]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/notify/inotify/inotify_user.c
inotify: fix locking around inotify watching in the idr
[net-next-2.6.git] / fs / notify / inotify / inotify_user.c
CommitLineData
2d9048e2
AG
1/*
2 * fs/inotify_user.c - inotify support for userspace
3 *
4 * Authors:
5 * John McCutchan <ttb@tentacle.dhs.org>
6 * Robert Love <rml@novell.com>
7 *
8 * Copyright (C) 2005 John McCutchan
9 * Copyright 2006 Hewlett-Packard Development Company, L.P.
10 *
63c882a0
EP
11 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
12 * inotify was largely rewriten to make use of the fsnotify infrastructure
13 *
2d9048e2
AG
14 * This program is free software; you can redistribute it and/or modify it
15 * under the terms of the GNU General Public License as published by the
16 * Free Software Foundation; either version 2, or (at your option) any
17 * later version.
18 *
19 * This program is distributed in the hope that it will be useful, but
20 * WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
23 */
24
2d9048e2 25#include <linux/file.h>
63c882a0
EP
26#include <linux/fs.h> /* struct inode */
27#include <linux/fsnotify_backend.h>
28#include <linux/idr.h>
29#include <linux/init.h> /* module_init */
2d9048e2 30#include <linux/inotify.h>
63c882a0
EP
31#include <linux/kernel.h> /* roundup() */
32#include <linux/magic.h> /* superblock magic number */
33#include <linux/mount.h> /* mntget */
34#include <linux/namei.h> /* LOOKUP_FOLLOW */
35#include <linux/path.h> /* struct path */
36#include <linux/sched.h> /* struct user */
37#include <linux/slab.h> /* struct kmem_cache */
2d9048e2 38#include <linux/syscalls.h>
63c882a0
EP
39#include <linux/types.h>
40#include <linux/uaccess.h>
41#include <linux/poll.h>
42#include <linux/wait.h>
2d9048e2 43
63c882a0 44#include "inotify.h"
2d9048e2 45
63c882a0 46#include <asm/ioctls.h>
2d9048e2
AG
47
48static struct vfsmount *inotify_mnt __read_mostly;
49
63c882a0
EP
50/* this just sits here and wastes global memory. used to just pad userspace messages with zeros */
51static struct inotify_event nul_inotify_event;
52
2d9048e2 53/* these are configurable via /proc/sys/fs/inotify/ */
3c828e49 54static int inotify_max_user_instances __read_mostly;
3c828e49 55static int inotify_max_queued_events __read_mostly;
63c882a0 56int inotify_max_user_watches __read_mostly;
2d9048e2 57
63c882a0
EP
58static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
59struct kmem_cache *event_priv_cachep __read_mostly;
2d9048e2
AG
60
61/*
63c882a0
EP
62 * When inotify registers a new group it increments this and uses that
63 * value as an offset to set the fsnotify group "name" and priority.
2d9048e2 64 */
63c882a0 65static atomic_t inotify_grp_num;
2d9048e2
AG
66
67#ifdef CONFIG_SYSCTL
68
69#include <linux/sysctl.h>
70
71static int zero;
72
73ctl_table inotify_table[] = {
74 {
75 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
76 .procname = "max_user_instances",
77 .data = &inotify_max_user_instances,
78 .maxlen = sizeof(int),
79 .mode = 0644,
80 .proc_handler = &proc_dointvec_minmax,
81 .strategy = &sysctl_intvec,
82 .extra1 = &zero,
83 },
84 {
85 .ctl_name = INOTIFY_MAX_USER_WATCHES,
86 .procname = "max_user_watches",
87 .data = &inotify_max_user_watches,
88 .maxlen = sizeof(int),
89 .mode = 0644,
90 .proc_handler = &proc_dointvec_minmax,
91 .strategy = &sysctl_intvec,
92 .extra1 = &zero,
93 },
94 {
95 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
96 .procname = "max_queued_events",
97 .data = &inotify_max_queued_events,
98 .maxlen = sizeof(int),
99 .mode = 0644,
100 .proc_handler = &proc_dointvec_minmax,
101 .strategy = &sysctl_intvec,
102 .extra1 = &zero
103 },
104 { .ctl_name = 0 }
105};
106#endif /* CONFIG_SYSCTL */
107
63c882a0 108static inline __u32 inotify_arg_to_mask(u32 arg)
1c17d18e 109{
63c882a0 110 __u32 mask;
1c17d18e 111
63c882a0
EP
112 /* everything should accept their own ignored and cares about children */
113 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
2d9048e2 114
63c882a0
EP
115 /* mask off the flags used to open the fd */
116 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
2d9048e2 117
63c882a0 118 return mask;
2d9048e2
AG
119}
120
63c882a0 121static inline u32 inotify_mask_to_arg(__u32 mask)
2d9048e2 122{
63c882a0
EP
123 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
124 IN_Q_OVERFLOW);
2d9048e2
AG
125}
126
63c882a0 127/* intofiy userspace file descriptor functions */
2d9048e2
AG
128static unsigned int inotify_poll(struct file *file, poll_table *wait)
129{
63c882a0 130 struct fsnotify_group *group = file->private_data;
2d9048e2
AG
131 int ret = 0;
132
63c882a0
EP
133 poll_wait(file, &group->notification_waitq, wait);
134 mutex_lock(&group->notification_mutex);
135 if (!fsnotify_notify_queue_is_empty(group))
2d9048e2 136 ret = POLLIN | POLLRDNORM;
63c882a0 137 mutex_unlock(&group->notification_mutex);
2d9048e2
AG
138
139 return ret;
140}
141
3632dee2
VN
142/*
143 * Get an inotify_kernel_event if one exists and is small
144 * enough to fit in "count". Return an error pointer if
145 * not large enough.
146 *
63c882a0 147 * Called with the group->notification_mutex held.
3632dee2 148 */
63c882a0
EP
149static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
150 size_t count)
3632dee2
VN
151{
152 size_t event_size = sizeof(struct inotify_event);
63c882a0 153 struct fsnotify_event *event;
3632dee2 154
63c882a0 155 if (fsnotify_notify_queue_is_empty(group))
3632dee2
VN
156 return NULL;
157
63c882a0
EP
158 event = fsnotify_peek_notify_event(group);
159
160 event_size += roundup(event->name_len, event_size);
3632dee2
VN
161
162 if (event_size > count)
163 return ERR_PTR(-EINVAL);
164
63c882a0
EP
165 /* held the notification_mutex the whole time, so this is the
166 * same event we peeked above */
167 fsnotify_remove_notify_event(group);
168
169 return event;
3632dee2
VN
170}
171
172/*
173 * Copy an event to user space, returning how much we copied.
174 *
175 * We already checked that the event size is smaller than the
176 * buffer we had in "get_one_event()" above.
177 */
63c882a0
EP
178static ssize_t copy_event_to_user(struct fsnotify_group *group,
179 struct fsnotify_event *event,
3632dee2
VN
180 char __user *buf)
181{
63c882a0
EP
182 struct inotify_event inotify_event;
183 struct fsnotify_event_private_data *fsn_priv;
184 struct inotify_event_private_data *priv;
3632dee2 185 size_t event_size = sizeof(struct inotify_event);
63c882a0
EP
186 size_t name_len;
187
188 /* we get the inotify watch descriptor from the event private data */
189 spin_lock(&event->lock);
190 fsn_priv = fsnotify_remove_priv_from_event(group, event);
191 spin_unlock(&event->lock);
192
193 if (!fsn_priv)
194 inotify_event.wd = -1;
195 else {
196 priv = container_of(fsn_priv, struct inotify_event_private_data,
197 fsnotify_event_priv_data);
198 inotify_event.wd = priv->wd;
199 inotify_free_event_priv(fsn_priv);
200 }
201
202 /* round up event->name_len so it is a multiple of event_size */
203 name_len = roundup(event->name_len, event_size);
204 inotify_event.len = name_len;
205
206 inotify_event.mask = inotify_mask_to_arg(event->mask);
207 inotify_event.cookie = event->sync_cookie;
3632dee2 208
63c882a0
EP
209 /* send the main event */
210 if (copy_to_user(buf, &inotify_event, event_size))
3632dee2
VN
211 return -EFAULT;
212
63c882a0 213 buf += event_size;
3632dee2 214
63c882a0
EP
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. I get my zeros from the nul_inotify_event.
219 */
220 if (name_len) {
221 unsigned int len_to_zero = name_len - event->name_len;
222 /* copy the path name */
223 if (copy_to_user(buf, event->file_name, event->name_len))
3632dee2 224 return -EFAULT;
63c882a0 225 buf += event->name_len;
3632dee2 226
63c882a0
EP
227 /* fill userspace with 0's from nul_inotify_event */
228 if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
229 return -EFAULT;
230 buf += len_to_zero;
231 event_size += name_len;
3632dee2 232 }
63c882a0 233
3632dee2
VN
234 return event_size;
235}
236
2d9048e2
AG
237static ssize_t inotify_read(struct file *file, char __user *buf,
238 size_t count, loff_t *pos)
239{
63c882a0
EP
240 struct fsnotify_group *group;
241 struct fsnotify_event *kevent;
2d9048e2
AG
242 char __user *start;
243 int ret;
244 DEFINE_WAIT(wait);
245
246 start = buf;
63c882a0 247 group = file->private_data;
2d9048e2
AG
248
249 while (1) {
63c882a0 250 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
2d9048e2 251
63c882a0
EP
252 mutex_lock(&group->notification_mutex);
253 kevent = get_one_event(group, count);
254 mutex_unlock(&group->notification_mutex);
2d9048e2 255
3632dee2
VN
256 if (kevent) {
257 ret = PTR_ERR(kevent);
258 if (IS_ERR(kevent))
259 break;
63c882a0
EP
260 ret = copy_event_to_user(group, kevent, buf);
261 fsnotify_put_event(kevent);
3632dee2
VN
262 if (ret < 0)
263 break;
264 buf += ret;
265 count -= ret;
266 continue;
2d9048e2
AG
267 }
268
3632dee2
VN
269 ret = -EAGAIN;
270 if (file->f_flags & O_NONBLOCK)
2d9048e2 271 break;
3632dee2
VN
272 ret = -EINTR;
273 if (signal_pending(current))
2d9048e2 274 break;
16dbc6c9 275
3632dee2 276 if (start != buf)
2d9048e2 277 break;
16dbc6c9 278
3632dee2 279 schedule();
2d9048e2 280 }
2d9048e2 281
63c882a0 282 finish_wait(&group->notification_waitq, &wait);
3632dee2
VN
283 if (start != buf && ret != -EFAULT)
284 ret = buf - start;
2d9048e2
AG
285 return ret;
286}
287
bcfbf84d
DA
288static int inotify_fasync(int fd, struct file *file, int on)
289{
63c882a0 290 struct fsnotify_group *group = file->private_data;
bcfbf84d 291
63c882a0 292 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
bcfbf84d
DA
293}
294
2d9048e2
AG
295static int inotify_release(struct inode *ignored, struct file *file)
296{
63c882a0 297 struct fsnotify_group *group = file->private_data;
bdae997f 298 struct user_struct *user = group->inotify_data.user;
2d9048e2 299
63c882a0 300 fsnotify_clear_marks_by_group(group);
2d9048e2 301
63c882a0
EP
302 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
303 fsnotify_put_group(group);
2d9048e2 304
bdae997f
KP
305 atomic_dec(&user->inotify_devs);
306
2d9048e2
AG
307 return 0;
308}
309
310static long inotify_ioctl(struct file *file, unsigned int cmd,
311 unsigned long arg)
312{
63c882a0
EP
313 struct fsnotify_group *group;
314 struct fsnotify_event_holder *holder;
315 struct fsnotify_event *event;
2d9048e2
AG
316 void __user *p;
317 int ret = -ENOTTY;
63c882a0 318 size_t send_len = 0;
2d9048e2 319
63c882a0 320 group = file->private_data;
2d9048e2
AG
321 p = (void __user *) arg;
322
323 switch (cmd) {
324 case FIONREAD:
63c882a0
EP
325 mutex_lock(&group->notification_mutex);
326 list_for_each_entry(holder, &group->notification_list, event_list) {
327 event = holder->event;
328 send_len += sizeof(struct inotify_event);
329 send_len += roundup(event->name_len,
330 sizeof(struct inotify_event));
331 }
332 mutex_unlock(&group->notification_mutex);
333 ret = put_user(send_len, (int __user *) p);
2d9048e2
AG
334 break;
335 }
336
337 return ret;
338}
339
340static const struct file_operations inotify_fops = {
63c882a0
EP
341 .poll = inotify_poll,
342 .read = inotify_read,
343 .fasync = inotify_fasync,
344 .release = inotify_release,
345 .unlocked_ioctl = inotify_ioctl,
2d9048e2
AG
346 .compat_ioctl = inotify_ioctl,
347};
348
2d9048e2 349
63c882a0
EP
350/*
351 * find_inode - resolve a user-given path to a specific inode
352 */
353static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
354{
355 int error;
356
357 error = user_path_at(AT_FDCWD, dirname, flags, path);
358 if (error)
359 return error;
360 /* you can only watch an inode if you have read permissions on it */
361 error = inode_permission(path->dentry->d_inode, MAY_READ);
362 if (error)
363 path_put(path);
364 return error;
365}
366
dead537d
EP
367/*
368 * Remove the mark from the idr (if present) and drop the reference
369 * on the mark because it was in the idr.
370 */
7e790dd5
EP
371static void inotify_remove_from_idr(struct fsnotify_group *group,
372 struct inotify_inode_mark_entry *ientry)
373{
374 struct idr *idr;
dead537d
EP
375 struct fsnotify_mark_entry *entry;
376 struct inotify_inode_mark_entry *found_ientry;
377 int wd;
7e790dd5
EP
378
379 spin_lock(&group->inotify_data.idr_lock);
380 idr = &group->inotify_data.idr;
dead537d
EP
381 wd = ientry->wd;
382
383 if (wd == -1)
384 goto out;
385
386 entry = idr_find(&group->inotify_data.idr, wd);
387 if (unlikely(!entry))
388 goto out;
389
390 found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
391 if (unlikely(found_ientry != ientry)) {
392 /* We found an entry in the idr with the right wd, but it's
393 * not the entry we were told to remove. eparis seriously
394 * fucked up somewhere. */
395 WARN_ON(1);
396 ientry->wd = -1;
397 goto out;
398 }
399
400 /* One ref for being in the idr, one ref held by the caller */
401 BUG_ON(atomic_read(&entry->refcnt) < 2);
402
403 idr_remove(idr, wd);
7e790dd5 404 ientry->wd = -1;
dead537d
EP
405
406 /* removed from the idr, drop that ref */
407 fsnotify_put_mark(entry);
408out:
409 spin_unlock(&group->inotify_data.idr_lock);
7e790dd5 410}
dead537d 411
63c882a0 412/*
dead537d 413 * Send IN_IGNORED for this wd, remove this wd from the idr.
63c882a0 414 */
528da3e9
EP
415void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
416 struct fsnotify_group *group)
63c882a0
EP
417{
418 struct inotify_inode_mark_entry *ientry;
f44aebcc 419 struct fsnotify_event *ignored_event;
63c882a0
EP
420 struct inotify_event_private_data *event_priv;
421 struct fsnotify_event_private_data *fsn_event_priv;
eef3a116 422 int ret;
63c882a0 423
f44aebcc
EP
424 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
425 FSNOTIFY_EVENT_NONE, NULL, 0,
426 GFP_NOFS);
427 if (!ignored_event)
428 return;
429
63c882a0
EP
430 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
431
f44aebcc 432 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
63c882a0
EP
433 if (unlikely(!event_priv))
434 goto skip_send_ignore;
435
436 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
437
438 fsn_event_priv->group = group;
439 event_priv->wd = ientry->wd;
440
eef3a116
EP
441 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
442 if (ret)
63c882a0
EP
443 inotify_free_event_priv(fsn_event_priv);
444
445skip_send_ignore:
446
f44aebcc
EP
447 /* matches the reference taken when the event was created */
448 fsnotify_put_event(ignored_event);
449
63c882a0 450 /* remove this entry from the idr */
7e790dd5 451 inotify_remove_from_idr(group, ientry);
63c882a0 452
5549f7cd 453 atomic_dec(&group->inotify_data.user->inotify_watches);
63c882a0
EP
454}
455
456/* ding dong the mark is dead */
457static void inotify_free_mark(struct fsnotify_mark_entry *entry)
458{
459 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
460
461 kmem_cache_free(inotify_inode_mark_cachep, ientry);
462}
463
52cef755
EP
464static int inotify_update_existing_watch(struct fsnotify_group *group,
465 struct inode *inode,
466 u32 arg)
63c882a0 467{
52cef755 468 struct fsnotify_mark_entry *entry;
63c882a0 469 struct inotify_inode_mark_entry *ientry;
63c882a0 470 __u32 old_mask, new_mask;
52cef755
EP
471 __u32 mask;
472 int add = (arg & IN_MASK_ADD);
473 int ret;
63c882a0
EP
474
475 /* don't allow invalid bits: we don't want flags set */
476 mask = inotify_arg_to_mask(arg);
477 if (unlikely(!mask))
478 return -EINVAL;
479
63c882a0
EP
480 spin_lock(&inode->i_lock);
481 entry = fsnotify_find_mark_entry(group, inode);
482 spin_unlock(&inode->i_lock);
52cef755
EP
483 if (!entry)
484 return -ENOENT;
7e790dd5 485
52cef755 486 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
75fe2b26 487
63c882a0
EP
488 spin_lock(&entry->lock);
489
490 old_mask = entry->mask;
491 if (add) {
492 entry->mask |= mask;
493 new_mask = entry->mask;
494 } else {
495 entry->mask = mask;
496 new_mask = entry->mask;
497 }
498
499 spin_unlock(&entry->lock);
500
501 if (old_mask != new_mask) {
502 /* more bits in old than in new? */
503 int dropped = (old_mask & ~new_mask);
504 /* more bits in this entry than the inode's mask? */
505 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
506 /* more bits in this entry than the group? */
507 int do_group = (new_mask & ~group->mask);
508
509 /* update the inode with this new entry */
510 if (dropped || do_inode)
511 fsnotify_recalc_inode_mask(inode);
512
513 /* update the group mask with the new mask */
514 if (dropped || do_group)
515 fsnotify_recalc_group_mask(group);
516 }
517
52cef755
EP
518 /* return the wd */
519 ret = ientry->wd;
520
521 /* match the get from fsnotify_find_mark_entry() */
75fe2b26
EP
522 fsnotify_put_mark(entry);
523
52cef755
EP
524 return ret;
525}
526
527static int inotify_new_watch(struct fsnotify_group *group,
528 struct inode *inode,
529 u32 arg)
530{
531 struct inotify_inode_mark_entry *tmp_ientry;
532 __u32 mask;
533 int ret;
534
535 /* don't allow invalid bits: we don't want flags set */
536 mask = inotify_arg_to_mask(arg);
537 if (unlikely(!mask))
538 return -EINVAL;
539
540 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
541 if (unlikely(!tmp_ientry))
542 return -ENOMEM;
543
544 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
545 tmp_ientry->fsn_entry.mask = mask;
546 tmp_ientry->wd = -1;
547
548 ret = -ENOSPC;
549 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
550 goto out_err;
551retry:
552 ret = -ENOMEM;
553 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
554 goto out_err;
555
556 spin_lock(&group->inotify_data.idr_lock);
557 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
558 group->inotify_data.last_wd,
559 &tmp_ientry->wd);
560 spin_unlock(&group->inotify_data.idr_lock);
561 if (ret) {
562 /* idr was out of memory allocate and try again */
563 if (ret == -EAGAIN)
564 goto retry;
565 goto out_err;
566 }
567
dead537d
EP
568 /* we put the mark on the idr, take a reference */
569 fsnotify_get_mark(&tmp_ientry->fsn_entry);
570
52cef755
EP
571 /* we are on the idr, now get on the inode */
572 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
573 if (ret) {
574 /* we failed to get on the inode, get off the idr */
575 inotify_remove_from_idr(group, tmp_ientry);
576 goto out_err;
577 }
578
52cef755
EP
579 /* update the idr hint, who cares about races, it's just a hint */
580 group->inotify_data.last_wd = tmp_ientry->wd;
581
582 /* increment the number of watches the user has */
583 atomic_inc(&group->inotify_data.user->inotify_watches);
584
585 /* return the watch descriptor for this new entry */
586 ret = tmp_ientry->wd;
587
588 /* match the ref from fsnotify_init_markentry() */
589 fsnotify_put_mark(&tmp_ientry->fsn_entry);
590
63c882a0 591out_err:
52cef755 592 if (ret < 0)
7e790dd5 593 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
52cef755
EP
594
595 return ret;
596}
597
598static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
599{
600 int ret = 0;
601
602retry:
603 /* try to update and existing watch with the new arg */
604 ret = inotify_update_existing_watch(group, inode, arg);
605 /* no mark present, try to add a new one */
606 if (ret == -ENOENT)
607 ret = inotify_new_watch(group, inode, arg);
608 /*
609 * inotify_new_watch could race with another thread which did an
610 * inotify_new_watch between the update_existing and the add watch
611 * here, go back and try to update an existing mark again.
612 */
613 if (ret == -EEXIST)
614 goto retry;
7e790dd5 615
63c882a0
EP
616 return ret;
617}
618
619static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
620{
621 struct fsnotify_group *group;
622 unsigned int grp_num;
623
624 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
625 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
626 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
627 if (IS_ERR(group))
628 return group;
629
630 group->max_events = max_events;
631
632 spin_lock_init(&group->inotify_data.idr_lock);
633 idr_init(&group->inotify_data.idr);
08e53fcb 634 group->inotify_data.last_wd = 1;
63c882a0
EP
635 group->inotify_data.user = user;
636 group->inotify_data.fa = NULL;
637
638 return group;
639}
640
641
642/* inotify syscalls */
938bb9f5 643SYSCALL_DEFINE1(inotify_init1, int, flags)
2d9048e2 644{
63c882a0 645 struct fsnotify_group *group;
2d9048e2
AG
646 struct user_struct *user;
647 struct file *filp;
648 int fd, ret;
649
e38b36f3
UD
650 /* Check the IN_* constants for consistency. */
651 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
652 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
653
510df2dd 654 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
4006553b
UD
655 return -EINVAL;
656
657 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2d9048e2
AG
658 if (fd < 0)
659 return fd;
660
661 filp = get_empty_filp();
662 if (!filp) {
663 ret = -ENFILE;
664 goto out_put_fd;
665 }
666
da9592ed 667 user = get_current_user();
2d9048e2
AG
668 if (unlikely(atomic_read(&user->inotify_devs) >=
669 inotify_max_user_instances)) {
670 ret = -EMFILE;
671 goto out_free_uid;
672 }
673
63c882a0
EP
674 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
675 group = inotify_new_group(user, inotify_max_queued_events);
676 if (IS_ERR(group)) {
677 ret = PTR_ERR(group);
2d9048e2
AG
678 goto out_free_uid;
679 }
680
2d9048e2 681 filp->f_op = &inotify_fops;
0f7fc9e4
JJS
682 filp->f_path.mnt = mntget(inotify_mnt);
683 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
684 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
2d9048e2 685 filp->f_mode = FMODE_READ;
510df2dd 686 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
63c882a0
EP
687 filp->private_data = group;
688
2d9048e2 689 atomic_inc(&user->inotify_devs);
63c882a0 690
2d9048e2
AG
691 fd_install(fd, filp);
692
693 return fd;
63c882a0 694
2d9048e2
AG
695out_free_uid:
696 free_uid(user);
697 put_filp(filp);
698out_put_fd:
699 put_unused_fd(fd);
700 return ret;
701}
702
938bb9f5 703SYSCALL_DEFINE0(inotify_init)
4006553b
UD
704{
705 return sys_inotify_init1(0);
706}
707
2e4d0924
HC
708SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
709 u32, mask)
2d9048e2 710{
63c882a0 711 struct fsnotify_group *group;
2d9048e2 712 struct inode *inode;
2d8f3038 713 struct path path;
2d9048e2
AG
714 struct file *filp;
715 int ret, fput_needed;
716 unsigned flags = 0;
717
718 filp = fget_light(fd, &fput_needed);
719 if (unlikely(!filp))
720 return -EBADF;
721
722 /* verify that this is indeed an inotify instance */
723 if (unlikely(filp->f_op != &inotify_fops)) {
724 ret = -EINVAL;
725 goto fput_and_out;
726 }
727
728 if (!(mask & IN_DONT_FOLLOW))
729 flags |= LOOKUP_FOLLOW;
730 if (mask & IN_ONLYDIR)
731 flags |= LOOKUP_DIRECTORY;
732
63c882a0
EP
733 ret = inotify_find_inode(pathname, &path, flags);
734 if (ret)
2d9048e2
AG
735 goto fput_and_out;
736
63c882a0 737 /* inode held in place by reference to path; group by fget on fd */
2d8f3038 738 inode = path.dentry->d_inode;
63c882a0 739 group = filp->private_data;
2d9048e2 740
63c882a0
EP
741 /* create/update an inode mark */
742 ret = inotify_update_watch(group, inode, mask);
743 if (unlikely(ret))
744 goto path_put_and_out;
2d9048e2 745
63c882a0 746path_put_and_out:
2d8f3038 747 path_put(&path);
2d9048e2
AG
748fput_and_out:
749 fput_light(filp, fput_needed);
750 return ret;
751}
752
2e4d0924 753SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
2d9048e2 754{
63c882a0
EP
755 struct fsnotify_group *group;
756 struct fsnotify_mark_entry *entry;
2d9048e2 757 struct file *filp;
63c882a0 758 int ret = 0, fput_needed;
2d9048e2
AG
759
760 filp = fget_light(fd, &fput_needed);
761 if (unlikely(!filp))
762 return -EBADF;
763
764 /* verify that this is indeed an inotify instance */
765 if (unlikely(filp->f_op != &inotify_fops)) {
766 ret = -EINVAL;
767 goto out;
768 }
769
63c882a0 770 group = filp->private_data;
2d9048e2 771
63c882a0
EP
772 spin_lock(&group->inotify_data.idr_lock);
773 entry = idr_find(&group->inotify_data.idr, wd);
774 if (unlikely(!entry)) {
775 spin_unlock(&group->inotify_data.idr_lock);
776 ret = -EINVAL;
777 goto out;
778 }
779 fsnotify_get_mark(entry);
780 spin_unlock(&group->inotify_data.idr_lock);
781
528da3e9 782 fsnotify_destroy_mark_by_entry(entry);
63c882a0 783 fsnotify_put_mark(entry);
2d9048e2
AG
784
785out:
786 fput_light(filp, fput_needed);
787 return ret;
788}
789
454e2398 790static int
2d9048e2 791inotify_get_sb(struct file_system_type *fs_type, int flags,
454e2398 792 const char *dev_name, void *data, struct vfsmount *mnt)
2d9048e2 793{
fd5eea42
AM
794 return get_sb_pseudo(fs_type, "inotify", NULL,
795 INOTIFYFS_SUPER_MAGIC, mnt);
2d9048e2
AG
796}
797
798static struct file_system_type inotify_fs_type = {
63c882a0
EP
799 .name = "inotifyfs",
800 .get_sb = inotify_get_sb,
801 .kill_sb = kill_anon_super,
2d9048e2
AG
802};
803
804/*
805 * inotify_user_setup - Our initialization function. Note that we cannnot return
806 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
807 * must result in panic().
808 */
809static int __init inotify_user_setup(void)
810{
811 int ret;
812
813 ret = register_filesystem(&inotify_fs_type);
814 if (unlikely(ret))
815 panic("inotify: register_filesystem returned %d!\n", ret);
816
817 inotify_mnt = kern_mount(&inotify_fs_type);
818 if (IS_ERR(inotify_mnt))
819 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
820
63c882a0
EP
821 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
822 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
63c882a0 823
2d9048e2
AG
824 inotify_max_queued_events = 16384;
825 inotify_max_user_instances = 128;
826 inotify_max_user_watches = 8192;
827
2d9048e2
AG
828 return 0;
829}
2d9048e2 830module_init(inotify_user_setup);