Commit | Line | Data |
---|---|---|
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 | |
c63181e6 | 19 | #include "../../mount.h" |
be77196b | 20 | #include "../fdinfo.h" |
c63181e6 | 21 | |
2529a0df | 22 | #define FANOTIFY_DEFAULT_MAX_EVENTS 16384 |
e7099d8a | 23 | #define FANOTIFY_DEFAULT_MAX_MARKS 8192 |
4afeff85 | 24 | #define FANOTIFY_DEFAULT_MAX_LISTENERS 128 |
2529a0df | 25 | |
33d3dfff | 26 | extern const struct fsnotify_ops fanotify_fsnotify_ops; |
11637e4b | 27 | |
2a3edf86 | 28 | static struct kmem_cache *fanotify_mark_cache __read_mostly; |
b2d87909 EP |
29 | static struct kmem_cache *fanotify_response_event_cache __read_mostly; |
30 | ||
31 | struct fanotify_response_event { | |
32 | struct list_head list; | |
33 | __s32 fd; | |
34 | struct fsnotify_event *event; | |
35 | }; | |
2a3edf86 | 36 | |
a1014f10 EP |
37 | /* |
38 | * Get an fsnotify notification event if one exists and is small | |
39 | * enough to fit in "count". Return an error pointer if the count | |
40 | * is not large enough. | |
41 | * | |
42 | * Called with the group->notification_mutex held. | |
43 | */ | |
44 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, | |
45 | size_t count) | |
46 | { | |
47 | BUG_ON(!mutex_is_locked(&group->notification_mutex)); | |
48 | ||
49 | pr_debug("%s: group=%p count=%zd\n", __func__, group, count); | |
50 | ||
51 | if (fsnotify_notify_queue_is_empty(group)) | |
52 | return NULL; | |
53 | ||
54 | if (FAN_EVENT_METADATA_LEN > count) | |
55 | return ERR_PTR(-EINVAL); | |
56 | ||
57 | /* held the notification_mutex the whole time, so this is the | |
58 | * same event we peeked above */ | |
59 | return fsnotify_remove_notify_event(group); | |
60 | } | |
61 | ||
352e3b24 AV |
62 | static int create_fd(struct fsnotify_group *group, |
63 | struct fsnotify_event *event, | |
64 | struct file **file) | |
a1014f10 EP |
65 | { |
66 | int client_fd; | |
a1014f10 EP |
67 | struct file *new_file; |
68 | ||
22aa425d | 69 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
a1014f10 EP |
70 | |
71 | client_fd = get_unused_fd(); | |
72 | if (client_fd < 0) | |
73 | return client_fd; | |
74 | ||
2069601b | 75 | if (event->data_type != FSNOTIFY_EVENT_PATH) { |
a1014f10 EP |
76 | WARN_ON(1); |
77 | put_unused_fd(client_fd); | |
78 | return -EINVAL; | |
79 | } | |
80 | ||
81 | /* | |
82 | * we need a new file handle for the userspace program so it can read even if it was | |
83 | * originally opened O_WRONLY. | |
84 | */ | |
a1014f10 EP |
85 | /* it's possible this event was an overflow event. in that case dentry and mnt |
86 | * are NULL; That's fine, just don't call dentry open */ | |
765927b2 AV |
87 | if (event->path.dentry && event->path.mnt) |
88 | new_file = dentry_open(&event->path, | |
80af2588 | 89 | group->fanotify_data.f_flags | FMODE_NONOTIFY, |
a1014f10 EP |
90 | current_cred()); |
91 | else | |
92 | new_file = ERR_PTR(-EOVERFLOW); | |
93 | if (IS_ERR(new_file)) { | |
94 | /* | |
95 | * we still send an event even if we can't open the file. this | |
96 | * can happen when say tasks are gone and we try to open their | |
97 | * /proc files or we try to open a WRONLY file like in sysfs | |
98 | * we just send the errno to userspace since there isn't much | |
99 | * else we can do. | |
100 | */ | |
101 | put_unused_fd(client_fd); | |
102 | client_fd = PTR_ERR(new_file); | |
103 | } else { | |
352e3b24 | 104 | *file = new_file; |
a1014f10 EP |
105 | } |
106 | ||
22aa425d | 107 | return client_fd; |
a1014f10 EP |
108 | } |
109 | ||
ecf6f5e7 | 110 | static int fill_event_metadata(struct fsnotify_group *group, |
a1014f10 | 111 | struct fanotify_event_metadata *metadata, |
352e3b24 AV |
112 | struct fsnotify_event *event, |
113 | struct file **file) | |
a1014f10 | 114 | { |
fdbf3cee LS |
115 | int ret = 0; |
116 | ||
a1014f10 EP |
117 | pr_debug("%s: group=%p metadata=%p event=%p\n", __func__, |
118 | group, metadata, event); | |
119 | ||
352e3b24 | 120 | *file = NULL; |
a1014f10 | 121 | metadata->event_len = FAN_EVENT_METADATA_LEN; |
7d131623 | 122 | metadata->metadata_len = FAN_EVENT_METADATA_LEN; |
a1014f10 | 123 | metadata->vers = FANOTIFY_METADATA_VERSION; |
33d3dfff | 124 | metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS; |
32c32632 | 125 | metadata->pid = pid_vnr(event->tgid); |
fdbf3cee LS |
126 | if (unlikely(event->mask & FAN_Q_OVERFLOW)) |
127 | metadata->fd = FAN_NOFD; | |
128 | else { | |
352e3b24 | 129 | metadata->fd = create_fd(group, event, file); |
fdbf3cee LS |
130 | if (metadata->fd < 0) |
131 | ret = metadata->fd; | |
132 | } | |
a1014f10 | 133 | |
fdbf3cee | 134 | return ret; |
a1014f10 EP |
135 | } |
136 | ||
b2d87909 EP |
137 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS |
138 | static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group, | |
139 | __s32 fd) | |
140 | { | |
141 | struct fanotify_response_event *re, *return_re = NULL; | |
142 | ||
143 | mutex_lock(&group->fanotify_data.access_mutex); | |
144 | list_for_each_entry(re, &group->fanotify_data.access_list, list) { | |
145 | if (re->fd != fd) | |
146 | continue; | |
147 | ||
148 | list_del_init(&re->list); | |
149 | return_re = re; | |
150 | break; | |
151 | } | |
152 | mutex_unlock(&group->fanotify_data.access_mutex); | |
153 | ||
154 | pr_debug("%s: found return_re=%p\n", __func__, return_re); | |
155 | ||
156 | return return_re; | |
157 | } | |
158 | ||
159 | static int process_access_response(struct fsnotify_group *group, | |
160 | struct fanotify_response *response_struct) | |
161 | { | |
162 | struct fanotify_response_event *re; | |
163 | __s32 fd = response_struct->fd; | |
164 | __u32 response = response_struct->response; | |
165 | ||
166 | pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group, | |
167 | fd, response); | |
168 | /* | |
169 | * make sure the response is valid, if invalid we do nothing and either | |
25985edc | 170 | * userspace can send a valid response or we will clean it up after the |
b2d87909 EP |
171 | * timeout |
172 | */ | |
173 | switch (response) { | |
174 | case FAN_ALLOW: | |
175 | case FAN_DENY: | |
176 | break; | |
177 | default: | |
178 | return -EINVAL; | |
179 | } | |
180 | ||
181 | if (fd < 0) | |
182 | return -EINVAL; | |
183 | ||
184 | re = dequeue_re(group, fd); | |
185 | if (!re) | |
186 | return -ENOENT; | |
187 | ||
188 | re->event->response = response; | |
189 | ||
190 | wake_up(&group->fanotify_data.access_waitq); | |
191 | ||
192 | kmem_cache_free(fanotify_response_event_cache, re); | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
197 | static int prepare_for_access_response(struct fsnotify_group *group, | |
198 | struct fsnotify_event *event, | |
199 | __s32 fd) | |
200 | { | |
201 | struct fanotify_response_event *re; | |
202 | ||
203 | if (!(event->mask & FAN_ALL_PERM_EVENTS)) | |
204 | return 0; | |
205 | ||
206 | re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL); | |
207 | if (!re) | |
208 | return -ENOMEM; | |
209 | ||
210 | re->event = event; | |
211 | re->fd = fd; | |
212 | ||
213 | mutex_lock(&group->fanotify_data.access_mutex); | |
2eebf582 | 214 | |
09e5f14e | 215 | if (atomic_read(&group->fanotify_data.bypass_perm)) { |
2eebf582 EP |
216 | mutex_unlock(&group->fanotify_data.access_mutex); |
217 | kmem_cache_free(fanotify_response_event_cache, re); | |
218 | event->response = FAN_ALLOW; | |
219 | return 0; | |
220 | } | |
221 | ||
b2d87909 EP |
222 | list_add_tail(&re->list, &group->fanotify_data.access_list); |
223 | mutex_unlock(&group->fanotify_data.access_mutex); | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
b2d87909 EP |
228 | #else |
229 | static int prepare_for_access_response(struct fsnotify_group *group, | |
230 | struct fsnotify_event *event, | |
231 | __s32 fd) | |
232 | { | |
233 | return 0; | |
234 | } | |
235 | ||
b2d87909 EP |
236 | #endif |
237 | ||
a1014f10 EP |
238 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
239 | struct fsnotify_event *event, | |
240 | char __user *buf) | |
241 | { | |
242 | struct fanotify_event_metadata fanotify_event_metadata; | |
352e3b24 | 243 | struct file *f; |
b2d87909 | 244 | int fd, ret; |
a1014f10 EP |
245 | |
246 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); | |
247 | ||
352e3b24 | 248 | ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f); |
ecf6f5e7 EP |
249 | if (ret < 0) |
250 | goto out; | |
b2d87909 | 251 | |
fdbf3cee | 252 | fd = fanotify_event_metadata.fd; |
b2d87909 | 253 | ret = -EFAULT; |
7d131623 EP |
254 | if (copy_to_user(buf, &fanotify_event_metadata, |
255 | fanotify_event_metadata.event_len)) | |
352e3b24 AV |
256 | goto out_close_fd; |
257 | ||
258 | ret = prepare_for_access_response(group, event, fd); | |
259 | if (ret) | |
260 | goto out_close_fd; | |
a1014f10 | 261 | |
3587b1b0 AV |
262 | if (fd != FAN_NOFD) |
263 | fd_install(fd, f); | |
7d131623 | 264 | return fanotify_event_metadata.event_len; |
b2d87909 | 265 | |
b2d87909 | 266 | out_close_fd: |
352e3b24 AV |
267 | if (fd != FAN_NOFD) { |
268 | put_unused_fd(fd); | |
269 | fput(f); | |
270 | } | |
ecf6f5e7 EP |
271 | out: |
272 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS | |
273 | if (event->mask & FAN_ALL_PERM_EVENTS) { | |
274 | event->response = FAN_DENY; | |
275 | wake_up(&group->fanotify_data.access_waitq); | |
276 | } | |
277 | #endif | |
b2d87909 | 278 | return ret; |
a1014f10 EP |
279 | } |
280 | ||
281 | /* intofiy userspace file descriptor functions */ | |
282 | static unsigned int fanotify_poll(struct file *file, poll_table *wait) | |
283 | { | |
284 | struct fsnotify_group *group = file->private_data; | |
285 | int ret = 0; | |
286 | ||
287 | poll_wait(file, &group->notification_waitq, wait); | |
288 | mutex_lock(&group->notification_mutex); | |
289 | if (!fsnotify_notify_queue_is_empty(group)) | |
290 | ret = POLLIN | POLLRDNORM; | |
291 | mutex_unlock(&group->notification_mutex); | |
292 | ||
293 | return ret; | |
294 | } | |
295 | ||
296 | static ssize_t fanotify_read(struct file *file, char __user *buf, | |
297 | size_t count, loff_t *pos) | |
298 | { | |
299 | struct fsnotify_group *group; | |
300 | struct fsnotify_event *kevent; | |
301 | char __user *start; | |
302 | int ret; | |
303 | DEFINE_WAIT(wait); | |
304 | ||
305 | start = buf; | |
306 | group = file->private_data; | |
307 | ||
308 | pr_debug("%s: group=%p\n", __func__, group); | |
309 | ||
310 | while (1) { | |
311 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); | |
312 | ||
313 | mutex_lock(&group->notification_mutex); | |
314 | kevent = get_one_event(group, count); | |
315 | mutex_unlock(&group->notification_mutex); | |
316 | ||
317 | if (kevent) { | |
318 | ret = PTR_ERR(kevent); | |
319 | if (IS_ERR(kevent)) | |
320 | break; | |
321 | ret = copy_event_to_user(group, kevent, buf); | |
322 | fsnotify_put_event(kevent); | |
323 | if (ret < 0) | |
324 | break; | |
325 | buf += ret; | |
326 | count -= ret; | |
327 | continue; | |
328 | } | |
329 | ||
330 | ret = -EAGAIN; | |
331 | if (file->f_flags & O_NONBLOCK) | |
332 | break; | |
1a5cea72 | 333 | ret = -ERESTARTSYS; |
a1014f10 EP |
334 | if (signal_pending(current)) |
335 | break; | |
336 | ||
337 | if (start != buf) | |
338 | break; | |
339 | ||
340 | schedule(); | |
341 | } | |
342 | ||
343 | finish_wait(&group->notification_waitq, &wait); | |
344 | if (start != buf && ret != -EFAULT) | |
345 | ret = buf - start; | |
346 | return ret; | |
347 | } | |
348 | ||
b2d87909 EP |
349 | static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) |
350 | { | |
351 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS | |
352 | struct fanotify_response response = { .fd = -1, .response = -1 }; | |
353 | struct fsnotify_group *group; | |
354 | int ret; | |
355 | ||
356 | group = file->private_data; | |
357 | ||
358 | if (count > sizeof(response)) | |
359 | count = sizeof(response); | |
360 | ||
361 | pr_debug("%s: group=%p count=%zu\n", __func__, group, count); | |
362 | ||
363 | if (copy_from_user(&response, buf, count)) | |
364 | return -EFAULT; | |
365 | ||
366 | ret = process_access_response(group, &response); | |
367 | if (ret < 0) | |
368 | count = ret; | |
369 | ||
370 | return count; | |
371 | #else | |
372 | return -EINVAL; | |
373 | #endif | |
374 | } | |
375 | ||
52c923dd EP |
376 | static int fanotify_release(struct inode *ignored, struct file *file) |
377 | { | |
378 | struct fsnotify_group *group = file->private_data; | |
52c923dd | 379 | |
2eebf582 | 380 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS |
19ba54f4 AM |
381 | struct fanotify_response_event *re, *lre; |
382 | ||
2eebf582 EP |
383 | mutex_lock(&group->fanotify_data.access_mutex); |
384 | ||
09e5f14e | 385 | atomic_inc(&group->fanotify_data.bypass_perm); |
2eebf582 EP |
386 | |
387 | list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) { | |
388 | pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group, | |
389 | re, re->event); | |
390 | ||
391 | list_del_init(&re->list); | |
392 | re->event->response = FAN_ALLOW; | |
393 | ||
394 | kmem_cache_free(fanotify_response_event_cache, re); | |
395 | } | |
396 | mutex_unlock(&group->fanotify_data.access_mutex); | |
397 | ||
398 | wake_up(&group->fanotify_data.access_waitq); | |
399 | #endif | |
0a6b6bd5 EP |
400 | |
401 | if (file->f_flags & FASYNC) | |
402 | fsnotify_fasync(-1, file, 0); | |
403 | ||
52c923dd | 404 | /* matches the fanotify_init->fsnotify_alloc_group */ |
d8153d4d | 405 | fsnotify_destroy_group(group); |
52c923dd EP |
406 | |
407 | return 0; | |
408 | } | |
409 | ||
a1014f10 EP |
410 | static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
411 | { | |
412 | struct fsnotify_group *group; | |
413 | struct fsnotify_event_holder *holder; | |
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 | mutex_lock(&group->notification_mutex); | |
425 | list_for_each_entry(holder, &group->notification_list, event_list) | |
426 | send_len += FAN_EVENT_METADATA_LEN; | |
427 | mutex_unlock(&group->notification_mutex); | |
428 | ret = put_user(send_len, (int __user *) p); | |
429 | break; | |
430 | } | |
431 | ||
432 | return ret; | |
433 | } | |
434 | ||
52c923dd | 435 | static const struct file_operations fanotify_fops = { |
be77196b | 436 | .show_fdinfo = fanotify_show_fdinfo, |
a1014f10 EP |
437 | .poll = fanotify_poll, |
438 | .read = fanotify_read, | |
b2d87909 | 439 | .write = fanotify_write, |
52c923dd EP |
440 | .fasync = NULL, |
441 | .release = fanotify_release, | |
a1014f10 EP |
442 | .unlocked_ioctl = fanotify_ioctl, |
443 | .compat_ioctl = fanotify_ioctl, | |
6038f373 | 444 | .llseek = noop_llseek, |
52c923dd EP |
445 | }; |
446 | ||
2a3edf86 EP |
447 | static void fanotify_free_mark(struct fsnotify_mark *fsn_mark) |
448 | { | |
449 | kmem_cache_free(fanotify_mark_cache, fsn_mark); | |
450 | } | |
451 | ||
452 | static 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) { | |
2903ff01 | 461 | struct fd f = fdget(dfd); |
2a3edf86 EP |
462 | |
463 | ret = -EBADF; | |
2903ff01 | 464 | if (!f.file) |
2a3edf86 EP |
465 | goto out; |
466 | ||
467 | ret = -ENOTDIR; | |
468 | if ((flags & FAN_MARK_ONLYDIR) && | |
496ad9aa | 469 | !(S_ISDIR(file_inode(f.file)->i_mode))) { |
2903ff01 | 470 | fdput(f); |
2a3edf86 EP |
471 | goto out; |
472 | } | |
473 | ||
2903ff01 | 474 | *path = f.file->f_path; |
2a3edf86 | 475 | path_get(path); |
2903ff01 | 476 | fdput(f); |
2a3edf86 EP |
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); | |
494 | out: | |
495 | return ret; | |
496 | } | |
497 | ||
b9e4e3bd EP |
498 | static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, |
499 | __u32 mask, | |
6dfbd149 LS |
500 | unsigned int flags, |
501 | int *destroy) | |
088b09b0 AG |
502 | { |
503 | __u32 oldmask; | |
504 | ||
505 | spin_lock(&fsn_mark->lock); | |
b9e4e3bd EP |
506 | if (!(flags & FAN_MARK_IGNORED_MASK)) { |
507 | oldmask = fsn_mark->mask; | |
508 | fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask)); | |
509 | } else { | |
510 | oldmask = fsn_mark->ignored_mask; | |
511 | fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask)); | |
512 | } | |
088b09b0 AG |
513 | spin_unlock(&fsn_mark->lock); |
514 | ||
6dfbd149 | 515 | *destroy = !(oldmask & ~mask); |
088b09b0 AG |
516 | |
517 | return mask & oldmask; | |
518 | } | |
519 | ||
f3640192 | 520 | static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group, |
b9e4e3bd EP |
521 | struct vfsmount *mnt, __u32 mask, |
522 | unsigned int flags) | |
88826276 EP |
523 | { |
524 | struct fsnotify_mark *fsn_mark = NULL; | |
088b09b0 | 525 | __u32 removed; |
6dfbd149 | 526 | int destroy_mark; |
88826276 | 527 | |
f3640192 AG |
528 | fsn_mark = fsnotify_find_vfsmount_mark(group, mnt); |
529 | if (!fsn_mark) | |
530 | return -ENOENT; | |
88826276 | 531 | |
6dfbd149 LS |
532 | removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, |
533 | &destroy_mark); | |
534 | if (destroy_mark) | |
e2a29943 | 535 | fsnotify_destroy_mark(fsn_mark, group); |
6dfbd149 | 536 | |
f3640192 | 537 | fsnotify_put_mark(fsn_mark); |
c63181e6 | 538 | if (removed & real_mount(mnt)->mnt_fsnotify_mask) |
f3640192 AG |
539 | fsnotify_recalc_vfsmount_mask(mnt); |
540 | ||
541 | return 0; | |
542 | } | |
2a3edf86 | 543 | |
f3640192 | 544 | static int fanotify_remove_inode_mark(struct fsnotify_group *group, |
b9e4e3bd EP |
545 | struct inode *inode, __u32 mask, |
546 | unsigned int flags) | |
f3640192 AG |
547 | { |
548 | struct fsnotify_mark *fsn_mark = NULL; | |
549 | __u32 removed; | |
6dfbd149 | 550 | int destroy_mark; |
f3640192 AG |
551 | |
552 | fsn_mark = fsnotify_find_inode_mark(group, inode); | |
88826276 EP |
553 | if (!fsn_mark) |
554 | return -ENOENT; | |
555 | ||
6dfbd149 LS |
556 | removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, |
557 | &destroy_mark); | |
558 | if (destroy_mark) | |
e2a29943 | 559 | fsnotify_destroy_mark(fsn_mark, group); |
5444e298 | 560 | /* matches the fsnotify_find_inode_mark() */ |
2a3edf86 | 561 | fsnotify_put_mark(fsn_mark); |
f3640192 AG |
562 | if (removed & inode->i_fsnotify_mask) |
563 | fsnotify_recalc_inode_mask(inode); | |
088b09b0 | 564 | |
2a3edf86 EP |
565 | return 0; |
566 | } | |
567 | ||
b9e4e3bd EP |
568 | static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, |
569 | __u32 mask, | |
570 | unsigned int flags) | |
912ee394 | 571 | { |
192ca4d1 | 572 | __u32 oldmask = -1; |
912ee394 AG |
573 | |
574 | spin_lock(&fsn_mark->lock); | |
b9e4e3bd EP |
575 | if (!(flags & FAN_MARK_IGNORED_MASK)) { |
576 | oldmask = fsn_mark->mask; | |
577 | fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask)); | |
578 | } else { | |
192ca4d1 EP |
579 | __u32 tmask = fsn_mark->ignored_mask | mask; |
580 | fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask); | |
c9778a98 EP |
581 | if (flags & FAN_MARK_IGNORED_SURV_MODIFY) |
582 | fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY; | |
b9e4e3bd | 583 | } |
8fcd6528 EP |
584 | |
585 | if (!(flags & FAN_MARK_ONDIR)) { | |
586 | __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR; | |
587 | fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask); | |
588 | } | |
589 | ||
912ee394 AG |
590 | spin_unlock(&fsn_mark->lock); |
591 | ||
592 | return mask & ~oldmask; | |
593 | } | |
594 | ||
52202dfb | 595 | static int fanotify_add_vfsmount_mark(struct fsnotify_group *group, |
b9e4e3bd EP |
596 | struct vfsmount *mnt, __u32 mask, |
597 | unsigned int flags) | |
2a3edf86 EP |
598 | { |
599 | struct fsnotify_mark *fsn_mark; | |
912ee394 | 600 | __u32 added; |
fa218ab9 | 601 | int ret = 0; |
2a3edf86 | 602 | |
88826276 EP |
603 | fsn_mark = fsnotify_find_vfsmount_mark(group, mnt); |
604 | if (!fsn_mark) { | |
e7099d8a EP |
605 | if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) |
606 | return -ENOSPC; | |
607 | ||
912ee394 AG |
608 | fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); |
609 | if (!fsn_mark) | |
52202dfb | 610 | return -ENOMEM; |
88826276 | 611 | |
912ee394 AG |
612 | fsnotify_init_mark(fsn_mark, fanotify_free_mark); |
613 | ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0); | |
fa218ab9 LS |
614 | if (ret) |
615 | goto err; | |
88826276 | 616 | } |
b9e4e3bd | 617 | added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); |
fa218ab9 | 618 | |
c63181e6 | 619 | if (added & ~real_mount(mnt)->mnt_fsnotify_mask) |
43709a28 | 620 | fsnotify_recalc_vfsmount_mask(mnt); |
fa218ab9 LS |
621 | err: |
622 | fsnotify_put_mark(fsn_mark); | |
623 | return ret; | |
88826276 EP |
624 | } |
625 | ||
52202dfb | 626 | static int fanotify_add_inode_mark(struct fsnotify_group *group, |
b9e4e3bd EP |
627 | struct inode *inode, __u32 mask, |
628 | unsigned int flags) | |
88826276 EP |
629 | { |
630 | struct fsnotify_mark *fsn_mark; | |
912ee394 | 631 | __u32 added; |
fa218ab9 | 632 | int ret = 0; |
88826276 EP |
633 | |
634 | pr_debug("%s: group=%p inode=%p\n", __func__, group, inode); | |
2a3edf86 | 635 | |
5322a59f EP |
636 | /* |
637 | * If some other task has this inode open for write we should not add | |
638 | * an ignored mark, unless that ignored mark is supposed to survive | |
639 | * modification changes anyway. | |
640 | */ | |
641 | if ((flags & FAN_MARK_IGNORED_MASK) && | |
642 | !(flags & FAN_MARK_IGNORED_SURV_MODIFY) && | |
643 | (atomic_read(&inode->i_writecount) > 0)) | |
644 | return 0; | |
645 | ||
5444e298 | 646 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
2a3edf86 | 647 | if (!fsn_mark) { |
e7099d8a EP |
648 | if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) |
649 | return -ENOSPC; | |
650 | ||
912ee394 AG |
651 | fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); |
652 | if (!fsn_mark) | |
52202dfb | 653 | return -ENOMEM; |
2a3edf86 | 654 | |
912ee394 AG |
655 | fsnotify_init_mark(fsn_mark, fanotify_free_mark); |
656 | ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0); | |
fa218ab9 LS |
657 | if (ret) |
658 | goto err; | |
2a3edf86 | 659 | } |
b9e4e3bd | 660 | added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); |
fa218ab9 | 661 | |
43709a28 EP |
662 | if (added & ~inode->i_fsnotify_mask) |
663 | fsnotify_recalc_inode_mask(inode); | |
fa218ab9 LS |
664 | err: |
665 | fsnotify_put_mark(fsn_mark); | |
666 | return ret; | |
88826276 | 667 | } |
2a3edf86 | 668 | |
52c923dd | 669 | /* fanotify syscalls */ |
08ae8938 | 670 | SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags) |
11637e4b | 671 | { |
52c923dd EP |
672 | struct fsnotify_group *group; |
673 | int f_flags, fd; | |
4afeff85 | 674 | struct user_struct *user; |
52c923dd | 675 | |
08ae8938 EP |
676 | pr_debug("%s: flags=%d event_f_flags=%d\n", |
677 | __func__, flags, event_f_flags); | |
52c923dd | 678 | |
52c923dd | 679 | if (!capable(CAP_SYS_ADMIN)) |
a2f13ad0 | 680 | return -EPERM; |
52c923dd EP |
681 | |
682 | if (flags & ~FAN_ALL_INIT_FLAGS) | |
683 | return -EINVAL; | |
684 | ||
4afeff85 EP |
685 | user = get_current_user(); |
686 | if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) { | |
687 | free_uid(user); | |
688 | return -EMFILE; | |
689 | } | |
690 | ||
b2d87909 | 691 | f_flags = O_RDWR | FMODE_NONOTIFY; |
52c923dd EP |
692 | if (flags & FAN_CLOEXEC) |
693 | f_flags |= O_CLOEXEC; | |
694 | if (flags & FAN_NONBLOCK) | |
695 | f_flags |= O_NONBLOCK; | |
696 | ||
697 | /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */ | |
698 | group = fsnotify_alloc_group(&fanotify_fsnotify_ops); | |
26379198 EP |
699 | if (IS_ERR(group)) { |
700 | free_uid(user); | |
52c923dd | 701 | return PTR_ERR(group); |
26379198 | 702 | } |
52c923dd | 703 | |
4afeff85 EP |
704 | group->fanotify_data.user = user; |
705 | atomic_inc(&user->fanotify_listeners); | |
706 | ||
80af2588 | 707 | group->fanotify_data.f_flags = event_f_flags; |
9e66e423 EP |
708 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS |
709 | mutex_init(&group->fanotify_data.access_mutex); | |
710 | init_waitqueue_head(&group->fanotify_data.access_waitq); | |
711 | INIT_LIST_HEAD(&group->fanotify_data.access_list); | |
09e5f14e | 712 | atomic_set(&group->fanotify_data.bypass_perm, 0); |
9e66e423 | 713 | #endif |
4231a235 EP |
714 | switch (flags & FAN_ALL_CLASS_BITS) { |
715 | case FAN_CLASS_NOTIF: | |
716 | group->priority = FS_PRIO_0; | |
717 | break; | |
718 | case FAN_CLASS_CONTENT: | |
719 | group->priority = FS_PRIO_1; | |
720 | break; | |
721 | case FAN_CLASS_PRE_CONTENT: | |
722 | group->priority = FS_PRIO_2; | |
723 | break; | |
724 | default: | |
725 | fd = -EINVAL; | |
d8153d4d | 726 | goto out_destroy_group; |
4231a235 | 727 | } |
cb2d429f | 728 | |
5dd03f55 EP |
729 | if (flags & FAN_UNLIMITED_QUEUE) { |
730 | fd = -EPERM; | |
731 | if (!capable(CAP_SYS_ADMIN)) | |
d8153d4d | 732 | goto out_destroy_group; |
5dd03f55 EP |
733 | group->max_events = UINT_MAX; |
734 | } else { | |
735 | group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS; | |
736 | } | |
2529a0df | 737 | |
ac7e22dc EP |
738 | if (flags & FAN_UNLIMITED_MARKS) { |
739 | fd = -EPERM; | |
740 | if (!capable(CAP_SYS_ADMIN)) | |
d8153d4d | 741 | goto out_destroy_group; |
ac7e22dc EP |
742 | group->fanotify_data.max_marks = UINT_MAX; |
743 | } else { | |
744 | group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS; | |
745 | } | |
e7099d8a | 746 | |
52c923dd EP |
747 | fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags); |
748 | if (fd < 0) | |
d8153d4d | 749 | goto out_destroy_group; |
52c923dd EP |
750 | |
751 | return fd; | |
752 | ||
d8153d4d LS |
753 | out_destroy_group: |
754 | fsnotify_destroy_group(group); | |
52c923dd | 755 | return fd; |
11637e4b | 756 | } |
bbaa4168 | 757 | |
9bbfc964 HC |
758 | SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags, |
759 | __u64 mask, int dfd, | |
760 | const char __user * pathname) | |
bbaa4168 | 761 | { |
0ff21db9 EP |
762 | struct inode *inode = NULL; |
763 | struct vfsmount *mnt = NULL; | |
2a3edf86 | 764 | struct fsnotify_group *group; |
2903ff01 | 765 | struct fd f; |
2a3edf86 | 766 | struct path path; |
2903ff01 | 767 | int ret; |
2a3edf86 EP |
768 | |
769 | pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n", | |
770 | __func__, fanotify_fd, flags, dfd, pathname, mask); | |
771 | ||
772 | /* we only use the lower 32 bits as of right now. */ | |
773 | if (mask & ((__u64)0xffffffff << 32)) | |
774 | return -EINVAL; | |
775 | ||
88380fe6 AG |
776 | if (flags & ~FAN_ALL_MARK_FLAGS) |
777 | return -EINVAL; | |
4d92604c | 778 | switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { |
1734dee4 | 779 | case FAN_MARK_ADD: /* fallthrough */ |
88380fe6 | 780 | case FAN_MARK_REMOVE: |
1734dee4 LS |
781 | if (!mask) |
782 | return -EINVAL; | |
4d92604c | 783 | case FAN_MARK_FLUSH: |
88380fe6 AG |
784 | break; |
785 | default: | |
786 | return -EINVAL; | |
787 | } | |
8fcd6528 EP |
788 | |
789 | if (mask & FAN_ONDIR) { | |
790 | flags |= FAN_MARK_ONDIR; | |
791 | mask &= ~FAN_ONDIR; | |
792 | } | |
793 | ||
b2d87909 EP |
794 | #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS |
795 | if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD)) | |
796 | #else | |
88380fe6 | 797 | if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD)) |
b2d87909 | 798 | #endif |
2a3edf86 EP |
799 | return -EINVAL; |
800 | ||
2903ff01 AV |
801 | f = fdget(fanotify_fd); |
802 | if (unlikely(!f.file)) | |
2a3edf86 EP |
803 | return -EBADF; |
804 | ||
805 | /* verify that this is indeed an fanotify instance */ | |
806 | ret = -EINVAL; | |
2903ff01 | 807 | if (unlikely(f.file->f_op != &fanotify_fops)) |
2a3edf86 | 808 | goto fput_and_out; |
2903ff01 | 809 | group = f.file->private_data; |
4231a235 EP |
810 | |
811 | /* | |
812 | * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not | |
813 | * allowed to set permissions events. | |
814 | */ | |
815 | ret = -EINVAL; | |
816 | if (mask & FAN_ALL_PERM_EVENTS && | |
817 | group->priority == FS_PRIO_0) | |
818 | goto fput_and_out; | |
2a3edf86 EP |
819 | |
820 | ret = fanotify_find_path(dfd, pathname, &path, flags); | |
821 | if (ret) | |
822 | goto fput_and_out; | |
823 | ||
824 | /* inode held in place by reference to path; group by fget on fd */ | |
eac8e9e8 | 825 | if (!(flags & FAN_MARK_MOUNT)) |
0ff21db9 EP |
826 | inode = path.dentry->d_inode; |
827 | else | |
828 | mnt = path.mnt; | |
2a3edf86 EP |
829 | |
830 | /* create/update an inode mark */ | |
4d92604c | 831 | switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { |
c6223f46 | 832 | case FAN_MARK_ADD: |
eac8e9e8 | 833 | if (flags & FAN_MARK_MOUNT) |
b9e4e3bd | 834 | ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags); |
0ff21db9 | 835 | else |
b9e4e3bd | 836 | ret = fanotify_add_inode_mark(group, inode, mask, flags); |
c6223f46 AG |
837 | break; |
838 | case FAN_MARK_REMOVE: | |
f3640192 | 839 | if (flags & FAN_MARK_MOUNT) |
b9e4e3bd | 840 | ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags); |
f3640192 | 841 | else |
b9e4e3bd | 842 | ret = fanotify_remove_inode_mark(group, inode, mask, flags); |
c6223f46 | 843 | break; |
4d92604c EP |
844 | case FAN_MARK_FLUSH: |
845 | if (flags & FAN_MARK_MOUNT) | |
846 | fsnotify_clear_vfsmount_marks_by_group(group); | |
847 | else | |
848 | fsnotify_clear_inode_marks_by_group(group); | |
4d92604c | 849 | break; |
c6223f46 AG |
850 | default: |
851 | ret = -EINVAL; | |
852 | } | |
2a3edf86 EP |
853 | |
854 | path_put(&path); | |
855 | fput_and_out: | |
2903ff01 | 856 | fdput(f); |
2a3edf86 EP |
857 | return ret; |
858 | } | |
859 | ||
9bbfc964 HC |
860 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
861 | asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask, | |
862 | long dfd, long pathname) | |
863 | { | |
864 | return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags, | |
865 | mask, (int) dfd, | |
866 | (const char __user *) pathname); | |
867 | } | |
868 | SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark); | |
869 | #endif | |
870 | ||
2a3edf86 | 871 | /* |
ae0e47f0 | 872 | * fanotify_user_setup - Our initialization function. Note that we cannot return |
2a3edf86 EP |
873 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
874 | * must result in panic(). | |
875 | */ | |
876 | static int __init fanotify_user_setup(void) | |
877 | { | |
878 | fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC); | |
b2d87909 EP |
879 | fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event, |
880 | SLAB_PANIC); | |
2a3edf86 EP |
881 | |
882 | return 0; | |
bbaa4168 | 883 | } |
2a3edf86 | 884 | device_initcall(fanotify_user_setup); |