]>
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, |
7050c488 | 362 | int *last_wd, |
000285de | 363 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
364 | { |
365 | int ret; | |
366 | ||
367 | do { | |
368 | if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) | |
369 | return -ENOMEM; | |
370 | ||
371 | spin_lock(idr_lock); | |
000285de EP |
372 | ret = idr_get_new_above(idr, i_mark, *last_wd + 1, |
373 | &i_mark->wd); | |
b7ba8371 | 374 | /* we added the mark to the idr, take a reference */ |
7050c488 | 375 | if (!ret) { |
000285de EP |
376 | *last_wd = i_mark->wd; |
377 | fsnotify_get_mark(&i_mark->fsn_mark); | |
7050c488 | 378 | } |
b7ba8371 EP |
379 | spin_unlock(idr_lock); |
380 | } while (ret == -EAGAIN); | |
381 | ||
382 | return ret; | |
383 | } | |
384 | ||
000285de | 385 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, |
b7ba8371 EP |
386 | int wd) |
387 | { | |
388 | struct idr *idr = &group->inotify_data.idr; | |
389 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 390 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
391 | |
392 | assert_spin_locked(idr_lock); | |
393 | ||
000285de EP |
394 | i_mark = idr_find(idr, wd); |
395 | if (i_mark) { | |
396 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
b7ba8371 | 397 | |
000285de | 398 | fsnotify_get_mark(fsn_mark); |
b7ba8371 | 399 | /* One ref for being in the idr, one ref we just took */ |
000285de | 400 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); |
b7ba8371 EP |
401 | } |
402 | ||
000285de | 403 | return i_mark; |
b7ba8371 EP |
404 | } |
405 | ||
000285de | 406 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, |
b7ba8371 EP |
407 | int wd) |
408 | { | |
000285de | 409 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
410 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
411 | ||
412 | spin_lock(idr_lock); | |
000285de | 413 | i_mark = inotify_idr_find_locked(group, wd); |
b7ba8371 EP |
414 | spin_unlock(idr_lock); |
415 | ||
000285de | 416 | return i_mark; |
b7ba8371 EP |
417 | } |
418 | ||
419 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
000285de | 420 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
421 | { |
422 | struct idr *idr = &group->inotify_data.idr; | |
423 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 424 | int wd = i_mark->wd; |
b7ba8371 EP |
425 | |
426 | assert_spin_locked(idr_lock); | |
427 | ||
428 | idr_remove(idr, wd); | |
429 | ||
430 | /* removed from the idr, drop that ref */ | |
000285de | 431 | fsnotify_put_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
432 | } |
433 | ||
dead537d EP |
434 | /* |
435 | * Remove the mark from the idr (if present) and drop the reference | |
436 | * on the mark because it was in the idr. | |
437 | */ | |
7e790dd5 | 438 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
000285de | 439 | struct inotify_inode_mark *i_mark) |
7e790dd5 | 440 | { |
b7ba8371 | 441 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
000285de | 442 | struct inotify_inode_mark *found_i_mark = NULL; |
dead537d | 443 | int wd; |
7e790dd5 | 444 | |
b7ba8371 | 445 | spin_lock(idr_lock); |
000285de | 446 | wd = i_mark->wd; |
dead537d | 447 | |
b7ba8371 | 448 | /* |
000285de | 449 | * does this i_mark think it is in the idr? we shouldn't get called |
b7ba8371 EP |
450 | * if it wasn't.... |
451 | */ | |
452 | if (wd == -1) { | |
000285de EP |
453 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" |
454 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
455 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 456 | goto out; |
b7ba8371 | 457 | } |
dead537d | 458 | |
b7ba8371 | 459 | /* Lets look in the idr to see if we find it */ |
000285de EP |
460 | found_i_mark = inotify_idr_find_locked(group, wd); |
461 | if (unlikely(!found_i_mark)) { | |
462 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
463 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
464 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 465 | goto out; |
b7ba8371 | 466 | } |
dead537d | 467 | |
b7ba8371 | 468 | /* |
000285de EP |
469 | * We found an mark in the idr at the right wd, but it's |
470 | * not the mark we were told to remove. eparis seriously | |
b7ba8371 EP |
471 | * fucked up somewhere. |
472 | */ | |
000285de EP |
473 | if (unlikely(found_i_mark != i_mark)) { |
474 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
475 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
476 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
477 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
478 | i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, | |
479 | found_i_mark->fsn_mark.group, | |
480 | found_i_mark->fsn_mark.i.inode); | |
dead537d EP |
481 | goto out; |
482 | } | |
483 | ||
b7ba8371 EP |
484 | /* |
485 | * One ref for being in the idr | |
486 | * one ref held by the caller trying to kill us | |
487 | * one ref grabbed by inotify_idr_find | |
488 | */ | |
000285de EP |
489 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { |
490 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
491 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
492 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
b7ba8371 EP |
493 | /* we can't really recover with bad ref cnting.. */ |
494 | BUG(); | |
495 | } | |
dead537d | 496 | |
000285de | 497 | do_inotify_remove_from_idr(group, i_mark); |
dead537d | 498 | out: |
b7ba8371 | 499 | /* match the ref taken by inotify_idr_find_locked() */ |
000285de EP |
500 | if (found_i_mark) |
501 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
502 | i_mark->wd = -1; | |
b7ba8371 | 503 | spin_unlock(idr_lock); |
7e790dd5 | 504 | } |
dead537d | 505 | |
63c882a0 | 506 | /* |
dead537d | 507 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 508 | */ |
000285de | 509 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, |
528da3e9 | 510 | struct fsnotify_group *group) |
63c882a0 | 511 | { |
000285de | 512 | struct inotify_inode_mark *i_mark; |
f70ab54c | 513 | struct fsnotify_event *ignored_event, *notify_event; |
63c882a0 EP |
514 | struct inotify_event_private_data *event_priv; |
515 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 516 | int ret; |
63c882a0 | 517 | |
8b99c3cc LS |
518 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
519 | ||
f44aebcc EP |
520 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
521 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
522 | GFP_NOFS); | |
523 | if (!ignored_event) | |
8b99c3cc | 524 | goto skip_send_ignore; |
63c882a0 | 525 | |
f44aebcc | 526 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
527 | if (unlikely(!event_priv)) |
528 | goto skip_send_ignore; | |
529 | ||
530 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
531 | ||
23e964c2 | 532 | fsnotify_get_group(group); |
63c882a0 | 533 | fsn_event_priv->group = group; |
000285de | 534 | event_priv->wd = i_mark->wd; |
63c882a0 | 535 | |
f70ab54c EP |
536 | notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); |
537 | if (notify_event) { | |
538 | if (IS_ERR(notify_event)) | |
539 | ret = PTR_ERR(notify_event); | |
540 | else | |
541 | fsnotify_put_event(notify_event); | |
63c882a0 | 542 | inotify_free_event_priv(fsn_event_priv); |
f70ab54c | 543 | } |
63c882a0 EP |
544 | |
545 | skip_send_ignore: | |
f44aebcc | 546 | /* matches the reference taken when the event was created */ |
8b99c3cc LS |
547 | if (ignored_event) |
548 | fsnotify_put_event(ignored_event); | |
f44aebcc | 549 | |
000285de EP |
550 | /* remove this mark from the idr */ |
551 | inotify_remove_from_idr(group, i_mark); | |
63c882a0 | 552 | |
5549f7cd | 553 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
554 | } |
555 | ||
556 | /* ding dong the mark is dead */ | |
000285de | 557 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) |
63c882a0 | 558 | { |
000285de | 559 | struct inotify_inode_mark *i_mark; |
31ddd326 | 560 | |
000285de | 561 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 562 | |
000285de | 563 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); |
63c882a0 EP |
564 | } |
565 | ||
52cef755 EP |
566 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
567 | struct inode *inode, | |
568 | u32 arg) | |
63c882a0 | 569 | { |
000285de EP |
570 | struct fsnotify_mark *fsn_mark; |
571 | struct inotify_inode_mark *i_mark; | |
63c882a0 | 572 | __u32 old_mask, new_mask; |
52cef755 EP |
573 | __u32 mask; |
574 | int add = (arg & IN_MASK_ADD); | |
575 | int ret; | |
63c882a0 EP |
576 | |
577 | /* don't allow invalid bits: we don't want flags set */ | |
578 | mask = inotify_arg_to_mask(arg); | |
63c882a0 | 579 | |
5444e298 | 580 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
000285de | 581 | if (!fsn_mark) |
52cef755 | 582 | return -ENOENT; |
7e790dd5 | 583 | |
000285de | 584 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
75fe2b26 | 585 | |
000285de | 586 | spin_lock(&fsn_mark->lock); |
63c882a0 | 587 | |
000285de | 588 | old_mask = fsn_mark->mask; |
90b1e7a5 EP |
589 | if (add) |
590 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
591 | else | |
592 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
593 | new_mask = fsn_mark->mask; | |
63c882a0 | 594 | |
000285de | 595 | spin_unlock(&fsn_mark->lock); |
63c882a0 EP |
596 | |
597 | if (old_mask != new_mask) { | |
598 | /* more bits in old than in new? */ | |
599 | int dropped = (old_mask & ~new_mask); | |
000285de | 600 | /* more bits in this fsn_mark than the inode's mask? */ |
63c882a0 | 601 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); |
63c882a0 | 602 | |
000285de | 603 | /* update the inode with this new fsn_mark */ |
63c882a0 EP |
604 | if (dropped || do_inode) |
605 | fsnotify_recalc_inode_mask(inode); | |
606 | ||
63c882a0 EP |
607 | } |
608 | ||
52cef755 | 609 | /* return the wd */ |
000285de | 610 | ret = i_mark->wd; |
52cef755 | 611 | |
d0775441 | 612 | /* match the get from fsnotify_find_mark() */ |
000285de | 613 | fsnotify_put_mark(fsn_mark); |
75fe2b26 | 614 | |
52cef755 EP |
615 | return ret; |
616 | } | |
617 | ||
618 | static int inotify_new_watch(struct fsnotify_group *group, | |
619 | struct inode *inode, | |
620 | u32 arg) | |
621 | { | |
000285de | 622 | struct inotify_inode_mark *tmp_i_mark; |
52cef755 EP |
623 | __u32 mask; |
624 | int ret; | |
b7ba8371 EP |
625 | struct idr *idr = &group->inotify_data.idr; |
626 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 EP |
627 | |
628 | /* don't allow invalid bits: we don't want flags set */ | |
629 | mask = inotify_arg_to_mask(arg); | |
52cef755 | 630 | |
000285de EP |
631 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); |
632 | if (unlikely(!tmp_i_mark)) | |
52cef755 EP |
633 | return -ENOMEM; |
634 | ||
000285de EP |
635 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); |
636 | tmp_i_mark->fsn_mark.mask = mask; | |
637 | tmp_i_mark->wd = -1; | |
52cef755 EP |
638 | |
639 | ret = -ENOSPC; | |
640 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
641 | goto out_err; | |
e0873344 | 642 | |
7050c488 | 643 | ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd, |
000285de | 644 | tmp_i_mark); |
b7ba8371 | 645 | if (ret) |
52cef755 | 646 | goto out_err; |
52cef755 EP |
647 | |
648 | /* we are on the idr, now get on the inode */ | |
5444e298 | 649 | ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0); |
52cef755 EP |
650 | if (ret) { |
651 | /* we failed to get on the inode, get off the idr */ | |
000285de | 652 | inotify_remove_from_idr(group, tmp_i_mark); |
52cef755 EP |
653 | goto out_err; |
654 | } | |
655 | ||
52cef755 EP |
656 | /* increment the number of watches the user has */ |
657 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
658 | ||
000285de EP |
659 | /* return the watch descriptor for this new mark */ |
660 | ret = tmp_i_mark->wd; | |
52cef755 | 661 | |
63c882a0 | 662 | out_err: |
000285de EP |
663 | /* match the ref from fsnotify_init_mark() */ |
664 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
52cef755 EP |
665 | |
666 | return ret; | |
667 | } | |
668 | ||
669 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
670 | { | |
671 | int ret = 0; | |
672 | ||
673 | retry: | |
674 | /* try to update and existing watch with the new arg */ | |
675 | ret = inotify_update_existing_watch(group, inode, arg); | |
676 | /* no mark present, try to add a new one */ | |
677 | if (ret == -ENOENT) | |
678 | ret = inotify_new_watch(group, inode, arg); | |
679 | /* | |
680 | * inotify_new_watch could race with another thread which did an | |
681 | * inotify_new_watch between the update_existing and the add watch | |
682 | * here, go back and try to update an existing mark again. | |
683 | */ | |
684 | if (ret == -EEXIST) | |
685 | goto retry; | |
7e790dd5 | 686 | |
63c882a0 EP |
687 | return ret; |
688 | } | |
689 | ||
d0de4dc5 | 690 | static struct fsnotify_group *inotify_new_group(unsigned int max_events) |
63c882a0 EP |
691 | { |
692 | struct fsnotify_group *group; | |
63c882a0 | 693 | |
0d2e2a1d | 694 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
695 | if (IS_ERR(group)) |
696 | return group; | |
697 | ||
698 | group->max_events = max_events; | |
699 | ||
700 | spin_lock_init(&group->inotify_data.idr_lock); | |
701 | idr_init(&group->inotify_data.idr); | |
9e572cc9 | 702 | group->inotify_data.last_wd = 0; |
d0de4dc5 EP |
703 | group->inotify_data.user = get_current_user(); |
704 | ||
705 | if (atomic_inc_return(&group->inotify_data.user->inotify_devs) > | |
706 | inotify_max_user_instances) { | |
d8153d4d | 707 | fsnotify_destroy_group(group); |
d0de4dc5 EP |
708 | return ERR_PTR(-EMFILE); |
709 | } | |
63c882a0 EP |
710 | |
711 | return group; | |
712 | } | |
713 | ||
714 | ||
715 | /* inotify syscalls */ | |
938bb9f5 | 716 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 717 | { |
63c882a0 | 718 | struct fsnotify_group *group; |
c44dcc56 | 719 | int ret; |
2d9048e2 | 720 | |
e38b36f3 UD |
721 | /* Check the IN_* constants for consistency. */ |
722 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
723 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
724 | ||
510df2dd | 725 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
726 | return -EINVAL; |
727 | ||
63c882a0 | 728 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
d0de4dc5 EP |
729 | group = inotify_new_group(inotify_max_queued_events); |
730 | if (IS_ERR(group)) | |
731 | return PTR_ERR(group); | |
825f9692 | 732 | |
c44dcc56 AV |
733 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
734 | O_RDONLY | flags); | |
d0de4dc5 | 735 | if (ret < 0) |
d8153d4d | 736 | fsnotify_destroy_group(group); |
825f9692 | 737 | |
2d9048e2 AG |
738 | return ret; |
739 | } | |
740 | ||
938bb9f5 | 741 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
742 | { |
743 | return sys_inotify_init1(0); | |
744 | } | |
745 | ||
2e4d0924 HC |
746 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
747 | u32, mask) | |
2d9048e2 | 748 | { |
63c882a0 | 749 | struct fsnotify_group *group; |
2d9048e2 | 750 | struct inode *inode; |
2d8f3038 | 751 | struct path path; |
2903ff01 AV |
752 | struct fd f; |
753 | int ret; | |
2d9048e2 AG |
754 | unsigned flags = 0; |
755 | ||
2903ff01 AV |
756 | f = fdget(fd); |
757 | if (unlikely(!f.file)) | |
2d9048e2 AG |
758 | return -EBADF; |
759 | ||
760 | /* verify that this is indeed an inotify instance */ | |
2903ff01 | 761 | if (unlikely(f.file->f_op != &inotify_fops)) { |
2d9048e2 AG |
762 | ret = -EINVAL; |
763 | goto fput_and_out; | |
764 | } | |
765 | ||
766 | if (!(mask & IN_DONT_FOLLOW)) | |
767 | flags |= LOOKUP_FOLLOW; | |
768 | if (mask & IN_ONLYDIR) | |
769 | flags |= LOOKUP_DIRECTORY; | |
770 | ||
63c882a0 EP |
771 | ret = inotify_find_inode(pathname, &path, flags); |
772 | if (ret) | |
2d9048e2 AG |
773 | goto fput_and_out; |
774 | ||
63c882a0 | 775 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 776 | inode = path.dentry->d_inode; |
2903ff01 | 777 | group = f.file->private_data; |
2d9048e2 | 778 | |
63c882a0 EP |
779 | /* create/update an inode mark */ |
780 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 781 | path_put(&path); |
2d9048e2 | 782 | fput_and_out: |
2903ff01 | 783 | fdput(f); |
2d9048e2 AG |
784 | return ret; |
785 | } | |
786 | ||
2e4d0924 | 787 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 788 | { |
63c882a0 | 789 | struct fsnotify_group *group; |
000285de | 790 | struct inotify_inode_mark *i_mark; |
2903ff01 AV |
791 | struct fd f; |
792 | int ret = 0; | |
2d9048e2 | 793 | |
2903ff01 AV |
794 | f = fdget(fd); |
795 | if (unlikely(!f.file)) | |
2d9048e2 AG |
796 | return -EBADF; |
797 | ||
798 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 | 799 | ret = -EINVAL; |
2903ff01 | 800 | if (unlikely(f.file->f_op != &inotify_fops)) |
2d9048e2 | 801 | goto out; |
2d9048e2 | 802 | |
2903ff01 | 803 | group = f.file->private_data; |
2d9048e2 | 804 | |
b7ba8371 | 805 | ret = -EINVAL; |
000285de EP |
806 | i_mark = inotify_idr_find(group, wd); |
807 | if (unlikely(!i_mark)) | |
63c882a0 | 808 | goto out; |
63c882a0 | 809 | |
b7ba8371 EP |
810 | ret = 0; |
811 | ||
e2a29943 | 812 | fsnotify_destroy_mark(&i_mark->fsn_mark, group); |
b7ba8371 EP |
813 | |
814 | /* match ref taken by inotify_idr_find */ | |
000285de | 815 | fsnotify_put_mark(&i_mark->fsn_mark); |
2d9048e2 AG |
816 | |
817 | out: | |
2903ff01 | 818 | fdput(f); |
2d9048e2 AG |
819 | return ret; |
820 | } | |
821 | ||
2d9048e2 | 822 | /* |
ae0e47f0 | 823 | * inotify_user_setup - Our initialization function. Note that we cannot return |
2d9048e2 AG |
824 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
825 | * must result in panic(). | |
826 | */ | |
827 | static int __init inotify_user_setup(void) | |
828 | { | |
f874e1ac EP |
829 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); |
830 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
831 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
832 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
833 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
834 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
835 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
836 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
837 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
838 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
839 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
840 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
841 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
842 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
843 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
844 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
b29866aa | 845 | BUILD_BUG_ON(IN_ISDIR != FS_ISDIR); |
f874e1ac EP |
846 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); |
847 | ||
848 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
849 | ||
000285de | 850 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); |
63c882a0 | 851 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); |
63c882a0 | 852 | |
2d9048e2 AG |
853 | inotify_max_queued_events = 16384; |
854 | inotify_max_user_instances = 128; | |
855 | inotify_max_user_watches = 8192; | |
856 | ||
2d9048e2 AG |
857 | return 0; |
858 | } | |
2d9048e2 | 859 | module_init(inotify_user_setup); |