fsnotify: implement ordering between notifiers
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / notify / inode_mark.c
blob4c29fcf557d13d21100030c31ab98aa44b555a62
1 /*
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)
7 * any later version.
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.
19 #include <linux/fs.h>
20 #include <linux/init.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/mutex.h>
24 #include <linux/spinlock.h>
25 #include <linux/writeback.h> /* for inode_lock */
27 #include <asm/atomic.h>
29 #include <linux/fsnotify_backend.h>
30 #include "fsnotify.h"
33 * Recalculate the mask of events relevant to a given inode locked.
35 static void fsnotify_recalc_inode_mask_locked(struct inode *inode)
37 struct fsnotify_mark *mark;
38 struct hlist_node *pos;
39 __u32 new_mask = 0;
41 assert_spin_locked(&inode->i_lock);
43 hlist_for_each_entry(mark, pos, &inode->i_fsnotify_marks, i.i_list)
44 new_mask |= mark->mask;
45 inode->i_fsnotify_mask = new_mask;
49 * Recalculate the inode->i_fsnotify_mask, or the mask of all FS_* event types
50 * any notifier is interested in hearing for this inode.
52 void fsnotify_recalc_inode_mask(struct inode *inode)
54 spin_lock(&inode->i_lock);
55 fsnotify_recalc_inode_mask_locked(inode);
56 spin_unlock(&inode->i_lock);
58 __fsnotify_update_child_dentry_flags(inode);
61 void fsnotify_destroy_inode_mark(struct fsnotify_mark *mark)
63 struct inode *inode = mark->i.inode;
65 assert_spin_locked(&mark->lock);
66 assert_spin_locked(&mark->group->mark_lock);
68 spin_lock(&inode->i_lock);
70 hlist_del_init_rcu(&mark->i.i_list);
71 mark->i.inode = NULL;
74 * this mark is now off the inode->i_fsnotify_marks list and we
75 * hold the inode->i_lock, so this is the perfect time to update the
76 * inode->i_fsnotify_mask
78 fsnotify_recalc_inode_mask_locked(inode);
80 spin_unlock(&inode->i_lock);
84 * Given an inode, destroy all of the marks associated with that inode.
86 void fsnotify_clear_marks_by_inode(struct inode *inode)
88 struct fsnotify_mark *mark, *lmark;
89 struct hlist_node *pos, *n;
90 LIST_HEAD(free_list);
92 spin_lock(&inode->i_lock);
93 hlist_for_each_entry_safe(mark, pos, n, &inode->i_fsnotify_marks, i.i_list) {
94 list_add(&mark->i.free_i_list, &free_list);
95 hlist_del_init_rcu(&mark->i.i_list);
96 fsnotify_get_mark(mark);
98 spin_unlock(&inode->i_lock);
100 list_for_each_entry_safe(mark, lmark, &free_list, i.free_i_list) {
101 fsnotify_destroy_mark(mark);
102 fsnotify_put_mark(mark);
107 * Given a group clear all of the inode marks associated with that group.
109 void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
111 fsnotify_clear_marks_by_group_flags(group, FSNOTIFY_MARK_FLAG_INODE);
115 * given a group and inode, find the mark associated with that combination.
116 * if found take a reference to that mark and return it, else return NULL
118 struct fsnotify_mark *fsnotify_find_inode_mark_locked(struct fsnotify_group *group,
119 struct inode *inode)
121 struct fsnotify_mark *mark;
122 struct hlist_node *pos;
124 assert_spin_locked(&inode->i_lock);
126 hlist_for_each_entry(mark, pos, &inode->i_fsnotify_marks, i.i_list) {
127 if (mark->group == group) {
128 fsnotify_get_mark(mark);
129 return mark;
132 return NULL;
136 * given a group and inode, find the mark associated with that combination.
137 * if found take a reference to that mark and return it, else return NULL
139 struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group,
140 struct inode *inode)
142 struct fsnotify_mark *mark;
144 spin_lock(&inode->i_lock);
145 mark = fsnotify_find_inode_mark_locked(group, inode);
146 spin_unlock(&inode->i_lock);
148 return mark;
152 * If we are setting a mark mask on an inode mark we should pin the inode
153 * in memory.
155 void fsnotify_set_inode_mark_mask_locked(struct fsnotify_mark *mark,
156 __u32 mask)
158 struct inode *inode;
160 assert_spin_locked(&mark->lock);
162 if (mask &&
163 mark->i.inode &&
164 !(mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED)) {
165 mark->flags |= FSNOTIFY_MARK_FLAG_OBJECT_PINNED;
166 inode = igrab(mark->i.inode);
168 * we shouldn't be able to get here if the inode wasn't
169 * already safely held in memory. But bug in case it
170 * ever is wrong.
172 BUG_ON(!inode);
177 * Attach an initialized mark to a given inode.
178 * These marks may be used for the fsnotify backend to determine which
179 * event types should be delivered to which group and for which inodes. These
180 * marks are ordered according to priority, highest number first, and then by
181 * the group's location in memory.
183 int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
184 struct fsnotify_group *group, struct inode *inode,
185 int allow_dups)
187 struct fsnotify_mark *lmark;
188 struct hlist_node *node, *last = NULL;
189 int ret = 0;
191 mark->flags |= FSNOTIFY_MARK_FLAG_INODE;
193 assert_spin_locked(&mark->lock);
194 assert_spin_locked(&group->mark_lock);
196 spin_lock(&inode->i_lock);
198 mark->i.inode = inode;
200 /* is mark the first mark? */
201 if (hlist_empty(&inode->i_fsnotify_marks)) {
202 hlist_add_head_rcu(&mark->i.i_list, &inode->i_fsnotify_marks);
203 goto out;
206 /* should mark be in the middle of the current list? */
207 hlist_for_each_entry(lmark, node, &inode->i_fsnotify_marks, i.i_list) {
208 last = node;
210 if ((lmark->group == group) && !allow_dups) {
211 ret = -EEXIST;
212 goto out;
215 if (mark->group->priority < lmark->group->priority)
216 continue;
218 if ((mark->group->priority == lmark->group->priority) &&
219 (mark->group < lmark->group))
220 continue;
222 hlist_add_before_rcu(&mark->i.i_list, &lmark->i.i_list);
223 goto out;
226 BUG_ON(last == NULL);
227 /* mark should be the last entry. last is the current last entry */
228 hlist_add_after_rcu(last, &mark->i.i_list);
229 out:
230 fsnotify_recalc_inode_mask_locked(inode);
231 spin_unlock(&inode->i_lock);
233 return ret;
237 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
238 * @list: list of inodes being unmounted (sb->s_inodes)
240 * Called with inode_lock held, protecting the unmounting super block's list
241 * of inodes, and with iprune_mutex held, keeping shrink_icache_memory() at bay.
242 * We temporarily drop inode_lock, however, and CAN block.
244 void fsnotify_unmount_inodes(struct list_head *list)
246 struct inode *inode, *next_i, *need_iput = NULL;
248 spin_lock(&inode_lock);
249 list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
250 struct inode *need_iput_tmp;
253 * We cannot __iget() an inode in state I_FREEING,
254 * I_WILL_FREE, or I_NEW which is fine because by that point
255 * the inode cannot have any associated watches.
257 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
258 continue;
261 * If i_count is zero, the inode cannot have any watches and
262 * doing an __iget/iput with MS_ACTIVE clear would actually
263 * evict all inodes with zero i_count from icache which is
264 * unnecessarily violent and may in fact be illegal to do.
266 if (!atomic_read(&inode->i_count))
267 continue;
269 need_iput_tmp = need_iput;
270 need_iput = NULL;
272 /* In case fsnotify_inode_delete() drops a reference. */
273 if (inode != need_iput_tmp)
274 __iget(inode);
275 else
276 need_iput_tmp = NULL;
278 /* In case the dropping of a reference would nuke next_i. */
279 if ((&next_i->i_sb_list != list) &&
280 atomic_read(&next_i->i_count) &&
281 !(next_i->i_state & (I_FREEING | I_WILL_FREE))) {
282 __iget(next_i);
283 need_iput = next_i;
287 * We can safely drop inode_lock here because we hold
288 * references on both inode and next_i. Also no new inodes
289 * will be added since the umount has begun. Finally,
290 * iprune_mutex keeps shrink_icache_memory() away.
292 spin_unlock(&inode_lock);
294 if (need_iput_tmp)
295 iput(need_iput_tmp);
297 /* for each watch, send FS_UNMOUNT and then remove it */
298 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
300 fsnotify_inode_delete(inode);
302 iput(inode);
304 spin_lock(&inode_lock);
306 spin_unlock(&inode_lock);