diff options
Diffstat (limited to 'fs/inotify.c')
-rw-r--r-- | fs/inotify.c | 1037 |
1 files changed, 1037 insertions, 0 deletions
diff --git a/fs/inotify.c b/fs/inotify.c new file mode 100644 index 000000000000..54757be888b6 --- /dev/null +++ b/fs/inotify.c | |||
@@ -0,0 +1,1037 @@ | |||
1 | /* | ||
2 | * fs/inotify.c - inode-based file event notifications | ||
3 | * | ||
4 | * Authors: | ||
5 | * John McCutchan <ttb@tentacle.dhs.org> | ||
6 | * Robert Love <rml@novell.com> | ||
7 | * | ||
8 | * Copyright (C) 2005 John McCutchan | ||
9 | * | ||
10 | * This program is free software; you can redistribute it and/or modify it | ||
11 | * under the terms of the GNU General Public License as published by the | ||
12 | * Free Software Foundation; either version 2, or (at your option) any | ||
13 | * later version. | ||
14 | * | ||
15 | * This program is distributed in the hope that it will be useful, but | ||
16 | * WITHOUT ANY WARRANTY; without even the implied warranty of | ||
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
18 | * General Public License for more details. | ||
19 | */ | ||
20 | |||
21 | #include <linux/module.h> | ||
22 | #include <linux/kernel.h> | ||
23 | #include <linux/sched.h> | ||
24 | #include <linux/spinlock.h> | ||
25 | #include <linux/idr.h> | ||
26 | #include <linux/slab.h> | ||
27 | #include <linux/fs.h> | ||
28 | #include <linux/file.h> | ||
29 | #include <linux/mount.h> | ||
30 | #include <linux/namei.h> | ||
31 | #include <linux/poll.h> | ||
32 | #include <linux/init.h> | ||
33 | #include <linux/list.h> | ||
34 | #include <linux/writeback.h> | ||
35 | #include <linux/inotify.h> | ||
36 | |||
37 | #include <asm/ioctls.h> | ||
38 | |||
39 | static atomic_t inotify_cookie; | ||
40 | |||
41 | static kmem_cache_t *watch_cachep; | ||
42 | static kmem_cache_t *event_cachep; | ||
43 | |||
44 | static struct vfsmount *inotify_mnt; | ||
45 | |||
46 | /* these are configurable via /proc/sys/fs/inotify/ */ | ||
47 | int inotify_max_user_instances; | ||
48 | int inotify_max_user_watches; | ||
49 | int inotify_max_queued_events; | ||
50 | |||
51 | /* | ||
52 | * Lock ordering: | ||
53 | * | ||
54 | * dentry->d_lock (used to keep d_move() away from dentry->d_parent) | ||
55 | * iprune_sem (synchronize shrink_icache_memory()) | ||
56 | * inode_lock (protects the super_block->s_inodes list) | ||
57 | * inode->inotify_sem (protects inode->inotify_watches and watches->i_list) | ||
58 | * inotify_dev->sem (protects inotify_device and watches->d_list) | ||
59 | */ | ||
60 | |||
61 | /* | ||
62 | * Lifetimes of the three main data structures--inotify_device, inode, and | ||
63 | * inotify_watch--are managed by reference count. | ||
64 | * | ||
65 | * inotify_device: Lifetime is from open until release. Additional references | ||
66 | * can bump the count via get_inotify_dev() and drop the count via | ||
67 | * put_inotify_dev(). | ||
68 | * | ||
69 | * inotify_watch: Lifetime is from create_watch() to destory_watch(). | ||
70 | * Additional references can bump the count via get_inotify_watch() and drop | ||
71 | * the count via put_inotify_watch(). | ||
72 | * | ||
73 | * inode: Pinned so long as the inode is associated with a watch, from | ||
74 | * create_watch() to put_inotify_watch(). | ||
75 | */ | ||
76 | |||
77 | /* | ||
78 | * struct inotify_device - represents an open instance of an inotify device | ||
79 | * | ||
80 | * This structure is protected by the semaphore 'sem'. | ||
81 | */ | ||
82 | struct inotify_device { | ||
83 | wait_queue_head_t wq; /* wait queue for i/o */ | ||
84 | struct idr idr; /* idr mapping wd -> watch */ | ||
85 | struct semaphore sem; /* protects this bad boy */ | ||
86 | struct list_head events; /* list of queued events */ | ||
87 | struct list_head watches; /* list of watches */ | ||
88 | atomic_t count; /* reference count */ | ||
89 | struct user_struct *user; /* user who opened this dev */ | ||
90 | unsigned int queue_size; /* size of the queue (bytes) */ | ||
91 | unsigned int event_count; /* number of pending events */ | ||
92 | unsigned int max_events; /* maximum number of events */ | ||
93 | }; | ||
94 | |||
95 | /* | ||
96 | * struct inotify_kernel_event - An inotify event, originating from a watch and | ||
97 | * queued for user-space. A list of these is attached to each instance of the | ||
98 | * device. In read(), this list is walked and all events that can fit in the | ||
99 | * buffer are returned. | ||
100 | * | ||
101 | * Protected by dev->sem of the device in which we are queued. | ||
102 | */ | ||
103 | struct inotify_kernel_event { | ||
104 | struct inotify_event event; /* the user-space event */ | ||
105 | struct list_head list; /* entry in inotify_device's list */ | ||
106 | char *name; /* filename, if any */ | ||
107 | }; | ||
108 | |||
109 | /* | ||
110 | * struct inotify_watch - represents a watch request on a specific inode | ||
111 | * | ||
112 | * d_list is protected by dev->sem of the associated watch->dev. | ||
113 | * i_list and mask are protected by inode->inotify_sem of the associated inode. | ||
114 | * dev, inode, and wd are never written to once the watch is created. | ||
115 | */ | ||
116 | struct inotify_watch { | ||
117 | struct list_head d_list; /* entry in inotify_device's list */ | ||
118 | struct list_head i_list; /* entry in inode's list */ | ||
119 | atomic_t count; /* reference count */ | ||
120 | struct inotify_device *dev; /* associated device */ | ||
121 | struct inode *inode; /* associated inode */ | ||
122 | s32 wd; /* watch descriptor */ | ||
123 | u32 mask; /* event mask for this watch */ | ||
124 | }; | ||
125 | |||
126 | #ifdef CONFIG_SYSCTL | ||
127 | |||
128 | #include <linux/sysctl.h> | ||
129 | |||
130 | static int zero; | ||
131 | |||
132 | ctl_table inotify_table[] = { | ||
133 | { | ||
134 | .ctl_name = INOTIFY_MAX_USER_INSTANCES, | ||
135 | .procname = "max_user_instances", | ||
136 | .data = &inotify_max_user_instances, | ||
137 | .maxlen = sizeof(int), | ||
138 | .mode = 0644, | ||
139 | .proc_handler = &proc_dointvec_minmax, | ||
140 | .strategy = &sysctl_intvec, | ||
141 | .extra1 = &zero, | ||
142 | }, | ||
143 | { | ||
144 | .ctl_name = INOTIFY_MAX_USER_WATCHES, | ||
145 | .procname = "max_user_watches", | ||
146 | .data = &inotify_max_user_watches, | ||
147 | .maxlen = sizeof(int), | ||
148 | .mode = 0644, | ||
149 | .proc_handler = &proc_dointvec_minmax, | ||
150 | .strategy = &sysctl_intvec, | ||
151 | .extra1 = &zero, | ||
152 | }, | ||
153 | { | ||
154 | .ctl_name = INOTIFY_MAX_QUEUED_EVENTS, | ||
155 | .procname = "max_queued_events", | ||
156 | .data = &inotify_max_queued_events, | ||
157 | .maxlen = sizeof(int), | ||
158 | .mode = 0644, | ||
159 | .proc_handler = &proc_dointvec_minmax, | ||
160 | .strategy = &sysctl_intvec, | ||
161 | .extra1 = &zero | ||
162 | }, | ||
163 | { .ctl_name = 0 } | ||
164 | }; | ||
165 | #endif /* CONFIG_SYSCTL */ | ||
166 | |||
167 | static inline void get_inotify_dev(struct inotify_device *dev) | ||
168 | { | ||
169 | atomic_inc(&dev->count); | ||
170 | } | ||
171 | |||
172 | static inline void put_inotify_dev(struct inotify_device *dev) | ||
173 | { | ||
174 | if (atomic_dec_and_test(&dev->count)) { | ||
175 | atomic_dec(&dev->user->inotify_devs); | ||
176 | free_uid(dev->user); | ||
177 | kfree(dev); | ||
178 | } | ||
179 | } | ||
180 | |||
181 | static inline void get_inotify_watch(struct inotify_watch *watch) | ||
182 | { | ||
183 | atomic_inc(&watch->count); | ||
184 | } | ||
185 | |||
186 | /* | ||
187 | * put_inotify_watch - decrements the ref count on a given watch. cleans up | ||
188 | * the watch and its references if the count reaches zero. | ||
189 | */ | ||
190 | static inline void put_inotify_watch(struct inotify_watch *watch) | ||
191 | { | ||
192 | if (atomic_dec_and_test(&watch->count)) { | ||
193 | put_inotify_dev(watch->dev); | ||
194 | iput(watch->inode); | ||
195 | kmem_cache_free(watch_cachep, watch); | ||
196 | } | ||
197 | } | ||
198 | |||
199 | /* | ||
200 | * kernel_event - create a new kernel event with the given parameters | ||
201 | * | ||
202 | * This function can sleep. | ||
203 | */ | ||
204 | static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie, | ||
205 | const char *name) | ||
206 | { | ||
207 | struct inotify_kernel_event *kevent; | ||
208 | |||
209 | kevent = kmem_cache_alloc(event_cachep, GFP_KERNEL); | ||
210 | if (unlikely(!kevent)) | ||
211 | return NULL; | ||
212 | |||
213 | /* we hand this out to user-space, so zero it just in case */ | ||
214 | memset(&kevent->event, 0, sizeof(struct inotify_event)); | ||
215 | |||
216 | kevent->event.wd = wd; | ||
217 | kevent->event.mask = mask; | ||
218 | kevent->event.cookie = cookie; | ||
219 | |||
220 | INIT_LIST_HEAD(&kevent->list); | ||
221 | |||
222 | if (name) { | ||
223 | size_t len, rem, event_size = sizeof(struct inotify_event); | ||
224 | |||
225 | /* | ||
226 | * We need to pad the filename so as to properly align an | ||
227 | * array of inotify_event structures. Because the structure is | ||
228 | * small and the common case is a small filename, we just round | ||
229 | * up to the next multiple of the structure's sizeof. This is | ||
230 | * simple and safe for all architectures. | ||
231 | */ | ||
232 | len = strlen(name) + 1; | ||
233 | rem = event_size - len; | ||
234 | if (len > event_size) { | ||
235 | rem = event_size - (len % event_size); | ||
236 | if (len % event_size == 0) | ||
237 | rem = 0; | ||
238 | } | ||
239 | |||
240 | kevent->name = kmalloc(len + rem, GFP_KERNEL); | ||
241 | if (unlikely(!kevent->name)) { | ||
242 | kmem_cache_free(event_cachep, kevent); | ||
243 | return NULL; | ||
244 | } | ||
245 | memcpy(kevent->name, name, len); | ||
246 | if (rem) | ||
247 | memset(kevent->name + len, 0, rem); | ||
248 | kevent->event.len = len + rem; | ||
249 | } else { | ||
250 | kevent->event.len = 0; | ||
251 | kevent->name = NULL; | ||
252 | } | ||
253 | |||
254 | return kevent; | ||
255 | } | ||
256 | |||
257 | /* | ||
258 | * inotify_dev_get_event - return the next event in the given dev's queue | ||
259 | * | ||
260 | * Caller must hold dev->sem. | ||
261 | */ | ||
262 | static inline struct inotify_kernel_event * | ||
263 | inotify_dev_get_event(struct inotify_device *dev) | ||
264 | { | ||
265 | return list_entry(dev->events.next, struct inotify_kernel_event, list); | ||
266 | } | ||
267 | |||
268 | /* | ||
269 | * inotify_dev_queue_event - add a new event to the given device | ||
270 | * | ||
271 | * Caller must hold dev->sem. Can sleep (calls kernel_event()). | ||
272 | */ | ||
273 | static void inotify_dev_queue_event(struct inotify_device *dev, | ||
274 | struct inotify_watch *watch, u32 mask, | ||
275 | u32 cookie, const char *name) | ||
276 | { | ||
277 | struct inotify_kernel_event *kevent, *last; | ||
278 | |||
279 | /* coalescing: drop this event if it is a dupe of the previous */ | ||
280 | last = inotify_dev_get_event(dev); | ||
281 | if (last && last->event.mask == mask && last->event.wd == watch->wd && | ||
282 | last->event.cookie == cookie) { | ||
283 | const char *lastname = last->name; | ||
284 | |||
285 | if (!name && !lastname) | ||
286 | return; | ||
287 | if (name && lastname && !strcmp(lastname, name)) | ||
288 | return; | ||
289 | } | ||
290 | |||
291 | /* the queue overflowed and we already sent the Q_OVERFLOW event */ | ||
292 | if (unlikely(dev->event_count > dev->max_events)) | ||
293 | return; | ||
294 | |||
295 | /* if the queue overflows, we need to notify user space */ | ||
296 | if (unlikely(dev->event_count == dev->max_events)) | ||
297 | kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL); | ||
298 | else | ||
299 | kevent = kernel_event(watch->wd, mask, cookie, name); | ||
300 | |||
301 | if (unlikely(!kevent)) | ||
302 | return; | ||
303 | |||
304 | /* queue the event and wake up anyone waiting */ | ||
305 | dev->event_count++; | ||
306 | dev->queue_size += sizeof(struct inotify_event) + kevent->event.len; | ||
307 | list_add_tail(&kevent->list, &dev->events); | ||
308 | wake_up_interruptible(&dev->wq); | ||
309 | } | ||
310 | |||
311 | /* | ||
312 | * remove_kevent - cleans up and ultimately frees the given kevent | ||
313 | * | ||
314 | * Caller must hold dev->sem. | ||
315 | */ | ||
316 | static void remove_kevent(struct inotify_device *dev, | ||
317 | struct inotify_kernel_event *kevent) | ||
318 | { | ||
319 | list_del(&kevent->list); | ||
320 | |||
321 | dev->event_count--; | ||
322 | dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len; | ||
323 | |||
324 | kfree(kevent->name); | ||
325 | kmem_cache_free(event_cachep, kevent); | ||
326 | } | ||
327 | |||
328 | /* | ||
329 | * inotify_dev_event_dequeue - destroy an event on the given device | ||
330 | * | ||
331 | * Caller must hold dev->sem. | ||
332 | */ | ||
333 | static void inotify_dev_event_dequeue(struct inotify_device *dev) | ||
334 | { | ||
335 | if (!list_empty(&dev->events)) { | ||
336 | struct inotify_kernel_event *kevent; | ||
337 | kevent = inotify_dev_get_event(dev); | ||
338 | remove_kevent(dev, kevent); | ||
339 | } | ||
340 | } | ||
341 | |||
342 | /* | ||
343 | * inotify_dev_get_wd - returns the next WD for use by the given dev | ||
344 | * | ||
345 | * Callers must hold dev->sem. This function can sleep. | ||
346 | */ | ||
347 | static int inotify_dev_get_wd(struct inotify_device *dev, | ||
348 | struct inotify_watch *watch) | ||
349 | { | ||
350 | int ret; | ||
351 | |||
352 | do { | ||
353 | if (unlikely(!idr_pre_get(&dev->idr, GFP_KERNEL))) | ||
354 | return -ENOSPC; | ||
355 | ret = idr_get_new(&dev->idr, watch, &watch->wd); | ||
356 | } while (ret == -EAGAIN); | ||
357 | |||
358 | return ret; | ||
359 | } | ||
360 | |||
361 | /* | ||
362 | * find_inode - resolve a user-given path to a specific inode and return a nd | ||
363 | */ | ||
364 | static int find_inode(const char __user *dirname, struct nameidata *nd) | ||
365 | { | ||
366 | int error; | ||
367 | |||
368 | error = __user_walk(dirname, LOOKUP_FOLLOW, nd); | ||
369 | if (error) | ||
370 | return error; | ||
371 | /* you can only watch an inode if you have read permissions on it */ | ||
372 | error = permission(nd->dentry->d_inode, MAY_READ, NULL); | ||
373 | if (error) | ||
374 | path_release (nd); | ||
375 | return error; | ||
376 | } | ||
377 | |||
378 | /* | ||
379 | * create_watch - creates a watch on the given device. | ||
380 | * | ||
381 | * Callers must hold dev->sem. Calls inotify_dev_get_wd() so may sleep. | ||
382 | * Both 'dev' and 'inode' (by way of nameidata) need to be pinned. | ||
383 | */ | ||
384 | static struct inotify_watch *create_watch(struct inotify_device *dev, | ||
385 | u32 mask, struct inode *inode) | ||
386 | { | ||
387 | struct inotify_watch *watch; | ||
388 | int ret; | ||
389 | |||
390 | if (atomic_read(&dev->user->inotify_watches) >= inotify_max_user_watches) | ||
391 | return ERR_PTR(-ENOSPC); | ||
392 | |||
393 | watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL); | ||
394 | if (unlikely(!watch)) | ||
395 | return ERR_PTR(-ENOMEM); | ||
396 | |||
397 | ret = inotify_dev_get_wd(dev, watch); | ||
398 | if (unlikely(ret)) { | ||
399 | kmem_cache_free(watch_cachep, watch); | ||
400 | return ERR_PTR(ret); | ||
401 | } | ||
402 | |||
403 | watch->mask = mask; | ||
404 | atomic_set(&watch->count, 0); | ||
405 | INIT_LIST_HEAD(&watch->d_list); | ||
406 | INIT_LIST_HEAD(&watch->i_list); | ||
407 | |||
408 | /* save a reference to device and bump the count to make it official */ | ||
409 | get_inotify_dev(dev); | ||
410 | watch->dev = dev; | ||
411 | |||
412 | /* | ||
413 | * Save a reference to the inode and bump the ref count to make it | ||
414 | * official. We hold a reference to nameidata, which makes this safe. | ||
415 | */ | ||
416 | watch->inode = igrab(inode); | ||
417 | |||
418 | /* bump our own count, corresponding to our entry in dev->watches */ | ||
419 | get_inotify_watch(watch); | ||
420 | |||
421 | atomic_inc(&dev->user->inotify_watches); | ||
422 | |||
423 | return watch; | ||
424 | } | ||
425 | |||
426 | /* | ||
427 | * inotify_find_dev - find the watch associated with the given inode and dev | ||
428 | * | ||
429 | * Callers must hold inode->inotify_sem. | ||
430 | */ | ||
431 | static struct inotify_watch *inode_find_dev(struct inode *inode, | ||
432 | struct inotify_device *dev) | ||
433 | { | ||
434 | struct inotify_watch *watch; | ||
435 | |||
436 | list_for_each_entry(watch, &inode->inotify_watches, i_list) { | ||
437 | if (watch->dev == dev) | ||
438 | return watch; | ||
439 | } | ||
440 | |||
441 | return NULL; | ||
442 | } | ||
443 | |||
444 | /* | ||
445 | * remove_watch_no_event - remove_watch() without the IN_IGNORED event. | ||
446 | */ | ||
447 | static void remove_watch_no_event(struct inotify_watch *watch, | ||
448 | struct inotify_device *dev) | ||
449 | { | ||
450 | list_del(&watch->i_list); | ||
451 | list_del(&watch->d_list); | ||
452 | |||
453 | atomic_dec(&dev->user->inotify_watches); | ||
454 | idr_remove(&dev->idr, watch->wd); | ||
455 | put_inotify_watch(watch); | ||
456 | } | ||
457 | |||
458 | /* | ||
459 | * remove_watch - Remove a watch from both the device and the inode. Sends | ||
460 | * the IN_IGNORED event to the given device signifying that the inode is no | ||
461 | * longer watched. | ||
462 | * | ||
463 | * Callers must hold both inode->inotify_sem and dev->sem. We drop a | ||
464 | * reference to the inode before returning. | ||
465 | * | ||
466 | * The inode is not iput() so as to remain atomic. If the inode needs to be | ||
467 | * iput(), the call returns one. Otherwise, it returns zero. | ||
468 | */ | ||
469 | static void remove_watch(struct inotify_watch *watch,struct inotify_device *dev) | ||
470 | { | ||
471 | inotify_dev_queue_event(dev, watch, IN_IGNORED, 0, NULL); | ||
472 | remove_watch_no_event(watch, dev); | ||
473 | } | ||
474 | |||
475 | /* | ||
476 | * inotify_inode_watched - returns nonzero if there are watches on this inode | ||
477 | * and zero otherwise. We call this lockless, we do not care if we race. | ||
478 | */ | ||
479 | static inline int inotify_inode_watched(struct inode *inode) | ||
480 | { | ||
481 | return !list_empty(&inode->inotify_watches); | ||
482 | } | ||
483 | |||
484 | /* Kernel API */ | ||
485 | |||
486 | /** | ||
487 | * inotify_inode_queue_event - queue an event to all watches on this inode | ||
488 | * @inode: inode event is originating from | ||
489 | * @mask: event mask describing this event | ||
490 | * @cookie: cookie for synchronization, or zero | ||
491 | * @name: filename, if any | ||
492 | */ | ||
493 | void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie, | ||
494 | const char *name) | ||
495 | { | ||
496 | struct inotify_watch *watch, *next; | ||
497 | |||
498 | if (!inotify_inode_watched(inode)) | ||
499 | return; | ||
500 | |||
501 | down(&inode->inotify_sem); | ||
502 | list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) { | ||
503 | u32 watch_mask = watch->mask; | ||
504 | if (watch_mask & mask) { | ||
505 | struct inotify_device *dev = watch->dev; | ||
506 | get_inotify_watch(watch); | ||
507 | down(&dev->sem); | ||
508 | inotify_dev_queue_event(dev, watch, mask, cookie, name); | ||
509 | if (watch_mask & IN_ONESHOT) | ||
510 | remove_watch_no_event(watch, dev); | ||
511 | up(&dev->sem); | ||
512 | put_inotify_watch(watch); | ||
513 | } | ||
514 | } | ||
515 | up(&inode->inotify_sem); | ||
516 | } | ||
517 | EXPORT_SYMBOL_GPL(inotify_inode_queue_event); | ||
518 | |||
519 | /** | ||
520 | * inotify_dentry_parent_queue_event - queue an event to a dentry's parent | ||
521 | * @dentry: the dentry in question, we queue against this dentry's parent | ||
522 | * @mask: event mask describing this event | ||
523 | * @cookie: cookie for synchronization, or zero | ||
524 | * @name: filename, if any | ||
525 | */ | ||
526 | void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask, | ||
527 | u32 cookie, const char *name) | ||
528 | { | ||
529 | struct dentry *parent; | ||
530 | struct inode *inode; | ||
531 | |||
532 | spin_lock(&dentry->d_lock); | ||
533 | parent = dentry->d_parent; | ||
534 | inode = parent->d_inode; | ||
535 | |||
536 | if (inotify_inode_watched(inode)) { | ||
537 | dget(parent); | ||
538 | spin_unlock(&dentry->d_lock); | ||
539 | inotify_inode_queue_event(inode, mask, cookie, name); | ||
540 | dput(parent); | ||
541 | } else | ||
542 | spin_unlock(&dentry->d_lock); | ||
543 | } | ||
544 | EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event); | ||
545 | |||
546 | /** | ||
547 | * inotify_get_cookie - return a unique cookie for use in synchronizing events. | ||
548 | */ | ||
549 | u32 inotify_get_cookie(void) | ||
550 | { | ||
551 | return atomic_inc_return(&inotify_cookie); | ||
552 | } | ||
553 | EXPORT_SYMBOL_GPL(inotify_get_cookie); | ||
554 | |||
555 | /** | ||
556 | * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes. | ||
557 | * @list: list of inodes being unmounted (sb->s_inodes) | ||
558 | * | ||
559 | * Called with inode_lock held, protecting the unmounting super block's list | ||
560 | * of inodes, and with iprune_sem held, keeping shrink_icache_memory() at bay. | ||
561 | * We temporarily drop inode_lock, however, and CAN block. | ||
562 | */ | ||
563 | void inotify_unmount_inodes(struct list_head *list) | ||
564 | { | ||
565 | struct inode *inode, *next_i, *need_iput = NULL; | ||
566 | |||
567 | list_for_each_entry_safe(inode, next_i, list, i_sb_list) { | ||
568 | struct inotify_watch *watch, *next_w; | ||
569 | struct inode *need_iput_tmp; | ||
570 | struct list_head *watches; | ||
571 | |||
572 | /* | ||
573 | * If i_count is zero, the inode cannot have any watches and | ||
574 | * doing an __iget/iput with MS_ACTIVE clear would actually | ||
575 | * evict all inodes with zero i_count from icache which is | ||
576 | * unnecessarily violent and may in fact be illegal to do. | ||
577 | */ | ||
578 | if (!atomic_read(&inode->i_count)) | ||
579 | continue; | ||
580 | |||
581 | /* | ||
582 | * We cannot __iget() an inode in state I_CLEAR, I_FREEING, or | ||
583 | * I_WILL_FREE which is fine because by that point the inode | ||
584 | * cannot have any associated watches. | ||
585 | */ | ||
586 | if (inode->i_state & (I_CLEAR | I_FREEING | I_WILL_FREE)) | ||
587 | continue; | ||
588 | |||
589 | need_iput_tmp = need_iput; | ||
590 | need_iput = NULL; | ||
591 | /* In case the remove_watch() drops a reference. */ | ||
592 | if (inode != need_iput_tmp) | ||
593 | __iget(inode); | ||
594 | else | ||
595 | need_iput_tmp = NULL; | ||
596 | /* In case the dropping of a reference would nuke next_i. */ | ||
597 | if ((&next_i->i_sb_list != list) && | ||
598 | atomic_read(&next_i->i_count) && | ||
599 | !(next_i->i_state & (I_CLEAR | I_FREEING | | ||
600 | I_WILL_FREE))) { | ||
601 | __iget(next_i); | ||
602 | need_iput = next_i; | ||
603 | } | ||
604 | |||
605 | /* | ||
606 | * We can safely drop inode_lock here because we hold | ||
607 | * references on both inode and next_i. Also no new inodes | ||
608 | * will be added since the umount has begun. Finally, | ||
609 | * iprune_sem keeps shrink_icache_memory() away. | ||
610 | */ | ||
611 | spin_unlock(&inode_lock); | ||
612 | |||
613 | if (need_iput_tmp) | ||
614 | iput(need_iput_tmp); | ||
615 | |||
616 | /* for each watch, send IN_UNMOUNT and then remove it */ | ||
617 | down(&inode->inotify_sem); | ||
618 | watches = &inode->inotify_watches; | ||
619 | list_for_each_entry_safe(watch, next_w, watches, i_list) { | ||
620 | struct inotify_device *dev = watch->dev; | ||
621 | down(&dev->sem); | ||
622 | inotify_dev_queue_event(dev, watch, IN_UNMOUNT,0,NULL); | ||
623 | remove_watch(watch, dev); | ||
624 | up(&dev->sem); | ||
625 | } | ||
626 | up(&inode->inotify_sem); | ||
627 | iput(inode); | ||
628 | |||
629 | spin_lock(&inode_lock); | ||
630 | } | ||
631 | } | ||
632 | EXPORT_SYMBOL_GPL(inotify_unmount_inodes); | ||
633 | |||
634 | /** | ||
635 | * inotify_inode_is_dead - an inode has been deleted, cleanup any watches | ||
636 | * @inode: inode that is about to be removed | ||
637 | */ | ||
638 | void inotify_inode_is_dead(struct inode *inode) | ||
639 | { | ||
640 | struct inotify_watch *watch, *next; | ||
641 | |||
642 | down(&inode->inotify_sem); | ||
643 | list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) { | ||
644 | struct inotify_device *dev = watch->dev; | ||
645 | down(&dev->sem); | ||
646 | remove_watch(watch, dev); | ||
647 | up(&dev->sem); | ||
648 | } | ||
649 | up(&inode->inotify_sem); | ||
650 | } | ||
651 | EXPORT_SYMBOL_GPL(inotify_inode_is_dead); | ||
652 | |||
653 | /* Device Interface */ | ||
654 | |||
655 | static unsigned int inotify_poll(struct file *file, poll_table *wait) | ||
656 | { | ||
657 | struct inotify_device *dev = file->private_data; | ||
658 | int ret = 0; | ||
659 | |||
660 | poll_wait(file, &dev->wq, wait); | ||
661 | down(&dev->sem); | ||
662 | if (!list_empty(&dev->events)) | ||
663 | ret = POLLIN | POLLRDNORM; | ||
664 | up(&dev->sem); | ||
665 | |||
666 | return ret; | ||
667 | } | ||
668 | |||
669 | static ssize_t inotify_read(struct file *file, char __user *buf, | ||
670 | size_t count, loff_t *pos) | ||
671 | { | ||
672 | size_t event_size = sizeof (struct inotify_event); | ||
673 | struct inotify_device *dev; | ||
674 | char __user *start; | ||
675 | int ret; | ||
676 | DEFINE_WAIT(wait); | ||
677 | |||
678 | start = buf; | ||
679 | dev = file->private_data; | ||
680 | |||
681 | while (1) { | ||
682 | int events; | ||
683 | |||
684 | prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE); | ||
685 | |||
686 | down(&dev->sem); | ||
687 | events = !list_empty(&dev->events); | ||
688 | up(&dev->sem); | ||
689 | if (events) { | ||
690 | ret = 0; | ||
691 | break; | ||
692 | } | ||
693 | |||
694 | if (file->f_flags & O_NONBLOCK) { | ||
695 | ret = -EAGAIN; | ||
696 | break; | ||
697 | } | ||
698 | |||
699 | if (signal_pending(current)) { | ||
700 | ret = -EINTR; | ||
701 | break; | ||
702 | } | ||
703 | |||
704 | schedule(); | ||
705 | } | ||
706 | |||
707 | finish_wait(&dev->wq, &wait); | ||
708 | if (ret) | ||
709 | return ret; | ||
710 | |||
711 | down(&dev->sem); | ||
712 | while (1) { | ||
713 | struct inotify_kernel_event *kevent; | ||
714 | |||
715 | ret = buf - start; | ||
716 | if (list_empty(&dev->events)) | ||
717 | break; | ||
718 | |||
719 | kevent = inotify_dev_get_event(dev); | ||
720 | if (event_size + kevent->event.len > count) | ||
721 | break; | ||
722 | |||
723 | if (copy_to_user(buf, &kevent->event, event_size)) { | ||
724 | ret = -EFAULT; | ||
725 | break; | ||
726 | } | ||
727 | buf += event_size; | ||
728 | count -= event_size; | ||
729 | |||
730 | if (kevent->name) { | ||
731 | if (copy_to_user(buf, kevent->name, kevent->event.len)){ | ||
732 | ret = -EFAULT; | ||
733 | break; | ||
734 | } | ||
735 | buf += kevent->event.len; | ||
736 | count -= kevent->event.len; | ||
737 | } | ||
738 | |||
739 | remove_kevent(dev, kevent); | ||
740 | } | ||
741 | up(&dev->sem); | ||
742 | |||
743 | return ret; | ||
744 | } | ||
745 | |||
746 | static int inotify_release(struct inode *ignored, struct file *file) | ||
747 | { | ||
748 | struct inotify_device *dev = file->private_data; | ||
749 | |||
750 | /* | ||
751 | * Destroy all of the watches on this device. Unfortunately, not very | ||
752 | * pretty. We cannot do a simple iteration over the list, because we | ||
753 | * do not know the inode until we iterate to the watch. But we need to | ||
754 | * hold inode->inotify_sem before dev->sem. The following works. | ||
755 | */ | ||
756 | while (1) { | ||
757 | struct inotify_watch *watch; | ||
758 | struct list_head *watches; | ||
759 | struct inode *inode; | ||
760 | |||
761 | down(&dev->sem); | ||
762 | watches = &dev->watches; | ||
763 | if (list_empty(watches)) { | ||
764 | up(&dev->sem); | ||
765 | break; | ||
766 | } | ||
767 | watch = list_entry(watches->next, struct inotify_watch, d_list); | ||
768 | get_inotify_watch(watch); | ||
769 | up(&dev->sem); | ||
770 | |||
771 | inode = watch->inode; | ||
772 | down(&inode->inotify_sem); | ||
773 | down(&dev->sem); | ||
774 | remove_watch_no_event(watch, dev); | ||
775 | up(&dev->sem); | ||
776 | up(&inode->inotify_sem); | ||
777 | put_inotify_watch(watch); | ||
778 | } | ||
779 | |||
780 | /* destroy all of the events on this device */ | ||
781 | down(&dev->sem); | ||
782 | while (!list_empty(&dev->events)) | ||
783 | inotify_dev_event_dequeue(dev); | ||
784 | up(&dev->sem); | ||
785 | |||
786 | /* free this device: the put matching the get in inotify_open() */ | ||
787 | put_inotify_dev(dev); | ||
788 | |||
789 | return 0; | ||
790 | } | ||
791 | |||
792 | /* | ||
793 | * inotify_ignore - handle the INOTIFY_IGNORE ioctl, asking that a given wd be | ||
794 | * removed from the device. | ||
795 | * | ||
796 | * Can sleep. | ||
797 | */ | ||
798 | static int inotify_ignore(struct inotify_device *dev, s32 wd) | ||
799 | { | ||
800 | struct inotify_watch *watch; | ||
801 | struct inode *inode; | ||
802 | |||
803 | down(&dev->sem); | ||
804 | watch = idr_find(&dev->idr, wd); | ||
805 | if (unlikely(!watch)) { | ||
806 | up(&dev->sem); | ||
807 | return -EINVAL; | ||
808 | } | ||
809 | get_inotify_watch(watch); | ||
810 | inode = watch->inode; | ||
811 | up(&dev->sem); | ||
812 | |||
813 | down(&inode->inotify_sem); | ||
814 | down(&dev->sem); | ||
815 | |||
816 | /* make sure that we did not race */ | ||
817 | watch = idr_find(&dev->idr, wd); | ||
818 | if (likely(watch)) | ||
819 | remove_watch(watch, dev); | ||
820 | |||
821 | up(&dev->sem); | ||
822 | up(&inode->inotify_sem); | ||
823 | put_inotify_watch(watch); | ||
824 | |||
825 | return 0; | ||
826 | } | ||
827 | |||
828 | static long inotify_ioctl(struct file *file, unsigned int cmd, | ||
829 | unsigned long arg) | ||
830 | { | ||
831 | struct inotify_device *dev; | ||
832 | void __user *p; | ||
833 | int ret = -ENOTTY; | ||
834 | |||
835 | dev = file->private_data; | ||
836 | p = (void __user *) arg; | ||
837 | |||
838 | switch (cmd) { | ||
839 | case FIONREAD: | ||
840 | ret = put_user(dev->queue_size, (int __user *) p); | ||
841 | break; | ||
842 | } | ||
843 | |||
844 | return ret; | ||
845 | } | ||
846 | |||
847 | static struct file_operations inotify_fops = { | ||
848 | .poll = inotify_poll, | ||
849 | .read = inotify_read, | ||
850 | .release = inotify_release, | ||
851 | .unlocked_ioctl = inotify_ioctl, | ||
852 | .compat_ioctl = inotify_ioctl, | ||
853 | }; | ||
854 | |||
855 | asmlinkage long sys_inotify_init(void) | ||
856 | { | ||
857 | struct inotify_device *dev; | ||
858 | struct user_struct *user; | ||
859 | int ret = -ENOTTY; | ||
860 | int fd; | ||
861 | struct file *filp; | ||
862 | |||
863 | fd = get_unused_fd(); | ||
864 | if (fd < 0) { | ||
865 | ret = fd; | ||
866 | goto out; | ||
867 | } | ||
868 | |||
869 | filp = get_empty_filp(); | ||
870 | if (!filp) { | ||
871 | put_unused_fd(fd); | ||
872 | ret = -ENFILE; | ||
873 | goto out; | ||
874 | } | ||
875 | filp->f_op = &inotify_fops; | ||
876 | filp->f_vfsmnt = mntget(inotify_mnt); | ||
877 | filp->f_dentry = dget(inotify_mnt->mnt_root); | ||
878 | filp->f_mapping = filp->f_dentry->d_inode->i_mapping; | ||
879 | filp->f_mode = FMODE_READ; | ||
880 | filp->f_flags = O_RDONLY; | ||
881 | |||
882 | user = get_uid(current->user); | ||
883 | |||
884 | if (unlikely(atomic_read(&user->inotify_devs) >= inotify_max_user_instances)) { | ||
885 | ret = -EMFILE; | ||
886 | goto out_err; | ||
887 | } | ||
888 | |||
889 | dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL); | ||
890 | if (unlikely(!dev)) { | ||
891 | ret = -ENOMEM; | ||
892 | goto out_err; | ||
893 | } | ||
894 | |||
895 | idr_init(&dev->idr); | ||
896 | INIT_LIST_HEAD(&dev->events); | ||
897 | INIT_LIST_HEAD(&dev->watches); | ||
898 | init_waitqueue_head(&dev->wq); | ||
899 | sema_init(&dev->sem, 1); | ||
900 | dev->event_count = 0; | ||
901 | dev->queue_size = 0; | ||
902 | dev->max_events = inotify_max_queued_events; | ||
903 | dev->user = user; | ||
904 | atomic_set(&dev->count, 0); | ||
905 | |||
906 | get_inotify_dev(dev); | ||
907 | atomic_inc(&user->inotify_devs); | ||
908 | |||
909 | filp->private_data = dev; | ||
910 | fd_install (fd, filp); | ||
911 | return fd; | ||
912 | out_err: | ||
913 | put_unused_fd (fd); | ||
914 | put_filp (filp); | ||
915 | free_uid(user); | ||
916 | out: | ||
917 | return ret; | ||
918 | } | ||
919 | |||
920 | asmlinkage long sys_inotify_add_watch(int fd, const char *path, u32 mask) | ||
921 | { | ||
922 | struct inotify_watch *watch, *old; | ||
923 | struct inode *inode; | ||
924 | struct inotify_device *dev; | ||
925 | struct nameidata nd; | ||
926 | struct file *filp; | ||
927 | int ret; | ||
928 | |||
929 | filp = fget(fd); | ||
930 | if (!filp) | ||
931 | return -EBADF; | ||
932 | |||
933 | dev = filp->private_data; | ||
934 | |||
935 | ret = find_inode((const char __user*) path, &nd); | ||
936 | if (ret) | ||
937 | goto fput_and_out; | ||
938 | |||
939 | /* Held in place by reference in nd */ | ||
940 | inode = nd.dentry->d_inode; | ||
941 | |||
942 | down(&inode->inotify_sem); | ||
943 | down(&dev->sem); | ||
944 | |||
945 | /* don't let user-space set invalid bits: we don't want flags set */ | ||
946 | mask &= IN_ALL_EVENTS; | ||
947 | if (!mask) { | ||
948 | ret = -EINVAL; | ||
949 | goto out; | ||
950 | } | ||
951 | |||
952 | /* | ||
953 | * Handle the case of re-adding a watch on an (inode,dev) pair that we | ||
954 | * are already watching. We just update the mask and return its wd. | ||
955 | */ | ||
956 | old = inode_find_dev(inode, dev); | ||
957 | if (unlikely(old)) { | ||
958 | old->mask = mask; | ||
959 | ret = old->wd; | ||
960 | goto out; | ||
961 | } | ||
962 | |||
963 | watch = create_watch(dev, mask, inode); | ||
964 | if (unlikely(IS_ERR(watch))) { | ||
965 | ret = PTR_ERR(watch); | ||
966 | goto out; | ||
967 | } | ||
968 | |||
969 | /* Add the watch to the device's and the inode's list */ | ||
970 | list_add(&watch->d_list, &dev->watches); | ||
971 | list_add(&watch->i_list, &inode->inotify_watches); | ||
972 | ret = watch->wd; | ||
973 | out: | ||
974 | path_release (&nd); | ||
975 | up(&dev->sem); | ||
976 | up(&inode->inotify_sem); | ||
977 | fput_and_out: | ||
978 | fput(filp); | ||
979 | return ret; | ||
980 | } | ||
981 | |||
982 | asmlinkage long sys_inotify_rm_watch(int fd, u32 wd) | ||
983 | { | ||
984 | struct file *filp; | ||
985 | struct inotify_device *dev; | ||
986 | int ret; | ||
987 | |||
988 | filp = fget(fd); | ||
989 | if (!filp) | ||
990 | return -EBADF; | ||
991 | dev = filp->private_data; | ||
992 | ret = inotify_ignore(dev, wd); | ||
993 | fput(filp); | ||
994 | |||
995 | return ret; | ||
996 | } | ||
997 | |||
998 | static struct super_block * | ||
999 | inotify_get_sb(struct file_system_type *fs_type, int flags, | ||
1000 | const char *dev_name, void *data) | ||
1001 | { | ||
1002 | return get_sb_pseudo(fs_type, "inotify", NULL, 0xBAD1DEA); | ||
1003 | } | ||
1004 | |||
1005 | static struct file_system_type inotify_fs_type = { | ||
1006 | .name = "inotifyfs", | ||
1007 | .get_sb = inotify_get_sb, | ||
1008 | .kill_sb = kill_anon_super, | ||
1009 | }; | ||
1010 | |||
1011 | /* | ||
1012 | * inotify_init - Our initialization function. Note that we cannnot return | ||
1013 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here | ||
1014 | * must result in panic(). | ||
1015 | */ | ||
1016 | static int __init inotify_init(void) | ||
1017 | { | ||
1018 | register_filesystem(&inotify_fs_type); | ||
1019 | inotify_mnt = kern_mount(&inotify_fs_type); | ||
1020 | |||
1021 | inotify_max_queued_events = 8192; | ||
1022 | inotify_max_user_instances = 8; | ||
1023 | inotify_max_user_watches = 8192; | ||
1024 | |||
1025 | atomic_set(&inotify_cookie, 0); | ||
1026 | |||
1027 | watch_cachep = kmem_cache_create("inotify_watch_cache", | ||
1028 | sizeof(struct inotify_watch), | ||
1029 | 0, SLAB_PANIC, NULL, NULL); | ||
1030 | event_cachep = kmem_cache_create("inotify_event_cache", | ||
1031 | sizeof(struct inotify_kernel_event), | ||
1032 | 0, SLAB_PANIC, NULL, NULL); | ||
1033 | |||
1034 | return 0; | ||
1035 | } | ||
1036 | |||
1037 | module_init(inotify_init); | ||