1 /*- 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61 /* 62 * Virtual memory mapping module. 63 */ 64 65 #include <sys/cdefs.h> 66 __FBSDID("$FreeBSD$"); 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/kernel.h> 71 #include <sys/ktr.h> 72 #include <sys/lock.h> 73 #include <sys/mutex.h> 74 #include <sys/proc.h> 75 #include <sys/vmmeter.h> 76 #include <sys/mman.h> 77 #include <sys/vnode.h> 78 #include <sys/racct.h> 79 #include <sys/resourcevar.h> 80 #include <sys/file.h> 81 #include <sys/sysctl.h> 82 #include <sys/sysent.h> 83 #include <sys/shm.h> 84 85 #include <vm/vm.h> 86 #include <vm/vm_param.h> 87 #include <vm/pmap.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_page.h> 90 #include <vm/vm_object.h> 91 #include <vm/vm_pager.h> 92 #include <vm/vm_kern.h> 93 #include <vm/vm_extern.h> 94 #include <vm/vnode_pager.h> 95 #include <vm/swap_pager.h> 96 #include <vm/uma.h> 97 98 /* 99 * Virtual memory maps provide for the mapping, protection, 100 * and sharing of virtual memory objects. In addition, 101 * this module provides for an efficient virtual copy of 102 * memory from one map to another. 103 * 104 * Synchronization is required prior to most operations. 105 * 106 * Maps consist of an ordered doubly-linked list of simple 107 * entries; a self-adjusting binary search tree of these 108 * entries is used to speed up lookups. 109 * 110 * Since portions of maps are specified by start/end addresses, 111 * which may not align with existing map entries, all 112 * routines merely "clip" entries to these start/end values. 113 * [That is, an entry is split into two, bordering at a 114 * start or end value.] Note that these clippings may not 115 * always be necessary (as the two resulting entries are then 116 * not changed); however, the clipping is done for convenience. 117 * 118 * As mentioned above, virtual copy operations are performed 119 * by copying VM object references from one map to 120 * another, and then marking both regions as copy-on-write. 121 */ 122 123 static struct mtx map_sleep_mtx; 124 static uma_zone_t mapentzone; 125 static uma_zone_t kmapentzone; 126 static uma_zone_t mapzone; 127 static uma_zone_t vmspace_zone; 128 static struct vm_object kmapentobj; 129 static int vmspace_zinit(void *mem, int size, int flags); 130 static void vmspace_zfini(void *mem, int size); 131 static int vm_map_zinit(void *mem, int ize, int flags); 132 static void vm_map_zfini(void *mem, int size); 133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 134 vm_offset_t max); 135 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 136 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 137 #ifdef INVARIANTS 138 static void vm_map_zdtor(void *mem, int size, void *arg); 139 static void vmspace_zdtor(void *mem, int size, void *arg); 140 #endif 141 142 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 143 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 144 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 145 146 /* 147 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 148 * stable. 149 */ 150 #define PROC_VMSPACE_LOCK(p) do { } while (0) 151 #define PROC_VMSPACE_UNLOCK(p) do { } while (0) 152 153 /* 154 * VM_MAP_RANGE_CHECK: [ internal use only ] 155 * 156 * Asserts that the starting and ending region 157 * addresses fall within the valid range of the map. 158 */ 159 #define VM_MAP_RANGE_CHECK(map, start, end) \ 160 { \ 161 if (start < vm_map_min(map)) \ 162 start = vm_map_min(map); \ 163 if (end > vm_map_max(map)) \ 164 end = vm_map_max(map); \ 165 if (start > end) \ 166 start = end; \ 167 } 168 169 /* 170 * vm_map_startup: 171 * 172 * Initialize the vm_map module. Must be called before 173 * any other vm_map routines. 174 * 175 * Map and entry structures are allocated from the general 176 * purpose memory pool with some exceptions: 177 * 178 * - The kernel map and kmem submap are allocated statically. 179 * - Kernel map entries are allocated out of a static pool. 180 * 181 * These restrictions are necessary since malloc() uses the 182 * maps and requires map entries. 183 */ 184 185 void 186 vm_map_startup(void) 187 { 188 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 189 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 190 #ifdef INVARIANTS 191 vm_map_zdtor, 192 #else 193 NULL, 194 #endif 195 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 196 uma_prealloc(mapzone, MAX_KMAP); 197 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 199 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 200 uma_prealloc(kmapentzone, MAX_KMAPENT); 201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 203 } 204 205 static void 206 vmspace_zfini(void *mem, int size) 207 { 208 struct vmspace *vm; 209 210 vm = (struct vmspace *)mem; 211 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 212 } 213 214 static int 215 vmspace_zinit(void *mem, int size, int flags) 216 { 217 struct vmspace *vm; 218 219 vm = (struct vmspace *)mem; 220 221 vm->vm_map.pmap = NULL; 222 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 223 return (0); 224 } 225 226 static void 227 vm_map_zfini(void *mem, int size) 228 { 229 vm_map_t map; 230 231 map = (vm_map_t)mem; 232 mtx_destroy(&map->system_mtx); 233 sx_destroy(&map->lock); 234 } 235 236 static int 237 vm_map_zinit(void *mem, int size, int flags) 238 { 239 vm_map_t map; 240 241 map = (vm_map_t)mem; 242 map->nentries = 0; 243 map->size = 0; 244 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 245 sx_init(&map->lock, "user map"); 246 return (0); 247 } 248 249 #ifdef INVARIANTS 250 static void 251 vmspace_zdtor(void *mem, int size, void *arg) 252 { 253 struct vmspace *vm; 254 255 vm = (struct vmspace *)mem; 256 257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 258 } 259 static void 260 vm_map_zdtor(void *mem, int size, void *arg) 261 { 262 vm_map_t map; 263 264 map = (vm_map_t)mem; 265 KASSERT(map->nentries == 0, 266 ("map %p nentries == %d on free.", 267 map, map->nentries)); 268 KASSERT(map->size == 0, 269 ("map %p size == %lu on free.", 270 map, (unsigned long)map->size)); 271 } 272 #endif /* INVARIANTS */ 273 274 /* 275 * Allocate a vmspace structure, including a vm_map and pmap, 276 * and initialize those structures. The refcnt is set to 1. 277 */ 278 struct vmspace * 279 vmspace_alloc(min, max) 280 vm_offset_t min, max; 281 { 282 struct vmspace *vm; 283 284 vm = uma_zalloc(vmspace_zone, M_WAITOK); 285 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) { 286 uma_zfree(vmspace_zone, vm); 287 return (NULL); 288 } 289 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 291 vm->vm_refcnt = 1; 292 vm->vm_shm = NULL; 293 vm->vm_swrss = 0; 294 vm->vm_tsize = 0; 295 vm->vm_dsize = 0; 296 vm->vm_ssize = 0; 297 vm->vm_taddr = 0; 298 vm->vm_daddr = 0; 299 vm->vm_maxsaddr = 0; 300 return (vm); 301 } 302 303 void 304 vm_init2(void) 305 { 306 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 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 } 317 318 static void 319 vmspace_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 } 332 333 static inline void 334 vmspace_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 } 357 358 void 359 vmspace_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 } 368 369 void 370 vmspace_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 } 381 382 void 383 vmspace_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. */ 434 435 struct vmspace * 436 vmspace_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 } 462 463 void 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 } 473 474 static void 475 vm_map_process_deferred(void) 476 { 477 struct thread *td; 478 vm_map_entry_t entry; 479 vm_object_t object; 480 481 td = curthread; 482 while ((entry = td->td_map_def_user) != NULL) { 483 td->td_map_def_user = entry->next; 484 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 485 /* 486 * Decrement the object's writemappings and 487 * possibly the vnode's v_writecount. 488 */ 489 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 490 ("Submap with writecount")); 491 object = entry->object.vm_object; 492 KASSERT(object != NULL, ("No object for writecount")); 493 vnode_pager_release_writecount(object, entry->start, 494 entry->end); 495 } 496 vm_map_entry_deallocate(entry, FALSE); 497 } 498 } 499 500 void 501 _vm_map_unlock(vm_map_t map, const char *file, int line) 502 { 503 504 if (map->system_map) 505 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 506 else { 507 sx_xunlock_(&map->lock, file, line); 508 vm_map_process_deferred(); 509 } 510 } 511 512 void 513 _vm_map_lock_read(vm_map_t map, const char *file, int line) 514 { 515 516 if (map->system_map) 517 mtx_lock_flags_(&map->system_mtx, 0, file, line); 518 else 519 sx_slock_(&map->lock, file, line); 520 } 521 522 void 523 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 524 { 525 526 if (map->system_map) 527 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 528 else { 529 sx_sunlock_(&map->lock, file, line); 530 vm_map_process_deferred(); 531 } 532 } 533 534 int 535 _vm_map_trylock(vm_map_t map, const char *file, int line) 536 { 537 int error; 538 539 error = map->system_map ? 540 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 541 !sx_try_xlock_(&map->lock, file, line); 542 if (error == 0) 543 map->timestamp++; 544 return (error == 0); 545 } 546 547 int 548 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 549 { 550 int error; 551 552 error = map->system_map ? 553 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 554 !sx_try_slock_(&map->lock, file, line); 555 return (error == 0); 556 } 557 558 /* 559 * _vm_map_lock_upgrade: [ internal use only ] 560 * 561 * Tries to upgrade a read (shared) lock on the specified map to a write 562 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 563 * non-zero value if the upgrade fails. If the upgrade fails, the map is 564 * returned without a read or write lock held. 565 * 566 * Requires that the map be read locked. 567 */ 568 int 569 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 570 { 571 unsigned int last_timestamp; 572 573 if (map->system_map) { 574 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 575 } else { 576 if (!sx_try_upgrade_(&map->lock, file, line)) { 577 last_timestamp = map->timestamp; 578 sx_sunlock_(&map->lock, file, line); 579 vm_map_process_deferred(); 580 /* 581 * If the map's timestamp does not change while the 582 * map is unlocked, then the upgrade succeeds. 583 */ 584 sx_xlock_(&map->lock, file, line); 585 if (last_timestamp != map->timestamp) { 586 sx_xunlock_(&map->lock, file, line); 587 return (1); 588 } 589 } 590 } 591 map->timestamp++; 592 return (0); 593 } 594 595 void 596 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 597 { 598 599 if (map->system_map) { 600 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 601 } else 602 sx_downgrade_(&map->lock, file, line); 603 } 604 605 /* 606 * vm_map_locked: 607 * 608 * Returns a non-zero value if the caller holds a write (exclusive) lock 609 * on the specified map and the value "0" otherwise. 610 */ 611 int 612 vm_map_locked(vm_map_t map) 613 { 614 615 if (map->system_map) 616 return (mtx_owned(&map->system_mtx)); 617 else 618 return (sx_xlocked(&map->lock)); 619 } 620 621 #ifdef INVARIANTS 622 static void 623 _vm_map_assert_locked(vm_map_t map, const char *file, int line) 624 { 625 626 if (map->system_map) 627 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 628 else 629 sx_assert_(&map->lock, SA_XLOCKED, file, line); 630 } 631 632 #define VM_MAP_ASSERT_LOCKED(map) \ 633 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 634 #else 635 #define VM_MAP_ASSERT_LOCKED(map) 636 #endif 637 638 /* 639 * _vm_map_unlock_and_wait: 640 * 641 * Atomically releases the lock on the specified map and puts the calling 642 * thread to sleep. The calling thread will remain asleep until either 643 * vm_map_wakeup() is performed on the map or the specified timeout is 644 * exceeded. 645 * 646 * WARNING! This function does not perform deferred deallocations of 647 * objects and map entries. Therefore, the calling thread is expected to 648 * reacquire the map lock after reawakening and later perform an ordinary 649 * unlock operation, such as vm_map_unlock(), before completing its 650 * operation on the map. 651 */ 652 int 653 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 654 { 655 656 mtx_lock(&map_sleep_mtx); 657 if (map->system_map) 658 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 659 else 660 sx_xunlock_(&map->lock, file, line); 661 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 662 timo)); 663 } 664 665 /* 666 * vm_map_wakeup: 667 * 668 * Awaken any threads that have slept on the map using 669 * vm_map_unlock_and_wait(). 670 */ 671 void 672 vm_map_wakeup(vm_map_t map) 673 { 674 675 /* 676 * Acquire and release map_sleep_mtx to prevent a wakeup() 677 * from being performed (and lost) between the map unlock 678 * and the msleep() in _vm_map_unlock_and_wait(). 679 */ 680 mtx_lock(&map_sleep_mtx); 681 mtx_unlock(&map_sleep_mtx); 682 wakeup(&map->root); 683 } 684 685 void 686 vm_map_busy(vm_map_t map) 687 { 688 689 VM_MAP_ASSERT_LOCKED(map); 690 map->busy++; 691 } 692 693 void 694 vm_map_unbusy(vm_map_t map) 695 { 696 697 VM_MAP_ASSERT_LOCKED(map); 698 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 699 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 700 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 701 wakeup(&map->busy); 702 } 703 } 704 705 void 706 vm_map_wait_busy(vm_map_t map) 707 { 708 709 VM_MAP_ASSERT_LOCKED(map); 710 while (map->busy) { 711 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 712 if (map->system_map) 713 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 714 else 715 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 716 } 717 map->timestamp++; 718 } 719 720 long 721 vmspace_resident_count(struct vmspace *vmspace) 722 { 723 return pmap_resident_count(vmspace_pmap(vmspace)); 724 } 725 726 long 727 vmspace_wired_count(struct vmspace *vmspace) 728 { 729 return pmap_wired_count(vmspace_pmap(vmspace)); 730 } 731 732 /* 733 * vm_map_create: 734 * 735 * Creates and returns a new empty VM map with 736 * the given physical map structure, and having 737 * the given lower and upper address bounds. 738 */ 739 vm_map_t 740 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 741 { 742 vm_map_t result; 743 744 result = uma_zalloc(mapzone, M_WAITOK); 745 CTR1(KTR_VM, "vm_map_create: %p", result); 746 _vm_map_init(result, pmap, min, max); 747 return (result); 748 } 749 750 /* 751 * Initialize an existing vm_map structure 752 * such as that in the vmspace structure. 753 */ 754 static void 755 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 756 { 757 758 map->header.next = map->header.prev = &map->header; 759 map->needs_wakeup = FALSE; 760 map->system_map = 0; 761 map->pmap = pmap; 762 map->min_offset = min; 763 map->max_offset = max; 764 map->flags = 0; 765 map->root = NULL; 766 map->timestamp = 0; 767 map->busy = 0; 768 } 769 770 void 771 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 772 { 773 774 _vm_map_init(map, pmap, min, max); 775 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 776 sx_init(&map->lock, "user map"); 777 } 778 779 /* 780 * vm_map_entry_dispose: [ internal use only ] 781 * 782 * Inverse of vm_map_entry_create. 783 */ 784 static void 785 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 786 { 787 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 788 } 789 790 /* 791 * vm_map_entry_create: [ internal use only ] 792 * 793 * Allocates a VM map entry for insertion. 794 * No entry fields are filled in. 795 */ 796 static vm_map_entry_t 797 vm_map_entry_create(vm_map_t map) 798 { 799 vm_map_entry_t new_entry; 800 801 if (map->system_map) 802 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 803 else 804 new_entry = uma_zalloc(mapentzone, M_WAITOK); 805 if (new_entry == NULL) 806 panic("vm_map_entry_create: kernel resources exhausted"); 807 return (new_entry); 808 } 809 810 /* 811 * vm_map_entry_set_behavior: 812 * 813 * Set the expected access behavior, either normal, random, or 814 * sequential. 815 */ 816 static inline void 817 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 818 { 819 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 820 (behavior & MAP_ENTRY_BEHAV_MASK); 821 } 822 823 /* 824 * vm_map_entry_set_max_free: 825 * 826 * Set the max_free field in a vm_map_entry. 827 */ 828 static inline void 829 vm_map_entry_set_max_free(vm_map_entry_t entry) 830 { 831 832 entry->max_free = entry->adj_free; 833 if (entry->left != NULL && entry->left->max_free > entry->max_free) 834 entry->max_free = entry->left->max_free; 835 if (entry->right != NULL && entry->right->max_free > entry->max_free) 836 entry->max_free = entry->right->max_free; 837 } 838 839 /* 840 * vm_map_entry_splay: 841 * 842 * The Sleator and Tarjan top-down splay algorithm with the 843 * following variation. Max_free must be computed bottom-up, so 844 * on the downward pass, maintain the left and right spines in 845 * reverse order. Then, make a second pass up each side to fix 846 * the pointers and compute max_free. The time bound is O(log n) 847 * amortized. 848 * 849 * The new root is the vm_map_entry containing "addr", or else an 850 * adjacent entry (lower or higher) if addr is not in the tree. 851 * 852 * The map must be locked, and leaves it so. 853 * 854 * Returns: the new root. 855 */ 856 static vm_map_entry_t 857 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 858 { 859 vm_map_entry_t llist, rlist; 860 vm_map_entry_t ltree, rtree; 861 vm_map_entry_t y; 862 863 /* Special case of empty tree. */ 864 if (root == NULL) 865 return (root); 866 867 /* 868 * Pass One: Splay down the tree until we find addr or a NULL 869 * pointer where addr would go. llist and rlist are the two 870 * sides in reverse order (bottom-up), with llist linked by 871 * the right pointer and rlist linked by the left pointer in 872 * the vm_map_entry. Wait until Pass Two to set max_free on 873 * the two spines. 874 */ 875 llist = NULL; 876 rlist = NULL; 877 for (;;) { 878 /* root is never NULL in here. */ 879 if (addr < root->start) { 880 y = root->left; 881 if (y == NULL) 882 break; 883 if (addr < y->start && y->left != NULL) { 884 /* Rotate right and put y on rlist. */ 885 root->left = y->right; 886 y->right = root; 887 vm_map_entry_set_max_free(root); 888 root = y->left; 889 y->left = rlist; 890 rlist = y; 891 } else { 892 /* Put root on rlist. */ 893 root->left = rlist; 894 rlist = root; 895 root = y; 896 } 897 } else if (addr >= root->end) { 898 y = root->right; 899 if (y == NULL) 900 break; 901 if (addr >= y->end && y->right != NULL) { 902 /* Rotate left and put y on llist. */ 903 root->right = y->left; 904 y->left = root; 905 vm_map_entry_set_max_free(root); 906 root = y->right; 907 y->right = llist; 908 llist = y; 909 } else { 910 /* Put root on llist. */ 911 root->right = llist; 912 llist = root; 913 root = y; 914 } 915 } else 916 break; 917 } 918 919 /* 920 * Pass Two: Walk back up the two spines, flip the pointers 921 * and set max_free. The subtrees of the root go at the 922 * bottom of llist and rlist. 923 */ 924 ltree = root->left; 925 while (llist != NULL) { 926 y = llist->right; 927 llist->right = ltree; 928 vm_map_entry_set_max_free(llist); 929 ltree = llist; 930 llist = y; 931 } 932 rtree = root->right; 933 while (rlist != NULL) { 934 y = rlist->left; 935 rlist->left = rtree; 936 vm_map_entry_set_max_free(rlist); 937 rtree = rlist; 938 rlist = y; 939 } 940 941 /* 942 * Final assembly: add ltree and rtree as subtrees of root. 943 */ 944 root->left = ltree; 945 root->right = rtree; 946 vm_map_entry_set_max_free(root); 947 948 return (root); 949 } 950 951 /* 952 * vm_map_entry_{un,}link: 953 * 954 * Insert/remove entries from maps. 955 */ 956 static void 957 vm_map_entry_link(vm_map_t map, 958 vm_map_entry_t after_where, 959 vm_map_entry_t entry) 960 { 961 962 CTR4(KTR_VM, 963 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 964 map->nentries, entry, after_where); 965 VM_MAP_ASSERT_LOCKED(map); 966 map->nentries++; 967 entry->prev = after_where; 968 entry->next = after_where->next; 969 entry->next->prev = entry; 970 after_where->next = entry; 971 972 if (after_where != &map->header) { 973 if (after_where != map->root) 974 vm_map_entry_splay(after_where->start, map->root); 975 entry->right = after_where->right; 976 entry->left = after_where; 977 after_where->right = NULL; 978 after_where->adj_free = entry->start - after_where->end; 979 vm_map_entry_set_max_free(after_where); 980 } else { 981 entry->right = map->root; 982 entry->left = NULL; 983 } 984 entry->adj_free = (entry->next == &map->header ? map->max_offset : 985 entry->next->start) - entry->end; 986 vm_map_entry_set_max_free(entry); 987 map->root = entry; 988 } 989 990 static void 991 vm_map_entry_unlink(vm_map_t map, 992 vm_map_entry_t entry) 993 { 994 vm_map_entry_t next, prev, root; 995 996 VM_MAP_ASSERT_LOCKED(map); 997 if (entry != map->root) 998 vm_map_entry_splay(entry->start, map->root); 999 if (entry->left == NULL) 1000 root = entry->right; 1001 else { 1002 root = vm_map_entry_splay(entry->start, entry->left); 1003 root->right = entry->right; 1004 root->adj_free = (entry->next == &map->header ? map->max_offset : 1005 entry->next->start) - root->end; 1006 vm_map_entry_set_max_free(root); 1007 } 1008 map->root = root; 1009 1010 prev = entry->prev; 1011 next = entry->next; 1012 next->prev = prev; 1013 prev->next = next; 1014 map->nentries--; 1015 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1016 map->nentries, entry); 1017 } 1018 1019 /* 1020 * vm_map_entry_resize_free: 1021 * 1022 * Recompute the amount of free space following a vm_map_entry 1023 * and propagate that value up the tree. Call this function after 1024 * resizing a map entry in-place, that is, without a call to 1025 * vm_map_entry_link() or _unlink(). 1026 * 1027 * The map must be locked, and leaves it so. 1028 */ 1029 static void 1030 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1031 { 1032 1033 /* 1034 * Using splay trees without parent pointers, propagating 1035 * max_free up the tree is done by moving the entry to the 1036 * root and making the change there. 1037 */ 1038 if (entry != map->root) 1039 map->root = vm_map_entry_splay(entry->start, map->root); 1040 1041 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1042 entry->next->start) - entry->end; 1043 vm_map_entry_set_max_free(entry); 1044 } 1045 1046 /* 1047 * vm_map_lookup_entry: [ internal use only ] 1048 * 1049 * Finds the map entry containing (or 1050 * immediately preceding) the specified address 1051 * in the given map; the entry is returned 1052 * in the "entry" parameter. The boolean 1053 * result indicates whether the address is 1054 * actually contained in the map. 1055 */ 1056 boolean_t 1057 vm_map_lookup_entry( 1058 vm_map_t map, 1059 vm_offset_t address, 1060 vm_map_entry_t *entry) /* OUT */ 1061 { 1062 vm_map_entry_t cur; 1063 boolean_t locked; 1064 1065 /* 1066 * If the map is empty, then the map entry immediately preceding 1067 * "address" is the map's header. 1068 */ 1069 cur = map->root; 1070 if (cur == NULL) 1071 *entry = &map->header; 1072 else if (address >= cur->start && cur->end > address) { 1073 *entry = cur; 1074 return (TRUE); 1075 } else if ((locked = vm_map_locked(map)) || 1076 sx_try_upgrade(&map->lock)) { 1077 /* 1078 * Splay requires a write lock on the map. However, it only 1079 * restructures the binary search tree; it does not otherwise 1080 * change the map. Thus, the map's timestamp need not change 1081 * on a temporary upgrade. 1082 */ 1083 map->root = cur = vm_map_entry_splay(address, cur); 1084 if (!locked) 1085 sx_downgrade(&map->lock); 1086 1087 /* 1088 * If "address" is contained within a map entry, the new root 1089 * is that map entry. Otherwise, the new root is a map entry 1090 * immediately before or after "address". 1091 */ 1092 if (address >= cur->start) { 1093 *entry = cur; 1094 if (cur->end > address) 1095 return (TRUE); 1096 } else 1097 *entry = cur->prev; 1098 } else 1099 /* 1100 * Since the map is only locked for read access, perform a 1101 * standard binary search tree lookup for "address". 1102 */ 1103 for (;;) { 1104 if (address < cur->start) { 1105 if (cur->left == NULL) { 1106 *entry = cur->prev; 1107 break; 1108 } 1109 cur = cur->left; 1110 } else if (cur->end > address) { 1111 *entry = cur; 1112 return (TRUE); 1113 } else { 1114 if (cur->right == NULL) { 1115 *entry = cur; 1116 break; 1117 } 1118 cur = cur->right; 1119 } 1120 } 1121 return (FALSE); 1122 } 1123 1124 /* 1125 * vm_map_insert: 1126 * 1127 * Inserts the given whole VM object into the target 1128 * map at the specified address range. The object's 1129 * size should match that of the address range. 1130 * 1131 * Requires that the map be locked, and leaves it so. 1132 * 1133 * If object is non-NULL, ref count must be bumped by caller 1134 * prior to making call to account for the new entry. 1135 */ 1136 int 1137 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1138 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1139 int cow) 1140 { 1141 vm_map_entry_t new_entry; 1142 vm_map_entry_t prev_entry; 1143 vm_map_entry_t temp_entry; 1144 vm_eflags_t protoeflags; 1145 struct ucred *cred; 1146 vm_inherit_t inheritance; 1147 boolean_t charge_prev_obj; 1148 1149 VM_MAP_ASSERT_LOCKED(map); 1150 1151 /* 1152 * Check that the start and end points are not bogus. 1153 */ 1154 if ((start < map->min_offset) || (end > map->max_offset) || 1155 (start >= end)) 1156 return (KERN_INVALID_ADDRESS); 1157 1158 /* 1159 * Find the entry prior to the proposed starting address; if it's part 1160 * of an existing entry, this range is bogus. 1161 */ 1162 if (vm_map_lookup_entry(map, start, &temp_entry)) 1163 return (KERN_NO_SPACE); 1164 1165 prev_entry = temp_entry; 1166 1167 /* 1168 * Assert that the next entry doesn't overlap the end point. 1169 */ 1170 if ((prev_entry->next != &map->header) && 1171 (prev_entry->next->start < end)) 1172 return (KERN_NO_SPACE); 1173 1174 protoeflags = 0; 1175 charge_prev_obj = FALSE; 1176 1177 if (cow & MAP_COPY_ON_WRITE) 1178 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1179 1180 if (cow & MAP_NOFAULT) { 1181 protoeflags |= MAP_ENTRY_NOFAULT; 1182 1183 KASSERT(object == NULL, 1184 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1185 } 1186 if (cow & MAP_DISABLE_SYNCER) 1187 protoeflags |= MAP_ENTRY_NOSYNC; 1188 if (cow & MAP_DISABLE_COREDUMP) 1189 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1190 if (cow & MAP_VN_WRITECOUNT) 1191 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1192 if (cow & MAP_INHERIT_SHARE) 1193 inheritance = VM_INHERIT_SHARE; 1194 else 1195 inheritance = VM_INHERIT_DEFAULT; 1196 1197 cred = NULL; 1198 KASSERT((object != kmem_object && object != kernel_object) || 1199 ((object == kmem_object || object == kernel_object) && 1200 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1201 ("kmem or kernel object and cow")); 1202 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1203 goto charged; 1204 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1205 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1206 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1207 return (KERN_RESOURCE_SHORTAGE); 1208 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1209 object->cred == NULL, 1210 ("OVERCOMMIT: vm_map_insert o %p", object)); 1211 cred = curthread->td_ucred; 1212 crhold(cred); 1213 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1214 charge_prev_obj = TRUE; 1215 } 1216 1217 charged: 1218 /* Expand the kernel pmap, if necessary. */ 1219 if (map == kernel_map && end > kernel_vm_end) 1220 pmap_growkernel(end); 1221 if (object != NULL) { 1222 /* 1223 * OBJ_ONEMAPPING must be cleared unless this mapping 1224 * is trivially proven to be the only mapping for any 1225 * of the object's pages. (Object granularity 1226 * reference counting is insufficient to recognize 1227 * aliases with precision.) 1228 */ 1229 VM_OBJECT_LOCK(object); 1230 if (object->ref_count > 1 || object->shadow_count != 0) 1231 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1232 VM_OBJECT_UNLOCK(object); 1233 } 1234 else if ((prev_entry != &map->header) && 1235 (prev_entry->eflags == protoeflags) && 1236 (prev_entry->end == start) && 1237 (prev_entry->wired_count == 0) && 1238 (prev_entry->cred == cred || 1239 (prev_entry->object.vm_object != NULL && 1240 (prev_entry->object.vm_object->cred == cred))) && 1241 vm_object_coalesce(prev_entry->object.vm_object, 1242 prev_entry->offset, 1243 (vm_size_t)(prev_entry->end - prev_entry->start), 1244 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1245 /* 1246 * We were able to extend the object. Determine if we 1247 * can extend the previous map entry to include the 1248 * new range as well. 1249 */ 1250 if ((prev_entry->inheritance == inheritance) && 1251 (prev_entry->protection == prot) && 1252 (prev_entry->max_protection == max)) { 1253 map->size += (end - prev_entry->end); 1254 prev_entry->end = end; 1255 vm_map_entry_resize_free(map, prev_entry); 1256 vm_map_simplify_entry(map, prev_entry); 1257 if (cred != NULL) 1258 crfree(cred); 1259 return (KERN_SUCCESS); 1260 } 1261 1262 /* 1263 * If we can extend the object but cannot extend the 1264 * map entry, we have to create a new map entry. We 1265 * must bump the ref count on the extended object to 1266 * account for it. object may be NULL. 1267 */ 1268 object = prev_entry->object.vm_object; 1269 offset = prev_entry->offset + 1270 (prev_entry->end - prev_entry->start); 1271 vm_object_reference(object); 1272 if (cred != NULL && object != NULL && object->cred != NULL && 1273 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1274 /* Object already accounts for this uid. */ 1275 crfree(cred); 1276 cred = NULL; 1277 } 1278 } 1279 1280 /* 1281 * NOTE: if conditionals fail, object can be NULL here. This occurs 1282 * in things like the buffer map where we manage kva but do not manage 1283 * backing objects. 1284 */ 1285 1286 /* 1287 * Create a new entry 1288 */ 1289 new_entry = vm_map_entry_create(map); 1290 new_entry->start = start; 1291 new_entry->end = end; 1292 new_entry->cred = NULL; 1293 1294 new_entry->eflags = protoeflags; 1295 new_entry->object.vm_object = object; 1296 new_entry->offset = offset; 1297 new_entry->avail_ssize = 0; 1298 1299 new_entry->inheritance = inheritance; 1300 new_entry->protection = prot; 1301 new_entry->max_protection = max; 1302 new_entry->wired_count = 0; 1303 1304 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1305 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1306 new_entry->cred = cred; 1307 1308 /* 1309 * Insert the new entry into the list 1310 */ 1311 vm_map_entry_link(map, prev_entry, new_entry); 1312 map->size += new_entry->end - new_entry->start; 1313 1314 /* 1315 * It may be possible to merge the new entry with the next and/or 1316 * previous entries. However, due to MAP_STACK_* being a hack, a 1317 * panic can result from merging such entries. 1318 */ 1319 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1320 vm_map_simplify_entry(map, new_entry); 1321 1322 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1323 vm_map_pmap_enter(map, start, prot, 1324 object, OFF_TO_IDX(offset), end - start, 1325 cow & MAP_PREFAULT_PARTIAL); 1326 } 1327 1328 return (KERN_SUCCESS); 1329 } 1330 1331 /* 1332 * vm_map_findspace: 1333 * 1334 * Find the first fit (lowest VM address) for "length" free bytes 1335 * beginning at address >= start in the given map. 1336 * 1337 * In a vm_map_entry, "adj_free" is the amount of free space 1338 * adjacent (higher address) to this entry, and "max_free" is the 1339 * maximum amount of contiguous free space in its subtree. This 1340 * allows finding a free region in one path down the tree, so 1341 * O(log n) amortized with splay trees. 1342 * 1343 * The map must be locked, and leaves it so. 1344 * 1345 * Returns: 0 on success, and starting address in *addr, 1346 * 1 if insufficient space. 1347 */ 1348 int 1349 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1350 vm_offset_t *addr) /* OUT */ 1351 { 1352 vm_map_entry_t entry; 1353 vm_offset_t st; 1354 1355 /* 1356 * Request must fit within min/max VM address and must avoid 1357 * address wrap. 1358 */ 1359 if (start < map->min_offset) 1360 start = map->min_offset; 1361 if (start + length > map->max_offset || start + length < start) 1362 return (1); 1363 1364 /* Empty tree means wide open address space. */ 1365 if (map->root == NULL) { 1366 *addr = start; 1367 return (0); 1368 } 1369 1370 /* 1371 * After splay, if start comes before root node, then there 1372 * must be a gap from start to the root. 1373 */ 1374 map->root = vm_map_entry_splay(start, map->root); 1375 if (start + length <= map->root->start) { 1376 *addr = start; 1377 return (0); 1378 } 1379 1380 /* 1381 * Root is the last node that might begin its gap before 1382 * start, and this is the last comparison where address 1383 * wrap might be a problem. 1384 */ 1385 st = (start > map->root->end) ? start : map->root->end; 1386 if (length <= map->root->end + map->root->adj_free - st) { 1387 *addr = st; 1388 return (0); 1389 } 1390 1391 /* With max_free, can immediately tell if no solution. */ 1392 entry = map->root->right; 1393 if (entry == NULL || length > entry->max_free) 1394 return (1); 1395 1396 /* 1397 * Search the right subtree in the order: left subtree, root, 1398 * right subtree (first fit). The previous splay implies that 1399 * all regions in the right subtree have addresses > start. 1400 */ 1401 while (entry != NULL) { 1402 if (entry->left != NULL && entry->left->max_free >= length) 1403 entry = entry->left; 1404 else if (entry->adj_free >= length) { 1405 *addr = entry->end; 1406 return (0); 1407 } else 1408 entry = entry->right; 1409 } 1410 1411 /* Can't get here, so panic if we do. */ 1412 panic("vm_map_findspace: max_free corrupt"); 1413 } 1414 1415 int 1416 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1417 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1418 vm_prot_t max, int cow) 1419 { 1420 vm_offset_t end; 1421 int result; 1422 1423 end = start + length; 1424 vm_map_lock(map); 1425 VM_MAP_RANGE_CHECK(map, start, end); 1426 (void) vm_map_delete(map, start, end); 1427 result = vm_map_insert(map, object, offset, start, end, prot, 1428 max, cow); 1429 vm_map_unlock(map); 1430 return (result); 1431 } 1432 1433 /* 1434 * vm_map_find finds an unallocated region in the target address 1435 * map with the given length. The search is defined to be 1436 * first-fit from the specified address; the region found is 1437 * returned in the same parameter. 1438 * 1439 * If object is non-NULL, ref count must be bumped by caller 1440 * prior to making call to account for the new entry. 1441 */ 1442 int 1443 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1444 vm_offset_t *addr, /* IN/OUT */ 1445 vm_size_t length, int find_space, vm_prot_t prot, 1446 vm_prot_t max, int cow) 1447 { 1448 vm_offset_t start; 1449 int result; 1450 1451 start = *addr; 1452 vm_map_lock(map); 1453 do { 1454 if (find_space != VMFS_NO_SPACE) { 1455 if (vm_map_findspace(map, start, length, addr)) { 1456 vm_map_unlock(map); 1457 return (KERN_NO_SPACE); 1458 } 1459 switch (find_space) { 1460 case VMFS_ALIGNED_SPACE: 1461 pmap_align_superpage(object, offset, addr, 1462 length); 1463 break; 1464 #ifdef VMFS_TLB_ALIGNED_SPACE 1465 case VMFS_TLB_ALIGNED_SPACE: 1466 pmap_align_tlb(addr); 1467 break; 1468 #endif 1469 default: 1470 break; 1471 } 1472 1473 start = *addr; 1474 } 1475 result = vm_map_insert(map, object, offset, start, start + 1476 length, prot, max, cow); 1477 } while (result == KERN_NO_SPACE && (find_space == VMFS_ALIGNED_SPACE 1478 #ifdef VMFS_TLB_ALIGNED_SPACE 1479 || find_space == VMFS_TLB_ALIGNED_SPACE 1480 #endif 1481 )); 1482 vm_map_unlock(map); 1483 return (result); 1484 } 1485 1486 /* 1487 * vm_map_simplify_entry: 1488 * 1489 * Simplify the given map entry by merging with either neighbor. This 1490 * routine also has the ability to merge with both neighbors. 1491 * 1492 * The map must be locked. 1493 * 1494 * This routine guarentees that the passed entry remains valid (though 1495 * possibly extended). When merging, this routine may delete one or 1496 * both neighbors. 1497 */ 1498 void 1499 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1500 { 1501 vm_map_entry_t next, prev; 1502 vm_size_t prevsize, esize; 1503 1504 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1505 return; 1506 1507 prev = entry->prev; 1508 if (prev != &map->header) { 1509 prevsize = prev->end - prev->start; 1510 if ( (prev->end == entry->start) && 1511 (prev->object.vm_object == entry->object.vm_object) && 1512 (!prev->object.vm_object || 1513 (prev->offset + prevsize == entry->offset)) && 1514 (prev->eflags == entry->eflags) && 1515 (prev->protection == entry->protection) && 1516 (prev->max_protection == entry->max_protection) && 1517 (prev->inheritance == entry->inheritance) && 1518 (prev->wired_count == entry->wired_count) && 1519 (prev->cred == entry->cred)) { 1520 vm_map_entry_unlink(map, prev); 1521 entry->start = prev->start; 1522 entry->offset = prev->offset; 1523 if (entry->prev != &map->header) 1524 vm_map_entry_resize_free(map, entry->prev); 1525 1526 /* 1527 * If the backing object is a vnode object, 1528 * vm_object_deallocate() calls vrele(). 1529 * However, vrele() does not lock the vnode 1530 * because the vnode has additional 1531 * references. Thus, the map lock can be kept 1532 * without causing a lock-order reversal with 1533 * the vnode lock. 1534 * 1535 * Since we count the number of virtual page 1536 * mappings in object->un_pager.vnp.writemappings, 1537 * the writemappings value should not be adjusted 1538 * when the entry is disposed of. 1539 */ 1540 if (prev->object.vm_object) 1541 vm_object_deallocate(prev->object.vm_object); 1542 if (prev->cred != NULL) 1543 crfree(prev->cred); 1544 vm_map_entry_dispose(map, prev); 1545 } 1546 } 1547 1548 next = entry->next; 1549 if (next != &map->header) { 1550 esize = entry->end - entry->start; 1551 if ((entry->end == next->start) && 1552 (next->object.vm_object == entry->object.vm_object) && 1553 (!entry->object.vm_object || 1554 (entry->offset + esize == next->offset)) && 1555 (next->eflags == entry->eflags) && 1556 (next->protection == entry->protection) && 1557 (next->max_protection == entry->max_protection) && 1558 (next->inheritance == entry->inheritance) && 1559 (next->wired_count == entry->wired_count) && 1560 (next->cred == entry->cred)) { 1561 vm_map_entry_unlink(map, next); 1562 entry->end = next->end; 1563 vm_map_entry_resize_free(map, entry); 1564 1565 /* 1566 * See comment above. 1567 */ 1568 if (next->object.vm_object) 1569 vm_object_deallocate(next->object.vm_object); 1570 if (next->cred != NULL) 1571 crfree(next->cred); 1572 vm_map_entry_dispose(map, next); 1573 } 1574 } 1575 } 1576 /* 1577 * vm_map_clip_start: [ internal use only ] 1578 * 1579 * Asserts that the given entry begins at or after 1580 * the specified address; if necessary, 1581 * it splits the entry into two. 1582 */ 1583 #define vm_map_clip_start(map, entry, startaddr) \ 1584 { \ 1585 if (startaddr > entry->start) \ 1586 _vm_map_clip_start(map, entry, startaddr); \ 1587 } 1588 1589 /* 1590 * This routine is called only when it is known that 1591 * the entry must be split. 1592 */ 1593 static void 1594 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1595 { 1596 vm_map_entry_t new_entry; 1597 1598 VM_MAP_ASSERT_LOCKED(map); 1599 1600 /* 1601 * Split off the front portion -- note that we must insert the new 1602 * entry BEFORE this one, so that this entry has the specified 1603 * starting address. 1604 */ 1605 vm_map_simplify_entry(map, entry); 1606 1607 /* 1608 * If there is no object backing this entry, we might as well create 1609 * one now. If we defer it, an object can get created after the map 1610 * is clipped, and individual objects will be created for the split-up 1611 * map. This is a bit of a hack, but is also about the best place to 1612 * put this improvement. 1613 */ 1614 if (entry->object.vm_object == NULL && !map->system_map) { 1615 vm_object_t object; 1616 object = vm_object_allocate(OBJT_DEFAULT, 1617 atop(entry->end - entry->start)); 1618 entry->object.vm_object = object; 1619 entry->offset = 0; 1620 if (entry->cred != NULL) { 1621 object->cred = entry->cred; 1622 object->charge = entry->end - entry->start; 1623 entry->cred = NULL; 1624 } 1625 } else if (entry->object.vm_object != NULL && 1626 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1627 entry->cred != NULL) { 1628 VM_OBJECT_LOCK(entry->object.vm_object); 1629 KASSERT(entry->object.vm_object->cred == NULL, 1630 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1631 entry->object.vm_object->cred = entry->cred; 1632 entry->object.vm_object->charge = entry->end - entry->start; 1633 VM_OBJECT_UNLOCK(entry->object.vm_object); 1634 entry->cred = NULL; 1635 } 1636 1637 new_entry = vm_map_entry_create(map); 1638 *new_entry = *entry; 1639 1640 new_entry->end = start; 1641 entry->offset += (start - entry->start); 1642 entry->start = start; 1643 if (new_entry->cred != NULL) 1644 crhold(entry->cred); 1645 1646 vm_map_entry_link(map, entry->prev, new_entry); 1647 1648 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1649 vm_object_reference(new_entry->object.vm_object); 1650 /* 1651 * The object->un_pager.vnp.writemappings for the 1652 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1653 * kept as is here. The virtual pages are 1654 * re-distributed among the clipped entries, so the sum is 1655 * left the same. 1656 */ 1657 } 1658 } 1659 1660 /* 1661 * vm_map_clip_end: [ internal use only ] 1662 * 1663 * Asserts that the given entry ends at or before 1664 * the specified address; if necessary, 1665 * it splits the entry into two. 1666 */ 1667 #define vm_map_clip_end(map, entry, endaddr) \ 1668 { \ 1669 if ((endaddr) < (entry->end)) \ 1670 _vm_map_clip_end((map), (entry), (endaddr)); \ 1671 } 1672 1673 /* 1674 * This routine is called only when it is known that 1675 * the entry must be split. 1676 */ 1677 static void 1678 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1679 { 1680 vm_map_entry_t new_entry; 1681 1682 VM_MAP_ASSERT_LOCKED(map); 1683 1684 /* 1685 * If there is no object backing this entry, we might as well create 1686 * one now. If we defer it, an object can get created after the map 1687 * is clipped, and individual objects will be created for the split-up 1688 * map. This is a bit of a hack, but is also about the best place to 1689 * put this improvement. 1690 */ 1691 if (entry->object.vm_object == NULL && !map->system_map) { 1692 vm_object_t object; 1693 object = vm_object_allocate(OBJT_DEFAULT, 1694 atop(entry->end - entry->start)); 1695 entry->object.vm_object = object; 1696 entry->offset = 0; 1697 if (entry->cred != NULL) { 1698 object->cred = entry->cred; 1699 object->charge = entry->end - entry->start; 1700 entry->cred = NULL; 1701 } 1702 } else if (entry->object.vm_object != NULL && 1703 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1704 entry->cred != NULL) { 1705 VM_OBJECT_LOCK(entry->object.vm_object); 1706 KASSERT(entry->object.vm_object->cred == NULL, 1707 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1708 entry->object.vm_object->cred = entry->cred; 1709 entry->object.vm_object->charge = entry->end - entry->start; 1710 VM_OBJECT_UNLOCK(entry->object.vm_object); 1711 entry->cred = NULL; 1712 } 1713 1714 /* 1715 * Create a new entry and insert it AFTER the specified entry 1716 */ 1717 new_entry = vm_map_entry_create(map); 1718 *new_entry = *entry; 1719 1720 new_entry->start = entry->end = end; 1721 new_entry->offset += (end - entry->start); 1722 if (new_entry->cred != NULL) 1723 crhold(entry->cred); 1724 1725 vm_map_entry_link(map, entry, new_entry); 1726 1727 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1728 vm_object_reference(new_entry->object.vm_object); 1729 } 1730 } 1731 1732 /* 1733 * vm_map_submap: [ kernel use only ] 1734 * 1735 * Mark the given range as handled by a subordinate map. 1736 * 1737 * This range must have been created with vm_map_find, 1738 * and no other operations may have been performed on this 1739 * range prior to calling vm_map_submap. 1740 * 1741 * Only a limited number of operations can be performed 1742 * within this rage after calling vm_map_submap: 1743 * vm_fault 1744 * [Don't try vm_map_copy!] 1745 * 1746 * To remove a submapping, one must first remove the 1747 * range from the superior map, and then destroy the 1748 * submap (if desired). [Better yet, don't try it.] 1749 */ 1750 int 1751 vm_map_submap( 1752 vm_map_t map, 1753 vm_offset_t start, 1754 vm_offset_t end, 1755 vm_map_t submap) 1756 { 1757 vm_map_entry_t entry; 1758 int result = KERN_INVALID_ARGUMENT; 1759 1760 vm_map_lock(map); 1761 1762 VM_MAP_RANGE_CHECK(map, start, end); 1763 1764 if (vm_map_lookup_entry(map, start, &entry)) { 1765 vm_map_clip_start(map, entry, start); 1766 } else 1767 entry = entry->next; 1768 1769 vm_map_clip_end(map, entry, end); 1770 1771 if ((entry->start == start) && (entry->end == end) && 1772 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1773 (entry->object.vm_object == NULL)) { 1774 entry->object.sub_map = submap; 1775 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1776 result = KERN_SUCCESS; 1777 } 1778 vm_map_unlock(map); 1779 1780 return (result); 1781 } 1782 1783 /* 1784 * The maximum number of pages to map 1785 */ 1786 #define MAX_INIT_PT 96 1787 1788 /* 1789 * vm_map_pmap_enter: 1790 * 1791 * Preload read-only mappings for the given object's resident pages into 1792 * the given map. This eliminates the soft faults on process startup and 1793 * immediately after an mmap(2). Because these are speculative mappings, 1794 * cached pages are not reactivated and mapped. 1795 */ 1796 void 1797 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1798 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1799 { 1800 vm_offset_t start; 1801 vm_page_t p, p_start; 1802 vm_pindex_t psize, tmpidx; 1803 1804 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1805 return; 1806 VM_OBJECT_LOCK(object); 1807 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1808 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1809 goto unlock_return; 1810 } 1811 1812 psize = atop(size); 1813 1814 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT && 1815 object->resident_page_count > MAX_INIT_PT) 1816 goto unlock_return; 1817 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); 1859 unlock_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 */ 1871 int 1872 vm_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 */ 2010 int 2011 vm_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 pindex; 2104 int count; 2105 2106 /* 2107 * madvise behaviors that are implemented in the underlying 2108 * vm_object. 2109 * 2110 * Since we don't clip the vm_map_entry, we have to clip 2111 * the vm_object pindex and count. 2112 */ 2113 for (current = entry; 2114 (current != &map->header) && (current->start < end); 2115 current = current->next 2116 ) { 2117 vm_offset_t useStart; 2118 2119 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2120 continue; 2121 2122 pindex = OFF_TO_IDX(current->offset); 2123 count = atop(current->end - current->start); 2124 useStart = current->start; 2125 2126 if (current->start < start) { 2127 pindex += atop(start - current->start); 2128 count -= atop(start - current->start); 2129 useStart = start; 2130 } 2131 if (current->end > end) 2132 count -= atop(current->end - end); 2133 2134 if (count <= 0) 2135 continue; 2136 2137 vm_object_madvise(current->object.vm_object, 2138 pindex, count, behav); 2139 if (behav == MADV_WILLNEED) { 2140 vm_map_pmap_enter(map, 2141 useStart, 2142 current->protection, 2143 current->object.vm_object, 2144 pindex, 2145 (count << PAGE_SHIFT), 2146 MAP_PREFAULT_MADVISE 2147 ); 2148 } 2149 } 2150 vm_map_unlock_read(map); 2151 } 2152 return (0); 2153 } 2154 2155 2156 /* 2157 * vm_map_inherit: 2158 * 2159 * Sets the inheritance of the specified address 2160 * range in the target map. Inheritance 2161 * affects how the map will be shared with 2162 * child maps at the time of vmspace_fork. 2163 */ 2164 int 2165 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2166 vm_inherit_t new_inheritance) 2167 { 2168 vm_map_entry_t entry; 2169 vm_map_entry_t temp_entry; 2170 2171 switch (new_inheritance) { 2172 case VM_INHERIT_NONE: 2173 case VM_INHERIT_COPY: 2174 case VM_INHERIT_SHARE: 2175 break; 2176 default: 2177 return (KERN_INVALID_ARGUMENT); 2178 } 2179 vm_map_lock(map); 2180 VM_MAP_RANGE_CHECK(map, start, end); 2181 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2182 entry = temp_entry; 2183 vm_map_clip_start(map, entry, start); 2184 } else 2185 entry = temp_entry->next; 2186 while ((entry != &map->header) && (entry->start < end)) { 2187 vm_map_clip_end(map, entry, end); 2188 entry->inheritance = new_inheritance; 2189 vm_map_simplify_entry(map, entry); 2190 entry = entry->next; 2191 } 2192 vm_map_unlock(map); 2193 return (KERN_SUCCESS); 2194 } 2195 2196 /* 2197 * vm_map_unwire: 2198 * 2199 * Implements both kernel and user unwiring. 2200 */ 2201 int 2202 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2203 int flags) 2204 { 2205 vm_map_entry_t entry, first_entry, tmp_entry; 2206 vm_offset_t saved_start; 2207 unsigned int last_timestamp; 2208 int rv; 2209 boolean_t need_wakeup, result, user_unwire; 2210 2211 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2212 vm_map_lock(map); 2213 VM_MAP_RANGE_CHECK(map, start, end); 2214 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2215 if (flags & VM_MAP_WIRE_HOLESOK) 2216 first_entry = first_entry->next; 2217 else { 2218 vm_map_unlock(map); 2219 return (KERN_INVALID_ADDRESS); 2220 } 2221 } 2222 last_timestamp = map->timestamp; 2223 entry = first_entry; 2224 while (entry != &map->header && entry->start < end) { 2225 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2226 /* 2227 * We have not yet clipped the entry. 2228 */ 2229 saved_start = (start >= entry->start) ? start : 2230 entry->start; 2231 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2232 if (vm_map_unlock_and_wait(map, 0)) { 2233 /* 2234 * Allow interruption of user unwiring? 2235 */ 2236 } 2237 vm_map_lock(map); 2238 if (last_timestamp+1 != map->timestamp) { 2239 /* 2240 * Look again for the entry because the map was 2241 * modified while it was unlocked. 2242 * Specifically, the entry may have been 2243 * clipped, merged, or deleted. 2244 */ 2245 if (!vm_map_lookup_entry(map, saved_start, 2246 &tmp_entry)) { 2247 if (flags & VM_MAP_WIRE_HOLESOK) 2248 tmp_entry = tmp_entry->next; 2249 else { 2250 if (saved_start == start) { 2251 /* 2252 * First_entry has been deleted. 2253 */ 2254 vm_map_unlock(map); 2255 return (KERN_INVALID_ADDRESS); 2256 } 2257 end = saved_start; 2258 rv = KERN_INVALID_ADDRESS; 2259 goto done; 2260 } 2261 } 2262 if (entry == first_entry) 2263 first_entry = tmp_entry; 2264 else 2265 first_entry = NULL; 2266 entry = tmp_entry; 2267 } 2268 last_timestamp = map->timestamp; 2269 continue; 2270 } 2271 vm_map_clip_start(map, entry, start); 2272 vm_map_clip_end(map, entry, end); 2273 /* 2274 * Mark the entry in case the map lock is released. (See 2275 * above.) 2276 */ 2277 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2278 /* 2279 * Check the map for holes in the specified region. 2280 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2281 */ 2282 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2283 (entry->end < end && (entry->next == &map->header || 2284 entry->next->start > entry->end))) { 2285 end = entry->end; 2286 rv = KERN_INVALID_ADDRESS; 2287 goto done; 2288 } 2289 /* 2290 * If system unwiring, require that the entry is system wired. 2291 */ 2292 if (!user_unwire && 2293 vm_map_entry_system_wired_count(entry) == 0) { 2294 end = entry->end; 2295 rv = KERN_INVALID_ARGUMENT; 2296 goto done; 2297 } 2298 entry = entry->next; 2299 } 2300 rv = KERN_SUCCESS; 2301 done: 2302 need_wakeup = FALSE; 2303 if (first_entry == NULL) { 2304 result = vm_map_lookup_entry(map, start, &first_entry); 2305 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2306 first_entry = first_entry->next; 2307 else 2308 KASSERT(result, ("vm_map_unwire: lookup failed")); 2309 } 2310 entry = first_entry; 2311 while (entry != &map->header && entry->start < end) { 2312 if (rv == KERN_SUCCESS && (!user_unwire || 2313 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2314 if (user_unwire) 2315 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2316 entry->wired_count--; 2317 if (entry->wired_count == 0) { 2318 /* 2319 * Retain the map lock. 2320 */ 2321 vm_fault_unwire(map, entry->start, entry->end, 2322 entry->object.vm_object != NULL && 2323 (entry->object.vm_object->type == OBJT_DEVICE || 2324 entry->object.vm_object->type == OBJT_SG)); 2325 } 2326 } 2327 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2328 ("vm_map_unwire: in-transition flag missing")); 2329 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2330 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2331 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2332 need_wakeup = TRUE; 2333 } 2334 vm_map_simplify_entry(map, entry); 2335 entry = entry->next; 2336 } 2337 vm_map_unlock(map); 2338 if (need_wakeup) 2339 vm_map_wakeup(map); 2340 return (rv); 2341 } 2342 2343 /* 2344 * vm_map_wire: 2345 * 2346 * Implements both kernel and user wiring. 2347 */ 2348 int 2349 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2350 int flags) 2351 { 2352 vm_map_entry_t entry, first_entry, tmp_entry; 2353 vm_offset_t saved_end, saved_start; 2354 unsigned int last_timestamp; 2355 int rv; 2356 boolean_t fictitious, need_wakeup, result, user_wire; 2357 vm_prot_t prot; 2358 2359 prot = 0; 2360 if (flags & VM_MAP_WIRE_WRITE) 2361 prot |= VM_PROT_WRITE; 2362 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2363 vm_map_lock(map); 2364 VM_MAP_RANGE_CHECK(map, start, end); 2365 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2366 if (flags & VM_MAP_WIRE_HOLESOK) 2367 first_entry = first_entry->next; 2368 else { 2369 vm_map_unlock(map); 2370 return (KERN_INVALID_ADDRESS); 2371 } 2372 } 2373 last_timestamp = map->timestamp; 2374 entry = first_entry; 2375 while (entry != &map->header && entry->start < end) { 2376 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2377 /* 2378 * We have not yet clipped the entry. 2379 */ 2380 saved_start = (start >= entry->start) ? start : 2381 entry->start; 2382 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2383 if (vm_map_unlock_and_wait(map, 0)) { 2384 /* 2385 * Allow interruption of user wiring? 2386 */ 2387 } 2388 vm_map_lock(map); 2389 if (last_timestamp + 1 != map->timestamp) { 2390 /* 2391 * Look again for the entry because the map was 2392 * modified while it was unlocked. 2393 * Specifically, the entry may have been 2394 * clipped, merged, or deleted. 2395 */ 2396 if (!vm_map_lookup_entry(map, saved_start, 2397 &tmp_entry)) { 2398 if (flags & VM_MAP_WIRE_HOLESOK) 2399 tmp_entry = tmp_entry->next; 2400 else { 2401 if (saved_start == start) { 2402 /* 2403 * first_entry has been deleted. 2404 */ 2405 vm_map_unlock(map); 2406 return (KERN_INVALID_ADDRESS); 2407 } 2408 end = saved_start; 2409 rv = KERN_INVALID_ADDRESS; 2410 goto done; 2411 } 2412 } 2413 if (entry == first_entry) 2414 first_entry = tmp_entry; 2415 else 2416 first_entry = NULL; 2417 entry = tmp_entry; 2418 } 2419 last_timestamp = map->timestamp; 2420 continue; 2421 } 2422 vm_map_clip_start(map, entry, start); 2423 vm_map_clip_end(map, entry, end); 2424 /* 2425 * Mark the entry in case the map lock is released. (See 2426 * above.) 2427 */ 2428 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2429 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2430 || (entry->protection & prot) != prot) { 2431 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2432 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2433 end = entry->end; 2434 rv = KERN_INVALID_ADDRESS; 2435 goto done; 2436 } 2437 goto next_entry; 2438 } 2439 if (entry->wired_count == 0) { 2440 entry->wired_count++; 2441 saved_start = entry->start; 2442 saved_end = entry->end; 2443 fictitious = entry->object.vm_object != NULL && 2444 (entry->object.vm_object->type == OBJT_DEVICE || 2445 entry->object.vm_object->type == OBJT_SG); 2446 /* 2447 * Release the map lock, relying on the in-transition 2448 * mark. Mark the map busy for fork. 2449 */ 2450 vm_map_busy(map); 2451 vm_map_unlock(map); 2452 rv = vm_fault_wire(map, saved_start, saved_end, 2453 fictitious); 2454 vm_map_lock(map); 2455 vm_map_unbusy(map); 2456 if (last_timestamp + 1 != map->timestamp) { 2457 /* 2458 * Look again for the entry because the map was 2459 * modified while it was unlocked. The entry 2460 * may have been clipped, but NOT merged or 2461 * deleted. 2462 */ 2463 result = vm_map_lookup_entry(map, saved_start, 2464 &tmp_entry); 2465 KASSERT(result, ("vm_map_wire: lookup failed")); 2466 if (entry == first_entry) 2467 first_entry = tmp_entry; 2468 else 2469 first_entry = NULL; 2470 entry = tmp_entry; 2471 while (entry->end < saved_end) { 2472 if (rv != KERN_SUCCESS) { 2473 KASSERT(entry->wired_count == 1, 2474 ("vm_map_wire: bad count")); 2475 entry->wired_count = -1; 2476 } 2477 entry = entry->next; 2478 } 2479 } 2480 last_timestamp = map->timestamp; 2481 if (rv != KERN_SUCCESS) { 2482 KASSERT(entry->wired_count == 1, 2483 ("vm_map_wire: bad count")); 2484 /* 2485 * Assign an out-of-range value to represent 2486 * the failure to wire this entry. 2487 */ 2488 entry->wired_count = -1; 2489 end = entry->end; 2490 goto done; 2491 } 2492 } else if (!user_wire || 2493 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2494 entry->wired_count++; 2495 } 2496 /* 2497 * Check the map for holes in the specified region. 2498 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2499 */ 2500 next_entry: 2501 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2502 (entry->end < end && (entry->next == &map->header || 2503 entry->next->start > entry->end))) { 2504 end = entry->end; 2505 rv = KERN_INVALID_ADDRESS; 2506 goto done; 2507 } 2508 entry = entry->next; 2509 } 2510 rv = KERN_SUCCESS; 2511 done: 2512 need_wakeup = FALSE; 2513 if (first_entry == NULL) { 2514 result = vm_map_lookup_entry(map, start, &first_entry); 2515 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2516 first_entry = first_entry->next; 2517 else 2518 KASSERT(result, ("vm_map_wire: lookup failed")); 2519 } 2520 entry = first_entry; 2521 while (entry != &map->header && entry->start < end) { 2522 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2523 goto next_entry_done; 2524 if (rv == KERN_SUCCESS) { 2525 if (user_wire) 2526 entry->eflags |= MAP_ENTRY_USER_WIRED; 2527 } else if (entry->wired_count == -1) { 2528 /* 2529 * Wiring failed on this entry. Thus, unwiring is 2530 * unnecessary. 2531 */ 2532 entry->wired_count = 0; 2533 } else { 2534 if (!user_wire || 2535 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2536 entry->wired_count--; 2537 if (entry->wired_count == 0) { 2538 /* 2539 * Retain the map lock. 2540 */ 2541 vm_fault_unwire(map, entry->start, entry->end, 2542 entry->object.vm_object != NULL && 2543 (entry->object.vm_object->type == OBJT_DEVICE || 2544 entry->object.vm_object->type == OBJT_SG)); 2545 } 2546 } 2547 next_entry_done: 2548 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2549 ("vm_map_wire: in-transition flag missing")); 2550 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED); 2551 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2552 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2553 need_wakeup = TRUE; 2554 } 2555 vm_map_simplify_entry(map, entry); 2556 entry = entry->next; 2557 } 2558 vm_map_unlock(map); 2559 if (need_wakeup) 2560 vm_map_wakeup(map); 2561 return (rv); 2562 } 2563 2564 /* 2565 * vm_map_sync 2566 * 2567 * Push any dirty cached pages in the address range to their pager. 2568 * If syncio is TRUE, dirty pages are written synchronously. 2569 * If invalidate is TRUE, any cached pages are freed as well. 2570 * 2571 * If the size of the region from start to end is zero, we are 2572 * supposed to flush all modified pages within the region containing 2573 * start. Unfortunately, a region can be split or coalesced with 2574 * neighboring regions, making it difficult to determine what the 2575 * original region was. Therefore, we approximate this requirement by 2576 * flushing the current region containing start. 2577 * 2578 * Returns an error if any part of the specified range is not mapped. 2579 */ 2580 int 2581 vm_map_sync( 2582 vm_map_t map, 2583 vm_offset_t start, 2584 vm_offset_t end, 2585 boolean_t syncio, 2586 boolean_t invalidate) 2587 { 2588 vm_map_entry_t current; 2589 vm_map_entry_t entry; 2590 vm_size_t size; 2591 vm_object_t object; 2592 vm_ooffset_t offset; 2593 unsigned int last_timestamp; 2594 2595 vm_map_lock_read(map); 2596 VM_MAP_RANGE_CHECK(map, start, end); 2597 if (!vm_map_lookup_entry(map, start, &entry)) { 2598 vm_map_unlock_read(map); 2599 return (KERN_INVALID_ADDRESS); 2600 } else if (start == end) { 2601 start = entry->start; 2602 end = entry->end; 2603 } 2604 /* 2605 * Make a first pass to check for user-wired memory and holes. 2606 */ 2607 for (current = entry; current != &map->header && current->start < end; 2608 current = current->next) { 2609 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2610 vm_map_unlock_read(map); 2611 return (KERN_INVALID_ARGUMENT); 2612 } 2613 if (end > current->end && 2614 (current->next == &map->header || 2615 current->end != current->next->start)) { 2616 vm_map_unlock_read(map); 2617 return (KERN_INVALID_ADDRESS); 2618 } 2619 } 2620 2621 if (invalidate) 2622 pmap_remove(map->pmap, start, end); 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 vm_object_sync(object, offset, size, syncio, invalidate); 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 (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 */ 2672 static void 2673 vm_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->type == OBJT_DEVICE || 2678 entry->object.vm_object->type == OBJT_SG)); 2679 entry->wired_count = 0; 2680 } 2681 2682 static void 2683 vm_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 */ 2696 static void 2697 vm_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 */ 2767 int 2768 vm_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 */ 2858 int 2859 vm_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 */ 2884 boolean_t 2885 vm_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 */ 2921 static void 2922 vm_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 */ 3046 static void 3047 vmspace_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 */ 3081 struct vmspace * 3082 vmspace_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 3236 int 3237 vm_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 init_ssize; 3243 int orient, rv; 3244 rlim_t 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 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 3262 3263 PROC_LOCK(curthread->td_proc); 3264 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 3265 PROC_UNLOCK(curthread->td_proc); 3266 3267 vm_map_lock(map); 3268 3269 /* If addr is already mapped, no go */ 3270 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3271 vm_map_unlock(map); 3272 return (KERN_NO_SPACE); 3273 } 3274 3275 /* If we would blow our VMEM resource limit, no go */ 3276 if (map->size + init_ssize > vmemlim) { 3277 vm_map_unlock(map); 3278 return (KERN_NO_SPACE); 3279 } 3280 3281 /* 3282 * If we can't accomodate max_ssize in the current mapping, no go. 3283 * However, we need to be aware that subsequent user mappings might 3284 * map into the space we have reserved for stack, and currently this 3285 * space is not protected. 3286 * 3287 * Hopefully we will at least detect this condition when we try to 3288 * grow the stack. 3289 */ 3290 if ((prev_entry->next != &map->header) && 3291 (prev_entry->next->start < addrbos + max_ssize)) { 3292 vm_map_unlock(map); 3293 return (KERN_NO_SPACE); 3294 } 3295 3296 /* 3297 * We initially map a stack of only init_ssize. We will grow as 3298 * needed later. Depending on the orientation of the stack (i.e. 3299 * the grow direction) we either map at the top of the range, the 3300 * bottom of the range or in the middle. 3301 * 3302 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3303 * and cow to be 0. Possibly we should eliminate these as input 3304 * parameters, and just pass these values here in the insert call. 3305 */ 3306 if (orient == MAP_STACK_GROWS_DOWN) 3307 bot = addrbos + max_ssize - init_ssize; 3308 else if (orient == MAP_STACK_GROWS_UP) 3309 bot = addrbos; 3310 else 3311 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3312 top = bot + init_ssize; 3313 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3314 3315 /* Now set the avail_ssize amount. */ 3316 if (rv == KERN_SUCCESS) { 3317 if (prev_entry != &map->header) 3318 vm_map_clip_end(map, prev_entry, bot); 3319 new_entry = prev_entry->next; 3320 if (new_entry->end != top || new_entry->start != bot) 3321 panic("Bad entry start/end for new stack entry"); 3322 3323 new_entry->avail_ssize = max_ssize - init_ssize; 3324 if (orient & MAP_STACK_GROWS_DOWN) 3325 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3326 if (orient & MAP_STACK_GROWS_UP) 3327 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3328 } 3329 3330 vm_map_unlock(map); 3331 return (rv); 3332 } 3333 3334 static int stack_guard_page = 0; 3335 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3336 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3337 &stack_guard_page, 0, 3338 "Insert stack guard page ahead of the growable segments."); 3339 3340 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3341 * desired address is already mapped, or if we successfully grow 3342 * the stack. Also returns KERN_SUCCESS if addr is outside the 3343 * stack range (this is strange, but preserves compatibility with 3344 * the grow function in vm_machdep.c). 3345 */ 3346 int 3347 vm_map_growstack(struct proc *p, vm_offset_t addr) 3348 { 3349 vm_map_entry_t next_entry, prev_entry; 3350 vm_map_entry_t new_entry, stack_entry; 3351 struct vmspace *vm = p->p_vmspace; 3352 vm_map_t map = &vm->vm_map; 3353 vm_offset_t end; 3354 size_t grow_amount, max_grow; 3355 rlim_t stacklim, vmemlim; 3356 int is_procstack, rv; 3357 struct ucred *cred; 3358 #ifdef notyet 3359 uint64_t limit; 3360 #endif 3361 #ifdef RACCT 3362 int error; 3363 #endif 3364 3365 Retry: 3366 PROC_LOCK(p); 3367 stacklim = lim_cur(p, RLIMIT_STACK); 3368 vmemlim = lim_cur(p, RLIMIT_VMEM); 3369 PROC_UNLOCK(p); 3370 3371 vm_map_lock_read(map); 3372 3373 /* If addr is already in the entry range, no need to grow.*/ 3374 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3375 vm_map_unlock_read(map); 3376 return (KERN_SUCCESS); 3377 } 3378 3379 next_entry = prev_entry->next; 3380 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3381 /* 3382 * This entry does not grow upwards. Since the address lies 3383 * beyond this entry, the next entry (if one exists) has to 3384 * be a downward growable entry. The entry list header is 3385 * never a growable entry, so it suffices to check the flags. 3386 */ 3387 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3388 vm_map_unlock_read(map); 3389 return (KERN_SUCCESS); 3390 } 3391 stack_entry = next_entry; 3392 } else { 3393 /* 3394 * This entry grows upward. If the next entry does not at 3395 * least grow downwards, this is the entry we need to grow. 3396 * otherwise we have two possible choices and we have to 3397 * select one. 3398 */ 3399 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3400 /* 3401 * We have two choices; grow the entry closest to 3402 * the address to minimize the amount of growth. 3403 */ 3404 if (addr - prev_entry->end <= next_entry->start - addr) 3405 stack_entry = prev_entry; 3406 else 3407 stack_entry = next_entry; 3408 } else 3409 stack_entry = prev_entry; 3410 } 3411 3412 if (stack_entry == next_entry) { 3413 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3414 KASSERT(addr < stack_entry->start, ("foo")); 3415 end = (prev_entry != &map->header) ? prev_entry->end : 3416 stack_entry->start - stack_entry->avail_ssize; 3417 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3418 max_grow = stack_entry->start - end; 3419 } else { 3420 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3421 KASSERT(addr >= stack_entry->end, ("foo")); 3422 end = (next_entry != &map->header) ? next_entry->start : 3423 stack_entry->end + stack_entry->avail_ssize; 3424 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3425 max_grow = end - stack_entry->end; 3426 } 3427 3428 if (grow_amount > stack_entry->avail_ssize) { 3429 vm_map_unlock_read(map); 3430 return (KERN_NO_SPACE); 3431 } 3432 3433 /* 3434 * If there is no longer enough space between the entries nogo, and 3435 * adjust the available space. Note: this should only happen if the 3436 * user has mapped into the stack area after the stack was created, 3437 * and is probably an error. 3438 * 3439 * This also effectively destroys any guard page the user might have 3440 * intended by limiting the stack size. 3441 */ 3442 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3443 if (vm_map_lock_upgrade(map)) 3444 goto Retry; 3445 3446 stack_entry->avail_ssize = max_grow; 3447 3448 vm_map_unlock(map); 3449 return (KERN_NO_SPACE); 3450 } 3451 3452 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3453 3454 /* 3455 * If this is the main process stack, see if we're over the stack 3456 * limit. 3457 */ 3458 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3459 vm_map_unlock_read(map); 3460 return (KERN_NO_SPACE); 3461 } 3462 #ifdef RACCT 3463 PROC_LOCK(p); 3464 if (is_procstack && 3465 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3466 PROC_UNLOCK(p); 3467 vm_map_unlock_read(map); 3468 return (KERN_NO_SPACE); 3469 } 3470 PROC_UNLOCK(p); 3471 #endif 3472 3473 /* Round up the grow amount modulo SGROWSIZ */ 3474 grow_amount = roundup (grow_amount, sgrowsiz); 3475 if (grow_amount > stack_entry->avail_ssize) 3476 grow_amount = stack_entry->avail_ssize; 3477 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3478 grow_amount = trunc_page((vm_size_t)stacklim) - 3479 ctob(vm->vm_ssize); 3480 } 3481 #ifdef notyet 3482 PROC_LOCK(p); 3483 limit = racct_get_available(p, RACCT_STACK); 3484 PROC_UNLOCK(p); 3485 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3486 grow_amount = limit - ctob(vm->vm_ssize); 3487 #endif 3488 3489 /* If we would blow our VMEM resource limit, no go */ 3490 if (map->size + grow_amount > vmemlim) { 3491 vm_map_unlock_read(map); 3492 rv = KERN_NO_SPACE; 3493 goto out; 3494 } 3495 #ifdef RACCT 3496 PROC_LOCK(p); 3497 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3498 PROC_UNLOCK(p); 3499 vm_map_unlock_read(map); 3500 rv = KERN_NO_SPACE; 3501 goto out; 3502 } 3503 PROC_UNLOCK(p); 3504 #endif 3505 3506 if (vm_map_lock_upgrade(map)) 3507 goto Retry; 3508 3509 if (stack_entry == next_entry) { 3510 /* 3511 * Growing downward. 3512 */ 3513 /* Get the preliminary new entry start value */ 3514 addr = stack_entry->start - grow_amount; 3515 3516 /* 3517 * If this puts us into the previous entry, cut back our 3518 * growth to the available space. Also, see the note above. 3519 */ 3520 if (addr < end) { 3521 stack_entry->avail_ssize = max_grow; 3522 addr = end; 3523 if (stack_guard_page) 3524 addr += PAGE_SIZE; 3525 } 3526 3527 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3528 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 3529 3530 /* Adjust the available stack space by the amount we grew. */ 3531 if (rv == KERN_SUCCESS) { 3532 if (prev_entry != &map->header) 3533 vm_map_clip_end(map, prev_entry, addr); 3534 new_entry = prev_entry->next; 3535 KASSERT(new_entry == stack_entry->prev, ("foo")); 3536 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3537 KASSERT(new_entry->start == addr, ("foo")); 3538 grow_amount = new_entry->end - new_entry->start; 3539 new_entry->avail_ssize = stack_entry->avail_ssize - 3540 grow_amount; 3541 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3542 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3543 } 3544 } else { 3545 /* 3546 * Growing upward. 3547 */ 3548 addr = stack_entry->end + grow_amount; 3549 3550 /* 3551 * If this puts us into the next entry, cut back our growth 3552 * to the available space. Also, see the note above. 3553 */ 3554 if (addr > end) { 3555 stack_entry->avail_ssize = end - stack_entry->end; 3556 addr = end; 3557 if (stack_guard_page) 3558 addr -= PAGE_SIZE; 3559 } 3560 3561 grow_amount = addr - stack_entry->end; 3562 cred = stack_entry->cred; 3563 if (cred == NULL && stack_entry->object.vm_object != NULL) 3564 cred = stack_entry->object.vm_object->cred; 3565 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3566 rv = KERN_NO_SPACE; 3567 /* Grow the underlying object if applicable. */ 3568 else if (stack_entry->object.vm_object == NULL || 3569 vm_object_coalesce(stack_entry->object.vm_object, 3570 stack_entry->offset, 3571 (vm_size_t)(stack_entry->end - stack_entry->start), 3572 (vm_size_t)grow_amount, cred != NULL)) { 3573 map->size += (addr - stack_entry->end); 3574 /* Update the current entry. */ 3575 stack_entry->end = addr; 3576 stack_entry->avail_ssize -= grow_amount; 3577 vm_map_entry_resize_free(map, stack_entry); 3578 rv = KERN_SUCCESS; 3579 3580 if (next_entry != &map->header) 3581 vm_map_clip_start(map, next_entry, addr); 3582 } else 3583 rv = KERN_FAILURE; 3584 } 3585 3586 if (rv == KERN_SUCCESS && is_procstack) 3587 vm->vm_ssize += btoc(grow_amount); 3588 3589 vm_map_unlock(map); 3590 3591 /* 3592 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3593 */ 3594 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3595 vm_map_wire(map, 3596 (stack_entry == next_entry) ? addr : addr - grow_amount, 3597 (stack_entry == next_entry) ? stack_entry->start : addr, 3598 (p->p_flag & P_SYSTEM) 3599 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3600 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3601 } 3602 3603 out: 3604 #ifdef RACCT 3605 if (rv != KERN_SUCCESS) { 3606 PROC_LOCK(p); 3607 error = racct_set(p, RACCT_VMEM, map->size); 3608 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3609 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3610 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3611 PROC_UNLOCK(p); 3612 } 3613 #endif 3614 3615 return (rv); 3616 } 3617 3618 /* 3619 * Unshare the specified VM space for exec. If other processes are 3620 * mapped to it, then create a new one. The new vmspace is null. 3621 */ 3622 int 3623 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3624 { 3625 struct vmspace *oldvmspace = p->p_vmspace; 3626 struct vmspace *newvmspace; 3627 3628 newvmspace = vmspace_alloc(minuser, maxuser); 3629 if (newvmspace == NULL) 3630 return (ENOMEM); 3631 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3632 /* 3633 * This code is written like this for prototype purposes. The 3634 * goal is to avoid running down the vmspace here, but let the 3635 * other process's that are still using the vmspace to finally 3636 * run it down. Even though there is little or no chance of blocking 3637 * here, it is a good idea to keep this form for future mods. 3638 */ 3639 PROC_VMSPACE_LOCK(p); 3640 p->p_vmspace = newvmspace; 3641 PROC_VMSPACE_UNLOCK(p); 3642 if (p == curthread->td_proc) 3643 pmap_activate(curthread); 3644 vmspace_free(oldvmspace); 3645 return (0); 3646 } 3647 3648 /* 3649 * Unshare the specified VM space for forcing COW. This 3650 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3651 */ 3652 int 3653 vmspace_unshare(struct proc *p) 3654 { 3655 struct vmspace *oldvmspace = p->p_vmspace; 3656 struct vmspace *newvmspace; 3657 vm_ooffset_t fork_charge; 3658 3659 if (oldvmspace->vm_refcnt == 1) 3660 return (0); 3661 fork_charge = 0; 3662 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3663 if (newvmspace == NULL) 3664 return (ENOMEM); 3665 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3666 vmspace_free(newvmspace); 3667 return (ENOMEM); 3668 } 3669 PROC_VMSPACE_LOCK(p); 3670 p->p_vmspace = newvmspace; 3671 PROC_VMSPACE_UNLOCK(p); 3672 if (p == curthread->td_proc) 3673 pmap_activate(curthread); 3674 vmspace_free(oldvmspace); 3675 return (0); 3676 } 3677 3678 /* 3679 * vm_map_lookup: 3680 * 3681 * Finds the VM object, offset, and 3682 * protection for a given virtual address in the 3683 * specified map, assuming a page fault of the 3684 * type specified. 3685 * 3686 * Leaves the map in question locked for read; return 3687 * values are guaranteed until a vm_map_lookup_done 3688 * call is performed. Note that the map argument 3689 * is in/out; the returned map must be used in 3690 * the call to vm_map_lookup_done. 3691 * 3692 * A handle (out_entry) is returned for use in 3693 * vm_map_lookup_done, to make that fast. 3694 * 3695 * If a lookup is requested with "write protection" 3696 * specified, the map may be changed to perform virtual 3697 * copying operations, although the data referenced will 3698 * remain the same. 3699 */ 3700 int 3701 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3702 vm_offset_t vaddr, 3703 vm_prot_t fault_typea, 3704 vm_map_entry_t *out_entry, /* OUT */ 3705 vm_object_t *object, /* OUT */ 3706 vm_pindex_t *pindex, /* OUT */ 3707 vm_prot_t *out_prot, /* OUT */ 3708 boolean_t *wired) /* OUT */ 3709 { 3710 vm_map_entry_t entry; 3711 vm_map_t map = *var_map; 3712 vm_prot_t prot; 3713 vm_prot_t fault_type = fault_typea; 3714 vm_object_t eobject; 3715 vm_size_t size; 3716 struct ucred *cred; 3717 3718 RetryLookup:; 3719 3720 vm_map_lock_read(map); 3721 3722 /* 3723 * Lookup the faulting address. 3724 */ 3725 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3726 vm_map_unlock_read(map); 3727 return (KERN_INVALID_ADDRESS); 3728 } 3729 3730 entry = *out_entry; 3731 3732 /* 3733 * Handle submaps. 3734 */ 3735 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3736 vm_map_t old_map = map; 3737 3738 *var_map = map = entry->object.sub_map; 3739 vm_map_unlock_read(old_map); 3740 goto RetryLookup; 3741 } 3742 3743 /* 3744 * Check whether this task is allowed to have this page. 3745 */ 3746 prot = entry->protection; 3747 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3748 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3749 vm_map_unlock_read(map); 3750 return (KERN_PROTECTION_FAILURE); 3751 } 3752 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3753 (entry->eflags & MAP_ENTRY_COW) && 3754 (fault_type & VM_PROT_WRITE)) { 3755 vm_map_unlock_read(map); 3756 return (KERN_PROTECTION_FAILURE); 3757 } 3758 3759 /* 3760 * If this page is not pageable, we have to get it for all possible 3761 * accesses. 3762 */ 3763 *wired = (entry->wired_count != 0); 3764 if (*wired) 3765 fault_type = entry->protection; 3766 size = entry->end - entry->start; 3767 /* 3768 * If the entry was copy-on-write, we either ... 3769 */ 3770 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3771 /* 3772 * If we want to write the page, we may as well handle that 3773 * now since we've got the map locked. 3774 * 3775 * If we don't need to write the page, we just demote the 3776 * permissions allowed. 3777 */ 3778 if ((fault_type & VM_PROT_WRITE) != 0 || 3779 (fault_typea & VM_PROT_COPY) != 0) { 3780 /* 3781 * Make a new object, and place it in the object 3782 * chain. Note that no new references have appeared 3783 * -- one just moved from the map to the new 3784 * object. 3785 */ 3786 if (vm_map_lock_upgrade(map)) 3787 goto RetryLookup; 3788 3789 if (entry->cred == NULL) { 3790 /* 3791 * The debugger owner is charged for 3792 * the memory. 3793 */ 3794 cred = curthread->td_ucred; 3795 crhold(cred); 3796 if (!swap_reserve_by_cred(size, cred)) { 3797 crfree(cred); 3798 vm_map_unlock(map); 3799 return (KERN_RESOURCE_SHORTAGE); 3800 } 3801 entry->cred = cred; 3802 } 3803 vm_object_shadow(&entry->object.vm_object, 3804 &entry->offset, size); 3805 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3806 eobject = entry->object.vm_object; 3807 if (eobject->cred != NULL) { 3808 /* 3809 * The object was not shadowed. 3810 */ 3811 swap_release_by_cred(size, entry->cred); 3812 crfree(entry->cred); 3813 entry->cred = NULL; 3814 } else if (entry->cred != NULL) { 3815 VM_OBJECT_LOCK(eobject); 3816 eobject->cred = entry->cred; 3817 eobject->charge = size; 3818 VM_OBJECT_UNLOCK(eobject); 3819 entry->cred = NULL; 3820 } 3821 3822 vm_map_lock_downgrade(map); 3823 } else { 3824 /* 3825 * We're attempting to read a copy-on-write page -- 3826 * don't allow writes. 3827 */ 3828 prot &= ~VM_PROT_WRITE; 3829 } 3830 } 3831 3832 /* 3833 * Create an object if necessary. 3834 */ 3835 if (entry->object.vm_object == NULL && 3836 !map->system_map) { 3837 if (vm_map_lock_upgrade(map)) 3838 goto RetryLookup; 3839 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3840 atop(size)); 3841 entry->offset = 0; 3842 if (entry->cred != NULL) { 3843 VM_OBJECT_LOCK(entry->object.vm_object); 3844 entry->object.vm_object->cred = entry->cred; 3845 entry->object.vm_object->charge = size; 3846 VM_OBJECT_UNLOCK(entry->object.vm_object); 3847 entry->cred = NULL; 3848 } 3849 vm_map_lock_downgrade(map); 3850 } 3851 3852 /* 3853 * Return the object/offset from this entry. If the entry was 3854 * copy-on-write or empty, it has been fixed up. 3855 */ 3856 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3857 *object = entry->object.vm_object; 3858 3859 *out_prot = prot; 3860 return (KERN_SUCCESS); 3861 } 3862 3863 /* 3864 * vm_map_lookup_locked: 3865 * 3866 * Lookup the faulting address. A version of vm_map_lookup that returns 3867 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3868 */ 3869 int 3870 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3871 vm_offset_t vaddr, 3872 vm_prot_t fault_typea, 3873 vm_map_entry_t *out_entry, /* OUT */ 3874 vm_object_t *object, /* OUT */ 3875 vm_pindex_t *pindex, /* OUT */ 3876 vm_prot_t *out_prot, /* OUT */ 3877 boolean_t *wired) /* OUT */ 3878 { 3879 vm_map_entry_t entry; 3880 vm_map_t map = *var_map; 3881 vm_prot_t prot; 3882 vm_prot_t fault_type = fault_typea; 3883 3884 /* 3885 * Lookup the faulting address. 3886 */ 3887 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3888 return (KERN_INVALID_ADDRESS); 3889 3890 entry = *out_entry; 3891 3892 /* 3893 * Fail if the entry refers to a submap. 3894 */ 3895 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3896 return (KERN_FAILURE); 3897 3898 /* 3899 * Check whether this task is allowed to have this page. 3900 */ 3901 prot = entry->protection; 3902 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3903 if ((fault_type & prot) != fault_type) 3904 return (KERN_PROTECTION_FAILURE); 3905 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3906 (entry->eflags & MAP_ENTRY_COW) && 3907 (fault_type & VM_PROT_WRITE)) 3908 return (KERN_PROTECTION_FAILURE); 3909 3910 /* 3911 * If this page is not pageable, we have to get it for all possible 3912 * accesses. 3913 */ 3914 *wired = (entry->wired_count != 0); 3915 if (*wired) 3916 fault_type = entry->protection; 3917 3918 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3919 /* 3920 * Fail if the entry was copy-on-write for a write fault. 3921 */ 3922 if (fault_type & VM_PROT_WRITE) 3923 return (KERN_FAILURE); 3924 /* 3925 * We're attempting to read a copy-on-write page -- 3926 * don't allow writes. 3927 */ 3928 prot &= ~VM_PROT_WRITE; 3929 } 3930 3931 /* 3932 * Fail if an object should be created. 3933 */ 3934 if (entry->object.vm_object == NULL && !map->system_map) 3935 return (KERN_FAILURE); 3936 3937 /* 3938 * Return the object/offset from this entry. If the entry was 3939 * copy-on-write or empty, it has been fixed up. 3940 */ 3941 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3942 *object = entry->object.vm_object; 3943 3944 *out_prot = prot; 3945 return (KERN_SUCCESS); 3946 } 3947 3948 /* 3949 * vm_map_lookup_done: 3950 * 3951 * Releases locks acquired by a vm_map_lookup 3952 * (according to the handle returned by that lookup). 3953 */ 3954 void 3955 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3956 { 3957 /* 3958 * Unlock the main-level map 3959 */ 3960 vm_map_unlock_read(map); 3961 } 3962 3963 #include "opt_ddb.h" 3964 #ifdef DDB 3965 #include <sys/kernel.h> 3966 3967 #include <ddb/ddb.h> 3968 3969 /* 3970 * vm_map_print: [ debug ] 3971 */ 3972 DB_SHOW_COMMAND(map, vm_map_print) 3973 { 3974 static int nlines; 3975 /* XXX convert args. */ 3976 vm_map_t map = (vm_map_t)addr; 3977 boolean_t full = have_addr; 3978 3979 vm_map_entry_t entry; 3980 3981 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3982 (void *)map, 3983 (void *)map->pmap, map->nentries, map->timestamp); 3984 nlines++; 3985 3986 if (!full && db_indent) 3987 return; 3988 3989 db_indent += 2; 3990 for (entry = map->header.next; entry != &map->header; 3991 entry = entry->next) { 3992 db_iprintf("map entry %p: start=%p, end=%p\n", 3993 (void *)entry, (void *)entry->start, (void *)entry->end); 3994 nlines++; 3995 { 3996 static char *inheritance_name[4] = 3997 {"share", "copy", "none", "donate_copy"}; 3998 3999 db_iprintf(" prot=%x/%x/%s", 4000 entry->protection, 4001 entry->max_protection, 4002 inheritance_name[(int)(unsigned char)entry->inheritance]); 4003 if (entry->wired_count != 0) 4004 db_printf(", wired"); 4005 } 4006 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4007 db_printf(", share=%p, offset=0x%jx\n", 4008 (void *)entry->object.sub_map, 4009 (uintmax_t)entry->offset); 4010 nlines++; 4011 if ((entry->prev == &map->header) || 4012 (entry->prev->object.sub_map != 4013 entry->object.sub_map)) { 4014 db_indent += 2; 4015 vm_map_print((db_expr_t)(intptr_t) 4016 entry->object.sub_map, 4017 full, 0, (char *)0); 4018 db_indent -= 2; 4019 } 4020 } else { 4021 if (entry->cred != NULL) 4022 db_printf(", ruid %d", entry->cred->cr_ruid); 4023 db_printf(", object=%p, offset=0x%jx", 4024 (void *)entry->object.vm_object, 4025 (uintmax_t)entry->offset); 4026 if (entry->object.vm_object && entry->object.vm_object->cred) 4027 db_printf(", obj ruid %d charge %jx", 4028 entry->object.vm_object->cred->cr_ruid, 4029 (uintmax_t)entry->object.vm_object->charge); 4030 if (entry->eflags & MAP_ENTRY_COW) 4031 db_printf(", copy (%s)", 4032 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4033 db_printf("\n"); 4034 nlines++; 4035 4036 if ((entry->prev == &map->header) || 4037 (entry->prev->object.vm_object != 4038 entry->object.vm_object)) { 4039 db_indent += 2; 4040 vm_object_print((db_expr_t)(intptr_t) 4041 entry->object.vm_object, 4042 full, 0, (char *)0); 4043 nlines += 4; 4044 db_indent -= 2; 4045 } 4046 } 4047 } 4048 db_indent -= 2; 4049 if (db_indent == 0) 4050 nlines = 0; 4051 } 4052 4053 4054 DB_SHOW_COMMAND(procvm, procvm) 4055 { 4056 struct proc *p; 4057 4058 if (have_addr) { 4059 p = (struct proc *) addr; 4060 } else { 4061 p = curproc; 4062 } 4063 4064 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4065 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4066 (void *)vmspace_pmap(p->p_vmspace)); 4067 4068 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 4069 } 4070 4071 #endif /* DDB */ 4072