xref: /openbmc/linux/fs/notify/inotify/inotify_user.c (revision 799a545b)
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> /* fs_initcall */
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_FUNC(wait, woken_wake_function);
231 
232 	start = buf;
233 	group = file->private_data;
234 
235 	add_wait_queue(&group->notification_waitq, &wait);
236 	while (1) {
237 		mutex_lock(&group->notification_mutex);
238 		kevent = get_one_event(group, count);
239 		mutex_unlock(&group->notification_mutex);
240 
241 		pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
242 
243 		if (kevent) {
244 			ret = PTR_ERR(kevent);
245 			if (IS_ERR(kevent))
246 				break;
247 			ret = copy_event_to_user(group, kevent, buf);
248 			fsnotify_destroy_event(group, kevent);
249 			if (ret < 0)
250 				break;
251 			buf += ret;
252 			count -= ret;
253 			continue;
254 		}
255 
256 		ret = -EAGAIN;
257 		if (file->f_flags & O_NONBLOCK)
258 			break;
259 		ret = -ERESTARTSYS;
260 		if (signal_pending(current))
261 			break;
262 
263 		if (start != buf)
264 			break;
265 
266 		wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
267 	}
268 	remove_wait_queue(&group->notification_waitq, &wait);
269 
270 	if (start != buf && ret != -EFAULT)
271 		ret = buf - start;
272 	return ret;
273 }
274 
275 static int inotify_release(struct inode *ignored, struct file *file)
276 {
277 	struct fsnotify_group *group = file->private_data;
278 
279 	pr_debug("%s: group=%p\n", __func__, group);
280 
281 	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
282 	fsnotify_destroy_group(group);
283 
284 	return 0;
285 }
286 
287 static long inotify_ioctl(struct file *file, unsigned int cmd,
288 			  unsigned long arg)
289 {
290 	struct fsnotify_group *group;
291 	struct fsnotify_event *fsn_event;
292 	void __user *p;
293 	int ret = -ENOTTY;
294 	size_t send_len = 0;
295 
296 	group = file->private_data;
297 	p = (void __user *) arg;
298 
299 	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
300 
301 	switch (cmd) {
302 	case FIONREAD:
303 		mutex_lock(&group->notification_mutex);
304 		list_for_each_entry(fsn_event, &group->notification_list,
305 				    list) {
306 			send_len += sizeof(struct inotify_event);
307 			send_len += round_event_name_len(fsn_event);
308 		}
309 		mutex_unlock(&group->notification_mutex);
310 		ret = put_user(send_len, (int __user *) p);
311 		break;
312 	}
313 
314 	return ret;
315 }
316 
317 static const struct file_operations inotify_fops = {
318 	.show_fdinfo	= inotify_show_fdinfo,
319 	.poll		= inotify_poll,
320 	.read		= inotify_read,
321 	.fasync		= fsnotify_fasync,
322 	.release	= inotify_release,
323 	.unlocked_ioctl	= inotify_ioctl,
324 	.compat_ioctl	= inotify_ioctl,
325 	.llseek		= noop_llseek,
326 };
327 
328 
329 /*
330  * find_inode - resolve a user-given path to a specific inode
331  */
332 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
333 {
334 	int error;
335 
336 	error = user_path_at(AT_FDCWD, dirname, flags, path);
337 	if (error)
338 		return error;
339 	/* you can only watch an inode if you have read permissions on it */
340 	error = inode_permission(path->dentry->d_inode, MAY_READ);
341 	if (error)
342 		path_put(path);
343 	return error;
344 }
345 
346 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
347 			      struct inotify_inode_mark *i_mark)
348 {
349 	int ret;
350 
351 	idr_preload(GFP_KERNEL);
352 	spin_lock(idr_lock);
353 
354 	ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
355 	if (ret >= 0) {
356 		/* we added the mark to the idr, take a reference */
357 		i_mark->wd = ret;
358 		fsnotify_get_mark(&i_mark->fsn_mark);
359 	}
360 
361 	spin_unlock(idr_lock);
362 	idr_preload_end();
363 	return ret < 0 ? ret : 0;
364 }
365 
366 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
367 								int wd)
368 {
369 	struct idr *idr = &group->inotify_data.idr;
370 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
371 	struct inotify_inode_mark *i_mark;
372 
373 	assert_spin_locked(idr_lock);
374 
375 	i_mark = idr_find(idr, wd);
376 	if (i_mark) {
377 		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
378 
379 		fsnotify_get_mark(fsn_mark);
380 		/* One ref for being in the idr, one ref we just took */
381 		BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
382 	}
383 
384 	return i_mark;
385 }
386 
387 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
388 							 int wd)
389 {
390 	struct inotify_inode_mark *i_mark;
391 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
392 
393 	spin_lock(idr_lock);
394 	i_mark = inotify_idr_find_locked(group, wd);
395 	spin_unlock(idr_lock);
396 
397 	return i_mark;
398 }
399 
400 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
401 				       struct inotify_inode_mark *i_mark)
402 {
403 	struct idr *idr = &group->inotify_data.idr;
404 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
405 	int wd = i_mark->wd;
406 
407 	assert_spin_locked(idr_lock);
408 
409 	idr_remove(idr, wd);
410 
411 	/* removed from the idr, drop that ref */
412 	fsnotify_put_mark(&i_mark->fsn_mark);
413 }
414 
415 /*
416  * Remove the mark from the idr (if present) and drop the reference
417  * on the mark because it was in the idr.
418  */
419 static void inotify_remove_from_idr(struct fsnotify_group *group,
420 				    struct inotify_inode_mark *i_mark)
421 {
422 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
423 	struct inotify_inode_mark *found_i_mark = NULL;
424 	int wd;
425 
426 	spin_lock(idr_lock);
427 	wd = i_mark->wd;
428 
429 	/*
430 	 * does this i_mark think it is in the idr?  we shouldn't get called
431 	 * if it wasn't....
432 	 */
433 	if (wd == -1) {
434 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
435 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
436 			i_mark->fsn_mark.group, i_mark->fsn_mark.inode);
437 		goto out;
438 	}
439 
440 	/* Lets look in the idr to see if we find it */
441 	found_i_mark = inotify_idr_find_locked(group, wd);
442 	if (unlikely(!found_i_mark)) {
443 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
444 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
445 			i_mark->fsn_mark.group, i_mark->fsn_mark.inode);
446 		goto out;
447 	}
448 
449 	/*
450 	 * We found an mark in the idr at the right wd, but it's
451 	 * not the mark we were told to remove.  eparis seriously
452 	 * fucked up somewhere.
453 	 */
454 	if (unlikely(found_i_mark != i_mark)) {
455 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
456 			"mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
457 			"found_i_mark->group=%p found_i_mark->inode=%p\n",
458 			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
459 			i_mark->fsn_mark.inode, found_i_mark, found_i_mark->wd,
460 			found_i_mark->fsn_mark.group,
461 			found_i_mark->fsn_mark.inode);
462 		goto out;
463 	}
464 
465 	/*
466 	 * One ref for being in the idr
467 	 * one ref held by the caller trying to kill us
468 	 * one ref grabbed by inotify_idr_find
469 	 */
470 	if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
471 		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
472 			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
473 			i_mark->fsn_mark.group, i_mark->fsn_mark.inode);
474 		/* we can't really recover with bad ref cnting.. */
475 		BUG();
476 	}
477 
478 	do_inotify_remove_from_idr(group, i_mark);
479 out:
480 	/* match the ref taken by inotify_idr_find_locked() */
481 	if (found_i_mark)
482 		fsnotify_put_mark(&found_i_mark->fsn_mark);
483 	i_mark->wd = -1;
484 	spin_unlock(idr_lock);
485 }
486 
487 /*
488  * Send IN_IGNORED for this wd, remove this wd from the idr.
489  */
490 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
491 				    struct fsnotify_group *group)
492 {
493 	struct inotify_inode_mark *i_mark;
494 
495 	/* Queue ignore event for the watch */
496 	inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED,
497 			     NULL, FSNOTIFY_EVENT_NONE, NULL, 0);
498 
499 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
500 	/* remove this mark from the idr */
501 	inotify_remove_from_idr(group, i_mark);
502 
503 	atomic_dec(&group->inotify_data.user->inotify_watches);
504 }
505 
506 /* ding dong the mark is dead */
507 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
508 {
509 	struct inotify_inode_mark *i_mark;
510 
511 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
512 
513 	kmem_cache_free(inotify_inode_mark_cachep, i_mark);
514 }
515 
516 static int inotify_update_existing_watch(struct fsnotify_group *group,
517 					 struct inode *inode,
518 					 u32 arg)
519 {
520 	struct fsnotify_mark *fsn_mark;
521 	struct inotify_inode_mark *i_mark;
522 	__u32 old_mask, new_mask;
523 	__u32 mask;
524 	int add = (arg & IN_MASK_ADD);
525 	int ret;
526 
527 	mask = inotify_arg_to_mask(arg);
528 
529 	fsn_mark = fsnotify_find_inode_mark(group, inode);
530 	if (!fsn_mark)
531 		return -ENOENT;
532 
533 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
534 
535 	spin_lock(&fsn_mark->lock);
536 
537 	old_mask = fsn_mark->mask;
538 	if (add)
539 		fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
540 	else
541 		fsnotify_set_mark_mask_locked(fsn_mark, mask);
542 	new_mask = fsn_mark->mask;
543 
544 	spin_unlock(&fsn_mark->lock);
545 
546 	if (old_mask != new_mask) {
547 		/* more bits in old than in new? */
548 		int dropped = (old_mask & ~new_mask);
549 		/* more bits in this fsn_mark than the inode's mask? */
550 		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
551 
552 		/* update the inode with this new fsn_mark */
553 		if (dropped || do_inode)
554 			fsnotify_recalc_inode_mask(inode);
555 
556 	}
557 
558 	/* return the wd */
559 	ret = i_mark->wd;
560 
561 	/* match the get from fsnotify_find_mark() */
562 	fsnotify_put_mark(fsn_mark);
563 
564 	return ret;
565 }
566 
567 static int inotify_new_watch(struct fsnotify_group *group,
568 			     struct inode *inode,
569 			     u32 arg)
570 {
571 	struct inotify_inode_mark *tmp_i_mark;
572 	__u32 mask;
573 	int ret;
574 	struct idr *idr = &group->inotify_data.idr;
575 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
576 
577 	mask = inotify_arg_to_mask(arg);
578 
579 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
580 	if (unlikely(!tmp_i_mark))
581 		return -ENOMEM;
582 
583 	fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
584 	tmp_i_mark->fsn_mark.mask = mask;
585 	tmp_i_mark->wd = -1;
586 
587 	ret = -ENOSPC;
588 	if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
589 		goto out_err;
590 
591 	ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
592 	if (ret)
593 		goto out_err;
594 
595 	/* we are on the idr, now get on the inode */
596 	ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, group, inode,
597 				       NULL, 0);
598 	if (ret) {
599 		/* we failed to get on the inode, get off the idr */
600 		inotify_remove_from_idr(group, tmp_i_mark);
601 		goto out_err;
602 	}
603 
604 	/* increment the number of watches the user has */
605 	atomic_inc(&group->inotify_data.user->inotify_watches);
606 
607 	/* return the watch descriptor for this new mark */
608 	ret = tmp_i_mark->wd;
609 
610 out_err:
611 	/* match the ref from fsnotify_init_mark() */
612 	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
613 
614 	return ret;
615 }
616 
617 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
618 {
619 	int ret = 0;
620 
621 	mutex_lock(&group->mark_mutex);
622 	/* try to update and existing watch with the new arg */
623 	ret = inotify_update_existing_watch(group, inode, arg);
624 	/* no mark present, try to add a new one */
625 	if (ret == -ENOENT)
626 		ret = inotify_new_watch(group, inode, arg);
627 	mutex_unlock(&group->mark_mutex);
628 
629 	return ret;
630 }
631 
632 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
633 {
634 	struct fsnotify_group *group;
635 	struct inotify_event_info *oevent;
636 
637 	group = fsnotify_alloc_group(&inotify_fsnotify_ops);
638 	if (IS_ERR(group))
639 		return group;
640 
641 	oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
642 	if (unlikely(!oevent)) {
643 		fsnotify_destroy_group(group);
644 		return ERR_PTR(-ENOMEM);
645 	}
646 	group->overflow_event = &oevent->fse;
647 	fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW);
648 	oevent->wd = -1;
649 	oevent->sync_cookie = 0;
650 	oevent->name_len = 0;
651 
652 	group->max_events = max_events;
653 
654 	spin_lock_init(&group->inotify_data.idr_lock);
655 	idr_init(&group->inotify_data.idr);
656 	group->inotify_data.user = get_current_user();
657 
658 	if (atomic_inc_return(&group->inotify_data.user->inotify_devs) >
659 	    inotify_max_user_instances) {
660 		fsnotify_destroy_group(group);
661 		return ERR_PTR(-EMFILE);
662 	}
663 
664 	return group;
665 }
666 
667 
668 /* inotify syscalls */
669 SYSCALL_DEFINE1(inotify_init1, int, flags)
670 {
671 	struct fsnotify_group *group;
672 	int ret;
673 
674 	/* Check the IN_* constants for consistency.  */
675 	BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
676 	BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
677 
678 	if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
679 		return -EINVAL;
680 
681 	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
682 	group = inotify_new_group(inotify_max_queued_events);
683 	if (IS_ERR(group))
684 		return PTR_ERR(group);
685 
686 	ret = anon_inode_getfd("inotify", &inotify_fops, group,
687 				  O_RDONLY | flags);
688 	if (ret < 0)
689 		fsnotify_destroy_group(group);
690 
691 	return ret;
692 }
693 
694 SYSCALL_DEFINE0(inotify_init)
695 {
696 	return sys_inotify_init1(0);
697 }
698 
699 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
700 		u32, mask)
701 {
702 	struct fsnotify_group *group;
703 	struct inode *inode;
704 	struct path path;
705 	struct fd f;
706 	int ret;
707 	unsigned flags = 0;
708 
709 	/*
710 	 * We share a lot of code with fs/dnotify.  We also share
711 	 * the bit layout between inotify's IN_* and the fsnotify
712 	 * FS_*.  This check ensures that only the inotify IN_*
713 	 * bits get passed in and set in watches/events.
714 	 */
715 	if (unlikely(mask & ~ALL_INOTIFY_BITS))
716 		return -EINVAL;
717 	/*
718 	 * Require at least one valid bit set in the mask.
719 	 * Without _something_ set, we would have no events to
720 	 * watch for.
721 	 */
722 	if (unlikely(!(mask & ALL_INOTIFY_BITS)))
723 		return -EINVAL;
724 
725 	f = fdget(fd);
726 	if (unlikely(!f.file))
727 		return -EBADF;
728 
729 	/* verify that this is indeed an inotify instance */
730 	if (unlikely(f.file->f_op != &inotify_fops)) {
731 		ret = -EINVAL;
732 		goto fput_and_out;
733 	}
734 
735 	if (!(mask & IN_DONT_FOLLOW))
736 		flags |= LOOKUP_FOLLOW;
737 	if (mask & IN_ONLYDIR)
738 		flags |= LOOKUP_DIRECTORY;
739 
740 	ret = inotify_find_inode(pathname, &path, flags);
741 	if (ret)
742 		goto fput_and_out;
743 
744 	/* inode held in place by reference to path; group by fget on fd */
745 	inode = path.dentry->d_inode;
746 	group = f.file->private_data;
747 
748 	/* create/update an inode mark */
749 	ret = inotify_update_watch(group, inode, mask);
750 	path_put(&path);
751 fput_and_out:
752 	fdput(f);
753 	return ret;
754 }
755 
756 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
757 {
758 	struct fsnotify_group *group;
759 	struct inotify_inode_mark *i_mark;
760 	struct fd f;
761 	int ret = 0;
762 
763 	f = fdget(fd);
764 	if (unlikely(!f.file))
765 		return -EBADF;
766 
767 	/* verify that this is indeed an inotify instance */
768 	ret = -EINVAL;
769 	if (unlikely(f.file->f_op != &inotify_fops))
770 		goto out;
771 
772 	group = f.file->private_data;
773 
774 	ret = -EINVAL;
775 	i_mark = inotify_idr_find(group, wd);
776 	if (unlikely(!i_mark))
777 		goto out;
778 
779 	ret = 0;
780 
781 	fsnotify_destroy_mark(&i_mark->fsn_mark, group);
782 
783 	/* match ref taken by inotify_idr_find */
784 	fsnotify_put_mark(&i_mark->fsn_mark);
785 
786 out:
787 	fdput(f);
788 	return ret;
789 }
790 
791 /*
792  * inotify_user_setup - Our initialization function.  Note that we cannot return
793  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
794  * must result in panic().
795  */
796 static int __init inotify_user_setup(void)
797 {
798 	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
799 	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
800 	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
801 	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
802 	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
803 	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
804 	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
805 	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
806 	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
807 	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
808 	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
809 	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
810 	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
811 	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
812 	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
813 	BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
814 	BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
815 	BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
816 
817 	BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
818 
819 	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
820 
821 	inotify_max_queued_events = 16384;
822 	inotify_max_user_instances = 128;
823 	inotify_max_user_watches = 8192;
824 
825 	return 0;
826 }
827 fs_initcall(inotify_user_setup);
828