vm_map.c revision 267956
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 267956 2014-06-27 11:42:51Z kib $"); 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/kernel.h> 71#include <sys/ktr.h> 72#include <sys/lock.h> 73#include <sys/mutex.h> 74#include <sys/proc.h> 75#include <sys/vmmeter.h> 76#include <sys/mman.h> 77#include <sys/vnode.h> 78#include <sys/racct.h> 79#include <sys/resourcevar.h> 80#include <sys/rwlock.h> 81#include <sys/file.h> 82#include <sys/sysctl.h> 83#include <sys/sysent.h> 84#include <sys/shm.h> 85 86#include <vm/vm.h> 87#include <vm/vm_param.h> 88#include <vm/pmap.h> 89#include <vm/vm_map.h> 90#include <vm/vm_page.h> 91#include <vm/vm_object.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_kern.h> 94#include <vm/vm_extern.h> 95#include <vm/vnode_pager.h> 96#include <vm/swap_pager.h> 97#include <vm/uma.h> 98 99/* 100 * Virtual memory maps provide for the mapping, protection, 101 * and sharing of virtual memory objects. In addition, 102 * this module provides for an efficient virtual copy of 103 * memory from one map to another. 104 * 105 * Synchronization is required prior to most operations. 106 * 107 * Maps consist of an ordered doubly-linked list of simple 108 * entries; a self-adjusting binary search tree of these 109 * entries is used to speed up lookups. 110 * 111 * Since portions of maps are specified by start/end addresses, 112 * which may not align with existing map entries, all 113 * routines merely "clip" entries to these start/end values. 114 * [That is, an entry is split into two, bordering at a 115 * start or end value.] Note that these clippings may not 116 * always be necessary (as the two resulting entries are then 117 * not changed); however, the clipping is done for convenience. 118 * 119 * As mentioned above, virtual copy operations are performed 120 * by copying VM object references from one map to 121 * another, and then marking both regions as copy-on-write. 122 */ 123 124static struct mtx map_sleep_mtx; 125static uma_zone_t mapentzone; 126static uma_zone_t kmapentzone; 127static uma_zone_t mapzone; 128static uma_zone_t vmspace_zone; 129static int vmspace_zinit(void *mem, int size, int flags); 130static int vm_map_zinit(void *mem, int ize, int flags); 131static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 132 vm_offset_t max); 133static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 134static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 135#ifdef INVARIANTS 136static void vm_map_zdtor(void *mem, int size, void *arg); 137static void vmspace_zdtor(void *mem, int size, void *arg); 138#endif 139static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, 140 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, 141 int cow); 142 143#define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 144 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 145 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 146 147/* 148 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 149 * stable. 150 */ 151#define PROC_VMSPACE_LOCK(p) do { } while (0) 152#define PROC_VMSPACE_UNLOCK(p) do { } while (0) 153 154/* 155 * VM_MAP_RANGE_CHECK: [ internal use only ] 156 * 157 * Asserts that the starting and ending region 158 * addresses fall within the valid range of the map. 159 */ 160#define VM_MAP_RANGE_CHECK(map, start, end) \ 161 { \ 162 if (start < vm_map_min(map)) \ 163 start = vm_map_min(map); \ 164 if (end > vm_map_max(map)) \ 165 end = vm_map_max(map); \ 166 if (start > end) \ 167 start = end; \ 168 } 169 170/* 171 * vm_map_startup: 172 * 173 * Initialize the vm_map module. Must be called before 174 * any other vm_map routines. 175 * 176 * Map and entry structures are allocated from the general 177 * purpose memory pool with some exceptions: 178 * 179 * - The kernel map and kmem submap are allocated statically. 180 * - Kernel map entries are allocated out of a static pool. 181 * 182 * These restrictions are necessary since malloc() uses the 183 * maps and requires map entries. 184 */ 185 186void 187vm_map_startup(void) 188{ 189 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 190 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 191#ifdef INVARIANTS 192 vm_map_zdtor, 193#else 194 NULL, 195#endif 196 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 197 uma_prealloc(mapzone, MAX_KMAP); 198 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 200 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 201 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 203 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 204#ifdef INVARIANTS 205 vmspace_zdtor, 206#else 207 NULL, 208#endif 209 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 210} 211 212static int 213vmspace_zinit(void *mem, int size, int flags) 214{ 215 struct vmspace *vm; 216 217 vm = (struct vmspace *)mem; 218 219 vm->vm_map.pmap = NULL; 220 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 221 PMAP_LOCK_INIT(vmspace_pmap(vm)); 222 return (0); 223} 224 225static int 226vm_map_zinit(void *mem, int size, int flags) 227{ 228 vm_map_t map; 229 230 map = (vm_map_t)mem; 231 memset(map, 0, sizeof(*map)); 232 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 233 sx_init(&map->lock, "vm map (user)"); 234 return (0); 235} 236 237#ifdef INVARIANTS 238static void 239vmspace_zdtor(void *mem, int size, void *arg) 240{ 241 struct vmspace *vm; 242 243 vm = (struct vmspace *)mem; 244 245 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 246} 247static void 248vm_map_zdtor(void *mem, int size, void *arg) 249{ 250 vm_map_t map; 251 252 map = (vm_map_t)mem; 253 KASSERT(map->nentries == 0, 254 ("map %p nentries == %d on free.", 255 map, map->nentries)); 256 KASSERT(map->size == 0, 257 ("map %p size == %lu on free.", 258 map, (unsigned long)map->size)); 259} 260#endif /* INVARIANTS */ 261 262/* 263 * Allocate a vmspace structure, including a vm_map and pmap, 264 * and initialize those structures. The refcnt is set to 1. 265 * 266 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 267 */ 268struct vmspace * 269vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 270{ 271 struct vmspace *vm; 272 273 vm = uma_zalloc(vmspace_zone, M_WAITOK); 274 275 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 276 277 if (pinit == NULL) 278 pinit = &pmap_pinit; 279 280 if (!pinit(vmspace_pmap(vm))) { 281 uma_zfree(vmspace_zone, vm); 282 return (NULL); 283 } 284 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 285 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 286 vm->vm_refcnt = 1; 287 vm->vm_shm = NULL; 288 vm->vm_swrss = 0; 289 vm->vm_tsize = 0; 290 vm->vm_dsize = 0; 291 vm->vm_ssize = 0; 292 vm->vm_taddr = 0; 293 vm->vm_daddr = 0; 294 vm->vm_maxsaddr = 0; 295 return (vm); 296} 297 298static void 299vmspace_container_reset(struct proc *p) 300{ 301 302#ifdef RACCT 303 PROC_LOCK(p); 304 racct_set(p, RACCT_DATA, 0); 305 racct_set(p, RACCT_STACK, 0); 306 racct_set(p, RACCT_RSS, 0); 307 racct_set(p, RACCT_MEMLOCK, 0); 308 racct_set(p, RACCT_VMEM, 0); 309 PROC_UNLOCK(p); 310#endif 311} 312 313static inline void 314vmspace_dofree(struct vmspace *vm) 315{ 316 317 CTR1(KTR_VM, "vmspace_free: %p", vm); 318 319 /* 320 * Make sure any SysV shm is freed, it might not have been in 321 * exit1(). 322 */ 323 shmexit(vm); 324 325 /* 326 * Lock the map, to wait out all other references to it. 327 * Delete all of the mappings and pages they hold, then call 328 * the pmap module to reclaim anything left. 329 */ 330 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 331 vm->vm_map.max_offset); 332 333 pmap_release(vmspace_pmap(vm)); 334 vm->vm_map.pmap = NULL; 335 uma_zfree(vmspace_zone, vm); 336} 337 338void 339vmspace_free(struct vmspace *vm) 340{ 341 342 if (vm->vm_refcnt == 0) 343 panic("vmspace_free: attempt to free already freed vmspace"); 344 345 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 346 vmspace_dofree(vm); 347} 348 349void 350vmspace_exitfree(struct proc *p) 351{ 352 struct vmspace *vm; 353 354 PROC_VMSPACE_LOCK(p); 355 vm = p->p_vmspace; 356 p->p_vmspace = NULL; 357 PROC_VMSPACE_UNLOCK(p); 358 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 359 vmspace_free(vm); 360} 361 362void 363vmspace_exit(struct thread *td) 364{ 365 int refcnt; 366 struct vmspace *vm; 367 struct proc *p; 368 369 /* 370 * Release user portion of address space. 371 * This releases references to vnodes, 372 * which could cause I/O if the file has been unlinked. 373 * Need to do this early enough that we can still sleep. 374 * 375 * The last exiting process to reach this point releases as 376 * much of the environment as it can. vmspace_dofree() is the 377 * slower fallback in case another process had a temporary 378 * reference to the vmspace. 379 */ 380 381 p = td->td_proc; 382 vm = p->p_vmspace; 383 atomic_add_int(&vmspace0.vm_refcnt, 1); 384 do { 385 refcnt = vm->vm_refcnt; 386 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 387 /* Switch now since other proc might free vmspace */ 388 PROC_VMSPACE_LOCK(p); 389 p->p_vmspace = &vmspace0; 390 PROC_VMSPACE_UNLOCK(p); 391 pmap_activate(td); 392 } 393 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 394 if (refcnt == 1) { 395 if (p->p_vmspace != vm) { 396 /* vmspace not yet freed, switch back */ 397 PROC_VMSPACE_LOCK(p); 398 p->p_vmspace = vm; 399 PROC_VMSPACE_UNLOCK(p); 400 pmap_activate(td); 401 } 402 pmap_remove_pages(vmspace_pmap(vm)); 403 /* Switch now since this proc will free vmspace */ 404 PROC_VMSPACE_LOCK(p); 405 p->p_vmspace = &vmspace0; 406 PROC_VMSPACE_UNLOCK(p); 407 pmap_activate(td); 408 vmspace_dofree(vm); 409 } 410 vmspace_container_reset(p); 411} 412 413/* Acquire reference to vmspace owned by another process. */ 414 415struct vmspace * 416vmspace_acquire_ref(struct proc *p) 417{ 418 struct vmspace *vm; 419 int refcnt; 420 421 PROC_VMSPACE_LOCK(p); 422 vm = p->p_vmspace; 423 if (vm == NULL) { 424 PROC_VMSPACE_UNLOCK(p); 425 return (NULL); 426 } 427 do { 428 refcnt = vm->vm_refcnt; 429 if (refcnt <= 0) { /* Avoid 0->1 transition */ 430 PROC_VMSPACE_UNLOCK(p); 431 return (NULL); 432 } 433 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 434 if (vm != p->p_vmspace) { 435 PROC_VMSPACE_UNLOCK(p); 436 vmspace_free(vm); 437 return (NULL); 438 } 439 PROC_VMSPACE_UNLOCK(p); 440 return (vm); 441} 442 443void 444_vm_map_lock(vm_map_t map, const char *file, int line) 445{ 446 447 if (map->system_map) 448 mtx_lock_flags_(&map->system_mtx, 0, file, line); 449 else 450 sx_xlock_(&map->lock, file, line); 451 map->timestamp++; 452} 453 454static void 455vm_map_process_deferred(void) 456{ 457 struct thread *td; 458 vm_map_entry_t entry, next; 459 vm_object_t object; 460 461 td = curthread; 462 entry = td->td_map_def_user; 463 td->td_map_def_user = NULL; 464 while (entry != NULL) { 465 next = entry->next; 466 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 467 /* 468 * Decrement the object's writemappings and 469 * possibly the vnode's v_writecount. 470 */ 471 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 472 ("Submap with writecount")); 473 object = entry->object.vm_object; 474 KASSERT(object != NULL, ("No object for writecount")); 475 vnode_pager_release_writecount(object, entry->start, 476 entry->end); 477 } 478 vm_map_entry_deallocate(entry, FALSE); 479 entry = next; 480 } 481} 482 483void 484_vm_map_unlock(vm_map_t map, const char *file, int line) 485{ 486 487 if (map->system_map) 488 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 489 else { 490 sx_xunlock_(&map->lock, file, line); 491 vm_map_process_deferred(); 492 } 493} 494 495void 496_vm_map_lock_read(vm_map_t map, const char *file, int line) 497{ 498 499 if (map->system_map) 500 mtx_lock_flags_(&map->system_mtx, 0, file, line); 501 else 502 sx_slock_(&map->lock, file, line); 503} 504 505void 506_vm_map_unlock_read(vm_map_t map, const char *file, int line) 507{ 508 509 if (map->system_map) 510 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 511 else { 512 sx_sunlock_(&map->lock, file, line); 513 vm_map_process_deferred(); 514 } 515} 516 517int 518_vm_map_trylock(vm_map_t map, const char *file, int line) 519{ 520 int error; 521 522 error = map->system_map ? 523 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 524 !sx_try_xlock_(&map->lock, file, line); 525 if (error == 0) 526 map->timestamp++; 527 return (error == 0); 528} 529 530int 531_vm_map_trylock_read(vm_map_t map, const char *file, int line) 532{ 533 int error; 534 535 error = map->system_map ? 536 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 537 !sx_try_slock_(&map->lock, file, line); 538 return (error == 0); 539} 540 541/* 542 * _vm_map_lock_upgrade: [ internal use only ] 543 * 544 * Tries to upgrade a read (shared) lock on the specified map to a write 545 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 546 * non-zero value if the upgrade fails. If the upgrade fails, the map is 547 * returned without a read or write lock held. 548 * 549 * Requires that the map be read locked. 550 */ 551int 552_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 553{ 554 unsigned int last_timestamp; 555 556 if (map->system_map) { 557 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 558 } else { 559 if (!sx_try_upgrade_(&map->lock, file, line)) { 560 last_timestamp = map->timestamp; 561 sx_sunlock_(&map->lock, file, line); 562 vm_map_process_deferred(); 563 /* 564 * If the map's timestamp does not change while the 565 * map is unlocked, then the upgrade succeeds. 566 */ 567 sx_xlock_(&map->lock, file, line); 568 if (last_timestamp != map->timestamp) { 569 sx_xunlock_(&map->lock, file, line); 570 return (1); 571 } 572 } 573 } 574 map->timestamp++; 575 return (0); 576} 577 578void 579_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 580{ 581 582 if (map->system_map) { 583 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 584 } else 585 sx_downgrade_(&map->lock, file, line); 586} 587 588/* 589 * vm_map_locked: 590 * 591 * Returns a non-zero value if the caller holds a write (exclusive) lock 592 * on the specified map and the value "0" otherwise. 593 */ 594int 595vm_map_locked(vm_map_t map) 596{ 597 598 if (map->system_map) 599 return (mtx_owned(&map->system_mtx)); 600 else 601 return (sx_xlocked(&map->lock)); 602} 603 604#ifdef INVARIANTS 605static void 606_vm_map_assert_locked(vm_map_t map, const char *file, int line) 607{ 608 609 if (map->system_map) 610 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 611 else 612 sx_assert_(&map->lock, SA_XLOCKED, file, line); 613} 614 615#define VM_MAP_ASSERT_LOCKED(map) \ 616 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 617#else 618#define VM_MAP_ASSERT_LOCKED(map) 619#endif 620 621/* 622 * _vm_map_unlock_and_wait: 623 * 624 * Atomically releases the lock on the specified map and puts the calling 625 * thread to sleep. The calling thread will remain asleep until either 626 * vm_map_wakeup() is performed on the map or the specified timeout is 627 * exceeded. 628 * 629 * WARNING! This function does not perform deferred deallocations of 630 * objects and map entries. Therefore, the calling thread is expected to 631 * reacquire the map lock after reawakening and later perform an ordinary 632 * unlock operation, such as vm_map_unlock(), before completing its 633 * operation on the map. 634 */ 635int 636_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 637{ 638 639 mtx_lock(&map_sleep_mtx); 640 if (map->system_map) 641 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 642 else 643 sx_xunlock_(&map->lock, file, line); 644 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 645 timo)); 646} 647 648/* 649 * vm_map_wakeup: 650 * 651 * Awaken any threads that have slept on the map using 652 * vm_map_unlock_and_wait(). 653 */ 654void 655vm_map_wakeup(vm_map_t map) 656{ 657 658 /* 659 * Acquire and release map_sleep_mtx to prevent a wakeup() 660 * from being performed (and lost) between the map unlock 661 * and the msleep() in _vm_map_unlock_and_wait(). 662 */ 663 mtx_lock(&map_sleep_mtx); 664 mtx_unlock(&map_sleep_mtx); 665 wakeup(&map->root); 666} 667 668void 669vm_map_busy(vm_map_t map) 670{ 671 672 VM_MAP_ASSERT_LOCKED(map); 673 map->busy++; 674} 675 676void 677vm_map_unbusy(vm_map_t map) 678{ 679 680 VM_MAP_ASSERT_LOCKED(map); 681 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 682 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 683 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 684 wakeup(&map->busy); 685 } 686} 687 688void 689vm_map_wait_busy(vm_map_t map) 690{ 691 692 VM_MAP_ASSERT_LOCKED(map); 693 while (map->busy) { 694 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 695 if (map->system_map) 696 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 697 else 698 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 699 } 700 map->timestamp++; 701} 702 703long 704vmspace_resident_count(struct vmspace *vmspace) 705{ 706 return pmap_resident_count(vmspace_pmap(vmspace)); 707} 708 709/* 710 * vm_map_create: 711 * 712 * Creates and returns a new empty VM map with 713 * the given physical map structure, and having 714 * the given lower and upper address bounds. 715 */ 716vm_map_t 717vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 718{ 719 vm_map_t result; 720 721 result = uma_zalloc(mapzone, M_WAITOK); 722 CTR1(KTR_VM, "vm_map_create: %p", result); 723 _vm_map_init(result, pmap, min, max); 724 return (result); 725} 726 727/* 728 * Initialize an existing vm_map structure 729 * such as that in the vmspace structure. 730 */ 731static void 732_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 733{ 734 735 map->header.next = map->header.prev = &map->header; 736 map->needs_wakeup = FALSE; 737 map->system_map = 0; 738 map->pmap = pmap; 739 map->min_offset = min; 740 map->max_offset = max; 741 map->flags = 0; 742 map->root = NULL; 743 map->timestamp = 0; 744 map->busy = 0; 745} 746 747void 748vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 749{ 750 751 _vm_map_init(map, pmap, min, max); 752 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 753 sx_init(&map->lock, "user map"); 754} 755 756/* 757 * vm_map_entry_dispose: [ internal use only ] 758 * 759 * Inverse of vm_map_entry_create. 760 */ 761static void 762vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 763{ 764 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 765} 766 767/* 768 * vm_map_entry_create: [ internal use only ] 769 * 770 * Allocates a VM map entry for insertion. 771 * No entry fields are filled in. 772 */ 773static vm_map_entry_t 774vm_map_entry_create(vm_map_t map) 775{ 776 vm_map_entry_t new_entry; 777 778 if (map->system_map) 779 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 780 else 781 new_entry = uma_zalloc(mapentzone, M_WAITOK); 782 if (new_entry == NULL) 783 panic("vm_map_entry_create: kernel resources exhausted"); 784 return (new_entry); 785} 786 787/* 788 * vm_map_entry_set_behavior: 789 * 790 * Set the expected access behavior, either normal, random, or 791 * sequential. 792 */ 793static inline void 794vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 795{ 796 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 797 (behavior & MAP_ENTRY_BEHAV_MASK); 798} 799 800/* 801 * vm_map_entry_set_max_free: 802 * 803 * Set the max_free field in a vm_map_entry. 804 */ 805static inline void 806vm_map_entry_set_max_free(vm_map_entry_t entry) 807{ 808 809 entry->max_free = entry->adj_free; 810 if (entry->left != NULL && entry->left->max_free > entry->max_free) 811 entry->max_free = entry->left->max_free; 812 if (entry->right != NULL && entry->right->max_free > entry->max_free) 813 entry->max_free = entry->right->max_free; 814} 815 816/* 817 * vm_map_entry_splay: 818 * 819 * The Sleator and Tarjan top-down splay algorithm with the 820 * following variation. Max_free must be computed bottom-up, so 821 * on the downward pass, maintain the left and right spines in 822 * reverse order. Then, make a second pass up each side to fix 823 * the pointers and compute max_free. The time bound is O(log n) 824 * amortized. 825 * 826 * The new root is the vm_map_entry containing "addr", or else an 827 * adjacent entry (lower or higher) if addr is not in the tree. 828 * 829 * The map must be locked, and leaves it so. 830 * 831 * Returns: the new root. 832 */ 833static vm_map_entry_t 834vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 835{ 836 vm_map_entry_t llist, rlist; 837 vm_map_entry_t ltree, rtree; 838 vm_map_entry_t y; 839 840 /* Special case of empty tree. */ 841 if (root == NULL) 842 return (root); 843 844 /* 845 * Pass One: Splay down the tree until we find addr or a NULL 846 * pointer where addr would go. llist and rlist are the two 847 * sides in reverse order (bottom-up), with llist linked by 848 * the right pointer and rlist linked by the left pointer in 849 * the vm_map_entry. Wait until Pass Two to set max_free on 850 * the two spines. 851 */ 852 llist = NULL; 853 rlist = NULL; 854 for (;;) { 855 /* root is never NULL in here. */ 856 if (addr < root->start) { 857 y = root->left; 858 if (y == NULL) 859 break; 860 if (addr < y->start && y->left != NULL) { 861 /* Rotate right and put y on rlist. */ 862 root->left = y->right; 863 y->right = root; 864 vm_map_entry_set_max_free(root); 865 root = y->left; 866 y->left = rlist; 867 rlist = y; 868 } else { 869 /* Put root on rlist. */ 870 root->left = rlist; 871 rlist = root; 872 root = y; 873 } 874 } else if (addr >= root->end) { 875 y = root->right; 876 if (y == NULL) 877 break; 878 if (addr >= y->end && y->right != NULL) { 879 /* Rotate left and put y on llist. */ 880 root->right = y->left; 881 y->left = root; 882 vm_map_entry_set_max_free(root); 883 root = y->right; 884 y->right = llist; 885 llist = y; 886 } else { 887 /* Put root on llist. */ 888 root->right = llist; 889 llist = root; 890 root = y; 891 } 892 } else 893 break; 894 } 895 896 /* 897 * Pass Two: Walk back up the two spines, flip the pointers 898 * and set max_free. The subtrees of the root go at the 899 * bottom of llist and rlist. 900 */ 901 ltree = root->left; 902 while (llist != NULL) { 903 y = llist->right; 904 llist->right = ltree; 905 vm_map_entry_set_max_free(llist); 906 ltree = llist; 907 llist = y; 908 } 909 rtree = root->right; 910 while (rlist != NULL) { 911 y = rlist->left; 912 rlist->left = rtree; 913 vm_map_entry_set_max_free(rlist); 914 rtree = rlist; 915 rlist = y; 916 } 917 918 /* 919 * Final assembly: add ltree and rtree as subtrees of root. 920 */ 921 root->left = ltree; 922 root->right = rtree; 923 vm_map_entry_set_max_free(root); 924 925 return (root); 926} 927 928/* 929 * vm_map_entry_{un,}link: 930 * 931 * Insert/remove entries from maps. 932 */ 933static void 934vm_map_entry_link(vm_map_t map, 935 vm_map_entry_t after_where, 936 vm_map_entry_t entry) 937{ 938 939 CTR4(KTR_VM, 940 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 941 map->nentries, entry, after_where); 942 VM_MAP_ASSERT_LOCKED(map); 943 KASSERT(after_where == &map->header || 944 after_where->end <= entry->start, 945 ("vm_map_entry_link: prev end %jx new start %jx overlap", 946 (uintmax_t)after_where->end, (uintmax_t)entry->start)); 947 KASSERT(after_where->next == &map->header || 948 entry->end <= after_where->next->start, 949 ("vm_map_entry_link: new end %jx next start %jx overlap", 950 (uintmax_t)entry->end, (uintmax_t)after_where->next->start)); 951 952 map->nentries++; 953 entry->prev = after_where; 954 entry->next = after_where->next; 955 entry->next->prev = entry; 956 after_where->next = entry; 957 958 if (after_where != &map->header) { 959 if (after_where != map->root) 960 vm_map_entry_splay(after_where->start, map->root); 961 entry->right = after_where->right; 962 entry->left = after_where; 963 after_where->right = NULL; 964 after_where->adj_free = entry->start - after_where->end; 965 vm_map_entry_set_max_free(after_where); 966 } else { 967 entry->right = map->root; 968 entry->left = NULL; 969 } 970 entry->adj_free = (entry->next == &map->header ? map->max_offset : 971 entry->next->start) - entry->end; 972 vm_map_entry_set_max_free(entry); 973 map->root = entry; 974} 975 976static void 977vm_map_entry_unlink(vm_map_t map, 978 vm_map_entry_t entry) 979{ 980 vm_map_entry_t next, prev, root; 981 982 VM_MAP_ASSERT_LOCKED(map); 983 if (entry != map->root) 984 vm_map_entry_splay(entry->start, map->root); 985 if (entry->left == NULL) 986 root = entry->right; 987 else { 988 root = vm_map_entry_splay(entry->start, entry->left); 989 root->right = entry->right; 990 root->adj_free = (entry->next == &map->header ? map->max_offset : 991 entry->next->start) - root->end; 992 vm_map_entry_set_max_free(root); 993 } 994 map->root = root; 995 996 prev = entry->prev; 997 next = entry->next; 998 next->prev = prev; 999 prev->next = next; 1000 map->nentries--; 1001 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1002 map->nentries, entry); 1003} 1004 1005/* 1006 * vm_map_entry_resize_free: 1007 * 1008 * Recompute the amount of free space following a vm_map_entry 1009 * and propagate that value up the tree. Call this function after 1010 * resizing a map entry in-place, that is, without a call to 1011 * vm_map_entry_link() or _unlink(). 1012 * 1013 * The map must be locked, and leaves it so. 1014 */ 1015static void 1016vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1017{ 1018 1019 /* 1020 * Using splay trees without parent pointers, propagating 1021 * max_free up the tree is done by moving the entry to the 1022 * root and making the change there. 1023 */ 1024 if (entry != map->root) 1025 map->root = vm_map_entry_splay(entry->start, map->root); 1026 1027 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1028 entry->next->start) - entry->end; 1029 vm_map_entry_set_max_free(entry); 1030} 1031 1032/* 1033 * vm_map_lookup_entry: [ internal use only ] 1034 * 1035 * Finds the map entry containing (or 1036 * immediately preceding) the specified address 1037 * in the given map; the entry is returned 1038 * in the "entry" parameter. The boolean 1039 * result indicates whether the address is 1040 * actually contained in the map. 1041 */ 1042boolean_t 1043vm_map_lookup_entry( 1044 vm_map_t map, 1045 vm_offset_t address, 1046 vm_map_entry_t *entry) /* OUT */ 1047{ 1048 vm_map_entry_t cur; 1049 boolean_t locked; 1050 1051 /* 1052 * If the map is empty, then the map entry immediately preceding 1053 * "address" is the map's header. 1054 */ 1055 cur = map->root; 1056 if (cur == NULL) 1057 *entry = &map->header; 1058 else if (address >= cur->start && cur->end > address) { 1059 *entry = cur; 1060 return (TRUE); 1061 } else if ((locked = vm_map_locked(map)) || 1062 sx_try_upgrade(&map->lock)) { 1063 /* 1064 * Splay requires a write lock on the map. However, it only 1065 * restructures the binary search tree; it does not otherwise 1066 * change the map. Thus, the map's timestamp need not change 1067 * on a temporary upgrade. 1068 */ 1069 map->root = cur = vm_map_entry_splay(address, cur); 1070 if (!locked) 1071 sx_downgrade(&map->lock); 1072 1073 /* 1074 * If "address" is contained within a map entry, the new root 1075 * is that map entry. Otherwise, the new root is a map entry 1076 * immediately before or after "address". 1077 */ 1078 if (address >= cur->start) { 1079 *entry = cur; 1080 if (cur->end > address) 1081 return (TRUE); 1082 } else 1083 *entry = cur->prev; 1084 } else 1085 /* 1086 * Since the map is only locked for read access, perform a 1087 * standard binary search tree lookup for "address". 1088 */ 1089 for (;;) { 1090 if (address < cur->start) { 1091 if (cur->left == NULL) { 1092 *entry = cur->prev; 1093 break; 1094 } 1095 cur = cur->left; 1096 } else if (cur->end > address) { 1097 *entry = cur; 1098 return (TRUE); 1099 } else { 1100 if (cur->right == NULL) { 1101 *entry = cur; 1102 break; 1103 } 1104 cur = cur->right; 1105 } 1106 } 1107 return (FALSE); 1108} 1109 1110/* 1111 * vm_map_insert: 1112 * 1113 * Inserts the given whole VM object into the target 1114 * map at the specified address range. The object's 1115 * size should match that of the address range. 1116 * 1117 * Requires that the map be locked, and leaves it so. 1118 * 1119 * If object is non-NULL, ref count must be bumped by caller 1120 * prior to making call to account for the new entry. 1121 */ 1122int 1123vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1124 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1125 int cow) 1126{ 1127 vm_map_entry_t new_entry; 1128 vm_map_entry_t prev_entry; 1129 vm_map_entry_t temp_entry; 1130 vm_eflags_t protoeflags; 1131 struct ucred *cred; 1132 vm_inherit_t inheritance; 1133 boolean_t charge_prev_obj; 1134 1135 VM_MAP_ASSERT_LOCKED(map); 1136 1137 /* 1138 * Check that the start and end points are not bogus. 1139 */ 1140 if ((start < map->min_offset) || (end > map->max_offset) || 1141 (start >= end)) 1142 return (KERN_INVALID_ADDRESS); 1143 1144 /* 1145 * Find the entry prior to the proposed starting address; if it's part 1146 * of an existing entry, this range is bogus. 1147 */ 1148 if (vm_map_lookup_entry(map, start, &temp_entry)) 1149 return (KERN_NO_SPACE); 1150 1151 prev_entry = temp_entry; 1152 1153 /* 1154 * Assert that the next entry doesn't overlap the end point. 1155 */ 1156 if ((prev_entry->next != &map->header) && 1157 (prev_entry->next->start < end)) 1158 return (KERN_NO_SPACE); 1159 1160 protoeflags = 0; 1161 charge_prev_obj = FALSE; 1162 1163 if (cow & MAP_COPY_ON_WRITE) 1164 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1165 1166 if (cow & MAP_NOFAULT) { 1167 protoeflags |= MAP_ENTRY_NOFAULT; 1168 1169 KASSERT(object == NULL, 1170 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1171 } 1172 if (cow & MAP_DISABLE_SYNCER) 1173 protoeflags |= MAP_ENTRY_NOSYNC; 1174 if (cow & MAP_DISABLE_COREDUMP) 1175 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1176 if (cow & MAP_VN_WRITECOUNT) 1177 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1178 if (cow & MAP_INHERIT_SHARE) 1179 inheritance = VM_INHERIT_SHARE; 1180 else 1181 inheritance = VM_INHERIT_DEFAULT; 1182 1183 cred = NULL; 1184 KASSERT((object != kmem_object && object != kernel_object) || 1185 ((object == kmem_object || object == kernel_object) && 1186 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1187 ("kmem or kernel object and cow")); 1188 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1189 goto charged; 1190 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1191 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1192 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1193 return (KERN_RESOURCE_SHORTAGE); 1194 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1195 object->cred == NULL, 1196 ("OVERCOMMIT: vm_map_insert o %p", object)); 1197 cred = curthread->td_ucred; 1198 crhold(cred); 1199 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1200 charge_prev_obj = TRUE; 1201 } 1202 1203charged: 1204 /* Expand the kernel pmap, if necessary. */ 1205 if (map == kernel_map && end > kernel_vm_end) 1206 pmap_growkernel(end); 1207 if (object != NULL) { 1208 /* 1209 * OBJ_ONEMAPPING must be cleared unless this mapping 1210 * is trivially proven to be the only mapping for any 1211 * of the object's pages. (Object granularity 1212 * reference counting is insufficient to recognize 1213 * aliases with precision.) 1214 */ 1215 VM_OBJECT_WLOCK(object); 1216 if (object->ref_count > 1 || object->shadow_count != 0) 1217 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1218 VM_OBJECT_WUNLOCK(object); 1219 } 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) && 1224 (prev_entry->wired_count == 0) && 1225 (prev_entry->cred == cred || 1226 (prev_entry->object.vm_object != NULL && 1227 (prev_entry->object.vm_object->cred == cred))) && 1228 vm_object_coalesce(prev_entry->object.vm_object, 1229 prev_entry->offset, 1230 (vm_size_t)(prev_entry->end - prev_entry->start), 1231 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 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 if (cred != NULL) 1245 crfree(cred); 1246 return (KERN_SUCCESS); 1247 } 1248 1249 /* 1250 * If we can extend the object but cannot extend the 1251 * map entry, we have to create a new map entry. We 1252 * must bump the ref count on the extended object to 1253 * account for it. object may be NULL. 1254 */ 1255 object = prev_entry->object.vm_object; 1256 offset = prev_entry->offset + 1257 (prev_entry->end - prev_entry->start); 1258 vm_object_reference(object); 1259 if (cred != NULL && object != NULL && object->cred != NULL && 1260 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1261 /* Object already accounts for this uid. */ 1262 crfree(cred); 1263 cred = NULL; 1264 } 1265 } 1266 1267 /* 1268 * NOTE: if conditionals fail, object can be NULL here. This occurs 1269 * in things like the buffer map where we manage kva but do not manage 1270 * backing objects. 1271 */ 1272 1273 /* 1274 * Create a new entry 1275 */ 1276 new_entry = vm_map_entry_create(map); 1277 new_entry->start = start; 1278 new_entry->end = end; 1279 new_entry->cred = NULL; 1280 1281 new_entry->eflags = protoeflags; 1282 new_entry->object.vm_object = object; 1283 new_entry->offset = offset; 1284 new_entry->avail_ssize = 0; 1285 1286 new_entry->inheritance = inheritance; 1287 new_entry->protection = prot; 1288 new_entry->max_protection = max; 1289 new_entry->wired_count = 0; 1290 new_entry->wiring_thread = NULL; 1291 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1292 new_entry->next_read = OFF_TO_IDX(offset); 1293 1294 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1295 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1296 new_entry->cred = cred; 1297 1298 /* 1299 * Insert the new entry into the list 1300 */ 1301 vm_map_entry_link(map, prev_entry, new_entry); 1302 map->size += new_entry->end - new_entry->start; 1303 1304 /* 1305 * It may be possible to merge the new entry with the next and/or 1306 * previous entries. However, due to MAP_STACK_* being a hack, a 1307 * panic can result from merging such entries. 1308 */ 1309 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1310 vm_map_simplify_entry(map, new_entry); 1311 1312 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1313 vm_map_pmap_enter(map, start, prot, 1314 object, OFF_TO_IDX(offset), end - start, 1315 cow & MAP_PREFAULT_PARTIAL); 1316 } 1317 1318 return (KERN_SUCCESS); 1319} 1320 1321/* 1322 * vm_map_findspace: 1323 * 1324 * Find the first fit (lowest VM address) for "length" free bytes 1325 * beginning at address >= start in the given map. 1326 * 1327 * In a vm_map_entry, "adj_free" is the amount of free space 1328 * adjacent (higher address) to this entry, and "max_free" is the 1329 * maximum amount of contiguous free space in its subtree. This 1330 * allows finding a free region in one path down the tree, so 1331 * O(log n) amortized with splay trees. 1332 * 1333 * The map must be locked, and leaves it so. 1334 * 1335 * Returns: 0 on success, and starting address in *addr, 1336 * 1 if insufficient space. 1337 */ 1338int 1339vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1340 vm_offset_t *addr) /* OUT */ 1341{ 1342 vm_map_entry_t entry; 1343 vm_offset_t st; 1344 1345 /* 1346 * Request must fit within min/max VM address and must avoid 1347 * address wrap. 1348 */ 1349 if (start < map->min_offset) 1350 start = map->min_offset; 1351 if (start + length > map->max_offset || start + length < start) 1352 return (1); 1353 1354 /* Empty tree means wide open address space. */ 1355 if (map->root == NULL) { 1356 *addr = start; 1357 return (0); 1358 } 1359 1360 /* 1361 * After splay, if start comes before root node, then there 1362 * must be a gap from start to the root. 1363 */ 1364 map->root = vm_map_entry_splay(start, map->root); 1365 if (start + length <= map->root->start) { 1366 *addr = start; 1367 return (0); 1368 } 1369 1370 /* 1371 * Root is the last node that might begin its gap before 1372 * start, and this is the last comparison where address 1373 * wrap might be a problem. 1374 */ 1375 st = (start > map->root->end) ? start : map->root->end; 1376 if (length <= map->root->end + map->root->adj_free - st) { 1377 *addr = st; 1378 return (0); 1379 } 1380 1381 /* With max_free, can immediately tell if no solution. */ 1382 entry = map->root->right; 1383 if (entry == NULL || length > entry->max_free) 1384 return (1); 1385 1386 /* 1387 * Search the right subtree in the order: left subtree, root, 1388 * right subtree (first fit). The previous splay implies that 1389 * all regions in the right subtree have addresses > start. 1390 */ 1391 while (entry != NULL) { 1392 if (entry->left != NULL && entry->left->max_free >= length) 1393 entry = entry->left; 1394 else if (entry->adj_free >= length) { 1395 *addr = entry->end; 1396 return (0); 1397 } else 1398 entry = entry->right; 1399 } 1400 1401 /* Can't get here, so panic if we do. */ 1402 panic("vm_map_findspace: max_free corrupt"); 1403} 1404 1405int 1406vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1407 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1408 vm_prot_t max, int cow) 1409{ 1410 vm_offset_t end; 1411 int result; 1412 1413 end = start + length; 1414 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1415 object == NULL, 1416 ("vm_map_fixed: non-NULL backing object for stack")); 1417 vm_map_lock(map); 1418 VM_MAP_RANGE_CHECK(map, start, end); 1419 if ((cow & MAP_CHECK_EXCL) == 0) 1420 vm_map_delete(map, start, end); 1421 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1422 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1423 prot, max, cow); 1424 } else { 1425 result = vm_map_insert(map, object, offset, start, end, 1426 prot, max, cow); 1427 } 1428 vm_map_unlock(map); 1429 return (result); 1430} 1431 1432/* 1433 * vm_map_find finds an unallocated region in the target address 1434 * map with the given length. The search is defined to be 1435 * first-fit from the specified address; the region found is 1436 * returned in the same parameter. 1437 * 1438 * If object is non-NULL, ref count must be bumped by caller 1439 * prior to making call to account for the new entry. 1440 */ 1441int 1442vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1443 vm_offset_t *addr, /* IN/OUT */ 1444 vm_size_t length, vm_offset_t max_addr, int find_space, 1445 vm_prot_t prot, vm_prot_t max, int cow) 1446{ 1447 vm_offset_t alignment, initial_addr, start; 1448 int result; 1449 1450 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1451 object == NULL, 1452 ("vm_map_find: non-NULL backing object for stack")); 1453 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1454 (object->flags & OBJ_COLORED) == 0)) 1455 find_space = VMFS_ANY_SPACE; 1456 if (find_space >> 8 != 0) { 1457 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1458 alignment = (vm_offset_t)1 << (find_space >> 8); 1459 } else 1460 alignment = 0; 1461 initial_addr = *addr; 1462again: 1463 start = initial_addr; 1464 vm_map_lock(map); 1465 do { 1466 if (find_space != VMFS_NO_SPACE) { 1467 if (vm_map_findspace(map, start, length, addr) || 1468 (max_addr != 0 && *addr + length > max_addr)) { 1469 vm_map_unlock(map); 1470 if (find_space == VMFS_OPTIMAL_SPACE) { 1471 find_space = VMFS_ANY_SPACE; 1472 goto again; 1473 } 1474 return (KERN_NO_SPACE); 1475 } 1476 switch (find_space) { 1477 case VMFS_SUPER_SPACE: 1478 case VMFS_OPTIMAL_SPACE: 1479 pmap_align_superpage(object, offset, addr, 1480 length); 1481 break; 1482 case VMFS_ANY_SPACE: 1483 break; 1484 default: 1485 if ((*addr & (alignment - 1)) != 0) { 1486 *addr &= ~(alignment - 1); 1487 *addr += alignment; 1488 } 1489 break; 1490 } 1491 1492 start = *addr; 1493 } 1494 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1495 result = vm_map_stack_locked(map, start, length, 1496 sgrowsiz, prot, max, cow); 1497 } else { 1498 result = vm_map_insert(map, object, offset, start, 1499 start + length, prot, max, cow); 1500 } 1501 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1502 find_space != VMFS_ANY_SPACE); 1503 vm_map_unlock(map); 1504 return (result); 1505} 1506 1507/* 1508 * vm_map_simplify_entry: 1509 * 1510 * Simplify the given map entry by merging with either neighbor. This 1511 * routine also has the ability to merge with both neighbors. 1512 * 1513 * The map must be locked. 1514 * 1515 * This routine guarentees that the passed entry remains valid (though 1516 * possibly extended). When merging, this routine may delete one or 1517 * both neighbors. 1518 */ 1519void 1520vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1521{ 1522 vm_map_entry_t next, prev; 1523 vm_size_t prevsize, esize; 1524 1525 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1526 return; 1527 1528 prev = entry->prev; 1529 if (prev != &map->header) { 1530 prevsize = prev->end - prev->start; 1531 if ( (prev->end == entry->start) && 1532 (prev->object.vm_object == entry->object.vm_object) && 1533 (!prev->object.vm_object || 1534 (prev->offset + prevsize == entry->offset)) && 1535 (prev->eflags == entry->eflags) && 1536 (prev->protection == entry->protection) && 1537 (prev->max_protection == entry->max_protection) && 1538 (prev->inheritance == entry->inheritance) && 1539 (prev->wired_count == entry->wired_count) && 1540 (prev->cred == entry->cred)) { 1541 vm_map_entry_unlink(map, prev); 1542 entry->start = prev->start; 1543 entry->offset = prev->offset; 1544 if (entry->prev != &map->header) 1545 vm_map_entry_resize_free(map, entry->prev); 1546 1547 /* 1548 * If the backing object is a vnode object, 1549 * vm_object_deallocate() calls vrele(). 1550 * However, vrele() does not lock the vnode 1551 * because the vnode has additional 1552 * references. Thus, the map lock can be kept 1553 * without causing a lock-order reversal with 1554 * the vnode lock. 1555 * 1556 * Since we count the number of virtual page 1557 * mappings in object->un_pager.vnp.writemappings, 1558 * the writemappings value should not be adjusted 1559 * when the entry is disposed of. 1560 */ 1561 if (prev->object.vm_object) 1562 vm_object_deallocate(prev->object.vm_object); 1563 if (prev->cred != NULL) 1564 crfree(prev->cred); 1565 vm_map_entry_dispose(map, prev); 1566 } 1567 } 1568 1569 next = entry->next; 1570 if (next != &map->header) { 1571 esize = entry->end - entry->start; 1572 if ((entry->end == next->start) && 1573 (next->object.vm_object == entry->object.vm_object) && 1574 (!entry->object.vm_object || 1575 (entry->offset + esize == next->offset)) && 1576 (next->eflags == entry->eflags) && 1577 (next->protection == entry->protection) && 1578 (next->max_protection == entry->max_protection) && 1579 (next->inheritance == entry->inheritance) && 1580 (next->wired_count == entry->wired_count) && 1581 (next->cred == entry->cred)) { 1582 vm_map_entry_unlink(map, next); 1583 entry->end = next->end; 1584 vm_map_entry_resize_free(map, entry); 1585 1586 /* 1587 * See comment above. 1588 */ 1589 if (next->object.vm_object) 1590 vm_object_deallocate(next->object.vm_object); 1591 if (next->cred != NULL) 1592 crfree(next->cred); 1593 vm_map_entry_dispose(map, next); 1594 } 1595 } 1596} 1597/* 1598 * vm_map_clip_start: [ internal use only ] 1599 * 1600 * Asserts that the given entry begins at or after 1601 * the specified address; if necessary, 1602 * it splits the entry into two. 1603 */ 1604#define vm_map_clip_start(map, entry, startaddr) \ 1605{ \ 1606 if (startaddr > entry->start) \ 1607 _vm_map_clip_start(map, entry, startaddr); \ 1608} 1609 1610/* 1611 * This routine is called only when it is known that 1612 * the entry must be split. 1613 */ 1614static void 1615_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1616{ 1617 vm_map_entry_t new_entry; 1618 1619 VM_MAP_ASSERT_LOCKED(map); 1620 1621 /* 1622 * Split off the front portion -- note that we must insert the new 1623 * entry BEFORE this one, so that this entry has the specified 1624 * starting address. 1625 */ 1626 vm_map_simplify_entry(map, entry); 1627 1628 /* 1629 * If there is no object backing this entry, we might as well create 1630 * one now. If we defer it, an object can get created after the map 1631 * is clipped, and individual objects will be created for the split-up 1632 * map. This is a bit of a hack, but is also about the best place to 1633 * put this improvement. 1634 */ 1635 if (entry->object.vm_object == NULL && !map->system_map) { 1636 vm_object_t object; 1637 object = vm_object_allocate(OBJT_DEFAULT, 1638 atop(entry->end - entry->start)); 1639 entry->object.vm_object = object; 1640 entry->offset = 0; 1641 if (entry->cred != NULL) { 1642 object->cred = entry->cred; 1643 object->charge = entry->end - entry->start; 1644 entry->cred = NULL; 1645 } 1646 } else if (entry->object.vm_object != NULL && 1647 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1648 entry->cred != NULL) { 1649 VM_OBJECT_WLOCK(entry->object.vm_object); 1650 KASSERT(entry->object.vm_object->cred == NULL, 1651 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1652 entry->object.vm_object->cred = entry->cred; 1653 entry->object.vm_object->charge = entry->end - entry->start; 1654 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1655 entry->cred = NULL; 1656 } 1657 1658 new_entry = vm_map_entry_create(map); 1659 *new_entry = *entry; 1660 1661 new_entry->end = start; 1662 entry->offset += (start - entry->start); 1663 entry->start = start; 1664 if (new_entry->cred != NULL) 1665 crhold(entry->cred); 1666 1667 vm_map_entry_link(map, entry->prev, new_entry); 1668 1669 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1670 vm_object_reference(new_entry->object.vm_object); 1671 /* 1672 * The object->un_pager.vnp.writemappings for the 1673 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1674 * kept as is here. The virtual pages are 1675 * re-distributed among the clipped entries, so the sum is 1676 * left the same. 1677 */ 1678 } 1679} 1680 1681/* 1682 * vm_map_clip_end: [ internal use only ] 1683 * 1684 * Asserts that the given entry ends at or before 1685 * the specified address; if necessary, 1686 * it splits the entry into two. 1687 */ 1688#define vm_map_clip_end(map, entry, endaddr) \ 1689{ \ 1690 if ((endaddr) < (entry->end)) \ 1691 _vm_map_clip_end((map), (entry), (endaddr)); \ 1692} 1693 1694/* 1695 * This routine is called only when it is known that 1696 * the entry must be split. 1697 */ 1698static void 1699_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1700{ 1701 vm_map_entry_t new_entry; 1702 1703 VM_MAP_ASSERT_LOCKED(map); 1704 1705 /* 1706 * If there is no object backing this entry, we might as well create 1707 * one now. If we defer it, an object can get created after the map 1708 * is clipped, and individual objects will be created for the split-up 1709 * map. This is a bit of a hack, but is also about the best place to 1710 * put this improvement. 1711 */ 1712 if (entry->object.vm_object == NULL && !map->system_map) { 1713 vm_object_t object; 1714 object = vm_object_allocate(OBJT_DEFAULT, 1715 atop(entry->end - entry->start)); 1716 entry->object.vm_object = object; 1717 entry->offset = 0; 1718 if (entry->cred != NULL) { 1719 object->cred = entry->cred; 1720 object->charge = entry->end - entry->start; 1721 entry->cred = NULL; 1722 } 1723 } else if (entry->object.vm_object != NULL && 1724 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1725 entry->cred != NULL) { 1726 VM_OBJECT_WLOCK(entry->object.vm_object); 1727 KASSERT(entry->object.vm_object->cred == NULL, 1728 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1729 entry->object.vm_object->cred = entry->cred; 1730 entry->object.vm_object->charge = entry->end - entry->start; 1731 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1732 entry->cred = NULL; 1733 } 1734 1735 /* 1736 * Create a new entry and insert it AFTER the specified entry 1737 */ 1738 new_entry = vm_map_entry_create(map); 1739 *new_entry = *entry; 1740 1741 new_entry->start = entry->end = end; 1742 new_entry->offset += (end - entry->start); 1743 if (new_entry->cred != NULL) 1744 crhold(entry->cred); 1745 1746 vm_map_entry_link(map, entry, new_entry); 1747 1748 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1749 vm_object_reference(new_entry->object.vm_object); 1750 } 1751} 1752 1753/* 1754 * vm_map_submap: [ kernel use only ] 1755 * 1756 * Mark the given range as handled by a subordinate map. 1757 * 1758 * This range must have been created with vm_map_find, 1759 * and no other operations may have been performed on this 1760 * range prior to calling vm_map_submap. 1761 * 1762 * Only a limited number of operations can be performed 1763 * within this rage after calling vm_map_submap: 1764 * vm_fault 1765 * [Don't try vm_map_copy!] 1766 * 1767 * To remove a submapping, one must first remove the 1768 * range from the superior map, and then destroy the 1769 * submap (if desired). [Better yet, don't try it.] 1770 */ 1771int 1772vm_map_submap( 1773 vm_map_t map, 1774 vm_offset_t start, 1775 vm_offset_t end, 1776 vm_map_t submap) 1777{ 1778 vm_map_entry_t entry; 1779 int result = KERN_INVALID_ARGUMENT; 1780 1781 vm_map_lock(map); 1782 1783 VM_MAP_RANGE_CHECK(map, start, end); 1784 1785 if (vm_map_lookup_entry(map, start, &entry)) { 1786 vm_map_clip_start(map, entry, start); 1787 } else 1788 entry = entry->next; 1789 1790 vm_map_clip_end(map, entry, end); 1791 1792 if ((entry->start == start) && (entry->end == end) && 1793 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1794 (entry->object.vm_object == NULL)) { 1795 entry->object.sub_map = submap; 1796 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1797 result = KERN_SUCCESS; 1798 } 1799 vm_map_unlock(map); 1800 1801 return (result); 1802} 1803 1804/* 1805 * The maximum number of pages to map 1806 */ 1807#define MAX_INIT_PT 96 1808 1809/* 1810 * vm_map_pmap_enter: 1811 * 1812 * Preload read-only mappings for the specified object's resident pages 1813 * into the target map. If "flags" is MAP_PREFAULT_PARTIAL, then only 1814 * the resident pages within the address range [addr, addr + ulmin(size, 1815 * ptoa(MAX_INIT_PT))) are mapped. Otherwise, all resident pages within 1816 * the specified address range are mapped. This eliminates many soft 1817 * faults on process startup and immediately after an mmap(2). Because 1818 * these are speculative mappings, cached pages are not reactivated and 1819 * mapped. 1820 */ 1821void 1822vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1823 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1824{ 1825 vm_offset_t start; 1826 vm_page_t p, p_start; 1827 vm_pindex_t psize, tmpidx; 1828 1829 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1830 return; 1831 VM_OBJECT_RLOCK(object); 1832 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1833 VM_OBJECT_RUNLOCK(object); 1834 VM_OBJECT_WLOCK(object); 1835 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1836 pmap_object_init_pt(map->pmap, addr, object, pindex, 1837 size); 1838 VM_OBJECT_WUNLOCK(object); 1839 return; 1840 } 1841 VM_OBJECT_LOCK_DOWNGRADE(object); 1842 } 1843 1844 psize = atop(size); 1845 if (psize > MAX_INIT_PT && (flags & MAP_PREFAULT_PARTIAL) != 0) 1846 psize = MAX_INIT_PT; 1847 if (psize + pindex > object->size) { 1848 if (object->size < pindex) { 1849 VM_OBJECT_RUNLOCK(object); 1850 return; 1851 } 1852 psize = object->size - pindex; 1853 } 1854 1855 start = 0; 1856 p_start = NULL; 1857 1858 p = vm_page_find_least(object, pindex); 1859 /* 1860 * Assert: the variable p is either (1) the page with the 1861 * least pindex greater than or equal to the parameter pindex 1862 * or (2) NULL. 1863 */ 1864 for (; 1865 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1866 p = TAILQ_NEXT(p, listq)) { 1867 /* 1868 * don't allow an madvise to blow away our really 1869 * free pages allocating pv entries. 1870 */ 1871 if ((flags & MAP_PREFAULT_MADVISE) && 1872 cnt.v_free_count < cnt.v_free_reserved) { 1873 psize = tmpidx; 1874 break; 1875 } 1876 if (p->valid == VM_PAGE_BITS_ALL) { 1877 if (p_start == NULL) { 1878 start = addr + ptoa(tmpidx); 1879 p_start = p; 1880 } 1881 } else if (p_start != NULL) { 1882 pmap_enter_object(map->pmap, start, addr + 1883 ptoa(tmpidx), p_start, prot); 1884 p_start = NULL; 1885 } 1886 } 1887 if (p_start != NULL) 1888 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1889 p_start, prot); 1890 VM_OBJECT_RUNLOCK(object); 1891} 1892 1893/* 1894 * vm_map_protect: 1895 * 1896 * Sets the protection of the specified address 1897 * region in the target map. If "set_max" is 1898 * specified, the maximum protection is to be set; 1899 * otherwise, only the current protection is affected. 1900 */ 1901int 1902vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1903 vm_prot_t new_prot, boolean_t set_max) 1904{ 1905 vm_map_entry_t current, entry; 1906 vm_object_t obj; 1907 struct ucred *cred; 1908 vm_prot_t old_prot; 1909 1910 if (start == end) 1911 return (KERN_SUCCESS); 1912 1913 vm_map_lock(map); 1914 1915 VM_MAP_RANGE_CHECK(map, start, end); 1916 1917 if (vm_map_lookup_entry(map, start, &entry)) { 1918 vm_map_clip_start(map, entry, start); 1919 } else { 1920 entry = entry->next; 1921 } 1922 1923 /* 1924 * Make a first pass to check for protection violations. 1925 */ 1926 current = entry; 1927 while ((current != &map->header) && (current->start < end)) { 1928 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1929 vm_map_unlock(map); 1930 return (KERN_INVALID_ARGUMENT); 1931 } 1932 if ((new_prot & current->max_protection) != new_prot) { 1933 vm_map_unlock(map); 1934 return (KERN_PROTECTION_FAILURE); 1935 } 1936 current = current->next; 1937 } 1938 1939 1940 /* 1941 * Do an accounting pass for private read-only mappings that 1942 * now will do cow due to allowed write (e.g. debugger sets 1943 * breakpoint on text segment) 1944 */ 1945 for (current = entry; (current != &map->header) && 1946 (current->start < end); current = current->next) { 1947 1948 vm_map_clip_end(map, current, end); 1949 1950 if (set_max || 1951 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1952 ENTRY_CHARGED(current)) { 1953 continue; 1954 } 1955 1956 cred = curthread->td_ucred; 1957 obj = current->object.vm_object; 1958 1959 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1960 if (!swap_reserve(current->end - current->start)) { 1961 vm_map_unlock(map); 1962 return (KERN_RESOURCE_SHORTAGE); 1963 } 1964 crhold(cred); 1965 current->cred = cred; 1966 continue; 1967 } 1968 1969 VM_OBJECT_WLOCK(obj); 1970 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1971 VM_OBJECT_WUNLOCK(obj); 1972 continue; 1973 } 1974 1975 /* 1976 * Charge for the whole object allocation now, since 1977 * we cannot distinguish between non-charged and 1978 * charged clipped mapping of the same object later. 1979 */ 1980 KASSERT(obj->charge == 0, 1981 ("vm_map_protect: object %p overcharged (entry %p)", 1982 obj, current)); 1983 if (!swap_reserve(ptoa(obj->size))) { 1984 VM_OBJECT_WUNLOCK(obj); 1985 vm_map_unlock(map); 1986 return (KERN_RESOURCE_SHORTAGE); 1987 } 1988 1989 crhold(cred); 1990 obj->cred = cred; 1991 obj->charge = ptoa(obj->size); 1992 VM_OBJECT_WUNLOCK(obj); 1993 } 1994 1995 /* 1996 * Go back and fix up protections. [Note that clipping is not 1997 * necessary the second time.] 1998 */ 1999 current = entry; 2000 while ((current != &map->header) && (current->start < end)) { 2001 old_prot = current->protection; 2002 2003 if (set_max) 2004 current->protection = 2005 (current->max_protection = new_prot) & 2006 old_prot; 2007 else 2008 current->protection = new_prot; 2009 2010 /* 2011 * For user wired map entries, the normal lazy evaluation of 2012 * write access upgrades through soft page faults is 2013 * undesirable. Instead, immediately copy any pages that are 2014 * copy-on-write and enable write access in the physical map. 2015 */ 2016 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2017 (current->protection & VM_PROT_WRITE) != 0 && 2018 (old_prot & VM_PROT_WRITE) == 0) 2019 vm_fault_copy_entry(map, map, current, current, NULL); 2020 2021 /* 2022 * When restricting access, update the physical map. Worry 2023 * about copy-on-write here. 2024 */ 2025 if ((old_prot & ~current->protection) != 0) { 2026#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2027 VM_PROT_ALL) 2028 pmap_protect(map->pmap, current->start, 2029 current->end, 2030 current->protection & MASK(current)); 2031#undef MASK 2032 } 2033 vm_map_simplify_entry(map, current); 2034 current = current->next; 2035 } 2036 vm_map_unlock(map); 2037 return (KERN_SUCCESS); 2038} 2039 2040/* 2041 * vm_map_madvise: 2042 * 2043 * This routine traverses a processes map handling the madvise 2044 * system call. Advisories are classified as either those effecting 2045 * the vm_map_entry structure, or those effecting the underlying 2046 * objects. 2047 */ 2048int 2049vm_map_madvise( 2050 vm_map_t map, 2051 vm_offset_t start, 2052 vm_offset_t end, 2053 int behav) 2054{ 2055 vm_map_entry_t current, entry; 2056 int modify_map = 0; 2057 2058 /* 2059 * Some madvise calls directly modify the vm_map_entry, in which case 2060 * we need to use an exclusive lock on the map and we need to perform 2061 * various clipping operations. Otherwise we only need a read-lock 2062 * on the map. 2063 */ 2064 switch(behav) { 2065 case MADV_NORMAL: 2066 case MADV_SEQUENTIAL: 2067 case MADV_RANDOM: 2068 case MADV_NOSYNC: 2069 case MADV_AUTOSYNC: 2070 case MADV_NOCORE: 2071 case MADV_CORE: 2072 if (start == end) 2073 return (KERN_SUCCESS); 2074 modify_map = 1; 2075 vm_map_lock(map); 2076 break; 2077 case MADV_WILLNEED: 2078 case MADV_DONTNEED: 2079 case MADV_FREE: 2080 if (start == end) 2081 return (KERN_SUCCESS); 2082 vm_map_lock_read(map); 2083 break; 2084 default: 2085 return (KERN_INVALID_ARGUMENT); 2086 } 2087 2088 /* 2089 * Locate starting entry and clip if necessary. 2090 */ 2091 VM_MAP_RANGE_CHECK(map, start, end); 2092 2093 if (vm_map_lookup_entry(map, start, &entry)) { 2094 if (modify_map) 2095 vm_map_clip_start(map, entry, start); 2096 } else { 2097 entry = entry->next; 2098 } 2099 2100 if (modify_map) { 2101 /* 2102 * madvise behaviors that are implemented in the vm_map_entry. 2103 * 2104 * We clip the vm_map_entry so that behavioral changes are 2105 * limited to the specified address range. 2106 */ 2107 for (current = entry; 2108 (current != &map->header) && (current->start < end); 2109 current = current->next 2110 ) { 2111 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2112 continue; 2113 2114 vm_map_clip_end(map, current, end); 2115 2116 switch (behav) { 2117 case MADV_NORMAL: 2118 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2119 break; 2120 case MADV_SEQUENTIAL: 2121 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2122 break; 2123 case MADV_RANDOM: 2124 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2125 break; 2126 case MADV_NOSYNC: 2127 current->eflags |= MAP_ENTRY_NOSYNC; 2128 break; 2129 case MADV_AUTOSYNC: 2130 current->eflags &= ~MAP_ENTRY_NOSYNC; 2131 break; 2132 case MADV_NOCORE: 2133 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2134 break; 2135 case MADV_CORE: 2136 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2137 break; 2138 default: 2139 break; 2140 } 2141 vm_map_simplify_entry(map, current); 2142 } 2143 vm_map_unlock(map); 2144 } else { 2145 vm_pindex_t pstart, pend; 2146 2147 /* 2148 * madvise behaviors that are implemented in the underlying 2149 * vm_object. 2150 * 2151 * Since we don't clip the vm_map_entry, we have to clip 2152 * the vm_object pindex and count. 2153 */ 2154 for (current = entry; 2155 (current != &map->header) && (current->start < end); 2156 current = current->next 2157 ) { 2158 vm_offset_t useEnd, useStart; 2159 2160 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2161 continue; 2162 2163 pstart = OFF_TO_IDX(current->offset); 2164 pend = pstart + atop(current->end - current->start); 2165 useStart = current->start; 2166 useEnd = current->end; 2167 2168 if (current->start < start) { 2169 pstart += atop(start - current->start); 2170 useStart = start; 2171 } 2172 if (current->end > end) { 2173 pend -= atop(current->end - end); 2174 useEnd = end; 2175 } 2176 2177 if (pstart >= pend) 2178 continue; 2179 2180 /* 2181 * Perform the pmap_advise() before clearing 2182 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2183 * concurrent pmap operation, such as pmap_remove(), 2184 * could clear a reference in the pmap and set 2185 * PGA_REFERENCED on the page before the pmap_advise() 2186 * had completed. Consequently, the page would appear 2187 * referenced based upon an old reference that 2188 * occurred before this pmap_advise() ran. 2189 */ 2190 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2191 pmap_advise(map->pmap, useStart, useEnd, 2192 behav); 2193 2194 vm_object_madvise(current->object.vm_object, pstart, 2195 pend, behav); 2196 if (behav == MADV_WILLNEED) { 2197 vm_map_pmap_enter(map, 2198 useStart, 2199 current->protection, 2200 current->object.vm_object, 2201 pstart, 2202 ptoa(pend - pstart), 2203 MAP_PREFAULT_MADVISE 2204 ); 2205 } 2206 } 2207 vm_map_unlock_read(map); 2208 } 2209 return (0); 2210} 2211 2212 2213/* 2214 * vm_map_inherit: 2215 * 2216 * Sets the inheritance of the specified address 2217 * range in the target map. Inheritance 2218 * affects how the map will be shared with 2219 * child maps at the time of vmspace_fork. 2220 */ 2221int 2222vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2223 vm_inherit_t new_inheritance) 2224{ 2225 vm_map_entry_t entry; 2226 vm_map_entry_t temp_entry; 2227 2228 switch (new_inheritance) { 2229 case VM_INHERIT_NONE: 2230 case VM_INHERIT_COPY: 2231 case VM_INHERIT_SHARE: 2232 break; 2233 default: 2234 return (KERN_INVALID_ARGUMENT); 2235 } 2236 if (start == end) 2237 return (KERN_SUCCESS); 2238 vm_map_lock(map); 2239 VM_MAP_RANGE_CHECK(map, start, end); 2240 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2241 entry = temp_entry; 2242 vm_map_clip_start(map, entry, start); 2243 } else 2244 entry = temp_entry->next; 2245 while ((entry != &map->header) && (entry->start < end)) { 2246 vm_map_clip_end(map, entry, end); 2247 entry->inheritance = new_inheritance; 2248 vm_map_simplify_entry(map, entry); 2249 entry = entry->next; 2250 } 2251 vm_map_unlock(map); 2252 return (KERN_SUCCESS); 2253} 2254 2255/* 2256 * vm_map_unwire: 2257 * 2258 * Implements both kernel and user unwiring. 2259 */ 2260int 2261vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2262 int flags) 2263{ 2264 vm_map_entry_t entry, first_entry, tmp_entry; 2265 vm_offset_t saved_start; 2266 unsigned int last_timestamp; 2267 int rv; 2268 boolean_t need_wakeup, result, user_unwire; 2269 2270 if (start == end) 2271 return (KERN_SUCCESS); 2272 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2273 vm_map_lock(map); 2274 VM_MAP_RANGE_CHECK(map, start, end); 2275 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2276 if (flags & VM_MAP_WIRE_HOLESOK) 2277 first_entry = first_entry->next; 2278 else { 2279 vm_map_unlock(map); 2280 return (KERN_INVALID_ADDRESS); 2281 } 2282 } 2283 last_timestamp = map->timestamp; 2284 entry = first_entry; 2285 while (entry != &map->header && entry->start < end) { 2286 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2287 /* 2288 * We have not yet clipped the entry. 2289 */ 2290 saved_start = (start >= entry->start) ? start : 2291 entry->start; 2292 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2293 if (vm_map_unlock_and_wait(map, 0)) { 2294 /* 2295 * Allow interruption of user unwiring? 2296 */ 2297 } 2298 vm_map_lock(map); 2299 if (last_timestamp+1 != map->timestamp) { 2300 /* 2301 * Look again for the entry because the map was 2302 * modified while it was unlocked. 2303 * Specifically, the entry may have been 2304 * clipped, merged, or deleted. 2305 */ 2306 if (!vm_map_lookup_entry(map, saved_start, 2307 &tmp_entry)) { 2308 if (flags & VM_MAP_WIRE_HOLESOK) 2309 tmp_entry = tmp_entry->next; 2310 else { 2311 if (saved_start == start) { 2312 /* 2313 * First_entry has been deleted. 2314 */ 2315 vm_map_unlock(map); 2316 return (KERN_INVALID_ADDRESS); 2317 } 2318 end = saved_start; 2319 rv = KERN_INVALID_ADDRESS; 2320 goto done; 2321 } 2322 } 2323 if (entry == first_entry) 2324 first_entry = tmp_entry; 2325 else 2326 first_entry = NULL; 2327 entry = tmp_entry; 2328 } 2329 last_timestamp = map->timestamp; 2330 continue; 2331 } 2332 vm_map_clip_start(map, entry, start); 2333 vm_map_clip_end(map, entry, end); 2334 /* 2335 * Mark the entry in case the map lock is released. (See 2336 * above.) 2337 */ 2338 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2339 entry->wiring_thread == NULL, 2340 ("owned map entry %p", entry)); 2341 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2342 entry->wiring_thread = curthread; 2343 /* 2344 * Check the map for holes in the specified region. 2345 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2346 */ 2347 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2348 (entry->end < end && (entry->next == &map->header || 2349 entry->next->start > entry->end))) { 2350 end = entry->end; 2351 rv = KERN_INVALID_ADDRESS; 2352 goto done; 2353 } 2354 /* 2355 * If system unwiring, require that the entry is system wired. 2356 */ 2357 if (!user_unwire && 2358 vm_map_entry_system_wired_count(entry) == 0) { 2359 end = entry->end; 2360 rv = KERN_INVALID_ARGUMENT; 2361 goto done; 2362 } 2363 entry = entry->next; 2364 } 2365 rv = KERN_SUCCESS; 2366done: 2367 need_wakeup = FALSE; 2368 if (first_entry == NULL) { 2369 result = vm_map_lookup_entry(map, start, &first_entry); 2370 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2371 first_entry = first_entry->next; 2372 else 2373 KASSERT(result, ("vm_map_unwire: lookup failed")); 2374 } 2375 for (entry = first_entry; entry != &map->header && entry->start < end; 2376 entry = entry->next) { 2377 /* 2378 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2379 * space in the unwired region could have been mapped 2380 * while the map lock was dropped for draining 2381 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2382 * could be simultaneously wiring this new mapping 2383 * entry. Detect these cases and skip any entries 2384 * marked as in transition by us. 2385 */ 2386 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2387 entry->wiring_thread != curthread) { 2388 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2389 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2390 continue; 2391 } 2392 2393 if (rv == KERN_SUCCESS && (!user_unwire || 2394 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2395 if (user_unwire) 2396 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2397 entry->wired_count--; 2398 if (entry->wired_count == 0) { 2399 /* 2400 * Retain the map lock. 2401 */ 2402 vm_fault_unwire(map, entry->start, entry->end, 2403 entry->object.vm_object != NULL && 2404 (entry->object.vm_object->flags & 2405 OBJ_FICTITIOUS) != 0); 2406 } 2407 } 2408 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2409 ("vm_map_unwire: in-transition flag missing %p", entry)); 2410 KASSERT(entry->wiring_thread == curthread, 2411 ("vm_map_unwire: alien wire %p", entry)); 2412 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2413 entry->wiring_thread = NULL; 2414 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2415 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2416 need_wakeup = TRUE; 2417 } 2418 vm_map_simplify_entry(map, entry); 2419 } 2420 vm_map_unlock(map); 2421 if (need_wakeup) 2422 vm_map_wakeup(map); 2423 return (rv); 2424} 2425 2426/* 2427 * vm_map_wire: 2428 * 2429 * Implements both kernel and user wiring. 2430 */ 2431int 2432vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2433 int flags) 2434{ 2435 vm_map_entry_t entry, first_entry, tmp_entry; 2436 vm_offset_t saved_end, saved_start; 2437 unsigned int last_timestamp; 2438 int rv; 2439 boolean_t fictitious, need_wakeup, result, user_wire; 2440 vm_prot_t prot; 2441 2442 if (start == end) 2443 return (KERN_SUCCESS); 2444 prot = 0; 2445 if (flags & VM_MAP_WIRE_WRITE) 2446 prot |= VM_PROT_WRITE; 2447 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2448 vm_map_lock(map); 2449 VM_MAP_RANGE_CHECK(map, start, end); 2450 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2451 if (flags & VM_MAP_WIRE_HOLESOK) 2452 first_entry = first_entry->next; 2453 else { 2454 vm_map_unlock(map); 2455 return (KERN_INVALID_ADDRESS); 2456 } 2457 } 2458 last_timestamp = map->timestamp; 2459 entry = first_entry; 2460 while (entry != &map->header && entry->start < end) { 2461 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2462 /* 2463 * We have not yet clipped the entry. 2464 */ 2465 saved_start = (start >= entry->start) ? start : 2466 entry->start; 2467 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2468 if (vm_map_unlock_and_wait(map, 0)) { 2469 /* 2470 * Allow interruption of user wiring? 2471 */ 2472 } 2473 vm_map_lock(map); 2474 if (last_timestamp + 1 != map->timestamp) { 2475 /* 2476 * Look again for the entry because the map was 2477 * modified while it was unlocked. 2478 * Specifically, the entry may have been 2479 * clipped, merged, or deleted. 2480 */ 2481 if (!vm_map_lookup_entry(map, saved_start, 2482 &tmp_entry)) { 2483 if (flags & VM_MAP_WIRE_HOLESOK) 2484 tmp_entry = tmp_entry->next; 2485 else { 2486 if (saved_start == start) { 2487 /* 2488 * first_entry has been deleted. 2489 */ 2490 vm_map_unlock(map); 2491 return (KERN_INVALID_ADDRESS); 2492 } 2493 end = saved_start; 2494 rv = KERN_INVALID_ADDRESS; 2495 goto done; 2496 } 2497 } 2498 if (entry == first_entry) 2499 first_entry = tmp_entry; 2500 else 2501 first_entry = NULL; 2502 entry = tmp_entry; 2503 } 2504 last_timestamp = map->timestamp; 2505 continue; 2506 } 2507 vm_map_clip_start(map, entry, start); 2508 vm_map_clip_end(map, entry, end); 2509 /* 2510 * Mark the entry in case the map lock is released. (See 2511 * above.) 2512 */ 2513 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2514 entry->wiring_thread == NULL, 2515 ("owned map entry %p", entry)); 2516 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2517 entry->wiring_thread = curthread; 2518 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2519 || (entry->protection & prot) != prot) { 2520 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2521 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2522 end = entry->end; 2523 rv = KERN_INVALID_ADDRESS; 2524 goto done; 2525 } 2526 goto next_entry; 2527 } 2528 if (entry->wired_count == 0) { 2529 entry->wired_count++; 2530 saved_start = entry->start; 2531 saved_end = entry->end; 2532 fictitious = entry->object.vm_object != NULL && 2533 (entry->object.vm_object->flags & 2534 OBJ_FICTITIOUS) != 0; 2535 /* 2536 * Release the map lock, relying on the in-transition 2537 * mark. Mark the map busy for fork. 2538 */ 2539 vm_map_busy(map); 2540 vm_map_unlock(map); 2541 rv = vm_fault_wire(map, saved_start, saved_end, 2542 fictitious); 2543 vm_map_lock(map); 2544 vm_map_unbusy(map); 2545 if (last_timestamp + 1 != map->timestamp) { 2546 /* 2547 * Look again for the entry because the map was 2548 * modified while it was unlocked. The entry 2549 * may have been clipped, but NOT merged or 2550 * deleted. 2551 */ 2552 result = vm_map_lookup_entry(map, saved_start, 2553 &tmp_entry); 2554 KASSERT(result, ("vm_map_wire: lookup failed")); 2555 if (entry == first_entry) 2556 first_entry = tmp_entry; 2557 else 2558 first_entry = NULL; 2559 entry = tmp_entry; 2560 while (entry->end < saved_end) { 2561 if (rv != KERN_SUCCESS) { 2562 KASSERT(entry->wired_count == 1, 2563 ("vm_map_wire: bad count")); 2564 entry->wired_count = -1; 2565 } 2566 entry = entry->next; 2567 } 2568 } 2569 last_timestamp = map->timestamp; 2570 if (rv != KERN_SUCCESS) { 2571 KASSERT(entry->wired_count == 1, 2572 ("vm_map_wire: bad count")); 2573 /* 2574 * Assign an out-of-range value to represent 2575 * the failure to wire this entry. 2576 */ 2577 entry->wired_count = -1; 2578 end = entry->end; 2579 goto done; 2580 } 2581 } else if (!user_wire || 2582 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2583 entry->wired_count++; 2584 } 2585 /* 2586 * Check the map for holes in the specified region. 2587 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2588 */ 2589 next_entry: 2590 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2591 (entry->end < end && (entry->next == &map->header || 2592 entry->next->start > entry->end))) { 2593 end = entry->end; 2594 rv = KERN_INVALID_ADDRESS; 2595 goto done; 2596 } 2597 entry = entry->next; 2598 } 2599 rv = KERN_SUCCESS; 2600done: 2601 need_wakeup = FALSE; 2602 if (first_entry == NULL) { 2603 result = vm_map_lookup_entry(map, start, &first_entry); 2604 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2605 first_entry = first_entry->next; 2606 else 2607 KASSERT(result, ("vm_map_wire: lookup failed")); 2608 } 2609 for (entry = first_entry; entry != &map->header && entry->start < end; 2610 entry = entry->next) { 2611 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2612 goto next_entry_done; 2613 2614 /* 2615 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2616 * space in the unwired region could have been mapped 2617 * while the map lock was dropped for faulting in the 2618 * pages or draining MAP_ENTRY_IN_TRANSITION. 2619 * Moreover, another thread could be simultaneously 2620 * wiring this new mapping entry. Detect these cases 2621 * and skip any entries marked as in transition by us. 2622 */ 2623 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2624 entry->wiring_thread != curthread) { 2625 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2626 ("vm_map_wire: !HOLESOK and new/changed entry")); 2627 continue; 2628 } 2629 2630 if (rv == KERN_SUCCESS) { 2631 if (user_wire) 2632 entry->eflags |= MAP_ENTRY_USER_WIRED; 2633 } else if (entry->wired_count == -1) { 2634 /* 2635 * Wiring failed on this entry. Thus, unwiring is 2636 * unnecessary. 2637 */ 2638 entry->wired_count = 0; 2639 } else { 2640 if (!user_wire || 2641 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2642 entry->wired_count--; 2643 if (entry->wired_count == 0) { 2644 /* 2645 * Retain the map lock. 2646 */ 2647 vm_fault_unwire(map, entry->start, entry->end, 2648 entry->object.vm_object != NULL && 2649 (entry->object.vm_object->flags & 2650 OBJ_FICTITIOUS) != 0); 2651 } 2652 } 2653 next_entry_done: 2654 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2655 ("vm_map_wire: in-transition flag missing %p", entry)); 2656 KASSERT(entry->wiring_thread == curthread, 2657 ("vm_map_wire: alien wire %p", entry)); 2658 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2659 MAP_ENTRY_WIRE_SKIPPED); 2660 entry->wiring_thread = NULL; 2661 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2662 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2663 need_wakeup = TRUE; 2664 } 2665 vm_map_simplify_entry(map, entry); 2666 } 2667 vm_map_unlock(map); 2668 if (need_wakeup) 2669 vm_map_wakeup(map); 2670 return (rv); 2671} 2672 2673/* 2674 * vm_map_sync 2675 * 2676 * Push any dirty cached pages in the address range to their pager. 2677 * If syncio is TRUE, dirty pages are written synchronously. 2678 * If invalidate is TRUE, any cached pages are freed as well. 2679 * 2680 * If the size of the region from start to end is zero, we are 2681 * supposed to flush all modified pages within the region containing 2682 * start. Unfortunately, a region can be split or coalesced with 2683 * neighboring regions, making it difficult to determine what the 2684 * original region was. Therefore, we approximate this requirement by 2685 * flushing the current region containing start. 2686 * 2687 * Returns an error if any part of the specified range is not mapped. 2688 */ 2689int 2690vm_map_sync( 2691 vm_map_t map, 2692 vm_offset_t start, 2693 vm_offset_t end, 2694 boolean_t syncio, 2695 boolean_t invalidate) 2696{ 2697 vm_map_entry_t current; 2698 vm_map_entry_t entry; 2699 vm_size_t size; 2700 vm_object_t object; 2701 vm_ooffset_t offset; 2702 unsigned int last_timestamp; 2703 boolean_t failed; 2704 2705 vm_map_lock_read(map); 2706 VM_MAP_RANGE_CHECK(map, start, end); 2707 if (!vm_map_lookup_entry(map, start, &entry)) { 2708 vm_map_unlock_read(map); 2709 return (KERN_INVALID_ADDRESS); 2710 } else if (start == end) { 2711 start = entry->start; 2712 end = entry->end; 2713 } 2714 /* 2715 * Make a first pass to check for user-wired memory and holes. 2716 */ 2717 for (current = entry; current != &map->header && current->start < end; 2718 current = current->next) { 2719 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2720 vm_map_unlock_read(map); 2721 return (KERN_INVALID_ARGUMENT); 2722 } 2723 if (end > current->end && 2724 (current->next == &map->header || 2725 current->end != current->next->start)) { 2726 vm_map_unlock_read(map); 2727 return (KERN_INVALID_ADDRESS); 2728 } 2729 } 2730 2731 if (invalidate) 2732 pmap_remove(map->pmap, start, end); 2733 failed = FALSE; 2734 2735 /* 2736 * Make a second pass, cleaning/uncaching pages from the indicated 2737 * objects as we go. 2738 */ 2739 for (current = entry; current != &map->header && current->start < end;) { 2740 offset = current->offset + (start - current->start); 2741 size = (end <= current->end ? end : current->end) - start; 2742 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2743 vm_map_t smap; 2744 vm_map_entry_t tentry; 2745 vm_size_t tsize; 2746 2747 smap = current->object.sub_map; 2748 vm_map_lock_read(smap); 2749 (void) vm_map_lookup_entry(smap, offset, &tentry); 2750 tsize = tentry->end - offset; 2751 if (tsize < size) 2752 size = tsize; 2753 object = tentry->object.vm_object; 2754 offset = tentry->offset + (offset - tentry->start); 2755 vm_map_unlock_read(smap); 2756 } else { 2757 object = current->object.vm_object; 2758 } 2759 vm_object_reference(object); 2760 last_timestamp = map->timestamp; 2761 vm_map_unlock_read(map); 2762 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2763 failed = TRUE; 2764 start += size; 2765 vm_object_deallocate(object); 2766 vm_map_lock_read(map); 2767 if (last_timestamp == map->timestamp || 2768 !vm_map_lookup_entry(map, start, ¤t)) 2769 current = current->next; 2770 } 2771 2772 vm_map_unlock_read(map); 2773 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2774} 2775 2776/* 2777 * vm_map_entry_unwire: [ internal use only ] 2778 * 2779 * Make the region specified by this entry pageable. 2780 * 2781 * The map in question should be locked. 2782 * [This is the reason for this routine's existence.] 2783 */ 2784static void 2785vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2786{ 2787 vm_fault_unwire(map, entry->start, entry->end, 2788 entry->object.vm_object != NULL && 2789 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2790 entry->wired_count = 0; 2791} 2792 2793static void 2794vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2795{ 2796 2797 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2798 vm_object_deallocate(entry->object.vm_object); 2799 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2800} 2801 2802/* 2803 * vm_map_entry_delete: [ internal use only ] 2804 * 2805 * Deallocate the given entry from the target map. 2806 */ 2807static void 2808vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2809{ 2810 vm_object_t object; 2811 vm_pindex_t offidxstart, offidxend, count, size1; 2812 vm_ooffset_t size; 2813 2814 vm_map_entry_unlink(map, entry); 2815 object = entry->object.vm_object; 2816 size = entry->end - entry->start; 2817 map->size -= size; 2818 2819 if (entry->cred != NULL) { 2820 swap_release_by_cred(size, entry->cred); 2821 crfree(entry->cred); 2822 } 2823 2824 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2825 (object != NULL)) { 2826 KASSERT(entry->cred == NULL || object->cred == NULL || 2827 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2828 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2829 count = OFF_TO_IDX(size); 2830 offidxstart = OFF_TO_IDX(entry->offset); 2831 offidxend = offidxstart + count; 2832 VM_OBJECT_WLOCK(object); 2833 if (object->ref_count != 1 && 2834 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2835 object == kernel_object || object == kmem_object)) { 2836 vm_object_collapse(object); 2837 2838 /* 2839 * The option OBJPR_NOTMAPPED can be passed here 2840 * because vm_map_delete() already performed 2841 * pmap_remove() on the only mapping to this range 2842 * of pages. 2843 */ 2844 vm_object_page_remove(object, offidxstart, offidxend, 2845 OBJPR_NOTMAPPED); 2846 if (object->type == OBJT_SWAP) 2847 swap_pager_freespace(object, offidxstart, count); 2848 if (offidxend >= object->size && 2849 offidxstart < object->size) { 2850 size1 = object->size; 2851 object->size = offidxstart; 2852 if (object->cred != NULL) { 2853 size1 -= object->size; 2854 KASSERT(object->charge >= ptoa(size1), 2855 ("vm_map_entry_delete: object->charge < 0")); 2856 swap_release_by_cred(ptoa(size1), object->cred); 2857 object->charge -= ptoa(size1); 2858 } 2859 } 2860 } 2861 VM_OBJECT_WUNLOCK(object); 2862 } else 2863 entry->object.vm_object = NULL; 2864 if (map->system_map) 2865 vm_map_entry_deallocate(entry, TRUE); 2866 else { 2867 entry->next = curthread->td_map_def_user; 2868 curthread->td_map_def_user = entry; 2869 } 2870} 2871 2872/* 2873 * vm_map_delete: [ internal use only ] 2874 * 2875 * Deallocates the given address range from the target 2876 * map. 2877 */ 2878int 2879vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2880{ 2881 vm_map_entry_t entry; 2882 vm_map_entry_t first_entry; 2883 2884 VM_MAP_ASSERT_LOCKED(map); 2885 if (start == end) 2886 return (KERN_SUCCESS); 2887 2888 /* 2889 * Find the start of the region, and clip it 2890 */ 2891 if (!vm_map_lookup_entry(map, start, &first_entry)) 2892 entry = first_entry->next; 2893 else { 2894 entry = first_entry; 2895 vm_map_clip_start(map, entry, start); 2896 } 2897 2898 /* 2899 * Step through all entries in this region 2900 */ 2901 while ((entry != &map->header) && (entry->start < end)) { 2902 vm_map_entry_t next; 2903 2904 /* 2905 * Wait for wiring or unwiring of an entry to complete. 2906 * Also wait for any system wirings to disappear on 2907 * user maps. 2908 */ 2909 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2910 (vm_map_pmap(map) != kernel_pmap && 2911 vm_map_entry_system_wired_count(entry) != 0)) { 2912 unsigned int last_timestamp; 2913 vm_offset_t saved_start; 2914 vm_map_entry_t tmp_entry; 2915 2916 saved_start = entry->start; 2917 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2918 last_timestamp = map->timestamp; 2919 (void) vm_map_unlock_and_wait(map, 0); 2920 vm_map_lock(map); 2921 if (last_timestamp + 1 != map->timestamp) { 2922 /* 2923 * Look again for the entry because the map was 2924 * modified while it was unlocked. 2925 * Specifically, the entry may have been 2926 * clipped, merged, or deleted. 2927 */ 2928 if (!vm_map_lookup_entry(map, saved_start, 2929 &tmp_entry)) 2930 entry = tmp_entry->next; 2931 else { 2932 entry = tmp_entry; 2933 vm_map_clip_start(map, entry, 2934 saved_start); 2935 } 2936 } 2937 continue; 2938 } 2939 vm_map_clip_end(map, entry, end); 2940 2941 next = entry->next; 2942 2943 /* 2944 * Unwire before removing addresses from the pmap; otherwise, 2945 * unwiring will put the entries back in the pmap. 2946 */ 2947 if (entry->wired_count != 0) { 2948 vm_map_entry_unwire(map, entry); 2949 } 2950 2951 pmap_remove(map->pmap, entry->start, entry->end); 2952 2953 /* 2954 * Delete the entry only after removing all pmap 2955 * entries pointing to its pages. (Otherwise, its 2956 * page frames may be reallocated, and any modify bits 2957 * will be set in the wrong object!) 2958 */ 2959 vm_map_entry_delete(map, entry); 2960 entry = next; 2961 } 2962 return (KERN_SUCCESS); 2963} 2964 2965/* 2966 * vm_map_remove: 2967 * 2968 * Remove the given address range from the target map. 2969 * This is the exported form of vm_map_delete. 2970 */ 2971int 2972vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2973{ 2974 int result; 2975 2976 vm_map_lock(map); 2977 VM_MAP_RANGE_CHECK(map, start, end); 2978 result = vm_map_delete(map, start, end); 2979 vm_map_unlock(map); 2980 return (result); 2981} 2982 2983/* 2984 * vm_map_check_protection: 2985 * 2986 * Assert that the target map allows the specified privilege on the 2987 * entire address region given. The entire region must be allocated. 2988 * 2989 * WARNING! This code does not and should not check whether the 2990 * contents of the region is accessible. For example a smaller file 2991 * might be mapped into a larger address space. 2992 * 2993 * NOTE! This code is also called by munmap(). 2994 * 2995 * The map must be locked. A read lock is sufficient. 2996 */ 2997boolean_t 2998vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2999 vm_prot_t protection) 3000{ 3001 vm_map_entry_t entry; 3002 vm_map_entry_t tmp_entry; 3003 3004 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3005 return (FALSE); 3006 entry = tmp_entry; 3007 3008 while (start < end) { 3009 if (entry == &map->header) 3010 return (FALSE); 3011 /* 3012 * No holes allowed! 3013 */ 3014 if (start < entry->start) 3015 return (FALSE); 3016 /* 3017 * Check protection associated with entry. 3018 */ 3019 if ((entry->protection & protection) != protection) 3020 return (FALSE); 3021 /* go to next entry */ 3022 start = entry->end; 3023 entry = entry->next; 3024 } 3025 return (TRUE); 3026} 3027 3028/* 3029 * vm_map_copy_entry: 3030 * 3031 * Copies the contents of the source entry to the destination 3032 * entry. The entries *must* be aligned properly. 3033 */ 3034static void 3035vm_map_copy_entry( 3036 vm_map_t src_map, 3037 vm_map_t dst_map, 3038 vm_map_entry_t src_entry, 3039 vm_map_entry_t dst_entry, 3040 vm_ooffset_t *fork_charge) 3041{ 3042 vm_object_t src_object; 3043 vm_map_entry_t fake_entry; 3044 vm_offset_t size; 3045 struct ucred *cred; 3046 int charged; 3047 3048 VM_MAP_ASSERT_LOCKED(dst_map); 3049 3050 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3051 return; 3052 3053 if (src_entry->wired_count == 0 || 3054 (src_entry->protection & VM_PROT_WRITE) == 0) { 3055 /* 3056 * If the source entry is marked needs_copy, it is already 3057 * write-protected. 3058 */ 3059 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3060 (src_entry->protection & VM_PROT_WRITE) != 0) { 3061 pmap_protect(src_map->pmap, 3062 src_entry->start, 3063 src_entry->end, 3064 src_entry->protection & ~VM_PROT_WRITE); 3065 } 3066 3067 /* 3068 * Make a copy of the object. 3069 */ 3070 size = src_entry->end - src_entry->start; 3071 if ((src_object = src_entry->object.vm_object) != NULL) { 3072 VM_OBJECT_WLOCK(src_object); 3073 charged = ENTRY_CHARGED(src_entry); 3074 if ((src_object->handle == NULL) && 3075 (src_object->type == OBJT_DEFAULT || 3076 src_object->type == OBJT_SWAP)) { 3077 vm_object_collapse(src_object); 3078 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3079 vm_object_split(src_entry); 3080 src_object = src_entry->object.vm_object; 3081 } 3082 } 3083 vm_object_reference_locked(src_object); 3084 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3085 if (src_entry->cred != NULL && 3086 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3087 KASSERT(src_object->cred == NULL, 3088 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3089 src_object)); 3090 src_object->cred = src_entry->cred; 3091 src_object->charge = size; 3092 } 3093 VM_OBJECT_WUNLOCK(src_object); 3094 dst_entry->object.vm_object = src_object; 3095 if (charged) { 3096 cred = curthread->td_ucred; 3097 crhold(cred); 3098 dst_entry->cred = cred; 3099 *fork_charge += size; 3100 if (!(src_entry->eflags & 3101 MAP_ENTRY_NEEDS_COPY)) { 3102 crhold(cred); 3103 src_entry->cred = cred; 3104 *fork_charge += size; 3105 } 3106 } 3107 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3108 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3109 dst_entry->offset = src_entry->offset; 3110 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3111 /* 3112 * MAP_ENTRY_VN_WRITECNT cannot 3113 * indicate write reference from 3114 * src_entry, since the entry is 3115 * marked as needs copy. Allocate a 3116 * fake entry that is used to 3117 * decrement object->un_pager.vnp.writecount 3118 * at the appropriate time. Attach 3119 * fake_entry to the deferred list. 3120 */ 3121 fake_entry = vm_map_entry_create(dst_map); 3122 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3123 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3124 vm_object_reference(src_object); 3125 fake_entry->object.vm_object = src_object; 3126 fake_entry->start = src_entry->start; 3127 fake_entry->end = src_entry->end; 3128 fake_entry->next = curthread->td_map_def_user; 3129 curthread->td_map_def_user = fake_entry; 3130 } 3131 } else { 3132 dst_entry->object.vm_object = NULL; 3133 dst_entry->offset = 0; 3134 if (src_entry->cred != NULL) { 3135 dst_entry->cred = curthread->td_ucred; 3136 crhold(dst_entry->cred); 3137 *fork_charge += size; 3138 } 3139 } 3140 3141 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3142 dst_entry->end - dst_entry->start, src_entry->start); 3143 } else { 3144 /* 3145 * We don't want to make writeable wired pages copy-on-write. 3146 * Immediately copy these pages into the new map by simulating 3147 * page faults. The new pages are pageable. 3148 */ 3149 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3150 fork_charge); 3151 } 3152} 3153 3154/* 3155 * vmspace_map_entry_forked: 3156 * Update the newly-forked vmspace each time a map entry is inherited 3157 * or copied. The values for vm_dsize and vm_tsize are approximate 3158 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3159 */ 3160static void 3161vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3162 vm_map_entry_t entry) 3163{ 3164 vm_size_t entrysize; 3165 vm_offset_t newend; 3166 3167 entrysize = entry->end - entry->start; 3168 vm2->vm_map.size += entrysize; 3169 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3170 vm2->vm_ssize += btoc(entrysize); 3171 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3172 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3173 newend = MIN(entry->end, 3174 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3175 vm2->vm_dsize += btoc(newend - entry->start); 3176 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3177 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3178 newend = MIN(entry->end, 3179 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3180 vm2->vm_tsize += btoc(newend - entry->start); 3181 } 3182} 3183 3184/* 3185 * vmspace_fork: 3186 * Create a new process vmspace structure and vm_map 3187 * based on those of an existing process. The new map 3188 * is based on the old map, according to the inheritance 3189 * values on the regions in that map. 3190 * 3191 * XXX It might be worth coalescing the entries added to the new vmspace. 3192 * 3193 * The source map must not be locked. 3194 */ 3195struct vmspace * 3196vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3197{ 3198 struct vmspace *vm2; 3199 vm_map_t new_map, old_map; 3200 vm_map_entry_t new_entry, old_entry; 3201 vm_object_t object; 3202 int locked; 3203 3204 old_map = &vm1->vm_map; 3205 /* Copy immutable fields of vm1 to vm2. */ 3206 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3207 if (vm2 == NULL) 3208 return (NULL); 3209 vm2->vm_taddr = vm1->vm_taddr; 3210 vm2->vm_daddr = vm1->vm_daddr; 3211 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3212 vm_map_lock(old_map); 3213 if (old_map->busy) 3214 vm_map_wait_busy(old_map); 3215 new_map = &vm2->vm_map; 3216 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3217 KASSERT(locked, ("vmspace_fork: lock failed")); 3218 3219 old_entry = old_map->header.next; 3220 3221 while (old_entry != &old_map->header) { 3222 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3223 panic("vm_map_fork: encountered a submap"); 3224 3225 switch (old_entry->inheritance) { 3226 case VM_INHERIT_NONE: 3227 break; 3228 3229 case VM_INHERIT_SHARE: 3230 /* 3231 * Clone the entry, creating the shared object if necessary. 3232 */ 3233 object = old_entry->object.vm_object; 3234 if (object == NULL) { 3235 object = vm_object_allocate(OBJT_DEFAULT, 3236 atop(old_entry->end - old_entry->start)); 3237 old_entry->object.vm_object = object; 3238 old_entry->offset = 0; 3239 if (old_entry->cred != NULL) { 3240 object->cred = old_entry->cred; 3241 object->charge = old_entry->end - 3242 old_entry->start; 3243 old_entry->cred = NULL; 3244 } 3245 } 3246 3247 /* 3248 * Add the reference before calling vm_object_shadow 3249 * to insure that a shadow object is created. 3250 */ 3251 vm_object_reference(object); 3252 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3253 vm_object_shadow(&old_entry->object.vm_object, 3254 &old_entry->offset, 3255 old_entry->end - old_entry->start); 3256 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3257 /* Transfer the second reference too. */ 3258 vm_object_reference( 3259 old_entry->object.vm_object); 3260 3261 /* 3262 * As in vm_map_simplify_entry(), the 3263 * vnode lock will not be acquired in 3264 * this call to vm_object_deallocate(). 3265 */ 3266 vm_object_deallocate(object); 3267 object = old_entry->object.vm_object; 3268 } 3269 VM_OBJECT_WLOCK(object); 3270 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3271 if (old_entry->cred != NULL) { 3272 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3273 object->cred = old_entry->cred; 3274 object->charge = old_entry->end - old_entry->start; 3275 old_entry->cred = NULL; 3276 } 3277 3278 /* 3279 * Assert the correct state of the vnode 3280 * v_writecount while the object is locked, to 3281 * not relock it later for the assertion 3282 * correctness. 3283 */ 3284 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3285 object->type == OBJT_VNODE) { 3286 KASSERT(((struct vnode *)object->handle)-> 3287 v_writecount > 0, 3288 ("vmspace_fork: v_writecount %p", object)); 3289 KASSERT(object->un_pager.vnp.writemappings > 0, 3290 ("vmspace_fork: vnp.writecount %p", 3291 object)); 3292 } 3293 VM_OBJECT_WUNLOCK(object); 3294 3295 /* 3296 * Clone the entry, referencing the shared object. 3297 */ 3298 new_entry = vm_map_entry_create(new_map); 3299 *new_entry = *old_entry; 3300 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3301 MAP_ENTRY_IN_TRANSITION); 3302 new_entry->wiring_thread = NULL; 3303 new_entry->wired_count = 0; 3304 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3305 vnode_pager_update_writecount(object, 3306 new_entry->start, new_entry->end); 3307 } 3308 3309 /* 3310 * Insert the entry into the new map -- we know we're 3311 * inserting at the end of the new map. 3312 */ 3313 vm_map_entry_link(new_map, new_map->header.prev, 3314 new_entry); 3315 vmspace_map_entry_forked(vm1, vm2, new_entry); 3316 3317 /* 3318 * Update the physical map 3319 */ 3320 pmap_copy(new_map->pmap, old_map->pmap, 3321 new_entry->start, 3322 (old_entry->end - old_entry->start), 3323 old_entry->start); 3324 break; 3325 3326 case VM_INHERIT_COPY: 3327 /* 3328 * Clone the entry and link into the map. 3329 */ 3330 new_entry = vm_map_entry_create(new_map); 3331 *new_entry = *old_entry; 3332 /* 3333 * Copied entry is COW over the old object. 3334 */ 3335 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3336 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3337 new_entry->wiring_thread = NULL; 3338 new_entry->wired_count = 0; 3339 new_entry->object.vm_object = NULL; 3340 new_entry->cred = NULL; 3341 vm_map_entry_link(new_map, new_map->header.prev, 3342 new_entry); 3343 vmspace_map_entry_forked(vm1, vm2, new_entry); 3344 vm_map_copy_entry(old_map, new_map, old_entry, 3345 new_entry, fork_charge); 3346 break; 3347 } 3348 old_entry = old_entry->next; 3349 } 3350 /* 3351 * Use inlined vm_map_unlock() to postpone handling the deferred 3352 * map entries, which cannot be done until both old_map and 3353 * new_map locks are released. 3354 */ 3355 sx_xunlock(&old_map->lock); 3356 sx_xunlock(&new_map->lock); 3357 vm_map_process_deferred(); 3358 3359 return (vm2); 3360} 3361 3362int 3363vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3364 vm_prot_t prot, vm_prot_t max, int cow) 3365{ 3366 vm_size_t growsize, init_ssize; 3367 rlim_t lmemlim, vmemlim; 3368 int rv; 3369 3370 growsize = sgrowsiz; 3371 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3372 vm_map_lock(map); 3373 PROC_LOCK(curproc); 3374 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3375 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3376 PROC_UNLOCK(curproc); 3377 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3378 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3379 rv = KERN_NO_SPACE; 3380 goto out; 3381 } 3382 } 3383 /* If we would blow our VMEM resource limit, no go */ 3384 if (map->size + init_ssize > vmemlim) { 3385 rv = KERN_NO_SPACE; 3386 goto out; 3387 } 3388 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3389 max, cow); 3390out: 3391 vm_map_unlock(map); 3392 return (rv); 3393} 3394 3395static int 3396vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3397 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3398{ 3399 vm_map_entry_t new_entry, prev_entry; 3400 vm_offset_t bot, top; 3401 vm_size_t init_ssize; 3402 int orient, rv; 3403 3404 /* 3405 * The stack orientation is piggybacked with the cow argument. 3406 * Extract it into orient and mask the cow argument so that we 3407 * don't pass it around further. 3408 * NOTE: We explicitly allow bi-directional stacks. 3409 */ 3410 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3411 KASSERT(orient != 0, ("No stack grow direction")); 3412 3413 if (addrbos < vm_map_min(map) || 3414 addrbos > vm_map_max(map) || 3415 addrbos + max_ssize < addrbos) 3416 return (KERN_NO_SPACE); 3417 3418 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3419 3420 /* If addr is already mapped, no go */ 3421 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3422 return (KERN_NO_SPACE); 3423 3424 /* 3425 * If we can't accomodate max_ssize in the current mapping, no go. 3426 * However, we need to be aware that subsequent user mappings might 3427 * map into the space we have reserved for stack, and currently this 3428 * space is not protected. 3429 * 3430 * Hopefully we will at least detect this condition when we try to 3431 * grow the stack. 3432 */ 3433 if ((prev_entry->next != &map->header) && 3434 (prev_entry->next->start < addrbos + max_ssize)) 3435 return (KERN_NO_SPACE); 3436 3437 /* 3438 * We initially map a stack of only init_ssize. We will grow as 3439 * needed later. Depending on the orientation of the stack (i.e. 3440 * the grow direction) we either map at the top of the range, the 3441 * bottom of the range or in the middle. 3442 * 3443 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3444 * and cow to be 0. Possibly we should eliminate these as input 3445 * parameters, and just pass these values here in the insert call. 3446 */ 3447 if (orient == MAP_STACK_GROWS_DOWN) 3448 bot = addrbos + max_ssize - init_ssize; 3449 else if (orient == MAP_STACK_GROWS_UP) 3450 bot = addrbos; 3451 else 3452 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3453 top = bot + init_ssize; 3454 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3455 3456 /* Now set the avail_ssize amount. */ 3457 if (rv == KERN_SUCCESS) { 3458 if (prev_entry != &map->header) 3459 vm_map_clip_end(map, prev_entry, bot); 3460 new_entry = prev_entry->next; 3461 if (new_entry->end != top || new_entry->start != bot) 3462 panic("Bad entry start/end for new stack entry"); 3463 3464 new_entry->avail_ssize = max_ssize - init_ssize; 3465 if (orient & MAP_STACK_GROWS_DOWN) 3466 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3467 if (orient & MAP_STACK_GROWS_UP) 3468 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3469 } 3470 3471 return (rv); 3472} 3473 3474static int stack_guard_page = 0; 3475TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3476SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3477 &stack_guard_page, 0, 3478 "Insert stack guard page ahead of the growable segments."); 3479 3480/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3481 * desired address is already mapped, or if we successfully grow 3482 * the stack. Also returns KERN_SUCCESS if addr is outside the 3483 * stack range (this is strange, but preserves compatibility with 3484 * the grow function in vm_machdep.c). 3485 */ 3486int 3487vm_map_growstack(struct proc *p, vm_offset_t addr) 3488{ 3489 vm_map_entry_t next_entry, prev_entry; 3490 vm_map_entry_t new_entry, stack_entry; 3491 struct vmspace *vm = p->p_vmspace; 3492 vm_map_t map = &vm->vm_map; 3493 vm_offset_t end; 3494 vm_size_t growsize; 3495 size_t grow_amount, max_grow; 3496 rlim_t lmemlim, stacklim, vmemlim; 3497 int is_procstack, rv; 3498 struct ucred *cred; 3499#ifdef notyet 3500 uint64_t limit; 3501#endif 3502#ifdef RACCT 3503 int error; 3504#endif 3505 3506Retry: 3507 PROC_LOCK(p); 3508 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3509 stacklim = lim_cur(p, RLIMIT_STACK); 3510 vmemlim = lim_cur(p, RLIMIT_VMEM); 3511 PROC_UNLOCK(p); 3512 3513 vm_map_lock_read(map); 3514 3515 /* If addr is already in the entry range, no need to grow.*/ 3516 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3517 vm_map_unlock_read(map); 3518 return (KERN_SUCCESS); 3519 } 3520 3521 next_entry = prev_entry->next; 3522 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3523 /* 3524 * This entry does not grow upwards. Since the address lies 3525 * beyond this entry, the next entry (if one exists) has to 3526 * be a downward growable entry. The entry list header is 3527 * never a growable entry, so it suffices to check the flags. 3528 */ 3529 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3530 vm_map_unlock_read(map); 3531 return (KERN_SUCCESS); 3532 } 3533 stack_entry = next_entry; 3534 } else { 3535 /* 3536 * This entry grows upward. If the next entry does not at 3537 * least grow downwards, this is the entry we need to grow. 3538 * otherwise we have two possible choices and we have to 3539 * select one. 3540 */ 3541 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3542 /* 3543 * We have two choices; grow the entry closest to 3544 * the address to minimize the amount of growth. 3545 */ 3546 if (addr - prev_entry->end <= next_entry->start - addr) 3547 stack_entry = prev_entry; 3548 else 3549 stack_entry = next_entry; 3550 } else 3551 stack_entry = prev_entry; 3552 } 3553 3554 if (stack_entry == next_entry) { 3555 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3556 KASSERT(addr < stack_entry->start, ("foo")); 3557 end = (prev_entry != &map->header) ? prev_entry->end : 3558 stack_entry->start - stack_entry->avail_ssize; 3559 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3560 max_grow = stack_entry->start - end; 3561 } else { 3562 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3563 KASSERT(addr >= stack_entry->end, ("foo")); 3564 end = (next_entry != &map->header) ? next_entry->start : 3565 stack_entry->end + stack_entry->avail_ssize; 3566 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3567 max_grow = end - stack_entry->end; 3568 } 3569 3570 if (grow_amount > stack_entry->avail_ssize) { 3571 vm_map_unlock_read(map); 3572 return (KERN_NO_SPACE); 3573 } 3574 3575 /* 3576 * If there is no longer enough space between the entries nogo, and 3577 * adjust the available space. Note: this should only happen if the 3578 * user has mapped into the stack area after the stack was created, 3579 * and is probably an error. 3580 * 3581 * This also effectively destroys any guard page the user might have 3582 * intended by limiting the stack size. 3583 */ 3584 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3585 if (vm_map_lock_upgrade(map)) 3586 goto Retry; 3587 3588 stack_entry->avail_ssize = max_grow; 3589 3590 vm_map_unlock(map); 3591 return (KERN_NO_SPACE); 3592 } 3593 3594 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3595 3596 /* 3597 * If this is the main process stack, see if we're over the stack 3598 * limit. 3599 */ 3600 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3601 vm_map_unlock_read(map); 3602 return (KERN_NO_SPACE); 3603 } 3604#ifdef RACCT 3605 PROC_LOCK(p); 3606 if (is_procstack && 3607 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3608 PROC_UNLOCK(p); 3609 vm_map_unlock_read(map); 3610 return (KERN_NO_SPACE); 3611 } 3612 PROC_UNLOCK(p); 3613#endif 3614 3615 /* Round up the grow amount modulo sgrowsiz */ 3616 growsize = sgrowsiz; 3617 grow_amount = roundup(grow_amount, growsize); 3618 if (grow_amount > stack_entry->avail_ssize) 3619 grow_amount = stack_entry->avail_ssize; 3620 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3621 grow_amount = trunc_page((vm_size_t)stacklim) - 3622 ctob(vm->vm_ssize); 3623 } 3624#ifdef notyet 3625 PROC_LOCK(p); 3626 limit = racct_get_available(p, RACCT_STACK); 3627 PROC_UNLOCK(p); 3628 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3629 grow_amount = limit - ctob(vm->vm_ssize); 3630#endif 3631 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3632 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3633 vm_map_unlock_read(map); 3634 rv = KERN_NO_SPACE; 3635 goto out; 3636 } 3637#ifdef RACCT 3638 PROC_LOCK(p); 3639 if (racct_set(p, RACCT_MEMLOCK, 3640 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3641 PROC_UNLOCK(p); 3642 vm_map_unlock_read(map); 3643 rv = KERN_NO_SPACE; 3644 goto out; 3645 } 3646 PROC_UNLOCK(p); 3647#endif 3648 } 3649 /* If we would blow our VMEM resource limit, no go */ 3650 if (map->size + grow_amount > vmemlim) { 3651 vm_map_unlock_read(map); 3652 rv = KERN_NO_SPACE; 3653 goto out; 3654 } 3655#ifdef RACCT 3656 PROC_LOCK(p); 3657 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3658 PROC_UNLOCK(p); 3659 vm_map_unlock_read(map); 3660 rv = KERN_NO_SPACE; 3661 goto out; 3662 } 3663 PROC_UNLOCK(p); 3664#endif 3665 3666 if (vm_map_lock_upgrade(map)) 3667 goto Retry; 3668 3669 if (stack_entry == next_entry) { 3670 /* 3671 * Growing downward. 3672 */ 3673 /* Get the preliminary new entry start value */ 3674 addr = stack_entry->start - grow_amount; 3675 3676 /* 3677 * If this puts us into the previous entry, cut back our 3678 * growth to the available space. Also, see the note above. 3679 */ 3680 if (addr < end) { 3681 stack_entry->avail_ssize = max_grow; 3682 addr = end; 3683 if (stack_guard_page) 3684 addr += PAGE_SIZE; 3685 } 3686 3687 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3688 next_entry->protection, next_entry->max_protection, 0); 3689 3690 /* Adjust the available stack space by the amount we grew. */ 3691 if (rv == KERN_SUCCESS) { 3692 if (prev_entry != &map->header) 3693 vm_map_clip_end(map, prev_entry, addr); 3694 new_entry = prev_entry->next; 3695 KASSERT(new_entry == stack_entry->prev, ("foo")); 3696 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3697 KASSERT(new_entry->start == addr, ("foo")); 3698 grow_amount = new_entry->end - new_entry->start; 3699 new_entry->avail_ssize = stack_entry->avail_ssize - 3700 grow_amount; 3701 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3702 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3703 } 3704 } else { 3705 /* 3706 * Growing upward. 3707 */ 3708 addr = stack_entry->end + grow_amount; 3709 3710 /* 3711 * If this puts us into the next entry, cut back our growth 3712 * to the available space. Also, see the note above. 3713 */ 3714 if (addr > end) { 3715 stack_entry->avail_ssize = end - stack_entry->end; 3716 addr = end; 3717 if (stack_guard_page) 3718 addr -= PAGE_SIZE; 3719 } 3720 3721 grow_amount = addr - stack_entry->end; 3722 cred = stack_entry->cred; 3723 if (cred == NULL && stack_entry->object.vm_object != NULL) 3724 cred = stack_entry->object.vm_object->cred; 3725 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3726 rv = KERN_NO_SPACE; 3727 /* Grow the underlying object if applicable. */ 3728 else if (stack_entry->object.vm_object == NULL || 3729 vm_object_coalesce(stack_entry->object.vm_object, 3730 stack_entry->offset, 3731 (vm_size_t)(stack_entry->end - stack_entry->start), 3732 (vm_size_t)grow_amount, cred != NULL)) { 3733 map->size += (addr - stack_entry->end); 3734 /* Update the current entry. */ 3735 stack_entry->end = addr; 3736 stack_entry->avail_ssize -= grow_amount; 3737 vm_map_entry_resize_free(map, stack_entry); 3738 rv = KERN_SUCCESS; 3739 3740 if (next_entry != &map->header) 3741 vm_map_clip_start(map, next_entry, addr); 3742 } else 3743 rv = KERN_FAILURE; 3744 } 3745 3746 if (rv == KERN_SUCCESS && is_procstack) 3747 vm->vm_ssize += btoc(grow_amount); 3748 3749 vm_map_unlock(map); 3750 3751 /* 3752 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3753 */ 3754 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3755 vm_map_wire(map, 3756 (stack_entry == next_entry) ? addr : addr - grow_amount, 3757 (stack_entry == next_entry) ? stack_entry->start : addr, 3758 (p->p_flag & P_SYSTEM) 3759 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3760 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3761 } 3762 3763out: 3764#ifdef RACCT 3765 if (rv != KERN_SUCCESS) { 3766 PROC_LOCK(p); 3767 error = racct_set(p, RACCT_VMEM, map->size); 3768 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3769 if (!old_mlock) { 3770 error = racct_set(p, RACCT_MEMLOCK, 3771 ptoa(pmap_wired_count(map->pmap))); 3772 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3773 } 3774 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3775 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3776 PROC_UNLOCK(p); 3777 } 3778#endif 3779 3780 return (rv); 3781} 3782 3783/* 3784 * Unshare the specified VM space for exec. If other processes are 3785 * mapped to it, then create a new one. The new vmspace is null. 3786 */ 3787int 3788vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3789{ 3790 struct vmspace *oldvmspace = p->p_vmspace; 3791 struct vmspace *newvmspace; 3792 3793 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3794 ("vmspace_exec recursed")); 3795 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3796 if (newvmspace == NULL) 3797 return (ENOMEM); 3798 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3799 /* 3800 * This code is written like this for prototype purposes. The 3801 * goal is to avoid running down the vmspace here, but let the 3802 * other process's that are still using the vmspace to finally 3803 * run it down. Even though there is little or no chance of blocking 3804 * here, it is a good idea to keep this form for future mods. 3805 */ 3806 PROC_VMSPACE_LOCK(p); 3807 p->p_vmspace = newvmspace; 3808 PROC_VMSPACE_UNLOCK(p); 3809 if (p == curthread->td_proc) 3810 pmap_activate(curthread); 3811 curthread->td_pflags |= TDP_EXECVMSPC; 3812 return (0); 3813} 3814 3815/* 3816 * Unshare the specified VM space for forcing COW. This 3817 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3818 */ 3819int 3820vmspace_unshare(struct proc *p) 3821{ 3822 struct vmspace *oldvmspace = p->p_vmspace; 3823 struct vmspace *newvmspace; 3824 vm_ooffset_t fork_charge; 3825 3826 if (oldvmspace->vm_refcnt == 1) 3827 return (0); 3828 fork_charge = 0; 3829 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3830 if (newvmspace == NULL) 3831 return (ENOMEM); 3832 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3833 vmspace_free(newvmspace); 3834 return (ENOMEM); 3835 } 3836 PROC_VMSPACE_LOCK(p); 3837 p->p_vmspace = newvmspace; 3838 PROC_VMSPACE_UNLOCK(p); 3839 if (p == curthread->td_proc) 3840 pmap_activate(curthread); 3841 vmspace_free(oldvmspace); 3842 return (0); 3843} 3844 3845/* 3846 * vm_map_lookup: 3847 * 3848 * Finds the VM object, offset, and 3849 * protection for a given virtual address in the 3850 * specified map, assuming a page fault of the 3851 * type specified. 3852 * 3853 * Leaves the map in question locked for read; return 3854 * values are guaranteed until a vm_map_lookup_done 3855 * call is performed. Note that the map argument 3856 * is in/out; the returned map must be used in 3857 * the call to vm_map_lookup_done. 3858 * 3859 * A handle (out_entry) is returned for use in 3860 * vm_map_lookup_done, to make that fast. 3861 * 3862 * If a lookup is requested with "write protection" 3863 * specified, the map may be changed to perform virtual 3864 * copying operations, although the data referenced will 3865 * remain the same. 3866 */ 3867int 3868vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3869 vm_offset_t vaddr, 3870 vm_prot_t fault_typea, 3871 vm_map_entry_t *out_entry, /* OUT */ 3872 vm_object_t *object, /* OUT */ 3873 vm_pindex_t *pindex, /* OUT */ 3874 vm_prot_t *out_prot, /* OUT */ 3875 boolean_t *wired) /* OUT */ 3876{ 3877 vm_map_entry_t entry; 3878 vm_map_t map = *var_map; 3879 vm_prot_t prot; 3880 vm_prot_t fault_type = fault_typea; 3881 vm_object_t eobject; 3882 vm_size_t size; 3883 struct ucred *cred; 3884 3885RetryLookup:; 3886 3887 vm_map_lock_read(map); 3888 3889 /* 3890 * Lookup the faulting address. 3891 */ 3892 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3893 vm_map_unlock_read(map); 3894 return (KERN_INVALID_ADDRESS); 3895 } 3896 3897 entry = *out_entry; 3898 3899 /* 3900 * Handle submaps. 3901 */ 3902 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3903 vm_map_t old_map = map; 3904 3905 *var_map = map = entry->object.sub_map; 3906 vm_map_unlock_read(old_map); 3907 goto RetryLookup; 3908 } 3909 3910 /* 3911 * Check whether this task is allowed to have this page. 3912 */ 3913 prot = entry->protection; 3914 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3915 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3916 vm_map_unlock_read(map); 3917 return (KERN_PROTECTION_FAILURE); 3918 } 3919 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3920 (entry->eflags & MAP_ENTRY_COW) && 3921 (fault_type & VM_PROT_WRITE)) { 3922 vm_map_unlock_read(map); 3923 return (KERN_PROTECTION_FAILURE); 3924 } 3925 if ((fault_typea & VM_PROT_COPY) != 0 && 3926 (entry->max_protection & VM_PROT_WRITE) == 0 && 3927 (entry->eflags & MAP_ENTRY_COW) == 0) { 3928 vm_map_unlock_read(map); 3929 return (KERN_PROTECTION_FAILURE); 3930 } 3931 3932 /* 3933 * If this page is not pageable, we have to get it for all possible 3934 * accesses. 3935 */ 3936 *wired = (entry->wired_count != 0); 3937 if (*wired) 3938 fault_type = entry->protection; 3939 size = entry->end - entry->start; 3940 /* 3941 * If the entry was copy-on-write, we either ... 3942 */ 3943 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3944 /* 3945 * If we want to write the page, we may as well handle that 3946 * now since we've got the map locked. 3947 * 3948 * If we don't need to write the page, we just demote the 3949 * permissions allowed. 3950 */ 3951 if ((fault_type & VM_PROT_WRITE) != 0 || 3952 (fault_typea & VM_PROT_COPY) != 0) { 3953 /* 3954 * Make a new object, and place it in the object 3955 * chain. Note that no new references have appeared 3956 * -- one just moved from the map to the new 3957 * object. 3958 */ 3959 if (vm_map_lock_upgrade(map)) 3960 goto RetryLookup; 3961 3962 if (entry->cred == NULL) { 3963 /* 3964 * The debugger owner is charged for 3965 * the memory. 3966 */ 3967 cred = curthread->td_ucred; 3968 crhold(cred); 3969 if (!swap_reserve_by_cred(size, cred)) { 3970 crfree(cred); 3971 vm_map_unlock(map); 3972 return (KERN_RESOURCE_SHORTAGE); 3973 } 3974 entry->cred = cred; 3975 } 3976 vm_object_shadow(&entry->object.vm_object, 3977 &entry->offset, size); 3978 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3979 eobject = entry->object.vm_object; 3980 if (eobject->cred != NULL) { 3981 /* 3982 * The object was not shadowed. 3983 */ 3984 swap_release_by_cred(size, entry->cred); 3985 crfree(entry->cred); 3986 entry->cred = NULL; 3987 } else if (entry->cred != NULL) { 3988 VM_OBJECT_WLOCK(eobject); 3989 eobject->cred = entry->cred; 3990 eobject->charge = size; 3991 VM_OBJECT_WUNLOCK(eobject); 3992 entry->cred = NULL; 3993 } 3994 3995 vm_map_lock_downgrade(map); 3996 } else { 3997 /* 3998 * We're attempting to read a copy-on-write page -- 3999 * don't allow writes. 4000 */ 4001 prot &= ~VM_PROT_WRITE; 4002 } 4003 } 4004 4005 /* 4006 * Create an object if necessary. 4007 */ 4008 if (entry->object.vm_object == NULL && 4009 !map->system_map) { 4010 if (vm_map_lock_upgrade(map)) 4011 goto RetryLookup; 4012 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4013 atop(size)); 4014 entry->offset = 0; 4015 if (entry->cred != NULL) { 4016 VM_OBJECT_WLOCK(entry->object.vm_object); 4017 entry->object.vm_object->cred = entry->cred; 4018 entry->object.vm_object->charge = size; 4019 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4020 entry->cred = NULL; 4021 } 4022 vm_map_lock_downgrade(map); 4023 } 4024 4025 /* 4026 * Return the object/offset from this entry. If the entry was 4027 * copy-on-write or empty, it has been fixed up. 4028 */ 4029 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4030 *object = entry->object.vm_object; 4031 4032 *out_prot = prot; 4033 return (KERN_SUCCESS); 4034} 4035 4036/* 4037 * vm_map_lookup_locked: 4038 * 4039 * Lookup the faulting address. A version of vm_map_lookup that returns 4040 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4041 */ 4042int 4043vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4044 vm_offset_t vaddr, 4045 vm_prot_t fault_typea, 4046 vm_map_entry_t *out_entry, /* OUT */ 4047 vm_object_t *object, /* OUT */ 4048 vm_pindex_t *pindex, /* OUT */ 4049 vm_prot_t *out_prot, /* OUT */ 4050 boolean_t *wired) /* OUT */ 4051{ 4052 vm_map_entry_t entry; 4053 vm_map_t map = *var_map; 4054 vm_prot_t prot; 4055 vm_prot_t fault_type = fault_typea; 4056 4057 /* 4058 * Lookup the faulting address. 4059 */ 4060 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4061 return (KERN_INVALID_ADDRESS); 4062 4063 entry = *out_entry; 4064 4065 /* 4066 * Fail if the entry refers to a submap. 4067 */ 4068 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4069 return (KERN_FAILURE); 4070 4071 /* 4072 * Check whether this task is allowed to have this page. 4073 */ 4074 prot = entry->protection; 4075 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4076 if ((fault_type & prot) != fault_type) 4077 return (KERN_PROTECTION_FAILURE); 4078 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 4079 (entry->eflags & MAP_ENTRY_COW) && 4080 (fault_type & VM_PROT_WRITE)) 4081 return (KERN_PROTECTION_FAILURE); 4082 4083 /* 4084 * If this page is not pageable, we have to get it for all possible 4085 * accesses. 4086 */ 4087 *wired = (entry->wired_count != 0); 4088 if (*wired) 4089 fault_type = entry->protection; 4090 4091 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4092 /* 4093 * Fail if the entry was copy-on-write for a write fault. 4094 */ 4095 if (fault_type & VM_PROT_WRITE) 4096 return (KERN_FAILURE); 4097 /* 4098 * We're attempting to read a copy-on-write page -- 4099 * don't allow writes. 4100 */ 4101 prot &= ~VM_PROT_WRITE; 4102 } 4103 4104 /* 4105 * Fail if an object should be created. 4106 */ 4107 if (entry->object.vm_object == NULL && !map->system_map) 4108 return (KERN_FAILURE); 4109 4110 /* 4111 * Return the object/offset from this entry. If the entry was 4112 * copy-on-write or empty, it has been fixed up. 4113 */ 4114 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4115 *object = entry->object.vm_object; 4116 4117 *out_prot = prot; 4118 return (KERN_SUCCESS); 4119} 4120 4121/* 4122 * vm_map_lookup_done: 4123 * 4124 * Releases locks acquired by a vm_map_lookup 4125 * (according to the handle returned by that lookup). 4126 */ 4127void 4128vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4129{ 4130 /* 4131 * Unlock the main-level map 4132 */ 4133 vm_map_unlock_read(map); 4134} 4135 4136#include "opt_ddb.h" 4137#ifdef DDB 4138#include <sys/kernel.h> 4139 4140#include <ddb/ddb.h> 4141 4142static void 4143vm_map_print(vm_map_t map) 4144{ 4145 vm_map_entry_t entry; 4146 4147 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4148 (void *)map, 4149 (void *)map->pmap, map->nentries, map->timestamp); 4150 4151 db_indent += 2; 4152 for (entry = map->header.next; entry != &map->header; 4153 entry = entry->next) { 4154 db_iprintf("map entry %p: start=%p, end=%p\n", 4155 (void *)entry, (void *)entry->start, (void *)entry->end); 4156 { 4157 static char *inheritance_name[4] = 4158 {"share", "copy", "none", "donate_copy"}; 4159 4160 db_iprintf(" prot=%x/%x/%s", 4161 entry->protection, 4162 entry->max_protection, 4163 inheritance_name[(int)(unsigned char)entry->inheritance]); 4164 if (entry->wired_count != 0) 4165 db_printf(", wired"); 4166 } 4167 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4168 db_printf(", share=%p, offset=0x%jx\n", 4169 (void *)entry->object.sub_map, 4170 (uintmax_t)entry->offset); 4171 if ((entry->prev == &map->header) || 4172 (entry->prev->object.sub_map != 4173 entry->object.sub_map)) { 4174 db_indent += 2; 4175 vm_map_print((vm_map_t)entry->object.sub_map); 4176 db_indent -= 2; 4177 } 4178 } else { 4179 if (entry->cred != NULL) 4180 db_printf(", ruid %d", entry->cred->cr_ruid); 4181 db_printf(", object=%p, offset=0x%jx", 4182 (void *)entry->object.vm_object, 4183 (uintmax_t)entry->offset); 4184 if (entry->object.vm_object && entry->object.vm_object->cred) 4185 db_printf(", obj ruid %d charge %jx", 4186 entry->object.vm_object->cred->cr_ruid, 4187 (uintmax_t)entry->object.vm_object->charge); 4188 if (entry->eflags & MAP_ENTRY_COW) 4189 db_printf(", copy (%s)", 4190 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4191 db_printf("\n"); 4192 4193 if ((entry->prev == &map->header) || 4194 (entry->prev->object.vm_object != 4195 entry->object.vm_object)) { 4196 db_indent += 2; 4197 vm_object_print((db_expr_t)(intptr_t) 4198 entry->object.vm_object, 4199 0, 0, (char *)0); 4200 db_indent -= 2; 4201 } 4202 } 4203 } 4204 db_indent -= 2; 4205} 4206 4207DB_SHOW_COMMAND(map, map) 4208{ 4209 4210 if (!have_addr) { 4211 db_printf("usage: show map <addr>\n"); 4212 return; 4213 } 4214 vm_map_print((vm_map_t)addr); 4215} 4216 4217DB_SHOW_COMMAND(procvm, procvm) 4218{ 4219 struct proc *p; 4220 4221 if (have_addr) { 4222 p = (struct proc *) addr; 4223 } else { 4224 p = curproc; 4225 } 4226 4227 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4228 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4229 (void *)vmspace_pmap(p->p_vmspace)); 4230 4231 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4232} 4233 4234#endif /* DDB */ 4235