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_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);
335 static int sysfs_ilookup_test(struct inode *inode, void *arg)
337 struct sysfs_dirent *sd = arg;
338 return inode->i_ino == sd->s_ino;
342 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
343 * @acxt: pointer to sysfs_addrm_cxt to be used
344 * @parent_sd: parent sysfs_dirent
346 * This function is called when the caller is about to add or
347 * remove sysfs_dirent under @parent_sd. This function acquires
348 * sysfs_mutex, grabs inode for @parent_sd if available and lock
349 * i_mutex of it. @acxt is used to keep and pass context to
350 * other addrm functions.
353 * Kernel thread context (may sleep). sysfs_mutex is locked on
354 * return. i_mutex of parent inode is locked on return if
357 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
358 struct sysfs_dirent *parent_sd)
362 memset(acxt, 0, sizeof(*acxt));
363 acxt->parent_sd = parent_sd;
365 /* Lookup parent inode. inode initialization and I_NEW
366 * clearing are protected by sysfs_mutex. By grabbing it and
367 * looking up with _nowait variant, inode state can be
368 * determined reliably.
370 mutex_lock(&sysfs_mutex);
372 inode = ilookup5_nowait(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test,
375 if (inode && !(inode->i_state & I_NEW)) {
376 /* parent inode available */
377 acxt->parent_inode = inode;
379 /* sysfs_mutex is below i_mutex in lock hierarchy.
380 * First, trylock i_mutex. If fails, unlock
381 * sysfs_mutex and lock them in order.
383 if (!mutex_trylock(&inode->i_mutex)) {
384 mutex_unlock(&sysfs_mutex);
385 mutex_lock(&inode->i_mutex);
386 mutex_lock(&sysfs_mutex);
393 * sysfs_add_one - add sysfs_dirent to parent
394 * @acxt: addrm context to use
395 * @sd: sysfs_dirent to be added
397 * Get @acxt->parent_sd and set sd->s_parent to it and increment
398 * nlink of parent inode if @sd is a directory and link into the
399 * children list of the parent.
401 * This function should be called between calls to
402 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
403 * passed the same @acxt as passed to sysfs_addrm_start().
406 * Determined by sysfs_addrm_start().
409 * 0 on success, -EEXIST if entry with the given name already
412 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
414 if (sysfs_find_dirent(acxt->parent_sd, sd->s_name)) {
415 printk(KERN_WARNING "sysfs: duplicate filename '%s' "
416 "can not be created\n", sd->s_name);
421 sd->s_parent = sysfs_get(acxt->parent_sd);
423 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
424 inc_nlink(acxt->parent_inode);
428 sysfs_link_sibling(sd);
434 * sysfs_remove_one - remove sysfs_dirent from parent
435 * @acxt: addrm context to use
436 * @sd: sysfs_dirent to be added
438 * Mark @sd removed and drop nlink of parent inode if @sd is a
439 * directory. @sd is unlinked from the children list.
441 * This function should be called between calls to
442 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
443 * passed the same @acxt as passed to sysfs_addrm_start().
446 * Determined by sysfs_addrm_start().
448 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
450 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
452 sysfs_unlink_sibling(sd);
454 sd->s_flags |= SYSFS_FLAG_REMOVED;
455 sd->s_sibling = acxt->removed;
458 if (sysfs_type(sd) == SYSFS_DIR && acxt->parent_inode)
459 drop_nlink(acxt->parent_inode);
465 * sysfs_drop_dentry - drop dentry for the specified sysfs_dirent
466 * @sd: target sysfs_dirent
468 * Drop dentry for @sd. @sd must have been unlinked from its
469 * parent on entry to this function such that it can't be looked
472 static void sysfs_drop_dentry(struct sysfs_dirent *sd)
475 struct dentry *dentry;
477 inode = ilookup(sysfs_sb, sd->s_ino);
481 /* Drop any existing dentries associated with sd.
483 * For the dentry to be properly freed we need to grab a
484 * reference to the dentry under the dcache lock, unhash it,
485 * and then put it. The playing with the dentry count allows
486 * dput to immediately free the dentry if it is not in use.
489 spin_lock(&dcache_lock);
490 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
491 if (d_unhashed(dentry))
494 spin_lock(&dentry->d_lock);
496 spin_unlock(&dentry->d_lock);
497 spin_unlock(&dcache_lock);
501 spin_unlock(&dcache_lock);
503 /* adjust nlink and update timestamp */
504 mutex_lock(&inode->i_mutex);
506 inode->i_ctime = CURRENT_TIME;
508 if (sysfs_type(sd) == SYSFS_DIR)
511 mutex_unlock(&inode->i_mutex);
517 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
518 * @acxt: addrm context to finish up
520 * Finish up sysfs_dirent add/remove. Resources acquired by
521 * sysfs_addrm_start() are released and removed sysfs_dirents are
522 * cleaned up. Timestamps on the parent inode are updated.
525 * All mutexes acquired by sysfs_addrm_start() are released.
527 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
529 /* release resources acquired by sysfs_addrm_start() */
530 mutex_unlock(&sysfs_mutex);
531 if (acxt->parent_inode) {
532 struct inode *inode = acxt->parent_inode;
534 /* if added/removed, update timestamps on the parent */
536 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
538 mutex_unlock(&inode->i_mutex);
542 /* kill removed sysfs_dirents */
543 while (acxt->removed) {
544 struct sysfs_dirent *sd = acxt->removed;
546 acxt->removed = sd->s_sibling;
547 sd->s_sibling = NULL;
549 sysfs_drop_dentry(sd);
550 sysfs_deactivate(sd);
556 * sysfs_find_dirent - find sysfs_dirent with the given name
557 * @parent_sd: sysfs_dirent to search under
558 * @name: name to look for
560 * Look for sysfs_dirent with name @name under @parent_sd.
563 * mutex_lock(sysfs_mutex)
566 * Pointer to sysfs_dirent if found, NULL if not.
568 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
569 const unsigned char *name)
571 struct sysfs_dirent *sd;
573 for (sd = parent_sd->s_children; sd; sd = sd->s_sibling)
574 if (!strcmp(sd->s_name, name))
580 * sysfs_get_dirent - find and get sysfs_dirent with the given name
581 * @parent_sd: sysfs_dirent to search under
582 * @name: name to look for
584 * Look for sysfs_dirent with name @name under @parent_sd and get
588 * Kernel thread context (may sleep). Grabs sysfs_mutex.
591 * Pointer to sysfs_dirent if found, NULL if not.
593 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
594 const unsigned char *name)
596 struct sysfs_dirent *sd;
598 mutex_lock(&sysfs_mutex);
599 sd = sysfs_find_dirent(parent_sd, name);
601 mutex_unlock(&sysfs_mutex);
606 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
607 const char *name, struct sysfs_dirent **p_sd)
609 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
610 struct sysfs_addrm_cxt acxt;
611 struct sysfs_dirent *sd;
615 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
618 sd->s_dir.kobj = kobj;
621 sysfs_addrm_start(&acxt, parent_sd);
622 rc = sysfs_add_one(&acxt, sd);
623 sysfs_addrm_finish(&acxt);
633 int sysfs_create_subdir(struct kobject *kobj, const char *name,
634 struct sysfs_dirent **p_sd)
636 return create_dir(kobj, kobj->sd, name, p_sd);
640 * sysfs_create_dir - create a directory for an object.
641 * @kobj: object we're creating directory for.
643 int sysfs_create_dir(struct kobject * kobj)
645 struct sysfs_dirent *parent_sd, *sd;
651 parent_sd = kobj->parent->sd;
653 parent_sd = &sysfs_root;
655 error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
661 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
662 struct nameidata *nd)
664 struct dentry *ret = NULL;
665 struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
666 struct sysfs_dirent *sd;
669 mutex_lock(&sysfs_mutex);
671 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
677 /* attach dentry and inode */
678 inode = sysfs_get_inode(sd);
680 ret = ERR_PTR(-ENOMEM);
684 /* instantiate and hash dentry */
685 dentry->d_op = &sysfs_dentry_ops;
686 dentry->d_fsdata = sysfs_get(sd);
687 d_instantiate(dentry, inode);
691 mutex_unlock(&sysfs_mutex);
695 const struct inode_operations sysfs_dir_inode_operations = {
696 .lookup = sysfs_lookup,
697 .setattr = sysfs_setattr,
700 static void remove_dir(struct sysfs_dirent *sd)
702 struct sysfs_addrm_cxt acxt;
704 sysfs_addrm_start(&acxt, sd->s_parent);
705 sysfs_remove_one(&acxt, sd);
706 sysfs_addrm_finish(&acxt);
709 void sysfs_remove_subdir(struct sysfs_dirent *sd)
715 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
717 struct sysfs_addrm_cxt acxt;
718 struct sysfs_dirent **pos;
723 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
724 sysfs_addrm_start(&acxt, dir_sd);
725 pos = &dir_sd->s_children;
727 struct sysfs_dirent *sd = *pos;
729 if (sysfs_type(sd) != SYSFS_DIR)
730 sysfs_remove_one(&acxt, sd);
732 pos = &(*pos)->s_sibling;
734 sysfs_addrm_finish(&acxt);
740 * sysfs_remove_dir - remove an object's directory.
743 * The only thing special about this is that we remove any files in
744 * the directory before we remove the directory, and we've inlined
745 * what used to be sysfs_rmdir() below, instead of calling separately.
748 void sysfs_remove_dir(struct kobject * kobj)
750 struct sysfs_dirent *sd = kobj->sd;
752 spin_lock(&sysfs_assoc_lock);
754 spin_unlock(&sysfs_assoc_lock);
756 __sysfs_remove_dir(sd);
759 int sysfs_rename_dir(struct kobject * kobj, const char *new_name)
761 struct sysfs_dirent *sd = kobj->sd;
762 struct dentry *parent = NULL;
763 struct dentry *old_dentry = NULL, *new_dentry = NULL;
764 const char *dup_name = NULL;
767 mutex_lock(&sysfs_rename_mutex);
770 if (strcmp(sd->s_name, new_name) == 0)
771 goto out; /* nothing to rename */
773 /* get the original dentry */
774 old_dentry = sysfs_get_dentry(sd);
775 if (IS_ERR(old_dentry)) {
776 error = PTR_ERR(old_dentry);
780 parent = old_dentry->d_parent;
782 /* lock parent and get dentry for new name */
783 mutex_lock(&parent->d_inode->i_mutex);
784 mutex_lock(&sysfs_mutex);
787 if (sysfs_find_dirent(sd->s_parent, new_name))
791 new_dentry = d_alloc_name(parent, new_name);
795 /* rename kobject and sysfs_dirent */
797 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
801 error = kobject_set_name(kobj, "%s", new_name);
805 dup_name = sd->s_name;
806 sd->s_name = new_name;
809 d_add(new_dentry, NULL);
810 d_move(old_dentry, new_dentry);
814 mutex_unlock(&sysfs_mutex);
815 mutex_unlock(&parent->d_inode->i_mutex);
820 mutex_unlock(&sysfs_rename_mutex);
824 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
826 struct sysfs_dirent *sd = kobj->sd;
827 struct sysfs_dirent *new_parent_sd;
828 struct dentry *old_parent, *new_parent = NULL;
829 struct dentry *old_dentry = NULL, *new_dentry = NULL;
832 mutex_lock(&sysfs_rename_mutex);
833 BUG_ON(!sd->s_parent);
834 new_parent_sd = new_parent_kobj->sd ? new_parent_kobj->sd : &sysfs_root;
837 if (sd->s_parent == new_parent_sd)
838 goto out; /* nothing to move */
841 old_dentry = sysfs_get_dentry(sd);
842 if (IS_ERR(old_dentry)) {
843 error = PTR_ERR(old_dentry);
846 old_parent = old_dentry->d_parent;
848 new_parent = sysfs_get_dentry(new_parent_sd);
849 if (IS_ERR(new_parent)) {
850 error = PTR_ERR(new_parent);
855 mutex_lock(&old_parent->d_inode->i_mutex);
856 if (!mutex_trylock(&new_parent->d_inode->i_mutex)) {
857 mutex_unlock(&old_parent->d_inode->i_mutex);
860 mutex_lock(&sysfs_mutex);
863 if (sysfs_find_dirent(new_parent_sd, sd->s_name))
867 new_dentry = d_alloc_name(new_parent, sd->s_name);
872 d_add(new_dentry, NULL);
873 d_move(old_dentry, new_dentry);
876 /* Remove from old parent's list and insert into new parent's list. */
877 sysfs_unlink_sibling(sd);
878 sysfs_get(new_parent_sd);
879 sysfs_put(sd->s_parent);
880 sd->s_parent = new_parent_sd;
881 sysfs_link_sibling(sd);
884 mutex_unlock(&sysfs_mutex);
885 mutex_unlock(&new_parent->d_inode->i_mutex);
886 mutex_unlock(&old_parent->d_inode->i_mutex);
891 mutex_unlock(&sysfs_rename_mutex);
895 /* Relationship between s_mode and the DT_xxx types */
896 static inline unsigned char dt_type(struct sysfs_dirent *sd)
898 return (sd->s_mode >> 12) & 15;
901 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
903 struct dentry *dentry = filp->f_path.dentry;
904 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
905 struct sysfs_dirent *pos;
908 if (filp->f_pos == 0) {
909 ino = parent_sd->s_ino;
910 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
913 if (filp->f_pos == 1) {
914 if (parent_sd->s_parent)
915 ino = parent_sd->s_parent->s_ino;
917 ino = parent_sd->s_ino;
918 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
921 if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
922 mutex_lock(&sysfs_mutex);
924 /* Skip the dentries we have already reported */
925 pos = parent_sd->s_children;
926 while (pos && (filp->f_pos > pos->s_ino))
927 pos = pos->s_sibling;
929 for ( ; pos; pos = pos->s_sibling) {
935 filp->f_pos = ino = pos->s_ino;
937 if (filldir(dirent, name, len, filp->f_pos, ino,
942 filp->f_pos = INT_MAX;
943 mutex_unlock(&sysfs_mutex);
949 const struct file_operations sysfs_dir_operations = {
950 .read = generic_read_dir,
951 .readdir = sysfs_readdir,