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/ */
48 static int inotify_max_queued_events __read_mostly;
49
50 struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
51
52 #ifdef CONFIG_SYSCTL
53
54 #include <linux/sysctl.h>
55
56 static int zero;
57
58 struct 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
inotify_arg_to_mask(u32 arg)87 static 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
inotify_mask_to_arg(__u32 mask)103 static 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 */
inotify_poll(struct file * file,poll_table * wait)110 static unsigned int inotify_poll(struct file *file, poll_table *wait)
111 {
112 struct fsnotify_group *group = file->private_data;
113 int 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 = POLLIN | POLLRDNORM;
119 spin_unlock(&group->notification_lock);
120
121 return ret;
122 }
123
round_event_name_len(struct fsnotify_event * fsn_event)124 static 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 */
get_one_event(struct fsnotify_group * group,size_t count)141 static 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 */
copy_event_to_user(struct fsnotify_group * group,struct fsnotify_event * fsn_event,char __user * buf)171 static 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
inotify_read(struct file * file,char __user * buf,size_t count,loff_t * pos)221 static 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
inotify_release(struct inode * ignored,struct file * file)273 static 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
inotify_ioctl(struct file * file,unsigned int cmd,unsigned long arg)285 static 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 }
311
312 return ret;
313 }
314
315 static const struct file_operations inotify_fops = {
316 .show_fdinfo = inotify_show_fdinfo,
317 .poll = inotify_poll,
318 .read = inotify_read,
319 .fasync = fsnotify_fasync,
320 .release = inotify_release,
321 .unlocked_ioctl = inotify_ioctl,
322 .compat_ioctl = inotify_ioctl,
323 .llseek = noop_llseek,
324 };
325
326
327 /*
328 * find_inode - resolve a user-given path to a specific inode
329 */
inotify_find_inode(const char __user * dirname,struct path * path,unsigned flags)330 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
331 {
332 int error;
333
334 error = user_path_at(AT_FDCWD, dirname, flags, path);
335 if (error)
336 return error;
337 /* you can only watch an inode if you have read permissions on it */
338 error = inode_permission2(path->mnt, path->dentry->d_inode, MAY_READ);
339 if (error)
340 path_put(path);
341 return error;
342 }
343
inotify_add_to_idr(struct idr * idr,spinlock_t * idr_lock,struct inotify_inode_mark * i_mark)344 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
345 struct inotify_inode_mark *i_mark)
346 {
347 int ret;
348
349 idr_preload(GFP_KERNEL);
350 spin_lock(idr_lock);
351
352 ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
353 if (ret >= 0) {
354 /* we added the mark to the idr, take a reference */
355 i_mark->wd = ret;
356 fsnotify_get_mark(&i_mark->fsn_mark);
357 }
358
359 spin_unlock(idr_lock);
360 idr_preload_end();
361 return ret < 0 ? ret : 0;
362 }
363
inotify_idr_find_locked(struct fsnotify_group * group,int wd)364 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
365 int wd)
366 {
367 struct idr *idr = &group->inotify_data.idr;
368 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
369 struct inotify_inode_mark *i_mark;
370
371 assert_spin_locked(idr_lock);
372
373 i_mark = idr_find(idr, wd);
374 if (i_mark) {
375 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
376
377 fsnotify_get_mark(fsn_mark);
378 /* One ref for being in the idr, one ref we just took */
379 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
380 }
381
382 return i_mark;
383 }
384
inotify_idr_find(struct fsnotify_group * group,int wd)385 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
386 int wd)
387 {
388 struct inotify_inode_mark *i_mark;
389 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
390
391 spin_lock(idr_lock);
392 i_mark = inotify_idr_find_locked(group, wd);
393 spin_unlock(idr_lock);
394
395 return i_mark;
396 }
397
398 /*
399 * Remove the mark from the idr (if present) and drop the reference
400 * on the mark because it was in the idr.
401 */
inotify_remove_from_idr(struct fsnotify_group * group,struct inotify_inode_mark * i_mark)402 static void inotify_remove_from_idr(struct fsnotify_group *group,
403 struct inotify_inode_mark *i_mark)
404 {
405 struct idr *idr = &group->inotify_data.idr;
406 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
407 struct inotify_inode_mark *found_i_mark = NULL;
408 int wd;
409
410 spin_lock(idr_lock);
411 wd = i_mark->wd;
412
413 /*
414 * does this i_mark think it is in the idr? we shouldn't get called
415 * if it wasn't....
416 */
417 if (wd == -1) {
418 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
419 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
420 goto out;
421 }
422
423 /* Lets look in the idr to see if we find it */
424 found_i_mark = inotify_idr_find_locked(group, wd);
425 if (unlikely(!found_i_mark)) {
426 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
427 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
428 goto out;
429 }
430
431 /*
432 * We found an mark in the idr at the right wd, but it's
433 * not the mark we were told to remove. eparis seriously
434 * fucked up somewhere.
435 */
436 if (unlikely(found_i_mark != i_mark)) {
437 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
438 "found_i_mark=%p found_i_mark->wd=%d "
439 "found_i_mark->group=%p\n", __func__, i_mark,
440 i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
441 found_i_mark->wd, found_i_mark->fsn_mark.group);
442 goto out;
443 }
444
445 /*
446 * One ref for being in the idr
447 * one ref grabbed by inotify_idr_find
448 */
449 if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 2)) {
450 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
451 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
452 /* we can't really recover with bad ref cnting.. */
453 BUG();
454 }
455
456 idr_remove(idr, wd);
457 /* Removed from the idr, drop that ref. */
458 fsnotify_put_mark(&i_mark->fsn_mark);
459 out:
460 i_mark->wd = -1;
461 spin_unlock(idr_lock);
462 /* match the ref taken by inotify_idr_find_locked() */
463 if (found_i_mark)
464 fsnotify_put_mark(&found_i_mark->fsn_mark);
465 }
466
467 /*
468 * Send IN_IGNORED for this wd, remove this wd from the idr.
469 */
inotify_ignored_and_remove_idr(struct fsnotify_mark * fsn_mark,struct fsnotify_group * group)470 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
471 struct fsnotify_group *group)
472 {
473 struct inotify_inode_mark *i_mark;
474
475 /* Queue ignore event for the watch */
476 inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED,
477 NULL, FSNOTIFY_EVENT_NONE, NULL, 0, NULL);
478
479 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
480 /* remove this mark from the idr */
481 inotify_remove_from_idr(group, i_mark);
482
483 dec_inotify_watches(group->inotify_data.ucounts);
484 }
485
inotify_update_existing_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)486 static int inotify_update_existing_watch(struct fsnotify_group *group,
487 struct inode *inode,
488 u32 arg)
489 {
490 struct fsnotify_mark *fsn_mark;
491 struct inotify_inode_mark *i_mark;
492 __u32 old_mask, new_mask;
493 __u32 mask;
494 int add = (arg & IN_MASK_ADD);
495 int ret;
496
497 mask = inotify_arg_to_mask(arg);
498
499 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
500 if (!fsn_mark)
501 return -ENOENT;
502
503 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
504
505 spin_lock(&fsn_mark->lock);
506 old_mask = fsn_mark->mask;
507 if (add)
508 fsn_mark->mask |= mask;
509 else
510 fsn_mark->mask = mask;
511 new_mask = fsn_mark->mask;
512 spin_unlock(&fsn_mark->lock);
513
514 if (old_mask != new_mask) {
515 /* more bits in old than in new? */
516 int dropped = (old_mask & ~new_mask);
517 /* more bits in this fsn_mark than the inode's mask? */
518 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
519
520 /* update the inode with this new fsn_mark */
521 if (dropped || do_inode)
522 fsnotify_recalc_mask(inode->i_fsnotify_marks);
523
524 }
525
526 /* return the wd */
527 ret = i_mark->wd;
528
529 /* match the get from fsnotify_find_mark() */
530 fsnotify_put_mark(fsn_mark);
531
532 return ret;
533 }
534
inotify_new_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)535 static int inotify_new_watch(struct fsnotify_group *group,
536 struct inode *inode,
537 u32 arg)
538 {
539 struct inotify_inode_mark *tmp_i_mark;
540 __u32 mask;
541 int ret;
542 struct idr *idr = &group->inotify_data.idr;
543 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
544
545 mask = inotify_arg_to_mask(arg);
546
547 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
548 if (unlikely(!tmp_i_mark))
549 return -ENOMEM;
550
551 fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
552 tmp_i_mark->fsn_mark.mask = mask;
553 tmp_i_mark->wd = -1;
554
555 ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
556 if (ret)
557 goto out_err;
558
559 /* increment the number of watches the user has */
560 if (!inc_inotify_watches(group->inotify_data.ucounts)) {
561 inotify_remove_from_idr(group, tmp_i_mark);
562 ret = -ENOSPC;
563 goto out_err;
564 }
565
566 /* we are on the idr, now get on the inode */
567 ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, inode, NULL, 0);
568 if (ret) {
569 /* we failed to get on the inode, get off the idr */
570 inotify_remove_from_idr(group, tmp_i_mark);
571 goto out_err;
572 }
573
574
575 /* return the watch descriptor for this new mark */
576 ret = tmp_i_mark->wd;
577
578 out_err:
579 /* match the ref from fsnotify_init_mark() */
580 fsnotify_put_mark(&tmp_i_mark->fsn_mark);
581
582 return ret;
583 }
584
inotify_update_watch(struct fsnotify_group * group,struct inode * inode,u32 arg)585 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
586 {
587 int ret = 0;
588
589 mutex_lock(&group->mark_mutex);
590 /* try to update and existing watch with the new arg */
591 ret = inotify_update_existing_watch(group, inode, arg);
592 /* no mark present, try to add a new one */
593 if (ret == -ENOENT)
594 ret = inotify_new_watch(group, inode, arg);
595 mutex_unlock(&group->mark_mutex);
596
597 return ret;
598 }
599
inotify_new_group(unsigned int max_events)600 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
601 {
602 struct fsnotify_group *group;
603 struct inotify_event_info *oevent;
604
605 group = fsnotify_alloc_group(&inotify_fsnotify_ops);
606 if (IS_ERR(group))
607 return group;
608
609 oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
610 if (unlikely(!oevent)) {
611 fsnotify_destroy_group(group);
612 return ERR_PTR(-ENOMEM);
613 }
614 group->overflow_event = &oevent->fse;
615 fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW);
616 oevent->wd = -1;
617 oevent->sync_cookie = 0;
618 oevent->name_len = 0;
619
620 group->max_events = max_events;
621
622 spin_lock_init(&group->inotify_data.idr_lock);
623 idr_init(&group->inotify_data.idr);
624 group->inotify_data.ucounts = inc_ucount(current_user_ns(),
625 current_euid(),
626 UCOUNT_INOTIFY_INSTANCES);
627
628 if (!group->inotify_data.ucounts) {
629 fsnotify_destroy_group(group);
630 return ERR_PTR(-EMFILE);
631 }
632
633 return group;
634 }
635
636
637 /* inotify syscalls */
SYSCALL_DEFINE1(inotify_init1,int,flags)638 SYSCALL_DEFINE1(inotify_init1, int, flags)
639 {
640 struct fsnotify_group *group;
641 int ret;
642
643 /* Check the IN_* constants for consistency. */
644 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
645 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
646
647 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
648 return -EINVAL;
649
650 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
651 group = inotify_new_group(inotify_max_queued_events);
652 if (IS_ERR(group))
653 return PTR_ERR(group);
654
655 ret = anon_inode_getfd("inotify", &inotify_fops, group,
656 O_RDONLY | flags);
657 if (ret < 0)
658 fsnotify_destroy_group(group);
659
660 return ret;
661 }
662
SYSCALL_DEFINE0(inotify_init)663 SYSCALL_DEFINE0(inotify_init)
664 {
665 return sys_inotify_init1(0);
666 }
667
SYSCALL_DEFINE3(inotify_add_watch,int,fd,const char __user *,pathname,u32,mask)668 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
669 u32, mask)
670 {
671 struct fsnotify_group *group;
672 struct inode *inode;
673 struct path path;
674 struct path alteredpath;
675 struct path *canonical_path = &path;
676 struct fd f;
677 int ret;
678 unsigned flags = 0;
679
680 /*
681 * We share a lot of code with fs/dnotify. We also share
682 * the bit layout between inotify's IN_* and the fsnotify
683 * FS_*. This check ensures that only the inotify IN_*
684 * bits get passed in and set in watches/events.
685 */
686 if (unlikely(mask & ~ALL_INOTIFY_BITS))
687 return -EINVAL;
688 /*
689 * Require at least one valid bit set in the mask.
690 * Without _something_ set, we would have no events to
691 * watch for.
692 */
693 if (unlikely(!(mask & ALL_INOTIFY_BITS)))
694 return -EINVAL;
695
696 f = fdget(fd);
697 if (unlikely(!f.file))
698 return -EBADF;
699
700 /* verify that this is indeed an inotify instance */
701 if (unlikely(f.file->f_op != &inotify_fops)) {
702 ret = -EINVAL;
703 goto fput_and_out;
704 }
705
706 if (!(mask & IN_DONT_FOLLOW))
707 flags |= LOOKUP_FOLLOW;
708 if (mask & IN_ONLYDIR)
709 flags |= LOOKUP_DIRECTORY;
710
711 ret = inotify_find_inode(pathname, &path, flags);
712 if (ret)
713 goto fput_and_out;
714
715 /* support stacked filesystems */
716 if(path.dentry && path.dentry->d_op) {
717 if (path.dentry->d_op->d_canonical_path) {
718 path.dentry->d_op->d_canonical_path(&path, &alteredpath);
719 canonical_path = &alteredpath;
720 path_put(&path);
721 }
722 }
723
724 /* inode held in place by reference to path; group by fget on fd */
725 inode = canonical_path->dentry->d_inode;
726 group = f.file->private_data;
727
728 /* create/update an inode mark */
729 ret = inotify_update_watch(group, inode, mask);
730 path_put(canonical_path);
731 fput_and_out:
732 fdput(f);
733 return ret;
734 }
735
SYSCALL_DEFINE2(inotify_rm_watch,int,fd,__s32,wd)736 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
737 {
738 struct fsnotify_group *group;
739 struct inotify_inode_mark *i_mark;
740 struct fd f;
741 int ret = 0;
742
743 f = fdget(fd);
744 if (unlikely(!f.file))
745 return -EBADF;
746
747 /* verify that this is indeed an inotify instance */
748 ret = -EINVAL;
749 if (unlikely(f.file->f_op != &inotify_fops))
750 goto out;
751
752 group = f.file->private_data;
753
754 ret = -EINVAL;
755 i_mark = inotify_idr_find(group, wd);
756 if (unlikely(!i_mark))
757 goto out;
758
759 ret = 0;
760
761 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
762
763 /* match ref taken by inotify_idr_find */
764 fsnotify_put_mark(&i_mark->fsn_mark);
765
766 out:
767 fdput(f);
768 return ret;
769 }
770
771 /*
772 * inotify_user_setup - Our initialization function. Note that we cannot return
773 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
774 * must result in panic().
775 */
inotify_user_setup(void)776 static int __init inotify_user_setup(void)
777 {
778 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
779 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
780 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
781 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
782 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
783 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
784 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
785 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
786 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
787 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
788 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
789 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
790 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
791 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
792 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
793 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
794 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
795 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
796
797 BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
798
799 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
800
801 inotify_max_queued_events = 16384;
802 init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
803 init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
804
805 return 0;
806 }
807 fs_initcall(inotify_user_setup);
808