Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later
2 : /*
3 : * fs/inotify_user.c - inotify support for userspace
4 : *
5 : * Authors:
6 : * John McCutchan <ttb@tentacle.dhs.org>
7 : * Robert Love <rml@novell.com>
8 : *
9 : * Copyright (C) 2005 John McCutchan
10 : * Copyright 2006 Hewlett-Packard Development Company, L.P.
11 : *
12 : * Copyright (C) 2009 Eric Paris <Red Hat Inc>
13 : * inotify was largely rewriten to make use of the fsnotify infrastructure
14 : */
15 :
16 : #include <linux/file.h>
17 : #include <linux/fs.h> /* struct inode */
18 : #include <linux/fsnotify_backend.h>
19 : #include <linux/idr.h>
20 : #include <linux/init.h> /* fs_initcall */
21 : #include <linux/inotify.h>
22 : #include <linux/kernel.h> /* roundup() */
23 : #include <linux/namei.h> /* LOOKUP_FOLLOW */
24 : #include <linux/sched/signal.h>
25 : #include <linux/slab.h> /* struct kmem_cache */
26 : #include <linux/syscalls.h>
27 : #include <linux/types.h>
28 : #include <linux/anon_inodes.h>
29 : #include <linux/uaccess.h>
30 : #include <linux/poll.h>
31 : #include <linux/wait.h>
32 : #include <linux/memcontrol.h>
33 : #include <linux/security.h>
34 :
35 : #include "inotify.h"
36 : #include "../fdinfo.h"
37 :
38 : #include <asm/ioctls.h>
39 :
40 : /*
41 : * An inotify watch requires allocating an inotify_inode_mark structure as
42 : * well as pinning the watched inode. Doubling the size of a VFS inode
43 : * should be more than enough to cover the additional filesystem inode
44 : * size increase.
45 : */
46 : #define INOTIFY_WATCH_COST (sizeof(struct inotify_inode_mark) + \
47 : 2 * sizeof(struct inode))
48 :
49 : /* configurable via /proc/sys/fs/inotify/ */
50 : static int inotify_max_queued_events __read_mostly;
51 :
52 : struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
53 :
54 : #ifdef CONFIG_SYSCTL
55 :
56 : #include <linux/sysctl.h>
57 :
58 : static long it_zero = 0;
59 : static long it_int_max = INT_MAX;
60 :
61 : static struct ctl_table inotify_table[] = {
62 : {
63 : .procname = "max_user_instances",
64 : .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
65 : .maxlen = sizeof(long),
66 : .mode = 0644,
67 : .proc_handler = proc_doulongvec_minmax,
68 : .extra1 = &it_zero,
69 : .extra2 = &it_int_max,
70 : },
71 : {
72 : .procname = "max_user_watches",
73 : .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
74 : .maxlen = sizeof(long),
75 : .mode = 0644,
76 : .proc_handler = proc_doulongvec_minmax,
77 : .extra1 = &it_zero,
78 : .extra2 = &it_int_max,
79 : },
80 : {
81 : .procname = "max_queued_events",
82 : .data = &inotify_max_queued_events,
83 : .maxlen = sizeof(int),
84 : .mode = 0644,
85 : .proc_handler = proc_dointvec_minmax,
86 : .extra1 = SYSCTL_ZERO
87 : },
88 : { }
89 : };
90 :
91 1 : static void __init inotify_sysctls_init(void)
92 : {
93 1 : register_sysctl("fs/inotify", inotify_table);
94 1 : }
95 :
96 : #else
97 : #define inotify_sysctls_init() do { } while (0)
98 : #endif /* CONFIG_SYSCTL */
99 :
100 : static inline __u32 inotify_arg_to_mask(struct inode *inode, u32 arg)
101 : {
102 : __u32 mask;
103 :
104 : /*
105 : * Everything should receive events when the inode is unmounted.
106 : * All directories care about children.
107 : */
108 0 : mask = (FS_UNMOUNT);
109 0 : if (S_ISDIR(inode->i_mode))
110 0 : mask |= FS_EVENT_ON_CHILD;
111 :
112 : /* mask off the flags used to open the fd */
113 0 : mask |= (arg & INOTIFY_USER_MASK);
114 :
115 : return mask;
116 : }
117 :
118 : #define INOTIFY_MARK_FLAGS \
119 : (FSNOTIFY_MARK_FLAG_EXCL_UNLINK | FSNOTIFY_MARK_FLAG_IN_ONESHOT)
120 :
121 : static inline unsigned int inotify_arg_to_flags(u32 arg)
122 : {
123 0 : unsigned int flags = 0;
124 :
125 0 : if (arg & IN_EXCL_UNLINK)
126 0 : flags |= FSNOTIFY_MARK_FLAG_EXCL_UNLINK;
127 0 : if (arg & IN_ONESHOT)
128 0 : flags |= FSNOTIFY_MARK_FLAG_IN_ONESHOT;
129 :
130 : return flags;
131 : }
132 :
133 : static inline u32 inotify_mask_to_arg(__u32 mask)
134 : {
135 0 : return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
136 : IN_Q_OVERFLOW);
137 : }
138 :
139 : /* inotify userspace file descriptor functions */
140 0 : static __poll_t inotify_poll(struct file *file, poll_table *wait)
141 : {
142 0 : struct fsnotify_group *group = file->private_data;
143 0 : __poll_t ret = 0;
144 :
145 0 : poll_wait(file, &group->notification_waitq, wait);
146 0 : spin_lock(&group->notification_lock);
147 0 : if (!fsnotify_notify_queue_is_empty(group))
148 0 : ret = EPOLLIN | EPOLLRDNORM;
149 0 : spin_unlock(&group->notification_lock);
150 :
151 0 : return ret;
152 : }
153 :
154 : static int round_event_name_len(struct fsnotify_event *fsn_event)
155 : {
156 : struct inotify_event_info *event;
157 :
158 0 : event = INOTIFY_E(fsn_event);
159 0 : if (!event->name_len)
160 : return 0;
161 0 : return roundup(event->name_len + 1, sizeof(struct inotify_event));
162 : }
163 :
164 : /*
165 : * Get an inotify_kernel_event if one exists and is small
166 : * enough to fit in "count". Return an error pointer if
167 : * not large enough.
168 : *
169 : * Called with the group->notification_lock held.
170 : */
171 0 : static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
172 : size_t count)
173 : {
174 0 : size_t event_size = sizeof(struct inotify_event);
175 : struct fsnotify_event *event;
176 :
177 0 : event = fsnotify_peek_first_event(group);
178 0 : if (!event)
179 : return NULL;
180 :
181 0 : pr_debug("%s: group=%p event=%p\n", __func__, group, event);
182 :
183 0 : event_size += round_event_name_len(event);
184 0 : if (event_size > count)
185 : return ERR_PTR(-EINVAL);
186 :
187 : /* held the notification_lock the whole time, so this is the
188 : * same event we peeked above */
189 0 : fsnotify_remove_first_event(group);
190 :
191 0 : return event;
192 : }
193 :
194 : /*
195 : * Copy an event to user space, returning how much we copied.
196 : *
197 : * We already checked that the event size is smaller than the
198 : * buffer we had in "get_one_event()" above.
199 : */
200 0 : static ssize_t copy_event_to_user(struct fsnotify_group *group,
201 : struct fsnotify_event *fsn_event,
202 : char __user *buf)
203 : {
204 : struct inotify_event inotify_event;
205 : struct inotify_event_info *event;
206 0 : size_t event_size = sizeof(struct inotify_event);
207 : size_t name_len;
208 : size_t pad_name_len;
209 :
210 0 : pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
211 :
212 0 : event = INOTIFY_E(fsn_event);
213 0 : name_len = event->name_len;
214 : /*
215 : * round up name length so it is a multiple of event_size
216 : * plus an extra byte for the terminating '\0'.
217 : */
218 0 : pad_name_len = round_event_name_len(fsn_event);
219 0 : inotify_event.len = pad_name_len;
220 0 : inotify_event.mask = inotify_mask_to_arg(event->mask);
221 0 : inotify_event.wd = event->wd;
222 0 : inotify_event.cookie = event->sync_cookie;
223 :
224 : /* send the main event */
225 0 : if (copy_to_user(buf, &inotify_event, event_size))
226 : return -EFAULT;
227 :
228 0 : buf += event_size;
229 :
230 : /*
231 : * fsnotify only stores the pathname, so here we have to send the pathname
232 : * and then pad that pathname out to a multiple of sizeof(inotify_event)
233 : * with zeros.
234 : */
235 0 : if (pad_name_len) {
236 : /* copy the path name */
237 0 : if (copy_to_user(buf, event->name, name_len))
238 : return -EFAULT;
239 0 : buf += name_len;
240 :
241 : /* fill userspace with 0's */
242 0 : if (clear_user(buf, pad_name_len - name_len))
243 : return -EFAULT;
244 0 : event_size += pad_name_len;
245 : }
246 :
247 0 : return event_size;
248 : }
249 :
250 0 : static ssize_t inotify_read(struct file *file, char __user *buf,
251 : size_t count, loff_t *pos)
252 : {
253 : struct fsnotify_group *group;
254 : struct fsnotify_event *kevent;
255 : char __user *start;
256 : int ret;
257 0 : DEFINE_WAIT_FUNC(wait, woken_wake_function);
258 :
259 0 : start = buf;
260 0 : group = file->private_data;
261 :
262 0 : add_wait_queue(&group->notification_waitq, &wait);
263 : while (1) {
264 0 : spin_lock(&group->notification_lock);
265 0 : kevent = get_one_event(group, count);
266 0 : spin_unlock(&group->notification_lock);
267 :
268 : pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
269 :
270 0 : if (kevent) {
271 0 : ret = PTR_ERR(kevent);
272 0 : if (IS_ERR(kevent))
273 : break;
274 0 : ret = copy_event_to_user(group, kevent, buf);
275 0 : fsnotify_destroy_event(group, kevent);
276 0 : if (ret < 0)
277 : break;
278 0 : buf += ret;
279 0 : count -= ret;
280 0 : continue;
281 : }
282 :
283 0 : ret = -EAGAIN;
284 0 : if (file->f_flags & O_NONBLOCK)
285 : break;
286 0 : ret = -ERESTARTSYS;
287 0 : if (signal_pending(current))
288 : break;
289 :
290 0 : if (start != buf)
291 : break;
292 :
293 0 : wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
294 : }
295 0 : remove_wait_queue(&group->notification_waitq, &wait);
296 :
297 0 : if (start != buf && ret != -EFAULT)
298 0 : ret = buf - start;
299 0 : return ret;
300 : }
301 :
302 0 : static int inotify_release(struct inode *ignored, struct file *file)
303 : {
304 0 : struct fsnotify_group *group = file->private_data;
305 :
306 : pr_debug("%s: group=%p\n", __func__, group);
307 :
308 : /* free this group, matching get was inotify_init->fsnotify_obtain_group */
309 0 : fsnotify_destroy_group(group);
310 :
311 0 : return 0;
312 : }
313 :
314 0 : static long inotify_ioctl(struct file *file, unsigned int cmd,
315 : unsigned long arg)
316 : {
317 : struct fsnotify_group *group;
318 : struct fsnotify_event *fsn_event;
319 : void __user *p;
320 0 : int ret = -ENOTTY;
321 0 : size_t send_len = 0;
322 :
323 0 : group = file->private_data;
324 0 : p = (void __user *) arg;
325 :
326 : pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
327 :
328 0 : switch (cmd) {
329 : case FIONREAD:
330 0 : spin_lock(&group->notification_lock);
331 0 : list_for_each_entry(fsn_event, &group->notification_list,
332 : list) {
333 0 : send_len += sizeof(struct inotify_event);
334 0 : send_len += round_event_name_len(fsn_event);
335 : }
336 0 : spin_unlock(&group->notification_lock);
337 0 : ret = put_user(send_len, (int __user *) p);
338 : break;
339 : #ifdef CONFIG_CHECKPOINT_RESTORE
340 : case INOTIFY_IOC_SETNEXTWD:
341 : ret = -EINVAL;
342 : if (arg >= 1 && arg <= INT_MAX) {
343 : struct inotify_group_private_data *data;
344 :
345 : data = &group->inotify_data;
346 : spin_lock(&data->idr_lock);
347 : idr_set_cursor(&data->idr, (unsigned int)arg);
348 : spin_unlock(&data->idr_lock);
349 : ret = 0;
350 : }
351 : break;
352 : #endif /* CONFIG_CHECKPOINT_RESTORE */
353 : }
354 :
355 0 : return ret;
356 : }
357 :
358 : static const struct file_operations inotify_fops = {
359 : .show_fdinfo = inotify_show_fdinfo,
360 : .poll = inotify_poll,
361 : .read = inotify_read,
362 : .fasync = fsnotify_fasync,
363 : .release = inotify_release,
364 : .unlocked_ioctl = inotify_ioctl,
365 : .compat_ioctl = inotify_ioctl,
366 : .llseek = noop_llseek,
367 : };
368 :
369 :
370 : /*
371 : * find_inode - resolve a user-given path to a specific inode
372 : */
373 0 : static int inotify_find_inode(const char __user *dirname, struct path *path,
374 : unsigned int flags, __u64 mask)
375 : {
376 : int error;
377 :
378 0 : error = user_path_at(AT_FDCWD, dirname, flags, path);
379 0 : if (error)
380 : return error;
381 : /* you can only watch an inode if you have read permissions on it */
382 0 : error = path_permission(path, MAY_READ);
383 0 : if (error) {
384 0 : path_put(path);
385 : return error;
386 : }
387 : error = security_path_notify(path, mask,
388 : FSNOTIFY_OBJ_TYPE_INODE);
389 : if (error)
390 : path_put(path);
391 :
392 : return error;
393 : }
394 :
395 0 : static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
396 : struct inotify_inode_mark *i_mark)
397 : {
398 : int ret;
399 :
400 0 : idr_preload(GFP_KERNEL);
401 0 : spin_lock(idr_lock);
402 :
403 0 : ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
404 0 : if (ret >= 0) {
405 : /* we added the mark to the idr, take a reference */
406 0 : i_mark->wd = ret;
407 0 : fsnotify_get_mark(&i_mark->fsn_mark);
408 : }
409 :
410 0 : spin_unlock(idr_lock);
411 : idr_preload_end();
412 0 : return ret < 0 ? ret : 0;
413 : }
414 :
415 0 : static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
416 : int wd)
417 : {
418 0 : struct idr *idr = &group->inotify_data.idr;
419 0 : spinlock_t *idr_lock = &group->inotify_data.idr_lock;
420 : struct inotify_inode_mark *i_mark;
421 :
422 : assert_spin_locked(idr_lock);
423 :
424 0 : i_mark = idr_find(idr, wd);
425 0 : if (i_mark) {
426 0 : struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
427 :
428 0 : fsnotify_get_mark(fsn_mark);
429 : /* One ref for being in the idr, one ref we just took */
430 0 : BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
431 : }
432 :
433 0 : return i_mark;
434 : }
435 :
436 : static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
437 : int wd)
438 : {
439 : struct inotify_inode_mark *i_mark;
440 0 : spinlock_t *idr_lock = &group->inotify_data.idr_lock;
441 :
442 0 : spin_lock(idr_lock);
443 0 : i_mark = inotify_idr_find_locked(group, wd);
444 0 : spin_unlock(idr_lock);
445 :
446 : return i_mark;
447 : }
448 :
449 : /*
450 : * Remove the mark from the idr (if present) and drop the reference
451 : * on the mark because it was in the idr.
452 : */
453 0 : static void inotify_remove_from_idr(struct fsnotify_group *group,
454 : struct inotify_inode_mark *i_mark)
455 : {
456 0 : struct idr *idr = &group->inotify_data.idr;
457 0 : spinlock_t *idr_lock = &group->inotify_data.idr_lock;
458 0 : struct inotify_inode_mark *found_i_mark = NULL;
459 : int wd;
460 :
461 0 : spin_lock(idr_lock);
462 0 : wd = i_mark->wd;
463 :
464 : /*
465 : * does this i_mark think it is in the idr? we shouldn't get called
466 : * if it wasn't....
467 : */
468 0 : if (wd == -1) {
469 0 : WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
470 : __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
471 : goto out;
472 : }
473 :
474 : /* Lets look in the idr to see if we find it */
475 0 : found_i_mark = inotify_idr_find_locked(group, wd);
476 0 : if (unlikely(!found_i_mark)) {
477 0 : WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
478 : __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
479 : goto out;
480 : }
481 :
482 : /*
483 : * We found an mark in the idr at the right wd, but it's
484 : * not the mark we were told to remove. eparis seriously
485 : * fucked up somewhere.
486 : */
487 0 : if (unlikely(found_i_mark != i_mark)) {
488 0 : WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
489 : "found_i_mark=%p found_i_mark->wd=%d "
490 : "found_i_mark->group=%p\n", __func__, i_mark,
491 : i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
492 : found_i_mark->wd, found_i_mark->fsn_mark.group);
493 : goto out;
494 : }
495 :
496 : /*
497 : * One ref for being in the idr
498 : * one ref grabbed by inotify_idr_find
499 : */
500 0 : if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
501 0 : printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
502 : __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
503 : /* we can't really recover with bad ref cnting.. */
504 0 : BUG();
505 : }
506 :
507 0 : idr_remove(idr, wd);
508 : /* Removed from the idr, drop that ref. */
509 0 : fsnotify_put_mark(&i_mark->fsn_mark);
510 : out:
511 0 : i_mark->wd = -1;
512 0 : spin_unlock(idr_lock);
513 : /* match the ref taken by inotify_idr_find_locked() */
514 0 : if (found_i_mark)
515 0 : fsnotify_put_mark(&found_i_mark->fsn_mark);
516 0 : }
517 :
518 : /*
519 : * Send IN_IGNORED for this wd, remove this wd from the idr.
520 : */
521 0 : void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
522 : struct fsnotify_group *group)
523 : {
524 : struct inotify_inode_mark *i_mark;
525 :
526 : /* Queue ignore event for the watch */
527 0 : inotify_handle_inode_event(fsn_mark, FS_IN_IGNORED, NULL, NULL, NULL,
528 : 0);
529 :
530 0 : i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
531 : /* remove this mark from the idr */
532 0 : inotify_remove_from_idr(group, i_mark);
533 :
534 0 : dec_inotify_watches(group->inotify_data.ucounts);
535 0 : }
536 :
537 0 : static int inotify_update_existing_watch(struct fsnotify_group *group,
538 : struct inode *inode,
539 : u32 arg)
540 : {
541 : struct fsnotify_mark *fsn_mark;
542 : struct inotify_inode_mark *i_mark;
543 : __u32 old_mask, new_mask;
544 0 : int replace = !(arg & IN_MASK_ADD);
545 0 : int create = (arg & IN_MASK_CREATE);
546 : int ret;
547 :
548 0 : fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
549 0 : if (!fsn_mark)
550 : return -ENOENT;
551 0 : else if (create) {
552 : ret = -EEXIST;
553 : goto out;
554 : }
555 :
556 0 : i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
557 :
558 0 : spin_lock(&fsn_mark->lock);
559 0 : old_mask = fsn_mark->mask;
560 0 : if (replace) {
561 0 : fsn_mark->mask = 0;
562 0 : fsn_mark->flags &= ~INOTIFY_MARK_FLAGS;
563 : }
564 0 : fsn_mark->mask |= inotify_arg_to_mask(inode, arg);
565 0 : fsn_mark->flags |= inotify_arg_to_flags(arg);
566 0 : new_mask = fsn_mark->mask;
567 0 : spin_unlock(&fsn_mark->lock);
568 :
569 0 : if (old_mask != new_mask) {
570 : /* more bits in old than in new? */
571 0 : int dropped = (old_mask & ~new_mask);
572 : /* more bits in this fsn_mark than the inode's mask? */
573 0 : int do_inode = (new_mask & ~inode->i_fsnotify_mask);
574 :
575 : /* update the inode with this new fsn_mark */
576 0 : if (dropped || do_inode)
577 0 : fsnotify_recalc_mask(inode->i_fsnotify_marks);
578 :
579 : }
580 :
581 : /* return the wd */
582 0 : ret = i_mark->wd;
583 :
584 : out:
585 : /* match the get from fsnotify_find_mark() */
586 0 : fsnotify_put_mark(fsn_mark);
587 :
588 0 : return ret;
589 : }
590 :
591 0 : static int inotify_new_watch(struct fsnotify_group *group,
592 : struct inode *inode,
593 : u32 arg)
594 : {
595 : struct inotify_inode_mark *tmp_i_mark;
596 : int ret;
597 0 : struct idr *idr = &group->inotify_data.idr;
598 0 : spinlock_t *idr_lock = &group->inotify_data.idr_lock;
599 :
600 0 : tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
601 0 : if (unlikely(!tmp_i_mark))
602 : return -ENOMEM;
603 :
604 0 : fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
605 0 : tmp_i_mark->fsn_mark.mask = inotify_arg_to_mask(inode, arg);
606 0 : tmp_i_mark->fsn_mark.flags = inotify_arg_to_flags(arg);
607 0 : tmp_i_mark->wd = -1;
608 :
609 0 : ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
610 0 : if (ret)
611 : goto out_err;
612 :
613 : /* increment the number of watches the user has */
614 0 : if (!inc_inotify_watches(group->inotify_data.ucounts)) {
615 0 : inotify_remove_from_idr(group, tmp_i_mark);
616 0 : ret = -ENOSPC;
617 0 : goto out_err;
618 : }
619 :
620 : /* we are on the idr, now get on the inode */
621 0 : ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
622 0 : if (ret) {
623 : /* we failed to get on the inode, get off the idr */
624 0 : inotify_remove_from_idr(group, tmp_i_mark);
625 0 : goto out_err;
626 : }
627 :
628 :
629 : /* return the watch descriptor for this new mark */
630 0 : ret = tmp_i_mark->wd;
631 :
632 : out_err:
633 : /* match the ref from fsnotify_init_mark() */
634 0 : fsnotify_put_mark(&tmp_i_mark->fsn_mark);
635 :
636 0 : return ret;
637 : }
638 :
639 0 : static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
640 : {
641 0 : int ret = 0;
642 :
643 0 : fsnotify_group_lock(group);
644 : /* try to update and existing watch with the new arg */
645 0 : ret = inotify_update_existing_watch(group, inode, arg);
646 : /* no mark present, try to add a new one */
647 0 : if (ret == -ENOENT)
648 0 : ret = inotify_new_watch(group, inode, arg);
649 0 : fsnotify_group_unlock(group);
650 :
651 0 : return ret;
652 : }
653 :
654 0 : static struct fsnotify_group *inotify_new_group(unsigned int max_events)
655 : {
656 : struct fsnotify_group *group;
657 : struct inotify_event_info *oevent;
658 :
659 0 : group = fsnotify_alloc_group(&inotify_fsnotify_ops,
660 : FSNOTIFY_GROUP_USER);
661 0 : if (IS_ERR(group))
662 : return group;
663 :
664 0 : oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL_ACCOUNT);
665 0 : if (unlikely(!oevent)) {
666 0 : fsnotify_destroy_group(group);
667 0 : return ERR_PTR(-ENOMEM);
668 : }
669 0 : group->overflow_event = &oevent->fse;
670 0 : fsnotify_init_event(group->overflow_event);
671 0 : oevent->mask = FS_Q_OVERFLOW;
672 0 : oevent->wd = -1;
673 0 : oevent->sync_cookie = 0;
674 0 : oevent->name_len = 0;
675 :
676 0 : group->max_events = max_events;
677 0 : group->memcg = get_mem_cgroup_from_mm(current->mm);
678 :
679 0 : spin_lock_init(&group->inotify_data.idr_lock);
680 0 : idr_init(&group->inotify_data.idr);
681 0 : group->inotify_data.ucounts = inc_ucount(current_user_ns(),
682 0 : current_euid(),
683 : UCOUNT_INOTIFY_INSTANCES);
684 :
685 0 : if (!group->inotify_data.ucounts) {
686 0 : fsnotify_destroy_group(group);
687 0 : return ERR_PTR(-EMFILE);
688 : }
689 :
690 : return group;
691 : }
692 :
693 :
694 : /* inotify syscalls */
695 0 : static int do_inotify_init(int flags)
696 : {
697 : struct fsnotify_group *group;
698 : int ret;
699 :
700 : /* Check the IN_* constants for consistency. */
701 : BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
702 : BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
703 :
704 0 : if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
705 : return -EINVAL;
706 :
707 : /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
708 0 : group = inotify_new_group(inotify_max_queued_events);
709 0 : if (IS_ERR(group))
710 0 : return PTR_ERR(group);
711 :
712 0 : ret = anon_inode_getfd("inotify", &inotify_fops, group,
713 : O_RDONLY | flags);
714 0 : if (ret < 0)
715 0 : fsnotify_destroy_group(group);
716 :
717 : return ret;
718 : }
719 :
720 0 : SYSCALL_DEFINE1(inotify_init1, int, flags)
721 : {
722 0 : return do_inotify_init(flags);
723 : }
724 :
725 0 : SYSCALL_DEFINE0(inotify_init)
726 : {
727 0 : return do_inotify_init(0);
728 : }
729 :
730 0 : SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
731 : u32, mask)
732 : {
733 : struct fsnotify_group *group;
734 : struct inode *inode;
735 : struct path path;
736 : struct fd f;
737 : int ret;
738 0 : unsigned flags = 0;
739 :
740 : /*
741 : * We share a lot of code with fs/dnotify. We also share
742 : * the bit layout between inotify's IN_* and the fsnotify
743 : * FS_*. This check ensures that only the inotify IN_*
744 : * bits get passed in and set in watches/events.
745 : */
746 0 : if (unlikely(mask & ~ALL_INOTIFY_BITS))
747 : return -EINVAL;
748 : /*
749 : * Require at least one valid bit set in the mask.
750 : * Without _something_ set, we would have no events to
751 : * watch for.
752 : */
753 0 : if (unlikely(!(mask & ALL_INOTIFY_BITS)))
754 : return -EINVAL;
755 :
756 0 : f = fdget(fd);
757 0 : if (unlikely(!f.file))
758 : return -EBADF;
759 :
760 : /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
761 0 : if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
762 : ret = -EINVAL;
763 : goto fput_and_out;
764 : }
765 :
766 : /* verify that this is indeed an inotify instance */
767 0 : if (unlikely(f.file->f_op != &inotify_fops)) {
768 : ret = -EINVAL;
769 : goto fput_and_out;
770 : }
771 :
772 0 : if (!(mask & IN_DONT_FOLLOW))
773 0 : flags |= LOOKUP_FOLLOW;
774 0 : if (mask & IN_ONLYDIR)
775 0 : flags |= LOOKUP_DIRECTORY;
776 :
777 0 : ret = inotify_find_inode(pathname, &path, flags,
778 : (mask & IN_ALL_EVENTS));
779 0 : if (ret)
780 : goto fput_and_out;
781 :
782 : /* inode held in place by reference to path; group by fget on fd */
783 0 : inode = path.dentry->d_inode;
784 0 : group = f.file->private_data;
785 :
786 : /* create/update an inode mark */
787 0 : ret = inotify_update_watch(group, inode, mask);
788 0 : path_put(&path);
789 : fput_and_out:
790 0 : fdput(f);
791 0 : return ret;
792 : }
793 :
794 0 : SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
795 : {
796 : struct fsnotify_group *group;
797 : struct inotify_inode_mark *i_mark;
798 : struct fd f;
799 0 : int ret = -EINVAL;
800 :
801 0 : f = fdget(fd);
802 0 : if (unlikely(!f.file))
803 : return -EBADF;
804 :
805 : /* verify that this is indeed an inotify instance */
806 0 : if (unlikely(f.file->f_op != &inotify_fops))
807 : goto out;
808 :
809 0 : group = f.file->private_data;
810 :
811 0 : i_mark = inotify_idr_find(group, wd);
812 0 : if (unlikely(!i_mark))
813 : goto out;
814 :
815 0 : ret = 0;
816 :
817 0 : fsnotify_destroy_mark(&i_mark->fsn_mark, group);
818 :
819 : /* match ref taken by inotify_idr_find */
820 0 : fsnotify_put_mark(&i_mark->fsn_mark);
821 :
822 : out:
823 0 : fdput(f);
824 0 : return ret;
825 : }
826 :
827 : /*
828 : * inotify_user_setup - Our initialization function. Note that we cannot return
829 : * error because we have compiled-in VFS hooks. So an (unlikely) failure here
830 : * must result in panic().
831 : */
832 1 : static int __init inotify_user_setup(void)
833 : {
834 : unsigned long watches_max;
835 : struct sysinfo si;
836 :
837 1 : si_meminfo(&si);
838 : /*
839 : * Allow up to 1% of addressable memory to be allocated for inotify
840 : * watches (per user) limited to the range [8192, 1048576].
841 : */
842 1 : watches_max = (((si.totalram - si.totalhigh) / 100) << PAGE_SHIFT) /
843 : INOTIFY_WATCH_COST;
844 1 : watches_max = clamp(watches_max, 8192UL, 1048576UL);
845 :
846 : BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
847 : BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
848 : BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
849 : BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
850 : BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
851 : BUILD_BUG_ON(IN_OPEN != FS_OPEN);
852 : BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
853 : BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
854 : BUILD_BUG_ON(IN_CREATE != FS_CREATE);
855 : BUILD_BUG_ON(IN_DELETE != FS_DELETE);
856 : BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
857 : BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
858 : BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
859 : BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
860 : BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
861 : BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
862 :
863 : BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
864 :
865 1 : inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
866 : SLAB_PANIC|SLAB_ACCOUNT);
867 :
868 1 : inotify_max_queued_events = 16384;
869 1 : init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
870 1 : init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = watches_max;
871 1 : inotify_sysctls_init();
872 :
873 1 : return 0;
874 : }
875 : fs_initcall(inotify_user_setup);
|
