1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * dir.c - Operations for configfs directories.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
22 * sysfs is Copyright (C) 2001, 2002, 2003 Patrick Mochel
24 * configfs Copyright (C) 2005 Oracle. All rights reserved.
30 #include <linux/mount.h>
31 #include <linux/module.h>
32 #include <linux/slab.h>
33 #include <linux/err.h>
35 #include <linux/configfs.h>
36 #include "configfs_internal.h"
38 DECLARE_RWSEM(configfs_rename_sem);
40 * Protects mutations of configfs_dirent linkage together with proper i_mutex
41 * Also protects mutations of symlinks linkage to target configfs_dirent
42 * Mutators of configfs_dirent linkage must *both* have the proper inode locked
43 * and configfs_dirent_lock locked, in that order.
44 * This allows one to safely traverse configfs_dirent trees and symlinks without
45 * having to lock inodes.
47 * Protects setting of CONFIGFS_USET_DROPPING: checking the flag
48 * unlocked is not reliable unless in detach_groups() called from
49 * rmdir()/unregister() and from configfs_attach_group()
51 DEFINE_SPINLOCK(configfs_dirent_lock);
53 static void configfs_d_iput(struct dentry * dentry,
56 struct configfs_dirent *sd = dentry->d_fsdata;
59 /* Coordinate with configfs_readdir */
60 spin_lock(&configfs_dirent_lock);
61 /* Coordinate with configfs_attach_attr where will increase
62 * sd->s_count and update sd->s_dentry to new allocated one.
63 * Only set sd->dentry to null when this dentry is the only
65 * If not do so, configfs_d_iput may run just after
66 * configfs_attach_attr and set sd->s_dentry to null
67 * even it's still in use.
69 if (atomic_read(&sd->s_count) <= 2)
72 spin_unlock(&configfs_dirent_lock);
78 const struct dentry_operations configfs_dentry_ops = {
79 .d_iput = configfs_d_iput,
80 .d_delete = always_delete_dentry,
86 * Helpers to make lockdep happy with our recursive locking of default groups'
87 * inodes (see configfs_attach_group() and configfs_detach_group()).
88 * We put default groups i_mutexes in separate classes according to their depth
89 * from the youngest non-default group ancestor.
91 * For a non-default group A having default groups A/B, A/C, and A/C/D, default
92 * groups A/B and A/C will have their inode's mutex in class
93 * default_group_class[0], and default group A/C/D will be in
94 * default_group_class[1].
96 * The lock classes are declared and assigned in inode.c, according to the
98 * The s_depth value is initialized to -1, adjusted to >= 0 when attaching
99 * default groups, and reset to -1 when all default groups are attached. During
100 * attachment, if configfs_create() sees s_depth > 0, the lock class of the new
101 * inode's mutex is set to default_group_class[s_depth - 1].
104 static void configfs_init_dirent_depth(struct configfs_dirent *sd)
109 static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
110 struct configfs_dirent *sd)
112 int parent_depth = parent_sd->s_depth;
114 if (parent_depth >= 0)
115 sd->s_depth = parent_depth + 1;
119 configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
122 * item's i_mutex class is already setup, so s_depth is now only
123 * used to set new sub-directories s_depth, which is always done
124 * with item's i_mutex locked.
127 * sd->s_depth == -1 iff we are a non default group.
128 * else (we are a default group) sd->s_depth > 0 (see
131 if (sd->s_depth == -1)
133 * We are a non default group and we are going to create
140 configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
142 /* We will not create default groups anymore. */
146 #else /* CONFIG_LOCKDEP */
148 static void configfs_init_dirent_depth(struct configfs_dirent *sd)
152 static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
153 struct configfs_dirent *sd)
158 configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
163 configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
167 #endif /* CONFIG_LOCKDEP */
170 * Allocates a new configfs_dirent and links it to the parent configfs_dirent
172 static struct configfs_dirent *configfs_new_dirent(struct configfs_dirent *parent_sd,
173 void *element, int type)
175 struct configfs_dirent * sd;
177 sd = kmem_cache_zalloc(configfs_dir_cachep, GFP_KERNEL);
179 return ERR_PTR(-ENOMEM);
181 atomic_set(&sd->s_count, 1);
182 INIT_LIST_HEAD(&sd->s_links);
183 INIT_LIST_HEAD(&sd->s_children);
184 sd->s_element = element;
186 configfs_init_dirent_depth(sd);
187 spin_lock(&configfs_dirent_lock);
188 if (parent_sd->s_type & CONFIGFS_USET_DROPPING) {
189 spin_unlock(&configfs_dirent_lock);
190 kmem_cache_free(configfs_dir_cachep, sd);
191 return ERR_PTR(-ENOENT);
193 list_add(&sd->s_sibling, &parent_sd->s_children);
194 spin_unlock(&configfs_dirent_lock);
201 * Return -EEXIST if there is already a configfs element with the same
202 * name for the same parent.
204 * called with parent inode's i_mutex held
206 static int configfs_dirent_exists(struct configfs_dirent *parent_sd,
207 const unsigned char *new)
209 struct configfs_dirent * sd;
211 list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
213 const unsigned char *existing = configfs_get_name(sd);
214 if (strcmp(existing, new))
225 int configfs_make_dirent(struct configfs_dirent * parent_sd,
226 struct dentry * dentry, void * element,
227 umode_t mode, int type)
229 struct configfs_dirent * sd;
231 sd = configfs_new_dirent(parent_sd, element, type);
236 sd->s_dentry = dentry;
238 dentry->d_fsdata = configfs_get(sd);
243 static int init_dir(struct inode * inode)
245 inode->i_op = &configfs_dir_inode_operations;
246 inode->i_fop = &configfs_dir_operations;
248 /* directory inodes start off with i_nlink == 2 (for "." entry) */
253 static int configfs_init_file(struct inode * inode)
255 inode->i_size = PAGE_SIZE;
256 inode->i_fop = &configfs_file_operations;
260 static int init_symlink(struct inode * inode)
262 inode->i_op = &configfs_symlink_inode_operations;
266 static int create_dir(struct config_item *k, struct dentry *d)
269 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
270 struct dentry *p = d->d_parent;
274 error = configfs_dirent_exists(p->d_fsdata, d->d_name.name);
276 error = configfs_make_dirent(p->d_fsdata, d, k, mode,
277 CONFIGFS_DIR | CONFIGFS_USET_CREATING);
279 configfs_set_dir_dirent_depth(p->d_fsdata, d->d_fsdata);
280 error = configfs_create(d, mode, init_dir);
282 inc_nlink(p->d_inode);
284 struct configfs_dirent *sd = d->d_fsdata;
286 spin_lock(&configfs_dirent_lock);
287 list_del_init(&sd->s_sibling);
288 spin_unlock(&configfs_dirent_lock);
298 * configfs_create_dir - create a directory for an config_item.
299 * @item: config_itemwe're creating directory for.
300 * @dentry: config_item's dentry.
302 * Note: user-created entries won't be allowed under this new directory
303 * until it is validated by configfs_dir_set_ready()
306 static int configfs_create_dir(struct config_item * item, struct dentry *dentry)
308 int error = create_dir(item, dentry);
310 item->ci_dentry = dentry;
315 * Allow userspace to create new entries under a new directory created with
316 * configfs_create_dir(), and under all of its chidlren directories recursively.
317 * @sd configfs_dirent of the new directory to validate
319 * Caller must hold configfs_dirent_lock.
321 static void configfs_dir_set_ready(struct configfs_dirent *sd)
323 struct configfs_dirent *child_sd;
325 sd->s_type &= ~CONFIGFS_USET_CREATING;
326 list_for_each_entry(child_sd, &sd->s_children, s_sibling)
327 if (child_sd->s_type & CONFIGFS_USET_CREATING)
328 configfs_dir_set_ready(child_sd);
332 * Check that a directory does not belong to a directory hierarchy being
333 * attached and not validated yet.
334 * @sd configfs_dirent of the directory to check
336 * @return non-zero iff the directory was validated
338 * Note: takes configfs_dirent_lock, so the result may change from false to true
339 * in two consecutive calls, but never from true to false.
341 int configfs_dirent_is_ready(struct configfs_dirent *sd)
345 spin_lock(&configfs_dirent_lock);
346 ret = !(sd->s_type & CONFIGFS_USET_CREATING);
347 spin_unlock(&configfs_dirent_lock);
352 int configfs_create_link(struct configfs_symlink *sl,
353 struct dentry *parent,
354 struct dentry *dentry)
357 umode_t mode = S_IFLNK | S_IRWXUGO;
359 err = configfs_make_dirent(parent->d_fsdata, dentry, sl, mode,
362 err = configfs_create(dentry, mode, init_symlink);
364 struct configfs_dirent *sd = dentry->d_fsdata;
366 spin_lock(&configfs_dirent_lock);
367 list_del_init(&sd->s_sibling);
368 spin_unlock(&configfs_dirent_lock);
376 static void remove_dir(struct dentry * d)
378 struct dentry * parent = dget(d->d_parent);
379 struct configfs_dirent * sd;
382 spin_lock(&configfs_dirent_lock);
383 list_del_init(&sd->s_sibling);
384 spin_unlock(&configfs_dirent_lock);
387 simple_rmdir(parent->d_inode,d);
389 pr_debug(" o %s removing done (%d)\n",d->d_name.name, d_count(d));
395 * configfs_remove_dir - remove an config_item's directory.
396 * @item: config_item we're removing.
398 * The only thing special about this is that we remove any files in
399 * the directory before we remove the directory, and we've inlined
400 * what used to be configfs_rmdir() below, instead of calling separately.
402 * Caller holds the mutex of the item's inode
405 static void configfs_remove_dir(struct config_item * item)
407 struct dentry * dentry = dget(item->ci_dentry);
414 * Drop reference from dget() on entrance.
420 /* attaches attribute's configfs_dirent to the dentry corresponding to the
423 static int configfs_attach_attr(struct configfs_dirent * sd, struct dentry * dentry)
425 struct configfs_attribute * attr = sd->s_element;
428 spin_lock(&configfs_dirent_lock);
429 dentry->d_fsdata = configfs_get(sd);
430 sd->s_dentry = dentry;
431 spin_unlock(&configfs_dirent_lock);
433 error = configfs_create(dentry, (attr->ca_mode & S_IALLUGO) | S_IFREG,
445 static struct dentry * configfs_lookup(struct inode *dir,
446 struct dentry *dentry,
449 struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
450 struct configfs_dirent * sd;
455 * Fake invisibility if dir belongs to a group/default groups hierarchy
458 * This forbids userspace to read/write attributes of items which may
459 * not complete their initialization, since the dentries of the
460 * attributes won't be instantiated.
463 if (!configfs_dirent_is_ready(parent_sd))
466 list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
467 if (sd->s_type & CONFIGFS_NOT_PINNED) {
468 const unsigned char * name = configfs_get_name(sd);
470 if (strcmp(name, dentry->d_name.name))
474 err = configfs_attach_attr(sd, dentry);
481 * If it doesn't exist and it isn't a NOT_PINNED item,
482 * it must be negative.
484 if (dentry->d_name.len > NAME_MAX)
485 return ERR_PTR(-ENAMETOOLONG);
495 * Only subdirectories count here. Files (CONFIGFS_NOT_PINNED) are
496 * attributes and are removed by rmdir(). We recurse, setting
497 * CONFIGFS_USET_DROPPING on all children that are candidates for
499 * If there is an error, the caller will reset the flags via
500 * configfs_detach_rollback().
502 static int configfs_detach_prep(struct dentry *dentry, struct mutex **wait_mutex)
504 struct configfs_dirent *parent_sd = dentry->d_fsdata;
505 struct configfs_dirent *sd;
508 /* Mark that we're trying to drop the group */
509 parent_sd->s_type |= CONFIGFS_USET_DROPPING;
512 if (!list_empty(&parent_sd->s_links))
516 list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
517 if (!sd->s_element ||
518 (sd->s_type & CONFIGFS_NOT_PINNED))
520 if (sd->s_type & CONFIGFS_USET_DEFAULT) {
521 /* Abort if racing with mkdir() */
522 if (sd->s_type & CONFIGFS_USET_IN_MKDIR) {
524 *wait_mutex = &sd->s_dentry->d_inode->i_mutex;
529 * Yup, recursive. If there's a problem, blame
530 * deep nesting of default_groups
532 ret = configfs_detach_prep(sd->s_dentry, wait_mutex);
546 * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was
549 static void configfs_detach_rollback(struct dentry *dentry)
551 struct configfs_dirent *parent_sd = dentry->d_fsdata;
552 struct configfs_dirent *sd;
554 parent_sd->s_type &= ~CONFIGFS_USET_DROPPING;
556 list_for_each_entry(sd, &parent_sd->s_children, s_sibling)
557 if (sd->s_type & CONFIGFS_USET_DEFAULT)
558 configfs_detach_rollback(sd->s_dentry);
561 static void detach_attrs(struct config_item * item)
563 struct dentry * dentry = dget(item->ci_dentry);
564 struct configfs_dirent * parent_sd;
565 struct configfs_dirent * sd, * tmp;
570 pr_debug("configfs %s: dropping attrs for dir\n",
571 dentry->d_name.name);
573 parent_sd = dentry->d_fsdata;
574 list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
575 if (!sd->s_element || !(sd->s_type & CONFIGFS_NOT_PINNED))
577 spin_lock(&configfs_dirent_lock);
578 list_del_init(&sd->s_sibling);
579 spin_unlock(&configfs_dirent_lock);
580 configfs_drop_dentry(sd, dentry);
585 * Drop reference from dget() on entrance.
590 static int populate_attrs(struct config_item *item)
592 struct config_item_type *t = item->ci_type;
593 struct configfs_attribute *attr;
600 for (i = 0; (attr = t->ct_attrs[i]) != NULL; i++) {
601 if ((error = configfs_create_file(item, attr)))
612 static int configfs_attach_group(struct config_item *parent_item,
613 struct config_item *item,
614 struct dentry *dentry);
615 static void configfs_detach_group(struct config_item *item);
617 static void detach_groups(struct config_group *group)
619 struct dentry * dentry = dget(group->cg_item.ci_dentry);
620 struct dentry *child;
621 struct configfs_dirent *parent_sd;
622 struct configfs_dirent *sd, *tmp;
627 parent_sd = dentry->d_fsdata;
628 list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
629 if (!sd->s_element ||
630 !(sd->s_type & CONFIGFS_USET_DEFAULT))
633 child = sd->s_dentry;
635 mutex_lock(&child->d_inode->i_mutex);
637 configfs_detach_group(sd->s_element);
638 child->d_inode->i_flags |= S_DEAD;
641 mutex_unlock(&child->d_inode->i_mutex);
648 * Drop reference from dget() on entrance.
654 * This fakes mkdir(2) on a default_groups[] entry. It
655 * creates a dentry, attachs it, and then does fixup
658 * We could, perhaps, tweak our parent's ->mkdir for a minute and
659 * try using vfs_mkdir. Just a thought.
661 static int create_default_group(struct config_group *parent_group,
662 struct config_group *group)
665 struct configfs_dirent *sd;
666 /* We trust the caller holds a reference to parent */
667 struct dentry *child, *parent = parent_group->cg_item.ci_dentry;
669 if (!group->cg_item.ci_name)
670 group->cg_item.ci_name = group->cg_item.ci_namebuf;
673 child = d_alloc_name(parent, group->cg_item.ci_name);
677 ret = configfs_attach_group(&parent_group->cg_item,
678 &group->cg_item, child);
680 sd = child->d_fsdata;
681 sd->s_type |= CONFIGFS_USET_DEFAULT;
683 BUG_ON(child->d_inode);
692 static int populate_groups(struct config_group *group)
694 struct config_group *new_group;
698 if (group->default_groups) {
699 for (i = 0; group->default_groups[i]; i++) {
700 new_group = group->default_groups[i];
702 ret = create_default_group(group, new_group);
704 detach_groups(group);
714 * All of link_obj/unlink_obj/link_group/unlink_group require that
715 * subsys->su_mutex is held.
718 static void unlink_obj(struct config_item *item)
720 struct config_group *group;
722 group = item->ci_group;
724 list_del_init(&item->ci_entry);
726 item->ci_group = NULL;
727 item->ci_parent = NULL;
729 /* Drop the reference for ci_entry */
730 config_item_put(item);
732 /* Drop the reference for ci_parent */
733 config_group_put(group);
737 static void link_obj(struct config_item *parent_item, struct config_item *item)
740 * Parent seems redundant with group, but it makes certain
741 * traversals much nicer.
743 item->ci_parent = parent_item;
746 * We hold a reference on the parent for the child's ci_parent
749 item->ci_group = config_group_get(to_config_group(parent_item));
750 list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
753 * We hold a reference on the child for ci_entry on the parent's
756 config_item_get(item);
759 static void unlink_group(struct config_group *group)
762 struct config_group *new_group;
764 if (group->default_groups) {
765 for (i = 0; group->default_groups[i]; i++) {
766 new_group = group->default_groups[i];
767 unlink_group(new_group);
771 group->cg_subsys = NULL;
772 unlink_obj(&group->cg_item);
775 static void link_group(struct config_group *parent_group, struct config_group *group)
778 struct config_group *new_group;
779 struct configfs_subsystem *subsys = NULL; /* gcc is a turd */
781 link_obj(&parent_group->cg_item, &group->cg_item);
783 if (parent_group->cg_subsys)
784 subsys = parent_group->cg_subsys;
785 else if (configfs_is_root(&parent_group->cg_item))
786 subsys = to_configfs_subsystem(group);
789 group->cg_subsys = subsys;
791 if (group->default_groups) {
792 for (i = 0; group->default_groups[i]; i++) {
793 new_group = group->default_groups[i];
794 link_group(group, new_group);
800 * The goal is that configfs_attach_item() (and
801 * configfs_attach_group()) can be called from either the VFS or this
802 * module. That is, they assume that the items have been created,
803 * the dentry allocated, and the dcache is all ready to go.
805 * If they fail, they must clean up after themselves as if they
806 * had never been called. The caller (VFS or local function) will
807 * handle cleaning up the dcache bits.
809 * configfs_detach_group() and configfs_detach_item() behave similarly on
810 * the way out. They assume that the proper semaphores are held, they
811 * clean up the configfs items, and they expect their callers will
812 * handle the dcache bits.
814 static int configfs_attach_item(struct config_item *parent_item,
815 struct config_item *item,
816 struct dentry *dentry)
820 ret = configfs_create_dir(item, dentry);
822 ret = populate_attrs(item);
825 * We are going to remove an inode and its dentry but
826 * the VFS may already have hit and used them. Thus,
827 * we must lock them as rmdir() would.
829 mutex_lock(&dentry->d_inode->i_mutex);
830 configfs_remove_dir(item);
831 dentry->d_inode->i_flags |= S_DEAD;
833 mutex_unlock(&dentry->d_inode->i_mutex);
841 /* Caller holds the mutex of the item's inode */
842 static void configfs_detach_item(struct config_item *item)
845 configfs_remove_dir(item);
848 static int configfs_attach_group(struct config_item *parent_item,
849 struct config_item *item,
850 struct dentry *dentry)
853 struct configfs_dirent *sd;
855 ret = configfs_attach_item(parent_item, item, dentry);
857 sd = dentry->d_fsdata;
858 sd->s_type |= CONFIGFS_USET_DIR;
861 * FYI, we're faking mkdir in populate_groups()
862 * We must lock the group's inode to avoid races with the VFS
863 * which can already hit the inode and try to add/remove entries
866 * We must also lock the inode to remove it safely in case of
867 * error, as rmdir() would.
869 mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD);
870 configfs_adjust_dir_dirent_depth_before_populate(sd);
871 ret = populate_groups(to_config_group(item));
873 configfs_detach_item(item);
874 dentry->d_inode->i_flags |= S_DEAD;
877 configfs_adjust_dir_dirent_depth_after_populate(sd);
878 mutex_unlock(&dentry->d_inode->i_mutex);
886 /* Caller holds the mutex of the group's inode */
887 static void configfs_detach_group(struct config_item *item)
889 detach_groups(to_config_group(item));
890 configfs_detach_item(item);
894 * After the item has been detached from the filesystem view, we are
895 * ready to tear it out of the hierarchy. Notify the client before
896 * we do that so they can perform any cleanup that requires
897 * navigating the hierarchy. A client does not need to provide this
898 * callback. The subsystem semaphore MUST be held by the caller, and
899 * references must be valid for both items. It also assumes the
900 * caller has validated ci_type.
902 static void client_disconnect_notify(struct config_item *parent_item,
903 struct config_item *item)
905 struct config_item_type *type;
907 type = parent_item->ci_type;
910 if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
911 type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
916 * Drop the initial reference from make_item()/make_group()
917 * This function assumes that reference is held on item
918 * and that item holds a valid reference to the parent. Also, it
919 * assumes the caller has validated ci_type.
921 static void client_drop_item(struct config_item *parent_item,
922 struct config_item *item)
924 struct config_item_type *type;
926 type = parent_item->ci_type;
930 * If ->drop_item() exists, it is responsible for the
933 if (type->ct_group_ops && type->ct_group_ops->drop_item)
934 type->ct_group_ops->drop_item(to_config_group(parent_item),
937 config_item_put(item);
941 static void configfs_dump_one(struct configfs_dirent *sd, int level)
943 printk(KERN_INFO "%*s\"%s\":\n", level, " ", configfs_get_name(sd));
945 #define type_print(_type) if (sd->s_type & _type) printk(KERN_INFO "%*s %s\n", level, " ", #_type);
946 type_print(CONFIGFS_ROOT);
947 type_print(CONFIGFS_DIR);
948 type_print(CONFIGFS_ITEM_ATTR);
949 type_print(CONFIGFS_ITEM_LINK);
950 type_print(CONFIGFS_USET_DIR);
951 type_print(CONFIGFS_USET_DEFAULT);
952 type_print(CONFIGFS_USET_DROPPING);
956 static int configfs_dump(struct configfs_dirent *sd, int level)
958 struct configfs_dirent *child_sd;
961 configfs_dump_one(sd, level);
963 if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
966 list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
967 ret = configfs_dump(child_sd, level + 2);
978 * configfs_depend_item() and configfs_undepend_item()
980 * WARNING: Do not call these from a configfs callback!
982 * This describes these functions and their helpers.
984 * Allow another kernel system to depend on a config_item. If this
985 * happens, the item cannot go away until the dependent can live without
986 * it. The idea is to give client modules as simple an interface as
987 * possible. When a system asks them to depend on an item, they just
988 * call configfs_depend_item(). If the item is live and the client
989 * driver is in good shape, we'll happily do the work for them.
991 * Why is the locking complex? Because configfs uses the VFS to handle
992 * all locking, but this function is called outside the normal
993 * VFS->configfs path. So it must take VFS locks to prevent the
994 * VFS->configfs stuff (configfs_mkdir(), configfs_rmdir(), etc). This is
995 * why you can't call these functions underneath configfs callbacks.
997 * Note, btw, that this can be called at *any* time, even when a configfs
998 * subsystem isn't registered, or when configfs is loading or unloading.
999 * Just like configfs_register_subsystem(). So we take the same
1000 * precautions. We pin the filesystem. We lock configfs_dirent_lock.
1001 * If we can find the target item in the
1002 * configfs tree, it must be part of the subsystem tree as well, so we
1003 * do not need the subsystem semaphore. Holding configfs_dirent_lock helps
1004 * locking out mkdir() and rmdir(), who might be racing us.
1008 * configfs_depend_prep()
1010 * Only subdirectories count here. Files (CONFIGFS_NOT_PINNED) are
1011 * attributes. This is similar but not the same to configfs_detach_prep().
1012 * Note that configfs_detach_prep() expects the parent to be locked when it
1013 * is called, but we lock the parent *inside* configfs_depend_prep(). We
1014 * do that so we can unlock it if we find nothing.
1016 * Here we do a depth-first search of the dentry hierarchy looking for
1018 * We deliberately ignore items tagged as dropping since they are virtually
1019 * dead, as well as items in the middle of attachment since they virtually
1020 * do not exist yet. This completes the locking out of racing mkdir() and
1022 * Note: subdirectories in the middle of attachment start with s_type =
1023 * CONFIGFS_DIR|CONFIGFS_USET_CREATING set by create_dir(). When
1024 * CONFIGFS_USET_CREATING is set, we ignore the item. The actual set of
1025 * s_type is in configfs_new_dirent(), which has configfs_dirent_lock.
1027 * If the target is not found, -ENOENT is bubbled up.
1029 * This adds a requirement that all config_items be unique!
1031 * This is recursive. There isn't
1032 * much on the stack, though, so folks that need this function - be careful
1033 * about your stack! Patches will be accepted to make it iterative.
1035 static int configfs_depend_prep(struct dentry *origin,
1036 struct config_item *target)
1038 struct configfs_dirent *child_sd, *sd;
1041 BUG_ON(!origin || !origin->d_fsdata);
1042 sd = origin->d_fsdata;
1044 if (sd->s_element == target) /* Boo-yah */
1047 list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
1048 if ((child_sd->s_type & CONFIGFS_DIR) &&
1049 !(child_sd->s_type & CONFIGFS_USET_DROPPING) &&
1050 !(child_sd->s_type & CONFIGFS_USET_CREATING)) {
1051 ret = configfs_depend_prep(child_sd->s_dentry,
1054 goto out; /* Child path boo-yah */
1058 /* We looped all our children and didn't find target */
1065 int configfs_depend_item(struct configfs_subsystem *subsys,
1066 struct config_item *target)
1069 struct configfs_dirent *p, *root_sd, *subsys_sd = NULL;
1070 struct config_item *s_item = &subsys->su_group.cg_item;
1071 struct dentry *root;
1074 * Pin the configfs filesystem. This means we can safely access
1075 * the root of the configfs filesystem.
1077 root = configfs_pin_fs();
1079 return PTR_ERR(root);
1082 * Next, lock the root directory. We're going to check that the
1083 * subsystem is really registered, and so we need to lock out
1084 * configfs_[un]register_subsystem().
1086 mutex_lock(&root->d_inode->i_mutex);
1088 root_sd = root->d_fsdata;
1090 list_for_each_entry(p, &root_sd->s_children, s_sibling) {
1091 if (p->s_type & CONFIGFS_DIR) {
1092 if (p->s_element == s_item) {
1104 /* Ok, now we can trust subsys/s_item */
1106 spin_lock(&configfs_dirent_lock);
1107 /* Scan the tree, return 0 if found */
1108 ret = configfs_depend_prep(subsys_sd->s_dentry, target);
1110 goto out_unlock_dirent_lock;
1113 * We are sure that the item is not about to be removed by rmdir(), and
1114 * not in the middle of attachment by mkdir().
1116 p = target->ci_dentry->d_fsdata;
1117 p->s_dependent_count += 1;
1119 out_unlock_dirent_lock:
1120 spin_unlock(&configfs_dirent_lock);
1122 mutex_unlock(&root->d_inode->i_mutex);
1125 * If we succeeded, the fs is pinned via other methods. If not,
1126 * we're done with it anyway. So release_fs() is always right.
1128 configfs_release_fs();
1132 EXPORT_SYMBOL(configfs_depend_item);
1135 * Release the dependent linkage. This is much simpler than
1136 * configfs_depend_item() because we know that that the client driver is
1137 * pinned, thus the subsystem is pinned, and therefore configfs is pinned.
1139 void configfs_undepend_item(struct configfs_subsystem *subsys,
1140 struct config_item *target)
1142 struct configfs_dirent *sd;
1145 * Since we can trust everything is pinned, we just need
1146 * configfs_dirent_lock.
1148 spin_lock(&configfs_dirent_lock);
1150 sd = target->ci_dentry->d_fsdata;
1151 BUG_ON(sd->s_dependent_count < 1);
1153 sd->s_dependent_count -= 1;
1156 * After this unlock, we cannot trust the item to stay alive!
1157 * DO NOT REFERENCE item after this unlock.
1159 spin_unlock(&configfs_dirent_lock);
1161 EXPORT_SYMBOL(configfs_undepend_item);
1163 static int configfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1167 struct config_group *group = NULL;
1168 struct config_item *item = NULL;
1169 struct config_item *parent_item;
1170 struct configfs_subsystem *subsys;
1171 struct configfs_dirent *sd;
1172 struct config_item_type *type;
1173 struct module *subsys_owner = NULL, *new_item_owner = NULL;
1176 sd = dentry->d_parent->d_fsdata;
1179 * Fake invisibility if dir belongs to a group/default groups hierarchy
1182 if (!configfs_dirent_is_ready(sd)) {
1187 if (!(sd->s_type & CONFIGFS_USET_DIR)) {
1192 /* Get a working ref for the duration of this function */
1193 parent_item = configfs_get_config_item(dentry->d_parent);
1194 type = parent_item->ci_type;
1195 subsys = to_config_group(parent_item)->cg_subsys;
1198 if (!type || !type->ct_group_ops ||
1199 (!type->ct_group_ops->make_group &&
1200 !type->ct_group_ops->make_item)) {
1201 ret = -EPERM; /* Lack-of-mkdir returns -EPERM */
1206 * The subsystem may belong to a different module than the item
1207 * being created. We don't want to safely pin the new item but
1208 * fail to pin the subsystem it sits under.
1210 if (!subsys->su_group.cg_item.ci_type) {
1214 subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1215 if (!try_module_get(subsys_owner)) {
1220 name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
1223 goto out_subsys_put;
1226 snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
1228 mutex_lock(&subsys->su_mutex);
1229 if (type->ct_group_ops->make_group) {
1230 group = type->ct_group_ops->make_group(to_config_group(parent_item), name);
1232 group = ERR_PTR(-ENOMEM);
1233 if (!IS_ERR(group)) {
1234 link_group(to_config_group(parent_item), group);
1235 item = &group->cg_item;
1237 ret = PTR_ERR(group);
1239 item = type->ct_group_ops->make_item(to_config_group(parent_item), name);
1241 item = ERR_PTR(-ENOMEM);
1243 link_obj(parent_item, item);
1245 ret = PTR_ERR(item);
1247 mutex_unlock(&subsys->su_mutex);
1252 * If ret != 0, then link_obj() was never called.
1253 * There are no extra references to clean up.
1255 goto out_subsys_put;
1259 * link_obj() has been called (via link_group() for groups).
1260 * From here on out, errors must clean that up.
1263 type = item->ci_type;
1269 new_item_owner = type->ct_owner;
1270 if (!try_module_get(new_item_owner)) {
1276 * I hate doing it this way, but if there is
1277 * an error, module_put() probably should
1278 * happen after any cleanup.
1283 * Make racing rmdir() fail if it did not tag parent with
1284 * CONFIGFS_USET_DROPPING
1285 * Note: if CONFIGFS_USET_DROPPING is already set, attach_group() will
1286 * fail and let rmdir() terminate correctly
1288 spin_lock(&configfs_dirent_lock);
1289 /* This will make configfs_detach_prep() fail */
1290 sd->s_type |= CONFIGFS_USET_IN_MKDIR;
1291 spin_unlock(&configfs_dirent_lock);
1294 ret = configfs_attach_group(parent_item, item, dentry);
1296 ret = configfs_attach_item(parent_item, item, dentry);
1298 spin_lock(&configfs_dirent_lock);
1299 sd->s_type &= ~CONFIGFS_USET_IN_MKDIR;
1301 configfs_dir_set_ready(dentry->d_fsdata);
1302 spin_unlock(&configfs_dirent_lock);
1306 /* Tear down everything we built up */
1307 mutex_lock(&subsys->su_mutex);
1309 client_disconnect_notify(parent_item, item);
1311 unlink_group(group);
1314 client_drop_item(parent_item, item);
1316 mutex_unlock(&subsys->su_mutex);
1319 module_put(new_item_owner);
1324 module_put(subsys_owner);
1328 * link_obj()/link_group() took a reference from child->parent,
1329 * so the parent is safely pinned. We can drop our working
1332 config_item_put(parent_item);
1338 static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
1340 struct config_item *parent_item;
1341 struct config_item *item;
1342 struct configfs_subsystem *subsys;
1343 struct configfs_dirent *sd;
1344 struct module *subsys_owner = NULL, *dead_item_owner = NULL;
1347 sd = dentry->d_fsdata;
1348 if (sd->s_type & CONFIGFS_USET_DEFAULT)
1351 /* Get a working ref until we have the child */
1352 parent_item = configfs_get_config_item(dentry->d_parent);
1353 subsys = to_config_group(parent_item)->cg_subsys;
1356 if (!parent_item->ci_type) {
1357 config_item_put(parent_item);
1361 /* configfs_mkdir() shouldn't have allowed this */
1362 BUG_ON(!subsys->su_group.cg_item.ci_type);
1363 subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1366 * Ensure that no racing symlink() will make detach_prep() fail while
1367 * the new link is temporarily attached
1370 struct mutex *wait_mutex;
1372 mutex_lock(&configfs_symlink_mutex);
1373 spin_lock(&configfs_dirent_lock);
1375 * Here's where we check for dependents. We're protected by
1376 * configfs_dirent_lock.
1377 * If no dependent, atomically tag the item as dropping.
1379 ret = sd->s_dependent_count ? -EBUSY : 0;
1381 ret = configfs_detach_prep(dentry, &wait_mutex);
1383 configfs_detach_rollback(dentry);
1385 spin_unlock(&configfs_dirent_lock);
1386 mutex_unlock(&configfs_symlink_mutex);
1389 if (ret != -EAGAIN) {
1390 config_item_put(parent_item);
1394 /* Wait until the racing operation terminates */
1395 mutex_lock(wait_mutex);
1396 mutex_unlock(wait_mutex);
1398 } while (ret == -EAGAIN);
1400 /* Get a working ref for the duration of this function */
1401 item = configfs_get_config_item(dentry);
1403 /* Drop reference from above, item already holds one. */
1404 config_item_put(parent_item);
1407 dead_item_owner = item->ci_type->ct_owner;
1409 if (sd->s_type & CONFIGFS_USET_DIR) {
1410 configfs_detach_group(item);
1412 mutex_lock(&subsys->su_mutex);
1413 client_disconnect_notify(parent_item, item);
1414 unlink_group(to_config_group(item));
1416 configfs_detach_item(item);
1418 mutex_lock(&subsys->su_mutex);
1419 client_disconnect_notify(parent_item, item);
1423 client_drop_item(parent_item, item);
1424 mutex_unlock(&subsys->su_mutex);
1426 /* Drop our reference from above */
1427 config_item_put(item);
1429 module_put(dead_item_owner);
1430 module_put(subsys_owner);
1435 const struct inode_operations configfs_dir_inode_operations = {
1436 .mkdir = configfs_mkdir,
1437 .rmdir = configfs_rmdir,
1438 .symlink = configfs_symlink,
1439 .unlink = configfs_unlink,
1440 .lookup = configfs_lookup,
1441 .setattr = configfs_setattr,
1444 const struct inode_operations configfs_root_inode_operations = {
1445 .lookup = configfs_lookup,
1446 .setattr = configfs_setattr,
1450 int configfs_rename_dir(struct config_item * item, const char *new_name)
1453 struct dentry * new_dentry, * parent;
1455 if (!strcmp(config_item_name(item), new_name))
1461 down_write(&configfs_rename_sem);
1462 parent = item->parent->dentry;
1464 mutex_lock(&parent->d_inode->i_mutex);
1466 new_dentry = lookup_one_len(new_name, parent, strlen(new_name));
1467 if (!IS_ERR(new_dentry)) {
1468 if (!new_dentry->d_inode) {
1469 error = config_item_set_name(item, "%s", new_name);
1471 d_add(new_dentry, NULL);
1472 d_move(item->dentry, new_dentry);
1475 d_delete(new_dentry);
1480 mutex_unlock(&parent->d_inode->i_mutex);
1481 up_write(&configfs_rename_sem);
1487 static int configfs_dir_open(struct inode *inode, struct file *file)
1489 struct dentry * dentry = file->f_path.dentry;
1490 struct configfs_dirent * parent_sd = dentry->d_fsdata;
1493 mutex_lock(&dentry->d_inode->i_mutex);
1495 * Fake invisibility if dir belongs to a group/default groups hierarchy
1499 if (configfs_dirent_is_ready(parent_sd)) {
1500 file->private_data = configfs_new_dirent(parent_sd, NULL, 0);
1501 if (IS_ERR(file->private_data))
1502 err = PTR_ERR(file->private_data);
1506 mutex_unlock(&dentry->d_inode->i_mutex);
1511 static int configfs_dir_close(struct inode *inode, struct file *file)
1513 struct dentry * dentry = file->f_path.dentry;
1514 struct configfs_dirent * cursor = file->private_data;
1516 mutex_lock(&dentry->d_inode->i_mutex);
1517 spin_lock(&configfs_dirent_lock);
1518 list_del_init(&cursor->s_sibling);
1519 spin_unlock(&configfs_dirent_lock);
1520 mutex_unlock(&dentry->d_inode->i_mutex);
1522 release_configfs_dirent(cursor);
1527 /* Relationship between s_mode and the DT_xxx types */
1528 static inline unsigned char dt_type(struct configfs_dirent *sd)
1530 return (sd->s_mode >> 12) & 15;
1533 static int configfs_readdir(struct file *file, struct dir_context *ctx)
1535 struct dentry *dentry = file->f_path.dentry;
1536 struct super_block *sb = dentry->d_sb;
1537 struct configfs_dirent * parent_sd = dentry->d_fsdata;
1538 struct configfs_dirent *cursor = file->private_data;
1539 struct list_head *p, *q = &cursor->s_sibling;
1542 if (!dir_emit_dots(file, ctx))
1544 if (ctx->pos == 2) {
1545 spin_lock(&configfs_dirent_lock);
1546 list_move(q, &parent_sd->s_children);
1547 spin_unlock(&configfs_dirent_lock);
1549 for (p = q->next; p != &parent_sd->s_children; p = p->next) {
1550 struct configfs_dirent *next;
1553 struct inode *inode = NULL;
1555 next = list_entry(p, struct configfs_dirent, s_sibling);
1556 if (!next->s_element)
1559 name = configfs_get_name(next);
1563 * We'll have a dentry and an inode for
1564 * PINNED items and for open attribute
1565 * files. We lock here to prevent a race
1566 * with configfs_d_iput() clearing
1567 * s_dentry before calling iput().
1569 * Why do we go to the trouble? If
1570 * someone has an attribute file open,
1571 * the inode number should match until
1572 * they close it. Beyond that, we don't
1575 spin_lock(&configfs_dirent_lock);
1576 dentry = next->s_dentry;
1578 inode = dentry->d_inode;
1581 spin_unlock(&configfs_dirent_lock);
1583 ino = iunique(sb, 2);
1585 if (!dir_emit(ctx, name, len, ino, dt_type(next)))
1588 spin_lock(&configfs_dirent_lock);
1590 spin_unlock(&configfs_dirent_lock);
1597 static loff_t configfs_dir_lseek(struct file *file, loff_t offset, int whence)
1599 struct dentry * dentry = file->f_path.dentry;
1601 mutex_lock(&dentry->d_inode->i_mutex);
1604 offset += file->f_pos;
1609 mutex_unlock(&file_inode(file)->i_mutex);
1612 if (offset != file->f_pos) {
1613 file->f_pos = offset;
1614 if (file->f_pos >= 2) {
1615 struct configfs_dirent *sd = dentry->d_fsdata;
1616 struct configfs_dirent *cursor = file->private_data;
1617 struct list_head *p;
1618 loff_t n = file->f_pos - 2;
1620 spin_lock(&configfs_dirent_lock);
1621 list_del(&cursor->s_sibling);
1622 p = sd->s_children.next;
1623 while (n && p != &sd->s_children) {
1624 struct configfs_dirent *next;
1625 next = list_entry(p, struct configfs_dirent,
1627 if (next->s_element)
1631 list_add_tail(&cursor->s_sibling, p);
1632 spin_unlock(&configfs_dirent_lock);
1635 mutex_unlock(&dentry->d_inode->i_mutex);
1639 const struct file_operations configfs_dir_operations = {
1640 .open = configfs_dir_open,
1641 .release = configfs_dir_close,
1642 .llseek = configfs_dir_lseek,
1643 .read = generic_read_dir,
1644 .iterate = configfs_readdir,
1647 int configfs_register_subsystem(struct configfs_subsystem *subsys)
1650 struct config_group *group = &subsys->su_group;
1651 struct dentry *dentry;
1652 struct dentry *root;
1653 struct configfs_dirent *sd;
1655 root = configfs_pin_fs();
1657 return PTR_ERR(root);
1659 if (!group->cg_item.ci_name)
1660 group->cg_item.ci_name = group->cg_item.ci_namebuf;
1662 sd = root->d_fsdata;
1663 link_group(to_config_group(sd->s_element), group);
1665 mutex_lock_nested(&root->d_inode->i_mutex, I_MUTEX_PARENT);
1668 dentry = d_alloc_name(root, group->cg_item.ci_name);
1670 d_add(dentry, NULL);
1672 err = configfs_attach_group(sd->s_element, &group->cg_item,
1675 BUG_ON(dentry->d_inode);
1679 spin_lock(&configfs_dirent_lock);
1680 configfs_dir_set_ready(dentry->d_fsdata);
1681 spin_unlock(&configfs_dirent_lock);
1685 mutex_unlock(&root->d_inode->i_mutex);
1688 unlink_group(group);
1689 configfs_release_fs();
1695 void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
1697 struct config_group *group = &subsys->su_group;
1698 struct dentry *dentry = group->cg_item.ci_dentry;
1699 struct dentry *root = dentry->d_sb->s_root;
1701 if (dentry->d_parent != root) {
1702 printk(KERN_ERR "configfs: Tried to unregister non-subsystem!\n");
1706 mutex_lock_nested(&root->d_inode->i_mutex,
1708 mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD);
1709 mutex_lock(&configfs_symlink_mutex);
1710 spin_lock(&configfs_dirent_lock);
1711 if (configfs_detach_prep(dentry, NULL)) {
1712 printk(KERN_ERR "configfs: Tried to unregister non-empty subsystem!\n");
1714 spin_unlock(&configfs_dirent_lock);
1715 mutex_unlock(&configfs_symlink_mutex);
1716 configfs_detach_group(&group->cg_item);
1717 dentry->d_inode->i_flags |= S_DEAD;
1719 mutex_unlock(&dentry->d_inode->i_mutex);
1723 mutex_unlock(&root->d_inode->i_mutex);
1727 unlink_group(group);
1728 configfs_release_fs();
1731 EXPORT_SYMBOL(configfs_register_subsystem);
1732 EXPORT_SYMBOL(configfs_unregister_subsystem);