vm_map.c revision 269072
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 269072 2014-07-24 16:29:44Z 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 if MAP_PREFAULT_PARTIAL is specified 1806 */ 1807#define MAX_INIT_PT 96 1808 1809/* 1810 * vm_map_pmap_enter: 1811 * 1812 * Preload the specified map's pmap with mappings to the specified 1813 * object's memory-resident pages. No further physical pages are 1814 * allocated, and no further virtual pages are retrieved from secondary 1815 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 1816 * limited number of page mappings are created at the low-end of the 1817 * specified address range. (For this purpose, a superpage mapping 1818 * counts as one page mapping.) Otherwise, all resident pages within 1819 * the specified address range are mapped. Because these mappings are 1820 * being created speculatively, cached pages are not reactivated and 1821 * mapped. 1822 */ 1823void 1824vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1825 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1826{ 1827 vm_offset_t start; 1828 vm_page_t p, p_start; 1829 vm_pindex_t mask, psize, threshold, tmpidx; 1830 1831 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1832 return; 1833 VM_OBJECT_RLOCK(object); 1834 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1835 VM_OBJECT_RUNLOCK(object); 1836 VM_OBJECT_WLOCK(object); 1837 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1838 pmap_object_init_pt(map->pmap, addr, object, pindex, 1839 size); 1840 VM_OBJECT_WUNLOCK(object); 1841 return; 1842 } 1843 VM_OBJECT_LOCK_DOWNGRADE(object); 1844 } 1845 1846 psize = atop(size); 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 threshold = MAX_INIT_PT; 1858 1859 p = vm_page_find_least(object, pindex); 1860 /* 1861 * Assert: the variable p is either (1) the page with the 1862 * least pindex greater than or equal to the parameter pindex 1863 * or (2) NULL. 1864 */ 1865 for (; 1866 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1867 p = TAILQ_NEXT(p, listq)) { 1868 /* 1869 * don't allow an madvise to blow away our really 1870 * free pages allocating pv entries. 1871 */ 1872 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 1873 cnt.v_free_count < cnt.v_free_reserved) || 1874 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 1875 tmpidx >= threshold)) { 1876 psize = tmpidx; 1877 break; 1878 } 1879 if (p->valid == VM_PAGE_BITS_ALL) { 1880 if (p_start == NULL) { 1881 start = addr + ptoa(tmpidx); 1882 p_start = p; 1883 } 1884 /* Jump ahead if a superpage mapping is possible. */ 1885 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 1886 (pagesizes[p->psind] - 1)) == 0) { 1887 mask = atop(pagesizes[p->psind]) - 1; 1888 if (tmpidx + mask < psize && 1889 vm_page_ps_is_valid(p)) { 1890 p += mask; 1891 threshold += mask; 1892 } 1893 } 1894 } else if (p_start != NULL) { 1895 pmap_enter_object(map->pmap, start, addr + 1896 ptoa(tmpidx), p_start, prot); 1897 p_start = NULL; 1898 } 1899 } 1900 if (p_start != NULL) 1901 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1902 p_start, prot); 1903 VM_OBJECT_RUNLOCK(object); 1904} 1905 1906/* 1907 * vm_map_protect: 1908 * 1909 * Sets the protection of the specified address 1910 * region in the target map. If "set_max" is 1911 * specified, the maximum protection is to be set; 1912 * otherwise, only the current protection is affected. 1913 */ 1914int 1915vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1916 vm_prot_t new_prot, boolean_t set_max) 1917{ 1918 vm_map_entry_t current, entry; 1919 vm_object_t obj; 1920 struct ucred *cred; 1921 vm_prot_t old_prot; 1922 1923 if (start == end) 1924 return (KERN_SUCCESS); 1925 1926 vm_map_lock(map); 1927 1928 VM_MAP_RANGE_CHECK(map, start, end); 1929 1930 if (vm_map_lookup_entry(map, start, &entry)) { 1931 vm_map_clip_start(map, entry, start); 1932 } else { 1933 entry = entry->next; 1934 } 1935 1936 /* 1937 * Make a first pass to check for protection violations. 1938 */ 1939 current = entry; 1940 while ((current != &map->header) && (current->start < end)) { 1941 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1942 vm_map_unlock(map); 1943 return (KERN_INVALID_ARGUMENT); 1944 } 1945 if ((new_prot & current->max_protection) != new_prot) { 1946 vm_map_unlock(map); 1947 return (KERN_PROTECTION_FAILURE); 1948 } 1949 current = current->next; 1950 } 1951 1952 1953 /* 1954 * Do an accounting pass for private read-only mappings that 1955 * now will do cow due to allowed write (e.g. debugger sets 1956 * breakpoint on text segment) 1957 */ 1958 for (current = entry; (current != &map->header) && 1959 (current->start < end); current = current->next) { 1960 1961 vm_map_clip_end(map, current, end); 1962 1963 if (set_max || 1964 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1965 ENTRY_CHARGED(current)) { 1966 continue; 1967 } 1968 1969 cred = curthread->td_ucred; 1970 obj = current->object.vm_object; 1971 1972 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1973 if (!swap_reserve(current->end - current->start)) { 1974 vm_map_unlock(map); 1975 return (KERN_RESOURCE_SHORTAGE); 1976 } 1977 crhold(cred); 1978 current->cred = cred; 1979 continue; 1980 } 1981 1982 VM_OBJECT_WLOCK(obj); 1983 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1984 VM_OBJECT_WUNLOCK(obj); 1985 continue; 1986 } 1987 1988 /* 1989 * Charge for the whole object allocation now, since 1990 * we cannot distinguish between non-charged and 1991 * charged clipped mapping of the same object later. 1992 */ 1993 KASSERT(obj->charge == 0, 1994 ("vm_map_protect: object %p overcharged (entry %p)", 1995 obj, current)); 1996 if (!swap_reserve(ptoa(obj->size))) { 1997 VM_OBJECT_WUNLOCK(obj); 1998 vm_map_unlock(map); 1999 return (KERN_RESOURCE_SHORTAGE); 2000 } 2001 2002 crhold(cred); 2003 obj->cred = cred; 2004 obj->charge = ptoa(obj->size); 2005 VM_OBJECT_WUNLOCK(obj); 2006 } 2007 2008 /* 2009 * Go back and fix up protections. [Note that clipping is not 2010 * necessary the second time.] 2011 */ 2012 current = entry; 2013 while ((current != &map->header) && (current->start < end)) { 2014 old_prot = current->protection; 2015 2016 if (set_max) 2017 current->protection = 2018 (current->max_protection = new_prot) & 2019 old_prot; 2020 else 2021 current->protection = new_prot; 2022 2023 /* 2024 * For user wired map entries, the normal lazy evaluation of 2025 * write access upgrades through soft page faults is 2026 * undesirable. Instead, immediately copy any pages that are 2027 * copy-on-write and enable write access in the physical map. 2028 */ 2029 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2030 (current->protection & VM_PROT_WRITE) != 0 && 2031 (old_prot & VM_PROT_WRITE) == 0) 2032 vm_fault_copy_entry(map, map, current, current, NULL); 2033 2034 /* 2035 * When restricting access, update the physical map. Worry 2036 * about copy-on-write here. 2037 */ 2038 if ((old_prot & ~current->protection) != 0) { 2039#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2040 VM_PROT_ALL) 2041 pmap_protect(map->pmap, current->start, 2042 current->end, 2043 current->protection & MASK(current)); 2044#undef MASK 2045 } 2046 vm_map_simplify_entry(map, current); 2047 current = current->next; 2048 } 2049 vm_map_unlock(map); 2050 return (KERN_SUCCESS); 2051} 2052 2053/* 2054 * vm_map_madvise: 2055 * 2056 * This routine traverses a processes map handling the madvise 2057 * system call. Advisories are classified as either those effecting 2058 * the vm_map_entry structure, or those effecting the underlying 2059 * objects. 2060 */ 2061int 2062vm_map_madvise( 2063 vm_map_t map, 2064 vm_offset_t start, 2065 vm_offset_t end, 2066 int behav) 2067{ 2068 vm_map_entry_t current, entry; 2069 int modify_map = 0; 2070 2071 /* 2072 * Some madvise calls directly modify the vm_map_entry, in which case 2073 * we need to use an exclusive lock on the map and we need to perform 2074 * various clipping operations. Otherwise we only need a read-lock 2075 * on the map. 2076 */ 2077 switch(behav) { 2078 case MADV_NORMAL: 2079 case MADV_SEQUENTIAL: 2080 case MADV_RANDOM: 2081 case MADV_NOSYNC: 2082 case MADV_AUTOSYNC: 2083 case MADV_NOCORE: 2084 case MADV_CORE: 2085 if (start == end) 2086 return (KERN_SUCCESS); 2087 modify_map = 1; 2088 vm_map_lock(map); 2089 break; 2090 case MADV_WILLNEED: 2091 case MADV_DONTNEED: 2092 case MADV_FREE: 2093 if (start == end) 2094 return (KERN_SUCCESS); 2095 vm_map_lock_read(map); 2096 break; 2097 default: 2098 return (KERN_INVALID_ARGUMENT); 2099 } 2100 2101 /* 2102 * Locate starting entry and clip if necessary. 2103 */ 2104 VM_MAP_RANGE_CHECK(map, start, end); 2105 2106 if (vm_map_lookup_entry(map, start, &entry)) { 2107 if (modify_map) 2108 vm_map_clip_start(map, entry, start); 2109 } else { 2110 entry = entry->next; 2111 } 2112 2113 if (modify_map) { 2114 /* 2115 * madvise behaviors that are implemented in the vm_map_entry. 2116 * 2117 * We clip the vm_map_entry so that behavioral changes are 2118 * limited to the specified address range. 2119 */ 2120 for (current = entry; 2121 (current != &map->header) && (current->start < end); 2122 current = current->next 2123 ) { 2124 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2125 continue; 2126 2127 vm_map_clip_end(map, current, end); 2128 2129 switch (behav) { 2130 case MADV_NORMAL: 2131 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2132 break; 2133 case MADV_SEQUENTIAL: 2134 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2135 break; 2136 case MADV_RANDOM: 2137 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2138 break; 2139 case MADV_NOSYNC: 2140 current->eflags |= MAP_ENTRY_NOSYNC; 2141 break; 2142 case MADV_AUTOSYNC: 2143 current->eflags &= ~MAP_ENTRY_NOSYNC; 2144 break; 2145 case MADV_NOCORE: 2146 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2147 break; 2148 case MADV_CORE: 2149 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2150 break; 2151 default: 2152 break; 2153 } 2154 vm_map_simplify_entry(map, current); 2155 } 2156 vm_map_unlock(map); 2157 } else { 2158 vm_pindex_t pstart, pend; 2159 2160 /* 2161 * madvise behaviors that are implemented in the underlying 2162 * vm_object. 2163 * 2164 * Since we don't clip the vm_map_entry, we have to clip 2165 * the vm_object pindex and count. 2166 */ 2167 for (current = entry; 2168 (current != &map->header) && (current->start < end); 2169 current = current->next 2170 ) { 2171 vm_offset_t useEnd, useStart; 2172 2173 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2174 continue; 2175 2176 pstart = OFF_TO_IDX(current->offset); 2177 pend = pstart + atop(current->end - current->start); 2178 useStart = current->start; 2179 useEnd = current->end; 2180 2181 if (current->start < start) { 2182 pstart += atop(start - current->start); 2183 useStart = start; 2184 } 2185 if (current->end > end) { 2186 pend -= atop(current->end - end); 2187 useEnd = end; 2188 } 2189 2190 if (pstart >= pend) 2191 continue; 2192 2193 /* 2194 * Perform the pmap_advise() before clearing 2195 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2196 * concurrent pmap operation, such as pmap_remove(), 2197 * could clear a reference in the pmap and set 2198 * PGA_REFERENCED on the page before the pmap_advise() 2199 * had completed. Consequently, the page would appear 2200 * referenced based upon an old reference that 2201 * occurred before this pmap_advise() ran. 2202 */ 2203 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2204 pmap_advise(map->pmap, useStart, useEnd, 2205 behav); 2206 2207 vm_object_madvise(current->object.vm_object, pstart, 2208 pend, behav); 2209 if (behav == MADV_WILLNEED) { 2210 vm_map_pmap_enter(map, 2211 useStart, 2212 current->protection, 2213 current->object.vm_object, 2214 pstart, 2215 ptoa(pend - pstart), 2216 MAP_PREFAULT_MADVISE 2217 ); 2218 } 2219 } 2220 vm_map_unlock_read(map); 2221 } 2222 return (0); 2223} 2224 2225 2226/* 2227 * vm_map_inherit: 2228 * 2229 * Sets the inheritance of the specified address 2230 * range in the target map. Inheritance 2231 * affects how the map will be shared with 2232 * child maps at the time of vmspace_fork. 2233 */ 2234int 2235vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2236 vm_inherit_t new_inheritance) 2237{ 2238 vm_map_entry_t entry; 2239 vm_map_entry_t temp_entry; 2240 2241 switch (new_inheritance) { 2242 case VM_INHERIT_NONE: 2243 case VM_INHERIT_COPY: 2244 case VM_INHERIT_SHARE: 2245 break; 2246 default: 2247 return (KERN_INVALID_ARGUMENT); 2248 } 2249 if (start == end) 2250 return (KERN_SUCCESS); 2251 vm_map_lock(map); 2252 VM_MAP_RANGE_CHECK(map, start, end); 2253 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2254 entry = temp_entry; 2255 vm_map_clip_start(map, entry, start); 2256 } else 2257 entry = temp_entry->next; 2258 while ((entry != &map->header) && (entry->start < end)) { 2259 vm_map_clip_end(map, entry, end); 2260 entry->inheritance = new_inheritance; 2261 vm_map_simplify_entry(map, entry); 2262 entry = entry->next; 2263 } 2264 vm_map_unlock(map); 2265 return (KERN_SUCCESS); 2266} 2267 2268/* 2269 * vm_map_unwire: 2270 * 2271 * Implements both kernel and user unwiring. 2272 */ 2273int 2274vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2275 int flags) 2276{ 2277 vm_map_entry_t entry, first_entry, tmp_entry; 2278 vm_offset_t saved_start; 2279 unsigned int last_timestamp; 2280 int rv; 2281 boolean_t need_wakeup, result, user_unwire; 2282 2283 if (start == end) 2284 return (KERN_SUCCESS); 2285 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2286 vm_map_lock(map); 2287 VM_MAP_RANGE_CHECK(map, start, end); 2288 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2289 if (flags & VM_MAP_WIRE_HOLESOK) 2290 first_entry = first_entry->next; 2291 else { 2292 vm_map_unlock(map); 2293 return (KERN_INVALID_ADDRESS); 2294 } 2295 } 2296 last_timestamp = map->timestamp; 2297 entry = first_entry; 2298 while (entry != &map->header && entry->start < end) { 2299 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2300 /* 2301 * We have not yet clipped the entry. 2302 */ 2303 saved_start = (start >= entry->start) ? start : 2304 entry->start; 2305 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2306 if (vm_map_unlock_and_wait(map, 0)) { 2307 /* 2308 * Allow interruption of user unwiring? 2309 */ 2310 } 2311 vm_map_lock(map); 2312 if (last_timestamp+1 != map->timestamp) { 2313 /* 2314 * Look again for the entry because the map was 2315 * modified while it was unlocked. 2316 * Specifically, the entry may have been 2317 * clipped, merged, or deleted. 2318 */ 2319 if (!vm_map_lookup_entry(map, saved_start, 2320 &tmp_entry)) { 2321 if (flags & VM_MAP_WIRE_HOLESOK) 2322 tmp_entry = tmp_entry->next; 2323 else { 2324 if (saved_start == start) { 2325 /* 2326 * First_entry has been deleted. 2327 */ 2328 vm_map_unlock(map); 2329 return (KERN_INVALID_ADDRESS); 2330 } 2331 end = saved_start; 2332 rv = KERN_INVALID_ADDRESS; 2333 goto done; 2334 } 2335 } 2336 if (entry == first_entry) 2337 first_entry = tmp_entry; 2338 else 2339 first_entry = NULL; 2340 entry = tmp_entry; 2341 } 2342 last_timestamp = map->timestamp; 2343 continue; 2344 } 2345 vm_map_clip_start(map, entry, start); 2346 vm_map_clip_end(map, entry, end); 2347 /* 2348 * Mark the entry in case the map lock is released. (See 2349 * above.) 2350 */ 2351 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2352 entry->wiring_thread == NULL, 2353 ("owned map entry %p", entry)); 2354 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2355 entry->wiring_thread = curthread; 2356 /* 2357 * Check the map for holes in the specified region. 2358 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2359 */ 2360 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2361 (entry->end < end && (entry->next == &map->header || 2362 entry->next->start > entry->end))) { 2363 end = entry->end; 2364 rv = KERN_INVALID_ADDRESS; 2365 goto done; 2366 } 2367 /* 2368 * If system unwiring, require that the entry is system wired. 2369 */ 2370 if (!user_unwire && 2371 vm_map_entry_system_wired_count(entry) == 0) { 2372 end = entry->end; 2373 rv = KERN_INVALID_ARGUMENT; 2374 goto done; 2375 } 2376 entry = entry->next; 2377 } 2378 rv = KERN_SUCCESS; 2379done: 2380 need_wakeup = FALSE; 2381 if (first_entry == NULL) { 2382 result = vm_map_lookup_entry(map, start, &first_entry); 2383 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2384 first_entry = first_entry->next; 2385 else 2386 KASSERT(result, ("vm_map_unwire: lookup failed")); 2387 } 2388 for (entry = first_entry; entry != &map->header && entry->start < end; 2389 entry = entry->next) { 2390 /* 2391 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2392 * space in the unwired region could have been mapped 2393 * while the map lock was dropped for draining 2394 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2395 * could be simultaneously wiring this new mapping 2396 * entry. Detect these cases and skip any entries 2397 * marked as in transition by us. 2398 */ 2399 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2400 entry->wiring_thread != curthread) { 2401 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2402 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2403 continue; 2404 } 2405 2406 if (rv == KERN_SUCCESS && (!user_unwire || 2407 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2408 if (user_unwire) 2409 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2410 entry->wired_count--; 2411 if (entry->wired_count == 0) { 2412 /* 2413 * Retain the map lock. 2414 */ 2415 vm_fault_unwire(map, entry->start, entry->end, 2416 entry->object.vm_object != NULL && 2417 (entry->object.vm_object->flags & 2418 OBJ_FICTITIOUS) != 0); 2419 } 2420 } 2421 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2422 ("vm_map_unwire: in-transition flag missing %p", entry)); 2423 KASSERT(entry->wiring_thread == curthread, 2424 ("vm_map_unwire: alien wire %p", entry)); 2425 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2426 entry->wiring_thread = NULL; 2427 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2428 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2429 need_wakeup = TRUE; 2430 } 2431 vm_map_simplify_entry(map, entry); 2432 } 2433 vm_map_unlock(map); 2434 if (need_wakeup) 2435 vm_map_wakeup(map); 2436 return (rv); 2437} 2438 2439/* 2440 * vm_map_wire: 2441 * 2442 * Implements both kernel and user wiring. 2443 */ 2444int 2445vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2446 int flags) 2447{ 2448 vm_map_entry_t entry, first_entry, tmp_entry; 2449 vm_offset_t saved_end, saved_start; 2450 unsigned int last_timestamp; 2451 int rv; 2452 boolean_t fictitious, need_wakeup, result, user_wire; 2453 vm_prot_t prot; 2454 2455 if (start == end) 2456 return (KERN_SUCCESS); 2457 prot = 0; 2458 if (flags & VM_MAP_WIRE_WRITE) 2459 prot |= VM_PROT_WRITE; 2460 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2461 vm_map_lock(map); 2462 VM_MAP_RANGE_CHECK(map, start, end); 2463 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2464 if (flags & VM_MAP_WIRE_HOLESOK) 2465 first_entry = first_entry->next; 2466 else { 2467 vm_map_unlock(map); 2468 return (KERN_INVALID_ADDRESS); 2469 } 2470 } 2471 last_timestamp = map->timestamp; 2472 entry = first_entry; 2473 while (entry != &map->header && entry->start < end) { 2474 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2475 /* 2476 * We have not yet clipped the entry. 2477 */ 2478 saved_start = (start >= entry->start) ? start : 2479 entry->start; 2480 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2481 if (vm_map_unlock_and_wait(map, 0)) { 2482 /* 2483 * Allow interruption of user wiring? 2484 */ 2485 } 2486 vm_map_lock(map); 2487 if (last_timestamp + 1 != map->timestamp) { 2488 /* 2489 * Look again for the entry because the map was 2490 * modified while it was unlocked. 2491 * Specifically, the entry may have been 2492 * clipped, merged, or deleted. 2493 */ 2494 if (!vm_map_lookup_entry(map, saved_start, 2495 &tmp_entry)) { 2496 if (flags & VM_MAP_WIRE_HOLESOK) 2497 tmp_entry = tmp_entry->next; 2498 else { 2499 if (saved_start == start) { 2500 /* 2501 * first_entry has been deleted. 2502 */ 2503 vm_map_unlock(map); 2504 return (KERN_INVALID_ADDRESS); 2505 } 2506 end = saved_start; 2507 rv = KERN_INVALID_ADDRESS; 2508 goto done; 2509 } 2510 } 2511 if (entry == first_entry) 2512 first_entry = tmp_entry; 2513 else 2514 first_entry = NULL; 2515 entry = tmp_entry; 2516 } 2517 last_timestamp = map->timestamp; 2518 continue; 2519 } 2520 vm_map_clip_start(map, entry, start); 2521 vm_map_clip_end(map, entry, end); 2522 /* 2523 * Mark the entry in case the map lock is released. (See 2524 * above.) 2525 */ 2526 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2527 entry->wiring_thread == NULL, 2528 ("owned map entry %p", entry)); 2529 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2530 entry->wiring_thread = curthread; 2531 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2532 || (entry->protection & prot) != prot) { 2533 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2534 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2535 end = entry->end; 2536 rv = KERN_INVALID_ADDRESS; 2537 goto done; 2538 } 2539 goto next_entry; 2540 } 2541 if (entry->wired_count == 0) { 2542 entry->wired_count++; 2543 saved_start = entry->start; 2544 saved_end = entry->end; 2545 fictitious = entry->object.vm_object != NULL && 2546 (entry->object.vm_object->flags & 2547 OBJ_FICTITIOUS) != 0; 2548 /* 2549 * Release the map lock, relying on the in-transition 2550 * mark. Mark the map busy for fork. 2551 */ 2552 vm_map_busy(map); 2553 vm_map_unlock(map); 2554 rv = vm_fault_wire(map, saved_start, saved_end, 2555 fictitious); 2556 vm_map_lock(map); 2557 vm_map_unbusy(map); 2558 if (last_timestamp + 1 != map->timestamp) { 2559 /* 2560 * Look again for the entry because the map was 2561 * modified while it was unlocked. The entry 2562 * may have been clipped, but NOT merged or 2563 * deleted. 2564 */ 2565 result = vm_map_lookup_entry(map, saved_start, 2566 &tmp_entry); 2567 KASSERT(result, ("vm_map_wire: lookup failed")); 2568 if (entry == first_entry) 2569 first_entry = tmp_entry; 2570 else 2571 first_entry = NULL; 2572 entry = tmp_entry; 2573 while (entry->end < saved_end) { 2574 if (rv != KERN_SUCCESS) { 2575 KASSERT(entry->wired_count == 1, 2576 ("vm_map_wire: bad count")); 2577 entry->wired_count = -1; 2578 } 2579 entry = entry->next; 2580 } 2581 } 2582 last_timestamp = map->timestamp; 2583 if (rv != KERN_SUCCESS) { 2584 KASSERT(entry->wired_count == 1, 2585 ("vm_map_wire: bad count")); 2586 /* 2587 * Assign an out-of-range value to represent 2588 * the failure to wire this entry. 2589 */ 2590 entry->wired_count = -1; 2591 end = entry->end; 2592 goto done; 2593 } 2594 } else if (!user_wire || 2595 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2596 entry->wired_count++; 2597 } 2598 /* 2599 * Check the map for holes in the specified region. 2600 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2601 */ 2602 next_entry: 2603 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2604 (entry->end < end && (entry->next == &map->header || 2605 entry->next->start > entry->end))) { 2606 end = entry->end; 2607 rv = KERN_INVALID_ADDRESS; 2608 goto done; 2609 } 2610 entry = entry->next; 2611 } 2612 rv = KERN_SUCCESS; 2613done: 2614 need_wakeup = FALSE; 2615 if (first_entry == NULL) { 2616 result = vm_map_lookup_entry(map, start, &first_entry); 2617 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2618 first_entry = first_entry->next; 2619 else 2620 KASSERT(result, ("vm_map_wire: lookup failed")); 2621 } 2622 for (entry = first_entry; entry != &map->header && entry->start < end; 2623 entry = entry->next) { 2624 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2625 goto next_entry_done; 2626 2627 /* 2628 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2629 * space in the unwired region could have been mapped 2630 * while the map lock was dropped for faulting in the 2631 * pages or draining MAP_ENTRY_IN_TRANSITION. 2632 * Moreover, another thread could be simultaneously 2633 * wiring this new mapping entry. Detect these cases 2634 * and skip any entries marked as in transition by us. 2635 */ 2636 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2637 entry->wiring_thread != curthread) { 2638 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2639 ("vm_map_wire: !HOLESOK and new/changed entry")); 2640 continue; 2641 } 2642 2643 if (rv == KERN_SUCCESS) { 2644 if (user_wire) 2645 entry->eflags |= MAP_ENTRY_USER_WIRED; 2646 } else if (entry->wired_count == -1) { 2647 /* 2648 * Wiring failed on this entry. Thus, unwiring is 2649 * unnecessary. 2650 */ 2651 entry->wired_count = 0; 2652 } else { 2653 if (!user_wire || 2654 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2655 entry->wired_count--; 2656 if (entry->wired_count == 0) { 2657 /* 2658 * Retain the map lock. 2659 */ 2660 vm_fault_unwire(map, entry->start, entry->end, 2661 entry->object.vm_object != NULL && 2662 (entry->object.vm_object->flags & 2663 OBJ_FICTITIOUS) != 0); 2664 } 2665 } 2666 next_entry_done: 2667 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2668 ("vm_map_wire: in-transition flag missing %p", entry)); 2669 KASSERT(entry->wiring_thread == curthread, 2670 ("vm_map_wire: alien wire %p", entry)); 2671 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2672 MAP_ENTRY_WIRE_SKIPPED); 2673 entry->wiring_thread = NULL; 2674 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2675 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2676 need_wakeup = TRUE; 2677 } 2678 vm_map_simplify_entry(map, entry); 2679 } 2680 vm_map_unlock(map); 2681 if (need_wakeup) 2682 vm_map_wakeup(map); 2683 return (rv); 2684} 2685 2686/* 2687 * vm_map_sync 2688 * 2689 * Push any dirty cached pages in the address range to their pager. 2690 * If syncio is TRUE, dirty pages are written synchronously. 2691 * If invalidate is TRUE, any cached pages are freed as well. 2692 * 2693 * If the size of the region from start to end is zero, we are 2694 * supposed to flush all modified pages within the region containing 2695 * start. Unfortunately, a region can be split or coalesced with 2696 * neighboring regions, making it difficult to determine what the 2697 * original region was. Therefore, we approximate this requirement by 2698 * flushing the current region containing start. 2699 * 2700 * Returns an error if any part of the specified range is not mapped. 2701 */ 2702int 2703vm_map_sync( 2704 vm_map_t map, 2705 vm_offset_t start, 2706 vm_offset_t end, 2707 boolean_t syncio, 2708 boolean_t invalidate) 2709{ 2710 vm_map_entry_t current; 2711 vm_map_entry_t entry; 2712 vm_size_t size; 2713 vm_object_t object; 2714 vm_ooffset_t offset; 2715 unsigned int last_timestamp; 2716 boolean_t failed; 2717 2718 vm_map_lock_read(map); 2719 VM_MAP_RANGE_CHECK(map, start, end); 2720 if (!vm_map_lookup_entry(map, start, &entry)) { 2721 vm_map_unlock_read(map); 2722 return (KERN_INVALID_ADDRESS); 2723 } else if (start == end) { 2724 start = entry->start; 2725 end = entry->end; 2726 } 2727 /* 2728 * Make a first pass to check for user-wired memory and holes. 2729 */ 2730 for (current = entry; current != &map->header && current->start < end; 2731 current = current->next) { 2732 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2733 vm_map_unlock_read(map); 2734 return (KERN_INVALID_ARGUMENT); 2735 } 2736 if (end > current->end && 2737 (current->next == &map->header || 2738 current->end != current->next->start)) { 2739 vm_map_unlock_read(map); 2740 return (KERN_INVALID_ADDRESS); 2741 } 2742 } 2743 2744 if (invalidate) 2745 pmap_remove(map->pmap, start, end); 2746 failed = FALSE; 2747 2748 /* 2749 * Make a second pass, cleaning/uncaching pages from the indicated 2750 * objects as we go. 2751 */ 2752 for (current = entry; current != &map->header && current->start < end;) { 2753 offset = current->offset + (start - current->start); 2754 size = (end <= current->end ? end : current->end) - start; 2755 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2756 vm_map_t smap; 2757 vm_map_entry_t tentry; 2758 vm_size_t tsize; 2759 2760 smap = current->object.sub_map; 2761 vm_map_lock_read(smap); 2762 (void) vm_map_lookup_entry(smap, offset, &tentry); 2763 tsize = tentry->end - offset; 2764 if (tsize < size) 2765 size = tsize; 2766 object = tentry->object.vm_object; 2767 offset = tentry->offset + (offset - tentry->start); 2768 vm_map_unlock_read(smap); 2769 } else { 2770 object = current->object.vm_object; 2771 } 2772 vm_object_reference(object); 2773 last_timestamp = map->timestamp; 2774 vm_map_unlock_read(map); 2775 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2776 failed = TRUE; 2777 start += size; 2778 vm_object_deallocate(object); 2779 vm_map_lock_read(map); 2780 if (last_timestamp == map->timestamp || 2781 !vm_map_lookup_entry(map, start, ¤t)) 2782 current = current->next; 2783 } 2784 2785 vm_map_unlock_read(map); 2786 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2787} 2788 2789/* 2790 * vm_map_entry_unwire: [ internal use only ] 2791 * 2792 * Make the region specified by this entry pageable. 2793 * 2794 * The map in question should be locked. 2795 * [This is the reason for this routine's existence.] 2796 */ 2797static void 2798vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2799{ 2800 vm_fault_unwire(map, entry->start, entry->end, 2801 entry->object.vm_object != NULL && 2802 (entry->object.vm_object->flags & OBJ_FICTITIOUS) != 0); 2803 entry->wired_count = 0; 2804} 2805 2806static void 2807vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2808{ 2809 2810 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2811 vm_object_deallocate(entry->object.vm_object); 2812 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2813} 2814 2815/* 2816 * vm_map_entry_delete: [ internal use only ] 2817 * 2818 * Deallocate the given entry from the target map. 2819 */ 2820static void 2821vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2822{ 2823 vm_object_t object; 2824 vm_pindex_t offidxstart, offidxend, count, size1; 2825 vm_ooffset_t size; 2826 2827 vm_map_entry_unlink(map, entry); 2828 object = entry->object.vm_object; 2829 size = entry->end - entry->start; 2830 map->size -= size; 2831 2832 if (entry->cred != NULL) { 2833 swap_release_by_cred(size, entry->cred); 2834 crfree(entry->cred); 2835 } 2836 2837 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2838 (object != NULL)) { 2839 KASSERT(entry->cred == NULL || object->cred == NULL || 2840 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2841 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2842 count = OFF_TO_IDX(size); 2843 offidxstart = OFF_TO_IDX(entry->offset); 2844 offidxend = offidxstart + count; 2845 VM_OBJECT_WLOCK(object); 2846 if (object->ref_count != 1 && 2847 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2848 object == kernel_object || object == kmem_object)) { 2849 vm_object_collapse(object); 2850 2851 /* 2852 * The option OBJPR_NOTMAPPED can be passed here 2853 * because vm_map_delete() already performed 2854 * pmap_remove() on the only mapping to this range 2855 * of pages. 2856 */ 2857 vm_object_page_remove(object, offidxstart, offidxend, 2858 OBJPR_NOTMAPPED); 2859 if (object->type == OBJT_SWAP) 2860 swap_pager_freespace(object, offidxstart, count); 2861 if (offidxend >= object->size && 2862 offidxstart < object->size) { 2863 size1 = object->size; 2864 object->size = offidxstart; 2865 if (object->cred != NULL) { 2866 size1 -= object->size; 2867 KASSERT(object->charge >= ptoa(size1), 2868 ("vm_map_entry_delete: object->charge < 0")); 2869 swap_release_by_cred(ptoa(size1), object->cred); 2870 object->charge -= ptoa(size1); 2871 } 2872 } 2873 } 2874 VM_OBJECT_WUNLOCK(object); 2875 } else 2876 entry->object.vm_object = NULL; 2877 if (map->system_map) 2878 vm_map_entry_deallocate(entry, TRUE); 2879 else { 2880 entry->next = curthread->td_map_def_user; 2881 curthread->td_map_def_user = entry; 2882 } 2883} 2884 2885/* 2886 * vm_map_delete: [ internal use only ] 2887 * 2888 * Deallocates the given address range from the target 2889 * map. 2890 */ 2891int 2892vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2893{ 2894 vm_map_entry_t entry; 2895 vm_map_entry_t first_entry; 2896 2897 VM_MAP_ASSERT_LOCKED(map); 2898 if (start == end) 2899 return (KERN_SUCCESS); 2900 2901 /* 2902 * Find the start of the region, and clip it 2903 */ 2904 if (!vm_map_lookup_entry(map, start, &first_entry)) 2905 entry = first_entry->next; 2906 else { 2907 entry = first_entry; 2908 vm_map_clip_start(map, entry, start); 2909 } 2910 2911 /* 2912 * Step through all entries in this region 2913 */ 2914 while ((entry != &map->header) && (entry->start < end)) { 2915 vm_map_entry_t next; 2916 2917 /* 2918 * Wait for wiring or unwiring of an entry to complete. 2919 * Also wait for any system wirings to disappear on 2920 * user maps. 2921 */ 2922 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2923 (vm_map_pmap(map) != kernel_pmap && 2924 vm_map_entry_system_wired_count(entry) != 0)) { 2925 unsigned int last_timestamp; 2926 vm_offset_t saved_start; 2927 vm_map_entry_t tmp_entry; 2928 2929 saved_start = entry->start; 2930 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2931 last_timestamp = map->timestamp; 2932 (void) vm_map_unlock_and_wait(map, 0); 2933 vm_map_lock(map); 2934 if (last_timestamp + 1 != map->timestamp) { 2935 /* 2936 * Look again for the entry because the map was 2937 * modified while it was unlocked. 2938 * Specifically, the entry may have been 2939 * clipped, merged, or deleted. 2940 */ 2941 if (!vm_map_lookup_entry(map, saved_start, 2942 &tmp_entry)) 2943 entry = tmp_entry->next; 2944 else { 2945 entry = tmp_entry; 2946 vm_map_clip_start(map, entry, 2947 saved_start); 2948 } 2949 } 2950 continue; 2951 } 2952 vm_map_clip_end(map, entry, end); 2953 2954 next = entry->next; 2955 2956 /* 2957 * Unwire before removing addresses from the pmap; otherwise, 2958 * unwiring will put the entries back in the pmap. 2959 */ 2960 if (entry->wired_count != 0) { 2961 vm_map_entry_unwire(map, entry); 2962 } 2963 2964 pmap_remove(map->pmap, entry->start, entry->end); 2965 2966 /* 2967 * Delete the entry only after removing all pmap 2968 * entries pointing to its pages. (Otherwise, its 2969 * page frames may be reallocated, and any modify bits 2970 * will be set in the wrong object!) 2971 */ 2972 vm_map_entry_delete(map, entry); 2973 entry = next; 2974 } 2975 return (KERN_SUCCESS); 2976} 2977 2978/* 2979 * vm_map_remove: 2980 * 2981 * Remove the given address range from the target map. 2982 * This is the exported form of vm_map_delete. 2983 */ 2984int 2985vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2986{ 2987 int result; 2988 2989 vm_map_lock(map); 2990 VM_MAP_RANGE_CHECK(map, start, end); 2991 result = vm_map_delete(map, start, end); 2992 vm_map_unlock(map); 2993 return (result); 2994} 2995 2996/* 2997 * vm_map_check_protection: 2998 * 2999 * Assert that the target map allows the specified privilege on the 3000 * entire address region given. The entire region must be allocated. 3001 * 3002 * WARNING! This code does not and should not check whether the 3003 * contents of the region is accessible. For example a smaller file 3004 * might be mapped into a larger address space. 3005 * 3006 * NOTE! This code is also called by munmap(). 3007 * 3008 * The map must be locked. A read lock is sufficient. 3009 */ 3010boolean_t 3011vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3012 vm_prot_t protection) 3013{ 3014 vm_map_entry_t entry; 3015 vm_map_entry_t tmp_entry; 3016 3017 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3018 return (FALSE); 3019 entry = tmp_entry; 3020 3021 while (start < end) { 3022 if (entry == &map->header) 3023 return (FALSE); 3024 /* 3025 * No holes allowed! 3026 */ 3027 if (start < entry->start) 3028 return (FALSE); 3029 /* 3030 * Check protection associated with entry. 3031 */ 3032 if ((entry->protection & protection) != protection) 3033 return (FALSE); 3034 /* go to next entry */ 3035 start = entry->end; 3036 entry = entry->next; 3037 } 3038 return (TRUE); 3039} 3040 3041/* 3042 * vm_map_copy_entry: 3043 * 3044 * Copies the contents of the source entry to the destination 3045 * entry. The entries *must* be aligned properly. 3046 */ 3047static void 3048vm_map_copy_entry( 3049 vm_map_t src_map, 3050 vm_map_t dst_map, 3051 vm_map_entry_t src_entry, 3052 vm_map_entry_t dst_entry, 3053 vm_ooffset_t *fork_charge) 3054{ 3055 vm_object_t src_object; 3056 vm_map_entry_t fake_entry; 3057 vm_offset_t size; 3058 struct ucred *cred; 3059 int charged; 3060 3061 VM_MAP_ASSERT_LOCKED(dst_map); 3062 3063 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3064 return; 3065 3066 if (src_entry->wired_count == 0 || 3067 (src_entry->protection & VM_PROT_WRITE) == 0) { 3068 /* 3069 * If the source entry is marked needs_copy, it is already 3070 * write-protected. 3071 */ 3072 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3073 (src_entry->protection & VM_PROT_WRITE) != 0) { 3074 pmap_protect(src_map->pmap, 3075 src_entry->start, 3076 src_entry->end, 3077 src_entry->protection & ~VM_PROT_WRITE); 3078 } 3079 3080 /* 3081 * Make a copy of the object. 3082 */ 3083 size = src_entry->end - src_entry->start; 3084 if ((src_object = src_entry->object.vm_object) != NULL) { 3085 VM_OBJECT_WLOCK(src_object); 3086 charged = ENTRY_CHARGED(src_entry); 3087 if ((src_object->handle == NULL) && 3088 (src_object->type == OBJT_DEFAULT || 3089 src_object->type == OBJT_SWAP)) { 3090 vm_object_collapse(src_object); 3091 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3092 vm_object_split(src_entry); 3093 src_object = src_entry->object.vm_object; 3094 } 3095 } 3096 vm_object_reference_locked(src_object); 3097 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3098 if (src_entry->cred != NULL && 3099 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3100 KASSERT(src_object->cred == NULL, 3101 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3102 src_object)); 3103 src_object->cred = src_entry->cred; 3104 src_object->charge = size; 3105 } 3106 VM_OBJECT_WUNLOCK(src_object); 3107 dst_entry->object.vm_object = src_object; 3108 if (charged) { 3109 cred = curthread->td_ucred; 3110 crhold(cred); 3111 dst_entry->cred = cred; 3112 *fork_charge += size; 3113 if (!(src_entry->eflags & 3114 MAP_ENTRY_NEEDS_COPY)) { 3115 crhold(cred); 3116 src_entry->cred = cred; 3117 *fork_charge += size; 3118 } 3119 } 3120 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3121 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3122 dst_entry->offset = src_entry->offset; 3123 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3124 /* 3125 * MAP_ENTRY_VN_WRITECNT cannot 3126 * indicate write reference from 3127 * src_entry, since the entry is 3128 * marked as needs copy. Allocate a 3129 * fake entry that is used to 3130 * decrement object->un_pager.vnp.writecount 3131 * at the appropriate time. Attach 3132 * fake_entry to the deferred list. 3133 */ 3134 fake_entry = vm_map_entry_create(dst_map); 3135 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3136 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3137 vm_object_reference(src_object); 3138 fake_entry->object.vm_object = src_object; 3139 fake_entry->start = src_entry->start; 3140 fake_entry->end = src_entry->end; 3141 fake_entry->next = curthread->td_map_def_user; 3142 curthread->td_map_def_user = fake_entry; 3143 } 3144 } else { 3145 dst_entry->object.vm_object = NULL; 3146 dst_entry->offset = 0; 3147 if (src_entry->cred != NULL) { 3148 dst_entry->cred = curthread->td_ucred; 3149 crhold(dst_entry->cred); 3150 *fork_charge += size; 3151 } 3152 } 3153 3154 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3155 dst_entry->end - dst_entry->start, src_entry->start); 3156 } else { 3157 /* 3158 * We don't want to make writeable wired pages copy-on-write. 3159 * Immediately copy these pages into the new map by simulating 3160 * page faults. The new pages are pageable. 3161 */ 3162 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3163 fork_charge); 3164 } 3165} 3166 3167/* 3168 * vmspace_map_entry_forked: 3169 * Update the newly-forked vmspace each time a map entry is inherited 3170 * or copied. The values for vm_dsize and vm_tsize are approximate 3171 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3172 */ 3173static void 3174vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3175 vm_map_entry_t entry) 3176{ 3177 vm_size_t entrysize; 3178 vm_offset_t newend; 3179 3180 entrysize = entry->end - entry->start; 3181 vm2->vm_map.size += entrysize; 3182 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3183 vm2->vm_ssize += btoc(entrysize); 3184 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3185 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3186 newend = MIN(entry->end, 3187 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3188 vm2->vm_dsize += btoc(newend - entry->start); 3189 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3190 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3191 newend = MIN(entry->end, 3192 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3193 vm2->vm_tsize += btoc(newend - entry->start); 3194 } 3195} 3196 3197/* 3198 * vmspace_fork: 3199 * Create a new process vmspace structure and vm_map 3200 * based on those of an existing process. The new map 3201 * is based on the old map, according to the inheritance 3202 * values on the regions in that map. 3203 * 3204 * XXX It might be worth coalescing the entries added to the new vmspace. 3205 * 3206 * The source map must not be locked. 3207 */ 3208struct vmspace * 3209vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3210{ 3211 struct vmspace *vm2; 3212 vm_map_t new_map, old_map; 3213 vm_map_entry_t new_entry, old_entry; 3214 vm_object_t object; 3215 int locked; 3216 3217 old_map = &vm1->vm_map; 3218 /* Copy immutable fields of vm1 to vm2. */ 3219 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3220 if (vm2 == NULL) 3221 return (NULL); 3222 vm2->vm_taddr = vm1->vm_taddr; 3223 vm2->vm_daddr = vm1->vm_daddr; 3224 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3225 vm_map_lock(old_map); 3226 if (old_map->busy) 3227 vm_map_wait_busy(old_map); 3228 new_map = &vm2->vm_map; 3229 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3230 KASSERT(locked, ("vmspace_fork: lock failed")); 3231 3232 old_entry = old_map->header.next; 3233 3234 while (old_entry != &old_map->header) { 3235 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3236 panic("vm_map_fork: encountered a submap"); 3237 3238 switch (old_entry->inheritance) { 3239 case VM_INHERIT_NONE: 3240 break; 3241 3242 case VM_INHERIT_SHARE: 3243 /* 3244 * Clone the entry, creating the shared object if necessary. 3245 */ 3246 object = old_entry->object.vm_object; 3247 if (object == NULL) { 3248 object = vm_object_allocate(OBJT_DEFAULT, 3249 atop(old_entry->end - old_entry->start)); 3250 old_entry->object.vm_object = object; 3251 old_entry->offset = 0; 3252 if (old_entry->cred != NULL) { 3253 object->cred = old_entry->cred; 3254 object->charge = old_entry->end - 3255 old_entry->start; 3256 old_entry->cred = NULL; 3257 } 3258 } 3259 3260 /* 3261 * Add the reference before calling vm_object_shadow 3262 * to insure that a shadow object is created. 3263 */ 3264 vm_object_reference(object); 3265 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3266 vm_object_shadow(&old_entry->object.vm_object, 3267 &old_entry->offset, 3268 old_entry->end - old_entry->start); 3269 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3270 /* Transfer the second reference too. */ 3271 vm_object_reference( 3272 old_entry->object.vm_object); 3273 3274 /* 3275 * As in vm_map_simplify_entry(), the 3276 * vnode lock will not be acquired in 3277 * this call to vm_object_deallocate(). 3278 */ 3279 vm_object_deallocate(object); 3280 object = old_entry->object.vm_object; 3281 } 3282 VM_OBJECT_WLOCK(object); 3283 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3284 if (old_entry->cred != NULL) { 3285 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3286 object->cred = old_entry->cred; 3287 object->charge = old_entry->end - old_entry->start; 3288 old_entry->cred = NULL; 3289 } 3290 3291 /* 3292 * Assert the correct state of the vnode 3293 * v_writecount while the object is locked, to 3294 * not relock it later for the assertion 3295 * correctness. 3296 */ 3297 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3298 object->type == OBJT_VNODE) { 3299 KASSERT(((struct vnode *)object->handle)-> 3300 v_writecount > 0, 3301 ("vmspace_fork: v_writecount %p", object)); 3302 KASSERT(object->un_pager.vnp.writemappings > 0, 3303 ("vmspace_fork: vnp.writecount %p", 3304 object)); 3305 } 3306 VM_OBJECT_WUNLOCK(object); 3307 3308 /* 3309 * Clone the entry, referencing the shared object. 3310 */ 3311 new_entry = vm_map_entry_create(new_map); 3312 *new_entry = *old_entry; 3313 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3314 MAP_ENTRY_IN_TRANSITION); 3315 new_entry->wiring_thread = NULL; 3316 new_entry->wired_count = 0; 3317 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3318 vnode_pager_update_writecount(object, 3319 new_entry->start, new_entry->end); 3320 } 3321 3322 /* 3323 * Insert the entry into the new map -- we know we're 3324 * inserting at the end of the new map. 3325 */ 3326 vm_map_entry_link(new_map, new_map->header.prev, 3327 new_entry); 3328 vmspace_map_entry_forked(vm1, vm2, new_entry); 3329 3330 /* 3331 * Update the physical map 3332 */ 3333 pmap_copy(new_map->pmap, old_map->pmap, 3334 new_entry->start, 3335 (old_entry->end - old_entry->start), 3336 old_entry->start); 3337 break; 3338 3339 case VM_INHERIT_COPY: 3340 /* 3341 * Clone the entry and link into the map. 3342 */ 3343 new_entry = vm_map_entry_create(new_map); 3344 *new_entry = *old_entry; 3345 /* 3346 * Copied entry is COW over the old object. 3347 */ 3348 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3349 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3350 new_entry->wiring_thread = NULL; 3351 new_entry->wired_count = 0; 3352 new_entry->object.vm_object = NULL; 3353 new_entry->cred = NULL; 3354 vm_map_entry_link(new_map, new_map->header.prev, 3355 new_entry); 3356 vmspace_map_entry_forked(vm1, vm2, new_entry); 3357 vm_map_copy_entry(old_map, new_map, old_entry, 3358 new_entry, fork_charge); 3359 break; 3360 } 3361 old_entry = old_entry->next; 3362 } 3363 /* 3364 * Use inlined vm_map_unlock() to postpone handling the deferred 3365 * map entries, which cannot be done until both old_map and 3366 * new_map locks are released. 3367 */ 3368 sx_xunlock(&old_map->lock); 3369 sx_xunlock(&new_map->lock); 3370 vm_map_process_deferred(); 3371 3372 return (vm2); 3373} 3374 3375int 3376vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3377 vm_prot_t prot, vm_prot_t max, int cow) 3378{ 3379 vm_size_t growsize, init_ssize; 3380 rlim_t lmemlim, vmemlim; 3381 int rv; 3382 3383 growsize = sgrowsiz; 3384 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3385 vm_map_lock(map); 3386 PROC_LOCK(curproc); 3387 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3388 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3389 PROC_UNLOCK(curproc); 3390 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3391 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3392 rv = KERN_NO_SPACE; 3393 goto out; 3394 } 3395 } 3396 /* If we would blow our VMEM resource limit, no go */ 3397 if (map->size + init_ssize > vmemlim) { 3398 rv = KERN_NO_SPACE; 3399 goto out; 3400 } 3401 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3402 max, cow); 3403out: 3404 vm_map_unlock(map); 3405 return (rv); 3406} 3407 3408static int 3409vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3410 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3411{ 3412 vm_map_entry_t new_entry, prev_entry; 3413 vm_offset_t bot, top; 3414 vm_size_t init_ssize; 3415 int orient, rv; 3416 3417 /* 3418 * The stack orientation is piggybacked with the cow argument. 3419 * Extract it into orient and mask the cow argument so that we 3420 * don't pass it around further. 3421 * NOTE: We explicitly allow bi-directional stacks. 3422 */ 3423 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3424 KASSERT(orient != 0, ("No stack grow direction")); 3425 3426 if (addrbos < vm_map_min(map) || 3427 addrbos > vm_map_max(map) || 3428 addrbos + max_ssize < addrbos) 3429 return (KERN_NO_SPACE); 3430 3431 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3432 3433 /* If addr is already mapped, no go */ 3434 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3435 return (KERN_NO_SPACE); 3436 3437 /* 3438 * If we can't accomodate max_ssize in the current mapping, no go. 3439 * However, we need to be aware that subsequent user mappings might 3440 * map into the space we have reserved for stack, and currently this 3441 * space is not protected. 3442 * 3443 * Hopefully we will at least detect this condition when we try to 3444 * grow the stack. 3445 */ 3446 if ((prev_entry->next != &map->header) && 3447 (prev_entry->next->start < addrbos + max_ssize)) 3448 return (KERN_NO_SPACE); 3449 3450 /* 3451 * We initially map a stack of only init_ssize. We will grow as 3452 * needed later. Depending on the orientation of the stack (i.e. 3453 * the grow direction) we either map at the top of the range, the 3454 * bottom of the range or in the middle. 3455 * 3456 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3457 * and cow to be 0. Possibly we should eliminate these as input 3458 * parameters, and just pass these values here in the insert call. 3459 */ 3460 if (orient == MAP_STACK_GROWS_DOWN) 3461 bot = addrbos + max_ssize - init_ssize; 3462 else if (orient == MAP_STACK_GROWS_UP) 3463 bot = addrbos; 3464 else 3465 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3466 top = bot + init_ssize; 3467 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3468 3469 /* Now set the avail_ssize amount. */ 3470 if (rv == KERN_SUCCESS) { 3471 if (prev_entry != &map->header) 3472 vm_map_clip_end(map, prev_entry, bot); 3473 new_entry = prev_entry->next; 3474 if (new_entry->end != top || new_entry->start != bot) 3475 panic("Bad entry start/end for new stack entry"); 3476 3477 new_entry->avail_ssize = max_ssize - init_ssize; 3478 if (orient & MAP_STACK_GROWS_DOWN) 3479 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3480 if (orient & MAP_STACK_GROWS_UP) 3481 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3482 } 3483 3484 return (rv); 3485} 3486 3487static int stack_guard_page = 0; 3488TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3489SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3490 &stack_guard_page, 0, 3491 "Insert stack guard page ahead of the growable segments."); 3492 3493/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3494 * desired address is already mapped, or if we successfully grow 3495 * the stack. Also returns KERN_SUCCESS if addr is outside the 3496 * stack range (this is strange, but preserves compatibility with 3497 * the grow function in vm_machdep.c). 3498 */ 3499int 3500vm_map_growstack(struct proc *p, vm_offset_t addr) 3501{ 3502 vm_map_entry_t next_entry, prev_entry; 3503 vm_map_entry_t new_entry, stack_entry; 3504 struct vmspace *vm = p->p_vmspace; 3505 vm_map_t map = &vm->vm_map; 3506 vm_offset_t end; 3507 vm_size_t growsize; 3508 size_t grow_amount, max_grow; 3509 rlim_t lmemlim, stacklim, vmemlim; 3510 int is_procstack, rv; 3511 struct ucred *cred; 3512#ifdef notyet 3513 uint64_t limit; 3514#endif 3515#ifdef RACCT 3516 int error; 3517#endif 3518 3519Retry: 3520 PROC_LOCK(p); 3521 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3522 stacklim = lim_cur(p, RLIMIT_STACK); 3523 vmemlim = lim_cur(p, RLIMIT_VMEM); 3524 PROC_UNLOCK(p); 3525 3526 vm_map_lock_read(map); 3527 3528 /* If addr is already in the entry range, no need to grow.*/ 3529 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3530 vm_map_unlock_read(map); 3531 return (KERN_SUCCESS); 3532 } 3533 3534 next_entry = prev_entry->next; 3535 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3536 /* 3537 * This entry does not grow upwards. Since the address lies 3538 * beyond this entry, the next entry (if one exists) has to 3539 * be a downward growable entry. The entry list header is 3540 * never a growable entry, so it suffices to check the flags. 3541 */ 3542 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3543 vm_map_unlock_read(map); 3544 return (KERN_SUCCESS); 3545 } 3546 stack_entry = next_entry; 3547 } else { 3548 /* 3549 * This entry grows upward. If the next entry does not at 3550 * least grow downwards, this is the entry we need to grow. 3551 * otherwise we have two possible choices and we have to 3552 * select one. 3553 */ 3554 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3555 /* 3556 * We have two choices; grow the entry closest to 3557 * the address to minimize the amount of growth. 3558 */ 3559 if (addr - prev_entry->end <= next_entry->start - addr) 3560 stack_entry = prev_entry; 3561 else 3562 stack_entry = next_entry; 3563 } else 3564 stack_entry = prev_entry; 3565 } 3566 3567 if (stack_entry == next_entry) { 3568 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3569 KASSERT(addr < stack_entry->start, ("foo")); 3570 end = (prev_entry != &map->header) ? prev_entry->end : 3571 stack_entry->start - stack_entry->avail_ssize; 3572 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3573 max_grow = stack_entry->start - end; 3574 } else { 3575 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3576 KASSERT(addr >= stack_entry->end, ("foo")); 3577 end = (next_entry != &map->header) ? next_entry->start : 3578 stack_entry->end + stack_entry->avail_ssize; 3579 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3580 max_grow = end - stack_entry->end; 3581 } 3582 3583 if (grow_amount > stack_entry->avail_ssize) { 3584 vm_map_unlock_read(map); 3585 return (KERN_NO_SPACE); 3586 } 3587 3588 /* 3589 * If there is no longer enough space between the entries nogo, and 3590 * adjust the available space. Note: this should only happen if the 3591 * user has mapped into the stack area after the stack was created, 3592 * and is probably an error. 3593 * 3594 * This also effectively destroys any guard page the user might have 3595 * intended by limiting the stack size. 3596 */ 3597 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3598 if (vm_map_lock_upgrade(map)) 3599 goto Retry; 3600 3601 stack_entry->avail_ssize = max_grow; 3602 3603 vm_map_unlock(map); 3604 return (KERN_NO_SPACE); 3605 } 3606 3607 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3608 3609 /* 3610 * If this is the main process stack, see if we're over the stack 3611 * limit. 3612 */ 3613 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3614 vm_map_unlock_read(map); 3615 return (KERN_NO_SPACE); 3616 } 3617#ifdef RACCT 3618 PROC_LOCK(p); 3619 if (is_procstack && 3620 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3621 PROC_UNLOCK(p); 3622 vm_map_unlock_read(map); 3623 return (KERN_NO_SPACE); 3624 } 3625 PROC_UNLOCK(p); 3626#endif 3627 3628 /* Round up the grow amount modulo sgrowsiz */ 3629 growsize = sgrowsiz; 3630 grow_amount = roundup(grow_amount, growsize); 3631 if (grow_amount > stack_entry->avail_ssize) 3632 grow_amount = stack_entry->avail_ssize; 3633 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3634 grow_amount = trunc_page((vm_size_t)stacklim) - 3635 ctob(vm->vm_ssize); 3636 } 3637#ifdef notyet 3638 PROC_LOCK(p); 3639 limit = racct_get_available(p, RACCT_STACK); 3640 PROC_UNLOCK(p); 3641 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3642 grow_amount = limit - ctob(vm->vm_ssize); 3643#endif 3644 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3645 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3646 vm_map_unlock_read(map); 3647 rv = KERN_NO_SPACE; 3648 goto out; 3649 } 3650#ifdef RACCT 3651 PROC_LOCK(p); 3652 if (racct_set(p, RACCT_MEMLOCK, 3653 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3654 PROC_UNLOCK(p); 3655 vm_map_unlock_read(map); 3656 rv = KERN_NO_SPACE; 3657 goto out; 3658 } 3659 PROC_UNLOCK(p); 3660#endif 3661 } 3662 /* If we would blow our VMEM resource limit, no go */ 3663 if (map->size + grow_amount > vmemlim) { 3664 vm_map_unlock_read(map); 3665 rv = KERN_NO_SPACE; 3666 goto out; 3667 } 3668#ifdef RACCT 3669 PROC_LOCK(p); 3670 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3671 PROC_UNLOCK(p); 3672 vm_map_unlock_read(map); 3673 rv = KERN_NO_SPACE; 3674 goto out; 3675 } 3676 PROC_UNLOCK(p); 3677#endif 3678 3679 if (vm_map_lock_upgrade(map)) 3680 goto Retry; 3681 3682 if (stack_entry == next_entry) { 3683 /* 3684 * Growing downward. 3685 */ 3686 /* Get the preliminary new entry start value */ 3687 addr = stack_entry->start - grow_amount; 3688 3689 /* 3690 * If this puts us into the previous entry, cut back our 3691 * growth to the available space. Also, see the note above. 3692 */ 3693 if (addr < end) { 3694 stack_entry->avail_ssize = max_grow; 3695 addr = end; 3696 if (stack_guard_page) 3697 addr += PAGE_SIZE; 3698 } 3699 3700 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3701 next_entry->protection, next_entry->max_protection, 0); 3702 3703 /* Adjust the available stack space by the amount we grew. */ 3704 if (rv == KERN_SUCCESS) { 3705 if (prev_entry != &map->header) 3706 vm_map_clip_end(map, prev_entry, addr); 3707 new_entry = prev_entry->next; 3708 KASSERT(new_entry == stack_entry->prev, ("foo")); 3709 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3710 KASSERT(new_entry->start == addr, ("foo")); 3711 grow_amount = new_entry->end - new_entry->start; 3712 new_entry->avail_ssize = stack_entry->avail_ssize - 3713 grow_amount; 3714 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3715 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3716 } 3717 } else { 3718 /* 3719 * Growing upward. 3720 */ 3721 addr = stack_entry->end + grow_amount; 3722 3723 /* 3724 * If this puts us into the next entry, cut back our growth 3725 * to the available space. Also, see the note above. 3726 */ 3727 if (addr > end) { 3728 stack_entry->avail_ssize = end - stack_entry->end; 3729 addr = end; 3730 if (stack_guard_page) 3731 addr -= PAGE_SIZE; 3732 } 3733 3734 grow_amount = addr - stack_entry->end; 3735 cred = stack_entry->cred; 3736 if (cred == NULL && stack_entry->object.vm_object != NULL) 3737 cred = stack_entry->object.vm_object->cred; 3738 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3739 rv = KERN_NO_SPACE; 3740 /* Grow the underlying object if applicable. */ 3741 else if (stack_entry->object.vm_object == NULL || 3742 vm_object_coalesce(stack_entry->object.vm_object, 3743 stack_entry->offset, 3744 (vm_size_t)(stack_entry->end - stack_entry->start), 3745 (vm_size_t)grow_amount, cred != NULL)) { 3746 map->size += (addr - stack_entry->end); 3747 /* Update the current entry. */ 3748 stack_entry->end = addr; 3749 stack_entry->avail_ssize -= grow_amount; 3750 vm_map_entry_resize_free(map, stack_entry); 3751 rv = KERN_SUCCESS; 3752 3753 if (next_entry != &map->header) 3754 vm_map_clip_start(map, next_entry, addr); 3755 } else 3756 rv = KERN_FAILURE; 3757 } 3758 3759 if (rv == KERN_SUCCESS && is_procstack) 3760 vm->vm_ssize += btoc(grow_amount); 3761 3762 vm_map_unlock(map); 3763 3764 /* 3765 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3766 */ 3767 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3768 vm_map_wire(map, 3769 (stack_entry == next_entry) ? addr : addr - grow_amount, 3770 (stack_entry == next_entry) ? stack_entry->start : addr, 3771 (p->p_flag & P_SYSTEM) 3772 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3773 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3774 } 3775 3776out: 3777#ifdef RACCT 3778 if (rv != KERN_SUCCESS) { 3779 PROC_LOCK(p); 3780 error = racct_set(p, RACCT_VMEM, map->size); 3781 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3782 if (!old_mlock) { 3783 error = racct_set(p, RACCT_MEMLOCK, 3784 ptoa(pmap_wired_count(map->pmap))); 3785 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3786 } 3787 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3788 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3789 PROC_UNLOCK(p); 3790 } 3791#endif 3792 3793 return (rv); 3794} 3795 3796/* 3797 * Unshare the specified VM space for exec. If other processes are 3798 * mapped to it, then create a new one. The new vmspace is null. 3799 */ 3800int 3801vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3802{ 3803 struct vmspace *oldvmspace = p->p_vmspace; 3804 struct vmspace *newvmspace; 3805 3806 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3807 ("vmspace_exec recursed")); 3808 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3809 if (newvmspace == NULL) 3810 return (ENOMEM); 3811 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3812 /* 3813 * This code is written like this for prototype purposes. The 3814 * goal is to avoid running down the vmspace here, but let the 3815 * other process's that are still using the vmspace to finally 3816 * run it down. Even though there is little or no chance of blocking 3817 * here, it is a good idea to keep this form for future mods. 3818 */ 3819 PROC_VMSPACE_LOCK(p); 3820 p->p_vmspace = newvmspace; 3821 PROC_VMSPACE_UNLOCK(p); 3822 if (p == curthread->td_proc) 3823 pmap_activate(curthread); 3824 curthread->td_pflags |= TDP_EXECVMSPC; 3825 return (0); 3826} 3827 3828/* 3829 * Unshare the specified VM space for forcing COW. This 3830 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3831 */ 3832int 3833vmspace_unshare(struct proc *p) 3834{ 3835 struct vmspace *oldvmspace = p->p_vmspace; 3836 struct vmspace *newvmspace; 3837 vm_ooffset_t fork_charge; 3838 3839 if (oldvmspace->vm_refcnt == 1) 3840 return (0); 3841 fork_charge = 0; 3842 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3843 if (newvmspace == NULL) 3844 return (ENOMEM); 3845 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3846 vmspace_free(newvmspace); 3847 return (ENOMEM); 3848 } 3849 PROC_VMSPACE_LOCK(p); 3850 p->p_vmspace = newvmspace; 3851 PROC_VMSPACE_UNLOCK(p); 3852 if (p == curthread->td_proc) 3853 pmap_activate(curthread); 3854 vmspace_free(oldvmspace); 3855 return (0); 3856} 3857 3858/* 3859 * vm_map_lookup: 3860 * 3861 * Finds the VM object, offset, and 3862 * protection for a given virtual address in the 3863 * specified map, assuming a page fault of the 3864 * type specified. 3865 * 3866 * Leaves the map in question locked for read; return 3867 * values are guaranteed until a vm_map_lookup_done 3868 * call is performed. Note that the map argument 3869 * is in/out; the returned map must be used in 3870 * the call to vm_map_lookup_done. 3871 * 3872 * A handle (out_entry) is returned for use in 3873 * vm_map_lookup_done, to make that fast. 3874 * 3875 * If a lookup is requested with "write protection" 3876 * specified, the map may be changed to perform virtual 3877 * copying operations, although the data referenced will 3878 * remain the same. 3879 */ 3880int 3881vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3882 vm_offset_t vaddr, 3883 vm_prot_t fault_typea, 3884 vm_map_entry_t *out_entry, /* OUT */ 3885 vm_object_t *object, /* OUT */ 3886 vm_pindex_t *pindex, /* OUT */ 3887 vm_prot_t *out_prot, /* OUT */ 3888 boolean_t *wired) /* OUT */ 3889{ 3890 vm_map_entry_t entry; 3891 vm_map_t map = *var_map; 3892 vm_prot_t prot; 3893 vm_prot_t fault_type = fault_typea; 3894 vm_object_t eobject; 3895 vm_size_t size; 3896 struct ucred *cred; 3897 3898RetryLookup:; 3899 3900 vm_map_lock_read(map); 3901 3902 /* 3903 * Lookup the faulting address. 3904 */ 3905 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3906 vm_map_unlock_read(map); 3907 return (KERN_INVALID_ADDRESS); 3908 } 3909 3910 entry = *out_entry; 3911 3912 /* 3913 * Handle submaps. 3914 */ 3915 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3916 vm_map_t old_map = map; 3917 3918 *var_map = map = entry->object.sub_map; 3919 vm_map_unlock_read(old_map); 3920 goto RetryLookup; 3921 } 3922 3923 /* 3924 * Check whether this task is allowed to have this page. 3925 */ 3926 prot = entry->protection; 3927 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3928 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3929 vm_map_unlock_read(map); 3930 return (KERN_PROTECTION_FAILURE); 3931 } 3932 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3933 (entry->eflags & MAP_ENTRY_COW) && 3934 (fault_type & VM_PROT_WRITE)) { 3935 vm_map_unlock_read(map); 3936 return (KERN_PROTECTION_FAILURE); 3937 } 3938 if ((fault_typea & VM_PROT_COPY) != 0 && 3939 (entry->max_protection & VM_PROT_WRITE) == 0 && 3940 (entry->eflags & MAP_ENTRY_COW) == 0) { 3941 vm_map_unlock_read(map); 3942 return (KERN_PROTECTION_FAILURE); 3943 } 3944 3945 /* 3946 * If this page is not pageable, we have to get it for all possible 3947 * accesses. 3948 */ 3949 *wired = (entry->wired_count != 0); 3950 if (*wired) 3951 fault_type = entry->protection; 3952 size = entry->end - entry->start; 3953 /* 3954 * If the entry was copy-on-write, we either ... 3955 */ 3956 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3957 /* 3958 * If we want to write the page, we may as well handle that 3959 * now since we've got the map locked. 3960 * 3961 * If we don't need to write the page, we just demote the 3962 * permissions allowed. 3963 */ 3964 if ((fault_type & VM_PROT_WRITE) != 0 || 3965 (fault_typea & VM_PROT_COPY) != 0) { 3966 /* 3967 * Make a new object, and place it in the object 3968 * chain. Note that no new references have appeared 3969 * -- one just moved from the map to the new 3970 * object. 3971 */ 3972 if (vm_map_lock_upgrade(map)) 3973 goto RetryLookup; 3974 3975 if (entry->cred == NULL) { 3976 /* 3977 * The debugger owner is charged for 3978 * the memory. 3979 */ 3980 cred = curthread->td_ucred; 3981 crhold(cred); 3982 if (!swap_reserve_by_cred(size, cred)) { 3983 crfree(cred); 3984 vm_map_unlock(map); 3985 return (KERN_RESOURCE_SHORTAGE); 3986 } 3987 entry->cred = cred; 3988 } 3989 vm_object_shadow(&entry->object.vm_object, 3990 &entry->offset, size); 3991 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3992 eobject = entry->object.vm_object; 3993 if (eobject->cred != NULL) { 3994 /* 3995 * The object was not shadowed. 3996 */ 3997 swap_release_by_cred(size, entry->cred); 3998 crfree(entry->cred); 3999 entry->cred = NULL; 4000 } else if (entry->cred != NULL) { 4001 VM_OBJECT_WLOCK(eobject); 4002 eobject->cred = entry->cred; 4003 eobject->charge = size; 4004 VM_OBJECT_WUNLOCK(eobject); 4005 entry->cred = NULL; 4006 } 4007 4008 vm_map_lock_downgrade(map); 4009 } else { 4010 /* 4011 * We're attempting to read a copy-on-write page -- 4012 * don't allow writes. 4013 */ 4014 prot &= ~VM_PROT_WRITE; 4015 } 4016 } 4017 4018 /* 4019 * Create an object if necessary. 4020 */ 4021 if (entry->object.vm_object == NULL && 4022 !map->system_map) { 4023 if (vm_map_lock_upgrade(map)) 4024 goto RetryLookup; 4025 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4026 atop(size)); 4027 entry->offset = 0; 4028 if (entry->cred != NULL) { 4029 VM_OBJECT_WLOCK(entry->object.vm_object); 4030 entry->object.vm_object->cred = entry->cred; 4031 entry->object.vm_object->charge = size; 4032 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4033 entry->cred = NULL; 4034 } 4035 vm_map_lock_downgrade(map); 4036 } 4037 4038 /* 4039 * Return the object/offset from this entry. If the entry was 4040 * copy-on-write or empty, it has been fixed up. 4041 */ 4042 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4043 *object = entry->object.vm_object; 4044 4045 *out_prot = prot; 4046 return (KERN_SUCCESS); 4047} 4048 4049/* 4050 * vm_map_lookup_locked: 4051 * 4052 * Lookup the faulting address. A version of vm_map_lookup that returns 4053 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4054 */ 4055int 4056vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4057 vm_offset_t vaddr, 4058 vm_prot_t fault_typea, 4059 vm_map_entry_t *out_entry, /* OUT */ 4060 vm_object_t *object, /* OUT */ 4061 vm_pindex_t *pindex, /* OUT */ 4062 vm_prot_t *out_prot, /* OUT */ 4063 boolean_t *wired) /* OUT */ 4064{ 4065 vm_map_entry_t entry; 4066 vm_map_t map = *var_map; 4067 vm_prot_t prot; 4068 vm_prot_t fault_type = fault_typea; 4069 4070 /* 4071 * Lookup the faulting address. 4072 */ 4073 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4074 return (KERN_INVALID_ADDRESS); 4075 4076 entry = *out_entry; 4077 4078 /* 4079 * Fail if the entry refers to a submap. 4080 */ 4081 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4082 return (KERN_FAILURE); 4083 4084 /* 4085 * Check whether this task is allowed to have this page. 4086 */ 4087 prot = entry->protection; 4088 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4089 if ((fault_type & prot) != fault_type) 4090 return (KERN_PROTECTION_FAILURE); 4091 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 4092 (entry->eflags & MAP_ENTRY_COW) && 4093 (fault_type & VM_PROT_WRITE)) 4094 return (KERN_PROTECTION_FAILURE); 4095 4096 /* 4097 * If this page is not pageable, we have to get it for all possible 4098 * accesses. 4099 */ 4100 *wired = (entry->wired_count != 0); 4101 if (*wired) 4102 fault_type = entry->protection; 4103 4104 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4105 /* 4106 * Fail if the entry was copy-on-write for a write fault. 4107 */ 4108 if (fault_type & VM_PROT_WRITE) 4109 return (KERN_FAILURE); 4110 /* 4111 * We're attempting to read a copy-on-write page -- 4112 * don't allow writes. 4113 */ 4114 prot &= ~VM_PROT_WRITE; 4115 } 4116 4117 /* 4118 * Fail if an object should be created. 4119 */ 4120 if (entry->object.vm_object == NULL && !map->system_map) 4121 return (KERN_FAILURE); 4122 4123 /* 4124 * Return the object/offset from this entry. If the entry was 4125 * copy-on-write or empty, it has been fixed up. 4126 */ 4127 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4128 *object = entry->object.vm_object; 4129 4130 *out_prot = prot; 4131 return (KERN_SUCCESS); 4132} 4133 4134/* 4135 * vm_map_lookup_done: 4136 * 4137 * Releases locks acquired by a vm_map_lookup 4138 * (according to the handle returned by that lookup). 4139 */ 4140void 4141vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4142{ 4143 /* 4144 * Unlock the main-level map 4145 */ 4146 vm_map_unlock_read(map); 4147} 4148 4149#include "opt_ddb.h" 4150#ifdef DDB 4151#include <sys/kernel.h> 4152 4153#include <ddb/ddb.h> 4154 4155static void 4156vm_map_print(vm_map_t map) 4157{ 4158 vm_map_entry_t entry; 4159 4160 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4161 (void *)map, 4162 (void *)map->pmap, map->nentries, map->timestamp); 4163 4164 db_indent += 2; 4165 for (entry = map->header.next; entry != &map->header; 4166 entry = entry->next) { 4167 db_iprintf("map entry %p: start=%p, end=%p\n", 4168 (void *)entry, (void *)entry->start, (void *)entry->end); 4169 { 4170 static char *inheritance_name[4] = 4171 {"share", "copy", "none", "donate_copy"}; 4172 4173 db_iprintf(" prot=%x/%x/%s", 4174 entry->protection, 4175 entry->max_protection, 4176 inheritance_name[(int)(unsigned char)entry->inheritance]); 4177 if (entry->wired_count != 0) 4178 db_printf(", wired"); 4179 } 4180 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4181 db_printf(", share=%p, offset=0x%jx\n", 4182 (void *)entry->object.sub_map, 4183 (uintmax_t)entry->offset); 4184 if ((entry->prev == &map->header) || 4185 (entry->prev->object.sub_map != 4186 entry->object.sub_map)) { 4187 db_indent += 2; 4188 vm_map_print((vm_map_t)entry->object.sub_map); 4189 db_indent -= 2; 4190 } 4191 } else { 4192 if (entry->cred != NULL) 4193 db_printf(", ruid %d", entry->cred->cr_ruid); 4194 db_printf(", object=%p, offset=0x%jx", 4195 (void *)entry->object.vm_object, 4196 (uintmax_t)entry->offset); 4197 if (entry->object.vm_object && entry->object.vm_object->cred) 4198 db_printf(", obj ruid %d charge %jx", 4199 entry->object.vm_object->cred->cr_ruid, 4200 (uintmax_t)entry->object.vm_object->charge); 4201 if (entry->eflags & MAP_ENTRY_COW) 4202 db_printf(", copy (%s)", 4203 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4204 db_printf("\n"); 4205 4206 if ((entry->prev == &map->header) || 4207 (entry->prev->object.vm_object != 4208 entry->object.vm_object)) { 4209 db_indent += 2; 4210 vm_object_print((db_expr_t)(intptr_t) 4211 entry->object.vm_object, 4212 0, 0, (char *)0); 4213 db_indent -= 2; 4214 } 4215 } 4216 } 4217 db_indent -= 2; 4218} 4219 4220DB_SHOW_COMMAND(map, map) 4221{ 4222 4223 if (!have_addr) { 4224 db_printf("usage: show map <addr>\n"); 4225 return; 4226 } 4227 vm_map_print((vm_map_t)addr); 4228} 4229 4230DB_SHOW_COMMAND(procvm, procvm) 4231{ 4232 struct proc *p; 4233 4234 if (have_addr) { 4235 p = (struct proc *) addr; 4236 } else { 4237 p = curproc; 4238 } 4239 4240 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4241 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4242 (void *)vmspace_pmap(p->p_vmspace)); 4243 4244 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4245} 4246 4247#endif /* DDB */ 4248