]>
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; |
0a24887a | 49 | static 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 | 53 | |
2d9048e2 AG |
54 | #ifdef CONFIG_SYSCTL |
55 | ||
56 | #include <linux/sysctl.h> | |
57 | ||
58 | static int zero; | |
59 | ||
60 | ctl_table inotify_table[] = { | |
61 | { | |
2d9048e2 AG |
62 | .procname = "max_user_instances", |
63 | .data = &inotify_max_user_instances, | |
64 | .maxlen = sizeof(int), | |
65 | .mode = 0644, | |
6d456111 | 66 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
67 | .extra1 = &zero, |
68 | }, | |
69 | { | |
2d9048e2 AG |
70 | .procname = "max_user_watches", |
71 | .data = &inotify_max_user_watches, | |
72 | .maxlen = sizeof(int), | |
73 | .mode = 0644, | |
6d456111 | 74 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
75 | .extra1 = &zero, |
76 | }, | |
77 | { | |
2d9048e2 AG |
78 | .procname = "max_queued_events", |
79 | .data = &inotify_max_queued_events, | |
80 | .maxlen = sizeof(int), | |
81 | .mode = 0644, | |
6d456111 | 82 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
83 | .extra1 = &zero |
84 | }, | |
ab09203e | 85 | { } |
2d9048e2 AG |
86 | }; |
87 | #endif /* CONFIG_SYSCTL */ | |
88 | ||
63c882a0 | 89 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 90 | { |
63c882a0 | 91 | __u32 mask; |
1c17d18e | 92 | |
611da04f EP |
93 | /* |
94 | * everything should accept their own ignored, cares about children, | |
95 | * and should receive events when the inode is unmounted | |
96 | */ | |
97 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); | |
2d9048e2 | 98 | |
63c882a0 | 99 | /* mask off the flags used to open the fd */ |
8c1934c8 | 100 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); |
2d9048e2 | 101 | |
63c882a0 | 102 | return mask; |
2d9048e2 AG |
103 | } |
104 | ||
63c882a0 | 105 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 106 | { |
63c882a0 EP |
107 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
108 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
109 | } |
110 | ||
63c882a0 | 111 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
112 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
113 | { | |
63c882a0 | 114 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
115 | int ret = 0; |
116 | ||
63c882a0 EP |
117 | poll_wait(file, &group->notification_waitq, wait); |
118 | mutex_lock(&group->notification_mutex); | |
119 | if (!fsnotify_notify_queue_is_empty(group)) | |
2d9048e2 | 120 | ret = POLLIN | POLLRDNORM; |
63c882a0 | 121 | mutex_unlock(&group->notification_mutex); |
2d9048e2 AG |
122 | |
123 | return ret; | |
124 | } | |
125 | ||
3632dee2 VN |
126 | /* |
127 | * Get an inotify_kernel_event if one exists and is small | |
128 | * enough to fit in "count". Return an error pointer if | |
129 | * not large enough. | |
130 | * | |
63c882a0 | 131 | * Called with the group->notification_mutex held. |
3632dee2 | 132 | */ |
63c882a0 EP |
133 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
134 | size_t count) | |
3632dee2 VN |
135 | { |
136 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 137 | struct fsnotify_event *event; |
3632dee2 | 138 | |
63c882a0 | 139 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
140 | return NULL; |
141 | ||
63c882a0 EP |
142 | event = fsnotify_peek_notify_event(group); |
143 | ||
5ba08e2e EP |
144 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
145 | ||
83cb10f0 EP |
146 | if (event->name_len) |
147 | event_size += roundup(event->name_len + 1, event_size); | |
3632dee2 VN |
148 | |
149 | if (event_size > count) | |
150 | return ERR_PTR(-EINVAL); | |
151 | ||
63c882a0 EP |
152 | /* held the notification_mutex the whole time, so this is the |
153 | * same event we peeked above */ | |
154 | fsnotify_remove_notify_event(group); | |
155 | ||
156 | return event; | |
3632dee2 VN |
157 | } |
158 | ||
159 | /* | |
160 | * Copy an event to user space, returning how much we copied. | |
161 | * | |
162 | * We already checked that the event size is smaller than the | |
163 | * buffer we had in "get_one_event()" above. | |
164 | */ | |
63c882a0 EP |
165 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
166 | struct fsnotify_event *event, | |
3632dee2 VN |
167 | char __user *buf) |
168 | { | |
63c882a0 EP |
169 | struct inotify_event inotify_event; |
170 | struct fsnotify_event_private_data *fsn_priv; | |
171 | struct inotify_event_private_data *priv; | |
3632dee2 | 172 | size_t event_size = sizeof(struct inotify_event); |
b962e731 | 173 | size_t name_len = 0; |
63c882a0 | 174 | |
5ba08e2e EP |
175 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
176 | ||
63c882a0 EP |
177 | /* we get the inotify watch descriptor from the event private data */ |
178 | spin_lock(&event->lock); | |
179 | fsn_priv = fsnotify_remove_priv_from_event(group, event); | |
180 | spin_unlock(&event->lock); | |
181 | ||
182 | if (!fsn_priv) | |
183 | inotify_event.wd = -1; | |
184 | else { | |
185 | priv = container_of(fsn_priv, struct inotify_event_private_data, | |
186 | fsnotify_event_priv_data); | |
187 | inotify_event.wd = priv->wd; | |
188 | inotify_free_event_priv(fsn_priv); | |
189 | } | |
190 | ||
b962e731 BR |
191 | /* |
192 | * round up event->name_len so it is a multiple of event_size | |
0db501bd EB |
193 | * plus an extra byte for the terminating '\0'. |
194 | */ | |
b962e731 BR |
195 | if (event->name_len) |
196 | name_len = roundup(event->name_len + 1, event_size); | |
63c882a0 EP |
197 | inotify_event.len = name_len; |
198 | ||
199 | inotify_event.mask = inotify_mask_to_arg(event->mask); | |
200 | inotify_event.cookie = event->sync_cookie; | |
3632dee2 | 201 | |
63c882a0 EP |
202 | /* send the main event */ |
203 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
204 | return -EFAULT; |
205 | ||
63c882a0 | 206 | buf += event_size; |
3632dee2 | 207 | |
63c882a0 EP |
208 | /* |
209 | * fsnotify only stores the pathname, so here we have to send the pathname | |
210 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
211 | * with zeros. I get my zeros from the nul_inotify_event. | |
212 | */ | |
213 | if (name_len) { | |
214 | unsigned int len_to_zero = name_len - event->name_len; | |
215 | /* copy the path name */ | |
216 | if (copy_to_user(buf, event->file_name, event->name_len)) | |
3632dee2 | 217 | return -EFAULT; |
63c882a0 | 218 | buf += event->name_len; |
3632dee2 | 219 | |
0db501bd EB |
220 | /* fill userspace with 0's */ |
221 | if (clear_user(buf, len_to_zero)) | |
63c882a0 EP |
222 | return -EFAULT; |
223 | buf += len_to_zero; | |
224 | event_size += name_len; | |
3632dee2 | 225 | } |
63c882a0 | 226 | |
3632dee2 VN |
227 | return event_size; |
228 | } | |
229 | ||
2d9048e2 AG |
230 | static ssize_t inotify_read(struct file *file, char __user *buf, |
231 | size_t count, loff_t *pos) | |
232 | { | |
63c882a0 EP |
233 | struct fsnotify_group *group; |
234 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
235 | char __user *start; |
236 | int ret; | |
237 | DEFINE_WAIT(wait); | |
238 | ||
239 | start = buf; | |
63c882a0 | 240 | group = file->private_data; |
2d9048e2 AG |
241 | |
242 | while (1) { | |
63c882a0 | 243 | prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); |
2d9048e2 | 244 | |
63c882a0 EP |
245 | mutex_lock(&group->notification_mutex); |
246 | kevent = get_one_event(group, count); | |
247 | mutex_unlock(&group->notification_mutex); | |
2d9048e2 | 248 | |
5ba08e2e EP |
249 | pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); |
250 | ||
3632dee2 VN |
251 | if (kevent) { |
252 | ret = PTR_ERR(kevent); | |
253 | if (IS_ERR(kevent)) | |
254 | break; | |
63c882a0 EP |
255 | ret = copy_event_to_user(group, kevent, buf); |
256 | fsnotify_put_event(kevent); | |
3632dee2 VN |
257 | if (ret < 0) |
258 | break; | |
259 | buf += ret; | |
260 | count -= ret; | |
261 | continue; | |
2d9048e2 AG |
262 | } |
263 | ||
3632dee2 VN |
264 | ret = -EAGAIN; |
265 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 266 | break; |
3632dee2 VN |
267 | ret = -EINTR; |
268 | if (signal_pending(current)) | |
2d9048e2 | 269 | break; |
16dbc6c9 | 270 | |
3632dee2 | 271 | if (start != buf) |
2d9048e2 | 272 | break; |
16dbc6c9 | 273 | |
3632dee2 | 274 | schedule(); |
2d9048e2 | 275 | } |
2d9048e2 | 276 | |
63c882a0 | 277 | finish_wait(&group->notification_waitq, &wait); |
3632dee2 VN |
278 | if (start != buf && ret != -EFAULT) |
279 | ret = buf - start; | |
2d9048e2 AG |
280 | return ret; |
281 | } | |
282 | ||
bcfbf84d DA |
283 | static int inotify_fasync(int fd, struct file *file, int on) |
284 | { | |
63c882a0 | 285 | struct fsnotify_group *group = file->private_data; |
bcfbf84d | 286 | |
63c882a0 | 287 | return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; |
bcfbf84d DA |
288 | } |
289 | ||
2d9048e2 AG |
290 | static int inotify_release(struct inode *ignored, struct file *file) |
291 | { | |
63c882a0 | 292 | struct fsnotify_group *group = file->private_data; |
2d9048e2 | 293 | |
5ba08e2e EP |
294 | pr_debug("%s: group=%p\n", __func__, group); |
295 | ||
63c882a0 | 296 | fsnotify_clear_marks_by_group(group); |
2d9048e2 | 297 | |
63c882a0 EP |
298 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
299 | fsnotify_put_group(group); | |
2d9048e2 AG |
300 | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
305 | unsigned long arg) | |
306 | { | |
63c882a0 EP |
307 | struct fsnotify_group *group; |
308 | struct fsnotify_event_holder *holder; | |
309 | struct fsnotify_event *event; | |
2d9048e2 AG |
310 | void __user *p; |
311 | int ret = -ENOTTY; | |
63c882a0 | 312 | size_t send_len = 0; |
2d9048e2 | 313 | |
63c882a0 | 314 | group = file->private_data; |
2d9048e2 AG |
315 | p = (void __user *) arg; |
316 | ||
5ba08e2e EP |
317 | pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); |
318 | ||
2d9048e2 AG |
319 | switch (cmd) { |
320 | case FIONREAD: | |
63c882a0 EP |
321 | mutex_lock(&group->notification_mutex); |
322 | list_for_each_entry(holder, &group->notification_list, event_list) { | |
323 | event = holder->event; | |
324 | send_len += sizeof(struct inotify_event); | |
83cb10f0 EP |
325 | if (event->name_len) |
326 | send_len += roundup(event->name_len + 1, | |
327 | sizeof(struct inotify_event)); | |
63c882a0 EP |
328 | } |
329 | mutex_unlock(&group->notification_mutex); | |
330 | ret = put_user(send_len, (int __user *) p); | |
2d9048e2 AG |
331 | break; |
332 | } | |
333 | ||
334 | return ret; | |
335 | } | |
336 | ||
337 | static const struct file_operations inotify_fops = { | |
63c882a0 EP |
338 | .poll = inotify_poll, |
339 | .read = inotify_read, | |
340 | .fasync = inotify_fasync, | |
341 | .release = inotify_release, | |
342 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 | 343 | .compat_ioctl = inotify_ioctl, |
6038f373 | 344 | .llseek = noop_llseek, |
2d9048e2 AG |
345 | }; |
346 | ||
2d9048e2 | 347 | |
63c882a0 EP |
348 | /* |
349 | * find_inode - resolve a user-given path to a specific inode | |
350 | */ | |
351 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
352 | { | |
353 | int error; | |
354 | ||
355 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
356 | if (error) | |
357 | return error; | |
358 | /* you can only watch an inode if you have read permissions on it */ | |
359 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
360 | if (error) | |
361 | path_put(path); | |
362 | return error; | |
363 | } | |
364 | ||
b7ba8371 | 365 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
7050c488 | 366 | int *last_wd, |
000285de | 367 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
368 | { |
369 | int ret; | |
370 | ||
371 | do { | |
372 | if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) | |
373 | return -ENOMEM; | |
374 | ||
375 | spin_lock(idr_lock); | |
000285de EP |
376 | ret = idr_get_new_above(idr, i_mark, *last_wd + 1, |
377 | &i_mark->wd); | |
b7ba8371 | 378 | /* we added the mark to the idr, take a reference */ |
7050c488 | 379 | if (!ret) { |
000285de EP |
380 | *last_wd = i_mark->wd; |
381 | fsnotify_get_mark(&i_mark->fsn_mark); | |
7050c488 | 382 | } |
b7ba8371 EP |
383 | spin_unlock(idr_lock); |
384 | } while (ret == -EAGAIN); | |
385 | ||
386 | return ret; | |
387 | } | |
388 | ||
000285de | 389 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, |
b7ba8371 EP |
390 | int wd) |
391 | { | |
392 | struct idr *idr = &group->inotify_data.idr; | |
393 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 394 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
395 | |
396 | assert_spin_locked(idr_lock); | |
397 | ||
000285de EP |
398 | i_mark = idr_find(idr, wd); |
399 | if (i_mark) { | |
400 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
b7ba8371 | 401 | |
000285de | 402 | fsnotify_get_mark(fsn_mark); |
b7ba8371 | 403 | /* One ref for being in the idr, one ref we just took */ |
000285de | 404 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); |
b7ba8371 EP |
405 | } |
406 | ||
000285de | 407 | return i_mark; |
b7ba8371 EP |
408 | } |
409 | ||
000285de | 410 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, |
b7ba8371 EP |
411 | int wd) |
412 | { | |
000285de | 413 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
414 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
415 | ||
416 | spin_lock(idr_lock); | |
000285de | 417 | i_mark = inotify_idr_find_locked(group, wd); |
b7ba8371 EP |
418 | spin_unlock(idr_lock); |
419 | ||
000285de | 420 | return i_mark; |
b7ba8371 EP |
421 | } |
422 | ||
423 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
000285de | 424 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
425 | { |
426 | struct idr *idr = &group->inotify_data.idr; | |
427 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 428 | int wd = i_mark->wd; |
b7ba8371 EP |
429 | |
430 | assert_spin_locked(idr_lock); | |
431 | ||
432 | idr_remove(idr, wd); | |
433 | ||
434 | /* removed from the idr, drop that ref */ | |
000285de | 435 | fsnotify_put_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
436 | } |
437 | ||
dead537d EP |
438 | /* |
439 | * Remove the mark from the idr (if present) and drop the reference | |
440 | * on the mark because it was in the idr. | |
441 | */ | |
7e790dd5 | 442 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
000285de | 443 | struct inotify_inode_mark *i_mark) |
7e790dd5 | 444 | { |
b7ba8371 | 445 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
000285de | 446 | struct inotify_inode_mark *found_i_mark = NULL; |
dead537d | 447 | int wd; |
7e790dd5 | 448 | |
b7ba8371 | 449 | spin_lock(idr_lock); |
000285de | 450 | wd = i_mark->wd; |
dead537d | 451 | |
b7ba8371 | 452 | /* |
000285de | 453 | * does this i_mark think it is in the idr? we shouldn't get called |
b7ba8371 EP |
454 | * if it wasn't.... |
455 | */ | |
456 | if (wd == -1) { | |
000285de EP |
457 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" |
458 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
459 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 460 | goto out; |
b7ba8371 | 461 | } |
dead537d | 462 | |
b7ba8371 | 463 | /* Lets look in the idr to see if we find it */ |
000285de EP |
464 | found_i_mark = inotify_idr_find_locked(group, wd); |
465 | if (unlikely(!found_i_mark)) { | |
466 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
467 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
468 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
dead537d | 469 | goto out; |
b7ba8371 | 470 | } |
dead537d | 471 | |
b7ba8371 | 472 | /* |
000285de EP |
473 | * We found an mark in the idr at the right wd, but it's |
474 | * not the mark we were told to remove. eparis seriously | |
b7ba8371 EP |
475 | * fucked up somewhere. |
476 | */ | |
000285de EP |
477 | if (unlikely(found_i_mark != i_mark)) { |
478 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
479 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
480 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
481 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
482 | i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, | |
483 | found_i_mark->fsn_mark.group, | |
484 | found_i_mark->fsn_mark.i.inode); | |
dead537d EP |
485 | goto out; |
486 | } | |
487 | ||
b7ba8371 EP |
488 | /* |
489 | * One ref for being in the idr | |
490 | * one ref held by the caller trying to kill us | |
491 | * one ref grabbed by inotify_idr_find | |
492 | */ | |
000285de EP |
493 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { |
494 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
495 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
496 | i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); | |
b7ba8371 EP |
497 | /* we can't really recover with bad ref cnting.. */ |
498 | BUG(); | |
499 | } | |
dead537d | 500 | |
000285de | 501 | do_inotify_remove_from_idr(group, i_mark); |
dead537d | 502 | out: |
b7ba8371 | 503 | /* match the ref taken by inotify_idr_find_locked() */ |
000285de EP |
504 | if (found_i_mark) |
505 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
506 | i_mark->wd = -1; | |
b7ba8371 | 507 | spin_unlock(idr_lock); |
7e790dd5 | 508 | } |
dead537d | 509 | |
63c882a0 | 510 | /* |
dead537d | 511 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 512 | */ |
000285de | 513 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, |
528da3e9 | 514 | struct fsnotify_group *group) |
63c882a0 | 515 | { |
000285de | 516 | struct inotify_inode_mark *i_mark; |
f70ab54c | 517 | struct fsnotify_event *ignored_event, *notify_event; |
63c882a0 EP |
518 | struct inotify_event_private_data *event_priv; |
519 | struct fsnotify_event_private_data *fsn_event_priv; | |
eef3a116 | 520 | int ret; |
63c882a0 | 521 | |
f44aebcc EP |
522 | ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, |
523 | FSNOTIFY_EVENT_NONE, NULL, 0, | |
524 | GFP_NOFS); | |
525 | if (!ignored_event) | |
526 | return; | |
527 | ||
000285de | 528 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 529 | |
f44aebcc | 530 | event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); |
63c882a0 EP |
531 | if (unlikely(!event_priv)) |
532 | goto skip_send_ignore; | |
533 | ||
534 | fsn_event_priv = &event_priv->fsnotify_event_priv_data; | |
535 | ||
536 | fsn_event_priv->group = group; | |
000285de | 537 | event_priv->wd = i_mark->wd; |
63c882a0 | 538 | |
f70ab54c EP |
539 | notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); |
540 | if (notify_event) { | |
541 | if (IS_ERR(notify_event)) | |
542 | ret = PTR_ERR(notify_event); | |
543 | else | |
544 | fsnotify_put_event(notify_event); | |
63c882a0 | 545 | inotify_free_event_priv(fsn_event_priv); |
f70ab54c | 546 | } |
63c882a0 EP |
547 | |
548 | skip_send_ignore: | |
549 | ||
f44aebcc EP |
550 | /* matches the reference taken when the event was created */ |
551 | fsnotify_put_event(ignored_event); | |
552 | ||
000285de EP |
553 | /* remove this mark from the idr */ |
554 | inotify_remove_from_idr(group, i_mark); | |
63c882a0 | 555 | |
5549f7cd | 556 | atomic_dec(&group->inotify_data.user->inotify_watches); |
63c882a0 EP |
557 | } |
558 | ||
559 | /* ding dong the mark is dead */ | |
000285de | 560 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) |
63c882a0 | 561 | { |
000285de | 562 | struct inotify_inode_mark *i_mark; |
31ddd326 | 563 | |
000285de | 564 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 565 | |
000285de | 566 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); |
63c882a0 EP |
567 | } |
568 | ||
52cef755 EP |
569 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
570 | struct inode *inode, | |
571 | u32 arg) | |
63c882a0 | 572 | { |
000285de EP |
573 | struct fsnotify_mark *fsn_mark; |
574 | struct inotify_inode_mark *i_mark; | |
63c882a0 | 575 | __u32 old_mask, new_mask; |
52cef755 EP |
576 | __u32 mask; |
577 | int add = (arg & IN_MASK_ADD); | |
578 | int ret; | |
63c882a0 EP |
579 | |
580 | /* don't allow invalid bits: we don't want flags set */ | |
581 | mask = inotify_arg_to_mask(arg); | |
44b350fc | 582 | if (unlikely(!(mask & IN_ALL_EVENTS))) |
63c882a0 EP |
583 | return -EINVAL; |
584 | ||
5444e298 | 585 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
000285de | 586 | if (!fsn_mark) |
52cef755 | 587 | return -ENOENT; |
7e790dd5 | 588 | |
000285de | 589 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
75fe2b26 | 590 | |
000285de | 591 | spin_lock(&fsn_mark->lock); |
63c882a0 | 592 | |
000285de | 593 | old_mask = fsn_mark->mask; |
90b1e7a5 EP |
594 | if (add) |
595 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
596 | else | |
597 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
598 | new_mask = fsn_mark->mask; | |
63c882a0 | 599 | |
000285de | 600 | spin_unlock(&fsn_mark->lock); |
63c882a0 EP |
601 | |
602 | if (old_mask != new_mask) { | |
603 | /* more bits in old than in new? */ | |
604 | int dropped = (old_mask & ~new_mask); | |
000285de | 605 | /* more bits in this fsn_mark than the inode's mask? */ |
63c882a0 | 606 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); |
63c882a0 | 607 | |
000285de | 608 | /* update the inode with this new fsn_mark */ |
63c882a0 EP |
609 | if (dropped || do_inode) |
610 | fsnotify_recalc_inode_mask(inode); | |
611 | ||
63c882a0 EP |
612 | } |
613 | ||
52cef755 | 614 | /* return the wd */ |
000285de | 615 | ret = i_mark->wd; |
52cef755 | 616 | |
d0775441 | 617 | /* match the get from fsnotify_find_mark() */ |
000285de | 618 | fsnotify_put_mark(fsn_mark); |
75fe2b26 | 619 | |
52cef755 EP |
620 | return ret; |
621 | } | |
622 | ||
623 | static int inotify_new_watch(struct fsnotify_group *group, | |
624 | struct inode *inode, | |
625 | u32 arg) | |
626 | { | |
000285de | 627 | struct inotify_inode_mark *tmp_i_mark; |
52cef755 EP |
628 | __u32 mask; |
629 | int ret; | |
b7ba8371 EP |
630 | struct idr *idr = &group->inotify_data.idr; |
631 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 EP |
632 | |
633 | /* don't allow invalid bits: we don't want flags set */ | |
634 | mask = inotify_arg_to_mask(arg); | |
44b350fc | 635 | if (unlikely(!(mask & IN_ALL_EVENTS))) |
52cef755 EP |
636 | return -EINVAL; |
637 | ||
000285de EP |
638 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); |
639 | if (unlikely(!tmp_i_mark)) | |
52cef755 EP |
640 | return -ENOMEM; |
641 | ||
000285de EP |
642 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); |
643 | tmp_i_mark->fsn_mark.mask = mask; | |
644 | tmp_i_mark->wd = -1; | |
52cef755 EP |
645 | |
646 | ret = -ENOSPC; | |
647 | if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) | |
648 | goto out_err; | |
e0873344 | 649 | |
7050c488 | 650 | ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd, |
000285de | 651 | tmp_i_mark); |
b7ba8371 | 652 | if (ret) |
52cef755 | 653 | goto out_err; |
52cef755 EP |
654 | |
655 | /* we are on the idr, now get on the inode */ | |
5444e298 | 656 | ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0); |
52cef755 EP |
657 | if (ret) { |
658 | /* we failed to get on the inode, get off the idr */ | |
000285de | 659 | inotify_remove_from_idr(group, tmp_i_mark); |
52cef755 EP |
660 | goto out_err; |
661 | } | |
662 | ||
52cef755 EP |
663 | /* increment the number of watches the user has */ |
664 | atomic_inc(&group->inotify_data.user->inotify_watches); | |
665 | ||
000285de EP |
666 | /* return the watch descriptor for this new mark */ |
667 | ret = tmp_i_mark->wd; | |
52cef755 | 668 | |
63c882a0 | 669 | out_err: |
000285de EP |
670 | /* match the ref from fsnotify_init_mark() */ |
671 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
52cef755 EP |
672 | |
673 | return ret; | |
674 | } | |
675 | ||
676 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
677 | { | |
678 | int ret = 0; | |
679 | ||
680 | retry: | |
681 | /* try to update and existing watch with the new arg */ | |
682 | ret = inotify_update_existing_watch(group, inode, arg); | |
683 | /* no mark present, try to add a new one */ | |
684 | if (ret == -ENOENT) | |
685 | ret = inotify_new_watch(group, inode, arg); | |
686 | /* | |
687 | * inotify_new_watch could race with another thread which did an | |
688 | * inotify_new_watch between the update_existing and the add watch | |
689 | * here, go back and try to update an existing mark again. | |
690 | */ | |
691 | if (ret == -EEXIST) | |
692 | goto retry; | |
7e790dd5 | 693 | |
63c882a0 EP |
694 | return ret; |
695 | } | |
696 | ||
d0de4dc5 | 697 | static struct fsnotify_group *inotify_new_group(unsigned int max_events) |
63c882a0 EP |
698 | { |
699 | struct fsnotify_group *group; | |
63c882a0 | 700 | |
0d2e2a1d | 701 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
702 | if (IS_ERR(group)) |
703 | return group; | |
704 | ||
705 | group->max_events = max_events; | |
706 | ||
707 | spin_lock_init(&group->inotify_data.idr_lock); | |
708 | idr_init(&group->inotify_data.idr); | |
9e572cc9 | 709 | group->inotify_data.last_wd = 0; |
63c882a0 | 710 | group->inotify_data.fa = NULL; |
d0de4dc5 EP |
711 | group->inotify_data.user = get_current_user(); |
712 | ||
713 | if (atomic_inc_return(&group->inotify_data.user->inotify_devs) > | |
714 | inotify_max_user_instances) { | |
715 | fsnotify_put_group(group); | |
716 | return ERR_PTR(-EMFILE); | |
717 | } | |
63c882a0 EP |
718 | |
719 | return group; | |
720 | } | |
721 | ||
722 | ||
723 | /* inotify syscalls */ | |
938bb9f5 | 724 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 725 | { |
63c882a0 | 726 | struct fsnotify_group *group; |
c44dcc56 | 727 | int ret; |
2d9048e2 | 728 | |
e38b36f3 UD |
729 | /* Check the IN_* constants for consistency. */ |
730 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
731 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
732 | ||
510df2dd | 733 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
734 | return -EINVAL; |
735 | ||
63c882a0 | 736 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
d0de4dc5 EP |
737 | group = inotify_new_group(inotify_max_queued_events); |
738 | if (IS_ERR(group)) | |
739 | return PTR_ERR(group); | |
825f9692 | 740 | |
c44dcc56 AV |
741 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
742 | O_RDONLY | flags); | |
d0de4dc5 EP |
743 | if (ret < 0) |
744 | fsnotify_put_group(group); | |
825f9692 | 745 | |
2d9048e2 AG |
746 | return ret; |
747 | } | |
748 | ||
938bb9f5 | 749 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
750 | { |
751 | return sys_inotify_init1(0); | |
752 | } | |
753 | ||
2e4d0924 HC |
754 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
755 | u32, mask) | |
2d9048e2 | 756 | { |
63c882a0 | 757 | struct fsnotify_group *group; |
2d9048e2 | 758 | struct inode *inode; |
2d8f3038 | 759 | struct path path; |
2903ff01 AV |
760 | struct fd f; |
761 | int ret; | |
2d9048e2 AG |
762 | unsigned flags = 0; |
763 | ||
2903ff01 AV |
764 | f = fdget(fd); |
765 | if (unlikely(!f.file)) | |
2d9048e2 AG |
766 | return -EBADF; |
767 | ||
768 | /* verify that this is indeed an inotify instance */ | |
2903ff01 | 769 | if (unlikely(f.file->f_op != &inotify_fops)) { |
2d9048e2 AG |
770 | ret = -EINVAL; |
771 | goto fput_and_out; | |
772 | } | |
773 | ||
774 | if (!(mask & IN_DONT_FOLLOW)) | |
775 | flags |= LOOKUP_FOLLOW; | |
776 | if (mask & IN_ONLYDIR) | |
777 | flags |= LOOKUP_DIRECTORY; | |
778 | ||
63c882a0 EP |
779 | ret = inotify_find_inode(pathname, &path, flags); |
780 | if (ret) | |
2d9048e2 AG |
781 | goto fput_and_out; |
782 | ||
63c882a0 | 783 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 784 | inode = path.dentry->d_inode; |
2903ff01 | 785 | group = f.file->private_data; |
2d9048e2 | 786 | |
63c882a0 EP |
787 | /* create/update an inode mark */ |
788 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 789 | path_put(&path); |
2d9048e2 | 790 | fput_and_out: |
2903ff01 | 791 | fdput(f); |
2d9048e2 AG |
792 | return ret; |
793 | } | |
794 | ||
2e4d0924 | 795 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 796 | { |
63c882a0 | 797 | struct fsnotify_group *group; |
000285de | 798 | struct inotify_inode_mark *i_mark; |
2903ff01 AV |
799 | struct fd f; |
800 | int ret = 0; | |
2d9048e2 | 801 | |
2903ff01 AV |
802 | f = fdget(fd); |
803 | if (unlikely(!f.file)) | |
2d9048e2 AG |
804 | return -EBADF; |
805 | ||
806 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 | 807 | ret = -EINVAL; |
2903ff01 | 808 | if (unlikely(f.file->f_op != &inotify_fops)) |
2d9048e2 | 809 | goto out; |
2d9048e2 | 810 | |
2903ff01 | 811 | group = f.file->private_data; |
2d9048e2 | 812 | |
b7ba8371 | 813 | ret = -EINVAL; |
000285de EP |
814 | i_mark = inotify_idr_find(group, wd); |
815 | if (unlikely(!i_mark)) | |
63c882a0 | 816 | goto out; |
63c882a0 | 817 | |
b7ba8371 EP |
818 | ret = 0; |
819 | ||
000285de | 820 | fsnotify_destroy_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
821 | |
822 | /* match ref taken by inotify_idr_find */ | |
000285de | 823 | fsnotify_put_mark(&i_mark->fsn_mark); |
2d9048e2 AG |
824 | |
825 | out: | |
2903ff01 | 826 | fdput(f); |
2d9048e2 AG |
827 | return ret; |
828 | } | |
829 | ||
2d9048e2 | 830 | /* |
ae0e47f0 | 831 | * inotify_user_setup - Our initialization function. Note that we cannot return |
2d9048e2 AG |
832 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
833 | * must result in panic(). | |
834 | */ | |
835 | static int __init inotify_user_setup(void) | |
836 | { | |
f874e1ac EP |
837 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); |
838 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
839 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
840 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
841 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
842 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
843 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
844 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
845 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
846 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
847 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
848 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
849 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
850 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
851 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
852 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
b29866aa | 853 | BUILD_BUG_ON(IN_ISDIR != FS_ISDIR); |
f874e1ac EP |
854 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); |
855 | ||
856 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
857 | ||
000285de | 858 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); |
63c882a0 | 859 | event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); |
63c882a0 | 860 | |
2d9048e2 AG |
861 | inotify_max_queued_events = 16384; |
862 | inotify_max_user_instances = 128; | |
863 | inotify_max_user_watches = 8192; | |
864 | ||
2d9048e2 AG |
865 | return 0; |
866 | } | |
2d9048e2 | 867 | module_init(inotify_user_setup); |