xref: /openbmc/linux/fs/notify/mark.c (revision 6774def6)
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 struct srcu_struct fsnotify_mark_srcu;
95 static DEFINE_SPINLOCK(destroy_lock);
96 static LIST_HEAD(destroy_list);
97 static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);
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 		if (mark->group)
108 			fsnotify_put_group(mark->group);
109 		mark->free_mark(mark);
110 	}
111 }
112 
113 /*
114  * Any time a mark is getting freed we end up here.
115  * The caller had better be holding a reference to this mark so we don't actually
116  * do the final put under the mark->lock
117  */
118 void fsnotify_destroy_mark_locked(struct fsnotify_mark *mark,
119 				  struct fsnotify_group *group)
120 {
121 	struct inode *inode = NULL;
122 
123 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
124 
125 	spin_lock(&mark->lock);
126 
127 	/* something else already called this function on this mark */
128 	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
129 		spin_unlock(&mark->lock);
130 		return;
131 	}
132 
133 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
134 
135 	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
136 		inode = mark->i.inode;
137 		fsnotify_destroy_inode_mark(mark);
138 	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
139 		fsnotify_destroy_vfsmount_mark(mark);
140 	else
141 		BUG();
142 
143 	list_del_init(&mark->g_list);
144 
145 	spin_unlock(&mark->lock);
146 
147 	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
148 		iput(inode);
149 	/* release lock temporarily */
150 	mutex_unlock(&group->mark_mutex);
151 
152 	spin_lock(&destroy_lock);
153 	list_add(&mark->destroy_list, &destroy_list);
154 	spin_unlock(&destroy_lock);
155 	wake_up(&destroy_waitq);
156 	/*
157 	 * We don't necessarily have a ref on mark from caller so the above destroy
158 	 * may have actually freed it, unless this group provides a 'freeing_mark'
159 	 * function which must be holding a reference.
160 	 */
161 
162 	/*
163 	 * Some groups like to know that marks are being freed.  This is a
164 	 * callback to the group function to let it know that this mark
165 	 * is being freed.
166 	 */
167 	if (group->ops->freeing_mark)
168 		group->ops->freeing_mark(mark, group);
169 
170 	/*
171 	 * __fsnotify_update_child_dentry_flags(inode);
172 	 *
173 	 * I really want to call that, but we can't, we have no idea if the inode
174 	 * still exists the second we drop the mark->lock.
175 	 *
176 	 * The next time an event arrive to this inode from one of it's children
177 	 * __fsnotify_parent will see that the inode doesn't care about it's
178 	 * children and will update all of these flags then.  So really this
179 	 * is just a lazy update (and could be a perf win...)
180 	 */
181 
182 	atomic_dec(&group->num_marks);
183 
184 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
185 }
186 
187 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
188 			   struct fsnotify_group *group)
189 {
190 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
191 	fsnotify_destroy_mark_locked(mark, group);
192 	mutex_unlock(&group->mark_mutex);
193 }
194 
195 void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
196 {
197 	assert_spin_locked(&mark->lock);
198 
199 	mark->mask = mask;
200 
201 	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
202 		fsnotify_set_inode_mark_mask_locked(mark, mask);
203 }
204 
205 void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
206 {
207 	assert_spin_locked(&mark->lock);
208 
209 	mark->ignored_mask = mask;
210 }
211 
212 /*
213  * Sorting function for lists of fsnotify marks.
214  *
215  * Fanotify supports different notification classes (reflected as priority of
216  * notification group). Events shall be passed to notification groups in
217  * decreasing priority order. To achieve this marks in notification lists for
218  * inodes and vfsmounts are sorted so that priorities of corresponding groups
219  * are descending.
220  *
221  * Furthermore correct handling of the ignore mask requires processing inode
222  * and vfsmount marks of each group together. Using the group address as
223  * further sort criterion provides a unique sorting order and thus we can
224  * merge inode and vfsmount lists of marks in linear time and find groups
225  * present in both lists.
226  *
227  * A return value of 1 signifies that b has priority over a.
228  * A return value of 0 signifies that the two marks have to be handled together.
229  * A return value of -1 signifies that a has priority over b.
230  */
231 int fsnotify_compare_groups(struct fsnotify_group *a, struct fsnotify_group *b)
232 {
233 	if (a == b)
234 		return 0;
235 	if (!a)
236 		return 1;
237 	if (!b)
238 		return -1;
239 	if (a->priority < b->priority)
240 		return 1;
241 	if (a->priority > b->priority)
242 		return -1;
243 	if (a < b)
244 		return 1;
245 	return -1;
246 }
247 
248 /*
249  * Attach an initialized mark to a given group and fs object.
250  * These marks may be used for the fsnotify backend to determine which
251  * event types should be delivered to which group.
252  */
253 int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
254 			     struct fsnotify_group *group, struct inode *inode,
255 			     struct vfsmount *mnt, int allow_dups)
256 {
257 	int ret = 0;
258 
259 	BUG_ON(inode && mnt);
260 	BUG_ON(!inode && !mnt);
261 	BUG_ON(!mutex_is_locked(&group->mark_mutex));
262 
263 	/*
264 	 * LOCKING ORDER!!!!
265 	 * group->mark_mutex
266 	 * mark->lock
267 	 * inode->i_lock
268 	 */
269 	spin_lock(&mark->lock);
270 	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;
271 
272 	fsnotify_get_group(group);
273 	mark->group = group;
274 	list_add(&mark->g_list, &group->marks_list);
275 	atomic_inc(&group->num_marks);
276 	fsnotify_get_mark(mark); /* for i_list and g_list */
277 
278 	if (inode) {
279 		ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
280 		if (ret)
281 			goto err;
282 	} else if (mnt) {
283 		ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
284 		if (ret)
285 			goto err;
286 	} else {
287 		BUG();
288 	}
289 
290 	/* this will pin the object if appropriate */
291 	fsnotify_set_mark_mask_locked(mark, mark->mask);
292 	spin_unlock(&mark->lock);
293 
294 	if (inode)
295 		__fsnotify_update_child_dentry_flags(inode);
296 
297 	return ret;
298 err:
299 	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
300 	list_del_init(&mark->g_list);
301 	fsnotify_put_group(group);
302 	mark->group = NULL;
303 	atomic_dec(&group->num_marks);
304 
305 	spin_unlock(&mark->lock);
306 
307 	spin_lock(&destroy_lock);
308 	list_add(&mark->destroy_list, &destroy_list);
309 	spin_unlock(&destroy_lock);
310 	wake_up(&destroy_waitq);
311 
312 	return ret;
313 }
314 
315 int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
316 		      struct inode *inode, struct vfsmount *mnt, int allow_dups)
317 {
318 	int ret;
319 	mutex_lock(&group->mark_mutex);
320 	ret = fsnotify_add_mark_locked(mark, group, inode, mnt, allow_dups);
321 	mutex_unlock(&group->mark_mutex);
322 	return ret;
323 }
324 
325 /*
326  * clear any marks in a group in which mark->flags & flags is true
327  */
328 void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
329 					 unsigned int flags)
330 {
331 	struct fsnotify_mark *lmark, *mark;
332 
333 	mutex_lock_nested(&group->mark_mutex, SINGLE_DEPTH_NESTING);
334 	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
335 		if (mark->flags & flags) {
336 			fsnotify_get_mark(mark);
337 			fsnotify_destroy_mark_locked(mark, group);
338 			fsnotify_put_mark(mark);
339 		}
340 	}
341 	mutex_unlock(&group->mark_mutex);
342 }
343 
344 /*
345  * Given a group, destroy all of the marks associated with that group.
346  */
347 void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
348 {
349 	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
350 }
351 
352 void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
353 {
354 	assert_spin_locked(&old->lock);
355 	new->i.inode = old->i.inode;
356 	new->m.mnt = old->m.mnt;
357 	if (old->group)
358 		fsnotify_get_group(old->group);
359 	new->group = old->group;
360 	new->mask = old->mask;
361 	new->free_mark = old->free_mark;
362 }
363 
364 /*
365  * Nothing fancy, just initialize lists and locks and counters.
366  */
367 void fsnotify_init_mark(struct fsnotify_mark *mark,
368 			void (*free_mark)(struct fsnotify_mark *mark))
369 {
370 	memset(mark, 0, sizeof(*mark));
371 	spin_lock_init(&mark->lock);
372 	atomic_set(&mark->refcnt, 1);
373 	mark->free_mark = free_mark;
374 }
375 
376 static int fsnotify_mark_destroy(void *ignored)
377 {
378 	struct fsnotify_mark *mark, *next;
379 	struct list_head private_destroy_list;
380 
381 	for (;;) {
382 		spin_lock(&destroy_lock);
383 		/* exchange the list head */
384 		list_replace_init(&destroy_list, &private_destroy_list);
385 		spin_unlock(&destroy_lock);
386 
387 		synchronize_srcu(&fsnotify_mark_srcu);
388 
389 		list_for_each_entry_safe(mark, next, &private_destroy_list, destroy_list) {
390 			list_del_init(&mark->destroy_list);
391 			fsnotify_put_mark(mark);
392 		}
393 
394 		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
395 	}
396 
397 	return 0;
398 }
399 
400 static int __init fsnotify_mark_init(void)
401 {
402 	struct task_struct *thread;
403 
404 	thread = kthread_run(fsnotify_mark_destroy, NULL,
405 			     "fsnotify_mark");
406 	if (IS_ERR(thread))
407 		panic("unable to start fsnotify mark destruction thread.");
408 
409 	return 0;
410 }
411 device_initcall(fsnotify_mark_init);
412