xref: /openbmc/linux/fs/notify/inotify/inotify_user.c (revision 588b48ca)
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 #include "../fdinfo.h"
44 
45 #include <asm/ioctls.h>
46 
47 /* these are configurable via /proc/sys/fs/inotify/ */
48 static int inotify_max_user_instances __read_mostly;
49 static int inotify_max_queued_events __read_mostly;
50 static int inotify_max_user_watches __read_mostly;
51 
52 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
53 
54 #ifdef CONFIG_SYSCTL
55 
56 #include <linux/sysctl.h>
57 
58 static int zero;
59 
60 struct 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 static int round_event_name_len(struct fsnotify_event *fsn_event)
127 {
128 	struct inotify_event_info *event;
129 
130 	event = INOTIFY_E(fsn_event);
131 	if (!event->name_len)
132 		return 0;
133 	return roundup(event->name_len + 1, sizeof(struct inotify_event));
134 }
135 
136 /*
137  * Get an inotify_kernel_event if one exists and is small
138  * enough to fit in "count". Return an error pointer if
139  * not large enough.
140  *
141  * Called with the group->notification_mutex held.
142  */
143 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
144 					    size_t count)
145 {
146 	size_t event_size = sizeof(struct inotify_event);
147 	struct fsnotify_event *event;
148 
149 	if (fsnotify_notify_queue_is_empty(group))
150 		return NULL;
151 
152 	event = fsnotify_peek_first_event(group);
153 
154 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
155 
156 	event_size += round_event_name_len(event);
157 	if (event_size > count)
158 		return ERR_PTR(-EINVAL);
159 
160 	/* held the notification_mutex the whole time, so this is the
161 	 * same event we peeked above */
162 	fsnotify_remove_first_event(group);
163 
164 	return event;
165 }
166 
167 /*
168  * Copy an event to user space, returning how much we copied.
169  *
170  * We already checked that the event size is smaller than the
171  * buffer we had in "get_one_event()" above.
172  */
173 static ssize_t copy_event_to_user(struct fsnotify_group *group,
174 				  struct fsnotify_event *fsn_event,
175 				  char __user *buf)
176 {
177 	struct inotify_event inotify_event;
178 	struct inotify_event_info *event;
179 	size_t event_size = sizeof(struct inotify_event);
180 	size_t name_len;
181 	size_t pad_name_len;
182 
183 	pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
184 
185 	event = INOTIFY_E(fsn_event);
186 	name_len = event->name_len;
187 	/*
188 	 * round up name length so it is a multiple of event_size
189 	 * plus an extra byte for the terminating '\0'.
190 	 */
191 	pad_name_len = round_event_name_len(fsn_event);
192 	inotify_event.len = pad_name_len;
193 	inotify_event.mask = inotify_mask_to_arg(fsn_event->mask);
194 	inotify_event.wd = event->wd;
195 	inotify_event.cookie = event->sync_cookie;
196 
197 	/* send the main event */
198 	if (copy_to_user(buf, &inotify_event, event_size))
199 		return -EFAULT;
200 
201 	buf += event_size;
202 
203 	/*
204 	 * fsnotify only stores the pathname, so here we have to send the pathname
205 	 * and then pad that pathname out to a multiple of sizeof(inotify_event)
206 	 * with zeros.
207 	 */
208 	if (pad_name_len) {
209 		/* copy the path name */
210 		if (copy_to_user(buf, event->name, name_len))
211 			return -EFAULT;
212 		buf += name_len;
213 
214 		/* fill userspace with 0's */
215 		if (clear_user(buf, pad_name_len - name_len))
216 			return -EFAULT;
217 		event_size += pad_name_len;
218 	}
219 
220 	return event_size;
221 }
222 
223 static ssize_t inotify_read(struct file *file, char __user *buf,
224 			    size_t count, loff_t *pos)
225 {
226 	struct fsnotify_group *group;
227 	struct fsnotify_event *kevent;
228 	char __user *start;
229 	int ret;
230 	DEFINE_WAIT(wait);
231 
232 	start = buf;
233 	group = file->private_data;
234 
235 	while (1) {
236 		prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
237 
238 		mutex_lock(&group->notification_mutex);
239 		kevent = get_one_event(group, count);
240 		mutex_unlock(&group->notification_mutex);
241 
242 		pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
243 
244 		if (kevent) {
245 			ret = PTR_ERR(kevent);
246 			if (IS_ERR(kevent))
247 				break;
248 			ret = copy_event_to_user(group, kevent, buf);
249 			fsnotify_destroy_event(group, kevent);
250 			if (ret < 0)
251 				break;
252 			buf += ret;
253 			count -= ret;
254 			continue;
255 		}
256 
257 		ret = -EAGAIN;
258 		if (file->f_flags & O_NONBLOCK)
259 			break;
260 		ret = -ERESTARTSYS;
261 		if (signal_pending(current))
262 			break;
263 
264 		if (start != buf)
265 			break;
266 
267 		schedule();
268 	}
269 
270 	finish_wait(&group->notification_waitq, &wait);
271 	if (start != buf && ret != -EFAULT)
272 		ret = buf - start;
273 	return ret;
274 }
275 
276 static int inotify_release(struct inode *ignored, struct file *file)
277 {
278 	struct fsnotify_group *group = file->private_data;
279 
280 	pr_debug("%s: group=%p\n", __func__, group);
281 
282 	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
283 	fsnotify_destroy_group(group);
284 
285 	return 0;
286 }
287 
288 static long inotify_ioctl(struct file *file, unsigned int cmd,
289 			  unsigned long arg)
290 {
291 	struct fsnotify_group *group;
292 	struct fsnotify_event *fsn_event;
293 	void __user *p;
294 	int ret = -ENOTTY;
295 	size_t send_len = 0;
296 
297 	group = file->private_data;
298 	p = (void __user *) arg;
299 
300 	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
301 
302 	switch (cmd) {
303 	case FIONREAD:
304 		mutex_lock(&group->notification_mutex);
305 		list_for_each_entry(fsn_event, &group->notification_list,
306 				    list) {
307 			send_len += sizeof(struct inotify_event);
308 			send_len += round_event_name_len(fsn_event);
309 		}
310 		mutex_unlock(&group->notification_mutex);
311 		ret = put_user(send_len, (int __user *) p);
312 		break;
313 	}
314 
315 	return ret;
316 }
317 
318 static const struct file_operations inotify_fops = {
319 	.show_fdinfo	= inotify_show_fdinfo,
320 	.poll		= inotify_poll,
321 	.read		= inotify_read,
322 	.fasync		= fsnotify_fasync,
323 	.release	= inotify_release,
324 	.unlocked_ioctl	= inotify_ioctl,
325 	.compat_ioctl	= inotify_ioctl,
326 	.llseek		= noop_llseek,
327 };
328 
329 
330 /*
331  * find_inode - resolve a user-given path to a specific inode
332  */
333 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
334 {
335 	int error;
336 
337 	error = user_path_at(AT_FDCWD, dirname, flags, path);
338 	if (error)
339 		return error;
340 	/* you can only watch an inode if you have read permissions on it */
341 	error = inode_permission(path->dentry->d_inode, MAY_READ);
342 	if (error)
343 		path_put(path);
344 	return error;
345 }
346 
347 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
348 			      struct inotify_inode_mark *i_mark)
349 {
350 	int ret;
351 
352 	idr_preload(GFP_KERNEL);
353 	spin_lock(idr_lock);
354 
355 	ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
356 	if (ret >= 0) {
357 		/* we added the mark to the idr, take a reference */
358 		i_mark->wd = ret;
359 		fsnotify_get_mark(&i_mark->fsn_mark);
360 	}
361 
362 	spin_unlock(idr_lock);
363 	idr_preload_end();
364 	return ret < 0 ? ret : 0;
365 }
366 
367 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
368 								int wd)
369 {
370 	struct idr *idr = &group->inotify_data.idr;
371 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
372 	struct inotify_inode_mark *i_mark;
373 
374 	assert_spin_locked(idr_lock);
375 
376 	i_mark = idr_find(idr, wd);
377 	if (i_mark) {
378 		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
379 
380 		fsnotify_get_mark(fsn_mark);
381 		/* One ref for being in the idr, one ref we just took */
382 		BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
383 	}
384 
385 	return i_mark;
386 }
387 
388 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
389 							 int wd)
390 {
391 	struct inotify_inode_mark *i_mark;
392 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
393 
394 	spin_lock(idr_lock);
395 	i_mark = inotify_idr_find_locked(group, wd);
396 	spin_unlock(idr_lock);
397 
398 	return i_mark;
399 }
400 
401 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
402 				       struct inotify_inode_mark *i_mark)
403 {
404 	struct idr *idr = &group->inotify_data.idr;
405 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
406 	int wd = i_mark->wd;
407 
408 	assert_spin_locked(idr_lock);
409 
410 	idr_remove(idr, wd);
411 
412 	/* removed from the idr, drop that ref */
413 	fsnotify_put_mark(&i_mark->fsn_mark);
414 }
415 
416 /*
417  * Remove the mark from the idr (if present) and drop the reference
418  * on the mark because it was in the idr.
419  */
420 static void inotify_remove_from_idr(struct fsnotify_group *group,
421 				    struct inotify_inode_mark *i_mark)
422 {
423 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
424 	struct inotify_inode_mark *found_i_mark = NULL;
425 	int wd;
426 
427 	spin_lock(idr_lock);
428 	wd = i_mark->wd;
429 
430 	/*
431 	 * does this i_mark think it is in the idr?  we shouldn't get called
432 	 * if it wasn't....
433 	 */
434 	if (wd == -1) {
435 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
436 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
437 			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
438 		goto out;
439 	}
440 
441 	/* Lets look in the idr to see if we find it */
442 	found_i_mark = inotify_idr_find_locked(group, wd);
443 	if (unlikely(!found_i_mark)) {
444 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
445 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
446 			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
447 		goto out;
448 	}
449 
450 	/*
451 	 * We found an mark in the idr at the right wd, but it's
452 	 * not the mark we were told to remove.  eparis seriously
453 	 * fucked up somewhere.
454 	 */
455 	if (unlikely(found_i_mark != i_mark)) {
456 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
457 			"mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
458 			"found_i_mark->group=%p found_i_mark->inode=%p\n",
459 			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
460 			i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
461 			found_i_mark->fsn_mark.group,
462 			found_i_mark->fsn_mark.i.inode);
463 		goto out;
464 	}
465 
466 	/*
467 	 * One ref for being in the idr
468 	 * one ref held by the caller trying to kill us
469 	 * one ref grabbed by inotify_idr_find
470 	 */
471 	if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
472 		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
473 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
474 			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
475 		/* we can't really recover with bad ref cnting.. */
476 		BUG();
477 	}
478 
479 	do_inotify_remove_from_idr(group, i_mark);
480 out:
481 	/* match the ref taken by inotify_idr_find_locked() */
482 	if (found_i_mark)
483 		fsnotify_put_mark(&found_i_mark->fsn_mark);
484 	i_mark->wd = -1;
485 	spin_unlock(idr_lock);
486 }
487 
488 /*
489  * Send IN_IGNORED for this wd, remove this wd from the idr.
490  */
491 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
492 				    struct fsnotify_group *group)
493 {
494 	struct inotify_inode_mark *i_mark;
495 
496 	/* Queue ignore event for the watch */
497 	inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED,
498 			     NULL, FSNOTIFY_EVENT_NONE, NULL, 0);
499 
500 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
501 	/* remove this mark from the idr */
502 	inotify_remove_from_idr(group, i_mark);
503 
504 	atomic_dec(&group->inotify_data.user->inotify_watches);
505 }
506 
507 /* ding dong the mark is dead */
508 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
509 {
510 	struct inotify_inode_mark *i_mark;
511 
512 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
513 
514 	kmem_cache_free(inotify_inode_mark_cachep, i_mark);
515 }
516 
517 static int inotify_update_existing_watch(struct fsnotify_group *group,
518 					 struct inode *inode,
519 					 u32 arg)
520 {
521 	struct fsnotify_mark *fsn_mark;
522 	struct inotify_inode_mark *i_mark;
523 	__u32 old_mask, new_mask;
524 	__u32 mask;
525 	int add = (arg & IN_MASK_ADD);
526 	int ret;
527 
528 	mask = inotify_arg_to_mask(arg);
529 
530 	fsn_mark = fsnotify_find_inode_mark(group, inode);
531 	if (!fsn_mark)
532 		return -ENOENT;
533 
534 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
535 
536 	spin_lock(&fsn_mark->lock);
537 
538 	old_mask = fsn_mark->mask;
539 	if (add)
540 		fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
541 	else
542 		fsnotify_set_mark_mask_locked(fsn_mark, mask);
543 	new_mask = fsn_mark->mask;
544 
545 	spin_unlock(&fsn_mark->lock);
546 
547 	if (old_mask != new_mask) {
548 		/* more bits in old than in new? */
549 		int dropped = (old_mask & ~new_mask);
550 		/* more bits in this fsn_mark than the inode's mask? */
551 		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
552 
553 		/* update the inode with this new fsn_mark */
554 		if (dropped || do_inode)
555 			fsnotify_recalc_inode_mask(inode);
556 
557 	}
558 
559 	/* return the wd */
560 	ret = i_mark->wd;
561 
562 	/* match the get from fsnotify_find_mark() */
563 	fsnotify_put_mark(fsn_mark);
564 
565 	return ret;
566 }
567 
568 static int inotify_new_watch(struct fsnotify_group *group,
569 			     struct inode *inode,
570 			     u32 arg)
571 {
572 	struct inotify_inode_mark *tmp_i_mark;
573 	__u32 mask;
574 	int ret;
575 	struct idr *idr = &group->inotify_data.idr;
576 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
577 
578 	mask = inotify_arg_to_mask(arg);
579 
580 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
581 	if (unlikely(!tmp_i_mark))
582 		return -ENOMEM;
583 
584 	fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
585 	tmp_i_mark->fsn_mark.mask = mask;
586 	tmp_i_mark->wd = -1;
587 
588 	ret = -ENOSPC;
589 	if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
590 		goto out_err;
591 
592 	ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
593 	if (ret)
594 		goto out_err;
595 
596 	/* we are on the idr, now get on the inode */
597 	ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, group, inode,
598 				       NULL, 0);
599 	if (ret) {
600 		/* we failed to get on the inode, get off the idr */
601 		inotify_remove_from_idr(group, tmp_i_mark);
602 		goto out_err;
603 	}
604 
605 	/* increment the number of watches the user has */
606 	atomic_inc(&group->inotify_data.user->inotify_watches);
607 
608 	/* return the watch descriptor for this new mark */
609 	ret = tmp_i_mark->wd;
610 
611 out_err:
612 	/* match the ref from fsnotify_init_mark() */
613 	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
614 
615 	return ret;
616 }
617 
618 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
619 {
620 	int ret = 0;
621 
622 	mutex_lock(&group->mark_mutex);
623 	/* try to update and existing watch with the new arg */
624 	ret = inotify_update_existing_watch(group, inode, arg);
625 	/* no mark present, try to add a new one */
626 	if (ret == -ENOENT)
627 		ret = inotify_new_watch(group, inode, arg);
628 	mutex_unlock(&group->mark_mutex);
629 
630 	return ret;
631 }
632 
633 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
634 {
635 	struct fsnotify_group *group;
636 	struct inotify_event_info *oevent;
637 
638 	group = fsnotify_alloc_group(&inotify_fsnotify_ops);
639 	if (IS_ERR(group))
640 		return group;
641 
642 	oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
643 	if (unlikely(!oevent)) {
644 		fsnotify_destroy_group(group);
645 		return ERR_PTR(-ENOMEM);
646 	}
647 	group->overflow_event = &oevent->fse;
648 	fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW);
649 	oevent->wd = -1;
650 	oevent->sync_cookie = 0;
651 	oevent->name_len = 0;
652 
653 	group->max_events = max_events;
654 
655 	spin_lock_init(&group->inotify_data.idr_lock);
656 	idr_init(&group->inotify_data.idr);
657 	group->inotify_data.user = get_current_user();
658 
659 	if (atomic_inc_return(&group->inotify_data.user->inotify_devs) >
660 	    inotify_max_user_instances) {
661 		fsnotify_destroy_group(group);
662 		return ERR_PTR(-EMFILE);
663 	}
664 
665 	return group;
666 }
667 
668 
669 /* inotify syscalls */
670 SYSCALL_DEFINE1(inotify_init1, int, flags)
671 {
672 	struct fsnotify_group *group;
673 	int ret;
674 
675 	/* Check the IN_* constants for consistency.  */
676 	BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
677 	BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
678 
679 	if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
680 		return -EINVAL;
681 
682 	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
683 	group = inotify_new_group(inotify_max_queued_events);
684 	if (IS_ERR(group))
685 		return PTR_ERR(group);
686 
687 	ret = anon_inode_getfd("inotify", &inotify_fops, group,
688 				  O_RDONLY | flags);
689 	if (ret < 0)
690 		fsnotify_destroy_group(group);
691 
692 	return ret;
693 }
694 
695 SYSCALL_DEFINE0(inotify_init)
696 {
697 	return sys_inotify_init1(0);
698 }
699 
700 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
701 		u32, mask)
702 {
703 	struct fsnotify_group *group;
704 	struct inode *inode;
705 	struct path path;
706 	struct fd f;
707 	int ret;
708 	unsigned flags = 0;
709 
710 	/* don't allow invalid bits: we don't want flags set */
711 	if (unlikely(!(mask & ALL_INOTIFY_BITS)))
712 		return -EINVAL;
713 
714 	f = fdget(fd);
715 	if (unlikely(!f.file))
716 		return -EBADF;
717 
718 	/* verify that this is indeed an inotify instance */
719 	if (unlikely(f.file->f_op != &inotify_fops)) {
720 		ret = -EINVAL;
721 		goto fput_and_out;
722 	}
723 
724 	if (!(mask & IN_DONT_FOLLOW))
725 		flags |= LOOKUP_FOLLOW;
726 	if (mask & IN_ONLYDIR)
727 		flags |= LOOKUP_DIRECTORY;
728 
729 	ret = inotify_find_inode(pathname, &path, flags);
730 	if (ret)
731 		goto fput_and_out;
732 
733 	/* inode held in place by reference to path; group by fget on fd */
734 	inode = path.dentry->d_inode;
735 	group = f.file->private_data;
736 
737 	/* create/update an inode mark */
738 	ret = inotify_update_watch(group, inode, mask);
739 	path_put(&path);
740 fput_and_out:
741 	fdput(f);
742 	return ret;
743 }
744 
745 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
746 {
747 	struct fsnotify_group *group;
748 	struct inotify_inode_mark *i_mark;
749 	struct fd f;
750 	int ret = 0;
751 
752 	f = fdget(fd);
753 	if (unlikely(!f.file))
754 		return -EBADF;
755 
756 	/* verify that this is indeed an inotify instance */
757 	ret = -EINVAL;
758 	if (unlikely(f.file->f_op != &inotify_fops))
759 		goto out;
760 
761 	group = f.file->private_data;
762 
763 	ret = -EINVAL;
764 	i_mark = inotify_idr_find(group, wd);
765 	if (unlikely(!i_mark))
766 		goto out;
767 
768 	ret = 0;
769 
770 	fsnotify_destroy_mark(&i_mark->fsn_mark, group);
771 
772 	/* match ref taken by inotify_idr_find */
773 	fsnotify_put_mark(&i_mark->fsn_mark);
774 
775 out:
776 	fdput(f);
777 	return ret;
778 }
779 
780 /*
781  * inotify_user_setup - Our initialization function.  Note that we cannot return
782  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
783  * must result in panic().
784  */
785 static int __init inotify_user_setup(void)
786 {
787 	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
788 	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
789 	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
790 	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
791 	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
792 	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
793 	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
794 	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
795 	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
796 	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
797 	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
798 	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
799 	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
800 	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
801 	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
802 	BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
803 	BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
804 	BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
805 
806 	BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
807 
808 	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
809 
810 	inotify_max_queued_events = 16384;
811 	inotify_max_user_instances = 128;
812 	inotify_max_user_watches = 8192;
813 
814 	return 0;
815 }
816 module_init(inotify_user_setup);
817