Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/fanotify.h>
3#include <linux/fcntl.h>
4#include <linux/file.h>
5#include <linux/fs.h>
6#include <linux/anon_inodes.h>
7#include <linux/fsnotify_backend.h>
8#include <linux/init.h>
9#include <linux/mount.h>
10#include <linux/namei.h>
11#include <linux/poll.h>
12#include <linux/security.h>
13#include <linux/syscalls.h>
14#include <linux/slab.h>
15#include <linux/types.h>
16#include <linux/uaccess.h>
17#include <linux/compat.h>
18#include <linux/sched/signal.h>
19
20#include <asm/ioctls.h>
21
22#include "../../mount.h"
23#include "../fdinfo.h"
24#include "fanotify.h"
25
26#define FANOTIFY_DEFAULT_MAX_EVENTS 16384
27#define FANOTIFY_DEFAULT_MAX_MARKS 8192
28#define FANOTIFY_DEFAULT_MAX_LISTENERS 128
29
30/*
31 * All flags that may be specified in parameter event_f_flags of fanotify_init.
32 *
33 * Internal and external open flags are stored together in field f_flags of
34 * struct file. Only external open flags shall be allowed in event_f_flags.
35 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
36 * excluded.
37 */
38#define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \
39 O_ACCMODE | O_APPEND | O_NONBLOCK | \
40 __O_SYNC | O_DSYNC | O_CLOEXEC | \
41 O_LARGEFILE | O_NOATIME )
42
43extern const struct fsnotify_ops fanotify_fsnotify_ops;
44
45struct kmem_cache *fanotify_mark_cache __read_mostly;
46struct kmem_cache *fanotify_event_cachep __read_mostly;
47struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
48
49/*
50 * Get an fsnotify notification event if one exists and is small
51 * enough to fit in "count". Return an error pointer if the count
52 * is not large enough.
53 *
54 * Called with the group->notification_lock held.
55 */
56static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
57 size_t count)
58{
59 assert_spin_locked(&group->notification_lock);
60
61 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
62
63 if (fsnotify_notify_queue_is_empty(group))
64 return NULL;
65
66 if (FAN_EVENT_METADATA_LEN > count)
67 return ERR_PTR(-EINVAL);
68
69 /* held the notification_lock the whole time, so this is the
70 * same event we peeked above */
71 return fsnotify_remove_first_event(group);
72}
73
74static int create_fd(struct fsnotify_group *group,
75 struct fanotify_event_info *event,
76 struct file **file)
77{
78 int client_fd;
79 struct file *new_file;
80
81 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
82
83 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
84 if (client_fd < 0)
85 return client_fd;
86
87 /*
88 * we need a new file handle for the userspace program so it can read even if it was
89 * originally opened O_WRONLY.
90 */
91 /* it's possible this event was an overflow event. in that case dentry and mnt
92 * are NULL; That's fine, just don't call dentry open */
93 if (event->path.dentry && event->path.mnt)
94 new_file = dentry_open(&event->path,
95 group->fanotify_data.f_flags | FMODE_NONOTIFY,
96 current_cred());
97 else
98 new_file = ERR_PTR(-EOVERFLOW);
99 if (IS_ERR(new_file)) {
100 /*
101 * we still send an event even if we can't open the file. this
102 * can happen when say tasks are gone and we try to open their
103 * /proc files or we try to open a WRONLY file like in sysfs
104 * we just send the errno to userspace since there isn't much
105 * else we can do.
106 */
107 put_unused_fd(client_fd);
108 client_fd = PTR_ERR(new_file);
109 } else {
110 *file = new_file;
111 }
112
113 return client_fd;
114}
115
116static int fill_event_metadata(struct fsnotify_group *group,
117 struct fanotify_event_metadata *metadata,
118 struct fsnotify_event *fsn_event,
119 struct file **file)
120{
121 int ret = 0;
122 struct fanotify_event_info *event;
123
124 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
125 group, metadata, fsn_event);
126
127 *file = NULL;
128 event = container_of(fsn_event, struct fanotify_event_info, fse);
129 metadata->event_len = FAN_EVENT_METADATA_LEN;
130 metadata->metadata_len = FAN_EVENT_METADATA_LEN;
131 metadata->vers = FANOTIFY_METADATA_VERSION;
132 metadata->reserved = 0;
133 metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
134 metadata->pid = pid_vnr(event->tgid);
135 if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
136 metadata->fd = FAN_NOFD;
137 else {
138 metadata->fd = create_fd(group, event, file);
139 if (metadata->fd < 0)
140 ret = metadata->fd;
141 }
142
143 return ret;
144}
145
146static struct fanotify_perm_event_info *dequeue_event(
147 struct fsnotify_group *group, int fd)
148{
149 struct fanotify_perm_event_info *event, *return_e = NULL;
150
151 spin_lock(&group->notification_lock);
152 list_for_each_entry(event, &group->fanotify_data.access_list,
153 fae.fse.list) {
154 if (event->fd != fd)
155 continue;
156
157 list_del_init(&event->fae.fse.list);
158 return_e = event;
159 break;
160 }
161 spin_unlock(&group->notification_lock);
162
163 pr_debug("%s: found return_re=%p\n", __func__, return_e);
164
165 return return_e;
166}
167
168static int process_access_response(struct fsnotify_group *group,
169 struct fanotify_response *response_struct)
170{
171 struct fanotify_perm_event_info *event;
172 int fd = response_struct->fd;
173 int response = response_struct->response;
174
175 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
176 fd, response);
177 /*
178 * make sure the response is valid, if invalid we do nothing and either
179 * userspace can send a valid response or we will clean it up after the
180 * timeout
181 */
182 switch (response & ~FAN_AUDIT) {
183 case FAN_ALLOW:
184 case FAN_DENY:
185 break;
186 default:
187 return -EINVAL;
188 }
189
190 if (fd < 0)
191 return -EINVAL;
192
193 if ((response & FAN_AUDIT) && !group->fanotify_data.audit)
194 return -EINVAL;
195
196 event = dequeue_event(group, fd);
197 if (!event)
198 return -ENOENT;
199
200 event->response = response;
201 wake_up(&group->fanotify_data.access_waitq);
202
203 return 0;
204}
205
206static ssize_t copy_event_to_user(struct fsnotify_group *group,
207 struct fsnotify_event *event,
208 char __user *buf)
209{
210 struct fanotify_event_metadata fanotify_event_metadata;
211 struct file *f;
212 int fd, ret;
213
214 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
215
216 ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
217 if (ret < 0)
218 return ret;
219
220 fd = fanotify_event_metadata.fd;
221 ret = -EFAULT;
222 if (copy_to_user(buf, &fanotify_event_metadata,
223 fanotify_event_metadata.event_len))
224 goto out_close_fd;
225
226 if (fanotify_is_perm_event(event->mask))
227 FANOTIFY_PE(event)->fd = fd;
228
229 if (fd != FAN_NOFD)
230 fd_install(fd, f);
231 return fanotify_event_metadata.event_len;
232
233out_close_fd:
234 if (fd != FAN_NOFD) {
235 put_unused_fd(fd);
236 fput(f);
237 }
238 return ret;
239}
240
241/* intofiy userspace file descriptor functions */
242static __poll_t fanotify_poll(struct file *file, poll_table *wait)
243{
244 struct fsnotify_group *group = file->private_data;
245 __poll_t ret = 0;
246
247 poll_wait(file, &group->notification_waitq, wait);
248 spin_lock(&group->notification_lock);
249 if (!fsnotify_notify_queue_is_empty(group))
250 ret = EPOLLIN | EPOLLRDNORM;
251 spin_unlock(&group->notification_lock);
252
253 return ret;
254}
255
256static ssize_t fanotify_read(struct file *file, char __user *buf,
257 size_t count, loff_t *pos)
258{
259 struct fsnotify_group *group;
260 struct fsnotify_event *kevent;
261 char __user *start;
262 int ret;
263 DEFINE_WAIT_FUNC(wait, woken_wake_function);
264
265 start = buf;
266 group = file->private_data;
267
268 pr_debug("%s: group=%p\n", __func__, group);
269
270 add_wait_queue(&group->notification_waitq, &wait);
271 while (1) {
272 spin_lock(&group->notification_lock);
273 kevent = get_one_event(group, count);
274 spin_unlock(&group->notification_lock);
275
276 if (IS_ERR(kevent)) {
277 ret = PTR_ERR(kevent);
278 break;
279 }
280
281 if (!kevent) {
282 ret = -EAGAIN;
283 if (file->f_flags & O_NONBLOCK)
284 break;
285
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 continue;
295 }
296
297 ret = copy_event_to_user(group, kevent, buf);
298 if (unlikely(ret == -EOPENSTALE)) {
299 /*
300 * We cannot report events with stale fd so drop it.
301 * Setting ret to 0 will continue the event loop and
302 * do the right thing if there are no more events to
303 * read (i.e. return bytes read, -EAGAIN or wait).
304 */
305 ret = 0;
306 }
307
308 /*
309 * Permission events get queued to wait for response. Other
310 * events can be destroyed now.
311 */
312 if (!fanotify_is_perm_event(kevent->mask)) {
313 fsnotify_destroy_event(group, kevent);
314 } else {
315 if (ret <= 0) {
316 FANOTIFY_PE(kevent)->response = FAN_DENY;
317 wake_up(&group->fanotify_data.access_waitq);
318 } else {
319 spin_lock(&group->notification_lock);
320 list_add_tail(&kevent->list,
321 &group->fanotify_data.access_list);
322 spin_unlock(&group->notification_lock);
323 }
324 }
325 if (ret < 0)
326 break;
327 buf += ret;
328 count -= ret;
329 }
330 remove_wait_queue(&group->notification_waitq, &wait);
331
332 if (start != buf && ret != -EFAULT)
333 ret = buf - start;
334 return ret;
335}
336
337static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
338{
339 struct fanotify_response response = { .fd = -1, .response = -1 };
340 struct fsnotify_group *group;
341 int ret;
342
343 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
344 return -EINVAL;
345
346 group = file->private_data;
347
348 if (count > sizeof(response))
349 count = sizeof(response);
350
351 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
352
353 if (copy_from_user(&response, buf, count))
354 return -EFAULT;
355
356 ret = process_access_response(group, &response);
357 if (ret < 0)
358 count = ret;
359
360 return count;
361}
362
363static int fanotify_release(struct inode *ignored, struct file *file)
364{
365 struct fsnotify_group *group = file->private_data;
366 struct fanotify_perm_event_info *event, *next;
367 struct fsnotify_event *fsn_event;
368
369 /*
370 * Stop new events from arriving in the notification queue. since
371 * userspace cannot use fanotify fd anymore, no event can enter or
372 * leave access_list by now either.
373 */
374 fsnotify_group_stop_queueing(group);
375
376 /*
377 * Process all permission events on access_list and notification queue
378 * and simulate reply from userspace.
379 */
380 spin_lock(&group->notification_lock);
381 list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
382 fae.fse.list) {
383 pr_debug("%s: found group=%p event=%p\n", __func__, group,
384 event);
385
386 list_del_init(&event->fae.fse.list);
387 event->response = FAN_ALLOW;
388 }
389
390 /*
391 * Destroy all non-permission events. For permission events just
392 * dequeue them and set the response. They will be freed once the
393 * response is consumed and fanotify_get_response() returns.
394 */
395 while (!fsnotify_notify_queue_is_empty(group)) {
396 fsn_event = fsnotify_remove_first_event(group);
397 if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
398 spin_unlock(&group->notification_lock);
399 fsnotify_destroy_event(group, fsn_event);
400 spin_lock(&group->notification_lock);
401 } else {
402 FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
403 }
404 }
405 spin_unlock(&group->notification_lock);
406
407 /* Response for all permission events it set, wakeup waiters */
408 wake_up(&group->fanotify_data.access_waitq);
409
410 /* matches the fanotify_init->fsnotify_alloc_group */
411 fsnotify_destroy_group(group);
412
413 return 0;
414}
415
416static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
417{
418 struct fsnotify_group *group;
419 struct fsnotify_event *fsn_event;
420 void __user *p;
421 int ret = -ENOTTY;
422 size_t send_len = 0;
423
424 group = file->private_data;
425
426 p = (void __user *) arg;
427
428 switch (cmd) {
429 case FIONREAD:
430 spin_lock(&group->notification_lock);
431 list_for_each_entry(fsn_event, &group->notification_list, list)
432 send_len += FAN_EVENT_METADATA_LEN;
433 spin_unlock(&group->notification_lock);
434 ret = put_user(send_len, (int __user *) p);
435 break;
436 }
437
438 return ret;
439}
440
441static const struct file_operations fanotify_fops = {
442 .show_fdinfo = fanotify_show_fdinfo,
443 .poll = fanotify_poll,
444 .read = fanotify_read,
445 .write = fanotify_write,
446 .fasync = NULL,
447 .release = fanotify_release,
448 .unlocked_ioctl = fanotify_ioctl,
449 .compat_ioctl = fanotify_ioctl,
450 .llseek = noop_llseek,
451};
452
453static int fanotify_find_path(int dfd, const char __user *filename,
454 struct path *path, unsigned int flags)
455{
456 int ret;
457
458 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
459 dfd, filename, flags);
460
461 if (filename == NULL) {
462 struct fd f = fdget(dfd);
463
464 ret = -EBADF;
465 if (!f.file)
466 goto out;
467
468 ret = -ENOTDIR;
469 if ((flags & FAN_MARK_ONLYDIR) &&
470 !(S_ISDIR(file_inode(f.file)->i_mode))) {
471 fdput(f);
472 goto out;
473 }
474
475 *path = f.file->f_path;
476 path_get(path);
477 fdput(f);
478 } else {
479 unsigned int lookup_flags = 0;
480
481 if (!(flags & FAN_MARK_DONT_FOLLOW))
482 lookup_flags |= LOOKUP_FOLLOW;
483 if (flags & FAN_MARK_ONLYDIR)
484 lookup_flags |= LOOKUP_DIRECTORY;
485
486 ret = user_path_at(dfd, filename, lookup_flags, path);
487 if (ret)
488 goto out;
489 }
490
491 /* you can only watch an inode if you have read permissions on it */
492 ret = inode_permission(path->dentry->d_inode, MAY_READ);
493 if (ret)
494 path_put(path);
495out:
496 return ret;
497}
498
499static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
500 __u32 mask,
501 unsigned int flags,
502 int *destroy)
503{
504 __u32 oldmask = 0;
505
506 spin_lock(&fsn_mark->lock);
507 if (!(flags & FAN_MARK_IGNORED_MASK)) {
508 __u32 tmask = fsn_mark->mask & ~mask;
509
510 if (flags & FAN_MARK_ONDIR)
511 tmask &= ~FAN_ONDIR;
512
513 oldmask = fsn_mark->mask;
514 fsn_mark->mask = tmask;
515 } else {
516 __u32 tmask = fsn_mark->ignored_mask & ~mask;
517 if (flags & FAN_MARK_ONDIR)
518 tmask &= ~FAN_ONDIR;
519 fsn_mark->ignored_mask = tmask;
520 }
521 *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
522 spin_unlock(&fsn_mark->lock);
523
524 return mask & oldmask;
525}
526
527static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
528 struct vfsmount *mnt, __u32 mask,
529 unsigned int flags)
530{
531 struct fsnotify_mark *fsn_mark = NULL;
532 __u32 removed;
533 int destroy_mark;
534
535 mutex_lock(&group->mark_mutex);
536 fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
537 group);
538 if (!fsn_mark) {
539 mutex_unlock(&group->mark_mutex);
540 return -ENOENT;
541 }
542
543 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
544 &destroy_mark);
545 if (removed & real_mount(mnt)->mnt_fsnotify_mask)
546 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
547 if (destroy_mark)
548 fsnotify_detach_mark(fsn_mark);
549 mutex_unlock(&group->mark_mutex);
550 if (destroy_mark)
551 fsnotify_free_mark(fsn_mark);
552
553 fsnotify_put_mark(fsn_mark);
554 return 0;
555}
556
557static int fanotify_remove_inode_mark(struct fsnotify_group *group,
558 struct inode *inode, __u32 mask,
559 unsigned int flags)
560{
561 struct fsnotify_mark *fsn_mark = NULL;
562 __u32 removed;
563 int destroy_mark;
564
565 mutex_lock(&group->mark_mutex);
566 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
567 if (!fsn_mark) {
568 mutex_unlock(&group->mark_mutex);
569 return -ENOENT;
570 }
571
572 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
573 &destroy_mark);
574 if (removed & inode->i_fsnotify_mask)
575 fsnotify_recalc_mask(inode->i_fsnotify_marks);
576 if (destroy_mark)
577 fsnotify_detach_mark(fsn_mark);
578 mutex_unlock(&group->mark_mutex);
579 if (destroy_mark)
580 fsnotify_free_mark(fsn_mark);
581
582 /* matches the fsnotify_find_mark() */
583 fsnotify_put_mark(fsn_mark);
584
585 return 0;
586}
587
588static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
589 __u32 mask,
590 unsigned int flags)
591{
592 __u32 oldmask = -1;
593
594 spin_lock(&fsn_mark->lock);
595 if (!(flags & FAN_MARK_IGNORED_MASK)) {
596 __u32 tmask = fsn_mark->mask | mask;
597
598 if (flags & FAN_MARK_ONDIR)
599 tmask |= FAN_ONDIR;
600
601 oldmask = fsn_mark->mask;
602 fsn_mark->mask = tmask;
603 } else {
604 __u32 tmask = fsn_mark->ignored_mask | mask;
605 if (flags & FAN_MARK_ONDIR)
606 tmask |= FAN_ONDIR;
607
608 fsn_mark->ignored_mask = tmask;
609 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
610 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
611 }
612 spin_unlock(&fsn_mark->lock);
613
614 return mask & ~oldmask;
615}
616
617static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
618 struct inode *inode,
619 struct vfsmount *mnt)
620{
621 struct fsnotify_mark *mark;
622 int ret;
623
624 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
625 return ERR_PTR(-ENOSPC);
626
627 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
628 if (!mark)
629 return ERR_PTR(-ENOMEM);
630
631 fsnotify_init_mark(mark, group);
632 ret = fsnotify_add_mark_locked(mark, inode, mnt, 0);
633 if (ret) {
634 fsnotify_put_mark(mark);
635 return ERR_PTR(ret);
636 }
637
638 return mark;
639}
640
641
642static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
643 struct vfsmount *mnt, __u32 mask,
644 unsigned int flags)
645{
646 struct fsnotify_mark *fsn_mark;
647 __u32 added;
648
649 mutex_lock(&group->mark_mutex);
650 fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
651 group);
652 if (!fsn_mark) {
653 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
654 if (IS_ERR(fsn_mark)) {
655 mutex_unlock(&group->mark_mutex);
656 return PTR_ERR(fsn_mark);
657 }
658 }
659 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
660 if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
661 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
662 mutex_unlock(&group->mark_mutex);
663
664 fsnotify_put_mark(fsn_mark);
665 return 0;
666}
667
668static int fanotify_add_inode_mark(struct fsnotify_group *group,
669 struct inode *inode, __u32 mask,
670 unsigned int flags)
671{
672 struct fsnotify_mark *fsn_mark;
673 __u32 added;
674
675 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
676
677 /*
678 * If some other task has this inode open for write we should not add
679 * an ignored mark, unless that ignored mark is supposed to survive
680 * modification changes anyway.
681 */
682 if ((flags & FAN_MARK_IGNORED_MASK) &&
683 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
684 (atomic_read(&inode->i_writecount) > 0))
685 return 0;
686
687 mutex_lock(&group->mark_mutex);
688 fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
689 if (!fsn_mark) {
690 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
691 if (IS_ERR(fsn_mark)) {
692 mutex_unlock(&group->mark_mutex);
693 return PTR_ERR(fsn_mark);
694 }
695 }
696 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
697 if (added & ~inode->i_fsnotify_mask)
698 fsnotify_recalc_mask(inode->i_fsnotify_marks);
699 mutex_unlock(&group->mark_mutex);
700
701 fsnotify_put_mark(fsn_mark);
702 return 0;
703}
704
705/* fanotify syscalls */
706SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
707{
708 struct fsnotify_group *group;
709 int f_flags, fd;
710 struct user_struct *user;
711 struct fanotify_event_info *oevent;
712
713 pr_debug("%s: flags=%d event_f_flags=%d\n",
714 __func__, flags, event_f_flags);
715
716 if (!capable(CAP_SYS_ADMIN))
717 return -EPERM;
718
719#ifdef CONFIG_AUDITSYSCALL
720 if (flags & ~(FAN_ALL_INIT_FLAGS | FAN_ENABLE_AUDIT))
721#else
722 if (flags & ~FAN_ALL_INIT_FLAGS)
723#endif
724 return -EINVAL;
725
726 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
727 return -EINVAL;
728
729 switch (event_f_flags & O_ACCMODE) {
730 case O_RDONLY:
731 case O_RDWR:
732 case O_WRONLY:
733 break;
734 default:
735 return -EINVAL;
736 }
737
738 user = get_current_user();
739 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
740 free_uid(user);
741 return -EMFILE;
742 }
743
744 f_flags = O_RDWR | FMODE_NONOTIFY;
745 if (flags & FAN_CLOEXEC)
746 f_flags |= O_CLOEXEC;
747 if (flags & FAN_NONBLOCK)
748 f_flags |= O_NONBLOCK;
749
750 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
751 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
752 if (IS_ERR(group)) {
753 free_uid(user);
754 return PTR_ERR(group);
755 }
756
757 group->fanotify_data.user = user;
758 atomic_inc(&user->fanotify_listeners);
759
760 oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL);
761 if (unlikely(!oevent)) {
762 fd = -ENOMEM;
763 goto out_destroy_group;
764 }
765 group->overflow_event = &oevent->fse;
766
767 if (force_o_largefile())
768 event_f_flags |= O_LARGEFILE;
769 group->fanotify_data.f_flags = event_f_flags;
770 init_waitqueue_head(&group->fanotify_data.access_waitq);
771 INIT_LIST_HEAD(&group->fanotify_data.access_list);
772 switch (flags & FAN_ALL_CLASS_BITS) {
773 case FAN_CLASS_NOTIF:
774 group->priority = FS_PRIO_0;
775 break;
776 case FAN_CLASS_CONTENT:
777 group->priority = FS_PRIO_1;
778 break;
779 case FAN_CLASS_PRE_CONTENT:
780 group->priority = FS_PRIO_2;
781 break;
782 default:
783 fd = -EINVAL;
784 goto out_destroy_group;
785 }
786
787 if (flags & FAN_UNLIMITED_QUEUE) {
788 fd = -EPERM;
789 if (!capable(CAP_SYS_ADMIN))
790 goto out_destroy_group;
791 group->max_events = UINT_MAX;
792 } else {
793 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
794 }
795
796 if (flags & FAN_UNLIMITED_MARKS) {
797 fd = -EPERM;
798 if (!capable(CAP_SYS_ADMIN))
799 goto out_destroy_group;
800 group->fanotify_data.max_marks = UINT_MAX;
801 } else {
802 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
803 }
804
805 if (flags & FAN_ENABLE_AUDIT) {
806 fd = -EPERM;
807 if (!capable(CAP_AUDIT_WRITE))
808 goto out_destroy_group;
809 group->fanotify_data.audit = true;
810 }
811
812 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
813 if (fd < 0)
814 goto out_destroy_group;
815
816 return fd;
817
818out_destroy_group:
819 fsnotify_destroy_group(group);
820 return fd;
821}
822
823static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
824 int dfd, const char __user *pathname)
825{
826 struct inode *inode = NULL;
827 struct vfsmount *mnt = NULL;
828 struct fsnotify_group *group;
829 struct fd f;
830 struct path path;
831 u32 valid_mask = FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD;
832 int ret;
833
834 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
835 __func__, fanotify_fd, flags, dfd, pathname, mask);
836
837 /* we only use the lower 32 bits as of right now. */
838 if (mask & ((__u64)0xffffffff << 32))
839 return -EINVAL;
840
841 if (flags & ~FAN_ALL_MARK_FLAGS)
842 return -EINVAL;
843 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
844 case FAN_MARK_ADD: /* fallthrough */
845 case FAN_MARK_REMOVE:
846 if (!mask)
847 return -EINVAL;
848 break;
849 case FAN_MARK_FLUSH:
850 if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
851 return -EINVAL;
852 break;
853 default:
854 return -EINVAL;
855 }
856
857 if (mask & FAN_ONDIR) {
858 flags |= FAN_MARK_ONDIR;
859 mask &= ~FAN_ONDIR;
860 }
861
862 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
863 valid_mask |= FAN_ALL_PERM_EVENTS;
864
865 if (mask & ~valid_mask)
866 return -EINVAL;
867
868 f = fdget(fanotify_fd);
869 if (unlikely(!f.file))
870 return -EBADF;
871
872 /* verify that this is indeed an fanotify instance */
873 ret = -EINVAL;
874 if (unlikely(f.file->f_op != &fanotify_fops))
875 goto fput_and_out;
876 group = f.file->private_data;
877
878 /*
879 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
880 * allowed to set permissions events.
881 */
882 ret = -EINVAL;
883 if (mask & FAN_ALL_PERM_EVENTS &&
884 group->priority == FS_PRIO_0)
885 goto fput_and_out;
886
887 if (flags & FAN_MARK_FLUSH) {
888 ret = 0;
889 if (flags & FAN_MARK_MOUNT)
890 fsnotify_clear_vfsmount_marks_by_group(group);
891 else
892 fsnotify_clear_inode_marks_by_group(group);
893 goto fput_and_out;
894 }
895
896 ret = fanotify_find_path(dfd, pathname, &path, flags);
897 if (ret)
898 goto fput_and_out;
899
900 /* inode held in place by reference to path; group by fget on fd */
901 if (!(flags & FAN_MARK_MOUNT))
902 inode = path.dentry->d_inode;
903 else
904 mnt = path.mnt;
905
906 /* create/update an inode mark */
907 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
908 case FAN_MARK_ADD:
909 if (flags & FAN_MARK_MOUNT)
910 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
911 else
912 ret = fanotify_add_inode_mark(group, inode, mask, flags);
913 break;
914 case FAN_MARK_REMOVE:
915 if (flags & FAN_MARK_MOUNT)
916 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
917 else
918 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
919 break;
920 default:
921 ret = -EINVAL;
922 }
923
924 path_put(&path);
925fput_and_out:
926 fdput(f);
927 return ret;
928}
929
930SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
931 __u64, mask, int, dfd,
932 const char __user *, pathname)
933{
934 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname);
935}
936
937#ifdef CONFIG_COMPAT
938COMPAT_SYSCALL_DEFINE6(fanotify_mark,
939 int, fanotify_fd, unsigned int, flags,
940 __u32, mask0, __u32, mask1, int, dfd,
941 const char __user *, pathname)
942{
943 return do_fanotify_mark(fanotify_fd, flags,
944#ifdef __BIG_ENDIAN
945 ((__u64)mask0 << 32) | mask1,
946#else
947 ((__u64)mask1 << 32) | mask0,
948#endif
949 dfd, pathname);
950}
951#endif
952
953/*
954 * fanotify_user_setup - Our initialization function. Note that we cannot return
955 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
956 * must result in panic().
957 */
958static int __init fanotify_user_setup(void)
959{
960 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
961 fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
962 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
963 fanotify_perm_event_cachep =
964 KMEM_CACHE(fanotify_perm_event_info, SLAB_PANIC);
965 }
966
967 return 0;
968}
969device_initcall(fanotify_user_setup);
1#include <linux/fanotify.h>
2#include <linux/fcntl.h>
3#include <linux/file.h>
4#include <linux/fs.h>
5#include <linux/anon_inodes.h>
6#include <linux/fsnotify_backend.h>
7#include <linux/init.h>
8#include <linux/mount.h>
9#include <linux/namei.h>
10#include <linux/poll.h>
11#include <linux/security.h>
12#include <linux/syscalls.h>
13#include <linux/slab.h>
14#include <linux/types.h>
15#include <linux/uaccess.h>
16#include <linux/compat.h>
17
18#include <asm/ioctls.h>
19
20#include "../../mount.h"
21#include "../fdinfo.h"
22#include "fanotify.h"
23
24#define FANOTIFY_DEFAULT_MAX_EVENTS 16384
25#define FANOTIFY_DEFAULT_MAX_MARKS 8192
26#define FANOTIFY_DEFAULT_MAX_LISTENERS 128
27
28/*
29 * All flags that may be specified in parameter event_f_flags of fanotify_init.
30 *
31 * Internal and external open flags are stored together in field f_flags of
32 * struct file. Only external open flags shall be allowed in event_f_flags.
33 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
34 * excluded.
35 */
36#define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \
37 O_ACCMODE | O_APPEND | O_NONBLOCK | \
38 __O_SYNC | O_DSYNC | O_CLOEXEC | \
39 O_LARGEFILE | O_NOATIME )
40
41extern const struct fsnotify_ops fanotify_fsnotify_ops;
42
43static struct kmem_cache *fanotify_mark_cache __read_mostly;
44struct kmem_cache *fanotify_event_cachep __read_mostly;
45struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
46
47/*
48 * Get an fsnotify notification event if one exists and is small
49 * enough to fit in "count". Return an error pointer if the count
50 * is not large enough.
51 *
52 * Called with the group->notification_lock held.
53 */
54static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
55 size_t count)
56{
57 assert_spin_locked(&group->notification_lock);
58
59 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
60
61 if (fsnotify_notify_queue_is_empty(group))
62 return NULL;
63
64 if (FAN_EVENT_METADATA_LEN > count)
65 return ERR_PTR(-EINVAL);
66
67 /* held the notification_lock the whole time, so this is the
68 * same event we peeked above */
69 return fsnotify_remove_first_event(group);
70}
71
72static int create_fd(struct fsnotify_group *group,
73 struct fanotify_event_info *event,
74 struct file **file)
75{
76 int client_fd;
77 struct file *new_file;
78
79 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
80
81 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
82 if (client_fd < 0)
83 return client_fd;
84
85 /*
86 * we need a new file handle for the userspace program so it can read even if it was
87 * originally opened O_WRONLY.
88 */
89 /* it's possible this event was an overflow event. in that case dentry and mnt
90 * are NULL; That's fine, just don't call dentry open */
91 if (event->path.dentry && event->path.mnt)
92 new_file = dentry_open(&event->path,
93 group->fanotify_data.f_flags | FMODE_NONOTIFY,
94 current_cred());
95 else
96 new_file = ERR_PTR(-EOVERFLOW);
97 if (IS_ERR(new_file)) {
98 /*
99 * we still send an event even if we can't open the file. this
100 * can happen when say tasks are gone and we try to open their
101 * /proc files or we try to open a WRONLY file like in sysfs
102 * we just send the errno to userspace since there isn't much
103 * else we can do.
104 */
105 put_unused_fd(client_fd);
106 client_fd = PTR_ERR(new_file);
107 } else {
108 *file = new_file;
109 }
110
111 return client_fd;
112}
113
114static int fill_event_metadata(struct fsnotify_group *group,
115 struct fanotify_event_metadata *metadata,
116 struct fsnotify_event *fsn_event,
117 struct file **file)
118{
119 int ret = 0;
120 struct fanotify_event_info *event;
121
122 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
123 group, metadata, fsn_event);
124
125 *file = NULL;
126 event = container_of(fsn_event, struct fanotify_event_info, fse);
127 metadata->event_len = FAN_EVENT_METADATA_LEN;
128 metadata->metadata_len = FAN_EVENT_METADATA_LEN;
129 metadata->vers = FANOTIFY_METADATA_VERSION;
130 metadata->reserved = 0;
131 metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
132 metadata->pid = pid_vnr(event->tgid);
133 if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
134 metadata->fd = FAN_NOFD;
135 else {
136 metadata->fd = create_fd(group, event, file);
137 if (metadata->fd < 0)
138 ret = metadata->fd;
139 }
140
141 return ret;
142}
143
144#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
145static struct fanotify_perm_event_info *dequeue_event(
146 struct fsnotify_group *group, int fd)
147{
148 struct fanotify_perm_event_info *event, *return_e = NULL;
149
150 spin_lock(&group->notification_lock);
151 list_for_each_entry(event, &group->fanotify_data.access_list,
152 fae.fse.list) {
153 if (event->fd != fd)
154 continue;
155
156 list_del_init(&event->fae.fse.list);
157 return_e = event;
158 break;
159 }
160 spin_unlock(&group->notification_lock);
161
162 pr_debug("%s: found return_re=%p\n", __func__, return_e);
163
164 return return_e;
165}
166
167static int process_access_response(struct fsnotify_group *group,
168 struct fanotify_response *response_struct)
169{
170 struct fanotify_perm_event_info *event;
171 int fd = response_struct->fd;
172 int response = response_struct->response;
173
174 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
175 fd, response);
176 /*
177 * make sure the response is valid, if invalid we do nothing and either
178 * userspace can send a valid response or we will clean it up after the
179 * timeout
180 */
181 switch (response) {
182 case FAN_ALLOW:
183 case FAN_DENY:
184 break;
185 default:
186 return -EINVAL;
187 }
188
189 if (fd < 0)
190 return -EINVAL;
191
192 event = dequeue_event(group, fd);
193 if (!event)
194 return -ENOENT;
195
196 event->response = response;
197 wake_up(&group->fanotify_data.access_waitq);
198
199 return 0;
200}
201#endif
202
203static ssize_t copy_event_to_user(struct fsnotify_group *group,
204 struct fsnotify_event *event,
205 char __user *buf)
206{
207 struct fanotify_event_metadata fanotify_event_metadata;
208 struct file *f;
209 int fd, ret;
210
211 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
212
213 ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
214 if (ret < 0)
215 return ret;
216
217 fd = fanotify_event_metadata.fd;
218 ret = -EFAULT;
219 if (copy_to_user(buf, &fanotify_event_metadata,
220 fanotify_event_metadata.event_len))
221 goto out_close_fd;
222
223#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
224 if (event->mask & FAN_ALL_PERM_EVENTS)
225 FANOTIFY_PE(event)->fd = fd;
226#endif
227
228 if (fd != FAN_NOFD)
229 fd_install(fd, f);
230 return fanotify_event_metadata.event_len;
231
232out_close_fd:
233 if (fd != FAN_NOFD) {
234 put_unused_fd(fd);
235 fput(f);
236 }
237 return ret;
238}
239
240/* intofiy userspace file descriptor functions */
241static unsigned int fanotify_poll(struct file *file, poll_table *wait)
242{
243 struct fsnotify_group *group = file->private_data;
244 int ret = 0;
245
246 poll_wait(file, &group->notification_waitq, wait);
247 spin_lock(&group->notification_lock);
248 if (!fsnotify_notify_queue_is_empty(group))
249 ret = POLLIN | POLLRDNORM;
250 spin_unlock(&group->notification_lock);
251
252 return ret;
253}
254
255static ssize_t fanotify_read(struct file *file, char __user *buf,
256 size_t count, loff_t *pos)
257{
258 struct fsnotify_group *group;
259 struct fsnotify_event *kevent;
260 char __user *start;
261 int ret;
262 DEFINE_WAIT_FUNC(wait, woken_wake_function);
263
264 start = buf;
265 group = file->private_data;
266
267 pr_debug("%s: group=%p\n", __func__, group);
268
269 add_wait_queue(&group->notification_waitq, &wait);
270 while (1) {
271 spin_lock(&group->notification_lock);
272 kevent = get_one_event(group, count);
273 spin_unlock(&group->notification_lock);
274
275 if (IS_ERR(kevent)) {
276 ret = PTR_ERR(kevent);
277 break;
278 }
279
280 if (!kevent) {
281 ret = -EAGAIN;
282 if (file->f_flags & O_NONBLOCK)
283 break;
284
285 ret = -ERESTARTSYS;
286 if (signal_pending(current))
287 break;
288
289 if (start != buf)
290 break;
291
292 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
293 continue;
294 }
295
296 ret = copy_event_to_user(group, kevent, buf);
297 /*
298 * Permission events get queued to wait for response. Other
299 * events can be destroyed now.
300 */
301 if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
302 fsnotify_destroy_event(group, kevent);
303 if (ret < 0)
304 break;
305 } else {
306#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
307 if (ret < 0) {
308 FANOTIFY_PE(kevent)->response = FAN_DENY;
309 wake_up(&group->fanotify_data.access_waitq);
310 break;
311 }
312 spin_lock(&group->notification_lock);
313 list_add_tail(&kevent->list,
314 &group->fanotify_data.access_list);
315 spin_unlock(&group->notification_lock);
316#endif
317 }
318 buf += ret;
319 count -= ret;
320 }
321 remove_wait_queue(&group->notification_waitq, &wait);
322
323 if (start != buf && ret != -EFAULT)
324 ret = buf - start;
325 return ret;
326}
327
328static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
329{
330#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
331 struct fanotify_response response = { .fd = -1, .response = -1 };
332 struct fsnotify_group *group;
333 int ret;
334
335 group = file->private_data;
336
337 if (count > sizeof(response))
338 count = sizeof(response);
339
340 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
341
342 if (copy_from_user(&response, buf, count))
343 return -EFAULT;
344
345 ret = process_access_response(group, &response);
346 if (ret < 0)
347 count = ret;
348
349 return count;
350#else
351 return -EINVAL;
352#endif
353}
354
355static int fanotify_release(struct inode *ignored, struct file *file)
356{
357 struct fsnotify_group *group = file->private_data;
358
359#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
360 struct fanotify_perm_event_info *event, *next;
361 struct fsnotify_event *fsn_event;
362
363 /*
364 * Stop new events from arriving in the notification queue. since
365 * userspace cannot use fanotify fd anymore, no event can enter or
366 * leave access_list by now either.
367 */
368 fsnotify_group_stop_queueing(group);
369
370 /*
371 * Process all permission events on access_list and notification queue
372 * and simulate reply from userspace.
373 */
374 spin_lock(&group->notification_lock);
375 list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
376 fae.fse.list) {
377 pr_debug("%s: found group=%p event=%p\n", __func__, group,
378 event);
379
380 list_del_init(&event->fae.fse.list);
381 event->response = FAN_ALLOW;
382 }
383
384 /*
385 * Destroy all non-permission events. For permission events just
386 * dequeue them and set the response. They will be freed once the
387 * response is consumed and fanotify_get_response() returns.
388 */
389 while (!fsnotify_notify_queue_is_empty(group)) {
390 fsn_event = fsnotify_remove_first_event(group);
391 if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
392 spin_unlock(&group->notification_lock);
393 fsnotify_destroy_event(group, fsn_event);
394 spin_lock(&group->notification_lock);
395 } else
396 FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
397 }
398 spin_unlock(&group->notification_lock);
399
400 /* Response for all permission events it set, wakeup waiters */
401 wake_up(&group->fanotify_data.access_waitq);
402#endif
403
404 /* matches the fanotify_init->fsnotify_alloc_group */
405 fsnotify_destroy_group(group);
406
407 return 0;
408}
409
410static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
411{
412 struct fsnotify_group *group;
413 struct fsnotify_event *fsn_event;
414 void __user *p;
415 int ret = -ENOTTY;
416 size_t send_len = 0;
417
418 group = file->private_data;
419
420 p = (void __user *) arg;
421
422 switch (cmd) {
423 case FIONREAD:
424 spin_lock(&group->notification_lock);
425 list_for_each_entry(fsn_event, &group->notification_list, list)
426 send_len += FAN_EVENT_METADATA_LEN;
427 spin_unlock(&group->notification_lock);
428 ret = put_user(send_len, (int __user *) p);
429 break;
430 }
431
432 return ret;
433}
434
435static const struct file_operations fanotify_fops = {
436 .show_fdinfo = fanotify_show_fdinfo,
437 .poll = fanotify_poll,
438 .read = fanotify_read,
439 .write = fanotify_write,
440 .fasync = NULL,
441 .release = fanotify_release,
442 .unlocked_ioctl = fanotify_ioctl,
443 .compat_ioctl = fanotify_ioctl,
444 .llseek = noop_llseek,
445};
446
447static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
448{
449 kmem_cache_free(fanotify_mark_cache, fsn_mark);
450}
451
452static int fanotify_find_path(int dfd, const char __user *filename,
453 struct path *path, unsigned int flags)
454{
455 int ret;
456
457 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
458 dfd, filename, flags);
459
460 if (filename == NULL) {
461 struct fd f = fdget(dfd);
462
463 ret = -EBADF;
464 if (!f.file)
465 goto out;
466
467 ret = -ENOTDIR;
468 if ((flags & FAN_MARK_ONLYDIR) &&
469 !(S_ISDIR(file_inode(f.file)->i_mode))) {
470 fdput(f);
471 goto out;
472 }
473
474 *path = f.file->f_path;
475 path_get(path);
476 fdput(f);
477 } else {
478 unsigned int lookup_flags = 0;
479
480 if (!(flags & FAN_MARK_DONT_FOLLOW))
481 lookup_flags |= LOOKUP_FOLLOW;
482 if (flags & FAN_MARK_ONLYDIR)
483 lookup_flags |= LOOKUP_DIRECTORY;
484
485 ret = user_path_at(dfd, filename, lookup_flags, path);
486 if (ret)
487 goto out;
488 }
489
490 /* you can only watch an inode if you have read permissions on it */
491 ret = inode_permission(path->dentry->d_inode, MAY_READ);
492 if (ret)
493 path_put(path);
494out:
495 return ret;
496}
497
498static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
499 __u32 mask,
500 unsigned int flags,
501 int *destroy)
502{
503 __u32 oldmask = 0;
504
505 spin_lock(&fsn_mark->lock);
506 if (!(flags & FAN_MARK_IGNORED_MASK)) {
507 __u32 tmask = fsn_mark->mask & ~mask;
508
509 if (flags & FAN_MARK_ONDIR)
510 tmask &= ~FAN_ONDIR;
511
512 oldmask = fsn_mark->mask;
513 fsnotify_set_mark_mask_locked(fsn_mark, tmask);
514 } else {
515 __u32 tmask = fsn_mark->ignored_mask & ~mask;
516 if (flags & FAN_MARK_ONDIR)
517 tmask &= ~FAN_ONDIR;
518
519 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
520 }
521 *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
522 spin_unlock(&fsn_mark->lock);
523
524 return mask & oldmask;
525}
526
527static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
528 struct vfsmount *mnt, __u32 mask,
529 unsigned int flags)
530{
531 struct fsnotify_mark *fsn_mark = NULL;
532 __u32 removed;
533 int destroy_mark;
534
535 mutex_lock(&group->mark_mutex);
536 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
537 if (!fsn_mark) {
538 mutex_unlock(&group->mark_mutex);
539 return -ENOENT;
540 }
541
542 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
543 &destroy_mark);
544 if (destroy_mark)
545 fsnotify_detach_mark(fsn_mark);
546 mutex_unlock(&group->mark_mutex);
547 if (destroy_mark)
548 fsnotify_free_mark(fsn_mark);
549
550 fsnotify_put_mark(fsn_mark);
551 if (removed & real_mount(mnt)->mnt_fsnotify_mask)
552 fsnotify_recalc_vfsmount_mask(mnt);
553
554 return 0;
555}
556
557static int fanotify_remove_inode_mark(struct fsnotify_group *group,
558 struct inode *inode, __u32 mask,
559 unsigned int flags)
560{
561 struct fsnotify_mark *fsn_mark = NULL;
562 __u32 removed;
563 int destroy_mark;
564
565 mutex_lock(&group->mark_mutex);
566 fsn_mark = fsnotify_find_inode_mark(group, inode);
567 if (!fsn_mark) {
568 mutex_unlock(&group->mark_mutex);
569 return -ENOENT;
570 }
571
572 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
573 &destroy_mark);
574 if (destroy_mark)
575 fsnotify_detach_mark(fsn_mark);
576 mutex_unlock(&group->mark_mutex);
577 if (destroy_mark)
578 fsnotify_free_mark(fsn_mark);
579
580 /* matches the fsnotify_find_inode_mark() */
581 fsnotify_put_mark(fsn_mark);
582 if (removed & inode->i_fsnotify_mask)
583 fsnotify_recalc_inode_mask(inode);
584
585 return 0;
586}
587
588static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
589 __u32 mask,
590 unsigned int flags)
591{
592 __u32 oldmask = -1;
593
594 spin_lock(&fsn_mark->lock);
595 if (!(flags & FAN_MARK_IGNORED_MASK)) {
596 __u32 tmask = fsn_mark->mask | mask;
597
598 if (flags & FAN_MARK_ONDIR)
599 tmask |= FAN_ONDIR;
600
601 oldmask = fsn_mark->mask;
602 fsnotify_set_mark_mask_locked(fsn_mark, tmask);
603 } else {
604 __u32 tmask = fsn_mark->ignored_mask | mask;
605 if (flags & FAN_MARK_ONDIR)
606 tmask |= FAN_ONDIR;
607
608 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
609 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
610 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
611 }
612 spin_unlock(&fsn_mark->lock);
613
614 return mask & ~oldmask;
615}
616
617static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
618 struct inode *inode,
619 struct vfsmount *mnt)
620{
621 struct fsnotify_mark *mark;
622 int ret;
623
624 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
625 return ERR_PTR(-ENOSPC);
626
627 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
628 if (!mark)
629 return ERR_PTR(-ENOMEM);
630
631 fsnotify_init_mark(mark, fanotify_free_mark);
632 ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
633 if (ret) {
634 fsnotify_put_mark(mark);
635 return ERR_PTR(ret);
636 }
637
638 return mark;
639}
640
641
642static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
643 struct vfsmount *mnt, __u32 mask,
644 unsigned int flags)
645{
646 struct fsnotify_mark *fsn_mark;
647 __u32 added;
648
649 mutex_lock(&group->mark_mutex);
650 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
651 if (!fsn_mark) {
652 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
653 if (IS_ERR(fsn_mark)) {
654 mutex_unlock(&group->mark_mutex);
655 return PTR_ERR(fsn_mark);
656 }
657 }
658 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
659 mutex_unlock(&group->mark_mutex);
660
661 if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
662 fsnotify_recalc_vfsmount_mask(mnt);
663
664 fsnotify_put_mark(fsn_mark);
665 return 0;
666}
667
668static int fanotify_add_inode_mark(struct fsnotify_group *group,
669 struct inode *inode, __u32 mask,
670 unsigned int flags)
671{
672 struct fsnotify_mark *fsn_mark;
673 __u32 added;
674
675 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
676
677 /*
678 * If some other task has this inode open for write we should not add
679 * an ignored mark, unless that ignored mark is supposed to survive
680 * modification changes anyway.
681 */
682 if ((flags & FAN_MARK_IGNORED_MASK) &&
683 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
684 (atomic_read(&inode->i_writecount) > 0))
685 return 0;
686
687 mutex_lock(&group->mark_mutex);
688 fsn_mark = fsnotify_find_inode_mark(group, inode);
689 if (!fsn_mark) {
690 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
691 if (IS_ERR(fsn_mark)) {
692 mutex_unlock(&group->mark_mutex);
693 return PTR_ERR(fsn_mark);
694 }
695 }
696 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
697 mutex_unlock(&group->mark_mutex);
698
699 if (added & ~inode->i_fsnotify_mask)
700 fsnotify_recalc_inode_mask(inode);
701
702 fsnotify_put_mark(fsn_mark);
703 return 0;
704}
705
706/* fanotify syscalls */
707SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
708{
709 struct fsnotify_group *group;
710 int f_flags, fd;
711 struct user_struct *user;
712 struct fanotify_event_info *oevent;
713
714 pr_debug("%s: flags=%d event_f_flags=%d\n",
715 __func__, flags, event_f_flags);
716
717 if (!capable(CAP_SYS_ADMIN))
718 return -EPERM;
719
720 if (flags & ~FAN_ALL_INIT_FLAGS)
721 return -EINVAL;
722
723 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
724 return -EINVAL;
725
726 switch (event_f_flags & O_ACCMODE) {
727 case O_RDONLY:
728 case O_RDWR:
729 case O_WRONLY:
730 break;
731 default:
732 return -EINVAL;
733 }
734
735 user = get_current_user();
736 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
737 free_uid(user);
738 return -EMFILE;
739 }
740
741 f_flags = O_RDWR | FMODE_NONOTIFY;
742 if (flags & FAN_CLOEXEC)
743 f_flags |= O_CLOEXEC;
744 if (flags & FAN_NONBLOCK)
745 f_flags |= O_NONBLOCK;
746
747 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
748 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
749 if (IS_ERR(group)) {
750 free_uid(user);
751 return PTR_ERR(group);
752 }
753
754 group->fanotify_data.user = user;
755 atomic_inc(&user->fanotify_listeners);
756
757 oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
758 if (unlikely(!oevent)) {
759 fd = -ENOMEM;
760 goto out_destroy_group;
761 }
762 group->overflow_event = &oevent->fse;
763
764 if (force_o_largefile())
765 event_f_flags |= O_LARGEFILE;
766 group->fanotify_data.f_flags = event_f_flags;
767#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
768 init_waitqueue_head(&group->fanotify_data.access_waitq);
769 INIT_LIST_HEAD(&group->fanotify_data.access_list);
770#endif
771 switch (flags & FAN_ALL_CLASS_BITS) {
772 case FAN_CLASS_NOTIF:
773 group->priority = FS_PRIO_0;
774 break;
775 case FAN_CLASS_CONTENT:
776 group->priority = FS_PRIO_1;
777 break;
778 case FAN_CLASS_PRE_CONTENT:
779 group->priority = FS_PRIO_2;
780 break;
781 default:
782 fd = -EINVAL;
783 goto out_destroy_group;
784 }
785
786 if (flags & FAN_UNLIMITED_QUEUE) {
787 fd = -EPERM;
788 if (!capable(CAP_SYS_ADMIN))
789 goto out_destroy_group;
790 group->max_events = UINT_MAX;
791 } else {
792 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
793 }
794
795 if (flags & FAN_UNLIMITED_MARKS) {
796 fd = -EPERM;
797 if (!capable(CAP_SYS_ADMIN))
798 goto out_destroy_group;
799 group->fanotify_data.max_marks = UINT_MAX;
800 } else {
801 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
802 }
803
804 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
805 if (fd < 0)
806 goto out_destroy_group;
807
808 return fd;
809
810out_destroy_group:
811 fsnotify_destroy_group(group);
812 return fd;
813}
814
815SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
816 __u64, mask, int, dfd,
817 const char __user *, pathname)
818{
819 struct inode *inode = NULL;
820 struct vfsmount *mnt = NULL;
821 struct fsnotify_group *group;
822 struct fd f;
823 struct path path;
824 int ret;
825
826 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
827 __func__, fanotify_fd, flags, dfd, pathname, mask);
828
829 /* we only use the lower 32 bits as of right now. */
830 if (mask & ((__u64)0xffffffff << 32))
831 return -EINVAL;
832
833 if (flags & ~FAN_ALL_MARK_FLAGS)
834 return -EINVAL;
835 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
836 case FAN_MARK_ADD: /* fallthrough */
837 case FAN_MARK_REMOVE:
838 if (!mask)
839 return -EINVAL;
840 break;
841 case FAN_MARK_FLUSH:
842 if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
843 return -EINVAL;
844 break;
845 default:
846 return -EINVAL;
847 }
848
849 if (mask & FAN_ONDIR) {
850 flags |= FAN_MARK_ONDIR;
851 mask &= ~FAN_ONDIR;
852 }
853
854#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
855 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
856#else
857 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
858#endif
859 return -EINVAL;
860
861 f = fdget(fanotify_fd);
862 if (unlikely(!f.file))
863 return -EBADF;
864
865 /* verify that this is indeed an fanotify instance */
866 ret = -EINVAL;
867 if (unlikely(f.file->f_op != &fanotify_fops))
868 goto fput_and_out;
869 group = f.file->private_data;
870
871 /*
872 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
873 * allowed to set permissions events.
874 */
875 ret = -EINVAL;
876 if (mask & FAN_ALL_PERM_EVENTS &&
877 group->priority == FS_PRIO_0)
878 goto fput_and_out;
879
880 if (flags & FAN_MARK_FLUSH) {
881 ret = 0;
882 if (flags & FAN_MARK_MOUNT)
883 fsnotify_clear_vfsmount_marks_by_group(group);
884 else
885 fsnotify_clear_inode_marks_by_group(group);
886 goto fput_and_out;
887 }
888
889 ret = fanotify_find_path(dfd, pathname, &path, flags);
890 if (ret)
891 goto fput_and_out;
892
893 /* inode held in place by reference to path; group by fget on fd */
894 if (!(flags & FAN_MARK_MOUNT))
895 inode = path.dentry->d_inode;
896 else
897 mnt = path.mnt;
898
899 /* create/update an inode mark */
900 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
901 case FAN_MARK_ADD:
902 if (flags & FAN_MARK_MOUNT)
903 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
904 else
905 ret = fanotify_add_inode_mark(group, inode, mask, flags);
906 break;
907 case FAN_MARK_REMOVE:
908 if (flags & FAN_MARK_MOUNT)
909 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
910 else
911 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
912 break;
913 default:
914 ret = -EINVAL;
915 }
916
917 path_put(&path);
918fput_and_out:
919 fdput(f);
920 return ret;
921}
922
923#ifdef CONFIG_COMPAT
924COMPAT_SYSCALL_DEFINE6(fanotify_mark,
925 int, fanotify_fd, unsigned int, flags,
926 __u32, mask0, __u32, mask1, int, dfd,
927 const char __user *, pathname)
928{
929 return sys_fanotify_mark(fanotify_fd, flags,
930#ifdef __BIG_ENDIAN
931 ((__u64)mask0 << 32) | mask1,
932#else
933 ((__u64)mask1 << 32) | mask0,
934#endif
935 dfd, pathname);
936}
937#endif
938
939/*
940 * fanotify_user_setup - Our initialization function. Note that we cannot return
941 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
942 * must result in panic().
943 */
944static int __init fanotify_user_setup(void)
945{
946 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
947 fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
948#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
949 fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
950 SLAB_PANIC);
951#endif
952
953 return 0;
954}
955device_initcall(fanotify_user_setup);