fs/notify/inotify/inotify_user.c

   1 /*
   2  * fs/inotify_user.c - inotify support for userspace
   3  *
   4  * Authors:
   5  *      John McCutchan  <ttb@tentacle.dhs.org>
   6  *      Robert Love     <rml@novell.com>
   7  *
   8  * Copyright (C) 2005 John McCutchan
   9  * Copyright 2006 Hewlett-Packard Development Company, L.P.
  10  *
  11  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
  12  * inotify was largely rewriten to make use of the fsnotify infrastructure
  13  *
  14  * This program is free software; you can redistribute it and/or modify it
  15  * under the terms of the GNU General Public License as published by the
  16  * Free Software Foundation; either version 2, or (at your option) any
  17  * later version.
  18  *
  19  * This program is distributed in the hope that it will be useful, but
  20  * WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  * General Public License for more details.
  23  */
  24
  25 #include <linux/file.h>
  26 #include <linux/fs.h> /* struct inode */
  27 #include <linux/fsnotify_backend.h>
  28 #include <linux/idr.h>
  29 #include <linux/init.h> /* module_init */
  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 */
  38 #include <linux/syscalls.h>
  39 #include <linux/types.h>
  40 #include <linux/uaccess.h>
  41 #include <linux/poll.h>
  42 #include <linux/wait.h>
  43
  44 #include "inotify.h"
  45
  46 #include <asm/ioctls.h>
  47
  48 static struct vfsmount *inotify_mnt __read_mostly;
  49
  50 /* this just sits here and wastes global memory.  used to just pad userspace messages with zeros */
  51 static struct inotify_event nul_inotify_event;
  52
  53 /* these are configurable via /proc/sys/fs/inotify/ */
  54 static int inotify_max_user_instances __read_mostly;
  55 static int inotify_max_queued_events __read_mostly;
  56 int inotify_max_user_watches __read_mostly;
  57
  58 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  59 struct kmem_cache *event_priv_cachep __read_mostly;
  60 static struct fsnotify_event *inotify_ignored_event;
  61
  62 /*
  63  * When inotify registers a new group it increments this and uses that
  64  * value as an offset to set the fsnotify group "name" and priority.
  65  */
  66 static atomic_t inotify_grp_num;
  67
  68 #ifdef CONFIG_SYSCTL
  69
  70 #include <linux/sysctl.h>
  71
  72 static int zero;
  73
  74 ctl_table inotify_table[] = {
  75         {
  76                 .ctl_name       = INOTIFY_MAX_USER_INSTANCES,
  77                 .procname       = "max_user_instances",
  78                 .data           = &inotify_max_user_instances,
  79                 .maxlen         = sizeof(int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &zero,
  84         },
  85         {
  86                 .ctl_name       = INOTIFY_MAX_USER_WATCHES,
  87                 .procname       = "max_user_watches",
  88                 .data           = &inotify_max_user_watches,
  89                 .maxlen         = sizeof(int),
  90                 .mode           = 0644,
  91                 .proc_handler   = &proc_dointvec_minmax,
  92                 .strategy       = &sysctl_intvec,
  93                 .extra1         = &zero,
  94         },
  95         {
  96                 .ctl_name       = INOTIFY_MAX_QUEUED_EVENTS,
  97                 .procname       = "max_queued_events",
  98                 .data           = &inotify_max_queued_events,
  99                 .maxlen         = sizeof(int),
 100                 .mode           = 0644,
 101                 .proc_handler   = &proc_dointvec_minmax,
 102                 .strategy       = &sysctl_intvec,
 103                 .extra1         = &zero
 104         },
 105         { .ctl_name = 0 }
 106 };
 107 #endif /* CONFIG_SYSCTL */
 108
 109 static inline __u32 inotify_arg_to_mask(u32 arg)
 110 {
 111         __u32 mask;
 112
 113         /* everything should accept their own ignored and cares about children */
 114         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
 115
 116         /* mask off the flags used to open the fd */
 117         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
 118
 119         return mask;
 120 }
 121
 122 static inline u32 inotify_mask_to_arg(__u32 mask)
 123 {
 124         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
 125                        IN_Q_OVERFLOW);
 126 }
 127
 128 /* intofiy userspace file descriptor functions */
 129 static unsigned int inotify_poll(struct file *file, poll_table *wait)
 130 {
 131         struct fsnotify_group *group = file->private_data;
 132         int ret = 0;
 133
 134         poll_wait(file, &group->notification_waitq, wait);
 135         mutex_lock(&group->notification_mutex);
 136         if (!fsnotify_notify_queue_is_empty(group))
 137                 ret = POLLIN | POLLRDNORM;
 138         mutex_unlock(&group->notification_mutex);
 139
 140         return ret;
 141 }
 142
 143 /*
 144  * Get an inotify_kernel_event if one exists and is small
 145  * enough to fit in "count". Return an error pointer if
 146  * not large enough.
 147  *
 148  * Called with the group->notification_mutex held.
 149  */
 150 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 151                                             size_t count)
 152 {
 153         size_t event_size = sizeof(struct inotify_event);
 154         struct fsnotify_event *event;
 155
 156         if (fsnotify_notify_queue_is_empty(group))
 157                 return NULL;
 158
 159         event = fsnotify_peek_notify_event(group);
 160
 161         event_size += roundup(event->name_len, event_size);
 162
 163         if (event_size > count)
 164                 return ERR_PTR(-EINVAL);
 165
 166         /* held the notification_mutex the whole time, so this is the
 167          * same event we peeked above */
 168         fsnotify_remove_notify_event(group);
 169
 170         return event;
 171 }
 172
 173 /*
 174  * Copy an event to user space, returning how much we copied.
 175  *
 176  * We already checked that the event size is smaller than the
 177  * buffer we had in "get_one_event()" above.
 178  */
 179 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 180                                   struct fsnotify_event *event,
 181                                   char __user *buf)
 182 {
 183         struct inotify_event inotify_event;
 184         struct fsnotify_event_private_data *fsn_priv;
 185         struct inotify_event_private_data *priv;
 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;
 209
 210         /* send the main event */
 211         if (copy_to_user(buf, &inotify_event, event_size))
 212                 return -EFAULT;
 213
 214         buf += event_size;
 215
 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))
 225                         return -EFAULT;
 226                 buf += event->name_len;
 227
 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;
 233         }
 234
 235         return event_size;
 236 }
 237
 238 static ssize_t inotify_read(struct file *file, char __user *buf,
 239                             size_t count, loff_t *pos)
 240 {
 241         struct fsnotify_group *group;
 242         struct fsnotify_event *kevent;
 243         char __user *start;
 244         int ret;
 245         DEFINE_WAIT(wait);
 246
 247         start = buf;
 248         group = file->private_data;
 249
 250         while (1) {
 251                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 252
 253                 mutex_lock(&group->notification_mutex);
 254                 kevent = get_one_event(group, count);
 255                 mutex_unlock(&group->notification_mutex);
 256
 257                 if (kevent) {
 258                         ret = PTR_ERR(kevent);
 259                         if (IS_ERR(kevent))
 260                                 break;
 261                         ret = copy_event_to_user(group, kevent, buf);
 262                         fsnotify_put_event(kevent);
 263                         if (ret < 0)
 264                                 break;
 265                         buf += ret;
 266                         count -= ret;
 267                         continue;
 268                 }
 269
 270                 ret = -EAGAIN;
 271                 if (file->f_flags & O_NONBLOCK)
 272                         break;
 273                 ret = -EINTR;
 274                 if (signal_pending(current))
 275                         break;
 276
 277                 if (start != buf)
 278                         break;
 279
 280                 schedule();
 281         }
 282
 283         finish_wait(&group->notification_waitq, &wait);
 284         if (start != buf && ret != -EFAULT)
 285                 ret = buf - start;
 286         return ret;
 287 }
 288
 289 static int inotify_fasync(int fd, struct file *file, int on)
 290 {
 291         struct fsnotify_group *group = file->private_data;
 292
 293         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 294 }
 295
 296 static int inotify_release(struct inode *ignored, struct file *file)
 297 {
 298         struct fsnotify_group *group = file->private_data;
 299
 300         fsnotify_clear_marks_by_group(group);
 301
 302         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 303         fsnotify_put_group(group);
 304
 305         return 0;
 306 }
 307
 308 static long inotify_ioctl(struct file *file, unsigned int cmd,
 309                           unsigned long arg)
 310 {
 311         struct fsnotify_group *group;
 312         struct fsnotify_event_holder *holder;
 313         struct fsnotify_event *event;
 314         void __user *p;
 315         int ret = -ENOTTY;
 316         size_t send_len = 0;
 317
 318         group = file->private_data;
 319         p = (void __user *) arg;
 320
 321         switch (cmd) {
 322         case FIONREAD:
 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);
 332                 break;
 333         }
 334
 335         return ret;
 336 }
 337
 338 static const struct file_operations inotify_fops = {
 339         .poll           = inotify_poll,
 340         .read           = inotify_read,
 341         .fasync         = inotify_fasync,
 342         .release        = inotify_release,
 343         .unlocked_ioctl = inotify_ioctl,
 344         .compat_ioctl   = inotify_ioctl,
 345 };
 346
 347
 348 /*
 349  * find_inode - resolve a user-given path to a specific inode
 350  */
 351 static 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  */
 379 void 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
 416 skip_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 */
 429 static 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
 436 static 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
 457 find_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;
 477 retry:
 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
 530 out_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
 539 static 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 */
 563 SYSCALL_DEFINE1(inotify_init1, int, flags)
 564 {
 565         struct fsnotify_group *group;
 566         struct user_struct *user;
 567         struct file *filp;
 568         int fd, ret;
 569
 570         /* Check the IN_* constants for consistency.  */
 571         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 572         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 573
 574         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 575                 return -EINVAL;
 576
 577         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 578         if (fd < 0)
 579                 return fd;
 580
 581         filp = get_empty_filp();
 582         if (!filp) {
 583                 ret = -ENFILE;
 584                 goto out_put_fd;
 585         }
 586
 587         user = get_current_user();
 588         if (unlikely(atomic_read(&user->inotify_devs) >=
 589                         inotify_max_user_instances)) {
 590                 ret = -EMFILE;
 591                 goto out_free_uid;
 592         }
 593
 594         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 595         group = inotify_new_group(user, inotify_max_queued_events);
 596         if (IS_ERR(group)) {
 597                 ret = PTR_ERR(group);
 598                 goto out_free_uid;
 599         }
 600
 601         filp->f_op = &inotify_fops;
 602         filp->f_path.mnt = mntget(inotify_mnt);
 603         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 604         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 605         filp->f_mode = FMODE_READ;
 606         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 607         filp->private_data = group;
 608
 609         atomic_inc(&user->inotify_devs);
 610
 611         fd_install(fd, filp);
 612
 613         return fd;
 614
 615 out_free_uid:
 616         free_uid(user);
 617         put_filp(filp);
 618 out_put_fd:
 619         put_unused_fd(fd);
 620         return ret;
 621 }
 622
 623 SYSCALL_DEFINE0(inotify_init)
 624 {
 625         return sys_inotify_init1(0);
 626 }
 627
 628 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 629                 u32, mask)
 630 {
 631         struct fsnotify_group *group;
 632         struct inode *inode;
 633         struct path path;
 634         struct file *filp;
 635         int ret, fput_needed;
 636         unsigned flags = 0;
 637
 638         filp = fget_light(fd, &fput_needed);
 639         if (unlikely(!filp))
 640                 return -EBADF;
 641
 642         /* verify that this is indeed an inotify instance */
 643         if (unlikely(filp->f_op != &inotify_fops)) {
 644                 ret = -EINVAL;
 645                 goto fput_and_out;
 646         }
 647
 648         if (!(mask & IN_DONT_FOLLOW))
 649                 flags |= LOOKUP_FOLLOW;
 650         if (mask & IN_ONLYDIR)
 651                 flags |= LOOKUP_DIRECTORY;
 652
 653         ret = inotify_find_inode(pathname, &path, flags);
 654         if (ret)
 655                 goto fput_and_out;
 656
 657         /* inode held in place by reference to path; group by fget on fd */
 658         inode = path.dentry->d_inode;
 659         group = filp->private_data;
 660
 661         /* create/update an inode mark */
 662         ret = inotify_update_watch(group, inode, mask);
 663         if (unlikely(ret))
 664                 goto path_put_and_out;
 665
 666 path_put_and_out:
 667         path_put(&path);
 668 fput_and_out:
 669         fput_light(filp, fput_needed);
 670         return ret;
 671 }
 672
 673 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 674 {
 675         struct fsnotify_group *group;
 676         struct fsnotify_mark_entry *entry;
 677         struct file *filp;
 678         int ret = 0, fput_needed;
 679
 680         filp = fget_light(fd, &fput_needed);
 681         if (unlikely(!filp))
 682                 return -EBADF;
 683
 684         /* verify that this is indeed an inotify instance */
 685         if (unlikely(filp->f_op != &inotify_fops)) {
 686                 ret = -EINVAL;
 687                 goto out;
 688         }
 689
 690         group = filp->private_data;
 691
 692         spin_lock(&group->inotify_data.idr_lock);
 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);
 704
 705 out:
 706         fput_light(filp, fput_needed);
 707         return ret;
 708 }
 709
 710 static int
 711 inotify_get_sb(struct file_system_type *fs_type, int flags,
 712                const char *dev_name, void *data, struct vfsmount *mnt)
 713 {
 714         return get_sb_pseudo(fs_type, "inotify", NULL,
 715                         INOTIFYFS_SUPER_MAGIC, mnt);
 716 }
 717
 718 static struct file_system_type inotify_fs_type = {
 719     .name       = "inotifyfs",
 720     .get_sb     = inotify_get_sb,
 721     .kill_sb    = kill_anon_super,
 722 };
 723
 724 /*
 725  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 726  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 727  * must result in panic().
 728  */
 729 static int __init inotify_user_setup(void)
 730 {
 731         int ret;
 732
 733         ret = register_filesystem(&inotify_fs_type);
 734         if (unlikely(ret))
 735                 panic("inotify: register_filesystem returned %d!\n", ret);
 736
 737         inotify_mnt = kern_mount(&inotify_fs_type);
 738         if (IS_ERR(inotify_mnt))
 739                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 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
 747         inotify_max_queued_events = 16384;
 748         inotify_max_user_instances = 128;
 749         inotify_max_user_watches = 8192;
 750
 751         return 0;
 752 }
 753 module_init(inotify_user_setup);