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 #include <linux/statfs.h> 21 #include <linux/exportfs.h> 22 23 #include <asm/ioctls.h> 24 25 #include "../../mount.h" 26 #include "../fdinfo.h" 27 #include "fanotify.h" 28 29 #define FANOTIFY_DEFAULT_MAX_EVENTS 16384 30 #define FANOTIFY_DEFAULT_MAX_MARKS 8192 31 #define FANOTIFY_DEFAULT_MAX_LISTENERS 128 32 33 /* 34 * All flags that may be specified in parameter event_f_flags of fanotify_init. 35 * 36 * Internal and external open flags are stored together in field f_flags of 37 * struct file. Only external open flags shall be allowed in event_f_flags. 38 * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be 39 * excluded. 40 */ 41 #define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \ 42 O_ACCMODE | O_APPEND | O_NONBLOCK | \ 43 __O_SYNC | O_DSYNC | O_CLOEXEC | \ 44 O_LARGEFILE | O_NOATIME ) 45 46 extern const struct fsnotify_ops fanotify_fsnotify_ops; 47 48 struct kmem_cache *fanotify_mark_cache __read_mostly; 49 struct kmem_cache *fanotify_fid_event_cachep __read_mostly; 50 struct kmem_cache *fanotify_path_event_cachep __read_mostly; 51 struct kmem_cache *fanotify_perm_event_cachep __read_mostly; 52 53 #define FANOTIFY_EVENT_ALIGN 4 54 #define FANOTIFY_INFO_HDR_LEN \ 55 (sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle)) 56 57 static int fanotify_fid_info_len(int fh_len, int name_len) 58 { 59 int info_len = fh_len; 60 61 if (name_len) 62 info_len += name_len + 1; 63 64 return roundup(FANOTIFY_INFO_HDR_LEN + info_len, FANOTIFY_EVENT_ALIGN); 65 } 66 67 static int fanotify_event_info_len(struct fanotify_event *event) 68 { 69 int info_len = 0; 70 int fh_len = fanotify_event_object_fh_len(event); 71 72 if (fh_len) 73 info_len += fanotify_fid_info_len(fh_len, 0); 74 75 if (fanotify_event_name_len(event)) { 76 struct fanotify_name_event *fne = FANOTIFY_NE(event); 77 78 info_len += fanotify_fid_info_len(fne->dir_fh.len, 79 fne->name_len); 80 } 81 82 return info_len; 83 } 84 85 /* 86 * Get an fanotify notification event if one exists and is small 87 * enough to fit in "count". Return an error pointer if the count 88 * is not large enough. When permission event is dequeued, its state is 89 * updated accordingly. 90 */ 91 static struct fanotify_event *get_one_event(struct fsnotify_group *group, 92 size_t count) 93 { 94 size_t event_size = FAN_EVENT_METADATA_LEN; 95 struct fanotify_event *event = NULL; 96 97 pr_debug("%s: group=%p count=%zd\n", __func__, group, count); 98 99 spin_lock(&group->notification_lock); 100 if (fsnotify_notify_queue_is_empty(group)) 101 goto out; 102 103 if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) { 104 event_size += fanotify_event_info_len( 105 FANOTIFY_E(fsnotify_peek_first_event(group))); 106 } 107 108 if (event_size > count) { 109 event = ERR_PTR(-EINVAL); 110 goto out; 111 } 112 event = FANOTIFY_E(fsnotify_remove_first_event(group)); 113 if (fanotify_is_perm_event(event->mask)) 114 FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED; 115 out: 116 spin_unlock(&group->notification_lock); 117 return event; 118 } 119 120 static int create_fd(struct fsnotify_group *group, struct path *path, 121 struct file **file) 122 { 123 int client_fd; 124 struct file *new_file; 125 126 client_fd = get_unused_fd_flags(group->fanotify_data.f_flags); 127 if (client_fd < 0) 128 return client_fd; 129 130 /* 131 * we need a new file handle for the userspace program so it can read even if it was 132 * originally opened O_WRONLY. 133 */ 134 new_file = dentry_open(path, 135 group->fanotify_data.f_flags | FMODE_NONOTIFY, 136 current_cred()); 137 if (IS_ERR(new_file)) { 138 /* 139 * we still send an event even if we can't open the file. this 140 * can happen when say tasks are gone and we try to open their 141 * /proc files or we try to open a WRONLY file like in sysfs 142 * we just send the errno to userspace since there isn't much 143 * else we can do. 144 */ 145 put_unused_fd(client_fd); 146 client_fd = PTR_ERR(new_file); 147 } else { 148 *file = new_file; 149 } 150 151 return client_fd; 152 } 153 154 /* 155 * Finish processing of permission event by setting it to ANSWERED state and 156 * drop group->notification_lock. 157 */ 158 static void finish_permission_event(struct fsnotify_group *group, 159 struct fanotify_perm_event *event, 160 unsigned int response) 161 __releases(&group->notification_lock) 162 { 163 bool destroy = false; 164 165 assert_spin_locked(&group->notification_lock); 166 event->response = response; 167 if (event->state == FAN_EVENT_CANCELED) 168 destroy = true; 169 else 170 event->state = FAN_EVENT_ANSWERED; 171 spin_unlock(&group->notification_lock); 172 if (destroy) 173 fsnotify_destroy_event(group, &event->fae.fse); 174 } 175 176 static int process_access_response(struct fsnotify_group *group, 177 struct fanotify_response *response_struct) 178 { 179 struct fanotify_perm_event *event; 180 int fd = response_struct->fd; 181 int response = response_struct->response; 182 183 pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group, 184 fd, response); 185 /* 186 * make sure the response is valid, if invalid we do nothing and either 187 * userspace can send a valid response or we will clean it up after the 188 * timeout 189 */ 190 switch (response & ~FAN_AUDIT) { 191 case FAN_ALLOW: 192 case FAN_DENY: 193 break; 194 default: 195 return -EINVAL; 196 } 197 198 if (fd < 0) 199 return -EINVAL; 200 201 if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT)) 202 return -EINVAL; 203 204 spin_lock(&group->notification_lock); 205 list_for_each_entry(event, &group->fanotify_data.access_list, 206 fae.fse.list) { 207 if (event->fd != fd) 208 continue; 209 210 list_del_init(&event->fae.fse.list); 211 finish_permission_event(group, event, response); 212 wake_up(&group->fanotify_data.access_waitq); 213 return 0; 214 } 215 spin_unlock(&group->notification_lock); 216 217 return -ENOENT; 218 } 219 220 static int copy_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh, 221 const char *name, size_t name_len, 222 char __user *buf, size_t count) 223 { 224 struct fanotify_event_info_fid info = { }; 225 struct file_handle handle = { }; 226 unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf; 227 size_t fh_len = fh ? fh->len : 0; 228 size_t info_len = fanotify_fid_info_len(fh_len, name_len); 229 size_t len = info_len; 230 231 pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n", 232 __func__, fh_len, name_len, info_len, count); 233 234 if (!fh_len || (name && !name_len)) 235 return 0; 236 237 if (WARN_ON_ONCE(len < sizeof(info) || len > count)) 238 return -EFAULT; 239 240 /* 241 * Copy event info fid header followed by variable sized file handle 242 * and optionally followed by variable sized filename. 243 */ 244 info.hdr.info_type = name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME : 245 FAN_EVENT_INFO_TYPE_FID; 246 info.hdr.len = len; 247 info.fsid = *fsid; 248 if (copy_to_user(buf, &info, sizeof(info))) 249 return -EFAULT; 250 251 buf += sizeof(info); 252 len -= sizeof(info); 253 if (WARN_ON_ONCE(len < sizeof(handle))) 254 return -EFAULT; 255 256 handle.handle_type = fh->type; 257 handle.handle_bytes = fh_len; 258 if (copy_to_user(buf, &handle, sizeof(handle))) 259 return -EFAULT; 260 261 buf += sizeof(handle); 262 len -= sizeof(handle); 263 if (WARN_ON_ONCE(len < fh_len)) 264 return -EFAULT; 265 266 /* 267 * For an inline fh and inline file name, copy through stack to exclude 268 * the copy from usercopy hardening protections. 269 */ 270 fh_buf = fanotify_fh_buf(fh); 271 if (fh_len <= FANOTIFY_INLINE_FH_LEN) { 272 memcpy(bounce, fh_buf, fh_len); 273 fh_buf = bounce; 274 } 275 if (copy_to_user(buf, fh_buf, fh_len)) 276 return -EFAULT; 277 278 buf += fh_len; 279 len -= fh_len; 280 281 if (name_len) { 282 /* Copy the filename with terminating null */ 283 name_len++; 284 if (WARN_ON_ONCE(len < name_len)) 285 return -EFAULT; 286 287 if (copy_to_user(buf, name, name_len)) 288 return -EFAULT; 289 290 buf += name_len; 291 len -= name_len; 292 } 293 294 /* Pad with 0's */ 295 WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN); 296 if (len > 0 && clear_user(buf, len)) 297 return -EFAULT; 298 299 return info_len; 300 } 301 302 static ssize_t copy_event_to_user(struct fsnotify_group *group, 303 struct fanotify_event *event, 304 char __user *buf, size_t count) 305 { 306 struct fanotify_event_metadata metadata; 307 struct path *path = fanotify_event_path(event); 308 struct file *f = NULL; 309 int ret, fd = FAN_NOFD; 310 311 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 312 313 metadata.event_len = FAN_EVENT_METADATA_LEN + 314 fanotify_event_info_len(event); 315 metadata.metadata_len = FAN_EVENT_METADATA_LEN; 316 metadata.vers = FANOTIFY_METADATA_VERSION; 317 metadata.reserved = 0; 318 metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS; 319 metadata.pid = pid_vnr(event->pid); 320 321 if (path && path->mnt && path->dentry) { 322 fd = create_fd(group, path, &f); 323 if (fd < 0) 324 return fd; 325 } 326 metadata.fd = fd; 327 328 ret = -EFAULT; 329 /* 330 * Sanity check copy size in case get_one_event() and 331 * fill_event_metadata() event_len sizes ever get out of sync. 332 */ 333 if (WARN_ON_ONCE(metadata.event_len > count)) 334 goto out_close_fd; 335 336 if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN)) 337 goto out_close_fd; 338 339 buf += FAN_EVENT_METADATA_LEN; 340 count -= FAN_EVENT_METADATA_LEN; 341 342 if (fanotify_is_perm_event(event->mask)) 343 FANOTIFY_PERM(event)->fd = fd; 344 345 if (f) 346 fd_install(fd, f); 347 348 /* Event info records order is: dir fid + name, child fid */ 349 if (fanotify_event_name_len(event)) { 350 struct fanotify_name_event *fne = FANOTIFY_NE(event); 351 352 ret = copy_info_to_user(fanotify_event_fsid(event), 353 fanotify_event_dir_fh(event), 354 fne->name, fne->name_len, 355 buf, count); 356 if (ret < 0) 357 return ret; 358 359 buf += ret; 360 count -= ret; 361 } 362 363 if (fanotify_event_object_fh_len(event)) { 364 ret = copy_info_to_user(fanotify_event_fsid(event), 365 fanotify_event_object_fh(event), 366 NULL, 0, buf, count); 367 if (ret < 0) 368 return ret; 369 370 buf += ret; 371 count -= ret; 372 } 373 374 return metadata.event_len; 375 376 out_close_fd: 377 if (fd != FAN_NOFD) { 378 put_unused_fd(fd); 379 fput(f); 380 } 381 return ret; 382 } 383 384 /* intofiy userspace file descriptor functions */ 385 static __poll_t fanotify_poll(struct file *file, poll_table *wait) 386 { 387 struct fsnotify_group *group = file->private_data; 388 __poll_t ret = 0; 389 390 poll_wait(file, &group->notification_waitq, wait); 391 spin_lock(&group->notification_lock); 392 if (!fsnotify_notify_queue_is_empty(group)) 393 ret = EPOLLIN | EPOLLRDNORM; 394 spin_unlock(&group->notification_lock); 395 396 return ret; 397 } 398 399 static ssize_t fanotify_read(struct file *file, char __user *buf, 400 size_t count, loff_t *pos) 401 { 402 struct fsnotify_group *group; 403 struct fanotify_event *event; 404 char __user *start; 405 int ret; 406 DEFINE_WAIT_FUNC(wait, woken_wake_function); 407 408 start = buf; 409 group = file->private_data; 410 411 pr_debug("%s: group=%p\n", __func__, group); 412 413 add_wait_queue(&group->notification_waitq, &wait); 414 while (1) { 415 event = get_one_event(group, count); 416 if (IS_ERR(event)) { 417 ret = PTR_ERR(event); 418 break; 419 } 420 421 if (!event) { 422 ret = -EAGAIN; 423 if (file->f_flags & O_NONBLOCK) 424 break; 425 426 ret = -ERESTARTSYS; 427 if (signal_pending(current)) 428 break; 429 430 if (start != buf) 431 break; 432 433 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 434 continue; 435 } 436 437 ret = copy_event_to_user(group, event, buf, count); 438 if (unlikely(ret == -EOPENSTALE)) { 439 /* 440 * We cannot report events with stale fd so drop it. 441 * Setting ret to 0 will continue the event loop and 442 * do the right thing if there are no more events to 443 * read (i.e. return bytes read, -EAGAIN or wait). 444 */ 445 ret = 0; 446 } 447 448 /* 449 * Permission events get queued to wait for response. Other 450 * events can be destroyed now. 451 */ 452 if (!fanotify_is_perm_event(event->mask)) { 453 fsnotify_destroy_event(group, &event->fse); 454 } else { 455 if (ret <= 0) { 456 spin_lock(&group->notification_lock); 457 finish_permission_event(group, 458 FANOTIFY_PERM(event), FAN_DENY); 459 wake_up(&group->fanotify_data.access_waitq); 460 } else { 461 spin_lock(&group->notification_lock); 462 list_add_tail(&event->fse.list, 463 &group->fanotify_data.access_list); 464 spin_unlock(&group->notification_lock); 465 } 466 } 467 if (ret < 0) 468 break; 469 buf += ret; 470 count -= ret; 471 } 472 remove_wait_queue(&group->notification_waitq, &wait); 473 474 if (start != buf && ret != -EFAULT) 475 ret = buf - start; 476 return ret; 477 } 478 479 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) 480 { 481 struct fanotify_response response = { .fd = -1, .response = -1 }; 482 struct fsnotify_group *group; 483 int ret; 484 485 if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 486 return -EINVAL; 487 488 group = file->private_data; 489 490 if (count > sizeof(response)) 491 count = sizeof(response); 492 493 pr_debug("%s: group=%p count=%zu\n", __func__, group, count); 494 495 if (copy_from_user(&response, buf, count)) 496 return -EFAULT; 497 498 ret = process_access_response(group, &response); 499 if (ret < 0) 500 count = ret; 501 502 return count; 503 } 504 505 static int fanotify_release(struct inode *ignored, struct file *file) 506 { 507 struct fsnotify_group *group = file->private_data; 508 509 /* 510 * Stop new events from arriving in the notification queue. since 511 * userspace cannot use fanotify fd anymore, no event can enter or 512 * leave access_list by now either. 513 */ 514 fsnotify_group_stop_queueing(group); 515 516 /* 517 * Process all permission events on access_list and notification queue 518 * and simulate reply from userspace. 519 */ 520 spin_lock(&group->notification_lock); 521 while (!list_empty(&group->fanotify_data.access_list)) { 522 struct fanotify_perm_event *event; 523 524 event = list_first_entry(&group->fanotify_data.access_list, 525 struct fanotify_perm_event, fae.fse.list); 526 list_del_init(&event->fae.fse.list); 527 finish_permission_event(group, event, FAN_ALLOW); 528 spin_lock(&group->notification_lock); 529 } 530 531 /* 532 * Destroy all non-permission events. For permission events just 533 * dequeue them and set the response. They will be freed once the 534 * response is consumed and fanotify_get_response() returns. 535 */ 536 while (!fsnotify_notify_queue_is_empty(group)) { 537 struct fanotify_event *event; 538 539 event = FANOTIFY_E(fsnotify_remove_first_event(group)); 540 if (!(event->mask & FANOTIFY_PERM_EVENTS)) { 541 spin_unlock(&group->notification_lock); 542 fsnotify_destroy_event(group, &event->fse); 543 } else { 544 finish_permission_event(group, FANOTIFY_PERM(event), 545 FAN_ALLOW); 546 } 547 spin_lock(&group->notification_lock); 548 } 549 spin_unlock(&group->notification_lock); 550 551 /* Response for all permission events it set, wakeup waiters */ 552 wake_up(&group->fanotify_data.access_waitq); 553 554 /* matches the fanotify_init->fsnotify_alloc_group */ 555 fsnotify_destroy_group(group); 556 557 return 0; 558 } 559 560 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 561 { 562 struct fsnotify_group *group; 563 struct fsnotify_event *fsn_event; 564 void __user *p; 565 int ret = -ENOTTY; 566 size_t send_len = 0; 567 568 group = file->private_data; 569 570 p = (void __user *) arg; 571 572 switch (cmd) { 573 case FIONREAD: 574 spin_lock(&group->notification_lock); 575 list_for_each_entry(fsn_event, &group->notification_list, list) 576 send_len += FAN_EVENT_METADATA_LEN; 577 spin_unlock(&group->notification_lock); 578 ret = put_user(send_len, (int __user *) p); 579 break; 580 } 581 582 return ret; 583 } 584 585 static const struct file_operations fanotify_fops = { 586 .show_fdinfo = fanotify_show_fdinfo, 587 .poll = fanotify_poll, 588 .read = fanotify_read, 589 .write = fanotify_write, 590 .fasync = NULL, 591 .release = fanotify_release, 592 .unlocked_ioctl = fanotify_ioctl, 593 .compat_ioctl = compat_ptr_ioctl, 594 .llseek = noop_llseek, 595 }; 596 597 static int fanotify_find_path(int dfd, const char __user *filename, 598 struct path *path, unsigned int flags, __u64 mask, 599 unsigned int obj_type) 600 { 601 int ret; 602 603 pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__, 604 dfd, filename, flags); 605 606 if (filename == NULL) { 607 struct fd f = fdget(dfd); 608 609 ret = -EBADF; 610 if (!f.file) 611 goto out; 612 613 ret = -ENOTDIR; 614 if ((flags & FAN_MARK_ONLYDIR) && 615 !(S_ISDIR(file_inode(f.file)->i_mode))) { 616 fdput(f); 617 goto out; 618 } 619 620 *path = f.file->f_path; 621 path_get(path); 622 fdput(f); 623 } else { 624 unsigned int lookup_flags = 0; 625 626 if (!(flags & FAN_MARK_DONT_FOLLOW)) 627 lookup_flags |= LOOKUP_FOLLOW; 628 if (flags & FAN_MARK_ONLYDIR) 629 lookup_flags |= LOOKUP_DIRECTORY; 630 631 ret = user_path_at(dfd, filename, lookup_flags, path); 632 if (ret) 633 goto out; 634 } 635 636 /* you can only watch an inode if you have read permissions on it */ 637 ret = inode_permission(path->dentry->d_inode, MAY_READ); 638 if (ret) { 639 path_put(path); 640 goto out; 641 } 642 643 ret = security_path_notify(path, mask, obj_type); 644 if (ret) 645 path_put(path); 646 647 out: 648 return ret; 649 } 650 651 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, 652 __u32 mask, 653 unsigned int flags, 654 int *destroy) 655 { 656 __u32 oldmask = 0; 657 658 spin_lock(&fsn_mark->lock); 659 if (!(flags & FAN_MARK_IGNORED_MASK)) { 660 oldmask = fsn_mark->mask; 661 fsn_mark->mask &= ~mask; 662 } else { 663 fsn_mark->ignored_mask &= ~mask; 664 } 665 *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask); 666 spin_unlock(&fsn_mark->lock); 667 668 return mask & oldmask; 669 } 670 671 static int fanotify_remove_mark(struct fsnotify_group *group, 672 fsnotify_connp_t *connp, __u32 mask, 673 unsigned int flags) 674 { 675 struct fsnotify_mark *fsn_mark = NULL; 676 __u32 removed; 677 int destroy_mark; 678 679 mutex_lock(&group->mark_mutex); 680 fsn_mark = fsnotify_find_mark(connp, group); 681 if (!fsn_mark) { 682 mutex_unlock(&group->mark_mutex); 683 return -ENOENT; 684 } 685 686 removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, 687 &destroy_mark); 688 if (removed & fsnotify_conn_mask(fsn_mark->connector)) 689 fsnotify_recalc_mask(fsn_mark->connector); 690 if (destroy_mark) 691 fsnotify_detach_mark(fsn_mark); 692 mutex_unlock(&group->mark_mutex); 693 if (destroy_mark) 694 fsnotify_free_mark(fsn_mark); 695 696 /* matches the fsnotify_find_mark() */ 697 fsnotify_put_mark(fsn_mark); 698 return 0; 699 } 700 701 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group, 702 struct vfsmount *mnt, __u32 mask, 703 unsigned int flags) 704 { 705 return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 706 mask, flags); 707 } 708 709 static int fanotify_remove_sb_mark(struct fsnotify_group *group, 710 struct super_block *sb, __u32 mask, 711 unsigned int flags) 712 { 713 return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask, flags); 714 } 715 716 static int fanotify_remove_inode_mark(struct fsnotify_group *group, 717 struct inode *inode, __u32 mask, 718 unsigned int flags) 719 { 720 return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask, 721 flags); 722 } 723 724 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, 725 __u32 mask, 726 unsigned int flags) 727 { 728 __u32 oldmask = -1; 729 730 spin_lock(&fsn_mark->lock); 731 if (!(flags & FAN_MARK_IGNORED_MASK)) { 732 oldmask = fsn_mark->mask; 733 fsn_mark->mask |= mask; 734 } else { 735 fsn_mark->ignored_mask |= mask; 736 if (flags & FAN_MARK_IGNORED_SURV_MODIFY) 737 fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY; 738 } 739 spin_unlock(&fsn_mark->lock); 740 741 return mask & ~oldmask; 742 } 743 744 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group, 745 fsnotify_connp_t *connp, 746 unsigned int type, 747 __kernel_fsid_t *fsid) 748 { 749 struct fsnotify_mark *mark; 750 int ret; 751 752 if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) 753 return ERR_PTR(-ENOSPC); 754 755 mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); 756 if (!mark) 757 return ERR_PTR(-ENOMEM); 758 759 fsnotify_init_mark(mark, group); 760 ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid); 761 if (ret) { 762 fsnotify_put_mark(mark); 763 return ERR_PTR(ret); 764 } 765 766 return mark; 767 } 768 769 770 static int fanotify_add_mark(struct fsnotify_group *group, 771 fsnotify_connp_t *connp, unsigned int type, 772 __u32 mask, unsigned int flags, 773 __kernel_fsid_t *fsid) 774 { 775 struct fsnotify_mark *fsn_mark; 776 __u32 added; 777 778 mutex_lock(&group->mark_mutex); 779 fsn_mark = fsnotify_find_mark(connp, group); 780 if (!fsn_mark) { 781 fsn_mark = fanotify_add_new_mark(group, connp, type, fsid); 782 if (IS_ERR(fsn_mark)) { 783 mutex_unlock(&group->mark_mutex); 784 return PTR_ERR(fsn_mark); 785 } 786 } 787 added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); 788 if (added & ~fsnotify_conn_mask(fsn_mark->connector)) 789 fsnotify_recalc_mask(fsn_mark->connector); 790 mutex_unlock(&group->mark_mutex); 791 792 fsnotify_put_mark(fsn_mark); 793 return 0; 794 } 795 796 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group, 797 struct vfsmount *mnt, __u32 mask, 798 unsigned int flags, __kernel_fsid_t *fsid) 799 { 800 return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, 801 FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid); 802 } 803 804 static int fanotify_add_sb_mark(struct fsnotify_group *group, 805 struct super_block *sb, __u32 mask, 806 unsigned int flags, __kernel_fsid_t *fsid) 807 { 808 return fanotify_add_mark(group, &sb->s_fsnotify_marks, 809 FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid); 810 } 811 812 static int fanotify_add_inode_mark(struct fsnotify_group *group, 813 struct inode *inode, __u32 mask, 814 unsigned int flags, __kernel_fsid_t *fsid) 815 { 816 pr_debug("%s: group=%p inode=%p\n", __func__, group, inode); 817 818 /* 819 * If some other task has this inode open for write we should not add 820 * an ignored mark, unless that ignored mark is supposed to survive 821 * modification changes anyway. 822 */ 823 if ((flags & FAN_MARK_IGNORED_MASK) && 824 !(flags & FAN_MARK_IGNORED_SURV_MODIFY) && 825 inode_is_open_for_write(inode)) 826 return 0; 827 828 return fanotify_add_mark(group, &inode->i_fsnotify_marks, 829 FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid); 830 } 831 832 /* fanotify syscalls */ 833 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags) 834 { 835 struct fsnotify_group *group; 836 int f_flags, fd; 837 struct user_struct *user; 838 struct fanotify_event *oevent; 839 840 pr_debug("%s: flags=%x event_f_flags=%x\n", 841 __func__, flags, event_f_flags); 842 843 if (!capable(CAP_SYS_ADMIN)) 844 return -EPERM; 845 846 #ifdef CONFIG_AUDITSYSCALL 847 if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT)) 848 #else 849 if (flags & ~FANOTIFY_INIT_FLAGS) 850 #endif 851 return -EINVAL; 852 853 if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS) 854 return -EINVAL; 855 856 switch (event_f_flags & O_ACCMODE) { 857 case O_RDONLY: 858 case O_RDWR: 859 case O_WRONLY: 860 break; 861 default: 862 return -EINVAL; 863 } 864 865 if ((flags & FAN_REPORT_FID) && 866 (flags & FANOTIFY_CLASS_BITS) != FAN_CLASS_NOTIF) 867 return -EINVAL; 868 869 user = get_current_user(); 870 if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) { 871 free_uid(user); 872 return -EMFILE; 873 } 874 875 f_flags = O_RDWR | FMODE_NONOTIFY; 876 if (flags & FAN_CLOEXEC) 877 f_flags |= O_CLOEXEC; 878 if (flags & FAN_NONBLOCK) 879 f_flags |= O_NONBLOCK; 880 881 /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */ 882 group = fsnotify_alloc_group(&fanotify_fsnotify_ops); 883 if (IS_ERR(group)) { 884 free_uid(user); 885 return PTR_ERR(group); 886 } 887 888 group->fanotify_data.user = user; 889 group->fanotify_data.flags = flags; 890 atomic_inc(&user->fanotify_listeners); 891 group->memcg = get_mem_cgroup_from_mm(current->mm); 892 893 oevent = fanotify_alloc_event(group, NULL, FS_Q_OVERFLOW, NULL, 894 FSNOTIFY_EVENT_NONE, NULL, NULL); 895 if (unlikely(!oevent)) { 896 fd = -ENOMEM; 897 goto out_destroy_group; 898 } 899 group->overflow_event = &oevent->fse; 900 901 if (force_o_largefile()) 902 event_f_flags |= O_LARGEFILE; 903 group->fanotify_data.f_flags = event_f_flags; 904 init_waitqueue_head(&group->fanotify_data.access_waitq); 905 INIT_LIST_HEAD(&group->fanotify_data.access_list); 906 switch (flags & FANOTIFY_CLASS_BITS) { 907 case FAN_CLASS_NOTIF: 908 group->priority = FS_PRIO_0; 909 break; 910 case FAN_CLASS_CONTENT: 911 group->priority = FS_PRIO_1; 912 break; 913 case FAN_CLASS_PRE_CONTENT: 914 group->priority = FS_PRIO_2; 915 break; 916 default: 917 fd = -EINVAL; 918 goto out_destroy_group; 919 } 920 921 if (flags & FAN_UNLIMITED_QUEUE) { 922 fd = -EPERM; 923 if (!capable(CAP_SYS_ADMIN)) 924 goto out_destroy_group; 925 group->max_events = UINT_MAX; 926 } else { 927 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS; 928 } 929 930 if (flags & FAN_UNLIMITED_MARKS) { 931 fd = -EPERM; 932 if (!capable(CAP_SYS_ADMIN)) 933 goto out_destroy_group; 934 group->fanotify_data.max_marks = UINT_MAX; 935 } else { 936 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS; 937 } 938 939 if (flags & FAN_ENABLE_AUDIT) { 940 fd = -EPERM; 941 if (!capable(CAP_AUDIT_WRITE)) 942 goto out_destroy_group; 943 } 944 945 fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags); 946 if (fd < 0) 947 goto out_destroy_group; 948 949 return fd; 950 951 out_destroy_group: 952 fsnotify_destroy_group(group); 953 return fd; 954 } 955 956 /* Check if filesystem can encode a unique fid */ 957 static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) 958 { 959 __kernel_fsid_t root_fsid; 960 int err; 961 962 /* 963 * Make sure path is not in filesystem with zero fsid (e.g. tmpfs). 964 */ 965 err = vfs_get_fsid(path->dentry, fsid); 966 if (err) 967 return err; 968 969 if (!fsid->val[0] && !fsid->val[1]) 970 return -ENODEV; 971 972 /* 973 * Make sure path is not inside a filesystem subvolume (e.g. btrfs) 974 * which uses a different fsid than sb root. 975 */ 976 err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid); 977 if (err) 978 return err; 979 980 if (root_fsid.val[0] != fsid->val[0] || 981 root_fsid.val[1] != fsid->val[1]) 982 return -EXDEV; 983 984 /* 985 * We need to make sure that the file system supports at least 986 * encoding a file handle so user can use name_to_handle_at() to 987 * compare fid returned with event to the file handle of watched 988 * objects. However, name_to_handle_at() requires that the 989 * filesystem also supports decoding file handles. 990 */ 991 if (!path->dentry->d_sb->s_export_op || 992 !path->dentry->d_sb->s_export_op->fh_to_dentry) 993 return -EOPNOTSUPP; 994 995 return 0; 996 } 997 998 static int fanotify_events_supported(struct path *path, __u64 mask) 999 { 1000 /* 1001 * Some filesystems such as 'proc' acquire unusual locks when opening 1002 * files. For them fanotify permission events have high chances of 1003 * deadlocking the system - open done when reporting fanotify event 1004 * blocks on this "unusual" lock while another process holding the lock 1005 * waits for fanotify permission event to be answered. Just disallow 1006 * permission events for such filesystems. 1007 */ 1008 if (mask & FANOTIFY_PERM_EVENTS && 1009 path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM) 1010 return -EINVAL; 1011 return 0; 1012 } 1013 1014 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask, 1015 int dfd, const char __user *pathname) 1016 { 1017 struct inode *inode = NULL; 1018 struct vfsmount *mnt = NULL; 1019 struct fsnotify_group *group; 1020 struct fd f; 1021 struct path path; 1022 __kernel_fsid_t __fsid, *fsid = NULL; 1023 u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS; 1024 unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS; 1025 unsigned int obj_type; 1026 int ret; 1027 1028 pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n", 1029 __func__, fanotify_fd, flags, dfd, pathname, mask); 1030 1031 /* we only use the lower 32 bits as of right now. */ 1032 if (mask & ((__u64)0xffffffff << 32)) 1033 return -EINVAL; 1034 1035 if (flags & ~FANOTIFY_MARK_FLAGS) 1036 return -EINVAL; 1037 1038 switch (mark_type) { 1039 case FAN_MARK_INODE: 1040 obj_type = FSNOTIFY_OBJ_TYPE_INODE; 1041 break; 1042 case FAN_MARK_MOUNT: 1043 obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT; 1044 break; 1045 case FAN_MARK_FILESYSTEM: 1046 obj_type = FSNOTIFY_OBJ_TYPE_SB; 1047 break; 1048 default: 1049 return -EINVAL; 1050 } 1051 1052 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { 1053 case FAN_MARK_ADD: /* fallthrough */ 1054 case FAN_MARK_REMOVE: 1055 if (!mask) 1056 return -EINVAL; 1057 break; 1058 case FAN_MARK_FLUSH: 1059 if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH)) 1060 return -EINVAL; 1061 break; 1062 default: 1063 return -EINVAL; 1064 } 1065 1066 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) 1067 valid_mask |= FANOTIFY_PERM_EVENTS; 1068 1069 if (mask & ~valid_mask) 1070 return -EINVAL; 1071 1072 f = fdget(fanotify_fd); 1073 if (unlikely(!f.file)) 1074 return -EBADF; 1075 1076 /* verify that this is indeed an fanotify instance */ 1077 ret = -EINVAL; 1078 if (unlikely(f.file->f_op != &fanotify_fops)) 1079 goto fput_and_out; 1080 group = f.file->private_data; 1081 1082 /* 1083 * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not 1084 * allowed to set permissions events. 1085 */ 1086 ret = -EINVAL; 1087 if (mask & FANOTIFY_PERM_EVENTS && 1088 group->priority == FS_PRIO_0) 1089 goto fput_and_out; 1090 1091 /* 1092 * Events with data type inode do not carry enough information to report 1093 * event->fd, so we do not allow setting a mask for inode events unless 1094 * group supports reporting fid. 1095 * inode events are not supported on a mount mark, because they do not 1096 * carry enough information (i.e. path) to be filtered by mount point. 1097 */ 1098 if (mask & FANOTIFY_INODE_EVENTS && 1099 (!FAN_GROUP_FLAG(group, FAN_REPORT_FID) || 1100 mark_type == FAN_MARK_MOUNT)) 1101 goto fput_and_out; 1102 1103 if (flags & FAN_MARK_FLUSH) { 1104 ret = 0; 1105 if (mark_type == FAN_MARK_MOUNT) 1106 fsnotify_clear_vfsmount_marks_by_group(group); 1107 else if (mark_type == FAN_MARK_FILESYSTEM) 1108 fsnotify_clear_sb_marks_by_group(group); 1109 else 1110 fsnotify_clear_inode_marks_by_group(group); 1111 goto fput_and_out; 1112 } 1113 1114 ret = fanotify_find_path(dfd, pathname, &path, flags, 1115 (mask & ALL_FSNOTIFY_EVENTS), obj_type); 1116 if (ret) 1117 goto fput_and_out; 1118 1119 if (flags & FAN_MARK_ADD) { 1120 ret = fanotify_events_supported(&path, mask); 1121 if (ret) 1122 goto path_put_and_out; 1123 } 1124 1125 if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) { 1126 ret = fanotify_test_fid(&path, &__fsid); 1127 if (ret) 1128 goto path_put_and_out; 1129 1130 fsid = &__fsid; 1131 } 1132 1133 /* inode held in place by reference to path; group by fget on fd */ 1134 if (mark_type == FAN_MARK_INODE) 1135 inode = path.dentry->d_inode; 1136 else 1137 mnt = path.mnt; 1138 1139 /* create/update an inode mark */ 1140 switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) { 1141 case FAN_MARK_ADD: 1142 if (mark_type == FAN_MARK_MOUNT) 1143 ret = fanotify_add_vfsmount_mark(group, mnt, mask, 1144 flags, fsid); 1145 else if (mark_type == FAN_MARK_FILESYSTEM) 1146 ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask, 1147 flags, fsid); 1148 else 1149 ret = fanotify_add_inode_mark(group, inode, mask, 1150 flags, fsid); 1151 break; 1152 case FAN_MARK_REMOVE: 1153 if (mark_type == FAN_MARK_MOUNT) 1154 ret = fanotify_remove_vfsmount_mark(group, mnt, mask, 1155 flags); 1156 else if (mark_type == FAN_MARK_FILESYSTEM) 1157 ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask, 1158 flags); 1159 else 1160 ret = fanotify_remove_inode_mark(group, inode, mask, 1161 flags); 1162 break; 1163 default: 1164 ret = -EINVAL; 1165 } 1166 1167 path_put_and_out: 1168 path_put(&path); 1169 fput_and_out: 1170 fdput(f); 1171 return ret; 1172 } 1173 1174 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags, 1175 __u64, mask, int, dfd, 1176 const char __user *, pathname) 1177 { 1178 return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname); 1179 } 1180 1181 #ifdef CONFIG_COMPAT 1182 COMPAT_SYSCALL_DEFINE6(fanotify_mark, 1183 int, fanotify_fd, unsigned int, flags, 1184 __u32, mask0, __u32, mask1, int, dfd, 1185 const char __user *, pathname) 1186 { 1187 return do_fanotify_mark(fanotify_fd, flags, 1188 #ifdef __BIG_ENDIAN 1189 ((__u64)mask0 << 32) | mask1, 1190 #else 1191 ((__u64)mask1 << 32) | mask0, 1192 #endif 1193 dfd, pathname); 1194 } 1195 #endif 1196 1197 /* 1198 * fanotify_user_setup - Our initialization function. Note that we cannot return 1199 * error because we have compiled-in VFS hooks. So an (unlikely) failure here 1200 * must result in panic(). 1201 */ 1202 static int __init fanotify_user_setup(void) 1203 { 1204 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 8); 1205 BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9); 1206 1207 fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, 1208 SLAB_PANIC|SLAB_ACCOUNT); 1209 fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event, 1210 SLAB_PANIC); 1211 fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event, 1212 SLAB_PANIC); 1213 if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) { 1214 fanotify_perm_event_cachep = 1215 KMEM_CACHE(fanotify_perm_event, SLAB_PANIC); 1216 } 1217 1218 return 0; 1219 } 1220 device_initcall(fanotify_user_setup); 1221