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
  23 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  24 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  25 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  26
  27 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  28
  29 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  30 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  31
  32 struct fanotify_response_event {
  33         struct list_head list;
  34         __s32 fd;
  35         struct fsnotify_event *event;
  36 };
  37
  38 /*
  39  * Get an fsnotify notification event if one exists and is small
  40  * enough to fit in "count". Return an error pointer if the count
  41  * is not large enough.
  42  *
  43  * Called with the group->notification_mutex held.
  44  */
  45 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  46                                             size_t count)
  47 {
  48         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  49
  50         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  51
  52         if (fsnotify_notify_queue_is_empty(group))
  53                 return NULL;
  54
  55         if (FAN_EVENT_METADATA_LEN > count)
  56                 return ERR_PTR(-EINVAL);
  57
  58         /* held the notification_mutex the whole time, so this is the
  59          * same event we peeked above */
  60         return fsnotify_remove_notify_event(group);
  61 }
  62
  63 static int create_fd(struct fsnotify_group *group,
  64                         struct fsnotify_event *event,
  65                         struct file **file)
  66 {
  67         int client_fd;
  68         struct file *new_file;
  69
  70         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  71
  72         client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
  73         if (client_fd < 0)
  74                 return client_fd;
  75
  76         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  77                 WARN_ON(1);
  78                 put_unused_fd(client_fd);
  79                 return -EINVAL;
  80         }
  81
  82         /*
  83          * we need a new file handle for the userspace program so it can read even if it was
  84          * originally opened O_WRONLY.
  85          */
  86         /* it's possible this event was an overflow event.  in that case dentry and mnt
  87          * are NULL;  That's fine, just don't call dentry open */
  88         if (event->path.dentry && event->path.mnt)
  89                 new_file = dentry_open(&event->path,
  90                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  91                                        current_cred());
  92         else
  93                 new_file = ERR_PTR(-EOVERFLOW);
  94         if (IS_ERR(new_file)) {
  95                 /*
  96                  * we still send an event even if we can't open the file.  this
  97                  * can happen when say tasks are gone and we try to open their
  98                  * /proc files or we try to open a WRONLY file like in sysfs
  99                  * we just send the errno to userspace since there isn't much
 100                  * else we can do.
 101                  */
 102                 put_unused_fd(client_fd);
 103                 client_fd = PTR_ERR(new_file);
 104         } else {
 105                 *file = new_file;
 106         }
 107
 108         return client_fd;
 109 }
 110
 111 static int fill_event_metadata(struct fsnotify_group *group,
 112                                    struct fanotify_event_metadata *metadata,
 113                                    struct fsnotify_event *event,
 114                                    struct file **file)
 115 {
 116         int ret = 0;
 117
 118         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 119                  group, metadata, event);
 120
 121         *file = NULL;
 122         metadata->event_len = FAN_EVENT_METADATA_LEN;
 123         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 124         metadata->vers = FANOTIFY_METADATA_VERSION;
 125         metadata->reserved = 0;
 126         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 127         metadata->pid = pid_vnr(event->tgid);
 128         if (unlikely(event->mask & FAN_Q_OVERFLOW))
 129                 metadata->fd = FAN_NOFD;
 130         else {
 131                 metadata->fd = create_fd(group, event, file);
 132                 if (metadata->fd < 0)
 133                         ret = metadata->fd;
 134         }
 135
 136         return ret;
 137 }
 138
 139 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 140 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 141                                                   __s32 fd)
 142 {
 143         struct fanotify_response_event *re, *return_re = NULL;
 144
 145         mutex_lock(&group->fanotify_data.access_mutex);
 146         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 147                 if (re->fd != fd)
 148                         continue;
 149
 150                 list_del_init(&re->list);
 151                 return_re = re;
 152                 break;
 153         }
 154         mutex_unlock(&group->fanotify_data.access_mutex);
 155
 156         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 157
 158         return return_re;
 159 }
 160
 161 static int process_access_response(struct fsnotify_group *group,
 162                                    struct fanotify_response *response_struct)
 163 {
 164         struct fanotify_response_event *re;
 165         __s32 fd = response_struct->fd;
 166         __u32 response = response_struct->response;
 167
 168         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 169                  fd, response);
 170         /*
 171          * make sure the response is valid, if invalid we do nothing and either
 172          * userspace can send a valid response or we will clean it up after the
 173          * timeout
 174          */
 175         switch (response) {
 176         case FAN_ALLOW:
 177         case FAN_DENY:
 178                 break;
 179         default:
 180                 return -EINVAL;
 181         }
 182
 183         if (fd < 0)
 184                 return -EINVAL;
 185
 186         re = dequeue_re(group, fd);
 187         if (!re)
 188                 return -ENOENT;
 189
 190         re->event->response = response;
 191
 192         wake_up(&group->fanotify_data.access_waitq);
 193
 194         kmem_cache_free(fanotify_response_event_cache, re);
 195
 196         return 0;
 197 }
 198
 199 static int prepare_for_access_response(struct fsnotify_group *group,
 200                                        struct fsnotify_event *event,
 201                                        __s32 fd)
 202 {
 203         struct fanotify_response_event *re;
 204
 205         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 206                 return 0;
 207
 208         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 209         if (!re)
 210                 return -ENOMEM;
 211
 212         re->event = event;
 213         re->fd = fd;
 214
 215         mutex_lock(&group->fanotify_data.access_mutex);
 216
 217         if (atomic_read(&group->fanotify_data.bypass_perm)) {
 218                 mutex_unlock(&group->fanotify_data.access_mutex);
 219                 kmem_cache_free(fanotify_response_event_cache, re);
 220                 event->response = FAN_ALLOW;
 221                 return 0;
 222         }
 223
 224         list_add_tail(&re->list, &group->fanotify_data.access_list);
 225         mutex_unlock(&group->fanotify_data.access_mutex);
 226
 227         return 0;
 228 }
 229
 230 #else
 231 static int prepare_for_access_response(struct fsnotify_group *group,
 232                                        struct fsnotify_event *event,
 233                                        __s32 fd)
 234 {
 235         return 0;
 236 }
 237
 238 #endif
 239
 240 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 241                                   struct fsnotify_event *event,
 242                                   char __user *buf)
 243 {
 244         struct fanotify_event_metadata fanotify_event_metadata;
 245         struct file *f;
 246         int fd, ret;
 247
 248         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 249
 250         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 251         if (ret < 0)
 252                 goto out;
 253
 254         fd = fanotify_event_metadata.fd;
 255         ret = -EFAULT;
 256         if (copy_to_user(buf, &fanotify_event_metadata,
 257                          fanotify_event_metadata.event_len))
 258                 goto out_close_fd;
 259
 260         ret = prepare_for_access_response(group, event, fd);
 261         if (ret)
 262                 goto out_close_fd;
 263
 264         if (fd != FAN_NOFD)
 265                 fd_install(fd, f);
 266         return fanotify_event_metadata.event_len;
 267
 268 out_close_fd:
 269         if (fd != FAN_NOFD) {
 270                 put_unused_fd(fd);
 271                 fput(f);
 272         }
 273 out:
 274 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 275         if (event->mask & FAN_ALL_PERM_EVENTS) {
 276                 event->response = FAN_DENY;
 277                 wake_up(&group->fanotify_data.access_waitq);
 278         }
 279 #endif
 280         return ret;
 281 }
 282
 283 /* intofiy userspace file descriptor functions */
 284 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 285 {
 286         struct fsnotify_group *group = file->private_data;
 287         int ret = 0;
 288
 289         poll_wait(file, &group->notification_waitq, wait);
 290         mutex_lock(&group->notification_mutex);
 291         if (!fsnotify_notify_queue_is_empty(group))
 292                 ret = POLLIN | POLLRDNORM;
 293         mutex_unlock(&group->notification_mutex);
 294
 295         return ret;
 296 }
 297
 298 static ssize_t fanotify_read(struct file *file, char __user *buf,
 299                              size_t count, loff_t *pos)
 300 {
 301         struct fsnotify_group *group;
 302         struct fsnotify_event *kevent;
 303         char __user *start;
 304         int ret;
 305         DEFINE_WAIT(wait);
 306
 307         start = buf;
 308         group = file->private_data;
 309
 310         pr_debug("%s: group=%p\n", __func__, group);
 311
 312         while (1) {
 313                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 314
 315                 mutex_lock(&group->notification_mutex);
 316                 kevent = get_one_event(group, count);
 317                 mutex_unlock(&group->notification_mutex);
 318
 319                 if (kevent) {
 320                         ret = PTR_ERR(kevent);
 321                         if (IS_ERR(kevent))
 322                                 break;
 323                         ret = copy_event_to_user(group, kevent, buf);
 324                         fsnotify_put_event(kevent);
 325                         if (ret < 0)
 326                                 break;
 327                         buf += ret;
 328                         count -= ret;
 329                         continue;
 330                 }
 331
 332                 ret = -EAGAIN;
 333                 if (file->f_flags & O_NONBLOCK)
 334                         break;
 335                 ret = -ERESTARTSYS;
 336                 if (signal_pending(current))
 337                         break;
 338
 339                 if (start != buf)
 340                         break;
 341
 342                 schedule();
 343         }
 344
 345         finish_wait(&group->notification_waitq, &wait);
 346         if (start != buf && ret != -EFAULT)
 347                 ret = buf - start;
 348         return ret;
 349 }
 350
 351 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 352 {
 353 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 354         struct fanotify_response response = { .fd = -1, .response = -1 };
 355         struct fsnotify_group *group;
 356         int ret;
 357
 358         group = file->private_data;
 359
 360         if (count > sizeof(response))
 361                 count = sizeof(response);
 362
 363         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 364
 365         if (copy_from_user(&response, buf, count))
 366                 return -EFAULT;
 367
 368         ret = process_access_response(group, &response);
 369         if (ret < 0)
 370                 count = ret;
 371
 372         return count;
 373 #else
 374         return -EINVAL;
 375 #endif
 376 }
 377
 378 static int fanotify_release(struct inode *ignored, struct file *file)
 379 {
 380         struct fsnotify_group *group = file->private_data;
 381
 382 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 383         struct fanotify_response_event *re, *lre;
 384
 385         mutex_lock(&group->fanotify_data.access_mutex);
 386
 387         atomic_inc(&group->fanotify_data.bypass_perm);
 388
 389         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 390                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 391                          re, re->event);
 392
 393                 list_del_init(&re->list);
 394                 re->event->response = FAN_ALLOW;
 395
 396                 kmem_cache_free(fanotify_response_event_cache, re);
 397         }
 398         mutex_unlock(&group->fanotify_data.access_mutex);
 399
 400         wake_up(&group->fanotify_data.access_waitq);
 401 #endif
 402
 403         if (file->f_flags & FASYNC)
 404                 fsnotify_fasync(-1, file, 0);
 405
 406         /* matches the fanotify_init->fsnotify_alloc_group */
 407         fsnotify_destroy_group(group);
 408
 409         return 0;
 410 }
 411
 412 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 413 {
 414         struct fsnotify_group *group;
 415         struct fsnotify_event_holder *holder;
 416         void __user *p;
 417         int ret = -ENOTTY;
 418         size_t send_len = 0;
 419
 420         group = file->private_data;
 421
 422         p = (void __user *) arg;
 423
 424         switch (cmd) {
 425         case FIONREAD:
 426                 mutex_lock(&group->notification_mutex);
 427                 list_for_each_entry(holder, &group->notification_list, event_list)
 428                         send_len += FAN_EVENT_METADATA_LEN;
 429                 mutex_unlock(&group->notification_mutex);
 430                 ret = put_user(send_len, (int __user *) p);
 431                 break;
 432         }
 433
 434         return ret;
 435 }
 436
 437 static const struct file_operations fanotify_fops = {
 438         .show_fdinfo    = fanotify_show_fdinfo,
 439         .poll           = fanotify_poll,
 440         .read           = fanotify_read,
 441         .write          = fanotify_write,
 442         .fasync         = NULL,
 443         .release        = fanotify_release,
 444         .unlocked_ioctl = fanotify_ioctl,
 445         .compat_ioctl   = fanotify_ioctl,
 446         .llseek         = noop_llseek,
 447 };
 448
 449 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 450 {
 451         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 452 }
 453
 454 static int fanotify_find_path(int dfd, const char __user *filename,
 455                               struct path *path, unsigned int flags)
 456 {
 457         int ret;
 458
 459         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 460                  dfd, filename, flags);
 461
 462         if (filename == NULL) {
 463                 struct fd f = fdget(dfd);
 464
 465                 ret = -EBADF;
 466                 if (!f.file)
 467                         goto out;
 468
 469                 ret = -ENOTDIR;
 470                 if ((flags & FAN_MARK_ONLYDIR) &&
 471                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 472                         fdput(f);
 473                         goto out;
 474                 }
 475
 476                 *path = f.file->f_path;
 477                 path_get(path);
 478                 fdput(f);
 479         } else {
 480                 unsigned int lookup_flags = 0;
 481
 482                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 483                         lookup_flags |= LOOKUP_FOLLOW;
 484                 if (flags & FAN_MARK_ONLYDIR)
 485                         lookup_flags |= LOOKUP_DIRECTORY;
 486
 487                 ret = user_path_at(dfd, filename, lookup_flags, path);
 488                 if (ret)
 489                         goto out;
 490         }
 491
 492         /* you can only watch an inode if you have read permissions on it */
 493         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 494         if (ret)
 495                 path_put(path);
 496 out:
 497         return ret;
 498 }
 499
 500 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 501                                             __u32 mask,
 502                                             unsigned int flags,
 503                                             int *destroy)
 504 {
 505         __u32 oldmask;
 506
 507         spin_lock(&fsn_mark->lock);
 508         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 509                 oldmask = fsn_mark->mask;
 510                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 511         } else {
 512                 oldmask = fsn_mark->ignored_mask;
 513                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 514         }
 515         spin_unlock(&fsn_mark->lock);
 516
 517         *destroy = !(oldmask & ~mask);
 518
 519         return mask & oldmask;
 520 }
 521
 522 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 523                                          struct vfsmount *mnt, __u32 mask,
 524                                          unsigned int flags)
 525 {
 526         struct fsnotify_mark *fsn_mark = NULL;
 527         __u32 removed;
 528         int destroy_mark;
 529
 530         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 531         if (!fsn_mark)
 532                 return -ENOENT;
 533
 534         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 535                                                  &destroy_mark);
 536         if (destroy_mark)
 537                 fsnotify_destroy_mark(fsn_mark, group);
 538
 539         fsnotify_put_mark(fsn_mark);
 540         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 541                 fsnotify_recalc_vfsmount_mask(mnt);
 542
 543         return 0;
 544 }
 545
 546 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 547                                       struct inode *inode, __u32 mask,
 548                                       unsigned int flags)
 549 {
 550         struct fsnotify_mark *fsn_mark = NULL;
 551         __u32 removed;
 552         int destroy_mark;
 553
 554         fsn_mark = fsnotify_find_inode_mark(group, inode);
 555         if (!fsn_mark)
 556                 return -ENOENT;
 557
 558         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 559                                                  &destroy_mark);
 560         if (destroy_mark)
 561                 fsnotify_destroy_mark(fsn_mark, group);
 562         /* matches the fsnotify_find_inode_mark() */
 563         fsnotify_put_mark(fsn_mark);
 564         if (removed & inode->i_fsnotify_mask)
 565                 fsnotify_recalc_inode_mask(inode);
 566
 567         return 0;
 568 }
 569
 570 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 571                                        __u32 mask,
 572                                        unsigned int flags)
 573 {
 574         __u32 oldmask = -1;
 575
 576         spin_lock(&fsn_mark->lock);
 577         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 578                 oldmask = fsn_mark->mask;
 579                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 580         } else {
 581                 __u32 tmask = fsn_mark->ignored_mask | mask;
 582                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 583                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 584                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 585         }
 586
 587         if (!(flags & FAN_MARK_ONDIR)) {
 588                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 589                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 590         }
 591
 592         spin_unlock(&fsn_mark->lock);
 593
 594         return mask & ~oldmask;
 595 }
 596
 597 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 598                                       struct vfsmount *mnt, __u32 mask,
 599                                       unsigned int flags)
 600 {
 601         struct fsnotify_mark *fsn_mark;
 602         __u32 added;
 603         int ret = 0;
 604
 605         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 606         if (!fsn_mark) {
 607                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 608                         return -ENOSPC;
 609
 610                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 611                 if (!fsn_mark)
 612                         return -ENOMEM;
 613
 614                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 615                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 616                 if (ret)
 617                         goto err;
 618         }
 619         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 620
 621         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 622                 fsnotify_recalc_vfsmount_mask(mnt);
 623 err:
 624         fsnotify_put_mark(fsn_mark);
 625         return ret;
 626 }
 627
 628 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 629                                    struct inode *inode, __u32 mask,
 630                                    unsigned int flags)
 631 {
 632         struct fsnotify_mark *fsn_mark;
 633         __u32 added;
 634         int ret = 0;
 635
 636         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 637
 638         /*
 639          * If some other task has this inode open for write we should not add
 640          * an ignored mark, unless that ignored mark is supposed to survive
 641          * modification changes anyway.
 642          */
 643         if ((flags & FAN_MARK_IGNORED_MASK) &&
 644             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 645             (atomic_read(&inode->i_writecount) > 0))
 646                 return 0;
 647
 648         fsn_mark = fsnotify_find_inode_mark(group, inode);
 649         if (!fsn_mark) {
 650                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 651                         return -ENOSPC;
 652
 653                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 654                 if (!fsn_mark)
 655                         return -ENOMEM;
 656
 657                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 658                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 659                 if (ret)
 660                         goto err;
 661         }
 662         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 663
 664         if (added & ~inode->i_fsnotify_mask)
 665                 fsnotify_recalc_inode_mask(inode);
 666 err:
 667         fsnotify_put_mark(fsn_mark);
 668         return ret;
 669 }
 670
 671 /* fanotify syscalls */
 672 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 673 {
 674         struct fsnotify_group *group;
 675         int f_flags, fd;
 676         struct user_struct *user;
 677
 678         pr_debug("%s: flags=%d event_f_flags=%d\n",
 679                 __func__, flags, event_f_flags);
 680
 681         if (!capable(CAP_SYS_ADMIN))
 682                 return -EPERM;
 683
 684         if (flags & ~FAN_ALL_INIT_FLAGS)
 685                 return -EINVAL;
 686
 687         user = get_current_user();
 688         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 689                 free_uid(user);
 690                 return -EMFILE;
 691         }
 692
 693         f_flags = O_RDWR | FMODE_NONOTIFY;
 694         if (flags & FAN_CLOEXEC)
 695                 f_flags |= O_CLOEXEC;
 696         if (flags & FAN_NONBLOCK)
 697                 f_flags |= O_NONBLOCK;
 698
 699         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 700         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 701         if (IS_ERR(group)) {
 702                 free_uid(user);
 703                 return PTR_ERR(group);
 704         }
 705
 706         group->fanotify_data.user = user;
 707         atomic_inc(&user->fanotify_listeners);
 708
 709         group->fanotify_data.f_flags = event_f_flags;
 710 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 711         mutex_init(&group->fanotify_data.access_mutex);
 712         init_waitqueue_head(&group->fanotify_data.access_waitq);
 713         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 714         atomic_set(&group->fanotify_data.bypass_perm, 0);
 715 #endif
 716         switch (flags & FAN_ALL_CLASS_BITS) {
 717         case FAN_CLASS_NOTIF:
 718                 group->priority = FS_PRIO_0;
 719                 break;
 720         case FAN_CLASS_CONTENT:
 721                 group->priority = FS_PRIO_1;
 722                 break;
 723         case FAN_CLASS_PRE_CONTENT:
 724                 group->priority = FS_PRIO_2;
 725                 break;
 726         default:
 727                 fd = -EINVAL;
 728                 goto out_destroy_group;
 729         }
 730
 731         if (flags & FAN_UNLIMITED_QUEUE) {
 732                 fd = -EPERM;
 733                 if (!capable(CAP_SYS_ADMIN))
 734                         goto out_destroy_group;
 735                 group->max_events = UINT_MAX;
 736         } else {
 737                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 738         }
 739
 740         if (flags & FAN_UNLIMITED_MARKS) {
 741                 fd = -EPERM;
 742                 if (!capable(CAP_SYS_ADMIN))
 743                         goto out_destroy_group;
 744                 group->fanotify_data.max_marks = UINT_MAX;
 745         } else {
 746                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 747         }
 748
 749         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 750         if (fd < 0)
 751                 goto out_destroy_group;
 752
 753         return fd;
 754
 755 out_destroy_group:
 756         fsnotify_destroy_group(group);
 757         return fd;
 758 }
 759
 760 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 761                               __u64, mask, int, dfd,
 762                               const char  __user *, pathname)
 763 {
 764         struct inode *inode = NULL;
 765         struct vfsmount *mnt = NULL;
 766         struct fsnotify_group *group;
 767         struct fd f;
 768         struct path path;
 769         int ret;
 770
 771         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 772                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 773
 774         /* we only use the lower 32 bits as of right now. */
 775         if (mask & ((__u64)0xffffffff << 32))
 776                 return -EINVAL;
 777
 778         if (flags & ~FAN_ALL_MARK_FLAGS)
 779                 return -EINVAL;
 780         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 781         case FAN_MARK_ADD:              /* fallthrough */
 782         case FAN_MARK_REMOVE:
 783                 if (!mask)
 784                         return -EINVAL;
 785         case FAN_MARK_FLUSH:
 786                 break;
 787         default:
 788                 return -EINVAL;
 789         }
 790
 791         if (mask & FAN_ONDIR) {
 792                 flags |= FAN_MARK_ONDIR;
 793                 mask &= ~FAN_ONDIR;
 794         }
 795
 796 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 797         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 798 #else
 799         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 800 #endif
 801                 return -EINVAL;
 802
 803         f = fdget(fanotify_fd);
 804         if (unlikely(!f.file))
 805                 return -EBADF;
 806
 807         /* verify that this is indeed an fanotify instance */
 808         ret = -EINVAL;
 809         if (unlikely(f.file->f_op != &fanotify_fops))
 810                 goto fput_and_out;
 811         group = f.file->private_data;
 812
 813         /*
 814          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 815          * allowed to set permissions events.
 816          */
 817         ret = -EINVAL;
 818         if (mask & FAN_ALL_PERM_EVENTS &&
 819             group->priority == FS_PRIO_0)
 820                 goto fput_and_out;
 821
 822         ret = fanotify_find_path(dfd, pathname, &path, flags);
 823         if (ret)
 824                 goto fput_and_out;
 825
 826         /* inode held in place by reference to path; group by fget on fd */
 827         if (!(flags & FAN_MARK_MOUNT))
 828                 inode = path.dentry->d_inode;
 829         else
 830                 mnt = path.mnt;
 831
 832         /* create/update an inode mark */
 833         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 834         case FAN_MARK_ADD:
 835                 if (flags & FAN_MARK_MOUNT)
 836                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 837                 else
 838                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 839                 break;
 840         case FAN_MARK_REMOVE:
 841                 if (flags & FAN_MARK_MOUNT)
 842                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 843                 else
 844                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 845                 break;
 846         case FAN_MARK_FLUSH:
 847                 if (flags & FAN_MARK_MOUNT)
 848                         fsnotify_clear_vfsmount_marks_by_group(group);
 849                 else
 850                         fsnotify_clear_inode_marks_by_group(group);
 851                 break;
 852         default:
 853                 ret = -EINVAL;
 854         }
 855
 856         path_put(&path);
 857 fput_and_out:
 858         fdput(f);
 859         return ret;
 860 }
 861
 862 #ifdef CONFIG_COMPAT
 863 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 864                                 int, fanotify_fd, unsigned int, flags,
 865                                 __u32, mask0, __u32, mask1, int, dfd,
 866                                 const char  __user *, pathname)
 867 {
 868         return sys_fanotify_mark(fanotify_fd, flags,
 869 #ifdef __BIG_ENDIAN
 870                                 ((__u64)mask0 << 32) | mask1,
 871 #else
 872                                 ((__u64)mask1 << 32) | mask0,
 873 #endif
 874                                  dfd, pathname);
 875 }
 876 #endif
 877
 878 /*
 879  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 880  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 881  * must result in panic().
 882  */
 883 static int __init fanotify_user_setup(void)
 884 {
 885         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 886         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 887                                                    SLAB_PANIC);
 888
 889         return 0;
 890 }
 891 device_initcall(fanotify_user_setup);