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 | 31 | #include <linux/kernel.h> /* roundup() */ |
63c882a0 | 32 | #include <linux/namei.h> /* LOOKUP_FOLLOW */ |
63c882a0 EP |
33 | #include <linux/sched.h> /* struct user */ |
34 | #include <linux/slab.h> /* struct kmem_cache */ | |
2d9048e2 | 35 | #include <linux/syscalls.h> |
63c882a0 | 36 | #include <linux/types.h> |
c44dcc56 | 37 | #include <linux/anon_inodes.h> |
63c882a0 EP |
38 | #include <linux/uaccess.h> |
39 | #include <linux/poll.h> | |
40 | #include <linux/wait.h> | |
2d9048e2 | 41 | |
63c882a0 | 42 | #include "inotify.h" |
2d9048e2 | 43 | |
63c882a0 | 44 | #include <asm/ioctls.h> |
2d9048e2 | 45 | |
2d9048e2 | 46 | /* these are configurable via /proc/sys/fs/inotify/ */ |
3c828e49 | 47 | static int inotify_max_user_instances __read_mostly; |
3c828e49 | 48 | static int inotify_max_queued_events __read_mostly; |
63c882a0 | 49 | int inotify_max_user_watches __read_mostly; |
2d9048e2 | 50 | |
63c882a0 EP |
51 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
52 | struct kmem_cache *event_priv_cachep __read_mostly; | |
2d9048e2 AG |
53 | |
54 | /* | |
63c882a0 EP |
55 | * When inotify registers a new group it increments this and uses that |
56 | * value as an offset to set the fsnotify group "name" and priority. | |
2d9048e2 | 57 | */ |
63c882a0 | 58 | static atomic_t inotify_grp_num; |
2d9048e2 AG |
59 | |
60 | #ifdef CONFIG_SYSCTL | |
61 | ||
62 | #include <linux/sysctl.h> | |
63 | ||
64 | static int zero; | |
65 | ||
66 | ctl_table inotify_table[] = { | |
67 | { | |
2d9048e2 AG |
68 | .procname = "max_user_instances", |
69 | .data = &inotify_max_user_instances, | |
70 | .maxlen = sizeof(int), | |
71 | .mode = 0644, | |
6d456111 | 72 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
73 | .extra1 = &zero, |
74 | }, | |
75 | { | |
2d9048e2 AG |
76 | .procname = "max_user_watches", |
77 | .data = &inotify_max_user_watches, | |
78 | .maxlen = sizeof(int), | |
79 | .mode = 0644, | |
6d456111 | 80 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
81 | .extra1 = &zero, |
82 | }, | |
83 | { | |
2d9048e2 AG |
84 | .procname = "max_queued_events", |
85 | .data = &inotify_max_queued_events, | |
86 | .maxlen = sizeof(int), | |
87 | .mode = 0644, | |
6d456111 | 88 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
89 | .extra1 = &zero |
90 | }, | |
ab09203e | 91 | { } |
2d9048e2 AG |
92 | }; |
93 | #endif /* CONFIG_SYSCTL */ | |
94 | ||
63c882a0 | 95 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 96 | { |
63c882a0 | 97 | __u32 mask; |
1c17d18e | 98 | |
63c882a0 EP |
99 | /* everything should accept their own ignored and cares about children */ |
100 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD); | |
2d9048e2 | 101 | |
63c882a0 EP |
102 | /* mask off the flags used to open the fd */ |
103 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT)); | |
2d9048e2 | 104 | |
63c882a0 | 105 | return mask; |
2d9048e2 AG |
106 | } |
107 | ||
63c882a0 | 108 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 109 | { |
63c882a0 EP |
110 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
111 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
112 | } |
113 | ||
63c882a0 | 114 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
115 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
116 | { | |
63c882a0 | 117 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
118 | int ret = 0; |
119 | ||
63c882a0 EP |
120 | poll_wait(file, &group->notification_waitq, wait); |
121 | mutex_lock(&group->notification_mutex); | |
122 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 123 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 124 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
125 | |
126 | return ret; | |
127 | } | |
128 | ||
3632dee2 VN |
129 | /* |
130 | * Get an inotify_kernel_event if one exists and is small | |
131 | * enough to fit in "count". Return an error pointer if | |
132 | * not large enough. | |
133 | * | |
63c882a0 | 134 | * Called with the group->notification_mutex held. |
3632dee2 | 135 | */ |
63c882a0 EP |
136 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
137 | size_t count) | |
3632dee2 VN |
138 | { |
139 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 140 | struct fsnotify_event *event; |
3632dee2 | 141 | |
63c882a0 | 142 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
143 | return NULL; |
144 | ||
63c882a0 EP |
145 | event = fsnotify_peek_notify_event(group); |
146 | ||
83cb10f0 EP |
147 | if (event->name_len) |
148 | event_size += roundup(event->name_len + 1, event_size); | |
3632dee2 VN |
149 | |
150 | if (event_size > count) | |
151 | return ERR_PTR(-EINVAL); | |
152 | ||
63c882a0 EP |
153 | /* held the notification_mutex the whole time, so this is the |
154 | * same event we peeked above */ | |
155 | fsnotify_remove_notify_event(group); | |
156 | ||
157 | return event; | |
3632dee2 VN |
158 | } |
159 | ||
160 | /* | |
161 | * Copy an event to user space, returning how much we copied. | |
162 | * | |
163 | * We already checked that the event size is smaller than the | |
164 | * buffer we had in "get_one_event()" above. | |
165 | */ | |
63c882a0 EP |
166 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
167 | struct fsnotify_event *event, | |
3632dee2 VN |
168 | char __user *buf) |
169 | { | |
63c882a0 EP |
170 | struct inotify_event inotify_event; |
171 | struct fsnotify_event_private_data *fsn_priv; | |
172 | struct inotify_event_private_data *priv; | |
3632dee2 | 173 | size_t event_size = sizeof(struct inotify_event); |
b962e731 | 174 | size_t name_len = 0; |
63c882a0 EP |
175 | |
176 | /* we get the inotify watch descriptor from the event private data */ | |
177 | spin_lock(&event->lock); | |
178 | fsn_priv = fsnotify_remove_priv_from_event(group, event); | |
179 | spin_unlock(&event->lock); | |
180 | ||
181 | if (!fsn_priv) | |
182 | inotify_event.wd = -1; | |
183 | else { | |
184 | priv = container_of(fsn_priv, struct inotify_event_private_data, | |
185 | fsnotify_event_priv_data); | |
186 | inotify_event.wd = priv->wd; | |
187 | inotify_free_event_priv(fsn_priv); | |
188 | } | |
189 | ||
b962e731 BR |
190 | /* |
191 | * round up event->name_len so it is a multiple of event_size | |
0db501bd EB |
192 | * plus an extra byte for the terminating '\0'. |
193 | */ | |
b962e731 BR |
194 | if (event->name_len) |
195 | name_len = roundup(event->name_len + 1, event_size); | |
63c882a0 EP |
196 | inotify_event.len = name_len; |
197 | ||
198 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
199 | inotify_event.cookie = event->sync_cookie; | |
3632dee2 | 200 | |
63c882a0 EP |
201 | /* send the main event */ |
202 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
203 | return -EFAULT; |
204 | ||
63c882a0 | 205 | buf += event_size; |
3632dee2 | 206 | |
63c882a0 EP |
207 | /* |
208 | * fsnotify only stores the pathname, so here we have to send the pathname | |
209 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
210 | * with zeros. I get my zeros from the nul_inotify_event. | |
211 | */ | |
212 | if (name_len) { | |
213 | unsigned int len_to_zero = name_len - event->name_len; | |
214 | /* copy the path name */ | |
215 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
3632dee2 | 216 | return -EFAULT; |
63c882a0 | 217 | buf += event->name_len; |
3632dee2 | 218 | |
0db501bd EB |
219 | /* fill userspace with 0's */ |
220 | if (clear_user(buf, len_to_zero)) | |
63c882a0 EP |
221 | return -EFAULT; |
222 | buf += len_to_zero; | |
223 | event_size += name_len; | |
3632dee2 | 224 | } |
63c882a0 | 225 | |
3632dee2 VN |
226 | return event_size; |
227 | } | |
228 | ||
2d9048e2 AG |
229 | static ssize_t inotify_read(struct file *file, char __user *buf, |
230 | size_t count, loff_t *pos) | |
231 | { | |
63c882a0 EP |
232 | struct fsnotify_group *group; |
233 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
234 | char __user *start; |
235 | int ret; | |
236 | DEFINE_WAIT(wait); | |
237 | ||
238 | start = buf; | |
63c882a0 | 239 | group = file->private_data; |
2d9048e2 AG |
240 | |
241 | while (1) { | |
63c882a0 | 242 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); |
2d9048e2 | 243 | |
63c882a0 EP |
244 | mutex_lock(&group->notification_mutex); |
245 | kevent = get_one_event(group, count); | |
246 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 247 | |
3632dee2 VN |
248 | if (kevent) { |
249 | ret = PTR_ERR(kevent); | |
250 | if (IS_ERR(kevent)) | |
251 | break; | |
63c882a0 EP |
252 | ret = copy_event_to_user(group, kevent, buf); |
253 | fsnotify_put_event(kevent); | |
3632dee2 VN |
254 | if (ret < 0) |
255 | break; | |
256 | buf += ret; | |
257 | count -= ret; | |
258 | continue; | |
2d9048e2 AG |
259 | } |
260 | ||
3632dee2 VN |
261 | ret = -EAGAIN; |
262 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 263 | break; |
3632dee2 VN |
264 | ret = -EINTR; |
265 | if (signal_pending(current)) | |
2d9048e2 | 266 | break; |
16dbc6c9 | 267 | |
3632dee2 | 268 | if (start != buf) |
2d9048e2 | 269 | break; |
16dbc6c9 | 270 | |
3632dee2 | 271 | schedule(); |
2d9048e2 | 272 | } |
2d9048e2 | 273 | |
63c882a0 | 274 | finish_wait(&group->notification_waitq, &wait); |
3632dee2 VN |
275 | if (start != buf && ret != -EFAULT) |
276 | ret = buf - start; | |
2d9048e2 AG |
277 | return ret; |
278 | } | |
279 | ||
bcfbf84d DA |
280 | static int inotify_fasync(int fd, struct file *file, int on) |
281 | { | |
63c882a0 | 282 | struct fsnotify_group *group = file->private_data; |
bcfbf84d | 283 | |
63c882a0 | 284 | return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; |
bcfbf84d DA |
285 | } |
286 | ||
2d9048e2 AG |
287 | static int inotify_release(struct inode *ignored, struct file *file) |
288 | { | |
63c882a0 | 289 | struct fsnotify_group *group = file->private_data; |
bdae997f | 290 | struct user_struct *user = group->inotify_data.user; |
2d9048e2 | 291 | |
63c882a0 | 292 | fsnotify_clear_marks_by_group(group); |
2d9048e2 | 293 | |
63c882a0 EP |
294 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
295 | fsnotify_put_group(group); | |
2d9048e2 | 296 | |
bdae997f KP |
297 | atomic_dec(&user->inotify_devs); |
298 | ||
2d9048e2 AG |
299 | return 0; |
300 | } | |
301 | ||
302 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
303 | unsigned long arg) | |
304 | { | |
63c882a0 EP |
305 | struct fsnotify_group *group; |
306 | struct fsnotify_event_holder *holder; | |
307 | struct fsnotify_event *event; | |
2d9048e2 AG |
308 | void __user *p; |
309 | int ret = -ENOTTY; | |
63c882a0 | 310 | size_t send_len = 0; |
2d9048e2 | 311 | |
63c882a0 | 312 | group = file->private_data; |
2d9048e2 AG |
313 | p = (void __user *) arg; |
314 | ||
315 | switch (cmd) { | |
316 | case FIONREAD: | |
63c882a0 EP |
317 | mutex_lock(&group->notification_mutex); |
318 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
319 | event = holder->event; | |
320 | send_len += sizeof(struct inotify_event); | |
83cb10f0 EP |
321 | if (event->name_len) |
322 | send_len += roundup(event->name_len + 1, | |
323 | sizeof(struct inotify_event)); | |
63c882a0 EP |
324 | } |
325 | mutex_unlock(&group->notification_mutex); | |
326 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
327 | break; |
328 | } | |
329 | ||
330 | return ret; | |
331 | } | |
332 | ||
333 | static const struct file_operations inotify_fops = { | |
63c882a0 EP |
334 | .poll = inotify_poll, |
335 | .read = inotify_read, | |
336 | .fasync = inotify_fasync, | |
337 | .release = inotify_release, | |
338 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 AG |
339 | .compat_ioctl = inotify_ioctl, |
340 | }; | |
341 | ||
2d9048e2 | 342 | |
63c882a0 EP |
343 | /* |
344 | * find_inode - resolve a user-given path to a specific inode | |
345 | */ | |
346 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
347 | { | |
348 | int error; | |
349 | ||
350 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
351 | if (error) | |
352 | return error; | |
353 | /* you can only watch an inode if you have read permissions on it */ | |
354 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
355 | if (error) | |
356 | path_put(path); | |
357 | return error; | |
358 | } | |
359 | ||
b7ba8371 | 360 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
7050c488 | 361 | int *last_wd, |
b7ba8371 EP |
362 | struct inotify_inode_mark_entry *ientry) |
363 | { | |
364 | int ret; | |
365 | ||
366 | do { | |
367 | if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) | |
368 | return -ENOMEM; | |
369 | ||
370 | spin_lock(idr_lock); | |
7050c488 | 371 | ret = idr_get_new_above(idr, ientry, *last_wd + 1, |
b7ba8371 EP |
372 | &ientry->wd); |
373 | /* we added the mark to the idr, take a reference */ | |
7050c488 | 374 | if (!ret) { |
b7ba8371 | 375 | fsnotify_get_mark(&ientry->fsn_entry); |
7050c488 EP |
376 | *last_wd = ientry->wd; |
377 | } | |
b7ba8371 EP |
378 | spin_unlock(idr_lock); |
379 | } while (ret == -EAGAIN); | |
380 | ||
381 | return ret; | |
382 | } | |
383 | ||
384 | static struct inotify_inode_mark_entry *inotify_idr_find_locked(struct fsnotify_group *group, | |
385 | int wd) | |
386 | { | |
387 | struct idr *idr = &group->inotify_data.idr; | |
388 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
389 | struct inotify_inode_mark_entry *ientry; | |
390 | ||
391 | assert_spin_locked(idr_lock); | |
392 | ||
393 | ientry = idr_find(idr, wd); | |
394 | if (ientry) { | |
395 | struct fsnotify_mark_entry *fsn_entry = &ientry->fsn_entry; | |
396 | ||
397 | fsnotify_get_mark(fsn_entry); | |
398 | /* One ref for being in the idr, one ref we just took */ | |
399 | BUG_ON(atomic_read(&fsn_entry->refcnt) < 2); | |
400 | } | |
401 | ||
402 | return ientry; | |
403 | } | |
404 | ||
405 | static struct inotify_inode_mark_entry *inotify_idr_find(struct fsnotify_group *group, | |
406 | int wd) | |
407 | { | |
408 | struct inotify_inode_mark_entry *ientry; | |
409 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
410 | ||
411 | spin_lock(idr_lock); | |
412 | ientry = inotify_idr_find_locked(group, wd); | |
413 | spin_unlock(idr_lock); | |
414 | ||
415 | return ientry; | |
416 | } | |
417 | ||
418 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
419 | struct inotify_inode_mark_entry *ientry) | |
420 | { | |
421 | struct idr *idr = &group->inotify_data.idr; | |
422 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
423 | int wd = ientry->wd; | |
424 | ||
425 | assert_spin_locked(idr_lock); | |
426 | ||
427 | idr_remove(idr, wd); | |
428 | ||
429 | /* removed from the idr, drop that ref */ | |
430 | fsnotify_put_mark(&ientry->fsn_entry); | |
431 | } | |
432 | ||
dead537d EP |
433 | /* |
434 | * Remove the mark from the idr (if present) and drop the reference | |
435 | * on the mark because it was in the idr. | |
436 | */ | |
7e790dd5 EP |
437 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
438 | struct inotify_inode_mark_entry *ientry) | |
439 | { | |
b7ba8371 EP |
440 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
441 | struct inotify_inode_mark_entry *found_ientry = NULL; | |
dead537d | 442 | int wd; |
7e790dd5 | 443 | |
b7ba8371 | 444 | spin_lock(idr_lock); |
dead537d EP |
445 | wd = ientry->wd; |
446 | ||
b7ba8371 EP |
447 | /* |
448 | * does this ientry think it is in the idr? we shouldn't get called | |
449 | * if it wasn't.... | |
450 | */ | |
451 | if (wd == -1) { | |
452 | printk(KERN_WARNING "%s: ientry=%p ientry->wd=%d ientry->group=%p" | |
453 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, | |
454 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
455 | WARN_ON(1); | |
dead537d | 456 | goto out; |
b7ba8371 | 457 | } |
dead537d | 458 | |
b7ba8371 EP |
459 | /* Lets look in the idr to see if we find it */ |
460 | found_ientry = inotify_idr_find_locked(group, wd); | |
461 | if (unlikely(!found_ientry)) { | |
462 | printk(KERN_WARNING "%s: ientry=%p ientry->wd=%d ientry->group=%p" | |
463 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, | |
464 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
465 | WARN_ON(1); | |
dead537d | 466 | goto out; |
b7ba8371 | 467 | } |
dead537d | 468 | |
b7ba8371 EP |
469 | /* |
470 | * We found an entry in the idr at the right wd, but it's | |
471 | * not the entry we were told to remove. eparis seriously | |
472 | * fucked up somewhere. | |
473 | */ | |
dead537d | 474 | if (unlikely(found_ientry != ientry)) { |
dead537d | 475 | WARN_ON(1); |
b7ba8371 EP |
476 | printk(KERN_WARNING "%s: ientry=%p ientry->wd=%d ientry->group=%p " |
477 | "entry->inode=%p found_ientry=%p found_ientry->wd=%d " | |
478 | "found_ientry->group=%p found_ientry->inode=%p\n", | |
479 | __func__, ientry, ientry->wd, ientry->fsn_entry.group, | |
480 | ientry->fsn_entry.inode, found_ientry, found_ientry->wd, | |
481 | found_ientry->fsn_entry.group, | |
482 | found_ientry->fsn_entry.inode); | |
dead537d EP |
483 | goto out; |
484 | } | |
485 | ||
b7ba8371 EP |
486 | /* |
487 | * One ref for being in the idr | |
488 | * one ref held by the caller trying to kill us | |
489 | * one ref grabbed by inotify_idr_find | |
490 | */ | |
491 | if (unlikely(atomic_read(&ientry->fsn_entry.refcnt) < 3)) { | |
492 | printk(KERN_WARNING "%s: ientry=%p ientry->wd=%d ientry->group=%p" | |
493 | " ientry->inode=%p\n", __func__, ientry, ientry->wd, | |
494 | ientry->fsn_entry.group, ientry->fsn_entry.inode); | |
495 | /* we can't really recover with bad ref cnting.. */ | |
496 | BUG(); | |
497 | } | |
dead537d | 498 | |
b7ba8371 | 499 | do_inotify_remove_from_idr(group, ientry); |
dead537d | 500 | out: |
b7ba8371 EP |
501 | /* match the ref taken by inotify_idr_find_locked() */ |
502 | if (found_ientry) | |
503 | fsnotify_put_mark(&found_ientry->fsn_entry); | |
504 | ientry->wd = -1; | |
505 | spin_unlock(idr_lock); | |
7e790dd5 | 506 | } |
dead537d | 507 | |
63c882a0 | 508 | /* |
dead537d | 509 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 510 | */ |
528da3e9 EP |
511 | void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry, |
512 | struct fsnotify_group *group) | |
63c882a0 EP |
513 | { |
514 | struct inotify_inode_mark_entry *ientry; | |
f44aebcc | 515 | struct fsnotify_event *ignored_event; |
63c882a0 EP |
516 | struct inotify_event_private_data *event_priv; |
517 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 518 | int ret; |
63c882a0 | 519 | |
f44aebcc EP |
520 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
521 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
522 | GFP_NOFS); | |
523 | if (!ignored_event) | |
524 | return; | |
525 | ||
63c882a0 EP |
526 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
527 | ||
f44aebcc | 528 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
529 | if (unlikely(!event_priv)) |
530 | goto skip_send_ignore; | |
531 | ||
532 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
533 | ||
534 | fsn_event_priv->group = group; | |
535 | event_priv->wd = ientry->wd; | |
536 | ||
eef3a116 EP |
537 | ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv); |
538 | if (ret) | |
63c882a0 EP |
539 | inotify_free_event_priv(fsn_event_priv); |
540 | ||
541 | skip_send_ignore: | |
542 | ||
f44aebcc EP |
543 | /* matches the reference taken when the event was created */ |
544 | fsnotify_put_event(ignored_event); | |
545 | ||
63c882a0 | 546 | /* remove this entry from the idr */ |
7e790dd5 | 547 | inotify_remove_from_idr(group, ientry); |
63c882a0 | 548 | |
5549f7cd | 549 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
550 | } |
551 | ||
552 | /* ding dong the mark is dead */ | |
553 | static void inotify_free_mark(struct fsnotify_mark_entry *entry) | |
554 | { | |
31ddd326 EP |
555 | struct inotify_inode_mark_entry *ientry; |
556 | ||
557 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); | |
63c882a0 EP |
558 | |
559 | kmem_cache_free(inotify_inode_mark_cachep, ientry); | |
560 | } | |
561 | ||
52cef755 EP |
562 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
563 | struct inode *inode, | |
564 | u32 arg) | |
63c882a0 | 565 | { |
52cef755 | 566 | struct fsnotify_mark_entry *entry; |
63c882a0 | 567 | struct inotify_inode_mark_entry *ientry; |
63c882a0 | 568 | __u32 old_mask, new_mask; |
52cef755 EP |
569 | __u32 mask; |
570 | int add = (arg & IN_MASK_ADD); | |
571 | int ret; | |
63c882a0 EP |
572 | |
573 | /* don't allow invalid bits: we don't want flags set */ | |
574 | mask = inotify_arg_to_mask(arg); | |
575 | if (unlikely(!mask)) | |
576 | return -EINVAL; | |
577 | ||
63c882a0 EP |
578 | spin_lock(&inode->i_lock); |
579 | entry = fsnotify_find_mark_entry(group, inode); | |
580 | spin_unlock(&inode->i_lock); | |
52cef755 EP |
581 | if (!entry) |
582 | return -ENOENT; | |
7e790dd5 | 583 | |
52cef755 | 584 | ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry); |
75fe2b26 | 585 | |
63c882a0 EP |
586 | spin_lock(&entry->lock); |
587 | ||
588 | old_mask = entry->mask; | |
589 | if (add) { | |
590 | entry->mask |= mask; | |
591 | new_mask = entry->mask; | |
592 | } else { | |
593 | entry->mask = mask; | |
594 | new_mask = entry->mask; | |
595 | } | |
596 | ||
597 | spin_unlock(&entry->lock); | |
598 | ||
599 | if (old_mask != new_mask) { | |
600 | /* more bits in old than in new? */ | |
601 | int dropped = (old_mask & ~new_mask); | |
602 | /* more bits in this entry than the inode's mask? */ | |
603 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); | |
604 | /* more bits in this entry than the group? */ | |
605 | int do_group = (new_mask & ~group->mask); | |
606 | ||
607 | /* update the inode with this new entry */ | |
608 | if (dropped || do_inode) | |
609 | fsnotify_recalc_inode_mask(inode); | |
610 | ||
611 | /* update the group mask with the new mask */ | |
612 | if (dropped || do_group) | |
613 | fsnotify_recalc_group_mask(group); | |
614 | } | |
615 | ||
52cef755 EP |
616 | /* return the wd */ |
617 | ret = ientry->wd; | |
618 | ||
619 | /* match the get from fsnotify_find_mark_entry() */ | |
75fe2b26 EP |
620 | fsnotify_put_mark(entry); |
621 | ||
52cef755 EP |
622 | return ret; |
623 | } | |
624 | ||
625 | static int inotify_new_watch(struct fsnotify_group *group, | |
626 | struct inode *inode, | |
627 | u32 arg) | |
628 | { | |
629 | struct inotify_inode_mark_entry *tmp_ientry; | |
630 | __u32 mask; | |
631 | int ret; | |
b7ba8371 EP |
632 | struct idr *idr = &group->inotify_data.idr; |
633 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 EP |
634 | |
635 | /* don't allow invalid bits: we don't want flags set */ | |
636 | mask = inotify_arg_to_mask(arg); | |
637 | if (unlikely(!mask)) | |
638 | return -EINVAL; | |
639 | ||
640 | tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); | |
641 | if (unlikely(!tmp_ientry)) | |
642 | return -ENOMEM; | |
643 | ||
644 | fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark); | |
645 | tmp_ientry->fsn_entry.mask = mask; | |
646 | tmp_ientry->wd = -1; | |
647 | ||
648 | ret = -ENOSPC; | |
649 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
650 | goto out_err; | |
e0873344 | 651 | |
7050c488 | 652 | ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd, |
b7ba8371 EP |
653 | tmp_ientry); |
654 | if (ret) | |
52cef755 | 655 | goto out_err; |
52cef755 EP |
656 | |
657 | /* we are on the idr, now get on the inode */ | |
40554c3d | 658 | ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode, 0); |
52cef755 EP |
659 | if (ret) { |
660 | /* we failed to get on the inode, get off the idr */ | |
661 | inotify_remove_from_idr(group, tmp_ientry); | |
662 | goto out_err; | |
663 | } | |
664 | ||
52cef755 EP |
665 | /* increment the number of watches the user has */ |
666 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
667 | ||
668 | /* return the watch descriptor for this new entry */ | |
669 | ret = tmp_ientry->wd; | |
670 | ||
750a8870 EP |
671 | /* if this mark added a new event update the group mask */ |
672 | if (mask & ~group->mask) | |
673 | fsnotify_recalc_group_mask(group); | |
674 | ||
63c882a0 | 675 | out_err: |
3dbc6fb6 EP |
676 | /* match the ref from fsnotify_init_markentry() */ |
677 | fsnotify_put_mark(&tmp_ientry->fsn_entry); | |
52cef755 EP |
678 | |
679 | return ret; | |
680 | } | |
681 | ||
682 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
683 | { | |
684 | int ret = 0; | |
685 | ||
686 | retry: | |
687 | /* try to update and existing watch with the new arg */ | |
688 | ret = inotify_update_existing_watch(group, inode, arg); | |
689 | /* no mark present, try to add a new one */ | |
690 | if (ret == -ENOENT) | |
691 | ret = inotify_new_watch(group, inode, arg); | |
692 | /* | |
693 | * inotify_new_watch could race with another thread which did an | |
694 | * inotify_new_watch between the update_existing and the add watch | |
695 | * here, go back and try to update an existing mark again. | |
696 | */ | |
697 | if (ret == -EEXIST) | |
698 | goto retry; | |
7e790dd5 | 699 | |
63c882a0 EP |
700 | return ret; |
701 | } | |
702 | ||
703 | static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events) | |
704 | { | |
705 | struct fsnotify_group *group; | |
706 | unsigned int grp_num; | |
707 | ||
708 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ | |
709 | grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num)); | |
710 | group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops); | |
711 | if (IS_ERR(group)) | |
712 | return group; | |
713 | ||
714 | group->max_events = max_events; | |
715 | ||
716 | spin_lock_init(&group->inotify_data.idr_lock); | |
717 | idr_init(&group->inotify_data.idr); | |
9e572cc9 | 718 | group->inotify_data.last_wd = 0; |
63c882a0 EP |
719 | group->inotify_data.user = user; |
720 | group->inotify_data.fa = NULL; | |
721 | ||
722 | return group; | |
723 | } | |
724 | ||
725 | ||
726 | /* inotify syscalls */ | |
938bb9f5 | 727 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 728 | { |
63c882a0 | 729 | struct fsnotify_group *group; |
2d9048e2 | 730 | struct user_struct *user; |
c44dcc56 | 731 | int ret; |
2d9048e2 | 732 | |
e38b36f3 UD |
733 | /* Check the IN_* constants for consistency. */ |
734 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
735 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
736 | ||
510df2dd | 737 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
738 | return -EINVAL; |
739 | ||
da9592ed | 740 | user = get_current_user(); |
2d9048e2 AG |
741 | if (unlikely(atomic_read(&user->inotify_devs) >= |
742 | inotify_max_user_instances)) { | |
743 | ret = -EMFILE; | |
744 | goto out_free_uid; | |
745 | } | |
746 | ||
63c882a0 EP |
747 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
748 | group = inotify_new_group(user, inotify_max_queued_events); | |
749 | if (IS_ERR(group)) { | |
750 | ret = PTR_ERR(group); | |
2d9048e2 AG |
751 | goto out_free_uid; |
752 | } | |
753 | ||
825f9692 AV |
754 | atomic_inc(&user->inotify_devs); |
755 | ||
c44dcc56 AV |
756 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
757 | O_RDONLY | flags); | |
758 | if (ret >= 0) | |
759 | return ret; | |
825f9692 | 760 | |
825f9692 | 761 | atomic_dec(&user->inotify_devs); |
2d9048e2 AG |
762 | out_free_uid: |
763 | free_uid(user); | |
2d9048e2 AG |
764 | return ret; |
765 | } | |
766 | ||
938bb9f5 | 767 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
768 | { |
769 | return sys_inotify_init1(0); | |
770 | } | |
771 | ||
2e4d0924 HC |
772 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
773 | u32, mask) | |
2d9048e2 | 774 | { |
63c882a0 | 775 | struct fsnotify_group *group; |
2d9048e2 | 776 | struct inode *inode; |
2d8f3038 | 777 | struct path path; |
2d9048e2 AG |
778 | struct file *filp; |
779 | int ret, fput_needed; | |
780 | unsigned flags = 0; | |
781 | ||
782 | filp = fget_light(fd, &fput_needed); | |
783 | if (unlikely(!filp)) | |
784 | return -EBADF; | |
785 | ||
786 | /* verify that this is indeed an inotify instance */ | |
787 | if (unlikely(filp->f_op != &inotify_fops)) { | |
788 | ret = -EINVAL; | |
789 | goto fput_and_out; | |
790 | } | |
791 | ||
792 | if (!(mask & IN_DONT_FOLLOW)) | |
793 | flags |= LOOKUP_FOLLOW; | |
794 | if (mask & IN_ONLYDIR) | |
795 | flags |= LOOKUP_DIRECTORY; | |
796 | ||
63c882a0 EP |
797 | ret = inotify_find_inode(pathname, &path, flags); |
798 | if (ret) | |
2d9048e2 AG |
799 | goto fput_and_out; |
800 | ||
63c882a0 | 801 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 802 | inode = path.dentry->d_inode; |
63c882a0 | 803 | group = filp->private_data; |
2d9048e2 | 804 | |
63c882a0 EP |
805 | /* create/update an inode mark */ |
806 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 807 | path_put(&path); |
2d9048e2 AG |
808 | fput_and_out: |
809 | fput_light(filp, fput_needed); | |
810 | return ret; | |
811 | } | |
812 | ||
2e4d0924 | 813 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 814 | { |
63c882a0 | 815 | struct fsnotify_group *group; |
b7ba8371 | 816 | struct inotify_inode_mark_entry *ientry; |
2d9048e2 | 817 | struct file *filp; |
63c882a0 | 818 | int ret = 0, fput_needed; |
2d9048e2 AG |
819 | |
820 | filp = fget_light(fd, &fput_needed); | |
821 | if (unlikely(!filp)) | |
822 | return -EBADF; | |
823 | ||
824 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 EP |
825 | ret = -EINVAL; |
826 | if (unlikely(filp->f_op != &inotify_fops)) | |
2d9048e2 | 827 | goto out; |
2d9048e2 | 828 | |
63c882a0 | 829 | group = filp->private_data; |
2d9048e2 | 830 | |
b7ba8371 EP |
831 | ret = -EINVAL; |
832 | ientry = inotify_idr_find(group, wd); | |
833 | if (unlikely(!ientry)) | |
63c882a0 | 834 | goto out; |
63c882a0 | 835 | |
b7ba8371 EP |
836 | ret = 0; |
837 | ||
838 | fsnotify_destroy_mark_by_entry(&ientry->fsn_entry); | |
839 | ||
840 | /* match ref taken by inotify_idr_find */ | |
841 | fsnotify_put_mark(&ientry->fsn_entry); | |
2d9048e2 AG |
842 | |
843 | out: | |
844 | fput_light(filp, fput_needed); | |
845 | return ret; | |
846 | } | |
847 | ||
2d9048e2 AG |
848 | /* |
849 | * inotify_user_setup - Our initialization function. Note that we cannnot return | |
850 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here | |
851 | * must result in panic(). | |
852 | */ | |
853 | static int __init inotify_user_setup(void) | |
854 | { | |
63c882a0 EP |
855 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC); |
856 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); | |
63c882a0 | 857 | |
2d9048e2 AG |
858 | inotify_max_queued_events = 16384; |
859 | inotify_max_user_instances = 128; | |
860 | inotify_max_user_watches = 8192; | |
861 | ||
2d9048e2 AG |
862 | return 0; |
863 | } | |
2d9048e2 | 864 | module_init(inotify_user_setup); |