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>
23 #include <linux/slab.h>
24 #include <linux/security.h>
27 DEFINE_MUTEX(sysfs_mutex);
28 DEFINE_SPINLOCK(sysfs_assoc_lock);
30 static DEFINE_SPINLOCK(sysfs_ino_lock);
31 static DEFINE_IDA(sysfs_ino_ida);
34 * sysfs_link_sibling - link sysfs_dirent into sibling list
35 * @sd: sysfs_dirent of interest
37 * Link @sd into its sibling list which starts from
38 * sd->s_parent->s_dir.children.
41 * mutex_lock(sysfs_mutex)
43 static void sysfs_link_sibling(struct sysfs_dirent *sd)
45 struct sysfs_dirent *parent_sd = sd->s_parent;
46 struct sysfs_dirent **pos;
48 BUG_ON(sd->s_sibling);
50 /* Store directory entries in order by ino. This allows
51 * readdir to properly restart without having to add a
52 * cursor into the s_dir.children list.
54 for (pos = &parent_sd->s_dir.children; *pos; pos = &(*pos)->s_sibling) {
55 if (sd->s_ino < (*pos)->s_ino)
63 * sysfs_unlink_sibling - unlink sysfs_dirent from sibling list
64 * @sd: sysfs_dirent of interest
66 * Unlink @sd from its sibling list which starts from
67 * sd->s_parent->s_dir.children.
70 * mutex_lock(sysfs_mutex)
72 static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
74 struct sysfs_dirent **pos;
76 for (pos = &sd->s_parent->s_dir.children; *pos;
77 pos = &(*pos)->s_sibling) {
87 * sysfs_get_active - get an active reference to sysfs_dirent
88 * @sd: sysfs_dirent to get an active reference to
90 * Get an active reference of @sd. This function is noop if @sd
94 * Pointer to @sd on success, NULL on failure.
96 static struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
104 v = atomic_read(&sd->s_active);
108 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
119 * sysfs_put_active - put an active reference to sysfs_dirent
120 * @sd: sysfs_dirent to put an active reference to
122 * Put an active reference to @sd. This function is noop if @sd
125 static void sysfs_put_active(struct sysfs_dirent *sd)
127 struct completion *cmpl;
133 v = atomic_dec_return(&sd->s_active);
134 if (likely(v != SD_DEACTIVATED_BIAS))
137 /* atomic_dec_return() is a mb(), we'll always see the updated
140 cmpl = (void *)sd->s_sibling;
145 * sysfs_get_active_two - get active references to sysfs_dirent and parent
146 * @sd: sysfs_dirent of interest
148 * Get active reference to @sd and its parent. Parent's active
149 * reference is grabbed first. This function is noop if @sd is
153 * Pointer to @sd on success, NULL on failure.
155 struct sysfs_dirent *sysfs_get_active_two(struct sysfs_dirent *sd)
158 if (sd->s_parent && unlikely(!sysfs_get_active(sd->s_parent)))
160 if (unlikely(!sysfs_get_active(sd))) {
161 sysfs_put_active(sd->s_parent);
169 * sysfs_put_active_two - put active references to sysfs_dirent and parent
170 * @sd: sysfs_dirent of interest
172 * Put active references to @sd and its parent. This function is
173 * noop if @sd is NULL.
175 void sysfs_put_active_two(struct sysfs_dirent *sd)
178 sysfs_put_active(sd);
179 sysfs_put_active(sd->s_parent);
184 * sysfs_deactivate - deactivate sysfs_dirent
185 * @sd: sysfs_dirent to deactivate
187 * Deny new active references and drain existing ones.
189 static void sysfs_deactivate(struct sysfs_dirent *sd)
191 DECLARE_COMPLETION_ONSTACK(wait);
194 BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
195 sd->s_sibling = (void *)&wait;
197 /* atomic_add_return() is a mb(), put_active() will always see
198 * the updated sd->s_sibling.
200 v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
202 if (v != SD_DEACTIVATED_BIAS)
203 wait_for_completion(&wait);
205 sd->s_sibling = NULL;
208 static int sysfs_alloc_ino(ino_t *pino)
213 spin_lock(&sysfs_ino_lock);
214 rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
215 spin_unlock(&sysfs_ino_lock);
218 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
227 static void sysfs_free_ino(ino_t ino)
229 spin_lock(&sysfs_ino_lock);
230 ida_remove(&sysfs_ino_ida, ino);
231 spin_unlock(&sysfs_ino_lock);
234 void release_sysfs_dirent(struct sysfs_dirent * sd)
236 struct sysfs_dirent *parent_sd;
239 /* Moving/renaming is always done while holding reference.
240 * sd->s_parent won't change beneath us.
242 parent_sd = sd->s_parent;
244 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
245 sysfs_put(sd->s_symlink.target_sd);
246 if (sysfs_type(sd) & SYSFS_COPY_NAME)
248 if (sd->s_iattr && sd->s_iattr->ia_secdata)
249 security_release_secctx(sd->s_iattr->ia_secdata,
250 sd->s_iattr->ia_secdata_len);
252 sysfs_free_ino(sd->s_ino);
253 kmem_cache_free(sysfs_dir_cachep, sd);
256 if (sd && atomic_dec_and_test(&sd->s_count))
260 static int sysfs_dentry_delete(struct dentry *dentry)
262 struct sysfs_dirent *sd = dentry->d_fsdata;
263 return !!(sd->s_flags & SYSFS_FLAG_REMOVED);
266 static int sysfs_dentry_revalidate(struct dentry *dentry, struct nameidata *nd)
268 struct sysfs_dirent *sd = dentry->d_fsdata;
271 mutex_lock(&sysfs_mutex);
273 /* The sysfs dirent has been deleted */
274 if (sd->s_flags & SYSFS_FLAG_REMOVED)
277 /* The sysfs dirent has been moved? */
278 if (dentry->d_parent->d_fsdata != sd->s_parent)
281 /* The sysfs dirent has been renamed */
282 if (strcmp(dentry->d_name.name, sd->s_name) != 0)
285 mutex_unlock(&sysfs_mutex);
289 /* Remove the dentry from the dcache hashes.
290 * If this is a deleted dentry we use d_drop instead of d_delete
291 * so sysfs doesn't need to cope with negative dentries.
293 * If this is a dentry that has simply been renamed we
294 * use d_drop to remove it from the dcache lookup on its
295 * old parent. If this dentry persists later when a lookup
296 * is performed at its new name the dentry will be readded
297 * to the dcache hashes.
299 is_dir = (sysfs_type(sd) == SYSFS_DIR);
300 mutex_unlock(&sysfs_mutex);
302 /* If we have submounts we must allow the vfs caches
303 * to lie about the state of the filesystem to prevent
304 * leaks and other nasty things.
306 if (have_submounts(dentry))
308 shrink_dcache_parent(dentry);
314 static void sysfs_dentry_iput(struct dentry *dentry, struct inode *inode)
316 struct sysfs_dirent * sd = dentry->d_fsdata;
322 static const struct dentry_operations sysfs_dentry_ops = {
323 .d_revalidate = sysfs_dentry_revalidate,
324 .d_delete = sysfs_dentry_delete,
325 .d_iput = sysfs_dentry_iput,
328 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
330 char *dup_name = NULL;
331 struct sysfs_dirent *sd;
333 if (type & SYSFS_COPY_NAME) {
334 name = dup_name = kstrdup(name, GFP_KERNEL);
339 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
343 if (sysfs_alloc_ino(&sd->s_ino))
346 atomic_set(&sd->s_count, 1);
347 atomic_set(&sd->s_active, 0);
356 kmem_cache_free(sysfs_dir_cachep, sd);
363 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
364 * @acxt: pointer to sysfs_addrm_cxt to be used
365 * @parent_sd: parent sysfs_dirent
367 * This function is called when the caller is about to add or
368 * remove sysfs_dirent under @parent_sd. This function acquires
369 * sysfs_mutex. @acxt is used to keep and pass context to
370 * other addrm functions.
373 * Kernel thread context (may sleep). sysfs_mutex is locked on
376 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
377 struct sysfs_dirent *parent_sd)
379 memset(acxt, 0, sizeof(*acxt));
380 acxt->parent_sd = parent_sd;
382 mutex_lock(&sysfs_mutex);
386 * __sysfs_add_one - add sysfs_dirent to parent without warning
387 * @acxt: addrm context to use
388 * @sd: sysfs_dirent to be added
390 * Get @acxt->parent_sd and set sd->s_parent to it and increment
391 * nlink of parent inode if @sd is a directory and link into the
392 * children list of the parent.
394 * This function should be called between calls to
395 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
396 * passed the same @acxt as passed to sysfs_addrm_start().
399 * Determined by sysfs_addrm_start().
402 * 0 on success, -EEXIST if entry with the given name already
405 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
407 struct sysfs_inode_attrs *ps_iattr;
409 if (sysfs_find_dirent(acxt->parent_sd, sd->s_name))
412 sd->s_parent = sysfs_get(acxt->parent_sd);
414 sysfs_link_sibling(sd);
416 /* Update timestamps on the parent */
417 ps_iattr = acxt->parent_sd->s_iattr;
419 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
420 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
427 * sysfs_pathname - return full path to sysfs dirent
428 * @sd: sysfs_dirent whose path we want
429 * @path: caller allocated buffer
431 * Gives the name "/" to the sysfs_root entry; any path returned
432 * is relative to wherever sysfs is mounted.
434 * XXX: does no error checking on @path size
436 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
439 sysfs_pathname(sd->s_parent, path);
442 strcat(path, sd->s_name);
447 * sysfs_add_one - add sysfs_dirent to parent
448 * @acxt: addrm context to use
449 * @sd: sysfs_dirent to be added
451 * Get @acxt->parent_sd and set sd->s_parent to it and increment
452 * nlink of parent inode if @sd is a directory and link into the
453 * children list of the parent.
455 * This function should be called between calls to
456 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
457 * passed the same @acxt as passed to sysfs_addrm_start().
460 * Determined by sysfs_addrm_start().
463 * 0 on success, -EEXIST if entry with the given name already
466 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
470 ret = __sysfs_add_one(acxt, sd);
471 if (ret == -EEXIST) {
472 char *path = kzalloc(PATH_MAX, GFP_KERNEL);
474 "sysfs: cannot create duplicate filename '%s'\n",
475 (path == NULL) ? sd->s_name :
476 strcat(strcat(sysfs_pathname(acxt->parent_sd, path), "/"),
485 * sysfs_remove_one - remove sysfs_dirent from parent
486 * @acxt: addrm context to use
487 * @sd: sysfs_dirent to be removed
489 * Mark @sd removed and drop nlink of parent inode if @sd is a
490 * directory. @sd is unlinked from the children list.
492 * This function should be called between calls to
493 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
494 * passed the same @acxt as passed to sysfs_addrm_start().
497 * Determined by sysfs_addrm_start().
499 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
501 struct sysfs_inode_attrs *ps_iattr;
503 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
505 sysfs_unlink_sibling(sd);
507 /* Update timestamps on the parent */
508 ps_iattr = acxt->parent_sd->s_iattr;
510 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
511 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
514 sd->s_flags |= SYSFS_FLAG_REMOVED;
515 sd->s_sibling = acxt->removed;
520 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
521 * @acxt: addrm context to finish up
523 * Finish up sysfs_dirent add/remove. Resources acquired by
524 * sysfs_addrm_start() are released and removed sysfs_dirents are
528 * sysfs_mutex is released.
530 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
532 /* release resources acquired by sysfs_addrm_start() */
533 mutex_unlock(&sysfs_mutex);
535 /* kill removed sysfs_dirents */
536 while (acxt->removed) {
537 struct sysfs_dirent *sd = acxt->removed;
539 acxt->removed = sd->s_sibling;
540 sd->s_sibling = NULL;
542 sysfs_deactivate(sd);
549 * sysfs_find_dirent - find sysfs_dirent with the given name
550 * @parent_sd: sysfs_dirent to search under
551 * @name: name to look for
553 * Look for sysfs_dirent with name @name under @parent_sd.
556 * mutex_lock(sysfs_mutex)
559 * Pointer to sysfs_dirent if found, NULL if not.
561 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
562 const unsigned char *name)
564 struct sysfs_dirent *sd;
566 for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
567 if (!strcmp(sd->s_name, name))
573 * sysfs_get_dirent - find and get sysfs_dirent with the given name
574 * @parent_sd: sysfs_dirent to search under
575 * @name: name to look for
577 * Look for sysfs_dirent with name @name under @parent_sd and get
581 * Kernel thread context (may sleep). Grabs sysfs_mutex.
584 * Pointer to sysfs_dirent if found, NULL if not.
586 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
587 const unsigned char *name)
589 struct sysfs_dirent *sd;
591 mutex_lock(&sysfs_mutex);
592 sd = sysfs_find_dirent(parent_sd, name);
594 mutex_unlock(&sysfs_mutex);
598 EXPORT_SYMBOL_GPL(sysfs_get_dirent);
600 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
601 const char *name, struct sysfs_dirent **p_sd)
603 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
604 struct sysfs_addrm_cxt acxt;
605 struct sysfs_dirent *sd;
609 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
612 sd->s_dir.kobj = kobj;
615 sysfs_addrm_start(&acxt, parent_sd);
616 rc = sysfs_add_one(&acxt, sd);
617 sysfs_addrm_finish(&acxt);
627 int sysfs_create_subdir(struct kobject *kobj, const char *name,
628 struct sysfs_dirent **p_sd)
630 return create_dir(kobj, kobj->sd, name, p_sd);
634 * sysfs_create_dir - create a directory for an object.
635 * @kobj: object we're creating directory for.
637 int sysfs_create_dir(struct kobject * kobj)
639 struct sysfs_dirent *parent_sd, *sd;
645 parent_sd = kobj->parent->sd;
647 parent_sd = &sysfs_root;
649 error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
655 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
656 struct nameidata *nd)
658 struct dentry *ret = NULL;
659 struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
660 struct sysfs_dirent *sd;
663 mutex_lock(&sysfs_mutex);
665 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
669 ret = ERR_PTR(-ENOENT);
673 /* attach dentry and inode */
674 inode = sysfs_get_inode(sd);
676 ret = ERR_PTR(-ENOMEM);
680 /* instantiate and hash dentry */
681 ret = d_find_alias(inode);
683 dentry->d_op = &sysfs_dentry_ops;
684 dentry->d_fsdata = sysfs_get(sd);
685 d_add(dentry, inode);
692 mutex_unlock(&sysfs_mutex);
696 const struct inode_operations sysfs_dir_inode_operations = {
697 .lookup = sysfs_lookup,
698 .permission = sysfs_permission,
699 .setattr = sysfs_setattr,
700 .getattr = sysfs_getattr,
701 .setxattr = sysfs_setxattr,
704 static void remove_dir(struct sysfs_dirent *sd)
706 struct sysfs_addrm_cxt acxt;
708 sysfs_addrm_start(&acxt, sd->s_parent);
709 sysfs_remove_one(&acxt, sd);
710 sysfs_addrm_finish(&acxt);
713 void sysfs_remove_subdir(struct sysfs_dirent *sd)
719 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
721 struct sysfs_addrm_cxt acxt;
722 struct sysfs_dirent **pos;
727 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
728 sysfs_addrm_start(&acxt, dir_sd);
729 pos = &dir_sd->s_dir.children;
731 struct sysfs_dirent *sd = *pos;
733 if (sysfs_type(sd) != SYSFS_DIR)
734 sysfs_remove_one(&acxt, sd);
736 pos = &(*pos)->s_sibling;
738 sysfs_addrm_finish(&acxt);
744 * sysfs_remove_dir - remove an object's directory.
747 * The only thing special about this is that we remove any files in
748 * the directory before we remove the directory, and we've inlined
749 * what used to be sysfs_rmdir() below, instead of calling separately.
752 void sysfs_remove_dir(struct kobject * kobj)
754 struct sysfs_dirent *sd = kobj->sd;
756 spin_lock(&sysfs_assoc_lock);
758 spin_unlock(&sysfs_assoc_lock);
760 __sysfs_remove_dir(sd);
763 int sysfs_rename(struct sysfs_dirent *sd,
764 struct sysfs_dirent *new_parent_sd, const char *new_name)
766 const char *dup_name = NULL;
769 mutex_lock(&sysfs_mutex);
772 if ((sd->s_parent == new_parent_sd) &&
773 (strcmp(sd->s_name, new_name) == 0))
774 goto out; /* nothing to rename */
777 if (sysfs_find_dirent(new_parent_sd, new_name))
780 /* rename sysfs_dirent */
781 if (strcmp(sd->s_name, new_name) != 0) {
783 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
787 dup_name = sd->s_name;
788 sd->s_name = new_name;
791 /* Remove from old parent's list and insert into new parent's list. */
792 if (sd->s_parent != new_parent_sd) {
793 sysfs_unlink_sibling(sd);
794 sysfs_get(new_parent_sd);
795 sysfs_put(sd->s_parent);
796 sd->s_parent = new_parent_sd;
797 sysfs_link_sibling(sd);
802 mutex_unlock(&sysfs_mutex);
807 int sysfs_rename_dir(struct kobject *kobj, const char *new_name)
809 return sysfs_rename(kobj->sd, kobj->sd->s_parent, new_name);
812 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
814 struct sysfs_dirent *sd = kobj->sd;
815 struct sysfs_dirent *new_parent_sd;
817 BUG_ON(!sd->s_parent);
818 new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
819 new_parent_kobj->sd : &sysfs_root;
821 return sysfs_rename(sd, new_parent_sd, sd->s_name);
824 /* Relationship between s_mode and the DT_xxx types */
825 static inline unsigned char dt_type(struct sysfs_dirent *sd)
827 return (sd->s_mode >> 12) & 15;
830 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
832 struct dentry *dentry = filp->f_path.dentry;
833 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
834 struct sysfs_dirent *pos;
837 if (filp->f_pos == 0) {
838 ino = parent_sd->s_ino;
839 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
842 if (filp->f_pos == 1) {
843 if (parent_sd->s_parent)
844 ino = parent_sd->s_parent->s_ino;
846 ino = parent_sd->s_ino;
847 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
850 if ((filp->f_pos > 1) && (filp->f_pos < INT_MAX)) {
851 mutex_lock(&sysfs_mutex);
853 /* Skip the dentries we have already reported */
854 pos = parent_sd->s_dir.children;
855 while (pos && (filp->f_pos > pos->s_ino))
856 pos = pos->s_sibling;
858 for ( ; pos; pos = pos->s_sibling) {
864 filp->f_pos = ino = pos->s_ino;
866 if (filldir(dirent, name, len, filp->f_pos, ino,
871 filp->f_pos = INT_MAX;
872 mutex_unlock(&sysfs_mutex);
878 const struct file_operations sysfs_dir_operations = {
879 .read = generic_read_dir,
880 .readdir = sysfs_readdir,
881 .llseek = generic_file_llseek,