]>
Commit | Line | Data |
---|---|---|
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 | * | |
11 | * Copyright (C) 2009 Eric Paris <Red Hat Inc> | |
12 | * inotify was largely rewriten to make use of the fsnotify infrastructure | |
13 | * | |
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 | ||
25 | #include <linux/file.h> | |
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 */ | |
30 | #include <linux/inotify.h> | |
31 | #include <linux/kernel.h> /* roundup() */ | |
32 | #include <linux/namei.h> /* LOOKUP_FOLLOW */ | |
33 | #include <linux/sched.h> /* struct user */ | |
34 | #include <linux/slab.h> /* struct kmem_cache */ | |
35 | #include <linux/syscalls.h> | |
36 | #include <linux/types.h> | |
37 | #include <linux/anon_inodes.h> | |
38 | #include <linux/uaccess.h> | |
39 | #include <linux/poll.h> | |
40 | #include <linux/wait.h> | |
41 | ||
42 | #include "inotify.h" | |
43 | ||
44 | #include <asm/ioctls.h> | |
45 | ||
46 | /* these are configurable via /proc/sys/fs/inotify/ */ | |
47 | static int inotify_max_user_instances __read_mostly; | |
48 | static int inotify_max_queued_events __read_mostly; | |
49 | static int inotify_max_user_watches __read_mostly; | |
50 | ||
51 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; | |
52 | struct kmem_cache *event_priv_cachep __read_mostly; | |
53 | ||
54 | #ifdef CONFIG_SYSCTL | |
55 | ||
56 | #include <linux/sysctl.h> | |
57 | ||
58 | static int zero; | |
59 | ||
60 | ctl_table inotify_table[] = { | |
61 | { | |
62 | .procname = "max_user_instances", | |
63 | .data = &inotify_max_user_instances, | |
64 | .maxlen = sizeof(int), | |
65 | .mode = 0644, | |
66 | .proc_handler = proc_dointvec_minmax, | |
67 | .extra1 = &zero, | |
68 | }, | |
69 | { | |
70 | .procname = "max_user_watches", | |
71 | .data = &inotify_max_user_watches, | |
72 | .maxlen = sizeof(int), | |
73 | .mode = 0644, | |
74 | .proc_handler = proc_dointvec_minmax, | |
75 | .extra1 = &zero, | |
76 | }, | |
77 | { | |
78 | .procname = "max_queued_events", | |
79 | .data = &inotify_max_queued_events, | |
80 | .maxlen = sizeof(int), | |
81 | .mode = 0644, | |
82 | .proc_handler = proc_dointvec_minmax, | |
83 | .extra1 = &zero | |
84 | }, | |
85 | { } | |
86 | }; | |
87 | #endif /* CONFIG_SYSCTL */ | |
88 | ||
89 | static inline __u32 inotify_arg_to_mask(u32 arg) | |
90 | { | |
91 | __u32 mask; | |
92 | ||
93 | /* | |
94 | * everything should accept their own ignored, cares about children, | |
95 | * and should receive events when the inode is unmounted | |
96 | */ | |
97 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); | |
98 | ||
99 | /* mask off the flags used to open the fd */ | |
100 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); | |
101 | ||
102 | return mask; | |
103 | } | |
104 | ||
105 | static inline u32 inotify_mask_to_arg(__u32 mask) | |
106 | { | |
107 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | | |
108 | IN_Q_OVERFLOW); | |
109 | } | |
110 | ||
111 | /* intofiy userspace file descriptor functions */ | |
112 | static unsigned int inotify_poll(struct file *file, poll_table *wait) | |
113 | { | |
114 | struct fsnotify_group *group = file->private_data; | |
115 | int ret = 0; | |
116 | ||
117 | poll_wait(file, &group->notification_waitq, wait); | |
118 | mutex_lock(&group->notification_mutex); | |
119 | if (!fsnotify_notify_queue_is_empty(group)) | |
120 | ret = POLLIN | POLLRDNORM; | |
121 | mutex_unlock(&group->notification_mutex); | |
122 | ||
123 | return ret; | |
124 | } | |
125 | ||
126 | /* | |
127 | * Get an inotify_kernel_event if one exists and is small | |
128 | * enough to fit in "count". Return an error pointer if | |
129 | * not large enough. | |
130 | * | |
131 | * Called with the group->notification_mutex held. | |
132 | */ | |
133 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, | |
134 | size_t count) | |
135 | { | |
136 | size_t event_size = sizeof(struct inotify_event); | |
137 | struct fsnotify_event *event; | |
138 | ||
139 | if (fsnotify_notify_queue_is_empty(group)) | |
140 | return NULL; | |
141 | ||
142 | event = fsnotify_peek_notify_event(group); | |
143 | ||
144 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); | |
145 | ||
146 | if (event->name_len) | |
147 | event_size += roundup(event->name_len + 1, event_size); | |
148 | ||
149 | if (event_size > count) | |
150 | return ERR_PTR(-EINVAL); | |
151 | ||
152 | /* held the notification_mutex the whole time, so this is the | |
153 | * same event we peeked above */ | |
154 | fsnotify_remove_notify_event(group); | |
155 | ||
156 | return event; | |
157 | } | |
158 | ||
159 | /* | |
160 | * Copy an event to user space, returning how much we copied. | |
161 | * | |
162 | * We already checked that the event size is smaller than the | |
163 | * buffer we had in "get_one_event()" above. | |
164 | */ | |
165 | static ssize_t copy_event_to_user(struct fsnotify_group *group, | |
166 | struct fsnotify_event *event, | |
167 | char __user *buf) | |
168 | { | |
169 | struct inotify_event inotify_event; | |
170 | struct fsnotify_event_private_data *fsn_priv; | |
171 | struct inotify_event_private_data *priv; | |
172 | size_t event_size = sizeof(struct inotify_event); | |
173 | size_t name_len = 0; | |
174 | ||
175 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); | |
176 | ||
177 | /* we get the inotify watch descriptor from the event private data */ | |
178 | spin_lock(&event->lock); | |
179 | fsn_priv = fsnotify_remove_priv_from_event(group, event); | |
180 | spin_unlock(&event->lock); | |
181 | ||
182 | if (!fsn_priv) | |
183 | inotify_event.wd = -1; | |
184 | else { | |
185 | priv = container_of(fsn_priv, struct inotify_event_private_data, | |
186 | fsnotify_event_priv_data); | |
187 | inotify_event.wd = priv->wd; | |
188 | inotify_free_event_priv(fsn_priv); | |
189 | } | |
190 | ||
191 | /* | |
192 | * round up event->name_len so it is a multiple of event_size | |
193 | * plus an extra byte for the terminating '\0'. | |
194 | */ | |
195 | if (event->name_len) | |
196 | name_len = roundup(event->name_len + 1, event_size); | |
197 | inotify_event.len = name_len; | |
198 | ||
199 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
200 | inotify_event.cookie = event->sync_cookie; | |
201 | ||
202 | /* send the main event */ | |
203 | if (copy_to_user(buf, &inotify_event, event_size)) | |
204 | return -EFAULT; | |
205 | ||
206 | buf += event_size; | |
207 | ||
208 | /* | |
209 | * fsnotify only stores the pathname, so here we have to send the pathname | |
210 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
211 | * with zeros. I get my zeros from the nul_inotify_event. | |
212 | */ | |
213 | if (name_len) { | |
214 | unsigned int len_to_zero = name_len - event->name_len; | |
215 | /* copy the path name */ | |
216 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
217 | return -EFAULT; | |
218 | buf += event->name_len; | |
219 | ||
220 | /* fill userspace with 0's */ | |
221 | if (clear_user(buf, len_to_zero)) | |
222 | return -EFAULT; | |
223 | buf += len_to_zero; | |
224 | event_size += name_len; | |
225 | } | |
226 | ||
227 | return event_size; | |
228 | } | |
229 | ||
230 | static ssize_t inotify_read(struct file *file, char __user *buf, | |
231 | size_t count, loff_t *pos) | |
232 | { | |
233 | struct fsnotify_group *group; | |
234 | struct fsnotify_event *kevent; | |
235 | char __user *start; | |
236 | int ret; | |
237 | DEFINE_WAIT(wait); | |
238 | ||
239 | start = buf; | |
240 | group = file->private_data; | |
241 | ||
242 | while (1) { | |
243 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); | |
244 | ||
245 | mutex_lock(&group->notification_mutex); | |
246 | kevent = get_one_event(group, count); | |
247 | mutex_unlock(&group->notification_mutex); | |
248 | ||
249 | pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); | |
250 | ||
251 | if (kevent) { | |
252 | ret = PTR_ERR(kevent); | |
253 | if (IS_ERR(kevent)) | |
254 | break; | |
255 | ret = copy_event_to_user(group, kevent, buf); | |
256 | fsnotify_put_event(kevent); | |
257 | if (ret < 0) | |
258 | break; | |
259 | buf += ret; | |
260 | count -= ret; | |
261 | continue; | |
262 | } | |
263 | ||
264 | ret = -EAGAIN; | |
265 | if (file->f_flags & O_NONBLOCK) | |
266 | break; | |
267 | ret = -EINTR; | |
268 | if (signal_pending(current)) | |
269 | break; | |
270 | ||
271 | if (start != buf) | |
272 | break; | |
273 | ||
274 | schedule(); | |
275 | } | |
276 | ||
277 | finish_wait(&group->notification_waitq, &wait); | |
278 | if (start != buf && ret != -EFAULT) | |
279 | ret = buf - start; | |
280 | return ret; | |
281 | } | |
282 | ||
283 | static int inotify_fasync(int fd, struct file *file, int on) | |
284 | { | |
285 | struct fsnotify_group *group = file->private_data; | |
286 | ||
287 | return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; | |
288 | } | |
289 | ||
290 | static int inotify_release(struct inode *ignored, struct file *file) | |
291 | { | |
292 | struct fsnotify_group *group = file->private_data; | |
293 | struct user_struct *user = group->inotify_data.user; | |
294 | ||
295 | pr_debug("%s: group=%p\n", __func__, group); | |
296 | ||
297 | fsnotify_clear_marks_by_group(group); | |
298 | ||
299 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ | |
300 | fsnotify_put_group(group); | |
301 | ||
302 | atomic_dec(&user->inotify_devs); | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
307 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
308 | unsigned long arg) | |
309 | { | |
310 | struct fsnotify_group *group; | |
311 | struct fsnotify_event_holder *holder; | |
312 | struct fsnotify_event *event; | |
313 | void __user *p; | |
314 | int ret = -ENOTTY; | |
315 | size_t send_len = 0; | |
316 | ||
317 | group = file->private_data; | |
318 | p = (void __user *) arg; | |
319 | ||
320 | pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); | |
321 | ||
322 | switch (cmd) { | |
323 | case FIONREAD: | |
324 | mutex_lock(&group->notification_mutex); | |
325 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
326 | event = holder->event; | |
327 | send_len += sizeof(struct inotify_event); | |
328 | if (event->name_len) | |
329 | send_len += roundup(event->name_len + 1, | |
330 | sizeof(struct inotify_event)); | |
331 | } | |
332 | mutex_unlock(&group->notification_mutex); | |
333 | ret = put_user(send_len, (int __user *) p); | |
334 | break; | |
335 | } | |
336 | ||
337 | return ret; | |
338 | } | |
339 | ||
340 | static const struct file_operations inotify_fops = { | |
341 | .poll = inotify_poll, | |
342 | .read = inotify_read, | |
343 | .fasync = inotify_fasync, | |
344 | .release = inotify_release, | |
345 | .unlocked_ioctl = inotify_ioctl, | |
346 | .compat_ioctl = inotify_ioctl, | |
347 | .llseek = noop_llseek, | |
348 | }; | |
349 | ||
350 | ||
351 | /* | |
352 | * find_inode - resolve a user-given path to a specific inode | |
353 | */ | |
354 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
355 | { | |
356 | int error; | |
357 | ||
358 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
359 | if (error) | |
360 | return error; | |
361 | /* you can only watch an inode if you have read permissions on it */ | |
362 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
363 | if (error) | |
364 | path_put(path); | |
365 | return error; | |
366 | } | |
367 | ||
368 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, | |
369 | int *last_wd, | |
370 | struct inotify_inode_mark *i_mark) | |
371 | { | |
372 | int ret; | |
373 | ||
374 | do { | |
375 | if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) | |
376 | return -ENOMEM; | |
377 | ||
378 | spin_lock(idr_lock); | |
379 | ret = idr_get_new_above(idr, i_mark, *last_wd + 1, | |
380 | &i_mark->wd); | |
381 | /* we added the mark to the idr, take a reference */ | |
382 | if (!ret) { | |
383 | *last_wd = i_mark->wd; | |
384 | fsnotify_get_mark(&i_mark->fsn_mark); | |
385 | } | |
386 | spin_unlock(idr_lock); | |
387 | } while (ret == -EAGAIN); | |
388 | ||
389 | return ret; | |
390 | } | |
391 | ||
392 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, | |
393 | int wd) | |
394 | { | |
395 | struct idr *idr = &group->inotify_data.idr; | |
396 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
397 | struct inotify_inode_mark *i_mark; | |
398 | ||
399 | assert_spin_locked(idr_lock); | |
400 | ||
401 | i_mark = idr_find(idr, wd); | |
402 | if (i_mark) { | |
403 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
404 | ||
405 | fsnotify_get_mark(fsn_mark); | |
406 | /* One ref for being in the idr, one ref we just took */ | |
407 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); | |
408 | } | |
409 | ||
410 | return i_mark; | |
411 | } | |
412 | ||
413 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, | |
414 | int wd) | |
415 | { | |
416 | struct inotify_inode_mark *i_mark; | |
417 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
418 | ||
419 | spin_lock(idr_lock); | |
420 | i_mark = inotify_idr_find_locked(group, wd); | |
421 | spin_unlock(idr_lock); | |
422 | ||
423 | return i_mark; | |
424 | } | |
425 | ||
426 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
427 | struct inotify_inode_mark *i_mark) | |
428 | { | |
429 | struct idr *idr = &group->inotify_data.idr; | |
430 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
431 | int wd = i_mark->wd; | |
432 | ||
433 | assert_spin_locked(idr_lock); | |
434 | ||
435 | idr_remove(idr, wd); | |
436 | ||
437 | /* removed from the idr, drop that ref */ | |
438 | fsnotify_put_mark(&i_mark->fsn_mark); | |
439 | } | |
440 | ||
441 | /* | |
442 | * Remove the mark from the idr (if present) and drop the reference | |
443 | * on the mark because it was in the idr. | |
444 | */ | |
445 | static void inotify_remove_from_idr(struct fsnotify_group *group, | |
446 | struct inotify_inode_mark *i_mark) | |
447 | { | |
448 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
449 | struct inotify_inode_mark *found_i_mark = NULL; | |
450 | int wd; | |
451 | ||
452 | spin_lock(idr_lock); | |
453 | wd = i_mark->wd; | |
454 | ||
455 | /* | |
456 | * does this i_mark think it is in the idr? we shouldn't get called | |
457 | * if it wasn't.... | |
458 | */ | |
459 | if (wd == -1) { | |
460 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
461 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
462 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
463 | goto out; | |
464 | } | |
465 | ||
466 | /* Lets look in the idr to see if we find it */ | |
467 | found_i_mark = inotify_idr_find_locked(group, wd); | |
468 | if (unlikely(!found_i_mark)) { | |
469 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
470 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
471 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
472 | goto out; | |
473 | } | |
474 | ||
475 | /* | |
476 | * We found an mark in the idr at the right wd, but it's | |
477 | * not the mark we were told to remove. eparis seriously | |
478 | * fucked up somewhere. | |
479 | */ | |
480 | if (unlikely(found_i_mark != i_mark)) { | |
481 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
482 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
483 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
484 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
485 | i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, | |
486 | found_i_mark->fsn_mark.group, | |
487 | found_i_mark->fsn_mark.i.inode); | |
488 | goto out; | |
489 | } | |
490 | ||
491 | /* | |
492 | * One ref for being in the idr | |
493 | * one ref held by the caller trying to kill us | |
494 | * one ref grabbed by inotify_idr_find | |
495 | */ | |
496 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { | |
497 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
498 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
499 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
500 | /* we can't really recover with bad ref cnting.. */ | |
501 | BUG(); | |
502 | } | |
503 | ||
504 | do_inotify_remove_from_idr(group, i_mark); | |
505 | out: | |
506 | /* match the ref taken by inotify_idr_find_locked() */ | |
507 | if (found_i_mark) | |
508 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
509 | i_mark->wd = -1; | |
510 | spin_unlock(idr_lock); | |
511 | } | |
512 | ||
513 | /* | |
514 | * Send IN_IGNORED for this wd, remove this wd from the idr. | |
515 | */ | |
516 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, | |
517 | struct fsnotify_group *group) | |
518 | { | |
519 | struct inotify_inode_mark *i_mark; | |
520 | struct fsnotify_event *ignored_event, *notify_event; | |
521 | struct inotify_event_private_data *event_priv; | |
522 | struct fsnotify_event_private_data *fsn_event_priv; | |
523 | int ret; | |
524 | ||
525 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, | |
526 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
527 | GFP_NOFS); | |
528 | if (!ignored_event) | |
529 | return; | |
530 | ||
531 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); | |
532 | ||
533 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); | |
534 | if (unlikely(!event_priv)) | |
535 | goto skip_send_ignore; | |
536 | ||
537 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
538 | ||
539 | fsn_event_priv->group = group; | |
540 | event_priv->wd = i_mark->wd; | |
541 | ||
542 | notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); | |
543 | if (notify_event) { | |
544 | if (IS_ERR(notify_event)) | |
545 | ret = PTR_ERR(notify_event); | |
546 | else | |
547 | fsnotify_put_event(notify_event); | |
548 | inotify_free_event_priv(fsn_event_priv); | |
549 | } | |
550 | ||
551 | skip_send_ignore: | |
552 | ||
553 | /* matches the reference taken when the event was created */ | |
554 | fsnotify_put_event(ignored_event); | |
555 | ||
556 | /* remove this mark from the idr */ | |
557 | inotify_remove_from_idr(group, i_mark); | |
558 | ||
559 | atomic_dec(&group->inotify_data.user->inotify_watches); | |
560 | } | |
561 | ||
562 | /* ding dong the mark is dead */ | |
563 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) | |
564 | { | |
565 | struct inotify_inode_mark *i_mark; | |
566 | ||
567 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); | |
568 | ||
569 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); | |
570 | } | |
571 | ||
572 | static int inotify_update_existing_watch(struct fsnotify_group *group, | |
573 | struct inode *inode, | |
574 | u32 arg) | |
575 | { | |
576 | struct fsnotify_mark *fsn_mark; | |
577 | struct inotify_inode_mark *i_mark; | |
578 | __u32 old_mask, new_mask; | |
579 | __u32 mask; | |
580 | int add = (arg & IN_MASK_ADD); | |
581 | int ret; | |
582 | ||
583 | /* don't allow invalid bits: we don't want flags set */ | |
584 | mask = inotify_arg_to_mask(arg); | |
585 | if (unlikely(!(mask & IN_ALL_EVENTS))) | |
586 | return -EINVAL; | |
587 | ||
588 | fsn_mark = fsnotify_find_inode_mark(group, inode); | |
589 | if (!fsn_mark) | |
590 | return -ENOENT; | |
591 | ||
592 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); | |
593 | ||
594 | spin_lock(&fsn_mark->lock); | |
595 | ||
596 | old_mask = fsn_mark->mask; | |
597 | if (add) | |
598 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
599 | else | |
600 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
601 | new_mask = fsn_mark->mask; | |
602 | ||
603 | spin_unlock(&fsn_mark->lock); | |
604 | ||
605 | if (old_mask != new_mask) { | |
606 | /* more bits in old than in new? */ | |
607 | int dropped = (old_mask & ~new_mask); | |
608 | /* more bits in this fsn_mark than the inode's mask? */ | |
609 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); | |
610 | ||
611 | /* update the inode with this new fsn_mark */ | |
612 | if (dropped || do_inode) | |
613 | fsnotify_recalc_inode_mask(inode); | |
614 | ||
615 | } | |
616 | ||
617 | /* return the wd */ | |
618 | ret = i_mark->wd; | |
619 | ||
620 | /* match the get from fsnotify_find_mark() */ | |
621 | fsnotify_put_mark(fsn_mark); | |
622 | ||
623 | return ret; | |
624 | } | |
625 | ||
626 | static int inotify_new_watch(struct fsnotify_group *group, | |
627 | struct inode *inode, | |
628 | u32 arg) | |
629 | { | |
630 | struct inotify_inode_mark *tmp_i_mark; | |
631 | __u32 mask; | |
632 | int ret; | |
633 | struct idr *idr = &group->inotify_data.idr; | |
634 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
635 | ||
636 | /* don't allow invalid bits: we don't want flags set */ | |
637 | mask = inotify_arg_to_mask(arg); | |
638 | if (unlikely(!(mask & IN_ALL_EVENTS))) | |
639 | return -EINVAL; | |
640 | ||
641 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); | |
642 | if (unlikely(!tmp_i_mark)) | |
643 | return -ENOMEM; | |
644 | ||
645 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); | |
646 | tmp_i_mark->fsn_mark.mask = mask; | |
647 | tmp_i_mark->wd = -1; | |
648 | ||
649 | ret = -ENOSPC; | |
650 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
651 | goto out_err; | |
652 | ||
653 | ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd, | |
654 | tmp_i_mark); | |
655 | if (ret) | |
656 | goto out_err; | |
657 | ||
658 | /* we are on the idr, now get on the inode */ | |
659 | ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0); | |
660 | if (ret) { | |
661 | /* we failed to get on the inode, get off the idr */ | |
662 | inotify_remove_from_idr(group, tmp_i_mark); | |
663 | goto out_err; | |
664 | } | |
665 | ||
666 | /* increment the number of watches the user has */ | |
667 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
668 | ||
669 | /* return the watch descriptor for this new mark */ | |
670 | ret = tmp_i_mark->wd; | |
671 | ||
672 | out_err: | |
673 | /* match the ref from fsnotify_init_mark() */ | |
674 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
675 | ||
676 | return ret; | |
677 | } | |
678 | ||
679 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
680 | { | |
681 | int ret = 0; | |
682 | ||
683 | retry: | |
684 | /* try to update and existing watch with the new arg */ | |
685 | ret = inotify_update_existing_watch(group, inode, arg); | |
686 | /* no mark present, try to add a new one */ | |
687 | if (ret == -ENOENT) | |
688 | ret = inotify_new_watch(group, inode, arg); | |
689 | /* | |
690 | * inotify_new_watch could race with another thread which did an | |
691 | * inotify_new_watch between the update_existing and the add watch | |
692 | * here, go back and try to update an existing mark again. | |
693 | */ | |
694 | if (ret == -EEXIST) | |
695 | goto retry; | |
696 | ||
697 | return ret; | |
698 | } | |
699 | ||
700 | static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events) | |
701 | { | |
702 | struct fsnotify_group *group; | |
703 | ||
704 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); | |
705 | if (IS_ERR(group)) | |
706 | return group; | |
707 | ||
708 | group->max_events = max_events; | |
709 | ||
710 | spin_lock_init(&group->inotify_data.idr_lock); | |
711 | idr_init(&group->inotify_data.idr); | |
712 | group->inotify_data.last_wd = 0; | |
713 | group->inotify_data.user = user; | |
714 | group->inotify_data.fa = NULL; | |
715 | ||
716 | return group; | |
717 | } | |
718 | ||
719 | ||
720 | /* inotify syscalls */ | |
721 | SYSCALL_DEFINE1(inotify_init1, int, flags) | |
722 | { | |
723 | struct fsnotify_group *group; | |
724 | struct user_struct *user; | |
725 | int ret; | |
726 | ||
727 | /* Check the IN_* constants for consistency. */ | |
728 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
729 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
730 | ||
731 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) | |
732 | return -EINVAL; | |
733 | ||
734 | user = get_current_user(); | |
735 | if (unlikely(atomic_read(&user->inotify_devs) >= | |
736 | inotify_max_user_instances)) { | |
737 | ret = -EMFILE; | |
738 | goto out_free_uid; | |
739 | } | |
740 | ||
741 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ | |
742 | group = inotify_new_group(user, inotify_max_queued_events); | |
743 | if (IS_ERR(group)) { | |
744 | ret = PTR_ERR(group); | |
745 | goto out_free_uid; | |
746 | } | |
747 | ||
748 | atomic_inc(&user->inotify_devs); | |
749 | ||
750 | ret = anon_inode_getfd("inotify", &inotify_fops, group, | |
751 | O_RDONLY | flags); | |
752 | if (ret >= 0) | |
753 | return ret; | |
754 | ||
755 | atomic_dec(&user->inotify_devs); | |
756 | out_free_uid: | |
757 | free_uid(user); | |
758 | return ret; | |
759 | } | |
760 | ||
761 | SYSCALL_DEFINE0(inotify_init) | |
762 | { | |
763 | return sys_inotify_init1(0); | |
764 | } | |
765 | ||
766 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, | |
767 | u32, mask) | |
768 | { | |
769 | struct fsnotify_group *group; | |
770 | struct inode *inode; | |
771 | struct path path; | |
772 | struct file *filp; | |
773 | int ret, fput_needed; | |
774 | unsigned flags = 0; | |
775 | ||
776 | filp = fget_light(fd, &fput_needed); | |
777 | if (unlikely(!filp)) | |
778 | return -EBADF; | |
779 | ||
780 | /* verify that this is indeed an inotify instance */ | |
781 | if (unlikely(filp->f_op != &inotify_fops)) { | |
782 | ret = -EINVAL; | |
783 | goto fput_and_out; | |
784 | } | |
785 | ||
786 | if (!(mask & IN_DONT_FOLLOW)) | |
787 | flags |= LOOKUP_FOLLOW; | |
788 | if (mask & IN_ONLYDIR) | |
789 | flags |= LOOKUP_DIRECTORY; | |
790 | ||
791 | ret = inotify_find_inode(pathname, &path, flags); | |
792 | if (ret) | |
793 | goto fput_and_out; | |
794 | ||
795 | /* inode held in place by reference to path; group by fget on fd */ | |
796 | inode = path.dentry->d_inode; | |
797 | group = filp->private_data; | |
798 | ||
799 | /* create/update an inode mark */ | |
800 | ret = inotify_update_watch(group, inode, mask); | |
801 | path_put(&path); | |
802 | fput_and_out: | |
803 | fput_light(filp, fput_needed); | |
804 | return ret; | |
805 | } | |
806 | ||
807 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) | |
808 | { | |
809 | struct fsnotify_group *group; | |
810 | struct inotify_inode_mark *i_mark; | |
811 | struct file *filp; | |
812 | int ret = 0, fput_needed; | |
813 | ||
814 | filp = fget_light(fd, &fput_needed); | |
815 | if (unlikely(!filp)) | |
816 | return -EBADF; | |
817 | ||
818 | /* verify that this is indeed an inotify instance */ | |
819 | ret = -EINVAL; | |
820 | if (unlikely(filp->f_op != &inotify_fops)) | |
821 | goto out; | |
822 | ||
823 | group = filp->private_data; | |
824 | ||
825 | ret = -EINVAL; | |
826 | i_mark = inotify_idr_find(group, wd); | |
827 | if (unlikely(!i_mark)) | |
828 | goto out; | |
829 | ||
830 | ret = 0; | |
831 | ||
832 | fsnotify_destroy_mark(&i_mark->fsn_mark); | |
833 | ||
834 | /* match ref taken by inotify_idr_find */ | |
835 | fsnotify_put_mark(&i_mark->fsn_mark); | |
836 | ||
837 | out: | |
838 | fput_light(filp, fput_needed); | |
839 | return ret; | |
840 | } | |
841 | ||
842 | /* | |
843 | * inotify_user_setup - Our initialization function. Note that we cannnot return | |
844 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here | |
845 | * must result in panic(). | |
846 | */ | |
847 | static int __init inotify_user_setup(void) | |
848 | { | |
849 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); | |
850 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
851 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
852 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
853 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
854 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
855 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
856 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
857 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
858 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
859 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
860 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
861 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
862 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
863 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
864 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
865 | BUILD_BUG_ON(IN_ISDIR != FS_IN_ISDIR); | |
866 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); | |
867 | ||
868 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
869 | ||
870 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); | |
871 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); | |
872 | ||
873 | inotify_max_queued_events = 16384; | |
874 | inotify_max_user_instances = 128; | |
875 | inotify_max_user_watches = 8192; | |
876 | ||
877 | return 0; | |
878 | } | |
879 | module_init(inotify_user_setup); |