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_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
21 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */
22 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
24 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
26 static bool kernfs_active(struct kernfs_node *kn)
28 lockdep_assert_held(&kernfs_root(kn)->kernfs_rwsem);
29 return atomic_read(&kn->active) >= 0;
32 static bool kernfs_lockdep(struct kernfs_node *kn)
34 #ifdef CONFIG_DEBUG_LOCK_ALLOC
35 return kn->flags & KERNFS_LOCKDEP;
41 static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
44 return strlcpy(buf, "(null)", buflen);
46 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
49 /* kernfs_node_depth - compute depth from @from to @to */
50 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
54 while (to->parent && to != from) {
61 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
62 struct kernfs_node *b)
65 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
70 da = kernfs_depth(ra->kn, a);
71 db = kernfs_depth(rb->kn, b);
82 /* worst case b and a will be the same at root */
92 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
93 * where kn_from is treated as root of the path.
94 * @kn_from: kernfs node which should be treated as root for the path
95 * @kn_to: kernfs node to which path is needed
96 * @buf: buffer to copy the path into
97 * @buflen: size of @buf
99 * We need to handle couple of scenarios here:
100 * [1] when @kn_from is an ancestor of @kn_to at some level
102 * kn_to: /n1/n2/n3/n4/n5
105 * [2] when @kn_from is on a different hierarchy and we need to find common
106 * ancestor between @kn_from and @kn_to.
107 * kn_from: /n1/n2/n3/n4
111 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
112 * kn_to: /n1/n2/n3 [depth=3]
115 * [3] when @kn_to is NULL result will be "(null)"
117 * Returns the length of the full path. If the full length is equal to or
118 * greater than @buflen, @buf contains the truncated path with the trailing
119 * '\0'. On error, -errno is returned.
121 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
122 struct kernfs_node *kn_from,
123 char *buf, size_t buflen)
125 struct kernfs_node *kn, *common;
126 const char parent_str[] = "/..";
127 size_t depth_from, depth_to, len = 0;
131 return strlcpy(buf, "(null)", buflen);
134 kn_from = kernfs_root(kn_to)->kn;
136 if (kn_from == kn_to)
137 return strlcpy(buf, "/", buflen);
142 common = kernfs_common_ancestor(kn_from, kn_to);
143 if (WARN_ON(!common))
146 depth_to = kernfs_depth(common, kn_to);
147 depth_from = kernfs_depth(common, kn_from);
151 for (i = 0; i < depth_from; i++)
152 len += strlcpy(buf + len, parent_str,
153 len < buflen ? buflen - len : 0);
155 /* Calculate how many bytes we need for the rest */
156 for (i = depth_to - 1; i >= 0; i--) {
157 for (kn = kn_to, j = 0; j < i; j++)
159 len += strlcpy(buf + len, "/",
160 len < buflen ? buflen - len : 0);
161 len += strlcpy(buf + len, kn->name,
162 len < buflen ? buflen - len : 0);
169 * kernfs_name - obtain the name of a given node
170 * @kn: kernfs_node of interest
171 * @buf: buffer to copy @kn's name into
172 * @buflen: size of @buf
174 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
175 * similar to strlcpy(). It returns the length of @kn's name and if @buf
176 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
178 * Fills buffer with "(null)" if @kn is NULL.
180 * This function can be called from any context.
182 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
187 spin_lock_irqsave(&kernfs_rename_lock, flags);
188 ret = kernfs_name_locked(kn, buf, buflen);
189 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
194 * kernfs_path_from_node - build path of node @to relative to @from.
195 * @from: parent kernfs_node relative to which we need to build the path
196 * @to: kernfs_node of interest
197 * @buf: buffer to copy @to's path into
198 * @buflen: size of @buf
200 * Builds @to's path relative to @from in @buf. @from and @to must
201 * be on the same kernfs-root. If @from is not parent of @to, then a relative
202 * path (which includes '..'s) as needed to reach from @from to @to is
205 * Returns the length of the full path. If the full length is equal to or
206 * greater than @buflen, @buf contains the truncated path with the trailing
207 * '\0'. On error, -errno is returned.
209 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
210 char *buf, size_t buflen)
215 spin_lock_irqsave(&kernfs_rename_lock, flags);
216 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
217 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
220 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
223 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
224 * @kn: kernfs_node of interest
226 * This function can be called from any context.
228 void pr_cont_kernfs_name(struct kernfs_node *kn)
232 spin_lock_irqsave(&kernfs_rename_lock, flags);
234 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
235 pr_cont("%s", kernfs_pr_cont_buf);
237 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
241 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
242 * @kn: kernfs_node of interest
244 * This function can be called from any context.
246 void pr_cont_kernfs_path(struct kernfs_node *kn)
251 spin_lock_irqsave(&kernfs_rename_lock, flags);
253 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
254 sizeof(kernfs_pr_cont_buf));
260 if (sz >= sizeof(kernfs_pr_cont_buf)) {
261 pr_cont("(name too long)");
265 pr_cont("%s", kernfs_pr_cont_buf);
268 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
272 * kernfs_get_parent - determine the parent node and pin it
273 * @kn: kernfs_node of interest
275 * Determines @kn's parent, pins and returns it. This function can be
276 * called from any context.
278 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
280 struct kernfs_node *parent;
283 spin_lock_irqsave(&kernfs_rename_lock, flags);
286 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
293 * @name: Null terminated string to hash
294 * @ns: Namespace tag to hash
296 * Returns 31 bit hash of ns + name (so it fits in an off_t )
298 static unsigned int kernfs_name_hash(const char *name, const void *ns)
300 unsigned long hash = init_name_hash(ns);
301 unsigned int len = strlen(name);
303 hash = partial_name_hash(*name++, hash);
304 hash = end_name_hash(hash);
306 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
314 static int kernfs_name_compare(unsigned int hash, const char *name,
315 const void *ns, const struct kernfs_node *kn)
325 return strcmp(name, kn->name);
328 static int kernfs_sd_compare(const struct kernfs_node *left,
329 const struct kernfs_node *right)
331 return kernfs_name_compare(left->hash, left->name, left->ns, right);
335 * kernfs_link_sibling - link kernfs_node into sibling rbtree
336 * @kn: kernfs_node of interest
338 * Link @kn into its sibling rbtree which starts from
339 * @kn->parent->dir.children.
342 * kernfs_rwsem held exclusive
345 * 0 on susccess -EEXIST on failure.
347 static int kernfs_link_sibling(struct kernfs_node *kn)
349 struct rb_node **node = &kn->parent->dir.children.rb_node;
350 struct rb_node *parent = NULL;
353 struct kernfs_node *pos;
356 pos = rb_to_kn(*node);
358 result = kernfs_sd_compare(kn, pos);
360 node = &pos->rb.rb_left;
362 node = &pos->rb.rb_right;
367 /* add new node and rebalance the tree */
368 rb_link_node(&kn->rb, parent, node);
369 rb_insert_color(&kn->rb, &kn->parent->dir.children);
371 /* successfully added, account subdir number */
372 if (kernfs_type(kn) == KERNFS_DIR)
373 kn->parent->dir.subdirs++;
374 kernfs_inc_rev(kn->parent);
380 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
381 * @kn: kernfs_node of interest
383 * Try to unlink @kn from its sibling rbtree which starts from
384 * kn->parent->dir.children. Returns %true if @kn was actually
385 * removed, %false if @kn wasn't on the rbtree.
388 * kernfs_rwsem held exclusive
390 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
392 if (RB_EMPTY_NODE(&kn->rb))
395 if (kernfs_type(kn) == KERNFS_DIR)
396 kn->parent->dir.subdirs--;
397 kernfs_inc_rev(kn->parent);
399 rb_erase(&kn->rb, &kn->parent->dir.children);
400 RB_CLEAR_NODE(&kn->rb);
405 * kernfs_get_active - get an active reference to kernfs_node
406 * @kn: kernfs_node to get an active reference to
408 * Get an active reference of @kn. This function is noop if @kn
412 * Pointer to @kn on success, NULL on failure.
414 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
419 if (!atomic_inc_unless_negative(&kn->active))
422 if (kernfs_lockdep(kn))
423 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
428 * kernfs_put_active - put an active reference to kernfs_node
429 * @kn: kernfs_node to put an active reference to
431 * Put an active reference to @kn. This function is noop if @kn
434 void kernfs_put_active(struct kernfs_node *kn)
441 if (kernfs_lockdep(kn))
442 rwsem_release(&kn->dep_map, _RET_IP_);
443 v = atomic_dec_return(&kn->active);
444 if (likely(v != KN_DEACTIVATED_BIAS))
447 wake_up_all(&kernfs_root(kn)->deactivate_waitq);
451 * kernfs_drain - drain kernfs_node
452 * @kn: kernfs_node to drain
454 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
455 * removers may invoke this function concurrently on @kn and all will
456 * return after draining is complete.
458 static void kernfs_drain(struct kernfs_node *kn)
459 __releases(&kernfs_root(kn)->kernfs_rwsem)
460 __acquires(&kernfs_root(kn)->kernfs_rwsem)
462 struct kernfs_root *root = kernfs_root(kn);
464 lockdep_assert_held_write(&root->kernfs_rwsem);
465 WARN_ON_ONCE(kernfs_active(kn));
467 up_write(&root->kernfs_rwsem);
469 if (kernfs_lockdep(kn)) {
470 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
471 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
472 lock_contended(&kn->dep_map, _RET_IP_);
475 /* but everyone should wait for draining */
476 wait_event(root->deactivate_waitq,
477 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
479 if (kernfs_lockdep(kn)) {
480 lock_acquired(&kn->dep_map, _RET_IP_);
481 rwsem_release(&kn->dep_map, _RET_IP_);
484 kernfs_drain_open_files(kn);
486 down_write(&root->kernfs_rwsem);
490 * kernfs_get - get a reference count on a kernfs_node
491 * @kn: the target kernfs_node
493 void kernfs_get(struct kernfs_node *kn)
496 WARN_ON(!atomic_read(&kn->count));
497 atomic_inc(&kn->count);
500 EXPORT_SYMBOL_GPL(kernfs_get);
503 * kernfs_put - put a reference count on a kernfs_node
504 * @kn: the target kernfs_node
506 * Put a reference count of @kn and destroy it if it reached zero.
508 void kernfs_put(struct kernfs_node *kn)
510 struct kernfs_node *parent;
511 struct kernfs_root *root;
513 if (!kn || !atomic_dec_and_test(&kn->count))
515 root = kernfs_root(kn);
518 * Moving/renaming is always done while holding reference.
519 * kn->parent won't change beneath us.
523 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
524 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
525 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
527 if (kernfs_type(kn) == KERNFS_LINK)
528 kernfs_put(kn->symlink.target_kn);
530 kfree_const(kn->name);
533 simple_xattrs_free(&kn->iattr->xattrs);
534 kmem_cache_free(kernfs_iattrs_cache, kn->iattr);
536 spin_lock(&kernfs_idr_lock);
537 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
538 spin_unlock(&kernfs_idr_lock);
539 kmem_cache_free(kernfs_node_cache, kn);
543 if (atomic_dec_and_test(&kn->count))
546 /* just released the root kn, free @root too */
547 idr_destroy(&root->ino_idr);
551 EXPORT_SYMBOL_GPL(kernfs_put);
554 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
555 * @dentry: the dentry in question
557 * Return the kernfs_node associated with @dentry. If @dentry is not a
558 * kernfs one, %NULL is returned.
560 * While the returned kernfs_node will stay accessible as long as @dentry
561 * is accessible, the returned node can be in any state and the caller is
562 * fully responsible for determining what's accessible.
564 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
566 if (dentry->d_sb->s_op == &kernfs_sops)
567 return kernfs_dentry_node(dentry);
571 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
572 struct kernfs_node *parent,
573 const char *name, umode_t mode,
574 kuid_t uid, kgid_t gid,
577 struct kernfs_node *kn;
581 name = kstrdup_const(name, GFP_KERNEL);
585 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
589 idr_preload(GFP_KERNEL);
590 spin_lock(&kernfs_idr_lock);
591 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
592 if (ret >= 0 && ret < root->last_id_lowbits)
594 id_highbits = root->id_highbits;
595 root->last_id_lowbits = ret;
596 spin_unlock(&kernfs_idr_lock);
601 kn->id = (u64)id_highbits << 32 | ret;
603 atomic_set(&kn->count, 1);
604 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
605 RB_CLEAR_NODE(&kn->rb);
611 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
612 struct iattr iattr = {
613 .ia_valid = ATTR_UID | ATTR_GID,
618 ret = __kernfs_setattr(kn, &iattr);
624 ret = security_kernfs_init_security(parent, kn);
632 idr_remove(&root->ino_idr, (u32)kernfs_ino(kn));
634 kmem_cache_free(kernfs_node_cache, kn);
640 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
641 const char *name, umode_t mode,
642 kuid_t uid, kgid_t gid,
645 struct kernfs_node *kn;
647 kn = __kernfs_new_node(kernfs_root(parent), parent,
648 name, mode, uid, gid, flags);
657 * kernfs_find_and_get_node_by_id - get kernfs_node from node id
658 * @root: the kernfs root
659 * @id: the target node id
661 * @id's lower 32bits encode ino and upper gen. If the gen portion is
662 * zero, all generations are matched.
665 * NULL on failure. Return a kernfs node with reference counter incremented
667 struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root,
670 struct kernfs_node *kn;
671 ino_t ino = kernfs_id_ino(id);
672 u32 gen = kernfs_id_gen(id);
674 spin_lock(&kernfs_idr_lock);
676 kn = idr_find(&root->ino_idr, (u32)ino);
680 if (sizeof(ino_t) >= sizeof(u64)) {
681 /* we looked up with the low 32bits, compare the whole */
682 if (kernfs_ino(kn) != ino)
685 /* 0 matches all generations */
686 if (unlikely(gen && kernfs_gen(kn) != gen))
691 * ACTIVATED is protected with kernfs_mutex but it was clear when
692 * @kn was added to idr and we just wanna see it set. No need to
695 if (unlikely(!(kn->flags & KERNFS_ACTIVATED) ||
696 !atomic_inc_not_zero(&kn->count)))
699 spin_unlock(&kernfs_idr_lock);
702 spin_unlock(&kernfs_idr_lock);
707 * kernfs_add_one - add kernfs_node to parent without warning
708 * @kn: kernfs_node to be added
710 * The caller must already have initialized @kn->parent. This
711 * function increments nlink of the parent's inode if @kn is a
712 * directory and link into the children list of the parent.
715 * 0 on success, -EEXIST if entry with the given name already
718 int kernfs_add_one(struct kernfs_node *kn)
720 struct kernfs_node *parent = kn->parent;
721 struct kernfs_root *root = kernfs_root(parent);
722 struct kernfs_iattrs *ps_iattr;
726 down_write(&root->kernfs_rwsem);
729 has_ns = kernfs_ns_enabled(parent);
730 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
731 has_ns ? "required" : "invalid", parent->name, kn->name))
734 if (kernfs_type(parent) != KERNFS_DIR)
738 if (parent->flags & KERNFS_EMPTY_DIR)
741 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
744 kn->hash = kernfs_name_hash(kn->name, kn->ns);
746 ret = kernfs_link_sibling(kn);
750 /* Update timestamps on the parent */
751 ps_iattr = parent->iattr;
753 ktime_get_real_ts64(&ps_iattr->ia_ctime);
754 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
757 up_write(&root->kernfs_rwsem);
760 * Activate the new node unless CREATE_DEACTIVATED is requested.
761 * If not activated here, the kernfs user is responsible for
762 * activating the node with kernfs_activate(). A node which hasn't
763 * been activated is not visible to userland and its removal won't
764 * trigger deactivation.
766 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
771 up_write(&root->kernfs_rwsem);
776 * kernfs_find_ns - find kernfs_node with the given name
777 * @parent: kernfs_node to search under
778 * @name: name to look for
779 * @ns: the namespace tag to use
781 * Look for kernfs_node with name @name under @parent. Returns pointer to
782 * the found kernfs_node on success, %NULL on failure.
784 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
785 const unsigned char *name,
788 struct rb_node *node = parent->dir.children.rb_node;
789 bool has_ns = kernfs_ns_enabled(parent);
792 lockdep_assert_held(&kernfs_root(parent)->kernfs_rwsem);
794 if (has_ns != (bool)ns) {
795 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
796 has_ns ? "required" : "invalid", parent->name, name);
800 hash = kernfs_name_hash(name, ns);
802 struct kernfs_node *kn;
806 result = kernfs_name_compare(hash, name, ns, kn);
808 node = node->rb_left;
810 node = node->rb_right;
817 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
818 const unsigned char *path,
824 lockdep_assert_held_read(&kernfs_root(parent)->kernfs_rwsem);
826 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
827 spin_lock_irq(&kernfs_rename_lock);
829 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
831 if (len >= sizeof(kernfs_pr_cont_buf)) {
832 spin_unlock_irq(&kernfs_rename_lock);
836 p = kernfs_pr_cont_buf;
838 while ((name = strsep(&p, "/")) && parent) {
841 parent = kernfs_find_ns(parent, name, ns);
844 spin_unlock_irq(&kernfs_rename_lock);
850 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
851 * @parent: kernfs_node to search under
852 * @name: name to look for
853 * @ns: the namespace tag to use
855 * Look for kernfs_node with name @name under @parent and get a reference
856 * if found. This function may sleep and returns pointer to the found
857 * kernfs_node on success, %NULL on failure.
859 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
860 const char *name, const void *ns)
862 struct kernfs_node *kn;
863 struct kernfs_root *root = kernfs_root(parent);
865 down_read(&root->kernfs_rwsem);
866 kn = kernfs_find_ns(parent, name, ns);
868 up_read(&root->kernfs_rwsem);
872 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
875 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
876 * @parent: kernfs_node to search under
877 * @path: path to look for
878 * @ns: the namespace tag to use
880 * Look for kernfs_node with path @path under @parent and get a reference
881 * if found. This function may sleep and returns pointer to the found
882 * kernfs_node on success, %NULL on failure.
884 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
885 const char *path, const void *ns)
887 struct kernfs_node *kn;
888 struct kernfs_root *root = kernfs_root(parent);
890 down_read(&root->kernfs_rwsem);
891 kn = kernfs_walk_ns(parent, path, ns);
893 up_read(&root->kernfs_rwsem);
899 * kernfs_create_root - create a new kernfs hierarchy
900 * @scops: optional syscall operations for the hierarchy
901 * @flags: KERNFS_ROOT_* flags
902 * @priv: opaque data associated with the new directory
904 * Returns the root of the new hierarchy on success, ERR_PTR() value on
907 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
908 unsigned int flags, void *priv)
910 struct kernfs_root *root;
911 struct kernfs_node *kn;
913 root = kzalloc(sizeof(*root), GFP_KERNEL);
915 return ERR_PTR(-ENOMEM);
917 idr_init(&root->ino_idr);
918 init_rwsem(&root->kernfs_rwsem);
919 INIT_LIST_HEAD(&root->supers);
922 * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino.
923 * High bits generation. The starting value for both ino and
924 * genenration is 1. Initialize upper 32bit allocation
927 if (sizeof(ino_t) >= sizeof(u64))
928 root->id_highbits = 0;
930 root->id_highbits = 1;
932 kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO,
933 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
936 idr_destroy(&root->ino_idr);
938 return ERR_PTR(-ENOMEM);
944 root->syscall_ops = scops;
947 init_waitqueue_head(&root->deactivate_waitq);
949 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
956 * kernfs_destroy_root - destroy a kernfs hierarchy
957 * @root: root of the hierarchy to destroy
959 * Destroy the hierarchy anchored at @root by removing all existing
960 * directories and destroying @root.
962 void kernfs_destroy_root(struct kernfs_root *root)
965 * kernfs_remove holds kernfs_rwsem from the root so the root
966 * shouldn't be freed during the operation.
968 kernfs_get(root->kn);
969 kernfs_remove(root->kn);
970 kernfs_put(root->kn); /* will also free @root */
974 * kernfs_root_to_node - return the kernfs_node associated with a kernfs_root
975 * @root: root to use to lookup
977 struct kernfs_node *kernfs_root_to_node(struct kernfs_root *root)
983 * kernfs_create_dir_ns - create a directory
984 * @parent: parent in which to create a new directory
985 * @name: name of the new directory
986 * @mode: mode of the new directory
987 * @uid: uid of the new directory
988 * @gid: gid of the new directory
989 * @priv: opaque data associated with the new directory
990 * @ns: optional namespace tag of the directory
992 * Returns the created node on success, ERR_PTR() value on failure.
994 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
995 const char *name, umode_t mode,
996 kuid_t uid, kgid_t gid,
997 void *priv, const void *ns)
999 struct kernfs_node *kn;
1003 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1004 uid, gid, KERNFS_DIR);
1006 return ERR_PTR(-ENOMEM);
1008 kn->dir.root = parent->dir.root;
1013 rc = kernfs_add_one(kn);
1022 * kernfs_create_empty_dir - create an always empty directory
1023 * @parent: parent in which to create a new directory
1024 * @name: name of the new directory
1026 * Returns the created node on success, ERR_PTR() value on failure.
1028 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1031 struct kernfs_node *kn;
1035 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1036 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1038 return ERR_PTR(-ENOMEM);
1040 kn->flags |= KERNFS_EMPTY_DIR;
1041 kn->dir.root = parent->dir.root;
1046 rc = kernfs_add_one(kn);
1054 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
1056 struct kernfs_node *kn;
1057 struct kernfs_root *root;
1059 if (flags & LOOKUP_RCU)
1062 /* Negative hashed dentry? */
1063 if (d_really_is_negative(dentry)) {
1064 struct kernfs_node *parent;
1066 /* If the kernfs parent node has changed discard and
1067 * proceed to ->lookup.
1069 spin_lock(&dentry->d_lock);
1070 parent = kernfs_dentry_node(dentry->d_parent);
1072 spin_unlock(&dentry->d_lock);
1073 root = kernfs_root(parent);
1074 down_read(&root->kernfs_rwsem);
1075 if (kernfs_dir_changed(parent, dentry)) {
1076 up_read(&root->kernfs_rwsem);
1079 up_read(&root->kernfs_rwsem);
1081 spin_unlock(&dentry->d_lock);
1083 /* The kernfs parent node hasn't changed, leave the
1084 * dentry negative and return success.
1089 kn = kernfs_dentry_node(dentry);
1090 root = kernfs_root(kn);
1091 down_read(&root->kernfs_rwsem);
1093 /* The kernfs node has been deactivated */
1094 if (!kernfs_active(kn))
1097 /* The kernfs node has been moved? */
1098 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
1101 /* The kernfs node has been renamed */
1102 if (strcmp(dentry->d_name.name, kn->name) != 0)
1105 /* The kernfs node has been moved to a different namespace */
1106 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
1107 kernfs_info(dentry->d_sb)->ns != kn->ns)
1110 up_read(&root->kernfs_rwsem);
1113 up_read(&root->kernfs_rwsem);
1117 const struct dentry_operations kernfs_dops = {
1118 .d_revalidate = kernfs_dop_revalidate,
1121 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1122 struct dentry *dentry,
1125 struct kernfs_node *parent = dir->i_private;
1126 struct kernfs_node *kn;
1127 struct kernfs_root *root;
1128 struct inode *inode = NULL;
1129 const void *ns = NULL;
1131 root = kernfs_root(parent);
1132 down_read(&root->kernfs_rwsem);
1133 if (kernfs_ns_enabled(parent))
1134 ns = kernfs_info(dir->i_sb)->ns;
1136 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1137 /* attach dentry and inode */
1139 /* Inactive nodes are invisible to the VFS so don't
1140 * create a negative.
1142 if (!kernfs_active(kn)) {
1143 up_read(&root->kernfs_rwsem);
1146 inode = kernfs_get_inode(dir->i_sb, kn);
1148 inode = ERR_PTR(-ENOMEM);
1151 * Needed for negative dentry validation.
1152 * The negative dentry can be created in kernfs_iop_lookup()
1153 * or transforms from positive dentry in dentry_unlink_inode()
1154 * called from vfs_rmdir().
1157 kernfs_set_rev(parent, dentry);
1158 up_read(&root->kernfs_rwsem);
1160 /* instantiate and hash (possibly negative) dentry */
1161 return d_splice_alias(inode, dentry);
1164 static int kernfs_iop_mkdir(struct user_namespace *mnt_userns,
1165 struct inode *dir, struct dentry *dentry,
1168 struct kernfs_node *parent = dir->i_private;
1169 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1172 if (!scops || !scops->mkdir)
1175 if (!kernfs_get_active(parent))
1178 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1180 kernfs_put_active(parent);
1184 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1186 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1187 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1190 if (!scops || !scops->rmdir)
1193 if (!kernfs_get_active(kn))
1196 ret = scops->rmdir(kn);
1198 kernfs_put_active(kn);
1202 static int kernfs_iop_rename(struct user_namespace *mnt_userns,
1203 struct inode *old_dir, struct dentry *old_dentry,
1204 struct inode *new_dir, struct dentry *new_dentry,
1207 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1208 struct kernfs_node *new_parent = new_dir->i_private;
1209 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1215 if (!scops || !scops->rename)
1218 if (!kernfs_get_active(kn))
1221 if (!kernfs_get_active(new_parent)) {
1222 kernfs_put_active(kn);
1226 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1228 kernfs_put_active(new_parent);
1229 kernfs_put_active(kn);
1233 const struct inode_operations kernfs_dir_iops = {
1234 .lookup = kernfs_iop_lookup,
1235 .permission = kernfs_iop_permission,
1236 .setattr = kernfs_iop_setattr,
1237 .getattr = kernfs_iop_getattr,
1238 .listxattr = kernfs_iop_listxattr,
1240 .mkdir = kernfs_iop_mkdir,
1241 .rmdir = kernfs_iop_rmdir,
1242 .rename = kernfs_iop_rename,
1245 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1247 struct kernfs_node *last;
1250 struct rb_node *rbn;
1254 if (kernfs_type(pos) != KERNFS_DIR)
1257 rbn = rb_first(&pos->dir.children);
1261 pos = rb_to_kn(rbn);
1268 * kernfs_next_descendant_post - find the next descendant for post-order walk
1269 * @pos: the current position (%NULL to initiate traversal)
1270 * @root: kernfs_node whose descendants to walk
1272 * Find the next descendant to visit for post-order traversal of @root's
1273 * descendants. @root is included in the iteration and the last node to be
1276 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1277 struct kernfs_node *root)
1279 struct rb_node *rbn;
1281 lockdep_assert_held_write(&kernfs_root(root)->kernfs_rwsem);
1283 /* if first iteration, visit leftmost descendant which may be root */
1285 return kernfs_leftmost_descendant(root);
1287 /* if we visited @root, we're done */
1291 /* if there's an unvisited sibling, visit its leftmost descendant */
1292 rbn = rb_next(&pos->rb);
1294 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1296 /* no sibling left, visit parent */
1301 * kernfs_activate - activate a node which started deactivated
1302 * @kn: kernfs_node whose subtree is to be activated
1304 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1305 * needs to be explicitly activated. A node which hasn't been activated
1306 * isn't visible to userland and deactivation is skipped during its
1307 * removal. This is useful to construct atomic init sequences where
1308 * creation of multiple nodes should either succeed or fail atomically.
1310 * The caller is responsible for ensuring that this function is not called
1311 * after kernfs_remove*() is invoked on @kn.
1313 void kernfs_activate(struct kernfs_node *kn)
1315 struct kernfs_node *pos;
1316 struct kernfs_root *root = kernfs_root(kn);
1318 down_write(&root->kernfs_rwsem);
1321 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1322 if (pos->flags & KERNFS_ACTIVATED)
1325 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1326 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1328 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1329 pos->flags |= KERNFS_ACTIVATED;
1332 up_write(&root->kernfs_rwsem);
1335 static void __kernfs_remove(struct kernfs_node *kn)
1337 struct kernfs_node *pos;
1339 lockdep_assert_held_write(&kernfs_root(kn)->kernfs_rwsem);
1342 * Short-circuit if non-root @kn has already finished removal.
1343 * This is for kernfs_remove_self() which plays with active ref
1346 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1349 pr_debug("kernfs %s: removing\n", kn->name);
1351 /* prevent any new usage under @kn by deactivating all nodes */
1353 while ((pos = kernfs_next_descendant_post(pos, kn)))
1354 if (kernfs_active(pos))
1355 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1357 /* deactivate and unlink the subtree node-by-node */
1359 pos = kernfs_leftmost_descendant(kn);
1362 * kernfs_drain() drops kernfs_rwsem temporarily and @pos's
1363 * base ref could have been put by someone else by the time
1364 * the function returns. Make sure it doesn't go away
1370 * Drain iff @kn was activated. This avoids draining and
1371 * its lockdep annotations for nodes which have never been
1372 * activated and allows embedding kernfs_remove() in create
1373 * error paths without worrying about draining.
1375 if (kn->flags & KERNFS_ACTIVATED)
1378 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1381 * kernfs_unlink_sibling() succeeds once per node. Use it
1382 * to decide who's responsible for cleanups.
1384 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1385 struct kernfs_iattrs *ps_iattr =
1386 pos->parent ? pos->parent->iattr : NULL;
1388 /* update timestamps on the parent */
1390 ktime_get_real_ts64(&ps_iattr->ia_ctime);
1391 ps_iattr->ia_mtime = ps_iattr->ia_ctime;
1398 } while (pos != kn);
1402 * kernfs_remove - remove a kernfs_node recursively
1403 * @kn: the kernfs_node to remove
1405 * Remove @kn along with all its subdirectories and files.
1407 void kernfs_remove(struct kernfs_node *kn)
1409 struct kernfs_root *root = kernfs_root(kn);
1411 down_write(&root->kernfs_rwsem);
1412 __kernfs_remove(kn);
1413 up_write(&root->kernfs_rwsem);
1417 * kernfs_break_active_protection - break out of active protection
1418 * @kn: the self kernfs_node
1420 * The caller must be running off of a kernfs operation which is invoked
1421 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1422 * this function must also be matched with an invocation of
1423 * kernfs_unbreak_active_protection().
1425 * This function releases the active reference of @kn the caller is
1426 * holding. Once this function is called, @kn may be removed at any point
1427 * and the caller is solely responsible for ensuring that the objects it
1428 * dereferences are accessible.
1430 void kernfs_break_active_protection(struct kernfs_node *kn)
1433 * Take out ourself out of the active ref dependency chain. If
1434 * we're called without an active ref, lockdep will complain.
1436 kernfs_put_active(kn);
1440 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1441 * @kn: the self kernfs_node
1443 * If kernfs_break_active_protection() was called, this function must be
1444 * invoked before finishing the kernfs operation. Note that while this
1445 * function restores the active reference, it doesn't and can't actually
1446 * restore the active protection - @kn may already or be in the process of
1447 * being removed. Once kernfs_break_active_protection() is invoked, that
1448 * protection is irreversibly gone for the kernfs operation instance.
1450 * While this function may be called at any point after
1451 * kernfs_break_active_protection() is invoked, its most useful location
1452 * would be right before the enclosing kernfs operation returns.
1454 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1457 * @kn->active could be in any state; however, the increment we do
1458 * here will be undone as soon as the enclosing kernfs operation
1459 * finishes and this temporary bump can't break anything. If @kn
1460 * is alive, nothing changes. If @kn is being deactivated, the
1461 * soon-to-follow put will either finish deactivation or restore
1462 * deactivated state. If @kn is already removed, the temporary
1463 * bump is guaranteed to be gone before @kn is released.
1465 atomic_inc(&kn->active);
1466 if (kernfs_lockdep(kn))
1467 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1471 * kernfs_remove_self - remove a kernfs_node from its own method
1472 * @kn: the self kernfs_node to remove
1474 * The caller must be running off of a kernfs operation which is invoked
1475 * with an active reference - e.g. one of kernfs_ops. This can be used to
1476 * implement a file operation which deletes itself.
1478 * For example, the "delete" file for a sysfs device directory can be
1479 * implemented by invoking kernfs_remove_self() on the "delete" file
1480 * itself. This function breaks the circular dependency of trying to
1481 * deactivate self while holding an active ref itself. It isn't necessary
1482 * to modify the usual removal path to use kernfs_remove_self(). The
1483 * "delete" implementation can simply invoke kernfs_remove_self() on self
1484 * before proceeding with the usual removal path. kernfs will ignore later
1485 * kernfs_remove() on self.
1487 * kernfs_remove_self() can be called multiple times concurrently on the
1488 * same kernfs_node. Only the first one actually performs removal and
1489 * returns %true. All others will wait until the kernfs operation which
1490 * won self-removal finishes and return %false. Note that the losers wait
1491 * for the completion of not only the winning kernfs_remove_self() but also
1492 * the whole kernfs_ops which won the arbitration. This can be used to
1493 * guarantee, for example, all concurrent writes to a "delete" file to
1494 * finish only after the whole operation is complete.
1496 bool kernfs_remove_self(struct kernfs_node *kn)
1499 struct kernfs_root *root = kernfs_root(kn);
1501 down_write(&root->kernfs_rwsem);
1502 kernfs_break_active_protection(kn);
1505 * SUICIDAL is used to arbitrate among competing invocations. Only
1506 * the first one will actually perform removal. When the removal
1507 * is complete, SUICIDED is set and the active ref is restored
1508 * while kernfs_rwsem for held exclusive. The ones which lost
1509 * arbitration waits for SUICIDED && drained which can happen only
1510 * after the enclosing kernfs operation which executed the winning
1511 * instance of kernfs_remove_self() finished.
1513 if (!(kn->flags & KERNFS_SUICIDAL)) {
1514 kn->flags |= KERNFS_SUICIDAL;
1515 __kernfs_remove(kn);
1516 kn->flags |= KERNFS_SUICIDED;
1519 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1523 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1525 if ((kn->flags & KERNFS_SUICIDED) &&
1526 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1529 up_write(&root->kernfs_rwsem);
1531 down_write(&root->kernfs_rwsem);
1533 finish_wait(waitq, &wait);
1534 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1539 * This must be done while kernfs_rwsem held exclusive; otherwise,
1540 * waiting for SUICIDED && deactivated could finish prematurely.
1542 kernfs_unbreak_active_protection(kn);
1544 up_write(&root->kernfs_rwsem);
1549 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1550 * @parent: parent of the target
1551 * @name: name of the kernfs_node to remove
1552 * @ns: namespace tag of the kernfs_node to remove
1554 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1555 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1557 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1560 struct kernfs_node *kn;
1561 struct kernfs_root *root;
1564 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1569 root = kernfs_root(parent);
1570 down_write(&root->kernfs_rwsem);
1572 kn = kernfs_find_ns(parent, name, ns);
1574 __kernfs_remove(kn);
1576 up_write(&root->kernfs_rwsem);
1585 * kernfs_rename_ns - move and rename a kernfs_node
1587 * @new_parent: new parent to put @sd under
1588 * @new_name: new name
1589 * @new_ns: new namespace tag
1591 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1592 const char *new_name, const void *new_ns)
1594 struct kernfs_node *old_parent;
1595 struct kernfs_root *root;
1596 const char *old_name = NULL;
1599 /* can't move or rename root */
1603 root = kernfs_root(kn);
1604 down_write(&root->kernfs_rwsem);
1607 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1608 (new_parent->flags & KERNFS_EMPTY_DIR))
1612 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1613 (strcmp(kn->name, new_name) == 0))
1614 goto out; /* nothing to rename */
1617 if (kernfs_find_ns(new_parent, new_name, new_ns))
1620 /* rename kernfs_node */
1621 if (strcmp(kn->name, new_name) != 0) {
1623 new_name = kstrdup_const(new_name, GFP_KERNEL);
1631 * Move to the appropriate place in the appropriate directories rbtree.
1633 kernfs_unlink_sibling(kn);
1634 kernfs_get(new_parent);
1636 /* rename_lock protects ->parent and ->name accessors */
1637 spin_lock_irq(&kernfs_rename_lock);
1639 old_parent = kn->parent;
1640 kn->parent = new_parent;
1644 old_name = kn->name;
1645 kn->name = new_name;
1648 spin_unlock_irq(&kernfs_rename_lock);
1650 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1651 kernfs_link_sibling(kn);
1653 kernfs_put(old_parent);
1654 kfree_const(old_name);
1658 up_write(&root->kernfs_rwsem);
1662 /* Relationship between mode and the DT_xxx types */
1663 static inline unsigned char dt_type(struct kernfs_node *kn)
1665 return (kn->mode >> 12) & 15;
1668 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1670 kernfs_put(filp->private_data);
1674 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1675 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1678 int valid = kernfs_active(pos) &&
1679 pos->parent == parent && hash == pos->hash;
1684 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1685 struct rb_node *node = parent->dir.children.rb_node;
1687 pos = rb_to_kn(node);
1689 if (hash < pos->hash)
1690 node = node->rb_left;
1691 else if (hash > pos->hash)
1692 node = node->rb_right;
1697 /* Skip over entries which are dying/dead or in the wrong namespace */
1698 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1699 struct rb_node *node = rb_next(&pos->rb);
1703 pos = rb_to_kn(node);
1708 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1709 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1711 pos = kernfs_dir_pos(ns, parent, ino, pos);
1714 struct rb_node *node = rb_next(&pos->rb);
1718 pos = rb_to_kn(node);
1719 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1724 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1726 struct dentry *dentry = file->f_path.dentry;
1727 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1728 struct kernfs_node *pos = file->private_data;
1729 struct kernfs_root *root;
1730 const void *ns = NULL;
1732 if (!dir_emit_dots(file, ctx))
1735 root = kernfs_root(parent);
1736 down_read(&root->kernfs_rwsem);
1738 if (kernfs_ns_enabled(parent))
1739 ns = kernfs_info(dentry->d_sb)->ns;
1741 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1743 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1744 const char *name = pos->name;
1745 unsigned int type = dt_type(pos);
1746 int len = strlen(name);
1747 ino_t ino = kernfs_ino(pos);
1749 ctx->pos = pos->hash;
1750 file->private_data = pos;
1753 up_read(&root->kernfs_rwsem);
1754 if (!dir_emit(ctx, name, len, ino, type))
1756 down_read(&root->kernfs_rwsem);
1758 up_read(&root->kernfs_rwsem);
1759 file->private_data = NULL;
1764 const struct file_operations kernfs_dir_fops = {
1765 .read = generic_read_dir,
1766 .iterate_shared = kernfs_fop_readdir,
1767 .release = kernfs_dir_fop_release,
1768 .llseek = generic_file_llseek,