Loading...
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 guaranteed 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/kthread.h>
89#include <linux/module.h>
90#include <linux/mutex.h>
91#include <linux/slab.h>
92#include <linux/spinlock.h>
93#include <linux/srcu.h>
94
95#include <linux/atomic.h>
96
97#include <linux/fsnotify_backend.h>
98#include "fsnotify.h"
99
100struct srcu_struct fsnotify_mark_srcu;
101static DEFINE_SPINLOCK(destroy_lock);
102static LIST_HEAD(destroy_list);
103static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);
104
105void fsnotify_get_mark(struct fsnotify_mark *mark)
106{
107 atomic_inc(&mark->refcnt);
108}
109
110void fsnotify_put_mark(struct fsnotify_mark *mark)
111{
112 if (atomic_dec_and_test(&mark->refcnt))
113 mark->free_mark(mark);
114}
115
116/*
117 * Any time a mark is getting freed we end up here.
118 * The caller had better be holding a reference to this mark so we don't actually
119 * do the final put under the mark->lock
120 */
121void fsnotify_destroy_mark(struct fsnotify_mark *mark)
122{
123 struct fsnotify_group *group;
124 struct inode *inode = NULL;
125
126 spin_lock(&mark->lock);
127
128 group = mark->group;
129
130 /* something else already called this function on this mark */
131 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
132 spin_unlock(&mark->lock);
133 return;
134 }
135
136 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
137
138 /* 1 from caller and 1 for being on i_list/g_list */
139 BUG_ON(atomic_read(&mark->refcnt) < 2);
140
141 spin_lock(&group->mark_lock);
142
143 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
144 inode = mark->i.inode;
145 fsnotify_destroy_inode_mark(mark);
146 } else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
147 fsnotify_destroy_vfsmount_mark(mark);
148 else
149 BUG();
150
151 list_del_init(&mark->g_list);
152
153 spin_unlock(&group->mark_lock);
154 spin_unlock(&mark->lock);
155
156 spin_lock(&destroy_lock);
157 list_add(&mark->destroy_list, &destroy_list);
158 spin_unlock(&destroy_lock);
159 wake_up(&destroy_waitq);
160
161 /*
162 * Some groups like to know that marks are being freed. This is a
163 * callback to the group function to let it know that this mark
164 * is being freed.
165 */
166 if (group->ops->freeing_mark)
167 group->ops->freeing_mark(mark, group);
168
169 /*
170 * __fsnotify_update_child_dentry_flags(inode);
171 *
172 * I really want to call that, but we can't, we have no idea if the inode
173 * still exists the second we drop the mark->lock.
174 *
175 * The next time an event arrive to this inode from one of it's children
176 * __fsnotify_parent will see that the inode doesn't care about it's
177 * children and will update all of these flags then. So really this
178 * is just a lazy update (and could be a perf win...)
179 */
180
181 if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
182 iput(inode);
183
184 /*
185 * it's possible that this group tried to destroy itself, but this
186 * this mark was simultaneously being freed by inode. If that's the
187 * case, we finish freeing the group here.
188 */
189 if (unlikely(atomic_dec_and_test(&group->num_marks)))
190 fsnotify_final_destroy_group(group);
191}
192
193void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
194{
195 assert_spin_locked(&mark->lock);
196
197 mark->mask = mask;
198
199 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
200 fsnotify_set_inode_mark_mask_locked(mark, mask);
201}
202
203void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
204{
205 assert_spin_locked(&mark->lock);
206
207 mark->ignored_mask = mask;
208}
209
210/*
211 * Attach an initialized mark to a given group and fs object.
212 * These marks may be used for the fsnotify backend to determine which
213 * event types should be delivered to which group.
214 */
215int fsnotify_add_mark(struct fsnotify_mark *mark,
216 struct fsnotify_group *group, struct inode *inode,
217 struct vfsmount *mnt, int allow_dups)
218{
219 int ret = 0;
220
221 BUG_ON(inode && mnt);
222 BUG_ON(!inode && !mnt);
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
233 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;
234
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;
244 } else if (mnt) {
245 ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
246 if (ret)
247 goto err;
248 } else {
249 BUG();
250 }
251
252 spin_unlock(&group->mark_lock);
253
254 /* this will pin the object if appropriate */
255 fsnotify_set_mark_mask_locked(mark, mark->mask);
256
257 spin_unlock(&mark->lock);
258
259 if (inode)
260 __fsnotify_update_child_dentry_flags(inode);
261
262 return ret;
263err:
264 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
265 list_del_init(&mark->g_list);
266 mark->group = NULL;
267 atomic_dec(&group->num_marks);
268
269 spin_unlock(&group->mark_lock);
270 spin_unlock(&mark->lock);
271
272 spin_lock(&destroy_lock);
273 list_add(&mark->destroy_list, &destroy_list);
274 spin_unlock(&destroy_lock);
275 wake_up(&destroy_waitq);
276
277 return ret;
278}
279
280/*
281 * clear any marks in a group in which mark->flags & flags is true
282 */
283void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
284 unsigned int flags)
285{
286 struct fsnotify_mark *lmark, *mark;
287 LIST_HEAD(free_list);
288
289 spin_lock(&group->mark_lock);
290 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
291 if (mark->flags & flags) {
292 list_add(&mark->free_g_list, &free_list);
293 list_del_init(&mark->g_list);
294 fsnotify_get_mark(mark);
295 }
296 }
297 spin_unlock(&group->mark_lock);
298
299 list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) {
300 fsnotify_destroy_mark(mark);
301 fsnotify_put_mark(mark);
302 }
303}
304
305/*
306 * Given a group, destroy all of the marks associated with that group.
307 */
308void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
309{
310 fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
311}
312
313void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
314{
315 assert_spin_locked(&old->lock);
316 new->i.inode = old->i.inode;
317 new->m.mnt = old->m.mnt;
318 new->group = old->group;
319 new->mask = old->mask;
320 new->free_mark = old->free_mark;
321}
322
323/*
324 * Nothing fancy, just initialize lists and locks and counters.
325 */
326void fsnotify_init_mark(struct fsnotify_mark *mark,
327 void (*free_mark)(struct fsnotify_mark *mark))
328{
329 memset(mark, 0, sizeof(*mark));
330 spin_lock_init(&mark->lock);
331 atomic_set(&mark->refcnt, 1);
332 mark->free_mark = free_mark;
333}
334
335static int fsnotify_mark_destroy(void *ignored)
336{
337 struct fsnotify_mark *mark, *next;
338 LIST_HEAD(private_destroy_list);
339
340 for (;;) {
341 spin_lock(&destroy_lock);
342 /* exchange the list head */
343 list_replace_init(&destroy_list, &private_destroy_list);
344 spin_unlock(&destroy_lock);
345
346 synchronize_srcu(&fsnotify_mark_srcu);
347
348 list_for_each_entry_safe(mark, next, &private_destroy_list, destroy_list) {
349 list_del_init(&mark->destroy_list);
350 fsnotify_put_mark(mark);
351 }
352
353 wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
354 }
355
356 return 0;
357}
358
359static int __init fsnotify_mark_init(void)
360{
361 struct task_struct *thread;
362
363 thread = kthread_run(fsnotify_mark_destroy, NULL,
364 "fsnotify_mark");
365 if (IS_ERR(thread))
366 panic("unable to start fsnotify mark destruction thread.");
367
368 return 0;
369}
370device_initcall(fsnotify_mark_init);
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 group->recnt and mark->refcnt tell how many "things" in the kernel
24 * currently are referencing the objects. Both kind of objects typically will
25 * live inside the kernel with a refcnt of 2, one for its creation and one for
26 * the reference a group and a mark hold to each other.
27 * If you are holding the appropriate locks, you can take a reference and the
28 * object itself is guaranteed to survive until the reference is dropped.
29 *
30 * LOCKING:
31 * There are 3 locks involved with fsnotify inode marks and they MUST be taken
32 * in order as follows:
33 *
34 * group->mark_mutex
35 * mark->lock
36 * inode->i_lock
37 *
38 * group->mark_mutex protects the marks_list anchored inside a given group and
39 * each mark is hooked via the g_list. It also protects the groups private
40 * data (i.e group limits).
41
42 * mark->lock protects the marks attributes like its masks and flags.
43 * Furthermore it protects the access to a reference of the group that the mark
44 * is assigned to as well as the access to a reference of the inode/vfsmount
45 * that is being watched by the mark.
46 *
47 * inode->i_lock protects the i_fsnotify_marks list anchored inside a
48 * given inode and each mark is hooked via the i_list. (and sorta the
49 * free_i_list)
50 *
51 *
52 * LIFETIME:
53 * Inode marks survive between when they are added to an inode and when their
54 * refcnt==0.
55 *
56 * The inode mark can be cleared for a number of different reasons including:
57 * - The inode is unlinked for the last time. (fsnotify_inode_remove)
58 * - The inode is being evicted from cache. (fsnotify_inode_delete)
59 * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes)
60 * - Something explicitly requests that it be removed. (fsnotify_destroy_mark)
61 * - The fsnotify_group associated with the mark is going away and all such marks
62 * need to be cleaned up. (fsnotify_clear_marks_by_group)
63 *
64 * Worst case we are given an inode and need to clean up all the marks on that
65 * inode. We take i_lock and walk the i_fsnotify_marks safely. For each
66 * mark on the list we take a reference (so the mark can't disappear under us).
67 * We remove that mark form the inode's list of marks and we add this mark to a
68 * private list anchored on the stack using i_free_list; we walk i_free_list
69 * and before we destroy the mark we make sure that we dont race with a
70 * concurrent destroy_group by getting a ref to the marks group and taking the
71 * groups mutex.
72
73 * Very similarly for freeing by group, except we use free_g_list.
74 *
75 * This has the very interesting property of being able to run concurrently with
76 * any (or all) other directions.
77 */
78
79#include <linux/fs.h>
80#include <linux/init.h>
81#include <linux/kernel.h>
82#include <linux/kthread.h>
83#include <linux/module.h>
84#include <linux/mutex.h>
85#include <linux/slab.h>
86#include <linux/spinlock.h>
87#include <linux/srcu.h>
88
89#include <linux/atomic.h>
90
91#include <linux/fsnotify_backend.h>
92#include "fsnotify.h"
93
94#define FSNOTIFY_REAPER_DELAY (1) /* 1 jiffy */
95
96struct srcu_struct fsnotify_mark_srcu;
97static DEFINE_SPINLOCK(destroy_lock);
98static LIST_HEAD(destroy_list);
99
100static void fsnotify_mark_destroy(struct work_struct *work);
101static DECLARE_DELAYED_WORK(reaper_work, fsnotify_mark_destroy);
102
103void fsnotify_get_mark(struct fsnotify_mark *mark)
104{
105 atomic_inc(&mark->refcnt);
106}
107
108void fsnotify_put_mark(struct fsnotify_mark *mark)
109{
110 if (atomic_dec_and_test(&mark->refcnt)) {
111 if (mark->group)
112 fsnotify_put_group(mark->group);
113 mark->free_mark(mark);
114 }
115}
116
117/* Calculate mask of events for a list of marks */
118u32 fsnotify_recalc_mask(struct hlist_head *head)
119{
120 u32 new_mask = 0;
121 struct fsnotify_mark *mark;
122
123 hlist_for_each_entry(mark, head, obj_list)
124 new_mask |= mark->mask;
125 return new_mask;
126}
127
128/*
129 * Remove mark from inode / vfsmount list, group list, drop inode reference
130 * if we got one.
131 *
132 * Must be called with group->mark_mutex held.
133 */
134void fsnotify_detach_mark(struct fsnotify_mark *mark)
135{
136 struct inode *inode = NULL;
137 struct fsnotify_group *group = mark->group;
138
139 BUG_ON(!mutex_is_locked(&group->mark_mutex));
140
141 spin_lock(&mark->lock);
142
143 /* something else already called this function on this mark */
144 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED)) {
145 spin_unlock(&mark->lock);
146 return;
147 }
148
149 mark->flags &= ~FSNOTIFY_MARK_FLAG_ATTACHED;
150
151 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
152 inode = mark->inode;
153 fsnotify_destroy_inode_mark(mark);
154 } else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
155 fsnotify_destroy_vfsmount_mark(mark);
156 else
157 BUG();
158 /*
159 * Note that we didn't update flags telling whether inode cares about
160 * what's happening with children. We update these flags from
161 * __fsnotify_parent() lazily when next event happens on one of our
162 * children.
163 */
164
165 list_del_init(&mark->g_list);
166
167 spin_unlock(&mark->lock);
168
169 if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
170 iput(inode);
171
172 atomic_dec(&group->num_marks);
173}
174
175/*
176 * Free fsnotify mark. The freeing is actually happening from a kthread which
177 * first waits for srcu period end. Caller must have a reference to the mark
178 * or be protected by fsnotify_mark_srcu.
179 */
180void fsnotify_free_mark(struct fsnotify_mark *mark)
181{
182 struct fsnotify_group *group = mark->group;
183
184 spin_lock(&mark->lock);
185 /* something else already called this function on this mark */
186 if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
187 spin_unlock(&mark->lock);
188 return;
189 }
190 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
191 spin_unlock(&mark->lock);
192
193 spin_lock(&destroy_lock);
194 list_add(&mark->g_list, &destroy_list);
195 spin_unlock(&destroy_lock);
196 queue_delayed_work(system_unbound_wq, &reaper_work,
197 FSNOTIFY_REAPER_DELAY);
198
199 /*
200 * Some groups like to know that marks are being freed. This is a
201 * callback to the group function to let it know that this mark
202 * is being freed.
203 */
204 if (group->ops->freeing_mark)
205 group->ops->freeing_mark(mark, group);
206}
207
208void fsnotify_destroy_mark(struct fsnotify_mark *mark,
209 struct fsnotify_group *group)
210{
211 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
212 fsnotify_detach_mark(mark);
213 mutex_unlock(&group->mark_mutex);
214 fsnotify_free_mark(mark);
215}
216
217void fsnotify_destroy_marks(struct hlist_head *head, spinlock_t *lock)
218{
219 struct fsnotify_mark *mark;
220
221 while (1) {
222 /*
223 * We have to be careful since we can race with e.g.
224 * fsnotify_clear_marks_by_group() and once we drop 'lock',
225 * mark can get removed from the obj_list and destroyed. But
226 * we are holding mark reference so mark cannot be freed and
227 * calling fsnotify_destroy_mark() more than once is fine.
228 */
229 spin_lock(lock);
230 if (hlist_empty(head)) {
231 spin_unlock(lock);
232 break;
233 }
234 mark = hlist_entry(head->first, struct fsnotify_mark, obj_list);
235 /*
236 * We don't update i_fsnotify_mask / mnt_fsnotify_mask here
237 * since inode / mount is going away anyway. So just remove
238 * mark from the list.
239 */
240 hlist_del_init_rcu(&mark->obj_list);
241 fsnotify_get_mark(mark);
242 spin_unlock(lock);
243 fsnotify_destroy_mark(mark, mark->group);
244 fsnotify_put_mark(mark);
245 }
246}
247
248void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
249{
250 assert_spin_locked(&mark->lock);
251
252 mark->mask = mask;
253
254 if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
255 fsnotify_set_inode_mark_mask_locked(mark, mask);
256}
257
258void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
259{
260 assert_spin_locked(&mark->lock);
261
262 mark->ignored_mask = mask;
263}
264
265/*
266 * Sorting function for lists of fsnotify marks.
267 *
268 * Fanotify supports different notification classes (reflected as priority of
269 * notification group). Events shall be passed to notification groups in
270 * decreasing priority order. To achieve this marks in notification lists for
271 * inodes and vfsmounts are sorted so that priorities of corresponding groups
272 * are descending.
273 *
274 * Furthermore correct handling of the ignore mask requires processing inode
275 * and vfsmount marks of each group together. Using the group address as
276 * further sort criterion provides a unique sorting order and thus we can
277 * merge inode and vfsmount lists of marks in linear time and find groups
278 * present in both lists.
279 *
280 * A return value of 1 signifies that b has priority over a.
281 * A return value of 0 signifies that the two marks have to be handled together.
282 * A return value of -1 signifies that a has priority over b.
283 */
284int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
285{
286 if (a == b)
287 return 0;
288 if (!a)
289 return 1;
290 if (!b)
291 return -1;
292 if (a->priority < b->priority)
293 return 1;
294 if (a->priority > b->priority)
295 return -1;
296 if (a < b)
297 return 1;
298 return -1;
299}
300
301/* Add mark into proper place in given list of marks */
302int fsnotify_add_mark_list(struct hlist_head *head, struct fsnotify_mark *mark,
303 int allow_dups)
304{
305 struct fsnotify_mark *lmark, *last = NULL;
306 int cmp;
307
308 /* is mark the first mark? */
309 if (hlist_empty(head)) {
310 hlist_add_head_rcu(&mark->obj_list, head);
311 return 0;
312 }
313
314 /* should mark be in the middle of the current list? */
315 hlist_for_each_entry(lmark, head, obj_list) {
316 last = lmark;
317
318 if ((lmark->group == mark->group) && !allow_dups)
319 return -EEXIST;
320
321 cmp = fsnotify_compare_groups(lmark->group, mark->group);
322 if (cmp >= 0) {
323 hlist_add_before_rcu(&mark->obj_list, &lmark->obj_list);
324 return 0;
325 }
326 }
327
328 BUG_ON(last == NULL);
329 /* mark should be the last entry. last is the current last entry */
330 hlist_add_behind_rcu(&mark->obj_list, &last->obj_list);
331 return 0;
332}
333
334/*
335 * Attach an initialized mark to a given group and fs object.
336 * These marks may be used for the fsnotify backend to determine which
337 * event types should be delivered to which group.
338 */
339int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
340 struct fsnotify_group *group, struct inode *inode,
341 struct vfsmount *mnt, int allow_dups)
342{
343 int ret = 0;
344
345 BUG_ON(inode && mnt);
346 BUG_ON(!inode && !mnt);
347 BUG_ON(!mutex_is_locked(&group->mark_mutex));
348
349 /*
350 * LOCKING ORDER!!!!
351 * group->mark_mutex
352 * mark->lock
353 * inode->i_lock
354 */
355 spin_lock(&mark->lock);
356 mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE | FSNOTIFY_MARK_FLAG_ATTACHED;
357
358 fsnotify_get_group(group);
359 mark->group = group;
360 list_add(&mark->g_list, &group->marks_list);
361 atomic_inc(&group->num_marks);
362 fsnotify_get_mark(mark); /* for i_list and g_list */
363
364 if (inode) {
365 ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
366 if (ret)
367 goto err;
368 } else if (mnt) {
369 ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
370 if (ret)
371 goto err;
372 } else {
373 BUG();
374 }
375
376 /* this will pin the object if appropriate */
377 fsnotify_set_mark_mask_locked(mark, mark->mask);
378 spin_unlock(&mark->lock);
379
380 if (inode)
381 __fsnotify_update_child_dentry_flags(inode);
382
383 return ret;
384err:
385 mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
386 list_del_init(&mark->g_list);
387 fsnotify_put_group(group);
388 mark->group = NULL;
389 atomic_dec(&group->num_marks);
390
391 spin_unlock(&mark->lock);
392
393 spin_lock(&destroy_lock);
394 list_add(&mark->g_list, &destroy_list);
395 spin_unlock(&destroy_lock);
396 queue_delayed_work(system_unbound_wq, &reaper_work,
397 FSNOTIFY_REAPER_DELAY);
398
399 return ret;
400}
401
402int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
403 struct inode *inode, struct vfsmount *mnt, int allow_dups)
404{
405 int ret;
406 mutex_lock(&group->mark_mutex);
407 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
408 mutex_unlock(&group->mark_mutex);
409 return ret;
410}
411
412/*
413 * Given a list of marks, find the mark associated with given group. If found
414 * take a reference to that mark and return it, else return NULL.
415 */
416struct fsnotify_mark *fsnotify_find_mark(struct hlist_head *head,
417 struct fsnotify_group *group)
418{
419 struct fsnotify_mark *mark;
420
421 hlist_for_each_entry(mark, head, obj_list) {
422 if (mark->group == group) {
423 fsnotify_get_mark(mark);
424 return mark;
425 }
426 }
427 return NULL;
428}
429
430/*
431 * clear any marks in a group in which mark->flags & flags is true
432 */
433void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
434 unsigned int flags)
435{
436 struct fsnotify_mark *lmark, *mark;
437 LIST_HEAD(to_free);
438
439 /*
440 * We have to be really careful here. Anytime we drop mark_mutex, e.g.
441 * fsnotify_clear_marks_by_inode() can come and free marks. Even in our
442 * to_free list so we have to use mark_mutex even when accessing that
443 * list. And freeing mark requires us to drop mark_mutex. So we can
444 * reliably free only the first mark in the list. That's why we first
445 * move marks to free to to_free list in one go and then free marks in
446 * to_free list one by one.
447 */
448 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
449 list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
450 if (mark->flags & flags)
451 list_move(&mark->g_list, &to_free);
452 }
453 mutex_unlock(&group->mark_mutex);
454
455 while (1) {
456 mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
457 if (list_empty(&to_free)) {
458 mutex_unlock(&group->mark_mutex);
459 break;
460 }
461 mark = list_first_entry(&to_free, struct fsnotify_mark, g_list);
462 fsnotify_get_mark(mark);
463 fsnotify_detach_mark(mark);
464 mutex_unlock(&group->mark_mutex);
465 fsnotify_free_mark(mark);
466 fsnotify_put_mark(mark);
467 }
468}
469
470/*
471 * Given a group, destroy all of the marks associated with that group.
472 */
473void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
474{
475 fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
476}
477
478void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
479{
480 assert_spin_locked(&old->lock);
481 new->inode = old->inode;
482 new->mnt = old->mnt;
483 if (old->group)
484 fsnotify_get_group(old->group);
485 new->group = old->group;
486 new->mask = old->mask;
487 new->free_mark = old->free_mark;
488}
489
490/*
491 * Nothing fancy, just initialize lists and locks and counters.
492 */
493void fsnotify_init_mark(struct fsnotify_mark *mark,
494 void (*free_mark)(struct fsnotify_mark *mark))
495{
496 memset(mark, 0, sizeof(*mark));
497 spin_lock_init(&mark->lock);
498 atomic_set(&mark->refcnt, 1);
499 mark->free_mark = free_mark;
500}
501
502static void fsnotify_mark_destroy(struct work_struct *work)
503{
504 struct fsnotify_mark *mark, *next;
505 struct list_head private_destroy_list;
506
507 spin_lock(&destroy_lock);
508 /* exchange the list head */
509 list_replace_init(&destroy_list, &private_destroy_list);
510 spin_unlock(&destroy_lock);
511
512 synchronize_srcu(&fsnotify_mark_srcu);
513
514 list_for_each_entry_safe(mark, next, &private_destroy_list, g_list) {
515 list_del_init(&mark->g_list);
516 fsnotify_put_mark(mark);
517 }
518}