1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/fanotify.h> 3 #include <linux/fcntl.h> 4 #include <linux/file.h> 5 #include <linux/fs.h> 6 #include <linux/anon_inodes.h> 7 #include <linux/fsnotify_backend.h> 8 #include <linux/init.h> 9 #include <linux/mount.h> 10 #include <linux/namei.h> 11 #include <linux/poll.h> 12 #include <linux/security.h> 13 #include <linux/syscalls.h> 14 #include <linux/slab.h> 15 #include <linux/types.h> 16 #include <linux/uaccess.h> 17 #include <linux/compat.h> 18 #include <linux/sched/signal.h> 19 #include <linux/memcontrol.h> 20 21 #include <asm/ioctls.h> 22 23 #include "../../mount.h" 24 #include "../fdinfo.h" 25 #include "fanotify.h" 26 27 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384 28 #define FANOTIFY_DEFAULT_MAX_MARKS 8192 29 #define FANOTIFY_DEFAULT_MAX_LISTENERS 128 30 31 /* 32 * All flags that may be specified in parameter event_f_flags of fanotify_init. 33 * 34 * Internal and external open flags are stored together in field f_flags of 35 * struct file. Only external open flags shall be allowed in event_f_flags. 36 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be 37 * excluded. 38 */ 39 #define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \ 40 O_ACCMODE | O_APPEND | O_NONBLOCK | \ 41 __O_SYNC | O_DSYNC | O_CLOEXEC | \ 42 O_LARGEFILE | O_NOATIME ) 43 44 extern const struct fsnotify_ops fanotify_fsnotify_ops; 45 46 struct kmem_cache *fanotify_mark_cache __read_mostly; 47 struct kmem_cache *fanotify_event_cachep __read_mostly; 48 struct kmem_cache *fanotify_perm_event_cachep __read_mostly; 49 50 /* 51 * Get an fsnotify notification event if one exists and is small 52 * enough to fit in "count". Return an error pointer if the count 53 * is not large enough. 54 * 55 * Called with the group->notification_lock held. 56 */ 57 static struct fsnotify_event *get_one_event(struct fsnotify_group *group, 58 size_t count) 59 { 60 assert_spin_locked(&group->notification_lock); 61 62 pr_debug("%s: group=%p count=%zd\n", __func__, group, count); 63 64 if (fsnotify_notify_queue_is_empty(group)) 65 return NULL; 66 67 if (FAN_EVENT_METADATA_LEN > count) 68 return ERR_PTR(-EINVAL); 69 70 /* held the notification_lock the whole time, so this is the 71 * same event we peeked above */ 72 return fsnotify_remove_first_event(group); 73 } 74 75 static int create_fd(struct fsnotify_group *group, 76 struct fanotify_event_info *event, 77 struct file **file) 78 { 79 int client_fd; 80 struct file *new_file; 81 82 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 83 84 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags); 85 if (client_fd < 0) 86 return client_fd; 87 88 /* 89 * we need a new file handle for the userspace program so it can read even if it was 90 * originally opened O_WRONLY. 91 */ 92 /* it's possible this event was an overflow event. in that case dentry and mnt 93 * are NULL; That's fine, just don't call dentry open */ 94 if (event->path.dentry && event->path.mnt) 95 new_file = dentry_open(&event->path, 96 group->fanotify_data.f_flags | FMODE_NONOTIFY, 97 current_cred()); 98 else 99 new_file = ERR_PTR(-EOVERFLOW); 100 if (IS_ERR(new_file)) { 101 /* 102 * we still send an event even if we can't open the file. this 103 * can happen when say tasks are gone and we try to open their 104 * /proc files or we try to open a WRONLY file like in sysfs 105 * we just send the errno to userspace since there isn't much 106 * else we can do. 107 */ 108 put_unused_fd(client_fd); 109 client_fd = PTR_ERR(new_file); 110 } else { 111 *file = new_file; 112 } 113 114 return client_fd; 115 } 116 117 static int fill_event_metadata(struct fsnotify_group *group, 118 struct fanotify_event_metadata *metadata, 119 struct fsnotify_event *fsn_event, 120 struct file **file) 121 { 122 int ret = 0; 123 struct fanotify_event_info *event; 124 125 pr_debug("%s: group=%p metadata=%p event=%p\n", __func__, 126 group, metadata, fsn_event); 127 128 *file = NULL; 129 event = container_of(fsn_event, struct fanotify_event_info, fse); 130 metadata->event_len = FAN_EVENT_METADATA_LEN; 131 metadata->metadata_len = FAN_EVENT_METADATA_LEN; 132 metadata->vers = FANOTIFY_METADATA_VERSION; 133 metadata->reserved = 0; 134 metadata->mask = fsn_event->mask & FANOTIFY_OUTGOING_EVENTS; 135 metadata->pid = pid_vnr(event->pid); 136 if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW)) 137 metadata->fd = FAN_NOFD; 138 else { 139 metadata->fd = create_fd(group, event, file); 140 if (metadata->fd < 0) 141 ret = metadata->fd; 142 } 143 144 return ret; 145 } 146 147 static struct fanotify_perm_event_info *dequeue_event( 148 struct fsnotify_group *group, int fd) 149 { 150 struct fanotify_perm_event_info *event, *return_e = NULL; 151 152 spin_lock(&group->notification_lock); 153 list_for_each_entry(event, &group->fanotify_data.access_list, 154 fae.fse.list) { 155 if (event->fd != fd) 156 continue; 157 158 list_del_init(&event->fae.fse.list); 159 return_e = event; 160 break; 161 } 162 spin_unlock(&group->notification_lock); 163 164 pr_debug("%s: found return_re=%p\n", __func__, return_e); 165 166 return return_e; 167 } 168 169 static int process_access_response(struct fsnotify_group *group, 170 struct fanotify_response *response_struct) 171 { 172 struct fanotify_perm_event_info *event; 173 int fd = response_struct->fd; 174 int response = response_struct->response; 175 176 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group, 177 fd, response); 178 /* 179 * make sure the response is valid, if invalid we do nothing and either 180 * userspace can send a valid response or we will clean it up after the 181 * timeout 182 */ 183 switch (response & ~FAN_AUDIT) { 184 case FAN_ALLOW: 185 case FAN_DENY: 186 break; 187 default: 188 return -EINVAL; 189 } 190 191 if (fd < 0) 192 return -EINVAL; 193 194 if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT)) 195 return -EINVAL; 196 197 event = dequeue_event(group, fd); 198 if (!event) 199 return -ENOENT; 200 201 event->response = response; 202 wake_up(&group->fanotify_data.access_waitq); 203 204 return 0; 205 } 206 207 static ssize_t copy_event_to_user(struct fsnotify_group *group, 208 struct fsnotify_event *event, 209 char __user *buf, size_t count) 210 { 211 struct fanotify_event_metadata fanotify_event_metadata; 212 struct file *f; 213 int fd, ret; 214 215 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 216 217 ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f); 218 if (ret < 0) 219 return ret; 220 221 fd = fanotify_event_metadata.fd; 222 ret = -EFAULT; 223 /* 224 * Sanity check copy size in case get_one_event() and 225 * fill_event_metadata() event_len sizes ever get out of sync. 226 */ 227 if (WARN_ON_ONCE(fanotify_event_metadata.event_len > count)) 228 goto out_close_fd; 229 if (copy_to_user(buf, &fanotify_event_metadata, 230 fanotify_event_metadata.event_len)) 231 goto out_close_fd; 232 233 if (fanotify_is_perm_event(event->mask)) 234 FANOTIFY_PE(event)->fd = fd; 235 236 if (fd != FAN_NOFD) 237 fd_install(fd, f); 238 return fanotify_event_metadata.event_len; 239 240 out_close_fd: 241 if (fd != FAN_NOFD) { 242 put_unused_fd(fd); 243 fput(f); 244 } 245 return ret; 246 } 247 248 /* intofiy userspace file descriptor functions */ 249 static __poll_t fanotify_poll(struct file *file, poll_table *wait) 250 { 251 struct fsnotify_group *group = file->private_data; 252 __poll_t ret = 0; 253 254 poll_wait(file, &group->notification_waitq, wait); 255 spin_lock(&group->notification_lock); 256 if (!fsnotify_notify_queue_is_empty(group)) 257 ret = EPOLLIN | EPOLLRDNORM; 258 spin_unlock(&group->notification_lock); 259 260 return ret; 261 } 262 263 static ssize_t fanotify_read(struct file *file, char __user *buf, 264 size_t count, loff_t *pos) 265 { 266 struct fsnotify_group *group; 267 struct fsnotify_event *kevent; 268 char __user *start; 269 int ret; 270 DEFINE_WAIT_FUNC(wait, woken_wake_function); 271 272 start = buf; 273 group = file->private_data; 274 275 pr_debug("%s: group=%p\n", __func__, group); 276 277 add_wait_queue(&group->notification_waitq, &wait); 278 while (1) { 279 spin_lock(&group->notification_lock); 280 kevent = get_one_event(group, count); 281 spin_unlock(&group->notification_lock); 282 283 if (IS_ERR(kevent)) { 284 ret = PTR_ERR(kevent); 285 break; 286 } 287 288 if (!kevent) { 289 ret = -EAGAIN; 290 if (file->f_flags & O_NONBLOCK) 291 break; 292 293 ret = -ERESTARTSYS; 294 if (signal_pending(current)) 295 break; 296 297 if (start != buf) 298 break; 299 300 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 301 continue; 302 } 303 304 ret = copy_event_to_user(group, kevent, buf, count); 305 if (unlikely(ret == -EOPENSTALE)) { 306 /* 307 * We cannot report events with stale fd so drop it. 308 * Setting ret to 0 will continue the event loop and 309 * do the right thing if there are no more events to 310 * read (i.e. return bytes read, -EAGAIN or wait). 311 */ 312 ret = 0; 313 } 314 315 /* 316 * Permission events get queued to wait for response. Other 317 * events can be destroyed now. 318 */ 319 if (!fanotify_is_perm_event(kevent->mask)) { 320 fsnotify_destroy_event(group, kevent); 321 } else { 322 if (ret <= 0) { 323 FANOTIFY_PE(kevent)->response = FAN_DENY; 324 wake_up(&group->fanotify_data.access_waitq); 325 } else { 326 spin_lock(&group->notification_lock); 327 list_add_tail(&kevent->list, 328 &group->fanotify_data.access_list); 329 spin_unlock(&group->notification_lock); 330 } 331 } 332 if (ret < 0) 333 break; 334 buf += ret; 335 count -= ret; 336 } 337 remove_wait_queue(&group->notification_waitq, &wait); 338 339 if (start != buf && ret != -EFAULT) 340 ret = buf - start; 341 return ret; 342 } 343 344 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) 345 { 346 struct fanotify_response response = { .fd = -1, .response = -1 }; 347 struct fsnotify_group *group; 348 int ret; 349 350 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 351 return -EINVAL; 352 353 group = file->private_data; 354 355 if (count > sizeof(response)) 356 count = sizeof(response); 357 358 pr_debug("%s: group=%p count=%zu\n", __func__, group, count); 359 360 if (copy_from_user(&response, buf, count)) 361 return -EFAULT; 362 363 ret = process_access_response(group, &response); 364 if (ret < 0) 365 count = ret; 366 367 return count; 368 } 369 370 static int fanotify_release(struct inode *ignored, struct file *file) 371 { 372 struct fsnotify_group *group = file->private_data; 373 struct fanotify_perm_event_info *event, *next; 374 struct fsnotify_event *fsn_event; 375 376 /* 377 * Stop new events from arriving in the notification queue. since 378 * userspace cannot use fanotify fd anymore, no event can enter or 379 * leave access_list by now either. 380 */ 381 fsnotify_group_stop_queueing(group); 382 383 /* 384 * Process all permission events on access_list and notification queue 385 * and simulate reply from userspace. 386 */ 387 spin_lock(&group->notification_lock); 388 list_for_each_entry_safe(event, next, &group->fanotify_data.access_list, 389 fae.fse.list) { 390 pr_debug("%s: found group=%p event=%p\n", __func__, group, 391 event); 392 393 list_del_init(&event->fae.fse.list); 394 event->response = FAN_ALLOW; 395 } 396 397 /* 398 * Destroy all non-permission events. For permission events just 399 * dequeue them and set the response. They will be freed once the 400 * response is consumed and fanotify_get_response() returns. 401 */ 402 while (!fsnotify_notify_queue_is_empty(group)) { 403 fsn_event = fsnotify_remove_first_event(group); 404 if (!(fsn_event->mask & FANOTIFY_PERM_EVENTS)) { 405 spin_unlock(&group->notification_lock); 406 fsnotify_destroy_event(group, fsn_event); 407 spin_lock(&group->notification_lock); 408 } else { 409 FANOTIFY_PE(fsn_event)->response = FAN_ALLOW; 410 } 411 } 412 spin_unlock(&group->notification_lock); 413 414 /* Response for all permission events it set, wakeup waiters */ 415 wake_up(&group->fanotify_data.access_waitq); 416 417 /* 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 *fsn_event; 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 spin_lock(&group->notification_lock); 438 list_for_each_entry(fsn_event, &group->notification_list, list) 439 send_len += FAN_EVENT_METADATA_LEN; 440 spin_unlock(&group->notification_lock); 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 .show_fdinfo = fanotify_show_fdinfo, 450 .poll = fanotify_poll, 451 .read = fanotify_read, 452 .write = fanotify_write, 453 .fasync = NULL, 454 .release = fanotify_release, 455 .unlocked_ioctl = fanotify_ioctl, 456 .compat_ioctl = fanotify_ioctl, 457 .llseek = noop_llseek, 458 }; 459 460 static int fanotify_find_path(int dfd, const char __user *filename, 461 struct path *path, unsigned int flags) 462 { 463 int ret; 464 465 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__, 466 dfd, filename, flags); 467 468 if (filename == NULL) { 469 struct fd f = fdget(dfd); 470 471 ret = -EBADF; 472 if (!f.file) 473 goto out; 474 475 ret = -ENOTDIR; 476 if ((flags & FAN_MARK_ONLYDIR) && 477 !(S_ISDIR(file_inode(f.file)->i_mode))) { 478 fdput(f); 479 goto out; 480 } 481 482 *path = f.file->f_path; 483 path_get(path); 484 fdput(f); 485 } else { 486 unsigned int lookup_flags = 0; 487 488 if (!(flags & FAN_MARK_DONT_FOLLOW)) 489 lookup_flags |= LOOKUP_FOLLOW; 490 if (flags & FAN_MARK_ONLYDIR) 491 lookup_flags |= LOOKUP_DIRECTORY; 492 493 ret = user_path_at(dfd, filename, lookup_flags, path); 494 if (ret) 495 goto out; 496 } 497 498 /* you can only watch an inode if you have read permissions on it */ 499 ret = inode_permission(path->dentry->d_inode, MAY_READ); 500 if (ret) 501 path_put(path); 502 out: 503 return ret; 504 } 505 506 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, 507 __u32 mask, 508 unsigned int flags, 509 int *destroy) 510 { 511 __u32 oldmask = 0; 512 513 spin_lock(&fsn_mark->lock); 514 if (!(flags & FAN_MARK_IGNORED_MASK)) { 515 oldmask = fsn_mark->mask; 516 fsn_mark->mask &= ~mask; 517 } else { 518 fsn_mark->ignored_mask &= ~mask; 519 } 520 *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask); 521 spin_unlock(&fsn_mark->lock); 522 523 return mask & oldmask; 524 } 525 526 static int fanotify_remove_mark(struct fsnotify_group *group, 527 fsnotify_connp_t *connp, __u32 mask, 528 unsigned int flags) 529 { 530 struct fsnotify_mark *fsn_mark = NULL; 531 __u32 removed; 532 int destroy_mark; 533 534 mutex_lock(&group->mark_mutex); 535 fsn_mark = fsnotify_find_mark(connp, group); 536 if (!fsn_mark) { 537 mutex_unlock(&group->mark_mutex); 538 return -ENOENT; 539 } 540 541 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, 542 &destroy_mark); 543 if (removed & fsnotify_conn_mask(fsn_mark->connector)) 544 fsnotify_recalc_mask(fsn_mark->connector); 545 if (destroy_mark) 546 fsnotify_detach_mark(fsn_mark); 547 mutex_unlock(&group->mark_mutex); 548 if (destroy_mark) 549 fsnotify_free_mark(fsn_mark); 550 551 /* matches the fsnotify_find_mark() */ 552 fsnotify_put_mark(fsn_mark); 553 return 0; 554 } 555 556 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group, 557 struct vfsmount *mnt, __u32 mask, 558 unsigned int flags) 559 { 560 return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 561 mask, flags); 562 } 563 564 static int fanotify_remove_sb_mark(struct fsnotify_group *group, 565 struct super_block *sb, __u32 mask, 566 unsigned int flags) 567 { 568 return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask, flags); 569 } 570 571 static int fanotify_remove_inode_mark(struct fsnotify_group *group, 572 struct inode *inode, __u32 mask, 573 unsigned int flags) 574 { 575 return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask, 576 flags); 577 } 578 579 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, 580 __u32 mask, 581 unsigned int flags) 582 { 583 __u32 oldmask = -1; 584 585 spin_lock(&fsn_mark->lock); 586 if (!(flags & FAN_MARK_IGNORED_MASK)) { 587 oldmask = fsn_mark->mask; 588 fsn_mark->mask |= mask; 589 } else { 590 fsn_mark->ignored_mask |= mask; 591 if (flags & FAN_MARK_IGNORED_SURV_MODIFY) 592 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY; 593 } 594 spin_unlock(&fsn_mark->lock); 595 596 return mask & ~oldmask; 597 } 598 599 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group, 600 fsnotify_connp_t *connp, 601 unsigned int type) 602 { 603 struct fsnotify_mark *mark; 604 int ret; 605 606 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) 607 return ERR_PTR(-ENOSPC); 608 609 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); 610 if (!mark) 611 return ERR_PTR(-ENOMEM); 612 613 fsnotify_init_mark(mark, group); 614 ret = fsnotify_add_mark_locked(mark, connp, type, 0); 615 if (ret) { 616 fsnotify_put_mark(mark); 617 return ERR_PTR(ret); 618 } 619 620 return mark; 621 } 622 623 624 static int fanotify_add_mark(struct fsnotify_group *group, 625 fsnotify_connp_t *connp, unsigned int type, 626 __u32 mask, unsigned int flags) 627 { 628 struct fsnotify_mark *fsn_mark; 629 __u32 added; 630 631 mutex_lock(&group->mark_mutex); 632 fsn_mark = fsnotify_find_mark(connp, group); 633 if (!fsn_mark) { 634 fsn_mark = fanotify_add_new_mark(group, connp, type); 635 if (IS_ERR(fsn_mark)) { 636 mutex_unlock(&group->mark_mutex); 637 return PTR_ERR(fsn_mark); 638 } 639 } 640 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); 641 if (added & ~fsnotify_conn_mask(fsn_mark->connector)) 642 fsnotify_recalc_mask(fsn_mark->connector); 643 mutex_unlock(&group->mark_mutex); 644 645 fsnotify_put_mark(fsn_mark); 646 return 0; 647 } 648 649 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group, 650 struct vfsmount *mnt, __u32 mask, 651 unsigned int flags) 652 { 653 return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 654 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags); 655 } 656 657 static int fanotify_add_sb_mark(struct fsnotify_group *group, 658 struct super_block *sb, __u32 mask, 659 unsigned int flags) 660 { 661 return fanotify_add_mark(group, &sb->s_fsnotify_marks, 662 FSNOTIFY_OBJ_TYPE_SB, mask, flags); 663 } 664 665 static int fanotify_add_inode_mark(struct fsnotify_group *group, 666 struct inode *inode, __u32 mask, 667 unsigned int flags) 668 { 669 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode); 670 671 /* 672 * If some other task has this inode open for write we should not add 673 * an ignored mark, unless that ignored mark is supposed to survive 674 * modification changes anyway. 675 */ 676 if ((flags & FAN_MARK_IGNORED_MASK) && 677 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) && 678 inode_is_open_for_write(inode)) 679 return 0; 680 681 return fanotify_add_mark(group, &inode->i_fsnotify_marks, 682 FSNOTIFY_OBJ_TYPE_INODE, mask, flags); 683 } 684 685 /* fanotify syscalls */ 686 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags) 687 { 688 struct fsnotify_group *group; 689 int f_flags, fd; 690 struct user_struct *user; 691 struct fanotify_event_info *oevent; 692 693 pr_debug("%s: flags=%x event_f_flags=%x\n", 694 __func__, flags, event_f_flags); 695 696 if (!capable(CAP_SYS_ADMIN)) 697 return -EPERM; 698 699 #ifdef CONFIG_AUDITSYSCALL 700 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT)) 701 #else 702 if (flags & ~FANOTIFY_INIT_FLAGS) 703 #endif 704 return -EINVAL; 705 706 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS) 707 return -EINVAL; 708 709 switch (event_f_flags & O_ACCMODE) { 710 case O_RDONLY: 711 case O_RDWR: 712 case O_WRONLY: 713 break; 714 default: 715 return -EINVAL; 716 } 717 718 user = get_current_user(); 719 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) { 720 free_uid(user); 721 return -EMFILE; 722 } 723 724 f_flags = O_RDWR | FMODE_NONOTIFY; 725 if (flags & FAN_CLOEXEC) 726 f_flags |= O_CLOEXEC; 727 if (flags & FAN_NONBLOCK) 728 f_flags |= O_NONBLOCK; 729 730 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */ 731 group = fsnotify_alloc_group(&fanotify_fsnotify_ops); 732 if (IS_ERR(group)) { 733 free_uid(user); 734 return PTR_ERR(group); 735 } 736 737 group->fanotify_data.user = user; 738 group->fanotify_data.flags = flags; 739 atomic_inc(&user->fanotify_listeners); 740 group->memcg = get_mem_cgroup_from_mm(current->mm); 741 742 oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL); 743 if (unlikely(!oevent)) { 744 fd = -ENOMEM; 745 goto out_destroy_group; 746 } 747 group->overflow_event = &oevent->fse; 748 749 if (force_o_largefile()) 750 event_f_flags |= O_LARGEFILE; 751 group->fanotify_data.f_flags = event_f_flags; 752 init_waitqueue_head(&group->fanotify_data.access_waitq); 753 INIT_LIST_HEAD(&group->fanotify_data.access_list); 754 switch (flags & FANOTIFY_CLASS_BITS) { 755 case FAN_CLASS_NOTIF: 756 group->priority = FS_PRIO_0; 757 break; 758 case FAN_CLASS_CONTENT: 759 group->priority = FS_PRIO_1; 760 break; 761 case FAN_CLASS_PRE_CONTENT: 762 group->priority = FS_PRIO_2; 763 break; 764 default: 765 fd = -EINVAL; 766 goto out_destroy_group; 767 } 768 769 if (flags & FAN_UNLIMITED_QUEUE) { 770 fd = -EPERM; 771 if (!capable(CAP_SYS_ADMIN)) 772 goto out_destroy_group; 773 group->max_events = UINT_MAX; 774 } else { 775 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS; 776 } 777 778 if (flags & FAN_UNLIMITED_MARKS) { 779 fd = -EPERM; 780 if (!capable(CAP_SYS_ADMIN)) 781 goto out_destroy_group; 782 group->fanotify_data.max_marks = UINT_MAX; 783 } else { 784 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS; 785 } 786 787 if (flags & FAN_ENABLE_AUDIT) { 788 fd = -EPERM; 789 if (!capable(CAP_AUDIT_WRITE)) 790 goto out_destroy_group; 791 } 792 793 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags); 794 if (fd < 0) 795 goto out_destroy_group; 796 797 return fd; 798 799 out_destroy_group: 800 fsnotify_destroy_group(group); 801 return fd; 802 } 803 804 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask, 805 int dfd, const char __user *pathname) 806 { 807 struct inode *inode = NULL; 808 struct vfsmount *mnt = NULL; 809 struct fsnotify_group *group; 810 struct fd f; 811 struct path path; 812 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS; 813 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS; 814 int ret; 815 816 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n", 817 __func__, fanotify_fd, flags, dfd, pathname, mask); 818 819 /* we only use the lower 32 bits as of right now. */ 820 if (mask & ((__u64)0xffffffff << 32)) 821 return -EINVAL; 822 823 if (flags & ~FANOTIFY_MARK_FLAGS) 824 return -EINVAL; 825 826 switch (mark_type) { 827 case FAN_MARK_INODE: 828 case FAN_MARK_MOUNT: 829 case FAN_MARK_FILESYSTEM: 830 break; 831 default: 832 return -EINVAL; 833 } 834 835 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { 836 case FAN_MARK_ADD: /* fallthrough */ 837 case FAN_MARK_REMOVE: 838 if (!mask) 839 return -EINVAL; 840 break; 841 case FAN_MARK_FLUSH: 842 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH)) 843 return -EINVAL; 844 break; 845 default: 846 return -EINVAL; 847 } 848 849 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 850 valid_mask |= FANOTIFY_PERM_EVENTS; 851 852 if (mask & ~valid_mask) 853 return -EINVAL; 854 855 f = fdget(fanotify_fd); 856 if (unlikely(!f.file)) 857 return -EBADF; 858 859 /* verify that this is indeed an fanotify instance */ 860 ret = -EINVAL; 861 if (unlikely(f.file->f_op != &fanotify_fops)) 862 goto fput_and_out; 863 group = f.file->private_data; 864 865 /* 866 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not 867 * allowed to set permissions events. 868 */ 869 ret = -EINVAL; 870 if (mask & FANOTIFY_PERM_EVENTS && 871 group->priority == FS_PRIO_0) 872 goto fput_and_out; 873 874 if (flags & FAN_MARK_FLUSH) { 875 ret = 0; 876 if (mark_type == FAN_MARK_MOUNT) 877 fsnotify_clear_vfsmount_marks_by_group(group); 878 else if (mark_type == FAN_MARK_FILESYSTEM) 879 fsnotify_clear_sb_marks_by_group(group); 880 else 881 fsnotify_clear_inode_marks_by_group(group); 882 goto fput_and_out; 883 } 884 885 ret = fanotify_find_path(dfd, pathname, &path, flags); 886 if (ret) 887 goto fput_and_out; 888 889 /* inode held in place by reference to path; group by fget on fd */ 890 if (mark_type == FAN_MARK_INODE) 891 inode = path.dentry->d_inode; 892 else 893 mnt = path.mnt; 894 895 /* create/update an inode mark */ 896 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) { 897 case FAN_MARK_ADD: 898 if (mark_type == FAN_MARK_MOUNT) 899 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags); 900 else if (mark_type == FAN_MARK_FILESYSTEM) 901 ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask, flags); 902 else 903 ret = fanotify_add_inode_mark(group, inode, mask, flags); 904 break; 905 case FAN_MARK_REMOVE: 906 if (mark_type == FAN_MARK_MOUNT) 907 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags); 908 else if (mark_type == FAN_MARK_FILESYSTEM) 909 ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask, flags); 910 else 911 ret = fanotify_remove_inode_mark(group, inode, mask, flags); 912 break; 913 default: 914 ret = -EINVAL; 915 } 916 917 path_put(&path); 918 fput_and_out: 919 fdput(f); 920 return ret; 921 } 922 923 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags, 924 __u64, mask, int, dfd, 925 const char __user *, pathname) 926 { 927 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname); 928 } 929 930 #ifdef CONFIG_COMPAT 931 COMPAT_SYSCALL_DEFINE6(fanotify_mark, 932 int, fanotify_fd, unsigned int, flags, 933 __u32, mask0, __u32, mask1, int, dfd, 934 const char __user *, pathname) 935 { 936 return do_fanotify_mark(fanotify_fd, flags, 937 #ifdef __BIG_ENDIAN 938 ((__u64)mask0 << 32) | mask1, 939 #else 940 ((__u64)mask1 << 32) | mask0, 941 #endif 942 dfd, pathname); 943 } 944 #endif 945 946 /* 947 * fanotify_user_setup - Our initialization function. Note that we cannot return 948 * error because we have compiled-in VFS hooks. So an (unlikely) failure here 949 * must result in panic(). 950 */ 951 static int __init fanotify_user_setup(void) 952 { 953 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 7); 954 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9); 955 956 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, 957 SLAB_PANIC|SLAB_ACCOUNT); 958 fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC); 959 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) { 960 fanotify_perm_event_cachep = 961 KMEM_CACHE(fanotify_perm_event_info, SLAB_PANIC); 962 } 963 964 return 0; 965 } 966 device_initcall(fanotify_user_setup); 967