fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/compat.h>
  17
  18 #include <asm/ioctls.h>
  19
  20 #include "../../mount.h"
  21 #include "../fdinfo.h"
  22 #include "fanotify.h"
  23
  24 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  25 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  26 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  27
  28 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  29
  30 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  31 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  32 struct kmem_cache *fanotify_event_cachep __read_mostly;
  33
  34 struct fanotify_response_event {
  35         struct list_head list;
  36         __s32 fd;
  37         struct fanotify_event_info *event;
  38 };
  39
  40 /*
  41  * Get an fsnotify notification event if one exists and is small
  42  * enough to fit in "count". Return an error pointer if the count
  43  * is not large enough.
  44  *
  45  * Called with the group->notification_mutex held.
  46  */
  47 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  48                                             size_t count)
  49 {
  50         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  51
  52         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  53
  54         if (fsnotify_notify_queue_is_empty(group))
  55                 return NULL;
  56
  57         if (FAN_EVENT_METADATA_LEN > count)
  58                 return ERR_PTR(-EINVAL);
  59
  60         /* held the notification_mutex the whole time, so this is the
  61          * same event we peeked above */
  62         return fsnotify_remove_notify_event(group);
  63 }
  64
  65 static int create_fd(struct fsnotify_group *group,
  66                      struct fanotify_event_info *event,
  67                      struct file **file)
  68 {
  69         int client_fd;
  70         struct file *new_file;
  71
  72         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  73
  74         client_fd = get_unused_fd();
  75         if (client_fd < 0)
  76                 return client_fd;
  77
  78         /*
  79          * we need a new file handle for the userspace program so it can read even if it was
  80          * originally opened O_WRONLY.
  81          */
  82         /* it's possible this event was an overflow event.  in that case dentry and mnt
  83          * are NULL;  That's fine, just don't call dentry open */
  84         if (event->path.dentry && event->path.mnt)
  85                 new_file = dentry_open(&event->path,
  86                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  87                                        current_cred());
  88         else
  89                 new_file = ERR_PTR(-EOVERFLOW);
  90         if (IS_ERR(new_file)) {
  91                 /*
  92                  * we still send an event even if we can't open the file.  this
  93                  * can happen when say tasks are gone and we try to open their
  94                  * /proc files or we try to open a WRONLY file like in sysfs
  95                  * we just send the errno to userspace since there isn't much
  96                  * else we can do.
  97                  */
  98                 put_unused_fd(client_fd);
  99                 client_fd = PTR_ERR(new_file);
 100         } else {
 101                 *file = new_file;
 102         }
 103
 104         return client_fd;
 105 }
 106
 107 static int fill_event_metadata(struct fsnotify_group *group,
 108                                struct fanotify_event_metadata *metadata,
 109                                struct fsnotify_event *fsn_event,
 110                                struct file **file)
 111 {
 112         int ret = 0;
 113         struct fanotify_event_info *event;
 114
 115         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 116                  group, metadata, fsn_event);
 117
 118         *file = NULL;
 119         event = container_of(fsn_event, struct fanotify_event_info, fse);
 120         metadata->event_len = FAN_EVENT_METADATA_LEN;
 121         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 122         metadata->vers = FANOTIFY_METADATA_VERSION;
 123         metadata->reserved = 0;
 124         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 125         metadata->pid = pid_vnr(event->tgid);
 126         if (unlikely(fsn_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 = FANOTIFY_E(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                 FANOTIFY_E(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                 FANOTIFY_E(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                         /*
 323                          * Permission events get destroyed after we
 324                          * receive response
 325                          */
 326                         if (!(kevent->mask & FAN_ALL_PERM_EVENTS))
 327                                 fsnotify_destroy_event(group, kevent);
 328                         if (ret < 0)
 329                                 break;
 330                         buf += ret;
 331                         count -= ret;
 332                         continue;
 333                 }
 334
 335                 ret = -EAGAIN;
 336                 if (file->f_flags & O_NONBLOCK)
 337                         break;
 338                 ret = -ERESTARTSYS;
 339                 if (signal_pending(current))
 340                         break;
 341
 342                 if (start != buf)
 343                         break;
 344
 345                 schedule();
 346         }
 347
 348         finish_wait(&group->notification_waitq, &wait);
 349         if (start != buf && ret != -EFAULT)
 350                 ret = buf - start;
 351         return ret;
 352 }
 353
 354 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 355 {
 356 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 357         struct fanotify_response response = { .fd = -1, .response = -1 };
 358         struct fsnotify_group *group;
 359         int ret;
 360
 361         group = file->private_data;
 362
 363         if (count > sizeof(response))
 364                 count = sizeof(response);
 365
 366         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 367
 368         if (copy_from_user(&response, buf, count))
 369                 return -EFAULT;
 370
 371         ret = process_access_response(group, &response);
 372         if (ret < 0)
 373                 count = ret;
 374
 375         return count;
 376 #else
 377         return -EINVAL;
 378 #endif
 379 }
 380
 381 static int fanotify_release(struct inode *ignored, struct file *file)
 382 {
 383         struct fsnotify_group *group = file->private_data;
 384
 385 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 386         struct fanotify_response_event *re, *lre;
 387
 388         mutex_lock(&group->fanotify_data.access_mutex);
 389
 390         atomic_inc(&group->fanotify_data.bypass_perm);
 391
 392         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 393                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 394                          re, re->event);
 395
 396                 list_del_init(&re->list);
 397                 re->event->response = FAN_ALLOW;
 398
 399                 kmem_cache_free(fanotify_response_event_cache, re);
 400         }
 401         mutex_unlock(&group->fanotify_data.access_mutex);
 402
 403         wake_up(&group->fanotify_data.access_waitq);
 404 #endif
 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 *fsn_event;
 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(fsn_event, &group->notification_list, 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         mutex_lock(&group->mark_mutex);
 531         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 532         if (!fsn_mark) {
 533                 mutex_unlock(&group->mark_mutex);
 534                 return -ENOENT;
 535         }
 536
 537         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 538                                                  &destroy_mark);
 539         if (destroy_mark)
 540                 fsnotify_destroy_mark_locked(fsn_mark, group);
 541         mutex_unlock(&group->mark_mutex);
 542
 543         fsnotify_put_mark(fsn_mark);
 544         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 545                 fsnotify_recalc_vfsmount_mask(mnt);
 546
 547         return 0;
 548 }
 549
 550 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 551                                       struct inode *inode, __u32 mask,
 552                                       unsigned int flags)
 553 {
 554         struct fsnotify_mark *fsn_mark = NULL;
 555         __u32 removed;
 556         int destroy_mark;
 557
 558         mutex_lock(&group->mark_mutex);
 559         fsn_mark = fsnotify_find_inode_mark(group, inode);
 560         if (!fsn_mark) {
 561                 mutex_unlock(&group->mark_mutex);
 562                 return -ENOENT;
 563         }
 564
 565         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 566                                                  &destroy_mark);
 567         if (destroy_mark)
 568                 fsnotify_destroy_mark_locked(fsn_mark, group);
 569         mutex_unlock(&group->mark_mutex);
 570
 571         /* matches the fsnotify_find_inode_mark() */
 572         fsnotify_put_mark(fsn_mark);
 573         if (removed & inode->i_fsnotify_mask)
 574                 fsnotify_recalc_inode_mask(inode);
 575
 576         return 0;
 577 }
 578
 579 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 580                                        __u32 mask,
 581                                        unsigned int flags)
 582 {
 583         __u32 oldmask = -1;
 584
 585         spin_lock(&fsn_mark->lock);
 586         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 587                 oldmask = fsn_mark->mask;
 588                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 589         } else {
 590                 __u32 tmask = fsn_mark->ignored_mask | mask;
 591                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 592                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 593                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 594         }
 595
 596         if (!(flags & FAN_MARK_ONDIR)) {
 597                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 598                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 599         }
 600
 601         spin_unlock(&fsn_mark->lock);
 602
 603         return mask & ~oldmask;
 604 }
 605
 606 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 607                                                    struct inode *inode,
 608                                                    struct vfsmount *mnt)
 609 {
 610         struct fsnotify_mark *mark;
 611         int ret;
 612
 613         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 614                 return ERR_PTR(-ENOSPC);
 615
 616         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 617         if (!mark)
 618                 return ERR_PTR(-ENOMEM);
 619
 620         fsnotify_init_mark(mark, fanotify_free_mark);
 621         ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
 622         if (ret) {
 623                 fsnotify_put_mark(mark);
 624                 return ERR_PTR(ret);
 625         }
 626
 627         return mark;
 628 }
 629
 630
 631 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 632                                       struct vfsmount *mnt, __u32 mask,
 633                                       unsigned int flags)
 634 {
 635         struct fsnotify_mark *fsn_mark;
 636         __u32 added;
 637
 638         mutex_lock(&group->mark_mutex);
 639         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 640         if (!fsn_mark) {
 641                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 642                 if (IS_ERR(fsn_mark)) {
 643                         mutex_unlock(&group->mark_mutex);
 644                         return PTR_ERR(fsn_mark);
 645                 }
 646         }
 647         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 648         mutex_unlock(&group->mark_mutex);
 649
 650         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 651                 fsnotify_recalc_vfsmount_mask(mnt);
 652
 653         fsnotify_put_mark(fsn_mark);
 654         return 0;
 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
 664         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 665
 666         /*
 667          * If some other task has this inode open for write we should not add
 668          * an ignored mark, unless that ignored mark is supposed to survive
 669          * modification changes anyway.
 670          */
 671         if ((flags & FAN_MARK_IGNORED_MASK) &&
 672             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 673             (atomic_read(&inode->i_writecount) > 0))
 674                 return 0;
 675
 676         mutex_lock(&group->mark_mutex);
 677         fsn_mark = fsnotify_find_inode_mark(group, inode);
 678         if (!fsn_mark) {
 679                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 680                 if (IS_ERR(fsn_mark)) {
 681                         mutex_unlock(&group->mark_mutex);
 682                         return PTR_ERR(fsn_mark);
 683                 }
 684         }
 685         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 686         mutex_unlock(&group->mark_mutex);
 687
 688         if (added & ~inode->i_fsnotify_mask)
 689                 fsnotify_recalc_inode_mask(inode);
 690
 691         fsnotify_put_mark(fsn_mark);
 692         return 0;
 693 }
 694
 695 /* fanotify syscalls */
 696 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 697 {
 698         struct fsnotify_group *group;
 699         int f_flags, fd;
 700         struct user_struct *user;
 701         struct fanotify_event_info *oevent;
 702
 703         pr_debug("%s: flags=%d event_f_flags=%d\n",
 704                 __func__, flags, event_f_flags);
 705
 706         if (!capable(CAP_SYS_ADMIN))
 707                 return -EPERM;
 708
 709         if (flags & ~FAN_ALL_INIT_FLAGS)
 710                 return -EINVAL;
 711
 712         user = get_current_user();
 713         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 714                 free_uid(user);
 715                 return -EMFILE;
 716         }
 717
 718         f_flags = O_RDWR | FMODE_NONOTIFY;
 719         if (flags & FAN_CLOEXEC)
 720                 f_flags |= O_CLOEXEC;
 721         if (flags & FAN_NONBLOCK)
 722                 f_flags |= O_NONBLOCK;
 723
 724         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 725         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 726         if (IS_ERR(group)) {
 727                 free_uid(user);
 728                 return PTR_ERR(group);
 729         }
 730
 731         group->fanotify_data.user = user;
 732         atomic_inc(&user->fanotify_listeners);
 733
 734         oevent = kmem_cache_alloc(fanotify_event_cachep, GFP_KERNEL);
 735         if (unlikely(!oevent)) {
 736                 fd = -ENOMEM;
 737                 goto out_destroy_group;
 738         }
 739         group->overflow_event = &oevent->fse;
 740         fsnotify_init_event(group->overflow_event, NULL, FS_Q_OVERFLOW);
 741         oevent->tgid = get_pid(task_tgid(current));
 742         oevent->path.mnt = NULL;
 743         oevent->path.dentry = NULL;
 744
 745         group->fanotify_data.f_flags = event_f_flags;
 746 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 747         oevent->response = 0;
 748         mutex_init(&group->fanotify_data.access_mutex);
 749         init_waitqueue_head(&group->fanotify_data.access_waitq);
 750         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 751         atomic_set(&group->fanotify_data.bypass_perm, 0);
 752 #endif
 753         switch (flags & FAN_ALL_CLASS_BITS) {
 754         case FAN_CLASS_NOTIF:
 755                 group->priority = FS_PRIO_0;
 756                 break;
 757         case FAN_CLASS_CONTENT:
 758                 group->priority = FS_PRIO_1;
 759                 break;
 760         case FAN_CLASS_PRE_CONTENT:
 761                 group->priority = FS_PRIO_2;
 762                 break;
 763         default:
 764                 fd = -EINVAL;
 765                 goto out_destroy_group;
 766         }
 767
 768         if (flags & FAN_UNLIMITED_QUEUE) {
 769                 fd = -EPERM;
 770                 if (!capable(CAP_SYS_ADMIN))
 771                         goto out_destroy_group;
 772                 group->max_events = UINT_MAX;
 773         } else {
 774                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 775         }
 776
 777         if (flags & FAN_UNLIMITED_MARKS) {
 778                 fd = -EPERM;
 779                 if (!capable(CAP_SYS_ADMIN))
 780                         goto out_destroy_group;
 781                 group->fanotify_data.max_marks = UINT_MAX;
 782         } else {
 783                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 784         }
 785
 786         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 787         if (fd < 0)
 788                 goto out_destroy_group;
 789
 790         return fd;
 791
 792 out_destroy_group:
 793         fsnotify_destroy_group(group);
 794         return fd;
 795 }
 796
 797 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 798                               __u64, mask, int, dfd,
 799                               const char  __user *, pathname)
 800 {
 801         struct inode *inode = NULL;
 802         struct vfsmount *mnt = NULL;
 803         struct fsnotify_group *group;
 804         struct fd f;
 805         struct path path;
 806         int ret;
 807
 808         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 809                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 810
 811         /* we only use the lower 32 bits as of right now. */
 812         if (mask & ((__u64)0xffffffff << 32))
 813                 return -EINVAL;
 814
 815         if (flags & ~FAN_ALL_MARK_FLAGS)
 816                 return -EINVAL;
 817         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 818         case FAN_MARK_ADD:              /* fallthrough */
 819         case FAN_MARK_REMOVE:
 820                 if (!mask)
 821                         return -EINVAL;
 822         case FAN_MARK_FLUSH:
 823                 break;
 824         default:
 825                 return -EINVAL;
 826         }
 827
 828         if (mask & FAN_ONDIR) {
 829                 flags |= FAN_MARK_ONDIR;
 830                 mask &= ~FAN_ONDIR;
 831         }
 832
 833 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 834         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 835 #else
 836         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 837 #endif
 838                 return -EINVAL;
 839
 840         f = fdget(fanotify_fd);
 841         if (unlikely(!f.file))
 842                 return -EBADF;
 843
 844         /* verify that this is indeed an fanotify instance */
 845         ret = -EINVAL;
 846         if (unlikely(f.file->f_op != &fanotify_fops))
 847                 goto fput_and_out;
 848         group = f.file->private_data;
 849
 850         /*
 851          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 852          * allowed to set permissions events.
 853          */
 854         ret = -EINVAL;
 855         if (mask & FAN_ALL_PERM_EVENTS &&
 856             group->priority == FS_PRIO_0)
 857                 goto fput_and_out;
 858
 859         ret = fanotify_find_path(dfd, pathname, &path, flags);
 860         if (ret)
 861                 goto fput_and_out;
 862
 863         /* inode held in place by reference to path; group by fget on fd */
 864         if (!(flags & FAN_MARK_MOUNT))
 865                 inode = path.dentry->d_inode;
 866         else
 867                 mnt = path.mnt;
 868
 869         /* create/update an inode mark */
 870         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 871         case FAN_MARK_ADD:
 872                 if (flags & FAN_MARK_MOUNT)
 873                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 874                 else
 875                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 876                 break;
 877         case FAN_MARK_REMOVE:
 878                 if (flags & FAN_MARK_MOUNT)
 879                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 880                 else
 881                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 882                 break;
 883         case FAN_MARK_FLUSH:
 884                 if (flags & FAN_MARK_MOUNT)
 885                         fsnotify_clear_vfsmount_marks_by_group(group);
 886                 else
 887                         fsnotify_clear_inode_marks_by_group(group);
 888                 break;
 889         default:
 890                 ret = -EINVAL;
 891         }
 892
 893         path_put(&path);
 894 fput_and_out:
 895         fdput(f);
 896         return ret;
 897 }
 898
 899 #ifdef CONFIG_COMPAT
 900 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 901                                 int, fanotify_fd, unsigned int, flags,
 902                                 __u32, mask0, __u32, mask1, int, dfd,
 903                                 const char  __user *, pathname)
 904 {
 905         return sys_fanotify_mark(fanotify_fd, flags,
 906 #ifdef __BIG_ENDIAN
 907                                 ((__u64)mask0 << 32) | mask1,
 908 #else
 909                                 ((__u64)mask1 << 32) | mask0,
 910 #endif
 911                                  dfd, pathname);
 912 }
 913 #endif
 914
 915 /*
 916  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 917  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 918  * must result in panic().
 919  */
 920 static int __init fanotify_user_setup(void)
 921 {
 922         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 923         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 924                                                    SLAB_PANIC);
 925         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 926
 927         return 0;
 928 }
 929 device_initcall(fanotify_user_setup);