2 * dir.c - Operations for sysfs directories.
8 #include <linux/mount.h>
9 #include <linux/module.h>
10 #include <linux/kobject.h>
11 #include <linux/namei.h>
12 #include <linux/idr.h>
13 #include <linux/completion.h>
14 #include <linux/mutex.h>
17 DEFINE_MUTEX(sysfs_mutex);
18 DEFINE_MUTEX(sysfs_rename_mutex);
19 spinlock_t sysfs_assoc_lock = SPIN_LOCK_UNLOCKED;
21 static spinlock_t sysfs_ino_lock = SPIN_LOCK_UNLOCKED;
22 static DEFINE_IDA(sysfs_ino_ida);
25 * sysfs_link_sibling - link sysfs_dirent into sibling list
26 * @sd: sysfs_dirent of interest
28 * Link @sd into its sibling list which starts from
29 * sd->s_parent->s_children.
32 * mutex_lock(sysfs_mutex)
34 static void sysfs_link_sibling(struct sysfs_dirent *sd)
36 struct sysfs_dirent *parent_sd = sd->s_parent;
37 struct sysfs_dirent **pos;
39 BUG_ON(sd->s_sibling);
41 /* Store directory entries in order by ino. This allows
42 * readdir to properly restart without having to add a
43 * cursor into the s_children list.
45 for (pos = &parent_sd->s_children; *pos; pos = &(*pos)->s_sibling) {
46 if (sd->s_ino < (*pos)->s_ino)
54 * sysfs_unlink_sibling - unlink sysfs_dirent from sibling list
55 * @sd: sysfs_dirent of interest
57 * Unlink @sd from its sibling list which starts from
58 * sd->s_parent->s_children.
61 * mutex_lock(sysfs_mutex)
63 static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
65 struct sysfs_dirent **pos;
67 for (pos = &sd->s_parent->s_children; *pos; pos = &(*pos)->s_sibling) {
77 * sysfs_get_dentry - get dentry for the given sysfs_dirent
78 * @sd: sysfs_dirent of interest
80 * Get dentry for @sd. Dentry is looked up if currently not
81 * present. This function descends from the root looking up
82 * dentry for each step.
85 * mutex_lock(sysfs_rename_mutex)
88 * Pointer to found dentry on success, ERR_PTR() value on error.
90 struct dentry *sysfs_get_dentry(struct sysfs_dirent *sd)
92 struct dentry *dentry = dget(sysfs_sb->s_root);
94 while (dentry->d_fsdata != sd) {
95 struct sysfs_dirent *cur;
96 struct dentry *parent;
98 /* find the first ancestor which hasn't been looked up */
100 while (cur->s_parent != dentry->d_fsdata)
105 mutex_lock(&parent->d_inode->i_mutex);
106 dentry = lookup_one_len_kern(cur->s_name, parent,
107 strlen(cur->s_name));
108 mutex_unlock(&parent->d_inode->i_mutex);
118 * sysfs_get_active - get an active reference to sysfs_dirent
119 * @sd: sysfs_dirent to get an active reference to
121 * Get an active reference of @sd. This function is noop if @sd
125 * Pointer to @sd on success, NULL on failure.
127 struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
135 v = atomic_read(&sd->s_active);
139 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
150 * sysfs_put_active - put an active reference to sysfs_dirent
151 * @sd: sysfs_dirent to put an active reference to
153 * Put an active reference to @sd. This function is noop if @sd
156 void sysfs_put_active(struct sysfs_dirent *sd)
158 struct completion *cmpl;
164 v = atomic_dec_return(&sd->s_active);
165 if (likely(v != SD_DEACTIVATED_BIAS))
168 /* atomic_dec_return() is a mb(), we'll always see the updated
171 cmpl = (void *)sd->s_sibling;
176 * sysfs_get_active_two - get active references to sysfs_dirent and parent
177 * @sd: sysfs_dirent of interest
179 * Get active reference to @sd and its parent. Parent's active
180 * reference is grabbed first. This function is noop if @sd is
184 * Pointer to @sd on success, NULL on failure.
186 struct sysfs_dirent *sysfs_get_active_two(struct sysfs_dirent *sd)
189 if (sd->s_parent && unlikely(!sysfs_get_active(sd->s_parent)))
191 if (unlikely(!sysfs_get_active(sd))) {
192 sysfs_put_active(sd->s_parent);
200 * sysfs_put_active_two - put active references to sysfs_dirent and parent
201 * @sd: sysfs_dirent of interest
203 * Put active references to @sd and its parent. This function is
204 * noop if @sd is NULL.
206 void sysfs_put_active_two(struct sysfs_dirent *sd)
209 sysfs_put_active(sd);
210 sysfs_put_active(sd->s_parent);
215 * sysfs_deactivate - deactivate sysfs_dirent
216 * @sd: sysfs_dirent to deactivate
218 * Deny new active references and drain existing ones.
220 static void sysfs_deactivate(struct sysfs_dirent *sd)
222 DECLARE_COMPLETION_ONSTACK(wait);
225 BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
226 sd->s_sibling = (void *)&wait;
228 /* atomic_add_return() is a mb(), put_active() will always see
229 * the updated sd->s_sibling.
231 v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
233 if (v != SD_DEACTIVATED_BIAS)
234 wait_for_completion(&wait);
236 sd->s_sibling = NULL;
239 static int sysfs_alloc_ino(ino_t *pino)
244 spin_lock(&sysfs_ino_lock);
245 rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
246 spin_unlock(&sysfs_ino_lock);
249 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
258 static void sysfs_free_ino(ino_t ino)
260 spin_lock(&sysfs_ino_lock);
261 ida_remove(&sysfs_ino_ida, ino);
262 spin_unlock(&sysfs_ino_lock);
265 void release_sysfs_dirent(struct sysfs_dirent * sd)
267 struct sysfs_dirent *parent_sd;
270 /* Moving/renaming is always done while holding reference.
271 * sd->s_parent won't change beneath us.
273 parent_sd = sd->s_parent;
275 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
276 sysfs_put(sd->s_elem.symlink.target_sd);
277 if (sysfs_type(sd) & SYSFS_COPY_NAME)
280 sysfs_free_ino(sd->s_ino);
281 kmem_cache_free(sysfs_dir_cachep, sd);
284 if (sd && atomic_dec_and_test(&sd->s_count))
288 static void sysfs_d_iput(struct dentry * dentry, struct inode * inode)
290 struct sysfs_dirent * sd = dentry->d_fsdata;
296 static struct dentry_operations sysfs_dentry_ops = {
297 .d_iput = sysfs_d_iput,
300 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
302 char *dup_name = NULL;
303 struct sysfs_dirent *sd;
305 if (type & SYSFS_COPY_NAME) {
306 name = dup_name = kstrdup(name, GFP_KERNEL);
311 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
315 if (sysfs_alloc_ino(&sd->s_ino))
318 atomic_set(&sd->s_count, 1);
319 atomic_set(&sd->s_active, 0);
320 atomic_set(&sd->s_event, 1);
329 kmem_cache_free(sysfs_dir_cachep, sd);
336 * sysfs_attach_dentry - associate sysfs_dirent with dentry
337 * @sd: target sysfs_dirent
338 * @dentry: dentry to associate
341 * mutex_lock(sysfs_mutex)
343 static void sysfs_attach_dentry(struct sysfs_dirent *sd, struct dentry *dentry)
345 dentry->d_op = &sysfs_dentry_ops;
346 dentry->d_fsdata = sysfs_get(sd);
350 static int sysfs_ilookup_test(struct inode *inode, void *arg)
352 struct sysfs_dirent *sd = arg;
353 return inode->i_ino == sd->s_ino;
357 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
358 * @acxt: pointer to sysfs_addrm_cxt to be used
359 * @parent_sd: parent sysfs_dirent
361 * This function is called when the caller is about to add or
362 * remove sysfs_dirent under @parent_sd. This function acquires
363 * sysfs_mutex, grabs inode for @parent_sd if available and lock
364 * i_mutex of it. @acxt is used to keep and pass context to
365 * other addrm functions.
368 * Kernel thread context (may sleep). sysfs_mutex is locked on
369 * return. i_mutex of parent inode is locked on return if
372 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
373 struct sysfs_dirent *parent_sd)
377 memset(acxt, 0, sizeof(*acxt));
378 acxt->parent_sd = parent_sd;
380 /* Lookup parent inode. inode initialization and I_NEW
381 * clearing are protected by sysfs_mutex. By grabbing it and
382 * looking up with _nowait variant, inode state can be
383 * determined reliably.
385 mutex_lock(&sysfs_mutex);
387 inode = ilookup5_nowait(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test,
390 if (inode && !(inode->i_state & I_NEW)) {
391 /* parent inode available */
392 acxt->parent_inode = inode;
394 /* sysfs_mutex is below i_mutex in lock hierarchy.
395 * First, trylock i_mutex. If fails, unlock
396 * sysfs_mutex and lock them in order.
398 if (!mutex_trylock(&inode->i_mutex)) {
399 mutex_unlock(&sysfs_mutex);
400 mutex_lock(&inode->i_mutex);
401 mutex_lock(&sysfs_mutex);
408 * sysfs_add_one - add sysfs_dirent to parent
409 * @acxt: addrm context to use
410 * @sd: sysfs_dirent to be added
412 * Get @acxt->parent_sd and set sd->s_parent to it and increment
413 * nlink of parent inode if @sd is a directory and link into the
414 * children list of the parent.
416 * This function should be called between calls to
417 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
418 * passed the same @acxt as passed to sysfs_addrm_start().
421 * Determined by sysfs_addrm_start().
424 * 0 on success, -EEXIST if entry with the given name already
427 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
429 if (sysfs_find_dirent(acxt->parent_sd, sd->s_name)) {
430 printk(KERN_WARNING "sysfs: duplicate filename '%s' "
431 "can not be created\n", sd->s_name);
436 sd->s_parent = sysfs_get(acxt->parent_sd);
438 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
439 inc_nlink(acxt->parent_inode);
443 sysfs_link_sibling(sd);
449 * sysfs_remove_one - remove sysfs_dirent from parent
450 * @acxt: addrm context to use
451 * @sd: sysfs_dirent to be added
453 * Mark @sd removed and drop nlink of parent inode if @sd is a
454 * directory. @sd is unlinked from the children list.
456 * This function should be called between calls to
457 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
458 * passed the same @acxt as passed to sysfs_addrm_start().
461 * Determined by sysfs_addrm_start().
463 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
465 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
467 sysfs_unlink_sibling(sd);
469 sd->s_flags |= SYSFS_FLAG_REMOVED;
470 sd->s_sibling = acxt->removed;
473 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
474 drop_nlink(acxt->parent_inode);
480 * sysfs_drop_dentry - drop dentry for the specified sysfs_dirent
481 * @sd: target sysfs_dirent
483 * Drop dentry for @sd. @sd must have been unlinked from its
484 * parent on entry to this function such that it can't be looked
487 static void sysfs_drop_dentry(struct sysfs_dirent *sd)
490 struct dentry *dentry;
492 inode = ilookup(sysfs_sb, sd->s_ino);
496 /* Drop any existing dentries associated with sd.
498 * For the dentry to be properly freed we need to grab a
499 * reference to the dentry under the dcache lock, unhash it,
500 * and then put it. The playing with the dentry count allows
501 * dput to immediately free the dentry if it is not in use.
504 spin_lock(&dcache_lock);
505 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
506 if (d_unhashed(dentry))
509 spin_lock(&dentry->d_lock);
511 spin_unlock(&dentry->d_lock);
512 spin_unlock(&dcache_lock);
516 spin_unlock(&dcache_lock);
518 /* adjust nlink and update timestamp */
519 mutex_lock(&inode->i_mutex);
521 inode->i_ctime = CURRENT_TIME;
523 if (sysfs_type(sd) == SYSFS_DIR)
526 mutex_unlock(&inode->i_mutex);
532 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
533 * @acxt: addrm context to finish up
535 * Finish up sysfs_dirent add/remove. Resources acquired by
536 * sysfs_addrm_start() are released and removed sysfs_dirents are
537 * cleaned up. Timestamps on the parent inode are updated.
540 * All mutexes acquired by sysfs_addrm_start() are released.
542 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
544 /* release resources acquired by sysfs_addrm_start() */
545 mutex_unlock(&sysfs_mutex);
546 if (acxt->parent_inode) {
547 struct inode *inode = acxt->parent_inode;
549 /* if added/removed, update timestamps on the parent */
551 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
553 mutex_unlock(&inode->i_mutex);
557 /* kill removed sysfs_dirents */
558 while (acxt->removed) {
559 struct sysfs_dirent *sd = acxt->removed;
561 acxt->removed = sd->s_sibling;
562 sd->s_sibling = NULL;
564 sysfs_drop_dentry(sd);
565 sysfs_deactivate(sd);
571 * sysfs_find_dirent - find sysfs_dirent with the given name
572 * @parent_sd: sysfs_dirent to search under
573 * @name: name to look for
575 * Look for sysfs_dirent with name @name under @parent_sd.
578 * mutex_lock(sysfs_mutex)
581 * Pointer to sysfs_dirent if found, NULL if not.
583 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
584 const unsigned char *name)
586 struct sysfs_dirent *sd;
588 for (sd = parent_sd->s_children; sd; sd = sd->s_sibling)
589 if (!strcmp(sd->s_name, name))
595 * sysfs_get_dirent - find and get sysfs_dirent with the given name
596 * @parent_sd: sysfs_dirent to search under
597 * @name: name to look for
599 * Look for sysfs_dirent with name @name under @parent_sd and get
603 * Kernel thread context (may sleep). Grabs sysfs_mutex.
606 * Pointer to sysfs_dirent if found, NULL if not.
608 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
609 const unsigned char *name)
611 struct sysfs_dirent *sd;
613 mutex_lock(&sysfs_mutex);
614 sd = sysfs_find_dirent(parent_sd, name);
616 mutex_unlock(&sysfs_mutex);
621 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
622 const char *name, struct sysfs_dirent **p_sd)
624 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
625 struct sysfs_addrm_cxt acxt;
626 struct sysfs_dirent *sd;
630 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
633 sd->s_elem.dir.kobj = kobj;
636 sysfs_addrm_start(&acxt, parent_sd);
637 rc = sysfs_add_one(&acxt, sd);
638 sysfs_addrm_finish(&acxt);
648 int sysfs_create_subdir(struct kobject *kobj, const char *name,
649 struct sysfs_dirent **p_sd)
651 return create_dir(kobj, kobj->sd, name, p_sd);
655 * sysfs_create_dir - create a directory for an object.
656 * @kobj: object we're creating directory for.
658 int sysfs_create_dir(struct kobject * kobj)
660 struct sysfs_dirent *parent_sd, *sd;
666 parent_sd = kobj->parent->sd;
668 parent_sd = &sysfs_root;
670 error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
676 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
677 struct nameidata *nd)
679 struct dentry *ret = NULL;
680 struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
681 struct sysfs_dirent *sd;
684 mutex_lock(&sysfs_mutex);
686 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
692 /* attach dentry and inode */
693 inode = sysfs_get_inode(sd);
695 ret = ERR_PTR(-ENOMEM);
699 d_instantiate(dentry, inode);
700 sysfs_attach_dentry(sd, dentry);
703 mutex_unlock(&sysfs_mutex);
707 const struct inode_operations sysfs_dir_inode_operations = {
708 .lookup = sysfs_lookup,
709 .setattr = sysfs_setattr,
712 static void remove_dir(struct sysfs_dirent *sd)
714 struct sysfs_addrm_cxt acxt;
716 sysfs_addrm_start(&acxt, sd->s_parent);
717 sysfs_remove_one(&acxt, sd);
718 sysfs_addrm_finish(&acxt);
721 void sysfs_remove_subdir(struct sysfs_dirent *sd)
727 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
729 struct sysfs_addrm_cxt acxt;
730 struct sysfs_dirent **pos;
735 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
736 sysfs_addrm_start(&acxt, dir_sd);
737 pos = &dir_sd->s_children;
739 struct sysfs_dirent *sd = *pos;
741 if (sysfs_type(sd) != SYSFS_DIR)
742 sysfs_remove_one(&acxt, sd);
744 pos = &(*pos)->s_sibling;
746 sysfs_addrm_finish(&acxt);
752 * sysfs_remove_dir - remove an object's directory.
755 * The only thing special about this is that we remove any files in
756 * the directory before we remove the directory, and we've inlined
757 * what used to be sysfs_rmdir() below, instead of calling separately.
760 void sysfs_remove_dir(struct kobject * kobj)
762 struct sysfs_dirent *sd = kobj->sd;
764 spin_lock(&sysfs_assoc_lock);
766 spin_unlock(&sysfs_assoc_lock);
768 __sysfs_remove_dir(sd);
771 int sysfs_rename_dir(struct kobject * kobj, const char *new_name)
773 struct sysfs_dirent *sd = kobj->sd;
774 struct dentry *parent = NULL;
775 struct dentry *old_dentry = NULL, *new_dentry = NULL;
776 const char *dup_name = NULL;
779 mutex_lock(&sysfs_rename_mutex);
782 if (strcmp(sd->s_name, new_name) == 0)
783 goto out; /* nothing to rename */
785 /* get the original dentry */
786 old_dentry = sysfs_get_dentry(sd);
787 if (IS_ERR(old_dentry)) {
788 error = PTR_ERR(old_dentry);
792 parent = old_dentry->d_parent;
794 /* lock parent and get dentry for new name */
795 mutex_lock(&parent->d_inode->i_mutex);
796 mutex_lock(&sysfs_mutex);
799 if (sysfs_find_dirent(sd->s_parent, new_name))
803 new_dentry = d_alloc_name(parent, new_name);
807 /* rename kobject and sysfs_dirent */
809 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
813 error = kobject_set_name(kobj, "%s", new_name);
817 dup_name = sd->s_name;
818 sd->s_name = new_name;
821 d_add(new_dentry, NULL);
822 d_move(old_dentry, new_dentry);
826 mutex_unlock(&sysfs_mutex);
827 mutex_unlock(&parent->d_inode->i_mutex);
832 mutex_unlock(&sysfs_rename_mutex);
836 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
838 struct sysfs_dirent *sd = kobj->sd;
839 struct sysfs_dirent *new_parent_sd;
840 struct dentry *old_parent, *new_parent = NULL;
841 struct dentry *old_dentry = NULL, *new_dentry = NULL;
844 mutex_lock(&sysfs_rename_mutex);
845 BUG_ON(!sd->s_parent);
846 new_parent_sd = new_parent_kobj->sd ? new_parent_kobj->sd : &sysfs_root;
849 if (sd->s_parent == new_parent_sd)
850 goto out; /* nothing to move */
853 old_dentry = sysfs_get_dentry(sd);
854 if (IS_ERR(old_dentry)) {
855 error = PTR_ERR(old_dentry);
858 old_parent = old_dentry->d_parent;
860 new_parent = sysfs_get_dentry(new_parent_sd);
861 if (IS_ERR(new_parent)) {
862 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);
888 /* Remove from old parent's list and insert into new parent's list. */
889 sysfs_unlink_sibling(sd);
890 sysfs_get(new_parent_sd);
891 sysfs_put(sd->s_parent);
892 sd->s_parent = new_parent_sd;
893 sysfs_link_sibling(sd);
896 mutex_unlock(&sysfs_mutex);
897 mutex_unlock(&new_parent->d_inode->i_mutex);
898 mutex_unlock(&old_parent->d_inode->i_mutex);
903 mutex_unlock(&sysfs_rename_mutex);
907 /* Relationship between s_mode and the DT_xxx types */
908 static inline unsigned char dt_type(struct sysfs_dirent *sd)
910 return (sd->s_mode >> 12) & 15;
913 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
915 struct dentry *dentry = filp->f_path.dentry;
916 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
917 struct sysfs_dirent *pos;
920 if (filp->f_pos == 0) {
921 ino = parent_sd->s_ino;
922 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
925 if (filp->f_pos == 1) {
926 if (parent_sd->s_parent)
927 ino = parent_sd->s_parent->s_ino;
929 ino = parent_sd->s_ino;
930 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
933 if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
934 mutex_lock(&sysfs_mutex);
936 /* Skip the dentries we have already reported */
937 pos = parent_sd->s_children;
938 while (pos && (filp->f_pos > pos->s_ino))
939 pos = pos->s_sibling;
941 for ( ; pos; pos = pos->s_sibling) {
947 filp->f_pos = ino = pos->s_ino;
949 if (filldir(dirent, name, len, filp->f_pos, ino,
954 filp->f_pos = INT_MAX;
955 mutex_unlock(&sysfs_mutex);
961 const struct file_operations sysfs_dir_operations = {
962 .read = generic_read_dir,
963 .readdir = sysfs_readdir,