vm_map.c revision 321359
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 321359 2017-07-22 04:57:51Z alc $"); 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 KASSERT((prot & ~max) == 0, 1147 ("prot %#x is not subset of max_prot %#x", prot, max)); 1148 1149 /* 1150 * Check that the start and end points are not bogus. 1151 */ 1152 if (start < map->min_offset || end > map->max_offset || start >= end) 1153 return (KERN_INVALID_ADDRESS); 1154 1155 /* 1156 * Find the entry prior to the proposed starting address; if it's part 1157 * of an existing entry, this range is bogus. 1158 */ 1159 if (vm_map_lookup_entry(map, start, &temp_entry)) 1160 return (KERN_NO_SPACE); 1161 1162 prev_entry = temp_entry; 1163 1164 /* 1165 * Assert that the next entry doesn't overlap the end point. 1166 */ 1167 if (prev_entry->next != &map->header && prev_entry->next->start < end) 1168 return (KERN_NO_SPACE); 1169 1170 protoeflags = 0; 1171 if (cow & MAP_COPY_ON_WRITE) 1172 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; 1173 if (cow & MAP_NOFAULT) 1174 protoeflags |= MAP_ENTRY_NOFAULT; 1175 if (cow & MAP_DISABLE_SYNCER) 1176 protoeflags |= MAP_ENTRY_NOSYNC; 1177 if (cow & MAP_DISABLE_COREDUMP) 1178 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1179 if (cow & MAP_STACK_GROWS_DOWN) 1180 protoeflags |= MAP_ENTRY_GROWS_DOWN; 1181 if (cow & MAP_STACK_GROWS_UP) 1182 protoeflags |= MAP_ENTRY_GROWS_UP; 1183 if (cow & MAP_VN_WRITECOUNT) 1184 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1185 if (cow & MAP_INHERIT_SHARE) 1186 inheritance = VM_INHERIT_SHARE; 1187 else 1188 inheritance = VM_INHERIT_DEFAULT; 1189 1190 cred = NULL; 1191 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1192 goto charged; 1193 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1194 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1195 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1196 return (KERN_RESOURCE_SHORTAGE); 1197 KASSERT(object == NULL || 1198 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 || 1199 object->cred == NULL, 1200 ("overcommit: vm_map_insert o %p", object)); 1201 cred = curthread->td_ucred; 1202 } 1203 1204charged: 1205 /* Expand the kernel pmap, if necessary. */ 1206 if (map == kernel_map && end > kernel_vm_end) 1207 pmap_growkernel(end); 1208 if (object != NULL) { 1209 /* 1210 * OBJ_ONEMAPPING must be cleared unless this mapping 1211 * is trivially proven to be the only mapping for any 1212 * of the object's pages. (Object granularity 1213 * reference counting is insufficient to recognize 1214 * aliases with precision.) 1215 */ 1216 VM_OBJECT_WLOCK(object); 1217 if (object->ref_count > 1 || object->shadow_count != 0) 1218 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1219 VM_OBJECT_WUNLOCK(object); 1220 } else if (prev_entry != &map->header && 1221 prev_entry->eflags == protoeflags && 1222 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 && 1223 prev_entry->end == start && prev_entry->wired_count == 0 && 1224 (prev_entry->cred == cred || 1225 (prev_entry->object.vm_object != NULL && 1226 prev_entry->object.vm_object->cred == cred)) && 1227 vm_object_coalesce(prev_entry->object.vm_object, 1228 prev_entry->offset, 1229 (vm_size_t)(prev_entry->end - prev_entry->start), 1230 (vm_size_t)(end - prev_entry->end), cred != NULL && 1231 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) { 1232 /* 1233 * We were able to extend the object. Determine if we 1234 * can extend the previous map entry to include the 1235 * new range as well. 1236 */ 1237 if (prev_entry->inheritance == inheritance && 1238 prev_entry->protection == prot && 1239 prev_entry->max_protection == max) { 1240 map->size += end - prev_entry->end; 1241 prev_entry->end = end; 1242 vm_map_entry_resize_free(map, prev_entry); 1243 vm_map_simplify_entry(map, prev_entry); 1244 return (KERN_SUCCESS); 1245 } 1246 1247 /* 1248 * If we can extend the object but cannot extend the 1249 * map entry, we have to create a new map entry. We 1250 * must bump the ref count on the extended object to 1251 * account for it. object may be NULL. 1252 */ 1253 object = prev_entry->object.vm_object; 1254 offset = prev_entry->offset + 1255 (prev_entry->end - prev_entry->start); 1256 vm_object_reference(object); 1257 if (cred != NULL && object != NULL && object->cred != NULL && 1258 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1259 /* Object already accounts for this uid. */ 1260 cred = NULL; 1261 } 1262 } 1263 if (cred != NULL) 1264 crhold(cred); 1265 1266 /* 1267 * Create a new entry 1268 */ 1269 new_entry = vm_map_entry_create(map); 1270 new_entry->start = start; 1271 new_entry->end = end; 1272 new_entry->cred = NULL; 1273 1274 new_entry->eflags = protoeflags; 1275 new_entry->object.vm_object = object; 1276 new_entry->offset = offset; 1277 new_entry->avail_ssize = 0; 1278 1279 new_entry->inheritance = inheritance; 1280 new_entry->protection = prot; 1281 new_entry->max_protection = max; 1282 new_entry->wired_count = 0; 1283 new_entry->wiring_thread = NULL; 1284 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1285 new_entry->next_read = OFF_TO_IDX(offset); 1286 1287 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1288 ("overcommit: vm_map_insert leaks vm_map %p", new_entry)); 1289 new_entry->cred = cred; 1290 1291 /* 1292 * Insert the new entry into the list 1293 */ 1294 vm_map_entry_link(map, prev_entry, new_entry); 1295 map->size += new_entry->end - new_entry->start; 1296 1297 /* 1298 * Try to coalesce the new entry with both the previous and next 1299 * entries in the list. Previously, we only attempted to coalesce 1300 * with the previous entry when object is NULL. Here, we handle the 1301 * other cases, which are less common. 1302 */ 1303 vm_map_simplify_entry(map, new_entry); 1304 1305 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) { 1306 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset), 1307 end - start, cow & MAP_PREFAULT_PARTIAL); 1308 } 1309 1310 return (KERN_SUCCESS); 1311} 1312 1313/* 1314 * vm_map_findspace: 1315 * 1316 * Find the first fit (lowest VM address) for "length" free bytes 1317 * beginning at address >= start in the given map. 1318 * 1319 * In a vm_map_entry, "adj_free" is the amount of free space 1320 * adjacent (higher address) to this entry, and "max_free" is the 1321 * maximum amount of contiguous free space in its subtree. This 1322 * allows finding a free region in one path down the tree, so 1323 * O(log n) amortized with splay trees. 1324 * 1325 * The map must be locked, and leaves it so. 1326 * 1327 * Returns: 0 on success, and starting address in *addr, 1328 * 1 if insufficient space. 1329 */ 1330int 1331vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1332 vm_offset_t *addr) /* OUT */ 1333{ 1334 vm_map_entry_t entry; 1335 vm_offset_t st; 1336 1337 /* 1338 * Request must fit within min/max VM address and must avoid 1339 * address wrap. 1340 */ 1341 if (start < map->min_offset) 1342 start = map->min_offset; 1343 if (start + length > map->max_offset || start + length < start) 1344 return (1); 1345 1346 /* Empty tree means wide open address space. */ 1347 if (map->root == NULL) { 1348 *addr = start; 1349 return (0); 1350 } 1351 1352 /* 1353 * After splay, if start comes before root node, then there 1354 * must be a gap from start to the root. 1355 */ 1356 map->root = vm_map_entry_splay(start, map->root); 1357 if (start + length <= map->root->start) { 1358 *addr = start; 1359 return (0); 1360 } 1361 1362 /* 1363 * Root is the last node that might begin its gap before 1364 * start, and this is the last comparison where address 1365 * wrap might be a problem. 1366 */ 1367 st = (start > map->root->end) ? start : map->root->end; 1368 if (length <= map->root->end + map->root->adj_free - st) { 1369 *addr = st; 1370 return (0); 1371 } 1372 1373 /* With max_free, can immediately tell if no solution. */ 1374 entry = map->root->right; 1375 if (entry == NULL || length > entry->max_free) 1376 return (1); 1377 1378 /* 1379 * Search the right subtree in the order: left subtree, root, 1380 * right subtree (first fit). The previous splay implies that 1381 * all regions in the right subtree have addresses > start. 1382 */ 1383 while (entry != NULL) { 1384 if (entry->left != NULL && entry->left->max_free >= length) 1385 entry = entry->left; 1386 else if (entry->adj_free >= length) { 1387 *addr = entry->end; 1388 return (0); 1389 } else 1390 entry = entry->right; 1391 } 1392 1393 /* Can't get here, so panic if we do. */ 1394 panic("vm_map_findspace: max_free corrupt"); 1395} 1396 1397int 1398vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1399 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1400 vm_prot_t max, int cow) 1401{ 1402 vm_offset_t end; 1403 int result; 1404 1405 end = start + length; 1406 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1407 object == NULL, 1408 ("vm_map_fixed: non-NULL backing object for stack")); 1409 vm_map_lock(map); 1410 VM_MAP_RANGE_CHECK(map, start, end); 1411 if ((cow & MAP_CHECK_EXCL) == 0) 1412 vm_map_delete(map, start, end); 1413 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1414 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1415 prot, max, cow); 1416 } else { 1417 result = vm_map_insert(map, object, offset, start, end, 1418 prot, max, cow); 1419 } 1420 vm_map_unlock(map); 1421 return (result); 1422} 1423 1424/* 1425 * vm_map_find finds an unallocated region in the target address 1426 * map with the given length. The search is defined to be 1427 * first-fit from the specified address; the region found is 1428 * returned in the same parameter. 1429 * 1430 * If object is non-NULL, ref count must be bumped by caller 1431 * prior to making call to account for the new entry. 1432 */ 1433int 1434vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1435 vm_offset_t *addr, /* IN/OUT */ 1436 vm_size_t length, vm_offset_t max_addr, int find_space, 1437 vm_prot_t prot, vm_prot_t max, int cow) 1438{ 1439 vm_offset_t alignment, initial_addr, start; 1440 int result; 1441 1442 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1443 object == NULL, 1444 ("vm_map_find: non-NULL backing object for stack")); 1445 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1446 (object->flags & OBJ_COLORED) == 0)) 1447 find_space = VMFS_ANY_SPACE; 1448 if (find_space >> 8 != 0) { 1449 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1450 alignment = (vm_offset_t)1 << (find_space >> 8); 1451 } else 1452 alignment = 0; 1453 initial_addr = *addr; 1454again: 1455 start = initial_addr; 1456 vm_map_lock(map); 1457 do { 1458 if (find_space != VMFS_NO_SPACE) { 1459 if (vm_map_findspace(map, start, length, addr) || 1460 (max_addr != 0 && *addr + length > max_addr)) { 1461 vm_map_unlock(map); 1462 if (find_space == VMFS_OPTIMAL_SPACE) { 1463 find_space = VMFS_ANY_SPACE; 1464 goto again; 1465 } 1466 return (KERN_NO_SPACE); 1467 } 1468 switch (find_space) { 1469 case VMFS_SUPER_SPACE: 1470 case VMFS_OPTIMAL_SPACE: 1471 pmap_align_superpage(object, offset, addr, 1472 length); 1473 break; 1474 case VMFS_ANY_SPACE: 1475 break; 1476 default: 1477 if ((*addr & (alignment - 1)) != 0) { 1478 *addr &= ~(alignment - 1); 1479 *addr += alignment; 1480 } 1481 break; 1482 } 1483 1484 start = *addr; 1485 } 1486 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1487 result = vm_map_stack_locked(map, start, length, 1488 sgrowsiz, prot, max, cow); 1489 } else { 1490 result = vm_map_insert(map, object, offset, start, 1491 start + length, prot, max, cow); 1492 } 1493 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1494 find_space != VMFS_ANY_SPACE); 1495 vm_map_unlock(map); 1496 return (result); 1497} 1498 1499/* 1500 * vm_map_simplify_entry: 1501 * 1502 * Simplify the given map entry by merging with either neighbor. This 1503 * routine also has the ability to merge with both neighbors. 1504 * 1505 * The map must be locked. 1506 * 1507 * This routine guarentees that the passed entry remains valid (though 1508 * possibly extended). When merging, this routine may delete one or 1509 * both neighbors. 1510 */ 1511void 1512vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1513{ 1514 vm_map_entry_t next, prev; 1515 vm_size_t prevsize, esize; 1516 1517 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | 1518 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0) 1519 return; 1520 1521 prev = entry->prev; 1522 if (prev != &map->header) { 1523 prevsize = prev->end - prev->start; 1524 if ( (prev->end == entry->start) && 1525 (prev->object.vm_object == entry->object.vm_object) && 1526 (!prev->object.vm_object || 1527 (prev->offset + prevsize == entry->offset)) && 1528 (prev->eflags == entry->eflags) && 1529 (prev->protection == entry->protection) && 1530 (prev->max_protection == entry->max_protection) && 1531 (prev->inheritance == entry->inheritance) && 1532 (prev->wired_count == entry->wired_count) && 1533 (prev->cred == entry->cred)) { 1534 vm_map_entry_unlink(map, prev); 1535 entry->start = prev->start; 1536 entry->offset = prev->offset; 1537 if (entry->prev != &map->header) 1538 vm_map_entry_resize_free(map, entry->prev); 1539 1540 /* 1541 * If the backing object is a vnode object, 1542 * vm_object_deallocate() calls vrele(). 1543 * However, vrele() does not lock the vnode 1544 * because the vnode has additional 1545 * references. Thus, the map lock can be kept 1546 * without causing a lock-order reversal with 1547 * the vnode lock. 1548 * 1549 * Since we count the number of virtual page 1550 * mappings in object->un_pager.vnp.writemappings, 1551 * the writemappings value should not be adjusted 1552 * when the entry is disposed of. 1553 */ 1554 if (prev->object.vm_object) 1555 vm_object_deallocate(prev->object.vm_object); 1556 if (prev->cred != NULL) 1557 crfree(prev->cred); 1558 vm_map_entry_dispose(map, prev); 1559 } 1560 } 1561 1562 next = entry->next; 1563 if (next != &map->header) { 1564 esize = entry->end - entry->start; 1565 if ((entry->end == next->start) && 1566 (next->object.vm_object == entry->object.vm_object) && 1567 (!entry->object.vm_object || 1568 (entry->offset + esize == next->offset)) && 1569 (next->eflags == entry->eflags) && 1570 (next->protection == entry->protection) && 1571 (next->max_protection == entry->max_protection) && 1572 (next->inheritance == entry->inheritance) && 1573 (next->wired_count == entry->wired_count) && 1574 (next->cred == entry->cred)) { 1575 vm_map_entry_unlink(map, next); 1576 entry->end = next->end; 1577 vm_map_entry_resize_free(map, entry); 1578 1579 /* 1580 * See comment above. 1581 */ 1582 if (next->object.vm_object) 1583 vm_object_deallocate(next->object.vm_object); 1584 if (next->cred != NULL) 1585 crfree(next->cred); 1586 vm_map_entry_dispose(map, next); 1587 } 1588 } 1589} 1590/* 1591 * vm_map_clip_start: [ internal use only ] 1592 * 1593 * Asserts that the given entry begins at or after 1594 * the specified address; if necessary, 1595 * it splits the entry into two. 1596 */ 1597#define vm_map_clip_start(map, entry, startaddr) \ 1598{ \ 1599 if (startaddr > entry->start) \ 1600 _vm_map_clip_start(map, entry, startaddr); \ 1601} 1602 1603/* 1604 * This routine is called only when it is known that 1605 * the entry must be split. 1606 */ 1607static void 1608_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1609{ 1610 vm_map_entry_t new_entry; 1611 1612 VM_MAP_ASSERT_LOCKED(map); 1613 1614 /* 1615 * Split off the front portion -- note that we must insert the new 1616 * entry BEFORE this one, so that this entry has the specified 1617 * starting address. 1618 */ 1619 vm_map_simplify_entry(map, entry); 1620 1621 /* 1622 * If there is no object backing this entry, we might as well create 1623 * one now. If we defer it, an object can get created after the map 1624 * is clipped, and individual objects will be created for the split-up 1625 * map. This is a bit of a hack, but is also about the best place to 1626 * put this improvement. 1627 */ 1628 if (entry->object.vm_object == NULL && !map->system_map) { 1629 vm_object_t object; 1630 object = vm_object_allocate(OBJT_DEFAULT, 1631 atop(entry->end - entry->start)); 1632 entry->object.vm_object = object; 1633 entry->offset = 0; 1634 if (entry->cred != NULL) { 1635 object->cred = entry->cred; 1636 object->charge = entry->end - entry->start; 1637 entry->cred = NULL; 1638 } 1639 } else if (entry->object.vm_object != NULL && 1640 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1641 entry->cred != NULL) { 1642 VM_OBJECT_WLOCK(entry->object.vm_object); 1643 KASSERT(entry->object.vm_object->cred == NULL, 1644 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1645 entry->object.vm_object->cred = entry->cred; 1646 entry->object.vm_object->charge = entry->end - entry->start; 1647 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1648 entry->cred = NULL; 1649 } 1650 1651 new_entry = vm_map_entry_create(map); 1652 *new_entry = *entry; 1653 1654 new_entry->end = start; 1655 entry->offset += (start - entry->start); 1656 entry->start = start; 1657 if (new_entry->cred != NULL) 1658 crhold(entry->cred); 1659 1660 vm_map_entry_link(map, entry->prev, new_entry); 1661 1662 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1663 vm_object_reference(new_entry->object.vm_object); 1664 /* 1665 * The object->un_pager.vnp.writemappings for the 1666 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1667 * kept as is here. The virtual pages are 1668 * re-distributed among the clipped entries, so the sum is 1669 * left the same. 1670 */ 1671 } 1672} 1673 1674/* 1675 * vm_map_clip_end: [ internal use only ] 1676 * 1677 * Asserts that the given entry ends at or before 1678 * the specified address; if necessary, 1679 * it splits the entry into two. 1680 */ 1681#define vm_map_clip_end(map, entry, endaddr) \ 1682{ \ 1683 if ((endaddr) < (entry->end)) \ 1684 _vm_map_clip_end((map), (entry), (endaddr)); \ 1685} 1686 1687/* 1688 * This routine is called only when it is known that 1689 * the entry must be split. 1690 */ 1691static void 1692_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1693{ 1694 vm_map_entry_t new_entry; 1695 1696 VM_MAP_ASSERT_LOCKED(map); 1697 1698 /* 1699 * If there is no object backing this entry, we might as well create 1700 * one now. If we defer it, an object can get created after the map 1701 * is clipped, and individual objects will be created for the split-up 1702 * map. This is a bit of a hack, but is also about the best place to 1703 * put this improvement. 1704 */ 1705 if (entry->object.vm_object == NULL && !map->system_map) { 1706 vm_object_t object; 1707 object = vm_object_allocate(OBJT_DEFAULT, 1708 atop(entry->end - entry->start)); 1709 entry->object.vm_object = object; 1710 entry->offset = 0; 1711 if (entry->cred != NULL) { 1712 object->cred = entry->cred; 1713 object->charge = entry->end - entry->start; 1714 entry->cred = NULL; 1715 } 1716 } else if (entry->object.vm_object != NULL && 1717 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1718 entry->cred != NULL) { 1719 VM_OBJECT_WLOCK(entry->object.vm_object); 1720 KASSERT(entry->object.vm_object->cred == NULL, 1721 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1722 entry->object.vm_object->cred = entry->cred; 1723 entry->object.vm_object->charge = entry->end - entry->start; 1724 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1725 entry->cred = NULL; 1726 } 1727 1728 /* 1729 * Create a new entry and insert it AFTER the specified entry 1730 */ 1731 new_entry = vm_map_entry_create(map); 1732 *new_entry = *entry; 1733 1734 new_entry->start = entry->end = end; 1735 new_entry->offset += (end - entry->start); 1736 if (new_entry->cred != NULL) 1737 crhold(entry->cred); 1738 1739 vm_map_entry_link(map, entry, new_entry); 1740 1741 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1742 vm_object_reference(new_entry->object.vm_object); 1743 } 1744} 1745 1746/* 1747 * vm_map_submap: [ kernel use only ] 1748 * 1749 * Mark the given range as handled by a subordinate map. 1750 * 1751 * This range must have been created with vm_map_find, 1752 * and no other operations may have been performed on this 1753 * range prior to calling vm_map_submap. 1754 * 1755 * Only a limited number of operations can be performed 1756 * within this rage after calling vm_map_submap: 1757 * vm_fault 1758 * [Don't try vm_map_copy!] 1759 * 1760 * To remove a submapping, one must first remove the 1761 * range from the superior map, and then destroy the 1762 * submap (if desired). [Better yet, don't try it.] 1763 */ 1764int 1765vm_map_submap( 1766 vm_map_t map, 1767 vm_offset_t start, 1768 vm_offset_t end, 1769 vm_map_t submap) 1770{ 1771 vm_map_entry_t entry; 1772 int result = KERN_INVALID_ARGUMENT; 1773 1774 vm_map_lock(map); 1775 1776 VM_MAP_RANGE_CHECK(map, start, end); 1777 1778 if (vm_map_lookup_entry(map, start, &entry)) { 1779 vm_map_clip_start(map, entry, start); 1780 } else 1781 entry = entry->next; 1782 1783 vm_map_clip_end(map, entry, end); 1784 1785 if ((entry->start == start) && (entry->end == end) && 1786 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1787 (entry->object.vm_object == NULL)) { 1788 entry->object.sub_map = submap; 1789 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1790 result = KERN_SUCCESS; 1791 } 1792 vm_map_unlock(map); 1793 1794 return (result); 1795} 1796 1797/* 1798 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified 1799 */ 1800#define MAX_INIT_PT 96 1801 1802/* 1803 * vm_map_pmap_enter: 1804 * 1805 * Preload the specified map's pmap with mappings to the specified 1806 * object's memory-resident pages. No further physical pages are 1807 * allocated, and no further virtual pages are retrieved from secondary 1808 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 1809 * limited number of page mappings are created at the low-end of the 1810 * specified address range. (For this purpose, a superpage mapping 1811 * counts as one page mapping.) Otherwise, all resident pages within 1812 * the specified address range are mapped. Because these mappings are 1813 * being created speculatively, cached pages are not reactivated and 1814 * mapped. 1815 */ 1816void 1817vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1818 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1819{ 1820 vm_offset_t start; 1821 vm_page_t p, p_start; 1822 vm_pindex_t mask, psize, threshold, tmpidx; 1823 1824 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1825 return; 1826 VM_OBJECT_RLOCK(object); 1827 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1828 VM_OBJECT_RUNLOCK(object); 1829 VM_OBJECT_WLOCK(object); 1830 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1831 pmap_object_init_pt(map->pmap, addr, object, pindex, 1832 size); 1833 VM_OBJECT_WUNLOCK(object); 1834 return; 1835 } 1836 VM_OBJECT_LOCK_DOWNGRADE(object); 1837 } 1838 1839 psize = atop(size); 1840 if (psize + pindex > object->size) { 1841 if (object->size < pindex) { 1842 VM_OBJECT_RUNLOCK(object); 1843 return; 1844 } 1845 psize = object->size - pindex; 1846 } 1847 1848 start = 0; 1849 p_start = NULL; 1850 threshold = MAX_INIT_PT; 1851 1852 p = vm_page_find_least(object, pindex); 1853 /* 1854 * Assert: the variable p is either (1) the page with the 1855 * least pindex greater than or equal to the parameter pindex 1856 * or (2) NULL. 1857 */ 1858 for (; 1859 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1860 p = TAILQ_NEXT(p, listq)) { 1861 /* 1862 * don't allow an madvise to blow away our really 1863 * free pages allocating pv entries. 1864 */ 1865 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 1866 cnt.v_free_count < cnt.v_free_reserved) || 1867 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 1868 tmpidx >= threshold)) { 1869 psize = tmpidx; 1870 break; 1871 } 1872 if (p->valid == VM_PAGE_BITS_ALL) { 1873 if (p_start == NULL) { 1874 start = addr + ptoa(tmpidx); 1875 p_start = p; 1876 } 1877 /* Jump ahead if a superpage mapping is possible. */ 1878 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 1879 (pagesizes[p->psind] - 1)) == 0) { 1880 mask = atop(pagesizes[p->psind]) - 1; 1881 if (tmpidx + mask < psize && 1882 vm_page_ps_is_valid(p)) { 1883 p += mask; 1884 threshold += mask; 1885 } 1886 } 1887 } else if (p_start != NULL) { 1888 pmap_enter_object(map->pmap, start, addr + 1889 ptoa(tmpidx), p_start, prot); 1890 p_start = NULL; 1891 } 1892 } 1893 if (p_start != NULL) 1894 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1895 p_start, prot); 1896 VM_OBJECT_RUNLOCK(object); 1897} 1898 1899/* 1900 * vm_map_protect: 1901 * 1902 * Sets the protection of the specified address 1903 * region in the target map. If "set_max" is 1904 * specified, the maximum protection is to be set; 1905 * otherwise, only the current protection is affected. 1906 */ 1907int 1908vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1909 vm_prot_t new_prot, boolean_t set_max) 1910{ 1911 vm_map_entry_t current, entry; 1912 vm_object_t obj; 1913 struct ucred *cred; 1914 vm_prot_t old_prot; 1915 1916 if (start == end) 1917 return (KERN_SUCCESS); 1918 1919 vm_map_lock(map); 1920 1921 VM_MAP_RANGE_CHECK(map, start, end); 1922 1923 if (vm_map_lookup_entry(map, start, &entry)) { 1924 vm_map_clip_start(map, entry, start); 1925 } else { 1926 entry = entry->next; 1927 } 1928 1929 /* 1930 * Make a first pass to check for protection violations. 1931 */ 1932 current = entry; 1933 while ((current != &map->header) && (current->start < end)) { 1934 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1935 vm_map_unlock(map); 1936 return (KERN_INVALID_ARGUMENT); 1937 } 1938 if ((new_prot & current->max_protection) != new_prot) { 1939 vm_map_unlock(map); 1940 return (KERN_PROTECTION_FAILURE); 1941 } 1942 current = current->next; 1943 } 1944 1945 1946 /* 1947 * Do an accounting pass for private read-only mappings that 1948 * now will do cow due to allowed write (e.g. debugger sets 1949 * breakpoint on text segment) 1950 */ 1951 for (current = entry; (current != &map->header) && 1952 (current->start < end); current = current->next) { 1953 1954 vm_map_clip_end(map, current, end); 1955 1956 if (set_max || 1957 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1958 ENTRY_CHARGED(current)) { 1959 continue; 1960 } 1961 1962 cred = curthread->td_ucred; 1963 obj = current->object.vm_object; 1964 1965 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1966 if (!swap_reserve(current->end - current->start)) { 1967 vm_map_unlock(map); 1968 return (KERN_RESOURCE_SHORTAGE); 1969 } 1970 crhold(cred); 1971 current->cred = cred; 1972 continue; 1973 } 1974 1975 VM_OBJECT_WLOCK(obj); 1976 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1977 VM_OBJECT_WUNLOCK(obj); 1978 continue; 1979 } 1980 1981 /* 1982 * Charge for the whole object allocation now, since 1983 * we cannot distinguish between non-charged and 1984 * charged clipped mapping of the same object later. 1985 */ 1986 KASSERT(obj->charge == 0, 1987 ("vm_map_protect: object %p overcharged (entry %p)", 1988 obj, current)); 1989 if (!swap_reserve(ptoa(obj->size))) { 1990 VM_OBJECT_WUNLOCK(obj); 1991 vm_map_unlock(map); 1992 return (KERN_RESOURCE_SHORTAGE); 1993 } 1994 1995 crhold(cred); 1996 obj->cred = cred; 1997 obj->charge = ptoa(obj->size); 1998 VM_OBJECT_WUNLOCK(obj); 1999 } 2000 2001 /* 2002 * Go back and fix up protections. [Note that clipping is not 2003 * necessary the second time.] 2004 */ 2005 current = entry; 2006 while ((current != &map->header) && (current->start < end)) { 2007 old_prot = current->protection; 2008 2009 if (set_max) 2010 current->protection = 2011 (current->max_protection = new_prot) & 2012 old_prot; 2013 else 2014 current->protection = new_prot; 2015 2016 /* 2017 * For user wired map entries, the normal lazy evaluation of 2018 * write access upgrades through soft page faults is 2019 * undesirable. Instead, immediately copy any pages that are 2020 * copy-on-write and enable write access in the physical map. 2021 */ 2022 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2023 (current->protection & VM_PROT_WRITE) != 0 && 2024 (old_prot & VM_PROT_WRITE) == 0) 2025 vm_fault_copy_entry(map, map, current, current, NULL); 2026 2027 /* 2028 * When restricting access, update the physical map. Worry 2029 * about copy-on-write here. 2030 */ 2031 if ((old_prot & ~current->protection) != 0) { 2032#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2033 VM_PROT_ALL) 2034 pmap_protect(map->pmap, current->start, 2035 current->end, 2036 current->protection & MASK(current)); 2037#undef MASK 2038 } 2039 vm_map_simplify_entry(map, current); 2040 current = current->next; 2041 } 2042 vm_map_unlock(map); 2043 return (KERN_SUCCESS); 2044} 2045 2046/* 2047 * vm_map_madvise: 2048 * 2049 * This routine traverses a processes map handling the madvise 2050 * system call. Advisories are classified as either those effecting 2051 * the vm_map_entry structure, or those effecting the underlying 2052 * objects. 2053 */ 2054int 2055vm_map_madvise( 2056 vm_map_t map, 2057 vm_offset_t start, 2058 vm_offset_t end, 2059 int behav) 2060{ 2061 vm_map_entry_t current, entry; 2062 int modify_map = 0; 2063 2064 /* 2065 * Some madvise calls directly modify the vm_map_entry, in which case 2066 * we need to use an exclusive lock on the map and we need to perform 2067 * various clipping operations. Otherwise we only need a read-lock 2068 * on the map. 2069 */ 2070 switch(behav) { 2071 case MADV_NORMAL: 2072 case MADV_SEQUENTIAL: 2073 case MADV_RANDOM: 2074 case MADV_NOSYNC: 2075 case MADV_AUTOSYNC: 2076 case MADV_NOCORE: 2077 case MADV_CORE: 2078 if (start == end) 2079 return (KERN_SUCCESS); 2080 modify_map = 1; 2081 vm_map_lock(map); 2082 break; 2083 case MADV_WILLNEED: 2084 case MADV_DONTNEED: 2085 case MADV_FREE: 2086 if (start == end) 2087 return (KERN_SUCCESS); 2088 vm_map_lock_read(map); 2089 break; 2090 default: 2091 return (KERN_INVALID_ARGUMENT); 2092 } 2093 2094 /* 2095 * Locate starting entry and clip if necessary. 2096 */ 2097 VM_MAP_RANGE_CHECK(map, start, end); 2098 2099 if (vm_map_lookup_entry(map, start, &entry)) { 2100 if (modify_map) 2101 vm_map_clip_start(map, entry, start); 2102 } else { 2103 entry = entry->next; 2104 } 2105 2106 if (modify_map) { 2107 /* 2108 * madvise behaviors that are implemented in the vm_map_entry. 2109 * 2110 * We clip the vm_map_entry so that behavioral changes are 2111 * limited to the specified address range. 2112 */ 2113 for (current = entry; 2114 (current != &map->header) && (current->start < end); 2115 current = current->next 2116 ) { 2117 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2118 continue; 2119 2120 vm_map_clip_end(map, current, end); 2121 2122 switch (behav) { 2123 case MADV_NORMAL: 2124 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2125 break; 2126 case MADV_SEQUENTIAL: 2127 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2128 break; 2129 case MADV_RANDOM: 2130 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2131 break; 2132 case MADV_NOSYNC: 2133 current->eflags |= MAP_ENTRY_NOSYNC; 2134 break; 2135 case MADV_AUTOSYNC: 2136 current->eflags &= ~MAP_ENTRY_NOSYNC; 2137 break; 2138 case MADV_NOCORE: 2139 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2140 break; 2141 case MADV_CORE: 2142 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2143 break; 2144 default: 2145 break; 2146 } 2147 vm_map_simplify_entry(map, current); 2148 } 2149 vm_map_unlock(map); 2150 } else { 2151 vm_pindex_t pstart, pend; 2152 2153 /* 2154 * madvise behaviors that are implemented in the underlying 2155 * vm_object. 2156 * 2157 * Since we don't clip the vm_map_entry, we have to clip 2158 * the vm_object pindex and count. 2159 */ 2160 for (current = entry; 2161 (current != &map->header) && (current->start < end); 2162 current = current->next 2163 ) { 2164 vm_offset_t useEnd, useStart; 2165 2166 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2167 continue; 2168 2169 pstart = OFF_TO_IDX(current->offset); 2170 pend = pstart + atop(current->end - current->start); 2171 useStart = current->start; 2172 useEnd = current->end; 2173 2174 if (current->start < start) { 2175 pstart += atop(start - current->start); 2176 useStart = start; 2177 } 2178 if (current->end > end) { 2179 pend -= atop(current->end - end); 2180 useEnd = end; 2181 } 2182 2183 if (pstart >= pend) 2184 continue; 2185 2186 /* 2187 * Perform the pmap_advise() before clearing 2188 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2189 * concurrent pmap operation, such as pmap_remove(), 2190 * could clear a reference in the pmap and set 2191 * PGA_REFERENCED on the page before the pmap_advise() 2192 * had completed. Consequently, the page would appear 2193 * referenced based upon an old reference that 2194 * occurred before this pmap_advise() ran. 2195 */ 2196 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2197 pmap_advise(map->pmap, useStart, useEnd, 2198 behav); 2199 2200 vm_object_madvise(current->object.vm_object, pstart, 2201 pend, behav); 2202 2203 /* 2204 * Pre-populate paging structures in the 2205 * WILLNEED case. For wired entries, the 2206 * paging structures are already populated. 2207 */ 2208 if (behav == MADV_WILLNEED && 2209 current->wired_count == 0) { 2210 vm_map_pmap_enter(map, 2211 useStart, 2212 current->protection, 2213 current->object.vm_object, 2214 pstart, 2215 ptoa(pend - pstart), 2216 MAP_PREFAULT_MADVISE 2217 ); 2218 } 2219 } 2220 vm_map_unlock_read(map); 2221 } 2222 return (0); 2223} 2224 2225 2226/* 2227 * vm_map_inherit: 2228 * 2229 * Sets the inheritance of the specified address 2230 * range in the target map. Inheritance 2231 * affects how the map will be shared with 2232 * child maps at the time of vmspace_fork. 2233 */ 2234int 2235vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2236 vm_inherit_t new_inheritance) 2237{ 2238 vm_map_entry_t entry; 2239 vm_map_entry_t temp_entry; 2240 2241 switch (new_inheritance) { 2242 case VM_INHERIT_NONE: 2243 case VM_INHERIT_COPY: 2244 case VM_INHERIT_SHARE: 2245 case VM_INHERIT_ZERO: 2246 break; 2247 default: 2248 return (KERN_INVALID_ARGUMENT); 2249 } 2250 if (start == end) 2251 return (KERN_SUCCESS); 2252 vm_map_lock(map); 2253 VM_MAP_RANGE_CHECK(map, start, end); 2254 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2255 entry = temp_entry; 2256 vm_map_clip_start(map, entry, start); 2257 } else 2258 entry = temp_entry->next; 2259 while ((entry != &map->header) && (entry->start < end)) { 2260 vm_map_clip_end(map, entry, end); 2261 entry->inheritance = new_inheritance; 2262 vm_map_simplify_entry(map, entry); 2263 entry = entry->next; 2264 } 2265 vm_map_unlock(map); 2266 return (KERN_SUCCESS); 2267} 2268 2269/* 2270 * vm_map_unwire: 2271 * 2272 * Implements both kernel and user unwiring. 2273 */ 2274int 2275vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2276 int flags) 2277{ 2278 vm_map_entry_t entry, first_entry, tmp_entry; 2279 vm_offset_t saved_start; 2280 unsigned int last_timestamp; 2281 int rv; 2282 boolean_t need_wakeup, result, user_unwire; 2283 2284 if (start == end) 2285 return (KERN_SUCCESS); 2286 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2287 vm_map_lock(map); 2288 VM_MAP_RANGE_CHECK(map, start, end); 2289 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2290 if (flags & VM_MAP_WIRE_HOLESOK) 2291 first_entry = first_entry->next; 2292 else { 2293 vm_map_unlock(map); 2294 return (KERN_INVALID_ADDRESS); 2295 } 2296 } 2297 last_timestamp = map->timestamp; 2298 entry = first_entry; 2299 while (entry != &map->header && entry->start < end) { 2300 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2301 /* 2302 * We have not yet clipped the entry. 2303 */ 2304 saved_start = (start >= entry->start) ? start : 2305 entry->start; 2306 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2307 if (vm_map_unlock_and_wait(map, 0)) { 2308 /* 2309 * Allow interruption of user unwiring? 2310 */ 2311 } 2312 vm_map_lock(map); 2313 if (last_timestamp+1 != map->timestamp) { 2314 /* 2315 * Look again for the entry because the map was 2316 * modified while it was unlocked. 2317 * Specifically, the entry may have been 2318 * clipped, merged, or deleted. 2319 */ 2320 if (!vm_map_lookup_entry(map, saved_start, 2321 &tmp_entry)) { 2322 if (flags & VM_MAP_WIRE_HOLESOK) 2323 tmp_entry = tmp_entry->next; 2324 else { 2325 if (saved_start == start) { 2326 /* 2327 * First_entry has been deleted. 2328 */ 2329 vm_map_unlock(map); 2330 return (KERN_INVALID_ADDRESS); 2331 } 2332 end = saved_start; 2333 rv = KERN_INVALID_ADDRESS; 2334 goto done; 2335 } 2336 } 2337 if (entry == first_entry) 2338 first_entry = tmp_entry; 2339 else 2340 first_entry = NULL; 2341 entry = tmp_entry; 2342 } 2343 last_timestamp = map->timestamp; 2344 continue; 2345 } 2346 vm_map_clip_start(map, entry, start); 2347 vm_map_clip_end(map, entry, end); 2348 /* 2349 * Mark the entry in case the map lock is released. (See 2350 * above.) 2351 */ 2352 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2353 entry->wiring_thread == NULL, 2354 ("owned map entry %p", entry)); 2355 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2356 entry->wiring_thread = curthread; 2357 /* 2358 * Check the map for holes in the specified region. 2359 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2360 */ 2361 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2362 (entry->end < end && (entry->next == &map->header || 2363 entry->next->start > entry->end))) { 2364 end = entry->end; 2365 rv = KERN_INVALID_ADDRESS; 2366 goto done; 2367 } 2368 /* 2369 * If system unwiring, require that the entry is system wired. 2370 */ 2371 if (!user_unwire && 2372 vm_map_entry_system_wired_count(entry) == 0) { 2373 end = entry->end; 2374 rv = KERN_INVALID_ARGUMENT; 2375 goto done; 2376 } 2377 entry = entry->next; 2378 } 2379 rv = KERN_SUCCESS; 2380done: 2381 need_wakeup = FALSE; 2382 if (first_entry == NULL) { 2383 result = vm_map_lookup_entry(map, start, &first_entry); 2384 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2385 first_entry = first_entry->next; 2386 else 2387 KASSERT(result, ("vm_map_unwire: lookup failed")); 2388 } 2389 for (entry = first_entry; entry != &map->header && entry->start < end; 2390 entry = entry->next) { 2391 /* 2392 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2393 * space in the unwired region could have been mapped 2394 * while the map lock was dropped for draining 2395 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2396 * could be simultaneously wiring this new mapping 2397 * entry. Detect these cases and skip any entries 2398 * marked as in transition by us. 2399 */ 2400 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2401 entry->wiring_thread != curthread) { 2402 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2403 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2404 continue; 2405 } 2406 2407 if (rv == KERN_SUCCESS && (!user_unwire || 2408 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2409 if (user_unwire) 2410 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2411 if (entry->wired_count == 1) 2412 vm_map_entry_unwire(map, entry); 2413 else 2414 entry->wired_count--; 2415 } 2416 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2417 ("vm_map_unwire: in-transition flag missing %p", entry)); 2418 KASSERT(entry->wiring_thread == curthread, 2419 ("vm_map_unwire: alien wire %p", entry)); 2420 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2421 entry->wiring_thread = NULL; 2422 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2423 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2424 need_wakeup = TRUE; 2425 } 2426 vm_map_simplify_entry(map, entry); 2427 } 2428 vm_map_unlock(map); 2429 if (need_wakeup) 2430 vm_map_wakeup(map); 2431 return (rv); 2432} 2433 2434/* 2435 * vm_map_wire_entry_failure: 2436 * 2437 * Handle a wiring failure on the given entry. 2438 * 2439 * The map should be locked. 2440 */ 2441static void 2442vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 2443 vm_offset_t failed_addr) 2444{ 2445 2446 VM_MAP_ASSERT_LOCKED(map); 2447 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && 2448 entry->wired_count == 1, 2449 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); 2450 KASSERT(failed_addr < entry->end, 2451 ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); 2452 2453 /* 2454 * If any pages at the start of this entry were successfully wired, 2455 * then unwire them. 2456 */ 2457 if (failed_addr > entry->start) { 2458 pmap_unwire(map->pmap, entry->start, failed_addr); 2459 vm_object_unwire(entry->object.vm_object, entry->offset, 2460 failed_addr - entry->start, PQ_ACTIVE); 2461 } 2462 2463 /* 2464 * Assign an out-of-range value to represent the failure to wire this 2465 * entry. 2466 */ 2467 entry->wired_count = -1; 2468} 2469 2470/* 2471 * vm_map_wire: 2472 * 2473 * Implements both kernel and user wiring. 2474 */ 2475int 2476vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2477 int flags) 2478{ 2479 vm_map_entry_t entry, first_entry, tmp_entry; 2480 vm_offset_t faddr, saved_end, saved_start; 2481 unsigned int last_timestamp; 2482 int rv; 2483 boolean_t need_wakeup, result, user_wire; 2484 vm_prot_t prot; 2485 2486 if (start == end) 2487 return (KERN_SUCCESS); 2488 prot = 0; 2489 if (flags & VM_MAP_WIRE_WRITE) 2490 prot |= VM_PROT_WRITE; 2491 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2492 vm_map_lock(map); 2493 VM_MAP_RANGE_CHECK(map, start, end); 2494 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2495 if (flags & VM_MAP_WIRE_HOLESOK) 2496 first_entry = first_entry->next; 2497 else { 2498 vm_map_unlock(map); 2499 return (KERN_INVALID_ADDRESS); 2500 } 2501 } 2502 last_timestamp = map->timestamp; 2503 entry = first_entry; 2504 while (entry != &map->header && entry->start < end) { 2505 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2506 /* 2507 * We have not yet clipped the entry. 2508 */ 2509 saved_start = (start >= entry->start) ? start : 2510 entry->start; 2511 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2512 if (vm_map_unlock_and_wait(map, 0)) { 2513 /* 2514 * Allow interruption of user wiring? 2515 */ 2516 } 2517 vm_map_lock(map); 2518 if (last_timestamp + 1 != map->timestamp) { 2519 /* 2520 * Look again for the entry because the map was 2521 * modified while it was unlocked. 2522 * Specifically, the entry may have been 2523 * clipped, merged, or deleted. 2524 */ 2525 if (!vm_map_lookup_entry(map, saved_start, 2526 &tmp_entry)) { 2527 if (flags & VM_MAP_WIRE_HOLESOK) 2528 tmp_entry = tmp_entry->next; 2529 else { 2530 if (saved_start == start) { 2531 /* 2532 * first_entry has been deleted. 2533 */ 2534 vm_map_unlock(map); 2535 return (KERN_INVALID_ADDRESS); 2536 } 2537 end = saved_start; 2538 rv = KERN_INVALID_ADDRESS; 2539 goto done; 2540 } 2541 } 2542 if (entry == first_entry) 2543 first_entry = tmp_entry; 2544 else 2545 first_entry = NULL; 2546 entry = tmp_entry; 2547 } 2548 last_timestamp = map->timestamp; 2549 continue; 2550 } 2551 vm_map_clip_start(map, entry, start); 2552 vm_map_clip_end(map, entry, end); 2553 /* 2554 * Mark the entry in case the map lock is released. (See 2555 * above.) 2556 */ 2557 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2558 entry->wiring_thread == NULL, 2559 ("owned map entry %p", entry)); 2560 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2561 entry->wiring_thread = curthread; 2562 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2563 || (entry->protection & prot) != prot) { 2564 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2565 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2566 end = entry->end; 2567 rv = KERN_INVALID_ADDRESS; 2568 goto done; 2569 } 2570 goto next_entry; 2571 } 2572 if (entry->wired_count == 0) { 2573 entry->wired_count++; 2574 saved_start = entry->start; 2575 saved_end = entry->end; 2576 2577 /* 2578 * Release the map lock, relying on the in-transition 2579 * mark. Mark the map busy for fork. 2580 */ 2581 vm_map_busy(map); 2582 vm_map_unlock(map); 2583 2584 faddr = saved_start; 2585 do { 2586 /* 2587 * Simulate a fault to get the page and enter 2588 * it into the physical map. 2589 */ 2590 if ((rv = vm_fault(map, faddr, VM_PROT_NONE, 2591 VM_FAULT_WIRE)) != KERN_SUCCESS) 2592 break; 2593 } while ((faddr += PAGE_SIZE) < saved_end); 2594 vm_map_lock(map); 2595 vm_map_unbusy(map); 2596 if (last_timestamp + 1 != map->timestamp) { 2597 /* 2598 * Look again for the entry because the map was 2599 * modified while it was unlocked. The entry 2600 * may have been clipped, but NOT merged or 2601 * deleted. 2602 */ 2603 result = vm_map_lookup_entry(map, saved_start, 2604 &tmp_entry); 2605 KASSERT(result, ("vm_map_wire: lookup failed")); 2606 if (entry == first_entry) 2607 first_entry = tmp_entry; 2608 else 2609 first_entry = NULL; 2610 entry = tmp_entry; 2611 while (entry->end < saved_end) { 2612 /* 2613 * In case of failure, handle entries 2614 * that were not fully wired here; 2615 * fully wired entries are handled 2616 * later. 2617 */ 2618 if (rv != KERN_SUCCESS && 2619 faddr < entry->end) 2620 vm_map_wire_entry_failure(map, 2621 entry, faddr); 2622 entry = entry->next; 2623 } 2624 } 2625 last_timestamp = map->timestamp; 2626 if (rv != KERN_SUCCESS) { 2627 vm_map_wire_entry_failure(map, entry, faddr); 2628 end = entry->end; 2629 goto done; 2630 } 2631 } else if (!user_wire || 2632 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2633 entry->wired_count++; 2634 } 2635 /* 2636 * Check the map for holes in the specified region. 2637 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2638 */ 2639 next_entry: 2640 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 && 2641 entry->end < end && (entry->next == &map->header || 2642 entry->next->start > entry->end)) { 2643 end = entry->end; 2644 rv = KERN_INVALID_ADDRESS; 2645 goto done; 2646 } 2647 entry = entry->next; 2648 } 2649 rv = KERN_SUCCESS; 2650done: 2651 need_wakeup = FALSE; 2652 if (first_entry == NULL) { 2653 result = vm_map_lookup_entry(map, start, &first_entry); 2654 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2655 first_entry = first_entry->next; 2656 else 2657 KASSERT(result, ("vm_map_wire: lookup failed")); 2658 } 2659 for (entry = first_entry; entry != &map->header && entry->start < end; 2660 entry = entry->next) { 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 not 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 ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2678 goto next_entry_done; 2679 2680 if (rv == KERN_SUCCESS) { 2681 if (user_wire) 2682 entry->eflags |= MAP_ENTRY_USER_WIRED; 2683 } else if (entry->wired_count == -1) { 2684 /* 2685 * Wiring failed on this entry. Thus, unwiring is 2686 * unnecessary. 2687 */ 2688 entry->wired_count = 0; 2689 } else if (!user_wire || 2690 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2691 /* 2692 * Undo the wiring. Wiring succeeded on this entry 2693 * but failed on a later entry. 2694 */ 2695 if (entry->wired_count == 1) 2696 vm_map_entry_unwire(map, entry); 2697 else 2698 entry->wired_count--; 2699 } 2700 next_entry_done: 2701 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2702 ("vm_map_wire: in-transition flag missing %p", entry)); 2703 KASSERT(entry->wiring_thread == curthread, 2704 ("vm_map_wire: alien wire %p", entry)); 2705 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2706 MAP_ENTRY_WIRE_SKIPPED); 2707 entry->wiring_thread = NULL; 2708 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2709 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2710 need_wakeup = TRUE; 2711 } 2712 vm_map_simplify_entry(map, entry); 2713 } 2714 vm_map_unlock(map); 2715 if (need_wakeup) 2716 vm_map_wakeup(map); 2717 return (rv); 2718} 2719 2720/* 2721 * vm_map_sync 2722 * 2723 * Push any dirty cached pages in the address range to their pager. 2724 * If syncio is TRUE, dirty pages are written synchronously. 2725 * If invalidate is TRUE, any cached pages are freed as well. 2726 * 2727 * If the size of the region from start to end is zero, we are 2728 * supposed to flush all modified pages within the region containing 2729 * start. Unfortunately, a region can be split or coalesced with 2730 * neighboring regions, making it difficult to determine what the 2731 * original region was. Therefore, we approximate this requirement by 2732 * flushing the current region containing start. 2733 * 2734 * Returns an error if any part of the specified range is not mapped. 2735 */ 2736int 2737vm_map_sync( 2738 vm_map_t map, 2739 vm_offset_t start, 2740 vm_offset_t end, 2741 boolean_t syncio, 2742 boolean_t invalidate) 2743{ 2744 vm_map_entry_t current; 2745 vm_map_entry_t entry; 2746 vm_size_t size; 2747 vm_object_t object; 2748 vm_ooffset_t offset; 2749 unsigned int last_timestamp; 2750 boolean_t failed; 2751 2752 vm_map_lock_read(map); 2753 VM_MAP_RANGE_CHECK(map, start, end); 2754 if (!vm_map_lookup_entry(map, start, &entry)) { 2755 vm_map_unlock_read(map); 2756 return (KERN_INVALID_ADDRESS); 2757 } else if (start == end) { 2758 start = entry->start; 2759 end = entry->end; 2760 } 2761 /* 2762 * Make a first pass to check for user-wired memory and holes. 2763 */ 2764 for (current = entry; current != &map->header && current->start < end; 2765 current = current->next) { 2766 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2767 vm_map_unlock_read(map); 2768 return (KERN_INVALID_ARGUMENT); 2769 } 2770 if (end > current->end && 2771 (current->next == &map->header || 2772 current->end != current->next->start)) { 2773 vm_map_unlock_read(map); 2774 return (KERN_INVALID_ADDRESS); 2775 } 2776 } 2777 2778 if (invalidate) 2779 pmap_remove(map->pmap, start, end); 2780 failed = FALSE; 2781 2782 /* 2783 * Make a second pass, cleaning/uncaching pages from the indicated 2784 * objects as we go. 2785 */ 2786 for (current = entry; current != &map->header && current->start < end;) { 2787 offset = current->offset + (start - current->start); 2788 size = (end <= current->end ? end : current->end) - start; 2789 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2790 vm_map_t smap; 2791 vm_map_entry_t tentry; 2792 vm_size_t tsize; 2793 2794 smap = current->object.sub_map; 2795 vm_map_lock_read(smap); 2796 (void) vm_map_lookup_entry(smap, offset, &tentry); 2797 tsize = tentry->end - offset; 2798 if (tsize < size) 2799 size = tsize; 2800 object = tentry->object.vm_object; 2801 offset = tentry->offset + (offset - tentry->start); 2802 vm_map_unlock_read(smap); 2803 } else { 2804 object = current->object.vm_object; 2805 } 2806 vm_object_reference(object); 2807 last_timestamp = map->timestamp; 2808 vm_map_unlock_read(map); 2809 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2810 failed = TRUE; 2811 start += size; 2812 vm_object_deallocate(object); 2813 vm_map_lock_read(map); 2814 if (last_timestamp == map->timestamp || 2815 !vm_map_lookup_entry(map, start, ¤t)) 2816 current = current->next; 2817 } 2818 2819 vm_map_unlock_read(map); 2820 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2821} 2822 2823/* 2824 * vm_map_entry_unwire: [ internal use only ] 2825 * 2826 * Make the region specified by this entry pageable. 2827 * 2828 * The map in question should be locked. 2829 * [This is the reason for this routine's existence.] 2830 */ 2831static void 2832vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2833{ 2834 2835 VM_MAP_ASSERT_LOCKED(map); 2836 KASSERT(entry->wired_count > 0, 2837 ("vm_map_entry_unwire: entry %p isn't wired", entry)); 2838 pmap_unwire(map->pmap, entry->start, entry->end); 2839 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end - 2840 entry->start, PQ_ACTIVE); 2841 entry->wired_count = 0; 2842} 2843 2844static void 2845vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2846{ 2847 2848 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2849 vm_object_deallocate(entry->object.vm_object); 2850 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2851} 2852 2853/* 2854 * vm_map_entry_delete: [ internal use only ] 2855 * 2856 * Deallocate the given entry from the target map. 2857 */ 2858static void 2859vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2860{ 2861 vm_object_t object; 2862 vm_pindex_t offidxstart, offidxend, count, size1; 2863 vm_ooffset_t size; 2864 2865 vm_map_entry_unlink(map, entry); 2866 object = entry->object.vm_object; 2867 size = entry->end - entry->start; 2868 map->size -= size; 2869 2870 if (entry->cred != NULL) { 2871 swap_release_by_cred(size, entry->cred); 2872 crfree(entry->cred); 2873 } 2874 2875 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2876 (object != NULL)) { 2877 KASSERT(entry->cred == NULL || object->cred == NULL || 2878 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2879 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2880 count = OFF_TO_IDX(size); 2881 offidxstart = OFF_TO_IDX(entry->offset); 2882 offidxend = offidxstart + count; 2883 VM_OBJECT_WLOCK(object); 2884 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT | 2885 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2886 object == kernel_object || object == kmem_object)) { 2887 vm_object_collapse(object); 2888 2889 /* 2890 * The option OBJPR_NOTMAPPED can be passed here 2891 * because vm_map_delete() already performed 2892 * pmap_remove() on the only mapping to this range 2893 * of pages. 2894 */ 2895 vm_object_page_remove(object, offidxstart, offidxend, 2896 OBJPR_NOTMAPPED); 2897 if (object->type == OBJT_SWAP) 2898 swap_pager_freespace(object, offidxstart, 2899 count); 2900 if (offidxend >= object->size && 2901 offidxstart < object->size) { 2902 size1 = object->size; 2903 object->size = offidxstart; 2904 if (object->cred != NULL) { 2905 size1 -= object->size; 2906 KASSERT(object->charge >= ptoa(size1), 2907 ("object %p charge < 0", object)); 2908 swap_release_by_cred(ptoa(size1), 2909 object->cred); 2910 object->charge -= ptoa(size1); 2911 } 2912 } 2913 } 2914 VM_OBJECT_WUNLOCK(object); 2915 } else 2916 entry->object.vm_object = NULL; 2917 if (map->system_map) 2918 vm_map_entry_deallocate(entry, TRUE); 2919 else { 2920 entry->next = curthread->td_map_def_user; 2921 curthread->td_map_def_user = entry; 2922 } 2923} 2924 2925/* 2926 * vm_map_delete: [ internal use only ] 2927 * 2928 * Deallocates the given address range from the target 2929 * map. 2930 */ 2931int 2932vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2933{ 2934 vm_map_entry_t entry; 2935 vm_map_entry_t first_entry; 2936 2937 VM_MAP_ASSERT_LOCKED(map); 2938 if (start == end) 2939 return (KERN_SUCCESS); 2940 2941 /* 2942 * Find the start of the region, and clip it 2943 */ 2944 if (!vm_map_lookup_entry(map, start, &first_entry)) 2945 entry = first_entry->next; 2946 else { 2947 entry = first_entry; 2948 vm_map_clip_start(map, entry, start); 2949 } 2950 2951 /* 2952 * Step through all entries in this region 2953 */ 2954 while ((entry != &map->header) && (entry->start < end)) { 2955 vm_map_entry_t next; 2956 2957 /* 2958 * Wait for wiring or unwiring of an entry to complete. 2959 * Also wait for any system wirings to disappear on 2960 * user maps. 2961 */ 2962 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2963 (vm_map_pmap(map) != kernel_pmap && 2964 vm_map_entry_system_wired_count(entry) != 0)) { 2965 unsigned int last_timestamp; 2966 vm_offset_t saved_start; 2967 vm_map_entry_t tmp_entry; 2968 2969 saved_start = entry->start; 2970 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2971 last_timestamp = map->timestamp; 2972 (void) vm_map_unlock_and_wait(map, 0); 2973 vm_map_lock(map); 2974 if (last_timestamp + 1 != map->timestamp) { 2975 /* 2976 * Look again for the entry because the map was 2977 * modified while it was unlocked. 2978 * Specifically, the entry may have been 2979 * clipped, merged, or deleted. 2980 */ 2981 if (!vm_map_lookup_entry(map, saved_start, 2982 &tmp_entry)) 2983 entry = tmp_entry->next; 2984 else { 2985 entry = tmp_entry; 2986 vm_map_clip_start(map, entry, 2987 saved_start); 2988 } 2989 } 2990 continue; 2991 } 2992 vm_map_clip_end(map, entry, end); 2993 2994 next = entry->next; 2995 2996 /* 2997 * Unwire before removing addresses from the pmap; otherwise, 2998 * unwiring will put the entries back in the pmap. 2999 */ 3000 if (entry->wired_count != 0) { 3001 vm_map_entry_unwire(map, entry); 3002 } 3003 3004 pmap_remove(map->pmap, entry->start, entry->end); 3005 3006 /* 3007 * Delete the entry only after removing all pmap 3008 * entries pointing to its pages. (Otherwise, its 3009 * page frames may be reallocated, and any modify bits 3010 * will be set in the wrong object!) 3011 */ 3012 vm_map_entry_delete(map, entry); 3013 entry = next; 3014 } 3015 return (KERN_SUCCESS); 3016} 3017 3018/* 3019 * vm_map_remove: 3020 * 3021 * Remove the given address range from the target map. 3022 * This is the exported form of vm_map_delete. 3023 */ 3024int 3025vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 3026{ 3027 int result; 3028 3029 vm_map_lock(map); 3030 VM_MAP_RANGE_CHECK(map, start, end); 3031 result = vm_map_delete(map, start, end); 3032 vm_map_unlock(map); 3033 return (result); 3034} 3035 3036/* 3037 * vm_map_check_protection: 3038 * 3039 * Assert that the target map allows the specified privilege on the 3040 * entire address region given. The entire region must be allocated. 3041 * 3042 * WARNING! This code does not and should not check whether the 3043 * contents of the region is accessible. For example a smaller file 3044 * might be mapped into a larger address space. 3045 * 3046 * NOTE! This code is also called by munmap(). 3047 * 3048 * The map must be locked. A read lock is sufficient. 3049 */ 3050boolean_t 3051vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3052 vm_prot_t protection) 3053{ 3054 vm_map_entry_t entry; 3055 vm_map_entry_t tmp_entry; 3056 3057 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3058 return (FALSE); 3059 entry = tmp_entry; 3060 3061 while (start < end) { 3062 if (entry == &map->header) 3063 return (FALSE); 3064 /* 3065 * No holes allowed! 3066 */ 3067 if (start < entry->start) 3068 return (FALSE); 3069 /* 3070 * Check protection associated with entry. 3071 */ 3072 if ((entry->protection & protection) != protection) 3073 return (FALSE); 3074 /* go to next entry */ 3075 start = entry->end; 3076 entry = entry->next; 3077 } 3078 return (TRUE); 3079} 3080 3081/* 3082 * vm_map_copy_entry: 3083 * 3084 * Copies the contents of the source entry to the destination 3085 * entry. The entries *must* be aligned properly. 3086 */ 3087static void 3088vm_map_copy_entry( 3089 vm_map_t src_map, 3090 vm_map_t dst_map, 3091 vm_map_entry_t src_entry, 3092 vm_map_entry_t dst_entry, 3093 vm_ooffset_t *fork_charge) 3094{ 3095 vm_object_t src_object; 3096 vm_map_entry_t fake_entry; 3097 vm_offset_t size; 3098 struct ucred *cred; 3099 int charged; 3100 3101 VM_MAP_ASSERT_LOCKED(dst_map); 3102 3103 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3104 return; 3105 3106 if (src_entry->wired_count == 0 || 3107 (src_entry->protection & VM_PROT_WRITE) == 0) { 3108 /* 3109 * If the source entry is marked needs_copy, it is already 3110 * write-protected. 3111 */ 3112 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3113 (src_entry->protection & VM_PROT_WRITE) != 0) { 3114 pmap_protect(src_map->pmap, 3115 src_entry->start, 3116 src_entry->end, 3117 src_entry->protection & ~VM_PROT_WRITE); 3118 } 3119 3120 /* 3121 * Make a copy of the object. 3122 */ 3123 size = src_entry->end - src_entry->start; 3124 if ((src_object = src_entry->object.vm_object) != NULL) { 3125 VM_OBJECT_WLOCK(src_object); 3126 charged = ENTRY_CHARGED(src_entry); 3127 if (src_object->handle == NULL && 3128 (src_object->type == OBJT_DEFAULT || 3129 src_object->type == OBJT_SWAP)) { 3130 vm_object_collapse(src_object); 3131 if ((src_object->flags & (OBJ_NOSPLIT | 3132 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3133 vm_object_split(src_entry); 3134 src_object = 3135 src_entry->object.vm_object; 3136 } 3137 } 3138 vm_object_reference_locked(src_object); 3139 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3140 if (src_entry->cred != NULL && 3141 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3142 KASSERT(src_object->cred == NULL, 3143 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3144 src_object)); 3145 src_object->cred = src_entry->cred; 3146 src_object->charge = size; 3147 } 3148 VM_OBJECT_WUNLOCK(src_object); 3149 dst_entry->object.vm_object = src_object; 3150 if (charged) { 3151 cred = curthread->td_ucred; 3152 crhold(cred); 3153 dst_entry->cred = cred; 3154 *fork_charge += size; 3155 if (!(src_entry->eflags & 3156 MAP_ENTRY_NEEDS_COPY)) { 3157 crhold(cred); 3158 src_entry->cred = cred; 3159 *fork_charge += size; 3160 } 3161 } 3162 src_entry->eflags |= MAP_ENTRY_COW | 3163 MAP_ENTRY_NEEDS_COPY; 3164 dst_entry->eflags |= MAP_ENTRY_COW | 3165 MAP_ENTRY_NEEDS_COPY; 3166 dst_entry->offset = src_entry->offset; 3167 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3168 /* 3169 * MAP_ENTRY_VN_WRITECNT cannot 3170 * indicate write reference from 3171 * src_entry, since the entry is 3172 * marked as needs copy. Allocate a 3173 * fake entry that is used to 3174 * decrement object->un_pager.vnp.writecount 3175 * at the appropriate time. Attach 3176 * fake_entry to the deferred list. 3177 */ 3178 fake_entry = vm_map_entry_create(dst_map); 3179 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3180 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3181 vm_object_reference(src_object); 3182 fake_entry->object.vm_object = src_object; 3183 fake_entry->start = src_entry->start; 3184 fake_entry->end = src_entry->end; 3185 fake_entry->next = curthread->td_map_def_user; 3186 curthread->td_map_def_user = fake_entry; 3187 } 3188 3189 pmap_copy(dst_map->pmap, src_map->pmap, 3190 dst_entry->start, dst_entry->end - dst_entry->start, 3191 src_entry->start); 3192 } else { 3193 dst_entry->object.vm_object = NULL; 3194 dst_entry->offset = 0; 3195 if (src_entry->cred != NULL) { 3196 dst_entry->cred = curthread->td_ucred; 3197 crhold(dst_entry->cred); 3198 *fork_charge += size; 3199 } 3200 } 3201 } else { 3202 /* 3203 * We don't want to make writeable wired pages copy-on-write. 3204 * Immediately copy these pages into the new map by simulating 3205 * page faults. The new pages are pageable. 3206 */ 3207 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3208 fork_charge); 3209 } 3210} 3211 3212/* 3213 * vmspace_map_entry_forked: 3214 * Update the newly-forked vmspace each time a map entry is inherited 3215 * or copied. The values for vm_dsize and vm_tsize are approximate 3216 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3217 */ 3218static void 3219vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3220 vm_map_entry_t entry) 3221{ 3222 vm_size_t entrysize; 3223 vm_offset_t newend; 3224 3225 entrysize = entry->end - entry->start; 3226 vm2->vm_map.size += entrysize; 3227 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3228 vm2->vm_ssize += btoc(entrysize); 3229 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3230 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3231 newend = MIN(entry->end, 3232 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3233 vm2->vm_dsize += btoc(newend - entry->start); 3234 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3235 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3236 newend = MIN(entry->end, 3237 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3238 vm2->vm_tsize += btoc(newend - entry->start); 3239 } 3240} 3241 3242/* 3243 * vmspace_fork: 3244 * Create a new process vmspace structure and vm_map 3245 * based on those of an existing process. The new map 3246 * is based on the old map, according to the inheritance 3247 * values on the regions in that map. 3248 * 3249 * XXX It might be worth coalescing the entries added to the new vmspace. 3250 * 3251 * The source map must not be locked. 3252 */ 3253struct vmspace * 3254vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3255{ 3256 struct vmspace *vm2; 3257 vm_map_t new_map, old_map; 3258 vm_map_entry_t new_entry, old_entry; 3259 vm_object_t object; 3260 int locked; 3261 3262 old_map = &vm1->vm_map; 3263 /* Copy immutable fields of vm1 to vm2. */ 3264 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3265 if (vm2 == NULL) 3266 return (NULL); 3267 vm2->vm_taddr = vm1->vm_taddr; 3268 vm2->vm_daddr = vm1->vm_daddr; 3269 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3270 vm_map_lock(old_map); 3271 if (old_map->busy) 3272 vm_map_wait_busy(old_map); 3273 new_map = &vm2->vm_map; 3274 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3275 KASSERT(locked, ("vmspace_fork: lock failed")); 3276 3277 old_entry = old_map->header.next; 3278 3279 while (old_entry != &old_map->header) { 3280 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3281 panic("vm_map_fork: encountered a submap"); 3282 3283 switch (old_entry->inheritance) { 3284 case VM_INHERIT_NONE: 3285 break; 3286 3287 case VM_INHERIT_SHARE: 3288 /* 3289 * Clone the entry, creating the shared object if necessary. 3290 */ 3291 object = old_entry->object.vm_object; 3292 if (object == NULL) { 3293 object = vm_object_allocate(OBJT_DEFAULT, 3294 atop(old_entry->end - old_entry->start)); 3295 old_entry->object.vm_object = object; 3296 old_entry->offset = 0; 3297 if (old_entry->cred != NULL) { 3298 object->cred = old_entry->cred; 3299 object->charge = old_entry->end - 3300 old_entry->start; 3301 old_entry->cred = NULL; 3302 } 3303 } 3304 3305 /* 3306 * Add the reference before calling vm_object_shadow 3307 * to insure that a shadow object is created. 3308 */ 3309 vm_object_reference(object); 3310 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3311 vm_object_shadow(&old_entry->object.vm_object, 3312 &old_entry->offset, 3313 old_entry->end - old_entry->start); 3314 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3315 /* Transfer the second reference too. */ 3316 vm_object_reference( 3317 old_entry->object.vm_object); 3318 3319 /* 3320 * As in vm_map_simplify_entry(), the 3321 * vnode lock will not be acquired in 3322 * this call to vm_object_deallocate(). 3323 */ 3324 vm_object_deallocate(object); 3325 object = old_entry->object.vm_object; 3326 } 3327 VM_OBJECT_WLOCK(object); 3328 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3329 if (old_entry->cred != NULL) { 3330 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3331 object->cred = old_entry->cred; 3332 object->charge = old_entry->end - old_entry->start; 3333 old_entry->cred = NULL; 3334 } 3335 3336 /* 3337 * Assert the correct state of the vnode 3338 * v_writecount while the object is locked, to 3339 * not relock it later for the assertion 3340 * correctness. 3341 */ 3342 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3343 object->type == OBJT_VNODE) { 3344 KASSERT(((struct vnode *)object->handle)-> 3345 v_writecount > 0, 3346 ("vmspace_fork: v_writecount %p", object)); 3347 KASSERT(object->un_pager.vnp.writemappings > 0, 3348 ("vmspace_fork: vnp.writecount %p", 3349 object)); 3350 } 3351 VM_OBJECT_WUNLOCK(object); 3352 3353 /* 3354 * Clone the entry, referencing the shared object. 3355 */ 3356 new_entry = vm_map_entry_create(new_map); 3357 *new_entry = *old_entry; 3358 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3359 MAP_ENTRY_IN_TRANSITION); 3360 new_entry->wiring_thread = NULL; 3361 new_entry->wired_count = 0; 3362 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3363 vnode_pager_update_writecount(object, 3364 new_entry->start, new_entry->end); 3365 } 3366 3367 /* 3368 * Insert the entry into the new map -- we know we're 3369 * inserting at the end of the new map. 3370 */ 3371 vm_map_entry_link(new_map, new_map->header.prev, 3372 new_entry); 3373 vmspace_map_entry_forked(vm1, vm2, new_entry); 3374 3375 /* 3376 * Update the physical map 3377 */ 3378 pmap_copy(new_map->pmap, old_map->pmap, 3379 new_entry->start, 3380 (old_entry->end - old_entry->start), 3381 old_entry->start); 3382 break; 3383 3384 case VM_INHERIT_COPY: 3385 /* 3386 * Clone the entry and link into the map. 3387 */ 3388 new_entry = vm_map_entry_create(new_map); 3389 *new_entry = *old_entry; 3390 /* 3391 * Copied entry is COW over the old object. 3392 */ 3393 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3394 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3395 new_entry->wiring_thread = NULL; 3396 new_entry->wired_count = 0; 3397 new_entry->object.vm_object = NULL; 3398 new_entry->cred = NULL; 3399 vm_map_entry_link(new_map, new_map->header.prev, 3400 new_entry); 3401 vmspace_map_entry_forked(vm1, vm2, new_entry); 3402 vm_map_copy_entry(old_map, new_map, old_entry, 3403 new_entry, fork_charge); 3404 break; 3405 3406 case VM_INHERIT_ZERO: 3407 /* 3408 * Create a new anonymous mapping entry modelled from 3409 * the old one. 3410 */ 3411 new_entry = vm_map_entry_create(new_map); 3412 memset(new_entry, 0, sizeof(*new_entry)); 3413 3414 new_entry->start = old_entry->start; 3415 new_entry->end = old_entry->end; 3416 new_entry->avail_ssize = old_entry->avail_ssize; 3417 new_entry->eflags = old_entry->eflags & 3418 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION | 3419 MAP_ENTRY_VN_WRITECNT); 3420 new_entry->protection = old_entry->protection; 3421 new_entry->max_protection = old_entry->max_protection; 3422 new_entry->inheritance = VM_INHERIT_ZERO; 3423 3424 vm_map_entry_link(new_map, new_map->header.prev, 3425 new_entry); 3426 vmspace_map_entry_forked(vm1, vm2, new_entry); 3427 3428 new_entry->cred = curthread->td_ucred; 3429 crhold(new_entry->cred); 3430 *fork_charge += (new_entry->end - new_entry->start); 3431 3432 break; 3433 } 3434 old_entry = old_entry->next; 3435 } 3436 /* 3437 * Use inlined vm_map_unlock() to postpone handling the deferred 3438 * map entries, which cannot be done until both old_map and 3439 * new_map locks are released. 3440 */ 3441 sx_xunlock(&old_map->lock); 3442 sx_xunlock(&new_map->lock); 3443 vm_map_process_deferred(); 3444 3445 return (vm2); 3446} 3447 3448/* 3449 * Create a process's stack for exec_new_vmspace(). This function is never 3450 * asked to wire the newly created stack. 3451 */ 3452int 3453vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3454 vm_prot_t prot, vm_prot_t max, int cow) 3455{ 3456 vm_size_t growsize, init_ssize; 3457 rlim_t vmemlim; 3458 int rv; 3459 3460 MPASS((map->flags & MAP_WIREFUTURE) == 0); 3461 growsize = sgrowsiz; 3462 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3463 vm_map_lock(map); 3464 PROC_LOCK(curproc); 3465 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3466 PROC_UNLOCK(curproc); 3467 /* If we would blow our VMEM resource limit, no go */ 3468 if (map->size + init_ssize > vmemlim) { 3469 rv = KERN_NO_SPACE; 3470 goto out; 3471 } 3472 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3473 max, cow); 3474out: 3475 vm_map_unlock(map); 3476 return (rv); 3477} 3478 3479static int 3480vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3481 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3482{ 3483 vm_map_entry_t new_entry, prev_entry; 3484 vm_offset_t bot, top; 3485 vm_size_t init_ssize; 3486 int orient, rv; 3487 3488 /* 3489 * The stack orientation is piggybacked with the cow argument. 3490 * Extract it into orient and mask the cow argument so that we 3491 * don't pass it around further. 3492 * NOTE: We explicitly allow bi-directional stacks. 3493 */ 3494 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3495 KASSERT(orient != 0, ("No stack grow direction")); 3496 3497 if (addrbos < vm_map_min(map) || 3498 addrbos > vm_map_max(map) || 3499 addrbos + max_ssize < addrbos) 3500 return (KERN_NO_SPACE); 3501 3502 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3503 3504 /* If addr is already mapped, no go */ 3505 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3506 return (KERN_NO_SPACE); 3507 3508 /* 3509 * If we can't accomodate max_ssize in the current mapping, no go. 3510 * However, we need to be aware that subsequent user mappings might 3511 * map into the space we have reserved for stack, and currently this 3512 * space is not protected. 3513 * 3514 * Hopefully we will at least detect this condition when we try to 3515 * grow the stack. 3516 */ 3517 if ((prev_entry->next != &map->header) && 3518 (prev_entry->next->start < addrbos + max_ssize)) 3519 return (KERN_NO_SPACE); 3520 3521 /* 3522 * We initially map a stack of only init_ssize. We will grow as 3523 * needed later. Depending on the orientation of the stack (i.e. 3524 * the grow direction) we either map at the top of the range, the 3525 * bottom of the range or in the middle. 3526 * 3527 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3528 * and cow to be 0. Possibly we should eliminate these as input 3529 * parameters, and just pass these values here in the insert call. 3530 */ 3531 if (orient == MAP_STACK_GROWS_DOWN) 3532 bot = addrbos + max_ssize - init_ssize; 3533 else if (orient == MAP_STACK_GROWS_UP) 3534 bot = addrbos; 3535 else 3536 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3537 top = bot + init_ssize; 3538 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3539 3540 /* Now set the avail_ssize amount. */ 3541 if (rv == KERN_SUCCESS) { 3542 new_entry = prev_entry->next; 3543 if (new_entry->end != top || new_entry->start != bot) 3544 panic("Bad entry start/end for new stack entry"); 3545 3546 new_entry->avail_ssize = max_ssize - init_ssize; 3547 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 || 3548 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, 3549 ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 3550 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 || 3551 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0, 3552 ("new entry lacks MAP_ENTRY_GROWS_UP")); 3553 } 3554 3555 return (rv); 3556} 3557 3558static int stack_guard_page = 0; 3559TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3560SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3561 &stack_guard_page, 0, 3562 "Insert stack guard page ahead of the growable segments."); 3563 3564/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3565 * desired address is already mapped, or if we successfully grow 3566 * the stack. Also returns KERN_SUCCESS if addr is outside the 3567 * stack range (this is strange, but preserves compatibility with 3568 * the grow function in vm_machdep.c). 3569 */ 3570int 3571vm_map_growstack(struct proc *p, vm_offset_t addr) 3572{ 3573 vm_map_entry_t next_entry, prev_entry; 3574 vm_map_entry_t new_entry, stack_entry; 3575 struct vmspace *vm = p->p_vmspace; 3576 vm_map_t map = &vm->vm_map; 3577 vm_offset_t end; 3578 vm_size_t growsize; 3579 size_t grow_amount, max_grow; 3580 rlim_t lmemlim, stacklim, vmemlim; 3581 int is_procstack, rv; 3582 struct ucred *cred; 3583#ifdef notyet 3584 uint64_t limit; 3585#endif 3586#ifdef RACCT 3587 int error; 3588#endif 3589 3590Retry: 3591 PROC_LOCK(p); 3592 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3593 stacklim = lim_cur(p, RLIMIT_STACK); 3594 vmemlim = lim_cur(p, RLIMIT_VMEM); 3595 PROC_UNLOCK(p); 3596 3597 vm_map_lock_read(map); 3598 3599 /* If addr is already in the entry range, no need to grow.*/ 3600 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3601 vm_map_unlock_read(map); 3602 return (KERN_SUCCESS); 3603 } 3604 3605 next_entry = prev_entry->next; 3606 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3607 /* 3608 * This entry does not grow upwards. Since the address lies 3609 * beyond this entry, the next entry (if one exists) has to 3610 * be a downward growable entry. The entry list header is 3611 * never a growable entry, so it suffices to check the flags. 3612 */ 3613 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3614 vm_map_unlock_read(map); 3615 return (KERN_SUCCESS); 3616 } 3617 stack_entry = next_entry; 3618 } else { 3619 /* 3620 * This entry grows upward. If the next entry does not at 3621 * least grow downwards, this is the entry we need to grow. 3622 * otherwise we have two possible choices and we have to 3623 * select one. 3624 */ 3625 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3626 /* 3627 * We have two choices; grow the entry closest to 3628 * the address to minimize the amount of growth. 3629 */ 3630 if (addr - prev_entry->end <= next_entry->start - addr) 3631 stack_entry = prev_entry; 3632 else 3633 stack_entry = next_entry; 3634 } else 3635 stack_entry = prev_entry; 3636 } 3637 3638 if (stack_entry == next_entry) { 3639 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3640 KASSERT(addr < stack_entry->start, ("foo")); 3641 end = (prev_entry != &map->header) ? prev_entry->end : 3642 stack_entry->start - stack_entry->avail_ssize; 3643 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3644 max_grow = stack_entry->start - end; 3645 } else { 3646 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3647 KASSERT(addr >= stack_entry->end, ("foo")); 3648 end = (next_entry != &map->header) ? next_entry->start : 3649 stack_entry->end + stack_entry->avail_ssize; 3650 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3651 max_grow = end - stack_entry->end; 3652 } 3653 3654 if (grow_amount > stack_entry->avail_ssize) { 3655 vm_map_unlock_read(map); 3656 return (KERN_NO_SPACE); 3657 } 3658 3659 /* 3660 * If there is no longer enough space between the entries nogo, and 3661 * adjust the available space. Note: this should only happen if the 3662 * user has mapped into the stack area after the stack was created, 3663 * and is probably an error. 3664 * 3665 * This also effectively destroys any guard page the user might have 3666 * intended by limiting the stack size. 3667 */ 3668 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3669 if (vm_map_lock_upgrade(map)) 3670 goto Retry; 3671 3672 stack_entry->avail_ssize = max_grow; 3673 3674 vm_map_unlock(map); 3675 return (KERN_NO_SPACE); 3676 } 3677 3678 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr && 3679 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0; 3680 3681 /* 3682 * If this is the main process stack, see if we're over the stack 3683 * limit. 3684 */ 3685 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3686 vm_map_unlock_read(map); 3687 return (KERN_NO_SPACE); 3688 } 3689#ifdef RACCT 3690 if (racct_enable) { 3691 PROC_LOCK(p); 3692 if (is_procstack && racct_set(p, RACCT_STACK, 3693 ctob(vm->vm_ssize) + grow_amount)) { 3694 PROC_UNLOCK(p); 3695 vm_map_unlock_read(map); 3696 return (KERN_NO_SPACE); 3697 } 3698 PROC_UNLOCK(p); 3699 } 3700#endif 3701 3702 /* Round up the grow amount modulo sgrowsiz */ 3703 growsize = sgrowsiz; 3704 grow_amount = roundup(grow_amount, growsize); 3705 if (grow_amount > stack_entry->avail_ssize) 3706 grow_amount = stack_entry->avail_ssize; 3707 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3708 grow_amount = trunc_page((vm_size_t)stacklim) - 3709 ctob(vm->vm_ssize); 3710 } 3711#ifdef notyet 3712 PROC_LOCK(p); 3713 limit = racct_get_available(p, RACCT_STACK); 3714 PROC_UNLOCK(p); 3715 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3716 grow_amount = limit - ctob(vm->vm_ssize); 3717#endif 3718 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3719 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3720 vm_map_unlock_read(map); 3721 rv = KERN_NO_SPACE; 3722 goto out; 3723 } 3724#ifdef RACCT 3725 if (racct_enable) { 3726 PROC_LOCK(p); 3727 if (racct_set(p, RACCT_MEMLOCK, 3728 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3729 PROC_UNLOCK(p); 3730 vm_map_unlock_read(map); 3731 rv = KERN_NO_SPACE; 3732 goto out; 3733 } 3734 PROC_UNLOCK(p); 3735 } 3736#endif 3737 } 3738 /* If we would blow our VMEM resource limit, no go */ 3739 if (map->size + grow_amount > vmemlim) { 3740 vm_map_unlock_read(map); 3741 rv = KERN_NO_SPACE; 3742 goto out; 3743 } 3744#ifdef RACCT 3745 if (racct_enable) { 3746 PROC_LOCK(p); 3747 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3748 PROC_UNLOCK(p); 3749 vm_map_unlock_read(map); 3750 rv = KERN_NO_SPACE; 3751 goto out; 3752 } 3753 PROC_UNLOCK(p); 3754 } 3755#endif 3756 3757 if (vm_map_lock_upgrade(map)) 3758 goto Retry; 3759 3760 if (stack_entry == next_entry) { 3761 /* 3762 * Growing downward. 3763 */ 3764 /* Get the preliminary new entry start value */ 3765 addr = stack_entry->start - grow_amount; 3766 3767 /* 3768 * If this puts us into the previous entry, cut back our 3769 * growth to the available space. Also, see the note above. 3770 */ 3771 if (addr < end) { 3772 stack_entry->avail_ssize = max_grow; 3773 addr = end; 3774 if (stack_guard_page) 3775 addr += PAGE_SIZE; 3776 } 3777 3778 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3779 next_entry->protection, next_entry->max_protection, 3780 MAP_STACK_GROWS_DOWN); 3781 3782 /* Adjust the available stack space by the amount we grew. */ 3783 if (rv == KERN_SUCCESS) { 3784 new_entry = prev_entry->next; 3785 KASSERT(new_entry == stack_entry->prev, ("foo")); 3786 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3787 KASSERT(new_entry->start == addr, ("foo")); 3788 KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 3789 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 3790 grow_amount = new_entry->end - new_entry->start; 3791 new_entry->avail_ssize = stack_entry->avail_ssize - 3792 grow_amount; 3793 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3794 } 3795 } else { 3796 /* 3797 * Growing upward. 3798 */ 3799 addr = stack_entry->end + grow_amount; 3800 3801 /* 3802 * If this puts us into the next entry, cut back our growth 3803 * to the available space. Also, see the note above. 3804 */ 3805 if (addr > end) { 3806 stack_entry->avail_ssize = end - stack_entry->end; 3807 addr = end; 3808 if (stack_guard_page) 3809 addr -= PAGE_SIZE; 3810 } 3811 3812 grow_amount = addr - stack_entry->end; 3813 cred = stack_entry->cred; 3814 if (cred == NULL && stack_entry->object.vm_object != NULL) 3815 cred = stack_entry->object.vm_object->cred; 3816 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3817 rv = KERN_NO_SPACE; 3818 /* Grow the underlying object if applicable. */ 3819 else if (stack_entry->object.vm_object == NULL || 3820 vm_object_coalesce(stack_entry->object.vm_object, 3821 stack_entry->offset, 3822 (vm_size_t)(stack_entry->end - stack_entry->start), 3823 (vm_size_t)grow_amount, cred != NULL)) { 3824 map->size += (addr - stack_entry->end); 3825 /* Update the current entry. */ 3826 stack_entry->end = addr; 3827 stack_entry->avail_ssize -= grow_amount; 3828 vm_map_entry_resize_free(map, stack_entry); 3829 rv = KERN_SUCCESS; 3830 3831 if (next_entry != &map->header) 3832 vm_map_clip_start(map, next_entry, addr); 3833 } else 3834 rv = KERN_FAILURE; 3835 } 3836 3837 if (rv == KERN_SUCCESS && is_procstack) 3838 vm->vm_ssize += btoc(grow_amount); 3839 3840 vm_map_unlock(map); 3841 3842 /* 3843 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3844 */ 3845 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3846 vm_map_wire(map, 3847 (stack_entry == next_entry) ? addr : addr - grow_amount, 3848 (stack_entry == next_entry) ? stack_entry->start : addr, 3849 (p->p_flag & P_SYSTEM) 3850 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3851 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3852 } 3853 3854out: 3855#ifdef RACCT 3856 if (racct_enable && rv != KERN_SUCCESS) { 3857 PROC_LOCK(p); 3858 error = racct_set(p, RACCT_VMEM, map->size); 3859 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3860 if (!old_mlock) { 3861 error = racct_set(p, RACCT_MEMLOCK, 3862 ptoa(pmap_wired_count(map->pmap))); 3863 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3864 } 3865 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3866 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3867 PROC_UNLOCK(p); 3868 } 3869#endif 3870 3871 return (rv); 3872} 3873 3874/* 3875 * Unshare the specified VM space for exec. If other processes are 3876 * mapped to it, then create a new one. The new vmspace is null. 3877 */ 3878int 3879vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3880{ 3881 struct vmspace *oldvmspace = p->p_vmspace; 3882 struct vmspace *newvmspace; 3883 3884 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3885 ("vmspace_exec recursed")); 3886 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3887 if (newvmspace == NULL) 3888 return (ENOMEM); 3889 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3890 /* 3891 * This code is written like this for prototype purposes. The 3892 * goal is to avoid running down the vmspace here, but let the 3893 * other process's that are still using the vmspace to finally 3894 * run it down. Even though there is little or no chance of blocking 3895 * here, it is a good idea to keep this form for future mods. 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 curthread->td_pflags |= TDP_EXECVMSPC; 3903 return (0); 3904} 3905 3906/* 3907 * Unshare the specified VM space for forcing COW. This 3908 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3909 */ 3910int 3911vmspace_unshare(struct proc *p) 3912{ 3913 struct vmspace *oldvmspace = p->p_vmspace; 3914 struct vmspace *newvmspace; 3915 vm_ooffset_t fork_charge; 3916 3917 if (oldvmspace->vm_refcnt == 1) 3918 return (0); 3919 fork_charge = 0; 3920 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3921 if (newvmspace == NULL) 3922 return (ENOMEM); 3923 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3924 vmspace_free(newvmspace); 3925 return (ENOMEM); 3926 } 3927 PROC_VMSPACE_LOCK(p); 3928 p->p_vmspace = newvmspace; 3929 PROC_VMSPACE_UNLOCK(p); 3930 if (p == curthread->td_proc) 3931 pmap_activate(curthread); 3932 vmspace_free(oldvmspace); 3933 return (0); 3934} 3935 3936/* 3937 * vm_map_lookup: 3938 * 3939 * Finds the VM object, offset, and 3940 * protection for a given virtual address in the 3941 * specified map, assuming a page fault of the 3942 * type specified. 3943 * 3944 * Leaves the map in question locked for read; return 3945 * values are guaranteed until a vm_map_lookup_done 3946 * call is performed. Note that the map argument 3947 * is in/out; the returned map must be used in 3948 * the call to vm_map_lookup_done. 3949 * 3950 * A handle (out_entry) is returned for use in 3951 * vm_map_lookup_done, to make that fast. 3952 * 3953 * If a lookup is requested with "write protection" 3954 * specified, the map may be changed to perform virtual 3955 * copying operations, although the data referenced will 3956 * remain the same. 3957 */ 3958int 3959vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3960 vm_offset_t vaddr, 3961 vm_prot_t fault_typea, 3962 vm_map_entry_t *out_entry, /* OUT */ 3963 vm_object_t *object, /* OUT */ 3964 vm_pindex_t *pindex, /* OUT */ 3965 vm_prot_t *out_prot, /* OUT */ 3966 boolean_t *wired) /* OUT */ 3967{ 3968 vm_map_entry_t entry; 3969 vm_map_t map = *var_map; 3970 vm_prot_t prot; 3971 vm_prot_t fault_type = fault_typea; 3972 vm_object_t eobject; 3973 vm_size_t size; 3974 struct ucred *cred; 3975 3976RetryLookup:; 3977 3978 vm_map_lock_read(map); 3979 3980 /* 3981 * Lookup the faulting address. 3982 */ 3983 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3984 vm_map_unlock_read(map); 3985 return (KERN_INVALID_ADDRESS); 3986 } 3987 3988 entry = *out_entry; 3989 3990 /* 3991 * Handle submaps. 3992 */ 3993 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3994 vm_map_t old_map = map; 3995 3996 *var_map = map = entry->object.sub_map; 3997 vm_map_unlock_read(old_map); 3998 goto RetryLookup; 3999 } 4000 4001 /* 4002 * Check whether this task is allowed to have this page. 4003 */ 4004 prot = entry->protection; 4005 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 4006 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 4007 vm_map_unlock_read(map); 4008 return (KERN_PROTECTION_FAILURE); 4009 } 4010 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags & 4011 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) != 4012 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY), 4013 ("entry %p flags %x", entry, entry->eflags)); 4014 if ((fault_typea & VM_PROT_COPY) != 0 && 4015 (entry->max_protection & VM_PROT_WRITE) == 0 && 4016 (entry->eflags & MAP_ENTRY_COW) == 0) { 4017 vm_map_unlock_read(map); 4018 return (KERN_PROTECTION_FAILURE); 4019 } 4020 4021 /* 4022 * If this page is not pageable, we have to get it for all possible 4023 * accesses. 4024 */ 4025 *wired = (entry->wired_count != 0); 4026 if (*wired) 4027 fault_type = entry->protection; 4028 size = entry->end - entry->start; 4029 /* 4030 * If the entry was copy-on-write, we either ... 4031 */ 4032 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4033 /* 4034 * If we want to write the page, we may as well handle that 4035 * now since we've got the map locked. 4036 * 4037 * If we don't need to write the page, we just demote the 4038 * permissions allowed. 4039 */ 4040 if ((fault_type & VM_PROT_WRITE) != 0 || 4041 (fault_typea & VM_PROT_COPY) != 0) { 4042 /* 4043 * Make a new object, and place it in the object 4044 * chain. Note that no new references have appeared 4045 * -- one just moved from the map to the new 4046 * object. 4047 */ 4048 if (vm_map_lock_upgrade(map)) 4049 goto RetryLookup; 4050 4051 if (entry->cred == NULL) { 4052 /* 4053 * The debugger owner is charged for 4054 * the memory. 4055 */ 4056 cred = curthread->td_ucred; 4057 crhold(cred); 4058 if (!swap_reserve_by_cred(size, cred)) { 4059 crfree(cred); 4060 vm_map_unlock(map); 4061 return (KERN_RESOURCE_SHORTAGE); 4062 } 4063 entry->cred = cred; 4064 } 4065 vm_object_shadow(&entry->object.vm_object, 4066 &entry->offset, size); 4067 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4068 eobject = entry->object.vm_object; 4069 if (eobject->cred != NULL) { 4070 /* 4071 * The object was not shadowed. 4072 */ 4073 swap_release_by_cred(size, entry->cred); 4074 crfree(entry->cred); 4075 entry->cred = NULL; 4076 } else if (entry->cred != NULL) { 4077 VM_OBJECT_WLOCK(eobject); 4078 eobject->cred = entry->cred; 4079 eobject->charge = size; 4080 VM_OBJECT_WUNLOCK(eobject); 4081 entry->cred = NULL; 4082 } 4083 4084 vm_map_lock_downgrade(map); 4085 } else { 4086 /* 4087 * We're attempting to read a copy-on-write page -- 4088 * don't allow writes. 4089 */ 4090 prot &= ~VM_PROT_WRITE; 4091 } 4092 } 4093 4094 /* 4095 * Create an object if necessary. 4096 */ 4097 if (entry->object.vm_object == NULL && 4098 !map->system_map) { 4099 if (vm_map_lock_upgrade(map)) 4100 goto RetryLookup; 4101 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4102 atop(size)); 4103 entry->offset = 0; 4104 if (entry->cred != NULL) { 4105 VM_OBJECT_WLOCK(entry->object.vm_object); 4106 entry->object.vm_object->cred = entry->cred; 4107 entry->object.vm_object->charge = size; 4108 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4109 entry->cred = NULL; 4110 } 4111 vm_map_lock_downgrade(map); 4112 } 4113 4114 /* 4115 * Return the object/offset from this entry. If the entry was 4116 * copy-on-write or empty, it has been fixed up. 4117 */ 4118 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4119 *object = entry->object.vm_object; 4120 4121 *out_prot = prot; 4122 return (KERN_SUCCESS); 4123} 4124 4125/* 4126 * vm_map_lookup_locked: 4127 * 4128 * Lookup the faulting address. A version of vm_map_lookup that returns 4129 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4130 */ 4131int 4132vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4133 vm_offset_t vaddr, 4134 vm_prot_t fault_typea, 4135 vm_map_entry_t *out_entry, /* OUT */ 4136 vm_object_t *object, /* OUT */ 4137 vm_pindex_t *pindex, /* OUT */ 4138 vm_prot_t *out_prot, /* OUT */ 4139 boolean_t *wired) /* OUT */ 4140{ 4141 vm_map_entry_t entry; 4142 vm_map_t map = *var_map; 4143 vm_prot_t prot; 4144 vm_prot_t fault_type = fault_typea; 4145 4146 /* 4147 * Lookup the faulting address. 4148 */ 4149 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4150 return (KERN_INVALID_ADDRESS); 4151 4152 entry = *out_entry; 4153 4154 /* 4155 * Fail if the entry refers to a submap. 4156 */ 4157 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4158 return (KERN_FAILURE); 4159 4160 /* 4161 * Check whether this task is allowed to have this page. 4162 */ 4163 prot = entry->protection; 4164 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4165 if ((fault_type & prot) != fault_type) 4166 return (KERN_PROTECTION_FAILURE); 4167 4168 /* 4169 * If this page is not pageable, we have to get it for all possible 4170 * accesses. 4171 */ 4172 *wired = (entry->wired_count != 0); 4173 if (*wired) 4174 fault_type = entry->protection; 4175 4176 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4177 /* 4178 * Fail if the entry was copy-on-write for a write fault. 4179 */ 4180 if (fault_type & VM_PROT_WRITE) 4181 return (KERN_FAILURE); 4182 /* 4183 * We're attempting to read a copy-on-write page -- 4184 * don't allow writes. 4185 */ 4186 prot &= ~VM_PROT_WRITE; 4187 } 4188 4189 /* 4190 * Fail if an object should be created. 4191 */ 4192 if (entry->object.vm_object == NULL && !map->system_map) 4193 return (KERN_FAILURE); 4194 4195 /* 4196 * Return the object/offset from this entry. If the entry was 4197 * copy-on-write or empty, it has been fixed up. 4198 */ 4199 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4200 *object = entry->object.vm_object; 4201 4202 *out_prot = prot; 4203 return (KERN_SUCCESS); 4204} 4205 4206/* 4207 * vm_map_lookup_done: 4208 * 4209 * Releases locks acquired by a vm_map_lookup 4210 * (according to the handle returned by that lookup). 4211 */ 4212void 4213vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4214{ 4215 /* 4216 * Unlock the main-level map 4217 */ 4218 vm_map_unlock_read(map); 4219} 4220 4221#include "opt_ddb.h" 4222#ifdef DDB 4223#include <sys/kernel.h> 4224 4225#include <ddb/ddb.h> 4226 4227static void 4228vm_map_print(vm_map_t map) 4229{ 4230 vm_map_entry_t entry; 4231 4232 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4233 (void *)map, 4234 (void *)map->pmap, map->nentries, map->timestamp); 4235 4236 db_indent += 2; 4237 for (entry = map->header.next; entry != &map->header; 4238 entry = entry->next) { 4239 db_iprintf("map entry %p: start=%p, end=%p\n", 4240 (void *)entry, (void *)entry->start, (void *)entry->end); 4241 { 4242 static char *inheritance_name[4] = 4243 {"share", "copy", "none", "donate_copy"}; 4244 4245 db_iprintf(" prot=%x/%x/%s", 4246 entry->protection, 4247 entry->max_protection, 4248 inheritance_name[(int)(unsigned char)entry->inheritance]); 4249 if (entry->wired_count != 0) 4250 db_printf(", wired"); 4251 } 4252 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4253 db_printf(", share=%p, offset=0x%jx\n", 4254 (void *)entry->object.sub_map, 4255 (uintmax_t)entry->offset); 4256 if ((entry->prev == &map->header) || 4257 (entry->prev->object.sub_map != 4258 entry->object.sub_map)) { 4259 db_indent += 2; 4260 vm_map_print((vm_map_t)entry->object.sub_map); 4261 db_indent -= 2; 4262 } 4263 } else { 4264 if (entry->cred != NULL) 4265 db_printf(", ruid %d", entry->cred->cr_ruid); 4266 db_printf(", object=%p, offset=0x%jx", 4267 (void *)entry->object.vm_object, 4268 (uintmax_t)entry->offset); 4269 if (entry->object.vm_object && entry->object.vm_object->cred) 4270 db_printf(", obj ruid %d charge %jx", 4271 entry->object.vm_object->cred->cr_ruid, 4272 (uintmax_t)entry->object.vm_object->charge); 4273 if (entry->eflags & MAP_ENTRY_COW) 4274 db_printf(", copy (%s)", 4275 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4276 db_printf("\n"); 4277 4278 if ((entry->prev == &map->header) || 4279 (entry->prev->object.vm_object != 4280 entry->object.vm_object)) { 4281 db_indent += 2; 4282 vm_object_print((db_expr_t)(intptr_t) 4283 entry->object.vm_object, 4284 0, 0, (char *)0); 4285 db_indent -= 2; 4286 } 4287 } 4288 } 4289 db_indent -= 2; 4290} 4291 4292DB_SHOW_COMMAND(map, map) 4293{ 4294 4295 if (!have_addr) { 4296 db_printf("usage: show map <addr>\n"); 4297 return; 4298 } 4299 vm_map_print((vm_map_t)addr); 4300} 4301 4302DB_SHOW_COMMAND(procvm, procvm) 4303{ 4304 struct proc *p; 4305 4306 if (have_addr) { 4307 p = db_lookup_proc(addr); 4308 } else { 4309 p = curproc; 4310 } 4311 4312 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4313 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4314 (void *)vmspace_pmap(p->p_vmspace)); 4315 4316 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4317} 4318 4319#endif /* DDB */ 4320