]>
Commit | Line | Data |
---|---|---|
2d9048e2 AG |
1 | /* |
2 | * fs/inotify_user.c - inotify support for userspace | |
3 | * | |
4 | * Authors: | |
5 | * John McCutchan <ttb@tentacle.dhs.org> | |
6 | * Robert Love <rml@novell.com> | |
7 | * | |
8 | * Copyright (C) 2005 John McCutchan | |
9 | * Copyright 2006 Hewlett-Packard Development Company, L.P. | |
10 | * | |
63c882a0 EP |
11 | * Copyright (C) 2009 Eric Paris <Red Hat Inc> |
12 | * inotify was largely rewriten to make use of the fsnotify infrastructure | |
13 | * | |
2d9048e2 AG |
14 | * This program is free software; you can redistribute it and/or modify it |
15 | * under the terms of the GNU General Public License as published by the | |
16 | * Free Software Foundation; either version 2, or (at your option) any | |
17 | * later version. | |
18 | * | |
19 | * This program is distributed in the hope that it will be useful, but | |
20 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
21 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
22 | * General Public License for more details. | |
23 | */ | |
24 | ||
2d9048e2 | 25 | #include <linux/file.h> |
63c882a0 EP |
26 | #include <linux/fs.h> /* struct inode */ |
27 | #include <linux/fsnotify_backend.h> | |
28 | #include <linux/idr.h> | |
29 | #include <linux/init.h> /* module_init */ | |
2d9048e2 | 30 | #include <linux/inotify.h> |
63c882a0 | 31 | #include <linux/kernel.h> /* roundup() */ |
63c882a0 | 32 | #include <linux/namei.h> /* LOOKUP_FOLLOW */ |
63c882a0 EP |
33 | #include <linux/sched.h> /* struct user */ |
34 | #include <linux/slab.h> /* struct kmem_cache */ | |
2d9048e2 | 35 | #include <linux/syscalls.h> |
63c882a0 | 36 | #include <linux/types.h> |
c44dcc56 | 37 | #include <linux/anon_inodes.h> |
63c882a0 EP |
38 | #include <linux/uaccess.h> |
39 | #include <linux/poll.h> | |
40 | #include <linux/wait.h> | |
2d9048e2 | 41 | |
63c882a0 | 42 | #include "inotify.h" |
be77196b | 43 | #include "../fdinfo.h" |
2d9048e2 | 44 | |
63c882a0 | 45 | #include <asm/ioctls.h> |
2d9048e2 | 46 | |
2d9048e2 | 47 | /* these are configurable via /proc/sys/fs/inotify/ */ |
3c828e49 | 48 | static int inotify_max_user_instances __read_mostly; |
3c828e49 | 49 | static int inotify_max_queued_events __read_mostly; |
0a24887a | 50 | static int inotify_max_user_watches __read_mostly; |
2d9048e2 | 51 | |
63c882a0 | 52 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
2d9048e2 | 53 | |
2d9048e2 AG |
54 | #ifdef CONFIG_SYSCTL |
55 | ||
56 | #include <linux/sysctl.h> | |
57 | ||
58 | static int zero; | |
59 | ||
92f778dd | 60 | struct ctl_table inotify_table[] = { |
2d9048e2 | 61 | { |
2d9048e2 AG |
62 | .procname = "max_user_instances", |
63 | .data = &inotify_max_user_instances, | |
64 | .maxlen = sizeof(int), | |
65 | .mode = 0644, | |
6d456111 | 66 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
67 | .extra1 = &zero, |
68 | }, | |
69 | { | |
2d9048e2 AG |
70 | .procname = "max_user_watches", |
71 | .data = &inotify_max_user_watches, | |
72 | .maxlen = sizeof(int), | |
73 | .mode = 0644, | |
6d456111 | 74 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
75 | .extra1 = &zero, |
76 | }, | |
77 | { | |
2d9048e2 AG |
78 | .procname = "max_queued_events", |
79 | .data = &inotify_max_queued_events, | |
80 | .maxlen = sizeof(int), | |
81 | .mode = 0644, | |
6d456111 | 82 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
83 | .extra1 = &zero |
84 | }, | |
ab09203e | 85 | { } |
2d9048e2 AG |
86 | }; |
87 | #endif /* CONFIG_SYSCTL */ | |
88 | ||
63c882a0 | 89 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 90 | { |
63c882a0 | 91 | __u32 mask; |
1c17d18e | 92 | |
611da04f EP |
93 | /* |
94 | * everything should accept their own ignored, cares about children, | |
95 | * and should receive events when the inode is unmounted | |
96 | */ | |
97 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); | |
2d9048e2 | 98 | |
63c882a0 | 99 | /* mask off the flags used to open the fd */ |
8c1934c8 | 100 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); |
2d9048e2 | 101 | |
63c882a0 | 102 | return mask; |
2d9048e2 AG |
103 | } |
104 | ||
63c882a0 | 105 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 106 | { |
63c882a0 EP |
107 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
108 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
109 | } |
110 | ||
63c882a0 | 111 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
112 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
113 | { | |
63c882a0 | 114 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
115 | int ret = 0; |
116 | ||
63c882a0 EP |
117 | poll_wait(file, &group->notification_waitq, wait); |
118 | mutex_lock(&group->notification_mutex); | |
119 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 120 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 121 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
122 | |
123 | return ret; | |
124 | } | |
125 | ||
7053aee2 | 126 | static int round_event_name_len(struct fsnotify_event *fsn_event) |
e9fe6904 | 127 | { |
7053aee2 JK |
128 | struct inotify_event_info *event; |
129 | ||
130 | event = INOTIFY_E(fsn_event); | |
e9fe6904 JK |
131 | if (!event->name_len) |
132 | return 0; | |
133 | return roundup(event->name_len + 1, sizeof(struct inotify_event)); | |
134 | } | |
135 | ||
3632dee2 VN |
136 | /* |
137 | * Get an inotify_kernel_event if one exists and is small | |
138 | * enough to fit in "count". Return an error pointer if | |
139 | * not large enough. | |
140 | * | |
63c882a0 | 141 | * Called with the group->notification_mutex held. |
3632dee2 | 142 | */ |
63c882a0 EP |
143 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
144 | size_t count) | |
3632dee2 VN |
145 | { |
146 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 147 | struct fsnotify_event *event; |
3632dee2 | 148 | |
63c882a0 | 149 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
150 | return NULL; |
151 | ||
8ba8fa91 | 152 | event = fsnotify_peek_first_event(group); |
63c882a0 | 153 | |
5ba08e2e EP |
154 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
155 | ||
e9fe6904 | 156 | event_size += round_event_name_len(event); |
3632dee2 VN |
157 | if (event_size > count) |
158 | return ERR_PTR(-EINVAL); | |
159 | ||
63c882a0 EP |
160 | /* held the notification_mutex the whole time, so this is the |
161 | * same event we peeked above */ | |
8ba8fa91 | 162 | fsnotify_remove_first_event(group); |
63c882a0 EP |
163 | |
164 | return event; | |
3632dee2 VN |
165 | } |
166 | ||
167 | /* | |
168 | * Copy an event to user space, returning how much we copied. | |
169 | * | |
170 | * We already checked that the event size is smaller than the | |
171 | * buffer we had in "get_one_event()" above. | |
172 | */ | |
63c882a0 | 173 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
7053aee2 | 174 | struct fsnotify_event *fsn_event, |
3632dee2 VN |
175 | char __user *buf) |
176 | { | |
63c882a0 | 177 | struct inotify_event inotify_event; |
7053aee2 | 178 | struct inotify_event_info *event; |
3632dee2 | 179 | size_t event_size = sizeof(struct inotify_event); |
e9fe6904 JK |
180 | size_t name_len; |
181 | size_t pad_name_len; | |
63c882a0 | 182 | |
7053aee2 | 183 | pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event); |
63c882a0 | 184 | |
7053aee2 | 185 | event = INOTIFY_E(fsn_event); |
e9fe6904 | 186 | name_len = event->name_len; |
b962e731 | 187 | /* |
e9fe6904 | 188 | * round up name length so it is a multiple of event_size |
0db501bd EB |
189 | * plus an extra byte for the terminating '\0'. |
190 | */ | |
7053aee2 | 191 | pad_name_len = round_event_name_len(fsn_event); |
e9fe6904 | 192 | inotify_event.len = pad_name_len; |
7053aee2 JK |
193 | inotify_event.mask = inotify_mask_to_arg(fsn_event->mask); |
194 | inotify_event.wd = event->wd; | |
63c882a0 | 195 | inotify_event.cookie = event->sync_cookie; |
3632dee2 | 196 | |
63c882a0 EP |
197 | /* send the main event */ |
198 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
199 | return -EFAULT; |
200 | ||
63c882a0 | 201 | buf += event_size; |
3632dee2 | 202 | |
63c882a0 EP |
203 | /* |
204 | * fsnotify only stores the pathname, so here we have to send the pathname | |
205 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
e9fe6904 | 206 | * with zeros. |
63c882a0 | 207 | */ |
e9fe6904 | 208 | if (pad_name_len) { |
63c882a0 | 209 | /* copy the path name */ |
7053aee2 | 210 | if (copy_to_user(buf, event->name, name_len)) |
3632dee2 | 211 | return -EFAULT; |
e9fe6904 | 212 | buf += name_len; |
3632dee2 | 213 | |
0db501bd | 214 | /* fill userspace with 0's */ |
e9fe6904 | 215 | if (clear_user(buf, pad_name_len - name_len)) |
63c882a0 | 216 | return -EFAULT; |
e9fe6904 | 217 | event_size += pad_name_len; |
3632dee2 | 218 | } |
63c882a0 | 219 | |
3632dee2 VN |
220 | return event_size; |
221 | } | |
222 | ||
2d9048e2 AG |
223 | static ssize_t inotify_read(struct file *file, char __user *buf, |
224 | size_t count, loff_t *pos) | |
225 | { | |
63c882a0 EP |
226 | struct fsnotify_group *group; |
227 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
228 | char __user *start; |
229 | int ret; | |
e23738a7 | 230 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
2d9048e2 AG |
231 | |
232 | start = buf; | |
63c882a0 | 233 | group = file->private_data; |
2d9048e2 | 234 | |
e23738a7 | 235 | add_wait_queue(&group->notification_waitq, &wait); |
2d9048e2 | 236 | while (1) { |
63c882a0 EP |
237 | mutex_lock(&group->notification_mutex); |
238 | kevent = get_one_event(group, count); | |
239 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 240 | |
5ba08e2e EP |
241 | pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); |
242 | ||
3632dee2 VN |
243 | if (kevent) { |
244 | ret = PTR_ERR(kevent); | |
245 | if (IS_ERR(kevent)) | |
246 | break; | |
63c882a0 | 247 | ret = copy_event_to_user(group, kevent, buf); |
7053aee2 | 248 | fsnotify_destroy_event(group, kevent); |
3632dee2 VN |
249 | if (ret < 0) |
250 | break; | |
251 | buf += ret; | |
252 | count -= ret; | |
253 | continue; | |
2d9048e2 AG |
254 | } |
255 | ||
3632dee2 VN |
256 | ret = -EAGAIN; |
257 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 258 | break; |
1ca39ab9 | 259 | ret = -ERESTARTSYS; |
3632dee2 | 260 | if (signal_pending(current)) |
2d9048e2 | 261 | break; |
16dbc6c9 | 262 | |
3632dee2 | 263 | if (start != buf) |
2d9048e2 | 264 | break; |
16dbc6c9 | 265 | |
e23738a7 | 266 | wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
2d9048e2 | 267 | } |
e23738a7 | 268 | remove_wait_queue(&group->notification_waitq, &wait); |
2d9048e2 | 269 | |
3632dee2 VN |
270 | if (start != buf && ret != -EFAULT) |
271 | ret = buf - start; | |
2d9048e2 AG |
272 | return ret; |
273 | } | |
274 | ||
275 | static int inotify_release(struct inode *ignored, struct file *file) | |
276 | { | |
63c882a0 | 277 | struct fsnotify_group *group = file->private_data; |
2d9048e2 | 278 | |
5ba08e2e EP |
279 | pr_debug("%s: group=%p\n", __func__, group); |
280 | ||
63c882a0 | 281 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
d8153d4d | 282 | fsnotify_destroy_group(group); |
2d9048e2 AG |
283 | |
284 | return 0; | |
285 | } | |
286 | ||
287 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
288 | unsigned long arg) | |
289 | { | |
63c882a0 | 290 | struct fsnotify_group *group; |
7053aee2 | 291 | struct fsnotify_event *fsn_event; |
2d9048e2 AG |
292 | void __user *p; |
293 | int ret = -ENOTTY; | |
63c882a0 | 294 | size_t send_len = 0; |
2d9048e2 | 295 | |
63c882a0 | 296 | group = file->private_data; |
2d9048e2 AG |
297 | p = (void __user *) arg; |
298 | ||
5ba08e2e EP |
299 | pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); |
300 | ||
2d9048e2 AG |
301 | switch (cmd) { |
302 | case FIONREAD: | |
63c882a0 | 303 | mutex_lock(&group->notification_mutex); |
7053aee2 JK |
304 | list_for_each_entry(fsn_event, &group->notification_list, |
305 | list) { | |
63c882a0 | 306 | send_len += sizeof(struct inotify_event); |
7053aee2 | 307 | send_len += round_event_name_len(fsn_event); |
63c882a0 EP |
308 | } |
309 | mutex_unlock(&group->notification_mutex); | |
310 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
311 | break; |
312 | } | |
313 | ||
314 | return ret; | |
315 | } | |
316 | ||
317 | static const struct file_operations inotify_fops = { | |
be77196b | 318 | .show_fdinfo = inotify_show_fdinfo, |
63c882a0 EP |
319 | .poll = inotify_poll, |
320 | .read = inotify_read, | |
0a6b6bd5 | 321 | .fasync = fsnotify_fasync, |
63c882a0 EP |
322 | .release = inotify_release, |
323 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 | 324 | .compat_ioctl = inotify_ioctl, |
6038f373 | 325 | .llseek = noop_llseek, |
2d9048e2 AG |
326 | }; |
327 | ||
2d9048e2 | 328 | |
63c882a0 EP |
329 | /* |
330 | * find_inode - resolve a user-given path to a specific inode | |
331 | */ | |
332 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
333 | { | |
334 | int error; | |
335 | ||
336 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
337 | if (error) | |
338 | return error; | |
339 | /* you can only watch an inode if you have read permissions on it */ | |
340 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
341 | if (error) | |
342 | path_put(path); | |
343 | return error; | |
344 | } | |
345 | ||
b7ba8371 | 346 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
000285de | 347 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
348 | { |
349 | int ret; | |
350 | ||
4542da63 TH |
351 | idr_preload(GFP_KERNEL); |
352 | spin_lock(idr_lock); | |
b7ba8371 | 353 | |
a66c04b4 | 354 | ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT); |
4542da63 | 355 | if (ret >= 0) { |
b7ba8371 | 356 | /* we added the mark to the idr, take a reference */ |
4542da63 | 357 | i_mark->wd = ret; |
4542da63 TH |
358 | fsnotify_get_mark(&i_mark->fsn_mark); |
359 | } | |
b7ba8371 | 360 | |
4542da63 TH |
361 | spin_unlock(idr_lock); |
362 | idr_preload_end(); | |
363 | return ret < 0 ? ret : 0; | |
b7ba8371 EP |
364 | } |
365 | ||
000285de | 366 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, |
b7ba8371 EP |
367 | int wd) |
368 | { | |
369 | struct idr *idr = &group->inotify_data.idr; | |
370 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 371 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
372 | |
373 | assert_spin_locked(idr_lock); | |
374 | ||
000285de EP |
375 | i_mark = idr_find(idr, wd); |
376 | if (i_mark) { | |
377 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
b7ba8371 | 378 | |
000285de | 379 | fsnotify_get_mark(fsn_mark); |
b7ba8371 | 380 | /* One ref for being in the idr, one ref we just took */ |
000285de | 381 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); |
b7ba8371 EP |
382 | } |
383 | ||
000285de | 384 | return i_mark; |
b7ba8371 EP |
385 | } |
386 | ||
000285de | 387 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, |
b7ba8371 EP |
388 | int wd) |
389 | { | |
000285de | 390 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
391 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
392 | ||
393 | spin_lock(idr_lock); | |
000285de | 394 | i_mark = inotify_idr_find_locked(group, wd); |
b7ba8371 EP |
395 | spin_unlock(idr_lock); |
396 | ||
000285de | 397 | return i_mark; |
b7ba8371 EP |
398 | } |
399 | ||
400 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
000285de | 401 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
402 | { |
403 | struct idr *idr = &group->inotify_data.idr; | |
404 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 405 | int wd = i_mark->wd; |
b7ba8371 EP |
406 | |
407 | assert_spin_locked(idr_lock); | |
408 | ||
409 | idr_remove(idr, wd); | |
410 | ||
411 | /* removed from the idr, drop that ref */ | |
000285de | 412 | fsnotify_put_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
413 | } |
414 | ||
dead537d EP |
415 | /* |
416 | * Remove the mark from the idr (if present) and drop the reference | |
417 | * on the mark because it was in the idr. | |
418 | */ | |
7e790dd5 | 419 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
000285de | 420 | struct inotify_inode_mark *i_mark) |
7e790dd5 | 421 | { |
b7ba8371 | 422 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
000285de | 423 | struct inotify_inode_mark *found_i_mark = NULL; |
dead537d | 424 | int wd; |
7e790dd5 | 425 | |
b7ba8371 | 426 | spin_lock(idr_lock); |
000285de | 427 | wd = i_mark->wd; |
dead537d | 428 | |
b7ba8371 | 429 | /* |
000285de | 430 | * does this i_mark think it is in the idr? we shouldn't get called |
b7ba8371 EP |
431 | * if it wasn't.... |
432 | */ | |
433 | if (wd == -1) { | |
000285de EP |
434 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" |
435 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
436 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 437 | goto out; |
b7ba8371 | 438 | } |
dead537d | 439 | |
b7ba8371 | 440 | /* Lets look in the idr to see if we find it */ |
000285de EP |
441 | found_i_mark = inotify_idr_find_locked(group, wd); |
442 | if (unlikely(!found_i_mark)) { | |
443 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
444 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
445 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 446 | goto out; |
b7ba8371 | 447 | } |
dead537d | 448 | |
b7ba8371 | 449 | /* |
000285de EP |
450 | * We found an mark in the idr at the right wd, but it's |
451 | * not the mark we were told to remove. eparis seriously | |
b7ba8371 EP |
452 | * fucked up somewhere. |
453 | */ | |
000285de EP |
454 | if (unlikely(found_i_mark != i_mark)) { |
455 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
456 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
457 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
458 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
459 | i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, | |
460 | found_i_mark->fsn_mark.group, | |
461 | found_i_mark->fsn_mark.i.inode); | |
dead537d EP |
462 | goto out; |
463 | } | |
464 | ||
b7ba8371 EP |
465 | /* |
466 | * One ref for being in the idr | |
467 | * one ref held by the caller trying to kill us | |
468 | * one ref grabbed by inotify_idr_find | |
469 | */ | |
000285de EP |
470 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { |
471 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
472 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
473 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
b7ba8371 EP |
474 | /* we can't really recover with bad ref cnting.. */ |
475 | BUG(); | |
476 | } | |
dead537d | 477 | |
000285de | 478 | do_inotify_remove_from_idr(group, i_mark); |
dead537d | 479 | out: |
b7ba8371 | 480 | /* match the ref taken by inotify_idr_find_locked() */ |
000285de EP |
481 | if (found_i_mark) |
482 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
483 | i_mark->wd = -1; | |
b7ba8371 | 484 | spin_unlock(idr_lock); |
7e790dd5 | 485 | } |
dead537d | 486 | |
63c882a0 | 487 | /* |
dead537d | 488 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 489 | */ |
000285de | 490 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, |
528da3e9 | 491 | struct fsnotify_group *group) |
63c882a0 | 492 | { |
000285de | 493 | struct inotify_inode_mark *i_mark; |
63c882a0 | 494 | |
7053aee2 JK |
495 | /* Queue ignore event for the watch */ |
496 | inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED, | |
45a22f4c | 497 | NULL, FSNOTIFY_EVENT_NONE, NULL, 0); |
f44aebcc | 498 | |
7053aee2 | 499 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
000285de EP |
500 | /* remove this mark from the idr */ |
501 | inotify_remove_from_idr(group, i_mark); | |
63c882a0 | 502 | |
5549f7cd | 503 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
504 | } |
505 | ||
506 | /* ding dong the mark is dead */ | |
000285de | 507 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) |
63c882a0 | 508 | { |
000285de | 509 | struct inotify_inode_mark *i_mark; |
31ddd326 | 510 | |
000285de | 511 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 512 | |
000285de | 513 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); |
63c882a0 EP |
514 | } |
515 | ||
52cef755 EP |
516 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
517 | struct inode *inode, | |
518 | u32 arg) | |
63c882a0 | 519 | { |
000285de EP |
520 | struct fsnotify_mark *fsn_mark; |
521 | struct inotify_inode_mark *i_mark; | |
63c882a0 | 522 | __u32 old_mask, new_mask; |
52cef755 EP |
523 | __u32 mask; |
524 | int add = (arg & IN_MASK_ADD); | |
525 | int ret; | |
63c882a0 | 526 | |
63c882a0 | 527 | mask = inotify_arg_to_mask(arg); |
63c882a0 | 528 | |
5444e298 | 529 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
000285de | 530 | if (!fsn_mark) |
52cef755 | 531 | return -ENOENT; |
7e790dd5 | 532 | |
000285de | 533 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
75fe2b26 | 534 | |
000285de | 535 | spin_lock(&fsn_mark->lock); |
63c882a0 | 536 | |
000285de | 537 | old_mask = fsn_mark->mask; |
90b1e7a5 EP |
538 | if (add) |
539 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
540 | else | |
541 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
542 | new_mask = fsn_mark->mask; | |
63c882a0 | 543 | |
000285de | 544 | spin_unlock(&fsn_mark->lock); |
63c882a0 EP |
545 | |
546 | if (old_mask != new_mask) { | |
547 | /* more bits in old than in new? */ | |
548 | int dropped = (old_mask & ~new_mask); | |
000285de | 549 | /* more bits in this fsn_mark than the inode's mask? */ |
63c882a0 | 550 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); |
63c882a0 | 551 | |
000285de | 552 | /* update the inode with this new fsn_mark */ |
63c882a0 EP |
553 | if (dropped || do_inode) |
554 | fsnotify_recalc_inode_mask(inode); | |
555 | ||
63c882a0 EP |
556 | } |
557 | ||
52cef755 | 558 | /* return the wd */ |
000285de | 559 | ret = i_mark->wd; |
52cef755 | 560 | |
d0775441 | 561 | /* match the get from fsnotify_find_mark() */ |
000285de | 562 | fsnotify_put_mark(fsn_mark); |
75fe2b26 | 563 | |
52cef755 EP |
564 | return ret; |
565 | } | |
566 | ||
567 | static int inotify_new_watch(struct fsnotify_group *group, | |
568 | struct inode *inode, | |
569 | u32 arg) | |
570 | { | |
000285de | 571 | struct inotify_inode_mark *tmp_i_mark; |
52cef755 EP |
572 | __u32 mask; |
573 | int ret; | |
b7ba8371 EP |
574 | struct idr *idr = &group->inotify_data.idr; |
575 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 | 576 | |
52cef755 | 577 | mask = inotify_arg_to_mask(arg); |
52cef755 | 578 | |
000285de EP |
579 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); |
580 | if (unlikely(!tmp_i_mark)) | |
52cef755 EP |
581 | return -ENOMEM; |
582 | ||
000285de EP |
583 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); |
584 | tmp_i_mark->fsn_mark.mask = mask; | |
585 | tmp_i_mark->wd = -1; | |
52cef755 EP |
586 | |
587 | ret = -ENOSPC; | |
588 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
589 | goto out_err; | |
e0873344 | 590 | |
a66c04b4 | 591 | ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark); |
b7ba8371 | 592 | if (ret) |
52cef755 | 593 | goto out_err; |
52cef755 EP |
594 | |
595 | /* we are on the idr, now get on the inode */ | |
e1e5a9f8 LS |
596 | ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, group, inode, |
597 | NULL, 0); | |
52cef755 EP |
598 | if (ret) { |
599 | /* we failed to get on the inode, get off the idr */ | |
000285de | 600 | inotify_remove_from_idr(group, tmp_i_mark); |
52cef755 EP |
601 | goto out_err; |
602 | } | |
603 | ||
52cef755 EP |
604 | /* increment the number of watches the user has */ |
605 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
606 | ||
000285de EP |
607 | /* return the watch descriptor for this new mark */ |
608 | ret = tmp_i_mark->wd; | |
52cef755 | 609 | |
63c882a0 | 610 | out_err: |
000285de EP |
611 | /* match the ref from fsnotify_init_mark() */ |
612 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
52cef755 EP |
613 | |
614 | return ret; | |
615 | } | |
616 | ||
617 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
618 | { | |
619 | int ret = 0; | |
620 | ||
e1e5a9f8 | 621 | mutex_lock(&group->mark_mutex); |
52cef755 EP |
622 | /* try to update and existing watch with the new arg */ |
623 | ret = inotify_update_existing_watch(group, inode, arg); | |
624 | /* no mark present, try to add a new one */ | |
625 | if (ret == -ENOENT) | |
626 | ret = inotify_new_watch(group, inode, arg); | |
e1e5a9f8 | 627 | mutex_unlock(&group->mark_mutex); |
7e790dd5 | 628 | |
63c882a0 EP |
629 | return ret; |
630 | } | |
631 | ||
d0de4dc5 | 632 | static struct fsnotify_group *inotify_new_group(unsigned int max_events) |
63c882a0 EP |
633 | { |
634 | struct fsnotify_group *group; | |
ff57cd58 | 635 | struct inotify_event_info *oevent; |
63c882a0 | 636 | |
0d2e2a1d | 637 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
638 | if (IS_ERR(group)) |
639 | return group; | |
640 | ||
ff57cd58 JK |
641 | oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL); |
642 | if (unlikely(!oevent)) { | |
643 | fsnotify_destroy_group(group); | |
644 | return ERR_PTR(-ENOMEM); | |
645 | } | |
646 | group->overflow_event = &oevent->fse; | |
647 | fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW); | |
648 | oevent->wd = -1; | |
649 | oevent->sync_cookie = 0; | |
650 | oevent->name_len = 0; | |
651 | ||
63c882a0 EP |
652 | group->max_events = max_events; |
653 | ||
654 | spin_lock_init(&group->inotify_data.idr_lock); | |
655 | idr_init(&group->inotify_data.idr); | |
d0de4dc5 EP |
656 | group->inotify_data.user = get_current_user(); |
657 | ||
658 | if (atomic_inc_return(&group->inotify_data.user->inotify_devs) > | |
659 | inotify_max_user_instances) { | |
d8153d4d | 660 | fsnotify_destroy_group(group); |
d0de4dc5 EP |
661 | return ERR_PTR(-EMFILE); |
662 | } | |
63c882a0 EP |
663 | |
664 | return group; | |
665 | } | |
666 | ||
667 | ||
668 | /* inotify syscalls */ | |
938bb9f5 | 669 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 670 | { |
63c882a0 | 671 | struct fsnotify_group *group; |
c44dcc56 | 672 | int ret; |
2d9048e2 | 673 | |
e38b36f3 UD |
674 | /* Check the IN_* constants for consistency. */ |
675 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
676 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
677 | ||
510df2dd | 678 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
679 | return -EINVAL; |
680 | ||
63c882a0 | 681 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
d0de4dc5 EP |
682 | group = inotify_new_group(inotify_max_queued_events); |
683 | if (IS_ERR(group)) | |
684 | return PTR_ERR(group); | |
825f9692 | 685 | |
c44dcc56 AV |
686 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
687 | O_RDONLY | flags); | |
d0de4dc5 | 688 | if (ret < 0) |
d8153d4d | 689 | fsnotify_destroy_group(group); |
825f9692 | 690 | |
2d9048e2 AG |
691 | return ret; |
692 | } | |
693 | ||
938bb9f5 | 694 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
695 | { |
696 | return sys_inotify_init1(0); | |
697 | } | |
698 | ||
2e4d0924 HC |
699 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
700 | u32, mask) | |
2d9048e2 | 701 | { |
63c882a0 | 702 | struct fsnotify_group *group; |
2d9048e2 | 703 | struct inode *inode; |
2d8f3038 | 704 | struct path path; |
2903ff01 AV |
705 | struct fd f; |
706 | int ret; | |
2d9048e2 AG |
707 | unsigned flags = 0; |
708 | ||
04df32fa ZH |
709 | /* don't allow invalid bits: we don't want flags set */ |
710 | if (unlikely(!(mask & ALL_INOTIFY_BITS))) | |
711 | return -EINVAL; | |
712 | ||
2903ff01 AV |
713 | f = fdget(fd); |
714 | if (unlikely(!f.file)) | |
2d9048e2 AG |
715 | return -EBADF; |
716 | ||
717 | /* verify that this is indeed an inotify instance */ | |
2903ff01 | 718 | if (unlikely(f.file->f_op != &inotify_fops)) { |
2d9048e2 AG |
719 | ret = -EINVAL; |
720 | goto fput_and_out; | |
721 | } | |
722 | ||
723 | if (!(mask & IN_DONT_FOLLOW)) | |
724 | flags |= LOOKUP_FOLLOW; | |
725 | if (mask & IN_ONLYDIR) | |
726 | flags |= LOOKUP_DIRECTORY; | |
727 | ||
63c882a0 EP |
728 | ret = inotify_find_inode(pathname, &path, flags); |
729 | if (ret) | |
2d9048e2 AG |
730 | goto fput_and_out; |
731 | ||
63c882a0 | 732 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 733 | inode = path.dentry->d_inode; |
2903ff01 | 734 | group = f.file->private_data; |
2d9048e2 | 735 | |
63c882a0 EP |
736 | /* create/update an inode mark */ |
737 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 738 | path_put(&path); |
2d9048e2 | 739 | fput_and_out: |
2903ff01 | 740 | fdput(f); |
2d9048e2 AG |
741 | return ret; |
742 | } | |
743 | ||
2e4d0924 | 744 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 745 | { |
63c882a0 | 746 | struct fsnotify_group *group; |
000285de | 747 | struct inotify_inode_mark *i_mark; |
2903ff01 AV |
748 | struct fd f; |
749 | int ret = 0; | |
2d9048e2 | 750 | |
2903ff01 AV |
751 | f = fdget(fd); |
752 | if (unlikely(!f.file)) | |
2d9048e2 AG |
753 | return -EBADF; |
754 | ||
755 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 | 756 | ret = -EINVAL; |
2903ff01 | 757 | if (unlikely(f.file->f_op != &inotify_fops)) |
2d9048e2 | 758 | goto out; |
2d9048e2 | 759 | |
2903ff01 | 760 | group = f.file->private_data; |
2d9048e2 | 761 | |
b7ba8371 | 762 | ret = -EINVAL; |
000285de EP |
763 | i_mark = inotify_idr_find(group, wd); |
764 | if (unlikely(!i_mark)) | |
63c882a0 | 765 | goto out; |
63c882a0 | 766 | |
b7ba8371 EP |
767 | ret = 0; |
768 | ||
e2a29943 | 769 | fsnotify_destroy_mark(&i_mark->fsn_mark, group); |
b7ba8371 EP |
770 | |
771 | /* match ref taken by inotify_idr_find */ | |
000285de | 772 | fsnotify_put_mark(&i_mark->fsn_mark); |
2d9048e2 AG |
773 | |
774 | out: | |
2903ff01 | 775 | fdput(f); |
2d9048e2 AG |
776 | return ret; |
777 | } | |
778 | ||
2d9048e2 | 779 | /* |
ae0e47f0 | 780 | * inotify_user_setup - Our initialization function. Note that we cannot return |
2d9048e2 AG |
781 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
782 | * must result in panic(). | |
783 | */ | |
784 | static int __init inotify_user_setup(void) | |
785 | { | |
f874e1ac EP |
786 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); |
787 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
788 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
789 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
790 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
791 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
792 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
793 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
794 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
795 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
796 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
797 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
798 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
799 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
800 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
801 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
b29866aa | 802 | BUILD_BUG_ON(IN_ISDIR != FS_ISDIR); |
f874e1ac EP |
803 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); |
804 | ||
805 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
806 | ||
000285de | 807 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); |
63c882a0 | 808 | |
2d9048e2 AG |
809 | inotify_max_queued_events = 16384; |
810 | inotify_max_user_instances = 128; | |
811 | inotify_max_user_watches = 8192; | |
812 | ||
2d9048e2 AG |
813 | return 0; |
814 | } | |
2d9048e2 | 815 | module_init(inotify_user_setup); |