]>
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" |
2d9048e2 | 43 | |
63c882a0 | 44 | #include <asm/ioctls.h> |
2d9048e2 | 45 | |
2d9048e2 | 46 | /* these are configurable via /proc/sys/fs/inotify/ */ |
3c828e49 | 47 | static int inotify_max_user_instances __read_mostly; |
3c828e49 | 48 | static int inotify_max_queued_events __read_mostly; |
63c882a0 | 49 | int inotify_max_user_watches __read_mostly; |
2d9048e2 | 50 | |
63c882a0 EP |
51 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
52 | struct kmem_cache *event_priv_cachep __read_mostly; | |
2d9048e2 | 53 | |
2d9048e2 AG |
54 | #ifdef CONFIG_SYSCTL |
55 | ||
56 | #include <linux/sysctl.h> | |
57 | ||
58 | static int zero; | |
59 | ||
60 | ctl_table inotify_table[] = { | |
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 | |
63c882a0 EP |
93 | /* everything should accept their own ignored and cares about children */ |
94 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD); | |
2d9048e2 | 95 | |
63c882a0 EP |
96 | /* mask off the flags used to open the fd */ |
97 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT)); | |
2d9048e2 | 98 | |
63c882a0 | 99 | return mask; |
2d9048e2 AG |
100 | } |
101 | ||
63c882a0 | 102 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 103 | { |
63c882a0 EP |
104 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
105 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
106 | } |
107 | ||
63c882a0 | 108 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
109 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
110 | { | |
63c882a0 | 111 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
112 | int ret = 0; |
113 | ||
63c882a0 EP |
114 | poll_wait(file, &group->notification_waitq, wait); |
115 | mutex_lock(&group->notification_mutex); | |
116 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 117 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 118 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
119 | |
120 | return ret; | |
121 | } | |
122 | ||
3632dee2 VN |
123 | /* |
124 | * Get an inotify_kernel_event if one exists and is small | |
125 | * enough to fit in "count". Return an error pointer if | |
126 | * not large enough. | |
127 | * | |
63c882a0 | 128 | * Called with the group->notification_mutex held. |
3632dee2 | 129 | */ |
63c882a0 EP |
130 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
131 | size_t count) | |
3632dee2 VN |
132 | { |
133 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 134 | struct fsnotify_event *event; |
3632dee2 | 135 | |
63c882a0 | 136 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
137 | return NULL; |
138 | ||
63c882a0 EP |
139 | event = fsnotify_peek_notify_event(group); |
140 | ||
83cb10f0 EP |
141 | if (event->name_len) |
142 | event_size += roundup(event->name_len + 1, event_size); | |
3632dee2 VN |
143 | |
144 | if (event_size > count) | |
145 | return ERR_PTR(-EINVAL); | |
146 | ||
63c882a0 EP |
147 | /* held the notification_mutex the whole time, so this is the |
148 | * same event we peeked above */ | |
149 | fsnotify_remove_notify_event(group); | |
150 | ||
151 | return event; | |
3632dee2 VN |
152 | } |
153 | ||
154 | /* | |
155 | * Copy an event to user space, returning how much we copied. | |
156 | * | |
157 | * We already checked that the event size is smaller than the | |
158 | * buffer we had in "get_one_event()" above. | |
159 | */ | |
63c882a0 EP |
160 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
161 | struct fsnotify_event *event, | |
3632dee2 VN |
162 | char __user *buf) |
163 | { | |
63c882a0 EP |
164 | struct inotify_event inotify_event; |
165 | struct fsnotify_event_private_data *fsn_priv; | |
166 | struct inotify_event_private_data *priv; | |
3632dee2 | 167 | size_t event_size = sizeof(struct inotify_event); |
b962e731 | 168 | size_t name_len = 0; |
63c882a0 EP |
169 | |
170 | /* we get the inotify watch descriptor from the event private data */ | |
171 | spin_lock(&event->lock); | |
172 | fsn_priv = fsnotify_remove_priv_from_event(group, event); | |
173 | spin_unlock(&event->lock); | |
174 | ||
175 | if (!fsn_priv) | |
176 | inotify_event.wd = -1; | |
177 | else { | |
178 | priv = container_of(fsn_priv, struct inotify_event_private_data, | |
179 | fsnotify_event_priv_data); | |
180 | inotify_event.wd = priv->wd; | |
181 | inotify_free_event_priv(fsn_priv); | |
182 | } | |
183 | ||
b962e731 BR |
184 | /* |
185 | * round up event->name_len so it is a multiple of event_size | |
0db501bd EB |
186 | * plus an extra byte for the terminating '\0'. |
187 | */ | |
b962e731 BR |
188 | if (event->name_len) |
189 | name_len = roundup(event->name_len + 1, event_size); | |
63c882a0 EP |
190 | inotify_event.len = name_len; |
191 | ||
192 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
193 | inotify_event.cookie = event->sync_cookie; | |
3632dee2 | 194 | |
63c882a0 EP |
195 | /* send the main event */ |
196 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
197 | return -EFAULT; |
198 | ||
63c882a0 | 199 | buf += event_size; |
3632dee2 | 200 | |
63c882a0 EP |
201 | /* |
202 | * fsnotify only stores the pathname, so here we have to send the pathname | |
203 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
204 | * with zeros. I get my zeros from the nul_inotify_event. | |
205 | */ | |
206 | if (name_len) { | |
207 | unsigned int len_to_zero = name_len - event->name_len; | |
208 | /* copy the path name */ | |
209 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
3632dee2 | 210 | return -EFAULT; |
63c882a0 | 211 | buf += event->name_len; |
3632dee2 | 212 | |
0db501bd EB |
213 | /* fill userspace with 0's */ |
214 | if (clear_user(buf, len_to_zero)) | |
63c882a0 EP |
215 | return -EFAULT; |
216 | buf += len_to_zero; | |
217 | event_size += 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; | |
230 | DEFINE_WAIT(wait); | |
231 | ||
232 | start = buf; | |
63c882a0 | 233 | group = file->private_data; |
2d9048e2 AG |
234 | |
235 | while (1) { | |
63c882a0 | 236 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); |
2d9048e2 | 237 | |
63c882a0 EP |
238 | mutex_lock(&group->notification_mutex); |
239 | kevent = get_one_event(group, count); | |
240 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 241 | |
3632dee2 VN |
242 | if (kevent) { |
243 | ret = PTR_ERR(kevent); | |
244 | if (IS_ERR(kevent)) | |
245 | break; | |
63c882a0 EP |
246 | ret = copy_event_to_user(group, kevent, buf); |
247 | fsnotify_put_event(kevent); | |
3632dee2 VN |
248 | if (ret < 0) |
249 | break; | |
250 | buf += ret; | |
251 | count -= ret; | |
252 | continue; | |
2d9048e2 AG |
253 | } |
254 | ||
3632dee2 VN |
255 | ret = -EAGAIN; |
256 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 257 | break; |
3632dee2 VN |
258 | ret = -EINTR; |
259 | if (signal_pending(current)) | |
2d9048e2 | 260 | break; |
16dbc6c9 | 261 | |
3632dee2 | 262 | if (start != buf) |
2d9048e2 | 263 | break; |
16dbc6c9 | 264 | |
3632dee2 | 265 | schedule(); |
2d9048e2 | 266 | } |
2d9048e2 | 267 | |
63c882a0 | 268 | finish_wait(&group->notification_waitq, &wait); |
3632dee2 VN |
269 | if (start != buf && ret != -EFAULT) |
270 | ret = buf - start; | |
2d9048e2 AG |
271 | return ret; |
272 | } | |
273 | ||
bcfbf84d DA |
274 | static int inotify_fasync(int fd, struct file *file, int on) |
275 | { | |
63c882a0 | 276 | struct fsnotify_group *group = file->private_data; |
bcfbf84d | 277 | |
63c882a0 | 278 | return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; |
bcfbf84d DA |
279 | } |
280 | ||
2d9048e2 AG |
281 | static int inotify_release(struct inode *ignored, struct file *file) |
282 | { | |
63c882a0 | 283 | struct fsnotify_group *group = file->private_data; |
bdae997f | 284 | struct user_struct *user = group->inotify_data.user; |
2d9048e2 | 285 | |
63c882a0 | 286 | fsnotify_clear_marks_by_group(group); |
2d9048e2 | 287 | |
63c882a0 EP |
288 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
289 | fsnotify_put_group(group); | |
2d9048e2 | 290 | |
bdae997f KP |
291 | atomic_dec(&user->inotify_devs); |
292 | ||
2d9048e2 AG |
293 | return 0; |
294 | } | |
295 | ||
296 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
297 | unsigned long arg) | |
298 | { | |
63c882a0 EP |
299 | struct fsnotify_group *group; |
300 | struct fsnotify_event_holder *holder; | |
301 | struct fsnotify_event *event; | |
2d9048e2 AG |
302 | void __user *p; |
303 | int ret = -ENOTTY; | |
63c882a0 | 304 | size_t send_len = 0; |
2d9048e2 | 305 | |
63c882a0 | 306 | group = file->private_data; |
2d9048e2 AG |
307 | p = (void __user *) arg; |
308 | ||
309 | switch (cmd) { | |
310 | case FIONREAD: | |
63c882a0 EP |
311 | mutex_lock(&group->notification_mutex); |
312 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
313 | event = holder->event; | |
314 | send_len += sizeof(struct inotify_event); | |
83cb10f0 EP |
315 | if (event->name_len) |
316 | send_len += roundup(event->name_len + 1, | |
317 | sizeof(struct inotify_event)); | |
63c882a0 EP |
318 | } |
319 | mutex_unlock(&group->notification_mutex); | |
320 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
321 | break; |
322 | } | |
323 | ||
324 | return ret; | |
325 | } | |
326 | ||
327 | static const struct file_operations inotify_fops = { | |
63c882a0 EP |
328 | .poll = inotify_poll, |
329 | .read = inotify_read, | |
330 | .fasync = inotify_fasync, | |
331 | .release = inotify_release, | |
332 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 AG |
333 | .compat_ioctl = inotify_ioctl, |
334 | }; | |
335 | ||
2d9048e2 | 336 | |
63c882a0 EP |
337 | /* |
338 | * find_inode - resolve a user-given path to a specific inode | |
339 | */ | |
340 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
341 | { | |
342 | int error; | |
343 | ||
344 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
345 | if (error) | |
346 | return error; | |
347 | /* you can only watch an inode if you have read permissions on it */ | |
348 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
349 | if (error) | |
350 | path_put(path); | |
351 | return error; | |
352 | } | |
353 | ||
b7ba8371 | 354 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
7050c488 | 355 | int *last_wd, |
b7ba8371 EP |
356 | struct inotify_inode_mark_entry *ientry) |
357 | { | |
358 | int ret; | |
359 | ||
360 | do { | |
361 | if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) | |
362 | return -ENOMEM; | |
363 | ||
364 | spin_lock(idr_lock); | |
7050c488 | 365 | ret = idr_get_new_above(idr, ientry, *last_wd + 1, |
b7ba8371 EP |
366 | &ientry->wd); |
367 | /* we added the mark to the idr, take a reference */ | |
7050c488 | 368 | if (!ret) { |
b7ba8371 | 369 | fsnotify_get_mark(&ientry->fsn_entry); |
7050c488 EP |
370 | *last_wd = ientry->wd; |
371 | } | |
b7ba8371 EP |
372 | spin_unlock(idr_lock); |
373 | } while (ret == -EAGAIN); | |
374 | ||
375 | return ret; | |
376 | } | |
377 | ||
378 | static struct inotify_inode_mark_entry *inotify_idr_find_locked(struct fsnotify_group *group, | |
379 | int wd) | |
380 | { | |
381 | struct idr *idr = &group->inotify_data.idr; | |
382 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
383 | struct inotify_inode_mark_entry *ientry; | |
384 | ||
385 | assert_spin_locked(idr_lock); | |
386 | ||
387 | ientry = idr_find(idr, wd); | |
388 | if (ientry) { | |
389 | struct fsnotify_mark_entry *fsn_entry = &ientry->fsn_entry; | |
390 | ||
391 | fsnotify_get_mark(fsn_entry); | |
392 | /* One ref for being in the idr, one ref we just took */ | |
393 | BUG_ON(atomic_read(&fsn_entry->refcnt) < 2); | |
394 | } | |
395 | ||
396 | return ientry; | |
397 | } | |
398 | ||
399 | static struct inotify_inode_mark_entry *inotify_idr_find(struct fsnotify_group *group, | |
400 | int wd) | |
401 | { | |
402 | struct inotify_inode_mark_entry *ientry; | |
403 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
404 | ||
405 | spin_lock(idr_lock); | |
406 | ientry = inotify_idr_find_locked(group, wd); | |
407 | spin_unlock(idr_lock); | |
408 | ||
409 | return ientry; | |
410 | } | |
411 | ||
412 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
413 | struct inotify_inode_mark_entry *ientry) | |
414 | { | |
415 | struct idr *idr = &group->inotify_data.idr; | |
416 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
417 | int wd = ientry->wd; | |
418 | ||
419 | assert_spin_locked(idr_lock); | |
420 | ||
421 | idr_remove(idr, wd); | |
422 | ||
423 | /* removed from the idr, drop that ref */ | |
424 | fsnotify_put_mark(&ientry->fsn_entry); | |
425 | } | |
426 | ||
dead537d EP |
427 | /* |
428 | * Remove the mark from the idr (if present) and drop the reference | |
429 | * on the mark because it was in the idr. | |
430 | */ | |
7e790dd5 EP |
431 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
432 | struct inotify_inode_mark_entry *ientry) | |
433 | { | |
b7ba8371 EP |
434 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
435 | struct inotify_inode_mark_entry *found_ientry = NULL; | |
dead537d | 436 | int wd; |
7e790dd5 | 437 | |
b7ba8371 | 438 | spin_lock(idr_lock); |
dead537d EP |
439 | wd = ientry->wd; |
440 | ||
b7ba8371 EP |
441 | /* |
442 | * does this ientry think it is in the idr? we shouldn't get called | |
443 | * if it wasn't.... | |
444 | */ | |
445 | if (wd == -1) { | |
d7f0ce4e | 446 | WARN_ONCE(1, "%s: ientry=%p ientry->wd=%d ientry->group=%p" |
b7ba8371 EP |
447 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, |
448 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
dead537d | 449 | goto out; |
b7ba8371 | 450 | } |
dead537d | 451 | |
b7ba8371 EP |
452 | /* Lets look in the idr to see if we find it */ |
453 | found_ientry = inotify_idr_find_locked(group, wd); | |
454 | if (unlikely(!found_ientry)) { | |
d7f0ce4e | 455 | WARN_ONCE(1, "%s: ientry=%p ientry->wd=%d ientry->group=%p" |
b7ba8371 EP |
456 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, |
457 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
dead537d | 458 | goto out; |
b7ba8371 | 459 | } |
dead537d | 460 | |
b7ba8371 EP |
461 | /* |
462 | * We found an entry in the idr at the right wd, but it's | |
463 | * not the entry we were told to remove. eparis seriously | |
464 | * fucked up somewhere. | |
465 | */ | |
dead537d | 466 | if (unlikely(found_ientry != ientry)) { |
d7f0ce4e | 467 | WARN_ONCE(1, "%s: ientry=%p ientry->wd=%d ientry->group=%p " |
b7ba8371 EP |
468 | "entry->inode=%p found_ientry=%p found_ientry->wd=%d " |
469 | "found_ientry->group=%p found_ientry->inode=%p\n", | |
470 | __func__, ientry, ientry->wd, ientry->fsn_entry.group, | |
471 | ientry->fsn_entry.inode, found_ientry, found_ientry->wd, | |
472 | found_ientry->fsn_entry.group, | |
473 | found_ientry->fsn_entry.inode); | |
dead537d EP |
474 | goto out; |
475 | } | |
476 | ||
b7ba8371 EP |
477 | /* |
478 | * One ref for being in the idr | |
479 | * one ref held by the caller trying to kill us | |
480 | * one ref grabbed by inotify_idr_find | |
481 | */ | |
482 | if (unlikely(atomic_read(&ientry->fsn_entry.refcnt) < 3)) { | |
d7f0ce4e | 483 | printk(KERN_ERR "%s: ientry=%p ientry->wd=%d ientry->group=%p" |
b7ba8371 EP |
484 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, |
485 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
486 | /* we can't really recover with bad ref cnting.. */ | |
487 | BUG(); | |
488 | } | |
dead537d | 489 | |
b7ba8371 | 490 | do_inotify_remove_from_idr(group, ientry); |
dead537d | 491 | out: |
b7ba8371 EP |
492 | /* match the ref taken by inotify_idr_find_locked() */ |
493 | if (found_ientry) | |
494 | fsnotify_put_mark(&found_ientry->fsn_entry); | |
495 | ientry->wd = -1; | |
496 | spin_unlock(idr_lock); | |
7e790dd5 | 497 | } |
dead537d | 498 | |
63c882a0 | 499 | /* |
dead537d | 500 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 501 | */ |
528da3e9 EP |
502 | void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry, |
503 | struct fsnotify_group *group) | |
63c882a0 EP |
504 | { |
505 | struct inotify_inode_mark_entry *ientry; | |
f44aebcc | 506 | struct fsnotify_event *ignored_event; |
63c882a0 EP |
507 | struct inotify_event_private_data *event_priv; |
508 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 509 | int ret; |
63c882a0 | 510 | |
f44aebcc EP |
511 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
512 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
513 | GFP_NOFS); | |
514 | if (!ignored_event) | |
515 | return; | |
516 | ||
63c882a0 EP |
517 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
518 | ||
f44aebcc | 519 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
520 | if (unlikely(!event_priv)) |
521 | goto skip_send_ignore; | |
522 | ||
523 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
524 | ||
525 | fsn_event_priv->group = group; | |
526 | event_priv->wd = ientry->wd; | |
527 | ||
74766bbf | 528 | ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); |
eef3a116 | 529 | if (ret) |
63c882a0 EP |
530 | inotify_free_event_priv(fsn_event_priv); |
531 | ||
532 | skip_send_ignore: | |
533 | ||
f44aebcc EP |
534 | /* matches the reference taken when the event was created */ |
535 | fsnotify_put_event(ignored_event); | |
536 | ||
63c882a0 | 537 | /* remove this entry from the idr */ |
7e790dd5 | 538 | inotify_remove_from_idr(group, ientry); |
63c882a0 | 539 | |
5549f7cd | 540 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
541 | } |
542 | ||
543 | /* ding dong the mark is dead */ | |
544 | static void inotify_free_mark(struct fsnotify_mark_entry *entry) | |
545 | { | |
31ddd326 EP |
546 | struct inotify_inode_mark_entry *ientry; |
547 | ||
548 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); | |
63c882a0 EP |
549 | |
550 | kmem_cache_free(inotify_inode_mark_cachep, ientry); | |
551 | } | |
552 | ||
52cef755 EP |
553 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
554 | struct inode *inode, | |
555 | u32 arg) | |
63c882a0 | 556 | { |
52cef755 | 557 | struct fsnotify_mark_entry *entry; |
63c882a0 | 558 | struct inotify_inode_mark_entry *ientry; |
63c882a0 | 559 | __u32 old_mask, new_mask; |
52cef755 EP |
560 | __u32 mask; |
561 | int add = (arg & IN_MASK_ADD); | |
562 | int ret; | |
63c882a0 EP |
563 | |
564 | /* don't allow invalid bits: we don't want flags set */ | |
565 | mask = inotify_arg_to_mask(arg); | |
566 | if (unlikely(!mask)) | |
567 | return -EINVAL; | |
568 | ||
63c882a0 EP |
569 | spin_lock(&inode->i_lock); |
570 | entry = fsnotify_find_mark_entry(group, inode); | |
571 | spin_unlock(&inode->i_lock); | |
52cef755 EP |
572 | if (!entry) |
573 | return -ENOENT; | |
7e790dd5 | 574 | |
52cef755 | 575 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
75fe2b26 | 576 | |
63c882a0 EP |
577 | spin_lock(&entry->lock); |
578 | ||
579 | old_mask = entry->mask; | |
580 | if (add) { | |
581 | entry->mask |= mask; | |
582 | new_mask = entry->mask; | |
583 | } else { | |
584 | entry->mask = mask; | |
585 | new_mask = entry->mask; | |
586 | } | |
587 | ||
588 | spin_unlock(&entry->lock); | |
589 | ||
590 | if (old_mask != new_mask) { | |
591 | /* more bits in old than in new? */ | |
592 | int dropped = (old_mask & ~new_mask); | |
593 | /* more bits in this entry than the inode's mask? */ | |
594 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); | |
595 | /* more bits in this entry than the group? */ | |
596 | int do_group = (new_mask & ~group->mask); | |
597 | ||
598 | /* update the inode with this new entry */ | |
599 | if (dropped || do_inode) | |
600 | fsnotify_recalc_inode_mask(inode); | |
601 | ||
602 | /* update the group mask with the new mask */ | |
603 | if (dropped || do_group) | |
604 | fsnotify_recalc_group_mask(group); | |
605 | } | |
606 | ||
52cef755 EP |
607 | /* return the wd */ |
608 | ret = ientry->wd; | |
609 | ||
610 | /* match the get from fsnotify_find_mark_entry() */ | |
75fe2b26 EP |
611 | fsnotify_put_mark(entry); |
612 | ||
52cef755 EP |
613 | return ret; |
614 | } | |
615 | ||
616 | static int inotify_new_watch(struct fsnotify_group *group, | |
617 | struct inode *inode, | |
618 | u32 arg) | |
619 | { | |
620 | struct inotify_inode_mark_entry *tmp_ientry; | |
621 | __u32 mask; | |
622 | int ret; | |
b7ba8371 EP |
623 | struct idr *idr = &group->inotify_data.idr; |
624 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 EP |
625 | |
626 | /* don't allow invalid bits: we don't want flags set */ | |
627 | mask = inotify_arg_to_mask(arg); | |
628 | if (unlikely(!mask)) | |
629 | return -EINVAL; | |
630 | ||
631 | tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); | |
632 | if (unlikely(!tmp_ientry)) | |
633 | return -ENOMEM; | |
634 | ||
635 | fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark); | |
636 | tmp_ientry->fsn_entry.mask = mask; | |
637 | tmp_ientry->wd = -1; | |
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, |
b7ba8371 EP |
644 | tmp_ientry); |
645 | if (ret) | |
52cef755 | 646 | goto out_err; |
52cef755 EP |
647 | |
648 | /* we are on the idr, now get on the inode */ | |
40554c3d | 649 | ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode, 0); |
52cef755 EP |
650 | if (ret) { |
651 | /* we failed to get on the inode, get off the idr */ | |
652 | inotify_remove_from_idr(group, tmp_ientry); | |
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 | ||
659 | /* return the watch descriptor for this new entry */ | |
660 | ret = tmp_ientry->wd; | |
661 | ||
750a8870 EP |
662 | /* if this mark added a new event update the group mask */ |
663 | if (mask & ~group->mask) | |
664 | fsnotify_recalc_group_mask(group); | |
665 | ||
63c882a0 | 666 | out_err: |
3dbc6fb6 EP |
667 | /* match the ref from fsnotify_init_markentry() */ |
668 | fsnotify_put_mark(&tmp_ientry->fsn_entry); | |
52cef755 EP |
669 | |
670 | return ret; | |
671 | } | |
672 | ||
673 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
674 | { | |
675 | int ret = 0; | |
676 | ||
677 | retry: | |
678 | /* try to update and existing watch with the new arg */ | |
679 | ret = inotify_update_existing_watch(group, inode, arg); | |
680 | /* no mark present, try to add a new one */ | |
681 | if (ret == -ENOENT) | |
682 | ret = inotify_new_watch(group, inode, arg); | |
683 | /* | |
684 | * inotify_new_watch could race with another thread which did an | |
685 | * inotify_new_watch between the update_existing and the add watch | |
686 | * here, go back and try to update an existing mark again. | |
687 | */ | |
688 | if (ret == -EEXIST) | |
689 | goto retry; | |
7e790dd5 | 690 | |
63c882a0 EP |
691 | return ret; |
692 | } | |
693 | ||
694 | static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events) | |
695 | { | |
696 | struct fsnotify_group *group; | |
63c882a0 | 697 | |
0d2e2a1d | 698 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
699 | if (IS_ERR(group)) |
700 | return group; | |
701 | ||
702 | group->max_events = max_events; | |
703 | ||
704 | spin_lock_init(&group->inotify_data.idr_lock); | |
705 | idr_init(&group->inotify_data.idr); | |
9e572cc9 | 706 | group->inotify_data.last_wd = 0; |
63c882a0 EP |
707 | group->inotify_data.user = user; |
708 | group->inotify_data.fa = NULL; | |
709 | ||
710 | return group; | |
711 | } | |
712 | ||
713 | ||
714 | /* inotify syscalls */ | |
938bb9f5 | 715 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 716 | { |
63c882a0 | 717 | struct fsnotify_group *group; |
2d9048e2 | 718 | struct user_struct *user; |
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 | ||
da9592ed | 728 | user = get_current_user(); |
2d9048e2 AG |
729 | if (unlikely(atomic_read(&user->inotify_devs) >= |
730 | inotify_max_user_instances)) { | |
731 | ret = -EMFILE; | |
732 | goto out_free_uid; | |
733 | } | |
734 | ||
63c882a0 EP |
735 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
736 | group = inotify_new_group(user, inotify_max_queued_events); | |
737 | if (IS_ERR(group)) { | |
738 | ret = PTR_ERR(group); | |
2d9048e2 AG |
739 | goto out_free_uid; |
740 | } | |
741 | ||
825f9692 AV |
742 | atomic_inc(&user->inotify_devs); |
743 | ||
c44dcc56 AV |
744 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
745 | O_RDONLY | flags); | |
746 | if (ret >= 0) | |
747 | return ret; | |
825f9692 | 748 | |
825f9692 | 749 | atomic_dec(&user->inotify_devs); |
2d9048e2 AG |
750 | out_free_uid: |
751 | free_uid(user); | |
2d9048e2 AG |
752 | return ret; |
753 | } | |
754 | ||
938bb9f5 | 755 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
756 | { |
757 | return sys_inotify_init1(0); | |
758 | } | |
759 | ||
2e4d0924 HC |
760 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
761 | u32, mask) | |
2d9048e2 | 762 | { |
63c882a0 | 763 | struct fsnotify_group *group; |
2d9048e2 | 764 | struct inode *inode; |
2d8f3038 | 765 | struct path path; |
2d9048e2 AG |
766 | struct file *filp; |
767 | int ret, fput_needed; | |
768 | unsigned flags = 0; | |
769 | ||
770 | filp = fget_light(fd, &fput_needed); | |
771 | if (unlikely(!filp)) | |
772 | return -EBADF; | |
773 | ||
774 | /* verify that this is indeed an inotify instance */ | |
775 | if (unlikely(filp->f_op != &inotify_fops)) { | |
776 | ret = -EINVAL; | |
777 | goto fput_and_out; | |
778 | } | |
779 | ||
780 | if (!(mask & IN_DONT_FOLLOW)) | |
781 | flags |= LOOKUP_FOLLOW; | |
782 | if (mask & IN_ONLYDIR) | |
783 | flags |= LOOKUP_DIRECTORY; | |
784 | ||
63c882a0 EP |
785 | ret = inotify_find_inode(pathname, &path, flags); |
786 | if (ret) | |
2d9048e2 AG |
787 | goto fput_and_out; |
788 | ||
63c882a0 | 789 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 790 | inode = path.dentry->d_inode; |
63c882a0 | 791 | group = filp->private_data; |
2d9048e2 | 792 | |
63c882a0 EP |
793 | /* create/update an inode mark */ |
794 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 795 | path_put(&path); |
2d9048e2 AG |
796 | fput_and_out: |
797 | fput_light(filp, fput_needed); | |
798 | return ret; | |
799 | } | |
800 | ||
2e4d0924 | 801 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 802 | { |
63c882a0 | 803 | struct fsnotify_group *group; |
b7ba8371 | 804 | struct inotify_inode_mark_entry *ientry; |
2d9048e2 | 805 | struct file *filp; |
63c882a0 | 806 | int ret = 0, fput_needed; |
2d9048e2 AG |
807 | |
808 | filp = fget_light(fd, &fput_needed); | |
809 | if (unlikely(!filp)) | |
810 | return -EBADF; | |
811 | ||
812 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 EP |
813 | ret = -EINVAL; |
814 | if (unlikely(filp->f_op != &inotify_fops)) | |
2d9048e2 | 815 | goto out; |
2d9048e2 | 816 | |
63c882a0 | 817 | group = filp->private_data; |
2d9048e2 | 818 | |
b7ba8371 EP |
819 | ret = -EINVAL; |
820 | ientry = inotify_idr_find(group, wd); | |
821 | if (unlikely(!ientry)) | |
63c882a0 | 822 | goto out; |
63c882a0 | 823 | |
b7ba8371 EP |
824 | ret = 0; |
825 | ||
826 | fsnotify_destroy_mark_by_entry(&ientry->fsn_entry); | |
827 | ||
828 | /* match ref taken by inotify_idr_find */ | |
829 | fsnotify_put_mark(&ientry->fsn_entry); | |
2d9048e2 AG |
830 | |
831 | out: | |
832 | fput_light(filp, fput_needed); | |
833 | return ret; | |
834 | } | |
835 | ||
2d9048e2 AG |
836 | /* |
837 | * inotify_user_setup - Our initialization function. Note that we cannnot return | |
838 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here | |
839 | * must result in panic(). | |
840 | */ | |
841 | static int __init inotify_user_setup(void) | |
842 | { | |
63c882a0 EP |
843 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC); |
844 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); | |
63c882a0 | 845 | |
2d9048e2 AG |
846 | inotify_max_queued_events = 16384; |
847 | inotify_max_user_instances = 128; | |
848 | inotify_max_user_watches = 8192; | |
849 | ||
2d9048e2 AG |
850 | return 0; |
851 | } | |
2d9048e2 | 852 | module_init(inotify_user_setup); |