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("$FreeBSD$"); 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/ktr.h> 71 #include <sys/lock.h> 72 #include <sys/mutex.h> 73 #include <sys/proc.h> 74 #include <sys/vmmeter.h> 75 #include <sys/mman.h> 76 #include <sys/vnode.h> 77 #include <sys/resourcevar.h> 78 #include <sys/file.h> 79 #include <sys/sysent.h> 80 #include <sys/shm.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_param.h> 84 #include <vm/pmap.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_page.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_pager.h> 89 #include <vm/vm_kern.h> 90 #include <vm/vm_extern.h> 91 #include <vm/swap_pager.h> 92 #include <vm/uma.h> 93 94 /* 95 * Virtual memory maps provide for the mapping, protection, 96 * and sharing of virtual memory objects. In addition, 97 * this module provides for an efficient virtual copy of 98 * memory from one map to another. 99 * 100 * Synchronization is required prior to most operations. 101 * 102 * Maps consist of an ordered doubly-linked list of simple 103 * entries; a self-adjusting binary search tree of these 104 * entries is used to speed up lookups. 105 * 106 * Since portions of maps are specified by start/end addresses, 107 * which may not align with existing map entries, all 108 * routines merely "clip" entries to these start/end values. 109 * [That is, an entry is split into two, bordering at a 110 * start or end value.] Note that these clippings may not 111 * always be necessary (as the two resulting entries are then 112 * not changed); however, the clipping is done for convenience. 113 * 114 * As mentioned above, virtual copy operations are performed 115 * by copying VM object references from one map to 116 * another, and then marking both regions as copy-on-write. 117 */ 118 119 /* 120 * vm_map_startup: 121 * 122 * Initialize the vm_map module. Must be called before 123 * any other vm_map routines. 124 * 125 * Map and entry structures are allocated from the general 126 * purpose memory pool with some exceptions: 127 * 128 * - The kernel map and kmem submap are allocated statically. 129 * - Kernel map entries are allocated out of a static pool. 130 * 131 * These restrictions are necessary since malloc() uses the 132 * maps and requires map entries. 133 */ 134 135 static struct mtx map_sleep_mtx; 136 static uma_zone_t mapentzone; 137 static uma_zone_t kmapentzone; 138 static uma_zone_t mapzone; 139 static uma_zone_t vmspace_zone; 140 static struct vm_object kmapentobj; 141 static int vmspace_zinit(void *mem, int size, int flags); 142 static void vmspace_zfini(void *mem, int size); 143 static int vm_map_zinit(void *mem, int ize, int flags); 144 static void vm_map_zfini(void *mem, int size); 145 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max); 146 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 147 #ifdef INVARIANTS 148 static void vm_map_zdtor(void *mem, int size, void *arg); 149 static void vmspace_zdtor(void *mem, int size, void *arg); 150 #endif 151 152 #define ENTRY_CHARGED(e) ((e)->uip != NULL || \ 153 ((e)->object.vm_object != NULL && (e)->object.vm_object->uip != NULL && \ 154 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 155 156 /* 157 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 158 * stable. 159 */ 160 #define PROC_VMSPACE_LOCK(p) do { } while (0) 161 #define PROC_VMSPACE_UNLOCK(p) do { } while (0) 162 163 /* 164 * VM_MAP_RANGE_CHECK: [ internal use only ] 165 * 166 * Asserts that the starting and ending region 167 * addresses fall within the valid range of the map. 168 */ 169 #define VM_MAP_RANGE_CHECK(map, start, end) \ 170 { \ 171 if (start < vm_map_min(map)) \ 172 start = vm_map_min(map); \ 173 if (end > vm_map_max(map)) \ 174 end = vm_map_max(map); \ 175 if (start > end) \ 176 start = end; \ 177 } 178 179 void 180 vm_map_startup(void) 181 { 182 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 183 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 184 #ifdef INVARIANTS 185 vm_map_zdtor, 186 #else 187 NULL, 188 #endif 189 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 190 uma_prealloc(mapzone, MAX_KMAP); 191 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 193 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 194 uma_prealloc(kmapentzone, MAX_KMAPENT); 195 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 196 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 197 } 198 199 static void 200 vmspace_zfini(void *mem, int size) 201 { 202 struct vmspace *vm; 203 204 vm = (struct vmspace *)mem; 205 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 206 } 207 208 static int 209 vmspace_zinit(void *mem, int size, int flags) 210 { 211 struct vmspace *vm; 212 213 vm = (struct vmspace *)mem; 214 215 vm->vm_map.pmap = NULL; 216 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 217 return (0); 218 } 219 220 static void 221 vm_map_zfini(void *mem, int size) 222 { 223 vm_map_t map; 224 225 map = (vm_map_t)mem; 226 mtx_destroy(&map->system_mtx); 227 sx_destroy(&map->lock); 228 } 229 230 static int 231 vm_map_zinit(void *mem, int size, int flags) 232 { 233 vm_map_t map; 234 235 map = (vm_map_t)mem; 236 map->nentries = 0; 237 map->size = 0; 238 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 239 sx_init(&map->lock, "user map"); 240 return (0); 241 } 242 243 #ifdef INVARIANTS 244 static void 245 vmspace_zdtor(void *mem, int size, void *arg) 246 { 247 struct vmspace *vm; 248 249 vm = (struct vmspace *)mem; 250 251 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 252 } 253 static void 254 vm_map_zdtor(void *mem, int size, void *arg) 255 { 256 vm_map_t map; 257 258 map = (vm_map_t)mem; 259 KASSERT(map->nentries == 0, 260 ("map %p nentries == %d on free.", 261 map, map->nentries)); 262 KASSERT(map->size == 0, 263 ("map %p size == %lu on free.", 264 map, (unsigned long)map->size)); 265 } 266 #endif /* INVARIANTS */ 267 268 /* 269 * Allocate a vmspace structure, including a vm_map and pmap, 270 * and initialize those structures. The refcnt is set to 1. 271 */ 272 struct vmspace * 273 vmspace_alloc(min, max) 274 vm_offset_t min, max; 275 { 276 struct vmspace *vm; 277 278 vm = uma_zalloc(vmspace_zone, M_WAITOK); 279 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) { 280 uma_zfree(vmspace_zone, vm); 281 return (NULL); 282 } 283 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 284 _vm_map_init(&vm->vm_map, min, max); 285 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ 286 vm->vm_refcnt = 1; 287 vm->vm_shm = NULL; 288 vm->vm_swrss = 0; 289 vm->vm_tsize = 0; 290 vm->vm_dsize = 0; 291 vm->vm_ssize = 0; 292 vm->vm_taddr = 0; 293 vm->vm_daddr = 0; 294 vm->vm_maxsaddr = 0; 295 return (vm); 296 } 297 298 void 299 vm_init2(void) 300 { 301 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 302 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 + 303 maxproc * 2 + maxfiles); 304 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 305 #ifdef INVARIANTS 306 vmspace_zdtor, 307 #else 308 NULL, 309 #endif 310 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 311 } 312 313 static inline void 314 vmspace_dofree(struct vmspace *vm) 315 { 316 CTR1(KTR_VM, "vmspace_free: %p", vm); 317 318 /* 319 * Make sure any SysV shm is freed, it might not have been in 320 * exit1(). 321 */ 322 shmexit(vm); 323 324 /* 325 * Lock the map, to wait out all other references to it. 326 * Delete all of the mappings and pages they hold, then call 327 * the pmap module to reclaim anything left. 328 */ 329 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 330 vm->vm_map.max_offset); 331 332 /* 333 * XXX Comment out the pmap_release call for now. The 334 * vmspace_zone is marked as UMA_ZONE_NOFREE, and bugs cause 335 * pmap.resident_count to be != 0 on exit sometimes. 336 */ 337 /* pmap_release(vmspace_pmap(vm)); */ 338 uma_zfree(vmspace_zone, vm); 339 } 340 341 void 342 vmspace_free(struct vmspace *vm) 343 { 344 int refcnt; 345 346 if (vm->vm_refcnt == 0) 347 panic("vmspace_free: attempt to free already freed vmspace"); 348 349 do 350 refcnt = vm->vm_refcnt; 351 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 352 if (refcnt == 1) 353 vmspace_dofree(vm); 354 } 355 356 void 357 vmspace_exitfree(struct proc *p) 358 { 359 struct vmspace *vm; 360 361 PROC_VMSPACE_LOCK(p); 362 vm = p->p_vmspace; 363 p->p_vmspace = NULL; 364 PROC_VMSPACE_UNLOCK(p); 365 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 366 vmspace_free(vm); 367 } 368 369 void 370 vmspace_exit(struct thread *td) 371 { 372 int refcnt; 373 struct vmspace *vm; 374 struct proc *p; 375 376 /* 377 * Release user portion of address space. 378 * This releases references to vnodes, 379 * which could cause I/O if the file has been unlinked. 380 * Need to do this early enough that we can still sleep. 381 * 382 * The last exiting process to reach this point releases as 383 * much of the environment as it can. vmspace_dofree() is the 384 * slower fallback in case another process had a temporary 385 * reference to the vmspace. 386 */ 387 388 p = td->td_proc; 389 vm = p->p_vmspace; 390 atomic_add_int(&vmspace0.vm_refcnt, 1); 391 do { 392 refcnt = vm->vm_refcnt; 393 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 394 /* Switch now since other proc might free vmspace */ 395 PROC_VMSPACE_LOCK(p); 396 p->p_vmspace = &vmspace0; 397 PROC_VMSPACE_UNLOCK(p); 398 pmap_activate(td); 399 } 400 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 401 if (refcnt == 1) { 402 if (p->p_vmspace != vm) { 403 /* vmspace not yet freed, switch back */ 404 PROC_VMSPACE_LOCK(p); 405 p->p_vmspace = vm; 406 PROC_VMSPACE_UNLOCK(p); 407 pmap_activate(td); 408 } 409 pmap_remove_pages(vmspace_pmap(vm)); 410 /* Switch now since this proc will free vmspace */ 411 PROC_VMSPACE_LOCK(p); 412 p->p_vmspace = &vmspace0; 413 PROC_VMSPACE_UNLOCK(p); 414 pmap_activate(td); 415 vmspace_dofree(vm); 416 } 417 } 418 419 /* Acquire reference to vmspace owned by another process. */ 420 421 struct vmspace * 422 vmspace_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 } 448 449 void 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 (void)_sx_xlock(&map->lock, 0, file, line); 457 map->timestamp++; 458 } 459 460 void 461 _vm_map_unlock(vm_map_t map, const char *file, int line) 462 { 463 vm_map_entry_t free_entry, entry; 464 vm_object_t object; 465 466 free_entry = map->deferred_freelist; 467 map->deferred_freelist = NULL; 468 469 if (map->system_map) 470 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 471 else 472 _sx_xunlock(&map->lock, file, line); 473 474 while (free_entry != NULL) { 475 entry = free_entry; 476 free_entry = free_entry->next; 477 478 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 479 object = entry->object.vm_object; 480 vm_object_deallocate(object); 481 } 482 483 vm_map_entry_dispose(map, entry); 484 } 485 } 486 487 void 488 _vm_map_lock_read(vm_map_t map, const char *file, int line) 489 { 490 491 if (map->system_map) 492 _mtx_lock_flags(&map->system_mtx, 0, file, line); 493 else 494 (void)_sx_slock(&map->lock, 0, file, line); 495 } 496 497 void 498 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 499 { 500 501 if (map->system_map) 502 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 503 else 504 _sx_sunlock(&map->lock, file, line); 505 } 506 507 int 508 _vm_map_trylock(vm_map_t map, const char *file, int line) 509 { 510 int error; 511 512 error = map->system_map ? 513 !_mtx_trylock(&map->system_mtx, 0, file, line) : 514 !_sx_try_xlock(&map->lock, file, line); 515 if (error == 0) 516 map->timestamp++; 517 return (error == 0); 518 } 519 520 int 521 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 522 { 523 int error; 524 525 error = map->system_map ? 526 !_mtx_trylock(&map->system_mtx, 0, file, line) : 527 !_sx_try_slock(&map->lock, file, line); 528 return (error == 0); 529 } 530 531 /* 532 * _vm_map_lock_upgrade: [ internal use only ] 533 * 534 * Tries to upgrade a read (shared) lock on the specified map to a write 535 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 536 * non-zero value if the upgrade fails. If the upgrade fails, the map is 537 * returned without a read or write lock held. 538 * 539 * Requires that the map be read locked. 540 */ 541 int 542 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 543 { 544 unsigned int last_timestamp; 545 546 if (map->system_map) { 547 #ifdef INVARIANTS 548 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 549 #endif 550 } else { 551 if (!_sx_try_upgrade(&map->lock, file, line)) { 552 last_timestamp = map->timestamp; 553 _sx_sunlock(&map->lock, file, line); 554 /* 555 * If the map's timestamp does not change while the 556 * map is unlocked, then the upgrade succeeds. 557 */ 558 (void)_sx_xlock(&map->lock, 0, file, line); 559 if (last_timestamp != map->timestamp) { 560 _sx_xunlock(&map->lock, file, line); 561 return (1); 562 } 563 } 564 } 565 map->timestamp++; 566 return (0); 567 } 568 569 void 570 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 571 { 572 573 if (map->system_map) { 574 #ifdef INVARIANTS 575 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 576 #endif 577 } else 578 _sx_downgrade(&map->lock, file, line); 579 } 580 581 /* 582 * vm_map_locked: 583 * 584 * Returns a non-zero value if the caller holds a write (exclusive) lock 585 * on the specified map and the value "0" otherwise. 586 */ 587 int 588 vm_map_locked(vm_map_t map) 589 { 590 591 if (map->system_map) 592 return (mtx_owned(&map->system_mtx)); 593 else 594 return (sx_xlocked(&map->lock)); 595 } 596 597 #ifdef INVARIANTS 598 static void 599 _vm_map_assert_locked(vm_map_t map, const char *file, int line) 600 { 601 602 if (map->system_map) 603 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 604 else 605 _sx_assert(&map->lock, SA_XLOCKED, file, line); 606 } 607 608 #if 0 609 static void 610 _vm_map_assert_locked_read(vm_map_t map, const char *file, int line) 611 { 612 613 if (map->system_map) 614 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 615 else 616 _sx_assert(&map->lock, SA_SLOCKED, file, line); 617 } 618 #endif 619 620 #define VM_MAP_ASSERT_LOCKED(map) \ 621 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 622 #define VM_MAP_ASSERT_LOCKED_READ(map) \ 623 _vm_map_assert_locked_read(map, LOCK_FILE, LOCK_LINE) 624 #else 625 #define VM_MAP_ASSERT_LOCKED(map) 626 #define VM_MAP_ASSERT_LOCKED_READ(map) 627 #endif 628 629 /* 630 * vm_map_unlock_and_wait: 631 */ 632 int 633 vm_map_unlock_and_wait(vm_map_t map, int timo) 634 { 635 636 mtx_lock(&map_sleep_mtx); 637 vm_map_unlock(map); 638 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", timo)); 639 } 640 641 /* 642 * vm_map_wakeup: 643 */ 644 void 645 vm_map_wakeup(vm_map_t map) 646 { 647 648 /* 649 * Acquire and release map_sleep_mtx to prevent a wakeup() 650 * from being performed (and lost) between the vm_map_unlock() 651 * and the msleep() in vm_map_unlock_and_wait(). 652 */ 653 mtx_lock(&map_sleep_mtx); 654 mtx_unlock(&map_sleep_mtx); 655 wakeup(&map->root); 656 } 657 658 long 659 vmspace_resident_count(struct vmspace *vmspace) 660 { 661 return pmap_resident_count(vmspace_pmap(vmspace)); 662 } 663 664 long 665 vmspace_wired_count(struct vmspace *vmspace) 666 { 667 return pmap_wired_count(vmspace_pmap(vmspace)); 668 } 669 670 /* 671 * vm_map_create: 672 * 673 * Creates and returns a new empty VM map with 674 * the given physical map structure, and having 675 * the given lower and upper address bounds. 676 */ 677 vm_map_t 678 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 679 { 680 vm_map_t result; 681 682 result = uma_zalloc(mapzone, M_WAITOK); 683 CTR1(KTR_VM, "vm_map_create: %p", result); 684 _vm_map_init(result, min, max); 685 result->pmap = pmap; 686 return (result); 687 } 688 689 /* 690 * Initialize an existing vm_map structure 691 * such as that in the vmspace structure. 692 * The pmap is set elsewhere. 693 */ 694 static void 695 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 696 { 697 698 map->header.next = map->header.prev = &map->header; 699 map->needs_wakeup = FALSE; 700 map->system_map = 0; 701 map->min_offset = min; 702 map->max_offset = max; 703 map->flags = 0; 704 map->root = NULL; 705 map->timestamp = 0; 706 map->deferred_freelist = NULL; 707 } 708 709 void 710 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 711 { 712 _vm_map_init(map, min, max); 713 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 714 sx_init(&map->lock, "user map"); 715 } 716 717 /* 718 * vm_map_entry_dispose: [ internal use only ] 719 * 720 * Inverse of vm_map_entry_create. 721 */ 722 static void 723 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 724 { 725 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 726 } 727 728 /* 729 * vm_map_entry_create: [ internal use only ] 730 * 731 * Allocates a VM map entry for insertion. 732 * No entry fields are filled in. 733 */ 734 static vm_map_entry_t 735 vm_map_entry_create(vm_map_t map) 736 { 737 vm_map_entry_t new_entry; 738 739 if (map->system_map) 740 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 741 else 742 new_entry = uma_zalloc(mapentzone, M_WAITOK); 743 if (new_entry == NULL) 744 panic("vm_map_entry_create: kernel resources exhausted"); 745 return (new_entry); 746 } 747 748 /* 749 * vm_map_entry_set_behavior: 750 * 751 * Set the expected access behavior, either normal, random, or 752 * sequential. 753 */ 754 static inline void 755 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 756 { 757 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 758 (behavior & MAP_ENTRY_BEHAV_MASK); 759 } 760 761 /* 762 * vm_map_entry_set_max_free: 763 * 764 * Set the max_free field in a vm_map_entry. 765 */ 766 static inline void 767 vm_map_entry_set_max_free(vm_map_entry_t entry) 768 { 769 770 entry->max_free = entry->adj_free; 771 if (entry->left != NULL && entry->left->max_free > entry->max_free) 772 entry->max_free = entry->left->max_free; 773 if (entry->right != NULL && entry->right->max_free > entry->max_free) 774 entry->max_free = entry->right->max_free; 775 } 776 777 /* 778 * vm_map_entry_splay: 779 * 780 * The Sleator and Tarjan top-down splay algorithm with the 781 * following variation. Max_free must be computed bottom-up, so 782 * on the downward pass, maintain the left and right spines in 783 * reverse order. Then, make a second pass up each side to fix 784 * the pointers and compute max_free. The time bound is O(log n) 785 * amortized. 786 * 787 * The new root is the vm_map_entry containing "addr", or else an 788 * adjacent entry (lower or higher) if addr is not in the tree. 789 * 790 * The map must be locked, and leaves it so. 791 * 792 * Returns: the new root. 793 */ 794 static vm_map_entry_t 795 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 796 { 797 vm_map_entry_t llist, rlist; 798 vm_map_entry_t ltree, rtree; 799 vm_map_entry_t y; 800 801 /* Special case of empty tree. */ 802 if (root == NULL) 803 return (root); 804 805 /* 806 * Pass One: Splay down the tree until we find addr or a NULL 807 * pointer where addr would go. llist and rlist are the two 808 * sides in reverse order (bottom-up), with llist linked by 809 * the right pointer and rlist linked by the left pointer in 810 * the vm_map_entry. Wait until Pass Two to set max_free on 811 * the two spines. 812 */ 813 llist = NULL; 814 rlist = NULL; 815 for (;;) { 816 /* root is never NULL in here. */ 817 if (addr < root->start) { 818 y = root->left; 819 if (y == NULL) 820 break; 821 if (addr < y->start && y->left != NULL) { 822 /* Rotate right and put y on rlist. */ 823 root->left = y->right; 824 y->right = root; 825 vm_map_entry_set_max_free(root); 826 root = y->left; 827 y->left = rlist; 828 rlist = y; 829 } else { 830 /* Put root on rlist. */ 831 root->left = rlist; 832 rlist = root; 833 root = y; 834 } 835 } else if (addr >= root->end) { 836 y = root->right; 837 if (y == NULL) 838 break; 839 if (addr >= y->end && y->right != NULL) { 840 /* Rotate left and put y on llist. */ 841 root->right = y->left; 842 y->left = root; 843 vm_map_entry_set_max_free(root); 844 root = y->right; 845 y->right = llist; 846 llist = y; 847 } else { 848 /* Put root on llist. */ 849 root->right = llist; 850 llist = root; 851 root = y; 852 } 853 } else 854 break; 855 } 856 857 /* 858 * Pass Two: Walk back up the two spines, flip the pointers 859 * and set max_free. The subtrees of the root go at the 860 * bottom of llist and rlist. 861 */ 862 ltree = root->left; 863 while (llist != NULL) { 864 y = llist->right; 865 llist->right = ltree; 866 vm_map_entry_set_max_free(llist); 867 ltree = llist; 868 llist = y; 869 } 870 rtree = root->right; 871 while (rlist != NULL) { 872 y = rlist->left; 873 rlist->left = rtree; 874 vm_map_entry_set_max_free(rlist); 875 rtree = rlist; 876 rlist = y; 877 } 878 879 /* 880 * Final assembly: add ltree and rtree as subtrees of root. 881 */ 882 root->left = ltree; 883 root->right = rtree; 884 vm_map_entry_set_max_free(root); 885 886 return (root); 887 } 888 889 /* 890 * vm_map_entry_{un,}link: 891 * 892 * Insert/remove entries from maps. 893 */ 894 static void 895 vm_map_entry_link(vm_map_t map, 896 vm_map_entry_t after_where, 897 vm_map_entry_t entry) 898 { 899 900 CTR4(KTR_VM, 901 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 902 map->nentries, entry, after_where); 903 VM_MAP_ASSERT_LOCKED(map); 904 map->nentries++; 905 entry->prev = after_where; 906 entry->next = after_where->next; 907 entry->next->prev = entry; 908 after_where->next = entry; 909 910 if (after_where != &map->header) { 911 if (after_where != map->root) 912 vm_map_entry_splay(after_where->start, map->root); 913 entry->right = after_where->right; 914 entry->left = after_where; 915 after_where->right = NULL; 916 after_where->adj_free = entry->start - after_where->end; 917 vm_map_entry_set_max_free(after_where); 918 } else { 919 entry->right = map->root; 920 entry->left = NULL; 921 } 922 entry->adj_free = (entry->next == &map->header ? map->max_offset : 923 entry->next->start) - entry->end; 924 vm_map_entry_set_max_free(entry); 925 map->root = entry; 926 } 927 928 static void 929 vm_map_entry_unlink(vm_map_t map, 930 vm_map_entry_t entry) 931 { 932 vm_map_entry_t next, prev, root; 933 934 VM_MAP_ASSERT_LOCKED(map); 935 if (entry != map->root) 936 vm_map_entry_splay(entry->start, map->root); 937 if (entry->left == NULL) 938 root = entry->right; 939 else { 940 root = vm_map_entry_splay(entry->start, entry->left); 941 root->right = entry->right; 942 root->adj_free = (entry->next == &map->header ? map->max_offset : 943 entry->next->start) - root->end; 944 vm_map_entry_set_max_free(root); 945 } 946 map->root = root; 947 948 prev = entry->prev; 949 next = entry->next; 950 next->prev = prev; 951 prev->next = next; 952 map->nentries--; 953 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 954 map->nentries, entry); 955 } 956 957 /* 958 * vm_map_entry_resize_free: 959 * 960 * Recompute the amount of free space following a vm_map_entry 961 * and propagate that value up the tree. Call this function after 962 * resizing a map entry in-place, that is, without a call to 963 * vm_map_entry_link() or _unlink(). 964 * 965 * The map must be locked, and leaves it so. 966 */ 967 static void 968 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 969 { 970 971 /* 972 * Using splay trees without parent pointers, propagating 973 * max_free up the tree is done by moving the entry to the 974 * root and making the change there. 975 */ 976 if (entry != map->root) 977 map->root = vm_map_entry_splay(entry->start, map->root); 978 979 entry->adj_free = (entry->next == &map->header ? map->max_offset : 980 entry->next->start) - entry->end; 981 vm_map_entry_set_max_free(entry); 982 } 983 984 /* 985 * vm_map_lookup_entry: [ internal use only ] 986 * 987 * Finds the map entry containing (or 988 * immediately preceding) the specified address 989 * in the given map; the entry is returned 990 * in the "entry" parameter. The boolean 991 * result indicates whether the address is 992 * actually contained in the map. 993 */ 994 boolean_t 995 vm_map_lookup_entry( 996 vm_map_t map, 997 vm_offset_t address, 998 vm_map_entry_t *entry) /* OUT */ 999 { 1000 vm_map_entry_t cur; 1001 boolean_t locked; 1002 1003 /* 1004 * If the map is empty, then the map entry immediately preceding 1005 * "address" is the map's header. 1006 */ 1007 cur = map->root; 1008 if (cur == NULL) 1009 *entry = &map->header; 1010 else if (address >= cur->start && cur->end > address) { 1011 *entry = cur; 1012 return (TRUE); 1013 } else if ((locked = vm_map_locked(map)) || 1014 sx_try_upgrade(&map->lock)) { 1015 /* 1016 * Splay requires a write lock on the map. However, it only 1017 * restructures the binary search tree; it does not otherwise 1018 * change the map. Thus, the map's timestamp need not change 1019 * on a temporary upgrade. 1020 */ 1021 map->root = cur = vm_map_entry_splay(address, cur); 1022 if (!locked) 1023 sx_downgrade(&map->lock); 1024 1025 /* 1026 * If "address" is contained within a map entry, the new root 1027 * is that map entry. Otherwise, the new root is a map entry 1028 * immediately before or after "address". 1029 */ 1030 if (address >= cur->start) { 1031 *entry = cur; 1032 if (cur->end > address) 1033 return (TRUE); 1034 } else 1035 *entry = cur->prev; 1036 } else 1037 /* 1038 * Since the map is only locked for read access, perform a 1039 * standard binary search tree lookup for "address". 1040 */ 1041 for (;;) { 1042 if (address < cur->start) { 1043 if (cur->left == NULL) { 1044 *entry = cur->prev; 1045 break; 1046 } 1047 cur = cur->left; 1048 } else if (cur->end > address) { 1049 *entry = cur; 1050 return (TRUE); 1051 } else { 1052 if (cur->right == NULL) { 1053 *entry = cur; 1054 break; 1055 } 1056 cur = cur->right; 1057 } 1058 } 1059 return (FALSE); 1060 } 1061 1062 /* 1063 * vm_map_insert: 1064 * 1065 * Inserts the given whole VM object into the target 1066 * map at the specified address range. The object's 1067 * size should match that of the address range. 1068 * 1069 * Requires that the map be locked, and leaves it so. 1070 * 1071 * If object is non-NULL, ref count must be bumped by caller 1072 * prior to making call to account for the new entry. 1073 */ 1074 int 1075 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1076 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 1077 int cow) 1078 { 1079 vm_map_entry_t new_entry; 1080 vm_map_entry_t prev_entry; 1081 vm_map_entry_t temp_entry; 1082 vm_eflags_t protoeflags; 1083 struct uidinfo *uip; 1084 boolean_t charge_prev_obj; 1085 1086 VM_MAP_ASSERT_LOCKED(map); 1087 1088 /* 1089 * Check that the start and end points are not bogus. 1090 */ 1091 if ((start < map->min_offset) || (end > map->max_offset) || 1092 (start >= end)) 1093 return (KERN_INVALID_ADDRESS); 1094 1095 /* 1096 * Find the entry prior to the proposed starting address; if it's part 1097 * of an existing entry, this range is bogus. 1098 */ 1099 if (vm_map_lookup_entry(map, start, &temp_entry)) 1100 return (KERN_NO_SPACE); 1101 1102 prev_entry = temp_entry; 1103 1104 /* 1105 * Assert that the next entry doesn't overlap the end point. 1106 */ 1107 if ((prev_entry->next != &map->header) && 1108 (prev_entry->next->start < end)) 1109 return (KERN_NO_SPACE); 1110 1111 protoeflags = 0; 1112 charge_prev_obj = FALSE; 1113 1114 if (cow & MAP_COPY_ON_WRITE) 1115 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 1116 1117 if (cow & MAP_NOFAULT) { 1118 protoeflags |= MAP_ENTRY_NOFAULT; 1119 1120 KASSERT(object == NULL, 1121 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1122 } 1123 if (cow & MAP_DISABLE_SYNCER) 1124 protoeflags |= MAP_ENTRY_NOSYNC; 1125 if (cow & MAP_DISABLE_COREDUMP) 1126 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1127 1128 uip = NULL; 1129 KASSERT((object != kmem_object && object != kernel_object) || 1130 ((object == kmem_object || object == kernel_object) && 1131 !(protoeflags & MAP_ENTRY_NEEDS_COPY)), 1132 ("kmem or kernel object and cow")); 1133 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) 1134 goto charged; 1135 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1136 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1137 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1138 return (KERN_RESOURCE_SHORTAGE); 1139 KASSERT(object == NULL || (cow & MAP_ENTRY_NEEDS_COPY) || 1140 object->uip == NULL, 1141 ("OVERCOMMIT: vm_map_insert o %p", object)); 1142 uip = curthread->td_ucred->cr_ruidinfo; 1143 uihold(uip); 1144 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY)) 1145 charge_prev_obj = TRUE; 1146 } 1147 1148 charged: 1149 if (object != NULL) { 1150 /* 1151 * OBJ_ONEMAPPING must be cleared unless this mapping 1152 * is trivially proven to be the only mapping for any 1153 * of the object's pages. (Object granularity 1154 * reference counting is insufficient to recognize 1155 * aliases with precision.) 1156 */ 1157 VM_OBJECT_LOCK(object); 1158 if (object->ref_count > 1 || object->shadow_count != 0) 1159 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1160 VM_OBJECT_UNLOCK(object); 1161 } 1162 else if ((prev_entry != &map->header) && 1163 (prev_entry->eflags == protoeflags) && 1164 (prev_entry->end == start) && 1165 (prev_entry->wired_count == 0) && 1166 (prev_entry->uip == uip || 1167 (prev_entry->object.vm_object != NULL && 1168 (prev_entry->object.vm_object->uip == uip))) && 1169 vm_object_coalesce(prev_entry->object.vm_object, 1170 prev_entry->offset, 1171 (vm_size_t)(prev_entry->end - prev_entry->start), 1172 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) { 1173 /* 1174 * We were able to extend the object. Determine if we 1175 * can extend the previous map entry to include the 1176 * new range as well. 1177 */ 1178 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && 1179 (prev_entry->protection == prot) && 1180 (prev_entry->max_protection == max)) { 1181 map->size += (end - prev_entry->end); 1182 prev_entry->end = end; 1183 vm_map_entry_resize_free(map, prev_entry); 1184 vm_map_simplify_entry(map, prev_entry); 1185 if (uip != NULL) 1186 uifree(uip); 1187 return (KERN_SUCCESS); 1188 } 1189 1190 /* 1191 * If we can extend the object but cannot extend the 1192 * map entry, we have to create a new map entry. We 1193 * must bump the ref count on the extended object to 1194 * account for it. object may be NULL. 1195 */ 1196 object = prev_entry->object.vm_object; 1197 offset = prev_entry->offset + 1198 (prev_entry->end - prev_entry->start); 1199 vm_object_reference(object); 1200 if (uip != NULL && object != NULL && object->uip != NULL && 1201 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1202 /* Object already accounts for this uid. */ 1203 uifree(uip); 1204 uip = NULL; 1205 } 1206 } 1207 1208 /* 1209 * NOTE: if conditionals fail, object can be NULL here. This occurs 1210 * in things like the buffer map where we manage kva but do not manage 1211 * backing objects. 1212 */ 1213 1214 /* 1215 * Create a new entry 1216 */ 1217 new_entry = vm_map_entry_create(map); 1218 new_entry->start = start; 1219 new_entry->end = end; 1220 new_entry->uip = NULL; 1221 1222 new_entry->eflags = protoeflags; 1223 new_entry->object.vm_object = object; 1224 new_entry->offset = offset; 1225 new_entry->avail_ssize = 0; 1226 1227 new_entry->inheritance = VM_INHERIT_DEFAULT; 1228 new_entry->protection = prot; 1229 new_entry->max_protection = max; 1230 new_entry->wired_count = 0; 1231 1232 KASSERT(uip == NULL || !ENTRY_CHARGED(new_entry), 1233 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); 1234 new_entry->uip = uip; 1235 1236 /* 1237 * Insert the new entry into the list 1238 */ 1239 vm_map_entry_link(map, prev_entry, new_entry); 1240 map->size += new_entry->end - new_entry->start; 1241 1242 #if 0 1243 /* 1244 * Temporarily removed to avoid MAP_STACK panic, due to 1245 * MAP_STACK being a huge hack. Will be added back in 1246 * when MAP_STACK (and the user stack mapping) is fixed. 1247 */ 1248 /* 1249 * It may be possible to simplify the entry 1250 */ 1251 vm_map_simplify_entry(map, new_entry); 1252 #endif 1253 1254 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1255 vm_map_pmap_enter(map, start, prot, 1256 object, OFF_TO_IDX(offset), end - start, 1257 cow & MAP_PREFAULT_PARTIAL); 1258 } 1259 1260 return (KERN_SUCCESS); 1261 } 1262 1263 /* 1264 * vm_map_findspace: 1265 * 1266 * Find the first fit (lowest VM address) for "length" free bytes 1267 * beginning at address >= start in the given map. 1268 * 1269 * In a vm_map_entry, "adj_free" is the amount of free space 1270 * adjacent (higher address) to this entry, and "max_free" is the 1271 * maximum amount of contiguous free space in its subtree. This 1272 * allows finding a free region in one path down the tree, so 1273 * O(log n) amortized with splay trees. 1274 * 1275 * The map must be locked, and leaves it so. 1276 * 1277 * Returns: 0 on success, and starting address in *addr, 1278 * 1 if insufficient space. 1279 */ 1280 int 1281 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1282 vm_offset_t *addr) /* OUT */ 1283 { 1284 vm_map_entry_t entry; 1285 vm_offset_t end, st; 1286 1287 /* 1288 * Request must fit within min/max VM address and must avoid 1289 * address wrap. 1290 */ 1291 if (start < map->min_offset) 1292 start = map->min_offset; 1293 if (start + length > map->max_offset || start + length < start) 1294 return (1); 1295 1296 /* Empty tree means wide open address space. */ 1297 if (map->root == NULL) { 1298 *addr = start; 1299 goto found; 1300 } 1301 1302 /* 1303 * After splay, if start comes before root node, then there 1304 * must be a gap from start to the root. 1305 */ 1306 map->root = vm_map_entry_splay(start, map->root); 1307 if (start + length <= map->root->start) { 1308 *addr = start; 1309 goto found; 1310 } 1311 1312 /* 1313 * Root is the last node that might begin its gap before 1314 * start, and this is the last comparison where address 1315 * wrap might be a problem. 1316 */ 1317 st = (start > map->root->end) ? start : map->root->end; 1318 if (length <= map->root->end + map->root->adj_free - st) { 1319 *addr = st; 1320 goto found; 1321 } 1322 1323 /* With max_free, can immediately tell if no solution. */ 1324 entry = map->root->right; 1325 if (entry == NULL || length > entry->max_free) 1326 return (1); 1327 1328 /* 1329 * Search the right subtree in the order: left subtree, root, 1330 * right subtree (first fit). The previous splay implies that 1331 * all regions in the right subtree have addresses > start. 1332 */ 1333 while (entry != NULL) { 1334 if (entry->left != NULL && entry->left->max_free >= length) 1335 entry = entry->left; 1336 else if (entry->adj_free >= length) { 1337 *addr = entry->end; 1338 goto found; 1339 } else 1340 entry = entry->right; 1341 } 1342 1343 /* Can't get here, so panic if we do. */ 1344 panic("vm_map_findspace: max_free corrupt"); 1345 1346 found: 1347 /* Expand the kernel pmap, if necessary. */ 1348 if (map == kernel_map) { 1349 end = round_page(*addr + length); 1350 if (end > kernel_vm_end) 1351 pmap_growkernel(end); 1352 } 1353 return (0); 1354 } 1355 1356 int 1357 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1358 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1359 vm_prot_t max, int cow) 1360 { 1361 vm_offset_t end; 1362 int result; 1363 1364 end = start + length; 1365 vm_map_lock(map); 1366 VM_MAP_RANGE_CHECK(map, start, end); 1367 (void) vm_map_delete(map, start, end); 1368 result = vm_map_insert(map, object, offset, start, end, prot, 1369 max, cow); 1370 vm_map_unlock(map); 1371 return (result); 1372 } 1373 1374 /* 1375 * vm_map_find finds an unallocated region in the target address 1376 * map with the given length. The search is defined to be 1377 * first-fit from the specified address; the region found is 1378 * returned in the same parameter. 1379 * 1380 * If object is non-NULL, ref count must be bumped by caller 1381 * prior to making call to account for the new entry. 1382 */ 1383 int 1384 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1385 vm_offset_t *addr, /* IN/OUT */ 1386 vm_size_t length, int find_space, vm_prot_t prot, 1387 vm_prot_t max, int cow) 1388 { 1389 vm_offset_t start; 1390 int result; 1391 1392 start = *addr; 1393 vm_map_lock(map); 1394 do { 1395 if (find_space != VMFS_NO_SPACE) { 1396 if (vm_map_findspace(map, start, length, addr)) { 1397 vm_map_unlock(map); 1398 return (KERN_NO_SPACE); 1399 } 1400 if (find_space == VMFS_ALIGNED_SPACE) 1401 pmap_align_superpage(object, offset, addr, 1402 length); 1403 start = *addr; 1404 } 1405 result = vm_map_insert(map, object, offset, start, start + 1406 length, prot, max, cow); 1407 } while (result == KERN_NO_SPACE && find_space == VMFS_ALIGNED_SPACE); 1408 vm_map_unlock(map); 1409 return (result); 1410 } 1411 1412 /* 1413 * vm_map_simplify_entry: 1414 * 1415 * Simplify the given map entry by merging with either neighbor. This 1416 * routine also has the ability to merge with both neighbors. 1417 * 1418 * The map must be locked. 1419 * 1420 * This routine guarentees that the passed entry remains valid (though 1421 * possibly extended). When merging, this routine may delete one or 1422 * both neighbors. 1423 */ 1424 void 1425 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1426 { 1427 vm_map_entry_t next, prev; 1428 vm_size_t prevsize, esize; 1429 1430 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1431 return; 1432 1433 prev = entry->prev; 1434 if (prev != &map->header) { 1435 prevsize = prev->end - prev->start; 1436 if ( (prev->end == entry->start) && 1437 (prev->object.vm_object == entry->object.vm_object) && 1438 (!prev->object.vm_object || 1439 (prev->offset + prevsize == entry->offset)) && 1440 (prev->eflags == entry->eflags) && 1441 (prev->protection == entry->protection) && 1442 (prev->max_protection == entry->max_protection) && 1443 (prev->inheritance == entry->inheritance) && 1444 (prev->wired_count == entry->wired_count) && 1445 (prev->uip == entry->uip)) { 1446 vm_map_entry_unlink(map, prev); 1447 entry->start = prev->start; 1448 entry->offset = prev->offset; 1449 if (entry->prev != &map->header) 1450 vm_map_entry_resize_free(map, entry->prev); 1451 1452 /* 1453 * If the backing object is a vnode object, 1454 * vm_object_deallocate() calls vrele(). 1455 * However, vrele() does not lock the vnode 1456 * because the vnode has additional 1457 * references. Thus, the map lock can be kept 1458 * without causing a lock-order reversal with 1459 * the vnode lock. 1460 */ 1461 if (prev->object.vm_object) 1462 vm_object_deallocate(prev->object.vm_object); 1463 if (prev->uip != NULL) 1464 uifree(prev->uip); 1465 vm_map_entry_dispose(map, prev); 1466 } 1467 } 1468 1469 next = entry->next; 1470 if (next != &map->header) { 1471 esize = entry->end - entry->start; 1472 if ((entry->end == next->start) && 1473 (next->object.vm_object == entry->object.vm_object) && 1474 (!entry->object.vm_object || 1475 (entry->offset + esize == next->offset)) && 1476 (next->eflags == entry->eflags) && 1477 (next->protection == entry->protection) && 1478 (next->max_protection == entry->max_protection) && 1479 (next->inheritance == entry->inheritance) && 1480 (next->wired_count == entry->wired_count) && 1481 (next->uip == entry->uip)) { 1482 vm_map_entry_unlink(map, next); 1483 entry->end = next->end; 1484 vm_map_entry_resize_free(map, entry); 1485 1486 /* 1487 * See comment above. 1488 */ 1489 if (next->object.vm_object) 1490 vm_object_deallocate(next->object.vm_object); 1491 if (next->uip != NULL) 1492 uifree(next->uip); 1493 vm_map_entry_dispose(map, next); 1494 } 1495 } 1496 } 1497 /* 1498 * vm_map_clip_start: [ internal use only ] 1499 * 1500 * Asserts that the given entry begins at or after 1501 * the specified address; if necessary, 1502 * it splits the entry into two. 1503 */ 1504 #define vm_map_clip_start(map, entry, startaddr) \ 1505 { \ 1506 if (startaddr > entry->start) \ 1507 _vm_map_clip_start(map, entry, startaddr); \ 1508 } 1509 1510 /* 1511 * This routine is called only when it is known that 1512 * the entry must be split. 1513 */ 1514 static void 1515 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1516 { 1517 vm_map_entry_t new_entry; 1518 1519 VM_MAP_ASSERT_LOCKED(map); 1520 1521 /* 1522 * Split off the front portion -- note that we must insert the new 1523 * entry BEFORE this one, so that this entry has the specified 1524 * starting address. 1525 */ 1526 vm_map_simplify_entry(map, entry); 1527 1528 /* 1529 * If there is no object backing this entry, we might as well create 1530 * one now. If we defer it, an object can get created after the map 1531 * is clipped, and individual objects will be created for the split-up 1532 * map. This is a bit of a hack, but is also about the best place to 1533 * put this improvement. 1534 */ 1535 if (entry->object.vm_object == NULL && !map->system_map) { 1536 vm_object_t object; 1537 object = vm_object_allocate(OBJT_DEFAULT, 1538 atop(entry->end - entry->start)); 1539 entry->object.vm_object = object; 1540 entry->offset = 0; 1541 if (entry->uip != NULL) { 1542 object->uip = entry->uip; 1543 object->charge = entry->end - entry->start; 1544 entry->uip = NULL; 1545 } 1546 } else if (entry->object.vm_object != NULL && 1547 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1548 entry->uip != NULL) { 1549 VM_OBJECT_LOCK(entry->object.vm_object); 1550 KASSERT(entry->object.vm_object->uip == NULL, 1551 ("OVERCOMMIT: vm_entry_clip_start: both uip e %p", entry)); 1552 entry->object.vm_object->uip = entry->uip; 1553 entry->object.vm_object->charge = entry->end - entry->start; 1554 VM_OBJECT_UNLOCK(entry->object.vm_object); 1555 entry->uip = NULL; 1556 } 1557 1558 new_entry = vm_map_entry_create(map); 1559 *new_entry = *entry; 1560 1561 new_entry->end = start; 1562 entry->offset += (start - entry->start); 1563 entry->start = start; 1564 if (new_entry->uip != NULL) 1565 uihold(entry->uip); 1566 1567 vm_map_entry_link(map, entry->prev, new_entry); 1568 1569 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1570 vm_object_reference(new_entry->object.vm_object); 1571 } 1572 } 1573 1574 /* 1575 * vm_map_clip_end: [ internal use only ] 1576 * 1577 * Asserts that the given entry ends at or before 1578 * the specified address; if necessary, 1579 * it splits the entry into two. 1580 */ 1581 #define vm_map_clip_end(map, entry, endaddr) \ 1582 { \ 1583 if ((endaddr) < (entry->end)) \ 1584 _vm_map_clip_end((map), (entry), (endaddr)); \ 1585 } 1586 1587 /* 1588 * This routine is called only when it is known that 1589 * the entry must be split. 1590 */ 1591 static void 1592 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1593 { 1594 vm_map_entry_t new_entry; 1595 1596 VM_MAP_ASSERT_LOCKED(map); 1597 1598 /* 1599 * If there is no object backing this entry, we might as well create 1600 * one now. If we defer it, an object can get created after the map 1601 * is clipped, and individual objects will be created for the split-up 1602 * map. This is a bit of a hack, but is also about the best place to 1603 * put this improvement. 1604 */ 1605 if (entry->object.vm_object == NULL && !map->system_map) { 1606 vm_object_t object; 1607 object = vm_object_allocate(OBJT_DEFAULT, 1608 atop(entry->end - entry->start)); 1609 entry->object.vm_object = object; 1610 entry->offset = 0; 1611 if (entry->uip != NULL) { 1612 object->uip = entry->uip; 1613 object->charge = entry->end - entry->start; 1614 entry->uip = NULL; 1615 } 1616 } else if (entry->object.vm_object != NULL && 1617 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1618 entry->uip != NULL) { 1619 VM_OBJECT_LOCK(entry->object.vm_object); 1620 KASSERT(entry->object.vm_object->uip == NULL, 1621 ("OVERCOMMIT: vm_entry_clip_end: both uip e %p", entry)); 1622 entry->object.vm_object->uip = entry->uip; 1623 entry->object.vm_object->charge = entry->end - entry->start; 1624 VM_OBJECT_UNLOCK(entry->object.vm_object); 1625 entry->uip = NULL; 1626 } 1627 1628 /* 1629 * Create a new entry and insert it AFTER the specified entry 1630 */ 1631 new_entry = vm_map_entry_create(map); 1632 *new_entry = *entry; 1633 1634 new_entry->start = entry->end = end; 1635 new_entry->offset += (end - entry->start); 1636 if (new_entry->uip != NULL) 1637 uihold(entry->uip); 1638 1639 vm_map_entry_link(map, entry, new_entry); 1640 1641 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1642 vm_object_reference(new_entry->object.vm_object); 1643 } 1644 } 1645 1646 /* 1647 * vm_map_submap: [ kernel use only ] 1648 * 1649 * Mark the given range as handled by a subordinate map. 1650 * 1651 * This range must have been created with vm_map_find, 1652 * and no other operations may have been performed on this 1653 * range prior to calling vm_map_submap. 1654 * 1655 * Only a limited number of operations can be performed 1656 * within this rage after calling vm_map_submap: 1657 * vm_fault 1658 * [Don't try vm_map_copy!] 1659 * 1660 * To remove a submapping, one must first remove the 1661 * range from the superior map, and then destroy the 1662 * submap (if desired). [Better yet, don't try it.] 1663 */ 1664 int 1665 vm_map_submap( 1666 vm_map_t map, 1667 vm_offset_t start, 1668 vm_offset_t end, 1669 vm_map_t submap) 1670 { 1671 vm_map_entry_t entry; 1672 int result = KERN_INVALID_ARGUMENT; 1673 1674 vm_map_lock(map); 1675 1676 VM_MAP_RANGE_CHECK(map, start, end); 1677 1678 if (vm_map_lookup_entry(map, start, &entry)) { 1679 vm_map_clip_start(map, entry, start); 1680 } else 1681 entry = entry->next; 1682 1683 vm_map_clip_end(map, entry, end); 1684 1685 if ((entry->start == start) && (entry->end == end) && 1686 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1687 (entry->object.vm_object == NULL)) { 1688 entry->object.sub_map = submap; 1689 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1690 result = KERN_SUCCESS; 1691 } 1692 vm_map_unlock(map); 1693 1694 return (result); 1695 } 1696 1697 /* 1698 * The maximum number of pages to map 1699 */ 1700 #define MAX_INIT_PT 96 1701 1702 /* 1703 * vm_map_pmap_enter: 1704 * 1705 * Preload read-only mappings for the given object's resident pages into 1706 * the given map. This eliminates the soft faults on process startup and 1707 * immediately after an mmap(2). Because these are speculative mappings, 1708 * cached pages are not reactivated and mapped. 1709 */ 1710 void 1711 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1712 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1713 { 1714 vm_offset_t start; 1715 vm_page_t p, p_start; 1716 vm_pindex_t psize, tmpidx; 1717 boolean_t are_queues_locked; 1718 1719 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1720 return; 1721 VM_OBJECT_LOCK(object); 1722 if (object->type == OBJT_DEVICE) { 1723 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1724 goto unlock_return; 1725 } 1726 1727 psize = atop(size); 1728 1729 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT && 1730 object->resident_page_count > MAX_INIT_PT) 1731 goto unlock_return; 1732 1733 if (psize + pindex > object->size) { 1734 if (object->size < pindex) 1735 goto unlock_return; 1736 psize = object->size - pindex; 1737 } 1738 1739 are_queues_locked = FALSE; 1740 start = 0; 1741 p_start = NULL; 1742 1743 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1744 if (p->pindex < pindex) { 1745 p = vm_page_splay(pindex, object->root); 1746 if ((object->root = p)->pindex < pindex) 1747 p = TAILQ_NEXT(p, listq); 1748 } 1749 } 1750 /* 1751 * Assert: the variable p is either (1) the page with the 1752 * least pindex greater than or equal to the parameter pindex 1753 * or (2) NULL. 1754 */ 1755 for (; 1756 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1757 p = TAILQ_NEXT(p, listq)) { 1758 /* 1759 * don't allow an madvise to blow away our really 1760 * free pages allocating pv entries. 1761 */ 1762 if ((flags & MAP_PREFAULT_MADVISE) && 1763 cnt.v_free_count < cnt.v_free_reserved) { 1764 psize = tmpidx; 1765 break; 1766 } 1767 if (p->valid == VM_PAGE_BITS_ALL) { 1768 if (p_start == NULL) { 1769 start = addr + ptoa(tmpidx); 1770 p_start = p; 1771 } 1772 } else if (p_start != NULL) { 1773 if (!are_queues_locked) { 1774 are_queues_locked = TRUE; 1775 vm_page_lock_queues(); 1776 } 1777 pmap_enter_object(map->pmap, start, addr + 1778 ptoa(tmpidx), p_start, prot); 1779 p_start = NULL; 1780 } 1781 } 1782 if (p_start != NULL) { 1783 if (!are_queues_locked) { 1784 are_queues_locked = TRUE; 1785 vm_page_lock_queues(); 1786 } 1787 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1788 p_start, prot); 1789 } 1790 if (are_queues_locked) 1791 vm_page_unlock_queues(); 1792 unlock_return: 1793 VM_OBJECT_UNLOCK(object); 1794 } 1795 1796 /* 1797 * vm_map_protect: 1798 * 1799 * Sets the protection of the specified address 1800 * region in the target map. If "set_max" is 1801 * specified, the maximum protection is to be set; 1802 * otherwise, only the current protection is affected. 1803 */ 1804 int 1805 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1806 vm_prot_t new_prot, boolean_t set_max) 1807 { 1808 vm_map_entry_t current; 1809 vm_map_entry_t entry; 1810 vm_object_t obj; 1811 struct uidinfo *uip; 1812 1813 vm_map_lock(map); 1814 1815 VM_MAP_RANGE_CHECK(map, start, end); 1816 1817 if (vm_map_lookup_entry(map, start, &entry)) { 1818 vm_map_clip_start(map, entry, start); 1819 } else { 1820 entry = entry->next; 1821 } 1822 1823 /* 1824 * Make a first pass to check for protection violations. 1825 */ 1826 current = entry; 1827 while ((current != &map->header) && (current->start < end)) { 1828 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1829 vm_map_unlock(map); 1830 return (KERN_INVALID_ARGUMENT); 1831 } 1832 if ((new_prot & current->max_protection) != new_prot) { 1833 vm_map_unlock(map); 1834 return (KERN_PROTECTION_FAILURE); 1835 } 1836 current = current->next; 1837 } 1838 1839 1840 /* 1841 * Do an accounting pass for private read-only mappings that 1842 * now will do cow due to allowed write (e.g. debugger sets 1843 * breakpoint on text segment) 1844 */ 1845 for (current = entry; (current != &map->header) && 1846 (current->start < end); current = current->next) { 1847 1848 vm_map_clip_end(map, current, end); 1849 1850 if (set_max || 1851 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 1852 ENTRY_CHARGED(current)) { 1853 continue; 1854 } 1855 1856 uip = curthread->td_ucred->cr_ruidinfo; 1857 obj = current->object.vm_object; 1858 1859 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 1860 if (!swap_reserve(current->end - current->start)) { 1861 vm_map_unlock(map); 1862 return (KERN_RESOURCE_SHORTAGE); 1863 } 1864 uihold(uip); 1865 current->uip = uip; 1866 continue; 1867 } 1868 1869 VM_OBJECT_LOCK(obj); 1870 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 1871 VM_OBJECT_UNLOCK(obj); 1872 continue; 1873 } 1874 1875 /* 1876 * Charge for the whole object allocation now, since 1877 * we cannot distinguish between non-charged and 1878 * charged clipped mapping of the same object later. 1879 */ 1880 KASSERT(obj->charge == 0, 1881 ("vm_map_protect: object %p overcharged\n", obj)); 1882 if (!swap_reserve(ptoa(obj->size))) { 1883 VM_OBJECT_UNLOCK(obj); 1884 vm_map_unlock(map); 1885 return (KERN_RESOURCE_SHORTAGE); 1886 } 1887 1888 uihold(uip); 1889 obj->uip = uip; 1890 obj->charge = ptoa(obj->size); 1891 VM_OBJECT_UNLOCK(obj); 1892 } 1893 1894 /* 1895 * Go back and fix up protections. [Note that clipping is not 1896 * necessary the second time.] 1897 */ 1898 current = entry; 1899 while ((current != &map->header) && (current->start < end)) { 1900 vm_prot_t old_prot; 1901 1902 old_prot = current->protection; 1903 if (set_max) 1904 current->protection = 1905 (current->max_protection = new_prot) & 1906 old_prot; 1907 else 1908 current->protection = new_prot; 1909 1910 /* 1911 * Update physical map if necessary. Worry about copy-on-write 1912 * here. 1913 */ 1914 if (current->protection != old_prot) { 1915 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1916 VM_PROT_ALL) 1917 pmap_protect(map->pmap, current->start, 1918 current->end, 1919 current->protection & MASK(current)); 1920 #undef MASK 1921 } 1922 vm_map_simplify_entry(map, current); 1923 current = current->next; 1924 } 1925 vm_map_unlock(map); 1926 return (KERN_SUCCESS); 1927 } 1928 1929 /* 1930 * vm_map_madvise: 1931 * 1932 * This routine traverses a processes map handling the madvise 1933 * system call. Advisories are classified as either those effecting 1934 * the vm_map_entry structure, or those effecting the underlying 1935 * objects. 1936 */ 1937 int 1938 vm_map_madvise( 1939 vm_map_t map, 1940 vm_offset_t start, 1941 vm_offset_t end, 1942 int behav) 1943 { 1944 vm_map_entry_t current, entry; 1945 int modify_map = 0; 1946 1947 /* 1948 * Some madvise calls directly modify the vm_map_entry, in which case 1949 * we need to use an exclusive lock on the map and we need to perform 1950 * various clipping operations. Otherwise we only need a read-lock 1951 * on the map. 1952 */ 1953 switch(behav) { 1954 case MADV_NORMAL: 1955 case MADV_SEQUENTIAL: 1956 case MADV_RANDOM: 1957 case MADV_NOSYNC: 1958 case MADV_AUTOSYNC: 1959 case MADV_NOCORE: 1960 case MADV_CORE: 1961 modify_map = 1; 1962 vm_map_lock(map); 1963 break; 1964 case MADV_WILLNEED: 1965 case MADV_DONTNEED: 1966 case MADV_FREE: 1967 vm_map_lock_read(map); 1968 break; 1969 default: 1970 return (KERN_INVALID_ARGUMENT); 1971 } 1972 1973 /* 1974 * Locate starting entry and clip if necessary. 1975 */ 1976 VM_MAP_RANGE_CHECK(map, start, end); 1977 1978 if (vm_map_lookup_entry(map, start, &entry)) { 1979 if (modify_map) 1980 vm_map_clip_start(map, entry, start); 1981 } else { 1982 entry = entry->next; 1983 } 1984 1985 if (modify_map) { 1986 /* 1987 * madvise behaviors that are implemented in the vm_map_entry. 1988 * 1989 * We clip the vm_map_entry so that behavioral changes are 1990 * limited to the specified address range. 1991 */ 1992 for (current = entry; 1993 (current != &map->header) && (current->start < end); 1994 current = current->next 1995 ) { 1996 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1997 continue; 1998 1999 vm_map_clip_end(map, current, end); 2000 2001 switch (behav) { 2002 case MADV_NORMAL: 2003 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2004 break; 2005 case MADV_SEQUENTIAL: 2006 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2007 break; 2008 case MADV_RANDOM: 2009 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2010 break; 2011 case MADV_NOSYNC: 2012 current->eflags |= MAP_ENTRY_NOSYNC; 2013 break; 2014 case MADV_AUTOSYNC: 2015 current->eflags &= ~MAP_ENTRY_NOSYNC; 2016 break; 2017 case MADV_NOCORE: 2018 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2019 break; 2020 case MADV_CORE: 2021 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2022 break; 2023 default: 2024 break; 2025 } 2026 vm_map_simplify_entry(map, current); 2027 } 2028 vm_map_unlock(map); 2029 } else { 2030 vm_pindex_t pindex; 2031 int count; 2032 2033 /* 2034 * madvise behaviors that are implemented in the underlying 2035 * vm_object. 2036 * 2037 * Since we don't clip the vm_map_entry, we have to clip 2038 * the vm_object pindex and count. 2039 */ 2040 for (current = entry; 2041 (current != &map->header) && (current->start < end); 2042 current = current->next 2043 ) { 2044 vm_offset_t useStart; 2045 2046 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2047 continue; 2048 2049 pindex = OFF_TO_IDX(current->offset); 2050 count = atop(current->end - current->start); 2051 useStart = current->start; 2052 2053 if (current->start < start) { 2054 pindex += atop(start - current->start); 2055 count -= atop(start - current->start); 2056 useStart = start; 2057 } 2058 if (current->end > end) 2059 count -= atop(current->end - end); 2060 2061 if (count <= 0) 2062 continue; 2063 2064 vm_object_madvise(current->object.vm_object, 2065 pindex, count, behav); 2066 if (behav == MADV_WILLNEED) { 2067 vm_map_pmap_enter(map, 2068 useStart, 2069 current->protection, 2070 current->object.vm_object, 2071 pindex, 2072 (count << PAGE_SHIFT), 2073 MAP_PREFAULT_MADVISE 2074 ); 2075 } 2076 } 2077 vm_map_unlock_read(map); 2078 } 2079 return (0); 2080 } 2081 2082 2083 /* 2084 * vm_map_inherit: 2085 * 2086 * Sets the inheritance of the specified address 2087 * range in the target map. Inheritance 2088 * affects how the map will be shared with 2089 * child maps at the time of vmspace_fork. 2090 */ 2091 int 2092 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2093 vm_inherit_t new_inheritance) 2094 { 2095 vm_map_entry_t entry; 2096 vm_map_entry_t temp_entry; 2097 2098 switch (new_inheritance) { 2099 case VM_INHERIT_NONE: 2100 case VM_INHERIT_COPY: 2101 case VM_INHERIT_SHARE: 2102 break; 2103 default: 2104 return (KERN_INVALID_ARGUMENT); 2105 } 2106 vm_map_lock(map); 2107 VM_MAP_RANGE_CHECK(map, start, end); 2108 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2109 entry = temp_entry; 2110 vm_map_clip_start(map, entry, start); 2111 } else 2112 entry = temp_entry->next; 2113 while ((entry != &map->header) && (entry->start < end)) { 2114 vm_map_clip_end(map, entry, end); 2115 entry->inheritance = new_inheritance; 2116 vm_map_simplify_entry(map, entry); 2117 entry = entry->next; 2118 } 2119 vm_map_unlock(map); 2120 return (KERN_SUCCESS); 2121 } 2122 2123 /* 2124 * vm_map_unwire: 2125 * 2126 * Implements both kernel and user unwiring. 2127 */ 2128 int 2129 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2130 int flags) 2131 { 2132 vm_map_entry_t entry, first_entry, tmp_entry; 2133 vm_offset_t saved_start; 2134 unsigned int last_timestamp; 2135 int rv; 2136 boolean_t need_wakeup, result, user_unwire; 2137 2138 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2139 vm_map_lock(map); 2140 VM_MAP_RANGE_CHECK(map, start, end); 2141 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2142 if (flags & VM_MAP_WIRE_HOLESOK) 2143 first_entry = first_entry->next; 2144 else { 2145 vm_map_unlock(map); 2146 return (KERN_INVALID_ADDRESS); 2147 } 2148 } 2149 last_timestamp = map->timestamp; 2150 entry = first_entry; 2151 while (entry != &map->header && entry->start < end) { 2152 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2153 /* 2154 * We have not yet clipped the entry. 2155 */ 2156 saved_start = (start >= entry->start) ? start : 2157 entry->start; 2158 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2159 if (vm_map_unlock_and_wait(map, 0)) { 2160 /* 2161 * Allow interruption of user unwiring? 2162 */ 2163 } 2164 vm_map_lock(map); 2165 if (last_timestamp+1 != map->timestamp) { 2166 /* 2167 * Look again for the entry because the map was 2168 * modified while it was unlocked. 2169 * Specifically, the entry may have been 2170 * clipped, merged, or deleted. 2171 */ 2172 if (!vm_map_lookup_entry(map, saved_start, 2173 &tmp_entry)) { 2174 if (flags & VM_MAP_WIRE_HOLESOK) 2175 tmp_entry = tmp_entry->next; 2176 else { 2177 if (saved_start == start) { 2178 /* 2179 * First_entry has been deleted. 2180 */ 2181 vm_map_unlock(map); 2182 return (KERN_INVALID_ADDRESS); 2183 } 2184 end = saved_start; 2185 rv = KERN_INVALID_ADDRESS; 2186 goto done; 2187 } 2188 } 2189 if (entry == first_entry) 2190 first_entry = tmp_entry; 2191 else 2192 first_entry = NULL; 2193 entry = tmp_entry; 2194 } 2195 last_timestamp = map->timestamp; 2196 continue; 2197 } 2198 vm_map_clip_start(map, entry, start); 2199 vm_map_clip_end(map, entry, end); 2200 /* 2201 * Mark the entry in case the map lock is released. (See 2202 * above.) 2203 */ 2204 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2205 /* 2206 * Check the map for holes in the specified region. 2207 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2208 */ 2209 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2210 (entry->end < end && (entry->next == &map->header || 2211 entry->next->start > entry->end))) { 2212 end = entry->end; 2213 rv = KERN_INVALID_ADDRESS; 2214 goto done; 2215 } 2216 /* 2217 * If system unwiring, require that the entry is system wired. 2218 */ 2219 if (!user_unwire && 2220 vm_map_entry_system_wired_count(entry) == 0) { 2221 end = entry->end; 2222 rv = KERN_INVALID_ARGUMENT; 2223 goto done; 2224 } 2225 entry = entry->next; 2226 } 2227 rv = KERN_SUCCESS; 2228 done: 2229 need_wakeup = FALSE; 2230 if (first_entry == NULL) { 2231 result = vm_map_lookup_entry(map, start, &first_entry); 2232 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2233 first_entry = first_entry->next; 2234 else 2235 KASSERT(result, ("vm_map_unwire: lookup failed")); 2236 } 2237 entry = first_entry; 2238 while (entry != &map->header && entry->start < end) { 2239 if (rv == KERN_SUCCESS && (!user_unwire || 2240 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2241 if (user_unwire) 2242 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2243 entry->wired_count--; 2244 if (entry->wired_count == 0) { 2245 /* 2246 * Retain the map lock. 2247 */ 2248 vm_fault_unwire(map, entry->start, entry->end, 2249 entry->object.vm_object != NULL && 2250 entry->object.vm_object->type == OBJT_DEVICE); 2251 } 2252 } 2253 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2254 ("vm_map_unwire: in-transition flag missing")); 2255 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2256 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2257 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2258 need_wakeup = TRUE; 2259 } 2260 vm_map_simplify_entry(map, entry); 2261 entry = entry->next; 2262 } 2263 vm_map_unlock(map); 2264 if (need_wakeup) 2265 vm_map_wakeup(map); 2266 return (rv); 2267 } 2268 2269 /* 2270 * vm_map_wire: 2271 * 2272 * Implements both kernel and user wiring. 2273 */ 2274 int 2275 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2276 int flags) 2277 { 2278 vm_map_entry_t entry, first_entry, tmp_entry; 2279 vm_offset_t saved_end, saved_start; 2280 unsigned int last_timestamp; 2281 int rv; 2282 boolean_t fictitious, need_wakeup, result, user_wire; 2283 2284 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2285 vm_map_lock(map); 2286 VM_MAP_RANGE_CHECK(map, start, end); 2287 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2288 if (flags & VM_MAP_WIRE_HOLESOK) 2289 first_entry = first_entry->next; 2290 else { 2291 vm_map_unlock(map); 2292 return (KERN_INVALID_ADDRESS); 2293 } 2294 } 2295 last_timestamp = map->timestamp; 2296 entry = first_entry; 2297 while (entry != &map->header && entry->start < end) { 2298 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2299 /* 2300 * We have not yet clipped the entry. 2301 */ 2302 saved_start = (start >= entry->start) ? start : 2303 entry->start; 2304 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2305 if (vm_map_unlock_and_wait(map, 0)) { 2306 /* 2307 * Allow interruption of user wiring? 2308 */ 2309 } 2310 vm_map_lock(map); 2311 if (last_timestamp + 1 != map->timestamp) { 2312 /* 2313 * Look again for the entry because the map was 2314 * modified while it was unlocked. 2315 * Specifically, the entry may have been 2316 * clipped, merged, or deleted. 2317 */ 2318 if (!vm_map_lookup_entry(map, saved_start, 2319 &tmp_entry)) { 2320 if (flags & VM_MAP_WIRE_HOLESOK) 2321 tmp_entry = tmp_entry->next; 2322 else { 2323 if (saved_start == start) { 2324 /* 2325 * first_entry has been deleted. 2326 */ 2327 vm_map_unlock(map); 2328 return (KERN_INVALID_ADDRESS); 2329 } 2330 end = saved_start; 2331 rv = KERN_INVALID_ADDRESS; 2332 goto done; 2333 } 2334 } 2335 if (entry == first_entry) 2336 first_entry = tmp_entry; 2337 else 2338 first_entry = NULL; 2339 entry = tmp_entry; 2340 } 2341 last_timestamp = map->timestamp; 2342 continue; 2343 } 2344 vm_map_clip_start(map, entry, start); 2345 vm_map_clip_end(map, entry, end); 2346 /* 2347 * Mark the entry in case the map lock is released. (See 2348 * above.) 2349 */ 2350 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2351 /* 2352 * 2353 */ 2354 if (entry->wired_count == 0) { 2355 if ((entry->protection & (VM_PROT_READ|VM_PROT_EXECUTE)) 2356 == 0) { 2357 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2358 end = entry->end; 2359 rv = KERN_INVALID_ADDRESS; 2360 goto done; 2361 } 2362 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2363 goto next_entry; 2364 } 2365 entry->wired_count++; 2366 saved_start = entry->start; 2367 saved_end = entry->end; 2368 fictitious = entry->object.vm_object != NULL && 2369 entry->object.vm_object->type == OBJT_DEVICE; 2370 /* 2371 * Release the map lock, relying on the in-transition 2372 * mark. 2373 */ 2374 vm_map_unlock(map); 2375 rv = vm_fault_wire(map, saved_start, saved_end, 2376 user_wire, fictitious); 2377 vm_map_lock(map); 2378 if (last_timestamp + 1 != map->timestamp) { 2379 /* 2380 * Look again for the entry because the map was 2381 * modified while it was unlocked. The entry 2382 * may have been clipped, but NOT merged or 2383 * deleted. 2384 */ 2385 result = vm_map_lookup_entry(map, saved_start, 2386 &tmp_entry); 2387 KASSERT(result, ("vm_map_wire: lookup failed")); 2388 if (entry == first_entry) 2389 first_entry = tmp_entry; 2390 else 2391 first_entry = NULL; 2392 entry = tmp_entry; 2393 while (entry->end < saved_end) { 2394 if (rv != KERN_SUCCESS) { 2395 KASSERT(entry->wired_count == 1, 2396 ("vm_map_wire: bad count")); 2397 entry->wired_count = -1; 2398 } 2399 entry = entry->next; 2400 } 2401 } 2402 last_timestamp = map->timestamp; 2403 if (rv != KERN_SUCCESS) { 2404 KASSERT(entry->wired_count == 1, 2405 ("vm_map_wire: bad count")); 2406 /* 2407 * Assign an out-of-range value to represent 2408 * the failure to wire this entry. 2409 */ 2410 entry->wired_count = -1; 2411 end = entry->end; 2412 goto done; 2413 } 2414 } else if (!user_wire || 2415 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2416 entry->wired_count++; 2417 } 2418 /* 2419 * Check the map for holes in the specified region. 2420 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2421 */ 2422 next_entry: 2423 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2424 (entry->end < end && (entry->next == &map->header || 2425 entry->next->start > entry->end))) { 2426 end = entry->end; 2427 rv = KERN_INVALID_ADDRESS; 2428 goto done; 2429 } 2430 entry = entry->next; 2431 } 2432 rv = KERN_SUCCESS; 2433 done: 2434 need_wakeup = FALSE; 2435 if (first_entry == NULL) { 2436 result = vm_map_lookup_entry(map, start, &first_entry); 2437 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2438 first_entry = first_entry->next; 2439 else 2440 KASSERT(result, ("vm_map_wire: lookup failed")); 2441 } 2442 entry = first_entry; 2443 while (entry != &map->header && entry->start < end) { 2444 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2445 goto next_entry_done; 2446 if (rv == KERN_SUCCESS) { 2447 if (user_wire) 2448 entry->eflags |= MAP_ENTRY_USER_WIRED; 2449 } else if (entry->wired_count == -1) { 2450 /* 2451 * Wiring failed on this entry. Thus, unwiring is 2452 * unnecessary. 2453 */ 2454 entry->wired_count = 0; 2455 } else { 2456 if (!user_wire || 2457 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2458 entry->wired_count--; 2459 if (entry->wired_count == 0) { 2460 /* 2461 * Retain the map lock. 2462 */ 2463 vm_fault_unwire(map, entry->start, entry->end, 2464 entry->object.vm_object != NULL && 2465 entry->object.vm_object->type == OBJT_DEVICE); 2466 } 2467 } 2468 next_entry_done: 2469 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2470 ("vm_map_wire: in-transition flag missing")); 2471 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED); 2472 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2473 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2474 need_wakeup = TRUE; 2475 } 2476 vm_map_simplify_entry(map, entry); 2477 entry = entry->next; 2478 } 2479 vm_map_unlock(map); 2480 if (need_wakeup) 2481 vm_map_wakeup(map); 2482 return (rv); 2483 } 2484 2485 /* 2486 * vm_map_sync 2487 * 2488 * Push any dirty cached pages in the address range to their pager. 2489 * If syncio is TRUE, dirty pages are written synchronously. 2490 * If invalidate is TRUE, any cached pages are freed as well. 2491 * 2492 * If the size of the region from start to end is zero, we are 2493 * supposed to flush all modified pages within the region containing 2494 * start. Unfortunately, a region can be split or coalesced with 2495 * neighboring regions, making it difficult to determine what the 2496 * original region was. Therefore, we approximate this requirement by 2497 * flushing the current region containing start. 2498 * 2499 * Returns an error if any part of the specified range is not mapped. 2500 */ 2501 int 2502 vm_map_sync( 2503 vm_map_t map, 2504 vm_offset_t start, 2505 vm_offset_t end, 2506 boolean_t syncio, 2507 boolean_t invalidate) 2508 { 2509 vm_map_entry_t current; 2510 vm_map_entry_t entry; 2511 vm_size_t size; 2512 vm_object_t object; 2513 vm_ooffset_t offset; 2514 unsigned int last_timestamp; 2515 2516 vm_map_lock_read(map); 2517 VM_MAP_RANGE_CHECK(map, start, end); 2518 if (!vm_map_lookup_entry(map, start, &entry)) { 2519 vm_map_unlock_read(map); 2520 return (KERN_INVALID_ADDRESS); 2521 } else if (start == end) { 2522 start = entry->start; 2523 end = entry->end; 2524 } 2525 /* 2526 * Make a first pass to check for user-wired memory and holes. 2527 */ 2528 for (current = entry; current != &map->header && current->start < end; 2529 current = current->next) { 2530 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2531 vm_map_unlock_read(map); 2532 return (KERN_INVALID_ARGUMENT); 2533 } 2534 if (end > current->end && 2535 (current->next == &map->header || 2536 current->end != current->next->start)) { 2537 vm_map_unlock_read(map); 2538 return (KERN_INVALID_ADDRESS); 2539 } 2540 } 2541 2542 if (invalidate) 2543 pmap_remove(map->pmap, start, end); 2544 2545 /* 2546 * Make a second pass, cleaning/uncaching pages from the indicated 2547 * objects as we go. 2548 */ 2549 for (current = entry; current != &map->header && current->start < end;) { 2550 offset = current->offset + (start - current->start); 2551 size = (end <= current->end ? end : current->end) - start; 2552 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2553 vm_map_t smap; 2554 vm_map_entry_t tentry; 2555 vm_size_t tsize; 2556 2557 smap = current->object.sub_map; 2558 vm_map_lock_read(smap); 2559 (void) vm_map_lookup_entry(smap, offset, &tentry); 2560 tsize = tentry->end - offset; 2561 if (tsize < size) 2562 size = tsize; 2563 object = tentry->object.vm_object; 2564 offset = tentry->offset + (offset - tentry->start); 2565 vm_map_unlock_read(smap); 2566 } else { 2567 object = current->object.vm_object; 2568 } 2569 vm_object_reference(object); 2570 last_timestamp = map->timestamp; 2571 vm_map_unlock_read(map); 2572 vm_object_sync(object, offset, size, syncio, invalidate); 2573 start += size; 2574 vm_object_deallocate(object); 2575 vm_map_lock_read(map); 2576 if (last_timestamp == map->timestamp || 2577 !vm_map_lookup_entry(map, start, ¤t)) 2578 current = current->next; 2579 } 2580 2581 vm_map_unlock_read(map); 2582 return (KERN_SUCCESS); 2583 } 2584 2585 /* 2586 * vm_map_entry_unwire: [ internal use only ] 2587 * 2588 * Make the region specified by this entry pageable. 2589 * 2590 * The map in question should be locked. 2591 * [This is the reason for this routine's existence.] 2592 */ 2593 static void 2594 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2595 { 2596 vm_fault_unwire(map, entry->start, entry->end, 2597 entry->object.vm_object != NULL && 2598 entry->object.vm_object->type == OBJT_DEVICE); 2599 entry->wired_count = 0; 2600 } 2601 2602 /* 2603 * vm_map_entry_delete: [ internal use only ] 2604 * 2605 * Deallocate the given entry from the target map. 2606 */ 2607 static void 2608 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2609 { 2610 vm_object_t object; 2611 vm_pindex_t offidxstart, offidxend, count, size1; 2612 vm_ooffset_t size; 2613 2614 vm_map_entry_unlink(map, entry); 2615 object = entry->object.vm_object; 2616 size = entry->end - entry->start; 2617 map->size -= size; 2618 2619 if (entry->uip != NULL) { 2620 swap_release_by_uid(size, entry->uip); 2621 uifree(entry->uip); 2622 } 2623 2624 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2625 (object != NULL)) { 2626 KASSERT(entry->uip == NULL || object->uip == NULL || 2627 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2628 ("OVERCOMMIT vm_map_entry_delete: both uip %p", entry)); 2629 count = OFF_TO_IDX(size); 2630 offidxstart = OFF_TO_IDX(entry->offset); 2631 offidxend = offidxstart + count; 2632 VM_OBJECT_LOCK(object); 2633 if (object->ref_count != 1 && 2634 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2635 object == kernel_object || object == kmem_object)) { 2636 vm_object_collapse(object); 2637 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2638 if (object->type == OBJT_SWAP) 2639 swap_pager_freespace(object, offidxstart, count); 2640 if (offidxend >= object->size && 2641 offidxstart < object->size) { 2642 size1 = object->size; 2643 object->size = offidxstart; 2644 if (object->uip != NULL) { 2645 size1 -= object->size; 2646 KASSERT(object->charge >= ptoa(size1), 2647 ("vm_map_entry_delete: object->charge < 0")); 2648 swap_release_by_uid(ptoa(size1), object->uip); 2649 object->charge -= ptoa(size1); 2650 } 2651 } 2652 } 2653 VM_OBJECT_UNLOCK(object); 2654 } else 2655 entry->object.vm_object = NULL; 2656 } 2657 2658 /* 2659 * vm_map_delete: [ internal use only ] 2660 * 2661 * Deallocates the given address range from the target 2662 * map. 2663 */ 2664 int 2665 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2666 { 2667 vm_map_entry_t entry; 2668 vm_map_entry_t first_entry; 2669 2670 VM_MAP_ASSERT_LOCKED(map); 2671 2672 /* 2673 * Find the start of the region, and clip it 2674 */ 2675 if (!vm_map_lookup_entry(map, start, &first_entry)) 2676 entry = first_entry->next; 2677 else { 2678 entry = first_entry; 2679 vm_map_clip_start(map, entry, start); 2680 } 2681 2682 /* 2683 * Step through all entries in this region 2684 */ 2685 while ((entry != &map->header) && (entry->start < end)) { 2686 vm_map_entry_t next; 2687 2688 /* 2689 * Wait for wiring or unwiring of an entry to complete. 2690 * Also wait for any system wirings to disappear on 2691 * user maps. 2692 */ 2693 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2694 (vm_map_pmap(map) != kernel_pmap && 2695 vm_map_entry_system_wired_count(entry) != 0)) { 2696 unsigned int last_timestamp; 2697 vm_offset_t saved_start; 2698 vm_map_entry_t tmp_entry; 2699 2700 saved_start = entry->start; 2701 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2702 last_timestamp = map->timestamp; 2703 (void) vm_map_unlock_and_wait(map, 0); 2704 vm_map_lock(map); 2705 if (last_timestamp + 1 != map->timestamp) { 2706 /* 2707 * Look again for the entry because the map was 2708 * modified while it was unlocked. 2709 * Specifically, the entry may have been 2710 * clipped, merged, or deleted. 2711 */ 2712 if (!vm_map_lookup_entry(map, saved_start, 2713 &tmp_entry)) 2714 entry = tmp_entry->next; 2715 else { 2716 entry = tmp_entry; 2717 vm_map_clip_start(map, entry, 2718 saved_start); 2719 } 2720 } 2721 continue; 2722 } 2723 vm_map_clip_end(map, entry, end); 2724 2725 next = entry->next; 2726 2727 /* 2728 * Unwire before removing addresses from the pmap; otherwise, 2729 * unwiring will put the entries back in the pmap. 2730 */ 2731 if (entry->wired_count != 0) { 2732 vm_map_entry_unwire(map, entry); 2733 } 2734 2735 pmap_remove(map->pmap, entry->start, entry->end); 2736 2737 /* 2738 * Delete the entry only after removing all pmap 2739 * entries pointing to its pages. (Otherwise, its 2740 * page frames may be reallocated, and any modify bits 2741 * will be set in the wrong object!) 2742 */ 2743 vm_map_entry_delete(map, entry); 2744 entry->next = map->deferred_freelist; 2745 map->deferred_freelist = entry; 2746 entry = next; 2747 } 2748 return (KERN_SUCCESS); 2749 } 2750 2751 /* 2752 * vm_map_remove: 2753 * 2754 * Remove the given address range from the target map. 2755 * This is the exported form of vm_map_delete. 2756 */ 2757 int 2758 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2759 { 2760 int result; 2761 2762 vm_map_lock(map); 2763 VM_MAP_RANGE_CHECK(map, start, end); 2764 result = vm_map_delete(map, start, end); 2765 vm_map_unlock(map); 2766 return (result); 2767 } 2768 2769 /* 2770 * vm_map_check_protection: 2771 * 2772 * Assert that the target map allows the specified privilege on the 2773 * entire address region given. The entire region must be allocated. 2774 * 2775 * WARNING! This code does not and should not check whether the 2776 * contents of the region is accessible. For example a smaller file 2777 * might be mapped into a larger address space. 2778 * 2779 * NOTE! This code is also called by munmap(). 2780 * 2781 * The map must be locked. A read lock is sufficient. 2782 */ 2783 boolean_t 2784 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2785 vm_prot_t protection) 2786 { 2787 vm_map_entry_t entry; 2788 vm_map_entry_t tmp_entry; 2789 2790 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2791 return (FALSE); 2792 entry = tmp_entry; 2793 2794 while (start < end) { 2795 if (entry == &map->header) 2796 return (FALSE); 2797 /* 2798 * No holes allowed! 2799 */ 2800 if (start < entry->start) 2801 return (FALSE); 2802 /* 2803 * Check protection associated with entry. 2804 */ 2805 if ((entry->protection & protection) != protection) 2806 return (FALSE); 2807 /* go to next entry */ 2808 start = entry->end; 2809 entry = entry->next; 2810 } 2811 return (TRUE); 2812 } 2813 2814 /* 2815 * vm_map_copy_entry: 2816 * 2817 * Copies the contents of the source entry to the destination 2818 * entry. The entries *must* be aligned properly. 2819 */ 2820 static void 2821 vm_map_copy_entry( 2822 vm_map_t src_map, 2823 vm_map_t dst_map, 2824 vm_map_entry_t src_entry, 2825 vm_map_entry_t dst_entry, 2826 vm_ooffset_t *fork_charge) 2827 { 2828 vm_object_t src_object; 2829 vm_offset_t size; 2830 struct uidinfo *uip; 2831 int charged; 2832 2833 VM_MAP_ASSERT_LOCKED(dst_map); 2834 2835 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2836 return; 2837 2838 if (src_entry->wired_count == 0) { 2839 2840 /* 2841 * If the source entry is marked needs_copy, it is already 2842 * write-protected. 2843 */ 2844 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2845 pmap_protect(src_map->pmap, 2846 src_entry->start, 2847 src_entry->end, 2848 src_entry->protection & ~VM_PROT_WRITE); 2849 } 2850 2851 /* 2852 * Make a copy of the object. 2853 */ 2854 size = src_entry->end - src_entry->start; 2855 if ((src_object = src_entry->object.vm_object) != NULL) { 2856 VM_OBJECT_LOCK(src_object); 2857 charged = ENTRY_CHARGED(src_entry); 2858 if ((src_object->handle == NULL) && 2859 (src_object->type == OBJT_DEFAULT || 2860 src_object->type == OBJT_SWAP)) { 2861 vm_object_collapse(src_object); 2862 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2863 vm_object_split(src_entry); 2864 src_object = src_entry->object.vm_object; 2865 } 2866 } 2867 vm_object_reference_locked(src_object); 2868 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2869 if (src_entry->uip != NULL && 2870 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 2871 KASSERT(src_object->uip == NULL, 2872 ("OVERCOMMIT: vm_map_copy_entry: uip %p", 2873 src_object)); 2874 src_object->uip = src_entry->uip; 2875 src_object->charge = size; 2876 } 2877 VM_OBJECT_UNLOCK(src_object); 2878 dst_entry->object.vm_object = src_object; 2879 if (charged) { 2880 uip = curthread->td_ucred->cr_ruidinfo; 2881 uihold(uip); 2882 dst_entry->uip = uip; 2883 *fork_charge += size; 2884 if (!(src_entry->eflags & 2885 MAP_ENTRY_NEEDS_COPY)) { 2886 uihold(uip); 2887 src_entry->uip = uip; 2888 *fork_charge += size; 2889 } 2890 } 2891 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2892 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2893 dst_entry->offset = src_entry->offset; 2894 } else { 2895 dst_entry->object.vm_object = NULL; 2896 dst_entry->offset = 0; 2897 if (src_entry->uip != NULL) { 2898 dst_entry->uip = curthread->td_ucred->cr_ruidinfo; 2899 uihold(dst_entry->uip); 2900 *fork_charge += size; 2901 } 2902 } 2903 2904 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2905 dst_entry->end - dst_entry->start, src_entry->start); 2906 } else { 2907 /* 2908 * Of course, wired down pages can't be set copy-on-write. 2909 * Cause wired pages to be copied into the new map by 2910 * simulating faults (the new pages are pageable) 2911 */ 2912 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2913 } 2914 } 2915 2916 /* 2917 * vmspace_map_entry_forked: 2918 * Update the newly-forked vmspace each time a map entry is inherited 2919 * or copied. The values for vm_dsize and vm_tsize are approximate 2920 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2921 */ 2922 static void 2923 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2924 vm_map_entry_t entry) 2925 { 2926 vm_size_t entrysize; 2927 vm_offset_t newend; 2928 2929 entrysize = entry->end - entry->start; 2930 vm2->vm_map.size += entrysize; 2931 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2932 vm2->vm_ssize += btoc(entrysize); 2933 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2934 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2935 newend = MIN(entry->end, 2936 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2937 vm2->vm_dsize += btoc(newend - entry->start); 2938 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2939 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2940 newend = MIN(entry->end, 2941 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2942 vm2->vm_tsize += btoc(newend - entry->start); 2943 } 2944 } 2945 2946 /* 2947 * vmspace_fork: 2948 * Create a new process vmspace structure and vm_map 2949 * based on those of an existing process. The new map 2950 * is based on the old map, according to the inheritance 2951 * values on the regions in that map. 2952 * 2953 * XXX It might be worth coalescing the entries added to the new vmspace. 2954 * 2955 * The source map must not be locked. 2956 */ 2957 struct vmspace * 2958 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 2959 { 2960 struct vmspace *vm2; 2961 vm_map_t old_map = &vm1->vm_map; 2962 vm_map_t new_map; 2963 vm_map_entry_t old_entry; 2964 vm_map_entry_t new_entry; 2965 vm_object_t object; 2966 int locked; 2967 2968 vm_map_lock(old_map); 2969 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2970 if (vm2 == NULL) 2971 goto unlock_and_return; 2972 vm2->vm_taddr = vm1->vm_taddr; 2973 vm2->vm_daddr = vm1->vm_daddr; 2974 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2975 new_map = &vm2->vm_map; /* XXX */ 2976 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 2977 KASSERT(locked, ("vmspace_fork: lock failed")); 2978 new_map->timestamp = 1; 2979 2980 old_entry = old_map->header.next; 2981 2982 while (old_entry != &old_map->header) { 2983 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2984 panic("vm_map_fork: encountered a submap"); 2985 2986 switch (old_entry->inheritance) { 2987 case VM_INHERIT_NONE: 2988 break; 2989 2990 case VM_INHERIT_SHARE: 2991 /* 2992 * Clone the entry, creating the shared object if necessary. 2993 */ 2994 object = old_entry->object.vm_object; 2995 if (object == NULL) { 2996 object = vm_object_allocate(OBJT_DEFAULT, 2997 atop(old_entry->end - old_entry->start)); 2998 old_entry->object.vm_object = object; 2999 old_entry->offset = 0; 3000 if (old_entry->uip != NULL) { 3001 object->uip = old_entry->uip; 3002 object->charge = old_entry->end - 3003 old_entry->start; 3004 old_entry->uip = NULL; 3005 } 3006 } 3007 3008 /* 3009 * Add the reference before calling vm_object_shadow 3010 * to insure that a shadow object is created. 3011 */ 3012 vm_object_reference(object); 3013 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3014 vm_object_shadow(&old_entry->object.vm_object, 3015 &old_entry->offset, 3016 atop(old_entry->end - old_entry->start)); 3017 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3018 /* Transfer the second reference too. */ 3019 vm_object_reference( 3020 old_entry->object.vm_object); 3021 3022 /* 3023 * As in vm_map_simplify_entry(), the 3024 * vnode lock will not be acquired in 3025 * this call to vm_object_deallocate(). 3026 */ 3027 vm_object_deallocate(object); 3028 object = old_entry->object.vm_object; 3029 } 3030 VM_OBJECT_LOCK(object); 3031 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3032 if (old_entry->uip != NULL) { 3033 KASSERT(object->uip == NULL, ("vmspace_fork both uip")); 3034 object->uip = old_entry->uip; 3035 object->charge = old_entry->end - old_entry->start; 3036 old_entry->uip = NULL; 3037 } 3038 VM_OBJECT_UNLOCK(object); 3039 3040 /* 3041 * Clone the entry, referencing the shared object. 3042 */ 3043 new_entry = vm_map_entry_create(new_map); 3044 *new_entry = *old_entry; 3045 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3046 MAP_ENTRY_IN_TRANSITION); 3047 new_entry->wired_count = 0; 3048 3049 /* 3050 * Insert the entry into the new map -- we know we're 3051 * inserting at the end of the new map. 3052 */ 3053 vm_map_entry_link(new_map, new_map->header.prev, 3054 new_entry); 3055 vmspace_map_entry_forked(vm1, vm2, new_entry); 3056 3057 /* 3058 * Update the physical map 3059 */ 3060 pmap_copy(new_map->pmap, old_map->pmap, 3061 new_entry->start, 3062 (old_entry->end - old_entry->start), 3063 old_entry->start); 3064 break; 3065 3066 case VM_INHERIT_COPY: 3067 /* 3068 * Clone the entry and link into the map. 3069 */ 3070 new_entry = vm_map_entry_create(new_map); 3071 *new_entry = *old_entry; 3072 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3073 MAP_ENTRY_IN_TRANSITION); 3074 new_entry->wired_count = 0; 3075 new_entry->object.vm_object = NULL; 3076 vm_map_entry_link(new_map, new_map->header.prev, 3077 new_entry); 3078 vmspace_map_entry_forked(vm1, vm2, new_entry); 3079 vm_map_copy_entry(old_map, new_map, old_entry, 3080 new_entry, fork_charge); 3081 break; 3082 } 3083 old_entry = old_entry->next; 3084 } 3085 unlock_and_return: 3086 vm_map_unlock(old_map); 3087 if (vm2 != NULL) 3088 vm_map_unlock(new_map); 3089 3090 return (vm2); 3091 } 3092 3093 int 3094 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3095 vm_prot_t prot, vm_prot_t max, int cow) 3096 { 3097 vm_map_entry_t new_entry, prev_entry; 3098 vm_offset_t bot, top; 3099 vm_size_t init_ssize; 3100 int orient, rv; 3101 rlim_t vmemlim; 3102 3103 /* 3104 * The stack orientation is piggybacked with the cow argument. 3105 * Extract it into orient and mask the cow argument so that we 3106 * don't pass it around further. 3107 * NOTE: We explicitly allow bi-directional stacks. 3108 */ 3109 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 3110 cow &= ~orient; 3111 KASSERT(orient != 0, ("No stack grow direction")); 3112 3113 if (addrbos < vm_map_min(map) || 3114 addrbos > vm_map_max(map) || 3115 addrbos + max_ssize < addrbos) 3116 return (KERN_NO_SPACE); 3117 3118 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 3119 3120 PROC_LOCK(curthread->td_proc); 3121 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 3122 PROC_UNLOCK(curthread->td_proc); 3123 3124 vm_map_lock(map); 3125 3126 /* If addr is already mapped, no go */ 3127 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 3128 vm_map_unlock(map); 3129 return (KERN_NO_SPACE); 3130 } 3131 3132 /* If we would blow our VMEM resource limit, no go */ 3133 if (map->size + init_ssize > vmemlim) { 3134 vm_map_unlock(map); 3135 return (KERN_NO_SPACE); 3136 } 3137 3138 /* 3139 * If we can't accomodate max_ssize in the current mapping, no go. 3140 * However, we need to be aware that subsequent user mappings might 3141 * map into the space we have reserved for stack, and currently this 3142 * space is not protected. 3143 * 3144 * Hopefully we will at least detect this condition when we try to 3145 * grow the stack. 3146 */ 3147 if ((prev_entry->next != &map->header) && 3148 (prev_entry->next->start < addrbos + max_ssize)) { 3149 vm_map_unlock(map); 3150 return (KERN_NO_SPACE); 3151 } 3152 3153 /* 3154 * We initially map a stack of only init_ssize. We will grow as 3155 * needed later. Depending on the orientation of the stack (i.e. 3156 * the grow direction) we either map at the top of the range, the 3157 * bottom of the range or in the middle. 3158 * 3159 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3160 * and cow to be 0. Possibly we should eliminate these as input 3161 * parameters, and just pass these values here in the insert call. 3162 */ 3163 if (orient == MAP_STACK_GROWS_DOWN) 3164 bot = addrbos + max_ssize - init_ssize; 3165 else if (orient == MAP_STACK_GROWS_UP) 3166 bot = addrbos; 3167 else 3168 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 3169 top = bot + init_ssize; 3170 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3171 3172 /* Now set the avail_ssize amount. */ 3173 if (rv == KERN_SUCCESS) { 3174 if (prev_entry != &map->header) 3175 vm_map_clip_end(map, prev_entry, bot); 3176 new_entry = prev_entry->next; 3177 if (new_entry->end != top || new_entry->start != bot) 3178 panic("Bad entry start/end for new stack entry"); 3179 3180 new_entry->avail_ssize = max_ssize - init_ssize; 3181 if (orient & MAP_STACK_GROWS_DOWN) 3182 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3183 if (orient & MAP_STACK_GROWS_UP) 3184 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 3185 } 3186 3187 vm_map_unlock(map); 3188 return (rv); 3189 } 3190 3191 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 3192 * desired address is already mapped, or if we successfully grow 3193 * the stack. Also returns KERN_SUCCESS if addr is outside the 3194 * stack range (this is strange, but preserves compatibility with 3195 * the grow function in vm_machdep.c). 3196 */ 3197 int 3198 vm_map_growstack(struct proc *p, vm_offset_t addr) 3199 { 3200 vm_map_entry_t next_entry, prev_entry; 3201 vm_map_entry_t new_entry, stack_entry; 3202 struct vmspace *vm = p->p_vmspace; 3203 vm_map_t map = &vm->vm_map; 3204 vm_offset_t end; 3205 size_t grow_amount, max_grow; 3206 rlim_t stacklim, vmemlim; 3207 int is_procstack, rv; 3208 struct uidinfo *uip; 3209 3210 Retry: 3211 PROC_LOCK(p); 3212 stacklim = lim_cur(p, RLIMIT_STACK); 3213 vmemlim = lim_cur(p, RLIMIT_VMEM); 3214 PROC_UNLOCK(p); 3215 3216 vm_map_lock_read(map); 3217 3218 /* If addr is already in the entry range, no need to grow.*/ 3219 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 3220 vm_map_unlock_read(map); 3221 return (KERN_SUCCESS); 3222 } 3223 3224 next_entry = prev_entry->next; 3225 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 3226 /* 3227 * This entry does not grow upwards. Since the address lies 3228 * beyond this entry, the next entry (if one exists) has to 3229 * be a downward growable entry. The entry list header is 3230 * never a growable entry, so it suffices to check the flags. 3231 */ 3232 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 3233 vm_map_unlock_read(map); 3234 return (KERN_SUCCESS); 3235 } 3236 stack_entry = next_entry; 3237 } else { 3238 /* 3239 * This entry grows upward. If the next entry does not at 3240 * least grow downwards, this is the entry we need to grow. 3241 * otherwise we have two possible choices and we have to 3242 * select one. 3243 */ 3244 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 3245 /* 3246 * We have two choices; grow the entry closest to 3247 * the address to minimize the amount of growth. 3248 */ 3249 if (addr - prev_entry->end <= next_entry->start - addr) 3250 stack_entry = prev_entry; 3251 else 3252 stack_entry = next_entry; 3253 } else 3254 stack_entry = prev_entry; 3255 } 3256 3257 if (stack_entry == next_entry) { 3258 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 3259 KASSERT(addr < stack_entry->start, ("foo")); 3260 end = (prev_entry != &map->header) ? prev_entry->end : 3261 stack_entry->start - stack_entry->avail_ssize; 3262 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 3263 max_grow = stack_entry->start - end; 3264 } else { 3265 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 3266 KASSERT(addr >= stack_entry->end, ("foo")); 3267 end = (next_entry != &map->header) ? next_entry->start : 3268 stack_entry->end + stack_entry->avail_ssize; 3269 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 3270 max_grow = end - stack_entry->end; 3271 } 3272 3273 if (grow_amount > stack_entry->avail_ssize) { 3274 vm_map_unlock_read(map); 3275 return (KERN_NO_SPACE); 3276 } 3277 3278 /* 3279 * If there is no longer enough space between the entries nogo, and 3280 * adjust the available space. Note: this should only happen if the 3281 * user has mapped into the stack area after the stack was created, 3282 * and is probably an error. 3283 * 3284 * This also effectively destroys any guard page the user might have 3285 * intended by limiting the stack size. 3286 */ 3287 if (grow_amount > max_grow) { 3288 if (vm_map_lock_upgrade(map)) 3289 goto Retry; 3290 3291 stack_entry->avail_ssize = max_grow; 3292 3293 vm_map_unlock(map); 3294 return (KERN_NO_SPACE); 3295 } 3296 3297 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 3298 3299 /* 3300 * If this is the main process stack, see if we're over the stack 3301 * limit. 3302 */ 3303 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3304 vm_map_unlock_read(map); 3305 return (KERN_NO_SPACE); 3306 } 3307 3308 /* Round up the grow amount modulo SGROWSIZ */ 3309 grow_amount = roundup (grow_amount, sgrowsiz); 3310 if (grow_amount > stack_entry->avail_ssize) 3311 grow_amount = stack_entry->avail_ssize; 3312 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3313 grow_amount = stacklim - ctob(vm->vm_ssize); 3314 } 3315 3316 /* If we would blow our VMEM resource limit, no go */ 3317 if (map->size + grow_amount > vmemlim) { 3318 vm_map_unlock_read(map); 3319 return (KERN_NO_SPACE); 3320 } 3321 3322 if (vm_map_lock_upgrade(map)) 3323 goto Retry; 3324 3325 if (stack_entry == next_entry) { 3326 /* 3327 * Growing downward. 3328 */ 3329 /* Get the preliminary new entry start value */ 3330 addr = stack_entry->start - grow_amount; 3331 3332 /* 3333 * If this puts us into the previous entry, cut back our 3334 * growth to the available space. Also, see the note above. 3335 */ 3336 if (addr < end) { 3337 stack_entry->avail_ssize = max_grow; 3338 addr = end; 3339 } 3340 3341 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 3342 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 3343 3344 /* Adjust the available stack space by the amount we grew. */ 3345 if (rv == KERN_SUCCESS) { 3346 if (prev_entry != &map->header) 3347 vm_map_clip_end(map, prev_entry, addr); 3348 new_entry = prev_entry->next; 3349 KASSERT(new_entry == stack_entry->prev, ("foo")); 3350 KASSERT(new_entry->end == stack_entry->start, ("foo")); 3351 KASSERT(new_entry->start == addr, ("foo")); 3352 grow_amount = new_entry->end - new_entry->start; 3353 new_entry->avail_ssize = stack_entry->avail_ssize - 3354 grow_amount; 3355 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 3356 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 3357 } 3358 } else { 3359 /* 3360 * Growing upward. 3361 */ 3362 addr = stack_entry->end + grow_amount; 3363 3364 /* 3365 * If this puts us into the next entry, cut back our growth 3366 * to the available space. Also, see the note above. 3367 */ 3368 if (addr > end) { 3369 stack_entry->avail_ssize = end - stack_entry->end; 3370 addr = end; 3371 } 3372 3373 grow_amount = addr - stack_entry->end; 3374 uip = stack_entry->uip; 3375 if (uip == NULL && stack_entry->object.vm_object != NULL) 3376 uip = stack_entry->object.vm_object->uip; 3377 if (uip != NULL && !swap_reserve_by_uid(grow_amount, uip)) 3378 rv = KERN_NO_SPACE; 3379 /* Grow the underlying object if applicable. */ 3380 else if (stack_entry->object.vm_object == NULL || 3381 vm_object_coalesce(stack_entry->object.vm_object, 3382 stack_entry->offset, 3383 (vm_size_t)(stack_entry->end - stack_entry->start), 3384 (vm_size_t)grow_amount, uip != NULL)) { 3385 map->size += (addr - stack_entry->end); 3386 /* Update the current entry. */ 3387 stack_entry->end = addr; 3388 stack_entry->avail_ssize -= grow_amount; 3389 vm_map_entry_resize_free(map, stack_entry); 3390 rv = KERN_SUCCESS; 3391 3392 if (next_entry != &map->header) 3393 vm_map_clip_start(map, next_entry, addr); 3394 } else 3395 rv = KERN_FAILURE; 3396 } 3397 3398 if (rv == KERN_SUCCESS && is_procstack) 3399 vm->vm_ssize += btoc(grow_amount); 3400 3401 vm_map_unlock(map); 3402 3403 /* 3404 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3405 */ 3406 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3407 vm_map_wire(map, 3408 (stack_entry == next_entry) ? addr : addr - grow_amount, 3409 (stack_entry == next_entry) ? stack_entry->start : addr, 3410 (p->p_flag & P_SYSTEM) 3411 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3412 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3413 } 3414 3415 return (rv); 3416 } 3417 3418 /* 3419 * Unshare the specified VM space for exec. If other processes are 3420 * mapped to it, then create a new one. The new vmspace is null. 3421 */ 3422 int 3423 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3424 { 3425 struct vmspace *oldvmspace = p->p_vmspace; 3426 struct vmspace *newvmspace; 3427 3428 newvmspace = vmspace_alloc(minuser, maxuser); 3429 if (newvmspace == NULL) 3430 return (ENOMEM); 3431 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3432 /* 3433 * This code is written like this for prototype purposes. The 3434 * goal is to avoid running down the vmspace here, but let the 3435 * other process's that are still using the vmspace to finally 3436 * run it down. Even though there is little or no chance of blocking 3437 * here, it is a good idea to keep this form for future mods. 3438 */ 3439 PROC_VMSPACE_LOCK(p); 3440 p->p_vmspace = newvmspace; 3441 PROC_VMSPACE_UNLOCK(p); 3442 if (p == curthread->td_proc) 3443 pmap_activate(curthread); 3444 vmspace_free(oldvmspace); 3445 return (0); 3446 } 3447 3448 /* 3449 * Unshare the specified VM space for forcing COW. This 3450 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3451 */ 3452 int 3453 vmspace_unshare(struct proc *p) 3454 { 3455 struct vmspace *oldvmspace = p->p_vmspace; 3456 struct vmspace *newvmspace; 3457 vm_ooffset_t fork_charge; 3458 3459 if (oldvmspace->vm_refcnt == 1) 3460 return (0); 3461 fork_charge = 0; 3462 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3463 if (newvmspace == NULL) 3464 return (ENOMEM); 3465 if (!swap_reserve_by_uid(fork_charge, p->p_ucred->cr_ruidinfo)) { 3466 vmspace_free(newvmspace); 3467 return (ENOMEM); 3468 } 3469 PROC_VMSPACE_LOCK(p); 3470 p->p_vmspace = newvmspace; 3471 PROC_VMSPACE_UNLOCK(p); 3472 if (p == curthread->td_proc) 3473 pmap_activate(curthread); 3474 vmspace_free(oldvmspace); 3475 return (0); 3476 } 3477 3478 /* 3479 * vm_map_lookup: 3480 * 3481 * Finds the VM object, offset, and 3482 * protection for a given virtual address in the 3483 * specified map, assuming a page fault of the 3484 * type specified. 3485 * 3486 * Leaves the map in question locked for read; return 3487 * values are guaranteed until a vm_map_lookup_done 3488 * call is performed. Note that the map argument 3489 * is in/out; the returned map must be used in 3490 * the call to vm_map_lookup_done. 3491 * 3492 * A handle (out_entry) is returned for use in 3493 * vm_map_lookup_done, to make that fast. 3494 * 3495 * If a lookup is requested with "write protection" 3496 * specified, the map may be changed to perform virtual 3497 * copying operations, although the data referenced will 3498 * remain the same. 3499 */ 3500 int 3501 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3502 vm_offset_t vaddr, 3503 vm_prot_t fault_typea, 3504 vm_map_entry_t *out_entry, /* OUT */ 3505 vm_object_t *object, /* OUT */ 3506 vm_pindex_t *pindex, /* OUT */ 3507 vm_prot_t *out_prot, /* OUT */ 3508 boolean_t *wired) /* OUT */ 3509 { 3510 vm_map_entry_t entry; 3511 vm_map_t map = *var_map; 3512 vm_prot_t prot; 3513 vm_prot_t fault_type = fault_typea; 3514 vm_object_t eobject; 3515 struct uidinfo *uip; 3516 vm_ooffset_t size; 3517 3518 RetryLookup:; 3519 3520 vm_map_lock_read(map); 3521 3522 /* 3523 * Lookup the faulting address. 3524 */ 3525 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 3526 vm_map_unlock_read(map); 3527 return (KERN_INVALID_ADDRESS); 3528 } 3529 3530 entry = *out_entry; 3531 3532 /* 3533 * Handle submaps. 3534 */ 3535 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3536 vm_map_t old_map = map; 3537 3538 *var_map = map = entry->object.sub_map; 3539 vm_map_unlock_read(old_map); 3540 goto RetryLookup; 3541 } 3542 3543 /* 3544 * Check whether this task is allowed to have this page. 3545 * Note the special case for MAP_ENTRY_COW 3546 * pages with an override. This is to implement a forced 3547 * COW for debuggers. 3548 */ 3549 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3550 prot = entry->max_protection; 3551 else 3552 prot = entry->protection; 3553 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3554 if ((fault_type & prot) != fault_type) { 3555 vm_map_unlock_read(map); 3556 return (KERN_PROTECTION_FAILURE); 3557 } 3558 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3559 (entry->eflags & MAP_ENTRY_COW) && 3560 (fault_type & VM_PROT_WRITE) && 3561 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3562 vm_map_unlock_read(map); 3563 return (KERN_PROTECTION_FAILURE); 3564 } 3565 3566 /* 3567 * If this page is not pageable, we have to get it for all possible 3568 * accesses. 3569 */ 3570 *wired = (entry->wired_count != 0); 3571 if (*wired) 3572 prot = fault_type = entry->protection; 3573 size = entry->end - entry->start; 3574 /* 3575 * If the entry was copy-on-write, we either ... 3576 */ 3577 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3578 /* 3579 * If we want to write the page, we may as well handle that 3580 * now since we've got the map locked. 3581 * 3582 * If we don't need to write the page, we just demote the 3583 * permissions allowed. 3584 */ 3585 if (fault_type & VM_PROT_WRITE) { 3586 /* 3587 * Make a new object, and place it in the object 3588 * chain. Note that no new references have appeared 3589 * -- one just moved from the map to the new 3590 * object. 3591 */ 3592 if (vm_map_lock_upgrade(map)) 3593 goto RetryLookup; 3594 3595 if (entry->uip == NULL) { 3596 /* 3597 * The debugger owner is charged for 3598 * the memory. 3599 */ 3600 uip = curthread->td_ucred->cr_ruidinfo; 3601 uihold(uip); 3602 if (!swap_reserve_by_uid(size, uip)) { 3603 uifree(uip); 3604 vm_map_unlock(map); 3605 return (KERN_RESOURCE_SHORTAGE); 3606 } 3607 entry->uip = uip; 3608 } 3609 vm_object_shadow( 3610 &entry->object.vm_object, 3611 &entry->offset, 3612 atop(size)); 3613 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3614 eobject = entry->object.vm_object; 3615 if (eobject->uip != NULL) { 3616 /* 3617 * The object was not shadowed. 3618 */ 3619 swap_release_by_uid(size, entry->uip); 3620 uifree(entry->uip); 3621 entry->uip = NULL; 3622 } else if (entry->uip != NULL) { 3623 VM_OBJECT_LOCK(eobject); 3624 eobject->uip = entry->uip; 3625 eobject->charge = size; 3626 VM_OBJECT_UNLOCK(eobject); 3627 entry->uip = NULL; 3628 } 3629 3630 vm_map_lock_downgrade(map); 3631 } else { 3632 /* 3633 * We're attempting to read a copy-on-write page -- 3634 * don't allow writes. 3635 */ 3636 prot &= ~VM_PROT_WRITE; 3637 } 3638 } 3639 3640 /* 3641 * Create an object if necessary. 3642 */ 3643 if (entry->object.vm_object == NULL && 3644 !map->system_map) { 3645 if (vm_map_lock_upgrade(map)) 3646 goto RetryLookup; 3647 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3648 atop(size)); 3649 entry->offset = 0; 3650 if (entry->uip != NULL) { 3651 VM_OBJECT_LOCK(entry->object.vm_object); 3652 entry->object.vm_object->uip = entry->uip; 3653 entry->object.vm_object->charge = size; 3654 VM_OBJECT_UNLOCK(entry->object.vm_object); 3655 entry->uip = NULL; 3656 } 3657 vm_map_lock_downgrade(map); 3658 } 3659 3660 /* 3661 * Return the object/offset from this entry. If the entry was 3662 * copy-on-write or empty, it has been fixed up. 3663 */ 3664 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3665 *object = entry->object.vm_object; 3666 3667 *out_prot = prot; 3668 return (KERN_SUCCESS); 3669 } 3670 3671 /* 3672 * vm_map_lookup_locked: 3673 * 3674 * Lookup the faulting address. A version of vm_map_lookup that returns 3675 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3676 */ 3677 int 3678 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3679 vm_offset_t vaddr, 3680 vm_prot_t fault_typea, 3681 vm_map_entry_t *out_entry, /* OUT */ 3682 vm_object_t *object, /* OUT */ 3683 vm_pindex_t *pindex, /* OUT */ 3684 vm_prot_t *out_prot, /* OUT */ 3685 boolean_t *wired) /* OUT */ 3686 { 3687 vm_map_entry_t entry; 3688 vm_map_t map = *var_map; 3689 vm_prot_t prot; 3690 vm_prot_t fault_type = fault_typea; 3691 3692 /* 3693 * Lookup the faulting address. 3694 */ 3695 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3696 return (KERN_INVALID_ADDRESS); 3697 3698 entry = *out_entry; 3699 3700 /* 3701 * Fail if the entry refers to a submap. 3702 */ 3703 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3704 return (KERN_FAILURE); 3705 3706 /* 3707 * Check whether this task is allowed to have this page. 3708 * Note the special case for MAP_ENTRY_COW 3709 * pages with an override. This is to implement a forced 3710 * COW for debuggers. 3711 */ 3712 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3713 prot = entry->max_protection; 3714 else 3715 prot = entry->protection; 3716 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3717 if ((fault_type & prot) != fault_type) 3718 return (KERN_PROTECTION_FAILURE); 3719 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3720 (entry->eflags & MAP_ENTRY_COW) && 3721 (fault_type & VM_PROT_WRITE) && 3722 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3723 return (KERN_PROTECTION_FAILURE); 3724 3725 /* 3726 * If this page is not pageable, we have to get it for all possible 3727 * accesses. 3728 */ 3729 *wired = (entry->wired_count != 0); 3730 if (*wired) 3731 prot = fault_type = entry->protection; 3732 3733 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3734 /* 3735 * Fail if the entry was copy-on-write for a write fault. 3736 */ 3737 if (fault_type & VM_PROT_WRITE) 3738 return (KERN_FAILURE); 3739 /* 3740 * We're attempting to read a copy-on-write page -- 3741 * don't allow writes. 3742 */ 3743 prot &= ~VM_PROT_WRITE; 3744 } 3745 3746 /* 3747 * Fail if an object should be created. 3748 */ 3749 if (entry->object.vm_object == NULL && !map->system_map) 3750 return (KERN_FAILURE); 3751 3752 /* 3753 * Return the object/offset from this entry. If the entry was 3754 * copy-on-write or empty, it has been fixed up. 3755 */ 3756 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3757 *object = entry->object.vm_object; 3758 3759 *out_prot = prot; 3760 return (KERN_SUCCESS); 3761 } 3762 3763 /* 3764 * vm_map_lookup_done: 3765 * 3766 * Releases locks acquired by a vm_map_lookup 3767 * (according to the handle returned by that lookup). 3768 */ 3769 void 3770 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3771 { 3772 /* 3773 * Unlock the main-level map 3774 */ 3775 vm_map_unlock_read(map); 3776 } 3777 3778 #include "opt_ddb.h" 3779 #ifdef DDB 3780 #include <sys/kernel.h> 3781 3782 #include <ddb/ddb.h> 3783 3784 /* 3785 * vm_map_print: [ debug ] 3786 */ 3787 DB_SHOW_COMMAND(map, vm_map_print) 3788 { 3789 static int nlines; 3790 /* XXX convert args. */ 3791 vm_map_t map = (vm_map_t)addr; 3792 boolean_t full = have_addr; 3793 3794 vm_map_entry_t entry; 3795 3796 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3797 (void *)map, 3798 (void *)map->pmap, map->nentries, map->timestamp); 3799 nlines++; 3800 3801 if (!full && db_indent) 3802 return; 3803 3804 db_indent += 2; 3805 for (entry = map->header.next; entry != &map->header; 3806 entry = entry->next) { 3807 db_iprintf("map entry %p: start=%p, end=%p\n", 3808 (void *)entry, (void *)entry->start, (void *)entry->end); 3809 nlines++; 3810 { 3811 static char *inheritance_name[4] = 3812 {"share", "copy", "none", "donate_copy"}; 3813 3814 db_iprintf(" prot=%x/%x/%s", 3815 entry->protection, 3816 entry->max_protection, 3817 inheritance_name[(int)(unsigned char)entry->inheritance]); 3818 if (entry->wired_count != 0) 3819 db_printf(", wired"); 3820 } 3821 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3822 db_printf(", share=%p, offset=0x%jx\n", 3823 (void *)entry->object.sub_map, 3824 (uintmax_t)entry->offset); 3825 nlines++; 3826 if ((entry->prev == &map->header) || 3827 (entry->prev->object.sub_map != 3828 entry->object.sub_map)) { 3829 db_indent += 2; 3830 vm_map_print((db_expr_t)(intptr_t) 3831 entry->object.sub_map, 3832 full, 0, (char *)0); 3833 db_indent -= 2; 3834 } 3835 } else { 3836 if (entry->uip != NULL) 3837 db_printf(", uip %d", entry->uip->ui_uid); 3838 db_printf(", object=%p, offset=0x%jx", 3839 (void *)entry->object.vm_object, 3840 (uintmax_t)entry->offset); 3841 if (entry->object.vm_object && entry->object.vm_object->uip) 3842 db_printf(", obj uip %d charge %jx", 3843 entry->object.vm_object->uip->ui_uid, 3844 (uintmax_t)entry->object.vm_object->charge); 3845 if (entry->eflags & MAP_ENTRY_COW) 3846 db_printf(", copy (%s)", 3847 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3848 db_printf("\n"); 3849 nlines++; 3850 3851 if ((entry->prev == &map->header) || 3852 (entry->prev->object.vm_object != 3853 entry->object.vm_object)) { 3854 db_indent += 2; 3855 vm_object_print((db_expr_t)(intptr_t) 3856 entry->object.vm_object, 3857 full, 0, (char *)0); 3858 nlines += 4; 3859 db_indent -= 2; 3860 } 3861 } 3862 } 3863 db_indent -= 2; 3864 if (db_indent == 0) 3865 nlines = 0; 3866 } 3867 3868 3869 DB_SHOW_COMMAND(procvm, procvm) 3870 { 3871 struct proc *p; 3872 3873 if (have_addr) { 3874 p = (struct proc *) addr; 3875 } else { 3876 p = curproc; 3877 } 3878 3879 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3880 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3881 (void *)vmspace_pmap(p->p_vmspace)); 3882 3883 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3884 } 3885 3886 #endif /* DDB */ 3887