fs/notify/inotify/inotify_user.c

   1 /*
   2  * fs/inotify_user.c - inotify support for userspace
   3  *
   4  * Authors:
   5  *      John McCutchan  <ttb@tentacle.dhs.org>
   6  *      Robert Love     <rml@novell.com>
   7  *
   8  * Copyright (C) 2005 John McCutchan
   9  * Copyright 2006 Hewlett-Packard Development Company, L.P.
  10  *
  11  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
  12  * inotify was largely rewriten to make use of the fsnotify infrastructure
  13  *
  14  * This program is free software; you can redistribute it and/or modify it
  15  * under the terms of the GNU General Public License as published by the
  16  * Free Software Foundation; either version 2, or (at your option) any
  17  * later version.
  18  *
  19  * This program is distributed in the hope that it will be useful, but
  20  * WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  * General Public License for more details.
  23  */
  24
  25 #include <linux/file.h>
  26 #include <linux/fs.h> /* struct inode */
  27 #include <linux/fsnotify_backend.h>
  28 #include <linux/idr.h>
  29 #include <linux/init.h> /* module_init */
  30 #include <linux/inotify.h>
  31 #include <linux/kernel.h> /* roundup() */
  32 #include <linux/magic.h> /* superblock magic number */
  33 #include <linux/mount.h> /* mntget */
  34 #include <linux/namei.h> /* LOOKUP_FOLLOW */
  35 #include <linux/path.h> /* struct path */
  36 #include <linux/sched.h> /* struct user */
  37 #include <linux/slab.h> /* struct kmem_cache */
  38 #include <linux/syscalls.h>
  39 #include <linux/types.h>
  40 #include <linux/uaccess.h>
  41 #include <linux/poll.h>
  42 #include <linux/wait.h>
  43
  44 #include "inotify.h"
  45
  46 #include <asm/ioctls.h>
  47
  48 static struct vfsmount *inotify_mnt __read_mostly;
  49
  50 /* this just sits here and wastes global memory.  used to just pad userspace messages with zeros */
  51 static struct inotify_event nul_inotify_event;
  52
  53 /* these are configurable via /proc/sys/fs/inotify/ */
  54 static int inotify_max_user_instances __read_mostly;
  55 static int inotify_max_queued_events __read_mostly;
  56 int inotify_max_user_watches __read_mostly;
  57
  58 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  59 struct kmem_cache *event_priv_cachep __read_mostly;
  60 static struct fsnotify_event *inotify_ignored_event;
  61
  62 /*
  63  * When inotify registers a new group it increments this and uses that
  64  * value as an offset to set the fsnotify group "name" and priority.
  65  */
  66 static atomic_t inotify_grp_num;
  67
  68 #ifdef CONFIG_SYSCTL
  69
  70 #include <linux/sysctl.h>
  71
  72 static int zero;
  73
  74 ctl_table inotify_table[] = {
  75         {
  76                 .ctl_name       = INOTIFY_MAX_USER_INSTANCES,
  77                 .procname       = "max_user_instances",
  78                 .data           = &inotify_max_user_instances,
  79                 .maxlen         = sizeof(int),
  80                 .mode           = 0644,
  81                 .proc_handler   = &proc_dointvec_minmax,
  82                 .strategy       = &sysctl_intvec,
  83                 .extra1         = &zero,
  84         },
  85         {
  86                 .ctl_name       = INOTIFY_MAX_USER_WATCHES,
  87                 .procname       = "max_user_watches",
  88                 .data           = &inotify_max_user_watches,
  89                 .maxlen         = sizeof(int),
  90                 .mode           = 0644,
  91                 .proc_handler   = &proc_dointvec_minmax,
  92                 .strategy       = &sysctl_intvec,
  93                 .extra1         = &zero,
  94         },
  95         {
  96                 .ctl_name       = INOTIFY_MAX_QUEUED_EVENTS,
  97                 .procname       = "max_queued_events",
  98                 .data           = &inotify_max_queued_events,
  99                 .maxlen         = sizeof(int),
 100                 .mode           = 0644,
 101                 .proc_handler   = &proc_dointvec_minmax,
 102                 .strategy       = &sysctl_intvec,
 103                 .extra1         = &zero
 104         },
 105         { .ctl_name = 0 }
 106 };
 107 #endif /* CONFIG_SYSCTL */
 108
 109 static inline __u32 inotify_arg_to_mask(u32 arg)
 110 {
 111         __u32 mask;
 112
 113         /* everything should accept their own ignored and cares about children */
 114         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
 115
 116         /* mask off the flags used to open the fd */
 117         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
 118
 119         return mask;
 120 }
 121
 122 static inline u32 inotify_mask_to_arg(__u32 mask)
 123 {
 124         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
 125                        IN_Q_OVERFLOW);
 126 }
 127
 128 /* intofiy userspace file descriptor functions */
 129 static unsigned int inotify_poll(struct file *file, poll_table *wait)
 130 {
 131         struct fsnotify_group *group = file->private_data;
 132         int ret = 0;
 133
 134         poll_wait(file, &group->notification_waitq, wait);
 135         mutex_lock(&group->notification_mutex);
 136         if (!fsnotify_notify_queue_is_empty(group))
 137                 ret = POLLIN | POLLRDNORM;
 138         mutex_unlock(&group->notification_mutex);
 139
 140         return ret;
 141 }
 142
 143 /*
 144  * Get an inotify_kernel_event if one exists and is small
 145  * enough to fit in "count". Return an error pointer if
 146  * not large enough.
 147  *
 148  * Called with the group->notification_mutex held.
 149  */
 150 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 151                                             size_t count)
 152 {
 153         size_t event_size = sizeof(struct inotify_event);
 154         struct fsnotify_event *event;
 155
 156         if (fsnotify_notify_queue_is_empty(group))
 157                 return NULL;
 158
 159         event = fsnotify_peek_notify_event(group);
 160
 161         event_size += roundup(event->name_len, event_size);
 162
 163         if (event_size > count)
 164                 return ERR_PTR(-EINVAL);
 165
 166         /* held the notification_mutex the whole time, so this is the
 167          * same event we peeked above */
 168         fsnotify_remove_notify_event(group);
 169
 170         return event;
 171 }
 172
 173 /*
 174  * Copy an event to user space, returning how much we copied.
 175  *
 176  * We already checked that the event size is smaller than the
 177  * buffer we had in "get_one_event()" above.
 178  */
 179 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 180                                   struct fsnotify_event *event,
 181                                   char __user *buf)
 182 {
 183         struct inotify_event inotify_event;
 184         struct fsnotify_event_private_data *fsn_priv;
 185         struct inotify_event_private_data *priv;
 186         size_t event_size = sizeof(struct inotify_event);
 187         size_t name_len;
 188
 189         /* we get the inotify watch descriptor from the event private data */
 190         spin_lock(&event->lock);
 191         fsn_priv = fsnotify_remove_priv_from_event(group, event);
 192         spin_unlock(&event->lock);
 193
 194         if (!fsn_priv)
 195                 inotify_event.wd = -1;
 196         else {
 197                 priv = container_of(fsn_priv, struct inotify_event_private_data,
 198                                     fsnotify_event_priv_data);
 199                 inotify_event.wd = priv->wd;
 200                 inotify_free_event_priv(fsn_priv);
 201         }
 202
 203         /* round up event->name_len so it is a multiple of event_size */
 204         name_len = roundup(event->name_len, event_size);
 205         inotify_event.len = name_len;
 206
 207         inotify_event.mask = inotify_mask_to_arg(event->mask);
 208         inotify_event.cookie = event->sync_cookie;
 209
 210         /* send the main event */
 211         if (copy_to_user(buf, &inotify_event, event_size))
 212                 return -EFAULT;
 213
 214         buf += event_size;
 215
 216         /*
 217          * fsnotify only stores the pathname, so here we have to send the pathname
 218          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 219          * with zeros.  I get my zeros from the nul_inotify_event.
 220          */
 221         if (name_len) {
 222                 unsigned int len_to_zero = name_len - event->name_len;
 223                 /* copy the path name */
 224                 if (copy_to_user(buf, event->file_name, event->name_len))
 225                         return -EFAULT;
 226                 buf += event->name_len;
 227
 228                 /* fill userspace with 0's from nul_inotify_event */
 229                 if (copy_to_user(buf, &nul_inotify_event, len_to_zero))
 230                         return -EFAULT;
 231                 buf += len_to_zero;
 232                 event_size += name_len;
 233         }
 234
 235         return event_size;
 236 }
 237
 238 static ssize_t inotify_read(struct file *file, char __user *buf,
 239                             size_t count, loff_t *pos)
 240 {
 241         struct fsnotify_group *group;
 242         struct fsnotify_event *kevent;
 243         char __user *start;
 244         int ret;
 245         DEFINE_WAIT(wait);
 246
 247         start = buf;
 248         group = file->private_data;
 249
 250         while (1) {
 251                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 252
 253                 mutex_lock(&group->notification_mutex);
 254                 kevent = get_one_event(group, count);
 255                 mutex_unlock(&group->notification_mutex);
 256
 257                 if (kevent) {
 258                         ret = PTR_ERR(kevent);
 259                         if (IS_ERR(kevent))
 260                                 break;
 261                         ret = copy_event_to_user(group, kevent, buf);
 262                         fsnotify_put_event(kevent);
 263                         if (ret < 0)
 264                                 break;
 265                         buf += ret;
 266                         count -= ret;
 267                         continue;
 268                 }
 269
 270                 ret = -EAGAIN;
 271                 if (file->f_flags & O_NONBLOCK)
 272                         break;
 273                 ret = -EINTR;
 274                 if (signal_pending(current))
 275                         break;
 276
 277                 if (start != buf)
 278                         break;
 279
 280                 schedule();
 281         }
 282
 283         finish_wait(&group->notification_waitq, &wait);
 284         if (start != buf && ret != -EFAULT)
 285                 ret = buf - start;
 286         return ret;
 287 }
 288
 289 static int inotify_fasync(int fd, struct file *file, int on)
 290 {
 291         struct fsnotify_group *group = file->private_data;
 292
 293         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 294 }
 295
 296 static int inotify_release(struct inode *ignored, struct file *file)
 297 {
 298         struct fsnotify_group *group = file->private_data;
 299         struct user_struct *user = group->inotify_data.user;
 300
 301         fsnotify_clear_marks_by_group(group);
 302
 303         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 304         fsnotify_put_group(group);
 305
 306         atomic_dec(&user->inotify_devs);
 307
 308         return 0;
 309 }
 310
 311 static long inotify_ioctl(struct file *file, unsigned int cmd,
 312                           unsigned long arg)
 313 {
 314         struct fsnotify_group *group;
 315         struct fsnotify_event_holder *holder;
 316         struct fsnotify_event *event;
 317         void __user *p;
 318         int ret = -ENOTTY;
 319         size_t send_len = 0;
 320
 321         group = file->private_data;
 322         p = (void __user *) arg;
 323
 324         switch (cmd) {
 325         case FIONREAD:
 326                 mutex_lock(&group->notification_mutex);
 327                 list_for_each_entry(holder, &group->notification_list, event_list) {
 328                         event = holder->event;
 329                         send_len += sizeof(struct inotify_event);
 330                         send_len += roundup(event->name_len,
 331                                              sizeof(struct inotify_event));
 332                 }
 333                 mutex_unlock(&group->notification_mutex);
 334                 ret = put_user(send_len, (int __user *) p);
 335                 break;
 336         }
 337
 338         return ret;
 339 }
 340
 341 static const struct file_operations inotify_fops = {
 342         .poll           = inotify_poll,
 343         .read           = inotify_read,
 344         .fasync         = inotify_fasync,
 345         .release        = inotify_release,
 346         .unlocked_ioctl = inotify_ioctl,
 347         .compat_ioctl   = inotify_ioctl,
 348 };
 349
 350
 351 /*
 352  * find_inode - resolve a user-given path to a specific inode
 353  */
 354 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 355 {
 356         int error;
 357
 358         error = user_path_at(AT_FDCWD, dirname, flags, path);
 359         if (error)
 360                 return error;
 361         /* you can only watch an inode if you have read permissions on it */
 362         error = inode_permission(path->dentry->d_inode, MAY_READ);
 363         if (error)
 364                 path_put(path);
 365         return error;
 366 }
 367
 368 static void inotify_remove_from_idr(struct fsnotify_group *group,
 369                                     struct inotify_inode_mark_entry *ientry)
 370 {
 371         struct idr *idr;
 372
 373         spin_lock(&group->inotify_data.idr_lock);
 374         idr = &group->inotify_data.idr;
 375         idr_remove(idr, ientry->wd);
 376         spin_unlock(&group->inotify_data.idr_lock);
 377         ientry->wd = -1;
 378 }
 379 /*
 380  * Send IN_IGNORED for this wd, remove this wd from the idr, and drop the
 381  * internal reference help on the mark because it is in the idr.
 382  */
 383 void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 384                                     struct fsnotify_group *group)
 385 {
 386         struct inotify_inode_mark_entry *ientry;
 387         struct inotify_event_private_data *event_priv;
 388         struct fsnotify_event_private_data *fsn_event_priv;
 389
 390         ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 391
 392         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_KERNEL);
 393         if (unlikely(!event_priv))
 394                 goto skip_send_ignore;
 395
 396         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 397
 398         fsn_event_priv->group = group;
 399         event_priv->wd = ientry->wd;
 400
 401         fsnotify_add_notify_event(group, inotify_ignored_event, fsn_event_priv);
 402
 403         /* did the private data get added? */
 404         if (list_empty(&fsn_event_priv->event_list))
 405                 inotify_free_event_priv(fsn_event_priv);
 406
 407 skip_send_ignore:
 408
 409         /* remove this entry from the idr */
 410         inotify_remove_from_idr(group, ientry);
 411
 412         /* removed from idr, drop that reference */
 413         fsnotify_put_mark(entry);
 414
 415         atomic_dec(&group->inotify_data.user->inotify_watches);
 416 }
 417
 418 /* ding dong the mark is dead */
 419 static void inotify_free_mark(struct fsnotify_mark_entry *entry)
 420 {
 421         struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
 422
 423         kmem_cache_free(inotify_inode_mark_cachep, ientry);
 424 }
 425
 426 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 427 {
 428         struct fsnotify_mark_entry *entry = NULL;
 429         struct inotify_inode_mark_entry *ientry;
 430         struct inotify_inode_mark_entry *tmp_ientry;
 431         int ret = 0;
 432         int add = (arg & IN_MASK_ADD);
 433         __u32 mask;
 434         __u32 old_mask, new_mask;
 435
 436         /* don't allow invalid bits: we don't want flags set */
 437         mask = inotify_arg_to_mask(arg);
 438         if (unlikely(!mask))
 439                 return -EINVAL;
 440
 441         tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 442         if (unlikely(!tmp_ientry))
 443                 return -ENOMEM;
 444         /* we set the mask at the end after attaching it */
 445         fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
 446         tmp_ientry->wd = -1;
 447
 448 find_entry:
 449         spin_lock(&inode->i_lock);
 450         entry = fsnotify_find_mark_entry(group, inode);
 451         spin_unlock(&inode->i_lock);
 452         if (entry) {
 453                 ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
 454         } else {
 455                 ret = -ENOSPC;
 456                 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 457                         goto out_err;
 458 retry:
 459                 ret = -ENOMEM;
 460                 if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
 461                         goto out_err;
 462
 463                 spin_lock(&group->inotify_data.idr_lock);
 464                 ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
 465                                         group->inotify_data.last_wd,
 466                                         &tmp_ientry->wd);
 467                 spin_unlock(&group->inotify_data.idr_lock);
 468                 if (ret) {
 469                         if (ret == -EAGAIN)
 470                                 goto retry;
 471                         goto out_err;
 472                 }
 473
 474                 ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
 475                 if (ret) {
 476                         inotify_remove_from_idr(group, tmp_ientry);
 477                         if (ret == -EEXIST)
 478                                 goto find_entry;
 479                         goto out_err;
 480                 }
 481
 482                 /* tmp_ientry has been added to the inode, so we are all set up.
 483                  * now we just need to make sure tmp_ientry doesn't get freed and
 484                  * we need to set up entry and ientry so the generic code can
 485                  * do its thing. */
 486                 ientry = tmp_ientry;
 487                 entry = &ientry->fsn_entry;
 488                 tmp_ientry = NULL;
 489
 490                 atomic_inc(&group->inotify_data.user->inotify_watches);
 491
 492                 /* update the idr hint */
 493                 group->inotify_data.last_wd = ientry->wd;
 494
 495                 /* we put the mark on the idr, take a reference */
 496                 fsnotify_get_mark(entry);
 497         }
 498
 499         ret = ientry->wd;
 500
 501         spin_lock(&entry->lock);
 502
 503         old_mask = entry->mask;
 504         if (add) {
 505                 entry->mask |= mask;
 506                 new_mask = entry->mask;
 507         } else {
 508                 entry->mask = mask;
 509                 new_mask = entry->mask;
 510         }
 511
 512         spin_unlock(&entry->lock);
 513
 514         if (old_mask != new_mask) {
 515                 /* more bits in old than in new? */
 516                 int dropped = (old_mask & ~new_mask);
 517                 /* more bits in this entry than the inode's mask? */
 518                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 519                 /* more bits in this entry than the group? */
 520                 int do_group = (new_mask & ~group->mask);
 521
 522                 /* update the inode with this new entry */
 523                 if (dropped || do_inode)
 524                         fsnotify_recalc_inode_mask(inode);
 525
 526                 /* update the group mask with the new mask */
 527                 if (dropped || do_group)
 528                         fsnotify_recalc_group_mask(group);
 529         }
 530
 531         /* this either matches fsnotify_find_mark_entry, or init_mark_entry
 532          * depending on which path we took... */
 533         fsnotify_put_mark(entry);
 534
 535 out_err:
 536         /* could be an error, could be that we found an existing mark */
 537         if (tmp_ientry) {
 538                 /* on the idr but didn't make it on the inode */
 539                 if (tmp_ientry->wd != -1)
 540                         inotify_remove_from_idr(group, tmp_ientry);
 541                 kmem_cache_free(inotify_inode_mark_cachep, tmp_ientry);
 542         }
 543
 544         return ret;
 545 }
 546
 547 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 548 {
 549         struct fsnotify_group *group;
 550         unsigned int grp_num;
 551
 552         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 553         grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
 554         group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
 555         if (IS_ERR(group))
 556                 return group;
 557
 558         group->max_events = max_events;
 559
 560         spin_lock_init(&group->inotify_data.idr_lock);
 561         idr_init(&group->inotify_data.idr);
 562         group->inotify_data.last_wd = 0;
 563         group->inotify_data.user = user;
 564         group->inotify_data.fa = NULL;
 565
 566         return group;
 567 }
 568
 569
 570 /* inotify syscalls */
 571 SYSCALL_DEFINE1(inotify_init1, int, flags)
 572 {
 573         struct fsnotify_group *group;
 574         struct user_struct *user;
 575         struct file *filp;
 576         int fd, ret;
 577
 578         /* Check the IN_* constants for consistency.  */
 579         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 580         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 581
 582         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 583                 return -EINVAL;
 584
 585         fd = get_unused_fd_flags(flags & O_CLOEXEC);
 586         if (fd < 0)
 587                 return fd;
 588
 589         filp = get_empty_filp();
 590         if (!filp) {
 591                 ret = -ENFILE;
 592                 goto out_put_fd;
 593         }
 594
 595         user = get_current_user();
 596         if (unlikely(atomic_read(&user->inotify_devs) >=
 597                         inotify_max_user_instances)) {
 598                 ret = -EMFILE;
 599                 goto out_free_uid;
 600         }
 601
 602         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 603         group = inotify_new_group(user, inotify_max_queued_events);
 604         if (IS_ERR(group)) {
 605                 ret = PTR_ERR(group);
 606                 goto out_free_uid;
 607         }
 608
 609         filp->f_op = &inotify_fops;
 610         filp->f_path.mnt = mntget(inotify_mnt);
 611         filp->f_path.dentry = dget(inotify_mnt->mnt_root);
 612         filp->f_mapping = filp->f_path.dentry->d_inode->i_mapping;
 613         filp->f_mode = FMODE_READ;
 614         filp->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 615         filp->private_data = group;
 616
 617         atomic_inc(&user->inotify_devs);
 618
 619         fd_install(fd, filp);
 620
 621         return fd;
 622
 623 out_free_uid:
 624         free_uid(user);
 625         put_filp(filp);
 626 out_put_fd:
 627         put_unused_fd(fd);
 628         return ret;
 629 }
 630
 631 SYSCALL_DEFINE0(inotify_init)
 632 {
 633         return sys_inotify_init1(0);
 634 }
 635
 636 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 637                 u32, mask)
 638 {
 639         struct fsnotify_group *group;
 640         struct inode *inode;
 641         struct path path;
 642         struct file *filp;
 643         int ret, fput_needed;
 644         unsigned flags = 0;
 645
 646         filp = fget_light(fd, &fput_needed);
 647         if (unlikely(!filp))
 648                 return -EBADF;
 649
 650         /* verify that this is indeed an inotify instance */
 651         if (unlikely(filp->f_op != &inotify_fops)) {
 652                 ret = -EINVAL;
 653                 goto fput_and_out;
 654         }
 655
 656         if (!(mask & IN_DONT_FOLLOW))
 657                 flags |= LOOKUP_FOLLOW;
 658         if (mask & IN_ONLYDIR)
 659                 flags |= LOOKUP_DIRECTORY;
 660
 661         ret = inotify_find_inode(pathname, &path, flags);
 662         if (ret)
 663                 goto fput_and_out;
 664
 665         /* inode held in place by reference to path; group by fget on fd */
 666         inode = path.dentry->d_inode;
 667         group = filp->private_data;
 668
 669         /* create/update an inode mark */
 670         ret = inotify_update_watch(group, inode, mask);
 671         if (unlikely(ret))
 672                 goto path_put_and_out;
 673
 674 path_put_and_out:
 675         path_put(&path);
 676 fput_and_out:
 677         fput_light(filp, fput_needed);
 678         return ret;
 679 }
 680
 681 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 682 {
 683         struct fsnotify_group *group;
 684         struct fsnotify_mark_entry *entry;
 685         struct file *filp;
 686         int ret = 0, fput_needed;
 687
 688         filp = fget_light(fd, &fput_needed);
 689         if (unlikely(!filp))
 690                 return -EBADF;
 691
 692         /* verify that this is indeed an inotify instance */
 693         if (unlikely(filp->f_op != &inotify_fops)) {
 694                 ret = -EINVAL;
 695                 goto out;
 696         }
 697
 698         group = filp->private_data;
 699
 700         spin_lock(&group->inotify_data.idr_lock);
 701         entry = idr_find(&group->inotify_data.idr, wd);
 702         if (unlikely(!entry)) {
 703                 spin_unlock(&group->inotify_data.idr_lock);
 704                 ret = -EINVAL;
 705                 goto out;
 706         }
 707         fsnotify_get_mark(entry);
 708         spin_unlock(&group->inotify_data.idr_lock);
 709
 710         fsnotify_destroy_mark_by_entry(entry);
 711         fsnotify_put_mark(entry);
 712
 713 out:
 714         fput_light(filp, fput_needed);
 715         return ret;
 716 }
 717
 718 static int
 719 inotify_get_sb(struct file_system_type *fs_type, int flags,
 720                const char *dev_name, void *data, struct vfsmount *mnt)
 721 {
 722         return get_sb_pseudo(fs_type, "inotify", NULL,
 723                         INOTIFYFS_SUPER_MAGIC, mnt);
 724 }
 725
 726 static struct file_system_type inotify_fs_type = {
 727     .name       = "inotifyfs",
 728     .get_sb     = inotify_get_sb,
 729     .kill_sb    = kill_anon_super,
 730 };
 731
 732 /*
 733  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 734  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 735  * must result in panic().
 736  */
 737 static int __init inotify_user_setup(void)
 738 {
 739         int ret;
 740
 741         ret = register_filesystem(&inotify_fs_type);
 742         if (unlikely(ret))
 743                 panic("inotify: register_filesystem returned %d!\n", ret);
 744
 745         inotify_mnt = kern_mount(&inotify_fs_type);
 746         if (IS_ERR(inotify_mnt))
 747                 panic("inotify: kern_mount ret %ld!\n", PTR_ERR(inotify_mnt));
 748
 749         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
 750         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 751         inotify_ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, FSNOTIFY_EVENT_NONE, NULL, 0);
 752         if (!inotify_ignored_event)
 753                 panic("unable to allocate the inotify ignored event\n");
 754
 755         inotify_max_queued_events = 16384;
 756         inotify_max_user_instances = 128;
 757         inotify_max_user_watches = 8192;
 758
 759         return 0;
 760 }
 761 module_init(inotify_user_setup);