2 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
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)
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.
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.
20 * fsnotify inode mark locking/lifetime/and refcnting
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.
30 * There are 3 spinlocks involved with fsnotify inode marks and they MUST
31 * be taken in order as follows:
37 * entry->lock protects 2 things, entry->group and entry->inode. You must hold
38 * that lock to dereference either of these things (they could be NULL even with
41 * group->mark_lock protects the mark_entries list anchored inside a given group
42 * and each entry 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.
46 * inode->i_lock protects the i_fsnotify_mark_entries list anchored inside a
47 * given inode and each entry is hooked via the i_list. (and sorta the
52 * Inode marks survive between when they are added to an inode and when their
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_by_entry)
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)
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_mark_entries 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.
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.
79 * Very similarly for freeing by group, except we use free_g_list.
81 * This has the very interesting property of being able to run concurrently with
82 * any (or all) other directions.
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>
93 #include <asm/atomic.h>
95 #include <linux/fsnotify_backend.h>
98 void fsnotify_get_mark(struct fsnotify_mark_entry
*entry
)
100 atomic_inc(&entry
->refcnt
);
103 void fsnotify_put_mark(struct fsnotify_mark_entry
*entry
)
105 if (atomic_dec_and_test(&entry
->refcnt
))
106 entry
->free_mark(entry
);
110 * Recalculate the mask of events relevant to a given inode locked.
112 static void fsnotify_recalc_inode_mask_locked(struct inode
*inode
)
114 struct fsnotify_mark_entry
*entry
;
115 struct hlist_node
*pos
;
118 assert_spin_locked(&inode
->i_lock
);
120 hlist_for_each_entry(entry
, pos
, &inode
->i_fsnotify_mark_entries
, i_list
)
121 new_mask
|= entry
->mask
;
122 inode
->i_fsnotify_mask
= new_mask
;
126 * Recalculate the inode->i_fsnotify_mask, or the mask of all FS_* event types
127 * any notifier is interested in hearing for this inode.
129 void fsnotify_recalc_inode_mask(struct inode
*inode
)
131 spin_lock(&inode
->i_lock
);
132 fsnotify_recalc_inode_mask_locked(inode
);
133 spin_unlock(&inode
->i_lock
);
137 * Any time a mark is getting freed we end up here.
138 * The caller had better be holding a reference to this mark so we don't actually
139 * do the final put under the entry->lock
141 void fsnotify_destroy_mark_by_entry(struct fsnotify_mark_entry
*entry
)
143 struct fsnotify_group
*group
;
146 spin_lock(&entry
->lock
);
148 group
= entry
->group
;
149 inode
= entry
->inode
;
151 BUG_ON(group
&& !inode
);
152 BUG_ON(!group
&& inode
);
154 /* if !group something else already marked this to die */
156 spin_unlock(&entry
->lock
);
160 /* 1 from caller and 1 for being on i_list/g_list */
161 BUG_ON(atomic_read(&entry
->refcnt
) < 2);
163 spin_lock(&group
->mark_lock
);
164 spin_lock(&inode
->i_lock
);
166 hlist_del_init(&entry
->i_list
);
169 list_del_init(&entry
->g_list
);
172 fsnotify_put_mark(entry
); /* for i_list and g_list */
175 * this mark is now off the inode->i_fsnotify_mark_entries list and we
176 * hold the inode->i_lock, so this is the perfect time to update the
177 * inode->i_fsnotify_mask
179 fsnotify_recalc_inode_mask_locked(inode
);
181 spin_unlock(&inode
->i_lock
);
182 spin_unlock(&group
->mark_lock
);
183 spin_unlock(&entry
->lock
);
186 * Some groups like to know that marks are being freed. This is a
187 * callback to the group function to let it know that this entry
190 group
->ops
->freeing_mark(entry
, group
);
193 * it's possible that this group tried to destroy itself, but this
194 * this mark was simultaneously being freed by inode. If that's the
195 * case, we finish freeing the group here.
197 if (unlikely(atomic_dec_and_test(&group
->num_marks
)))
198 fsnotify_final_destroy_group(group
);
202 * Given a group, destroy all of the marks associated with that group.
204 void fsnotify_clear_marks_by_group(struct fsnotify_group
*group
)
206 struct fsnotify_mark_entry
*lentry
, *entry
;
207 LIST_HEAD(free_list
);
209 spin_lock(&group
->mark_lock
);
210 list_for_each_entry_safe(entry
, lentry
, &group
->mark_entries
, g_list
) {
211 list_add(&entry
->free_g_list
, &free_list
);
212 list_del_init(&entry
->g_list
);
213 fsnotify_get_mark(entry
);
215 spin_unlock(&group
->mark_lock
);
217 list_for_each_entry_safe(entry
, lentry
, &free_list
, free_g_list
) {
218 fsnotify_destroy_mark_by_entry(entry
);
219 fsnotify_put_mark(entry
);
224 * Given an inode, destroy all of the marks associated with that inode.
226 void fsnotify_clear_marks_by_inode(struct inode
*inode
)
228 struct fsnotify_mark_entry
*entry
, *lentry
;
229 struct hlist_node
*pos
, *n
;
230 LIST_HEAD(free_list
);
232 spin_lock(&inode
->i_lock
);
233 hlist_for_each_entry_safe(entry
, pos
, n
, &inode
->i_fsnotify_mark_entries
, i_list
) {
234 list_add(&entry
->free_i_list
, &free_list
);
235 hlist_del_init(&entry
->i_list
);
236 fsnotify_get_mark(entry
);
238 spin_unlock(&inode
->i_lock
);
240 list_for_each_entry_safe(entry
, lentry
, &free_list
, free_i_list
) {
241 fsnotify_destroy_mark_by_entry(entry
);
242 fsnotify_put_mark(entry
);
247 * given a group and inode, find the mark associated with that combination.
248 * if found take a reference to that mark and return it, else return NULL
250 struct fsnotify_mark_entry
*fsnotify_find_mark_entry(struct fsnotify_group
*group
,
253 struct fsnotify_mark_entry
*entry
;
254 struct hlist_node
*pos
;
256 assert_spin_locked(&inode
->i_lock
);
258 hlist_for_each_entry(entry
, pos
, &inode
->i_fsnotify_mark_entries
, i_list
) {
259 if (entry
->group
== group
) {
260 fsnotify_get_mark(entry
);
268 * Nothing fancy, just initialize lists and locks and counters.
270 void fsnotify_init_mark(struct fsnotify_mark_entry
*entry
,
271 void (*free_mark
)(struct fsnotify_mark_entry
*entry
))
274 spin_lock_init(&entry
->lock
);
275 atomic_set(&entry
->refcnt
, 1);
276 INIT_HLIST_NODE(&entry
->i_list
);
280 entry
->free_mark
= free_mark
;
284 * Attach an initialized mark entry to a given group and inode.
285 * These marks may be used for the fsnotify backend to determine which
286 * event types should be delivered to which group and for which inodes.
288 int fsnotify_add_mark(struct fsnotify_mark_entry
*entry
,
289 struct fsnotify_group
*group
, struct inode
*inode
)
291 struct fsnotify_mark_entry
*lentry
;
300 spin_lock(&entry
->lock
);
301 spin_lock(&group
->mark_lock
);
302 spin_lock(&inode
->i_lock
);
304 entry
->group
= group
;
305 entry
->inode
= inode
;
307 lentry
= fsnotify_find_mark_entry(group
, inode
);
309 hlist_add_head(&entry
->i_list
, &inode
->i_fsnotify_mark_entries
);
310 list_add(&entry
->g_list
, &group
->mark_entries
);
312 fsnotify_get_mark(entry
); /* for i_list and g_list */
314 atomic_inc(&group
->num_marks
);
316 fsnotify_recalc_inode_mask_locked(inode
);
319 spin_unlock(&inode
->i_lock
);
320 spin_unlock(&group
->mark_lock
);
321 spin_unlock(&entry
->lock
);
325 fsnotify_put_mark(lentry
);
This page took 0.038929 seconds and 6 git commands to generate.