1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25#include <linux/file.h>
26#include <linux/fs.h>
27#include <linux/fsnotify_backend.h>
28#include <linux/idr.h>
29#include <linux/init.h>
30#include <linux/inotify.h>
31#include <linux/kernel.h>
32#include <linux/magic.h>
33#include <linux/mount.h>
34#include <linux/namei.h>
35#include <linux/path.h>
36#include <linux/sched.h>
37#include <linux/slab.h>
38#include <linux/syscalls.h>
39#include <linux/types.h>
40#include <linux/uaccess.h>
41#include <linux/poll.h>
42#include <linux/wait.h>
43
44#include "inotify.h"
45
46#include <asm/ioctls.h>
47
48static struct vfsmount *inotify_mnt __read_mostly;
49
50
51static int inotify_max_user_instances __read_mostly;
52static int inotify_max_queued_events __read_mostly;
53int inotify_max_user_watches __read_mostly;
54
55static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
56struct kmem_cache *event_priv_cachep __read_mostly;
57
58
59
60
61
62static atomic_t inotify_grp_num;
63
64#ifdef CONFIG_SYSCTL
65
66#include <linux/sysctl.h>
67
68static int zero;
69
70ctl_table inotify_table[] = {
71 {
72 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
73 .procname = "max_user_instances",
74 .data = &inotify_max_user_instances,
75 .maxlen = sizeof(int),
76 .mode = 0644,
77 .proc_handler = &proc_dointvec_minmax,
78 .strategy = &sysctl_intvec,
79 .extra1 = &zero,
80 },
81 {
82 .ctl_name = INOTIFY_MAX_USER_WATCHES,
83 .procname = "max_user_watches",
84 .data = &inotify_max_user_watches,
85 .maxlen = sizeof(int),
86 .mode = 0644,
87 .proc_handler = &proc_dointvec_minmax,
88 .strategy = &sysctl_intvec,
89 .extra1 = &zero,
90 },
91 {
92 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
93 .procname = "max_queued_events",
94 .data = &inotify_max_queued_events,
95 .maxlen = sizeof(int),
96 .mode = 0644,
97 .proc_handler = &proc_dointvec_minmax,
98 .strategy = &sysctl_intvec,
99 .extra1 = &zero
100 },
101 { .ctl_name = 0 }
102};
103#endif
104
105static inline __u32 inotify_arg_to_mask(u32 arg)
106{
107 __u32 mask;
108
109
110 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
111
112
113 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
114
115 return mask;
116}
117
118static inline u32 inotify_mask_to_arg(__u32 mask)
119{
120 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
121 IN_Q_OVERFLOW);
122}
123
124
125static unsigned int inotify_poll(struct file *file, poll_table *wait)
126{
127 struct fsnotify_group *group = file->private_data;
128 int ret = 0;
129
130 poll_wait(file, &group->notification_waitq, wait);
131 mutex_lock(&group->notification_mutex);
132 if (!fsnotify_notify_queue_is_empty(group))
133 ret = POLLIN | POLLRDNORM;
134 mutex_unlock(&group->notification_mutex);
135
136 return ret;
137}
138
139
140
141
142
143
144
145
146static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
147 size_t count)
148{
149 size_t event_size = sizeof(struct inotify_event);
150 struct fsnotify_event *event;
151
152 if (fsnotify_notify_queue_is_empty(group))
153 return NULL;
154
155 event = fsnotify_peek_notify_event(group);
156
157 if (event->name_len)
158 event_size += roundup(event->name_len + 1, event_size);
159
160 if (event_size > count)
161 return ERR_PTR(-EINVAL);
162
163
164
165 fsnotify_remove_notify_event(group);
166
167 return event;
168}
169
170
171
172
173
174
175
176static ssize_t copy_event_to_user(struct fsnotify_group *group,
177 struct fsnotify_event *event,
178 char __user *buf)
179{
180 struct inotify_event inotify_event;
181 struct fsnotify_event_private_data *fsn_priv;
182 struct inotify_event_private_data *priv;
183 size_t event_size = sizeof(struct inotify_event);
184 size_t name_len = 0;
185
186
187 spin_lock(&event->lock);
188 fsn_priv = fsnotify_remove_priv_from_event(group, event);
189 spin_unlock(&event->lock);
190
191 if (!fsn_priv)
192 inotify_event.wd = -1;
193 else {
194 priv = container_of(fsn_priv, struct inotify_event_private_data,
195 fsnotify_event_priv_data);
196 inotify_event.wd = priv->wd;
197 inotify_free_event_priv(fsn_priv);
198 }
199
200
201
202
203
204 if (event->name_len)
205 name_len = roundup(event->name_len + 1, event_size);
206 inotify_event.len = name_len;
207
208 inotify_event.mask = inotify_mask_to_arg(event->mask);
209 inotify_event.cookie = event->sync_cookie;
210
211
212 if (copy_to_user(buf, &inotify_event, event_size))
213 return -EFAULT;
214
215 buf += event_size;
216
217
218
219
220
221
222 if (name_len) {
223 unsigned int len_to_zero = name_len - event->name_len;
224
225 if (copy_to_user(buf, event->file_name, event->name_len))
226 return -EFAULT;
227 buf += event->name_len;
228
229
230 if (clear_user(buf, len_to_zero))
231 return -EFAULT;
232 buf += len_to_zero;
233 event_size += name_len;
234 }
235
236 return event_size;
237}
238
239static ssize_t inotify_read(struct file *file, char __user *buf,
240 size_t count, loff_t *pos)
241{
242 struct fsnotify_group *group;
243 struct fsnotify_event *kevent;
244 char __user *start;
245 int ret;
246 DEFINE_WAIT(wait);
247
248 start = buf;
249 group = file->private_data;
250
251 while (1) {
252 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
253
254 mutex_lock(&group->notification_mutex);
255 kevent = get_one_event(group, count);
256 mutex_unlock(&group->notification_mutex);
257
258 if (kevent) {
259 ret = PTR_ERR(kevent);
260 if (IS_ERR(kevent))
261 break;
262 ret = copy_event_to_user(group, kevent, buf);
263 fsnotify_put_event(kevent);
264 if (ret < 0)
265 break;
266 buf += ret;
267 count -= ret;
268 continue;
269 }
270
271 ret = -EAGAIN;
272 if (file->f_flags & O_NONBLOCK)
273 break;
274 ret = -EINTR;
275 if (signal_pending(current))
276 break;
277
278 if (start != buf)
279 break;
280
281 schedule();
282 }
283
284 finish_wait(&group->notification_waitq, &wait);
285 if (start != buf && ret != -EFAULT)
286 ret = buf - start;
287 return ret;
288}
289
290static int inotify_fasync(int fd, struct file *file, int on)
291{
292 struct fsnotify_group *group = file->private_data;
293
294 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
295}
296
297static int inotify_release(struct inode *ignored, struct file *file)
298{
299 struct fsnotify_group *group = file->private_data;
300 struct user_struct *user = group->inotify_data.user;
301
302 fsnotify_clear_marks_by_group(group);
303
304
305 fsnotify_put_group(group);
306
307 atomic_dec(&user->inotify_devs);
308
309 return 0;
310}
311
312static long inotify_ioctl(struct file *file, unsigned int cmd,
313 unsigned long arg)
314{
315 struct fsnotify_group *group;
316 struct fsnotify_event_holder *holder;
317 struct fsnotify_event *event;
318 void __user *p;
319 int ret = -ENOTTY;
320 size_t send_len = 0;
321
322 group = file->private_data;
323 p = (void __user *) arg;
324
325 switch (cmd) {
326 case FIONREAD:
327 mutex_lock(&group->notification_mutex);
328 list_for_each_entry(holder, &group->notification_list, event_list) {
329 event = holder->event;
330 send_len += sizeof(struct inotify_event);
331 if (event->name_len)
332 send_len += roundup(event->name_len + 1,
333 sizeof(struct inotify_event));
334 }
335 mutex_unlock(&group->notification_mutex);
336 ret = put_user(send_len, (int __user *) p);
337 break;
338 }
339
340 return ret;
341}
342
343static const struct file_operations inotify_fops = {
344 .poll = inotify_poll,
345 .read = inotify_read,
346 .fasync = inotify_fasync,
347 .release = inotify_release,
348 .unlocked_ioctl = inotify_ioctl,
349 .compat_ioctl = inotify_ioctl,
350};
351
352
353
354
355
356static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
357{
358 int error;
359
360 error = user_path_at(AT_FDCWD, dirname, flags, path);
361 if (error)
362 return error;
363
364 error = inode_permission(path->dentry->d_inode, MAY_READ);
365 if (error)
366 path_put(path);
367 return error;
368}
369
370
371
372
373
374static void inotify_remove_from_idr(struct fsnotify_group *group,
375 struct inotify_inode_mark_entry *ientry)
376{
377 struct idr *idr;
378 struct fsnotify_mark_entry *entry;
379 struct inotify_inode_mark_entry *found_ientry;
380 int wd;
381
382 spin_lock(&group->inotify_data.idr_lock);
383 idr = &group->inotify_data.idr;
384 wd = ientry->wd;
385
386 if (wd == -1)
387 goto out;
388
389 entry = idr_find(&group->inotify_data.idr, wd);
390 if (unlikely(!entry))
391 goto out;
392
393 found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
394 if (unlikely(found_ientry != ientry)) {
395
396
397
398 WARN_ON(1);
399 ientry->wd = -1;
400 goto out;
401 }
402
403
404 BUG_ON(atomic_read(&entry->refcnt) < 2);
405
406 idr_remove(idr, wd);
407 ientry->wd = -1;
408
409
410 fsnotify_put_mark(entry);
411out:
412 spin_unlock(&group->inotify_data.idr_lock);
413}
414
415
416
417
418void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
419 struct fsnotify_group *group)
420{
421 struct inotify_inode_mark_entry *ientry;
422 struct fsnotify_event *ignored_event;
423 struct inotify_event_private_data *event_priv;
424 struct fsnotify_event_private_data *fsn_event_priv;
425 int ret;
426
427 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
428 FSNOTIFY_EVENT_NONE, NULL, 0,
429 GFP_NOFS);
430 if (!ignored_event)
431 return;
432
433 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
434
435 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
436 if (unlikely(!event_priv))
437 goto skip_send_ignore;
438
439 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
440
441 fsn_event_priv->group = group;
442 event_priv->wd = ientry->wd;
443
444 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
445 if (ret)
446 inotify_free_event_priv(fsn_event_priv);
447
448skip_send_ignore:
449
450
451 fsnotify_put_event(ignored_event);
452
453
454 inotify_remove_from_idr(group, ientry);
455
456 atomic_dec(&group->inotify_data.user->inotify_watches);
457}
458
459
460static void inotify_free_mark(struct fsnotify_mark_entry *entry)
461{
462 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
463
464 kmem_cache_free(inotify_inode_mark_cachep, ientry);
465}
466
467static int inotify_update_existing_watch(struct fsnotify_group *group,
468 struct inode *inode,
469 u32 arg)
470{
471 struct fsnotify_mark_entry *entry;
472 struct inotify_inode_mark_entry *ientry;
473 __u32 old_mask, new_mask;
474 __u32 mask;
475 int add = (arg & IN_MASK_ADD);
476 int ret;
477
478
479 mask = inotify_arg_to_mask(arg);
480 if (unlikely(!mask))
481 return -EINVAL;
482
483 spin_lock(&inode->i_lock);
484 entry = fsnotify_find_mark_entry(group, inode);
485 spin_unlock(&inode->i_lock);
486 if (!entry)
487 return -ENOENT;
488
489 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
490
491 spin_lock(&entry->lock);
492
493 old_mask = entry->mask;
494 if (add) {
495 entry->mask |= mask;
496 new_mask = entry->mask;
497 } else {
498 entry->mask = mask;
499 new_mask = entry->mask;
500 }
501
502 spin_unlock(&entry->lock);
503
504 if (old_mask != new_mask) {
505
506 int dropped = (old_mask & ~new_mask);
507
508 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
509
510 int do_group = (new_mask & ~group->mask);
511
512
513 if (dropped || do_inode)
514 fsnotify_recalc_inode_mask(inode);
515
516
517 if (dropped || do_group)
518 fsnotify_recalc_group_mask(group);
519 }
520
521
522 ret = ientry->wd;
523
524
525 fsnotify_put_mark(entry);
526
527 return ret;
528}
529
530static int inotify_new_watch(struct fsnotify_group *group,
531 struct inode *inode,
532 u32 arg)
533{
534 struct inotify_inode_mark_entry *tmp_ientry;
535 __u32 mask;
536 int ret;
537
538
539 mask = inotify_arg_to_mask(arg);
540 if (unlikely(!mask))
541 return -EINVAL;
542
543 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
544 if (unlikely(!tmp_ientry))
545 return -ENOMEM;
546
547 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
548 tmp_ientry->fsn_entry.mask = mask;
549 tmp_ientry->wd = -1;
550
551 ret = -ENOSPC;
552 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
553 goto out_err;
554retry:
555 ret = -ENOMEM;
556 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
557 goto out_err;
558
559 spin_lock(&group->inotify_data.idr_lock);
560 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
561 group->inotify_data.last_wd,
562 &tmp_ientry->wd);
563 spin_unlock(&group->inotify_data.idr_lock);
564 if (ret) {
565
566 if (ret == -EAGAIN)
567 goto retry;
568 goto out_err;
569 }
570
571
572 fsnotify_get_mark(&tmp_ientry->fsn_entry);
573
574
575 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
576 if (ret) {
577
578 inotify_remove_from_idr(group, tmp_ientry);
579 goto out_err;
580 }
581
582
583 group->inotify_data.last_wd = tmp_ientry->wd;
584
585
586 atomic_inc(&group->inotify_data.user->inotify_watches);
587
588
589 ret = tmp_ientry->wd;
590
591
592 fsnotify_put_mark(&tmp_ientry->fsn_entry);
593
594
595 if (mask & ~group->mask)
596 fsnotify_recalc_group_mask(group);
597
598out_err:
599 if (ret < 0)
600 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
601
602 return ret;
603}
604
605static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
606{
607 int ret = 0;
608
609retry:
610
611 ret = inotify_update_existing_watch(group, inode, arg);
612
613 if (ret == -ENOENT)
614 ret = inotify_new_watch(group, inode, arg);
615
616
617
618
619
620 if (ret == -EEXIST)
621 goto retry;
622
623 return ret;
624}
625
626static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
627{
628 struct fsnotify_group *group;
629 unsigned int grp_num;
630
631
632 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
633 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
634 if (IS_ERR(group))
635 return group;
636
637 group->max_events = max_events;
638
639 spin_lock_init(&group->inotify_data.idr_lock);
640 idr_init(&group->inotify_data.idr);
641 group->inotify_data.last_wd = 1;
642 group->inotify_data.user = user;
643 group->inotify_data.fa = NULL;
644
645 return group;
646}
647
648
649
650SYSCALL_DEFINE1(inotify_init1, int, flags)
651{
652 struct fsnotify_group *group;
653 struct user_struct *user;
654 struct file *filp;
655 int fd, ret;
656
657
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 fd = get_unused_fd_flags(flags & O_CLOEXEC);
665 if (fd < 0)
666 return fd;
667
668 filp = get_empty_filp();
669 if (!filp) {
670 ret = -ENFILE;
671 goto out_put_fd;
672 }
673
674 user = get_current_user();
675 if (unlikely(atomic_read(&user->inotify_devs) >=
676 inotify_max_user_instances)) {
677 ret = -EMFILE;
678 goto out_free_uid;
679 }
680
681
682 group = inotify_new_group(user, inotify_max_queued_events);
683 if (IS_ERR(group)) {
684 ret = PTR_ERR(group);
685 goto out_free_uid;
686 }
687
688 filp->f_op = &inotify_fops;
689 filp->f_path.mnt = mntget(inotify_mnt);
690 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
691 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
692 filp->f_mode = FMODE_READ;
693 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
694 filp->private_data = group;
695
696 atomic_inc(&user->inotify_devs);
697
698 fd_install(fd, filp);
699
700 return fd;
701
702out_free_uid:
703 free_uid(user);
704 put_filp(filp);
705out_put_fd:
706 put_unused_fd(fd);
707 return ret;
708}
709
710SYSCALL_DEFINE0(inotify_init)
711{
712 return sys_inotify_init1(0);
713}
714
715SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
716 u32, mask)
717{
718 struct fsnotify_group *group;
719 struct inode *inode;
720 struct path path;
721 struct file *filp;
722 int ret, fput_needed;
723 unsigned flags = 0;
724
725 filp = fget_light(fd, &fput_needed);
726 if (unlikely(!filp))
727 return -EBADF;
728
729
730 if (unlikely(filp->f_op != &inotify_fops)) {
731 ret = -EINVAL;
732 goto fput_and_out;
733 }
734
735 if (!(mask & IN_DONT_FOLLOW))
736 flags |= LOOKUP_FOLLOW;
737 if (mask & IN_ONLYDIR)
738 flags |= LOOKUP_DIRECTORY;
739
740 ret = inotify_find_inode(pathname, &path, flags);
741 if (ret)
742 goto fput_and_out;
743
744
745 inode = path.dentry->d_inode;
746 group = filp->private_data;
747
748
749 ret = inotify_update_watch(group, inode, mask);
750 if (unlikely(ret))
751 goto path_put_and_out;
752
753path_put_and_out:
754 path_put(&path);
755fput_and_out:
756 fput_light(filp, fput_needed);
757 return ret;
758}
759
760SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
761{
762 struct fsnotify_group *group;
763 struct fsnotify_mark_entry *entry;
764 struct file *filp;
765 int ret = 0, fput_needed;
766
767 filp = fget_light(fd, &fput_needed);
768 if (unlikely(!filp))
769 return -EBADF;
770
771
772 if (unlikely(filp->f_op != &inotify_fops)) {
773 ret = -EINVAL;
774 goto out;
775 }
776
777 group = filp->private_data;
778
779 spin_lock(&group->inotify_data.idr_lock);
780 entry = idr_find(&group->inotify_data.idr, wd);
781 if (unlikely(!entry)) {
782 spin_unlock(&group->inotify_data.idr_lock);
783 ret = -EINVAL;
784 goto out;
785 }
786 fsnotify_get_mark(entry);
787 spin_unlock(&group->inotify_data.idr_lock);
788
789 fsnotify_destroy_mark_by_entry(entry);
790 fsnotify_put_mark(entry);
791
792out:
793 fput_light(filp, fput_needed);
794 return ret;
795}
796
797static int
798inotify_get_sb(struct file_system_type *fs_type, int flags,
799 const char *dev_name, void *data, struct vfsmount *mnt)
800{
801 return get_sb_pseudo(fs_type, "inotify", NULL,
802 INOTIFYFS_SUPER_MAGIC, mnt);
803}
804
805static struct file_system_type inotify_fs_type = {
806 .name = "inotifyfs",
807 .get_sb = inotify_get_sb,
808 .kill_sb = kill_anon_super,
809};
810
811
812
813
814
815
816static int __init inotify_user_setup(void)
817{
818 int ret;
819
820 ret = register_filesystem(&inotify_fs_type);
821 if (unlikely(ret))
822 panic("inotify: register_filesystem returned %d!\n", ret);
823
824 inotify_mnt = kern_mount(&inotify_fs_type);
825 if (IS_ERR(inotify_mnt))
826 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
827
828 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
829 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
830
831 inotify_max_queued_events = 16384;
832 inotify_max_user_instances = 128;
833 inotify_max_user_watches = 8192;
834
835 return 0;
836}
837module_init(inotify_user_setup);
838