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 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 201 } 202 203 static int 204 vmspace_zinit(void *mem, int size, int flags) 205 { 206 struct vmspace *vm; 207 208 vm = (struct vmspace *)mem; 209 210 vm->vm_map.pmap = NULL; 211 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 212 return (0); 213 } 214 215 static void 216 vm_map_zfini(void *mem, int size) 217 { 218 vm_map_t map; 219 220 map = (vm_map_t)mem; 221 mtx_destroy(&map->system_mtx); 222 sx_destroy(&map->lock); 223 } 224 225 static int 226 vm_map_zinit(void *mem, int size, int flags) 227 { 228 vm_map_t map; 229 230 map = (vm_map_t)mem; 231 map->nentries = 0; 232 map->size = 0; 233 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 234 sx_init(&map->lock, "user map"); 235 return (0); 236 } 237 238 #ifdef INVARIANTS 239 static void 240 vmspace_zdtor(void *mem, int size, void *arg) 241 { 242 struct vmspace *vm; 243 244 vm = (struct vmspace *)mem; 245 246 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 247 } 248 static void 249 vm_map_zdtor(void *mem, int size, void *arg) 250 { 251 vm_map_t map; 252 253 map = (vm_map_t)mem; 254 KASSERT(map->nentries == 0, 255 ("map %p nentries == %d on free.", 256 map, map->nentries)); 257 KASSERT(map->size == 0, 258 ("map %p size == %lu on free.", 259 map, (unsigned long)map->size)); 260 } 261 #endif /* INVARIANTS */ 262 263 /* 264 * Allocate a vmspace structure, including a vm_map and pmap, 265 * and initialize those structures. The refcnt is set to 1. 266 */ 267 struct vmspace * 268 vmspace_alloc(min, max) 269 vm_offset_t min, max; 270 { 271 struct vmspace *vm; 272 273 vm = uma_zalloc(vmspace_zone, M_WAITOK); 274 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) { 275 uma_zfree(vmspace_zone, vm); 276 return (NULL); 277 } 278 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 279 _vm_map_init(&vm->vm_map, min, max); 280 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ 281 vm->vm_refcnt = 1; 282 vm->vm_shm = NULL; 283 vm->vm_swrss = 0; 284 vm->vm_tsize = 0; 285 vm->vm_dsize = 0; 286 vm->vm_ssize = 0; 287 vm->vm_taddr = 0; 288 vm->vm_daddr = 0; 289 vm->vm_maxsaddr = 0; 290 return (vm); 291 } 292 293 void 294 vm_init2(void) 295 { 296 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 297 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 + 298 maxproc * 2 + maxfiles); 299 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 300 #ifdef INVARIANTS 301 vmspace_zdtor, 302 #else 303 NULL, 304 #endif 305 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 306 } 307 308 static inline void 309 vmspace_dofree(struct vmspace *vm) 310 { 311 CTR1(KTR_VM, "vmspace_free: %p", vm); 312 313 /* 314 * Make sure any SysV shm is freed, it might not have been in 315 * exit1(). 316 */ 317 shmexit(vm); 318 319 /* 320 * Lock the map, to wait out all other references to it. 321 * Delete all of the mappings and pages they hold, then call 322 * the pmap module to reclaim anything left. 323 */ 324 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 325 vm->vm_map.max_offset); 326 327 /* 328 * XXX Comment out the pmap_release call for now. The 329 * vmspace_zone is marked as UMA_ZONE_NOFREE, and bugs cause 330 * pmap.resident_count to be != 0 on exit sometimes. 331 */ 332 /* pmap_release(vmspace_pmap(vm)); */ 333 uma_zfree(vmspace_zone, vm); 334 } 335 336 void 337 vmspace_free(struct vmspace *vm) 338 { 339 int refcnt; 340 341 if (vm->vm_refcnt == 0) 342 panic("vmspace_free: attempt to free already freed vmspace"); 343 344 do 345 refcnt = vm->vm_refcnt; 346 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 347 if (refcnt == 1) 348 vmspace_dofree(vm); 349 } 350 351 void 352 vmspace_exitfree(struct proc *p) 353 { 354 struct vmspace *vm; 355 356 PROC_VMSPACE_LOCK(p); 357 vm = p->p_vmspace; 358 p->p_vmspace = NULL; 359 PROC_VMSPACE_UNLOCK(p); 360 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 361 vmspace_free(vm); 362 } 363 364 void 365 vmspace_exit(struct thread *td) 366 { 367 int refcnt; 368 struct vmspace *vm; 369 struct proc *p; 370 371 /* 372 * Release user portion of address space. 373 * This releases references to vnodes, 374 * which could cause I/O if the file has been unlinked. 375 * Need to do this early enough that we can still sleep. 376 * 377 * The last exiting process to reach this point releases as 378 * much of the environment as it can. vmspace_dofree() is the 379 * slower fallback in case another process had a temporary 380 * reference to the vmspace. 381 */ 382 383 p = td->td_proc; 384 vm = p->p_vmspace; 385 atomic_add_int(&vmspace0.vm_refcnt, 1); 386 do { 387 refcnt = vm->vm_refcnt; 388 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 389 /* Switch now since other proc might free vmspace */ 390 PROC_VMSPACE_LOCK(p); 391 p->p_vmspace = &vmspace0; 392 PROC_VMSPACE_UNLOCK(p); 393 pmap_activate(td); 394 } 395 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 396 if (refcnt == 1) { 397 if (p->p_vmspace != vm) { 398 /* vmspace not yet freed, switch back */ 399 PROC_VMSPACE_LOCK(p); 400 p->p_vmspace = vm; 401 PROC_VMSPACE_UNLOCK(p); 402 pmap_activate(td); 403 } 404 pmap_remove_pages(vmspace_pmap(vm)); 405 /* Switch now since this proc will free vmspace */ 406 PROC_VMSPACE_LOCK(p); 407 p->p_vmspace = &vmspace0; 408 PROC_VMSPACE_UNLOCK(p); 409 pmap_activate(td); 410 vmspace_dofree(vm); 411 } 412 } 413 414 /* Acquire reference to vmspace owned by another process. */ 415 416 struct vmspace * 417 vmspace_acquire_ref(struct proc *p) 418 { 419 struct vmspace *vm; 420 int refcnt; 421 422 PROC_VMSPACE_LOCK(p); 423 vm = p->p_vmspace; 424 if (vm == NULL) { 425 PROC_VMSPACE_UNLOCK(p); 426 return (NULL); 427 } 428 do { 429 refcnt = vm->vm_refcnt; 430 if (refcnt <= 0) { /* Avoid 0->1 transition */ 431 PROC_VMSPACE_UNLOCK(p); 432 return (NULL); 433 } 434 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 435 if (vm != p->p_vmspace) { 436 PROC_VMSPACE_UNLOCK(p); 437 vmspace_free(vm); 438 return (NULL); 439 } 440 PROC_VMSPACE_UNLOCK(p); 441 return (vm); 442 } 443 444 void 445 _vm_map_lock(vm_map_t map, const char *file, int line) 446 { 447 448 if (map->system_map) 449 _mtx_lock_flags(&map->system_mtx, 0, file, line); 450 else 451 (void)_sx_xlock(&map->lock, 0, file, line); 452 map->timestamp++; 453 } 454 455 void 456 _vm_map_unlock(vm_map_t map, const char *file, int line) 457 { 458 459 if (map->system_map) 460 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 461 else 462 _sx_xunlock(&map->lock, file, line); 463 } 464 465 void 466 _vm_map_lock_read(vm_map_t map, const char *file, int line) 467 { 468 469 if (map->system_map) 470 _mtx_lock_flags(&map->system_mtx, 0, file, line); 471 else 472 (void)_sx_xlock(&map->lock, 0, file, line); 473 } 474 475 void 476 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 477 { 478 479 if (map->system_map) 480 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 481 else 482 _sx_xunlock(&map->lock, file, line); 483 } 484 485 int 486 _vm_map_trylock(vm_map_t map, const char *file, int line) 487 { 488 int error; 489 490 error = map->system_map ? 491 !_mtx_trylock(&map->system_mtx, 0, file, line) : 492 !_sx_try_xlock(&map->lock, file, line); 493 if (error == 0) 494 map->timestamp++; 495 return (error == 0); 496 } 497 498 int 499 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 500 { 501 int error; 502 503 error = map->system_map ? 504 !_mtx_trylock(&map->system_mtx, 0, file, line) : 505 !_sx_try_xlock(&map->lock, file, line); 506 return (error == 0); 507 } 508 509 int 510 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 511 { 512 513 #ifdef INVARIANTS 514 if (map->system_map) { 515 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 516 } else 517 _sx_assert(&map->lock, SX_XLOCKED, file, line); 518 #endif 519 map->timestamp++; 520 return (0); 521 } 522 523 void 524 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 525 { 526 527 #ifdef INVARIANTS 528 if (map->system_map) { 529 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 530 } else 531 _sx_assert(&map->lock, SX_XLOCKED, file, line); 532 #endif 533 } 534 535 /* 536 * vm_map_unlock_and_wait: 537 */ 538 int 539 vm_map_unlock_and_wait(vm_map_t map, int timo) 540 { 541 542 mtx_lock(&map_sleep_mtx); 543 vm_map_unlock(map); 544 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", timo)); 545 } 546 547 /* 548 * vm_map_wakeup: 549 */ 550 void 551 vm_map_wakeup(vm_map_t map) 552 { 553 554 /* 555 * Acquire and release map_sleep_mtx to prevent a wakeup() 556 * from being performed (and lost) between the vm_map_unlock() 557 * and the msleep() in vm_map_unlock_and_wait(). 558 */ 559 mtx_lock(&map_sleep_mtx); 560 mtx_unlock(&map_sleep_mtx); 561 wakeup(&map->root); 562 } 563 564 long 565 vmspace_resident_count(struct vmspace *vmspace) 566 { 567 return pmap_resident_count(vmspace_pmap(vmspace)); 568 } 569 570 long 571 vmspace_wired_count(struct vmspace *vmspace) 572 { 573 return pmap_wired_count(vmspace_pmap(vmspace)); 574 } 575 576 /* 577 * vm_map_create: 578 * 579 * Creates and returns a new empty VM map with 580 * the given physical map structure, and having 581 * the given lower and upper address bounds. 582 */ 583 vm_map_t 584 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 585 { 586 vm_map_t result; 587 588 result = uma_zalloc(mapzone, M_WAITOK); 589 CTR1(KTR_VM, "vm_map_create: %p", result); 590 _vm_map_init(result, min, max); 591 result->pmap = pmap; 592 return (result); 593 } 594 595 /* 596 * Initialize an existing vm_map structure 597 * such as that in the vmspace structure. 598 * The pmap is set elsewhere. 599 */ 600 static void 601 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 602 { 603 604 map->header.next = map->header.prev = &map->header; 605 map->needs_wakeup = FALSE; 606 map->system_map = 0; 607 map->min_offset = min; 608 map->max_offset = max; 609 map->flags = 0; 610 map->root = NULL; 611 map->timestamp = 0; 612 } 613 614 void 615 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 616 { 617 _vm_map_init(map, min, max); 618 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 619 sx_init(&map->lock, "user map"); 620 } 621 622 /* 623 * vm_map_entry_dispose: [ internal use only ] 624 * 625 * Inverse of vm_map_entry_create. 626 */ 627 static void 628 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 629 { 630 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 631 } 632 633 /* 634 * vm_map_entry_create: [ internal use only ] 635 * 636 * Allocates a VM map entry for insertion. 637 * No entry fields are filled in. 638 */ 639 static vm_map_entry_t 640 vm_map_entry_create(vm_map_t map) 641 { 642 vm_map_entry_t new_entry; 643 644 if (map->system_map) 645 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 646 else 647 new_entry = uma_zalloc(mapentzone, M_WAITOK); 648 if (new_entry == NULL) 649 panic("vm_map_entry_create: kernel resources exhausted"); 650 return (new_entry); 651 } 652 653 /* 654 * vm_map_entry_set_behavior: 655 * 656 * Set the expected access behavior, either normal, random, or 657 * sequential. 658 */ 659 static inline void 660 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 661 { 662 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 663 (behavior & MAP_ENTRY_BEHAV_MASK); 664 } 665 666 /* 667 * vm_map_entry_set_max_free: 668 * 669 * Set the max_free field in a vm_map_entry. 670 */ 671 static inline void 672 vm_map_entry_set_max_free(vm_map_entry_t entry) 673 { 674 675 entry->max_free = entry->adj_free; 676 if (entry->left != NULL && entry->left->max_free > entry->max_free) 677 entry->max_free = entry->left->max_free; 678 if (entry->right != NULL && entry->right->max_free > entry->max_free) 679 entry->max_free = entry->right->max_free; 680 } 681 682 /* 683 * vm_map_entry_splay: 684 * 685 * The Sleator and Tarjan top-down splay algorithm with the 686 * following variation. Max_free must be computed bottom-up, so 687 * on the downward pass, maintain the left and right spines in 688 * reverse order. Then, make a second pass up each side to fix 689 * the pointers and compute max_free. The time bound is O(log n) 690 * amortized. 691 * 692 * The new root is the vm_map_entry containing "addr", or else an 693 * adjacent entry (lower or higher) if addr is not in the tree. 694 * 695 * The map must be locked, and leaves it so. 696 * 697 * Returns: the new root. 698 */ 699 static vm_map_entry_t 700 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 701 { 702 vm_map_entry_t llist, rlist; 703 vm_map_entry_t ltree, rtree; 704 vm_map_entry_t y; 705 706 /* Special case of empty tree. */ 707 if (root == NULL) 708 return (root); 709 710 /* 711 * Pass One: Splay down the tree until we find addr or a NULL 712 * pointer where addr would go. llist and rlist are the two 713 * sides in reverse order (bottom-up), with llist linked by 714 * the right pointer and rlist linked by the left pointer in 715 * the vm_map_entry. Wait until Pass Two to set max_free on 716 * the two spines. 717 */ 718 llist = NULL; 719 rlist = NULL; 720 for (;;) { 721 /* root is never NULL in here. */ 722 if (addr < root->start) { 723 y = root->left; 724 if (y == NULL) 725 break; 726 if (addr < y->start && y->left != NULL) { 727 /* Rotate right and put y on rlist. */ 728 root->left = y->right; 729 y->right = root; 730 vm_map_entry_set_max_free(root); 731 root = y->left; 732 y->left = rlist; 733 rlist = y; 734 } else { 735 /* Put root on rlist. */ 736 root->left = rlist; 737 rlist = root; 738 root = y; 739 } 740 } else { 741 y = root->right; 742 if (addr < root->end || y == NULL) 743 break; 744 if (addr >= y->end && y->right != NULL) { 745 /* Rotate left and put y on llist. */ 746 root->right = y->left; 747 y->left = root; 748 vm_map_entry_set_max_free(root); 749 root = y->right; 750 y->right = llist; 751 llist = y; 752 } else { 753 /* Put root on llist. */ 754 root->right = llist; 755 llist = root; 756 root = y; 757 } 758 } 759 } 760 761 /* 762 * Pass Two: Walk back up the two spines, flip the pointers 763 * and set max_free. The subtrees of the root go at the 764 * bottom of llist and rlist. 765 */ 766 ltree = root->left; 767 while (llist != NULL) { 768 y = llist->right; 769 llist->right = ltree; 770 vm_map_entry_set_max_free(llist); 771 ltree = llist; 772 llist = y; 773 } 774 rtree = root->right; 775 while (rlist != NULL) { 776 y = rlist->left; 777 rlist->left = rtree; 778 vm_map_entry_set_max_free(rlist); 779 rtree = rlist; 780 rlist = y; 781 } 782 783 /* 784 * Final assembly: add ltree and rtree as subtrees of root. 785 */ 786 root->left = ltree; 787 root->right = rtree; 788 vm_map_entry_set_max_free(root); 789 790 return (root); 791 } 792 793 /* 794 * vm_map_entry_{un,}link: 795 * 796 * Insert/remove entries from maps. 797 */ 798 static void 799 vm_map_entry_link(vm_map_t map, 800 vm_map_entry_t after_where, 801 vm_map_entry_t entry) 802 { 803 804 CTR4(KTR_VM, 805 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 806 map->nentries, entry, after_where); 807 map->nentries++; 808 entry->prev = after_where; 809 entry->next = after_where->next; 810 entry->next->prev = entry; 811 after_where->next = entry; 812 813 if (after_where != &map->header) { 814 if (after_where != map->root) 815 vm_map_entry_splay(after_where->start, map->root); 816 entry->right = after_where->right; 817 entry->left = after_where; 818 after_where->right = NULL; 819 after_where->adj_free = entry->start - after_where->end; 820 vm_map_entry_set_max_free(after_where); 821 } else { 822 entry->right = map->root; 823 entry->left = NULL; 824 } 825 entry->adj_free = (entry->next == &map->header ? map->max_offset : 826 entry->next->start) - entry->end; 827 vm_map_entry_set_max_free(entry); 828 map->root = entry; 829 } 830 831 static void 832 vm_map_entry_unlink(vm_map_t map, 833 vm_map_entry_t entry) 834 { 835 vm_map_entry_t next, prev, root; 836 837 if (entry != map->root) 838 vm_map_entry_splay(entry->start, map->root); 839 if (entry->left == NULL) 840 root = entry->right; 841 else { 842 root = vm_map_entry_splay(entry->start, entry->left); 843 root->right = entry->right; 844 root->adj_free = (entry->next == &map->header ? map->max_offset : 845 entry->next->start) - root->end; 846 vm_map_entry_set_max_free(root); 847 } 848 map->root = root; 849 850 prev = entry->prev; 851 next = entry->next; 852 next->prev = prev; 853 prev->next = next; 854 map->nentries--; 855 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 856 map->nentries, entry); 857 } 858 859 /* 860 * vm_map_entry_resize_free: 861 * 862 * Recompute the amount of free space following a vm_map_entry 863 * and propagate that value up the tree. Call this function after 864 * resizing a map entry in-place, that is, without a call to 865 * vm_map_entry_link() or _unlink(). 866 * 867 * The map must be locked, and leaves it so. 868 */ 869 static void 870 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 871 { 872 873 /* 874 * Using splay trees without parent pointers, propagating 875 * max_free up the tree is done by moving the entry to the 876 * root and making the change there. 877 */ 878 if (entry != map->root) 879 map->root = vm_map_entry_splay(entry->start, map->root); 880 881 entry->adj_free = (entry->next == &map->header ? map->max_offset : 882 entry->next->start) - entry->end; 883 vm_map_entry_set_max_free(entry); 884 } 885 886 /* 887 * vm_map_lookup_entry: [ internal use only ] 888 * 889 * Finds the map entry containing (or 890 * immediately preceding) the specified address 891 * in the given map; the entry is returned 892 * in the "entry" parameter. The boolean 893 * result indicates whether the address is 894 * actually contained in the map. 895 */ 896 boolean_t 897 vm_map_lookup_entry( 898 vm_map_t map, 899 vm_offset_t address, 900 vm_map_entry_t *entry) /* OUT */ 901 { 902 vm_map_entry_t cur; 903 904 cur = vm_map_entry_splay(address, map->root); 905 if (cur == NULL) 906 *entry = &map->header; 907 else { 908 map->root = cur; 909 910 if (address >= cur->start) { 911 *entry = cur; 912 if (cur->end > address) 913 return (TRUE); 914 } else 915 *entry = cur->prev; 916 } 917 return (FALSE); 918 } 919 920 /* 921 * vm_map_insert: 922 * 923 * Inserts the given whole VM object into the target 924 * map at the specified address range. The object's 925 * size should match that of the address range. 926 * 927 * Requires that the map be locked, and leaves it so. 928 * 929 * If object is non-NULL, ref count must be bumped by caller 930 * prior to making call to account for the new entry. 931 */ 932 int 933 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 934 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 935 int cow) 936 { 937 vm_map_entry_t new_entry; 938 vm_map_entry_t prev_entry; 939 vm_map_entry_t temp_entry; 940 vm_eflags_t protoeflags; 941 942 /* 943 * Check that the start and end points are not bogus. 944 */ 945 if ((start < map->min_offset) || (end > map->max_offset) || 946 (start >= end)) 947 return (KERN_INVALID_ADDRESS); 948 949 /* 950 * Find the entry prior to the proposed starting address; if it's part 951 * of an existing entry, this range is bogus. 952 */ 953 if (vm_map_lookup_entry(map, start, &temp_entry)) 954 return (KERN_NO_SPACE); 955 956 prev_entry = temp_entry; 957 958 /* 959 * Assert that the next entry doesn't overlap the end point. 960 */ 961 if ((prev_entry->next != &map->header) && 962 (prev_entry->next->start < end)) 963 return (KERN_NO_SPACE); 964 965 protoeflags = 0; 966 967 if (cow & MAP_COPY_ON_WRITE) 968 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 969 970 if (cow & MAP_NOFAULT) { 971 protoeflags |= MAP_ENTRY_NOFAULT; 972 973 KASSERT(object == NULL, 974 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 975 } 976 if (cow & MAP_DISABLE_SYNCER) 977 protoeflags |= MAP_ENTRY_NOSYNC; 978 if (cow & MAP_DISABLE_COREDUMP) 979 protoeflags |= MAP_ENTRY_NOCOREDUMP; 980 981 if (object != NULL) { 982 /* 983 * OBJ_ONEMAPPING must be cleared unless this mapping 984 * is trivially proven to be the only mapping for any 985 * of the object's pages. (Object granularity 986 * reference counting is insufficient to recognize 987 * aliases with precision.) 988 */ 989 VM_OBJECT_LOCK(object); 990 if (object->ref_count > 1 || object->shadow_count != 0) 991 vm_object_clear_flag(object, OBJ_ONEMAPPING); 992 VM_OBJECT_UNLOCK(object); 993 } 994 else if ((prev_entry != &map->header) && 995 (prev_entry->eflags == protoeflags) && 996 (prev_entry->end == start) && 997 (prev_entry->wired_count == 0) && 998 ((prev_entry->object.vm_object == NULL) || 999 vm_object_coalesce(prev_entry->object.vm_object, 1000 prev_entry->offset, 1001 (vm_size_t)(prev_entry->end - prev_entry->start), 1002 (vm_size_t)(end - prev_entry->end)))) { 1003 /* 1004 * We were able to extend the object. Determine if we 1005 * can extend the previous map entry to include the 1006 * new range as well. 1007 */ 1008 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && 1009 (prev_entry->protection == prot) && 1010 (prev_entry->max_protection == max)) { 1011 map->size += (end - prev_entry->end); 1012 prev_entry->end = end; 1013 vm_map_entry_resize_free(map, prev_entry); 1014 vm_map_simplify_entry(map, prev_entry); 1015 return (KERN_SUCCESS); 1016 } 1017 1018 /* 1019 * If we can extend the object but cannot extend the 1020 * map entry, we have to create a new map entry. We 1021 * must bump the ref count on the extended object to 1022 * account for it. object may be NULL. 1023 */ 1024 object = prev_entry->object.vm_object; 1025 offset = prev_entry->offset + 1026 (prev_entry->end - prev_entry->start); 1027 vm_object_reference(object); 1028 } 1029 1030 /* 1031 * NOTE: if conditionals fail, object can be NULL here. This occurs 1032 * in things like the buffer map where we manage kva but do not manage 1033 * backing objects. 1034 */ 1035 1036 /* 1037 * Create a new entry 1038 */ 1039 new_entry = vm_map_entry_create(map); 1040 new_entry->start = start; 1041 new_entry->end = end; 1042 1043 new_entry->eflags = protoeflags; 1044 new_entry->object.vm_object = object; 1045 new_entry->offset = offset; 1046 new_entry->avail_ssize = 0; 1047 1048 new_entry->inheritance = VM_INHERIT_DEFAULT; 1049 new_entry->protection = prot; 1050 new_entry->max_protection = max; 1051 new_entry->wired_count = 0; 1052 1053 /* 1054 * Insert the new entry into the list 1055 */ 1056 vm_map_entry_link(map, prev_entry, new_entry); 1057 map->size += new_entry->end - new_entry->start; 1058 1059 #if 0 1060 /* 1061 * Temporarily removed to avoid MAP_STACK panic, due to 1062 * MAP_STACK being a huge hack. Will be added back in 1063 * when MAP_STACK (and the user stack mapping) is fixed. 1064 */ 1065 /* 1066 * It may be possible to simplify the entry 1067 */ 1068 vm_map_simplify_entry(map, new_entry); 1069 #endif 1070 1071 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1072 vm_map_pmap_enter(map, start, prot, 1073 object, OFF_TO_IDX(offset), end - start, 1074 cow & MAP_PREFAULT_PARTIAL); 1075 } 1076 1077 return (KERN_SUCCESS); 1078 } 1079 1080 /* 1081 * vm_map_findspace: 1082 * 1083 * Find the first fit (lowest VM address) for "length" free bytes 1084 * beginning at address >= start in the given map. 1085 * 1086 * In a vm_map_entry, "adj_free" is the amount of free space 1087 * adjacent (higher address) to this entry, and "max_free" is the 1088 * maximum amount of contiguous free space in its subtree. This 1089 * allows finding a free region in one path down the tree, so 1090 * O(log n) amortized with splay trees. 1091 * 1092 * The map must be locked, and leaves it so. 1093 * 1094 * Returns: 0 on success, and starting address in *addr, 1095 * 1 if insufficient space. 1096 */ 1097 int 1098 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1099 vm_offset_t *addr) /* OUT */ 1100 { 1101 vm_map_entry_t entry; 1102 vm_offset_t end, st; 1103 1104 /* 1105 * Request must fit within min/max VM address and must avoid 1106 * address wrap. 1107 */ 1108 if (start < map->min_offset) 1109 start = map->min_offset; 1110 if (start + length > map->max_offset || start + length < start) 1111 return (1); 1112 1113 /* Empty tree means wide open address space. */ 1114 if (map->root == NULL) { 1115 *addr = start; 1116 goto found; 1117 } 1118 1119 /* 1120 * After splay, if start comes before root node, then there 1121 * must be a gap from start to the root. 1122 */ 1123 map->root = vm_map_entry_splay(start, map->root); 1124 if (start + length <= map->root->start) { 1125 *addr = start; 1126 goto found; 1127 } 1128 1129 /* 1130 * Root is the last node that might begin its gap before 1131 * start, and this is the last comparison where address 1132 * wrap might be a problem. 1133 */ 1134 st = (start > map->root->end) ? start : map->root->end; 1135 if (length <= map->root->end + map->root->adj_free - st) { 1136 *addr = st; 1137 goto found; 1138 } 1139 1140 /* With max_free, can immediately tell if no solution. */ 1141 entry = map->root->right; 1142 if (entry == NULL || length > entry->max_free) 1143 return (1); 1144 1145 /* 1146 * Search the right subtree in the order: left subtree, root, 1147 * right subtree (first fit). The previous splay implies that 1148 * all regions in the right subtree have addresses > start. 1149 */ 1150 while (entry != NULL) { 1151 if (entry->left != NULL && entry->left->max_free >= length) 1152 entry = entry->left; 1153 else if (entry->adj_free >= length) { 1154 *addr = entry->end; 1155 goto found; 1156 } else 1157 entry = entry->right; 1158 } 1159 1160 /* Can't get here, so panic if we do. */ 1161 panic("vm_map_findspace: max_free corrupt"); 1162 1163 found: 1164 /* Expand the kernel pmap, if necessary. */ 1165 if (map == kernel_map) { 1166 end = round_page(*addr + length); 1167 if (end > kernel_vm_end) 1168 pmap_growkernel(end); 1169 } 1170 return (0); 1171 } 1172 1173 int 1174 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1175 vm_offset_t *addr /* IN/OUT */, vm_size_t length, vm_prot_t prot, 1176 vm_prot_t max, int cow) 1177 { 1178 vm_offset_t start, end; 1179 int result; 1180 1181 start = *addr; 1182 vm_map_lock(map); 1183 end = start + length; 1184 VM_MAP_RANGE_CHECK(map, start, end); 1185 (void) vm_map_delete(map, start, end); 1186 result = vm_map_insert(map, object, offset, start, end, prot, 1187 max, cow); 1188 vm_map_unlock(map); 1189 return (result); 1190 } 1191 1192 /* 1193 * vm_map_find finds an unallocated region in the target address 1194 * map with the given length. The search is defined to be 1195 * first-fit from the specified address; the region found is 1196 * returned in the same parameter. 1197 * 1198 * If object is non-NULL, ref count must be bumped by caller 1199 * prior to making call to account for the new entry. 1200 */ 1201 int 1202 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1203 vm_offset_t *addr, /* IN/OUT */ 1204 vm_size_t length, boolean_t find_space, vm_prot_t prot, 1205 vm_prot_t max, int cow) 1206 { 1207 vm_offset_t start; 1208 int result; 1209 1210 start = *addr; 1211 vm_map_lock(map); 1212 if (find_space) { 1213 if (vm_map_findspace(map, start, length, addr)) { 1214 vm_map_unlock(map); 1215 return (KERN_NO_SPACE); 1216 } 1217 start = *addr; 1218 } 1219 result = vm_map_insert(map, object, offset, 1220 start, start + length, prot, max, cow); 1221 vm_map_unlock(map); 1222 return (result); 1223 } 1224 1225 /* 1226 * vm_map_simplify_entry: 1227 * 1228 * Simplify the given map entry by merging with either neighbor. This 1229 * routine also has the ability to merge with both neighbors. 1230 * 1231 * The map must be locked. 1232 * 1233 * This routine guarentees that the passed entry remains valid (though 1234 * possibly extended). When merging, this routine may delete one or 1235 * both neighbors. 1236 */ 1237 void 1238 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1239 { 1240 vm_map_entry_t next, prev; 1241 vm_size_t prevsize, esize; 1242 1243 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1244 return; 1245 1246 prev = entry->prev; 1247 if (prev != &map->header) { 1248 prevsize = prev->end - prev->start; 1249 if ( (prev->end == entry->start) && 1250 (prev->object.vm_object == entry->object.vm_object) && 1251 (!prev->object.vm_object || 1252 (prev->offset + prevsize == entry->offset)) && 1253 (prev->eflags == entry->eflags) && 1254 (prev->protection == entry->protection) && 1255 (prev->max_protection == entry->max_protection) && 1256 (prev->inheritance == entry->inheritance) && 1257 (prev->wired_count == entry->wired_count)) { 1258 vm_map_entry_unlink(map, prev); 1259 entry->start = prev->start; 1260 entry->offset = prev->offset; 1261 if (entry->prev != &map->header) 1262 vm_map_entry_resize_free(map, entry->prev); 1263 if (prev->object.vm_object) 1264 vm_object_deallocate(prev->object.vm_object); 1265 vm_map_entry_dispose(map, prev); 1266 } 1267 } 1268 1269 next = entry->next; 1270 if (next != &map->header) { 1271 esize = entry->end - entry->start; 1272 if ((entry->end == next->start) && 1273 (next->object.vm_object == entry->object.vm_object) && 1274 (!entry->object.vm_object || 1275 (entry->offset + esize == next->offset)) && 1276 (next->eflags == entry->eflags) && 1277 (next->protection == entry->protection) && 1278 (next->max_protection == entry->max_protection) && 1279 (next->inheritance == entry->inheritance) && 1280 (next->wired_count == entry->wired_count)) { 1281 vm_map_entry_unlink(map, next); 1282 entry->end = next->end; 1283 vm_map_entry_resize_free(map, entry); 1284 if (next->object.vm_object) 1285 vm_object_deallocate(next->object.vm_object); 1286 vm_map_entry_dispose(map, next); 1287 } 1288 } 1289 } 1290 /* 1291 * vm_map_clip_start: [ internal use only ] 1292 * 1293 * Asserts that the given entry begins at or after 1294 * the specified address; if necessary, 1295 * it splits the entry into two. 1296 */ 1297 #define vm_map_clip_start(map, entry, startaddr) \ 1298 { \ 1299 if (startaddr > entry->start) \ 1300 _vm_map_clip_start(map, entry, startaddr); \ 1301 } 1302 1303 /* 1304 * This routine is called only when it is known that 1305 * the entry must be split. 1306 */ 1307 static void 1308 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1309 { 1310 vm_map_entry_t new_entry; 1311 1312 /* 1313 * Split off the front portion -- note that we must insert the new 1314 * entry BEFORE this one, so that this entry has the specified 1315 * starting address. 1316 */ 1317 vm_map_simplify_entry(map, entry); 1318 1319 /* 1320 * If there is no object backing this entry, we might as well create 1321 * one now. If we defer it, an object can get created after the map 1322 * is clipped, and individual objects will be created for the split-up 1323 * map. This is a bit of a hack, but is also about the best place to 1324 * put this improvement. 1325 */ 1326 if (entry->object.vm_object == NULL && !map->system_map) { 1327 vm_object_t object; 1328 object = vm_object_allocate(OBJT_DEFAULT, 1329 atop(entry->end - entry->start)); 1330 entry->object.vm_object = object; 1331 entry->offset = 0; 1332 } 1333 1334 new_entry = vm_map_entry_create(map); 1335 *new_entry = *entry; 1336 1337 new_entry->end = start; 1338 entry->offset += (start - entry->start); 1339 entry->start = start; 1340 1341 vm_map_entry_link(map, entry->prev, new_entry); 1342 1343 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1344 vm_object_reference(new_entry->object.vm_object); 1345 } 1346 } 1347 1348 /* 1349 * vm_map_clip_end: [ internal use only ] 1350 * 1351 * Asserts that the given entry ends at or before 1352 * the specified address; if necessary, 1353 * it splits the entry into two. 1354 */ 1355 #define vm_map_clip_end(map, entry, endaddr) \ 1356 { \ 1357 if ((endaddr) < (entry->end)) \ 1358 _vm_map_clip_end((map), (entry), (endaddr)); \ 1359 } 1360 1361 /* 1362 * This routine is called only when it is known that 1363 * the entry must be split. 1364 */ 1365 static void 1366 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1367 { 1368 vm_map_entry_t new_entry; 1369 1370 /* 1371 * If there is no object backing this entry, we might as well create 1372 * one now. If we defer it, an object can get created after the map 1373 * is clipped, and individual objects will be created for the split-up 1374 * map. This is a bit of a hack, but is also about the best place to 1375 * put this improvement. 1376 */ 1377 if (entry->object.vm_object == NULL && !map->system_map) { 1378 vm_object_t object; 1379 object = vm_object_allocate(OBJT_DEFAULT, 1380 atop(entry->end - entry->start)); 1381 entry->object.vm_object = object; 1382 entry->offset = 0; 1383 } 1384 1385 /* 1386 * Create a new entry and insert it AFTER the specified entry 1387 */ 1388 new_entry = vm_map_entry_create(map); 1389 *new_entry = *entry; 1390 1391 new_entry->start = entry->end = end; 1392 new_entry->offset += (end - entry->start); 1393 1394 vm_map_entry_link(map, entry, new_entry); 1395 1396 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1397 vm_object_reference(new_entry->object.vm_object); 1398 } 1399 } 1400 1401 /* 1402 * vm_map_submap: [ kernel use only ] 1403 * 1404 * Mark the given range as handled by a subordinate map. 1405 * 1406 * This range must have been created with vm_map_find, 1407 * and no other operations may have been performed on this 1408 * range prior to calling vm_map_submap. 1409 * 1410 * Only a limited number of operations can be performed 1411 * within this rage after calling vm_map_submap: 1412 * vm_fault 1413 * [Don't try vm_map_copy!] 1414 * 1415 * To remove a submapping, one must first remove the 1416 * range from the superior map, and then destroy the 1417 * submap (if desired). [Better yet, don't try it.] 1418 */ 1419 int 1420 vm_map_submap( 1421 vm_map_t map, 1422 vm_offset_t start, 1423 vm_offset_t end, 1424 vm_map_t submap) 1425 { 1426 vm_map_entry_t entry; 1427 int result = KERN_INVALID_ARGUMENT; 1428 1429 vm_map_lock(map); 1430 1431 VM_MAP_RANGE_CHECK(map, start, end); 1432 1433 if (vm_map_lookup_entry(map, start, &entry)) { 1434 vm_map_clip_start(map, entry, start); 1435 } else 1436 entry = entry->next; 1437 1438 vm_map_clip_end(map, entry, end); 1439 1440 if ((entry->start == start) && (entry->end == end) && 1441 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1442 (entry->object.vm_object == NULL)) { 1443 entry->object.sub_map = submap; 1444 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1445 result = KERN_SUCCESS; 1446 } 1447 vm_map_unlock(map); 1448 1449 return (result); 1450 } 1451 1452 /* 1453 * The maximum number of pages to map 1454 */ 1455 #define MAX_INIT_PT 96 1456 1457 /* 1458 * vm_map_pmap_enter: 1459 * 1460 * Preload read-only mappings for the given object's resident pages into 1461 * the given map. This eliminates the soft faults on process startup and 1462 * immediately after an mmap(2). Unless the given flags include 1463 * MAP_PREFAULT_MADVISE, cached pages are not reactivated and mapped. 1464 */ 1465 void 1466 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1467 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1468 { 1469 vm_offset_t start; 1470 vm_page_t p, p_start; 1471 vm_pindex_t psize, tmpidx; 1472 boolean_t are_queues_locked; 1473 1474 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1475 return; 1476 VM_OBJECT_LOCK(object); 1477 if (object->type == OBJT_DEVICE) { 1478 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1479 goto unlock_return; 1480 } 1481 1482 psize = atop(size); 1483 1484 if (object->type != OBJT_VNODE || 1485 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 1486 (object->resident_page_count > MAX_INIT_PT))) { 1487 goto unlock_return; 1488 } 1489 1490 if (psize + pindex > object->size) { 1491 if (object->size < pindex) 1492 goto unlock_return; 1493 psize = object->size - pindex; 1494 } 1495 1496 are_queues_locked = FALSE; 1497 start = 0; 1498 p_start = NULL; 1499 1500 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1501 if (p->pindex < pindex) { 1502 p = vm_page_splay(pindex, object->root); 1503 if ((object->root = p)->pindex < pindex) 1504 p = TAILQ_NEXT(p, listq); 1505 } 1506 } 1507 /* 1508 * Assert: the variable p is either (1) the page with the 1509 * least pindex greater than or equal to the parameter pindex 1510 * or (2) NULL. 1511 */ 1512 for (; 1513 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1514 p = TAILQ_NEXT(p, listq)) { 1515 /* 1516 * don't allow an madvise to blow away our really 1517 * free pages allocating pv entries. 1518 */ 1519 if ((flags & MAP_PREFAULT_MADVISE) && 1520 cnt.v_free_count < cnt.v_free_reserved) { 1521 psize = tmpidx; 1522 break; 1523 } 1524 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 1525 (p->busy == 0)) { 1526 if (p_start == NULL) { 1527 start = addr + ptoa(tmpidx); 1528 p_start = p; 1529 } 1530 } else if (p_start != NULL) { 1531 if (!are_queues_locked) { 1532 are_queues_locked = TRUE; 1533 vm_page_lock_queues(); 1534 } 1535 pmap_enter_object(map->pmap, start, addr + 1536 ptoa(tmpidx), p_start, prot); 1537 p_start = NULL; 1538 } 1539 } 1540 if (p_start != NULL) { 1541 if (!are_queues_locked) { 1542 are_queues_locked = TRUE; 1543 vm_page_lock_queues(); 1544 } 1545 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1546 p_start, prot); 1547 } 1548 if (are_queues_locked) 1549 vm_page_unlock_queues(); 1550 unlock_return: 1551 VM_OBJECT_UNLOCK(object); 1552 } 1553 1554 /* 1555 * vm_map_protect: 1556 * 1557 * Sets the protection of the specified address 1558 * region in the target map. If "set_max" is 1559 * specified, the maximum protection is to be set; 1560 * otherwise, only the current protection is affected. 1561 */ 1562 int 1563 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1564 vm_prot_t new_prot, boolean_t set_max) 1565 { 1566 vm_map_entry_t current; 1567 vm_map_entry_t entry; 1568 1569 vm_map_lock(map); 1570 1571 VM_MAP_RANGE_CHECK(map, start, end); 1572 1573 if (vm_map_lookup_entry(map, start, &entry)) { 1574 vm_map_clip_start(map, entry, start); 1575 } else { 1576 entry = entry->next; 1577 } 1578 1579 /* 1580 * Make a first pass to check for protection violations. 1581 */ 1582 current = entry; 1583 while ((current != &map->header) && (current->start < end)) { 1584 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1585 vm_map_unlock(map); 1586 return (KERN_INVALID_ARGUMENT); 1587 } 1588 if ((new_prot & current->max_protection) != new_prot) { 1589 vm_map_unlock(map); 1590 return (KERN_PROTECTION_FAILURE); 1591 } 1592 current = current->next; 1593 } 1594 1595 /* 1596 * Go back and fix up protections. [Note that clipping is not 1597 * necessary the second time.] 1598 */ 1599 current = entry; 1600 while ((current != &map->header) && (current->start < end)) { 1601 vm_prot_t old_prot; 1602 1603 vm_map_clip_end(map, current, end); 1604 1605 old_prot = current->protection; 1606 if (set_max) 1607 current->protection = 1608 (current->max_protection = new_prot) & 1609 old_prot; 1610 else 1611 current->protection = new_prot; 1612 1613 /* 1614 * Update physical map if necessary. Worry about copy-on-write 1615 * here -- CHECK THIS XXX 1616 */ 1617 if (current->protection != old_prot) { 1618 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1619 VM_PROT_ALL) 1620 pmap_protect(map->pmap, current->start, 1621 current->end, 1622 current->protection & MASK(current)); 1623 #undef MASK 1624 } 1625 vm_map_simplify_entry(map, current); 1626 current = current->next; 1627 } 1628 vm_map_unlock(map); 1629 return (KERN_SUCCESS); 1630 } 1631 1632 /* 1633 * vm_map_madvise: 1634 * 1635 * This routine traverses a processes map handling the madvise 1636 * system call. Advisories are classified as either those effecting 1637 * the vm_map_entry structure, or those effecting the underlying 1638 * objects. 1639 */ 1640 int 1641 vm_map_madvise( 1642 vm_map_t map, 1643 vm_offset_t start, 1644 vm_offset_t end, 1645 int behav) 1646 { 1647 vm_map_entry_t current, entry; 1648 int modify_map = 0; 1649 1650 /* 1651 * Some madvise calls directly modify the vm_map_entry, in which case 1652 * we need to use an exclusive lock on the map and we need to perform 1653 * various clipping operations. Otherwise we only need a read-lock 1654 * on the map. 1655 */ 1656 switch(behav) { 1657 case MADV_NORMAL: 1658 case MADV_SEQUENTIAL: 1659 case MADV_RANDOM: 1660 case MADV_NOSYNC: 1661 case MADV_AUTOSYNC: 1662 case MADV_NOCORE: 1663 case MADV_CORE: 1664 modify_map = 1; 1665 vm_map_lock(map); 1666 break; 1667 case MADV_WILLNEED: 1668 case MADV_DONTNEED: 1669 case MADV_FREE: 1670 vm_map_lock_read(map); 1671 break; 1672 default: 1673 return (KERN_INVALID_ARGUMENT); 1674 } 1675 1676 /* 1677 * Locate starting entry and clip if necessary. 1678 */ 1679 VM_MAP_RANGE_CHECK(map, start, end); 1680 1681 if (vm_map_lookup_entry(map, start, &entry)) { 1682 if (modify_map) 1683 vm_map_clip_start(map, entry, start); 1684 } else { 1685 entry = entry->next; 1686 } 1687 1688 if (modify_map) { 1689 /* 1690 * madvise behaviors that are implemented in the vm_map_entry. 1691 * 1692 * We clip the vm_map_entry so that behavioral changes are 1693 * limited to the specified address range. 1694 */ 1695 for (current = entry; 1696 (current != &map->header) && (current->start < end); 1697 current = current->next 1698 ) { 1699 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1700 continue; 1701 1702 vm_map_clip_end(map, current, end); 1703 1704 switch (behav) { 1705 case MADV_NORMAL: 1706 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 1707 break; 1708 case MADV_SEQUENTIAL: 1709 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 1710 break; 1711 case MADV_RANDOM: 1712 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 1713 break; 1714 case MADV_NOSYNC: 1715 current->eflags |= MAP_ENTRY_NOSYNC; 1716 break; 1717 case MADV_AUTOSYNC: 1718 current->eflags &= ~MAP_ENTRY_NOSYNC; 1719 break; 1720 case MADV_NOCORE: 1721 current->eflags |= MAP_ENTRY_NOCOREDUMP; 1722 break; 1723 case MADV_CORE: 1724 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 1725 break; 1726 default: 1727 break; 1728 } 1729 vm_map_simplify_entry(map, current); 1730 } 1731 vm_map_unlock(map); 1732 } else { 1733 vm_pindex_t pindex; 1734 int count; 1735 1736 /* 1737 * madvise behaviors that are implemented in the underlying 1738 * vm_object. 1739 * 1740 * Since we don't clip the vm_map_entry, we have to clip 1741 * the vm_object pindex and count. 1742 */ 1743 for (current = entry; 1744 (current != &map->header) && (current->start < end); 1745 current = current->next 1746 ) { 1747 vm_offset_t useStart; 1748 1749 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1750 continue; 1751 1752 pindex = OFF_TO_IDX(current->offset); 1753 count = atop(current->end - current->start); 1754 useStart = current->start; 1755 1756 if (current->start < start) { 1757 pindex += atop(start - current->start); 1758 count -= atop(start - current->start); 1759 useStart = start; 1760 } 1761 if (current->end > end) 1762 count -= atop(current->end - end); 1763 1764 if (count <= 0) 1765 continue; 1766 1767 vm_object_madvise(current->object.vm_object, 1768 pindex, count, behav); 1769 if (behav == MADV_WILLNEED) { 1770 vm_map_pmap_enter(map, 1771 useStart, 1772 current->protection, 1773 current->object.vm_object, 1774 pindex, 1775 (count << PAGE_SHIFT), 1776 MAP_PREFAULT_MADVISE 1777 ); 1778 } 1779 } 1780 vm_map_unlock_read(map); 1781 } 1782 return (0); 1783 } 1784 1785 1786 /* 1787 * vm_map_inherit: 1788 * 1789 * Sets the inheritance of the specified address 1790 * range in the target map. Inheritance 1791 * affects how the map will be shared with 1792 * child maps at the time of vm_map_fork. 1793 */ 1794 int 1795 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 1796 vm_inherit_t new_inheritance) 1797 { 1798 vm_map_entry_t entry; 1799 vm_map_entry_t temp_entry; 1800 1801 switch (new_inheritance) { 1802 case VM_INHERIT_NONE: 1803 case VM_INHERIT_COPY: 1804 case VM_INHERIT_SHARE: 1805 break; 1806 default: 1807 return (KERN_INVALID_ARGUMENT); 1808 } 1809 vm_map_lock(map); 1810 VM_MAP_RANGE_CHECK(map, start, end); 1811 if (vm_map_lookup_entry(map, start, &temp_entry)) { 1812 entry = temp_entry; 1813 vm_map_clip_start(map, entry, start); 1814 } else 1815 entry = temp_entry->next; 1816 while ((entry != &map->header) && (entry->start < end)) { 1817 vm_map_clip_end(map, entry, end); 1818 entry->inheritance = new_inheritance; 1819 vm_map_simplify_entry(map, entry); 1820 entry = entry->next; 1821 } 1822 vm_map_unlock(map); 1823 return (KERN_SUCCESS); 1824 } 1825 1826 /* 1827 * vm_map_unwire: 1828 * 1829 * Implements both kernel and user unwiring. 1830 */ 1831 int 1832 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1833 int flags) 1834 { 1835 vm_map_entry_t entry, first_entry, tmp_entry; 1836 vm_offset_t saved_start; 1837 unsigned int last_timestamp; 1838 int rv; 1839 boolean_t need_wakeup, result, user_unwire; 1840 1841 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1842 vm_map_lock(map); 1843 VM_MAP_RANGE_CHECK(map, start, end); 1844 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1845 if (flags & VM_MAP_WIRE_HOLESOK) 1846 first_entry = first_entry->next; 1847 else { 1848 vm_map_unlock(map); 1849 return (KERN_INVALID_ADDRESS); 1850 } 1851 } 1852 last_timestamp = map->timestamp; 1853 entry = first_entry; 1854 while (entry != &map->header && entry->start < end) { 1855 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1856 /* 1857 * We have not yet clipped the entry. 1858 */ 1859 saved_start = (start >= entry->start) ? start : 1860 entry->start; 1861 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1862 if (vm_map_unlock_and_wait(map, 0)) { 1863 /* 1864 * Allow interruption of user unwiring? 1865 */ 1866 } 1867 vm_map_lock(map); 1868 if (last_timestamp+1 != map->timestamp) { 1869 /* 1870 * Look again for the entry because the map was 1871 * modified while it was unlocked. 1872 * Specifically, the entry may have been 1873 * clipped, merged, or deleted. 1874 */ 1875 if (!vm_map_lookup_entry(map, saved_start, 1876 &tmp_entry)) { 1877 if (flags & VM_MAP_WIRE_HOLESOK) 1878 tmp_entry = tmp_entry->next; 1879 else { 1880 if (saved_start == start) { 1881 /* 1882 * First_entry has been deleted. 1883 */ 1884 vm_map_unlock(map); 1885 return (KERN_INVALID_ADDRESS); 1886 } 1887 end = saved_start; 1888 rv = KERN_INVALID_ADDRESS; 1889 goto done; 1890 } 1891 } 1892 if (entry == first_entry) 1893 first_entry = tmp_entry; 1894 else 1895 first_entry = NULL; 1896 entry = tmp_entry; 1897 } 1898 last_timestamp = map->timestamp; 1899 continue; 1900 } 1901 vm_map_clip_start(map, entry, start); 1902 vm_map_clip_end(map, entry, end); 1903 /* 1904 * Mark the entry in case the map lock is released. (See 1905 * above.) 1906 */ 1907 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1908 /* 1909 * Check the map for holes in the specified region. 1910 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 1911 */ 1912 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 1913 (entry->end < end && (entry->next == &map->header || 1914 entry->next->start > entry->end))) { 1915 end = entry->end; 1916 rv = KERN_INVALID_ADDRESS; 1917 goto done; 1918 } 1919 /* 1920 * If system unwiring, require that the entry is system wired. 1921 */ 1922 if (!user_unwire && 1923 vm_map_entry_system_wired_count(entry) == 0) { 1924 end = entry->end; 1925 rv = KERN_INVALID_ARGUMENT; 1926 goto done; 1927 } 1928 entry = entry->next; 1929 } 1930 rv = KERN_SUCCESS; 1931 done: 1932 need_wakeup = FALSE; 1933 if (first_entry == NULL) { 1934 result = vm_map_lookup_entry(map, start, &first_entry); 1935 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 1936 first_entry = first_entry->next; 1937 else 1938 KASSERT(result, ("vm_map_unwire: lookup failed")); 1939 } 1940 entry = first_entry; 1941 while (entry != &map->header && entry->start < end) { 1942 if (rv == KERN_SUCCESS && (!user_unwire || 1943 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 1944 if (user_unwire) 1945 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1946 entry->wired_count--; 1947 if (entry->wired_count == 0) { 1948 /* 1949 * Retain the map lock. 1950 */ 1951 vm_fault_unwire(map, entry->start, entry->end, 1952 entry->object.vm_object != NULL && 1953 entry->object.vm_object->type == OBJT_DEVICE); 1954 } 1955 } 1956 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 1957 ("vm_map_unwire: in-transition flag missing")); 1958 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 1959 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 1960 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 1961 need_wakeup = TRUE; 1962 } 1963 vm_map_simplify_entry(map, entry); 1964 entry = entry->next; 1965 } 1966 vm_map_unlock(map); 1967 if (need_wakeup) 1968 vm_map_wakeup(map); 1969 return (rv); 1970 } 1971 1972 /* 1973 * vm_map_wire: 1974 * 1975 * Implements both kernel and user wiring. 1976 */ 1977 int 1978 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1979 int flags) 1980 { 1981 vm_map_entry_t entry, first_entry, tmp_entry; 1982 vm_offset_t saved_end, saved_start; 1983 unsigned int last_timestamp; 1984 int rv; 1985 boolean_t fictitious, need_wakeup, result, user_wire; 1986 1987 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1988 vm_map_lock(map); 1989 VM_MAP_RANGE_CHECK(map, start, end); 1990 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1991 if (flags & VM_MAP_WIRE_HOLESOK) 1992 first_entry = first_entry->next; 1993 else { 1994 vm_map_unlock(map); 1995 return (KERN_INVALID_ADDRESS); 1996 } 1997 } 1998 last_timestamp = map->timestamp; 1999 entry = first_entry; 2000 while (entry != &map->header && entry->start < end) { 2001 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2002 /* 2003 * We have not yet clipped the entry. 2004 */ 2005 saved_start = (start >= entry->start) ? start : 2006 entry->start; 2007 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2008 if (vm_map_unlock_and_wait(map, 0)) { 2009 /* 2010 * Allow interruption of user wiring? 2011 */ 2012 } 2013 vm_map_lock(map); 2014 if (last_timestamp + 1 != map->timestamp) { 2015 /* 2016 * Look again for the entry because the map was 2017 * modified while it was unlocked. 2018 * Specifically, the entry may have been 2019 * clipped, merged, or deleted. 2020 */ 2021 if (!vm_map_lookup_entry(map, saved_start, 2022 &tmp_entry)) { 2023 if (flags & VM_MAP_WIRE_HOLESOK) 2024 tmp_entry = tmp_entry->next; 2025 else { 2026 if (saved_start == start) { 2027 /* 2028 * first_entry has been deleted. 2029 */ 2030 vm_map_unlock(map); 2031 return (KERN_INVALID_ADDRESS); 2032 } 2033 end = saved_start; 2034 rv = KERN_INVALID_ADDRESS; 2035 goto done; 2036 } 2037 } 2038 if (entry == first_entry) 2039 first_entry = tmp_entry; 2040 else 2041 first_entry = NULL; 2042 entry = tmp_entry; 2043 } 2044 last_timestamp = map->timestamp; 2045 continue; 2046 } 2047 vm_map_clip_start(map, entry, start); 2048 vm_map_clip_end(map, entry, end); 2049 /* 2050 * Mark the entry in case the map lock is released. (See 2051 * above.) 2052 */ 2053 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2054 /* 2055 * 2056 */ 2057 if (entry->wired_count == 0) { 2058 entry->wired_count++; 2059 saved_start = entry->start; 2060 saved_end = entry->end; 2061 fictitious = entry->object.vm_object != NULL && 2062 entry->object.vm_object->type == OBJT_DEVICE; 2063 /* 2064 * Release the map lock, relying on the in-transition 2065 * mark. 2066 */ 2067 vm_map_unlock(map); 2068 rv = vm_fault_wire(map, saved_start, saved_end, 2069 user_wire, fictitious); 2070 vm_map_lock(map); 2071 if (last_timestamp + 1 != map->timestamp) { 2072 /* 2073 * Look again for the entry because the map was 2074 * modified while it was unlocked. The entry 2075 * may have been clipped, but NOT merged or 2076 * deleted. 2077 */ 2078 result = vm_map_lookup_entry(map, saved_start, 2079 &tmp_entry); 2080 KASSERT(result, ("vm_map_wire: lookup failed")); 2081 if (entry == first_entry) 2082 first_entry = tmp_entry; 2083 else 2084 first_entry = NULL; 2085 entry = tmp_entry; 2086 while (entry->end < saved_end) { 2087 if (rv != KERN_SUCCESS) { 2088 KASSERT(entry->wired_count == 1, 2089 ("vm_map_wire: bad count")); 2090 entry->wired_count = -1; 2091 } 2092 entry = entry->next; 2093 } 2094 } 2095 last_timestamp = map->timestamp; 2096 if (rv != KERN_SUCCESS) { 2097 KASSERT(entry->wired_count == 1, 2098 ("vm_map_wire: bad count")); 2099 /* 2100 * Assign an out-of-range value to represent 2101 * the failure to wire this entry. 2102 */ 2103 entry->wired_count = -1; 2104 end = entry->end; 2105 goto done; 2106 } 2107 } else if (!user_wire || 2108 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2109 entry->wired_count++; 2110 } 2111 /* 2112 * Check the map for holes in the specified region. 2113 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2114 */ 2115 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2116 (entry->end < end && (entry->next == &map->header || 2117 entry->next->start > entry->end))) { 2118 end = entry->end; 2119 rv = KERN_INVALID_ADDRESS; 2120 goto done; 2121 } 2122 entry = entry->next; 2123 } 2124 rv = KERN_SUCCESS; 2125 done: 2126 need_wakeup = FALSE; 2127 if (first_entry == NULL) { 2128 result = vm_map_lookup_entry(map, start, &first_entry); 2129 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2130 first_entry = first_entry->next; 2131 else 2132 KASSERT(result, ("vm_map_wire: lookup failed")); 2133 } 2134 entry = first_entry; 2135 while (entry != &map->header && entry->start < end) { 2136 if (rv == KERN_SUCCESS) { 2137 if (user_wire) 2138 entry->eflags |= MAP_ENTRY_USER_WIRED; 2139 } else if (entry->wired_count == -1) { 2140 /* 2141 * Wiring failed on this entry. Thus, unwiring is 2142 * unnecessary. 2143 */ 2144 entry->wired_count = 0; 2145 } else { 2146 if (!user_wire || 2147 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2148 entry->wired_count--; 2149 if (entry->wired_count == 0) { 2150 /* 2151 * Retain the map lock. 2152 */ 2153 vm_fault_unwire(map, entry->start, entry->end, 2154 entry->object.vm_object != NULL && 2155 entry->object.vm_object->type == OBJT_DEVICE); 2156 } 2157 } 2158 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2159 ("vm_map_wire: in-transition flag missing")); 2160 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2161 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2162 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2163 need_wakeup = TRUE; 2164 } 2165 vm_map_simplify_entry(map, entry); 2166 entry = entry->next; 2167 } 2168 vm_map_unlock(map); 2169 if (need_wakeup) 2170 vm_map_wakeup(map); 2171 return (rv); 2172 } 2173 2174 /* 2175 * vm_map_sync 2176 * 2177 * Push any dirty cached pages in the address range to their pager. 2178 * If syncio is TRUE, dirty pages are written synchronously. 2179 * If invalidate is TRUE, any cached pages are freed as well. 2180 * 2181 * If the size of the region from start to end is zero, we are 2182 * supposed to flush all modified pages within the region containing 2183 * start. Unfortunately, a region can be split or coalesced with 2184 * neighboring regions, making it difficult to determine what the 2185 * original region was. Therefore, we approximate this requirement by 2186 * flushing the current region containing start. 2187 * 2188 * Returns an error if any part of the specified range is not mapped. 2189 */ 2190 int 2191 vm_map_sync( 2192 vm_map_t map, 2193 vm_offset_t start, 2194 vm_offset_t end, 2195 boolean_t syncio, 2196 boolean_t invalidate) 2197 { 2198 vm_map_entry_t current; 2199 vm_map_entry_t entry; 2200 vm_size_t size; 2201 vm_object_t object; 2202 vm_ooffset_t offset; 2203 2204 vm_map_lock_read(map); 2205 VM_MAP_RANGE_CHECK(map, start, end); 2206 if (!vm_map_lookup_entry(map, start, &entry)) { 2207 vm_map_unlock_read(map); 2208 return (KERN_INVALID_ADDRESS); 2209 } else if (start == end) { 2210 start = entry->start; 2211 end = entry->end; 2212 } 2213 /* 2214 * Make a first pass to check for user-wired memory and holes. 2215 */ 2216 for (current = entry; current != &map->header && current->start < end; 2217 current = current->next) { 2218 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2219 vm_map_unlock_read(map); 2220 return (KERN_INVALID_ARGUMENT); 2221 } 2222 if (end > current->end && 2223 (current->next == &map->header || 2224 current->end != current->next->start)) { 2225 vm_map_unlock_read(map); 2226 return (KERN_INVALID_ADDRESS); 2227 } 2228 } 2229 2230 if (invalidate) 2231 pmap_remove(map->pmap, start, end); 2232 2233 /* 2234 * Make a second pass, cleaning/uncaching pages from the indicated 2235 * objects as we go. 2236 */ 2237 for (current = entry; current != &map->header && current->start < end; 2238 current = current->next) { 2239 offset = current->offset + (start - current->start); 2240 size = (end <= current->end ? end : current->end) - start; 2241 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2242 vm_map_t smap; 2243 vm_map_entry_t tentry; 2244 vm_size_t tsize; 2245 2246 smap = current->object.sub_map; 2247 vm_map_lock_read(smap); 2248 (void) vm_map_lookup_entry(smap, offset, &tentry); 2249 tsize = tentry->end - offset; 2250 if (tsize < size) 2251 size = tsize; 2252 object = tentry->object.vm_object; 2253 offset = tentry->offset + (offset - tentry->start); 2254 vm_map_unlock_read(smap); 2255 } else { 2256 object = current->object.vm_object; 2257 } 2258 vm_object_sync(object, offset, size, syncio, invalidate); 2259 start += size; 2260 } 2261 2262 vm_map_unlock_read(map); 2263 return (KERN_SUCCESS); 2264 } 2265 2266 /* 2267 * vm_map_entry_unwire: [ internal use only ] 2268 * 2269 * Make the region specified by this entry pageable. 2270 * 2271 * The map in question should be locked. 2272 * [This is the reason for this routine's existence.] 2273 */ 2274 static void 2275 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2276 { 2277 vm_fault_unwire(map, entry->start, entry->end, 2278 entry->object.vm_object != NULL && 2279 entry->object.vm_object->type == OBJT_DEVICE); 2280 entry->wired_count = 0; 2281 } 2282 2283 /* 2284 * vm_map_entry_delete: [ internal use only ] 2285 * 2286 * Deallocate the given entry from the target map. 2287 */ 2288 static void 2289 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2290 { 2291 vm_object_t object; 2292 vm_pindex_t offidxstart, offidxend, count; 2293 2294 vm_map_entry_unlink(map, entry); 2295 map->size -= entry->end - entry->start; 2296 2297 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2298 (object = entry->object.vm_object) != NULL) { 2299 count = OFF_TO_IDX(entry->end - entry->start); 2300 offidxstart = OFF_TO_IDX(entry->offset); 2301 offidxend = offidxstart + count; 2302 VM_OBJECT_LOCK(object); 2303 if (object->ref_count != 1 && 2304 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2305 object == kernel_object || object == kmem_object)) { 2306 vm_object_collapse(object); 2307 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2308 if (object->type == OBJT_SWAP) 2309 swap_pager_freespace(object, offidxstart, count); 2310 if (offidxend >= object->size && 2311 offidxstart < object->size) 2312 object->size = offidxstart; 2313 } 2314 VM_OBJECT_UNLOCK(object); 2315 vm_object_deallocate(object); 2316 } 2317 2318 vm_map_entry_dispose(map, entry); 2319 } 2320 2321 /* 2322 * vm_map_delete: [ internal use only ] 2323 * 2324 * Deallocates the given address range from the target 2325 * map. 2326 */ 2327 int 2328 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2329 { 2330 vm_map_entry_t entry; 2331 vm_map_entry_t first_entry; 2332 2333 /* 2334 * Find the start of the region, and clip it 2335 */ 2336 if (!vm_map_lookup_entry(map, start, &first_entry)) 2337 entry = first_entry->next; 2338 else { 2339 entry = first_entry; 2340 vm_map_clip_start(map, entry, start); 2341 } 2342 2343 /* 2344 * Step through all entries in this region 2345 */ 2346 while ((entry != &map->header) && (entry->start < end)) { 2347 vm_map_entry_t next; 2348 2349 /* 2350 * Wait for wiring or unwiring of an entry to complete. 2351 * Also wait for any system wirings to disappear on 2352 * user maps. 2353 */ 2354 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2355 (vm_map_pmap(map) != kernel_pmap && 2356 vm_map_entry_system_wired_count(entry) != 0)) { 2357 unsigned int last_timestamp; 2358 vm_offset_t saved_start; 2359 vm_map_entry_t tmp_entry; 2360 2361 saved_start = entry->start; 2362 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2363 last_timestamp = map->timestamp; 2364 (void) vm_map_unlock_and_wait(map, 0); 2365 vm_map_lock(map); 2366 if (last_timestamp + 1 != map->timestamp) { 2367 /* 2368 * Look again for the entry because the map was 2369 * modified while it was unlocked. 2370 * Specifically, the entry may have been 2371 * clipped, merged, or deleted. 2372 */ 2373 if (!vm_map_lookup_entry(map, saved_start, 2374 &tmp_entry)) 2375 entry = tmp_entry->next; 2376 else { 2377 entry = tmp_entry; 2378 vm_map_clip_start(map, entry, 2379 saved_start); 2380 } 2381 } 2382 continue; 2383 } 2384 vm_map_clip_end(map, entry, end); 2385 2386 next = entry->next; 2387 2388 /* 2389 * Unwire before removing addresses from the pmap; otherwise, 2390 * unwiring will put the entries back in the pmap. 2391 */ 2392 if (entry->wired_count != 0) { 2393 vm_map_entry_unwire(map, entry); 2394 } 2395 2396 pmap_remove(map->pmap, entry->start, entry->end); 2397 2398 /* 2399 * Delete the entry (which may delete the object) only after 2400 * removing all pmap entries pointing to its pages. 2401 * (Otherwise, its page frames may be reallocated, and any 2402 * modify bits will be set in the wrong object!) 2403 */ 2404 vm_map_entry_delete(map, entry); 2405 entry = next; 2406 } 2407 return (KERN_SUCCESS); 2408 } 2409 2410 /* 2411 * vm_map_remove: 2412 * 2413 * Remove the given address range from the target map. 2414 * This is the exported form of vm_map_delete. 2415 */ 2416 int 2417 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2418 { 2419 int result; 2420 2421 vm_map_lock(map); 2422 VM_MAP_RANGE_CHECK(map, start, end); 2423 result = vm_map_delete(map, start, end); 2424 vm_map_unlock(map); 2425 return (result); 2426 } 2427 2428 /* 2429 * vm_map_check_protection: 2430 * 2431 * Assert that the target map allows the specified privilege on the 2432 * entire address region given. The entire region must be allocated. 2433 * 2434 * WARNING! This code does not and should not check whether the 2435 * contents of the region is accessible. For example a smaller file 2436 * might be mapped into a larger address space. 2437 * 2438 * NOTE! This code is also called by munmap(). 2439 * 2440 * The map must be locked. A read lock is sufficient. 2441 */ 2442 boolean_t 2443 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2444 vm_prot_t protection) 2445 { 2446 vm_map_entry_t entry; 2447 vm_map_entry_t tmp_entry; 2448 2449 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2450 return (FALSE); 2451 entry = tmp_entry; 2452 2453 while (start < end) { 2454 if (entry == &map->header) 2455 return (FALSE); 2456 /* 2457 * No holes allowed! 2458 */ 2459 if (start < entry->start) 2460 return (FALSE); 2461 /* 2462 * Check protection associated with entry. 2463 */ 2464 if ((entry->protection & protection) != protection) 2465 return (FALSE); 2466 /* go to next entry */ 2467 start = entry->end; 2468 entry = entry->next; 2469 } 2470 return (TRUE); 2471 } 2472 2473 /* 2474 * vm_map_copy_entry: 2475 * 2476 * Copies the contents of the source entry to the destination 2477 * entry. The entries *must* be aligned properly. 2478 */ 2479 static void 2480 vm_map_copy_entry( 2481 vm_map_t src_map, 2482 vm_map_t dst_map, 2483 vm_map_entry_t src_entry, 2484 vm_map_entry_t dst_entry) 2485 { 2486 vm_object_t src_object; 2487 2488 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2489 return; 2490 2491 if (src_entry->wired_count == 0) { 2492 2493 /* 2494 * If the source entry is marked needs_copy, it is already 2495 * write-protected. 2496 */ 2497 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2498 pmap_protect(src_map->pmap, 2499 src_entry->start, 2500 src_entry->end, 2501 src_entry->protection & ~VM_PROT_WRITE); 2502 } 2503 2504 /* 2505 * Make a copy of the object. 2506 */ 2507 if ((src_object = src_entry->object.vm_object) != NULL) { 2508 VM_OBJECT_LOCK(src_object); 2509 if ((src_object->handle == NULL) && 2510 (src_object->type == OBJT_DEFAULT || 2511 src_object->type == OBJT_SWAP)) { 2512 vm_object_collapse(src_object); 2513 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2514 vm_object_split(src_entry); 2515 src_object = src_entry->object.vm_object; 2516 } 2517 } 2518 vm_object_reference_locked(src_object); 2519 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2520 VM_OBJECT_UNLOCK(src_object); 2521 dst_entry->object.vm_object = src_object; 2522 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2523 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2524 dst_entry->offset = src_entry->offset; 2525 } else { 2526 dst_entry->object.vm_object = NULL; 2527 dst_entry->offset = 0; 2528 } 2529 2530 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2531 dst_entry->end - dst_entry->start, src_entry->start); 2532 } else { 2533 /* 2534 * Of course, wired down pages can't be set copy-on-write. 2535 * Cause wired pages to be copied into the new map by 2536 * simulating faults (the new pages are pageable) 2537 */ 2538 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2539 } 2540 } 2541 2542 /* 2543 * vmspace_map_entry_forked: 2544 * Update the newly-forked vmspace each time a map entry is inherited 2545 * or copied. The values for vm_dsize and vm_tsize are approximate 2546 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2547 */ 2548 static void 2549 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2550 vm_map_entry_t entry) 2551 { 2552 vm_size_t entrysize; 2553 vm_offset_t newend; 2554 2555 entrysize = entry->end - entry->start; 2556 vm2->vm_map.size += entrysize; 2557 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2558 vm2->vm_ssize += btoc(entrysize); 2559 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2560 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2561 newend = MIN(entry->end, 2562 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2563 vm2->vm_dsize += btoc(newend - entry->start); 2564 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2565 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2566 newend = MIN(entry->end, 2567 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2568 vm2->vm_tsize += btoc(newend - entry->start); 2569 } 2570 } 2571 2572 /* 2573 * vmspace_fork: 2574 * Create a new process vmspace structure and vm_map 2575 * based on those of an existing process. The new map 2576 * is based on the old map, according to the inheritance 2577 * values on the regions in that map. 2578 * 2579 * XXX It might be worth coalescing the entries added to the new vmspace. 2580 * 2581 * The source map must not be locked. 2582 */ 2583 struct vmspace * 2584 vmspace_fork(struct vmspace *vm1) 2585 { 2586 struct vmspace *vm2; 2587 vm_map_t old_map = &vm1->vm_map; 2588 vm_map_t new_map; 2589 vm_map_entry_t old_entry; 2590 vm_map_entry_t new_entry; 2591 vm_object_t object; 2592 2593 vm_map_lock(old_map); 2594 2595 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2596 if (vm2 == NULL) 2597 goto unlock_and_return; 2598 vm2->vm_taddr = vm1->vm_taddr; 2599 vm2->vm_daddr = vm1->vm_daddr; 2600 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2601 new_map = &vm2->vm_map; /* XXX */ 2602 new_map->timestamp = 1; 2603 2604 old_entry = old_map->header.next; 2605 2606 while (old_entry != &old_map->header) { 2607 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2608 panic("vm_map_fork: encountered a submap"); 2609 2610 switch (old_entry->inheritance) { 2611 case VM_INHERIT_NONE: 2612 break; 2613 2614 case VM_INHERIT_SHARE: 2615 /* 2616 * Clone the entry, creating the shared object if necessary. 2617 */ 2618 object = old_entry->object.vm_object; 2619 if (object == NULL) { 2620 object = vm_object_allocate(OBJT_DEFAULT, 2621 atop(old_entry->end - old_entry->start)); 2622 old_entry->object.vm_object = object; 2623 old_entry->offset = 0; 2624 } 2625 2626 /* 2627 * Add the reference before calling vm_object_shadow 2628 * to insure that a shadow object is created. 2629 */ 2630 vm_object_reference(object); 2631 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 2632 vm_object_shadow(&old_entry->object.vm_object, 2633 &old_entry->offset, 2634 atop(old_entry->end - old_entry->start)); 2635 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 2636 /* Transfer the second reference too. */ 2637 vm_object_reference( 2638 old_entry->object.vm_object); 2639 vm_object_deallocate(object); 2640 object = old_entry->object.vm_object; 2641 } 2642 VM_OBJECT_LOCK(object); 2643 vm_object_clear_flag(object, OBJ_ONEMAPPING); 2644 VM_OBJECT_UNLOCK(object); 2645 2646 /* 2647 * Clone the entry, referencing the shared object. 2648 */ 2649 new_entry = vm_map_entry_create(new_map); 2650 *new_entry = *old_entry; 2651 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2652 new_entry->wired_count = 0; 2653 2654 /* 2655 * Insert the entry into the new map -- we know we're 2656 * inserting at the end of the new map. 2657 */ 2658 vm_map_entry_link(new_map, new_map->header.prev, 2659 new_entry); 2660 vmspace_map_entry_forked(vm1, vm2, new_entry); 2661 2662 /* 2663 * Update the physical map 2664 */ 2665 pmap_copy(new_map->pmap, old_map->pmap, 2666 new_entry->start, 2667 (old_entry->end - old_entry->start), 2668 old_entry->start); 2669 break; 2670 2671 case VM_INHERIT_COPY: 2672 /* 2673 * Clone the entry and link into the map. 2674 */ 2675 new_entry = vm_map_entry_create(new_map); 2676 *new_entry = *old_entry; 2677 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2678 new_entry->wired_count = 0; 2679 new_entry->object.vm_object = NULL; 2680 vm_map_entry_link(new_map, new_map->header.prev, 2681 new_entry); 2682 vmspace_map_entry_forked(vm1, vm2, new_entry); 2683 vm_map_copy_entry(old_map, new_map, old_entry, 2684 new_entry); 2685 break; 2686 } 2687 old_entry = old_entry->next; 2688 } 2689 unlock_and_return: 2690 vm_map_unlock(old_map); 2691 2692 return (vm2); 2693 } 2694 2695 int 2696 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 2697 vm_prot_t prot, vm_prot_t max, int cow) 2698 { 2699 vm_map_entry_t new_entry, prev_entry; 2700 vm_offset_t bot, top; 2701 vm_size_t init_ssize; 2702 int orient, rv; 2703 rlim_t vmemlim; 2704 2705 /* 2706 * The stack orientation is piggybacked with the cow argument. 2707 * Extract it into orient and mask the cow argument so that we 2708 * don't pass it around further. 2709 * NOTE: We explicitly allow bi-directional stacks. 2710 */ 2711 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 2712 cow &= ~orient; 2713 KASSERT(orient != 0, ("No stack grow direction")); 2714 2715 if (addrbos < vm_map_min(map) || 2716 addrbos > vm_map_max(map) || 2717 addrbos + max_ssize < addrbos) 2718 return (KERN_NO_SPACE); 2719 2720 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 2721 2722 PROC_LOCK(curthread->td_proc); 2723 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 2724 PROC_UNLOCK(curthread->td_proc); 2725 2726 vm_map_lock(map); 2727 2728 /* If addr is already mapped, no go */ 2729 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 2730 vm_map_unlock(map); 2731 return (KERN_NO_SPACE); 2732 } 2733 2734 /* If we would blow our VMEM resource limit, no go */ 2735 if (map->size + init_ssize > vmemlim) { 2736 vm_map_unlock(map); 2737 return (KERN_NO_SPACE); 2738 } 2739 2740 /* 2741 * If we can't accomodate max_ssize in the current mapping, no go. 2742 * However, we need to be aware that subsequent user mappings might 2743 * map into the space we have reserved for stack, and currently this 2744 * space is not protected. 2745 * 2746 * Hopefully we will at least detect this condition when we try to 2747 * grow the stack. 2748 */ 2749 if ((prev_entry->next != &map->header) && 2750 (prev_entry->next->start < addrbos + max_ssize)) { 2751 vm_map_unlock(map); 2752 return (KERN_NO_SPACE); 2753 } 2754 2755 /* 2756 * We initially map a stack of only init_ssize. We will grow as 2757 * needed later. Depending on the orientation of the stack (i.e. 2758 * the grow direction) we either map at the top of the range, the 2759 * bottom of the range or in the middle. 2760 * 2761 * Note: we would normally expect prot and max to be VM_PROT_ALL, 2762 * and cow to be 0. Possibly we should eliminate these as input 2763 * parameters, and just pass these values here in the insert call. 2764 */ 2765 if (orient == MAP_STACK_GROWS_DOWN) 2766 bot = addrbos + max_ssize - init_ssize; 2767 else if (orient == MAP_STACK_GROWS_UP) 2768 bot = addrbos; 2769 else 2770 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 2771 top = bot + init_ssize; 2772 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 2773 2774 /* Now set the avail_ssize amount. */ 2775 if (rv == KERN_SUCCESS) { 2776 if (prev_entry != &map->header) 2777 vm_map_clip_end(map, prev_entry, bot); 2778 new_entry = prev_entry->next; 2779 if (new_entry->end != top || new_entry->start != bot) 2780 panic("Bad entry start/end for new stack entry"); 2781 2782 new_entry->avail_ssize = max_ssize - init_ssize; 2783 if (orient & MAP_STACK_GROWS_DOWN) 2784 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2785 if (orient & MAP_STACK_GROWS_UP) 2786 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 2787 } 2788 2789 vm_map_unlock(map); 2790 return (rv); 2791 } 2792 2793 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 2794 * desired address is already mapped, or if we successfully grow 2795 * the stack. Also returns KERN_SUCCESS if addr is outside the 2796 * stack range (this is strange, but preserves compatibility with 2797 * the grow function in vm_machdep.c). 2798 */ 2799 int 2800 vm_map_growstack(struct proc *p, vm_offset_t addr) 2801 { 2802 vm_map_entry_t next_entry, prev_entry; 2803 vm_map_entry_t new_entry, stack_entry; 2804 struct vmspace *vm = p->p_vmspace; 2805 vm_map_t map = &vm->vm_map; 2806 vm_offset_t end; 2807 size_t grow_amount, max_grow; 2808 rlim_t stacklim, vmemlim; 2809 int is_procstack, rv; 2810 2811 Retry: 2812 PROC_LOCK(p); 2813 stacklim = lim_cur(p, RLIMIT_STACK); 2814 vmemlim = lim_cur(p, RLIMIT_VMEM); 2815 PROC_UNLOCK(p); 2816 2817 vm_map_lock_read(map); 2818 2819 /* If addr is already in the entry range, no need to grow.*/ 2820 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 2821 vm_map_unlock_read(map); 2822 return (KERN_SUCCESS); 2823 } 2824 2825 next_entry = prev_entry->next; 2826 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 2827 /* 2828 * This entry does not grow upwards. Since the address lies 2829 * beyond this entry, the next entry (if one exists) has to 2830 * be a downward growable entry. The entry list header is 2831 * never a growable entry, so it suffices to check the flags. 2832 */ 2833 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 2834 vm_map_unlock_read(map); 2835 return (KERN_SUCCESS); 2836 } 2837 stack_entry = next_entry; 2838 } else { 2839 /* 2840 * This entry grows upward. If the next entry does not at 2841 * least grow downwards, this is the entry we need to grow. 2842 * otherwise we have two possible choices and we have to 2843 * select one. 2844 */ 2845 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 2846 /* 2847 * We have two choices; grow the entry closest to 2848 * the address to minimize the amount of growth. 2849 */ 2850 if (addr - prev_entry->end <= next_entry->start - addr) 2851 stack_entry = prev_entry; 2852 else 2853 stack_entry = next_entry; 2854 } else 2855 stack_entry = prev_entry; 2856 } 2857 2858 if (stack_entry == next_entry) { 2859 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 2860 KASSERT(addr < stack_entry->start, ("foo")); 2861 end = (prev_entry != &map->header) ? prev_entry->end : 2862 stack_entry->start - stack_entry->avail_ssize; 2863 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 2864 max_grow = stack_entry->start - end; 2865 } else { 2866 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 2867 KASSERT(addr >= stack_entry->end, ("foo")); 2868 end = (next_entry != &map->header) ? next_entry->start : 2869 stack_entry->end + stack_entry->avail_ssize; 2870 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 2871 max_grow = end - stack_entry->end; 2872 } 2873 2874 if (grow_amount > stack_entry->avail_ssize) { 2875 vm_map_unlock_read(map); 2876 return (KERN_NO_SPACE); 2877 } 2878 2879 /* 2880 * If there is no longer enough space between the entries nogo, and 2881 * adjust the available space. Note: this should only happen if the 2882 * user has mapped into the stack area after the stack was created, 2883 * and is probably an error. 2884 * 2885 * This also effectively destroys any guard page the user might have 2886 * intended by limiting the stack size. 2887 */ 2888 if (grow_amount > max_grow) { 2889 if (vm_map_lock_upgrade(map)) 2890 goto Retry; 2891 2892 stack_entry->avail_ssize = max_grow; 2893 2894 vm_map_unlock(map); 2895 return (KERN_NO_SPACE); 2896 } 2897 2898 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 2899 2900 /* 2901 * If this is the main process stack, see if we're over the stack 2902 * limit. 2903 */ 2904 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2905 vm_map_unlock_read(map); 2906 return (KERN_NO_SPACE); 2907 } 2908 2909 /* Round up the grow amount modulo SGROWSIZ */ 2910 grow_amount = roundup (grow_amount, sgrowsiz); 2911 if (grow_amount > stack_entry->avail_ssize) 2912 grow_amount = stack_entry->avail_ssize; 2913 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2914 grow_amount = stacklim - ctob(vm->vm_ssize); 2915 } 2916 2917 /* If we would blow our VMEM resource limit, no go */ 2918 if (map->size + grow_amount > vmemlim) { 2919 vm_map_unlock_read(map); 2920 return (KERN_NO_SPACE); 2921 } 2922 2923 if (vm_map_lock_upgrade(map)) 2924 goto Retry; 2925 2926 if (stack_entry == next_entry) { 2927 /* 2928 * Growing downward. 2929 */ 2930 /* Get the preliminary new entry start value */ 2931 addr = stack_entry->start - grow_amount; 2932 2933 /* 2934 * If this puts us into the previous entry, cut back our 2935 * growth to the available space. Also, see the note above. 2936 */ 2937 if (addr < end) { 2938 stack_entry->avail_ssize = max_grow; 2939 addr = end; 2940 } 2941 2942 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 2943 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 2944 2945 /* Adjust the available stack space by the amount we grew. */ 2946 if (rv == KERN_SUCCESS) { 2947 if (prev_entry != &map->header) 2948 vm_map_clip_end(map, prev_entry, addr); 2949 new_entry = prev_entry->next; 2950 KASSERT(new_entry == stack_entry->prev, ("foo")); 2951 KASSERT(new_entry->end == stack_entry->start, ("foo")); 2952 KASSERT(new_entry->start == addr, ("foo")); 2953 grow_amount = new_entry->end - new_entry->start; 2954 new_entry->avail_ssize = stack_entry->avail_ssize - 2955 grow_amount; 2956 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 2957 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2958 } 2959 } else { 2960 /* 2961 * Growing upward. 2962 */ 2963 addr = stack_entry->end + grow_amount; 2964 2965 /* 2966 * If this puts us into the next entry, cut back our growth 2967 * to the available space. Also, see the note above. 2968 */ 2969 if (addr > end) { 2970 stack_entry->avail_ssize = end - stack_entry->end; 2971 addr = end; 2972 } 2973 2974 grow_amount = addr - stack_entry->end; 2975 2976 /* Grow the underlying object if applicable. */ 2977 if (stack_entry->object.vm_object == NULL || 2978 vm_object_coalesce(stack_entry->object.vm_object, 2979 stack_entry->offset, 2980 (vm_size_t)(stack_entry->end - stack_entry->start), 2981 (vm_size_t)grow_amount)) { 2982 map->size += (addr - stack_entry->end); 2983 /* Update the current entry. */ 2984 stack_entry->end = addr; 2985 stack_entry->avail_ssize -= grow_amount; 2986 vm_map_entry_resize_free(map, stack_entry); 2987 rv = KERN_SUCCESS; 2988 2989 if (next_entry != &map->header) 2990 vm_map_clip_start(map, next_entry, addr); 2991 } else 2992 rv = KERN_FAILURE; 2993 } 2994 2995 if (rv == KERN_SUCCESS && is_procstack) 2996 vm->vm_ssize += btoc(grow_amount); 2997 2998 vm_map_unlock(map); 2999 3000 /* 3001 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3002 */ 3003 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 3004 vm_map_wire(map, 3005 (stack_entry == next_entry) ? addr : addr - grow_amount, 3006 (stack_entry == next_entry) ? stack_entry->start : addr, 3007 (p->p_flag & P_SYSTEM) 3008 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 3009 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 3010 } 3011 3012 return (rv); 3013 } 3014 3015 /* 3016 * Unshare the specified VM space for exec. If other processes are 3017 * mapped to it, then create a new one. The new vmspace is null. 3018 */ 3019 int 3020 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3021 { 3022 struct vmspace *oldvmspace = p->p_vmspace; 3023 struct vmspace *newvmspace; 3024 3025 newvmspace = vmspace_alloc(minuser, maxuser); 3026 if (newvmspace == NULL) 3027 return (ENOMEM); 3028 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3029 /* 3030 * This code is written like this for prototype purposes. The 3031 * goal is to avoid running down the vmspace here, but let the 3032 * other process's that are still using the vmspace to finally 3033 * run it down. Even though there is little or no chance of blocking 3034 * here, it is a good idea to keep this form for future mods. 3035 */ 3036 PROC_VMSPACE_LOCK(p); 3037 p->p_vmspace = newvmspace; 3038 PROC_VMSPACE_UNLOCK(p); 3039 if (p == curthread->td_proc) /* XXXKSE ? */ 3040 pmap_activate(curthread); 3041 vmspace_free(oldvmspace); 3042 return (0); 3043 } 3044 3045 /* 3046 * Unshare the specified VM space for forcing COW. This 3047 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3048 */ 3049 int 3050 vmspace_unshare(struct proc *p) 3051 { 3052 struct vmspace *oldvmspace = p->p_vmspace; 3053 struct vmspace *newvmspace; 3054 3055 if (oldvmspace->vm_refcnt == 1) 3056 return (0); 3057 newvmspace = vmspace_fork(oldvmspace); 3058 if (newvmspace == NULL) 3059 return (ENOMEM); 3060 PROC_VMSPACE_LOCK(p); 3061 p->p_vmspace = newvmspace; 3062 PROC_VMSPACE_UNLOCK(p); 3063 if (p == curthread->td_proc) /* XXXKSE ? */ 3064 pmap_activate(curthread); 3065 vmspace_free(oldvmspace); 3066 return (0); 3067 } 3068 3069 /* 3070 * vm_map_lookup: 3071 * 3072 * Finds the VM object, offset, and 3073 * protection for a given virtual address in the 3074 * specified map, assuming a page fault of the 3075 * type specified. 3076 * 3077 * Leaves the map in question locked for read; return 3078 * values are guaranteed until a vm_map_lookup_done 3079 * call is performed. Note that the map argument 3080 * is in/out; the returned map must be used in 3081 * the call to vm_map_lookup_done. 3082 * 3083 * A handle (out_entry) is returned for use in 3084 * vm_map_lookup_done, to make that fast. 3085 * 3086 * If a lookup is requested with "write protection" 3087 * specified, the map may be changed to perform virtual 3088 * copying operations, although the data referenced will 3089 * remain the same. 3090 */ 3091 int 3092 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3093 vm_offset_t vaddr, 3094 vm_prot_t fault_typea, 3095 vm_map_entry_t *out_entry, /* OUT */ 3096 vm_object_t *object, /* OUT */ 3097 vm_pindex_t *pindex, /* OUT */ 3098 vm_prot_t *out_prot, /* OUT */ 3099 boolean_t *wired) /* OUT */ 3100 { 3101 vm_map_entry_t entry; 3102 vm_map_t map = *var_map; 3103 vm_prot_t prot; 3104 vm_prot_t fault_type = fault_typea; 3105 3106 RetryLookup:; 3107 /* 3108 * Lookup the faulting address. 3109 */ 3110 3111 vm_map_lock_read(map); 3112 #define RETURN(why) \ 3113 { \ 3114 vm_map_unlock_read(map); \ 3115 return (why); \ 3116 } 3117 3118 /* 3119 * If the map has an interesting hint, try it before calling full 3120 * blown lookup routine. 3121 */ 3122 entry = map->root; 3123 *out_entry = entry; 3124 if (entry == NULL || 3125 (vaddr < entry->start) || (vaddr >= entry->end)) { 3126 /* 3127 * Entry was either not a valid hint, or the vaddr was not 3128 * contained in the entry, so do a full lookup. 3129 */ 3130 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3131 RETURN(KERN_INVALID_ADDRESS); 3132 3133 entry = *out_entry; 3134 } 3135 3136 /* 3137 * Handle submaps. 3138 */ 3139 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3140 vm_map_t old_map = map; 3141 3142 *var_map = map = entry->object.sub_map; 3143 vm_map_unlock_read(old_map); 3144 goto RetryLookup; 3145 } 3146 3147 /* 3148 * Check whether this task is allowed to have this page. 3149 * Note the special case for MAP_ENTRY_COW 3150 * pages with an override. This is to implement a forced 3151 * COW for debuggers. 3152 */ 3153 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3154 prot = entry->max_protection; 3155 else 3156 prot = entry->protection; 3157 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3158 if ((fault_type & prot) != fault_type) { 3159 RETURN(KERN_PROTECTION_FAILURE); 3160 } 3161 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3162 (entry->eflags & MAP_ENTRY_COW) && 3163 (fault_type & VM_PROT_WRITE) && 3164 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3165 RETURN(KERN_PROTECTION_FAILURE); 3166 } 3167 3168 /* 3169 * If this page is not pageable, we have to get it for all possible 3170 * accesses. 3171 */ 3172 *wired = (entry->wired_count != 0); 3173 if (*wired) 3174 prot = fault_type = entry->protection; 3175 3176 /* 3177 * If the entry was copy-on-write, we either ... 3178 */ 3179 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3180 /* 3181 * If we want to write the page, we may as well handle that 3182 * now since we've got the map locked. 3183 * 3184 * If we don't need to write the page, we just demote the 3185 * permissions allowed. 3186 */ 3187 if (fault_type & VM_PROT_WRITE) { 3188 /* 3189 * Make a new object, and place it in the object 3190 * chain. Note that no new references have appeared 3191 * -- one just moved from the map to the new 3192 * object. 3193 */ 3194 if (vm_map_lock_upgrade(map)) 3195 goto RetryLookup; 3196 3197 vm_object_shadow( 3198 &entry->object.vm_object, 3199 &entry->offset, 3200 atop(entry->end - entry->start)); 3201 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3202 3203 vm_map_lock_downgrade(map); 3204 } else { 3205 /* 3206 * We're attempting to read a copy-on-write page -- 3207 * don't allow writes. 3208 */ 3209 prot &= ~VM_PROT_WRITE; 3210 } 3211 } 3212 3213 /* 3214 * Create an object if necessary. 3215 */ 3216 if (entry->object.vm_object == NULL && 3217 !map->system_map) { 3218 if (vm_map_lock_upgrade(map)) 3219 goto RetryLookup; 3220 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3221 atop(entry->end - entry->start)); 3222 entry->offset = 0; 3223 vm_map_lock_downgrade(map); 3224 } 3225 3226 /* 3227 * Return the object/offset from this entry. If the entry was 3228 * copy-on-write or empty, it has been fixed up. 3229 */ 3230 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3231 *object = entry->object.vm_object; 3232 3233 *out_prot = prot; 3234 return (KERN_SUCCESS); 3235 3236 #undef RETURN 3237 } 3238 3239 /* 3240 * vm_map_lookup_locked: 3241 * 3242 * Lookup the faulting address. A version of vm_map_lookup that returns 3243 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3244 */ 3245 int 3246 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3247 vm_offset_t vaddr, 3248 vm_prot_t fault_typea, 3249 vm_map_entry_t *out_entry, /* OUT */ 3250 vm_object_t *object, /* OUT */ 3251 vm_pindex_t *pindex, /* OUT */ 3252 vm_prot_t *out_prot, /* OUT */ 3253 boolean_t *wired) /* OUT */ 3254 { 3255 vm_map_entry_t entry; 3256 vm_map_t map = *var_map; 3257 vm_prot_t prot; 3258 vm_prot_t fault_type = fault_typea; 3259 3260 /* 3261 * If the map has an interesting hint, try it before calling full 3262 * blown lookup routine. 3263 */ 3264 entry = map->root; 3265 *out_entry = entry; 3266 if (entry == NULL || 3267 (vaddr < entry->start) || (vaddr >= entry->end)) { 3268 /* 3269 * Entry was either not a valid hint, or the vaddr was not 3270 * contained in the entry, so do a full lookup. 3271 */ 3272 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3273 return (KERN_INVALID_ADDRESS); 3274 3275 entry = *out_entry; 3276 } 3277 3278 /* 3279 * Fail if the entry refers to a submap. 3280 */ 3281 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3282 return (KERN_FAILURE); 3283 3284 /* 3285 * Check whether this task is allowed to have this page. 3286 * Note the special case for MAP_ENTRY_COW 3287 * pages with an override. This is to implement a forced 3288 * COW for debuggers. 3289 */ 3290 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3291 prot = entry->max_protection; 3292 else 3293 prot = entry->protection; 3294 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3295 if ((fault_type & prot) != fault_type) 3296 return (KERN_PROTECTION_FAILURE); 3297 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3298 (entry->eflags & MAP_ENTRY_COW) && 3299 (fault_type & VM_PROT_WRITE) && 3300 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3301 return (KERN_PROTECTION_FAILURE); 3302 3303 /* 3304 * If this page is not pageable, we have to get it for all possible 3305 * accesses. 3306 */ 3307 *wired = (entry->wired_count != 0); 3308 if (*wired) 3309 prot = fault_type = entry->protection; 3310 3311 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3312 /* 3313 * Fail if the entry was copy-on-write for a write fault. 3314 */ 3315 if (fault_type & VM_PROT_WRITE) 3316 return (KERN_FAILURE); 3317 /* 3318 * We're attempting to read a copy-on-write page -- 3319 * don't allow writes. 3320 */ 3321 prot &= ~VM_PROT_WRITE; 3322 } 3323 3324 /* 3325 * Fail if an object should be created. 3326 */ 3327 if (entry->object.vm_object == NULL && !map->system_map) 3328 return (KERN_FAILURE); 3329 3330 /* 3331 * Return the object/offset from this entry. If the entry was 3332 * copy-on-write or empty, it has been fixed up. 3333 */ 3334 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3335 *object = entry->object.vm_object; 3336 3337 *out_prot = prot; 3338 return (KERN_SUCCESS); 3339 } 3340 3341 /* 3342 * vm_map_lookup_done: 3343 * 3344 * Releases locks acquired by a vm_map_lookup 3345 * (according to the handle returned by that lookup). 3346 */ 3347 void 3348 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3349 { 3350 /* 3351 * Unlock the main-level map 3352 */ 3353 vm_map_unlock_read(map); 3354 } 3355 3356 #include "opt_ddb.h" 3357 #ifdef DDB 3358 #include <sys/kernel.h> 3359 3360 #include <ddb/ddb.h> 3361 3362 /* 3363 * vm_map_print: [ debug ] 3364 */ 3365 DB_SHOW_COMMAND(map, vm_map_print) 3366 { 3367 static int nlines; 3368 /* XXX convert args. */ 3369 vm_map_t map = (vm_map_t)addr; 3370 boolean_t full = have_addr; 3371 3372 vm_map_entry_t entry; 3373 3374 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3375 (void *)map, 3376 (void *)map->pmap, map->nentries, map->timestamp); 3377 nlines++; 3378 3379 if (!full && db_indent) 3380 return; 3381 3382 db_indent += 2; 3383 for (entry = map->header.next; entry != &map->header; 3384 entry = entry->next) { 3385 db_iprintf("map entry %p: start=%p, end=%p\n", 3386 (void *)entry, (void *)entry->start, (void *)entry->end); 3387 nlines++; 3388 { 3389 static char *inheritance_name[4] = 3390 {"share", "copy", "none", "donate_copy"}; 3391 3392 db_iprintf(" prot=%x/%x/%s", 3393 entry->protection, 3394 entry->max_protection, 3395 inheritance_name[(int)(unsigned char)entry->inheritance]); 3396 if (entry->wired_count != 0) 3397 db_printf(", wired"); 3398 } 3399 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3400 db_printf(", share=%p, offset=0x%jx\n", 3401 (void *)entry->object.sub_map, 3402 (uintmax_t)entry->offset); 3403 nlines++; 3404 if ((entry->prev == &map->header) || 3405 (entry->prev->object.sub_map != 3406 entry->object.sub_map)) { 3407 db_indent += 2; 3408 vm_map_print((db_expr_t)(intptr_t) 3409 entry->object.sub_map, 3410 full, 0, (char *)0); 3411 db_indent -= 2; 3412 } 3413 } else { 3414 db_printf(", object=%p, offset=0x%jx", 3415 (void *)entry->object.vm_object, 3416 (uintmax_t)entry->offset); 3417 if (entry->eflags & MAP_ENTRY_COW) 3418 db_printf(", copy (%s)", 3419 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3420 db_printf("\n"); 3421 nlines++; 3422 3423 if ((entry->prev == &map->header) || 3424 (entry->prev->object.vm_object != 3425 entry->object.vm_object)) { 3426 db_indent += 2; 3427 vm_object_print((db_expr_t)(intptr_t) 3428 entry->object.vm_object, 3429 full, 0, (char *)0); 3430 nlines += 4; 3431 db_indent -= 2; 3432 } 3433 } 3434 } 3435 db_indent -= 2; 3436 if (db_indent == 0) 3437 nlines = 0; 3438 } 3439 3440 3441 DB_SHOW_COMMAND(procvm, procvm) 3442 { 3443 struct proc *p; 3444 3445 if (have_addr) { 3446 p = (struct proc *) addr; 3447 } else { 3448 p = curproc; 3449 } 3450 3451 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3452 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3453 (void *)vmspace_pmap(p->p_vmspace)); 3454 3455 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3456 } 3457 3458 #endif /* DDB */ 3459