vm_map.c revision 279474
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 279474 2015-03-01 04:39:07Z rstone $"); 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/kernel.h> 71#include <sys/ktr.h> 72#include <sys/lock.h> 73#include <sys/mutex.h> 74#include <sys/proc.h> 75#include <sys/vmmeter.h> 76#include <sys/mman.h> 77#include <sys/vnode.h> 78#include <sys/racct.h> 79#include <sys/resourcevar.h> 80#include <sys/rwlock.h> 81#include <sys/file.h> 82#include <sys/sysctl.h> 83#include <sys/sysent.h> 84#include <sys/shm.h> 85 86#include <vm/vm.h> 87#include <vm/vm_param.h> 88#include <vm/pmap.h> 89#include <vm/vm_map.h> 90#include <vm/vm_page.h> 91#include <vm/vm_object.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_kern.h> 94#include <vm/vm_extern.h> 95#include <vm/vnode_pager.h> 96#include <vm/swap_pager.h> 97#include <vm/uma.h> 98 99/* 100 * Virtual memory maps provide for the mapping, protection, 101 * and sharing of virtual memory objects. In addition, 102 * this module provides for an efficient virtual copy of 103 * memory from one map to another. 104 * 105 * Synchronization is required prior to most operations. 106 * 107 * Maps consist of an ordered doubly-linked list of simple 108 * entries; a self-adjusting binary search tree of these 109 * entries is used to speed up lookups. 110 * 111 * Since portions of maps are specified by start/end addresses, 112 * which may not align with existing map entries, all 113 * routines merely "clip" entries to these start/end values. 114 * [That is, an entry is split into two, bordering at a 115 * start or end value.] Note that these clippings may not 116 * always be necessary (as the two resulting entries are then 117 * not changed); however, the clipping is done for convenience. 118 * 119 * As mentioned above, virtual copy operations are performed 120 * by copying VM object references from one map to 121 * another, and then marking both regions as copy-on-write. 122 */ 123 124static struct mtx map_sleep_mtx; 125static uma_zone_t mapentzone; 126static uma_zone_t kmapentzone; 127static uma_zone_t mapzone; 128static uma_zone_t vmspace_zone; 129static int vmspace_zinit(void *mem, int size, int flags); 130static int vm_map_zinit(void *mem, int ize, int flags); 131static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 132 vm_offset_t max); 133static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 134static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 135static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry); 136#ifdef INVARIANTS 137static void vm_map_zdtor(void *mem, int size, void *arg); 138static void vmspace_zdtor(void *mem, int size, void *arg); 139#endif 140static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, 141 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, 142 int cow); 143static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 144 vm_offset_t failed_addr); 145 146#define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 147 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 148 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 149 150/* 151 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 152 * stable. 153 */ 154#define PROC_VMSPACE_LOCK(p) do { } while (0) 155#define PROC_VMSPACE_UNLOCK(p) do { } while (0) 156 157/* 158 * VM_MAP_RANGE_CHECK: [ internal use only ] 159 * 160 * Asserts that the starting and ending region 161 * addresses fall within the valid range of the map. 162 */ 163#define VM_MAP_RANGE_CHECK(map, start, end) \ 164 { \ 165 if (start < vm_map_min(map)) \ 166 start = vm_map_min(map); \ 167 if (end > vm_map_max(map)) \ 168 end = vm_map_max(map); \ 169 if (start > end) \ 170 start = end; \ 171 } 172 173/* 174 * vm_map_startup: 175 * 176 * Initialize the vm_map module. Must be called before 177 * any other vm_map routines. 178 * 179 * Map and entry structures are allocated from the general 180 * purpose memory pool with some exceptions: 181 * 182 * - The kernel map and kmem submap are allocated statically. 183 * - Kernel map entries are allocated out of a static pool. 184 * 185 * These restrictions are necessary since malloc() uses the 186 * maps and requires map entries. 187 */ 188 189void 190vm_map_startup(void) 191{ 192 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 193 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 194#ifdef INVARIANTS 195 vm_map_zdtor, 196#else 197 NULL, 198#endif 199 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 200 uma_prealloc(mapzone, MAX_KMAP); 201 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 203 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 204 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 205 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 206 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 207#ifdef INVARIANTS 208 vmspace_zdtor, 209#else 210 NULL, 211#endif 212 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 213} 214 215static int 216vmspace_zinit(void *mem, int size, int flags) 217{ 218 struct vmspace *vm; 219 220 vm = (struct vmspace *)mem; 221 222 vm->vm_map.pmap = NULL; 223 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 224 PMAP_LOCK_INIT(vmspace_pmap(vm)); 225 return (0); 226} 227 228static int 229vm_map_zinit(void *mem, int size, int flags) 230{ 231 vm_map_t map; 232 233 map = (vm_map_t)mem; 234 memset(map, 0, sizeof(*map)); 235 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 236 sx_init(&map->lock, "vm map (user)"); 237 return (0); 238} 239 240#ifdef INVARIANTS 241static void 242vmspace_zdtor(void *mem, int size, void *arg) 243{ 244 struct vmspace *vm; 245 246 vm = (struct vmspace *)mem; 247 248 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 249} 250static void 251vm_map_zdtor(void *mem, int size, void *arg) 252{ 253 vm_map_t map; 254 255 map = (vm_map_t)mem; 256 KASSERT(map->nentries == 0, 257 ("map %p nentries == %d on free.", 258 map, map->nentries)); 259 KASSERT(map->size == 0, 260 ("map %p size == %lu on free.", 261 map, (unsigned long)map->size)); 262} 263#endif /* INVARIANTS */ 264 265/* 266 * Allocate a vmspace structure, including a vm_map and pmap, 267 * and initialize those structures. The refcnt is set to 1. 268 * 269 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 270 */ 271struct vmspace * 272vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 273{ 274 struct vmspace *vm; 275 276 vm = uma_zalloc(vmspace_zone, M_WAITOK); 277 278 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 279 280 if (pinit == NULL) 281 pinit = &pmap_pinit; 282 283 if (!pinit(vmspace_pmap(vm))) { 284 uma_zfree(vmspace_zone, vm); 285 return (NULL); 286 } 287 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 288 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 289 vm->vm_refcnt = 1; 290 vm->vm_shm = NULL; 291 vm->vm_swrss = 0; 292 vm->vm_tsize = 0; 293 vm->vm_dsize = 0; 294 vm->vm_ssize = 0; 295 vm->vm_taddr = 0; 296 vm->vm_daddr = 0; 297 vm->vm_maxsaddr = 0; 298 return (vm); 299} 300 301static void 302vmspace_container_reset(struct proc *p) 303{ 304 305#ifdef RACCT 306 PROC_LOCK(p); 307 racct_set(p, RACCT_DATA, 0); 308 racct_set(p, RACCT_STACK, 0); 309 racct_set(p, RACCT_RSS, 0); 310 racct_set(p, RACCT_MEMLOCK, 0); 311 racct_set(p, RACCT_VMEM, 0); 312 PROC_UNLOCK(p); 313#endif 314} 315 316static inline void 317vmspace_dofree(struct vmspace *vm) 318{ 319 320 CTR1(KTR_VM, "vmspace_free: %p", vm); 321 322 /* 323 * Make sure any SysV shm is freed, it might not have been in 324 * exit1(). 325 */ 326 shmexit(vm); 327 328 /* 329 * Lock the map, to wait out all other references to it. 330 * Delete all of the mappings and pages they hold, then call 331 * the pmap module to reclaim anything left. 332 */ 333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 334 vm->vm_map.max_offset); 335 336 pmap_release(vmspace_pmap(vm)); 337 vm->vm_map.pmap = NULL; 338 uma_zfree(vmspace_zone, vm); 339} 340 341void 342vmspace_free(struct vmspace *vm) 343{ 344 345 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 346 "vmspace_free() called with non-sleepable lock held"); 347 348 if (vm->vm_refcnt == 0) 349 panic("vmspace_free: attempt to free already freed vmspace"); 350 351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 352 vmspace_dofree(vm); 353} 354 355void 356vmspace_exitfree(struct proc *p) 357{ 358 struct vmspace *vm; 359 360 PROC_VMSPACE_LOCK(p); 361 vm = p->p_vmspace; 362 p->p_vmspace = NULL; 363 PROC_VMSPACE_UNLOCK(p); 364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 365 vmspace_free(vm); 366} 367 368void 369vmspace_exit(struct thread *td) 370{ 371 int refcnt; 372 struct vmspace *vm; 373 struct proc *p; 374 375 /* 376 * Release user portion of address space. 377 * This releases references to vnodes, 378 * which could cause I/O if the file has been unlinked. 379 * Need to do this early enough that we can still sleep. 380 * 381 * The last exiting process to reach this point releases as 382 * much of the environment as it can. vmspace_dofree() is the 383 * slower fallback in case another process had a temporary 384 * reference to the vmspace. 385 */ 386 387 p = td->td_proc; 388 vm = p->p_vmspace; 389 atomic_add_int(&vmspace0.vm_refcnt, 1); 390 do { 391 refcnt = vm->vm_refcnt; 392 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 393 /* Switch now since other proc might free vmspace */ 394 PROC_VMSPACE_LOCK(p); 395 p->p_vmspace = &vmspace0; 396 PROC_VMSPACE_UNLOCK(p); 397 pmap_activate(td); 398 } 399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 400 if (refcnt == 1) { 401 if (p->p_vmspace != vm) { 402 /* vmspace not yet freed, switch back */ 403 PROC_VMSPACE_LOCK(p); 404 p->p_vmspace = vm; 405 PROC_VMSPACE_UNLOCK(p); 406 pmap_activate(td); 407 } 408 pmap_remove_pages(vmspace_pmap(vm)); 409 /* Switch now since this proc will free vmspace */ 410 PROC_VMSPACE_LOCK(p); 411 p->p_vmspace = &vmspace0; 412 PROC_VMSPACE_UNLOCK(p); 413 pmap_activate(td); 414 vmspace_dofree(vm); 415 } 416 vmspace_container_reset(p); 417} 418 419/* Acquire reference to vmspace owned by another process. */ 420 421struct vmspace * 422vmspace_acquire_ref(struct proc *p) 423{ 424 struct vmspace *vm; 425 int refcnt; 426 427 PROC_VMSPACE_LOCK(p); 428 vm = p->p_vmspace; 429 if (vm == NULL) { 430 PROC_VMSPACE_UNLOCK(p); 431 return (NULL); 432 } 433 do { 434 refcnt = vm->vm_refcnt; 435 if (refcnt <= 0) { /* Avoid 0->1 transition */ 436 PROC_VMSPACE_UNLOCK(p); 437 return (NULL); 438 } 439 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 440 if (vm != p->p_vmspace) { 441 PROC_VMSPACE_UNLOCK(p); 442 vmspace_free(vm); 443 return (NULL); 444 } 445 PROC_VMSPACE_UNLOCK(p); 446 return (vm); 447} 448 449void 450_vm_map_lock(vm_map_t map, const char *file, int line) 451{ 452 453 if (map->system_map) 454 mtx_lock_flags_(&map->system_mtx, 0, file, line); 455 else 456 sx_xlock_(&map->lock, file, line); 457 map->timestamp++; 458} 459 460static void 461vm_map_process_deferred(void) 462{ 463 struct thread *td; 464 vm_map_entry_t entry, next; 465 vm_object_t object; 466 467 td = curthread; 468 entry = td->td_map_def_user; 469 td->td_map_def_user = NULL; 470 while (entry != NULL) { 471 next = entry->next; 472 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 473 /* 474 * Decrement the object's writemappings and 475 * possibly the vnode's v_writecount. 476 */ 477 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 478 ("Submap with writecount")); 479 object = entry->object.vm_object; 480 KASSERT(object != NULL, ("No object for writecount")); 481 vnode_pager_release_writecount(object, entry->start, 482 entry->end); 483 } 484 vm_map_entry_deallocate(entry, FALSE); 485 entry = next; 486 } 487} 488 489void 490_vm_map_unlock(vm_map_t map, const char *file, int line) 491{ 492 493 if (map->system_map) 494 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 495 else { 496 sx_xunlock_(&map->lock, file, line); 497 vm_map_process_deferred(); 498 } 499} 500 501void 502_vm_map_lock_read(vm_map_t map, const char *file, int line) 503{ 504 505 if (map->system_map) 506 mtx_lock_flags_(&map->system_mtx, 0, file, line); 507 else 508 sx_slock_(&map->lock, file, line); 509} 510 511void 512_vm_map_unlock_read(vm_map_t map, const char *file, int line) 513{ 514 515 if (map->system_map) 516 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 517 else { 518 sx_sunlock_(&map->lock, file, line); 519 vm_map_process_deferred(); 520 } 521} 522 523int 524_vm_map_trylock(vm_map_t map, const char *file, int line) 525{ 526 int error; 527 528 error = map->system_map ? 529 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 530 !sx_try_xlock_(&map->lock, file, line); 531 if (error == 0) 532 map->timestamp++; 533 return (error == 0); 534} 535 536int 537_vm_map_trylock_read(vm_map_t map, const char *file, int line) 538{ 539 int error; 540 541 error = map->system_map ? 542 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 543 !sx_try_slock_(&map->lock, file, line); 544 return (error == 0); 545} 546 547/* 548 * _vm_map_lock_upgrade: [ internal use only ] 549 * 550 * Tries to upgrade a read (shared) lock on the specified map to a write 551 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 552 * non-zero value if the upgrade fails. If the upgrade fails, the map is 553 * returned without a read or write lock held. 554 * 555 * Requires that the map be read locked. 556 */ 557int 558_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 559{ 560 unsigned int last_timestamp; 561 562 if (map->system_map) { 563 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 564 } else { 565 if (!sx_try_upgrade_(&map->lock, file, line)) { 566 last_timestamp = map->timestamp; 567 sx_sunlock_(&map->lock, file, line); 568 vm_map_process_deferred(); 569 /* 570 * If the map's timestamp does not change while the 571 * map is unlocked, then the upgrade succeeds. 572 */ 573 sx_xlock_(&map->lock, file, line); 574 if (last_timestamp != map->timestamp) { 575 sx_xunlock_(&map->lock, file, line); 576 return (1); 577 } 578 } 579 } 580 map->timestamp++; 581 return (0); 582} 583 584void 585_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 586{ 587 588 if (map->system_map) { 589 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 590 } else 591 sx_downgrade_(&map->lock, file, line); 592} 593 594/* 595 * vm_map_locked: 596 * 597 * Returns a non-zero value if the caller holds a write (exclusive) lock 598 * on the specified map and the value "0" otherwise. 599 */ 600int 601vm_map_locked(vm_map_t map) 602{ 603 604 if (map->system_map) 605 return (mtx_owned(&map->system_mtx)); 606 else 607 return (sx_xlocked(&map->lock)); 608} 609 610#ifdef INVARIANTS 611static void 612_vm_map_assert_locked(vm_map_t map, const char *file, int line) 613{ 614 615 if (map->system_map) 616 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 617 else 618 sx_assert_(&map->lock, SA_XLOCKED, file, line); 619} 620 621#define VM_MAP_ASSERT_LOCKED(map) \ 622 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 623#else 624#define VM_MAP_ASSERT_LOCKED(map) 625#endif 626 627/* 628 * _vm_map_unlock_and_wait: 629 * 630 * Atomically releases the lock on the specified map and puts the calling 631 * thread to sleep. The calling thread will remain asleep until either 632 * vm_map_wakeup() is performed on the map or the specified timeout is 633 * exceeded. 634 * 635 * WARNING! This function does not perform deferred deallocations of 636 * objects and map entries. Therefore, the calling thread is expected to 637 * reacquire the map lock after reawakening and later perform an ordinary 638 * unlock operation, such as vm_map_unlock(), before completing its 639 * operation on the map. 640 */ 641int 642_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 643{ 644 645 mtx_lock(&map_sleep_mtx); 646 if (map->system_map) 647 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 648 else 649 sx_xunlock_(&map->lock, file, line); 650 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 651 timo)); 652} 653 654/* 655 * vm_map_wakeup: 656 * 657 * Awaken any threads that have slept on the map using 658 * vm_map_unlock_and_wait(). 659 */ 660void 661vm_map_wakeup(vm_map_t map) 662{ 663 664 /* 665 * Acquire and release map_sleep_mtx to prevent a wakeup() 666 * from being performed (and lost) between the map unlock 667 * and the msleep() in _vm_map_unlock_and_wait(). 668 */ 669 mtx_lock(&map_sleep_mtx); 670 mtx_unlock(&map_sleep_mtx); 671 wakeup(&map->root); 672} 673 674void 675vm_map_busy(vm_map_t map) 676{ 677 678 VM_MAP_ASSERT_LOCKED(map); 679 map->busy++; 680} 681 682void 683vm_map_unbusy(vm_map_t map) 684{ 685 686 VM_MAP_ASSERT_LOCKED(map); 687 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 688 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 689 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 690 wakeup(&map->busy); 691 } 692} 693 694void 695vm_map_wait_busy(vm_map_t map) 696{ 697 698 VM_MAP_ASSERT_LOCKED(map); 699 while (map->busy) { 700 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 701 if (map->system_map) 702 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 703 else 704 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 705 } 706 map->timestamp++; 707} 708 709long 710vmspace_resident_count(struct vmspace *vmspace) 711{ 712 return pmap_resident_count(vmspace_pmap(vmspace)); 713} 714 715/* 716 * vm_map_create: 717 * 718 * Creates and returns a new empty VM map with 719 * the given physical map structure, and having 720 * the given lower and upper address bounds. 721 */ 722vm_map_t 723vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 724{ 725 vm_map_t result; 726 727 result = uma_zalloc(mapzone, M_WAITOK); 728 CTR1(KTR_VM, "vm_map_create: %p", result); 729 _vm_map_init(result, pmap, min, max); 730 return (result); 731} 732 733/* 734 * Initialize an existing vm_map structure 735 * such as that in the vmspace structure. 736 */ 737static void 738_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 739{ 740 741 map->header.next = map->header.prev = &map->header; 742 map->needs_wakeup = FALSE; 743 map->system_map = 0; 744 map->pmap = pmap; 745 map->min_offset = min; 746 map->max_offset = max; 747 map->flags = 0; 748 map->root = NULL; 749 map->timestamp = 0; 750 map->busy = 0; 751} 752 753void 754vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 755{ 756 757 _vm_map_init(map, pmap, min, max); 758 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 759 sx_init(&map->lock, "user map"); 760} 761 762/* 763 * vm_map_entry_dispose: [ internal use only ] 764 * 765 * Inverse of vm_map_entry_create. 766 */ 767static void 768vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 769{ 770 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 771} 772 773/* 774 * vm_map_entry_create: [ internal use only ] 775 * 776 * Allocates a VM map entry for insertion. 777 * No entry fields are filled in. 778 */ 779static vm_map_entry_t 780vm_map_entry_create(vm_map_t map) 781{ 782 vm_map_entry_t new_entry; 783 784 if (map->system_map) 785 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 786 else 787 new_entry = uma_zalloc(mapentzone, M_WAITOK); 788 if (new_entry == NULL) 789 panic("vm_map_entry_create: kernel resources exhausted"); 790 return (new_entry); 791} 792 793/* 794 * vm_map_entry_set_behavior: 795 * 796 * Set the expected access behavior, either normal, random, or 797 * sequential. 798 */ 799static inline void 800vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 801{ 802 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 803 (behavior & MAP_ENTRY_BEHAV_MASK); 804} 805 806/* 807 * vm_map_entry_set_max_free: 808 * 809 * Set the max_free field in a vm_map_entry. 810 */ 811static inline void 812vm_map_entry_set_max_free(vm_map_entry_t entry) 813{ 814 815 entry->max_free = entry->adj_free; 816 if (entry->left != NULL && entry->left->max_free > entry->max_free) 817 entry->max_free = entry->left->max_free; 818 if (entry->right != NULL && entry->right->max_free > entry->max_free) 819 entry->max_free = entry->right->max_free; 820} 821 822/* 823 * vm_map_entry_splay: 824 * 825 * The Sleator and Tarjan top-down splay algorithm with the 826 * following variation. Max_free must be computed bottom-up, so 827 * on the downward pass, maintain the left and right spines in 828 * reverse order. Then, make a second pass up each side to fix 829 * the pointers and compute max_free. The time bound is O(log n) 830 * amortized. 831 * 832 * The new root is the vm_map_entry containing "addr", or else an 833 * adjacent entry (lower or higher) if addr is not in the tree. 834 * 835 * The map must be locked, and leaves it so. 836 * 837 * Returns: the new root. 838 */ 839static vm_map_entry_t 840vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 841{ 842 vm_map_entry_t llist, rlist; 843 vm_map_entry_t ltree, rtree; 844 vm_map_entry_t y; 845 846 /* Special case of empty tree. */ 847 if (root == NULL) 848 return (root); 849 850 /* 851 * Pass One: Splay down the tree until we find addr or a NULL 852 * pointer where addr would go. llist and rlist are the two 853 * sides in reverse order (bottom-up), with llist linked by 854 * the right pointer and rlist linked by the left pointer in 855 * the vm_map_entry. Wait until Pass Two to set max_free on 856 * the two spines. 857 */ 858 llist = NULL; 859 rlist = NULL; 860 for (;;) { 861 /* root is never NULL in here. */ 862 if (addr < root->start) { 863 y = root->left; 864 if (y == NULL) 865 break; 866 if (addr < y->start && y->left != NULL) { 867 /* Rotate right and put y on rlist. */ 868 root->left = y->right; 869 y->right = root; 870 vm_map_entry_set_max_free(root); 871 root = y->left; 872 y->left = rlist; 873 rlist = y; 874 } else { 875 /* Put root on rlist. */ 876 root->left = rlist; 877 rlist = root; 878 root = y; 879 } 880 } else if (addr >= root->end) { 881 y = root->right; 882 if (y == NULL) 883 break; 884 if (addr >= y->end && y->right != NULL) { 885 /* Rotate left and put y on llist. */ 886 root->right = y->left; 887 y->left = root; 888 vm_map_entry_set_max_free(root); 889 root = y->right; 890 y->right = llist; 891 llist = y; 892 } else { 893 /* Put root on llist. */ 894 root->right = llist; 895 llist = root; 896 root = y; 897 } 898 } else 899 break; 900 } 901 902 /* 903 * Pass Two: Walk back up the two spines, flip the pointers 904 * and set max_free. The subtrees of the root go at the 905 * bottom of llist and rlist. 906 */ 907 ltree = root->left; 908 while (llist != NULL) { 909 y = llist->right; 910 llist->right = ltree; 911 vm_map_entry_set_max_free(llist); 912 ltree = llist; 913 llist = y; 914 } 915 rtree = root->right; 916 while (rlist != NULL) { 917 y = rlist->left; 918 rlist->left = rtree; 919 vm_map_entry_set_max_free(rlist); 920 rtree = rlist; 921 rlist = y; 922 } 923 924 /* 925 * Final assembly: add ltree and rtree as subtrees of root. 926 */ 927 root->left = ltree; 928 root->right = rtree; 929 vm_map_entry_set_max_free(root); 930 931 return (root); 932} 933 934/* 935 * vm_map_entry_{un,}link: 936 * 937 * Insert/remove entries from maps. 938 */ 939static void 940vm_map_entry_link(vm_map_t map, 941 vm_map_entry_t after_where, 942 vm_map_entry_t entry) 943{ 944 945 CTR4(KTR_VM, 946 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 947 map->nentries, entry, after_where); 948 VM_MAP_ASSERT_LOCKED(map); 949 KASSERT(after_where == &map->header || 950 after_where->end <= entry->start, 951 ("vm_map_entry_link: prev end %jx new start %jx overlap", 952 (uintmax_t)after_where->end, (uintmax_t)entry->start)); 953 KASSERT(after_where->next == &map->header || 954 entry->end <= after_where->next->start, 955 ("vm_map_entry_link: new end %jx next start %jx overlap", 956 (uintmax_t)entry->end, (uintmax_t)after_where->next->start)); 957 958 map->nentries++; 959 entry->prev = after_where; 960 entry->next = after_where->next; 961 entry->next->prev = entry; 962 after_where->next = entry; 963 964 if (after_where != &map->header) { 965 if (after_where != map->root) 966 vm_map_entry_splay(after_where->start, map->root); 967 entry->right = after_where->right; 968 entry->left = after_where; 969 after_where->right = NULL; 970 after_where->adj_free = entry->start - after_where->end; 971 vm_map_entry_set_max_free(after_where); 972 } else { 973 entry->right = map->root; 974 entry->left = NULL; 975 } 976 entry->adj_free = (entry->next == &map->header ? map->max_offset : 977 entry->next->start) - entry->end; 978 vm_map_entry_set_max_free(entry); 979 map->root = entry; 980} 981 982static void 983vm_map_entry_unlink(vm_map_t map, 984 vm_map_entry_t entry) 985{ 986 vm_map_entry_t next, prev, root; 987 988 VM_MAP_ASSERT_LOCKED(map); 989 if (entry != map->root) 990 vm_map_entry_splay(entry->start, map->root); 991 if (entry->left == NULL) 992 root = entry->right; 993 else { 994 root = vm_map_entry_splay(entry->start, entry->left); 995 root->right = entry->right; 996 root->adj_free = (entry->next == &map->header ? map->max_offset : 997 entry->next->start) - root->end; 998 vm_map_entry_set_max_free(root); 999 } 1000 map->root = root; 1001 1002 prev = entry->prev; 1003 next = entry->next; 1004 next->prev = prev; 1005 prev->next = next; 1006 map->nentries--; 1007 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1008 map->nentries, entry); 1009} 1010 1011/* 1012 * vm_map_entry_resize_free: 1013 * 1014 * Recompute the amount of free space following a vm_map_entry 1015 * and propagate that value up the tree. Call this function after 1016 * resizing a map entry in-place, that is, without a call to 1017 * vm_map_entry_link() or _unlink(). 1018 * 1019 * The map must be locked, and leaves it so. 1020 */ 1021static void 1022vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1023{ 1024 1025 /* 1026 * Using splay trees without parent pointers, propagating 1027 * max_free up the tree is done by moving the entry to the 1028 * root and making the change there. 1029 */ 1030 if (entry != map->root) 1031 map->root = vm_map_entry_splay(entry->start, map->root); 1032 1033 entry->adj_free = (entry->next == &map->header ? map->max_offset : 1034 entry->next->start) - entry->end; 1035 vm_map_entry_set_max_free(entry); 1036} 1037 1038/* 1039 * vm_map_lookup_entry: [ internal use only ] 1040 * 1041 * Finds the map entry containing (or 1042 * immediately preceding) the specified address 1043 * in the given map; the entry is returned 1044 * in the "entry" parameter. The boolean 1045 * result indicates whether the address is 1046 * actually contained in the map. 1047 */ 1048boolean_t 1049vm_map_lookup_entry( 1050 vm_map_t map, 1051 vm_offset_t address, 1052 vm_map_entry_t *entry) /* OUT */ 1053{ 1054 vm_map_entry_t cur; 1055 boolean_t locked; 1056 1057 /* 1058 * If the map is empty, then the map entry immediately preceding 1059 * "address" is the map's header. 1060 */ 1061 cur = map->root; 1062 if (cur == NULL) 1063 *entry = &map->header; 1064 else if (address >= cur->start && cur->end > address) { 1065 *entry = cur; 1066 return (TRUE); 1067 } else if ((locked = vm_map_locked(map)) || 1068 sx_try_upgrade(&map->lock)) { 1069 /* 1070 * Splay requires a write lock on the map. However, it only 1071 * restructures the binary search tree; it does not otherwise 1072 * change the map. Thus, the map's timestamp need not change 1073 * on a temporary upgrade. 1074 */ 1075 map->root = cur = vm_map_entry_splay(address, cur); 1076 if (!locked) 1077 sx_downgrade(&map->lock); 1078 1079 /* 1080 * If "address" is contained within a map entry, the new root 1081 * is that map entry. Otherwise, the new root is a map entry 1082 * immediately before or after "address". 1083 */ 1084 if (address >= cur->start) { 1085 *entry = cur; 1086 if (cur->end > address) 1087 return (TRUE); 1088 } else 1089 *entry = cur->prev; 1090 } else 1091 /* 1092 * Since the map is only locked for read access, perform a 1093 * standard binary search tree lookup for "address". 1094 */ 1095 for (;;) { 1096 if (address < cur->start) { 1097 if (cur->left == NULL) { 1098 *entry = cur->prev; 1099 break; 1100 } 1101 cur = cur->left; 1102 } else if (cur->end > address) { 1103 *entry = cur; 1104 return (TRUE); 1105 } else { 1106 if (cur->right == NULL) { 1107 *entry = cur; 1108 break; 1109 } 1110 cur = cur->right; 1111 } 1112 } 1113 return (FALSE); 1114} 1115 1116/* 1117 * vm_map_insert: 1118 * 1119 * Inserts the given whole VM object into the target 1120 * map at the specified address range. The object's 1121 * size should match that of the address range. 1122 * 1123 * Requires that the map be locked, and leaves it so. 1124 * 1125 * If object is non-NULL, ref count must be bumped by caller 1126 * prior to making call to account for the new entry. 1127 */ 1128int 1129vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1130 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1131 int cow) 1132{ 1133 vm_map_entry_t new_entry; 1134 vm_map_entry_t prev_entry; 1135 vm_map_entry_t temp_entry; 1136 vm_eflags_t protoeflags; 1137 struct ucred *cred; 1138 vm_inherit_t inheritance; 1139 boolean_t charge_prev_obj; 1140 1141 VM_MAP_ASSERT_LOCKED(map); 1142 1143 /* 1144 * Check that the start and end points are not bogus. 1145 */ 1146 if ((start < map->min_offset) || (end > map->max_offset) || 1147 (start >= end)) 1148 return (KERN_INVALID_ADDRESS); 1149 1150 /* 1151 * Find the entry prior to the proposed starting address; if it's part 1152 * of an existing entry, this range is bogus. 1153 */ 1154 if (vm_map_lookup_entry(map, start, &temp_entry)) 1155 return (KERN_NO_SPACE); 1156 1157 prev_entry = temp_entry; 1158 1159 /* 1160 * Assert that the next entry doesn't overlap the end point. 1161 */ 1162 if ((prev_entry->next != &map->header) && 1163 (prev_entry->next->start < end)) 1164 return (KERN_NO_SPACE); 1165 1166 protoeflags = 0; 1167 charge_prev_obj = FALSE; 1168 1169 if (cow & MAP_COPY_ON_WRITE) 1170 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1171 1172 if (cow & MAP_NOFAULT) { 1173 protoeflags |= MAP_ENTRY_NOFAULT; 1174 1175 KASSERT(object == NULL, 1176 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1177 } 1178 if (cow & MAP_DISABLE_SYNCER) 1179 protoeflags |= MAP_ENTRY_NOSYNC; 1180 if (cow & MAP_DISABLE_COREDUMP) 1181 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1182 if (cow & MAP_VN_WRITECOUNT) 1183 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1184 if (cow & MAP_INHERIT_SHARE) 1185 inheritance = VM_INHERIT_SHARE; 1186 else 1187 inheritance = VM_INHERIT_DEFAULT; 1188 1189 cred = NULL; 1190 KASSERT((object != kmem_object && object != kernel_object) || 1191 ((object == kmem_object || object == kernel_object) && 1192 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1193 ("kmem or kernel object and cow")); 1194 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1195 goto charged; 1196 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1197 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1198 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1199 return (KERN_RESOURCE_SHORTAGE); 1200 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || 1201 object->cred == NULL, 1202 ("OVERCOMMIT: vm_map_insert o %p", object)); 1203 cred = curthread->td_ucred; 1204 crhold(cred); 1205 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1206 charge_prev_obj = TRUE; 1207 } 1208 1209charged: 1210 /* Expand the kernel pmap, if necessary. */ 1211 if (map == kernel_map && end > kernel_vm_end) 1212 pmap_growkernel(end); 1213 if (object != NULL) { 1214 /* 1215 * OBJ_ONEMAPPING must be cleared unless this mapping 1216 * is trivially proven to be the only mapping for any 1217 * of the object's pages. (Object granularity 1218 * reference counting is insufficient to recognize 1219 * aliases with precision.) 1220 */ 1221 VM_OBJECT_WLOCK(object); 1222 if (object->ref_count > 1 || object->shadow_count != 0) 1223 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1224 VM_OBJECT_WUNLOCK(object); 1225 } 1226 else if ((prev_entry != &map->header) && 1227 (prev_entry->eflags == protoeflags) && 1228 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 && 1229 (prev_entry->end == start) && 1230 (prev_entry->wired_count == 0) && 1231 (prev_entry->cred == cred || 1232 (prev_entry->object.vm_object != NULL && 1233 (prev_entry->object.vm_object->cred == cred))) && 1234 vm_object_coalesce(prev_entry->object.vm_object, 1235 prev_entry->offset, 1236 (vm_size_t)(prev_entry->end - prev_entry->start), 1237 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1238 /* 1239 * We were able to extend the object. Determine if we 1240 * can extend the previous map entry to include the 1241 * new range as well. 1242 */ 1243 if ((prev_entry->inheritance == inheritance) && 1244 (prev_entry->protection == prot) && 1245 (prev_entry->max_protection == max)) { 1246 map->size += (end - prev_entry->end); 1247 prev_entry->end = end; 1248 vm_map_entry_resize_free(map, prev_entry); 1249 vm_map_simplify_entry(map, prev_entry); 1250 if (cred != NULL) 1251 crfree(cred); 1252 return (KERN_SUCCESS); 1253 } 1254 1255 /* 1256 * If we can extend the object but cannot extend the 1257 * map entry, we have to create a new map entry. We 1258 * must bump the ref count on the extended object to 1259 * account for it. object may be NULL. 1260 */ 1261 object = prev_entry->object.vm_object; 1262 offset = prev_entry->offset + 1263 (prev_entry->end - prev_entry->start); 1264 vm_object_reference(object); 1265 if (cred != NULL && object != NULL && object->cred != NULL && 1266 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1267 /* Object already accounts for this uid. */ 1268 crfree(cred); 1269 cred = NULL; 1270 } 1271 } 1272 1273 /* 1274 * NOTE: if conditionals fail, object can be NULL here. This occurs 1275 * in things like the buffer map where we manage kva but do not manage 1276 * backing objects. 1277 */ 1278 1279 /* 1280 * Create a new entry 1281 */ 1282 new_entry = vm_map_entry_create(map); 1283 new_entry->start = start; 1284 new_entry->end = end; 1285 new_entry->cred = NULL; 1286 1287 new_entry->eflags = protoeflags; 1288 new_entry->object.vm_object = object; 1289 new_entry->offset = offset; 1290 new_entry->avail_ssize = 0; 1291 1292 new_entry->inheritance = inheritance; 1293 new_entry->protection = prot; 1294 new_entry->max_protection = max; 1295 new_entry->wired_count = 0; 1296 new_entry->wiring_thread = NULL; 1297 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1298 new_entry->next_read = OFF_TO_IDX(offset); 1299 1300 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1301 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1302 new_entry->cred = cred; 1303 1304 /* 1305 * Insert the new entry into the list 1306 */ 1307 vm_map_entry_link(map, prev_entry, new_entry); 1308 map->size += new_entry->end - new_entry->start; 1309 1310 /* 1311 * It may be possible to merge the new entry with the next and/or 1312 * previous entries. However, due to MAP_STACK_* being a hack, a 1313 * panic can result from merging such entries. 1314 */ 1315 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0) 1316 vm_map_simplify_entry(map, new_entry); 1317 1318 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1319 vm_map_pmap_enter(map, start, prot, 1320 object, OFF_TO_IDX(offset), end - start, 1321 cow & MAP_PREFAULT_PARTIAL); 1322 } 1323 1324 return (KERN_SUCCESS); 1325} 1326 1327/* 1328 * vm_map_findspace: 1329 * 1330 * Find the first fit (lowest VM address) for "length" free bytes 1331 * beginning at address >= start in the given map. 1332 * 1333 * In a vm_map_entry, "adj_free" is the amount of free space 1334 * adjacent (higher address) to this entry, and "max_free" is the 1335 * maximum amount of contiguous free space in its subtree. This 1336 * allows finding a free region in one path down the tree, so 1337 * O(log n) amortized with splay trees. 1338 * 1339 * The map must be locked, and leaves it so. 1340 * 1341 * Returns: 0 on success, and starting address in *addr, 1342 * 1 if insufficient space. 1343 */ 1344int 1345vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1346 vm_offset_t *addr) /* OUT */ 1347{ 1348 vm_map_entry_t entry; 1349 vm_offset_t st; 1350 1351 /* 1352 * Request must fit within min/max VM address and must avoid 1353 * address wrap. 1354 */ 1355 if (start < map->min_offset) 1356 start = map->min_offset; 1357 if (start + length > map->max_offset || start + length < start) 1358 return (1); 1359 1360 /* Empty tree means wide open address space. */ 1361 if (map->root == NULL) { 1362 *addr = start; 1363 return (0); 1364 } 1365 1366 /* 1367 * After splay, if start comes before root node, then there 1368 * must be a gap from start to the root. 1369 */ 1370 map->root = vm_map_entry_splay(start, map->root); 1371 if (start + length <= map->root->start) { 1372 *addr = start; 1373 return (0); 1374 } 1375 1376 /* 1377 * Root is the last node that might begin its gap before 1378 * start, and this is the last comparison where address 1379 * wrap might be a problem. 1380 */ 1381 st = (start > map->root->end) ? start : map->root->end; 1382 if (length <= map->root->end + map->root->adj_free - st) { 1383 *addr = st; 1384 return (0); 1385 } 1386 1387 /* With max_free, can immediately tell if no solution. */ 1388 entry = map->root->right; 1389 if (entry == NULL || length > entry->max_free) 1390 return (1); 1391 1392 /* 1393 * Search the right subtree in the order: left subtree, root, 1394 * right subtree (first fit). The previous splay implies that 1395 * all regions in the right subtree have addresses > start. 1396 */ 1397 while (entry != NULL) { 1398 if (entry->left != NULL && entry->left->max_free >= length) 1399 entry = entry->left; 1400 else if (entry->adj_free >= length) { 1401 *addr = entry->end; 1402 return (0); 1403 } else 1404 entry = entry->right; 1405 } 1406 1407 /* Can't get here, so panic if we do. */ 1408 panic("vm_map_findspace: max_free corrupt"); 1409} 1410 1411int 1412vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1413 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1414 vm_prot_t max, int cow) 1415{ 1416 vm_offset_t end; 1417 int result; 1418 1419 end = start + length; 1420 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1421 object == NULL, 1422 ("vm_map_fixed: non-NULL backing object for stack")); 1423 vm_map_lock(map); 1424 VM_MAP_RANGE_CHECK(map, start, end); 1425 if ((cow & MAP_CHECK_EXCL) == 0) 1426 vm_map_delete(map, start, end); 1427 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1428 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1429 prot, max, cow); 1430 } else { 1431 result = vm_map_insert(map, object, offset, start, end, 1432 prot, max, cow); 1433 } 1434 vm_map_unlock(map); 1435 return (result); 1436} 1437 1438/* 1439 * vm_map_find finds an unallocated region in the target address 1440 * map with the given length. The search is defined to be 1441 * first-fit from the specified address; the region found is 1442 * returned in the same parameter. 1443 * 1444 * If object is non-NULL, ref count must be bumped by caller 1445 * prior to making call to account for the new entry. 1446 */ 1447int 1448vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1449 vm_offset_t *addr, /* IN/OUT */ 1450 vm_size_t length, vm_offset_t max_addr, int find_space, 1451 vm_prot_t prot, vm_prot_t max, int cow) 1452{ 1453 vm_offset_t alignment, initial_addr, start; 1454 int result; 1455 1456 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1457 object == NULL, 1458 ("vm_map_find: non-NULL backing object for stack")); 1459 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1460 (object->flags & OBJ_COLORED) == 0)) 1461 find_space = VMFS_ANY_SPACE; 1462 if (find_space >> 8 != 0) { 1463 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1464 alignment = (vm_offset_t)1 << (find_space >> 8); 1465 } else 1466 alignment = 0; 1467 initial_addr = *addr; 1468again: 1469 start = initial_addr; 1470 vm_map_lock(map); 1471 do { 1472 if (find_space != VMFS_NO_SPACE) { 1473 if (vm_map_findspace(map, start, length, addr) || 1474 (max_addr != 0 && *addr + length > max_addr)) { 1475 vm_map_unlock(map); 1476 if (find_space == VMFS_OPTIMAL_SPACE) { 1477 find_space = VMFS_ANY_SPACE; 1478 goto again; 1479 } 1480 return (KERN_NO_SPACE); 1481 } 1482 switch (find_space) { 1483 case VMFS_SUPER_SPACE: 1484 case VMFS_OPTIMAL_SPACE: 1485 pmap_align_superpage(object, offset, addr, 1486 length); 1487 break; 1488 case VMFS_ANY_SPACE: 1489 break; 1490 default: 1491 if ((*addr & (alignment - 1)) != 0) { 1492 *addr &= ~(alignment - 1); 1493 *addr += alignment; 1494 } 1495 break; 1496 } 1497 1498 start = *addr; 1499 } 1500 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1501 result = vm_map_stack_locked(map, start, length, 1502 sgrowsiz, prot, max, cow); 1503 } else { 1504 result = vm_map_insert(map, object, offset, start, 1505 start + length, prot, max, cow); 1506 } 1507 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1508 find_space != VMFS_ANY_SPACE); 1509 vm_map_unlock(map); 1510 return (result); 1511} 1512 1513/* 1514 * vm_map_simplify_entry: 1515 * 1516 * Simplify the given map entry by merging with either neighbor. This 1517 * routine also has the ability to merge with both neighbors. 1518 * 1519 * The map must be locked. 1520 * 1521 * This routine guarentees that the passed entry remains valid (though 1522 * possibly extended). When merging, this routine may delete one or 1523 * both neighbors. 1524 */ 1525void 1526vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1527{ 1528 vm_map_entry_t next, prev; 1529 vm_size_t prevsize, esize; 1530 1531 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1532 return; 1533 1534 prev = entry->prev; 1535 if (prev != &map->header) { 1536 prevsize = prev->end - prev->start; 1537 if ( (prev->end == entry->start) && 1538 (prev->object.vm_object == entry->object.vm_object) && 1539 (!prev->object.vm_object || 1540 (prev->offset + prevsize == entry->offset)) && 1541 (prev->eflags == entry->eflags) && 1542 (prev->protection == entry->protection) && 1543 (prev->max_protection == entry->max_protection) && 1544 (prev->inheritance == entry->inheritance) && 1545 (prev->wired_count == entry->wired_count) && 1546 (prev->cred == entry->cred)) { 1547 vm_map_entry_unlink(map, prev); 1548 entry->start = prev->start; 1549 entry->offset = prev->offset; 1550 if (entry->prev != &map->header) 1551 vm_map_entry_resize_free(map, entry->prev); 1552 1553 /* 1554 * If the backing object is a vnode object, 1555 * vm_object_deallocate() calls vrele(). 1556 * However, vrele() does not lock the vnode 1557 * because the vnode has additional 1558 * references. Thus, the map lock can be kept 1559 * without causing a lock-order reversal with 1560 * the vnode lock. 1561 * 1562 * Since we count the number of virtual page 1563 * mappings in object->un_pager.vnp.writemappings, 1564 * the writemappings value should not be adjusted 1565 * when the entry is disposed of. 1566 */ 1567 if (prev->object.vm_object) 1568 vm_object_deallocate(prev->object.vm_object); 1569 if (prev->cred != NULL) 1570 crfree(prev->cred); 1571 vm_map_entry_dispose(map, prev); 1572 } 1573 } 1574 1575 next = entry->next; 1576 if (next != &map->header) { 1577 esize = entry->end - entry->start; 1578 if ((entry->end == next->start) && 1579 (next->object.vm_object == entry->object.vm_object) && 1580 (!entry->object.vm_object || 1581 (entry->offset + esize == next->offset)) && 1582 (next->eflags == entry->eflags) && 1583 (next->protection == entry->protection) && 1584 (next->max_protection == entry->max_protection) && 1585 (next->inheritance == entry->inheritance) && 1586 (next->wired_count == entry->wired_count) && 1587 (next->cred == entry->cred)) { 1588 vm_map_entry_unlink(map, next); 1589 entry->end = next->end; 1590 vm_map_entry_resize_free(map, entry); 1591 1592 /* 1593 * See comment above. 1594 */ 1595 if (next->object.vm_object) 1596 vm_object_deallocate(next->object.vm_object); 1597 if (next->cred != NULL) 1598 crfree(next->cred); 1599 vm_map_entry_dispose(map, next); 1600 } 1601 } 1602} 1603/* 1604 * vm_map_clip_start: [ internal use only ] 1605 * 1606 * Asserts that the given entry begins at or after 1607 * the specified address; if necessary, 1608 * it splits the entry into two. 1609 */ 1610#define vm_map_clip_start(map, entry, startaddr) \ 1611{ \ 1612 if (startaddr > entry->start) \ 1613 _vm_map_clip_start(map, entry, startaddr); \ 1614} 1615 1616/* 1617 * This routine is called only when it is known that 1618 * the entry must be split. 1619 */ 1620static void 1621_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1622{ 1623 vm_map_entry_t new_entry; 1624 1625 VM_MAP_ASSERT_LOCKED(map); 1626 1627 /* 1628 * Split off the front portion -- note that we must insert the new 1629 * entry BEFORE this one, so that this entry has the specified 1630 * starting address. 1631 */ 1632 vm_map_simplify_entry(map, entry); 1633 1634 /* 1635 * If there is no object backing this entry, we might as well create 1636 * one now. If we defer it, an object can get created after the map 1637 * is clipped, and individual objects will be created for the split-up 1638 * map. This is a bit of a hack, but is also about the best place to 1639 * put this improvement. 1640 */ 1641 if (entry->object.vm_object == NULL && !map->system_map) { 1642 vm_object_t object; 1643 object = vm_object_allocate(OBJT_DEFAULT, 1644 atop(entry->end - entry->start)); 1645 entry->object.vm_object = object; 1646 entry->offset = 0; 1647 if (entry->cred != NULL) { 1648 object->cred = entry->cred; 1649 object->charge = entry->end - entry->start; 1650 entry->cred = NULL; 1651 } 1652 } else if (entry->object.vm_object != NULL && 1653 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1654 entry->cred != NULL) { 1655 VM_OBJECT_WLOCK(entry->object.vm_object); 1656 KASSERT(entry->object.vm_object->cred == NULL, 1657 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1658 entry->object.vm_object->cred = entry->cred; 1659 entry->object.vm_object->charge = entry->end - entry->start; 1660 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1661 entry->cred = NULL; 1662 } 1663 1664 new_entry = vm_map_entry_create(map); 1665 *new_entry = *entry; 1666 1667 new_entry->end = start; 1668 entry->offset += (start - entry->start); 1669 entry->start = start; 1670 if (new_entry->cred != NULL) 1671 crhold(entry->cred); 1672 1673 vm_map_entry_link(map, entry->prev, new_entry); 1674 1675 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1676 vm_object_reference(new_entry->object.vm_object); 1677 /* 1678 * The object->un_pager.vnp.writemappings for the 1679 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1680 * kept as is here. The virtual pages are 1681 * re-distributed among the clipped entries, so the sum is 1682 * left the same. 1683 */ 1684 } 1685} 1686 1687/* 1688 * vm_map_clip_end: [ internal use only ] 1689 * 1690 * Asserts that the given entry ends at or before 1691 * the specified address; if necessary, 1692 * it splits the entry into two. 1693 */ 1694#define vm_map_clip_end(map, entry, endaddr) \ 1695{ \ 1696 if ((endaddr) < (entry->end)) \ 1697 _vm_map_clip_end((map), (entry), (endaddr)); \ 1698} 1699 1700/* 1701 * This routine is called only when it is known that 1702 * the entry must be split. 1703 */ 1704static void 1705_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1706{ 1707 vm_map_entry_t new_entry; 1708 1709 VM_MAP_ASSERT_LOCKED(map); 1710 1711 /* 1712 * If there is no object backing this entry, we might as well create 1713 * one now. If we defer it, an object can get created after the map 1714 * is clipped, and individual objects will be created for the split-up 1715 * map. This is a bit of a hack, but is also about the best place to 1716 * put this improvement. 1717 */ 1718 if (entry->object.vm_object == NULL && !map->system_map) { 1719 vm_object_t object; 1720 object = vm_object_allocate(OBJT_DEFAULT, 1721 atop(entry->end - entry->start)); 1722 entry->object.vm_object = object; 1723 entry->offset = 0; 1724 if (entry->cred != NULL) { 1725 object->cred = entry->cred; 1726 object->charge = entry->end - entry->start; 1727 entry->cred = NULL; 1728 } 1729 } else if (entry->object.vm_object != NULL && 1730 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1731 entry->cred != NULL) { 1732 VM_OBJECT_WLOCK(entry->object.vm_object); 1733 KASSERT(entry->object.vm_object->cred == NULL, 1734 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1735 entry->object.vm_object->cred = entry->cred; 1736 entry->object.vm_object->charge = entry->end - entry->start; 1737 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1738 entry->cred = NULL; 1739 } 1740 1741 /* 1742 * Create a new entry and insert it AFTER the specified entry 1743 */ 1744 new_entry = vm_map_entry_create(map); 1745 *new_entry = *entry; 1746 1747 new_entry->start = entry->end = end; 1748 new_entry->offset += (end - entry->start); 1749 if (new_entry->cred != NULL) 1750 crhold(entry->cred); 1751 1752 vm_map_entry_link(map, entry, new_entry); 1753 1754 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1755 vm_object_reference(new_entry->object.vm_object); 1756 } 1757} 1758 1759/* 1760 * vm_map_submap: [ kernel use only ] 1761 * 1762 * Mark the given range as handled by a subordinate map. 1763 * 1764 * This range must have been created with vm_map_find, 1765 * and no other operations may have been performed on this 1766 * range prior to calling vm_map_submap. 1767 * 1768 * Only a limited number of operations can be performed 1769 * within this rage after calling vm_map_submap: 1770 * vm_fault 1771 * [Don't try vm_map_copy!] 1772 * 1773 * To remove a submapping, one must first remove the 1774 * range from the superior map, and then destroy the 1775 * submap (if desired). [Better yet, don't try it.] 1776 */ 1777int 1778vm_map_submap( 1779 vm_map_t map, 1780 vm_offset_t start, 1781 vm_offset_t end, 1782 vm_map_t submap) 1783{ 1784 vm_map_entry_t entry; 1785 int result = KERN_INVALID_ARGUMENT; 1786 1787 vm_map_lock(map); 1788 1789 VM_MAP_RANGE_CHECK(map, start, end); 1790 1791 if (vm_map_lookup_entry(map, start, &entry)) { 1792 vm_map_clip_start(map, entry, start); 1793 } else 1794 entry = entry->next; 1795 1796 vm_map_clip_end(map, entry, end); 1797 1798 if ((entry->start == start) && (entry->end == end) && 1799 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1800 (entry->object.vm_object == NULL)) { 1801 entry->object.sub_map = submap; 1802 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1803 result = KERN_SUCCESS; 1804 } 1805 vm_map_unlock(map); 1806 1807 return (result); 1808} 1809 1810/* 1811 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified 1812 */ 1813#define MAX_INIT_PT 96 1814 1815/* 1816 * vm_map_pmap_enter: 1817 * 1818 * Preload the specified map's pmap with mappings to the specified 1819 * object's memory-resident pages. No further physical pages are 1820 * allocated, and no further virtual pages are retrieved from secondary 1821 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 1822 * limited number of page mappings are created at the low-end of the 1823 * specified address range. (For this purpose, a superpage mapping 1824 * counts as one page mapping.) Otherwise, all resident pages within 1825 * the specified address range are mapped. Because these mappings are 1826 * being created speculatively, cached pages are not reactivated and 1827 * mapped. 1828 */ 1829void 1830vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1831 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1832{ 1833 vm_offset_t start; 1834 vm_page_t p, p_start; 1835 vm_pindex_t mask, psize, threshold, tmpidx; 1836 1837 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1838 return; 1839 VM_OBJECT_RLOCK(object); 1840 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1841 VM_OBJECT_RUNLOCK(object); 1842 VM_OBJECT_WLOCK(object); 1843 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1844 pmap_object_init_pt(map->pmap, addr, object, pindex, 1845 size); 1846 VM_OBJECT_WUNLOCK(object); 1847 return; 1848 } 1849 VM_OBJECT_LOCK_DOWNGRADE(object); 1850 } 1851 1852 psize = atop(size); 1853 if (psize + pindex > object->size) { 1854 if (object->size < pindex) { 1855 VM_OBJECT_RUNLOCK(object); 1856 return; 1857 } 1858 psize = object->size - pindex; 1859 } 1860 1861 start = 0; 1862 p_start = NULL; 1863 threshold = MAX_INIT_PT; 1864 1865 p = vm_page_find_least(object, pindex); 1866 /* 1867 * Assert: the variable p is either (1) the page with the 1868 * least pindex greater than or equal to the parameter pindex 1869 * or (2) NULL. 1870 */ 1871 for (; 1872 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1873 p = TAILQ_NEXT(p, listq)) { 1874 /* 1875 * don't allow an madvise to blow away our really 1876 * free pages allocating pv entries. 1877 */ 1878 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 1879 cnt.v_free_count < cnt.v_free_reserved) || 1880 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 1881 tmpidx >= threshold)) { 1882 psize = tmpidx; 1883 break; 1884 } 1885 if (p->valid == VM_PAGE_BITS_ALL) { 1886 if (p_start == NULL) { 1887 start = addr + ptoa(tmpidx); 1888 p_start = p; 1889 } 1890 /* Jump ahead if a superpage mapping is possible. */ 1891 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 1892 (pagesizes[p->psind] - 1)) == 0) { 1893 mask = atop(pagesizes[p->psind]) - 1; 1894 if (tmpidx + mask < psize && 1895 vm_page_ps_is_valid(p)) { 1896 p += mask; 1897 threshold += mask; 1898 } 1899 } 1900 } else if (p_start != NULL) { 1901 pmap_enter_object(map->pmap, start, addr + 1902 ptoa(tmpidx), p_start, prot); 1903 p_start = NULL; 1904 } 1905 } 1906 if (p_start != NULL) 1907 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1908 p_start, prot); 1909 VM_OBJECT_RUNLOCK(object); 1910} 1911 1912/* 1913 * vm_map_protect: 1914 * 1915 * Sets the protection of the specified address 1916 * region in the target map. If "set_max" is 1917 * specified, the maximum protection is to be set; 1918 * otherwise, only the current protection is affected. 1919 */ 1920int 1921vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1922 vm_prot_t new_prot, boolean_t set_max) 1923{ 1924 vm_map_entry_t current, entry; 1925 vm_object_t obj; 1926 struct ucred *cred; 1927 vm_prot_t old_prot; 1928 1929 if (start == end) 1930 return (KERN_SUCCESS); 1931 1932 vm_map_lock(map); 1933 1934 VM_MAP_RANGE_CHECK(map, start, end); 1935 1936 if (vm_map_lookup_entry(map, start, &entry)) { 1937 vm_map_clip_start(map, entry, start); 1938 } else { 1939 entry = entry->next; 1940 } 1941 1942 /* 1943 * Make a first pass to check for protection violations. 1944 */ 1945 current = entry; 1946 while ((current != &map->header) && (current->start < end)) { 1947 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1948 vm_map_unlock(map); 1949 return (KERN_INVALID_ARGUMENT); 1950 } 1951 if ((new_prot & current->max_protection) != new_prot) { 1952 vm_map_unlock(map); 1953 return (KERN_PROTECTION_FAILURE); 1954 } 1955 current = current->next; 1956 } 1957 1958 1959 /* 1960 * Do an accounting pass for private read-only mappings that 1961 * now will do cow due to allowed write (e.g. debugger sets 1962 * breakpoint on text segment) 1963 */ 1964 for (current = entry; (current != &map->header) && 1965 (current->start < end); current = current->next) { 1966 1967 vm_map_clip_end(map, current, end); 1968 1969 if (set_max || 1970 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1971 ENTRY_CHARGED(current)) { 1972 continue; 1973 } 1974 1975 cred = curthread->td_ucred; 1976 obj = current->object.vm_object; 1977 1978 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1979 if (!swap_reserve(current->end - current->start)) { 1980 vm_map_unlock(map); 1981 return (KERN_RESOURCE_SHORTAGE); 1982 } 1983 crhold(cred); 1984 current->cred = cred; 1985 continue; 1986 } 1987 1988 VM_OBJECT_WLOCK(obj); 1989 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1990 VM_OBJECT_WUNLOCK(obj); 1991 continue; 1992 } 1993 1994 /* 1995 * Charge for the whole object allocation now, since 1996 * we cannot distinguish between non-charged and 1997 * charged clipped mapping of the same object later. 1998 */ 1999 KASSERT(obj->charge == 0, 2000 ("vm_map_protect: object %p overcharged (entry %p)", 2001 obj, current)); 2002 if (!swap_reserve(ptoa(obj->size))) { 2003 VM_OBJECT_WUNLOCK(obj); 2004 vm_map_unlock(map); 2005 return (KERN_RESOURCE_SHORTAGE); 2006 } 2007 2008 crhold(cred); 2009 obj->cred = cred; 2010 obj->charge = ptoa(obj->size); 2011 VM_OBJECT_WUNLOCK(obj); 2012 } 2013 2014 /* 2015 * Go back and fix up protections. [Note that clipping is not 2016 * necessary the second time.] 2017 */ 2018 current = entry; 2019 while ((current != &map->header) && (current->start < end)) { 2020 old_prot = current->protection; 2021 2022 if (set_max) 2023 current->protection = 2024 (current->max_protection = new_prot) & 2025 old_prot; 2026 else 2027 current->protection = new_prot; 2028 2029 /* 2030 * For user wired map entries, the normal lazy evaluation of 2031 * write access upgrades through soft page faults is 2032 * undesirable. Instead, immediately copy any pages that are 2033 * copy-on-write and enable write access in the physical map. 2034 */ 2035 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2036 (current->protection & VM_PROT_WRITE) != 0 && 2037 (old_prot & VM_PROT_WRITE) == 0) 2038 vm_fault_copy_entry(map, map, current, current, NULL); 2039 2040 /* 2041 * When restricting access, update the physical map. Worry 2042 * about copy-on-write here. 2043 */ 2044 if ((old_prot & ~current->protection) != 0) { 2045#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2046 VM_PROT_ALL) 2047 pmap_protect(map->pmap, current->start, 2048 current->end, 2049 current->protection & MASK(current)); 2050#undef MASK 2051 } 2052 vm_map_simplify_entry(map, current); 2053 current = current->next; 2054 } 2055 vm_map_unlock(map); 2056 return (KERN_SUCCESS); 2057} 2058 2059/* 2060 * vm_map_madvise: 2061 * 2062 * This routine traverses a processes map handling the madvise 2063 * system call. Advisories are classified as either those effecting 2064 * the vm_map_entry structure, or those effecting the underlying 2065 * objects. 2066 */ 2067int 2068vm_map_madvise( 2069 vm_map_t map, 2070 vm_offset_t start, 2071 vm_offset_t end, 2072 int behav) 2073{ 2074 vm_map_entry_t current, entry; 2075 int modify_map = 0; 2076 2077 /* 2078 * Some madvise calls directly modify the vm_map_entry, in which case 2079 * we need to use an exclusive lock on the map and we need to perform 2080 * various clipping operations. Otherwise we only need a read-lock 2081 * on the map. 2082 */ 2083 switch(behav) { 2084 case MADV_NORMAL: 2085 case MADV_SEQUENTIAL: 2086 case MADV_RANDOM: 2087 case MADV_NOSYNC: 2088 case MADV_AUTOSYNC: 2089 case MADV_NOCORE: 2090 case MADV_CORE: 2091 if (start == end) 2092 return (KERN_SUCCESS); 2093 modify_map = 1; 2094 vm_map_lock(map); 2095 break; 2096 case MADV_WILLNEED: 2097 case MADV_DONTNEED: 2098 case MADV_FREE: 2099 if (start == end) 2100 return (KERN_SUCCESS); 2101 vm_map_lock_read(map); 2102 break; 2103 default: 2104 return (KERN_INVALID_ARGUMENT); 2105 } 2106 2107 /* 2108 * Locate starting entry and clip if necessary. 2109 */ 2110 VM_MAP_RANGE_CHECK(map, start, end); 2111 2112 if (vm_map_lookup_entry(map, start, &entry)) { 2113 if (modify_map) 2114 vm_map_clip_start(map, entry, start); 2115 } else { 2116 entry = entry->next; 2117 } 2118 2119 if (modify_map) { 2120 /* 2121 * madvise behaviors that are implemented in the vm_map_entry. 2122 * 2123 * We clip the vm_map_entry so that behavioral changes are 2124 * limited to the specified address range. 2125 */ 2126 for (current = entry; 2127 (current != &map->header) && (current->start < end); 2128 current = current->next 2129 ) { 2130 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2131 continue; 2132 2133 vm_map_clip_end(map, current, end); 2134 2135 switch (behav) { 2136 case MADV_NORMAL: 2137 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2138 break; 2139 case MADV_SEQUENTIAL: 2140 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2141 break; 2142 case MADV_RANDOM: 2143 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2144 break; 2145 case MADV_NOSYNC: 2146 current->eflags |= MAP_ENTRY_NOSYNC; 2147 break; 2148 case MADV_AUTOSYNC: 2149 current->eflags &= ~MAP_ENTRY_NOSYNC; 2150 break; 2151 case MADV_NOCORE: 2152 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2153 break; 2154 case MADV_CORE: 2155 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2156 break; 2157 default: 2158 break; 2159 } 2160 vm_map_simplify_entry(map, current); 2161 } 2162 vm_map_unlock(map); 2163 } else { 2164 vm_pindex_t pstart, pend; 2165 2166 /* 2167 * madvise behaviors that are implemented in the underlying 2168 * vm_object. 2169 * 2170 * Since we don't clip the vm_map_entry, we have to clip 2171 * the vm_object pindex and count. 2172 */ 2173 for (current = entry; 2174 (current != &map->header) && (current->start < end); 2175 current = current->next 2176 ) { 2177 vm_offset_t useEnd, useStart; 2178 2179 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2180 continue; 2181 2182 pstart = OFF_TO_IDX(current->offset); 2183 pend = pstart + atop(current->end - current->start); 2184 useStart = current->start; 2185 useEnd = current->end; 2186 2187 if (current->start < start) { 2188 pstart += atop(start - current->start); 2189 useStart = start; 2190 } 2191 if (current->end > end) { 2192 pend -= atop(current->end - end); 2193 useEnd = end; 2194 } 2195 2196 if (pstart >= pend) 2197 continue; 2198 2199 /* 2200 * Perform the pmap_advise() before clearing 2201 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2202 * concurrent pmap operation, such as pmap_remove(), 2203 * could clear a reference in the pmap and set 2204 * PGA_REFERENCED on the page before the pmap_advise() 2205 * had completed. Consequently, the page would appear 2206 * referenced based upon an old reference that 2207 * occurred before this pmap_advise() ran. 2208 */ 2209 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2210 pmap_advise(map->pmap, useStart, useEnd, 2211 behav); 2212 2213 vm_object_madvise(current->object.vm_object, pstart, 2214 pend, behav); 2215 2216 /* 2217 * Pre-populate paging structures in the 2218 * WILLNEED case. For wired entries, the 2219 * paging structures are already populated. 2220 */ 2221 if (behav == MADV_WILLNEED && 2222 current->wired_count == 0) { 2223 vm_map_pmap_enter(map, 2224 useStart, 2225 current->protection, 2226 current->object.vm_object, 2227 pstart, 2228 ptoa(pend - pstart), 2229 MAP_PREFAULT_MADVISE 2230 ); 2231 } 2232 } 2233 vm_map_unlock_read(map); 2234 } 2235 return (0); 2236} 2237 2238 2239/* 2240 * vm_map_inherit: 2241 * 2242 * Sets the inheritance of the specified address 2243 * range in the target map. Inheritance 2244 * affects how the map will be shared with 2245 * child maps at the time of vmspace_fork. 2246 */ 2247int 2248vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2249 vm_inherit_t new_inheritance) 2250{ 2251 vm_map_entry_t entry; 2252 vm_map_entry_t temp_entry; 2253 2254 switch (new_inheritance) { 2255 case VM_INHERIT_NONE: 2256 case VM_INHERIT_COPY: 2257 case VM_INHERIT_SHARE: 2258 break; 2259 default: 2260 return (KERN_INVALID_ARGUMENT); 2261 } 2262 if (start == end) 2263 return (KERN_SUCCESS); 2264 vm_map_lock(map); 2265 VM_MAP_RANGE_CHECK(map, start, end); 2266 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2267 entry = temp_entry; 2268 vm_map_clip_start(map, entry, start); 2269 } else 2270 entry = temp_entry->next; 2271 while ((entry != &map->header) && (entry->start < end)) { 2272 vm_map_clip_end(map, entry, end); 2273 entry->inheritance = new_inheritance; 2274 vm_map_simplify_entry(map, entry); 2275 entry = entry->next; 2276 } 2277 vm_map_unlock(map); 2278 return (KERN_SUCCESS); 2279} 2280 2281/* 2282 * vm_map_unwire: 2283 * 2284 * Implements both kernel and user unwiring. 2285 */ 2286int 2287vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2288 int flags) 2289{ 2290 vm_map_entry_t entry, first_entry, tmp_entry; 2291 vm_offset_t saved_start; 2292 unsigned int last_timestamp; 2293 int rv; 2294 boolean_t need_wakeup, result, user_unwire; 2295 2296 if (start == end) 2297 return (KERN_SUCCESS); 2298 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2299 vm_map_lock(map); 2300 VM_MAP_RANGE_CHECK(map, start, end); 2301 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2302 if (flags & VM_MAP_WIRE_HOLESOK) 2303 first_entry = first_entry->next; 2304 else { 2305 vm_map_unlock(map); 2306 return (KERN_INVALID_ADDRESS); 2307 } 2308 } 2309 last_timestamp = map->timestamp; 2310 entry = first_entry; 2311 while (entry != &map->header && entry->start < end) { 2312 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2313 /* 2314 * We have not yet clipped the entry. 2315 */ 2316 saved_start = (start >= entry->start) ? start : 2317 entry->start; 2318 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2319 if (vm_map_unlock_and_wait(map, 0)) { 2320 /* 2321 * Allow interruption of user unwiring? 2322 */ 2323 } 2324 vm_map_lock(map); 2325 if (last_timestamp+1 != map->timestamp) { 2326 /* 2327 * Look again for the entry because the map was 2328 * modified while it was unlocked. 2329 * Specifically, the entry may have been 2330 * clipped, merged, or deleted. 2331 */ 2332 if (!vm_map_lookup_entry(map, saved_start, 2333 &tmp_entry)) { 2334 if (flags & VM_MAP_WIRE_HOLESOK) 2335 tmp_entry = tmp_entry->next; 2336 else { 2337 if (saved_start == start) { 2338 /* 2339 * First_entry has been deleted. 2340 */ 2341 vm_map_unlock(map); 2342 return (KERN_INVALID_ADDRESS); 2343 } 2344 end = saved_start; 2345 rv = KERN_INVALID_ADDRESS; 2346 goto done; 2347 } 2348 } 2349 if (entry == first_entry) 2350 first_entry = tmp_entry; 2351 else 2352 first_entry = NULL; 2353 entry = tmp_entry; 2354 } 2355 last_timestamp = map->timestamp; 2356 continue; 2357 } 2358 vm_map_clip_start(map, entry, start); 2359 vm_map_clip_end(map, entry, end); 2360 /* 2361 * Mark the entry in case the map lock is released. (See 2362 * above.) 2363 */ 2364 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2365 entry->wiring_thread == NULL, 2366 ("owned map entry %p", entry)); 2367 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2368 entry->wiring_thread = curthread; 2369 /* 2370 * Check the map for holes in the specified region. 2371 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2372 */ 2373 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2374 (entry->end < end && (entry->next == &map->header || 2375 entry->next->start > entry->end))) { 2376 end = entry->end; 2377 rv = KERN_INVALID_ADDRESS; 2378 goto done; 2379 } 2380 /* 2381 * If system unwiring, require that the entry is system wired. 2382 */ 2383 if (!user_unwire && 2384 vm_map_entry_system_wired_count(entry) == 0) { 2385 end = entry->end; 2386 rv = KERN_INVALID_ARGUMENT; 2387 goto done; 2388 } 2389 entry = entry->next; 2390 } 2391 rv = KERN_SUCCESS; 2392done: 2393 need_wakeup = FALSE; 2394 if (first_entry == NULL) { 2395 result = vm_map_lookup_entry(map, start, &first_entry); 2396 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2397 first_entry = first_entry->next; 2398 else 2399 KASSERT(result, ("vm_map_unwire: lookup failed")); 2400 } 2401 for (entry = first_entry; entry != &map->header && entry->start < end; 2402 entry = entry->next) { 2403 /* 2404 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2405 * space in the unwired region could have been mapped 2406 * while the map lock was dropped for draining 2407 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2408 * could be simultaneously wiring this new mapping 2409 * entry. Detect these cases and skip any entries 2410 * marked as in transition by us. 2411 */ 2412 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2413 entry->wiring_thread != curthread) { 2414 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2415 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2416 continue; 2417 } 2418 2419 if (rv == KERN_SUCCESS && (!user_unwire || 2420 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2421 if (user_unwire) 2422 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2423 if (entry->wired_count == 1) 2424 vm_map_entry_unwire(map, entry); 2425 else 2426 entry->wired_count--; 2427 } 2428 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2429 ("vm_map_unwire: in-transition flag missing %p", entry)); 2430 KASSERT(entry->wiring_thread == curthread, 2431 ("vm_map_unwire: alien wire %p", entry)); 2432 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2433 entry->wiring_thread = NULL; 2434 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2435 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2436 need_wakeup = TRUE; 2437 } 2438 vm_map_simplify_entry(map, entry); 2439 } 2440 vm_map_unlock(map); 2441 if (need_wakeup) 2442 vm_map_wakeup(map); 2443 return (rv); 2444} 2445 2446/* 2447 * vm_map_wire_entry_failure: 2448 * 2449 * Handle a wiring failure on the given entry. 2450 * 2451 * The map should be locked. 2452 */ 2453static void 2454vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 2455 vm_offset_t failed_addr) 2456{ 2457 2458 VM_MAP_ASSERT_LOCKED(map); 2459 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && 2460 entry->wired_count == 1, 2461 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); 2462 KASSERT(failed_addr < entry->end, 2463 ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); 2464 2465 /* 2466 * If any pages at the start of this entry were successfully wired, 2467 * then unwire them. 2468 */ 2469 if (failed_addr > entry->start) { 2470 pmap_unwire(map->pmap, entry->start, failed_addr); 2471 vm_object_unwire(entry->object.vm_object, entry->offset, 2472 failed_addr - entry->start, PQ_ACTIVE); 2473 } 2474 2475 /* 2476 * Assign an out-of-range value to represent the failure to wire this 2477 * entry. 2478 */ 2479 entry->wired_count = -1; 2480} 2481 2482/* 2483 * vm_map_wire: 2484 * 2485 * Implements both kernel and user wiring. 2486 */ 2487int 2488vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2489 int flags) 2490{ 2491 vm_map_entry_t entry, first_entry, tmp_entry; 2492 vm_offset_t faddr, saved_end, saved_start; 2493 unsigned int last_timestamp; 2494 int rv; 2495 boolean_t need_wakeup, result, user_wire; 2496 vm_prot_t prot; 2497 2498 if (start == end) 2499 return (KERN_SUCCESS); 2500 prot = 0; 2501 if (flags & VM_MAP_WIRE_WRITE) 2502 prot |= VM_PROT_WRITE; 2503 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2504 vm_map_lock(map); 2505 VM_MAP_RANGE_CHECK(map, start, end); 2506 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2507 if (flags & VM_MAP_WIRE_HOLESOK) 2508 first_entry = first_entry->next; 2509 else { 2510 vm_map_unlock(map); 2511 return (KERN_INVALID_ADDRESS); 2512 } 2513 } 2514 last_timestamp = map->timestamp; 2515 entry = first_entry; 2516 while (entry != &map->header && entry->start < end) { 2517 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2518 /* 2519 * We have not yet clipped the entry. 2520 */ 2521 saved_start = (start >= entry->start) ? start : 2522 entry->start; 2523 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2524 if (vm_map_unlock_and_wait(map, 0)) { 2525 /* 2526 * Allow interruption of user wiring? 2527 */ 2528 } 2529 vm_map_lock(map); 2530 if (last_timestamp + 1 != map->timestamp) { 2531 /* 2532 * Look again for the entry because the map was 2533 * modified while it was unlocked. 2534 * Specifically, the entry may have been 2535 * clipped, merged, or deleted. 2536 */ 2537 if (!vm_map_lookup_entry(map, saved_start, 2538 &tmp_entry)) { 2539 if (flags & VM_MAP_WIRE_HOLESOK) 2540 tmp_entry = tmp_entry->next; 2541 else { 2542 if (saved_start == start) { 2543 /* 2544 * first_entry has been deleted. 2545 */ 2546 vm_map_unlock(map); 2547 return (KERN_INVALID_ADDRESS); 2548 } 2549 end = saved_start; 2550 rv = KERN_INVALID_ADDRESS; 2551 goto done; 2552 } 2553 } 2554 if (entry == first_entry) 2555 first_entry = tmp_entry; 2556 else 2557 first_entry = NULL; 2558 entry = tmp_entry; 2559 } 2560 last_timestamp = map->timestamp; 2561 continue; 2562 } 2563 vm_map_clip_start(map, entry, start); 2564 vm_map_clip_end(map, entry, end); 2565 /* 2566 * Mark the entry in case the map lock is released. (See 2567 * above.) 2568 */ 2569 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2570 entry->wiring_thread == NULL, 2571 ("owned map entry %p", entry)); 2572 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2573 entry->wiring_thread = curthread; 2574 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2575 || (entry->protection & prot) != prot) { 2576 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2577 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2578 end = entry->end; 2579 rv = KERN_INVALID_ADDRESS; 2580 goto done; 2581 } 2582 goto next_entry; 2583 } 2584 if (entry->wired_count == 0) { 2585 entry->wired_count++; 2586 saved_start = entry->start; 2587 saved_end = entry->end; 2588 2589 /* 2590 * Release the map lock, relying on the in-transition 2591 * mark. Mark the map busy for fork. 2592 */ 2593 vm_map_busy(map); 2594 vm_map_unlock(map); 2595 2596 faddr = saved_start; 2597 do { 2598 /* 2599 * Simulate a fault to get the page and enter 2600 * it into the physical map. 2601 */ 2602 if ((rv = vm_fault(map, faddr, VM_PROT_NONE, 2603 VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS) 2604 break; 2605 } while ((faddr += PAGE_SIZE) < saved_end); 2606 vm_map_lock(map); 2607 vm_map_unbusy(map); 2608 if (last_timestamp + 1 != map->timestamp) { 2609 /* 2610 * Look again for the entry because the map was 2611 * modified while it was unlocked. The entry 2612 * may have been clipped, but NOT merged or 2613 * deleted. 2614 */ 2615 result = vm_map_lookup_entry(map, saved_start, 2616 &tmp_entry); 2617 KASSERT(result, ("vm_map_wire: lookup failed")); 2618 if (entry == first_entry) 2619 first_entry = tmp_entry; 2620 else 2621 first_entry = NULL; 2622 entry = tmp_entry; 2623 while (entry->end < saved_end) { 2624 /* 2625 * In case of failure, handle entries 2626 * that were not fully wired here; 2627 * fully wired entries are handled 2628 * later. 2629 */ 2630 if (rv != KERN_SUCCESS && 2631 faddr < entry->end) 2632 vm_map_wire_entry_failure(map, 2633 entry, faddr); 2634 entry = entry->next; 2635 } 2636 } 2637 last_timestamp = map->timestamp; 2638 if (rv != KERN_SUCCESS) { 2639 vm_map_wire_entry_failure(map, entry, faddr); 2640 end = entry->end; 2641 goto done; 2642 } 2643 } else if (!user_wire || 2644 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2645 entry->wired_count++; 2646 } 2647 /* 2648 * Check the map for holes in the specified region. 2649 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2650 */ 2651 next_entry: 2652 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2653 (entry->end < end && (entry->next == &map->header || 2654 entry->next->start > entry->end))) { 2655 end = entry->end; 2656 rv = KERN_INVALID_ADDRESS; 2657 goto done; 2658 } 2659 entry = entry->next; 2660 } 2661 rv = KERN_SUCCESS; 2662done: 2663 need_wakeup = FALSE; 2664 if (first_entry == NULL) { 2665 result = vm_map_lookup_entry(map, start, &first_entry); 2666 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2667 first_entry = first_entry->next; 2668 else 2669 KASSERT(result, ("vm_map_wire: lookup failed")); 2670 } 2671 for (entry = first_entry; entry != &map->header && entry->start < end; 2672 entry = entry->next) { 2673 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2674 goto next_entry_done; 2675 2676 /* 2677 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2678 * space in the unwired region could have been mapped 2679 * while the map lock was dropped for faulting in the 2680 * pages or draining MAP_ENTRY_IN_TRANSITION. 2681 * Moreover, another thread could be simultaneously 2682 * wiring this new mapping entry. Detect these cases 2683 * and skip any entries marked as in transition by us. 2684 */ 2685 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2686 entry->wiring_thread != curthread) { 2687 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2688 ("vm_map_wire: !HOLESOK and new/changed entry")); 2689 continue; 2690 } 2691 2692 if (rv == KERN_SUCCESS) { 2693 if (user_wire) 2694 entry->eflags |= MAP_ENTRY_USER_WIRED; 2695 } else if (entry->wired_count == -1) { 2696 /* 2697 * Wiring failed on this entry. Thus, unwiring is 2698 * unnecessary. 2699 */ 2700 entry->wired_count = 0; 2701 } else if (!user_wire || 2702 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2703 /* 2704 * Undo the wiring. Wiring succeeded on this entry 2705 * but failed on a later entry. 2706 */ 2707 if (entry->wired_count == 1) 2708 vm_map_entry_unwire(map, entry); 2709 else 2710 entry->wired_count--; 2711 } 2712 next_entry_done: 2713 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2714 ("vm_map_wire: in-transition flag missing %p", entry)); 2715 KASSERT(entry->wiring_thread == curthread, 2716 ("vm_map_wire: alien wire %p", entry)); 2717 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2718 MAP_ENTRY_WIRE_SKIPPED); 2719 entry->wiring_thread = NULL; 2720 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2721 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2722 need_wakeup = TRUE; 2723 } 2724 vm_map_simplify_entry(map, entry); 2725 } 2726 vm_map_unlock(map); 2727 if (need_wakeup) 2728 vm_map_wakeup(map); 2729 return (rv); 2730} 2731 2732/* 2733 * vm_map_sync 2734 * 2735 * Push any dirty cached pages in the address range to their pager. 2736 * If syncio is TRUE, dirty pages are written synchronously. 2737 * If invalidate is TRUE, any cached pages are freed as well. 2738 * 2739 * If the size of the region from start to end is zero, we are 2740 * supposed to flush all modified pages within the region containing 2741 * start. Unfortunately, a region can be split or coalesced with 2742 * neighboring regions, making it difficult to determine what the 2743 * original region was. Therefore, we approximate this requirement by 2744 * flushing the current region containing start. 2745 * 2746 * Returns an error if any part of the specified range is not mapped. 2747 */ 2748int 2749vm_map_sync( 2750 vm_map_t map, 2751 vm_offset_t start, 2752 vm_offset_t end, 2753 boolean_t syncio, 2754 boolean_t invalidate) 2755{ 2756 vm_map_entry_t current; 2757 vm_map_entry_t entry; 2758 vm_size_t size; 2759 vm_object_t object; 2760 vm_ooffset_t offset; 2761 unsigned int last_timestamp; 2762 boolean_t failed; 2763 2764 vm_map_lock_read(map); 2765 VM_MAP_RANGE_CHECK(map, start, end); 2766 if (!vm_map_lookup_entry(map, start, &entry)) { 2767 vm_map_unlock_read(map); 2768 return (KERN_INVALID_ADDRESS); 2769 } else if (start == end) { 2770 start = entry->start; 2771 end = entry->end; 2772 } 2773 /* 2774 * Make a first pass to check for user-wired memory and holes. 2775 */ 2776 for (current = entry; current != &map->header && current->start < end; 2777 current = current->next) { 2778 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2779 vm_map_unlock_read(map); 2780 return (KERN_INVALID_ARGUMENT); 2781 } 2782 if (end > current->end && 2783 (current->next == &map->header || 2784 current->end != current->next->start)) { 2785 vm_map_unlock_read(map); 2786 return (KERN_INVALID_ADDRESS); 2787 } 2788 } 2789 2790 if (invalidate) 2791 pmap_remove(map->pmap, start, end); 2792 failed = FALSE; 2793 2794 /* 2795 * Make a second pass, cleaning/uncaching pages from the indicated 2796 * objects as we go. 2797 */ 2798 for (current = entry; current != &map->header && current->start < end;) { 2799 offset = current->offset + (start - current->start); 2800 size = (end <= current->end ? end : current->end) - start; 2801 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2802 vm_map_t smap; 2803 vm_map_entry_t tentry; 2804 vm_size_t tsize; 2805 2806 smap = current->object.sub_map; 2807 vm_map_lock_read(smap); 2808 (void) vm_map_lookup_entry(smap, offset, &tentry); 2809 tsize = tentry->end - offset; 2810 if (tsize < size) 2811 size = tsize; 2812 object = tentry->object.vm_object; 2813 offset = tentry->offset + (offset - tentry->start); 2814 vm_map_unlock_read(smap); 2815 } else { 2816 object = current->object.vm_object; 2817 } 2818 vm_object_reference(object); 2819 last_timestamp = map->timestamp; 2820 vm_map_unlock_read(map); 2821 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2822 failed = TRUE; 2823 start += size; 2824 vm_object_deallocate(object); 2825 vm_map_lock_read(map); 2826 if (last_timestamp == map->timestamp || 2827 !vm_map_lookup_entry(map, start, ¤t)) 2828 current = current->next; 2829 } 2830 2831 vm_map_unlock_read(map); 2832 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2833} 2834 2835/* 2836 * vm_map_entry_unwire: [ internal use only ] 2837 * 2838 * Make the region specified by this entry pageable. 2839 * 2840 * The map in question should be locked. 2841 * [This is the reason for this routine's existence.] 2842 */ 2843static void 2844vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2845{ 2846 2847 VM_MAP_ASSERT_LOCKED(map); 2848 KASSERT(entry->wired_count > 0, 2849 ("vm_map_entry_unwire: entry %p isn't wired", entry)); 2850 pmap_unwire(map->pmap, entry->start, entry->end); 2851 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end - 2852 entry->start, PQ_ACTIVE); 2853 entry->wired_count = 0; 2854} 2855 2856static void 2857vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2858{ 2859 2860 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2861 vm_object_deallocate(entry->object.vm_object); 2862 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2863} 2864 2865/* 2866 * vm_map_entry_delete: [ internal use only ] 2867 * 2868 * Deallocate the given entry from the target map. 2869 */ 2870static void 2871vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2872{ 2873 vm_object_t object; 2874 vm_pindex_t offidxstart, offidxend, count, size1; 2875 vm_ooffset_t size; 2876 2877 vm_map_entry_unlink(map, entry); 2878 object = entry->object.vm_object; 2879 size = entry->end - entry->start; 2880 map->size -= size; 2881 2882 if (entry->cred != NULL) { 2883 swap_release_by_cred(size, entry->cred); 2884 crfree(entry->cred); 2885 } 2886 2887 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2888 (object != NULL)) { 2889 KASSERT(entry->cred == NULL || object->cred == NULL || 2890 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2891 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2892 count = OFF_TO_IDX(size); 2893 offidxstart = OFF_TO_IDX(entry->offset); 2894 offidxend = offidxstart + count; 2895 VM_OBJECT_WLOCK(object); 2896 if (object->ref_count != 1 && 2897 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2898 object == kernel_object || object == kmem_object)) { 2899 vm_object_collapse(object); 2900 2901 /* 2902 * The option OBJPR_NOTMAPPED can be passed here 2903 * because vm_map_delete() already performed 2904 * pmap_remove() on the only mapping to this range 2905 * of pages. 2906 */ 2907 vm_object_page_remove(object, offidxstart, offidxend, 2908 OBJPR_NOTMAPPED); 2909 if (object->type == OBJT_SWAP) 2910 swap_pager_freespace(object, offidxstart, count); 2911 if (offidxend >= object->size && 2912 offidxstart < object->size) { 2913 size1 = object->size; 2914 object->size = offidxstart; 2915 if (object->cred != NULL) { 2916 size1 -= object->size; 2917 KASSERT(object->charge >= ptoa(size1), 2918 ("vm_map_entry_delete: object->charge < 0")); 2919 swap_release_by_cred(ptoa(size1), object->cred); 2920 object->charge -= ptoa(size1); 2921 } 2922 } 2923 } 2924 VM_OBJECT_WUNLOCK(object); 2925 } else 2926 entry->object.vm_object = NULL; 2927 if (map->system_map) 2928 vm_map_entry_deallocate(entry, TRUE); 2929 else { 2930 entry->next = curthread->td_map_def_user; 2931 curthread->td_map_def_user = entry; 2932 } 2933} 2934 2935/* 2936 * vm_map_delete: [ internal use only ] 2937 * 2938 * Deallocates the given address range from the target 2939 * map. 2940 */ 2941int 2942vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2943{ 2944 vm_map_entry_t entry; 2945 vm_map_entry_t first_entry; 2946 2947 VM_MAP_ASSERT_LOCKED(map); 2948 if (start == end) 2949 return (KERN_SUCCESS); 2950 2951 /* 2952 * Find the start of the region, and clip it 2953 */ 2954 if (!vm_map_lookup_entry(map, start, &first_entry)) 2955 entry = first_entry->next; 2956 else { 2957 entry = first_entry; 2958 vm_map_clip_start(map, entry, start); 2959 } 2960 2961 /* 2962 * Step through all entries in this region 2963 */ 2964 while ((entry != &map->header) && (entry->start < end)) { 2965 vm_map_entry_t next; 2966 2967 /* 2968 * Wait for wiring or unwiring of an entry to complete. 2969 * Also wait for any system wirings to disappear on 2970 * user maps. 2971 */ 2972 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2973 (vm_map_pmap(map) != kernel_pmap && 2974 vm_map_entry_system_wired_count(entry) != 0)) { 2975 unsigned int last_timestamp; 2976 vm_offset_t saved_start; 2977 vm_map_entry_t tmp_entry; 2978 2979 saved_start = entry->start; 2980 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2981 last_timestamp = map->timestamp; 2982 (void) vm_map_unlock_and_wait(map, 0); 2983 vm_map_lock(map); 2984 if (last_timestamp + 1 != map->timestamp) { 2985 /* 2986 * Look again for the entry because the map was 2987 * modified while it was unlocked. 2988 * Specifically, the entry may have been 2989 * clipped, merged, or deleted. 2990 */ 2991 if (!vm_map_lookup_entry(map, saved_start, 2992 &tmp_entry)) 2993 entry = tmp_entry->next; 2994 else { 2995 entry = tmp_entry; 2996 vm_map_clip_start(map, entry, 2997 saved_start); 2998 } 2999 } 3000 continue; 3001 } 3002 vm_map_clip_end(map, entry, end); 3003 3004 next = entry->next; 3005 3006 /* 3007 * Unwire before removing addresses from the pmap; otherwise, 3008 * unwiring will put the entries back in the pmap. 3009 */ 3010 if (entry->wired_count != 0) { 3011 vm_map_entry_unwire(map, entry); 3012 } 3013 3014 pmap_remove(map->pmap, entry->start, entry->end); 3015 3016 /* 3017 * Delete the entry only after removing all pmap 3018 * entries pointing to its pages. (Otherwise, its 3019 * page frames may be reallocated, and any modify bits 3020 * will be set in the wrong object!) 3021 */ 3022 vm_map_entry_delete(map, entry); 3023 entry = next; 3024 } 3025 return (KERN_SUCCESS); 3026} 3027 3028/* 3029 * vm_map_remove: 3030 * 3031 * Remove the given address range from the target map. 3032 * This is the exported form of vm_map_delete. 3033 */ 3034int 3035vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 3036{ 3037 int result; 3038 3039 vm_map_lock(map); 3040 VM_MAP_RANGE_CHECK(map, start, end); 3041 result = vm_map_delete(map, start, end); 3042 vm_map_unlock(map); 3043 return (result); 3044} 3045 3046/* 3047 * vm_map_check_protection: 3048 * 3049 * Assert that the target map allows the specified privilege on the 3050 * entire address region given. The entire region must be allocated. 3051 * 3052 * WARNING! This code does not and should not check whether the 3053 * contents of the region is accessible. For example a smaller file 3054 * might be mapped into a larger address space. 3055 * 3056 * NOTE! This code is also called by munmap(). 3057 * 3058 * The map must be locked. A read lock is sufficient. 3059 */ 3060boolean_t 3061vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3062 vm_prot_t protection) 3063{ 3064 vm_map_entry_t entry; 3065 vm_map_entry_t tmp_entry; 3066 3067 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3068 return (FALSE); 3069 entry = tmp_entry; 3070 3071 while (start < end) { 3072 if (entry == &map->header) 3073 return (FALSE); 3074 /* 3075 * No holes allowed! 3076 */ 3077 if (start < entry->start) 3078 return (FALSE); 3079 /* 3080 * Check protection associated with entry. 3081 */ 3082 if ((entry->protection & protection) != protection) 3083 return (FALSE); 3084 /* go to next entry */ 3085 start = entry->end; 3086 entry = entry->next; 3087 } 3088 return (TRUE); 3089} 3090 3091/* 3092 * vm_map_copy_entry: 3093 * 3094 * Copies the contents of the source entry to the destination 3095 * entry. The entries *must* be aligned properly. 3096 */ 3097static void 3098vm_map_copy_entry( 3099 vm_map_t src_map, 3100 vm_map_t dst_map, 3101 vm_map_entry_t src_entry, 3102 vm_map_entry_t dst_entry, 3103 vm_ooffset_t *fork_charge) 3104{ 3105 vm_object_t src_object; 3106 vm_map_entry_t fake_entry; 3107 vm_offset_t size; 3108 struct ucred *cred; 3109 int charged; 3110 3111 VM_MAP_ASSERT_LOCKED(dst_map); 3112 3113 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3114 return; 3115 3116 if (src_entry->wired_count == 0 || 3117 (src_entry->protection & VM_PROT_WRITE) == 0) { 3118 /* 3119 * If the source entry is marked needs_copy, it is already 3120 * write-protected. 3121 */ 3122 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3123 (src_entry->protection & VM_PROT_WRITE) != 0) { 3124 pmap_protect(src_map->pmap, 3125 src_entry->start, 3126 src_entry->end, 3127 src_entry->protection & ~VM_PROT_WRITE); 3128 } 3129 3130 /* 3131 * Make a copy of the object. 3132 */ 3133 size = src_entry->end - src_entry->start; 3134 if ((src_object = src_entry->object.vm_object) != NULL) { 3135 VM_OBJECT_WLOCK(src_object); 3136 charged = ENTRY_CHARGED(src_entry); 3137 if ((src_object->handle == NULL) && 3138 (src_object->type == OBJT_DEFAULT || 3139 src_object->type == OBJT_SWAP)) { 3140 vm_object_collapse(src_object); 3141 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3142 vm_object_split(src_entry); 3143 src_object = src_entry->object.vm_object; 3144 } 3145 } 3146 vm_object_reference_locked(src_object); 3147 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3148 if (src_entry->cred != NULL && 3149 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3150 KASSERT(src_object->cred == NULL, 3151 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3152 src_object)); 3153 src_object->cred = src_entry->cred; 3154 src_object->charge = size; 3155 } 3156 VM_OBJECT_WUNLOCK(src_object); 3157 dst_entry->object.vm_object = src_object; 3158 if (charged) { 3159 cred = curthread->td_ucred; 3160 crhold(cred); 3161 dst_entry->cred = cred; 3162 *fork_charge += size; 3163 if (!(src_entry->eflags & 3164 MAP_ENTRY_NEEDS_COPY)) { 3165 crhold(cred); 3166 src_entry->cred = cred; 3167 *fork_charge += size; 3168 } 3169 } 3170 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3171 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 3172 dst_entry->offset = src_entry->offset; 3173 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3174 /* 3175 * MAP_ENTRY_VN_WRITECNT cannot 3176 * indicate write reference from 3177 * src_entry, since the entry is 3178 * marked as needs copy. Allocate a 3179 * fake entry that is used to 3180 * decrement object->un_pager.vnp.writecount 3181 * at the appropriate time. Attach 3182 * fake_entry to the deferred list. 3183 */ 3184 fake_entry = vm_map_entry_create(dst_map); 3185 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3186 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3187 vm_object_reference(src_object); 3188 fake_entry->object.vm_object = src_object; 3189 fake_entry->start = src_entry->start; 3190 fake_entry->end = src_entry->end; 3191 fake_entry->next = curthread->td_map_def_user; 3192 curthread->td_map_def_user = fake_entry; 3193 } 3194 } else { 3195 dst_entry->object.vm_object = NULL; 3196 dst_entry->offset = 0; 3197 if (src_entry->cred != NULL) { 3198 dst_entry->cred = curthread->td_ucred; 3199 crhold(dst_entry->cred); 3200 *fork_charge += size; 3201 } 3202 } 3203 3204 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 3205 dst_entry->end - dst_entry->start, src_entry->start); 3206 } else { 3207 /* 3208 * We don't want to make writeable wired pages copy-on-write. 3209 * Immediately copy these pages into the new map by simulating 3210 * page faults. The new pages are pageable. 3211 */ 3212 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3213 fork_charge); 3214 } 3215} 3216 3217/* 3218 * vmspace_map_entry_forked: 3219 * Update the newly-forked vmspace each time a map entry is inherited 3220 * or copied. The values for vm_dsize and vm_tsize are approximate 3221 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3222 */ 3223static void 3224vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3225 vm_map_entry_t entry) 3226{ 3227 vm_size_t entrysize; 3228 vm_offset_t newend; 3229 3230 entrysize = entry->end - entry->start; 3231 vm2->vm_map.size += entrysize; 3232 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3233 vm2->vm_ssize += btoc(entrysize); 3234 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3235 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3236 newend = MIN(entry->end, 3237 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3238 vm2->vm_dsize += btoc(newend - entry->start); 3239 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3240 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3241 newend = MIN(entry->end, 3242 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3243 vm2->vm_tsize += btoc(newend - entry->start); 3244 } 3245} 3246 3247/* 3248 * vmspace_fork: 3249 * Create a new process vmspace structure and vm_map 3250 * based on those of an existing process. The new map 3251 * is based on the old map, according to the inheritance 3252 * values on the regions in that map. 3253 * 3254 * XXX It might be worth coalescing the entries added to the new vmspace. 3255 * 3256 * The source map must not be locked. 3257 */ 3258struct vmspace * 3259vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3260{ 3261 struct vmspace *vm2; 3262 vm_map_t new_map, old_map; 3263 vm_map_entry_t new_entry, old_entry; 3264 vm_object_t object; 3265 int locked; 3266 3267 old_map = &vm1->vm_map; 3268 /* Copy immutable fields of vm1 to vm2. */ 3269 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); 3270 if (vm2 == NULL) 3271 return (NULL); 3272 vm2->vm_taddr = vm1->vm_taddr; 3273 vm2->vm_daddr = vm1->vm_daddr; 3274 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3275 vm_map_lock(old_map); 3276 if (old_map->busy) 3277 vm_map_wait_busy(old_map); 3278 new_map = &vm2->vm_map; 3279 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3280 KASSERT(locked, ("vmspace_fork: lock failed")); 3281 3282 old_entry = old_map->header.next; 3283 3284 while (old_entry != &old_map->header) { 3285 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3286 panic("vm_map_fork: encountered a submap"); 3287 3288 switch (old_entry->inheritance) { 3289 case VM_INHERIT_NONE: 3290 break; 3291 3292 case VM_INHERIT_SHARE: 3293 /* 3294 * Clone the entry, creating the shared object if necessary. 3295 */ 3296 object = old_entry->object.vm_object; 3297 if (object == NULL) { 3298 object = vm_object_allocate(OBJT_DEFAULT, 3299 atop(old_entry->end - old_entry->start)); 3300 old_entry->object.vm_object = object; 3301 old_entry->offset = 0; 3302 if (old_entry->cred != NULL) { 3303 object->cred = old_entry->cred; 3304 object->charge = old_entry->end - 3305 old_entry->start; 3306 old_entry->cred = NULL; 3307 } 3308 } 3309 3310 /* 3311 * Add the reference before calling vm_object_shadow 3312 * to insure that a shadow object is created. 3313 */ 3314 vm_object_reference(object); 3315 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3316 vm_object_shadow(&old_entry->object.vm_object, 3317 &old_entry->offset, 3318 old_entry->end - old_entry->start); 3319 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3320 /* Transfer the second reference too. */ 3321 vm_object_reference( 3322 old_entry->object.vm_object); 3323 3324 /* 3325 * As in vm_map_simplify_entry(), the 3326 * vnode lock will not be acquired in 3327 * this call to vm_object_deallocate(). 3328 */ 3329 vm_object_deallocate(object); 3330 object = old_entry->object.vm_object; 3331 } 3332 VM_OBJECT_WLOCK(object); 3333 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3334 if (old_entry->cred != NULL) { 3335 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3336 object->cred = old_entry->cred; 3337 object->charge = old_entry->end - old_entry->start; 3338 old_entry->cred = NULL; 3339 } 3340 3341 /* 3342 * Assert the correct state of the vnode 3343 * v_writecount while the object is locked, to 3344 * not relock it later for the assertion 3345 * correctness. 3346 */ 3347 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3348 object->type == OBJT_VNODE) { 3349 KASSERT(((struct vnode *)object->handle)-> 3350 v_writecount > 0, 3351 ("vmspace_fork: v_writecount %p", object)); 3352 KASSERT(object->un_pager.vnp.writemappings > 0, 3353 ("vmspace_fork: vnp.writecount %p", 3354 object)); 3355 } 3356 VM_OBJECT_WUNLOCK(object); 3357 3358 /* 3359 * Clone the entry, referencing the shared object. 3360 */ 3361 new_entry = vm_map_entry_create(new_map); 3362 *new_entry = *old_entry; 3363 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3364 MAP_ENTRY_IN_TRANSITION); 3365 new_entry->wiring_thread = NULL; 3366 new_entry->wired_count = 0; 3367 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3368 vnode_pager_update_writecount(object, 3369 new_entry->start, new_entry->end); 3370 } 3371 3372 /* 3373 * Insert the entry into the new map -- we know we're 3374 * inserting at the end of the new map. 3375 */ 3376 vm_map_entry_link(new_map, new_map->header.prev, 3377 new_entry); 3378 vmspace_map_entry_forked(vm1, vm2, new_entry); 3379 3380 /* 3381 * Update the physical map 3382 */ 3383 pmap_copy(new_map->pmap, old_map->pmap, 3384 new_entry->start, 3385 (old_entry->end - old_entry->start), 3386 old_entry->start); 3387 break; 3388 3389 case VM_INHERIT_COPY: 3390 /* 3391 * Clone the entry and link into the map. 3392 */ 3393 new_entry = vm_map_entry_create(new_map); 3394 *new_entry = *old_entry; 3395 /* 3396 * Copied entry is COW over the old object. 3397 */ 3398 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3399 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3400 new_entry->wiring_thread = NULL; 3401 new_entry->wired_count = 0; 3402 new_entry->object.vm_object = NULL; 3403 new_entry->cred = NULL; 3404 vm_map_entry_link(new_map, new_map->header.prev, 3405 new_entry); 3406 vmspace_map_entry_forked(vm1, vm2, new_entry); 3407 vm_map_copy_entry(old_map, new_map, old_entry, 3408 new_entry, fork_charge); 3409 break; 3410 } 3411 old_entry = old_entry->next; 3412 } 3413 /* 3414 * Use inlined vm_map_unlock() to postpone handling the deferred 3415 * map entries, which cannot be done until both old_map and 3416 * new_map locks are released. 3417 */ 3418 sx_xunlock(&old_map->lock); 3419 sx_xunlock(&new_map->lock); 3420 vm_map_process_deferred(); 3421 3422 return (vm2); 3423} 3424 3425int 3426vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3427 vm_prot_t prot, vm_prot_t max, int cow) 3428{ 3429 vm_size_t growsize, init_ssize; 3430 rlim_t lmemlim, vmemlim; 3431 int rv; 3432 3433 growsize = sgrowsiz; 3434 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3435 vm_map_lock(map); 3436 PROC_LOCK(curproc); 3437 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); 3438 vmemlim = lim_cur(curproc, RLIMIT_VMEM); 3439 PROC_UNLOCK(curproc); 3440 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3441 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { 3442 rv = KERN_NO_SPACE; 3443 goto out; 3444 } 3445 } 3446 /* If we would blow our VMEM resource limit, no go */ 3447 if (map->size + init_ssize > vmemlim) { 3448 rv = KERN_NO_SPACE; 3449 goto out; 3450 } 3451 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3452 max, cow); 3453out: 3454 vm_map_unlock(map); 3455 return (rv); 3456} 3457 3458static int 3459vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3460 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3461{ 3462 vm_map_entry_t new_entry, prev_entry; 3463 vm_offset_t bot, top; 3464 vm_size_t init_ssize; 3465 int orient, rv; 3466 3467 /* 3468 * The stack orientation is piggybacked with the cow argument. 3469 * Extract it into orient and mask the cow argument so that we 3470 * don't pass it around further. 3471 * NOTE: We explicitly allow bi-directional stacks. 3472 */ 3473 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3474 KASSERT(orient != 0, ("No stack grow direction")); 3475 3476 if (addrbos < vm_map_min(map) || 3477 addrbos > vm_map_max(map) || 3478 addrbos + max_ssize < addrbos) 3479 return (KERN_NO_SPACE); 3480 3481 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3482 3483 /* If addr is already mapped, no go */ 3484 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3485 return (KERN_NO_SPACE); 3486 3487 /* 3488 * If we can't accomodate max_ssize in the current mapping, no go. 3489 * However, we need to be aware that subsequent user mappings might 3490 * map into the space we have reserved for stack, and currently this 3491 * space is not protected. 3492 * 3493 * Hopefully we will at least detect this condition when we try to 3494 * grow the stack. 3495 */ 3496 if ((prev_entry->next != &map->header) && 3497 (prev_entry->next->start < addrbos + max_ssize)) 3498 return (KERN_NO_SPACE); 3499 3500 /* 3501 * We initially map a stack of only init_ssize. We will grow as 3502 * needed later. Depending on the orientation of the stack (i.e. 3503 * the grow direction) we either map at the top of the range, the 3504 * bottom of the range or in the middle. 3505 * 3506 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3507 * and cow to be 0. Possibly we should eliminate these as input 3508 * parameters, and just pass these values here in the insert call. 3509 */ 3510 if (orient == MAP_STACK_GROWS_DOWN) 3511 bot = addrbos + max_ssize - init_ssize; 3512 else if (orient == MAP_STACK_GROWS_UP) 3513 bot = addrbos; 3514 else 3515 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3516 top = bot + init_ssize; 3517 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3518 3519 /* Now set the avail_ssize amount. */ 3520 if (rv == KERN_SUCCESS) { 3521 if (prev_entry != &map->header) 3522 vm_map_clip_end(map, prev_entry, bot); 3523 new_entry = prev_entry->next; 3524 if (new_entry->end != top || new_entry->start != bot) 3525 panic("Bad entry start/end for new stack entry"); 3526 3527 new_entry->avail_ssize = max_ssize - init_ssize; 3528 if (orient & MAP_STACK_GROWS_DOWN) 3529 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3530 if (orient & MAP_STACK_GROWS_UP) 3531 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3532 } 3533 3534 return (rv); 3535} 3536 3537static int stack_guard_page = 0; 3538TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page); 3539SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW, 3540 &stack_guard_page, 0, 3541 "Insert stack guard page ahead of the growable segments."); 3542 3543/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3544 * desired address is already mapped, or if we successfully grow 3545 * the stack. Also returns KERN_SUCCESS if addr is outside the 3546 * stack range (this is strange, but preserves compatibility with 3547 * the grow function in vm_machdep.c). 3548 */ 3549int 3550vm_map_growstack(struct proc *p, vm_offset_t addr) 3551{ 3552 vm_map_entry_t next_entry, prev_entry; 3553 vm_map_entry_t new_entry, stack_entry; 3554 struct vmspace *vm = p->p_vmspace; 3555 vm_map_t map = &vm->vm_map; 3556 vm_offset_t end; 3557 vm_size_t growsize; 3558 size_t grow_amount, max_grow; 3559 rlim_t lmemlim, stacklim, vmemlim; 3560 int is_procstack, rv; 3561 struct ucred *cred; 3562#ifdef notyet 3563 uint64_t limit; 3564#endif 3565#ifdef RACCT 3566 int error; 3567#endif 3568 3569Retry: 3570 PROC_LOCK(p); 3571 lmemlim = lim_cur(p, RLIMIT_MEMLOCK); 3572 stacklim = lim_cur(p, RLIMIT_STACK); 3573 vmemlim = lim_cur(p, RLIMIT_VMEM); 3574 PROC_UNLOCK(p); 3575 3576 vm_map_lock_read(map); 3577 3578 /* If addr is already in the entry range, no need to grow.*/ 3579 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3580 vm_map_unlock_read(map); 3581 return (KERN_SUCCESS); 3582 } 3583 3584 next_entry = prev_entry->next; 3585 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3586 /* 3587 * This entry does not grow upwards. Since the address lies 3588 * beyond this entry, the next entry (if one exists) has to 3589 * be a downward growable entry. The entry list header is 3590 * never a growable entry, so it suffices to check the flags. 3591 */ 3592 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3593 vm_map_unlock_read(map); 3594 return (KERN_SUCCESS); 3595 } 3596 stack_entry = next_entry; 3597 } else { 3598 /* 3599 * This entry grows upward. If the next entry does not at 3600 * least grow downwards, this is the entry we need to grow. 3601 * otherwise we have two possible choices and we have to 3602 * select one. 3603 */ 3604 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3605 /* 3606 * We have two choices; grow the entry closest to 3607 * the address to minimize the amount of growth. 3608 */ 3609 if (addr - prev_entry->end <= next_entry->start - addr) 3610 stack_entry = prev_entry; 3611 else 3612 stack_entry = next_entry; 3613 } else 3614 stack_entry = prev_entry; 3615 } 3616 3617 if (stack_entry == next_entry) { 3618 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3619 KASSERT(addr < stack_entry->start, ("foo")); 3620 end = (prev_entry != &map->header) ? prev_entry->end : 3621 stack_entry->start - stack_entry->avail_ssize; 3622 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3623 max_grow = stack_entry->start - end; 3624 } else { 3625 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3626 KASSERT(addr >= stack_entry->end, ("foo")); 3627 end = (next_entry != &map->header) ? next_entry->start : 3628 stack_entry->end + stack_entry->avail_ssize; 3629 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3630 max_grow = end - stack_entry->end; 3631 } 3632 3633 if (grow_amount > stack_entry->avail_ssize) { 3634 vm_map_unlock_read(map); 3635 return (KERN_NO_SPACE); 3636 } 3637 3638 /* 3639 * If there is no longer enough space between the entries nogo, and 3640 * adjust the available space. Note: this should only happen if the 3641 * user has mapped into the stack area after the stack was created, 3642 * and is probably an error. 3643 * 3644 * This also effectively destroys any guard page the user might have 3645 * intended by limiting the stack size. 3646 */ 3647 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { 3648 if (vm_map_lock_upgrade(map)) 3649 goto Retry; 3650 3651 stack_entry->avail_ssize = max_grow; 3652 3653 vm_map_unlock(map); 3654 return (KERN_NO_SPACE); 3655 } 3656 3657 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3658 3659 /* 3660 * If this is the main process stack, see if we're over the stack 3661 * limit. 3662 */ 3663 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3664 vm_map_unlock_read(map); 3665 return (KERN_NO_SPACE); 3666 } 3667#ifdef RACCT 3668 PROC_LOCK(p); 3669 if (is_procstack && 3670 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { 3671 PROC_UNLOCK(p); 3672 vm_map_unlock_read(map); 3673 return (KERN_NO_SPACE); 3674 } 3675 PROC_UNLOCK(p); 3676#endif 3677 3678 /* Round up the grow amount modulo sgrowsiz */ 3679 growsize = sgrowsiz; 3680 grow_amount = roundup(grow_amount, growsize); 3681 if (grow_amount > stack_entry->avail_ssize) 3682 grow_amount = stack_entry->avail_ssize; 3683 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3684 grow_amount = trunc_page((vm_size_t)stacklim) - 3685 ctob(vm->vm_ssize); 3686 } 3687#ifdef notyet 3688 PROC_LOCK(p); 3689 limit = racct_get_available(p, RACCT_STACK); 3690 PROC_UNLOCK(p); 3691 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3692 grow_amount = limit - ctob(vm->vm_ssize); 3693#endif 3694 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 3695 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3696 vm_map_unlock_read(map); 3697 rv = KERN_NO_SPACE; 3698 goto out; 3699 } 3700#ifdef RACCT 3701 PROC_LOCK(p); 3702 if (racct_set(p, RACCT_MEMLOCK, 3703 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3704 PROC_UNLOCK(p); 3705 vm_map_unlock_read(map); 3706 rv = KERN_NO_SPACE; 3707 goto out; 3708 } 3709 PROC_UNLOCK(p); 3710#endif 3711 } 3712 /* If we would blow our VMEM resource limit, no go */ 3713 if (map->size + grow_amount > vmemlim) { 3714 vm_map_unlock_read(map); 3715 rv = KERN_NO_SPACE; 3716 goto out; 3717 } 3718#ifdef RACCT 3719 PROC_LOCK(p); 3720 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3721 PROC_UNLOCK(p); 3722 vm_map_unlock_read(map); 3723 rv = KERN_NO_SPACE; 3724 goto out; 3725 } 3726 PROC_UNLOCK(p); 3727#endif 3728 3729 if (vm_map_lock_upgrade(map)) 3730 goto Retry; 3731 3732 if (stack_entry == next_entry) { 3733 /* 3734 * Growing downward. 3735 */ 3736 /* Get the preliminary new entry start value */ 3737 addr = stack_entry->start - grow_amount; 3738 3739 /* 3740 * If this puts us into the previous entry, cut back our 3741 * growth to the available space. Also, see the note above. 3742 */ 3743 if (addr < end) { 3744 stack_entry->avail_ssize = max_grow; 3745 addr = end; 3746 if (stack_guard_page) 3747 addr += PAGE_SIZE; 3748 } 3749 3750 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3751 next_entry->protection, next_entry->max_protection, 0); 3752 3753 /* Adjust the available stack space by the amount we grew. */ 3754 if (rv == KERN_SUCCESS) { 3755 if (prev_entry != &map->header) 3756 vm_map_clip_end(map, prev_entry, addr); 3757 new_entry = prev_entry->next; 3758 KASSERT(new_entry == stack_entry->prev, ("foo")); 3759 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3760 KASSERT(new_entry->start == addr, ("foo")); 3761 grow_amount = new_entry->end - new_entry->start; 3762 new_entry->avail_ssize = stack_entry->avail_ssize - 3763 grow_amount; 3764 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3765 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3766 } 3767 } else { 3768 /* 3769 * Growing upward. 3770 */ 3771 addr = stack_entry->end + grow_amount; 3772 3773 /* 3774 * If this puts us into the next entry, cut back our growth 3775 * to the available space. Also, see the note above. 3776 */ 3777 if (addr > end) { 3778 stack_entry->avail_ssize = end - stack_entry->end; 3779 addr = end; 3780 if (stack_guard_page) 3781 addr -= PAGE_SIZE; 3782 } 3783 3784 grow_amount = addr - stack_entry->end; 3785 cred = stack_entry->cred; 3786 if (cred == NULL && stack_entry->object.vm_object != NULL) 3787 cred = stack_entry->object.vm_object->cred; 3788 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3789 rv = KERN_NO_SPACE; 3790 /* Grow the underlying object if applicable. */ 3791 else if (stack_entry->object.vm_object == NULL || 3792 vm_object_coalesce(stack_entry->object.vm_object, 3793 stack_entry->offset, 3794 (vm_size_t)(stack_entry->end - stack_entry->start), 3795 (vm_size_t)grow_amount, cred != NULL)) { 3796 map->size += (addr - stack_entry->end); 3797 /* Update the current entry. */ 3798 stack_entry->end = addr; 3799 stack_entry->avail_ssize -= grow_amount; 3800 vm_map_entry_resize_free(map, stack_entry); 3801 rv = KERN_SUCCESS; 3802 3803 if (next_entry != &map->header) 3804 vm_map_clip_start(map, next_entry, addr); 3805 } else 3806 rv = KERN_FAILURE; 3807 } 3808 3809 if (rv == KERN_SUCCESS && is_procstack) 3810 vm->vm_ssize += btoc(grow_amount); 3811 3812 vm_map_unlock(map); 3813 3814 /* 3815 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3816 */ 3817 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3818 vm_map_wire(map, 3819 (stack_entry == next_entry) ? addr : addr - grow_amount, 3820 (stack_entry == next_entry) ? stack_entry->start : addr, 3821 (p->p_flag & P_SYSTEM) 3822 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3823 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3824 } 3825 3826out: 3827#ifdef RACCT 3828 if (rv != KERN_SUCCESS) { 3829 PROC_LOCK(p); 3830 error = racct_set(p, RACCT_VMEM, map->size); 3831 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3832 if (!old_mlock) { 3833 error = racct_set(p, RACCT_MEMLOCK, 3834 ptoa(pmap_wired_count(map->pmap))); 3835 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3836 } 3837 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3838 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3839 PROC_UNLOCK(p); 3840 } 3841#endif 3842 3843 return (rv); 3844} 3845 3846/* 3847 * Unshare the specified VM space for exec. If other processes are 3848 * mapped to it, then create a new one. The new vmspace is null. 3849 */ 3850int 3851vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3852{ 3853 struct vmspace *oldvmspace = p->p_vmspace; 3854 struct vmspace *newvmspace; 3855 3856 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3857 ("vmspace_exec recursed")); 3858 newvmspace = vmspace_alloc(minuser, maxuser, NULL); 3859 if (newvmspace == NULL) 3860 return (ENOMEM); 3861 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3862 /* 3863 * This code is written like this for prototype purposes. The 3864 * goal is to avoid running down the vmspace here, but let the 3865 * other process's that are still using the vmspace to finally 3866 * run it down. Even though there is little or no chance of blocking 3867 * here, it is a good idea to keep this form for future mods. 3868 */ 3869 PROC_VMSPACE_LOCK(p); 3870 p->p_vmspace = newvmspace; 3871 PROC_VMSPACE_UNLOCK(p); 3872 if (p == curthread->td_proc) 3873 pmap_activate(curthread); 3874 curthread->td_pflags |= TDP_EXECVMSPC; 3875 return (0); 3876} 3877 3878/* 3879 * Unshare the specified VM space for forcing COW. This 3880 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3881 */ 3882int 3883vmspace_unshare(struct proc *p) 3884{ 3885 struct vmspace *oldvmspace = p->p_vmspace; 3886 struct vmspace *newvmspace; 3887 vm_ooffset_t fork_charge; 3888 3889 if (oldvmspace->vm_refcnt == 1) 3890 return (0); 3891 fork_charge = 0; 3892 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3893 if (newvmspace == NULL) 3894 return (ENOMEM); 3895 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3896 vmspace_free(newvmspace); 3897 return (ENOMEM); 3898 } 3899 PROC_VMSPACE_LOCK(p); 3900 p->p_vmspace = newvmspace; 3901 PROC_VMSPACE_UNLOCK(p); 3902 if (p == curthread->td_proc) 3903 pmap_activate(curthread); 3904 vmspace_free(oldvmspace); 3905 return (0); 3906} 3907 3908/* 3909 * vm_map_lookup: 3910 * 3911 * Finds the VM object, offset, and 3912 * protection for a given virtual address in the 3913 * specified map, assuming a page fault of the 3914 * type specified. 3915 * 3916 * Leaves the map in question locked for read; return 3917 * values are guaranteed until a vm_map_lookup_done 3918 * call is performed. Note that the map argument 3919 * is in/out; the returned map must be used in 3920 * the call to vm_map_lookup_done. 3921 * 3922 * A handle (out_entry) is returned for use in 3923 * vm_map_lookup_done, to make that fast. 3924 * 3925 * If a lookup is requested with "write protection" 3926 * specified, the map may be changed to perform virtual 3927 * copying operations, although the data referenced will 3928 * remain the same. 3929 */ 3930int 3931vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3932 vm_offset_t vaddr, 3933 vm_prot_t fault_typea, 3934 vm_map_entry_t *out_entry, /* OUT */ 3935 vm_object_t *object, /* OUT */ 3936 vm_pindex_t *pindex, /* OUT */ 3937 vm_prot_t *out_prot, /* OUT */ 3938 boolean_t *wired) /* OUT */ 3939{ 3940 vm_map_entry_t entry; 3941 vm_map_t map = *var_map; 3942 vm_prot_t prot; 3943 vm_prot_t fault_type = fault_typea; 3944 vm_object_t eobject; 3945 vm_size_t size; 3946 struct ucred *cred; 3947 3948RetryLookup:; 3949 3950 vm_map_lock_read(map); 3951 3952 /* 3953 * Lookup the faulting address. 3954 */ 3955 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3956 vm_map_unlock_read(map); 3957 return (KERN_INVALID_ADDRESS); 3958 } 3959 3960 entry = *out_entry; 3961 3962 /* 3963 * Handle submaps. 3964 */ 3965 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3966 vm_map_t old_map = map; 3967 3968 *var_map = map = entry->object.sub_map; 3969 vm_map_unlock_read(old_map); 3970 goto RetryLookup; 3971 } 3972 3973 /* 3974 * Check whether this task is allowed to have this page. 3975 */ 3976 prot = entry->protection; 3977 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3978 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 3979 vm_map_unlock_read(map); 3980 return (KERN_PROTECTION_FAILURE); 3981 } 3982 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3983 (entry->eflags & MAP_ENTRY_COW) && 3984 (fault_type & VM_PROT_WRITE)) { 3985 vm_map_unlock_read(map); 3986 return (KERN_PROTECTION_FAILURE); 3987 } 3988 if ((fault_typea & VM_PROT_COPY) != 0 && 3989 (entry->max_protection & VM_PROT_WRITE) == 0 && 3990 (entry->eflags & MAP_ENTRY_COW) == 0) { 3991 vm_map_unlock_read(map); 3992 return (KERN_PROTECTION_FAILURE); 3993 } 3994 3995 /* 3996 * If this page is not pageable, we have to get it for all possible 3997 * accesses. 3998 */ 3999 *wired = (entry->wired_count != 0); 4000 if (*wired) 4001 fault_type = entry->protection; 4002 size = entry->end - entry->start; 4003 /* 4004 * If the entry was copy-on-write, we either ... 4005 */ 4006 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4007 /* 4008 * If we want to write the page, we may as well handle that 4009 * now since we've got the map locked. 4010 * 4011 * If we don't need to write the page, we just demote the 4012 * permissions allowed. 4013 */ 4014 if ((fault_type & VM_PROT_WRITE) != 0 || 4015 (fault_typea & VM_PROT_COPY) != 0) { 4016 /* 4017 * Make a new object, and place it in the object 4018 * chain. Note that no new references have appeared 4019 * -- one just moved from the map to the new 4020 * object. 4021 */ 4022 if (vm_map_lock_upgrade(map)) 4023 goto RetryLookup; 4024 4025 if (entry->cred == NULL) { 4026 /* 4027 * The debugger owner is charged for 4028 * the memory. 4029 */ 4030 cred = curthread->td_ucred; 4031 crhold(cred); 4032 if (!swap_reserve_by_cred(size, cred)) { 4033 crfree(cred); 4034 vm_map_unlock(map); 4035 return (KERN_RESOURCE_SHORTAGE); 4036 } 4037 entry->cred = cred; 4038 } 4039 vm_object_shadow(&entry->object.vm_object, 4040 &entry->offset, size); 4041 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4042 eobject = entry->object.vm_object; 4043 if (eobject->cred != NULL) { 4044 /* 4045 * The object was not shadowed. 4046 */ 4047 swap_release_by_cred(size, entry->cred); 4048 crfree(entry->cred); 4049 entry->cred = NULL; 4050 } else if (entry->cred != NULL) { 4051 VM_OBJECT_WLOCK(eobject); 4052 eobject->cred = entry->cred; 4053 eobject->charge = size; 4054 VM_OBJECT_WUNLOCK(eobject); 4055 entry->cred = NULL; 4056 } 4057 4058 vm_map_lock_downgrade(map); 4059 } else { 4060 /* 4061 * We're attempting to read a copy-on-write page -- 4062 * don't allow writes. 4063 */ 4064 prot &= ~VM_PROT_WRITE; 4065 } 4066 } 4067 4068 /* 4069 * Create an object if necessary. 4070 */ 4071 if (entry->object.vm_object == NULL && 4072 !map->system_map) { 4073 if (vm_map_lock_upgrade(map)) 4074 goto RetryLookup; 4075 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4076 atop(size)); 4077 entry->offset = 0; 4078 if (entry->cred != NULL) { 4079 VM_OBJECT_WLOCK(entry->object.vm_object); 4080 entry->object.vm_object->cred = entry->cred; 4081 entry->object.vm_object->charge = size; 4082 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4083 entry->cred = NULL; 4084 } 4085 vm_map_lock_downgrade(map); 4086 } 4087 4088 /* 4089 * Return the object/offset from this entry. If the entry was 4090 * copy-on-write or empty, it has been fixed up. 4091 */ 4092 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4093 *object = entry->object.vm_object; 4094 4095 *out_prot = prot; 4096 return (KERN_SUCCESS); 4097} 4098 4099/* 4100 * vm_map_lookup_locked: 4101 * 4102 * Lookup the faulting address. A version of vm_map_lookup that returns 4103 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4104 */ 4105int 4106vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4107 vm_offset_t vaddr, 4108 vm_prot_t fault_typea, 4109 vm_map_entry_t *out_entry, /* OUT */ 4110 vm_object_t *object, /* OUT */ 4111 vm_pindex_t *pindex, /* OUT */ 4112 vm_prot_t *out_prot, /* OUT */ 4113 boolean_t *wired) /* OUT */ 4114{ 4115 vm_map_entry_t entry; 4116 vm_map_t map = *var_map; 4117 vm_prot_t prot; 4118 vm_prot_t fault_type = fault_typea; 4119 4120 /* 4121 * Lookup the faulting address. 4122 */ 4123 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4124 return (KERN_INVALID_ADDRESS); 4125 4126 entry = *out_entry; 4127 4128 /* 4129 * Fail if the entry refers to a submap. 4130 */ 4131 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4132 return (KERN_FAILURE); 4133 4134 /* 4135 * Check whether this task is allowed to have this page. 4136 */ 4137 prot = entry->protection; 4138 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4139 if ((fault_type & prot) != fault_type) 4140 return (KERN_PROTECTION_FAILURE); 4141 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 4142 (entry->eflags & MAP_ENTRY_COW) && 4143 (fault_type & VM_PROT_WRITE)) 4144 return (KERN_PROTECTION_FAILURE); 4145 4146 /* 4147 * If this page is not pageable, we have to get it for all possible 4148 * accesses. 4149 */ 4150 *wired = (entry->wired_count != 0); 4151 if (*wired) 4152 fault_type = entry->protection; 4153 4154 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4155 /* 4156 * Fail if the entry was copy-on-write for a write fault. 4157 */ 4158 if (fault_type & VM_PROT_WRITE) 4159 return (KERN_FAILURE); 4160 /* 4161 * We're attempting to read a copy-on-write page -- 4162 * don't allow writes. 4163 */ 4164 prot &= ~VM_PROT_WRITE; 4165 } 4166 4167 /* 4168 * Fail if an object should be created. 4169 */ 4170 if (entry->object.vm_object == NULL && !map->system_map) 4171 return (KERN_FAILURE); 4172 4173 /* 4174 * Return the object/offset from this entry. If the entry was 4175 * copy-on-write or empty, it has been fixed up. 4176 */ 4177 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4178 *object = entry->object.vm_object; 4179 4180 *out_prot = prot; 4181 return (KERN_SUCCESS); 4182} 4183 4184/* 4185 * vm_map_lookup_done: 4186 * 4187 * Releases locks acquired by a vm_map_lookup 4188 * (according to the handle returned by that lookup). 4189 */ 4190void 4191vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4192{ 4193 /* 4194 * Unlock the main-level map 4195 */ 4196 vm_map_unlock_read(map); 4197} 4198 4199#include "opt_ddb.h" 4200#ifdef DDB 4201#include <sys/kernel.h> 4202 4203#include <ddb/ddb.h> 4204 4205static void 4206vm_map_print(vm_map_t map) 4207{ 4208 vm_map_entry_t entry; 4209 4210 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4211 (void *)map, 4212 (void *)map->pmap, map->nentries, map->timestamp); 4213 4214 db_indent += 2; 4215 for (entry = map->header.next; entry != &map->header; 4216 entry = entry->next) { 4217 db_iprintf("map entry %p: start=%p, end=%p\n", 4218 (void *)entry, (void *)entry->start, (void *)entry->end); 4219 { 4220 static char *inheritance_name[4] = 4221 {"share", "copy", "none", "donate_copy"}; 4222 4223 db_iprintf(" prot=%x/%x/%s", 4224 entry->protection, 4225 entry->max_protection, 4226 inheritance_name[(int)(unsigned char)entry->inheritance]); 4227 if (entry->wired_count != 0) 4228 db_printf(", wired"); 4229 } 4230 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4231 db_printf(", share=%p, offset=0x%jx\n", 4232 (void *)entry->object.sub_map, 4233 (uintmax_t)entry->offset); 4234 if ((entry->prev == &map->header) || 4235 (entry->prev->object.sub_map != 4236 entry->object.sub_map)) { 4237 db_indent += 2; 4238 vm_map_print((vm_map_t)entry->object.sub_map); 4239 db_indent -= 2; 4240 } 4241 } else { 4242 if (entry->cred != NULL) 4243 db_printf(", ruid %d", entry->cred->cr_ruid); 4244 db_printf(", object=%p, offset=0x%jx", 4245 (void *)entry->object.vm_object, 4246 (uintmax_t)entry->offset); 4247 if (entry->object.vm_object && entry->object.vm_object->cred) 4248 db_printf(", obj ruid %d charge %jx", 4249 entry->object.vm_object->cred->cr_ruid, 4250 (uintmax_t)entry->object.vm_object->charge); 4251 if (entry->eflags & MAP_ENTRY_COW) 4252 db_printf(", copy (%s)", 4253 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4254 db_printf("\n"); 4255 4256 if ((entry->prev == &map->header) || 4257 (entry->prev->object.vm_object != 4258 entry->object.vm_object)) { 4259 db_indent += 2; 4260 vm_object_print((db_expr_t)(intptr_t) 4261 entry->object.vm_object, 4262 0, 0, (char *)0); 4263 db_indent -= 2; 4264 } 4265 } 4266 } 4267 db_indent -= 2; 4268} 4269 4270DB_SHOW_COMMAND(map, map) 4271{ 4272 4273 if (!have_addr) { 4274 db_printf("usage: show map <addr>\n"); 4275 return; 4276 } 4277 vm_map_print((vm_map_t)addr); 4278} 4279 4280DB_SHOW_COMMAND(procvm, procvm) 4281{ 4282 struct proc *p; 4283 4284 if (have_addr) { 4285 p = (struct proc *) addr; 4286 } else { 4287 p = curproc; 4288 } 4289 4290 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4291 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4292 (void *)vmspace_pmap(p->p_vmspace)); 4293 4294 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4295} 4296 4297#endif /* DDB */ 4298