]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/notify/fanotify/fanotify_user.c
fanotify: allow userspace to override max queue depth
[net-next-2.6.git] / fs / notify / fanotify / fanotify_user.c
CommitLineData
33d3dfff 1#include <linux/fanotify.h>
11637e4b 2#include <linux/fcntl.h>
2a3edf86 3#include <linux/file.h>
11637e4b 4#include <linux/fs.h>
52c923dd 5#include <linux/anon_inodes.h>
11637e4b 6#include <linux/fsnotify_backend.h>
2a3edf86 7#include <linux/init.h>
a1014f10 8#include <linux/mount.h>
2a3edf86 9#include <linux/namei.h>
a1014f10 10#include <linux/poll.h>
11637e4b
EP
11#include <linux/security.h>
12#include <linux/syscalls.h>
e4e047a2 13#include <linux/slab.h>
2a3edf86 14#include <linux/types.h>
a1014f10
EP
15#include <linux/uaccess.h>
16
17#include <asm/ioctls.h>
11637e4b 18
2529a0df
EP
19#define FANOTIFY_DEFAULT_MAX_EVENTS 16384
20
33d3dfff 21extern const struct fsnotify_ops fanotify_fsnotify_ops;
11637e4b 22
2a3edf86 23static struct kmem_cache *fanotify_mark_cache __read_mostly;
b2d87909
EP
24static struct kmem_cache *fanotify_response_event_cache __read_mostly;
25
26struct fanotify_response_event {
27 struct list_head list;
28 __s32 fd;
29 struct fsnotify_event *event;
30};
2a3edf86 31
a1014f10
EP
32/*
33 * Get an fsnotify notification event if one exists and is small
34 * enough to fit in "count". Return an error pointer if the count
35 * is not large enough.
36 *
37 * Called with the group->notification_mutex held.
38 */
39static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
40 size_t count)
41{
42 BUG_ON(!mutex_is_locked(&group->notification_mutex));
43
44 pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
45
46 if (fsnotify_notify_queue_is_empty(group))
47 return NULL;
48
49 if (FAN_EVENT_METADATA_LEN > count)
50 return ERR_PTR(-EINVAL);
51
52 /* held the notification_mutex the whole time, so this is the
53 * same event we peeked above */
54 return fsnotify_remove_notify_event(group);
55}
56
22aa425d 57static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
a1014f10
EP
58{
59 int client_fd;
60 struct dentry *dentry;
61 struct vfsmount *mnt;
62 struct file *new_file;
63
22aa425d 64 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
a1014f10
EP
65
66 client_fd = get_unused_fd();
67 if (client_fd < 0)
68 return client_fd;
69
2069601b 70 if (event->data_type != FSNOTIFY_EVENT_PATH) {
a1014f10
EP
71 WARN_ON(1);
72 put_unused_fd(client_fd);
73 return -EINVAL;
74 }
75
76 /*
77 * we need a new file handle for the userspace program so it can read even if it was
78 * originally opened O_WRONLY.
79 */
2069601b
LT
80 dentry = dget(event->path.dentry);
81 mnt = mntget(event->path.mnt);
a1014f10
EP
82 /* it's possible this event was an overflow event. in that case dentry and mnt
83 * are NULL; That's fine, just don't call dentry open */
84 if (dentry && mnt)
85 new_file = dentry_open(dentry, mnt,
80af2588 86 group->fanotify_data.f_flags | FMODE_NONOTIFY,
a1014f10
EP
87 current_cred());
88 else
89 new_file = ERR_PTR(-EOVERFLOW);
90 if (IS_ERR(new_file)) {
91 /*
92 * we still send an event even if we can't open the file. this
93 * can happen when say tasks are gone and we try to open their
94 * /proc files or we try to open a WRONLY file like in sysfs
95 * we just send the errno to userspace since there isn't much
96 * else we can do.
97 */
98 put_unused_fd(client_fd);
99 client_fd = PTR_ERR(new_file);
100 } else {
101 fd_install(client_fd, new_file);
102 }
103
22aa425d 104 return client_fd;
a1014f10
EP
105}
106
107static ssize_t fill_event_metadata(struct fsnotify_group *group,
108 struct fanotify_event_metadata *metadata,
109 struct fsnotify_event *event)
110{
111 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
112 group, metadata, event);
113
114 metadata->event_len = FAN_EVENT_METADATA_LEN;
115 metadata->vers = FANOTIFY_METADATA_VERSION;
33d3dfff 116 metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
32c32632 117 metadata->pid = pid_vnr(event->tgid);
22aa425d 118 metadata->fd = create_fd(group, event);
a1014f10 119
22aa425d 120 return metadata->fd;
a1014f10
EP
121}
122
b2d87909
EP
123#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
124static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
125 __s32 fd)
126{
127 struct fanotify_response_event *re, *return_re = NULL;
128
129 mutex_lock(&group->fanotify_data.access_mutex);
130 list_for_each_entry(re, &group->fanotify_data.access_list, list) {
131 if (re->fd != fd)
132 continue;
133
134 list_del_init(&re->list);
135 return_re = re;
136 break;
137 }
138 mutex_unlock(&group->fanotify_data.access_mutex);
139
140 pr_debug("%s: found return_re=%p\n", __func__, return_re);
141
142 return return_re;
143}
144
145static int process_access_response(struct fsnotify_group *group,
146 struct fanotify_response *response_struct)
147{
148 struct fanotify_response_event *re;
149 __s32 fd = response_struct->fd;
150 __u32 response = response_struct->response;
151
152 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
153 fd, response);
154 /*
155 * make sure the response is valid, if invalid we do nothing and either
156 * userspace can send a valid responce or we will clean it up after the
157 * timeout
158 */
159 switch (response) {
160 case FAN_ALLOW:
161 case FAN_DENY:
162 break;
163 default:
164 return -EINVAL;
165 }
166
167 if (fd < 0)
168 return -EINVAL;
169
170 re = dequeue_re(group, fd);
171 if (!re)
172 return -ENOENT;
173
174 re->event->response = response;
175
176 wake_up(&group->fanotify_data.access_waitq);
177
178 kmem_cache_free(fanotify_response_event_cache, re);
179
180 return 0;
181}
182
183static int prepare_for_access_response(struct fsnotify_group *group,
184 struct fsnotify_event *event,
185 __s32 fd)
186{
187 struct fanotify_response_event *re;
188
189 if (!(event->mask & FAN_ALL_PERM_EVENTS))
190 return 0;
191
192 re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
193 if (!re)
194 return -ENOMEM;
195
196 re->event = event;
197 re->fd = fd;
198
199 mutex_lock(&group->fanotify_data.access_mutex);
2eebf582
EP
200
201 if (group->fanotify_data.bypass_perm) {
202 mutex_unlock(&group->fanotify_data.access_mutex);
203 kmem_cache_free(fanotify_response_event_cache, re);
204 event->response = FAN_ALLOW;
205 return 0;
206 }
207
b2d87909
EP
208 list_add_tail(&re->list, &group->fanotify_data.access_list);
209 mutex_unlock(&group->fanotify_data.access_mutex);
210
211 return 0;
212}
213
214static void remove_access_response(struct fsnotify_group *group,
215 struct fsnotify_event *event,
216 __s32 fd)
217{
218 struct fanotify_response_event *re;
219
220 if (!(event->mask & FAN_ALL_PERM_EVENTS))
221 return;
222
223 re = dequeue_re(group, fd);
224 if (!re)
225 return;
226
227 BUG_ON(re->event != event);
228
229 kmem_cache_free(fanotify_response_event_cache, re);
230
231 return;
232}
233#else
234static int prepare_for_access_response(struct fsnotify_group *group,
235 struct fsnotify_event *event,
236 __s32 fd)
237{
238 return 0;
239}
240
241static void remove_access_response(struct fsnotify_group *group,
242 struct fsnotify_event *event,
243 __s32 fd)
244{
8860f060 245 return;
b2d87909
EP
246}
247#endif
248
a1014f10
EP
249static ssize_t copy_event_to_user(struct fsnotify_group *group,
250 struct fsnotify_event *event,
251 char __user *buf)
252{
253 struct fanotify_event_metadata fanotify_event_metadata;
b2d87909 254 int fd, ret;
a1014f10
EP
255
256 pr_debug("%s: group=%p event=%p\n", __func__, group, event);
257
b2d87909
EP
258 fd = fill_event_metadata(group, &fanotify_event_metadata, event);
259 if (fd < 0)
260 return fd;
261
262 ret = prepare_for_access_response(group, event, fd);
263 if (ret)
264 goto out_close_fd;
a1014f10 265
b2d87909 266 ret = -EFAULT;
a1014f10 267 if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
b2d87909 268 goto out_kill_access_response;
a1014f10
EP
269
270 return FAN_EVENT_METADATA_LEN;
b2d87909
EP
271
272out_kill_access_response:
273 remove_access_response(group, event, fd);
274out_close_fd:
275 sys_close(fd);
276 return ret;
a1014f10
EP
277}
278
279/* intofiy userspace file descriptor functions */
280static unsigned int fanotify_poll(struct file *file, poll_table *wait)
281{
282 struct fsnotify_group *group = file->private_data;
283 int ret = 0;
284
285 poll_wait(file, &group->notification_waitq, wait);
286 mutex_lock(&group->notification_mutex);
287 if (!fsnotify_notify_queue_is_empty(group))
288 ret = POLLIN | POLLRDNORM;
289 mutex_unlock(&group->notification_mutex);
290
291 return ret;
292}
293
294static ssize_t fanotify_read(struct file *file, char __user *buf,
295 size_t count, loff_t *pos)
296{
297 struct fsnotify_group *group;
298 struct fsnotify_event *kevent;
299 char __user *start;
300 int ret;
301 DEFINE_WAIT(wait);
302
303 start = buf;
304 group = file->private_data;
305
306 pr_debug("%s: group=%p\n", __func__, group);
307
308 while (1) {
309 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
310
311 mutex_lock(&group->notification_mutex);
312 kevent = get_one_event(group, count);
313 mutex_unlock(&group->notification_mutex);
314
315 if (kevent) {
316 ret = PTR_ERR(kevent);
317 if (IS_ERR(kevent))
318 break;
319 ret = copy_event_to_user(group, kevent, buf);
320 fsnotify_put_event(kevent);
321 if (ret < 0)
322 break;
323 buf += ret;
324 count -= ret;
325 continue;
326 }
327
328 ret = -EAGAIN;
329 if (file->f_flags & O_NONBLOCK)
330 break;
331 ret = -EINTR;
332 if (signal_pending(current))
333 break;
334
335 if (start != buf)
336 break;
337
338 schedule();
339 }
340
341 finish_wait(&group->notification_waitq, &wait);
342 if (start != buf && ret != -EFAULT)
343 ret = buf - start;
344 return ret;
345}
346
b2d87909
EP
347static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
348{
349#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
350 struct fanotify_response response = { .fd = -1, .response = -1 };
351 struct fsnotify_group *group;
352 int ret;
353
354 group = file->private_data;
355
356 if (count > sizeof(response))
357 count = sizeof(response);
358
359 pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
360
361 if (copy_from_user(&response, buf, count))
362 return -EFAULT;
363
364 ret = process_access_response(group, &response);
365 if (ret < 0)
366 count = ret;
367
368 return count;
369#else
370 return -EINVAL;
371#endif
372}
373
52c923dd
EP
374static int fanotify_release(struct inode *ignored, struct file *file)
375{
376 struct fsnotify_group *group = file->private_data;
2eebf582 377 struct fanotify_response_event *re, *lre;
52c923dd
EP
378
379 pr_debug("%s: file=%p group=%p\n", __func__, file, group);
380
2eebf582
EP
381#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
382 mutex_lock(&group->fanotify_data.access_mutex);
383
384 group->fanotify_data.bypass_perm = true;
385
386 list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
387 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
388 re, re->event);
389
390 list_del_init(&re->list);
391 re->event->response = FAN_ALLOW;
392
393 kmem_cache_free(fanotify_response_event_cache, re);
394 }
395 mutex_unlock(&group->fanotify_data.access_mutex);
396
397 wake_up(&group->fanotify_data.access_waitq);
398#endif
52c923dd
EP
399 /* matches the fanotify_init->fsnotify_alloc_group */
400 fsnotify_put_group(group);
401
402 return 0;
403}
404
a1014f10
EP
405static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
406{
407 struct fsnotify_group *group;
408 struct fsnotify_event_holder *holder;
409 void __user *p;
410 int ret = -ENOTTY;
411 size_t send_len = 0;
412
413 group = file->private_data;
414
415 p = (void __user *) arg;
416
417 switch (cmd) {
418 case FIONREAD:
419 mutex_lock(&group->notification_mutex);
420 list_for_each_entry(holder, &group->notification_list, event_list)
421 send_len += FAN_EVENT_METADATA_LEN;
422 mutex_unlock(&group->notification_mutex);
423 ret = put_user(send_len, (int __user *) p);
424 break;
425 }
426
427 return ret;
428}
429
52c923dd 430static const struct file_operations fanotify_fops = {
a1014f10
EP
431 .poll = fanotify_poll,
432 .read = fanotify_read,
b2d87909 433 .write = fanotify_write,
52c923dd
EP
434 .fasync = NULL,
435 .release = fanotify_release,
a1014f10
EP
436 .unlocked_ioctl = fanotify_ioctl,
437 .compat_ioctl = fanotify_ioctl,
6038f373 438 .llseek = noop_llseek,
52c923dd
EP
439};
440
2a3edf86
EP
441static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
442{
443 kmem_cache_free(fanotify_mark_cache, fsn_mark);
444}
445
446static int fanotify_find_path(int dfd, const char __user *filename,
447 struct path *path, unsigned int flags)
448{
449 int ret;
450
451 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
452 dfd, filename, flags);
453
454 if (filename == NULL) {
455 struct file *file;
456 int fput_needed;
457
458 ret = -EBADF;
459 file = fget_light(dfd, &fput_needed);
460 if (!file)
461 goto out;
462
463 ret = -ENOTDIR;
464 if ((flags & FAN_MARK_ONLYDIR) &&
465 !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
466 fput_light(file, fput_needed);
467 goto out;
468 }
469
470 *path = file->f_path;
471 path_get(path);
472 fput_light(file, fput_needed);
473 } else {
474 unsigned int lookup_flags = 0;
475
476 if (!(flags & FAN_MARK_DONT_FOLLOW))
477 lookup_flags |= LOOKUP_FOLLOW;
478 if (flags & FAN_MARK_ONLYDIR)
479 lookup_flags |= LOOKUP_DIRECTORY;
480
481 ret = user_path_at(dfd, filename, lookup_flags, path);
482 if (ret)
483 goto out;
484 }
485
486 /* you can only watch an inode if you have read permissions on it */
487 ret = inode_permission(path->dentry->d_inode, MAY_READ);
488 if (ret)
489 path_put(path);
490out:
491 return ret;
492}
493
b9e4e3bd
EP
494static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
495 __u32 mask,
496 unsigned int flags)
088b09b0
AG
497{
498 __u32 oldmask;
499
500 spin_lock(&fsn_mark->lock);
b9e4e3bd
EP
501 if (!(flags & FAN_MARK_IGNORED_MASK)) {
502 oldmask = fsn_mark->mask;
503 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
504 } else {
505 oldmask = fsn_mark->ignored_mask;
506 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
507 }
088b09b0
AG
508 spin_unlock(&fsn_mark->lock);
509
510 if (!(oldmask & ~mask))
511 fsnotify_destroy_mark(fsn_mark);
512
513 return mask & oldmask;
514}
515
f3640192 516static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
b9e4e3bd
EP
517 struct vfsmount *mnt, __u32 mask,
518 unsigned int flags)
88826276
EP
519{
520 struct fsnotify_mark *fsn_mark = NULL;
088b09b0 521 __u32 removed;
88826276 522
f3640192
AG
523 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
524 if (!fsn_mark)
525 return -ENOENT;
88826276 526
b9e4e3bd 527 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
f3640192 528 fsnotify_put_mark(fsn_mark);
f3640192
AG
529 if (removed & mnt->mnt_fsnotify_mask)
530 fsnotify_recalc_vfsmount_mask(mnt);
531
532 return 0;
533}
2a3edf86 534
f3640192 535static int fanotify_remove_inode_mark(struct fsnotify_group *group,
b9e4e3bd
EP
536 struct inode *inode, __u32 mask,
537 unsigned int flags)
f3640192
AG
538{
539 struct fsnotify_mark *fsn_mark = NULL;
540 __u32 removed;
541
542 fsn_mark = fsnotify_find_inode_mark(group, inode);
88826276
EP
543 if (!fsn_mark)
544 return -ENOENT;
545
b9e4e3bd 546 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
5444e298 547 /* matches the fsnotify_find_inode_mark() */
2a3edf86 548 fsnotify_put_mark(fsn_mark);
f3640192
AG
549 if (removed & inode->i_fsnotify_mask)
550 fsnotify_recalc_inode_mask(inode);
088b09b0 551
2a3edf86
EP
552 return 0;
553}
554
b9e4e3bd
EP
555static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
556 __u32 mask,
557 unsigned int flags)
912ee394
AG
558{
559 __u32 oldmask;
560
561 spin_lock(&fsn_mark->lock);
b9e4e3bd
EP
562 if (!(flags & FAN_MARK_IGNORED_MASK)) {
563 oldmask = fsn_mark->mask;
564 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
565 } else {
566 oldmask = fsn_mark->ignored_mask;
567 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask | mask));
c9778a98
EP
568 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
569 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
b9e4e3bd 570 }
912ee394
AG
571 spin_unlock(&fsn_mark->lock);
572
573 return mask & ~oldmask;
574}
575
52202dfb 576static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
b9e4e3bd
EP
577 struct vfsmount *mnt, __u32 mask,
578 unsigned int flags)
2a3edf86
EP
579{
580 struct fsnotify_mark *fsn_mark;
912ee394 581 __u32 added;
2a3edf86 582
88826276
EP
583 fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
584 if (!fsn_mark) {
88826276
EP
585 int ret;
586
912ee394
AG
587 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
588 if (!fsn_mark)
52202dfb 589 return -ENOMEM;
88826276 590
912ee394
AG
591 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
592 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
88826276 593 if (ret) {
912ee394 594 fanotify_free_mark(fsn_mark);
52202dfb 595 return ret;
88826276 596 }
88826276 597 }
b9e4e3bd 598 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
52202dfb 599 fsnotify_put_mark(fsn_mark);
43709a28
EP
600 if (added & ~mnt->mnt_fsnotify_mask)
601 fsnotify_recalc_vfsmount_mask(mnt);
602
52202dfb 603 return 0;
88826276
EP
604}
605
52202dfb 606static int fanotify_add_inode_mark(struct fsnotify_group *group,
b9e4e3bd
EP
607 struct inode *inode, __u32 mask,
608 unsigned int flags)
88826276
EP
609{
610 struct fsnotify_mark *fsn_mark;
912ee394 611 __u32 added;
88826276
EP
612
613 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
2a3edf86 614
5322a59f
EP
615 /*
616 * If some other task has this inode open for write we should not add
617 * an ignored mark, unless that ignored mark is supposed to survive
618 * modification changes anyway.
619 */
620 if ((flags & FAN_MARK_IGNORED_MASK) &&
621 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
622 (atomic_read(&inode->i_writecount) > 0))
623 return 0;
624
5444e298 625 fsn_mark = fsnotify_find_inode_mark(group, inode);
2a3edf86 626 if (!fsn_mark) {
88826276 627 int ret;
2a3edf86 628
912ee394
AG
629 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
630 if (!fsn_mark)
52202dfb 631 return -ENOMEM;
2a3edf86 632
912ee394
AG
633 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
634 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
2a3edf86 635 if (ret) {
912ee394 636 fanotify_free_mark(fsn_mark);
52202dfb 637 return ret;
2a3edf86 638 }
2a3edf86 639 }
b9e4e3bd 640 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
52202dfb 641 fsnotify_put_mark(fsn_mark);
43709a28
EP
642 if (added & ~inode->i_fsnotify_mask)
643 fsnotify_recalc_inode_mask(inode);
52202dfb 644 return 0;
88826276 645}
2a3edf86 646
52c923dd 647/* fanotify syscalls */
08ae8938 648SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
11637e4b 649{
52c923dd
EP
650 struct fsnotify_group *group;
651 int f_flags, fd;
652
08ae8938
EP
653 pr_debug("%s: flags=%d event_f_flags=%d\n",
654 __func__, flags, event_f_flags);
52c923dd 655
52c923dd 656 if (!capable(CAP_SYS_ADMIN))
a2f13ad0 657 return -EPERM;
52c923dd
EP
658
659 if (flags & ~FAN_ALL_INIT_FLAGS)
660 return -EINVAL;
661
b2d87909 662 f_flags = O_RDWR | FMODE_NONOTIFY;
52c923dd
EP
663 if (flags & FAN_CLOEXEC)
664 f_flags |= O_CLOEXEC;
665 if (flags & FAN_NONBLOCK)
666 f_flags |= O_NONBLOCK;
667
668 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */
669 group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
670 if (IS_ERR(group))
671 return PTR_ERR(group);
672
80af2588 673 group->fanotify_data.f_flags = event_f_flags;
9e66e423
EP
674#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
675 mutex_init(&group->fanotify_data.access_mutex);
676 init_waitqueue_head(&group->fanotify_data.access_waitq);
677 INIT_LIST_HEAD(&group->fanotify_data.access_list);
678#endif
4231a235
EP
679 switch (flags & FAN_ALL_CLASS_BITS) {
680 case FAN_CLASS_NOTIF:
681 group->priority = FS_PRIO_0;
682 break;
683 case FAN_CLASS_CONTENT:
684 group->priority = FS_PRIO_1;
685 break;
686 case FAN_CLASS_PRE_CONTENT:
687 group->priority = FS_PRIO_2;
688 break;
689 default:
690 fd = -EINVAL;
691 goto out_put_group;
692 }
cb2d429f 693
5dd03f55
EP
694 if (flags & FAN_UNLIMITED_QUEUE) {
695 fd = -EPERM;
696 if (!capable(CAP_SYS_ADMIN))
697 goto out_put_group;
698 group->max_events = UINT_MAX;
699 } else {
700 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
701 }
2529a0df 702
52c923dd
EP
703 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
704 if (fd < 0)
705 goto out_put_group;
706
707 return fd;
708
709out_put_group:
710 fsnotify_put_group(group);
711 return fd;
11637e4b 712}
bbaa4168 713
9bbfc964
HC
714SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
715 __u64 mask, int dfd,
716 const char __user * pathname)
bbaa4168 717{
0ff21db9
EP
718 struct inode *inode = NULL;
719 struct vfsmount *mnt = NULL;
2a3edf86
EP
720 struct fsnotify_group *group;
721 struct file *filp;
722 struct path path;
723 int ret, fput_needed;
724
725 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
726 __func__, fanotify_fd, flags, dfd, pathname, mask);
727
728 /* we only use the lower 32 bits as of right now. */
729 if (mask & ((__u64)0xffffffff << 32))
730 return -EINVAL;
731
88380fe6
AG
732 if (flags & ~FAN_ALL_MARK_FLAGS)
733 return -EINVAL;
4d92604c 734 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
88380fe6
AG
735 case FAN_MARK_ADD:
736 case FAN_MARK_REMOVE:
4d92604c 737 case FAN_MARK_FLUSH:
88380fe6
AG
738 break;
739 default:
740 return -EINVAL;
741 }
b2d87909
EP
742#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
743 if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
744#else
88380fe6 745 if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
b2d87909 746#endif
2a3edf86
EP
747 return -EINVAL;
748
749 filp = fget_light(fanotify_fd, &fput_needed);
750 if (unlikely(!filp))
751 return -EBADF;
752
753 /* verify that this is indeed an fanotify instance */
754 ret = -EINVAL;
755 if (unlikely(filp->f_op != &fanotify_fops))
756 goto fput_and_out;
4231a235
EP
757 group = filp->private_data;
758
759 /*
760 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not
761 * allowed to set permissions events.
762 */
763 ret = -EINVAL;
764 if (mask & FAN_ALL_PERM_EVENTS &&
765 group->priority == FS_PRIO_0)
766 goto fput_and_out;
2a3edf86
EP
767
768 ret = fanotify_find_path(dfd, pathname, &path, flags);
769 if (ret)
770 goto fput_and_out;
771
772 /* inode held in place by reference to path; group by fget on fd */
eac8e9e8 773 if (!(flags & FAN_MARK_MOUNT))
0ff21db9
EP
774 inode = path.dentry->d_inode;
775 else
776 mnt = path.mnt;
2a3edf86
EP
777
778 /* create/update an inode mark */
4d92604c 779 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
c6223f46 780 case FAN_MARK_ADD:
eac8e9e8 781 if (flags & FAN_MARK_MOUNT)
b9e4e3bd 782 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
0ff21db9 783 else
b9e4e3bd 784 ret = fanotify_add_inode_mark(group, inode, mask, flags);
c6223f46
AG
785 break;
786 case FAN_MARK_REMOVE:
f3640192 787 if (flags & FAN_MARK_MOUNT)
b9e4e3bd 788 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
f3640192 789 else
b9e4e3bd 790 ret = fanotify_remove_inode_mark(group, inode, mask, flags);
c6223f46 791 break;
4d92604c
EP
792 case FAN_MARK_FLUSH:
793 if (flags & FAN_MARK_MOUNT)
794 fsnotify_clear_vfsmount_marks_by_group(group);
795 else
796 fsnotify_clear_inode_marks_by_group(group);
4d92604c 797 break;
c6223f46
AG
798 default:
799 ret = -EINVAL;
800 }
2a3edf86
EP
801
802 path_put(&path);
803fput_and_out:
804 fput_light(filp, fput_needed);
805 return ret;
806}
807
9bbfc964
HC
808#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
809asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
810 long dfd, long pathname)
811{
812 return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
813 mask, (int) dfd,
814 (const char __user *) pathname);
815}
816SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
817#endif
818
2a3edf86
EP
819/*
820 * fanotify_user_setup - Our initialization function. Note that we cannnot return
821 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
822 * must result in panic().
823 */
824static int __init fanotify_user_setup(void)
825{
826 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
b2d87909
EP
827 fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
828 SLAB_PANIC);
2a3edf86
EP
829
830 return 0;
bbaa4168 831}
2a3edf86 832device_initcall(fanotify_user_setup);