2 * fs/sysfs/dir.c - sysfs core and dir operation implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007 Tejun Heo <teheo@suse.de>
8 * This file is released under the GPLv2.
10 * Please see Documentation/filesystems/sysfs.txt for more information.
16 #include <linux/mount.h>
17 #include <linux/module.h>
18 #include <linux/kobject.h>
19 #include <linux/namei.h>
20 #include <linux/idr.h>
21 #include <linux/completion.h>
22 #include <linux/mutex.h>
25 DEFINE_MUTEX(sysfs_mutex);
26 DEFINE_MUTEX(sysfs_rename_mutex);
27 DEFINE_SPINLOCK(sysfs_assoc_lock);
29 static DEFINE_SPINLOCK(sysfs_ino_lock);
30 static DEFINE_IDA(sysfs_ino_ida);
33 * sysfs_link_sibling - link sysfs_dirent into sibling list
34 * @sd: sysfs_dirent of interest
36 * Link @sd into its sibling list which starts from
37 * sd->s_parent->s_dir.children.
40 * mutex_lock(sysfs_mutex)
42 static void sysfs_link_sibling(struct sysfs_dirent *sd)
44 struct sysfs_dirent *parent_sd = sd->s_parent;
45 struct sysfs_dirent **pos;
47 BUG_ON(sd->s_sibling);
49 /* Store directory entries in order by ino. This allows
50 * readdir to properly restart without having to add a
51 * cursor into the s_dir.children list.
53 for (pos = &parent_sd->s_dir.children; *pos; pos = &(*pos)->s_sibling) {
54 if (sd->s_ino < (*pos)->s_ino)
62 * sysfs_unlink_sibling - unlink sysfs_dirent from sibling list
63 * @sd: sysfs_dirent of interest
65 * Unlink @sd from its sibling list which starts from
66 * sd->s_parent->s_dir.children.
69 * mutex_lock(sysfs_mutex)
71 static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
73 struct sysfs_dirent **pos;
75 for (pos = &sd->s_parent->s_dir.children; *pos;
76 pos = &(*pos)->s_sibling) {
86 * sysfs_get_dentry - get dentry for the given sysfs_dirent
87 * @sd: sysfs_dirent of interest
89 * Get dentry for @sd. Dentry is looked up if currently not
90 * present. This function descends from the root looking up
91 * dentry for each step.
94 * mutex_lock(sysfs_rename_mutex)
97 * Pointer to found dentry on success, ERR_PTR() value on error.
99 struct dentry *sysfs_get_dentry(struct sysfs_dirent *sd)
101 struct dentry *dentry = dget(sysfs_sb->s_root);
103 while (dentry->d_fsdata != sd) {
104 struct sysfs_dirent *cur;
105 struct dentry *parent;
107 /* find the first ancestor which hasn't been looked up */
109 while (cur->s_parent != dentry->d_fsdata)
114 mutex_lock(&parent->d_inode->i_mutex);
115 dentry = lookup_one_noperm(cur->s_name, parent);
116 mutex_unlock(&parent->d_inode->i_mutex);
126 * sysfs_get_active - get an active reference to sysfs_dirent
127 * @sd: sysfs_dirent to get an active reference to
129 * Get an active reference of @sd. This function is noop if @sd
133 * Pointer to @sd on success, NULL on failure.
135 static struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
143 v = atomic_read(&sd->s_active);
147 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
158 * sysfs_put_active - put an active reference to sysfs_dirent
159 * @sd: sysfs_dirent to put an active reference to
161 * Put an active reference to @sd. This function is noop if @sd
164 static void sysfs_put_active(struct sysfs_dirent *sd)
166 struct completion *cmpl;
172 v = atomic_dec_return(&sd->s_active);
173 if (likely(v != SD_DEACTIVATED_BIAS))
176 /* atomic_dec_return() is a mb(), we'll always see the updated
179 cmpl = (void *)sd->s_sibling;
184 * sysfs_get_active_two - get active references to sysfs_dirent and parent
185 * @sd: sysfs_dirent of interest
187 * Get active reference to @sd and its parent. Parent's active
188 * reference is grabbed first. This function is noop if @sd is
192 * Pointer to @sd on success, NULL on failure.
194 struct sysfs_dirent *sysfs_get_active_two(struct sysfs_dirent *sd)
197 if (sd->s_parent && unlikely(!sysfs_get_active(sd->s_parent)))
199 if (unlikely(!sysfs_get_active(sd))) {
200 sysfs_put_active(sd->s_parent);
208 * sysfs_put_active_two - put active references to sysfs_dirent and parent
209 * @sd: sysfs_dirent of interest
211 * Put active references to @sd and its parent. This function is
212 * noop if @sd is NULL.
214 void sysfs_put_active_two(struct sysfs_dirent *sd)
217 sysfs_put_active(sd);
218 sysfs_put_active(sd->s_parent);
223 * sysfs_deactivate - deactivate sysfs_dirent
224 * @sd: sysfs_dirent to deactivate
226 * Deny new active references and drain existing ones.
228 static void sysfs_deactivate(struct sysfs_dirent *sd)
230 DECLARE_COMPLETION_ONSTACK(wait);
233 BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
234 sd->s_sibling = (void *)&wait;
236 /* atomic_add_return() is a mb(), put_active() will always see
237 * the updated sd->s_sibling.
239 v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
241 if (v != SD_DEACTIVATED_BIAS)
242 wait_for_completion(&wait);
244 sd->s_sibling = NULL;
247 static int sysfs_alloc_ino(ino_t *pino)
252 spin_lock(&sysfs_ino_lock);
253 rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
254 spin_unlock(&sysfs_ino_lock);
257 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
266 static void sysfs_free_ino(ino_t ino)
268 spin_lock(&sysfs_ino_lock);
269 ida_remove(&sysfs_ino_ida, ino);
270 spin_unlock(&sysfs_ino_lock);
273 void release_sysfs_dirent(struct sysfs_dirent * sd)
275 struct sysfs_dirent *parent_sd;
278 /* Moving/renaming is always done while holding reference.
279 * sd->s_parent won't change beneath us.
281 parent_sd = sd->s_parent;
283 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
284 sysfs_put(sd->s_symlink.target_sd);
285 if (sysfs_type(sd) & SYSFS_COPY_NAME)
288 sysfs_free_ino(sd->s_ino);
289 kmem_cache_free(sysfs_dir_cachep, sd);
292 if (sd && atomic_dec_and_test(&sd->s_count))
296 static void sysfs_d_iput(struct dentry * dentry, struct inode * inode)
298 struct sysfs_dirent * sd = dentry->d_fsdata;
304 static struct dentry_operations sysfs_dentry_ops = {
305 .d_iput = sysfs_d_iput,
308 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
310 char *dup_name = NULL;
311 struct sysfs_dirent *sd;
313 if (type & SYSFS_COPY_NAME) {
314 name = dup_name = kstrdup(name, GFP_KERNEL);
319 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
323 if (sysfs_alloc_ino(&sd->s_ino))
326 atomic_set(&sd->s_count, 1);
327 atomic_set(&sd->s_active, 0);
336 kmem_cache_free(sysfs_dir_cachep, sd);
342 static int sysfs_ilookup_test(struct inode *inode, void *arg)
344 struct sysfs_dirent *sd = arg;
345 return inode->i_ino == sd->s_ino;
349 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
350 * @acxt: pointer to sysfs_addrm_cxt to be used
351 * @parent_sd: parent sysfs_dirent
353 * This function is called when the caller is about to add or
354 * remove sysfs_dirent under @parent_sd. This function acquires
355 * sysfs_mutex, grabs inode for @parent_sd if available and lock
356 * i_mutex of it. @acxt is used to keep and pass context to
357 * other addrm functions.
360 * Kernel thread context (may sleep). sysfs_mutex is locked on
361 * return. i_mutex of parent inode is locked on return if
364 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
365 struct sysfs_dirent *parent_sd)
369 memset(acxt, 0, sizeof(*acxt));
370 acxt->parent_sd = parent_sd;
372 /* Lookup parent inode. inode initialization and I_NEW
373 * clearing are protected by sysfs_mutex. By grabbing it and
374 * looking up with _nowait variant, inode state can be
375 * determined reliably.
377 mutex_lock(&sysfs_mutex);
379 inode = ilookup5_nowait(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test,
382 if (inode && !(inode->i_state & I_NEW)) {
383 /* parent inode available */
384 acxt->parent_inode = inode;
386 /* sysfs_mutex is below i_mutex in lock hierarchy.
387 * First, trylock i_mutex. If fails, unlock
388 * sysfs_mutex and lock them in order.
390 if (!mutex_trylock(&inode->i_mutex)) {
391 mutex_unlock(&sysfs_mutex);
392 mutex_lock(&inode->i_mutex);
393 mutex_lock(&sysfs_mutex);
400 * sysfs_add_one - add sysfs_dirent to parent
401 * @acxt: addrm context to use
402 * @sd: sysfs_dirent to be added
404 * Get @acxt->parent_sd and set sd->s_parent to it and increment
405 * nlink of parent inode if @sd is a directory and link into the
406 * children list of the parent.
408 * This function should be called between calls to
409 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
410 * passed the same @acxt as passed to sysfs_addrm_start().
413 * Determined by sysfs_addrm_start().
416 * 0 on success, -EEXIST if entry with the given name already
419 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
421 if (sysfs_find_dirent(acxt->parent_sd, sd->s_name)) {
422 printk(KERN_WARNING "sysfs: duplicate filename '%s' "
423 "can not be created\n", sd->s_name);
428 sd->s_parent = sysfs_get(acxt->parent_sd);
430 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
431 inc_nlink(acxt->parent_inode);
435 sysfs_link_sibling(sd);
441 * sysfs_remove_one - remove sysfs_dirent from parent
442 * @acxt: addrm context to use
443 * @sd: sysfs_dirent to be removed
445 * Mark @sd removed and drop nlink of parent inode if @sd is a
446 * directory. @sd is unlinked from the children list.
448 * This function should be called between calls to
449 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
450 * passed the same @acxt as passed to sysfs_addrm_start().
453 * Determined by sysfs_addrm_start().
455 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
457 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
459 sysfs_unlink_sibling(sd);
461 sd->s_flags |= SYSFS_FLAG_REMOVED;
462 sd->s_sibling = acxt->removed;
465 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
466 drop_nlink(acxt->parent_inode);
472 * sysfs_drop_dentry - drop dentry for the specified sysfs_dirent
473 * @sd: target sysfs_dirent
475 * Drop dentry for @sd. @sd must have been unlinked from its
476 * parent on entry to this function such that it can't be looked
479 static void sysfs_drop_dentry(struct sysfs_dirent *sd)
482 struct dentry *dentry;
484 inode = ilookup(sysfs_sb, sd->s_ino);
488 /* Drop any existing dentries associated with sd.
490 * For the dentry to be properly freed we need to grab a
491 * reference to the dentry under the dcache lock, unhash it,
492 * and then put it. The playing with the dentry count allows
493 * dput to immediately free the dentry if it is not in use.
496 spin_lock(&dcache_lock);
497 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
498 if (d_unhashed(dentry))
501 spin_lock(&dentry->d_lock);
503 spin_unlock(&dentry->d_lock);
504 spin_unlock(&dcache_lock);
508 spin_unlock(&dcache_lock);
510 /* adjust nlink and update timestamp */
511 mutex_lock(&inode->i_mutex);
513 inode->i_ctime = CURRENT_TIME;
515 if (sysfs_type(sd) == SYSFS_DIR)
518 mutex_unlock(&inode->i_mutex);
524 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
525 * @acxt: addrm context to finish up
527 * Finish up sysfs_dirent add/remove. Resources acquired by
528 * sysfs_addrm_start() are released and removed sysfs_dirents are
529 * cleaned up. Timestamps on the parent inode are updated.
532 * All mutexes acquired by sysfs_addrm_start() are released.
534 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
536 /* release resources acquired by sysfs_addrm_start() */
537 mutex_unlock(&sysfs_mutex);
538 if (acxt->parent_inode) {
539 struct inode *inode = acxt->parent_inode;
541 /* if added/removed, update timestamps on the parent */
543 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
545 mutex_unlock(&inode->i_mutex);
549 /* kill removed sysfs_dirents */
550 while (acxt->removed) {
551 struct sysfs_dirent *sd = acxt->removed;
553 acxt->removed = sd->s_sibling;
554 sd->s_sibling = NULL;
556 sysfs_drop_dentry(sd);
557 sysfs_deactivate(sd);
563 * sysfs_find_dirent - find sysfs_dirent with the given name
564 * @parent_sd: sysfs_dirent to search under
565 * @name: name to look for
567 * Look for sysfs_dirent with name @name under @parent_sd.
570 * mutex_lock(sysfs_mutex)
573 * Pointer to sysfs_dirent if found, NULL if not.
575 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
576 const unsigned char *name)
578 struct sysfs_dirent *sd;
580 for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
581 if (!strcmp(sd->s_name, name))
587 * sysfs_get_dirent - find and get sysfs_dirent with the given name
588 * @parent_sd: sysfs_dirent to search under
589 * @name: name to look for
591 * Look for sysfs_dirent with name @name under @parent_sd and get
595 * Kernel thread context (may sleep). Grabs sysfs_mutex.
598 * Pointer to sysfs_dirent if found, NULL if not.
600 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
601 const unsigned char *name)
603 struct sysfs_dirent *sd;
605 mutex_lock(&sysfs_mutex);
606 sd = sysfs_find_dirent(parent_sd, name);
608 mutex_unlock(&sysfs_mutex);
613 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
614 const char *name, struct sysfs_dirent **p_sd)
616 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
617 struct sysfs_addrm_cxt acxt;
618 struct sysfs_dirent *sd;
622 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
625 sd->s_dir.kobj = kobj;
628 sysfs_addrm_start(&acxt, parent_sd);
629 rc = sysfs_add_one(&acxt, sd);
630 sysfs_addrm_finish(&acxt);
640 int sysfs_create_subdir(struct kobject *kobj, const char *name,
641 struct sysfs_dirent **p_sd)
643 return create_dir(kobj, kobj->sd, name, p_sd);
647 * sysfs_create_dir - create a directory for an object.
648 * @kobj: object we're creating directory for.
650 int sysfs_create_dir(struct kobject * kobj)
652 struct sysfs_dirent *parent_sd, *sd;
658 parent_sd = kobj->parent->sd;
660 parent_sd = &sysfs_root;
662 error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
668 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
669 struct nameidata *nd)
671 struct dentry *ret = NULL;
672 struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
673 struct sysfs_dirent *sd;
676 mutex_lock(&sysfs_mutex);
678 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
682 ret = ERR_PTR(-ENOENT);
686 /* attach dentry and inode */
687 inode = sysfs_get_inode(sd);
689 ret = ERR_PTR(-ENOMEM);
693 /* instantiate and hash dentry */
694 dentry->d_op = &sysfs_dentry_ops;
695 dentry->d_fsdata = sysfs_get(sd);
696 d_instantiate(dentry, inode);
700 mutex_unlock(&sysfs_mutex);
704 const struct inode_operations sysfs_dir_inode_operations = {
705 .lookup = sysfs_lookup,
706 .setattr = sysfs_setattr,
709 static void remove_dir(struct sysfs_dirent *sd)
711 struct sysfs_addrm_cxt acxt;
713 sysfs_addrm_start(&acxt, sd->s_parent);
714 sysfs_remove_one(&acxt, sd);
715 sysfs_addrm_finish(&acxt);
718 void sysfs_remove_subdir(struct sysfs_dirent *sd)
724 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
726 struct sysfs_addrm_cxt acxt;
727 struct sysfs_dirent **pos;
732 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
733 sysfs_addrm_start(&acxt, dir_sd);
734 pos = &dir_sd->s_dir.children;
736 struct sysfs_dirent *sd = *pos;
738 if (sysfs_type(sd) != SYSFS_DIR)
739 sysfs_remove_one(&acxt, sd);
741 pos = &(*pos)->s_sibling;
743 sysfs_addrm_finish(&acxt);
749 * sysfs_remove_dir - remove an object's directory.
752 * The only thing special about this is that we remove any files in
753 * the directory before we remove the directory, and we've inlined
754 * what used to be sysfs_rmdir() below, instead of calling separately.
757 void sysfs_remove_dir(struct kobject * kobj)
759 struct sysfs_dirent *sd = kobj->sd;
761 spin_lock(&sysfs_assoc_lock);
763 spin_unlock(&sysfs_assoc_lock);
765 __sysfs_remove_dir(sd);
768 int sysfs_rename_dir(struct kobject * kobj, const char *new_name)
770 struct sysfs_dirent *sd = kobj->sd;
771 struct dentry *parent = NULL;
772 struct dentry *old_dentry = NULL, *new_dentry = NULL;
773 const char *dup_name = NULL;
776 mutex_lock(&sysfs_rename_mutex);
779 if (strcmp(sd->s_name, new_name) == 0)
780 goto out; /* nothing to rename */
782 /* get the original dentry */
783 old_dentry = sysfs_get_dentry(sd);
784 if (IS_ERR(old_dentry)) {
785 error = PTR_ERR(old_dentry);
790 parent = old_dentry->d_parent;
792 /* lock parent and get dentry for new name */
793 mutex_lock(&parent->d_inode->i_mutex);
794 mutex_lock(&sysfs_mutex);
797 if (sysfs_find_dirent(sd->s_parent, new_name))
801 new_dentry = d_alloc_name(parent, new_name);
805 /* rename kobject and sysfs_dirent */
807 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
811 error = kobject_set_name(kobj, "%s", new_name);
815 dup_name = sd->s_name;
816 sd->s_name = new_name;
819 d_add(new_dentry, NULL);
820 d_move(old_dentry, new_dentry);
824 mutex_unlock(&sysfs_mutex);
825 mutex_unlock(&parent->d_inode->i_mutex);
830 mutex_unlock(&sysfs_rename_mutex);
834 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
836 struct sysfs_dirent *sd = kobj->sd;
837 struct sysfs_dirent *new_parent_sd;
838 struct dentry *old_parent, *new_parent = NULL;
839 struct dentry *old_dentry = NULL, *new_dentry = NULL;
842 mutex_lock(&sysfs_rename_mutex);
843 BUG_ON(!sd->s_parent);
844 new_parent_sd = new_parent_kobj->sd ? new_parent_kobj->sd : &sysfs_root;
847 if (sd->s_parent == new_parent_sd)
848 goto out; /* nothing to move */
851 old_dentry = sysfs_get_dentry(sd);
852 if (IS_ERR(old_dentry)) {
853 error = PTR_ERR(old_dentry);
857 old_parent = old_dentry->d_parent;
859 new_parent = sysfs_get_dentry(new_parent_sd);
860 if (IS_ERR(new_parent)) {
861 error = PTR_ERR(new_parent);
867 mutex_lock(&old_parent->d_inode->i_mutex);
868 if (!mutex_trylock(&new_parent->d_inode->i_mutex)) {
869 mutex_unlock(&old_parent->d_inode->i_mutex);
872 mutex_lock(&sysfs_mutex);
875 if (sysfs_find_dirent(new_parent_sd, sd->s_name))
879 new_dentry = d_alloc_name(new_parent, sd->s_name);
884 d_add(new_dentry, NULL);
885 d_move(old_dentry, new_dentry);
887 /* Remove from old parent's list and insert into new parent's list. */
888 sysfs_unlink_sibling(sd);
889 sysfs_get(new_parent_sd);
890 sysfs_put(sd->s_parent);
891 sd->s_parent = new_parent_sd;
892 sysfs_link_sibling(sd);
895 mutex_unlock(&sysfs_mutex);
896 mutex_unlock(&new_parent->d_inode->i_mutex);
897 mutex_unlock(&old_parent->d_inode->i_mutex);
902 mutex_unlock(&sysfs_rename_mutex);
906 /* Relationship between s_mode and the DT_xxx types */
907 static inline unsigned char dt_type(struct sysfs_dirent *sd)
909 return (sd->s_mode >> 12) & 15;
912 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
914 struct dentry *dentry = filp->f_path.dentry;
915 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
916 struct sysfs_dirent *pos;
919 if (filp->f_pos == 0) {
920 ino = parent_sd->s_ino;
921 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
924 if (filp->f_pos == 1) {
925 if (parent_sd->s_parent)
926 ino = parent_sd->s_parent->s_ino;
928 ino = parent_sd->s_ino;
929 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
932 if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
933 mutex_lock(&sysfs_mutex);
935 /* Skip the dentries we have already reported */
936 pos = parent_sd->s_dir.children;
937 while (pos && (filp->f_pos > pos->s_ino))
938 pos = pos->s_sibling;
940 for ( ; pos; pos = pos->s_sibling) {
946 filp->f_pos = ino = pos->s_ino;
948 if (filldir(dirent, name, len, filp->f_pos, ino,
953 filp->f_pos = INT_MAX;
954 mutex_unlock(&sysfs_mutex);
960 const struct file_operations sysfs_dir_operations = {
961 .read = generic_read_dir,
962 .readdir = sysfs_readdir,