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 static DEFINE_RWLOCK(kernfs_rename_lock); /* kn->parent and ->name */
22 * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
23 * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
24 * will perform wakeups when releasing console_sem. Holding rename_lock
25 * will introduce deadlock if the scheduler reads the kernfs_name in the
28 static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
29 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by pr_cont_lock */
30 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
32 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
34 static bool __kernfs_active(struct kernfs_node *kn)
36 return atomic_read(&kn->active) >= 0;
39 static bool kernfs_active(struct kernfs_node *kn)
41 lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
42 return __kernfs_active(kn);
45 static bool kernfs_lockdep(struct kernfs_node *kn)
47 #ifdef CONFIG_DEBUG_LOCK_ALLOC
48 return kn->flags & KERNFS_LOCKDEP;
54 static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
57 return strlcpy(buf, "(null)", buflen);
59 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
62 /* kernfs_node_depth - compute depth from @from to @to */
63 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
67 while (to->parent && to != from) {
74 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
75 struct kernfs_node *b)
78 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
83 da = kernfs_depth(ra->kn, a);
84 db = kernfs_depth(rb->kn, b);
95 /* worst case b and a will be the same at root */
105 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
106 * where kn_from is treated as root of the path.
107 * @kn_from: kernfs node which should be treated as root for the path
108 * @kn_to: kernfs node to which path is needed
109 * @buf: buffer to copy the path into
110 * @buflen: size of @buf
112 * We need to handle couple of scenarios here:
113 * [1] when @kn_from is an ancestor of @kn_to at some level
115 * kn_to: /n1/n2/n3/n4/n5
118 * [2] when @kn_from is on a different hierarchy and we need to find common
119 * ancestor between @kn_from and @kn_to.
120 * kn_from: /n1/n2/n3/n4
124 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
125 * kn_to: /n1/n2/n3 [depth=3]
128 * [3] when @kn_to is %NULL result will be "(null)"
130 * Return: the length of the full path. If the full length is equal to or
131 * greater than @buflen, @buf contains the truncated path with the trailing
132 * '\0'. On error, -errno is returned.
134 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
135 struct kernfs_node *kn_from,
136 char *buf, size_t buflen)
138 struct kernfs_node *kn, *common;
139 const char parent_str[] = "/..";
140 size_t depth_from, depth_to, len = 0;
144 return strlcpy(buf, "(null)", buflen);
147 kn_from = kernfs_root(kn_to)->kn;
149 if (kn_from == kn_to)
150 return strlcpy(buf, "/", buflen);
152 common = kernfs_common_ancestor(kn_from, kn_to);
153 if (WARN_ON(!common))
156 depth_to = kernfs_depth(common, kn_to);
157 depth_from = kernfs_depth(common, kn_from);
161 for (i = 0; i < depth_from; i++)
162 len += strlcpy(buf + len, parent_str,
163 len < buflen ? buflen - len : 0);
165 /* Calculate how many bytes we need for the rest */
166 for (i = depth_to - 1; i >= 0; i--) {
167 for (kn = kn_to, j = 0; j < i; j++)
169 len += strlcpy(buf + len, "/",
170 len < buflen ? buflen - len : 0);
171 len += strlcpy(buf + len, kn->name,
172 len < buflen ? buflen - len : 0);
179 * kernfs_name - obtain the name of a given node
180 * @kn: kernfs_node of interest
181 * @buf: buffer to copy @kn's name into
182 * @buflen: size of @buf
184 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
185 * similar to strlcpy().
187 * Fills buffer with "(null)" if @kn is %NULL.
189 * Return: the length of @kn's name and if @buf isn't long enough,
190 * it's filled up to @buflen-1 and nul terminated.
192 * This function can be called from any context.
194 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
199 read_lock_irqsave(&kernfs_rename_lock, flags);
200 ret = kernfs_name_locked(kn, buf, buflen);
201 read_unlock_irqrestore(&kernfs_rename_lock, flags);
206 * kernfs_path_from_node - build path of node @to relative to @from.
207 * @from: parent kernfs_node relative to which we need to build the path
208 * @to: kernfs_node of interest
209 * @buf: buffer to copy @to's path into
210 * @buflen: size of @buf
212 * Builds @to's path relative to @from in @buf. @from and @to must
213 * be on the same kernfs-root. If @from is not parent of @to, then a relative
214 * path (which includes '..'s) as needed to reach from @from to @to is
217 * Return: the length of the full path. If the full length is equal to or
218 * greater than @buflen, @buf contains the truncated path with the trailing
219 * '\0'. On error, -errno is returned.
221 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
222 char *buf, size_t buflen)
227 read_lock_irqsave(&kernfs_rename_lock, flags);
228 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
229 read_unlock_irqrestore(&kernfs_rename_lock, flags);
232 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
235 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
236 * @kn: kernfs_node of interest
238 * This function can be called from any context.
240 void pr_cont_kernfs_name(struct kernfs_node *kn)
244 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
246 kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
247 pr_cont("%s", kernfs_pr_cont_buf);
249 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
253 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
254 * @kn: kernfs_node of interest
256 * This function can be called from any context.
258 void pr_cont_kernfs_path(struct kernfs_node *kn)
263 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
265 sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
266 sizeof(kernfs_pr_cont_buf));
272 if (sz >= sizeof(kernfs_pr_cont_buf)) {
273 pr_cont("(name too long)");
277 pr_cont("%s", kernfs_pr_cont_buf);
280 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
284 * kernfs_get_parent - determine the parent node and pin it
285 * @kn: kernfs_node of interest
287 * Determines @kn's parent, pins and returns it. This function can be
288 * called from any context.
290 * Return: parent node of @kn
292 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
294 struct kernfs_node *parent;
297 read_lock_irqsave(&kernfs_rename_lock, flags);
300 read_unlock_irqrestore(&kernfs_rename_lock, flags);
306 * kernfs_name_hash - calculate hash of @ns + @name
307 * @name: Null terminated string to hash
308 * @ns: Namespace tag to hash
310 * Return: 31-bit hash of ns + name (so it fits in an off_t)
312 static unsigned int kernfs_name_hash(const char *name, const void *ns)
314 unsigned long hash = init_name_hash(ns);
315 unsigned int len = strlen(name);
317 hash = partial_name_hash(*name++, hash);
318 hash = end_name_hash(hash);
320 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
328 static int kernfs_name_compare(unsigned int hash, const char *name,
329 const void *ns, const struct kernfs_node *kn)
339 return strcmp(name, kn->name);
342 static int kernfs_sd_compare(const struct kernfs_node *left,
343 const struct kernfs_node *right)
345 return kernfs_name_compare(left->hash, left->name, left->ns, right);
349 * kernfs_link_sibling - link kernfs_node into sibling rbtree
350 * @kn: kernfs_node of interest
352 * Link @kn into its sibling rbtree which starts from
353 * @kn->parent->dir.children.
356 * kernfs_rwsem held exclusive
359 * %0 on success, -EEXIST on failure.
361 static int kernfs_link_sibling(struct kernfs_node *kn)
363 struct rb_node **node = &kn->parent->dir.children.rb_node;
364 struct rb_node *parent = NULL;
367 struct kernfs_node *pos;
370 pos = rb_to_kn(*node);
372 result = kernfs_sd_compare(kn, pos);
374 node = &pos->rb.rb_left;
376 node = &pos->rb.rb_right;
381 /* add new node and rebalance the tree */
382 rb_link_node(&kn->rb, parent, node);
383 rb_insert_color(&kn->rb, &kn->parent->dir.children);
385 /* successfully added, account subdir number */
386 if (kernfs_type(kn) == KERNFS_DIR)
387 kn->parent->dir.subdirs++;
388 kernfs_inc_rev(kn->parent);
394 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
395 * @kn: kernfs_node of interest
397 * Try to unlink @kn from its sibling rbtree which starts from
398 * kn->parent->dir.children.
400 * Return: %true if @kn was actually removed,
401 * %false if @kn wasn't on the rbtree.
404 * kernfs_rwsem held exclusive
406 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
408 if (RB_EMPTY_NODE(&kn->rb))
411 if (kernfs_type(kn) == KERNFS_DIR)
412 kn->parent->dir.subdirs--;
413 kernfs_inc_rev(kn->parent);
415 rb_erase(&kn->rb, &kn->parent->dir.children);
416 RB_CLEAR_NODE(&kn->rb);
421 * kernfs_get_active - get an active reference to kernfs_node
422 * @kn: kernfs_node to get an active reference to
424 * Get an active reference of @kn. This function is noop if @kn
428 * Pointer to @kn on success, %NULL on failure.
430 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
435 if (!atomic_inc_unless_negative(&kn->active))
438 if (kernfs_lockdep(kn))
439 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
444 * kernfs_put_active - put an active reference to kernfs_node
445 * @kn: kernfs_node to put an active reference to
447 * Put an active reference to @kn. This function is noop if @kn
450 void kernfs_put_active(struct kernfs_node *kn)
457 if (kernfs_lockdep(kn))
458 rwsem_release(&kn->dep_map, _RET_IP_);
459 v = atomic_dec_return(&kn->active);
460 if (likely(v != KN_DEACTIVATED_BIAS))
463 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
467 * kernfs_drain - drain kernfs_node
468 * @kn: kernfs_node to drain
470 * Drain existing usages and nuke all existing mmaps of @kn. Multiple
471 * removers may invoke this function concurrently on @kn and all will
472 * return after draining is complete.
474 static void kernfs_drain(struct kernfs_node *kn)
475 __releases(&kernfs_root(kn)->kernfs_rwsem)
476 __acquires(&kernfs_root(kn)->kernfs_rwsem)
478 struct kernfs_root *root = kernfs_root(kn);
480 lockdep_assert_held_write(&root->kernfs_rwsem);
481 WARN_ON_ONCE(kernfs_active(kn));
484 * Skip draining if already fully drained. This avoids draining and its
485 * lockdep annotations for nodes which have never been activated
486 * allowing embedding kernfs_remove() in create error paths without
487 * worrying about draining.
489 if (atomic_read(&kn->active) == KN_DEACTIVATED_BIAS &&
490 !kernfs_should_drain_open_files(kn))
493 up_write(&root->kernfs_rwsem);
495 if (kernfs_lockdep(kn)) {
496 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
497 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
498 lock_contended(&kn->dep_map, _RET_IP_);
501 wait_event(root->deactivate_waitq,
502 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
504 if (kernfs_lockdep(kn)) {
505 lock_acquired(&kn->dep_map, _RET_IP_);
506 rwsem_release(&kn->dep_map, _RET_IP_);
509 if (kernfs_should_drain_open_files(kn))
510 kernfs_drain_open_files(kn);
512 down_write(&root->kernfs_rwsem);
516 * kernfs_get - get a reference count on a kernfs_node
517 * @kn: the target kernfs_node
519 void kernfs_get(struct kernfs_node *kn)
522 WARN_ON(!atomic_read(&kn->count));
523 atomic_inc(&kn->count);
526 EXPORT_SYMBOL_GPL(kernfs_get);
529 * kernfs_put - put a reference count on a kernfs_node
530 * @kn: the target kernfs_node
532 * Put a reference count of @kn and destroy it if it reached zero.
534 void kernfs_put(struct kernfs_node *kn)
536 struct kernfs_node *parent;
537 struct kernfs_root *root;
539 if (!kn || !atomic_dec_and_test(&kn->count))
541 root = kernfs_root(kn);
544 * Moving/renaming is always done while holding reference.
545 * kn->parent won't change beneath us.
549 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
550 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
551 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
553 if (kernfs_type(kn) == KERNFS_LINK)
554 kernfs_put(kn->symlink.target_kn);
556 kfree_const(kn->name);
559 simple_xattrs_free(&kn->iattr->xattrs);
560 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
562 spin_lock(&kernfs_idr_lock);
563 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
564 spin_unlock(&kernfs_idr_lock);
565 kmem_cache_free(kernfs_node_cache, kn);
569 if (atomic_dec_and_test(&kn->count))
572 /* just released the root kn, free @root too */
573 idr_destroy(&root->ino_idr);
577 EXPORT_SYMBOL_GPL(kernfs_put);
580 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
581 * @dentry: the dentry in question
583 * Return: the kernfs_node associated with @dentry. If @dentry is not a
584 * kernfs one, %NULL is returned.
586 * While the returned kernfs_node will stay accessible as long as @dentry
587 * is accessible, the returned node can be in any state and the caller is
588 * fully responsible for determining what's accessible.
590 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
592 if (dentry->d_sb->s_op == &kernfs_sops)
593 return kernfs_dentry_node(dentry);
597 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
598 struct kernfs_node *parent,
599 const char *name, umode_t mode,
600 kuid_t uid, kgid_t gid,
603 struct kernfs_node *kn;
607 name = kstrdup_const(name, GFP_KERNEL);
611 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
615 idr_preload(GFP_KERNEL);
616 spin_lock(&kernfs_idr_lock);
617 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
618 if (ret >= 0 && ret < root->last_id_lowbits)
620 id_highbits = root->id_highbits;
621 root->last_id_lowbits = ret;
622 spin_unlock(&kernfs_idr_lock);
627 kn->id = (u64)id_highbits << 32 | ret;
629 atomic_set(&kn->count, 1);
630 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
631 RB_CLEAR_NODE(&kn->rb);
637 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
638 struct iattr iattr = {
639 .ia_valid = ATTR_UID | ATTR_GID,
644 ret = __kernfs_setattr(kn, &iattr);
650 ret = security_kernfs_init_security(parent, kn);
658 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
660 kmem_cache_free(kernfs_node_cache, kn);
666 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
667 const char *name, umode_t mode,
668 kuid_t uid, kgid_t gid,
671 struct kernfs_node *kn;
673 kn = __kernfs_new_node(kernfs_root(parent), parent,
674 name, mode, uid, gid, flags);
683 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
684 * @root: the kernfs root
685 * @id: the target node id
687 * @id's lower 32bits encode ino and upper gen. If the gen portion is
688 * zero, all generations are matched.
690 * Return: %NULL on failure,
691 * otherwise a kernfs node with reference counter incremented.
693 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
696 struct kernfs_node *kn;
697 ino_t ino = kernfs_id_ino(id);
698 u32 gen = kernfs_id_gen(id);
700 spin_lock(&kernfs_idr_lock);
702 kn = idr_find(&root->ino_idr, (u32)ino);
706 if (sizeof(ino_t) >= sizeof(u64)) {
707 /* we looked up with the low 32bits, compare the whole */
708 if (kernfs_ino(kn) != ino)
711 /* 0 matches all generations */
712 if (unlikely(gen && kernfs_gen(kn) != gen))
717 * We should fail if @kn has never been activated and guarantee success
718 * if the caller knows that @kn is active. Both can be achieved by
719 * __kernfs_active() which tests @kn->active without kernfs_rwsem.
721 if (unlikely(!__kernfs_active(kn) || !atomic_inc_not_zero(&kn->count)))
724 spin_unlock(&kernfs_idr_lock);
727 spin_unlock(&kernfs_idr_lock);
732 * kernfs_add_one - add kernfs_node to parent without warning
733 * @kn: kernfs_node to be added
735 * The caller must already have initialized @kn->parent. This
736 * function increments nlink of the parent's inode if @kn is a
737 * directory and link into the children list of the parent.
740 * %0 on success, -EEXIST if entry with the given name already
743 int kernfs_add_one(struct kernfs_node *kn)
745 struct kernfs_node *parent = kn->parent;
746 struct kernfs_root *root = kernfs_root(parent);
747 struct kernfs_iattrs *ps_iattr;
751 down_write(&root->kernfs_rwsem);
754 has_ns = kernfs_ns_enabled(parent);
755 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
756 has_ns ? "required" : "invalid", parent->name, kn->name))
759 if (kernfs_type(parent) != KERNFS_DIR)
763 if (parent->flags & (KERNFS_REMOVING | KERNFS_EMPTY_DIR))
766 kn->hash = kernfs_name_hash(kn->name, kn->ns);
768 ret = kernfs_link_sibling(kn);
772 /* Update timestamps on the parent */
773 down_write(&root->kernfs_iattr_rwsem);
775 ps_iattr = parent->iattr;
777 ktime_get_real_ts64(&ps_iattr->ia_ctime);
778 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
781 up_write(&root->kernfs_iattr_rwsem);
782 up_write(&root->kernfs_rwsem);
785 * Activate the new node unless CREATE_DEACTIVATED is requested.
786 * If not activated here, the kernfs user is responsible for
787 * activating the node with kernfs_activate(). A node which hasn't
788 * been activated is not visible to userland and its removal won't
789 * trigger deactivation.
791 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
796 up_write(&root->kernfs_rwsem);
801 * kernfs_find_ns - find kernfs_node with the given name
802 * @parent: kernfs_node to search under
803 * @name: name to look for
804 * @ns: the namespace tag to use
806 * Look for kernfs_node with name @name under @parent.
808 * Return: pointer to the found kernfs_node on success, %NULL on failure.
810 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
811 const unsigned char *name,
814 struct rb_node *node = parent->dir.children.rb_node;
815 bool has_ns = kernfs_ns_enabled(parent);
818 lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
820 if (has_ns != (bool)ns) {
821 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
822 has_ns ? "required" : "invalid", parent->name, name);
826 hash = kernfs_name_hash(name, ns);
828 struct kernfs_node *kn;
832 result = kernfs_name_compare(hash, name, ns, kn);
834 node = node->rb_left;
836 node = node->rb_right;
843 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
844 const unsigned char *path,
850 lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
852 spin_lock_irq(&kernfs_pr_cont_lock);
854 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
856 if (len >= sizeof(kernfs_pr_cont_buf)) {
857 spin_unlock_irq(&kernfs_pr_cont_lock);
861 p = kernfs_pr_cont_buf;
863 while ((name = strsep(&p, "/")) && parent) {
866 parent = kernfs_find_ns(parent, name, ns);
869 spin_unlock_irq(&kernfs_pr_cont_lock);
875 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
876 * @parent: kernfs_node to search under
877 * @name: name to look for
878 * @ns: the namespace tag to use
880 * Look for kernfs_node with name @name under @parent and get a reference
881 * if found. This function may sleep.
883 * Return: pointer to the found kernfs_node on success, %NULL on failure.
885 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
886 const char *name, const void *ns)
888 struct kernfs_node *kn;
889 struct kernfs_root *root = kernfs_root(parent);
891 down_read(&root->kernfs_rwsem);
892 kn = kernfs_find_ns(parent, name, ns);
894 up_read(&root->kernfs_rwsem);
898 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
901 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
902 * @parent: kernfs_node to search under
903 * @path: path to look for
904 * @ns: the namespace tag to use
906 * Look for kernfs_node with path @path under @parent and get a reference
907 * if found. This function may sleep.
909 * Return: pointer to the found kernfs_node on success, %NULL on failure.
911 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
912 const char *path, const void *ns)
914 struct kernfs_node *kn;
915 struct kernfs_root *root = kernfs_root(parent);
917 down_read(&root->kernfs_rwsem);
918 kn = kernfs_walk_ns(parent, path, ns);
920 up_read(&root->kernfs_rwsem);
926 * kernfs_create_root - create a new kernfs hierarchy
927 * @scops: optional syscall operations for the hierarchy
928 * @flags: KERNFS_ROOT_* flags
929 * @priv: opaque data associated with the new directory
931 * Return: the root of the new hierarchy on success, ERR_PTR() value on
934 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
935 unsigned int flags, void *priv)
937 struct kernfs_root *root;
938 struct kernfs_node *kn;
940 root = kzalloc(sizeof(*root), GFP_KERNEL);
942 return ERR_PTR(-ENOMEM);
944 idr_init(&root->ino_idr);
945 init_rwsem(&root->kernfs_rwsem);
946 init_rwsem(&root->kernfs_iattr_rwsem);
947 init_rwsem(&root->kernfs_supers_rwsem);
948 INIT_LIST_HEAD(&root->supers);
951 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
952 * High bits generation. The starting value for both ino and
953 * genenration is 1. Initialize upper 32bit allocation
956 if (sizeof(ino_t) >= sizeof(u64))
957 root->id_highbits = 0;
959 root->id_highbits = 1;
961 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
962 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
965 idr_destroy(&root->ino_idr);
967 return ERR_PTR(-ENOMEM);
973 root->syscall_ops = scops;
976 init_waitqueue_head(&root->deactivate_waitq);
978 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
985 * kernfs_destroy_root - destroy a kernfs hierarchy
986 * @root: root of the hierarchy to destroy
988 * Destroy the hierarchy anchored at @root by removing all existing
989 * directories and destroying @root.
991 void kernfs_destroy_root(struct kernfs_root *root)
994 * kernfs_remove holds kernfs_rwsem from the root so the root
995 * shouldn't be freed during the operation.
997 kernfs_get(root->kn);
998 kernfs_remove(root->kn);
999 kernfs_put(root->kn); /* will also free @root */
1003 * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
1004 * @root: root to use to lookup
1006 * Return: @root's kernfs_node
1008 struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
1014 * kernfs_create_dir_ns - create a directory
1015 * @parent: parent in which to create a new directory
1016 * @name: name of the new directory
1017 * @mode: mode of the new directory
1018 * @uid: uid of the new directory
1019 * @gid: gid of the new directory
1020 * @priv: opaque data associated with the new directory
1021 * @ns: optional namespace tag of the directory
1023 * Return: the created node on success, ERR_PTR() value on failure.
1025 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1026 const char *name, umode_t mode,
1027 kuid_t uid, kgid_t gid,
1028 void *priv, const void *ns)
1030 struct kernfs_node *kn;
1034 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1035 uid, gid, KERNFS_DIR);
1037 return ERR_PTR(-ENOMEM);
1039 kn->dir.root = parent->dir.root;
1044 rc = kernfs_add_one(kn);
1053 * kernfs_create_empty_dir - create an always empty directory
1054 * @parent: parent in which to create a new directory
1055 * @name: name of the new directory
1057 * Return: the created node on success, ERR_PTR() value on failure.
1059 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1062 struct kernfs_node *kn;
1066 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1067 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1069 return ERR_PTR(-ENOMEM);
1071 kn->flags |= KERNFS_EMPTY_DIR;
1072 kn->dir.root = parent->dir.root;
1077 rc = kernfs_add_one(kn);
1085 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1087 struct kernfs_node *kn;
1088 struct kernfs_root *root;
1090 if (flags & LOOKUP_RCU)
1093 /* Negative hashed dentry? */
1094 if (d_really_is_negative(dentry)) {
1095 struct kernfs_node *parent;
1097 /* If the kernfs parent node has changed discard and
1098 * proceed to ->lookup.
1100 * There's nothing special needed here when getting the
1101 * dentry parent, even if a concurrent rename is in
1102 * progress. That's because the dentry is negative so
1103 * it can only be the target of the rename and it will
1104 * be doing a d_move() not a replace. Consequently the
1105 * dentry d_parent won't change over the d_move().
1107 * Also kernfs negative dentries transitioning from
1108 * negative to positive during revalidate won't happen
1109 * because they are invalidated on containing directory
1110 * changes and the lookup re-done so that a new positive
1111 * dentry can be properly created.
1113 root = kernfs_root_from_sb(dentry->d_sb);
1114 down_read(&root->kernfs_rwsem);
1115 parent = kernfs_dentry_node(dentry->d_parent);
1117 if (kernfs_dir_changed(parent, dentry)) {
1118 up_read(&root->kernfs_rwsem);
1122 up_read(&root->kernfs_rwsem);
1124 /* The kernfs parent node hasn't changed, leave the
1125 * dentry negative and return success.
1130 kn = kernfs_dentry_node(dentry);
1131 root = kernfs_root(kn);
1132 down_read(&root->kernfs_rwsem);
1134 /* The kernfs node has been deactivated */
1135 if (!kernfs_active(kn))
1138 /* The kernfs node has been moved? */
1139 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1142 /* The kernfs node has been renamed */
1143 if (strcmp(dentry->d_name.name, kn->name) != 0)
1146 /* The kernfs node has been moved to a different namespace */
1147 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1148 kernfs_info(dentry->d_sb)->ns != kn->ns)
1151 up_read(&root->kernfs_rwsem);
1154 up_read(&root->kernfs_rwsem);
1158 const struct dentry_operations kernfs_dops = {
1159 .d_revalidate = kernfs_dop_revalidate,
1162 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1163 struct dentry *dentry,
1166 struct kernfs_node *parent = dir->i_private;
1167 struct kernfs_node *kn;
1168 struct kernfs_root *root;
1169 struct inode *inode = NULL;
1170 const void *ns = NULL;
1172 root = kernfs_root(parent);
1173 down_read(&root->kernfs_rwsem);
1174 if (kernfs_ns_enabled(parent))
1175 ns = kernfs_info(dir->i_sb)->ns;
1177 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1178 /* attach dentry and inode */
1180 /* Inactive nodes are invisible to the VFS so don't
1181 * create a negative.
1183 if (!kernfs_active(kn)) {
1184 up_read(&root->kernfs_rwsem);
1187 inode = kernfs_get_inode(dir->i_sb, kn);
1189 inode = ERR_PTR(-ENOMEM);
1192 * Needed for negative dentry validation.
1193 * The negative dentry can be created in kernfs_iop_lookup()
1194 * or transforms from positive dentry in dentry_unlink_inode()
1195 * called from vfs_rmdir().
1198 kernfs_set_rev(parent, dentry);
1199 up_read(&root->kernfs_rwsem);
1201 /* instantiate and hash (possibly negative) dentry */
1202 return d_splice_alias(inode, dentry);
1205 static int kernfs_iop_mkdir(struct mnt_idmap *idmap,
1206 struct inode *dir, struct dentry *dentry,
1209 struct kernfs_node *parent = dir->i_private;
1210 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1213 if (!scops || !scops->mkdir)
1216 if (!kernfs_get_active(parent))
1219 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1221 kernfs_put_active(parent);
1225 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1227 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1228 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1231 if (!scops || !scops->rmdir)
1234 if (!kernfs_get_active(kn))
1237 ret = scops->rmdir(kn);
1239 kernfs_put_active(kn);
1243 static int kernfs_iop_rename(struct mnt_idmap *idmap,
1244 struct inode *old_dir, struct dentry *old_dentry,
1245 struct inode *new_dir, struct dentry *new_dentry,
1248 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1249 struct kernfs_node *new_parent = new_dir->i_private;
1250 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1256 if (!scops || !scops->rename)
1259 if (!kernfs_get_active(kn))
1262 if (!kernfs_get_active(new_parent)) {
1263 kernfs_put_active(kn);
1267 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1269 kernfs_put_active(new_parent);
1270 kernfs_put_active(kn);
1274 const struct inode_operations kernfs_dir_iops = {
1275 .lookup = kernfs_iop_lookup,
1276 .permission = kernfs_iop_permission,
1277 .setattr = kernfs_iop_setattr,
1278 .getattr = kernfs_iop_getattr,
1279 .listxattr = kernfs_iop_listxattr,
1281 .mkdir = kernfs_iop_mkdir,
1282 .rmdir = kernfs_iop_rmdir,
1283 .rename = kernfs_iop_rename,
1286 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1288 struct kernfs_node *last;
1291 struct rb_node *rbn;
1295 if (kernfs_type(pos) != KERNFS_DIR)
1298 rbn = rb_first(&pos->dir.children);
1302 pos = rb_to_kn(rbn);
1309 * kernfs_next_descendant_post - find the next descendant for post-order walk
1310 * @pos: the current position (%NULL to initiate traversal)
1311 * @root: kernfs_node whose descendants to walk
1313 * Find the next descendant to visit for post-order traversal of @root's
1314 * descendants. @root is included in the iteration and the last node to be
1317 * Return: the next descendant to visit or %NULL when done.
1319 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1320 struct kernfs_node *root)
1322 struct rb_node *rbn;
1324 lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1326 /* if first iteration, visit leftmost descendant which may be root */
1328 return kernfs_leftmost_descendant(root);
1330 /* if we visited @root, we're done */
1334 /* if there's an unvisited sibling, visit its leftmost descendant */
1335 rbn = rb_next(&pos->rb);
1337 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1339 /* no sibling left, visit parent */
1343 static void kernfs_activate_one(struct kernfs_node *kn)
1345 lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1347 kn->flags |= KERNFS_ACTIVATED;
1349 if (kernfs_active(kn) || (kn->flags & (KERNFS_HIDDEN | KERNFS_REMOVING)))
1352 WARN_ON_ONCE(kn->parent && RB_EMPTY_NODE(&kn->rb));
1353 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1355 atomic_sub(KN_DEACTIVATED_BIAS, &kn->active);
1359 * kernfs_activate - activate a node which started deactivated
1360 * @kn: kernfs_node whose subtree is to be activated
1362 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1363 * needs to be explicitly activated. A node which hasn't been activated
1364 * isn't visible to userland and deactivation is skipped during its
1365 * removal. This is useful to construct atomic init sequences where
1366 * creation of multiple nodes should either succeed or fail atomically.
1368 * The caller is responsible for ensuring that this function is not called
1369 * after kernfs_remove*() is invoked on @kn.
1371 void kernfs_activate(struct kernfs_node *kn)
1373 struct kernfs_node *pos;
1374 struct kernfs_root *root = kernfs_root(kn);
1376 down_write(&root->kernfs_rwsem);
1379 while ((pos = kernfs_next_descendant_post(pos, kn)))
1380 kernfs_activate_one(pos);
1382 up_write(&root->kernfs_rwsem);
1386 * kernfs_show - show or hide a node
1387 * @kn: kernfs_node to show or hide
1388 * @show: whether to show or hide
1390 * If @show is %false, @kn is marked hidden and deactivated. A hidden node is
1391 * ignored in future activaitons. If %true, the mark is removed and activation
1392 * state is restored. This function won't implicitly activate a new node in a
1393 * %KERNFS_ROOT_CREATE_DEACTIVATED root which hasn't been activated yet.
1395 * To avoid recursion complexities, directories aren't supported for now.
1397 void kernfs_show(struct kernfs_node *kn, bool show)
1399 struct kernfs_root *root = kernfs_root(kn);
1401 if (WARN_ON_ONCE(kernfs_type(kn) == KERNFS_DIR))
1404 down_write(&root->kernfs_rwsem);
1407 kn->flags &= ~KERNFS_HIDDEN;
1408 if (kn->flags & KERNFS_ACTIVATED)
1409 kernfs_activate_one(kn);
1411 kn->flags |= KERNFS_HIDDEN;
1412 if (kernfs_active(kn))
1413 atomic_add(KN_DEACTIVATED_BIAS, &kn->active);
1417 up_write(&root->kernfs_rwsem);
1420 static void __kernfs_remove(struct kernfs_node *kn)
1422 struct kernfs_node *pos;
1424 /* Short-circuit if non-root @kn has already finished removal. */
1428 lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1431 * This is for kernfs_remove_self() which plays with active ref
1434 if (kn->parent && RB_EMPTY_NODE(&kn->rb))
1437 pr_debug("kernfs %s: removing\n", kn->name);
1439 /* prevent new usage by marking all nodes removing and deactivating */
1441 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1442 pos->flags |= KERNFS_REMOVING;
1443 if (kernfs_active(pos))
1444 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1447 /* deactivate and unlink the subtree node-by-node */
1449 pos = kernfs_leftmost_descendant(kn);
1452 * kernfs_drain() may drop kernfs_rwsem temporarily and @pos's
1453 * base ref could have been put by someone else by the time
1454 * the function returns. Make sure it doesn't go away
1462 * kernfs_unlink_sibling() succeeds once per node. Use it
1463 * to decide who's responsible for cleanups.
1465 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1466 struct kernfs_iattrs *ps_iattr =
1467 pos->parent ? pos->parent->iattr : NULL;
1469 /* update timestamps on the parent */
1470 down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1473 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1474 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1477 up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
1482 } while (pos != kn);
1486 * kernfs_remove - remove a kernfs_node recursively
1487 * @kn: the kernfs_node to remove
1489 * Remove @kn along with all its subdirectories and files.
1491 void kernfs_remove(struct kernfs_node *kn)
1493 struct kernfs_root *root;
1498 root = kernfs_root(kn);
1500 down_write(&root->kernfs_rwsem);
1501 __kernfs_remove(kn);
1502 up_write(&root->kernfs_rwsem);
1506 * kernfs_break_active_protection - break out of active protection
1507 * @kn: the self kernfs_node
1509 * The caller must be running off of a kernfs operation which is invoked
1510 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1511 * this function must also be matched with an invocation of
1512 * kernfs_unbreak_active_protection().
1514 * This function releases the active reference of @kn the caller is
1515 * holding. Once this function is called, @kn may be removed at any point
1516 * and the caller is solely responsible for ensuring that the objects it
1517 * dereferences are accessible.
1519 void kernfs_break_active_protection(struct kernfs_node *kn)
1522 * Take out ourself out of the active ref dependency chain. If
1523 * we're called without an active ref, lockdep will complain.
1525 kernfs_put_active(kn);
1529 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1530 * @kn: the self kernfs_node
1532 * If kernfs_break_active_protection() was called, this function must be
1533 * invoked before finishing the kernfs operation. Note that while this
1534 * function restores the active reference, it doesn't and can't actually
1535 * restore the active protection - @kn may already or be in the process of
1536 * being removed. Once kernfs_break_active_protection() is invoked, that
1537 * protection is irreversibly gone for the kernfs operation instance.
1539 * While this function may be called at any point after
1540 * kernfs_break_active_protection() is invoked, its most useful location
1541 * would be right before the enclosing kernfs operation returns.
1543 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1546 * @kn->active could be in any state; however, the increment we do
1547 * here will be undone as soon as the enclosing kernfs operation
1548 * finishes and this temporary bump can't break anything. If @kn
1549 * is alive, nothing changes. If @kn is being deactivated, the
1550 * soon-to-follow put will either finish deactivation or restore
1551 * deactivated state. If @kn is already removed, the temporary
1552 * bump is guaranteed to be gone before @kn is released.
1554 atomic_inc(&kn->active);
1555 if (kernfs_lockdep(kn))
1556 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1560 * kernfs_remove_self - remove a kernfs_node from its own method
1561 * @kn: the self kernfs_node to remove
1563 * The caller must be running off of a kernfs operation which is invoked
1564 * with an active reference - e.g. one of kernfs_ops. This can be used to
1565 * implement a file operation which deletes itself.
1567 * For example, the "delete" file for a sysfs device directory can be
1568 * implemented by invoking kernfs_remove_self() on the "delete" file
1569 * itself. This function breaks the circular dependency of trying to
1570 * deactivate self while holding an active ref itself. It isn't necessary
1571 * to modify the usual removal path to use kernfs_remove_self(). The
1572 * "delete" implementation can simply invoke kernfs_remove_self() on self
1573 * before proceeding with the usual removal path. kernfs will ignore later
1574 * kernfs_remove() on self.
1576 * kernfs_remove_self() can be called multiple times concurrently on the
1577 * same kernfs_node. Only the first one actually performs removal and
1578 * returns %true. All others will wait until the kernfs operation which
1579 * won self-removal finishes and return %false. Note that the losers wait
1580 * for the completion of not only the winning kernfs_remove_self() but also
1581 * the whole kernfs_ops which won the arbitration. This can be used to
1582 * guarantee, for example, all concurrent writes to a "delete" file to
1583 * finish only after the whole operation is complete.
1585 * Return: %true if @kn is removed by this call, otherwise %false.
1587 bool kernfs_remove_self(struct kernfs_node *kn)
1590 struct kernfs_root *root = kernfs_root(kn);
1592 down_write(&root->kernfs_rwsem);
1593 kernfs_break_active_protection(kn);
1596 * SUICIDAL is used to arbitrate among competing invocations. Only
1597 * the first one will actually perform removal. When the removal
1598 * is complete, SUICIDED is set and the active ref is restored
1599 * while kernfs_rwsem for held exclusive. The ones which lost
1600 * arbitration waits for SUICIDED && drained which can happen only
1601 * after the enclosing kernfs operation which executed the winning
1602 * instance of kernfs_remove_self() finished.
1604 if (!(kn->flags & KERNFS_SUICIDAL)) {
1605 kn->flags |= KERNFS_SUICIDAL;
1606 __kernfs_remove(kn);
1607 kn->flags |= KERNFS_SUICIDED;
1610 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1614 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1616 if ((kn->flags & KERNFS_SUICIDED) &&
1617 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1620 up_write(&root->kernfs_rwsem);
1622 down_write(&root->kernfs_rwsem);
1624 finish_wait(waitq, &wait);
1625 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1630 * This must be done while kernfs_rwsem held exclusive; otherwise,
1631 * waiting for SUICIDED && deactivated could finish prematurely.
1633 kernfs_unbreak_active_protection(kn);
1635 up_write(&root->kernfs_rwsem);
1640 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1641 * @parent: parent of the target
1642 * @name: name of the kernfs_node to remove
1643 * @ns: namespace tag of the kernfs_node to remove
1645 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1647 * Return: %0 on success, -ENOENT if such entry doesn't exist.
1649 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1652 struct kernfs_node *kn;
1653 struct kernfs_root *root;
1656 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1661 root = kernfs_root(parent);
1662 down_write(&root->kernfs_rwsem);
1664 kn = kernfs_find_ns(parent, name, ns);
1667 __kernfs_remove(kn);
1671 up_write(&root->kernfs_rwsem);
1680 * kernfs_rename_ns - move and rename a kernfs_node
1682 * @new_parent: new parent to put @sd under
1683 * @new_name: new name
1684 * @new_ns: new namespace tag
1686 * Return: %0 on success, -errno on failure.
1688 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1689 const char *new_name, const void *new_ns)
1691 struct kernfs_node *old_parent;
1692 struct kernfs_root *root;
1693 const char *old_name = NULL;
1696 /* can't move or rename root */
1700 root = kernfs_root(kn);
1701 down_write(&root->kernfs_rwsem);
1704 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1705 (new_parent->flags & KERNFS_EMPTY_DIR))
1709 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1710 (strcmp(kn->name, new_name) == 0))
1711 goto out; /* nothing to rename */
1714 if (kernfs_find_ns(new_parent, new_name, new_ns))
1717 /* rename kernfs_node */
1718 if (strcmp(kn->name, new_name) != 0) {
1720 new_name = kstrdup_const(new_name, GFP_KERNEL);
1728 * Move to the appropriate place in the appropriate directories rbtree.
1730 kernfs_unlink_sibling(kn);
1731 kernfs_get(new_parent);
1733 /* rename_lock protects ->parent and ->name accessors */
1734 write_lock_irq(&kernfs_rename_lock);
1736 old_parent = kn->parent;
1737 kn->parent = new_parent;
1741 old_name = kn->name;
1742 kn->name = new_name;
1745 write_unlock_irq(&kernfs_rename_lock);
1747 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1748 kernfs_link_sibling(kn);
1750 kernfs_put(old_parent);
1751 kfree_const(old_name);
1755 up_write(&root->kernfs_rwsem);
1759 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1761 kernfs_put(filp->private_data);
1765 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1766 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1769 int valid = kernfs_active(pos) &&
1770 pos->parent == parent && hash == pos->hash;
1775 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1776 struct rb_node *node = parent->dir.children.rb_node;
1778 pos = rb_to_kn(node);
1780 if (hash < pos->hash)
1781 node = node->rb_left;
1782 else if (hash > pos->hash)
1783 node = node->rb_right;
1788 /* Skip over entries which are dying/dead or in the wrong namespace */
1789 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1790 struct rb_node *node = rb_next(&pos->rb);
1794 pos = rb_to_kn(node);
1799 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1800 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1802 pos = kernfs_dir_pos(ns, parent, ino, pos);
1805 struct rb_node *node = rb_next(&pos->rb);
1809 pos = rb_to_kn(node);
1810 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1815 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1817 struct dentry *dentry = file->f_path.dentry;
1818 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1819 struct kernfs_node *pos = file->private_data;
1820 struct kernfs_root *root;
1821 const void *ns = NULL;
1823 if (!dir_emit_dots(file, ctx))
1826 root = kernfs_root(parent);
1827 down_read(&root->kernfs_rwsem);
1829 if (kernfs_ns_enabled(parent))
1830 ns = kernfs_info(dentry->d_sb)->ns;
1832 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1834 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1835 const char *name = pos->name;
1836 unsigned int type = fs_umode_to_dtype(pos->mode);
1837 int len = strlen(name);
1838 ino_t ino = kernfs_ino(pos);
1840 ctx->pos = pos->hash;
1841 file->private_data = pos;
1844 up_read(&root->kernfs_rwsem);
1845 if (!dir_emit(ctx, name, len, ino, type))
1847 down_read(&root->kernfs_rwsem);
1849 up_read(&root->kernfs_rwsem);
1850 file->private_data = NULL;
1855 const struct file_operations kernfs_dir_fops = {
1856 .read = generic_read_dir,
1857 .iterate_shared = kernfs_fop_readdir,
1858 .release = kernfs_dir_fop_release,
1859 .llseek = generic_file_llseek,