 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1588 (new_parent->flags & KERNFS_EMPTY_DIR))
1592 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1593 (strcmp(kn->name, new_name) == 0))
1594 goto out; /* nothing to rename */
1597 if (kernfs_find_ns(new_parent, new_name, new_ns))
1600 /* rename kernfs_node */
1601 if (strcmp(kn->name, new_name) != 0) {
1603 new_name = kstrdup_const(new_name, GFP_KERNEL);
1611 * Move to the appropriate place in the appropriate directories rbtree.
1613 kernfs_unlink_sibling(kn);
1614 kernfs_get(new_parent);
1616 /* rename_lock protects ->parent and ->name accessors */
1617 spin_lock_irq(&kernfs_rename_lock);
1619 old_parent = kn->parent;
1620 kn->parent = new_parent;
1624 old_name = kn->name;
1625 kn->name = new_name;
1628 spin_unlock_irq(&kernfs_rename_lock);
1630 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1631 kernfs_link_sibling(kn);
1633 kernfs_put(old_parent);
1634 kfree_const(old_name);
1638 up_write(&kernfs_rwsem);
1642 /* Relationship between mode and the DT_xxx types */
1643 static inline unsigned char dt_type(struct kernfs_node *kn)
1645 return (kn->mode >> 12) & 15;
1648 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1650 kernfs_put(filp->private_data);
1654 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1655 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1658 int valid = kernfs_active(pos) &&
1659 pos->parent == parent && hash == pos->hash;
1664 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1665 struct rb_node *node = parent->dir.children.rb_node;
1667 pos = rb_to_kn(node);
1669 if (hash < pos->hash)
1670 node = node->rb_left;
1671 else if (hash > pos->hash)
1672 node = node->rb_right;
1677 /* Skip over entries which are dying/dead or in the wrong namespace */
1678 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1679 struct rb_node *node = rb_next(&pos->rb);
1683 pos = rb_to_kn(node);
1688 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1689 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1691 pos = kernfs_dir_pos(ns, parent, ino, pos);
1694 struct rb_node *node = rb_next(&pos->rb);
1698 pos = rb_to_kn(node);
1699 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1704 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1706 struct dentry *dentry = file->f_path.dentry;
1707 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1708 struct kernfs_node *pos = file->private_data;
1709 const void *ns = NULL;
1711 if (!dir_emit_dots(file, ctx))
1713 down_read(&kernfs_rwsem);
1715 if (kernfs_ns_enabled(parent))
1716 ns = kernfs_info(dentry->d_sb)->ns;
1718 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1720 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1721 const char *name = pos->name;
1722 unsigned int type = dt_type(pos);
1723 int len = strlen(name);
1724 ino_t ino = kernfs_ino(pos);
1726 ctx->pos = pos->hash;
1727 file->private_data = pos;
1730 up_read(&kernfs_rwsem);
1731 if (!dir_emit(ctx, name, len, ino, type))
1733 down_read(&kernfs_rwsem);
1735 up_read(&kernfs_rwsem);
1736 file->private_data = NULL;
1741 const struct file_operations kernfs_dir_fops = {
1742 .read = generic_read_dir,
1743 .iterate_shared = kernfs_fop_readdir,
1744 .release = kernfs_dir_fop_release,
1745 .llseek = generic_file_llseek,