fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16
  17 #include <asm/ioctls.h>
  18
  19 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  20 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  21 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  22
  23 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  24
  25 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  26 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  27
  28 struct fanotify_response_event {
  29         struct list_head list;
  30         __s32 fd;
  31         struct fsnotify_event *event;
  32 };
  33
  34 /*
  35  * Get an fsnotify notification event if one exists and is small
  36  * enough to fit in "count". Return an error pointer if the count
  37  * is not large enough.
  38  *
  39  * Called with the group->notification_mutex held.
  40  */
  41 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  42                                             size_t count)
  43 {
  44         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  45
  46         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  47
  48         if (fsnotify_notify_queue_is_empty(group))
  49                 return NULL;
  50
  51         if (FAN_EVENT_METADATA_LEN > count)
  52                 return ERR_PTR(-EINVAL);
  53
  54         /* held the notification_mutex the whole time, so this is the
  55          * same event we peeked above */
  56         return fsnotify_remove_notify_event(group);
  57 }
  58
  59 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
  60 {
  61         int client_fd;
  62         struct dentry *dentry;
  63         struct vfsmount *mnt;
  64         struct file *new_file;
  65
  66         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  67
  68         client_fd = get_unused_fd();
  69         if (client_fd < 0)
  70                 return client_fd;
  71
  72         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  73                 WARN_ON(1);
  74                 put_unused_fd(client_fd);
  75                 return -EINVAL;
  76         }
  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         dentry = dget(event->path.dentry);
  83         mnt = mntget(event->path.mnt);
  84         /* it's possible this event was an overflow event.  in that case dentry and mnt
  85          * are NULL;  That's fine, just don't call dentry open */
  86         if (dentry && mnt)
  87                 new_file = dentry_open(dentry, mnt,
  88                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  89                                        current_cred());
  90         else
  91                 new_file = ERR_PTR(-EOVERFLOW);
  92         if (IS_ERR(new_file)) {
  93                 /*
  94                  * we still send an event even if we can't open the file.  this
  95                  * can happen when say tasks are gone and we try to open their
  96                  * /proc files or we try to open a WRONLY file like in sysfs
  97                  * we just send the errno to userspace since there isn't much
  98                  * else we can do.
  99                  */
 100                 put_unused_fd(client_fd);
 101                 client_fd = PTR_ERR(new_file);
 102         } else {
 103                 fd_install(client_fd, new_file);
 104         }
 105
 106         return client_fd;
 107 }
 108
 109 static ssize_t fill_event_metadata(struct fsnotify_group *group,
 110                                    struct fanotify_event_metadata *metadata,
 111                                    struct fsnotify_event *event)
 112 {
 113         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 114                  group, metadata, event);
 115
 116         metadata->event_len = FAN_EVENT_METADATA_LEN;
 117         metadata->vers = FANOTIFY_METADATA_VERSION;
 118         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 119         metadata->pid = pid_vnr(event->tgid);
 120         metadata->fd = create_fd(group, event);
 121
 122         return metadata->fd;
 123 }
 124
 125 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 126 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 127                                                   __s32 fd)
 128 {
 129         struct fanotify_response_event *re, *return_re = NULL;
 130
 131         mutex_lock(&group->fanotify_data.access_mutex);
 132         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 133                 if (re->fd != fd)
 134                         continue;
 135
 136                 list_del_init(&re->list);
 137                 return_re = re;
 138                 break;
 139         }
 140         mutex_unlock(&group->fanotify_data.access_mutex);
 141
 142         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 143
 144         return return_re;
 145 }
 146
 147 static int process_access_response(struct fsnotify_group *group,
 148                                    struct fanotify_response *response_struct)
 149 {
 150         struct fanotify_response_event *re;
 151         __s32 fd = response_struct->fd;
 152         __u32 response = response_struct->response;
 153
 154         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 155                  fd, response);
 156         /*
 157          * make sure the response is valid, if invalid we do nothing and either
 158          * userspace can send a valid responce or we will clean it up after the
 159          * timeout
 160          */
 161         switch (response) {
 162         case FAN_ALLOW:
 163         case FAN_DENY:
 164                 break;
 165         default:
 166                 return -EINVAL;
 167         }
 168
 169         if (fd < 0)
 170                 return -EINVAL;
 171
 172         re = dequeue_re(group, fd);
 173         if (!re)
 174                 return -ENOENT;
 175
 176         re->event->response = response;
 177
 178         wake_up(&group->fanotify_data.access_waitq);
 179
 180         kmem_cache_free(fanotify_response_event_cache, re);
 181
 182         return 0;
 183 }
 184
 185 static int prepare_for_access_response(struct fsnotify_group *group,
 186                                        struct fsnotify_event *event,
 187                                        __s32 fd)
 188 {
 189         struct fanotify_response_event *re;
 190
 191         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 192                 return 0;
 193
 194         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 195         if (!re)
 196                 return -ENOMEM;
 197
 198         re->event = event;
 199         re->fd = fd;
 200
 201         mutex_lock(&group->fanotify_data.access_mutex);
 202
 203         if (group->fanotify_data.bypass_perm) {
 204                 mutex_unlock(&group->fanotify_data.access_mutex);
 205                 kmem_cache_free(fanotify_response_event_cache, re);
 206                 event->response = FAN_ALLOW;
 207                 return 0;
 208         }
 209
 210         list_add_tail(&re->list, &group->fanotify_data.access_list);
 211         mutex_unlock(&group->fanotify_data.access_mutex);
 212
 213         return 0;
 214 }
 215
 216 static void remove_access_response(struct fsnotify_group *group,
 217                                    struct fsnotify_event *event,
 218                                    __s32 fd)
 219 {
 220         struct fanotify_response_event *re;
 221
 222         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 223                 return;
 224
 225         re = dequeue_re(group, fd);
 226         if (!re)
 227                 return;
 228
 229         BUG_ON(re->event != event);
 230
 231         kmem_cache_free(fanotify_response_event_cache, re);
 232
 233         return;
 234 }
 235 #else
 236 static int prepare_for_access_response(struct fsnotify_group *group,
 237                                        struct fsnotify_event *event,
 238                                        __s32 fd)
 239 {
 240         return 0;
 241 }
 242
 243 static void remove_access_response(struct fsnotify_group *group,
 244                                    struct fsnotify_event *event,
 245                                    __s32 fd)
 246 {
 247         return;
 248 }
 249 #endif
 250
 251 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 252                                   struct fsnotify_event *event,
 253                                   char __user *buf)
 254 {
 255         struct fanotify_event_metadata fanotify_event_metadata;
 256         int fd, ret;
 257
 258         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 259
 260         fd = fill_event_metadata(group, &fanotify_event_metadata, event);
 261         if (fd < 0)
 262                 return fd;
 263
 264         ret = prepare_for_access_response(group, event, fd);
 265         if (ret)
 266                 goto out_close_fd;
 267
 268         ret = -EFAULT;
 269         if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
 270                 goto out_kill_access_response;
 271
 272         return FAN_EVENT_METADATA_LEN;
 273
 274 out_kill_access_response:
 275         remove_access_response(group, event, fd);
 276 out_close_fd:
 277         sys_close(fd);
 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 = -EINTR;
 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         struct fanotify_response_event *re, *lre;
 380
 381         pr_debug("%s: file=%p group=%p\n", __func__, file, group);
 382
 383 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 384         mutex_lock(&group->fanotify_data.access_mutex);
 385
 386         group->fanotify_data.bypass_perm = true;
 387
 388         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 389                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 390                          re, re->event);
 391
 392                 list_del_init(&re->list);
 393                 re->event->response = FAN_ALLOW;
 394
 395                 kmem_cache_free(fanotify_response_event_cache, re);
 396         }
 397         mutex_unlock(&group->fanotify_data.access_mutex);
 398
 399         wake_up(&group->fanotify_data.access_waitq);
 400 #endif
 401         /* matches the fanotify_init->fsnotify_alloc_group */
 402         fsnotify_put_group(group);
 403
 404         return 0;
 405 }
 406
 407 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 408 {
 409         struct fsnotify_group *group;
 410         struct fsnotify_event_holder *holder;
 411         void __user *p;
 412         int ret = -ENOTTY;
 413         size_t send_len = 0;
 414
 415         group = file->private_data;
 416
 417         p = (void __user *) arg;
 418
 419         switch (cmd) {
 420         case FIONREAD:
 421                 mutex_lock(&group->notification_mutex);
 422                 list_for_each_entry(holder, &group->notification_list, event_list)
 423                         send_len += FAN_EVENT_METADATA_LEN;
 424                 mutex_unlock(&group->notification_mutex);
 425                 ret = put_user(send_len, (int __user *) p);
 426                 break;
 427         }
 428
 429         return ret;
 430 }
 431
 432 static const struct file_operations fanotify_fops = {
 433         .poll           = fanotify_poll,
 434         .read           = fanotify_read,
 435         .write          = fanotify_write,
 436         .fasync         = NULL,
 437         .release        = fanotify_release,
 438         .unlocked_ioctl = fanotify_ioctl,
 439         .compat_ioctl   = fanotify_ioctl,
 440         .llseek         = noop_llseek,
 441 };
 442
 443 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 444 {
 445         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 446 }
 447
 448 static int fanotify_find_path(int dfd, const char __user *filename,
 449                               struct path *path, unsigned int flags)
 450 {
 451         int ret;
 452
 453         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 454                  dfd, filename, flags);
 455
 456         if (filename == NULL) {
 457                 struct file *file;
 458                 int fput_needed;
 459
 460                 ret = -EBADF;
 461                 file = fget_light(dfd, &fput_needed);
 462                 if (!file)
 463                         goto out;
 464
 465                 ret = -ENOTDIR;
 466                 if ((flags & FAN_MARK_ONLYDIR) &&
 467                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 468                         fput_light(file, fput_needed);
 469                         goto out;
 470                 }
 471
 472                 *path = file->f_path;
 473                 path_get(path);
 474                 fput_light(file, fput_needed);
 475         } else {
 476                 unsigned int lookup_flags = 0;
 477
 478                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 479                         lookup_flags |= LOOKUP_FOLLOW;
 480                 if (flags & FAN_MARK_ONLYDIR)
 481                         lookup_flags |= LOOKUP_DIRECTORY;
 482
 483                 ret = user_path_at(dfd, filename, lookup_flags, path);
 484                 if (ret)
 485                         goto out;
 486         }
 487
 488         /* you can only watch an inode if you have read permissions on it */
 489         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 490         if (ret)
 491                 path_put(path);
 492 out:
 493         return ret;
 494 }
 495
 496 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 497                                             __u32 mask,
 498                                             unsigned int flags)
 499 {
 500         __u32 oldmask;
 501
 502         spin_lock(&fsn_mark->lock);
 503         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 504                 oldmask = fsn_mark->mask;
 505                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 506         } else {
 507                 oldmask = fsn_mark->ignored_mask;
 508                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 509         }
 510         spin_unlock(&fsn_mark->lock);
 511
 512         if (!(oldmask & ~mask))
 513                 fsnotify_destroy_mark(fsn_mark);
 514
 515         return mask & oldmask;
 516 }
 517
 518 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 519                                          struct vfsmount *mnt, __u32 mask,
 520                                          unsigned int flags)
 521 {
 522         struct fsnotify_mark *fsn_mark = NULL;
 523         __u32 removed;
 524
 525         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 526         if (!fsn_mark)
 527                 return -ENOENT;
 528
 529         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 530         fsnotify_put_mark(fsn_mark);
 531         if (removed & mnt->mnt_fsnotify_mask)
 532                 fsnotify_recalc_vfsmount_mask(mnt);
 533
 534         return 0;
 535 }
 536
 537 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 538                                       struct inode *inode, __u32 mask,
 539                                       unsigned int flags)
 540 {
 541         struct fsnotify_mark *fsn_mark = NULL;
 542         __u32 removed;
 543
 544         fsn_mark = fsnotify_find_inode_mark(group, inode);
 545         if (!fsn_mark)
 546                 return -ENOENT;
 547
 548         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 549         /* matches the fsnotify_find_inode_mark() */
 550         fsnotify_put_mark(fsn_mark);
 551         if (removed & inode->i_fsnotify_mask)
 552                 fsnotify_recalc_inode_mask(inode);
 553
 554         return 0;
 555 }
 556
 557 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 558                                        __u32 mask,
 559                                        unsigned int flags)
 560 {
 561         __u32 oldmask = -1;
 562
 563         spin_lock(&fsn_mark->lock);
 564         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 565                 oldmask = fsn_mark->mask;
 566                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 567         } else {
 568                 __u32 tmask = fsn_mark->ignored_mask | mask;
 569                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 570                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 571                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 572         }
 573
 574         if (!(flags & FAN_MARK_ONDIR)) {
 575                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 576                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 577         }
 578
 579         spin_unlock(&fsn_mark->lock);
 580
 581         return mask & ~oldmask;
 582 }
 583
 584 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 585                                       struct vfsmount *mnt, __u32 mask,
 586                                       unsigned int flags)
 587 {
 588         struct fsnotify_mark *fsn_mark;
 589         __u32 added;
 590
 591         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 592         if (!fsn_mark) {
 593                 int ret;
 594
 595                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 596                         return -ENOSPC;
 597
 598                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 599                 if (!fsn_mark)
 600                         return -ENOMEM;
 601
 602                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 603                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 604                 if (ret) {
 605                         fanotify_free_mark(fsn_mark);
 606                         return ret;
 607                 }
 608         }
 609         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 610         fsnotify_put_mark(fsn_mark);
 611         if (added & ~mnt->mnt_fsnotify_mask)
 612                 fsnotify_recalc_vfsmount_mask(mnt);
 613
 614         return 0;
 615 }
 616
 617 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 618                                    struct inode *inode, __u32 mask,
 619                                    unsigned int flags)
 620 {
 621         struct fsnotify_mark *fsn_mark;
 622         __u32 added;
 623
 624         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 625
 626         /*
 627          * If some other task has this inode open for write we should not add
 628          * an ignored mark, unless that ignored mark is supposed to survive
 629          * modification changes anyway.
 630          */
 631         if ((flags & FAN_MARK_IGNORED_MASK) &&
 632             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 633             (atomic_read(&inode->i_writecount) > 0))
 634                 return 0;
 635
 636         fsn_mark = fsnotify_find_inode_mark(group, inode);
 637         if (!fsn_mark) {
 638                 int ret;
 639
 640                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 641                         return -ENOSPC;
 642
 643                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 644                 if (!fsn_mark)
 645                         return -ENOMEM;
 646
 647                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 648                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 649                 if (ret) {
 650                         fanotify_free_mark(fsn_mark);
 651                         return ret;
 652                 }
 653         }
 654         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 655         fsnotify_put_mark(fsn_mark);
 656         if (added & ~inode->i_fsnotify_mask)
 657                 fsnotify_recalc_inode_mask(inode);
 658         return 0;
 659 }
 660
 661 /* fanotify syscalls */
 662 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 663 {
 664         struct fsnotify_group *group;
 665         int f_flags, fd;
 666         struct user_struct *user;
 667
 668         pr_debug("%s: flags=%d event_f_flags=%d\n",
 669                 __func__, flags, event_f_flags);
 670
 671         if (!capable(CAP_SYS_ADMIN))
 672                 return -EPERM;
 673
 674         if (flags & ~FAN_ALL_INIT_FLAGS)
 675                 return -EINVAL;
 676
 677         user = get_current_user();
 678         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 679                 free_uid(user);
 680                 return -EMFILE;
 681         }
 682
 683         f_flags = O_RDWR | FMODE_NONOTIFY;
 684         if (flags & FAN_CLOEXEC)
 685                 f_flags |= O_CLOEXEC;
 686         if (flags & FAN_NONBLOCK)
 687                 f_flags |= O_NONBLOCK;
 688
 689         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 690         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 691         if (IS_ERR(group))
 692                 return PTR_ERR(group);
 693
 694         group->fanotify_data.user = user;
 695         atomic_inc(&user->fanotify_listeners);
 696
 697         group->fanotify_data.f_flags = event_f_flags;
 698 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 699         mutex_init(&group->fanotify_data.access_mutex);
 700         init_waitqueue_head(&group->fanotify_data.access_waitq);
 701         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 702 #endif
 703         switch (flags & FAN_ALL_CLASS_BITS) {
 704         case FAN_CLASS_NOTIF:
 705                 group->priority = FS_PRIO_0;
 706                 break;
 707         case FAN_CLASS_CONTENT:
 708                 group->priority = FS_PRIO_1;
 709                 break;
 710         case FAN_CLASS_PRE_CONTENT:
 711                 group->priority = FS_PRIO_2;
 712                 break;
 713         default:
 714                 fd = -EINVAL;
 715                 goto out_put_group;
 716         }
 717
 718         if (flags & FAN_UNLIMITED_QUEUE) {
 719                 fd = -EPERM;
 720                 if (!capable(CAP_SYS_ADMIN))
 721                         goto out_put_group;
 722                 group->max_events = UINT_MAX;
 723         } else {
 724                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 725         }
 726
 727         if (flags & FAN_UNLIMITED_MARKS) {
 728                 fd = -EPERM;
 729                 if (!capable(CAP_SYS_ADMIN))
 730                         goto out_put_group;
 731                 group->fanotify_data.max_marks = UINT_MAX;
 732         } else {
 733                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 734         }
 735
 736         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 737         if (fd < 0)
 738                 goto out_put_group;
 739
 740         return fd;
 741
 742 out_put_group:
 743         fsnotify_put_group(group);
 744         return fd;
 745 }
 746
 747 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 748                               __u64 mask, int dfd,
 749                               const char  __user * pathname)
 750 {
 751         struct inode *inode = NULL;
 752         struct vfsmount *mnt = NULL;
 753         struct fsnotify_group *group;
 754         struct file *filp;
 755         struct path path;
 756         int ret, fput_needed;
 757
 758         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 759                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 760
 761         /* we only use the lower 32 bits as of right now. */
 762         if (mask & ((__u64)0xffffffff << 32))
 763                 return -EINVAL;
 764
 765         if (flags & ~FAN_ALL_MARK_FLAGS)
 766                 return -EINVAL;
 767         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 768         case FAN_MARK_ADD:
 769         case FAN_MARK_REMOVE:
 770         case FAN_MARK_FLUSH:
 771                 break;
 772         default:
 773                 return -EINVAL;
 774         }
 775
 776         if (mask & FAN_ONDIR) {
 777                 flags |= FAN_MARK_ONDIR;
 778                 mask &= ~FAN_ONDIR;
 779         }
 780
 781 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 782         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 783 #else
 784         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 785 #endif
 786                 return -EINVAL;
 787
 788         filp = fget_light(fanotify_fd, &fput_needed);
 789         if (unlikely(!filp))
 790                 return -EBADF;
 791
 792         /* verify that this is indeed an fanotify instance */
 793         ret = -EINVAL;
 794         if (unlikely(filp->f_op != &fanotify_fops))
 795                 goto fput_and_out;
 796         group = filp->private_data;
 797
 798         /*
 799          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 800          * allowed to set permissions events.
 801          */
 802         ret = -EINVAL;
 803         if (mask & FAN_ALL_PERM_EVENTS &&
 804             group->priority == FS_PRIO_0)
 805                 goto fput_and_out;
 806
 807         ret = fanotify_find_path(dfd, pathname, &path, flags);
 808         if (ret)
 809                 goto fput_and_out;
 810
 811         /* inode held in place by reference to path; group by fget on fd */
 812         if (!(flags & FAN_MARK_MOUNT))
 813                 inode = path.dentry->d_inode;
 814         else
 815                 mnt = path.mnt;
 816
 817         /* create/update an inode mark */
 818         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 819         case FAN_MARK_ADD:
 820                 if (flags & FAN_MARK_MOUNT)
 821                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 822                 else
 823                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 824                 break;
 825         case FAN_MARK_REMOVE:
 826                 if (flags & FAN_MARK_MOUNT)
 827                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 828                 else
 829                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 830                 break;
 831         case FAN_MARK_FLUSH:
 832                 if (flags & FAN_MARK_MOUNT)
 833                         fsnotify_clear_vfsmount_marks_by_group(group);
 834                 else
 835                         fsnotify_clear_inode_marks_by_group(group);
 836                 break;
 837         default:
 838                 ret = -EINVAL;
 839         }
 840
 841         path_put(&path);
 842 fput_and_out:
 843         fput_light(filp, fput_needed);
 844         return ret;
 845 }
 846
 847 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 848 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 849                                   long dfd, long pathname)
 850 {
 851         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 852                                   mask, (int) dfd,
 853                                   (const char  __user *) pathname);
 854 }
 855 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 856 #endif
 857
 858 /*
 859  * fanotify_user_setup - Our initialization function.  Note that we cannnot return
 860  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 861  * must result in panic().
 862  */
 863 static int __init fanotify_user_setup(void)
 864 {
 865         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 866         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 867                                                    SLAB_PANIC);
 868
 869         return 0;
 870 }
 871 device_initcall(fanotify_user_setup);