fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16
  17 #include <asm/ioctls.h>
  18
  19 #include "../../mount.h"
  20
  21 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  22 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  23 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  24
  25 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  26
  27 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  28 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  29
  30 struct fanotify_response_event {
  31         struct list_head list;
  32         __s32 fd;
  33         struct fsnotify_event *event;
  34 };
  35
  36 /*
  37  * Get an fsnotify notification event if one exists and is small
  38  * enough to fit in "count". Return an error pointer if the count
  39  * is not large enough.
  40  *
  41  * Called with the group->notification_mutex held.
  42  */
  43 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  44                                             size_t count)
  45 {
  46         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  47
  48         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  49
  50         if (fsnotify_notify_queue_is_empty(group))
  51                 return NULL;
  52
  53         if (FAN_EVENT_METADATA_LEN > count)
  54                 return ERR_PTR(-EINVAL);
  55
  56         /* held the notification_mutex the whole time, so this is the
  57          * same event we peeked above */
  58         return fsnotify_remove_notify_event(group);
  59 }
  60
  61 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
  62 {
  63         int client_fd;
  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         /* 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                 fd_install(client_fd, 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 *event)
 110 {
 111         int ret = 0;
 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->metadata_len = FAN_EVENT_METADATA_LEN;
 118         metadata->vers = FANOTIFY_METADATA_VERSION;
 119         metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
 120         metadata->pid = pid_vnr(event->tgid);
 121         if (unlikely(event->mask & FAN_Q_OVERFLOW))
 122                 metadata->fd = FAN_NOFD;
 123         else {
 124                 metadata->fd = create_fd(group, event);
 125                 if (metadata->fd < 0)
 126                         ret = metadata->fd;
 127         }
 128
 129         return ret;
 130 }
 131
 132 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 133 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 134                                                   __s32 fd)
 135 {
 136         struct fanotify_response_event *re, *return_re = NULL;
 137
 138         mutex_lock(&group->fanotify_data.access_mutex);
 139         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 140                 if (re->fd != fd)
 141                         continue;
 142
 143                 list_del_init(&re->list);
 144                 return_re = re;
 145                 break;
 146         }
 147         mutex_unlock(&group->fanotify_data.access_mutex);
 148
 149         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 150
 151         return return_re;
 152 }
 153
 154 static int process_access_response(struct fsnotify_group *group,
 155                                    struct fanotify_response *response_struct)
 156 {
 157         struct fanotify_response_event *re;
 158         __s32 fd = response_struct->fd;
 159         __u32 response = response_struct->response;
 160
 161         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 162                  fd, response);
 163         /*
 164          * make sure the response is valid, if invalid we do nothing and either
 165          * userspace can send a valid response or we will clean it up after the
 166          * timeout
 167          */
 168         switch (response) {
 169         case FAN_ALLOW:
 170         case FAN_DENY:
 171                 break;
 172         default:
 173                 return -EINVAL;
 174         }
 175
 176         if (fd < 0)
 177                 return -EINVAL;
 178
 179         re = dequeue_re(group, fd);
 180         if (!re)
 181                 return -ENOENT;
 182
 183         re->event->response = response;
 184
 185         wake_up(&group->fanotify_data.access_waitq);
 186
 187         kmem_cache_free(fanotify_response_event_cache, re);
 188
 189         return 0;
 190 }
 191
 192 static int prepare_for_access_response(struct fsnotify_group *group,
 193                                        struct fsnotify_event *event,
 194                                        __s32 fd)
 195 {
 196         struct fanotify_response_event *re;
 197
 198         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 199                 return 0;
 200
 201         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 202         if (!re)
 203                 return -ENOMEM;
 204
 205         re->event = event;
 206         re->fd = fd;
 207
 208         mutex_lock(&group->fanotify_data.access_mutex);
 209
 210         if (atomic_read(&group->fanotify_data.bypass_perm)) {
 211                 mutex_unlock(&group->fanotify_data.access_mutex);
 212                 kmem_cache_free(fanotify_response_event_cache, re);
 213                 event->response = FAN_ALLOW;
 214                 return 0;
 215         }
 216
 217         list_add_tail(&re->list, &group->fanotify_data.access_list);
 218         mutex_unlock(&group->fanotify_data.access_mutex);
 219
 220         return 0;
 221 }
 222
 223 static void remove_access_response(struct fsnotify_group *group,
 224                                    struct fsnotify_event *event,
 225                                    __s32 fd)
 226 {
 227         struct fanotify_response_event *re;
 228
 229         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 230                 return;
 231
 232         re = dequeue_re(group, fd);
 233         if (!re)
 234                 return;
 235
 236         BUG_ON(re->event != event);
 237
 238         kmem_cache_free(fanotify_response_event_cache, re);
 239
 240         return;
 241 }
 242 #else
 243 static int prepare_for_access_response(struct fsnotify_group *group,
 244                                        struct fsnotify_event *event,
 245                                        __s32 fd)
 246 {
 247         return 0;
 248 }
 249
 250 static void remove_access_response(struct fsnotify_group *group,
 251                                    struct fsnotify_event *event,
 252                                    __s32 fd)
 253 {
 254         return;
 255 }
 256 #endif
 257
 258 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 259                                   struct fsnotify_event *event,
 260                                   char __user *buf)
 261 {
 262         struct fanotify_event_metadata fanotify_event_metadata;
 263         int fd, ret;
 264
 265         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 266
 267         ret = fill_event_metadata(group, &fanotify_event_metadata, event);
 268         if (ret < 0)
 269                 goto out;
 270
 271         fd = fanotify_event_metadata.fd;
 272         ret = prepare_for_access_response(group, event, fd);
 273         if (ret)
 274                 goto out_close_fd;
 275
 276         ret = -EFAULT;
 277         if (copy_to_user(buf, &fanotify_event_metadata,
 278                          fanotify_event_metadata.event_len))
 279                 goto out_kill_access_response;
 280
 281         return fanotify_event_metadata.event_len;
 282
 283 out_kill_access_response:
 284         remove_access_response(group, event, fd);
 285 out_close_fd:
 286         if (fd != FAN_NOFD)
 287                 sys_close(fd);
 288 out:
 289 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 290         if (event->mask & FAN_ALL_PERM_EVENTS) {
 291                 event->response = FAN_DENY;
 292                 wake_up(&group->fanotify_data.access_waitq);
 293         }
 294 #endif
 295         return ret;
 296 }
 297
 298 /* intofiy userspace file descriptor functions */
 299 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 300 {
 301         struct fsnotify_group *group = file->private_data;
 302         int ret = 0;
 303
 304         poll_wait(file, &group->notification_waitq, wait);
 305         mutex_lock(&group->notification_mutex);
 306         if (!fsnotify_notify_queue_is_empty(group))
 307                 ret = POLLIN | POLLRDNORM;
 308         mutex_unlock(&group->notification_mutex);
 309
 310         return ret;
 311 }
 312
 313 static ssize_t fanotify_read(struct file *file, char __user *buf,
 314                              size_t count, loff_t *pos)
 315 {
 316         struct fsnotify_group *group;
 317         struct fsnotify_event *kevent;
 318         char __user *start;
 319         int ret;
 320         DEFINE_WAIT(wait);
 321
 322         start = buf;
 323         group = file->private_data;
 324
 325         pr_debug("%s: group=%p\n", __func__, group);
 326
 327         while (1) {
 328                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 329
 330                 mutex_lock(&group->notification_mutex);
 331                 kevent = get_one_event(group, count);
 332                 mutex_unlock(&group->notification_mutex);
 333
 334                 if (kevent) {
 335                         ret = PTR_ERR(kevent);
 336                         if (IS_ERR(kevent))
 337                                 break;
 338                         ret = copy_event_to_user(group, kevent, buf);
 339                         fsnotify_put_event(kevent);
 340                         if (ret < 0)
 341                                 break;
 342                         buf += ret;
 343                         count -= ret;
 344                         continue;
 345                 }
 346
 347                 ret = -EAGAIN;
 348                 if (file->f_flags & O_NONBLOCK)
 349                         break;
 350                 ret = -ERESTARTSYS;
 351                 if (signal_pending(current))
 352                         break;
 353
 354                 if (start != buf)
 355                         break;
 356
 357                 schedule();
 358         }
 359
 360         finish_wait(&group->notification_waitq, &wait);
 361         if (start != buf && ret != -EFAULT)
 362                 ret = buf - start;
 363         return ret;
 364 }
 365
 366 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 367 {
 368 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 369         struct fanotify_response response = { .fd = -1, .response = -1 };
 370         struct fsnotify_group *group;
 371         int ret;
 372
 373         group = file->private_data;
 374
 375         if (count > sizeof(response))
 376                 count = sizeof(response);
 377
 378         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 379
 380         if (copy_from_user(&response, buf, count))
 381                 return -EFAULT;
 382
 383         ret = process_access_response(group, &response);
 384         if (ret < 0)
 385                 count = ret;
 386
 387         return count;
 388 #else
 389         return -EINVAL;
 390 #endif
 391 }
 392
 393 static int fanotify_release(struct inode *ignored, struct file *file)
 394 {
 395         struct fsnotify_group *group = file->private_data;
 396
 397 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 398         struct fanotify_response_event *re, *lre;
 399
 400         mutex_lock(&group->fanotify_data.access_mutex);
 401
 402         atomic_inc(&group->fanotify_data.bypass_perm);
 403
 404         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 405                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 406                          re, re->event);
 407
 408                 list_del_init(&re->list);
 409                 re->event->response = FAN_ALLOW;
 410
 411                 kmem_cache_free(fanotify_response_event_cache, re);
 412         }
 413         mutex_unlock(&group->fanotify_data.access_mutex);
 414
 415         wake_up(&group->fanotify_data.access_waitq);
 416 #endif
 417         /* matches the fanotify_init->fsnotify_alloc_group */
 418         fsnotify_destroy_group(group);
 419
 420         return 0;
 421 }
 422
 423 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 424 {
 425         struct fsnotify_group *group;
 426         struct fsnotify_event_holder *holder;
 427         void __user *p;
 428         int ret = -ENOTTY;
 429         size_t send_len = 0;
 430
 431         group = file->private_data;
 432
 433         p = (void __user *) arg;
 434
 435         switch (cmd) {
 436         case FIONREAD:
 437                 mutex_lock(&group->notification_mutex);
 438                 list_for_each_entry(holder, &group->notification_list, event_list)
 439                         send_len += FAN_EVENT_METADATA_LEN;
 440                 mutex_unlock(&group->notification_mutex);
 441                 ret = put_user(send_len, (int __user *) p);
 442                 break;
 443         }
 444
 445         return ret;
 446 }
 447
 448 static const struct file_operations fanotify_fops = {
 449         .poll           = fanotify_poll,
 450         .read           = fanotify_read,
 451         .write          = fanotify_write,
 452         .fasync         = NULL,
 453         .release        = fanotify_release,
 454         .unlocked_ioctl = fanotify_ioctl,
 455         .compat_ioctl   = fanotify_ioctl,
 456         .llseek         = noop_llseek,
 457 };
 458
 459 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 460 {
 461         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 462 }
 463
 464 static int fanotify_find_path(int dfd, const char __user *filename,
 465                               struct path *path, unsigned int flags)
 466 {
 467         int ret;
 468
 469         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 470                  dfd, filename, flags);
 471
 472         if (filename == NULL) {
 473                 struct file *file;
 474                 int fput_needed;
 475
 476                 ret = -EBADF;
 477                 file = fget_light(dfd, &fput_needed);
 478                 if (!file)
 479                         goto out;
 480
 481                 ret = -ENOTDIR;
 482                 if ((flags & FAN_MARK_ONLYDIR) &&
 483                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 484                         fput_light(file, fput_needed);
 485                         goto out;
 486                 }
 487
 488                 *path = file->f_path;
 489                 path_get(path);
 490                 fput_light(file, fput_needed);
 491         } else {
 492                 unsigned int lookup_flags = 0;
 493
 494                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 495                         lookup_flags |= LOOKUP_FOLLOW;
 496                 if (flags & FAN_MARK_ONLYDIR)
 497                         lookup_flags |= LOOKUP_DIRECTORY;
 498
 499                 ret = user_path_at(dfd, filename, lookup_flags, path);
 500                 if (ret)
 501                         goto out;
 502         }
 503
 504         /* you can only watch an inode if you have read permissions on it */
 505         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 506         if (ret)
 507                 path_put(path);
 508 out:
 509         return ret;
 510 }
 511
 512 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 513                                             __u32 mask,
 514                                             unsigned int flags,
 515                                             int *destroy)
 516 {
 517         __u32 oldmask;
 518
 519         spin_lock(&fsn_mark->lock);
 520         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 521                 oldmask = fsn_mark->mask;
 522                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 523         } else {
 524                 oldmask = fsn_mark->ignored_mask;
 525                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 526         }
 527         spin_unlock(&fsn_mark->lock);
 528
 529         *destroy = !(oldmask & ~mask);
 530
 531         return mask & oldmask;
 532 }
 533
 534 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 535                                          struct vfsmount *mnt, __u32 mask,
 536                                          unsigned int flags)
 537 {
 538         struct fsnotify_mark *fsn_mark = NULL;
 539         __u32 removed;
 540         int destroy_mark;
 541
 542         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 543         if (!fsn_mark)
 544                 return -ENOENT;
 545
 546         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 547                                                  &destroy_mark);
 548         if (destroy_mark)
 549                 fsnotify_destroy_mark(fsn_mark, group);
 550
 551         fsnotify_put_mark(fsn_mark);
 552         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 553                 fsnotify_recalc_vfsmount_mask(mnt);
 554
 555         return 0;
 556 }
 557
 558 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 559                                       struct inode *inode, __u32 mask,
 560                                       unsigned int flags)
 561 {
 562         struct fsnotify_mark *fsn_mark = NULL;
 563         __u32 removed;
 564         int destroy_mark;
 565
 566         fsn_mark = fsnotify_find_inode_mark(group, inode);
 567         if (!fsn_mark)
 568                 return -ENOENT;
 569
 570         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 571                                                  &destroy_mark);
 572         if (destroy_mark)
 573                 fsnotify_destroy_mark(fsn_mark, group);
 574         /* matches the fsnotify_find_inode_mark() */
 575         fsnotify_put_mark(fsn_mark);
 576         if (removed & inode->i_fsnotify_mask)
 577                 fsnotify_recalc_inode_mask(inode);
 578
 579         return 0;
 580 }
 581
 582 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 583                                        __u32 mask,
 584                                        unsigned int flags)
 585 {
 586         __u32 oldmask = -1;
 587
 588         spin_lock(&fsn_mark->lock);
 589         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 590                 oldmask = fsn_mark->mask;
 591                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 592         } else {
 593                 __u32 tmask = fsn_mark->ignored_mask | mask;
 594                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 595                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 596                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 597         }
 598
 599         if (!(flags & FAN_MARK_ONDIR)) {
 600                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 601                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 602         }
 603
 604         spin_unlock(&fsn_mark->lock);
 605
 606         return mask & ~oldmask;
 607 }
 608
 609 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 610                                       struct vfsmount *mnt, __u32 mask,
 611                                       unsigned int flags)
 612 {
 613         struct fsnotify_mark *fsn_mark;
 614         __u32 added;
 615         int ret = 0;
 616
 617         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 618         if (!fsn_mark) {
 619                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 620                         return -ENOSPC;
 621
 622                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 623                 if (!fsn_mark)
 624                         return -ENOMEM;
 625
 626                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 627                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 628                 if (ret)
 629                         goto err;
 630         }
 631         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 632
 633         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 634                 fsnotify_recalc_vfsmount_mask(mnt);
 635 err:
 636         fsnotify_put_mark(fsn_mark);
 637         return ret;
 638 }
 639
 640 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 641                                    struct inode *inode, __u32 mask,
 642                                    unsigned int flags)
 643 {
 644         struct fsnotify_mark *fsn_mark;
 645         __u32 added;
 646         int ret = 0;
 647
 648         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 649
 650         /*
 651          * If some other task has this inode open for write we should not add
 652          * an ignored mark, unless that ignored mark is supposed to survive
 653          * modification changes anyway.
 654          */
 655         if ((flags & FAN_MARK_IGNORED_MASK) &&
 656             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 657             (atomic_read(&inode->i_writecount) > 0))
 658                 return 0;
 659
 660         fsn_mark = fsnotify_find_inode_mark(group, inode);
 661         if (!fsn_mark) {
 662                 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 663                         return -ENOSPC;
 664
 665                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 666                 if (!fsn_mark)
 667                         return -ENOMEM;
 668
 669                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 670                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 671                 if (ret)
 672                         goto err;
 673         }
 674         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 675
 676         if (added & ~inode->i_fsnotify_mask)
 677                 fsnotify_recalc_inode_mask(inode);
 678 err:
 679         fsnotify_put_mark(fsn_mark);
 680         return ret;
 681 }
 682
 683 /* fanotify syscalls */
 684 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 685 {
 686         struct fsnotify_group *group;
 687         int f_flags, fd;
 688         struct user_struct *user;
 689
 690         pr_debug("%s: flags=%d event_f_flags=%d\n",
 691                 __func__, flags, event_f_flags);
 692
 693         if (!capable(CAP_SYS_ADMIN))
 694                 return -EPERM;
 695
 696         if (flags & ~FAN_ALL_INIT_FLAGS)
 697                 return -EINVAL;
 698
 699         user = get_current_user();
 700         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 701                 free_uid(user);
 702                 return -EMFILE;
 703         }
 704
 705         f_flags = O_RDWR | FMODE_NONOTIFY;
 706         if (flags & FAN_CLOEXEC)
 707                 f_flags |= O_CLOEXEC;
 708         if (flags & FAN_NONBLOCK)
 709                 f_flags |= O_NONBLOCK;
 710
 711         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 712         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 713         if (IS_ERR(group)) {
 714                 free_uid(user);
 715                 return PTR_ERR(group);
 716         }
 717
 718         group->fanotify_data.user = user;
 719         atomic_inc(&user->fanotify_listeners);
 720
 721         group->fanotify_data.f_flags = event_f_flags;
 722 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 723         mutex_init(&group->fanotify_data.access_mutex);
 724         init_waitqueue_head(&group->fanotify_data.access_waitq);
 725         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 726         atomic_set(&group->fanotify_data.bypass_perm, 0);
 727 #endif
 728         switch (flags & FAN_ALL_CLASS_BITS) {
 729         case FAN_CLASS_NOTIF:
 730                 group->priority = FS_PRIO_0;
 731                 break;
 732         case FAN_CLASS_CONTENT:
 733                 group->priority = FS_PRIO_1;
 734                 break;
 735         case FAN_CLASS_PRE_CONTENT:
 736                 group->priority = FS_PRIO_2;
 737                 break;
 738         default:
 739                 fd = -EINVAL;
 740                 goto out_destroy_group;
 741         }
 742
 743         if (flags & FAN_UNLIMITED_QUEUE) {
 744                 fd = -EPERM;
 745                 if (!capable(CAP_SYS_ADMIN))
 746                         goto out_destroy_group;
 747                 group->max_events = UINT_MAX;
 748         } else {
 749                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 750         }
 751
 752         if (flags & FAN_UNLIMITED_MARKS) {
 753                 fd = -EPERM;
 754                 if (!capable(CAP_SYS_ADMIN))
 755                         goto out_destroy_group;
 756                 group->fanotify_data.max_marks = UINT_MAX;
 757         } else {
 758                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 759         }
 760
 761         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 762         if (fd < 0)
 763                 goto out_destroy_group;
 764
 765         return fd;
 766
 767 out_destroy_group:
 768         fsnotify_destroy_group(group);
 769         return fd;
 770 }
 771
 772 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 773                               __u64 mask, int dfd,
 774                               const char  __user * pathname)
 775 {
 776         struct inode *inode = NULL;
 777         struct vfsmount *mnt = NULL;
 778         struct fsnotify_group *group;
 779         struct file *filp;
 780         struct path path;
 781         int ret, fput_needed;
 782
 783         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 784                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 785
 786         /* we only use the lower 32 bits as of right now. */
 787         if (mask & ((__u64)0xffffffff << 32))
 788                 return -EINVAL;
 789
 790         if (flags & ~FAN_ALL_MARK_FLAGS)
 791                 return -EINVAL;
 792         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 793         case FAN_MARK_ADD:              /* fallthrough */
 794         case FAN_MARK_REMOVE:
 795                 if (!mask)
 796                         return -EINVAL;
 797         case FAN_MARK_FLUSH:
 798                 break;
 799         default:
 800                 return -EINVAL;
 801         }
 802
 803         if (mask & FAN_ONDIR) {
 804                 flags |= FAN_MARK_ONDIR;
 805                 mask &= ~FAN_ONDIR;
 806         }
 807
 808 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 809         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 810 #else
 811         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 812 #endif
 813                 return -EINVAL;
 814
 815         filp = fget_light(fanotify_fd, &fput_needed);
 816         if (unlikely(!filp))
 817                 return -EBADF;
 818
 819         /* verify that this is indeed an fanotify instance */
 820         ret = -EINVAL;
 821         if (unlikely(filp->f_op != &fanotify_fops))
 822                 goto fput_and_out;
 823         group = filp->private_data;
 824
 825         /*
 826          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 827          * allowed to set permissions events.
 828          */
 829         ret = -EINVAL;
 830         if (mask & FAN_ALL_PERM_EVENTS &&
 831             group->priority == FS_PRIO_0)
 832                 goto fput_and_out;
 833
 834         ret = fanotify_find_path(dfd, pathname, &path, flags);
 835         if (ret)
 836                 goto fput_and_out;
 837
 838         /* inode held in place by reference to path; group by fget on fd */
 839         if (!(flags & FAN_MARK_MOUNT))
 840                 inode = path.dentry->d_inode;
 841         else
 842                 mnt = path.mnt;
 843
 844         /* create/update an inode mark */
 845         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 846         case FAN_MARK_ADD:
 847                 if (flags & FAN_MARK_MOUNT)
 848                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 849                 else
 850                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 851                 break;
 852         case FAN_MARK_REMOVE:
 853                 if (flags & FAN_MARK_MOUNT)
 854                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 855                 else
 856                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 857                 break;
 858         case FAN_MARK_FLUSH:
 859                 if (flags & FAN_MARK_MOUNT)
 860                         fsnotify_clear_vfsmount_marks_by_group(group);
 861                 else
 862                         fsnotify_clear_inode_marks_by_group(group);
 863                 break;
 864         default:
 865                 ret = -EINVAL;
 866         }
 867
 868         path_put(&path);
 869 fput_and_out:
 870         fput_light(filp, fput_needed);
 871         return ret;
 872 }
 873
 874 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 875 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 876                                   long dfd, long pathname)
 877 {
 878         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 879                                   mask, (int) dfd,
 880                                   (const char  __user *) pathname);
 881 }
 882 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 883 #endif
 884
 885 /*
 886  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 887  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 888  * must result in panic().
 889  */
 890 static int __init fanotify_user_setup(void)
 891 {
 892         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 893         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 894                                                    SLAB_PANIC);
 895
 896         return 0;
 897 }
 898 device_initcall(fanotify_user_setup);