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 void 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}
251
252static void configfs_init_file(struct inode * inode)
253{
254	inode->i_size = PAGE_SIZE;
255	inode->i_fop = &configfs_file_operations;
256}
257
258static void init_symlink(struct inode * inode)
259{
260	inode->i_op = &configfs_symlink_inode_operations;
261}
262
263/**
264 *	configfs_create_dir - create a directory for an config_item.
265 *	@item:		config_itemwe're creating directory for.
266 *	@dentry:	config_item's dentry.
267 *
268 *	Note: user-created entries won't be allowed under this new directory
269 *	until it is validated by configfs_dir_set_ready()
270 */
271
272static int configfs_create_dir(struct config_item *item, struct dentry *dentry)
273{
274	int error;
275	umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
276	struct dentry *p = dentry->d_parent;
277
278	BUG_ON(!item);
279
280	error = configfs_dirent_exists(p->d_fsdata, dentry->d_name.name);
281	if (unlikely(error))
282		return error;
283
284	error = configfs_make_dirent(p->d_fsdata, dentry, item, mode,
285				     CONFIGFS_DIR | CONFIGFS_USET_CREATING);
286	if (unlikely(error))
287		return error;
288
289	configfs_set_dir_dirent_depth(p->d_fsdata, dentry->d_fsdata);
290	error = configfs_create(dentry, mode, init_dir);
291	if (!error) {
292		inc_nlink(d_inode(p));
293		item->ci_dentry = dentry;
294	} else {
295		struct configfs_dirent *sd = dentry->d_fsdata;
296		if (sd) {
297			spin_lock(&configfs_dirent_lock);
298			list_del_init(&sd->s_sibling);
299			spin_unlock(&configfs_dirent_lock);
300			configfs_put(sd);
301		}
302	}
303	return error;
304}
305
306/*
307 * Allow userspace to create new entries under a new directory created with
308 * configfs_create_dir(), and under all of its chidlren directories recursively.
309 * @sd		configfs_dirent of the new directory to validate
310 *
311 * Caller must hold configfs_dirent_lock.
312 */
313static void configfs_dir_set_ready(struct configfs_dirent *sd)
314{
315	struct configfs_dirent *child_sd;
316
317	sd->s_type &= ~CONFIGFS_USET_CREATING;
318	list_for_each_entry(child_sd, &sd->s_children, s_sibling)
319		if (child_sd->s_type & CONFIGFS_USET_CREATING)
320			configfs_dir_set_ready(child_sd);
321}
322
323/*
324 * Check that a directory does not belong to a directory hierarchy being
325 * attached and not validated yet.
326 * @sd		configfs_dirent of the directory to check
327 *
328 * @return	non-zero iff the directory was validated
329 *
330 * Note: takes configfs_dirent_lock, so the result may change from false to true
331 * in two consecutive calls, but never from true to false.
332 */
333int configfs_dirent_is_ready(struct configfs_dirent *sd)
334{
335	int ret;
336
337	spin_lock(&configfs_dirent_lock);
338	ret = !(sd->s_type & CONFIGFS_USET_CREATING);
339	spin_unlock(&configfs_dirent_lock);
340
341	return ret;
342}
343
344int configfs_create_link(struct configfs_symlink *sl,
345			 struct dentry *parent,
346			 struct dentry *dentry)
347{
348	int err = 0;
349	umode_t mode = S_IFLNK | S_IRWXUGO;
350
351	err = configfs_make_dirent(parent->d_fsdata, dentry, sl, mode,
352				   CONFIGFS_ITEM_LINK);
353	if (!err) {
354		err = configfs_create(dentry, mode, init_symlink);
355		if (err) {
356			struct configfs_dirent *sd = dentry->d_fsdata;
357			if (sd) {
358				spin_lock(&configfs_dirent_lock);
359				list_del_init(&sd->s_sibling);
360				spin_unlock(&configfs_dirent_lock);
361				configfs_put(sd);
362			}
363		}
364	}
365	return err;
366}
367
368static void remove_dir(struct dentry * d)
369{
370	struct dentry * parent = dget(d->d_parent);
371	struct configfs_dirent * sd;
372
373	sd = d->d_fsdata;
374	spin_lock(&configfs_dirent_lock);
375	list_del_init(&sd->s_sibling);
376	spin_unlock(&configfs_dirent_lock);
377	configfs_put(sd);
378	if (d_really_is_positive(d))
379		simple_rmdir(d_inode(parent),d);
380
381	pr_debug(" o %pd removing done (%d)\n", d, d_count(d));
382
383	dput(parent);
384}
385
386/**
387 * configfs_remove_dir - remove an config_item's directory.
388 * @item:	config_item we're removing.
389 *
390 * The only thing special about this is that we remove any files in
391 * the directory before we remove the directory, and we've inlined
392 * what used to be configfs_rmdir() below, instead of calling separately.
393 *
394 * Caller holds the mutex of the item's inode
395 */
396
397static void configfs_remove_dir(struct config_item * item)
398{
399	struct dentry * dentry = dget(item->ci_dentry);
400
401	if (!dentry)
402		return;
403
404	remove_dir(dentry);
405	/**
406	 * Drop reference from dget() on entrance.
407	 */
408	dput(dentry);
409}
410
411
412/* attaches attribute's configfs_dirent to the dentry corresponding to the
413 * attribute file
414 */
415static int configfs_attach_attr(struct configfs_dirent * sd, struct dentry * dentry)
416{
417	struct configfs_attribute * attr = sd->s_element;
418	int error;
419
420	spin_lock(&configfs_dirent_lock);
421	dentry->d_fsdata = configfs_get(sd);
422	sd->s_dentry = dentry;
423	spin_unlock(&configfs_dirent_lock);
424
425	error = configfs_create(dentry, (attr->ca_mode & S_IALLUGO) | S_IFREG,
426				configfs_init_file);
427	if (error) {
428		configfs_put(sd);
429		return error;
430	}
431
432	d_rehash(dentry);
433
434	return 0;
435}
436
437static struct dentry * configfs_lookup(struct inode *dir,
438				       struct dentry *dentry,
439				       unsigned int flags)
440{
441	struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata;
442	struct configfs_dirent * sd;
443	int found = 0;
444	int err;
445
446	/*
447	 * Fake invisibility if dir belongs to a group/default groups hierarchy
448	 * being attached
449	 *
450	 * This forbids userspace to read/write attributes of items which may
451	 * not complete their initialization, since the dentries of the
452	 * attributes won't be instantiated.
453	 */
454	err = -ENOENT;
455	if (!configfs_dirent_is_ready(parent_sd))
456		goto out;
457
458	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
459		if (sd->s_type & CONFIGFS_NOT_PINNED) {
460			const unsigned char * name = configfs_get_name(sd);
461
462			if (strcmp(name, dentry->d_name.name))
463				continue;
464
465			found = 1;
466			err = configfs_attach_attr(sd, dentry);
467			break;
468		}
469	}
470
471	if (!found) {
472		/*
473		 * If it doesn't exist and it isn't a NOT_PINNED item,
474		 * it must be negative.
475		 */
476		if (dentry->d_name.len > NAME_MAX)
477			return ERR_PTR(-ENAMETOOLONG);
478		d_add(dentry, NULL);
479		return NULL;
480	}
481
482out:
483	return ERR_PTR(err);
484}
485
486/*
487 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
488 * attributes and are removed by rmdir().  We recurse, setting
489 * CONFIGFS_USET_DROPPING on all children that are candidates for
490 * default detach.
491 * If there is an error, the caller will reset the flags via
492 * configfs_detach_rollback().
493 */
494static int configfs_detach_prep(struct dentry *dentry, struct mutex **wait_mutex)
495{
496	struct configfs_dirent *parent_sd = dentry->d_fsdata;
497	struct configfs_dirent *sd;
498	int ret;
499
500	/* Mark that we're trying to drop the group */
501	parent_sd->s_type |= CONFIGFS_USET_DROPPING;
502
503	ret = -EBUSY;
504	if (!list_empty(&parent_sd->s_links))
505		goto out;
506
507	ret = 0;
508	list_for_each_entry(sd, &parent_sd->s_children, s_sibling) {
509		if (!sd->s_element ||
510		    (sd->s_type & CONFIGFS_NOT_PINNED))
511			continue;
512		if (sd->s_type & CONFIGFS_USET_DEFAULT) {
513			/* Abort if racing with mkdir() */
514			if (sd->s_type & CONFIGFS_USET_IN_MKDIR) {
515				if (wait_mutex)
516					*wait_mutex = &d_inode(sd->s_dentry)->i_mutex;
517				return -EAGAIN;
518			}
519
520			/*
521			 * Yup, recursive.  If there's a problem, blame
522			 * deep nesting of default_groups
523			 */
524			ret = configfs_detach_prep(sd->s_dentry, wait_mutex);
525			if (!ret)
526				continue;
527		} else
528			ret = -ENOTEMPTY;
529
530		break;
531	}
532
533out:
534	return ret;
535}
536
537/*
538 * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was
539 * set.
540 */
541static void configfs_detach_rollback(struct dentry *dentry)
542{
543	struct configfs_dirent *parent_sd = dentry->d_fsdata;
544	struct configfs_dirent *sd;
545
546	parent_sd->s_type &= ~CONFIGFS_USET_DROPPING;
547
548	list_for_each_entry(sd, &parent_sd->s_children, s_sibling)
549		if (sd->s_type & CONFIGFS_USET_DEFAULT)
550			configfs_detach_rollback(sd->s_dentry);
551}
552
553static void detach_attrs(struct config_item * item)
554{
555	struct dentry * dentry = dget(item->ci_dentry);
556	struct configfs_dirent * parent_sd;
557	struct configfs_dirent * sd, * tmp;
558
559	if (!dentry)
560		return;
561
562	pr_debug("configfs %s: dropping attrs for  dir\n",
563		 dentry->d_name.name);
564
565	parent_sd = dentry->d_fsdata;
566	list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
567		if (!sd->s_element || !(sd->s_type & CONFIGFS_NOT_PINNED))
568			continue;
569		spin_lock(&configfs_dirent_lock);
570		list_del_init(&sd->s_sibling);
571		spin_unlock(&configfs_dirent_lock);
572		configfs_drop_dentry(sd, dentry);
573		configfs_put(sd);
574	}
575
576	/**
577	 * Drop reference from dget() on entrance.
578	 */
579	dput(dentry);
580}
581
582static int populate_attrs(struct config_item *item)
583{
584	struct config_item_type *t = item->ci_type;
585	struct configfs_attribute *attr;
586	int error = 0;
587	int i;
588
589	if (!t)
590		return -EINVAL;
591	if (t->ct_attrs) {
592		for (i = 0; (attr = t->ct_attrs[i]) != NULL; i++) {
593			if ((error = configfs_create_file(item, attr)))
594				break;
595		}
596	}
597
598	if (error)
599		detach_attrs(item);
600
601	return error;
602}
603
604static int configfs_attach_group(struct config_item *parent_item,
605				 struct config_item *item,
606				 struct dentry *dentry);
607static void configfs_detach_group(struct config_item *item);
608
609static void detach_groups(struct config_group *group)
610{
611	struct dentry * dentry = dget(group->cg_item.ci_dentry);
612	struct dentry *child;
613	struct configfs_dirent *parent_sd;
614	struct configfs_dirent *sd, *tmp;
615
616	if (!dentry)
617		return;
618
619	parent_sd = dentry->d_fsdata;
620	list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) {
621		if (!sd->s_element ||
622		    !(sd->s_type & CONFIGFS_USET_DEFAULT))
623			continue;
624
625		child = sd->s_dentry;
626
627		mutex_lock(&d_inode(child)->i_mutex);
628
629		configfs_detach_group(sd->s_element);
630		d_inode(child)->i_flags |= S_DEAD;
631		dont_mount(child);
632
633		mutex_unlock(&d_inode(child)->i_mutex);
634
635		d_delete(child);
636		dput(child);
637	}
638
639	/**
640	 * Drop reference from dget() on entrance.
641	 */
642	dput(dentry);
643}
644
645/*
646 * This fakes mkdir(2) on a default_groups[] entry.  It
647 * creates a dentry, attachs it, and then does fixup
648 * on the sd->s_type.
649 *
650 * We could, perhaps, tweak our parent's ->mkdir for a minute and
651 * try using vfs_mkdir.  Just a thought.
652 */
653static int create_default_group(struct config_group *parent_group,
654				struct config_group *group)
655{
656	int ret;
657	struct configfs_dirent *sd;
658	/* We trust the caller holds a reference to parent */
659	struct dentry *child, *parent = parent_group->cg_item.ci_dentry;
660
661	if (!group->cg_item.ci_name)
662		group->cg_item.ci_name = group->cg_item.ci_namebuf;
663
664	ret = -ENOMEM;
665	child = d_alloc_name(parent, group->cg_item.ci_name);
666	if (child) {
667		d_add(child, NULL);
668
669		ret = configfs_attach_group(&parent_group->cg_item,
670					    &group->cg_item, child);
671		if (!ret) {
672			sd = child->d_fsdata;
673			sd->s_type |= CONFIGFS_USET_DEFAULT;
674		} else {
675			BUG_ON(d_inode(child));
676			d_drop(child);
677			dput(child);
678		}
679	}
680
681	return ret;
682}
683
684static int populate_groups(struct config_group *group)
685{
686	struct config_group *new_group;
687	int ret = 0;
688	int i;
689
690	if (group->default_groups) {
691		for (i = 0; group->default_groups[i]; i++) {
692			new_group = group->default_groups[i];
693
694			ret = create_default_group(group, new_group);
695			if (ret) {
696				detach_groups(group);
697				break;
698			}
699		}
700	}
701
702	return ret;
703}
704
705/*
706 * All of link_obj/unlink_obj/link_group/unlink_group require that
707 * subsys->su_mutex is held.
708 */
709
710static void unlink_obj(struct config_item *item)
711{
712	struct config_group *group;
713
714	group = item->ci_group;
715	if (group) {
716		list_del_init(&item->ci_entry);
717
718		item->ci_group = NULL;
719		item->ci_parent = NULL;
720
721		/* Drop the reference for ci_entry */
722		config_item_put(item);
723
724		/* Drop the reference for ci_parent */
725		config_group_put(group);
726	}
727}
728
729static void link_obj(struct config_item *parent_item, struct config_item *item)
730{
731	/*
732	 * Parent seems redundant with group, but it makes certain
733	 * traversals much nicer.
734	 */
735	item->ci_parent = parent_item;
736
737	/*
738	 * We hold a reference on the parent for the child's ci_parent
739	 * link.
740	 */
741	item->ci_group = config_group_get(to_config_group(parent_item));
742	list_add_tail(&item->ci_entry, &item->ci_group->cg_children);
743
744	/*
745	 * We hold a reference on the child for ci_entry on the parent's
746	 * cg_children
747	 */
748	config_item_get(item);
749}
750
751static void unlink_group(struct config_group *group)
752{
753	int i;
754	struct config_group *new_group;
755
756	if (group->default_groups) {
757		for (i = 0; group->default_groups[i]; i++) {
758			new_group = group->default_groups[i];
759			unlink_group(new_group);
760		}
761	}
762
763	group->cg_subsys = NULL;
764	unlink_obj(&group->cg_item);
765}
766
767static void link_group(struct config_group *parent_group, struct config_group *group)
768{
769	int i;
770	struct config_group *new_group;
771	struct configfs_subsystem *subsys = NULL; /* gcc is a turd */
772
773	link_obj(&parent_group->cg_item, &group->cg_item);
774
775	if (parent_group->cg_subsys)
776		subsys = parent_group->cg_subsys;
777	else if (configfs_is_root(&parent_group->cg_item))
778		subsys = to_configfs_subsystem(group);
779	else
780		BUG();
781	group->cg_subsys = subsys;
782
783	if (group->default_groups) {
784		for (i = 0; group->default_groups[i]; i++) {
785			new_group = group->default_groups[i];
786			link_group(group, new_group);
787		}
788	}
789}
790
791/*
792 * The goal is that configfs_attach_item() (and
793 * configfs_attach_group()) can be called from either the VFS or this
794 * module.  That is, they assume that the items have been created,
795 * the dentry allocated, and the dcache is all ready to go.
796 *
797 * If they fail, they must clean up after themselves as if they
798 * had never been called.  The caller (VFS or local function) will
799 * handle cleaning up the dcache bits.
800 *
801 * configfs_detach_group() and configfs_detach_item() behave similarly on
802 * the way out.  They assume that the proper semaphores are held, they
803 * clean up the configfs items, and they expect their callers will
804 * handle the dcache bits.
805 */
806static int configfs_attach_item(struct config_item *parent_item,
807				struct config_item *item,
808				struct dentry *dentry)
809{
810	int ret;
811
812	ret = configfs_create_dir(item, dentry);
813	if (!ret) {
814		ret = populate_attrs(item);
815		if (ret) {
816			/*
817			 * We are going to remove an inode and its dentry but
818			 * the VFS may already have hit and used them. Thus,
819			 * we must lock them as rmdir() would.
820			 */
821			mutex_lock(&d_inode(dentry)->i_mutex);
822			configfs_remove_dir(item);
823			d_inode(dentry)->i_flags |= S_DEAD;
824			dont_mount(dentry);
825			mutex_unlock(&d_inode(dentry)->i_mutex);
826			d_delete(dentry);
827		}
828	}
829
830	return ret;
831}
832
833/* Caller holds the mutex of the item's inode */
834static void configfs_detach_item(struct config_item *item)
835{
836	detach_attrs(item);
837	configfs_remove_dir(item);
838}
839
840static int configfs_attach_group(struct config_item *parent_item,
841				 struct config_item *item,
842				 struct dentry *dentry)
843{
844	int ret;
845	struct configfs_dirent *sd;
846
847	ret = configfs_attach_item(parent_item, item, dentry);
848	if (!ret) {
849		sd = dentry->d_fsdata;
850		sd->s_type |= CONFIGFS_USET_DIR;
851
852		/*
853		 * FYI, we're faking mkdir in populate_groups()
854		 * We must lock the group's inode to avoid races with the VFS
855		 * which can already hit the inode and try to add/remove entries
856		 * under it.
857		 *
858		 * We must also lock the inode to remove it safely in case of
859		 * error, as rmdir() would.
860		 */
861		mutex_lock_nested(&d_inode(dentry)->i_mutex, I_MUTEX_CHILD);
862		configfs_adjust_dir_dirent_depth_before_populate(sd);
863		ret = populate_groups(to_config_group(item));
864		if (ret) {
865			configfs_detach_item(item);
866			d_inode(dentry)->i_flags |= S_DEAD;
867			dont_mount(dentry);
868		}
869		configfs_adjust_dir_dirent_depth_after_populate(sd);
870		mutex_unlock(&d_inode(dentry)->i_mutex);
871		if (ret)
872			d_delete(dentry);
873	}
874
875	return ret;
876}
877
878/* Caller holds the mutex of the group's inode */
879static void configfs_detach_group(struct config_item *item)
880{
881	detach_groups(to_config_group(item));
882	configfs_detach_item(item);
883}
884
885/*
886 * After the item has been detached from the filesystem view, we are
887 * ready to tear it out of the hierarchy.  Notify the client before
888 * we do that so they can perform any cleanup that requires
889 * navigating the hierarchy.  A client does not need to provide this
890 * callback.  The subsystem semaphore MUST be held by the caller, and
891 * references must be valid for both items.  It also assumes the
892 * caller has validated ci_type.
893 */
894static void client_disconnect_notify(struct config_item *parent_item,
895				     struct config_item *item)
896{
897	struct config_item_type *type;
898
899	type = parent_item->ci_type;
900	BUG_ON(!type);
901
902	if (type->ct_group_ops && type->ct_group_ops->disconnect_notify)
903		type->ct_group_ops->disconnect_notify(to_config_group(parent_item),
904						      item);
905}
906
907/*
908 * Drop the initial reference from make_item()/make_group()
909 * This function assumes that reference is held on item
910 * and that item holds a valid reference to the parent.  Also, it
911 * assumes the caller has validated ci_type.
912 */
913static void client_drop_item(struct config_item *parent_item,
914			     struct config_item *item)
915{
916	struct config_item_type *type;
917
918	type = parent_item->ci_type;
919	BUG_ON(!type);
920
921	/*
922	 * If ->drop_item() exists, it is responsible for the
923	 * config_item_put().
924	 */
925	if (type->ct_group_ops && type->ct_group_ops->drop_item)
926		type->ct_group_ops->drop_item(to_config_group(parent_item),
927					      item);
928	else
929		config_item_put(item);
930}
931
932#ifdef DEBUG
933static void configfs_dump_one(struct configfs_dirent *sd, int level)
934{
935	pr_info("%*s\"%s\":\n", level, " ", configfs_get_name(sd));
936
937#define type_print(_type) if (sd->s_type & _type) pr_info("%*s %s\n", level, " ", #_type);
938	type_print(CONFIGFS_ROOT);
939	type_print(CONFIGFS_DIR);
940	type_print(CONFIGFS_ITEM_ATTR);
941	type_print(CONFIGFS_ITEM_LINK);
942	type_print(CONFIGFS_USET_DIR);
943	type_print(CONFIGFS_USET_DEFAULT);
944	type_print(CONFIGFS_USET_DROPPING);
945#undef type_print
946}
947
948static int configfs_dump(struct configfs_dirent *sd, int level)
949{
950	struct configfs_dirent *child_sd;
951	int ret = 0;
952
953	configfs_dump_one(sd, level);
954
955	if (!(sd->s_type & (CONFIGFS_DIR|CONFIGFS_ROOT)))
956		return 0;
957
958	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
959		ret = configfs_dump(child_sd, level + 2);
960		if (ret)
961			break;
962	}
963
964	return ret;
965}
966#endif
967
968
969/*
970 * configfs_depend_item() and configfs_undepend_item()
971 *
972 * WARNING: Do not call these from a configfs callback!
973 *
974 * This describes these functions and their helpers.
975 *
976 * Allow another kernel system to depend on a config_item.  If this
977 * happens, the item cannot go away until the dependent can live without
978 * it.  The idea is to give client modules as simple an interface as
979 * possible.  When a system asks them to depend on an item, they just
980 * call configfs_depend_item().  If the item is live and the client
981 * driver is in good shape, we'll happily do the work for them.
982 *
983 * Why is the locking complex?  Because configfs uses the VFS to handle
984 * all locking, but this function is called outside the normal
985 * VFS->configfs path.  So it must take VFS locks to prevent the
986 * VFS->configfs stuff (configfs_mkdir(), configfs_rmdir(), etc).  This is
987 * why you can't call these functions underneath configfs callbacks.
988 *
989 * Note, btw, that this can be called at *any* time, even when a configfs
990 * subsystem isn't registered, or when configfs is loading or unloading.
991 * Just like configfs_register_subsystem().  So we take the same
992 * precautions.  We pin the filesystem.  We lock configfs_dirent_lock.
993 * If we can find the target item in the
994 * configfs tree, it must be part of the subsystem tree as well, so we
995 * do not need the subsystem semaphore.  Holding configfs_dirent_lock helps
996 * locking out mkdir() and rmdir(), who might be racing us.
997 */
998
999/*
1000 * configfs_depend_prep()
1001 *
1002 * Only subdirectories count here.  Files (CONFIGFS_NOT_PINNED) are
1003 * attributes.  This is similar but not the same to configfs_detach_prep().
1004 * Note that configfs_detach_prep() expects the parent to be locked when it
1005 * is called, but we lock the parent *inside* configfs_depend_prep().  We
1006 * do that so we can unlock it if we find nothing.
1007 *
1008 * Here we do a depth-first search of the dentry hierarchy looking for
1009 * our object.
1010 * We deliberately ignore items tagged as dropping since they are virtually
1011 * dead, as well as items in the middle of attachment since they virtually
1012 * do not exist yet. This completes the locking out of racing mkdir() and
1013 * rmdir().
1014 * Note: subdirectories in the middle of attachment start with s_type =
1015 * CONFIGFS_DIR|CONFIGFS_USET_CREATING set by create_dir().  When
1016 * CONFIGFS_USET_CREATING is set, we ignore the item.  The actual set of
1017 * s_type is in configfs_new_dirent(), which has configfs_dirent_lock.
1018 *
1019 * If the target is not found, -ENOENT is bubbled up.
1020 *
1021 * This adds a requirement that all config_items be unique!
1022 *
1023 * This is recursive.  There isn't
1024 * much on the stack, though, so folks that need this function - be careful
1025 * about your stack!  Patches will be accepted to make it iterative.
1026 */
1027static int configfs_depend_prep(struct dentry *origin,
1028				struct config_item *target)
1029{
1030	struct configfs_dirent *child_sd, *sd;
1031	int ret = 0;
1032
1033	BUG_ON(!origin || !origin->d_fsdata);
1034	sd = origin->d_fsdata;
1035
1036	if (sd->s_element == target)  /* Boo-yah */
1037		goto out;
1038
1039	list_for_each_entry(child_sd, &sd->s_children, s_sibling) {
1040		if ((child_sd->s_type & CONFIGFS_DIR) &&
1041		    !(child_sd->s_type & CONFIGFS_USET_DROPPING) &&
1042		    !(child_sd->s_type & CONFIGFS_USET_CREATING)) {
1043			ret = configfs_depend_prep(child_sd->s_dentry,
1044						   target);
1045			if (!ret)
1046				goto out;  /* Child path boo-yah */
1047		}
1048	}
1049
1050	/* We looped all our children and didn't find target */
1051	ret = -ENOENT;
1052
1053out:
1054	return ret;
1055}
1056
1057int configfs_depend_item(struct configfs_subsystem *subsys,
1058			 struct config_item *target)
1059{
1060	int ret;
1061	struct configfs_dirent *p, *root_sd, *subsys_sd = NULL;
1062	struct config_item *s_item = &subsys->su_group.cg_item;
1063	struct dentry *root;
1064
1065	/*
1066	 * Pin the configfs filesystem.  This means we can safely access
1067	 * the root of the configfs filesystem.
1068	 */
1069	root = configfs_pin_fs();
1070	if (IS_ERR(root))
1071		return PTR_ERR(root);
1072
1073	/*
1074	 * Next, lock the root directory.  We're going to check that the
1075	 * subsystem is really registered, and so we need to lock out
1076	 * configfs_[un]register_subsystem().
1077	 */
1078	mutex_lock(&d_inode(root)->i_mutex);
1079
1080	root_sd = root->d_fsdata;
1081
1082	list_for_each_entry(p, &root_sd->s_children, s_sibling) {
1083		if (p->s_type & CONFIGFS_DIR) {
1084			if (p->s_element == s_item) {
1085				subsys_sd = p;
1086				break;
1087			}
1088		}
1089	}
1090
1091	if (!subsys_sd) {
1092		ret = -ENOENT;
1093		goto out_unlock_fs;
1094	}
1095
1096	/* Ok, now we can trust subsys/s_item */
1097
1098	spin_lock(&configfs_dirent_lock);
1099	/* Scan the tree, return 0 if found */
1100	ret = configfs_depend_prep(subsys_sd->s_dentry, target);
1101	if (ret)
1102		goto out_unlock_dirent_lock;
1103
1104	/*
1105	 * We are sure that the item is not about to be removed by rmdir(), and
1106	 * not in the middle of attachment by mkdir().
1107	 */
1108	p = target->ci_dentry->d_fsdata;
1109	p->s_dependent_count += 1;
1110
1111out_unlock_dirent_lock:
1112	spin_unlock(&configfs_dirent_lock);
1113out_unlock_fs:
1114	mutex_unlock(&d_inode(root)->i_mutex);
1115
1116	/*
1117	 * If we succeeded, the fs is pinned via other methods.  If not,
1118	 * we're done with it anyway.  So release_fs() is always right.
1119	 */
1120	configfs_release_fs();
1121
1122	return ret;
1123}
1124EXPORT_SYMBOL(configfs_depend_item);
1125
1126/*
1127 * Release the dependent linkage.  This is much simpler than
1128 * configfs_depend_item() because we know that that the client driver is
1129 * pinned, thus the subsystem is pinned, and therefore configfs is pinned.
1130 */
1131void configfs_undepend_item(struct configfs_subsystem *subsys,
1132			    struct config_item *target)
1133{
1134	struct configfs_dirent *sd;
1135
1136	/*
1137	 * Since we can trust everything is pinned, we just need
1138	 * configfs_dirent_lock.
1139	 */
1140	spin_lock(&configfs_dirent_lock);
1141
1142	sd = target->ci_dentry->d_fsdata;
1143	BUG_ON(sd->s_dependent_count < 1);
1144
1145	sd->s_dependent_count -= 1;
1146
1147	/*
1148	 * After this unlock, we cannot trust the item to stay alive!
1149	 * DO NOT REFERENCE item after this unlock.
1150	 */
1151	spin_unlock(&configfs_dirent_lock);
1152}
1153EXPORT_SYMBOL(configfs_undepend_item);
1154
1155static int configfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1156{
1157	int ret = 0;
1158	int module_got = 0;
1159	struct config_group *group = NULL;
1160	struct config_item *item = NULL;
1161	struct config_item *parent_item;
1162	struct configfs_subsystem *subsys;
1163	struct configfs_dirent *sd;
1164	struct config_item_type *type;
1165	struct module *subsys_owner = NULL, *new_item_owner = NULL;
1166	char *name;
1167
1168	sd = dentry->d_parent->d_fsdata;
1169
1170	/*
1171	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1172	 * being attached
1173	 */
1174	if (!configfs_dirent_is_ready(sd)) {
1175		ret = -ENOENT;
1176		goto out;
1177	}
1178
1179	if (!(sd->s_type & CONFIGFS_USET_DIR)) {
1180		ret = -EPERM;
1181		goto out;
1182	}
1183
1184	/* Get a working ref for the duration of this function */
1185	parent_item = configfs_get_config_item(dentry->d_parent);
1186	type = parent_item->ci_type;
1187	subsys = to_config_group(parent_item)->cg_subsys;
1188	BUG_ON(!subsys);
1189
1190	if (!type || !type->ct_group_ops ||
1191	    (!type->ct_group_ops->make_group &&
1192	     !type->ct_group_ops->make_item)) {
1193		ret = -EPERM;  /* Lack-of-mkdir returns -EPERM */
1194		goto out_put;
1195	}
1196
1197	/*
1198	 * The subsystem may belong to a different module than the item
1199	 * being created.  We don't want to safely pin the new item but
1200	 * fail to pin the subsystem it sits under.
1201	 */
1202	if (!subsys->su_group.cg_item.ci_type) {
1203		ret = -EINVAL;
1204		goto out_put;
1205	}
1206	subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1207	if (!try_module_get(subsys_owner)) {
1208		ret = -EINVAL;
1209		goto out_put;
1210	}
1211
1212	name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL);
1213	if (!name) {
1214		ret = -ENOMEM;
1215		goto out_subsys_put;
1216	}
1217
1218	snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name);
1219
1220	mutex_lock(&subsys->su_mutex);
1221	if (type->ct_group_ops->make_group) {
1222		group = type->ct_group_ops->make_group(to_config_group(parent_item), name);
1223		if (!group)
1224			group = ERR_PTR(-ENOMEM);
1225		if (!IS_ERR(group)) {
1226			link_group(to_config_group(parent_item), group);
1227			item = &group->cg_item;
1228		} else
1229			ret = PTR_ERR(group);
1230	} else {
1231		item = type->ct_group_ops->make_item(to_config_group(parent_item), name);
1232		if (!item)
1233			item = ERR_PTR(-ENOMEM);
1234		if (!IS_ERR(item))
1235			link_obj(parent_item, item);
1236		else
1237			ret = PTR_ERR(item);
1238	}
1239	mutex_unlock(&subsys->su_mutex);
1240
1241	kfree(name);
1242	if (ret) {
1243		/*
1244		 * If ret != 0, then link_obj() was never called.
1245		 * There are no extra references to clean up.
1246		 */
1247		goto out_subsys_put;
1248	}
1249
1250	/*
1251	 * link_obj() has been called (via link_group() for groups).
1252	 * From here on out, errors must clean that up.
1253	 */
1254
1255	type = item->ci_type;
1256	if (!type) {
1257		ret = -EINVAL;
1258		goto out_unlink;
1259	}
1260
1261	new_item_owner = type->ct_owner;
1262	if (!try_module_get(new_item_owner)) {
1263		ret = -EINVAL;
1264		goto out_unlink;
1265	}
1266
1267	/*
1268	 * I hate doing it this way, but if there is
1269	 * an error,  module_put() probably should
1270	 * happen after any cleanup.
1271	 */
1272	module_got = 1;
1273
1274	/*
1275	 * Make racing rmdir() fail if it did not tag parent with
1276	 * CONFIGFS_USET_DROPPING
1277	 * Note: if CONFIGFS_USET_DROPPING is already set, attach_group() will
1278	 * fail and let rmdir() terminate correctly
1279	 */
1280	spin_lock(&configfs_dirent_lock);
1281	/* This will make configfs_detach_prep() fail */
1282	sd->s_type |= CONFIGFS_USET_IN_MKDIR;
1283	spin_unlock(&configfs_dirent_lock);
1284
1285	if (group)
1286		ret = configfs_attach_group(parent_item, item, dentry);
1287	else
1288		ret = configfs_attach_item(parent_item, item, dentry);
1289
1290	spin_lock(&configfs_dirent_lock);
1291	sd->s_type &= ~CONFIGFS_USET_IN_MKDIR;
1292	if (!ret)
1293		configfs_dir_set_ready(dentry->d_fsdata);
1294	spin_unlock(&configfs_dirent_lock);
1295
1296out_unlink:
1297	if (ret) {
1298		/* Tear down everything we built up */
1299		mutex_lock(&subsys->su_mutex);
1300
1301		client_disconnect_notify(parent_item, item);
1302		if (group)
1303			unlink_group(group);
1304		else
1305			unlink_obj(item);
1306		client_drop_item(parent_item, item);
1307
1308		mutex_unlock(&subsys->su_mutex);
1309
1310		if (module_got)
1311			module_put(new_item_owner);
1312	}
1313
1314out_subsys_put:
1315	if (ret)
1316		module_put(subsys_owner);
1317
1318out_put:
1319	/*
1320	 * link_obj()/link_group() took a reference from child->parent,
1321	 * so the parent is safely pinned.  We can drop our working
1322	 * reference.
1323	 */
1324	config_item_put(parent_item);
1325
1326out:
1327	return ret;
1328}
1329
1330static int configfs_rmdir(struct inode *dir, struct dentry *dentry)
1331{
1332	struct config_item *parent_item;
1333	struct config_item *item;
1334	struct configfs_subsystem *subsys;
1335	struct configfs_dirent *sd;
1336	struct module *subsys_owner = NULL, *dead_item_owner = NULL;
1337	int ret;
1338
1339	sd = dentry->d_fsdata;
1340	if (sd->s_type & CONFIGFS_USET_DEFAULT)
1341		return -EPERM;
1342
1343	/* Get a working ref until we have the child */
1344	parent_item = configfs_get_config_item(dentry->d_parent);
1345	subsys = to_config_group(parent_item)->cg_subsys;
1346	BUG_ON(!subsys);
1347
1348	if (!parent_item->ci_type) {
1349		config_item_put(parent_item);
1350		return -EINVAL;
1351	}
1352
1353	/* configfs_mkdir() shouldn't have allowed this */
1354	BUG_ON(!subsys->su_group.cg_item.ci_type);
1355	subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner;
1356
1357	/*
1358	 * Ensure that no racing symlink() will make detach_prep() fail while
1359	 * the new link is temporarily attached
1360	 */
1361	do {
1362		struct mutex *wait_mutex;
1363
1364		mutex_lock(&configfs_symlink_mutex);
1365		spin_lock(&configfs_dirent_lock);
1366		/*
1367		 * Here's where we check for dependents.  We're protected by
1368		 * configfs_dirent_lock.
1369		 * If no dependent, atomically tag the item as dropping.
1370		 */
1371		ret = sd->s_dependent_count ? -EBUSY : 0;
1372		if (!ret) {
1373			ret = configfs_detach_prep(dentry, &wait_mutex);
1374			if (ret)
1375				configfs_detach_rollback(dentry);
1376		}
1377		spin_unlock(&configfs_dirent_lock);
1378		mutex_unlock(&configfs_symlink_mutex);
1379
1380		if (ret) {
1381			if (ret != -EAGAIN) {
1382				config_item_put(parent_item);
1383				return ret;
1384			}
1385
1386			/* Wait until the racing operation terminates */
1387			mutex_lock(wait_mutex);
1388			mutex_unlock(wait_mutex);
1389		}
1390	} while (ret == -EAGAIN);
1391
1392	/* Get a working ref for the duration of this function */
1393	item = configfs_get_config_item(dentry);
1394
1395	/* Drop reference from above, item already holds one. */
1396	config_item_put(parent_item);
1397
1398	if (item->ci_type)
1399		dead_item_owner = item->ci_type->ct_owner;
1400
1401	if (sd->s_type & CONFIGFS_USET_DIR) {
1402		configfs_detach_group(item);
1403
1404		mutex_lock(&subsys->su_mutex);
1405		client_disconnect_notify(parent_item, item);
1406		unlink_group(to_config_group(item));
1407	} else {
1408		configfs_detach_item(item);
1409
1410		mutex_lock(&subsys->su_mutex);
1411		client_disconnect_notify(parent_item, item);
1412		unlink_obj(item);
1413	}
1414
1415	client_drop_item(parent_item, item);
1416	mutex_unlock(&subsys->su_mutex);
1417
1418	/* Drop our reference from above */
1419	config_item_put(item);
1420
1421	module_put(dead_item_owner);
1422	module_put(subsys_owner);
1423
1424	return 0;
1425}
1426
1427const struct inode_operations configfs_dir_inode_operations = {
1428	.mkdir		= configfs_mkdir,
1429	.rmdir		= configfs_rmdir,
1430	.symlink	= configfs_symlink,
1431	.unlink		= configfs_unlink,
1432	.lookup		= configfs_lookup,
1433	.setattr	= configfs_setattr,
1434};
1435
1436const struct inode_operations configfs_root_inode_operations = {
1437	.lookup		= configfs_lookup,
1438	.setattr	= configfs_setattr,
1439};
1440
1441#if 0
1442int configfs_rename_dir(struct config_item * item, const char *new_name)
1443{
1444	int error = 0;
1445	struct dentry * new_dentry, * parent;
1446
1447	if (!strcmp(config_item_name(item), new_name))
1448		return -EINVAL;
1449
1450	if (!item->parent)
1451		return -EINVAL;
1452
1453	down_write(&configfs_rename_sem);
1454	parent = item->parent->dentry;
1455
1456	mutex_lock(&d_inode(parent)->i_mutex);
1457
1458	new_dentry = lookup_one_len(new_name, parent, strlen(new_name));
1459	if (!IS_ERR(new_dentry)) {
1460		if (d_really_is_negative(new_dentry)) {
1461			error = config_item_set_name(item, "%s", new_name);
1462			if (!error) {
1463				d_add(new_dentry, NULL);
1464				d_move(item->dentry, new_dentry);
1465			}
1466			else
1467				d_delete(new_dentry);
1468		} else
1469			error = -EEXIST;
1470		dput(new_dentry);
1471	}
1472	mutex_unlock(&d_inode(parent)->i_mutex);
1473	up_write(&configfs_rename_sem);
1474
1475	return error;
1476}
1477#endif
1478
1479static int configfs_dir_open(struct inode *inode, struct file *file)
1480{
1481	struct dentry * dentry = file->f_path.dentry;
1482	struct configfs_dirent * parent_sd = dentry->d_fsdata;
1483	int err;
1484
1485	mutex_lock(&d_inode(dentry)->i_mutex);
1486	/*
1487	 * Fake invisibility if dir belongs to a group/default groups hierarchy
1488	 * being attached
1489	 */
1490	err = -ENOENT;
1491	if (configfs_dirent_is_ready(parent_sd)) {
1492		file->private_data = configfs_new_dirent(parent_sd, NULL, 0);
1493		if (IS_ERR(file->private_data))
1494			err = PTR_ERR(file->private_data);
1495		else
1496			err = 0;
1497	}
1498	mutex_unlock(&d_inode(dentry)->i_mutex);
1499
1500	return err;
1501}
1502
1503static int configfs_dir_close(struct inode *inode, struct file *file)
1504{
1505	struct dentry * dentry = file->f_path.dentry;
1506	struct configfs_dirent * cursor = file->private_data;
1507
1508	mutex_lock(&d_inode(dentry)->i_mutex);
1509	spin_lock(&configfs_dirent_lock);
1510	list_del_init(&cursor->s_sibling);
1511	spin_unlock(&configfs_dirent_lock);
1512	mutex_unlock(&d_inode(dentry)->i_mutex);
1513
1514	release_configfs_dirent(cursor);
1515
1516	return 0;
1517}
1518
1519/* Relationship between s_mode and the DT_xxx types */
1520static inline unsigned char dt_type(struct configfs_dirent *sd)
1521{
1522	return (sd->s_mode >> 12) & 15;
1523}
1524
1525static int configfs_readdir(struct file *file, struct dir_context *ctx)
1526{
1527	struct dentry *dentry = file->f_path.dentry;
1528	struct super_block *sb = dentry->d_sb;
1529	struct configfs_dirent * parent_sd = dentry->d_fsdata;
1530	struct configfs_dirent *cursor = file->private_data;
1531	struct list_head *p, *q = &cursor->s_sibling;
1532	ino_t ino = 0;
1533
1534	if (!dir_emit_dots(file, ctx))
1535		return 0;
1536	if (ctx->pos == 2) {
1537		spin_lock(&configfs_dirent_lock);
1538		list_move(q, &parent_sd->s_children);
1539		spin_unlock(&configfs_dirent_lock);
1540	}
1541	for (p = q->next; p != &parent_sd->s_children; p = p->next) {
1542		struct configfs_dirent *next;
1543		const char *name;
1544		int len;
1545		struct inode *inode = NULL;
1546
1547		next = list_entry(p, struct configfs_dirent, s_sibling);
1548		if (!next->s_element)
1549			continue;
1550
1551		name = configfs_get_name(next);
1552		len = strlen(name);
1553
1554		/*
1555		 * We'll have a dentry and an inode for
1556		 * PINNED items and for open attribute
1557		 * files.  We lock here to prevent a race
1558		 * with configfs_d_iput() clearing
1559		 * s_dentry before calling iput().
1560		 *
1561		 * Why do we go to the trouble?  If
1562		 * someone has an attribute file open,
1563		 * the inode number should match until
1564		 * they close it.  Beyond that, we don't
1565		 * care.
1566		 */
1567		spin_lock(&configfs_dirent_lock);
1568		dentry = next->s_dentry;
1569		if (dentry)
1570			inode = d_inode(dentry);
1571		if (inode)
1572			ino = inode->i_ino;
1573		spin_unlock(&configfs_dirent_lock);
1574		if (!inode)
1575			ino = iunique(sb, 2);
1576
1577		if (!dir_emit(ctx, name, len, ino, dt_type(next)))
1578			return 0;
1579
1580		spin_lock(&configfs_dirent_lock);
1581		list_move(q, p);
1582		spin_unlock(&configfs_dirent_lock);
1583		p = q;
1584		ctx->pos++;
1585	}
1586	return 0;
1587}
1588
1589static loff_t configfs_dir_lseek(struct file *file, loff_t offset, int whence)
1590{
1591	struct dentry * dentry = file->f_path.dentry;
1592
1593	mutex_lock(&d_inode(dentry)->i_mutex);
1594	switch (whence) {
1595		case 1:
1596			offset += file->f_pos;
1597		case 0:
1598			if (offset >= 0)
1599				break;
1600		default:
1601			mutex_unlock(&d_inode(dentry)->i_mutex);
1602			return -EINVAL;
1603	}
1604	if (offset != file->f_pos) {
1605		file->f_pos = offset;
1606		if (file->f_pos >= 2) {
1607			struct configfs_dirent *sd = dentry->d_fsdata;
1608			struct configfs_dirent *cursor = file->private_data;
1609			struct list_head *p;
1610			loff_t n = file->f_pos - 2;
1611
1612			spin_lock(&configfs_dirent_lock);
1613			list_del(&cursor->s_sibling);
1614			p = sd->s_children.next;
1615			while (n && p != &sd->s_children) {
1616				struct configfs_dirent *next;
1617				next = list_entry(p, struct configfs_dirent,
1618						   s_sibling);
1619				if (next->s_element)
1620					n--;
1621				p = p->next;
1622			}
1623			list_add_tail(&cursor->s_sibling, p);
1624			spin_unlock(&configfs_dirent_lock);
1625		}
1626	}
1627	mutex_unlock(&d_inode(dentry)->i_mutex);
1628	return offset;
1629}
1630
1631const struct file_operations configfs_dir_operations = {
1632	.open		= configfs_dir_open,
1633	.release	= configfs_dir_close,
1634	.llseek		= configfs_dir_lseek,
1635	.read		= generic_read_dir,
1636	.iterate	= configfs_readdir,
1637};
1638
1639int configfs_register_subsystem(struct configfs_subsystem *subsys)
1640{
1641	int err;
1642	struct config_group *group = &subsys->su_group;
1643	struct dentry *dentry;
1644	struct dentry *root;
1645	struct configfs_dirent *sd;
1646
1647	root = configfs_pin_fs();
1648	if (IS_ERR(root))
1649		return PTR_ERR(root);
1650
1651	if (!group->cg_item.ci_name)
1652		group->cg_item.ci_name = group->cg_item.ci_namebuf;
1653
1654	sd = root->d_fsdata;
1655	link_group(to_config_group(sd->s_element), group);
1656
1657	mutex_lock_nested(&d_inode(root)->i_mutex, I_MUTEX_PARENT);
1658
1659	err = -ENOMEM;
1660	dentry = d_alloc_name(root, group->cg_item.ci_name);
1661	if (dentry) {
1662		d_add(dentry, NULL);
1663
1664		err = configfs_attach_group(sd->s_element, &group->cg_item,
1665					    dentry);
1666		if (err) {
1667			BUG_ON(d_inode(dentry));
1668			d_drop(dentry);
1669			dput(dentry);
1670		} else {
1671			spin_lock(&configfs_dirent_lock);
1672			configfs_dir_set_ready(dentry->d_fsdata);
1673			spin_unlock(&configfs_dirent_lock);
1674		}
1675	}
1676
1677	mutex_unlock(&d_inode(root)->i_mutex);
1678
1679	if (err) {
1680		unlink_group(group);
1681		configfs_release_fs();
1682	}
1683
1684	return err;
1685}
1686
1687void configfs_unregister_subsystem(struct configfs_subsystem *subsys)
1688{
1689	struct config_group *group = &subsys->su_group;
1690	struct dentry *dentry = group->cg_item.ci_dentry;
1691	struct dentry *root = dentry->d_sb->s_root;
1692
1693	if (dentry->d_parent != root) {
1694		pr_err("Tried to unregister non-subsystem!\n");
1695		return;
1696	}
1697
1698	mutex_lock_nested(&d_inode(root)->i_mutex,
1699			  I_MUTEX_PARENT);
1700	mutex_lock_nested(&d_inode(dentry)->i_mutex, I_MUTEX_CHILD);
1701	mutex_lock(&configfs_symlink_mutex);
1702	spin_lock(&configfs_dirent_lock);
1703	if (configfs_detach_prep(dentry, NULL)) {
1704		pr_err("Tried to unregister non-empty subsystem!\n");
1705	}
1706	spin_unlock(&configfs_dirent_lock);
1707	mutex_unlock(&configfs_symlink_mutex);
1708	configfs_detach_group(&group->cg_item);
1709	d_inode(dentry)->i_flags |= S_DEAD;
1710	dont_mount(dentry);
1711	mutex_unlock(&d_inode(dentry)->i_mutex);
1712
1713	d_delete(dentry);
1714
1715	mutex_unlock(&d_inode(root)->i_mutex);
1716
1717	dput(dentry);
1718
1719	unlink_group(group);
1720	configfs_release_fs();
1721}
1722
1723EXPORT_SYMBOL(configfs_register_subsystem);
1724EXPORT_SYMBOL(configfs_unregister_subsystem);
1725