1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dir.c - Operations for configfs directories.
5 *
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.
10 *
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.
15 *
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.
20 *
21 * Based on sysfs:
22 * 	sysfs is Copyright (C) 2001, 2002, 2003 Patrick Mochel
23 *
24 * configfs Copyright (C) 2005 Oracle.  All rights reserved.
25 */
26
27#undef DEBUG
28
29#include <linux/fs.h>
30#include <linux/mount.h>
31#include <linux/module.h>
32#include <linux/slab.h>
33#include <linux/err.h>
34
35#include <linux/configfs.h>
36#include "configfs_internal.h"
37
38DECLARE_RWSEM(configfs_rename_sem);
39/*
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.
46 *
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()
50 */
51DEFINE_SPINLOCK(configfs_dirent_lock);
52
53static void configfs_d_iput(struct dentry * dentry,
54			    struct inode * inode)
55{
56	struct configfs_dirent *sd = dentry->d_fsdata;
57
58	if (sd) {
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
64		 * sd owner.
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.
68		 */
69		if (atomic_read(&sd->s_count) <= 2)
70			sd->s_dentry = NULL;
71
72		spin_unlock(&configfs_dirent_lock);
73		configfs_put(sd);
74	}
75	iput(inode);
76}
77
78const struct dentry_operations configfs_dentry_ops = {
79	.d_iput		= configfs_d_iput,
80	.d_delete	= always_delete_dentry,
81};
82
83#ifdef CONFIG_LOCKDEP
84
85/*
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.
90 *
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].
95 *
96 * The lock classes are declared and assigned in inode.c, according to the
97 * s_depth value.
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].
102 */
103
104static void configfs_init_dirent_depth(struct configfs_dirent *sd)
105{
106	sd->s_depth = -1;
107}
108
109static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
110					  struct configfs_dirent *sd)
111{
112	int parent_depth = parent_sd->s_depth;
113
114	if (parent_depth >= 0)
115		sd->s_depth = parent_depth + 1;
116}
117
118static void
119configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
120{
121	/*
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.
125	 */
126	/*
127	 *  sd->s_depth == -1 iff we are a non default group.
128	 *  else (we are a default group) sd->s_depth > 0 (see
129	 *  create_dir()).
130	 */
131	if (sd->s_depth == -1)
132		/*
133		 * We are a non default group and we are going to create
134		 * default groups.
135		 */
136		sd->s_depth = 0;
137}
138
139static void
140configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
141{
142	/* We will not create default groups anymore. */
143	sd->s_depth = -1;
144}
145
146#else /* CONFIG_LOCKDEP */
147
148static void configfs_init_dirent_depth(struct configfs_dirent *sd)
149{
150}
151
152static void configfs_set_dir_dirent_depth(struct configfs_dirent *parent_sd,
153					  struct configfs_dirent *sd)
154{
155}
156
157static void
158configfs_adjust_dir_dirent_depth_before_populate(struct configfs_dirent *sd)
159{
160}
161
162static void
163configfs_adjust_dir_dirent_depth_after_populate(struct configfs_dirent *sd)
164{
165}
166
167#endif /* CONFIG_LOCKDEP */
168
169/*
170 * Allocates a new configfs_dirent and links it to the parent configfs_dirent
171 */
172static struct configfs_dirent *configfs_new_dirent(struct configfs_dirent *parent_sd,
173						   void *element, int type)
174{
175	struct configfs_dirent * sd;
176
177	sd = kmem_cache_zalloc(configfs_dir_cachep, GFP_KERNEL);
178	if (!sd)
179		return ERR_PTR(-ENOMEM);
180
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;
185	sd->s_type = type;
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);
192	}
193	list_add(&sd->s_sibling, &parent_sd->s_children);
194	spin_unlock(&configfs_dirent_lock);
195
196	return sd;
197}
198
199/*
200 *
201 * Return -EEXIST if there is already a configfs element with the same
202 * name for the same parent.
203 *
204 * called with parent inode's i_mutex held
205 */
206static int configfs_dirent_exists(struct configfs_dirent *parent_sd,
207				  const unsigned char *new)
208{
209	struct configfs_dirent * sd;
210
211	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
212		if (sd->s_element) {
213			const unsigned char *existing = configfs_get_name(sd);
214			if (strcmp(existing, new))
215				continue;
216			else
217				return -EEXIST;
218		}
219	}
220
221	return 0;
222}
223
224
225int configfs_make_dirent(struct configfs_dirent * parent_sd,
226			 struct dentry * dentry, void * element,
227			 umode_t mode, int type)
228{
229	struct configfs_dirent * sd;
230
231	sd = configfs_new_dirent(parent_sd, element, type);
232	if (IS_ERR(sd))
233		return PTR_ERR(sd);
234
235	sd->s_mode = mode;
236	sd->s_dentry = dentry;
237	if (dentry)
238		dentry->d_fsdata = configfs_get(sd);
239
240	return 0;
241}
242
243static int init_dir(struct inode * inode)
244{
245	inode->i_op = &configfs_dir_inode_operations;
246	inode->i_fop = &configfs_dir_operations;
247
248	/* directory inodes start off with i_nlink == 2 (for "." entry) */
249	inc_nlink(inode);
250	return 0;
251}
252
253static int configfs_init_file(struct inode * inode)
254{
255	inode->i_size = PAGE_SIZE;
256	inode->i_fop = &configfs_file_operations;
257	return 0;
258}
259
260static int init_symlink(struct inode * inode)
261{
262	inode->i_op = &configfs_symlink_inode_operations;
263	return 0;
264}
265
266static int create_dir(struct config_item *k, struct dentry *d)
267{
268	int error;
269	umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
270	struct dentry *p = d->d_parent;
271
272	BUG_ON(!k);
273
274	error = configfs_dirent_exists(p->d_fsdata, d->d_name.name);
275	if (!error)
276		error = configfs_make_dirent(p->d_fsdata, d, k, mode,
277					     CONFIGFS_DIR | CONFIGFS_USET_CREATING);
278	if (!error) {
279		configfs_set_dir_dirent_depth(p->d_fsdata, d->d_fsdata);
280		error = configfs_create(d, mode, init_dir);
281		if (!error) {
282			inc_nlink(p->d_inode);
283		} else {
284			struct configfs_dirent *sd = d->d_fsdata;
285			if (sd) {
286				spin_lock(&configfs_dirent_lock);
287				list_del_init(&sd->s_sibling);
288				spin_unlock(&configfs_dirent_lock);
289				configfs_put(sd);
290			}
291		}
292	}
293	return error;
294}
295
296
297/**
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.
301 *
302 *	Note: user-created entries won't be allowed under this new directory
303 *	until it is validated by configfs_dir_set_ready()
304 */
305
306static int configfs_create_dir(struct config_item * item, struct dentry *dentry)
307{
308	int error = create_dir(item, dentry);
309	if (!error)
310		item->ci_dentry = dentry;
311	return error;
312}
313
314/*
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
318 *
319 * Caller must hold configfs_dirent_lock.
320 */
321static void configfs_dir_set_ready(struct configfs_dirent *sd)
322{
323	struct configfs_dirent *child_sd;
324
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);
329}
330
331/*
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
335 *
336 * @return	non-zero iff the directory was validated
337 *
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.
340 */
341int configfs_dirent_is_ready(struct configfs_dirent *sd)
342{
343	int ret;
344
345	spin_lock(&configfs_dirent_lock);
346	ret = !(sd->s_type & CONFIGFS_USET_CREATING);
347	spin_unlock(&configfs_dirent_lock);
348
349	return ret;
350}
351
352int configfs_create_link(struct configfs_symlink *sl,
353			 struct dentry *parent,
354			 struct dentry *dentry)
355{
356	int err = 0;
357	umode_t mode = S_IFLNK | S_IRWXUGO;
358
359	err = configfs_make_dirent(parent->d_fsdata, dentry, sl, mode,
360				   CONFIGFS_ITEM_LINK);
361	if (!err) {
362		err = configfs_create(dentry, mode, init_symlink);
363		if (err) {
364			struct configfs_dirent *sd = dentry->d_fsdata;
365			if (sd) {
366				spin_lock(&configfs_dirent_lock);
367				list_del_init(&sd->s_sibling);
368				spin_unlock(&configfs_dirent_lock);
369				configfs_put(sd);
370			}
371		}
372	}
373	return err;
374}
375
376static void remove_dir(struct dentry * d)
377{
378	struct dentry * parent = dget(d->d_parent);
379	struct configfs_dirent * sd;
380
381	sd = d->d_fsdata;
382	spin_lock(&configfs_dirent_lock);
383	list_del_init(&sd->s_sibling);
384	spin_unlock(&configfs_dirent_lock);
385	configfs_put(sd);
386	if (d->d_inode)
387		simple_rmdir(parent->d_inode,d);
388
389	pr_debug(" o %s removing done (%d)\n",d->d_name.name, d_count(d));
390
391	dput(parent);
392}
393
394/**
395 * configfs_remove_dir - remove an config_item's directory.
396 * @item:	config_item we're removing.
397 *
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.
401 *
402 * Caller holds the mutex of the item's inode
403 */
404
405static void configfs_remove_dir(struct config_item * item)
406{
407	struct dentry * dentry = dget(item->ci_dentry);
408
409	if (!dentry)
410		return;
411
412	remove_dir(dentry);
413	/**
414	 * Drop reference from dget() on entrance.
415	 */
416	dput(dentry);
417}
418
419
420/* attaches attribute's configfs_dirent to the dentry corresponding to the
421 * attribute file
422 */
423static int configfs_attach_attr(struct configfs_dirent * sd, struct dentry * dentry)
424{
425	struct configfs_attribute * attr = sd->s_element;
426	int error;
427
428	spin_lock(&configfs_dirent_lock);
429	dentry->d_fsdata = configfs_get(sd);
430	sd->s_dentry = dentry;
431	spin_unlock(&configfs_dirent_lock);
432
433	error = configfs_create(dentry, (attr->ca_mode & S_IALLUGO) | S_IFREG,
434				configfs_init_file);
435	if (error) {
436		configfs_put(sd);
437		return error;
438	}
439
440	d_rehash(dentry);
441
442	return 0;
443}
444
445static struct dentry * configfs_lookup(struct inode *dir,
446				       struct dentry *dentry,
447				       unsigned int flags)
448{
449	struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
450	struct configfs_dirent * sd;
451	int found = 0;
452	int err;
453
454	/*
455	 * Fake invisibility if dir belongs to a group/default groups hierarchy
456	 * being attached
457	 *
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.
461	 */
462	err = -ENOENT;
463	if (!configfs_dirent_is_ready(parent_sd))
464		goto out;
465
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);
469
470			if (strcmp(name, dentry->d_name.name))
471				continue;
472
473			found = 1;
474			err = configfs_attach_attr(sd, dentry);
475			break;
476		}
477	}
478
479	if (!found) {
480		/*
481		 * If it doesn't exist and it isn't a NOT_PINNED item,
482		 * it must be negative.
483		 */
484		if (dentry->d_name.len > NAME_MAX)
485			return ERR_PTR(-ENAMETOOLONG);
486		d_add(dentry, NULL);
487		return NULL;
488	}
489
490out:
491	return ERR_PTR(err);
492}
493
494/*
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
498 * default detach.
499 * If there is an error, the caller will reset the flags via
500 * configfs_detach_rollback().
501 */
502static int configfs_detach_prep(struct dentry *dentry, struct mutex **wait_mutex)
503{
504	struct configfs_dirent *parent_sd = dentry->d_fsdata;
505	struct configfs_dirent *sd;
506	int ret;
507
508	/* Mark that we're trying to drop the group */
509	parent_sd->s_type |= CONFIGFS_USET_DROPPING;
510
511	ret = -EBUSY;
512	if (!list_empty(&parent_sd->s_links))
513		goto out;
514
515	ret = 0;
516	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
517		if (!sd->s_element ||
518		    (sd->s_type & CONFIGFS_NOT_PINNED))
519			continue;
520		if (sd->s_type & CONFIGFS_USET_DEFAULT) {
521			/* Abort if racing with mkdir() */
522			if (sd->s_type & CONFIGFS_USET_IN_MKDIR) {
523				if (wait_mutex)
524					*wait_mutex = &sd->s_dentry->d_inode->i_mutex;
525				return -EAGAIN;
526			}
527
528			/*
529			 * Yup, recursive.  If there's a problem, blame
530			 * deep nesting of default_groups
531			 */
532			ret = configfs_detach_prep(sd->s_dentry, wait_mutex);
533			if (!ret)
534				continue;
535		} else
536			ret = -ENOTEMPTY;
537
538		break;
539	}
540
541out:
542	return ret;
543}
544
545/*
546 * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was
547 * set.
548 */
549static void configfs_detach_rollback(struct dentry *dentry)
550{
551	struct configfs_dirent *parent_sd = dentry->d_fsdata;
552	struct configfs_dirent *sd;
553
554	parent_sd->s_type &= ~CONFIGFS_USET_DROPPING;
555
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);
559}
560
561static void detach_attrs(struct config_item * item)
562{
563	struct dentry * dentry = dget(item->ci_dentry);
564	struct configfs_dirent * parent_sd;
565	struct configfs_dirent * sd, * tmp;
566
567	if (!dentry)
568		return;
569
570	pr_debug("configfs %s: dropping attrs for  dir\n",
571		 dentry->d_name.name);
572
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))
576			continue;
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);
581		configfs_put(sd);
582	}
583
584	/**
585	 * Drop reference from dget() on entrance.
586	 */
587	dput(dentry);
588}
589
590static int populate_attrs(struct config_item *item)
591{
592	struct config_item_type *t = item->ci_type;
593	struct configfs_attribute *attr;
594	int error = 0;
595	int i;
596
597	if (!t)
598		return -EINVAL;
599	if (t->ct_attrs) {
600		for (i = 0; (attr = t->ct_attrs[i]) != NULL; i++) {
601			if ((error = configfs_create_file(item, attr)))
602				break;
603		}
604	}
605
606	if (error)
607		detach_attrs(item);
608
609	return error;
610}
611
612static int configfs_attach_group(struct config_item *parent_item,
613				 struct config_item *item,
614				 struct dentry *dentry);
615static void configfs_detach_group(struct config_item *item);
616
617static void detach_groups(struct config_group *group)
618{
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;
623
624	if (!dentry)
625		return;
626
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))
631			continue;
632
633		child = sd->s_dentry;
634
635		mutex_lock(&child->d_inode->i_mutex);
636
637		configfs_detach_group(sd->s_element);
638		child->d_inode->i_flags |= S_DEAD;
639		dont_mount(child);
640
641		mutex_unlock(&child->d_inode->i_mutex);
642
643		d_delete(child);
644		dput(child);
645	}
646
647	/**
648	 * Drop reference from dget() on entrance.
649	 */
650	dput(dentry);
651}
652
653/*
654 * This fakes mkdir(2) on a default_groups[] entry.  It
655 * creates a dentry, attachs it, and then does fixup
656 * on the sd->s_type.
657 *
658 * We could, perhaps, tweak our parent's ->mkdir for a minute and
659 * try using vfs_mkdir.  Just a thought.
660 */
661static int create_default_group(struct config_group *parent_group,
662				struct config_group *group)
663{
664	int ret;
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;
668
669	if (!group->cg_item.ci_name)
670		group->cg_item.ci_name = group->cg_item.ci_namebuf;
671
672	ret = -ENOMEM;
673	child = d_alloc_name(parent, group->cg_item.ci_name);
674	if (child) {
675		d_add(child, NULL);
676
677		ret = configfs_attach_group(&parent_group->cg_item,
678					    &group->cg_item, child);
679		if (!ret) {
680			sd = child->d_fsdata;
681			sd->s_type |= CONFIGFS_USET_DEFAULT;
682		} else {
683			BUG_ON(child->d_inode);
684			d_drop(child);
685			dput(child);
686		}
687	}
688
689	return ret;
690}
691
692static int populate_groups(struct config_group *group)
693{
694	struct config_group *new_group;
695	int ret = 0;
696	int i;
697
698	if (group->default_groups) {
699		for (i = 0; group->default_groups[i]; i++) {
700			new_group = group->default_groups[i];
701
702			ret = create_default_group(group, new_group);
703			if (ret) {
704				detach_groups(group);
705				break;
706			}
707		}
708	}
709
710	return ret;
711}
712
713/*
714 * All of link_obj/unlink_obj/link_group/unlink_group require that
715 * subsys->su_mutex is held.
716 */
717
718static void unlink_obj(struct config_item *item)
719{
720	struct config_group *group;
721
722	group = item->ci_group;
723	if (group) {
724		list_del_init(&item->ci_entry);
725
726		item->ci_group = NULL;
727		item->ci_parent = NULL;
728
729		/* Drop the reference for ci_entry */
730		config_item_put(item);
731
732		/* Drop the reference for ci_parent */
733		config_group_put(group);
734	}
735}
736
737static void link_obj(struct config_item *parent_item, struct config_item *item)
738{
739	/*
740	 * Parent seems redundant with group, but it makes certain
741	 * traversals much nicer.
742	 */
743	item->ci_parent = parent_item;
744
745	/*
746	 * We hold a reference on the parent for the child's ci_parent
747	 * link.
748	 */
749	item->ci_group = config_group_get(to_config_group(parent_item));
750	list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
751
752	/*
753	 * We hold a reference on the child for ci_entry on the parent's
754	 * cg_children
755	 */
756	config_item_get(item);
757}
758
759static void unlink_group(struct config_group *group)
760{
761	int i;
762	struct config_group *new_group;
763
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);
768		}
769	}
770
771	group->cg_subsys = NULL;
772	unlink_obj(&group->cg_item);
773}
774
775static void link_group(struct config_group *parent_group, struct config_group *group)
776{
777	int i;
778	struct config_group *new_group;
779	struct configfs_subsystem *subsys = NULL; /* gcc is a turd */
780
781	link_obj(&parent_group->cg_item, &group->cg_item);
782
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);
787	else
788		BUG();
789	group->cg_subsys = subsys;
790
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);
795		}
796	}
797}
798
799/*
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.
804 *
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.
808 *
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.
813 */
814static int configfs_attach_item(struct config_item *parent_item,
815				struct config_item *item,
816				struct dentry *dentry)
817{
818	int ret;
819
820	ret = configfs_create_dir(item, dentry);
821	if (!ret) {
822		ret = populate_attrs(item);
823		if (ret) {
824			/*
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.
828			 */
829			mutex_lock(&dentry->d_inode->i_mutex);
830			configfs_remove_dir(item);
831			dentry->d_inode->i_flags |= S_DEAD;
832			dont_mount(dentry);
833			mutex_unlock(&dentry->d_inode->i_mutex);
834			d_delete(dentry);
835		}
836	}
837
838	return ret;
839}
840
841/* Caller holds the mutex of the item's inode */
842static void configfs_detach_item(struct config_item *item)
843{
844	detach_attrs(item);
845	configfs_remove_dir(item);
846}
847
848static int configfs_attach_group(struct config_item *parent_item,
849				 struct config_item *item,
850				 struct dentry *dentry)
851{
852	int ret;
853	struct configfs_dirent *sd;
854
855	ret = configfs_attach_item(parent_item, item, dentry);
856	if (!ret) {
857		sd = dentry->d_fsdata;
858		sd->s_type |= CONFIGFS_USET_DIR;
859
860		/*
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
864		 * under it.
865		 *
866		 * We must also lock the inode to remove it safely in case of
867		 * error, as rmdir() would.
868		 */
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));
872		if (ret) {
873			configfs_detach_item(item);
874			dentry->d_inode->i_flags |= S_DEAD;
875			dont_mount(dentry);
876		}
877		configfs_adjust_dir_dirent_depth_after_populate(sd);
878		mutex_unlock(&dentry->d_inode->i_mutex);
879		if (ret)
880			d_delete(dentry);
881	}
882
883	return ret;
884}
885
886/* Caller holds the mutex of the group's inode */
887static void configfs_detach_group(struct config_item *item)
888{
889	detach_groups(to_config_group(item));
890	configfs_detach_item(item);
891}
892
893/*
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.
901 */
902static void client_disconnect_notify(struct config_item *parent_item,
903				     struct config_item *item)
904{
905	struct config_item_type *type;
906
907	type = parent_item->ci_type;
908	BUG_ON(!type);
909
910	if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
911		type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
912						      item);
913}
914
915/*
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.
920 */
921static void client_drop_item(struct config_item *parent_item,
922			     struct config_item *item)
923{
924	struct config_item_type *type;
925
926	type = parent_item->ci_type;
927	BUG_ON(!type);
928
929	/*
930	 * If ->drop_item() exists, it is responsible for the
931	 * config_item_put().
932	 */
933	if (type->ct_group_ops && type->ct_group_ops->drop_item)
934		type->ct_group_ops->drop_item(to_config_group(parent_item),
935					      item);
936	else
937		config_item_put(item);
938}
939
940#ifdef DEBUG
941static void configfs_dump_one(struct configfs_dirent *sd, int level)
942{
943	pr_info("%*s\"%s\":\n", level, " ", configfs_get_name(sd));
944
945#define type_print(_type) if (sd->s_type & _type) pr_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);
953#undef type_print
954}
955
956static int configfs_dump(struct configfs_dirent *sd, int level)
957{
958	struct configfs_dirent *child_sd;
959	int ret = 0;
960
961	configfs_dump_one(sd, level);
962
963	if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
964		return 0;
965
966	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
967		ret = configfs_dump(child_sd, level + 2);
968		if (ret)
969			break;
970	}
971
972	return ret;
973}
974#endif
975
976
977/*
978 * configfs_depend_item() and configfs_undepend_item()
979 *
980 * WARNING: Do not call these from a configfs callback!
981 *
982 * This describes these functions and their helpers.
983 *
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.
990 *
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.
996 *
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.
1005 */
1006
1007/*
1008 * configfs_depend_prep()
1009 *
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.
1015 *
1016 * Here we do a depth-first search of the dentry hierarchy looking for
1017 * our object.
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
1021 * rmdir().
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.
1026 *
1027 * If the target is not found, -ENOENT is bubbled up.
1028 *
1029 * This adds a requirement that all config_items be unique!
1030 *
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.
1034 */
1035static int configfs_depend_prep(struct dentry *origin,
1036				struct config_item *target)
1037{
1038	struct configfs_dirent *child_sd, *sd;
1039	int ret = 0;
1040
1041	BUG_ON(!origin || !origin->d_fsdata);
1042	sd = origin->d_fsdata;
1043
1044	if (sd->s_element == target)  /* Boo-yah */
1045		goto out;
1046
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,
1052						   target);
1053			if (!ret)
1054				goto out;  /* Child path boo-yah */
1055		}
1056	}
1057
1058	/* We looped all our children and didn't find target */
1059	ret = -ENOENT;
1060
1061out:
1062	return ret;
1063}
1064
1065int configfs_depend_item(struct configfs_subsystem *subsys,
1066			 struct config_item *target)
1067{
1068	int ret;
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;
1072
1073	/*
1074	 * Pin the configfs filesystem.  This means we can safely access
1075	 * the root of the configfs filesystem.
1076	 */
1077	root = configfs_pin_fs();
1078	if (IS_ERR(root))
1079		return PTR_ERR(root);
1080
1081	/*
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().
1085	 */
1086	mutex_lock(&root->d_inode->i_mutex);
1087
1088	root_sd = root->d_fsdata;
1089
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) {
1093				subsys_sd = p;
1094				break;
1095			}
1096		}
1097	}
1098
1099	if (!subsys_sd) {
1100		ret = -ENOENT;
1101		goto out_unlock_fs;
1102	}
1103
1104	/* Ok, now we can trust subsys/s_item */
1105
1106	spin_lock(&configfs_dirent_lock);
1107	/* Scan the tree, return 0 if found */
1108	ret = configfs_depend_prep(subsys_sd->s_dentry, target);
1109	if (ret)
1110		goto out_unlock_dirent_lock;
1111
1112	/*
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().
1115	 */
1116	p = target->ci_dentry->d_fsdata;
1117	p->s_dependent_count += 1;
1118
1119out_unlock_dirent_lock:
1120	spin_unlock(&configfs_dirent_lock);
1121out_unlock_fs:
1122	mutex_unlock(&root->d_inode->i_mutex);
1123
1124	/*
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.
1127	 */
1128	configfs_release_fs();
1129
1130	return ret;
1131}
1132EXPORT_SYMBOL(configfs_depend_item);
1133
1134/*
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.
1138 */
1139void configfs_undepend_item(struct configfs_subsystem *subsys,
1140			    struct config_item *target)
1141{
1142	struct configfs_dirent *sd;
1143
1144	/*
1145	 * Since we can trust everything is pinned, we just need
1146	 * configfs_dirent_lock.
1147	 */
1148	spin_lock(&configfs_dirent_lock);
1149
1150	sd = target->ci_dentry->d_fsdata;
1151	BUG_ON(sd->s_dependent_count < 1);
1152
1153	sd->s_dependent_count -= 1;
1154
1155	/*
1156	 * After this unlock, we cannot trust the item to stay alive!
1157	 * DO NOT REFERENCE item after this unlock.
1158	 */
1159	spin_unlock(&configfs_dirent_lock);
1160}
1161EXPORT_SYMBOL(configfs_undepend_item);
1162
1163static int configfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1164{
1165	int ret = 0;
1166	int module_got = 0;
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;
1174	char *name;
1175
1176	sd = dentry->d_parent->d_fsdata;
1177
1178	/*
1179	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1180	 * being attached
1181	 */
1182	if (!configfs_dirent_is_ready(sd)) {
1183		ret = -ENOENT;
1184		goto out;
1185	}
1186
1187	if (!(sd->s_type & CONFIGFS_USET_DIR)) {
1188		ret = -EPERM;
1189		goto out;
1190	}
1191
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;
1196	BUG_ON(!subsys);
1197
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 */
1202		goto out_put;
1203	}
1204
1205	/*
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.
1209	 */
1210	if (!subsys->su_group.cg_item.ci_type) {
1211		ret = -EINVAL;
1212		goto out_put;
1213	}
1214	subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1215	if (!try_module_get(subsys_owner)) {
1216		ret = -EINVAL;
1217		goto out_put;
1218	}
1219
1220	name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
1221	if (!name) {
1222		ret = -ENOMEM;
1223		goto out_subsys_put;
1224	}
1225
1226	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
1227
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);
1231		if (!group)
1232			group = ERR_PTR(-ENOMEM);
1233		if (!IS_ERR(group)) {
1234			link_group(to_config_group(parent_item), group);
1235			item = &group->cg_item;
1236		} else
1237			ret = PTR_ERR(group);
1238	} else {
1239		item = type->ct_group_ops->make_item(to_config_group(parent_item), name);
1240		if (!item)
1241			item = ERR_PTR(-ENOMEM);
1242		if (!IS_ERR(item))
1243			link_obj(parent_item, item);
1244		else
1245			ret = PTR_ERR(item);
1246	}
1247	mutex_unlock(&subsys->su_mutex);
1248
1249	kfree(name);
1250	if (ret) {
1251		/*
1252		 * If ret != 0, then link_obj() was never called.
1253		 * There are no extra references to clean up.
1254		 */
1255		goto out_subsys_put;
1256	}
1257
1258	/*
1259	 * link_obj() has been called (via link_group() for groups).
1260	 * From here on out, errors must clean that up.
1261	 */
1262
1263	type = item->ci_type;
1264	if (!type) {
1265		ret = -EINVAL;
1266		goto out_unlink;
1267	}
1268
1269	new_item_owner = type->ct_owner;
1270	if (!try_module_get(new_item_owner)) {
1271		ret = -EINVAL;
1272		goto out_unlink;
1273	}
1274
1275	/*
1276	 * I hate doing it this way, but if there is
1277	 * an error,  module_put() probably should
1278	 * happen after any cleanup.
1279	 */
1280	module_got = 1;
1281
1282	/*
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
1287	 */
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);
1292
1293	if (group)
1294		ret = configfs_attach_group(parent_item, item, dentry);
1295	else
1296		ret = configfs_attach_item(parent_item, item, dentry);
1297
1298	spin_lock(&configfs_dirent_lock);
1299	sd->s_type &= ~CONFIGFS_USET_IN_MKDIR;
1300	if (!ret)
1301		configfs_dir_set_ready(dentry->d_fsdata);
1302	spin_unlock(&configfs_dirent_lock);
1303
1304out_unlink:
1305	if (ret) {
1306		/* Tear down everything we built up */
1307		mutex_lock(&subsys->su_mutex);
1308
1309		client_disconnect_notify(parent_item, item);
1310		if (group)
1311			unlink_group(group);
1312		else
1313			unlink_obj(item);
1314		client_drop_item(parent_item, item);
1315
1316		mutex_unlock(&subsys->su_mutex);
1317
1318		if (module_got)
1319			module_put(new_item_owner);
1320	}
1321
1322out_subsys_put:
1323	if (ret)
1324		module_put(subsys_owner);
1325
1326out_put:
1327	/*
1328	 * link_obj()/link_group() took a reference from child->parent,
1329	 * so the parent is safely pinned.  We can drop our working
1330	 * reference.
1331	 */
1332	config_item_put(parent_item);
1333
1334out:
1335	return ret;
1336}
1337
1338static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
1339{
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;
1345	int ret;
1346
1347	sd = dentry->d_fsdata;
1348	if (sd->s_type & CONFIGFS_USET_DEFAULT)
1349		return -EPERM;
1350
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;
1354	BUG_ON(!subsys);
1355
1356	if (!parent_item->ci_type) {
1357		config_item_put(parent_item);
1358		return -EINVAL;
1359	}
1360
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;
1364
1365	/*
1366	 * Ensure that no racing symlink() will make detach_prep() fail while
1367	 * the new link is temporarily attached
1368	 */
1369	do {
1370		struct mutex *wait_mutex;
1371
1372		mutex_lock(&configfs_symlink_mutex);
1373		spin_lock(&configfs_dirent_lock);
1374		/*
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.
1378		 */
1379		ret = sd->s_dependent_count ? -EBUSY : 0;
1380		if (!ret) {
1381			ret = configfs_detach_prep(dentry, &wait_mutex);
1382			if (ret)
1383				configfs_detach_rollback(dentry);
1384		}
1385		spin_unlock(&configfs_dirent_lock);
1386		mutex_unlock(&configfs_symlink_mutex);
1387
1388		if (ret) {
1389			if (ret != -EAGAIN) {
1390				config_item_put(parent_item);
1391				return ret;
1392			}
1393
1394			/* Wait until the racing operation terminates */
1395			mutex_lock(wait_mutex);
1396			mutex_unlock(wait_mutex);
1397		}
1398	} while (ret == -EAGAIN);
1399
1400	/* Get a working ref for the duration of this function */
1401	item = configfs_get_config_item(dentry);
1402
1403	/* Drop reference from above, item already holds one. */
1404	config_item_put(parent_item);
1405
1406	if (item->ci_type)
1407		dead_item_owner = item->ci_type->ct_owner;
1408
1409	if (sd->s_type & CONFIGFS_USET_DIR) {
1410		configfs_detach_group(item);
1411
1412		mutex_lock(&subsys->su_mutex);
1413		client_disconnect_notify(parent_item, item);
1414		unlink_group(to_config_group(item));
1415	} else {
1416		configfs_detach_item(item);
1417
1418		mutex_lock(&subsys->su_mutex);
1419		client_disconnect_notify(parent_item, item);
1420		unlink_obj(item);
1421	}
1422
1423	client_drop_item(parent_item, item);
1424	mutex_unlock(&subsys->su_mutex);
1425
1426	/* Drop our reference from above */
1427	config_item_put(item);
1428
1429	module_put(dead_item_owner);
1430	module_put(subsys_owner);
1431
1432	return 0;
1433}
1434
1435const 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,
1442};
1443
1444const struct inode_operations configfs_root_inode_operations = {
1445	.lookup		= configfs_lookup,
1446	.setattr	= configfs_setattr,
1447};
1448
1449#if 0
1450int configfs_rename_dir(struct config_item * item, const char *new_name)
1451{
1452	int error = 0;
1453	struct dentry * new_dentry, * parent;
1454
1455	if (!strcmp(config_item_name(item), new_name))
1456		return -EINVAL;
1457
1458	if (!item->parent)
1459		return -EINVAL;
1460
1461	down_write(&configfs_rename_sem);
1462	parent = item->parent->dentry;
1463
1464	mutex_lock(&parent->d_inode->i_mutex);
1465
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);
1470			if (!error) {
1471				d_add(new_dentry, NULL);
1472				d_move(item->dentry, new_dentry);
1473			}
1474			else
1475				d_delete(new_dentry);
1476		} else
1477			error = -EEXIST;
1478		dput(new_dentry);
1479	}
1480	mutex_unlock(&parent->d_inode->i_mutex);
1481	up_write(&configfs_rename_sem);
1482
1483	return error;
1484}
1485#endif
1486
1487static int configfs_dir_open(struct inode *inode, struct file *file)
1488{
1489	struct dentry * dentry = file->f_path.dentry;
1490	struct configfs_dirent * parent_sd = dentry->d_fsdata;
1491	int err;
1492
1493	mutex_lock(&dentry->d_inode->i_mutex);
1494	/*
1495	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1496	 * being attached
1497	 */
1498	err = -ENOENT;
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);
1503		else
1504			err = 0;
1505	}
1506	mutex_unlock(&dentry->d_inode->i_mutex);
1507
1508	return err;
1509}
1510
1511static int configfs_dir_close(struct inode *inode, struct file *file)
1512{
1513	struct dentry * dentry = file->f_path.dentry;
1514	struct configfs_dirent * cursor = file->private_data;
1515
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);
1521
1522	release_configfs_dirent(cursor);
1523
1524	return 0;
1525}
1526
1527/* Relationship between s_mode and the DT_xxx types */
1528static inline unsigned char dt_type(struct configfs_dirent *sd)
1529{
1530	return (sd->s_mode >> 12) & 15;
1531}
1532
1533static int configfs_readdir(struct file *file, struct dir_context *ctx)
1534{
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;
1540	ino_t ino = 0;
1541
1542	if (!dir_emit_dots(file, ctx))
1543		return 0;
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);
1548	}
1549	for (p = q->next; p != &parent_sd->s_children; p = p->next) {
1550		struct configfs_dirent *next;
1551		const char *name;
1552		int len;
1553		struct inode *inode = NULL;
1554
1555		next = list_entry(p, struct configfs_dirent, s_sibling);
1556		if (!next->s_element)
1557			continue;
1558
1559		name = configfs_get_name(next);
1560		len = strlen(name);
1561
1562		/*
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().
1568		 *
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
1573		 * care.
1574		 */
1575		spin_lock(&configfs_dirent_lock);
1576		dentry = next->s_dentry;
1577		if (dentry)
1578			inode = dentry->d_inode;
1579		if (inode)
1580			ino = inode->i_ino;
1581		spin_unlock(&configfs_dirent_lock);
1582		if (!inode)
1583			ino = iunique(sb, 2);
1584
1585		if (!dir_emit(ctx, name, len, ino, dt_type(next)))
1586			return 0;
1587
1588		spin_lock(&configfs_dirent_lock);
1589		list_move(q, p);
1590		spin_unlock(&configfs_dirent_lock);
1591		p = q;
1592		ctx->pos++;
1593	}
1594	return 0;
1595}
1596
1597static loff_t configfs_dir_lseek(struct file *file, loff_t offset, int whence)
1598{
1599	struct dentry * dentry = file->f_path.dentry;
1600
1601	mutex_lock(&dentry->d_inode->i_mutex);
1602	switch (whence) {
1603		case 1:
1604			offset += file->f_pos;
1605		case 0:
1606			if (offset >= 0)
1607				break;
1608		default:
1609			mutex_unlock(&file_inode(file)->i_mutex);
1610			return -EINVAL;
1611	}
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;
1619
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,
1626						   s_sibling);
1627				if (next->s_element)
1628					n--;
1629				p = p->next;
1630			}
1631			list_add_tail(&cursor->s_sibling, p);
1632			spin_unlock(&configfs_dirent_lock);
1633		}
1634	}
1635	mutex_unlock(&dentry->d_inode->i_mutex);
1636	return offset;
1637}
1638
1639const 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,
1645};
1646
1647int configfs_register_subsystem(struct configfs_subsystem *subsys)
1648{
1649	int err;
1650	struct config_group *group = &subsys->su_group;
1651	struct dentry *dentry;
1652	struct dentry *root;
1653	struct configfs_dirent *sd;
1654
1655	root = configfs_pin_fs();
1656	if (IS_ERR(root))
1657		return PTR_ERR(root);
1658
1659	if (!group->cg_item.ci_name)
1660		group->cg_item.ci_name = group->cg_item.ci_namebuf;
1661
1662	sd = root->d_fsdata;
1663	link_group(to_config_group(sd->s_element), group);
1664
1665	mutex_lock_nested(&root->d_inode->i_mutex, I_MUTEX_PARENT);
1666
1667	err = -ENOMEM;
1668	dentry = d_alloc_name(root, group->cg_item.ci_name);
1669	if (dentry) {
1670		d_add(dentry, NULL);
1671
1672		err = configfs_attach_group(sd->s_element, &group->cg_item,
1673					    dentry);
1674		if (err) {
1675			BUG_ON(dentry->d_inode);
1676			d_drop(dentry);
1677			dput(dentry);
1678		} else {
1679			spin_lock(&configfs_dirent_lock);
1680			configfs_dir_set_ready(dentry->d_fsdata);
1681			spin_unlock(&configfs_dirent_lock);
1682		}
1683	}
1684
1685	mutex_unlock(&root->d_inode->i_mutex);
1686
1687	if (err) {
1688		unlink_group(group);
1689		configfs_release_fs();
1690	}
1691
1692	return err;
1693}
1694
1695void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
1696{
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;
1700
1701	if (dentry->d_parent != root) {
1702		pr_err("Tried to unregister non-subsystem!\n");
1703		return;
1704	}
1705
1706	mutex_lock_nested(&root->d_inode->i_mutex,
1707			  I_MUTEX_PARENT);
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		pr_err("Tried to unregister non-empty subsystem!\n");
1713	}
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;
1718	dont_mount(dentry);
1719	mutex_unlock(&dentry->d_inode->i_mutex);
1720
1721	d_delete(dentry);
1722
1723	mutex_unlock(&root->d_inode->i_mutex);
1724
1725	dput(dentry);
1726
1727	unlink_group(group);
1728	configfs_release_fs();
1729}
1730
1731EXPORT_SYMBOL(configfs_register_subsystem);
1732EXPORT_SYMBOL(configfs_unregister_subsystem);
1733