307 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 + 308 maxproc * 2 + maxfiles); 309 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 310#ifdef INVARIANTS 311 vmspace_zdtor, 312#else 313 NULL, 314#endif 315 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 316} 317 318static void 319vmspace_container_reset(struct proc *p) 320{ 321 322#ifdef RACCT 323 PROC_LOCK(p); 324 racct_set(p, RACCT_DATA, 0); 325 racct_set(p, RACCT_STACK, 0); 326 racct_set(p, RACCT_RSS, 0); 327 racct_set(p, RACCT_MEMLOCK, 0); 328 racct_set(p, RACCT_VMEM, 0); 329 PROC_UNLOCK(p); 330#endif 331} 332 333static inline void 334vmspace_dofree(struct vmspace *vm) 335{ 336 337 CTR1(KTR_VM, "vmspace_free: %p", vm); 338 339 /* 340 * Make sure any SysV shm is freed, it might not have been in 341 * exit1(). 342 */ 343 shmexit(vm); 344 345 /* 346 * Lock the map, to wait out all other references to it. 347 * Delete all of the mappings and pages they hold, then call 348 * the pmap module to reclaim anything left. 349 */ 350 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 351 vm->vm_map.max_offset); 352 353 pmap_release(vmspace_pmap(vm)); 354 vm->vm_map.pmap = NULL; 355 uma_zfree(vmspace_zone, vm); 356} 357 358void 359vmspace_free(struct vmspace *vm) 360{ 361 362 if (vm->vm_refcnt == 0) 363 panic("vmspace_free: attempt to free already freed vmspace"); 364 365 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 366 vmspace_dofree(vm); 367} 368 369void 370vmspace_exitfree(struct proc *p) 371{ 372 struct vmspace *vm; 373 374 PROC_VMSPACE_LOCK(p); 375 vm = p->p_vmspace; 376 p->p_vmspace = NULL; 377 PROC_VMSPACE_UNLOCK(p); 378 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 379 vmspace_free(vm); 380} 381 382void 383vmspace_exit(struct thread *td) 384{ 385 int refcnt; 386 struct vmspace *vm; 387 struct proc *p; 388 389 /* 390 * Release user portion of address space. 391 * This releases references to vnodes, 392 * which could cause I/O if the file has been unlinked. 393 * Need to do this early enough that we can still sleep. 394 * 395 * The last exiting process to reach this point releases as 396 * much of the environment as it can. vmspace_dofree() is the 397 * slower fallback in case another process had a temporary 398 * reference to the vmspace. 399 */ 400 401 p = td->td_proc; 402 vm = p->p_vmspace; 403 atomic_add_int(&vmspace0.vm_refcnt, 1); 404 do { 405 refcnt = vm->vm_refcnt; 406 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 407 /* Switch now since other proc might free vmspace */ 408 PROC_VMSPACE_LOCK(p); 409 p->p_vmspace = &vmspace0; 410 PROC_VMSPACE_UNLOCK(p); 411 pmap_activate(td); 412 } 413 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 414 if (refcnt == 1) { 415 if (p->p_vmspace != vm) { 416 /* vmspace not yet freed, switch back */ 417 PROC_VMSPACE_LOCK(p); 418 p->p_vmspace = vm; 419 PROC_VMSPACE_UNLOCK(p); 420 pmap_activate(td); 421 } 422 pmap_remove_pages(vmspace_pmap(vm)); 423 /* Switch now since this proc will free vmspace */ 424 PROC_VMSPACE_LOCK(p); 425 p->p_vmspace = &vmspace0; 426 PROC_VMSPACE_UNLOCK(p); 427 pmap_activate(td); 428 vmspace_dofree(vm); 429 } 430 vmspace_container_reset(p); 431} 432 433/* Acquire reference to vmspace owned by another process. */ 434 435struct vmspace * 436vmspace_acquire_ref(struct proc *p) 437{ 438 struct vmspace *vm; 439 int refcnt; 440 441 PROC_VMSPACE_LOCK(p); 442 vm = p->p_vmspace; 443 if (vm == NULL) { 444 PROC_VMSPACE_UNLOCK(p); 445 return (NULL); 446 } 447 do { 448 refcnt = vm->vm_refcnt; 449 if (refcnt <= 0) { /* Avoid 0->1 transition */ 450 PROC_VMSPACE_UNLOCK(p); 451 return (NULL); 452 } 453 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 454 if (vm != p->p_vmspace) { 455 PROC_VMSPACE_UNLOCK(p); 456 vmspace_free(vm); 457 return (NULL); 458 } 459 PROC_VMSPACE_UNLOCK(p); 460 return (vm); 461} 462 463void 464_vm_map_lock(vm_map_t map, const char *file, int line) 465{ 466 467 if (map->system_map) 468 mtx_lock_flags_(&map->system_mtx, 0, file, line); 469 else 470 sx_xlock_(&map->lock, file, line); 471 map->timestamp++; 472} 473 474static void 475vm_map_process_deferred(void) 476{ 477 struct thread *td; 478 vm_map_entry_t entry, next; 479 vm_object_t object; 480 481 td = curthread; 482 entry = td->td_map_def_user; 483 td->td_map_def_user = NULL; 484 while (entry != NULL) { 485 next = entry->next; 486 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 487 /* 488 * Decrement the object's writemappings and 489 * possibly the vnode's v_writecount. 490 */ 491 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 492 ("Submap with writecount")); 493 object = entry->object.vm_object; 494 KASSERT(object != NULL, ("No object for writecount")); 495 vnode_pager_release_writecount(object, entry->start, 496 entry->end); 497 } 498 vm_map_entry_deallocate(entry, FALSE); 499 entry = next; 500 } 501} 502 503void 504_vm_map_unlock(vm_map_t map, const char *file, int line) 505{ 506 507 if (map->system_map) 508 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 509 else { 510 sx_xunlock_(&map->lock, file, line); 511 vm_map_process_deferred(); 512 } 513} 514 515void 516_vm_map_lock_read(vm_map_t map, const char *file, int line) 517{ 518 519 if (map->system_map) 520 mtx_lock_flags_(&map->system_mtx, 0, file, line); 521 else 522 sx_slock_(&map->lock, file, line); 523} 524 525void 526_vm_map_unlock_read(vm_map_t map, const char *file, int line) 527{ 528 529 if (map->system_map) 530 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 531 else { 532 sx_sunlock_(&map->lock, file, line); 533 vm_map_process_deferred(); 534 } 535} 536 537int 538_vm_map_trylock(vm_map_t map, const char *file, int line) 539{ 540 int error; 541 542 error = map->system_map ? 543 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 544 !sx_try_xlock_(&map->lock, file, line); 545 if (error == 0) 546 map->timestamp++; 547 return (error == 0); 548} 549 550int 551_vm_map_trylock_read(vm_map_t map, const char *file, int line) 552{ 553 int error; 554 555 error = map->system_map ? 556 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 557 !sx_try_slock_(&map->lock, file, line); 558 return (error == 0); 559} 560 561/* 562 * _vm_map_lock_upgrade: [ internal use only ] 563 * 564 * Tries to upgrade a read (shared) lock on the specified map to a write 565 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 566 * non-zero value if the upgrade fails. If the upgrade fails, the map is 567 * returned without a read or write lock held. 568 * 569 * Requires that the map be read locked. 570 */ 571int 572_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 573{ 574 unsigned int last_timestamp; 575 576 if (map->system_map) { 577 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 578 } else { 579 if (!sx_try_upgrade_(&map->lock, file, line)) { 580 last_timestamp = map->timestamp; 581 sx_sunlock_(&map->lock, file, line); 582 vm_map_process_deferred(); 583 /* 584 * If the map's timestamp does not change while the 585 * map is unlocked, then the upgrade succeeds. 586 */ 587 sx_xlock_(&map->lock, file, line); 588 if (last_timestamp != map->timestamp) { 589 sx_xunlock_(&map->lock, file, line); 590 return (1); 591 } 592 } 593 } 594 map->timestamp++; 595 return (0); 596} 597 598void 599_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 600{ 601 602 if (map->system_map) { 603 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 604 } else 605 sx_downgrade_(&map->lock, file, line); 606} 607 608/* 609 * vm_map_locked: 610 * 611 * Returns a non-zero value if the caller holds a write (exclusive) lock 612 * on the specified map and the value "0" otherwise. 613 */ 614int 615vm_map_locked(vm_map_t map) 616{ 617 618 if (map->system_map) 619 return (mtx_owned(&map->system_mtx)); 620 else 621 return (sx_xlocked(&map->lock)); 622} 623 624#ifdef INVARIANTS 625static void 626_vm_map_assert_locked(vm_map_t map, const char *file, int line) 627{ 628 629 if (map->system_map) 630 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 631 else 632 sx_assert_(&map->lock, SA_XLOCKED, file, line); 633} 634 635#define VM_MAP_ASSERT_LOCKED(map) \ 636 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 637#else 638#define VM_MAP_ASSERT_LOCKED(map) 639#endif 640 641/* 642 * _vm_map_unlock_and_wait: 643 * 644 * Atomically releases the lock on the specified map and puts the calling 645 * thread to sleep. The calling thread will remain asleep until either 646 * vm_map_wakeup() is performed on the map or the specified timeout is 647 * exceeded. 648 * 649 * WARNING! This function does not perform deferred deallocations of 650 * objects and map entries. Therefore, the calling thread is expected to 651 * reacquire the map lock after reawakening and later perform an ordinary 652 * unlock operation, such as vm_map_unlock(), before completing its 653 * operation on the map. 654 */ 655int 656_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 657{ 658 659 mtx_lock(&map_sleep_mtx); 660 if (map->system_map) 661 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 662 else 663 sx_xunlock_(&map->lock, file, line); 664 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 665 timo)); 666} 667 668/* 669 * vm_map_wakeup: 670 * 671 * Awaken any threads that have slept on the map using 672 * vm_map_unlock_and_wait(). 673 */ 674void 675vm_map_wakeup(vm_map_t map) 676{ 677 678 /* 679 * Acquire and release map_sleep_mtx to prevent a wakeup() 680 * from being performed (and lost) between the map unlock 681 * and the msleep() in _vm_map_unlock_and_wait(). 682 */ 683 mtx_lock(&map_sleep_mtx); 684 mtx_unlock(&map_sleep_mtx); 685 wakeup(&map->root); 686} 687 688void 689vm_map_busy(vm_map_t map) 690{ 691 692 VM_MAP_ASSERT_LOCKED(map); 693 map->busy++; 694} 695 696void 697vm_map_unbusy(vm_map_t map) 698{ 699 700 VM_MAP_ASSERT_LOCKED(map); 701 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 702 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 703 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 704 wakeup(&map->busy); 705 } 706} 707 708void 709vm_map_wait_busy(vm_map_t map) 710{ 711 712 VM_MAP_ASSERT_LOCKED(map); 713 while (map->busy) { 714 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 715 if (map->system_map) 716 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 717 else 718 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 719 } 720 map->timestamp++; 721} 722 723long 724vmspace_resident_count(struct vmspace *vmspace) 725{ 726 return pmap_resident_count(vmspace_pmap(vmspace)); 727} 728 729/* 730 * vm_map_create: 731 * 732 * Creates and returns a new empty VM map with 733 * the given physical map structure, and having 734 * the given lower and upper address bounds. 735 */ 736vm_map_t 737vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 738{ 739 vm_map_t result; 740 741 result = uma_zalloc(mapzone, M_WAITOK); 742 CTR1(KTR_VM, "vm_map_create: %p", result); 743 _vm_map_init(result, pmap, min, max); 744 return (result); 745} 746 747/* 748 * Initialize an existing vm_map structure 749 * such as that in the vmspace structure. 750 */ 751static void 752_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 753{ 754 755 map->header.next = map->header.prev = &map->header; 756 map->needs_wakeup = FALSE; 757 map->system_map = 0; 758 map->pmap = pmap; 759 map->min_offset = min; 760 map->max_offset = max; 761 map->flags = 0; 762 map->root = NULL; 763 map->timestamp = 0; 764 map->busy = 0; 765} 766 767void 768vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 769{ 770 771 _vm_map_init(map, pmap, min, max); 772 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 773 sx_init(&map->lock, "user map"); 774} 775 776/* 777 * vm_map_entry_dispose: [ internal use only ] 778 * 779 * Inverse of vm_map_entry_create. 780 */ 781static void 782vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 783{ 784 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 785} 786 787/* 788 * vm_map_entry_create: [ internal use only ] 789 * 790 * Allocates a VM map entry for insertion. 791 * No entry fields are filled in. 792 */ 793static vm_map_entry_t 794vm_map_entry_create(vm_map_t map) 795{ 796 vm_map_entry_t new_entry; 797 798 if (map->system_map) 799 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 800 else 801 new_entry = uma_zalloc(mapentzone, M_WAITOK); 802 if (new_entry == NULL) 803 panic("vm_map_entry_create: kernel resources exhausted"); 804 return (new_entry); 805} 806 807/* 808 * vm_map_entry_set_behavior: 809 * 810 * Set the expected access behavior, either normal, random, or 811 * sequential. 812 */ 813static inline void 814vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 815{ 816 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 817 (behavior & MAP_ENTRY_BEHAV_MASK); 818} 819 820/* 821 * vm_map_entry_set_max_free: 822 * 823 * Set the max_free field in a vm_map_entry. 824 */ 825static inline void 826vm_map_entry_set_max_free(vm_map_entry_t entry) 827{ 828 829 entry->max_free = entry->adj_free; 830 if (entry->left != NULL && entry->left->max_free > entry->max_free) 831 entry->max_free = entry->left->max_free; 832 if (entry->right != NULL && entry->right->max_free > entry->max_free) 833 entry->max_free = entry->right->max_free; 834} 835 836/* 837 * vm_map_entry_splay: 838 * 839 * The Sleator and Tarjan top-down splay algorithm with the 840 * following variation. Max_free must be computed bottom-up, so 841 * on the downward pass, maintain the left and right spines in 842 * reverse order. Then, make a second pass up each side to fix 843 * the pointers and compute max_free. The time bound is O(log n) 844 * amortized. 845 * 846 * The new root is the vm_map_entry containing "addr", or else an 847 * adjacent entry (lower or higher) if addr is not in the tree. 848 * 849 * The map must be locked, and leaves it so. 850 * 851 * Returns: the new root. 852 */ 853static vm_map_entry_t 854vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 855{ 856 vm_map_entry_t llist, rlist; 857 vm_map_entry_t ltree, rtree; 858 vm_map_entry_t y; 859 860 /* Special case of empty tree. */ 861 if (root == NULL) 862 return (root); 863 864 /* 865 * Pass One: Splay down the tree until we find addr or a NULL 866 * pointer where addr would go. llist and rlist are the two 867 * sides in reverse order (bottom-up), with llist linked by 868 * the right pointer and rlist linked by the left pointer in 869 * the vm_map_entry. Wait until Pass Two to set max_free on 870 * the two spines. 871 */ 872 llist = NULL; 873 rlist = NULL; 874 for (;;) { 875 /* root is never NULL in here. */ 876 if (addr < root->start) { 877 y = root->left; 878 if (y == NULL) 879 break; 880 if (addr < y->start && y->left != NULL) { 881 /* Rotate right and put y on rlist. */ 882 root->left = y->right; 883 y->right = root; 884 vm_map_entry_set_max_free(root); 885 root = y->left; 886 y->left = rlist; 887 rlist = y; 888 } else { 889 /* Put root on rlist. */ 890 root->left = rlist; 891 rlist = root; 892 root = y; 893 } 894 } else if (addr >= root->end) { 895 y = root->right; 896 if (y == NULL) 897 break; 898 if (addr >= y->end && y->right != NULL) { 899 /* Rotate left and put y on llist. */ 900 root->right = y->left; 901 y->left = root; 902 vm_map_entry_set_max_free(root); 903 root = y->right; 904 y->right = llist; 905 llist = y; 906 } else { 907 /* Put root on llist. */ 908 root->right = llist; 909 llist = root; 910 root = y; 911 } 912 } else 913 break; 914 } 915 916 /* 917 * Pass Two: Walk back up the two spines, flip the pointers 918 * and set max_free. The subtrees of the root go at the 919 * bottom of llist and rlist. 920 */ 921 ltree = root->left; 922 while (llist != NULL) { 923 y = llist->right; 924 llist->right = ltree; 925 vm_map_entry_set_max_free(llist); 926 ltree = llist; 927 llist = y; 928 } 929 rtree = root->right; 930 while (rlist != NULL) { 931 y = rlist->left; 932 rlist->left = rtree; 933 vm_map_entry_set_max_free(rlist); 934 rtree = rlist; 935 rlist = y; 936 } 937 938 /* 939 * Final assembly: add ltree and rtree as subtrees of root. 940 */ 941 root->left = ltree; 942 root->right = rtree; 943 vm_map_entry_set_max_free(root); 944 945 return (root); 946} 947 948/* 949 * vm_map_entry_{un,}link: 950 * 951 * Insert/remove entries from maps. 952 */ 953static void 954vm_map_entry_link(vm_map_t map, 955 vm_map_entry_t after_where, 956 vm_map_entry_t entry) 957{ 958 959 CTR4(KTR_VM, 960 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 961 map->nentries, entry, after_where); 962 VM_MAP_ASSERT_LOCKED(map); 963 map->nentries++; 964 entry->prev = after_where; 965 entry->next = after_where->next; 966 entry->next->prev = entry; 967 after_where->next = entry; 968 969 if (after_where != &map->header) { 970 if (after_where != map->root) 971 vm_map_entry_splay(after_where->start, map->root); 972 entry->right = after_where->right; 973 entry->left = after_where; 974 after_where->right = NULL; 975 after_where->adj_free = entry->start - after_where->end; 976 vm_map_entry_set_max_free(after_where); 977 } else { 978 entry->right = map->root; 979 entry->left = NULL; 980 } 981 entry->adj_free = (entry->next == &map->header ? map->max_offset : 982 entry->next->start) - entry->end; 983 vm_map_entry_set_max_free(entry); 984 map->root = entry; 985} 986 987static void 988vm_map_entry_unlink(vm_map_t map, 989 vm_map_entry_t entry) 990{ 991 vm_map_entry_t next, prev, root; 992 993 VM_MAP_ASSERT_LOCKED(map); 994 if (entry != map->root) 995 vm_map_entry_splay(entry->start, map->root); 996 if (entry->left == NULL) 997 root = entry->right; 998 else { 999 root = vm_map_entry_splay(entry->start, entry->left); 1000 root->right = entry->right; 1001 root->adj_free = (entry->next == &map->header ? map->max_offset : 1002 entry->next->start) - root->end; 1003 vm_map_entry_set_max_free(root); 1004 } 1005 map->root = root; 1006 1007 prev = entry->prev; 1008 next = entry->next; 1009 next->prev = prev; 1010 prev->next = next; 1011 map->nentries--; 1012 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1013 map->nentries, entry); 1014} 1015 1016/* 1017 * vm_map_entry_resize_free: 1018 * 1019 * Recompute the amount of free space following a vm_map_entry 1020 * and propagate that value up the tree. Call this function after 1021 * resizing a map entry in-place, that is, without a call to 1022 * vm_map_entry_link() or _unlink(). 1023 * 1024 * The map must be locked, and leaves it so. 1025 */ 1026static void 1027vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1028{ 1029 1030 /* 1031 * Using splay trees without parent pointers, propagating 1032 * max_free up the tree is done by moving the entry to the 1033 * root and making the change there. 1034 */ 1035 if (entry != map->root) 1036 map->root = vm_map_entry_splay(entry->start, map->root); 1037 1038 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1039 entry->next->start) - entry->end; 1040 vm_map_entry_set_max_free(entry); 1041} 1042 1043/* 1044 * vm_map_lookup_entry: [ internal use only ] 1045 * 1046 * Finds the map entry containing (or 1047 * immediately preceding) the specified address 1048 * in the given map; the entry is returned 1049 * in the "entry" parameter. The boolean 1050 * result indicates whether the address is 1051 * actually contained in the map. 1052 */ 1053boolean_t 1054vm_map_lookup_entry( 1055 vm_map_t map, 1056 vm_offset_t address, 1057 vm_map_entry_t *entry) /* OUT */ 1058{ 1059 vm_map_entry_t cur; 1060 boolean_t locked; 1061 1062 /* 1063 * If the map is empty, then the map entry immediately preceding 1064 * "address" is the map's header. 1065 */ 1066 cur = map->root; 1067 if (cur == NULL) 1068 *entry = &map->header; 1069 else if (address >= cur->start && cur->end > address) { 1070 *entry = cur; 1071 return (TRUE); 1072 } else if ((locked = vm_map_locked(map)) || 1073 sx_try_upgrade(&map->lock)) { 1074 /* 1075 * Splay requires a write lock on the map. However, it only 1076 * restructures the binary search tree; it does not otherwise 1077 * change the map. Thus, the map's timestamp need not change 1078 * on a temporary upgrade. 1079 */ 1080 map->root = cur = vm_map_entry_splay(address, cur); 1081 if (!locked) 1082 sx_downgrade(&map->lock); 1083 1084 /* 1085 * If "address" is contained within a map entry, the new root 1086 * is that map entry. Otherwise, the new root is a map entry 1087 * immediately before or after "address". 1088 */ 1089 if (address >= cur->start) { 1090 *entry = cur; 1091 if (cur->end > address) 1092 return (TRUE); 1093 } else 1094 *entry = cur->prev; 1095 } else 1096 /* 1097 * Since the map is only locked for read access, perform a 1098 * standard binary search tree lookup for "address". 1099 */ 1100 for (;;) { 1101 if (address < cur->start) { 1102 if (cur->left == NULL) { 1103 *entry = cur->prev; 1104 break; 1105 } 1106 cur = cur->left; 1107 } else if (cur->end > address) { 1108 *entry = cur; 1109 return (TRUE); 1110 } else { 1111 if (cur->right == NULL) { 1112 *entry = cur; 1113 break; 1114 } 1115 cur = cur->right; 1116 } 1117 } 1118 return (FALSE); 1119} 1120 1121/* 1122 * vm_map_insert: 1123 * 1124 * Inserts the given whole VM object into the target 1125 * map at the specified address range. The object's 1126 * size should match that of the address range. 1127 * 1128 * Requires that the map be locked, and leaves it so. 1129 * 1130 * If object is non-NULL, ref count must be bumped by caller 1131 * prior to making call to account for the new entry. 1132 */ 1133int 1134vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1135 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1136 int cow) 1137{ 1138 vm_map_entry_t new_entry; 1139 vm_map_entry_t prev_entry; 1140 vm_map_entry_t temp_entry; 1141 vm_eflags_t protoeflags; 1142 struct ucred *cred; 1143 vm_inherit_t inheritance; 1144 boolean_t charge_prev_obj; 1145 1146 VM_MAP_ASSERT_LOCKED(map); 1147 1148 /* 1149 * Check that the start and end points are not bogus. 1150 */ 1151 if ((start < map->min_offset) || (end > map->max_offset) || 1152 (start >= end)) 1153 return (KERN_INVALID_ADDRESS); 1154 1155 /* 1156 * Find the entry prior to the proposed starting address; if it's part 1157 * of an existing entry, this range is bogus. 1158 */ 1159 if (vm_map_lookup_entry(map, start, &temp_entry)) 1160 return (KERN_NO_SPACE); 1161 1162 prev_entry = temp_entry; 1163 1164 /* 1165 * Assert that the next entry doesn't overlap the end point. 1166 */ 1167 if ((prev_entry->next != &map->header) && 1168 (prev_entry->next->start < end)) 1169 return (KERN_NO_SPACE); 1170 1171 protoeflags = 0; 1172 charge_prev_obj = FALSE; 1173 1174 if (cow & MAP_COPY_ON_WRITE) 1175 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1176 1177 if (cow & MAP_NOFAULT) { 1178 protoeflags |= MAP_ENTRY_NOFAULT; 1179 1180 KASSERT(object == NULL, 1181 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1182 } 1183 if (cow & MAP_DISABLE_SYNCER) 1184 protoeflags |= MAP_ENTRY_NOSYNC; 1185 if (cow & MAP_DISABLE_COREDUMP) 1186 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1187 if (cow & MAP_VN_WRITECOUNT) 1188 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1189 if (cow & MAP_INHERIT_SHARE) 1190 inheritance = VM_INHERIT_SHARE; 1191 else 1192 inheritance = VM_INHERIT_DEFAULT; 1193 1194 cred = NULL; 1195 KASSERT((object != kmem_object && object != kernel_object) || 1196 ((object == kmem_object || object == kernel_object) && 1197 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1198 ("kmem or kernel object and cow")); 1199 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1200 goto charged; 1201 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1202 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1203 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1204 return (KERN_RESOURCE_SHORTAGE); 1205 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1206 object->cred == NULL, 1207 ("OVERCOMMIT: vm_map_insert o %p", object)); 1208 cred = curthread->td_ucred; 1209 crhold(cred); 1210 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1211 charge_prev_obj = TRUE; 1212 } 1213 1214charged: 1215 /* Expand the kernel pmap, if necessary. */ 1216 if (map == kernel_map && end > kernel_vm_end) 1217 pmap_growkernel(end); 1218 if (object != NULL) { 1219 /* 1220 * OBJ_ONEMAPPING must be cleared unless this mapping 1221 * is trivially proven to be the only mapping for any 1222 * of the object's pages. (Object granularity 1223 * reference counting is insufficient to recognize 1224 * aliases with precision.) 1225 */ 1226 VM_OBJECT_LOCK(object); 1227 if (object->ref_count > 1 || object->shadow_count != 0) 1228 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1229 VM_OBJECT_UNLOCK(object); 1230 } 1231 else if ((prev_entry != &map->header) && 1232 (prev_entry->eflags == protoeflags) && 1233 (prev_entry->end == start) && 1234 (prev_entry->wired_count == 0) && 1235 (prev_entry->cred == cred || 1236 (prev_entry->object.vm_object != NULL && 1237 (prev_entry->object.vm_object->cred == cred))) && 1238 vm_object_coalesce(prev_entry->object.vm_object, 1239 prev_entry->offset, 1240 (vm_size_t)(prev_entry->end - prev_entry->start), 1241 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1242 /* 1243 * We were able to extend the object. Determine if we 1244 * can extend the previous map entry to include the 1245 * new range as well. 1246 */ 1247 if ((prev_entry->inheritance == inheritance) && 1248 (prev_entry->protection == prot) && 1249 (prev_entry->max_protection == max)) { 1250 map->size += (end - prev_entry->end); 1251 prev_entry->end = end; 1252 vm_map_entry_resize_free(map, prev_entry); 1253 vm_map_simplify_entry(map, prev_entry); 1254 if (cred != NULL) 1255 crfree(cred); 1256 return (KERN_SUCCESS); 1257 } 1258 1259 /* 1260 * If we can extend the object but cannot extend the 1261 * map entry, we have to create a new map entry. We 1262 * must bump the ref count on the extended object to 1263 * account for it. object may be NULL. 1264 */ 1265 object = prev_entry->object.vm_object; 1266 offset = prev_entry->offset + 1267 (prev_entry->end - prev_entry->start); 1268 vm_object_reference(object); 1269 if (cred != NULL && object != NULL && object->cred != NULL && 1270 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1271 /* Object already accounts for this uid. */ 1272 crfree(cred); 1273 cred = NULL; 1274 } 1275 } 1276 1277 /* 1278 * NOTE: if conditionals fail, object can be NULL here. This occurs 1279 * in things like the buffer map where we manage kva but do not manage 1280 * backing objects. 1281 */ 1282 1283 /* 1284 * Create a new entry 1285 */ 1286 new_entry = vm_map_entry_create(map); 1287 new_entry->start = start; 1288 new_entry->end = end; 1289 new_entry->cred = NULL; 1290 1291 new_entry->eflags = protoeflags; 1292 new_entry->object.vm_object = object; 1293 new_entry->offset = offset; 1294 new_entry->avail_ssize = 0; 1295 1296 new_entry->inheritance = inheritance; 1297 new_entry->protection = prot; 1298 new_entry->max_protection = max; 1299 new_entry->wired_count = 0; 1300 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1301 new_entry->next_read = OFF_TO_IDX(offset); 1302 1303 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1304 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1305 new_entry->cred = cred; 1306 1307 /* 1308 * Insert the new entry into the list 1309 */ 1310 vm_map_entry_link(map, prev_entry, new_entry); 1311 map->size += new_entry->end - new_entry->start; 1312 1313 /* 1314 * It may be possible to merge the new entry with the next and/or 1315 * previous entries. However, due to MAP_STACK_* being a hack, a 1316 * panic can result from merging such entries. 1317 */ 1318 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1319 vm_map_simplify_entry(map, new_entry); 1320 1321 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1322 vm_map_pmap_enter(map, start, prot, 1323 object, OFF_TO_IDX(offset), end - start, 1324 cow & MAP_PREFAULT_PARTIAL); 1325 } 1326 1327 return (KERN_SUCCESS); 1328} 1329 1330/* 1331 * vm_map_findspace: 1332 * 1333 * Find the first fit (lowest VM address) for "length" free bytes 1334 * beginning at address >= start in the given map. 1335 * 1336 * In a vm_map_entry, "adj_free" is the amount of free space 1337 * adjacent (higher address) to this entry, and "max_free" is the 1338 * maximum amount of contiguous free space in its subtree. This 1339 * allows finding a free region in one path down the tree, so 1340 * O(log n) amortized with splay trees. 1341 * 1342 * The map must be locked, and leaves it so. 1343 * 1344 * Returns: 0 on success, and starting address in *addr, 1345 * 1 if insufficient space. 1346 */ 1347int 1348vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1349 vm_offset_t *addr) /* OUT */ 1350{ 1351 vm_map_entry_t entry; 1352 vm_offset_t st; 1353 1354 /* 1355 * Request must fit within min/max VM address and must avoid 1356 * address wrap. 1357 */ 1358 if (start < map->min_offset) 1359 start = map->min_offset; 1360 if (start + length > map->max_offset || start + length < start) 1361 return (1); 1362 1363 /* Empty tree means wide open address space. */ 1364 if (map->root == NULL) { 1365 *addr = start; 1366 return (0); 1367 } 1368 1369 /* 1370 * After splay, if start comes before root node, then there 1371 * must be a gap from start to the root. 1372 */ 1373 map->root = vm_map_entry_splay(start, map->root); 1374 if (start + length <= map->root->start) { 1375 *addr = start; 1376 return (0); 1377 } 1378 1379 /* 1380 * Root is the last node that might begin its gap before 1381 * start, and this is the last comparison where address 1382 * wrap might be a problem. 1383 */ 1384 st = (start > map->root->end) ? start : map->root->end; 1385 if (length <= map->root->end + map->root->adj_free - st) { 1386 *addr = st; 1387 return (0); 1388 } 1389 1390 /* With max_free, can immediately tell if no solution. */ 1391 entry = map->root->right; 1392 if (entry == NULL || length > entry->max_free) 1393 return (1); 1394 1395 /* 1396 * Search the right subtree in the order: left subtree, root, 1397 * right subtree (first fit). The previous splay implies that 1398 * all regions in the right subtree have addresses > start. 1399 */ 1400 while (entry != NULL) { 1401 if (entry->left != NULL && entry->left->max_free >= length) 1402 entry = entry->left; 1403 else if (entry->adj_free >= length) { 1404 *addr = entry->end; 1405 return (0); 1406 } else 1407 entry = entry->right; 1408 } 1409 1410 /* Can't get here, so panic if we do. */ 1411 panic("vm_map_findspace: max_free corrupt"); 1412} 1413 1414int 1415vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1416 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1417 vm_prot_t max, int cow) 1418{ 1419 vm_offset_t end; 1420 int result; 1421 1422 end = start + length; 1423 vm_map_lock(map); 1424 VM_MAP_RANGE_CHECK(map, start, end); 1425 (void) vm_map_delete(map, start, end); 1426 result = vm_map_insert(map, object, offset, start, end, prot, 1427 max, cow); 1428 vm_map_unlock(map); 1429 return (result); 1430} 1431 1432/* 1433 * vm_map_find finds an unallocated region in the target address 1434 * map with the given length. The search is defined to be 1435 * first-fit from the specified address; the region found is 1436 * returned in the same parameter. 1437 * 1438 * If object is non-NULL, ref count must be bumped by caller 1439 * prior to making call to account for the new entry. 1440 */ 1441int 1442vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1443 vm_offset_t *addr, /* IN/OUT */ 1444 vm_size_t length, int find_space, vm_prot_t prot, 1445 vm_prot_t max, int cow) 1446{ 1447 vm_offset_t start; 1448 int result; 1449 1450 start = *addr; 1451 vm_map_lock(map); 1452 do { 1453 if (find_space != VMFS_NO_SPACE) { 1454 if (vm_map_findspace(map, start, length, addr)) { 1455 vm_map_unlock(map); 1456 return (KERN_NO_SPACE); 1457 } 1458 switch (find_space) { 1459 case VMFS_ALIGNED_SPACE: 1460 pmap_align_superpage(object, offset, addr, 1461 length); 1462 break; 1463#ifdef VMFS_TLB_ALIGNED_SPACE 1464 case VMFS_TLB_ALIGNED_SPACE: 1465 pmap_align_tlb(addr); 1466 break; 1467#endif 1468 default: 1469 break; 1470 } 1471 1472 start = *addr; 1473 } 1474 result = vm_map_insert(map, object, offset, start, start + 1475 length, prot, max, cow); 1476 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE 1477#ifdef VMFS_TLB_ALIGNED_SPACE 1478 || find_space == VMFS_TLB_ALIGNED_SPACE 1479#endif 1480 )); 1481 vm_map_unlock(map); 1482 return (result); 1483} 1484 1485/* 1486 * vm_map_simplify_entry: 1487 * 1488 * Simplify the given map entry by merging with either neighbor. This 1489 * routine also has the ability to merge with both neighbors. 1490 * 1491 * The map must be locked. 1492 * 1493 * This routine guarentees that the passed entry remains valid (though 1494 * possibly extended). When merging, this routine may delete one or 1495 * both neighbors. 1496 */ 1497void 1498vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1499{ 1500 vm_map_entry_t next, prev; 1501 vm_size_t prevsize, esize; 1502 1503 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1504 return; 1505 1506 prev = entry->prev; 1507 if (prev != &map->header) { 1508 prevsize = prev->end - prev->start; 1509 if ( (prev->end == entry->start) && 1510 (prev->object.vm_object == entry->object.vm_object) && 1511 (!prev->object.vm_object || 1512 (prev->offset + prevsize == entry->offset)) && 1513 (prev->eflags == entry->eflags) && 1514 (prev->protection == entry->protection) && 1515 (prev->max_protection == entry->max_protection) && 1516 (prev->inheritance == entry->inheritance) && 1517 (prev->wired_count == entry->wired_count) && 1518 (prev->cred == entry->cred)) { 1519 vm_map_entry_unlink(map, prev); 1520 entry->start = prev->start; 1521 entry->offset = prev->offset; 1522 if (entry->prev != &map->header) 1523 vm_map_entry_resize_free(map, entry->prev); 1524 1525 /* 1526 * If the backing object is a vnode object, 1527 * vm_object_deallocate() calls vrele(). 1528 * However, vrele() does not lock the vnode 1529 * because the vnode has additional 1530 * references. Thus, the map lock can be kept 1531 * without causing a lock-order reversal with 1532 * the vnode lock. 1533 * 1534 * Since we count the number of virtual page 1535 * mappings in object->un_pager.vnp.writemappings, 1536 * the writemappings value should not be adjusted 1537 * when the entry is disposed of. 1538 */ 1539 if (prev->object.vm_object) 1540 vm_object_deallocate(prev->object.vm_object); 1541 if (prev->cred != NULL) 1542 crfree(prev->cred); 1543 vm_map_entry_dispose(map, prev); 1544 } 1545 } 1546 1547 next = entry->next; 1548 if (next != &map->header) { 1549 esize = entry->end - entry->start; 1550 if ((entry->end == next->start) && 1551 (next->object.vm_object == entry->object.vm_object) && 1552 (!entry->object.vm_object || 1553 (entry->offset + esize == next->offset)) && 1554 (next->eflags == entry->eflags) && 1555 (next->protection == entry->protection) && 1556 (next->max_protection == entry->max_protection) && 1557 (next->inheritance == entry->inheritance) && 1558 (next->wired_count == entry->wired_count) && 1559 (next->cred == entry->cred)) { 1560 vm_map_entry_unlink(map, next); 1561 entry->end = next->end; 1562 vm_map_entry_resize_free(map, entry); 1563 1564 /* 1565 * See comment above. 1566 */ 1567 if (next->object.vm_object) 1568 vm_object_deallocate(next->object.vm_object); 1569 if (next->cred != NULL) 1570 crfree(next->cred); 1571 vm_map_entry_dispose(map, next); 1572 } 1573 } 1574} 1575/* 1576 * vm_map_clip_start: [ internal use only ] 1577 * 1578 * Asserts that the given entry begins at or after 1579 * the specified address; if necessary, 1580 * it splits the entry into two. 1581 */ 1582#define vm_map_clip_start(map, entry, startaddr) \ 1583{ \ 1584 if (startaddr > entry->start) \ 1585 _vm_map_clip_start(map, entry, startaddr); \ 1586} 1587 1588/* 1589 * This routine is called only when it is known that 1590 * the entry must be split. 1591 */ 1592static void 1593_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1594{ 1595 vm_map_entry_t new_entry; 1596 1597 VM_MAP_ASSERT_LOCKED(map); 1598 1599 /* 1600 * Split off the front portion -- note that we must insert the new 1601 * entry BEFORE this one, so that this entry has the specified 1602 * starting address. 1603 */ 1604 vm_map_simplify_entry(map, entry); 1605 1606 /* 1607 * If there is no object backing this entry, we might as well create 1608 * one now. If we defer it, an object can get created after the map 1609 * is clipped, and individual objects will be created for the split-up 1610 * map. This is a bit of a hack, but is also about the best place to 1611 * put this improvement. 1612 */ 1613 if (entry->object.vm_object == NULL && !map->system_map) { 1614 vm_object_t object; 1615 object = vm_object_allocate(OBJT_DEFAULT, 1616 atop(entry->end - entry->start)); 1617 entry->object.vm_object = object; 1618 entry->offset = 0; 1619 if (entry->cred != NULL) { 1620 object->cred = entry->cred; 1621 object->charge = entry->end - entry->start; 1622 entry->cred = NULL; 1623 } 1624 } else if (entry->object.vm_object != NULL && 1625 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1626 entry->cred != NULL) { 1627 VM_OBJECT_LOCK(entry->object.vm_object); 1628 KASSERT(entry->object.vm_object->cred == NULL, 1629 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1630 entry->object.vm_object->cred = entry->cred; 1631 entry->object.vm_object->charge = entry->end - entry->start; 1632 VM_OBJECT_UNLOCK(entry->object.vm_object); 1633 entry->cred = NULL; 1634 } 1635 1636 new_entry = vm_map_entry_create(map); 1637 *new_entry = *entry; 1638 1639 new_entry->end = start; 1640 entry->offset += (start - entry->start); 1641 entry->start = start; 1642 if (new_entry->cred != NULL) 1643 crhold(entry->cred); 1644 1645 vm_map_entry_link(map, entry->prev, new_entry); 1646 1647 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1648 vm_object_reference(new_entry->object.vm_object); 1649 /* 1650 * The object->un_pager.vnp.writemappings for the 1651 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1652 * kept as is here. The virtual pages are 1653 * re-distributed among the clipped entries, so the sum is 1654 * left the same. 1655 */ 1656 } 1657} 1658 1659/* 1660 * vm_map_clip_end: [ internal use only ] 1661 * 1662 * Asserts that the given entry ends at or before 1663 * the specified address; if necessary, 1664 * it splits the entry into two. 1665 */ 1666#define vm_map_clip_end(map, entry, endaddr) \ 1667{ \ 1668 if ((endaddr) < (entry->end)) \ 1669 _vm_map_clip_end((map), (entry), (endaddr)); \ 1670} 1671 1672/* 1673 * This routine is called only when it is known that 1674 * the entry must be split. 1675 */ 1676static void 1677_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1678{ 1679 vm_map_entry_t new_entry; 1680 1681 VM_MAP_ASSERT_LOCKED(map); 1682 1683 /* 1684 * If there is no object backing this entry, we might as well create 1685 * one now. If we defer it, an object can get created after the map 1686 * is clipped, and individual objects will be created for the split-up 1687 * map. This is a bit of a hack, but is also about the best place to 1688 * put this improvement. 1689 */ 1690 if (entry->object.vm_object == NULL && !map->system_map) { 1691 vm_object_t object; 1692 object = vm_object_allocate(OBJT_DEFAULT, 1693 atop(entry->end - entry->start)); 1694 entry->object.vm_object = object; 1695 entry->offset = 0; 1696 if (entry->cred != NULL) { 1697 object->cred = entry->cred; 1698 object->charge = entry->end - entry->start; 1699 entry->cred = NULL; 1700 } 1701 } else if (entry->object.vm_object != NULL && 1702 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1703 entry->cred != NULL) { 1704 VM_OBJECT_LOCK(entry->object.vm_object); 1705 KASSERT(entry->object.vm_object->cred == NULL, 1706 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1707 entry->object.vm_object->cred = entry->cred; 1708 entry->object.vm_object->charge = entry->end - entry->start; 1709 VM_OBJECT_UNLOCK(entry->object.vm_object); 1710 entry->cred = NULL; 1711 } 1712 1713 /* 1714 * Create a new entry and insert it AFTER the specified entry 1715 */ 1716 new_entry = vm_map_entry_create(map); 1717 *new_entry = *entry; 1718 1719 new_entry->start = entry->end = end; 1720 new_entry->offset += (end - entry->start); 1721 if (new_entry->cred != NULL) 1722 crhold(entry->cred); 1723 1724 vm_map_entry_link(map, entry, new_entry); 1725 1726 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1727 vm_object_reference(new_entry->object.vm_object); 1728 } 1729} 1730 1731/* 1732 * vm_map_submap: [ kernel use only ] 1733 * 1734 * Mark the given range as handled by a subordinate map. 1735 * 1736 * This range must have been created with vm_map_find, 1737 * and no other operations may have been performed on this 1738 * range prior to calling vm_map_submap. 1739 * 1740 * Only a limited number of operations can be performed 1741 * within this rage after calling vm_map_submap: 1742 * vm_fault 1743 * [Don't try vm_map_copy!] 1744 * 1745 * To remove a submapping, one must first remove the 1746 * range from the superior map, and then destroy the 1747 * submap (if desired). [Better yet, don't try it.] 1748 */ 1749int 1750vm_map_submap( 1751 vm_map_t map, 1752 vm_offset_t start, 1753 vm_offset_t end, 1754 vm_map_t submap) 1755{ 1756 vm_map_entry_t entry; 1757 int result = KERN_INVALID_ARGUMENT; 1758 1759 vm_map_lock(map); 1760 1761 VM_MAP_RANGE_CHECK(map, start, end); 1762 1763 if (vm_map_lookup_entry(map, start, &entry)) { 1764 vm_map_clip_start(map, entry, start); 1765 } else 1766 entry = entry->next; 1767 1768 vm_map_clip_end(map, entry, end); 1769 1770 if ((entry->start == start) && (entry->end == end) && 1771 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1772 (entry->object.vm_object == NULL)) { 1773 entry->object.sub_map = submap; 1774 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1775 result = KERN_SUCCESS; 1776 } 1777 vm_map_unlock(map); 1778 1779 return (result); 1780} 1781 1782/* 1783 * The maximum number of pages to map 1784 */ 1785#define MAX_INIT_PT 96 1786 1787/* 1788 * vm_map_pmap_enter: 1789 * 1790 * Preload read-only mappings for the specified object's resident pages 1791 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only 1792 * the resident pages within the address range [addr, addr + ulmin(size, 1793 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within 1794 * the specified address range are mapped. This eliminates many soft 1795 * faults on process startup and immediately after an mmap(2). Because 1796 * these are speculative mappings, cached pages are not reactivated and 1797 * mapped. 1798 */ 1799void 1800vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1801 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1802{ 1803 vm_offset_t start; 1804 vm_page_t p, p_start; 1805 vm_pindex_t psize, tmpidx; 1806 1807 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1808 return; 1809 VM_OBJECT_LOCK(object); 1810 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1811 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1812 goto unlock_return; 1813 } 1814 1815 psize = atop(size); 1816 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0) 1817 psize = MAX_INIT_PT; 1818 if (psize + pindex > object->size) { 1819 if (object->size < pindex) 1820 goto unlock_return; 1821 psize = object->size - pindex; 1822 } 1823 1824 start = 0; 1825 p_start = NULL; 1826 1827 p = vm_page_find_least(object, pindex); 1828 /* 1829 * Assert: the variable p is either (1) the page with the 1830 * least pindex greater than or equal to the parameter pindex 1831 * or (2) NULL. 1832 */ 1833 for (; 1834 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1835 p = TAILQ_NEXT(p, listq)) { 1836 /* 1837 * don't allow an madvise to blow away our really 1838 * free pages allocating pv entries. 1839 */ 1840 if ((flags & MAP_PREFAULT_MADVISE) && 1841 cnt.v_free_count < cnt.v_free_reserved) { 1842 psize = tmpidx; 1843 break; 1844 } 1845 if (p->valid == VM_PAGE_BITS_ALL) { 1846 if (p_start == NULL) { 1847 start = addr + ptoa(tmpidx); 1848 p_start = p; 1849 } 1850 } else if (p_start != NULL) { 1851 pmap_enter_object(map->pmap, start, addr + 1852 ptoa(tmpidx), p_start, prot); 1853 p_start = NULL; 1854 } 1855 } 1856 if (p_start != NULL) 1857 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1858 p_start, prot); 1859unlock_return: 1860 VM_OBJECT_UNLOCK(object); 1861} 1862 1863/* 1864 * vm_map_protect: 1865 * 1866 * Sets the protection of the specified address 1867 * region in the target map. If "set_max" is 1868 * specified, the maximum protection is to be set; 1869 * otherwise, only the current protection is affected. 1870 */ 1871int 1872vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1873 vm_prot_t new_prot, boolean_t set_max) 1874{ 1875 vm_map_entry_t current, entry; 1876 vm_object_t obj; 1877 struct ucred *cred; 1878 vm_prot_t old_prot; 1879 1880 vm_map_lock(map); 1881 1882 VM_MAP_RANGE_CHECK(map, start, end); 1883 1884 if (vm_map_lookup_entry(map, start, &entry)) { 1885 vm_map_clip_start(map, entry, start); 1886 } else { 1887 entry = entry->next; 1888 } 1889 1890 /* 1891 * Make a first pass to check for protection violations. 1892 */ 1893 current = entry; 1894 while ((current != &map->header) && (current->start < end)) { 1895 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1896 vm_map_unlock(map); 1897 return (KERN_INVALID_ARGUMENT); 1898 } 1899 if ((new_prot & current->max_protection) != new_prot) { 1900 vm_map_unlock(map); 1901 return (KERN_PROTECTION_FAILURE); 1902 } 1903 current = current->next; 1904 } 1905 1906 1907 /* 1908 * Do an accounting pass for private read-only mappings that 1909 * now will do cow due to allowed write (e.g. debugger sets 1910 * breakpoint on text segment) 1911 */ 1912 for (current = entry; (current != &map->header) && 1913 (current->start < end); current = current->next) { 1914 1915 vm_map_clip_end(map, current, end); 1916 1917 if (set_max || 1918 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1919 ENTRY_CHARGED(current)) { 1920 continue; 1921 } 1922 1923 cred = curthread->td_ucred; 1924 obj = current->object.vm_object; 1925 1926 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1927 if (!swap_reserve(current->end - current->start)) { 1928 vm_map_unlock(map); 1929 return (KERN_RESOURCE_SHORTAGE); 1930 } 1931 crhold(cred); 1932 current->cred = cred; 1933 continue; 1934 } 1935 1936 VM_OBJECT_LOCK(obj); 1937 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1938 VM_OBJECT_UNLOCK(obj); 1939 continue; 1940 } 1941 1942 /* 1943 * Charge for the whole object allocation now, since 1944 * we cannot distinguish between non-charged and 1945 * charged clipped mapping of the same object later. 1946 */ 1947 KASSERT(obj->charge == 0, 1948 ("vm_map_protect: object %p overcharged\n", obj)); 1949 if (!swap_reserve(ptoa(obj->size))) { 1950 VM_OBJECT_UNLOCK(obj); 1951 vm_map_unlock(map); 1952 return (KERN_RESOURCE_SHORTAGE); 1953 } 1954 1955 crhold(cred); 1956 obj->cred = cred; 1957 obj->charge = ptoa(obj->size); 1958 VM_OBJECT_UNLOCK(obj); 1959 } 1960 1961 /* 1962 * Go back and fix up protections. [Note that clipping is not 1963 * necessary the second time.] 1964 */ 1965 current = entry; 1966 while ((current != &map->header) && (current->start < end)) { 1967 old_prot = current->protection; 1968 1969 if (set_max) 1970 current->protection = 1971 (current->max_protection = new_prot) & 1972 old_prot; 1973 else 1974 current->protection = new_prot; 1975 1976 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED)) 1977 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) && 1978 (current->protection & VM_PROT_WRITE) != 0 && 1979 (old_prot & VM_PROT_WRITE) == 0) { 1980 vm_fault_copy_entry(map, map, current, current, NULL); 1981 } 1982 1983 /* 1984 * When restricting access, update the physical map. Worry 1985 * about copy-on-write here. 1986 */ 1987 if ((old_prot & ~current->protection) != 0) { 1988#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1989 VM_PROT_ALL) 1990 pmap_protect(map->pmap, current->start, 1991 current->end, 1992 current->protection & MASK(current)); 1993#undef MASK 1994 } 1995 vm_map_simplify_entry(map, current); 1996 current = current->next; 1997 } 1998 vm_map_unlock(map); 1999 return (KERN_SUCCESS); 2000} 2001 2002/* 2003 * vm_map_madvise: 2004 * 2005 * This routine traverses a processes map handling the madvise 2006 * system call. Advisories are classified as either those effecting 2007 * the vm_map_entry structure, or those effecting the underlying 2008 * objects. 2009 */ 2010int 2011vm_map_madvise( 2012 vm_map_t map, 2013 vm_offset_t start, 2014 vm_offset_t end, 2015 int behav) 2016{ 2017 vm_map_entry_t current, entry; 2018 int modify_map = 0; 2019 2020 /* 2021 * Some madvise calls directly modify the vm_map_entry, in which case 2022 * we need to use an exclusive lock on the map and we need to perform 2023 * various clipping operations. Otherwise we only need a read-lock 2024 * on the map. 2025 */ 2026 switch(behav) { 2027 case MADV_NORMAL: 2028 case MADV_SEQUENTIAL: 2029 case MADV_RANDOM: 2030 case MADV_NOSYNC: 2031 case MADV_AUTOSYNC: 2032 case MADV_NOCORE: 2033 case MADV_CORE: 2034 modify_map = 1; 2035 vm_map_lock(map); 2036 break; 2037 case MADV_WILLNEED: 2038 case MADV_DONTNEED: 2039 case MADV_FREE: 2040 vm_map_lock_read(map); 2041 break; 2042 default: 2043 return (KERN_INVALID_ARGUMENT); 2044 } 2045 2046 /* 2047 * Locate starting entry and clip if necessary. 2048 */ 2049 VM_MAP_RANGE_CHECK(map, start, end); 2050 2051 if (vm_map_lookup_entry(map, start, &entry)) { 2052 if (modify_map) 2053 vm_map_clip_start(map, entry, start); 2054 } else { 2055 entry = entry->next; 2056 } 2057 2058 if (modify_map) { 2059 /* 2060 * madvise behaviors that are implemented in the vm_map_entry. 2061 * 2062 * We clip the vm_map_entry so that behavioral changes are 2063 * limited to the specified address range. 2064 */ 2065 for (current = entry; 2066 (current != &map->header) && (current->start < end); 2067 current = current->next 2068 ) { 2069 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2070 continue; 2071 2072 vm_map_clip_end(map, current, end); 2073 2074 switch (behav) { 2075 case MADV_NORMAL: 2076 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2077 break; 2078 case MADV_SEQUENTIAL: 2079 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2080 break; 2081 case MADV_RANDOM: 2082 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2083 break; 2084 case MADV_NOSYNC: 2085 current->eflags |= MAP_ENTRY_NOSYNC; 2086 break; 2087 case MADV_AUTOSYNC: 2088 current->eflags &= ~MAP_ENTRY_NOSYNC; 2089 break; 2090 case MADV_NOCORE: 2091 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2092 break; 2093 case MADV_CORE: 2094 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2095 break; 2096 default: 2097 break; 2098 } 2099 vm_map_simplify_entry(map, current); 2100 } 2101 vm_map_unlock(map); 2102 } else { 2103 vm_pindex_t pstart, pend; 2104 2105 /* 2106 * madvise behaviors that are implemented in the underlying 2107 * vm_object. 2108 * 2109 * Since we don't clip the vm_map_entry, we have to clip 2110 * the vm_object pindex and count. 2111 */ 2112 for (current = entry; 2113 (current != &map->header) && (current->start < end); 2114 current = current->next 2115 ) { 2116 vm_offset_t useStart; 2117 2118 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2119 continue; 2120 2121 pstart = OFF_TO_IDX(current->offset); 2122 pend = pstart + atop(current->end - current->start); 2123 useStart = current->start; 2124 2125 if (current->start < start) { 2126 pstart += atop(start - current->start); 2127 useStart = start; 2128 } 2129 if (current->end > end) 2130 pend -= atop(current->end - end); 2131 2132 if (pstart >= pend) 2133 continue; 2134 2135 vm_object_madvise(current->object.vm_object, pstart, 2136 pend, behav); 2137 if (behav == MADV_WILLNEED) { 2138 vm_map_pmap_enter(map, 2139 useStart, 2140 current->protection, 2141 current->object.vm_object, 2142 pstart, 2143 ptoa(pend - pstart), 2144 MAP_PREFAULT_MADVISE 2145 ); 2146 } 2147 } 2148 vm_map_unlock_read(map); 2149 } 2150 return (0); 2151} 2152 2153 2154/* 2155 * vm_map_inherit: 2156 * 2157 * Sets the inheritance of the specified address 2158 * range in the target map. Inheritance 2159 * affects how the map will be shared with 2160 * child maps at the time of vmspace_fork. 2161 */ 2162int 2163vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2164 vm_inherit_t new_inheritance) 2165{ 2166 vm_map_entry_t entry; 2167 vm_map_entry_t temp_entry; 2168 2169 switch (new_inheritance) { 2170 case VM_INHERIT_NONE: 2171 case VM_INHERIT_COPY: 2172 case VM_INHERIT_SHARE: 2173 break; 2174 default: 2175 return (KERN_INVALID_ARGUMENT); 2176 } 2177 vm_map_lock(map); 2178 VM_MAP_RANGE_CHECK(map, start, end); 2179 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2180 entry = temp_entry; 2181 vm_map_clip_start(map, entry, start); 2182 } else 2183 entry = temp_entry->next; 2184 while ((entry != &map->header) && (entry->start < end)) { 2185 vm_map_clip_end(map, entry, end); 2186 entry->inheritance = new_inheritance; 2187 vm_map_simplify_entry(map, entry); 2188 entry = entry->next; 2189 } 2190 vm_map_unlock(map); 2191 return (KERN_SUCCESS); 2192} 2193 2194/* 2195 * vm_map_unwire: 2196 * 2197 * Implements both kernel and user unwiring. 2198 */ 2199int 2200vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2201 int flags) 2202{ 2203 vm_map_entry_t entry, first_entry, tmp_entry; 2204 vm_offset_t saved_start; 2205 unsigned int last_timestamp; 2206 int rv; 2207 boolean_t need_wakeup, result, user_unwire; 2208 2209 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2210 vm_map_lock(map); 2211 VM_MAP_RANGE_CHECK(map, start, end); 2212 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2213 if (flags & VM_MAP_WIRE_HOLESOK) 2214 first_entry = first_entry->next; 2215 else { 2216 vm_map_unlock(map); 2217 return (KERN_INVALID_ADDRESS); 2218 } 2219 } 2220 last_timestamp = map->timestamp; 2221 entry = first_entry; 2222 while (entry != &map->header && entry->start < end) { 2223 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2224 /* 2225 * We have not yet clipped the entry. 2226 */ 2227 saved_start = (start >= entry->start) ? start : 2228 entry->start; 2229 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2230 if (vm_map_unlock_and_wait(map, 0)) { 2231 /* 2232 * Allow interruption of user unwiring? 2233 */ 2234 } 2235 vm_map_lock(map); 2236 if (last_timestamp+1 != map->timestamp) { 2237 /* 2238 * Look again for the entry because the map was 2239 * modified while it was unlocked. 2240 * Specifically, the entry may have been 2241 * clipped, merged, or deleted. 2242 */ 2243 if (!vm_map_lookup_entry(map, saved_start, 2244 &tmp_entry)) { 2245 if (flags & VM_MAP_WIRE_HOLESOK) 2246 tmp_entry = tmp_entry->next; 2247 else { 2248 if (saved_start == start) { 2249 /* 2250 * First_entry has been deleted. 2251 */ 2252 vm_map_unlock(map); 2253 return (KERN_INVALID_ADDRESS); 2254 } 2255 end = saved_start; 2256 rv = KERN_INVALID_ADDRESS; 2257 goto done; 2258 } 2259 } 2260 if (entry == first_entry) 2261 first_entry = tmp_entry; 2262 else 2263 first_entry = NULL; 2264 entry = tmp_entry; 2265 } 2266 last_timestamp = map->timestamp; 2267 continue; 2268 } 2269 vm_map_clip_start(map, entry, start); 2270 vm_map_clip_end(map, entry, end); 2271 /* 2272 * Mark the entry in case the map lock is released. (See 2273 * above.) 2274 */ 2275 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2276 /* 2277 * Check the map for holes in the specified region. 2278 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2279 */ 2280 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2281 (entry->end < end && (entry->next == &map->header || 2282 entry->next->start > entry->end))) { 2283 end = entry->end; 2284 rv = KERN_INVALID_ADDRESS; 2285 goto done; 2286 } 2287 /* 2288 * If system unwiring, require that the entry is system wired. 2289 */ 2290 if (!user_unwire && 2291 vm_map_entry_system_wired_count(entry) == 0) { 2292 end = entry->end; 2293 rv = KERN_INVALID_ARGUMENT; 2294 goto done; 2295 } 2296 entry = entry->next; 2297 } 2298 rv = KERN_SUCCESS; 2299done: 2300 need_wakeup = FALSE; 2301 if (first_entry == NULL) { 2302 result = vm_map_lookup_entry(map, start, &first_entry); 2303 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2304 first_entry = first_entry->next; 2305 else 2306 KASSERT(result, ("vm_map_unwire: lookup failed")); 2307 } 2308 entry = first_entry; 2309 while (entry != &map->header && entry->start < end) { 2310 if (rv == KERN_SUCCESS && (!user_unwire || 2311 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2312 if (user_unwire) 2313 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2314 entry->wired_count--; 2315 if (entry->wired_count == 0) { 2316 /* 2317 * Retain the map lock. 2318 */ 2319 vm_fault_unwire(map, entry->start, entry->end, 2320 entry->object.vm_object != NULL && 2321 (entry->object.vm_object->flags & 2322 OBJ_FICTITIOUS) != 0); 2323 } 2324 } 2325 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2326 ("vm_map_unwire: in-transition flag missing")); 2327 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2328 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2329 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2330 need_wakeup = TRUE; 2331 } 2332 vm_map_simplify_entry(map, entry); 2333 entry = entry->next; 2334 } 2335 vm_map_unlock(map); 2336 if (need_wakeup) 2337 vm_map_wakeup(map); 2338 return (rv); 2339} 2340 2341/* 2342 * vm_map_wire: 2343 * 2344 * Implements both kernel and user wiring. 2345 */ 2346int 2347vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2348 int flags) 2349{ 2350 vm_map_entry_t entry, first_entry, tmp_entry; 2351 vm_offset_t saved_end, saved_start; 2352 unsigned int last_timestamp; 2353 int rv; 2354 boolean_t fictitious, need_wakeup, result, user_wire; 2355 vm_prot_t prot; 2356 2357 prot = 0; 2358 if (flags & VM_MAP_WIRE_WRITE) 2359 prot |= VM_PROT_WRITE; 2360 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2361 vm_map_lock(map); 2362 VM_MAP_RANGE_CHECK(map, start, end); 2363 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2364 if (flags & VM_MAP_WIRE_HOLESOK) 2365 first_entry = first_entry->next; 2366 else { 2367 vm_map_unlock(map); 2368 return (KERN_INVALID_ADDRESS); 2369 } 2370 } 2371 last_timestamp = map->timestamp; 2372 entry = first_entry; 2373 while (entry != &map->header && entry->start < end) { 2374 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2375 /* 2376 * We have not yet clipped the entry. 2377 */ 2378 saved_start = (start >= entry->start) ? start : 2379 entry->start; 2380 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2381 if (vm_map_unlock_and_wait(map, 0)) { 2382 /* 2383 * Allow interruption of user wiring? 2384 */ 2385 } 2386 vm_map_lock(map); 2387 if (last_timestamp + 1 != map->timestamp) { 2388 /* 2389 * Look again for the entry because the map was 2390 * modified while it was unlocked. 2391 * Specifically, the entry may have been 2392 * clipped, merged, or deleted. 2393 */ 2394 if (!vm_map_lookup_entry(map, saved_start, 2395 &tmp_entry)) { 2396 if (flags & VM_MAP_WIRE_HOLESOK) 2397 tmp_entry = tmp_entry->next; 2398 else { 2399 if (saved_start == start) { 2400 /* 2401 * first_entry has been deleted. 2402 */ 2403 vm_map_unlock(map); 2404 return (KERN_INVALID_ADDRESS); 2405 } 2406 end = saved_start; 2407 rv = KERN_INVALID_ADDRESS; 2408 goto done; 2409 } 2410 } 2411 if (entry == first_entry) 2412 first_entry = tmp_entry; 2413 else 2414 first_entry = NULL; 2415 entry = tmp_entry; 2416 } 2417 last_timestamp = map->timestamp; 2418 continue; 2419 } 2420 vm_map_clip_start(map, entry, start); 2421 vm_map_clip_end(map, entry, end); 2422 /* 2423 * Mark the entry in case the map lock is released. (See 2424 * above.) 2425 */ 2426 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2427 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2428 || (entry->protection & prot) != prot) { 2429 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2430 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2431 end = entry->end; 2432 rv = KERN_INVALID_ADDRESS; 2433 goto done; 2434 } 2435 goto next_entry; 2436 } 2437 if (entry->wired_count == 0) { 2438 entry->wired_count++; 2439 saved_start = entry->start; 2440 saved_end = entry->end; 2441 fictitious = entry->object.vm_object != NULL && 2442 (entry->object.vm_object->flags & 2443 OBJ_FICTITIOUS) != 0; 2444 /* 2445 * Release the map lock, relying on the in-transition 2446 * mark. Mark the map busy for fork. 2447 */ 2448 vm_map_busy(map); 2449 vm_map_unlock(map); 2450 rv = vm_fault_wire(map, saved_start, saved_end, 2451 fictitious); 2452 vm_map_lock(map); 2453 vm_map_unbusy(map); 2454 if (last_timestamp + 1 != map->timestamp) { 2455 /* 2456 * Look again for the entry because the map was 2457 * modified while it was unlocked. The entry 2458 * may have been clipped, but NOT merged or 2459 * deleted. 2460 */ 2461 result = vm_map_lookup_entry(map, saved_start, 2462 &tmp_entry); 2463 KASSERT(result, ("vm_map_wire: lookup failed")); 2464 if (entry == first_entry) 2465 first_entry = tmp_entry; 2466 else 2467 first_entry = NULL; 2468 entry = tmp_entry; 2469 while (entry->end < saved_end) { 2470 if (rv != KERN_SUCCESS) { 2471 KASSERT(entry->wired_count == 1, 2472 ("vm_map_wire: bad count")); 2473 entry->wired_count = -1; 2474 } 2475 entry = entry->next; 2476 } 2477 } 2478 last_timestamp = map->timestamp; 2479 if (rv != KERN_SUCCESS) { 2480 KASSERT(entry->wired_count == 1, 2481 ("vm_map_wire: bad count")); 2482 /* 2483 * Assign an out-of-range value to represent 2484 * the failure to wire this entry. 2485 */ 2486 entry->wired_count = -1; 2487 end = entry->end; 2488 goto done; 2489 } 2490 } else if (!user_wire || 2491 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2492 entry->wired_count++; 2493 } 2494 /* 2495 * Check the map for holes in the specified region. 2496 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2497 */ 2498 next_entry: 2499 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2500 (entry->end < end && (entry->next == &map->header || 2501 entry->next->start > entry->end))) { 2502 end = entry->end; 2503 rv = KERN_INVALID_ADDRESS; 2504 goto done; 2505 } 2506 entry = entry->next; 2507 } 2508 rv = KERN_SUCCESS; 2509done: 2510 need_wakeup = FALSE; 2511 if (first_entry == NULL) { 2512 result = vm_map_lookup_entry(map, start, &first_entry); 2513 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2514 first_entry = first_entry->next; 2515 else 2516 KASSERT(result, ("vm_map_wire: lookup failed")); 2517 } 2518 entry = first_entry; 2519 while (entry != &map->header && entry->start < end) { 2520 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2521 goto next_entry_done; 2522 if (rv == KERN_SUCCESS) { 2523 if (user_wire) 2524 entry->eflags |= MAP_ENTRY_USER_WIRED; 2525 } else if (entry->wired_count == -1) { 2526 /* 2527 * Wiring failed on this entry. Thus, unwiring is 2528 * unnecessary. 2529 */ 2530 entry->wired_count = 0; 2531 } else { 2532 if (!user_wire || 2533 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2534 entry->wired_count--; 2535 if (entry->wired_count == 0) { 2536 /* 2537 * Retain the map lock. 2538 */ 2539 vm_fault_unwire(map, entry->start, entry->end, 2540 entry->object.vm_object != NULL && 2541 (entry->object.vm_object->flags & 2542 OBJ_FICTITIOUS) != 0); 2543 } 2544 } 2545 next_entry_done: 2546 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2547 ("vm_map_wire: in-transition flag missing")); 2548 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED); 2549 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2550 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2551 need_wakeup = TRUE; 2552 } 2553 vm_map_simplify_entry(map, entry); 2554 entry = entry->next; 2555 } 2556 vm_map_unlock(map); 2557 if (need_wakeup) 2558 vm_map_wakeup(map); 2559 return (rv); 2560} 2561 2562/* 2563 * vm_map_sync 2564 * 2565 * Push any dirty cached pages in the address range to their pager. 2566 * If syncio is TRUE, dirty pages are written synchronously. 2567 * If invalidate is TRUE, any cached pages are freed as well. 2568 * 2569 * If the size of the region from start to end is zero, we are 2570 * supposed to flush all modified pages within the region containing 2571 * start. Unfortunately, a region can be split or coalesced with 2572 * neighboring regions, making it difficult to determine what the 2573 * original region was. Therefore, we approximate this requirement by 2574 * flushing the current region containing start. 2575 * 2576 * Returns an error if any part of the specified range is not mapped. 2577 */ 2578int 2579vm_map_sync( 2580 vm_map_t map, 2581 vm_offset_t start, 2582 vm_offset_t end, 2583 boolean_t syncio, 2584 boolean_t invalidate) 2585{ 2586 vm_map_entry_t current; 2587 vm_map_entry_t entry; 2588 vm_size_t size; 2589 vm_object_t object; 2590 vm_ooffset_t offset; 2591 unsigned int last_timestamp; 2592 boolean_t failed; 2593 2594 vm_map_lock_read(map); 2595 VM_MAP_RANGE_CHECK(map, start, end); 2596 if (!vm_map_lookup_entry(map, start, &entry)) { 2597 vm_map_unlock_read(map); 2598 return (KERN_INVALID_ADDRESS); 2599 } else if (start == end) { 2600 start = entry->start; 2601 end = entry->end; 2602 } 2603 /* 2604 * Make a first pass to check for user-wired memory and holes. 2605 */ 2606 for (current = entry; current != &map->header && current->start < end; 2607 current = current->next) { 2608 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2609 vm_map_unlock_read(map); 2610 return (KERN_INVALID_ARGUMENT); 2611 } 2612 if (end > current->end && 2613 (current->next == &map->header || 2614 current->end != current->next->start)) { 2615 vm_map_unlock_read(map); 2616 return (KERN_INVALID_ADDRESS); 2617 } 2618 } 2619 2620 if (invalidate) 2621 pmap_remove(map->pmap, start, end); 2622 failed = FALSE; 2623 2624 /* 2625 * Make a second pass, cleaning/uncaching pages from the indicated 2626 * objects as we go. 2627 */ 2628 for (current = entry; current != &map->header && current->start < end;) { 2629 offset = current->offset + (start - current->start); 2630 size = (end <= current->end ? end : current->end) - start; 2631 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2632 vm_map_t smap; 2633 vm_map_entry_t tentry; 2634 vm_size_t tsize; 2635 2636 smap = current->object.sub_map; 2637 vm_map_lock_read(smap); 2638 (void) vm_map_lookup_entry(smap, offset, &tentry); 2639 tsize = tentry->end - offset; 2640 if (tsize < size) 2641 size = tsize; 2642 object = tentry->object.vm_object; 2643 offset = tentry->offset + (offset - tentry->start); 2644 vm_map_unlock_read(smap); 2645 } else { 2646 object = current->object.vm_object; 2647 } 2648 vm_object_reference(object); 2649 last_timestamp = map->timestamp; 2650 vm_map_unlock_read(map); 2651 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2652 failed = TRUE; 2653 start += size; 2654 vm_object_deallocate(object); 2655 vm_map_lock_read(map); 2656 if (last_timestamp == map->timestamp || 2657 !vm_map_lookup_entry(map, start, ¤t)) 2658 current = current->next; 2659 } 2660 2661 vm_map_unlock_read(map); 2662 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2663} 2664 2665/* 2666 * vm_map_entry_unwire: [ internal use only ] 2667 * 2668 * Make the region specified by this entry pageable. 2669 * 2670 * The map in question should be locked. 2671 * [This is the reason for this routine's existence.] 2672 */ 2673static void 2674vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2675{ 2676 vm_fault_unwire(map, entry->start, entry->end, 2677 entry->object.vm_object != NULL && 2678 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2679 entry->wired_count = 0; 2680} 2681 2682static void 2683vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2684{ 2685 2686 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2687 vm_object_deallocate(entry->object.vm_object); 2688 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2689} 2690 2691/* 2692 * vm_map_entry_delete: [ internal use only ] 2693 * 2694 * Deallocate the given entry from the target map. 2695 */ 2696static void 2697vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2698{ 2699 vm_object_t object; 2700 vm_pindex_t offidxstart, offidxend, count, size1; 2701 vm_ooffset_t size; 2702 2703 vm_map_entry_unlink(map, entry); 2704 object = entry->object.vm_object; 2705 size = entry->end - entry->start; 2706 map->size -= size; 2707 2708 if (entry->cred != NULL) { 2709 swap_release_by_cred(size, entry->cred); 2710 crfree(entry->cred); 2711 } 2712 2713 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2714 (object != NULL)) { 2715 KASSERT(entry->cred == NULL || object->cred == NULL || 2716 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2717 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2718 count = OFF_TO_IDX(size); 2719 offidxstart = OFF_TO_IDX(entry->offset); 2720 offidxend = offidxstart + count; 2721 VM_OBJECT_LOCK(object); 2722 if (object->ref_count != 1 && 2723 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2724 object == kernel_object || object == kmem_object)) { 2725 vm_object_collapse(object); 2726 2727 /* 2728 * The option OBJPR_NOTMAPPED can be passed here 2729 * because vm_map_delete() already performed 2730 * pmap_remove() on the only mapping to this range 2731 * of pages. 2732 */ 2733 vm_object_page_remove(object, offidxstart, offidxend, 2734 OBJPR_NOTMAPPED); 2735 if (object->type == OBJT_SWAP) 2736 swap_pager_freespace(object, offidxstart, count); 2737 if (offidxend >= object->size && 2738 offidxstart < object->size) { 2739 size1 = object->size; 2740 object->size = offidxstart; 2741 if (object->cred != NULL) { 2742 size1 -= object->size; 2743 KASSERT(object->charge >= ptoa(size1), 2744 ("vm_map_entry_delete: object->charge < 0")); 2745 swap_release_by_cred(ptoa(size1), object->cred); 2746 object->charge -= ptoa(size1); 2747 } 2748 } 2749 } 2750 VM_OBJECT_UNLOCK(object); 2751 } else 2752 entry->object.vm_object = NULL; 2753 if (map->system_map) 2754 vm_map_entry_deallocate(entry, TRUE); 2755 else { 2756 entry->next = curthread->td_map_def_user; 2757 curthread->td_map_def_user = entry; 2758 } 2759} 2760 2761/* 2762 * vm_map_delete: [ internal use only ] 2763 * 2764 * Deallocates the given address range from the target 2765 * map. 2766 */ 2767int 2768vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2769{ 2770 vm_map_entry_t entry; 2771 vm_map_entry_t first_entry; 2772 2773 VM_MAP_ASSERT_LOCKED(map); 2774 2775 /* 2776 * Find the start of the region, and clip it 2777 */ 2778 if (!vm_map_lookup_entry(map, start, &first_entry)) 2779 entry = first_entry->next; 2780 else { 2781 entry = first_entry; 2782 vm_map_clip_start(map, entry, start); 2783 } 2784 2785 /* 2786 * Step through all entries in this region 2787 */ 2788 while ((entry != &map->header) && (entry->start < end)) { 2789 vm_map_entry_t next; 2790 2791 /* 2792 * Wait for wiring or unwiring of an entry to complete. 2793 * Also wait for any system wirings to disappear on 2794 * user maps. 2795 */ 2796 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2797 (vm_map_pmap(map) != kernel_pmap && 2798 vm_map_entry_system_wired_count(entry) != 0)) { 2799 unsigned int last_timestamp; 2800 vm_offset_t saved_start; 2801 vm_map_entry_t tmp_entry; 2802 2803 saved_start = entry->start; 2804 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2805 last_timestamp = map->timestamp; 2806 (void) vm_map_unlock_and_wait(map, 0); 2807 vm_map_lock(map); 2808 if (last_timestamp + 1 != map->timestamp) { 2809 /* 2810 * Look again for the entry because the map was 2811 * modified while it was unlocked. 2812 * Specifically, the entry may have been 2813 * clipped, merged, or deleted. 2814 */ 2815 if (!vm_map_lookup_entry(map, saved_start, 2816 &tmp_entry)) 2817 entry = tmp_entry->next; 2818 else { 2819 entry = tmp_entry; 2820 vm_map_clip_start(map, entry, 2821 saved_start); 2822 } 2823 } 2824 continue; 2825 } 2826 vm_map_clip_end(map, entry, end); 2827 2828 next = entry->next; 2829 2830 /* 2831 * Unwire before removing addresses from the pmap; otherwise, 2832 * unwiring will put the entries back in the pmap. 2833 */ 2834 if (entry->wired_count != 0) { 2835 vm_map_entry_unwire(map, entry); 2836 } 2837 2838 pmap_remove(map->pmap, entry->start, entry->end); 2839 2840 /* 2841 * Delete the entry only after removing all pmap 2842 * entries pointing to its pages. (Otherwise, its 2843 * page frames may be reallocated, and any modify bits 2844 * will be set in the wrong object!) 2845 */ 2846 vm_map_entry_delete(map, entry); 2847 entry = next; 2848 } 2849 return (KERN_SUCCESS); 2850} 2851 2852/* 2853 * vm_map_remove: 2854 * 2855 * Remove the given address range from the target map. 2856 * This is the exported form of vm_map_delete. 2857 */ 2858int 2859vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2860{ 2861 int result; 2862 2863 vm_map_lock(map); 2864 VM_MAP_RANGE_CHECK(map, start, end); 2865 result = vm_map_delete(map, start, end); 2866 vm_map_unlock(map); 2867 return (result); 2868} 2869 2870/* 2871 * vm_map_check_protection: 2872 * 2873 * Assert that the target map allows the specified privilege on the 2874 * entire address region given. The entire region must be allocated. 2875 * 2876 * WARNING! This code does not and should not check whether the 2877 * contents of the region is accessible. For example a smaller file 2878 * might be mapped into a larger address space. 2879 * 2880 * NOTE! This code is also called by munmap(). 2881 * 2882 * The map must be locked. A read lock is sufficient. 2883 */ 2884boolean_t 2885vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2886 vm_prot_t protection) 2887{ 2888 vm_map_entry_t entry; 2889 vm_map_entry_t tmp_entry; 2890 2891 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2892 return (FALSE); 2893 entry = tmp_entry; 2894 2895 while (start < end) { 2896 if (entry == &map->header) 2897 return (FALSE); 2898 /* 2899 * No holes allowed! 2900 */ 2901 if (start < entry->start) 2902 return (FALSE); 2903 /* 2904 * Check protection associated with entry. 2905 */ 2906 if ((entry->protection & protection) != protection) 2907 return (FALSE); 2908 /* go to next entry */ 2909 start = entry->end; 2910 entry = entry->next; 2911 } 2912 return (TRUE); 2913} 2914 2915/* 2916 * vm_map_copy_entry: 2917 * 2918 * Copies the contents of the source entry to the destination 2919 * entry. The entries *must* be aligned properly. 2920 */ 2921static void 2922vm_map_copy_entry( 2923 vm_map_t src_map, 2924 vm_map_t dst_map, 2925 vm_map_entry_t src_entry, 2926 vm_map_entry_t dst_entry, 2927 vm_ooffset_t *fork_charge) 2928{ 2929 vm_object_t src_object; 2930 vm_map_entry_t fake_entry; 2931 vm_offset_t size; 2932 struct ucred *cred; 2933 int charged; 2934 2935 VM_MAP_ASSERT_LOCKED(dst_map); 2936 2937 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2938 return; 2939 2940 if (src_entry->wired_count == 0) { 2941 2942 /* 2943 * If the source entry is marked needs_copy, it is already 2944 * write-protected. 2945 */ 2946 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2947 pmap_protect(src_map->pmap, 2948 src_entry->start, 2949 src_entry->end, 2950 src_entry->protection & ~VM_PROT_WRITE); 2951 } 2952 2953 /* 2954 * Make a copy of the object. 2955 */ 2956 size = src_entry->end - src_entry->start; 2957 if ((src_object = src_entry->object.vm_object) != NULL) { 2958 VM_OBJECT_LOCK(src_object); 2959 charged = ENTRY_CHARGED(src_entry); 2960 if ((src_object->handle == NULL) && 2961 (src_object->type == OBJT_DEFAULT || 2962 src_object->type == OBJT_SWAP)) { 2963 vm_object_collapse(src_object); 2964 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2965 vm_object_split(src_entry); 2966 src_object = src_entry->object.vm_object; 2967 } 2968 } 2969 vm_object_reference_locked(src_object); 2970 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2971 if (src_entry->cred != NULL && 2972 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 2973 KASSERT(src_object->cred == NULL, 2974 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 2975 src_object)); 2976 src_object->cred = src_entry->cred; 2977 src_object->charge = size; 2978 } 2979 VM_OBJECT_UNLOCK(src_object); 2980 dst_entry->object.vm_object = src_object; 2981 if (charged) { 2982 cred = curthread->td_ucred; 2983 crhold(cred); 2984 dst_entry->cred = cred; 2985 *fork_charge += size; 2986 if (!(src_entry->eflags & 2987 MAP_ENTRY_NEEDS_COPY)) { 2988 crhold(cred); 2989 src_entry->cred = cred; 2990 *fork_charge += size; 2991 } 2992 } 2993 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2994 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2995 dst_entry->offset = src_entry->offset; 2996 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 2997 /* 2998 * MAP_ENTRY_VN_WRITECNT cannot 2999 * indicate write reference from 3000 * src_entry, since the entry is 3001 * marked as needs copy. Allocate a 3002 * fake entry that is used to 3003 * decrement object->un_pager.vnp.writecount 3004 * at the appropriate time. Attach 3005 * fake_entry to the deferred list. 3006 */ 3007 fake_entry = vm_map_entry_create(dst_map); 3008 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3009 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3010 vm_object_reference(src_object); 3011 fake_entry->object.vm_object = src_object; 3012 fake_entry->start = src_entry->start; 3013 fake_entry->end = src_entry->end; 3014 fake_entry->next = curthread->td_map_def_user; 3015 curthread->td_map_def_user = fake_entry; 3016 } 3017 } else { 3018 dst_entry->object.vm_object = NULL; 3019 dst_entry->offset = 0; 3020 if (src_entry->cred != NULL) { 3021 dst_entry->cred = curthread->td_ucred; 3022 crhold(dst_entry->cred); 3023 *fork_charge += size; 3024 } 3025 } 3026 3027 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3028 dst_entry->end - dst_entry->start, src_entry->start); 3029 } else { 3030 /* 3031 * Of course, wired down pages can't be set copy-on-write. 3032 * Cause wired pages to be copied into the new map by 3033 * simulating faults (the new pages are pageable) 3034 */ 3035 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3036 fork_charge); 3037 } 3038} 3039 3040/* 3041 * vmspace_map_entry_forked: 3042 * Update the newly-forked vmspace each time a map entry is inherited 3043 * or copied. The values for vm_dsize and vm_tsize are approximate 3044 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3045 */ 3046static void 3047vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3048 vm_map_entry_t entry) 3049{ 3050 vm_size_t entrysize; 3051 vm_offset_t newend; 3052 3053 entrysize = entry->end - entry->start; 3054 vm2->vm_map.size += entrysize; 3055 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3056 vm2->vm_ssize += btoc(entrysize); 3057 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3058 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3059 newend = MIN(entry->end, 3060 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3061 vm2->vm_dsize += btoc(newend - entry->start); 3062 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3063 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3064 newend = MIN(entry->end, 3065 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3066 vm2->vm_tsize += btoc(newend - entry->start); 3067 } 3068} 3069 3070/* 3071 * vmspace_fork: 3072 * Create a new process vmspace structure and vm_map 3073 * based on those of an existing process. The new map 3074 * is based on the old map, according to the inheritance 3075 * values on the regions in that map. 3076 * 3077 * XXX It might be worth coalescing the entries added to the new vmspace. 3078 * 3079 * The source map must not be locked. 3080 */ 3081struct vmspace * 3082vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3083{ 3084 struct vmspace *vm2; 3085 vm_map_t new_map, old_map; 3086 vm_map_entry_t new_entry, old_entry; 3087 vm_object_t object; 3088 int locked; 3089 3090 old_map = &vm1->vm_map; 3091 /* Copy immutable fields of vm1 to vm2. */ 3092 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 3093 if (vm2 == NULL) 3094 return (NULL); 3095 vm2->vm_taddr = vm1->vm_taddr; 3096 vm2->vm_daddr = vm1->vm_daddr; 3097 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3098 vm_map_lock(old_map); 3099 if (old_map->busy) 3100 vm_map_wait_busy(old_map); 3101 new_map = &vm2->vm_map; 3102 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3103 KASSERT(locked, ("vmspace_fork: lock failed")); 3104 3105 old_entry = old_map->header.next; 3106 3107 while (old_entry != &old_map->header) { 3108 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3109 panic("vm_map_fork: encountered a submap"); 3110 3111 switch (old_entry->inheritance) { 3112 case VM_INHERIT_NONE: 3113 break; 3114 3115 case VM_INHERIT_SHARE: 3116 /* 3117 * Clone the entry, creating the shared object if necessary. 3118 */ 3119 object = old_entry->object.vm_object; 3120 if (object == NULL) { 3121 object = vm_object_allocate(OBJT_DEFAULT, 3122 atop(old_entry->end - old_entry->start)); 3123 old_entry->object.vm_object = object; 3124 old_entry->offset = 0; 3125 if (old_entry->cred != NULL) { 3126 object->cred = old_entry->cred; 3127 object->charge = old_entry->end - 3128 old_entry->start; 3129 old_entry->cred = NULL; 3130 } 3131 } 3132 3133 /* 3134 * Add the reference before calling vm_object_shadow 3135 * to insure that a shadow object is created. 3136 */ 3137 vm_object_reference(object); 3138 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3139 vm_object_shadow(&old_entry->object.vm_object, 3140 &old_entry->offset, 3141 old_entry->end - old_entry->start); 3142 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3143 /* Transfer the second reference too. */ 3144 vm_object_reference( 3145 old_entry->object.vm_object); 3146 3147 /* 3148 * As in vm_map_simplify_entry(), the 3149 * vnode lock will not be acquired in 3150 * this call to vm_object_deallocate(). 3151 */ 3152 vm_object_deallocate(object); 3153 object = old_entry->object.vm_object; 3154 } 3155 VM_OBJECT_LOCK(object); 3156 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3157 if (old_entry->cred != NULL) { 3158 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3159 object->cred = old_entry->cred; 3160 object->charge = old_entry->end - old_entry->start; 3161 old_entry->cred = NULL; 3162 } 3163 VM_OBJECT_UNLOCK(object); 3164 3165 /* 3166 * Clone the entry, referencing the shared object. 3167 */ 3168 new_entry = vm_map_entry_create(new_map); 3169 *new_entry = *old_entry; 3170 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3171 MAP_ENTRY_IN_TRANSITION); 3172 new_entry->wired_count = 0; 3173 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3174 object = new_entry->object.vm_object; 3175 KASSERT(((struct vnode *)object->handle)-> 3176 v_writecount > 0, 3177 ("vmspace_fork: v_writecount")); 3178 KASSERT(object->un_pager.vnp.writemappings > 0, 3179 ("vmspace_fork: vnp.writecount")); 3180 vnode_pager_update_writecount(object, 3181 new_entry->start, new_entry->end); 3182 } 3183 3184 /* 3185 * Insert the entry into the new map -- we know we're 3186 * inserting at the end of the new map. 3187 */ 3188 vm_map_entry_link(new_map, new_map->header.prev, 3189 new_entry); 3190 vmspace_map_entry_forked(vm1, vm2, new_entry); 3191 3192 /* 3193 * Update the physical map 3194 */ 3195 pmap_copy(new_map->pmap, old_map->pmap, 3196 new_entry->start, 3197 (old_entry->end - old_entry->start), 3198 old_entry->start); 3199 break; 3200 3201 case VM_INHERIT_COPY: 3202 /* 3203 * Clone the entry and link into the map. 3204 */ 3205 new_entry = vm_map_entry_create(new_map); 3206 *new_entry = *old_entry; 3207 /* 3208 * Copied entry is COW over the old object. 3209 */ 3210 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3211 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3212 new_entry->wired_count = 0; 3213 new_entry->object.vm_object = NULL; 3214 new_entry->cred = NULL; 3215 vm_map_entry_link(new_map, new_map->header.prev, 3216 new_entry); 3217 vmspace_map_entry_forked(vm1, vm2, new_entry); 3218 vm_map_copy_entry(old_map, new_map, old_entry, 3219 new_entry, fork_charge); 3220 break; 3221 } 3222 old_entry = old_entry->next; 3223 } 3224 /* 3225 * Use inlined vm_map_unlock() to postpone handling the deferred 3226 * map entries, which cannot be done until both old_map and 3227 * new_map locks are released. 3228 */ 3229 sx_xunlock(&old_map->lock); 3230 sx_xunlock(&new_map->lock); 3231 vm_map_process_deferred(); 3232 3233 return (vm2); 3234} 3235 3236int 3237vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3238 vm_prot_t prot, vm_prot_t max, int cow) 3239{ 3240 vm_map_entry_t new_entry, prev_entry; 3241 vm_offset_t bot, top; 3242 vm_size_t growsize, init_ssize; 3243 int orient, rv; 3244 rlim_t lmemlim, vmemlim; 3245 3246 /* 3247 * The stack orientation is piggybacked with the cow argument. 3248 * Extract it into orient and mask the cow argument so that we 3249 * don't pass it around further. 3250 * NOTE: We explicitly allow bi-directional stacks. 3251 */ 3252 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3253 cow &= ~orient; 3254 KASSERT(orient != 0, ("No stack grow direction")); 3255 3256 if (addrbos < vm_map_min(map) || 3257 addrbos > vm_map_max(map) || 3258 addrbos + max_ssize < addrbos) 3259 return (KERN_NO_SPACE); 3260 3261 growsize = sgrowsiz; 3262 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3263 3264 PROC_LOCK(curproc); 3265 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3266 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3267 PROC_UNLOCK(curproc); 3268 3269 vm_map_lock(map); 3270 3271 /* If addr is already mapped, no go */ 3272 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3273 vm_map_unlock(map); 3274 return (KERN_NO_SPACE); 3275 } 3276 3277 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3278 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3279 vm_map_unlock(map); 3280 return (KERN_NO_SPACE); 3281 } 3282 } 3283 3284 /* If we would blow our VMEM resource limit, no go */ 3285 if (map->size + init_ssize > vmemlim) { 3286 vm_map_unlock(map); 3287 return (KERN_NO_SPACE); 3288 } 3289 3290 /* 3291 * If we can't accomodate max_ssize in the current mapping, no go. 3292 * However, we need to be aware that subsequent user mappings might 3293 * map into the space we have reserved for stack, and currently this 3294 * space is not protected. 3295 * 3296 * Hopefully we will at least detect this condition when we try to 3297 * grow the stack. 3298 */ 3299 if ((prev_entry->next != &map->header) && 3300 (prev_entry->next->start < addrbos + max_ssize)) { 3301 vm_map_unlock(map); 3302 return (KERN_NO_SPACE); 3303 } 3304 3305 /* 3306 * We initially map a stack of only init_ssize. We will grow as 3307 * needed later. Depending on the orientation of the stack (i.e. 3308 * the grow direction) we either map at the top of the range, the 3309 * bottom of the range or in the middle. 3310 * 3311 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3312 * and cow to be 0. Possibly we should eliminate these as input 3313 * parameters, and just pass these values here in the insert call. 3314 */ 3315 if (orient == MAP_STACK_GROWS_DOWN) 3316 bot = addrbos + max_ssize - init_ssize; 3317 else if (orient == MAP_STACK_GROWS_UP) 3318 bot = addrbos; 3319 else 3320 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3321 top = bot + init_ssize; 3322 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3323 3324 /* Now set the avail_ssize amount. */ 3325 if (rv == KERN_SUCCESS) { 3326 if (prev_entry != &map->header) 3327 vm_map_clip_end(map, prev_entry, bot); 3328 new_entry = prev_entry->next; 3329 if (new_entry->end != top || new_entry->start != bot) 3330 panic("Bad entry start/end for new stack entry"); 3331 3332 new_entry->avail_ssize = max_ssize - init_ssize; 3333 if (orient & MAP_STACK_GROWS_DOWN) 3334 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3335 if (orient & MAP_STACK_GROWS_UP) 3336 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3337 } 3338 3339 vm_map_unlock(map); 3340 return (rv); 3341} 3342 3343static int stack_guard_page = 0; 3344TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3345SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3346 &stack_guard_page, 0, 3347 "Insert stack guard page ahead of the growable segments."); 3348 3349/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3350 * desired address is already mapped, or if we successfully grow 3351 * the stack. Also returns KERN_SUCCESS if addr is outside the 3352 * stack range (this is strange, but preserves compatibility with 3353 * the grow function in vm_machdep.c). 3354 */ 3355int 3356vm_map_growstack(struct proc *p, vm_offset_t addr) 3357{ 3358 vm_map_entry_t next_entry, prev_entry; 3359 vm_map_entry_t new_entry, stack_entry; 3360 struct vmspace *vm = p->p_vmspace; 3361 vm_map_t map = &vm->vm_map; 3362 vm_offset_t end; 3363 vm_size_t growsize; 3364 size_t grow_amount, max_grow; 3365 rlim_t lmemlim, stacklim, vmemlim; 3366 int is_procstack, rv; 3367 struct ucred *cred; 3368#ifdef notyet 3369 uint64_t limit; 3370#endif 3371#ifdef RACCT 3372 int error; 3373#endif 3374 3375Retry: 3376 PROC_LOCK(p); 3377 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3378 stacklim = lim_cur(p, RLIMIT_STACK); 3379 vmemlim = lim_cur(p, RLIMIT_VMEM); 3380 PROC_UNLOCK(p); 3381 3382 vm_map_lock_read(map); 3383 3384 /* If addr is already in the entry range, no need to grow.*/ 3385 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3386 vm_map_unlock_read(map); 3387 return (KERN_SUCCESS); 3388 } 3389 3390 next_entry = prev_entry->next; 3391 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3392 /* 3393 * This entry does not grow upwards. Since the address lies 3394 * beyond this entry, the next entry (if one exists) has to 3395 * be a downward growable entry. The entry list header is 3396 * never a growable entry, so it suffices to check the flags. 3397 */ 3398 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3399 vm_map_unlock_read(map); 3400 return (KERN_SUCCESS); 3401 } 3402 stack_entry = next_entry; 3403 } else { 3404 /* 3405 * This entry grows upward. If the next entry does not at 3406 * least grow downwards, this is the entry we need to grow. 3407 * otherwise we have two possible choices and we have to 3408 * select one. 3409 */ 3410 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3411 /* 3412 * We have two choices; grow the entry closest to 3413 * the address to minimize the amount of growth. 3414 */ 3415 if (addr - prev_entry->end <= next_entry->start - addr) 3416 stack_entry = prev_entry; 3417 else 3418 stack_entry = next_entry; 3419 } else 3420 stack_entry = prev_entry; 3421 } 3422 3423 if (stack_entry == next_entry) { 3424 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3425 KASSERT(addr < stack_entry->start, ("foo")); 3426 end = (prev_entry != &map->header) ? prev_entry->end : 3427 stack_entry->start - stack_entry->avail_ssize; 3428 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3429 max_grow = stack_entry->start - end; 3430 } else { 3431 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3432 KASSERT(addr >= stack_entry->end, ("foo")); 3433 end = (next_entry != &map->header) ? next_entry->start : 3434 stack_entry->end + stack_entry->avail_ssize; 3435 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3436 max_grow = end - stack_entry->end; 3437 } 3438 3439 if (grow_amount > stack_entry->avail_ssize) { 3440 vm_map_unlock_read(map); 3441 return (KERN_NO_SPACE); 3442 } 3443 3444 /* 3445 * If there is no longer enough space between the entries nogo, and 3446 * adjust the available space. Note: this should only happen if the 3447 * user has mapped into the stack area after the stack was created, 3448 * and is probably an error. 3449 * 3450 * This also effectively destroys any guard page the user might have 3451 * intended by limiting the stack size. 3452 */ 3453 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3454 if (vm_map_lock_upgrade(map)) 3455 goto Retry; 3456 3457 stack_entry->avail_ssize = max_grow; 3458 3459 vm_map_unlock(map); 3460 return (KERN_NO_SPACE); 3461 } 3462 3463 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3464 3465 /* 3466 * If this is the main process stack, see if we're over the stack 3467 * limit. 3468 */ 3469 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3470 vm_map_unlock_read(map); 3471 return (KERN_NO_SPACE); 3472 } 3473#ifdef RACCT 3474 PROC_LOCK(p); 3475 if (is_procstack && 3476 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3477 PROC_UNLOCK(p); 3478 vm_map_unlock_read(map); 3479 return (KERN_NO_SPACE); 3480 } 3481 PROC_UNLOCK(p); 3482#endif 3483 3484 /* Round up the grow amount modulo sgrowsiz */ 3485 growsize = sgrowsiz; 3486 grow_amount = roundup(grow_amount, growsize); 3487 if (grow_amount > stack_entry->avail_ssize) 3488 grow_amount = stack_entry->avail_ssize; 3489 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3490 grow_amount = trunc_page((vm_size_t)stacklim) - 3491 ctob(vm->vm_ssize); 3492 } 3493#ifdef notyet 3494 PROC_LOCK(p); 3495 limit = racct_get_available(p, RACCT_STACK); 3496 PROC_UNLOCK(p); 3497 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3498 grow_amount = limit - ctob(vm->vm_ssize); 3499#endif 3500 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3501 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3502 vm_map_unlock_read(map); 3503 rv = KERN_NO_SPACE; 3504 goto out; 3505 } 3506#ifdef RACCT 3507 PROC_LOCK(p); 3508 if (racct_set(p, RACCT_MEMLOCK, 3509 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3510 PROC_UNLOCK(p); 3511 vm_map_unlock_read(map); 3512 rv = KERN_NO_SPACE; 3513 goto out; 3514 } 3515 PROC_UNLOCK(p); 3516#endif 3517 } 3518 /* If we would blow our VMEM resource limit, no go */ 3519 if (map->size + grow_amount > vmemlim) { 3520 vm_map_unlock_read(map); 3521 rv = KERN_NO_SPACE; 3522 goto out; 3523 } 3524#ifdef RACCT 3525 PROC_LOCK(p); 3526 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3527 PROC_UNLOCK(p); 3528 vm_map_unlock_read(map); 3529 rv = KERN_NO_SPACE; 3530 goto out; 3531 } 3532 PROC_UNLOCK(p); 3533#endif 3534 3535 if (vm_map_lock_upgrade(map)) 3536 goto Retry; 3537 3538 if (stack_entry == next_entry) { 3539 /* 3540 * Growing downward. 3541 */ 3542 /* Get the preliminary new entry start value */ 3543 addr = stack_entry->start - grow_amount; 3544 3545 /* 3546 * If this puts us into the previous entry, cut back our 3547 * growth to the available space. Also, see the note above. 3548 */ 3549 if (addr < end) { 3550 stack_entry->avail_ssize = max_grow; 3551 addr = end; 3552 if (stack_guard_page) 3553 addr += PAGE_SIZE; 3554 } 3555 3556 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3557 next_entry->protection, next_entry->max_protection, 0); 3558 3559 /* Adjust the available stack space by the amount we grew. */ 3560 if (rv == KERN_SUCCESS) { 3561 if (prev_entry != &map->header) 3562 vm_map_clip_end(map, prev_entry, addr); 3563 new_entry = prev_entry->next; 3564 KASSERT(new_entry == stack_entry->prev, ("foo")); 3565 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3566 KASSERT(new_entry->start == addr, ("foo")); 3567 grow_amount = new_entry->end - new_entry->start; 3568 new_entry->avail_ssize = stack_entry->avail_ssize - 3569 grow_amount; 3570 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3571 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3572 } 3573 } else { 3574 /* 3575 * Growing upward. 3576 */ 3577 addr = stack_entry->end + grow_amount; 3578 3579 /* 3580 * If this puts us into the next entry, cut back our growth 3581 * to the available space. Also, see the note above. 3582 */ 3583 if (addr > end) { 3584 stack_entry->avail_ssize = end - stack_entry->end; 3585 addr = end; 3586 if (stack_guard_page) 3587 addr -= PAGE_SIZE; 3588 } 3589 3590 grow_amount = addr - stack_entry->end; 3591 cred = stack_entry->cred; 3592 if (cred == NULL && stack_entry->object.vm_object != NULL) 3593 cred = stack_entry->object.vm_object->cred; 3594 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3595 rv = KERN_NO_SPACE; 3596 /* Grow the underlying object if applicable. */ 3597 else if (stack_entry->object.vm_object == NULL || 3598 vm_object_coalesce(stack_entry->object.vm_object, 3599 stack_entry->offset, 3600 (vm_size_t)(stack_entry->end - stack_entry->start), 3601 (vm_size_t)grow_amount, cred != NULL)) { 3602 map->size += (addr - stack_entry->end); 3603 /* Update the current entry. */ 3604 stack_entry->end = addr; 3605 stack_entry->avail_ssize -= grow_amount; 3606 vm_map_entry_resize_free(map, stack_entry); 3607 rv = KERN_SUCCESS; 3608 3609 if (next_entry != &map->header) 3610 vm_map_clip_start(map, next_entry, addr); 3611 } else 3612 rv = KERN_FAILURE; 3613 } 3614 3615 if (rv == KERN_SUCCESS && is_procstack) 3616 vm->vm_ssize += btoc(grow_amount); 3617 3618 vm_map_unlock(map); 3619 3620 /* 3621 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3622 */ 3623 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3624 vm_map_wire(map, 3625 (stack_entry == next_entry) ? addr : addr - grow_amount, 3626 (stack_entry == next_entry) ? stack_entry->start : addr, 3627 (p->p_flag & P_SYSTEM) 3628 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3629 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3630 } 3631 3632out: 3633#ifdef RACCT 3634 if (rv != KERN_SUCCESS) { 3635 PROC_LOCK(p); 3636 error = racct_set(p, RACCT_VMEM, map->size); 3637 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3638 if (!old_mlock) { 3639 error = racct_set(p, RACCT_MEMLOCK, 3640 ptoa(pmap_wired_count(map->pmap))); 3641 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3642 } 3643 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3644 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3645 PROC_UNLOCK(p); 3646 } 3647#endif 3648 3649 return (rv); 3650} 3651 3652/* 3653 * Unshare the specified VM space for exec. If other processes are 3654 * mapped to it, then create a new one. The new vmspace is null. 3655 */ 3656int 3657vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3658{ 3659 struct vmspace *oldvmspace = p->p_vmspace; 3660 struct vmspace *newvmspace; 3661 3662 newvmspace = vmspace_alloc(minuser, maxuser); 3663 if (newvmspace == NULL) 3664 return (ENOMEM); 3665 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3666 /* 3667 * This code is written like this for prototype purposes. The 3668 * goal is to avoid running down the vmspace here, but let the 3669 * other process's that are still using the vmspace to finally 3670 * run it down. Even though there is little or no chance of blocking 3671 * here, it is a good idea to keep this form for future mods. 3672 */ 3673 PROC_VMSPACE_LOCK(p); 3674 p->p_vmspace = newvmspace; 3675 PROC_VMSPACE_UNLOCK(p); 3676 if (p == curthread->td_proc) 3677 pmap_activate(curthread); 3678 vmspace_free(oldvmspace); 3679 return (0); 3680} 3681 3682/* 3683 * Unshare the specified VM space for forcing COW. This 3684 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3685 */ 3686int 3687vmspace_unshare(struct proc *p) 3688{ 3689 struct vmspace *oldvmspace = p->p_vmspace; 3690 struct vmspace *newvmspace; 3691 vm_ooffset_t fork_charge; 3692 3693 if (oldvmspace->vm_refcnt == 1) 3694 return (0); 3695 fork_charge = 0; 3696 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3697 if (newvmspace == NULL) 3698 return (ENOMEM); 3699 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3700 vmspace_free(newvmspace); 3701 return (ENOMEM); 3702 } 3703 PROC_VMSPACE_LOCK(p); 3704 p->p_vmspace = newvmspace; 3705 PROC_VMSPACE_UNLOCK(p); 3706 if (p == curthread->td_proc) 3707 pmap_activate(curthread); 3708 vmspace_free(oldvmspace); 3709 return (0); 3710} 3711 3712/* 3713 * vm_map_lookup: 3714 * 3715 * Finds the VM object, offset, and 3716 * protection for a given virtual address in the 3717 * specified map, assuming a page fault of the 3718 * type specified. 3719 * 3720 * Leaves the map in question locked for read; return 3721 * values are guaranteed until a vm_map_lookup_done 3722 * call is performed. Note that the map argument 3723 * is in/out; the returned map must be used in 3724 * the call to vm_map_lookup_done. 3725 * 3726 * A handle (out_entry) is returned for use in 3727 * vm_map_lookup_done, to make that fast. 3728 * 3729 * If a lookup is requested with "write protection" 3730 * specified, the map may be changed to perform virtual 3731 * copying operations, although the data referenced will 3732 * remain the same. 3733 */ 3734int 3735vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3736 vm_offset_t vaddr, 3737 vm_prot_t fault_typea, 3738 vm_map_entry_t *out_entry, /* OUT */ 3739 vm_object_t *object, /* OUT */ 3740 vm_pindex_t *pindex, /* OUT */ 3741 vm_prot_t *out_prot, /* OUT */ 3742 boolean_t *wired) /* OUT */ 3743{ 3744 vm_map_entry_t entry; 3745 vm_map_t map = *var_map; 3746 vm_prot_t prot; 3747 vm_prot_t fault_type = fault_typea; 3748 vm_object_t eobject; 3749 vm_size_t size; 3750 struct ucred *cred; 3751 3752RetryLookup:; 3753 3754 vm_map_lock_read(map); 3755 3756 /* 3757 * Lookup the faulting address. 3758 */ 3759 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3760 vm_map_unlock_read(map); 3761 return (KERN_INVALID_ADDRESS); 3762 } 3763 3764 entry = *out_entry; 3765 3766 /* 3767 * Handle submaps. 3768 */ 3769 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3770 vm_map_t old_map = map; 3771 3772 *var_map = map = entry->object.sub_map; 3773 vm_map_unlock_read(old_map); 3774 goto RetryLookup; 3775 } 3776 3777 /* 3778 * Check whether this task is allowed to have this page. 3779 */ 3780 prot = entry->protection; 3781 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3782 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3783 vm_map_unlock_read(map); 3784 return (KERN_PROTECTION_FAILURE); 3785 } 3786 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3787 (entry->eflags & MAP_ENTRY_COW) && 3788 (fault_type & VM_PROT_WRITE)) { 3789 vm_map_unlock_read(map); 3790 return (KERN_PROTECTION_FAILURE); 3791 } 3792 3793 /* 3794 * If this page is not pageable, we have to get it for all possible 3795 * accesses. 3796 */ 3797 *wired = (entry->wired_count != 0); 3798 if (*wired) 3799 fault_type = entry->protection; 3800 size = entry->end - entry->start; 3801 /* 3802 * If the entry was copy-on-write, we either ... 3803 */ 3804 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3805 /* 3806 * If we want to write the page, we may as well handle that 3807 * now since we've got the map locked. 3808 * 3809 * If we don't need to write the page, we just demote the 3810 * permissions allowed. 3811 */ 3812 if ((fault_type & VM_PROT_WRITE) != 0 || 3813 (fault_typea & VM_PROT_COPY) != 0) { 3814 /* 3815 * Make a new object, and place it in the object 3816 * chain. Note that no new references have appeared 3817 * -- one just moved from the map to the new 3818 * object. 3819 */ 3820 if (vm_map_lock_upgrade(map)) 3821 goto RetryLookup; 3822 3823 if (entry->cred == NULL) { 3824 /* 3825 * The debugger owner is charged for 3826 * the memory. 3827 */ 3828 cred = curthread->td_ucred; 3829 crhold(cred); 3830 if (!swap_reserve_by_cred(size, cred)) { 3831 crfree(cred); 3832 vm_map_unlock(map); 3833 return (KERN_RESOURCE_SHORTAGE); 3834 } 3835 entry->cred = cred; 3836 } 3837 vm_object_shadow(&entry->object.vm_object, 3838 &entry->offset, size); 3839 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3840 eobject = entry->object.vm_object; 3841 if (eobject->cred != NULL) { 3842 /* 3843 * The object was not shadowed. 3844 */ 3845 swap_release_by_cred(size, entry->cred); 3846 crfree(entry->cred); 3847 entry->cred = NULL; 3848 } else if (entry->cred != NULL) { 3849 VM_OBJECT_LOCK(eobject); 3850 eobject->cred = entry->cred; 3851 eobject->charge = size; 3852 VM_OBJECT_UNLOCK(eobject); 3853 entry->cred = NULL; 3854 } 3855 3856 vm_map_lock_downgrade(map); 3857 } else { 3858 /* 3859 * We're attempting to read a copy-on-write page -- 3860 * don't allow writes. 3861 */ 3862 prot &= ~VM_PROT_WRITE; 3863 } 3864 } 3865 3866 /* 3867 * Create an object if necessary. 3868 */ 3869 if (entry->object.vm_object == NULL && 3870 !map->system_map) { 3871 if (vm_map_lock_upgrade(map)) 3872 goto RetryLookup; 3873 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3874 atop(size)); 3875 entry->offset = 0; 3876 if (entry->cred != NULL) { 3877 VM_OBJECT_LOCK(entry->object.vm_object); 3878 entry->object.vm_object->cred = entry->cred; 3879 entry->object.vm_object->charge = size; 3880 VM_OBJECT_UNLOCK(entry->object.vm_object); 3881 entry->cred = NULL; 3882 } 3883 vm_map_lock_downgrade(map); 3884 } 3885 3886 /* 3887 * Return the object/offset from this entry. If the entry was 3888 * copy-on-write or empty, it has been fixed up. 3889 */ 3890 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3891 *object = entry->object.vm_object; 3892 3893 *out_prot = prot; 3894 return (KERN_SUCCESS); 3895} 3896 3897/* 3898 * vm_map_lookup_locked: 3899 * 3900 * Lookup the faulting address. A version of vm_map_lookup that returns 3901 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3902 */ 3903int 3904vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3905 vm_offset_t vaddr, 3906 vm_prot_t fault_typea, 3907 vm_map_entry_t *out_entry, /* OUT */ 3908 vm_object_t *object, /* OUT */ 3909 vm_pindex_t *pindex, /* OUT */ 3910 vm_prot_t *out_prot, /* OUT */ 3911 boolean_t *wired) /* OUT */ 3912{ 3913 vm_map_entry_t entry; 3914 vm_map_t map = *var_map; 3915 vm_prot_t prot; 3916 vm_prot_t fault_type = fault_typea; 3917 3918 /* 3919 * Lookup the faulting address. 3920 */ 3921 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3922 return (KERN_INVALID_ADDRESS); 3923 3924 entry = *out_entry; 3925 3926 /* 3927 * Fail if the entry refers to a submap. 3928 */ 3929 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3930 return (KERN_FAILURE); 3931 3932 /* 3933 * Check whether this task is allowed to have this page. 3934 */ 3935 prot = entry->protection; 3936 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3937 if ((fault_type & prot) != fault_type) 3938 return (KERN_PROTECTION_FAILURE); 3939 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3940 (entry->eflags & MAP_ENTRY_COW) && 3941 (fault_type & VM_PROT_WRITE)) 3942 return (KERN_PROTECTION_FAILURE); 3943 3944 /* 3945 * If this page is not pageable, we have to get it for all possible 3946 * accesses. 3947 */ 3948 *wired = (entry->wired_count != 0); 3949 if (*wired) 3950 fault_type = entry->protection; 3951 3952 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3953 /* 3954 * Fail if the entry was copy-on-write for a write fault. 3955 */ 3956 if (fault_type & VM_PROT_WRITE) 3957 return (KERN_FAILURE); 3958 /* 3959 * We're attempting to read a copy-on-write page -- 3960 * don't allow writes. 3961 */ 3962 prot &= ~VM_PROT_WRITE; 3963 } 3964 3965 /* 3966 * Fail if an object should be created. 3967 */ 3968 if (entry->object.vm_object == NULL && !map->system_map) 3969 return (KERN_FAILURE); 3970 3971 /* 3972 * Return the object/offset from this entry. If the entry was 3973 * copy-on-write or empty, it has been fixed up. 3974 */ 3975 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3976 *object = entry->object.vm_object; 3977 3978 *out_prot = prot; 3979 return (KERN_SUCCESS); 3980} 3981 3982/* 3983 * vm_map_lookup_done: 3984 * 3985 * Releases locks acquired by a vm_map_lookup 3986 * (according to the handle returned by that lookup). 3987 */ 3988void 3989vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3990{ 3991 /* 3992 * Unlock the main-level map 3993 */ 3994 vm_map_unlock_read(map); 3995} 3996 3997#include "opt_ddb.h" 3998#ifdef DDB 3999#include <sys/kernel.h> 4000 4001#include <ddb/ddb.h> 4002 4003static void 4004vm_map_print(vm_map_t map) 4005{ 4006 vm_map_entry_t entry; 4007 4008 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4009 (void *)map, 4010 (void *)map->pmap, map->nentries, map->timestamp); 4011 4012 db_indent += 2; 4013 for (entry = map->header.next; entry != &map->header; 4014 entry = entry->next) { 4015 db_iprintf("map entry %p: start=%p, end=%p\n", 4016 (void *)entry, (void *)entry->start, (void *)entry->end); 4017 { 4018 static char *inheritance_name[4] = 4019 {"share", "copy", "none", "donate_copy"}; 4020 4021 db_iprintf(" prot=%x/%x/%s", 4022 entry->protection, 4023 entry->max_protection, 4024 inheritance_name[(int)(unsigned char)entry->inheritance]); 4025 if (entry->wired_count != 0) 4026 db_printf(", wired"); 4027 } 4028 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4029 db_printf(", share=%p, offset=0x%jx\n", 4030 (void *)entry->object.sub_map, 4031 (uintmax_t)entry->offset); 4032 if ((entry->prev == &map->header) || 4033 (entry->prev->object.sub_map != 4034 entry->object.sub_map)) { 4035 db_indent += 2; 4036 vm_map_print((vm_map_t)entry->object.sub_map); 4037 db_indent -= 2; 4038 } 4039 } else { 4040 if (entry->cred != NULL) 4041 db_printf(", ruid %d", entry->cred->cr_ruid); 4042 db_printf(", object=%p, offset=0x%jx", 4043 (void *)entry->object.vm_object, 4044 (uintmax_t)entry->offset); 4045 if (entry->object.vm_object && entry->object.vm_object->cred) 4046 db_printf(", obj ruid %d charge %jx", 4047 entry->object.vm_object->cred->cr_ruid, 4048 (uintmax_t)entry->object.vm_object->charge); 4049 if (entry->eflags & MAP_ENTRY_COW) 4050 db_printf(", copy (%s)", 4051 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4052 db_printf("\n"); 4053 4054 if ((entry->prev == &map->header) || 4055 (entry->prev->object.vm_object != 4056 entry->object.vm_object)) { 4057 db_indent += 2; 4058 vm_object_print((db_expr_t)(intptr_t) 4059 entry->object.vm_object, 4060 1, 0, (char *)0); 4061 db_indent -= 2; 4062 } 4063 } 4064 } 4065 db_indent -= 2; 4066} 4067 4068DB_SHOW_COMMAND(map, map) 4069{ 4070 4071 if (!have_addr) { 4072 db_printf("usage: show map <addr>\n"); 4073 return; 4074 } 4075 vm_map_print((vm_map_t)addr); 4076} 4077 4078DB_SHOW_COMMAND(procvm, procvm) 4079{ 4080 struct proc *p; 4081 4082 if (have_addr) { 4083 p = (struct proc *) addr; 4084 } else { 4085 p = curproc; 4086 } 4087 4088 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4089 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4090 (void *)vmspace_pmap(p->p_vmspace)); 4091 4092 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4093} 4094 4095#endif /* DDB */
| 306 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 + 307 maxproc * 2 + maxfiles); 308 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 309#ifdef INVARIANTS 310 vmspace_zdtor, 311#else 312 NULL, 313#endif 314 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 315} 316 317static void 318vmspace_container_reset(struct proc *p) 319{ 320 321#ifdef RACCT 322 PROC_LOCK(p); 323 racct_set(p, RACCT_DATA, 0); 324 racct_set(p, RACCT_STACK, 0); 325 racct_set(p, RACCT_RSS, 0); 326 racct_set(p, RACCT_MEMLOCK, 0); 327 racct_set(p, RACCT_VMEM, 0); 328 PROC_UNLOCK(p); 329#endif 330} 331 332static inline void 333vmspace_dofree(struct vmspace *vm) 334{ 335 336 CTR1(KTR_VM, "vmspace_free: %p", vm); 337 338 /* 339 * Make sure any SysV shm is freed, it might not have been in 340 * exit1(). 341 */ 342 shmexit(vm); 343 344 /* 345 * Lock the map, to wait out all other references to it. 346 * Delete all of the mappings and pages they hold, then call 347 * the pmap module to reclaim anything left. 348 */ 349 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 350 vm->vm_map.max_offset); 351 352 pmap_release(vmspace_pmap(vm)); 353 vm->vm_map.pmap = NULL; 354 uma_zfree(vmspace_zone, vm); 355} 356 357void 358vmspace_free(struct vmspace *vm) 359{ 360 361 if (vm->vm_refcnt == 0) 362 panic("vmspace_free: attempt to free already freed vmspace"); 363 364 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 365 vmspace_dofree(vm); 366} 367 368void 369vmspace_exitfree(struct proc *p) 370{ 371 struct vmspace *vm; 372 373 PROC_VMSPACE_LOCK(p); 374 vm = p->p_vmspace; 375 p->p_vmspace = NULL; 376 PROC_VMSPACE_UNLOCK(p); 377 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 378 vmspace_free(vm); 379} 380 381void 382vmspace_exit(struct thread *td) 383{ 384 int refcnt; 385 struct vmspace *vm; 386 struct proc *p; 387 388 /* 389 * Release user portion of address space. 390 * This releases references to vnodes, 391 * which could cause I/O if the file has been unlinked. 392 * Need to do this early enough that we can still sleep. 393 * 394 * The last exiting process to reach this point releases as 395 * much of the environment as it can. vmspace_dofree() is the 396 * slower fallback in case another process had a temporary 397 * reference to the vmspace. 398 */ 399 400 p = td->td_proc; 401 vm = p->p_vmspace; 402 atomic_add_int(&vmspace0.vm_refcnt, 1); 403 do { 404 refcnt = vm->vm_refcnt; 405 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 406 /* Switch now since other proc might free vmspace */ 407 PROC_VMSPACE_LOCK(p); 408 p->p_vmspace = &vmspace0; 409 PROC_VMSPACE_UNLOCK(p); 410 pmap_activate(td); 411 } 412 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 413 if (refcnt == 1) { 414 if (p->p_vmspace != vm) { 415 /* vmspace not yet freed, switch back */ 416 PROC_VMSPACE_LOCK(p); 417 p->p_vmspace = vm; 418 PROC_VMSPACE_UNLOCK(p); 419 pmap_activate(td); 420 } 421 pmap_remove_pages(vmspace_pmap(vm)); 422 /* Switch now since this proc will free vmspace */ 423 PROC_VMSPACE_LOCK(p); 424 p->p_vmspace = &vmspace0; 425 PROC_VMSPACE_UNLOCK(p); 426 pmap_activate(td); 427 vmspace_dofree(vm); 428 } 429 vmspace_container_reset(p); 430} 431 432/* Acquire reference to vmspace owned by another process. */ 433 434struct vmspace * 435vmspace_acquire_ref(struct proc *p) 436{ 437 struct vmspace *vm; 438 int refcnt; 439 440 PROC_VMSPACE_LOCK(p); 441 vm = p->p_vmspace; 442 if (vm == NULL) { 443 PROC_VMSPACE_UNLOCK(p); 444 return (NULL); 445 } 446 do { 447 refcnt = vm->vm_refcnt; 448 if (refcnt <= 0) { /* Avoid 0->1 transition */ 449 PROC_VMSPACE_UNLOCK(p); 450 return (NULL); 451 } 452 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 453 if (vm != p->p_vmspace) { 454 PROC_VMSPACE_UNLOCK(p); 455 vmspace_free(vm); 456 return (NULL); 457 } 458 PROC_VMSPACE_UNLOCK(p); 459 return (vm); 460} 461 462void 463_vm_map_lock(vm_map_t map, const char *file, int line) 464{ 465 466 if (map->system_map) 467 mtx_lock_flags_(&map->system_mtx, 0, file, line); 468 else 469 sx_xlock_(&map->lock, file, line); 470 map->timestamp++; 471} 472 473static void 474vm_map_process_deferred(void) 475{ 476 struct thread *td; 477 vm_map_entry_t entry, next; 478 vm_object_t object; 479 480 td = curthread; 481 entry = td->td_map_def_user; 482 td->td_map_def_user = NULL; 483 while (entry != NULL) { 484 next = entry->next; 485 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 486 /* 487 * Decrement the object's writemappings and 488 * possibly the vnode's v_writecount. 489 */ 490 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 491 ("Submap with writecount")); 492 object = entry->object.vm_object; 493 KASSERT(object != NULL, ("No object for writecount")); 494 vnode_pager_release_writecount(object, entry->start, 495 entry->end); 496 } 497 vm_map_entry_deallocate(entry, FALSE); 498 entry = next; 499 } 500} 501 502void 503_vm_map_unlock(vm_map_t map, const char *file, int line) 504{ 505 506 if (map->system_map) 507 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 508 else { 509 sx_xunlock_(&map->lock, file, line); 510 vm_map_process_deferred(); 511 } 512} 513 514void 515_vm_map_lock_read(vm_map_t map, const char *file, int line) 516{ 517 518 if (map->system_map) 519 mtx_lock_flags_(&map->system_mtx, 0, file, line); 520 else 521 sx_slock_(&map->lock, file, line); 522} 523 524void 525_vm_map_unlock_read(vm_map_t map, const char *file, int line) 526{ 527 528 if (map->system_map) 529 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 530 else { 531 sx_sunlock_(&map->lock, file, line); 532 vm_map_process_deferred(); 533 } 534} 535 536int 537_vm_map_trylock(vm_map_t map, const char *file, int line) 538{ 539 int error; 540 541 error = map->system_map ? 542 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 543 !sx_try_xlock_(&map->lock, file, line); 544 if (error == 0) 545 map->timestamp++; 546 return (error == 0); 547} 548 549int 550_vm_map_trylock_read(vm_map_t map, const char *file, int line) 551{ 552 int error; 553 554 error = map->system_map ? 555 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 556 !sx_try_slock_(&map->lock, file, line); 557 return (error == 0); 558} 559 560/* 561 * _vm_map_lock_upgrade: [ internal use only ] 562 * 563 * Tries to upgrade a read (shared) lock on the specified map to a write 564 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 565 * non-zero value if the upgrade fails. If the upgrade fails, the map is 566 * returned without a read or write lock held. 567 * 568 * Requires that the map be read locked. 569 */ 570int 571_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 572{ 573 unsigned int last_timestamp; 574 575 if (map->system_map) { 576 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 577 } else { 578 if (!sx_try_upgrade_(&map->lock, file, line)) { 579 last_timestamp = map->timestamp; 580 sx_sunlock_(&map->lock, file, line); 581 vm_map_process_deferred(); 582 /* 583 * If the map's timestamp does not change while the 584 * map is unlocked, then the upgrade succeeds. 585 */ 586 sx_xlock_(&map->lock, file, line); 587 if (last_timestamp != map->timestamp) { 588 sx_xunlock_(&map->lock, file, line); 589 return (1); 590 } 591 } 592 } 593 map->timestamp++; 594 return (0); 595} 596 597void 598_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 599{ 600 601 if (map->system_map) { 602 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 603 } else 604 sx_downgrade_(&map->lock, file, line); 605} 606 607/* 608 * vm_map_locked: 609 * 610 * Returns a non-zero value if the caller holds a write (exclusive) lock 611 * on the specified map and the value "0" otherwise. 612 */ 613int 614vm_map_locked(vm_map_t map) 615{ 616 617 if (map->system_map) 618 return (mtx_owned(&map->system_mtx)); 619 else 620 return (sx_xlocked(&map->lock)); 621} 622 623#ifdef INVARIANTS 624static void 625_vm_map_assert_locked(vm_map_t map, const char *file, int line) 626{ 627 628 if (map->system_map) 629 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 630 else 631 sx_assert_(&map->lock, SA_XLOCKED, file, line); 632} 633 634#define VM_MAP_ASSERT_LOCKED(map) \ 635 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 636#else 637#define VM_MAP_ASSERT_LOCKED(map) 638#endif 639 640/* 641 * _vm_map_unlock_and_wait: 642 * 643 * Atomically releases the lock on the specified map and puts the calling 644 * thread to sleep. The calling thread will remain asleep until either 645 * vm_map_wakeup() is performed on the map or the specified timeout is 646 * exceeded. 647 * 648 * WARNING! This function does not perform deferred deallocations of 649 * objects and map entries. Therefore, the calling thread is expected to 650 * reacquire the map lock after reawakening and later perform an ordinary 651 * unlock operation, such as vm_map_unlock(), before completing its 652 * operation on the map. 653 */ 654int 655_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 656{ 657 658 mtx_lock(&map_sleep_mtx); 659 if (map->system_map) 660 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 661 else 662 sx_xunlock_(&map->lock, file, line); 663 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 664 timo)); 665} 666 667/* 668 * vm_map_wakeup: 669 * 670 * Awaken any threads that have slept on the map using 671 * vm_map_unlock_and_wait(). 672 */ 673void 674vm_map_wakeup(vm_map_t map) 675{ 676 677 /* 678 * Acquire and release map_sleep_mtx to prevent a wakeup() 679 * from being performed (and lost) between the map unlock 680 * and the msleep() in _vm_map_unlock_and_wait(). 681 */ 682 mtx_lock(&map_sleep_mtx); 683 mtx_unlock(&map_sleep_mtx); 684 wakeup(&map->root); 685} 686 687void 688vm_map_busy(vm_map_t map) 689{ 690 691 VM_MAP_ASSERT_LOCKED(map); 692 map->busy++; 693} 694 695void 696vm_map_unbusy(vm_map_t map) 697{ 698 699 VM_MAP_ASSERT_LOCKED(map); 700 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 701 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 702 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 703 wakeup(&map->busy); 704 } 705} 706 707void 708vm_map_wait_busy(vm_map_t map) 709{ 710 711 VM_MAP_ASSERT_LOCKED(map); 712 while (map->busy) { 713 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 714 if (map->system_map) 715 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 716 else 717 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 718 } 719 map->timestamp++; 720} 721 722long 723vmspace_resident_count(struct vmspace *vmspace) 724{ 725 return pmap_resident_count(vmspace_pmap(vmspace)); 726} 727 728/* 729 * vm_map_create: 730 * 731 * Creates and returns a new empty VM map with 732 * the given physical map structure, and having 733 * the given lower and upper address bounds. 734 */ 735vm_map_t 736vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 737{ 738 vm_map_t result; 739 740 result = uma_zalloc(mapzone, M_WAITOK); 741 CTR1(KTR_VM, "vm_map_create: %p", result); 742 _vm_map_init(result, pmap, min, max); 743 return (result); 744} 745 746/* 747 * Initialize an existing vm_map structure 748 * such as that in the vmspace structure. 749 */ 750static void 751_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 752{ 753 754 map->header.next = map->header.prev = &map->header; 755 map->needs_wakeup = FALSE; 756 map->system_map = 0; 757 map->pmap = pmap; 758 map->min_offset = min; 759 map->max_offset = max; 760 map->flags = 0; 761 map->root = NULL; 762 map->timestamp = 0; 763 map->busy = 0; 764} 765 766void 767vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 768{ 769 770 _vm_map_init(map, pmap, min, max); 771 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 772 sx_init(&map->lock, "user map"); 773} 774 775/* 776 * vm_map_entry_dispose: [ internal use only ] 777 * 778 * Inverse of vm_map_entry_create. 779 */ 780static void 781vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 782{ 783 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 784} 785 786/* 787 * vm_map_entry_create: [ internal use only ] 788 * 789 * Allocates a VM map entry for insertion. 790 * No entry fields are filled in. 791 */ 792static vm_map_entry_t 793vm_map_entry_create(vm_map_t map) 794{ 795 vm_map_entry_t new_entry; 796 797 if (map->system_map) 798 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 799 else 800 new_entry = uma_zalloc(mapentzone, M_WAITOK); 801 if (new_entry == NULL) 802 panic("vm_map_entry_create: kernel resources exhausted"); 803 return (new_entry); 804} 805 806/* 807 * vm_map_entry_set_behavior: 808 * 809 * Set the expected access behavior, either normal, random, or 810 * sequential. 811 */ 812static inline void 813vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 814{ 815 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 816 (behavior & MAP_ENTRY_BEHAV_MASK); 817} 818 819/* 820 * vm_map_entry_set_max_free: 821 * 822 * Set the max_free field in a vm_map_entry. 823 */ 824static inline void 825vm_map_entry_set_max_free(vm_map_entry_t entry) 826{ 827 828 entry->max_free = entry->adj_free; 829 if (entry->left != NULL && entry->left->max_free > entry->max_free) 830 entry->max_free = entry->left->max_free; 831 if (entry->right != NULL && entry->right->max_free > entry->max_free) 832 entry->max_free = entry->right->max_free; 833} 834 835/* 836 * vm_map_entry_splay: 837 * 838 * The Sleator and Tarjan top-down splay algorithm with the 839 * following variation. Max_free must be computed bottom-up, so 840 * on the downward pass, maintain the left and right spines in 841 * reverse order. Then, make a second pass up each side to fix 842 * the pointers and compute max_free. The time bound is O(log n) 843 * amortized. 844 * 845 * The new root is the vm_map_entry containing "addr", or else an 846 * adjacent entry (lower or higher) if addr is not in the tree. 847 * 848 * The map must be locked, and leaves it so. 849 * 850 * Returns: the new root. 851 */ 852static vm_map_entry_t 853vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 854{ 855 vm_map_entry_t llist, rlist; 856 vm_map_entry_t ltree, rtree; 857 vm_map_entry_t y; 858 859 /* Special case of empty tree. */ 860 if (root == NULL) 861 return (root); 862 863 /* 864 * Pass One: Splay down the tree until we find addr or a NULL 865 * pointer where addr would go. llist and rlist are the two 866 * sides in reverse order (bottom-up), with llist linked by 867 * the right pointer and rlist linked by the left pointer in 868 * the vm_map_entry. Wait until Pass Two to set max_free on 869 * the two spines. 870 */ 871 llist = NULL; 872 rlist = NULL; 873 for (;;) { 874 /* root is never NULL in here. */ 875 if (addr < root->start) { 876 y = root->left; 877 if (y == NULL) 878 break; 879 if (addr < y->start && y->left != NULL) { 880 /* Rotate right and put y on rlist. */ 881 root->left = y->right; 882 y->right = root; 883 vm_map_entry_set_max_free(root); 884 root = y->left; 885 y->left = rlist; 886 rlist = y; 887 } else { 888 /* Put root on rlist. */ 889 root->left = rlist; 890 rlist = root; 891 root = y; 892 } 893 } else if (addr >= root->end) { 894 y = root->right; 895 if (y == NULL) 896 break; 897 if (addr >= y->end && y->right != NULL) { 898 /* Rotate left and put y on llist. */ 899 root->right = y->left; 900 y->left = root; 901 vm_map_entry_set_max_free(root); 902 root = y->right; 903 y->right = llist; 904 llist = y; 905 } else { 906 /* Put root on llist. */ 907 root->right = llist; 908 llist = root; 909 root = y; 910 } 911 } else 912 break; 913 } 914 915 /* 916 * Pass Two: Walk back up the two spines, flip the pointers 917 * and set max_free. The subtrees of the root go at the 918 * bottom of llist and rlist. 919 */ 920 ltree = root->left; 921 while (llist != NULL) { 922 y = llist->right; 923 llist->right = ltree; 924 vm_map_entry_set_max_free(llist); 925 ltree = llist; 926 llist = y; 927 } 928 rtree = root->right; 929 while (rlist != NULL) { 930 y = rlist->left; 931 rlist->left = rtree; 932 vm_map_entry_set_max_free(rlist); 933 rtree = rlist; 934 rlist = y; 935 } 936 937 /* 938 * Final assembly: add ltree and rtree as subtrees of root. 939 */ 940 root->left = ltree; 941 root->right = rtree; 942 vm_map_entry_set_max_free(root); 943 944 return (root); 945} 946 947/* 948 * vm_map_entry_{un,}link: 949 * 950 * Insert/remove entries from maps. 951 */ 952static void 953vm_map_entry_link(vm_map_t map, 954 vm_map_entry_t after_where, 955 vm_map_entry_t entry) 956{ 957 958 CTR4(KTR_VM, 959 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 960 map->nentries, entry, after_where); 961 VM_MAP_ASSERT_LOCKED(map); 962 map->nentries++; 963 entry->prev = after_where; 964 entry->next = after_where->next; 965 entry->next->prev = entry; 966 after_where->next = entry; 967 968 if (after_where != &map->header) { 969 if (after_where != map->root) 970 vm_map_entry_splay(after_where->start, map->root); 971 entry->right = after_where->right; 972 entry->left = after_where; 973 after_where->right = NULL; 974 after_where->adj_free = entry->start - after_where->end; 975 vm_map_entry_set_max_free(after_where); 976 } else { 977 entry->right = map->root; 978 entry->left = NULL; 979 } 980 entry->adj_free = (entry->next == &map->header ? map->max_offset : 981 entry->next->start) - entry->end; 982 vm_map_entry_set_max_free(entry); 983 map->root = entry; 984} 985 986static void 987vm_map_entry_unlink(vm_map_t map, 988 vm_map_entry_t entry) 989{ 990 vm_map_entry_t next, prev, root; 991 992 VM_MAP_ASSERT_LOCKED(map); 993 if (entry != map->root) 994 vm_map_entry_splay(entry->start, map->root); 995 if (entry->left == NULL) 996 root = entry->right; 997 else { 998 root = vm_map_entry_splay(entry->start, entry->left); 999 root->right = entry->right; 1000 root->adj_free = (entry->next == &map->header ? map->max_offset : 1001 entry->next->start) - root->end; 1002 vm_map_entry_set_max_free(root); 1003 } 1004 map->root = root; 1005 1006 prev = entry->prev; 1007 next = entry->next; 1008 next->prev = prev; 1009 prev->next = next; 1010 map->nentries--; 1011 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1012 map->nentries, entry); 1013} 1014 1015/* 1016 * vm_map_entry_resize_free: 1017 * 1018 * Recompute the amount of free space following a vm_map_entry 1019 * and propagate that value up the tree. Call this function after 1020 * resizing a map entry in-place, that is, without a call to 1021 * vm_map_entry_link() or _unlink(). 1022 * 1023 * The map must be locked, and leaves it so. 1024 */ 1025static void 1026vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1027{ 1028 1029 /* 1030 * Using splay trees without parent pointers, propagating 1031 * max_free up the tree is done by moving the entry to the 1032 * root and making the change there. 1033 */ 1034 if (entry != map->root) 1035 map->root = vm_map_entry_splay(entry->start, map->root); 1036 1037 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1038 entry->next->start) - entry->end; 1039 vm_map_entry_set_max_free(entry); 1040} 1041 1042/* 1043 * vm_map_lookup_entry: [ internal use only ] 1044 * 1045 * Finds the map entry containing (or 1046 * immediately preceding) the specified address 1047 * in the given map; the entry is returned 1048 * in the "entry" parameter. The boolean 1049 * result indicates whether the address is 1050 * actually contained in the map. 1051 */ 1052boolean_t 1053vm_map_lookup_entry( 1054 vm_map_t map, 1055 vm_offset_t address, 1056 vm_map_entry_t *entry) /* OUT */ 1057{ 1058 vm_map_entry_t cur; 1059 boolean_t locked; 1060 1061 /* 1062 * If the map is empty, then the map entry immediately preceding 1063 * "address" is the map's header. 1064 */ 1065 cur = map->root; 1066 if (cur == NULL) 1067 *entry = &map->header; 1068 else if (address >= cur->start && cur->end > address) { 1069 *entry = cur; 1070 return (TRUE); 1071 } else if ((locked = vm_map_locked(map)) || 1072 sx_try_upgrade(&map->lock)) { 1073 /* 1074 * Splay requires a write lock on the map. However, it only 1075 * restructures the binary search tree; it does not otherwise 1076 * change the map. Thus, the map's timestamp need not change 1077 * on a temporary upgrade. 1078 */ 1079 map->root = cur = vm_map_entry_splay(address, cur); 1080 if (!locked) 1081 sx_downgrade(&map->lock); 1082 1083 /* 1084 * If "address" is contained within a map entry, the new root 1085 * is that map entry. Otherwise, the new root is a map entry 1086 * immediately before or after "address". 1087 */ 1088 if (address >= cur->start) { 1089 *entry = cur; 1090 if (cur->end > address) 1091 return (TRUE); 1092 } else 1093 *entry = cur->prev; 1094 } else 1095 /* 1096 * Since the map is only locked for read access, perform a 1097 * standard binary search tree lookup for "address". 1098 */ 1099 for (;;) { 1100 if (address < cur->start) { 1101 if (cur->left == NULL) { 1102 *entry = cur->prev; 1103 break; 1104 } 1105 cur = cur->left; 1106 } else if (cur->end > address) { 1107 *entry = cur; 1108 return (TRUE); 1109 } else { 1110 if (cur->right == NULL) { 1111 *entry = cur; 1112 break; 1113 } 1114 cur = cur->right; 1115 } 1116 } 1117 return (FALSE); 1118} 1119 1120/* 1121 * vm_map_insert: 1122 * 1123 * Inserts the given whole VM object into the target 1124 * map at the specified address range. The object's 1125 * size should match that of the address range. 1126 * 1127 * Requires that the map be locked, and leaves it so. 1128 * 1129 * If object is non-NULL, ref count must be bumped by caller 1130 * prior to making call to account for the new entry. 1131 */ 1132int 1133vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1134 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1135 int cow) 1136{ 1137 vm_map_entry_t new_entry; 1138 vm_map_entry_t prev_entry; 1139 vm_map_entry_t temp_entry; 1140 vm_eflags_t protoeflags; 1141 struct ucred *cred; 1142 vm_inherit_t inheritance; 1143 boolean_t charge_prev_obj; 1144 1145 VM_MAP_ASSERT_LOCKED(map); 1146 1147 /* 1148 * Check that the start and end points are not bogus. 1149 */ 1150 if ((start < map->min_offset) || (end > map->max_offset) || 1151 (start >= end)) 1152 return (KERN_INVALID_ADDRESS); 1153 1154 /* 1155 * Find the entry prior to the proposed starting address; if it's part 1156 * of an existing entry, this range is bogus. 1157 */ 1158 if (vm_map_lookup_entry(map, start, &temp_entry)) 1159 return (KERN_NO_SPACE); 1160 1161 prev_entry = temp_entry; 1162 1163 /* 1164 * Assert that the next entry doesn't overlap the end point. 1165 */ 1166 if ((prev_entry->next != &map->header) && 1167 (prev_entry->next->start < end)) 1168 return (KERN_NO_SPACE); 1169 1170 protoeflags = 0; 1171 charge_prev_obj = FALSE; 1172 1173 if (cow & MAP_COPY_ON_WRITE) 1174 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1175 1176 if (cow & MAP_NOFAULT) { 1177 protoeflags |= MAP_ENTRY_NOFAULT; 1178 1179 KASSERT(object == NULL, 1180 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1181 } 1182 if (cow & MAP_DISABLE_SYNCER) 1183 protoeflags |= MAP_ENTRY_NOSYNC; 1184 if (cow & MAP_DISABLE_COREDUMP) 1185 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1186 if (cow & MAP_VN_WRITECOUNT) 1187 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1188 if (cow & MAP_INHERIT_SHARE) 1189 inheritance = VM_INHERIT_SHARE; 1190 else 1191 inheritance = VM_INHERIT_DEFAULT; 1192 1193 cred = NULL; 1194 KASSERT((object != kmem_object && object != kernel_object) || 1195 ((object == kmem_object || object == kernel_object) && 1196 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1197 ("kmem or kernel object and cow")); 1198 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1199 goto charged; 1200 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1201 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1202 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1203 return (KERN_RESOURCE_SHORTAGE); 1204 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1205 object->cred == NULL, 1206 ("OVERCOMMIT: vm_map_insert o %p", object)); 1207 cred = curthread->td_ucred; 1208 crhold(cred); 1209 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1210 charge_prev_obj = TRUE; 1211 } 1212 1213charged: 1214 /* Expand the kernel pmap, if necessary. */ 1215 if (map == kernel_map && end > kernel_vm_end) 1216 pmap_growkernel(end); 1217 if (object != NULL) { 1218 /* 1219 * OBJ_ONEMAPPING must be cleared unless this mapping 1220 * is trivially proven to be the only mapping for any 1221 * of the object's pages. (Object granularity 1222 * reference counting is insufficient to recognize 1223 * aliases with precision.) 1224 */ 1225 VM_OBJECT_LOCK(object); 1226 if (object->ref_count > 1 || object->shadow_count != 0) 1227 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1228 VM_OBJECT_UNLOCK(object); 1229 } 1230 else if ((prev_entry != &map->header) && 1231 (prev_entry->eflags == protoeflags) && 1232 (prev_entry->end == start) && 1233 (prev_entry->wired_count == 0) && 1234 (prev_entry->cred == cred || 1235 (prev_entry->object.vm_object != NULL && 1236 (prev_entry->object.vm_object->cred == cred))) && 1237 vm_object_coalesce(prev_entry->object.vm_object, 1238 prev_entry->offset, 1239 (vm_size_t)(prev_entry->end - prev_entry->start), 1240 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1241 /* 1242 * We were able to extend the object. Determine if we 1243 * can extend the previous map entry to include the 1244 * new range as well. 1245 */ 1246 if ((prev_entry->inheritance == inheritance) && 1247 (prev_entry->protection == prot) && 1248 (prev_entry->max_protection == max)) { 1249 map->size += (end - prev_entry->end); 1250 prev_entry->end = end; 1251 vm_map_entry_resize_free(map, prev_entry); 1252 vm_map_simplify_entry(map, prev_entry); 1253 if (cred != NULL) 1254 crfree(cred); 1255 return (KERN_SUCCESS); 1256 } 1257 1258 /* 1259 * If we can extend the object but cannot extend the 1260 * map entry, we have to create a new map entry. We 1261 * must bump the ref count on the extended object to 1262 * account for it. object may be NULL. 1263 */ 1264 object = prev_entry->object.vm_object; 1265 offset = prev_entry->offset + 1266 (prev_entry->end - prev_entry->start); 1267 vm_object_reference(object); 1268 if (cred != NULL && object != NULL && object->cred != NULL && 1269 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1270 /* Object already accounts for this uid. */ 1271 crfree(cred); 1272 cred = NULL; 1273 } 1274 } 1275 1276 /* 1277 * NOTE: if conditionals fail, object can be NULL here. This occurs 1278 * in things like the buffer map where we manage kva but do not manage 1279 * backing objects. 1280 */ 1281 1282 /* 1283 * Create a new entry 1284 */ 1285 new_entry = vm_map_entry_create(map); 1286 new_entry->start = start; 1287 new_entry->end = end; 1288 new_entry->cred = NULL; 1289 1290 new_entry->eflags = protoeflags; 1291 new_entry->object.vm_object = object; 1292 new_entry->offset = offset; 1293 new_entry->avail_ssize = 0; 1294 1295 new_entry->inheritance = inheritance; 1296 new_entry->protection = prot; 1297 new_entry->max_protection = max; 1298 new_entry->wired_count = 0; 1299 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1300 new_entry->next_read = OFF_TO_IDX(offset); 1301 1302 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1303 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1304 new_entry->cred = cred; 1305 1306 /* 1307 * Insert the new entry into the list 1308 */ 1309 vm_map_entry_link(map, prev_entry, new_entry); 1310 map->size += new_entry->end - new_entry->start; 1311 1312 /* 1313 * It may be possible to merge the new entry with the next and/or 1314 * previous entries. However, due to MAP_STACK_* being a hack, a 1315 * panic can result from merging such entries. 1316 */ 1317 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1318 vm_map_simplify_entry(map, new_entry); 1319 1320 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1321 vm_map_pmap_enter(map, start, prot, 1322 object, OFF_TO_IDX(offset), end - start, 1323 cow & MAP_PREFAULT_PARTIAL); 1324 } 1325 1326 return (KERN_SUCCESS); 1327} 1328 1329/* 1330 * vm_map_findspace: 1331 * 1332 * Find the first fit (lowest VM address) for "length" free bytes 1333 * beginning at address >= start in the given map. 1334 * 1335 * In a vm_map_entry, "adj_free" is the amount of free space 1336 * adjacent (higher address) to this entry, and "max_free" is the 1337 * maximum amount of contiguous free space in its subtree. This 1338 * allows finding a free region in one path down the tree, so 1339 * O(log n) amortized with splay trees. 1340 * 1341 * The map must be locked, and leaves it so. 1342 * 1343 * Returns: 0 on success, and starting address in *addr, 1344 * 1 if insufficient space. 1345 */ 1346int 1347vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1348 vm_offset_t *addr) /* OUT */ 1349{ 1350 vm_map_entry_t entry; 1351 vm_offset_t st; 1352 1353 /* 1354 * Request must fit within min/max VM address and must avoid 1355 * address wrap. 1356 */ 1357 if (start < map->min_offset) 1358 start = map->min_offset; 1359 if (start + length > map->max_offset || start + length < start) 1360 return (1); 1361 1362 /* Empty tree means wide open address space. */ 1363 if (map->root == NULL) { 1364 *addr = start; 1365 return (0); 1366 } 1367 1368 /* 1369 * After splay, if start comes before root node, then there 1370 * must be a gap from start to the root. 1371 */ 1372 map->root = vm_map_entry_splay(start, map->root); 1373 if (start + length <= map->root->start) { 1374 *addr = start; 1375 return (0); 1376 } 1377 1378 /* 1379 * Root is the last node that might begin its gap before 1380 * start, and this is the last comparison where address 1381 * wrap might be a problem. 1382 */ 1383 st = (start > map->root->end) ? start : map->root->end; 1384 if (length <= map->root->end + map->root->adj_free - st) { 1385 *addr = st; 1386 return (0); 1387 } 1388 1389 /* With max_free, can immediately tell if no solution. */ 1390 entry = map->root->right; 1391 if (entry == NULL || length > entry->max_free) 1392 return (1); 1393 1394 /* 1395 * Search the right subtree in the order: left subtree, root, 1396 * right subtree (first fit). The previous splay implies that 1397 * all regions in the right subtree have addresses > start. 1398 */ 1399 while (entry != NULL) { 1400 if (entry->left != NULL && entry->left->max_free >= length) 1401 entry = entry->left; 1402 else if (entry->adj_free >= length) { 1403 *addr = entry->end; 1404 return (0); 1405 } else 1406 entry = entry->right; 1407 } 1408 1409 /* Can't get here, so panic if we do. */ 1410 panic("vm_map_findspace: max_free corrupt"); 1411} 1412 1413int 1414vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1415 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1416 vm_prot_t max, int cow) 1417{ 1418 vm_offset_t end; 1419 int result; 1420 1421 end = start + length; 1422 vm_map_lock(map); 1423 VM_MAP_RANGE_CHECK(map, start, end); 1424 (void) vm_map_delete(map, start, end); 1425 result = vm_map_insert(map, object, offset, start, end, prot, 1426 max, cow); 1427 vm_map_unlock(map); 1428 return (result); 1429} 1430 1431/* 1432 * vm_map_find finds an unallocated region in the target address 1433 * map with the given length. The search is defined to be 1434 * first-fit from the specified address; the region found is 1435 * returned in the same parameter. 1436 * 1437 * If object is non-NULL, ref count must be bumped by caller 1438 * prior to making call to account for the new entry. 1439 */ 1440int 1441vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1442 vm_offset_t *addr, /* IN/OUT */ 1443 vm_size_t length, int find_space, vm_prot_t prot, 1444 vm_prot_t max, int cow) 1445{ 1446 vm_offset_t start; 1447 int result; 1448 1449 start = *addr; 1450 vm_map_lock(map); 1451 do { 1452 if (find_space != VMFS_NO_SPACE) { 1453 if (vm_map_findspace(map, start, length, addr)) { 1454 vm_map_unlock(map); 1455 return (KERN_NO_SPACE); 1456 } 1457 switch (find_space) { 1458 case VMFS_ALIGNED_SPACE: 1459 pmap_align_superpage(object, offset, addr, 1460 length); 1461 break; 1462#ifdef VMFS_TLB_ALIGNED_SPACE 1463 case VMFS_TLB_ALIGNED_SPACE: 1464 pmap_align_tlb(addr); 1465 break; 1466#endif 1467 default: 1468 break; 1469 } 1470 1471 start = *addr; 1472 } 1473 result = vm_map_insert(map, object, offset, start, start + 1474 length, prot, max, cow); 1475 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE 1476#ifdef VMFS_TLB_ALIGNED_SPACE 1477 || find_space == VMFS_TLB_ALIGNED_SPACE 1478#endif 1479 )); 1480 vm_map_unlock(map); 1481 return (result); 1482} 1483 1484/* 1485 * vm_map_simplify_entry: 1486 * 1487 * Simplify the given map entry by merging with either neighbor. This 1488 * routine also has the ability to merge with both neighbors. 1489 * 1490 * The map must be locked. 1491 * 1492 * This routine guarentees that the passed entry remains valid (though 1493 * possibly extended). When merging, this routine may delete one or 1494 * both neighbors. 1495 */ 1496void 1497vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1498{ 1499 vm_map_entry_t next, prev; 1500 vm_size_t prevsize, esize; 1501 1502 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1503 return; 1504 1505 prev = entry->prev; 1506 if (prev != &map->header) { 1507 prevsize = prev->end - prev->start; 1508 if ( (prev->end == entry->start) && 1509 (prev->object.vm_object == entry->object.vm_object) && 1510 (!prev->object.vm_object || 1511 (prev->offset + prevsize == entry->offset)) && 1512 (prev->eflags == entry->eflags) && 1513 (prev->protection == entry->protection) && 1514 (prev->max_protection == entry->max_protection) && 1515 (prev->inheritance == entry->inheritance) && 1516 (prev->wired_count == entry->wired_count) && 1517 (prev->cred == entry->cred)) { 1518 vm_map_entry_unlink(map, prev); 1519 entry->start = prev->start; 1520 entry->offset = prev->offset; 1521 if (entry->prev != &map->header) 1522 vm_map_entry_resize_free(map, entry->prev); 1523 1524 /* 1525 * If the backing object is a vnode object, 1526 * vm_object_deallocate() calls vrele(). 1527 * However, vrele() does not lock the vnode 1528 * because the vnode has additional 1529 * references. Thus, the map lock can be kept 1530 * without causing a lock-order reversal with 1531 * the vnode lock. 1532 * 1533 * Since we count the number of virtual page 1534 * mappings in object->un_pager.vnp.writemappings, 1535 * the writemappings value should not be adjusted 1536 * when the entry is disposed of. 1537 */ 1538 if (prev->object.vm_object) 1539 vm_object_deallocate(prev->object.vm_object); 1540 if (prev->cred != NULL) 1541 crfree(prev->cred); 1542 vm_map_entry_dispose(map, prev); 1543 } 1544 } 1545 1546 next = entry->next; 1547 if (next != &map->header) { 1548 esize = entry->end - entry->start; 1549 if ((entry->end == next->start) && 1550 (next->object.vm_object == entry->object.vm_object) && 1551 (!entry->object.vm_object || 1552 (entry->offset + esize == next->offset)) && 1553 (next->eflags == entry->eflags) && 1554 (next->protection == entry->protection) && 1555 (next->max_protection == entry->max_protection) && 1556 (next->inheritance == entry->inheritance) && 1557 (next->wired_count == entry->wired_count) && 1558 (next->cred == entry->cred)) { 1559 vm_map_entry_unlink(map, next); 1560 entry->end = next->end; 1561 vm_map_entry_resize_free(map, entry); 1562 1563 /* 1564 * See comment above. 1565 */ 1566 if (next->object.vm_object) 1567 vm_object_deallocate(next->object.vm_object); 1568 if (next->cred != NULL) 1569 crfree(next->cred); 1570 vm_map_entry_dispose(map, next); 1571 } 1572 } 1573} 1574/* 1575 * vm_map_clip_start: [ internal use only ] 1576 * 1577 * Asserts that the given entry begins at or after 1578 * the specified address; if necessary, 1579 * it splits the entry into two. 1580 */ 1581#define vm_map_clip_start(map, entry, startaddr) \ 1582{ \ 1583 if (startaddr > entry->start) \ 1584 _vm_map_clip_start(map, entry, startaddr); \ 1585} 1586 1587/* 1588 * This routine is called only when it is known that 1589 * the entry must be split. 1590 */ 1591static void 1592_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1593{ 1594 vm_map_entry_t new_entry; 1595 1596 VM_MAP_ASSERT_LOCKED(map); 1597 1598 /* 1599 * Split off the front portion -- note that we must insert the new 1600 * entry BEFORE this one, so that this entry has the specified 1601 * starting address. 1602 */ 1603 vm_map_simplify_entry(map, entry); 1604 1605 /* 1606 * If there is no object backing this entry, we might as well create 1607 * one now. If we defer it, an object can get created after the map 1608 * is clipped, and individual objects will be created for the split-up 1609 * map. This is a bit of a hack, but is also about the best place to 1610 * put this improvement. 1611 */ 1612 if (entry->object.vm_object == NULL && !map->system_map) { 1613 vm_object_t object; 1614 object = vm_object_allocate(OBJT_DEFAULT, 1615 atop(entry->end - entry->start)); 1616 entry->object.vm_object = object; 1617 entry->offset = 0; 1618 if (entry->cred != NULL) { 1619 object->cred = entry->cred; 1620 object->charge = entry->end - entry->start; 1621 entry->cred = NULL; 1622 } 1623 } else if (entry->object.vm_object != NULL && 1624 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1625 entry->cred != NULL) { 1626 VM_OBJECT_LOCK(entry->object.vm_object); 1627 KASSERT(entry->object.vm_object->cred == NULL, 1628 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1629 entry->object.vm_object->cred = entry->cred; 1630 entry->object.vm_object->charge = entry->end - entry->start; 1631 VM_OBJECT_UNLOCK(entry->object.vm_object); 1632 entry->cred = NULL; 1633 } 1634 1635 new_entry = vm_map_entry_create(map); 1636 *new_entry = *entry; 1637 1638 new_entry->end = start; 1639 entry->offset += (start - entry->start); 1640 entry->start = start; 1641 if (new_entry->cred != NULL) 1642 crhold(entry->cred); 1643 1644 vm_map_entry_link(map, entry->prev, new_entry); 1645 1646 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1647 vm_object_reference(new_entry->object.vm_object); 1648 /* 1649 * The object->un_pager.vnp.writemappings for the 1650 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1651 * kept as is here. The virtual pages are 1652 * re-distributed among the clipped entries, so the sum is 1653 * left the same. 1654 */ 1655 } 1656} 1657 1658/* 1659 * vm_map_clip_end: [ internal use only ] 1660 * 1661 * Asserts that the given entry ends at or before 1662 * the specified address; if necessary, 1663 * it splits the entry into two. 1664 */ 1665#define vm_map_clip_end(map, entry, endaddr) \ 1666{ \ 1667 if ((endaddr) < (entry->end)) \ 1668 _vm_map_clip_end((map), (entry), (endaddr)); \ 1669} 1670 1671/* 1672 * This routine is called only when it is known that 1673 * the entry must be split. 1674 */ 1675static void 1676_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1677{ 1678 vm_map_entry_t new_entry; 1679 1680 VM_MAP_ASSERT_LOCKED(map); 1681 1682 /* 1683 * If there is no object backing this entry, we might as well create 1684 * one now. If we defer it, an object can get created after the map 1685 * is clipped, and individual objects will be created for the split-up 1686 * map. This is a bit of a hack, but is also about the best place to 1687 * put this improvement. 1688 */ 1689 if (entry->object.vm_object == NULL && !map->system_map) { 1690 vm_object_t object; 1691 object = vm_object_allocate(OBJT_DEFAULT, 1692 atop(entry->end - entry->start)); 1693 entry->object.vm_object = object; 1694 entry->offset = 0; 1695 if (entry->cred != NULL) { 1696 object->cred = entry->cred; 1697 object->charge = entry->end - entry->start; 1698 entry->cred = NULL; 1699 } 1700 } else if (entry->object.vm_object != NULL && 1701 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1702 entry->cred != NULL) { 1703 VM_OBJECT_LOCK(entry->object.vm_object); 1704 KASSERT(entry->object.vm_object->cred == NULL, 1705 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1706 entry->object.vm_object->cred = entry->cred; 1707 entry->object.vm_object->charge = entry->end - entry->start; 1708 VM_OBJECT_UNLOCK(entry->object.vm_object); 1709 entry->cred = NULL; 1710 } 1711 1712 /* 1713 * Create a new entry and insert it AFTER the specified entry 1714 */ 1715 new_entry = vm_map_entry_create(map); 1716 *new_entry = *entry; 1717 1718 new_entry->start = entry->end = end; 1719 new_entry->offset += (end - entry->start); 1720 if (new_entry->cred != NULL) 1721 crhold(entry->cred); 1722 1723 vm_map_entry_link(map, entry, new_entry); 1724 1725 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1726 vm_object_reference(new_entry->object.vm_object); 1727 } 1728} 1729 1730/* 1731 * vm_map_submap: [ kernel use only ] 1732 * 1733 * Mark the given range as handled by a subordinate map. 1734 * 1735 * This range must have been created with vm_map_find, 1736 * and no other operations may have been performed on this 1737 * range prior to calling vm_map_submap. 1738 * 1739 * Only a limited number of operations can be performed 1740 * within this rage after calling vm_map_submap: 1741 * vm_fault 1742 * [Don't try vm_map_copy!] 1743 * 1744 * To remove a submapping, one must first remove the 1745 * range from the superior map, and then destroy the 1746 * submap (if desired). [Better yet, don't try it.] 1747 */ 1748int 1749vm_map_submap( 1750 vm_map_t map, 1751 vm_offset_t start, 1752 vm_offset_t end, 1753 vm_map_t submap) 1754{ 1755 vm_map_entry_t entry; 1756 int result = KERN_INVALID_ARGUMENT; 1757 1758 vm_map_lock(map); 1759 1760 VM_MAP_RANGE_CHECK(map, start, end); 1761 1762 if (vm_map_lookup_entry(map, start, &entry)) { 1763 vm_map_clip_start(map, entry, start); 1764 } else 1765 entry = entry->next; 1766 1767 vm_map_clip_end(map, entry, end); 1768 1769 if ((entry->start == start) && (entry->end == end) && 1770 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1771 (entry->object.vm_object == NULL)) { 1772 entry->object.sub_map = submap; 1773 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1774 result = KERN_SUCCESS; 1775 } 1776 vm_map_unlock(map); 1777 1778 return (result); 1779} 1780 1781/* 1782 * The maximum number of pages to map 1783 */ 1784#define MAX_INIT_PT 96 1785 1786/* 1787 * vm_map_pmap_enter: 1788 * 1789 * Preload read-only mappings for the specified object's resident pages 1790 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only 1791 * the resident pages within the address range [addr, addr + ulmin(size, 1792 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within 1793 * the specified address range are mapped. This eliminates many soft 1794 * faults on process startup and immediately after an mmap(2). Because 1795 * these are speculative mappings, cached pages are not reactivated and 1796 * mapped. 1797 */ 1798void 1799vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1800 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1801{ 1802 vm_offset_t start; 1803 vm_page_t p, p_start; 1804 vm_pindex_t psize, tmpidx; 1805 1806 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1807 return; 1808 VM_OBJECT_LOCK(object); 1809 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1810 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1811 goto unlock_return; 1812 } 1813 1814 psize = atop(size); 1815 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0) 1816 psize = MAX_INIT_PT; 1817 if (psize + pindex > object->size) { 1818 if (object->size < pindex) 1819 goto unlock_return; 1820 psize = object->size - pindex; 1821 } 1822 1823 start = 0; 1824 p_start = NULL; 1825 1826 p = vm_page_find_least(object, pindex); 1827 /* 1828 * Assert: the variable p is either (1) the page with the 1829 * least pindex greater than or equal to the parameter pindex 1830 * or (2) NULL. 1831 */ 1832 for (; 1833 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1834 p = TAILQ_NEXT(p, listq)) { 1835 /* 1836 * don't allow an madvise to blow away our really 1837 * free pages allocating pv entries. 1838 */ 1839 if ((flags & MAP_PREFAULT_MADVISE) && 1840 cnt.v_free_count < cnt.v_free_reserved) { 1841 psize = tmpidx; 1842 break; 1843 } 1844 if (p->valid == VM_PAGE_BITS_ALL) { 1845 if (p_start == NULL) { 1846 start = addr + ptoa(tmpidx); 1847 p_start = p; 1848 } 1849 } else if (p_start != NULL) { 1850 pmap_enter_object(map->pmap, start, addr + 1851 ptoa(tmpidx), p_start, prot); 1852 p_start = NULL; 1853 } 1854 } 1855 if (p_start != NULL) 1856 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1857 p_start, prot); 1858unlock_return: 1859 VM_OBJECT_UNLOCK(object); 1860} 1861 1862/* 1863 * vm_map_protect: 1864 * 1865 * Sets the protection of the specified address 1866 * region in the target map. If "set_max" is 1867 * specified, the maximum protection is to be set; 1868 * otherwise, only the current protection is affected. 1869 */ 1870int 1871vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1872 vm_prot_t new_prot, boolean_t set_max) 1873{ 1874 vm_map_entry_t current, entry; 1875 vm_object_t obj; 1876 struct ucred *cred; 1877 vm_prot_t old_prot; 1878 1879 vm_map_lock(map); 1880 1881 VM_MAP_RANGE_CHECK(map, start, end); 1882 1883 if (vm_map_lookup_entry(map, start, &entry)) { 1884 vm_map_clip_start(map, entry, start); 1885 } else { 1886 entry = entry->next; 1887 } 1888 1889 /* 1890 * Make a first pass to check for protection violations. 1891 */ 1892 current = entry; 1893 while ((current != &map->header) && (current->start < end)) { 1894 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1895 vm_map_unlock(map); 1896 return (KERN_INVALID_ARGUMENT); 1897 } 1898 if ((new_prot & current->max_protection) != new_prot) { 1899 vm_map_unlock(map); 1900 return (KERN_PROTECTION_FAILURE); 1901 } 1902 current = current->next; 1903 } 1904 1905 1906 /* 1907 * Do an accounting pass for private read-only mappings that 1908 * now will do cow due to allowed write (e.g. debugger sets 1909 * breakpoint on text segment) 1910 */ 1911 for (current = entry; (current != &map->header) && 1912 (current->start < end); current = current->next) { 1913 1914 vm_map_clip_end(map, current, end); 1915 1916 if (set_max || 1917 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1918 ENTRY_CHARGED(current)) { 1919 continue; 1920 } 1921 1922 cred = curthread->td_ucred; 1923 obj = current->object.vm_object; 1924 1925 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1926 if (!swap_reserve(current->end - current->start)) { 1927 vm_map_unlock(map); 1928 return (KERN_RESOURCE_SHORTAGE); 1929 } 1930 crhold(cred); 1931 current->cred = cred; 1932 continue; 1933 } 1934 1935 VM_OBJECT_LOCK(obj); 1936 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1937 VM_OBJECT_UNLOCK(obj); 1938 continue; 1939 } 1940 1941 /* 1942 * Charge for the whole object allocation now, since 1943 * we cannot distinguish between non-charged and 1944 * charged clipped mapping of the same object later. 1945 */ 1946 KASSERT(obj->charge == 0, 1947 ("vm_map_protect: object %p overcharged\n", obj)); 1948 if (!swap_reserve(ptoa(obj->size))) { 1949 VM_OBJECT_UNLOCK(obj); 1950 vm_map_unlock(map); 1951 return (KERN_RESOURCE_SHORTAGE); 1952 } 1953 1954 crhold(cred); 1955 obj->cred = cred; 1956 obj->charge = ptoa(obj->size); 1957 VM_OBJECT_UNLOCK(obj); 1958 } 1959 1960 /* 1961 * Go back and fix up protections. [Note that clipping is not 1962 * necessary the second time.] 1963 */ 1964 current = entry; 1965 while ((current != &map->header) && (current->start < end)) { 1966 old_prot = current->protection; 1967 1968 if (set_max) 1969 current->protection = 1970 (current->max_protection = new_prot) & 1971 old_prot; 1972 else 1973 current->protection = new_prot; 1974 1975 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED)) 1976 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) && 1977 (current->protection & VM_PROT_WRITE) != 0 && 1978 (old_prot & VM_PROT_WRITE) == 0) { 1979 vm_fault_copy_entry(map, map, current, current, NULL); 1980 } 1981 1982 /* 1983 * When restricting access, update the physical map. Worry 1984 * about copy-on-write here. 1985 */ 1986 if ((old_prot & ~current->protection) != 0) { 1987#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1988 VM_PROT_ALL) 1989 pmap_protect(map->pmap, current->start, 1990 current->end, 1991 current->protection & MASK(current)); 1992#undef MASK 1993 } 1994 vm_map_simplify_entry(map, current); 1995 current = current->next; 1996 } 1997 vm_map_unlock(map); 1998 return (KERN_SUCCESS); 1999} 2000 2001/* 2002 * vm_map_madvise: 2003 * 2004 * This routine traverses a processes map handling the madvise 2005 * system call. Advisories are classified as either those effecting 2006 * the vm_map_entry structure, or those effecting the underlying 2007 * objects. 2008 */ 2009int 2010vm_map_madvise( 2011 vm_map_t map, 2012 vm_offset_t start, 2013 vm_offset_t end, 2014 int behav) 2015{ 2016 vm_map_entry_t current, entry; 2017 int modify_map = 0; 2018 2019 /* 2020 * Some madvise calls directly modify the vm_map_entry, in which case 2021 * we need to use an exclusive lock on the map and we need to perform 2022 * various clipping operations. Otherwise we only need a read-lock 2023 * on the map. 2024 */ 2025 switch(behav) { 2026 case MADV_NORMAL: 2027 case MADV_SEQUENTIAL: 2028 case MADV_RANDOM: 2029 case MADV_NOSYNC: 2030 case MADV_AUTOSYNC: 2031 case MADV_NOCORE: 2032 case MADV_CORE: 2033 modify_map = 1; 2034 vm_map_lock(map); 2035 break; 2036 case MADV_WILLNEED: 2037 case MADV_DONTNEED: 2038 case MADV_FREE: 2039 vm_map_lock_read(map); 2040 break; 2041 default: 2042 return (KERN_INVALID_ARGUMENT); 2043 } 2044 2045 /* 2046 * Locate starting entry and clip if necessary. 2047 */ 2048 VM_MAP_RANGE_CHECK(map, start, end); 2049 2050 if (vm_map_lookup_entry(map, start, &entry)) { 2051 if (modify_map) 2052 vm_map_clip_start(map, entry, start); 2053 } else { 2054 entry = entry->next; 2055 } 2056 2057 if (modify_map) { 2058 /* 2059 * madvise behaviors that are implemented in the vm_map_entry. 2060 * 2061 * We clip the vm_map_entry so that behavioral changes are 2062 * limited to the specified address range. 2063 */ 2064 for (current = entry; 2065 (current != &map->header) && (current->start < end); 2066 current = current->next 2067 ) { 2068 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2069 continue; 2070 2071 vm_map_clip_end(map, current, end); 2072 2073 switch (behav) { 2074 case MADV_NORMAL: 2075 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2076 break; 2077 case MADV_SEQUENTIAL: 2078 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2079 break; 2080 case MADV_RANDOM: 2081 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2082 break; 2083 case MADV_NOSYNC: 2084 current->eflags |= MAP_ENTRY_NOSYNC; 2085 break; 2086 case MADV_AUTOSYNC: 2087 current->eflags &= ~MAP_ENTRY_NOSYNC; 2088 break; 2089 case MADV_NOCORE: 2090 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2091 break; 2092 case MADV_CORE: 2093 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2094 break; 2095 default: 2096 break; 2097 } 2098 vm_map_simplify_entry(map, current); 2099 } 2100 vm_map_unlock(map); 2101 } else { 2102 vm_pindex_t pstart, pend; 2103 2104 /* 2105 * madvise behaviors that are implemented in the underlying 2106 * vm_object. 2107 * 2108 * Since we don't clip the vm_map_entry, we have to clip 2109 * the vm_object pindex and count. 2110 */ 2111 for (current = entry; 2112 (current != &map->header) && (current->start < end); 2113 current = current->next 2114 ) { 2115 vm_offset_t useStart; 2116 2117 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2118 continue; 2119 2120 pstart = OFF_TO_IDX(current->offset); 2121 pend = pstart + atop(current->end - current->start); 2122 useStart = current->start; 2123 2124 if (current->start < start) { 2125 pstart += atop(start - current->start); 2126 useStart = start; 2127 } 2128 if (current->end > end) 2129 pend -= atop(current->end - end); 2130 2131 if (pstart >= pend) 2132 continue; 2133 2134 vm_object_madvise(current->object.vm_object, pstart, 2135 pend, behav); 2136 if (behav == MADV_WILLNEED) { 2137 vm_map_pmap_enter(map, 2138 useStart, 2139 current->protection, 2140 current->object.vm_object, 2141 pstart, 2142 ptoa(pend - pstart), 2143 MAP_PREFAULT_MADVISE 2144 ); 2145 } 2146 } 2147 vm_map_unlock_read(map); 2148 } 2149 return (0); 2150} 2151 2152 2153/* 2154 * vm_map_inherit: 2155 * 2156 * Sets the inheritance of the specified address 2157 * range in the target map. Inheritance 2158 * affects how the map will be shared with 2159 * child maps at the time of vmspace_fork. 2160 */ 2161int 2162vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2163 vm_inherit_t new_inheritance) 2164{ 2165 vm_map_entry_t entry; 2166 vm_map_entry_t temp_entry; 2167 2168 switch (new_inheritance) { 2169 case VM_INHERIT_NONE: 2170 case VM_INHERIT_COPY: 2171 case VM_INHERIT_SHARE: 2172 break; 2173 default: 2174 return (KERN_INVALID_ARGUMENT); 2175 } 2176 vm_map_lock(map); 2177 VM_MAP_RANGE_CHECK(map, start, end); 2178 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2179 entry = temp_entry; 2180 vm_map_clip_start(map, entry, start); 2181 } else 2182 entry = temp_entry->next; 2183 while ((entry != &map->header) && (entry->start < end)) { 2184 vm_map_clip_end(map, entry, end); 2185 entry->inheritance = new_inheritance; 2186 vm_map_simplify_entry(map, entry); 2187 entry = entry->next; 2188 } 2189 vm_map_unlock(map); 2190 return (KERN_SUCCESS); 2191} 2192 2193/* 2194 * vm_map_unwire: 2195 * 2196 * Implements both kernel and user unwiring. 2197 */ 2198int 2199vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2200 int flags) 2201{ 2202 vm_map_entry_t entry, first_entry, tmp_entry; 2203 vm_offset_t saved_start; 2204 unsigned int last_timestamp; 2205 int rv; 2206 boolean_t need_wakeup, result, user_unwire; 2207 2208 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2209 vm_map_lock(map); 2210 VM_MAP_RANGE_CHECK(map, start, end); 2211 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2212 if (flags & VM_MAP_WIRE_HOLESOK) 2213 first_entry = first_entry->next; 2214 else { 2215 vm_map_unlock(map); 2216 return (KERN_INVALID_ADDRESS); 2217 } 2218 } 2219 last_timestamp = map->timestamp; 2220 entry = first_entry; 2221 while (entry != &map->header && entry->start < end) { 2222 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2223 /* 2224 * We have not yet clipped the entry. 2225 */ 2226 saved_start = (start >= entry->start) ? start : 2227 entry->start; 2228 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2229 if (vm_map_unlock_and_wait(map, 0)) { 2230 /* 2231 * Allow interruption of user unwiring? 2232 */ 2233 } 2234 vm_map_lock(map); 2235 if (last_timestamp+1 != map->timestamp) { 2236 /* 2237 * Look again for the entry because the map was 2238 * modified while it was unlocked. 2239 * Specifically, the entry may have been 2240 * clipped, merged, or deleted. 2241 */ 2242 if (!vm_map_lookup_entry(map, saved_start, 2243 &tmp_entry)) { 2244 if (flags & VM_MAP_WIRE_HOLESOK) 2245 tmp_entry = tmp_entry->next; 2246 else { 2247 if (saved_start == start) { 2248 /* 2249 * First_entry has been deleted. 2250 */ 2251 vm_map_unlock(map); 2252 return (KERN_INVALID_ADDRESS); 2253 } 2254 end = saved_start; 2255 rv = KERN_INVALID_ADDRESS; 2256 goto done; 2257 } 2258 } 2259 if (entry == first_entry) 2260 first_entry = tmp_entry; 2261 else 2262 first_entry = NULL; 2263 entry = tmp_entry; 2264 } 2265 last_timestamp = map->timestamp; 2266 continue; 2267 } 2268 vm_map_clip_start(map, entry, start); 2269 vm_map_clip_end(map, entry, end); 2270 /* 2271 * Mark the entry in case the map lock is released. (See 2272 * above.) 2273 */ 2274 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2275 /* 2276 * Check the map for holes in the specified region. 2277 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2278 */ 2279 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2280 (entry->end < end && (entry->next == &map->header || 2281 entry->next->start > entry->end))) { 2282 end = entry->end; 2283 rv = KERN_INVALID_ADDRESS; 2284 goto done; 2285 } 2286 /* 2287 * If system unwiring, require that the entry is system wired. 2288 */ 2289 if (!user_unwire && 2290 vm_map_entry_system_wired_count(entry) == 0) { 2291 end = entry->end; 2292 rv = KERN_INVALID_ARGUMENT; 2293 goto done; 2294 } 2295 entry = entry->next; 2296 } 2297 rv = KERN_SUCCESS; 2298done: 2299 need_wakeup = FALSE; 2300 if (first_entry == NULL) { 2301 result = vm_map_lookup_entry(map, start, &first_entry); 2302 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2303 first_entry = first_entry->next; 2304 else 2305 KASSERT(result, ("vm_map_unwire: lookup failed")); 2306 } 2307 entry = first_entry; 2308 while (entry != &map->header && entry->start < end) { 2309 if (rv == KERN_SUCCESS && (!user_unwire || 2310 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2311 if (user_unwire) 2312 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2313 entry->wired_count--; 2314 if (entry->wired_count == 0) { 2315 /* 2316 * Retain the map lock. 2317 */ 2318 vm_fault_unwire(map, entry->start, entry->end, 2319 entry->object.vm_object != NULL && 2320 (entry->object.vm_object->flags & 2321 OBJ_FICTITIOUS) != 0); 2322 } 2323 } 2324 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2325 ("vm_map_unwire: in-transition flag missing")); 2326 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2327 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2328 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2329 need_wakeup = TRUE; 2330 } 2331 vm_map_simplify_entry(map, entry); 2332 entry = entry->next; 2333 } 2334 vm_map_unlock(map); 2335 if (need_wakeup) 2336 vm_map_wakeup(map); 2337 return (rv); 2338} 2339 2340/* 2341 * vm_map_wire: 2342 * 2343 * Implements both kernel and user wiring. 2344 */ 2345int 2346vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2347 int flags) 2348{ 2349 vm_map_entry_t entry, first_entry, tmp_entry; 2350 vm_offset_t saved_end, saved_start; 2351 unsigned int last_timestamp; 2352 int rv; 2353 boolean_t fictitious, need_wakeup, result, user_wire; 2354 vm_prot_t prot; 2355 2356 prot = 0; 2357 if (flags & VM_MAP_WIRE_WRITE) 2358 prot |= VM_PROT_WRITE; 2359 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2360 vm_map_lock(map); 2361 VM_MAP_RANGE_CHECK(map, start, end); 2362 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2363 if (flags & VM_MAP_WIRE_HOLESOK) 2364 first_entry = first_entry->next; 2365 else { 2366 vm_map_unlock(map); 2367 return (KERN_INVALID_ADDRESS); 2368 } 2369 } 2370 last_timestamp = map->timestamp; 2371 entry = first_entry; 2372 while (entry != &map->header && entry->start < end) { 2373 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2374 /* 2375 * We have not yet clipped the entry. 2376 */ 2377 saved_start = (start >= entry->start) ? start : 2378 entry->start; 2379 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2380 if (vm_map_unlock_and_wait(map, 0)) { 2381 /* 2382 * Allow interruption of user wiring? 2383 */ 2384 } 2385 vm_map_lock(map); 2386 if (last_timestamp + 1 != map->timestamp) { 2387 /* 2388 * Look again for the entry because the map was 2389 * modified while it was unlocked. 2390 * Specifically, the entry may have been 2391 * clipped, merged, or deleted. 2392 */ 2393 if (!vm_map_lookup_entry(map, saved_start, 2394 &tmp_entry)) { 2395 if (flags & VM_MAP_WIRE_HOLESOK) 2396 tmp_entry = tmp_entry->next; 2397 else { 2398 if (saved_start == start) { 2399 /* 2400 * first_entry has been deleted. 2401 */ 2402 vm_map_unlock(map); 2403 return (KERN_INVALID_ADDRESS); 2404 } 2405 end = saved_start; 2406 rv = KERN_INVALID_ADDRESS; 2407 goto done; 2408 } 2409 } 2410 if (entry == first_entry) 2411 first_entry = tmp_entry; 2412 else 2413 first_entry = NULL; 2414 entry = tmp_entry; 2415 } 2416 last_timestamp = map->timestamp; 2417 continue; 2418 } 2419 vm_map_clip_start(map, entry, start); 2420 vm_map_clip_end(map, entry, end); 2421 /* 2422 * Mark the entry in case the map lock is released. (See 2423 * above.) 2424 */ 2425 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2426 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2427 || (entry->protection & prot) != prot) { 2428 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2429 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2430 end = entry->end; 2431 rv = KERN_INVALID_ADDRESS; 2432 goto done; 2433 } 2434 goto next_entry; 2435 } 2436 if (entry->wired_count == 0) { 2437 entry->wired_count++; 2438 saved_start = entry->start; 2439 saved_end = entry->end; 2440 fictitious = entry->object.vm_object != NULL && 2441 (entry->object.vm_object->flags & 2442 OBJ_FICTITIOUS) != 0; 2443 /* 2444 * Release the map lock, relying on the in-transition 2445 * mark. Mark the map busy for fork. 2446 */ 2447 vm_map_busy(map); 2448 vm_map_unlock(map); 2449 rv = vm_fault_wire(map, saved_start, saved_end, 2450 fictitious); 2451 vm_map_lock(map); 2452 vm_map_unbusy(map); 2453 if (last_timestamp + 1 != map->timestamp) { 2454 /* 2455 * Look again for the entry because the map was 2456 * modified while it was unlocked. The entry 2457 * may have been clipped, but NOT merged or 2458 * deleted. 2459 */ 2460 result = vm_map_lookup_entry(map, saved_start, 2461 &tmp_entry); 2462 KASSERT(result, ("vm_map_wire: lookup failed")); 2463 if (entry == first_entry) 2464 first_entry = tmp_entry; 2465 else 2466 first_entry = NULL; 2467 entry = tmp_entry; 2468 while (entry->end < saved_end) { 2469 if (rv != KERN_SUCCESS) { 2470 KASSERT(entry->wired_count == 1, 2471 ("vm_map_wire: bad count")); 2472 entry->wired_count = -1; 2473 } 2474 entry = entry->next; 2475 } 2476 } 2477 last_timestamp = map->timestamp; 2478 if (rv != KERN_SUCCESS) { 2479 KASSERT(entry->wired_count == 1, 2480 ("vm_map_wire: bad count")); 2481 /* 2482 * Assign an out-of-range value to represent 2483 * the failure to wire this entry. 2484 */ 2485 entry->wired_count = -1; 2486 end = entry->end; 2487 goto done; 2488 } 2489 } else if (!user_wire || 2490 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2491 entry->wired_count++; 2492 } 2493 /* 2494 * Check the map for holes in the specified region. 2495 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2496 */ 2497 next_entry: 2498 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2499 (entry->end < end && (entry->next == &map->header || 2500 entry->next->start > entry->end))) { 2501 end = entry->end; 2502 rv = KERN_INVALID_ADDRESS; 2503 goto done; 2504 } 2505 entry = entry->next; 2506 } 2507 rv = KERN_SUCCESS; 2508done: 2509 need_wakeup = FALSE; 2510 if (first_entry == NULL) { 2511 result = vm_map_lookup_entry(map, start, &first_entry); 2512 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2513 first_entry = first_entry->next; 2514 else 2515 KASSERT(result, ("vm_map_wire: lookup failed")); 2516 } 2517 entry = first_entry; 2518 while (entry != &map->header && entry->start < end) { 2519 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2520 goto next_entry_done; 2521 if (rv == KERN_SUCCESS) { 2522 if (user_wire) 2523 entry->eflags |= MAP_ENTRY_USER_WIRED; 2524 } else if (entry->wired_count == -1) { 2525 /* 2526 * Wiring failed on this entry. Thus, unwiring is 2527 * unnecessary. 2528 */ 2529 entry->wired_count = 0; 2530 } else { 2531 if (!user_wire || 2532 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2533 entry->wired_count--; 2534 if (entry->wired_count == 0) { 2535 /* 2536 * Retain the map lock. 2537 */ 2538 vm_fault_unwire(map, entry->start, entry->end, 2539 entry->object.vm_object != NULL && 2540 (entry->object.vm_object->flags & 2541 OBJ_FICTITIOUS) != 0); 2542 } 2543 } 2544 next_entry_done: 2545 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2546 ("vm_map_wire: in-transition flag missing")); 2547 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED); 2548 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2549 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2550 need_wakeup = TRUE; 2551 } 2552 vm_map_simplify_entry(map, entry); 2553 entry = entry->next; 2554 } 2555 vm_map_unlock(map); 2556 if (need_wakeup) 2557 vm_map_wakeup(map); 2558 return (rv); 2559} 2560 2561/* 2562 * vm_map_sync 2563 * 2564 * Push any dirty cached pages in the address range to their pager. 2565 * If syncio is TRUE, dirty pages are written synchronously. 2566 * If invalidate is TRUE, any cached pages are freed as well. 2567 * 2568 * If the size of the region from start to end is zero, we are 2569 * supposed to flush all modified pages within the region containing 2570 * start. Unfortunately, a region can be split or coalesced with 2571 * neighboring regions, making it difficult to determine what the 2572 * original region was. Therefore, we approximate this requirement by 2573 * flushing the current region containing start. 2574 * 2575 * Returns an error if any part of the specified range is not mapped. 2576 */ 2577int 2578vm_map_sync( 2579 vm_map_t map, 2580 vm_offset_t start, 2581 vm_offset_t end, 2582 boolean_t syncio, 2583 boolean_t invalidate) 2584{ 2585 vm_map_entry_t current; 2586 vm_map_entry_t entry; 2587 vm_size_t size; 2588 vm_object_t object; 2589 vm_ooffset_t offset; 2590 unsigned int last_timestamp; 2591 boolean_t failed; 2592 2593 vm_map_lock_read(map); 2594 VM_MAP_RANGE_CHECK(map, start, end); 2595 if (!vm_map_lookup_entry(map, start, &entry)) { 2596 vm_map_unlock_read(map); 2597 return (KERN_INVALID_ADDRESS); 2598 } else if (start == end) { 2599 start = entry->start; 2600 end = entry->end; 2601 } 2602 /* 2603 * Make a first pass to check for user-wired memory and holes. 2604 */ 2605 for (current = entry; current != &map->header && current->start < end; 2606 current = current->next) { 2607 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2608 vm_map_unlock_read(map); 2609 return (KERN_INVALID_ARGUMENT); 2610 } 2611 if (end > current->end && 2612 (current->next == &map->header || 2613 current->end != current->next->start)) { 2614 vm_map_unlock_read(map); 2615 return (KERN_INVALID_ADDRESS); 2616 } 2617 } 2618 2619 if (invalidate) 2620 pmap_remove(map->pmap, start, end); 2621 failed = FALSE; 2622 2623 /* 2624 * Make a second pass, cleaning/uncaching pages from the indicated 2625 * objects as we go. 2626 */ 2627 for (current = entry; current != &map->header && current->start < end;) { 2628 offset = current->offset + (start - current->start); 2629 size = (end <= current->end ? end : current->end) - start; 2630 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2631 vm_map_t smap; 2632 vm_map_entry_t tentry; 2633 vm_size_t tsize; 2634 2635 smap = current->object.sub_map; 2636 vm_map_lock_read(smap); 2637 (void) vm_map_lookup_entry(smap, offset, &tentry); 2638 tsize = tentry->end - offset; 2639 if (tsize < size) 2640 size = tsize; 2641 object = tentry->object.vm_object; 2642 offset = tentry->offset + (offset - tentry->start); 2643 vm_map_unlock_read(smap); 2644 } else { 2645 object = current->object.vm_object; 2646 } 2647 vm_object_reference(object); 2648 last_timestamp = map->timestamp; 2649 vm_map_unlock_read(map); 2650 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2651 failed = TRUE; 2652 start += size; 2653 vm_object_deallocate(object); 2654 vm_map_lock_read(map); 2655 if (last_timestamp == map->timestamp || 2656 !vm_map_lookup_entry(map, start, ¤t)) 2657 current = current->next; 2658 } 2659 2660 vm_map_unlock_read(map); 2661 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2662} 2663 2664/* 2665 * vm_map_entry_unwire: [ internal use only ] 2666 * 2667 * Make the region specified by this entry pageable. 2668 * 2669 * The map in question should be locked. 2670 * [This is the reason for this routine's existence.] 2671 */ 2672static void 2673vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2674{ 2675 vm_fault_unwire(map, entry->start, entry->end, 2676 entry->object.vm_object != NULL && 2677 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2678 entry->wired_count = 0; 2679} 2680 2681static void 2682vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2683{ 2684 2685 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2686 vm_object_deallocate(entry->object.vm_object); 2687 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2688} 2689 2690/* 2691 * vm_map_entry_delete: [ internal use only ] 2692 * 2693 * Deallocate the given entry from the target map. 2694 */ 2695static void 2696vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2697{ 2698 vm_object_t object; 2699 vm_pindex_t offidxstart, offidxend, count, size1; 2700 vm_ooffset_t size; 2701 2702 vm_map_entry_unlink(map, entry); 2703 object = entry->object.vm_object; 2704 size = entry->end - entry->start; 2705 map->size -= size; 2706 2707 if (entry->cred != NULL) { 2708 swap_release_by_cred(size, entry->cred); 2709 crfree(entry->cred); 2710 } 2711 2712 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2713 (object != NULL)) { 2714 KASSERT(entry->cred == NULL || object->cred == NULL || 2715 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2716 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2717 count = OFF_TO_IDX(size); 2718 offidxstart = OFF_TO_IDX(entry->offset); 2719 offidxend = offidxstart + count; 2720 VM_OBJECT_LOCK(object); 2721 if (object->ref_count != 1 && 2722 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2723 object == kernel_object || object == kmem_object)) { 2724 vm_object_collapse(object); 2725 2726 /* 2727 * The option OBJPR_NOTMAPPED can be passed here 2728 * because vm_map_delete() already performed 2729 * pmap_remove() on the only mapping to this range 2730 * of pages. 2731 */ 2732 vm_object_page_remove(object, offidxstart, offidxend, 2733 OBJPR_NOTMAPPED); 2734 if (object->type == OBJT_SWAP) 2735 swap_pager_freespace(object, offidxstart, count); 2736 if (offidxend >= object->size && 2737 offidxstart < object->size) { 2738 size1 = object->size; 2739 object->size = offidxstart; 2740 if (object->cred != NULL) { 2741 size1 -= object->size; 2742 KASSERT(object->charge >= ptoa(size1), 2743 ("vm_map_entry_delete: object->charge < 0")); 2744 swap_release_by_cred(ptoa(size1), object->cred); 2745 object->charge -= ptoa(size1); 2746 } 2747 } 2748 } 2749 VM_OBJECT_UNLOCK(object); 2750 } else 2751 entry->object.vm_object = NULL; 2752 if (map->system_map) 2753 vm_map_entry_deallocate(entry, TRUE); 2754 else { 2755 entry->next = curthread->td_map_def_user; 2756 curthread->td_map_def_user = entry; 2757 } 2758} 2759 2760/* 2761 * vm_map_delete: [ internal use only ] 2762 * 2763 * Deallocates the given address range from the target 2764 * map. 2765 */ 2766int 2767vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2768{ 2769 vm_map_entry_t entry; 2770 vm_map_entry_t first_entry; 2771 2772 VM_MAP_ASSERT_LOCKED(map); 2773 2774 /* 2775 * Find the start of the region, and clip it 2776 */ 2777 if (!vm_map_lookup_entry(map, start, &first_entry)) 2778 entry = first_entry->next; 2779 else { 2780 entry = first_entry; 2781 vm_map_clip_start(map, entry, start); 2782 } 2783 2784 /* 2785 * Step through all entries in this region 2786 */ 2787 while ((entry != &map->header) && (entry->start < end)) { 2788 vm_map_entry_t next; 2789 2790 /* 2791 * Wait for wiring or unwiring of an entry to complete. 2792 * Also wait for any system wirings to disappear on 2793 * user maps. 2794 */ 2795 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2796 (vm_map_pmap(map) != kernel_pmap && 2797 vm_map_entry_system_wired_count(entry) != 0)) { 2798 unsigned int last_timestamp; 2799 vm_offset_t saved_start; 2800 vm_map_entry_t tmp_entry; 2801 2802 saved_start = entry->start; 2803 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2804 last_timestamp = map->timestamp; 2805 (void) vm_map_unlock_and_wait(map, 0); 2806 vm_map_lock(map); 2807 if (last_timestamp + 1 != map->timestamp) { 2808 /* 2809 * Look again for the entry because the map was 2810 * modified while it was unlocked. 2811 * Specifically, the entry may have been 2812 * clipped, merged, or deleted. 2813 */ 2814 if (!vm_map_lookup_entry(map, saved_start, 2815 &tmp_entry)) 2816 entry = tmp_entry->next; 2817 else { 2818 entry = tmp_entry; 2819 vm_map_clip_start(map, entry, 2820 saved_start); 2821 } 2822 } 2823 continue; 2824 } 2825 vm_map_clip_end(map, entry, end); 2826 2827 next = entry->next; 2828 2829 /* 2830 * Unwire before removing addresses from the pmap; otherwise, 2831 * unwiring will put the entries back in the pmap. 2832 */ 2833 if (entry->wired_count != 0) { 2834 vm_map_entry_unwire(map, entry); 2835 } 2836 2837 pmap_remove(map->pmap, entry->start, entry->end); 2838 2839 /* 2840 * Delete the entry only after removing all pmap 2841 * entries pointing to its pages. (Otherwise, its 2842 * page frames may be reallocated, and any modify bits 2843 * will be set in the wrong object!) 2844 */ 2845 vm_map_entry_delete(map, entry); 2846 entry = next; 2847 } 2848 return (KERN_SUCCESS); 2849} 2850 2851/* 2852 * vm_map_remove: 2853 * 2854 * Remove the given address range from the target map. 2855 * This is the exported form of vm_map_delete. 2856 */ 2857int 2858vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2859{ 2860 int result; 2861 2862 vm_map_lock(map); 2863 VM_MAP_RANGE_CHECK(map, start, end); 2864 result = vm_map_delete(map, start, end); 2865 vm_map_unlock(map); 2866 return (result); 2867} 2868 2869/* 2870 * vm_map_check_protection: 2871 * 2872 * Assert that the target map allows the specified privilege on the 2873 * entire address region given. The entire region must be allocated. 2874 * 2875 * WARNING! This code does not and should not check whether the 2876 * contents of the region is accessible. For example a smaller file 2877 * might be mapped into a larger address space. 2878 * 2879 * NOTE! This code is also called by munmap(). 2880 * 2881 * The map must be locked. A read lock is sufficient. 2882 */ 2883boolean_t 2884vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2885 vm_prot_t protection) 2886{ 2887 vm_map_entry_t entry; 2888 vm_map_entry_t tmp_entry; 2889 2890 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2891 return (FALSE); 2892 entry = tmp_entry; 2893 2894 while (start < end) { 2895 if (entry == &map->header) 2896 return (FALSE); 2897 /* 2898 * No holes allowed! 2899 */ 2900 if (start < entry->start) 2901 return (FALSE); 2902 /* 2903 * Check protection associated with entry. 2904 */ 2905 if ((entry->protection & protection) != protection) 2906 return (FALSE); 2907 /* go to next entry */ 2908 start = entry->end; 2909 entry = entry->next; 2910 } 2911 return (TRUE); 2912} 2913 2914/* 2915 * vm_map_copy_entry: 2916 * 2917 * Copies the contents of the source entry to the destination 2918 * entry. The entries *must* be aligned properly. 2919 */ 2920static void 2921vm_map_copy_entry( 2922 vm_map_t src_map, 2923 vm_map_t dst_map, 2924 vm_map_entry_t src_entry, 2925 vm_map_entry_t dst_entry, 2926 vm_ooffset_t *fork_charge) 2927{ 2928 vm_object_t src_object; 2929 vm_map_entry_t fake_entry; 2930 vm_offset_t size; 2931 struct ucred *cred; 2932 int charged; 2933 2934 VM_MAP_ASSERT_LOCKED(dst_map); 2935 2936 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2937 return; 2938 2939 if (src_entry->wired_count == 0) { 2940 2941 /* 2942 * If the source entry is marked needs_copy, it is already 2943 * write-protected. 2944 */ 2945 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2946 pmap_protect(src_map->pmap, 2947 src_entry->start, 2948 src_entry->end, 2949 src_entry->protection & ~VM_PROT_WRITE); 2950 } 2951 2952 /* 2953 * Make a copy of the object. 2954 */ 2955 size = src_entry->end - src_entry->start; 2956 if ((src_object = src_entry->object.vm_object) != NULL) { 2957 VM_OBJECT_LOCK(src_object); 2958 charged = ENTRY_CHARGED(src_entry); 2959 if ((src_object->handle == NULL) && 2960 (src_object->type == OBJT_DEFAULT || 2961 src_object->type == OBJT_SWAP)) { 2962 vm_object_collapse(src_object); 2963 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2964 vm_object_split(src_entry); 2965 src_object = src_entry->object.vm_object; 2966 } 2967 } 2968 vm_object_reference_locked(src_object); 2969 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2970 if (src_entry->cred != NULL && 2971 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 2972 KASSERT(src_object->cred == NULL, 2973 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 2974 src_object)); 2975 src_object->cred = src_entry->cred; 2976 src_object->charge = size; 2977 } 2978 VM_OBJECT_UNLOCK(src_object); 2979 dst_entry->object.vm_object = src_object; 2980 if (charged) { 2981 cred = curthread->td_ucred; 2982 crhold(cred); 2983 dst_entry->cred = cred; 2984 *fork_charge += size; 2985 if (!(src_entry->eflags & 2986 MAP_ENTRY_NEEDS_COPY)) { 2987 crhold(cred); 2988 src_entry->cred = cred; 2989 *fork_charge += size; 2990 } 2991 } 2992 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2993 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2994 dst_entry->offset = src_entry->offset; 2995 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 2996 /* 2997 * MAP_ENTRY_VN_WRITECNT cannot 2998 * indicate write reference from 2999 * src_entry, since the entry is 3000 * marked as needs copy. Allocate a 3001 * fake entry that is used to 3002 * decrement object->un_pager.vnp.writecount 3003 * at the appropriate time. Attach 3004 * fake_entry to the deferred list. 3005 */ 3006 fake_entry = vm_map_entry_create(dst_map); 3007 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3008 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3009 vm_object_reference(src_object); 3010 fake_entry->object.vm_object = src_object; 3011 fake_entry->start = src_entry->start; 3012 fake_entry->end = src_entry->end; 3013 fake_entry->next = curthread->td_map_def_user; 3014 curthread->td_map_def_user = fake_entry; 3015 } 3016 } else { 3017 dst_entry->object.vm_object = NULL; 3018 dst_entry->offset = 0; 3019 if (src_entry->cred != NULL) { 3020 dst_entry->cred = curthread->td_ucred; 3021 crhold(dst_entry->cred); 3022 *fork_charge += size; 3023 } 3024 } 3025 3026 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3027 dst_entry->end - dst_entry->start, src_entry->start); 3028 } else { 3029 /* 3030 * Of course, wired down pages can't be set copy-on-write. 3031 * Cause wired pages to be copied into the new map by 3032 * simulating faults (the new pages are pageable) 3033 */ 3034 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3035 fork_charge); 3036 } 3037} 3038 3039/* 3040 * vmspace_map_entry_forked: 3041 * Update the newly-forked vmspace each time a map entry is inherited 3042 * or copied. The values for vm_dsize and vm_tsize are approximate 3043 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3044 */ 3045static void 3046vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3047 vm_map_entry_t entry) 3048{ 3049 vm_size_t entrysize; 3050 vm_offset_t newend; 3051 3052 entrysize = entry->end - entry->start; 3053 vm2->vm_map.size += entrysize; 3054 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3055 vm2->vm_ssize += btoc(entrysize); 3056 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3057 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3058 newend = MIN(entry->end, 3059 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3060 vm2->vm_dsize += btoc(newend - entry->start); 3061 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3062 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3063 newend = MIN(entry->end, 3064 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3065 vm2->vm_tsize += btoc(newend - entry->start); 3066 } 3067} 3068 3069/* 3070 * vmspace_fork: 3071 * Create a new process vmspace structure and vm_map 3072 * based on those of an existing process. The new map 3073 * is based on the old map, according to the inheritance 3074 * values on the regions in that map. 3075 * 3076 * XXX It might be worth coalescing the entries added to the new vmspace. 3077 * 3078 * The source map must not be locked. 3079 */ 3080struct vmspace * 3081vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3082{ 3083 struct vmspace *vm2; 3084 vm_map_t new_map, old_map; 3085 vm_map_entry_t new_entry, old_entry; 3086 vm_object_t object; 3087 int locked; 3088 3089 old_map = &vm1->vm_map; 3090 /* Copy immutable fields of vm1 to vm2. */ 3091 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 3092 if (vm2 == NULL) 3093 return (NULL); 3094 vm2->vm_taddr = vm1->vm_taddr; 3095 vm2->vm_daddr = vm1->vm_daddr; 3096 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3097 vm_map_lock(old_map); 3098 if (old_map->busy) 3099 vm_map_wait_busy(old_map); 3100 new_map = &vm2->vm_map; 3101 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3102 KASSERT(locked, ("vmspace_fork: lock failed")); 3103 3104 old_entry = old_map->header.next; 3105 3106 while (old_entry != &old_map->header) { 3107 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3108 panic("vm_map_fork: encountered a submap"); 3109 3110 switch (old_entry->inheritance) { 3111 case VM_INHERIT_NONE: 3112 break; 3113 3114 case VM_INHERIT_SHARE: 3115 /* 3116 * Clone the entry, creating the shared object if necessary. 3117 */ 3118 object = old_entry->object.vm_object; 3119 if (object == NULL) { 3120 object = vm_object_allocate(OBJT_DEFAULT, 3121 atop(old_entry->end - old_entry->start)); 3122 old_entry->object.vm_object = object; 3123 old_entry->offset = 0; 3124 if (old_entry->cred != NULL) { 3125 object->cred = old_entry->cred; 3126 object->charge = old_entry->end - 3127 old_entry->start; 3128 old_entry->cred = NULL; 3129 } 3130 } 3131 3132 /* 3133 * Add the reference before calling vm_object_shadow 3134 * to insure that a shadow object is created. 3135 */ 3136 vm_object_reference(object); 3137 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3138 vm_object_shadow(&old_entry->object.vm_object, 3139 &old_entry->offset, 3140 old_entry->end - old_entry->start); 3141 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3142 /* Transfer the second reference too. */ 3143 vm_object_reference( 3144 old_entry->object.vm_object); 3145 3146 /* 3147 * As in vm_map_simplify_entry(), the 3148 * vnode lock will not be acquired in 3149 * this call to vm_object_deallocate(). 3150 */ 3151 vm_object_deallocate(object); 3152 object = old_entry->object.vm_object; 3153 } 3154 VM_OBJECT_LOCK(object); 3155 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3156 if (old_entry->cred != NULL) { 3157 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3158 object->cred = old_entry->cred; 3159 object->charge = old_entry->end - old_entry->start; 3160 old_entry->cred = NULL; 3161 } 3162 VM_OBJECT_UNLOCK(object); 3163 3164 /* 3165 * Clone the entry, referencing the shared object. 3166 */ 3167 new_entry = vm_map_entry_create(new_map); 3168 *new_entry = *old_entry; 3169 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3170 MAP_ENTRY_IN_TRANSITION); 3171 new_entry->wired_count = 0; 3172 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3173 object = new_entry->object.vm_object; 3174 KASSERT(((struct vnode *)object->handle)-> 3175 v_writecount > 0, 3176 ("vmspace_fork: v_writecount")); 3177 KASSERT(object->un_pager.vnp.writemappings > 0, 3178 ("vmspace_fork: vnp.writecount")); 3179 vnode_pager_update_writecount(object, 3180 new_entry->start, new_entry->end); 3181 } 3182 3183 /* 3184 * Insert the entry into the new map -- we know we're 3185 * inserting at the end of the new map. 3186 */ 3187 vm_map_entry_link(new_map, new_map->header.prev, 3188 new_entry); 3189 vmspace_map_entry_forked(vm1, vm2, new_entry); 3190 3191 /* 3192 * Update the physical map 3193 */ 3194 pmap_copy(new_map->pmap, old_map->pmap, 3195 new_entry->start, 3196 (old_entry->end - old_entry->start), 3197 old_entry->start); 3198 break; 3199 3200 case VM_INHERIT_COPY: 3201 /* 3202 * Clone the entry and link into the map. 3203 */ 3204 new_entry = vm_map_entry_create(new_map); 3205 *new_entry = *old_entry; 3206 /* 3207 * Copied entry is COW over the old object. 3208 */ 3209 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3210 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3211 new_entry->wired_count = 0; 3212 new_entry->object.vm_object = NULL; 3213 new_entry->cred = NULL; 3214 vm_map_entry_link(new_map, new_map->header.prev, 3215 new_entry); 3216 vmspace_map_entry_forked(vm1, vm2, new_entry); 3217 vm_map_copy_entry(old_map, new_map, old_entry, 3218 new_entry, fork_charge); 3219 break; 3220 } 3221 old_entry = old_entry->next; 3222 } 3223 /* 3224 * Use inlined vm_map_unlock() to postpone handling the deferred 3225 * map entries, which cannot be done until both old_map and 3226 * new_map locks are released. 3227 */ 3228 sx_xunlock(&old_map->lock); 3229 sx_xunlock(&new_map->lock); 3230 vm_map_process_deferred(); 3231 3232 return (vm2); 3233} 3234 3235int 3236vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3237 vm_prot_t prot, vm_prot_t max, int cow) 3238{ 3239 vm_map_entry_t new_entry, prev_entry; 3240 vm_offset_t bot, top; 3241 vm_size_t growsize, init_ssize; 3242 int orient, rv; 3243 rlim_t lmemlim, vmemlim; 3244 3245 /* 3246 * The stack orientation is piggybacked with the cow argument. 3247 * Extract it into orient and mask the cow argument so that we 3248 * don't pass it around further. 3249 * NOTE: We explicitly allow bi-directional stacks. 3250 */ 3251 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3252 cow &= ~orient; 3253 KASSERT(orient != 0, ("No stack grow direction")); 3254 3255 if (addrbos < vm_map_min(map) || 3256 addrbos > vm_map_max(map) || 3257 addrbos + max_ssize < addrbos) 3258 return (KERN_NO_SPACE); 3259 3260 growsize = sgrowsiz; 3261 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3262 3263 PROC_LOCK(curproc); 3264 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3265 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3266 PROC_UNLOCK(curproc); 3267 3268 vm_map_lock(map); 3269 3270 /* If addr is already mapped, no go */ 3271 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3272 vm_map_unlock(map); 3273 return (KERN_NO_SPACE); 3274 } 3275 3276 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3277 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3278 vm_map_unlock(map); 3279 return (KERN_NO_SPACE); 3280 } 3281 } 3282 3283 /* If we would blow our VMEM resource limit, no go */ 3284 if (map->size + init_ssize > vmemlim) { 3285 vm_map_unlock(map); 3286 return (KERN_NO_SPACE); 3287 } 3288 3289 /* 3290 * If we can't accomodate max_ssize in the current mapping, no go. 3291 * However, we need to be aware that subsequent user mappings might 3292 * map into the space we have reserved for stack, and currently this 3293 * space is not protected. 3294 * 3295 * Hopefully we will at least detect this condition when we try to 3296 * grow the stack. 3297 */ 3298 if ((prev_entry->next != &map->header) && 3299 (prev_entry->next->start < addrbos + max_ssize)) { 3300 vm_map_unlock(map); 3301 return (KERN_NO_SPACE); 3302 } 3303 3304 /* 3305 * We initially map a stack of only init_ssize. We will grow as 3306 * needed later. Depending on the orientation of the stack (i.e. 3307 * the grow direction) we either map at the top of the range, the 3308 * bottom of the range or in the middle. 3309 * 3310 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3311 * and cow to be 0. Possibly we should eliminate these as input 3312 * parameters, and just pass these values here in the insert call. 3313 */ 3314 if (orient == MAP_STACK_GROWS_DOWN) 3315 bot = addrbos + max_ssize - init_ssize; 3316 else if (orient == MAP_STACK_GROWS_UP) 3317 bot = addrbos; 3318 else 3319 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3320 top = bot + init_ssize; 3321 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3322 3323 /* Now set the avail_ssize amount. */ 3324 if (rv == KERN_SUCCESS) { 3325 if (prev_entry != &map->header) 3326 vm_map_clip_end(map, prev_entry, bot); 3327 new_entry = prev_entry->next; 3328 if (new_entry->end != top || new_entry->start != bot) 3329 panic("Bad entry start/end for new stack entry"); 3330 3331 new_entry->avail_ssize = max_ssize - init_ssize; 3332 if (orient & MAP_STACK_GROWS_DOWN) 3333 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3334 if (orient & MAP_STACK_GROWS_UP) 3335 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3336 } 3337 3338 vm_map_unlock(map); 3339 return (rv); 3340} 3341 3342static int stack_guard_page = 0; 3343TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3344SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3345 &stack_guard_page, 0, 3346 "Insert stack guard page ahead of the growable segments."); 3347 3348/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3349 * desired address is already mapped, or if we successfully grow 3350 * the stack. Also returns KERN_SUCCESS if addr is outside the 3351 * stack range (this is strange, but preserves compatibility with 3352 * the grow function in vm_machdep.c). 3353 */ 3354int 3355vm_map_growstack(struct proc *p, vm_offset_t addr) 3356{ 3357 vm_map_entry_t next_entry, prev_entry; 3358 vm_map_entry_t new_entry, stack_entry; 3359 struct vmspace *vm = p->p_vmspace; 3360 vm_map_t map = &vm->vm_map; 3361 vm_offset_t end; 3362 vm_size_t growsize; 3363 size_t grow_amount, max_grow; 3364 rlim_t lmemlim, stacklim, vmemlim; 3365 int is_procstack, rv; 3366 struct ucred *cred; 3367#ifdef notyet 3368 uint64_t limit; 3369#endif 3370#ifdef RACCT 3371 int error; 3372#endif 3373 3374Retry: 3375 PROC_LOCK(p); 3376 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3377 stacklim = lim_cur(p, RLIMIT_STACK); 3378 vmemlim = lim_cur(p, RLIMIT_VMEM); 3379 PROC_UNLOCK(p); 3380 3381 vm_map_lock_read(map); 3382 3383 /* If addr is already in the entry range, no need to grow.*/ 3384 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3385 vm_map_unlock_read(map); 3386 return (KERN_SUCCESS); 3387 } 3388 3389 next_entry = prev_entry->next; 3390 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3391 /* 3392 * This entry does not grow upwards. Since the address lies 3393 * beyond this entry, the next entry (if one exists) has to 3394 * be a downward growable entry. The entry list header is 3395 * never a growable entry, so it suffices to check the flags. 3396 */ 3397 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3398 vm_map_unlock_read(map); 3399 return (KERN_SUCCESS); 3400 } 3401 stack_entry = next_entry; 3402 } else { 3403 /* 3404 * This entry grows upward. If the next entry does not at 3405 * least grow downwards, this is the entry we need to grow. 3406 * otherwise we have two possible choices and we have to 3407 * select one. 3408 */ 3409 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3410 /* 3411 * We have two choices; grow the entry closest to 3412 * the address to minimize the amount of growth. 3413 */ 3414 if (addr - prev_entry->end <= next_entry->start - addr) 3415 stack_entry = prev_entry; 3416 else 3417 stack_entry = next_entry; 3418 } else 3419 stack_entry = prev_entry; 3420 } 3421 3422 if (stack_entry == next_entry) { 3423 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3424 KASSERT(addr < stack_entry->start, ("foo")); 3425 end = (prev_entry != &map->header) ? prev_entry->end : 3426 stack_entry->start - stack_entry->avail_ssize; 3427 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3428 max_grow = stack_entry->start - end; 3429 } else { 3430 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3431 KASSERT(addr >= stack_entry->end, ("foo")); 3432 end = (next_entry != &map->header) ? next_entry->start : 3433 stack_entry->end + stack_entry->avail_ssize; 3434 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3435 max_grow = end - stack_entry->end; 3436 } 3437 3438 if (grow_amount > stack_entry->avail_ssize) { 3439 vm_map_unlock_read(map); 3440 return (KERN_NO_SPACE); 3441 } 3442 3443 /* 3444 * If there is no longer enough space between the entries nogo, and 3445 * adjust the available space. Note: this should only happen if the 3446 * user has mapped into the stack area after the stack was created, 3447 * and is probably an error. 3448 * 3449 * This also effectively destroys any guard page the user might have 3450 * intended by limiting the stack size. 3451 */ 3452 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3453 if (vm_map_lock_upgrade(map)) 3454 goto Retry; 3455 3456 stack_entry->avail_ssize = max_grow; 3457 3458 vm_map_unlock(map); 3459 return (KERN_NO_SPACE); 3460 } 3461 3462 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3463 3464 /* 3465 * If this is the main process stack, see if we're over the stack 3466 * limit. 3467 */ 3468 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3469 vm_map_unlock_read(map); 3470 return (KERN_NO_SPACE); 3471 } 3472#ifdef RACCT 3473 PROC_LOCK(p); 3474 if (is_procstack && 3475 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3476 PROC_UNLOCK(p); 3477 vm_map_unlock_read(map); 3478 return (KERN_NO_SPACE); 3479 } 3480 PROC_UNLOCK(p); 3481#endif 3482 3483 /* Round up the grow amount modulo sgrowsiz */ 3484 growsize = sgrowsiz; 3485 grow_amount = roundup(grow_amount, growsize); 3486 if (grow_amount > stack_entry->avail_ssize) 3487 grow_amount = stack_entry->avail_ssize; 3488 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3489 grow_amount = trunc_page((vm_size_t)stacklim) - 3490 ctob(vm->vm_ssize); 3491 } 3492#ifdef notyet 3493 PROC_LOCK(p); 3494 limit = racct_get_available(p, RACCT_STACK); 3495 PROC_UNLOCK(p); 3496 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3497 grow_amount = limit - ctob(vm->vm_ssize); 3498#endif 3499 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3500 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3501 vm_map_unlock_read(map); 3502 rv = KERN_NO_SPACE; 3503 goto out; 3504 } 3505#ifdef RACCT 3506 PROC_LOCK(p); 3507 if (racct_set(p, RACCT_MEMLOCK, 3508 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3509 PROC_UNLOCK(p); 3510 vm_map_unlock_read(map); 3511 rv = KERN_NO_SPACE; 3512 goto out; 3513 } 3514 PROC_UNLOCK(p); 3515#endif 3516 } 3517 /* If we would blow our VMEM resource limit, no go */ 3518 if (map->size + grow_amount > vmemlim) { 3519 vm_map_unlock_read(map); 3520 rv = KERN_NO_SPACE; 3521 goto out; 3522 } 3523#ifdef RACCT 3524 PROC_LOCK(p); 3525 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3526 PROC_UNLOCK(p); 3527 vm_map_unlock_read(map); 3528 rv = KERN_NO_SPACE; 3529 goto out; 3530 } 3531 PROC_UNLOCK(p); 3532#endif 3533 3534 if (vm_map_lock_upgrade(map)) 3535 goto Retry; 3536 3537 if (stack_entry == next_entry) { 3538 /* 3539 * Growing downward. 3540 */ 3541 /* Get the preliminary new entry start value */ 3542 addr = stack_entry->start - grow_amount; 3543 3544 /* 3545 * If this puts us into the previous entry, cut back our 3546 * growth to the available space. Also, see the note above. 3547 */ 3548 if (addr < end) { 3549 stack_entry->avail_ssize = max_grow; 3550 addr = end; 3551 if (stack_guard_page) 3552 addr += PAGE_SIZE; 3553 } 3554 3555 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3556 next_entry->protection, next_entry->max_protection, 0); 3557 3558 /* Adjust the available stack space by the amount we grew. */ 3559 if (rv == KERN_SUCCESS) { 3560 if (prev_entry != &map->header) 3561 vm_map_clip_end(map, prev_entry, addr); 3562 new_entry = prev_entry->next; 3563 KASSERT(new_entry == stack_entry->prev, ("foo")); 3564 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3565 KASSERT(new_entry->start == addr, ("foo")); 3566 grow_amount = new_entry->end - new_entry->start; 3567 new_entry->avail_ssize = stack_entry->avail_ssize - 3568 grow_amount; 3569 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3570 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3571 } 3572 } else { 3573 /* 3574 * Growing upward. 3575 */ 3576 addr = stack_entry->end + grow_amount; 3577 3578 /* 3579 * If this puts us into the next entry, cut back our growth 3580 * to the available space. Also, see the note above. 3581 */ 3582 if (addr > end) { 3583 stack_entry->avail_ssize = end - stack_entry->end; 3584 addr = end; 3585 if (stack_guard_page) 3586 addr -= PAGE_SIZE; 3587 } 3588 3589 grow_amount = addr - stack_entry->end; 3590 cred = stack_entry->cred; 3591 if (cred == NULL && stack_entry->object.vm_object != NULL) 3592 cred = stack_entry->object.vm_object->cred; 3593 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3594 rv = KERN_NO_SPACE; 3595 /* Grow the underlying object if applicable. */ 3596 else if (stack_entry->object.vm_object == NULL || 3597 vm_object_coalesce(stack_entry->object.vm_object, 3598 stack_entry->offset, 3599 (vm_size_t)(stack_entry->end - stack_entry->start), 3600 (vm_size_t)grow_amount, cred != NULL)) { 3601 map->size += (addr - stack_entry->end); 3602 /* Update the current entry. */ 3603 stack_entry->end = addr; 3604 stack_entry->avail_ssize -= grow_amount; 3605 vm_map_entry_resize_free(map, stack_entry); 3606 rv = KERN_SUCCESS; 3607 3608 if (next_entry != &map->header) 3609 vm_map_clip_start(map, next_entry, addr); 3610 } else 3611 rv = KERN_FAILURE; 3612 } 3613 3614 if (rv == KERN_SUCCESS && is_procstack) 3615 vm->vm_ssize += btoc(grow_amount); 3616 3617 vm_map_unlock(map); 3618 3619 /* 3620 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3621 */ 3622 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3623 vm_map_wire(map, 3624 (stack_entry == next_entry) ? addr : addr - grow_amount, 3625 (stack_entry == next_entry) ? stack_entry->start : addr, 3626 (p->p_flag & P_SYSTEM) 3627 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3628 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3629 } 3630 3631out: 3632#ifdef RACCT 3633 if (rv != KERN_SUCCESS) { 3634 PROC_LOCK(p); 3635 error = racct_set(p, RACCT_VMEM, map->size); 3636 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3637 if (!old_mlock) { 3638 error = racct_set(p, RACCT_MEMLOCK, 3639 ptoa(pmap_wired_count(map->pmap))); 3640 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3641 } 3642 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3643 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3644 PROC_UNLOCK(p); 3645 } 3646#endif 3647 3648 return (rv); 3649} 3650 3651/* 3652 * Unshare the specified VM space for exec. If other processes are 3653 * mapped to it, then create a new one. The new vmspace is null. 3654 */ 3655int 3656vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3657{ 3658 struct vmspace *oldvmspace = p->p_vmspace; 3659 struct vmspace *newvmspace; 3660 3661 newvmspace = vmspace_alloc(minuser, maxuser); 3662 if (newvmspace == NULL) 3663 return (ENOMEM); 3664 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3665 /* 3666 * This code is written like this for prototype purposes. The 3667 * goal is to avoid running down the vmspace here, but let the 3668 * other process's that are still using the vmspace to finally 3669 * run it down. Even though there is little or no chance of blocking 3670 * here, it is a good idea to keep this form for future mods. 3671 */ 3672 PROC_VMSPACE_LOCK(p); 3673 p->p_vmspace = newvmspace; 3674 PROC_VMSPACE_UNLOCK(p); 3675 if (p == curthread->td_proc) 3676 pmap_activate(curthread); 3677 vmspace_free(oldvmspace); 3678 return (0); 3679} 3680 3681/* 3682 * Unshare the specified VM space for forcing COW. This 3683 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3684 */ 3685int 3686vmspace_unshare(struct proc *p) 3687{ 3688 struct vmspace *oldvmspace = p->p_vmspace; 3689 struct vmspace *newvmspace; 3690 vm_ooffset_t fork_charge; 3691 3692 if (oldvmspace->vm_refcnt == 1) 3693 return (0); 3694 fork_charge = 0; 3695 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3696 if (newvmspace == NULL) 3697 return (ENOMEM); 3698 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3699 vmspace_free(newvmspace); 3700 return (ENOMEM); 3701 } 3702 PROC_VMSPACE_LOCK(p); 3703 p->p_vmspace = newvmspace; 3704 PROC_VMSPACE_UNLOCK(p); 3705 if (p == curthread->td_proc) 3706 pmap_activate(curthread); 3707 vmspace_free(oldvmspace); 3708 return (0); 3709} 3710 3711/* 3712 * vm_map_lookup: 3713 * 3714 * Finds the VM object, offset, and 3715 * protection for a given virtual address in the 3716 * specified map, assuming a page fault of the 3717 * type specified. 3718 * 3719 * Leaves the map in question locked for read; return 3720 * values are guaranteed until a vm_map_lookup_done 3721 * call is performed. Note that the map argument 3722 * is in/out; the returned map must be used in 3723 * the call to vm_map_lookup_done. 3724 * 3725 * A handle (out_entry) is returned for use in 3726 * vm_map_lookup_done, to make that fast. 3727 * 3728 * If a lookup is requested with "write protection" 3729 * specified, the map may be changed to perform virtual 3730 * copying operations, although the data referenced will 3731 * remain the same. 3732 */ 3733int 3734vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3735 vm_offset_t vaddr, 3736 vm_prot_t fault_typea, 3737 vm_map_entry_t *out_entry, /* OUT */ 3738 vm_object_t *object, /* OUT */ 3739 vm_pindex_t *pindex, /* OUT */ 3740 vm_prot_t *out_prot, /* OUT */ 3741 boolean_t *wired) /* OUT */ 3742{ 3743 vm_map_entry_t entry; 3744 vm_map_t map = *var_map; 3745 vm_prot_t prot; 3746 vm_prot_t fault_type = fault_typea; 3747 vm_object_t eobject; 3748 vm_size_t size; 3749 struct ucred *cred; 3750 3751RetryLookup:; 3752 3753 vm_map_lock_read(map); 3754 3755 /* 3756 * Lookup the faulting address. 3757 */ 3758 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3759 vm_map_unlock_read(map); 3760 return (KERN_INVALID_ADDRESS); 3761 } 3762 3763 entry = *out_entry; 3764 3765 /* 3766 * Handle submaps. 3767 */ 3768 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3769 vm_map_t old_map = map; 3770 3771 *var_map = map = entry->object.sub_map; 3772 vm_map_unlock_read(old_map); 3773 goto RetryLookup; 3774 } 3775 3776 /* 3777 * Check whether this task is allowed to have this page. 3778 */ 3779 prot = entry->protection; 3780 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3781 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3782 vm_map_unlock_read(map); 3783 return (KERN_PROTECTION_FAILURE); 3784 } 3785 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3786 (entry->eflags & MAP_ENTRY_COW) && 3787 (fault_type & VM_PROT_WRITE)) { 3788 vm_map_unlock_read(map); 3789 return (KERN_PROTECTION_FAILURE); 3790 } 3791 3792 /* 3793 * If this page is not pageable, we have to get it for all possible 3794 * accesses. 3795 */ 3796 *wired = (entry->wired_count != 0); 3797 if (*wired) 3798 fault_type = entry->protection; 3799 size = entry->end - entry->start; 3800 /* 3801 * If the entry was copy-on-write, we either ... 3802 */ 3803 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3804 /* 3805 * If we want to write the page, we may as well handle that 3806 * now since we've got the map locked. 3807 * 3808 * If we don't need to write the page, we just demote the 3809 * permissions allowed. 3810 */ 3811 if ((fault_type & VM_PROT_WRITE) != 0 || 3812 (fault_typea & VM_PROT_COPY) != 0) { 3813 /* 3814 * Make a new object, and place it in the object 3815 * chain. Note that no new references have appeared 3816 * -- one just moved from the map to the new 3817 * object. 3818 */ 3819 if (vm_map_lock_upgrade(map)) 3820 goto RetryLookup; 3821 3822 if (entry->cred == NULL) { 3823 /* 3824 * The debugger owner is charged for 3825 * the memory. 3826 */ 3827 cred = curthread->td_ucred; 3828 crhold(cred); 3829 if (!swap_reserve_by_cred(size, cred)) { 3830 crfree(cred); 3831 vm_map_unlock(map); 3832 return (KERN_RESOURCE_SHORTAGE); 3833 } 3834 entry->cred = cred; 3835 } 3836 vm_object_shadow(&entry->object.vm_object, 3837 &entry->offset, size); 3838 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3839 eobject = entry->object.vm_object; 3840 if (eobject->cred != NULL) { 3841 /* 3842 * The object was not shadowed. 3843 */ 3844 swap_release_by_cred(size, entry->cred); 3845 crfree(entry->cred); 3846 entry->cred = NULL; 3847 } else if (entry->cred != NULL) { 3848 VM_OBJECT_LOCK(eobject); 3849 eobject->cred = entry->cred; 3850 eobject->charge = size; 3851 VM_OBJECT_UNLOCK(eobject); 3852 entry->cred = NULL; 3853 } 3854 3855 vm_map_lock_downgrade(map); 3856 } else { 3857 /* 3858 * We're attempting to read a copy-on-write page -- 3859 * don't allow writes. 3860 */ 3861 prot &= ~VM_PROT_WRITE; 3862 } 3863 } 3864 3865 /* 3866 * Create an object if necessary. 3867 */ 3868 if (entry->object.vm_object == NULL && 3869 !map->system_map) { 3870 if (vm_map_lock_upgrade(map)) 3871 goto RetryLookup; 3872 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3873 atop(size)); 3874 entry->offset = 0; 3875 if (entry->cred != NULL) { 3876 VM_OBJECT_LOCK(entry->object.vm_object); 3877 entry->object.vm_object->cred = entry->cred; 3878 entry->object.vm_object->charge = size; 3879 VM_OBJECT_UNLOCK(entry->object.vm_object); 3880 entry->cred = NULL; 3881 } 3882 vm_map_lock_downgrade(map); 3883 } 3884 3885 /* 3886 * Return the object/offset from this entry. If the entry was 3887 * copy-on-write or empty, it has been fixed up. 3888 */ 3889 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3890 *object = entry->object.vm_object; 3891 3892 *out_prot = prot; 3893 return (KERN_SUCCESS); 3894} 3895 3896/* 3897 * vm_map_lookup_locked: 3898 * 3899 * Lookup the faulting address. A version of vm_map_lookup that returns 3900 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3901 */ 3902int 3903vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3904 vm_offset_t vaddr, 3905 vm_prot_t fault_typea, 3906 vm_map_entry_t *out_entry, /* OUT */ 3907 vm_object_t *object, /* OUT */ 3908 vm_pindex_t *pindex, /* OUT */ 3909 vm_prot_t *out_prot, /* OUT */ 3910 boolean_t *wired) /* OUT */ 3911{ 3912 vm_map_entry_t entry; 3913 vm_map_t map = *var_map; 3914 vm_prot_t prot; 3915 vm_prot_t fault_type = fault_typea; 3916 3917 /* 3918 * Lookup the faulting address. 3919 */ 3920 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3921 return (KERN_INVALID_ADDRESS); 3922 3923 entry = *out_entry; 3924 3925 /* 3926 * Fail if the entry refers to a submap. 3927 */ 3928 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3929 return (KERN_FAILURE); 3930 3931 /* 3932 * Check whether this task is allowed to have this page. 3933 */ 3934 prot = entry->protection; 3935 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3936 if ((fault_type & prot) != fault_type) 3937 return (KERN_PROTECTION_FAILURE); 3938 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3939 (entry->eflags & MAP_ENTRY_COW) && 3940 (fault_type & VM_PROT_WRITE)) 3941 return (KERN_PROTECTION_FAILURE); 3942 3943 /* 3944 * If this page is not pageable, we have to get it for all possible 3945 * accesses. 3946 */ 3947 *wired = (entry->wired_count != 0); 3948 if (*wired) 3949 fault_type = entry->protection; 3950 3951 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3952 /* 3953 * Fail if the entry was copy-on-write for a write fault. 3954 */ 3955 if (fault_type & VM_PROT_WRITE) 3956 return (KERN_FAILURE); 3957 /* 3958 * We're attempting to read a copy-on-write page -- 3959 * don't allow writes. 3960 */ 3961 prot &= ~VM_PROT_WRITE; 3962 } 3963 3964 /* 3965 * Fail if an object should be created. 3966 */ 3967 if (entry->object.vm_object == NULL && !map->system_map) 3968 return (KERN_FAILURE); 3969 3970 /* 3971 * Return the object/offset from this entry. If the entry was 3972 * copy-on-write or empty, it has been fixed up. 3973 */ 3974 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3975 *object = entry->object.vm_object; 3976 3977 *out_prot = prot; 3978 return (KERN_SUCCESS); 3979} 3980 3981/* 3982 * vm_map_lookup_done: 3983 * 3984 * Releases locks acquired by a vm_map_lookup 3985 * (according to the handle returned by that lookup). 3986 */ 3987void 3988vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3989{ 3990 /* 3991 * Unlock the main-level map 3992 */ 3993 vm_map_unlock_read(map); 3994} 3995 3996#include "opt_ddb.h" 3997#ifdef DDB 3998#include <sys/kernel.h> 3999 4000#include <ddb/ddb.h> 4001 4002static void 4003vm_map_print(vm_map_t map) 4004{ 4005 vm_map_entry_t entry; 4006 4007 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4008 (void *)map, 4009 (void *)map->pmap, map->nentries, map->timestamp); 4010 4011 db_indent += 2; 4012 for (entry = map->header.next; entry != &map->header; 4013 entry = entry->next) { 4014 db_iprintf("map entry %p: start=%p, end=%p\n", 4015 (void *)entry, (void *)entry->start, (void *)entry->end); 4016 { 4017 static char *inheritance_name[4] = 4018 {"share", "copy", "none", "donate_copy"}; 4019 4020 db_iprintf(" prot=%x/%x/%s", 4021 entry->protection, 4022 entry->max_protection, 4023 inheritance_name[(int)(unsigned char)entry->inheritance]); 4024 if (entry->wired_count != 0) 4025 db_printf(", wired"); 4026 } 4027 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4028 db_printf(", share=%p, offset=0x%jx\n", 4029 (void *)entry->object.sub_map, 4030 (uintmax_t)entry->offset); 4031 if ((entry->prev == &map->header) || 4032 (entry->prev->object.sub_map != 4033 entry->object.sub_map)) { 4034 db_indent += 2; 4035 vm_map_print((vm_map_t)entry->object.sub_map); 4036 db_indent -= 2; 4037 } 4038 } else { 4039 if (entry->cred != NULL) 4040 db_printf(", ruid %d", entry->cred->cr_ruid); 4041 db_printf(", object=%p, offset=0x%jx", 4042 (void *)entry->object.vm_object, 4043 (uintmax_t)entry->offset); 4044 if (entry->object.vm_object && entry->object.vm_object->cred) 4045 db_printf(", obj ruid %d charge %jx", 4046 entry->object.vm_object->cred->cr_ruid, 4047 (uintmax_t)entry->object.vm_object->charge); 4048 if (entry->eflags & MAP_ENTRY_COW) 4049 db_printf(", copy (%s)", 4050 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4051 db_printf("\n"); 4052 4053 if ((entry->prev == &map->header) || 4054 (entry->prev->object.vm_object != 4055 entry->object.vm_object)) { 4056 db_indent += 2; 4057 vm_object_print((db_expr_t)(intptr_t) 4058 entry->object.vm_object, 4059 1, 0, (char *)0); 4060 db_indent -= 2; 4061 } 4062 } 4063 } 4064 db_indent -= 2; 4065} 4066 4067DB_SHOW_COMMAND(map, map) 4068{ 4069 4070 if (!have_addr) { 4071 db_printf("usage: show map <addr>\n"); 4072 return; 4073 } 4074 vm_map_print((vm_map_t)addr); 4075} 4076 4077DB_SHOW_COMMAND(procvm, procvm) 4078{ 4079 struct proc *p; 4080 4081 if (have_addr) { 4082 p = (struct proc *) addr; 4083 } else { 4084 p = curproc; 4085 } 4086 4087 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4088 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4089 (void *)vmspace_pmap(p->p_vmspace)); 4090 4091 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4092} 4093 4094#endif /* DDB */
|