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