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 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);
109 if (likely(t == v)) {
110 rwsem_acquire_read(&sd->dep_map, 0, 1, _RET_IP_);
121 * sysfs_put_active - put an active reference to sysfs_dirent
122 * @sd: sysfs_dirent to put an active reference to
124 * Put an active reference to @sd. This function is noop if @sd
127 void sysfs_put_active(struct sysfs_dirent *sd)
129 struct completion *cmpl;
135 rwsem_release(&sd->dep_map, 1, _RET_IP_);
136 v = atomic_dec_return(&sd->s_active);
137 if (likely(v != SD_DEACTIVATED_BIAS))
140 /* atomic_dec_return() is a mb(), we'll always see the updated
143 cmpl = (void *)sd->s_sibling;
148 * sysfs_deactivate - deactivate sysfs_dirent
149 * @sd: sysfs_dirent to deactivate
151 * Deny new active references and drain existing ones.
153 static void sysfs_deactivate(struct sysfs_dirent *sd)
155 DECLARE_COMPLETION_ONSTACK(wait);
158 BUG_ON(sd->s_sibling || !(sd->s_flags & SYSFS_FLAG_REMOVED));
159 sd->s_sibling = (void *)&wait;
161 rwsem_acquire(&sd->dep_map, 0, 0, _RET_IP_);
162 /* atomic_add_return() is a mb(), put_active() will always see
163 * the updated sd->s_sibling.
165 v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
167 if (v != SD_DEACTIVATED_BIAS) {
168 lock_contended(&sd->dep_map, _RET_IP_);
169 wait_for_completion(&wait);
172 sd->s_sibling = NULL;
174 lock_acquired(&sd->dep_map, _RET_IP_);
175 rwsem_release(&sd->dep_map, 1, _RET_IP_);
178 static int sysfs_alloc_ino(ino_t *pino)
183 spin_lock(&sysfs_ino_lock);
184 rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
185 spin_unlock(&sysfs_ino_lock);
188 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
197 static void sysfs_free_ino(ino_t ino)
199 spin_lock(&sysfs_ino_lock);
200 ida_remove(&sysfs_ino_ida, ino);
201 spin_unlock(&sysfs_ino_lock);
204 void release_sysfs_dirent(struct sysfs_dirent * sd)
206 struct sysfs_dirent *parent_sd;
209 /* Moving/renaming is always done while holding reference.
210 * sd->s_parent won't change beneath us.
212 parent_sd = sd->s_parent;
214 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
215 sysfs_put(sd->s_symlink.target_sd);
216 if (sysfs_type(sd) & SYSFS_COPY_NAME)
218 if (sd->s_iattr && sd->s_iattr->ia_secdata)
219 security_release_secctx(sd->s_iattr->ia_secdata,
220 sd->s_iattr->ia_secdata_len);
222 sysfs_free_ino(sd->s_ino);
223 kmem_cache_free(sysfs_dir_cachep, sd);
226 if (sd && atomic_dec_and_test(&sd->s_count))
230 static int sysfs_dentry_delete(struct dentry *dentry)
232 struct sysfs_dirent *sd = dentry->d_fsdata;
233 return !!(sd->s_flags & SYSFS_FLAG_REMOVED);
236 static int sysfs_dentry_revalidate(struct dentry *dentry, struct nameidata *nd)
238 struct sysfs_dirent *sd = dentry->d_fsdata;
241 mutex_lock(&sysfs_mutex);
243 /* The sysfs dirent has been deleted */
244 if (sd->s_flags & SYSFS_FLAG_REMOVED)
247 /* The sysfs dirent has been moved? */
248 if (dentry->d_parent->d_fsdata != sd->s_parent)
251 /* The sysfs dirent has been renamed */
252 if (strcmp(dentry->d_name.name, sd->s_name) != 0)
255 mutex_unlock(&sysfs_mutex);
259 /* Remove the dentry from the dcache hashes.
260 * If this is a deleted dentry we use d_drop instead of d_delete
261 * so sysfs doesn't need to cope with negative dentries.
263 * If this is a dentry that has simply been renamed we
264 * use d_drop to remove it from the dcache lookup on its
265 * old parent. If this dentry persists later when a lookup
266 * is performed at its new name the dentry will be readded
267 * to the dcache hashes.
269 is_dir = (sysfs_type(sd) == SYSFS_DIR);
270 mutex_unlock(&sysfs_mutex);
272 /* If we have submounts we must allow the vfs caches
273 * to lie about the state of the filesystem to prevent
274 * leaks and other nasty things.
276 if (have_submounts(dentry))
278 shrink_dcache_parent(dentry);
284 static void sysfs_dentry_iput(struct dentry *dentry, struct inode *inode)
286 struct sysfs_dirent * sd = dentry->d_fsdata;
292 static const struct dentry_operations sysfs_dentry_ops = {
293 .d_revalidate = sysfs_dentry_revalidate,
294 .d_delete = sysfs_dentry_delete,
295 .d_iput = sysfs_dentry_iput,
298 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
300 char *dup_name = NULL;
301 struct sysfs_dirent *sd;
303 if (type & SYSFS_COPY_NAME) {
304 name = dup_name = kstrdup(name, GFP_KERNEL);
309 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
313 if (sysfs_alloc_ino(&sd->s_ino))
316 atomic_set(&sd->s_count, 1);
317 atomic_set(&sd->s_active, 0);
318 sysfs_dirent_init_lockdep(sd);
327 kmem_cache_free(sysfs_dir_cachep, sd);
334 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
335 * @acxt: pointer to sysfs_addrm_cxt to be used
336 * @parent_sd: parent sysfs_dirent
338 * This function is called when the caller is about to add or
339 * remove sysfs_dirent under @parent_sd. This function acquires
340 * sysfs_mutex. @acxt is used to keep and pass context to
341 * other addrm functions.
344 * Kernel thread context (may sleep). sysfs_mutex is locked on
347 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
348 struct sysfs_dirent *parent_sd)
350 memset(acxt, 0, sizeof(*acxt));
351 acxt->parent_sd = parent_sd;
353 mutex_lock(&sysfs_mutex);
357 * __sysfs_add_one - add sysfs_dirent to parent without warning
358 * @acxt: addrm context to use
359 * @sd: sysfs_dirent to be added
361 * Get @acxt->parent_sd and set sd->s_parent to it and increment
362 * nlink of parent inode if @sd is a directory and link into the
363 * children list of the parent.
365 * This function should be called between calls to
366 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
367 * passed the same @acxt as passed to sysfs_addrm_start().
370 * Determined by sysfs_addrm_start().
373 * 0 on success, -EEXIST if entry with the given name already
376 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
378 struct sysfs_inode_attrs *ps_iattr;
380 if (sysfs_find_dirent(acxt->parent_sd, sd->s_name))
383 sd->s_parent = sysfs_get(acxt->parent_sd);
385 sysfs_link_sibling(sd);
387 /* Update timestamps on the parent */
388 ps_iattr = acxt->parent_sd->s_iattr;
390 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
391 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
398 * sysfs_pathname - return full path to sysfs dirent
399 * @sd: sysfs_dirent whose path we want
400 * @path: caller allocated buffer
402 * Gives the name "/" to the sysfs_root entry; any path returned
403 * is relative to wherever sysfs is mounted.
405 * XXX: does no error checking on @path size
407 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
410 sysfs_pathname(sd->s_parent, path);
413 strcat(path, sd->s_name);
418 * sysfs_add_one - add sysfs_dirent to parent
419 * @acxt: addrm context to use
420 * @sd: sysfs_dirent to be added
422 * Get @acxt->parent_sd and set sd->s_parent to it and increment
423 * nlink of parent inode if @sd is a directory and link into the
424 * children list of the parent.
426 * This function should be called between calls to
427 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
428 * passed the same @acxt as passed to sysfs_addrm_start().
431 * Determined by sysfs_addrm_start().
434 * 0 on success, -EEXIST if entry with the given name already
437 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
441 ret = __sysfs_add_one(acxt, sd);
442 if (ret == -EEXIST) {
443 char *path = kzalloc(PATH_MAX, GFP_KERNEL);
445 "sysfs: cannot create duplicate filename '%s'\n",
446 (path == NULL) ? sd->s_name :
447 strcat(strcat(sysfs_pathname(acxt->parent_sd, path), "/"),
456 * sysfs_remove_one - remove sysfs_dirent from parent
457 * @acxt: addrm context to use
458 * @sd: sysfs_dirent to be removed
460 * Mark @sd removed and drop nlink of parent inode if @sd is a
461 * directory. @sd is unlinked from the children list.
463 * This function should be called between calls to
464 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
465 * passed the same @acxt as passed to sysfs_addrm_start().
468 * Determined by sysfs_addrm_start().
470 void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
472 struct sysfs_inode_attrs *ps_iattr;
474 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
476 sysfs_unlink_sibling(sd);
478 /* Update timestamps on the parent */
479 ps_iattr = acxt->parent_sd->s_iattr;
481 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
482 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
485 sd->s_flags |= SYSFS_FLAG_REMOVED;
486 sd->s_sibling = acxt->removed;
491 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
492 * @acxt: addrm context to finish up
494 * Finish up sysfs_dirent add/remove. Resources acquired by
495 * sysfs_addrm_start() are released and removed sysfs_dirents are
499 * sysfs_mutex is released.
501 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
503 /* release resources acquired by sysfs_addrm_start() */
504 mutex_unlock(&sysfs_mutex);
506 /* kill removed sysfs_dirents */
507 while (acxt->removed) {
508 struct sysfs_dirent *sd = acxt->removed;
510 acxt->removed = sd->s_sibling;
511 sd->s_sibling = NULL;
513 sysfs_deactivate(sd);
520 * sysfs_find_dirent - find sysfs_dirent with the given name
521 * @parent_sd: sysfs_dirent to search under
522 * @name: name to look for
524 * Look for sysfs_dirent with name @name under @parent_sd.
527 * mutex_lock(sysfs_mutex)
530 * Pointer to sysfs_dirent if found, NULL if not.
532 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
533 const unsigned char *name)
535 struct sysfs_dirent *sd;
537 for (sd = parent_sd->s_dir.children; sd; sd = sd->s_sibling)
538 if (!strcmp(sd->s_name, name))
544 * sysfs_get_dirent - find and get sysfs_dirent with the given name
545 * @parent_sd: sysfs_dirent to search under
546 * @name: name to look for
548 * Look for sysfs_dirent with name @name under @parent_sd and get
552 * Kernel thread context (may sleep). Grabs sysfs_mutex.
555 * Pointer to sysfs_dirent if found, NULL if not.
557 struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
558 const unsigned char *name)
560 struct sysfs_dirent *sd;
562 mutex_lock(&sysfs_mutex);
563 sd = sysfs_find_dirent(parent_sd, name);
565 mutex_unlock(&sysfs_mutex);
569 EXPORT_SYMBOL_GPL(sysfs_get_dirent);
571 static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
572 const char *name, struct sysfs_dirent **p_sd)
574 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
575 struct sysfs_addrm_cxt acxt;
576 struct sysfs_dirent *sd;
580 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
583 sd->s_dir.kobj = kobj;
586 sysfs_addrm_start(&acxt, parent_sd);
587 rc = sysfs_add_one(&acxt, sd);
588 sysfs_addrm_finish(&acxt);
598 int sysfs_create_subdir(struct kobject *kobj, const char *name,
599 struct sysfs_dirent **p_sd)
601 return create_dir(kobj, kobj->sd, name, p_sd);
605 * sysfs_create_dir - create a directory for an object.
606 * @kobj: object we're creating directory for.
608 int sysfs_create_dir(struct kobject * kobj)
610 struct sysfs_dirent *parent_sd, *sd;
616 parent_sd = kobj->parent->sd;
618 parent_sd = &sysfs_root;
620 error = create_dir(kobj, parent_sd, kobject_name(kobj), &sd);
626 static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
627 struct nameidata *nd)
629 struct dentry *ret = NULL;
630 struct sysfs_dirent *parent_sd = dentry->d_parent->d_fsdata;
631 struct sysfs_dirent *sd;
634 mutex_lock(&sysfs_mutex);
636 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name);
640 ret = ERR_PTR(-ENOENT);
644 /* attach dentry and inode */
645 inode = sysfs_get_inode(sd);
647 ret = ERR_PTR(-ENOMEM);
651 /* instantiate and hash dentry */
652 ret = d_find_alias(inode);
654 dentry->d_op = &sysfs_dentry_ops;
655 dentry->d_fsdata = sysfs_get(sd);
656 d_add(dentry, inode);
663 mutex_unlock(&sysfs_mutex);
667 const struct inode_operations sysfs_dir_inode_operations = {
668 .lookup = sysfs_lookup,
669 .permission = sysfs_permission,
670 .setattr = sysfs_setattr,
671 .getattr = sysfs_getattr,
672 .setxattr = sysfs_setxattr,
675 static void remove_dir(struct sysfs_dirent *sd)
677 struct sysfs_addrm_cxt acxt;
679 sysfs_addrm_start(&acxt, sd->s_parent);
680 sysfs_remove_one(&acxt, sd);
681 sysfs_addrm_finish(&acxt);
684 void sysfs_remove_subdir(struct sysfs_dirent *sd)
690 static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
692 struct sysfs_addrm_cxt acxt;
693 struct sysfs_dirent **pos;
698 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
699 sysfs_addrm_start(&acxt, dir_sd);
700 pos = &dir_sd->s_dir.children;
702 struct sysfs_dirent *sd = *pos;
704 if (sysfs_type(sd) != SYSFS_DIR)
705 sysfs_remove_one(&acxt, sd);
707 pos = &(*pos)->s_sibling;
709 sysfs_addrm_finish(&acxt);
715 * sysfs_remove_dir - remove an object's directory.
718 * The only thing special about this is that we remove any files in
719 * the directory before we remove the directory, and we've inlined
720 * what used to be sysfs_rmdir() below, instead of calling separately.
723 void sysfs_remove_dir(struct kobject * kobj)
725 struct sysfs_dirent *sd = kobj->sd;
727 spin_lock(&sysfs_assoc_lock);
729 spin_unlock(&sysfs_assoc_lock);
731 __sysfs_remove_dir(sd);
734 int sysfs_rename(struct sysfs_dirent *sd,
735 struct sysfs_dirent *new_parent_sd, const char *new_name)
737 const char *dup_name = NULL;
740 mutex_lock(&sysfs_mutex);
743 if ((sd->s_parent == new_parent_sd) &&
744 (strcmp(sd->s_name, new_name) == 0))
745 goto out; /* nothing to rename */
748 if (sysfs_find_dirent(new_parent_sd, new_name))
751 /* rename sysfs_dirent */
752 if (strcmp(sd->s_name, new_name) != 0) {
754 new_name = dup_name = kstrdup(new_name, GFP_KERNEL);
758 dup_name = sd->s_name;
759 sd->s_name = new_name;
762 /* Remove from old parent's list and insert into new parent's list. */
763 if (sd->s_parent != new_parent_sd) {
764 sysfs_unlink_sibling(sd);
765 sysfs_get(new_parent_sd);
766 sysfs_put(sd->s_parent);
767 sd->s_parent = new_parent_sd;
768 sysfs_link_sibling(sd);
773 mutex_unlock(&sysfs_mutex);
778 int sysfs_rename_dir(struct kobject *kobj, const char *new_name)
780 return sysfs_rename(kobj->sd, kobj->sd->s_parent, new_name);
783 int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
785 struct sysfs_dirent *sd = kobj->sd;
786 struct sysfs_dirent *new_parent_sd;
788 BUG_ON(!sd->s_parent);
789 new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
790 new_parent_kobj->sd : &sysfs_root;
792 return sysfs_rename(sd, new_parent_sd, sd->s_name);
795 /* Relationship between s_mode and the DT_xxx types */
796 static inline unsigned char dt_type(struct sysfs_dirent *sd)
798 return (sd->s_mode >> 12) & 15;
801 static int sysfs_dir_release(struct inode *inode, struct file *filp)
803 sysfs_put(filp->private_data);
807 static struct sysfs_dirent *sysfs_dir_pos(struct sysfs_dirent *parent_sd,
808 ino_t ino, struct sysfs_dirent *pos)
811 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
812 pos->s_parent == parent_sd &&
819 if ((ino > 1) && (ino < INT_MAX)) {
820 pos = parent_sd->s_dir.children;
821 while (pos && (ino > pos->s_ino))
822 pos = pos->s_sibling;
827 static struct sysfs_dirent *sysfs_dir_next_pos(struct sysfs_dirent *parent_sd,
828 ino_t ino, struct sysfs_dirent *pos)
830 pos = sysfs_dir_pos(parent_sd, ino, pos);
832 pos = pos->s_sibling;
836 static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
838 struct dentry *dentry = filp->f_path.dentry;
839 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
840 struct sysfs_dirent *pos = filp->private_data;
843 if (filp->f_pos == 0) {
844 ino = parent_sd->s_ino;
845 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
848 if (filp->f_pos == 1) {
849 if (parent_sd->s_parent)
850 ino = parent_sd->s_parent->s_ino;
852 ino = parent_sd->s_ino;
853 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
856 mutex_lock(&sysfs_mutex);
857 for (pos = sysfs_dir_pos(parent_sd, filp->f_pos, pos);
859 pos = sysfs_dir_next_pos(parent_sd, filp->f_pos, pos)) {
869 filp->private_data = sysfs_get(pos);
871 mutex_unlock(&sysfs_mutex);
872 ret = filldir(dirent, name, len, filp->f_pos, ino, type);
873 mutex_lock(&sysfs_mutex);
877 mutex_unlock(&sysfs_mutex);
878 if ((filp->f_pos > 1) && !pos) { /* EOF */
879 filp->f_pos = INT_MAX;
880 filp->private_data = NULL;
886 const struct file_operations sysfs_dir_operations = {
887 .read = generic_read_dir,
888 .readdir = sysfs_readdir,
889 .release = sysfs_dir_release,
890 .llseek = generic_file_llseek,