2 * fs/kernfs/dir.c - kernfs directory implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
8 * This file is released under the GPLv2.
11 #include <linux/sched.h>
13 #include <linux/namei.h>
14 #include <linux/idr.h>
15 #include <linux/slab.h>
16 #include <linux/security.h>
17 #include <linux/hash.h>
19 #include "kernfs-internal.h"
21 DEFINE_MUTEX(kernfs_mutex);
22 static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
24 * Don't use rename_lock to piggy back on pr_cont_buf. We don't want to
25 * call pr_cont() while holding rename_lock. Because sometimes pr_cont()
26 * will perform wakeups when releasing console_sem. Holding rename_lock
27 * will introduce deadlock if the scheduler reads the kernfs_name in the
30 static DEFINE_SPINLOCK(kernfs_pr_cont_lock);
31 static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by pr_cont_lock */
32 static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */
34 #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb)
36 static bool kernfs_active(struct kernfs_node *kn)
38 lockdep_assert_held(&kernfs_mutex);
39 return atomic_read(&kn->active) >= 0;
42 static bool kernfs_lockdep(struct kernfs_node *kn)
44 #ifdef CONFIG_DEBUG_LOCK_ALLOC
45 return kn->flags & KERNFS_LOCKDEP;
51 static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
54 return strlcpy(buf, "(null)", buflen);
56 return strlcpy(buf, kn->parent ? kn->name : "/", buflen);
59 /* kernfs_node_depth - compute depth from @from to @to */
60 static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
64 while (to->parent && to != from) {
71 static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a,
72 struct kernfs_node *b)
75 struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b);
80 da = kernfs_depth(ra->kn, a);
81 db = kernfs_depth(rb->kn, b);
92 /* worst case b and a will be the same at root */
102 * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to,
103 * where kn_from is treated as root of the path.
104 * @kn_from: kernfs node which should be treated as root for the path
105 * @kn_to: kernfs node to which path is needed
106 * @buf: buffer to copy the path into
107 * @buflen: size of @buf
109 * We need to handle couple of scenarios here:
110 * [1] when @kn_from is an ancestor of @kn_to at some level
112 * kn_to: /n1/n2/n3/n4/n5
115 * [2] when @kn_from is on a different hierarchy and we need to find common
116 * ancestor between @kn_from and @kn_to.
117 * kn_from: /n1/n2/n3/n4
121 * kn_from: /n1/n2/n3/n4/n5 [depth=5]
122 * kn_to: /n1/n2/n3 [depth=3]
125 * [3] when @kn_to is NULL result will be "(null)"
127 * Returns the length of the full path. If the full length is equal to or
128 * greater than @buflen, @buf contains the truncated path with the trailing
129 * '\0'. On error, -errno is returned.
131 static int kernfs_path_from_node_locked(struct kernfs_node *kn_to,
132 struct kernfs_node *kn_from,
133 char *buf, size_t buflen)
135 struct kernfs_node *kn, *common;
136 const char parent_str[] = "/..";
137 size_t depth_from, depth_to, len = 0;
141 return strlcpy(buf, "(null)", buflen);
144 kn_from = kernfs_root(kn_to)->kn;
146 if (kn_from == kn_to)
147 return strlcpy(buf, "/", buflen);
149 common = kernfs_common_ancestor(kn_from, kn_to);
150 if (WARN_ON(!common))
153 depth_to = kernfs_depth(common, kn_to);
154 depth_from = kernfs_depth(common, kn_from);
159 for (i = 0; i < depth_from; i++)
160 len += strlcpy(buf + len, parent_str,
161 len < buflen ? buflen - len : 0);
163 /* Calculate how many bytes we need for the rest */
164 for (i = depth_to - 1; i >= 0; i--) {
165 for (kn = kn_to, j = 0; j < i; j++)
167 len += strlcpy(buf + len, "/",
168 len < buflen ? buflen - len : 0);
169 len += strlcpy(buf + len, kn->name,
170 len < buflen ? buflen - len : 0);
177 * kernfs_name - obtain the name of a given node
178 * @kn: kernfs_node of interest
179 * @buf: buffer to copy @kn's name into
180 * @buflen: size of @buf
182 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
183 * similar to strlcpy(). It returns the length of @kn's name and if @buf
184 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
186 * Fills buffer with "(null)" if @kn is NULL.
188 * This function can be called from any context.
190 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
195 spin_lock_irqsave(&kernfs_rename_lock, flags);
196 ret = kernfs_name_locked(kn, buf, buflen);
197 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
202 * kernfs_path_from_node - build path of node @to relative to @from.
203 * @from: parent kernfs_node relative to which we need to build the path
204 * @to: kernfs_node of interest
205 * @buf: buffer to copy @to's path into
206 * @buflen: size of @buf
208 * Builds @to's path relative to @from in @buf. @from and @to must
209 * be on the same kernfs-root. If @from is not parent of @to, then a relative
210 * path (which includes '..'s) as needed to reach from @from to @to is
213 * Returns the length of the full path. If the full length is equal to or
214 * greater than @buflen, @buf contains the truncated path with the trailing
215 * '\0'. On error, -errno is returned.
217 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
218 char *buf, size_t buflen)
223 spin_lock_irqsave(&kernfs_rename_lock, flags);
224 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
225 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
228 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
231 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
232 * @kn: kernfs_node of interest
234 * This function can be called from any context.
236 void pr_cont_kernfs_name(struct kernfs_node *kn)
240 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
242 kernfs_name(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
243 pr_cont("%s", kernfs_pr_cont_buf);
245 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
249 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
250 * @kn: kernfs_node of interest
252 * This function can be called from any context.
254 void pr_cont_kernfs_path(struct kernfs_node *kn)
259 spin_lock_irqsave(&kernfs_pr_cont_lock, flags);
261 sz = kernfs_path_from_node(kn, NULL, kernfs_pr_cont_buf,
262 sizeof(kernfs_pr_cont_buf));
268 if (sz >= sizeof(kernfs_pr_cont_buf)) {
269 pr_cont("(name too long)");
273 pr_cont("%s", kernfs_pr_cont_buf);
276 spin_unlock_irqrestore(&kernfs_pr_cont_lock, flags);
280 * kernfs_get_parent - determine the parent node and pin it
281 * @kn: kernfs_node of interest
283 * Determines @kn's parent, pins and returns it. This function can be
284 * called from any context.
286 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
288 struct kernfs_node *parent;
291 spin_lock_irqsave(&kernfs_rename_lock, flags);
294 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
301 * @name: Null terminated string to hash
302 * @ns: Namespace tag to hash
304 * Returns 31 bit hash of ns + name (so it fits in an off_t )
306 static unsigned int kernfs_name_hash(const char *name, const void *ns)
308 unsigned long hash = init_name_hash(ns);
309 unsigned int len = strlen(name);
311 hash = partial_name_hash(*name++, hash);
312 hash = end_name_hash(hash);
314 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
322 static int kernfs_name_compare(unsigned int hash, const char *name,
323 const void *ns, const struct kernfs_node *kn)
333 return strcmp(name, kn->name);
336 static int kernfs_sd_compare(const struct kernfs_node *left,
337 const struct kernfs_node *right)
339 return kernfs_name_compare(left->hash, left->name, left->ns, right);
343 * kernfs_link_sibling - link kernfs_node into sibling rbtree
344 * @kn: kernfs_node of interest
346 * Link @kn into its sibling rbtree which starts from
347 * @kn->parent->dir.children.
350 * mutex_lock(kernfs_mutex)
353 * 0 on susccess -EEXIST on failure.
355 static int kernfs_link_sibling(struct kernfs_node *kn)
357 struct rb_node **node = &kn->parent->dir.children.rb_node;
358 struct rb_node *parent = NULL;
361 struct kernfs_node *pos;
364 pos = rb_to_kn(*node);
366 result = kernfs_sd_compare(kn, pos);
368 node = &pos->rb.rb_left;
370 node = &pos->rb.rb_right;
375 /* add new node and rebalance the tree */
376 rb_link_node(&kn->rb, parent, node);
377 rb_insert_color(&kn->rb, &kn->parent->dir.children);
379 /* successfully added, account subdir number */
380 if (kernfs_type(kn) == KERNFS_DIR)
381 kn->parent->dir.subdirs++;
387 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
388 * @kn: kernfs_node of interest
390 * Try to unlink @kn from its sibling rbtree which starts from
391 * kn->parent->dir.children. Returns %true if @kn was actually
392 * removed, %false if @kn wasn't on the rbtree.
395 * mutex_lock(kernfs_mutex)
397 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
399 if (RB_EMPTY_NODE(&kn->rb))
402 if (kernfs_type(kn) == KERNFS_DIR)
403 kn->parent->dir.subdirs--;
405 rb_erase(&kn->rb, &kn->parent->dir.children);
406 RB_CLEAR_NODE(&kn->rb);
411 * kernfs_get_active - get an active reference to kernfs_node
412 * @kn: kernfs_node to get an active reference to
414 * Get an active reference of @kn. This function is noop if @kn
418 * Pointer to @kn on success, NULL on failure.
420 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
425 if (!atomic_inc_unless_negative(&kn->active))
428 if (kernfs_lockdep(kn))
429 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
434 * kernfs_put_active - put an active reference to kernfs_node
435 * @kn: kernfs_node to put an active reference to
437 * Put an active reference to @kn. This function is noop if @kn
440 void kernfs_put_active(struct kernfs_node *kn)
442 struct kernfs_root *root = kernfs_root(kn);
448 if (kernfs_lockdep(kn))
449 rwsem_release(&kn->dep_map, 1, _RET_IP_);
450 v = atomic_dec_return(&kn->active);
451 if (likely(v != KN_DEACTIVATED_BIAS))
454 wake_up_all(&root->deactivate_waitq);
458 * kernfs_drain - drain kernfs_node
459 * @kn: kernfs_node to drain
461 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
462 * removers may invoke this function concurrently on @kn and all will
463 * return after draining is complete.
465 static void kernfs_drain(struct kernfs_node *kn)
466 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
468 struct kernfs_root *root = kernfs_root(kn);
470 lockdep_assert_held(&kernfs_mutex);
471 WARN_ON_ONCE(kernfs_active(kn));
473 mutex_unlock(&kernfs_mutex);
475 if (kernfs_lockdep(kn)) {
476 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
477 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
478 lock_contended(&kn->dep_map, _RET_IP_);
481 /* but everyone should wait for draining */
482 wait_event(root->deactivate_waitq,
483 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
485 if (kernfs_lockdep(kn)) {
486 lock_acquired(&kn->dep_map, _RET_IP_);
487 rwsem_release(&kn->dep_map, 1, _RET_IP_);
490 kernfs_drain_open_files(kn);
492 mutex_lock(&kernfs_mutex);
496 * kernfs_get - get a reference count on a kernfs_node
497 * @kn: the target kernfs_node
499 void kernfs_get(struct kernfs_node *kn)
502 WARN_ON(!atomic_read(&kn->count));
503 atomic_inc(&kn->count);
506 EXPORT_SYMBOL_GPL(kernfs_get);
509 * kernfs_put - put a reference count on a kernfs_node
510 * @kn: the target kernfs_node
512 * Put a reference count of @kn and destroy it if it reached zero.
514 void kernfs_put(struct kernfs_node *kn)
516 struct kernfs_node *parent;
517 struct kernfs_root *root;
520 * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino
521 * depends on this to filter reused stale node
523 if (!kn || !atomic_dec_and_test(&kn->count))
525 root = kernfs_root(kn);
528 * Moving/renaming is always done while holding reference.
529 * kn->parent won't change beneath us.
533 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
534 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
535 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
537 if (kernfs_type(kn) == KERNFS_LINK)
538 kernfs_put(kn->symlink.target_kn);
540 kfree_const(kn->name);
543 if (kn->iattr->ia_secdata)
544 security_release_secctx(kn->iattr->ia_secdata,
545 kn->iattr->ia_secdata_len);
546 simple_xattrs_free(&kn->iattr->xattrs);
549 spin_lock(&kernfs_idr_lock);
550 idr_remove(&root->ino_idr, kn->id.ino);
551 spin_unlock(&kernfs_idr_lock);
552 kmem_cache_free(kernfs_node_cache, kn);
556 if (atomic_dec_and_test(&kn->count))
559 /* just released the root kn, free @root too */
560 idr_destroy(&root->ino_idr);
564 EXPORT_SYMBOL_GPL(kernfs_put);
566 static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags)
568 struct kernfs_node *kn;
570 if (flags & LOOKUP_RCU)
573 /* Always perform fresh lookup for negatives */
574 if (d_really_is_negative(dentry))
575 goto out_bad_unlocked;
577 kn = kernfs_dentry_node(dentry);
578 mutex_lock(&kernfs_mutex);
580 /* The kernfs node has been deactivated */
581 if (!kernfs_active(kn))
584 /* The kernfs node has been moved? */
585 if (kernfs_dentry_node(dentry->d_parent) != kn->parent)
588 /* The kernfs node has been renamed */
589 if (strcmp(dentry->d_name.name, kn->name) != 0)
592 /* The kernfs node has been moved to a different namespace */
593 if (kn->parent && kernfs_ns_enabled(kn->parent) &&
594 kernfs_info(dentry->d_sb)->ns != kn->ns)
597 mutex_unlock(&kernfs_mutex);
600 mutex_unlock(&kernfs_mutex);
605 const struct dentry_operations kernfs_dops = {
606 .d_revalidate = kernfs_dop_revalidate,
610 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
611 * @dentry: the dentry in question
613 * Return the kernfs_node associated with @dentry. If @dentry is not a
614 * kernfs one, %NULL is returned.
616 * While the returned kernfs_node will stay accessible as long as @dentry
617 * is accessible, the returned node can be in any state and the caller is
618 * fully responsible for determining what's accessible.
620 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
622 if (dentry->d_sb->s_op == &kernfs_sops &&
623 !d_really_is_negative(dentry))
624 return kernfs_dentry_node(dentry);
628 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
629 const char *name, umode_t mode,
630 kuid_t uid, kgid_t gid,
633 struct kernfs_node *kn;
637 name = kstrdup_const(name, GFP_KERNEL);
641 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
645 idr_preload(GFP_KERNEL);
646 spin_lock(&kernfs_idr_lock);
647 ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
648 if (ret >= 0 && ret < root->last_ino)
649 root->next_generation++;
650 gen = root->next_generation;
651 root->last_ino = ret;
652 spin_unlock(&kernfs_idr_lock);
657 kn->id.generation = gen;
660 * set ino first. This RELEASE is paired with atomic_inc_not_zero in
661 * kernfs_find_and_get_node_by_ino
663 atomic_set_release(&kn->count, 1);
664 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
665 RB_CLEAR_NODE(&kn->rb);
671 if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) {
672 struct iattr iattr = {
673 .ia_valid = ATTR_UID | ATTR_GID,
678 ret = __kernfs_setattr(kn, &iattr);
686 idr_remove(&root->ino_idr, kn->id.ino);
688 kmem_cache_free(kernfs_node_cache, kn);
694 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
695 const char *name, umode_t mode,
696 kuid_t uid, kgid_t gid,
699 struct kernfs_node *kn;
701 kn = __kernfs_new_node(kernfs_root(parent),
702 name, mode, uid, gid, flags);
711 * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number
712 * @root: the kernfs root
716 * NULL on failure. Return a kernfs node with reference counter incremented
718 struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root,
721 struct kernfs_node *kn;
724 kn = idr_find(&root->ino_idr, ino);
729 * Since kernfs_node is freed in RCU, it's possible an old node for ino
730 * is freed, but reused before RCU grace period. But a freed node (see
731 * kernfs_put) or an incompletedly initialized node (see
732 * __kernfs_new_node) should have 'count' 0. We can use this fact to
733 * filter out such node.
735 if (!atomic_inc_not_zero(&kn->count)) {
741 * The node could be a new node or a reused node. If it's a new node,
742 * we are ok. If it's reused because of RCU (because of
743 * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino'
744 * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate,
745 * hence we can use 'ino' to filter stale node.
747 if (kn->id.ino != ino)
759 * kernfs_add_one - add kernfs_node to parent without warning
760 * @kn: kernfs_node to be added
762 * The caller must already have initialized @kn->parent. This
763 * function increments nlink of the parent's inode if @kn is a
764 * directory and link into the children list of the parent.
767 * 0 on success, -EEXIST if entry with the given name already
770 int kernfs_add_one(struct kernfs_node *kn)
772 struct kernfs_node *parent = kn->parent;
773 struct kernfs_iattrs *ps_iattr;
777 mutex_lock(&kernfs_mutex);
780 has_ns = kernfs_ns_enabled(parent);
781 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
782 has_ns ? "required" : "invalid", parent->name, kn->name))
785 if (kernfs_type(parent) != KERNFS_DIR)
789 if (parent->flags & KERNFS_EMPTY_DIR)
792 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
795 kn->hash = kernfs_name_hash(kn->name, kn->ns);
797 ret = kernfs_link_sibling(kn);
801 /* Update timestamps on the parent */
802 ps_iattr = parent->iattr;
804 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
805 ktime_get_real_ts64(&ps_iattrs->ia_ctime);
806 ps_iattrs->ia_mtime = ps_iattrs->ia_ctime;
809 mutex_unlock(&kernfs_mutex);
812 * Activate the new node unless CREATE_DEACTIVATED is requested.
813 * If not activated here, the kernfs user is responsible for
814 * activating the node with kernfs_activate(). A node which hasn't
815 * been activated is not visible to userland and its removal won't
816 * trigger deactivation.
818 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
823 mutex_unlock(&kernfs_mutex);
828 * kernfs_find_ns - find kernfs_node with the given name
829 * @parent: kernfs_node to search under
830 * @name: name to look for
831 * @ns: the namespace tag to use
833 * Look for kernfs_node with name @name under @parent. Returns pointer to
834 * the found kernfs_node on success, %NULL on failure.
836 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
837 const unsigned char *name,
840 struct rb_node *node = parent->dir.children.rb_node;
841 bool has_ns = kernfs_ns_enabled(parent);
844 lockdep_assert_held(&kernfs_mutex);
846 if (has_ns != (bool)ns) {
847 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
848 has_ns ? "required" : "invalid", parent->name, name);
852 hash = kernfs_name_hash(name, ns);
854 struct kernfs_node *kn;
858 result = kernfs_name_compare(hash, name, ns, kn);
860 node = node->rb_left;
862 node = node->rb_right;
869 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
870 const unsigned char *path,
876 lockdep_assert_held(&kernfs_mutex);
878 spin_lock_irq(&kernfs_pr_cont_lock);
880 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
882 if (len >= sizeof(kernfs_pr_cont_buf)) {
883 spin_unlock_irq(&kernfs_pr_cont_lock);
887 p = kernfs_pr_cont_buf;
889 while ((name = strsep(&p, "/")) && parent) {
892 parent = kernfs_find_ns(parent, name, ns);
895 spin_unlock_irq(&kernfs_pr_cont_lock);
901 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
902 * @parent: kernfs_node to search under
903 * @name: name to look for
904 * @ns: the namespace tag to use
906 * Look for kernfs_node with name @name under @parent and get a reference
907 * if found. This function may sleep and returns pointer to the found
908 * kernfs_node on success, %NULL on failure.
910 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
911 const char *name, const void *ns)
913 struct kernfs_node *kn;
915 mutex_lock(&kernfs_mutex);
916 kn = kernfs_find_ns(parent, name, ns);
918 mutex_unlock(&kernfs_mutex);
922 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
925 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
926 * @parent: kernfs_node to search under
927 * @path: path to look for
928 * @ns: the namespace tag to use
930 * Look for kernfs_node with path @path under @parent and get a reference
931 * if found. This function may sleep and returns pointer to the found
932 * kernfs_node on success, %NULL on failure.
934 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
935 const char *path, const void *ns)
937 struct kernfs_node *kn;
939 mutex_lock(&kernfs_mutex);
940 kn = kernfs_walk_ns(parent, path, ns);
942 mutex_unlock(&kernfs_mutex);
948 * kernfs_create_root - create a new kernfs hierarchy
949 * @scops: optional syscall operations for the hierarchy
950 * @flags: KERNFS_ROOT_* flags
951 * @priv: opaque data associated with the new directory
953 * Returns the root of the new hierarchy on success, ERR_PTR() value on
956 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
957 unsigned int flags, void *priv)
959 struct kernfs_root *root;
960 struct kernfs_node *kn;
962 root = kzalloc(sizeof(*root), GFP_KERNEL);
964 return ERR_PTR(-ENOMEM);
966 idr_init(&root->ino_idr);
967 INIT_LIST_HEAD(&root->supers);
968 root->next_generation = 1;
970 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
971 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
974 idr_destroy(&root->ino_idr);
976 return ERR_PTR(-ENOMEM);
982 root->syscall_ops = scops;
985 init_waitqueue_head(&root->deactivate_waitq);
987 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
994 * kernfs_destroy_root - destroy a kernfs hierarchy
995 * @root: root of the hierarchy to destroy
997 * Destroy the hierarchy anchored at @root by removing all existing
998 * directories and destroying @root.
1000 void kernfs_destroy_root(struct kernfs_root *root)
1002 kernfs_remove(root->kn); /* will also free @root */
1006 * kernfs_create_dir_ns - create a directory
1007 * @parent: parent in which to create a new directory
1008 * @name: name of the new directory
1009 * @mode: mode of the new directory
1010 * @uid: uid of the new directory
1011 * @gid: gid of the new directory
1012 * @priv: opaque data associated with the new directory
1013 * @ns: optional namespace tag of the directory
1015 * Returns the created node on success, ERR_PTR() value on failure.
1017 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
1018 const char *name, umode_t mode,
1019 kuid_t uid, kgid_t gid,
1020 void *priv, const void *ns)
1022 struct kernfs_node *kn;
1026 kn = kernfs_new_node(parent, name, mode | S_IFDIR,
1027 uid, gid, KERNFS_DIR);
1029 return ERR_PTR(-ENOMEM);
1031 kn->dir.root = parent->dir.root;
1036 rc = kernfs_add_one(kn);
1045 * kernfs_create_empty_dir - create an always empty directory
1046 * @parent: parent in which to create a new directory
1047 * @name: name of the new directory
1049 * Returns the created node on success, ERR_PTR() value on failure.
1051 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
1054 struct kernfs_node *kn;
1058 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR,
1059 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR);
1061 return ERR_PTR(-ENOMEM);
1063 kn->flags |= KERNFS_EMPTY_DIR;
1064 kn->dir.root = parent->dir.root;
1069 rc = kernfs_add_one(kn);
1077 static struct dentry *kernfs_iop_lookup(struct inode *dir,
1078 struct dentry *dentry,
1082 struct kernfs_node *parent = dir->i_private;
1083 struct kernfs_node *kn;
1084 struct inode *inode;
1085 const void *ns = NULL;
1087 mutex_lock(&kernfs_mutex);
1089 if (kernfs_ns_enabled(parent))
1090 ns = kernfs_info(dir->i_sb)->ns;
1092 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
1095 if (!kn || !kernfs_active(kn)) {
1100 /* attach dentry and inode */
1101 inode = kernfs_get_inode(dir->i_sb, kn);
1103 ret = ERR_PTR(-ENOMEM);
1107 /* instantiate and hash dentry */
1108 ret = d_splice_alias(inode, dentry);
1110 mutex_unlock(&kernfs_mutex);
1114 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1117 struct kernfs_node *parent = dir->i_private;
1118 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1121 if (!scops || !scops->mkdir)
1124 if (!kernfs_get_active(parent))
1127 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1129 kernfs_put_active(parent);
1133 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1135 struct kernfs_node *kn = kernfs_dentry_node(dentry);
1136 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1139 if (!scops || !scops->rmdir)
1142 if (!kernfs_get_active(kn))
1145 ret = scops->rmdir(kn);
1147 kernfs_put_active(kn);
1151 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1152 struct inode *new_dir, struct dentry *new_dentry,
1155 struct kernfs_node *kn = kernfs_dentry_node(old_dentry);
1156 struct kernfs_node *new_parent = new_dir->i_private;
1157 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1163 if (!scops || !scops->rename)
1166 if (!kernfs_get_active(kn))
1169 if (!kernfs_get_active(new_parent)) {
1170 kernfs_put_active(kn);
1174 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1176 kernfs_put_active(new_parent);
1177 kernfs_put_active(kn);
1181 const struct inode_operations kernfs_dir_iops = {
1182 .lookup = kernfs_iop_lookup,
1183 .permission = kernfs_iop_permission,
1184 .setattr = kernfs_iop_setattr,
1185 .getattr = kernfs_iop_getattr,
1186 .listxattr = kernfs_iop_listxattr,
1188 .mkdir = kernfs_iop_mkdir,
1189 .rmdir = kernfs_iop_rmdir,
1190 .rename = kernfs_iop_rename,
1193 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1195 struct kernfs_node *last;
1198 struct rb_node *rbn;
1202 if (kernfs_type(pos) != KERNFS_DIR)
1205 rbn = rb_first(&pos->dir.children);
1209 pos = rb_to_kn(rbn);
1216 * kernfs_next_descendant_post - find the next descendant for post-order walk
1217 * @pos: the current position (%NULL to initiate traversal)
1218 * @root: kernfs_node whose descendants to walk
1220 * Find the next descendant to visit for post-order traversal of @root's
1221 * descendants. @root is included in the iteration and the last node to be
1224 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1225 struct kernfs_node *root)
1227 struct rb_node *rbn;
1229 lockdep_assert_held(&kernfs_mutex);
1231 /* if first iteration, visit leftmost descendant which may be root */
1233 return kernfs_leftmost_descendant(root);
1235 /* if we visited @root, we're done */
1239 /* if there's an unvisited sibling, visit its leftmost descendant */
1240 rbn = rb_next(&pos->rb);
1242 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1244 /* no sibling left, visit parent */
1249 * kernfs_activate - activate a node which started deactivated
1250 * @kn: kernfs_node whose subtree is to be activated
1252 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1253 * needs to be explicitly activated. A node which hasn't been activated
1254 * isn't visible to userland and deactivation is skipped during its
1255 * removal. This is useful to construct atomic init sequences where
1256 * creation of multiple nodes should either succeed or fail atomically.
1258 * The caller is responsible for ensuring that this function is not called
1259 * after kernfs_remove*() is invoked on @kn.
1261 void kernfs_activate(struct kernfs_node *kn)
1263 struct kernfs_node *pos;
1265 mutex_lock(&kernfs_mutex);
1268 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1269 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1272 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1273 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1275 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1276 pos->flags |= KERNFS_ACTIVATED;
1279 mutex_unlock(&kernfs_mutex);
1282 static void __kernfs_remove(struct kernfs_node *kn)
1284 struct kernfs_node *pos;
1286 lockdep_assert_held(&kernfs_mutex);
1289 * Short-circuit if non-root @kn has already finished removal.
1290 * This is for kernfs_remove_self() which plays with active ref
1293 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1296 pr_debug("kernfs %s: removing\n", kn->name);
1298 /* prevent any new usage under @kn by deactivating all nodes */
1300 while ((pos = kernfs_next_descendant_post(pos, kn)))
1301 if (kernfs_active(pos))
1302 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1304 /* deactivate and unlink the subtree node-by-node */
1306 pos = kernfs_leftmost_descendant(kn);
1309 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1310 * base ref could have been put by someone else by the time
1311 * the function returns. Make sure it doesn't go away
1317 * Drain iff @kn was activated. This avoids draining and
1318 * its lockdep annotations for nodes which have never been
1319 * activated and allows embedding kernfs_remove() in create
1320 * error paths without worrying about draining.
1322 if (kn->flags & KERNFS_ACTIVATED)
1325 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1328 * kernfs_unlink_sibling() succeeds once per node. Use it
1329 * to decide who's responsible for cleanups.
1331 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1332 struct kernfs_iattrs *ps_iattr =
1333 pos->parent ? pos->parent->iattr : NULL;
1335 /* update timestamps on the parent */
1337 ktime_get_real_ts64(&ps_iattr->ia_iattr.ia_ctime);
1338 ps_iattr->ia_iattr.ia_mtime =
1339 ps_iattr->ia_iattr.ia_ctime;
1346 } while (pos != kn);
1350 * kernfs_remove - remove a kernfs_node recursively
1351 * @kn: the kernfs_node to remove
1353 * Remove @kn along with all its subdirectories and files.
1355 void kernfs_remove(struct kernfs_node *kn)
1357 mutex_lock(&kernfs_mutex);
1358 __kernfs_remove(kn);
1359 mutex_unlock(&kernfs_mutex);
1363 * kernfs_break_active_protection - break out of active protection
1364 * @kn: the self kernfs_node
1366 * The caller must be running off of a kernfs operation which is invoked
1367 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1368 * this function must also be matched with an invocation of
1369 * kernfs_unbreak_active_protection().
1371 * This function releases the active reference of @kn the caller is
1372 * holding. Once this function is called, @kn may be removed at any point
1373 * and the caller is solely responsible for ensuring that the objects it
1374 * dereferences are accessible.
1376 void kernfs_break_active_protection(struct kernfs_node *kn)
1379 * Take out ourself out of the active ref dependency chain. If
1380 * we're called without an active ref, lockdep will complain.
1382 kernfs_put_active(kn);
1386 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1387 * @kn: the self kernfs_node
1389 * If kernfs_break_active_protection() was called, this function must be
1390 * invoked before finishing the kernfs operation. Note that while this
1391 * function restores the active reference, it doesn't and can't actually
1392 * restore the active protection - @kn may already or be in the process of
1393 * being removed. Once kernfs_break_active_protection() is invoked, that
1394 * protection is irreversibly gone for the kernfs operation instance.
1396 * While this function may be called at any point after
1397 * kernfs_break_active_protection() is invoked, its most useful location
1398 * would be right before the enclosing kernfs operation returns.
1400 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1403 * @kn->active could be in any state; however, the increment we do
1404 * here will be undone as soon as the enclosing kernfs operation
1405 * finishes and this temporary bump can't break anything. If @kn
1406 * is alive, nothing changes. If @kn is being deactivated, the
1407 * soon-to-follow put will either finish deactivation or restore
1408 * deactivated state. If @kn is already removed, the temporary
1409 * bump is guaranteed to be gone before @kn is released.
1411 atomic_inc(&kn->active);
1412 if (kernfs_lockdep(kn))
1413 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1417 * kernfs_remove_self - remove a kernfs_node from its own method
1418 * @kn: the self kernfs_node to remove
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. This can be used to
1422 * implement a file operation which deletes itself.
1424 * For example, the "delete" file for a sysfs device directory can be
1425 * implemented by invoking kernfs_remove_self() on the "delete" file
1426 * itself. This function breaks the circular dependency of trying to
1427 * deactivate self while holding an active ref itself. It isn't necessary
1428 * to modify the usual removal path to use kernfs_remove_self(). The
1429 * "delete" implementation can simply invoke kernfs_remove_self() on self
1430 * before proceeding with the usual removal path. kernfs will ignore later
1431 * kernfs_remove() on self.
1433 * kernfs_remove_self() can be called multiple times concurrently on the
1434 * same kernfs_node. Only the first one actually performs removal and
1435 * returns %true. All others will wait until the kernfs operation which
1436 * won self-removal finishes and return %false. Note that the losers wait
1437 * for the completion of not only the winning kernfs_remove_self() but also
1438 * the whole kernfs_ops which won the arbitration. This can be used to
1439 * guarantee, for example, all concurrent writes to a "delete" file to
1440 * finish only after the whole operation is complete.
1442 bool kernfs_remove_self(struct kernfs_node *kn)
1446 mutex_lock(&kernfs_mutex);
1447 kernfs_break_active_protection(kn);
1450 * SUICIDAL is used to arbitrate among competing invocations. Only
1451 * the first one will actually perform removal. When the removal
1452 * is complete, SUICIDED is set and the active ref is restored
1453 * while holding kernfs_mutex. The ones which lost arbitration
1454 * waits for SUICDED && drained which can happen only after the
1455 * enclosing kernfs operation which executed the winning instance
1456 * of kernfs_remove_self() finished.
1458 if (!(kn->flags & KERNFS_SUICIDAL)) {
1459 kn->flags |= KERNFS_SUICIDAL;
1460 __kernfs_remove(kn);
1461 kn->flags |= KERNFS_SUICIDED;
1464 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1468 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1470 if ((kn->flags & KERNFS_SUICIDED) &&
1471 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1474 mutex_unlock(&kernfs_mutex);
1476 mutex_lock(&kernfs_mutex);
1478 finish_wait(waitq, &wait);
1479 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1484 * This must be done while holding kernfs_mutex; otherwise, waiting
1485 * for SUICIDED && deactivated could finish prematurely.
1487 kernfs_unbreak_active_protection(kn);
1489 mutex_unlock(&kernfs_mutex);
1494 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1495 * @parent: parent of the target
1496 * @name: name of the kernfs_node to remove
1497 * @ns: namespace tag of the kernfs_node to remove
1499 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1500 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1502 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1505 struct kernfs_node *kn;
1508 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1513 mutex_lock(&kernfs_mutex);
1515 kn = kernfs_find_ns(parent, name, ns);
1518 __kernfs_remove(kn);
1522 mutex_unlock(&kernfs_mutex);
1531 * kernfs_rename_ns - move and rename a kernfs_node
1533 * @new_parent: new parent to put @sd under
1534 * @new_name: new name
1535 * @new_ns: new namespace tag
1537 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1538 const char *new_name, const void *new_ns)
1540 struct kernfs_node *old_parent;
1541 const char *old_name = NULL;
1544 /* can't move or rename root */
1548 mutex_lock(&kernfs_mutex);
1551 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1552 (new_parent->flags & KERNFS_EMPTY_DIR))
1556 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1557 (strcmp(kn->name, new_name) == 0))
1558 goto out; /* nothing to rename */
1561 if (kernfs_find_ns(new_parent, new_name, new_ns))
1564 /* rename kernfs_node */
1565 if (strcmp(kn->name, new_name) != 0) {
1567 new_name = kstrdup_const(new_name, GFP_KERNEL);
1575 * Move to the appropriate place in the appropriate directories rbtree.
1577 kernfs_unlink_sibling(kn);
1578 kernfs_get(new_parent);
1580 /* rename_lock protects ->parent and ->name accessors */
1581 spin_lock_irq(&kernfs_rename_lock);
1583 old_parent = kn->parent;
1584 kn->parent = new_parent;
1588 old_name = kn->name;
1589 kn->name = new_name;
1592 spin_unlock_irq(&kernfs_rename_lock);
1594 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1595 kernfs_link_sibling(kn);
1597 kernfs_put(old_parent);
1598 kfree_const(old_name);
1602 mutex_unlock(&kernfs_mutex);
1606 /* Relationship between s_mode and the DT_xxx types */
1607 static inline unsigned char dt_type(struct kernfs_node *kn)
1609 return (kn->mode >> 12) & 15;
1612 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1614 kernfs_put(filp->private_data);
1618 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1619 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1622 int valid = kernfs_active(pos) &&
1623 pos->parent == parent && hash == pos->hash;
1628 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1629 struct rb_node *node = parent->dir.children.rb_node;
1631 pos = rb_to_kn(node);
1633 if (hash < pos->hash)
1634 node = node->rb_left;
1635 else if (hash > pos->hash)
1636 node = node->rb_right;
1641 /* Skip over entries which are dying/dead or in the wrong namespace */
1642 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1643 struct rb_node *node = rb_next(&pos->rb);
1647 pos = rb_to_kn(node);
1652 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1653 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1655 pos = kernfs_dir_pos(ns, parent, ino, pos);
1658 struct rb_node *node = rb_next(&pos->rb);
1662 pos = rb_to_kn(node);
1663 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1668 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1670 struct dentry *dentry = file->f_path.dentry;
1671 struct kernfs_node *parent = kernfs_dentry_node(dentry);
1672 struct kernfs_node *pos = file->private_data;
1673 const void *ns = NULL;
1675 if (!dir_emit_dots(file, ctx))
1677 mutex_lock(&kernfs_mutex);
1679 if (kernfs_ns_enabled(parent))
1680 ns = kernfs_info(dentry->d_sb)->ns;
1682 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1684 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1685 const char *name = pos->name;
1686 unsigned int type = dt_type(pos);
1687 int len = strlen(name);
1688 ino_t ino = pos->id.ino;
1690 ctx->pos = pos->hash;
1691 file->private_data = pos;
1694 mutex_unlock(&kernfs_mutex);
1695 if (!dir_emit(ctx, name, len, ino, type))
1697 mutex_lock(&kernfs_mutex);
1699 mutex_unlock(&kernfs_mutex);
1700 file->private_data = NULL;
1705 const struct file_operations kernfs_dir_fops = {
1706 .read = generic_read_dir,
1707 .iterate_shared = kernfs_fop_readdir,
1708 .release = kernfs_dir_fop_release,
1709 .llseek = generic_file_llseek,