xref: /openbmc/linux/fs/notify/inotify/inotify_user.c (revision c05787b4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * fs/inotify_user.c - inotify support for userspace
4  *
5  * Authors:
6  *	John McCutchan	<ttb@tentacle.dhs.org>
7  *	Robert Love	<rml@novell.com>
8  *
9  * Copyright (C) 2005 John McCutchan
10  * Copyright 2006 Hewlett-Packard Development Company, L.P.
11  *
12  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
13  * inotify was largely rewriten to make use of the fsnotify infrastructure
14  */
15 
16 #include <linux/file.h>
17 #include <linux/fs.h> /* struct inode */
18 #include <linux/fsnotify_backend.h>
19 #include <linux/idr.h>
20 #include <linux/init.h> /* fs_initcall */
21 #include <linux/inotify.h>
22 #include <linux/kernel.h> /* roundup() */
23 #include <linux/namei.h> /* LOOKUP_FOLLOW */
24 #include <linux/sched/signal.h>
25 #include <linux/slab.h> /* struct kmem_cache */
26 #include <linux/syscalls.h>
27 #include <linux/types.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/uaccess.h>
30 #include <linux/poll.h>
31 #include <linux/wait.h>
32 #include <linux/memcontrol.h>
33 #include <linux/security.h>
34 
35 #include "inotify.h"
36 #include "../fdinfo.h"
37 
38 #include <asm/ioctls.h>
39 
40 /*
41  * An inotify watch requires allocating an inotify_inode_mark structure as
42  * well as pinning the watched inode. Doubling the size of a VFS inode
43  * should be more than enough to cover the additional filesystem inode
44  * size increase.
45  */
46 #define INOTIFY_WATCH_COST	(sizeof(struct inotify_inode_mark) + \
47 				 2 * sizeof(struct inode))
48 
49 /* configurable via /proc/sys/fs/inotify/ */
50 static int inotify_max_queued_events __read_mostly;
51 
52 struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
53 
54 #ifdef CONFIG_SYSCTL
55 
56 #include <linux/sysctl.h>
57 
58 static long it_zero = 0;
59 static long it_int_max = INT_MAX;
60 
61 static struct ctl_table inotify_table[] = {
62 	{
63 		.procname	= "max_user_instances",
64 		.data		= &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
65 		.maxlen		= sizeof(long),
66 		.mode		= 0644,
67 		.proc_handler	= proc_doulongvec_minmax,
68 		.extra1		= &it_zero,
69 		.extra2		= &it_int_max,
70 	},
71 	{
72 		.procname	= "max_user_watches",
73 		.data		= &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
74 		.maxlen		= sizeof(long),
75 		.mode		= 0644,
76 		.proc_handler	= proc_doulongvec_minmax,
77 		.extra1		= &it_zero,
78 		.extra2		= &it_int_max,
79 	},
80 	{
81 		.procname	= "max_queued_events",
82 		.data		= &inotify_max_queued_events,
83 		.maxlen		= sizeof(int),
84 		.mode		= 0644,
85 		.proc_handler	= proc_dointvec_minmax,
86 		.extra1		= SYSCTL_ZERO
87 	},
88 	{ }
89 };
90 
inotify_sysctls_init(void)91 static void __init inotify_sysctls_init(void)
92 {
93 	register_sysctl("fs/inotify", inotify_table);
94 }
95 
96 #else
97 #define inotify_sysctls_init() do { } while (0)
98 #endif /* CONFIG_SYSCTL */
99 
inotify_arg_to_mask(struct inode * inode,u32 arg)100 static inline __u32 inotify_arg_to_mask(struct inode *inode, u32 arg)
101 {
102 	__u32 mask;
103 
104 	/*
105 	 * Everything should receive events when the inode is unmounted.
106 	 * All directories care about children.
107 	 */
108 	mask = (FS_UNMOUNT);
109 	if (S_ISDIR(inode->i_mode))
110 		mask |= FS_EVENT_ON_CHILD;
111 
112 	/* mask off the flags used to open the fd */
113 	mask |= (arg & INOTIFY_USER_MASK);
114 
115 	return mask;
116 }
117 
118 #define INOTIFY_MARK_FLAGS \
119 	(FSNOTIFY_MARK_FLAG_EXCL_UNLINK | FSNOTIFY_MARK_FLAG_IN_ONESHOT)
120 
inotify_arg_to_flags(u32 arg)121 static inline unsigned int inotify_arg_to_flags(u32 arg)
122 {
123 	unsigned int flags = 0;
124 
125 	if (arg & IN_EXCL_UNLINK)
126 		flags |= FSNOTIFY_MARK_FLAG_EXCL_UNLINK;
127 	if (arg & IN_ONESHOT)
128 		flags |= FSNOTIFY_MARK_FLAG_IN_ONESHOT;
129 
130 	return flags;
131 }
132 
inotify_mask_to_arg(__u32 mask)133 static inline u32 inotify_mask_to_arg(__u32 mask)
134 {
135 	return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
136 		       IN_Q_OVERFLOW);
137 }
138 
139 /* inotify userspace file descriptor functions */
inotify_poll(struct file * file,poll_table * wait)140 static __poll_t inotify_poll(struct file *file, poll_table *wait)
141 {
142 	struct fsnotify_group *group = file->private_data;
143 	__poll_t ret = 0;
144 
145 	poll_wait(file, &group->notification_waitq, wait);
146 	spin_lock(&group->notification_lock);
147 	if (!fsnotify_notify_queue_is_empty(group))
148 		ret = EPOLLIN | EPOLLRDNORM;
149 	spin_unlock(&group->notification_lock);
150 
151 	return ret;
152 }
153 
round_event_name_len(struct fsnotify_event * fsn_event)154 static int round_event_name_len(struct fsnotify_event *fsn_event)
155 {
156 	struct inotify_event_info *event;
157 
158 	event = INOTIFY_E(fsn_event);
159 	if (!event->name_len)
160 		return 0;
161 	return roundup(event->name_len + 1, sizeof(struct inotify_event));
162 }
163 
164 /*
165  * Get an inotify_kernel_event if one exists and is small
166  * enough to fit in "count". Return an error pointer if
167  * not large enough.
168  *
169  * Called with the group->notification_lock held.
170  */
get_one_event(struct fsnotify_group * group,size_t count)171 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
172 					    size_t count)
173 {
174 	size_t event_size = sizeof(struct inotify_event);
175 	struct fsnotify_event *event;
176 
177 	event = fsnotify_peek_first_event(group);
178 	if (!event)
179 		return NULL;
180 
181 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
182 
183 	event_size += round_event_name_len(event);
184 	if (event_size > count)
185 		return ERR_PTR(-EINVAL);
186 
187 	/* held the notification_lock the whole time, so this is the
188 	 * same event we peeked above */
189 	fsnotify_remove_first_event(group);
190 
191 	return event;
192 }
193 
194 /*
195  * Copy an event to user space, returning how much we copied.
196  *
197  * We already checked that the event size is smaller than the
198  * buffer we had in "get_one_event()" above.
199  */
copy_event_to_user(struct fsnotify_group * group,struct fsnotify_event * fsn_event,char __user * buf)200 static ssize_t copy_event_to_user(struct fsnotify_group *group,
201 				  struct fsnotify_event *fsn_event,
202 				  char __user *buf)
203 {
204 	struct inotify_event inotify_event;
205 	struct inotify_event_info *event;
206 	size_t event_size = sizeof(struct inotify_event);
207 	size_t name_len;
208 	size_t pad_name_len;
209 
210 	pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
211 
212 	event = INOTIFY_E(fsn_event);
213 	name_len = event->name_len;
214 	/*
215 	 * round up name length so it is a multiple of event_size
216 	 * plus an extra byte for the terminating '\0'.
217 	 */
218 	pad_name_len = round_event_name_len(fsn_event);
219 	inotify_event.len = pad_name_len;
220 	inotify_event.mask = inotify_mask_to_arg(event->mask);
221 	inotify_event.wd = event->wd;
222 	inotify_event.cookie = event->sync_cookie;
223 
224 	/* send the main event */
225 	if (copy_to_user(buf, &inotify_event, event_size))
226 		return -EFAULT;
227 
228 	buf += event_size;
229 
230 	/*
231 	 * fsnotify only stores the pathname, so here we have to send the pathname
232 	 * and then pad that pathname out to a multiple of sizeof(inotify_event)
233 	 * with zeros.
234 	 */
235 	if (pad_name_len) {
236 		/* copy the path name */
237 		if (copy_to_user(buf, event->name, name_len))
238 			return -EFAULT;
239 		buf += name_len;
240 
241 		/* fill userspace with 0's */
242 		if (clear_user(buf, pad_name_len - name_len))
243 			return -EFAULT;
244 		event_size += pad_name_len;
245 	}
246 
247 	return event_size;
248 }
249 
inotify_read(struct file * file,char __user * buf,size_t count,loff_t * pos)250 static ssize_t inotify_read(struct file *file, char __user *buf,
251 			    size_t count, loff_t *pos)
252 {
253 	struct fsnotify_group *group;
254 	struct fsnotify_event *kevent;
255 	char __user *start;
256 	int ret;
257 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
258 
259 	start = buf;
260 	group = file->private_data;
261 
262 	add_wait_queue(&group->notification_waitq, &wait);
263 	while (1) {
264 		spin_lock(&group->notification_lock);
265 		kevent = get_one_event(group, count);
266 		spin_unlock(&group->notification_lock);
267 
268 		pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
269 
270 		if (kevent) {
271 			ret = PTR_ERR(kevent);
272 			if (IS_ERR(kevent))
273 				break;
274 			ret = copy_event_to_user(group, kevent, buf);
275 			fsnotify_destroy_event(group, kevent);
276 			if (ret < 0)
277 				break;
278 			buf += ret;
279 			count -= ret;
280 			continue;
281 		}
282 
283 		ret = -EAGAIN;
284 		if (file->f_flags & O_NONBLOCK)
285 			break;
286 		ret = -ERESTARTSYS;
287 		if (signal_pending(current))
288 			break;
289 
290 		if (start != buf)
291 			break;
292 
293 		wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
294 	}
295 	remove_wait_queue(&group->notification_waitq, &wait);
296 
297 	if (start != buf && ret != -EFAULT)
298 		ret = buf - start;
299 	return ret;
300 }
301 
inotify_release(struct inode * ignored,struct file * file)302 static int inotify_release(struct inode *ignored, struct file *file)
303 {
304 	struct fsnotify_group *group = file->private_data;
305 
306 	pr_debug("%s: group=%p\n", __func__, group);
307 
308 	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
309 	fsnotify_destroy_group(group);
310 
311 	return 0;
312 }
313 
inotify_ioctl(struct file * file,unsigned int cmd,unsigned long arg)314 static long inotify_ioctl(struct file *file, unsigned int cmd,
315 			  unsigned long arg)
316 {
317 	struct fsnotify_group *group;
318 	struct fsnotify_event *fsn_event;
319 	void __user *p;
320 	int ret = -ENOTTY;
321 	size_t send_len = 0;
322 
323 	group = file->private_data;
324 	p = (void __user *) arg;
325 
326 	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
327 
328 	switch (cmd) {
329 	case FIONREAD:
330 		spin_lock(&group->notification_lock);
331 		list_for_each_entry(fsn_event, &group->notification_list,
332 				    list) {
333 			send_len += sizeof(struct inotify_event);
334 			send_len += round_event_name_len(fsn_event);
335 		}
336 		spin_unlock(&group->notification_lock);
337 		ret = put_user(send_len, (int __user *) p);
338 		break;
339 #ifdef CONFIG_CHECKPOINT_RESTORE
340 	case INOTIFY_IOC_SETNEXTWD:
341 		ret = -EINVAL;
342 		if (arg >= 1 && arg <= INT_MAX) {
343 			struct inotify_group_private_data *data;
344 
345 			data = &group->inotify_data;
346 			spin_lock(&data->idr_lock);
347 			idr_set_cursor(&data->idr, (unsigned int)arg);
348 			spin_unlock(&data->idr_lock);
349 			ret = 0;
350 		}
351 		break;
352 #endif /* CONFIG_CHECKPOINT_RESTORE */
353 	}
354 
355 	return ret;
356 }
357 
358 static const struct file_operations inotify_fops = {
359 	.show_fdinfo	= inotify_show_fdinfo,
360 	.poll		= inotify_poll,
361 	.read		= inotify_read,
362 	.fasync		= fsnotify_fasync,
363 	.release	= inotify_release,
364 	.unlocked_ioctl	= inotify_ioctl,
365 	.compat_ioctl	= inotify_ioctl,
366 	.llseek		= noop_llseek,
367 };
368 
369 
370 /*
371  * find_inode - resolve a user-given path to a specific inode
372  */
inotify_find_inode(const char __user * dirname,struct path * path,unsigned int flags,__u64 mask)373 static int inotify_find_inode(const char __user *dirname, struct path *path,
374 						unsigned int flags, __u64 mask)
375 {
376 	int error;
377 
378 	error = user_path_at(AT_FDCWD, dirname, flags, path);
379 	if (error)
380 		return error;
381 	/* you can only watch an inode if you have read permissions on it */
382 	error = path_permission(path, MAY_READ);
383 	if (error) {
384 		path_put(path);
385 		return error;
386 	}
387 	error = security_path_notify(path, mask,
388 				FSNOTIFY_OBJ_TYPE_INODE);
389 	if (error)
390 		path_put(path);
391 
392 	return error;
393 }
394 
inotify_add_to_idr(struct idr * idr,spinlock_t * idr_lock,struct inotify_inode_mark * i_mark)395 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
396 			      struct inotify_inode_mark *i_mark)
397 {
398 	int ret;
399 
400 	idr_preload(GFP_KERNEL);
401 	spin_lock(idr_lock);
402 
403 	ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
404 	if (ret >= 0) {
405 		/* we added the mark to the idr, take a reference */
406 		i_mark->wd = ret;
407 		fsnotify_get_mark(&i_mark->fsn_mark);
408 	}
409 
410 	spin_unlock(idr_lock);
411 	idr_preload_end();
412 	return ret < 0 ? ret : 0;
413 }
414 
inotify_idr_find_locked(struct fsnotify_group * group,int wd)415 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
416 								int wd)
417 {
418 	struct idr *idr = &group->inotify_data.idr;
419 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
420 	struct inotify_inode_mark *i_mark;
421 
422 	assert_spin_locked(idr_lock);
423 
424 	i_mark = idr_find(idr, wd);
425 	if (i_mark) {
426 		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
427 
428 		fsnotify_get_mark(fsn_mark);
429 		/* One ref for being in the idr, one ref we just took */
430 		BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
431 	}
432 
433 	return i_mark;
434 }
435 
inotify_idr_find(struct fsnotify_group * group,int wd)436 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
437 							 int wd)
438 {
439 	struct inotify_inode_mark *i_mark;
440 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
441 
442 	spin_lock(idr_lock);
443 	i_mark = inotify_idr_find_locked(group, wd);
444 	spin_unlock(idr_lock);
445 
446 	return i_mark;
447 }
448 
449 /*
450  * Remove the mark from the idr (if present) and drop the reference
451  * on the mark because it was in the idr.
452  */
inotify_remove_from_idr(struct fsnotify_group * group,struct inotify_inode_mark * i_mark)453 static void inotify_remove_from_idr(struct fsnotify_group *group,
454 				    struct inotify_inode_mark *i_mark)
455 {
456 	struct idr *idr = &group->inotify_data.idr;
457 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
458 	struct inotify_inode_mark *found_i_mark = NULL;
459 	int wd;
460 
461 	spin_lock(idr_lock);
462 	wd = i_mark->wd;
463 
464 	/*
465 	 * does this i_mark think it is in the idr?  we shouldn't get called
466 	 * if it wasn't....
467 	 */
468 	if (wd == -1) {
469 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
470 			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
471 		goto out;
472 	}
473 
474 	/* Lets look in the idr to see if we find it */
475 	found_i_mark = inotify_idr_find_locked(group, wd);
476 	if (unlikely(!found_i_mark)) {
477 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
478 			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
479 		goto out;
480 	}
481 
482 	/*
483 	 * We found an mark in the idr at the right wd, but it's
484 	 * not the mark we were told to remove.  eparis seriously
485 	 * fucked up somewhere.
486 	 */
487 	if (unlikely(found_i_mark != i_mark)) {
488 		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
489 			"found_i_mark=%p found_i_mark->wd=%d "
490 			"found_i_mark->group=%p\n", __func__, i_mark,
491 			i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
492 			found_i_mark->wd, found_i_mark->fsn_mark.group);
493 		goto out;
494 	}
495 
496 	/*
497 	 * One ref for being in the idr
498 	 * one ref grabbed by inotify_idr_find
499 	 */
500 	if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
501 		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
502 			 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
503 		/* we can't really recover with bad ref cnting.. */
504 		BUG();
505 	}
506 
507 	idr_remove(idr, wd);
508 	/* Removed from the idr, drop that ref. */
509 	fsnotify_put_mark(&i_mark->fsn_mark);
510 out:
511 	i_mark->wd = -1;
512 	spin_unlock(idr_lock);
513 	/* match the ref taken by inotify_idr_find_locked() */
514 	if (found_i_mark)
515 		fsnotify_put_mark(&found_i_mark->fsn_mark);
516 }
517 
518 /*
519  * Send IN_IGNORED for this wd, remove this wd from the idr.
520  */
inotify_ignored_and_remove_idr(struct fsnotify_mark * fsn_mark,struct fsnotify_group * group)521 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
522 				    struct fsnotify_group *group)
523 {
524 	struct inotify_inode_mark *i_mark;
525 
526 	/* Queue ignore event for the watch */
527 	inotify_handle_inode_event(fsn_mark, FS_IN_IGNORED, NULL, NULL, NULL,
528 				   0);
529 
530 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
531 	/* remove this mark from the idr */
532 	inotify_remove_from_idr(group, i_mark);
533 
534 	dec_inotify_watches(group->inotify_data.ucounts);
535 }
536 
inotify_update_existing_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)537 static int inotify_update_existing_watch(struct fsnotify_group *group,
538 					 struct inode *inode,
539 					 u32 arg)
540 {
541 	struct fsnotify_mark *fsn_mark;
542 	struct inotify_inode_mark *i_mark;
543 	__u32 old_mask, new_mask;
544 	int replace = !(arg & IN_MASK_ADD);
545 	int create = (arg & IN_MASK_CREATE);
546 	int ret;
547 
548 	fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
549 	if (!fsn_mark)
550 		return -ENOENT;
551 	else if (create) {
552 		ret = -EEXIST;
553 		goto out;
554 	}
555 
556 	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
557 
558 	spin_lock(&fsn_mark->lock);
559 	old_mask = fsn_mark->mask;
560 	if (replace) {
561 		fsn_mark->mask = 0;
562 		fsn_mark->flags &= ~INOTIFY_MARK_FLAGS;
563 	}
564 	fsn_mark->mask |= inotify_arg_to_mask(inode, arg);
565 	fsn_mark->flags |= inotify_arg_to_flags(arg);
566 	new_mask = fsn_mark->mask;
567 	spin_unlock(&fsn_mark->lock);
568 
569 	if (old_mask != new_mask) {
570 		/* more bits in old than in new? */
571 		int dropped = (old_mask & ~new_mask);
572 		/* more bits in this fsn_mark than the inode's mask? */
573 		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
574 
575 		/* update the inode with this new fsn_mark */
576 		if (dropped || do_inode)
577 			fsnotify_recalc_mask(inode->i_fsnotify_marks);
578 
579 	}
580 
581 	/* return the wd */
582 	ret = i_mark->wd;
583 
584 out:
585 	/* match the get from fsnotify_find_mark() */
586 	fsnotify_put_mark(fsn_mark);
587 
588 	return ret;
589 }
590 
inotify_new_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)591 static int inotify_new_watch(struct fsnotify_group *group,
592 			     struct inode *inode,
593 			     u32 arg)
594 {
595 	struct inotify_inode_mark *tmp_i_mark;
596 	int ret;
597 	struct idr *idr = &group->inotify_data.idr;
598 	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
599 
600 	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
601 	if (unlikely(!tmp_i_mark))
602 		return -ENOMEM;
603 
604 	fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
605 	tmp_i_mark->fsn_mark.mask = inotify_arg_to_mask(inode, arg);
606 	tmp_i_mark->fsn_mark.flags = inotify_arg_to_flags(arg);
607 	tmp_i_mark->wd = -1;
608 
609 	ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
610 	if (ret)
611 		goto out_err;
612 
613 	/* increment the number of watches the user has */
614 	if (!inc_inotify_watches(group->inotify_data.ucounts)) {
615 		inotify_remove_from_idr(group, tmp_i_mark);
616 		ret = -ENOSPC;
617 		goto out_err;
618 	}
619 
620 	/* we are on the idr, now get on the inode */
621 	ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
622 	if (ret) {
623 		/* we failed to get on the inode, get off the idr */
624 		inotify_remove_from_idr(group, tmp_i_mark);
625 		goto out_err;
626 	}
627 
628 
629 	/* return the watch descriptor for this new mark */
630 	ret = tmp_i_mark->wd;
631 
632 out_err:
633 	/* match the ref from fsnotify_init_mark() */
634 	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
635 
636 	return ret;
637 }
638 
inotify_update_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)639 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
640 {
641 	int ret = 0;
642 
643 	fsnotify_group_lock(group);
644 	/* try to update and existing watch with the new arg */
645 	ret = inotify_update_existing_watch(group, inode, arg);
646 	/* no mark present, try to add a new one */
647 	if (ret == -ENOENT)
648 		ret = inotify_new_watch(group, inode, arg);
649 	fsnotify_group_unlock(group);
650 
651 	return ret;
652 }
653 
inotify_new_group(unsigned int max_events)654 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
655 {
656 	struct fsnotify_group *group;
657 	struct inotify_event_info *oevent;
658 
659 	group = fsnotify_alloc_group(&inotify_fsnotify_ops,
660 				     FSNOTIFY_GROUP_USER);
661 	if (IS_ERR(group))
662 		return group;
663 
664 	oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL_ACCOUNT);
665 	if (unlikely(!oevent)) {
666 		fsnotify_destroy_group(group);
667 		return ERR_PTR(-ENOMEM);
668 	}
669 	group->overflow_event = &oevent->fse;
670 	fsnotify_init_event(group->overflow_event);
671 	oevent->mask = FS_Q_OVERFLOW;
672 	oevent->wd = -1;
673 	oevent->sync_cookie = 0;
674 	oevent->name_len = 0;
675 
676 	group->max_events = max_events;
677 	group->memcg = get_mem_cgroup_from_mm(current->mm);
678 
679 	spin_lock_init(&group->inotify_data.idr_lock);
680 	idr_init(&group->inotify_data.idr);
681 	group->inotify_data.ucounts = inc_ucount(current_user_ns(),
682 						 current_euid(),
683 						 UCOUNT_INOTIFY_INSTANCES);
684 
685 	if (!group->inotify_data.ucounts) {
686 		fsnotify_destroy_group(group);
687 		return ERR_PTR(-EMFILE);
688 	}
689 
690 	return group;
691 }
692 
693 
694 /* inotify syscalls */
do_inotify_init(int flags)695 static int do_inotify_init(int flags)
696 {
697 	struct fsnotify_group *group;
698 	int ret;
699 
700 	/* Check the IN_* constants for consistency.  */
701 	BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
702 	BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
703 
704 	if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
705 		return -EINVAL;
706 
707 	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
708 	group = inotify_new_group(inotify_max_queued_events);
709 	if (IS_ERR(group))
710 		return PTR_ERR(group);
711 
712 	ret = anon_inode_getfd("inotify", &inotify_fops, group,
713 				  O_RDONLY | flags);
714 	if (ret < 0)
715 		fsnotify_destroy_group(group);
716 
717 	return ret;
718 }
719 
SYSCALL_DEFINE1(inotify_init1,int,flags)720 SYSCALL_DEFINE1(inotify_init1, int, flags)
721 {
722 	return do_inotify_init(flags);
723 }
724 
SYSCALL_DEFINE0(inotify_init)725 SYSCALL_DEFINE0(inotify_init)
726 {
727 	return do_inotify_init(0);
728 }
729 
SYSCALL_DEFINE3(inotify_add_watch,int,fd,const char __user *,pathname,u32,mask)730 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
731 		u32, mask)
732 {
733 	struct fsnotify_group *group;
734 	struct inode *inode;
735 	struct path path;
736 	struct fd f;
737 	int ret;
738 	unsigned flags = 0;
739 
740 	/*
741 	 * We share a lot of code with fs/dnotify.  We also share
742 	 * the bit layout between inotify's IN_* and the fsnotify
743 	 * FS_*.  This check ensures that only the inotify IN_*
744 	 * bits get passed in and set in watches/events.
745 	 */
746 	if (unlikely(mask & ~ALL_INOTIFY_BITS))
747 		return -EINVAL;
748 	/*
749 	 * Require at least one valid bit set in the mask.
750 	 * Without _something_ set, we would have no events to
751 	 * watch for.
752 	 */
753 	if (unlikely(!(mask & ALL_INOTIFY_BITS)))
754 		return -EINVAL;
755 
756 	f = fdget(fd);
757 	if (unlikely(!f.file))
758 		return -EBADF;
759 
760 	/* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
761 	if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
762 		ret = -EINVAL;
763 		goto fput_and_out;
764 	}
765 
766 	/* verify that this is indeed an inotify instance */
767 	if (unlikely(f.file->f_op != &inotify_fops)) {
768 		ret = -EINVAL;
769 		goto fput_and_out;
770 	}
771 
772 	if (!(mask & IN_DONT_FOLLOW))
773 		flags |= LOOKUP_FOLLOW;
774 	if (mask & IN_ONLYDIR)
775 		flags |= LOOKUP_DIRECTORY;
776 
777 	ret = inotify_find_inode(pathname, &path, flags,
778 			(mask & IN_ALL_EVENTS));
779 	if (ret)
780 		goto fput_and_out;
781 
782 	/* inode held in place by reference to path; group by fget on fd */
783 	inode = path.dentry->d_inode;
784 	group = f.file->private_data;
785 
786 	/* create/update an inode mark */
787 	ret = inotify_update_watch(group, inode, mask);
788 	path_put(&path);
789 fput_and_out:
790 	fdput(f);
791 	return ret;
792 }
793 
SYSCALL_DEFINE2(inotify_rm_watch,int,fd,__s32,wd)794 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
795 {
796 	struct fsnotify_group *group;
797 	struct inotify_inode_mark *i_mark;
798 	struct fd f;
799 	int ret = -EINVAL;
800 
801 	f = fdget(fd);
802 	if (unlikely(!f.file))
803 		return -EBADF;
804 
805 	/* verify that this is indeed an inotify instance */
806 	if (unlikely(f.file->f_op != &inotify_fops))
807 		goto out;
808 
809 	group = f.file->private_data;
810 
811 	i_mark = inotify_idr_find(group, wd);
812 	if (unlikely(!i_mark))
813 		goto out;
814 
815 	ret = 0;
816 
817 	fsnotify_destroy_mark(&i_mark->fsn_mark, group);
818 
819 	/* match ref taken by inotify_idr_find */
820 	fsnotify_put_mark(&i_mark->fsn_mark);
821 
822 out:
823 	fdput(f);
824 	return ret;
825 }
826 
827 /*
828  * inotify_user_setup - Our initialization function.  Note that we cannot return
829  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
830  * must result in panic().
831  */
inotify_user_setup(void)832 static int __init inotify_user_setup(void)
833 {
834 	unsigned long watches_max;
835 	struct sysinfo si;
836 
837 	si_meminfo(&si);
838 	/*
839 	 * Allow up to 1% of addressable memory to be allocated for inotify
840 	 * watches (per user) limited to the range [8192, 1048576].
841 	 */
842 	watches_max = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
843 			INOTIFY_WATCH_COST;
844 	watches_max = clamp(watches_max, 8192UL, 1048576UL);
845 
846 	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
847 	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
848 	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
849 	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
850 	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
851 	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
852 	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
853 	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
854 	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
855 	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
856 	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
857 	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
858 	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
859 	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
860 	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
861 	BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
862 
863 	BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
864 
865 	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
866 					       SLAB_PANIC|SLAB_ACCOUNT);
867 
868 	inotify_max_queued_events = 16384;
869 	init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
870 	init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = watches_max;
871 	inotify_sysctls_init();
872 
873 	return 0;
874 }
875 fs_initcall(inotify_user_setup);
876