root/fs/notify/inotify/inotify_user.c

DEFINITIONS

1. inotify_arg_to_mask
2. inotify_mask_to_arg
3. inotify_poll
4. round_event_name_len
5. get_one_event
6. copy_event_to_user
7. inotify_read
8. inotify_release
9. inotify_ioctl
10. inotify_find_inode
11. inotify_add_to_idr
12. inotify_idr_find_locked
13. inotify_idr_find
14. inotify_remove_from_idr
15. inotify_ignored_and_remove_idr
16. inotify_update_existing_watch
17. inotify_new_watch
18. inotify_update_watch
19. inotify_new_group
20. do_inotify_init
21. SYSCALL_DEFINE1
22. SYSCALL_DEFINE0
23. SYSCALL_DEFINE3
24. SYSCALL_DEFINE2
25. inotify_user_setup

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * fs/inotify_user.c - inotify support for userspace
   4  *
   5  * Authors:
   6  *      John McCutchan  <ttb@tentacle.dhs.org>
   7  *      Robert Love     <rml@novell.com>
   8  *
   9  * Copyright (C) 2005 John McCutchan
  10  * Copyright 2006 Hewlett-Packard Development Company, L.P.
  11  *
  12  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
  13  * inotify was largely rewriten to make use of the fsnotify infrastructure
  14  */
  15 
  16 #include <linux/file.h>
  17 #include <linux/fs.h> /* struct inode */
  18 #include <linux/fsnotify_backend.h>
  19 #include <linux/idr.h>
  20 #include <linux/init.h> /* fs_initcall */
  21 #include <linux/inotify.h>
  22 #include <linux/kernel.h> /* roundup() */
  23 #include <linux/namei.h> /* LOOKUP_FOLLOW */
  24 #include <linux/sched/signal.h>
  25 #include <linux/slab.h> /* struct kmem_cache */
  26 #include <linux/syscalls.h>
  27 #include <linux/types.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/uaccess.h>
  30 #include <linux/poll.h>
  31 #include <linux/wait.h>
  32 #include <linux/memcontrol.h>
  33 #include <linux/security.h>
  34 
  35 #include "inotify.h"
  36 #include "../fdinfo.h"
  37 
  38 #include <asm/ioctls.h>
  39 
  40 /* configurable via /proc/sys/fs/inotify/ */
  41 static int inotify_max_queued_events __read_mostly;
  42 
  43 struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
  44 
  45 #ifdef CONFIG_SYSCTL
  46 
  47 #include <linux/sysctl.h>
  48 
  49 struct ctl_table inotify_table[] = {
  50         {
  51                 .procname       = "max_user_instances",
  52                 .data           = &init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES],
  53                 .maxlen         = sizeof(int),
  54                 .mode           = 0644,
  55                 .proc_handler   = proc_dointvec_minmax,
  56                 .extra1         = SYSCTL_ZERO,
  57         },
  58         {
  59                 .procname       = "max_user_watches",
  60                 .data           = &init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES],
  61                 .maxlen         = sizeof(int),
  62                 .mode           = 0644,
  63                 .proc_handler   = proc_dointvec_minmax,
  64                 .extra1         = SYSCTL_ZERO,
  65         },
  66         {
  67                 .procname       = "max_queued_events",
  68                 .data           = &inotify_max_queued_events,
  69                 .maxlen         = sizeof(int),
  70                 .mode           = 0644,
  71                 .proc_handler   = proc_dointvec_minmax,
  72                 .extra1         = SYSCTL_ZERO
  73         },
  74         { }
  75 };
  76 #endif /* CONFIG_SYSCTL */
  77 
  78 static inline __u32 inotify_arg_to_mask(u32 arg)
  79 {
  80         __u32 mask;
  81 
  82         /*
  83          * everything should accept their own ignored, cares about children,
  84          * and should receive events when the inode is unmounted
  85          */
  86         mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
  87 
  88         /* mask off the flags used to open the fd */
  89         mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
  90 
  91         return mask;
  92 }
  93 
  94 static inline u32 inotify_mask_to_arg(__u32 mask)
  95 {
  96         return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED |
  97                        IN_Q_OVERFLOW);
  98 }
  99 
 100 /* intofiy userspace file descriptor functions */
 101 static __poll_t inotify_poll(struct file *file, poll_table *wait)
 102 {
 103         struct fsnotify_group *group = file->private_data;
 104         __poll_t ret = 0;
 105 
 106         poll_wait(file, &group->notification_waitq, wait);
 107         spin_lock(&group->notification_lock);
 108         if (!fsnotify_notify_queue_is_empty(group))
 109                 ret = EPOLLIN | EPOLLRDNORM;
 110         spin_unlock(&group->notification_lock);
 111 
 112         return ret;
 113 }
 114 
 115 static int round_event_name_len(struct fsnotify_event *fsn_event)
 116 {
 117         struct inotify_event_info *event;
 118 
 119         event = INOTIFY_E(fsn_event);
 120         if (!event->name_len)
 121                 return 0;
 122         return roundup(event->name_len + 1, sizeof(struct inotify_event));
 123 }
 124 
 125 /*
 126  * Get an inotify_kernel_event if one exists and is small
 127  * enough to fit in "count". Return an error pointer if
 128  * not large enough.
 129  *
 130  * Called with the group->notification_lock held.
 131  */
 132 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 133                                             size_t count)
 134 {
 135         size_t event_size = sizeof(struct inotify_event);
 136         struct fsnotify_event *event;
 137 
 138         if (fsnotify_notify_queue_is_empty(group))
 139                 return NULL;
 140 
 141         event = fsnotify_peek_first_event(group);
 142 
 143         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 144 
 145         event_size += round_event_name_len(event);
 146         if (event_size > count)
 147                 return ERR_PTR(-EINVAL);
 148 
 149         /* held the notification_lock the whole time, so this is the
 150          * same event we peeked above */
 151         fsnotify_remove_first_event(group);
 152 
 153         return event;
 154 }
 155 
 156 /*
 157  * Copy an event to user space, returning how much we copied.
 158  *
 159  * We already checked that the event size is smaller than the
 160  * buffer we had in "get_one_event()" above.
 161  */
 162 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 163                                   struct fsnotify_event *fsn_event,
 164                                   char __user *buf)
 165 {
 166         struct inotify_event inotify_event;
 167         struct inotify_event_info *event;
 168         size_t event_size = sizeof(struct inotify_event);
 169         size_t name_len;
 170         size_t pad_name_len;
 171 
 172         pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
 173 
 174         event = INOTIFY_E(fsn_event);
 175         name_len = event->name_len;
 176         /*
 177          * round up name length so it is a multiple of event_size
 178          * plus an extra byte for the terminating '\0'.
 179          */
 180         pad_name_len = round_event_name_len(fsn_event);
 181         inotify_event.len = pad_name_len;
 182         inotify_event.mask = inotify_mask_to_arg(event->mask);
 183         inotify_event.wd = event->wd;
 184         inotify_event.cookie = event->sync_cookie;
 185 
 186         /* send the main event */
 187         if (copy_to_user(buf, &inotify_event, event_size))
 188                 return -EFAULT;
 189 
 190         buf += event_size;
 191 
 192         /*
 193          * fsnotify only stores the pathname, so here we have to send the pathname
 194          * and then pad that pathname out to a multiple of sizeof(inotify_event)
 195          * with zeros.
 196          */
 197         if (pad_name_len) {
 198                 /* copy the path name */
 199                 if (copy_to_user(buf, event->name, name_len))
 200                         return -EFAULT;
 201                 buf += name_len;
 202 
 203                 /* fill userspace with 0's */
 204                 if (clear_user(buf, pad_name_len - name_len))
 205                         return -EFAULT;
 206                 event_size += pad_name_len;
 207         }
 208 
 209         return event_size;
 210 }
 211 
 212 static ssize_t inotify_read(struct file *file, char __user *buf,
 213                             size_t count, loff_t *pos)
 214 {
 215         struct fsnotify_group *group;
 216         struct fsnotify_event *kevent;
 217         char __user *start;
 218         int ret;
 219         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 220 
 221         start = buf;
 222         group = file->private_data;
 223 
 224         add_wait_queue(&group->notification_waitq, &wait);
 225         while (1) {
 226                 spin_lock(&group->notification_lock);
 227                 kevent = get_one_event(group, count);
 228                 spin_unlock(&group->notification_lock);
 229 
 230                 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
 231 
 232                 if (kevent) {
 233                         ret = PTR_ERR(kevent);
 234                         if (IS_ERR(kevent))
 235                                 break;
 236                         ret = copy_event_to_user(group, kevent, buf);
 237                         fsnotify_destroy_event(group, kevent);
 238                         if (ret < 0)
 239                                 break;
 240                         buf += ret;
 241                         count -= ret;
 242                         continue;
 243                 }
 244 
 245                 ret = -EAGAIN;
 246                 if (file->f_flags & O_NONBLOCK)
 247                         break;
 248                 ret = -ERESTARTSYS;
 249                 if (signal_pending(current))
 250                         break;
 251 
 252                 if (start != buf)
 253                         break;
 254 
 255                 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
 256         }
 257         remove_wait_queue(&group->notification_waitq, &wait);
 258 
 259         if (start != buf && ret != -EFAULT)
 260                 ret = buf - start;
 261         return ret;
 262 }
 263 
 264 static int inotify_release(struct inode *ignored, struct file *file)
 265 {
 266         struct fsnotify_group *group = file->private_data;
 267 
 268         pr_debug("%s: group=%p\n", __func__, group);
 269 
 270         /* free this group, matching get was inotify_init->fsnotify_obtain_group */
 271         fsnotify_destroy_group(group);
 272 
 273         return 0;
 274 }
 275 
 276 static long inotify_ioctl(struct file *file, unsigned int cmd,
 277                           unsigned long arg)
 278 {
 279         struct fsnotify_group *group;
 280         struct fsnotify_event *fsn_event;
 281         void __user *p;
 282         int ret = -ENOTTY;
 283         size_t send_len = 0;
 284 
 285         group = file->private_data;
 286         p = (void __user *) arg;
 287 
 288         pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
 289 
 290         switch (cmd) {
 291         case FIONREAD:
 292                 spin_lock(&group->notification_lock);
 293                 list_for_each_entry(fsn_event, &group->notification_list,
 294                                     list) {
 295                         send_len += sizeof(struct inotify_event);
 296                         send_len += round_event_name_len(fsn_event);
 297                 }
 298                 spin_unlock(&group->notification_lock);
 299                 ret = put_user(send_len, (int __user *) p);
 300                 break;
 301 #ifdef CONFIG_CHECKPOINT_RESTORE
 302         case INOTIFY_IOC_SETNEXTWD:
 303                 ret = -EINVAL;
 304                 if (arg >= 1 && arg <= INT_MAX) {
 305                         struct inotify_group_private_data *data;
 306 
 307                         data = &group->inotify_data;
 308                         spin_lock(&data->idr_lock);
 309                         idr_set_cursor(&data->idr, (unsigned int)arg);
 310                         spin_unlock(&data->idr_lock);
 311                         ret = 0;
 312                 }
 313                 break;
 314 #endif /* CONFIG_CHECKPOINT_RESTORE */
 315         }
 316 
 317         return ret;
 318 }
 319 
 320 static const struct file_operations inotify_fops = {
 321         .show_fdinfo    = inotify_show_fdinfo,
 322         .poll           = inotify_poll,
 323         .read           = inotify_read,
 324         .fasync         = fsnotify_fasync,
 325         .release        = inotify_release,
 326         .unlocked_ioctl = inotify_ioctl,
 327         .compat_ioctl   = inotify_ioctl,
 328         .llseek         = noop_llseek,
 329 };
 330 
 331 
 332 /*
 333  * find_inode - resolve a user-given path to a specific inode
 334  */
 335 static int inotify_find_inode(const char __user *dirname, struct path *path,
 336                                                 unsigned int flags, __u64 mask)
 337 {
 338         int error;
 339 
 340         error = user_path_at(AT_FDCWD, dirname, flags, path);
 341         if (error)
 342                 return error;
 343         /* you can only watch an inode if you have read permissions on it */
 344         error = inode_permission(path->dentry->d_inode, MAY_READ);
 345         if (error) {
 346                 path_put(path);
 347                 return error;
 348         }
 349         error = security_path_notify(path, mask,
 350                                 FSNOTIFY_OBJ_TYPE_INODE);
 351         if (error)
 352                 path_put(path);
 353 
 354         return error;
 355 }
 356 
 357 static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
 358                               struct inotify_inode_mark *i_mark)
 359 {
 360         int ret;
 361 
 362         idr_preload(GFP_KERNEL);
 363         spin_lock(idr_lock);
 364 
 365         ret = idr_alloc_cyclic(idr, i_mark, 1, 0, GFP_NOWAIT);
 366         if (ret >= 0) {
 367                 /* we added the mark to the idr, take a reference */
 368                 i_mark->wd = ret;
 369                 fsnotify_get_mark(&i_mark->fsn_mark);
 370         }
 371 
 372         spin_unlock(idr_lock);
 373         idr_preload_end();
 374         return ret < 0 ? ret : 0;
 375 }
 376 
 377 static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
 378                                                                 int wd)
 379 {
 380         struct idr *idr = &group->inotify_data.idr;
 381         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 382         struct inotify_inode_mark *i_mark;
 383 
 384         assert_spin_locked(idr_lock);
 385 
 386         i_mark = idr_find(idr, wd);
 387         if (i_mark) {
 388                 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
 389 
 390                 fsnotify_get_mark(fsn_mark);
 391                 /* One ref for being in the idr, one ref we just took */
 392                 BUG_ON(refcount_read(&fsn_mark->refcnt) < 2);
 393         }
 394 
 395         return i_mark;
 396 }
 397 
 398 static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
 399                                                          int wd)
 400 {
 401         struct inotify_inode_mark *i_mark;
 402         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 403 
 404         spin_lock(idr_lock);
 405         i_mark = inotify_idr_find_locked(group, wd);
 406         spin_unlock(idr_lock);
 407 
 408         return i_mark;
 409 }
 410 
 411 /*
 412  * Remove the mark from the idr (if present) and drop the reference
 413  * on the mark because it was in the idr.
 414  */
 415 static void inotify_remove_from_idr(struct fsnotify_group *group,
 416                                     struct inotify_inode_mark *i_mark)
 417 {
 418         struct idr *idr = &group->inotify_data.idr;
 419         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 420         struct inotify_inode_mark *found_i_mark = NULL;
 421         int wd;
 422 
 423         spin_lock(idr_lock);
 424         wd = i_mark->wd;
 425 
 426         /*
 427          * does this i_mark think it is in the idr?  we shouldn't get called
 428          * if it wasn't....
 429          */
 430         if (wd == -1) {
 431                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 432                         __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 433                 goto out;
 434         }
 435 
 436         /* Lets look in the idr to see if we find it */
 437         found_i_mark = inotify_idr_find_locked(group, wd);
 438         if (unlikely(!found_i_mark)) {
 439                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 440                         __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 441                 goto out;
 442         }
 443 
 444         /*
 445          * We found an mark in the idr at the right wd, but it's
 446          * not the mark we were told to remove.  eparis seriously
 447          * fucked up somewhere.
 448          */
 449         if (unlikely(found_i_mark != i_mark)) {
 450                 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
 451                         "found_i_mark=%p found_i_mark->wd=%d "
 452                         "found_i_mark->group=%p\n", __func__, i_mark,
 453                         i_mark->wd, i_mark->fsn_mark.group, found_i_mark,
 454                         found_i_mark->wd, found_i_mark->fsn_mark.group);
 455                 goto out;
 456         }
 457 
 458         /*
 459          * One ref for being in the idr
 460          * one ref grabbed by inotify_idr_find
 461          */
 462         if (unlikely(refcount_read(&i_mark->fsn_mark.refcnt) < 2)) {
 463                 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p\n",
 464                          __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group);
 465                 /* we can't really recover with bad ref cnting.. */
 466                 BUG();
 467         }
 468 
 469         idr_remove(idr, wd);
 470         /* Removed from the idr, drop that ref. */
 471         fsnotify_put_mark(&i_mark->fsn_mark);
 472 out:
 473         i_mark->wd = -1;
 474         spin_unlock(idr_lock);
 475         /* match the ref taken by inotify_idr_find_locked() */
 476         if (found_i_mark)
 477                 fsnotify_put_mark(&found_i_mark->fsn_mark);
 478 }
 479 
 480 /*
 481  * Send IN_IGNORED for this wd, remove this wd from the idr.
 482  */
 483 void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
 484                                     struct fsnotify_group *group)
 485 {
 486         struct inotify_inode_mark *i_mark;
 487         struct fsnotify_iter_info iter_info = { };
 488 
 489         fsnotify_iter_set_report_type_mark(&iter_info, FSNOTIFY_OBJ_TYPE_INODE,
 490                                            fsn_mark);
 491 
 492         /* Queue ignore event for the watch */
 493         inotify_handle_event(group, NULL, FS_IN_IGNORED, NULL,
 494                              FSNOTIFY_EVENT_NONE, NULL, 0, &iter_info);
 495 
 496         i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 497         /* remove this mark from the idr */
 498         inotify_remove_from_idr(group, i_mark);
 499 
 500         dec_inotify_watches(group->inotify_data.ucounts);
 501 }
 502 
 503 static int inotify_update_existing_watch(struct fsnotify_group *group,
 504                                          struct inode *inode,
 505                                          u32 arg)
 506 {
 507         struct fsnotify_mark *fsn_mark;
 508         struct inotify_inode_mark *i_mark;
 509         __u32 old_mask, new_mask;
 510         __u32 mask;
 511         int add = (arg & IN_MASK_ADD);
 512         int create = (arg & IN_MASK_CREATE);
 513         int ret;
 514 
 515         mask = inotify_arg_to_mask(arg);
 516 
 517         fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 518         if (!fsn_mark)
 519                 return -ENOENT;
 520         else if (create) {
 521                 ret = -EEXIST;
 522                 goto out;
 523         }
 524 
 525         i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 526 
 527         spin_lock(&fsn_mark->lock);
 528         old_mask = fsn_mark->mask;
 529         if (add)
 530                 fsn_mark->mask |= mask;
 531         else
 532                 fsn_mark->mask = mask;
 533         new_mask = fsn_mark->mask;
 534         spin_unlock(&fsn_mark->lock);
 535 
 536         if (old_mask != new_mask) {
 537                 /* more bits in old than in new? */
 538                 int dropped = (old_mask & ~new_mask);
 539                 /* more bits in this fsn_mark than the inode's mask? */
 540                 int do_inode = (new_mask & ~inode->i_fsnotify_mask);
 541 
 542                 /* update the inode with this new fsn_mark */
 543                 if (dropped || do_inode)
 544                         fsnotify_recalc_mask(inode->i_fsnotify_marks);
 545 
 546         }
 547 
 548         /* return the wd */
 549         ret = i_mark->wd;
 550 
 551 out:
 552         /* match the get from fsnotify_find_mark() */
 553         fsnotify_put_mark(fsn_mark);
 554 
 555         return ret;
 556 }
 557 
 558 static int inotify_new_watch(struct fsnotify_group *group,
 559                              struct inode *inode,
 560                              u32 arg)
 561 {
 562         struct inotify_inode_mark *tmp_i_mark;
 563         __u32 mask;
 564         int ret;
 565         struct idr *idr = &group->inotify_data.idr;
 566         spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 567 
 568         mask = inotify_arg_to_mask(arg);
 569 
 570         tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
 571         if (unlikely(!tmp_i_mark))
 572                 return -ENOMEM;
 573 
 574         fsnotify_init_mark(&tmp_i_mark->fsn_mark, group);
 575         tmp_i_mark->fsn_mark.mask = mask;
 576         tmp_i_mark->wd = -1;
 577 
 578         ret = inotify_add_to_idr(idr, idr_lock, tmp_i_mark);
 579         if (ret)
 580                 goto out_err;
 581 
 582         /* increment the number of watches the user has */
 583         if (!inc_inotify_watches(group->inotify_data.ucounts)) {
 584                 inotify_remove_from_idr(group, tmp_i_mark);
 585                 ret = -ENOSPC;
 586                 goto out_err;
 587         }
 588 
 589         /* we are on the idr, now get on the inode */
 590         ret = fsnotify_add_inode_mark_locked(&tmp_i_mark->fsn_mark, inode, 0);
 591         if (ret) {
 592                 /* we failed to get on the inode, get off the idr */
 593                 inotify_remove_from_idr(group, tmp_i_mark);
 594                 goto out_err;
 595         }
 596 
 597 
 598         /* return the watch descriptor for this new mark */
 599         ret = tmp_i_mark->wd;
 600 
 601 out_err:
 602         /* match the ref from fsnotify_init_mark() */
 603         fsnotify_put_mark(&tmp_i_mark->fsn_mark);
 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         mutex_lock(&group->mark_mutex);
 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         mutex_unlock(&group->mark_mutex);
 619 
 620         return ret;
 621 }
 622 
 623 static struct fsnotify_group *inotify_new_group(unsigned int max_events)
 624 {
 625         struct fsnotify_group *group;
 626         struct inotify_event_info *oevent;
 627 
 628         group = fsnotify_alloc_group(&inotify_fsnotify_ops);
 629         if (IS_ERR(group))
 630                 return group;
 631 
 632         oevent = kmalloc(sizeof(struct inotify_event_info), GFP_KERNEL);
 633         if (unlikely(!oevent)) {
 634                 fsnotify_destroy_group(group);
 635                 return ERR_PTR(-ENOMEM);
 636         }
 637         group->overflow_event = &oevent->fse;
 638         fsnotify_init_event(group->overflow_event, 0);
 639         oevent->mask = FS_Q_OVERFLOW;
 640         oevent->wd = -1;
 641         oevent->sync_cookie = 0;
 642         oevent->name_len = 0;
 643 
 644         group->max_events = max_events;
 645         group->memcg = get_mem_cgroup_from_mm(current->mm);
 646 
 647         spin_lock_init(&group->inotify_data.idr_lock);
 648         idr_init(&group->inotify_data.idr);
 649         group->inotify_data.ucounts = inc_ucount(current_user_ns(),
 650                                                  current_euid(),
 651                                                  UCOUNT_INOTIFY_INSTANCES);
 652 
 653         if (!group->inotify_data.ucounts) {
 654                 fsnotify_destroy_group(group);
 655                 return ERR_PTR(-EMFILE);
 656         }
 657 
 658         return group;
 659 }
 660 
 661 
 662 /* inotify syscalls */
 663 static int do_inotify_init(int flags)
 664 {
 665         struct fsnotify_group *group;
 666         int ret;
 667 
 668         /* Check the IN_* constants for consistency.  */
 669         BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC);
 670         BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK);
 671 
 672         if (flags & ~(IN_CLOEXEC | IN_NONBLOCK))
 673                 return -EINVAL;
 674 
 675         /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
 676         group = inotify_new_group(inotify_max_queued_events);
 677         if (IS_ERR(group))
 678                 return PTR_ERR(group);
 679 
 680         ret = anon_inode_getfd("inotify", &inotify_fops, group,
 681                                   O_RDONLY | flags);
 682         if (ret < 0)
 683                 fsnotify_destroy_group(group);
 684 
 685         return ret;
 686 }
 687 
 688 SYSCALL_DEFINE1(inotify_init1, int, flags)
 689 {
 690         return do_inotify_init(flags);
 691 }
 692 
 693 SYSCALL_DEFINE0(inotify_init)
 694 {
 695         return do_inotify_init(0);
 696 }
 697 
 698 SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 699                 u32, mask)
 700 {
 701         struct fsnotify_group *group;
 702         struct inode *inode;
 703         struct path path;
 704         struct fd f;
 705         int ret;
 706         unsigned flags = 0;
 707 
 708         /*
 709          * We share a lot of code with fs/dnotify.  We also share
 710          * the bit layout between inotify's IN_* and the fsnotify
 711          * FS_*.  This check ensures that only the inotify IN_*
 712          * bits get passed in and set in watches/events.
 713          */
 714         if (unlikely(mask & ~ALL_INOTIFY_BITS))
 715                 return -EINVAL;
 716         /*
 717          * Require at least one valid bit set in the mask.
 718          * Without _something_ set, we would have no events to
 719          * watch for.
 720          */
 721         if (unlikely(!(mask & ALL_INOTIFY_BITS)))
 722                 return -EINVAL;
 723 
 724         f = fdget(fd);
 725         if (unlikely(!f.file))
 726                 return -EBADF;
 727 
 728         /* IN_MASK_ADD and IN_MASK_CREATE don't make sense together */
 729         if (unlikely((mask & IN_MASK_ADD) && (mask & IN_MASK_CREATE))) {
 730                 ret = -EINVAL;
 731                 goto fput_and_out;
 732         }
 733 
 734         /* verify that this is indeed an inotify instance */
 735         if (unlikely(f.file->f_op != &inotify_fops)) {
 736                 ret = -EINVAL;
 737                 goto fput_and_out;
 738         }
 739 
 740         if (!(mask & IN_DONT_FOLLOW))
 741                 flags |= LOOKUP_FOLLOW;
 742         if (mask & IN_ONLYDIR)
 743                 flags |= LOOKUP_DIRECTORY;
 744 
 745         ret = inotify_find_inode(pathname, &path, flags,
 746                         (mask & IN_ALL_EVENTS));
 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 = f.file->private_data;
 753 
 754         /* create/update an inode mark */
 755         ret = inotify_update_watch(group, inode, mask);
 756         path_put(&path);
 757 fput_and_out:
 758         fdput(f);
 759         return ret;
 760 }
 761 
 762 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 763 {
 764         struct fsnotify_group *group;
 765         struct inotify_inode_mark *i_mark;
 766         struct fd f;
 767         int ret = 0;
 768 
 769         f = fdget(fd);
 770         if (unlikely(!f.file))
 771                 return -EBADF;
 772 
 773         /* verify that this is indeed an inotify instance */
 774         ret = -EINVAL;
 775         if (unlikely(f.file->f_op != &inotify_fops))
 776                 goto out;
 777 
 778         group = f.file->private_data;
 779 
 780         ret = -EINVAL;
 781         i_mark = inotify_idr_find(group, wd);
 782         if (unlikely(!i_mark))
 783                 goto out;
 784 
 785         ret = 0;
 786 
 787         fsnotify_destroy_mark(&i_mark->fsn_mark, group);
 788 
 789         /* match ref taken by inotify_idr_find */
 790         fsnotify_put_mark(&i_mark->fsn_mark);
 791 
 792 out:
 793         fdput(f);
 794         return ret;
 795 }
 796 
 797 /*
 798  * inotify_user_setup - Our initialization function.  Note that we cannot return
 799  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 800  * must result in panic().
 801  */
 802 static int __init inotify_user_setup(void)
 803 {
 804         BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
 805         BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
 806         BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
 807         BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
 808         BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
 809         BUILD_BUG_ON(IN_OPEN != FS_OPEN);
 810         BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
 811         BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
 812         BUILD_BUG_ON(IN_CREATE != FS_CREATE);
 813         BUILD_BUG_ON(IN_DELETE != FS_DELETE);
 814         BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
 815         BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
 816         BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
 817         BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
 818         BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
 819         BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
 820         BUILD_BUG_ON(IN_ISDIR != FS_ISDIR);
 821         BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
 822 
 823         BUILD_BUG_ON(HWEIGHT32(ALL_INOTIFY_BITS) != 22);
 824 
 825         inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark,
 826                                                SLAB_PANIC|SLAB_ACCOUNT);
 827 
 828         inotify_max_queued_events = 16384;
 829         init_user_ns.ucount_max[UCOUNT_INOTIFY_INSTANCES] = 128;
 830         init_user_ns.ucount_max[UCOUNT_INOTIFY_WATCHES] = 8192;
 831 
 832         return 0;
 833 }
 834 fs_initcall(inotify_user_setup);