fs/notify/notification.c

   1 /*
   2  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
   3  *
   4  *  This program is free software; you can redistribute it and/or modify
   5  *  it under the terms of the GNU General Public License as published by
   6  *  the Free Software Foundation; either version 2, or (at your option)
   7  *  any later version.
   8  *
   9  *  This program is distributed in the hope that it will be useful,
  10  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  *  GNU General Public License for more details.
  13  *
  14  *  You should have received a copy of the GNU General Public License
  15  *  along with this program; see the file COPYING.  If not, write to
  16  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  17  */
  18
  19 /*
  20  * Basic idea behind the notification queue: An fsnotify group (like inotify)
  21  * sends the userspace notification about events asyncronously some time after
  22  * the event happened.  When inotify gets an event it will need to add that
  23  * event to the group notify queue.  Since a single event might need to be on
  24  * multiple group's notification queues we can't add the event directly to each
  25  * queue and instead add a small "event_holder" to each queue.  This event_holder
  26  * has a pointer back to the original event.  Since the majority of events are
  27  * going to end up on one, and only one, notification queue we embed one
  28  * event_holder into each event.  This means we have a single allocation instead
  29  * of always needing two.  If the embedded event_holder is already in use by
  30  * another group a new event_holder (from fsnotify_event_holder_cachep) will be
  31  * allocated and used.
  32  */
  33
  34 #include <linux/fs.h>
  35 #include <linux/init.h>
  36 #include <linux/kernel.h>
  37 #include <linux/list.h>
  38 #include <linux/module.h>
  39 #include <linux/mount.h>
  40 #include <linux/mutex.h>
  41 #include <linux/namei.h>
  42 #include <linux/path.h>
  43 #include <linux/slab.h>
  44 #include <linux/spinlock.h>
  45
  46 #include <asm/atomic.h>
  47
  48 #include <linux/fsnotify_backend.h>
  49 #include "fsnotify.h"
  50
  51 static struct kmem_cache *fsnotify_event_cachep;
  52 static struct kmem_cache *fsnotify_event_holder_cachep;
  53 /*
  54  * This is a magic event we send when the q is too full.  Since it doesn't
  55  * hold real event information we just keep one system wide and use it any time
  56  * it is needed.  It's refcnt is set 1 at kernel init time and will never
  57  * get set to 0 so it will never get 'freed'
  58  */
  59 static struct fsnotify_event *q_overflow_event;
  60 static atomic_t fsnotify_sync_cookie = ATOMIC_INIT(0);
  61
  62 /**
  63  * fsnotify_get_cookie - return a unique cookie for use in synchronizing events.
  64  * Called from fsnotify_move, which is inlined into filesystem modules.
  65  */
  66 u32 fsnotify_get_cookie(void)
  67 {
  68         return atomic_inc_return(&fsnotify_sync_cookie);
  69 }
  70 EXPORT_SYMBOL_GPL(fsnotify_get_cookie);
  71
  72 /* return true if the notify queue is empty, false otherwise */
  73 bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
  74 {
  75         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  76         return list_empty(&group->notification_list) ? true : false;
  77 }
  78
  79 void fsnotify_get_event(struct fsnotify_event *event)
  80 {
  81         atomic_inc(&event->refcnt);
  82 }
  83
  84 void fsnotify_put_event(struct fsnotify_event *event)
  85 {
  86         if (!event)
  87                 return;
  88
  89         if (atomic_dec_and_test(&event->refcnt)) {
  90                 if (event->data_type == FSNOTIFY_EVENT_PATH)
  91                         path_put(&event->path);
  92
  93                 BUG_ON(!list_empty(&event->private_data_list));
  94
  95                 kfree(event->file_name);
  96                 kmem_cache_free(fsnotify_event_cachep, event);
  97         }
  98 }
  99
 100 struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
 101 {
 102         return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
 103 }
 104
 105 void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
 106 {
 107         if (holder)
 108                 kmem_cache_free(fsnotify_event_holder_cachep, holder);
 109 }
 110
 111 /*
 112  * Find the private data that the group previously attached to this event when
 113  * the group added the event to the notification queue (fsnotify_add_notify_event)
 114  */
 115 struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event)
 116 {
 117         struct fsnotify_event_private_data *lpriv;
 118         struct fsnotify_event_private_data *priv = NULL;
 119
 120         assert_spin_locked(&event->lock);
 121
 122         list_for_each_entry(lpriv, &event->private_data_list, event_list) {
 123                 if (lpriv->group == group) {
 124                         priv = lpriv;
 125                         list_del(&priv->event_list);
 126                         break;
 127                 }
 128         }
 129         return priv;
 130 }
 131
 132 /*
 133  * Add an event to the group notification queue.  The group can later pull this
 134  * event off the queue to deal with.  If the event is successfully added to the
 135  * group's notification queue, a reference is taken on event.
 136  */
 137 int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
 138                               struct fsnotify_event_private_data *priv,
 139                               int (*merge)(struct list_head *, struct fsnotify_event *))
 140 {
 141         struct fsnotify_event_holder *holder = NULL;
 142         struct list_head *list = &group->notification_list;
 143         int rc = 0;
 144
 145         /*
 146          * There is one fsnotify_event_holder embedded inside each fsnotify_event.
 147          * Check if we expect to be able to use that holder.  If not alloc a new
 148          * holder.
 149          * For the overflow event it's possible that something will use the in
 150          * event holder before we get the lock so we may need to jump back and
 151          * alloc a new holder, this can't happen for most events...
 152          */
 153         if (!list_empty(&event->holder.event_list)) {
 154 alloc_holder:
 155                 holder = fsnotify_alloc_event_holder();
 156                 if (!holder)
 157                         return -ENOMEM;
 158         }
 159
 160         mutex_lock(&group->notification_mutex);
 161
 162         if (group->q_len >= group->max_events) {
 163                 event = q_overflow_event;
 164                 rc = -EOVERFLOW;
 165                 /* sorry, no private data on the overflow event */
 166                 priv = NULL;
 167         }
 168
 169         if (!list_empty(list) && merge) {
 170                 int ret;
 171
 172                 ret = merge(list, event);
 173                 if (ret) {
 174                         mutex_unlock(&group->notification_mutex);
 175                         if (holder != &event->holder)
 176                                 fsnotify_destroy_event_holder(holder);
 177                         return ret;
 178                 }
 179         }
 180
 181         spin_lock(&event->lock);
 182
 183         if (list_empty(&event->holder.event_list)) {
 184                 if (unlikely(holder))
 185                         fsnotify_destroy_event_holder(holder);
 186                 holder = &event->holder;
 187         } else if (unlikely(!holder)) {
 188                 /* between the time we checked above and got the lock the in
 189                  * event holder was used, go back and get a new one */
 190                 spin_unlock(&event->lock);
 191                 mutex_unlock(&group->notification_mutex);
 192                 goto alloc_holder;
 193         }
 194
 195         group->q_len++;
 196         holder->event = event;
 197
 198         fsnotify_get_event(event);
 199         list_add_tail(&holder->event_list, list);
 200         if (priv)
 201                 list_add_tail(&priv->event_list, &event->private_data_list);
 202         spin_unlock(&event->lock);
 203         mutex_unlock(&group->notification_mutex);
 204
 205         wake_up(&group->notification_waitq);
 206         return rc;
 207 }
 208
 209 /*
 210  * Remove and return the first event from the notification list.  There is a
 211  * reference held on this event since it was on the list.  It is the responsibility
 212  * of the caller to drop this reference.
 213  */
 214 struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
 215 {
 216         struct fsnotify_event *event;
 217         struct fsnotify_event_holder *holder;
 218
 219         BUG_ON(!mutex_is_locked(&group->notification_mutex));
 220
 221         holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
 222
 223         event = holder->event;
 224
 225         spin_lock(&event->lock);
 226         holder->event = NULL;
 227         list_del_init(&holder->event_list);
 228         spin_unlock(&event->lock);
 229
 230         /* event == holder means we are referenced through the in event holder */
 231         if (holder != &event->holder)
 232                 fsnotify_destroy_event_holder(holder);
 233
 234         group->q_len--;
 235
 236         return event;
 237 }
 238
 239 /*
 240  * This will not remove the event, that must be done with fsnotify_remove_notify_event()
 241  */
 242 struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
 243 {
 244         struct fsnotify_event *event;
 245         struct fsnotify_event_holder *holder;
 246
 247         BUG_ON(!mutex_is_locked(&group->notification_mutex));
 248
 249         holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
 250         event = holder->event;
 251
 252         return event;
 253 }
 254
 255 /*
 256  * Called when a group is being torn down to clean up any outstanding
 257  * event notifications.
 258  */
 259 void fsnotify_flush_notify(struct fsnotify_group *group)
 260 {
 261         struct fsnotify_event *event;
 262         struct fsnotify_event_private_data *priv;
 263
 264         mutex_lock(&group->notification_mutex);
 265         while (!fsnotify_notify_queue_is_empty(group)) {
 266                 event = fsnotify_remove_notify_event(group);
 267                 /* if they don't implement free_event_priv they better not have attached any */
 268                 if (group->ops->free_event_priv) {
 269                         spin_lock(&event->lock);
 270                         priv = fsnotify_remove_priv_from_event(group, event);
 271                         spin_unlock(&event->lock);
 272                         if (priv)
 273                                 group->ops->free_event_priv(priv);
 274                 }
 275                 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
 276         }
 277         mutex_unlock(&group->notification_mutex);
 278 }
 279
 280 static void initialize_event(struct fsnotify_event *event)
 281 {
 282         INIT_LIST_HEAD(&event->holder.event_list);
 283         atomic_set(&event->refcnt, 1);
 284
 285         spin_lock_init(&event->lock);
 286
 287         event->data_type = FSNOTIFY_EVENT_NONE;
 288
 289         INIT_LIST_HEAD(&event->private_data_list);
 290 }
 291
 292 /*
 293  * fsnotify_create_event - Allocate a new event which will be sent to each
 294  * group's handle_event function if the group was interested in this
 295  * particular event.
 296  *
 297  * @to_tell the inode which is supposed to receive the event (sometimes a
 298  *      parent of the inode to which the event happened.
 299  * @mask what actually happened.
 300  * @data pointer to the object which was actually affected
 301  * @data_type flag indication if the data is a file, path, inode, nothing...
 302  * @name the filename, if available
 303  */
 304 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
 305                                              int data_type, const char *name, u32 cookie,
 306                                              gfp_t gfp)
 307 {
 308         struct fsnotify_event *event;
 309
 310         event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
 311         if (!event)
 312                 return NULL;
 313
 314         initialize_event(event);
 315
 316         if (name) {
 317                 event->file_name = kstrdup(name, gfp);
 318                 if (!event->file_name) {
 319                         kmem_cache_free(fsnotify_event_cachep, event);
 320                         return NULL;
 321                 }
 322                 event->name_len = strlen(event->file_name);
 323         }
 324
 325         event->sync_cookie = cookie;
 326         event->to_tell = to_tell;
 327
 328         switch (data_type) {
 329         case FSNOTIFY_EVENT_FILE: {
 330                 struct file *file = data;
 331                 struct path *path = &file->f_path;
 332                 event->path.dentry = path->dentry;
 333                 event->path.mnt = path->mnt;
 334                 path_get(&event->path);
 335                 event->data_type = FSNOTIFY_EVENT_PATH;
 336                 break;
 337         }
 338         case FSNOTIFY_EVENT_PATH: {
 339                 struct path *path = data;
 340                 event->path.dentry = path->dentry;
 341                 event->path.mnt = path->mnt;
 342                 path_get(&event->path);
 343                 event->data_type = FSNOTIFY_EVENT_PATH;
 344                 break;
 345         }
 346         case FSNOTIFY_EVENT_INODE:
 347                 event->inode = data;
 348                 event->data_type = FSNOTIFY_EVENT_INODE;
 349                 break;
 350         case FSNOTIFY_EVENT_NONE:
 351                 event->inode = NULL;
 352                 event->path.dentry = NULL;
 353                 event->path.mnt = NULL;
 354                 break;
 355         default:
 356                 BUG();
 357         }
 358
 359         event->mask = mask;
 360
 361         return event;
 362 }
 363
 364 __init int fsnotify_notification_init(void)
 365 {
 366         fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
 367         fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
 368
 369         q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
 370                                                  FSNOTIFY_EVENT_NONE, NULL, 0,
 371                                                  GFP_KERNEL);
 372         if (!q_overflow_event)
 373                 panic("unable to allocate fsnotify q_overflow_event\n");
 374
 375         return 0;
 376 }
 377 subsys_initcall(fsnotify_notification_init);
 378