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