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
  61 /*
  62  * When inotify registers a new group it increments this and uses that
  63  * value as an offset to set the fsnotify group "name" and priority.
  64  */
  65 static atomic_t inotify_grp_num;
  66
  67 #ifdef CONFIG_SYSCTL
  68
  69 #include <linux/sysctl.h>
  70
  71 static int zero;
  72
  73 ctl_table inotify_table[] = {
  74         {
  75                 .ctl_name       = INOTIFY_MAX_USER_INSTANCES,
  76                 .procname       = "max_user_instances",
  77                 .data           = &inotify_max_user_instances,
  78                 .maxlen         = sizeof(int),
  79                 .mode           = 0644,
  80                 .proc_handler   = &proc_dointvec_minmax,
  81                 .strategy       = &sysctl_intvec,
  82                 .extra1         = &zero,
  83         },
  84         {
  85                 .ctl_name       = INOTIFY_MAX_USER_WATCHES,
  86                 .procname       = "max_user_watches",
  87                 .data           = &inotify_max_user_watches,
  88                 .maxlen         = sizeof(int),
  89                 .mode           = 0644,
  90                 .proc_handler   = &proc_dointvec_minmax,
  91                 .strategy       = &sysctl_intvec,
  92                 .extra1         = &zero,
  93         },
  94         {
  95                 .ctl_name       = INOTIFY_MAX_QUEUED_EVENTS,
  96                 .procname       = "max_queued_events",
  97                 .data           = &inotify_max_queued_events,
  98                 .maxlen         = sizeof(int),
  99                 .mode           = 0644,
 100                 .proc_handler   = &proc_dointvec_minmax,
 101                 .strategy       = &sysctl_intvec,
 102                 .extra1         = &zero
 103         },
 104         { .ctl_name = 0 }
 105 };
 106 #endif /* CONFIG_SYSCTL */
 107
 108 static inline __u32 inotify_arg_to_mask(u32 arg)
 109 {
 110         __u32 mask;
 111
 112         /* everything should accept their own ignored and cares about children */
 113         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
 114
 115         /* mask off the flags used to open the fd */
 116         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
 117
 118         return mask;
 119 }
 120
 121 static inline u32 inotify_mask_to_arg(__u32 mask)
 122 {
 123         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
 124                        IN_Q_OVERFLOW);
 125 }
 126
 127 /* intofiy userspace file descriptor functions */
 128 static unsigned int inotify_poll(struct file *file, poll_table *wait)
 129 {
 130         struct fsnotify_group *group = file->private_data;
 131         int ret = 0;
 132
 133         poll_wait(file, &group->notification_waitq, wait);
 134         mutex_lock(&group->notification_mutex);
 135         if (!fsnotify_notify_queue_is_empty(group))
 136                 ret = POLLIN | POLLRDNORM;
 137         mutex_unlock(&group->notification_mutex);
 138
 139         return ret;
 140 }
 141
 142 /*
 143  * Get an inotify_kernel_event if one exists and is small
 144  * enough to fit in "count". Return an error pointer if
 145  * not large enough.
 146  *
 147  * Called with the group->notification_mutex held.
 148  */
 149 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 150                                             size_t count)
 151 {
 152         size_t event_size = sizeof(struct inotify_event);
 153         struct fsnotify_event *event;
 154
 155         if (fsnotify_notify_queue_is_empty(group))
 156                 return NULL;
 157
 158         event = fsnotify_peek_notify_event(group);
 159
 160         event_size += roundup(event->name_len, event_size);
 161
 162         if (event_size > count)
 163                 return ERR_PTR(-EINVAL);
 164
 165         /* held the notification_mutex the whole time, so this is the
 166          * same event we peeked above */
 167         fsnotify_remove_notify_event(group);
 168
 169         return event;
 170 }
 171
 172 /*
 173  * Copy an event to user space, returning how much we copied.
 174  *
 175  * We already checked that the event size is smaller than the
 176  * buffer we had in "get_one_event()" above.
 177  */
 178 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 179                                   struct fsnotify_event *event,
 180                                   char __user *buf)
 181 {
 182         struct inotify_event inotify_event;
 183         struct fsnotify_event_private_data *fsn_priv;
 184         struct inotify_event_private_data *priv;
 185         size_t event_size = sizeof(struct inotify_event);
 186         size_t name_len;
 187
 188         /* we get the inotify watch descriptor from the event private data */
 189         spin_lock(&event->lock);
 190         fsn_priv = fsnotify_remove_priv_from_event(group, event);
 191         spin_unlock(&event->lock);
 192
 193         if (!fsn_priv)
 194                 inotify_event.wd = -1;
 195         else {
 196                 priv = container_of(fsn_priv, struct inotify_event_private_data,
 197                                     fsnotify_event_priv_data);
 198                 inotify_event.wd = priv->wd;
 199                 inotify_free_event_priv(fsn_priv);
 200         }
 201
 202         /* round up event->name_len so it is a multiple of event_size */
 203         name_len = roundup(event->name_len, event_size);
 204         inotify_event.len = name_len;
 205
 206         inotify_event.mask = inotify_mask_to_arg(event->mask);
 207         inotify_event.cookie = event->sync_cookie;
 208
 209         /* send the main event */
 210         if (copy_to_user(buf, &inotify_event, event_size))
 211                 return -EFAULT;
 212
 213         buf += event_size;
 214
 215         /*
 216          * fsnotify only stores the pathname, so here we have to send the pathname
 217          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 218          * with zeros.  I get my zeros from the nul_inotify_event.
 219          */
 220         if (name_len) {
 221                 unsigned int len_to_zero = name_len - event->name_len;
 222                 /* copy the path name */
 223                 if (copy_to_user(buf, event->file_name, event->name_len))
 224                         return -EFAULT;
 225                 buf += event->name_len;
 226
 227                 /* fill userspace with 0's from nul_inotify_event */
 228                 if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
 229                         return -EFAULT;
 230                 buf += len_to_zero;
 231                 event_size += name_len;
 232         }
 233
 234         return event_size;
 235 }
 236
 237 static ssize_t inotify_read(struct file *file, char __user *buf,
 238                             size_t count, loff_t *pos)
 239 {
 240         struct fsnotify_group *group;
 241         struct fsnotify_event *kevent;
 242         char __user *start;
 243         int ret;
 244         DEFINE_WAIT(wait);
 245
 246         start = buf;
 247         group = file->private_data;
 248
 249         while (1) {
 250                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 251
 252                 mutex_lock(&group->notification_mutex);
 253                 kevent = get_one_event(group, count);
 254                 mutex_unlock(&group->notification_mutex);
 255
 256                 if (kevent) {
 257                         ret = PTR_ERR(kevent);
 258                         if (IS_ERR(kevent))
 259                                 break;
 260                         ret = copy_event_to_user(group, kevent, buf);
 261                         fsnotify_put_event(kevent);
 262                         if (ret < 0)
 263                                 break;
 264                         buf += ret;
 265                         count -= ret;
 266                         continue;
 267                 }
 268
 269                 ret = -EAGAIN;
 270                 if (file->f_flags & O_NONBLOCK)
 271                         break;
 272                 ret = -EINTR;
 273                 if (signal_pending(current))
 274                         break;
 275
 276                 if (start != buf)
 277                         break;
 278
 279                 schedule();
 280         }
 281
 282         finish_wait(&group->notification_waitq, &wait);
 283         if (start != buf && ret != -EFAULT)
 284                 ret = buf - start;
 285         return ret;
 286 }
 287
 288 static int inotify_fasync(int fd, struct file *file, int on)
 289 {
 290         struct fsnotify_group *group = file->private_data;
 291
 292         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 293 }
 294
 295 static int inotify_release(struct inode *ignored, struct file *file)
 296 {
 297         struct fsnotify_group *group = file->private_data;
 298         struct user_struct *user = group->inotify_data.user;
 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         atomic_dec(&user->inotify_devs);
 306
 307         return 0;
 308 }
 309
 310 static long inotify_ioctl(struct file *file, unsigned int cmd,
 311                           unsigned long arg)
 312 {
 313         struct fsnotify_group *group;
 314         struct fsnotify_event_holder *holder;
 315         struct fsnotify_event *event;
 316         void __user *p;
 317         int ret = -ENOTTY;
 318         size_t send_len = 0;
 319
 320         group = file->private_data;
 321         p = (void __user *) arg;
 322
 323         switch (cmd) {
 324         case FIONREAD:
 325                 mutex_lock(&group->notification_mutex);
 326                 list_for_each_entry(holder, &group->notification_list, event_list) {
 327                         event = holder->event;
 328                         send_len += sizeof(struct inotify_event);
 329                         send_len += roundup(event->name_len,
 330                                              sizeof(struct inotify_event));
 331                 }
 332                 mutex_unlock(&group->notification_mutex);
 333                 ret = put_user(send_len, (int __user *) p);
 334                 break;
 335         }
 336
 337         return ret;
 338 }
 339
 340 static const struct file_operations inotify_fops = {
 341         .poll           = inotify_poll,
 342         .read           = inotify_read,
 343         .fasync         = inotify_fasync,
 344         .release        = inotify_release,
 345         .unlocked_ioctl = inotify_ioctl,
 346         .compat_ioctl   = inotify_ioctl,
 347 };
 348
 349
 350 /*
 351  * find_inode - resolve a user-given path to a specific inode
 352  */
 353 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 354 {
 355         int error;
 356
 357         error = user_path_at(AT_FDCWD, dirname, flags, path);
 358         if (error)
 359                 return error;
 360         /* you can only watch an inode if you have read permissions on it */
 361         error = inode_permission(path->dentry->d_inode, MAY_READ);
 362         if (error)
 363                 path_put(path);
 364         return error;
 365 }
 366
 367 static void inotify_remove_from_idr(struct fsnotify_group *group,
 368                                     struct inotify_inode_mark_entry *ientry)
 369 {
 370         struct idr *idr;
 371
 372         spin_lock(&group->inotify_data.idr_lock);
 373         idr = &group->inotify_data.idr;
 374         idr_remove(idr, ientry->wd);
 375         spin_unlock(&group->inotify_data.idr_lock);
 376         ientry->wd = -1;
 377 }
 378 /*
 379  * Send IN_IGNORED for this wd, remove this wd from the idr, and drop the
 380  * internal reference help on the mark because it is in the idr.
 381  */
 382 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 383                                     struct fsnotify_group *group)
 384 {
 385         struct inotify_inode_mark_entry *ientry;
 386         struct fsnotify_event *ignored_event;
 387         struct inotify_event_private_data *event_priv;
 388         struct fsnotify_event_private_data *fsn_event_priv;
 389
 390         ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 391                                               FSNOTIFY_EVENT_NONE, NULL, 0,
 392                                               GFP_NOFS);
 393         if (!ignored_event)
 394                 return;
 395
 396         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 397
 398         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 399         if (unlikely(!event_priv))
 400                 goto skip_send_ignore;
 401
 402         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 403
 404         fsn_event_priv->group = group;
 405         event_priv->wd = ientry->wd;
 406
 407         fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
 408
 409         /* did the private data get added? */
 410         if (list_empty(&fsn_event_priv->event_list))
 411                 inotify_free_event_priv(fsn_event_priv);
 412
 413 skip_send_ignore:
 414
 415         /* matches the reference taken when the event was created */
 416         fsnotify_put_event(ignored_event);
 417
 418         /* remove this entry from the idr */
 419         inotify_remove_from_idr(group, ientry);
 420
 421         /* removed from idr, drop that reference */
 422         fsnotify_put_mark(entry);
 423
 424         atomic_dec(&group->inotify_data.user->inotify_watches);
 425 }
 426
 427 /* ding dong the mark is dead */
 428 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 429 {
 430         struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
 431
 432         kmem_cache_free(inotify_inode_mark_cachep, ientry);
 433 }
 434
 435 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 436 {
 437         struct fsnotify_mark_entry *entry = NULL;
 438         struct inotify_inode_mark_entry *ientry;
 439         struct inotify_inode_mark_entry *tmp_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         tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 451         if (unlikely(!tmp_ientry))
 452                 return -ENOMEM;
 453         /* we set the mask at the end after attaching it */
 454         fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
 455         tmp_ientry->wd = -1;
 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                 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 463         } else {
 464                 ret = -ENOSPC;
 465                 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 466                         goto out_err;
 467 retry:
 468                 ret = -ENOMEM;
 469                 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
 470                         goto out_err;
 471
 472                 spin_lock(&group->inotify_data.idr_lock);
 473                 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
 474                                         group->inotify_data.last_wd,
 475                                         &tmp_ientry->wd);
 476                 spin_unlock(&group->inotify_data.idr_lock);
 477                 if (ret) {
 478                         if (ret == -EAGAIN)
 479                                 goto retry;
 480                         goto out_err;
 481                 }
 482
 483                 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 484                 if (ret) {
 485                         inotify_remove_from_idr(group, tmp_ientry);
 486                         if (ret == -EEXIST)
 487                                 goto find_entry;
 488                         goto out_err;
 489                 }
 490
 491                 /* tmp_ientry has been added to the inode, so we are all set up.
 492                  * now we just need to make sure tmp_ientry doesn't get freed and
 493                  * we need to set up entry and ientry so the generic code can
 494                  * do its thing. */
 495                 ientry = tmp_ientry;
 496                 entry = &ientry->fsn_entry;
 497                 tmp_ientry = NULL;
 498
 499                 atomic_inc(&group->inotify_data.user->inotify_watches);
 500
 501                 /* update the idr hint */
 502                 group->inotify_data.last_wd = ientry->wd;
 503
 504                 /* we put the mark on the idr, take a reference */
 505                 fsnotify_get_mark(entry);
 506         }
 507
 508         ret = ientry->wd;
 509
 510         spin_lock(&entry->lock);
 511
 512         old_mask = entry->mask;
 513         if (add) {
 514                 entry->mask |= mask;
 515                 new_mask = entry->mask;
 516         } else {
 517                 entry->mask = mask;
 518                 new_mask = entry->mask;
 519         }
 520
 521         spin_unlock(&entry->lock);
 522
 523         if (old_mask != new_mask) {
 524                 /* more bits in old than in new? */
 525                 int dropped = (old_mask & ~new_mask);
 526                 /* more bits in this entry than the inode's mask? */
 527                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 528                 /* more bits in this entry than the group? */
 529                 int do_group = (new_mask & ~group->mask);
 530
 531                 /* update the inode with this new entry */
 532                 if (dropped || do_inode)
 533                         fsnotify_recalc_inode_mask(inode);
 534
 535                 /* update the group mask with the new mask */
 536                 if (dropped || do_group)
 537                         fsnotify_recalc_group_mask(group);
 538         }
 539
 540         /* this either matches fsnotify_find_mark_entry, or init_mark_entry
 541          * depending on which path we took... */
 542         fsnotify_put_mark(entry);
 543
 544 out_err:
 545         /* could be an error, could be that we found an existing mark */
 546         if (tmp_ientry) {
 547                 /* on the idr but didn't make it on the inode */
 548                 if (tmp_ientry->wd != -1)
 549                         inotify_remove_from_idr(group, tmp_ientry);
 550                 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 551         }
 552
 553         return ret;
 554 }
 555
 556 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 557 {
 558         struct fsnotify_group *group;
 559         unsigned int grp_num;
 560
 561         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 562         grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
 563         group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
 564         if (IS_ERR(group))
 565                 return group;
 566
 567         group->max_events = max_events;
 568
 569         spin_lock_init(&group->inotify_data.idr_lock);
 570         idr_init(&group->inotify_data.idr);
 571         group->inotify_data.last_wd = 0;
 572         group->inotify_data.user = user;
 573         group->inotify_data.fa = NULL;
 574
 575         return group;
 576 }
 577
 578
 579 /* inotify syscalls */
 580 SYSCALL_DEFINE1(inotify_init1, int, flags)
 581 {
 582         struct fsnotify_group *group;
 583         struct user_struct *user;
 584         struct file *filp;
 585         int fd, ret;
 586
 587         /* Check the IN_* constants for consistency.  */
 588         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 589         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 590
 591         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 592                 return -EINVAL;
 593
 594         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 595         if (fd < 0)
 596                 return fd;
 597
 598         filp = get_empty_filp();
 599         if (!filp) {
 600                 ret = -ENFILE;
 601                 goto out_put_fd;
 602         }
 603
 604         user = get_current_user();
 605         if (unlikely(atomic_read(&user->inotify_devs) >=
 606                         inotify_max_user_instances)) {
 607                 ret = -EMFILE;
 608                 goto out_free_uid;
 609         }
 610
 611         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 612         group = inotify_new_group(user, inotify_max_queued_events);
 613         if (IS_ERR(group)) {
 614                 ret = PTR_ERR(group);
 615                 goto out_free_uid;
 616         }
 617
 618         filp->f_op = &inotify_fops;
 619         filp->f_path.mnt = mntget(inotify_mnt);
 620         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 621         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 622         filp->f_mode = FMODE_READ;
 623         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 624         filp->private_data = group;
 625
 626         atomic_inc(&user->inotify_devs);
 627
 628         fd_install(fd, filp);
 629
 630         return fd;
 631
 632 out_free_uid:
 633         free_uid(user);
 634         put_filp(filp);
 635 out_put_fd:
 636         put_unused_fd(fd);
 637         return ret;
 638 }
 639
 640 SYSCALL_DEFINE0(inotify_init)
 641 {
 642         return sys_inotify_init1(0);
 643 }
 644
 645 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 646                 u32, mask)
 647 {
 648         struct fsnotify_group *group;
 649         struct inode *inode;
 650         struct path path;
 651         struct file *filp;
 652         int ret, fput_needed;
 653         unsigned flags = 0;
 654
 655         filp = fget_light(fd, &fput_needed);
 656         if (unlikely(!filp))
 657                 return -EBADF;
 658
 659         /* verify that this is indeed an inotify instance */
 660         if (unlikely(filp->f_op != &inotify_fops)) {
 661                 ret = -EINVAL;
 662                 goto fput_and_out;
 663         }
 664
 665         if (!(mask & IN_DONT_FOLLOW))
 666                 flags |= LOOKUP_FOLLOW;
 667         if (mask & IN_ONLYDIR)
 668                 flags |= LOOKUP_DIRECTORY;
 669
 670         ret = inotify_find_inode(pathname, &path, flags);
 671         if (ret)
 672                 goto fput_and_out;
 673
 674         /* inode held in place by reference to path; group by fget on fd */
 675         inode = path.dentry->d_inode;
 676         group = filp->private_data;
 677
 678         /* create/update an inode mark */
 679         ret = inotify_update_watch(group, inode, mask);
 680         if (unlikely(ret))
 681                 goto path_put_and_out;
 682
 683 path_put_and_out:
 684         path_put(&path);
 685 fput_and_out:
 686         fput_light(filp, fput_needed);
 687         return ret;
 688 }
 689
 690 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 691 {
 692         struct fsnotify_group *group;
 693         struct fsnotify_mark_entry *entry;
 694         struct file *filp;
 695         int ret = 0, fput_needed;
 696
 697         filp = fget_light(fd, &fput_needed);
 698         if (unlikely(!filp))
 699                 return -EBADF;
 700
 701         /* verify that this is indeed an inotify instance */
 702         if (unlikely(filp->f_op != &inotify_fops)) {
 703                 ret = -EINVAL;
 704                 goto out;
 705         }
 706
 707         group = filp->private_data;
 708
 709         spin_lock(&group->inotify_data.idr_lock);
 710         entry = idr_find(&group->inotify_data.idr, wd);
 711         if (unlikely(!entry)) {
 712                 spin_unlock(&group->inotify_data.idr_lock);
 713                 ret = -EINVAL;
 714                 goto out;
 715         }
 716         fsnotify_get_mark(entry);
 717         spin_unlock(&group->inotify_data.idr_lock);
 718
 719         fsnotify_destroy_mark_by_entry(entry);
 720         fsnotify_put_mark(entry);
 721
 722 out:
 723         fput_light(filp, fput_needed);
 724         return ret;
 725 }
 726
 727 static int
 728 inotify_get_sb(struct file_system_type *fs_type, int flags,
 729                const char *dev_name, void *data, struct vfsmount *mnt)
 730 {
 731         return get_sb_pseudo(fs_type, "inotify", NULL,
 732                         INOTIFYFS_SUPER_MAGIC, mnt);
 733 }
 734
 735 static struct file_system_type inotify_fs_type = {
 736     .name       = "inotifyfs",
 737     .get_sb     = inotify_get_sb,
 738     .kill_sb    = kill_anon_super,
 739 };
 740
 741 /*
 742  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 743  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 744  * must result in panic().
 745  */
 746 static int __init inotify_user_setup(void)
 747 {
 748         int ret;
 749
 750         ret = register_filesystem(&inotify_fs_type);
 751         if (unlikely(ret))
 752                 panic("inotify: register_filesystem returned %d!\n", ret);
 753
 754         inotify_mnt = kern_mount(&inotify_fs_type);
 755         if (IS_ERR(inotify_mnt))
 756                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 757
 758         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
 759         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 760
 761         inotify_max_queued_events = 16384;
 762         inotify_max_user_instances = 128;
 763         inotify_max_user_watches = 8192;
 764
 765         return 0;
 766 }
 767 module_init(inotify_user_setup);