blob: 367c487c014b6f62db3c04811ad56e3bec4cc0df [file] [log] [blame]
Robert Love0eeca282005-07-12 17:06:03 -04001/*
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>
Robert Love0eeca282005-07-12 17:06:03 -040032#include <linux/init.h>
33#include <linux/list.h>
34#include <linux/writeback.h>
35#include <linux/inotify.h>
Arnd Bergmann5131cf12006-01-18 17:43:04 -080036#include <linux/syscalls.h>
Robert Love0eeca282005-07-12 17:06:03 -040037
38#include <asm/ioctls.h>
39
40static atomic_t inotify_cookie;
41
Eric Dumazetfa3536c2006-03-26 01:37:24 -080042static kmem_cache_t *watch_cachep __read_mostly;
43static kmem_cache_t *event_cachep __read_mostly;
Robert Love0eeca282005-07-12 17:06:03 -040044
Eric Dumazetfa3536c2006-03-26 01:37:24 -080045static struct vfsmount *inotify_mnt __read_mostly;
Robert Love0eeca282005-07-12 17:06:03 -040046
Robert Love0399cb02005-07-13 12:38:18 -040047/* these are configurable via /proc/sys/fs/inotify/ */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080048int inotify_max_user_instances __read_mostly;
49int inotify_max_user_watches __read_mostly;
50int inotify_max_queued_events __read_mostly;
Robert Love0eeca282005-07-12 17:06:03 -040051
52/*
53 * Lock ordering:
54 *
55 * dentry->d_lock (used to keep d_move() away from dentry->d_parent)
Ingo Molnarf24075b2006-03-23 03:00:34 -080056 * iprune_mutex (synchronize shrink_icache_memory())
Robert Love0eeca282005-07-12 17:06:03 -040057 * inode_lock (protects the super_block->s_inodes list)
Ingo Molnard4f9af92006-03-23 03:00:30 -080058 * inode->inotify_mutex (protects inode->inotify_watches and watches->i_list)
59 * inotify_dev->mutex (protects inotify_device and watches->d_list)
Robert Love0eeca282005-07-12 17:06:03 -040060 */
61
62/*
63 * Lifetimes of the three main data structures--inotify_device, inode, and
64 * inotify_watch--are managed by reference count.
65 *
Robert Loveb6807162005-07-25 15:07:13 -040066 * inotify_device: Lifetime is from inotify_init() until release. Additional
67 * references can bump the count via get_inotify_dev() and drop the count via
Robert Love0eeca282005-07-12 17:06:03 -040068 * put_inotify_dev().
69 *
70 * inotify_watch: Lifetime is from create_watch() to destory_watch().
71 * Additional references can bump the count via get_inotify_watch() and drop
72 * the count via put_inotify_watch().
73 *
74 * inode: Pinned so long as the inode is associated with a watch, from
75 * create_watch() to put_inotify_watch().
76 */
77
78/*
Robert Loveb6807162005-07-25 15:07:13 -040079 * struct inotify_device - represents an inotify instance
Robert Love0eeca282005-07-12 17:06:03 -040080 *
Ingo Molnard4f9af92006-03-23 03:00:30 -080081 * This structure is protected by the mutex 'mutex'.
Robert Love0eeca282005-07-12 17:06:03 -040082 */
83struct inotify_device {
84 wait_queue_head_t wq; /* wait queue for i/o */
85 struct idr idr; /* idr mapping wd -> watch */
Ingo Molnard4f9af92006-03-23 03:00:30 -080086 struct mutex mutex; /* protects this bad boy */
Robert Love0eeca282005-07-12 17:06:03 -040087 struct list_head events; /* list of queued events */
88 struct list_head watches; /* list of watches */
89 atomic_t count; /* reference count */
90 struct user_struct *user; /* user who opened this dev */
91 unsigned int queue_size; /* size of the queue (bytes) */
92 unsigned int event_count; /* number of pending events */
93 unsigned int max_events; /* maximum number of events */
John McCutchanb9c55d22005-08-01 11:00:45 -040094 u32 last_wd; /* the last wd allocated */
Robert Love0eeca282005-07-12 17:06:03 -040095};
96
97/*
98 * struct inotify_kernel_event - An inotify event, originating from a watch and
99 * queued for user-space. A list of these is attached to each instance of the
100 * device. In read(), this list is walked and all events that can fit in the
101 * buffer are returned.
102 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800103 * Protected by dev->mutex of the device in which we are queued.
Robert Love0eeca282005-07-12 17:06:03 -0400104 */
105struct inotify_kernel_event {
106 struct inotify_event event; /* the user-space event */
107 struct list_head list; /* entry in inotify_device's list */
108 char *name; /* filename, if any */
109};
110
111/*
112 * struct inotify_watch - represents a watch request on a specific inode
113 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800114 * d_list is protected by dev->mutex of the associated watch->dev.
115 * i_list and mask are protected by inode->inotify_mutex of the associated inode.
Robert Love0eeca282005-07-12 17:06:03 -0400116 * dev, inode, and wd are never written to once the watch is created.
117 */
118struct inotify_watch {
119 struct list_head d_list; /* entry in inotify_device's list */
120 struct list_head i_list; /* entry in inode's list */
121 atomic_t count; /* reference count */
122 struct inotify_device *dev; /* associated device */
123 struct inode *inode; /* associated inode */
124 s32 wd; /* watch descriptor */
125 u32 mask; /* event mask for this watch */
126};
127
Robert Love0399cb02005-07-13 12:38:18 -0400128#ifdef CONFIG_SYSCTL
129
130#include <linux/sysctl.h>
131
132static int zero;
133
134ctl_table inotify_table[] = {
135 {
136 .ctl_name = INOTIFY_MAX_USER_INSTANCES,
137 .procname = "max_user_instances",
138 .data = &inotify_max_user_instances,
139 .maxlen = sizeof(int),
140 .mode = 0644,
141 .proc_handler = &proc_dointvec_minmax,
142 .strategy = &sysctl_intvec,
143 .extra1 = &zero,
144 },
145 {
146 .ctl_name = INOTIFY_MAX_USER_WATCHES,
147 .procname = "max_user_watches",
148 .data = &inotify_max_user_watches,
149 .maxlen = sizeof(int),
150 .mode = 0644,
151 .proc_handler = &proc_dointvec_minmax,
152 .strategy = &sysctl_intvec,
153 .extra1 = &zero,
154 },
155 {
156 .ctl_name = INOTIFY_MAX_QUEUED_EVENTS,
157 .procname = "max_queued_events",
158 .data = &inotify_max_queued_events,
159 .maxlen = sizeof(int),
160 .mode = 0644,
161 .proc_handler = &proc_dointvec_minmax,
162 .strategy = &sysctl_intvec,
163 .extra1 = &zero
164 },
165 { .ctl_name = 0 }
166};
167#endif /* CONFIG_SYSCTL */
168
Robert Love0eeca282005-07-12 17:06:03 -0400169static inline void get_inotify_dev(struct inotify_device *dev)
170{
171 atomic_inc(&dev->count);
172}
173
174static inline void put_inotify_dev(struct inotify_device *dev)
175{
176 if (atomic_dec_and_test(&dev->count)) {
177 atomic_dec(&dev->user->inotify_devs);
178 free_uid(dev->user);
Andrew Morton8d3b3592005-10-23 12:57:18 -0700179 idr_destroy(&dev->idr);
Robert Love0eeca282005-07-12 17:06:03 -0400180 kfree(dev);
181 }
182}
183
184static inline void get_inotify_watch(struct inotify_watch *watch)
185{
186 atomic_inc(&watch->count);
187}
188
189/*
190 * put_inotify_watch - decrements the ref count on a given watch. cleans up
191 * the watch and its references if the count reaches zero.
192 */
193static inline void put_inotify_watch(struct inotify_watch *watch)
194{
195 if (atomic_dec_and_test(&watch->count)) {
196 put_inotify_dev(watch->dev);
197 iput(watch->inode);
198 kmem_cache_free(watch_cachep, watch);
199 }
200}
201
202/*
203 * kernel_event - create a new kernel event with the given parameters
204 *
205 * This function can sleep.
206 */
207static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
208 const char *name)
209{
210 struct inotify_kernel_event *kevent;
211
212 kevent = kmem_cache_alloc(event_cachep, GFP_KERNEL);
213 if (unlikely(!kevent))
214 return NULL;
215
216 /* we hand this out to user-space, so zero it just in case */
217 memset(&kevent->event, 0, sizeof(struct inotify_event));
218
219 kevent->event.wd = wd;
220 kevent->event.mask = mask;
221 kevent->event.cookie = cookie;
222
223 INIT_LIST_HEAD(&kevent->list);
224
225 if (name) {
226 size_t len, rem, event_size = sizeof(struct inotify_event);
227
228 /*
229 * We need to pad the filename so as to properly align an
230 * array of inotify_event structures. Because the structure is
231 * small and the common case is a small filename, we just round
232 * up to the next multiple of the structure's sizeof. This is
233 * simple and safe for all architectures.
234 */
235 len = strlen(name) + 1;
236 rem = event_size - len;
237 if (len > event_size) {
238 rem = event_size - (len % event_size);
239 if (len % event_size == 0)
240 rem = 0;
241 }
242
243 kevent->name = kmalloc(len + rem, GFP_KERNEL);
244 if (unlikely(!kevent->name)) {
245 kmem_cache_free(event_cachep, kevent);
246 return NULL;
247 }
248 memcpy(kevent->name, name, len);
249 if (rem)
250 memset(kevent->name + len, 0, rem);
251 kevent->event.len = len + rem;
252 } else {
253 kevent->event.len = 0;
254 kevent->name = NULL;
255 }
256
257 return kevent;
258}
259
260/*
261 * inotify_dev_get_event - return the next event in the given dev's queue
262 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800263 * Caller must hold dev->mutex.
Robert Love0eeca282005-07-12 17:06:03 -0400264 */
265static inline struct inotify_kernel_event *
266inotify_dev_get_event(struct inotify_device *dev)
267{
268 return list_entry(dev->events.next, struct inotify_kernel_event, list);
269}
270
271/*
272 * inotify_dev_queue_event - add a new event to the given device
273 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800274 * Caller must hold dev->mutex. Can sleep (calls kernel_event()).
Robert Love0eeca282005-07-12 17:06:03 -0400275 */
276static void inotify_dev_queue_event(struct inotify_device *dev,
277 struct inotify_watch *watch, u32 mask,
278 u32 cookie, const char *name)
279{
280 struct inotify_kernel_event *kevent, *last;
281
282 /* coalescing: drop this event if it is a dupe of the previous */
283 last = inotify_dev_get_event(dev);
284 if (last && last->event.mask == mask && last->event.wd == watch->wd &&
285 last->event.cookie == cookie) {
286 const char *lastname = last->name;
287
288 if (!name && !lastname)
289 return;
290 if (name && lastname && !strcmp(lastname, name))
291 return;
292 }
293
294 /* the queue overflowed and we already sent the Q_OVERFLOW event */
295 if (unlikely(dev->event_count > dev->max_events))
296 return;
297
298 /* if the queue overflows, we need to notify user space */
299 if (unlikely(dev->event_count == dev->max_events))
300 kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL);
301 else
302 kevent = kernel_event(watch->wd, mask, cookie, name);
303
304 if (unlikely(!kevent))
305 return;
306
307 /* queue the event and wake up anyone waiting */
308 dev->event_count++;
309 dev->queue_size += sizeof(struct inotify_event) + kevent->event.len;
310 list_add_tail(&kevent->list, &dev->events);
311 wake_up_interruptible(&dev->wq);
312}
313
314/*
315 * remove_kevent - cleans up and ultimately frees the given kevent
316 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800317 * Caller must hold dev->mutex.
Robert Love0eeca282005-07-12 17:06:03 -0400318 */
319static void remove_kevent(struct inotify_device *dev,
320 struct inotify_kernel_event *kevent)
321{
322 list_del(&kevent->list);
323
324 dev->event_count--;
325 dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
326
327 kfree(kevent->name);
328 kmem_cache_free(event_cachep, kevent);
329}
330
331/*
332 * inotify_dev_event_dequeue - destroy an event on the given device
333 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800334 * Caller must hold dev->mutex.
Robert Love0eeca282005-07-12 17:06:03 -0400335 */
336static void inotify_dev_event_dequeue(struct inotify_device *dev)
337{
338 if (!list_empty(&dev->events)) {
339 struct inotify_kernel_event *kevent;
340 kevent = inotify_dev_get_event(dev);
341 remove_kevent(dev, kevent);
342 }
343}
344
345/*
346 * inotify_dev_get_wd - returns the next WD for use by the given dev
347 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800348 * Callers must hold dev->mutex. This function can sleep.
Robert Love0eeca282005-07-12 17:06:03 -0400349 */
350static int inotify_dev_get_wd(struct inotify_device *dev,
351 struct inotify_watch *watch)
352{
353 int ret;
354
355 do {
356 if (unlikely(!idr_pre_get(&dev->idr, GFP_KERNEL)))
357 return -ENOSPC;
John McCutchan7c657f22005-08-26 14:02:04 -0400358 ret = idr_get_new_above(&dev->idr, watch, dev->last_wd+1, &watch->wd);
Robert Love0eeca282005-07-12 17:06:03 -0400359 } while (ret == -EAGAIN);
360
361 return ret;
362}
363
364/*
365 * find_inode - resolve a user-given path to a specific inode and return a nd
366 */
John McCutchan8140a502005-12-12 00:37:14 -0800367static int find_inode(const char __user *dirname, struct nameidata *nd,
368 unsigned flags)
Robert Love0eeca282005-07-12 17:06:03 -0400369{
370 int error;
371
John McCutchan8140a502005-12-12 00:37:14 -0800372 error = __user_walk(dirname, flags, nd);
Robert Love0eeca282005-07-12 17:06:03 -0400373 if (error)
374 return error;
375 /* you can only watch an inode if you have read permissions on it */
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800376 error = vfs_permission(nd, MAY_READ);
Robert Love0eeca282005-07-12 17:06:03 -0400377 if (error)
Robert Loveb6807162005-07-25 15:07:13 -0400378 path_release(nd);
Robert Love0eeca282005-07-12 17:06:03 -0400379 return error;
380}
381
382/*
Nick Pigginc32ccd82006-03-25 03:07:09 -0800383 * inotify_inode_watched - returns nonzero if there are watches on this inode
384 * and zero otherwise. We call this lockless, we do not care if we race.
385 */
386static inline int inotify_inode_watched(struct inode *inode)
387{
388 return !list_empty(&inode->inotify_watches);
389}
390
391/*
392 * Get child dentry flag into synch with parent inode.
393 * Flag should always be clear for negative dentrys.
394 */
395static void set_dentry_child_flags(struct inode *inode, int watched)
396{
397 struct dentry *alias;
398
399 spin_lock(&dcache_lock);
400 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
401 struct dentry *child;
402
403 list_for_each_entry(child, &alias->d_subdirs, d_u.d_child) {
404 if (!child->d_inode) {
405 WARN_ON(child->d_flags & DCACHE_INOTIFY_PARENT_WATCHED);
406 continue;
407 }
408 spin_lock(&child->d_lock);
409 if (watched) {
410 WARN_ON(child->d_flags &
411 DCACHE_INOTIFY_PARENT_WATCHED);
412 child->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
413 } else {
414 WARN_ON(!(child->d_flags &
415 DCACHE_INOTIFY_PARENT_WATCHED));
416 child->d_flags&=~DCACHE_INOTIFY_PARENT_WATCHED;
417 }
418 spin_unlock(&child->d_lock);
419 }
420 }
421 spin_unlock(&dcache_lock);
422}
423
424/*
Robert Love0eeca282005-07-12 17:06:03 -0400425 * create_watch - creates a watch on the given device.
426 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800427 * Callers must hold dev->mutex. Calls inotify_dev_get_wd() so may sleep.
Robert Love0eeca282005-07-12 17:06:03 -0400428 * Both 'dev' and 'inode' (by way of nameidata) need to be pinned.
429 */
430static struct inotify_watch *create_watch(struct inotify_device *dev,
431 u32 mask, struct inode *inode)
432{
433 struct inotify_watch *watch;
434 int ret;
435
Robert Loveb6807162005-07-25 15:07:13 -0400436 if (atomic_read(&dev->user->inotify_watches) >=
437 inotify_max_user_watches)
Robert Love0eeca282005-07-12 17:06:03 -0400438 return ERR_PTR(-ENOSPC);
439
440 watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL);
441 if (unlikely(!watch))
442 return ERR_PTR(-ENOMEM);
443
444 ret = inotify_dev_get_wd(dev, watch);
445 if (unlikely(ret)) {
446 kmem_cache_free(watch_cachep, watch);
447 return ERR_PTR(ret);
448 }
449
Robert Love0bf955c2005-08-15 12:27:54 -0400450 dev->last_wd = watch->wd;
Robert Love0eeca282005-07-12 17:06:03 -0400451 watch->mask = mask;
452 atomic_set(&watch->count, 0);
453 INIT_LIST_HEAD(&watch->d_list);
454 INIT_LIST_HEAD(&watch->i_list);
455
456 /* save a reference to device and bump the count to make it official */
457 get_inotify_dev(dev);
458 watch->dev = dev;
459
460 /*
461 * Save a reference to the inode and bump the ref count to make it
462 * official. We hold a reference to nameidata, which makes this safe.
463 */
464 watch->inode = igrab(inode);
465
466 /* bump our own count, corresponding to our entry in dev->watches */
467 get_inotify_watch(watch);
468
469 atomic_inc(&dev->user->inotify_watches);
470
471 return watch;
472}
473
474/*
475 * inotify_find_dev - find the watch associated with the given inode and dev
476 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800477 * Callers must hold inode->inotify_mutex.
Robert Love0eeca282005-07-12 17:06:03 -0400478 */
479static struct inotify_watch *inode_find_dev(struct inode *inode,
480 struct inotify_device *dev)
481{
482 struct inotify_watch *watch;
483
484 list_for_each_entry(watch, &inode->inotify_watches, i_list) {
485 if (watch->dev == dev)
486 return watch;
487 }
488
489 return NULL;
490}
491
492/*
493 * remove_watch_no_event - remove_watch() without the IN_IGNORED event.
494 */
495static void remove_watch_no_event(struct inotify_watch *watch,
496 struct inotify_device *dev)
497{
498 list_del(&watch->i_list);
499 list_del(&watch->d_list);
500
Nick Pigginc32ccd82006-03-25 03:07:09 -0800501 if (!inotify_inode_watched(watch->inode))
502 set_dentry_child_flags(watch->inode, 0);
503
Robert Love0eeca282005-07-12 17:06:03 -0400504 atomic_dec(&dev->user->inotify_watches);
505 idr_remove(&dev->idr, watch->wd);
506 put_inotify_watch(watch);
507}
508
509/*
510 * remove_watch - Remove a watch from both the device and the inode. Sends
511 * the IN_IGNORED event to the given device signifying that the inode is no
512 * longer watched.
513 *
Ingo Molnard4f9af92006-03-23 03:00:30 -0800514 * Callers must hold both inode->inotify_mutex and dev->mutex. We drop a
Robert Love0eeca282005-07-12 17:06:03 -0400515 * reference to the inode before returning.
516 *
517 * The inode is not iput() so as to remain atomic. If the inode needs to be
518 * iput(), the call returns one. Otherwise, it returns zero.
519 */
520static void remove_watch(struct inotify_watch *watch,struct inotify_device *dev)
521{
522 inotify_dev_queue_event(dev, watch, IN_IGNORED, 0, NULL);
523 remove_watch_no_event(watch, dev);
524}
525
Nick Pigginc32ccd82006-03-25 03:07:09 -0800526/* Kernel API */
527
Robert Love0eeca282005-07-12 17:06:03 -0400528/*
Nick Pigginc32ccd82006-03-25 03:07:09 -0800529 * inotify_d_instantiate - instantiate dcache entry for inode
Robert Love0eeca282005-07-12 17:06:03 -0400530 */
Nick Pigginc32ccd82006-03-25 03:07:09 -0800531void inotify_d_instantiate(struct dentry *entry, struct inode *inode)
Robert Love0eeca282005-07-12 17:06:03 -0400532{
Nick Pigginc32ccd82006-03-25 03:07:09 -0800533 struct dentry *parent;
534
535 if (!inode)
536 return;
537
538 WARN_ON(entry->d_flags & DCACHE_INOTIFY_PARENT_WATCHED);
539 spin_lock(&entry->d_lock);
540 parent = entry->d_parent;
541 if (inotify_inode_watched(parent->d_inode))
542 entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
543 spin_unlock(&entry->d_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400544}
545
Nick Pigginc32ccd82006-03-25 03:07:09 -0800546/*
547 * inotify_d_move - dcache entry has been moved
548 */
549void inotify_d_move(struct dentry *entry)
550{
551 struct dentry *parent;
552
553 parent = entry->d_parent;
554 if (inotify_inode_watched(parent->d_inode))
555 entry->d_flags |= DCACHE_INOTIFY_PARENT_WATCHED;
556 else
557 entry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
558}
Robert Love0eeca282005-07-12 17:06:03 -0400559
560/**
561 * inotify_inode_queue_event - queue an event to all watches on this inode
562 * @inode: inode event is originating from
563 * @mask: event mask describing this event
564 * @cookie: cookie for synchronization, or zero
565 * @name: filename, if any
566 */
567void inotify_inode_queue_event(struct inode *inode, u32 mask, u32 cookie,
568 const char *name)
569{
570 struct inotify_watch *watch, *next;
571
572 if (!inotify_inode_watched(inode))
573 return;
574
Ingo Molnard4f9af92006-03-23 03:00:30 -0800575 mutex_lock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400576 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
577 u32 watch_mask = watch->mask;
578 if (watch_mask & mask) {
579 struct inotify_device *dev = watch->dev;
580 get_inotify_watch(watch);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800581 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400582 inotify_dev_queue_event(dev, watch, mask, cookie, name);
583 if (watch_mask & IN_ONESHOT)
584 remove_watch_no_event(watch, dev);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800585 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400586 put_inotify_watch(watch);
587 }
588 }
Ingo Molnard4f9af92006-03-23 03:00:30 -0800589 mutex_unlock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400590}
591EXPORT_SYMBOL_GPL(inotify_inode_queue_event);
592
593/**
594 * inotify_dentry_parent_queue_event - queue an event to a dentry's parent
595 * @dentry: the dentry in question, we queue against this dentry's parent
596 * @mask: event mask describing this event
597 * @cookie: cookie for synchronization, or zero
598 * @name: filename, if any
599 */
600void inotify_dentry_parent_queue_event(struct dentry *dentry, u32 mask,
601 u32 cookie, const char *name)
602{
603 struct dentry *parent;
604 struct inode *inode;
605
Nick Pigginc32ccd82006-03-25 03:07:09 -0800606 if (!(dentry->d_flags & DCACHE_INOTIFY_PARENT_WATCHED))
John McCutchan820249b2005-09-06 15:16:38 -0700607 return;
608
Robert Love0eeca282005-07-12 17:06:03 -0400609 spin_lock(&dentry->d_lock);
610 parent = dentry->d_parent;
611 inode = parent->d_inode;
612
613 if (inotify_inode_watched(inode)) {
614 dget(parent);
615 spin_unlock(&dentry->d_lock);
616 inotify_inode_queue_event(inode, mask, cookie, name);
617 dput(parent);
618 } else
619 spin_unlock(&dentry->d_lock);
620}
621EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event);
622
623/**
624 * inotify_get_cookie - return a unique cookie for use in synchronizing events.
625 */
626u32 inotify_get_cookie(void)
627{
628 return atomic_inc_return(&inotify_cookie);
629}
630EXPORT_SYMBOL_GPL(inotify_get_cookie);
631
632/**
633 * inotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
634 * @list: list of inodes being unmounted (sb->s_inodes)
635 *
636 * Called with inode_lock held, protecting the unmounting super block's list
Ingo Molnarf24075b2006-03-23 03:00:34 -0800637 * of inodes, and with iprune_mutex held, keeping shrink_icache_memory() at bay.
Robert Love0eeca282005-07-12 17:06:03 -0400638 * We temporarily drop inode_lock, however, and CAN block.
639 */
640void inotify_unmount_inodes(struct list_head *list)
641{
642 struct inode *inode, *next_i, *need_iput = NULL;
643
644 list_for_each_entry_safe(inode, next_i, list, i_sb_list) {
645 struct inotify_watch *watch, *next_w;
646 struct inode *need_iput_tmp;
647 struct list_head *watches;
648
649 /*
650 * If i_count is zero, the inode cannot have any watches and
651 * doing an __iget/iput with MS_ACTIVE clear would actually
652 * evict all inodes with zero i_count from icache which is
653 * unnecessarily violent and may in fact be illegal to do.
654 */
655 if (!atomic_read(&inode->i_count))
656 continue;
657
658 /*
659 * We cannot __iget() an inode in state I_CLEAR, I_FREEING, or
660 * I_WILL_FREE which is fine because by that point the inode
661 * cannot have any associated watches.
662 */
663 if (inode->i_state & (I_CLEAR | I_FREEING | I_WILL_FREE))
664 continue;
665
666 need_iput_tmp = need_iput;
667 need_iput = NULL;
668 /* In case the remove_watch() drops a reference. */
669 if (inode != need_iput_tmp)
670 __iget(inode);
671 else
672 need_iput_tmp = NULL;
673 /* In case the dropping of a reference would nuke next_i. */
674 if ((&next_i->i_sb_list != list) &&
675 atomic_read(&next_i->i_count) &&
676 !(next_i->i_state & (I_CLEAR | I_FREEING |
677 I_WILL_FREE))) {
678 __iget(next_i);
679 need_iput = next_i;
680 }
681
682 /*
683 * We can safely drop inode_lock here because we hold
684 * references on both inode and next_i. Also no new inodes
685 * will be added since the umount has begun. Finally,
Ingo Molnarf24075b2006-03-23 03:00:34 -0800686 * iprune_mutex keeps shrink_icache_memory() away.
Robert Love0eeca282005-07-12 17:06:03 -0400687 */
688 spin_unlock(&inode_lock);
689
690 if (need_iput_tmp)
691 iput(need_iput_tmp);
692
693 /* for each watch, send IN_UNMOUNT and then remove it */
Ingo Molnard4f9af92006-03-23 03:00:30 -0800694 mutex_lock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400695 watches = &inode->inotify_watches;
696 list_for_each_entry_safe(watch, next_w, watches, i_list) {
697 struct inotify_device *dev = watch->dev;
Ingo Molnard4f9af92006-03-23 03:00:30 -0800698 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400699 inotify_dev_queue_event(dev, watch, IN_UNMOUNT,0,NULL);
700 remove_watch(watch, dev);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800701 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400702 }
Ingo Molnard4f9af92006-03-23 03:00:30 -0800703 mutex_unlock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400704 iput(inode);
705
706 spin_lock(&inode_lock);
707 }
708}
709EXPORT_SYMBOL_GPL(inotify_unmount_inodes);
710
711/**
712 * inotify_inode_is_dead - an inode has been deleted, cleanup any watches
713 * @inode: inode that is about to be removed
714 */
715void inotify_inode_is_dead(struct inode *inode)
716{
717 struct inotify_watch *watch, *next;
718
Ingo Molnard4f9af92006-03-23 03:00:30 -0800719 mutex_lock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400720 list_for_each_entry_safe(watch, next, &inode->inotify_watches, i_list) {
721 struct inotify_device *dev = watch->dev;
Ingo Molnard4f9af92006-03-23 03:00:30 -0800722 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400723 remove_watch(watch, dev);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800724 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400725 }
Ingo Molnard4f9af92006-03-23 03:00:30 -0800726 mutex_unlock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400727}
728EXPORT_SYMBOL_GPL(inotify_inode_is_dead);
729
730/* Device Interface */
731
732static unsigned int inotify_poll(struct file *file, poll_table *wait)
733{
734 struct inotify_device *dev = file->private_data;
735 int ret = 0;
736
737 poll_wait(file, &dev->wq, wait);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800738 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400739 if (!list_empty(&dev->events))
740 ret = POLLIN | POLLRDNORM;
Ingo Molnard4f9af92006-03-23 03:00:30 -0800741 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400742
743 return ret;
744}
745
746static ssize_t inotify_read(struct file *file, char __user *buf,
747 size_t count, loff_t *pos)
748{
749 size_t event_size = sizeof (struct inotify_event);
750 struct inotify_device *dev;
751 char __user *start;
752 int ret;
753 DEFINE_WAIT(wait);
754
755 start = buf;
756 dev = file->private_data;
757
758 while (1) {
759 int events;
760
761 prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE);
762
Ingo Molnard4f9af92006-03-23 03:00:30 -0800763 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400764 events = !list_empty(&dev->events);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800765 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400766 if (events) {
767 ret = 0;
768 break;
769 }
770
771 if (file->f_flags & O_NONBLOCK) {
772 ret = -EAGAIN;
773 break;
774 }
775
776 if (signal_pending(current)) {
777 ret = -EINTR;
778 break;
779 }
780
781 schedule();
782 }
783
784 finish_wait(&dev->wq, &wait);
785 if (ret)
786 return ret;
787
Ingo Molnard4f9af92006-03-23 03:00:30 -0800788 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400789 while (1) {
790 struct inotify_kernel_event *kevent;
791
792 ret = buf - start;
793 if (list_empty(&dev->events))
794 break;
795
796 kevent = inotify_dev_get_event(dev);
797 if (event_size + kevent->event.len > count)
798 break;
799
800 if (copy_to_user(buf, &kevent->event, event_size)) {
801 ret = -EFAULT;
802 break;
803 }
804 buf += event_size;
805 count -= event_size;
806
807 if (kevent->name) {
808 if (copy_to_user(buf, kevent->name, kevent->event.len)){
809 ret = -EFAULT;
810 break;
811 }
812 buf += kevent->event.len;
813 count -= kevent->event.len;
814 }
815
816 remove_kevent(dev, kevent);
817 }
Ingo Molnard4f9af92006-03-23 03:00:30 -0800818 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400819
820 return ret;
821}
822
823static int inotify_release(struct inode *ignored, struct file *file)
824{
825 struct inotify_device *dev = file->private_data;
826
827 /*
828 * Destroy all of the watches on this device. Unfortunately, not very
829 * pretty. We cannot do a simple iteration over the list, because we
830 * do not know the inode until we iterate to the watch. But we need to
Ingo Molnard4f9af92006-03-23 03:00:30 -0800831 * hold inode->inotify_mutex before dev->mutex. The following works.
Robert Love0eeca282005-07-12 17:06:03 -0400832 */
833 while (1) {
834 struct inotify_watch *watch;
835 struct list_head *watches;
836 struct inode *inode;
837
Ingo Molnard4f9af92006-03-23 03:00:30 -0800838 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400839 watches = &dev->watches;
840 if (list_empty(watches)) {
Ingo Molnard4f9af92006-03-23 03:00:30 -0800841 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400842 break;
843 }
844 watch = list_entry(watches->next, struct inotify_watch, d_list);
845 get_inotify_watch(watch);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800846 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400847
848 inode = watch->inode;
Ingo Molnard4f9af92006-03-23 03:00:30 -0800849 mutex_lock(&inode->inotify_mutex);
850 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400851 remove_watch_no_event(watch, dev);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800852 mutex_unlock(&dev->mutex);
853 mutex_unlock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400854 put_inotify_watch(watch);
855 }
856
857 /* destroy all of the events on this device */
Ingo Molnard4f9af92006-03-23 03:00:30 -0800858 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400859 while (!list_empty(&dev->events))
860 inotify_dev_event_dequeue(dev);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800861 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400862
Robert Loveb6807162005-07-25 15:07:13 -0400863 /* free this device: the put matching the get in inotify_init() */
Robert Love0eeca282005-07-12 17:06:03 -0400864 put_inotify_dev(dev);
865
866 return 0;
867}
868
869/*
Robert Loveb6807162005-07-25 15:07:13 -0400870 * inotify_ignore - remove a given wd from this inotify instance.
Robert Love0eeca282005-07-12 17:06:03 -0400871 *
872 * Can sleep.
873 */
874static int inotify_ignore(struct inotify_device *dev, s32 wd)
875{
876 struct inotify_watch *watch;
877 struct inode *inode;
878
Ingo Molnard4f9af92006-03-23 03:00:30 -0800879 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400880 watch = idr_find(&dev->idr, wd);
881 if (unlikely(!watch)) {
Ingo Molnard4f9af92006-03-23 03:00:30 -0800882 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400883 return -EINVAL;
884 }
885 get_inotify_watch(watch);
886 inode = watch->inode;
Ingo Molnard4f9af92006-03-23 03:00:30 -0800887 mutex_unlock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400888
Ingo Molnard4f9af92006-03-23 03:00:30 -0800889 mutex_lock(&inode->inotify_mutex);
890 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400891
892 /* make sure that we did not race */
893 watch = idr_find(&dev->idr, wd);
894 if (likely(watch))
895 remove_watch(watch, dev);
896
Ingo Molnard4f9af92006-03-23 03:00:30 -0800897 mutex_unlock(&dev->mutex);
898 mutex_unlock(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400899 put_inotify_watch(watch);
900
901 return 0;
902}
903
904static long inotify_ioctl(struct file *file, unsigned int cmd,
905 unsigned long arg)
906{
907 struct inotify_device *dev;
908 void __user *p;
909 int ret = -ENOTTY;
910
911 dev = file->private_data;
912 p = (void __user *) arg;
913
914 switch (cmd) {
915 case FIONREAD:
916 ret = put_user(dev->queue_size, (int __user *) p);
917 break;
918 }
919
920 return ret;
921}
922
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800923static const struct file_operations inotify_fops = {
Robert Love0eeca282005-07-12 17:06:03 -0400924 .poll = inotify_poll,
925 .read = inotify_read,
926 .release = inotify_release,
927 .unlocked_ioctl = inotify_ioctl,
928 .compat_ioctl = inotify_ioctl,
929};
930
931asmlinkage long sys_inotify_init(void)
932{
933 struct inotify_device *dev;
934 struct user_struct *user;
Robert Loveb6807162005-07-25 15:07:13 -0400935 struct file *filp;
936 int fd, ret;
Robert Love0eeca282005-07-12 17:06:03 -0400937
938 fd = get_unused_fd();
Robert Loveb6807162005-07-25 15:07:13 -0400939 if (fd < 0)
940 return fd;
Robert Love0eeca282005-07-12 17:06:03 -0400941
942 filp = get_empty_filp();
943 if (!filp) {
Robert Love0eeca282005-07-12 17:06:03 -0400944 ret = -ENFILE;
Robert Love5eb22cb2005-07-25 15:12:19 -0400945 goto out_put_fd;
Robert Love0eeca282005-07-12 17:06:03 -0400946 }
Robert Love0eeca282005-07-12 17:06:03 -0400947
948 user = get_uid(current->user);
Robert Loveb6807162005-07-25 15:07:13 -0400949 if (unlikely(atomic_read(&user->inotify_devs) >=
950 inotify_max_user_instances)) {
Robert Love0eeca282005-07-12 17:06:03 -0400951 ret = -EMFILE;
Robert Love5eb22cb2005-07-25 15:12:19 -0400952 goto out_free_uid;
Robert Love0eeca282005-07-12 17:06:03 -0400953 }
954
955 dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL);
956 if (unlikely(!dev)) {
957 ret = -ENOMEM;
Robert Love5eb22cb2005-07-25 15:12:19 -0400958 goto out_free_uid;
Robert Love0eeca282005-07-12 17:06:03 -0400959 }
960
Robert Loveb6807162005-07-25 15:07:13 -0400961 filp->f_op = &inotify_fops;
962 filp->f_vfsmnt = mntget(inotify_mnt);
963 filp->f_dentry = dget(inotify_mnt->mnt_root);
964 filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
965 filp->f_mode = FMODE_READ;
966 filp->f_flags = O_RDONLY;
967 filp->private_data = dev;
968
Robert Love0eeca282005-07-12 17:06:03 -0400969 idr_init(&dev->idr);
970 INIT_LIST_HEAD(&dev->events);
971 INIT_LIST_HEAD(&dev->watches);
972 init_waitqueue_head(&dev->wq);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800973 mutex_init(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400974 dev->event_count = 0;
975 dev->queue_size = 0;
976 dev->max_events = inotify_max_queued_events;
977 dev->user = user;
John McCutchanb9c55d22005-08-01 11:00:45 -0400978 dev->last_wd = 0;
Robert Love0eeca282005-07-12 17:06:03 -0400979 atomic_set(&dev->count, 0);
980
981 get_inotify_dev(dev);
982 atomic_inc(&user->inotify_devs);
Robert Loveb6807162005-07-25 15:07:13 -0400983 fd_install(fd, filp);
Robert Love0eeca282005-07-12 17:06:03 -0400984
Robert Love0eeca282005-07-12 17:06:03 -0400985 return fd;
Robert Love5eb22cb2005-07-25 15:12:19 -0400986out_free_uid:
Robert Love0eeca282005-07-12 17:06:03 -0400987 free_uid(user);
Robert Love5eb22cb2005-07-25 15:12:19 -0400988 put_filp(filp);
989out_put_fd:
990 put_unused_fd(fd);
Robert Love0eeca282005-07-12 17:06:03 -0400991 return ret;
992}
993
Robert Loveb6807162005-07-25 15:07:13 -0400994asmlinkage long sys_inotify_add_watch(int fd, const char __user *path, u32 mask)
Robert Love0eeca282005-07-12 17:06:03 -0400995{
996 struct inotify_watch *watch, *old;
997 struct inode *inode;
998 struct inotify_device *dev;
999 struct nameidata nd;
1000 struct file *filp;
Robert Love33ea2f52005-07-25 15:08:37 -04001001 int ret, fput_needed;
John McCutchan7ea60402005-09-06 15:18:02 -07001002 int mask_add = 0;
John McCutchan8140a502005-12-12 00:37:14 -08001003 unsigned flags = 0;
Robert Love0eeca282005-07-12 17:06:03 -04001004
Robert Love33ea2f52005-07-25 15:08:37 -04001005 filp = fget_light(fd, &fput_needed);
1006 if (unlikely(!filp))
Robert Love0eeca282005-07-12 17:06:03 -04001007 return -EBADF;
1008
Robert Love783bc292005-07-25 15:10:08 -04001009 /* verify that this is indeed an inotify instance */
1010 if (unlikely(filp->f_op != &inotify_fops)) {
1011 ret = -EINVAL;
1012 goto fput_and_out;
1013 }
1014
John McCutchan8140a502005-12-12 00:37:14 -08001015 if (!(mask & IN_DONT_FOLLOW))
1016 flags |= LOOKUP_FOLLOW;
1017 if (mask & IN_ONLYDIR)
1018 flags |= LOOKUP_DIRECTORY;
1019
1020 ret = find_inode(path, &nd, flags);
Robert Loveb6807162005-07-25 15:07:13 -04001021 if (unlikely(ret))
Robert Love0eeca282005-07-12 17:06:03 -04001022 goto fput_and_out;
1023
Robert Loveb6807162005-07-25 15:07:13 -04001024 /* inode held in place by reference to nd; dev by fget on fd */
Robert Love0eeca282005-07-12 17:06:03 -04001025 inode = nd.dentry->d_inode;
Robert Loveb6807162005-07-25 15:07:13 -04001026 dev = filp->private_data;
Robert Love0eeca282005-07-12 17:06:03 -04001027
Ingo Molnard4f9af92006-03-23 03:00:30 -08001028 mutex_lock(&inode->inotify_mutex);
1029 mutex_lock(&dev->mutex);
Robert Love0eeca282005-07-12 17:06:03 -04001030
John McCutchan7ea60402005-09-06 15:18:02 -07001031 if (mask & IN_MASK_ADD)
1032 mask_add = 1;
1033
Robert Love0eeca282005-07-12 17:06:03 -04001034 /* don't let user-space set invalid bits: we don't want flags set */
Robert Loveb5173112006-02-07 12:58:45 -08001035 mask &= IN_ALL_EVENTS | IN_ONESHOT;
Robert Loveb6807162005-07-25 15:07:13 -04001036 if (unlikely(!mask)) {
Robert Love0eeca282005-07-12 17:06:03 -04001037 ret = -EINVAL;
1038 goto out;
1039 }
1040
1041 /*
1042 * Handle the case of re-adding a watch on an (inode,dev) pair that we
1043 * are already watching. We just update the mask and return its wd.
1044 */
1045 old = inode_find_dev(inode, dev);
1046 if (unlikely(old)) {
John McCutchan7ea60402005-09-06 15:18:02 -07001047 if (mask_add)
1048 old->mask |= mask;
1049 else
1050 old->mask = mask;
Robert Love0eeca282005-07-12 17:06:03 -04001051 ret = old->wd;
1052 goto out;
1053 }
1054
1055 watch = create_watch(dev, mask, inode);
1056 if (unlikely(IS_ERR(watch))) {
1057 ret = PTR_ERR(watch);
1058 goto out;
1059 }
1060
Nick Pigginc32ccd82006-03-25 03:07:09 -08001061 if (!inotify_inode_watched(inode))
1062 set_dentry_child_flags(inode, 1);
1063
Robert Love0eeca282005-07-12 17:06:03 -04001064 /* Add the watch to the device's and the inode's list */
1065 list_add(&watch->d_list, &dev->watches);
1066 list_add(&watch->i_list, &inode->inotify_watches);
1067 ret = watch->wd;
1068out:
Ingo Molnard4f9af92006-03-23 03:00:30 -08001069 mutex_unlock(&dev->mutex);
1070 mutex_unlock(&inode->inotify_mutex);
Robert Love5eb22cb2005-07-25 15:12:19 -04001071 path_release(&nd);
Robert Love0eeca282005-07-12 17:06:03 -04001072fput_and_out:
Robert Love33ea2f52005-07-25 15:08:37 -04001073 fput_light(filp, fput_needed);
Robert Love0eeca282005-07-12 17:06:03 -04001074 return ret;
1075}
1076
1077asmlinkage long sys_inotify_rm_watch(int fd, u32 wd)
1078{
1079 struct file *filp;
1080 struct inotify_device *dev;
Robert Love33ea2f52005-07-25 15:08:37 -04001081 int ret, fput_needed;
Robert Love0eeca282005-07-12 17:06:03 -04001082
Robert Love33ea2f52005-07-25 15:08:37 -04001083 filp = fget_light(fd, &fput_needed);
1084 if (unlikely(!filp))
Robert Love0eeca282005-07-12 17:06:03 -04001085 return -EBADF;
Robert Love783bc292005-07-25 15:10:08 -04001086
1087 /* verify that this is indeed an inotify instance */
1088 if (unlikely(filp->f_op != &inotify_fops)) {
1089 ret = -EINVAL;
1090 goto out;
1091 }
1092
Robert Love0eeca282005-07-12 17:06:03 -04001093 dev = filp->private_data;
Robert Love9a556e82005-07-13 13:49:23 -04001094 ret = inotify_ignore(dev, wd);
Robert Love9a556e82005-07-13 13:49:23 -04001095
Robert Love783bc292005-07-25 15:10:08 -04001096out:
1097 fput_light(filp, fput_needed);
Robert Love0eeca282005-07-12 17:06:03 -04001098 return ret;
1099}
1100
1101static struct super_block *
1102inotify_get_sb(struct file_system_type *fs_type, int flags,
1103 const char *dev_name, void *data)
1104{
1105 return get_sb_pseudo(fs_type, "inotify", NULL, 0xBAD1DEA);
1106}
1107
1108static struct file_system_type inotify_fs_type = {
1109 .name = "inotifyfs",
1110 .get_sb = inotify_get_sb,
1111 .kill_sb = kill_anon_super,
1112};
1113
1114/*
Robert Loveb6807162005-07-25 15:07:13 -04001115 * inotify_setup - Our initialization function. Note that we cannnot return
Robert Love0eeca282005-07-12 17:06:03 -04001116 * error because we have compiled-in VFS hooks. So an (unlikely) failure here
1117 * must result in panic().
1118 */
Robert Loveb6807162005-07-25 15:07:13 -04001119static int __init inotify_setup(void)
Robert Love0eeca282005-07-12 17:06:03 -04001120{
Robert Lovee5ca8442005-07-25 15:17:34 -04001121 int ret;
1122
1123 ret = register_filesystem(&inotify_fs_type);
1124 if (unlikely(ret))
1125 panic("inotify: register_filesystem returned %d!\n", ret);
1126
Robert Love0eeca282005-07-12 17:06:03 -04001127 inotify_mnt = kern_mount(&inotify_fs_type);
Andrew Morton89373de2005-07-26 14:08:38 -07001128 if (IS_ERR(inotify_mnt))
Robert Lovee5ca8442005-07-25 15:17:34 -04001129 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
Robert Love0eeca282005-07-12 17:06:03 -04001130
Robert Love1b2ccf02005-07-25 15:13:43 -04001131 inotify_max_queued_events = 16384;
1132 inotify_max_user_instances = 128;
Robert Love0eeca282005-07-12 17:06:03 -04001133 inotify_max_user_watches = 8192;
1134
1135 atomic_set(&inotify_cookie, 0);
1136
1137 watch_cachep = kmem_cache_create("inotify_watch_cache",
1138 sizeof(struct inotify_watch),
1139 0, SLAB_PANIC, NULL, NULL);
1140 event_cachep = kmem_cache_create("inotify_event_cache",
1141 sizeof(struct inotify_kernel_event),
1142 0, SLAB_PANIC, NULL, NULL);
1143
Robert Love0eeca282005-07-12 17:06:03 -04001144 return 0;
1145}
1146
Robert Loveb6807162005-07-25 15:07:13 -04001147module_init(inotify_setup);