aboutsummaryrefslogtreecommitdiffstats
path: root/fs/notify/inotify/inotify_user.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/notify/inotify/inotify_user.c')
-rw-r--r--fs/notify/inotify/inotify_user.c837
1 files changed, 399 insertions, 438 deletions
diff --git a/fs/notify/inotify/inotify_user.c b/fs/notify/inotify/inotify_user.c
index 1634319e2404..982a412ac5bc 100644
--- a/fs/notify/inotify/inotify_user.c
+++ b/fs/notify/inotify/inotify_user.c
@@ -8,6 +8,9 @@
8 * Copyright (C) 2005 John McCutchan 8 * Copyright (C) 2005 John McCutchan
9 * Copyright 2006 Hewlett-Packard Development Company, L.P. 9 * Copyright 2006 Hewlett-Packard Development Company, L.P.
10 * 10 *
11 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
12 * inotify was largely rewriten to make use of the fsnotify infrastructure
13 *
11 * This program is free software; you can redistribute it and/or modify it 14 * This program is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the 15 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation; either version 2, or (at your option) any 16 * Free Software Foundation; either version 2, or (at your option) any
@@ -19,94 +22,48 @@
19 * General Public License for more details. 22 * General Public License for more details.
20 */ 23 */
21 24
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/slab.h>
25#include <linux/fs.h>
26#include <linux/file.h> 25#include <linux/file.h>
27#include <linux/mount.h> 26#include <linux/fs.h> /* struct inode */
28#include <linux/namei.h> 27#include <linux/fsnotify_backend.h>
29#include <linux/poll.h> 28#include <linux/idr.h>
30#include <linux/init.h> 29#include <linux/init.h> /* module_init */
31#include <linux/list.h>
32#include <linux/inotify.h> 30#include <linux/inotify.h>
31#include <linux/kernel.h> /* roundup() */
32#include <linux/magic.h> /* superblock magic number */
33#include <linux/mount.h> /* mntget */
34#include <linux/namei.h> /* LOOKUP_FOLLOW */
35#include <linux/path.h> /* struct path */
36#include <linux/sched.h> /* struct user */
37#include <linux/slab.h> /* struct kmem_cache */
33#include <linux/syscalls.h> 38#include <linux/syscalls.h>
34#include <linux/magic.h> 39#include <linux/types.h>
40#include <linux/uaccess.h>
41#include <linux/poll.h>
42#include <linux/wait.h>
35 43
36#include <asm/ioctls.h> 44#include "inotify.h"
37 45
38static struct kmem_cache *watch_cachep __read_mostly; 46#include <asm/ioctls.h>
39static struct kmem_cache *event_cachep __read_mostly;
40 47
41static struct vfsmount *inotify_mnt __read_mostly; 48static struct vfsmount *inotify_mnt __read_mostly;
42 49
50/* this just sits here and wastes global memory. used to just pad userspace messages with zeros */
51static struct inotify_event nul_inotify_event;
52
43/* these are configurable via /proc/sys/fs/inotify/ */ 53/* these are configurable via /proc/sys/fs/inotify/ */
44static int inotify_max_user_instances __read_mostly; 54static int inotify_max_user_instances __read_mostly;
45static int inotify_max_user_watches __read_mostly;
46static int inotify_max_queued_events __read_mostly; 55static int inotify_max_queued_events __read_mostly;
56int inotify_max_user_watches __read_mostly;
47 57
48/* 58static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
49 * Lock ordering: 59struct kmem_cache *event_priv_cachep __read_mostly;
50 * 60static struct fsnotify_event *inotify_ignored_event;
51 * inotify_dev->up_mutex (ensures we don't re-add the same watch)
52 * inode->inotify_mutex (protects inode's watch list)
53 * inotify_handle->mutex (protects inotify_handle's watch list)
54 * inotify_dev->ev_mutex (protects device's event queue)
55 */
56 61
57/* 62/*
58 * Lifetimes of the main data structures: 63 * When inotify registers a new group it increments this and uses that
59 * 64 * value as an offset to set the fsnotify group "name" and priority.
60 * inotify_device: Lifetime is managed by reference count, from
61 * sys_inotify_init() until release. Additional references can bump the count
62 * via get_inotify_dev() and drop the count via put_inotify_dev().
63 *
64 * inotify_user_watch: Lifetime is from create_watch() to the receipt of an
65 * IN_IGNORED event from inotify, or when using IN_ONESHOT, to receipt of the
66 * first event, or to inotify_destroy().
67 */ 65 */
68 66static atomic_t inotify_grp_num;
69/*
70 * struct inotify_device - represents an inotify instance
71 *
72 * This structure is protected by the mutex 'mutex'.
73 */
74struct inotify_device {
75 wait_queue_head_t wq; /* wait queue for i/o */
76 struct mutex ev_mutex; /* protects event queue */
77 struct mutex up_mutex; /* synchronizes watch updates */
78 struct list_head events; /* list of queued events */
79 struct user_struct *user; /* user who opened this dev */
80 struct inotify_handle *ih; /* inotify handle */
81 struct fasync_struct *fa; /* async notification */
82 atomic_t count; /* reference count */
83 unsigned int queue_size; /* size of the queue (bytes) */
84 unsigned int event_count; /* number of pending events */
85 unsigned int max_events; /* maximum number of events */
86};
87
88/*
89 * struct inotify_kernel_event - An inotify event, originating from a watch and
90 * queued for user-space. A list of these is attached to each instance of the
91 * device. In read(), this list is walked and all events that can fit in the
92 * buffer are returned.
93 *
94 * Protected by dev->ev_mutex of the device in which we are queued.
95 */
96struct inotify_kernel_event {
97 struct inotify_event event; /* the user-space event */
98 struct list_head list; /* entry in inotify_device's list */
99 char *name; /* filename, if any */
100};
101
102/*
103 * struct inotify_user_watch - our version of an inotify_watch, we add
104 * a reference to the associated inotify_device.
105 */
106struct inotify_user_watch {
107 struct inotify_device *dev; /* associated device */
108 struct inotify_watch wdata; /* inotify watch data */
109};
110 67
111#ifdef CONFIG_SYSCTL 68#ifdef CONFIG_SYSCTL
112 69
@@ -149,280 +106,36 @@ ctl_table inotify_table[] = {
149}; 106};
150#endif /* CONFIG_SYSCTL */ 107#endif /* CONFIG_SYSCTL */
151 108
152static inline void get_inotify_dev(struct inotify_device *dev) 109static inline __u32 inotify_arg_to_mask(u32 arg)
153{
154 atomic_inc(&dev->count);
155}
156
157static inline void put_inotify_dev(struct inotify_device *dev)
158{
159 if (atomic_dec_and_test(&dev->count)) {
160 atomic_dec(&dev->user->inotify_devs);
161 free_uid(dev->user);
162 kfree(dev);
163 }
164}
165
166/*
167 * free_inotify_user_watch - cleans up the watch and its references
168 */
169static void free_inotify_user_watch(struct inotify_watch *w)
170{
171 struct inotify_user_watch *watch;
172 struct inotify_device *dev;
173
174 watch = container_of(w, struct inotify_user_watch, wdata);
175 dev = watch->dev;
176
177 atomic_dec(&dev->user->inotify_watches);
178 put_inotify_dev(dev);
179 kmem_cache_free(watch_cachep, watch);
180}
181
182/*
183 * kernel_event - create a new kernel event with the given parameters
184 *
185 * This function can sleep.
186 */
187static struct inotify_kernel_event * kernel_event(s32 wd, u32 mask, u32 cookie,
188 const char *name)
189{
190 struct inotify_kernel_event *kevent;
191
192 kevent = kmem_cache_alloc(event_cachep, GFP_NOFS);
193 if (unlikely(!kevent))
194 return NULL;
195
196 /* we hand this out to user-space, so zero it just in case */
197 memset(&kevent->event, 0, sizeof(struct inotify_event));
198
199 kevent->event.wd = wd;
200 kevent->event.mask = mask;
201 kevent->event.cookie = cookie;
202
203 INIT_LIST_HEAD(&kevent->list);
204
205 if (name) {
206 size_t len, rem, event_size = sizeof(struct inotify_event);
207
208 /*
209 * We need to pad the filename so as to properly align an
210 * array of inotify_event structures. Because the structure is
211 * small and the common case is a small filename, we just round
212 * up to the next multiple of the structure's sizeof. This is
213 * simple and safe for all architectures.
214 */
215 len = strlen(name) + 1;
216 rem = event_size - len;
217 if (len > event_size) {
218 rem = event_size - (len % event_size);
219 if (len % event_size == 0)
220 rem = 0;
221 }
222
223 kevent->name = kmalloc(len + rem, GFP_NOFS);
224 if (unlikely(!kevent->name)) {
225 kmem_cache_free(event_cachep, kevent);
226 return NULL;
227 }
228 memcpy(kevent->name, name, len);
229 if (rem)
230 memset(kevent->name + len, 0, rem);
231 kevent->event.len = len + rem;
232 } else {
233 kevent->event.len = 0;
234 kevent->name = NULL;
235 }
236
237 return kevent;
238}
239
240/*
241 * inotify_dev_get_event - return the next event in the given dev's queue
242 *
243 * Caller must hold dev->ev_mutex.
244 */
245static inline struct inotify_kernel_event *
246inotify_dev_get_event(struct inotify_device *dev)
247{
248 return list_entry(dev->events.next, struct inotify_kernel_event, list);
249}
250
251/*
252 * inotify_dev_get_last_event - return the last event in the given dev's queue
253 *
254 * Caller must hold dev->ev_mutex.
255 */
256static inline struct inotify_kernel_event *
257inotify_dev_get_last_event(struct inotify_device *dev)
258{ 110{
259 if (list_empty(&dev->events)) 111 __u32 mask;
260 return NULL;
261 return list_entry(dev->events.prev, struct inotify_kernel_event, list);
262}
263 112
264/* 113 /* everything should accept their own ignored and cares about children */
265 * inotify_dev_queue_event - event handler registered with core inotify, adds 114 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
266 * a new event to the given device
267 *
268 * Can sleep (calls kernel_event()).
269 */
270static void inotify_dev_queue_event(struct inotify_watch *w, u32 wd, u32 mask,
271 u32 cookie, const char *name,
272 struct inode *ignored)
273{
274 struct inotify_user_watch *watch;
275 struct inotify_device *dev;
276 struct inotify_kernel_event *kevent, *last;
277 115
278 watch = container_of(w, struct inotify_user_watch, wdata); 116 /* mask off the flags used to open the fd */
279 dev = watch->dev; 117 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
280 118
281 mutex_lock(&dev->ev_mutex); 119 return mask;
282
283 /* we can safely put the watch as we don't reference it while
284 * generating the event
285 */
286 if (mask & IN_IGNORED || w->mask & IN_ONESHOT)
287 put_inotify_watch(w); /* final put */
288
289 /* coalescing: drop this event if it is a dupe of the previous */
290 last = inotify_dev_get_last_event(dev);
291 if (last && last->event.mask == mask && last->event.wd == wd &&
292 last->event.cookie == cookie) {
293 const char *lastname = last->name;
294
295 if (!name && !lastname)
296 goto out;
297 if (name && lastname && !strcmp(lastname, name))
298 goto out;
299 }
300
301 /* the queue overflowed and we already sent the Q_OVERFLOW event */
302 if (unlikely(dev->event_count > dev->max_events))
303 goto out;
304
305 /* if the queue overflows, we need to notify user space */
306 if (unlikely(dev->event_count == dev->max_events))
307 kevent = kernel_event(-1, IN_Q_OVERFLOW, cookie, NULL);
308 else
309 kevent = kernel_event(wd, mask, cookie, name);
310
311 if (unlikely(!kevent))
312 goto out;
313
314 /* queue the event and wake up anyone waiting */
315 dev->event_count++;
316 dev->queue_size += sizeof(struct inotify_event) + kevent->event.len;
317 list_add_tail(&kevent->list, &dev->events);
318 wake_up_interruptible(&dev->wq);
319 kill_fasync(&dev->fa, SIGIO, POLL_IN);
320
321out:
322 mutex_unlock(&dev->ev_mutex);
323}
324
325/*
326 * remove_kevent - cleans up the given kevent
327 *
328 * Caller must hold dev->ev_mutex.
329 */
330static void remove_kevent(struct inotify_device *dev,
331 struct inotify_kernel_event *kevent)
332{
333 list_del(&kevent->list);
334
335 dev->event_count--;
336 dev->queue_size -= sizeof(struct inotify_event) + kevent->event.len;
337}
338
339/*
340 * free_kevent - frees the given kevent.
341 */
342static void free_kevent(struct inotify_kernel_event *kevent)
343{
344 kfree(kevent->name);
345 kmem_cache_free(event_cachep, kevent);
346}
347
348/*
349 * inotify_dev_event_dequeue - destroy an event on the given device
350 *
351 * Caller must hold dev->ev_mutex.
352 */
353static void inotify_dev_event_dequeue(struct inotify_device *dev)
354{
355 if (!list_empty(&dev->events)) {
356 struct inotify_kernel_event *kevent;
357 kevent = inotify_dev_get_event(dev);
358 remove_kevent(dev, kevent);
359 free_kevent(kevent);
360 }
361}
362
363/*
364 * find_inode - resolve a user-given path to a specific inode
365 */
366static int find_inode(const char __user *dirname, struct path *path,
367 unsigned flags)
368{
369 int error;
370
371 error = user_path_at(AT_FDCWD, dirname, flags, path);
372 if (error)
373 return error;
374 /* you can only watch an inode if you have read permissions on it */
375 error = inode_permission(path->dentry->d_inode, MAY_READ);
376 if (error)
377 path_put(path);
378 return error;
379} 120}
380 121
381/* 122static inline u32 inotify_mask_to_arg(__u32 mask)
382 * create_watch - creates a watch on the given device.
383 *
384 * Callers must hold dev->up_mutex.
385 */
386static int create_watch(struct inotify_device *dev, struct inode *inode,
387 u32 mask)
388{ 123{
389 struct inotify_user_watch *watch; 124 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
390 int ret; 125 IN_Q_OVERFLOW);
391
392 if (atomic_read(&dev->user->inotify_watches) >=
393 inotify_max_user_watches)
394 return -ENOSPC;
395
396 watch = kmem_cache_alloc(watch_cachep, GFP_KERNEL);
397 if (unlikely(!watch))
398 return -ENOMEM;
399
400 /* save a reference to device and bump the count to make it official */
401 get_inotify_dev(dev);
402 watch->dev = dev;
403
404 atomic_inc(&dev->user->inotify_watches);
405
406 inotify_init_watch(&watch->wdata);
407 ret = inotify_add_watch(dev->ih, &watch->wdata, inode, mask);
408 if (ret < 0)
409 free_inotify_user_watch(&watch->wdata);
410
411 return ret;
412} 126}
413 127
414/* Device Interface */ 128/* intofiy userspace file descriptor functions */
415
416static unsigned int inotify_poll(struct file *file, poll_table *wait) 129static unsigned int inotify_poll(struct file *file, poll_table *wait)
417{ 130{
418 struct inotify_device *dev = file->private_data; 131 struct fsnotify_group *group = file->private_data;
419 int ret = 0; 132 int ret = 0;
420 133
421 poll_wait(file, &dev->wq, wait); 134 poll_wait(file, &group->notification_waitq, wait);
422 mutex_lock(&dev->ev_mutex); 135 mutex_lock(&group->notification_mutex);
423 if (!list_empty(&dev->events)) 136 if (!fsnotify_notify_queue_is_empty(group))
424 ret = POLLIN | POLLRDNORM; 137 ret = POLLIN | POLLRDNORM;
425 mutex_unlock(&dev->ev_mutex); 138 mutex_unlock(&group->notification_mutex);
426 139
427 return ret; 140 return ret;
428} 141}
@@ -432,26 +145,29 @@ static unsigned int inotify_poll(struct file *file, poll_table *wait)
432 * enough to fit in "count". Return an error pointer if 145 * enough to fit in "count". Return an error pointer if
433 * not large enough. 146 * not large enough.
434 * 147 *
435 * Called with the device ev_mutex held. 148 * Called with the group->notification_mutex held.
436 */ 149 */
437static struct inotify_kernel_event *get_one_event(struct inotify_device *dev, 150static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
438 size_t count) 151 size_t count)
439{ 152{
440 size_t event_size = sizeof(struct inotify_event); 153 size_t event_size = sizeof(struct inotify_event);
441 struct inotify_kernel_event *kevent; 154 struct fsnotify_event *event;
442 155
443 if (list_empty(&dev->events)) 156 if (fsnotify_notify_queue_is_empty(group))
444 return NULL; 157 return NULL;
445 158
446 kevent = inotify_dev_get_event(dev); 159 event = fsnotify_peek_notify_event(group);
447 if (kevent->name) 160
448 event_size += kevent->event.len; 161 event_size += roundup(event->name_len, event_size);
449 162
450 if (event_size > count) 163 if (event_size > count)
451 return ERR_PTR(-EINVAL); 164 return ERR_PTR(-EINVAL);
452 165
453 remove_kevent(dev, kevent); 166 /* held the notification_mutex the whole time, so this is the
454 return kevent; 167 * same event we peeked above */
168 fsnotify_remove_notify_event(group);
169
170 return event;
455} 171}
456 172
457/* 173/*
@@ -460,51 +176,90 @@ static struct inotify_kernel_event *get_one_event(struct inotify_device *dev,
460 * We already checked that the event size is smaller than the 176 * We already checked that the event size is smaller than the
461 * buffer we had in "get_one_event()" above. 177 * buffer we had in "get_one_event()" above.
462 */ 178 */
463static ssize_t copy_event_to_user(struct inotify_kernel_event *kevent, 179static ssize_t copy_event_to_user(struct fsnotify_group *group,
180 struct fsnotify_event *event,
464 char __user *buf) 181 char __user *buf)
465{ 182{
183 struct inotify_event inotify_event;
184 struct fsnotify_event_private_data *fsn_priv;
185 struct inotify_event_private_data *priv;
466 size_t event_size = sizeof(struct inotify_event); 186 size_t event_size = sizeof(struct inotify_event);
187 size_t name_len;
188
189 /* we get the inotify watch descriptor from the event private data */
190 spin_lock(&event->lock);
191 fsn_priv = fsnotify_remove_priv_from_event(group, event);
192 spin_unlock(&event->lock);
193
194 if (!fsn_priv)
195 inotify_event.wd = -1;
196 else {
197 priv = container_of(fsn_priv, struct inotify_event_private_data,
198 fsnotify_event_priv_data);
199 inotify_event.wd = priv->wd;
200 inotify_free_event_priv(fsn_priv);
201 }
202
203 /* round up event->name_len so it is a multiple of event_size */
204 name_len = roundup(event->name_len, event_size);
205 inotify_event.len = name_len;
206
207 inotify_event.mask = inotify_mask_to_arg(event->mask);
208 inotify_event.cookie = event->sync_cookie;
467 209
468 if (copy_to_user(buf, &kevent->event, event_size)) 210 /* send the main event */
211 if (copy_to_user(buf, &inotify_event, event_size))
469 return -EFAULT; 212 return -EFAULT;
470 213
471 if (kevent->name) { 214 buf += event_size;
472 buf += event_size;
473 215
474 if (copy_to_user(buf, kevent->name, kevent->event.len)) 216 /*
217 * fsnotify only stores the pathname, so here we have to send the pathname
218 * and then pad that pathname out to a multiple of sizeof(inotify_event)
219 * with zeros. I get my zeros from the nul_inotify_event.
220 */
221 if (name_len) {
222 unsigned int len_to_zero = name_len - event->name_len;
223 /* copy the path name */
224 if (copy_to_user(buf, event->file_name, event->name_len))
475 return -EFAULT; 225 return -EFAULT;
226 buf += event->name_len;
476 227
477 event_size += kevent->event.len; 228 /* fill userspace with 0's from nul_inotify_event */
229 if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
230 return -EFAULT;
231 buf += len_to_zero;
232 event_size += name_len;
478 } 233 }
234
479 return event_size; 235 return event_size;
480} 236}
481 237
482static ssize_t inotify_read(struct file *file, char __user *buf, 238static ssize_t inotify_read(struct file *file, char __user *buf,
483 size_t count, loff_t *pos) 239 size_t count, loff_t *pos)
484{ 240{
485 struct inotify_device *dev; 241 struct fsnotify_group *group;
242 struct fsnotify_event *kevent;
486 char __user *start; 243 char __user *start;
487 int ret; 244 int ret;
488 DEFINE_WAIT(wait); 245 DEFINE_WAIT(wait);
489 246
490 start = buf; 247 start = buf;
491 dev = file->private_data; 248 group = file->private_data;
492 249
493 while (1) { 250 while (1) {
494 struct inotify_kernel_event *kevent; 251 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
495 252
496 prepare_to_wait(&dev->wq, &wait, TASK_INTERRUPTIBLE); 253 mutex_lock(&group->notification_mutex);
497 254 kevent = get_one_event(group, count);
498 mutex_lock(&dev->ev_mutex); 255 mutex_unlock(&group->notification_mutex);
499 kevent = get_one_event(dev, count);
500 mutex_unlock(&dev->ev_mutex);
501 256
502 if (kevent) { 257 if (kevent) {
503 ret = PTR_ERR(kevent); 258 ret = PTR_ERR(kevent);
504 if (IS_ERR(kevent)) 259 if (IS_ERR(kevent))
505 break; 260 break;
506 ret = copy_event_to_user(kevent, buf); 261 ret = copy_event_to_user(group, kevent, buf);
507 free_kevent(kevent); 262 fsnotify_put_event(kevent);
508 if (ret < 0) 263 if (ret < 0)
509 break; 264 break;
510 buf += ret; 265 buf += ret;
@@ -525,7 +280,7 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
525 schedule(); 280 schedule();
526 } 281 }
527 282
528 finish_wait(&dev->wq, &wait); 283 finish_wait(&group->notification_waitq, &wait);
529 if (start != buf && ret != -EFAULT) 284 if (start != buf && ret != -EFAULT)
530 ret = buf - start; 285 ret = buf - start;
531 return ret; 286 return ret;
@@ -533,25 +288,19 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
533 288
534static int inotify_fasync(int fd, struct file *file, int on) 289static int inotify_fasync(int fd, struct file *file, int on)
535{ 290{
536 struct inotify_device *dev = file->private_data; 291 struct fsnotify_group *group = file->private_data;
537 292
538 return fasync_helper(fd, file, on, &dev->fa) >= 0 ? 0 : -EIO; 293 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
539} 294}
540 295
541static int inotify_release(struct inode *ignored, struct file *file) 296static int inotify_release(struct inode *ignored, struct file *file)
542{ 297{
543 struct inotify_device *dev = file->private_data; 298 struct fsnotify_group *group = file->private_data;
544
545 inotify_destroy(dev->ih);
546 299
547 /* destroy all of the events on this device */ 300 fsnotify_clear_marks_by_group(group);
548 mutex_lock(&dev->ev_mutex);
549 while (!list_empty(&dev->events))
550 inotify_dev_event_dequeue(dev);
551 mutex_unlock(&dev->ev_mutex);
552 301
553 /* free this device: the put matching the get in inotify_init() */ 302 /* free this group, matching get was inotify_init->fsnotify_obtain_group */
554 put_inotify_dev(dev); 303 fsnotify_put_group(group);
555 304
556 return 0; 305 return 0;
557} 306}
@@ -559,16 +308,27 @@ static int inotify_release(struct inode *ignored, struct file *file)
559static long inotify_ioctl(struct file *file, unsigned int cmd, 308static long inotify_ioctl(struct file *file, unsigned int cmd,
560 unsigned long arg) 309 unsigned long arg)
561{ 310{
562 struct inotify_device *dev; 311 struct fsnotify_group *group;
312 struct fsnotify_event_holder *holder;
313 struct fsnotify_event *event;
563 void __user *p; 314 void __user *p;
564 int ret = -ENOTTY; 315 int ret = -ENOTTY;
316 size_t send_len = 0;
565 317
566 dev = file->private_data; 318 group = file->private_data;
567 p = (void __user *) arg; 319 p = (void __user *) arg;
568 320
569 switch (cmd) { 321 switch (cmd) {
570 case FIONREAD: 322 case FIONREAD:
571 ret = put_user(dev->queue_size, (int __user *) p); 323 mutex_lock(&group->notification_mutex);
324 list_for_each_entry(holder, &group->notification_list, event_list) {
325 event = holder->event;
326 send_len += sizeof(struct inotify_event);
327 send_len += roundup(event->name_len,
328 sizeof(struct inotify_event));
329 }
330 mutex_unlock(&group->notification_mutex);
331 ret = put_user(send_len, (int __user *) p);
572 break; 332 break;
573 } 333 }
574 334
@@ -576,23 +336,233 @@ static long inotify_ioctl(struct file *file, unsigned int cmd,
576} 336}
577 337
578static const struct file_operations inotify_fops = { 338static const struct file_operations inotify_fops = {
579 .poll = inotify_poll, 339 .poll = inotify_poll,
580 .read = inotify_read, 340 .read = inotify_read,
581 .fasync = inotify_fasync, 341 .fasync = inotify_fasync,
582 .release = inotify_release, 342 .release = inotify_release,
583 .unlocked_ioctl = inotify_ioctl, 343 .unlocked_ioctl = inotify_ioctl,
584 .compat_ioctl = inotify_ioctl, 344 .compat_ioctl = inotify_ioctl,
585}; 345};
586 346
587static const struct inotify_operations inotify_user_ops = {
588 .handle_event = inotify_dev_queue_event,
589 .destroy_watch = free_inotify_user_watch,
590};
591 347
348/*
349 * find_inode - resolve a user-given path to a specific inode
350 */
351static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
352{
353 int error;
354
355 error = user_path_at(AT_FDCWD, dirname, flags, path);
356 if (error)
357 return error;
358 /* you can only watch an inode if you have read permissions on it */
359 error = inode_permission(path->dentry->d_inode, MAY_READ);
360 if (error)
361 path_put(path);
362 return error;
363}
364
365/*
366 * When, for whatever reason, inotify is done with a mark (or what used to be a
367 * watch) we need to remove that watch from the idr and we need to send IN_IGNORED
368 * for the given wd.
369 *
370 * There is a bit of recursion here. The loop looks like:
371 * inotify_destroy_mark_entry -> fsnotify_destroy_mark_by_entry ->
372 * inotify_freeing_mark -> inotify_destory_mark_entry -> restart
373 * But the loop is broken in 2 places. fsnotify_destroy_mark_by_entry sets
374 * entry->group = NULL before the call to inotify_freeing_mark, so the if (egroup)
375 * test below will not call back to fsnotify again. But even if that test wasn't
376 * there this would still be safe since fsnotify_destroy_mark_by_entry() is
377 * safe from recursion.
378 */
379void inotify_destroy_mark_entry(struct fsnotify_mark_entry *entry, struct fsnotify_group *group)
380{
381 struct inotify_inode_mark_entry *ientry;
382 struct inotify_event_private_data *event_priv;
383 struct fsnotify_event_private_data *fsn_event_priv;
384 struct fsnotify_group *egroup;
385 struct idr *idr;
386
387 spin_lock(&entry->lock);
388 egroup = entry->group;
389
390 /* if egroup we aren't really done and something might still send events
391 * for this inode, on the callback we'll send the IN_IGNORED */
392 if (egroup) {
393 spin_unlock(&entry->lock);
394 fsnotify_destroy_mark_by_entry(entry);
395 return;
396 }
397 spin_unlock(&entry->lock);
398
399 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
400
401 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_KERNEL);
402 if (unlikely(!event_priv))
403 goto skip_send_ignore;
404
405 fsn_event_priv = &event_priv->fsnotify_event_priv_data;
406
407 fsn_event_priv->group = group;
408 event_priv->wd = ientry->wd;
409
410 fsnotify_add_notify_event(group, inotify_ignored_event, fsn_event_priv);
411
412 /* did the private data get added? */
413 if (list_empty(&fsn_event_priv->event_list))
414 inotify_free_event_priv(fsn_event_priv);
415
416skip_send_ignore:
417
418 /* remove this entry from the idr */
419 spin_lock(&group->inotify_data.idr_lock);
420 idr = &group->inotify_data.idr;
421 idr_remove(idr, ientry->wd);
422 spin_unlock(&group->inotify_data.idr_lock);
423
424 /* removed from idr, drop that reference */
425 fsnotify_put_mark(entry);
426}
427
428/* ding dong the mark is dead */
429static void inotify_free_mark(struct fsnotify_mark_entry *entry)
430{
431 struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
432
433 kmem_cache_free(inotify_inode_mark_cachep, ientry);
434}
435
436static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
437{
438 struct fsnotify_mark_entry *entry = NULL;
439 struct inotify_inode_mark_entry *ientry;
440 int ret = 0;
441 int add = (arg & IN_MASK_ADD);
442 __u32 mask;
443 __u32 old_mask, new_mask;
444
445 /* don't allow invalid bits: we don't want flags set */
446 mask = inotify_arg_to_mask(arg);
447 if (unlikely(!mask))
448 return -EINVAL;
449
450 ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
451 if (unlikely(!ientry))
452 return -ENOMEM;
453 /* we set the mask at the end after attaching it */
454 fsnotify_init_mark(&ientry->fsn_entry, inotify_free_mark);
455 ientry->wd = 0;
456
457find_entry:
458 spin_lock(&inode->i_lock);
459 entry = fsnotify_find_mark_entry(group, inode);
460 spin_unlock(&inode->i_lock);
461 if (entry) {
462 kmem_cache_free(inotify_inode_mark_cachep, ientry);
463 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
464 } else {
465 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) {
466 ret = -ENOSPC;
467 goto out_err;
468 }
469
470 ret = fsnotify_add_mark(&ientry->fsn_entry, group, inode);
471 if (ret == -EEXIST)
472 goto find_entry;
473 else if (ret)
474 goto out_err;
475
476 entry = &ientry->fsn_entry;
477retry:
478 ret = -ENOMEM;
479 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
480 goto out_err;
481
482 spin_lock(&group->inotify_data.idr_lock);
483 /* if entry is added to the idr we keep the reference obtained
484 * through fsnotify_mark_add. remember to drop this reference
485 * when entry is removed from idr */
486 ret = idr_get_new_above(&group->inotify_data.idr, entry,
487 ++group->inotify_data.last_wd,
488 &ientry->wd);
489 spin_unlock(&group->inotify_data.idr_lock);
490 if (ret) {
491 if (ret == -EAGAIN)
492 goto retry;
493 goto out_err;
494 }
495 atomic_inc(&group->inotify_data.user->inotify_watches);
496 }
497
498 spin_lock(&entry->lock);
499
500 old_mask = entry->mask;
501 if (add) {
502 entry->mask |= mask;
503 new_mask = entry->mask;
504 } else {
505 entry->mask = mask;
506 new_mask = entry->mask;
507 }
508
509 spin_unlock(&entry->lock);
510
511 if (old_mask != new_mask) {
512 /* more bits in old than in new? */
513 int dropped = (old_mask & ~new_mask);
514 /* more bits in this entry than the inode's mask? */
515 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
516 /* more bits in this entry than the group? */
517 int do_group = (new_mask & ~group->mask);
518
519 /* update the inode with this new entry */
520 if (dropped || do_inode)
521 fsnotify_recalc_inode_mask(inode);
522
523 /* update the group mask with the new mask */
524 if (dropped || do_group)
525 fsnotify_recalc_group_mask(group);
526 }
527
528 return ientry->wd;
529
530out_err:
531 /* see this isn't supposed to happen, just kill the watch */
532 if (entry) {
533 fsnotify_destroy_mark_by_entry(entry);
534 fsnotify_put_mark(entry);
535 }
536 return ret;
537}
538
539static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
540{
541 struct fsnotify_group *group;
542 unsigned int grp_num;
543
544 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
545 grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
546 group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
547 if (IS_ERR(group))
548 return group;
549
550 group->max_events = max_events;
551
552 spin_lock_init(&group->inotify_data.idr_lock);
553 idr_init(&group->inotify_data.idr);
554 group->inotify_data.last_wd = 0;
555 group->inotify_data.user = user;
556 group->inotify_data.fa = NULL;
557
558 return group;
559}
560
561
562/* inotify syscalls */
592SYSCALL_DEFINE1(inotify_init1, int, flags) 563SYSCALL_DEFINE1(inotify_init1, int, flags)
593{ 564{
594 struct inotify_device *dev; 565 struct fsnotify_group *group;
595 struct inotify_handle *ih;
596 struct user_struct *user; 566 struct user_struct *user;
597 struct file *filp; 567 struct file *filp;
598 int fd, ret; 568 int fd, ret;
@@ -621,45 +591,27 @@ SYSCALL_DEFINE1(inotify_init1, int, flags)
621 goto out_free_uid; 591 goto out_free_uid;
622 } 592 }
623 593
624 dev = kmalloc(sizeof(struct inotify_device), GFP_KERNEL); 594 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
625 if (unlikely(!dev)) { 595 group = inotify_new_group(user, inotify_max_queued_events);
626 ret = -ENOMEM; 596 if (IS_ERR(group)) {
597 ret = PTR_ERR(group);
627 goto out_free_uid; 598 goto out_free_uid;
628 } 599 }
629 600
630 ih = inotify_init(&inotify_user_ops);
631 if (IS_ERR(ih)) {
632 ret = PTR_ERR(ih);
633 goto out_free_dev;
634 }
635 dev->ih = ih;
636 dev->fa = NULL;
637
638 filp->f_op = &inotify_fops; 601 filp->f_op = &inotify_fops;
639 filp->f_path.mnt = mntget(inotify_mnt); 602 filp->f_path.mnt = mntget(inotify_mnt);
640 filp->f_path.dentry = dget(inotify_mnt->mnt_root); 603 filp->f_path.dentry = dget(inotify_mnt->mnt_root);
641 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping; 604 filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
642 filp->f_mode = FMODE_READ; 605 filp->f_mode = FMODE_READ;
643 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK); 606 filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
644 filp->private_data = dev; 607 filp->private_data = group;
645 608
646 INIT_LIST_HEAD(&dev->events);
647 init_waitqueue_head(&dev->wq);
648 mutex_init(&dev->ev_mutex);
649 mutex_init(&dev->up_mutex);
650 dev->event_count = 0;
651 dev->queue_size = 0;
652 dev->max_events = inotify_max_queued_events;
653 dev->user = user;
654 atomic_set(&dev->count, 0);
655
656 get_inotify_dev(dev);
657 atomic_inc(&user->inotify_devs); 609 atomic_inc(&user->inotify_devs);
610
658 fd_install(fd, filp); 611 fd_install(fd, filp);
659 612
660 return fd; 613 return fd;
661out_free_dev: 614
662 kfree(dev);
663out_free_uid: 615out_free_uid:
664 free_uid(user); 616 free_uid(user);
665 put_filp(filp); 617 put_filp(filp);
@@ -676,8 +628,8 @@ SYSCALL_DEFINE0(inotify_init)
676SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, 628SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
677 u32, mask) 629 u32, mask)
678{ 630{
631 struct fsnotify_group *group;
679 struct inode *inode; 632 struct inode *inode;
680 struct inotify_device *dev;
681 struct path path; 633 struct path path;
682 struct file *filp; 634 struct file *filp;
683 int ret, fput_needed; 635 int ret, fput_needed;
@@ -698,20 +650,20 @@ SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
698 if (mask & IN_ONLYDIR) 650 if (mask & IN_ONLYDIR)
699 flags |= LOOKUP_DIRECTORY; 651 flags |= LOOKUP_DIRECTORY;
700 652
701 ret = find_inode(pathname, &path, flags); 653 ret = inotify_find_inode(pathname, &path, flags);
702 if (unlikely(ret)) 654 if (ret)
703 goto fput_and_out; 655 goto fput_and_out;
704 656
705 /* inode held in place by reference to path; dev by fget on fd */ 657 /* inode held in place by reference to path; group by fget on fd */
706 inode = path.dentry->d_inode; 658 inode = path.dentry->d_inode;
707 dev = filp->private_data; 659 group = filp->private_data;
708 660
709 mutex_lock(&dev->up_mutex); 661 /* create/update an inode mark */
710 ret = inotify_find_update_watch(dev->ih, inode, mask); 662 ret = inotify_update_watch(group, inode, mask);
711 if (ret == -ENOENT) 663 if (unlikely(ret))
712 ret = create_watch(dev, inode, mask); 664 goto path_put_and_out;
713 mutex_unlock(&dev->up_mutex);
714 665
666path_put_and_out:
715 path_put(&path); 667 path_put(&path);
716fput_and_out: 668fput_and_out:
717 fput_light(filp, fput_needed); 669 fput_light(filp, fput_needed);
@@ -720,9 +672,10 @@ fput_and_out:
720 672
721SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) 673SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
722{ 674{
675 struct fsnotify_group *group;
676 struct fsnotify_mark_entry *entry;
723 struct file *filp; 677 struct file *filp;
724 struct inotify_device *dev; 678 int ret = 0, fput_needed;
725 int ret, fput_needed;
726 679
727 filp = fget_light(fd, &fput_needed); 680 filp = fget_light(fd, &fput_needed);
728 if (unlikely(!filp)) 681 if (unlikely(!filp))
@@ -734,10 +687,20 @@ SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
734 goto out; 687 goto out;
735 } 688 }
736 689
737 dev = filp->private_data; 690 group = filp->private_data;
738 691
739 /* we free our watch data when we get IN_IGNORED */ 692 spin_lock(&group->inotify_data.idr_lock);
740 ret = inotify_rm_wd(dev->ih, wd); 693 entry = idr_find(&group->inotify_data.idr, wd);
694 if (unlikely(!entry)) {
695 spin_unlock(&group->inotify_data.idr_lock);
696 ret = -EINVAL;
697 goto out;
698 }
699 fsnotify_get_mark(entry);
700 spin_unlock(&group->inotify_data.idr_lock);
701
702 inotify_destroy_mark_entry(entry, group);
703 fsnotify_put_mark(entry);
741 704
742out: 705out:
743 fput_light(filp, fput_needed); 706 fput_light(filp, fput_needed);
@@ -753,9 +716,9 @@ inotify_get_sb(struct file_system_type *fs_type, int flags,
753} 716}
754 717
755static struct file_system_type inotify_fs_type = { 718static struct file_system_type inotify_fs_type = {
756 .name = "inotifyfs", 719 .name = "inotifyfs",
757 .get_sb = inotify_get_sb, 720 .get_sb = inotify_get_sb,
758 .kill_sb = kill_anon_super, 721 .kill_sb = kill_anon_super,
759}; 722};
760 723
761/* 724/*
@@ -775,18 +738,16 @@ static int __init inotify_user_setup(void)
775 if (IS_ERR(inotify_mnt)) 738 if (IS_ERR(inotify_mnt))
776 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt)); 739 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
777 740
741 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
742 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
743 inotify_ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, FSNOTIFY_EVENT_NONE, NULL, 0);
744 if (!inotify_ignored_event)
745 panic("unable to allocate the inotify ignored event\n");
746
778 inotify_max_queued_events = 16384; 747 inotify_max_queued_events = 16384;
779 inotify_max_user_instances = 128; 748 inotify_max_user_instances = 128;
780 inotify_max_user_watches = 8192; 749 inotify_max_user_watches = 8192;
781 750
782 watch_cachep = kmem_cache_create("inotify_watch_cache",
783 sizeof(struct inotify_user_watch),
784 0, SLAB_PANIC, NULL);
785 event_cachep = kmem_cache_create("inotify_event_cache",
786 sizeof(struct inotify_kernel_event),
787 0, SLAB_PANIC, NULL);
788
789 return 0; 751 return 0;
790} 752}
791
792module_init(inotify_user_setup); 753module_init(inotify_user_setup);