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 = -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         group->fanotify_data.bypass_perm = true;
 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         /* matches the fanotify_init->fsnotify_alloc_group */
 401         fsnotify_put_group(group);
 402
 403         return 0;
 404 }
 405
 406 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 407 {
 408         struct fsnotify_group *group;
 409         struct fsnotify_event_holder *holder;
 410         void __user *p;
 411         int ret = -ENOTTY;
 412         size_t send_len = 0;
 413
 414         group = file->private_data;
 415
 416         p = (void __user *) arg;
 417
 418         switch (cmd) {
 419         case FIONREAD:
 420                 mutex_lock(&group->notification_mutex);
 421                 list_for_each_entry(holder, &group->notification_list, event_list)
 422                         send_len += FAN_EVENT_METADATA_LEN;
 423                 mutex_unlock(&group->notification_mutex);
 424                 ret = put_user(send_len, (int __user *) p);
 425                 break;
 426         }
 427
 428         return ret;
 429 }
 430
 431 static const struct file_operations fanotify_fops = {
 432         .poll           = fanotify_poll,
 433         .read           = fanotify_read,
 434         .write          = fanotify_write,
 435         .fasync         = NULL,
 436         .release        = fanotify_release,
 437         .unlocked_ioctl = fanotify_ioctl,
 438         .compat_ioctl   = fanotify_ioctl,
 439         .llseek         = noop_llseek,
 440 };
 441
 442 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 443 {
 444         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 445 }
 446
 447 static int fanotify_find_path(int dfd, const char __user *filename,
 448                               struct path *path, unsigned int flags)
 449 {
 450         int ret;
 451
 452         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 453                  dfd, filename, flags);
 454
 455         if (filename == NULL) {
 456                 struct file *file;
 457                 int fput_needed;
 458
 459                 ret = -EBADF;
 460                 file = fget_light(dfd, &fput_needed);
 461                 if (!file)
 462                         goto out;
 463
 464                 ret = -ENOTDIR;
 465                 if ((flags & FAN_MARK_ONLYDIR) &&
 466                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 467                         fput_light(file, fput_needed);
 468                         goto out;
 469                 }
 470
 471                 *path = file->f_path;
 472                 path_get(path);
 473                 fput_light(file, fput_needed);
 474         } else {
 475                 unsigned int lookup_flags = 0;
 476
 477                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 478                         lookup_flags |= LOOKUP_FOLLOW;
 479                 if (flags & FAN_MARK_ONLYDIR)
 480                         lookup_flags |= LOOKUP_DIRECTORY;
 481
 482                 ret = user_path_at(dfd, filename, lookup_flags, path);
 483                 if (ret)
 484                         goto out;
 485         }
 486
 487         /* you can only watch an inode if you have read permissions on it */
 488         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 489         if (ret)
 490                 path_put(path);
 491 out:
 492         return ret;
 493 }
 494
 495 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 496                                             __u32 mask,
 497                                             unsigned int flags)
 498 {
 499         __u32 oldmask;
 500
 501         spin_lock(&fsn_mark->lock);
 502         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 503                 oldmask = fsn_mark->mask;
 504                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 505         } else {
 506                 oldmask = fsn_mark->ignored_mask;
 507                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 508         }
 509         spin_unlock(&fsn_mark->lock);
 510
 511         if (!(oldmask & ~mask))
 512                 fsnotify_destroy_mark(fsn_mark);
 513
 514         return mask & oldmask;
 515 }
 516
 517 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 518                                          struct vfsmount *mnt, __u32 mask,
 519                                          unsigned int flags)
 520 {
 521         struct fsnotify_mark *fsn_mark = NULL;
 522         __u32 removed;
 523
 524         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 525         if (!fsn_mark)
 526                 return -ENOENT;
 527
 528         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 529         fsnotify_put_mark(fsn_mark);
 530         if (removed & mnt->mnt_fsnotify_mask)
 531                 fsnotify_recalc_vfsmount_mask(mnt);
 532
 533         return 0;
 534 }
 535
 536 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 537                                       struct inode *inode, __u32 mask,
 538                                       unsigned int flags)
 539 {
 540         struct fsnotify_mark *fsn_mark = NULL;
 541         __u32 removed;
 542
 543         fsn_mark = fsnotify_find_inode_mark(group, inode);
 544         if (!fsn_mark)
 545                 return -ENOENT;
 546
 547         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
 548         /* matches the fsnotify_find_inode_mark() */
 549         fsnotify_put_mark(fsn_mark);
 550         if (removed & inode->i_fsnotify_mask)
 551                 fsnotify_recalc_inode_mask(inode);
 552
 553         return 0;
 554 }
 555
 556 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 557                                        __u32 mask,
 558                                        unsigned int flags)
 559 {
 560         __u32 oldmask = -1;
 561
 562         spin_lock(&fsn_mark->lock);
 563         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 564                 oldmask = fsn_mark->mask;
 565                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 566         } else {
 567                 __u32 tmask = fsn_mark->ignored_mask | mask;
 568                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 569                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 570                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 571         }
 572
 573         if (!(flags & FAN_MARK_ONDIR)) {
 574                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 575                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 576         }
 577
 578         spin_unlock(&fsn_mark->lock);
 579
 580         return mask & ~oldmask;
 581 }
 582
 583 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 584                                       struct vfsmount *mnt, __u32 mask,
 585                                       unsigned int flags)
 586 {
 587         struct fsnotify_mark *fsn_mark;
 588         __u32 added;
 589
 590         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 591         if (!fsn_mark) {
 592                 int ret;
 593
 594                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 595                         return -ENOSPC;
 596
 597                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 598                 if (!fsn_mark)
 599                         return -ENOMEM;
 600
 601                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 602                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 603                 if (ret) {
 604                         fanotify_free_mark(fsn_mark);
 605                         return ret;
 606                 }
 607         }
 608         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 609         fsnotify_put_mark(fsn_mark);
 610         if (added & ~mnt->mnt_fsnotify_mask)
 611                 fsnotify_recalc_vfsmount_mask(mnt);
 612
 613         return 0;
 614 }
 615
 616 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 617                                    struct inode *inode, __u32 mask,
 618                                    unsigned int flags)
 619 {
 620         struct fsnotify_mark *fsn_mark;
 621         __u32 added;
 622
 623         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 624
 625         /*
 626          * If some other task has this inode open for write we should not add
 627          * an ignored mark, unless that ignored mark is supposed to survive
 628          * modification changes anyway.
 629          */
 630         if ((flags & FAN_MARK_IGNORED_MASK) &&
 631             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 632             (atomic_read(&inode->i_writecount) > 0))
 633                 return 0;
 634
 635         fsn_mark = fsnotify_find_inode_mark(group, inode);
 636         if (!fsn_mark) {
 637                 int ret;
 638
 639                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 640                         return -ENOSPC;
 641
 642                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 643                 if (!fsn_mark)
 644                         return -ENOMEM;
 645
 646                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 647                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 648                 if (ret) {
 649                         fanotify_free_mark(fsn_mark);
 650                         return ret;
 651                 }
 652         }
 653         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 654         fsnotify_put_mark(fsn_mark);
 655         if (added & ~inode->i_fsnotify_mask)
 656                 fsnotify_recalc_inode_mask(inode);
 657         return 0;
 658 }
 659
 660 /* fanotify syscalls */
 661 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 662 {
 663         struct fsnotify_group *group;
 664         int f_flags, fd;
 665         struct user_struct *user;
 666
 667         pr_debug("%s: flags=%d event_f_flags=%d\n",
 668                 __func__, flags, event_f_flags);
 669
 670         if (!capable(CAP_SYS_ADMIN))
 671                 return -EPERM;
 672
 673         if (flags & ~FAN_ALL_INIT_FLAGS)
 674                 return -EINVAL;
 675
 676         user = get_current_user();
 677         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 678                 free_uid(user);
 679                 return -EMFILE;
 680         }
 681
 682         f_flags = O_RDWR | FMODE_NONOTIFY;
 683         if (flags & FAN_CLOEXEC)
 684                 f_flags |= O_CLOEXEC;
 685         if (flags & FAN_NONBLOCK)
 686                 f_flags |= O_NONBLOCK;
 687
 688         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 689         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 690         if (IS_ERR(group))
 691                 return PTR_ERR(group);
 692
 693         group->fanotify_data.user = user;
 694         atomic_inc(&user->fanotify_listeners);
 695
 696         group->fanotify_data.f_flags = event_f_flags;
 697 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 698         mutex_init(&group->fanotify_data.access_mutex);
 699         init_waitqueue_head(&group->fanotify_data.access_waitq);
 700         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 701 #endif
 702         switch (flags & FAN_ALL_CLASS_BITS) {
 703         case FAN_CLASS_NOTIF:
 704                 group->priority = FS_PRIO_0;
 705                 break;
 706         case FAN_CLASS_CONTENT:
 707                 group->priority = FS_PRIO_1;
 708                 break;
 709         case FAN_CLASS_PRE_CONTENT:
 710                 group->priority = FS_PRIO_2;
 711                 break;
 712         default:
 713                 fd = -EINVAL;
 714                 goto out_put_group;
 715         }
 716
 717         if (flags & FAN_UNLIMITED_QUEUE) {
 718                 fd = -EPERM;
 719                 if (!capable(CAP_SYS_ADMIN))
 720                         goto out_put_group;
 721                 group->max_events = UINT_MAX;
 722         } else {
 723                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 724         }
 725
 726         if (flags & FAN_UNLIMITED_MARKS) {
 727                 fd = -EPERM;
 728                 if (!capable(CAP_SYS_ADMIN))
 729                         goto out_put_group;
 730                 group->fanotify_data.max_marks = UINT_MAX;
 731         } else {
 732                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 733         }
 734
 735         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 736         if (fd < 0)
 737                 goto out_put_group;
 738
 739         return fd;
 740
 741 out_put_group:
 742         fsnotify_put_group(group);
 743         return fd;
 744 }
 745
 746 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 747                               __u64 mask, int dfd,
 748                               const char  __user * pathname)
 749 {
 750         struct inode *inode = NULL;
 751         struct vfsmount *mnt = NULL;
 752         struct fsnotify_group *group;
 753         struct file *filp;
 754         struct path path;
 755         int ret, fput_needed;
 756
 757         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 758                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 759
 760         /* we only use the lower 32 bits as of right now. */
 761         if (mask & ((__u64)0xffffffff << 32))
 762                 return -EINVAL;
 763
 764         if (flags & ~FAN_ALL_MARK_FLAGS)
 765                 return -EINVAL;
 766         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 767         case FAN_MARK_ADD:
 768         case FAN_MARK_REMOVE:
 769         case FAN_MARK_FLUSH:
 770                 break;
 771         default:
 772                 return -EINVAL;
 773         }
 774
 775         if (mask & FAN_ONDIR) {
 776                 flags |= FAN_MARK_ONDIR;
 777                 mask &= ~FAN_ONDIR;
 778         }
 779
 780 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 781         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 782 #else
 783         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 784 #endif
 785                 return -EINVAL;
 786
 787         filp = fget_light(fanotify_fd, &fput_needed);
 788         if (unlikely(!filp))
 789                 return -EBADF;
 790
 791         /* verify that this is indeed an fanotify instance */
 792         ret = -EINVAL;
 793         if (unlikely(filp->f_op != &fanotify_fops))
 794                 goto fput_and_out;
 795         group = filp->private_data;
 796
 797         /*
 798          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 799          * allowed to set permissions events.
 800          */
 801         ret = -EINVAL;
 802         if (mask & FAN_ALL_PERM_EVENTS &&
 803             group->priority == FS_PRIO_0)
 804                 goto fput_and_out;
 805
 806         ret = fanotify_find_path(dfd, pathname, &path, flags);
 807         if (ret)
 808                 goto fput_and_out;
 809
 810         /* inode held in place by reference to path; group by fget on fd */
 811         if (!(flags & FAN_MARK_MOUNT))
 812                 inode = path.dentry->d_inode;
 813         else
 814                 mnt = path.mnt;
 815
 816         /* create/update an inode mark */
 817         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 818         case FAN_MARK_ADD:
 819                 if (flags & FAN_MARK_MOUNT)
 820                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 821                 else
 822                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 823                 break;
 824         case FAN_MARK_REMOVE:
 825                 if (flags & FAN_MARK_MOUNT)
 826                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 827                 else
 828                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 829                 break;
 830         case FAN_MARK_FLUSH:
 831                 if (flags & FAN_MARK_MOUNT)
 832                         fsnotify_clear_vfsmount_marks_by_group(group);
 833                 else
 834                         fsnotify_clear_inode_marks_by_group(group);
 835                 break;
 836         default:
 837                 ret = -EINVAL;
 838         }
 839
 840         path_put(&path);
 841 fput_and_out:
 842         fput_light(filp, fput_needed);
 843         return ret;
 844 }
 845
 846 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 847 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 848                                   long dfd, long pathname)
 849 {
 850         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 851                                   mask, (int) dfd,
 852                                   (const char  __user *) pathname);
 853 }
 854 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 855 #endif
 856
 857 /*
 858  * fanotify_user_setup - Our initialization function.  Note that we cannnot return
 859  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 860  * must result in panic().
 861  */
 862 static int __init fanotify_user_setup(void)
 863 {
 864         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 865         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 866                                                    SLAB_PANIC);
 867
 868         return 0;
 869 }
 870 device_initcall(fanotify_user_setup);