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#include <linux/dcache.h>
20#include <linux/fs.h>
21#include <linux/gfp.h>
22#include <linux/init.h>
23#include <linux/module.h>
24#include <linux/mount.h>
25#include <linux/srcu.h>
26
27#include <linux/fsnotify_backend.h>
28#include "fsnotify.h"
29
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
39void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
40{
41 fsnotify_clear_marks_by_mount(mnt);
42}
43
44/**
45 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
46 * @sb: superblock being unmounted.
47 *
48 * Called during unmount with no locks held, so needs to be safe against
49 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
50 */
51void fsnotify_unmount_inodes(struct super_block *sb)
52{
53 struct inode *inode, *iput_inode = NULL;
54
55 spin_lock(&sb->s_inode_list_lock);
56 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
57 /*
58 * We cannot __iget() an inode in state I_FREEING,
59 * I_WILL_FREE, or I_NEW which is fine because by that point
60 * the inode cannot have any associated watches.
61 */
62 spin_lock(&inode->i_lock);
63 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
64 spin_unlock(&inode->i_lock);
65 continue;
66 }
67
68 /*
69 * If i_count is zero, the inode cannot have any watches and
70 * doing an __iget/iput with SB_ACTIVE clear would actually
71 * evict all inodes with zero i_count from icache which is
72 * unnecessarily violent and may in fact be illegal to do.
73 */
74 if (!atomic_read(&inode->i_count)) {
75 spin_unlock(&inode->i_lock);
76 continue;
77 }
78
79 __iget(inode);
80 spin_unlock(&inode->i_lock);
81 spin_unlock(&sb->s_inode_list_lock);
82
83 if (iput_inode)
84 iput(iput_inode);
85
86 /* for each watch, send FS_UNMOUNT and then remove it */
87 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
88
89 fsnotify_inode_delete(inode);
90
91 iput_inode = inode;
92
93 spin_lock(&sb->s_inode_list_lock);
94 }
95 spin_unlock(&sb->s_inode_list_lock);
96
97 if (iput_inode)
98 iput(iput_inode);
99}
100
101/*
102 * Given an inode, first check if we care what happens to our children. Inotify
103 * and dnotify both tell their parents about events. If we care about any event
104 * on a child we run all of our children and set a dentry flag saying that the
105 * parent cares. Thus when an event happens on a child it can quickly tell if
106 * if there is a need to find a parent and send the event to the parent.
107 */
108void __fsnotify_update_child_dentry_flags(struct inode *inode)
109{
110 struct dentry *alias;
111 int watched;
112
113 if (!S_ISDIR(inode->i_mode))
114 return;
115
116 /* determine if the children should tell inode about their events */
117 watched = fsnotify_inode_watches_children(inode);
118
119 spin_lock(&inode->i_lock);
120 /* run all of the dentries associated with this inode. Since this is a
121 * directory, there damn well better only be one item on this list */
122 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
123 struct dentry *child;
124
125 /* run all of the children of the original inode and fix their
126 * d_flags to indicate parental interest (their parent is the
127 * original inode) */
128 spin_lock(&alias->d_lock);
129 list_for_each_entry(child, &alias->d_subdirs, d_child) {
130 if (!child->d_inode)
131 continue;
132
133 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
134 if (watched)
135 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
136 else
137 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
138 spin_unlock(&child->d_lock);
139 }
140 spin_unlock(&alias->d_lock);
141 }
142 spin_unlock(&inode->i_lock);
143}
144
145/* Notify this dentry's parent about a child's events. */
146int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
147{
148 struct dentry *parent;
149 struct inode *p_inode;
150 int ret = 0;
151
152 if (!dentry)
153 dentry = path->dentry;
154
155 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
156 return 0;
157
158 parent = dget_parent(dentry);
159 p_inode = parent->d_inode;
160
161 if (unlikely(!fsnotify_inode_watches_children(p_inode)))
162 __fsnotify_update_child_dentry_flags(p_inode);
163 else if (p_inode->i_fsnotify_mask & mask) {
164 struct name_snapshot name;
165
166 /* we are notifying a parent so come up with the new mask which
167 * specifies these are events which came from a child. */
168 mask |= FS_EVENT_ON_CHILD;
169
170 take_dentry_name_snapshot(&name, dentry);
171 if (path)
172 ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
173 name.name, 0);
174 else
175 ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
176 name.name, 0);
177 release_dentry_name_snapshot(&name);
178 }
179
180 dput(parent);
181
182 return ret;
183}
184EXPORT_SYMBOL_GPL(__fsnotify_parent);
185
186static int send_to_group(struct inode *to_tell,
187 struct fsnotify_mark *inode_mark,
188 struct fsnotify_mark *vfsmount_mark,
189 __u32 mask, const void *data,
190 int data_is, u32 cookie,
191 const unsigned char *file_name,
192 struct fsnotify_iter_info *iter_info)
193{
194 struct fsnotify_group *group = NULL;
195 __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
196 __u32 marks_mask = 0;
197 __u32 marks_ignored_mask = 0;
198
199 if (unlikely(!inode_mark && !vfsmount_mark)) {
200 BUG();
201 return 0;
202 }
203
204 /* clear ignored on inode modification */
205 if (mask & FS_MODIFY) {
206 if (inode_mark &&
207 !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
208 inode_mark->ignored_mask = 0;
209 if (vfsmount_mark &&
210 !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
211 vfsmount_mark->ignored_mask = 0;
212 }
213
214 /* does the inode mark tell us to do something? */
215 if (inode_mark) {
216 group = inode_mark->group;
217 marks_mask |= inode_mark->mask;
218 marks_ignored_mask |= inode_mark->ignored_mask;
219 }
220
221 /* does the vfsmount_mark tell us to do something? */
222 if (vfsmount_mark) {
223 group = vfsmount_mark->group;
224 marks_mask |= vfsmount_mark->mask;
225 marks_ignored_mask |= vfsmount_mark->ignored_mask;
226 }
227
228 pr_debug("%s: group=%p to_tell=%p mask=%x inode_mark=%p"
229 " vfsmount_mark=%p marks_mask=%x marks_ignored_mask=%x"
230 " data=%p data_is=%d cookie=%d\n",
231 __func__, group, to_tell, mask, inode_mark, vfsmount_mark,
232 marks_mask, marks_ignored_mask, data,
233 data_is, cookie);
234
235 if (!(test_mask & marks_mask & ~marks_ignored_mask))
236 return 0;
237
238 return group->ops->handle_event(group, to_tell, inode_mark,
239 vfsmount_mark, mask, data, data_is,
240 file_name, cookie, iter_info);
241}
242
243static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
244{
245 struct fsnotify_mark_connector *conn;
246 struct hlist_node *node = NULL;
247
248 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
249 if (conn)
250 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
251
252 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
253}
254
255static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
256{
257 struct hlist_node *node = NULL;
258
259 if (mark)
260 node = srcu_dereference(mark->obj_list.next,
261 &fsnotify_mark_srcu);
262
263 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
264}
265
266/*
267 * This is the main call to fsnotify. The VFS calls into hook specific functions
268 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
269 * out to all of the registered fsnotify_group. Those groups can then use the
270 * notification event in whatever means they feel necessary.
271 */
272int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
273 const unsigned char *file_name, u32 cookie)
274{
275 struct fsnotify_iter_info iter_info = {};
276 struct mount *mnt;
277 int ret = 0;
278 /* global tests shouldn't care about events on child only the specific event */
279 __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
280
281 if (data_is == FSNOTIFY_EVENT_PATH)
282 mnt = real_mount(((const struct path *)data)->mnt);
283 else
284 mnt = NULL;
285
286 /*
287 * Optimization: srcu_read_lock() has a memory barrier which can
288 * be expensive. It protects walking the *_fsnotify_marks lists.
289 * However, if we do not walk the lists, we do not have to do
290 * SRCU because we have no references to any objects and do not
291 * need SRCU to keep them "alive".
292 */
293 if (!to_tell->i_fsnotify_marks &&
294 (!mnt || !mnt->mnt_fsnotify_marks))
295 return 0;
296 /*
297 * if this is a modify event we may need to clear the ignored masks
298 * otherwise return if neither the inode nor the vfsmount care about
299 * this type of event.
300 */
301 if (!(mask & FS_MODIFY) &&
302 !(test_mask & to_tell->i_fsnotify_mask) &&
303 !(mnt && test_mask & mnt->mnt_fsnotify_mask))
304 return 0;
305
306 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
307
308 if ((mask & FS_MODIFY) ||
309 (test_mask & to_tell->i_fsnotify_mask)) {
310 iter_info.inode_mark =
311 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
312 }
313
314 if (mnt && ((mask & FS_MODIFY) ||
315 (test_mask & mnt->mnt_fsnotify_mask))) {
316 iter_info.inode_mark =
317 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
318 iter_info.vfsmount_mark =
319 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
320 }
321
322 /*
323 * We need to merge inode & vfsmount mark lists so that inode mark
324 * ignore masks are properly reflected for mount mark notifications.
325 * That's why this traversal is so complicated...
326 */
327 while (iter_info.inode_mark || iter_info.vfsmount_mark) {
328 struct fsnotify_mark *inode_mark = iter_info.inode_mark;
329 struct fsnotify_mark *vfsmount_mark = iter_info.vfsmount_mark;
330
331 if (inode_mark && vfsmount_mark) {
332 int cmp = fsnotify_compare_groups(inode_mark->group,
333 vfsmount_mark->group);
334 if (cmp > 0)
335 inode_mark = NULL;
336 else if (cmp < 0)
337 vfsmount_mark = NULL;
338 }
339
340 ret = send_to_group(to_tell, inode_mark, vfsmount_mark, mask,
341 data, data_is, cookie, file_name,
342 &iter_info);
343
344 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
345 goto out;
346
347 if (inode_mark)
348 iter_info.inode_mark =
349 fsnotify_next_mark(iter_info.inode_mark);
350 if (vfsmount_mark)
351 iter_info.vfsmount_mark =
352 fsnotify_next_mark(iter_info.vfsmount_mark);
353 }
354 ret = 0;
355out:
356 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
357
358 return ret;
359}
360EXPORT_SYMBOL_GPL(fsnotify);
361
362extern struct kmem_cache *fsnotify_mark_connector_cachep;
363
364static __init int fsnotify_init(void)
365{
366 int ret;
367
368 BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23);
369
370 ret = init_srcu_struct(&fsnotify_mark_srcu);
371 if (ret)
372 panic("initializing fsnotify_mark_srcu");
373
374 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
375 SLAB_PANIC);
376
377 return 0;
378}
379core_initcall(fsnotify_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
4 */
5
6#include <linux/dcache.h>
7#include <linux/fs.h>
8#include <linux/gfp.h>
9#include <linux/init.h>
10#include <linux/module.h>
11#include <linux/mount.h>
12#include <linux/srcu.h>
13
14#include <linux/fsnotify_backend.h>
15#include "fsnotify.h"
16
17/*
18 * Clear all of the marks on an inode when it is being evicted from core
19 */
20void __fsnotify_inode_delete(struct inode *inode)
21{
22 fsnotify_clear_marks_by_inode(inode);
23}
24EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
25
26void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
27{
28 fsnotify_clear_marks_by_mount(mnt);
29}
30
31/**
32 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
33 * @sb: superblock being unmounted.
34 *
35 * Called during unmount with no locks held, so needs to be safe against
36 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
37 */
38static void fsnotify_unmount_inodes(struct super_block *sb)
39{
40 struct inode *inode, *iput_inode = NULL;
41
42 spin_lock(&sb->s_inode_list_lock);
43 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
44 /*
45 * We cannot __iget() an inode in state I_FREEING,
46 * I_WILL_FREE, or I_NEW which is fine because by that point
47 * the inode cannot have any associated watches.
48 */
49 spin_lock(&inode->i_lock);
50 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
51 spin_unlock(&inode->i_lock);
52 continue;
53 }
54
55 /*
56 * If i_count is zero, the inode cannot have any watches and
57 * doing an __iget/iput with SB_ACTIVE clear would actually
58 * evict all inodes with zero i_count from icache which is
59 * unnecessarily violent and may in fact be illegal to do.
60 */
61 if (!atomic_read(&inode->i_count)) {
62 spin_unlock(&inode->i_lock);
63 continue;
64 }
65
66 __iget(inode);
67 spin_unlock(&inode->i_lock);
68 spin_unlock(&sb->s_inode_list_lock);
69
70 if (iput_inode)
71 iput(iput_inode);
72
73 /* for each watch, send FS_UNMOUNT and then remove it */
74 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
75
76 fsnotify_inode_delete(inode);
77
78 iput_inode = inode;
79
80 spin_lock(&sb->s_inode_list_lock);
81 }
82 spin_unlock(&sb->s_inode_list_lock);
83
84 if (iput_inode)
85 iput(iput_inode);
86 /* Wait for outstanding inode references from connectors */
87 wait_var_event(&sb->s_fsnotify_inode_refs,
88 !atomic_long_read(&sb->s_fsnotify_inode_refs));
89}
90
91void fsnotify_sb_delete(struct super_block *sb)
92{
93 fsnotify_unmount_inodes(sb);
94 fsnotify_clear_marks_by_sb(sb);
95}
96
97/*
98 * Given an inode, first check if we care what happens to our children. Inotify
99 * and dnotify both tell their parents about events. If we care about any event
100 * on a child we run all of our children and set a dentry flag saying that the
101 * parent cares. Thus when an event happens on a child it can quickly tell if
102 * if there is a need to find a parent and send the event to the parent.
103 */
104void __fsnotify_update_child_dentry_flags(struct inode *inode)
105{
106 struct dentry *alias;
107 int watched;
108
109 if (!S_ISDIR(inode->i_mode))
110 return;
111
112 /* determine if the children should tell inode about their events */
113 watched = fsnotify_inode_watches_children(inode);
114
115 spin_lock(&inode->i_lock);
116 /* run all of the dentries associated with this inode. Since this is a
117 * directory, there damn well better only be one item on this list */
118 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
119 struct dentry *child;
120
121 /* run all of the children of the original inode and fix their
122 * d_flags to indicate parental interest (their parent is the
123 * original inode) */
124 spin_lock(&alias->d_lock);
125 list_for_each_entry(child, &alias->d_subdirs, d_child) {
126 if (!child->d_inode)
127 continue;
128
129 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
130 if (watched)
131 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
132 else
133 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
134 spin_unlock(&child->d_lock);
135 }
136 spin_unlock(&alias->d_lock);
137 }
138 spin_unlock(&inode->i_lock);
139}
140
141/* Notify this dentry's parent about a child's events. */
142int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
143{
144 struct dentry *parent;
145 struct inode *p_inode;
146 int ret = 0;
147
148 if (!dentry)
149 dentry = path->dentry;
150
151 if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
152 return 0;
153
154 parent = dget_parent(dentry);
155 p_inode = parent->d_inode;
156
157 if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
158 __fsnotify_update_child_dentry_flags(p_inode);
159 } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
160 struct name_snapshot name;
161
162 /* we are notifying a parent so come up with the new mask which
163 * specifies these are events which came from a child. */
164 mask |= FS_EVENT_ON_CHILD;
165
166 take_dentry_name_snapshot(&name, dentry);
167 if (path)
168 ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
169 &name.name, 0);
170 else
171 ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
172 &name.name, 0);
173 release_dentry_name_snapshot(&name);
174 }
175
176 dput(parent);
177
178 return ret;
179}
180EXPORT_SYMBOL_GPL(__fsnotify_parent);
181
182static int send_to_group(struct inode *to_tell,
183 __u32 mask, const void *data,
184 int data_is, u32 cookie,
185 const struct qstr *file_name,
186 struct fsnotify_iter_info *iter_info)
187{
188 struct fsnotify_group *group = NULL;
189 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
190 __u32 marks_mask = 0;
191 __u32 marks_ignored_mask = 0;
192 struct fsnotify_mark *mark;
193 int type;
194
195 if (WARN_ON(!iter_info->report_mask))
196 return 0;
197
198 /* clear ignored on inode modification */
199 if (mask & FS_MODIFY) {
200 fsnotify_foreach_obj_type(type) {
201 if (!fsnotify_iter_should_report_type(iter_info, type))
202 continue;
203 mark = iter_info->marks[type];
204 if (mark &&
205 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
206 mark->ignored_mask = 0;
207 }
208 }
209
210 fsnotify_foreach_obj_type(type) {
211 if (!fsnotify_iter_should_report_type(iter_info, type))
212 continue;
213 mark = iter_info->marks[type];
214 /* does the object mark tell us to do something? */
215 if (mark) {
216 group = mark->group;
217 marks_mask |= mark->mask;
218 marks_ignored_mask |= mark->ignored_mask;
219 }
220 }
221
222 pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x"
223 " data=%p data_is=%d cookie=%d\n",
224 __func__, group, to_tell, mask, marks_mask, marks_ignored_mask,
225 data, data_is, cookie);
226
227 if (!(test_mask & marks_mask & ~marks_ignored_mask))
228 return 0;
229
230 return group->ops->handle_event(group, to_tell, mask, data, data_is,
231 file_name, cookie, iter_info);
232}
233
234static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
235{
236 struct fsnotify_mark_connector *conn;
237 struct hlist_node *node = NULL;
238
239 conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
240 if (conn)
241 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
242
243 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
244}
245
246static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
247{
248 struct hlist_node *node = NULL;
249
250 if (mark)
251 node = srcu_dereference(mark->obj_list.next,
252 &fsnotify_mark_srcu);
253
254 return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
255}
256
257/*
258 * iter_info is a multi head priority queue of marks.
259 * Pick a subset of marks from queue heads, all with the
260 * same group and set the report_mask for selected subset.
261 * Returns the report_mask of the selected subset.
262 */
263static unsigned int fsnotify_iter_select_report_types(
264 struct fsnotify_iter_info *iter_info)
265{
266 struct fsnotify_group *max_prio_group = NULL;
267 struct fsnotify_mark *mark;
268 int type;
269
270 /* Choose max prio group among groups of all queue heads */
271 fsnotify_foreach_obj_type(type) {
272 mark = iter_info->marks[type];
273 if (mark &&
274 fsnotify_compare_groups(max_prio_group, mark->group) > 0)
275 max_prio_group = mark->group;
276 }
277
278 if (!max_prio_group)
279 return 0;
280
281 /* Set the report mask for marks from same group as max prio group */
282 iter_info->report_mask = 0;
283 fsnotify_foreach_obj_type(type) {
284 mark = iter_info->marks[type];
285 if (mark &&
286 fsnotify_compare_groups(max_prio_group, mark->group) == 0)
287 fsnotify_iter_set_report_type(iter_info, type);
288 }
289
290 return iter_info->report_mask;
291}
292
293/*
294 * Pop from iter_info multi head queue, the marks that were iterated in the
295 * current iteration step.
296 */
297static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
298{
299 int type;
300
301 fsnotify_foreach_obj_type(type) {
302 if (fsnotify_iter_should_report_type(iter_info, type))
303 iter_info->marks[type] =
304 fsnotify_next_mark(iter_info->marks[type]);
305 }
306}
307
308/*
309 * This is the main call to fsnotify. The VFS calls into hook specific functions
310 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
311 * out to all of the registered fsnotify_group. Those groups can then use the
312 * notification event in whatever means they feel necessary.
313 */
314int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
315 const struct qstr *file_name, u32 cookie)
316{
317 struct fsnotify_iter_info iter_info = {};
318 struct super_block *sb = to_tell->i_sb;
319 struct mount *mnt = NULL;
320 __u32 mnt_or_sb_mask = sb->s_fsnotify_mask;
321 int ret = 0;
322 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
323
324 if (data_is == FSNOTIFY_EVENT_PATH) {
325 mnt = real_mount(((const struct path *)data)->mnt);
326 mnt_or_sb_mask |= mnt->mnt_fsnotify_mask;
327 }
328 /* An event "on child" is not intended for a mount/sb mark */
329 if (mask & FS_EVENT_ON_CHILD)
330 mnt_or_sb_mask = 0;
331
332 /*
333 * Optimization: srcu_read_lock() has a memory barrier which can
334 * be expensive. It protects walking the *_fsnotify_marks lists.
335 * However, if we do not walk the lists, we do not have to do
336 * SRCU because we have no references to any objects and do not
337 * need SRCU to keep them "alive".
338 */
339 if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
340 (!mnt || !mnt->mnt_fsnotify_marks))
341 return 0;
342 /*
343 * if this is a modify event we may need to clear the ignored masks
344 * otherwise return if neither the inode nor the vfsmount/sb care about
345 * this type of event.
346 */
347 if (!(mask & FS_MODIFY) &&
348 !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask)))
349 return 0;
350
351 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
352
353 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
354 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
355 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
356 fsnotify_first_mark(&sb->s_fsnotify_marks);
357 if (mnt) {
358 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
359 fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
360 }
361
362 /*
363 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
364 * ignore masks are properly reflected for mount/sb mark notifications.
365 * That's why this traversal is so complicated...
366 */
367 while (fsnotify_iter_select_report_types(&iter_info)) {
368 ret = send_to_group(to_tell, mask, data, data_is, cookie,
369 file_name, &iter_info);
370
371 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
372 goto out;
373
374 fsnotify_iter_next(&iter_info);
375 }
376 ret = 0;
377out:
378 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
379
380 return ret;
381}
382EXPORT_SYMBOL_GPL(fsnotify);
383
384extern struct kmem_cache *fsnotify_mark_connector_cachep;
385
386static __init int fsnotify_init(void)
387{
388 int ret;
389
390 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
391
392 ret = init_srcu_struct(&fsnotify_mark_srcu);
393 if (ret)
394 panic("initializing fsnotify_mark_srcu");
395
396 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
397 SLAB_PANIC);
398
399 return 0;
400}
401core_initcall(fsnotify_init);