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 EP |
52 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
53 | struct kmem_cache *event_priv_cachep __read_mostly; | |
2d9048e2 | 54 | |
2d9048e2 AG |
55 | #ifdef CONFIG_SYSCTL |
56 | ||
57 | #include <linux/sysctl.h> | |
58 | ||
59 | static int zero; | |
60 | ||
61 | ctl_table inotify_table[] = { | |
62 | { | |
2d9048e2 AG |
63 | .procname = "max_user_instances", |
64 | .data = &inotify_max_user_instances, | |
65 | .maxlen = sizeof(int), | |
66 | .mode = 0644, | |
6d456111 | 67 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
68 | .extra1 = &zero, |
69 | }, | |
70 | { | |
2d9048e2 AG |
71 | .procname = "max_user_watches", |
72 | .data = &inotify_max_user_watches, | |
73 | .maxlen = sizeof(int), | |
74 | .mode = 0644, | |
6d456111 | 75 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
76 | .extra1 = &zero, |
77 | }, | |
78 | { | |
2d9048e2 AG |
79 | .procname = "max_queued_events", |
80 | .data = &inotify_max_queued_events, | |
81 | .maxlen = sizeof(int), | |
82 | .mode = 0644, | |
6d456111 | 83 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
84 | .extra1 = &zero |
85 | }, | |
ab09203e | 86 | { } |
2d9048e2 AG |
87 | }; |
88 | #endif /* CONFIG_SYSCTL */ | |
89 | ||
63c882a0 | 90 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 91 | { |
63c882a0 | 92 | __u32 mask; |
1c17d18e | 93 | |
611da04f EP |
94 | /* |
95 | * everything should accept their own ignored, cares about children, | |
96 | * and should receive events when the inode is unmounted | |
97 | */ | |
98 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); | |
2d9048e2 | 99 | |
63c882a0 | 100 | /* mask off the flags used to open the fd */ |
8c1934c8 | 101 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); |
2d9048e2 | 102 | |
63c882a0 | 103 | return mask; |
2d9048e2 AG |
104 | } |
105 | ||
63c882a0 | 106 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 107 | { |
63c882a0 EP |
108 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
109 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
110 | } |
111 | ||
63c882a0 | 112 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
113 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
114 | { | |
63c882a0 | 115 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
116 | int ret = 0; |
117 | ||
63c882a0 EP |
118 | poll_wait(file, &group->notification_waitq, wait); |
119 | mutex_lock(&group->notification_mutex); | |
120 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 121 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 122 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
123 | |
124 | return ret; | |
125 | } | |
126 | ||
3632dee2 VN |
127 | /* |
128 | * Get an inotify_kernel_event if one exists and is small | |
129 | * enough to fit in "count". Return an error pointer if | |
130 | * not large enough. | |
131 | * | |
63c882a0 | 132 | * Called with the group->notification_mutex held. |
3632dee2 | 133 | */ |
63c882a0 EP |
134 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
135 | size_t count) | |
3632dee2 VN |
136 | { |
137 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 138 | struct fsnotify_event *event; |
3632dee2 | 139 | |
63c882a0 | 140 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
141 | return NULL; |
142 | ||
63c882a0 EP |
143 | event = fsnotify_peek_notify_event(group); |
144 | ||
5ba08e2e EP |
145 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
146 | ||
83cb10f0 EP |
147 | if (event->name_len) |
148 | event_size += roundup(event->name_len + 1, event_size); | |
3632dee2 VN |
149 | |
150 | if (event_size > count) | |
151 | return ERR_PTR(-EINVAL); | |
152 | ||
63c882a0 EP |
153 | /* held the notification_mutex the whole time, so this is the |
154 | * same event we peeked above */ | |
155 | fsnotify_remove_notify_event(group); | |
156 | ||
157 | return event; | |
3632dee2 VN |
158 | } |
159 | ||
160 | /* | |
161 | * Copy an event to user space, returning how much we copied. | |
162 | * | |
163 | * We already checked that the event size is smaller than the | |
164 | * buffer we had in "get_one_event()" above. | |
165 | */ | |
63c882a0 EP |
166 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
167 | struct fsnotify_event *event, | |
3632dee2 VN |
168 | char __user *buf) |
169 | { | |
63c882a0 EP |
170 | struct inotify_event inotify_event; |
171 | struct fsnotify_event_private_data *fsn_priv; | |
172 | struct inotify_event_private_data *priv; | |
3632dee2 | 173 | size_t event_size = sizeof(struct inotify_event); |
b962e731 | 174 | size_t name_len = 0; |
63c882a0 | 175 | |
5ba08e2e EP |
176 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
177 | ||
63c882a0 EP |
178 | /* we get the inotify watch descriptor from the event private data */ |
179 | spin_lock(&event->lock); | |
180 | fsn_priv = fsnotify_remove_priv_from_event(group, event); | |
181 | spin_unlock(&event->lock); | |
182 | ||
183 | if (!fsn_priv) | |
184 | inotify_event.wd = -1; | |
185 | else { | |
186 | priv = container_of(fsn_priv, struct inotify_event_private_data, | |
187 | fsnotify_event_priv_data); | |
188 | inotify_event.wd = priv->wd; | |
189 | inotify_free_event_priv(fsn_priv); | |
190 | } | |
191 | ||
b962e731 BR |
192 | /* |
193 | * round up event->name_len so it is a multiple of event_size | |
0db501bd EB |
194 | * plus an extra byte for the terminating '\0'. |
195 | */ | |
b962e731 BR |
196 | if (event->name_len) |
197 | name_len = roundup(event->name_len + 1, event_size); | |
63c882a0 EP |
198 | inotify_event.len = name_len; |
199 | ||
200 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
201 | inotify_event.cookie = event->sync_cookie; | |
3632dee2 | 202 | |
63c882a0 EP |
203 | /* send the main event */ |
204 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
205 | return -EFAULT; |
206 | ||
63c882a0 | 207 | buf += event_size; |
3632dee2 | 208 | |
63c882a0 EP |
209 | /* |
210 | * fsnotify only stores the pathname, so here we have to send the pathname | |
211 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
212 | * with zeros. I get my zeros from the nul_inotify_event. | |
213 | */ | |
214 | if (name_len) { | |
215 | unsigned int len_to_zero = name_len - event->name_len; | |
216 | /* copy the path name */ | |
217 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
3632dee2 | 218 | return -EFAULT; |
63c882a0 | 219 | buf += event->name_len; |
3632dee2 | 220 | |
0db501bd EB |
221 | /* fill userspace with 0's */ |
222 | if (clear_user(buf, len_to_zero)) | |
63c882a0 EP |
223 | return -EFAULT; |
224 | buf += len_to_zero; | |
225 | event_size += name_len; | |
3632dee2 | 226 | } |
63c882a0 | 227 | |
3632dee2 VN |
228 | return event_size; |
229 | } | |
230 | ||
2d9048e2 AG |
231 | static ssize_t inotify_read(struct file *file, char __user *buf, |
232 | size_t count, loff_t *pos) | |
233 | { | |
63c882a0 EP |
234 | struct fsnotify_group *group; |
235 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
236 | char __user *start; |
237 | int ret; | |
238 | DEFINE_WAIT(wait); | |
239 | ||
240 | start = buf; | |
63c882a0 | 241 | group = file->private_data; |
2d9048e2 AG |
242 | |
243 | while (1) { | |
63c882a0 | 244 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); |
2d9048e2 | 245 | |
63c882a0 EP |
246 | mutex_lock(&group->notification_mutex); |
247 | kevent = get_one_event(group, count); | |
248 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 249 | |
5ba08e2e EP |
250 | pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); |
251 | ||
3632dee2 VN |
252 | if (kevent) { |
253 | ret = PTR_ERR(kevent); | |
254 | if (IS_ERR(kevent)) | |
255 | break; | |
63c882a0 EP |
256 | ret = copy_event_to_user(group, kevent, buf); |
257 | fsnotify_put_event(kevent); | |
3632dee2 VN |
258 | if (ret < 0) |
259 | break; | |
260 | buf += ret; | |
261 | count -= ret; | |
262 | continue; | |
2d9048e2 AG |
263 | } |
264 | ||
3632dee2 VN |
265 | ret = -EAGAIN; |
266 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 267 | break; |
1ca39ab9 | 268 | ret = -ERESTARTSYS; |
3632dee2 | 269 | if (signal_pending(current)) |
2d9048e2 | 270 | break; |
16dbc6c9 | 271 | |
3632dee2 | 272 | if (start != buf) |
2d9048e2 | 273 | break; |
16dbc6c9 | 274 | |
3632dee2 | 275 | schedule(); |
2d9048e2 | 276 | } |
2d9048e2 | 277 | |
63c882a0 | 278 | finish_wait(&group->notification_waitq, &wait); |
3632dee2 VN |
279 | if (start != buf && ret != -EFAULT) |
280 | ret = buf - start; | |
2d9048e2 AG |
281 | return ret; |
282 | } | |
283 | ||
284 | static int inotify_release(struct inode *ignored, struct file *file) | |
285 | { | |
63c882a0 | 286 | struct fsnotify_group *group = file->private_data; |
2d9048e2 | 287 | |
5ba08e2e EP |
288 | pr_debug("%s: group=%p\n", __func__, group); |
289 | ||
0a6b6bd5 EP |
290 | if (file->f_flags & FASYNC) |
291 | fsnotify_fasync(-1, file, 0); | |
2d9048e2 | 292 | |
63c882a0 | 293 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
d8153d4d | 294 | fsnotify_destroy_group(group); |
2d9048e2 AG |
295 | |
296 | return 0; | |
297 | } | |
298 | ||
299 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
300 | unsigned long arg) | |
301 | { | |
63c882a0 EP |
302 | struct fsnotify_group *group; |
303 | struct fsnotify_event_holder *holder; | |
304 | struct fsnotify_event *event; | |
2d9048e2 AG |
305 | void __user *p; |
306 | int ret = -ENOTTY; | |
63c882a0 | 307 | size_t send_len = 0; |
2d9048e2 | 308 | |
63c882a0 | 309 | group = file->private_data; |
2d9048e2 AG |
310 | p = (void __user *) arg; |
311 | ||
5ba08e2e EP |
312 | pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); |
313 | ||
2d9048e2 AG |
314 | switch (cmd) { |
315 | case FIONREAD: | |
63c882a0 EP |
316 | mutex_lock(&group->notification_mutex); |
317 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
318 | event = holder->event; | |
319 | send_len += sizeof(struct inotify_event); | |
83cb10f0 EP |
320 | if (event->name_len) |
321 | send_len += roundup(event->name_len + 1, | |
322 | sizeof(struct inotify_event)); | |
63c882a0 EP |
323 | } |
324 | mutex_unlock(&group->notification_mutex); | |
325 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
326 | break; |
327 | } | |
328 | ||
329 | return ret; | |
330 | } | |
331 | ||
332 | static const struct file_operations inotify_fops = { | |
be77196b | 333 | .show_fdinfo = inotify_show_fdinfo, |
63c882a0 EP |
334 | .poll = inotify_poll, |
335 | .read = inotify_read, | |
0a6b6bd5 | 336 | .fasync = fsnotify_fasync, |
63c882a0 EP |
337 | .release = inotify_release, |
338 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 | 339 | .compat_ioctl = inotify_ioctl, |
6038f373 | 340 | .llseek = noop_llseek, |
2d9048e2 AG |
341 | }; |
342 | ||
2d9048e2 | 343 | |
63c882a0 EP |
344 | /* |
345 | * find_inode - resolve a user-given path to a specific inode | |
346 | */ | |
347 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
348 | { | |
349 | int error; | |
350 | ||
351 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
352 | if (error) | |
353 | return error; | |
354 | /* you can only watch an inode if you have read permissions on it */ | |
355 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
356 | if (error) | |
357 | path_put(path); | |
358 | return error; | |
359 | } | |
360 | ||
b7ba8371 | 361 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
000285de | 362 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
363 | { |
364 | int ret; | |
365 | ||
4542da63 TH |
366 | idr_preload(GFP_KERNEL); |
367 | spin_lock(idr_lock); | |
b7ba8371 | 368 | |
a66c04b4 | 369 | ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT); |
4542da63 | 370 | if (ret >= 0) { |
b7ba8371 | 371 | /* we added the mark to the idr, take a reference */ |
4542da63 | 372 | i_mark->wd = ret; |
4542da63 TH |
373 | fsnotify_get_mark(&i_mark->fsn_mark); |
374 | } | |
b7ba8371 | 375 | |
4542da63 TH |
376 | spin_unlock(idr_lock); |
377 | idr_preload_end(); | |
378 | return ret < 0 ? ret : 0; | |
b7ba8371 EP |
379 | } |
380 | ||
000285de | 381 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, |
b7ba8371 EP |
382 | int wd) |
383 | { | |
384 | struct idr *idr = &group->inotify_data.idr; | |
385 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 386 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
387 | |
388 | assert_spin_locked(idr_lock); | |
389 | ||
000285de EP |
390 | i_mark = idr_find(idr, wd); |
391 | if (i_mark) { | |
392 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
b7ba8371 | 393 | |
000285de | 394 | fsnotify_get_mark(fsn_mark); |
b7ba8371 | 395 | /* One ref for being in the idr, one ref we just took */ |
000285de | 396 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); |
b7ba8371 EP |
397 | } |
398 | ||
000285de | 399 | return i_mark; |
b7ba8371 EP |
400 | } |
401 | ||
000285de | 402 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, |
b7ba8371 EP |
403 | int wd) |
404 | { | |
000285de | 405 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
406 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
407 | ||
408 | spin_lock(idr_lock); | |
000285de | 409 | i_mark = inotify_idr_find_locked(group, wd); |
b7ba8371 EP |
410 | spin_unlock(idr_lock); |
411 | ||
000285de | 412 | return i_mark; |
b7ba8371 EP |
413 | } |
414 | ||
415 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
000285de | 416 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
417 | { |
418 | struct idr *idr = &group->inotify_data.idr; | |
419 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 420 | int wd = i_mark->wd; |
b7ba8371 EP |
421 | |
422 | assert_spin_locked(idr_lock); | |
423 | ||
424 | idr_remove(idr, wd); | |
425 | ||
426 | /* removed from the idr, drop that ref */ | |
000285de | 427 | fsnotify_put_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
428 | } |
429 | ||
dead537d EP |
430 | /* |
431 | * Remove the mark from the idr (if present) and drop the reference | |
432 | * on the mark because it was in the idr. | |
433 | */ | |
7e790dd5 | 434 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
000285de | 435 | struct inotify_inode_mark *i_mark) |
7e790dd5 | 436 | { |
b7ba8371 | 437 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
000285de | 438 | struct inotify_inode_mark *found_i_mark = NULL; |
dead537d | 439 | int wd; |
7e790dd5 | 440 | |
b7ba8371 | 441 | spin_lock(idr_lock); |
000285de | 442 | wd = i_mark->wd; |
dead537d | 443 | |
b7ba8371 | 444 | /* |
000285de | 445 | * does this i_mark think it is in the idr? we shouldn't get called |
b7ba8371 EP |
446 | * if it wasn't.... |
447 | */ | |
448 | if (wd == -1) { | |
000285de EP |
449 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" |
450 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
451 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 452 | goto out; |
b7ba8371 | 453 | } |
dead537d | 454 | |
b7ba8371 | 455 | /* Lets look in the idr to see if we find it */ |
000285de EP |
456 | found_i_mark = inotify_idr_find_locked(group, wd); |
457 | if (unlikely(!found_i_mark)) { | |
458 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
459 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
460 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 461 | goto out; |
b7ba8371 | 462 | } |
dead537d | 463 | |
b7ba8371 | 464 | /* |
000285de EP |
465 | * We found an mark in the idr at the right wd, but it's |
466 | * not the mark we were told to remove. eparis seriously | |
b7ba8371 EP |
467 | * fucked up somewhere. |
468 | */ | |
000285de EP |
469 | if (unlikely(found_i_mark != i_mark)) { |
470 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
471 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
472 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
473 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
474 | i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, | |
475 | found_i_mark->fsn_mark.group, | |
476 | found_i_mark->fsn_mark.i.inode); | |
dead537d EP |
477 | goto out; |
478 | } | |
479 | ||
b7ba8371 EP |
480 | /* |
481 | * One ref for being in the idr | |
482 | * one ref held by the caller trying to kill us | |
483 | * one ref grabbed by inotify_idr_find | |
484 | */ | |
000285de EP |
485 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { |
486 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
487 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
488 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
b7ba8371 EP |
489 | /* we can't really recover with bad ref cnting.. */ |
490 | BUG(); | |
491 | } | |
dead537d | 492 | |
000285de | 493 | do_inotify_remove_from_idr(group, i_mark); |
dead537d | 494 | out: |
b7ba8371 | 495 | /* match the ref taken by inotify_idr_find_locked() */ |
000285de EP |
496 | if (found_i_mark) |
497 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
498 | i_mark->wd = -1; | |
b7ba8371 | 499 | spin_unlock(idr_lock); |
7e790dd5 | 500 | } |
dead537d | 501 | |
63c882a0 | 502 | /* |
dead537d | 503 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 504 | */ |
000285de | 505 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, |
528da3e9 | 506 | struct fsnotify_group *group) |
63c882a0 | 507 | { |
000285de | 508 | struct inotify_inode_mark *i_mark; |
f70ab54c | 509 | struct fsnotify_event *ignored_event, *notify_event; |
63c882a0 EP |
510 | struct inotify_event_private_data *event_priv; |
511 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 512 | int ret; |
63c882a0 | 513 | |
8b99c3cc LS |
514 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
515 | ||
f44aebcc EP |
516 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
517 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
518 | GFP_NOFS); | |
519 | if (!ignored_event) | |
8b99c3cc | 520 | goto skip_send_ignore; |
63c882a0 | 521 | |
f44aebcc | 522 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
523 | if (unlikely(!event_priv)) |
524 | goto skip_send_ignore; | |
525 | ||
526 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
527 | ||
23e964c2 | 528 | fsnotify_get_group(group); |
63c882a0 | 529 | fsn_event_priv->group = group; |
000285de | 530 | event_priv->wd = i_mark->wd; |
63c882a0 | 531 | |
f70ab54c EP |
532 | notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); |
533 | if (notify_event) { | |
534 | if (IS_ERR(notify_event)) | |
535 | ret = PTR_ERR(notify_event); | |
536 | else | |
537 | fsnotify_put_event(notify_event); | |
63c882a0 | 538 | inotify_free_event_priv(fsn_event_priv); |
f70ab54c | 539 | } |
63c882a0 EP |
540 | |
541 | skip_send_ignore: | |
f44aebcc | 542 | /* matches the reference taken when the event was created */ |
8b99c3cc LS |
543 | if (ignored_event) |
544 | fsnotify_put_event(ignored_event); | |
f44aebcc | 545 | |
000285de EP |
546 | /* remove this mark from the idr */ |
547 | inotify_remove_from_idr(group, i_mark); | |
63c882a0 | 548 | |
5549f7cd | 549 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
550 | } |
551 | ||
552 | /* ding dong the mark is dead */ | |
000285de | 553 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) |
63c882a0 | 554 | { |
000285de | 555 | struct inotify_inode_mark *i_mark; |
31ddd326 | 556 | |
000285de | 557 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 558 | |
000285de | 559 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); |
63c882a0 EP |
560 | } |
561 | ||
52cef755 EP |
562 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
563 | struct inode *inode, | |
564 | u32 arg) | |
63c882a0 | 565 | { |
000285de EP |
566 | struct fsnotify_mark *fsn_mark; |
567 | struct inotify_inode_mark *i_mark; | |
63c882a0 | 568 | __u32 old_mask, new_mask; |
52cef755 EP |
569 | __u32 mask; |
570 | int add = (arg & IN_MASK_ADD); | |
571 | int ret; | |
63c882a0 EP |
572 | |
573 | /* don't allow invalid bits: we don't want flags set */ | |
574 | mask = inotify_arg_to_mask(arg); | |
63c882a0 | 575 | |
5444e298 | 576 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
000285de | 577 | if (!fsn_mark) |
52cef755 | 578 | return -ENOENT; |
7e790dd5 | 579 | |
000285de | 580 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
75fe2b26 | 581 | |
000285de | 582 | spin_lock(&fsn_mark->lock); |
63c882a0 | 583 | |
000285de | 584 | old_mask = fsn_mark->mask; |
90b1e7a5 EP |
585 | if (add) |
586 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
587 | else | |
588 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
589 | new_mask = fsn_mark->mask; | |
63c882a0 | 590 | |
000285de | 591 | spin_unlock(&fsn_mark->lock); |
63c882a0 EP |
592 | |
593 | if (old_mask != new_mask) { | |
594 | /* more bits in old than in new? */ | |
595 | int dropped = (old_mask & ~new_mask); | |
000285de | 596 | /* more bits in this fsn_mark than the inode's mask? */ |
63c882a0 | 597 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); |
63c882a0 | 598 | |
000285de | 599 | /* update the inode with this new fsn_mark */ |
63c882a0 EP |
600 | if (dropped || do_inode) |
601 | fsnotify_recalc_inode_mask(inode); | |
602 | ||
63c882a0 EP |
603 | } |
604 | ||
52cef755 | 605 | /* return the wd */ |
000285de | 606 | ret = i_mark->wd; |
52cef755 | 607 | |
d0775441 | 608 | /* match the get from fsnotify_find_mark() */ |
000285de | 609 | fsnotify_put_mark(fsn_mark); |
75fe2b26 | 610 | |
52cef755 EP |
611 | return ret; |
612 | } | |
613 | ||
614 | static int inotify_new_watch(struct fsnotify_group *group, | |
615 | struct inode *inode, | |
616 | u32 arg) | |
617 | { | |
000285de | 618 | struct inotify_inode_mark *tmp_i_mark; |
52cef755 EP |
619 | __u32 mask; |
620 | int ret; | |
b7ba8371 EP |
621 | struct idr *idr = &group->inotify_data.idr; |
622 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 EP |
623 | |
624 | /* don't allow invalid bits: we don't want flags set */ | |
625 | mask = inotify_arg_to_mask(arg); | |
52cef755 | 626 | |
000285de EP |
627 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); |
628 | if (unlikely(!tmp_i_mark)) | |
52cef755 EP |
629 | return -ENOMEM; |
630 | ||
000285de EP |
631 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); |
632 | tmp_i_mark->fsn_mark.mask = mask; | |
633 | tmp_i_mark->wd = -1; | |
52cef755 EP |
634 | |
635 | ret = -ENOSPC; | |
636 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
637 | goto out_err; | |
e0873344 | 638 | |
a66c04b4 | 639 | ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark); |
b7ba8371 | 640 | if (ret) |
52cef755 | 641 | goto out_err; |
52cef755 EP |
642 | |
643 | /* we are on the idr, now get on the inode */ | |
5444e298 | 644 | ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0); |
52cef755 EP |
645 | if (ret) { |
646 | /* we failed to get on the inode, get off the idr */ | |
000285de | 647 | inotify_remove_from_idr(group, tmp_i_mark); |
52cef755 EP |
648 | goto out_err; |
649 | } | |
650 | ||
52cef755 EP |
651 | /* increment the number of watches the user has */ |
652 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
653 | ||
000285de EP |
654 | /* return the watch descriptor for this new mark */ |
655 | ret = tmp_i_mark->wd; | |
52cef755 | 656 | |
63c882a0 | 657 | out_err: |
000285de EP |
658 | /* match the ref from fsnotify_init_mark() */ |
659 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
52cef755 EP |
660 | |
661 | return ret; | |
662 | } | |
663 | ||
664 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
665 | { | |
666 | int ret = 0; | |
667 | ||
668 | retry: | |
669 | /* try to update and existing watch with the new arg */ | |
670 | ret = inotify_update_existing_watch(group, inode, arg); | |
671 | /* no mark present, try to add a new one */ | |
672 | if (ret == -ENOENT) | |
673 | ret = inotify_new_watch(group, inode, arg); | |
674 | /* | |
675 | * inotify_new_watch could race with another thread which did an | |
676 | * inotify_new_watch between the update_existing and the add watch | |
677 | * here, go back and try to update an existing mark again. | |
678 | */ | |
679 | if (ret == -EEXIST) | |
680 | goto retry; | |
7e790dd5 | 681 | |
63c882a0 EP |
682 | return ret; |
683 | } | |
684 | ||
d0de4dc5 | 685 | static struct fsnotify_group *inotify_new_group(unsigned int max_events) |
63c882a0 EP |
686 | { |
687 | struct fsnotify_group *group; | |
63c882a0 | 688 | |
0d2e2a1d | 689 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
690 | if (IS_ERR(group)) |
691 | return group; | |
692 | ||
693 | group->max_events = max_events; | |
694 | ||
695 | spin_lock_init(&group->inotify_data.idr_lock); | |
696 | idr_init(&group->inotify_data.idr); | |
d0de4dc5 EP |
697 | group->inotify_data.user = get_current_user(); |
698 | ||
699 | if (atomic_inc_return(&group->inotify_data.user->inotify_devs) > | |
700 | inotify_max_user_instances) { | |
d8153d4d | 701 | fsnotify_destroy_group(group); |
d0de4dc5 EP |
702 | return ERR_PTR(-EMFILE); |
703 | } | |
63c882a0 EP |
704 | |
705 | return group; | |
706 | } | |
707 | ||
708 | ||
709 | /* inotify syscalls */ | |
938bb9f5 | 710 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 711 | { |
63c882a0 | 712 | struct fsnotify_group *group; |
c44dcc56 | 713 | int ret; |
2d9048e2 | 714 | |
e38b36f3 UD |
715 | /* Check the IN_* constants for consistency. */ |
716 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
717 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
718 | ||
510df2dd | 719 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
720 | return -EINVAL; |
721 | ||
63c882a0 | 722 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
d0de4dc5 EP |
723 | group = inotify_new_group(inotify_max_queued_events); |
724 | if (IS_ERR(group)) | |
725 | return PTR_ERR(group); | |
825f9692 | 726 | |
c44dcc56 AV |
727 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
728 | O_RDONLY | flags); | |
d0de4dc5 | 729 | if (ret < 0) |
d8153d4d | 730 | fsnotify_destroy_group(group); |
825f9692 | 731 | |
2d9048e2 AG |
732 | return ret; |
733 | } | |
734 | ||
938bb9f5 | 735 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
736 | { |
737 | return sys_inotify_init1(0); | |
738 | } | |
739 | ||
2e4d0924 HC |
740 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
741 | u32, mask) | |
2d9048e2 | 742 | { |
63c882a0 | 743 | struct fsnotify_group *group; |
2d9048e2 | 744 | struct inode *inode; |
2d8f3038 | 745 | struct path path; |
2903ff01 AV |
746 | struct fd f; |
747 | int ret; | |
2d9048e2 AG |
748 | unsigned flags = 0; |
749 | ||
2903ff01 AV |
750 | f = fdget(fd); |
751 | if (unlikely(!f.file)) | |
2d9048e2 AG |
752 | return -EBADF; |
753 | ||
754 | /* verify that this is indeed an inotify instance */ | |
2903ff01 | 755 | if (unlikely(f.file->f_op != &inotify_fops)) { |
2d9048e2 AG |
756 | ret = -EINVAL; |
757 | goto fput_and_out; | |
758 | } | |
759 | ||
760 | if (!(mask & IN_DONT_FOLLOW)) | |
761 | flags |= LOOKUP_FOLLOW; | |
762 | if (mask & IN_ONLYDIR) | |
763 | flags |= LOOKUP_DIRECTORY; | |
764 | ||
63c882a0 EP |
765 | ret = inotify_find_inode(pathname, &path, flags); |
766 | if (ret) | |
2d9048e2 AG |
767 | goto fput_and_out; |
768 | ||
63c882a0 | 769 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 770 | inode = path.dentry->d_inode; |
2903ff01 | 771 | group = f.file->private_data; |
2d9048e2 | 772 | |
63c882a0 EP |
773 | /* create/update an inode mark */ |
774 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 775 | path_put(&path); |
2d9048e2 | 776 | fput_and_out: |
2903ff01 | 777 | fdput(f); |
2d9048e2 AG |
778 | return ret; |
779 | } | |
780 | ||
2e4d0924 | 781 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 782 | { |
63c882a0 | 783 | struct fsnotify_group *group; |
000285de | 784 | struct inotify_inode_mark *i_mark; |
2903ff01 AV |
785 | struct fd f; |
786 | int ret = 0; | |
2d9048e2 | 787 | |
2903ff01 AV |
788 | f = fdget(fd); |
789 | if (unlikely(!f.file)) | |
2d9048e2 AG |
790 | return -EBADF; |
791 | ||
792 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 | 793 | ret = -EINVAL; |
2903ff01 | 794 | if (unlikely(f.file->f_op != &inotify_fops)) |
2d9048e2 | 795 | goto out; |
2d9048e2 | 796 | |
2903ff01 | 797 | group = f.file->private_data; |
2d9048e2 | 798 | |
b7ba8371 | 799 | ret = -EINVAL; |
000285de EP |
800 | i_mark = inotify_idr_find(group, wd); |
801 | if (unlikely(!i_mark)) | |
63c882a0 | 802 | goto out; |
63c882a0 | 803 | |
b7ba8371 EP |
804 | ret = 0; |
805 | ||
e2a29943 | 806 | fsnotify_destroy_mark(&i_mark->fsn_mark, group); |
b7ba8371 EP |
807 | |
808 | /* match ref taken by inotify_idr_find */ | |
000285de | 809 | fsnotify_put_mark(&i_mark->fsn_mark); |
2d9048e2 AG |
810 | |
811 | out: | |
2903ff01 | 812 | fdput(f); |
2d9048e2 AG |
813 | return ret; |
814 | } | |
815 | ||
2d9048e2 | 816 | /* |
ae0e47f0 | 817 | * inotify_user_setup - Our initialization function. Note that we cannot return |
2d9048e2 AG |
818 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
819 | * must result in panic(). | |
820 | */ | |
821 | static int __init inotify_user_setup(void) | |
822 | { | |
f874e1ac EP |
823 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); |
824 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
825 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
826 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
827 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
828 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
829 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
830 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
831 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
832 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
833 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
834 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
835 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
836 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
837 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
838 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
b29866aa | 839 | BUILD_BUG_ON(IN_ISDIR != FS_ISDIR); |
f874e1ac EP |
840 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); |
841 | ||
842 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
843 | ||
000285de | 844 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); |
63c882a0 | 845 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); |
63c882a0 | 846 | |
2d9048e2 AG |
847 | inotify_max_queued_events = 16384; |
848 | inotify_max_user_instances = 128; | |
849 | inotify_max_user_watches = 8192; | |
850 | ||
2d9048e2 AG |
851 | return 0; |
852 | } | |
2d9048e2 | 853 | module_init(inotify_user_setup); |