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 /* these are configurable via /proc/sys/fs/inotify/ */
  51 static int inotify_max_user_instances __read_mostly;
  52 static int inotify_max_queued_events __read_mostly;
  53 int inotify_max_user_watches __read_mostly;
  54
  55 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  56 struct kmem_cache *event_priv_cachep __read_mostly;
  57
  58 /*
  59  * When inotify registers a new group it increments this and uses that
  60  * value as an offset to set the fsnotify group "name" and priority.
  61  */
  62 static atomic_t inotify_grp_num;
  63
  64 #ifdef CONFIG_SYSCTL
  65
  66 #include <linux/sysctl.h>
  67
  68 static int zero;
  69
  70 ctl_table inotify_table[] = {
  71         {
  72                 .ctl_name       = INOTIFY_MAX_USER_INSTANCES,
  73                 .procname       = "max_user_instances",
  74                 .data           = &inotify_max_user_instances,
  75                 .maxlen         = sizeof(int),
  76                 .mode           = 0644,
  77                 .proc_handler   = &proc_dointvec_minmax,
  78                 .strategy       = &sysctl_intvec,
  79                 .extra1         = &zero,
  80         },
  81         {
  82                 .ctl_name       = INOTIFY_MAX_USER_WATCHES,
  83                 .procname       = "max_user_watches",
  84                 .data           = &inotify_max_user_watches,
  85                 .maxlen         = sizeof(int),
  86                 .mode           = 0644,
  87                 .proc_handler   = &proc_dointvec_minmax,
  88                 .strategy       = &sysctl_intvec,
  89                 .extra1         = &zero,
  90         },
  91         {
  92                 .ctl_name       = INOTIFY_MAX_QUEUED_EVENTS,
  93                 .procname       = "max_queued_events",
  94                 .data           = &inotify_max_queued_events,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec_minmax,
  98                 .strategy       = &sysctl_intvec,
  99                 .extra1         = &zero
 100         },
 101         { .ctl_name = 0 }
 102 };
 103 #endif /* CONFIG_SYSCTL */
 104
 105 static inline __u32 inotify_arg_to_mask(u32 arg)
 106 {
 107         __u32 mask;
 108
 109         /* everything should accept their own ignored and cares about children */
 110         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
 111
 112         /* mask off the flags used to open the fd */
 113         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
 114
 115         return mask;
 116 }
 117
 118 static inline u32 inotify_mask_to_arg(__u32 mask)
 119 {
 120         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
 121                        IN_Q_OVERFLOW);
 122 }
 123
 124 /* intofiy userspace file descriptor functions */
 125 static unsigned int inotify_poll(struct file *file, poll_table *wait)
 126 {
 127         struct fsnotify_group *group = file->private_data;
 128         int ret = 0;
 129
 130         poll_wait(file, &group->notification_waitq, wait);
 131         mutex_lock(&group->notification_mutex);
 132         if (!fsnotify_notify_queue_is_empty(group))
 133                 ret = POLLIN | POLLRDNORM;
 134         mutex_unlock(&group->notification_mutex);
 135
 136         return ret;
 137 }
 138
 139 /*
 140  * Get an inotify_kernel_event if one exists and is small
 141  * enough to fit in "count". Return an error pointer if
 142  * not large enough.
 143  *
 144  * Called with the group->notification_mutex held.
 145  */
 146 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 147                                             size_t count)
 148 {
 149         size_t event_size = sizeof(struct inotify_event);
 150         struct fsnotify_event *event;
 151
 152         if (fsnotify_notify_queue_is_empty(group))
 153                 return NULL;
 154
 155         event = fsnotify_peek_notify_event(group);
 156
 157         event_size += roundup(event->name_len, event_size);
 158
 159         if (event_size > count)
 160                 return ERR_PTR(-EINVAL);
 161
 162         /* held the notification_mutex the whole time, so this is the
 163          * same event we peeked above */
 164         fsnotify_remove_notify_event(group);
 165
 166         return event;
 167 }
 168
 169 /*
 170  * Copy an event to user space, returning how much we copied.
 171  *
 172  * We already checked that the event size is smaller than the
 173  * buffer we had in "get_one_event()" above.
 174  */
 175 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 176                                   struct fsnotify_event *event,
 177                                   char __user *buf)
 178 {
 179         struct inotify_event inotify_event;
 180         struct fsnotify_event_private_data *fsn_priv;
 181         struct inotify_event_private_data *priv;
 182         size_t event_size = sizeof(struct inotify_event);
 183         size_t name_len;
 184
 185         /* we get the inotify watch descriptor from the event private data */
 186         spin_lock(&event->lock);
 187         fsn_priv = fsnotify_remove_priv_from_event(group, event);
 188         spin_unlock(&event->lock);
 189
 190         if (!fsn_priv)
 191                 inotify_event.wd = -1;
 192         else {
 193                 priv = container_of(fsn_priv, struct inotify_event_private_data,
 194                                     fsnotify_event_priv_data);
 195                 inotify_event.wd = priv->wd;
 196                 inotify_free_event_priv(fsn_priv);
 197         }
 198
 199         /* round up event->name_len so it is a multiple of event_size
 200          * plus an extra byte for the terminating '\0'.
 201          */
 202         name_len = roundup(event->name_len + 1, event_size);
 203         inotify_event.len = name_len;
 204
 205         inotify_event.mask = inotify_mask_to_arg(event->mask);
 206         inotify_event.cookie = event->sync_cookie;
 207
 208         /* send the main event */
 209         if (copy_to_user(buf, &inotify_event, event_size))
 210                 return -EFAULT;
 211
 212         buf += event_size;
 213
 214         /*
 215          * fsnotify only stores the pathname, so here we have to send the pathname
 216          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 217          * with zeros.  I get my zeros from the nul_inotify_event.
 218          */
 219         if (name_len) {
 220                 unsigned int len_to_zero = name_len - event->name_len;
 221                 /* copy the path name */
 222                 if (copy_to_user(buf, event->file_name, event->name_len))
 223                         return -EFAULT;
 224                 buf += event->name_len;
 225
 226                 /* fill userspace with 0's */
 227                 if (clear_user(buf, len_to_zero))
 228                         return -EFAULT;
 229                 buf += len_to_zero;
 230                 event_size += name_len;
 231         }
 232
 233         return event_size;
 234 }
 235
 236 static ssize_t inotify_read(struct file *file, char __user *buf,
 237                             size_t count, loff_t *pos)
 238 {
 239         struct fsnotify_group *group;
 240         struct fsnotify_event *kevent;
 241         char __user *start;
 242         int ret;
 243         DEFINE_WAIT(wait);
 244
 245         start = buf;
 246         group = file->private_data;
 247
 248         while (1) {
 249                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 250
 251                 mutex_lock(&group->notification_mutex);
 252                 kevent = get_one_event(group, count);
 253                 mutex_unlock(&group->notification_mutex);
 254
 255                 if (kevent) {
 256                         ret = PTR_ERR(kevent);
 257                         if (IS_ERR(kevent))
 258                                 break;
 259                         ret = copy_event_to_user(group, kevent, buf);
 260                         fsnotify_put_event(kevent);
 261                         if (ret < 0)
 262                                 break;
 263                         buf += ret;
 264                         count -= ret;
 265                         continue;
 266                 }
 267
 268                 ret = -EAGAIN;
 269                 if (file->f_flags & O_NONBLOCK)
 270                         break;
 271                 ret = -EINTR;
 272                 if (signal_pending(current))
 273                         break;
 274
 275                 if (start != buf)
 276                         break;
 277
 278                 schedule();
 279         }
 280
 281         finish_wait(&group->notification_waitq, &wait);
 282         if (start != buf && ret != -EFAULT)
 283                 ret = buf - start;
 284         return ret;
 285 }
 286
 287 static int inotify_fasync(int fd, struct file *file, int on)
 288 {
 289         struct fsnotify_group *group = file->private_data;
 290
 291         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 292 }
 293
 294 static int inotify_release(struct inode *ignored, struct file *file)
 295 {
 296         struct fsnotify_group *group = file->private_data;
 297         struct user_struct *user = group->inotify_data.user;
 298
 299         fsnotify_clear_marks_by_group(group);
 300
 301         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 302         fsnotify_put_group(group);
 303
 304         atomic_dec(&user->inotify_devs);
 305
 306         return 0;
 307 }
 308
 309 static long inotify_ioctl(struct file *file, unsigned int cmd,
 310                           unsigned long arg)
 311 {
 312         struct fsnotify_group *group;
 313         struct fsnotify_event_holder *holder;
 314         struct fsnotify_event *event;
 315         void __user *p;
 316         int ret = -ENOTTY;
 317         size_t send_len = 0;
 318
 319         group = file->private_data;
 320         p = (void __user *) arg;
 321
 322         switch (cmd) {
 323         case FIONREAD:
 324                 mutex_lock(&group->notification_mutex);
 325                 list_for_each_entry(holder, &group->notification_list, event_list) {
 326                         event = holder->event;
 327                         send_len += sizeof(struct inotify_event);
 328                         send_len += roundup(event->name_len,
 329                                              sizeof(struct inotify_event));
 330                 }
 331                 mutex_unlock(&group->notification_mutex);
 332                 ret = put_user(send_len, (int __user *) p);
 333                 break;
 334         }
 335
 336         return ret;
 337 }
 338
 339 static const struct file_operations inotify_fops = {
 340         .poll           = inotify_poll,
 341         .read           = inotify_read,
 342         .fasync         = inotify_fasync,
 343         .release        = inotify_release,
 344         .unlocked_ioctl = inotify_ioctl,
 345         .compat_ioctl   = inotify_ioctl,
 346 };
 347
 348
 349 /*
 350  * find_inode - resolve a user-given path to a specific inode
 351  */
 352 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 353 {
 354         int error;
 355
 356         error = user_path_at(AT_FDCWD, dirname, flags, path);
 357         if (error)
 358                 return error;
 359         /* you can only watch an inode if you have read permissions on it */
 360         error = inode_permission(path->dentry->d_inode, MAY_READ);
 361         if (error)
 362                 path_put(path);
 363         return error;
 364 }
 365
 366 /*
 367  * Remove the mark from the idr (if present) and drop the reference
 368  * on the mark because it was in the idr.
 369  */
 370 static void inotify_remove_from_idr(struct fsnotify_group *group,
 371                                     struct inotify_inode_mark_entry *ientry)
 372 {
 373         struct idr *idr;
 374         struct fsnotify_mark_entry *entry;
 375         struct inotify_inode_mark_entry *found_ientry;
 376         int wd;
 377
 378         spin_lock(&group->inotify_data.idr_lock);
 379         idr = &group->inotify_data.idr;
 380         wd = ientry->wd;
 381
 382         if (wd == -1)
 383                 goto out;
 384
 385         entry = idr_find(&group->inotify_data.idr, wd);
 386         if (unlikely(!entry))
 387                 goto out;
 388
 389         found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 390         if (unlikely(found_ientry != ientry)) {
 391                 /* We found an entry in the idr with the right wd, but it's
 392                  * not the entry we were told to remove.  eparis seriously
 393                  * fucked up somewhere. */
 394                 WARN_ON(1);
 395                 ientry->wd = -1;
 396                 goto out;
 397         }
 398
 399         /* One ref for being in the idr, one ref held by the caller */
 400         BUG_ON(atomic_read(&entry->refcnt) < 2);
 401
 402         idr_remove(idr, wd);
 403         ientry->wd = -1;
 404
 405         /* removed from the idr, drop that ref */
 406         fsnotify_put_mark(entry);
 407 out:
 408         spin_unlock(&group->inotify_data.idr_lock);
 409 }
 410
 411 /*
 412  * Send IN_IGNORED for this wd, remove this wd from the idr.
 413  */
 414 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 415                                     struct fsnotify_group *group)
 416 {
 417         struct inotify_inode_mark_entry *ientry;
 418         struct fsnotify_event *ignored_event;
 419         struct inotify_event_private_data *event_priv;
 420         struct fsnotify_event_private_data *fsn_event_priv;
 421         int ret;
 422
 423         ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 424                                               FSNOTIFY_EVENT_NONE, NULL, 0,
 425                                               GFP_NOFS);
 426         if (!ignored_event)
 427                 return;
 428
 429         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 430
 431         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 432         if (unlikely(!event_priv))
 433                 goto skip_send_ignore;
 434
 435         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 436
 437         fsn_event_priv->group = group;
 438         event_priv->wd = ientry->wd;
 439
 440         ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
 441         if (ret)
 442                 inotify_free_event_priv(fsn_event_priv);
 443
 444 skip_send_ignore:
 445
 446         /* matches the reference taken when the event was created */
 447         fsnotify_put_event(ignored_event);
 448
 449         /* remove this entry from the idr */
 450         inotify_remove_from_idr(group, ientry);
 451
 452         atomic_dec(&group->inotify_data.user->inotify_watches);
 453 }
 454
 455 /* ding dong the mark is dead */
 456 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 457 {
 458         struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
 459
 460         kmem_cache_free(inotify_inode_mark_cachep, ientry);
 461 }
 462
 463 static int inotify_update_existing_watch(struct fsnotify_group *group,
 464                                          struct inode *inode,
 465                                          u32 arg)
 466 {
 467         struct fsnotify_mark_entry *entry;
 468         struct inotify_inode_mark_entry *ientry;
 469         __u32 old_mask, new_mask;
 470         __u32 mask;
 471         int add = (arg & IN_MASK_ADD);
 472         int ret;
 473
 474         /* don't allow invalid bits: we don't want flags set */
 475         mask = inotify_arg_to_mask(arg);
 476         if (unlikely(!mask))
 477                 return -EINVAL;
 478
 479         spin_lock(&inode->i_lock);
 480         entry = fsnotify_find_mark_entry(group, inode);
 481         spin_unlock(&inode->i_lock);
 482         if (!entry)
 483                 return -ENOENT;
 484
 485         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 486
 487         spin_lock(&entry->lock);
 488
 489         old_mask = entry->mask;
 490         if (add) {
 491                 entry->mask |= mask;
 492                 new_mask = entry->mask;
 493         } else {
 494                 entry->mask = mask;
 495                 new_mask = entry->mask;
 496         }
 497
 498         spin_unlock(&entry->lock);
 499
 500         if (old_mask != new_mask) {
 501                 /* more bits in old than in new? */
 502                 int dropped = (old_mask & ~new_mask);
 503                 /* more bits in this entry than the inode's mask? */
 504                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 505                 /* more bits in this entry than the group? */
 506                 int do_group = (new_mask & ~group->mask);
 507
 508                 /* update the inode with this new entry */
 509                 if (dropped || do_inode)
 510                         fsnotify_recalc_inode_mask(inode);
 511
 512                 /* update the group mask with the new mask */
 513                 if (dropped || do_group)
 514                         fsnotify_recalc_group_mask(group);
 515         }
 516
 517         /* return the wd */
 518         ret = ientry->wd;
 519
 520         /* match the get from fsnotify_find_mark_entry() */
 521         fsnotify_put_mark(entry);
 522
 523         return ret;
 524 }
 525
 526 static int inotify_new_watch(struct fsnotify_group *group,
 527                              struct inode *inode,
 528                              u32 arg)
 529 {
 530         struct inotify_inode_mark_entry *tmp_ientry;
 531         __u32 mask;
 532         int ret;
 533
 534         /* don't allow invalid bits: we don't want flags set */
 535         mask = inotify_arg_to_mask(arg);
 536         if (unlikely(!mask))
 537                 return -EINVAL;
 538
 539         tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 540         if (unlikely(!tmp_ientry))
 541                 return -ENOMEM;
 542
 543         fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
 544         tmp_ientry->fsn_entry.mask = mask;
 545         tmp_ientry->wd = -1;
 546
 547         ret = -ENOSPC;
 548         if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 549                 goto out_err;
 550 retry:
 551         ret = -ENOMEM;
 552         if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
 553                 goto out_err;
 554
 555         spin_lock(&group->inotify_data.idr_lock);
 556         ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
 557                                 group->inotify_data.last_wd,
 558                                 &tmp_ientry->wd);
 559         spin_unlock(&group->inotify_data.idr_lock);
 560         if (ret) {
 561                 /* idr was out of memory allocate and try again */
 562                 if (ret == -EAGAIN)
 563                         goto retry;
 564                 goto out_err;
 565         }
 566
 567         /* we put the mark on the idr, take a reference */
 568         fsnotify_get_mark(&tmp_ientry->fsn_entry);
 569
 570         /* we are on the idr, now get on the inode */
 571         ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 572         if (ret) {
 573                 /* we failed to get on the inode, get off the idr */
 574                 inotify_remove_from_idr(group, tmp_ientry);
 575                 goto out_err;
 576         }
 577
 578         /* update the idr hint, who cares about races, it's just a hint */
 579         group->inotify_data.last_wd = tmp_ientry->wd;
 580
 581         /* increment the number of watches the user has */
 582         atomic_inc(&group->inotify_data.user->inotify_watches);
 583
 584         /* return the watch descriptor for this new entry */
 585         ret = tmp_ientry->wd;
 586
 587         /* match the ref from fsnotify_init_markentry() */
 588         fsnotify_put_mark(&tmp_ientry->fsn_entry);
 589
 590 out_err:
 591         if (ret < 0)
 592                 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 593
 594         return ret;
 595 }
 596
 597 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 598 {
 599         int ret = 0;
 600
 601 retry:
 602         /* try to update and existing watch with the new arg */
 603         ret = inotify_update_existing_watch(group, inode, arg);
 604         /* no mark present, try to add a new one */
 605         if (ret == -ENOENT)
 606                 ret = inotify_new_watch(group, inode, arg);
 607         /*
 608          * inotify_new_watch could race with another thread which did an
 609          * inotify_new_watch between the update_existing and the add watch
 610          * here, go back and try to update an existing mark again.
 611          */
 612         if (ret == -EEXIST)
 613                 goto retry;
 614
 615         return ret;
 616 }
 617
 618 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 619 {
 620         struct fsnotify_group *group;
 621         unsigned int grp_num;
 622
 623         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 624         grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
 625         group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
 626         if (IS_ERR(group))
 627                 return group;
 628
 629         group->max_events = max_events;
 630
 631         spin_lock_init(&group->inotify_data.idr_lock);
 632         idr_init(&group->inotify_data.idr);
 633         group->inotify_data.last_wd = 1;
 634         group->inotify_data.user = user;
 635         group->inotify_data.fa = NULL;
 636
 637         return group;
 638 }
 639
 640
 641 /* inotify syscalls */
 642 SYSCALL_DEFINE1(inotify_init1, int, flags)
 643 {
 644         struct fsnotify_group *group;
 645         struct user_struct *user;
 646         struct file *filp;
 647         int fd, ret;
 648
 649         /* Check the IN_* constants for consistency.  */
 650         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 651         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 652
 653         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 654                 return -EINVAL;
 655
 656         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 657         if (fd < 0)
 658                 return fd;
 659
 660         filp = get_empty_filp();
 661         if (!filp) {
 662                 ret = -ENFILE;
 663                 goto out_put_fd;
 664         }
 665
 666         user = get_current_user();
 667         if (unlikely(atomic_read(&user->inotify_devs) >=
 668                         inotify_max_user_instances)) {
 669                 ret = -EMFILE;
 670                 goto out_free_uid;
 671         }
 672
 673         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 674         group = inotify_new_group(user, inotify_max_queued_events);
 675         if (IS_ERR(group)) {
 676                 ret = PTR_ERR(group);
 677                 goto out_free_uid;
 678         }
 679
 680         filp->f_op = &inotify_fops;
 681         filp->f_path.mnt = mntget(inotify_mnt);
 682         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 683         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 684         filp->f_mode = FMODE_READ;
 685         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 686         filp->private_data = group;
 687
 688         atomic_inc(&user->inotify_devs);
 689
 690         fd_install(fd, filp);
 691
 692         return fd;
 693
 694 out_free_uid:
 695         free_uid(user);
 696         put_filp(filp);
 697 out_put_fd:
 698         put_unused_fd(fd);
 699         return ret;
 700 }
 701
 702 SYSCALL_DEFINE0(inotify_init)
 703 {
 704         return sys_inotify_init1(0);
 705 }
 706
 707 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 708                 u32, mask)
 709 {
 710         struct fsnotify_group *group;
 711         struct inode *inode;
 712         struct path path;
 713         struct file *filp;
 714         int ret, fput_needed;
 715         unsigned flags = 0;
 716
 717         filp = fget_light(fd, &fput_needed);
 718         if (unlikely(!filp))
 719                 return -EBADF;
 720
 721         /* verify that this is indeed an inotify instance */
 722         if (unlikely(filp->f_op != &inotify_fops)) {
 723                 ret = -EINVAL;
 724                 goto fput_and_out;
 725         }
 726
 727         if (!(mask & IN_DONT_FOLLOW))
 728                 flags |= LOOKUP_FOLLOW;
 729         if (mask & IN_ONLYDIR)
 730                 flags |= LOOKUP_DIRECTORY;
 731
 732         ret = inotify_find_inode(pathname, &path, flags);
 733         if (ret)
 734                 goto fput_and_out;
 735
 736         /* inode held in place by reference to path; group by fget on fd */
 737         inode = path.dentry->d_inode;
 738         group = filp->private_data;
 739
 740         /* create/update an inode mark */
 741         ret = inotify_update_watch(group, inode, mask);
 742         if (unlikely(ret))
 743                 goto path_put_and_out;
 744
 745 path_put_and_out:
 746         path_put(&path);
 747 fput_and_out:
 748         fput_light(filp, fput_needed);
 749         return ret;
 750 }
 751
 752 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 753 {
 754         struct fsnotify_group *group;
 755         struct fsnotify_mark_entry *entry;
 756         struct file *filp;
 757         int ret = 0, fput_needed;
 758
 759         filp = fget_light(fd, &fput_needed);
 760         if (unlikely(!filp))
 761                 return -EBADF;
 762
 763         /* verify that this is indeed an inotify instance */
 764         if (unlikely(filp->f_op != &inotify_fops)) {
 765                 ret = -EINVAL;
 766                 goto out;
 767         }
 768
 769         group = filp->private_data;
 770
 771         spin_lock(&group->inotify_data.idr_lock);
 772         entry = idr_find(&group->inotify_data.idr, wd);
 773         if (unlikely(!entry)) {
 774                 spin_unlock(&group->inotify_data.idr_lock);
 775                 ret = -EINVAL;
 776                 goto out;
 777         }
 778         fsnotify_get_mark(entry);
 779         spin_unlock(&group->inotify_data.idr_lock);
 780
 781         fsnotify_destroy_mark_by_entry(entry);
 782         fsnotify_put_mark(entry);
 783
 784 out:
 785         fput_light(filp, fput_needed);
 786         return ret;
 787 }
 788
 789 static int
 790 inotify_get_sb(struct file_system_type *fs_type, int flags,
 791                const char *dev_name, void *data, struct vfsmount *mnt)
 792 {
 793         return get_sb_pseudo(fs_type, "inotify", NULL,
 794                         INOTIFYFS_SUPER_MAGIC, mnt);
 795 }
 796
 797 static struct file_system_type inotify_fs_type = {
 798     .name       = "inotifyfs",
 799     .get_sb     = inotify_get_sb,
 800     .kill_sb    = kill_anon_super,
 801 };
 802
 803 /*
 804  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 805  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 806  * must result in panic().
 807  */
 808 static int __init inotify_user_setup(void)
 809 {
 810         int ret;
 811
 812         ret = register_filesystem(&inotify_fs_type);
 813         if (unlikely(ret))
 814                 panic("inotify: register_filesystem returned %d!\n", ret);
 815
 816         inotify_mnt = kern_mount(&inotify_fs_type);
 817         if (IS_ERR(inotify_mnt))
 818                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 819
 820         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
 821         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 822
 823         inotify_max_queued_events = 16384;
 824         inotify_max_user_instances = 128;
 825         inotify_max_user_watches = 8192;
 826
 827         return 0;
 828 }
 829 module_init(inotify_user_setup);