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) 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 if (copy_to_user(buf, &fanotify_event_metadata, 224 fanotify_event_metadata.event_len)) 225 goto out_close_fd; 226 227 if (fanotify_is_perm_event(event->mask)) 228 FANOTIFY_PE(event)->fd = fd; 229 230 if (fd != FAN_NOFD) 231 fd_install(fd, f); 232 return fanotify_event_metadata.event_len; 233 234 out_close_fd: 235 if (fd != FAN_NOFD) { 236 put_unused_fd(fd); 237 fput(f); 238 } 239 return ret; 240 } 241 242 /* intofiy userspace file descriptor functions */ 243 static __poll_t fanotify_poll(struct file *file, poll_table *wait) 244 { 245 struct fsnotify_group *group = file->private_data; 246 __poll_t ret = 0; 247 248 poll_wait(file, &group->notification_waitq, wait); 249 spin_lock(&group->notification_lock); 250 if (!fsnotify_notify_queue_is_empty(group)) 251 ret = EPOLLIN | EPOLLRDNORM; 252 spin_unlock(&group->notification_lock); 253 254 return ret; 255 } 256 257 static ssize_t fanotify_read(struct file *file, char __user *buf, 258 size_t count, loff_t *pos) 259 { 260 struct fsnotify_group *group; 261 struct fsnotify_event *kevent; 262 char __user *start; 263 int ret; 264 DEFINE_WAIT_FUNC(wait, woken_wake_function); 265 266 start = buf; 267 group = file->private_data; 268 269 pr_debug("%s: group=%p\n", __func__, group); 270 271 add_wait_queue(&group->notification_waitq, &wait); 272 while (1) { 273 spin_lock(&group->notification_lock); 274 kevent = get_one_event(group, count); 275 spin_unlock(&group->notification_lock); 276 277 if (IS_ERR(kevent)) { 278 ret = PTR_ERR(kevent); 279 break; 280 } 281 282 if (!kevent) { 283 ret = -EAGAIN; 284 if (file->f_flags & O_NONBLOCK) 285 break; 286 287 ret = -ERESTARTSYS; 288 if (signal_pending(current)) 289 break; 290 291 if (start != buf) 292 break; 293 294 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 295 continue; 296 } 297 298 ret = copy_event_to_user(group, kevent, buf); 299 if (unlikely(ret == -EOPENSTALE)) { 300 /* 301 * We cannot report events with stale fd so drop it. 302 * Setting ret to 0 will continue the event loop and 303 * do the right thing if there are no more events to 304 * read (i.e. return bytes read, -EAGAIN or wait). 305 */ 306 ret = 0; 307 } 308 309 /* 310 * Permission events get queued to wait for response. Other 311 * events can be destroyed now. 312 */ 313 if (!fanotify_is_perm_event(kevent->mask)) { 314 fsnotify_destroy_event(group, kevent); 315 } else { 316 if (ret <= 0) { 317 FANOTIFY_PE(kevent)->response = FAN_DENY; 318 wake_up(&group->fanotify_data.access_waitq); 319 } else { 320 spin_lock(&group->notification_lock); 321 list_add_tail(&kevent->list, 322 &group->fanotify_data.access_list); 323 spin_unlock(&group->notification_lock); 324 } 325 } 326 if (ret < 0) 327 break; 328 buf += ret; 329 count -= ret; 330 } 331 remove_wait_queue(&group->notification_waitq, &wait); 332 333 if (start != buf && ret != -EFAULT) 334 ret = buf - start; 335 return ret; 336 } 337 338 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) 339 { 340 struct fanotify_response response = { .fd = -1, .response = -1 }; 341 struct fsnotify_group *group; 342 int ret; 343 344 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 345 return -EINVAL; 346 347 group = file->private_data; 348 349 if (count > sizeof(response)) 350 count = sizeof(response); 351 352 pr_debug("%s: group=%p count=%zu\n", __func__, group, count); 353 354 if (copy_from_user(&response, buf, count)) 355 return -EFAULT; 356 357 ret = process_access_response(group, &response); 358 if (ret < 0) 359 count = ret; 360 361 return count; 362 } 363 364 static int fanotify_release(struct inode *ignored, struct file *file) 365 { 366 struct fsnotify_group *group = file->private_data; 367 struct fanotify_perm_event_info *event, *next; 368 struct fsnotify_event *fsn_event; 369 370 /* 371 * Stop new events from arriving in the notification queue. since 372 * userspace cannot use fanotify fd anymore, no event can enter or 373 * leave access_list by now either. 374 */ 375 fsnotify_group_stop_queueing(group); 376 377 /* 378 * Process all permission events on access_list and notification queue 379 * and simulate reply from userspace. 380 */ 381 spin_lock(&group->notification_lock); 382 list_for_each_entry_safe(event, next, &group->fanotify_data.access_list, 383 fae.fse.list) { 384 pr_debug("%s: found group=%p event=%p\n", __func__, group, 385 event); 386 387 list_del_init(&event->fae.fse.list); 388 event->response = FAN_ALLOW; 389 } 390 391 /* 392 * Destroy all non-permission events. For permission events just 393 * dequeue them and set the response. They will be freed once the 394 * response is consumed and fanotify_get_response() returns. 395 */ 396 while (!fsnotify_notify_queue_is_empty(group)) { 397 fsn_event = fsnotify_remove_first_event(group); 398 if (!(fsn_event->mask & FANOTIFY_PERM_EVENTS)) { 399 spin_unlock(&group->notification_lock); 400 fsnotify_destroy_event(group, fsn_event); 401 spin_lock(&group->notification_lock); 402 } else { 403 FANOTIFY_PE(fsn_event)->response = FAN_ALLOW; 404 } 405 } 406 spin_unlock(&group->notification_lock); 407 408 /* Response for all permission events it set, wakeup waiters */ 409 wake_up(&group->fanotify_data.access_waitq); 410 411 /* matches the fanotify_init->fsnotify_alloc_group */ 412 fsnotify_destroy_group(group); 413 414 return 0; 415 } 416 417 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 418 { 419 struct fsnotify_group *group; 420 struct fsnotify_event *fsn_event; 421 void __user *p; 422 int ret = -ENOTTY; 423 size_t send_len = 0; 424 425 group = file->private_data; 426 427 p = (void __user *) arg; 428 429 switch (cmd) { 430 case FIONREAD: 431 spin_lock(&group->notification_lock); 432 list_for_each_entry(fsn_event, &group->notification_list, list) 433 send_len += FAN_EVENT_METADATA_LEN; 434 spin_unlock(&group->notification_lock); 435 ret = put_user(send_len, (int __user *) p); 436 break; 437 } 438 439 return ret; 440 } 441 442 static const struct file_operations fanotify_fops = { 443 .show_fdinfo = fanotify_show_fdinfo, 444 .poll = fanotify_poll, 445 .read = fanotify_read, 446 .write = fanotify_write, 447 .fasync = NULL, 448 .release = fanotify_release, 449 .unlocked_ioctl = fanotify_ioctl, 450 .compat_ioctl = fanotify_ioctl, 451 .llseek = noop_llseek, 452 }; 453 454 static int fanotify_find_path(int dfd, const char __user *filename, 455 struct path *path, unsigned int flags) 456 { 457 int ret; 458 459 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__, 460 dfd, filename, flags); 461 462 if (filename == NULL) { 463 struct fd f = fdget(dfd); 464 465 ret = -EBADF; 466 if (!f.file) 467 goto out; 468 469 ret = -ENOTDIR; 470 if ((flags & FAN_MARK_ONLYDIR) && 471 !(S_ISDIR(file_inode(f.file)->i_mode))) { 472 fdput(f); 473 goto out; 474 } 475 476 *path = f.file->f_path; 477 path_get(path); 478 fdput(f); 479 } else { 480 unsigned int lookup_flags = 0; 481 482 if (!(flags & FAN_MARK_DONT_FOLLOW)) 483 lookup_flags |= LOOKUP_FOLLOW; 484 if (flags & FAN_MARK_ONLYDIR) 485 lookup_flags |= LOOKUP_DIRECTORY; 486 487 ret = user_path_at(dfd, filename, lookup_flags, path); 488 if (ret) 489 goto out; 490 } 491 492 /* you can only watch an inode if you have read permissions on it */ 493 ret = inode_permission(path->dentry->d_inode, MAY_READ); 494 if (ret) 495 path_put(path); 496 out: 497 return ret; 498 } 499 500 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, 501 __u32 mask, 502 unsigned int flags, 503 int *destroy) 504 { 505 __u32 oldmask = 0; 506 507 spin_lock(&fsn_mark->lock); 508 if (!(flags & FAN_MARK_IGNORED_MASK)) { 509 oldmask = fsn_mark->mask; 510 fsn_mark->mask &= ~mask; 511 } else { 512 fsn_mark->ignored_mask &= ~mask; 513 } 514 *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask); 515 spin_unlock(&fsn_mark->lock); 516 517 return mask & oldmask; 518 } 519 520 static int fanotify_remove_mark(struct fsnotify_group *group, 521 fsnotify_connp_t *connp, __u32 mask, 522 unsigned int flags) 523 { 524 struct fsnotify_mark *fsn_mark = NULL; 525 __u32 removed; 526 int destroy_mark; 527 528 mutex_lock(&group->mark_mutex); 529 fsn_mark = fsnotify_find_mark(connp, group); 530 if (!fsn_mark) { 531 mutex_unlock(&group->mark_mutex); 532 return -ENOENT; 533 } 534 535 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, 536 &destroy_mark); 537 if (removed & fsnotify_conn_mask(fsn_mark->connector)) 538 fsnotify_recalc_mask(fsn_mark->connector); 539 if (destroy_mark) 540 fsnotify_detach_mark(fsn_mark); 541 mutex_unlock(&group->mark_mutex); 542 if (destroy_mark) 543 fsnotify_free_mark(fsn_mark); 544 545 /* matches the fsnotify_find_mark() */ 546 fsnotify_put_mark(fsn_mark); 547 return 0; 548 } 549 550 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group, 551 struct vfsmount *mnt, __u32 mask, 552 unsigned int flags) 553 { 554 return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 555 mask, flags); 556 } 557 558 static int fanotify_remove_sb_mark(struct fsnotify_group *group, 559 struct super_block *sb, __u32 mask, 560 unsigned int flags) 561 { 562 return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask, flags); 563 } 564 565 static int fanotify_remove_inode_mark(struct fsnotify_group *group, 566 struct inode *inode, __u32 mask, 567 unsigned int flags) 568 { 569 return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask, 570 flags); 571 } 572 573 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, 574 __u32 mask, 575 unsigned int flags) 576 { 577 __u32 oldmask = -1; 578 579 spin_lock(&fsn_mark->lock); 580 if (!(flags & FAN_MARK_IGNORED_MASK)) { 581 oldmask = fsn_mark->mask; 582 fsn_mark->mask |= mask; 583 } else { 584 fsn_mark->ignored_mask |= mask; 585 if (flags & FAN_MARK_IGNORED_SURV_MODIFY) 586 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY; 587 } 588 spin_unlock(&fsn_mark->lock); 589 590 return mask & ~oldmask; 591 } 592 593 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group, 594 fsnotify_connp_t *connp, 595 unsigned int type) 596 { 597 struct fsnotify_mark *mark; 598 int ret; 599 600 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) 601 return ERR_PTR(-ENOSPC); 602 603 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); 604 if (!mark) 605 return ERR_PTR(-ENOMEM); 606 607 fsnotify_init_mark(mark, group); 608 ret = fsnotify_add_mark_locked(mark, connp, type, 0); 609 if (ret) { 610 fsnotify_put_mark(mark); 611 return ERR_PTR(ret); 612 } 613 614 return mark; 615 } 616 617 618 static int fanotify_add_mark(struct fsnotify_group *group, 619 fsnotify_connp_t *connp, unsigned int type, 620 __u32 mask, unsigned int flags) 621 { 622 struct fsnotify_mark *fsn_mark; 623 __u32 added; 624 625 mutex_lock(&group->mark_mutex); 626 fsn_mark = fsnotify_find_mark(connp, group); 627 if (!fsn_mark) { 628 fsn_mark = fanotify_add_new_mark(group, connp, type); 629 if (IS_ERR(fsn_mark)) { 630 mutex_unlock(&group->mark_mutex); 631 return PTR_ERR(fsn_mark); 632 } 633 } 634 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); 635 if (added & ~fsnotify_conn_mask(fsn_mark->connector)) 636 fsnotify_recalc_mask(fsn_mark->connector); 637 mutex_unlock(&group->mark_mutex); 638 639 fsnotify_put_mark(fsn_mark); 640 return 0; 641 } 642 643 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group, 644 struct vfsmount *mnt, __u32 mask, 645 unsigned int flags) 646 { 647 return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 648 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags); 649 } 650 651 static int fanotify_add_sb_mark(struct fsnotify_group *group, 652 struct super_block *sb, __u32 mask, 653 unsigned int flags) 654 { 655 return fanotify_add_mark(group, &sb->s_fsnotify_marks, 656 FSNOTIFY_OBJ_TYPE_SB, mask, flags); 657 } 658 659 static int fanotify_add_inode_mark(struct fsnotify_group *group, 660 struct inode *inode, __u32 mask, 661 unsigned int flags) 662 { 663 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode); 664 665 /* 666 * If some other task has this inode open for write we should not add 667 * an ignored mark, unless that ignored mark is supposed to survive 668 * modification changes anyway. 669 */ 670 if ((flags & FAN_MARK_IGNORED_MASK) && 671 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) && 672 (atomic_read(&inode->i_writecount) > 0)) 673 return 0; 674 675 return fanotify_add_mark(group, &inode->i_fsnotify_marks, 676 FSNOTIFY_OBJ_TYPE_INODE, mask, flags); 677 } 678 679 /* fanotify syscalls */ 680 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags) 681 { 682 struct fsnotify_group *group; 683 int f_flags, fd; 684 struct user_struct *user; 685 struct fanotify_event_info *oevent; 686 687 pr_debug("%s: flags=%x event_f_flags=%x\n", 688 __func__, flags, event_f_flags); 689 690 if (!capable(CAP_SYS_ADMIN)) 691 return -EPERM; 692 693 #ifdef CONFIG_AUDITSYSCALL 694 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT)) 695 #else 696 if (flags & ~FANOTIFY_INIT_FLAGS) 697 #endif 698 return -EINVAL; 699 700 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS) 701 return -EINVAL; 702 703 switch (event_f_flags & O_ACCMODE) { 704 case O_RDONLY: 705 case O_RDWR: 706 case O_WRONLY: 707 break; 708 default: 709 return -EINVAL; 710 } 711 712 user = get_current_user(); 713 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) { 714 free_uid(user); 715 return -EMFILE; 716 } 717 718 f_flags = O_RDWR | FMODE_NONOTIFY; 719 if (flags & FAN_CLOEXEC) 720 f_flags |= O_CLOEXEC; 721 if (flags & FAN_NONBLOCK) 722 f_flags |= O_NONBLOCK; 723 724 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */ 725 group = fsnotify_alloc_group(&fanotify_fsnotify_ops); 726 if (IS_ERR(group)) { 727 free_uid(user); 728 return PTR_ERR(group); 729 } 730 731 group->fanotify_data.user = user; 732 group->fanotify_data.flags = flags; 733 atomic_inc(&user->fanotify_listeners); 734 group->memcg = get_mem_cgroup_from_mm(current->mm); 735 736 oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL); 737 if (unlikely(!oevent)) { 738 fd = -ENOMEM; 739 goto out_destroy_group; 740 } 741 group->overflow_event = &oevent->fse; 742 743 if (force_o_largefile()) 744 event_f_flags |= O_LARGEFILE; 745 group->fanotify_data.f_flags = event_f_flags; 746 init_waitqueue_head(&group->fanotify_data.access_waitq); 747 INIT_LIST_HEAD(&group->fanotify_data.access_list); 748 switch (flags & FANOTIFY_CLASS_BITS) { 749 case FAN_CLASS_NOTIF: 750 group->priority = FS_PRIO_0; 751 break; 752 case FAN_CLASS_CONTENT: 753 group->priority = FS_PRIO_1; 754 break; 755 case FAN_CLASS_PRE_CONTENT: 756 group->priority = FS_PRIO_2; 757 break; 758 default: 759 fd = -EINVAL; 760 goto out_destroy_group; 761 } 762 763 if (flags & FAN_UNLIMITED_QUEUE) { 764 fd = -EPERM; 765 if (!capable(CAP_SYS_ADMIN)) 766 goto out_destroy_group; 767 group->max_events = UINT_MAX; 768 } else { 769 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS; 770 } 771 772 if (flags & FAN_UNLIMITED_MARKS) { 773 fd = -EPERM; 774 if (!capable(CAP_SYS_ADMIN)) 775 goto out_destroy_group; 776 group->fanotify_data.max_marks = UINT_MAX; 777 } else { 778 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS; 779 } 780 781 if (flags & FAN_ENABLE_AUDIT) { 782 fd = -EPERM; 783 if (!capable(CAP_AUDIT_WRITE)) 784 goto out_destroy_group; 785 } 786 787 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags); 788 if (fd < 0) 789 goto out_destroy_group; 790 791 return fd; 792 793 out_destroy_group: 794 fsnotify_destroy_group(group); 795 return fd; 796 } 797 798 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask, 799 int dfd, const char __user *pathname) 800 { 801 struct inode *inode = NULL; 802 struct vfsmount *mnt = NULL; 803 struct fsnotify_group *group; 804 struct fd f; 805 struct path path; 806 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS; 807 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS; 808 int ret; 809 810 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n", 811 __func__, fanotify_fd, flags, dfd, pathname, mask); 812 813 /* we only use the lower 32 bits as of right now. */ 814 if (mask & ((__u64)0xffffffff << 32)) 815 return -EINVAL; 816 817 if (flags & ~FANOTIFY_MARK_FLAGS) 818 return -EINVAL; 819 820 switch (mark_type) { 821 case FAN_MARK_INODE: 822 case FAN_MARK_MOUNT: 823 case FAN_MARK_FILESYSTEM: 824 break; 825 default: 826 return -EINVAL; 827 } 828 829 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { 830 case FAN_MARK_ADD: /* fallthrough */ 831 case FAN_MARK_REMOVE: 832 if (!mask) 833 return -EINVAL; 834 break; 835 case FAN_MARK_FLUSH: 836 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH)) 837 return -EINVAL; 838 break; 839 default: 840 return -EINVAL; 841 } 842 843 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 844 valid_mask |= FANOTIFY_PERM_EVENTS; 845 846 if (mask & ~valid_mask) 847 return -EINVAL; 848 849 f = fdget(fanotify_fd); 850 if (unlikely(!f.file)) 851 return -EBADF; 852 853 /* verify that this is indeed an fanotify instance */ 854 ret = -EINVAL; 855 if (unlikely(f.file->f_op != &fanotify_fops)) 856 goto fput_and_out; 857 group = f.file->private_data; 858 859 /* 860 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not 861 * allowed to set permissions events. 862 */ 863 ret = -EINVAL; 864 if (mask & FANOTIFY_PERM_EVENTS && 865 group->priority == FS_PRIO_0) 866 goto fput_and_out; 867 868 if (flags & FAN_MARK_FLUSH) { 869 ret = 0; 870 if (mark_type == FAN_MARK_MOUNT) 871 fsnotify_clear_vfsmount_marks_by_group(group); 872 else if (mark_type == FAN_MARK_FILESYSTEM) 873 fsnotify_clear_sb_marks_by_group(group); 874 else 875 fsnotify_clear_inode_marks_by_group(group); 876 goto fput_and_out; 877 } 878 879 ret = fanotify_find_path(dfd, pathname, &path, flags); 880 if (ret) 881 goto fput_and_out; 882 883 /* inode held in place by reference to path; group by fget on fd */ 884 if (mark_type == FAN_MARK_INODE) 885 inode = path.dentry->d_inode; 886 else 887 mnt = path.mnt; 888 889 /* create/update an inode mark */ 890 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) { 891 case FAN_MARK_ADD: 892 if (mark_type == FAN_MARK_MOUNT) 893 ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags); 894 else if (mark_type == FAN_MARK_FILESYSTEM) 895 ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask, flags); 896 else 897 ret = fanotify_add_inode_mark(group, inode, mask, flags); 898 break; 899 case FAN_MARK_REMOVE: 900 if (mark_type == FAN_MARK_MOUNT) 901 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags); 902 else if (mark_type == FAN_MARK_FILESYSTEM) 903 ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask, flags); 904 else 905 ret = fanotify_remove_inode_mark(group, inode, mask, flags); 906 break; 907 default: 908 ret = -EINVAL; 909 } 910 911 path_put(&path); 912 fput_and_out: 913 fdput(f); 914 return ret; 915 } 916 917 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags, 918 __u64, mask, int, dfd, 919 const char __user *, pathname) 920 { 921 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname); 922 } 923 924 #ifdef CONFIG_COMPAT 925 COMPAT_SYSCALL_DEFINE6(fanotify_mark, 926 int, fanotify_fd, unsigned int, flags, 927 __u32, mask0, __u32, mask1, int, dfd, 928 const char __user *, pathname) 929 { 930 return do_fanotify_mark(fanotify_fd, flags, 931 #ifdef __BIG_ENDIAN 932 ((__u64)mask0 << 32) | mask1, 933 #else 934 ((__u64)mask1 << 32) | mask0, 935 #endif 936 dfd, pathname); 937 } 938 #endif 939 940 /* 941 * fanotify_user_setup - Our initialization function. Note that we cannot return 942 * error because we have compiled-in VFS hooks. So an (unlikely) failure here 943 * must result in panic(). 944 */ 945 static int __init fanotify_user_setup(void) 946 { 947 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 7); 948 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9); 949 950 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, 951 SLAB_PANIC|SLAB_ACCOUNT); 952 fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC); 953 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) { 954 fanotify_perm_event_cachep = 955 KMEM_CACHE(fanotify_perm_event_info, SLAB_PANIC); 956 } 957 958 return 0; 959 } 960 device_initcall(fanotify_user_setup); 961