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/namei.h> /* LOOKUP_FOLLOW */
  33 #include <linux/sched.h> /* struct user */
  34 #include <linux/slab.h> /* struct kmem_cache */
  35 #include <linux/syscalls.h>
  36 #include <linux/types.h>
  37 #include <linux/anon_inodes.h>
  38 #include <linux/uaccess.h>
  39 #include <linux/poll.h>
  40 #include <linux/wait.h>
  41
  42 #include "inotify.h"
  43
  44 #include <asm/ioctls.h>
  45
  46 /* these are configurable via /proc/sys/fs/inotify/ */
  47 static int inotify_max_user_instances __read_mostly;
  48 static int inotify_max_queued_events __read_mostly;
  49 static int inotify_max_user_watches __read_mostly;
  50
  51 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  52 struct kmem_cache *event_priv_cachep __read_mostly;
  53
  54 #ifdef CONFIG_SYSCTL
  55
  56 #include <linux/sysctl.h>
  57
  58 static int zero;
  59
  60 ctl_table inotify_table[] = {
  61         {
  62                 .procname       = "max_user_instances",
  63                 .data           = &inotify_max_user_instances,
  64                 .maxlen         = sizeof(int),
  65                 .mode           = 0644,
  66                 .proc_handler   = proc_dointvec_minmax,
  67                 .extra1         = &zero,
  68         },
  69         {
  70                 .procname       = "max_user_watches",
  71                 .data           = &inotify_max_user_watches,
  72                 .maxlen         = sizeof(int),
  73                 .mode           = 0644,
  74                 .proc_handler   = proc_dointvec_minmax,
  75                 .extra1         = &zero,
  76         },
  77         {
  78                 .procname       = "max_queued_events",
  79                 .data           = &inotify_max_queued_events,
  80                 .maxlen         = sizeof(int),
  81                 .mode           = 0644,
  82                 .proc_handler   = proc_dointvec_minmax,
  83                 .extra1         = &zero
  84         },
  85         { }
  86 };
  87 #endif /* CONFIG_SYSCTL */
  88
  89 static inline __u32 inotify_arg_to_mask(u32 arg)
  90 {
  91         __u32 mask;
  92
  93         /* everything should accept their own ignored and cares about children */
  94         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
  95
  96         /* mask off the flags used to open the fd */
  97         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
  98
  99         return mask;
 100 }
 101
 102 static inline u32 inotify_mask_to_arg(__u32 mask)
 103 {
 104         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
 105                        IN_Q_OVERFLOW);
 106 }
 107
 108 /* intofiy userspace file descriptor functions */
 109 static unsigned int inotify_poll(struct file *file, poll_table *wait)
 110 {
 111         struct fsnotify_group *group = file->private_data;
 112         int ret = 0;
 113
 114         poll_wait(file, &group->notification_waitq, wait);
 115         mutex_lock(&group->notification_mutex);
 116         if (!fsnotify_notify_queue_is_empty(group))
 117                 ret = POLLIN | POLLRDNORM;
 118         mutex_unlock(&group->notification_mutex);
 119
 120         return ret;
 121 }
 122
 123 /*
 124  * Get an inotify_kernel_event if one exists and is small
 125  * enough to fit in "count". Return an error pointer if
 126  * not large enough.
 127  *
 128  * Called with the group->notification_mutex held.
 129  */
 130 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 131                                             size_t count)
 132 {
 133         size_t event_size = sizeof(struct inotify_event);
 134         struct fsnotify_event *event;
 135
 136         if (fsnotify_notify_queue_is_empty(group))
 137                 return NULL;
 138
 139         event = fsnotify_peek_notify_event(group);
 140
 141         if (event->name_len)
 142                 event_size += roundup(event->name_len + 1, event_size);
 143
 144         if (event_size > count)
 145                 return ERR_PTR(-EINVAL);
 146
 147         /* held the notification_mutex the whole time, so this is the
 148          * same event we peeked above */
 149         fsnotify_remove_notify_event(group);
 150
 151         return event;
 152 }
 153
 154 /*
 155  * Copy an event to user space, returning how much we copied.
 156  *
 157  * We already checked that the event size is smaller than the
 158  * buffer we had in "get_one_event()" above.
 159  */
 160 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 161                                   struct fsnotify_event *event,
 162                                   char __user *buf)
 163 {
 164         struct inotify_event inotify_event;
 165         struct fsnotify_event_private_data *fsn_priv;
 166         struct inotify_event_private_data *priv;
 167         size_t event_size = sizeof(struct inotify_event);
 168         size_t name_len = 0;
 169
 170         /* we get the inotify watch descriptor from the event private data */
 171         spin_lock(&event->lock);
 172         fsn_priv = fsnotify_remove_priv_from_event(group, event);
 173         spin_unlock(&event->lock);
 174
 175         if (!fsn_priv)
 176                 inotify_event.wd = -1;
 177         else {
 178                 priv = container_of(fsn_priv, struct inotify_event_private_data,
 179                                     fsnotify_event_priv_data);
 180                 inotify_event.wd = priv->wd;
 181                 inotify_free_event_priv(fsn_priv);
 182         }
 183
 184         /*
 185          * round up event->name_len so it is a multiple of event_size
 186          * plus an extra byte for the terminating '\0'.
 187          */
 188         if (event->name_len)
 189                 name_len = roundup(event->name_len + 1, event_size);
 190         inotify_event.len = name_len;
 191
 192         inotify_event.mask = inotify_mask_to_arg(event->mask);
 193         inotify_event.cookie = event->sync_cookie;
 194
 195         /* send the main event */
 196         if (copy_to_user(buf, &inotify_event, event_size))
 197                 return -EFAULT;
 198
 199         buf += event_size;
 200
 201         /*
 202          * fsnotify only stores the pathname, so here we have to send the pathname
 203          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 204          * with zeros.  I get my zeros from the nul_inotify_event.
 205          */
 206         if (name_len) {
 207                 unsigned int len_to_zero = name_len - event->name_len;
 208                 /* copy the path name */
 209                 if (copy_to_user(buf, event->file_name, event->name_len))
 210                         return -EFAULT;
 211                 buf += event->name_len;
 212
 213                 /* fill userspace with 0's */
 214                 if (clear_user(buf, len_to_zero))
 215                         return -EFAULT;
 216                 buf += len_to_zero;
 217                 event_size += name_len;
 218         }
 219
 220         return event_size;
 221 }
 222
 223 static ssize_t inotify_read(struct file *file, char __user *buf,
 224                             size_t count, loff_t *pos)
 225 {
 226         struct fsnotify_group *group;
 227         struct fsnotify_event *kevent;
 228         char __user *start;
 229         int ret;
 230         DEFINE_WAIT(wait);
 231
 232         start = buf;
 233         group = file->private_data;
 234
 235         while (1) {
 236                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 237
 238                 mutex_lock(&group->notification_mutex);
 239                 kevent = get_one_event(group, count);
 240                 mutex_unlock(&group->notification_mutex);
 241
 242                 if (kevent) {
 243                         ret = PTR_ERR(kevent);
 244                         if (IS_ERR(kevent))
 245                                 break;
 246                         ret = copy_event_to_user(group, kevent, buf);
 247                         fsnotify_put_event(kevent);
 248                         if (ret < 0)
 249                                 break;
 250                         buf += ret;
 251                         count -= ret;
 252                         continue;
 253                 }
 254
 255                 ret = -EAGAIN;
 256                 if (file->f_flags & O_NONBLOCK)
 257                         break;
 258                 ret = -EINTR;
 259                 if (signal_pending(current))
 260                         break;
 261
 262                 if (start != buf)
 263                         break;
 264
 265                 schedule();
 266         }
 267
 268         finish_wait(&group->notification_waitq, &wait);
 269         if (start != buf && ret != -EFAULT)
 270                 ret = buf - start;
 271         return ret;
 272 }
 273
 274 static int inotify_fasync(int fd, struct file *file, int on)
 275 {
 276         struct fsnotify_group *group = file->private_data;
 277
 278         return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO;
 279 }
 280
 281 static int inotify_release(struct inode *ignored, struct file *file)
 282 {
 283         struct fsnotify_group *group = file->private_data;
 284         struct user_struct *user = group->inotify_data.user;
 285
 286         fsnotify_clear_marks_by_group(group);
 287
 288         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 289         fsnotify_put_group(group);
 290
 291         atomic_dec(&user->inotify_devs);
 292
 293         return 0;
 294 }
 295
 296 static long inotify_ioctl(struct file *file, unsigned int cmd,
 297                           unsigned long arg)
 298 {
 299         struct fsnotify_group *group;
 300         struct fsnotify_event_holder *holder;
 301         struct fsnotify_event *event;
 302         void __user *p;
 303         int ret = -ENOTTY;
 304         size_t send_len = 0;
 305
 306         group = file->private_data;
 307         p = (void __user *) arg;
 308
 309         switch (cmd) {
 310         case FIONREAD:
 311                 mutex_lock(&group->notification_mutex);
 312                 list_for_each_entry(holder, &group->notification_list, event_list) {
 313                         event = holder->event;
 314                         send_len += sizeof(struct inotify_event);
 315                         if (event->name_len)
 316                                 send_len += roundup(event->name_len + 1,
 317                                                 sizeof(struct inotify_event));
 318                 }
 319                 mutex_unlock(&group->notification_mutex);
 320                 ret = put_user(send_len, (int __user *) p);
 321                 break;
 322         }
 323
 324         return ret;
 325 }
 326
 327 static const struct file_operations inotify_fops = {
 328         .poll           = inotify_poll,
 329         .read           = inotify_read,
 330         .fasync         = inotify_fasync,
 331         .release        = inotify_release,
 332         .unlocked_ioctl = inotify_ioctl,
 333         .compat_ioctl   = inotify_ioctl,
 334 };
 335
 336
 337 /*
 338  * find_inode - resolve a user-given path to a specific inode
 339  */
 340 static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags)
 341 {
 342         int error;
 343
 344         error = user_path_at(AT_FDCWD, dirname, flags, path);
 345         if (error)
 346                 return error;
 347         /* you can only watch an inode if you have read permissions on it */
 348         error = inode_permission(path->dentry->d_inode, MAY_READ);
 349         if (error)
 350                 path_put(path);
 351         return error;
 352 }
 353
 354 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
 355                               int *last_wd,
 356                               struct inotify_inode_mark *i_mark)
 357 {
 358         int ret;
 359
 360         do {
 361                 if (unlikely(!idr_pre_get(idr, GFP_KERNEL)))
 362                         return -ENOMEM;
 363
 364                 spin_lock(idr_lock);
 365                 ret = idr_get_new_above(idr, i_mark, *last_wd + 1,
 366                                         &i_mark->wd);
 367                 /* we added the mark to the idr, take a reference */
 368                 if (!ret) {
 369                         *last_wd = i_mark->wd;
 370                         fsnotify_get_mark(&i_mark->fsn_mark);
 371                 }
 372                 spin_unlock(idr_lock);
 373         } while (ret == -EAGAIN);
 374
 375         return ret;
 376 }
 377
 378 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
 379                                                                 int wd)
 380 {
 381         struct idr *idr = &group->inotify_data.idr;
 382         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 383         struct inotify_inode_mark *i_mark;
 384
 385         assert_spin_locked(idr_lock);
 386
 387         i_mark = idr_find(idr, wd);
 388         if (i_mark) {
 389                 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
 390
 391                 fsnotify_get_mark(fsn_mark);
 392                 /* One ref for being in the idr, one ref we just took */
 393                 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
 394         }
 395
 396         return i_mark;
 397 }
 398
 399 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
 400                                                          int wd)
 401 {
 402         struct inotify_inode_mark *i_mark;
 403         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 404
 405         spin_lock(idr_lock);
 406         i_mark = inotify_idr_find_locked(group, wd);
 407         spin_unlock(idr_lock);
 408
 409         return i_mark;
 410 }
 411
 412 static void do_inotify_remove_from_idr(struct fsnotify_group *group,
 413                                        struct inotify_inode_mark *i_mark)
 414 {
 415         struct idr *idr = &group->inotify_data.idr;
 416         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 417         int wd = i_mark->wd;
 418
 419         assert_spin_locked(idr_lock);
 420
 421         idr_remove(idr, wd);
 422
 423         /* removed from the idr, drop that ref */
 424         fsnotify_put_mark(&i_mark->fsn_mark);
 425 }
 426
 427 /*
 428  * Remove the mark from the idr (if present) and drop the reference
 429  * on the mark because it was in the idr.
 430  */
 431 static void inotify_remove_from_idr(struct fsnotify_group *group,
 432                                     struct inotify_inode_mark *i_mark)
 433 {
 434         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 435         struct inotify_inode_mark *found_i_mark = NULL;
 436         int wd;
 437
 438         spin_lock(idr_lock);
 439         wd = i_mark->wd;
 440
 441         /*
 442          * does this i_mark think it is in the idr?  we shouldn't get called
 443          * if it wasn't....
 444          */
 445         if (wd == -1) {
 446                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
 447                         " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
 448                         i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
 449                 goto out;
 450         }
 451
 452         /* Lets look in the idr to see if we find it */
 453         found_i_mark = inotify_idr_find_locked(group, wd);
 454         if (unlikely(!found_i_mark)) {
 455                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
 456                         " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
 457                         i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
 458                 goto out;
 459         }
 460
 461         /*
 462          * We found an mark in the idr at the right wd, but it's
 463          * not the mark we were told to remove.  eparis seriously
 464          * fucked up somewhere.
 465          */
 466         if (unlikely(found_i_mark != i_mark)) {
 467                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
 468                         "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
 469                         "found_i_mark->group=%p found_i_mark->inode=%p\n",
 470                         __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
 471                         i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
 472                         found_i_mark->fsn_mark.group,
 473                         found_i_mark->fsn_mark.i.inode);
 474                 goto out;
 475         }
 476
 477         /*
 478          * One ref for being in the idr
 479          * one ref held by the caller trying to kill us
 480          * one ref grabbed by inotify_idr_find
 481          */
 482         if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
 483                 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
 484                         " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
 485                         i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
 486                 /* we can't really recover with bad ref cnting.. */
 487                 BUG();
 488         }
 489
 490         do_inotify_remove_from_idr(group, i_mark);
 491 out:
 492         /* match the ref taken by inotify_idr_find_locked() */
 493         if (found_i_mark)
 494                 fsnotify_put_mark(&found_i_mark->fsn_mark);
 495         i_mark->wd = -1;
 496         spin_unlock(idr_lock);
 497 }
 498
 499 /*
 500  * Send IN_IGNORED for this wd, remove this wd from the idr.
 501  */
 502 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
 503                                     struct fsnotify_group *group)
 504 {
 505         struct inotify_inode_mark *i_mark;
 506         struct fsnotify_event *ignored_event;
 507         struct inotify_event_private_data *event_priv;
 508         struct fsnotify_event_private_data *fsn_event_priv;
 509         int ret;
 510
 511         ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL,
 512                                               FSNOTIFY_EVENT_NONE, NULL, 0,
 513                                               GFP_NOFS);
 514         if (!ignored_event)
 515                 return;
 516
 517         i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 518
 519         event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 520         if (unlikely(!event_priv))
 521                 goto skip_send_ignore;
 522
 523         fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 524
 525         fsn_event_priv->group = group;
 526         event_priv->wd = i_mark->wd;
 527
 528         ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL, NULL);
 529         if (ret)
 530                 inotify_free_event_priv(fsn_event_priv);
 531
 532 skip_send_ignore:
 533
 534         /* matches the reference taken when the event was created */
 535         fsnotify_put_event(ignored_event);
 536
 537         /* remove this mark from the idr */
 538         inotify_remove_from_idr(group, i_mark);
 539
 540         atomic_dec(&group->inotify_data.user->inotify_watches);
 541 }
 542
 543 /* ding dong the mark is dead */
 544 static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
 545 {
 546         struct inotify_inode_mark *i_mark;
 547
 548         i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 549
 550         kmem_cache_free(inotify_inode_mark_cachep, i_mark);
 551 }
 552
 553 static int inotify_update_existing_watch(struct fsnotify_group *group,
 554                                          struct inode *inode,
 555                                          u32 arg)
 556 {
 557         struct fsnotify_mark *fsn_mark;
 558         struct inotify_inode_mark *i_mark;
 559         __u32 old_mask, new_mask;
 560         __u32 mask;
 561         int add = (arg & IN_MASK_ADD);
 562         int ret;
 563
 564         /* don't allow invalid bits: we don't want flags set */
 565         mask = inotify_arg_to_mask(arg);
 566         if (unlikely(!mask))
 567                 return -EINVAL;
 568
 569         fsn_mark = fsnotify_find_inode_mark(group, inode);
 570         if (!fsn_mark)
 571                 return -ENOENT;
 572
 573         i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 574
 575         spin_lock(&fsn_mark->lock);
 576
 577         old_mask = fsn_mark->mask;
 578         if (add)
 579                 fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
 580         else
 581                 fsnotify_set_mark_mask_locked(fsn_mark, mask);
 582         new_mask = fsn_mark->mask;
 583
 584         spin_unlock(&fsn_mark->lock);
 585
 586         if (old_mask != new_mask) {
 587                 /* more bits in old than in new? */
 588                 int dropped = (old_mask & ~new_mask);
 589                 /* more bits in this fsn_mark than the inode's mask? */
 590                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 591                 /* more bits in this fsn_mark than the group? */
 592                 int do_group = (new_mask & ~group->mask);
 593
 594                 /* update the inode with this new fsn_mark */
 595                 if (dropped || do_inode)
 596                         fsnotify_recalc_inode_mask(inode);
 597
 598                 /* update the group mask with the new mask */
 599                 if (dropped || do_group)
 600                         fsnotify_recalc_group_mask(group);
 601         }
 602
 603         /* return the wd */
 604         ret = i_mark->wd;
 605
 606         /* match the get from fsnotify_find_mark() */
 607         fsnotify_put_mark(fsn_mark);
 608
 609         return ret;
 610 }
 611
 612 static int inotify_new_watch(struct fsnotify_group *group,
 613                              struct inode *inode,
 614                              u32 arg)
 615 {
 616         struct inotify_inode_mark *tmp_i_mark;
 617         __u32 mask;
 618         int ret;
 619         struct idr *idr = &group->inotify_data.idr;
 620         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 621
 622         /* don't allow invalid bits: we don't want flags set */
 623         mask = inotify_arg_to_mask(arg);
 624         if (unlikely(!mask))
 625                 return -EINVAL;
 626
 627         tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 628         if (unlikely(!tmp_i_mark))
 629                 return -ENOMEM;
 630
 631         fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
 632         tmp_i_mark->fsn_mark.mask = mask;
 633         tmp_i_mark->wd = -1;
 634
 635         ret = -ENOSPC;
 636         if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 637                 goto out_err;
 638
 639         ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd,
 640                                  tmp_i_mark);
 641         if (ret)
 642                 goto out_err;
 643
 644         /* we are on the idr, now get on the inode */
 645         ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0);
 646         if (ret) {
 647                 /* we failed to get on the inode, get off the idr */
 648                 inotify_remove_from_idr(group, tmp_i_mark);
 649                 goto out_err;
 650         }
 651
 652         /* increment the number of watches the user has */
 653         atomic_inc(&group->inotify_data.user->inotify_watches);
 654
 655         /* return the watch descriptor for this new mark */
 656         ret = tmp_i_mark->wd;
 657
 658         /* if this mark added a new event update the group mask */
 659         if (mask & ~group->mask)
 660                 fsnotify_recalc_group_mask(group);
 661
 662 out_err:
 663         /* match the ref from fsnotify_init_mark() */
 664         fsnotify_put_mark(&tmp_i_mark->fsn_mark);
 665
 666         return ret;
 667 }
 668
 669 static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg)
 670 {
 671         int ret = 0;
 672
 673 retry:
 674         /* try to update and existing watch with the new arg */
 675         ret = inotify_update_existing_watch(group, inode, arg);
 676         /* no mark present, try to add a new one */
 677         if (ret == -ENOENT)
 678                 ret = inotify_new_watch(group, inode, arg);
 679         /*
 680          * inotify_new_watch could race with another thread which did an
 681          * inotify_new_watch between the update_existing and the add watch
 682          * here, go back and try to update an existing mark again.
 683          */
 684         if (ret == -EEXIST)
 685                 goto retry;
 686
 687         return ret;
 688 }
 689
 690 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 691 {
 692         struct fsnotify_group *group;
 693
 694         group = fsnotify_alloc_group(&inotify_fsnotify_ops);
 695         if (IS_ERR(group))
 696                 return group;
 697
 698         group->max_events = max_events;
 699
 700         spin_lock_init(&group->inotify_data.idr_lock);
 701         idr_init(&group->inotify_data.idr);
 702         group->inotify_data.last_wd = 0;
 703         group->inotify_data.user = user;
 704         group->inotify_data.fa = NULL;
 705
 706         return group;
 707 }
 708
 709
 710 /* inotify syscalls */
 711 SYSCALL_DEFINE1(inotify_init1, int, flags)
 712 {
 713         struct fsnotify_group *group;
 714         struct user_struct *user;
 715         int ret;
 716
 717         /* Check the IN_* constants for consistency.  */
 718         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 719         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 720
 721         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 722                 return -EINVAL;
 723
 724         user = get_current_user();
 725         if (unlikely(atomic_read(&user->inotify_devs) >=
 726                         inotify_max_user_instances)) {
 727                 ret = -EMFILE;
 728                 goto out_free_uid;
 729         }
 730
 731         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 732         group = inotify_new_group(user, inotify_max_queued_events);
 733         if (IS_ERR(group)) {
 734                 ret = PTR_ERR(group);
 735                 goto out_free_uid;
 736         }
 737
 738         atomic_inc(&user->inotify_devs);
 739
 740         ret = anon_inode_getfd("inotify", &inotify_fops, group,
 741                                   O_RDONLY | flags);
 742         if (ret >= 0)
 743                 return ret;
 744
 745         atomic_dec(&user->inotify_devs);
 746 out_free_uid:
 747         free_uid(user);
 748         return ret;
 749 }
 750
 751 SYSCALL_DEFINE0(inotify_init)
 752 {
 753         return sys_inotify_init1(0);
 754 }
 755
 756 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 757                 u32, mask)
 758 {
 759         struct fsnotify_group *group;
 760         struct inode *inode;
 761         struct path path;
 762         struct file *filp;
 763         int ret, fput_needed;
 764         unsigned flags = 0;
 765
 766         filp = fget_light(fd, &fput_needed);
 767         if (unlikely(!filp))
 768                 return -EBADF;
 769
 770         /* verify that this is indeed an inotify instance */
 771         if (unlikely(filp->f_op != &inotify_fops)) {
 772                 ret = -EINVAL;
 773                 goto fput_and_out;
 774         }
 775
 776         if (!(mask & IN_DONT_FOLLOW))
 777                 flags |= LOOKUP_FOLLOW;
 778         if (mask & IN_ONLYDIR)
 779                 flags |= LOOKUP_DIRECTORY;
 780
 781         ret = inotify_find_inode(pathname, &path, flags);
 782         if (ret)
 783                 goto fput_and_out;
 784
 785         /* inode held in place by reference to path; group by fget on fd */
 786         inode = path.dentry->d_inode;
 787         group = filp->private_data;
 788
 789         /* create/update an inode mark */
 790         ret = inotify_update_watch(group, inode, mask);
 791         path_put(&path);
 792 fput_and_out:
 793         fput_light(filp, fput_needed);
 794         return ret;
 795 }
 796
 797 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 798 {
 799         struct fsnotify_group *group;
 800         struct inotify_inode_mark *i_mark;
 801         struct file *filp;
 802         int ret = 0, fput_needed;
 803
 804         filp = fget_light(fd, &fput_needed);
 805         if (unlikely(!filp))
 806                 return -EBADF;
 807
 808         /* verify that this is indeed an inotify instance */
 809         ret = -EINVAL;
 810         if (unlikely(filp->f_op != &inotify_fops))
 811                 goto out;
 812
 813         group = filp->private_data;
 814
 815         ret = -EINVAL;
 816         i_mark = inotify_idr_find(group, wd);
 817         if (unlikely(!i_mark))
 818                 goto out;
 819
 820         ret = 0;
 821
 822         fsnotify_destroy_mark(&i_mark->fsn_mark);
 823
 824         /* match ref taken by inotify_idr_find */
 825         fsnotify_put_mark(&i_mark->fsn_mark);
 826
 827 out:
 828         fput_light(filp, fput_needed);
 829         return ret;
 830 }
 831
 832 /*
 833  * inotify_user_setup - Our initialization function.  Note that we cannnot return
 834  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 835  * must result in panic().
 836  */
 837 static int __init inotify_user_setup(void)
 838 {
 839         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
 840         event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 841
 842         inotify_max_queued_events = 16384;
 843         inotify_max_user_instances = 128;
 844         inotify_max_user_watches = 8192;
 845
 846         return 0;
 847 }
 848 module_init(inotify_user_setup);