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
  17 #include <asm/ioctls.h>
  18
  19 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  20 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  21 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  22
  23 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  24
  25 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  26 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  27
  28 struct fanotify_response_event {
  29         struct list_head list;
  30         __s32 fd;
  31         struct fsnotify_event *event;
  32 };
  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, struct fsnotify_event *event)
  60 {
  61         int client_fd;
  62         struct dentry *dentry;
  63         struct vfsmount *mnt;
  64         struct file *new_file;
  65         unsigned int flags;
  66
  67         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  68
  69         client_fd = get_unused_fd();
  70         if (client_fd < 0)
  71                 return client_fd;
  72
  73         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  74                 WARN_ON(1);
  75                 put_unused_fd(client_fd);
  76                 return -EINVAL;
  77         }
  78
  79         /*
  80          * we need a new file handle for the userspace program so it can read even if it was
  81          * originally opened O_WRONLY.
  82          */
  83         dentry = dget(event->path.dentry);
  84         mnt = mntget(event->path.mnt);
  85         /* it's possible this event was an overflow event.  in that case dentry and mnt
  86          * are NULL;  That's fine, just don't call dentry open */
  87         if (dentry && mnt) {
  88                 flags = group->fanotify_data.f_flags;
  89                 new_file = dentry_open(dentry, mnt, flags, current_cred());
  90                 /*
  91                  * Attempt fallback to read-only access if writable was not possible
  92                  * in order to at least provide something to the listener.
  93                  */
  94                 if (IS_ERR(new_file) && group->fanotify_data.readonly_fallback) {
  95                         flags &= ~O_ACCMODE;
  96                         flags |= O_RDONLY;
  97                         new_file = dentry_open(dentry, mnt, flags,
  98                                                current_cred());
  99                 }
 100         } else {
 101                 new_file = ERR_PTR(-EOVERFLOW);
 102         }
 103         if (IS_ERR(new_file)) {
 104                 /*
 105                  * we still send an event even if we can't open the file.  this
 106                  * can happen when say tasks are gone and we try to open their
 107                  * /proc files or we try to open a WRONLY file like in sysfs
 108                  * we just send the errno to userspace since there isn't much
 109                  * else we can do.
 110                  */
 111                 put_unused_fd(client_fd);
 112                 client_fd = PTR_ERR(new_file);
 113         } else {
 114                 fd_install(client_fd, new_file);
 115         }
 116
 117         return client_fd;
 118 }
 119
 120 static int fill_event_metadata(struct fsnotify_group *group,
 121                                    struct fanotify_event_metadata *metadata,
 122                                    struct fsnotify_event *event)
 123 {
 124         int ret = 0;
 125
 126         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 127                  group, metadata, event);
 128
 129         metadata->event_len = FAN_EVENT_METADATA_LEN;
 130         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 131         metadata->vers = FANOTIFY_METADATA_VERSION;
 132         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 133         metadata->pid = pid_vnr(event->tgid);
 134         if (unlikely(event->mask & FAN_Q_OVERFLOW))
 135                 metadata->fd = FAN_NOFD;
 136         else {
 137                 metadata->fd = create_fd(group, event);
 138                 if (metadata->fd < 0)
 139                         ret = metadata->fd;
 140         }
 141
 142         return ret;
 143 }
 144
 145 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 146 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 147                                                   __s32 fd)
 148 {
 149         struct fanotify_response_event *re, *return_re = NULL;
 150
 151         mutex_lock(&group->fanotify_data.access_mutex);
 152         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 153                 if (re->fd != fd)
 154                         continue;
 155
 156                 list_del_init(&re->list);
 157                 return_re = re;
 158                 break;
 159         }
 160         mutex_unlock(&group->fanotify_data.access_mutex);
 161
 162         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 163
 164         return return_re;
 165 }
 166
 167 static int process_access_response(struct fsnotify_group *group,
 168                                    struct fanotify_response *response_struct)
 169 {
 170         struct fanotify_response_event *re;
 171         __s32 fd = response_struct->fd;
 172         __u32 response = response_struct->response;
 173
 174         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 175                  fd, response);
 176         /*
 177          * make sure the response is valid, if invalid we do nothing and either
 178          * userspace can send a valid response or we will clean it up after the
 179          * timeout
 180          */
 181         switch (response) {
 182         case FAN_ALLOW:
 183         case FAN_DENY:
 184                 break;
 185         default:
 186                 return -EINVAL;
 187         }
 188
 189         if (fd < 0)
 190                 return -EINVAL;
 191
 192         re = dequeue_re(group, fd);
 193         if (!re)
 194                 return -ENOENT;
 195
 196         re->event->response = response;
 197
 198         wake_up(&group->fanotify_data.access_waitq);
 199
 200         kmem_cache_free(fanotify_response_event_cache, re);
 201
 202         return 0;
 203 }
 204
 205 static int prepare_for_access_response(struct fsnotify_group *group,
 206                                        struct fsnotify_event *event,
 207                                        __s32 fd)
 208 {
 209         struct fanotify_response_event *re;
 210
 211         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 212                 return 0;
 213
 214         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 215         if (!re)
 216                 return -ENOMEM;
 217
 218         re->event = event;
 219         re->fd = fd;
 220
 221         mutex_lock(&group->fanotify_data.access_mutex);
 222         list_add_tail(&re->list, &group->fanotify_data.access_list);
 223         mutex_unlock(&group->fanotify_data.access_mutex);
 224
 225         return 0;
 226 }
 227
 228 static void remove_access_response(struct fsnotify_group *group,
 229                                    struct fsnotify_event *event,
 230                                    __s32 fd)
 231 {
 232         struct fanotify_response_event *re;
 233
 234         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 235                 return;
 236
 237         re = dequeue_re(group, fd);
 238         if (!re)
 239                 return;
 240
 241         BUG_ON(re->event != event);
 242
 243         kmem_cache_free(fanotify_response_event_cache, re);
 244
 245         return;
 246 }
 247 #else
 248 static int prepare_for_access_response(struct fsnotify_group *group,
 249                                        struct fsnotify_event *event,
 250                                        __s32 fd)
 251 {
 252         return 0;
 253 }
 254
 255 static void remove_access_response(struct fsnotify_group *group,
 256                                    struct fsnotify_event *event,
 257                                    __s32 fd)
 258 {
 259         return;
 260 }
 261 #endif
 262
 263 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 264                                   struct fsnotify_event *event,
 265                                   char __user *buf)
 266 {
 267         struct fanotify_event_metadata fanotify_event_metadata;
 268         int fd, ret;
 269
 270         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 271
 272         ret = fill_event_metadata(group, &fanotify_event_metadata, event);
 273         if (ret < 0)
 274                 goto out;
 275
 276         fd = fanotify_event_metadata.fd;
 277         ret = prepare_for_access_response(group, event, fd);
 278         if (ret)
 279                 goto out_close_fd;
 280
 281         ret = -EFAULT;
 282         if (copy_to_user(buf, &fanotify_event_metadata,
 283                          fanotify_event_metadata.event_len))
 284                 goto out_kill_access_response;
 285
 286         return fanotify_event_metadata.event_len;
 287
 288 out_kill_access_response:
 289         remove_access_response(group, event, fd);
 290 out_close_fd:
 291         if (fd != FAN_NOFD)
 292                 sys_close(fd);
 293 out:
 294 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 295         if (event->mask & FAN_ALL_PERM_EVENTS) {
 296                 event->response = FAN_DENY;
 297                 wake_up(&group->fanotify_data.access_waitq);
 298         }
 299 #endif
 300         return ret;
 301 }
 302
 303 /* intofiy userspace file descriptor functions */
 304 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 305 {
 306         struct fsnotify_group *group = file->private_data;
 307         int ret = 0;
 308
 309         poll_wait(file, &group->notification_waitq, wait);
 310         mutex_lock(&group->notification_mutex);
 311         if (!fsnotify_notify_queue_is_empty(group))
 312                 ret = POLLIN | POLLRDNORM;
 313         mutex_unlock(&group->notification_mutex);
 314
 315         return ret;
 316 }
 317
 318 static ssize_t fanotify_read(struct file *file, char __user *buf,
 319                              size_t count, loff_t *pos)
 320 {
 321         struct fsnotify_group *group;
 322         struct fsnotify_event *kevent;
 323         char __user *start;
 324         int ret;
 325         DEFINE_WAIT(wait);
 326
 327         start = buf;
 328         group = file->private_data;
 329
 330         pr_debug("%s: group=%p\n", __func__, group);
 331
 332         while (1) {
 333                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 334
 335                 mutex_lock(&group->notification_mutex);
 336                 kevent = get_one_event(group, count);
 337                 mutex_unlock(&group->notification_mutex);
 338
 339                 if (kevent) {
 340                         ret = PTR_ERR(kevent);
 341                         if (IS_ERR(kevent))
 342                                 break;
 343                         ret = copy_event_to_user(group, kevent, buf);
 344                         fsnotify_put_event(kevent);
 345                         if (ret < 0)
 346                                 break;
 347                         buf += ret;
 348                         count -= ret;
 349                         continue;
 350                 }
 351
 352                 ret = -EAGAIN;
 353                 if (file->f_flags & O_NONBLOCK)
 354                         break;
 355                 ret = -ERESTARTSYS;
 356                 if (signal_pending(current))
 357                         break;
 358
 359                 if (start != buf)
 360                         break;
 361
 362                 schedule();
 363         }
 364
 365         finish_wait(&group->notification_waitq, &wait);
 366         if (start != buf && ret != -EFAULT)
 367                 ret = buf - start;
 368         return ret;
 369 }
 370
 371 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 372 {
 373 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 374         struct fanotify_response response = { .fd = -1, .response = -1 };
 375         struct fsnotify_group *group;
 376         int ret;
 377
 378         group = file->private_data;
 379
 380         if (count > sizeof(response))
 381                 count = sizeof(response);
 382
 383         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 384
 385         if (copy_from_user(&response, buf, count))
 386                 return -EFAULT;
 387
 388         ret = process_access_response(group, &response);
 389         if (ret < 0)
 390                 count = ret;
 391
 392         return count;
 393 #else
 394         return -EINVAL;
 395 #endif
 396 }
 397
 398 static int fanotify_release(struct inode *ignored, struct file *file)
 399 {
 400         struct fsnotify_group *group = file->private_data;
 401
 402 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 403         struct fanotify_response_event *re, *lre;
 404
 405         mutex_lock(&group->fanotify_data.access_mutex);
 406
 407         atomic_inc(&group->fanotify_data.bypass_perm);
 408
 409         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 410                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 411                          re, re->event);
 412
 413                 list_del_init(&re->list);
 414                 re->event->response = FAN_ALLOW;
 415
 416                 kmem_cache_free(fanotify_response_event_cache, re);
 417         }
 418         mutex_unlock(&group->fanotify_data.access_mutex);
 419
 420         wake_up(&group->fanotify_data.access_waitq);
 421 #endif
 422         /* matches the fanotify_init->fsnotify_alloc_group */
 423         fsnotify_put_group(group);
 424
 425         return 0;
 426 }
 427
 428 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 429 {
 430         struct fsnotify_group *group;
 431         struct fsnotify_event_holder *holder;
 432         void __user *p;
 433         int ret = -ENOTTY;
 434         size_t send_len = 0;
 435
 436         group = file->private_data;
 437
 438         p = (void __user *) arg;
 439
 440         switch (cmd) {
 441         case FIONREAD:
 442                 mutex_lock(&group->notification_mutex);
 443                 list_for_each_entry(holder, &group->notification_list, event_list)
 444                         send_len += FAN_EVENT_METADATA_LEN;
 445                 mutex_unlock(&group->notification_mutex);
 446                 ret = put_user(send_len, (int __user *) p);
 447                 break;
 448         }
 449
 450         return ret;
 451 }
 452
 453 static const struct file_operations fanotify_fops = {
 454         .poll           = fanotify_poll,
 455         .read           = fanotify_read,
 456         .write          = fanotify_write,
 457         .fasync         = NULL,
 458         .release        = fanotify_release,
 459         .unlocked_ioctl = fanotify_ioctl,
 460         .compat_ioctl   = fanotify_ioctl,
 461         .llseek         = noop_llseek,
 462 };
 463
 464 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 465 {
 466         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 467 }
 468
 469 static int fanotify_find_path(int dfd, const char __user *filename,
 470                               struct path *path, unsigned int flags)
 471 {
 472         int ret;
 473
 474         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 475                  dfd, filename, flags);
 476
 477         if (filename == NULL) {
 478                 struct file *file;
 479                 int fput_needed;
 480
 481                 ret = -EBADF;
 482                 file = fget_light(dfd, &fput_needed);
 483                 if (!file)
 484                         goto out;
 485
 486                 ret = -ENOTDIR;
 487                 if ((flags & FAN_MARK_ONLYDIR) &&
 488                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 489                         fput_light(file, fput_needed);
 490                         goto out;
 491                 }
 492
 493                 *path = file->f_path;
 494                 path_get(path);
 495                 fput_light(file, fput_needed);
 496         } else {
 497                 unsigned int lookup_flags = 0;
 498
 499                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 500                         lookup_flags |= LOOKUP_FOLLOW;
 501                 if (flags & FAN_MARK_ONLYDIR)
 502                         lookup_flags |= LOOKUP_DIRECTORY;
 503
 504                 ret = user_path_at(dfd, filename, lookup_flags, path);
 505                 if (ret)
 506                         goto out;
 507         }
 508
 509         /* you can only watch an inode if you have read permissions on it */
 510         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 511         if (ret)
 512                 path_put(path);
 513 out:
 514         return ret;
 515 }
 516
 517 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 518                                             __u32 mask,
 519                                             unsigned int flags)
 520 {
 521         __u32 oldmask;
 522         int destroy_mark;
 523
 524         spin_lock(&fsn_mark->lock);
 525         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 526                 oldmask = fsn_mark->mask;
 527                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 528         } else {
 529                 oldmask = fsn_mark->ignored_mask;
 530                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 531         }
 532         destroy_mark = (!fsn_mark->mask && !fsn_mark->ignored_mask);
 533         spin_unlock(&fsn_mark->lock);
 534
 535         if (destroy_mark)
 536                 fsnotify_destroy_mark(fsn_mark);
 537
 538         return mask & oldmask;
 539 }
 540
 541 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 542                                          struct vfsmount *mnt, __u32 mask,
 543                                          unsigned int flags)
 544 {
 545         struct fsnotify_mark *fsn_mark = NULL;
 546         __u32 removed;
 547         int ret;
 548
 549         mutex_lock(&group->mutex);
 550         ret = -ENOENT;
 551         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 552         if (!fsn_mark)
 553                 goto err;
 554
 555         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 556         fsnotify_put_mark(fsn_mark);
 557         if (removed & mnt->mnt_fsnotify_mask)
 558                 fsnotify_recalc_vfsmount_mask(mnt);
 559         ret = 0;
 560 err:
 561         mutex_unlock(&group->mutex);
 562
 563         return ret;
 564 }
 565
 566 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 567                                       struct inode *inode, __u32 mask,
 568                                       unsigned int flags)
 569 {
 570         struct fsnotify_mark *fsn_mark = NULL;
 571         __u32 removed;
 572         int ret;
 573
 574         mutex_lock(&group->mutex);
 575         ret = -ENOENT;
 576         fsn_mark = fsnotify_find_inode_mark(group, inode);
 577         if (!fsn_mark)
 578                 goto err;
 579
 580         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 581         /* matches the fsnotify_find_inode_mark() */
 582         fsnotify_put_mark(fsn_mark);
 583         if (removed & inode->i_fsnotify_mask)
 584                 fsnotify_recalc_inode_mask(inode);
 585         ret = 0;
 586 err:
 587         mutex_unlock(&group->mutex);
 588
 589         return ret;
 590 }
 591
 592 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 593                                        __u32 mask,
 594                                        unsigned int flags)
 595 {
 596         __u32 oldmask = -1;
 597
 598         spin_lock(&fsn_mark->lock);
 599         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 600                 oldmask = fsn_mark->mask;
 601                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 602         } else {
 603                 __u32 tmask = fsn_mark->ignored_mask | mask;
 604                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 605                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 606                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 607         }
 608
 609         if (!(flags & FAN_MARK_ONDIR)) {
 610                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 611                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 612         }
 613
 614         spin_unlock(&fsn_mark->lock);
 615
 616         return mask & ~oldmask;
 617 }
 618
 619 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 620                                       struct vfsmount *mnt, __u32 mask,
 621                                       unsigned int flags)
 622 {
 623         struct fsnotify_mark *fsn_mark;
 624         __u32 added;
 625         int ret;
 626
 627         mutex_lock(&group->mutex);
 628         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 629         if (!fsn_mark) {
 630                 ret = -ENOSPC;
 631                 if (atomic_read(&group->num_marks) >
 632                     group->fanotify_data.max_marks)
 633                         goto err;
 634
 635                 ret = -ENOMEM;
 636                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 637                 if (!fsn_mark)
 638                         goto err;
 639
 640                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 641                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 642                 if (ret)
 643                         goto err2;
 644         }
 645         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 646
 647         if (added & ~mnt->mnt_fsnotify_mask)
 648                 fsnotify_recalc_vfsmount_mask(mnt);
 649         ret = 0;
 650 err2:
 651         fsnotify_put_mark(fsn_mark);
 652 err:
 653         mutex_unlock(&group->mutex);
 654         return ret;
 655 }
 656
 657 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 658                                    struct inode *inode, __u32 mask,
 659                                    unsigned int flags)
 660 {
 661         struct fsnotify_mark *fsn_mark;
 662         __u32 added;
 663         int ret;
 664
 665         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 666
 667         /*
 668          * If some other task has this inode open for write we should not add
 669          * an ignored mark, unless that ignored mark is supposed to survive
 670          * modification changes anyway.
 671          */
 672         if ((flags & FAN_MARK_IGNORED_MASK) &&
 673             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 674             (atomic_read(&inode->i_writecount) > 0))
 675                 return 0;
 676
 677         mutex_lock(&group->mutex);
 678         fsn_mark = fsnotify_find_inode_mark(group, inode);
 679         if (!fsn_mark) {
 680                 ret = -ENOSPC;
 681                 if (atomic_read(&group->num_marks) >
 682                     group->fanotify_data.max_marks)
 683                         goto err;
 684
 685                 ret = -ENOMEM;
 686                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 687                 if (!fsn_mark)
 688                         goto err;
 689
 690                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 691                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 692                 if (ret)
 693                         goto err2;
 694         }
 695         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 696
 697         if (added & ~inode->i_fsnotify_mask)
 698                 fsnotify_recalc_inode_mask(inode);
 699         ret = 0;
 700 err2:
 701         fsnotify_put_mark(fsn_mark);
 702 err:
 703         mutex_unlock(&group->mutex);
 704         return ret;
 705 }
 706
 707 /* fanotify syscalls */
 708 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 709 {
 710         struct fsnotify_group *group;
 711         int f_flags, fd;
 712         struct user_struct *user;
 713
 714         pr_debug("%s: flags=%d event_f_flags=%d\n",
 715                 __func__, flags, event_f_flags);
 716
 717         if (!capable(CAP_SYS_ADMIN))
 718                 return -EPERM;
 719
 720         if (flags & ~FAN_ALL_INIT_FLAGS)
 721                 return -EINVAL;
 722
 723         user = get_current_user();
 724         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 725                 free_uid(user);
 726                 return -EMFILE;
 727         }
 728
 729         f_flags = O_RDWR | FMODE_NONOTIFY;
 730         if (flags & FAN_CLOEXEC)
 731                 f_flags |= O_CLOEXEC;
 732         if (flags & FAN_NONBLOCK)
 733                 f_flags |= O_NONBLOCK;
 734
 735         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 736         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 737         if (IS_ERR(group)) {
 738                 free_uid(user);
 739                 return PTR_ERR(group);
 740         }
 741
 742         group->fanotify_data.user = user;
 743         atomic_inc(&user->fanotify_listeners);
 744
 745         group->fanotify_data.f_flags = event_f_flags | FMODE_NONOTIFY;
 746 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 747         mutex_init(&group->fanotify_data.access_mutex);
 748         init_waitqueue_head(&group->fanotify_data.access_waitq);
 749         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 750         atomic_set(&group->fanotify_data.bypass_perm, 0);
 751 #endif
 752         switch (flags & FAN_ALL_CLASS_BITS) {
 753         case FAN_CLASS_NOTIF:
 754                 group->priority = FS_PRIO_0;
 755                 break;
 756         case FAN_CLASS_CONTENT:
 757                 group->priority = FS_PRIO_1;
 758                 break;
 759         case FAN_CLASS_PRE_CONTENT:
 760                 group->priority = FS_PRIO_2;
 761                 break;
 762         default:
 763                 fd = -EINVAL;
 764                 goto out_put_group;
 765         }
 766
 767         if (flags & FAN_UNLIMITED_QUEUE) {
 768                 fd = -EPERM;
 769                 if (!capable(CAP_SYS_ADMIN))
 770                         goto out_put_group;
 771                 group->max_events = UINT_MAX;
 772         } else {
 773                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 774         }
 775
 776         if (flags & FAN_UNLIMITED_MARKS) {
 777                 fd = -EPERM;
 778                 if (!capable(CAP_SYS_ADMIN))
 779                         goto out_put_group;
 780                 group->fanotify_data.max_marks = UINT_MAX;
 781         } else {
 782                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 783         }
 784
 785         fd = -EINVAL;
 786         if (flags & FAN_READONLY_FALLBACK) {
 787                 if ((event_f_flags & O_ACCMODE) == O_RDWR)
 788                         group->fanotify_data.readonly_fallback = true;
 789                 else
 790                         goto out_put_group;
 791         }
 792
 793         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 794         if (fd < 0)
 795                 goto out_put_group;
 796
 797         return fd;
 798
 799 out_put_group:
 800         fsnotify_put_group(group);
 801         return fd;
 802 }
 803
 804 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 805                               __u64 mask, int dfd,
 806                               const char  __user * pathname)
 807 {
 808         struct inode *inode = NULL;
 809         struct vfsmount *mnt = NULL;
 810         struct fsnotify_group *group;
 811         struct file *filp;
 812         struct path path;
 813         int ret, fput_needed;
 814
 815         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 816                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 817
 818         /* we only use the lower 32 bits as of right now. */
 819         if (mask & ((__u64)0xffffffff << 32))
 820                 return -EINVAL;
 821
 822         if (flags & ~FAN_ALL_MARK_FLAGS)
 823                 return -EINVAL;
 824         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 825         case FAN_MARK_ADD:              /* fallthrough */
 826         case FAN_MARK_REMOVE:
 827                 if (!mask)
 828                         return -EINVAL;
 829         case FAN_MARK_FLUSH:
 830                 break;
 831         default:
 832                 return -EINVAL;
 833         }
 834
 835         if (mask & FAN_ONDIR) {
 836                 flags |= FAN_MARK_ONDIR;
 837                 mask &= ~FAN_ONDIR;
 838         }
 839
 840 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 841         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 842 #else
 843         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 844 #endif
 845                 return -EINVAL;
 846
 847         filp = fget_light(fanotify_fd, &fput_needed);
 848         if (unlikely(!filp))
 849                 return -EBADF;
 850
 851         /* verify that this is indeed an fanotify instance */
 852         ret = -EINVAL;
 853         if (unlikely(filp->f_op != &fanotify_fops))
 854                 goto fput_and_out;
 855         group = filp->private_data;
 856
 857         /*
 858          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 859          * allowed to set permissions events.
 860          */
 861         ret = -EINVAL;
 862         if (mask & FAN_ALL_PERM_EVENTS &&
 863             group->priority == FS_PRIO_0)
 864                 goto fput_and_out;
 865
 866         ret = fanotify_find_path(dfd, pathname, &path, flags);
 867         if (ret)
 868                 goto fput_and_out;
 869
 870         /* inode held in place by reference to path; group by fget on fd */
 871         if (!(flags & FAN_MARK_MOUNT))
 872                 inode = path.dentry->d_inode;
 873         else
 874                 mnt = path.mnt;
 875
 876         /* create/update an inode mark */
 877         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 878         case FAN_MARK_ADD:
 879                 if (flags & FAN_MARK_MOUNT)
 880                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 881                 else
 882                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 883                 break;
 884         case FAN_MARK_REMOVE:
 885                 if (flags & FAN_MARK_MOUNT)
 886                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 887                 else
 888                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 889                 break;
 890         case FAN_MARK_FLUSH:
 891                 if (flags & FAN_MARK_MOUNT)
 892                         fsnotify_clear_vfsmount_marks_by_group(group);
 893                 else
 894                         fsnotify_clear_inode_marks_by_group(group);
 895                 break;
 896         default:
 897                 ret = -EINVAL;
 898         }
 899
 900         path_put(&path);
 901 fput_and_out:
 902         fput_light(filp, fput_needed);
 903         return ret;
 904 }
 905
 906 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 907 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 908                                   long dfd, long pathname)
 909 {
 910         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 911                                   mask, (int) dfd,
 912                                   (const char  __user *) pathname);
 913 }
 914 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 915 #endif
 916
 917 /*
 918  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 919  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 920  * must result in panic().
 921  */
 922 static int __init fanotify_user_setup(void)
 923 {
 924         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 925         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 926                                                    SLAB_PANIC);
 927
 928         return 0;
 929 }
 930 device_initcall(fanotify_user_setup);