Commit | Line | Data |
---|---|---|
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 | |
48 | static struct vfsmount *inotify_mnt __read_mostly; | |
49 | ||
50 | /* these are configurable via /proc/sys/fs/inotify/ */ | |
3c828e49 | 51 | static int inotify_max_user_instances __read_mostly; |
3c828e49 | 52 | static int inotify_max_queued_events __read_mostly; |
63c882a0 | 53 | int inotify_max_user_watches __read_mostly; |
2d9048e2 | 54 | |
63c882a0 EP |
55 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
56 | struct kmem_cache *event_priv_cachep __read_mostly; | |
2d9048e2 AG |
57 | |
58 | /* | |
63c882a0 EP |
59 | * 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. | |
2d9048e2 | 61 | */ |
63c882a0 | 62 | static atomic_t inotify_grp_num; |
2d9048e2 AG |
63 | |
64 | #ifdef CONFIG_SYSCTL | |
65 | ||
66 | #include <linux/sysctl.h> | |
67 | ||
68 | static int zero; | |
69 | ||
70 | ctl_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 | ||
63c882a0 | 105 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 106 | { |
63c882a0 | 107 | __u32 mask; |
1c17d18e | 108 | |
63c882a0 EP |
109 | /* everything should accept their own ignored and cares about children */ |
110 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD); | |
2d9048e2 | 111 | |
63c882a0 EP |
112 | /* mask off the flags used to open the fd */ |
113 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT)); | |
2d9048e2 | 114 | |
63c882a0 | 115 | return mask; |
2d9048e2 AG |
116 | } |
117 | ||
63c882a0 | 118 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 119 | { |
63c882a0 EP |
120 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
121 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
122 | } |
123 | ||
63c882a0 | 124 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
125 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
126 | { | |
63c882a0 | 127 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
128 | int ret = 0; |
129 | ||
63c882a0 EP |
130 | poll_wait(file, &group->notification_waitq, wait); |
131 | mutex_lock(&group->notification_mutex); | |
132 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 133 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 134 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
135 | |
136 | return ret; | |
137 | } | |
138 | ||
3632dee2 VN |
139 | /* |
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 | * | |
63c882a0 | 144 | * Called with the group->notification_mutex held. |
3632dee2 | 145 | */ |
63c882a0 EP |
146 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
147 | size_t count) | |
3632dee2 VN |
148 | { |
149 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 150 | struct fsnotify_event *event; |
3632dee2 | 151 | |
63c882a0 | 152 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
153 | return NULL; |
154 | ||
63c882a0 EP |
155 | event = fsnotify_peek_notify_event(group); |
156 | ||
157 | event_size += roundup(event->name_len, event_size); | |
3632dee2 VN |
158 | |
159 | if (event_size > count) | |
160 | return ERR_PTR(-EINVAL); | |
161 | ||
63c882a0 EP |
162 | /* 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; | |
3632dee2 VN |
167 | } |
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 | */ | |
63c882a0 EP |
175 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
176 | struct fsnotify_event *event, | |
3632dee2 VN |
177 | char __user *buf) |
178 | { | |
63c882a0 EP |
179 | struct inotify_event inotify_event; |
180 | struct fsnotify_event_private_data *fsn_priv; | |
181 | struct inotify_event_private_data *priv; | |
3632dee2 | 182 | size_t event_size = sizeof(struct inotify_event); |
63c882a0 EP |
183 | size_t name_len; |
184 | ||
185 | /* 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 | ||
0db501bd EB |
199 | /* round up event->name_len so it is a multiple of event_size |
200 | * plus an extra byte for the terminating '\0'. | |
201 | */ | |
202 | name_len = roundup(event->name_len + 1, event_size); | |
63c882a0 EP |
203 | inotify_event.len = name_len; |
204 | ||
205 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
206 | inotify_event.cookie = event->sync_cookie; | |
3632dee2 | 207 | |
63c882a0 EP |
208 | /* send the main event */ |
209 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
210 | return -EFAULT; |
211 | ||
63c882a0 | 212 | buf += event_size; |
3632dee2 | 213 | |
63c882a0 EP |
214 | /* |
215 | * fsnotify only stores the pathname, so here we have to send the pathname | |
216 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
217 | * with zeros. I get my zeros from the nul_inotify_event. | |
218 | */ | |
219 | if (name_len) { | |
220 | unsigned int len_to_zero = name_len - event->name_len; | |
221 | /* copy the path name */ | |
222 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
3632dee2 | 223 | return -EFAULT; |
63c882a0 | 224 | buf += event->name_len; |
3632dee2 | 225 | |
0db501bd EB |
226 | /* fill userspace with 0's */ |
227 | if (clear_user(buf, len_to_zero)) | |
63c882a0 EP |
228 | return -EFAULT; |
229 | buf += len_to_zero; | |
230 | event_size += name_len; | |
3632dee2 | 231 | } |
63c882a0 | 232 | |
3632dee2 VN |
233 | return event_size; |
234 | } | |
235 | ||
2d9048e2 AG |
236 | static ssize_t inotify_read(struct file *file, char __user *buf, |
237 | size_t count, loff_t *pos) | |
238 | { | |
63c882a0 EP |
239 | struct fsnotify_group *group; |
240 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
241 | char __user *start; |
242 | int ret; | |
243 | DEFINE_WAIT(wait); | |
244 | ||
245 | start = buf; | |
63c882a0 | 246 | group = file->private_data; |
2d9048e2 AG |
247 | |
248 | while (1) { | |
63c882a0 | 249 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); |
2d9048e2 | 250 | |
63c882a0 EP |
251 | mutex_lock(&group->notification_mutex); |
252 | kevent = get_one_event(group, count); | |
253 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 254 | |
3632dee2 VN |
255 | if (kevent) { |
256 | ret = PTR_ERR(kevent); | |
257 | if (IS_ERR(kevent)) | |
258 | break; | |
63c882a0 EP |
259 | ret = copy_event_to_user(group, kevent, buf); |
260 | fsnotify_put_event(kevent); | |
3632dee2 VN |
261 | if (ret < 0) |
262 | break; | |
263 | buf += ret; | |
264 | count -= ret; | |
265 | continue; | |
2d9048e2 AG |
266 | } |
267 | ||
3632dee2 VN |
268 | ret = -EAGAIN; |
269 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 270 | break; |
3632dee2 VN |
271 | ret = -EINTR; |
272 | if (signal_pending(current)) | |
2d9048e2 | 273 | break; |
16dbc6c9 | 274 | |
3632dee2 | 275 | if (start != buf) |
2d9048e2 | 276 | break; |
16dbc6c9 | 277 | |
3632dee2 | 278 | schedule(); |
2d9048e2 | 279 | } |
2d9048e2 | 280 | |
63c882a0 | 281 | finish_wait(&group->notification_waitq, &wait); |
3632dee2 VN |
282 | if (start != buf && ret != -EFAULT) |
283 | ret = buf - start; | |
2d9048e2 AG |
284 | return ret; |
285 | } | |
286 | ||
bcfbf84d DA |
287 | static int inotify_fasync(int fd, struct file *file, int on) |
288 | { | |
63c882a0 | 289 | struct fsnotify_group *group = file->private_data; |
bcfbf84d | 290 | |
63c882a0 | 291 | return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; |
bcfbf84d DA |
292 | } |
293 | ||
2d9048e2 AG |
294 | static int inotify_release(struct inode *ignored, struct file *file) |
295 | { | |
63c882a0 | 296 | struct fsnotify_group *group = file->private_data; |
bdae997f | 297 | struct user_struct *user = group->inotify_data.user; |
2d9048e2 | 298 | |
63c882a0 | 299 | fsnotify_clear_marks_by_group(group); |
2d9048e2 | 300 | |
63c882a0 EP |
301 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
302 | fsnotify_put_group(group); | |
2d9048e2 | 303 | |
bdae997f KP |
304 | atomic_dec(&user->inotify_devs); |
305 | ||
2d9048e2 AG |
306 | return 0; |
307 | } | |
308 | ||
309 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
310 | unsigned long arg) | |
311 | { | |
63c882a0 EP |
312 | struct fsnotify_group *group; |
313 | struct fsnotify_event_holder *holder; | |
314 | struct fsnotify_event *event; | |
2d9048e2 AG |
315 | void __user *p; |
316 | int ret = -ENOTTY; | |
63c882a0 | 317 | size_t send_len = 0; |
2d9048e2 | 318 | |
63c882a0 | 319 | group = file->private_data; |
2d9048e2 AG |
320 | p = (void __user *) arg; |
321 | ||
322 | switch (cmd) { | |
323 | case FIONREAD: | |
63c882a0 EP |
324 | mutex_lock(&group->notification_mutex); |
325 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
326 | event = holder->event; | |
327 | send_len += sizeof(struct inotify_event); | |
328 | send_len += roundup(event->name_len, | |
329 | sizeof(struct inotify_event)); | |
330 | } | |
331 | mutex_unlock(&group->notification_mutex); | |
332 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
333 | break; |
334 | } | |
335 | ||
336 | return ret; | |
337 | } | |
338 | ||
339 | static const struct file_operations inotify_fops = { | |
63c882a0 EP |
340 | .poll = inotify_poll, |
341 | .read = inotify_read, | |
342 | .fasync = inotify_fasync, | |
343 | .release = inotify_release, | |
344 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 AG |
345 | .compat_ioctl = inotify_ioctl, |
346 | }; | |
347 | ||
2d9048e2 | 348 | |
63c882a0 EP |
349 | /* |
350 | * find_inode - resolve a user-given path to a specific inode | |
351 | */ | |
352 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
353 | { | |
354 | int error; | |
355 | ||
356 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
357 | if (error) | |
358 | return error; | |
359 | /* you can only watch an inode if you have read permissions on it */ | |
360 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
361 | if (error) | |
362 | path_put(path); | |
363 | return error; | |
364 | } | |
365 | ||
dead537d EP |
366 | /* |
367 | * Remove the mark from the idr (if present) and drop the reference | |
368 | * on the mark because it was in the idr. | |
369 | */ | |
7e790dd5 EP |
370 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
371 | struct inotify_inode_mark_entry *ientry) | |
372 | { | |
373 | struct idr *idr; | |
dead537d EP |
374 | struct fsnotify_mark_entry *entry; |
375 | struct inotify_inode_mark_entry *found_ientry; | |
376 | int wd; | |
7e790dd5 EP |
377 | |
378 | spin_lock(&group->inotify_data.idr_lock); | |
379 | idr = &group->inotify_data.idr; | |
dead537d EP |
380 | wd = ientry->wd; |
381 | ||
382 | if (wd == -1) | |
383 | goto out; | |
384 | ||
385 | entry = idr_find(&group->inotify_data.idr, wd); | |
386 | if (unlikely(!entry)) | |
387 | goto out; | |
388 | ||
389 | found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); | |
390 | if (unlikely(found_ientry != ientry)) { | |
391 | /* We found an entry in the idr with the right wd, but it's | |
392 | * not the entry we were told to remove. eparis seriously | |
393 | * fucked up somewhere. */ | |
394 | WARN_ON(1); | |
395 | ientry->wd = -1; | |
396 | goto out; | |
397 | } | |
398 | ||
399 | /* One ref for being in the idr, one ref held by the caller */ | |
400 | BUG_ON(atomic_read(&entry->refcnt) < 2); | |
401 | ||
402 | idr_remove(idr, wd); | |
7e790dd5 | 403 | ientry->wd = -1; |
dead537d EP |
404 | |
405 | /* removed from the idr, drop that ref */ | |
406 | fsnotify_put_mark(entry); | |
407 | out: | |
408 | spin_unlock(&group->inotify_data.idr_lock); | |
7e790dd5 | 409 | } |
dead537d | 410 | |
63c882a0 | 411 | /* |
dead537d | 412 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 413 | */ |
528da3e9 EP |
414 | void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry, |
415 | struct fsnotify_group *group) | |
63c882a0 EP |
416 | { |
417 | struct inotify_inode_mark_entry *ientry; | |
f44aebcc | 418 | struct fsnotify_event *ignored_event; |
63c882a0 EP |
419 | struct inotify_event_private_data *event_priv; |
420 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 421 | int ret; |
63c882a0 | 422 | |
f44aebcc EP |
423 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
424 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
425 | GFP_NOFS); | |
426 | if (!ignored_event) | |
427 | return; | |
428 | ||
63c882a0 EP |
429 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
430 | ||
f44aebcc | 431 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
432 | if (unlikely(!event_priv)) |
433 | goto skip_send_ignore; | |
434 | ||
435 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
436 | ||
437 | fsn_event_priv->group = group; | |
438 | event_priv->wd = ientry->wd; | |
439 | ||
eef3a116 EP |
440 | ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv); |
441 | if (ret) | |
63c882a0 EP |
442 | inotify_free_event_priv(fsn_event_priv); |
443 | ||
444 | skip_send_ignore: | |
445 | ||
f44aebcc EP |
446 | /* matches the reference taken when the event was created */ |
447 | fsnotify_put_event(ignored_event); | |
448 | ||
63c882a0 | 449 | /* remove this entry from the idr */ |
7e790dd5 | 450 | inotify_remove_from_idr(group, ientry); |
63c882a0 | 451 | |
5549f7cd | 452 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
453 | } |
454 | ||
455 | /* ding dong the mark is dead */ | |
456 | static void inotify_free_mark(struct fsnotify_mark_entry *entry) | |
457 | { | |
458 | struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry; | |
459 | ||
460 | kmem_cache_free(inotify_inode_mark_cachep, ientry); | |
461 | } | |
462 | ||
52cef755 EP |
463 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
464 | struct inode *inode, | |
465 | u32 arg) | |
63c882a0 | 466 | { |
52cef755 | 467 | struct fsnotify_mark_entry *entry; |
63c882a0 | 468 | struct inotify_inode_mark_entry *ientry; |
63c882a0 | 469 | __u32 old_mask, new_mask; |
52cef755 EP |
470 | __u32 mask; |
471 | int add = (arg & IN_MASK_ADD); | |
472 | int ret; | |
63c882a0 EP |
473 | |
474 | /* don't allow invalid bits: we don't want flags set */ | |
475 | mask = inotify_arg_to_mask(arg); | |
476 | if (unlikely(!mask)) | |
477 | return -EINVAL; | |
478 | ||
63c882a0 EP |
479 | spin_lock(&inode->i_lock); |
480 | entry = fsnotify_find_mark_entry(group, inode); | |
481 | spin_unlock(&inode->i_lock); | |
52cef755 EP |
482 | if (!entry) |
483 | return -ENOENT; | |
7e790dd5 | 484 | |
52cef755 | 485 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
75fe2b26 | 486 | |
63c882a0 EP |
487 | spin_lock(&entry->lock); |
488 | ||
489 | old_mask = entry->mask; | |
490 | if (add) { | |
491 | entry->mask |= mask; | |
492 | new_mask = entry->mask; | |
493 | } else { | |
494 | entry->mask = mask; | |
495 | new_mask = entry->mask; | |
496 | } | |
497 | ||
498 | spin_unlock(&entry->lock); | |
499 | ||
500 | if (old_mask != new_mask) { | |
501 | /* more bits in old than in new? */ | |
502 | int dropped = (old_mask & ~new_mask); | |
503 | /* more bits in this entry than the inode's mask? */ | |
504 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); | |
505 | /* more bits in this entry than the group? */ | |
506 | int do_group = (new_mask & ~group->mask); | |
507 | ||
508 | /* update the inode with this new entry */ | |
509 | if (dropped || do_inode) | |
510 | fsnotify_recalc_inode_mask(inode); | |
511 | ||
512 | /* update the group mask with the new mask */ | |
513 | if (dropped || do_group) | |
514 | fsnotify_recalc_group_mask(group); | |
515 | } | |
516 | ||
52cef755 EP |
517 | /* return the wd */ |
518 | ret = ientry->wd; | |
519 | ||
520 | /* match the get from fsnotify_find_mark_entry() */ | |
75fe2b26 EP |
521 | fsnotify_put_mark(entry); |
522 | ||
52cef755 EP |
523 | return ret; |
524 | } | |
525 | ||
526 | static int inotify_new_watch(struct fsnotify_group *group, | |
527 | struct inode *inode, | |
528 | u32 arg) | |
529 | { | |
530 | struct inotify_inode_mark_entry *tmp_ientry; | |
531 | __u32 mask; | |
532 | int ret; | |
533 | ||
534 | /* don't allow invalid bits: we don't want flags set */ | |
535 | mask = inotify_arg_to_mask(arg); | |
536 | if (unlikely(!mask)) | |
537 | return -EINVAL; | |
538 | ||
539 | tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); | |
540 | if (unlikely(!tmp_ientry)) | |
541 | return -ENOMEM; | |
542 | ||
543 | fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark); | |
544 | tmp_ientry->fsn_entry.mask = mask; | |
545 | tmp_ientry->wd = -1; | |
546 | ||
547 | ret = -ENOSPC; | |
548 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
549 | goto out_err; | |
550 | retry: | |
551 | ret = -ENOMEM; | |
552 | if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL))) | |
553 | goto out_err; | |
554 | ||
555 | spin_lock(&group->inotify_data.idr_lock); | |
556 | ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry, | |
557 | group->inotify_data.last_wd, | |
558 | &tmp_ientry->wd); | |
559 | spin_unlock(&group->inotify_data.idr_lock); | |
560 | if (ret) { | |
561 | /* idr was out of memory allocate and try again */ | |
562 | if (ret == -EAGAIN) | |
563 | goto retry; | |
564 | goto out_err; | |
565 | } | |
566 | ||
dead537d EP |
567 | /* we put the mark on the idr, take a reference */ |
568 | fsnotify_get_mark(&tmp_ientry->fsn_entry); | |
569 | ||
52cef755 EP |
570 | /* we are on the idr, now get on the inode */ |
571 | ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode); | |
572 | if (ret) { | |
573 | /* we failed to get on the inode, get off the idr */ | |
574 | inotify_remove_from_idr(group, tmp_ientry); | |
575 | goto out_err; | |
576 | } | |
577 | ||
52cef755 EP |
578 | /* update the idr hint, who cares about races, it's just a hint */ |
579 | group->inotify_data.last_wd = tmp_ientry->wd; | |
580 | ||
581 | /* increment the number of watches the user has */ | |
582 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
583 | ||
584 | /* return the watch descriptor for this new entry */ | |
585 | ret = tmp_ientry->wd; | |
586 | ||
587 | /* match the ref from fsnotify_init_markentry() */ | |
588 | fsnotify_put_mark(&tmp_ientry->fsn_entry); | |
589 | ||
63c882a0 | 590 | out_err: |
52cef755 | 591 | if (ret < 0) |
7e790dd5 | 592 | kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry); |
52cef755 EP |
593 | |
594 | return ret; | |
595 | } | |
596 | ||
597 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
598 | { | |
599 | int ret = 0; | |
600 | ||
601 | retry: | |
602 | /* try to update and existing watch with the new arg */ | |
603 | ret = inotify_update_existing_watch(group, inode, arg); | |
604 | /* no mark present, try to add a new one */ | |
605 | if (ret == -ENOENT) | |
606 | ret = inotify_new_watch(group, inode, arg); | |
607 | /* | |
608 | * inotify_new_watch could race with another thread which did an | |
609 | * inotify_new_watch between the update_existing and the add watch | |
610 | * here, go back and try to update an existing mark again. | |
611 | */ | |
612 | if (ret == -EEXIST) | |
613 | goto retry; | |
7e790dd5 | 614 | |
63c882a0 EP |
615 | return ret; |
616 | } | |
617 | ||
618 | static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events) | |
619 | { | |
620 | struct fsnotify_group *group; | |
621 | unsigned int grp_num; | |
622 | ||
623 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ | |
624 | grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num)); | |
625 | group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops); | |
626 | if (IS_ERR(group)) | |
627 | return group; | |
628 | ||
629 | group->max_events = max_events; | |
630 | ||
631 | spin_lock_init(&group->inotify_data.idr_lock); | |
632 | idr_init(&group->inotify_data.idr); | |
08e53fcb | 633 | group->inotify_data.last_wd = 1; |
63c882a0 EP |
634 | group->inotify_data.user = user; |
635 | group->inotify_data.fa = NULL; | |
636 | ||
637 | return group; | |
638 | } | |
639 | ||
640 | ||
641 | /* inotify syscalls */ | |
938bb9f5 | 642 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 643 | { |
63c882a0 | 644 | struct fsnotify_group *group; |
2d9048e2 AG |
645 | struct user_struct *user; |
646 | struct file *filp; | |
647 | int fd, ret; | |
648 | ||
e38b36f3 UD |
649 | /* Check the IN_* constants for consistency. */ |
650 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
651 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
652 | ||
510df2dd | 653 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
654 | return -EINVAL; |
655 | ||
656 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2d9048e2 AG |
657 | if (fd < 0) |
658 | return fd; | |
659 | ||
660 | filp = get_empty_filp(); | |
661 | if (!filp) { | |
662 | ret = -ENFILE; | |
663 | goto out_put_fd; | |
664 | } | |
665 | ||
da9592ed | 666 | user = get_current_user(); |
2d9048e2 AG |
667 | if (unlikely(atomic_read(&user->inotify_devs) >= |
668 | inotify_max_user_instances)) { | |
669 | ret = -EMFILE; | |
670 | goto out_free_uid; | |
671 | } | |
672 | ||
63c882a0 EP |
673 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
674 | group = inotify_new_group(user, inotify_max_queued_events); | |
675 | if (IS_ERR(group)) { | |
676 | ret = PTR_ERR(group); | |
2d9048e2 AG |
677 | goto out_free_uid; |
678 | } | |
679 | ||
2d9048e2 | 680 | filp->f_op = &inotify_fops; |
0f7fc9e4 JJS |
681 | filp->f_path.mnt = mntget(inotify_mnt); |
682 | filp->f_path.dentry = dget(inotify_mnt->mnt_root); | |
683 | filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping; | |
2d9048e2 | 684 | filp->f_mode = FMODE_READ; |
510df2dd | 685 | filp->f_flags = O_RDONLY | (flags & O_NONBLOCK); |
63c882a0 EP |
686 | filp->private_data = group; |
687 | ||
2d9048e2 | 688 | atomic_inc(&user->inotify_devs); |
63c882a0 | 689 | |
2d9048e2 AG |
690 | fd_install(fd, filp); |
691 | ||
692 | return fd; | |
63c882a0 | 693 | |
2d9048e2 AG |
694 | out_free_uid: |
695 | free_uid(user); | |
696 | put_filp(filp); | |
697 | out_put_fd: | |
698 | put_unused_fd(fd); | |
699 | return ret; | |
700 | } | |
701 | ||
938bb9f5 | 702 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
703 | { |
704 | return sys_inotify_init1(0); | |
705 | } | |
706 | ||
2e4d0924 HC |
707 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
708 | u32, mask) | |
2d9048e2 | 709 | { |
63c882a0 | 710 | struct fsnotify_group *group; |
2d9048e2 | 711 | struct inode *inode; |
2d8f3038 | 712 | struct path path; |
2d9048e2 AG |
713 | struct file *filp; |
714 | int ret, fput_needed; | |
715 | unsigned flags = 0; | |
716 | ||
717 | filp = fget_light(fd, &fput_needed); | |
718 | if (unlikely(!filp)) | |
719 | return -EBADF; | |
720 | ||
721 | /* verify that this is indeed an inotify instance */ | |
722 | if (unlikely(filp->f_op != &inotify_fops)) { | |
723 | ret = -EINVAL; | |
724 | goto fput_and_out; | |
725 | } | |
726 | ||
727 | if (!(mask & IN_DONT_FOLLOW)) | |
728 | flags |= LOOKUP_FOLLOW; | |
729 | if (mask & IN_ONLYDIR) | |
730 | flags |= LOOKUP_DIRECTORY; | |
731 | ||
63c882a0 EP |
732 | ret = inotify_find_inode(pathname, &path, flags); |
733 | if (ret) | |
2d9048e2 AG |
734 | goto fput_and_out; |
735 | ||
63c882a0 | 736 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 737 | inode = path.dentry->d_inode; |
63c882a0 | 738 | group = filp->private_data; |
2d9048e2 | 739 | |
63c882a0 EP |
740 | /* create/update an inode mark */ |
741 | ret = inotify_update_watch(group, inode, mask); | |
742 | if (unlikely(ret)) | |
743 | goto path_put_and_out; | |
2d9048e2 | 744 | |
63c882a0 | 745 | path_put_and_out: |
2d8f3038 | 746 | path_put(&path); |
2d9048e2 AG |
747 | fput_and_out: |
748 | fput_light(filp, fput_needed); | |
749 | return ret; | |
750 | } | |
751 | ||
2e4d0924 | 752 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 753 | { |
63c882a0 EP |
754 | struct fsnotify_group *group; |
755 | struct fsnotify_mark_entry *entry; | |
2d9048e2 | 756 | struct file *filp; |
63c882a0 | 757 | int ret = 0, fput_needed; |
2d9048e2 AG |
758 | |
759 | filp = fget_light(fd, &fput_needed); | |
760 | if (unlikely(!filp)) | |
761 | return -EBADF; | |
762 | ||
763 | /* verify that this is indeed an inotify instance */ | |
764 | if (unlikely(filp->f_op != &inotify_fops)) { | |
765 | ret = -EINVAL; | |
766 | goto out; | |
767 | } | |
768 | ||
63c882a0 | 769 | group = filp->private_data; |
2d9048e2 | 770 | |
63c882a0 EP |
771 | spin_lock(&group->inotify_data.idr_lock); |
772 | entry = idr_find(&group->inotify_data.idr, wd); | |
773 | if (unlikely(!entry)) { | |
774 | spin_unlock(&group->inotify_data.idr_lock); | |
775 | ret = -EINVAL; | |
776 | goto out; | |
777 | } | |
778 | fsnotify_get_mark(entry); | |
779 | spin_unlock(&group->inotify_data.idr_lock); | |
780 | ||
528da3e9 | 781 | fsnotify_destroy_mark_by_entry(entry); |
63c882a0 | 782 | fsnotify_put_mark(entry); |
2d9048e2 AG |
783 | |
784 | out: | |
785 | fput_light(filp, fput_needed); | |
786 | return ret; | |
787 | } | |
788 | ||
454e2398 | 789 | static int |
2d9048e2 | 790 | inotify_get_sb(struct file_system_type *fs_type, int flags, |
454e2398 | 791 | const char *dev_name, void *data, struct vfsmount *mnt) |
2d9048e2 | 792 | { |
fd5eea42 AM |
793 | return get_sb_pseudo(fs_type, "inotify", NULL, |
794 | INOTIFYFS_SUPER_MAGIC, mnt); | |
2d9048e2 AG |
795 | } |
796 | ||
797 | static struct file_system_type inotify_fs_type = { | |
63c882a0 EP |
798 | .name = "inotifyfs", |
799 | .get_sb = inotify_get_sb, | |
800 | .kill_sb = kill_anon_super, | |
2d9048e2 AG |
801 | }; |
802 | ||
803 | /* | |
804 | * inotify_user_setup - Our initialization function. Note that we cannnot return | |
805 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here | |
806 | * must result in panic(). | |
807 | */ | |
808 | static int __init inotify_user_setup(void) | |
809 | { | |
810 | int ret; | |
811 | ||
812 | ret = register_filesystem(&inotify_fs_type); | |
813 | if (unlikely(ret)) | |
814 | panic("inotify: register_filesystem returned %d!\n", ret); | |
815 | ||
816 | inotify_mnt = kern_mount(&inotify_fs_type); | |
817 | if (IS_ERR(inotify_mnt)) | |
818 | panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt)); | |
819 | ||
63c882a0 EP |
820 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC); |
821 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); | |
63c882a0 | 822 | |
2d9048e2 AG |
823 | inotify_max_queued_events = 16384; |
824 | inotify_max_user_instances = 128; | |
825 | inotify_max_user_watches = 8192; | |
826 | ||
2d9048e2 AG |
827 | return 0; |
828 | } | |
2d9048e2 | 829 | module_init(inotify_user_setup); |