[net-next-2.6.git] / fs / notify / mark.c

/*
 *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2, or (at your option)
 *  any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; see the file COPYING.  If not, write to
 *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/*
 * fsnotify inode mark locking/lifetime/and refcnting
 *
 * REFCNT:
 * The mark->refcnt tells how many "things" in the kernel currently are
 * referencing this object.  The object typically will live inside the kernel
 * with a refcnt of 2, one for each list it is on (i_list, g_list).  Any task
 * which can find this object holding the appropriete locks, can take a reference
 * and the object itself is guarenteed to survive until the reference is dropped.
 *
 * LOCKING:
 * There are 3 spinlocks involved with fsnotify inode marks and they MUST
 * be taken in order as follows:
 *
 * mark->lock
 * group->mark_lock
 * inode->i_lock
 *
 * mark->lock protects 2 things, mark->group and mark->inode.  You must hold
 * that lock to dereference either of these things (they could be NULL even with
 * the lock)
 *
 * group->mark_lock protects the marks_list anchored inside a given group
 * and each mark is hooked via the g_list.  It also sorta protects the
 * free_g_list, which when used is anchored by a private list on the stack of the
 * task which held the group->mark_lock.
 *
 * inode->i_lock protects the i_fsnotify_marks list anchored inside a
 * given inode and each mark is hooked via the i_list. (and sorta the
 * free_i_list)
 *
 *
 * LIFETIME:
 * Inode marks survive between when they are added to an inode and when their
 * refcnt==0.
 *
 * The inode mark can be cleared for a number of different reasons including:
 * - The inode is unlinked for the last time.  (fsnotify_inode_remove)
 * - The inode is being evicted from cache. (fsnotify_inode_delete)
 * - The fs the inode is on is unmounted.  (fsnotify_inode_delete/fsnotify_unmount_inodes)
 * - Something explicitly requests that it be removed.  (fsnotify_destroy_mark)
 * - The fsnotify_group associated with the mark is going away and all such marks
 *   need to be cleaned up. (fsnotify_clear_marks_by_group)
 *
 * Worst case we are given an inode and need to clean up all the marks on that
 * inode.  We take i_lock and walk the i_fsnotify_marks safely.  For each
 * mark on the list we take a reference (so the mark can't disappear under us).
 * We remove that mark form the inode's list of marks and we add this mark to a
 * private list anchored on the stack using i_free_list;  At this point we no
 * longer fear anything finding the mark using the inode's list of marks.
 *
 * We can safely and locklessly run the private list on the stack of everything
 * we just unattached from the original inode.  For each mark on the private list
 * we grab the mark-> and can thus dereference mark->group and mark->inode.  If
 * we see the group and inode are not NULL we take those locks.  Now holding all
 * 3 locks we can completely remove the mark from other tasks finding it in the
 * future.  Remember, 10 things might already be referencing this mark, but they
 * better be holding a ref.  We drop our reference we took before we unhooked it
 * from the inode.  When the ref hits 0 we can free the mark.
 *
 * Very similarly for freeing by group, except we use free_g_list.
 *
 * This has the very interesting property of being able to run concurrently with
 * any (or all) other directions.
 */

#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/writeback.h> /* for inode_lock */

#include <asm/atomic.h>

#include <linux/fsnotify_backend.h>
#include "fsnotify.h"

void fsnotify_get_mark(struct fsnotify_mark *mark)
{
	atomic_inc(&mark->refcnt);
}

void fsnotify_put_mark(struct fsnotify_mark *mark)
{
	if (atomic_dec_and_test(&mark->refcnt))
		mark->free_mark(mark);
}

/*
 * Any time a mark is getting freed we end up here.
 * The caller had better be holding a reference to this mark so we don't actually
 * do the final put under the mark->lock
 */
void fsnotify_destroy_mark(struct fsnotify_mark *mark)
{
	struct fsnotify_group *group;
	struct inode *inode = NULL;

	spin_lock(&mark->lock);

	group = mark->group;

	/* something else already called this function on this mark */
	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
		spin_unlock(&mark->lock);
		return;
	}

	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;

	/* 1 from caller and 1 for being on i_list/g_list */
	BUG_ON(atomic_read(&mark->refcnt) < 2);

	spin_lock(&group->mark_lock);

	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
		inode = mark->i.inode;
		fsnotify_destroy_inode_mark(mark);
	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
		fsnotify_destroy_vfsmount_mark(mark);
	else
		BUG();

	list_del_init(&mark->g_list);

	fsnotify_put_mark(mark); /* for i_list and g_list */

	spin_unlock(&group->mark_lock);
	spin_unlock(&mark->lock);

	/*
	 * Some groups like to know that marks are being freed.  This is a
	 * callback to the group function to let it know that this mark
	 * is being freed.
	 */
	if (group->ops->freeing_mark)
		group->ops->freeing_mark(mark, group);

	/*
	 * __fsnotify_update_child_dentry_flags(inode);
	 *
	 * I really want to call that, but we can't, we have no idea if the inode
	 * still exists the second we drop the mark->lock.
	 *
	 * The next time an event arrive to this inode from one of it's children
	 * __fsnotify_parent will see that the inode doesn't care about it's
	 * children and will update all of these flags then.  So really this
	 * is just a lazy update (and could be a perf win...)
	 */

	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
		iput(inode);

	/*
	 * it's possible that this group tried to destroy itself, but this
	 * this mark was simultaneously being freed by inode.  If that's the
	 * case, we finish freeing the group here.
	 */
	if (unlikely(atomic_dec_and_test(&group->num_marks)))
		fsnotify_final_destroy_group(group);
}

void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
{
	assert_spin_locked(&mark->lock);

	mark->mask = mask;

	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
		fsnotify_set_inode_mark_mask_locked(mark, mask);
}

void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
{
	assert_spin_locked(&mark->lock);

	mark->ignored_mask = mask;
}

/*
 * Attach an initialized mark to a given group and fs object.
 * These marks may be used for the fsnotify backend to determine which
 * event types should be delivered to which group.
 */
int fsnotify_add_mark(struct fsnotify_mark *mark,
		      struct fsnotify_group *group, struct inode *inode,
		      struct vfsmount *mnt, int allow_dups)
{
	int ret = 0;

	BUG_ON(inode && mnt);
	BUG_ON(!inode && !mnt);

	/*
	 * if this group isn't being testing for inode type events we need
	 * to start testing
	 */
	if (inode && unlikely(list_empty(&group->inode_group_list)))
		fsnotify_add_inode_group(group);
	else if (mnt && unlikely(list_empty(&group->vfsmount_group_list)))
		fsnotify_add_vfsmount_group(group);

	/*
	 * LOCKING ORDER!!!!
	 * mark->lock
	 * group->mark_lock
	 * inode->i_lock
	 */
	spin_lock(&mark->lock);
	spin_lock(&group->mark_lock);

	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;

	mark->group = group;
	list_add(&mark->g_list, &group->marks_list);
	atomic_inc(&group->num_marks);
	fsnotify_get_mark(mark); /* for i_list and g_list */

	if (inode) {
		ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
		if (ret)
			goto err;
	} else if (mnt) {
		ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
		if (ret)
			goto err;
	} else {
		BUG();
	}

	spin_unlock(&group->mark_lock);

	/* this will pin the object if appropriate */
	fsnotify_set_mark_mask_locked(mark, mark->mask);

	spin_unlock(&mark->lock);

	if (inode)
		__fsnotify_update_child_dentry_flags(inode);

	return ret;
err:
	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
	list_del_init(&mark->g_list);
	atomic_dec(&group->num_marks);
	fsnotify_put_mark(mark);

	spin_unlock(&group->mark_lock);
	spin_unlock(&mark->lock);

	return ret;
}

/*
 * clear any marks in a group in which mark->flags & flags is true
 */
void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
					 unsigned int flags)
{
	struct fsnotify_mark *lmark, *mark;
	LIST_HEAD(free_list);

	spin_lock(&group->mark_lock);
	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
		if (mark->flags & flags) {
			list_add(&mark->free_g_list, &free_list);
			list_del_init(&mark->g_list);
			fsnotify_get_mark(mark);
		}
	}
	spin_unlock(&group->mark_lock);

	list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) {
		fsnotify_destroy_mark(mark);
		fsnotify_put_mark(mark);
	}
}

/*
 * Given a group, destroy all of the marks associated with that group.
 */
void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
{
	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
}

void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
{
	assert_spin_locked(&old->lock);
	new->i.inode = old->i.inode;
	new->m.mnt = old->m.mnt;
	new->group = old->group;
	new->mask = old->mask;
	new->free_mark = old->free_mark;
}

/*
 * Nothing fancy, just initialize lists and locks and counters.
 */
void fsnotify_init_mark(struct fsnotify_mark *mark,
			void (*free_mark)(struct fsnotify_mark *mark))
{
	memset(mark, 0, sizeof(*mark));
	spin_lock_init(&mark->lock);
	atomic_set(&mark->refcnt, 1);
	mark->free_mark = free_mark;
}
Commit	Line	Data
5444e298 EP	1	/*
	2	* Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
	3	*
	4	* This program is free software; you can redistribute it and/or modify
	5	* it under the terms of the GNU General Public License as published by
	6	* the Free Software Foundation; either version 2, or (at your option)
	7	* any later version.
	8	*
	9	* This program is distributed in the hope that it will be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	12	* GNU General Public License for more details.
	13	*
	14	* You should have received a copy of the GNU General Public License
	15	* along with this program; see the file COPYING. If not, write to
	16	* the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
	17	*/
	18
	19	/*
	20	* fsnotify inode mark locking/lifetime/and refcnting
	21	*
	22	* REFCNT:
	23	* The mark->refcnt tells how many "things" in the kernel currently are
	24	* referencing this object. The object typically will live inside the kernel
	25	* with a refcnt of 2, one for each list it is on (i_list, g_list). Any task
	26	* which can find this object holding the appropriete locks, can take a reference
	27	* and the object itself is guarenteed to survive until the reference is dropped.
	28	*
	29	* LOCKING:
	30	* There are 3 spinlocks involved with fsnotify inode marks and they MUST
	31	* be taken in order as follows:
	32	*
	33	* mark->lock
	34	* group->mark_lock
	35	* inode->i_lock
	36	*
	37	* mark->lock protects 2 things, mark->group and mark->inode. You must hold
	38	* that lock to dereference either of these things (they could be NULL even with
	39	* the lock)
	40	*
	41	* group->mark_lock protects the marks_list anchored inside a given group
	42	* and each mark is hooked via the g_list. It also sorta protects the
	43	* free_g_list, which when used is anchored by a private list on the stack of the
	44	* task which held the group->mark_lock.
	45	*
	46	* inode->i_lock protects the i_fsnotify_marks list anchored inside a
	47	* given inode and each mark is hooked via the i_list. (and sorta the
	48	* free_i_list)
	49	*
	50	*
	51	* LIFETIME:
	52	* Inode marks survive between when they are added to an inode and when their
	53	* refcnt==0.
	54	*
	55	* The inode mark can be cleared for a number of different reasons including:
	56	* - The inode is unlinked for the last time. (fsnotify_inode_remove)
	57	* - The inode is being evicted from cache. (fsnotify_inode_delete)
	58	* - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes)
	59	* - Something explicitly requests that it be removed. (fsnotify_destroy_mark)
	60	* - The fsnotify_group associated with the mark is going away and all such marks
	61	* need to be cleaned up. (fsnotify_clear_marks_by_group)
	62	*
	63	* Worst case we are given an inode and need to clean up all the marks on that
	64	* inode. We take i_lock and walk the i_fsnotify_marks safely. For each
65	* mark on the list we take a reference (so the mark can't disappear under us).
66	* We remove that mark form the inode's list of marks and we add this mark to a
67	* private list anchored on the stack using i_free_list; At this point we no
68	* longer fear anything finding the mark using the inode's list of marks.
69	*
70	* We can safely and locklessly run the private list on the stack of everything
71	* we just unattached from the original inode. For each mark on the private list
72	* we grab the mark-> and can thus dereference mark->group and mark->inode. If
73	* we see the group and inode are not NULL we take those locks. Now holding all
74	* 3 locks we can completely remove the mark from other tasks finding it in the
75	* future. Remember, 10 things might already be referencing this mark, but they
76	* better be holding a ref. We drop our reference we took before we unhooked it
77	* from the inode. When the ref hits 0 we can free the mark.
78	*
79	* Very similarly for freeing by group, except we use free_g_list.
80	*
81	* This has the very interesting property of being able to run concurrently with
82	* any (or all) other directions.
83	*/
84
85	#include <linux/fs.h>
86	#include <linux/init.h>
87	#include <linux/kernel.h>
88	#include <linux/module.h>
89	#include <linux/mutex.h>
90	#include <linux/slab.h>
91	#include <linux/spinlock.h>
92	#include <linux/writeback.h> /* for inode_lock */
93
94	#include <asm/atomic.h>
95
96	#include <linux/fsnotify_backend.h>
97	#include "fsnotify.h"
98
99	void fsnotify_get_mark(struct fsnotify_mark *mark)
100	{
101	atomic_inc(&mark->refcnt);
102	}
103
104	void fsnotify_put_mark(struct fsnotify_mark *mark)
105	{
106	if (atomic_dec_and_test(&mark->refcnt))
107	mark->free_mark(mark);
108	}
109
110	/*
111	* Any time a mark is getting freed we end up here.
112	* The caller had better be holding a reference to this mark so we don't actually
113	* do the final put under the mark->lock
114	*/
115	void fsnotify_destroy_mark(struct fsnotify_mark *mark)
116	{
117	struct fsnotify_group *group;
0d48b7f0	118	struct inode *inode = NULL;
5444e298 EP	119
	120	spin_lock(&mark->lock);
	121
	122	group = mark->group;
5444e298	123
700307a2 EP	124	/* something else already called this function on this mark */
700307a2 EP	125	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
5444e298 EP	126	spin_unlock(&mark->lock);
	127	return;
	128	}
	129
700307a2 EP	130	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
700307a2 EP	131
5444e298 EP	132	/* 1 from caller and 1 for being on i_list/g_list */
	133	BUG_ON(atomic_read(&mark->refcnt) < 2);
	134
	135	spin_lock(&group->mark_lock);
	136
0d48b7f0	137	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
0d48b7f0	138	inode = mark->i.inode;
b31d397e	139	fsnotify_destroy_inode_mark(mark);
0d48b7f0 EP	140	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
0d48b7f0 EP	141	fsnotify_destroy_vfsmount_mark(mark);
5444e298 EP	142	else
	143	BUG();
	144
	145	list_del_init(&mark->g_list);
5444e298 EP	146
	147	fsnotify_put_mark(mark); /* for i_list and g_list */
	148
	149	spin_unlock(&group->mark_lock);
	150	spin_unlock(&mark->lock);
	151
	152	/*
	153	* Some groups like to know that marks are being freed. This is a
	154	* callback to the group function to let it know that this mark
	155	* is being freed.
	156	*/
	157	if (group->ops->freeing_mark)
	158	group->ops->freeing_mark(mark, group);
	159
	160	/*
	161	* __fsnotify_update_child_dentry_flags(inode);
	162	*
	163	* I really want to call that, but we can't, we have no idea if the inode
	164	* still exists the second we drop the mark->lock.
	165	*
	166	* The next time an event arrive to this inode from one of it's children
	167	* __fsnotify_parent will see that the inode doesn't care about it's
	168	* children and will update all of these flags then. So really this
	169	* is just a lazy update (and could be a perf win...)
	170	*/
	171
90b1e7a5	172	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
0d48b7f0	173	iput(inode);
5444e298 EP	174
	175	/*
	176	* it's possible that this group tried to destroy itself, but this
	177	* this mark was simultaneously being freed by inode. If that's the
	178	* case, we finish freeing the group here.
	179	*/
	180	if (unlikely(atomic_dec_and_test(&group->num_marks)))
	181	fsnotify_final_destroy_group(group);
	182	}
	183
90b1e7a5 EP	184	void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
	185	{
	186	assert_spin_locked(&mark->lock);
	187
	188	mark->mask = mask;
	189
	190	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
	191	fsnotify_set_inode_mark_mask_locked(mark, mask);
	192	}
	193
33af5e32 EP	194	void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
	195	{
	196	assert_spin_locked(&mark->lock);
	197
	198	mark->ignored_mask = mask;
	199	}
90b1e7a5	200
5444e298 EP	201	/*
	202	* Attach an initialized mark to a given group and fs object.
	203	* These marks may be used for the fsnotify backend to determine which
	204	* event types should be delivered to which group.
	205	*/
	206	int fsnotify_add_mark(struct fsnotify_mark *mark,
	207	struct fsnotify_group group, struct inode inode,
	208	struct vfsmount *mnt, int allow_dups)
	209	{
	210	int ret = 0;
	211
5444e298 EP	212	BUG_ON(inode && mnt);
	213	BUG_ON(!inode && !mnt);
	214
	215	/*
	216	* if this group isn't being testing for inode type events we need
	217	* to start testing
	218	*/
	219	if (inode && unlikely(list_empty(&group->inode_group_list)))
	220	fsnotify_add_inode_group(group);
	221	else if (mnt && unlikely(list_empty(&group->vfsmount_group_list)))
	222	fsnotify_add_vfsmount_group(group);
	223
	224	/*
	225	* LOCKING ORDER!!!!
	226	* mark->lock
	227	* group->mark_lock
	228	* inode->i_lock
	229	*/
	230	spin_lock(&mark->lock);
	231	spin_lock(&group->mark_lock);
	232
700307a2 EP	233	mark->flags \|= FSNOTIFY_MARK_FLAG_ALIVE;
700307a2 EP	234
5444e298 EP	235	mark->group = group;
	236	list_add(&mark->g_list, &group->marks_list);
	237	atomic_inc(&group->num_marks);
	238	fsnotify_get_mark(mark); /* for i_list and g_list */
	239
	240	if (inode) {
	241	ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
	242	if (ret)
	243	goto err;
0d48b7f0 EP	244	} else if (mnt) {
	245	ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
	246	if (ret)
	247	goto err;
5444e298 EP	248	} else {
	249	BUG();
	250	}
	251
	252	spin_unlock(&group->mark_lock);
90b1e7a5 EP	253
	254	/* this will pin the object if appropriate */
	255	fsnotify_set_mark_mask_locked(mark, mark->mask);
	256
5444e298 EP	257	spin_unlock(&mark->lock);
	258
	259	if (inode)
	260	__fsnotify_update_child_dentry_flags(inode);
	261
	262	return ret;
	263	err:
700307a2	264	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
5444e298 EP	265	list_del_init(&mark->g_list);
	266	atomic_dec(&group->num_marks);
	267	fsnotify_put_mark(mark);
	268
	269	spin_unlock(&group->mark_lock);
	270	spin_unlock(&mark->lock);
	271
	272	return ret;
	273	}
	274
	275	/*
4d92604c	276	* clear any marks in a group in which mark->flags & flags is true
5444e298	277	*/
4d92604c EP	278	void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
4d92604c EP	279	unsigned int flags)
5444e298 EP	280	{
	281	struct fsnotify_mark lmark, mark;
	282	LIST_HEAD(free_list);
	283
	284	spin_lock(&group->mark_lock);
	285	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
4d92604c EP	286	if (mark->flags & flags) {
	287	list_add(&mark->free_g_list, &free_list);
	288	list_del_init(&mark->g_list);
	289	fsnotify_get_mark(mark);
	290	}
5444e298 EP	291	}
	292	spin_unlock(&group->mark_lock);
	293
	294	list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) {
	295	fsnotify_destroy_mark(mark);
	296	fsnotify_put_mark(mark);
	297	}
	298	}
	299
4d92604c EP	300	/*
	301	* Given a group, destroy all of the marks associated with that group.
	302	*/
	303	void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
	304	{
	305	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
	306	}
	307
5444e298 EP	308	void fsnotify_duplicate_mark(struct fsnotify_mark new, struct fsnotify_mark old)
	309	{
	310	assert_spin_locked(&old->lock);
	311	new->i.inode = old->i.inode;
	312	new->m.mnt = old->m.mnt;
	313	new->group = old->group;
	314	new->mask = old->mask;
	315	new->free_mark = old->free_mark;
	316	}
	317
	318	/*
	319	* Nothing fancy, just initialize lists and locks and counters.
	320	*/
	321	void fsnotify_init_mark(struct fsnotify_mark *mark,
	322	void (free_mark)(struct fsnotify_mark mark))
	323	{
ba643f04	324	memset(mark, 0, sizeof(*mark));
5444e298 EP	325	spin_lock_init(&mark->lock);
5444e298 EP	326	atomic_set(&mark->refcnt, 1);
5444e298 EP	327	mark->free_mark = free_mark;
5444e298 EP	328	}