blob: b547ae17b46124db884913bf94bb534d1048f18c [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
50/* these are configurable via /proc/sys/fs/inotify/ */
Harvey Harrison3c828e42008-02-14 19:31:21 -080051static int inotify_max_user_instances __read_mostly;
Harvey Harrison3c828e42008-02-14 19:31:21 -080052static int inotify_max_queued_events __read_mostly;
Eric Paris63c882a2009-05-21 17:02:01 -040053int inotify_max_user_watches __read_mostly;
54
55static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
56struct kmem_cache *event_priv_cachep __read_mostly;
Amy Griffis2d9048e2006-06-01 13:10:59 -070057
58/*
Eric Paris63c882a2009-05-21 17:02:01 -040059 * When inotify registers a new group it increments this and uses that
60 * value as an offset to set the fsnotify group "name" and priority.
Amy Griffis2d9048e2006-06-01 13:10:59 -070061 */
Eric Paris63c882a2009-05-21 17:02:01 -040062static atomic_t inotify_grp_num;
Amy Griffis2d9048e2006-06-01 13:10:59 -070063
64#ifdef CONFIG_SYSCTL
65
66#include <linux/sysctl.h>
67
68static int zero;
69
70ctl_table inotify_table[] = {
71 {
72 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
73 .procname = "max_user_instances",
74 .data = &inotify_max_user_instances,
75 .maxlen = sizeof(int),
76 .mode = 0644,
77 .proc_handler = &proc_dointvec_minmax,
78 .strategy = &sysctl_intvec,
79 .extra1 = &zero,
80 },
81 {
82 .ctl_name = INOTIFY_MAX_USER_WATCHES,
83 .procname = "max_user_watches",
84 .data = &inotify_max_user_watches,
85 .maxlen = sizeof(int),
86 .mode = 0644,
87 .proc_handler = &proc_dointvec_minmax,
88 .strategy = &sysctl_intvec,
89 .extra1 = &zero,
90 },
91 {
92 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
93 .procname = "max_queued_events",
94 .data = &inotify_max_queued_events,
95 .maxlen = sizeof(int),
96 .mode = 0644,
97 .proc_handler = &proc_dointvec_minmax,
98 .strategy = &sysctl_intvec,
99 .extra1 = &zero
100 },
101 { .ctl_name = 0 }
102};
103#endif /* CONFIG_SYSCTL */
104
Eric Paris63c882a2009-05-21 17:02:01 -0400105static inline __u32 inotify_arg_to_mask(u32 arg)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700106{
Eric Paris63c882a2009-05-21 17:02:01 -0400107 __u32 mask;
108
109 /* everything should accept their own ignored and cares about children */
110 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
111
112 /* mask off the flags used to open the fd */
113 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
114
115 return mask;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700116}
117
Eric Paris63c882a2009-05-21 17:02:01 -0400118static inline u32 inotify_mask_to_arg(__u32 mask)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700119{
Eric Paris63c882a2009-05-21 17:02:01 -0400120 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
121 IN_Q_OVERFLOW);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700122}
123
Eric Paris63c882a2009-05-21 17:02:01 -0400124/* intofiy userspace file descriptor functions */
Amy Griffis2d9048e2006-06-01 13:10:59 -0700125static unsigned int inotify_poll(struct file *file, poll_table *wait)
126{
Eric Paris63c882a2009-05-21 17:02:01 -0400127 struct fsnotify_group *group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700128 int ret = 0;
129
Eric Paris63c882a2009-05-21 17:02:01 -0400130 poll_wait(file, &group->notification_waitq, wait);
131 mutex_lock(&group->notification_mutex);
132 if (!fsnotify_notify_queue_is_empty(group))
Amy Griffis2d9048e2006-06-01 13:10:59 -0700133 ret = POLLIN | POLLRDNORM;
Eric Paris63c882a2009-05-21 17:02:01 -0400134 mutex_unlock(&group->notification_mutex);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700135
136 return ret;
137}
138
Vegard Nossum3632dee2009-01-22 15:29:45 +0100139/*
140 * Get an inotify_kernel_event if one exists and is small
141 * enough to fit in "count". Return an error pointer if
142 * not large enough.
143 *
Eric Paris63c882a2009-05-21 17:02:01 -0400144 * Called with the group->notification_mutex held.
Vegard Nossum3632dee2009-01-22 15:29:45 +0100145 */
Eric Paris63c882a2009-05-21 17:02:01 -0400146static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
147 size_t count)
Vegard Nossum3632dee2009-01-22 15:29:45 +0100148{
149 size_t event_size = sizeof(struct inotify_event);
Eric Paris63c882a2009-05-21 17:02:01 -0400150 struct fsnotify_event *event;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100151
Eric Paris63c882a2009-05-21 17:02:01 -0400152 if (fsnotify_notify_queue_is_empty(group))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100153 return NULL;
154
Eric Paris63c882a2009-05-21 17:02:01 -0400155 event = fsnotify_peek_notify_event(group);
156
157 event_size += roundup(event->name_len, event_size);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100158
159 if (event_size > count)
160 return ERR_PTR(-EINVAL);
161
Eric Paris63c882a2009-05-21 17:02:01 -0400162 /* held the notification_mutex the whole time, so this is the
163 * same event we peeked above */
164 fsnotify_remove_notify_event(group);
165
166 return event;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100167}
168
169/*
170 * Copy an event to user space, returning how much we copied.
171 *
172 * We already checked that the event size is smaller than the
173 * buffer we had in "get_one_event()" above.
174 */
Eric Paris63c882a2009-05-21 17:02:01 -0400175static ssize_t copy_event_to_user(struct fsnotify_group *group,
176 struct fsnotify_event *event,
Vegard Nossum3632dee2009-01-22 15:29:45 +0100177 char __user *buf)
178{
Eric Paris63c882a2009-05-21 17:02:01 -0400179 struct inotify_event inotify_event;
180 struct fsnotify_event_private_data *fsn_priv;
181 struct inotify_event_private_data *priv;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100182 size_t event_size = sizeof(struct inotify_event);
Brian Rogersb962e732009-08-28 10:00:05 -0400183 size_t name_len = 0;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100184
Eric Paris63c882a2009-05-21 17:02:01 -0400185 /* we get the inotify watch descriptor from the event private data */
186 spin_lock(&event->lock);
187 fsn_priv = fsnotify_remove_priv_from_event(group, event);
188 spin_unlock(&event->lock);
189
190 if (!fsn_priv)
191 inotify_event.wd = -1;
192 else {
193 priv = container_of(fsn_priv, struct inotify_event_private_data,
194 fsnotify_event_priv_data);
195 inotify_event.wd = priv->wd;
196 inotify_free_event_priv(fsn_priv);
197 }
198
Brian Rogersb962e732009-08-28 10:00:05 -0400199 /*
200 * round up event->name_len so it is a multiple of event_size
Eric W. Biederman0db501b2009-08-27 03:20:04 -0700201 * plus an extra byte for the terminating '\0'.
202 */
Brian Rogersb962e732009-08-28 10:00:05 -0400203 if (event->name_len)
204 name_len = roundup(event->name_len + 1, event_size);
Eric Paris63c882a2009-05-21 17:02:01 -0400205 inotify_event.len = name_len;
206
207 inotify_event.mask = inotify_mask_to_arg(event->mask);
208 inotify_event.cookie = event->sync_cookie;
209
210 /* send the main event */
211 if (copy_to_user(buf, &inotify_event, event_size))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100212 return -EFAULT;
213
Eric Paris63c882a2009-05-21 17:02:01 -0400214 buf += event_size;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100215
Eric Paris63c882a2009-05-21 17:02:01 -0400216 /*
217 * fsnotify only stores the pathname, so here we have to send the pathname
218 * and then pad that pathname out to a multiple of sizeof(inotify_event)
219 * with zeros. I get my zeros from the nul_inotify_event.
220 */
221 if (name_len) {
222 unsigned int len_to_zero = name_len - event->name_len;
223 /* copy the path name */
224 if (copy_to_user(buf, event->file_name, event->name_len))
Vegard Nossum3632dee2009-01-22 15:29:45 +0100225 return -EFAULT;
Eric Paris63c882a2009-05-21 17:02:01 -0400226 buf += event->name_len;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100227
Eric W. Biederman0db501b2009-08-27 03:20:04 -0700228 /* fill userspace with 0's */
229 if (clear_user(buf, len_to_zero))
Eric Paris63c882a2009-05-21 17:02:01 -0400230 return -EFAULT;
231 buf += len_to_zero;
232 event_size += name_len;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100233 }
Eric Paris63c882a2009-05-21 17:02:01 -0400234
Vegard Nossum3632dee2009-01-22 15:29:45 +0100235 return event_size;
236}
237
Amy Griffis2d9048e2006-06-01 13:10:59 -0700238static ssize_t inotify_read(struct file *file, char __user *buf,
239 size_t count, loff_t *pos)
240{
Eric Paris63c882a2009-05-21 17:02:01 -0400241 struct fsnotify_group *group;
242 struct fsnotify_event *kevent;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700243 char __user *start;
244 int ret;
245 DEFINE_WAIT(wait);
246
247 start = buf;
Eric Paris63c882a2009-05-21 17:02:01 -0400248 group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700249
250 while (1) {
Eric Paris63c882a2009-05-21 17:02:01 -0400251 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700252
Eric Paris63c882a2009-05-21 17:02:01 -0400253 mutex_lock(&group->notification_mutex);
254 kevent = get_one_event(group, count);
255 mutex_unlock(&group->notification_mutex);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700256
Vegard Nossum3632dee2009-01-22 15:29:45 +0100257 if (kevent) {
258 ret = PTR_ERR(kevent);
259 if (IS_ERR(kevent))
260 break;
Eric Paris63c882a2009-05-21 17:02:01 -0400261 ret = copy_event_to_user(group, kevent, buf);
262 fsnotify_put_event(kevent);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100263 if (ret < 0)
264 break;
265 buf += ret;
266 count -= ret;
267 continue;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700268 }
269
Vegard Nossum3632dee2009-01-22 15:29:45 +0100270 ret = -EAGAIN;
271 if (file->f_flags & O_NONBLOCK)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700272 break;
Vegard Nossum3632dee2009-01-22 15:29:45 +0100273 ret = -EINTR;
274 if (signal_pending(current))
275 break;
276
277 if (start != buf)
278 break;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700279
280 schedule();
281 }
282
Eric Paris63c882a2009-05-21 17:02:01 -0400283 finish_wait(&group->notification_waitq, &wait);
Vegard Nossum3632dee2009-01-22 15:29:45 +0100284 if (start != buf && ret != -EFAULT)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700285 ret = buf - start;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700286 return ret;
287}
288
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800289static int inotify_fasync(int fd, struct file *file, int on)
290{
Eric Paris63c882a2009-05-21 17:02:01 -0400291 struct fsnotify_group *group = file->private_data;
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800292
Eric Paris63c882a2009-05-21 17:02:01 -0400293 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
Dmitry Antipovbcfbf842008-02-06 01:36:19 -0800294}
295
Amy Griffis2d9048e2006-06-01 13:10:59 -0700296static int inotify_release(struct inode *ignored, struct file *file)
297{
Eric Paris63c882a2009-05-21 17:02:01 -0400298 struct fsnotify_group *group = file->private_data;
Keith Packardbdae9972009-07-01 21:56:38 -0700299 struct user_struct *user = group->inotify_data.user;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700300
Eric Paris63c882a2009-05-21 17:02:01 -0400301 fsnotify_clear_marks_by_group(group);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700302
Eric Paris63c882a2009-05-21 17:02:01 -0400303 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
304 fsnotify_put_group(group);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700305
Keith Packardbdae9972009-07-01 21:56:38 -0700306 atomic_dec(&user->inotify_devs);
307
Amy Griffis2d9048e2006-06-01 13:10:59 -0700308 return 0;
309}
310
311static long inotify_ioctl(struct file *file, unsigned int cmd,
312 unsigned long arg)
313{
Eric Paris63c882a2009-05-21 17:02:01 -0400314 struct fsnotify_group *group;
315 struct fsnotify_event_holder *holder;
316 struct fsnotify_event *event;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700317 void __user *p;
318 int ret = -ENOTTY;
Eric Paris63c882a2009-05-21 17:02:01 -0400319 size_t send_len = 0;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700320
Eric Paris63c882a2009-05-21 17:02:01 -0400321 group = file->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700322 p = (void __user *) arg;
323
324 switch (cmd) {
325 case FIONREAD:
Eric Paris63c882a2009-05-21 17:02:01 -0400326 mutex_lock(&group->notification_mutex);
327 list_for_each_entry(holder, &group->notification_list, event_list) {
328 event = holder->event;
329 send_len += sizeof(struct inotify_event);
330 send_len += roundup(event->name_len,
331 sizeof(struct inotify_event));
332 }
333 mutex_unlock(&group->notification_mutex);
334 ret = put_user(send_len, (int __user *) p);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700335 break;
336 }
337
338 return ret;
339}
340
341static const struct file_operations inotify_fops = {
Eric Paris63c882a2009-05-21 17:02:01 -0400342 .poll = inotify_poll,
343 .read = inotify_read,
344 .fasync = inotify_fasync,
345 .release = inotify_release,
346 .unlocked_ioctl = inotify_ioctl,
Amy Griffis2d9048e2006-06-01 13:10:59 -0700347 .compat_ioctl = inotify_ioctl,
348};
349
Amy Griffis2d9048e2006-06-01 13:10:59 -0700350
Eric Paris63c882a2009-05-21 17:02:01 -0400351/*
352 * find_inode - resolve a user-given path to a specific inode
353 */
354static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
355{
356 int error;
357
358 error = user_path_at(AT_FDCWD, dirname, flags, path);
359 if (error)
360 return error;
361 /* you can only watch an inode if you have read permissions on it */
362 error = inode_permission(path->dentry->d_inode, MAY_READ);
363 if (error)
364 path_put(path);
365 return error;
366}
367
Eric Parisdead5372009-08-24 16:03:35 -0400368/*
369 * Remove the mark from the idr (if present) and drop the reference
370 * on the mark because it was in the idr.
371 */
Eric Paris7e790dd2009-07-07 10:28:24 -0400372static void inotify_remove_from_idr(struct fsnotify_group *group,
373 struct inotify_inode_mark_entry *ientry)
374{
375 struct idr *idr;
Eric Parisdead5372009-08-24 16:03:35 -0400376 struct fsnotify_mark_entry *entry;
377 struct inotify_inode_mark_entry *found_ientry;
378 int wd;
Eric Paris7e790dd2009-07-07 10:28:24 -0400379
380 spin_lock(&group->inotify_data.idr_lock);
381 idr = &group->inotify_data.idr;
Eric Parisdead5372009-08-24 16:03:35 -0400382 wd = ientry->wd;
383
384 if (wd == -1)
385 goto out;
386
387 entry = idr_find(&group->inotify_data.idr, wd);
388 if (unlikely(!entry))
389 goto out;
390
391 found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
392 if (unlikely(found_ientry != ientry)) {
393 /* We found an entry in the idr with the right wd, but it's
394 * not the entry we were told to remove. eparis seriously
395 * fucked up somewhere. */
396 WARN_ON(1);
397 ientry->wd = -1;
398 goto out;
399 }
400
401 /* One ref for being in the idr, one ref held by the caller */
402 BUG_ON(atomic_read(&entry->refcnt) < 2);
403
404 idr_remove(idr, wd);
Eric Paris7e790dd2009-07-07 10:28:24 -0400405 ientry->wd = -1;
Eric Parisdead5372009-08-24 16:03:35 -0400406
407 /* removed from the idr, drop that ref */
408 fsnotify_put_mark(entry);
409out:
410 spin_unlock(&group->inotify_data.idr_lock);
Eric Paris7e790dd2009-07-07 10:28:24 -0400411}
Eric Parisdead5372009-08-24 16:03:35 -0400412
Eric Paris63c882a2009-05-21 17:02:01 -0400413/*
Eric Parisdead5372009-08-24 16:03:35 -0400414 * Send IN_IGNORED for this wd, remove this wd from the idr.
Eric Paris63c882a2009-05-21 17:02:01 -0400415 */
Eric Paris528da3e2009-06-12 16:04:26 -0400416void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
417 struct fsnotify_group *group)
Eric Paris63c882a2009-05-21 17:02:01 -0400418{
419 struct inotify_inode_mark_entry *ientry;
Eric Parisf44aebc2009-07-15 15:49:52 -0400420 struct fsnotify_event *ignored_event;
Eric Paris63c882a2009-05-21 17:02:01 -0400421 struct inotify_event_private_data *event_priv;
422 struct fsnotify_event_private_data *fsn_event_priv;
Eric Pariseef3a112009-08-16 21:51:44 -0400423 int ret;
Eric Paris63c882a2009-05-21 17:02:01 -0400424
Eric Parisf44aebc2009-07-15 15:49:52 -0400425 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
426 FSNOTIFY_EVENT_NONE, NULL, 0,
427 GFP_NOFS);
428 if (!ignored_event)
429 return;
430
Eric Paris63c882a2009-05-21 17:02:01 -0400431 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
432
Eric Parisf44aebc2009-07-15 15:49:52 -0400433 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
Eric Paris63c882a2009-05-21 17:02:01 -0400434 if (unlikely(!event_priv))
435 goto skip_send_ignore;
436
437 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
438
439 fsn_event_priv->group = group;
440 event_priv->wd = ientry->wd;
441
Eric Pariseef3a112009-08-16 21:51:44 -0400442 ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
443 if (ret)
Eric Paris63c882a2009-05-21 17:02:01 -0400444 inotify_free_event_priv(fsn_event_priv);
445
446skip_send_ignore:
447
Eric Parisf44aebc2009-07-15 15:49:52 -0400448 /* matches the reference taken when the event was created */
449 fsnotify_put_event(ignored_event);
450
Eric Paris63c882a2009-05-21 17:02:01 -0400451 /* remove this entry from the idr */
Eric Paris7e790dd2009-07-07 10:28:24 -0400452 inotify_remove_from_idr(group, ientry);
Eric Paris63c882a2009-05-21 17:02:01 -0400453
Eric Paris5549f7c2009-07-07 10:28:23 -0400454 atomic_dec(&group->inotify_data.user->inotify_watches);
Eric Paris63c882a2009-05-21 17:02:01 -0400455}
456
457/* ding dong the mark is dead */
458static void inotify_free_mark(struct fsnotify_mark_entry *entry)
459{
460 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
461
462 kmem_cache_free(inotify_inode_mark_cachep, ientry);
463}
464
Eric Paris52cef752009-08-24 16:03:35 -0400465static int inotify_update_existing_watch(struct fsnotify_group *group,
466 struct inode *inode,
467 u32 arg)
Eric Paris63c882a2009-05-21 17:02:01 -0400468{
Eric Paris52cef752009-08-24 16:03:35 -0400469 struct fsnotify_mark_entry *entry;
Eric Paris63c882a2009-05-21 17:02:01 -0400470 struct inotify_inode_mark_entry *ientry;
Eric Paris63c882a2009-05-21 17:02:01 -0400471 __u32 old_mask, new_mask;
Eric Paris52cef752009-08-24 16:03:35 -0400472 __u32 mask;
473 int add = (arg & IN_MASK_ADD);
474 int ret;
Eric Paris63c882a2009-05-21 17:02:01 -0400475
476 /* don't allow invalid bits: we don't want flags set */
477 mask = inotify_arg_to_mask(arg);
478 if (unlikely(!mask))
479 return -EINVAL;
480
Eric Paris63c882a2009-05-21 17:02:01 -0400481 spin_lock(&inode->i_lock);
482 entry = fsnotify_find_mark_entry(group, inode);
483 spin_unlock(&inode->i_lock);
Eric Paris52cef752009-08-24 16:03:35 -0400484 if (!entry)
485 return -ENOENT;
Eric Paris63c882a2009-05-21 17:02:01 -0400486
Eric Paris52cef752009-08-24 16:03:35 -0400487 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
Eric Paris75fe2b22009-07-07 10:28:23 -0400488
Eric Paris63c882a2009-05-21 17:02:01 -0400489 spin_lock(&entry->lock);
490
491 old_mask = entry->mask;
492 if (add) {
493 entry->mask |= mask;
494 new_mask = entry->mask;
495 } else {
496 entry->mask = mask;
497 new_mask = entry->mask;
498 }
499
500 spin_unlock(&entry->lock);
501
502 if (old_mask != new_mask) {
503 /* more bits in old than in new? */
504 int dropped = (old_mask & ~new_mask);
505 /* more bits in this entry than the inode's mask? */
506 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
507 /* more bits in this entry than the group? */
508 int do_group = (new_mask & ~group->mask);
509
510 /* update the inode with this new entry */
511 if (dropped || do_inode)
512 fsnotify_recalc_inode_mask(inode);
513
514 /* update the group mask with the new mask */
515 if (dropped || do_group)
516 fsnotify_recalc_group_mask(group);
517 }
518
Eric Paris52cef752009-08-24 16:03:35 -0400519 /* return the wd */
520 ret = ientry->wd;
521
522 /* match the get from fsnotify_find_mark_entry() */
Eric Paris75fe2b22009-07-07 10:28:23 -0400523 fsnotify_put_mark(entry);
524
Eric Paris52cef752009-08-24 16:03:35 -0400525 return ret;
526}
527
528static int inotify_new_watch(struct fsnotify_group *group,
529 struct inode *inode,
530 u32 arg)
531{
532 struct inotify_inode_mark_entry *tmp_ientry;
533 __u32 mask;
534 int ret;
535
536 /* don't allow invalid bits: we don't want flags set */
537 mask = inotify_arg_to_mask(arg);
538 if (unlikely(!mask))
539 return -EINVAL;
540
541 tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
542 if (unlikely(!tmp_ientry))
543 return -ENOMEM;
544
545 fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
546 tmp_ientry->fsn_entry.mask = mask;
547 tmp_ientry->wd = -1;
548
549 ret = -ENOSPC;
550 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
551 goto out_err;
552retry:
553 ret = -ENOMEM;
554 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
555 goto out_err;
556
557 spin_lock(&group->inotify_data.idr_lock);
558 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
559 group->inotify_data.last_wd,
560 &tmp_ientry->wd);
561 spin_unlock(&group->inotify_data.idr_lock);
562 if (ret) {
563 /* idr was out of memory allocate and try again */
564 if (ret == -EAGAIN)
565 goto retry;
566 goto out_err;
Eric Paris63c882a2009-05-21 17:02:01 -0400567 }
Eric Paris7e790dd2009-07-07 10:28:24 -0400568
Eric Parisdead5372009-08-24 16:03:35 -0400569 /* we put the mark on the idr, take a reference */
570 fsnotify_get_mark(&tmp_ientry->fsn_entry);
571
Eric Paris52cef752009-08-24 16:03:35 -0400572 /* we are on the idr, now get on the inode */
573 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
574 if (ret) {
575 /* we failed to get on the inode, get off the idr */
576 inotify_remove_from_idr(group, tmp_ientry);
577 goto out_err;
578 }
579
Eric Paris52cef752009-08-24 16:03:35 -0400580 /* update the idr hint, who cares about races, it's just a hint */
581 group->inotify_data.last_wd = tmp_ientry->wd;
582
583 /* increment the number of watches the user has */
584 atomic_inc(&group->inotify_data.user->inotify_watches);
585
586 /* return the watch descriptor for this new entry */
587 ret = tmp_ientry->wd;
588
589 /* match the ref from fsnotify_init_markentry() */
590 fsnotify_put_mark(&tmp_ientry->fsn_entry);
591
592out_err:
593 if (ret < 0)
594 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
595
596 return ret;
597}
598
599static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
600{
601 int ret = 0;
602
603retry:
604 /* try to update and existing watch with the new arg */
605 ret = inotify_update_existing_watch(group, inode, arg);
606 /* no mark present, try to add a new one */
607 if (ret == -ENOENT)
608 ret = inotify_new_watch(group, inode, arg);
609 /*
610 * inotify_new_watch could race with another thread which did an
611 * inotify_new_watch between the update_existing and the add watch
612 * here, go back and try to update an existing mark again.
613 */
614 if (ret == -EEXIST)
615 goto retry;
616
Eric Paris63c882a2009-05-21 17:02:01 -0400617 return ret;
618}
619
620static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
621{
622 struct fsnotify_group *group;
623 unsigned int grp_num;
624
625 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
626 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
627 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
628 if (IS_ERR(group))
629 return group;
630
631 group->max_events = max_events;
632
633 spin_lock_init(&group->inotify_data.idr_lock);
634 idr_init(&group->inotify_data.idr);
Eric Paris08e53fc2009-08-16 21:51:55 -0400635 group->inotify_data.last_wd = 1;
Eric Paris63c882a2009-05-21 17:02:01 -0400636 group->inotify_data.user = user;
637 group->inotify_data.fa = NULL;
638
639 return group;
640}
641
642
643/* inotify syscalls */
Heiko Carstens938bb9f2009-01-14 14:14:30 +0100644SYSCALL_DEFINE1(inotify_init1, int, flags)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700645{
Eric Paris63c882a2009-05-21 17:02:01 -0400646 struct fsnotify_group *group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700647 struct user_struct *user;
648 struct file *filp;
649 int fd, ret;
650
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700651 /* Check the IN_* constants for consistency. */
652 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
653 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
654
Ulrich Drepper510df2d2008-07-23 21:29:41 -0700655 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
Ulrich Drepper40065532008-07-23 21:29:32 -0700656 return -EINVAL;
657
658 fd = get_unused_fd_flags(flags & O_CLOEXEC);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700659 if (fd < 0)
660 return fd;
661
662 filp = get_empty_filp();
663 if (!filp) {
664 ret = -ENFILE;
665 goto out_put_fd;
666 }
667
David Howellsda9592e2008-11-14 10:39:05 +1100668 user = get_current_user();
Amy Griffis2d9048e2006-06-01 13:10:59 -0700669 if (unlikely(atomic_read(&user->inotify_devs) >=
670 inotify_max_user_instances)) {
671 ret = -EMFILE;
672 goto out_free_uid;
673 }
674
Eric Paris63c882a2009-05-21 17:02:01 -0400675 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
676 group = inotify_new_group(user, inotify_max_queued_events);
677 if (IS_ERR(group)) {
678 ret = PTR_ERR(group);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700679 goto out_free_uid;
680 }
681
Amy Griffis2d9048e2006-06-01 13:10:59 -0700682 filp->f_op = &inotify_fops;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800683 filp->f_path.mnt = mntget(inotify_mnt);
684 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
685 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700686 filp->f_mode = FMODE_READ;
Ulrich Drepper510df2d2008-07-23 21:29:41 -0700687 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Eric Paris63c882a2009-05-21 17:02:01 -0400688 filp->private_data = group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700689
Amy Griffis2d9048e2006-06-01 13:10:59 -0700690 atomic_inc(&user->inotify_devs);
Eric Paris63c882a2009-05-21 17:02:01 -0400691
Amy Griffis2d9048e2006-06-01 13:10:59 -0700692 fd_install(fd, filp);
693
694 return fd;
Eric Paris63c882a2009-05-21 17:02:01 -0400695
Amy Griffis2d9048e2006-06-01 13:10:59 -0700696out_free_uid:
697 free_uid(user);
698 put_filp(filp);
699out_put_fd:
700 put_unused_fd(fd);
701 return ret;
702}
703
Heiko Carstens938bb9f2009-01-14 14:14:30 +0100704SYSCALL_DEFINE0(inotify_init)
Ulrich Drepper40065532008-07-23 21:29:32 -0700705{
706 return sys_inotify_init1(0);
707}
708
Heiko Carstens2e4d0922009-01-14 14:14:31 +0100709SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
710 u32, mask)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700711{
Eric Paris63c882a2009-05-21 17:02:01 -0400712 struct fsnotify_group *group;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700713 struct inode *inode;
Al Viro2d8f3032008-07-22 09:59:21 -0400714 struct path path;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700715 struct file *filp;
716 int ret, fput_needed;
717 unsigned flags = 0;
718
719 filp = fget_light(fd, &fput_needed);
720 if (unlikely(!filp))
721 return -EBADF;
722
723 /* verify that this is indeed an inotify instance */
724 if (unlikely(filp->f_op != &inotify_fops)) {
725 ret = -EINVAL;
726 goto fput_and_out;
727 }
728
729 if (!(mask & IN_DONT_FOLLOW))
730 flags |= LOOKUP_FOLLOW;
731 if (mask & IN_ONLYDIR)
732 flags |= LOOKUP_DIRECTORY;
733
Eric Paris63c882a2009-05-21 17:02:01 -0400734 ret = inotify_find_inode(pathname, &path, flags);
735 if (ret)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700736 goto fput_and_out;
737
Eric Paris63c882a2009-05-21 17:02:01 -0400738 /* inode held in place by reference to path; group by fget on fd */
Al Viro2d8f3032008-07-22 09:59:21 -0400739 inode = path.dentry->d_inode;
Eric Paris63c882a2009-05-21 17:02:01 -0400740 group = filp->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700741
Eric Paris63c882a2009-05-21 17:02:01 -0400742 /* create/update an inode mark */
743 ret = inotify_update_watch(group, inode, mask);
744 if (unlikely(ret))
745 goto path_put_and_out;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700746
Eric Paris63c882a2009-05-21 17:02:01 -0400747path_put_and_out:
Al Viro2d8f3032008-07-22 09:59:21 -0400748 path_put(&path);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700749fput_and_out:
750 fput_light(filp, fput_needed);
751 return ret;
752}
753
Heiko Carstens2e4d0922009-01-14 14:14:31 +0100754SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700755{
Eric Paris63c882a2009-05-21 17:02:01 -0400756 struct fsnotify_group *group;
757 struct fsnotify_mark_entry *entry;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700758 struct file *filp;
Eric Paris63c882a2009-05-21 17:02:01 -0400759 int ret = 0, fput_needed;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700760
761 filp = fget_light(fd, &fput_needed);
762 if (unlikely(!filp))
763 return -EBADF;
764
765 /* verify that this is indeed an inotify instance */
766 if (unlikely(filp->f_op != &inotify_fops)) {
767 ret = -EINVAL;
768 goto out;
769 }
770
Eric Paris63c882a2009-05-21 17:02:01 -0400771 group = filp->private_data;
Amy Griffis2d9048e2006-06-01 13:10:59 -0700772
Eric Paris63c882a2009-05-21 17:02:01 -0400773 spin_lock(&group->inotify_data.idr_lock);
774 entry = idr_find(&group->inotify_data.idr, wd);
775 if (unlikely(!entry)) {
776 spin_unlock(&group->inotify_data.idr_lock);
777 ret = -EINVAL;
778 goto out;
779 }
780 fsnotify_get_mark(entry);
781 spin_unlock(&group->inotify_data.idr_lock);
782
Eric Paris528da3e2009-06-12 16:04:26 -0400783 fsnotify_destroy_mark_by_entry(entry);
Eric Paris63c882a2009-05-21 17:02:01 -0400784 fsnotify_put_mark(entry);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700785
786out:
787 fput_light(filp, fput_needed);
788 return ret;
789}
790
David Howells454e2392006-06-23 02:02:57 -0700791static int
Amy Griffis2d9048e2006-06-01 13:10:59 -0700792inotify_get_sb(struct file_system_type *fs_type, int flags,
David Howells454e2392006-06-23 02:02:57 -0700793 const char *dev_name, void *data, struct vfsmount *mnt)
Amy Griffis2d9048e2006-06-01 13:10:59 -0700794{
Andrey Mirkinfd5eea42007-10-16 23:30:13 -0700795 return get_sb_pseudo(fs_type, "inotify", NULL,
796 INOTIFYFS_SUPER_MAGIC, mnt);
Amy Griffis2d9048e2006-06-01 13:10:59 -0700797}
798
799static struct file_system_type inotify_fs_type = {
Eric Paris63c882a2009-05-21 17:02:01 -0400800 .name = "inotifyfs",
801 .get_sb = inotify_get_sb,
802 .kill_sb = kill_anon_super,
Amy Griffis2d9048e2006-06-01 13:10:59 -0700803};
804
805/*
806 * inotify_user_setup - Our initialization function. Note that we cannnot return
807 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
808 * must result in panic().
809 */
810static int __init inotify_user_setup(void)
811{
812 int ret;
813
814 ret = register_filesystem(&inotify_fs_type);
815 if (unlikely(ret))
816 panic("inotify: register_filesystem returned %d!\n", ret);
817
818 inotify_mnt = kern_mount(&inotify_fs_type);
819 if (IS_ERR(inotify_mnt))
820 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
821
Eric Paris63c882a2009-05-21 17:02:01 -0400822 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
823 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
Eric Paris63c882a2009-05-21 17:02:01 -0400824
Amy Griffis2d9048e2006-06-01 13:10:59 -0700825 inotify_max_queued_events = 16384;
826 inotify_max_user_instances = 128;
827 inotify_max_user_watches = 8192;
828
Amy Griffis2d9048e2006-06-01 13:10:59 -0700829 return 0;
830}
Amy Griffis2d9048e2006-06-01 13:10:59 -0700831module_init(inotify_user_setup);