]> git.ipfire.org Git - people/ms/linux.git/blame - fs/notify/fsnotify.c
fsnotify: Provide framework for dropping SRCU lock in ->handle_event
[people/ms/linux.git] / fs / notify / fsnotify.c
CommitLineData
90586523
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#include <linux/dcache.h>
20#include <linux/fs.h>
5a0e3ad6 21#include <linux/gfp.h>
90586523
EP
22#include <linux/init.h>
23#include <linux/module.h>
7131485a 24#include <linux/mount.h>
90586523
EP
25#include <linux/srcu.h>
26
27#include <linux/fsnotify_backend.h>
28#include "fsnotify.h"
29
3be25f49
EP
30/*
31 * Clear all of the marks on an inode when it is being evicted from core
32 */
33void __fsnotify_inode_delete(struct inode *inode)
34{
35 fsnotify_clear_marks_by_inode(inode);
36}
37EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
38
ca9c726e
AG
39void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
40{
41 fsnotify_clear_marks_by_mount(mnt);
42}
43
c28f7e56
EP
44/*
45 * Given an inode, first check if we care what happens to our children. Inotify
46 * and dnotify both tell their parents about events. If we care about any event
47 * on a child we run all of our children and set a dentry flag saying that the
48 * parent cares. Thus when an event happens on a child it can quickly tell if
49 * if there is a need to find a parent and send the event to the parent.
50 */
51void __fsnotify_update_child_dentry_flags(struct inode *inode)
52{
53 struct dentry *alias;
54 int watched;
55
56 if (!S_ISDIR(inode->i_mode))
57 return;
58
59 /* determine if the children should tell inode about their events */
60 watched = fsnotify_inode_watches_children(inode);
61
873feea0 62 spin_lock(&inode->i_lock);
c28f7e56
EP
63 /* run all of the dentries associated with this inode. Since this is a
64 * directory, there damn well better only be one item on this list */
946e51f2 65 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
c28f7e56
EP
66 struct dentry *child;
67
68 /* run all of the children of the original inode and fix their
69 * d_flags to indicate parental interest (their parent is the
70 * original inode) */
2fd6b7f5 71 spin_lock(&alias->d_lock);
946e51f2 72 list_for_each_entry(child, &alias->d_subdirs, d_child) {
c28f7e56
EP
73 if (!child->d_inode)
74 continue;
75
2fd6b7f5 76 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
c28f7e56
EP
77 if (watched)
78 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
79 else
80 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
81 spin_unlock(&child->d_lock);
82 }
2fd6b7f5 83 spin_unlock(&alias->d_lock);
c28f7e56 84 }
873feea0 85 spin_unlock(&inode->i_lock);
c28f7e56
EP
86}
87
88/* Notify this dentry's parent about a child's events. */
12c7f9dc 89int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
c28f7e56
EP
90{
91 struct dentry *parent;
92 struct inode *p_inode;
52420392 93 int ret = 0;
c28f7e56 94
72acc854 95 if (!dentry)
2069601b 96 dentry = path->dentry;
28c60e37 97
c28f7e56 98 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
52420392 99 return 0;
c28f7e56 100
4d4eb366 101 parent = dget_parent(dentry);
c28f7e56
EP
102 p_inode = parent->d_inode;
103
4d4eb366
CH
104 if (unlikely(!fsnotify_inode_watches_children(p_inode)))
105 __fsnotify_update_child_dentry_flags(p_inode);
106 else if (p_inode->i_fsnotify_mask & mask) {
c28f7e56
EP
107 /* we are notifying a parent so come up with the new mask which
108 * specifies these are events which came from a child. */
109 mask |= FS_EVENT_ON_CHILD;
110
2069601b 111 if (path)
52420392
EP
112 ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
113 dentry->d_name.name, 0);
28c60e37 114 else
52420392
EP
115 ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
116 dentry->d_name.name, 0);
c28f7e56
EP
117 }
118
4d4eb366 119 dput(parent);
52420392
EP
120
121 return ret;
c28f7e56
EP
122}
123EXPORT_SYMBOL_GPL(__fsnotify_parent);
124
fd657170 125static int send_to_group(struct inode *to_tell,
ce8f76fb
EP
126 struct fsnotify_mark *inode_mark,
127 struct fsnotify_mark *vfsmount_mark,
e637835e 128 __u32 mask, const void *data,
613a807f 129 int data_is, u32 cookie,
7053aee2 130 const unsigned char *file_name)
7131485a 131{
faa9560a 132 struct fsnotify_group *group = NULL;
84e1ab4d
EP
133 __u32 inode_test_mask = 0;
134 __u32 vfsmount_test_mask = 0;
613a807f 135
faa9560a
EP
136 if (unlikely(!inode_mark && !vfsmount_mark)) {
137 BUG();
138 return 0;
139 }
ce8f76fb
EP
140
141 /* clear ignored on inode modification */
142 if (mask & FS_MODIFY) {
143 if (inode_mark &&
144 !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
145 inode_mark->ignored_mask = 0;
146 if (vfsmount_mark &&
147 !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
148 vfsmount_mark->ignored_mask = 0;
149 }
5ba08e2e 150
ce8f76fb
EP
151 /* does the inode mark tell us to do something? */
152 if (inode_mark) {
faa9560a 153 group = inode_mark->group;
84e1ab4d 154 inode_test_mask = (mask & ~FS_EVENT_ON_CHILD);
ce8f76fb
EP
155 inode_test_mask &= inode_mark->mask;
156 inode_test_mask &= ~inode_mark->ignored_mask;
157 }
613a807f 158
ce8f76fb
EP
159 /* does the vfsmount_mark tell us to do something? */
160 if (vfsmount_mark) {
84e1ab4d 161 vfsmount_test_mask = (mask & ~FS_EVENT_ON_CHILD);
faa9560a 162 group = vfsmount_mark->group;
ce8f76fb
EP
163 vfsmount_test_mask &= vfsmount_mark->mask;
164 vfsmount_test_mask &= ~vfsmount_mark->ignored_mask;
165 if (inode_mark)
166 vfsmount_test_mask &= ~inode_mark->ignored_mask;
167 }
168
fd657170 169 pr_debug("%s: group=%p to_tell=%p mask=%x inode_mark=%p"
84e1ab4d 170 " inode_test_mask=%x vfsmount_mark=%p vfsmount_test_mask=%x"
7053aee2 171 " data=%p data_is=%d cookie=%d\n",
fd657170 172 __func__, group, to_tell, mask, inode_mark,
84e1ab4d 173 inode_test_mask, vfsmount_mark, vfsmount_test_mask, data,
7053aee2 174 data_is, cookie);
faa9560a 175
ce8f76fb 176 if (!inode_test_mask && !vfsmount_test_mask)
613a807f
EP
177 return 0;
178
7053aee2
JK
179 return group->ops->handle_event(group, to_tell, inode_mark,
180 vfsmount_mark, mask, data, data_is,
45a22f4c 181 file_name, cookie);
7131485a
EP
182}
183
90586523
EP
184/*
185 * This is the main call to fsnotify. The VFS calls into hook specific functions
186 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
187 * out to all of the registered fsnotify_group. Those groups can then use the
188 * notification event in whatever means they feel necessary.
189 */
e637835e 190int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
c4ec54b4 191 const unsigned char *file_name, u32 cookie)
90586523 192{
84e1ab4d 193 struct hlist_node *inode_node = NULL, *vfsmount_node = NULL;
613a807f
EP
194 struct fsnotify_mark *inode_mark = NULL, *vfsmount_mark = NULL;
195 struct fsnotify_group *inode_group, *vfsmount_group;
9dd813c1 196 struct fsnotify_mark_connector *inode_conn, *vfsmount_conn;
c63181e6 197 struct mount *mnt;
c4ec54b4 198 int idx, ret = 0;
e42e2773
EP
199 /* global tests shouldn't care about events on child only the specific event */
200 __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
90586523 201
2069601b 202 if (data_is == FSNOTIFY_EVENT_PATH)
e637835e 203 mnt = real_mount(((const struct path *)data)->mnt);
613a807f
EP
204 else
205 mnt = NULL;
206
7c49b861
DH
207 /*
208 * Optimization: srcu_read_lock() has a memory barrier which can
209 * be expensive. It protects walking the *_fsnotify_marks lists.
210 * However, if we do not walk the lists, we do not have to do
211 * SRCU because we have no references to any objects and do not
212 * need SRCU to keep them "alive".
213 */
9dd813c1
JK
214 if (!to_tell->i_fsnotify_marks &&
215 (!mnt || !mnt->mnt_fsnotify_marks))
7c49b861 216 return 0;
613a807f
EP
217 /*
218 * if this is a modify event we may need to clear the ignored masks
219 * otherwise return if neither the inode nor the vfsmount care about
220 * this type of event.
221 */
222 if (!(mask & FS_MODIFY) &&
223 !(test_mask & to_tell->i_fsnotify_mask) &&
224 !(mnt && test_mask & mnt->mnt_fsnotify_mask))
225 return 0;
3a9fb89f 226
75c1be48 227 idx = srcu_read_lock(&fsnotify_mark_srcu);
7131485a 228
613a807f 229 if ((mask & FS_MODIFY) ||
9dd813c1 230 (test_mask & to_tell->i_fsnotify_mask)) {
08991e83
JK
231 inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
232 &fsnotify_mark_srcu);
9dd813c1
JK
233 if (inode_conn)
234 inode_node = srcu_dereference(inode_conn->list.first,
235 &fsnotify_mark_srcu);
236 }
613a807f 237
84e1ab4d
EP
238 if (mnt && ((mask & FS_MODIFY) ||
239 (test_mask & mnt->mnt_fsnotify_mask))) {
08991e83
JK
240 inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
241 &fsnotify_mark_srcu);
9dd813c1
JK
242 if (inode_conn)
243 inode_node = srcu_dereference(inode_conn->list.first,
244 &fsnotify_mark_srcu);
08991e83
JK
245 vfsmount_conn = srcu_dereference(mnt->mnt_fsnotify_marks,
246 &fsnotify_mark_srcu);
9dd813c1
JK
247 if (vfsmount_conn)
248 vfsmount_node = srcu_dereference(
249 vfsmount_conn->list.first,
250 &fsnotify_mark_srcu);
90586523 251 }
75c1be48 252
8edc6e16
JK
253 /*
254 * We need to merge inode & vfsmount mark lists so that inode mark
255 * ignore masks are properly reflected for mount mark notifications.
256 * That's why this traversal is so complicated...
257 */
613a807f 258 while (inode_node || vfsmount_node) {
8edc6e16
JK
259 inode_group = NULL;
260 inode_mark = NULL;
261 vfsmount_group = NULL;
262 vfsmount_mark = NULL;
5f3f259f 263
613a807f
EP
264 if (inode_node) {
265 inode_mark = hlist_entry(srcu_dereference(inode_node, &fsnotify_mark_srcu),
0809ab69 266 struct fsnotify_mark, obj_list);
613a807f 267 inode_group = inode_mark->group;
f72adfd5 268 }
613a807f
EP
269
270 if (vfsmount_node) {
271 vfsmount_mark = hlist_entry(srcu_dereference(vfsmount_node, &fsnotify_mark_srcu),
0809ab69 272 struct fsnotify_mark, obj_list);
613a807f 273 vfsmount_group = vfsmount_mark->group;
f72adfd5 274 }
613a807f 275
8edc6e16
JK
276 if (inode_group && vfsmount_group) {
277 int cmp = fsnotify_compare_groups(inode_group,
278 vfsmount_group);
279 if (cmp > 0) {
280 inode_group = NULL;
281 inode_mark = NULL;
282 } else if (cmp < 0) {
283 vfsmount_group = NULL;
284 vfsmount_mark = NULL;
285 }
7131485a 286 }
8edc6e16
JK
287 ret = send_to_group(to_tell, inode_mark, vfsmount_mark, mask,
288 data, data_is, cookie, file_name);
613a807f 289
ff8bcbd0
EP
290 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
291 goto out;
292
92b4678e 293 if (inode_group)
ce8f76fb
EP
294 inode_node = srcu_dereference(inode_node->next,
295 &fsnotify_mark_srcu);
92b4678e 296 if (vfsmount_group)
ce8f76fb
EP
297 vfsmount_node = srcu_dereference(vfsmount_node->next,
298 &fsnotify_mark_srcu);
7131485a 299 }
ff8bcbd0
EP
300 ret = 0;
301out:
75c1be48 302 srcu_read_unlock(&fsnotify_mark_srcu, idx);
c4ec54b4 303
98b5c10d 304 return ret;
90586523
EP
305}
306EXPORT_SYMBOL_GPL(fsnotify);
307
9dd813c1
JK
308extern struct kmem_cache *fsnotify_mark_connector_cachep;
309
90586523
EP
310static __init int fsnotify_init(void)
311{
75c1be48
EP
312 int ret;
313
20dee624
EP
314 BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23);
315
75c1be48
EP
316 ret = init_srcu_struct(&fsnotify_mark_srcu);
317 if (ret)
318 panic("initializing fsnotify_mark_srcu");
319
9dd813c1
JK
320 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
321 SLAB_PANIC);
322
75c1be48 323 return 0;
90586523 324}
75c1be48 325core_initcall(fsnotify_init);