vm_map.c revision 312397
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 312397 2017-01-18 22:40:22Z jhb $"); 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); 135static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry); 136#ifdef INVARIANTS 137static void vm_map_zdtor(void *mem, int size, void *arg); 138static void vmspace_zdtor(void *mem, int size, void *arg); 139#endif 140static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, 141 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, 142 int cow); 143static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 144 vm_offset_t failed_addr); 145 146#define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 147 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 148 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 149 150/* 151 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 152 * stable. 153 */ 154#define PROC_VMSPACE_LOCK(p) do { } while (0) 155#define PROC_VMSPACE_UNLOCK(p) do { } while (0) 156 157/* 158 * VM_MAP_RANGE_CHECK: [ internal use only ] 159 * 160 * Asserts that the starting and ending region 161 * addresses fall within the valid range of the map. 162 */ 163#define VM_MAP_RANGE_CHECK(map, start, end) \ 164 { \ 165 if (start < vm_map_min(map)) \ 166 start = vm_map_min(map); \ 167 if (end > vm_map_max(map)) \ 168 end = vm_map_max(map); \ 169 if (start > end) \ 170 start = end; \ 171 } 172 173/* 174 * vm_map_startup: 175 * 176 * Initialize the vm_map module. Must be called before 177 * any other vm_map routines. 178 * 179 * Map and entry structures are allocated from the general 180 * purpose memory pool with some exceptions: 181 * 182 * - The kernel map and kmem submap are allocated statically. 183 * - Kernel map entries are allocated out of a static pool. 184 * 185 * These restrictions are necessary since malloc() uses the 186 * maps and requires map entries. 187 */ 188 189void 190vm_map_startup(void) 191{ 192 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 193 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 194#ifdef INVARIANTS 195 vm_map_zdtor, 196#else 197 NULL, 198#endif 199 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 200 uma_prealloc(mapzone, MAX_KMAP); 201 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 203 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 204 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 205 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 206 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 207#ifdef INVARIANTS 208 vmspace_zdtor, 209#else 210 NULL, 211#endif 212 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 213} 214 215static int 216vmspace_zinit(void *mem, int size, int flags) 217{ 218 struct vmspace *vm; 219 220 vm = (struct vmspace *)mem; 221 222 vm->vm_map.pmap = NULL; 223 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 224 PMAP_LOCK_INIT(vmspace_pmap(vm)); 225 return (0); 226} 227 228static int 229vm_map_zinit(void *mem, int size, int flags) 230{ 231 vm_map_t map; 232 233 map = (vm_map_t)mem; 234 memset(map, 0, sizeof(*map)); 235 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 236 sx_init(&map->lock, "vm map (user)"); 237 return (0); 238} 239 240#ifdef INVARIANTS 241static void 242vmspace_zdtor(void *mem, int size, void *arg) 243{ 244 struct vmspace *vm; 245 246 vm = (struct vmspace *)mem; 247 248 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 249} 250static void 251vm_map_zdtor(void *mem, int size, void *arg) 252{ 253 vm_map_t map; 254 255 map = (vm_map_t)mem; 256 KASSERT(map->nentries == 0, 257 ("map %p nentries == %d on free.", 258 map, map->nentries)); 259 KASSERT(map->size == 0, 260 ("map %p size == %lu on free.", 261 map, (unsigned long)map->size)); 262} 263#endif /* INVARIANTS */ 264 265/* 266 * Allocate a vmspace structure, including a vm_map and pmap, 267 * and initialize those structures. The refcnt is set to 1. 268 * 269 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 270 */ 271struct vmspace * 272vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 273{ 274 struct vmspace *vm; 275 276 vm = uma_zalloc(vmspace_zone, M_WAITOK); 277 278 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 279 280 if (pinit == NULL) 281 pinit = &pmap_pinit; 282 283 if (!pinit(vmspace_pmap(vm))) { 284 uma_zfree(vmspace_zone, vm); 285 return (NULL); 286 } 287 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 288 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 289 vm->vm_refcnt = 1; 290 vm->vm_shm = NULL; 291 vm->vm_swrss = 0; 292 vm->vm_tsize = 0; 293 vm->vm_dsize = 0; 294 vm->vm_ssize = 0; 295 vm->vm_taddr = 0; 296 vm->vm_daddr = 0; 297 vm->vm_maxsaddr = 0; 298 return (vm); 299} 300 301#ifdef RACCT 302static void 303vmspace_container_reset(struct proc *p) 304{ 305 306 PROC_LOCK(p); 307 racct_set(p, RACCT_DATA, 0); 308 racct_set(p, RACCT_STACK, 0); 309 racct_set(p, RACCT_RSS, 0); 310 racct_set(p, RACCT_MEMLOCK, 0); 311 racct_set(p, RACCT_VMEM, 0); 312 PROC_UNLOCK(p); 313} 314#endif 315 316static inline void 317vmspace_dofree(struct vmspace *vm) 318{ 319 320 CTR1(KTR_VM, "vmspace_free: %p", vm); 321 322 /* 323 * Make sure any SysV shm is freed, it might not have been in 324 * exit1(). 325 */ 326 shmexit(vm); 327 328 /* 329 * Lock the map, to wait out all other references to it. 330 * Delete all of the mappings and pages they hold, then call 331 * the pmap module to reclaim anything left. 332 */ 333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 334 vm->vm_map.max_offset); 335 336 pmap_release(vmspace_pmap(vm)); 337 vm->vm_map.pmap = NULL; 338 uma_zfree(vmspace_zone, vm); 339} 340 341void 342vmspace_free(struct vmspace *vm) 343{ 344 345 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 346 "vmspace_free() called with non-sleepable lock held"); 347 348 if (vm->vm_refcnt == 0) 349 panic("vmspace_free: attempt to free already freed vmspace"); 350 351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 352 vmspace_dofree(vm); 353} 354 355void 356vmspace_exitfree(struct proc *p) 357{ 358 struct vmspace *vm; 359 360 PROC_VMSPACE_LOCK(p); 361 vm = p->p_vmspace; 362 p->p_vmspace = NULL; 363 PROC_VMSPACE_UNLOCK(p); 364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 365 vmspace_free(vm); 366} 367 368void 369vmspace_exit(struct thread *td) 370{ 371 int refcnt; 372 struct vmspace *vm; 373 struct proc *p; 374 375 /* 376 * Release user portion of address space. 377 * This releases references to vnodes, 378 * which could cause I/O if the file has been unlinked. 379 * Need to do this early enough that we can still sleep. 380 * 381 * The last exiting process to reach this point releases as 382 * much of the environment as it can. vmspace_dofree() is the 383 * slower fallback in case another process had a temporary 384 * reference to the vmspace. 385 */ 386 387 p = td->td_proc; 388 vm = p->p_vmspace; 389 atomic_add_int(&vmspace0.vm_refcnt, 1); 390 do { 391 refcnt = vm->vm_refcnt; 392 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 393 /* Switch now since other proc might free vmspace */ 394 PROC_VMSPACE_LOCK(p); 395 p->p_vmspace = &vmspace0; 396 PROC_VMSPACE_UNLOCK(p); 397 pmap_activate(td); 398 } 399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 400 if (refcnt == 1) { 401 if (p->p_vmspace != vm) { 402 /* vmspace not yet freed, switch back */ 403 PROC_VMSPACE_LOCK(p); 404 p->p_vmspace = vm; 405 PROC_VMSPACE_UNLOCK(p); 406 pmap_activate(td); 407 } 408 pmap_remove_pages(vmspace_pmap(vm)); 409 /* Switch now since this proc will free vmspace */ 410 PROC_VMSPACE_LOCK(p); 411 p->p_vmspace = &vmspace0; 412 PROC_VMSPACE_UNLOCK(p); 413 pmap_activate(td); 414 vmspace_dofree(vm); 415 } 416#ifdef RACCT 417 if (racct_enable) 418 vmspace_container_reset(p); 419#endif 420} 421 422/* Acquire reference to vmspace owned by another process. */ 423 424struct vmspace * 425vmspace_acquire_ref(struct proc *p) 426{ 427 struct vmspace *vm; 428 int refcnt; 429 430 PROC_VMSPACE_LOCK(p); 431 vm = p->p_vmspace; 432 if (vm == NULL) { 433 PROC_VMSPACE_UNLOCK(p); 434 return (NULL); 435 } 436 do { 437 refcnt = vm->vm_refcnt; 438 if (refcnt <= 0) { /* Avoid 0->1 transition */ 439 PROC_VMSPACE_UNLOCK(p); 440 return (NULL); 441 } 442 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 443 if (vm != p->p_vmspace) { 444 PROC_VMSPACE_UNLOCK(p); 445 vmspace_free(vm); 446 return (NULL); 447 } 448 PROC_VMSPACE_UNLOCK(p); 449 return (vm); 450} 451 452void 453_vm_map_lock(vm_map_t map, const char *file, int line) 454{ 455 456 if (map->system_map) 457 mtx_lock_flags_(&map->system_mtx, 0, file, line); 458 else 459 sx_xlock_(&map->lock, file, line); 460 map->timestamp++; 461} 462 463static void 464vm_map_process_deferred(void) 465{ 466 struct thread *td; 467 vm_map_entry_t entry, next; 468 vm_object_t object; 469 470 td = curthread; 471 entry = td->td_map_def_user; 472 td->td_map_def_user = NULL; 473 while (entry != NULL) { 474 next = entry->next; 475 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 476 /* 477 * Decrement the object's writemappings and 478 * possibly the vnode's v_writecount. 479 */ 480 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 481 ("Submap with writecount")); 482 object = entry->object.vm_object; 483 KASSERT(object != NULL, ("No object for writecount")); 484 vnode_pager_release_writecount(object, entry->start, 485 entry->end); 486 } 487 vm_map_entry_deallocate(entry, FALSE); 488 entry = next; 489 } 490} 491 492void 493_vm_map_unlock(vm_map_t map, const char *file, int line) 494{ 495 496 if (map->system_map) 497 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 498 else { 499 sx_xunlock_(&map->lock, file, line); 500 vm_map_process_deferred(); 501 } 502} 503 504void 505_vm_map_lock_read(vm_map_t map, const char *file, int line) 506{ 507 508 if (map->system_map) 509 mtx_lock_flags_(&map->system_mtx, 0, file, line); 510 else 511 sx_slock_(&map->lock, file, line); 512} 513 514void 515_vm_map_unlock_read(vm_map_t map, const char *file, int line) 516{ 517 518 if (map->system_map) 519 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 520 else { 521 sx_sunlock_(&map->lock, file, line); 522 vm_map_process_deferred(); 523 } 524} 525 526int 527_vm_map_trylock(vm_map_t map, const char *file, int line) 528{ 529 int error; 530 531 error = map->system_map ? 532 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 533 !sx_try_xlock_(&map->lock, file, line); 534 if (error == 0) 535 map->timestamp++; 536 return (error == 0); 537} 538 539int 540_vm_map_trylock_read(vm_map_t map, const char *file, int line) 541{ 542 int error; 543 544 error = map->system_map ? 545 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 546 !sx_try_slock_(&map->lock, file, line); 547 return (error == 0); 548} 549 550/* 551 * _vm_map_lock_upgrade: [ internal use only ] 552 * 553 * Tries to upgrade a read (shared) lock on the specified map to a write 554 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 555 * non-zero value if the upgrade fails. If the upgrade fails, the map is 556 * returned without a read or write lock held. 557 * 558 * Requires that the map be read locked. 559 */ 560int 561_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 562{ 563 unsigned int last_timestamp; 564 565 if (map->system_map) { 566 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 567 } else { 568 if (!sx_try_upgrade_(&map->lock, file, line)) { 569 last_timestamp = map->timestamp; 570 sx_sunlock_(&map->lock, file, line); 571 vm_map_process_deferred(); 572 /* 573 * If the map's timestamp does not change while the 574 * map is unlocked, then the upgrade succeeds. 575 */ 576 sx_xlock_(&map->lock, file, line); 577 if (last_timestamp != map->timestamp) { 578 sx_xunlock_(&map->lock, file, line); 579 return (1); 580 } 581 } 582 } 583 map->timestamp++; 584 return (0); 585} 586 587void 588_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 589{ 590 591 if (map->system_map) { 592 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 593 } else 594 sx_downgrade_(&map->lock, file, line); 595} 596 597/* 598 * vm_map_locked: 599 * 600 * Returns a non-zero value if the caller holds a write (exclusive) lock 601 * on the specified map and the value "0" otherwise. 602 */ 603int 604vm_map_locked(vm_map_t map) 605{ 606 607 if (map->system_map) 608 return (mtx_owned(&map->system_mtx)); 609 else 610 return (sx_xlocked(&map->lock)); 611} 612 613#ifdef INVARIANTS 614static void 615_vm_map_assert_locked(vm_map_t map, const char *file, int line) 616{ 617 618 if (map->system_map) 619 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 620 else 621 sx_assert_(&map->lock, SA_XLOCKED, file, line); 622} 623 624#define VM_MAP_ASSERT_LOCKED(map) \ 625 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 626#else 627#define VM_MAP_ASSERT_LOCKED(map) 628#endif 629 630/* 631 * _vm_map_unlock_and_wait: 632 * 633 * Atomically releases the lock on the specified map and puts the calling 634 * thread to sleep. The calling thread will remain asleep until either 635 * vm_map_wakeup() is performed on the map or the specified timeout is 636 * exceeded. 637 * 638 * WARNING! This function does not perform deferred deallocations of 639 * objects and map entries. Therefore, the calling thread is expected to 640 * reacquire the map lock after reawakening and later perform an ordinary 641 * unlock operation, such as vm_map_unlock(), before completing its 642 * operation on the map. 643 */ 644int 645_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 646{ 647 648 mtx_lock(&map_sleep_mtx); 649 if (map->system_map) 650 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 651 else 652 sx_xunlock_(&map->lock, file, line); 653 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 654 timo)); 655} 656 657/* 658 * vm_map_wakeup: 659 * 660 * Awaken any threads that have slept on the map using 661 * vm_map_unlock_and_wait(). 662 */ 663void 664vm_map_wakeup(vm_map_t map) 665{ 666 667 /* 668 * Acquire and release map_sleep_mtx to prevent a wakeup() 669 * from being performed (and lost) between the map unlock 670 * and the msleep() in _vm_map_unlock_and_wait(). 671 */ 672 mtx_lock(&map_sleep_mtx); 673 mtx_unlock(&map_sleep_mtx); 674 wakeup(&map->root); 675} 676 677void 678vm_map_busy(vm_map_t map) 679{ 680 681 VM_MAP_ASSERT_LOCKED(map); 682 map->busy++; 683} 684 685void 686vm_map_unbusy(vm_map_t map) 687{ 688 689 VM_MAP_ASSERT_LOCKED(map); 690 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 691 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 692 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 693 wakeup(&map->busy); 694 } 695} 696 697void 698vm_map_wait_busy(vm_map_t map) 699{ 700 701 VM_MAP_ASSERT_LOCKED(map); 702 while (map->busy) { 703 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 704 if (map->system_map) 705 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 706 else 707 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 708 } 709 map->timestamp++; 710} 711 712long 713vmspace_resident_count(struct vmspace *vmspace) 714{ 715 return pmap_resident_count(vmspace_pmap(vmspace)); 716} 717 718/* 719 * vm_map_create: 720 * 721 * Creates and returns a new empty VM map with 722 * the given physical map structure, and having 723 * the given lower and upper address bounds. 724 */ 725vm_map_t 726vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 727{ 728 vm_map_t result; 729 730 result = uma_zalloc(mapzone, M_WAITOK); 731 CTR1(KTR_VM, "vm_map_create: %p", result); 732 _vm_map_init(result, pmap, min, max); 733 return (result); 734} 735 736/* 737 * Initialize an existing vm_map structure 738 * such as that in the vmspace structure. 739 */ 740static void 741_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 742{ 743 744 map->header.next = map->header.prev = &map->header; 745 map->needs_wakeup = FALSE; 746 map->system_map = 0; 747 map->pmap = pmap; 748 map->min_offset = min; 749 map->max_offset = max; 750 map->flags = 0; 751 map->root = NULL; 752 map->timestamp = 0; 753 map->busy = 0; 754} 755 756void 757vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 758{ 759 760 _vm_map_init(map, pmap, min, max); 761 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 762 sx_init(&map->lock, "user map"); 763} 764 765/* 766 * vm_map_entry_dispose: [ internal use only ] 767 * 768 * Inverse of vm_map_entry_create. 769 */ 770static void 771vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 772{ 773 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 774} 775 776/* 777 * vm_map_entry_create: [ internal use only ] 778 * 779 * Allocates a VM map entry for insertion. 780 * No entry fields are filled in. 781 */ 782static vm_map_entry_t 783vm_map_entry_create(vm_map_t map) 784{ 785 vm_map_entry_t new_entry; 786 787 if (map->system_map) 788 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 789 else 790 new_entry = uma_zalloc(mapentzone, M_WAITOK); 791 if (new_entry == NULL) 792 panic("vm_map_entry_create: kernel resources exhausted"); 793 return (new_entry); 794} 795 796/* 797 * vm_map_entry_set_behavior: 798 * 799 * Set the expected access behavior, either normal, random, or 800 * sequential. 801 */ 802static inline void 803vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 804{ 805 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 806 (behavior & MAP_ENTRY_BEHAV_MASK); 807} 808 809/* 810 * vm_map_entry_set_max_free: 811 * 812 * Set the max_free field in a vm_map_entry. 813 */ 814static inline void 815vm_map_entry_set_max_free(vm_map_entry_t entry) 816{ 817 818 entry->max_free = entry->adj_free; 819 if (entry->left != NULL && entry->left->max_free > entry->max_free) 820 entry->max_free = entry->left->max_free; 821 if (entry->right != NULL && entry->right->max_free > entry->max_free) 822 entry->max_free = entry->right->max_free; 823} 824 825/* 826 * vm_map_entry_splay: 827 * 828 * The Sleator and Tarjan top-down splay algorithm with the 829 * following variation. Max_free must be computed bottom-up, so 830 * on the downward pass, maintain the left and right spines in 831 * reverse order. Then, make a second pass up each side to fix 832 * the pointers and compute max_free. The time bound is O(log n) 833 * amortized. 834 * 835 * The new root is the vm_map_entry containing "addr", or else an 836 * adjacent entry (lower or higher) if addr is not in the tree. 837 * 838 * The map must be locked, and leaves it so. 839 * 840 * Returns: the new root. 841 */ 842static vm_map_entry_t 843vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 844{ 845 vm_map_entry_t llist, rlist; 846 vm_map_entry_t ltree, rtree; 847 vm_map_entry_t y; 848 849 /* Special case of empty tree. */ 850 if (root == NULL) 851 return (root); 852 853 /* 854 * Pass One: Splay down the tree until we find addr or a NULL 855 * pointer where addr would go. llist and rlist are the two 856 * sides in reverse order (bottom-up), with llist linked by 857 * the right pointer and rlist linked by the left pointer in 858 * the vm_map_entry. Wait until Pass Two to set max_free on 859 * the two spines. 860 */ 861 llist = NULL; 862 rlist = NULL; 863 for (;;) { 864 /* root is never NULL in here. */ 865 if (addr < root->start) { 866 y = root->left; 867 if (y == NULL) 868 break; 869 if (addr < y->start && y->left != NULL) { 870 /* Rotate right and put y on rlist. */ 871 root->left = y->right; 872 y->right = root; 873 vm_map_entry_set_max_free(root); 874 root = y->left; 875 y->left = rlist; 876 rlist = y; 877 } else { 878 /* Put root on rlist. */ 879 root->left = rlist; 880 rlist = root; 881 root = y; 882 } 883 } else if (addr >= root->end) { 884 y = root->right; 885 if (y == NULL) 886 break; 887 if (addr >= y->end && y->right != NULL) { 888 /* Rotate left and put y on llist. */ 889 root->right = y->left; 890 y->left = root; 891 vm_map_entry_set_max_free(root); 892 root = y->right; 893 y->right = llist; 894 llist = y; 895 } else { 896 /* Put root on llist. */ 897 root->right = llist; 898 llist = root; 899 root = y; 900 } 901 } else 902 break; 903 } 904 905 /* 906 * Pass Two: Walk back up the two spines, flip the pointers 907 * and set max_free. The subtrees of the root go at the 908 * bottom of llist and rlist. 909 */ 910 ltree = root->left; 911 while (llist != NULL) { 912 y = llist->right; 913 llist->right = ltree; 914 vm_map_entry_set_max_free(llist); 915 ltree = llist; 916 llist = y; 917 } 918 rtree = root->right; 919 while (rlist != NULL) { 920 y = rlist->left; 921 rlist->left = rtree; 922 vm_map_entry_set_max_free(rlist); 923 rtree = rlist; 924 rlist = y; 925 } 926 927 /* 928 * Final assembly: add ltree and rtree as subtrees of root. 929 */ 930 root->left = ltree; 931 root->right = rtree; 932 vm_map_entry_set_max_free(root); 933 934 return (root); 935} 936 937/* 938 * vm_map_entry_{un,}link: 939 * 940 * Insert/remove entries from maps. 941 */ 942static void 943vm_map_entry_link(vm_map_t map, 944 vm_map_entry_t after_where, 945 vm_map_entry_t entry) 946{ 947 948 CTR4(KTR_VM, 949 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 950 map->nentries, entry, after_where); 951 VM_MAP_ASSERT_LOCKED(map); 952 KASSERT(after_where == &map->header || 953 after_where->end <= entry->start, 954 ("vm_map_entry_link: prev end %jx new start %jx overlap", 955 (uintmax_t)after_where->end, (uintmax_t)entry->start)); 956 KASSERT(after_where->next == &map->header || 957 entry->end <= after_where->next->start, 958 ("vm_map_entry_link: new end %jx next start %jx overlap", 959 (uintmax_t)entry->end, (uintmax_t)after_where->next->start)); 960 961 map->nentries++; 962 entry->prev = after_where; 963 entry->next = after_where->next; 964 entry->next->prev = entry; 965 after_where->next = entry; 966 967 if (after_where != &map->header) { 968 if (after_where != map->root) 969 vm_map_entry_splay(after_where->start, map->root); 970 entry->right = after_where->right; 971 entry->left = after_where; 972 after_where->right = NULL; 973 after_where->adj_free = entry->start - after_where->end; 974 vm_map_entry_set_max_free(after_where); 975 } else { 976 entry->right = map->root; 977 entry->left = NULL; 978 } 979 entry->adj_free = (entry->next == &map->header ? map->max_offset : 980 entry->next->start) - entry->end; 981 vm_map_entry_set_max_free(entry); 982 map->root = entry; 983} 984 985static void 986vm_map_entry_unlink(vm_map_t map, 987 vm_map_entry_t entry) 988{ 989 vm_map_entry_t next, prev, root; 990 991 VM_MAP_ASSERT_LOCKED(map); 992 if (entry != map->root) 993 vm_map_entry_splay(entry->start, map->root); 994 if (entry->left == NULL) 995 root = entry->right; 996 else { 997 root = vm_map_entry_splay(entry->start, entry->left); 998 root->right = entry->right; 999 root->adj_free = (entry->next == &map->header ? map->max_offset : 1000 entry->next->start) - root->end; 1001 vm_map_entry_set_max_free(root); 1002 } 1003 map->root = root; 1004 1005 prev = entry->prev; 1006 next = entry->next; 1007 next->prev = prev; 1008 prev->next = next; 1009 map->nentries--; 1010 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1011 map->nentries, entry); 1012} 1013 1014/* 1015 * vm_map_entry_resize_free: 1016 * 1017 * Recompute the amount of free space following a vm_map_entry 1018 * and propagate that value up the tree. Call this function after 1019 * resizing a map entry in-place, that is, without a call to 1020 * vm_map_entry_link() or _unlink(). 1021 * 1022 * The map must be locked, and leaves it so. 1023 */ 1024static void 1025vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1026{ 1027 1028 /* 1029 * Using splay trees without parent pointers, propagating 1030 * max_free up the tree is done by moving the entry to the 1031 * root and making the change there. 1032 */ 1033 if (entry != map->root) 1034 map->root = vm_map_entry_splay(entry->start, map->root); 1035 1036 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1037 entry->next->start) - entry->end; 1038 vm_map_entry_set_max_free(entry); 1039} 1040 1041/* 1042 * vm_map_lookup_entry: [ internal use only ] 1043 * 1044 * Finds the map entry containing (or 1045 * immediately preceding) the specified address 1046 * in the given map; the entry is returned 1047 * in the "entry" parameter. The boolean 1048 * result indicates whether the address is 1049 * actually contained in the map. 1050 */ 1051boolean_t 1052vm_map_lookup_entry( 1053 vm_map_t map, 1054 vm_offset_t address, 1055 vm_map_entry_t *entry) /* OUT */ 1056{ 1057 vm_map_entry_t cur; 1058 boolean_t locked; 1059 1060 /* 1061 * If the map is empty, then the map entry immediately preceding 1062 * "address" is the map's header. 1063 */ 1064 cur = map->root; 1065 if (cur == NULL) 1066 *entry = &map->header; 1067 else if (address >= cur->start && cur->end > address) { 1068 *entry = cur; 1069 return (TRUE); 1070 } else if ((locked = vm_map_locked(map)) || 1071 sx_try_upgrade(&map->lock)) { 1072 /* 1073 * Splay requires a write lock on the map. However, it only 1074 * restructures the binary search tree; it does not otherwise 1075 * change the map. Thus, the map's timestamp need not change 1076 * on a temporary upgrade. 1077 */ 1078 map->root = cur = vm_map_entry_splay(address, cur); 1079 if (!locked) 1080 sx_downgrade(&map->lock); 1081 1082 /* 1083 * If "address" is contained within a map entry, the new root 1084 * is that map entry. Otherwise, the new root is a map entry 1085 * immediately before or after "address". 1086 */ 1087 if (address >= cur->start) { 1088 *entry = cur; 1089 if (cur->end > address) 1090 return (TRUE); 1091 } else 1092 *entry = cur->prev; 1093 } else 1094 /* 1095 * Since the map is only locked for read access, perform a 1096 * standard binary search tree lookup for "address". 1097 */ 1098 for (;;) { 1099 if (address < cur->start) { 1100 if (cur->left == NULL) { 1101 *entry = cur->prev; 1102 break; 1103 } 1104 cur = cur->left; 1105 } else if (cur->end > address) { 1106 *entry = cur; 1107 return (TRUE); 1108 } else { 1109 if (cur->right == NULL) { 1110 *entry = cur; 1111 break; 1112 } 1113 cur = cur->right; 1114 } 1115 } 1116 return (FALSE); 1117} 1118 1119/* 1120 * vm_map_insert: 1121 * 1122 * Inserts the given whole VM object into the target 1123 * map at the specified address range. The object's 1124 * size should match that of the address range. 1125 * 1126 * Requires that the map be locked, and leaves it so. 1127 * 1128 * If object is non-NULL, ref count must be bumped by caller 1129 * prior to making call to account for the new entry. 1130 */ 1131int 1132vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1133 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow) 1134{ 1135 vm_map_entry_t new_entry, prev_entry, temp_entry; 1136 struct ucred *cred; 1137 vm_eflags_t protoeflags; 1138 vm_inherit_t inheritance; 1139 1140 VM_MAP_ASSERT_LOCKED(map); 1141 KASSERT((object != kmem_object && object != kernel_object) || 1142 (cow & MAP_COPY_ON_WRITE) == 0, 1143 ("vm_map_insert: kmem or kernel object and COW")); 1144 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0, 1145 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1146 1147 /* 1148 * Check that the start and end points are not bogus. 1149 */ 1150 if (start < map->min_offset || end > map->max_offset || start >= end) 1151 return (KERN_INVALID_ADDRESS); 1152 1153 /* 1154 * Find the entry prior to the proposed starting address; if it's part 1155 * of an existing entry, this range is bogus. 1156 */ 1157 if (vm_map_lookup_entry(map, start, &temp_entry)) 1158 return (KERN_NO_SPACE); 1159 1160 prev_entry = temp_entry; 1161 1162 /* 1163 * Assert that the next entry doesn't overlap the end point. 1164 */ 1165 if (prev_entry->next != &map->header && prev_entry->next->start < end) 1166 return (KERN_NO_SPACE); 1167 1168 protoeflags = 0; 1169 if (cow & MAP_COPY_ON_WRITE) 1170 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; 1171 if (cow & MAP_NOFAULT) 1172 protoeflags |= MAP_ENTRY_NOFAULT; 1173 if (cow & MAP_DISABLE_SYNCER) 1174 protoeflags |= MAP_ENTRY_NOSYNC; 1175 if (cow & MAP_DISABLE_COREDUMP) 1176 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1177 if (cow & MAP_STACK_GROWS_DOWN) 1178 protoeflags |= MAP_ENTRY_GROWS_DOWN; 1179 if (cow & MAP_STACK_GROWS_UP) 1180 protoeflags |= MAP_ENTRY_GROWS_UP; 1181 if (cow & MAP_VN_WRITECOUNT) 1182 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1183 if (cow & MAP_INHERIT_SHARE) 1184 inheritance = VM_INHERIT_SHARE; 1185 else 1186 inheritance = VM_INHERIT_DEFAULT; 1187 1188 cred = NULL; 1189 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1190 goto charged; 1191 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1192 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1193 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1194 return (KERN_RESOURCE_SHORTAGE); 1195 KASSERT(object == NULL || 1196 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 || 1197 object->cred == NULL, 1198 ("overcommit: vm_map_insert o %p", object)); 1199 cred = curthread->td_ucred; 1200 } 1201 1202charged: 1203 /* Expand the kernel pmap, if necessary. */ 1204 if (map == kernel_map && end > kernel_vm_end) 1205 pmap_growkernel(end); 1206 if (object != NULL) { 1207 /* 1208 * OBJ_ONEMAPPING must be cleared unless this mapping 1209 * is trivially proven to be the only mapping for any 1210 * of the object's pages. (Object granularity 1211 * reference counting is insufficient to recognize 1212 * aliases with precision.) 1213 */ 1214 VM_OBJECT_WLOCK(object); 1215 if (object->ref_count > 1 || object->shadow_count != 0) 1216 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1217 VM_OBJECT_WUNLOCK(object); 1218 } else if (prev_entry != &map->header && 1219 prev_entry->eflags == protoeflags && 1220 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 && 1221 prev_entry->end == start && prev_entry->wired_count == 0 && 1222 (prev_entry->cred == cred || 1223 (prev_entry->object.vm_object != NULL && 1224 prev_entry->object.vm_object->cred == cred)) && 1225 vm_object_coalesce(prev_entry->object.vm_object, 1226 prev_entry->offset, 1227 (vm_size_t)(prev_entry->end - prev_entry->start), 1228 (vm_size_t)(end - prev_entry->end), cred != NULL && 1229 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) { 1230 /* 1231 * We were able to extend the object. Determine if we 1232 * can extend the previous map entry to include the 1233 * new range as well. 1234 */ 1235 if (prev_entry->inheritance == inheritance && 1236 prev_entry->protection == prot && 1237 prev_entry->max_protection == max) { 1238 map->size += end - prev_entry->end; 1239 prev_entry->end = end; 1240 vm_map_entry_resize_free(map, prev_entry); 1241 vm_map_simplify_entry(map, prev_entry); 1242 return (KERN_SUCCESS); 1243 } 1244 1245 /* 1246 * If we can extend the object but cannot extend the 1247 * map entry, we have to create a new map entry. We 1248 * must bump the ref count on the extended object to 1249 * account for it. object may be NULL. 1250 */ 1251 object = prev_entry->object.vm_object; 1252 offset = prev_entry->offset + 1253 (prev_entry->end - prev_entry->start); 1254 vm_object_reference(object); 1255 if (cred != NULL && object != NULL && object->cred != NULL && 1256 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1257 /* Object already accounts for this uid. */ 1258 cred = NULL; 1259 } 1260 } 1261 if (cred != NULL) 1262 crhold(cred); 1263 1264 /* 1265 * Create a new entry 1266 */ 1267 new_entry = vm_map_entry_create(map); 1268 new_entry->start = start; 1269 new_entry->end = end; 1270 new_entry->cred = NULL; 1271 1272 new_entry->eflags = protoeflags; 1273 new_entry->object.vm_object = object; 1274 new_entry->offset = offset; 1275 new_entry->avail_ssize = 0; 1276 1277 new_entry->inheritance = inheritance; 1278 new_entry->protection = prot; 1279 new_entry->max_protection = max; 1280 new_entry->wired_count = 0; 1281 new_entry->wiring_thread = NULL; 1282 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1283 new_entry->next_read = OFF_TO_IDX(offset); 1284 1285 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1286 ("overcommit: vm_map_insert leaks vm_map %p", new_entry)); 1287 new_entry->cred = cred; 1288 1289 /* 1290 * Insert the new entry into the list 1291 */ 1292 vm_map_entry_link(map, prev_entry, new_entry); 1293 map->size += new_entry->end - new_entry->start; 1294 1295 /* 1296 * Try to coalesce the new entry with both the previous and next 1297 * entries in the list. Previously, we only attempted to coalesce 1298 * with the previous entry when object is NULL. Here, we handle the 1299 * other cases, which are less common. 1300 */ 1301 vm_map_simplify_entry(map, new_entry); 1302 1303 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) { 1304 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset), 1305 end - start, cow & MAP_PREFAULT_PARTIAL); 1306 } 1307 1308 return (KERN_SUCCESS); 1309} 1310 1311/* 1312 * vm_map_findspace: 1313 * 1314 * Find the first fit (lowest VM address) for "length" free bytes 1315 * beginning at address >= start in the given map. 1316 * 1317 * In a vm_map_entry, "adj_free" is the amount of free space 1318 * adjacent (higher address) to this entry, and "max_free" is the 1319 * maximum amount of contiguous free space in its subtree. This 1320 * allows finding a free region in one path down the tree, so 1321 * O(log n) amortized with splay trees. 1322 * 1323 * The map must be locked, and leaves it so. 1324 * 1325 * Returns: 0 on success, and starting address in *addr, 1326 * 1 if insufficient space. 1327 */ 1328int 1329vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1330 vm_offset_t *addr) /* OUT */ 1331{ 1332 vm_map_entry_t entry; 1333 vm_offset_t st; 1334 1335 /* 1336 * Request must fit within min/max VM address and must avoid 1337 * address wrap. 1338 */ 1339 if (start < map->min_offset) 1340 start = map->min_offset; 1341 if (start + length > map->max_offset || start + length < start) 1342 return (1); 1343 1344 /* Empty tree means wide open address space. */ 1345 if (map->root == NULL) { 1346 *addr = start; 1347 return (0); 1348 } 1349 1350 /* 1351 * After splay, if start comes before root node, then there 1352 * must be a gap from start to the root. 1353 */ 1354 map->root = vm_map_entry_splay(start, map->root); 1355 if (start + length <= map->root->start) { 1356 *addr = start; 1357 return (0); 1358 } 1359 1360 /* 1361 * Root is the last node that might begin its gap before 1362 * start, and this is the last comparison where address 1363 * wrap might be a problem. 1364 */ 1365 st = (start > map->root->end) ? start : map->root->end; 1366 if (length <= map->root->end + map->root->adj_free - st) { 1367 *addr = st; 1368 return (0); 1369 } 1370 1371 /* With max_free, can immediately tell if no solution. */ 1372 entry = map->root->right; 1373 if (entry == NULL || length > entry->max_free) 1374 return (1); 1375 1376 /* 1377 * Search the right subtree in the order: left subtree, root, 1378 * right subtree (first fit). The previous splay implies that 1379 * all regions in the right subtree have addresses > start. 1380 */ 1381 while (entry != NULL) { 1382 if (entry->left != NULL && entry->left->max_free >= length) 1383 entry = entry->left; 1384 else if (entry->adj_free >= length) { 1385 *addr = entry->end; 1386 return (0); 1387 } else 1388 entry = entry->right; 1389 } 1390 1391 /* Can't get here, so panic if we do. */ 1392 panic("vm_map_findspace: max_free corrupt"); 1393} 1394 1395int 1396vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1397 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1398 vm_prot_t max, int cow) 1399{ 1400 vm_offset_t end; 1401 int result; 1402 1403 end = start + length; 1404 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1405 object == NULL, 1406 ("vm_map_fixed: non-NULL backing object for stack")); 1407 vm_map_lock(map); 1408 VM_MAP_RANGE_CHECK(map, start, end); 1409 if ((cow & MAP_CHECK_EXCL) == 0) 1410 vm_map_delete(map, start, end); 1411 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1412 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1413 prot, max, cow); 1414 } else { 1415 result = vm_map_insert(map, object, offset, start, end, 1416 prot, max, cow); 1417 } 1418 vm_map_unlock(map); 1419 return (result); 1420} 1421 1422/* 1423 * vm_map_find finds an unallocated region in the target address 1424 * map with the given length. The search is defined to be 1425 * first-fit from the specified address; the region found is 1426 * returned in the same parameter. 1427 * 1428 * If object is non-NULL, ref count must be bumped by caller 1429 * prior to making call to account for the new entry. 1430 */ 1431int 1432vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1433 vm_offset_t *addr, /* IN/OUT */ 1434 vm_size_t length, vm_offset_t max_addr, int find_space, 1435 vm_prot_t prot, vm_prot_t max, int cow) 1436{ 1437 vm_offset_t alignment, initial_addr, start; 1438 int result; 1439 1440 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1441 object == NULL, 1442 ("vm_map_find: non-NULL backing object for stack")); 1443 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1444 (object->flags & OBJ_COLORED) == 0)) 1445 find_space = VMFS_ANY_SPACE; 1446 if (find_space >> 8 != 0) { 1447 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1448 alignment = (vm_offset_t)1 << (find_space >> 8); 1449 } else 1450 alignment = 0; 1451 initial_addr = *addr; 1452again: 1453 start = initial_addr; 1454 vm_map_lock(map); 1455 do { 1456 if (find_space != VMFS_NO_SPACE) { 1457 if (vm_map_findspace(map, start, length, addr) || 1458 (max_addr != 0 && *addr + length > max_addr)) { 1459 vm_map_unlock(map); 1460 if (find_space == VMFS_OPTIMAL_SPACE) { 1461 find_space = VMFS_ANY_SPACE; 1462 goto again; 1463 } 1464 return (KERN_NO_SPACE); 1465 } 1466 switch (find_space) { 1467 case VMFS_SUPER_SPACE: 1468 case VMFS_OPTIMAL_SPACE: 1469 pmap_align_superpage(object, offset, addr, 1470 length); 1471 break; 1472 case VMFS_ANY_SPACE: 1473 break; 1474 default: 1475 if ((*addr & (alignment - 1)) != 0) { 1476 *addr &= ~(alignment - 1); 1477 *addr += alignment; 1478 } 1479 break; 1480 } 1481 1482 start = *addr; 1483 } 1484 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1485 result = vm_map_stack_locked(map, start, length, 1486 sgrowsiz, prot, max, cow); 1487 } else { 1488 result = vm_map_insert(map, object, offset, start, 1489 start + length, prot, max, cow); 1490 } 1491 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1492 find_space != VMFS_ANY_SPACE); 1493 vm_map_unlock(map); 1494 return (result); 1495} 1496 1497/* 1498 * vm_map_simplify_entry: 1499 * 1500 * Simplify the given map entry by merging with either neighbor. This 1501 * routine also has the ability to merge with both neighbors. 1502 * 1503 * The map must be locked. 1504 * 1505 * This routine guarentees that the passed entry remains valid (though 1506 * possibly extended). When merging, this routine may delete one or 1507 * both neighbors. 1508 */ 1509void 1510vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1511{ 1512 vm_map_entry_t next, prev; 1513 vm_size_t prevsize, esize; 1514 1515 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | 1516 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0) 1517 return; 1518 1519 prev = entry->prev; 1520 if (prev != &map->header) { 1521 prevsize = prev->end - prev->start; 1522 if ( (prev->end == entry->start) && 1523 (prev->object.vm_object == entry->object.vm_object) && 1524 (!prev->object.vm_object || 1525 (prev->offset + prevsize == entry->offset)) && 1526 (prev->eflags == entry->eflags) && 1527 (prev->protection == entry->protection) && 1528 (prev->max_protection == entry->max_protection) && 1529 (prev->inheritance == entry->inheritance) && 1530 (prev->wired_count == entry->wired_count) && 1531 (prev->cred == entry->cred)) { 1532 vm_map_entry_unlink(map, prev); 1533 entry->start = prev->start; 1534 entry->offset = prev->offset; 1535 if (entry->prev != &map->header) 1536 vm_map_entry_resize_free(map, entry->prev); 1537 1538 /* 1539 * If the backing object is a vnode object, 1540 * vm_object_deallocate() calls vrele(). 1541 * However, vrele() does not lock the vnode 1542 * because the vnode has additional 1543 * references. Thus, the map lock can be kept 1544 * without causing a lock-order reversal with 1545 * the vnode lock. 1546 * 1547 * Since we count the number of virtual page 1548 * mappings in object->un_pager.vnp.writemappings, 1549 * the writemappings value should not be adjusted 1550 * when the entry is disposed of. 1551 */ 1552 if (prev->object.vm_object) 1553 vm_object_deallocate(prev->object.vm_object); 1554 if (prev->cred != NULL) 1555 crfree(prev->cred); 1556 vm_map_entry_dispose(map, prev); 1557 } 1558 } 1559 1560 next = entry->next; 1561 if (next != &map->header) { 1562 esize = entry->end - entry->start; 1563 if ((entry->end == next->start) && 1564 (next->object.vm_object == entry->object.vm_object) && 1565 (!entry->object.vm_object || 1566 (entry->offset + esize == next->offset)) && 1567 (next->eflags == entry->eflags) && 1568 (next->protection == entry->protection) && 1569 (next->max_protection == entry->max_protection) && 1570 (next->inheritance == entry->inheritance) && 1571 (next->wired_count == entry->wired_count) && 1572 (next->cred == entry->cred)) { 1573 vm_map_entry_unlink(map, next); 1574 entry->end = next->end; 1575 vm_map_entry_resize_free(map, entry); 1576 1577 /* 1578 * See comment above. 1579 */ 1580 if (next->object.vm_object) 1581 vm_object_deallocate(next->object.vm_object); 1582 if (next->cred != NULL) 1583 crfree(next->cred); 1584 vm_map_entry_dispose(map, next); 1585 } 1586 } 1587} 1588/* 1589 * vm_map_clip_start: [ internal use only ] 1590 * 1591 * Asserts that the given entry begins at or after 1592 * the specified address; if necessary, 1593 * it splits the entry into two. 1594 */ 1595#define vm_map_clip_start(map, entry, startaddr) \ 1596{ \ 1597 if (startaddr > entry->start) \ 1598 _vm_map_clip_start(map, entry, startaddr); \ 1599} 1600 1601/* 1602 * This routine is called only when it is known that 1603 * the entry must be split. 1604 */ 1605static void 1606_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1607{ 1608 vm_map_entry_t new_entry; 1609 1610 VM_MAP_ASSERT_LOCKED(map); 1611 1612 /* 1613 * Split off the front portion -- note that we must insert the new 1614 * entry BEFORE this one, so that this entry has the specified 1615 * starting address. 1616 */ 1617 vm_map_simplify_entry(map, entry); 1618 1619 /* 1620 * If there is no object backing this entry, we might as well create 1621 * one now. If we defer it, an object can get created after the map 1622 * is clipped, and individual objects will be created for the split-up 1623 * map. This is a bit of a hack, but is also about the best place to 1624 * put this improvement. 1625 */ 1626 if (entry->object.vm_object == NULL && !map->system_map) { 1627 vm_object_t object; 1628 object = vm_object_allocate(OBJT_DEFAULT, 1629 atop(entry->end - entry->start)); 1630 entry->object.vm_object = object; 1631 entry->offset = 0; 1632 if (entry->cred != NULL) { 1633 object->cred = entry->cred; 1634 object->charge = entry->end - entry->start; 1635 entry->cred = NULL; 1636 } 1637 } else if (entry->object.vm_object != NULL && 1638 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1639 entry->cred != NULL) { 1640 VM_OBJECT_WLOCK(entry->object.vm_object); 1641 KASSERT(entry->object.vm_object->cred == NULL, 1642 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1643 entry->object.vm_object->cred = entry->cred; 1644 entry->object.vm_object->charge = entry->end - entry->start; 1645 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1646 entry->cred = NULL; 1647 } 1648 1649 new_entry = vm_map_entry_create(map); 1650 *new_entry = *entry; 1651 1652 new_entry->end = start; 1653 entry->offset += (start - entry->start); 1654 entry->start = start; 1655 if (new_entry->cred != NULL) 1656 crhold(entry->cred); 1657 1658 vm_map_entry_link(map, entry->prev, new_entry); 1659 1660 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1661 vm_object_reference(new_entry->object.vm_object); 1662 /* 1663 * The object->un_pager.vnp.writemappings for the 1664 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1665 * kept as is here. The virtual pages are 1666 * re-distributed among the clipped entries, so the sum is 1667 * left the same. 1668 */ 1669 } 1670} 1671 1672/* 1673 * vm_map_clip_end: [ internal use only ] 1674 * 1675 * Asserts that the given entry ends at or before 1676 * the specified address; if necessary, 1677 * it splits the entry into two. 1678 */ 1679#define vm_map_clip_end(map, entry, endaddr) \ 1680{ \ 1681 if ((endaddr) < (entry->end)) \ 1682 _vm_map_clip_end((map), (entry), (endaddr)); \ 1683} 1684 1685/* 1686 * This routine is called only when it is known that 1687 * the entry must be split. 1688 */ 1689static void 1690_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1691{ 1692 vm_map_entry_t new_entry; 1693 1694 VM_MAP_ASSERT_LOCKED(map); 1695 1696 /* 1697 * If there is no object backing this entry, we might as well create 1698 * one now. If we defer it, an object can get created after the map 1699 * is clipped, and individual objects will be created for the split-up 1700 * map. This is a bit of a hack, but is also about the best place to 1701 * put this improvement. 1702 */ 1703 if (entry->object.vm_object == NULL && !map->system_map) { 1704 vm_object_t object; 1705 object = vm_object_allocate(OBJT_DEFAULT, 1706 atop(entry->end - entry->start)); 1707 entry->object.vm_object = object; 1708 entry->offset = 0; 1709 if (entry->cred != NULL) { 1710 object->cred = entry->cred; 1711 object->charge = entry->end - entry->start; 1712 entry->cred = NULL; 1713 } 1714 } else if (entry->object.vm_object != NULL && 1715 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1716 entry->cred != NULL) { 1717 VM_OBJECT_WLOCK(entry->object.vm_object); 1718 KASSERT(entry->object.vm_object->cred == NULL, 1719 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1720 entry->object.vm_object->cred = entry->cred; 1721 entry->object.vm_object->charge = entry->end - entry->start; 1722 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1723 entry->cred = NULL; 1724 } 1725 1726 /* 1727 * Create a new entry and insert it AFTER the specified entry 1728 */ 1729 new_entry = vm_map_entry_create(map); 1730 *new_entry = *entry; 1731 1732 new_entry->start = entry->end = end; 1733 new_entry->offset += (end - entry->start); 1734 if (new_entry->cred != NULL) 1735 crhold(entry->cred); 1736 1737 vm_map_entry_link(map, entry, new_entry); 1738 1739 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1740 vm_object_reference(new_entry->object.vm_object); 1741 } 1742} 1743 1744/* 1745 * vm_map_submap: [ kernel use only ] 1746 * 1747 * Mark the given range as handled by a subordinate map. 1748 * 1749 * This range must have been created with vm_map_find, 1750 * and no other operations may have been performed on this 1751 * range prior to calling vm_map_submap. 1752 * 1753 * Only a limited number of operations can be performed 1754 * within this rage after calling vm_map_submap: 1755 * vm_fault 1756 * [Don't try vm_map_copy!] 1757 * 1758 * To remove a submapping, one must first remove the 1759 * range from the superior map, and then destroy the 1760 * submap (if desired). [Better yet, don't try it.] 1761 */ 1762int 1763vm_map_submap( 1764 vm_map_t map, 1765 vm_offset_t start, 1766 vm_offset_t end, 1767 vm_map_t submap) 1768{ 1769 vm_map_entry_t entry; 1770 int result = KERN_INVALID_ARGUMENT; 1771 1772 vm_map_lock(map); 1773 1774 VM_MAP_RANGE_CHECK(map, start, end); 1775 1776 if (vm_map_lookup_entry(map, start, &entry)) { 1777 vm_map_clip_start(map, entry, start); 1778 } else 1779 entry = entry->next; 1780 1781 vm_map_clip_end(map, entry, end); 1782 1783 if ((entry->start == start) && (entry->end == end) && 1784 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1785 (entry->object.vm_object == NULL)) { 1786 entry->object.sub_map = submap; 1787 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1788 result = KERN_SUCCESS; 1789 } 1790 vm_map_unlock(map); 1791 1792 return (result); 1793} 1794 1795/* 1796 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified 1797 */ 1798#define MAX_INIT_PT 96 1799 1800/* 1801 * vm_map_pmap_enter: 1802 * 1803 * Preload the specified map's pmap with mappings to the specified 1804 * object's memory-resident pages. No further physical pages are 1805 * allocated, and no further virtual pages are retrieved from secondary 1806 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 1807 * limited number of page mappings are created at the low-end of the 1808 * specified address range. (For this purpose, a superpage mapping 1809 * counts as one page mapping.) Otherwise, all resident pages within 1810 * the specified address range are mapped. Because these mappings are 1811 * being created speculatively, cached pages are not reactivated and 1812 * mapped. 1813 */ 1814void 1815vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1816 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1817{ 1818 vm_offset_t start; 1819 vm_page_t p, p_start; 1820 vm_pindex_t mask, psize, threshold, tmpidx; 1821 1822 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1823 return; 1824 VM_OBJECT_RLOCK(object); 1825 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1826 VM_OBJECT_RUNLOCK(object); 1827 VM_OBJECT_WLOCK(object); 1828 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1829 pmap_object_init_pt(map->pmap, addr, object, pindex, 1830 size); 1831 VM_OBJECT_WUNLOCK(object); 1832 return; 1833 } 1834 VM_OBJECT_LOCK_DOWNGRADE(object); 1835 } 1836 1837 psize = atop(size); 1838 if (psize + pindex > object->size) { 1839 if (object->size < pindex) { 1840 VM_OBJECT_RUNLOCK(object); 1841 return; 1842 } 1843 psize = object->size - pindex; 1844 } 1845 1846 start = 0; 1847 p_start = NULL; 1848 threshold = MAX_INIT_PT; 1849 1850 p = vm_page_find_least(object, pindex); 1851 /* 1852 * Assert: the variable p is either (1) the page with the 1853 * least pindex greater than or equal to the parameter pindex 1854 * or (2) NULL. 1855 */ 1856 for (; 1857 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1858 p = TAILQ_NEXT(p, listq)) { 1859 /* 1860 * don't allow an madvise to blow away our really 1861 * free pages allocating pv entries. 1862 */ 1863 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 1864 cnt.v_free_count < cnt.v_free_reserved) || 1865 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 1866 tmpidx >= threshold)) { 1867 psize = tmpidx; 1868 break; 1869 } 1870 if (p->valid == VM_PAGE_BITS_ALL) { 1871 if (p_start == NULL) { 1872 start = addr + ptoa(tmpidx); 1873 p_start = p; 1874 } 1875 /* Jump ahead if a superpage mapping is possible. */ 1876 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 1877 (pagesizes[p->psind] - 1)) == 0) { 1878 mask = atop(pagesizes[p->psind]) - 1; 1879 if (tmpidx + mask < psize && 1880 vm_page_ps_is_valid(p)) { 1881 p += mask; 1882 threshold += mask; 1883 } 1884 } 1885 } else if (p_start != NULL) { 1886 pmap_enter_object(map->pmap, start, addr + 1887 ptoa(tmpidx), p_start, prot); 1888 p_start = NULL; 1889 } 1890 } 1891 if (p_start != NULL) 1892 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1893 p_start, prot); 1894 VM_OBJECT_RUNLOCK(object); 1895} 1896 1897/* 1898 * vm_map_protect: 1899 * 1900 * Sets the protection of the specified address 1901 * region in the target map. If "set_max" is 1902 * specified, the maximum protection is to be set; 1903 * otherwise, only the current protection is affected. 1904 */ 1905int 1906vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1907 vm_prot_t new_prot, boolean_t set_max) 1908{ 1909 vm_map_entry_t current, entry; 1910 vm_object_t obj; 1911 struct ucred *cred; 1912 vm_prot_t old_prot; 1913 1914 if (start == end) 1915 return (KERN_SUCCESS); 1916 1917 vm_map_lock(map); 1918 1919 VM_MAP_RANGE_CHECK(map, start, end); 1920 1921 if (vm_map_lookup_entry(map, start, &entry)) { 1922 vm_map_clip_start(map, entry, start); 1923 } else { 1924 entry = entry->next; 1925 } 1926 1927 /* 1928 * Make a first pass to check for protection violations. 1929 */ 1930 current = entry; 1931 while ((current != &map->header) && (current->start < end)) { 1932 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1933 vm_map_unlock(map); 1934 return (KERN_INVALID_ARGUMENT); 1935 } 1936 if ((new_prot & current->max_protection) != new_prot) { 1937 vm_map_unlock(map); 1938 return (KERN_PROTECTION_FAILURE); 1939 } 1940 current = current->next; 1941 } 1942 1943 1944 /* 1945 * Do an accounting pass for private read-only mappings that 1946 * now will do cow due to allowed write (e.g. debugger sets 1947 * breakpoint on text segment) 1948 */ 1949 for (current = entry; (current != &map->header) && 1950 (current->start < end); current = current->next) { 1951 1952 vm_map_clip_end(map, current, end); 1953 1954 if (set_max || 1955 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1956 ENTRY_CHARGED(current)) { 1957 continue; 1958 } 1959 1960 cred = curthread->td_ucred; 1961 obj = current->object.vm_object; 1962 1963 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1964 if (!swap_reserve(current->end - current->start)) { 1965 vm_map_unlock(map); 1966 return (KERN_RESOURCE_SHORTAGE); 1967 } 1968 crhold(cred); 1969 current->cred = cred; 1970 continue; 1971 } 1972 1973 VM_OBJECT_WLOCK(obj); 1974 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1975 VM_OBJECT_WUNLOCK(obj); 1976 continue; 1977 } 1978 1979 /* 1980 * Charge for the whole object allocation now, since 1981 * we cannot distinguish between non-charged and 1982 * charged clipped mapping of the same object later. 1983 */ 1984 KASSERT(obj->charge == 0, 1985 ("vm_map_protect: object %p overcharged (entry %p)", 1986 obj, current)); 1987 if (!swap_reserve(ptoa(obj->size))) { 1988 VM_OBJECT_WUNLOCK(obj); 1989 vm_map_unlock(map); 1990 return (KERN_RESOURCE_SHORTAGE); 1991 } 1992 1993 crhold(cred); 1994 obj->cred = cred; 1995 obj->charge = ptoa(obj->size); 1996 VM_OBJECT_WUNLOCK(obj); 1997 } 1998 1999 /* 2000 * Go back and fix up protections. [Note that clipping is not 2001 * necessary the second time.] 2002 */ 2003 current = entry; 2004 while ((current != &map->header) && (current->start < end)) { 2005 old_prot = current->protection; 2006 2007 if (set_max) 2008 current->protection = 2009 (current->max_protection = new_prot) & 2010 old_prot; 2011 else 2012 current->protection = new_prot; 2013 2014 /* 2015 * For user wired map entries, the normal lazy evaluation of 2016 * write access upgrades through soft page faults is 2017 * undesirable. Instead, immediately copy any pages that are 2018 * copy-on-write and enable write access in the physical map. 2019 */ 2020 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2021 (current->protection & VM_PROT_WRITE) != 0 && 2022 (old_prot & VM_PROT_WRITE) == 0) 2023 vm_fault_copy_entry(map, map, current, current, NULL); 2024 2025 /* 2026 * When restricting access, update the physical map. Worry 2027 * about copy-on-write here. 2028 */ 2029 if ((old_prot & ~current->protection) != 0) { 2030#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2031 VM_PROT_ALL) 2032 pmap_protect(map->pmap, current->start, 2033 current->end, 2034 current->protection & MASK(current)); 2035#undef MASK 2036 } 2037 vm_map_simplify_entry(map, current); 2038 current = current->next; 2039 } 2040 vm_map_unlock(map); 2041 return (KERN_SUCCESS); 2042} 2043 2044/* 2045 * vm_map_madvise: 2046 * 2047 * This routine traverses a processes map handling the madvise 2048 * system call. Advisories are classified as either those effecting 2049 * the vm_map_entry structure, or those effecting the underlying 2050 * objects. 2051 */ 2052int 2053vm_map_madvise( 2054 vm_map_t map, 2055 vm_offset_t start, 2056 vm_offset_t end, 2057 int behav) 2058{ 2059 vm_map_entry_t current, entry; 2060 int modify_map = 0; 2061 2062 /* 2063 * Some madvise calls directly modify the vm_map_entry, in which case 2064 * we need to use an exclusive lock on the map and we need to perform 2065 * various clipping operations. Otherwise we only need a read-lock 2066 * on the map. 2067 */ 2068 switch(behav) { 2069 case MADV_NORMAL: 2070 case MADV_SEQUENTIAL: 2071 case MADV_RANDOM: 2072 case MADV_NOSYNC: 2073 case MADV_AUTOSYNC: 2074 case MADV_NOCORE: 2075 case MADV_CORE: 2076 if (start == end) 2077 return (KERN_SUCCESS); 2078 modify_map = 1; 2079 vm_map_lock(map); 2080 break; 2081 case MADV_WILLNEED: 2082 case MADV_DONTNEED: 2083 case MADV_FREE: 2084 if (start == end) 2085 return (KERN_SUCCESS); 2086 vm_map_lock_read(map); 2087 break; 2088 default: 2089 return (KERN_INVALID_ARGUMENT); 2090 } 2091 2092 /* 2093 * Locate starting entry and clip if necessary. 2094 */ 2095 VM_MAP_RANGE_CHECK(map, start, end); 2096 2097 if (vm_map_lookup_entry(map, start, &entry)) { 2098 if (modify_map) 2099 vm_map_clip_start(map, entry, start); 2100 } else { 2101 entry = entry->next; 2102 } 2103 2104 if (modify_map) { 2105 /* 2106 * madvise behaviors that are implemented in the vm_map_entry. 2107 * 2108 * We clip the vm_map_entry so that behavioral changes are 2109 * limited to the specified address range. 2110 */ 2111 for (current = entry; 2112 (current != &map->header) && (current->start < end); 2113 current = current->next 2114 ) { 2115 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2116 continue; 2117 2118 vm_map_clip_end(map, current, end); 2119 2120 switch (behav) { 2121 case MADV_NORMAL: 2122 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2123 break; 2124 case MADV_SEQUENTIAL: 2125 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2126 break; 2127 case MADV_RANDOM: 2128 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2129 break; 2130 case MADV_NOSYNC: 2131 current->eflags |= MAP_ENTRY_NOSYNC; 2132 break; 2133 case MADV_AUTOSYNC: 2134 current->eflags &= ~MAP_ENTRY_NOSYNC; 2135 break; 2136 case MADV_NOCORE: 2137 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2138 break; 2139 case MADV_CORE: 2140 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2141 break; 2142 default: 2143 break; 2144 } 2145 vm_map_simplify_entry(map, current); 2146 } 2147 vm_map_unlock(map); 2148 } else { 2149 vm_pindex_t pstart, pend; 2150 2151 /* 2152 * madvise behaviors that are implemented in the underlying 2153 * vm_object. 2154 * 2155 * Since we don't clip the vm_map_entry, we have to clip 2156 * the vm_object pindex and count. 2157 */ 2158 for (current = entry; 2159 (current != &map->header) && (current->start < end); 2160 current = current->next 2161 ) { 2162 vm_offset_t useEnd, useStart; 2163 2164 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2165 continue; 2166 2167 pstart = OFF_TO_IDX(current->offset); 2168 pend = pstart + atop(current->end - current->start); 2169 useStart = current->start; 2170 useEnd = current->end; 2171 2172 if (current->start < start) { 2173 pstart += atop(start - current->start); 2174 useStart = start; 2175 } 2176 if (current->end > end) { 2177 pend -= atop(current->end - end); 2178 useEnd = end; 2179 } 2180 2181 if (pstart >= pend) 2182 continue; 2183 2184 /* 2185 * Perform the pmap_advise() before clearing 2186 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2187 * concurrent pmap operation, such as pmap_remove(), 2188 * could clear a reference in the pmap and set 2189 * PGA_REFERENCED on the page before the pmap_advise() 2190 * had completed. Consequently, the page would appear 2191 * referenced based upon an old reference that 2192 * occurred before this pmap_advise() ran. 2193 */ 2194 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2195 pmap_advise(map->pmap, useStart, useEnd, 2196 behav); 2197 2198 vm_object_madvise(current->object.vm_object, pstart, 2199 pend, behav); 2200 2201 /* 2202 * Pre-populate paging structures in the 2203 * WILLNEED case. For wired entries, the 2204 * paging structures are already populated. 2205 */ 2206 if (behav == MADV_WILLNEED && 2207 current->wired_count == 0) { 2208 vm_map_pmap_enter(map, 2209 useStart, 2210 current->protection, 2211 current->object.vm_object, 2212 pstart, 2213 ptoa(pend - pstart), 2214 MAP_PREFAULT_MADVISE 2215 ); 2216 } 2217 } 2218 vm_map_unlock_read(map); 2219 } 2220 return (0); 2221} 2222 2223 2224/* 2225 * vm_map_inherit: 2226 * 2227 * Sets the inheritance of the specified address 2228 * range in the target map. Inheritance 2229 * affects how the map will be shared with 2230 * child maps at the time of vmspace_fork. 2231 */ 2232int 2233vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2234 vm_inherit_t new_inheritance) 2235{ 2236 vm_map_entry_t entry; 2237 vm_map_entry_t temp_entry; 2238 2239 switch (new_inheritance) { 2240 case VM_INHERIT_NONE: 2241 case VM_INHERIT_COPY: 2242 case VM_INHERIT_SHARE: 2243 break; 2244 default: 2245 return (KERN_INVALID_ARGUMENT); 2246 } 2247 if (start == end) 2248 return (KERN_SUCCESS); 2249 vm_map_lock(map); 2250 VM_MAP_RANGE_CHECK(map, start, end); 2251 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2252 entry = temp_entry; 2253 vm_map_clip_start(map, entry, start); 2254 } else 2255 entry = temp_entry->next; 2256 while ((entry != &map->header) && (entry->start < end)) { 2257 vm_map_clip_end(map, entry, end); 2258 entry->inheritance = new_inheritance; 2259 vm_map_simplify_entry(map, entry); 2260 entry = entry->next; 2261 } 2262 vm_map_unlock(map); 2263 return (KERN_SUCCESS); 2264} 2265 2266/* 2267 * vm_map_unwire: 2268 * 2269 * Implements both kernel and user unwiring. 2270 */ 2271int 2272vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2273 int flags) 2274{ 2275 vm_map_entry_t entry, first_entry, tmp_entry; 2276 vm_offset_t saved_start; 2277 unsigned int last_timestamp; 2278 int rv; 2279 boolean_t need_wakeup, result, user_unwire; 2280 2281 if (start == end) 2282 return (KERN_SUCCESS); 2283 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2284 vm_map_lock(map); 2285 VM_MAP_RANGE_CHECK(map, start, end); 2286 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2287 if (flags & VM_MAP_WIRE_HOLESOK) 2288 first_entry = first_entry->next; 2289 else { 2290 vm_map_unlock(map); 2291 return (KERN_INVALID_ADDRESS); 2292 } 2293 } 2294 last_timestamp = map->timestamp; 2295 entry = first_entry; 2296 while (entry != &map->header && entry->start < end) { 2297 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2298 /* 2299 * We have not yet clipped the entry. 2300 */ 2301 saved_start = (start >= entry->start) ? start : 2302 entry->start; 2303 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2304 if (vm_map_unlock_and_wait(map, 0)) { 2305 /* 2306 * Allow interruption of user unwiring? 2307 */ 2308 } 2309 vm_map_lock(map); 2310 if (last_timestamp+1 != map->timestamp) { 2311 /* 2312 * Look again for the entry because the map was 2313 * modified while it was unlocked. 2314 * Specifically, the entry may have been 2315 * clipped, merged, or deleted. 2316 */ 2317 if (!vm_map_lookup_entry(map, saved_start, 2318 &tmp_entry)) { 2319 if (flags & VM_MAP_WIRE_HOLESOK) 2320 tmp_entry = tmp_entry->next; 2321 else { 2322 if (saved_start == start) { 2323 /* 2324 * First_entry has been deleted. 2325 */ 2326 vm_map_unlock(map); 2327 return (KERN_INVALID_ADDRESS); 2328 } 2329 end = saved_start; 2330 rv = KERN_INVALID_ADDRESS; 2331 goto done; 2332 } 2333 } 2334 if (entry == first_entry) 2335 first_entry = tmp_entry; 2336 else 2337 first_entry = NULL; 2338 entry = tmp_entry; 2339 } 2340 last_timestamp = map->timestamp; 2341 continue; 2342 } 2343 vm_map_clip_start(map, entry, start); 2344 vm_map_clip_end(map, entry, end); 2345 /* 2346 * Mark the entry in case the map lock is released. (See 2347 * above.) 2348 */ 2349 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2350 entry->wiring_thread == NULL, 2351 ("owned map entry %p", entry)); 2352 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2353 entry->wiring_thread = curthread; 2354 /* 2355 * Check the map for holes in the specified region. 2356 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2357 */ 2358 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2359 (entry->end < end && (entry->next == &map->header || 2360 entry->next->start > entry->end))) { 2361 end = entry->end; 2362 rv = KERN_INVALID_ADDRESS; 2363 goto done; 2364 } 2365 /* 2366 * If system unwiring, require that the entry is system wired. 2367 */ 2368 if (!user_unwire && 2369 vm_map_entry_system_wired_count(entry) == 0) { 2370 end = entry->end; 2371 rv = KERN_INVALID_ARGUMENT; 2372 goto done; 2373 } 2374 entry = entry->next; 2375 } 2376 rv = KERN_SUCCESS; 2377done: 2378 need_wakeup = FALSE; 2379 if (first_entry == NULL) { 2380 result = vm_map_lookup_entry(map, start, &first_entry); 2381 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2382 first_entry = first_entry->next; 2383 else 2384 KASSERT(result, ("vm_map_unwire: lookup failed")); 2385 } 2386 for (entry = first_entry; entry != &map->header && entry->start < end; 2387 entry = entry->next) { 2388 /* 2389 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2390 * space in the unwired region could have been mapped 2391 * while the map lock was dropped for draining 2392 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2393 * could be simultaneously wiring this new mapping 2394 * entry. Detect these cases and skip any entries 2395 * marked as in transition by us. 2396 */ 2397 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2398 entry->wiring_thread != curthread) { 2399 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2400 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2401 continue; 2402 } 2403 2404 if (rv == KERN_SUCCESS && (!user_unwire || 2405 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2406 if (user_unwire) 2407 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2408 if (entry->wired_count == 1) 2409 vm_map_entry_unwire(map, entry); 2410 else 2411 entry->wired_count--; 2412 } 2413 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2414 ("vm_map_unwire: in-transition flag missing %p", entry)); 2415 KASSERT(entry->wiring_thread == curthread, 2416 ("vm_map_unwire: alien wire %p", entry)); 2417 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2418 entry->wiring_thread = NULL; 2419 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2420 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2421 need_wakeup = TRUE; 2422 } 2423 vm_map_simplify_entry(map, entry); 2424 } 2425 vm_map_unlock(map); 2426 if (need_wakeup) 2427 vm_map_wakeup(map); 2428 return (rv); 2429} 2430 2431/* 2432 * vm_map_wire_entry_failure: 2433 * 2434 * Handle a wiring failure on the given entry. 2435 * 2436 * The map should be locked. 2437 */ 2438static void 2439vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 2440 vm_offset_t failed_addr) 2441{ 2442 2443 VM_MAP_ASSERT_LOCKED(map); 2444 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && 2445 entry->wired_count == 1, 2446 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); 2447 KASSERT(failed_addr < entry->end, 2448 ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); 2449 2450 /* 2451 * If any pages at the start of this entry were successfully wired, 2452 * then unwire them. 2453 */ 2454 if (failed_addr > entry->start) { 2455 pmap_unwire(map->pmap, entry->start, failed_addr); 2456 vm_object_unwire(entry->object.vm_object, entry->offset, 2457 failed_addr - entry->start, PQ_ACTIVE); 2458 } 2459 2460 /* 2461 * Assign an out-of-range value to represent the failure to wire this 2462 * entry. 2463 */ 2464 entry->wired_count = -1; 2465} 2466 2467/* 2468 * vm_map_wire: 2469 * 2470 * Implements both kernel and user wiring. 2471 */ 2472int 2473vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2474 int flags) 2475{ 2476 vm_map_entry_t entry, first_entry, tmp_entry; 2477 vm_offset_t faddr, saved_end, saved_start; 2478 unsigned int last_timestamp; 2479 int rv; 2480 boolean_t need_wakeup, result, user_wire; 2481 vm_prot_t prot; 2482 2483 if (start == end) 2484 return (KERN_SUCCESS); 2485 prot = 0; 2486 if (flags & VM_MAP_WIRE_WRITE) 2487 prot |= VM_PROT_WRITE; 2488 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2489 vm_map_lock(map); 2490 VM_MAP_RANGE_CHECK(map, start, end); 2491 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2492 if (flags & VM_MAP_WIRE_HOLESOK) 2493 first_entry = first_entry->next; 2494 else { 2495 vm_map_unlock(map); 2496 return (KERN_INVALID_ADDRESS); 2497 } 2498 } 2499 last_timestamp = map->timestamp; 2500 entry = first_entry; 2501 while (entry != &map->header && entry->start < end) { 2502 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2503 /* 2504 * We have not yet clipped the entry. 2505 */ 2506 saved_start = (start >= entry->start) ? start : 2507 entry->start; 2508 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2509 if (vm_map_unlock_and_wait(map, 0)) { 2510 /* 2511 * Allow interruption of user wiring? 2512 */ 2513 } 2514 vm_map_lock(map); 2515 if (last_timestamp + 1 != map->timestamp) { 2516 /* 2517 * Look again for the entry because the map was 2518 * modified while it was unlocked. 2519 * Specifically, the entry may have been 2520 * clipped, merged, or deleted. 2521 */ 2522 if (!vm_map_lookup_entry(map, saved_start, 2523 &tmp_entry)) { 2524 if (flags & VM_MAP_WIRE_HOLESOK) 2525 tmp_entry = tmp_entry->next; 2526 else { 2527 if (saved_start == start) { 2528 /* 2529 * first_entry has been deleted. 2530 */ 2531 vm_map_unlock(map); 2532 return (KERN_INVALID_ADDRESS); 2533 } 2534 end = saved_start; 2535 rv = KERN_INVALID_ADDRESS; 2536 goto done; 2537 } 2538 } 2539 if (entry == first_entry) 2540 first_entry = tmp_entry; 2541 else 2542 first_entry = NULL; 2543 entry = tmp_entry; 2544 } 2545 last_timestamp = map->timestamp; 2546 continue; 2547 } 2548 vm_map_clip_start(map, entry, start); 2549 vm_map_clip_end(map, entry, end); 2550 /* 2551 * Mark the entry in case the map lock is released. (See 2552 * above.) 2553 */ 2554 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2555 entry->wiring_thread == NULL, 2556 ("owned map entry %p", entry)); 2557 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2558 entry->wiring_thread = curthread; 2559 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2560 || (entry->protection & prot) != prot) { 2561 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2562 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2563 end = entry->end; 2564 rv = KERN_INVALID_ADDRESS; 2565 goto done; 2566 } 2567 goto next_entry; 2568 } 2569 if (entry->wired_count == 0) { 2570 entry->wired_count++; 2571 saved_start = entry->start; 2572 saved_end = entry->end; 2573 2574 /* 2575 * Release the map lock, relying on the in-transition 2576 * mark. Mark the map busy for fork. 2577 */ 2578 vm_map_busy(map); 2579 vm_map_unlock(map); 2580 2581 faddr = saved_start; 2582 do { 2583 /* 2584 * Simulate a fault to get the page and enter 2585 * it into the physical map. 2586 */ 2587 if ((rv = vm_fault(map, faddr, VM_PROT_NONE, 2588 VM_FAULT_WIRE)) != KERN_SUCCESS) 2589 break; 2590 } while ((faddr += PAGE_SIZE) < saved_end); 2591 vm_map_lock(map); 2592 vm_map_unbusy(map); 2593 if (last_timestamp + 1 != map->timestamp) { 2594 /* 2595 * Look again for the entry because the map was 2596 * modified while it was unlocked. The entry 2597 * may have been clipped, but NOT merged or 2598 * deleted. 2599 */ 2600 result = vm_map_lookup_entry(map, saved_start, 2601 &tmp_entry); 2602 KASSERT(result, ("vm_map_wire: lookup failed")); 2603 if (entry == first_entry) 2604 first_entry = tmp_entry; 2605 else 2606 first_entry = NULL; 2607 entry = tmp_entry; 2608 while (entry->end < saved_end) { 2609 /* 2610 * In case of failure, handle entries 2611 * that were not fully wired here; 2612 * fully wired entries are handled 2613 * later. 2614 */ 2615 if (rv != KERN_SUCCESS && 2616 faddr < entry->end) 2617 vm_map_wire_entry_failure(map, 2618 entry, faddr); 2619 entry = entry->next; 2620 } 2621 } 2622 last_timestamp = map->timestamp; 2623 if (rv != KERN_SUCCESS) { 2624 vm_map_wire_entry_failure(map, entry, faddr); 2625 end = entry->end; 2626 goto done; 2627 } 2628 } else if (!user_wire || 2629 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2630 entry->wired_count++; 2631 } 2632 /* 2633 * Check the map for holes in the specified region. 2634 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2635 */ 2636 next_entry: 2637 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2638 (entry->end < end && (entry->next == &map->header || 2639 entry->next->start > entry->end))) { 2640 end = entry->end; 2641 rv = KERN_INVALID_ADDRESS; 2642 goto done; 2643 } 2644 entry = entry->next; 2645 } 2646 rv = KERN_SUCCESS; 2647done: 2648 need_wakeup = FALSE; 2649 if (first_entry == NULL) { 2650 result = vm_map_lookup_entry(map, start, &first_entry); 2651 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2652 first_entry = first_entry->next; 2653 else 2654 KASSERT(result, ("vm_map_wire: lookup failed")); 2655 } 2656 for (entry = first_entry; entry != &map->header && entry->start < end; 2657 entry = entry->next) { 2658 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2659 goto next_entry_done; 2660 2661 /* 2662 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2663 * space in the unwired region could have been mapped 2664 * while the map lock was dropped for faulting in the 2665 * pages or draining MAP_ENTRY_IN_TRANSITION. 2666 * Moreover, another thread could be simultaneously 2667 * wiring this new mapping entry. Detect these cases 2668 * and skip any entries marked as in transition by us. 2669 */ 2670 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2671 entry->wiring_thread != curthread) { 2672 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2673 ("vm_map_wire: !HOLESOK and new/changed entry")); 2674 continue; 2675 } 2676 2677 if (rv == KERN_SUCCESS) { 2678 if (user_wire) 2679 entry->eflags |= MAP_ENTRY_USER_WIRED; 2680 } else if (entry->wired_count == -1) { 2681 /* 2682 * Wiring failed on this entry. Thus, unwiring is 2683 * unnecessary. 2684 */ 2685 entry->wired_count = 0; 2686 } else if (!user_wire || 2687 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2688 /* 2689 * Undo the wiring. Wiring succeeded on this entry 2690 * but failed on a later entry. 2691 */ 2692 if (entry->wired_count == 1) 2693 vm_map_entry_unwire(map, entry); 2694 else 2695 entry->wired_count--; 2696 } 2697 next_entry_done: 2698 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2699 ("vm_map_wire: in-transition flag missing %p", entry)); 2700 KASSERT(entry->wiring_thread == curthread, 2701 ("vm_map_wire: alien wire %p", entry)); 2702 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2703 MAP_ENTRY_WIRE_SKIPPED); 2704 entry->wiring_thread = NULL; 2705 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2706 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2707 need_wakeup = TRUE; 2708 } 2709 vm_map_simplify_entry(map, entry); 2710 } 2711 vm_map_unlock(map); 2712 if (need_wakeup) 2713 vm_map_wakeup(map); 2714 return (rv); 2715} 2716 2717/* 2718 * vm_map_sync 2719 * 2720 * Push any dirty cached pages in the address range to their pager. 2721 * If syncio is TRUE, dirty pages are written synchronously. 2722 * If invalidate is TRUE, any cached pages are freed as well. 2723 * 2724 * If the size of the region from start to end is zero, we are 2725 * supposed to flush all modified pages within the region containing 2726 * start. Unfortunately, a region can be split or coalesced with 2727 * neighboring regions, making it difficult to determine what the 2728 * original region was. Therefore, we approximate this requirement by 2729 * flushing the current region containing start. 2730 * 2731 * Returns an error if any part of the specified range is not mapped. 2732 */ 2733int 2734vm_map_sync( 2735 vm_map_t map, 2736 vm_offset_t start, 2737 vm_offset_t end, 2738 boolean_t syncio, 2739 boolean_t invalidate) 2740{ 2741 vm_map_entry_t current; 2742 vm_map_entry_t entry; 2743 vm_size_t size; 2744 vm_object_t object; 2745 vm_ooffset_t offset; 2746 unsigned int last_timestamp; 2747 boolean_t failed; 2748 2749 vm_map_lock_read(map); 2750 VM_MAP_RANGE_CHECK(map, start, end); 2751 if (!vm_map_lookup_entry(map, start, &entry)) { 2752 vm_map_unlock_read(map); 2753 return (KERN_INVALID_ADDRESS); 2754 } else if (start == end) { 2755 start = entry->start; 2756 end = entry->end; 2757 } 2758 /* 2759 * Make a first pass to check for user-wired memory and holes. 2760 */ 2761 for (current = entry; current != &map->header && current->start < end; 2762 current = current->next) { 2763 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2764 vm_map_unlock_read(map); 2765 return (KERN_INVALID_ARGUMENT); 2766 } 2767 if (end > current->end && 2768 (current->next == &map->header || 2769 current->end != current->next->start)) { 2770 vm_map_unlock_read(map); 2771 return (KERN_INVALID_ADDRESS); 2772 } 2773 } 2774 2775 if (invalidate) 2776 pmap_remove(map->pmap, start, end); 2777 failed = FALSE; 2778 2779 /* 2780 * Make a second pass, cleaning/uncaching pages from the indicated 2781 * objects as we go. 2782 */ 2783 for (current = entry; current != &map->header && current->start < end;) { 2784 offset = current->offset + (start - current->start); 2785 size = (end <= current->end ? end : current->end) - start; 2786 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2787 vm_map_t smap; 2788 vm_map_entry_t tentry; 2789 vm_size_t tsize; 2790 2791 smap = current->object.sub_map; 2792 vm_map_lock_read(smap); 2793 (void) vm_map_lookup_entry(smap, offset, &tentry); 2794 tsize = tentry->end - offset; 2795 if (tsize < size) 2796 size = tsize; 2797 object = tentry->object.vm_object; 2798 offset = tentry->offset + (offset - tentry->start); 2799 vm_map_unlock_read(smap); 2800 } else { 2801 object = current->object.vm_object; 2802 } 2803 vm_object_reference(object); 2804 last_timestamp = map->timestamp; 2805 vm_map_unlock_read(map); 2806 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2807 failed = TRUE; 2808 start += size; 2809 vm_object_deallocate(object); 2810 vm_map_lock_read(map); 2811 if (last_timestamp == map->timestamp || 2812 !vm_map_lookup_entry(map, start, ¤t)) 2813 current = current->next; 2814 } 2815 2816 vm_map_unlock_read(map); 2817 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2818} 2819 2820/* 2821 * vm_map_entry_unwire: [ internal use only ] 2822 * 2823 * Make the region specified by this entry pageable. 2824 * 2825 * The map in question should be locked. 2826 * [This is the reason for this routine's existence.] 2827 */ 2828static void 2829vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2830{ 2831 2832 VM_MAP_ASSERT_LOCKED(map); 2833 KASSERT(entry->wired_count > 0, 2834 ("vm_map_entry_unwire: entry %p isn't wired", entry)); 2835 pmap_unwire(map->pmap, entry->start, entry->end); 2836 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end - 2837 entry->start, PQ_ACTIVE); 2838 entry->wired_count = 0; 2839} 2840 2841static void 2842vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2843{ 2844 2845 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2846 vm_object_deallocate(entry->object.vm_object); 2847 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2848} 2849 2850/* 2851 * vm_map_entry_delete: [ internal use only ] 2852 * 2853 * Deallocate the given entry from the target map. 2854 */ 2855static void 2856vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2857{ 2858 vm_object_t object; 2859 vm_pindex_t offidxstart, offidxend, count, size1; 2860 vm_ooffset_t size; 2861 2862 vm_map_entry_unlink(map, entry); 2863 object = entry->object.vm_object; 2864 size = entry->end - entry->start; 2865 map->size -= size; 2866 2867 if (entry->cred != NULL) { 2868 swap_release_by_cred(size, entry->cred); 2869 crfree(entry->cred); 2870 } 2871 2872 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2873 (object != NULL)) { 2874 KASSERT(entry->cred == NULL || object->cred == NULL || 2875 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2876 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2877 count = OFF_TO_IDX(size); 2878 offidxstart = OFF_TO_IDX(entry->offset); 2879 offidxend = offidxstart + count; 2880 VM_OBJECT_WLOCK(object); 2881 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT | 2882 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2883 object == kernel_object || object == kmem_object)) { 2884 vm_object_collapse(object); 2885 2886 /* 2887 * The option OBJPR_NOTMAPPED can be passed here 2888 * because vm_map_delete() already performed 2889 * pmap_remove() on the only mapping to this range 2890 * of pages. 2891 */ 2892 vm_object_page_remove(object, offidxstart, offidxend, 2893 OBJPR_NOTMAPPED); 2894 if (object->type == OBJT_SWAP) 2895 swap_pager_freespace(object, offidxstart, 2896 count); 2897 if (offidxend >= object->size && 2898 offidxstart < object->size) { 2899 size1 = object->size; 2900 object->size = offidxstart; 2901 if (object->cred != NULL) { 2902 size1 -= object->size; 2903 KASSERT(object->charge >= ptoa(size1), 2904 ("object %p charge < 0", object)); 2905 swap_release_by_cred(ptoa(size1), 2906 object->cred); 2907 object->charge -= ptoa(size1); 2908 } 2909 } 2910 } 2911 VM_OBJECT_WUNLOCK(object); 2912 } else 2913 entry->object.vm_object = NULL; 2914 if (map->system_map) 2915 vm_map_entry_deallocate(entry, TRUE); 2916 else { 2917 entry->next = curthread->td_map_def_user; 2918 curthread->td_map_def_user = entry; 2919 } 2920} 2921 2922/* 2923 * vm_map_delete: [ internal use only ] 2924 * 2925 * Deallocates the given address range from the target 2926 * map. 2927 */ 2928int 2929vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2930{ 2931 vm_map_entry_t entry; 2932 vm_map_entry_t first_entry; 2933 2934 VM_MAP_ASSERT_LOCKED(map); 2935 if (start == end) 2936 return (KERN_SUCCESS); 2937 2938 /* 2939 * Find the start of the region, and clip it 2940 */ 2941 if (!vm_map_lookup_entry(map, start, &first_entry)) 2942 entry = first_entry->next; 2943 else { 2944 entry = first_entry; 2945 vm_map_clip_start(map, entry, start); 2946 } 2947 2948 /* 2949 * Step through all entries in this region 2950 */ 2951 while ((entry != &map->header) && (entry->start < end)) { 2952 vm_map_entry_t next; 2953 2954 /* 2955 * Wait for wiring or unwiring of an entry to complete. 2956 * Also wait for any system wirings to disappear on 2957 * user maps. 2958 */ 2959 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2960 (vm_map_pmap(map) != kernel_pmap && 2961 vm_map_entry_system_wired_count(entry) != 0)) { 2962 unsigned int last_timestamp; 2963 vm_offset_t saved_start; 2964 vm_map_entry_t tmp_entry; 2965 2966 saved_start = entry->start; 2967 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2968 last_timestamp = map->timestamp; 2969 (void) vm_map_unlock_and_wait(map, 0); 2970 vm_map_lock(map); 2971 if (last_timestamp + 1 != map->timestamp) { 2972 /* 2973 * Look again for the entry because the map was 2974 * modified while it was unlocked. 2975 * Specifically, the entry may have been 2976 * clipped, merged, or deleted. 2977 */ 2978 if (!vm_map_lookup_entry(map, saved_start, 2979 &tmp_entry)) 2980 entry = tmp_entry->next; 2981 else { 2982 entry = tmp_entry; 2983 vm_map_clip_start(map, entry, 2984 saved_start); 2985 } 2986 } 2987 continue; 2988 } 2989 vm_map_clip_end(map, entry, end); 2990 2991 next = entry->next; 2992 2993 /* 2994 * Unwire before removing addresses from the pmap; otherwise, 2995 * unwiring will put the entries back in the pmap. 2996 */ 2997 if (entry->wired_count != 0) { 2998 vm_map_entry_unwire(map, entry); 2999 } 3000 3001 pmap_remove(map->pmap, entry->start, entry->end); 3002 3003 /* 3004 * Delete the entry only after removing all pmap 3005 * entries pointing to its pages. (Otherwise, its 3006 * page frames may be reallocated, and any modify bits 3007 * will be set in the wrong object!) 3008 */ 3009 vm_map_entry_delete(map, entry); 3010 entry = next; 3011 } 3012 return (KERN_SUCCESS); 3013} 3014 3015/* 3016 * vm_map_remove: 3017 * 3018 * Remove the given address range from the target map. 3019 * This is the exported form of vm_map_delete. 3020 */ 3021int 3022vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 3023{ 3024 int result; 3025 3026 vm_map_lock(map); 3027 VM_MAP_RANGE_CHECK(map, start, end); 3028 result = vm_map_delete(map, start, end); 3029 vm_map_unlock(map); 3030 return (result); 3031} 3032 3033/* 3034 * vm_map_check_protection: 3035 * 3036 * Assert that the target map allows the specified privilege on the 3037 * entire address region given. The entire region must be allocated. 3038 * 3039 * WARNING! This code does not and should not check whether the 3040 * contents of the region is accessible. For example a smaller file 3041 * might be mapped into a larger address space. 3042 * 3043 * NOTE! This code is also called by munmap(). 3044 * 3045 * The map must be locked. A read lock is sufficient. 3046 */ 3047boolean_t 3048vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3049 vm_prot_t protection) 3050{ 3051 vm_map_entry_t entry; 3052 vm_map_entry_t tmp_entry; 3053 3054 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3055 return (FALSE); 3056 entry = tmp_entry; 3057 3058 while (start < end) { 3059 if (entry == &map->header) 3060 return (FALSE); 3061 /* 3062 * No holes allowed! 3063 */ 3064 if (start < entry->start) 3065 return (FALSE); 3066 /* 3067 * Check protection associated with entry. 3068 */ 3069 if ((entry->protection & protection) != protection) 3070 return (FALSE); 3071 /* go to next entry */ 3072 start = entry->end; 3073 entry = entry->next; 3074 } 3075 return (TRUE); 3076} 3077 3078/* 3079 * vm_map_copy_entry: 3080 * 3081 * Copies the contents of the source entry to the destination 3082 * entry. The entries *must* be aligned properly. 3083 */ 3084static void 3085vm_map_copy_entry( 3086 vm_map_t src_map, 3087 vm_map_t dst_map, 3088 vm_map_entry_t src_entry, 3089 vm_map_entry_t dst_entry, 3090 vm_ooffset_t *fork_charge) 3091{ 3092 vm_object_t src_object; 3093 vm_map_entry_t fake_entry; 3094 vm_offset_t size; 3095 struct ucred *cred; 3096 int charged; 3097 3098 VM_MAP_ASSERT_LOCKED(dst_map); 3099 3100 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3101 return; 3102 3103 if (src_entry->wired_count == 0 || 3104 (src_entry->protection & VM_PROT_WRITE) == 0) { 3105 /* 3106 * If the source entry is marked needs_copy, it is already 3107 * write-protected. 3108 */ 3109 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3110 (src_entry->protection & VM_PROT_WRITE) != 0) { 3111 pmap_protect(src_map->pmap, 3112 src_entry->start, 3113 src_entry->end, 3114 src_entry->protection & ~VM_PROT_WRITE); 3115 } 3116 3117 /* 3118 * Make a copy of the object. 3119 */ 3120 size = src_entry->end - src_entry->start; 3121 if ((src_object = src_entry->object.vm_object) != NULL) { 3122 VM_OBJECT_WLOCK(src_object); 3123 charged = ENTRY_CHARGED(src_entry); 3124 if (src_object->handle == NULL && 3125 (src_object->type == OBJT_DEFAULT || 3126 src_object->type == OBJT_SWAP)) { 3127 vm_object_collapse(src_object); 3128 if ((src_object->flags & (OBJ_NOSPLIT | 3129 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3130 vm_object_split(src_entry); 3131 src_object = 3132 src_entry->object.vm_object; 3133 } 3134 } 3135 vm_object_reference_locked(src_object); 3136 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3137 if (src_entry->cred != NULL && 3138 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3139 KASSERT(src_object->cred == NULL, 3140 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3141 src_object)); 3142 src_object->cred = src_entry->cred; 3143 src_object->charge = size; 3144 } 3145 VM_OBJECT_WUNLOCK(src_object); 3146 dst_entry->object.vm_object = src_object; 3147 if (charged) { 3148 cred = curthread->td_ucred; 3149 crhold(cred); 3150 dst_entry->cred = cred; 3151 *fork_charge += size; 3152 if (!(src_entry->eflags & 3153 MAP_ENTRY_NEEDS_COPY)) { 3154 crhold(cred); 3155 src_entry->cred = cred; 3156 *fork_charge += size; 3157 } 3158 } 3159 src_entry->eflags |= MAP_ENTRY_COW | 3160 MAP_ENTRY_NEEDS_COPY; 3161 dst_entry->eflags |= MAP_ENTRY_COW | 3162 MAP_ENTRY_NEEDS_COPY; 3163 dst_entry->offset = src_entry->offset; 3164 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3165 /* 3166 * MAP_ENTRY_VN_WRITECNT cannot 3167 * indicate write reference from 3168 * src_entry, since the entry is 3169 * marked as needs copy. Allocate a 3170 * fake entry that is used to 3171 * decrement object->un_pager.vnp.writecount 3172 * at the appropriate time. Attach 3173 * fake_entry to the deferred list. 3174 */ 3175 fake_entry = vm_map_entry_create(dst_map); 3176 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3177 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3178 vm_object_reference(src_object); 3179 fake_entry->object.vm_object = src_object; 3180 fake_entry->start = src_entry->start; 3181 fake_entry->end = src_entry->end; 3182 fake_entry->next = curthread->td_map_def_user; 3183 curthread->td_map_def_user = fake_entry; 3184 } 3185 } else { 3186 dst_entry->object.vm_object = NULL; 3187 dst_entry->offset = 0; 3188 if (src_entry->cred != NULL) { 3189 dst_entry->cred = curthread->td_ucred; 3190 crhold(dst_entry->cred); 3191 *fork_charge += size; 3192 } 3193 } 3194 3195 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3196 dst_entry->end - dst_entry->start, src_entry->start); 3197 } else { 3198 /* 3199 * We don't want to make writeable wired pages copy-on-write. 3200 * Immediately copy these pages into the new map by simulating 3201 * page faults. The new pages are pageable. 3202 */ 3203 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3204 fork_charge); 3205 } 3206} 3207 3208/* 3209 * vmspace_map_entry_forked: 3210 * Update the newly-forked vmspace each time a map entry is inherited 3211 * or copied. The values for vm_dsize and vm_tsize are approximate 3212 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3213 */ 3214static void 3215vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3216 vm_map_entry_t entry) 3217{ 3218 vm_size_t entrysize; 3219 vm_offset_t newend; 3220 3221 entrysize = entry->end - entry->start; 3222 vm2->vm_map.size += entrysize; 3223 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3224 vm2->vm_ssize += btoc(entrysize); 3225 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3226 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3227 newend = MIN(entry->end, 3228 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3229 vm2->vm_dsize += btoc(newend - entry->start); 3230 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3231 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3232 newend = MIN(entry->end, 3233 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3234 vm2->vm_tsize += btoc(newend - entry->start); 3235 } 3236} 3237 3238/* 3239 * vmspace_fork: 3240 * Create a new process vmspace structure and vm_map 3241 * based on those of an existing process. The new map 3242 * is based on the old map, according to the inheritance 3243 * values on the regions in that map. 3244 * 3245 * XXX It might be worth coalescing the entries added to the new vmspace. 3246 * 3247 * The source map must not be locked. 3248 */ 3249struct vmspace * 3250vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3251{ 3252 struct vmspace *vm2; 3253 vm_map_t new_map, old_map; 3254 vm_map_entry_t new_entry, old_entry; 3255 vm_object_t object; 3256 int locked; 3257 3258 old_map = &vm1->vm_map; 3259 /* Copy immutable fields of vm1 to vm2. */ 3260 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3261 if (vm2 == NULL) 3262 return (NULL); 3263 vm2->vm_taddr = vm1->vm_taddr; 3264 vm2->vm_daddr = vm1->vm_daddr; 3265 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3266 vm_map_lock(old_map); 3267 if (old_map->busy) 3268 vm_map_wait_busy(old_map); 3269 new_map = &vm2->vm_map; 3270 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3271 KASSERT(locked, ("vmspace_fork: lock failed")); 3272 3273 old_entry = old_map->header.next; 3274 3275 while (old_entry != &old_map->header) { 3276 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3277 panic("vm_map_fork: encountered a submap"); 3278 3279 switch (old_entry->inheritance) { 3280 case VM_INHERIT_NONE: 3281 break; 3282 3283 case VM_INHERIT_SHARE: 3284 /* 3285 * Clone the entry, creating the shared object if necessary. 3286 */ 3287 object = old_entry->object.vm_object; 3288 if (object == NULL) { 3289 object = vm_object_allocate(OBJT_DEFAULT, 3290 atop(old_entry->end - old_entry->start)); 3291 old_entry->object.vm_object = object; 3292 old_entry->offset = 0; 3293 if (old_entry->cred != NULL) { 3294 object->cred = old_entry->cred; 3295 object->charge = old_entry->end - 3296 old_entry->start; 3297 old_entry->cred = NULL; 3298 } 3299 } 3300 3301 /* 3302 * Add the reference before calling vm_object_shadow 3303 * to insure that a shadow object is created. 3304 */ 3305 vm_object_reference(object); 3306 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3307 vm_object_shadow(&old_entry->object.vm_object, 3308 &old_entry->offset, 3309 old_entry->end - old_entry->start); 3310 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3311 /* Transfer the second reference too. */ 3312 vm_object_reference( 3313 old_entry->object.vm_object); 3314 3315 /* 3316 * As in vm_map_simplify_entry(), the 3317 * vnode lock will not be acquired in 3318 * this call to vm_object_deallocate(). 3319 */ 3320 vm_object_deallocate(object); 3321 object = old_entry->object.vm_object; 3322 } 3323 VM_OBJECT_WLOCK(object); 3324 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3325 if (old_entry->cred != NULL) { 3326 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3327 object->cred = old_entry->cred; 3328 object->charge = old_entry->end - old_entry->start; 3329 old_entry->cred = NULL; 3330 } 3331 3332 /* 3333 * Assert the correct state of the vnode 3334 * v_writecount while the object is locked, to 3335 * not relock it later for the assertion 3336 * correctness. 3337 */ 3338 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3339 object->type == OBJT_VNODE) { 3340 KASSERT(((struct vnode *)object->handle)-> 3341 v_writecount > 0, 3342 ("vmspace_fork: v_writecount %p", object)); 3343 KASSERT(object->un_pager.vnp.writemappings > 0, 3344 ("vmspace_fork: vnp.writecount %p", 3345 object)); 3346 } 3347 VM_OBJECT_WUNLOCK(object); 3348 3349 /* 3350 * Clone the entry, referencing the shared object. 3351 */ 3352 new_entry = vm_map_entry_create(new_map); 3353 *new_entry = *old_entry; 3354 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3355 MAP_ENTRY_IN_TRANSITION); 3356 new_entry->wiring_thread = NULL; 3357 new_entry->wired_count = 0; 3358 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3359 vnode_pager_update_writecount(object, 3360 new_entry->start, new_entry->end); 3361 } 3362 3363 /* 3364 * Insert the entry into the new map -- we know we're 3365 * inserting at the end of the new map. 3366 */ 3367 vm_map_entry_link(new_map, new_map->header.prev, 3368 new_entry); 3369 vmspace_map_entry_forked(vm1, vm2, new_entry); 3370 3371 /* 3372 * Update the physical map 3373 */ 3374 pmap_copy(new_map->pmap, old_map->pmap, 3375 new_entry->start, 3376 (old_entry->end - old_entry->start), 3377 old_entry->start); 3378 break; 3379 3380 case VM_INHERIT_COPY: 3381 /* 3382 * Clone the entry and link into the map. 3383 */ 3384 new_entry = vm_map_entry_create(new_map); 3385 *new_entry = *old_entry; 3386 /* 3387 * Copied entry is COW over the old object. 3388 */ 3389 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3390 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3391 new_entry->wiring_thread = NULL; 3392 new_entry->wired_count = 0; 3393 new_entry->object.vm_object = NULL; 3394 new_entry->cred = NULL; 3395 vm_map_entry_link(new_map, new_map->header.prev, 3396 new_entry); 3397 vmspace_map_entry_forked(vm1, vm2, new_entry); 3398 vm_map_copy_entry(old_map, new_map, old_entry, 3399 new_entry, fork_charge); 3400 break; 3401 } 3402 old_entry = old_entry->next; 3403 } 3404 /* 3405 * Use inlined vm_map_unlock() to postpone handling the deferred 3406 * map entries, which cannot be done until both old_map and 3407 * new_map locks are released. 3408 */ 3409 sx_xunlock(&old_map->lock); 3410 sx_xunlock(&new_map->lock); 3411 vm_map_process_deferred(); 3412 3413 return (vm2); 3414} 3415 3416int 3417vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3418 vm_prot_t prot, vm_prot_t max, int cow) 3419{ 3420 vm_size_t growsize, init_ssize; 3421 rlim_t lmemlim, vmemlim; 3422 int rv; 3423 3424 growsize = sgrowsiz; 3425 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3426 vm_map_lock(map); 3427 PROC_LOCK(curproc); 3428 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3429 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3430 PROC_UNLOCK(curproc); 3431 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3432 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3433 rv = KERN_NO_SPACE; 3434 goto out; 3435 } 3436 } 3437 /* If we would blow our VMEM resource limit, no go */ 3438 if (map->size + init_ssize > vmemlim) { 3439 rv = KERN_NO_SPACE; 3440 goto out; 3441 } 3442 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3443 max, cow); 3444out: 3445 vm_map_unlock(map); 3446 return (rv); 3447} 3448 3449static int 3450vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3451 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3452{ 3453 vm_map_entry_t new_entry, prev_entry; 3454 vm_offset_t bot, top; 3455 vm_size_t init_ssize; 3456 int orient, rv; 3457 3458 /* 3459 * The stack orientation is piggybacked with the cow argument. 3460 * Extract it into orient and mask the cow argument so that we 3461 * don't pass it around further. 3462 * NOTE: We explicitly allow bi-directional stacks. 3463 */ 3464 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3465 KASSERT(orient != 0, ("No stack grow direction")); 3466 3467 if (addrbos < vm_map_min(map) || 3468 addrbos > vm_map_max(map) || 3469 addrbos + max_ssize < addrbos) 3470 return (KERN_NO_SPACE); 3471 3472 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3473 3474 /* If addr is already mapped, no go */ 3475 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3476 return (KERN_NO_SPACE); 3477 3478 /* 3479 * If we can't accomodate max_ssize in the current mapping, no go. 3480 * However, we need to be aware that subsequent user mappings might 3481 * map into the space we have reserved for stack, and currently this 3482 * space is not protected. 3483 * 3484 * Hopefully we will at least detect this condition when we try to 3485 * grow the stack. 3486 */ 3487 if ((prev_entry->next != &map->header) && 3488 (prev_entry->next->start < addrbos + max_ssize)) 3489 return (KERN_NO_SPACE); 3490 3491 /* 3492 * We initially map a stack of only init_ssize. We will grow as 3493 * needed later. Depending on the orientation of the stack (i.e. 3494 * the grow direction) we either map at the top of the range, the 3495 * bottom of the range or in the middle. 3496 * 3497 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3498 * and cow to be 0. Possibly we should eliminate these as input 3499 * parameters, and just pass these values here in the insert call. 3500 */ 3501 if (orient == MAP_STACK_GROWS_DOWN) 3502 bot = addrbos + max_ssize - init_ssize; 3503 else if (orient == MAP_STACK_GROWS_UP) 3504 bot = addrbos; 3505 else 3506 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3507 top = bot + init_ssize; 3508 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3509 3510 /* Now set the avail_ssize amount. */ 3511 if (rv == KERN_SUCCESS) { 3512 new_entry = prev_entry->next; 3513 if (new_entry->end != top || new_entry->start != bot) 3514 panic("Bad entry start/end for new stack entry"); 3515 3516 new_entry->avail_ssize = max_ssize - init_ssize; 3517 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 || 3518 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, 3519 ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 3520 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 || 3521 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0, 3522 ("new entry lacks MAP_ENTRY_GROWS_UP")); 3523 } 3524 3525 return (rv); 3526} 3527 3528static int stack_guard_page = 0; 3529TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3530SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3531 &stack_guard_page, 0, 3532 "Insert stack guard page ahead of the growable segments."); 3533 3534/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3535 * desired address is already mapped, or if we successfully grow 3536 * the stack. Also returns KERN_SUCCESS if addr is outside the 3537 * stack range (this is strange, but preserves compatibility with 3538 * the grow function in vm_machdep.c). 3539 */ 3540int 3541vm_map_growstack(struct proc *p, vm_offset_t addr) 3542{ 3543 vm_map_entry_t next_entry, prev_entry; 3544 vm_map_entry_t new_entry, stack_entry; 3545 struct vmspace *vm = p->p_vmspace; 3546 vm_map_t map = &vm->vm_map; 3547 vm_offset_t end; 3548 vm_size_t growsize; 3549 size_t grow_amount, max_grow; 3550 rlim_t lmemlim, stacklim, vmemlim; 3551 int is_procstack, rv; 3552 struct ucred *cred; 3553#ifdef notyet 3554 uint64_t limit; 3555#endif 3556#ifdef RACCT 3557 int error; 3558#endif 3559 3560Retry: 3561 PROC_LOCK(p); 3562 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3563 stacklim = lim_cur(p, RLIMIT_STACK); 3564 vmemlim = lim_cur(p, RLIMIT_VMEM); 3565 PROC_UNLOCK(p); 3566 3567 vm_map_lock_read(map); 3568 3569 /* If addr is already in the entry range, no need to grow.*/ 3570 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3571 vm_map_unlock_read(map); 3572 return (KERN_SUCCESS); 3573 } 3574 3575 next_entry = prev_entry->next; 3576 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3577 /* 3578 * This entry does not grow upwards. Since the address lies 3579 * beyond this entry, the next entry (if one exists) has to 3580 * be a downward growable entry. The entry list header is 3581 * never a growable entry, so it suffices to check the flags. 3582 */ 3583 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3584 vm_map_unlock_read(map); 3585 return (KERN_SUCCESS); 3586 } 3587 stack_entry = next_entry; 3588 } else { 3589 /* 3590 * This entry grows upward. If the next entry does not at 3591 * least grow downwards, this is the entry we need to grow. 3592 * otherwise we have two possible choices and we have to 3593 * select one. 3594 */ 3595 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3596 /* 3597 * We have two choices; grow the entry closest to 3598 * the address to minimize the amount of growth. 3599 */ 3600 if (addr - prev_entry->end <= next_entry->start - addr) 3601 stack_entry = prev_entry; 3602 else 3603 stack_entry = next_entry; 3604 } else 3605 stack_entry = prev_entry; 3606 } 3607 3608 if (stack_entry == next_entry) { 3609 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3610 KASSERT(addr < stack_entry->start, ("foo")); 3611 end = (prev_entry != &map->header) ? prev_entry->end : 3612 stack_entry->start - stack_entry->avail_ssize; 3613 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3614 max_grow = stack_entry->start - end; 3615 } else { 3616 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3617 KASSERT(addr >= stack_entry->end, ("foo")); 3618 end = (next_entry != &map->header) ? next_entry->start : 3619 stack_entry->end + stack_entry->avail_ssize; 3620 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3621 max_grow = end - stack_entry->end; 3622 } 3623 3624 if (grow_amount > stack_entry->avail_ssize) { 3625 vm_map_unlock_read(map); 3626 return (KERN_NO_SPACE); 3627 } 3628 3629 /* 3630 * If there is no longer enough space between the entries nogo, and 3631 * adjust the available space. Note: this should only happen if the 3632 * user has mapped into the stack area after the stack was created, 3633 * and is probably an error. 3634 * 3635 * This also effectively destroys any guard page the user might have 3636 * intended by limiting the stack size. 3637 */ 3638 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3639 if (vm_map_lock_upgrade(map)) 3640 goto Retry; 3641 3642 stack_entry->avail_ssize = max_grow; 3643 3644 vm_map_unlock(map); 3645 return (KERN_NO_SPACE); 3646 } 3647 3648 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr && 3649 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0; 3650 3651 /* 3652 * If this is the main process stack, see if we're over the stack 3653 * limit. 3654 */ 3655 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3656 vm_map_unlock_read(map); 3657 return (KERN_NO_SPACE); 3658 } 3659#ifdef RACCT 3660 if (racct_enable) { 3661 PROC_LOCK(p); 3662 if (is_procstack && racct_set(p, RACCT_STACK, 3663 ctob(vm->vm_ssize) + grow_amount)) { 3664 PROC_UNLOCK(p); 3665 vm_map_unlock_read(map); 3666 return (KERN_NO_SPACE); 3667 } 3668 PROC_UNLOCK(p); 3669 } 3670#endif 3671 3672 /* Round up the grow amount modulo sgrowsiz */ 3673 growsize = sgrowsiz; 3674 grow_amount = roundup(grow_amount, growsize); 3675 if (grow_amount > stack_entry->avail_ssize) 3676 grow_amount = stack_entry->avail_ssize; 3677 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3678 grow_amount = trunc_page((vm_size_t)stacklim) - 3679 ctob(vm->vm_ssize); 3680 } 3681#ifdef notyet 3682 PROC_LOCK(p); 3683 limit = racct_get_available(p, RACCT_STACK); 3684 PROC_UNLOCK(p); 3685 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3686 grow_amount = limit - ctob(vm->vm_ssize); 3687#endif 3688 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3689 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3690 vm_map_unlock_read(map); 3691 rv = KERN_NO_SPACE; 3692 goto out; 3693 } 3694#ifdef RACCT 3695 if (racct_enable) { 3696 PROC_LOCK(p); 3697 if (racct_set(p, RACCT_MEMLOCK, 3698 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3699 PROC_UNLOCK(p); 3700 vm_map_unlock_read(map); 3701 rv = KERN_NO_SPACE; 3702 goto out; 3703 } 3704 PROC_UNLOCK(p); 3705 } 3706#endif 3707 } 3708 /* If we would blow our VMEM resource limit, no go */ 3709 if (map->size + grow_amount > vmemlim) { 3710 vm_map_unlock_read(map); 3711 rv = KERN_NO_SPACE; 3712 goto out; 3713 } 3714#ifdef RACCT 3715 if (racct_enable) { 3716 PROC_LOCK(p); 3717 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3718 PROC_UNLOCK(p); 3719 vm_map_unlock_read(map); 3720 rv = KERN_NO_SPACE; 3721 goto out; 3722 } 3723 PROC_UNLOCK(p); 3724 } 3725#endif 3726 3727 if (vm_map_lock_upgrade(map)) 3728 goto Retry; 3729 3730 if (stack_entry == next_entry) { 3731 /* 3732 * Growing downward. 3733 */ 3734 /* Get the preliminary new entry start value */ 3735 addr = stack_entry->start - grow_amount; 3736 3737 /* 3738 * If this puts us into the previous entry, cut back our 3739 * growth to the available space. Also, see the note above. 3740 */ 3741 if (addr < end) { 3742 stack_entry->avail_ssize = max_grow; 3743 addr = end; 3744 if (stack_guard_page) 3745 addr += PAGE_SIZE; 3746 } 3747 3748 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3749 next_entry->protection, next_entry->max_protection, 3750 MAP_STACK_GROWS_DOWN); 3751 3752 /* Adjust the available stack space by the amount we grew. */ 3753 if (rv == KERN_SUCCESS) { 3754 new_entry = prev_entry->next; 3755 KASSERT(new_entry == stack_entry->prev, ("foo")); 3756 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3757 KASSERT(new_entry->start == addr, ("foo")); 3758 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 3759 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 3760 grow_amount = new_entry->end - new_entry->start; 3761 new_entry->avail_ssize = stack_entry->avail_ssize - 3762 grow_amount; 3763 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3764 } 3765 } else { 3766 /* 3767 * Growing upward. 3768 */ 3769 addr = stack_entry->end + grow_amount; 3770 3771 /* 3772 * If this puts us into the next entry, cut back our growth 3773 * to the available space. Also, see the note above. 3774 */ 3775 if (addr > end) { 3776 stack_entry->avail_ssize = end - stack_entry->end; 3777 addr = end; 3778 if (stack_guard_page) 3779 addr -= PAGE_SIZE; 3780 } 3781 3782 grow_amount = addr - stack_entry->end; 3783 cred = stack_entry->cred; 3784 if (cred == NULL && stack_entry->object.vm_object != NULL) 3785 cred = stack_entry->object.vm_object->cred; 3786 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3787 rv = KERN_NO_SPACE; 3788 /* Grow the underlying object if applicable. */ 3789 else if (stack_entry->object.vm_object == NULL || 3790 vm_object_coalesce(stack_entry->object.vm_object, 3791 stack_entry->offset, 3792 (vm_size_t)(stack_entry->end - stack_entry->start), 3793 (vm_size_t)grow_amount, cred != NULL)) { 3794 map->size += (addr - stack_entry->end); 3795 /* Update the current entry. */ 3796 stack_entry->end = addr; 3797 stack_entry->avail_ssize -= grow_amount; 3798 vm_map_entry_resize_free(map, stack_entry); 3799 rv = KERN_SUCCESS; 3800 3801 if (next_entry != &map->header) 3802 vm_map_clip_start(map, next_entry, addr); 3803 } else 3804 rv = KERN_FAILURE; 3805 } 3806 3807 if (rv == KERN_SUCCESS && is_procstack) 3808 vm->vm_ssize += btoc(grow_amount); 3809 3810 vm_map_unlock(map); 3811 3812 /* 3813 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3814 */ 3815 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3816 vm_map_wire(map, 3817 (stack_entry == next_entry) ? addr : addr - grow_amount, 3818 (stack_entry == next_entry) ? stack_entry->start : addr, 3819 (p->p_flag & P_SYSTEM) 3820 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3821 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3822 } 3823 3824out: 3825#ifdef RACCT 3826 if (racct_enable && rv != KERN_SUCCESS) { 3827 PROC_LOCK(p); 3828 error = racct_set(p, RACCT_VMEM, map->size); 3829 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3830 if (!old_mlock) { 3831 error = racct_set(p, RACCT_MEMLOCK, 3832 ptoa(pmap_wired_count(map->pmap))); 3833 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3834 } 3835 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3836 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3837 PROC_UNLOCK(p); 3838 } 3839#endif 3840 3841 return (rv); 3842} 3843 3844/* 3845 * Unshare the specified VM space for exec. If other processes are 3846 * mapped to it, then create a new one. The new vmspace is null. 3847 */ 3848int 3849vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3850{ 3851 struct vmspace *oldvmspace = p->p_vmspace; 3852 struct vmspace *newvmspace; 3853 3854 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3855 ("vmspace_exec recursed")); 3856 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3857 if (newvmspace == NULL) 3858 return (ENOMEM); 3859 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3860 /* 3861 * This code is written like this for prototype purposes. The 3862 * goal is to avoid running down the vmspace here, but let the 3863 * other process's that are still using the vmspace to finally 3864 * run it down. Even though there is little or no chance of blocking 3865 * here, it is a good idea to keep this form for future mods. 3866 */ 3867 PROC_VMSPACE_LOCK(p); 3868 p->p_vmspace = newvmspace; 3869 PROC_VMSPACE_UNLOCK(p); 3870 if (p == curthread->td_proc) 3871 pmap_activate(curthread); 3872 curthread->td_pflags |= TDP_EXECVMSPC; 3873 return (0); 3874} 3875 3876/* 3877 * Unshare the specified VM space for forcing COW. This 3878 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3879 */ 3880int 3881vmspace_unshare(struct proc *p) 3882{ 3883 struct vmspace *oldvmspace = p->p_vmspace; 3884 struct vmspace *newvmspace; 3885 vm_ooffset_t fork_charge; 3886 3887 if (oldvmspace->vm_refcnt == 1) 3888 return (0); 3889 fork_charge = 0; 3890 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3891 if (newvmspace == NULL) 3892 return (ENOMEM); 3893 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3894 vmspace_free(newvmspace); 3895 return (ENOMEM); 3896 } 3897 PROC_VMSPACE_LOCK(p); 3898 p->p_vmspace = newvmspace; 3899 PROC_VMSPACE_UNLOCK(p); 3900 if (p == curthread->td_proc) 3901 pmap_activate(curthread); 3902 vmspace_free(oldvmspace); 3903 return (0); 3904} 3905 3906/* 3907 * vm_map_lookup: 3908 * 3909 * Finds the VM object, offset, and 3910 * protection for a given virtual address in the 3911 * specified map, assuming a page fault of the 3912 * type specified. 3913 * 3914 * Leaves the map in question locked for read; return 3915 * values are guaranteed until a vm_map_lookup_done 3916 * call is performed. Note that the map argument 3917 * is in/out; the returned map must be used in 3918 * the call to vm_map_lookup_done. 3919 * 3920 * A handle (out_entry) is returned for use in 3921 * vm_map_lookup_done, to make that fast. 3922 * 3923 * If a lookup is requested with "write protection" 3924 * specified, the map may be changed to perform virtual 3925 * copying operations, although the data referenced will 3926 * remain the same. 3927 */ 3928int 3929vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3930 vm_offset_t vaddr, 3931 vm_prot_t fault_typea, 3932 vm_map_entry_t *out_entry, /* OUT */ 3933 vm_object_t *object, /* OUT */ 3934 vm_pindex_t *pindex, /* OUT */ 3935 vm_prot_t *out_prot, /* OUT */ 3936 boolean_t *wired) /* OUT */ 3937{ 3938 vm_map_entry_t entry; 3939 vm_map_t map = *var_map; 3940 vm_prot_t prot; 3941 vm_prot_t fault_type = fault_typea; 3942 vm_object_t eobject; 3943 vm_size_t size; 3944 struct ucred *cred; 3945 3946RetryLookup:; 3947 3948 vm_map_lock_read(map); 3949 3950 /* 3951 * Lookup the faulting address. 3952 */ 3953 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3954 vm_map_unlock_read(map); 3955 return (KERN_INVALID_ADDRESS); 3956 } 3957 3958 entry = *out_entry; 3959 3960 /* 3961 * Handle submaps. 3962 */ 3963 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3964 vm_map_t old_map = map; 3965 3966 *var_map = map = entry->object.sub_map; 3967 vm_map_unlock_read(old_map); 3968 goto RetryLookup; 3969 } 3970 3971 /* 3972 * Check whether this task is allowed to have this page. 3973 */ 3974 prot = entry->protection; 3975 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3976 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3977 vm_map_unlock_read(map); 3978 return (KERN_PROTECTION_FAILURE); 3979 } 3980 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags & 3981 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) != 3982 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY), 3983 ("entry %p flags %x", entry, entry->eflags)); 3984 if ((fault_typea & VM_PROT_COPY) != 0 && 3985 (entry->max_protection & VM_PROT_WRITE) == 0 && 3986 (entry->eflags & MAP_ENTRY_COW) == 0) { 3987 vm_map_unlock_read(map); 3988 return (KERN_PROTECTION_FAILURE); 3989 } 3990 3991 /* 3992 * If this page is not pageable, we have to get it for all possible 3993 * accesses. 3994 */ 3995 *wired = (entry->wired_count != 0); 3996 if (*wired) 3997 fault_type = entry->protection; 3998 size = entry->end - entry->start; 3999 /* 4000 * If the entry was copy-on-write, we either ... 4001 */ 4002 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4003 /* 4004 * If we want to write the page, we may as well handle that 4005 * now since we've got the map locked. 4006 * 4007 * If we don't need to write the page, we just demote the 4008 * permissions allowed. 4009 */ 4010 if ((fault_type & VM_PROT_WRITE) != 0 || 4011 (fault_typea & VM_PROT_COPY) != 0) { 4012 /* 4013 * Make a new object, and place it in the object 4014 * chain. Note that no new references have appeared 4015 * -- one just moved from the map to the new 4016 * object. 4017 */ 4018 if (vm_map_lock_upgrade(map)) 4019 goto RetryLookup; 4020 4021 if (entry->cred == NULL) { 4022 /* 4023 * The debugger owner is charged for 4024 * the memory. 4025 */ 4026 cred = curthread->td_ucred; 4027 crhold(cred); 4028 if (!swap_reserve_by_cred(size, cred)) { 4029 crfree(cred); 4030 vm_map_unlock(map); 4031 return (KERN_RESOURCE_SHORTAGE); 4032 } 4033 entry->cred = cred; 4034 } 4035 vm_object_shadow(&entry->object.vm_object, 4036 &entry->offset, size); 4037 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4038 eobject = entry->object.vm_object; 4039 if (eobject->cred != NULL) { 4040 /* 4041 * The object was not shadowed. 4042 */ 4043 swap_release_by_cred(size, entry->cred); 4044 crfree(entry->cred); 4045 entry->cred = NULL; 4046 } else if (entry->cred != NULL) { 4047 VM_OBJECT_WLOCK(eobject); 4048 eobject->cred = entry->cred; 4049 eobject->charge = size; 4050 VM_OBJECT_WUNLOCK(eobject); 4051 entry->cred = NULL; 4052 } 4053 4054 vm_map_lock_downgrade(map); 4055 } else { 4056 /* 4057 * We're attempting to read a copy-on-write page -- 4058 * don't allow writes. 4059 */ 4060 prot &= ~VM_PROT_WRITE; 4061 } 4062 } 4063 4064 /* 4065 * Create an object if necessary. 4066 */ 4067 if (entry->object.vm_object == NULL && 4068 !map->system_map) { 4069 if (vm_map_lock_upgrade(map)) 4070 goto RetryLookup; 4071 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4072 atop(size)); 4073 entry->offset = 0; 4074 if (entry->cred != NULL) { 4075 VM_OBJECT_WLOCK(entry->object.vm_object); 4076 entry->object.vm_object->cred = entry->cred; 4077 entry->object.vm_object->charge = size; 4078 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4079 entry->cred = NULL; 4080 } 4081 vm_map_lock_downgrade(map); 4082 } 4083 4084 /* 4085 * Return the object/offset from this entry. If the entry was 4086 * copy-on-write or empty, it has been fixed up. 4087 */ 4088 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4089 *object = entry->object.vm_object; 4090 4091 *out_prot = prot; 4092 return (KERN_SUCCESS); 4093} 4094 4095/* 4096 * vm_map_lookup_locked: 4097 * 4098 * Lookup the faulting address. A version of vm_map_lookup that returns 4099 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4100 */ 4101int 4102vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4103 vm_offset_t vaddr, 4104 vm_prot_t fault_typea, 4105 vm_map_entry_t *out_entry, /* OUT */ 4106 vm_object_t *object, /* OUT */ 4107 vm_pindex_t *pindex, /* OUT */ 4108 vm_prot_t *out_prot, /* OUT */ 4109 boolean_t *wired) /* OUT */ 4110{ 4111 vm_map_entry_t entry; 4112 vm_map_t map = *var_map; 4113 vm_prot_t prot; 4114 vm_prot_t fault_type = fault_typea; 4115 4116 /* 4117 * Lookup the faulting address. 4118 */ 4119 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4120 return (KERN_INVALID_ADDRESS); 4121 4122 entry = *out_entry; 4123 4124 /* 4125 * Fail if the entry refers to a submap. 4126 */ 4127 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4128 return (KERN_FAILURE); 4129 4130 /* 4131 * Check whether this task is allowed to have this page. 4132 */ 4133 prot = entry->protection; 4134 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4135 if ((fault_type & prot) != fault_type) 4136 return (KERN_PROTECTION_FAILURE); 4137 4138 /* 4139 * If this page is not pageable, we have to get it for all possible 4140 * accesses. 4141 */ 4142 *wired = (entry->wired_count != 0); 4143 if (*wired) 4144 fault_type = entry->protection; 4145 4146 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4147 /* 4148 * Fail if the entry was copy-on-write for a write fault. 4149 */ 4150 if (fault_type & VM_PROT_WRITE) 4151 return (KERN_FAILURE); 4152 /* 4153 * We're attempting to read a copy-on-write page -- 4154 * don't allow writes. 4155 */ 4156 prot &= ~VM_PROT_WRITE; 4157 } 4158 4159 /* 4160 * Fail if an object should be created. 4161 */ 4162 if (entry->object.vm_object == NULL && !map->system_map) 4163 return (KERN_FAILURE); 4164 4165 /* 4166 * Return the object/offset from this entry. If the entry was 4167 * copy-on-write or empty, it has been fixed up. 4168 */ 4169 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4170 *object = entry->object.vm_object; 4171 4172 *out_prot = prot; 4173 return (KERN_SUCCESS); 4174} 4175 4176/* 4177 * vm_map_lookup_done: 4178 * 4179 * Releases locks acquired by a vm_map_lookup 4180 * (according to the handle returned by that lookup). 4181 */ 4182void 4183vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4184{ 4185 /* 4186 * Unlock the main-level map 4187 */ 4188 vm_map_unlock_read(map); 4189} 4190 4191#include "opt_ddb.h" 4192#ifdef DDB 4193#include <sys/kernel.h> 4194 4195#include <ddb/ddb.h> 4196 4197static void 4198vm_map_print(vm_map_t map) 4199{ 4200 vm_map_entry_t entry; 4201 4202 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4203 (void *)map, 4204 (void *)map->pmap, map->nentries, map->timestamp); 4205 4206 db_indent += 2; 4207 for (entry = map->header.next; entry != &map->header; 4208 entry = entry->next) { 4209 db_iprintf("map entry %p: start=%p, end=%p\n", 4210 (void *)entry, (void *)entry->start, (void *)entry->end); 4211 { 4212 static char *inheritance_name[4] = 4213 {"share", "copy", "none", "donate_copy"}; 4214 4215 db_iprintf(" prot=%x/%x/%s", 4216 entry->protection, 4217 entry->max_protection, 4218 inheritance_name[(int)(unsigned char)entry->inheritance]); 4219 if (entry->wired_count != 0) 4220 db_printf(", wired"); 4221 } 4222 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4223 db_printf(", share=%p, offset=0x%jx\n", 4224 (void *)entry->object.sub_map, 4225 (uintmax_t)entry->offset); 4226 if ((entry->prev == &map->header) || 4227 (entry->prev->object.sub_map != 4228 entry->object.sub_map)) { 4229 db_indent += 2; 4230 vm_map_print((vm_map_t)entry->object.sub_map); 4231 db_indent -= 2; 4232 } 4233 } else { 4234 if (entry->cred != NULL) 4235 db_printf(", ruid %d", entry->cred->cr_ruid); 4236 db_printf(", object=%p, offset=0x%jx", 4237 (void *)entry->object.vm_object, 4238 (uintmax_t)entry->offset); 4239 if (entry->object.vm_object && entry->object.vm_object->cred) 4240 db_printf(", obj ruid %d charge %jx", 4241 entry->object.vm_object->cred->cr_ruid, 4242 (uintmax_t)entry->object.vm_object->charge); 4243 if (entry->eflags & MAP_ENTRY_COW) 4244 db_printf(", copy (%s)", 4245 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4246 db_printf("\n"); 4247 4248 if ((entry->prev == &map->header) || 4249 (entry->prev->object.vm_object != 4250 entry->object.vm_object)) { 4251 db_indent += 2; 4252 vm_object_print((db_expr_t)(intptr_t) 4253 entry->object.vm_object, 4254 0, 0, (char *)0); 4255 db_indent -= 2; 4256 } 4257 } 4258 } 4259 db_indent -= 2; 4260} 4261 4262DB_SHOW_COMMAND(map, map) 4263{ 4264 4265 if (!have_addr) { 4266 db_printf("usage: show map <addr>\n"); 4267 return; 4268 } 4269 vm_map_print((vm_map_t)addr); 4270} 4271 4272DB_SHOW_COMMAND(procvm, procvm) 4273{ 4274 struct proc *p; 4275 4276 if (have_addr) { 4277 p = db_lookup_proc(addr); 4278 } else { 4279 p = curproc; 4280 } 4281 4282 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4283 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4284 (void *)vmspace_pmap(p->p_vmspace)); 4285 4286 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4287} 4288 4289#endif /* DDB */ 4290