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 #include "../../mount.h"
  20 #include "../fdinfo.h"
  21
  22 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  23 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  24 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  25
  26 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  27
  28 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  29 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  30
  31 struct fanotify_response_event {
  32         struct list_head list;
  33         __s32 fd;
  34         struct fsnotify_event *event;
  35 };
  36
  37 /*
  38  * Get an fsnotify notification event if one exists and is small
  39  * enough to fit in "count". Return an error pointer if the count
  40  * is not large enough.
  41  *
  42  * Called with the group->notification_mutex held.
  43  */
  44 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  45                                             size_t count)
  46 {
  47         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  48
  49         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  50
  51         if (fsnotify_notify_queue_is_empty(group))
  52                 return NULL;
  53
  54         if (FAN_EVENT_METADATA_LEN > count)
  55                 return ERR_PTR(-EINVAL);
  56
  57         /* held the notification_mutex the whole time, so this is the
  58          * same event we peeked above */
  59         return fsnotify_remove_notify_event(group);
  60 }
  61
  62 static int create_fd(struct fsnotify_group *group,
  63                         struct fsnotify_event *event,
  64                         struct file **file)
  65 {
  66         int client_fd;
  67         struct file *new_file;
  68
  69         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  70
  71         client_fd = get_unused_fd();
  72         if (client_fd < 0)
  73                 return client_fd;
  74
  75         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  76                 WARN_ON(1);
  77                 put_unused_fd(client_fd);
  78                 return -EINVAL;
  79         }
  80
  81         /*
  82          * we need a new file handle for the userspace program so it can read even if it was
  83          * originally opened O_WRONLY.
  84          */
  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 (event->path.dentry && event->path.mnt)
  88                 new_file = dentry_open(&event->path,
  89                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  90                                        current_cred());
  91         else
  92                 new_file = ERR_PTR(-EOVERFLOW);
  93         if (IS_ERR(new_file)) {
  94                 /*
  95                  * we still send an event even if we can't open the file.  this
  96                  * can happen when say tasks are gone and we try to open their
  97                  * /proc files or we try to open a WRONLY file like in sysfs
  98                  * we just send the errno to userspace since there isn't much
  99                  * else we can do.
 100                  */
 101                 put_unused_fd(client_fd);
 102                 client_fd = PTR_ERR(new_file);
 103         } else {
 104                 *file = new_file;
 105         }
 106
 107         return client_fd;
 108 }
 109
 110 static int fill_event_metadata(struct fsnotify_group *group,
 111                                    struct fanotify_event_metadata *metadata,
 112                                    struct fsnotify_event *event,
 113                                    struct file **file)
 114 {
 115         int ret = 0;
 116
 117         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 118                  group, metadata, event);
 119
 120         *file = NULL;
 121         metadata->event_len = FAN_EVENT_METADATA_LEN;
 122         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 123         metadata->vers = FANOTIFY_METADATA_VERSION;
 124         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 125         metadata->pid = pid_vnr(event->tgid);
 126         if (unlikely(event->mask & FAN_Q_OVERFLOW))
 127                 metadata->fd = FAN_NOFD;
 128         else {
 129                 metadata->fd = create_fd(group, event, file);
 130                 if (metadata->fd < 0)
 131                         ret = metadata->fd;
 132         }
 133
 134         return ret;
 135 }
 136
 137 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 138 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 139                                                   __s32 fd)
 140 {
 141         struct fanotify_response_event *re, *return_re = NULL;
 142
 143         mutex_lock(&group->fanotify_data.access_mutex);
 144         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 145                 if (re->fd != fd)
 146                         continue;
 147
 148                 list_del_init(&re->list);
 149                 return_re = re;
 150                 break;
 151         }
 152         mutex_unlock(&group->fanotify_data.access_mutex);
 153
 154         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 155
 156         return return_re;
 157 }
 158
 159 static int process_access_response(struct fsnotify_group *group,
 160                                    struct fanotify_response *response_struct)
 161 {
 162         struct fanotify_response_event *re;
 163         __s32 fd = response_struct->fd;
 164         __u32 response = response_struct->response;
 165
 166         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 167                  fd, response);
 168         /*
 169          * make sure the response is valid, if invalid we do nothing and either
 170          * userspace can send a valid response or we will clean it up after the
 171          * timeout
 172          */
 173         switch (response) {
 174         case FAN_ALLOW:
 175         case FAN_DENY:
 176                 break;
 177         default:
 178                 return -EINVAL;
 179         }
 180
 181         if (fd < 0)
 182                 return -EINVAL;
 183
 184         re = dequeue_re(group, fd);
 185         if (!re)
 186                 return -ENOENT;
 187
 188         re->event->response = response;
 189
 190         wake_up(&group->fanotify_data.access_waitq);
 191
 192         kmem_cache_free(fanotify_response_event_cache, re);
 193
 194         return 0;
 195 }
 196
 197 static int prepare_for_access_response(struct fsnotify_group *group,
 198                                        struct fsnotify_event *event,
 199                                        __s32 fd)
 200 {
 201         struct fanotify_response_event *re;
 202
 203         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 204                 return 0;
 205
 206         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 207         if (!re)
 208                 return -ENOMEM;
 209
 210         re->event = event;
 211         re->fd = fd;
 212
 213         mutex_lock(&group->fanotify_data.access_mutex);
 214
 215         if (atomic_read(&group->fanotify_data.bypass_perm)) {
 216                 mutex_unlock(&group->fanotify_data.access_mutex);
 217                 kmem_cache_free(fanotify_response_event_cache, re);
 218                 event->response = FAN_ALLOW;
 219                 return 0;
 220         }
 221
 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 #else
 229 static int prepare_for_access_response(struct fsnotify_group *group,
 230                                        struct fsnotify_event *event,
 231                                        __s32 fd)
 232 {
 233         return 0;
 234 }
 235
 236 #endif
 237
 238 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 239                                   struct fsnotify_event *event,
 240                                   char __user *buf)
 241 {
 242         struct fanotify_event_metadata fanotify_event_metadata;
 243         struct file *f;
 244         int fd, ret;
 245
 246         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 247
 248         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 249         if (ret < 0)
 250                 goto out;
 251
 252         fd = fanotify_event_metadata.fd;
 253         ret = -EFAULT;
 254         if (copy_to_user(buf, &fanotify_event_metadata,
 255                          fanotify_event_metadata.event_len))
 256                 goto out_close_fd;
 257
 258         ret = prepare_for_access_response(group, event, fd);
 259         if (ret)
 260                 goto out_close_fd;
 261
 262         if (fd != FAN_NOFD)
 263                 fd_install(fd, f);
 264         return fanotify_event_metadata.event_len;
 265
 266 out_close_fd:
 267         if (fd != FAN_NOFD) {
 268                 put_unused_fd(fd);
 269                 fput(f);
 270         }
 271 out:
 272 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 273         if (event->mask & FAN_ALL_PERM_EVENTS) {
 274                 event->response = FAN_DENY;
 275                 wake_up(&group->fanotify_data.access_waitq);
 276         }
 277 #endif
 278         return ret;
 279 }
 280
 281 /* intofiy userspace file descriptor functions */
 282 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 283 {
 284         struct fsnotify_group *group = file->private_data;
 285         int ret = 0;
 286
 287         poll_wait(file, &group->notification_waitq, wait);
 288         mutex_lock(&group->notification_mutex);
 289         if (!fsnotify_notify_queue_is_empty(group))
 290                 ret = POLLIN | POLLRDNORM;
 291         mutex_unlock(&group->notification_mutex);
 292
 293         return ret;
 294 }
 295
 296 static ssize_t fanotify_read(struct file *file, char __user *buf,
 297                              size_t count, loff_t *pos)
 298 {
 299         struct fsnotify_group *group;
 300         struct fsnotify_event *kevent;
 301         char __user *start;
 302         int ret;
 303         DEFINE_WAIT(wait);
 304
 305         start = buf;
 306         group = file->private_data;
 307
 308         pr_debug("%s: group=%p\n", __func__, group);
 309
 310         while (1) {
 311                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 312
 313                 mutex_lock(&group->notification_mutex);
 314                 kevent = get_one_event(group, count);
 315                 mutex_unlock(&group->notification_mutex);
 316
 317                 if (kevent) {
 318                         ret = PTR_ERR(kevent);
 319                         if (IS_ERR(kevent))
 320                                 break;
 321                         ret = copy_event_to_user(group, kevent, buf);
 322                         fsnotify_put_event(kevent);
 323                         if (ret < 0)
 324                                 break;
 325                         buf += ret;
 326                         count -= ret;
 327                         continue;
 328                 }
 329
 330                 ret = -EAGAIN;
 331                 if (file->f_flags & O_NONBLOCK)
 332                         break;
 333                 ret = -ERESTARTSYS;
 334                 if (signal_pending(current))
 335                         break;
 336
 337                 if (start != buf)
 338                         break;
 339
 340                 schedule();
 341         }
 342
 343         finish_wait(&group->notification_waitq, &wait);
 344         if (start != buf && ret != -EFAULT)
 345                 ret = buf - start;
 346         return ret;
 347 }
 348
 349 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 350 {
 351 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 352         struct fanotify_response response = { .fd = -1, .response = -1 };
 353         struct fsnotify_group *group;
 354         int ret;
 355
 356         group = file->private_data;
 357
 358         if (count > sizeof(response))
 359                 count = sizeof(response);
 360
 361         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 362
 363         if (copy_from_user(&response, buf, count))
 364                 return -EFAULT;
 365
 366         ret = process_access_response(group, &response);
 367         if (ret < 0)
 368                 count = ret;
 369
 370         return count;
 371 #else
 372         return -EINVAL;
 373 #endif
 374 }
 375
 376 static int fanotify_release(struct inode *ignored, struct file *file)
 377 {
 378         struct fsnotify_group *group = file->private_data;
 379
 380 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 381         struct fanotify_response_event *re, *lre;
 382
 383         mutex_lock(&group->fanotify_data.access_mutex);
 384
 385         atomic_inc(&group->fanotify_data.bypass_perm);
 386
 387         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 388                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 389                          re, re->event);
 390
 391                 list_del_init(&re->list);
 392                 re->event->response = FAN_ALLOW;
 393
 394                 kmem_cache_free(fanotify_response_event_cache, re);
 395         }
 396         mutex_unlock(&group->fanotify_data.access_mutex);
 397
 398         wake_up(&group->fanotify_data.access_waitq);
 399 #endif
 400
 401         if (file->f_flags & FASYNC)
 402                 fsnotify_fasync(-1, file, 0);
 403
 404         /* matches the fanotify_init->fsnotify_alloc_group */
 405         fsnotify_destroy_group(group);
 406
 407         return 0;
 408 }
 409
 410 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 411 {
 412         struct fsnotify_group *group;
 413         struct fsnotify_event_holder *holder;
 414         void __user *p;
 415         int ret = -ENOTTY;
 416         size_t send_len = 0;
 417
 418         group = file->private_data;
 419
 420         p = (void __user *) arg;
 421
 422         switch (cmd) {
 423         case FIONREAD:
 424                 mutex_lock(&group->notification_mutex);
 425                 list_for_each_entry(holder, &group->notification_list, event_list)
 426                         send_len += FAN_EVENT_METADATA_LEN;
 427                 mutex_unlock(&group->notification_mutex);
 428                 ret = put_user(send_len, (int __user *) p);
 429                 break;
 430         }
 431
 432         return ret;
 433 }
 434
 435 static const struct file_operations fanotify_fops = {
 436         .show_fdinfo    = fanotify_show_fdinfo,
 437         .poll           = fanotify_poll,
 438         .read           = fanotify_read,
 439         .write          = fanotify_write,
 440         .fasync         = NULL,
 441         .release        = fanotify_release,
 442         .unlocked_ioctl = fanotify_ioctl,
 443         .compat_ioctl   = fanotify_ioctl,
 444         .llseek         = noop_llseek,
 445 };
 446
 447 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 448 {
 449         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 450 }
 451
 452 static int fanotify_find_path(int dfd, const char __user *filename,
 453                               struct path *path, unsigned int flags)
 454 {
 455         int ret;
 456
 457         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 458                  dfd, filename, flags);
 459
 460         if (filename == NULL) {
 461                 struct fd f = fdget(dfd);
 462
 463                 ret = -EBADF;
 464                 if (!f.file)
 465                         goto out;
 466
 467                 ret = -ENOTDIR;
 468                 if ((flags & FAN_MARK_ONLYDIR) &&
 469                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 470                         fdput(f);
 471                         goto out;
 472                 }
 473
 474                 *path = f.file->f_path;
 475                 path_get(path);
 476                 fdput(f);
 477         } else {
 478                 unsigned int lookup_flags = 0;
 479
 480                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 481                         lookup_flags |= LOOKUP_FOLLOW;
 482                 if (flags & FAN_MARK_ONLYDIR)
 483                         lookup_flags |= LOOKUP_DIRECTORY;
 484
 485                 ret = user_path_at(dfd, filename, lookup_flags, path);
 486                 if (ret)
 487                         goto out;
 488         }
 489
 490         /* you can only watch an inode if you have read permissions on it */
 491         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 492         if (ret)
 493                 path_put(path);
 494 out:
 495         return ret;
 496 }
 497
 498 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 499                                             __u32 mask,
 500                                             unsigned int flags,
 501                                             int *destroy)
 502 {
 503         __u32 oldmask;
 504
 505         spin_lock(&fsn_mark->lock);
 506         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 507                 oldmask = fsn_mark->mask;
 508                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 509         } else {
 510                 oldmask = fsn_mark->ignored_mask;
 511                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 512         }
 513         spin_unlock(&fsn_mark->lock);
 514
 515         *destroy = !(oldmask & ~mask);
 516
 517         return mask & oldmask;
 518 }
 519
 520 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 521                                          struct vfsmount *mnt, __u32 mask,
 522                                          unsigned int flags)
 523 {
 524         struct fsnotify_mark *fsn_mark = NULL;
 525         __u32 removed;
 526         int destroy_mark;
 527
 528         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 529         if (!fsn_mark)
 530                 return -ENOENT;
 531
 532         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 533                                                  &destroy_mark);
 534         if (destroy_mark)
 535                 fsnotify_destroy_mark(fsn_mark, group);
 536
 537         fsnotify_put_mark(fsn_mark);
 538         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 539                 fsnotify_recalc_vfsmount_mask(mnt);
 540
 541         return 0;
 542 }
 543
 544 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 545                                       struct inode *inode, __u32 mask,
 546                                       unsigned int flags)
 547 {
 548         struct fsnotify_mark *fsn_mark = NULL;
 549         __u32 removed;
 550         int destroy_mark;
 551
 552         fsn_mark = fsnotify_find_inode_mark(group, inode);
 553         if (!fsn_mark)
 554                 return -ENOENT;
 555
 556         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 557                                                  &destroy_mark);
 558         if (destroy_mark)
 559                 fsnotify_destroy_mark(fsn_mark, group);
 560         /* matches the fsnotify_find_inode_mark() */
 561         fsnotify_put_mark(fsn_mark);
 562         if (removed & inode->i_fsnotify_mask)
 563                 fsnotify_recalc_inode_mask(inode);
 564
 565         return 0;
 566 }
 567
 568 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 569                                        __u32 mask,
 570                                        unsigned int flags)
 571 {
 572         __u32 oldmask = -1;
 573
 574         spin_lock(&fsn_mark->lock);
 575         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 576                 oldmask = fsn_mark->mask;
 577                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 578         } else {
 579                 __u32 tmask = fsn_mark->ignored_mask | mask;
 580                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 581                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 582                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 583         }
 584
 585         if (!(flags & FAN_MARK_ONDIR)) {
 586                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 587                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 588         }
 589
 590         spin_unlock(&fsn_mark->lock);
 591
 592         return mask & ~oldmask;
 593 }
 594
 595 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 596                                       struct vfsmount *mnt, __u32 mask,
 597                                       unsigned int flags)
 598 {
 599         struct fsnotify_mark *fsn_mark;
 600         __u32 added;
 601         int ret = 0;
 602
 603         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 604         if (!fsn_mark) {
 605                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 606                         return -ENOSPC;
 607
 608                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 609                 if (!fsn_mark)
 610                         return -ENOMEM;
 611
 612                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 613                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 614                 if (ret)
 615                         goto err;
 616         }
 617         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 618
 619         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 620                 fsnotify_recalc_vfsmount_mask(mnt);
 621 err:
 622         fsnotify_put_mark(fsn_mark);
 623         return ret;
 624 }
 625
 626 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 627                                    struct inode *inode, __u32 mask,
 628                                    unsigned int flags)
 629 {
 630         struct fsnotify_mark *fsn_mark;
 631         __u32 added;
 632         int ret = 0;
 633
 634         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 635
 636         /*
 637          * If some other task has this inode open for write we should not add
 638          * an ignored mark, unless that ignored mark is supposed to survive
 639          * modification changes anyway.
 640          */
 641         if ((flags & FAN_MARK_IGNORED_MASK) &&
 642             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 643             (atomic_read(&inode->i_writecount) > 0))
 644                 return 0;
 645
 646         fsn_mark = fsnotify_find_inode_mark(group, inode);
 647         if (!fsn_mark) {
 648                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 649                         return -ENOSPC;
 650
 651                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 652                 if (!fsn_mark)
 653                         return -ENOMEM;
 654
 655                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 656                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 657                 if (ret)
 658                         goto err;
 659         }
 660         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 661
 662         if (added & ~inode->i_fsnotify_mask)
 663                 fsnotify_recalc_inode_mask(inode);
 664 err:
 665         fsnotify_put_mark(fsn_mark);
 666         return ret;
 667 }
 668
 669 /* fanotify syscalls */
 670 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 671 {
 672         struct fsnotify_group *group;
 673         int f_flags, fd;
 674         struct user_struct *user;
 675
 676         pr_debug("%s: flags=%d event_f_flags=%d\n",
 677                 __func__, flags, event_f_flags);
 678
 679         if (!capable(CAP_SYS_ADMIN))
 680                 return -EPERM;
 681
 682         if (flags & ~FAN_ALL_INIT_FLAGS)
 683                 return -EINVAL;
 684
 685         user = get_current_user();
 686         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 687                 free_uid(user);
 688                 return -EMFILE;
 689         }
 690
 691         f_flags = O_RDWR | FMODE_NONOTIFY;
 692         if (flags & FAN_CLOEXEC)
 693                 f_flags |= O_CLOEXEC;
 694         if (flags & FAN_NONBLOCK)
 695                 f_flags |= O_NONBLOCK;
 696
 697         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 698         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 699         if (IS_ERR(group)) {
 700                 free_uid(user);
 701                 return PTR_ERR(group);
 702         }
 703
 704         group->fanotify_data.user = user;
 705         atomic_inc(&user->fanotify_listeners);
 706
 707         group->fanotify_data.f_flags = event_f_flags;
 708 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 709         mutex_init(&group->fanotify_data.access_mutex);
 710         init_waitqueue_head(&group->fanotify_data.access_waitq);
 711         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 712         atomic_set(&group->fanotify_data.bypass_perm, 0);
 713 #endif
 714         switch (flags & FAN_ALL_CLASS_BITS) {
 715         case FAN_CLASS_NOTIF:
 716                 group->priority = FS_PRIO_0;
 717                 break;
 718         case FAN_CLASS_CONTENT:
 719                 group->priority = FS_PRIO_1;
 720                 break;
 721         case FAN_CLASS_PRE_CONTENT:
 722                 group->priority = FS_PRIO_2;
 723                 break;
 724         default:
 725                 fd = -EINVAL;
 726                 goto out_destroy_group;
 727         }
 728
 729         if (flags & FAN_UNLIMITED_QUEUE) {
 730                 fd = -EPERM;
 731                 if (!capable(CAP_SYS_ADMIN))
 732                         goto out_destroy_group;
 733                 group->max_events = UINT_MAX;
 734         } else {
 735                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 736         }
 737
 738         if (flags & FAN_UNLIMITED_MARKS) {
 739                 fd = -EPERM;
 740                 if (!capable(CAP_SYS_ADMIN))
 741                         goto out_destroy_group;
 742                 group->fanotify_data.max_marks = UINT_MAX;
 743         } else {
 744                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 745         }
 746
 747         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 748         if (fd < 0)
 749                 goto out_destroy_group;
 750
 751         return fd;
 752
 753 out_destroy_group:
 754         fsnotify_destroy_group(group);
 755         return fd;
 756 }
 757
 758 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 759                               __u64 mask, int dfd,
 760                               const char  __user * pathname)
 761 {
 762         struct inode *inode = NULL;
 763         struct vfsmount *mnt = NULL;
 764         struct fsnotify_group *group;
 765         struct fd f;
 766         struct path path;
 767         int ret;
 768
 769         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 770                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 771
 772         /* we only use the lower 32 bits as of right now. */
 773         if (mask & ((__u64)0xffffffff << 32))
 774                 return -EINVAL;
 775
 776         if (flags & ~FAN_ALL_MARK_FLAGS)
 777                 return -EINVAL;
 778         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 779         case FAN_MARK_ADD:              /* fallthrough */
 780         case FAN_MARK_REMOVE:
 781                 if (!mask)
 782                         return -EINVAL;
 783         case FAN_MARK_FLUSH:
 784                 break;
 785         default:
 786                 return -EINVAL;
 787         }
 788
 789         if (mask & FAN_ONDIR) {
 790                 flags |= FAN_MARK_ONDIR;
 791                 mask &= ~FAN_ONDIR;
 792         }
 793
 794 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 795         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 796 #else
 797         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 798 #endif
 799                 return -EINVAL;
 800
 801         f = fdget(fanotify_fd);
 802         if (unlikely(!f.file))
 803                 return -EBADF;
 804
 805         /* verify that this is indeed an fanotify instance */
 806         ret = -EINVAL;
 807         if (unlikely(f.file->f_op != &fanotify_fops))
 808                 goto fput_and_out;
 809         group = f.file->private_data;
 810
 811         /*
 812          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 813          * allowed to set permissions events.
 814          */
 815         ret = -EINVAL;
 816         if (mask & FAN_ALL_PERM_EVENTS &&
 817             group->priority == FS_PRIO_0)
 818                 goto fput_and_out;
 819
 820         ret = fanotify_find_path(dfd, pathname, &path, flags);
 821         if (ret)
 822                 goto fput_and_out;
 823
 824         /* inode held in place by reference to path; group by fget on fd */
 825         if (!(flags & FAN_MARK_MOUNT))
 826                 inode = path.dentry->d_inode;
 827         else
 828                 mnt = path.mnt;
 829
 830         /* create/update an inode mark */
 831         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 832         case FAN_MARK_ADD:
 833                 if (flags & FAN_MARK_MOUNT)
 834                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 835                 else
 836                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 837                 break;
 838         case FAN_MARK_REMOVE:
 839                 if (flags & FAN_MARK_MOUNT)
 840                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 841                 else
 842                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 843                 break;
 844         case FAN_MARK_FLUSH:
 845                 if (flags & FAN_MARK_MOUNT)
 846                         fsnotify_clear_vfsmount_marks_by_group(group);
 847                 else
 848                         fsnotify_clear_inode_marks_by_group(group);
 849                 break;
 850         default:
 851                 ret = -EINVAL;
 852         }
 853
 854         path_put(&path);
 855 fput_and_out:
 856         fdput(f);
 857         return ret;
 858 }
 859
 860 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 861 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 862                                   long dfd, long pathname)
 863 {
 864         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 865                                   mask, (int) dfd,
 866                                   (const char  __user *) pathname);
 867 }
 868 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 869 #endif
 870
 871 /*
 872  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 873  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 874  * must result in panic().
 875  */
 876 static int __init fanotify_user_setup(void)
 877 {
 878         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 879         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 880                                                    SLAB_PANIC);
 881
 882         return 0;
 883 }
 884 device_initcall(fanotify_user_setup);