]>
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> | |
c013d5a4 | 29 | #include <linux/init.h> /* fs_initcall */ |
2d9048e2 | 30 | #include <linux/inotify.h> |
63c882a0 | 31 | #include <linux/kernel.h> /* roundup() */ |
63c882a0 | 32 | #include <linux/namei.h> /* LOOKUP_FOLLOW */ |
174cd4b1 | 33 | #include <linux/sched/signal.h> |
63c882a0 | 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" |
be77196b | 43 | #include "../fdinfo.h" |
2d9048e2 | 44 | |
63c882a0 | 45 | #include <asm/ioctls.h> |
2d9048e2 | 46 | |
1cce1eea | 47 | /* configurable via /proc/sys/fs/inotify/ */ |
3c828e49 | 48 | static int inotify_max_queued_events __read_mostly; |
2d9048e2 | 49 | |
63c882a0 | 50 | static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; |
2d9048e2 | 51 | |
2d9048e2 AG |
52 | #ifdef CONFIG_SYSCTL |
53 | ||
54 | #include <linux/sysctl.h> | |
55 | ||
56 | static int zero; | |
57 | ||
92f778dd | 58 | struct ctl_table inotify_table[] = { |
2d9048e2 | 59 | { |
2d9048e2 | 60 | .procname = "max_user_instances", |
1cce1eea | 61 | .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES], |
2d9048e2 AG |
62 | .maxlen = sizeof(int), |
63 | .mode = 0644, | |
6d456111 | 64 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
65 | .extra1 = &zero, |
66 | }, | |
67 | { | |
2d9048e2 | 68 | .procname = "max_user_watches", |
1cce1eea | 69 | .data = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES], |
2d9048e2 AG |
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_queued_events", |
77 | .data = &inotify_max_queued_events, | |
78 | .maxlen = sizeof(int), | |
79 | .mode = 0644, | |
6d456111 | 80 | .proc_handler = proc_dointvec_minmax, |
2d9048e2 AG |
81 | .extra1 = &zero |
82 | }, | |
ab09203e | 83 | { } |
2d9048e2 AG |
84 | }; |
85 | #endif /* CONFIG_SYSCTL */ | |
86 | ||
63c882a0 | 87 | static inline __u32 inotify_arg_to_mask(u32 arg) |
1c17d18e | 88 | { |
63c882a0 | 89 | __u32 mask; |
1c17d18e | 90 | |
611da04f EP |
91 | /* |
92 | * everything should accept their own ignored, cares about children, | |
93 | * and should receive events when the inode is unmounted | |
94 | */ | |
95 | mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); | |
2d9048e2 | 96 | |
63c882a0 | 97 | /* mask off the flags used to open the fd */ |
8c1934c8 | 98 | mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); |
2d9048e2 | 99 | |
63c882a0 | 100 | return mask; |
2d9048e2 AG |
101 | } |
102 | ||
63c882a0 | 103 | static inline u32 inotify_mask_to_arg(__u32 mask) |
2d9048e2 | 104 | { |
63c882a0 EP |
105 | return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | |
106 | IN_Q_OVERFLOW); | |
2d9048e2 AG |
107 | } |
108 | ||
63c882a0 | 109 | /* intofiy userspace file descriptor functions */ |
2d9048e2 AG |
110 | static unsigned int inotify_poll(struct file *file, poll_table *wait) |
111 | { | |
63c882a0 | 112 | struct fsnotify_group *group = file->private_data; |
2d9048e2 AG |
113 | int ret = 0; |
114 | ||
63c882a0 | 115 | poll_wait(file, &group->notification_waitq, wait); |
c21dbe20 | 116 | spin_lock(&group->notification_lock); |
63c882a0 | 117 | if (!fsnotify_notify_queue_is_empty(group)) |
2d9048e2 | 118 | ret = POLLIN | POLLRDNORM; |
c21dbe20 | 119 | spin_unlock(&group->notification_lock); |
2d9048e2 AG |
120 | |
121 | return ret; | |
122 | } | |
123 | ||
7053aee2 | 124 | static int round_event_name_len(struct fsnotify_event *fsn_event) |
e9fe6904 | 125 | { |
7053aee2 JK |
126 | struct inotify_event_info *event; |
127 | ||
128 | event = INOTIFY_E(fsn_event); | |
e9fe6904 JK |
129 | if (!event->name_len) |
130 | return 0; | |
131 | return roundup(event->name_len + 1, sizeof(struct inotify_event)); | |
132 | } | |
133 | ||
3632dee2 VN |
134 | /* |
135 | * Get an inotify_kernel_event if one exists and is small | |
136 | * enough to fit in "count". Return an error pointer if | |
137 | * not large enough. | |
138 | * | |
c21dbe20 | 139 | * Called with the group->notification_lock held. |
3632dee2 | 140 | */ |
63c882a0 EP |
141 | static struct fsnotify_event *get_one_event(struct fsnotify_group *group, |
142 | size_t count) | |
3632dee2 VN |
143 | { |
144 | size_t event_size = sizeof(struct inotify_event); | |
63c882a0 | 145 | struct fsnotify_event *event; |
3632dee2 | 146 | |
63c882a0 | 147 | if (fsnotify_notify_queue_is_empty(group)) |
3632dee2 VN |
148 | return NULL; |
149 | ||
8ba8fa91 | 150 | event = fsnotify_peek_first_event(group); |
63c882a0 | 151 | |
5ba08e2e EP |
152 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
153 | ||
e9fe6904 | 154 | event_size += round_event_name_len(event); |
3632dee2 VN |
155 | if (event_size > count) |
156 | return ERR_PTR(-EINVAL); | |
157 | ||
c21dbe20 | 158 | /* held the notification_lock the whole time, so this is the |
63c882a0 | 159 | * same event we peeked above */ |
8ba8fa91 | 160 | fsnotify_remove_first_event(group); |
63c882a0 EP |
161 | |
162 | return event; | |
3632dee2 VN |
163 | } |
164 | ||
165 | /* | |
166 | * Copy an event to user space, returning how much we copied. | |
167 | * | |
168 | * We already checked that the event size is smaller than the | |
169 | * buffer we had in "get_one_event()" above. | |
170 | */ | |
63c882a0 | 171 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
7053aee2 | 172 | struct fsnotify_event *fsn_event, |
3632dee2 VN |
173 | char __user *buf) |
174 | { | |
63c882a0 | 175 | struct inotify_event inotify_event; |
7053aee2 | 176 | struct inotify_event_info *event; |
3632dee2 | 177 | size_t event_size = sizeof(struct inotify_event); |
e9fe6904 JK |
178 | size_t name_len; |
179 | size_t pad_name_len; | |
63c882a0 | 180 | |
7053aee2 | 181 | pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event); |
63c882a0 | 182 | |
7053aee2 | 183 | event = INOTIFY_E(fsn_event); |
e9fe6904 | 184 | name_len = event->name_len; |
b962e731 | 185 | /* |
e9fe6904 | 186 | * round up name length so it is a multiple of event_size |
0db501bd EB |
187 | * plus an extra byte for the terminating '\0'. |
188 | */ | |
7053aee2 | 189 | pad_name_len = round_event_name_len(fsn_event); |
e9fe6904 | 190 | inotify_event.len = pad_name_len; |
7053aee2 JK |
191 | inotify_event.mask = inotify_mask_to_arg(fsn_event->mask); |
192 | inotify_event.wd = event->wd; | |
63c882a0 | 193 | inotify_event.cookie = event->sync_cookie; |
3632dee2 | 194 | |
63c882a0 EP |
195 | /* send the main event */ |
196 | if (copy_to_user(buf, &inotify_event, event_size)) | |
3632dee2 VN |
197 | return -EFAULT; |
198 | ||
63c882a0 | 199 | buf += event_size; |
3632dee2 | 200 | |
63c882a0 EP |
201 | /* |
202 | * fsnotify only stores the pathname, so here we have to send the pathname | |
203 | * and then pad that pathname out to a multiple of sizeof(inotify_event) | |
e9fe6904 | 204 | * with zeros. |
63c882a0 | 205 | */ |
e9fe6904 | 206 | if (pad_name_len) { |
63c882a0 | 207 | /* copy the path name */ |
7053aee2 | 208 | if (copy_to_user(buf, event->name, name_len)) |
3632dee2 | 209 | return -EFAULT; |
e9fe6904 | 210 | buf += name_len; |
3632dee2 | 211 | |
0db501bd | 212 | /* fill userspace with 0's */ |
e9fe6904 | 213 | if (clear_user(buf, pad_name_len - name_len)) |
63c882a0 | 214 | return -EFAULT; |
e9fe6904 | 215 | event_size += pad_name_len; |
3632dee2 | 216 | } |
63c882a0 | 217 | |
3632dee2 VN |
218 | return event_size; |
219 | } | |
220 | ||
2d9048e2 AG |
221 | static ssize_t inotify_read(struct file *file, char __user *buf, |
222 | size_t count, loff_t *pos) | |
223 | { | |
63c882a0 EP |
224 | struct fsnotify_group *group; |
225 | struct fsnotify_event *kevent; | |
2d9048e2 AG |
226 | char __user *start; |
227 | int ret; | |
e23738a7 | 228 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
2d9048e2 AG |
229 | |
230 | start = buf; | |
63c882a0 | 231 | group = file->private_data; |
2d9048e2 | 232 | |
e23738a7 | 233 | add_wait_queue(&group->notification_waitq, &wait); |
2d9048e2 | 234 | while (1) { |
c21dbe20 | 235 | spin_lock(&group->notification_lock); |
63c882a0 | 236 | kevent = get_one_event(group, count); |
c21dbe20 | 237 | spin_unlock(&group->notification_lock); |
2d9048e2 | 238 | |
5ba08e2e EP |
239 | pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); |
240 | ||
3632dee2 VN |
241 | if (kevent) { |
242 | ret = PTR_ERR(kevent); | |
243 | if (IS_ERR(kevent)) | |
244 | break; | |
63c882a0 | 245 | ret = copy_event_to_user(group, kevent, buf); |
7053aee2 | 246 | fsnotify_destroy_event(group, kevent); |
3632dee2 VN |
247 | if (ret < 0) |
248 | break; | |
249 | buf += ret; | |
250 | count -= ret; | |
251 | continue; | |
2d9048e2 AG |
252 | } |
253 | ||
3632dee2 VN |
254 | ret = -EAGAIN; |
255 | if (file->f_flags & O_NONBLOCK) | |
2d9048e2 | 256 | break; |
1ca39ab9 | 257 | ret = -ERESTARTSYS; |
3632dee2 | 258 | if (signal_pending(current)) |
2d9048e2 | 259 | break; |
16dbc6c9 | 260 | |
3632dee2 | 261 | if (start != buf) |
2d9048e2 | 262 | break; |
16dbc6c9 | 263 | |
e23738a7 | 264 | wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
2d9048e2 | 265 | } |
e23738a7 | 266 | remove_wait_queue(&group->notification_waitq, &wait); |
2d9048e2 | 267 | |
3632dee2 VN |
268 | if (start != buf && ret != -EFAULT) |
269 | ret = buf - start; | |
2d9048e2 AG |
270 | return ret; |
271 | } | |
272 | ||
273 | static int inotify_release(struct inode *ignored, struct file *file) | |
274 | { | |
63c882a0 | 275 | struct fsnotify_group *group = file->private_data; |
2d9048e2 | 276 | |
5ba08e2e EP |
277 | pr_debug("%s: group=%p\n", __func__, group); |
278 | ||
63c882a0 | 279 | /* free this group, matching get was inotify_init->fsnotify_obtain_group */ |
d8153d4d | 280 | fsnotify_destroy_group(group); |
2d9048e2 AG |
281 | |
282 | return 0; | |
283 | } | |
284 | ||
285 | static long inotify_ioctl(struct file *file, unsigned int cmd, | |
286 | unsigned long arg) | |
287 | { | |
63c882a0 | 288 | struct fsnotify_group *group; |
7053aee2 | 289 | struct fsnotify_event *fsn_event; |
2d9048e2 AG |
290 | void __user *p; |
291 | int ret = -ENOTTY; | |
63c882a0 | 292 | size_t send_len = 0; |
2d9048e2 | 293 | |
63c882a0 | 294 | group = file->private_data; |
2d9048e2 AG |
295 | p = (void __user *) arg; |
296 | ||
5ba08e2e EP |
297 | pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); |
298 | ||
2d9048e2 AG |
299 | switch (cmd) { |
300 | case FIONREAD: | |
c21dbe20 | 301 | spin_lock(&group->notification_lock); |
7053aee2 JK |
302 | list_for_each_entry(fsn_event, &group->notification_list, |
303 | list) { | |
63c882a0 | 304 | send_len += sizeof(struct inotify_event); |
7053aee2 | 305 | send_len += round_event_name_len(fsn_event); |
63c882a0 | 306 | } |
c21dbe20 | 307 | spin_unlock(&group->notification_lock); |
63c882a0 | 308 | ret = put_user(send_len, (int __user *) p); |
2d9048e2 AG |
309 | break; |
310 | } | |
311 | ||
312 | return ret; | |
313 | } | |
314 | ||
315 | static const struct file_operations inotify_fops = { | |
be77196b | 316 | .show_fdinfo = inotify_show_fdinfo, |
63c882a0 EP |
317 | .poll = inotify_poll, |
318 | .read = inotify_read, | |
0a6b6bd5 | 319 | .fasync = fsnotify_fasync, |
63c882a0 EP |
320 | .release = inotify_release, |
321 | .unlocked_ioctl = inotify_ioctl, | |
2d9048e2 | 322 | .compat_ioctl = inotify_ioctl, |
6038f373 | 323 | .llseek = noop_llseek, |
2d9048e2 AG |
324 | }; |
325 | ||
2d9048e2 | 326 | |
63c882a0 EP |
327 | /* |
328 | * find_inode - resolve a user-given path to a specific inode | |
329 | */ | |
330 | static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) | |
331 | { | |
332 | int error; | |
333 | ||
334 | error = user_path_at(AT_FDCWD, dirname, flags, path); | |
335 | if (error) | |
336 | return error; | |
337 | /* you can only watch an inode if you have read permissions on it */ | |
338 | error = inode_permission(path->dentry->d_inode, MAY_READ); | |
339 | if (error) | |
340 | path_put(path); | |
341 | return error; | |
342 | } | |
343 | ||
b7ba8371 | 344 | static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, |
000285de | 345 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
346 | { |
347 | int ret; | |
348 | ||
4542da63 TH |
349 | idr_preload(GFP_KERNEL); |
350 | spin_lock(idr_lock); | |
b7ba8371 | 351 | |
a66c04b4 | 352 | ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT); |
4542da63 | 353 | if (ret >= 0) { |
b7ba8371 | 354 | /* we added the mark to the idr, take a reference */ |
4542da63 | 355 | i_mark->wd = ret; |
4542da63 TH |
356 | fsnotify_get_mark(&i_mark->fsn_mark); |
357 | } | |
b7ba8371 | 358 | |
4542da63 TH |
359 | spin_unlock(idr_lock); |
360 | idr_preload_end(); | |
361 | return ret < 0 ? ret : 0; | |
b7ba8371 EP |
362 | } |
363 | ||
000285de | 364 | static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, |
b7ba8371 EP |
365 | int wd) |
366 | { | |
367 | struct idr *idr = &group->inotify_data.idr; | |
368 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 369 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
370 | |
371 | assert_spin_locked(idr_lock); | |
372 | ||
000285de EP |
373 | i_mark = idr_find(idr, wd); |
374 | if (i_mark) { | |
375 | struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; | |
b7ba8371 | 376 | |
000285de | 377 | fsnotify_get_mark(fsn_mark); |
b7ba8371 | 378 | /* One ref for being in the idr, one ref we just took */ |
000285de | 379 | BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); |
b7ba8371 EP |
380 | } |
381 | ||
000285de | 382 | return i_mark; |
b7ba8371 EP |
383 | } |
384 | ||
000285de | 385 | static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, |
b7ba8371 EP |
386 | int wd) |
387 | { | |
000285de | 388 | struct inotify_inode_mark *i_mark; |
b7ba8371 EP |
389 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
390 | ||
391 | spin_lock(idr_lock); | |
000285de | 392 | i_mark = inotify_idr_find_locked(group, wd); |
b7ba8371 EP |
393 | spin_unlock(idr_lock); |
394 | ||
000285de | 395 | return i_mark; |
b7ba8371 EP |
396 | } |
397 | ||
398 | static void do_inotify_remove_from_idr(struct fsnotify_group *group, | |
000285de | 399 | struct inotify_inode_mark *i_mark) |
b7ba8371 EP |
400 | { |
401 | struct idr *idr = &group->inotify_data.idr; | |
402 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
000285de | 403 | int wd = i_mark->wd; |
b7ba8371 EP |
404 | |
405 | assert_spin_locked(idr_lock); | |
406 | ||
407 | idr_remove(idr, wd); | |
408 | ||
409 | /* removed from the idr, drop that ref */ | |
000285de | 410 | fsnotify_put_mark(&i_mark->fsn_mark); |
b7ba8371 EP |
411 | } |
412 | ||
dead537d EP |
413 | /* |
414 | * Remove the mark from the idr (if present) and drop the reference | |
415 | * on the mark because it was in the idr. | |
416 | */ | |
7e790dd5 | 417 | static void inotify_remove_from_idr(struct fsnotify_group *group, |
000285de | 418 | struct inotify_inode_mark *i_mark) |
7e790dd5 | 419 | { |
b7ba8371 | 420 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; |
000285de | 421 | struct inotify_inode_mark *found_i_mark = NULL; |
dead537d | 422 | int wd; |
7e790dd5 | 423 | |
b7ba8371 | 424 | spin_lock(idr_lock); |
000285de | 425 | wd = i_mark->wd; |
dead537d | 426 | |
b7ba8371 | 427 | /* |
000285de | 428 | * does this i_mark think it is in the idr? we shouldn't get called |
b7ba8371 EP |
429 | * if it wasn't.... |
430 | */ | |
431 | if (wd == -1) { | |
000285de EP |
432 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" |
433 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
0809ab69 | 434 | i_mark->fsn_mark.group, i_mark->fsn_mark.inode); |
dead537d | 435 | goto out; |
b7ba8371 | 436 | } |
dead537d | 437 | |
b7ba8371 | 438 | /* Lets look in the idr to see if we find it */ |
000285de EP |
439 | found_i_mark = inotify_idr_find_locked(group, wd); |
440 | if (unlikely(!found_i_mark)) { | |
441 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
442 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
0809ab69 | 443 | i_mark->fsn_mark.group, i_mark->fsn_mark.inode); |
dead537d | 444 | goto out; |
b7ba8371 | 445 | } |
dead537d | 446 | |
b7ba8371 | 447 | /* |
000285de EP |
448 | * We found an mark in the idr at the right wd, but it's |
449 | * not the mark we were told to remove. eparis seriously | |
b7ba8371 EP |
450 | * fucked up somewhere. |
451 | */ | |
000285de EP |
452 | if (unlikely(found_i_mark != i_mark)) { |
453 | WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " | |
454 | "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " | |
455 | "found_i_mark->group=%p found_i_mark->inode=%p\n", | |
456 | __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, | |
0809ab69 | 457 | i_mark->fsn_mark.inode, found_i_mark, found_i_mark->wd, |
000285de | 458 | found_i_mark->fsn_mark.group, |
0809ab69 | 459 | found_i_mark->fsn_mark.inode); |
dead537d EP |
460 | goto out; |
461 | } | |
462 | ||
b7ba8371 EP |
463 | /* |
464 | * One ref for being in the idr | |
465 | * one ref held by the caller trying to kill us | |
466 | * one ref grabbed by inotify_idr_find | |
467 | */ | |
000285de EP |
468 | if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { |
469 | printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" | |
470 | " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, | |
0809ab69 | 471 | i_mark->fsn_mark.group, i_mark->fsn_mark.inode); |
b7ba8371 EP |
472 | /* we can't really recover with bad ref cnting.. */ |
473 | BUG(); | |
474 | } | |
dead537d | 475 | |
000285de | 476 | do_inotify_remove_from_idr(group, i_mark); |
dead537d | 477 | out: |
b7ba8371 | 478 | /* match the ref taken by inotify_idr_find_locked() */ |
000285de EP |
479 | if (found_i_mark) |
480 | fsnotify_put_mark(&found_i_mark->fsn_mark); | |
481 | i_mark->wd = -1; | |
b7ba8371 | 482 | spin_unlock(idr_lock); |
7e790dd5 | 483 | } |
dead537d | 484 | |
63c882a0 | 485 | /* |
dead537d | 486 | * Send IN_IGNORED for this wd, remove this wd from the idr. |
63c882a0 | 487 | */ |
000285de | 488 | void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, |
528da3e9 | 489 | struct fsnotify_group *group) |
63c882a0 | 490 | { |
000285de | 491 | struct inotify_inode_mark *i_mark; |
63c882a0 | 492 | |
7053aee2 JK |
493 | /* Queue ignore event for the watch */ |
494 | inotify_handle_event(group, NULL, fsn_mark, NULL, FS_IN_IGNORED, | |
45a22f4c | 495 | NULL, FSNOTIFY_EVENT_NONE, NULL, 0); |
f44aebcc | 496 | |
7053aee2 | 497 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
000285de EP |
498 | /* remove this mark from the idr */ |
499 | inotify_remove_from_idr(group, i_mark); | |
63c882a0 | 500 | |
1cce1eea | 501 | dec_inotify_watches(group->inotify_data.ucounts); |
63c882a0 EP |
502 | } |
503 | ||
504 | /* ding dong the mark is dead */ | |
000285de | 505 | static void inotify_free_mark(struct fsnotify_mark *fsn_mark) |
63c882a0 | 506 | { |
000285de | 507 | struct inotify_inode_mark *i_mark; |
31ddd326 | 508 | |
000285de | 509 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
63c882a0 | 510 | |
000285de | 511 | kmem_cache_free(inotify_inode_mark_cachep, i_mark); |
63c882a0 EP |
512 | } |
513 | ||
52cef755 EP |
514 | static int inotify_update_existing_watch(struct fsnotify_group *group, |
515 | struct inode *inode, | |
516 | u32 arg) | |
63c882a0 | 517 | { |
000285de EP |
518 | struct fsnotify_mark *fsn_mark; |
519 | struct inotify_inode_mark *i_mark; | |
63c882a0 | 520 | __u32 old_mask, new_mask; |
52cef755 EP |
521 | __u32 mask; |
522 | int add = (arg & IN_MASK_ADD); | |
523 | int ret; | |
63c882a0 | 524 | |
63c882a0 | 525 | mask = inotify_arg_to_mask(arg); |
63c882a0 | 526 | |
5444e298 | 527 | fsn_mark = fsnotify_find_inode_mark(group, inode); |
000285de | 528 | if (!fsn_mark) |
52cef755 | 529 | return -ENOENT; |
7e790dd5 | 530 | |
000285de | 531 | i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); |
75fe2b26 | 532 | |
000285de | 533 | spin_lock(&fsn_mark->lock); |
63c882a0 | 534 | |
000285de | 535 | old_mask = fsn_mark->mask; |
90b1e7a5 EP |
536 | if (add) |
537 | fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); | |
538 | else | |
539 | fsnotify_set_mark_mask_locked(fsn_mark, mask); | |
540 | new_mask = fsn_mark->mask; | |
63c882a0 | 541 | |
000285de | 542 | spin_unlock(&fsn_mark->lock); |
63c882a0 EP |
543 | |
544 | if (old_mask != new_mask) { | |
545 | /* more bits in old than in new? */ | |
546 | int dropped = (old_mask & ~new_mask); | |
000285de | 547 | /* more bits in this fsn_mark than the inode's mask? */ |
63c882a0 | 548 | int do_inode = (new_mask & ~inode->i_fsnotify_mask); |
63c882a0 | 549 | |
000285de | 550 | /* update the inode with this new fsn_mark */ |
63c882a0 EP |
551 | if (dropped || do_inode) |
552 | fsnotify_recalc_inode_mask(inode); | |
553 | ||
63c882a0 EP |
554 | } |
555 | ||
52cef755 | 556 | /* return the wd */ |
000285de | 557 | ret = i_mark->wd; |
52cef755 | 558 | |
d0775441 | 559 | /* match the get from fsnotify_find_mark() */ |
000285de | 560 | fsnotify_put_mark(fsn_mark); |
75fe2b26 | 561 | |
52cef755 EP |
562 | return ret; |
563 | } | |
564 | ||
565 | static int inotify_new_watch(struct fsnotify_group *group, | |
566 | struct inode *inode, | |
567 | u32 arg) | |
568 | { | |
000285de | 569 | struct inotify_inode_mark *tmp_i_mark; |
52cef755 EP |
570 | __u32 mask; |
571 | int ret; | |
b7ba8371 EP |
572 | struct idr *idr = &group->inotify_data.idr; |
573 | spinlock_t *idr_lock = &group->inotify_data.idr_lock; | |
52cef755 | 574 | |
52cef755 | 575 | mask = inotify_arg_to_mask(arg); |
52cef755 | 576 | |
000285de EP |
577 | tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); |
578 | if (unlikely(!tmp_i_mark)) | |
52cef755 EP |
579 | return -ENOMEM; |
580 | ||
000285de EP |
581 | fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); |
582 | tmp_i_mark->fsn_mark.mask = mask; | |
583 | tmp_i_mark->wd = -1; | |
52cef755 | 584 | |
a66c04b4 | 585 | ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark); |
b7ba8371 | 586 | if (ret) |
52cef755 | 587 | goto out_err; |
52cef755 | 588 | |
1cce1eea NB |
589 | /* increment the number of watches the user has */ |
590 | if (!inc_inotify_watches(group->inotify_data.ucounts)) { | |
591 | inotify_remove_from_idr(group, tmp_i_mark); | |
592 | ret = -ENOSPC; | |
593 | goto out_err; | |
594 | } | |
595 | ||
52cef755 | 596 | /* we are on the idr, now get on the inode */ |
e1e5a9f8 LS |
597 | ret = fsnotify_add_mark_locked(&tmp_i_mark->fsn_mark, group, inode, |
598 | NULL, 0); | |
52cef755 EP |
599 | if (ret) { |
600 | /* we failed to get on the inode, get off the idr */ | |
000285de | 601 | inotify_remove_from_idr(group, tmp_i_mark); |
52cef755 EP |
602 | goto out_err; |
603 | } | |
604 | ||
52cef755 | 605 | |
000285de EP |
606 | /* return the watch descriptor for this new mark */ |
607 | ret = tmp_i_mark->wd; | |
52cef755 | 608 | |
63c882a0 | 609 | out_err: |
000285de EP |
610 | /* match the ref from fsnotify_init_mark() */ |
611 | fsnotify_put_mark(&tmp_i_mark->fsn_mark); | |
52cef755 EP |
612 | |
613 | return ret; | |
614 | } | |
615 | ||
616 | static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) | |
617 | { | |
618 | int ret = 0; | |
619 | ||
e1e5a9f8 | 620 | mutex_lock(&group->mark_mutex); |
52cef755 EP |
621 | /* try to update and existing watch with the new arg */ |
622 | ret = inotify_update_existing_watch(group, inode, arg); | |
623 | /* no mark present, try to add a new one */ | |
624 | if (ret == -ENOENT) | |
625 | ret = inotify_new_watch(group, inode, arg); | |
e1e5a9f8 | 626 | mutex_unlock(&group->mark_mutex); |
7e790dd5 | 627 | |
63c882a0 EP |
628 | return ret; |
629 | } | |
630 | ||
d0de4dc5 | 631 | static struct fsnotify_group *inotify_new_group(unsigned int max_events) |
63c882a0 EP |
632 | { |
633 | struct fsnotify_group *group; | |
ff57cd58 | 634 | struct inotify_event_info *oevent; |
63c882a0 | 635 | |
0d2e2a1d | 636 | group = fsnotify_alloc_group(&inotify_fsnotify_ops); |
63c882a0 EP |
637 | if (IS_ERR(group)) |
638 | return group; | |
639 | ||
ff57cd58 JK |
640 | oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL); |
641 | if (unlikely(!oevent)) { | |
642 | fsnotify_destroy_group(group); | |
643 | return ERR_PTR(-ENOMEM); | |
644 | } | |
645 | group->overflow_event = &oevent->fse; | |
646 | fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW); | |
647 | oevent->wd = -1; | |
648 | oevent->sync_cookie = 0; | |
649 | oevent->name_len = 0; | |
650 | ||
63c882a0 EP |
651 | group->max_events = max_events; |
652 | ||
653 | spin_lock_init(&group->inotify_data.idr_lock); | |
654 | idr_init(&group->inotify_data.idr); | |
1cce1eea NB |
655 | group->inotify_data.ucounts = inc_ucount(current_user_ns(), |
656 | current_euid(), | |
657 | UCOUNT_INOTIFY_INSTANCES); | |
d0de4dc5 | 658 | |
1cce1eea | 659 | if (!group->inotify_data.ucounts) { |
d8153d4d | 660 | fsnotify_destroy_group(group); |
d0de4dc5 EP |
661 | return ERR_PTR(-EMFILE); |
662 | } | |
63c882a0 EP |
663 | |
664 | return group; | |
665 | } | |
666 | ||
667 | ||
668 | /* inotify syscalls */ | |
938bb9f5 | 669 | SYSCALL_DEFINE1(inotify_init1, int, flags) |
2d9048e2 | 670 | { |
63c882a0 | 671 | struct fsnotify_group *group; |
c44dcc56 | 672 | int ret; |
2d9048e2 | 673 | |
e38b36f3 UD |
674 | /* Check the IN_* constants for consistency. */ |
675 | BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); | |
676 | BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); | |
677 | ||
510df2dd | 678 | if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) |
4006553b UD |
679 | return -EINVAL; |
680 | ||
63c882a0 | 681 | /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ |
d0de4dc5 EP |
682 | group = inotify_new_group(inotify_max_queued_events); |
683 | if (IS_ERR(group)) | |
684 | return PTR_ERR(group); | |
825f9692 | 685 | |
c44dcc56 AV |
686 | ret = anon_inode_getfd("inotify", &inotify_fops, group, |
687 | O_RDONLY | flags); | |
d0de4dc5 | 688 | if (ret < 0) |
d8153d4d | 689 | fsnotify_destroy_group(group); |
825f9692 | 690 | |
2d9048e2 AG |
691 | return ret; |
692 | } | |
693 | ||
938bb9f5 | 694 | SYSCALL_DEFINE0(inotify_init) |
4006553b UD |
695 | { |
696 | return sys_inotify_init1(0); | |
697 | } | |
698 | ||
2e4d0924 HC |
699 | SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, |
700 | u32, mask) | |
2d9048e2 | 701 | { |
63c882a0 | 702 | struct fsnotify_group *group; |
2d9048e2 | 703 | struct inode *inode; |
2d8f3038 | 704 | struct path path; |
2903ff01 AV |
705 | struct fd f; |
706 | int ret; | |
2d9048e2 AG |
707 | unsigned flags = 0; |
708 | ||
d30e2c05 DH |
709 | /* |
710 | * We share a lot of code with fs/dnotify. We also share | |
711 | * the bit layout between inotify's IN_* and the fsnotify | |
712 | * FS_*. This check ensures that only the inotify IN_* | |
713 | * bits get passed in and set in watches/events. | |
714 | */ | |
715 | if (unlikely(mask & ~ALL_INOTIFY_BITS)) | |
716 | return -EINVAL; | |
717 | /* | |
718 | * Require at least one valid bit set in the mask. | |
719 | * Without _something_ set, we would have no events to | |
720 | * watch for. | |
721 | */ | |
04df32fa ZH |
722 | if (unlikely(!(mask & ALL_INOTIFY_BITS))) |
723 | return -EINVAL; | |
724 | ||
2903ff01 AV |
725 | f = fdget(fd); |
726 | if (unlikely(!f.file)) | |
2d9048e2 AG |
727 | return -EBADF; |
728 | ||
729 | /* verify that this is indeed an inotify instance */ | |
2903ff01 | 730 | if (unlikely(f.file->f_op != &inotify_fops)) { |
2d9048e2 AG |
731 | ret = -EINVAL; |
732 | goto fput_and_out; | |
733 | } | |
734 | ||
735 | if (!(mask & IN_DONT_FOLLOW)) | |
736 | flags |= LOOKUP_FOLLOW; | |
737 | if (mask & IN_ONLYDIR) | |
738 | flags |= LOOKUP_DIRECTORY; | |
739 | ||
63c882a0 EP |
740 | ret = inotify_find_inode(pathname, &path, flags); |
741 | if (ret) | |
2d9048e2 AG |
742 | goto fput_and_out; |
743 | ||
63c882a0 | 744 | /* inode held in place by reference to path; group by fget on fd */ |
2d8f3038 | 745 | inode = path.dentry->d_inode; |
2903ff01 | 746 | group = f.file->private_data; |
2d9048e2 | 747 | |
63c882a0 EP |
748 | /* create/update an inode mark */ |
749 | ret = inotify_update_watch(group, inode, mask); | |
2d8f3038 | 750 | path_put(&path); |
2d9048e2 | 751 | fput_and_out: |
2903ff01 | 752 | fdput(f); |
2d9048e2 AG |
753 | return ret; |
754 | } | |
755 | ||
2e4d0924 | 756 | SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) |
2d9048e2 | 757 | { |
63c882a0 | 758 | struct fsnotify_group *group; |
000285de | 759 | struct inotify_inode_mark *i_mark; |
2903ff01 AV |
760 | struct fd f; |
761 | int ret = 0; | |
2d9048e2 | 762 | |
2903ff01 AV |
763 | f = fdget(fd); |
764 | if (unlikely(!f.file)) | |
2d9048e2 AG |
765 | return -EBADF; |
766 | ||
767 | /* verify that this is indeed an inotify instance */ | |
b7ba8371 | 768 | ret = -EINVAL; |
2903ff01 | 769 | if (unlikely(f.file->f_op != &inotify_fops)) |
2d9048e2 | 770 | goto out; |
2d9048e2 | 771 | |
2903ff01 | 772 | group = f.file->private_data; |
2d9048e2 | 773 | |
b7ba8371 | 774 | ret = -EINVAL; |
000285de EP |
775 | i_mark = inotify_idr_find(group, wd); |
776 | if (unlikely(!i_mark)) | |
63c882a0 | 777 | goto out; |
63c882a0 | 778 | |
b7ba8371 EP |
779 | ret = 0; |
780 | ||
e2a29943 | 781 | fsnotify_destroy_mark(&i_mark->fsn_mark, group); |
b7ba8371 EP |
782 | |
783 | /* match ref taken by inotify_idr_find */ | |
000285de | 784 | fsnotify_put_mark(&i_mark->fsn_mark); |
2d9048e2 AG |
785 | |
786 | out: | |
2903ff01 | 787 | fdput(f); |
2d9048e2 AG |
788 | return ret; |
789 | } | |
790 | ||
2d9048e2 | 791 | /* |
ae0e47f0 | 792 | * inotify_user_setup - Our initialization function. Note that we cannot return |
2d9048e2 AG |
793 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
794 | * must result in panic(). | |
795 | */ | |
796 | static int __init inotify_user_setup(void) | |
797 | { | |
f874e1ac EP |
798 | BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); |
799 | BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); | |
800 | BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); | |
801 | BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); | |
802 | BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); | |
803 | BUILD_BUG_ON(IN_OPEN != FS_OPEN); | |
804 | BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); | |
805 | BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); | |
806 | BUILD_BUG_ON(IN_CREATE != FS_CREATE); | |
807 | BUILD_BUG_ON(IN_DELETE != FS_DELETE); | |
808 | BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); | |
809 | BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); | |
810 | BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); | |
811 | BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); | |
812 | BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); | |
813 | BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); | |
b29866aa | 814 | BUILD_BUG_ON(IN_ISDIR != FS_ISDIR); |
f874e1ac EP |
815 | BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); |
816 | ||
817 | BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); | |
818 | ||
000285de | 819 | inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); |
63c882a0 | 820 | |
2d9048e2 | 821 | inotify_max_queued_events = 16384; |
1cce1eea NB |
822 | init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128; |
823 | init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192; | |
2d9048e2 | 824 | |
2d9048e2 AG |
825 | return 0; |
826 | } | |
c013d5a4 | 827 | fs_initcall(inotify_user_setup); |