blob: ce1f5823e2c2e3a20dd96d1f2e72eecd972b787d [file] [log] [blame]
Amy Griffis2d9048e2006-06-01 13:10:59 -07001/*
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 *
Eric Paris63c882a2009-05-21 17:02:01 -040011 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
12 * inotify was largely rewriten to make use of the fsnotify infrastructure
13 *
Amy Griffis2d9048e2006-06-01 13:10:59 -070014 * 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
Amy Griffis2d9048e2006-06-01 13:10:59 -070025#include <linux/file.h>
Eric Paris63c882a2009-05-21 17:02:01 -040026#include <linux/fs.h> /* struct inode */
27#include <linux/fsnotify_backend.h>
28#include <linux/idr.h>
29#include <linux/init.h> /* module_init */
Amy Griffis2d9048e2006-06-01 13:10:59 -070030#include <linux/inotify.h>
Eric Paris63c882a2009-05-21 17:02:01 -040031#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 */
Amy Griffis2d9048e2006-06-01 13:10:59 -070038#include <linux/syscalls.h>
Eric Paris63c882a2009-05-21 17:02:01 -040039#include <linux/types.h>
40#include <linux/uaccess.h>
41#include <linux/poll.h>
42#include <linux/wait.h>
43
44#include "inotify.h"
Amy Griffis2d9048e2006-06-01 13:10:59 -070045
46#include <asm/ioctls.h>
47
Amy Griffis2d9048e2006-06-01 13:10:59 -070048static struct vfsmount *inotify_mnt __read_mostly;
49
Eric Paris63c882a2009-05-21 17:02:01 -040050/* this just sits here and wastes global memory. used to just pad userspace messages with zeros */
51static struct inotify_event nul_inotify_event;
52
Amy Griffis2d9048e2006-06-01 13:10:59 -070053/* these are configurable via /proc/sys/fs/inotify/ */
Harvey Harrison3c828e42008-02-14 19:31:21 -080054static int inotify_max_user_instances __read_mostly;
Harvey Harrison3c828e42008-02-14 19:31:21 -080055static int inotify_max_queued_events __read_mostly;
Eric Paris63c882a2009-05-21 17:02:01 -040056int inotify_max_user_watches __read_mostly;
57
58static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
59struct kmem_cache *event_priv_cachep __read_mostly;
Amy Griffis2d9048e2006-06-01 13:10:59 -070060
61/*
Eric Paris63c882a2009-05-21 17:02:01 -040062 * 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.
Amy Griffis2d9048e2006-06-01 13:10:59 -070064 */
Eric Paris63c882a2009-05-21 17:02:01 -040065static atomic_t inotify_grp_num;
Amy Griffis2d9048e2006-06-01 13:10:59 -070066
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
Eric Paris63c882a2009-05-21 17:02:01 -0400108static inline __u32 inotify_arg_to_mask(u32 arg)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700109{
Eric Paris63c882a2009-05-21 17:02:01 -0400110 __u32 mask;
111
112 /* everything should accept their own ignored and cares about children */
113 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
114
115 /* mask off the flags used to open the fd */
116 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
117
118 return mask;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700119}
120
Eric Paris63c882a2009-05-21 17:02:01 -0400121static inline u32 inotify_mask_to_arg(__u32 mask)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700122{
Eric Paris63c882a2009-05-21 17:02:01 -0400123 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
124 IN_Q_OVERFLOW);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700125}
126
Eric Paris63c882a2009-05-21 17:02:01 -0400127/* intofiy userspace file descriptor functions */
Amy Griffis2d9048e2006-06-01 13:10:59 -0700128static unsigned int inotify_poll(struct file *file, poll_table *wait)
129{
Eric Paris63c882a2009-05-21 17:02:01 -0400130 struct fsnotify_group *group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700131 int ret = 0;
132
Eric Paris63c882a2009-05-21 17:02:01 -0400133 poll_wait(file, &group->notification_waitq, wait);
134 mutex_lock(&group->notification_mutex);
135 if (!fsnotify_notify_queue_is_empty(group))
Amy Griffis2d9048e2006-06-01 13:10:59 -0700136 ret = POLLIN | POLLRDNORM;
Eric Paris63c882a2009-05-21 17:02:01 -0400137 mutex_unlock(&group->notification_mutex);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700138
139 return ret;
140}
141
Vegard Nossum3632dee2009-01-22 15:29:45 +0100142/*
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 *
Eric Paris63c882a2009-05-21 17:02:01 -0400147 * Called with the group->notification_mutex held.
Vegard Nossum3632dee2009-01-22 15:29:45 +0100148 */
Eric Paris63c882a2009-05-21 17:02:01 -0400149static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
150 size_t count)
Vegard Nossum3632dee2009-01-22 15:29:45 +0100151{
152 size_t event_size = sizeof(struct inotify_event);
Eric Paris63c882a2009-05-21 17:02:01 -0400153 struct fsnotify_event *event;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100154
Eric Paris63c882a2009-05-21 17:02:01 -0400155 if (fsnotify_notify_queue_is_empty(group))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100156 return NULL;
157
Eric Paris63c882a2009-05-21 17:02:01 -0400158 event = fsnotify_peek_notify_event(group);
159
160 event_size += roundup(event->name_len, event_size);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100161
162 if (event_size > count)
163 return ERR_PTR(-EINVAL);
164
Eric Paris63c882a2009-05-21 17:02:01 -0400165 /* 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;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100170}
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 */
Eric Paris63c882a2009-05-21 17:02:01 -0400178static ssize_t copy_event_to_user(struct fsnotify_group *group,
179 struct fsnotify_event *event,
Vegard Nossum3632dee2009-01-22 15:29:45 +0100180 char __user *buf)
181{
Eric Paris63c882a2009-05-21 17:02:01 -0400182 struct inotify_event inotify_event;
183 struct fsnotify_event_private_data *fsn_priv;
184 struct inotify_event_private_data *priv;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100185 size_t event_size = sizeof(struct inotify_event);
Eric Paris63c882a2009-05-21 17:02:01 -0400186 size_t name_len;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100187
Eric Paris63c882a2009-05-21 17:02:01 -0400188 /* 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;
208
209 /* send the main event */
210 if (copy_to_user(buf, &inotify_event, event_size))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100211 return -EFAULT;
212
Eric Paris63c882a2009-05-21 17:02:01 -0400213 buf += event_size;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100214
Eric Paris63c882a2009-05-21 17:02:01 -0400215 /*
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))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100224 return -EFAULT;
Eric Paris63c882a2009-05-21 17:02:01 -0400225 buf += event->name_len;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100226
Eric Paris63c882a2009-05-21 17:02:01 -0400227 /* 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;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100232 }
Eric Paris63c882a2009-05-21 17:02:01 -0400233
Vegard Nossum3632dee2009-01-22 15:29:45 +0100234 return event_size;
235}
236
Amy Griffis2d9048e2006-06-01 13:10:59 -0700237static ssize_t inotify_read(struct file *file, char __user *buf,
238 size_t count, loff_t *pos)
239{
Eric Paris63c882a2009-05-21 17:02:01 -0400240 struct fsnotify_group *group;
241 struct fsnotify_event *kevent;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700242 char __user *start;
243 int ret;
244 DEFINE_WAIT(wait);
245
246 start = buf;
Eric Paris63c882a2009-05-21 17:02:01 -0400247 group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700248
249 while (1) {
Eric Paris63c882a2009-05-21 17:02:01 -0400250 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700251
Eric Paris63c882a2009-05-21 17:02:01 -0400252 mutex_lock(&group->notification_mutex);
253 kevent = get_one_event(group, count);
254 mutex_unlock(&group->notification_mutex);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700255
Vegard Nossum3632dee2009-01-22 15:29:45 +0100256 if (kevent) {
257 ret = PTR_ERR(kevent);
258 if (IS_ERR(kevent))
259 break;
Eric Paris63c882a2009-05-21 17:02:01 -0400260 ret = copy_event_to_user(group, kevent, buf);
261 fsnotify_put_event(kevent);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100262 if (ret < 0)
263 break;
264 buf += ret;
265 count -= ret;
266 continue;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700267 }
268
Vegard Nossum3632dee2009-01-22 15:29:45 +0100269 ret = -EAGAIN;
270 if (file->f_flags & O_NONBLOCK)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700271 break;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100272 ret = -EINTR;
273 if (signal_pending(current))
274 break;
275
276 if (start != buf)
277 break;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700278
279 schedule();
280 }
281
Eric Paris63c882a2009-05-21 17:02:01 -0400282 finish_wait(&group->notification_waitq, &wait);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100283 if (start != buf && ret != -EFAULT)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700284 ret = buf - start;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700285 return ret;
286}
287
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800288static int inotify_fasync(int fd, struct file *file, int on)
289{
Eric Paris63c882a2009-05-21 17:02:01 -0400290 struct fsnotify_group *group = file->private_data;
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800291
Eric Paris63c882a2009-05-21 17:02:01 -0400292 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800293}
294
Amy Griffis2d9048e2006-06-01 13:10:59 -0700295static int inotify_release(struct inode *ignored, struct file *file)
296{
Eric Paris63c882a2009-05-21 17:02:01 -0400297 struct fsnotify_group *group = file->private_data;
Keith Packardbdae9972009-07-01 21:56:38 -0700298 struct user_struct *user = group->inotify_data.user;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700299
Eric Paris63c882a2009-05-21 17:02:01 -0400300 fsnotify_clear_marks_by_group(group);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700301
Eric Paris63c882a2009-05-21 17:02:01 -0400302 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
303 fsnotify_put_group(group);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700304
Keith Packardbdae9972009-07-01 21:56:38 -0700305 atomic_dec(&user->inotify_devs);
306
Amy Griffis2d9048e2006-06-01 13:10:59 -0700307 return 0;
308}
309
310static long inotify_ioctl(struct file *file, unsigned int cmd,
311 unsigned long arg)
312{
Eric Paris63c882a2009-05-21 17:02:01 -0400313 struct fsnotify_group *group;
314 struct fsnotify_event_holder *holder;
315 struct fsnotify_event *event;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700316 void __user *p;
317 int ret = -ENOTTY;
Eric Paris63c882a2009-05-21 17:02:01 -0400318 size_t send_len = 0;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700319
Eric Paris63c882a2009-05-21 17:02:01 -0400320 group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700321 p = (void __user *) arg;
322
323 switch (cmd) {
324 case FIONREAD:
Eric Paris63c882a2009-05-21 17:02:01 -0400325 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);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700334 break;
335 }
336
337 return ret;
338}
339
340static const struct file_operations inotify_fops = {
Eric Paris63c882a2009-05-21 17:02:01 -0400341 .poll = inotify_poll,
342 .read = inotify_read,
343 .fasync = inotify_fasync,
344 .release = inotify_release,
345 .unlocked_ioctl = inotify_ioctl,
Amy Griffis2d9048e2006-06-01 13:10:59 -0700346 .compat_ioctl = inotify_ioctl,
347};
348
Amy Griffis2d9048e2006-06-01 13:10:59 -0700349
Eric Paris63c882a2009-05-21 17:02:01 -0400350/*
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
Eric Parisdead5372009-08-24 16:03:35 -0400367/*
368 * Remove the mark from the idr (if present) and drop the reference
369 * on the mark because it was in the idr.
370 */
Eric Paris7e790dd2009-07-07 10:28:24 -0400371static void inotify_remove_from_idr(struct fsnotify_group *group,
372 struct inotify_inode_mark_entry *ientry)
373{
374 struct idr *idr;
Eric Parisdead5372009-08-24 16:03:35 -0400375 struct fsnotify_mark_entry *entry;
376 struct inotify_inode_mark_entry *found_ientry;
377 int wd;
Eric Paris7e790dd2009-07-07 10:28:24 -0400378
379 spin_lock(&group->inotify_data.idr_lock);
380 idr = &group->inotify_data.idr;
Eric Parisdead5372009-08-24 16:03:35 -0400381 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);
Eric Paris7e790dd2009-07-07 10:28:24 -0400404 ientry->wd = -1;
Eric Parisdead5372009-08-24 16:03:35 -0400405
406 /* removed from the idr, drop that ref */
407 fsnotify_put_mark(entry);
408out:
409 spin_unlock(&group->inotify_data.idr_lock);
Eric Paris7e790dd2009-07-07 10:28:24 -0400410}
Eric Parisdead5372009-08-24 16:03:35 -0400411
Eric Paris63c882a2009-05-21 17:02:01 -0400412/*
Eric Parisdead5372009-08-24 16:03:35 -0400413 * Send IN_IGNORED for this wd, remove this wd from the idr.
Eric Paris63c882a2009-05-21 17:02:01 -0400414 */
Eric Paris528da3e2009-06-12 16:04:26 -0400415void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
416 struct fsnotify_group *group)
Eric Paris63c882a2009-05-21 17:02:01 -0400417{
418 struct inotify_inode_mark_entry *ientry;
Eric Parisf44aebc2009-07-15 15:49:52 -0400419 struct fsnotify_event *ignored_event;
Eric Paris63c882a2009-05-21 17:02:01 -0400420 struct inotify_event_private_data *event_priv;
421 struct fsnotify_event_private_data *fsn_event_priv;
Eric Pariseef3a112009-08-16 21:51:44 -0400422 int ret;
Eric Paris63c882a2009-05-21 17:02:01 -0400423
Eric Parisf44aebc2009-07-15 15:49:52 -0400424 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
Eric Paris63c882a2009-05-21 17:02:01 -0400430 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
431
Eric Parisf44aebc2009-07-15 15:49:52 -0400432 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
Eric Paris63c882a2009-05-21 17:02:01 -0400433 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
Eric Pariseef3a112009-08-16 21:51:44 -0400441 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
442 if (ret)
Eric Paris63c882a2009-05-21 17:02:01 -0400443 inotify_free_event_priv(fsn_event_priv);
444
445skip_send_ignore:
446
Eric Parisf44aebc2009-07-15 15:49:52 -0400447 /* matches the reference taken when the event was created */
448 fsnotify_put_event(ignored_event);
449
Eric Paris63c882a2009-05-21 17:02:01 -0400450 /* remove this entry from the idr */
Eric Paris7e790dd2009-07-07 10:28:24 -0400451 inotify_remove_from_idr(group, ientry);
Eric Paris63c882a2009-05-21 17:02:01 -0400452
Eric Paris5549f7c2009-07-07 10:28:23 -0400453 atomic_dec(&group->inotify_data.user->inotify_watches);
Eric Paris63c882a2009-05-21 17:02:01 -0400454}
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
Eric Paris52cef752009-08-24 16:03:35 -0400464static int inotify_update_existing_watch(struct fsnotify_group *group,
465 struct inode *inode,
466 u32 arg)
Eric Paris63c882a2009-05-21 17:02:01 -0400467{
Eric Paris52cef752009-08-24 16:03:35 -0400468 struct fsnotify_mark_entry *entry;
Eric Paris63c882a2009-05-21 17:02:01 -0400469 struct inotify_inode_mark_entry *ientry;
Eric Paris63c882a2009-05-21 17:02:01 -0400470 __u32 old_mask, new_mask;
Eric Paris52cef752009-08-24 16:03:35 -0400471 __u32 mask;
472 int add = (arg & IN_MASK_ADD);
473 int ret;
Eric Paris63c882a2009-05-21 17:02:01 -0400474
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
Eric Paris63c882a2009-05-21 17:02:01 -0400480 spin_lock(&inode->i_lock);
481 entry = fsnotify_find_mark_entry(group, inode);
482 spin_unlock(&inode->i_lock);
Eric Paris52cef752009-08-24 16:03:35 -0400483 if (!entry)
484 return -ENOENT;
Eric Paris63c882a2009-05-21 17:02:01 -0400485
Eric Paris52cef752009-08-24 16:03:35 -0400486 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
Eric Paris75fe2b262009-07-07 10:28:23 -0400487
Eric Paris63c882a2009-05-21 17:02:01 -0400488 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
Eric Paris52cef752009-08-24 16:03:35 -0400518 /* return the wd */
519 ret = ientry->wd;
520
521 /* match the get from fsnotify_find_mark_entry() */
Eric Paris75fe2b262009-07-07 10:28:23 -0400522 fsnotify_put_mark(entry);
523
Eric Paris52cef752009-08-24 16:03:35 -0400524 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;
Eric Paris63c882a2009-05-21 17:02:01 -0400566 }
Eric Paris7e790dd2009-07-07 10:28:24 -0400567
Eric Parisdead5372009-08-24 16:03:35 -0400568 /* we put the mark on the idr, take a reference */
569 fsnotify_get_mark(&tmp_ientry->fsn_entry);
570
Eric Paris52cef752009-08-24 16:03:35 -0400571 /* 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
Eric Paris52cef752009-08-24 16:03:35 -0400579 /* 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
591out_err:
592 if (ret < 0)
593 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
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;
615
Eric Paris63c882a2009-05-21 17:02:01 -0400616 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);
Eric Paris08e53fc2009-08-16 21:51:55 -0400634 group->inotify_data.last_wd = 1;
Eric Paris63c882a2009-05-21 17:02:01 -0400635 group->inotify_data.user = user;
636 group->inotify_data.fa = NULL;
637
638 return group;
639}
640
641
642/* inotify syscalls */
Heiko Carstens938bb9f2009-01-14 14:14:30 +0100643SYSCALL_DEFINE1(inotify_init1, int, flags)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700644{
Eric Paris63c882a2009-05-21 17:02:01 -0400645 struct fsnotify_group *group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700646 struct user_struct *user;
647 struct file *filp;
648 int fd, ret;
649
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700650 /* Check the IN_* constants for consistency. */
651 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
652 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
653
Ulrich Drepper510df2d2008-07-23 21:29:41 -0700654 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
Ulrich Drepper40065532008-07-23 21:29:32 -0700655 return -EINVAL;
656
657 fd = get_unused_fd_flags(flags & O_CLOEXEC);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700658 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
David Howellsda9592e2008-11-14 10:39:05 +1100667 user = get_current_user();
Amy Griffis2d9048e2006-06-01 13:10:59 -0700668 if (unlikely(atomic_read(&user->inotify_devs) >=
669 inotify_max_user_instances)) {
670 ret = -EMFILE;
671 goto out_free_uid;
672 }
673
Eric Paris63c882a2009-05-21 17:02:01 -0400674 /* 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);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700678 goto out_free_uid;
679 }
680
Amy Griffis2d9048e2006-06-01 13:10:59 -0700681 filp->f_op = &inotify_fops;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800682 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;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700685 filp->f_mode = FMODE_READ;
Ulrich Drepper510df2d2008-07-23 21:29:41 -0700686 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Eric Paris63c882a2009-05-21 17:02:01 -0400687 filp->private_data = group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700688
Amy Griffis2d9048e2006-06-01 13:10:59 -0700689 atomic_inc(&user->inotify_devs);
Eric Paris63c882a2009-05-21 17:02:01 -0400690
Amy Griffis2d9048e2006-06-01 13:10:59 -0700691 fd_install(fd, filp);
692
693 return fd;
Eric Paris63c882a2009-05-21 17:02:01 -0400694
Amy Griffis2d9048e2006-06-01 13:10:59 -0700695out_free_uid:
696 free_uid(user);
697 put_filp(filp);
698out_put_fd:
699 put_unused_fd(fd);
700 return ret;
701}
702
Heiko Carstens938bb9f2009-01-14 14:14:30 +0100703SYSCALL_DEFINE0(inotify_init)
Ulrich Drepper40065532008-07-23 21:29:32 -0700704{
705 return sys_inotify_init1(0);
706}
707
Heiko Carstens2e4d0922009-01-14 14:14:31 +0100708SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
709 u32, mask)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700710{
Eric Paris63c882a2009-05-21 17:02:01 -0400711 struct fsnotify_group *group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700712 struct inode *inode;
Al Viro2d8f3032008-07-22 09:59:21 -0400713 struct path path;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700714 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
Eric Paris63c882a2009-05-21 17:02:01 -0400733 ret = inotify_find_inode(pathname, &path, flags);
734 if (ret)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700735 goto fput_and_out;
736
Eric Paris63c882a2009-05-21 17:02:01 -0400737 /* inode held in place by reference to path; group by fget on fd */
Al Viro2d8f3032008-07-22 09:59:21 -0400738 inode = path.dentry->d_inode;
Eric Paris63c882a2009-05-21 17:02:01 -0400739 group = filp->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700740
Eric Paris63c882a2009-05-21 17:02:01 -0400741 /* create/update an inode mark */
742 ret = inotify_update_watch(group, inode, mask);
743 if (unlikely(ret))
744 goto path_put_and_out;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700745
Eric Paris63c882a2009-05-21 17:02:01 -0400746path_put_and_out:
Al Viro2d8f3032008-07-22 09:59:21 -0400747 path_put(&path);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700748fput_and_out:
749 fput_light(filp, fput_needed);
750 return ret;
751}
752
Heiko Carstens2e4d0922009-01-14 14:14:31 +0100753SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700754{
Eric Paris63c882a2009-05-21 17:02:01 -0400755 struct fsnotify_group *group;
756 struct fsnotify_mark_entry *entry;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700757 struct file *filp;
Eric Paris63c882a2009-05-21 17:02:01 -0400758 int ret = 0, fput_needed;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700759
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
Eric Paris63c882a2009-05-21 17:02:01 -0400770 group = filp->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700771
Eric Paris63c882a2009-05-21 17:02:01 -0400772 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
Eric Paris528da3e2009-06-12 16:04:26 -0400782 fsnotify_destroy_mark_by_entry(entry);
Eric Paris63c882a2009-05-21 17:02:01 -0400783 fsnotify_put_mark(entry);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700784
785out:
786 fput_light(filp, fput_needed);
787 return ret;
788}
789
David Howells454e2392006-06-23 02:02:57 -0700790static int
Amy Griffis2d9048e2006-06-01 13:10:59 -0700791inotify_get_sb(struct file_system_type *fs_type, int flags,
David Howells454e2392006-06-23 02:02:57 -0700792 const char *dev_name, void *data, struct vfsmount *mnt)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700793{
Andrey Mirkinfd5eea42007-10-16 23:30:13 -0700794 return get_sb_pseudo(fs_type, "inotify", NULL,
795 INOTIFYFS_SUPER_MAGIC, mnt);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700796}
797
798static struct file_system_type inotify_fs_type = {
Eric Paris63c882a2009-05-21 17:02:01 -0400799 .name = "inotifyfs",
800 .get_sb = inotify_get_sb,
801 .kill_sb = kill_anon_super,
Amy Griffis2d9048e2006-06-01 13:10:59 -0700802};
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
Eric Paris63c882a2009-05-21 17:02:01 -0400821 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
822 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
Eric Paris63c882a2009-05-21 17:02:01 -0400823
Amy Griffis2d9048e2006-06-01 13:10:59 -0700824 inotify_max_queued_events = 16384;
825 inotify_max_user_instances = 128;
826 inotify_max_user_watches = 8192;
827
Amy Griffis2d9048e2006-06-01 13:10:59 -0700828 return 0;
829}
Amy Griffis2d9048e2006-06-01 13:10:59 -0700830module_init(inotify_user_setup);