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         if (event->name_len)
 158                 event_size += roundup(event->name_len + 1, event_size);
 159
 160         if (event_size > count)
 161                 return ERR_PTR(-EINVAL);
 162
 163         /* held the notification_mutex the whole time, so this is the
 164          * same event we peeked above */
 165         fsnotify_remove_notify_event(group);
 166
 167         return event;
 168 }
 169
 170 /*
 171  * Copy an event to user space, returning how much we copied.
 172  *
 173  * We already checked that the event size is smaller than the
 174  * buffer we had in "get_one_event()" above.
 175  */
 176 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 177                                   struct fsnotify_event *event,
 178                                   char __user *buf)
 179 {
 180         struct inotify_event inotify_event;
 181         struct fsnotify_event_private_data *fsn_priv;
 182         struct inotify_event_private_data *priv;
 183         size_t event_size = sizeof(struct inotify_event);
 184         size_t name_len = 0;
 185
 186         /* we get the inotify watch descriptor from the event private data */
 187         spin_lock(&event->lock);
 188         fsn_priv = fsnotify_remove_priv_from_event(group, event);
 189         spin_unlock(&event->lock);
 190
 191         if (!fsn_priv)
 192                 inotify_event.wd = -1;
 193         else {
 194                 priv = container_of(fsn_priv, struct inotify_event_private_data,
 195                                     fsnotify_event_priv_data);
 196                 inotify_event.wd = priv->wd;
 197                 inotify_free_event_priv(fsn_priv);
 198         }
 199
 200         /*
 201          * round up event->name_len so it is a multiple of event_size
 202          * plus an extra byte for the terminating '\0'.
 203          */
 204         if (event->name_len)
 205                 name_len = roundup(event->name_len + 1, event_size);
 206         inotify_event.len = name_len;
 207
 208         inotify_event.mask = inotify_mask_to_arg(event->mask);
 209         inotify_event.cookie = event->sync_cookie;
 210
 211         /* send the main event */
 212         if (copy_to_user(buf, &inotify_event, event_size))
 213                 return -EFAULT;
 214
 215         buf += event_size;
 216
 217         /*
 218          * fsnotify only stores the pathname, so here we have to send the pathname
 219          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 220          * with zeros.  I get my zeros from the nul_inotify_event.
 221          */
 222         if (name_len) {
 223                 unsigned int len_to_zero = name_len - event->name_len;
 224                 /* copy the path name */
 225                 if (copy_to_user(buf, event->file_name, event->name_len))
 226                         return -EFAULT;
 227                 buf += event->name_len;
 228
 229                 /* fill userspace with 0's */
 230                 if (clear_user(buf, len_to_zero))
 231                         return -EFAULT;
 232                 buf += len_to_zero;
 233                 event_size += name_len;
 234         }
 235
 236         return event_size;
 237 }
 238
 239 static ssize_t inotify_read(struct file *file, char __user *buf,
 240                             size_t count, loff_t *pos)
 241 {
 242         struct fsnotify_group *group;
 243         struct fsnotify_event *kevent;
 244         char __user *start;
 245         int ret;
 246         DEFINE_WAIT(wait);
 247
 248         start = buf;
 249         group = file->private_data;
 250
 251         while (1) {
 252                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 253
 254                 mutex_lock(&group->notification_mutex);
 255                 kevent = get_one_event(group, count);
 256                 mutex_unlock(&group->notification_mutex);
 257
 258                 if (kevent) {
 259                         ret = PTR_ERR(kevent);
 260                         if (IS_ERR(kevent))
 261                                 break;
 262                         ret = copy_event_to_user(group, kevent, buf);
 263                         fsnotify_put_event(kevent);
 264                         if (ret < 0)
 265                                 break;
 266                         buf += ret;
 267                         count -= ret;
 268                         continue;
 269                 }
 270
 271                 ret = -EAGAIN;
 272                 if (file->f_flags & O_NONBLOCK)
 273                         break;
 274                 ret = -EINTR;
 275                 if (signal_pending(current))
 276                         break;
 277
 278                 if (start != buf)
 279                         break;
 280
 281                 schedule();
 282         }
 283
 284         finish_wait(&group->notification_waitq, &wait);
 285         if (start != buf && ret != -EFAULT)
 286                 ret = buf - start;
 287         return ret;
 288 }
 289
 290 static int inotify_fasync(int fd, struct file *file, int on)
 291 {
 292         struct fsnotify_group *group = file->private_data;
 293
 294         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 295 }
 296
 297 static int inotify_release(struct inode *ignored, struct file *file)
 298 {
 299         struct fsnotify_group *group = file->private_data;
 300         struct user_struct *user = group->inotify_data.user;
 301
 302         fsnotify_clear_marks_by_group(group);
 303
 304         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 305         fsnotify_put_group(group);
 306
 307         atomic_dec(&user->inotify_devs);
 308
 309         return 0;
 310 }
 311
 312 static long inotify_ioctl(struct file *file, unsigned int cmd,
 313                           unsigned long arg)
 314 {
 315         struct fsnotify_group *group;
 316         struct fsnotify_event_holder *holder;
 317         struct fsnotify_event *event;
 318         void __user *p;
 319         int ret = -ENOTTY;
 320         size_t send_len = 0;
 321
 322         group = file->private_data;
 323         p = (void __user *) arg;
 324
 325         switch (cmd) {
 326         case FIONREAD:
 327                 mutex_lock(&group->notification_mutex);
 328                 list_for_each_entry(holder, &group->notification_list, event_list) {
 329                         event = holder->event;
 330                         send_len += sizeof(struct inotify_event);
 331                         if (event->name_len)
 332                                 send_len += roundup(event->name_len + 1,
 333                                                 sizeof(struct inotify_event));
 334                 }
 335                 mutex_unlock(&group->notification_mutex);
 336                 ret = put_user(send_len, (int __user *) p);
 337                 break;
 338         }
 339
 340         return ret;
 341 }
 342
 343 static const struct file_operations inotify_fops = {
 344         .poll           = inotify_poll,
 345         .read           = inotify_read,
 346         .fasync         = inotify_fasync,
 347         .release        = inotify_release,
 348         .unlocked_ioctl = inotify_ioctl,
 349         .compat_ioctl   = inotify_ioctl,
 350 };
 351
 352
 353 /*
 354  * find_inode - resolve a user-given path to a specific inode
 355  */
 356 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 357 {
 358         int error;
 359
 360         error = user_path_at(AT_FDCWD, dirname, flags, path);
 361         if (error)
 362                 return error;
 363         /* you can only watch an inode if you have read permissions on it */
 364         error = inode_permission(path->dentry->d_inode, MAY_READ);
 365         if (error)
 366                 path_put(path);
 367         return error;
 368 }
 369
 370 /*
 371  * Remove the mark from the idr (if present) and drop the reference
 372  * on the mark because it was in the idr.
 373  */
 374 static void inotify_remove_from_idr(struct fsnotify_group *group,
 375                                     struct inotify_inode_mark_entry *ientry)
 376 {
 377         struct idr *idr;
 378         struct fsnotify_mark_entry *entry;
 379         struct inotify_inode_mark_entry *found_ientry;
 380         int wd;
 381
 382         spin_lock(&group->inotify_data.idr_lock);
 383         idr = &group->inotify_data.idr;
 384         wd = ientry->wd;
 385
 386         if (wd == -1)
 387                 goto out;
 388
 389         entry = idr_find(&group->inotify_data.idr, wd);
 390         if (unlikely(!entry))
 391                 goto out;
 392
 393         found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 394         if (unlikely(found_ientry != ientry)) {
 395                 /* We found an entry in the idr with the right wd, but it's
 396                  * not the entry we were told to remove.  eparis seriously
 397                  * fucked up somewhere. */
 398                 WARN_ON(1);
 399                 ientry->wd = -1;
 400                 goto out;
 401         }
 402
 403         /* One ref for being in the idr, one ref held by the caller */
 404         BUG_ON(atomic_read(&entry->refcnt) < 2);
 405
 406         idr_remove(idr, wd);
 407         ientry->wd = -1;
 408
 409         /* removed from the idr, drop that ref */
 410         fsnotify_put_mark(entry);
 411 out:
 412         spin_unlock(&group->inotify_data.idr_lock);
 413 }
 414
 415 /*
 416  * Send IN_IGNORED for this wd, remove this wd from the idr.
 417  */
 418 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 419                                     struct fsnotify_group *group)
 420 {
 421         struct inotify_inode_mark_entry *ientry;
 422         struct fsnotify_event *ignored_event;
 423         struct inotify_event_private_data *event_priv;
 424         struct fsnotify_event_private_data *fsn_event_priv;
 425         int ret;
 426
 427         ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 428                                               FSNOTIFY_EVENT_NONE, NULL, 0,
 429                                               GFP_NOFS);
 430         if (!ignored_event)
 431                 return;
 432
 433         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 434
 435         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 436         if (unlikely(!event_priv))
 437                 goto skip_send_ignore;
 438
 439         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 440
 441         fsn_event_priv->group = group;
 442         event_priv->wd = ientry->wd;
 443
 444         ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
 445         if (ret)
 446                 inotify_free_event_priv(fsn_event_priv);
 447
 448 skip_send_ignore:
 449
 450         /* matches the reference taken when the event was created */
 451         fsnotify_put_event(ignored_event);
 452
 453         /* remove this entry from the idr */
 454         inotify_remove_from_idr(group, ientry);
 455
 456         atomic_dec(&group->inotify_data.user->inotify_watches);
 457 }
 458
 459 /* ding dong the mark is dead */
 460 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 461 {
 462         struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
 463
 464         kmem_cache_free(inotify_inode_mark_cachep, ientry);
 465 }
 466
 467 static int inotify_update_existing_watch(struct fsnotify_group *group,
 468                                          struct inode *inode,
 469                                          u32 arg)
 470 {
 471         struct fsnotify_mark_entry *entry;
 472         struct inotify_inode_mark_entry *ientry;
 473         __u32 old_mask, new_mask;
 474         __u32 mask;
 475         int add = (arg & IN_MASK_ADD);
 476         int ret;
 477
 478         /* don't allow invalid bits: we don't want flags set */
 479         mask = inotify_arg_to_mask(arg);
 480         if (unlikely(!mask))
 481                 return -EINVAL;
 482
 483         spin_lock(&inode->i_lock);
 484         entry = fsnotify_find_mark_entry(group, inode);
 485         spin_unlock(&inode->i_lock);
 486         if (!entry)
 487                 return -ENOENT;
 488
 489         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 490
 491         spin_lock(&entry->lock);
 492
 493         old_mask = entry->mask;
 494         if (add) {
 495                 entry->mask |= mask;
 496                 new_mask = entry->mask;
 497         } else {
 498                 entry->mask = mask;
 499                 new_mask = entry->mask;
 500         }
 501
 502         spin_unlock(&entry->lock);
 503
 504         if (old_mask != new_mask) {
 505                 /* more bits in old than in new? */
 506                 int dropped = (old_mask & ~new_mask);
 507                 /* more bits in this entry than the inode's mask? */
 508                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 509                 /* more bits in this entry than the group? */
 510                 int do_group = (new_mask & ~group->mask);
 511
 512                 /* update the inode with this new entry */
 513                 if (dropped || do_inode)
 514                         fsnotify_recalc_inode_mask(inode);
 515
 516                 /* update the group mask with the new mask */
 517                 if (dropped || do_group)
 518                         fsnotify_recalc_group_mask(group);
 519         }
 520
 521         /* return the wd */
 522         ret = ientry->wd;
 523
 524         /* match the get from fsnotify_find_mark_entry() */
 525         fsnotify_put_mark(entry);
 526
 527         return ret;
 528 }
 529
 530 static int inotify_new_watch(struct fsnotify_group *group,
 531                              struct inode *inode,
 532                              u32 arg)
 533 {
 534         struct inotify_inode_mark_entry *tmp_ientry;
 535         __u32 mask;
 536         int ret;
 537
 538         /* don't allow invalid bits: we don't want flags set */
 539         mask = inotify_arg_to_mask(arg);
 540         if (unlikely(!mask))
 541                 return -EINVAL;
 542
 543         tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 544         if (unlikely(!tmp_ientry))
 545                 return -ENOMEM;
 546
 547         fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
 548         tmp_ientry->fsn_entry.mask = mask;
 549         tmp_ientry->wd = -1;
 550
 551         ret = -ENOSPC;
 552         if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 553                 goto out_err;
 554 retry:
 555         ret = -ENOMEM;
 556         if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
 557                 goto out_err;
 558
 559         /* we are putting the mark on the idr, take a reference */
 560         fsnotify_get_mark(&tmp_ientry->fsn_entry);
 561
 562         spin_lock(&group->inotify_data.idr_lock);
 563         ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
 564                                 group->inotify_data.last_wd+1,
 565                                 &tmp_ientry->wd);
 566         spin_unlock(&group->inotify_data.idr_lock);
 567         if (ret) {
 568                 /* we didn't get on the idr, drop the idr reference */
 569                 fsnotify_put_mark(&tmp_ientry->fsn_entry);
 570
 571                 /* idr was out of memory allocate and try again */
 572                 if (ret == -EAGAIN)
 573                         goto retry;
 574                 goto out_err;
 575         }
 576
 577         /* we are on the idr, now get on the inode */
 578         ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 579         if (ret) {
 580                 /* we failed to get on the inode, get off the idr */
 581                 inotify_remove_from_idr(group, tmp_ientry);
 582                 goto out_err;
 583         }
 584
 585         /* update the idr hint, who cares about races, it's just a hint */
 586         group->inotify_data.last_wd = tmp_ientry->wd;
 587
 588         /* increment the number of watches the user has */
 589         atomic_inc(&group->inotify_data.user->inotify_watches);
 590
 591         /* return the watch descriptor for this new entry */
 592         ret = tmp_ientry->wd;
 593
 594         /* match the ref from fsnotify_init_markentry() */
 595         fsnotify_put_mark(&tmp_ientry->fsn_entry);
 596
 597         /* if this mark added a new event update the group mask */
 598         if (mask & ~group->mask)
 599                 fsnotify_recalc_group_mask(group);
 600
 601 out_err:
 602         if (ret < 0)
 603                 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 604
 605         return ret;
 606 }
 607
 608 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 609 {
 610         int ret = 0;
 611
 612 retry:
 613         /* try to update and existing watch with the new arg */
 614         ret = inotify_update_existing_watch(group, inode, arg);
 615         /* no mark present, try to add a new one */
 616         if (ret == -ENOENT)
 617                 ret = inotify_new_watch(group, inode, arg);
 618         /*
 619          * inotify_new_watch could race with another thread which did an
 620          * inotify_new_watch between the update_existing and the add watch
 621          * here, go back and try to update an existing mark again.
 622          */
 623         if (ret == -EEXIST)
 624                 goto retry;
 625
 626         return ret;
 627 }
 628
 629 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 630 {
 631         struct fsnotify_group *group;
 632         unsigned int grp_num;
 633
 634         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 635         grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
 636         group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
 637         if (IS_ERR(group))
 638                 return group;
 639
 640         group->max_events = max_events;
 641
 642         spin_lock_init(&group->inotify_data.idr_lock);
 643         idr_init(&group->inotify_data.idr);
 644         group->inotify_data.last_wd = 0;
 645         group->inotify_data.user = user;
 646         group->inotify_data.fa = NULL;
 647
 648         return group;
 649 }
 650
 651
 652 /* inotify syscalls */
 653 SYSCALL_DEFINE1(inotify_init1, int, flags)
 654 {
 655         struct fsnotify_group *group;
 656         struct user_struct *user;
 657         struct file *filp;
 658         int fd, ret;
 659
 660         /* Check the IN_* constants for consistency.  */
 661         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 662         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 663
 664         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 665                 return -EINVAL;
 666
 667         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 668         if (fd < 0)
 669                 return fd;
 670
 671         filp = get_empty_filp();
 672         if (!filp) {
 673                 ret = -ENFILE;
 674                 goto out_put_fd;
 675         }
 676
 677         user = get_current_user();
 678         if (unlikely(atomic_read(&user->inotify_devs) >=
 679                         inotify_max_user_instances)) {
 680                 ret = -EMFILE;
 681                 goto out_free_uid;
 682         }
 683
 684         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 685         group = inotify_new_group(user, inotify_max_queued_events);
 686         if (IS_ERR(group)) {
 687                 ret = PTR_ERR(group);
 688                 goto out_free_uid;
 689         }
 690
 691         filp->f_op = &inotify_fops;
 692         filp->f_path.mnt = mntget(inotify_mnt);
 693         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 694         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 695         filp->f_mode = FMODE_READ;
 696         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 697         filp->private_data = group;
 698
 699         atomic_inc(&user->inotify_devs);
 700
 701         fd_install(fd, filp);
 702
 703         return fd;
 704
 705 out_free_uid:
 706         free_uid(user);
 707         put_filp(filp);
 708 out_put_fd:
 709         put_unused_fd(fd);
 710         return ret;
 711 }
 712
 713 SYSCALL_DEFINE0(inotify_init)
 714 {
 715         return sys_inotify_init1(0);
 716 }
 717
 718 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 719                 u32, mask)
 720 {
 721         struct fsnotify_group *group;
 722         struct inode *inode;
 723         struct path path;
 724         struct file *filp;
 725         int ret, fput_needed;
 726         unsigned flags = 0;
 727
 728         filp = fget_light(fd, &fput_needed);
 729         if (unlikely(!filp))
 730                 return -EBADF;
 731
 732         /* verify that this is indeed an inotify instance */
 733         if (unlikely(filp->f_op != &inotify_fops)) {
 734                 ret = -EINVAL;
 735                 goto fput_and_out;
 736         }
 737
 738         if (!(mask & IN_DONT_FOLLOW))
 739                 flags |= LOOKUP_FOLLOW;
 740         if (mask & IN_ONLYDIR)
 741                 flags |= LOOKUP_DIRECTORY;
 742
 743         ret = inotify_find_inode(pathname, &path, flags);
 744         if (ret)
 745                 goto fput_and_out;
 746
 747         /* inode held in place by reference to path; group by fget on fd */
 748         inode = path.dentry->d_inode;
 749         group = filp->private_data;
 750
 751         /* create/update an inode mark */
 752         ret = inotify_update_watch(group, inode, mask);
 753         if (unlikely(ret))
 754                 goto path_put_and_out;
 755
 756 path_put_and_out:
 757         path_put(&path);
 758 fput_and_out:
 759         fput_light(filp, fput_needed);
 760         return ret;
 761 }
 762
 763 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 764 {
 765         struct fsnotify_group *group;
 766         struct fsnotify_mark_entry *entry;
 767         struct file *filp;
 768         int ret = 0, fput_needed;
 769
 770         filp = fget_light(fd, &fput_needed);
 771         if (unlikely(!filp))
 772                 return -EBADF;
 773
 774         /* verify that this is indeed an inotify instance */
 775         if (unlikely(filp->f_op != &inotify_fops)) {
 776                 ret = -EINVAL;
 777                 goto out;
 778         }
 779
 780         group = filp->private_data;
 781
 782         spin_lock(&group->inotify_data.idr_lock);
 783         entry = idr_find(&group->inotify_data.idr, wd);
 784         if (unlikely(!entry)) {
 785                 spin_unlock(&group->inotify_data.idr_lock);
 786                 ret = -EINVAL;
 787                 goto out;
 788         }
 789         fsnotify_get_mark(entry);
 790         spin_unlock(&group->inotify_data.idr_lock);
 791
 792         fsnotify_destroy_mark_by_entry(entry);
 793         fsnotify_put_mark(entry);
 794
 795 out:
 796         fput_light(filp, fput_needed);
 797         return ret;
 798 }
 799
 800 static int
 801 inotify_get_sb(struct file_system_type *fs_type, int flags,
 802                const char *dev_name, void *data, struct vfsmount *mnt)
 803 {
 804         return get_sb_pseudo(fs_type, "inotify", NULL,
 805                         INOTIFYFS_SUPER_MAGIC, mnt);
 806 }
 807
 808 static struct file_system_type inotify_fs_type = {
 809     .name       = "inotifyfs",
 810     .get_sb     = inotify_get_sb,
 811     .kill_sb    = kill_anon_super,
 812 };
 813
 814 /*
 815  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 816  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 817  * must result in panic().
 818  */
 819 static int __init inotify_user_setup(void)
 820 {
 821         int ret;
 822
 823         ret = register_filesystem(&inotify_fs_type);
 824         if (unlikely(ret))
 825                 panic("inotify: register_filesystem returned %d!\n", ret);
 826
 827         inotify_mnt = kern_mount(&inotify_fs_type);
 828         if (IS_ERR(inotify_mnt))
 829                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 830
 831         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
 832         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 833
 834         inotify_max_queued_events = 16384;
 835         inotify_max_user_instances = 128;
 836         inotify_max_user_watches = 8192;
 837
 838         return 0;
 839 }
 840 module_init(inotify_user_setup);