vm_map.c revision 266302
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 266302 2014-05-17 11:29:32Z 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 (cow & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) == 0 && 1211 (prev_entry->end == start) && 1212 (prev_entry->wired_count == 0) && 1213 (prev_entry->cred == cred || 1214 (prev_entry->object.vm_object != NULL && 1215 (prev_entry->object.vm_object->cred == cred))) && 1216 vm_object_coalesce(prev_entry->object.vm_object, 1217 prev_entry->offset, 1218 (vm_size_t)(prev_entry->end - prev_entry->start), 1219 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1220 /* 1221 * We were able to extend the object. Determine if we 1222 * can extend the previous map entry to include the 1223 * new range as well. 1224 */ 1225 if ((prev_entry->inheritance == inheritance) && 1226 (prev_entry->protection == prot) && 1227 (prev_entry->max_protection == max)) { 1228 map->size += (end - prev_entry->end); 1229 prev_entry->end = end; 1230 vm_map_entry_resize_free(map, prev_entry); 1231 vm_map_simplify_entry(map, prev_entry); 1232 if (cred != NULL) 1233 crfree(cred); 1234 return (KERN_SUCCESS); 1235 } 1236 1237 /* 1238 * If we can extend the object but cannot extend the 1239 * map entry, we have to create a new map entry. We 1240 * must bump the ref count on the extended object to 1241 * account for it. object may be NULL. 1242 */ 1243 object = prev_entry->object.vm_object; 1244 offset = prev_entry->offset + 1245 (prev_entry->end - prev_entry->start); 1246 vm_object_reference(object); 1247 if (cred != NULL && object != NULL && object->cred != NULL && 1248 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1249 /* Object already accounts for this uid. */ 1250 crfree(cred); 1251 cred = NULL; 1252 } 1253 } 1254 1255 /* 1256 * NOTE: if conditionals fail, object can be NULL here. This occurs 1257 * in things like the buffer map where we manage kva but do not manage 1258 * backing objects. 1259 */ 1260 1261 /* 1262 * Create a new entry 1263 */ 1264 new_entry = vm_map_entry_create(map); 1265 new_entry->start = start; 1266 new_entry->end = end; 1267 new_entry->cred = NULL; 1268 1269 new_entry->eflags = protoeflags; 1270 new_entry->object.vm_object = object; 1271 new_entry->offset = offset; 1272 new_entry->avail_ssize = 0; 1273 1274 new_entry->inheritance = inheritance; 1275 new_entry->protection = prot; 1276 new_entry->max_protection = max; 1277 new_entry->wired_count = 0; 1278 new_entry->wiring_thread = NULL; 1279 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1280 new_entry->next_read = OFF_TO_IDX(offset); 1281 1282 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1283 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1284 new_entry->cred = cred; 1285 1286 /* 1287 * Insert the new entry into the list 1288 */ 1289 vm_map_entry_link(map, prev_entry, new_entry); 1290 map->size += new_entry->end - new_entry->start; 1291 1292 /* 1293 * It may be possible to merge the new entry with the next and/or 1294 * previous entries. However, due to MAP_STACK_* being a hack, a 1295 * panic can result from merging such entries. 1296 */ 1297 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1298 vm_map_simplify_entry(map, new_entry); 1299 1300 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1301 vm_map_pmap_enter(map, start, prot, 1302 object, OFF_TO_IDX(offset), end - start, 1303 cow & MAP_PREFAULT_PARTIAL); 1304 } 1305 1306 return (KERN_SUCCESS); 1307} 1308 1309/* 1310 * vm_map_findspace: 1311 * 1312 * Find the first fit (lowest VM address) for "length" free bytes 1313 * beginning at address >= start in the given map. 1314 * 1315 * In a vm_map_entry, "adj_free" is the amount of free space 1316 * adjacent (higher address) to this entry, and "max_free" is the 1317 * maximum amount of contiguous free space in its subtree. This 1318 * allows finding a free region in one path down the tree, so 1319 * O(log n) amortized with splay trees. 1320 * 1321 * The map must be locked, and leaves it so. 1322 * 1323 * Returns: 0 on success, and starting address in *addr, 1324 * 1 if insufficient space. 1325 */ 1326int 1327vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1328 vm_offset_t *addr) /* OUT */ 1329{ 1330 vm_map_entry_t entry; 1331 vm_offset_t st; 1332 1333 /* 1334 * Request must fit within min/max VM address and must avoid 1335 * address wrap. 1336 */ 1337 if (start < map->min_offset) 1338 start = map->min_offset; 1339 if (start + length > map->max_offset || start + length < start) 1340 return (1); 1341 1342 /* Empty tree means wide open address space. */ 1343 if (map->root == NULL) { 1344 *addr = start; 1345 return (0); 1346 } 1347 1348 /* 1349 * After splay, if start comes before root node, then there 1350 * must be a gap from start to the root. 1351 */ 1352 map->root = vm_map_entry_splay(start, map->root); 1353 if (start + length <= map->root->start) { 1354 *addr = start; 1355 return (0); 1356 } 1357 1358 /* 1359 * Root is the last node that might begin its gap before 1360 * start, and this is the last comparison where address 1361 * wrap might be a problem. 1362 */ 1363 st = (start > map->root->end) ? start : map->root->end; 1364 if (length <= map->root->end + map->root->adj_free - st) { 1365 *addr = st; 1366 return (0); 1367 } 1368 1369 /* With max_free, can immediately tell if no solution. */ 1370 entry = map->root->right; 1371 if (entry == NULL || length > entry->max_free) 1372 return (1); 1373 1374 /* 1375 * Search the right subtree in the order: left subtree, root, 1376 * right subtree (first fit). The previous splay implies that 1377 * all regions in the right subtree have addresses > start. 1378 */ 1379 while (entry != NULL) { 1380 if (entry->left != NULL && entry->left->max_free >= length) 1381 entry = entry->left; 1382 else if (entry->adj_free >= length) { 1383 *addr = entry->end; 1384 return (0); 1385 } else 1386 entry = entry->right; 1387 } 1388 1389 /* Can't get here, so panic if we do. */ 1390 panic("vm_map_findspace: max_free corrupt"); 1391} 1392 1393int 1394vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1395 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1396 vm_prot_t max, int cow) 1397{ 1398 vm_offset_t end; 1399 int result; 1400 1401 end = start + length; 1402 vm_map_lock(map); 1403 VM_MAP_RANGE_CHECK(map, start, end); 1404 (void) vm_map_delete(map, start, end); 1405 result = vm_map_insert(map, object, offset, start, end, prot, 1406 max, cow); 1407 vm_map_unlock(map); 1408 return (result); 1409} 1410 1411/* 1412 * vm_map_find finds an unallocated region in the target address 1413 * map with the given length. The search is defined to be 1414 * first-fit from the specified address; the region found is 1415 * returned in the same parameter. 1416 * 1417 * If object is non-NULL, ref count must be bumped by caller 1418 * prior to making call to account for the new entry. 1419 */ 1420int 1421vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1422 vm_offset_t *addr, /* IN/OUT */ 1423 vm_size_t length, vm_offset_t max_addr, int find_space, 1424 vm_prot_t prot, vm_prot_t max, int cow) 1425{ 1426 vm_offset_t alignment, initial_addr, start; 1427 int result; 1428 1429 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1430 (object->flags & OBJ_COLORED) == 0)) 1431 find_space = VMFS_ANY_SPACE; 1432 if (find_space >> 8 != 0) { 1433 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1434 alignment = (vm_offset_t)1 << (find_space >> 8); 1435 } else 1436 alignment = 0; 1437 initial_addr = *addr; 1438again: 1439 start = initial_addr; 1440 vm_map_lock(map); 1441 do { 1442 if (find_space != VMFS_NO_SPACE) { 1443 if (vm_map_findspace(map, start, length, addr) || 1444 (max_addr != 0 && *addr + length > max_addr)) { 1445 vm_map_unlock(map); 1446 if (find_space == VMFS_OPTIMAL_SPACE) { 1447 find_space = VMFS_ANY_SPACE; 1448 goto again; 1449 } 1450 return (KERN_NO_SPACE); 1451 } 1452 switch (find_space) { 1453 case VMFS_SUPER_SPACE: 1454 case VMFS_OPTIMAL_SPACE: 1455 pmap_align_superpage(object, offset, addr, 1456 length); 1457 break; 1458 case VMFS_ANY_SPACE: 1459 break; 1460 default: 1461 if ((*addr & (alignment - 1)) != 0) { 1462 *addr &= ~(alignment - 1); 1463 *addr += alignment; 1464 } 1465 break; 1466 } 1467 1468 start = *addr; 1469 } 1470 result = vm_map_insert(map, object, offset, start, start + 1471 length, prot, max, cow); 1472 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1473 find_space != VMFS_ANY_SPACE); 1474 vm_map_unlock(map); 1475 return (result); 1476} 1477 1478/* 1479 * vm_map_simplify_entry: 1480 * 1481 * Simplify the given map entry by merging with either neighbor. This 1482 * routine also has the ability to merge with both neighbors. 1483 * 1484 * The map must be locked. 1485 * 1486 * This routine guarentees that the passed entry remains valid (though 1487 * possibly extended). When merging, this routine may delete one or 1488 * both neighbors. 1489 */ 1490void 1491vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1492{ 1493 vm_map_entry_t next, prev; 1494 vm_size_t prevsize, esize; 1495 1496 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1497 return; 1498 1499 prev = entry->prev; 1500 if (prev != &map->header) { 1501 prevsize = prev->end - prev->start; 1502 if ( (prev->end == entry->start) && 1503 (prev->object.vm_object == entry->object.vm_object) && 1504 (!prev->object.vm_object || 1505 (prev->offset + prevsize == entry->offset)) && 1506 (prev->eflags == entry->eflags) && 1507 (prev->protection == entry->protection) && 1508 (prev->max_protection == entry->max_protection) && 1509 (prev->inheritance == entry->inheritance) && 1510 (prev->wired_count == entry->wired_count) && 1511 (prev->cred == entry->cred)) { 1512 vm_map_entry_unlink(map, prev); 1513 entry->start = prev->start; 1514 entry->offset = prev->offset; 1515 if (entry->prev != &map->header) 1516 vm_map_entry_resize_free(map, entry->prev); 1517 1518 /* 1519 * If the backing object is a vnode object, 1520 * vm_object_deallocate() calls vrele(). 1521 * However, vrele() does not lock the vnode 1522 * because the vnode has additional 1523 * references. Thus, the map lock can be kept 1524 * without causing a lock-order reversal with 1525 * the vnode lock. 1526 * 1527 * Since we count the number of virtual page 1528 * mappings in object->un_pager.vnp.writemappings, 1529 * the writemappings value should not be adjusted 1530 * when the entry is disposed of. 1531 */ 1532 if (prev->object.vm_object) 1533 vm_object_deallocate(prev->object.vm_object); 1534 if (prev->cred != NULL) 1535 crfree(prev->cred); 1536 vm_map_entry_dispose(map, prev); 1537 } 1538 } 1539 1540 next = entry->next; 1541 if (next != &map->header) { 1542 esize = entry->end - entry->start; 1543 if ((entry->end == next->start) && 1544 (next->object.vm_object == entry->object.vm_object) && 1545 (!entry->object.vm_object || 1546 (entry->offset + esize == next->offset)) && 1547 (next->eflags == entry->eflags) && 1548 (next->protection == entry->protection) && 1549 (next->max_protection == entry->max_protection) && 1550 (next->inheritance == entry->inheritance) && 1551 (next->wired_count == entry->wired_count) && 1552 (next->cred == entry->cred)) { 1553 vm_map_entry_unlink(map, next); 1554 entry->end = next->end; 1555 vm_map_entry_resize_free(map, entry); 1556 1557 /* 1558 * See comment above. 1559 */ 1560 if (next->object.vm_object) 1561 vm_object_deallocate(next->object.vm_object); 1562 if (next->cred != NULL) 1563 crfree(next->cred); 1564 vm_map_entry_dispose(map, next); 1565 } 1566 } 1567} 1568/* 1569 * vm_map_clip_start: [ internal use only ] 1570 * 1571 * Asserts that the given entry begins at or after 1572 * the specified address; if necessary, 1573 * it splits the entry into two. 1574 */ 1575#define vm_map_clip_start(map, entry, startaddr) \ 1576{ \ 1577 if (startaddr > entry->start) \ 1578 _vm_map_clip_start(map, entry, startaddr); \ 1579} 1580 1581/* 1582 * This routine is called only when it is known that 1583 * the entry must be split. 1584 */ 1585static void 1586_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1587{ 1588 vm_map_entry_t new_entry; 1589 1590 VM_MAP_ASSERT_LOCKED(map); 1591 1592 /* 1593 * Split off the front portion -- note that we must insert the new 1594 * entry BEFORE this one, so that this entry has the specified 1595 * starting address. 1596 */ 1597 vm_map_simplify_entry(map, entry); 1598 1599 /* 1600 * If there is no object backing this entry, we might as well create 1601 * one now. If we defer it, an object can get created after the map 1602 * is clipped, and individual objects will be created for the split-up 1603 * map. This is a bit of a hack, but is also about the best place to 1604 * put this improvement. 1605 */ 1606 if (entry->object.vm_object == NULL && !map->system_map) { 1607 vm_object_t object; 1608 object = vm_object_allocate(OBJT_DEFAULT, 1609 atop(entry->end - entry->start)); 1610 entry->object.vm_object = object; 1611 entry->offset = 0; 1612 if (entry->cred != NULL) { 1613 object->cred = entry->cred; 1614 object->charge = entry->end - entry->start; 1615 entry->cred = NULL; 1616 } 1617 } else if (entry->object.vm_object != NULL && 1618 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1619 entry->cred != NULL) { 1620 VM_OBJECT_WLOCK(entry->object.vm_object); 1621 KASSERT(entry->object.vm_object->cred == NULL, 1622 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1623 entry->object.vm_object->cred = entry->cred; 1624 entry->object.vm_object->charge = entry->end - entry->start; 1625 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1626 entry->cred = NULL; 1627 } 1628 1629 new_entry = vm_map_entry_create(map); 1630 *new_entry = *entry; 1631 1632 new_entry->end = start; 1633 entry->offset += (start - entry->start); 1634 entry->start = start; 1635 if (new_entry->cred != NULL) 1636 crhold(entry->cred); 1637 1638 vm_map_entry_link(map, entry->prev, new_entry); 1639 1640 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1641 vm_object_reference(new_entry->object.vm_object); 1642 /* 1643 * The object->un_pager.vnp.writemappings for the 1644 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1645 * kept as is here. The virtual pages are 1646 * re-distributed among the clipped entries, so the sum is 1647 * left the same. 1648 */ 1649 } 1650} 1651 1652/* 1653 * vm_map_clip_end: [ internal use only ] 1654 * 1655 * Asserts that the given entry ends at or before 1656 * the specified address; if necessary, 1657 * it splits the entry into two. 1658 */ 1659#define vm_map_clip_end(map, entry, endaddr) \ 1660{ \ 1661 if ((endaddr) < (entry->end)) \ 1662 _vm_map_clip_end((map), (entry), (endaddr)); \ 1663} 1664 1665/* 1666 * This routine is called only when it is known that 1667 * the entry must be split. 1668 */ 1669static void 1670_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1671{ 1672 vm_map_entry_t new_entry; 1673 1674 VM_MAP_ASSERT_LOCKED(map); 1675 1676 /* 1677 * If there is no object backing this entry, we might as well create 1678 * one now. If we defer it, an object can get created after the map 1679 * is clipped, and individual objects will be created for the split-up 1680 * map. This is a bit of a hack, but is also about the best place to 1681 * put this improvement. 1682 */ 1683 if (entry->object.vm_object == NULL && !map->system_map) { 1684 vm_object_t object; 1685 object = vm_object_allocate(OBJT_DEFAULT, 1686 atop(entry->end - entry->start)); 1687 entry->object.vm_object = object; 1688 entry->offset = 0; 1689 if (entry->cred != NULL) { 1690 object->cred = entry->cred; 1691 object->charge = entry->end - entry->start; 1692 entry->cred = NULL; 1693 } 1694 } else if (entry->object.vm_object != NULL && 1695 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1696 entry->cred != NULL) { 1697 VM_OBJECT_WLOCK(entry->object.vm_object); 1698 KASSERT(entry->object.vm_object->cred == NULL, 1699 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1700 entry->object.vm_object->cred = entry->cred; 1701 entry->object.vm_object->charge = entry->end - entry->start; 1702 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1703 entry->cred = NULL; 1704 } 1705 1706 /* 1707 * Create a new entry and insert it AFTER the specified entry 1708 */ 1709 new_entry = vm_map_entry_create(map); 1710 *new_entry = *entry; 1711 1712 new_entry->start = entry->end = end; 1713 new_entry->offset += (end - entry->start); 1714 if (new_entry->cred != NULL) 1715 crhold(entry->cred); 1716 1717 vm_map_entry_link(map, entry, new_entry); 1718 1719 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1720 vm_object_reference(new_entry->object.vm_object); 1721 } 1722} 1723 1724/* 1725 * vm_map_submap: [ kernel use only ] 1726 * 1727 * Mark the given range as handled by a subordinate map. 1728 * 1729 * This range must have been created with vm_map_find, 1730 * and no other operations may have been performed on this 1731 * range prior to calling vm_map_submap. 1732 * 1733 * Only a limited number of operations can be performed 1734 * within this rage after calling vm_map_submap: 1735 * vm_fault 1736 * [Don't try vm_map_copy!] 1737 * 1738 * To remove a submapping, one must first remove the 1739 * range from the superior map, and then destroy the 1740 * submap (if desired). [Better yet, don't try it.] 1741 */ 1742int 1743vm_map_submap( 1744 vm_map_t map, 1745 vm_offset_t start, 1746 vm_offset_t end, 1747 vm_map_t submap) 1748{ 1749 vm_map_entry_t entry; 1750 int result = KERN_INVALID_ARGUMENT; 1751 1752 vm_map_lock(map); 1753 1754 VM_MAP_RANGE_CHECK(map, start, end); 1755 1756 if (vm_map_lookup_entry(map, start, &entry)) { 1757 vm_map_clip_start(map, entry, start); 1758 } else 1759 entry = entry->next; 1760 1761 vm_map_clip_end(map, entry, end); 1762 1763 if ((entry->start == start) && (entry->end == end) && 1764 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1765 (entry->object.vm_object == NULL)) { 1766 entry->object.sub_map = submap; 1767 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1768 result = KERN_SUCCESS; 1769 } 1770 vm_map_unlock(map); 1771 1772 return (result); 1773} 1774 1775/* 1776 * The maximum number of pages to map 1777 */ 1778#define MAX_INIT_PT 96 1779 1780/* 1781 * vm_map_pmap_enter: 1782 * 1783 * Preload read-only mappings for the specified object's resident pages 1784 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only 1785 * the resident pages within the address range [addr, addr + ulmin(size, 1786 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within 1787 * the specified address range are mapped. This eliminates many soft 1788 * faults on process startup and immediately after an mmap(2). Because 1789 * these are speculative mappings, cached pages are not reactivated and 1790 * mapped. 1791 */ 1792void 1793vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1794 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1795{ 1796 vm_offset_t start; 1797 vm_page_t p, p_start; 1798 vm_pindex_t psize, tmpidx; 1799 1800 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1801 return; 1802 VM_OBJECT_RLOCK(object); 1803 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1804 VM_OBJECT_RUNLOCK(object); 1805 VM_OBJECT_WLOCK(object); 1806 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1807 pmap_object_init_pt(map->pmap, addr, object, pindex, 1808 size); 1809 VM_OBJECT_WUNLOCK(object); 1810 return; 1811 } 1812 VM_OBJECT_LOCK_DOWNGRADE(object); 1813 } 1814 1815 psize = atop(size); 1816 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0) 1817 psize = MAX_INIT_PT; 1818 if (psize + pindex > object->size) { 1819 if (object->size < pindex) { 1820 VM_OBJECT_RUNLOCK(object); 1821 return; 1822 } 1823 psize = object->size - pindex; 1824 } 1825 1826 start = 0; 1827 p_start = NULL; 1828 1829 p = vm_page_find_least(object, pindex); 1830 /* 1831 * Assert: the variable p is either (1) the page with the 1832 * least pindex greater than or equal to the parameter pindex 1833 * or (2) NULL. 1834 */ 1835 for (; 1836 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1837 p = TAILQ_NEXT(p, listq)) { 1838 /* 1839 * don't allow an madvise to blow away our really 1840 * free pages allocating pv entries. 1841 */ 1842 if ((flags & MAP_PREFAULT_MADVISE) && 1843 cnt.v_free_count < cnt.v_free_reserved) { 1844 psize = tmpidx; 1845 break; 1846 } 1847 if (p->valid == VM_PAGE_BITS_ALL) { 1848 if (p_start == NULL) { 1849 start = addr + ptoa(tmpidx); 1850 p_start = p; 1851 } 1852 } else if (p_start != NULL) { 1853 pmap_enter_object(map->pmap, start, addr + 1854 ptoa(tmpidx), p_start, prot); 1855 p_start = NULL; 1856 } 1857 } 1858 if (p_start != NULL) 1859 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1860 p_start, prot); 1861 VM_OBJECT_RUNLOCK(object); 1862} 1863 1864/* 1865 * vm_map_protect: 1866 * 1867 * Sets the protection of the specified address 1868 * region in the target map. If "set_max" is 1869 * specified, the maximum protection is to be set; 1870 * otherwise, only the current protection is affected. 1871 */ 1872int 1873vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1874 vm_prot_t new_prot, boolean_t set_max) 1875{ 1876 vm_map_entry_t current, entry; 1877 vm_object_t obj; 1878 struct ucred *cred; 1879 vm_prot_t old_prot; 1880 1881 if (start == end) 1882 return (KERN_SUCCESS); 1883 1884 vm_map_lock(map); 1885 1886 VM_MAP_RANGE_CHECK(map, start, end); 1887 1888 if (vm_map_lookup_entry(map, start, &entry)) { 1889 vm_map_clip_start(map, entry, start); 1890 } else { 1891 entry = entry->next; 1892 } 1893 1894 /* 1895 * Make a first pass to check for protection violations. 1896 */ 1897 current = entry; 1898 while ((current != &map->header) && (current->start < end)) { 1899 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1900 vm_map_unlock(map); 1901 return (KERN_INVALID_ARGUMENT); 1902 } 1903 if ((new_prot & current->max_protection) != new_prot) { 1904 vm_map_unlock(map); 1905 return (KERN_PROTECTION_FAILURE); 1906 } 1907 current = current->next; 1908 } 1909 1910 1911 /* 1912 * Do an accounting pass for private read-only mappings that 1913 * now will do cow due to allowed write (e.g. debugger sets 1914 * breakpoint on text segment) 1915 */ 1916 for (current = entry; (current != &map->header) && 1917 (current->start < end); current = current->next) { 1918 1919 vm_map_clip_end(map, current, end); 1920 1921 if (set_max || 1922 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1923 ENTRY_CHARGED(current)) { 1924 continue; 1925 } 1926 1927 cred = curthread->td_ucred; 1928 obj = current->object.vm_object; 1929 1930 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1931 if (!swap_reserve(current->end - current->start)) { 1932 vm_map_unlock(map); 1933 return (KERN_RESOURCE_SHORTAGE); 1934 } 1935 crhold(cred); 1936 current->cred = cred; 1937 continue; 1938 } 1939 1940 VM_OBJECT_WLOCK(obj); 1941 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1942 VM_OBJECT_WUNLOCK(obj); 1943 continue; 1944 } 1945 1946 /* 1947 * Charge for the whole object allocation now, since 1948 * we cannot distinguish between non-charged and 1949 * charged clipped mapping of the same object later. 1950 */ 1951 KASSERT(obj->charge == 0, 1952 ("vm_map_protect: object %p overcharged (entry %p)", 1953 obj, current)); 1954 if (!swap_reserve(ptoa(obj->size))) { 1955 VM_OBJECT_WUNLOCK(obj); 1956 vm_map_unlock(map); 1957 return (KERN_RESOURCE_SHORTAGE); 1958 } 1959 1960 crhold(cred); 1961 obj->cred = cred; 1962 obj->charge = ptoa(obj->size); 1963 VM_OBJECT_WUNLOCK(obj); 1964 } 1965 1966 /* 1967 * Go back and fix up protections. [Note that clipping is not 1968 * necessary the second time.] 1969 */ 1970 current = entry; 1971 while ((current != &map->header) && (current->start < end)) { 1972 old_prot = current->protection; 1973 1974 if (set_max) 1975 current->protection = 1976 (current->max_protection = new_prot) & 1977 old_prot; 1978 else 1979 current->protection = new_prot; 1980 1981 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED)) 1982 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) && 1983 (current->protection & VM_PROT_WRITE) != 0 && 1984 (old_prot & VM_PROT_WRITE) == 0) { 1985 vm_fault_copy_entry(map, map, current, current, NULL); 1986 } 1987 1988 /* 1989 * When restricting access, update the physical map. Worry 1990 * about copy-on-write here. 1991 */ 1992 if ((old_prot & ~current->protection) != 0) { 1993#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1994 VM_PROT_ALL) 1995 pmap_protect(map->pmap, current->start, 1996 current->end, 1997 current->protection & MASK(current)); 1998#undef MASK 1999 } 2000 vm_map_simplify_entry(map, current); 2001 current = current->next; 2002 } 2003 vm_map_unlock(map); 2004 return (KERN_SUCCESS); 2005} 2006 2007/* 2008 * vm_map_madvise: 2009 * 2010 * This routine traverses a processes map handling the madvise 2011 * system call. Advisories are classified as either those effecting 2012 * the vm_map_entry structure, or those effecting the underlying 2013 * objects. 2014 */ 2015int 2016vm_map_madvise( 2017 vm_map_t map, 2018 vm_offset_t start, 2019 vm_offset_t end, 2020 int behav) 2021{ 2022 vm_map_entry_t current, entry; 2023 int modify_map = 0; 2024 2025 /* 2026 * Some madvise calls directly modify the vm_map_entry, in which case 2027 * we need to use an exclusive lock on the map and we need to perform 2028 * various clipping operations. Otherwise we only need a read-lock 2029 * on the map. 2030 */ 2031 switch(behav) { 2032 case MADV_NORMAL: 2033 case MADV_SEQUENTIAL: 2034 case MADV_RANDOM: 2035 case MADV_NOSYNC: 2036 case MADV_AUTOSYNC: 2037 case MADV_NOCORE: 2038 case MADV_CORE: 2039 if (start == end) 2040 return (KERN_SUCCESS); 2041 modify_map = 1; 2042 vm_map_lock(map); 2043 break; 2044 case MADV_WILLNEED: 2045 case MADV_DONTNEED: 2046 case MADV_FREE: 2047 if (start == end) 2048 return (KERN_SUCCESS); 2049 vm_map_lock_read(map); 2050 break; 2051 default: 2052 return (KERN_INVALID_ARGUMENT); 2053 } 2054 2055 /* 2056 * Locate starting entry and clip if necessary. 2057 */ 2058 VM_MAP_RANGE_CHECK(map, start, end); 2059 2060 if (vm_map_lookup_entry(map, start, &entry)) { 2061 if (modify_map) 2062 vm_map_clip_start(map, entry, start); 2063 } else { 2064 entry = entry->next; 2065 } 2066 2067 if (modify_map) { 2068 /* 2069 * madvise behaviors that are implemented in the vm_map_entry. 2070 * 2071 * We clip the vm_map_entry so that behavioral changes are 2072 * limited to the specified address range. 2073 */ 2074 for (current = entry; 2075 (current != &map->header) && (current->start < end); 2076 current = current->next 2077 ) { 2078 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2079 continue; 2080 2081 vm_map_clip_end(map, current, end); 2082 2083 switch (behav) { 2084 case MADV_NORMAL: 2085 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2086 break; 2087 case MADV_SEQUENTIAL: 2088 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2089 break; 2090 case MADV_RANDOM: 2091 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2092 break; 2093 case MADV_NOSYNC: 2094 current->eflags |= MAP_ENTRY_NOSYNC; 2095 break; 2096 case MADV_AUTOSYNC: 2097 current->eflags &= ~MAP_ENTRY_NOSYNC; 2098 break; 2099 case MADV_NOCORE: 2100 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2101 break; 2102 case MADV_CORE: 2103 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2104 break; 2105 default: 2106 break; 2107 } 2108 vm_map_simplify_entry(map, current); 2109 } 2110 vm_map_unlock(map); 2111 } else { 2112 vm_pindex_t pstart, pend; 2113 2114 /* 2115 * madvise behaviors that are implemented in the underlying 2116 * vm_object. 2117 * 2118 * Since we don't clip the vm_map_entry, we have to clip 2119 * the vm_object pindex and count. 2120 */ 2121 for (current = entry; 2122 (current != &map->header) && (current->start < end); 2123 current = current->next 2124 ) { 2125 vm_offset_t useEnd, useStart; 2126 2127 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2128 continue; 2129 2130 pstart = OFF_TO_IDX(current->offset); 2131 pend = pstart + atop(current->end - current->start); 2132 useStart = current->start; 2133 useEnd = current->end; 2134 2135 if (current->start < start) { 2136 pstart += atop(start - current->start); 2137 useStart = start; 2138 } 2139 if (current->end > end) { 2140 pend -= atop(current->end - end); 2141 useEnd = end; 2142 } 2143 2144 if (pstart >= pend) 2145 continue; 2146 2147 /* 2148 * Perform the pmap_advise() before clearing 2149 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2150 * concurrent pmap operation, such as pmap_remove(), 2151 * could clear a reference in the pmap and set 2152 * PGA_REFERENCED on the page before the pmap_advise() 2153 * had completed. Consequently, the page would appear 2154 * referenced based upon an old reference that 2155 * occurred before this pmap_advise() ran. 2156 */ 2157 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2158 pmap_advise(map->pmap, useStart, useEnd, 2159 behav); 2160 2161 vm_object_madvise(current->object.vm_object, pstart, 2162 pend, behav); 2163 if (behav == MADV_WILLNEED) { 2164 vm_map_pmap_enter(map, 2165 useStart, 2166 current->protection, 2167 current->object.vm_object, 2168 pstart, 2169 ptoa(pend - pstart), 2170 MAP_PREFAULT_MADVISE 2171 ); 2172 } 2173 } 2174 vm_map_unlock_read(map); 2175 } 2176 return (0); 2177} 2178 2179 2180/* 2181 * vm_map_inherit: 2182 * 2183 * Sets the inheritance of the specified address 2184 * range in the target map. Inheritance 2185 * affects how the map will be shared with 2186 * child maps at the time of vmspace_fork. 2187 */ 2188int 2189vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2190 vm_inherit_t new_inheritance) 2191{ 2192 vm_map_entry_t entry; 2193 vm_map_entry_t temp_entry; 2194 2195 switch (new_inheritance) { 2196 case VM_INHERIT_NONE: 2197 case VM_INHERIT_COPY: 2198 case VM_INHERIT_SHARE: 2199 break; 2200 default: 2201 return (KERN_INVALID_ARGUMENT); 2202 } 2203 if (start == end) 2204 return (KERN_SUCCESS); 2205 vm_map_lock(map); 2206 VM_MAP_RANGE_CHECK(map, start, end); 2207 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2208 entry = temp_entry; 2209 vm_map_clip_start(map, entry, start); 2210 } else 2211 entry = temp_entry->next; 2212 while ((entry != &map->header) && (entry->start < end)) { 2213 vm_map_clip_end(map, entry, end); 2214 entry->inheritance = new_inheritance; 2215 vm_map_simplify_entry(map, entry); 2216 entry = entry->next; 2217 } 2218 vm_map_unlock(map); 2219 return (KERN_SUCCESS); 2220} 2221 2222/* 2223 * vm_map_unwire: 2224 * 2225 * Implements both kernel and user unwiring. 2226 */ 2227int 2228vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2229 int flags) 2230{ 2231 vm_map_entry_t entry, first_entry, tmp_entry; 2232 vm_offset_t saved_start; 2233 unsigned int last_timestamp; 2234 int rv; 2235 boolean_t need_wakeup, result, user_unwire; 2236 2237 if (start == end) 2238 return (KERN_SUCCESS); 2239 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2240 vm_map_lock(map); 2241 VM_MAP_RANGE_CHECK(map, start, end); 2242 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2243 if (flags & VM_MAP_WIRE_HOLESOK) 2244 first_entry = first_entry->next; 2245 else { 2246 vm_map_unlock(map); 2247 return (KERN_INVALID_ADDRESS); 2248 } 2249 } 2250 last_timestamp = map->timestamp; 2251 entry = first_entry; 2252 while (entry != &map->header && entry->start < end) { 2253 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2254 /* 2255 * We have not yet clipped the entry. 2256 */ 2257 saved_start = (start >= entry->start) ? start : 2258 entry->start; 2259 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2260 if (vm_map_unlock_and_wait(map, 0)) { 2261 /* 2262 * Allow interruption of user unwiring? 2263 */ 2264 } 2265 vm_map_lock(map); 2266 if (last_timestamp+1 != map->timestamp) { 2267 /* 2268 * Look again for the entry because the map was 2269 * modified while it was unlocked. 2270 * Specifically, the entry may have been 2271 * clipped, merged, or deleted. 2272 */ 2273 if (!vm_map_lookup_entry(map, saved_start, 2274 &tmp_entry)) { 2275 if (flags & VM_MAP_WIRE_HOLESOK) 2276 tmp_entry = tmp_entry->next; 2277 else { 2278 if (saved_start == start) { 2279 /* 2280 * First_entry has been deleted. 2281 */ 2282 vm_map_unlock(map); 2283 return (KERN_INVALID_ADDRESS); 2284 } 2285 end = saved_start; 2286 rv = KERN_INVALID_ADDRESS; 2287 goto done; 2288 } 2289 } 2290 if (entry == first_entry) 2291 first_entry = tmp_entry; 2292 else 2293 first_entry = NULL; 2294 entry = tmp_entry; 2295 } 2296 last_timestamp = map->timestamp; 2297 continue; 2298 } 2299 vm_map_clip_start(map, entry, start); 2300 vm_map_clip_end(map, entry, end); 2301 /* 2302 * Mark the entry in case the map lock is released. (See 2303 * above.) 2304 */ 2305 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2306 entry->wiring_thread == NULL, 2307 ("owned map entry %p", entry)); 2308 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2309 entry->wiring_thread = curthread; 2310 /* 2311 * Check the map for holes in the specified region. 2312 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2313 */ 2314 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2315 (entry->end < end && (entry->next == &map->header || 2316 entry->next->start > entry->end))) { 2317 end = entry->end; 2318 rv = KERN_INVALID_ADDRESS; 2319 goto done; 2320 } 2321 /* 2322 * If system unwiring, require that the entry is system wired. 2323 */ 2324 if (!user_unwire && 2325 vm_map_entry_system_wired_count(entry) == 0) { 2326 end = entry->end; 2327 rv = KERN_INVALID_ARGUMENT; 2328 goto done; 2329 } 2330 entry = entry->next; 2331 } 2332 rv = KERN_SUCCESS; 2333done: 2334 need_wakeup = FALSE; 2335 if (first_entry == NULL) { 2336 result = vm_map_lookup_entry(map, start, &first_entry); 2337 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2338 first_entry = first_entry->next; 2339 else 2340 KASSERT(result, ("vm_map_unwire: lookup failed")); 2341 } 2342 for (entry = first_entry; entry != &map->header && entry->start < end; 2343 entry = entry->next) { 2344 /* 2345 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2346 * space in the unwired region could have been mapped 2347 * while the map lock was dropped for draining 2348 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2349 * could be simultaneously wiring this new mapping 2350 * entry. Detect these cases and skip any entries 2351 * marked as in transition by us. 2352 */ 2353 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2354 entry->wiring_thread != curthread) { 2355 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2356 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2357 continue; 2358 } 2359 2360 if (rv == KERN_SUCCESS && (!user_unwire || 2361 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2362 if (user_unwire) 2363 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2364 entry->wired_count--; 2365 if (entry->wired_count == 0) { 2366 /* 2367 * Retain the map lock. 2368 */ 2369 vm_fault_unwire(map, entry->start, entry->end, 2370 entry->object.vm_object != NULL && 2371 (entry->object.vm_object->flags & 2372 OBJ_FICTITIOUS) != 0); 2373 } 2374 } 2375 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2376 ("vm_map_unwire: in-transition flag missing %p", entry)); 2377 KASSERT(entry->wiring_thread == curthread, 2378 ("vm_map_unwire: alien wire %p", entry)); 2379 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2380 entry->wiring_thread = NULL; 2381 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2382 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2383 need_wakeup = TRUE; 2384 } 2385 vm_map_simplify_entry(map, entry); 2386 } 2387 vm_map_unlock(map); 2388 if (need_wakeup) 2389 vm_map_wakeup(map); 2390 return (rv); 2391} 2392 2393/* 2394 * vm_map_wire: 2395 * 2396 * Implements both kernel and user wiring. 2397 */ 2398int 2399vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2400 int flags) 2401{ 2402 vm_map_entry_t entry, first_entry, tmp_entry; 2403 vm_offset_t saved_end, saved_start; 2404 unsigned int last_timestamp; 2405 int rv; 2406 boolean_t fictitious, need_wakeup, result, user_wire; 2407 vm_prot_t prot; 2408 2409 if (start == end) 2410 return (KERN_SUCCESS); 2411 prot = 0; 2412 if (flags & VM_MAP_WIRE_WRITE) 2413 prot |= VM_PROT_WRITE; 2414 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2415 vm_map_lock(map); 2416 VM_MAP_RANGE_CHECK(map, start, end); 2417 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2418 if (flags & VM_MAP_WIRE_HOLESOK) 2419 first_entry = first_entry->next; 2420 else { 2421 vm_map_unlock(map); 2422 return (KERN_INVALID_ADDRESS); 2423 } 2424 } 2425 last_timestamp = map->timestamp; 2426 entry = first_entry; 2427 while (entry != &map->header && entry->start < end) { 2428 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2429 /* 2430 * We have not yet clipped the entry. 2431 */ 2432 saved_start = (start >= entry->start) ? start : 2433 entry->start; 2434 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2435 if (vm_map_unlock_and_wait(map, 0)) { 2436 /* 2437 * Allow interruption of user wiring? 2438 */ 2439 } 2440 vm_map_lock(map); 2441 if (last_timestamp + 1 != map->timestamp) { 2442 /* 2443 * Look again for the entry because the map was 2444 * modified while it was unlocked. 2445 * Specifically, the entry may have been 2446 * clipped, merged, or deleted. 2447 */ 2448 if (!vm_map_lookup_entry(map, saved_start, 2449 &tmp_entry)) { 2450 if (flags & VM_MAP_WIRE_HOLESOK) 2451 tmp_entry = tmp_entry->next; 2452 else { 2453 if (saved_start == start) { 2454 /* 2455 * first_entry has been deleted. 2456 */ 2457 vm_map_unlock(map); 2458 return (KERN_INVALID_ADDRESS); 2459 } 2460 end = saved_start; 2461 rv = KERN_INVALID_ADDRESS; 2462 goto done; 2463 } 2464 } 2465 if (entry == first_entry) 2466 first_entry = tmp_entry; 2467 else 2468 first_entry = NULL; 2469 entry = tmp_entry; 2470 } 2471 last_timestamp = map->timestamp; 2472 continue; 2473 } 2474 vm_map_clip_start(map, entry, start); 2475 vm_map_clip_end(map, entry, end); 2476 /* 2477 * Mark the entry in case the map lock is released. (See 2478 * above.) 2479 */ 2480 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2481 entry->wiring_thread == NULL, 2482 ("owned map entry %p", entry)); 2483 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2484 entry->wiring_thread = curthread; 2485 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2486 || (entry->protection & prot) != prot) { 2487 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2488 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2489 end = entry->end; 2490 rv = KERN_INVALID_ADDRESS; 2491 goto done; 2492 } 2493 goto next_entry; 2494 } 2495 if (entry->wired_count == 0) { 2496 entry->wired_count++; 2497 saved_start = entry->start; 2498 saved_end = entry->end; 2499 fictitious = entry->object.vm_object != NULL && 2500 (entry->object.vm_object->flags & 2501 OBJ_FICTITIOUS) != 0; 2502 /* 2503 * Release the map lock, relying on the in-transition 2504 * mark. Mark the map busy for fork. 2505 */ 2506 vm_map_busy(map); 2507 vm_map_unlock(map); 2508 rv = vm_fault_wire(map, saved_start, saved_end, 2509 fictitious); 2510 vm_map_lock(map); 2511 vm_map_unbusy(map); 2512 if (last_timestamp + 1 != map->timestamp) { 2513 /* 2514 * Look again for the entry because the map was 2515 * modified while it was unlocked. The entry 2516 * may have been clipped, but NOT merged or 2517 * deleted. 2518 */ 2519 result = vm_map_lookup_entry(map, saved_start, 2520 &tmp_entry); 2521 KASSERT(result, ("vm_map_wire: lookup failed")); 2522 if (entry == first_entry) 2523 first_entry = tmp_entry; 2524 else 2525 first_entry = NULL; 2526 entry = tmp_entry; 2527 while (entry->end < saved_end) { 2528 if (rv != KERN_SUCCESS) { 2529 KASSERT(entry->wired_count == 1, 2530 ("vm_map_wire: bad count")); 2531 entry->wired_count = -1; 2532 } 2533 entry = entry->next; 2534 } 2535 } 2536 last_timestamp = map->timestamp; 2537 if (rv != KERN_SUCCESS) { 2538 KASSERT(entry->wired_count == 1, 2539 ("vm_map_wire: bad count")); 2540 /* 2541 * Assign an out-of-range value to represent 2542 * the failure to wire this entry. 2543 */ 2544 entry->wired_count = -1; 2545 end = entry->end; 2546 goto done; 2547 } 2548 } else if (!user_wire || 2549 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2550 entry->wired_count++; 2551 } 2552 /* 2553 * Check the map for holes in the specified region. 2554 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2555 */ 2556 next_entry: 2557 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2558 (entry->end < end && (entry->next == &map->header || 2559 entry->next->start > entry->end))) { 2560 end = entry->end; 2561 rv = KERN_INVALID_ADDRESS; 2562 goto done; 2563 } 2564 entry = entry->next; 2565 } 2566 rv = KERN_SUCCESS; 2567done: 2568 need_wakeup = FALSE; 2569 if (first_entry == NULL) { 2570 result = vm_map_lookup_entry(map, start, &first_entry); 2571 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2572 first_entry = first_entry->next; 2573 else 2574 KASSERT(result, ("vm_map_wire: lookup failed")); 2575 } 2576 for (entry = first_entry; entry != &map->header && entry->start < end; 2577 entry = entry->next) { 2578 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2579 goto next_entry_done; 2580 2581 /* 2582 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2583 * space in the unwired region could have been mapped 2584 * while the map lock was dropped for faulting in the 2585 * pages or draining MAP_ENTRY_IN_TRANSITION. 2586 * Moreover, another thread could be simultaneously 2587 * wiring this new mapping entry. Detect these cases 2588 * and skip any entries marked as in transition by us. 2589 */ 2590 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2591 entry->wiring_thread != curthread) { 2592 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2593 ("vm_map_wire: !HOLESOK and new/changed entry")); 2594 continue; 2595 } 2596 2597 if (rv == KERN_SUCCESS) { 2598 if (user_wire) 2599 entry->eflags |= MAP_ENTRY_USER_WIRED; 2600 } else if (entry->wired_count == -1) { 2601 /* 2602 * Wiring failed on this entry. Thus, unwiring is 2603 * unnecessary. 2604 */ 2605 entry->wired_count = 0; 2606 } else { 2607 if (!user_wire || 2608 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2609 entry->wired_count--; 2610 if (entry->wired_count == 0) { 2611 /* 2612 * Retain the map lock. 2613 */ 2614 vm_fault_unwire(map, entry->start, entry->end, 2615 entry->object.vm_object != NULL && 2616 (entry->object.vm_object->flags & 2617 OBJ_FICTITIOUS) != 0); 2618 } 2619 } 2620 next_entry_done: 2621 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2622 ("vm_map_wire: in-transition flag missing %p", entry)); 2623 KASSERT(entry->wiring_thread == curthread, 2624 ("vm_map_wire: alien wire %p", entry)); 2625 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2626 MAP_ENTRY_WIRE_SKIPPED); 2627 entry->wiring_thread = NULL; 2628 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2629 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2630 need_wakeup = TRUE; 2631 } 2632 vm_map_simplify_entry(map, entry); 2633 } 2634 vm_map_unlock(map); 2635 if (need_wakeup) 2636 vm_map_wakeup(map); 2637 return (rv); 2638} 2639 2640/* 2641 * vm_map_sync 2642 * 2643 * Push any dirty cached pages in the address range to their pager. 2644 * If syncio is TRUE, dirty pages are written synchronously. 2645 * If invalidate is TRUE, any cached pages are freed as well. 2646 * 2647 * If the size of the region from start to end is zero, we are 2648 * supposed to flush all modified pages within the region containing 2649 * start. Unfortunately, a region can be split or coalesced with 2650 * neighboring regions, making it difficult to determine what the 2651 * original region was. Therefore, we approximate this requirement by 2652 * flushing the current region containing start. 2653 * 2654 * Returns an error if any part of the specified range is not mapped. 2655 */ 2656int 2657vm_map_sync( 2658 vm_map_t map, 2659 vm_offset_t start, 2660 vm_offset_t end, 2661 boolean_t syncio, 2662 boolean_t invalidate) 2663{ 2664 vm_map_entry_t current; 2665 vm_map_entry_t entry; 2666 vm_size_t size; 2667 vm_object_t object; 2668 vm_ooffset_t offset; 2669 unsigned int last_timestamp; 2670 boolean_t failed; 2671 2672 vm_map_lock_read(map); 2673 VM_MAP_RANGE_CHECK(map, start, end); 2674 if (!vm_map_lookup_entry(map, start, &entry)) { 2675 vm_map_unlock_read(map); 2676 return (KERN_INVALID_ADDRESS); 2677 } else if (start == end) { 2678 start = entry->start; 2679 end = entry->end; 2680 } 2681 /* 2682 * Make a first pass to check for user-wired memory and holes. 2683 */ 2684 for (current = entry; current != &map->header && current->start < end; 2685 current = current->next) { 2686 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2687 vm_map_unlock_read(map); 2688 return (KERN_INVALID_ARGUMENT); 2689 } 2690 if (end > current->end && 2691 (current->next == &map->header || 2692 current->end != current->next->start)) { 2693 vm_map_unlock_read(map); 2694 return (KERN_INVALID_ADDRESS); 2695 } 2696 } 2697 2698 if (invalidate) 2699 pmap_remove(map->pmap, start, end); 2700 failed = FALSE; 2701 2702 /* 2703 * Make a second pass, cleaning/uncaching pages from the indicated 2704 * objects as we go. 2705 */ 2706 for (current = entry; current != &map->header && current->start < end;) { 2707 offset = current->offset + (start - current->start); 2708 size = (end <= current->end ? end : current->end) - start; 2709 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2710 vm_map_t smap; 2711 vm_map_entry_t tentry; 2712 vm_size_t tsize; 2713 2714 smap = current->object.sub_map; 2715 vm_map_lock_read(smap); 2716 (void) vm_map_lookup_entry(smap, offset, &tentry); 2717 tsize = tentry->end - offset; 2718 if (tsize < size) 2719 size = tsize; 2720 object = tentry->object.vm_object; 2721 offset = tentry->offset + (offset - tentry->start); 2722 vm_map_unlock_read(smap); 2723 } else { 2724 object = current->object.vm_object; 2725 } 2726 vm_object_reference(object); 2727 last_timestamp = map->timestamp; 2728 vm_map_unlock_read(map); 2729 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2730 failed = TRUE; 2731 start += size; 2732 vm_object_deallocate(object); 2733 vm_map_lock_read(map); 2734 if (last_timestamp == map->timestamp || 2735 !vm_map_lookup_entry(map, start, ¤t)) 2736 current = current->next; 2737 } 2738 2739 vm_map_unlock_read(map); 2740 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2741} 2742 2743/* 2744 * vm_map_entry_unwire: [ internal use only ] 2745 * 2746 * Make the region specified by this entry pageable. 2747 * 2748 * The map in question should be locked. 2749 * [This is the reason for this routine's existence.] 2750 */ 2751static void 2752vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2753{ 2754 vm_fault_unwire(map, entry->start, entry->end, 2755 entry->object.vm_object != NULL && 2756 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2757 entry->wired_count = 0; 2758} 2759 2760static void 2761vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2762{ 2763 2764 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2765 vm_object_deallocate(entry->object.vm_object); 2766 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2767} 2768 2769/* 2770 * vm_map_entry_delete: [ internal use only ] 2771 * 2772 * Deallocate the given entry from the target map. 2773 */ 2774static void 2775vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2776{ 2777 vm_object_t object; 2778 vm_pindex_t offidxstart, offidxend, count, size1; 2779 vm_ooffset_t size; 2780 2781 vm_map_entry_unlink(map, entry); 2782 object = entry->object.vm_object; 2783 size = entry->end - entry->start; 2784 map->size -= size; 2785 2786 if (entry->cred != NULL) { 2787 swap_release_by_cred(size, entry->cred); 2788 crfree(entry->cred); 2789 } 2790 2791 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2792 (object != NULL)) { 2793 KASSERT(entry->cred == NULL || object->cred == NULL || 2794 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2795 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2796 count = OFF_TO_IDX(size); 2797 offidxstart = OFF_TO_IDX(entry->offset); 2798 offidxend = offidxstart + count; 2799 VM_OBJECT_WLOCK(object); 2800 if (object->ref_count != 1 && 2801 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2802 object == kernel_object || object == kmem_object)) { 2803 vm_object_collapse(object); 2804 2805 /* 2806 * The option OBJPR_NOTMAPPED can be passed here 2807 * because vm_map_delete() already performed 2808 * pmap_remove() on the only mapping to this range 2809 * of pages. 2810 */ 2811 vm_object_page_remove(object, offidxstart, offidxend, 2812 OBJPR_NOTMAPPED); 2813 if (object->type == OBJT_SWAP) 2814 swap_pager_freespace(object, offidxstart, count); 2815 if (offidxend >= object->size && 2816 offidxstart < object->size) { 2817 size1 = object->size; 2818 object->size = offidxstart; 2819 if (object->cred != NULL) { 2820 size1 -= object->size; 2821 KASSERT(object->charge >= ptoa(size1), 2822 ("vm_map_entry_delete: object->charge < 0")); 2823 swap_release_by_cred(ptoa(size1), object->cred); 2824 object->charge -= ptoa(size1); 2825 } 2826 } 2827 } 2828 VM_OBJECT_WUNLOCK(object); 2829 } else 2830 entry->object.vm_object = NULL; 2831 if (map->system_map) 2832 vm_map_entry_deallocate(entry, TRUE); 2833 else { 2834 entry->next = curthread->td_map_def_user; 2835 curthread->td_map_def_user = entry; 2836 } 2837} 2838 2839/* 2840 * vm_map_delete: [ internal use only ] 2841 * 2842 * Deallocates the given address range from the target 2843 * map. 2844 */ 2845int 2846vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2847{ 2848 vm_map_entry_t entry; 2849 vm_map_entry_t first_entry; 2850 2851 VM_MAP_ASSERT_LOCKED(map); 2852 if (start == end) 2853 return (KERN_SUCCESS); 2854 2855 /* 2856 * Find the start of the region, and clip it 2857 */ 2858 if (!vm_map_lookup_entry(map, start, &first_entry)) 2859 entry = first_entry->next; 2860 else { 2861 entry = first_entry; 2862 vm_map_clip_start(map, entry, start); 2863 } 2864 2865 /* 2866 * Step through all entries in this region 2867 */ 2868 while ((entry != &map->header) && (entry->start < end)) { 2869 vm_map_entry_t next; 2870 2871 /* 2872 * Wait for wiring or unwiring of an entry to complete. 2873 * Also wait for any system wirings to disappear on 2874 * user maps. 2875 */ 2876 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2877 (vm_map_pmap(map) != kernel_pmap && 2878 vm_map_entry_system_wired_count(entry) != 0)) { 2879 unsigned int last_timestamp; 2880 vm_offset_t saved_start; 2881 vm_map_entry_t tmp_entry; 2882 2883 saved_start = entry->start; 2884 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2885 last_timestamp = map->timestamp; 2886 (void) vm_map_unlock_and_wait(map, 0); 2887 vm_map_lock(map); 2888 if (last_timestamp + 1 != map->timestamp) { 2889 /* 2890 * Look again for the entry because the map was 2891 * modified while it was unlocked. 2892 * Specifically, the entry may have been 2893 * clipped, merged, or deleted. 2894 */ 2895 if (!vm_map_lookup_entry(map, saved_start, 2896 &tmp_entry)) 2897 entry = tmp_entry->next; 2898 else { 2899 entry = tmp_entry; 2900 vm_map_clip_start(map, entry, 2901 saved_start); 2902 } 2903 } 2904 continue; 2905 } 2906 vm_map_clip_end(map, entry, end); 2907 2908 next = entry->next; 2909 2910 /* 2911 * Unwire before removing addresses from the pmap; otherwise, 2912 * unwiring will put the entries back in the pmap. 2913 */ 2914 if (entry->wired_count != 0) { 2915 vm_map_entry_unwire(map, entry); 2916 } 2917 2918 pmap_remove(map->pmap, entry->start, entry->end); 2919 2920 /* 2921 * Delete the entry only after removing all pmap 2922 * entries pointing to its pages. (Otherwise, its 2923 * page frames may be reallocated, and any modify bits 2924 * will be set in the wrong object!) 2925 */ 2926 vm_map_entry_delete(map, entry); 2927 entry = next; 2928 } 2929 return (KERN_SUCCESS); 2930} 2931 2932/* 2933 * vm_map_remove: 2934 * 2935 * Remove the given address range from the target map. 2936 * This is the exported form of vm_map_delete. 2937 */ 2938int 2939vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2940{ 2941 int result; 2942 2943 vm_map_lock(map); 2944 VM_MAP_RANGE_CHECK(map, start, end); 2945 result = vm_map_delete(map, start, end); 2946 vm_map_unlock(map); 2947 return (result); 2948} 2949 2950/* 2951 * vm_map_check_protection: 2952 * 2953 * Assert that the target map allows the specified privilege on the 2954 * entire address region given. The entire region must be allocated. 2955 * 2956 * WARNING! This code does not and should not check whether the 2957 * contents of the region is accessible. For example a smaller file 2958 * might be mapped into a larger address space. 2959 * 2960 * NOTE! This code is also called by munmap(). 2961 * 2962 * The map must be locked. A read lock is sufficient. 2963 */ 2964boolean_t 2965vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2966 vm_prot_t protection) 2967{ 2968 vm_map_entry_t entry; 2969 vm_map_entry_t tmp_entry; 2970 2971 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2972 return (FALSE); 2973 entry = tmp_entry; 2974 2975 while (start < end) { 2976 if (entry == &map->header) 2977 return (FALSE); 2978 /* 2979 * No holes allowed! 2980 */ 2981 if (start < entry->start) 2982 return (FALSE); 2983 /* 2984 * Check protection associated with entry. 2985 */ 2986 if ((entry->protection & protection) != protection) 2987 return (FALSE); 2988 /* go to next entry */ 2989 start = entry->end; 2990 entry = entry->next; 2991 } 2992 return (TRUE); 2993} 2994 2995/* 2996 * vm_map_copy_entry: 2997 * 2998 * Copies the contents of the source entry to the destination 2999 * entry. The entries *must* be aligned properly. 3000 */ 3001static void 3002vm_map_copy_entry( 3003 vm_map_t src_map, 3004 vm_map_t dst_map, 3005 vm_map_entry_t src_entry, 3006 vm_map_entry_t dst_entry, 3007 vm_ooffset_t *fork_charge) 3008{ 3009 vm_object_t src_object; 3010 vm_map_entry_t fake_entry; 3011 vm_offset_t size; 3012 struct ucred *cred; 3013 int charged; 3014 3015 VM_MAP_ASSERT_LOCKED(dst_map); 3016 3017 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3018 return; 3019 3020 if (src_entry->wired_count == 0) { 3021 3022 /* 3023 * If the source entry is marked needs_copy, it is already 3024 * write-protected. 3025 */ 3026 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 3027 pmap_protect(src_map->pmap, 3028 src_entry->start, 3029 src_entry->end, 3030 src_entry->protection & ~VM_PROT_WRITE); 3031 } 3032 3033 /* 3034 * Make a copy of the object. 3035 */ 3036 size = src_entry->end - src_entry->start; 3037 if ((src_object = src_entry->object.vm_object) != NULL) { 3038 VM_OBJECT_WLOCK(src_object); 3039 charged = ENTRY_CHARGED(src_entry); 3040 if ((src_object->handle == NULL) && 3041 (src_object->type == OBJT_DEFAULT || 3042 src_object->type == OBJT_SWAP)) { 3043 vm_object_collapse(src_object); 3044 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3045 vm_object_split(src_entry); 3046 src_object = src_entry->object.vm_object; 3047 } 3048 } 3049 vm_object_reference_locked(src_object); 3050 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3051 if (src_entry->cred != NULL && 3052 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3053 KASSERT(src_object->cred == NULL, 3054 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3055 src_object)); 3056 src_object->cred = src_entry->cred; 3057 src_object->charge = size; 3058 } 3059 VM_OBJECT_WUNLOCK(src_object); 3060 dst_entry->object.vm_object = src_object; 3061 if (charged) { 3062 cred = curthread->td_ucred; 3063 crhold(cred); 3064 dst_entry->cred = cred; 3065 *fork_charge += size; 3066 if (!(src_entry->eflags & 3067 MAP_ENTRY_NEEDS_COPY)) { 3068 crhold(cred); 3069 src_entry->cred = cred; 3070 *fork_charge += size; 3071 } 3072 } 3073 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3074 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3075 dst_entry->offset = src_entry->offset; 3076 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3077 /* 3078 * MAP_ENTRY_VN_WRITECNT cannot 3079 * indicate write reference from 3080 * src_entry, since the entry is 3081 * marked as needs copy. Allocate a 3082 * fake entry that is used to 3083 * decrement object->un_pager.vnp.writecount 3084 * at the appropriate time. Attach 3085 * fake_entry to the deferred list. 3086 */ 3087 fake_entry = vm_map_entry_create(dst_map); 3088 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3089 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3090 vm_object_reference(src_object); 3091 fake_entry->object.vm_object = src_object; 3092 fake_entry->start = src_entry->start; 3093 fake_entry->end = src_entry->end; 3094 fake_entry->next = curthread->td_map_def_user; 3095 curthread->td_map_def_user = fake_entry; 3096 } 3097 } else { 3098 dst_entry->object.vm_object = NULL; 3099 dst_entry->offset = 0; 3100 if (src_entry->cred != NULL) { 3101 dst_entry->cred = curthread->td_ucred; 3102 crhold(dst_entry->cred); 3103 *fork_charge += size; 3104 } 3105 } 3106 3107 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3108 dst_entry->end - dst_entry->start, src_entry->start); 3109 } else { 3110 /* 3111 * Of course, wired down pages can't be set copy-on-write. 3112 * Cause wired pages to be copied into the new map by 3113 * simulating faults (the new pages are pageable) 3114 */ 3115 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3116 fork_charge); 3117 } 3118} 3119 3120/* 3121 * vmspace_map_entry_forked: 3122 * Update the newly-forked vmspace each time a map entry is inherited 3123 * or copied. The values for vm_dsize and vm_tsize are approximate 3124 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3125 */ 3126static void 3127vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3128 vm_map_entry_t entry) 3129{ 3130 vm_size_t entrysize; 3131 vm_offset_t newend; 3132 3133 entrysize = entry->end - entry->start; 3134 vm2->vm_map.size += entrysize; 3135 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3136 vm2->vm_ssize += btoc(entrysize); 3137 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3138 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3139 newend = MIN(entry->end, 3140 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3141 vm2->vm_dsize += btoc(newend - entry->start); 3142 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3143 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3144 newend = MIN(entry->end, 3145 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3146 vm2->vm_tsize += btoc(newend - entry->start); 3147 } 3148} 3149 3150/* 3151 * vmspace_fork: 3152 * Create a new process vmspace structure and vm_map 3153 * based on those of an existing process. The new map 3154 * is based on the old map, according to the inheritance 3155 * values on the regions in that map. 3156 * 3157 * XXX It might be worth coalescing the entries added to the new vmspace. 3158 * 3159 * The source map must not be locked. 3160 */ 3161struct vmspace * 3162vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3163{ 3164 struct vmspace *vm2; 3165 vm_map_t new_map, old_map; 3166 vm_map_entry_t new_entry, old_entry; 3167 vm_object_t object; 3168 int locked; 3169 3170 old_map = &vm1->vm_map; 3171 /* Copy immutable fields of vm1 to vm2. */ 3172 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3173 if (vm2 == NULL) 3174 return (NULL); 3175 vm2->vm_taddr = vm1->vm_taddr; 3176 vm2->vm_daddr = vm1->vm_daddr; 3177 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3178 vm_map_lock(old_map); 3179 if (old_map->busy) 3180 vm_map_wait_busy(old_map); 3181 new_map = &vm2->vm_map; 3182 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3183 KASSERT(locked, ("vmspace_fork: lock failed")); 3184 3185 old_entry = old_map->header.next; 3186 3187 while (old_entry != &old_map->header) { 3188 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3189 panic("vm_map_fork: encountered a submap"); 3190 3191 switch (old_entry->inheritance) { 3192 case VM_INHERIT_NONE: 3193 break; 3194 3195 case VM_INHERIT_SHARE: 3196 /* 3197 * Clone the entry, creating the shared object if necessary. 3198 */ 3199 object = old_entry->object.vm_object; 3200 if (object == NULL) { 3201 object = vm_object_allocate(OBJT_DEFAULT, 3202 atop(old_entry->end - old_entry->start)); 3203 old_entry->object.vm_object = object; 3204 old_entry->offset = 0; 3205 if (old_entry->cred != NULL) { 3206 object->cred = old_entry->cred; 3207 object->charge = old_entry->end - 3208 old_entry->start; 3209 old_entry->cred = NULL; 3210 } 3211 } 3212 3213 /* 3214 * Add the reference before calling vm_object_shadow 3215 * to insure that a shadow object is created. 3216 */ 3217 vm_object_reference(object); 3218 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3219 vm_object_shadow(&old_entry->object.vm_object, 3220 &old_entry->offset, 3221 old_entry->end - old_entry->start); 3222 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3223 /* Transfer the second reference too. */ 3224 vm_object_reference( 3225 old_entry->object.vm_object); 3226 3227 /* 3228 * As in vm_map_simplify_entry(), the 3229 * vnode lock will not be acquired in 3230 * this call to vm_object_deallocate(). 3231 */ 3232 vm_object_deallocate(object); 3233 object = old_entry->object.vm_object; 3234 } 3235 VM_OBJECT_WLOCK(object); 3236 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3237 if (old_entry->cred != NULL) { 3238 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3239 object->cred = old_entry->cred; 3240 object->charge = old_entry->end - old_entry->start; 3241 old_entry->cred = NULL; 3242 } 3243 3244 /* 3245 * Assert the correct state of the vnode 3246 * v_writecount while the object is locked, to 3247 * not relock it later for the assertion 3248 * correctness. 3249 */ 3250 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3251 object->type == OBJT_VNODE) { 3252 KASSERT(((struct vnode *)object->handle)-> 3253 v_writecount > 0, 3254 ("vmspace_fork: v_writecount %p", object)); 3255 KASSERT(object->un_pager.vnp.writemappings > 0, 3256 ("vmspace_fork: vnp.writecount %p", 3257 object)); 3258 } 3259 VM_OBJECT_WUNLOCK(object); 3260 3261 /* 3262 * Clone the entry, referencing the shared object. 3263 */ 3264 new_entry = vm_map_entry_create(new_map); 3265 *new_entry = *old_entry; 3266 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3267 MAP_ENTRY_IN_TRANSITION); 3268 new_entry->wiring_thread = NULL; 3269 new_entry->wired_count = 0; 3270 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3271 vnode_pager_update_writecount(object, 3272 new_entry->start, new_entry->end); 3273 } 3274 3275 /* 3276 * Insert the entry into the new map -- we know we're 3277 * inserting at the end of the new map. 3278 */ 3279 vm_map_entry_link(new_map, new_map->header.prev, 3280 new_entry); 3281 vmspace_map_entry_forked(vm1, vm2, new_entry); 3282 3283 /* 3284 * Update the physical map 3285 */ 3286 pmap_copy(new_map->pmap, old_map->pmap, 3287 new_entry->start, 3288 (old_entry->end - old_entry->start), 3289 old_entry->start); 3290 break; 3291 3292 case VM_INHERIT_COPY: 3293 /* 3294 * Clone the entry and link into the map. 3295 */ 3296 new_entry = vm_map_entry_create(new_map); 3297 *new_entry = *old_entry; 3298 /* 3299 * Copied entry is COW over the old object. 3300 */ 3301 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3302 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3303 new_entry->wiring_thread = NULL; 3304 new_entry->wired_count = 0; 3305 new_entry->object.vm_object = NULL; 3306 new_entry->cred = NULL; 3307 vm_map_entry_link(new_map, new_map->header.prev, 3308 new_entry); 3309 vmspace_map_entry_forked(vm1, vm2, new_entry); 3310 vm_map_copy_entry(old_map, new_map, old_entry, 3311 new_entry, fork_charge); 3312 break; 3313 } 3314 old_entry = old_entry->next; 3315 } 3316 /* 3317 * Use inlined vm_map_unlock() to postpone handling the deferred 3318 * map entries, which cannot be done until both old_map and 3319 * new_map locks are released. 3320 */ 3321 sx_xunlock(&old_map->lock); 3322 sx_xunlock(&new_map->lock); 3323 vm_map_process_deferred(); 3324 3325 return (vm2); 3326} 3327 3328int 3329vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3330 vm_prot_t prot, vm_prot_t max, int cow) 3331{ 3332 vm_map_entry_t new_entry, prev_entry; 3333 vm_offset_t bot, top; 3334 vm_size_t growsize, init_ssize; 3335 int orient, rv; 3336 rlim_t lmemlim, vmemlim; 3337 3338 /* 3339 * The stack orientation is piggybacked with the cow argument. 3340 * Extract it into orient and mask the cow argument so that we 3341 * don't pass it around further. 3342 * NOTE: We explicitly allow bi-directional stacks. 3343 */ 3344 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3345 KASSERT(orient != 0, ("No stack grow direction")); 3346 3347 if (addrbos < vm_map_min(map) || 3348 addrbos > vm_map_max(map) || 3349 addrbos + max_ssize < addrbos) 3350 return (KERN_NO_SPACE); 3351 3352 growsize = sgrowsiz; 3353 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3354 3355 PROC_LOCK(curproc); 3356 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3357 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3358 PROC_UNLOCK(curproc); 3359 3360 vm_map_lock(map); 3361 3362 /* If addr is already mapped, no go */ 3363 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3364 vm_map_unlock(map); 3365 return (KERN_NO_SPACE); 3366 } 3367 3368 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3369 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3370 vm_map_unlock(map); 3371 return (KERN_NO_SPACE); 3372 } 3373 } 3374 3375 /* If we would blow our VMEM resource limit, no go */ 3376 if (map->size + init_ssize > vmemlim) { 3377 vm_map_unlock(map); 3378 return (KERN_NO_SPACE); 3379 } 3380 3381 /* 3382 * If we can't accomodate max_ssize in the current mapping, no go. 3383 * However, we need to be aware that subsequent user mappings might 3384 * map into the space we have reserved for stack, and currently this 3385 * space is not protected. 3386 * 3387 * Hopefully we will at least detect this condition when we try to 3388 * grow the stack. 3389 */ 3390 if ((prev_entry->next != &map->header) && 3391 (prev_entry->next->start < addrbos + max_ssize)) { 3392 vm_map_unlock(map); 3393 return (KERN_NO_SPACE); 3394 } 3395 3396 /* 3397 * We initially map a stack of only init_ssize. We will grow as 3398 * needed later. Depending on the orientation of the stack (i.e. 3399 * the grow direction) we either map at the top of the range, the 3400 * bottom of the range or in the middle. 3401 * 3402 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3403 * and cow to be 0. Possibly we should eliminate these as input 3404 * parameters, and just pass these values here in the insert call. 3405 */ 3406 if (orient == MAP_STACK_GROWS_DOWN) 3407 bot = addrbos + max_ssize - init_ssize; 3408 else if (orient == MAP_STACK_GROWS_UP) 3409 bot = addrbos; 3410 else 3411 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3412 top = bot + init_ssize; 3413 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3414 3415 /* Now set the avail_ssize amount. */ 3416 if (rv == KERN_SUCCESS) { 3417 if (prev_entry != &map->header) 3418 vm_map_clip_end(map, prev_entry, bot); 3419 new_entry = prev_entry->next; 3420 if (new_entry->end != top || new_entry->start != bot) 3421 panic("Bad entry start/end for new stack entry"); 3422 3423 new_entry->avail_ssize = max_ssize - init_ssize; 3424 if (orient & MAP_STACK_GROWS_DOWN) 3425 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3426 if (orient & MAP_STACK_GROWS_UP) 3427 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3428 } 3429 3430 vm_map_unlock(map); 3431 return (rv); 3432} 3433 3434static int stack_guard_page = 0; 3435TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3436SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3437 &stack_guard_page, 0, 3438 "Insert stack guard page ahead of the growable segments."); 3439 3440/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3441 * desired address is already mapped, or if we successfully grow 3442 * the stack. Also returns KERN_SUCCESS if addr is outside the 3443 * stack range (this is strange, but preserves compatibility with 3444 * the grow function in vm_machdep.c). 3445 */ 3446int 3447vm_map_growstack(struct proc *p, vm_offset_t addr) 3448{ 3449 vm_map_entry_t next_entry, prev_entry; 3450 vm_map_entry_t new_entry, stack_entry; 3451 struct vmspace *vm = p->p_vmspace; 3452 vm_map_t map = &vm->vm_map; 3453 vm_offset_t end; 3454 vm_size_t growsize; 3455 size_t grow_amount, max_grow; 3456 rlim_t lmemlim, stacklim, vmemlim; 3457 int is_procstack, rv; 3458 struct ucred *cred; 3459#ifdef notyet 3460 uint64_t limit; 3461#endif 3462#ifdef RACCT 3463 int error; 3464#endif 3465 3466Retry: 3467 PROC_LOCK(p); 3468 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3469 stacklim = lim_cur(p, RLIMIT_STACK); 3470 vmemlim = lim_cur(p, RLIMIT_VMEM); 3471 PROC_UNLOCK(p); 3472 3473 vm_map_lock_read(map); 3474 3475 /* If addr is already in the entry range, no need to grow.*/ 3476 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3477 vm_map_unlock_read(map); 3478 return (KERN_SUCCESS); 3479 } 3480 3481 next_entry = prev_entry->next; 3482 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3483 /* 3484 * This entry does not grow upwards. Since the address lies 3485 * beyond this entry, the next entry (if one exists) has to 3486 * be a downward growable entry. The entry list header is 3487 * never a growable entry, so it suffices to check the flags. 3488 */ 3489 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3490 vm_map_unlock_read(map); 3491 return (KERN_SUCCESS); 3492 } 3493 stack_entry = next_entry; 3494 } else { 3495 /* 3496 * This entry grows upward. If the next entry does not at 3497 * least grow downwards, this is the entry we need to grow. 3498 * otherwise we have two possible choices and we have to 3499 * select one. 3500 */ 3501 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3502 /* 3503 * We have two choices; grow the entry closest to 3504 * the address to minimize the amount of growth. 3505 */ 3506 if (addr - prev_entry->end <= next_entry->start - addr) 3507 stack_entry = prev_entry; 3508 else 3509 stack_entry = next_entry; 3510 } else 3511 stack_entry = prev_entry; 3512 } 3513 3514 if (stack_entry == next_entry) { 3515 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3516 KASSERT(addr < stack_entry->start, ("foo")); 3517 end = (prev_entry != &map->header) ? prev_entry->end : 3518 stack_entry->start - stack_entry->avail_ssize; 3519 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3520 max_grow = stack_entry->start - end; 3521 } else { 3522 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3523 KASSERT(addr >= stack_entry->end, ("foo")); 3524 end = (next_entry != &map->header) ? next_entry->start : 3525 stack_entry->end + stack_entry->avail_ssize; 3526 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3527 max_grow = end - stack_entry->end; 3528 } 3529 3530 if (grow_amount > stack_entry->avail_ssize) { 3531 vm_map_unlock_read(map); 3532 return (KERN_NO_SPACE); 3533 } 3534 3535 /* 3536 * If there is no longer enough space between the entries nogo, and 3537 * adjust the available space. Note: this should only happen if the 3538 * user has mapped into the stack area after the stack was created, 3539 * and is probably an error. 3540 * 3541 * This also effectively destroys any guard page the user might have 3542 * intended by limiting the stack size. 3543 */ 3544 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3545 if (vm_map_lock_upgrade(map)) 3546 goto Retry; 3547 3548 stack_entry->avail_ssize = max_grow; 3549 3550 vm_map_unlock(map); 3551 return (KERN_NO_SPACE); 3552 } 3553 3554 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3555 3556 /* 3557 * If this is the main process stack, see if we're over the stack 3558 * limit. 3559 */ 3560 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3561 vm_map_unlock_read(map); 3562 return (KERN_NO_SPACE); 3563 } 3564#ifdef RACCT 3565 PROC_LOCK(p); 3566 if (is_procstack && 3567 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3568 PROC_UNLOCK(p); 3569 vm_map_unlock_read(map); 3570 return (KERN_NO_SPACE); 3571 } 3572 PROC_UNLOCK(p); 3573#endif 3574 3575 /* Round up the grow amount modulo sgrowsiz */ 3576 growsize = sgrowsiz; 3577 grow_amount = roundup(grow_amount, growsize); 3578 if (grow_amount > stack_entry->avail_ssize) 3579 grow_amount = stack_entry->avail_ssize; 3580 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3581 grow_amount = trunc_page((vm_size_t)stacklim) - 3582 ctob(vm->vm_ssize); 3583 } 3584#ifdef notyet 3585 PROC_LOCK(p); 3586 limit = racct_get_available(p, RACCT_STACK); 3587 PROC_UNLOCK(p); 3588 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3589 grow_amount = limit - ctob(vm->vm_ssize); 3590#endif 3591 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3592 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3593 vm_map_unlock_read(map); 3594 rv = KERN_NO_SPACE; 3595 goto out; 3596 } 3597#ifdef RACCT 3598 PROC_LOCK(p); 3599 if (racct_set(p, RACCT_MEMLOCK, 3600 ptoa(pmap_wired_count(map->pmap)) + 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 we would blow our VMEM resource limit, no go */ 3610 if (map->size + grow_amount > vmemlim) { 3611 vm_map_unlock_read(map); 3612 rv = KERN_NO_SPACE; 3613 goto out; 3614 } 3615#ifdef RACCT 3616 PROC_LOCK(p); 3617 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3618 PROC_UNLOCK(p); 3619 vm_map_unlock_read(map); 3620 rv = KERN_NO_SPACE; 3621 goto out; 3622 } 3623 PROC_UNLOCK(p); 3624#endif 3625 3626 if (vm_map_lock_upgrade(map)) 3627 goto Retry; 3628 3629 if (stack_entry == next_entry) { 3630 /* 3631 * Growing downward. 3632 */ 3633 /* Get the preliminary new entry start value */ 3634 addr = stack_entry->start - grow_amount; 3635 3636 /* 3637 * If this puts us into the previous entry, cut back our 3638 * growth to the available space. Also, see the note above. 3639 */ 3640 if (addr < end) { 3641 stack_entry->avail_ssize = max_grow; 3642 addr = end; 3643 if (stack_guard_page) 3644 addr += PAGE_SIZE; 3645 } 3646 3647 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3648 next_entry->protection, next_entry->max_protection, 0); 3649 3650 /* Adjust the available stack space by the amount we grew. */ 3651 if (rv == KERN_SUCCESS) { 3652 if (prev_entry != &map->header) 3653 vm_map_clip_end(map, prev_entry, addr); 3654 new_entry = prev_entry->next; 3655 KASSERT(new_entry == stack_entry->prev, ("foo")); 3656 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3657 KASSERT(new_entry->start == addr, ("foo")); 3658 grow_amount = new_entry->end - new_entry->start; 3659 new_entry->avail_ssize = stack_entry->avail_ssize - 3660 grow_amount; 3661 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3662 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3663 } 3664 } else { 3665 /* 3666 * Growing upward. 3667 */ 3668 addr = stack_entry->end + grow_amount; 3669 3670 /* 3671 * If this puts us into the next entry, cut back our growth 3672 * to the available space. Also, see the note above. 3673 */ 3674 if (addr > end) { 3675 stack_entry->avail_ssize = end - stack_entry->end; 3676 addr = end; 3677 if (stack_guard_page) 3678 addr -= PAGE_SIZE; 3679 } 3680 3681 grow_amount = addr - stack_entry->end; 3682 cred = stack_entry->cred; 3683 if (cred == NULL && stack_entry->object.vm_object != NULL) 3684 cred = stack_entry->object.vm_object->cred; 3685 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3686 rv = KERN_NO_SPACE; 3687 /* Grow the underlying object if applicable. */ 3688 else if (stack_entry->object.vm_object == NULL || 3689 vm_object_coalesce(stack_entry->object.vm_object, 3690 stack_entry->offset, 3691 (vm_size_t)(stack_entry->end - stack_entry->start), 3692 (vm_size_t)grow_amount, cred != NULL)) { 3693 map->size += (addr - stack_entry->end); 3694 /* Update the current entry. */ 3695 stack_entry->end = addr; 3696 stack_entry->avail_ssize -= grow_amount; 3697 vm_map_entry_resize_free(map, stack_entry); 3698 rv = KERN_SUCCESS; 3699 3700 if (next_entry != &map->header) 3701 vm_map_clip_start(map, next_entry, addr); 3702 } else 3703 rv = KERN_FAILURE; 3704 } 3705 3706 if (rv == KERN_SUCCESS && is_procstack) 3707 vm->vm_ssize += btoc(grow_amount); 3708 3709 vm_map_unlock(map); 3710 3711 /* 3712 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3713 */ 3714 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3715 vm_map_wire(map, 3716 (stack_entry == next_entry) ? addr : addr - grow_amount, 3717 (stack_entry == next_entry) ? stack_entry->start : addr, 3718 (p->p_flag & P_SYSTEM) 3719 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3720 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3721 } 3722 3723out: 3724#ifdef RACCT 3725 if (rv != KERN_SUCCESS) { 3726 PROC_LOCK(p); 3727 error = racct_set(p, RACCT_VMEM, map->size); 3728 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3729 if (!old_mlock) { 3730 error = racct_set(p, RACCT_MEMLOCK, 3731 ptoa(pmap_wired_count(map->pmap))); 3732 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3733 } 3734 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3735 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3736 PROC_UNLOCK(p); 3737 } 3738#endif 3739 3740 return (rv); 3741} 3742 3743/* 3744 * Unshare the specified VM space for exec. If other processes are 3745 * mapped to it, then create a new one. The new vmspace is null. 3746 */ 3747int 3748vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3749{ 3750 struct vmspace *oldvmspace = p->p_vmspace; 3751 struct vmspace *newvmspace; 3752 3753 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3754 if (newvmspace == NULL) 3755 return (ENOMEM); 3756 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3757 /* 3758 * This code is written like this for prototype purposes. The 3759 * goal is to avoid running down the vmspace here, but let the 3760 * other process's that are still using the vmspace to finally 3761 * run it down. Even though there is little or no chance of blocking 3762 * here, it is a good idea to keep this form for future mods. 3763 */ 3764 PROC_VMSPACE_LOCK(p); 3765 p->p_vmspace = newvmspace; 3766 PROC_VMSPACE_UNLOCK(p); 3767 if (p == curthread->td_proc) 3768 pmap_activate(curthread); 3769 vmspace_free(oldvmspace); 3770 return (0); 3771} 3772 3773/* 3774 * Unshare the specified VM space for forcing COW. This 3775 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3776 */ 3777int 3778vmspace_unshare(struct proc *p) 3779{ 3780 struct vmspace *oldvmspace = p->p_vmspace; 3781 struct vmspace *newvmspace; 3782 vm_ooffset_t fork_charge; 3783 3784 if (oldvmspace->vm_refcnt == 1) 3785 return (0); 3786 fork_charge = 0; 3787 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3788 if (newvmspace == NULL) 3789 return (ENOMEM); 3790 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3791 vmspace_free(newvmspace); 3792 return (ENOMEM); 3793 } 3794 PROC_VMSPACE_LOCK(p); 3795 p->p_vmspace = newvmspace; 3796 PROC_VMSPACE_UNLOCK(p); 3797 if (p == curthread->td_proc) 3798 pmap_activate(curthread); 3799 vmspace_free(oldvmspace); 3800 return (0); 3801} 3802 3803/* 3804 * vm_map_lookup: 3805 * 3806 * Finds the VM object, offset, and 3807 * protection for a given virtual address in the 3808 * specified map, assuming a page fault of the 3809 * type specified. 3810 * 3811 * Leaves the map in question locked for read; return 3812 * values are guaranteed until a vm_map_lookup_done 3813 * call is performed. Note that the map argument 3814 * is in/out; the returned map must be used in 3815 * the call to vm_map_lookup_done. 3816 * 3817 * A handle (out_entry) is returned for use in 3818 * vm_map_lookup_done, to make that fast. 3819 * 3820 * If a lookup is requested with "write protection" 3821 * specified, the map may be changed to perform virtual 3822 * copying operations, although the data referenced will 3823 * remain the same. 3824 */ 3825int 3826vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3827 vm_offset_t vaddr, 3828 vm_prot_t fault_typea, 3829 vm_map_entry_t *out_entry, /* OUT */ 3830 vm_object_t *object, /* OUT */ 3831 vm_pindex_t *pindex, /* OUT */ 3832 vm_prot_t *out_prot, /* OUT */ 3833 boolean_t *wired) /* OUT */ 3834{ 3835 vm_map_entry_t entry; 3836 vm_map_t map = *var_map; 3837 vm_prot_t prot; 3838 vm_prot_t fault_type = fault_typea; 3839 vm_object_t eobject; 3840 vm_size_t size; 3841 struct ucred *cred; 3842 3843RetryLookup:; 3844 3845 vm_map_lock_read(map); 3846 3847 /* 3848 * Lookup the faulting address. 3849 */ 3850 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3851 vm_map_unlock_read(map); 3852 return (KERN_INVALID_ADDRESS); 3853 } 3854 3855 entry = *out_entry; 3856 3857 /* 3858 * Handle submaps. 3859 */ 3860 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3861 vm_map_t old_map = map; 3862 3863 *var_map = map = entry->object.sub_map; 3864 vm_map_unlock_read(old_map); 3865 goto RetryLookup; 3866 } 3867 3868 /* 3869 * Check whether this task is allowed to have this page. 3870 */ 3871 prot = entry->protection; 3872 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3873 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3874 vm_map_unlock_read(map); 3875 return (KERN_PROTECTION_FAILURE); 3876 } 3877 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3878 (entry->eflags & MAP_ENTRY_COW) && 3879 (fault_type & VM_PROT_WRITE)) { 3880 vm_map_unlock_read(map); 3881 return (KERN_PROTECTION_FAILURE); 3882 } 3883 if ((fault_typea & VM_PROT_COPY) != 0 && 3884 (entry->max_protection & VM_PROT_WRITE) == 0 && 3885 (entry->eflags & MAP_ENTRY_COW) == 0) { 3886 vm_map_unlock_read(map); 3887 return (KERN_PROTECTION_FAILURE); 3888 } 3889 3890 /* 3891 * If this page is not pageable, we have to get it for all possible 3892 * accesses. 3893 */ 3894 *wired = (entry->wired_count != 0); 3895 if (*wired) 3896 fault_type = entry->protection; 3897 size = entry->end - entry->start; 3898 /* 3899 * If the entry was copy-on-write, we either ... 3900 */ 3901 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3902 /* 3903 * If we want to write the page, we may as well handle that 3904 * now since we've got the map locked. 3905 * 3906 * If we don't need to write the page, we just demote the 3907 * permissions allowed. 3908 */ 3909 if ((fault_type & VM_PROT_WRITE) != 0 || 3910 (fault_typea & VM_PROT_COPY) != 0) { 3911 /* 3912 * Make a new object, and place it in the object 3913 * chain. Note that no new references have appeared 3914 * -- one just moved from the map to the new 3915 * object. 3916 */ 3917 if (vm_map_lock_upgrade(map)) 3918 goto RetryLookup; 3919 3920 if (entry->cred == NULL) { 3921 /* 3922 * The debugger owner is charged for 3923 * the memory. 3924 */ 3925 cred = curthread->td_ucred; 3926 crhold(cred); 3927 if (!swap_reserve_by_cred(size, cred)) { 3928 crfree(cred); 3929 vm_map_unlock(map); 3930 return (KERN_RESOURCE_SHORTAGE); 3931 } 3932 entry->cred = cred; 3933 } 3934 vm_object_shadow(&entry->object.vm_object, 3935 &entry->offset, size); 3936 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3937 eobject = entry->object.vm_object; 3938 if (eobject->cred != NULL) { 3939 /* 3940 * The object was not shadowed. 3941 */ 3942 swap_release_by_cred(size, entry->cred); 3943 crfree(entry->cred); 3944 entry->cred = NULL; 3945 } else if (entry->cred != NULL) { 3946 VM_OBJECT_WLOCK(eobject); 3947 eobject->cred = entry->cred; 3948 eobject->charge = size; 3949 VM_OBJECT_WUNLOCK(eobject); 3950 entry->cred = NULL; 3951 } 3952 3953 vm_map_lock_downgrade(map); 3954 } else { 3955 /* 3956 * We're attempting to read a copy-on-write page -- 3957 * don't allow writes. 3958 */ 3959 prot &= ~VM_PROT_WRITE; 3960 } 3961 } 3962 3963 /* 3964 * Create an object if necessary. 3965 */ 3966 if (entry->object.vm_object == NULL && 3967 !map->system_map) { 3968 if (vm_map_lock_upgrade(map)) 3969 goto RetryLookup; 3970 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3971 atop(size)); 3972 entry->offset = 0; 3973 if (entry->cred != NULL) { 3974 VM_OBJECT_WLOCK(entry->object.vm_object); 3975 entry->object.vm_object->cred = entry->cred; 3976 entry->object.vm_object->charge = size; 3977 VM_OBJECT_WUNLOCK(entry->object.vm_object); 3978 entry->cred = NULL; 3979 } 3980 vm_map_lock_downgrade(map); 3981 } 3982 3983 /* 3984 * Return the object/offset from this entry. If the entry was 3985 * copy-on-write or empty, it has been fixed up. 3986 */ 3987 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3988 *object = entry->object.vm_object; 3989 3990 *out_prot = prot; 3991 return (KERN_SUCCESS); 3992} 3993 3994/* 3995 * vm_map_lookup_locked: 3996 * 3997 * Lookup the faulting address. A version of vm_map_lookup that returns 3998 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3999 */ 4000int 4001vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4002 vm_offset_t vaddr, 4003 vm_prot_t fault_typea, 4004 vm_map_entry_t *out_entry, /* OUT */ 4005 vm_object_t *object, /* OUT */ 4006 vm_pindex_t *pindex, /* OUT */ 4007 vm_prot_t *out_prot, /* OUT */ 4008 boolean_t *wired) /* OUT */ 4009{ 4010 vm_map_entry_t entry; 4011 vm_map_t map = *var_map; 4012 vm_prot_t prot; 4013 vm_prot_t fault_type = fault_typea; 4014 4015 /* 4016 * Lookup the faulting address. 4017 */ 4018 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4019 return (KERN_INVALID_ADDRESS); 4020 4021 entry = *out_entry; 4022 4023 /* 4024 * Fail if the entry refers to a submap. 4025 */ 4026 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4027 return (KERN_FAILURE); 4028 4029 /* 4030 * Check whether this task is allowed to have this page. 4031 */ 4032 prot = entry->protection; 4033 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4034 if ((fault_type & prot) != fault_type) 4035 return (KERN_PROTECTION_FAILURE); 4036 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 4037 (entry->eflags & MAP_ENTRY_COW) && 4038 (fault_type & VM_PROT_WRITE)) 4039 return (KERN_PROTECTION_FAILURE); 4040 4041 /* 4042 * If this page is not pageable, we have to get it for all possible 4043 * accesses. 4044 */ 4045 *wired = (entry->wired_count != 0); 4046 if (*wired) 4047 fault_type = entry->protection; 4048 4049 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4050 /* 4051 * Fail if the entry was copy-on-write for a write fault. 4052 */ 4053 if (fault_type & VM_PROT_WRITE) 4054 return (KERN_FAILURE); 4055 /* 4056 * We're attempting to read a copy-on-write page -- 4057 * don't allow writes. 4058 */ 4059 prot &= ~VM_PROT_WRITE; 4060 } 4061 4062 /* 4063 * Fail if an object should be created. 4064 */ 4065 if (entry->object.vm_object == NULL && !map->system_map) 4066 return (KERN_FAILURE); 4067 4068 /* 4069 * Return the object/offset from this entry. If the entry was 4070 * copy-on-write or empty, it has been fixed up. 4071 */ 4072 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4073 *object = entry->object.vm_object; 4074 4075 *out_prot = prot; 4076 return (KERN_SUCCESS); 4077} 4078 4079/* 4080 * vm_map_lookup_done: 4081 * 4082 * Releases locks acquired by a vm_map_lookup 4083 * (according to the handle returned by that lookup). 4084 */ 4085void 4086vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4087{ 4088 /* 4089 * Unlock the main-level map 4090 */ 4091 vm_map_unlock_read(map); 4092} 4093 4094#include "opt_ddb.h" 4095#ifdef DDB 4096#include <sys/kernel.h> 4097 4098#include <ddb/ddb.h> 4099 4100static void 4101vm_map_print(vm_map_t map) 4102{ 4103 vm_map_entry_t entry; 4104 4105 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4106 (void *)map, 4107 (void *)map->pmap, map->nentries, map->timestamp); 4108 4109 db_indent += 2; 4110 for (entry = map->header.next; entry != &map->header; 4111 entry = entry->next) { 4112 db_iprintf("map entry %p: start=%p, end=%p\n", 4113 (void *)entry, (void *)entry->start, (void *)entry->end); 4114 { 4115 static char *inheritance_name[4] = 4116 {"share", "copy", "none", "donate_copy"}; 4117 4118 db_iprintf(" prot=%x/%x/%s", 4119 entry->protection, 4120 entry->max_protection, 4121 inheritance_name[(int)(unsigned char)entry->inheritance]); 4122 if (entry->wired_count != 0) 4123 db_printf(", wired"); 4124 } 4125 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4126 db_printf(", share=%p, offset=0x%jx\n", 4127 (void *)entry->object.sub_map, 4128 (uintmax_t)entry->offset); 4129 if ((entry->prev == &map->header) || 4130 (entry->prev->object.sub_map != 4131 entry->object.sub_map)) { 4132 db_indent += 2; 4133 vm_map_print((vm_map_t)entry->object.sub_map); 4134 db_indent -= 2; 4135 } 4136 } else { 4137 if (entry->cred != NULL) 4138 db_printf(", ruid %d", entry->cred->cr_ruid); 4139 db_printf(", object=%p, offset=0x%jx", 4140 (void *)entry->object.vm_object, 4141 (uintmax_t)entry->offset); 4142 if (entry->object.vm_object && entry->object.vm_object->cred) 4143 db_printf(", obj ruid %d charge %jx", 4144 entry->object.vm_object->cred->cr_ruid, 4145 (uintmax_t)entry->object.vm_object->charge); 4146 if (entry->eflags & MAP_ENTRY_COW) 4147 db_printf(", copy (%s)", 4148 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4149 db_printf("\n"); 4150 4151 if ((entry->prev == &map->header) || 4152 (entry->prev->object.vm_object != 4153 entry->object.vm_object)) { 4154 db_indent += 2; 4155 vm_object_print((db_expr_t)(intptr_t) 4156 entry->object.vm_object, 4157 0, 0, (char *)0); 4158 db_indent -= 2; 4159 } 4160 } 4161 } 4162 db_indent -= 2; 4163} 4164 4165DB_SHOW_COMMAND(map, map) 4166{ 4167 4168 if (!have_addr) { 4169 db_printf("usage: show map <addr>\n"); 4170 return; 4171 } 4172 vm_map_print((vm_map_t)addr); 4173} 4174 4175DB_SHOW_COMMAND(procvm, procvm) 4176{ 4177 struct proc *p; 4178 4179 if (have_addr) { 4180 p = (struct proc *) addr; 4181 } else { 4182 p = curproc; 4183 } 4184 4185 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4186 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4187 (void *)vmspace_pmap(p->p_vmspace)); 4188 4189 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4190} 4191 4192#endif /* DDB */ 4193