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         /*
 110          * everything should accept their own ignored, cares about children,
 111          * and should receive events when the inode is unmounted
 112          */
 113         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
 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         if (event->name_len)
 161                 event_size += roundup(event->name_len + 1, 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 = 0;
 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         /*
 204          * round up event->name_len so it is a multiple of event_size
 205          * plus an extra byte for the terminating '\0'.
 206          */
 207         if (event->name_len)
 208                 name_len = roundup(event->name_len + 1, event_size);
 209         inotify_event.len = name_len;
 210
 211         inotify_event.mask = inotify_mask_to_arg(event->mask);
 212         inotify_event.cookie = event->sync_cookie;
 213
 214         /* send the main event */
 215         if (copy_to_user(buf, &inotify_event, event_size))
 216                 return -EFAULT;
 217
 218         buf += event_size;
 219
 220         /*
 221          * fsnotify only stores the pathname, so here we have to send the pathname
 222          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 223          * with zeros.  I get my zeros from the nul_inotify_event.
 224          */
 225         if (name_len) {
 226                 unsigned int len_to_zero = name_len - event->name_len;
 227                 /* copy the path name */
 228                 if (copy_to_user(buf, event->file_name, event->name_len))
 229                         return -EFAULT;
 230                 buf += event->name_len;
 231
 232                 /* fill userspace with 0's */
 233                 if (clear_user(buf, len_to_zero))
 234                         return -EFAULT;
 235                 buf += len_to_zero;
 236                 event_size += name_len;
 237         }
 238
 239         return event_size;
 240 }
 241
 242 static ssize_t inotify_read(struct file *file, char __user *buf,
 243                             size_t count, loff_t *pos)
 244 {
 245         struct fsnotify_group *group;
 246         struct fsnotify_event *kevent;
 247         char __user *start;
 248         int ret;
 249         DEFINE_WAIT(wait);
 250
 251         start = buf;
 252         group = file->private_data;
 253
 254         while (1) {
 255                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 256
 257                 mutex_lock(&group->notification_mutex);
 258                 kevent = get_one_event(group, count);
 259                 mutex_unlock(&group->notification_mutex);
 260
 261                 if (kevent) {
 262                         ret = PTR_ERR(kevent);
 263                         if (IS_ERR(kevent))
 264                                 break;
 265                         ret = copy_event_to_user(group, kevent, buf);
 266                         fsnotify_put_event(kevent);
 267                         if (ret < 0)
 268                                 break;
 269                         buf += ret;
 270                         count -= ret;
 271                         continue;
 272                 }
 273
 274                 ret = -EAGAIN;
 275                 if (file->f_flags & O_NONBLOCK)
 276                         break;
 277                 ret = -EINTR;
 278                 if (signal_pending(current))
 279                         break;
 280
 281                 if (start != buf)
 282                         break;
 283
 284                 schedule();
 285         }
 286
 287         finish_wait(&group->notification_waitq, &wait);
 288         if (start != buf && ret != -EFAULT)
 289                 ret = buf - start;
 290         return ret;
 291 }
 292
 293 static int inotify_fasync(int fd, struct file *file, int on)
 294 {
 295         struct fsnotify_group *group = file->private_data;
 296
 297         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 298 }
 299
 300 static int inotify_release(struct inode *ignored, struct file *file)
 301 {
 302         struct fsnotify_group *group = file->private_data;
 303         struct user_struct *user = group->inotify_data.user;
 304
 305         fsnotify_clear_marks_by_group(group);
 306
 307         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 308         fsnotify_put_group(group);
 309
 310         atomic_dec(&user->inotify_devs);
 311
 312         return 0;
 313 }
 314
 315 static long inotify_ioctl(struct file *file, unsigned int cmd,
 316                           unsigned long arg)
 317 {
 318         struct fsnotify_group *group;
 319         struct fsnotify_event_holder *holder;
 320         struct fsnotify_event *event;
 321         void __user *p;
 322         int ret = -ENOTTY;
 323         size_t send_len = 0;
 324
 325         group = file->private_data;
 326         p = (void __user *) arg;
 327
 328         switch (cmd) {
 329         case FIONREAD:
 330                 mutex_lock(&group->notification_mutex);
 331                 list_for_each_entry(holder, &group->notification_list, event_list) {
 332                         event = holder->event;
 333                         send_len += sizeof(struct inotify_event);
 334                         if (event->name_len)
 335                                 send_len += roundup(event->name_len + 1,
 336                                                 sizeof(struct inotify_event));
 337                 }
 338                 mutex_unlock(&group->notification_mutex);
 339                 ret = put_user(send_len, (int __user *) p);
 340                 break;
 341         }
 342
 343         return ret;
 344 }
 345
 346 static const struct file_operations inotify_fops = {
 347         .poll           = inotify_poll,
 348         .read           = inotify_read,
 349         .fasync         = inotify_fasync,
 350         .release        = inotify_release,
 351         .unlocked_ioctl = inotify_ioctl,
 352         .compat_ioctl   = inotify_ioctl,
 353 };
 354
 355
 356 /*
 357  * find_inode - resolve a user-given path to a specific inode
 358  */
 359 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 360 {
 361         int error;
 362
 363         error = user_path_at(AT_FDCWD, dirname, flags, path);
 364         if (error)
 365                 return error;
 366         /* you can only watch an inode if you have read permissions on it */
 367         error = inode_permission(path->dentry->d_inode, MAY_READ);
 368         if (error)
 369                 path_put(path);
 370         return error;
 371 }
 372
 373 /*
 374  * Remove the mark from the idr (if present) and drop the reference
 375  * on the mark because it was in the idr.
 376  */
 377 static void inotify_remove_from_idr(struct fsnotify_group *group,
 378                                     struct inotify_inode_mark_entry *ientry)
 379 {
 380         struct idr *idr;
 381         struct fsnotify_mark_entry *entry;
 382         struct inotify_inode_mark_entry *found_ientry;
 383         int wd;
 384
 385         spin_lock(&group->inotify_data.idr_lock);
 386         idr = &group->inotify_data.idr;
 387         wd = ientry->wd;
 388
 389         if (wd == -1)
 390                 goto out;
 391
 392         entry = idr_find(&group->inotify_data.idr, wd);
 393         if (unlikely(!entry))
 394                 goto out;
 395
 396         found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 397         if (unlikely(found_ientry != ientry)) {
 398                 /* We found an entry in the idr with the right wd, but it's
 399                  * not the entry we were told to remove.  eparis seriously
 400                  * fucked up somewhere. */
 401                 WARN_ON(1);
 402                 ientry->wd = -1;
 403                 goto out;
 404         }
 405
 406         /* One ref for being in the idr, one ref held by the caller */
 407         BUG_ON(atomic_read(&entry->refcnt) < 2);
 408
 409         idr_remove(idr, wd);
 410         ientry->wd = -1;
 411
 412         /* removed from the idr, drop that ref */
 413         fsnotify_put_mark(entry);
 414 out:
 415         spin_unlock(&group->inotify_data.idr_lock);
 416 }
 417
 418 /*
 419  * Send IN_IGNORED for this wd, remove this wd from the idr.
 420  */
 421 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 422                                     struct fsnotify_group *group)
 423 {
 424         struct inotify_inode_mark_entry *ientry;
 425         struct fsnotify_event *ignored_event;
 426         struct inotify_event_private_data *event_priv;
 427         struct fsnotify_event_private_data *fsn_event_priv;
 428         int ret;
 429
 430         ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 431                                               FSNOTIFY_EVENT_NONE, NULL, 0,
 432                                               GFP_NOFS);
 433         if (!ignored_event)
 434                 return;
 435
 436         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 437
 438         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 439         if (unlikely(!event_priv))
 440                 goto skip_send_ignore;
 441
 442         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 443
 444         fsn_event_priv->group = group;
 445         event_priv->wd = ientry->wd;
 446
 447         ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
 448         if (ret)
 449                 inotify_free_event_priv(fsn_event_priv);
 450
 451 skip_send_ignore:
 452
 453         /* matches the reference taken when the event was created */
 454         fsnotify_put_event(ignored_event);
 455
 456         /* remove this entry from the idr */
 457         inotify_remove_from_idr(group, ientry);
 458
 459         atomic_dec(&group->inotify_data.user->inotify_watches);
 460 }
 461
 462 /* ding dong the mark is dead */
 463 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 464 {
 465         struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
 466
 467         kmem_cache_free(inotify_inode_mark_cachep, ientry);
 468 }
 469
 470 static int inotify_update_existing_watch(struct fsnotify_group *group,
 471                                          struct inode *inode,
 472                                          u32 arg)
 473 {
 474         struct fsnotify_mark_entry *entry;
 475         struct inotify_inode_mark_entry *ientry;
 476         __u32 old_mask, new_mask;
 477         __u32 mask;
 478         int add = (arg & IN_MASK_ADD);
 479         int ret;
 480
 481         /* don't allow invalid bits: we don't want flags set */
 482         mask = inotify_arg_to_mask(arg);
 483         if (unlikely(!mask))
 484                 return -EINVAL;
 485
 486         spin_lock(&inode->i_lock);
 487         entry = fsnotify_find_mark_entry(group, inode);
 488         spin_unlock(&inode->i_lock);
 489         if (!entry)
 490                 return -ENOENT;
 491
 492         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 493
 494         spin_lock(&entry->lock);
 495
 496         old_mask = entry->mask;
 497         if (add) {
 498                 entry->mask |= mask;
 499                 new_mask = entry->mask;
 500         } else {
 501                 entry->mask = mask;
 502                 new_mask = entry->mask;
 503         }
 504
 505         spin_unlock(&entry->lock);
 506
 507         if (old_mask != new_mask) {
 508                 /* more bits in old than in new? */
 509                 int dropped = (old_mask & ~new_mask);
 510                 /* more bits in this entry than the inode's mask? */
 511                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 512                 /* more bits in this entry than the group? */
 513                 int do_group = (new_mask & ~group->mask);
 514
 515                 /* update the inode with this new entry */
 516                 if (dropped || do_inode)
 517                         fsnotify_recalc_inode_mask(inode);
 518
 519                 /* update the group mask with the new mask */
 520                 if (dropped || do_group)
 521                         fsnotify_recalc_group_mask(group);
 522         }
 523
 524         /* return the wd */
 525         ret = ientry->wd;
 526
 527         /* match the get from fsnotify_find_mark_entry() */
 528         fsnotify_put_mark(entry);
 529
 530         return ret;
 531 }
 532
 533 static int inotify_new_watch(struct fsnotify_group *group,
 534                              struct inode *inode,
 535                              u32 arg)
 536 {
 537         struct inotify_inode_mark_entry *tmp_ientry;
 538         __u32 mask;
 539         int ret;
 540
 541         /* don't allow invalid bits: we don't want flags set */
 542         mask = inotify_arg_to_mask(arg);
 543         if (unlikely(!mask))
 544                 return -EINVAL;
 545
 546         tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 547         if (unlikely(!tmp_ientry))
 548                 return -ENOMEM;
 549
 550         fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
 551         tmp_ientry->fsn_entry.mask = mask;
 552         tmp_ientry->wd = -1;
 553
 554         ret = -ENOSPC;
 555         if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 556                 goto out_err;
 557 retry:
 558         ret = -ENOMEM;
 559         if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
 560                 goto out_err;
 561
 562         /* we are putting the mark on the idr, take a reference */
 563         fsnotify_get_mark(&tmp_ientry->fsn_entry);
 564
 565         spin_lock(&group->inotify_data.idr_lock);
 566         ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
 567                                 group->inotify_data.last_wd+1,
 568                                 &tmp_ientry->wd);
 569         spin_unlock(&group->inotify_data.idr_lock);
 570         if (ret) {
 571                 /* we didn't get on the idr, drop the idr reference */
 572                 fsnotify_put_mark(&tmp_ientry->fsn_entry);
 573
 574                 /* idr was out of memory allocate and try again */
 575                 if (ret == -EAGAIN)
 576                         goto retry;
 577                 goto out_err;
 578         }
 579
 580         /* we are on the idr, now get on the inode */
 581         ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 582         if (ret) {
 583                 /* we failed to get on the inode, get off the idr */
 584                 inotify_remove_from_idr(group, tmp_ientry);
 585                 goto out_err;
 586         }
 587
 588         /* update the idr hint, who cares about races, it's just a hint */
 589         group->inotify_data.last_wd = tmp_ientry->wd;
 590
 591         /* increment the number of watches the user has */
 592         atomic_inc(&group->inotify_data.user->inotify_watches);
 593
 594         /* return the watch descriptor for this new entry */
 595         ret = tmp_ientry->wd;
 596
 597         /* match the ref from fsnotify_init_markentry() */
 598         fsnotify_put_mark(&tmp_ientry->fsn_entry);
 599
 600         /* if this mark added a new event update the group mask */
 601         if (mask & ~group->mask)
 602                 fsnotify_recalc_group_mask(group);
 603
 604 out_err:
 605         if (ret < 0)
 606                 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 607
 608         return ret;
 609 }
 610
 611 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 612 {
 613         int ret = 0;
 614
 615 retry:
 616         /* try to update and existing watch with the new arg */
 617         ret = inotify_update_existing_watch(group, inode, arg);
 618         /* no mark present, try to add a new one */
 619         if (ret == -ENOENT)
 620                 ret = inotify_new_watch(group, inode, arg);
 621         /*
 622          * inotify_new_watch could race with another thread which did an
 623          * inotify_new_watch between the update_existing and the add watch
 624          * here, go back and try to update an existing mark again.
 625          */
 626         if (ret == -EEXIST)
 627                 goto retry;
 628
 629         return ret;
 630 }
 631
 632 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 633 {
 634         struct fsnotify_group *group;
 635         unsigned int grp_num;
 636
 637         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 638         grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
 639         group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
 640         if (IS_ERR(group))
 641                 return group;
 642
 643         group->max_events = max_events;
 644
 645         spin_lock_init(&group->inotify_data.idr_lock);
 646         idr_init(&group->inotify_data.idr);
 647         group->inotify_data.last_wd = 0;
 648         group->inotify_data.user = user;
 649         group->inotify_data.fa = NULL;
 650
 651         return group;
 652 }
 653
 654
 655 /* inotify syscalls */
 656 SYSCALL_DEFINE1(inotify_init1, int, flags)
 657 {
 658         struct fsnotify_group *group;
 659         struct user_struct *user;
 660         struct file *filp;
 661         int fd, ret;
 662
 663         /* Check the IN_* constants for consistency.  */
 664         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 665         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 666
 667         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 668                 return -EINVAL;
 669
 670         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 671         if (fd < 0)
 672                 return fd;
 673
 674         filp = get_empty_filp();
 675         if (!filp) {
 676                 ret = -ENFILE;
 677                 goto out_put_fd;
 678         }
 679
 680         user = get_current_user();
 681         if (unlikely(atomic_read(&user->inotify_devs) >=
 682                         inotify_max_user_instances)) {
 683                 ret = -EMFILE;
 684                 goto out_free_uid;
 685         }
 686
 687         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 688         group = inotify_new_group(user, inotify_max_queued_events);
 689         if (IS_ERR(group)) {
 690                 ret = PTR_ERR(group);
 691                 goto out_free_uid;
 692         }
 693
 694         filp->f_op = &inotify_fops;
 695         filp->f_path.mnt = mntget(inotify_mnt);
 696         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 697         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 698         filp->f_mode = FMODE_READ;
 699         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 700         filp->private_data = group;
 701
 702         atomic_inc(&user->inotify_devs);
 703
 704         fd_install(fd, filp);
 705
 706         return fd;
 707
 708 out_free_uid:
 709         free_uid(user);
 710         put_filp(filp);
 711 out_put_fd:
 712         put_unused_fd(fd);
 713         return ret;
 714 }
 715
 716 SYSCALL_DEFINE0(inotify_init)
 717 {
 718         return sys_inotify_init1(0);
 719 }
 720
 721 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 722                 u32, mask)
 723 {
 724         struct fsnotify_group *group;
 725         struct inode *inode;
 726         struct path path;
 727         struct file *filp;
 728         int ret, fput_needed;
 729         unsigned flags = 0;
 730
 731         filp = fget_light(fd, &fput_needed);
 732         if (unlikely(!filp))
 733                 return -EBADF;
 734
 735         /* verify that this is indeed an inotify instance */
 736         if (unlikely(filp->f_op != &inotify_fops)) {
 737                 ret = -EINVAL;
 738                 goto fput_and_out;
 739         }
 740
 741         if (!(mask & IN_DONT_FOLLOW))
 742                 flags |= LOOKUP_FOLLOW;
 743         if (mask & IN_ONLYDIR)
 744                 flags |= LOOKUP_DIRECTORY;
 745
 746         ret = inotify_find_inode(pathname, &path, flags);
 747         if (ret)
 748                 goto fput_and_out;
 749
 750         /* inode held in place by reference to path; group by fget on fd */
 751         inode = path.dentry->d_inode;
 752         group = filp->private_data;
 753
 754         /* create/update an inode mark */
 755         ret = inotify_update_watch(group, inode, mask);
 756         if (unlikely(ret))
 757                 goto path_put_and_out;
 758
 759 path_put_and_out:
 760         path_put(&path);
 761 fput_and_out:
 762         fput_light(filp, fput_needed);
 763         return ret;
 764 }
 765
 766 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 767 {
 768         struct fsnotify_group *group;
 769         struct fsnotify_mark_entry *entry;
 770         struct file *filp;
 771         int ret = 0, fput_needed;
 772
 773         filp = fget_light(fd, &fput_needed);
 774         if (unlikely(!filp))
 775                 return -EBADF;
 776
 777         /* verify that this is indeed an inotify instance */
 778         if (unlikely(filp->f_op != &inotify_fops)) {
 779                 ret = -EINVAL;
 780                 goto out;
 781         }
 782
 783         group = filp->private_data;
 784
 785         spin_lock(&group->inotify_data.idr_lock);
 786         entry = idr_find(&group->inotify_data.idr, wd);
 787         if (unlikely(!entry)) {
 788                 spin_unlock(&group->inotify_data.idr_lock);
 789                 ret = -EINVAL;
 790                 goto out;
 791         }
 792         fsnotify_get_mark(entry);
 793         spin_unlock(&group->inotify_data.idr_lock);
 794
 795         fsnotify_destroy_mark_by_entry(entry);
 796         fsnotify_put_mark(entry);
 797
 798 out:
 799         fput_light(filp, fput_needed);
 800         return ret;
 801 }
 802
 803 static int
 804 inotify_get_sb(struct file_system_type *fs_type, int flags,
 805                const char *dev_name, void *data, struct vfsmount *mnt)
 806 {
 807         return get_sb_pseudo(fs_type, "inotify", NULL,
 808                         INOTIFYFS_SUPER_MAGIC, mnt);
 809 }
 810
 811 static struct file_system_type inotify_fs_type = {
 812     .name       = "inotifyfs",
 813     .get_sb     = inotify_get_sb,
 814     .kill_sb    = kill_anon_super,
 815 };
 816
 817 /*
 818  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 819  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 820  * must result in panic().
 821  */
 822 static int __init inotify_user_setup(void)
 823 {
 824         int ret;
 825
 826         ret = register_filesystem(&inotify_fs_type);
 827         if (unlikely(ret))
 828                 panic("inotify: register_filesystem returned %d!\n", ret);
 829
 830         inotify_mnt = kern_mount(&inotify_fs_type);
 831         if (IS_ERR(inotify_mnt))
 832                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 833
 834         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
 835         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 836
 837         inotify_max_queued_events = 16384;
 838         inotify_max_user_instances = 128;
 839         inotify_max_user_watches = 8192;
 840
 841         return 0;
 842 }
 843 module_init(inotify_user_setup);