Linux Audio

Check our new training course

Loading...
v4.17
  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/signal.h>
 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/* configurable via /proc/sys/fs/inotify/ */
 
 48static int inotify_max_queued_events __read_mostly;
 
 49
 50struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
 
 51
 52#ifdef CONFIG_SYSCTL
 53
 54#include <linux/sysctl.h>
 55
 56static int zero;
 57
 58struct ctl_table inotify_table[] = {
 59	{
 60		.procname	= "max_user_instances",
 61		.data		= &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
 62		.maxlen		= sizeof(int),
 63		.mode		= 0644,
 64		.proc_handler	= proc_dointvec_minmax,
 65		.extra1		= &zero,
 66	},
 67	{
 68		.procname	= "max_user_watches",
 69		.data		= &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
 70		.maxlen		= sizeof(int),
 71		.mode		= 0644,
 72		.proc_handler	= proc_dointvec_minmax,
 73		.extra1		= &zero,
 74	},
 75	{
 76		.procname	= "max_queued_events",
 77		.data		= &inotify_max_queued_events,
 78		.maxlen		= sizeof(int),
 79		.mode		= 0644,
 80		.proc_handler	= proc_dointvec_minmax,
 81		.extra1		= &zero
 82	},
 83	{ }
 84};
 85#endif /* CONFIG_SYSCTL */
 86
 87static inline __u32 inotify_arg_to_mask(u32 arg)
 88{
 89	__u32 mask;
 90
 91	/*
 92	 * everything should accept their own ignored, cares about children,
 93	 * and should receive events when the inode is unmounted
 94	 */
 95	mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
 96
 97	/* mask off the flags used to open the fd */
 98	mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
 99
100	return mask;
101}
102
103static inline u32 inotify_mask_to_arg(__u32 mask)
104{
105	return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
106		       IN_Q_OVERFLOW);
107}
108
109/* intofiy userspace file descriptor functions */
110static __poll_t inotify_poll(struct file *file, poll_table *wait)
111{
112	struct fsnotify_group *group = file->private_data;
113	__poll_t ret = 0;
114
115	poll_wait(file, &group->notification_waitq, wait);
116	spin_lock(&group->notification_lock);
117	if (!fsnotify_notify_queue_is_empty(group))
118		ret = EPOLLIN | EPOLLRDNORM;
119	spin_unlock(&group->notification_lock);
120
121	return ret;
122}
123
124static int round_event_name_len(struct fsnotify_event *fsn_event)
125{
126	struct inotify_event_info *event;
127
128	event = INOTIFY_E(fsn_event);
129	if (!event->name_len)
130		return 0;
131	return roundup(event->name_len + 1, sizeof(struct inotify_event));
132}
133
134/*
135 * Get an inotify_kernel_event if one exists and is small
136 * enough to fit in "count". Return an error pointer if
137 * not large enough.
138 *
139 * Called with the group->notification_lock held.
140 */
141static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
142					    size_t count)
143{
144	size_t event_size = sizeof(struct inotify_event);
145	struct fsnotify_event *event;
146
147	if (fsnotify_notify_queue_is_empty(group))
148		return NULL;
149
150	event = fsnotify_peek_first_event(group);
151
152	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
153
154	event_size += round_event_name_len(event);
 
 
155	if (event_size > count)
156		return ERR_PTR(-EINVAL);
157
158	/* held the notification_lock the whole time, so this is the
159	 * same event we peeked above */
160	fsnotify_remove_first_event(group);
161
162	return event;
163}
164
165/*
166 * Copy an event to user space, returning how much we copied.
167 *
168 * We already checked that the event size is smaller than the
169 * buffer we had in "get_one_event()" above.
170 */
171static ssize_t copy_event_to_user(struct fsnotify_group *group,
172				  struct fsnotify_event *fsn_event,
173				  char __user *buf)
174{
175	struct inotify_event inotify_event;
176	struct inotify_event_info *event;
 
177	size_t event_size = sizeof(struct inotify_event);
178	size_t name_len;
179	size_t pad_name_len;
180
181	pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
182
183	event = INOTIFY_E(fsn_event);
184	name_len = event->name_len;
185	/*
186	 * round up name length so it is a multiple of event_size
187	 * plus an extra byte for the terminating '\0'.
188	 */
189	pad_name_len = round_event_name_len(fsn_event);
190	inotify_event.len = pad_name_len;
191	inotify_event.mask = inotify_mask_to_arg(fsn_event->mask);
192	inotify_event.wd = event->wd;
 
193	inotify_event.cookie = event->sync_cookie;
194
195	/* send the main event */
196	if (copy_to_user(buf, &inotify_event, event_size))
197		return -EFAULT;
198
199	buf += event_size;
200
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.
205	 */
206	if (pad_name_len) {
 
207		/* copy the path name */
208		if (copy_to_user(buf, event->name, name_len))
209			return -EFAULT;
210		buf += name_len;
211
212		/* fill userspace with 0's */
213		if (clear_user(buf, pad_name_len - name_len))
214			return -EFAULT;
215		event_size += pad_name_len;
 
216	}
217
218	return event_size;
219}
220
221static ssize_t inotify_read(struct file *file, char __user *buf,
222			    size_t count, loff_t *pos)
223{
224	struct fsnotify_group *group;
225	struct fsnotify_event *kevent;
226	char __user *start;
227	int ret;
228	DEFINE_WAIT_FUNC(wait, woken_wake_function);
229
230	start = buf;
231	group = file->private_data;
232
233	add_wait_queue(&group->notification_waitq, &wait);
234	while (1) {
235		spin_lock(&group->notification_lock);
 
 
236		kevent = get_one_event(group, count);
237		spin_unlock(&group->notification_lock);
238
239		pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
240
241		if (kevent) {
242			ret = PTR_ERR(kevent);
243			if (IS_ERR(kevent))
244				break;
245			ret = copy_event_to_user(group, kevent, buf);
246			fsnotify_destroy_event(group, kevent);
247			if (ret < 0)
248				break;
249			buf += ret;
250			count -= ret;
251			continue;
252		}
253
254		ret = -EAGAIN;
255		if (file->f_flags & O_NONBLOCK)
256			break;
257		ret = -ERESTARTSYS;
258		if (signal_pending(current))
259			break;
260
261		if (start != buf)
262			break;
263
264		wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
265	}
266	remove_wait_queue(&group->notification_waitq, &wait);
267
 
268	if (start != buf && ret != -EFAULT)
269		ret = buf - start;
270	return ret;
271}
272
 
 
 
 
 
 
 
273static int inotify_release(struct inode *ignored, struct file *file)
274{
275	struct fsnotify_group *group = file->private_data;
276
277	pr_debug("%s: group=%p\n", __func__, group);
278
 
 
279	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
280	fsnotify_destroy_group(group);
281
282	return 0;
283}
284
285static long inotify_ioctl(struct file *file, unsigned int cmd,
286			  unsigned long arg)
287{
288	struct fsnotify_group *group;
289	struct fsnotify_event *fsn_event;
 
290	void __user *p;
291	int ret = -ENOTTY;
292	size_t send_len = 0;
293
294	group = file->private_data;
295	p = (void __user *) arg;
296
297	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
298
299	switch (cmd) {
300	case FIONREAD:
301		spin_lock(&group->notification_lock);
302		list_for_each_entry(fsn_event, &group->notification_list,
303				    list) {
304			send_len += sizeof(struct inotify_event);
305			send_len += round_event_name_len(fsn_event);
 
 
306		}
307		spin_unlock(&group->notification_lock);
308		ret = put_user(send_len, (int __user *) p);
309		break;
310#ifdef CONFIG_CHECKPOINT_RESTORE
311	case INOTIFY_IOC_SETNEXTWD:
312		ret = -EINVAL;
313		if (arg >= 1 && arg <= INT_MAX) {
314			struct inotify_group_private_data *data;
315
316			data = &group->inotify_data;
317			spin_lock(&data->idr_lock);
318			idr_set_cursor(&data->idr, (unsigned int)arg);
319			spin_unlock(&data->idr_lock);
320			ret = 0;
321		}
322		break;
323#endif /* CONFIG_CHECKPOINT_RESTORE */
324	}
325
326	return ret;
327}
328
329static const struct file_operations inotify_fops = {
330	.show_fdinfo	= inotify_show_fdinfo,
331	.poll		= inotify_poll,
332	.read		= inotify_read,
333	.fasync		= fsnotify_fasync,
334	.release	= inotify_release,
335	.unlocked_ioctl	= inotify_ioctl,
336	.compat_ioctl	= inotify_ioctl,
337	.llseek		= noop_llseek,
338};
339
340
341/*
342 * find_inode - resolve a user-given path to a specific inode
343 */
344static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
345{
346	int error;
347
348	error = user_path_at(AT_FDCWD, dirname, flags, path);
349	if (error)
350		return error;
351	/* you can only watch an inode if you have read permissions on it */
352	error = inode_permission(path->dentry->d_inode, MAY_READ);
353	if (error)
354		path_put(path);
355	return error;
356}
357
358static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
 
359			      struct inotify_inode_mark *i_mark)
360{
361	int ret;
362
363	idr_preload(GFP_KERNEL);
364	spin_lock(idr_lock);
365
366	ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
367	if (ret >= 0) {
 
 
368		/* we added the mark to the idr, take a reference */
369		i_mark->wd = ret;
370		fsnotify_get_mark(&i_mark->fsn_mark);
371	}
 
 
 
372
373	spin_unlock(idr_lock);
374	idr_preload_end();
375	return ret < 0 ? ret : 0;
376}
377
378static struct inotify_inode_mark *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 *i_mark;
384
385	assert_spin_locked(idr_lock);
386
387	i_mark = idr_find(idr, wd);
388	if (i_mark) {
389		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
390
391		fsnotify_get_mark(fsn_mark);
392		/* One ref for being in the idr, one ref we just took */
393		BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
394	}
395
396	return i_mark;
397}
398
399static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
400							 int wd)
401{
402	struct inotify_inode_mark *i_mark;
403	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
404
405	spin_lock(idr_lock);
406	i_mark = inotify_idr_find_locked(group, wd);
407	spin_unlock(idr_lock);
408
409	return i_mark;
410}
411
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
412/*
413 * Remove the mark from the idr (if present) and drop the reference
414 * on the mark because it was in the idr.
415 */
416static void inotify_remove_from_idr(struct fsnotify_group *group,
417				    struct inotify_inode_mark *i_mark)
418{
419	struct idr *idr = &group->inotify_data.idr;
420	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
421	struct inotify_inode_mark *found_i_mark = NULL;
422	int wd;
423
424	spin_lock(idr_lock);
425	wd = i_mark->wd;
426
427	/*
428	 * does this i_mark think it is in the idr?  we shouldn't get called
429	 * if it wasn't....
430	 */
431	if (wd == -1) {
432		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
433			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 
434		goto out;
435	}
436
437	/* Lets look in the idr to see if we find it */
438	found_i_mark = inotify_idr_find_locked(group, wd);
439	if (unlikely(!found_i_mark)) {
440		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
441			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 
442		goto out;
443	}
444
445	/*
446	 * We found an mark in the idr at the right wd, but it's
447	 * not the mark we were told to remove.  eparis seriously
448	 * fucked up somewhere.
449	 */
450	if (unlikely(found_i_mark != i_mark)) {
451		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
452			"found_i_mark=%p found_i_mark->wd=%d "
453			"found_i_mark->group=%p\n", __func__, i_mark,
454			i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
455			found_i_mark->wd, found_i_mark->fsn_mark.group);
 
 
456		goto out;
457	}
458
459	/*
460	 * One ref for being in the idr
 
461	 * one ref grabbed by inotify_idr_find
462	 */
463	if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
464		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
465			 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 
466		/* we can't really recover with bad ref cnting.. */
467		BUG();
468	}
469
470	idr_remove(idr, wd);
471	/* Removed from the idr, drop that ref. */
472	fsnotify_put_mark(&i_mark->fsn_mark);
473out:
474	i_mark->wd = -1;
475	spin_unlock(idr_lock);
476	/* match the ref taken by inotify_idr_find_locked() */
477	if (found_i_mark)
478		fsnotify_put_mark(&found_i_mark->fsn_mark);
 
 
479}
480
481/*
482 * Send IN_IGNORED for this wd, remove this wd from the idr.
483 */
484void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
485				    struct fsnotify_group *group)
486{
487	struct inotify_inode_mark *i_mark;
 
 
 
 
488
489	/* Queue ignore event for the watch */
490	inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED,
491			     NULL, FSNOTIFY_EVENT_NONE, NULL, 0, NULL);
 
 
492
493	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
494	/* remove this mark from the idr */
495	inotify_remove_from_idr(group, i_mark);
496
497	dec_inotify_watches(group->inotify_data.ucounts);
 
 
 
 
 
 
 
 
 
 
498}
499
500static int inotify_update_existing_watch(struct fsnotify_group *group,
501					 struct inode *inode,
502					 u32 arg)
503{
504	struct fsnotify_mark *fsn_mark;
505	struct inotify_inode_mark *i_mark;
506	__u32 old_mask, new_mask;
507	__u32 mask;
508	int add = (arg & IN_MASK_ADD);
509	int ret;
510
 
511	mask = inotify_arg_to_mask(arg);
 
 
512
513	fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
514	if (!fsn_mark)
515		return -ENOENT;
516
517	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
518
519	spin_lock(&fsn_mark->lock);
 
520	old_mask = fsn_mark->mask;
521	if (add)
522		fsn_mark->mask |= mask;
523	else
524		fsn_mark->mask = mask;
525	new_mask = fsn_mark->mask;
 
526	spin_unlock(&fsn_mark->lock);
527
528	if (old_mask != new_mask) {
529		/* more bits in old than in new? */
530		int dropped = (old_mask & ~new_mask);
531		/* more bits in this fsn_mark than the inode's mask? */
532		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
533
534		/* update the inode with this new fsn_mark */
535		if (dropped || do_inode)
536			fsnotify_recalc_mask(inode->i_fsnotify_marks);
537
538	}
539
540	/* return the wd */
541	ret = i_mark->wd;
542
543	/* match the get from fsnotify_find_mark() */
544	fsnotify_put_mark(fsn_mark);
545
546	return ret;
547}
548
549static int inotify_new_watch(struct fsnotify_group *group,
550			     struct inode *inode,
551			     u32 arg)
552{
553	struct inotify_inode_mark *tmp_i_mark;
554	__u32 mask;
555	int ret;
556	struct idr *idr = &group->inotify_data.idr;
557	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
558
 
559	mask = inotify_arg_to_mask(arg);
 
 
560
561	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
562	if (unlikely(!tmp_i_mark))
563		return -ENOMEM;
564
565	fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
566	tmp_i_mark->fsn_mark.mask = mask;
567	tmp_i_mark->wd = -1;
568
569	ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
570	if (ret)
571		goto out_err;
572
573	/* increment the number of watches the user has */
574	if (!inc_inotify_watches(group->inotify_data.ucounts)) {
575		inotify_remove_from_idr(group, tmp_i_mark);
576		ret = -ENOSPC;
577		goto out_err;
578	}
579
580	/* we are on the idr, now get on the inode */
581	ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, inode, NULL, 0);
582	if (ret) {
583		/* we failed to get on the inode, get off the idr */
584		inotify_remove_from_idr(group, tmp_i_mark);
585		goto out_err;
586	}
587
 
 
588
589	/* return the watch descriptor for this new mark */
590	ret = tmp_i_mark->wd;
591
592out_err:
593	/* match the ref from fsnotify_init_mark() */
594	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
595
596	return ret;
597}
598
599static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
600{
601	int ret = 0;
602
603	mutex_lock(&group->mark_mutex);
604	/* try to update and existing watch with the new arg */
605	ret = inotify_update_existing_watch(group, inode, arg);
606	/* no mark present, try to add a new one */
607	if (ret == -ENOENT)
608		ret = inotify_new_watch(group, inode, arg);
609	mutex_unlock(&group->mark_mutex);
 
 
 
 
 
 
610
611	return ret;
612}
613
614static struct fsnotify_group *inotify_new_group(unsigned int max_events)
615{
616	struct fsnotify_group *group;
617	struct inotify_event_info *oevent;
618
619	group = fsnotify_alloc_group(&inotify_fsnotify_ops);
620	if (IS_ERR(group))
621		return group;
622
623	oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
624	if (unlikely(!oevent)) {
625		fsnotify_destroy_group(group);
626		return ERR_PTR(-ENOMEM);
627	}
628	group->overflow_event = &oevent->fse;
629	fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW);
630	oevent->wd = -1;
631	oevent->sync_cookie = 0;
632	oevent->name_len = 0;
633
634	group->max_events = max_events;
635
636	spin_lock_init(&group->inotify_data.idr_lock);
637	idr_init(&group->inotify_data.idr);
638	group->inotify_data.ucounts = inc_ucount(current_user_ns(),
639						 current_euid(),
640						 UCOUNT_INOTIFY_INSTANCES);
641
642	if (!group->inotify_data.ucounts) {
643		fsnotify_destroy_group(group);
 
644		return ERR_PTR(-EMFILE);
645	}
646
647	return group;
648}
649
650
651/* inotify syscalls */
652static int do_inotify_init(int flags)
653{
654	struct fsnotify_group *group;
655	int ret;
656
657	/* Check the IN_* constants for consistency.  */
658	BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
659	BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
660
661	if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
662		return -EINVAL;
663
664	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
665	group = inotify_new_group(inotify_max_queued_events);
666	if (IS_ERR(group))
667		return PTR_ERR(group);
668
669	ret = anon_inode_getfd("inotify", &inotify_fops, group,
670				  O_RDONLY | flags);
671	if (ret < 0)
672		fsnotify_destroy_group(group);
673
674	return ret;
675}
676
677SYSCALL_DEFINE1(inotify_init1, int, flags)
678{
679	return do_inotify_init(flags);
680}
681
682SYSCALL_DEFINE0(inotify_init)
683{
684	return do_inotify_init(0);
685}
686
687SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
688		u32, mask)
689{
690	struct fsnotify_group *group;
691	struct inode *inode;
692	struct path path;
693	struct fd f;
694	int ret;
695	unsigned flags = 0;
696
697	/*
698	 * We share a lot of code with fs/dnotify.  We also share
699	 * the bit layout between inotify's IN_* and the fsnotify
700	 * FS_*.  This check ensures that only the inotify IN_*
701	 * bits get passed in and set in watches/events.
702	 */
703	if (unlikely(mask & ~ALL_INOTIFY_BITS))
704		return -EINVAL;
705	/*
706	 * Require at least one valid bit set in the mask.
707	 * Without _something_ set, we would have no events to
708	 * watch for.
709	 */
710	if (unlikely(!(mask & ALL_INOTIFY_BITS)))
711		return -EINVAL;
712
713	f = fdget(fd);
714	if (unlikely(!f.file))
715		return -EBADF;
716
717	/* verify that this is indeed an inotify instance */
718	if (unlikely(f.file->f_op != &inotify_fops)) {
719		ret = -EINVAL;
720		goto fput_and_out;
721	}
722
723	if (!(mask & IN_DONT_FOLLOW))
724		flags |= LOOKUP_FOLLOW;
725	if (mask & IN_ONLYDIR)
726		flags |= LOOKUP_DIRECTORY;
727
728	ret = inotify_find_inode(pathname, &path, flags);
729	if (ret)
730		goto fput_and_out;
731
732	/* inode held in place by reference to path; group by fget on fd */
733	inode = path.dentry->d_inode;
734	group = f.file->private_data;
735
736	/* create/update an inode mark */
737	ret = inotify_update_watch(group, inode, mask);
738	path_put(&path);
739fput_and_out:
740	fdput(f);
741	return ret;
742}
743
744SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
745{
746	struct fsnotify_group *group;
747	struct inotify_inode_mark *i_mark;
748	struct fd f;
749	int ret = 0;
750
751	f = fdget(fd);
752	if (unlikely(!f.file))
753		return -EBADF;
754
755	/* verify that this is indeed an inotify instance */
756	ret = -EINVAL;
757	if (unlikely(f.file->f_op != &inotify_fops))
758		goto out;
759
760	group = f.file->private_data;
761
762	ret = -EINVAL;
763	i_mark = inotify_idr_find(group, wd);
764	if (unlikely(!i_mark))
765		goto out;
766
767	ret = 0;
768
769	fsnotify_destroy_mark(&i_mark->fsn_mark, group);
770
771	/* match ref taken by inotify_idr_find */
772	fsnotify_put_mark(&i_mark->fsn_mark);
773
774out:
775	fdput(f);
776	return ret;
777}
778
779/*
780 * inotify_user_setup - Our initialization function.  Note that we cannot return
781 * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
782 * must result in panic().
783 */
784static int __init inotify_user_setup(void)
785{
786	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
787	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
788	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
789	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
790	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
791	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
792	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
793	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
794	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
795	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
796	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
797	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
798	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
799	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
800	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
801	BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
802	BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
803	BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
804
805	BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
806
807	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
 
808
809	inotify_max_queued_events = 16384;
810	init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
811	init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
812
813	return 0;
814}
815fs_initcall(inotify_user_setup);
v3.1
  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/ */
 47static int inotify_max_user_instances __read_mostly;
 48static int inotify_max_queued_events __read_mostly;
 49static int inotify_max_user_watches __read_mostly;
 50
 51static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
 52struct kmem_cache *event_priv_cachep __read_mostly;
 53
 54#ifdef CONFIG_SYSCTL
 55
 56#include <linux/sysctl.h>
 57
 58static int zero;
 59
 60ctl_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
 89static 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
105static 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 */
112static 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 */
133static 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 */
165static 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
230static 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
283static 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
290static int inotify_release(struct inode *ignored, struct file *file)
291{
292	struct fsnotify_group *group = file->private_data;
293
294	pr_debug("%s: group=%p\n", __func__, group);
295
296	fsnotify_clear_marks_by_group(group);
297
298	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
299	fsnotify_put_group(group);
300
301	return 0;
302}
303
304static long inotify_ioctl(struct file *file, unsigned int cmd,
305			  unsigned long arg)
306{
307	struct fsnotify_group *group;
308	struct fsnotify_event_holder *holder;
309	struct fsnotify_event *event;
310	void __user *p;
311	int ret = -ENOTTY;
312	size_t send_len = 0;
313
314	group = file->private_data;
315	p = (void __user *) arg;
316
317	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
318
319	switch (cmd) {
320	case FIONREAD:
321		mutex_lock(&group->notification_mutex);
322		list_for_each_entry(holder, &group->notification_list, event_list) {
323			event = holder->event;
324			send_len += sizeof(struct inotify_event);
325			if (event->name_len)
326				send_len += roundup(event->name_len + 1,
327						sizeof(struct inotify_event));
328		}
329		mutex_unlock(&group->notification_mutex);
330		ret = put_user(send_len, (int __user *) p);
331		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
332	}
333
334	return ret;
335}
336
337static const struct file_operations inotify_fops = {
 
338	.poll		= inotify_poll,
339	.read		= inotify_read,
340	.fasync		= inotify_fasync,
341	.release	= inotify_release,
342	.unlocked_ioctl	= inotify_ioctl,
343	.compat_ioctl	= inotify_ioctl,
344	.llseek		= noop_llseek,
345};
346
347
348/*
349 * find_inode - resolve a user-given path to a specific inode
350 */
351static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
352{
353	int error;
354
355	error = user_path_at(AT_FDCWD, dirname, flags, path);
356	if (error)
357		return error;
358	/* you can only watch an inode if you have read permissions on it */
359	error = inode_permission(path->dentry->d_inode, MAY_READ);
360	if (error)
361		path_put(path);
362	return error;
363}
364
365static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
366			      int *last_wd,
367			      struct inotify_inode_mark *i_mark)
368{
369	int ret;
370
371	do {
372		if (unlikely(!idr_pre_get(idr, GFP_KERNEL)))
373			return -ENOMEM;
374
375		spin_lock(idr_lock);
376		ret = idr_get_new_above(idr, i_mark, *last_wd + 1,
377					&i_mark->wd);
378		/* we added the mark to the idr, take a reference */
379		if (!ret) {
380			*last_wd = i_mark->wd;
381			fsnotify_get_mark(&i_mark->fsn_mark);
382		}
383		spin_unlock(idr_lock);
384	} while (ret == -EAGAIN);
385
386	return ret;
 
 
387}
388
389static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
390								int wd)
391{
392	struct idr *idr = &group->inotify_data.idr;
393	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
394	struct inotify_inode_mark *i_mark;
395
396	assert_spin_locked(idr_lock);
397
398	i_mark = idr_find(idr, wd);
399	if (i_mark) {
400		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
401
402		fsnotify_get_mark(fsn_mark);
403		/* One ref for being in the idr, one ref we just took */
404		BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
405	}
406
407	return i_mark;
408}
409
410static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
411							 int wd)
412{
413	struct inotify_inode_mark *i_mark;
414	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
415
416	spin_lock(idr_lock);
417	i_mark = inotify_idr_find_locked(group, wd);
418	spin_unlock(idr_lock);
419
420	return i_mark;
421}
422
423static void do_inotify_remove_from_idr(struct fsnotify_group *group,
424				       struct inotify_inode_mark *i_mark)
425{
426	struct idr *idr = &group->inotify_data.idr;
427	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
428	int wd = i_mark->wd;
429
430	assert_spin_locked(idr_lock);
431
432	idr_remove(idr, wd);
433
434	/* removed from the idr, drop that ref */
435	fsnotify_put_mark(&i_mark->fsn_mark);
436}
437
438/*
439 * Remove the mark from the idr (if present) and drop the reference
440 * on the mark because it was in the idr.
441 */
442static void inotify_remove_from_idr(struct fsnotify_group *group,
443				    struct inotify_inode_mark *i_mark)
444{
 
445	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
446	struct inotify_inode_mark *found_i_mark = NULL;
447	int wd;
448
449	spin_lock(idr_lock);
450	wd = i_mark->wd;
451
452	/*
453	 * does this i_mark think it is in the idr?  we shouldn't get called
454	 * if it wasn't....
455	 */
456	if (wd == -1) {
457		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
458			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
459			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
460		goto out;
461	}
462
463	/* Lets look in the idr to see if we find it */
464	found_i_mark = inotify_idr_find_locked(group, wd);
465	if (unlikely(!found_i_mark)) {
466		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
467			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
468			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
469		goto out;
470	}
471
472	/*
473	 * We found an mark in the idr at the right wd, but it's
474	 * not the mark we were told to remove.  eparis seriously
475	 * fucked up somewhere.
476	 */
477	if (unlikely(found_i_mark != i_mark)) {
478		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
479			"mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
480			"found_i_mark->group=%p found_i_mark->inode=%p\n",
481			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
482			i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
483			found_i_mark->fsn_mark.group,
484			found_i_mark->fsn_mark.i.inode);
485		goto out;
486	}
487
488	/*
489	 * One ref for being in the idr
490	 * one ref held by the caller trying to kill us
491	 * one ref grabbed by inotify_idr_find
492	 */
493	if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
494		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
495			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
496			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
497		/* we can't really recover with bad ref cnting.. */
498		BUG();
499	}
500
501	do_inotify_remove_from_idr(group, i_mark);
 
 
502out:
 
 
503	/* match the ref taken by inotify_idr_find_locked() */
504	if (found_i_mark)
505		fsnotify_put_mark(&found_i_mark->fsn_mark);
506	i_mark->wd = -1;
507	spin_unlock(idr_lock);
508}
509
510/*
511 * Send IN_IGNORED for this wd, remove this wd from the idr.
512 */
513void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
514				    struct fsnotify_group *group)
515{
516	struct inotify_inode_mark *i_mark;
517	struct fsnotify_event *ignored_event, *notify_event;
518	struct inotify_event_private_data *event_priv;
519	struct fsnotify_event_private_data *fsn_event_priv;
520	int ret;
521
522	ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
523					      FSNOTIFY_EVENT_NONE, NULL, 0,
524					      GFP_NOFS);
525	if (!ignored_event)
526		return;
527
528	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
529
530	event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
531	if (unlikely(!event_priv))
532		goto skip_send_ignore;
533
534	fsn_event_priv = &event_priv->fsnotify_event_priv_data;
535
536	fsn_event_priv->group = group;
537	event_priv->wd = i_mark->wd;
538
539	notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL);
540	if (notify_event) {
541		if (IS_ERR(notify_event))
542			ret = PTR_ERR(notify_event);
543		else
544			fsnotify_put_event(notify_event);
545		inotify_free_event_priv(fsn_event_priv);
546	}
547
548skip_send_ignore:
549
550	/* matches the reference taken when the event was created */
551	fsnotify_put_event(ignored_event);
552
553	/* remove this mark from the idr */
554	inotify_remove_from_idr(group, i_mark);
555
556	atomic_dec(&group->inotify_data.user->inotify_watches);
557}
558
559/* ding dong the mark is dead */
560static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
561{
562	struct inotify_inode_mark *i_mark;
563
564	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
565
566	kmem_cache_free(inotify_inode_mark_cachep, i_mark);
567}
568
569static int inotify_update_existing_watch(struct fsnotify_group *group,
570					 struct inode *inode,
571					 u32 arg)
572{
573	struct fsnotify_mark *fsn_mark;
574	struct inotify_inode_mark *i_mark;
575	__u32 old_mask, new_mask;
576	__u32 mask;
577	int add = (arg & IN_MASK_ADD);
578	int ret;
579
580	/* don't allow invalid bits: we don't want flags set */
581	mask = inotify_arg_to_mask(arg);
582	if (unlikely(!(mask & IN_ALL_EVENTS)))
583		return -EINVAL;
584
585	fsn_mark = fsnotify_find_inode_mark(group, inode);
586	if (!fsn_mark)
587		return -ENOENT;
588
589	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
590
591	spin_lock(&fsn_mark->lock);
592
593	old_mask = fsn_mark->mask;
594	if (add)
595		fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
596	else
597		fsnotify_set_mark_mask_locked(fsn_mark, mask);
598	new_mask = fsn_mark->mask;
599
600	spin_unlock(&fsn_mark->lock);
601
602	if (old_mask != new_mask) {
603		/* more bits in old than in new? */
604		int dropped = (old_mask & ~new_mask);
605		/* more bits in this fsn_mark than the inode's mask? */
606		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
607
608		/* update the inode with this new fsn_mark */
609		if (dropped || do_inode)
610			fsnotify_recalc_inode_mask(inode);
611
612	}
613
614	/* return the wd */
615	ret = i_mark->wd;
616
617	/* match the get from fsnotify_find_mark() */
618	fsnotify_put_mark(fsn_mark);
619
620	return ret;
621}
622
623static int inotify_new_watch(struct fsnotify_group *group,
624			     struct inode *inode,
625			     u32 arg)
626{
627	struct inotify_inode_mark *tmp_i_mark;
628	__u32 mask;
629	int ret;
630	struct idr *idr = &group->inotify_data.idr;
631	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
632
633	/* don't allow invalid bits: we don't want flags set */
634	mask = inotify_arg_to_mask(arg);
635	if (unlikely(!(mask & IN_ALL_EVENTS)))
636		return -EINVAL;
637
638	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
639	if (unlikely(!tmp_i_mark))
640		return -ENOMEM;
641
642	fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
643	tmp_i_mark->fsn_mark.mask = mask;
644	tmp_i_mark->wd = -1;
645
646	ret = -ENOSPC;
647	if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
648		goto out_err;
649
650	ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd,
651				 tmp_i_mark);
652	if (ret)
 
653		goto out_err;
 
654
655	/* we are on the idr, now get on the inode */
656	ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0);
657	if (ret) {
658		/* we failed to get on the inode, get off the idr */
659		inotify_remove_from_idr(group, tmp_i_mark);
660		goto out_err;
661	}
662
663	/* increment the number of watches the user has */
664	atomic_inc(&group->inotify_data.user->inotify_watches);
665
666	/* return the watch descriptor for this new mark */
667	ret = tmp_i_mark->wd;
668
669out_err:
670	/* match the ref from fsnotify_init_mark() */
671	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
672
673	return ret;
674}
675
676static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
677{
678	int ret = 0;
679
680retry:
681	/* try to update and existing watch with the new arg */
682	ret = inotify_update_existing_watch(group, inode, arg);
683	/* no mark present, try to add a new one */
684	if (ret == -ENOENT)
685		ret = inotify_new_watch(group, inode, arg);
686	/*
687	 * inotify_new_watch could race with another thread which did an
688	 * inotify_new_watch between the update_existing and the add watch
689	 * here, go back and try to update an existing mark again.
690	 */
691	if (ret == -EEXIST)
692		goto retry;
693
694	return ret;
695}
696
697static struct fsnotify_group *inotify_new_group(unsigned int max_events)
698{
699	struct fsnotify_group *group;
 
700
701	group = fsnotify_alloc_group(&inotify_fsnotify_ops);
702	if (IS_ERR(group))
703		return group;
704
 
 
 
 
 
 
 
 
 
 
 
705	group->max_events = max_events;
706
707	spin_lock_init(&group->inotify_data.idr_lock);
708	idr_init(&group->inotify_data.idr);
709	group->inotify_data.last_wd = 0;
710	group->inotify_data.fa = NULL;
711	group->inotify_data.user = get_current_user();
712
713	if (atomic_inc_return(&group->inotify_data.user->inotify_devs) >
714	    inotify_max_user_instances) {
715		fsnotify_put_group(group);
716		return ERR_PTR(-EMFILE);
717	}
718
719	return group;
720}
721
722
723/* inotify syscalls */
724SYSCALL_DEFINE1(inotify_init1, int, flags)
725{
726	struct fsnotify_group *group;
727	int ret;
728
729	/* Check the IN_* constants for consistency.  */
730	BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
731	BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
732
733	if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
734		return -EINVAL;
735
736	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
737	group = inotify_new_group(inotify_max_queued_events);
738	if (IS_ERR(group))
739		return PTR_ERR(group);
740
741	ret = anon_inode_getfd("inotify", &inotify_fops, group,
742				  O_RDONLY | flags);
743	if (ret < 0)
744		fsnotify_put_group(group);
745
746	return ret;
747}
748
 
 
 
 
 
749SYSCALL_DEFINE0(inotify_init)
750{
751	return sys_inotify_init1(0);
752}
753
754SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
755		u32, mask)
756{
757	struct fsnotify_group *group;
758	struct inode *inode;
759	struct path path;
760	struct file *filp;
761	int ret, fput_needed;
762	unsigned flags = 0;
763
764	filp = fget_light(fd, &fput_needed);
765	if (unlikely(!filp))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
766		return -EBADF;
767
768	/* verify that this is indeed an inotify instance */
769	if (unlikely(filp->f_op != &inotify_fops)) {
770		ret = -EINVAL;
771		goto fput_and_out;
772	}
773
774	if (!(mask & IN_DONT_FOLLOW))
775		flags |= LOOKUP_FOLLOW;
776	if (mask & IN_ONLYDIR)
777		flags |= LOOKUP_DIRECTORY;
778
779	ret = inotify_find_inode(pathname, &path, flags);
780	if (ret)
781		goto fput_and_out;
782
783	/* inode held in place by reference to path; group by fget on fd */
784	inode = path.dentry->d_inode;
785	group = filp->private_data;
786
787	/* create/update an inode mark */
788	ret = inotify_update_watch(group, inode, mask);
789	path_put(&path);
790fput_and_out:
791	fput_light(filp, fput_needed);
792	return ret;
793}
794
795SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
796{
797	struct fsnotify_group *group;
798	struct inotify_inode_mark *i_mark;
799	struct file *filp;
800	int ret = 0, fput_needed;
801
802	filp = fget_light(fd, &fput_needed);
803	if (unlikely(!filp))
804		return -EBADF;
805
806	/* verify that this is indeed an inotify instance */
807	ret = -EINVAL;
808	if (unlikely(filp->f_op != &inotify_fops))
809		goto out;
810
811	group = filp->private_data;
812
813	ret = -EINVAL;
814	i_mark = inotify_idr_find(group, wd);
815	if (unlikely(!i_mark))
816		goto out;
817
818	ret = 0;
819
820	fsnotify_destroy_mark(&i_mark->fsn_mark);
821
822	/* match ref taken by inotify_idr_find */
823	fsnotify_put_mark(&i_mark->fsn_mark);
824
825out:
826	fput_light(filp, fput_needed);
827	return ret;
828}
829
830/*
831 * inotify_user_setup - Our initialization function.  Note that we cannot return
832 * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
833 * must result in panic().
834 */
835static int __init inotify_user_setup(void)
836{
837	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
838	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
839	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
840	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
841	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
842	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
843	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
844	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
845	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
846	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
847	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
848	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
849	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
850	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
851	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
852	BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
853	BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
854	BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
855
856	BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
857
858	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
859	event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
860
861	inotify_max_queued_events = 16384;
862	inotify_max_user_instances = 128;
863	inotify_max_user_watches = 8192;
864
865	return 0;
866}
867module_init(inotify_user_setup);