fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/compat.h>
  17
  18 #include <asm/ioctls.h>
  19
  20 #include "../../mount.h"
  21 #include "../fdinfo.h"
  22 #include "fanotify.h"
  23
  24 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  25 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  26 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  27
  28 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  29
  30 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  31 struct kmem_cache *fanotify_event_cachep __read_mostly;
  32 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
  33
  34 /*
  35  * Get an fsnotify notification event if one exists and is small
  36  * enough to fit in "count". Return an error pointer if the count
  37  * is not large enough.
  38  *
  39  * Called with the group->notification_mutex held.
  40  */
  41 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  42                                             size_t count)
  43 {
  44         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  45
  46         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  47
  48         if (fsnotify_notify_queue_is_empty(group))
  49                 return NULL;
  50
  51         if (FAN_EVENT_METADATA_LEN > count)
  52                 return ERR_PTR(-EINVAL);
  53
  54         /* held the notification_mutex the whole time, so this is the
  55          * same event we peeked above */
  56         return fsnotify_remove_notify_event(group);
  57 }
  58
  59 static int create_fd(struct fsnotify_group *group,
  60                      struct fanotify_event_info *event,
  61                      struct file **file)
  62 {
  63         int client_fd;
  64         struct file *new_file;
  65
  66         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  67
  68         client_fd = get_unused_fd();
  69         if (client_fd < 0)
  70                 return client_fd;
  71
  72         /*
  73          * we need a new file handle for the userspace program so it can read even if it was
  74          * originally opened O_WRONLY.
  75          */
  76         /* it's possible this event was an overflow event.  in that case dentry and mnt
  77          * are NULL;  That's fine, just don't call dentry open */
  78         if (event->path.dentry && event->path.mnt)
  79                 new_file = dentry_open(&event->path,
  80                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  81                                        current_cred());
  82         else
  83                 new_file = ERR_PTR(-EOVERFLOW);
  84         if (IS_ERR(new_file)) {
  85                 /*
  86                  * we still send an event even if we can't open the file.  this
  87                  * can happen when say tasks are gone and we try to open their
  88                  * /proc files or we try to open a WRONLY file like in sysfs
  89                  * we just send the errno to userspace since there isn't much
  90                  * else we can do.
  91                  */
  92                 put_unused_fd(client_fd);
  93                 client_fd = PTR_ERR(new_file);
  94         } else {
  95                 *file = new_file;
  96         }
  97
  98         return client_fd;
  99 }
 100
 101 static int fill_event_metadata(struct fsnotify_group *group,
 102                                struct fanotify_event_metadata *metadata,
 103                                struct fsnotify_event *fsn_event,
 104                                struct file **file)
 105 {
 106         int ret = 0;
 107         struct fanotify_event_info *event;
 108
 109         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 110                  group, metadata, fsn_event);
 111
 112         *file = NULL;
 113         event = container_of(fsn_event, struct fanotify_event_info, fse);
 114         metadata->event_len = FAN_EVENT_METADATA_LEN;
 115         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 116         metadata->vers = FANOTIFY_METADATA_VERSION;
 117         metadata->reserved = 0;
 118         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 119         metadata->pid = pid_vnr(event->tgid);
 120         if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
 121                 metadata->fd = FAN_NOFD;
 122         else {
 123                 metadata->fd = create_fd(group, event, file);
 124                 if (metadata->fd < 0)
 125                         ret = metadata->fd;
 126         }
 127
 128         return ret;
 129 }
 130
 131 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 132 static struct fanotify_perm_event_info *dequeue_event(
 133                                 struct fsnotify_group *group, int fd)
 134 {
 135         struct fanotify_perm_event_info *event, *return_e = NULL;
 136
 137         spin_lock(&group->fanotify_data.access_lock);
 138         list_for_each_entry(event, &group->fanotify_data.access_list,
 139                             fae.fse.list) {
 140                 if (event->fd != fd)
 141                         continue;
 142
 143                 list_del_init(&event->fae.fse.list);
 144                 return_e = event;
 145                 break;
 146         }
 147         spin_unlock(&group->fanotify_data.access_lock);
 148
 149         pr_debug("%s: found return_re=%p\n", __func__, return_e);
 150
 151         return return_e;
 152 }
 153
 154 static int process_access_response(struct fsnotify_group *group,
 155                                    struct fanotify_response *response_struct)
 156 {
 157         struct fanotify_perm_event_info *event;
 158         int fd = response_struct->fd;
 159         int response = response_struct->response;
 160
 161         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 162                  fd, response);
 163         /*
 164          * make sure the response is valid, if invalid we do nothing and either
 165          * userspace can send a valid response or we will clean it up after the
 166          * timeout
 167          */
 168         switch (response) {
 169         case FAN_ALLOW:
 170         case FAN_DENY:
 171                 break;
 172         default:
 173                 return -EINVAL;
 174         }
 175
 176         if (fd < 0)
 177                 return -EINVAL;
 178
 179         event = dequeue_event(group, fd);
 180         if (!event)
 181                 return -ENOENT;
 182
 183         event->response = response;
 184         wake_up(&group->fanotify_data.access_waitq);
 185
 186         return 0;
 187 }
 188 #endif
 189
 190 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 191                                   struct fsnotify_event *event,
 192                                   char __user *buf)
 193 {
 194         struct fanotify_event_metadata fanotify_event_metadata;
 195         struct file *f;
 196         int fd, ret;
 197
 198         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 199
 200         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 201         if (ret < 0)
 202                 return ret;
 203
 204         fd = fanotify_event_metadata.fd;
 205         ret = -EFAULT;
 206         if (copy_to_user(buf, &fanotify_event_metadata,
 207                          fanotify_event_metadata.event_len))
 208                 goto out_close_fd;
 209
 210 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 211         if (event->mask & FAN_ALL_PERM_EVENTS)
 212                 FANOTIFY_PE(event)->fd = fd;
 213 #endif
 214
 215         if (fd != FAN_NOFD)
 216                 fd_install(fd, f);
 217         return fanotify_event_metadata.event_len;
 218
 219 out_close_fd:
 220         if (fd != FAN_NOFD) {
 221                 put_unused_fd(fd);
 222                 fput(f);
 223         }
 224         return ret;
 225 }
 226
 227 /* intofiy userspace file descriptor functions */
 228 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 229 {
 230         struct fsnotify_group *group = file->private_data;
 231         int ret = 0;
 232
 233         poll_wait(file, &group->notification_waitq, wait);
 234         mutex_lock(&group->notification_mutex);
 235         if (!fsnotify_notify_queue_is_empty(group))
 236                 ret = POLLIN | POLLRDNORM;
 237         mutex_unlock(&group->notification_mutex);
 238
 239         return ret;
 240 }
 241
 242 static ssize_t fanotify_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         pr_debug("%s: group=%p\n", __func__, group);
 255
 256         while (1) {
 257                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 258
 259                 mutex_lock(&group->notification_mutex);
 260                 kevent = get_one_event(group, count);
 261                 mutex_unlock(&group->notification_mutex);
 262
 263                 if (IS_ERR(kevent)) {
 264                         ret = PTR_ERR(kevent);
 265                         break;
 266                 }
 267
 268                 if (!kevent) {
 269                         ret = -EAGAIN;
 270                         if (file->f_flags & O_NONBLOCK)
 271                                 break;
 272
 273                         ret = -ERESTARTSYS;
 274                         if (signal_pending(current))
 275                                 break;
 276
 277                         if (start != buf)
 278                                 break;
 279                         schedule();
 280                         continue;
 281                 }
 282
 283                 ret = copy_event_to_user(group, kevent, buf);
 284                 /*
 285                  * Permission events get queued to wait for response.  Other
 286                  * events can be destroyed now.
 287                  */
 288                 if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
 289                         fsnotify_destroy_event(group, kevent);
 290                         if (ret < 0)
 291                                 break;
 292                 } else {
 293 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 294                         if (ret < 0) {
 295                                 FANOTIFY_PE(kevent)->response = FAN_DENY;
 296                                 wake_up(&group->fanotify_data.access_waitq);
 297                                 break;
 298                         }
 299                         spin_lock(&group->fanotify_data.access_lock);
 300                         list_add_tail(&kevent->list,
 301                                       &group->fanotify_data.access_list);
 302                         spin_unlock(&group->fanotify_data.access_lock);
 303 #endif
 304                 }
 305                 buf += ret;
 306                 count -= ret;
 307         }
 308
 309         finish_wait(&group->notification_waitq, &wait);
 310         if (start != buf && ret != -EFAULT)
 311                 ret = buf - start;
 312         return ret;
 313 }
 314
 315 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 316 {
 317 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 318         struct fanotify_response response = { .fd = -1, .response = -1 };
 319         struct fsnotify_group *group;
 320         int ret;
 321
 322         group = file->private_data;
 323
 324         if (count > sizeof(response))
 325                 count = sizeof(response);
 326
 327         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 328
 329         if (copy_from_user(&response, buf, count))
 330                 return -EFAULT;
 331
 332         ret = process_access_response(group, &response);
 333         if (ret < 0)
 334                 count = ret;
 335
 336         return count;
 337 #else
 338         return -EINVAL;
 339 #endif
 340 }
 341
 342 static int fanotify_release(struct inode *ignored, struct file *file)
 343 {
 344         struct fsnotify_group *group = file->private_data;
 345
 346 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 347         struct fanotify_perm_event_info *event, *next;
 348
 349         spin_lock(&group->fanotify_data.access_lock);
 350
 351         atomic_inc(&group->fanotify_data.bypass_perm);
 352
 353         list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
 354                                  fae.fse.list) {
 355                 pr_debug("%s: found group=%p event=%p\n", __func__, group,
 356                          event);
 357
 358                 list_del_init(&event->fae.fse.list);
 359                 event->response = FAN_ALLOW;
 360         }
 361         spin_unlock(&group->fanotify_data.access_lock);
 362
 363         wake_up(&group->fanotify_data.access_waitq);
 364 #endif
 365
 366         /* matches the fanotify_init->fsnotify_alloc_group */
 367         fsnotify_destroy_group(group);
 368
 369         return 0;
 370 }
 371
 372 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 373 {
 374         struct fsnotify_group *group;
 375         struct fsnotify_event *fsn_event;
 376         void __user *p;
 377         int ret = -ENOTTY;
 378         size_t send_len = 0;
 379
 380         group = file->private_data;
 381
 382         p = (void __user *) arg;
 383
 384         switch (cmd) {
 385         case FIONREAD:
 386                 mutex_lock(&group->notification_mutex);
 387                 list_for_each_entry(fsn_event, &group->notification_list, list)
 388                         send_len += FAN_EVENT_METADATA_LEN;
 389                 mutex_unlock(&group->notification_mutex);
 390                 ret = put_user(send_len, (int __user *) p);
 391                 break;
 392         }
 393
 394         return ret;
 395 }
 396
 397 static const struct file_operations fanotify_fops = {
 398         .show_fdinfo    = fanotify_show_fdinfo,
 399         .poll           = fanotify_poll,
 400         .read           = fanotify_read,
 401         .write          = fanotify_write,
 402         .fasync         = NULL,
 403         .release        = fanotify_release,
 404         .unlocked_ioctl = fanotify_ioctl,
 405         .compat_ioctl   = fanotify_ioctl,
 406         .llseek         = noop_llseek,
 407 };
 408
 409 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 410 {
 411         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 412 }
 413
 414 static int fanotify_find_path(int dfd, const char __user *filename,
 415                               struct path *path, unsigned int flags)
 416 {
 417         int ret;
 418
 419         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 420                  dfd, filename, flags);
 421
 422         if (filename == NULL) {
 423                 struct fd f = fdget(dfd);
 424
 425                 ret = -EBADF;
 426                 if (!f.file)
 427                         goto out;
 428
 429                 ret = -ENOTDIR;
 430                 if ((flags & FAN_MARK_ONLYDIR) &&
 431                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 432                         fdput(f);
 433                         goto out;
 434                 }
 435
 436                 *path = f.file->f_path;
 437                 path_get(path);
 438                 fdput(f);
 439         } else {
 440                 unsigned int lookup_flags = 0;
 441
 442                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 443                         lookup_flags |= LOOKUP_FOLLOW;
 444                 if (flags & FAN_MARK_ONLYDIR)
 445                         lookup_flags |= LOOKUP_DIRECTORY;
 446
 447                 ret = user_path_at(dfd, filename, lookup_flags, path);
 448                 if (ret)
 449                         goto out;
 450         }
 451
 452         /* you can only watch an inode if you have read permissions on it */
 453         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 454         if (ret)
 455                 path_put(path);
 456 out:
 457         return ret;
 458 }
 459
 460 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 461                                             __u32 mask,
 462                                             unsigned int flags,
 463                                             int *destroy)
 464 {
 465         __u32 oldmask;
 466
 467         spin_lock(&fsn_mark->lock);
 468         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 469                 oldmask = fsn_mark->mask;
 470                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 471         } else {
 472                 oldmask = fsn_mark->ignored_mask;
 473                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 474         }
 475         spin_unlock(&fsn_mark->lock);
 476
 477         *destroy = !(oldmask & ~mask);
 478
 479         return mask & oldmask;
 480 }
 481
 482 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 483                                          struct vfsmount *mnt, __u32 mask,
 484                                          unsigned int flags)
 485 {
 486         struct fsnotify_mark *fsn_mark = NULL;
 487         __u32 removed;
 488         int destroy_mark;
 489
 490         mutex_lock(&group->mark_mutex);
 491         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 492         if (!fsn_mark) {
 493                 mutex_unlock(&group->mark_mutex);
 494                 return -ENOENT;
 495         }
 496
 497         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 498                                                  &destroy_mark);
 499         if (destroy_mark)
 500                 fsnotify_destroy_mark_locked(fsn_mark, group);
 501         mutex_unlock(&group->mark_mutex);
 502
 503         fsnotify_put_mark(fsn_mark);
 504         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 505                 fsnotify_recalc_vfsmount_mask(mnt);
 506
 507         return 0;
 508 }
 509
 510 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 511                                       struct inode *inode, __u32 mask,
 512                                       unsigned int flags)
 513 {
 514         struct fsnotify_mark *fsn_mark = NULL;
 515         __u32 removed;
 516         int destroy_mark;
 517
 518         mutex_lock(&group->mark_mutex);
 519         fsn_mark = fsnotify_find_inode_mark(group, inode);
 520         if (!fsn_mark) {
 521                 mutex_unlock(&group->mark_mutex);
 522                 return -ENOENT;
 523         }
 524
 525         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 526                                                  &destroy_mark);
 527         if (destroy_mark)
 528                 fsnotify_destroy_mark_locked(fsn_mark, group);
 529         mutex_unlock(&group->mark_mutex);
 530
 531         /* matches the fsnotify_find_inode_mark() */
 532         fsnotify_put_mark(fsn_mark);
 533         if (removed & inode->i_fsnotify_mask)
 534                 fsnotify_recalc_inode_mask(inode);
 535
 536         return 0;
 537 }
 538
 539 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 540                                        __u32 mask,
 541                                        unsigned int flags)
 542 {
 543         __u32 oldmask = -1;
 544
 545         spin_lock(&fsn_mark->lock);
 546         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 547                 oldmask = fsn_mark->mask;
 548                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 549         } else {
 550                 __u32 tmask = fsn_mark->ignored_mask | mask;
 551                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 552                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 553                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 554         }
 555
 556         if (!(flags & FAN_MARK_ONDIR)) {
 557                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 558                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 559         }
 560
 561         spin_unlock(&fsn_mark->lock);
 562
 563         return mask & ~oldmask;
 564 }
 565
 566 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 567                                                    struct inode *inode,
 568                                                    struct vfsmount *mnt)
 569 {
 570         struct fsnotify_mark *mark;
 571         int ret;
 572
 573         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 574                 return ERR_PTR(-ENOSPC);
 575
 576         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 577         if (!mark)
 578                 return ERR_PTR(-ENOMEM);
 579
 580         fsnotify_init_mark(mark, fanotify_free_mark);
 581         ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
 582         if (ret) {
 583                 fsnotify_put_mark(mark);
 584                 return ERR_PTR(ret);
 585         }
 586
 587         return mark;
 588 }
 589
 590
 591 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 592                                       struct vfsmount *mnt, __u32 mask,
 593                                       unsigned int flags)
 594 {
 595         struct fsnotify_mark *fsn_mark;
 596         __u32 added;
 597
 598         mutex_lock(&group->mark_mutex);
 599         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 600         if (!fsn_mark) {
 601                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 602                 if (IS_ERR(fsn_mark)) {
 603                         mutex_unlock(&group->mark_mutex);
 604                         return PTR_ERR(fsn_mark);
 605                 }
 606         }
 607         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 608         mutex_unlock(&group->mark_mutex);
 609
 610         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 611                 fsnotify_recalc_vfsmount_mask(mnt);
 612
 613         fsnotify_put_mark(fsn_mark);
 614         return 0;
 615 }
 616
 617 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 618                                    struct inode *inode, __u32 mask,
 619                                    unsigned int flags)
 620 {
 621         struct fsnotify_mark *fsn_mark;
 622         __u32 added;
 623
 624         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 625
 626         /*
 627          * If some other task has this inode open for write we should not add
 628          * an ignored mark, unless that ignored mark is supposed to survive
 629          * modification changes anyway.
 630          */
 631         if ((flags & FAN_MARK_IGNORED_MASK) &&
 632             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 633             (atomic_read(&inode->i_writecount) > 0))
 634                 return 0;
 635
 636         mutex_lock(&group->mark_mutex);
 637         fsn_mark = fsnotify_find_inode_mark(group, inode);
 638         if (!fsn_mark) {
 639                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 640                 if (IS_ERR(fsn_mark)) {
 641                         mutex_unlock(&group->mark_mutex);
 642                         return PTR_ERR(fsn_mark);
 643                 }
 644         }
 645         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 646         mutex_unlock(&group->mark_mutex);
 647
 648         if (added & ~inode->i_fsnotify_mask)
 649                 fsnotify_recalc_inode_mask(inode);
 650
 651         fsnotify_put_mark(fsn_mark);
 652         return 0;
 653 }
 654
 655 /* fanotify syscalls */
 656 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 657 {
 658         struct fsnotify_group *group;
 659         int f_flags, fd;
 660         struct user_struct *user;
 661         struct fanotify_event_info *oevent;
 662
 663         pr_debug("%s: flags=%d event_f_flags=%d\n",
 664                 __func__, flags, event_f_flags);
 665
 666         if (!capable(CAP_SYS_ADMIN))
 667                 return -EPERM;
 668
 669         if (flags & ~FAN_ALL_INIT_FLAGS)
 670                 return -EINVAL;
 671
 672         user = get_current_user();
 673         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 674                 free_uid(user);
 675                 return -EMFILE;
 676         }
 677
 678         f_flags = O_RDWR | FMODE_NONOTIFY;
 679         if (flags & FAN_CLOEXEC)
 680                 f_flags |= O_CLOEXEC;
 681         if (flags & FAN_NONBLOCK)
 682                 f_flags |= O_NONBLOCK;
 683
 684         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 685         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 686         if (IS_ERR(group)) {
 687                 free_uid(user);
 688                 return PTR_ERR(group);
 689         }
 690
 691         group->fanotify_data.user = user;
 692         atomic_inc(&user->fanotify_listeners);
 693
 694         oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
 695         if (unlikely(!oevent)) {
 696                 fd = -ENOMEM;
 697                 goto out_destroy_group;
 698         }
 699         group->overflow_event = &oevent->fse;
 700
 701         if (force_o_largefile())
 702                 event_f_flags |= O_LARGEFILE;
 703         group->fanotify_data.f_flags = event_f_flags;
 704 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 705         spin_lock_init(&group->fanotify_data.access_lock);
 706         init_waitqueue_head(&group->fanotify_data.access_waitq);
 707         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 708         atomic_set(&group->fanotify_data.bypass_perm, 0);
 709 #endif
 710         switch (flags & FAN_ALL_CLASS_BITS) {
 711         case FAN_CLASS_NOTIF:
 712                 group->priority = FS_PRIO_0;
 713                 break;
 714         case FAN_CLASS_CONTENT:
 715                 group->priority = FS_PRIO_1;
 716                 break;
 717         case FAN_CLASS_PRE_CONTENT:
 718                 group->priority = FS_PRIO_2;
 719                 break;
 720         default:
 721                 fd = -EINVAL;
 722                 goto out_destroy_group;
 723         }
 724
 725         if (flags & FAN_UNLIMITED_QUEUE) {
 726                 fd = -EPERM;
 727                 if (!capable(CAP_SYS_ADMIN))
 728                         goto out_destroy_group;
 729                 group->max_events = UINT_MAX;
 730         } else {
 731                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 732         }
 733
 734         if (flags & FAN_UNLIMITED_MARKS) {
 735                 fd = -EPERM;
 736                 if (!capable(CAP_SYS_ADMIN))
 737                         goto out_destroy_group;
 738                 group->fanotify_data.max_marks = UINT_MAX;
 739         } else {
 740                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 741         }
 742
 743         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 744         if (fd < 0)
 745                 goto out_destroy_group;
 746
 747         return fd;
 748
 749 out_destroy_group:
 750         fsnotify_destroy_group(group);
 751         return fd;
 752 }
 753
 754 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 755                               __u64, mask, int, dfd,
 756                               const char  __user *, pathname)
 757 {
 758         struct inode *inode = NULL;
 759         struct vfsmount *mnt = NULL;
 760         struct fsnotify_group *group;
 761         struct fd f;
 762         struct path path;
 763         int ret;
 764
 765         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 766                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 767
 768         /* we only use the lower 32 bits as of right now. */
 769         if (mask & ((__u64)0xffffffff << 32))
 770                 return -EINVAL;
 771
 772         if (flags & ~FAN_ALL_MARK_FLAGS)
 773                 return -EINVAL;
 774         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 775         case FAN_MARK_ADD:              /* fallthrough */
 776         case FAN_MARK_REMOVE:
 777                 if (!mask)
 778                         return -EINVAL;
 779         case FAN_MARK_FLUSH:
 780                 break;
 781         default:
 782                 return -EINVAL;
 783         }
 784
 785         if (mask & FAN_ONDIR) {
 786                 flags |= FAN_MARK_ONDIR;
 787                 mask &= ~FAN_ONDIR;
 788         }
 789
 790 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 791         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 792 #else
 793         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 794 #endif
 795                 return -EINVAL;
 796
 797         f = fdget(fanotify_fd);
 798         if (unlikely(!f.file))
 799                 return -EBADF;
 800
 801         /* verify that this is indeed an fanotify instance */
 802         ret = -EINVAL;
 803         if (unlikely(f.file->f_op != &fanotify_fops))
 804                 goto fput_and_out;
 805         group = f.file->private_data;
 806
 807         /*
 808          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 809          * allowed to set permissions events.
 810          */
 811         ret = -EINVAL;
 812         if (mask & FAN_ALL_PERM_EVENTS &&
 813             group->priority == FS_PRIO_0)
 814                 goto fput_and_out;
 815
 816         ret = fanotify_find_path(dfd, pathname, &path, flags);
 817         if (ret)
 818                 goto fput_and_out;
 819
 820         /* inode held in place by reference to path; group by fget on fd */
 821         if (!(flags & FAN_MARK_MOUNT))
 822                 inode = path.dentry->d_inode;
 823         else
 824                 mnt = path.mnt;
 825
 826         /* create/update an inode mark */
 827         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 828         case FAN_MARK_ADD:
 829                 if (flags & FAN_MARK_MOUNT)
 830                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 831                 else
 832                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 833                 break;
 834         case FAN_MARK_REMOVE:
 835                 if (flags & FAN_MARK_MOUNT)
 836                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 837                 else
 838                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 839                 break;
 840         case FAN_MARK_FLUSH:
 841                 if (flags & FAN_MARK_MOUNT)
 842                         fsnotify_clear_vfsmount_marks_by_group(group);
 843                 else
 844                         fsnotify_clear_inode_marks_by_group(group);
 845                 break;
 846         default:
 847                 ret = -EINVAL;
 848         }
 849
 850         path_put(&path);
 851 fput_and_out:
 852         fdput(f);
 853         return ret;
 854 }
 855
 856 #ifdef CONFIG_COMPAT
 857 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 858                                 int, fanotify_fd, unsigned int, flags,
 859                                 __u32, mask0, __u32, mask1, int, dfd,
 860                                 const char  __user *, pathname)
 861 {
 862         return sys_fanotify_mark(fanotify_fd, flags,
 863 #ifdef __BIG_ENDIAN
 864                                 ((__u64)mask0 << 32) | mask1,
 865 #else
 866                                 ((__u64)mask1 << 32) | mask0,
 867 #endif
 868                                  dfd, pathname);
 869 }
 870 #endif
 871
 872 /*
 873  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 874  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 875  * must result in panic().
 876  */
 877 static int __init fanotify_user_setup(void)
 878 {
 879         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 880         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 881 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 882         fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
 883                                                 SLAB_PANIC);
 884 #endif
 885
 886         return 0;
 887 }
 888 device_initcall(fanotify_user_setup);