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 single hint is used to speed up lookups. 104 * 105 * Since portions of maps are specified by start/end addresses, 106 * which may not align with existing map entries, all 107 * routines merely "clip" entries to these start/end values. 108 * [That is, an entry is split into two, bordering at a 109 * start or end value.] Note that these clippings may not 110 * always be necessary (as the two resulting entries are then 111 * not changed); however, the clipping is done for convenience. 112 * 113 * As mentioned above, virtual copy operations are performed 114 * by copying VM object references from one map to 115 * another, and then marking both regions as copy-on-write. 116 */ 117 118 /* 119 * vm_map_startup: 120 * 121 * Initialize the vm_map module. Must be called before 122 * any other vm_map routines. 123 * 124 * Map and entry structures are allocated from the general 125 * purpose memory pool with some exceptions: 126 * 127 * - The kernel map and kmem submap are allocated statically. 128 * - Kernel map entries are allocated out of a static pool. 129 * 130 * These restrictions are necessary since malloc() uses the 131 * maps and requires map entries. 132 */ 133 134 static struct mtx map_sleep_mtx; 135 static uma_zone_t mapentzone; 136 static uma_zone_t kmapentzone; 137 static uma_zone_t mapzone; 138 static uma_zone_t vmspace_zone; 139 static struct vm_object kmapentobj; 140 static int vmspace_zinit(void *mem, int size, int flags); 141 static void vmspace_zfini(void *mem, int size); 142 static int vm_map_zinit(void *mem, int ize, int flags); 143 static void vm_map_zfini(void *mem, int size); 144 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max); 145 146 #ifdef INVARIANTS 147 static void vm_map_zdtor(void *mem, int size, void *arg); 148 static void vmspace_zdtor(void *mem, int size, void *arg); 149 #endif 150 151 /* 152 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 153 * stable. 154 */ 155 #define PROC_VMSPACE_LOCK(p) do { } while (0) 156 #define PROC_VMSPACE_UNLOCK(p) do { } while (0) 157 158 /* 159 * VM_MAP_RANGE_CHECK: [ internal use only ] 160 * 161 * Asserts that the starting and ending region 162 * addresses fall within the valid range of the map. 163 */ 164 #define VM_MAP_RANGE_CHECK(map, start, end) \ 165 { \ 166 if (start < vm_map_min(map)) \ 167 start = vm_map_min(map); \ 168 if (end > vm_map_max(map)) \ 169 end = vm_map_max(map); \ 170 if (start > end) \ 171 start = end; \ 172 } 173 174 void 175 vm_map_startup(void) 176 { 177 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 178 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 179 #ifdef INVARIANTS 180 vm_map_zdtor, 181 #else 182 NULL, 183 #endif 184 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 185 uma_prealloc(mapzone, MAX_KMAP); 186 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 187 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 188 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 189 uma_prealloc(kmapentzone, MAX_KMAPENT); 190 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 191 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 192 } 193 194 static void 195 vmspace_zfini(void *mem, int size) 196 { 197 struct vmspace *vm; 198 199 vm = (struct vmspace *)mem; 200 pmap_release(vmspace_pmap(vm)); 201 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 202 } 203 204 static int 205 vmspace_zinit(void *mem, int size, int flags) 206 { 207 struct vmspace *vm; 208 209 vm = (struct vmspace *)mem; 210 211 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 212 pmap_pinit(vmspace_pmap(vm)); 213 return (0); 214 } 215 216 static void 217 vm_map_zfini(void *mem, int size) 218 { 219 vm_map_t map; 220 221 map = (vm_map_t)mem; 222 mtx_destroy(&map->system_mtx); 223 sx_destroy(&map->lock); 224 } 225 226 static int 227 vm_map_zinit(void *mem, int size, int flags) 228 { 229 vm_map_t map; 230 231 map = (vm_map_t)mem; 232 map->nentries = 0; 233 map->size = 0; 234 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 235 sx_init(&map->lock, "user map"); 236 return (0); 237 } 238 239 #ifdef INVARIANTS 240 static void 241 vmspace_zdtor(void *mem, int size, void *arg) 242 { 243 struct vmspace *vm; 244 245 vm = (struct vmspace *)mem; 246 247 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 248 } 249 static void 250 vm_map_zdtor(void *mem, int size, void *arg) 251 { 252 vm_map_t map; 253 254 map = (vm_map_t)mem; 255 KASSERT(map->nentries == 0, 256 ("map %p nentries == %d on free.", 257 map, map->nentries)); 258 KASSERT(map->size == 0, 259 ("map %p size == %lu on free.", 260 map, (unsigned long)map->size)); 261 } 262 #endif /* INVARIANTS */ 263 264 /* 265 * Allocate a vmspace structure, including a vm_map and pmap, 266 * and initialize those structures. The refcnt is set to 1. 267 */ 268 struct vmspace * 269 vmspace_alloc(min, max) 270 vm_offset_t min, max; 271 { 272 struct vmspace *vm; 273 274 vm = uma_zalloc(vmspace_zone, M_WAITOK); 275 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 276 _vm_map_init(&vm->vm_map, min, max); 277 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ 278 vm->vm_refcnt = 1; 279 vm->vm_shm = NULL; 280 vm->vm_swrss = 0; 281 vm->vm_tsize = 0; 282 vm->vm_dsize = 0; 283 vm->vm_ssize = 0; 284 vm->vm_taddr = 0; 285 vm->vm_daddr = 0; 286 vm->vm_maxsaddr = 0; 287 return (vm); 288 } 289 290 void 291 vm_init2(void) 292 { 293 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 294 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 + 295 maxproc * 2 + maxfiles); 296 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 297 #ifdef INVARIANTS 298 vmspace_zdtor, 299 #else 300 NULL, 301 #endif 302 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 303 } 304 305 static inline void 306 vmspace_dofree(struct vmspace *vm) 307 { 308 CTR1(KTR_VM, "vmspace_free: %p", vm); 309 310 /* 311 * Make sure any SysV shm is freed, it might not have been in 312 * exit1(). 313 */ 314 shmexit(vm); 315 316 /* 317 * Lock the map, to wait out all other references to it. 318 * Delete all of the mappings and pages they hold, then call 319 * the pmap module to reclaim anything left. 320 */ 321 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 322 vm->vm_map.max_offset); 323 324 uma_zfree(vmspace_zone, vm); 325 } 326 327 void 328 vmspace_free(struct vmspace *vm) 329 { 330 int refcnt; 331 332 if (vm->vm_refcnt == 0) 333 panic("vmspace_free: attempt to free already freed vmspace"); 334 335 do 336 refcnt = vm->vm_refcnt; 337 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 338 if (refcnt == 1) 339 vmspace_dofree(vm); 340 } 341 342 void 343 vmspace_exitfree(struct proc *p) 344 { 345 struct vmspace *vm; 346 347 PROC_VMSPACE_LOCK(p); 348 vm = p->p_vmspace; 349 p->p_vmspace = NULL; 350 PROC_VMSPACE_UNLOCK(p); 351 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 352 vmspace_free(vm); 353 } 354 355 void 356 vmspace_exit(struct thread *td) 357 { 358 int refcnt; 359 struct vmspace *vm; 360 struct proc *p; 361 362 /* 363 * Release user portion of address space. 364 * This releases references to vnodes, 365 * which could cause I/O if the file has been unlinked. 366 * Need to do this early enough that we can still sleep. 367 * 368 * The last exiting process to reach this point releases as 369 * much of the environment as it can. vmspace_dofree() is the 370 * slower fallback in case another process had a temporary 371 * reference to the vmspace. 372 */ 373 374 p = td->td_proc; 375 vm = p->p_vmspace; 376 atomic_add_int(&vmspace0.vm_refcnt, 1); 377 do { 378 refcnt = vm->vm_refcnt; 379 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 380 /* Switch now since other proc might free vmspace */ 381 PROC_VMSPACE_LOCK(p); 382 p->p_vmspace = &vmspace0; 383 PROC_VMSPACE_UNLOCK(p); 384 pmap_activate(td); 385 } 386 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 387 if (refcnt == 1) { 388 if (p->p_vmspace != vm) { 389 /* vmspace not yet freed, switch back */ 390 PROC_VMSPACE_LOCK(p); 391 p->p_vmspace = vm; 392 PROC_VMSPACE_UNLOCK(p); 393 pmap_activate(td); 394 } 395 pmap_remove_pages(vmspace_pmap(vm)); 396 /* Switch now since this proc will free vmspace */ 397 PROC_VMSPACE_LOCK(p); 398 p->p_vmspace = &vmspace0; 399 PROC_VMSPACE_UNLOCK(p); 400 pmap_activate(td); 401 vmspace_dofree(vm); 402 } 403 } 404 405 /* Acquire reference to vmspace owned by another process. */ 406 407 struct vmspace * 408 vmspace_acquire_ref(struct proc *p) 409 { 410 struct vmspace *vm; 411 int refcnt; 412 413 PROC_VMSPACE_LOCK(p); 414 vm = p->p_vmspace; 415 if (vm == NULL) { 416 PROC_VMSPACE_UNLOCK(p); 417 return (NULL); 418 } 419 do { 420 refcnt = vm->vm_refcnt; 421 if (refcnt <= 0) { /* Avoid 0->1 transition */ 422 PROC_VMSPACE_UNLOCK(p); 423 return (NULL); 424 } 425 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 426 if (vm != p->p_vmspace) { 427 PROC_VMSPACE_UNLOCK(p); 428 vmspace_free(vm); 429 return (NULL); 430 } 431 PROC_VMSPACE_UNLOCK(p); 432 return (vm); 433 } 434 435 void 436 _vm_map_lock(vm_map_t map, const char *file, int line) 437 { 438 439 if (map->system_map) 440 _mtx_lock_flags(&map->system_mtx, 0, file, line); 441 else 442 (void)_sx_xlock(&map->lock, 0, file, line); 443 map->timestamp++; 444 } 445 446 void 447 _vm_map_unlock(vm_map_t map, const char *file, int line) 448 { 449 450 if (map->system_map) 451 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 452 else 453 _sx_xunlock(&map->lock, file, line); 454 } 455 456 void 457 _vm_map_lock_read(vm_map_t map, const char *file, int line) 458 { 459 460 if (map->system_map) 461 _mtx_lock_flags(&map->system_mtx, 0, file, line); 462 else 463 (void)_sx_xlock(&map->lock, 0, file, line); 464 } 465 466 void 467 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 468 { 469 470 if (map->system_map) 471 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 472 else 473 _sx_xunlock(&map->lock, file, line); 474 } 475 476 int 477 _vm_map_trylock(vm_map_t map, const char *file, int line) 478 { 479 int error; 480 481 error = map->system_map ? 482 !_mtx_trylock(&map->system_mtx, 0, file, line) : 483 !_sx_try_xlock(&map->lock, file, line); 484 if (error == 0) 485 map->timestamp++; 486 return (error == 0); 487 } 488 489 int 490 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 491 { 492 int error; 493 494 error = map->system_map ? 495 !_mtx_trylock(&map->system_mtx, 0, file, line) : 496 !_sx_try_xlock(&map->lock, file, line); 497 return (error == 0); 498 } 499 500 int 501 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 502 { 503 504 #ifdef INVARIANTS 505 if (map->system_map) { 506 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 507 } else 508 _sx_assert(&map->lock, SX_XLOCKED, file, line); 509 #endif 510 map->timestamp++; 511 return (0); 512 } 513 514 void 515 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 516 { 517 518 #ifdef INVARIANTS 519 if (map->system_map) { 520 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 521 } else 522 _sx_assert(&map->lock, SX_XLOCKED, file, line); 523 #endif 524 } 525 526 /* 527 * vm_map_unlock_and_wait: 528 */ 529 int 530 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait) 531 { 532 533 mtx_lock(&map_sleep_mtx); 534 vm_map_unlock(map); 535 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0)); 536 } 537 538 /* 539 * vm_map_wakeup: 540 */ 541 void 542 vm_map_wakeup(vm_map_t map) 543 { 544 545 /* 546 * Acquire and release map_sleep_mtx to prevent a wakeup() 547 * from being performed (and lost) between the vm_map_unlock() 548 * and the msleep() in vm_map_unlock_and_wait(). 549 */ 550 mtx_lock(&map_sleep_mtx); 551 mtx_unlock(&map_sleep_mtx); 552 wakeup(&map->root); 553 } 554 555 long 556 vmspace_resident_count(struct vmspace *vmspace) 557 { 558 return pmap_resident_count(vmspace_pmap(vmspace)); 559 } 560 561 long 562 vmspace_wired_count(struct vmspace *vmspace) 563 { 564 return pmap_wired_count(vmspace_pmap(vmspace)); 565 } 566 567 /* 568 * vm_map_create: 569 * 570 * Creates and returns a new empty VM map with 571 * the given physical map structure, and having 572 * the given lower and upper address bounds. 573 */ 574 vm_map_t 575 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 576 { 577 vm_map_t result; 578 579 result = uma_zalloc(mapzone, M_WAITOK); 580 CTR1(KTR_VM, "vm_map_create: %p", result); 581 _vm_map_init(result, min, max); 582 result->pmap = pmap; 583 return (result); 584 } 585 586 /* 587 * Initialize an existing vm_map structure 588 * such as that in the vmspace structure. 589 * The pmap is set elsewhere. 590 */ 591 static void 592 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 593 { 594 595 map->header.next = map->header.prev = &map->header; 596 map->needs_wakeup = FALSE; 597 map->system_map = 0; 598 map->min_offset = min; 599 map->max_offset = max; 600 map->flags = 0; 601 map->root = NULL; 602 map->timestamp = 0; 603 } 604 605 void 606 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 607 { 608 _vm_map_init(map, min, max); 609 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 610 sx_init(&map->lock, "user map"); 611 } 612 613 /* 614 * vm_map_entry_dispose: [ internal use only ] 615 * 616 * Inverse of vm_map_entry_create. 617 */ 618 static void 619 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 620 { 621 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 622 } 623 624 /* 625 * vm_map_entry_create: [ internal use only ] 626 * 627 * Allocates a VM map entry for insertion. 628 * No entry fields are filled in. 629 */ 630 static vm_map_entry_t 631 vm_map_entry_create(vm_map_t map) 632 { 633 vm_map_entry_t new_entry; 634 635 if (map->system_map) 636 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 637 else 638 new_entry = uma_zalloc(mapentzone, M_WAITOK); 639 if (new_entry == NULL) 640 panic("vm_map_entry_create: kernel resources exhausted"); 641 return (new_entry); 642 } 643 644 /* 645 * vm_map_entry_set_behavior: 646 * 647 * Set the expected access behavior, either normal, random, or 648 * sequential. 649 */ 650 static inline void 651 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 652 { 653 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 654 (behavior & MAP_ENTRY_BEHAV_MASK); 655 } 656 657 /* 658 * vm_map_entry_set_max_free: 659 * 660 * Set the max_free field in a vm_map_entry. 661 */ 662 static inline void 663 vm_map_entry_set_max_free(vm_map_entry_t entry) 664 { 665 666 entry->max_free = entry->adj_free; 667 if (entry->left != NULL && entry->left->max_free > entry->max_free) 668 entry->max_free = entry->left->max_free; 669 if (entry->right != NULL && entry->right->max_free > entry->max_free) 670 entry->max_free = entry->right->max_free; 671 } 672 673 /* 674 * vm_map_entry_splay: 675 * 676 * The Sleator and Tarjan top-down splay algorithm with the 677 * following variation. Max_free must be computed bottom-up, so 678 * on the downward pass, maintain the left and right spines in 679 * reverse order. Then, make a second pass up each side to fix 680 * the pointers and compute max_free. The time bound is O(log n) 681 * amortized. 682 * 683 * The new root is the vm_map_entry containing "addr", or else an 684 * adjacent entry (lower or higher) if addr is not in the tree. 685 * 686 * The map must be locked, and leaves it so. 687 * 688 * Returns: the new root. 689 */ 690 static vm_map_entry_t 691 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 692 { 693 vm_map_entry_t llist, rlist; 694 vm_map_entry_t ltree, rtree; 695 vm_map_entry_t y; 696 697 /* Special case of empty tree. */ 698 if (root == NULL) 699 return (root); 700 701 /* 702 * Pass One: Splay down the tree until we find addr or a NULL 703 * pointer where addr would go. llist and rlist are the two 704 * sides in reverse order (bottom-up), with llist linked by 705 * the right pointer and rlist linked by the left pointer in 706 * the vm_map_entry. Wait until Pass Two to set max_free on 707 * the two spines. 708 */ 709 llist = NULL; 710 rlist = NULL; 711 for (;;) { 712 /* root is never NULL in here. */ 713 if (addr < root->start) { 714 y = root->left; 715 if (y == NULL) 716 break; 717 if (addr < y->start && y->left != NULL) { 718 /* Rotate right and put y on rlist. */ 719 root->left = y->right; 720 y->right = root; 721 vm_map_entry_set_max_free(root); 722 root = y->left; 723 y->left = rlist; 724 rlist = y; 725 } else { 726 /* Put root on rlist. */ 727 root->left = rlist; 728 rlist = root; 729 root = y; 730 } 731 } else { 732 y = root->right; 733 if (addr < root->end || y == NULL) 734 break; 735 if (addr >= y->end && y->right != NULL) { 736 /* Rotate left and put y on llist. */ 737 root->right = y->left; 738 y->left = root; 739 vm_map_entry_set_max_free(root); 740 root = y->right; 741 y->right = llist; 742 llist = y; 743 } else { 744 /* Put root on llist. */ 745 root->right = llist; 746 llist = root; 747 root = y; 748 } 749 } 750 } 751 752 /* 753 * Pass Two: Walk back up the two spines, flip the pointers 754 * and set max_free. The subtrees of the root go at the 755 * bottom of llist and rlist. 756 */ 757 ltree = root->left; 758 while (llist != NULL) { 759 y = llist->right; 760 llist->right = ltree; 761 vm_map_entry_set_max_free(llist); 762 ltree = llist; 763 llist = y; 764 } 765 rtree = root->right; 766 while (rlist != NULL) { 767 y = rlist->left; 768 rlist->left = rtree; 769 vm_map_entry_set_max_free(rlist); 770 rtree = rlist; 771 rlist = y; 772 } 773 774 /* 775 * Final assembly: add ltree and rtree as subtrees of root. 776 */ 777 root->left = ltree; 778 root->right = rtree; 779 vm_map_entry_set_max_free(root); 780 781 return (root); 782 } 783 784 /* 785 * vm_map_entry_{un,}link: 786 * 787 * Insert/remove entries from maps. 788 */ 789 static void 790 vm_map_entry_link(vm_map_t map, 791 vm_map_entry_t after_where, 792 vm_map_entry_t entry) 793 { 794 795 CTR4(KTR_VM, 796 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 797 map->nentries, entry, after_where); 798 map->nentries++; 799 entry->prev = after_where; 800 entry->next = after_where->next; 801 entry->next->prev = entry; 802 after_where->next = entry; 803 804 if (after_where != &map->header) { 805 if (after_where != map->root) 806 vm_map_entry_splay(after_where->start, map->root); 807 entry->right = after_where->right; 808 entry->left = after_where; 809 after_where->right = NULL; 810 after_where->adj_free = entry->start - after_where->end; 811 vm_map_entry_set_max_free(after_where); 812 } else { 813 entry->right = map->root; 814 entry->left = NULL; 815 } 816 entry->adj_free = (entry->next == &map->header ? map->max_offset : 817 entry->next->start) - entry->end; 818 vm_map_entry_set_max_free(entry); 819 map->root = entry; 820 } 821 822 static void 823 vm_map_entry_unlink(vm_map_t map, 824 vm_map_entry_t entry) 825 { 826 vm_map_entry_t next, prev, root; 827 828 if (entry != map->root) 829 vm_map_entry_splay(entry->start, map->root); 830 if (entry->left == NULL) 831 root = entry->right; 832 else { 833 root = vm_map_entry_splay(entry->start, entry->left); 834 root->right = entry->right; 835 root->adj_free = (entry->next == &map->header ? map->max_offset : 836 entry->next->start) - root->end; 837 vm_map_entry_set_max_free(root); 838 } 839 map->root = root; 840 841 prev = entry->prev; 842 next = entry->next; 843 next->prev = prev; 844 prev->next = next; 845 map->nentries--; 846 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 847 map->nentries, entry); 848 } 849 850 /* 851 * vm_map_entry_resize_free: 852 * 853 * Recompute the amount of free space following a vm_map_entry 854 * and propagate that value up the tree. Call this function after 855 * resizing a map entry in-place, that is, without a call to 856 * vm_map_entry_link() or _unlink(). 857 * 858 * The map must be locked, and leaves it so. 859 */ 860 static void 861 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 862 { 863 864 /* 865 * Using splay trees without parent pointers, propagating 866 * max_free up the tree is done by moving the entry to the 867 * root and making the change there. 868 */ 869 if (entry != map->root) 870 map->root = vm_map_entry_splay(entry->start, map->root); 871 872 entry->adj_free = (entry->next == &map->header ? map->max_offset : 873 entry->next->start) - entry->end; 874 vm_map_entry_set_max_free(entry); 875 } 876 877 /* 878 * vm_map_lookup_entry: [ internal use only ] 879 * 880 * Finds the map entry containing (or 881 * immediately preceding) the specified address 882 * in the given map; the entry is returned 883 * in the "entry" parameter. The boolean 884 * result indicates whether the address is 885 * actually contained in the map. 886 */ 887 boolean_t 888 vm_map_lookup_entry( 889 vm_map_t map, 890 vm_offset_t address, 891 vm_map_entry_t *entry) /* OUT */ 892 { 893 vm_map_entry_t cur; 894 895 cur = vm_map_entry_splay(address, map->root); 896 if (cur == NULL) 897 *entry = &map->header; 898 else { 899 map->root = cur; 900 901 if (address >= cur->start) { 902 *entry = cur; 903 if (cur->end > address) 904 return (TRUE); 905 } else 906 *entry = cur->prev; 907 } 908 return (FALSE); 909 } 910 911 /* 912 * vm_map_insert: 913 * 914 * Inserts the given whole VM object into the target 915 * map at the specified address range. The object's 916 * size should match that of the address range. 917 * 918 * Requires that the map be locked, and leaves it so. 919 * 920 * If object is non-NULL, ref count must be bumped by caller 921 * prior to making call to account for the new entry. 922 */ 923 int 924 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 925 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 926 int cow) 927 { 928 vm_map_entry_t new_entry; 929 vm_map_entry_t prev_entry; 930 vm_map_entry_t temp_entry; 931 vm_eflags_t protoeflags; 932 933 /* 934 * Check that the start and end points are not bogus. 935 */ 936 if ((start < map->min_offset) || (end > map->max_offset) || 937 (start >= end)) 938 return (KERN_INVALID_ADDRESS); 939 940 /* 941 * Find the entry prior to the proposed starting address; if it's part 942 * of an existing entry, this range is bogus. 943 */ 944 if (vm_map_lookup_entry(map, start, &temp_entry)) 945 return (KERN_NO_SPACE); 946 947 prev_entry = temp_entry; 948 949 /* 950 * Assert that the next entry doesn't overlap the end point. 951 */ 952 if ((prev_entry->next != &map->header) && 953 (prev_entry->next->start < end)) 954 return (KERN_NO_SPACE); 955 956 protoeflags = 0; 957 958 if (cow & MAP_COPY_ON_WRITE) 959 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 960 961 if (cow & MAP_NOFAULT) { 962 protoeflags |= MAP_ENTRY_NOFAULT; 963 964 KASSERT(object == NULL, 965 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 966 } 967 if (cow & MAP_DISABLE_SYNCER) 968 protoeflags |= MAP_ENTRY_NOSYNC; 969 if (cow & MAP_DISABLE_COREDUMP) 970 protoeflags |= MAP_ENTRY_NOCOREDUMP; 971 972 if (object != NULL) { 973 /* 974 * OBJ_ONEMAPPING must be cleared unless this mapping 975 * is trivially proven to be the only mapping for any 976 * of the object's pages. (Object granularity 977 * reference counting is insufficient to recognize 978 * aliases with precision.) 979 */ 980 VM_OBJECT_LOCK(object); 981 if (object->ref_count > 1 || object->shadow_count != 0) 982 vm_object_clear_flag(object, OBJ_ONEMAPPING); 983 VM_OBJECT_UNLOCK(object); 984 } 985 else if ((prev_entry != &map->header) && 986 (prev_entry->eflags == protoeflags) && 987 (prev_entry->end == start) && 988 (prev_entry->wired_count == 0) && 989 ((prev_entry->object.vm_object == NULL) || 990 vm_object_coalesce(prev_entry->object.vm_object, 991 prev_entry->offset, 992 (vm_size_t)(prev_entry->end - prev_entry->start), 993 (vm_size_t)(end - prev_entry->end)))) { 994 /* 995 * We were able to extend the object. Determine if we 996 * can extend the previous map entry to include the 997 * new range as well. 998 */ 999 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && 1000 (prev_entry->protection == prot) && 1001 (prev_entry->max_protection == max)) { 1002 map->size += (end - prev_entry->end); 1003 prev_entry->end = end; 1004 vm_map_entry_resize_free(map, prev_entry); 1005 vm_map_simplify_entry(map, prev_entry); 1006 return (KERN_SUCCESS); 1007 } 1008 1009 /* 1010 * If we can extend the object but cannot extend the 1011 * map entry, we have to create a new map entry. We 1012 * must bump the ref count on the extended object to 1013 * account for it. object may be NULL. 1014 */ 1015 object = prev_entry->object.vm_object; 1016 offset = prev_entry->offset + 1017 (prev_entry->end - prev_entry->start); 1018 vm_object_reference(object); 1019 } 1020 1021 /* 1022 * NOTE: if conditionals fail, object can be NULL here. This occurs 1023 * in things like the buffer map where we manage kva but do not manage 1024 * backing objects. 1025 */ 1026 1027 /* 1028 * Create a new entry 1029 */ 1030 new_entry = vm_map_entry_create(map); 1031 new_entry->start = start; 1032 new_entry->end = end; 1033 1034 new_entry->eflags = protoeflags; 1035 new_entry->object.vm_object = object; 1036 new_entry->offset = offset; 1037 new_entry->avail_ssize = 0; 1038 1039 new_entry->inheritance = VM_INHERIT_DEFAULT; 1040 new_entry->protection = prot; 1041 new_entry->max_protection = max; 1042 new_entry->wired_count = 0; 1043 1044 /* 1045 * Insert the new entry into the list 1046 */ 1047 vm_map_entry_link(map, prev_entry, new_entry); 1048 map->size += new_entry->end - new_entry->start; 1049 1050 #if 0 1051 /* 1052 * Temporarily removed to avoid MAP_STACK panic, due to 1053 * MAP_STACK being a huge hack. Will be added back in 1054 * when MAP_STACK (and the user stack mapping) is fixed. 1055 */ 1056 /* 1057 * It may be possible to simplify the entry 1058 */ 1059 vm_map_simplify_entry(map, new_entry); 1060 #endif 1061 1062 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1063 vm_map_pmap_enter(map, start, prot, 1064 object, OFF_TO_IDX(offset), end - start, 1065 cow & MAP_PREFAULT_PARTIAL); 1066 } 1067 1068 return (KERN_SUCCESS); 1069 } 1070 1071 /* 1072 * vm_map_findspace: 1073 * 1074 * Find the first fit (lowest VM address) for "length" free bytes 1075 * beginning at address >= start in the given map. 1076 * 1077 * In a vm_map_entry, "adj_free" is the amount of free space 1078 * adjacent (higher address) to this entry, and "max_free" is the 1079 * maximum amount of contiguous free space in its subtree. This 1080 * allows finding a free region in one path down the tree, so 1081 * O(log n) amortized with splay trees. 1082 * 1083 * The map must be locked, and leaves it so. 1084 * 1085 * Returns: 0 on success, and starting address in *addr, 1086 * 1 if insufficient space. 1087 */ 1088 int 1089 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1090 vm_offset_t *addr) /* OUT */ 1091 { 1092 vm_map_entry_t entry; 1093 vm_offset_t end, st; 1094 1095 /* 1096 * Request must fit within min/max VM address and must avoid 1097 * address wrap. 1098 */ 1099 if (start < map->min_offset) 1100 start = map->min_offset; 1101 if (start + length > map->max_offset || start + length < start) 1102 return (1); 1103 1104 /* Empty tree means wide open address space. */ 1105 if (map->root == NULL) { 1106 *addr = start; 1107 goto found; 1108 } 1109 1110 /* 1111 * After splay, if start comes before root node, then there 1112 * must be a gap from start to the root. 1113 */ 1114 map->root = vm_map_entry_splay(start, map->root); 1115 if (start + length <= map->root->start) { 1116 *addr = start; 1117 goto found; 1118 } 1119 1120 /* 1121 * Root is the last node that might begin its gap before 1122 * start, and this is the last comparison where address 1123 * wrap might be a problem. 1124 */ 1125 st = (start > map->root->end) ? start : map->root->end; 1126 if (length <= map->root->end + map->root->adj_free - st) { 1127 *addr = st; 1128 goto found; 1129 } 1130 1131 /* With max_free, can immediately tell if no solution. */ 1132 entry = map->root->right; 1133 if (entry == NULL || length > entry->max_free) 1134 return (1); 1135 1136 /* 1137 * Search the right subtree in the order: left subtree, root, 1138 * right subtree (first fit). The previous splay implies that 1139 * all regions in the right subtree have addresses > start. 1140 */ 1141 while (entry != NULL) { 1142 if (entry->left != NULL && entry->left->max_free >= length) 1143 entry = entry->left; 1144 else if (entry->adj_free >= length) { 1145 *addr = entry->end; 1146 goto found; 1147 } else 1148 entry = entry->right; 1149 } 1150 1151 /* Can't get here, so panic if we do. */ 1152 panic("vm_map_findspace: max_free corrupt"); 1153 1154 found: 1155 /* Expand the kernel pmap, if necessary. */ 1156 if (map == kernel_map) { 1157 end = round_page(*addr + length); 1158 if (end > kernel_vm_end) 1159 pmap_growkernel(end); 1160 } 1161 return (0); 1162 } 1163 1164 int 1165 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1166 vm_offset_t *addr /* IN/OUT */, vm_size_t length, vm_prot_t prot, 1167 vm_prot_t max, int cow) 1168 { 1169 vm_offset_t start, end; 1170 int result; 1171 1172 start = *addr; 1173 vm_map_lock(map); 1174 end = start + length; 1175 VM_MAP_RANGE_CHECK(map, start, end); 1176 (void) vm_map_delete(map, start, end); 1177 result = vm_map_insert(map, object, offset, start, end, prot, 1178 max, cow); 1179 vm_map_unlock(map); 1180 return (result); 1181 } 1182 1183 /* 1184 * vm_map_find finds an unallocated region in the target address 1185 * map with the given length. The search is defined to be 1186 * first-fit from the specified address; the region found is 1187 * returned in the same parameter. 1188 * 1189 * If object is non-NULL, ref count must be bumped by caller 1190 * prior to making call to account for the new entry. 1191 */ 1192 int 1193 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1194 vm_offset_t *addr, /* IN/OUT */ 1195 vm_size_t length, boolean_t find_space, vm_prot_t prot, 1196 vm_prot_t max, int cow) 1197 { 1198 vm_offset_t start; 1199 int result; 1200 1201 start = *addr; 1202 vm_map_lock(map); 1203 if (find_space) { 1204 if (vm_map_findspace(map, start, length, addr)) { 1205 vm_map_unlock(map); 1206 return (KERN_NO_SPACE); 1207 } 1208 start = *addr; 1209 } 1210 result = vm_map_insert(map, object, offset, 1211 start, start + length, prot, max, cow); 1212 vm_map_unlock(map); 1213 return (result); 1214 } 1215 1216 /* 1217 * vm_map_simplify_entry: 1218 * 1219 * Simplify the given map entry by merging with either neighbor. This 1220 * routine also has the ability to merge with both neighbors. 1221 * 1222 * The map must be locked. 1223 * 1224 * This routine guarentees that the passed entry remains valid (though 1225 * possibly extended). When merging, this routine may delete one or 1226 * both neighbors. 1227 */ 1228 void 1229 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1230 { 1231 vm_map_entry_t next, prev; 1232 vm_size_t prevsize, esize; 1233 1234 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1235 return; 1236 1237 prev = entry->prev; 1238 if (prev != &map->header) { 1239 prevsize = prev->end - prev->start; 1240 if ( (prev->end == entry->start) && 1241 (prev->object.vm_object == entry->object.vm_object) && 1242 (!prev->object.vm_object || 1243 (prev->offset + prevsize == entry->offset)) && 1244 (prev->eflags == entry->eflags) && 1245 (prev->protection == entry->protection) && 1246 (prev->max_protection == entry->max_protection) && 1247 (prev->inheritance == entry->inheritance) && 1248 (prev->wired_count == entry->wired_count)) { 1249 vm_map_entry_unlink(map, prev); 1250 entry->start = prev->start; 1251 entry->offset = prev->offset; 1252 if (entry->prev != &map->header) 1253 vm_map_entry_resize_free(map, entry->prev); 1254 if (prev->object.vm_object) 1255 vm_object_deallocate(prev->object.vm_object); 1256 vm_map_entry_dispose(map, prev); 1257 } 1258 } 1259 1260 next = entry->next; 1261 if (next != &map->header) { 1262 esize = entry->end - entry->start; 1263 if ((entry->end == next->start) && 1264 (next->object.vm_object == entry->object.vm_object) && 1265 (!entry->object.vm_object || 1266 (entry->offset + esize == next->offset)) && 1267 (next->eflags == entry->eflags) && 1268 (next->protection == entry->protection) && 1269 (next->max_protection == entry->max_protection) && 1270 (next->inheritance == entry->inheritance) && 1271 (next->wired_count == entry->wired_count)) { 1272 vm_map_entry_unlink(map, next); 1273 entry->end = next->end; 1274 vm_map_entry_resize_free(map, entry); 1275 if (next->object.vm_object) 1276 vm_object_deallocate(next->object.vm_object); 1277 vm_map_entry_dispose(map, next); 1278 } 1279 } 1280 } 1281 /* 1282 * vm_map_clip_start: [ internal use only ] 1283 * 1284 * Asserts that the given entry begins at or after 1285 * the specified address; if necessary, 1286 * it splits the entry into two. 1287 */ 1288 #define vm_map_clip_start(map, entry, startaddr) \ 1289 { \ 1290 if (startaddr > entry->start) \ 1291 _vm_map_clip_start(map, entry, startaddr); \ 1292 } 1293 1294 /* 1295 * This routine is called only when it is known that 1296 * the entry must be split. 1297 */ 1298 static void 1299 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1300 { 1301 vm_map_entry_t new_entry; 1302 1303 /* 1304 * Split off the front portion -- note that we must insert the new 1305 * entry BEFORE this one, so that this entry has the specified 1306 * starting address. 1307 */ 1308 vm_map_simplify_entry(map, entry); 1309 1310 /* 1311 * If there is no object backing this entry, we might as well create 1312 * one now. If we defer it, an object can get created after the map 1313 * is clipped, and individual objects will be created for the split-up 1314 * map. This is a bit of a hack, but is also about the best place to 1315 * put this improvement. 1316 */ 1317 if (entry->object.vm_object == NULL && !map->system_map) { 1318 vm_object_t object; 1319 object = vm_object_allocate(OBJT_DEFAULT, 1320 atop(entry->end - entry->start)); 1321 entry->object.vm_object = object; 1322 entry->offset = 0; 1323 } 1324 1325 new_entry = vm_map_entry_create(map); 1326 *new_entry = *entry; 1327 1328 new_entry->end = start; 1329 entry->offset += (start - entry->start); 1330 entry->start = start; 1331 1332 vm_map_entry_link(map, entry->prev, new_entry); 1333 1334 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1335 vm_object_reference(new_entry->object.vm_object); 1336 } 1337 } 1338 1339 /* 1340 * vm_map_clip_end: [ internal use only ] 1341 * 1342 * Asserts that the given entry ends at or before 1343 * the specified address; if necessary, 1344 * it splits the entry into two. 1345 */ 1346 #define vm_map_clip_end(map, entry, endaddr) \ 1347 { \ 1348 if ((endaddr) < (entry->end)) \ 1349 _vm_map_clip_end((map), (entry), (endaddr)); \ 1350 } 1351 1352 /* 1353 * This routine is called only when it is known that 1354 * the entry must be split. 1355 */ 1356 static void 1357 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1358 { 1359 vm_map_entry_t new_entry; 1360 1361 /* 1362 * If there is no object backing this entry, we might as well create 1363 * one now. If we defer it, an object can get created after the map 1364 * is clipped, and individual objects will be created for the split-up 1365 * map. This is a bit of a hack, but is also about the best place to 1366 * put this improvement. 1367 */ 1368 if (entry->object.vm_object == NULL && !map->system_map) { 1369 vm_object_t object; 1370 object = vm_object_allocate(OBJT_DEFAULT, 1371 atop(entry->end - entry->start)); 1372 entry->object.vm_object = object; 1373 entry->offset = 0; 1374 } 1375 1376 /* 1377 * Create a new entry and insert it AFTER the specified entry 1378 */ 1379 new_entry = vm_map_entry_create(map); 1380 *new_entry = *entry; 1381 1382 new_entry->start = entry->end = end; 1383 new_entry->offset += (end - entry->start); 1384 1385 vm_map_entry_link(map, entry, new_entry); 1386 1387 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1388 vm_object_reference(new_entry->object.vm_object); 1389 } 1390 } 1391 1392 /* 1393 * vm_map_submap: [ kernel use only ] 1394 * 1395 * Mark the given range as handled by a subordinate map. 1396 * 1397 * This range must have been created with vm_map_find, 1398 * and no other operations may have been performed on this 1399 * range prior to calling vm_map_submap. 1400 * 1401 * Only a limited number of operations can be performed 1402 * within this rage after calling vm_map_submap: 1403 * vm_fault 1404 * [Don't try vm_map_copy!] 1405 * 1406 * To remove a submapping, one must first remove the 1407 * range from the superior map, and then destroy the 1408 * submap (if desired). [Better yet, don't try it.] 1409 */ 1410 int 1411 vm_map_submap( 1412 vm_map_t map, 1413 vm_offset_t start, 1414 vm_offset_t end, 1415 vm_map_t submap) 1416 { 1417 vm_map_entry_t entry; 1418 int result = KERN_INVALID_ARGUMENT; 1419 1420 vm_map_lock(map); 1421 1422 VM_MAP_RANGE_CHECK(map, start, end); 1423 1424 if (vm_map_lookup_entry(map, start, &entry)) { 1425 vm_map_clip_start(map, entry, start); 1426 } else 1427 entry = entry->next; 1428 1429 vm_map_clip_end(map, entry, end); 1430 1431 if ((entry->start == start) && (entry->end == end) && 1432 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1433 (entry->object.vm_object == NULL)) { 1434 entry->object.sub_map = submap; 1435 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1436 result = KERN_SUCCESS; 1437 } 1438 vm_map_unlock(map); 1439 1440 return (result); 1441 } 1442 1443 /* 1444 * The maximum number of pages to map 1445 */ 1446 #define MAX_INIT_PT 96 1447 1448 /* 1449 * vm_map_pmap_enter: 1450 * 1451 * Preload read-only mappings for the given object's resident pages into 1452 * the given map. This eliminates the soft faults on process startup and 1453 * immediately after an mmap(2). Unless the given flags include 1454 * MAP_PREFAULT_MADVISE, cached pages are not reactivated and mapped. 1455 */ 1456 void 1457 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1458 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1459 { 1460 vm_offset_t start; 1461 vm_page_t p, p_start; 1462 vm_pindex_t psize, tmpidx; 1463 boolean_t are_queues_locked; 1464 1465 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1466 return; 1467 VM_OBJECT_LOCK(object); 1468 if (object->type == OBJT_DEVICE) { 1469 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1470 goto unlock_return; 1471 } 1472 1473 psize = atop(size); 1474 1475 if (object->type != OBJT_VNODE || 1476 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 1477 (object->resident_page_count > MAX_INIT_PT))) { 1478 goto unlock_return; 1479 } 1480 1481 if (psize + pindex > object->size) { 1482 if (object->size < pindex) 1483 goto unlock_return; 1484 psize = object->size - pindex; 1485 } 1486 1487 are_queues_locked = FALSE; 1488 start = 0; 1489 p_start = NULL; 1490 1491 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1492 if (p->pindex < pindex) { 1493 p = vm_page_splay(pindex, object->root); 1494 if ((object->root = p)->pindex < pindex) 1495 p = TAILQ_NEXT(p, listq); 1496 } 1497 } 1498 /* 1499 * Assert: the variable p is either (1) the page with the 1500 * least pindex greater than or equal to the parameter pindex 1501 * or (2) NULL. 1502 */ 1503 for (; 1504 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1505 p = TAILQ_NEXT(p, listq)) { 1506 /* 1507 * don't allow an madvise to blow away our really 1508 * free pages allocating pv entries. 1509 */ 1510 if ((flags & MAP_PREFAULT_MADVISE) && 1511 cnt.v_free_count < cnt.v_free_reserved) { 1512 psize = tmpidx; 1513 break; 1514 } 1515 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 1516 (p->busy == 0)) { 1517 if (p_start == NULL) { 1518 start = addr + ptoa(tmpidx); 1519 p_start = p; 1520 } 1521 } else if (p_start != NULL) { 1522 if (!are_queues_locked) { 1523 are_queues_locked = TRUE; 1524 vm_page_lock_queues(); 1525 } 1526 pmap_enter_object(map->pmap, start, addr + 1527 ptoa(tmpidx), p_start, prot); 1528 p_start = NULL; 1529 } 1530 } 1531 if (p_start != NULL) { 1532 if (!are_queues_locked) { 1533 are_queues_locked = TRUE; 1534 vm_page_lock_queues(); 1535 } 1536 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1537 p_start, prot); 1538 } 1539 if (are_queues_locked) 1540 vm_page_unlock_queues(); 1541 unlock_return: 1542 VM_OBJECT_UNLOCK(object); 1543 } 1544 1545 /* 1546 * vm_map_protect: 1547 * 1548 * Sets the protection of the specified address 1549 * region in the target map. If "set_max" is 1550 * specified, the maximum protection is to be set; 1551 * otherwise, only the current protection is affected. 1552 */ 1553 int 1554 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1555 vm_prot_t new_prot, boolean_t set_max) 1556 { 1557 vm_map_entry_t current; 1558 vm_map_entry_t entry; 1559 1560 vm_map_lock(map); 1561 1562 VM_MAP_RANGE_CHECK(map, start, end); 1563 1564 if (vm_map_lookup_entry(map, start, &entry)) { 1565 vm_map_clip_start(map, entry, start); 1566 } else { 1567 entry = entry->next; 1568 } 1569 1570 /* 1571 * Make a first pass to check for protection violations. 1572 */ 1573 current = entry; 1574 while ((current != &map->header) && (current->start < end)) { 1575 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1576 vm_map_unlock(map); 1577 return (KERN_INVALID_ARGUMENT); 1578 } 1579 if ((new_prot & current->max_protection) != new_prot) { 1580 vm_map_unlock(map); 1581 return (KERN_PROTECTION_FAILURE); 1582 } 1583 current = current->next; 1584 } 1585 1586 /* 1587 * Go back and fix up protections. [Note that clipping is not 1588 * necessary the second time.] 1589 */ 1590 current = entry; 1591 while ((current != &map->header) && (current->start < end)) { 1592 vm_prot_t old_prot; 1593 1594 vm_map_clip_end(map, current, end); 1595 1596 old_prot = current->protection; 1597 if (set_max) 1598 current->protection = 1599 (current->max_protection = new_prot) & 1600 old_prot; 1601 else 1602 current->protection = new_prot; 1603 1604 /* 1605 * Update physical map if necessary. Worry about copy-on-write 1606 * here -- CHECK THIS XXX 1607 */ 1608 if (current->protection != old_prot) { 1609 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1610 VM_PROT_ALL) 1611 pmap_protect(map->pmap, current->start, 1612 current->end, 1613 current->protection & MASK(current)); 1614 #undef MASK 1615 } 1616 vm_map_simplify_entry(map, current); 1617 current = current->next; 1618 } 1619 vm_map_unlock(map); 1620 return (KERN_SUCCESS); 1621 } 1622 1623 /* 1624 * vm_map_madvise: 1625 * 1626 * This routine traverses a processes map handling the madvise 1627 * system call. Advisories are classified as either those effecting 1628 * the vm_map_entry structure, or those effecting the underlying 1629 * objects. 1630 */ 1631 int 1632 vm_map_madvise( 1633 vm_map_t map, 1634 vm_offset_t start, 1635 vm_offset_t end, 1636 int behav) 1637 { 1638 vm_map_entry_t current, entry; 1639 int modify_map = 0; 1640 1641 /* 1642 * Some madvise calls directly modify the vm_map_entry, in which case 1643 * we need to use an exclusive lock on the map and we need to perform 1644 * various clipping operations. Otherwise we only need a read-lock 1645 * on the map. 1646 */ 1647 switch(behav) { 1648 case MADV_NORMAL: 1649 case MADV_SEQUENTIAL: 1650 case MADV_RANDOM: 1651 case MADV_NOSYNC: 1652 case MADV_AUTOSYNC: 1653 case MADV_NOCORE: 1654 case MADV_CORE: 1655 modify_map = 1; 1656 vm_map_lock(map); 1657 break; 1658 case MADV_WILLNEED: 1659 case MADV_DONTNEED: 1660 case MADV_FREE: 1661 vm_map_lock_read(map); 1662 break; 1663 default: 1664 return (KERN_INVALID_ARGUMENT); 1665 } 1666 1667 /* 1668 * Locate starting entry and clip if necessary. 1669 */ 1670 VM_MAP_RANGE_CHECK(map, start, end); 1671 1672 if (vm_map_lookup_entry(map, start, &entry)) { 1673 if (modify_map) 1674 vm_map_clip_start(map, entry, start); 1675 } else { 1676 entry = entry->next; 1677 } 1678 1679 if (modify_map) { 1680 /* 1681 * madvise behaviors that are implemented in the vm_map_entry. 1682 * 1683 * We clip the vm_map_entry so that behavioral changes are 1684 * limited to the specified address range. 1685 */ 1686 for (current = entry; 1687 (current != &map->header) && (current->start < end); 1688 current = current->next 1689 ) { 1690 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1691 continue; 1692 1693 vm_map_clip_end(map, current, end); 1694 1695 switch (behav) { 1696 case MADV_NORMAL: 1697 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 1698 break; 1699 case MADV_SEQUENTIAL: 1700 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 1701 break; 1702 case MADV_RANDOM: 1703 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 1704 break; 1705 case MADV_NOSYNC: 1706 current->eflags |= MAP_ENTRY_NOSYNC; 1707 break; 1708 case MADV_AUTOSYNC: 1709 current->eflags &= ~MAP_ENTRY_NOSYNC; 1710 break; 1711 case MADV_NOCORE: 1712 current->eflags |= MAP_ENTRY_NOCOREDUMP; 1713 break; 1714 case MADV_CORE: 1715 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 1716 break; 1717 default: 1718 break; 1719 } 1720 vm_map_simplify_entry(map, current); 1721 } 1722 vm_map_unlock(map); 1723 } else { 1724 vm_pindex_t pindex; 1725 int count; 1726 1727 /* 1728 * madvise behaviors that are implemented in the underlying 1729 * vm_object. 1730 * 1731 * Since we don't clip the vm_map_entry, we have to clip 1732 * the vm_object pindex and count. 1733 */ 1734 for (current = entry; 1735 (current != &map->header) && (current->start < end); 1736 current = current->next 1737 ) { 1738 vm_offset_t useStart; 1739 1740 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1741 continue; 1742 1743 pindex = OFF_TO_IDX(current->offset); 1744 count = atop(current->end - current->start); 1745 useStart = current->start; 1746 1747 if (current->start < start) { 1748 pindex += atop(start - current->start); 1749 count -= atop(start - current->start); 1750 useStart = start; 1751 } 1752 if (current->end > end) 1753 count -= atop(current->end - end); 1754 1755 if (count <= 0) 1756 continue; 1757 1758 vm_object_madvise(current->object.vm_object, 1759 pindex, count, behav); 1760 if (behav == MADV_WILLNEED) { 1761 vm_map_pmap_enter(map, 1762 useStart, 1763 current->protection, 1764 current->object.vm_object, 1765 pindex, 1766 (count << PAGE_SHIFT), 1767 MAP_PREFAULT_MADVISE 1768 ); 1769 } 1770 } 1771 vm_map_unlock_read(map); 1772 } 1773 return (0); 1774 } 1775 1776 1777 /* 1778 * vm_map_inherit: 1779 * 1780 * Sets the inheritance of the specified address 1781 * range in the target map. Inheritance 1782 * affects how the map will be shared with 1783 * child maps at the time of vm_map_fork. 1784 */ 1785 int 1786 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 1787 vm_inherit_t new_inheritance) 1788 { 1789 vm_map_entry_t entry; 1790 vm_map_entry_t temp_entry; 1791 1792 switch (new_inheritance) { 1793 case VM_INHERIT_NONE: 1794 case VM_INHERIT_COPY: 1795 case VM_INHERIT_SHARE: 1796 break; 1797 default: 1798 return (KERN_INVALID_ARGUMENT); 1799 } 1800 vm_map_lock(map); 1801 VM_MAP_RANGE_CHECK(map, start, end); 1802 if (vm_map_lookup_entry(map, start, &temp_entry)) { 1803 entry = temp_entry; 1804 vm_map_clip_start(map, entry, start); 1805 } else 1806 entry = temp_entry->next; 1807 while ((entry != &map->header) && (entry->start < end)) { 1808 vm_map_clip_end(map, entry, end); 1809 entry->inheritance = new_inheritance; 1810 vm_map_simplify_entry(map, entry); 1811 entry = entry->next; 1812 } 1813 vm_map_unlock(map); 1814 return (KERN_SUCCESS); 1815 } 1816 1817 /* 1818 * vm_map_unwire: 1819 * 1820 * Implements both kernel and user unwiring. 1821 */ 1822 int 1823 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1824 int flags) 1825 { 1826 vm_map_entry_t entry, first_entry, tmp_entry; 1827 vm_offset_t saved_start; 1828 unsigned int last_timestamp; 1829 int rv; 1830 boolean_t need_wakeup, result, user_unwire; 1831 1832 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1833 vm_map_lock(map); 1834 VM_MAP_RANGE_CHECK(map, start, end); 1835 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1836 if (flags & VM_MAP_WIRE_HOLESOK) 1837 first_entry = first_entry->next; 1838 else { 1839 vm_map_unlock(map); 1840 return (KERN_INVALID_ADDRESS); 1841 } 1842 } 1843 last_timestamp = map->timestamp; 1844 entry = first_entry; 1845 while (entry != &map->header && entry->start < end) { 1846 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1847 /* 1848 * We have not yet clipped the entry. 1849 */ 1850 saved_start = (start >= entry->start) ? start : 1851 entry->start; 1852 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1853 if (vm_map_unlock_and_wait(map, user_unwire)) { 1854 /* 1855 * Allow interruption of user unwiring? 1856 */ 1857 } 1858 vm_map_lock(map); 1859 if (last_timestamp+1 != map->timestamp) { 1860 /* 1861 * Look again for the entry because the map was 1862 * modified while it was unlocked. 1863 * Specifically, the entry may have been 1864 * clipped, merged, or deleted. 1865 */ 1866 if (!vm_map_lookup_entry(map, saved_start, 1867 &tmp_entry)) { 1868 if (flags & VM_MAP_WIRE_HOLESOK) 1869 tmp_entry = tmp_entry->next; 1870 else { 1871 if (saved_start == start) { 1872 /* 1873 * First_entry has been deleted. 1874 */ 1875 vm_map_unlock(map); 1876 return (KERN_INVALID_ADDRESS); 1877 } 1878 end = saved_start; 1879 rv = KERN_INVALID_ADDRESS; 1880 goto done; 1881 } 1882 } 1883 if (entry == first_entry) 1884 first_entry = tmp_entry; 1885 else 1886 first_entry = NULL; 1887 entry = tmp_entry; 1888 } 1889 last_timestamp = map->timestamp; 1890 continue; 1891 } 1892 vm_map_clip_start(map, entry, start); 1893 vm_map_clip_end(map, entry, end); 1894 /* 1895 * Mark the entry in case the map lock is released. (See 1896 * above.) 1897 */ 1898 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1899 /* 1900 * Check the map for holes in the specified region. 1901 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 1902 */ 1903 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 1904 (entry->end < end && (entry->next == &map->header || 1905 entry->next->start > entry->end))) { 1906 end = entry->end; 1907 rv = KERN_INVALID_ADDRESS; 1908 goto done; 1909 } 1910 /* 1911 * If system unwiring, require that the entry is system wired. 1912 */ 1913 if (!user_unwire && 1914 vm_map_entry_system_wired_count(entry) == 0) { 1915 end = entry->end; 1916 rv = KERN_INVALID_ARGUMENT; 1917 goto done; 1918 } 1919 entry = entry->next; 1920 } 1921 rv = KERN_SUCCESS; 1922 done: 1923 need_wakeup = FALSE; 1924 if (first_entry == NULL) { 1925 result = vm_map_lookup_entry(map, start, &first_entry); 1926 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 1927 first_entry = first_entry->next; 1928 else 1929 KASSERT(result, ("vm_map_unwire: lookup failed")); 1930 } 1931 entry = first_entry; 1932 while (entry != &map->header && entry->start < end) { 1933 if (rv == KERN_SUCCESS && (!user_unwire || 1934 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 1935 if (user_unwire) 1936 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1937 entry->wired_count--; 1938 if (entry->wired_count == 0) { 1939 /* 1940 * Retain the map lock. 1941 */ 1942 vm_fault_unwire(map, entry->start, entry->end, 1943 entry->object.vm_object != NULL && 1944 entry->object.vm_object->type == OBJT_DEVICE); 1945 } 1946 } 1947 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 1948 ("vm_map_unwire: in-transition flag missing")); 1949 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 1950 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 1951 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 1952 need_wakeup = TRUE; 1953 } 1954 vm_map_simplify_entry(map, entry); 1955 entry = entry->next; 1956 } 1957 vm_map_unlock(map); 1958 if (need_wakeup) 1959 vm_map_wakeup(map); 1960 return (rv); 1961 } 1962 1963 /* 1964 * vm_map_wire: 1965 * 1966 * Implements both kernel and user wiring. 1967 */ 1968 int 1969 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1970 int flags) 1971 { 1972 vm_map_entry_t entry, first_entry, tmp_entry; 1973 vm_offset_t saved_end, saved_start; 1974 unsigned int last_timestamp; 1975 int rv; 1976 boolean_t fictitious, need_wakeup, result, user_wire; 1977 1978 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1979 vm_map_lock(map); 1980 VM_MAP_RANGE_CHECK(map, start, end); 1981 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1982 if (flags & VM_MAP_WIRE_HOLESOK) 1983 first_entry = first_entry->next; 1984 else { 1985 vm_map_unlock(map); 1986 return (KERN_INVALID_ADDRESS); 1987 } 1988 } 1989 last_timestamp = map->timestamp; 1990 entry = first_entry; 1991 while (entry != &map->header && entry->start < end) { 1992 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1993 /* 1994 * We have not yet clipped the entry. 1995 */ 1996 saved_start = (start >= entry->start) ? start : 1997 entry->start; 1998 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1999 if (vm_map_unlock_and_wait(map, user_wire)) { 2000 /* 2001 * Allow interruption of user wiring? 2002 */ 2003 } 2004 vm_map_lock(map); 2005 if (last_timestamp + 1 != map->timestamp) { 2006 /* 2007 * Look again for the entry because the map was 2008 * modified while it was unlocked. 2009 * Specifically, the entry may have been 2010 * clipped, merged, or deleted. 2011 */ 2012 if (!vm_map_lookup_entry(map, saved_start, 2013 &tmp_entry)) { 2014 if (flags & VM_MAP_WIRE_HOLESOK) 2015 tmp_entry = tmp_entry->next; 2016 else { 2017 if (saved_start == start) { 2018 /* 2019 * first_entry has been deleted. 2020 */ 2021 vm_map_unlock(map); 2022 return (KERN_INVALID_ADDRESS); 2023 } 2024 end = saved_start; 2025 rv = KERN_INVALID_ADDRESS; 2026 goto done; 2027 } 2028 } 2029 if (entry == first_entry) 2030 first_entry = tmp_entry; 2031 else 2032 first_entry = NULL; 2033 entry = tmp_entry; 2034 } 2035 last_timestamp = map->timestamp; 2036 continue; 2037 } 2038 vm_map_clip_start(map, entry, start); 2039 vm_map_clip_end(map, entry, end); 2040 /* 2041 * Mark the entry in case the map lock is released. (See 2042 * above.) 2043 */ 2044 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2045 /* 2046 * 2047 */ 2048 if (entry->wired_count == 0) { 2049 entry->wired_count++; 2050 saved_start = entry->start; 2051 saved_end = entry->end; 2052 fictitious = entry->object.vm_object != NULL && 2053 entry->object.vm_object->type == OBJT_DEVICE; 2054 /* 2055 * Release the map lock, relying on the in-transition 2056 * mark. 2057 */ 2058 vm_map_unlock(map); 2059 rv = vm_fault_wire(map, saved_start, saved_end, 2060 user_wire, fictitious); 2061 vm_map_lock(map); 2062 if (last_timestamp + 1 != map->timestamp) { 2063 /* 2064 * Look again for the entry because the map was 2065 * modified while it was unlocked. The entry 2066 * may have been clipped, but NOT merged or 2067 * deleted. 2068 */ 2069 result = vm_map_lookup_entry(map, saved_start, 2070 &tmp_entry); 2071 KASSERT(result, ("vm_map_wire: lookup failed")); 2072 if (entry == first_entry) 2073 first_entry = tmp_entry; 2074 else 2075 first_entry = NULL; 2076 entry = tmp_entry; 2077 while (entry->end < saved_end) { 2078 if (rv != KERN_SUCCESS) { 2079 KASSERT(entry->wired_count == 1, 2080 ("vm_map_wire: bad count")); 2081 entry->wired_count = -1; 2082 } 2083 entry = entry->next; 2084 } 2085 } 2086 last_timestamp = map->timestamp; 2087 if (rv != KERN_SUCCESS) { 2088 KASSERT(entry->wired_count == 1, 2089 ("vm_map_wire: bad count")); 2090 /* 2091 * Assign an out-of-range value to represent 2092 * the failure to wire this entry. 2093 */ 2094 entry->wired_count = -1; 2095 end = entry->end; 2096 goto done; 2097 } 2098 } else if (!user_wire || 2099 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2100 entry->wired_count++; 2101 } 2102 /* 2103 * Check the map for holes in the specified region. 2104 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2105 */ 2106 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2107 (entry->end < end && (entry->next == &map->header || 2108 entry->next->start > entry->end))) { 2109 end = entry->end; 2110 rv = KERN_INVALID_ADDRESS; 2111 goto done; 2112 } 2113 entry = entry->next; 2114 } 2115 rv = KERN_SUCCESS; 2116 done: 2117 need_wakeup = FALSE; 2118 if (first_entry == NULL) { 2119 result = vm_map_lookup_entry(map, start, &first_entry); 2120 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2121 first_entry = first_entry->next; 2122 else 2123 KASSERT(result, ("vm_map_wire: lookup failed")); 2124 } 2125 entry = first_entry; 2126 while (entry != &map->header && entry->start < end) { 2127 if (rv == KERN_SUCCESS) { 2128 if (user_wire) 2129 entry->eflags |= MAP_ENTRY_USER_WIRED; 2130 } else if (entry->wired_count == -1) { 2131 /* 2132 * Wiring failed on this entry. Thus, unwiring is 2133 * unnecessary. 2134 */ 2135 entry->wired_count = 0; 2136 } else { 2137 if (!user_wire || 2138 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2139 entry->wired_count--; 2140 if (entry->wired_count == 0) { 2141 /* 2142 * Retain the map lock. 2143 */ 2144 vm_fault_unwire(map, entry->start, entry->end, 2145 entry->object.vm_object != NULL && 2146 entry->object.vm_object->type == OBJT_DEVICE); 2147 } 2148 } 2149 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2150 ("vm_map_wire: in-transition flag missing")); 2151 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2152 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2153 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2154 need_wakeup = TRUE; 2155 } 2156 vm_map_simplify_entry(map, entry); 2157 entry = entry->next; 2158 } 2159 vm_map_unlock(map); 2160 if (need_wakeup) 2161 vm_map_wakeup(map); 2162 return (rv); 2163 } 2164 2165 /* 2166 * vm_map_sync 2167 * 2168 * Push any dirty cached pages in the address range to their pager. 2169 * If syncio is TRUE, dirty pages are written synchronously. 2170 * If invalidate is TRUE, any cached pages are freed as well. 2171 * 2172 * If the size of the region from start to end is zero, we are 2173 * supposed to flush all modified pages within the region containing 2174 * start. Unfortunately, a region can be split or coalesced with 2175 * neighboring regions, making it difficult to determine what the 2176 * original region was. Therefore, we approximate this requirement by 2177 * flushing the current region containing start. 2178 * 2179 * Returns an error if any part of the specified range is not mapped. 2180 */ 2181 int 2182 vm_map_sync( 2183 vm_map_t map, 2184 vm_offset_t start, 2185 vm_offset_t end, 2186 boolean_t syncio, 2187 boolean_t invalidate) 2188 { 2189 vm_map_entry_t current; 2190 vm_map_entry_t entry; 2191 vm_size_t size; 2192 vm_object_t object; 2193 vm_ooffset_t offset; 2194 2195 vm_map_lock_read(map); 2196 VM_MAP_RANGE_CHECK(map, start, end); 2197 if (!vm_map_lookup_entry(map, start, &entry)) { 2198 vm_map_unlock_read(map); 2199 return (KERN_INVALID_ADDRESS); 2200 } else if (start == end) { 2201 start = entry->start; 2202 end = entry->end; 2203 } 2204 /* 2205 * Make a first pass to check for user-wired memory and holes. 2206 */ 2207 for (current = entry; current->start < end; current = current->next) { 2208 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2209 vm_map_unlock_read(map); 2210 return (KERN_INVALID_ARGUMENT); 2211 } 2212 if (end > current->end && 2213 (current->next == &map->header || 2214 current->end != current->next->start)) { 2215 vm_map_unlock_read(map); 2216 return (KERN_INVALID_ADDRESS); 2217 } 2218 } 2219 2220 if (invalidate) 2221 pmap_remove(map->pmap, start, end); 2222 2223 /* 2224 * Make a second pass, cleaning/uncaching pages from the indicated 2225 * objects as we go. 2226 */ 2227 for (current = entry; current->start < end; current = current->next) { 2228 offset = current->offset + (start - current->start); 2229 size = (end <= current->end ? end : current->end) - start; 2230 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2231 vm_map_t smap; 2232 vm_map_entry_t tentry; 2233 vm_size_t tsize; 2234 2235 smap = current->object.sub_map; 2236 vm_map_lock_read(smap); 2237 (void) vm_map_lookup_entry(smap, offset, &tentry); 2238 tsize = tentry->end - offset; 2239 if (tsize < size) 2240 size = tsize; 2241 object = tentry->object.vm_object; 2242 offset = tentry->offset + (offset - tentry->start); 2243 vm_map_unlock_read(smap); 2244 } else { 2245 object = current->object.vm_object; 2246 } 2247 vm_object_sync(object, offset, size, syncio, invalidate); 2248 start += size; 2249 } 2250 2251 vm_map_unlock_read(map); 2252 return (KERN_SUCCESS); 2253 } 2254 2255 /* 2256 * vm_map_entry_unwire: [ internal use only ] 2257 * 2258 * Make the region specified by this entry pageable. 2259 * 2260 * The map in question should be locked. 2261 * [This is the reason for this routine's existence.] 2262 */ 2263 static void 2264 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2265 { 2266 vm_fault_unwire(map, entry->start, entry->end, 2267 entry->object.vm_object != NULL && 2268 entry->object.vm_object->type == OBJT_DEVICE); 2269 entry->wired_count = 0; 2270 } 2271 2272 /* 2273 * vm_map_entry_delete: [ internal use only ] 2274 * 2275 * Deallocate the given entry from the target map. 2276 */ 2277 static void 2278 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2279 { 2280 vm_object_t object; 2281 vm_pindex_t offidxstart, offidxend, count; 2282 2283 vm_map_entry_unlink(map, entry); 2284 map->size -= entry->end - entry->start; 2285 2286 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2287 (object = entry->object.vm_object) != NULL) { 2288 count = OFF_TO_IDX(entry->end - entry->start); 2289 offidxstart = OFF_TO_IDX(entry->offset); 2290 offidxend = offidxstart + count; 2291 VM_OBJECT_LOCK(object); 2292 if (object->ref_count != 1 && 2293 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2294 object == kernel_object || object == kmem_object)) { 2295 vm_object_collapse(object); 2296 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2297 if (object->type == OBJT_SWAP) 2298 swap_pager_freespace(object, offidxstart, count); 2299 if (offidxend >= object->size && 2300 offidxstart < object->size) 2301 object->size = offidxstart; 2302 } 2303 VM_OBJECT_UNLOCK(object); 2304 vm_object_deallocate(object); 2305 } 2306 2307 vm_map_entry_dispose(map, entry); 2308 } 2309 2310 /* 2311 * vm_map_delete: [ internal use only ] 2312 * 2313 * Deallocates the given address range from the target 2314 * map. 2315 */ 2316 int 2317 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2318 { 2319 vm_map_entry_t entry; 2320 vm_map_entry_t first_entry; 2321 2322 /* 2323 * Find the start of the region, and clip it 2324 */ 2325 if (!vm_map_lookup_entry(map, start, &first_entry)) 2326 entry = first_entry->next; 2327 else { 2328 entry = first_entry; 2329 vm_map_clip_start(map, entry, start); 2330 } 2331 2332 /* 2333 * Step through all entries in this region 2334 */ 2335 while ((entry != &map->header) && (entry->start < end)) { 2336 vm_map_entry_t next; 2337 2338 /* 2339 * Wait for wiring or unwiring of an entry to complete. 2340 * Also wait for any system wirings to disappear on 2341 * user maps. 2342 */ 2343 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2344 (vm_map_pmap(map) != kernel_pmap && 2345 vm_map_entry_system_wired_count(entry) != 0)) { 2346 unsigned int last_timestamp; 2347 vm_offset_t saved_start; 2348 vm_map_entry_t tmp_entry; 2349 2350 saved_start = entry->start; 2351 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2352 last_timestamp = map->timestamp; 2353 (void) vm_map_unlock_and_wait(map, FALSE); 2354 vm_map_lock(map); 2355 if (last_timestamp + 1 != map->timestamp) { 2356 /* 2357 * Look again for the entry because the map was 2358 * modified while it was unlocked. 2359 * Specifically, the entry may have been 2360 * clipped, merged, or deleted. 2361 */ 2362 if (!vm_map_lookup_entry(map, saved_start, 2363 &tmp_entry)) 2364 entry = tmp_entry->next; 2365 else { 2366 entry = tmp_entry; 2367 vm_map_clip_start(map, entry, 2368 saved_start); 2369 } 2370 } 2371 continue; 2372 } 2373 vm_map_clip_end(map, entry, end); 2374 2375 next = entry->next; 2376 2377 /* 2378 * Unwire before removing addresses from the pmap; otherwise, 2379 * unwiring will put the entries back in the pmap. 2380 */ 2381 if (entry->wired_count != 0) { 2382 vm_map_entry_unwire(map, entry); 2383 } 2384 2385 pmap_remove(map->pmap, entry->start, entry->end); 2386 2387 /* 2388 * Delete the entry (which may delete the object) only after 2389 * removing all pmap entries pointing to its pages. 2390 * (Otherwise, its page frames may be reallocated, and any 2391 * modify bits will be set in the wrong object!) 2392 */ 2393 vm_map_entry_delete(map, entry); 2394 entry = next; 2395 } 2396 return (KERN_SUCCESS); 2397 } 2398 2399 /* 2400 * vm_map_remove: 2401 * 2402 * Remove the given address range from the target map. 2403 * This is the exported form of vm_map_delete. 2404 */ 2405 int 2406 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2407 { 2408 int result; 2409 2410 vm_map_lock(map); 2411 VM_MAP_RANGE_CHECK(map, start, end); 2412 result = vm_map_delete(map, start, end); 2413 vm_map_unlock(map); 2414 return (result); 2415 } 2416 2417 /* 2418 * vm_map_check_protection: 2419 * 2420 * Assert that the target map allows the specified privilege on the 2421 * entire address region given. The entire region must be allocated. 2422 * 2423 * WARNING! This code does not and should not check whether the 2424 * contents of the region is accessible. For example a smaller file 2425 * might be mapped into a larger address space. 2426 * 2427 * NOTE! This code is also called by munmap(). 2428 * 2429 * The map must be locked. A read lock is sufficient. 2430 */ 2431 boolean_t 2432 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2433 vm_prot_t protection) 2434 { 2435 vm_map_entry_t entry; 2436 vm_map_entry_t tmp_entry; 2437 2438 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2439 return (FALSE); 2440 entry = tmp_entry; 2441 2442 while (start < end) { 2443 if (entry == &map->header) 2444 return (FALSE); 2445 /* 2446 * No holes allowed! 2447 */ 2448 if (start < entry->start) 2449 return (FALSE); 2450 /* 2451 * Check protection associated with entry. 2452 */ 2453 if ((entry->protection & protection) != protection) 2454 return (FALSE); 2455 /* go to next entry */ 2456 start = entry->end; 2457 entry = entry->next; 2458 } 2459 return (TRUE); 2460 } 2461 2462 /* 2463 * vm_map_copy_entry: 2464 * 2465 * Copies the contents of the source entry to the destination 2466 * entry. The entries *must* be aligned properly. 2467 */ 2468 static void 2469 vm_map_copy_entry( 2470 vm_map_t src_map, 2471 vm_map_t dst_map, 2472 vm_map_entry_t src_entry, 2473 vm_map_entry_t dst_entry) 2474 { 2475 vm_object_t src_object; 2476 2477 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2478 return; 2479 2480 if (src_entry->wired_count == 0) { 2481 2482 /* 2483 * If the source entry is marked needs_copy, it is already 2484 * write-protected. 2485 */ 2486 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2487 pmap_protect(src_map->pmap, 2488 src_entry->start, 2489 src_entry->end, 2490 src_entry->protection & ~VM_PROT_WRITE); 2491 } 2492 2493 /* 2494 * Make a copy of the object. 2495 */ 2496 if ((src_object = src_entry->object.vm_object) != NULL) { 2497 VM_OBJECT_LOCK(src_object); 2498 if ((src_object->handle == NULL) && 2499 (src_object->type == OBJT_DEFAULT || 2500 src_object->type == OBJT_SWAP)) { 2501 vm_object_collapse(src_object); 2502 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2503 vm_object_split(src_entry); 2504 src_object = src_entry->object.vm_object; 2505 } 2506 } 2507 vm_object_reference_locked(src_object); 2508 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2509 VM_OBJECT_UNLOCK(src_object); 2510 dst_entry->object.vm_object = src_object; 2511 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2512 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2513 dst_entry->offset = src_entry->offset; 2514 } else { 2515 dst_entry->object.vm_object = NULL; 2516 dst_entry->offset = 0; 2517 } 2518 2519 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2520 dst_entry->end - dst_entry->start, src_entry->start); 2521 } else { 2522 /* 2523 * Of course, wired down pages can't be set copy-on-write. 2524 * Cause wired pages to be copied into the new map by 2525 * simulating faults (the new pages are pageable) 2526 */ 2527 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2528 } 2529 } 2530 2531 /* 2532 * vmspace_map_entry_forked: 2533 * Update the newly-forked vmspace each time a map entry is inherited 2534 * or copied. The values for vm_dsize and vm_tsize are approximate 2535 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2536 */ 2537 static void 2538 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2539 vm_map_entry_t entry) 2540 { 2541 vm_size_t entrysize; 2542 vm_offset_t newend; 2543 2544 entrysize = entry->end - entry->start; 2545 vm2->vm_map.size += entrysize; 2546 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2547 vm2->vm_ssize += btoc(entrysize); 2548 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2549 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2550 newend = MIN(entry->end, 2551 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2552 vm2->vm_dsize += btoc(newend - entry->start); 2553 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2554 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2555 newend = MIN(entry->end, 2556 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2557 vm2->vm_tsize += btoc(newend - entry->start); 2558 } 2559 } 2560 2561 /* 2562 * vmspace_fork: 2563 * Create a new process vmspace structure and vm_map 2564 * based on those of an existing process. The new map 2565 * is based on the old map, according to the inheritance 2566 * values on the regions in that map. 2567 * 2568 * XXX It might be worth coalescing the entries added to the new vmspace. 2569 * 2570 * The source map must not be locked. 2571 */ 2572 struct vmspace * 2573 vmspace_fork(struct vmspace *vm1) 2574 { 2575 struct vmspace *vm2; 2576 vm_map_t old_map = &vm1->vm_map; 2577 vm_map_t new_map; 2578 vm_map_entry_t old_entry; 2579 vm_map_entry_t new_entry; 2580 vm_object_t object; 2581 2582 vm_map_lock(old_map); 2583 2584 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2585 vm2->vm_taddr = vm1->vm_taddr; 2586 vm2->vm_daddr = vm1->vm_daddr; 2587 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2588 new_map = &vm2->vm_map; /* XXX */ 2589 new_map->timestamp = 1; 2590 2591 old_entry = old_map->header.next; 2592 2593 while (old_entry != &old_map->header) { 2594 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2595 panic("vm_map_fork: encountered a submap"); 2596 2597 switch (old_entry->inheritance) { 2598 case VM_INHERIT_NONE: 2599 break; 2600 2601 case VM_INHERIT_SHARE: 2602 /* 2603 * Clone the entry, creating the shared object if necessary. 2604 */ 2605 object = old_entry->object.vm_object; 2606 if (object == NULL) { 2607 object = vm_object_allocate(OBJT_DEFAULT, 2608 atop(old_entry->end - old_entry->start)); 2609 old_entry->object.vm_object = object; 2610 old_entry->offset = 0; 2611 } 2612 2613 /* 2614 * Add the reference before calling vm_object_shadow 2615 * to insure that a shadow object is created. 2616 */ 2617 vm_object_reference(object); 2618 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 2619 vm_object_shadow(&old_entry->object.vm_object, 2620 &old_entry->offset, 2621 atop(old_entry->end - old_entry->start)); 2622 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 2623 /* Transfer the second reference too. */ 2624 vm_object_reference( 2625 old_entry->object.vm_object); 2626 vm_object_deallocate(object); 2627 object = old_entry->object.vm_object; 2628 } 2629 VM_OBJECT_LOCK(object); 2630 vm_object_clear_flag(object, OBJ_ONEMAPPING); 2631 VM_OBJECT_UNLOCK(object); 2632 2633 /* 2634 * Clone the entry, referencing the shared object. 2635 */ 2636 new_entry = vm_map_entry_create(new_map); 2637 *new_entry = *old_entry; 2638 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2639 new_entry->wired_count = 0; 2640 2641 /* 2642 * Insert the entry into the new map -- we know we're 2643 * inserting at the end of the new map. 2644 */ 2645 vm_map_entry_link(new_map, new_map->header.prev, 2646 new_entry); 2647 vmspace_map_entry_forked(vm1, vm2, new_entry); 2648 2649 /* 2650 * Update the physical map 2651 */ 2652 pmap_copy(new_map->pmap, old_map->pmap, 2653 new_entry->start, 2654 (old_entry->end - old_entry->start), 2655 old_entry->start); 2656 break; 2657 2658 case VM_INHERIT_COPY: 2659 /* 2660 * Clone the entry and link into the map. 2661 */ 2662 new_entry = vm_map_entry_create(new_map); 2663 *new_entry = *old_entry; 2664 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2665 new_entry->wired_count = 0; 2666 new_entry->object.vm_object = NULL; 2667 vm_map_entry_link(new_map, new_map->header.prev, 2668 new_entry); 2669 vmspace_map_entry_forked(vm1, vm2, new_entry); 2670 vm_map_copy_entry(old_map, new_map, old_entry, 2671 new_entry); 2672 break; 2673 } 2674 old_entry = old_entry->next; 2675 } 2676 2677 vm_map_unlock(old_map); 2678 2679 return (vm2); 2680 } 2681 2682 int 2683 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 2684 vm_prot_t prot, vm_prot_t max, int cow) 2685 { 2686 vm_map_entry_t new_entry, prev_entry; 2687 vm_offset_t bot, top; 2688 vm_size_t init_ssize; 2689 int orient, rv; 2690 rlim_t vmemlim; 2691 2692 /* 2693 * The stack orientation is piggybacked with the cow argument. 2694 * Extract it into orient and mask the cow argument so that we 2695 * don't pass it around further. 2696 * NOTE: We explicitly allow bi-directional stacks. 2697 */ 2698 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 2699 cow &= ~orient; 2700 KASSERT(orient != 0, ("No stack grow direction")); 2701 2702 if (addrbos < vm_map_min(map) || addrbos > map->max_offset) 2703 return (KERN_NO_SPACE); 2704 2705 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 2706 2707 PROC_LOCK(curthread->td_proc); 2708 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 2709 PROC_UNLOCK(curthread->td_proc); 2710 2711 vm_map_lock(map); 2712 2713 /* If addr is already mapped, no go */ 2714 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 2715 vm_map_unlock(map); 2716 return (KERN_NO_SPACE); 2717 } 2718 2719 /* If we would blow our VMEM resource limit, no go */ 2720 if (map->size + init_ssize > vmemlim) { 2721 vm_map_unlock(map); 2722 return (KERN_NO_SPACE); 2723 } 2724 2725 /* 2726 * If we can't accomodate max_ssize in the current mapping, no go. 2727 * However, we need to be aware that subsequent user mappings might 2728 * map into the space we have reserved for stack, and currently this 2729 * space is not protected. 2730 * 2731 * Hopefully we will at least detect this condition when we try to 2732 * grow the stack. 2733 */ 2734 if ((prev_entry->next != &map->header) && 2735 (prev_entry->next->start < addrbos + max_ssize)) { 2736 vm_map_unlock(map); 2737 return (KERN_NO_SPACE); 2738 } 2739 2740 /* 2741 * We initially map a stack of only init_ssize. We will grow as 2742 * needed later. Depending on the orientation of the stack (i.e. 2743 * the grow direction) we either map at the top of the range, the 2744 * bottom of the range or in the middle. 2745 * 2746 * Note: we would normally expect prot and max to be VM_PROT_ALL, 2747 * and cow to be 0. Possibly we should eliminate these as input 2748 * parameters, and just pass these values here in the insert call. 2749 */ 2750 if (orient == MAP_STACK_GROWS_DOWN) 2751 bot = addrbos + max_ssize - init_ssize; 2752 else if (orient == MAP_STACK_GROWS_UP) 2753 bot = addrbos; 2754 else 2755 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 2756 top = bot + init_ssize; 2757 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 2758 2759 /* Now set the avail_ssize amount. */ 2760 if (rv == KERN_SUCCESS) { 2761 if (prev_entry != &map->header) 2762 vm_map_clip_end(map, prev_entry, bot); 2763 new_entry = prev_entry->next; 2764 if (new_entry->end != top || new_entry->start != bot) 2765 panic("Bad entry start/end for new stack entry"); 2766 2767 new_entry->avail_ssize = max_ssize - init_ssize; 2768 if (orient & MAP_STACK_GROWS_DOWN) 2769 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2770 if (orient & MAP_STACK_GROWS_UP) 2771 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 2772 } 2773 2774 vm_map_unlock(map); 2775 return (rv); 2776 } 2777 2778 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 2779 * desired address is already mapped, or if we successfully grow 2780 * the stack. Also returns KERN_SUCCESS if addr is outside the 2781 * stack range (this is strange, but preserves compatibility with 2782 * the grow function in vm_machdep.c). 2783 */ 2784 int 2785 vm_map_growstack(struct proc *p, vm_offset_t addr) 2786 { 2787 vm_map_entry_t next_entry, prev_entry; 2788 vm_map_entry_t new_entry, stack_entry; 2789 struct vmspace *vm = p->p_vmspace; 2790 vm_map_t map = &vm->vm_map; 2791 vm_offset_t end; 2792 size_t grow_amount, max_grow; 2793 rlim_t stacklim, vmemlim; 2794 int is_procstack, rv; 2795 2796 Retry: 2797 PROC_LOCK(p); 2798 stacklim = lim_cur(p, RLIMIT_STACK); 2799 vmemlim = lim_cur(p, RLIMIT_VMEM); 2800 PROC_UNLOCK(p); 2801 2802 vm_map_lock_read(map); 2803 2804 /* If addr is already in the entry range, no need to grow.*/ 2805 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 2806 vm_map_unlock_read(map); 2807 return (KERN_SUCCESS); 2808 } 2809 2810 next_entry = prev_entry->next; 2811 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 2812 /* 2813 * This entry does not grow upwards. Since the address lies 2814 * beyond this entry, the next entry (if one exists) has to 2815 * be a downward growable entry. The entry list header is 2816 * never a growable entry, so it suffices to check the flags. 2817 */ 2818 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 2819 vm_map_unlock_read(map); 2820 return (KERN_SUCCESS); 2821 } 2822 stack_entry = next_entry; 2823 } else { 2824 /* 2825 * This entry grows upward. If the next entry does not at 2826 * least grow downwards, this is the entry we need to grow. 2827 * otherwise we have two possible choices and we have to 2828 * select one. 2829 */ 2830 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 2831 /* 2832 * We have two choices; grow the entry closest to 2833 * the address to minimize the amount of growth. 2834 */ 2835 if (addr - prev_entry->end <= next_entry->start - addr) 2836 stack_entry = prev_entry; 2837 else 2838 stack_entry = next_entry; 2839 } else 2840 stack_entry = prev_entry; 2841 } 2842 2843 if (stack_entry == next_entry) { 2844 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 2845 KASSERT(addr < stack_entry->start, ("foo")); 2846 end = (prev_entry != &map->header) ? prev_entry->end : 2847 stack_entry->start - stack_entry->avail_ssize; 2848 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 2849 max_grow = stack_entry->start - end; 2850 } else { 2851 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 2852 KASSERT(addr >= stack_entry->end, ("foo")); 2853 end = (next_entry != &map->header) ? next_entry->start : 2854 stack_entry->end + stack_entry->avail_ssize; 2855 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 2856 max_grow = end - stack_entry->end; 2857 } 2858 2859 if (grow_amount > stack_entry->avail_ssize) { 2860 vm_map_unlock_read(map); 2861 return (KERN_NO_SPACE); 2862 } 2863 2864 /* 2865 * If there is no longer enough space between the entries nogo, and 2866 * adjust the available space. Note: this should only happen if the 2867 * user has mapped into the stack area after the stack was created, 2868 * and is probably an error. 2869 * 2870 * This also effectively destroys any guard page the user might have 2871 * intended by limiting the stack size. 2872 */ 2873 if (grow_amount > max_grow) { 2874 if (vm_map_lock_upgrade(map)) 2875 goto Retry; 2876 2877 stack_entry->avail_ssize = max_grow; 2878 2879 vm_map_unlock(map); 2880 return (KERN_NO_SPACE); 2881 } 2882 2883 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 2884 2885 /* 2886 * If this is the main process stack, see if we're over the stack 2887 * limit. 2888 */ 2889 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2890 vm_map_unlock_read(map); 2891 return (KERN_NO_SPACE); 2892 } 2893 2894 /* Round up the grow amount modulo SGROWSIZ */ 2895 grow_amount = roundup (grow_amount, sgrowsiz); 2896 if (grow_amount > stack_entry->avail_ssize) 2897 grow_amount = stack_entry->avail_ssize; 2898 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2899 grow_amount = stacklim - ctob(vm->vm_ssize); 2900 } 2901 2902 /* If we would blow our VMEM resource limit, no go */ 2903 if (map->size + grow_amount > vmemlim) { 2904 vm_map_unlock_read(map); 2905 return (KERN_NO_SPACE); 2906 } 2907 2908 if (vm_map_lock_upgrade(map)) 2909 goto Retry; 2910 2911 if (stack_entry == next_entry) { 2912 /* 2913 * Growing downward. 2914 */ 2915 /* Get the preliminary new entry start value */ 2916 addr = stack_entry->start - grow_amount; 2917 2918 /* 2919 * If this puts us into the previous entry, cut back our 2920 * growth to the available space. Also, see the note above. 2921 */ 2922 if (addr < end) { 2923 stack_entry->avail_ssize = max_grow; 2924 addr = end; 2925 } 2926 2927 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 2928 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 2929 2930 /* Adjust the available stack space by the amount we grew. */ 2931 if (rv == KERN_SUCCESS) { 2932 if (prev_entry != &map->header) 2933 vm_map_clip_end(map, prev_entry, addr); 2934 new_entry = prev_entry->next; 2935 KASSERT(new_entry == stack_entry->prev, ("foo")); 2936 KASSERT(new_entry->end == stack_entry->start, ("foo")); 2937 KASSERT(new_entry->start == addr, ("foo")); 2938 grow_amount = new_entry->end - new_entry->start; 2939 new_entry->avail_ssize = stack_entry->avail_ssize - 2940 grow_amount; 2941 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 2942 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2943 } 2944 } else { 2945 /* 2946 * Growing upward. 2947 */ 2948 addr = stack_entry->end + grow_amount; 2949 2950 /* 2951 * If this puts us into the next entry, cut back our growth 2952 * to the available space. Also, see the note above. 2953 */ 2954 if (addr > end) { 2955 stack_entry->avail_ssize = end - stack_entry->end; 2956 addr = end; 2957 } 2958 2959 grow_amount = addr - stack_entry->end; 2960 2961 /* Grow the underlying object if applicable. */ 2962 if (stack_entry->object.vm_object == NULL || 2963 vm_object_coalesce(stack_entry->object.vm_object, 2964 stack_entry->offset, 2965 (vm_size_t)(stack_entry->end - stack_entry->start), 2966 (vm_size_t)grow_amount)) { 2967 map->size += (addr - stack_entry->end); 2968 /* Update the current entry. */ 2969 stack_entry->end = addr; 2970 stack_entry->avail_ssize -= grow_amount; 2971 vm_map_entry_resize_free(map, stack_entry); 2972 rv = KERN_SUCCESS; 2973 2974 if (next_entry != &map->header) 2975 vm_map_clip_start(map, next_entry, addr); 2976 } else 2977 rv = KERN_FAILURE; 2978 } 2979 2980 if (rv == KERN_SUCCESS && is_procstack) 2981 vm->vm_ssize += btoc(grow_amount); 2982 2983 vm_map_unlock(map); 2984 2985 /* 2986 * Heed the MAP_WIREFUTURE flag if it was set for this process. 2987 */ 2988 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 2989 vm_map_wire(map, 2990 (stack_entry == next_entry) ? addr : addr - grow_amount, 2991 (stack_entry == next_entry) ? stack_entry->start : addr, 2992 (p->p_flag & P_SYSTEM) 2993 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 2994 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 2995 } 2996 2997 return (rv); 2998 } 2999 3000 /* 3001 * Unshare the specified VM space for exec. If other processes are 3002 * mapped to it, then create a new one. The new vmspace is null. 3003 */ 3004 void 3005 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3006 { 3007 struct vmspace *oldvmspace = p->p_vmspace; 3008 struct vmspace *newvmspace; 3009 3010 newvmspace = vmspace_alloc(minuser, maxuser); 3011 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3012 /* 3013 * This code is written like this for prototype purposes. The 3014 * goal is to avoid running down the vmspace here, but let the 3015 * other process's that are still using the vmspace to finally 3016 * run it down. Even though there is little or no chance of blocking 3017 * here, it is a good idea to keep this form for future mods. 3018 */ 3019 PROC_VMSPACE_LOCK(p); 3020 p->p_vmspace = newvmspace; 3021 PROC_VMSPACE_UNLOCK(p); 3022 if (p == curthread->td_proc) /* XXXKSE ? */ 3023 pmap_activate(curthread); 3024 vmspace_free(oldvmspace); 3025 } 3026 3027 /* 3028 * Unshare the specified VM space for forcing COW. This 3029 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3030 */ 3031 void 3032 vmspace_unshare(struct proc *p) 3033 { 3034 struct vmspace *oldvmspace = p->p_vmspace; 3035 struct vmspace *newvmspace; 3036 3037 if (oldvmspace->vm_refcnt == 1) 3038 return; 3039 newvmspace = vmspace_fork(oldvmspace); 3040 PROC_VMSPACE_LOCK(p); 3041 p->p_vmspace = newvmspace; 3042 PROC_VMSPACE_UNLOCK(p); 3043 if (p == curthread->td_proc) /* XXXKSE ? */ 3044 pmap_activate(curthread); 3045 vmspace_free(oldvmspace); 3046 } 3047 3048 /* 3049 * vm_map_lookup: 3050 * 3051 * Finds the VM object, offset, and 3052 * protection for a given virtual address in the 3053 * specified map, assuming a page fault of the 3054 * type specified. 3055 * 3056 * Leaves the map in question locked for read; return 3057 * values are guaranteed until a vm_map_lookup_done 3058 * call is performed. Note that the map argument 3059 * is in/out; the returned map must be used in 3060 * the call to vm_map_lookup_done. 3061 * 3062 * A handle (out_entry) is returned for use in 3063 * vm_map_lookup_done, to make that fast. 3064 * 3065 * If a lookup is requested with "write protection" 3066 * specified, the map may be changed to perform virtual 3067 * copying operations, although the data referenced will 3068 * remain the same. 3069 */ 3070 int 3071 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3072 vm_offset_t vaddr, 3073 vm_prot_t fault_typea, 3074 vm_map_entry_t *out_entry, /* OUT */ 3075 vm_object_t *object, /* OUT */ 3076 vm_pindex_t *pindex, /* OUT */ 3077 vm_prot_t *out_prot, /* OUT */ 3078 boolean_t *wired) /* OUT */ 3079 { 3080 vm_map_entry_t entry; 3081 vm_map_t map = *var_map; 3082 vm_prot_t prot; 3083 vm_prot_t fault_type = fault_typea; 3084 3085 RetryLookup:; 3086 /* 3087 * Lookup the faulting address. 3088 */ 3089 3090 vm_map_lock_read(map); 3091 #define RETURN(why) \ 3092 { \ 3093 vm_map_unlock_read(map); \ 3094 return (why); \ 3095 } 3096 3097 /* 3098 * If the map has an interesting hint, try it before calling full 3099 * blown lookup routine. 3100 */ 3101 entry = map->root; 3102 *out_entry = entry; 3103 if (entry == NULL || 3104 (vaddr < entry->start) || (vaddr >= entry->end)) { 3105 /* 3106 * Entry was either not a valid hint, or the vaddr was not 3107 * contained in the entry, so do a full lookup. 3108 */ 3109 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3110 RETURN(KERN_INVALID_ADDRESS); 3111 3112 entry = *out_entry; 3113 } 3114 3115 /* 3116 * Handle submaps. 3117 */ 3118 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3119 vm_map_t old_map = map; 3120 3121 *var_map = map = entry->object.sub_map; 3122 vm_map_unlock_read(old_map); 3123 goto RetryLookup; 3124 } 3125 3126 /* 3127 * Check whether this task is allowed to have this page. 3128 * Note the special case for MAP_ENTRY_COW 3129 * pages with an override. This is to implement a forced 3130 * COW for debuggers. 3131 */ 3132 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3133 prot = entry->max_protection; 3134 else 3135 prot = entry->protection; 3136 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3137 if ((fault_type & prot) != fault_type) { 3138 RETURN(KERN_PROTECTION_FAILURE); 3139 } 3140 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3141 (entry->eflags & MAP_ENTRY_COW) && 3142 (fault_type & VM_PROT_WRITE) && 3143 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3144 RETURN(KERN_PROTECTION_FAILURE); 3145 } 3146 3147 /* 3148 * If this page is not pageable, we have to get it for all possible 3149 * accesses. 3150 */ 3151 *wired = (entry->wired_count != 0); 3152 if (*wired) 3153 prot = fault_type = entry->protection; 3154 3155 /* 3156 * If the entry was copy-on-write, we either ... 3157 */ 3158 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3159 /* 3160 * If we want to write the page, we may as well handle that 3161 * now since we've got the map locked. 3162 * 3163 * If we don't need to write the page, we just demote the 3164 * permissions allowed. 3165 */ 3166 if (fault_type & VM_PROT_WRITE) { 3167 /* 3168 * Make a new object, and place it in the object 3169 * chain. Note that no new references have appeared 3170 * -- one just moved from the map to the new 3171 * object. 3172 */ 3173 if (vm_map_lock_upgrade(map)) 3174 goto RetryLookup; 3175 3176 vm_object_shadow( 3177 &entry->object.vm_object, 3178 &entry->offset, 3179 atop(entry->end - entry->start)); 3180 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3181 3182 vm_map_lock_downgrade(map); 3183 } else { 3184 /* 3185 * We're attempting to read a copy-on-write page -- 3186 * don't allow writes. 3187 */ 3188 prot &= ~VM_PROT_WRITE; 3189 } 3190 } 3191 3192 /* 3193 * Create an object if necessary. 3194 */ 3195 if (entry->object.vm_object == NULL && 3196 !map->system_map) { 3197 if (vm_map_lock_upgrade(map)) 3198 goto RetryLookup; 3199 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3200 atop(entry->end - entry->start)); 3201 entry->offset = 0; 3202 vm_map_lock_downgrade(map); 3203 } 3204 3205 /* 3206 * Return the object/offset from this entry. If the entry was 3207 * copy-on-write or empty, it has been fixed up. 3208 */ 3209 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3210 *object = entry->object.vm_object; 3211 3212 *out_prot = prot; 3213 return (KERN_SUCCESS); 3214 3215 #undef RETURN 3216 } 3217 3218 /* 3219 * vm_map_lookup_locked: 3220 * 3221 * Lookup the faulting address. A version of vm_map_lookup that returns 3222 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3223 */ 3224 int 3225 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3226 vm_offset_t vaddr, 3227 vm_prot_t fault_typea, 3228 vm_map_entry_t *out_entry, /* OUT */ 3229 vm_object_t *object, /* OUT */ 3230 vm_pindex_t *pindex, /* OUT */ 3231 vm_prot_t *out_prot, /* OUT */ 3232 boolean_t *wired) /* OUT */ 3233 { 3234 vm_map_entry_t entry; 3235 vm_map_t map = *var_map; 3236 vm_prot_t prot; 3237 vm_prot_t fault_type = fault_typea; 3238 3239 /* 3240 * If the map has an interesting hint, try it before calling full 3241 * blown lookup routine. 3242 */ 3243 entry = map->root; 3244 *out_entry = entry; 3245 if (entry == NULL || 3246 (vaddr < entry->start) || (vaddr >= entry->end)) { 3247 /* 3248 * Entry was either not a valid hint, or the vaddr was not 3249 * contained in the entry, so do a full lookup. 3250 */ 3251 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3252 return (KERN_INVALID_ADDRESS); 3253 3254 entry = *out_entry; 3255 } 3256 3257 /* 3258 * Fail if the entry refers to a submap. 3259 */ 3260 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3261 return (KERN_FAILURE); 3262 3263 /* 3264 * Check whether this task is allowed to have this page. 3265 * Note the special case for MAP_ENTRY_COW 3266 * pages with an override. This is to implement a forced 3267 * COW for debuggers. 3268 */ 3269 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3270 prot = entry->max_protection; 3271 else 3272 prot = entry->protection; 3273 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3274 if ((fault_type & prot) != fault_type) 3275 return (KERN_PROTECTION_FAILURE); 3276 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3277 (entry->eflags & MAP_ENTRY_COW) && 3278 (fault_type & VM_PROT_WRITE) && 3279 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3280 return (KERN_PROTECTION_FAILURE); 3281 3282 /* 3283 * If this page is not pageable, we have to get it for all possible 3284 * accesses. 3285 */ 3286 *wired = (entry->wired_count != 0); 3287 if (*wired) 3288 prot = fault_type = entry->protection; 3289 3290 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3291 /* 3292 * Fail if the entry was copy-on-write for a write fault. 3293 */ 3294 if (fault_type & VM_PROT_WRITE) 3295 return (KERN_FAILURE); 3296 /* 3297 * We're attempting to read a copy-on-write page -- 3298 * don't allow writes. 3299 */ 3300 prot &= ~VM_PROT_WRITE; 3301 } 3302 3303 /* 3304 * Fail if an object should be created. 3305 */ 3306 if (entry->object.vm_object == NULL && !map->system_map) 3307 return (KERN_FAILURE); 3308 3309 /* 3310 * Return the object/offset from this entry. If the entry was 3311 * copy-on-write or empty, it has been fixed up. 3312 */ 3313 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3314 *object = entry->object.vm_object; 3315 3316 *out_prot = prot; 3317 return (KERN_SUCCESS); 3318 } 3319 3320 /* 3321 * vm_map_lookup_done: 3322 * 3323 * Releases locks acquired by a vm_map_lookup 3324 * (according to the handle returned by that lookup). 3325 */ 3326 void 3327 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3328 { 3329 /* 3330 * Unlock the main-level map 3331 */ 3332 vm_map_unlock_read(map); 3333 } 3334 3335 #include "opt_ddb.h" 3336 #ifdef DDB 3337 #include <sys/kernel.h> 3338 3339 #include <ddb/ddb.h> 3340 3341 /* 3342 * vm_map_print: [ debug ] 3343 */ 3344 DB_SHOW_COMMAND(map, vm_map_print) 3345 { 3346 static int nlines; 3347 /* XXX convert args. */ 3348 vm_map_t map = (vm_map_t)addr; 3349 boolean_t full = have_addr; 3350 3351 vm_map_entry_t entry; 3352 3353 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3354 (void *)map, 3355 (void *)map->pmap, map->nentries, map->timestamp); 3356 nlines++; 3357 3358 if (!full && db_indent) 3359 return; 3360 3361 db_indent += 2; 3362 for (entry = map->header.next; entry != &map->header; 3363 entry = entry->next) { 3364 db_iprintf("map entry %p: start=%p, end=%p\n", 3365 (void *)entry, (void *)entry->start, (void *)entry->end); 3366 nlines++; 3367 { 3368 static char *inheritance_name[4] = 3369 {"share", "copy", "none", "donate_copy"}; 3370 3371 db_iprintf(" prot=%x/%x/%s", 3372 entry->protection, 3373 entry->max_protection, 3374 inheritance_name[(int)(unsigned char)entry->inheritance]); 3375 if (entry->wired_count != 0) 3376 db_printf(", wired"); 3377 } 3378 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3379 db_printf(", share=%p, offset=0x%jx\n", 3380 (void *)entry->object.sub_map, 3381 (uintmax_t)entry->offset); 3382 nlines++; 3383 if ((entry->prev == &map->header) || 3384 (entry->prev->object.sub_map != 3385 entry->object.sub_map)) { 3386 db_indent += 2; 3387 vm_map_print((db_expr_t)(intptr_t) 3388 entry->object.sub_map, 3389 full, 0, (char *)0); 3390 db_indent -= 2; 3391 } 3392 } else { 3393 db_printf(", object=%p, offset=0x%jx", 3394 (void *)entry->object.vm_object, 3395 (uintmax_t)entry->offset); 3396 if (entry->eflags & MAP_ENTRY_COW) 3397 db_printf(", copy (%s)", 3398 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3399 db_printf("\n"); 3400 nlines++; 3401 3402 if ((entry->prev == &map->header) || 3403 (entry->prev->object.vm_object != 3404 entry->object.vm_object)) { 3405 db_indent += 2; 3406 vm_object_print((db_expr_t)(intptr_t) 3407 entry->object.vm_object, 3408 full, 0, (char *)0); 3409 nlines += 4; 3410 db_indent -= 2; 3411 } 3412 } 3413 } 3414 db_indent -= 2; 3415 if (db_indent == 0) 3416 nlines = 0; 3417 } 3418 3419 3420 DB_SHOW_COMMAND(procvm, procvm) 3421 { 3422 struct proc *p; 3423 3424 if (have_addr) { 3425 p = (struct proc *) addr; 3426 } else { 3427 p = curproc; 3428 } 3429 3430 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3431 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3432 (void *)vmspace_pmap(p->p_vmspace)); 3433 3434 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3435 } 3436 3437 #endif /* DDB */ 3438