vm_map.c revision 259297
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: stable/10/sys/vm/vm_map.c 259297 2013-12-13 06:25:08Z kib $"); 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/rwlock.h> 81#include <sys/file.h> 82#include <sys/sysctl.h> 83#include <sys/sysent.h> 84#include <sys/shm.h> 85 86#include <vm/vm.h> 87#include <vm/vm_param.h> 88#include <vm/pmap.h> 89#include <vm/vm_map.h> 90#include <vm/vm_page.h> 91#include <vm/vm_object.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_kern.h> 94#include <vm/vm_extern.h> 95#include <vm/vnode_pager.h> 96#include <vm/swap_pager.h> 97#include <vm/uma.h> 98 99/* 100 * Virtual memory maps provide for the mapping, protection, 101 * and sharing of virtual memory objects. In addition, 102 * this module provides for an efficient virtual copy of 103 * memory from one map to another. 104 * 105 * Synchronization is required prior to most operations. 106 * 107 * Maps consist of an ordered doubly-linked list of simple 108 * entries; a self-adjusting binary search tree of these 109 * entries is used to speed up lookups. 110 * 111 * Since portions of maps are specified by start/end addresses, 112 * which may not align with existing map entries, all 113 * routines merely "clip" entries to these start/end values. 114 * [That is, an entry is split into two, bordering at a 115 * start or end value.] Note that these clippings may not 116 * always be necessary (as the two resulting entries are then 117 * not changed); however, the clipping is done for convenience. 118 * 119 * As mentioned above, virtual copy operations are performed 120 * by copying VM object references from one map to 121 * another, and then marking both regions as copy-on-write. 122 */ 123 124static struct mtx map_sleep_mtx; 125static uma_zone_t mapentzone; 126static uma_zone_t kmapentzone; 127static uma_zone_t mapzone; 128static uma_zone_t vmspace_zone; 129static int vmspace_zinit(void *mem, int size, int flags); 130static int vm_map_zinit(void *mem, int ize, int flags); 131static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 132 vm_offset_t max); 133static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 134static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 135#ifdef INVARIANTS 136static void vm_map_zdtor(void *mem, int size, void *arg); 137static void vmspace_zdtor(void *mem, int size, void *arg); 138#endif 139 140#define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 141 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 142 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 143 144/* 145 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 146 * stable. 147 */ 148#define PROC_VMSPACE_LOCK(p) do { } while (0) 149#define PROC_VMSPACE_UNLOCK(p) do { } while (0) 150 151/* 152 * VM_MAP_RANGE_CHECK: [ internal use only ] 153 * 154 * Asserts that the starting and ending region 155 * addresses fall within the valid range of the map. 156 */ 157#define VM_MAP_RANGE_CHECK(map, start, end) \ 158 { \ 159 if (start < vm_map_min(map)) \ 160 start = vm_map_min(map); \ 161 if (end > vm_map_max(map)) \ 162 end = vm_map_max(map); \ 163 if (start > end) \ 164 start = end; \ 165 } 166 167/* 168 * vm_map_startup: 169 * 170 * Initialize the vm_map module. Must be called before 171 * any other vm_map routines. 172 * 173 * Map and entry structures are allocated from the general 174 * purpose memory pool with some exceptions: 175 * 176 * - The kernel map and kmem submap are allocated statically. 177 * - Kernel map entries are allocated out of a static pool. 178 * 179 * These restrictions are necessary since malloc() uses the 180 * maps and requires map entries. 181 */ 182 183void 184vm_map_startup(void) 185{ 186 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 187 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 188#ifdef INVARIANTS 189 vm_map_zdtor, 190#else 191 NULL, 192#endif 193 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 194 uma_prealloc(mapzone, MAX_KMAP); 195 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 196 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 197 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 198 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 200 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 201#ifdef INVARIANTS 202 vmspace_zdtor, 203#else 204 NULL, 205#endif 206 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 207} 208 209static int 210vmspace_zinit(void *mem, int size, int flags) 211{ 212 struct vmspace *vm; 213 214 vm = (struct vmspace *)mem; 215 216 vm->vm_map.pmap = NULL; 217 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 218 PMAP_LOCK_INIT(vmspace_pmap(vm)); 219 return (0); 220} 221 222static int 223vm_map_zinit(void *mem, int size, int flags) 224{ 225 vm_map_t map; 226 227 map = (vm_map_t)mem; 228 memset(map, 0, sizeof(*map)); 229 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 230 sx_init(&map->lock, "vm map (user)"); 231 return (0); 232} 233 234#ifdef INVARIANTS 235static void 236vmspace_zdtor(void *mem, int size, void *arg) 237{ 238 struct vmspace *vm; 239 240 vm = (struct vmspace *)mem; 241 242 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 243} 244static void 245vm_map_zdtor(void *mem, int size, void *arg) 246{ 247 vm_map_t map; 248 249 map = (vm_map_t)mem; 250 KASSERT(map->nentries == 0, 251 ("map %p nentries == %d on free.", 252 map, map->nentries)); 253 KASSERT(map->size == 0, 254 ("map %p size == %lu on free.", 255 map, (unsigned long)map->size)); 256} 257#endif /* INVARIANTS */ 258 259/* 260 * Allocate a vmspace structure, including a vm_map and pmap, 261 * and initialize those structures. The refcnt is set to 1. 262 * 263 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 264 */ 265struct vmspace * 266vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 267{ 268 struct vmspace *vm; 269 270 vm = uma_zalloc(vmspace_zone, M_WAITOK); 271 272 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 273 274 if (pinit == NULL) 275 pinit = &pmap_pinit; 276 277 if (!pinit(vmspace_pmap(vm))) { 278 uma_zfree(vmspace_zone, vm); 279 return (NULL); 280 } 281 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 282 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 283 vm->vm_refcnt = 1; 284 vm->vm_shm = NULL; 285 vm->vm_swrss = 0; 286 vm->vm_tsize = 0; 287 vm->vm_dsize = 0; 288 vm->vm_ssize = 0; 289 vm->vm_taddr = 0; 290 vm->vm_daddr = 0; 291 vm->vm_maxsaddr = 0; 292 return (vm); 293} 294 295static void 296vmspace_container_reset(struct proc *p) 297{ 298 299#ifdef RACCT 300 PROC_LOCK(p); 301 racct_set(p, RACCT_DATA, 0); 302 racct_set(p, RACCT_STACK, 0); 303 racct_set(p, RACCT_RSS, 0); 304 racct_set(p, RACCT_MEMLOCK, 0); 305 racct_set(p, RACCT_VMEM, 0); 306 PROC_UNLOCK(p); 307#endif 308} 309 310static inline void 311vmspace_dofree(struct vmspace *vm) 312{ 313 314 CTR1(KTR_VM, "vmspace_free: %p", vm); 315 316 /* 317 * Make sure any SysV shm is freed, it might not have been in 318 * exit1(). 319 */ 320 shmexit(vm); 321 322 /* 323 * Lock the map, to wait out all other references to it. 324 * Delete all of the mappings and pages they hold, then call 325 * the pmap module to reclaim anything left. 326 */ 327 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 328 vm->vm_map.max_offset); 329 330 pmap_release(vmspace_pmap(vm)); 331 vm->vm_map.pmap = NULL; 332 uma_zfree(vmspace_zone, vm); 333} 334 335void 336vmspace_free(struct vmspace *vm) 337{ 338 339 if (vm->vm_refcnt == 0) 340 panic("vmspace_free: attempt to free already freed vmspace"); 341 342 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 343 vmspace_dofree(vm); 344} 345 346void 347vmspace_exitfree(struct proc *p) 348{ 349 struct vmspace *vm; 350 351 PROC_VMSPACE_LOCK(p); 352 vm = p->p_vmspace; 353 p->p_vmspace = NULL; 354 PROC_VMSPACE_UNLOCK(p); 355 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 356 vmspace_free(vm); 357} 358 359void 360vmspace_exit(struct thread *td) 361{ 362 int refcnt; 363 struct vmspace *vm; 364 struct proc *p; 365 366 /* 367 * Release user portion of address space. 368 * This releases references to vnodes, 369 * which could cause I/O if the file has been unlinked. 370 * Need to do this early enough that we can still sleep. 371 * 372 * The last exiting process to reach this point releases as 373 * much of the environment as it can. vmspace_dofree() is the 374 * slower fallback in case another process had a temporary 375 * reference to the vmspace. 376 */ 377 378 p = td->td_proc; 379 vm = p->p_vmspace; 380 atomic_add_int(&vmspace0.vm_refcnt, 1); 381 do { 382 refcnt = vm->vm_refcnt; 383 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 384 /* Switch now since other proc might free vmspace */ 385 PROC_VMSPACE_LOCK(p); 386 p->p_vmspace = &vmspace0; 387 PROC_VMSPACE_UNLOCK(p); 388 pmap_activate(td); 389 } 390 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 391 if (refcnt == 1) { 392 if (p->p_vmspace != vm) { 393 /* vmspace not yet freed, switch back */ 394 PROC_VMSPACE_LOCK(p); 395 p->p_vmspace = vm; 396 PROC_VMSPACE_UNLOCK(p); 397 pmap_activate(td); 398 } 399 pmap_remove_pages(vmspace_pmap(vm)); 400 /* Switch now since this proc will free vmspace */ 401 PROC_VMSPACE_LOCK(p); 402 p->p_vmspace = &vmspace0; 403 PROC_VMSPACE_UNLOCK(p); 404 pmap_activate(td); 405 vmspace_dofree(vm); 406 } 407 vmspace_container_reset(p); 408} 409 410/* Acquire reference to vmspace owned by another process. */ 411 412struct vmspace * 413vmspace_acquire_ref(struct proc *p) 414{ 415 struct vmspace *vm; 416 int refcnt; 417 418 PROC_VMSPACE_LOCK(p); 419 vm = p->p_vmspace; 420 if (vm == NULL) { 421 PROC_VMSPACE_UNLOCK(p); 422 return (NULL); 423 } 424 do { 425 refcnt = vm->vm_refcnt; 426 if (refcnt <= 0) { /* Avoid 0->1 transition */ 427 PROC_VMSPACE_UNLOCK(p); 428 return (NULL); 429 } 430 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 431 if (vm != p->p_vmspace) { 432 PROC_VMSPACE_UNLOCK(p); 433 vmspace_free(vm); 434 return (NULL); 435 } 436 PROC_VMSPACE_UNLOCK(p); 437 return (vm); 438} 439 440void 441_vm_map_lock(vm_map_t map, const char *file, int line) 442{ 443 444 if (map->system_map) 445 mtx_lock_flags_(&map->system_mtx, 0, file, line); 446 else 447 sx_xlock_(&map->lock, file, line); 448 map->timestamp++; 449} 450 451static void 452vm_map_process_deferred(void) 453{ 454 struct thread *td; 455 vm_map_entry_t entry, next; 456 vm_object_t object; 457 458 td = curthread; 459 entry = td->td_map_def_user; 460 td->td_map_def_user = NULL; 461 while (entry != NULL) { 462 next = entry->next; 463 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 464 /* 465 * Decrement the object's writemappings and 466 * possibly the vnode's v_writecount. 467 */ 468 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 469 ("Submap with writecount")); 470 object = entry->object.vm_object; 471 KASSERT(object != NULL, ("No object for writecount")); 472 vnode_pager_release_writecount(object, entry->start, 473 entry->end); 474 } 475 vm_map_entry_deallocate(entry, FALSE); 476 entry = next; 477 } 478} 479 480void 481_vm_map_unlock(vm_map_t map, const char *file, int line) 482{ 483 484 if (map->system_map) 485 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 486 else { 487 sx_xunlock_(&map->lock, file, line); 488 vm_map_process_deferred(); 489 } 490} 491 492void 493_vm_map_lock_read(vm_map_t map, const char *file, int line) 494{ 495 496 if (map->system_map) 497 mtx_lock_flags_(&map->system_mtx, 0, file, line); 498 else 499 sx_slock_(&map->lock, file, line); 500} 501 502void 503_vm_map_unlock_read(vm_map_t map, const char *file, int line) 504{ 505 506 if (map->system_map) 507 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 508 else { 509 sx_sunlock_(&map->lock, file, line); 510 vm_map_process_deferred(); 511 } 512} 513 514int 515_vm_map_trylock(vm_map_t map, const char *file, int line) 516{ 517 int error; 518 519 error = map->system_map ? 520 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 521 !sx_try_xlock_(&map->lock, file, line); 522 if (error == 0) 523 map->timestamp++; 524 return (error == 0); 525} 526 527int 528_vm_map_trylock_read(vm_map_t map, const char *file, int line) 529{ 530 int error; 531 532 error = map->system_map ? 533 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 534 !sx_try_slock_(&map->lock, file, line); 535 return (error == 0); 536} 537 538/* 539 * _vm_map_lock_upgrade: [ internal use only ] 540 * 541 * Tries to upgrade a read (shared) lock on the specified map to a write 542 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 543 * non-zero value if the upgrade fails. If the upgrade fails, the map is 544 * returned without a read or write lock held. 545 * 546 * Requires that the map be read locked. 547 */ 548int 549_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 550{ 551 unsigned int last_timestamp; 552 553 if (map->system_map) { 554 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 555 } else { 556 if (!sx_try_upgrade_(&map->lock, file, line)) { 557 last_timestamp = map->timestamp; 558 sx_sunlock_(&map->lock, file, line); 559 vm_map_process_deferred(); 560 /* 561 * If the map's timestamp does not change while the 562 * map is unlocked, then the upgrade succeeds. 563 */ 564 sx_xlock_(&map->lock, file, line); 565 if (last_timestamp != map->timestamp) { 566 sx_xunlock_(&map->lock, file, line); 567 return (1); 568 } 569 } 570 } 571 map->timestamp++; 572 return (0); 573} 574 575void 576_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 577{ 578 579 if (map->system_map) { 580 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 581 } else 582 sx_downgrade_(&map->lock, file, line); 583} 584 585/* 586 * vm_map_locked: 587 * 588 * Returns a non-zero value if the caller holds a write (exclusive) lock 589 * on the specified map and the value "0" otherwise. 590 */ 591int 592vm_map_locked(vm_map_t map) 593{ 594 595 if (map->system_map) 596 return (mtx_owned(&map->system_mtx)); 597 else 598 return (sx_xlocked(&map->lock)); 599} 600 601#ifdef INVARIANTS 602static void 603_vm_map_assert_locked(vm_map_t map, const char *file, int line) 604{ 605 606 if (map->system_map) 607 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 608 else 609 sx_assert_(&map->lock, SA_XLOCKED, file, line); 610} 611 612#define VM_MAP_ASSERT_LOCKED(map) \ 613 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 614#else 615#define VM_MAP_ASSERT_LOCKED(map) 616#endif 617 618/* 619 * _vm_map_unlock_and_wait: 620 * 621 * Atomically releases the lock on the specified map and puts the calling 622 * thread to sleep. The calling thread will remain asleep until either 623 * vm_map_wakeup() is performed on the map or the specified timeout is 624 * exceeded. 625 * 626 * WARNING! This function does not perform deferred deallocations of 627 * objects and map entries. Therefore, the calling thread is expected to 628 * reacquire the map lock after reawakening and later perform an ordinary 629 * unlock operation, such as vm_map_unlock(), before completing its 630 * operation on the map. 631 */ 632int 633_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 634{ 635 636 mtx_lock(&map_sleep_mtx); 637 if (map->system_map) 638 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 639 else 640 sx_xunlock_(&map->lock, file, line); 641 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 642 timo)); 643} 644 645/* 646 * vm_map_wakeup: 647 * 648 * Awaken any threads that have slept on the map using 649 * vm_map_unlock_and_wait(). 650 */ 651void 652vm_map_wakeup(vm_map_t map) 653{ 654 655 /* 656 * Acquire and release map_sleep_mtx to prevent a wakeup() 657 * from being performed (and lost) between the map unlock 658 * and the msleep() in _vm_map_unlock_and_wait(). 659 */ 660 mtx_lock(&map_sleep_mtx); 661 mtx_unlock(&map_sleep_mtx); 662 wakeup(&map->root); 663} 664 665void 666vm_map_busy(vm_map_t map) 667{ 668 669 VM_MAP_ASSERT_LOCKED(map); 670 map->busy++; 671} 672 673void 674vm_map_unbusy(vm_map_t map) 675{ 676 677 VM_MAP_ASSERT_LOCKED(map); 678 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 679 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 680 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 681 wakeup(&map->busy); 682 } 683} 684 685void 686vm_map_wait_busy(vm_map_t map) 687{ 688 689 VM_MAP_ASSERT_LOCKED(map); 690 while (map->busy) { 691 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 692 if (map->system_map) 693 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 694 else 695 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 696 } 697 map->timestamp++; 698} 699 700long 701vmspace_resident_count(struct vmspace *vmspace) 702{ 703 return pmap_resident_count(vmspace_pmap(vmspace)); 704} 705 706/* 707 * vm_map_create: 708 * 709 * Creates and returns a new empty VM map with 710 * the given physical map structure, and having 711 * the given lower and upper address bounds. 712 */ 713vm_map_t 714vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 715{ 716 vm_map_t result; 717 718 result = uma_zalloc(mapzone, M_WAITOK); 719 CTR1(KTR_VM, "vm_map_create: %p", result); 720 _vm_map_init(result, pmap, min, max); 721 return (result); 722} 723 724/* 725 * Initialize an existing vm_map structure 726 * such as that in the vmspace structure. 727 */ 728static void 729_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 730{ 731 732 map->header.next = map->header.prev = &map->header; 733 map->needs_wakeup = FALSE; 734 map->system_map = 0; 735 map->pmap = pmap; 736 map->min_offset = min; 737 map->max_offset = max; 738 map->flags = 0; 739 map->root = NULL; 740 map->timestamp = 0; 741 map->busy = 0; 742} 743 744void 745vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 746{ 747 748 _vm_map_init(map, pmap, min, max); 749 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 750 sx_init(&map->lock, "user map"); 751} 752 753/* 754 * vm_map_entry_dispose: [ internal use only ] 755 * 756 * Inverse of vm_map_entry_create. 757 */ 758static void 759vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 760{ 761 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 762} 763 764/* 765 * vm_map_entry_create: [ internal use only ] 766 * 767 * Allocates a VM map entry for insertion. 768 * No entry fields are filled in. 769 */ 770static vm_map_entry_t 771vm_map_entry_create(vm_map_t map) 772{ 773 vm_map_entry_t new_entry; 774 775 if (map->system_map) 776 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 777 else 778 new_entry = uma_zalloc(mapentzone, M_WAITOK); 779 if (new_entry == NULL) 780 panic("vm_map_entry_create: kernel resources exhausted"); 781 return (new_entry); 782} 783 784/* 785 * vm_map_entry_set_behavior: 786 * 787 * Set the expected access behavior, either normal, random, or 788 * sequential. 789 */ 790static inline void 791vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 792{ 793 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 794 (behavior & MAP_ENTRY_BEHAV_MASK); 795} 796 797/* 798 * vm_map_entry_set_max_free: 799 * 800 * Set the max_free field in a vm_map_entry. 801 */ 802static inline void 803vm_map_entry_set_max_free(vm_map_entry_t entry) 804{ 805 806 entry->max_free = entry->adj_free; 807 if (entry->left != NULL && entry->left->max_free > entry->max_free) 808 entry->max_free = entry->left->max_free; 809 if (entry->right != NULL && entry->right->max_free > entry->max_free) 810 entry->max_free = entry->right->max_free; 811} 812 813/* 814 * vm_map_entry_splay: 815 * 816 * The Sleator and Tarjan top-down splay algorithm with the 817 * following variation. Max_free must be computed bottom-up, so 818 * on the downward pass, maintain the left and right spines in 819 * reverse order. Then, make a second pass up each side to fix 820 * the pointers and compute max_free. The time bound is O(log n) 821 * amortized. 822 * 823 * The new root is the vm_map_entry containing "addr", or else an 824 * adjacent entry (lower or higher) if addr is not in the tree. 825 * 826 * The map must be locked, and leaves it so. 827 * 828 * Returns: the new root. 829 */ 830static vm_map_entry_t 831vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 832{ 833 vm_map_entry_t llist, rlist; 834 vm_map_entry_t ltree, rtree; 835 vm_map_entry_t y; 836 837 /* Special case of empty tree. */ 838 if (root == NULL) 839 return (root); 840 841 /* 842 * Pass One: Splay down the tree until we find addr or a NULL 843 * pointer where addr would go. llist and rlist are the two 844 * sides in reverse order (bottom-up), with llist linked by 845 * the right pointer and rlist linked by the left pointer in 846 * the vm_map_entry. Wait until Pass Two to set max_free on 847 * the two spines. 848 */ 849 llist = NULL; 850 rlist = NULL; 851 for (;;) { 852 /* root is never NULL in here. */ 853 if (addr < root->start) { 854 y = root->left; 855 if (y == NULL) 856 break; 857 if (addr < y->start && y->left != NULL) { 858 /* Rotate right and put y on rlist. */ 859 root->left = y->right; 860 y->right = root; 861 vm_map_entry_set_max_free(root); 862 root = y->left; 863 y->left = rlist; 864 rlist = y; 865 } else { 866 /* Put root on rlist. */ 867 root->left = rlist; 868 rlist = root; 869 root = y; 870 } 871 } else if (addr >= root->end) { 872 y = root->right; 873 if (y == NULL) 874 break; 875 if (addr >= y->end && y->right != NULL) { 876 /* Rotate left and put y on llist. */ 877 root->right = y->left; 878 y->left = root; 879 vm_map_entry_set_max_free(root); 880 root = y->right; 881 y->right = llist; 882 llist = y; 883 } else { 884 /* Put root on llist. */ 885 root->right = llist; 886 llist = root; 887 root = y; 888 } 889 } else 890 break; 891 } 892 893 /* 894 * Pass Two: Walk back up the two spines, flip the pointers 895 * and set max_free. The subtrees of the root go at the 896 * bottom of llist and rlist. 897 */ 898 ltree = root->left; 899 while (llist != NULL) { 900 y = llist->right; 901 llist->right = ltree; 902 vm_map_entry_set_max_free(llist); 903 ltree = llist; 904 llist = y; 905 } 906 rtree = root->right; 907 while (rlist != NULL) { 908 y = rlist->left; 909 rlist->left = rtree; 910 vm_map_entry_set_max_free(rlist); 911 rtree = rlist; 912 rlist = y; 913 } 914 915 /* 916 * Final assembly: add ltree and rtree as subtrees of root. 917 */ 918 root->left = ltree; 919 root->right = rtree; 920 vm_map_entry_set_max_free(root); 921 922 return (root); 923} 924 925/* 926 * vm_map_entry_{un,}link: 927 * 928 * Insert/remove entries from maps. 929 */ 930static void 931vm_map_entry_link(vm_map_t map, 932 vm_map_entry_t after_where, 933 vm_map_entry_t entry) 934{ 935 936 CTR4(KTR_VM, 937 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 938 map->nentries, entry, after_where); 939 VM_MAP_ASSERT_LOCKED(map); 940 map->nentries++; 941 entry->prev = after_where; 942 entry->next = after_where->next; 943 entry->next->prev = entry; 944 after_where->next = entry; 945 946 if (after_where != &map->header) { 947 if (after_where != map->root) 948 vm_map_entry_splay(after_where->start, map->root); 949 entry->right = after_where->right; 950 entry->left = after_where; 951 after_where->right = NULL; 952 after_where->adj_free = entry->start - after_where->end; 953 vm_map_entry_set_max_free(after_where); 954 } else { 955 entry->right = map->root; 956 entry->left = NULL; 957 } 958 entry->adj_free = (entry->next == &map->header ? map->max_offset : 959 entry->next->start) - entry->end; 960 vm_map_entry_set_max_free(entry); 961 map->root = entry; 962} 963 964static void 965vm_map_entry_unlink(vm_map_t map, 966 vm_map_entry_t entry) 967{ 968 vm_map_entry_t next, prev, root; 969 970 VM_MAP_ASSERT_LOCKED(map); 971 if (entry != map->root) 972 vm_map_entry_splay(entry->start, map->root); 973 if (entry->left == NULL) 974 root = entry->right; 975 else { 976 root = vm_map_entry_splay(entry->start, entry->left); 977 root->right = entry->right; 978 root->adj_free = (entry->next == &map->header ? map->max_offset : 979 entry->next->start) - root->end; 980 vm_map_entry_set_max_free(root); 981 } 982 map->root = root; 983 984 prev = entry->prev; 985 next = entry->next; 986 next->prev = prev; 987 prev->next = next; 988 map->nentries--; 989 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 990 map->nentries, entry); 991} 992 993/* 994 * vm_map_entry_resize_free: 995 * 996 * Recompute the amount of free space following a vm_map_entry 997 * and propagate that value up the tree. Call this function after 998 * resizing a map entry in-place, that is, without a call to 999 * vm_map_entry_link() or _unlink(). 1000 * 1001 * The map must be locked, and leaves it so. 1002 */ 1003static void 1004vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1005{ 1006 1007 /* 1008 * Using splay trees without parent pointers, propagating 1009 * max_free up the tree is done by moving the entry to the 1010 * root and making the change there. 1011 */ 1012 if (entry != map->root) 1013 map->root = vm_map_entry_splay(entry->start, map->root); 1014 1015 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1016 entry->next->start) - entry->end; 1017 vm_map_entry_set_max_free(entry); 1018} 1019 1020/* 1021 * vm_map_lookup_entry: [ internal use only ] 1022 * 1023 * Finds the map entry containing (or 1024 * immediately preceding) the specified address 1025 * in the given map; the entry is returned 1026 * in the "entry" parameter. The boolean 1027 * result indicates whether the address is 1028 * actually contained in the map. 1029 */ 1030boolean_t 1031vm_map_lookup_entry( 1032 vm_map_t map, 1033 vm_offset_t address, 1034 vm_map_entry_t *entry) /* OUT */ 1035{ 1036 vm_map_entry_t cur; 1037 boolean_t locked; 1038 1039 /* 1040 * If the map is empty, then the map entry immediately preceding 1041 * "address" is the map's header. 1042 */ 1043 cur = map->root; 1044 if (cur == NULL) 1045 *entry = &map->header; 1046 else if (address >= cur->start && cur->end > address) { 1047 *entry = cur; 1048 return (TRUE); 1049 } else if ((locked = vm_map_locked(map)) || 1050 sx_try_upgrade(&map->lock)) { 1051 /* 1052 * Splay requires a write lock on the map. However, it only 1053 * restructures the binary search tree; it does not otherwise 1054 * change the map. Thus, the map's timestamp need not change 1055 * on a temporary upgrade. 1056 */ 1057 map->root = cur = vm_map_entry_splay(address, cur); 1058 if (!locked) 1059 sx_downgrade(&map->lock); 1060 1061 /* 1062 * If "address" is contained within a map entry, the new root 1063 * is that map entry. Otherwise, the new root is a map entry 1064 * immediately before or after "address". 1065 */ 1066 if (address >= cur->start) { 1067 *entry = cur; 1068 if (cur->end > address) 1069 return (TRUE); 1070 } else 1071 *entry = cur->prev; 1072 } else 1073 /* 1074 * Since the map is only locked for read access, perform a 1075 * standard binary search tree lookup for "address". 1076 */ 1077 for (;;) { 1078 if (address < cur->start) { 1079 if (cur->left == NULL) { 1080 *entry = cur->prev; 1081 break; 1082 } 1083 cur = cur->left; 1084 } else if (cur->end > address) { 1085 *entry = cur; 1086 return (TRUE); 1087 } else { 1088 if (cur->right == NULL) { 1089 *entry = cur; 1090 break; 1091 } 1092 cur = cur->right; 1093 } 1094 } 1095 return (FALSE); 1096} 1097 1098/* 1099 * vm_map_insert: 1100 * 1101 * Inserts the given whole VM object into the target 1102 * map at the specified address range. The object's 1103 * size should match that of the address range. 1104 * 1105 * Requires that the map be locked, and leaves it so. 1106 * 1107 * If object is non-NULL, ref count must be bumped by caller 1108 * prior to making call to account for the new entry. 1109 */ 1110int 1111vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1112 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1113 int cow) 1114{ 1115 vm_map_entry_t new_entry; 1116 vm_map_entry_t prev_entry; 1117 vm_map_entry_t temp_entry; 1118 vm_eflags_t protoeflags; 1119 struct ucred *cred; 1120 vm_inherit_t inheritance; 1121 boolean_t charge_prev_obj; 1122 1123 VM_MAP_ASSERT_LOCKED(map); 1124 1125 /* 1126 * Check that the start and end points are not bogus. 1127 */ 1128 if ((start < map->min_offset) || (end > map->max_offset) || 1129 (start >= end)) 1130 return (KERN_INVALID_ADDRESS); 1131 1132 /* 1133 * Find the entry prior to the proposed starting address; if it's part 1134 * of an existing entry, this range is bogus. 1135 */ 1136 if (vm_map_lookup_entry(map, start, &temp_entry)) 1137 return (KERN_NO_SPACE); 1138 1139 prev_entry = temp_entry; 1140 1141 /* 1142 * Assert that the next entry doesn't overlap the end point. 1143 */ 1144 if ((prev_entry->next != &map->header) && 1145 (prev_entry->next->start < end)) 1146 return (KERN_NO_SPACE); 1147 1148 protoeflags = 0; 1149 charge_prev_obj = FALSE; 1150 1151 if (cow & MAP_COPY_ON_WRITE) 1152 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1153 1154 if (cow & MAP_NOFAULT) { 1155 protoeflags |= MAP_ENTRY_NOFAULT; 1156 1157 KASSERT(object == NULL, 1158 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1159 } 1160 if (cow & MAP_DISABLE_SYNCER) 1161 protoeflags |= MAP_ENTRY_NOSYNC; 1162 if (cow & MAP_DISABLE_COREDUMP) 1163 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1164 if (cow & MAP_VN_WRITECOUNT) 1165 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1166 if (cow & MAP_INHERIT_SHARE) 1167 inheritance = VM_INHERIT_SHARE; 1168 else 1169 inheritance = VM_INHERIT_DEFAULT; 1170 1171 cred = NULL; 1172 KASSERT((object != kmem_object && object != kernel_object) || 1173 ((object == kmem_object || object == kernel_object) && 1174 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1175 ("kmem or kernel object and cow")); 1176 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1177 goto charged; 1178 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1179 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1180 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1181 return (KERN_RESOURCE_SHORTAGE); 1182 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1183 object->cred == NULL, 1184 ("OVERCOMMIT: vm_map_insert o %p", object)); 1185 cred = curthread->td_ucred; 1186 crhold(cred); 1187 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1188 charge_prev_obj = TRUE; 1189 } 1190 1191charged: 1192 /* Expand the kernel pmap, if necessary. */ 1193 if (map == kernel_map && end > kernel_vm_end) 1194 pmap_growkernel(end); 1195 if (object != NULL) { 1196 /* 1197 * OBJ_ONEMAPPING must be cleared unless this mapping 1198 * is trivially proven to be the only mapping for any 1199 * of the object's pages. (Object granularity 1200 * reference counting is insufficient to recognize 1201 * aliases with precision.) 1202 */ 1203 VM_OBJECT_WLOCK(object); 1204 if (object->ref_count > 1 || object->shadow_count != 0) 1205 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1206 VM_OBJECT_WUNLOCK(object); 1207 } 1208 else if ((prev_entry != &map->header) && 1209 (prev_entry->eflags == protoeflags) && 1210 (prev_entry->end == start) && 1211 (prev_entry->wired_count == 0) && 1212 (prev_entry->cred == cred || 1213 (prev_entry->object.vm_object != NULL && 1214 (prev_entry->object.vm_object->cred == cred))) && 1215 vm_object_coalesce(prev_entry->object.vm_object, 1216 prev_entry->offset, 1217 (vm_size_t)(prev_entry->end - prev_entry->start), 1218 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1219 /* 1220 * We were able to extend the object. Determine if we 1221 * can extend the previous map entry to include the 1222 * new range as well. 1223 */ 1224 if ((prev_entry->inheritance == inheritance) && 1225 (prev_entry->protection == prot) && 1226 (prev_entry->max_protection == max)) { 1227 map->size += (end - prev_entry->end); 1228 prev_entry->end = end; 1229 vm_map_entry_resize_free(map, prev_entry); 1230 vm_map_simplify_entry(map, prev_entry); 1231 if (cred != NULL) 1232 crfree(cred); 1233 return (KERN_SUCCESS); 1234 } 1235 1236 /* 1237 * If we can extend the object but cannot extend the 1238 * map entry, we have to create a new map entry. We 1239 * must bump the ref count on the extended object to 1240 * account for it. object may be NULL. 1241 */ 1242 object = prev_entry->object.vm_object; 1243 offset = prev_entry->offset + 1244 (prev_entry->end - prev_entry->start); 1245 vm_object_reference(object); 1246 if (cred != NULL && object != NULL && object->cred != NULL && 1247 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1248 /* Object already accounts for this uid. */ 1249 crfree(cred); 1250 cred = NULL; 1251 } 1252 } 1253 1254 /* 1255 * NOTE: if conditionals fail, object can be NULL here. This occurs 1256 * in things like the buffer map where we manage kva but do not manage 1257 * backing objects. 1258 */ 1259 1260 /* 1261 * Create a new entry 1262 */ 1263 new_entry = vm_map_entry_create(map); 1264 new_entry->start = start; 1265 new_entry->end = end; 1266 new_entry->cred = NULL; 1267 1268 new_entry->eflags = protoeflags; 1269 new_entry->object.vm_object = object; 1270 new_entry->offset = offset; 1271 new_entry->avail_ssize = 0; 1272 1273 new_entry->inheritance = inheritance; 1274 new_entry->protection = prot; 1275 new_entry->max_protection = max; 1276 new_entry->wired_count = 0; 1277 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1278 new_entry->next_read = OFF_TO_IDX(offset); 1279 1280 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1281 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1282 new_entry->cred = cred; 1283 1284 /* 1285 * Insert the new entry into the list 1286 */ 1287 vm_map_entry_link(map, prev_entry, new_entry); 1288 map->size += new_entry->end - new_entry->start; 1289 1290 /* 1291 * It may be possible to merge the new entry with the next and/or 1292 * previous entries. However, due to MAP_STACK_* being a hack, a 1293 * panic can result from merging such entries. 1294 */ 1295 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1296 vm_map_simplify_entry(map, new_entry); 1297 1298 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1299 vm_map_pmap_enter(map, start, prot, 1300 object, OFF_TO_IDX(offset), end - start, 1301 cow & MAP_PREFAULT_PARTIAL); 1302 } 1303 1304 return (KERN_SUCCESS); 1305} 1306 1307/* 1308 * vm_map_findspace: 1309 * 1310 * Find the first fit (lowest VM address) for "length" free bytes 1311 * beginning at address >= start in the given map. 1312 * 1313 * In a vm_map_entry, "adj_free" is the amount of free space 1314 * adjacent (higher address) to this entry, and "max_free" is the 1315 * maximum amount of contiguous free space in its subtree. This 1316 * allows finding a free region in one path down the tree, so 1317 * O(log n) amortized with splay trees. 1318 * 1319 * The map must be locked, and leaves it so. 1320 * 1321 * Returns: 0 on success, and starting address in *addr, 1322 * 1 if insufficient space. 1323 */ 1324int 1325vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1326 vm_offset_t *addr) /* OUT */ 1327{ 1328 vm_map_entry_t entry; 1329 vm_offset_t st; 1330 1331 /* 1332 * Request must fit within min/max VM address and must avoid 1333 * address wrap. 1334 */ 1335 if (start < map->min_offset) 1336 start = map->min_offset; 1337 if (start + length > map->max_offset || start + length < start) 1338 return (1); 1339 1340 /* Empty tree means wide open address space. */ 1341 if (map->root == NULL) { 1342 *addr = start; 1343 return (0); 1344 } 1345 1346 /* 1347 * After splay, if start comes before root node, then there 1348 * must be a gap from start to the root. 1349 */ 1350 map->root = vm_map_entry_splay(start, map->root); 1351 if (start + length <= map->root->start) { 1352 *addr = start; 1353 return (0); 1354 } 1355 1356 /* 1357 * Root is the last node that might begin its gap before 1358 * start, and this is the last comparison where address 1359 * wrap might be a problem. 1360 */ 1361 st = (start > map->root->end) ? start : map->root->end; 1362 if (length <= map->root->end + map->root->adj_free - st) { 1363 *addr = st; 1364 return (0); 1365 } 1366 1367 /* With max_free, can immediately tell if no solution. */ 1368 entry = map->root->right; 1369 if (entry == NULL || length > entry->max_free) 1370 return (1); 1371 1372 /* 1373 * Search the right subtree in the order: left subtree, root, 1374 * right subtree (first fit). The previous splay implies that 1375 * all regions in the right subtree have addresses > start. 1376 */ 1377 while (entry != NULL) { 1378 if (entry->left != NULL && entry->left->max_free >= length) 1379 entry = entry->left; 1380 else if (entry->adj_free >= length) { 1381 *addr = entry->end; 1382 return (0); 1383 } else 1384 entry = entry->right; 1385 } 1386 1387 /* Can't get here, so panic if we do. */ 1388 panic("vm_map_findspace: max_free corrupt"); 1389} 1390 1391int 1392vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1393 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1394 vm_prot_t max, int cow) 1395{ 1396 vm_offset_t end; 1397 int result; 1398 1399 end = start + length; 1400 vm_map_lock(map); 1401 VM_MAP_RANGE_CHECK(map, start, end); 1402 (void) vm_map_delete(map, start, end); 1403 result = vm_map_insert(map, object, offset, start, end, prot, 1404 max, cow); 1405 vm_map_unlock(map); 1406 return (result); 1407} 1408 1409/* 1410 * vm_map_find finds an unallocated region in the target address 1411 * map with the given length. The search is defined to be 1412 * first-fit from the specified address; the region found is 1413 * returned in the same parameter. 1414 * 1415 * If object is non-NULL, ref count must be bumped by caller 1416 * prior to making call to account for the new entry. 1417 */ 1418int 1419vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1420 vm_offset_t *addr, /* IN/OUT */ 1421 vm_size_t length, vm_offset_t max_addr, int find_space, 1422 vm_prot_t prot, vm_prot_t max, int cow) 1423{ 1424 vm_offset_t alignment, initial_addr, start; 1425 int result; 1426 1427 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1428 (object->flags & OBJ_COLORED) == 0)) 1429 find_space = VMFS_ANY_SPACE; 1430 if (find_space >> 8 != 0) { 1431 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1432 alignment = (vm_offset_t)1 << (find_space >> 8); 1433 } else 1434 alignment = 0; 1435 initial_addr = *addr; 1436again: 1437 start = initial_addr; 1438 vm_map_lock(map); 1439 do { 1440 if (find_space != VMFS_NO_SPACE) { 1441 if (vm_map_findspace(map, start, length, addr) || 1442 (max_addr != 0 && *addr + length > max_addr)) { 1443 vm_map_unlock(map); 1444 if (find_space == VMFS_OPTIMAL_SPACE) { 1445 find_space = VMFS_ANY_SPACE; 1446 goto again; 1447 } 1448 return (KERN_NO_SPACE); 1449 } 1450 switch (find_space) { 1451 case VMFS_SUPER_SPACE: 1452 case VMFS_OPTIMAL_SPACE: 1453 pmap_align_superpage(object, offset, addr, 1454 length); 1455 break; 1456 case VMFS_ANY_SPACE: 1457 break; 1458 default: 1459 if ((*addr & (alignment - 1)) != 0) { 1460 *addr &= ~(alignment - 1); 1461 *addr += alignment; 1462 } 1463 break; 1464 } 1465 1466 start = *addr; 1467 } 1468 result = vm_map_insert(map, object, offset, start, start + 1469 length, prot, max, cow); 1470 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1471 find_space != VMFS_ANY_SPACE); 1472 vm_map_unlock(map); 1473 return (result); 1474} 1475 1476/* 1477 * vm_map_simplify_entry: 1478 * 1479 * Simplify the given map entry by merging with either neighbor. This 1480 * routine also has the ability to merge with both neighbors. 1481 * 1482 * The map must be locked. 1483 * 1484 * This routine guarentees that the passed entry remains valid (though 1485 * possibly extended). When merging, this routine may delete one or 1486 * both neighbors. 1487 */ 1488void 1489vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1490{ 1491 vm_map_entry_t next, prev; 1492 vm_size_t prevsize, esize; 1493 1494 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1495 return; 1496 1497 prev = entry->prev; 1498 if (prev != &map->header) { 1499 prevsize = prev->end - prev->start; 1500 if ( (prev->end == entry->start) && 1501 (prev->object.vm_object == entry->object.vm_object) && 1502 (!prev->object.vm_object || 1503 (prev->offset + prevsize == entry->offset)) && 1504 (prev->eflags == entry->eflags) && 1505 (prev->protection == entry->protection) && 1506 (prev->max_protection == entry->max_protection) && 1507 (prev->inheritance == entry->inheritance) && 1508 (prev->wired_count == entry->wired_count) && 1509 (prev->cred == entry->cred)) { 1510 vm_map_entry_unlink(map, prev); 1511 entry->start = prev->start; 1512 entry->offset = prev->offset; 1513 if (entry->prev != &map->header) 1514 vm_map_entry_resize_free(map, entry->prev); 1515 1516 /* 1517 * If the backing object is a vnode object, 1518 * vm_object_deallocate() calls vrele(). 1519 * However, vrele() does not lock the vnode 1520 * because the vnode has additional 1521 * references. Thus, the map lock can be kept 1522 * without causing a lock-order reversal with 1523 * the vnode lock. 1524 * 1525 * Since we count the number of virtual page 1526 * mappings in object->un_pager.vnp.writemappings, 1527 * the writemappings value should not be adjusted 1528 * when the entry is disposed of. 1529 */ 1530 if (prev->object.vm_object) 1531 vm_object_deallocate(prev->object.vm_object); 1532 if (prev->cred != NULL) 1533 crfree(prev->cred); 1534 vm_map_entry_dispose(map, prev); 1535 } 1536 } 1537 1538 next = entry->next; 1539 if (next != &map->header) { 1540 esize = entry->end - entry->start; 1541 if ((entry->end == next->start) && 1542 (next->object.vm_object == entry->object.vm_object) && 1543 (!entry->object.vm_object || 1544 (entry->offset + esize == next->offset)) && 1545 (next->eflags == entry->eflags) && 1546 (next->protection == entry->protection) && 1547 (next->max_protection == entry->max_protection) && 1548 (next->inheritance == entry->inheritance) && 1549 (next->wired_count == entry->wired_count) && 1550 (next->cred == entry->cred)) { 1551 vm_map_entry_unlink(map, next); 1552 entry->end = next->end; 1553 vm_map_entry_resize_free(map, entry); 1554 1555 /* 1556 * See comment above. 1557 */ 1558 if (next->object.vm_object) 1559 vm_object_deallocate(next->object.vm_object); 1560 if (next->cred != NULL) 1561 crfree(next->cred); 1562 vm_map_entry_dispose(map, next); 1563 } 1564 } 1565} 1566/* 1567 * vm_map_clip_start: [ internal use only ] 1568 * 1569 * Asserts that the given entry begins at or after 1570 * the specified address; if necessary, 1571 * it splits the entry into two. 1572 */ 1573#define vm_map_clip_start(map, entry, startaddr) \ 1574{ \ 1575 if (startaddr > entry->start) \ 1576 _vm_map_clip_start(map, entry, startaddr); \ 1577} 1578 1579/* 1580 * This routine is called only when it is known that 1581 * the entry must be split. 1582 */ 1583static void 1584_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1585{ 1586 vm_map_entry_t new_entry; 1587 1588 VM_MAP_ASSERT_LOCKED(map); 1589 1590 /* 1591 * Split off the front portion -- note that we must insert the new 1592 * entry BEFORE this one, so that this entry has the specified 1593 * starting address. 1594 */ 1595 vm_map_simplify_entry(map, entry); 1596 1597 /* 1598 * If there is no object backing this entry, we might as well create 1599 * one now. If we defer it, an object can get created after the map 1600 * is clipped, and individual objects will be created for the split-up 1601 * map. This is a bit of a hack, but is also about the best place to 1602 * put this improvement. 1603 */ 1604 if (entry->object.vm_object == NULL && !map->system_map) { 1605 vm_object_t object; 1606 object = vm_object_allocate(OBJT_DEFAULT, 1607 atop(entry->end - entry->start)); 1608 entry->object.vm_object = object; 1609 entry->offset = 0; 1610 if (entry->cred != NULL) { 1611 object->cred = entry->cred; 1612 object->charge = entry->end - entry->start; 1613 entry->cred = NULL; 1614 } 1615 } else if (entry->object.vm_object != NULL && 1616 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1617 entry->cred != NULL) { 1618 VM_OBJECT_WLOCK(entry->object.vm_object); 1619 KASSERT(entry->object.vm_object->cred == NULL, 1620 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1621 entry->object.vm_object->cred = entry->cred; 1622 entry->object.vm_object->charge = entry->end - entry->start; 1623 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1624 entry->cred = NULL; 1625 } 1626 1627 new_entry = vm_map_entry_create(map); 1628 *new_entry = *entry; 1629 1630 new_entry->end = start; 1631 entry->offset += (start - entry->start); 1632 entry->start = start; 1633 if (new_entry->cred != NULL) 1634 crhold(entry->cred); 1635 1636 vm_map_entry_link(map, entry->prev, new_entry); 1637 1638 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1639 vm_object_reference(new_entry->object.vm_object); 1640 /* 1641 * The object->un_pager.vnp.writemappings for the 1642 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1643 * kept as is here. The virtual pages are 1644 * re-distributed among the clipped entries, so the sum is 1645 * left the same. 1646 */ 1647 } 1648} 1649 1650/* 1651 * vm_map_clip_end: [ internal use only ] 1652 * 1653 * Asserts that the given entry ends at or before 1654 * the specified address; if necessary, 1655 * it splits the entry into two. 1656 */ 1657#define vm_map_clip_end(map, entry, endaddr) \ 1658{ \ 1659 if ((endaddr) < (entry->end)) \ 1660 _vm_map_clip_end((map), (entry), (endaddr)); \ 1661} 1662 1663/* 1664 * This routine is called only when it is known that 1665 * the entry must be split. 1666 */ 1667static void 1668_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1669{ 1670 vm_map_entry_t new_entry; 1671 1672 VM_MAP_ASSERT_LOCKED(map); 1673 1674 /* 1675 * If there is no object backing this entry, we might as well create 1676 * one now. If we defer it, an object can get created after the map 1677 * is clipped, and individual objects will be created for the split-up 1678 * map. This is a bit of a hack, but is also about the best place to 1679 * put this improvement. 1680 */ 1681 if (entry->object.vm_object == NULL && !map->system_map) { 1682 vm_object_t object; 1683 object = vm_object_allocate(OBJT_DEFAULT, 1684 atop(entry->end - entry->start)); 1685 entry->object.vm_object = object; 1686 entry->offset = 0; 1687 if (entry->cred != NULL) { 1688 object->cred = entry->cred; 1689 object->charge = entry->end - entry->start; 1690 entry->cred = NULL; 1691 } 1692 } else if (entry->object.vm_object != NULL && 1693 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1694 entry->cred != NULL) { 1695 VM_OBJECT_WLOCK(entry->object.vm_object); 1696 KASSERT(entry->object.vm_object->cred == NULL, 1697 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1698 entry->object.vm_object->cred = entry->cred; 1699 entry->object.vm_object->charge = entry->end - entry->start; 1700 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1701 entry->cred = NULL; 1702 } 1703 1704 /* 1705 * Create a new entry and insert it AFTER the specified entry 1706 */ 1707 new_entry = vm_map_entry_create(map); 1708 *new_entry = *entry; 1709 1710 new_entry->start = entry->end = end; 1711 new_entry->offset += (end - entry->start); 1712 if (new_entry->cred != NULL) 1713 crhold(entry->cred); 1714 1715 vm_map_entry_link(map, entry, new_entry); 1716 1717 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1718 vm_object_reference(new_entry->object.vm_object); 1719 } 1720} 1721 1722/* 1723 * vm_map_submap: [ kernel use only ] 1724 * 1725 * Mark the given range as handled by a subordinate map. 1726 * 1727 * This range must have been created with vm_map_find, 1728 * and no other operations may have been performed on this 1729 * range prior to calling vm_map_submap. 1730 * 1731 * Only a limited number of operations can be performed 1732 * within this rage after calling vm_map_submap: 1733 * vm_fault 1734 * [Don't try vm_map_copy!] 1735 * 1736 * To remove a submapping, one must first remove the 1737 * range from the superior map, and then destroy the 1738 * submap (if desired). [Better yet, don't try it.] 1739 */ 1740int 1741vm_map_submap( 1742 vm_map_t map, 1743 vm_offset_t start, 1744 vm_offset_t end, 1745 vm_map_t submap) 1746{ 1747 vm_map_entry_t entry; 1748 int result = KERN_INVALID_ARGUMENT; 1749 1750 vm_map_lock(map); 1751 1752 VM_MAP_RANGE_CHECK(map, start, end); 1753 1754 if (vm_map_lookup_entry(map, start, &entry)) { 1755 vm_map_clip_start(map, entry, start); 1756 } else 1757 entry = entry->next; 1758 1759 vm_map_clip_end(map, entry, end); 1760 1761 if ((entry->start == start) && (entry->end == end) && 1762 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1763 (entry->object.vm_object == NULL)) { 1764 entry->object.sub_map = submap; 1765 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1766 result = KERN_SUCCESS; 1767 } 1768 vm_map_unlock(map); 1769 1770 return (result); 1771} 1772 1773/* 1774 * The maximum number of pages to map 1775 */ 1776#define MAX_INIT_PT 96 1777 1778/* 1779 * vm_map_pmap_enter: 1780 * 1781 * Preload read-only mappings for the specified object's resident pages 1782 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only 1783 * the resident pages within the address range [addr, addr + ulmin(size, 1784 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within 1785 * the specified address range are mapped. This eliminates many soft 1786 * faults on process startup and immediately after an mmap(2). Because 1787 * these are speculative mappings, cached pages are not reactivated and 1788 * mapped. 1789 */ 1790void 1791vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1792 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1793{ 1794 vm_offset_t start; 1795 vm_page_t p, p_start; 1796 vm_pindex_t psize, tmpidx; 1797 1798 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1799 return; 1800 VM_OBJECT_RLOCK(object); 1801 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1802 VM_OBJECT_RUNLOCK(object); 1803 VM_OBJECT_WLOCK(object); 1804 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1805 pmap_object_init_pt(map->pmap, addr, object, pindex, 1806 size); 1807 VM_OBJECT_WUNLOCK(object); 1808 return; 1809 } 1810 VM_OBJECT_LOCK_DOWNGRADE(object); 1811 } 1812 1813 psize = atop(size); 1814 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0) 1815 psize = MAX_INIT_PT; 1816 if (psize + pindex > object->size) { 1817 if (object->size < pindex) { 1818 VM_OBJECT_RUNLOCK(object); 1819 return; 1820 } 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 VM_OBJECT_RUNLOCK(object); 1860} 1861 1862/* 1863 * vm_map_protect: 1864 * 1865 * Sets the protection of the specified address 1866 * region in the target map. If "set_max" is 1867 * specified, the maximum protection is to be set; 1868 * otherwise, only the current protection is affected. 1869 */ 1870int 1871vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1872 vm_prot_t new_prot, boolean_t set_max) 1873{ 1874 vm_map_entry_t current, entry; 1875 vm_object_t obj; 1876 struct ucred *cred; 1877 vm_prot_t old_prot; 1878 1879 vm_map_lock(map); 1880 1881 VM_MAP_RANGE_CHECK(map, start, end); 1882 1883 if (vm_map_lookup_entry(map, start, &entry)) { 1884 vm_map_clip_start(map, entry, start); 1885 } else { 1886 entry = entry->next; 1887 } 1888 1889 /* 1890 * Make a first pass to check for protection violations. 1891 */ 1892 current = entry; 1893 while ((current != &map->header) && (current->start < end)) { 1894 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1895 vm_map_unlock(map); 1896 return (KERN_INVALID_ARGUMENT); 1897 } 1898 if ((new_prot & current->max_protection) != new_prot) { 1899 vm_map_unlock(map); 1900 return (KERN_PROTECTION_FAILURE); 1901 } 1902 current = current->next; 1903 } 1904 1905 1906 /* 1907 * Do an accounting pass for private read-only mappings that 1908 * now will do cow due to allowed write (e.g. debugger sets 1909 * breakpoint on text segment) 1910 */ 1911 for (current = entry; (current != &map->header) && 1912 (current->start < end); current = current->next) { 1913 1914 vm_map_clip_end(map, current, end); 1915 1916 if (set_max || 1917 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1918 ENTRY_CHARGED(current)) { 1919 continue; 1920 } 1921 1922 cred = curthread->td_ucred; 1923 obj = current->object.vm_object; 1924 1925 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1926 if (!swap_reserve(current->end - current->start)) { 1927 vm_map_unlock(map); 1928 return (KERN_RESOURCE_SHORTAGE); 1929 } 1930 crhold(cred); 1931 current->cred = cred; 1932 continue; 1933 } 1934 1935 VM_OBJECT_WLOCK(obj); 1936 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1937 VM_OBJECT_WUNLOCK(obj); 1938 continue; 1939 } 1940 1941 /* 1942 * Charge for the whole object allocation now, since 1943 * we cannot distinguish between non-charged and 1944 * charged clipped mapping of the same object later. 1945 */ 1946 KASSERT(obj->charge == 0, 1947 ("vm_map_protect: object %p overcharged\n", obj)); 1948 if (!swap_reserve(ptoa(obj->size))) { 1949 VM_OBJECT_WUNLOCK(obj); 1950 vm_map_unlock(map); 1951 return (KERN_RESOURCE_SHORTAGE); 1952 } 1953 1954 crhold(cred); 1955 obj->cred = cred; 1956 obj->charge = ptoa(obj->size); 1957 VM_OBJECT_WUNLOCK(obj); 1958 } 1959 1960 /* 1961 * Go back and fix up protections. [Note that clipping is not 1962 * necessary the second time.] 1963 */ 1964 current = entry; 1965 while ((current != &map->header) && (current->start < end)) { 1966 old_prot = current->protection; 1967 1968 if (set_max) 1969 current->protection = 1970 (current->max_protection = new_prot) & 1971 old_prot; 1972 else 1973 current->protection = new_prot; 1974 1975 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED)) 1976 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) && 1977 (current->protection & VM_PROT_WRITE) != 0 && 1978 (old_prot & VM_PROT_WRITE) == 0) { 1979 vm_fault_copy_entry(map, map, current, current, NULL); 1980 } 1981 1982 /* 1983 * When restricting access, update the physical map. Worry 1984 * about copy-on-write here. 1985 */ 1986 if ((old_prot & ~current->protection) != 0) { 1987#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1988 VM_PROT_ALL) 1989 pmap_protect(map->pmap, current->start, 1990 current->end, 1991 current->protection & MASK(current)); 1992#undef MASK 1993 } 1994 vm_map_simplify_entry(map, current); 1995 current = current->next; 1996 } 1997 vm_map_unlock(map); 1998 return (KERN_SUCCESS); 1999} 2000 2001/* 2002 * vm_map_madvise: 2003 * 2004 * This routine traverses a processes map handling the madvise 2005 * system call. Advisories are classified as either those effecting 2006 * the vm_map_entry structure, or those effecting the underlying 2007 * objects. 2008 */ 2009int 2010vm_map_madvise( 2011 vm_map_t map, 2012 vm_offset_t start, 2013 vm_offset_t end, 2014 int behav) 2015{ 2016 vm_map_entry_t current, entry; 2017 int modify_map = 0; 2018 2019 /* 2020 * Some madvise calls directly modify the vm_map_entry, in which case 2021 * we need to use an exclusive lock on the map and we need to perform 2022 * various clipping operations. Otherwise we only need a read-lock 2023 * on the map. 2024 */ 2025 switch(behav) { 2026 case MADV_NORMAL: 2027 case MADV_SEQUENTIAL: 2028 case MADV_RANDOM: 2029 case MADV_NOSYNC: 2030 case MADV_AUTOSYNC: 2031 case MADV_NOCORE: 2032 case MADV_CORE: 2033 modify_map = 1; 2034 vm_map_lock(map); 2035 break; 2036 case MADV_WILLNEED: 2037 case MADV_DONTNEED: 2038 case MADV_FREE: 2039 vm_map_lock_read(map); 2040 break; 2041 default: 2042 return (KERN_INVALID_ARGUMENT); 2043 } 2044 2045 /* 2046 * Locate starting entry and clip if necessary. 2047 */ 2048 VM_MAP_RANGE_CHECK(map, start, end); 2049 2050 if (vm_map_lookup_entry(map, start, &entry)) { 2051 if (modify_map) 2052 vm_map_clip_start(map, entry, start); 2053 } else { 2054 entry = entry->next; 2055 } 2056 2057 if (modify_map) { 2058 /* 2059 * madvise behaviors that are implemented in the vm_map_entry. 2060 * 2061 * We clip the vm_map_entry so that behavioral changes are 2062 * limited to the specified address range. 2063 */ 2064 for (current = entry; 2065 (current != &map->header) && (current->start < end); 2066 current = current->next 2067 ) { 2068 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2069 continue; 2070 2071 vm_map_clip_end(map, current, end); 2072 2073 switch (behav) { 2074 case MADV_NORMAL: 2075 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2076 break; 2077 case MADV_SEQUENTIAL: 2078 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2079 break; 2080 case MADV_RANDOM: 2081 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2082 break; 2083 case MADV_NOSYNC: 2084 current->eflags |= MAP_ENTRY_NOSYNC; 2085 break; 2086 case MADV_AUTOSYNC: 2087 current->eflags &= ~MAP_ENTRY_NOSYNC; 2088 break; 2089 case MADV_NOCORE: 2090 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2091 break; 2092 case MADV_CORE: 2093 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2094 break; 2095 default: 2096 break; 2097 } 2098 vm_map_simplify_entry(map, current); 2099 } 2100 vm_map_unlock(map); 2101 } else { 2102 vm_pindex_t pstart, pend; 2103 2104 /* 2105 * madvise behaviors that are implemented in the underlying 2106 * vm_object. 2107 * 2108 * Since we don't clip the vm_map_entry, we have to clip 2109 * the vm_object pindex and count. 2110 */ 2111 for (current = entry; 2112 (current != &map->header) && (current->start < end); 2113 current = current->next 2114 ) { 2115 vm_offset_t useEnd, useStart; 2116 2117 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2118 continue; 2119 2120 pstart = OFF_TO_IDX(current->offset); 2121 pend = pstart + atop(current->end - current->start); 2122 useStart = current->start; 2123 useEnd = current->end; 2124 2125 if (current->start < start) { 2126 pstart += atop(start - current->start); 2127 useStart = start; 2128 } 2129 if (current->end > end) { 2130 pend -= atop(current->end - end); 2131 useEnd = end; 2132 } 2133 2134 if (pstart >= pend) 2135 continue; 2136 2137 /* 2138 * Perform the pmap_advise() before clearing 2139 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2140 * concurrent pmap operation, such as pmap_remove(), 2141 * could clear a reference in the pmap and set 2142 * PGA_REFERENCED on the page before the pmap_advise() 2143 * had completed. Consequently, the page would appear 2144 * referenced based upon an old reference that 2145 * occurred before this pmap_advise() ran. 2146 */ 2147 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2148 pmap_advise(map->pmap, useStart, useEnd, 2149 behav); 2150 2151 vm_object_madvise(current->object.vm_object, pstart, 2152 pend, behav); 2153 if (behav == MADV_WILLNEED) { 2154 vm_map_pmap_enter(map, 2155 useStart, 2156 current->protection, 2157 current->object.vm_object, 2158 pstart, 2159 ptoa(pend - pstart), 2160 MAP_PREFAULT_MADVISE 2161 ); 2162 } 2163 } 2164 vm_map_unlock_read(map); 2165 } 2166 return (0); 2167} 2168 2169 2170/* 2171 * vm_map_inherit: 2172 * 2173 * Sets the inheritance of the specified address 2174 * range in the target map. Inheritance 2175 * affects how the map will be shared with 2176 * child maps at the time of vmspace_fork. 2177 */ 2178int 2179vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2180 vm_inherit_t new_inheritance) 2181{ 2182 vm_map_entry_t entry; 2183 vm_map_entry_t temp_entry; 2184 2185 switch (new_inheritance) { 2186 case VM_INHERIT_NONE: 2187 case VM_INHERIT_COPY: 2188 case VM_INHERIT_SHARE: 2189 break; 2190 default: 2191 return (KERN_INVALID_ARGUMENT); 2192 } 2193 vm_map_lock(map); 2194 VM_MAP_RANGE_CHECK(map, start, end); 2195 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2196 entry = temp_entry; 2197 vm_map_clip_start(map, entry, start); 2198 } else 2199 entry = temp_entry->next; 2200 while ((entry != &map->header) && (entry->start < end)) { 2201 vm_map_clip_end(map, entry, end); 2202 entry->inheritance = new_inheritance; 2203 vm_map_simplify_entry(map, entry); 2204 entry = entry->next; 2205 } 2206 vm_map_unlock(map); 2207 return (KERN_SUCCESS); 2208} 2209 2210/* 2211 * vm_map_unwire: 2212 * 2213 * Implements both kernel and user unwiring. 2214 */ 2215int 2216vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2217 int flags) 2218{ 2219 vm_map_entry_t entry, first_entry, tmp_entry; 2220 vm_offset_t saved_start; 2221 unsigned int last_timestamp; 2222 int rv; 2223 boolean_t need_wakeup, result, user_unwire; 2224 2225 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2226 vm_map_lock(map); 2227 VM_MAP_RANGE_CHECK(map, start, end); 2228 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2229 if (flags & VM_MAP_WIRE_HOLESOK) 2230 first_entry = first_entry->next; 2231 else { 2232 vm_map_unlock(map); 2233 return (KERN_INVALID_ADDRESS); 2234 } 2235 } 2236 last_timestamp = map->timestamp; 2237 entry = first_entry; 2238 while (entry != &map->header && entry->start < end) { 2239 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2240 /* 2241 * We have not yet clipped the entry. 2242 */ 2243 saved_start = (start >= entry->start) ? start : 2244 entry->start; 2245 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2246 if (vm_map_unlock_and_wait(map, 0)) { 2247 /* 2248 * Allow interruption of user unwiring? 2249 */ 2250 } 2251 vm_map_lock(map); 2252 if (last_timestamp+1 != map->timestamp) { 2253 /* 2254 * Look again for the entry because the map was 2255 * modified while it was unlocked. 2256 * Specifically, the entry may have been 2257 * clipped, merged, or deleted. 2258 */ 2259 if (!vm_map_lookup_entry(map, saved_start, 2260 &tmp_entry)) { 2261 if (flags & VM_MAP_WIRE_HOLESOK) 2262 tmp_entry = tmp_entry->next; 2263 else { 2264 if (saved_start == start) { 2265 /* 2266 * First_entry has been deleted. 2267 */ 2268 vm_map_unlock(map); 2269 return (KERN_INVALID_ADDRESS); 2270 } 2271 end = saved_start; 2272 rv = KERN_INVALID_ADDRESS; 2273 goto done; 2274 } 2275 } 2276 if (entry == first_entry) 2277 first_entry = tmp_entry; 2278 else 2279 first_entry = NULL; 2280 entry = tmp_entry; 2281 } 2282 last_timestamp = map->timestamp; 2283 continue; 2284 } 2285 vm_map_clip_start(map, entry, start); 2286 vm_map_clip_end(map, entry, end); 2287 /* 2288 * Mark the entry in case the map lock is released. (See 2289 * above.) 2290 */ 2291 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2292 entry->wiring_thread == NULL, 2293 ("owned map entry %p", entry)); 2294 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2295 entry->wiring_thread = curthread; 2296 /* 2297 * Check the map for holes in the specified region. 2298 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2299 */ 2300 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2301 (entry->end < end && (entry->next == &map->header || 2302 entry->next->start > entry->end))) { 2303 end = entry->end; 2304 rv = KERN_INVALID_ADDRESS; 2305 goto done; 2306 } 2307 /* 2308 * If system unwiring, require that the entry is system wired. 2309 */ 2310 if (!user_unwire && 2311 vm_map_entry_system_wired_count(entry) == 0) { 2312 end = entry->end; 2313 rv = KERN_INVALID_ARGUMENT; 2314 goto done; 2315 } 2316 entry = entry->next; 2317 } 2318 rv = KERN_SUCCESS; 2319done: 2320 need_wakeup = FALSE; 2321 if (first_entry == NULL) { 2322 result = vm_map_lookup_entry(map, start, &first_entry); 2323 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2324 first_entry = first_entry->next; 2325 else 2326 KASSERT(result, ("vm_map_unwire: lookup failed")); 2327 } 2328 for (entry = first_entry; entry != &map->header && entry->start < end; 2329 entry = entry->next) { 2330 /* 2331 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2332 * space in the unwired region could have been mapped 2333 * while the map lock was dropped for draining 2334 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2335 * could be simultaneously wiring this new mapping 2336 * entry. Detect these cases and skip any entries 2337 * marked as in transition by us. 2338 */ 2339 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2340 entry->wiring_thread != curthread) { 2341 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2342 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2343 continue; 2344 } 2345 2346 if (rv == KERN_SUCCESS && (!user_unwire || 2347 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2348 if (user_unwire) 2349 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2350 entry->wired_count--; 2351 if (entry->wired_count == 0) { 2352 /* 2353 * Retain the map lock. 2354 */ 2355 vm_fault_unwire(map, entry->start, entry->end, 2356 entry->object.vm_object != NULL && 2357 (entry->object.vm_object->flags & 2358 OBJ_FICTITIOUS) != 0); 2359 } 2360 } 2361 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2362 ("vm_map_unwire: in-transition flag missing %p", entry)); 2363 KASSERT(entry->wiring_thread == curthread, 2364 ("vm_map_unwire: alien wire %p", entry)); 2365 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2366 entry->wiring_thread = NULL; 2367 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2368 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2369 need_wakeup = TRUE; 2370 } 2371 vm_map_simplify_entry(map, entry); 2372 } 2373 vm_map_unlock(map); 2374 if (need_wakeup) 2375 vm_map_wakeup(map); 2376 return (rv); 2377} 2378 2379/* 2380 * vm_map_wire: 2381 * 2382 * Implements both kernel and user wiring. 2383 */ 2384int 2385vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2386 int flags) 2387{ 2388 vm_map_entry_t entry, first_entry, tmp_entry; 2389 vm_offset_t saved_end, saved_start; 2390 unsigned int last_timestamp; 2391 int rv; 2392 boolean_t fictitious, need_wakeup, result, user_wire; 2393 vm_prot_t prot; 2394 2395 prot = 0; 2396 if (flags & VM_MAP_WIRE_WRITE) 2397 prot |= VM_PROT_WRITE; 2398 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2399 vm_map_lock(map); 2400 VM_MAP_RANGE_CHECK(map, start, end); 2401 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2402 if (flags & VM_MAP_WIRE_HOLESOK) 2403 first_entry = first_entry->next; 2404 else { 2405 vm_map_unlock(map); 2406 return (KERN_INVALID_ADDRESS); 2407 } 2408 } 2409 last_timestamp = map->timestamp; 2410 entry = first_entry; 2411 while (entry != &map->header && entry->start < end) { 2412 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2413 /* 2414 * We have not yet clipped the entry. 2415 */ 2416 saved_start = (start >= entry->start) ? start : 2417 entry->start; 2418 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2419 if (vm_map_unlock_and_wait(map, 0)) { 2420 /* 2421 * Allow interruption of user wiring? 2422 */ 2423 } 2424 vm_map_lock(map); 2425 if (last_timestamp + 1 != map->timestamp) { 2426 /* 2427 * Look again for the entry because the map was 2428 * modified while it was unlocked. 2429 * Specifically, the entry may have been 2430 * clipped, merged, or deleted. 2431 */ 2432 if (!vm_map_lookup_entry(map, saved_start, 2433 &tmp_entry)) { 2434 if (flags & VM_MAP_WIRE_HOLESOK) 2435 tmp_entry = tmp_entry->next; 2436 else { 2437 if (saved_start == start) { 2438 /* 2439 * first_entry has been deleted. 2440 */ 2441 vm_map_unlock(map); 2442 return (KERN_INVALID_ADDRESS); 2443 } 2444 end = saved_start; 2445 rv = KERN_INVALID_ADDRESS; 2446 goto done; 2447 } 2448 } 2449 if (entry == first_entry) 2450 first_entry = tmp_entry; 2451 else 2452 first_entry = NULL; 2453 entry = tmp_entry; 2454 } 2455 last_timestamp = map->timestamp; 2456 continue; 2457 } 2458 vm_map_clip_start(map, entry, start); 2459 vm_map_clip_end(map, entry, end); 2460 /* 2461 * Mark the entry in case the map lock is released. (See 2462 * above.) 2463 */ 2464 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2465 entry->wiring_thread == NULL, 2466 ("owned map entry %p", entry)); 2467 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2468 entry->wiring_thread = curthread; 2469 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2470 || (entry->protection & prot) != prot) { 2471 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2472 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2473 end = entry->end; 2474 rv = KERN_INVALID_ADDRESS; 2475 goto done; 2476 } 2477 goto next_entry; 2478 } 2479 if (entry->wired_count == 0) { 2480 entry->wired_count++; 2481 saved_start = entry->start; 2482 saved_end = entry->end; 2483 fictitious = entry->object.vm_object != NULL && 2484 (entry->object.vm_object->flags & 2485 OBJ_FICTITIOUS) != 0; 2486 /* 2487 * Release the map lock, relying on the in-transition 2488 * mark. Mark the map busy for fork. 2489 */ 2490 vm_map_busy(map); 2491 vm_map_unlock(map); 2492 rv = vm_fault_wire(map, saved_start, saved_end, 2493 fictitious); 2494 vm_map_lock(map); 2495 vm_map_unbusy(map); 2496 if (last_timestamp + 1 != map->timestamp) { 2497 /* 2498 * Look again for the entry because the map was 2499 * modified while it was unlocked. The entry 2500 * may have been clipped, but NOT merged or 2501 * deleted. 2502 */ 2503 result = vm_map_lookup_entry(map, saved_start, 2504 &tmp_entry); 2505 KASSERT(result, ("vm_map_wire: lookup failed")); 2506 if (entry == first_entry) 2507 first_entry = tmp_entry; 2508 else 2509 first_entry = NULL; 2510 entry = tmp_entry; 2511 while (entry->end < saved_end) { 2512 if (rv != KERN_SUCCESS) { 2513 KASSERT(entry->wired_count == 1, 2514 ("vm_map_wire: bad count")); 2515 entry->wired_count = -1; 2516 } 2517 entry = entry->next; 2518 } 2519 } 2520 last_timestamp = map->timestamp; 2521 if (rv != KERN_SUCCESS) { 2522 KASSERT(entry->wired_count == 1, 2523 ("vm_map_wire: bad count")); 2524 /* 2525 * Assign an out-of-range value to represent 2526 * the failure to wire this entry. 2527 */ 2528 entry->wired_count = -1; 2529 end = entry->end; 2530 goto done; 2531 } 2532 } else if (!user_wire || 2533 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2534 entry->wired_count++; 2535 } 2536 /* 2537 * Check the map for holes in the specified region. 2538 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2539 */ 2540 next_entry: 2541 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2542 (entry->end < end && (entry->next == &map->header || 2543 entry->next->start > entry->end))) { 2544 end = entry->end; 2545 rv = KERN_INVALID_ADDRESS; 2546 goto done; 2547 } 2548 entry = entry->next; 2549 } 2550 rv = KERN_SUCCESS; 2551done: 2552 need_wakeup = FALSE; 2553 if (first_entry == NULL) { 2554 result = vm_map_lookup_entry(map, start, &first_entry); 2555 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2556 first_entry = first_entry->next; 2557 else 2558 KASSERT(result, ("vm_map_wire: lookup failed")); 2559 } 2560 for (entry = first_entry; entry != &map->header && entry->start < end; 2561 entry = entry->next) { 2562 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2563 goto next_entry_done; 2564 2565 /* 2566 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2567 * space in the unwired region could have been mapped 2568 * while the map lock was dropped for faulting in the 2569 * pages or draining MAP_ENTRY_IN_TRANSITION. 2570 * Moreover, another thread could be simultaneously 2571 * wiring this new mapping entry. Detect these cases 2572 * and skip any entries marked as in transition by us. 2573 */ 2574 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2575 entry->wiring_thread != curthread) { 2576 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2577 ("vm_map_wire: !HOLESOK and new/changed entry")); 2578 continue; 2579 } 2580 2581 if (rv == KERN_SUCCESS) { 2582 if (user_wire) 2583 entry->eflags |= MAP_ENTRY_USER_WIRED; 2584 } else if (entry->wired_count == -1) { 2585 /* 2586 * Wiring failed on this entry. Thus, unwiring is 2587 * unnecessary. 2588 */ 2589 entry->wired_count = 0; 2590 } else { 2591 if (!user_wire || 2592 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2593 entry->wired_count--; 2594 if (entry->wired_count == 0) { 2595 /* 2596 * Retain the map lock. 2597 */ 2598 vm_fault_unwire(map, entry->start, entry->end, 2599 entry->object.vm_object != NULL && 2600 (entry->object.vm_object->flags & 2601 OBJ_FICTITIOUS) != 0); 2602 } 2603 } 2604 next_entry_done: 2605 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2606 ("vm_map_wire: in-transition flag missing %p", entry)); 2607 KASSERT(entry->wiring_thread == curthread, 2608 ("vm_map_wire: alien wire %p", entry)); 2609 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2610 MAP_ENTRY_WIRE_SKIPPED); 2611 entry->wiring_thread = NULL; 2612 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2613 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2614 need_wakeup = TRUE; 2615 } 2616 vm_map_simplify_entry(map, entry); 2617 } 2618 vm_map_unlock(map); 2619 if (need_wakeup) 2620 vm_map_wakeup(map); 2621 return (rv); 2622} 2623 2624/* 2625 * vm_map_sync 2626 * 2627 * Push any dirty cached pages in the address range to their pager. 2628 * If syncio is TRUE, dirty pages are written synchronously. 2629 * If invalidate is TRUE, any cached pages are freed as well. 2630 * 2631 * If the size of the region from start to end is zero, we are 2632 * supposed to flush all modified pages within the region containing 2633 * start. Unfortunately, a region can be split or coalesced with 2634 * neighboring regions, making it difficult to determine what the 2635 * original region was. Therefore, we approximate this requirement by 2636 * flushing the current region containing start. 2637 * 2638 * Returns an error if any part of the specified range is not mapped. 2639 */ 2640int 2641vm_map_sync( 2642 vm_map_t map, 2643 vm_offset_t start, 2644 vm_offset_t end, 2645 boolean_t syncio, 2646 boolean_t invalidate) 2647{ 2648 vm_map_entry_t current; 2649 vm_map_entry_t entry; 2650 vm_size_t size; 2651 vm_object_t object; 2652 vm_ooffset_t offset; 2653 unsigned int last_timestamp; 2654 boolean_t failed; 2655 2656 vm_map_lock_read(map); 2657 VM_MAP_RANGE_CHECK(map, start, end); 2658 if (!vm_map_lookup_entry(map, start, &entry)) { 2659 vm_map_unlock_read(map); 2660 return (KERN_INVALID_ADDRESS); 2661 } else if (start == end) { 2662 start = entry->start; 2663 end = entry->end; 2664 } 2665 /* 2666 * Make a first pass to check for user-wired memory and holes. 2667 */ 2668 for (current = entry; current != &map->header && current->start < end; 2669 current = current->next) { 2670 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2671 vm_map_unlock_read(map); 2672 return (KERN_INVALID_ARGUMENT); 2673 } 2674 if (end > current->end && 2675 (current->next == &map->header || 2676 current->end != current->next->start)) { 2677 vm_map_unlock_read(map); 2678 return (KERN_INVALID_ADDRESS); 2679 } 2680 } 2681 2682 if (invalidate) 2683 pmap_remove(map->pmap, start, end); 2684 failed = FALSE; 2685 2686 /* 2687 * Make a second pass, cleaning/uncaching pages from the indicated 2688 * objects as we go. 2689 */ 2690 for (current = entry; current != &map->header && current->start < end;) { 2691 offset = current->offset + (start - current->start); 2692 size = (end <= current->end ? end : current->end) - start; 2693 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2694 vm_map_t smap; 2695 vm_map_entry_t tentry; 2696 vm_size_t tsize; 2697 2698 smap = current->object.sub_map; 2699 vm_map_lock_read(smap); 2700 (void) vm_map_lookup_entry(smap, offset, &tentry); 2701 tsize = tentry->end - offset; 2702 if (tsize < size) 2703 size = tsize; 2704 object = tentry->object.vm_object; 2705 offset = tentry->offset + (offset - tentry->start); 2706 vm_map_unlock_read(smap); 2707 } else { 2708 object = current->object.vm_object; 2709 } 2710 vm_object_reference(object); 2711 last_timestamp = map->timestamp; 2712 vm_map_unlock_read(map); 2713 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2714 failed = TRUE; 2715 start += size; 2716 vm_object_deallocate(object); 2717 vm_map_lock_read(map); 2718 if (last_timestamp == map->timestamp || 2719 !vm_map_lookup_entry(map, start, ¤t)) 2720 current = current->next; 2721 } 2722 2723 vm_map_unlock_read(map); 2724 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2725} 2726 2727/* 2728 * vm_map_entry_unwire: [ internal use only ] 2729 * 2730 * Make the region specified by this entry pageable. 2731 * 2732 * The map in question should be locked. 2733 * [This is the reason for this routine's existence.] 2734 */ 2735static void 2736vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2737{ 2738 vm_fault_unwire(map, entry->start, entry->end, 2739 entry->object.vm_object != NULL && 2740 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2741 entry->wired_count = 0; 2742} 2743 2744static void 2745vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2746{ 2747 2748 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2749 vm_object_deallocate(entry->object.vm_object); 2750 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2751} 2752 2753/* 2754 * vm_map_entry_delete: [ internal use only ] 2755 * 2756 * Deallocate the given entry from the target map. 2757 */ 2758static void 2759vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2760{ 2761 vm_object_t object; 2762 vm_pindex_t offidxstart, offidxend, count, size1; 2763 vm_ooffset_t size; 2764 2765 vm_map_entry_unlink(map, entry); 2766 object = entry->object.vm_object; 2767 size = entry->end - entry->start; 2768 map->size -= size; 2769 2770 if (entry->cred != NULL) { 2771 swap_release_by_cred(size, entry->cred); 2772 crfree(entry->cred); 2773 } 2774 2775 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2776 (object != NULL)) { 2777 KASSERT(entry->cred == NULL || object->cred == NULL || 2778 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2779 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2780 count = OFF_TO_IDX(size); 2781 offidxstart = OFF_TO_IDX(entry->offset); 2782 offidxend = offidxstart + count; 2783 VM_OBJECT_WLOCK(object); 2784 if (object->ref_count != 1 && 2785 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2786 object == kernel_object || object == kmem_object)) { 2787 vm_object_collapse(object); 2788 2789 /* 2790 * The option OBJPR_NOTMAPPED can be passed here 2791 * because vm_map_delete() already performed 2792 * pmap_remove() on the only mapping to this range 2793 * of pages. 2794 */ 2795 vm_object_page_remove(object, offidxstart, offidxend, 2796 OBJPR_NOTMAPPED); 2797 if (object->type == OBJT_SWAP) 2798 swap_pager_freespace(object, offidxstart, count); 2799 if (offidxend >= object->size && 2800 offidxstart < object->size) { 2801 size1 = object->size; 2802 object->size = offidxstart; 2803 if (object->cred != NULL) { 2804 size1 -= object->size; 2805 KASSERT(object->charge >= ptoa(size1), 2806 ("vm_map_entry_delete: object->charge < 0")); 2807 swap_release_by_cred(ptoa(size1), object->cred); 2808 object->charge -= ptoa(size1); 2809 } 2810 } 2811 } 2812 VM_OBJECT_WUNLOCK(object); 2813 } else 2814 entry->object.vm_object = NULL; 2815 if (map->system_map) 2816 vm_map_entry_deallocate(entry, TRUE); 2817 else { 2818 entry->next = curthread->td_map_def_user; 2819 curthread->td_map_def_user = entry; 2820 } 2821} 2822 2823/* 2824 * vm_map_delete: [ internal use only ] 2825 * 2826 * Deallocates the given address range from the target 2827 * map. 2828 */ 2829int 2830vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2831{ 2832 vm_map_entry_t entry; 2833 vm_map_entry_t first_entry; 2834 2835 VM_MAP_ASSERT_LOCKED(map); 2836 2837 /* 2838 * Find the start of the region, and clip it 2839 */ 2840 if (!vm_map_lookup_entry(map, start, &first_entry)) 2841 entry = first_entry->next; 2842 else { 2843 entry = first_entry; 2844 vm_map_clip_start(map, entry, start); 2845 } 2846 2847 /* 2848 * Step through all entries in this region 2849 */ 2850 while ((entry != &map->header) && (entry->start < end)) { 2851 vm_map_entry_t next; 2852 2853 /* 2854 * Wait for wiring or unwiring of an entry to complete. 2855 * Also wait for any system wirings to disappear on 2856 * user maps. 2857 */ 2858 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2859 (vm_map_pmap(map) != kernel_pmap && 2860 vm_map_entry_system_wired_count(entry) != 0)) { 2861 unsigned int last_timestamp; 2862 vm_offset_t saved_start; 2863 vm_map_entry_t tmp_entry; 2864 2865 saved_start = entry->start; 2866 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2867 last_timestamp = map->timestamp; 2868 (void) vm_map_unlock_and_wait(map, 0); 2869 vm_map_lock(map); 2870 if (last_timestamp + 1 != map->timestamp) { 2871 /* 2872 * Look again for the entry because the map was 2873 * modified while it was unlocked. 2874 * Specifically, the entry may have been 2875 * clipped, merged, or deleted. 2876 */ 2877 if (!vm_map_lookup_entry(map, saved_start, 2878 &tmp_entry)) 2879 entry = tmp_entry->next; 2880 else { 2881 entry = tmp_entry; 2882 vm_map_clip_start(map, entry, 2883 saved_start); 2884 } 2885 } 2886 continue; 2887 } 2888 vm_map_clip_end(map, entry, end); 2889 2890 next = entry->next; 2891 2892 /* 2893 * Unwire before removing addresses from the pmap; otherwise, 2894 * unwiring will put the entries back in the pmap. 2895 */ 2896 if (entry->wired_count != 0) { 2897 vm_map_entry_unwire(map, entry); 2898 } 2899 2900 pmap_remove(map->pmap, entry->start, entry->end); 2901 2902 /* 2903 * Delete the entry only after removing all pmap 2904 * entries pointing to its pages. (Otherwise, its 2905 * page frames may be reallocated, and any modify bits 2906 * will be set in the wrong object!) 2907 */ 2908 vm_map_entry_delete(map, entry); 2909 entry = next; 2910 } 2911 return (KERN_SUCCESS); 2912} 2913 2914/* 2915 * vm_map_remove: 2916 * 2917 * Remove the given address range from the target map. 2918 * This is the exported form of vm_map_delete. 2919 */ 2920int 2921vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2922{ 2923 int result; 2924 2925 vm_map_lock(map); 2926 VM_MAP_RANGE_CHECK(map, start, end); 2927 result = vm_map_delete(map, start, end); 2928 vm_map_unlock(map); 2929 return (result); 2930} 2931 2932/* 2933 * vm_map_check_protection: 2934 * 2935 * Assert that the target map allows the specified privilege on the 2936 * entire address region given. The entire region must be allocated. 2937 * 2938 * WARNING! This code does not and should not check whether the 2939 * contents of the region is accessible. For example a smaller file 2940 * might be mapped into a larger address space. 2941 * 2942 * NOTE! This code is also called by munmap(). 2943 * 2944 * The map must be locked. A read lock is sufficient. 2945 */ 2946boolean_t 2947vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2948 vm_prot_t protection) 2949{ 2950 vm_map_entry_t entry; 2951 vm_map_entry_t tmp_entry; 2952 2953 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2954 return (FALSE); 2955 entry = tmp_entry; 2956 2957 while (start < end) { 2958 if (entry == &map->header) 2959 return (FALSE); 2960 /* 2961 * No holes allowed! 2962 */ 2963 if (start < entry->start) 2964 return (FALSE); 2965 /* 2966 * Check protection associated with entry. 2967 */ 2968 if ((entry->protection & protection) != protection) 2969 return (FALSE); 2970 /* go to next entry */ 2971 start = entry->end; 2972 entry = entry->next; 2973 } 2974 return (TRUE); 2975} 2976 2977/* 2978 * vm_map_copy_entry: 2979 * 2980 * Copies the contents of the source entry to the destination 2981 * entry. The entries *must* be aligned properly. 2982 */ 2983static void 2984vm_map_copy_entry( 2985 vm_map_t src_map, 2986 vm_map_t dst_map, 2987 vm_map_entry_t src_entry, 2988 vm_map_entry_t dst_entry, 2989 vm_ooffset_t *fork_charge) 2990{ 2991 vm_object_t src_object; 2992 vm_map_entry_t fake_entry; 2993 vm_offset_t size; 2994 struct ucred *cred; 2995 int charged; 2996 2997 VM_MAP_ASSERT_LOCKED(dst_map); 2998 2999 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3000 return; 3001 3002 if (src_entry->wired_count == 0) { 3003 3004 /* 3005 * If the source entry is marked needs_copy, it is already 3006 * write-protected. 3007 */ 3008 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 3009 pmap_protect(src_map->pmap, 3010 src_entry->start, 3011 src_entry->end, 3012 src_entry->protection & ~VM_PROT_WRITE); 3013 } 3014 3015 /* 3016 * Make a copy of the object. 3017 */ 3018 size = src_entry->end - src_entry->start; 3019 if ((src_object = src_entry->object.vm_object) != NULL) { 3020 VM_OBJECT_WLOCK(src_object); 3021 charged = ENTRY_CHARGED(src_entry); 3022 if ((src_object->handle == NULL) && 3023 (src_object->type == OBJT_DEFAULT || 3024 src_object->type == OBJT_SWAP)) { 3025 vm_object_collapse(src_object); 3026 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3027 vm_object_split(src_entry); 3028 src_object = src_entry->object.vm_object; 3029 } 3030 } 3031 vm_object_reference_locked(src_object); 3032 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3033 if (src_entry->cred != NULL && 3034 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3035 KASSERT(src_object->cred == NULL, 3036 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3037 src_object)); 3038 src_object->cred = src_entry->cred; 3039 src_object->charge = size; 3040 } 3041 VM_OBJECT_WUNLOCK(src_object); 3042 dst_entry->object.vm_object = src_object; 3043 if (charged) { 3044 cred = curthread->td_ucred; 3045 crhold(cred); 3046 dst_entry->cred = cred; 3047 *fork_charge += size; 3048 if (!(src_entry->eflags & 3049 MAP_ENTRY_NEEDS_COPY)) { 3050 crhold(cred); 3051 src_entry->cred = cred; 3052 *fork_charge += size; 3053 } 3054 } 3055 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3056 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3057 dst_entry->offset = src_entry->offset; 3058 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3059 /* 3060 * MAP_ENTRY_VN_WRITECNT cannot 3061 * indicate write reference from 3062 * src_entry, since the entry is 3063 * marked as needs copy. Allocate a 3064 * fake entry that is used to 3065 * decrement object->un_pager.vnp.writecount 3066 * at the appropriate time. Attach 3067 * fake_entry to the deferred list. 3068 */ 3069 fake_entry = vm_map_entry_create(dst_map); 3070 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3071 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3072 vm_object_reference(src_object); 3073 fake_entry->object.vm_object = src_object; 3074 fake_entry->start = src_entry->start; 3075 fake_entry->end = src_entry->end; 3076 fake_entry->next = curthread->td_map_def_user; 3077 curthread->td_map_def_user = fake_entry; 3078 } 3079 } else { 3080 dst_entry->object.vm_object = NULL; 3081 dst_entry->offset = 0; 3082 if (src_entry->cred != NULL) { 3083 dst_entry->cred = curthread->td_ucred; 3084 crhold(dst_entry->cred); 3085 *fork_charge += size; 3086 } 3087 } 3088 3089 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3090 dst_entry->end - dst_entry->start, src_entry->start); 3091 } else { 3092 /* 3093 * Of course, wired down pages can't be set copy-on-write. 3094 * Cause wired pages to be copied into the new map by 3095 * simulating faults (the new pages are pageable) 3096 */ 3097 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3098 fork_charge); 3099 } 3100} 3101 3102/* 3103 * vmspace_map_entry_forked: 3104 * Update the newly-forked vmspace each time a map entry is inherited 3105 * or copied. The values for vm_dsize and vm_tsize are approximate 3106 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3107 */ 3108static void 3109vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3110 vm_map_entry_t entry) 3111{ 3112 vm_size_t entrysize; 3113 vm_offset_t newend; 3114 3115 entrysize = entry->end - entry->start; 3116 vm2->vm_map.size += entrysize; 3117 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3118 vm2->vm_ssize += btoc(entrysize); 3119 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3120 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3121 newend = MIN(entry->end, 3122 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3123 vm2->vm_dsize += btoc(newend - entry->start); 3124 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3125 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3126 newend = MIN(entry->end, 3127 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3128 vm2->vm_tsize += btoc(newend - entry->start); 3129 } 3130} 3131 3132/* 3133 * vmspace_fork: 3134 * Create a new process vmspace structure and vm_map 3135 * based on those of an existing process. The new map 3136 * is based on the old map, according to the inheritance 3137 * values on the regions in that map. 3138 * 3139 * XXX It might be worth coalescing the entries added to the new vmspace. 3140 * 3141 * The source map must not be locked. 3142 */ 3143struct vmspace * 3144vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3145{ 3146 struct vmspace *vm2; 3147 vm_map_t new_map, old_map; 3148 vm_map_entry_t new_entry, old_entry; 3149 vm_object_t object; 3150 int locked; 3151 3152 old_map = &vm1->vm_map; 3153 /* Copy immutable fields of vm1 to vm2. */ 3154 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3155 if (vm2 == NULL) 3156 return (NULL); 3157 vm2->vm_taddr = vm1->vm_taddr; 3158 vm2->vm_daddr = vm1->vm_daddr; 3159 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3160 vm_map_lock(old_map); 3161 if (old_map->busy) 3162 vm_map_wait_busy(old_map); 3163 new_map = &vm2->vm_map; 3164 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3165 KASSERT(locked, ("vmspace_fork: lock failed")); 3166 3167 old_entry = old_map->header.next; 3168 3169 while (old_entry != &old_map->header) { 3170 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3171 panic("vm_map_fork: encountered a submap"); 3172 3173 switch (old_entry->inheritance) { 3174 case VM_INHERIT_NONE: 3175 break; 3176 3177 case VM_INHERIT_SHARE: 3178 /* 3179 * Clone the entry, creating the shared object if necessary. 3180 */ 3181 object = old_entry->object.vm_object; 3182 if (object == NULL) { 3183 object = vm_object_allocate(OBJT_DEFAULT, 3184 atop(old_entry->end - old_entry->start)); 3185 old_entry->object.vm_object = object; 3186 old_entry->offset = 0; 3187 if (old_entry->cred != NULL) { 3188 object->cred = old_entry->cred; 3189 object->charge = old_entry->end - 3190 old_entry->start; 3191 old_entry->cred = NULL; 3192 } 3193 } 3194 3195 /* 3196 * Add the reference before calling vm_object_shadow 3197 * to insure that a shadow object is created. 3198 */ 3199 vm_object_reference(object); 3200 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3201 vm_object_shadow(&old_entry->object.vm_object, 3202 &old_entry->offset, 3203 old_entry->end - old_entry->start); 3204 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3205 /* Transfer the second reference too. */ 3206 vm_object_reference( 3207 old_entry->object.vm_object); 3208 3209 /* 3210 * As in vm_map_simplify_entry(), the 3211 * vnode lock will not be acquired in 3212 * this call to vm_object_deallocate(). 3213 */ 3214 vm_object_deallocate(object); 3215 object = old_entry->object.vm_object; 3216 } 3217 VM_OBJECT_WLOCK(object); 3218 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3219 if (old_entry->cred != NULL) { 3220 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3221 object->cred = old_entry->cred; 3222 object->charge = old_entry->end - old_entry->start; 3223 old_entry->cred = NULL; 3224 } 3225 3226 /* 3227 * Assert the correct state of the vnode 3228 * v_writecount while the object is locked, to 3229 * not relock it later for the assertion 3230 * correctness. 3231 */ 3232 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3233 object->type == OBJT_VNODE) { 3234 KASSERT(((struct vnode *)object->handle)-> 3235 v_writecount > 0, 3236 ("vmspace_fork: v_writecount %p", object)); 3237 KASSERT(object->un_pager.vnp.writemappings > 0, 3238 ("vmspace_fork: vnp.writecount %p", 3239 object)); 3240 } 3241 VM_OBJECT_WUNLOCK(object); 3242 3243 /* 3244 * Clone the entry, referencing the shared object. 3245 */ 3246 new_entry = vm_map_entry_create(new_map); 3247 *new_entry = *old_entry; 3248 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3249 MAP_ENTRY_IN_TRANSITION); 3250 new_entry->wiring_thread = NULL; 3251 new_entry->wired_count = 0; 3252 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3253 vnode_pager_update_writecount(object, 3254 new_entry->start, new_entry->end); 3255 } 3256 3257 /* 3258 * Insert the entry into the new map -- we know we're 3259 * inserting at the end of the new map. 3260 */ 3261 vm_map_entry_link(new_map, new_map->header.prev, 3262 new_entry); 3263 vmspace_map_entry_forked(vm1, vm2, new_entry); 3264 3265 /* 3266 * Update the physical map 3267 */ 3268 pmap_copy(new_map->pmap, old_map->pmap, 3269 new_entry->start, 3270 (old_entry->end - old_entry->start), 3271 old_entry->start); 3272 break; 3273 3274 case VM_INHERIT_COPY: 3275 /* 3276 * Clone the entry and link into the map. 3277 */ 3278 new_entry = vm_map_entry_create(new_map); 3279 *new_entry = *old_entry; 3280 /* 3281 * Copied entry is COW over the old object. 3282 */ 3283 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3284 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3285 new_entry->wiring_thread = NULL; 3286 new_entry->wired_count = 0; 3287 new_entry->object.vm_object = NULL; 3288 new_entry->cred = NULL; 3289 vm_map_entry_link(new_map, new_map->header.prev, 3290 new_entry); 3291 vmspace_map_entry_forked(vm1, vm2, new_entry); 3292 vm_map_copy_entry(old_map, new_map, old_entry, 3293 new_entry, fork_charge); 3294 break; 3295 } 3296 old_entry = old_entry->next; 3297 } 3298 /* 3299 * Use inlined vm_map_unlock() to postpone handling the deferred 3300 * map entries, which cannot be done until both old_map and 3301 * new_map locks are released. 3302 */ 3303 sx_xunlock(&old_map->lock); 3304 sx_xunlock(&new_map->lock); 3305 vm_map_process_deferred(); 3306 3307 return (vm2); 3308} 3309 3310int 3311vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3312 vm_prot_t prot, vm_prot_t max, int cow) 3313{ 3314 vm_map_entry_t new_entry, prev_entry; 3315 vm_offset_t bot, top; 3316 vm_size_t growsize, init_ssize; 3317 int orient, rv; 3318 rlim_t lmemlim, vmemlim; 3319 3320 /* 3321 * The stack orientation is piggybacked with the cow argument. 3322 * Extract it into orient and mask the cow argument so that we 3323 * don't pass it around further. 3324 * NOTE: We explicitly allow bi-directional stacks. 3325 */ 3326 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3327 cow &= ~orient; 3328 KASSERT(orient != 0, ("No stack grow direction")); 3329 3330 if (addrbos < vm_map_min(map) || 3331 addrbos > vm_map_max(map) || 3332 addrbos + max_ssize < addrbos) 3333 return (KERN_NO_SPACE); 3334 3335 growsize = sgrowsiz; 3336 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3337 3338 PROC_LOCK(curproc); 3339 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3340 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3341 PROC_UNLOCK(curproc); 3342 3343 vm_map_lock(map); 3344 3345 /* If addr is already mapped, no go */ 3346 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3347 vm_map_unlock(map); 3348 return (KERN_NO_SPACE); 3349 } 3350 3351 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3352 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3353 vm_map_unlock(map); 3354 return (KERN_NO_SPACE); 3355 } 3356 } 3357 3358 /* If we would blow our VMEM resource limit, no go */ 3359 if (map->size + init_ssize > vmemlim) { 3360 vm_map_unlock(map); 3361 return (KERN_NO_SPACE); 3362 } 3363 3364 /* 3365 * If we can't accomodate max_ssize in the current mapping, no go. 3366 * However, we need to be aware that subsequent user mappings might 3367 * map into the space we have reserved for stack, and currently this 3368 * space is not protected. 3369 * 3370 * Hopefully we will at least detect this condition when we try to 3371 * grow the stack. 3372 */ 3373 if ((prev_entry->next != &map->header) && 3374 (prev_entry->next->start < addrbos + max_ssize)) { 3375 vm_map_unlock(map); 3376 return (KERN_NO_SPACE); 3377 } 3378 3379 /* 3380 * We initially map a stack of only init_ssize. We will grow as 3381 * needed later. Depending on the orientation of the stack (i.e. 3382 * the grow direction) we either map at the top of the range, the 3383 * bottom of the range or in the middle. 3384 * 3385 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3386 * and cow to be 0. Possibly we should eliminate these as input 3387 * parameters, and just pass these values here in the insert call. 3388 */ 3389 if (orient == MAP_STACK_GROWS_DOWN) 3390 bot = addrbos + max_ssize - init_ssize; 3391 else if (orient == MAP_STACK_GROWS_UP) 3392 bot = addrbos; 3393 else 3394 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3395 top = bot + init_ssize; 3396 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3397 3398 /* Now set the avail_ssize amount. */ 3399 if (rv == KERN_SUCCESS) { 3400 if (prev_entry != &map->header) 3401 vm_map_clip_end(map, prev_entry, bot); 3402 new_entry = prev_entry->next; 3403 if (new_entry->end != top || new_entry->start != bot) 3404 panic("Bad entry start/end for new stack entry"); 3405 3406 new_entry->avail_ssize = max_ssize - init_ssize; 3407 if (orient & MAP_STACK_GROWS_DOWN) 3408 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3409 if (orient & MAP_STACK_GROWS_UP) 3410 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3411 } 3412 3413 vm_map_unlock(map); 3414 return (rv); 3415} 3416 3417static int stack_guard_page = 0; 3418TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3419SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3420 &stack_guard_page, 0, 3421 "Insert stack guard page ahead of the growable segments."); 3422 3423/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3424 * desired address is already mapped, or if we successfully grow 3425 * the stack. Also returns KERN_SUCCESS if addr is outside the 3426 * stack range (this is strange, but preserves compatibility with 3427 * the grow function in vm_machdep.c). 3428 */ 3429int 3430vm_map_growstack(struct proc *p, vm_offset_t addr) 3431{ 3432 vm_map_entry_t next_entry, prev_entry; 3433 vm_map_entry_t new_entry, stack_entry; 3434 struct vmspace *vm = p->p_vmspace; 3435 vm_map_t map = &vm->vm_map; 3436 vm_offset_t end; 3437 vm_size_t growsize; 3438 size_t grow_amount, max_grow; 3439 rlim_t lmemlim, stacklim, vmemlim; 3440 int is_procstack, rv; 3441 struct ucred *cred; 3442#ifdef notyet 3443 uint64_t limit; 3444#endif 3445#ifdef RACCT 3446 int error; 3447#endif 3448 3449Retry: 3450 PROC_LOCK(p); 3451 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3452 stacklim = lim_cur(p, RLIMIT_STACK); 3453 vmemlim = lim_cur(p, RLIMIT_VMEM); 3454 PROC_UNLOCK(p); 3455 3456 vm_map_lock_read(map); 3457 3458 /* If addr is already in the entry range, no need to grow.*/ 3459 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3460 vm_map_unlock_read(map); 3461 return (KERN_SUCCESS); 3462 } 3463 3464 next_entry = prev_entry->next; 3465 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3466 /* 3467 * This entry does not grow upwards. Since the address lies 3468 * beyond this entry, the next entry (if one exists) has to 3469 * be a downward growable entry. The entry list header is 3470 * never a growable entry, so it suffices to check the flags. 3471 */ 3472 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3473 vm_map_unlock_read(map); 3474 return (KERN_SUCCESS); 3475 } 3476 stack_entry = next_entry; 3477 } else { 3478 /* 3479 * This entry grows upward. If the next entry does not at 3480 * least grow downwards, this is the entry we need to grow. 3481 * otherwise we have two possible choices and we have to 3482 * select one. 3483 */ 3484 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3485 /* 3486 * We have two choices; grow the entry closest to 3487 * the address to minimize the amount of growth. 3488 */ 3489 if (addr - prev_entry->end <= next_entry->start - addr) 3490 stack_entry = prev_entry; 3491 else 3492 stack_entry = next_entry; 3493 } else 3494 stack_entry = prev_entry; 3495 } 3496 3497 if (stack_entry == next_entry) { 3498 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3499 KASSERT(addr < stack_entry->start, ("foo")); 3500 end = (prev_entry != &map->header) ? prev_entry->end : 3501 stack_entry->start - stack_entry->avail_ssize; 3502 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3503 max_grow = stack_entry->start - end; 3504 } else { 3505 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3506 KASSERT(addr >= stack_entry->end, ("foo")); 3507 end = (next_entry != &map->header) ? next_entry->start : 3508 stack_entry->end + stack_entry->avail_ssize; 3509 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3510 max_grow = end - stack_entry->end; 3511 } 3512 3513 if (grow_amount > stack_entry->avail_ssize) { 3514 vm_map_unlock_read(map); 3515 return (KERN_NO_SPACE); 3516 } 3517 3518 /* 3519 * If there is no longer enough space between the entries nogo, and 3520 * adjust the available space. Note: this should only happen if the 3521 * user has mapped into the stack area after the stack was created, 3522 * and is probably an error. 3523 * 3524 * This also effectively destroys any guard page the user might have 3525 * intended by limiting the stack size. 3526 */ 3527 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3528 if (vm_map_lock_upgrade(map)) 3529 goto Retry; 3530 3531 stack_entry->avail_ssize = max_grow; 3532 3533 vm_map_unlock(map); 3534 return (KERN_NO_SPACE); 3535 } 3536 3537 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3538 3539 /* 3540 * If this is the main process stack, see if we're over the stack 3541 * limit. 3542 */ 3543 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3544 vm_map_unlock_read(map); 3545 return (KERN_NO_SPACE); 3546 } 3547#ifdef RACCT 3548 PROC_LOCK(p); 3549 if (is_procstack && 3550 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3551 PROC_UNLOCK(p); 3552 vm_map_unlock_read(map); 3553 return (KERN_NO_SPACE); 3554 } 3555 PROC_UNLOCK(p); 3556#endif 3557 3558 /* Round up the grow amount modulo sgrowsiz */ 3559 growsize = sgrowsiz; 3560 grow_amount = roundup(grow_amount, growsize); 3561 if (grow_amount > stack_entry->avail_ssize) 3562 grow_amount = stack_entry->avail_ssize; 3563 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3564 grow_amount = trunc_page((vm_size_t)stacklim) - 3565 ctob(vm->vm_ssize); 3566 } 3567#ifdef notyet 3568 PROC_LOCK(p); 3569 limit = racct_get_available(p, RACCT_STACK); 3570 PROC_UNLOCK(p); 3571 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3572 grow_amount = limit - ctob(vm->vm_ssize); 3573#endif 3574 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3575 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3576 vm_map_unlock_read(map); 3577 rv = KERN_NO_SPACE; 3578 goto out; 3579 } 3580#ifdef RACCT 3581 PROC_LOCK(p); 3582 if (racct_set(p, RACCT_MEMLOCK, 3583 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3584 PROC_UNLOCK(p); 3585 vm_map_unlock_read(map); 3586 rv = KERN_NO_SPACE; 3587 goto out; 3588 } 3589 PROC_UNLOCK(p); 3590#endif 3591 } 3592 /* If we would blow our VMEM resource limit, no go */ 3593 if (map->size + grow_amount > vmemlim) { 3594 vm_map_unlock_read(map); 3595 rv = KERN_NO_SPACE; 3596 goto out; 3597 } 3598#ifdef RACCT 3599 PROC_LOCK(p); 3600 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3601 PROC_UNLOCK(p); 3602 vm_map_unlock_read(map); 3603 rv = KERN_NO_SPACE; 3604 goto out; 3605 } 3606 PROC_UNLOCK(p); 3607#endif 3608 3609 if (vm_map_lock_upgrade(map)) 3610 goto Retry; 3611 3612 if (stack_entry == next_entry) { 3613 /* 3614 * Growing downward. 3615 */ 3616 /* Get the preliminary new entry start value */ 3617 addr = stack_entry->start - grow_amount; 3618 3619 /* 3620 * If this puts us into the previous entry, cut back our 3621 * growth to the available space. Also, see the note above. 3622 */ 3623 if (addr < end) { 3624 stack_entry->avail_ssize = max_grow; 3625 addr = end; 3626 if (stack_guard_page) 3627 addr += PAGE_SIZE; 3628 } 3629 3630 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3631 next_entry->protection, next_entry->max_protection, 0); 3632 3633 /* Adjust the available stack space by the amount we grew. */ 3634 if (rv == KERN_SUCCESS) { 3635 if (prev_entry != &map->header) 3636 vm_map_clip_end(map, prev_entry, addr); 3637 new_entry = prev_entry->next; 3638 KASSERT(new_entry == stack_entry->prev, ("foo")); 3639 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3640 KASSERT(new_entry->start == addr, ("foo")); 3641 grow_amount = new_entry->end - new_entry->start; 3642 new_entry->avail_ssize = stack_entry->avail_ssize - 3643 grow_amount; 3644 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3645 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3646 } 3647 } else { 3648 /* 3649 * Growing upward. 3650 */ 3651 addr = stack_entry->end + grow_amount; 3652 3653 /* 3654 * If this puts us into the next entry, cut back our growth 3655 * to the available space. Also, see the note above. 3656 */ 3657 if (addr > end) { 3658 stack_entry->avail_ssize = end - stack_entry->end; 3659 addr = end; 3660 if (stack_guard_page) 3661 addr -= PAGE_SIZE; 3662 } 3663 3664 grow_amount = addr - stack_entry->end; 3665 cred = stack_entry->cred; 3666 if (cred == NULL && stack_entry->object.vm_object != NULL) 3667 cred = stack_entry->object.vm_object->cred; 3668 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3669 rv = KERN_NO_SPACE; 3670 /* Grow the underlying object if applicable. */ 3671 else if (stack_entry->object.vm_object == NULL || 3672 vm_object_coalesce(stack_entry->object.vm_object, 3673 stack_entry->offset, 3674 (vm_size_t)(stack_entry->end - stack_entry->start), 3675 (vm_size_t)grow_amount, cred != NULL)) { 3676 map->size += (addr - stack_entry->end); 3677 /* Update the current entry. */ 3678 stack_entry->end = addr; 3679 stack_entry->avail_ssize -= grow_amount; 3680 vm_map_entry_resize_free(map, stack_entry); 3681 rv = KERN_SUCCESS; 3682 3683 if (next_entry != &map->header) 3684 vm_map_clip_start(map, next_entry, addr); 3685 } else 3686 rv = KERN_FAILURE; 3687 } 3688 3689 if (rv == KERN_SUCCESS && is_procstack) 3690 vm->vm_ssize += btoc(grow_amount); 3691 3692 vm_map_unlock(map); 3693 3694 /* 3695 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3696 */ 3697 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3698 vm_map_wire(map, 3699 (stack_entry == next_entry) ? addr : addr - grow_amount, 3700 (stack_entry == next_entry) ? stack_entry->start : addr, 3701 (p->p_flag & P_SYSTEM) 3702 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3703 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3704 } 3705 3706out: 3707#ifdef RACCT 3708 if (rv != KERN_SUCCESS) { 3709 PROC_LOCK(p); 3710 error = racct_set(p, RACCT_VMEM, map->size); 3711 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3712 if (!old_mlock) { 3713 error = racct_set(p, RACCT_MEMLOCK, 3714 ptoa(pmap_wired_count(map->pmap))); 3715 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3716 } 3717 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3718 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3719 PROC_UNLOCK(p); 3720 } 3721#endif 3722 3723 return (rv); 3724} 3725 3726/* 3727 * Unshare the specified VM space for exec. If other processes are 3728 * mapped to it, then create a new one. The new vmspace is null. 3729 */ 3730int 3731vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3732{ 3733 struct vmspace *oldvmspace = p->p_vmspace; 3734 struct vmspace *newvmspace; 3735 3736 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3737 if (newvmspace == NULL) 3738 return (ENOMEM); 3739 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3740 /* 3741 * This code is written like this for prototype purposes. The 3742 * goal is to avoid running down the vmspace here, but let the 3743 * other process's that are still using the vmspace to finally 3744 * run it down. Even though there is little or no chance of blocking 3745 * here, it is a good idea to keep this form for future mods. 3746 */ 3747 PROC_VMSPACE_LOCK(p); 3748 p->p_vmspace = newvmspace; 3749 PROC_VMSPACE_UNLOCK(p); 3750 if (p == curthread->td_proc) 3751 pmap_activate(curthread); 3752 vmspace_free(oldvmspace); 3753 return (0); 3754} 3755 3756/* 3757 * Unshare the specified VM space for forcing COW. This 3758 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3759 */ 3760int 3761vmspace_unshare(struct proc *p) 3762{ 3763 struct vmspace *oldvmspace = p->p_vmspace; 3764 struct vmspace *newvmspace; 3765 vm_ooffset_t fork_charge; 3766 3767 if (oldvmspace->vm_refcnt == 1) 3768 return (0); 3769 fork_charge = 0; 3770 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3771 if (newvmspace == NULL) 3772 return (ENOMEM); 3773 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3774 vmspace_free(newvmspace); 3775 return (ENOMEM); 3776 } 3777 PROC_VMSPACE_LOCK(p); 3778 p->p_vmspace = newvmspace; 3779 PROC_VMSPACE_UNLOCK(p); 3780 if (p == curthread->td_proc) 3781 pmap_activate(curthread); 3782 vmspace_free(oldvmspace); 3783 return (0); 3784} 3785 3786/* 3787 * vm_map_lookup: 3788 * 3789 * Finds the VM object, offset, and 3790 * protection for a given virtual address in the 3791 * specified map, assuming a page fault of the 3792 * type specified. 3793 * 3794 * Leaves the map in question locked for read; return 3795 * values are guaranteed until a vm_map_lookup_done 3796 * call is performed. Note that the map argument 3797 * is in/out; the returned map must be used in 3798 * the call to vm_map_lookup_done. 3799 * 3800 * A handle (out_entry) is returned for use in 3801 * vm_map_lookup_done, to make that fast. 3802 * 3803 * If a lookup is requested with "write protection" 3804 * specified, the map may be changed to perform virtual 3805 * copying operations, although the data referenced will 3806 * remain the same. 3807 */ 3808int 3809vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3810 vm_offset_t vaddr, 3811 vm_prot_t fault_typea, 3812 vm_map_entry_t *out_entry, /* OUT */ 3813 vm_object_t *object, /* OUT */ 3814 vm_pindex_t *pindex, /* OUT */ 3815 vm_prot_t *out_prot, /* OUT */ 3816 boolean_t *wired) /* OUT */ 3817{ 3818 vm_map_entry_t entry; 3819 vm_map_t map = *var_map; 3820 vm_prot_t prot; 3821 vm_prot_t fault_type = fault_typea; 3822 vm_object_t eobject; 3823 vm_size_t size; 3824 struct ucred *cred; 3825 3826RetryLookup:; 3827 3828 vm_map_lock_read(map); 3829 3830 /* 3831 * Lookup the faulting address. 3832 */ 3833 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3834 vm_map_unlock_read(map); 3835 return (KERN_INVALID_ADDRESS); 3836 } 3837 3838 entry = *out_entry; 3839 3840 /* 3841 * Handle submaps. 3842 */ 3843 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3844 vm_map_t old_map = map; 3845 3846 *var_map = map = entry->object.sub_map; 3847 vm_map_unlock_read(old_map); 3848 goto RetryLookup; 3849 } 3850 3851 /* 3852 * Check whether this task is allowed to have this page. 3853 */ 3854 prot = entry->protection; 3855 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3856 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3857 vm_map_unlock_read(map); 3858 return (KERN_PROTECTION_FAILURE); 3859 } 3860 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3861 (entry->eflags & MAP_ENTRY_COW) && 3862 (fault_type & VM_PROT_WRITE)) { 3863 vm_map_unlock_read(map); 3864 return (KERN_PROTECTION_FAILURE); 3865 } 3866 if ((fault_typea & VM_PROT_COPY) != 0 && 3867 (entry->max_protection & VM_PROT_WRITE) == 0 && 3868 (entry->eflags & MAP_ENTRY_COW) == 0) { 3869 vm_map_unlock_read(map); 3870 return (KERN_PROTECTION_FAILURE); 3871 } 3872 3873 /* 3874 * If this page is not pageable, we have to get it for all possible 3875 * accesses. 3876 */ 3877 *wired = (entry->wired_count != 0); 3878 if (*wired) 3879 fault_type = entry->protection; 3880 size = entry->end - entry->start; 3881 /* 3882 * If the entry was copy-on-write, we either ... 3883 */ 3884 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3885 /* 3886 * If we want to write the page, we may as well handle that 3887 * now since we've got the map locked. 3888 * 3889 * If we don't need to write the page, we just demote the 3890 * permissions allowed. 3891 */ 3892 if ((fault_type & VM_PROT_WRITE) != 0 || 3893 (fault_typea & VM_PROT_COPY) != 0) { 3894 /* 3895 * Make a new object, and place it in the object 3896 * chain. Note that no new references have appeared 3897 * -- one just moved from the map to the new 3898 * object. 3899 */ 3900 if (vm_map_lock_upgrade(map)) 3901 goto RetryLookup; 3902 3903 if (entry->cred == NULL) { 3904 /* 3905 * The debugger owner is charged for 3906 * the memory. 3907 */ 3908 cred = curthread->td_ucred; 3909 crhold(cred); 3910 if (!swap_reserve_by_cred(size, cred)) { 3911 crfree(cred); 3912 vm_map_unlock(map); 3913 return (KERN_RESOURCE_SHORTAGE); 3914 } 3915 entry->cred = cred; 3916 } 3917 vm_object_shadow(&entry->object.vm_object, 3918 &entry->offset, size); 3919 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3920 eobject = entry->object.vm_object; 3921 if (eobject->cred != NULL) { 3922 /* 3923 * The object was not shadowed. 3924 */ 3925 swap_release_by_cred(size, entry->cred); 3926 crfree(entry->cred); 3927 entry->cred = NULL; 3928 } else if (entry->cred != NULL) { 3929 VM_OBJECT_WLOCK(eobject); 3930 eobject->cred = entry->cred; 3931 eobject->charge = size; 3932 VM_OBJECT_WUNLOCK(eobject); 3933 entry->cred = NULL; 3934 } 3935 3936 vm_map_lock_downgrade(map); 3937 } else { 3938 /* 3939 * We're attempting to read a copy-on-write page -- 3940 * don't allow writes. 3941 */ 3942 prot &= ~VM_PROT_WRITE; 3943 } 3944 } 3945 3946 /* 3947 * Create an object if necessary. 3948 */ 3949 if (entry->object.vm_object == NULL && 3950 !map->system_map) { 3951 if (vm_map_lock_upgrade(map)) 3952 goto RetryLookup; 3953 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3954 atop(size)); 3955 entry->offset = 0; 3956 if (entry->cred != NULL) { 3957 VM_OBJECT_WLOCK(entry->object.vm_object); 3958 entry->object.vm_object->cred = entry->cred; 3959 entry->object.vm_object->charge = size; 3960 VM_OBJECT_WUNLOCK(entry->object.vm_object); 3961 entry->cred = NULL; 3962 } 3963 vm_map_lock_downgrade(map); 3964 } 3965 3966 /* 3967 * Return the object/offset from this entry. If the entry was 3968 * copy-on-write or empty, it has been fixed up. 3969 */ 3970 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3971 *object = entry->object.vm_object; 3972 3973 *out_prot = prot; 3974 return (KERN_SUCCESS); 3975} 3976 3977/* 3978 * vm_map_lookup_locked: 3979 * 3980 * Lookup the faulting address. A version of vm_map_lookup that returns 3981 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3982 */ 3983int 3984vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3985 vm_offset_t vaddr, 3986 vm_prot_t fault_typea, 3987 vm_map_entry_t *out_entry, /* OUT */ 3988 vm_object_t *object, /* OUT */ 3989 vm_pindex_t *pindex, /* OUT */ 3990 vm_prot_t *out_prot, /* OUT */ 3991 boolean_t *wired) /* OUT */ 3992{ 3993 vm_map_entry_t entry; 3994 vm_map_t map = *var_map; 3995 vm_prot_t prot; 3996 vm_prot_t fault_type = fault_typea; 3997 3998 /* 3999 * Lookup the faulting address. 4000 */ 4001 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4002 return (KERN_INVALID_ADDRESS); 4003 4004 entry = *out_entry; 4005 4006 /* 4007 * Fail if the entry refers to a submap. 4008 */ 4009 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4010 return (KERN_FAILURE); 4011 4012 /* 4013 * Check whether this task is allowed to have this page. 4014 */ 4015 prot = entry->protection; 4016 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4017 if ((fault_type & prot) != fault_type) 4018 return (KERN_PROTECTION_FAILURE); 4019 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 4020 (entry->eflags & MAP_ENTRY_COW) && 4021 (fault_type & VM_PROT_WRITE)) 4022 return (KERN_PROTECTION_FAILURE); 4023 4024 /* 4025 * If this page is not pageable, we have to get it for all possible 4026 * accesses. 4027 */ 4028 *wired = (entry->wired_count != 0); 4029 if (*wired) 4030 fault_type = entry->protection; 4031 4032 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4033 /* 4034 * Fail if the entry was copy-on-write for a write fault. 4035 */ 4036 if (fault_type & VM_PROT_WRITE) 4037 return (KERN_FAILURE); 4038 /* 4039 * We're attempting to read a copy-on-write page -- 4040 * don't allow writes. 4041 */ 4042 prot &= ~VM_PROT_WRITE; 4043 } 4044 4045 /* 4046 * Fail if an object should be created. 4047 */ 4048 if (entry->object.vm_object == NULL && !map->system_map) 4049 return (KERN_FAILURE); 4050 4051 /* 4052 * Return the object/offset from this entry. If the entry was 4053 * copy-on-write or empty, it has been fixed up. 4054 */ 4055 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4056 *object = entry->object.vm_object; 4057 4058 *out_prot = prot; 4059 return (KERN_SUCCESS); 4060} 4061 4062/* 4063 * vm_map_lookup_done: 4064 * 4065 * Releases locks acquired by a vm_map_lookup 4066 * (according to the handle returned by that lookup). 4067 */ 4068void 4069vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4070{ 4071 /* 4072 * Unlock the main-level map 4073 */ 4074 vm_map_unlock_read(map); 4075} 4076 4077#include "opt_ddb.h" 4078#ifdef DDB 4079#include <sys/kernel.h> 4080 4081#include <ddb/ddb.h> 4082 4083static void 4084vm_map_print(vm_map_t map) 4085{ 4086 vm_map_entry_t entry; 4087 4088 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4089 (void *)map, 4090 (void *)map->pmap, map->nentries, map->timestamp); 4091 4092 db_indent += 2; 4093 for (entry = map->header.next; entry != &map->header; 4094 entry = entry->next) { 4095 db_iprintf("map entry %p: start=%p, end=%p\n", 4096 (void *)entry, (void *)entry->start, (void *)entry->end); 4097 { 4098 static char *inheritance_name[4] = 4099 {"share", "copy", "none", "donate_copy"}; 4100 4101 db_iprintf(" prot=%x/%x/%s", 4102 entry->protection, 4103 entry->max_protection, 4104 inheritance_name[(int)(unsigned char)entry->inheritance]); 4105 if (entry->wired_count != 0) 4106 db_printf(", wired"); 4107 } 4108 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4109 db_printf(", share=%p, offset=0x%jx\n", 4110 (void *)entry->object.sub_map, 4111 (uintmax_t)entry->offset); 4112 if ((entry->prev == &map->header) || 4113 (entry->prev->object.sub_map != 4114 entry->object.sub_map)) { 4115 db_indent += 2; 4116 vm_map_print((vm_map_t)entry->object.sub_map); 4117 db_indent -= 2; 4118 } 4119 } else { 4120 if (entry->cred != NULL) 4121 db_printf(", ruid %d", entry->cred->cr_ruid); 4122 db_printf(", object=%p, offset=0x%jx", 4123 (void *)entry->object.vm_object, 4124 (uintmax_t)entry->offset); 4125 if (entry->object.vm_object && entry->object.vm_object->cred) 4126 db_printf(", obj ruid %d charge %jx", 4127 entry->object.vm_object->cred->cr_ruid, 4128 (uintmax_t)entry->object.vm_object->charge); 4129 if (entry->eflags & MAP_ENTRY_COW) 4130 db_printf(", copy (%s)", 4131 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4132 db_printf("\n"); 4133 4134 if ((entry->prev == &map->header) || 4135 (entry->prev->object.vm_object != 4136 entry->object.vm_object)) { 4137 db_indent += 2; 4138 vm_object_print((db_expr_t)(intptr_t) 4139 entry->object.vm_object, 4140 1, 0, (char *)0); 4141 db_indent -= 2; 4142 } 4143 } 4144 } 4145 db_indent -= 2; 4146} 4147 4148DB_SHOW_COMMAND(map, map) 4149{ 4150 4151 if (!have_addr) { 4152 db_printf("usage: show map <addr>\n"); 4153 return; 4154 } 4155 vm_map_print((vm_map_t)addr); 4156} 4157 4158DB_SHOW_COMMAND(procvm, procvm) 4159{ 4160 struct proc *p; 4161 4162 if (have_addr) { 4163 p = (struct proc *) addr; 4164 } else { 4165 p = curproc; 4166 } 4167 4168 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4169 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4170 (void *)vmspace_pmap(p->p_vmspace)); 4171 4172 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4173} 4174 4175#endif /* DDB */ 4176