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 <linux/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                 pr_debug("%s: event=%p\n", __func__, event);
  91
  92                 if (event->data_type == FSNOTIFY_EVENT_PATH)
  93                         path_put(&event->path);
  94
  95                 BUG_ON(!list_empty(&event->private_data_list));
  96
  97                 kfree(event->file_name);
  98                 put_pid(event->tgid);
  99                 kmem_cache_free(fsnotify_event_cachep, event);
 100         }
 101 }
 102
 103 struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
 104 {
 105         return kmem_cache_alloc(fsnotify_event_holder_cachep, GFP_KERNEL);
 106 }
 107
 108 void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
 109 {
 110         if (holder)
 111                 kmem_cache_free(fsnotify_event_holder_cachep, holder);
 112 }
 113
 114 /*
 115  * Find the private data that the group previously attached to this event when
 116  * the group added the event to the notification queue (fsnotify_add_notify_event)
 117  */
 118 struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group, struct fsnotify_event *event)
 119 {
 120         struct fsnotify_event_private_data *lpriv;
 121         struct fsnotify_event_private_data *priv = NULL;
 122
 123         assert_spin_locked(&event->lock);
 124
 125         list_for_each_entry(lpriv, &event->private_data_list, event_list) {
 126                 if (lpriv->group == group) {
 127                         priv = lpriv;
 128                         list_del(&priv->event_list);
 129                         break;
 130                 }
 131         }
 132         return priv;
 133 }
 134
 135 /*
 136  * Add an event to the group notification queue.  The group can later pull this
 137  * event off the queue to deal with.  If the event is successfully added to the
 138  * group's notification queue, a reference is taken on event.
 139  */
 140 struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
 141                                                  struct fsnotify_event_private_data *priv,
 142                                                  struct fsnotify_event *(*merge)(struct list_head *,
 143                                                                                  struct fsnotify_event *))
 144 {
 145         struct fsnotify_event *return_event = NULL;
 146         struct fsnotify_event_holder *holder = NULL;
 147         struct list_head *list = &group->notification_list;
 148
 149         pr_debug("%s: group=%p event=%p priv=%p\n", __func__, group, event, priv);
 150
 151         /*
 152          * There is one fsnotify_event_holder embedded inside each fsnotify_event.
 153          * Check if we expect to be able to use that holder.  If not alloc a new
 154          * holder.
 155          * For the overflow event it's possible that something will use the in
 156          * event holder before we get the lock so we may need to jump back and
 157          * alloc a new holder, this can't happen for most events...
 158          */
 159         if (!list_empty(&event->holder.event_list)) {
 160 alloc_holder:
 161                 holder = fsnotify_alloc_event_holder();
 162                 if (!holder)
 163                         return ERR_PTR(-ENOMEM);
 164         }
 165
 166         mutex_lock(&group->notification_mutex);
 167
 168         if (group->q_len >= group->max_events) {
 169                 event = q_overflow_event;
 170
 171                 /*
 172                  * we need to return the overflow event
 173                  * which means we need a ref
 174                  */
 175                 fsnotify_get_event(event);
 176                 return_event = event;
 177
 178                 /* sorry, no private data on the overflow event */
 179                 priv = NULL;
 180         }
 181
 182         if (!list_empty(list) && merge) {
 183                 struct fsnotify_event *tmp;
 184
 185                 tmp = merge(list, event);
 186                 if (tmp) {
 187                         mutex_unlock(&group->notification_mutex);
 188
 189                         if (return_event)
 190                                 fsnotify_put_event(return_event);
 191                         if (holder != &event->holder)
 192                                 fsnotify_destroy_event_holder(holder);
 193                         return tmp;
 194                 }
 195         }
 196
 197         spin_lock(&event->lock);
 198
 199         if (list_empty(&event->holder.event_list)) {
 200                 if (unlikely(holder))
 201                         fsnotify_destroy_event_holder(holder);
 202                 holder = &event->holder;
 203         } else if (unlikely(!holder)) {
 204                 /* between the time we checked above and got the lock the in
 205                  * event holder was used, go back and get a new one */
 206                 spin_unlock(&event->lock);
 207                 mutex_unlock(&group->notification_mutex);
 208
 209                 if (return_event) {
 210                         fsnotify_put_event(return_event);
 211                         return_event = NULL;
 212                 }
 213
 214                 goto alloc_holder;
 215         }
 216
 217         group->q_len++;
 218         holder->event = event;
 219
 220         fsnotify_get_event(event);
 221         list_add_tail(&holder->event_list, list);
 222         if (priv)
 223                 list_add_tail(&priv->event_list, &event->private_data_list);
 224         spin_unlock(&event->lock);
 225         mutex_unlock(&group->notification_mutex);
 226
 227         wake_up(&group->notification_waitq);
 228         return return_event;
 229 }
 230
 231 /*
 232  * Remove and return the first event from the notification list.  There is a
 233  * reference held on this event since it was on the list.  It is the responsibility
 234  * of the caller to drop this reference.
 235  */
 236 struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group)
 237 {
 238         struct fsnotify_event *event;
 239         struct fsnotify_event_holder *holder;
 240
 241         BUG_ON(!mutex_is_locked(&group->notification_mutex));
 242
 243         pr_debug("%s: group=%p\n", __func__, group);
 244
 245         holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
 246
 247         event = holder->event;
 248
 249         spin_lock(&event->lock);
 250         holder->event = NULL;
 251         list_del_init(&holder->event_list);
 252         spin_unlock(&event->lock);
 253
 254         /* event == holder means we are referenced through the in event holder */
 255         if (holder != &event->holder)
 256                 fsnotify_destroy_event_holder(holder);
 257
 258         group->q_len--;
 259
 260         return event;
 261 }
 262
 263 /*
 264  * This will not remove the event, that must be done with fsnotify_remove_notify_event()
 265  */
 266 struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group)
 267 {
 268         struct fsnotify_event *event;
 269         struct fsnotify_event_holder *holder;
 270
 271         BUG_ON(!mutex_is_locked(&group->notification_mutex));
 272
 273         holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
 274         event = holder->event;
 275
 276         return event;
 277 }
 278
 279 /*
 280  * Called when a group is being torn down to clean up any outstanding
 281  * event notifications.
 282  */
 283 void fsnotify_flush_notify(struct fsnotify_group *group)
 284 {
 285         struct fsnotify_event *event;
 286         struct fsnotify_event_private_data *priv;
 287
 288         mutex_lock(&group->notification_mutex);
 289         while (!fsnotify_notify_queue_is_empty(group)) {
 290                 event = fsnotify_remove_notify_event(group);
 291                 /* if they don't implement free_event_priv they better not have attached any */
 292                 if (group->ops->free_event_priv) {
 293                         spin_lock(&event->lock);
 294                         priv = fsnotify_remove_priv_from_event(group, event);
 295                         spin_unlock(&event->lock);
 296                         if (priv)
 297                                 group->ops->free_event_priv(priv);
 298                 }
 299                 fsnotify_put_event(event); /* matches fsnotify_add_notify_event */
 300         }
 301         mutex_unlock(&group->notification_mutex);
 302 }
 303
 304 static void initialize_event(struct fsnotify_event *event)
 305 {
 306         INIT_LIST_HEAD(&event->holder.event_list);
 307         atomic_set(&event->refcnt, 1);
 308
 309         spin_lock_init(&event->lock);
 310
 311         INIT_LIST_HEAD(&event->private_data_list);
 312 }
 313
 314 /*
 315  * Caller damn well better be holding whatever mutex is protecting the
 316  * old_holder->event_list and the new_event must be a clean event which
 317  * cannot be found anywhere else in the kernel.
 318  */
 319 int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
 320                            struct fsnotify_event *new_event)
 321 {
 322         struct fsnotify_event *old_event = old_holder->event;
 323         struct fsnotify_event_holder *new_holder = &new_event->holder;
 324
 325         enum event_spinlock_class {
 326                 SPINLOCK_OLD,
 327                 SPINLOCK_NEW,
 328         };
 329
 330         pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, new_event);
 331
 332         /*
 333          * if the new_event's embedded holder is in use someone
 334          * screwed up and didn't give us a clean new event.
 335          */
 336         BUG_ON(!list_empty(&new_holder->event_list));
 337
 338         spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
 339         spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
 340
 341         new_holder->event = new_event;
 342         list_replace_init(&old_holder->event_list, &new_holder->event_list);
 343
 344         spin_unlock(&new_event->lock);
 345         spin_unlock(&old_event->lock);
 346
 347         /* event == holder means we are referenced through the in event holder */
 348         if (old_holder != &old_event->holder)
 349                 fsnotify_destroy_event_holder(old_holder);
 350
 351         fsnotify_get_event(new_event); /* on the list take reference */
 352         fsnotify_put_event(old_event); /* off the list, drop reference */
 353
 354         return 0;
 355 }
 356
 357 struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event)
 358 {
 359         struct fsnotify_event *event;
 360
 361         event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
 362         if (!event)
 363                 return NULL;
 364
 365         pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, event);
 366
 367         memcpy(event, old_event, sizeof(*event));
 368         initialize_event(event);
 369
 370         if (event->name_len) {
 371                 event->file_name = kstrdup(old_event->file_name, GFP_KERNEL);
 372                 if (!event->file_name) {
 373                         kmem_cache_free(fsnotify_event_cachep, event);
 374                         return NULL;
 375                 }
 376         }
 377         event->tgid = get_pid(old_event->tgid);
 378         if (event->data_type == FSNOTIFY_EVENT_PATH)
 379                 path_get(&event->path);
 380
 381         return event;
 382 }
 383
 384 /*
 385  * fsnotify_create_event - Allocate a new event which will be sent to each
 386  * group's handle_event function if the group was interested in this
 387  * particular event.
 388  *
 389  * @to_tell the inode which is supposed to receive the event (sometimes a
 390  *      parent of the inode to which the event happened.
 391  * @mask what actually happened.
 392  * @data pointer to the object which was actually affected
 393  * @data_type flag indication if the data is a file, path, inode, nothing...
 394  * @name the filename, if available
 395  */
 396 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
 397                                              int data_type, const unsigned char *name,
 398                                              u32 cookie, gfp_t gfp)
 399 {
 400         struct fsnotify_event *event;
 401
 402         event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
 403         if (!event)
 404                 return NULL;
 405
 406         pr_debug("%s: event=%p to_tell=%p mask=%x data=%p data_type=%d\n",
 407                  __func__, event, to_tell, mask, data, data_type);
 408
 409         initialize_event(event);
 410
 411         if (name) {
 412                 event->file_name = kstrdup(name, gfp);
 413                 if (!event->file_name) {
 414                         kmem_cache_free(fsnotify_event_cachep, event);
 415                         return NULL;
 416                 }
 417                 event->name_len = strlen(event->file_name);
 418         }
 419
 420         event->tgid = get_pid(task_tgid(current));
 421         event->sync_cookie = cookie;
 422         event->to_tell = to_tell;
 423         event->data_type = data_type;
 424
 425         switch (data_type) {
 426         case FSNOTIFY_EVENT_PATH: {
 427                 struct path *path = data;
 428                 event->path.dentry = path->dentry;
 429                 event->path.mnt = path->mnt;
 430                 path_get(&event->path);
 431                 break;
 432         }
 433         case FSNOTIFY_EVENT_INODE:
 434                 event->inode = data;
 435                 break;
 436         case FSNOTIFY_EVENT_NONE:
 437                 event->inode = NULL;
 438                 event->path.dentry = NULL;
 439                 event->path.mnt = NULL;
 440                 break;
 441         default:
 442                 BUG();
 443         }
 444
 445         event->mask = mask;
 446
 447         return event;
 448 }
 449
 450 static __init int fsnotify_notification_init(void)
 451 {
 452         fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
 453         fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
 454
 455         q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
 456                                                  FSNOTIFY_EVENT_NONE, NULL, 0,
 457                                                  GFP_KERNEL);
 458         if (!q_overflow_event)
 459                 panic("unable to allocate fsnotify q_overflow_event\n");
 460
 461         return 0;
 462 }
 463 subsys_initcall(fsnotify_notification_init);