1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16#include <linux/file.h>
17#include <linux/fs.h>
18#include <linux/fsnotify_backend.h>
19#include <linux/idr.h>
20#include <linux/init.h>
21#include <linux/inotify.h>
22#include <linux/kernel.h>
23#include <linux/namei.h>
24#include <linux/sched/signal.h>
25#include <linux/slab.h>
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
34#include "inotify.h"
35#include "../fdinfo.h"
36
37#include <asm/ioctls.h>
38
39
40static int inotify_max_queued_events __read_mostly;
41
42struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
43
44#ifdef CONFIG_SYSCTL
45
46#include <linux/sysctl.h>
47
48struct ctl_table inotify_table[] = {
49 {
50 .procname = "max_user_instances",
51 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
52 .maxlen = sizeof(int),
53 .mode = 0644,
54 .proc_handler = proc_dointvec_minmax,
55 .extra1 = SYSCTL_ZERO,
56 },
57 {
58 .procname = "max_user_watches",
59 .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
60 .maxlen = sizeof(int),
61 .mode = 0644,
62 .proc_handler = proc_dointvec_minmax,
63 .extra1 = SYSCTL_ZERO,
64 },
65 {
66 .procname = "max_queued_events",
67 .data = &inotify_max_queued_events,
68 .maxlen = sizeof(int),
69 .mode = 0644,
70 .proc_handler = proc_dointvec_minmax,
71 .extra1 = SYSCTL_ZERO
72 },
73 { }
74};
75#endif
76
77static inline __u32 inotify_arg_to_mask(u32 arg)
78{
79 __u32 mask;
80
81
82
83
84
85 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
86
87
88 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
89
90 return mask;
91}
92
93static inline u32 inotify_mask_to_arg(__u32 mask)
94{
95 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
96 IN_Q_OVERFLOW);
97}
98
99
100static __poll_t inotify_poll(struct file *file, poll_table *wait)
101{
102 struct fsnotify_group *group = file->private_data;
103 __poll_t ret = 0;
104
105 poll_wait(file, &group->notification_waitq, wait);
106 spin_lock(&group->notification_lock);
107 if (!fsnotify_notify_queue_is_empty(group))
108 ret = EPOLLIN | EPOLLRDNORM;
109 spin_unlock(&group->notification_lock);
110
111 return ret;
112}
113
114static int round_event_name_len(struct fsnotify_event *fsn_event)
115{
116 struct inotify_event_info *event;
117
118 event = INOTIFY_E(fsn_event);
119 if (!event->name_len)
120 return 0;
121 return roundup(event->name_len + 1, sizeof(struct inotify_event));
122}
123
124
125
126
127
128
129
130
131static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
132 size_t count)
133{
134 size_t event_size = sizeof(struct inotify_event);
135 struct fsnotify_event *event;
136
137 if (fsnotify_notify_queue_is_empty(group))
138 return NULL;
139
140 event = fsnotify_peek_first_event(group);
141
142 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
143
144 event_size += round_event_name_len(event);
145 if (event_size > count)
146 return ERR_PTR(-EINVAL);
147
148
149
150 fsnotify_remove_first_event(group);
151
152 return event;
153}
154
155
156
157
158
159
160
161static ssize_t copy_event_to_user(struct fsnotify_group *group,
162 struct fsnotify_event *fsn_event,
163 char __user *buf)
164{
165 struct inotify_event inotify_event;
166 struct inotify_event_info *event;
167 size_t event_size = sizeof(struct inotify_event);
168 size_t name_len;
169 size_t pad_name_len;
170
171 pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
172
173 event = INOTIFY_E(fsn_event);
174 name_len = event->name_len;
175
176
177
178
179 pad_name_len = round_event_name_len(fsn_event);
180 inotify_event.len = pad_name_len;
181 inotify_event.mask = inotify_mask_to_arg(event->mask);
182 inotify_event.wd = event->wd;
183 inotify_event.cookie = event->sync_cookie;
184
185
186 if (copy_to_user(buf, &inotify_event, event_size))
187 return -EFAULT;
188
189 buf += event_size;
190
191
192
193
194
195
196 if (pad_name_len) {
197
198 if (copy_to_user(buf, event->name, name_len))
199 return -EFAULT;
200 buf += name_len;
201
202
203 if (clear_user(buf, pad_name_len - name_len))
204 return -EFAULT;
205 event_size += pad_name_len;
206 }
207
208 return event_size;
209}
210
211static ssize_t inotify_read(struct file *file, char __user *buf,
212 size_t count, loff_t *pos)
213{
214 struct fsnotify_group *group;
215 struct fsnotify_event *kevent;
216 char __user *start;
217 int ret;
218 DEFINE_WAIT_FUNC(wait, woken_wake_function);
219
220 start = buf;
221 group = file->private_data;
222
223 add_wait_queue(&group->notification_waitq, &wait);
224 while (1) {
225 spin_lock(&group->notification_lock);
226 kevent = get_one_event(group, count);
227 spin_unlock(&group->notification_lock);
228
229 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
230
231 if (kevent) {
232 ret = PTR_ERR(kevent);
233 if (IS_ERR(kevent))
234 break;
235 ret = copy_event_to_user(group, kevent, buf);
236 fsnotify_destroy_event(group, kevent);
237 if (ret < 0)
238 break;
239 buf += ret;
240 count -= ret;
241 continue;
242 }
243
244 ret = -EAGAIN;
245 if (file->f_flags & O_NONBLOCK)
246 break;
247 ret = -ERESTARTSYS;
248 if (signal_pending(current))
249 break;
250
251 if (start != buf)
252 break;
253
254 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
255 }
256 remove_wait_queue(&group->notification_waitq, &wait);
257
258 if (start != buf && ret != -EFAULT)
259 ret = buf - start;
260 return ret;
261}
262
263static int inotify_release(struct inode *ignored, struct file *file)
264{
265 struct fsnotify_group *group = file->private_data;
266
267 pr_debug("%s: group=%p\n", __func__, group);
268
269
270 fsnotify_destroy_group(group);
271
272 return 0;
273}
274
275static long inotify_ioctl(struct file *file, unsigned int cmd,
276 unsigned long arg)
277{
278 struct fsnotify_group *group;
279 struct fsnotify_event *fsn_event;
280 void __user *p;
281 int ret = -ENOTTY;
282 size_t send_len = 0;
283
284 group = file->private_data;
285 p = (void __user *) arg;
286
287 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
288
289 switch (cmd) {
290 case FIONREAD:
291 spin_lock(&group->notification_lock);
292 list_for_each_entry(fsn_event, &group->notification_list,
293 list) {
294 send_len += sizeof(struct inotify_event);
295 send_len += round_event_name_len(fsn_event);
296 }
297 spin_unlock(&group->notification_lock);
298 ret = put_user(send_len, (int __user *) p);
299 break;
300#ifdef CONFIG_CHECKPOINT_RESTORE
301 case INOTIFY_IOC_SETNEXTWD:
302 ret = -EINVAL;
303 if (arg >= 1 && arg <= INT_MAX) {
304 struct inotify_group_private_data *data;
305
306 data = &group->inotify_data;
307 spin_lock(&data->idr_lock);
308 idr_set_cursor(&data->idr, (unsigned int)arg);
309 spin_unlock(&data->idr_lock);
310 ret = 0;
311 }
312 break;
313#endif
314 }
315
316 return ret;
317}
318
319static const struct file_operations inotify_fops = {
320 .show_fdinfo = inotify_show_fdinfo,
321 .poll = inotify_poll,
322 .read = inotify_read,
323 .fasync = fsnotify_fasync,
324 .release = inotify_release,
325 .unlocked_ioctl = inotify_ioctl,
326 .compat_ioctl = inotify_ioctl,
327 .llseek = noop_llseek,
328};
329
330
331
332
333
334static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
335{
336 int error;
337
338 error = user_path_at(AT_FDCWD, dirname, flags, path);
339 if (error)
340 return error;
341
342 error = inode_permission(path->dentry->d_inode, MAY_READ);
343 if (error)
344 path_put(path);
345 return error;
346}
347
348static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
349 struct inotify_inode_mark *i_mark)
350{
351 int ret;
352
353 idr_preload(GFP_KERNEL);
354 spin_lock(idr_lock);
355
356 ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
357 if (ret >= 0) {
358
359 i_mark->wd = ret;
360 fsnotify_get_mark(&i_mark->fsn_mark);
361 }
362
363 spin_unlock(idr_lock);
364 idr_preload_end();
365 return ret < 0 ? ret : 0;
366}
367
368static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
369 int wd)
370{
371 struct idr *idr = &group->inotify_data.idr;
372 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
373 struct inotify_inode_mark *i_mark;
374
375 assert_spin_locked(idr_lock);
376
377 i_mark = idr_find(idr, wd);
378 if (i_mark) {
379 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
380
381 fsnotify_get_mark(fsn_mark);
382
383 BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
384 }
385
386 return i_mark;
387}
388
389static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
390 int wd)
391{
392 struct inotify_inode_mark *i_mark;
393 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
394
395 spin_lock(idr_lock);
396 i_mark = inotify_idr_find_locked(group, wd);
397 spin_unlock(idr_lock);
398
399 return i_mark;
400}
401
402
403
404
405
406static void inotify_remove_from_idr(struct fsnotify_group *group,
407 struct inotify_inode_mark *i_mark)
408{
409 struct idr *idr = &group->inotify_data.idr;
410 spinlock_t *idr_lock = &group->inotify_data.idr_lock;
411 struct inotify_inode_mark *found_i_mark = NULL;
412 int wd;
413
414 spin_lock(idr_lock);
415 wd = i_mark->wd;
416
417
418
419
420
421 if (wd == -1) {
422 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
423 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
424 goto out;
425 }
426
427
428 found_i_mark = inotify_idr_find_locked(group, wd);
429 if (unlikely(!found_i_mark)) {
430 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
431 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
432 goto out;
433 }
434
435
436
437
438
439
440 if (unlikely(found_i_mark != i_mark)) {
441 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
442 "found_i_mark=%p found_i_mark->wd=%d "
443 "found_i_mark->group=%p\n", __func__, i_mark,
444 i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
445 found_i_mark->wd, found_i_mark->fsn_mark.group);
446 goto out;
447 }
448
449
450
451
452
453 if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
454 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
455 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
456
457 BUG();
458 }
459
460 idr_remove(idr, wd);
461
462 fsnotify_put_mark(&i_mark->fsn_mark);
463out:
464 i_mark->wd = -1;
465 spin_unlock(idr_lock);
466
467 if (found_i_mark)
468 fsnotify_put_mark(&found_i_mark->fsn_mark);
469}
470
471
472
473
474void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
475 struct fsnotify_group *group)
476{
477 struct inotify_inode_mark *i_mark;
478 struct fsnotify_iter_info iter_info = { };
479
480 fsnotify_iter_set_report_type_mark(&iter_info, FSNOTIFY_OBJ_TYPE_INODE,
481 fsn_mark);
482
483
484 inotify_handle_event(group, NULL, FS_IN_IGNORED, NULL,
485 FSNOTIFY_EVENT_NONE, NULL, 0, &iter_info);
486
487 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
488
489 inotify_remove_from_idr(group, i_mark);
490
491 dec_inotify_watches(group->inotify_data.ucounts);
492}
493
494static int inotify_update_existing_watch(struct fsnotify_group *group,
495 struct inode *inode,
496 u32 arg)
497{
498 struct fsnotify_mark *fsn_mark;
499 struct inotify_inode_mark *i_mark;
500 __u32 old_mask, new_mask;
501 __u32 mask;
502 int add = (arg & IN_MASK_ADD);
503 int create = (arg & IN_MASK_CREATE);
504 int ret;
505
506 mask = inotify_arg_to_mask(arg);
507
508 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
509 if (!fsn_mark)
510 return -ENOENT;
511 else if (create) {
512 ret = -EEXIST;
513 goto out;
514 }
515
516 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
517
518 spin_lock(&fsn_mark->lock);
519 old_mask = fsn_mark->mask;
520 if (add)
521 fsn_mark->mask |= mask;
522 else
523 fsn_mark->mask = mask;
524 new_mask = fsn_mark->mask;
525 spin_unlock(&fsn_mark->lock);
526
527 if (old_mask != new_mask) {
528
529 int dropped = (old_mask & ~new_mask);
530
531 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
532
533
534 if (dropped || do_inode)
535 fsnotify_recalc_mask(inode->i_fsnotify_marks);
536
537 }
538
539
540 ret = i_mark->wd;
541
542out:
543
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
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
581 ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
582 if (ret) {
583
584 inotify_remove_from_idr(group, tmp_i_mark);
585 goto out_err;
586 }
587
588
589
590 ret = tmp_i_mark->wd;
591
592out_err:
593
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
605 ret = inotify_update_existing_watch(group, inode, arg);
606
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);
630 oevent->mask = FS_Q_OVERFLOW;
631 oevent->wd = -1;
632 oevent->sync_cookie = 0;
633 oevent->name_len = 0;
634
635 group->max_events = max_events;
636 group->memcg = get_mem_cgroup_from_mm(current->mm);
637
638 spin_lock_init(&group->inotify_data.idr_lock);
639 idr_init(&group->inotify_data.idr);
640 group->inotify_data.ucounts = inc_ucount(current_user_ns(),
641 current_euid(),
642 UCOUNT_INOTIFY_INSTANCES);
643
644 if (!group->inotify_data.ucounts) {
645 fsnotify_destroy_group(group);
646 return ERR_PTR(-EMFILE);
647 }
648
649 return group;
650}
651
652
653
654static int do_inotify_init(int flags)
655{
656 struct fsnotify_group *group;
657 int ret;
658
659
660 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
661 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
662
663 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
664 return -EINVAL;
665
666
667 group = inotify_new_group(inotify_max_queued_events);
668 if (IS_ERR(group))
669 return PTR_ERR(group);
670
671 ret = anon_inode_getfd("inotify", &inotify_fops, group,
672 O_RDONLY | flags);
673 if (ret < 0)
674 fsnotify_destroy_group(group);
675
676 return ret;
677}
678
679SYSCALL_DEFINE1(inotify_init1, int, flags)
680{
681 return do_inotify_init(flags);
682}
683
684SYSCALL_DEFINE0(inotify_init)
685{
686 return do_inotify_init(0);
687}
688
689SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
690 u32, mask)
691{
692 struct fsnotify_group *group;
693 struct inode *inode;
694 struct path path;
695 struct fd f;
696 int ret;
697 unsigned flags = 0;
698
699
700
701
702
703
704
705 if (unlikely(mask & ~ALL_INOTIFY_BITS))
706 return -EINVAL;
707
708
709
710
711
712 if (unlikely(!(mask & ALL_INOTIFY_BITS)))
713 return -EINVAL;
714
715 f = fdget(fd);
716 if (unlikely(!f.file))
717 return -EBADF;
718
719
720 if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
721 ret = -EINVAL;
722 goto fput_and_out;
723 }
724
725
726 if (unlikely(f.file->f_op != &inotify_fops)) {
727 ret = -EINVAL;
728 goto fput_and_out;
729 }
730
731 if (!(mask & IN_DONT_FOLLOW))
732 flags |= LOOKUP_FOLLOW;
733 if (mask & IN_ONLYDIR)
734 flags |= LOOKUP_DIRECTORY;
735
736 ret = inotify_find_inode(pathname, &path, flags);
737 if (ret)
738 goto fput_and_out;
739
740
741 inode = path.dentry->d_inode;
742 group = f.file->private_data;
743
744
745 ret = inotify_update_watch(group, inode, mask);
746 path_put(&path);
747fput_and_out:
748 fdput(f);
749 return ret;
750}
751
752SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
753{
754 struct fsnotify_group *group;
755 struct inotify_inode_mark *i_mark;
756 struct fd f;
757 int ret = 0;
758
759 f = fdget(fd);
760 if (unlikely(!f.file))
761 return -EBADF;
762
763
764 ret = -EINVAL;
765 if (unlikely(f.file->f_op != &inotify_fops))
766 goto out;
767
768 group = f.file->private_data;
769
770 ret = -EINVAL;
771 i_mark = inotify_idr_find(group, wd);
772 if (unlikely(!i_mark))
773 goto out;
774
775 ret = 0;
776
777 fsnotify_destroy_mark(&i_mark->fsn_mark, group);
778
779
780 fsnotify_put_mark(&i_mark->fsn_mark);
781
782out:
783 fdput(f);
784 return ret;
785}
786
787
788
789
790
791
792static int __init inotify_user_setup(void)
793{
794 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
795 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
796 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
797 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
798 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
799 BUILD_BUG_ON(IN_OPEN != FS_OPEN);
800 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
801 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
802 BUILD_BUG_ON(IN_CREATE != FS_CREATE);
803 BUILD_BUG_ON(IN_DELETE != FS_DELETE);
804 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
805 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
806 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
807 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
808 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
809 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
810 BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
811 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
812
813 BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
814
815 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
816 SLAB_PANIC|SLAB_ACCOUNT);
817
818 inotify_max_queued_events = 16384;
819 init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
820 init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
821
822 return 0;
823}
824fs_initcall(inotify_user_setup);
825