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 void 159 vm_map_startup(void) 160 { 161 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 162 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 163 #ifdef INVARIANTS 164 vm_map_zdtor, 165 #else 166 NULL, 167 #endif 168 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 169 uma_prealloc(mapzone, MAX_KMAP); 170 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 171 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 172 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 173 uma_prealloc(kmapentzone, MAX_KMAPENT); 174 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 175 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 176 } 177 178 static void 179 vmspace_zfini(void *mem, int size) 180 { 181 struct vmspace *vm; 182 183 vm = (struct vmspace *)mem; 184 pmap_release(vmspace_pmap(vm)); 185 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 186 } 187 188 static int 189 vmspace_zinit(void *mem, int size, int flags) 190 { 191 struct vmspace *vm; 192 193 vm = (struct vmspace *)mem; 194 195 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 196 pmap_pinit(vmspace_pmap(vm)); 197 return (0); 198 } 199 200 static void 201 vm_map_zfini(void *mem, int size) 202 { 203 vm_map_t map; 204 205 map = (vm_map_t)mem; 206 mtx_destroy(&map->system_mtx); 207 sx_destroy(&map->lock); 208 } 209 210 static int 211 vm_map_zinit(void *mem, int size, int flags) 212 { 213 vm_map_t map; 214 215 map = (vm_map_t)mem; 216 map->nentries = 0; 217 map->size = 0; 218 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 219 sx_init(&map->lock, "user map"); 220 return (0); 221 } 222 223 #ifdef INVARIANTS 224 static void 225 vmspace_zdtor(void *mem, int size, void *arg) 226 { 227 struct vmspace *vm; 228 229 vm = (struct vmspace *)mem; 230 231 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 232 } 233 static void 234 vm_map_zdtor(void *mem, int size, void *arg) 235 { 236 vm_map_t map; 237 238 map = (vm_map_t)mem; 239 KASSERT(map->nentries == 0, 240 ("map %p nentries == %d on free.", 241 map, map->nentries)); 242 KASSERT(map->size == 0, 243 ("map %p size == %lu on free.", 244 map, (unsigned long)map->size)); 245 } 246 #endif /* INVARIANTS */ 247 248 /* 249 * Allocate a vmspace structure, including a vm_map and pmap, 250 * and initialize those structures. The refcnt is set to 1. 251 */ 252 struct vmspace * 253 vmspace_alloc(min, max) 254 vm_offset_t min, max; 255 { 256 struct vmspace *vm; 257 258 vm = uma_zalloc(vmspace_zone, M_WAITOK); 259 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 260 _vm_map_init(&vm->vm_map, min, max); 261 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ 262 vm->vm_refcnt = 1; 263 vm->vm_shm = NULL; 264 vm->vm_swrss = 0; 265 vm->vm_tsize = 0; 266 vm->vm_dsize = 0; 267 vm->vm_ssize = 0; 268 vm->vm_taddr = 0; 269 vm->vm_daddr = 0; 270 vm->vm_maxsaddr = 0; 271 return (vm); 272 } 273 274 void 275 vm_init2(void) 276 { 277 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 278 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 + 279 maxproc * 2 + maxfiles); 280 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 281 #ifdef INVARIANTS 282 vmspace_zdtor, 283 #else 284 NULL, 285 #endif 286 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 287 } 288 289 static inline void 290 vmspace_dofree(struct vmspace *vm) 291 { 292 CTR1(KTR_VM, "vmspace_free: %p", vm); 293 294 /* 295 * Make sure any SysV shm is freed, it might not have been in 296 * exit1(). 297 */ 298 shmexit(vm); 299 300 /* 301 * Lock the map, to wait out all other references to it. 302 * Delete all of the mappings and pages they hold, then call 303 * the pmap module to reclaim anything left. 304 */ 305 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, 306 vm->vm_map.max_offset); 307 308 uma_zfree(vmspace_zone, vm); 309 } 310 311 void 312 vmspace_free(struct vmspace *vm) 313 { 314 int refcnt; 315 316 if (vm->vm_refcnt == 0) 317 panic("vmspace_free: attempt to free already freed vmspace"); 318 319 do 320 refcnt = vm->vm_refcnt; 321 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 322 if (refcnt == 1) 323 vmspace_dofree(vm); 324 } 325 326 void 327 vmspace_exitfree(struct proc *p) 328 { 329 struct vmspace *vm; 330 331 PROC_VMSPACE_LOCK(p); 332 vm = p->p_vmspace; 333 p->p_vmspace = NULL; 334 PROC_VMSPACE_UNLOCK(p); 335 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 336 vmspace_free(vm); 337 } 338 339 void 340 vmspace_exit(struct thread *td) 341 { 342 int refcnt; 343 struct vmspace *vm; 344 struct proc *p; 345 346 /* 347 * Release user portion of address space. 348 * This releases references to vnodes, 349 * which could cause I/O if the file has been unlinked. 350 * Need to do this early enough that we can still sleep. 351 * 352 * The last exiting process to reach this point releases as 353 * much of the environment as it can. vmspace_dofree() is the 354 * slower fallback in case another process had a temporary 355 * reference to the vmspace. 356 */ 357 358 p = td->td_proc; 359 vm = p->p_vmspace; 360 atomic_add_int(&vmspace0.vm_refcnt, 1); 361 do { 362 refcnt = vm->vm_refcnt; 363 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 364 /* Switch now since other proc might free vmspace */ 365 PROC_VMSPACE_LOCK(p); 366 p->p_vmspace = &vmspace0; 367 PROC_VMSPACE_UNLOCK(p); 368 pmap_activate(td); 369 } 370 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 371 if (refcnt == 1) { 372 if (p->p_vmspace != vm) { 373 /* vmspace not yet freed, switch back */ 374 PROC_VMSPACE_LOCK(p); 375 p->p_vmspace = vm; 376 PROC_VMSPACE_UNLOCK(p); 377 pmap_activate(td); 378 } 379 pmap_remove_pages(vmspace_pmap(vm)); 380 /* Switch now since this proc will free vmspace */ 381 PROC_VMSPACE_LOCK(p); 382 p->p_vmspace = &vmspace0; 383 PROC_VMSPACE_UNLOCK(p); 384 pmap_activate(td); 385 vmspace_dofree(vm); 386 } 387 } 388 389 /* Acquire reference to vmspace owned by another process. */ 390 391 struct vmspace * 392 vmspace_acquire_ref(struct proc *p) 393 { 394 struct vmspace *vm; 395 int refcnt; 396 397 PROC_VMSPACE_LOCK(p); 398 vm = p->p_vmspace; 399 if (vm == NULL) { 400 PROC_VMSPACE_UNLOCK(p); 401 return (NULL); 402 } 403 do { 404 refcnt = vm->vm_refcnt; 405 if (refcnt <= 0) { /* Avoid 0->1 transition */ 406 PROC_VMSPACE_UNLOCK(p); 407 return (NULL); 408 } 409 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 410 if (vm != p->p_vmspace) { 411 PROC_VMSPACE_UNLOCK(p); 412 vmspace_free(vm); 413 return (NULL); 414 } 415 PROC_VMSPACE_UNLOCK(p); 416 return (vm); 417 } 418 419 void 420 _vm_map_lock(vm_map_t map, const char *file, int line) 421 { 422 423 if (map->system_map) 424 _mtx_lock_flags(&map->system_mtx, 0, file, line); 425 else 426 _sx_xlock(&map->lock, file, line); 427 map->timestamp++; 428 } 429 430 void 431 _vm_map_unlock(vm_map_t map, const char *file, int line) 432 { 433 434 if (map->system_map) 435 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 436 else 437 _sx_xunlock(&map->lock, file, line); 438 } 439 440 void 441 _vm_map_lock_read(vm_map_t map, const char *file, int line) 442 { 443 444 if (map->system_map) 445 _mtx_lock_flags(&map->system_mtx, 0, file, line); 446 else 447 _sx_xlock(&map->lock, file, line); 448 } 449 450 void 451 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 452 { 453 454 if (map->system_map) 455 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 456 else 457 _sx_xunlock(&map->lock, file, line); 458 } 459 460 int 461 _vm_map_trylock(vm_map_t map, const char *file, int line) 462 { 463 int error; 464 465 error = map->system_map ? 466 !_mtx_trylock(&map->system_mtx, 0, file, line) : 467 !_sx_try_xlock(&map->lock, file, line); 468 if (error == 0) 469 map->timestamp++; 470 return (error == 0); 471 } 472 473 int 474 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 475 { 476 int error; 477 478 error = map->system_map ? 479 !_mtx_trylock(&map->system_mtx, 0, file, line) : 480 !_sx_try_xlock(&map->lock, file, line); 481 return (error == 0); 482 } 483 484 int 485 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 486 { 487 488 #ifdef INVARIANTS 489 if (map->system_map) { 490 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 491 } else 492 _sx_assert(&map->lock, SX_XLOCKED, file, line); 493 #endif 494 map->timestamp++; 495 return (0); 496 } 497 498 void 499 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 500 { 501 502 #ifdef INVARIANTS 503 if (map->system_map) { 504 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 505 } else 506 _sx_assert(&map->lock, SX_XLOCKED, file, line); 507 #endif 508 } 509 510 /* 511 * vm_map_unlock_and_wait: 512 */ 513 int 514 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait) 515 { 516 517 mtx_lock(&map_sleep_mtx); 518 vm_map_unlock(map); 519 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0)); 520 } 521 522 /* 523 * vm_map_wakeup: 524 */ 525 void 526 vm_map_wakeup(vm_map_t map) 527 { 528 529 /* 530 * Acquire and release map_sleep_mtx to prevent a wakeup() 531 * from being performed (and lost) between the vm_map_unlock() 532 * and the msleep() in vm_map_unlock_and_wait(). 533 */ 534 mtx_lock(&map_sleep_mtx); 535 mtx_unlock(&map_sleep_mtx); 536 wakeup(&map->root); 537 } 538 539 long 540 vmspace_resident_count(struct vmspace *vmspace) 541 { 542 return pmap_resident_count(vmspace_pmap(vmspace)); 543 } 544 545 long 546 vmspace_wired_count(struct vmspace *vmspace) 547 { 548 return pmap_wired_count(vmspace_pmap(vmspace)); 549 } 550 551 /* 552 * vm_map_create: 553 * 554 * Creates and returns a new empty VM map with 555 * the given physical map structure, and having 556 * the given lower and upper address bounds. 557 */ 558 vm_map_t 559 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 560 { 561 vm_map_t result; 562 563 result = uma_zalloc(mapzone, M_WAITOK); 564 CTR1(KTR_VM, "vm_map_create: %p", result); 565 _vm_map_init(result, min, max); 566 result->pmap = pmap; 567 return (result); 568 } 569 570 /* 571 * Initialize an existing vm_map structure 572 * such as that in the vmspace structure. 573 * The pmap is set elsewhere. 574 */ 575 static void 576 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 577 { 578 579 map->header.next = map->header.prev = &map->header; 580 map->needs_wakeup = FALSE; 581 map->system_map = 0; 582 map->min_offset = min; 583 map->max_offset = max; 584 map->flags = 0; 585 map->root = NULL; 586 map->timestamp = 0; 587 } 588 589 void 590 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 591 { 592 _vm_map_init(map, min, max); 593 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 594 sx_init(&map->lock, "user map"); 595 } 596 597 /* 598 * vm_map_entry_dispose: [ internal use only ] 599 * 600 * Inverse of vm_map_entry_create. 601 */ 602 static void 603 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 604 { 605 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 606 } 607 608 /* 609 * vm_map_entry_create: [ internal use only ] 610 * 611 * Allocates a VM map entry for insertion. 612 * No entry fields are filled in. 613 */ 614 static vm_map_entry_t 615 vm_map_entry_create(vm_map_t map) 616 { 617 vm_map_entry_t new_entry; 618 619 if (map->system_map) 620 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 621 else 622 new_entry = uma_zalloc(mapentzone, M_WAITOK); 623 if (new_entry == NULL) 624 panic("vm_map_entry_create: kernel resources exhausted"); 625 return (new_entry); 626 } 627 628 /* 629 * vm_map_entry_set_behavior: 630 * 631 * Set the expected access behavior, either normal, random, or 632 * sequential. 633 */ 634 static inline void 635 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 636 { 637 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 638 (behavior & MAP_ENTRY_BEHAV_MASK); 639 } 640 641 /* 642 * vm_map_entry_set_max_free: 643 * 644 * Set the max_free field in a vm_map_entry. 645 */ 646 static inline void 647 vm_map_entry_set_max_free(vm_map_entry_t entry) 648 { 649 650 entry->max_free = entry->adj_free; 651 if (entry->left != NULL && entry->left->max_free > entry->max_free) 652 entry->max_free = entry->left->max_free; 653 if (entry->right != NULL && entry->right->max_free > entry->max_free) 654 entry->max_free = entry->right->max_free; 655 } 656 657 /* 658 * vm_map_entry_splay: 659 * 660 * The Sleator and Tarjan top-down splay algorithm with the 661 * following variation. Max_free must be computed bottom-up, so 662 * on the downward pass, maintain the left and right spines in 663 * reverse order. Then, make a second pass up each side to fix 664 * the pointers and compute max_free. The time bound is O(log n) 665 * amortized. 666 * 667 * The new root is the vm_map_entry containing "addr", or else an 668 * adjacent entry (lower or higher) if addr is not in the tree. 669 * 670 * The map must be locked, and leaves it so. 671 * 672 * Returns: the new root. 673 */ 674 static vm_map_entry_t 675 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 676 { 677 vm_map_entry_t llist, rlist; 678 vm_map_entry_t ltree, rtree; 679 vm_map_entry_t y; 680 681 /* Special case of empty tree. */ 682 if (root == NULL) 683 return (root); 684 685 /* 686 * Pass One: Splay down the tree until we find addr or a NULL 687 * pointer where addr would go. llist and rlist are the two 688 * sides in reverse order (bottom-up), with llist linked by 689 * the right pointer and rlist linked by the left pointer in 690 * the vm_map_entry. Wait until Pass Two to set max_free on 691 * the two spines. 692 */ 693 llist = NULL; 694 rlist = NULL; 695 for (;;) { 696 /* root is never NULL in here. */ 697 if (addr < root->start) { 698 y = root->left; 699 if (y == NULL) 700 break; 701 if (addr < y->start && y->left != NULL) { 702 /* Rotate right and put y on rlist. */ 703 root->left = y->right; 704 y->right = root; 705 vm_map_entry_set_max_free(root); 706 root = y->left; 707 y->left = rlist; 708 rlist = y; 709 } else { 710 /* Put root on rlist. */ 711 root->left = rlist; 712 rlist = root; 713 root = y; 714 } 715 } else { 716 y = root->right; 717 if (addr < root->end || y == NULL) 718 break; 719 if (addr >= y->end && y->right != NULL) { 720 /* Rotate left and put y on llist. */ 721 root->right = y->left; 722 y->left = root; 723 vm_map_entry_set_max_free(root); 724 root = y->right; 725 y->right = llist; 726 llist = y; 727 } else { 728 /* Put root on llist. */ 729 root->right = llist; 730 llist = root; 731 root = y; 732 } 733 } 734 } 735 736 /* 737 * Pass Two: Walk back up the two spines, flip the pointers 738 * and set max_free. The subtrees of the root go at the 739 * bottom of llist and rlist. 740 */ 741 ltree = root->left; 742 while (llist != NULL) { 743 y = llist->right; 744 llist->right = ltree; 745 vm_map_entry_set_max_free(llist); 746 ltree = llist; 747 llist = y; 748 } 749 rtree = root->right; 750 while (rlist != NULL) { 751 y = rlist->left; 752 rlist->left = rtree; 753 vm_map_entry_set_max_free(rlist); 754 rtree = rlist; 755 rlist = y; 756 } 757 758 /* 759 * Final assembly: add ltree and rtree as subtrees of root. 760 */ 761 root->left = ltree; 762 root->right = rtree; 763 vm_map_entry_set_max_free(root); 764 765 return (root); 766 } 767 768 /* 769 * vm_map_entry_{un,}link: 770 * 771 * Insert/remove entries from maps. 772 */ 773 static void 774 vm_map_entry_link(vm_map_t map, 775 vm_map_entry_t after_where, 776 vm_map_entry_t entry) 777 { 778 779 CTR4(KTR_VM, 780 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 781 map->nentries, entry, after_where); 782 map->nentries++; 783 entry->prev = after_where; 784 entry->next = after_where->next; 785 entry->next->prev = entry; 786 after_where->next = entry; 787 788 if (after_where != &map->header) { 789 if (after_where != map->root) 790 vm_map_entry_splay(after_where->start, map->root); 791 entry->right = after_where->right; 792 entry->left = after_where; 793 after_where->right = NULL; 794 after_where->adj_free = entry->start - after_where->end; 795 vm_map_entry_set_max_free(after_where); 796 } else { 797 entry->right = map->root; 798 entry->left = NULL; 799 } 800 entry->adj_free = (entry->next == &map->header ? map->max_offset : 801 entry->next->start) - entry->end; 802 vm_map_entry_set_max_free(entry); 803 map->root = entry; 804 } 805 806 static void 807 vm_map_entry_unlink(vm_map_t map, 808 vm_map_entry_t entry) 809 { 810 vm_map_entry_t next, prev, root; 811 812 if (entry != map->root) 813 vm_map_entry_splay(entry->start, map->root); 814 if (entry->left == NULL) 815 root = entry->right; 816 else { 817 root = vm_map_entry_splay(entry->start, entry->left); 818 root->right = entry->right; 819 root->adj_free = (entry->next == &map->header ? map->max_offset : 820 entry->next->start) - root->end; 821 vm_map_entry_set_max_free(root); 822 } 823 map->root = root; 824 825 prev = entry->prev; 826 next = entry->next; 827 next->prev = prev; 828 prev->next = next; 829 map->nentries--; 830 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 831 map->nentries, entry); 832 } 833 834 /* 835 * vm_map_entry_resize_free: 836 * 837 * Recompute the amount of free space following a vm_map_entry 838 * and propagate that value up the tree. Call this function after 839 * resizing a map entry in-place, that is, without a call to 840 * vm_map_entry_link() or _unlink(). 841 * 842 * The map must be locked, and leaves it so. 843 */ 844 static void 845 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 846 { 847 848 /* 849 * Using splay trees without parent pointers, propagating 850 * max_free up the tree is done by moving the entry to the 851 * root and making the change there. 852 */ 853 if (entry != map->root) 854 map->root = vm_map_entry_splay(entry->start, map->root); 855 856 entry->adj_free = (entry->next == &map->header ? map->max_offset : 857 entry->next->start) - entry->end; 858 vm_map_entry_set_max_free(entry); 859 } 860 861 /* 862 * vm_map_lookup_entry: [ internal use only ] 863 * 864 * Finds the map entry containing (or 865 * immediately preceding) the specified address 866 * in the given map; the entry is returned 867 * in the "entry" parameter. The boolean 868 * result indicates whether the address is 869 * actually contained in the map. 870 */ 871 boolean_t 872 vm_map_lookup_entry( 873 vm_map_t map, 874 vm_offset_t address, 875 vm_map_entry_t *entry) /* OUT */ 876 { 877 vm_map_entry_t cur; 878 879 cur = vm_map_entry_splay(address, map->root); 880 if (cur == NULL) 881 *entry = &map->header; 882 else { 883 map->root = cur; 884 885 if (address >= cur->start) { 886 *entry = cur; 887 if (cur->end > address) 888 return (TRUE); 889 } else 890 *entry = cur->prev; 891 } 892 return (FALSE); 893 } 894 895 /* 896 * vm_map_insert: 897 * 898 * Inserts the given whole VM object into the target 899 * map at the specified address range. The object's 900 * size should match that of the address range. 901 * 902 * Requires that the map be locked, and leaves it so. 903 * 904 * If object is non-NULL, ref count must be bumped by caller 905 * prior to making call to account for the new entry. 906 */ 907 int 908 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 909 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 910 int cow) 911 { 912 vm_map_entry_t new_entry; 913 vm_map_entry_t prev_entry; 914 vm_map_entry_t temp_entry; 915 vm_eflags_t protoeflags; 916 917 /* 918 * Check that the start and end points are not bogus. 919 */ 920 if ((start < map->min_offset) || (end > map->max_offset) || 921 (start >= end)) 922 return (KERN_INVALID_ADDRESS); 923 924 /* 925 * Find the entry prior to the proposed starting address; if it's part 926 * of an existing entry, this range is bogus. 927 */ 928 if (vm_map_lookup_entry(map, start, &temp_entry)) 929 return (KERN_NO_SPACE); 930 931 prev_entry = temp_entry; 932 933 /* 934 * Assert that the next entry doesn't overlap the end point. 935 */ 936 if ((prev_entry->next != &map->header) && 937 (prev_entry->next->start < end)) 938 return (KERN_NO_SPACE); 939 940 protoeflags = 0; 941 942 if (cow & MAP_COPY_ON_WRITE) 943 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 944 945 if (cow & MAP_NOFAULT) { 946 protoeflags |= MAP_ENTRY_NOFAULT; 947 948 KASSERT(object == NULL, 949 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 950 } 951 if (cow & MAP_DISABLE_SYNCER) 952 protoeflags |= MAP_ENTRY_NOSYNC; 953 if (cow & MAP_DISABLE_COREDUMP) 954 protoeflags |= MAP_ENTRY_NOCOREDUMP; 955 956 if (object != NULL) { 957 /* 958 * OBJ_ONEMAPPING must be cleared unless this mapping 959 * is trivially proven to be the only mapping for any 960 * of the object's pages. (Object granularity 961 * reference counting is insufficient to recognize 962 * aliases with precision.) 963 */ 964 VM_OBJECT_LOCK(object); 965 if (object->ref_count > 1 || object->shadow_count != 0) 966 vm_object_clear_flag(object, OBJ_ONEMAPPING); 967 VM_OBJECT_UNLOCK(object); 968 } 969 else if ((prev_entry != &map->header) && 970 (prev_entry->eflags == protoeflags) && 971 (prev_entry->end == start) && 972 (prev_entry->wired_count == 0) && 973 ((prev_entry->object.vm_object == NULL) || 974 vm_object_coalesce(prev_entry->object.vm_object, 975 prev_entry->offset, 976 (vm_size_t)(prev_entry->end - prev_entry->start), 977 (vm_size_t)(end - prev_entry->end)))) { 978 /* 979 * We were able to extend the object. Determine if we 980 * can extend the previous map entry to include the 981 * new range as well. 982 */ 983 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && 984 (prev_entry->protection == prot) && 985 (prev_entry->max_protection == max)) { 986 map->size += (end - prev_entry->end); 987 prev_entry->end = end; 988 vm_map_entry_resize_free(map, prev_entry); 989 vm_map_simplify_entry(map, prev_entry); 990 return (KERN_SUCCESS); 991 } 992 993 /* 994 * If we can extend the object but cannot extend the 995 * map entry, we have to create a new map entry. We 996 * must bump the ref count on the extended object to 997 * account for it. object may be NULL. 998 */ 999 object = prev_entry->object.vm_object; 1000 offset = prev_entry->offset + 1001 (prev_entry->end - prev_entry->start); 1002 vm_object_reference(object); 1003 } 1004 1005 /* 1006 * NOTE: if conditionals fail, object can be NULL here. This occurs 1007 * in things like the buffer map where we manage kva but do not manage 1008 * backing objects. 1009 */ 1010 1011 /* 1012 * Create a new entry 1013 */ 1014 new_entry = vm_map_entry_create(map); 1015 new_entry->start = start; 1016 new_entry->end = end; 1017 1018 new_entry->eflags = protoeflags; 1019 new_entry->object.vm_object = object; 1020 new_entry->offset = offset; 1021 new_entry->avail_ssize = 0; 1022 1023 new_entry->inheritance = VM_INHERIT_DEFAULT; 1024 new_entry->protection = prot; 1025 new_entry->max_protection = max; 1026 new_entry->wired_count = 0; 1027 1028 /* 1029 * Insert the new entry into the list 1030 */ 1031 vm_map_entry_link(map, prev_entry, new_entry); 1032 map->size += new_entry->end - new_entry->start; 1033 1034 #if 0 1035 /* 1036 * Temporarily removed to avoid MAP_STACK panic, due to 1037 * MAP_STACK being a huge hack. Will be added back in 1038 * when MAP_STACK (and the user stack mapping) is fixed. 1039 */ 1040 /* 1041 * It may be possible to simplify the entry 1042 */ 1043 vm_map_simplify_entry(map, new_entry); 1044 #endif 1045 1046 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 1047 vm_map_pmap_enter(map, start, prot, 1048 object, OFF_TO_IDX(offset), end - start, 1049 cow & MAP_PREFAULT_PARTIAL); 1050 } 1051 1052 return (KERN_SUCCESS); 1053 } 1054 1055 /* 1056 * vm_map_findspace: 1057 * 1058 * Find the first fit (lowest VM address) for "length" free bytes 1059 * beginning at address >= start in the given map. 1060 * 1061 * In a vm_map_entry, "adj_free" is the amount of free space 1062 * adjacent (higher address) to this entry, and "max_free" is the 1063 * maximum amount of contiguous free space in its subtree. This 1064 * allows finding a free region in one path down the tree, so 1065 * O(log n) amortized with splay trees. 1066 * 1067 * The map must be locked, and leaves it so. 1068 * 1069 * Returns: 0 on success, and starting address in *addr, 1070 * 1 if insufficient space. 1071 */ 1072 int 1073 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1074 vm_offset_t *addr) /* OUT */ 1075 { 1076 vm_map_entry_t entry; 1077 vm_offset_t end, st; 1078 1079 /* 1080 * Request must fit within min/max VM address and must avoid 1081 * address wrap. 1082 */ 1083 if (start < map->min_offset) 1084 start = map->min_offset; 1085 if (start + length > map->max_offset || start + length < start) 1086 return (1); 1087 1088 /* Empty tree means wide open address space. */ 1089 if (map->root == NULL) { 1090 *addr = start; 1091 goto found; 1092 } 1093 1094 /* 1095 * After splay, if start comes before root node, then there 1096 * must be a gap from start to the root. 1097 */ 1098 map->root = vm_map_entry_splay(start, map->root); 1099 if (start + length <= map->root->start) { 1100 *addr = start; 1101 goto found; 1102 } 1103 1104 /* 1105 * Root is the last node that might begin its gap before 1106 * start, and this is the last comparison where address 1107 * wrap might be a problem. 1108 */ 1109 st = (start > map->root->end) ? start : map->root->end; 1110 if (length <= map->root->end + map->root->adj_free - st) { 1111 *addr = st; 1112 goto found; 1113 } 1114 1115 /* With max_free, can immediately tell if no solution. */ 1116 entry = map->root->right; 1117 if (entry == NULL || length > entry->max_free) 1118 return (1); 1119 1120 /* 1121 * Search the right subtree in the order: left subtree, root, 1122 * right subtree (first fit). The previous splay implies that 1123 * all regions in the right subtree have addresses > start. 1124 */ 1125 while (entry != NULL) { 1126 if (entry->left != NULL && entry->left->max_free >= length) 1127 entry = entry->left; 1128 else if (entry->adj_free >= length) { 1129 *addr = entry->end; 1130 goto found; 1131 } else 1132 entry = entry->right; 1133 } 1134 1135 /* Can't get here, so panic if we do. */ 1136 panic("vm_map_findspace: max_free corrupt"); 1137 1138 found: 1139 /* Expand the kernel pmap, if necessary. */ 1140 if (map == kernel_map) { 1141 end = round_page(*addr + length); 1142 if (end > kernel_vm_end) 1143 pmap_growkernel(end); 1144 } 1145 return (0); 1146 } 1147 1148 /* 1149 * vm_map_find finds an unallocated region in the target address 1150 * map with the given length. The search is defined to be 1151 * first-fit from the specified address; the region found is 1152 * returned in the same parameter. 1153 * 1154 * If object is non-NULL, ref count must be bumped by caller 1155 * prior to making call to account for the new entry. 1156 */ 1157 int 1158 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1159 vm_offset_t *addr, /* IN/OUT */ 1160 vm_size_t length, boolean_t find_space, vm_prot_t prot, 1161 vm_prot_t max, int cow) 1162 { 1163 vm_offset_t start; 1164 int result; 1165 1166 start = *addr; 1167 vm_map_lock(map); 1168 if (find_space) { 1169 if (vm_map_findspace(map, start, length, addr)) { 1170 vm_map_unlock(map); 1171 return (KERN_NO_SPACE); 1172 } 1173 start = *addr; 1174 } 1175 result = vm_map_insert(map, object, offset, 1176 start, start + length, prot, max, cow); 1177 vm_map_unlock(map); 1178 return (result); 1179 } 1180 1181 /* 1182 * vm_map_simplify_entry: 1183 * 1184 * Simplify the given map entry by merging with either neighbor. This 1185 * routine also has the ability to merge with both neighbors. 1186 * 1187 * The map must be locked. 1188 * 1189 * This routine guarentees that the passed entry remains valid (though 1190 * possibly extended). When merging, this routine may delete one or 1191 * both neighbors. 1192 */ 1193 void 1194 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1195 { 1196 vm_map_entry_t next, prev; 1197 vm_size_t prevsize, esize; 1198 1199 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1200 return; 1201 1202 prev = entry->prev; 1203 if (prev != &map->header) { 1204 prevsize = prev->end - prev->start; 1205 if ( (prev->end == entry->start) && 1206 (prev->object.vm_object == entry->object.vm_object) && 1207 (!prev->object.vm_object || 1208 (prev->offset + prevsize == entry->offset)) && 1209 (prev->eflags == entry->eflags) && 1210 (prev->protection == entry->protection) && 1211 (prev->max_protection == entry->max_protection) && 1212 (prev->inheritance == entry->inheritance) && 1213 (prev->wired_count == entry->wired_count)) { 1214 vm_map_entry_unlink(map, prev); 1215 entry->start = prev->start; 1216 entry->offset = prev->offset; 1217 if (entry->prev != &map->header) 1218 vm_map_entry_resize_free(map, entry->prev); 1219 if (prev->object.vm_object) 1220 vm_object_deallocate(prev->object.vm_object); 1221 vm_map_entry_dispose(map, prev); 1222 } 1223 } 1224 1225 next = entry->next; 1226 if (next != &map->header) { 1227 esize = entry->end - entry->start; 1228 if ((entry->end == next->start) && 1229 (next->object.vm_object == entry->object.vm_object) && 1230 (!entry->object.vm_object || 1231 (entry->offset + esize == next->offset)) && 1232 (next->eflags == entry->eflags) && 1233 (next->protection == entry->protection) && 1234 (next->max_protection == entry->max_protection) && 1235 (next->inheritance == entry->inheritance) && 1236 (next->wired_count == entry->wired_count)) { 1237 vm_map_entry_unlink(map, next); 1238 entry->end = next->end; 1239 vm_map_entry_resize_free(map, entry); 1240 if (next->object.vm_object) 1241 vm_object_deallocate(next->object.vm_object); 1242 vm_map_entry_dispose(map, next); 1243 } 1244 } 1245 } 1246 /* 1247 * vm_map_clip_start: [ internal use only ] 1248 * 1249 * Asserts that the given entry begins at or after 1250 * the specified address; if necessary, 1251 * it splits the entry into two. 1252 */ 1253 #define vm_map_clip_start(map, entry, startaddr) \ 1254 { \ 1255 if (startaddr > entry->start) \ 1256 _vm_map_clip_start(map, entry, startaddr); \ 1257 } 1258 1259 /* 1260 * This routine is called only when it is known that 1261 * the entry must be split. 1262 */ 1263 static void 1264 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1265 { 1266 vm_map_entry_t new_entry; 1267 1268 /* 1269 * Split off the front portion -- note that we must insert the new 1270 * entry BEFORE this one, so that this entry has the specified 1271 * starting address. 1272 */ 1273 vm_map_simplify_entry(map, entry); 1274 1275 /* 1276 * If there is no object backing this entry, we might as well create 1277 * one now. If we defer it, an object can get created after the map 1278 * is clipped, and individual objects will be created for the split-up 1279 * map. This is a bit of a hack, but is also about the best place to 1280 * put this improvement. 1281 */ 1282 if (entry->object.vm_object == NULL && !map->system_map) { 1283 vm_object_t object; 1284 object = vm_object_allocate(OBJT_DEFAULT, 1285 atop(entry->end - entry->start)); 1286 entry->object.vm_object = object; 1287 entry->offset = 0; 1288 } 1289 1290 new_entry = vm_map_entry_create(map); 1291 *new_entry = *entry; 1292 1293 new_entry->end = start; 1294 entry->offset += (start - entry->start); 1295 entry->start = start; 1296 1297 vm_map_entry_link(map, entry->prev, new_entry); 1298 1299 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1300 vm_object_reference(new_entry->object.vm_object); 1301 } 1302 } 1303 1304 /* 1305 * vm_map_clip_end: [ internal use only ] 1306 * 1307 * Asserts that the given entry ends at or before 1308 * the specified address; if necessary, 1309 * it splits the entry into two. 1310 */ 1311 #define vm_map_clip_end(map, entry, endaddr) \ 1312 { \ 1313 if ((endaddr) < (entry->end)) \ 1314 _vm_map_clip_end((map), (entry), (endaddr)); \ 1315 } 1316 1317 /* 1318 * This routine is called only when it is known that 1319 * the entry must be split. 1320 */ 1321 static void 1322 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1323 { 1324 vm_map_entry_t new_entry; 1325 1326 /* 1327 * If there is no object backing this entry, we might as well create 1328 * one now. If we defer it, an object can get created after the map 1329 * is clipped, and individual objects will be created for the split-up 1330 * map. This is a bit of a hack, but is also about the best place to 1331 * put this improvement. 1332 */ 1333 if (entry->object.vm_object == NULL && !map->system_map) { 1334 vm_object_t object; 1335 object = vm_object_allocate(OBJT_DEFAULT, 1336 atop(entry->end - entry->start)); 1337 entry->object.vm_object = object; 1338 entry->offset = 0; 1339 } 1340 1341 /* 1342 * Create a new entry and insert it AFTER the specified entry 1343 */ 1344 new_entry = vm_map_entry_create(map); 1345 *new_entry = *entry; 1346 1347 new_entry->start = entry->end = end; 1348 new_entry->offset += (end - entry->start); 1349 1350 vm_map_entry_link(map, entry, new_entry); 1351 1352 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1353 vm_object_reference(new_entry->object.vm_object); 1354 } 1355 } 1356 1357 /* 1358 * VM_MAP_RANGE_CHECK: [ internal use only ] 1359 * 1360 * Asserts that the starting and ending region 1361 * addresses fall within the valid range of the map. 1362 */ 1363 #define VM_MAP_RANGE_CHECK(map, start, end) \ 1364 { \ 1365 if (start < vm_map_min(map)) \ 1366 start = vm_map_min(map); \ 1367 if (end > vm_map_max(map)) \ 1368 end = vm_map_max(map); \ 1369 if (start > end) \ 1370 start = end; \ 1371 } 1372 1373 /* 1374 * vm_map_submap: [ kernel use only ] 1375 * 1376 * Mark the given range as handled by a subordinate map. 1377 * 1378 * This range must have been created with vm_map_find, 1379 * and no other operations may have been performed on this 1380 * range prior to calling vm_map_submap. 1381 * 1382 * Only a limited number of operations can be performed 1383 * within this rage after calling vm_map_submap: 1384 * vm_fault 1385 * [Don't try vm_map_copy!] 1386 * 1387 * To remove a submapping, one must first remove the 1388 * range from the superior map, and then destroy the 1389 * submap (if desired). [Better yet, don't try it.] 1390 */ 1391 int 1392 vm_map_submap( 1393 vm_map_t map, 1394 vm_offset_t start, 1395 vm_offset_t end, 1396 vm_map_t submap) 1397 { 1398 vm_map_entry_t entry; 1399 int result = KERN_INVALID_ARGUMENT; 1400 1401 vm_map_lock(map); 1402 1403 VM_MAP_RANGE_CHECK(map, start, end); 1404 1405 if (vm_map_lookup_entry(map, start, &entry)) { 1406 vm_map_clip_start(map, entry, start); 1407 } else 1408 entry = entry->next; 1409 1410 vm_map_clip_end(map, entry, end); 1411 1412 if ((entry->start == start) && (entry->end == end) && 1413 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1414 (entry->object.vm_object == NULL)) { 1415 entry->object.sub_map = submap; 1416 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1417 result = KERN_SUCCESS; 1418 } 1419 vm_map_unlock(map); 1420 1421 return (result); 1422 } 1423 1424 /* 1425 * The maximum number of pages to map 1426 */ 1427 #define MAX_INIT_PT 96 1428 1429 /* 1430 * vm_map_pmap_enter: 1431 * 1432 * Preload read-only mappings for the given object into the specified 1433 * map. This eliminates the soft faults on process startup and 1434 * immediately after an mmap(2). 1435 */ 1436 void 1437 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1438 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1439 { 1440 vm_offset_t start, tmpidx; 1441 int psize; 1442 vm_page_t p, p_start; 1443 boolean_t are_queues_locked; 1444 1445 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1446 return; 1447 VM_OBJECT_LOCK(object); 1448 if (object->type == OBJT_DEVICE) { 1449 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1450 goto unlock_return; 1451 } 1452 1453 psize = atop(size); 1454 1455 if (object->type != OBJT_VNODE || 1456 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 1457 (object->resident_page_count > MAX_INIT_PT))) { 1458 goto unlock_return; 1459 } 1460 1461 if (psize + pindex > object->size) { 1462 if (object->size < pindex) 1463 goto unlock_return; 1464 psize = object->size - pindex; 1465 } 1466 1467 are_queues_locked = FALSE; 1468 start = 0; 1469 p_start = NULL; 1470 1471 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1472 if (p->pindex < pindex) { 1473 p = vm_page_splay(pindex, object->root); 1474 if ((object->root = p)->pindex < pindex) 1475 p = TAILQ_NEXT(p, listq); 1476 } 1477 } 1478 /* 1479 * Assert: the variable p is either (1) the page with the 1480 * least pindex greater than or equal to the parameter pindex 1481 * or (2) NULL. 1482 */ 1483 for (; 1484 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1485 p = TAILQ_NEXT(p, listq)) { 1486 /* 1487 * don't allow an madvise to blow away our really 1488 * free pages allocating pv entries. 1489 */ 1490 if ((flags & MAP_PREFAULT_MADVISE) && 1491 cnt.v_free_count < cnt.v_free_reserved) { 1492 psize = tmpidx; 1493 break; 1494 } 1495 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 1496 (p->busy == 0) && 1497 (p->flags & PG_FICTITIOUS) == 0) { 1498 if (p_start == NULL) { 1499 start = addr + ptoa(tmpidx); 1500 p_start = p; 1501 } 1502 if (!are_queues_locked) { 1503 are_queues_locked = TRUE; 1504 vm_page_lock_queues(); 1505 } 1506 if (VM_PAGE_INQUEUE1(p, PQ_CACHE)) 1507 vm_page_deactivate(p); 1508 } else if (p_start != NULL) { 1509 pmap_enter_object(map->pmap, start, addr + 1510 ptoa(tmpidx), p_start, prot); 1511 p_start = NULL; 1512 } 1513 } 1514 if (p_start != NULL) 1515 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1516 p_start, prot); 1517 if (are_queues_locked) 1518 vm_page_unlock_queues(); 1519 unlock_return: 1520 VM_OBJECT_UNLOCK(object); 1521 } 1522 1523 /* 1524 * vm_map_protect: 1525 * 1526 * Sets the protection of the specified address 1527 * region in the target map. If "set_max" is 1528 * specified, the maximum protection is to be set; 1529 * otherwise, only the current protection is affected. 1530 */ 1531 int 1532 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1533 vm_prot_t new_prot, boolean_t set_max) 1534 { 1535 vm_map_entry_t current; 1536 vm_map_entry_t entry; 1537 1538 vm_map_lock(map); 1539 1540 VM_MAP_RANGE_CHECK(map, start, end); 1541 1542 if (vm_map_lookup_entry(map, start, &entry)) { 1543 vm_map_clip_start(map, entry, start); 1544 } else { 1545 entry = entry->next; 1546 } 1547 1548 /* 1549 * Make a first pass to check for protection violations. 1550 */ 1551 current = entry; 1552 while ((current != &map->header) && (current->start < end)) { 1553 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1554 vm_map_unlock(map); 1555 return (KERN_INVALID_ARGUMENT); 1556 } 1557 if ((new_prot & current->max_protection) != new_prot) { 1558 vm_map_unlock(map); 1559 return (KERN_PROTECTION_FAILURE); 1560 } 1561 current = current->next; 1562 } 1563 1564 /* 1565 * Go back and fix up protections. [Note that clipping is not 1566 * necessary the second time.] 1567 */ 1568 current = entry; 1569 while ((current != &map->header) && (current->start < end)) { 1570 vm_prot_t old_prot; 1571 1572 vm_map_clip_end(map, current, end); 1573 1574 old_prot = current->protection; 1575 if (set_max) 1576 current->protection = 1577 (current->max_protection = new_prot) & 1578 old_prot; 1579 else 1580 current->protection = new_prot; 1581 1582 /* 1583 * Update physical map if necessary. Worry about copy-on-write 1584 * here -- CHECK THIS XXX 1585 */ 1586 if (current->protection != old_prot) { 1587 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1588 VM_PROT_ALL) 1589 pmap_protect(map->pmap, current->start, 1590 current->end, 1591 current->protection & MASK(current)); 1592 #undef MASK 1593 } 1594 vm_map_simplify_entry(map, current); 1595 current = current->next; 1596 } 1597 vm_map_unlock(map); 1598 return (KERN_SUCCESS); 1599 } 1600 1601 /* 1602 * vm_map_madvise: 1603 * 1604 * This routine traverses a processes map handling the madvise 1605 * system call. Advisories are classified as either those effecting 1606 * the vm_map_entry structure, or those effecting the underlying 1607 * objects. 1608 */ 1609 int 1610 vm_map_madvise( 1611 vm_map_t map, 1612 vm_offset_t start, 1613 vm_offset_t end, 1614 int behav) 1615 { 1616 vm_map_entry_t current, entry; 1617 int modify_map = 0; 1618 1619 /* 1620 * Some madvise calls directly modify the vm_map_entry, in which case 1621 * we need to use an exclusive lock on the map and we need to perform 1622 * various clipping operations. Otherwise we only need a read-lock 1623 * on the map. 1624 */ 1625 switch(behav) { 1626 case MADV_NORMAL: 1627 case MADV_SEQUENTIAL: 1628 case MADV_RANDOM: 1629 case MADV_NOSYNC: 1630 case MADV_AUTOSYNC: 1631 case MADV_NOCORE: 1632 case MADV_CORE: 1633 modify_map = 1; 1634 vm_map_lock(map); 1635 break; 1636 case MADV_WILLNEED: 1637 case MADV_DONTNEED: 1638 case MADV_FREE: 1639 vm_map_lock_read(map); 1640 break; 1641 default: 1642 return (KERN_INVALID_ARGUMENT); 1643 } 1644 1645 /* 1646 * Locate starting entry and clip if necessary. 1647 */ 1648 VM_MAP_RANGE_CHECK(map, start, end); 1649 1650 if (vm_map_lookup_entry(map, start, &entry)) { 1651 if (modify_map) 1652 vm_map_clip_start(map, entry, start); 1653 } else { 1654 entry = entry->next; 1655 } 1656 1657 if (modify_map) { 1658 /* 1659 * madvise behaviors that are implemented in the vm_map_entry. 1660 * 1661 * We clip the vm_map_entry so that behavioral changes are 1662 * limited to the specified address range. 1663 */ 1664 for (current = entry; 1665 (current != &map->header) && (current->start < end); 1666 current = current->next 1667 ) { 1668 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1669 continue; 1670 1671 vm_map_clip_end(map, current, end); 1672 1673 switch (behav) { 1674 case MADV_NORMAL: 1675 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 1676 break; 1677 case MADV_SEQUENTIAL: 1678 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 1679 break; 1680 case MADV_RANDOM: 1681 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 1682 break; 1683 case MADV_NOSYNC: 1684 current->eflags |= MAP_ENTRY_NOSYNC; 1685 break; 1686 case MADV_AUTOSYNC: 1687 current->eflags &= ~MAP_ENTRY_NOSYNC; 1688 break; 1689 case MADV_NOCORE: 1690 current->eflags |= MAP_ENTRY_NOCOREDUMP; 1691 break; 1692 case MADV_CORE: 1693 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 1694 break; 1695 default: 1696 break; 1697 } 1698 vm_map_simplify_entry(map, current); 1699 } 1700 vm_map_unlock(map); 1701 } else { 1702 vm_pindex_t pindex; 1703 int count; 1704 1705 /* 1706 * madvise behaviors that are implemented in the underlying 1707 * vm_object. 1708 * 1709 * Since we don't clip the vm_map_entry, we have to clip 1710 * the vm_object pindex and count. 1711 */ 1712 for (current = entry; 1713 (current != &map->header) && (current->start < end); 1714 current = current->next 1715 ) { 1716 vm_offset_t useStart; 1717 1718 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1719 continue; 1720 1721 pindex = OFF_TO_IDX(current->offset); 1722 count = atop(current->end - current->start); 1723 useStart = current->start; 1724 1725 if (current->start < start) { 1726 pindex += atop(start - current->start); 1727 count -= atop(start - current->start); 1728 useStart = start; 1729 } 1730 if (current->end > end) 1731 count -= atop(current->end - end); 1732 1733 if (count <= 0) 1734 continue; 1735 1736 vm_object_madvise(current->object.vm_object, 1737 pindex, count, behav); 1738 if (behav == MADV_WILLNEED) { 1739 vm_map_pmap_enter(map, 1740 useStart, 1741 current->protection, 1742 current->object.vm_object, 1743 pindex, 1744 (count << PAGE_SHIFT), 1745 MAP_PREFAULT_MADVISE 1746 ); 1747 } 1748 } 1749 vm_map_unlock_read(map); 1750 } 1751 return (0); 1752 } 1753 1754 1755 /* 1756 * vm_map_inherit: 1757 * 1758 * Sets the inheritance of the specified address 1759 * range in the target map. Inheritance 1760 * affects how the map will be shared with 1761 * child maps at the time of vm_map_fork. 1762 */ 1763 int 1764 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 1765 vm_inherit_t new_inheritance) 1766 { 1767 vm_map_entry_t entry; 1768 vm_map_entry_t temp_entry; 1769 1770 switch (new_inheritance) { 1771 case VM_INHERIT_NONE: 1772 case VM_INHERIT_COPY: 1773 case VM_INHERIT_SHARE: 1774 break; 1775 default: 1776 return (KERN_INVALID_ARGUMENT); 1777 } 1778 vm_map_lock(map); 1779 VM_MAP_RANGE_CHECK(map, start, end); 1780 if (vm_map_lookup_entry(map, start, &temp_entry)) { 1781 entry = temp_entry; 1782 vm_map_clip_start(map, entry, start); 1783 } else 1784 entry = temp_entry->next; 1785 while ((entry != &map->header) && (entry->start < end)) { 1786 vm_map_clip_end(map, entry, end); 1787 entry->inheritance = new_inheritance; 1788 vm_map_simplify_entry(map, entry); 1789 entry = entry->next; 1790 } 1791 vm_map_unlock(map); 1792 return (KERN_SUCCESS); 1793 } 1794 1795 /* 1796 * vm_map_unwire: 1797 * 1798 * Implements both kernel and user unwiring. 1799 */ 1800 int 1801 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1802 int flags) 1803 { 1804 vm_map_entry_t entry, first_entry, tmp_entry; 1805 vm_offset_t saved_start; 1806 unsigned int last_timestamp; 1807 int rv; 1808 boolean_t need_wakeup, result, user_unwire; 1809 1810 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1811 vm_map_lock(map); 1812 VM_MAP_RANGE_CHECK(map, start, end); 1813 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1814 if (flags & VM_MAP_WIRE_HOLESOK) 1815 first_entry = first_entry->next; 1816 else { 1817 vm_map_unlock(map); 1818 return (KERN_INVALID_ADDRESS); 1819 } 1820 } 1821 last_timestamp = map->timestamp; 1822 entry = first_entry; 1823 while (entry != &map->header && entry->start < end) { 1824 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1825 /* 1826 * We have not yet clipped the entry. 1827 */ 1828 saved_start = (start >= entry->start) ? start : 1829 entry->start; 1830 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1831 if (vm_map_unlock_and_wait(map, user_unwire)) { 1832 /* 1833 * Allow interruption of user unwiring? 1834 */ 1835 } 1836 vm_map_lock(map); 1837 if (last_timestamp+1 != map->timestamp) { 1838 /* 1839 * Look again for the entry because the map was 1840 * modified while it was unlocked. 1841 * Specifically, the entry may have been 1842 * clipped, merged, or deleted. 1843 */ 1844 if (!vm_map_lookup_entry(map, saved_start, 1845 &tmp_entry)) { 1846 if (flags & VM_MAP_WIRE_HOLESOK) 1847 tmp_entry = tmp_entry->next; 1848 else { 1849 if (saved_start == start) { 1850 /* 1851 * First_entry has been deleted. 1852 */ 1853 vm_map_unlock(map); 1854 return (KERN_INVALID_ADDRESS); 1855 } 1856 end = saved_start; 1857 rv = KERN_INVALID_ADDRESS; 1858 goto done; 1859 } 1860 } 1861 if (entry == first_entry) 1862 first_entry = tmp_entry; 1863 else 1864 first_entry = NULL; 1865 entry = tmp_entry; 1866 } 1867 last_timestamp = map->timestamp; 1868 continue; 1869 } 1870 vm_map_clip_start(map, entry, start); 1871 vm_map_clip_end(map, entry, end); 1872 /* 1873 * Mark the entry in case the map lock is released. (See 1874 * above.) 1875 */ 1876 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1877 /* 1878 * Check the map for holes in the specified region. 1879 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 1880 */ 1881 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 1882 (entry->end < end && (entry->next == &map->header || 1883 entry->next->start > entry->end))) { 1884 end = entry->end; 1885 rv = KERN_INVALID_ADDRESS; 1886 goto done; 1887 } 1888 /* 1889 * If system unwiring, require that the entry is system wired. 1890 */ 1891 if (!user_unwire && 1892 vm_map_entry_system_wired_count(entry) == 0) { 1893 end = entry->end; 1894 rv = KERN_INVALID_ARGUMENT; 1895 goto done; 1896 } 1897 entry = entry->next; 1898 } 1899 rv = KERN_SUCCESS; 1900 done: 1901 need_wakeup = FALSE; 1902 if (first_entry == NULL) { 1903 result = vm_map_lookup_entry(map, start, &first_entry); 1904 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 1905 first_entry = first_entry->next; 1906 else 1907 KASSERT(result, ("vm_map_unwire: lookup failed")); 1908 } 1909 entry = first_entry; 1910 while (entry != &map->header && entry->start < end) { 1911 if (rv == KERN_SUCCESS && (!user_unwire || 1912 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 1913 if (user_unwire) 1914 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1915 entry->wired_count--; 1916 if (entry->wired_count == 0) { 1917 /* 1918 * Retain the map lock. 1919 */ 1920 vm_fault_unwire(map, entry->start, entry->end, 1921 entry->object.vm_object != NULL && 1922 entry->object.vm_object->type == OBJT_DEVICE); 1923 } 1924 } 1925 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 1926 ("vm_map_unwire: in-transition flag missing")); 1927 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 1928 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 1929 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 1930 need_wakeup = TRUE; 1931 } 1932 vm_map_simplify_entry(map, entry); 1933 entry = entry->next; 1934 } 1935 vm_map_unlock(map); 1936 if (need_wakeup) 1937 vm_map_wakeup(map); 1938 return (rv); 1939 } 1940 1941 /* 1942 * vm_map_wire: 1943 * 1944 * Implements both kernel and user wiring. 1945 */ 1946 int 1947 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1948 int flags) 1949 { 1950 vm_map_entry_t entry, first_entry, tmp_entry; 1951 vm_offset_t saved_end, saved_start; 1952 unsigned int last_timestamp; 1953 int rv; 1954 boolean_t fictitious, need_wakeup, result, user_wire; 1955 1956 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1957 vm_map_lock(map); 1958 VM_MAP_RANGE_CHECK(map, start, end); 1959 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1960 if (flags & VM_MAP_WIRE_HOLESOK) 1961 first_entry = first_entry->next; 1962 else { 1963 vm_map_unlock(map); 1964 return (KERN_INVALID_ADDRESS); 1965 } 1966 } 1967 last_timestamp = map->timestamp; 1968 entry = first_entry; 1969 while (entry != &map->header && entry->start < end) { 1970 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1971 /* 1972 * We have not yet clipped the entry. 1973 */ 1974 saved_start = (start >= entry->start) ? start : 1975 entry->start; 1976 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1977 if (vm_map_unlock_and_wait(map, user_wire)) { 1978 /* 1979 * Allow interruption of user wiring? 1980 */ 1981 } 1982 vm_map_lock(map); 1983 if (last_timestamp + 1 != map->timestamp) { 1984 /* 1985 * Look again for the entry because the map was 1986 * modified while it was unlocked. 1987 * Specifically, the entry may have been 1988 * clipped, merged, or deleted. 1989 */ 1990 if (!vm_map_lookup_entry(map, saved_start, 1991 &tmp_entry)) { 1992 if (flags & VM_MAP_WIRE_HOLESOK) 1993 tmp_entry = tmp_entry->next; 1994 else { 1995 if (saved_start == start) { 1996 /* 1997 * first_entry has been deleted. 1998 */ 1999 vm_map_unlock(map); 2000 return (KERN_INVALID_ADDRESS); 2001 } 2002 end = saved_start; 2003 rv = KERN_INVALID_ADDRESS; 2004 goto done; 2005 } 2006 } 2007 if (entry == first_entry) 2008 first_entry = tmp_entry; 2009 else 2010 first_entry = NULL; 2011 entry = tmp_entry; 2012 } 2013 last_timestamp = map->timestamp; 2014 continue; 2015 } 2016 vm_map_clip_start(map, entry, start); 2017 vm_map_clip_end(map, entry, end); 2018 /* 2019 * Mark the entry in case the map lock is released. (See 2020 * above.) 2021 */ 2022 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2023 /* 2024 * 2025 */ 2026 if (entry->wired_count == 0) { 2027 entry->wired_count++; 2028 saved_start = entry->start; 2029 saved_end = entry->end; 2030 fictitious = entry->object.vm_object != NULL && 2031 entry->object.vm_object->type == OBJT_DEVICE; 2032 /* 2033 * Release the map lock, relying on the in-transition 2034 * mark. 2035 */ 2036 vm_map_unlock(map); 2037 rv = vm_fault_wire(map, saved_start, saved_end, 2038 user_wire, fictitious); 2039 vm_map_lock(map); 2040 if (last_timestamp + 1 != map->timestamp) { 2041 /* 2042 * Look again for the entry because the map was 2043 * modified while it was unlocked. The entry 2044 * may have been clipped, but NOT merged or 2045 * deleted. 2046 */ 2047 result = vm_map_lookup_entry(map, saved_start, 2048 &tmp_entry); 2049 KASSERT(result, ("vm_map_wire: lookup failed")); 2050 if (entry == first_entry) 2051 first_entry = tmp_entry; 2052 else 2053 first_entry = NULL; 2054 entry = tmp_entry; 2055 while (entry->end < saved_end) { 2056 if (rv != KERN_SUCCESS) { 2057 KASSERT(entry->wired_count == 1, 2058 ("vm_map_wire: bad count")); 2059 entry->wired_count = -1; 2060 } 2061 entry = entry->next; 2062 } 2063 } 2064 last_timestamp = map->timestamp; 2065 if (rv != KERN_SUCCESS) { 2066 KASSERT(entry->wired_count == 1, 2067 ("vm_map_wire: bad count")); 2068 /* 2069 * Assign an out-of-range value to represent 2070 * the failure to wire this entry. 2071 */ 2072 entry->wired_count = -1; 2073 end = entry->end; 2074 goto done; 2075 } 2076 } else if (!user_wire || 2077 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2078 entry->wired_count++; 2079 } 2080 /* 2081 * Check the map for holes in the specified region. 2082 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2083 */ 2084 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2085 (entry->end < end && (entry->next == &map->header || 2086 entry->next->start > entry->end))) { 2087 end = entry->end; 2088 rv = KERN_INVALID_ADDRESS; 2089 goto done; 2090 } 2091 entry = entry->next; 2092 } 2093 rv = KERN_SUCCESS; 2094 done: 2095 need_wakeup = FALSE; 2096 if (first_entry == NULL) { 2097 result = vm_map_lookup_entry(map, start, &first_entry); 2098 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2099 first_entry = first_entry->next; 2100 else 2101 KASSERT(result, ("vm_map_wire: lookup failed")); 2102 } 2103 entry = first_entry; 2104 while (entry != &map->header && entry->start < end) { 2105 if (rv == KERN_SUCCESS) { 2106 if (user_wire) 2107 entry->eflags |= MAP_ENTRY_USER_WIRED; 2108 } else if (entry->wired_count == -1) { 2109 /* 2110 * Wiring failed on this entry. Thus, unwiring is 2111 * unnecessary. 2112 */ 2113 entry->wired_count = 0; 2114 } else { 2115 if (!user_wire || 2116 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2117 entry->wired_count--; 2118 if (entry->wired_count == 0) { 2119 /* 2120 * Retain the map lock. 2121 */ 2122 vm_fault_unwire(map, entry->start, entry->end, 2123 entry->object.vm_object != NULL && 2124 entry->object.vm_object->type == OBJT_DEVICE); 2125 } 2126 } 2127 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2128 ("vm_map_wire: in-transition flag missing")); 2129 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2130 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2131 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2132 need_wakeup = TRUE; 2133 } 2134 vm_map_simplify_entry(map, entry); 2135 entry = entry->next; 2136 } 2137 vm_map_unlock(map); 2138 if (need_wakeup) 2139 vm_map_wakeup(map); 2140 return (rv); 2141 } 2142 2143 /* 2144 * vm_map_sync 2145 * 2146 * Push any dirty cached pages in the address range to their pager. 2147 * If syncio is TRUE, dirty pages are written synchronously. 2148 * If invalidate is TRUE, any cached pages are freed as well. 2149 * 2150 * If the size of the region from start to end is zero, we are 2151 * supposed to flush all modified pages within the region containing 2152 * start. Unfortunately, a region can be split or coalesced with 2153 * neighboring regions, making it difficult to determine what the 2154 * original region was. Therefore, we approximate this requirement by 2155 * flushing the current region containing start. 2156 * 2157 * Returns an error if any part of the specified range is not mapped. 2158 */ 2159 int 2160 vm_map_sync( 2161 vm_map_t map, 2162 vm_offset_t start, 2163 vm_offset_t end, 2164 boolean_t syncio, 2165 boolean_t invalidate) 2166 { 2167 vm_map_entry_t current; 2168 vm_map_entry_t entry; 2169 vm_size_t size; 2170 vm_object_t object; 2171 vm_ooffset_t offset; 2172 2173 vm_map_lock_read(map); 2174 VM_MAP_RANGE_CHECK(map, start, end); 2175 if (!vm_map_lookup_entry(map, start, &entry)) { 2176 vm_map_unlock_read(map); 2177 return (KERN_INVALID_ADDRESS); 2178 } else if (start == end) { 2179 start = entry->start; 2180 end = entry->end; 2181 } 2182 /* 2183 * Make a first pass to check for user-wired memory and holes. 2184 */ 2185 for (current = entry; current->start < end; current = current->next) { 2186 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2187 vm_map_unlock_read(map); 2188 return (KERN_INVALID_ARGUMENT); 2189 } 2190 if (end > current->end && 2191 (current->next == &map->header || 2192 current->end != current->next->start)) { 2193 vm_map_unlock_read(map); 2194 return (KERN_INVALID_ADDRESS); 2195 } 2196 } 2197 2198 if (invalidate) 2199 pmap_remove(map->pmap, start, end); 2200 2201 /* 2202 * Make a second pass, cleaning/uncaching pages from the indicated 2203 * objects as we go. 2204 */ 2205 for (current = entry; current->start < end; current = current->next) { 2206 offset = current->offset + (start - current->start); 2207 size = (end <= current->end ? end : current->end) - start; 2208 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2209 vm_map_t smap; 2210 vm_map_entry_t tentry; 2211 vm_size_t tsize; 2212 2213 smap = current->object.sub_map; 2214 vm_map_lock_read(smap); 2215 (void) vm_map_lookup_entry(smap, offset, &tentry); 2216 tsize = tentry->end - offset; 2217 if (tsize < size) 2218 size = tsize; 2219 object = tentry->object.vm_object; 2220 offset = tentry->offset + (offset - tentry->start); 2221 vm_map_unlock_read(smap); 2222 } else { 2223 object = current->object.vm_object; 2224 } 2225 vm_object_sync(object, offset, size, syncio, invalidate); 2226 start += size; 2227 } 2228 2229 vm_map_unlock_read(map); 2230 return (KERN_SUCCESS); 2231 } 2232 2233 /* 2234 * vm_map_entry_unwire: [ internal use only ] 2235 * 2236 * Make the region specified by this entry pageable. 2237 * 2238 * The map in question should be locked. 2239 * [This is the reason for this routine's existence.] 2240 */ 2241 static void 2242 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2243 { 2244 vm_fault_unwire(map, entry->start, entry->end, 2245 entry->object.vm_object != NULL && 2246 entry->object.vm_object->type == OBJT_DEVICE); 2247 entry->wired_count = 0; 2248 } 2249 2250 /* 2251 * vm_map_entry_delete: [ internal use only ] 2252 * 2253 * Deallocate the given entry from the target map. 2254 */ 2255 static void 2256 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2257 { 2258 vm_object_t object; 2259 vm_pindex_t offidxstart, offidxend, count; 2260 2261 vm_map_entry_unlink(map, entry); 2262 map->size -= entry->end - entry->start; 2263 2264 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2265 (object = entry->object.vm_object) != NULL) { 2266 count = OFF_TO_IDX(entry->end - entry->start); 2267 offidxstart = OFF_TO_IDX(entry->offset); 2268 offidxend = offidxstart + count; 2269 VM_OBJECT_LOCK(object); 2270 if (object->ref_count != 1 && 2271 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2272 object == kernel_object || object == kmem_object)) { 2273 vm_object_collapse(object); 2274 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2275 if (object->type == OBJT_SWAP) 2276 swap_pager_freespace(object, offidxstart, count); 2277 if (offidxend >= object->size && 2278 offidxstart < object->size) 2279 object->size = offidxstart; 2280 } 2281 VM_OBJECT_UNLOCK(object); 2282 vm_object_deallocate(object); 2283 } 2284 2285 vm_map_entry_dispose(map, entry); 2286 } 2287 2288 /* 2289 * vm_map_delete: [ internal use only ] 2290 * 2291 * Deallocates the given address range from the target 2292 * map. 2293 */ 2294 int 2295 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2296 { 2297 vm_map_entry_t entry; 2298 vm_map_entry_t first_entry; 2299 2300 /* 2301 * Find the start of the region, and clip it 2302 */ 2303 if (!vm_map_lookup_entry(map, start, &first_entry)) 2304 entry = first_entry->next; 2305 else { 2306 entry = first_entry; 2307 vm_map_clip_start(map, entry, start); 2308 } 2309 2310 /* 2311 * Step through all entries in this region 2312 */ 2313 while ((entry != &map->header) && (entry->start < end)) { 2314 vm_map_entry_t next; 2315 2316 /* 2317 * Wait for wiring or unwiring of an entry to complete. 2318 * Also wait for any system wirings to disappear on 2319 * user maps. 2320 */ 2321 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2322 (vm_map_pmap(map) != kernel_pmap && 2323 vm_map_entry_system_wired_count(entry) != 0)) { 2324 unsigned int last_timestamp; 2325 vm_offset_t saved_start; 2326 vm_map_entry_t tmp_entry; 2327 2328 saved_start = entry->start; 2329 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2330 last_timestamp = map->timestamp; 2331 (void) vm_map_unlock_and_wait(map, FALSE); 2332 vm_map_lock(map); 2333 if (last_timestamp + 1 != map->timestamp) { 2334 /* 2335 * Look again for the entry because the map was 2336 * modified while it was unlocked. 2337 * Specifically, the entry may have been 2338 * clipped, merged, or deleted. 2339 */ 2340 if (!vm_map_lookup_entry(map, saved_start, 2341 &tmp_entry)) 2342 entry = tmp_entry->next; 2343 else { 2344 entry = tmp_entry; 2345 vm_map_clip_start(map, entry, 2346 saved_start); 2347 } 2348 } 2349 continue; 2350 } 2351 vm_map_clip_end(map, entry, end); 2352 2353 next = entry->next; 2354 2355 /* 2356 * Unwire before removing addresses from the pmap; otherwise, 2357 * unwiring will put the entries back in the pmap. 2358 */ 2359 if (entry->wired_count != 0) { 2360 vm_map_entry_unwire(map, entry); 2361 } 2362 2363 pmap_remove(map->pmap, entry->start, entry->end); 2364 2365 /* 2366 * Delete the entry (which may delete the object) only after 2367 * removing all pmap entries pointing to its pages. 2368 * (Otherwise, its page frames may be reallocated, and any 2369 * modify bits will be set in the wrong object!) 2370 */ 2371 vm_map_entry_delete(map, entry); 2372 entry = next; 2373 } 2374 return (KERN_SUCCESS); 2375 } 2376 2377 /* 2378 * vm_map_remove: 2379 * 2380 * Remove the given address range from the target map. 2381 * This is the exported form of vm_map_delete. 2382 */ 2383 int 2384 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2385 { 2386 int result; 2387 2388 vm_map_lock(map); 2389 VM_MAP_RANGE_CHECK(map, start, end); 2390 result = vm_map_delete(map, start, end); 2391 vm_map_unlock(map); 2392 return (result); 2393 } 2394 2395 /* 2396 * vm_map_check_protection: 2397 * 2398 * Assert that the target map allows the specified privilege on the 2399 * entire address region given. The entire region must be allocated. 2400 * 2401 * WARNING! This code does not and should not check whether the 2402 * contents of the region is accessible. For example a smaller file 2403 * might be mapped into a larger address space. 2404 * 2405 * NOTE! This code is also called by munmap(). 2406 * 2407 * The map must be locked. A read lock is sufficient. 2408 */ 2409 boolean_t 2410 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2411 vm_prot_t protection) 2412 { 2413 vm_map_entry_t entry; 2414 vm_map_entry_t tmp_entry; 2415 2416 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2417 return (FALSE); 2418 entry = tmp_entry; 2419 2420 while (start < end) { 2421 if (entry == &map->header) 2422 return (FALSE); 2423 /* 2424 * No holes allowed! 2425 */ 2426 if (start < entry->start) 2427 return (FALSE); 2428 /* 2429 * Check protection associated with entry. 2430 */ 2431 if ((entry->protection & protection) != protection) 2432 return (FALSE); 2433 /* go to next entry */ 2434 start = entry->end; 2435 entry = entry->next; 2436 } 2437 return (TRUE); 2438 } 2439 2440 /* 2441 * vm_map_copy_entry: 2442 * 2443 * Copies the contents of the source entry to the destination 2444 * entry. The entries *must* be aligned properly. 2445 */ 2446 static void 2447 vm_map_copy_entry( 2448 vm_map_t src_map, 2449 vm_map_t dst_map, 2450 vm_map_entry_t src_entry, 2451 vm_map_entry_t dst_entry) 2452 { 2453 vm_object_t src_object; 2454 2455 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2456 return; 2457 2458 if (src_entry->wired_count == 0) { 2459 2460 /* 2461 * If the source entry is marked needs_copy, it is already 2462 * write-protected. 2463 */ 2464 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2465 pmap_protect(src_map->pmap, 2466 src_entry->start, 2467 src_entry->end, 2468 src_entry->protection & ~VM_PROT_WRITE); 2469 } 2470 2471 /* 2472 * Make a copy of the object. 2473 */ 2474 if ((src_object = src_entry->object.vm_object) != NULL) { 2475 VM_OBJECT_LOCK(src_object); 2476 if ((src_object->handle == NULL) && 2477 (src_object->type == OBJT_DEFAULT || 2478 src_object->type == OBJT_SWAP)) { 2479 vm_object_collapse(src_object); 2480 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2481 vm_object_split(src_entry); 2482 src_object = src_entry->object.vm_object; 2483 } 2484 } 2485 vm_object_reference_locked(src_object); 2486 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2487 VM_OBJECT_UNLOCK(src_object); 2488 dst_entry->object.vm_object = src_object; 2489 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2490 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2491 dst_entry->offset = src_entry->offset; 2492 } else { 2493 dst_entry->object.vm_object = NULL; 2494 dst_entry->offset = 0; 2495 } 2496 2497 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2498 dst_entry->end - dst_entry->start, src_entry->start); 2499 } else { 2500 /* 2501 * Of course, wired down pages can't be set copy-on-write. 2502 * Cause wired pages to be copied into the new map by 2503 * simulating faults (the new pages are pageable) 2504 */ 2505 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2506 } 2507 } 2508 2509 /* 2510 * vmspace_map_entry_forked: 2511 * Update the newly-forked vmspace each time a map entry is inherited 2512 * or copied. The values for vm_dsize and vm_tsize are approximate 2513 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2514 */ 2515 static void 2516 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2517 vm_map_entry_t entry) 2518 { 2519 vm_size_t entrysize; 2520 vm_offset_t newend; 2521 2522 entrysize = entry->end - entry->start; 2523 vm2->vm_map.size += entrysize; 2524 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2525 vm2->vm_ssize += btoc(entrysize); 2526 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2527 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2528 newend = MIN(entry->end, 2529 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2530 vm2->vm_dsize += btoc(newend - entry->start); 2531 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2532 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2533 newend = MIN(entry->end, 2534 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2535 vm2->vm_tsize += btoc(newend - entry->start); 2536 } 2537 } 2538 2539 /* 2540 * vmspace_fork: 2541 * Create a new process vmspace structure and vm_map 2542 * based on those of an existing process. The new map 2543 * is based on the old map, according to the inheritance 2544 * values on the regions in that map. 2545 * 2546 * XXX It might be worth coalescing the entries added to the new vmspace. 2547 * 2548 * The source map must not be locked. 2549 */ 2550 struct vmspace * 2551 vmspace_fork(struct vmspace *vm1) 2552 { 2553 struct vmspace *vm2; 2554 vm_map_t old_map = &vm1->vm_map; 2555 vm_map_t new_map; 2556 vm_map_entry_t old_entry; 2557 vm_map_entry_t new_entry; 2558 vm_object_t object; 2559 2560 vm_map_lock(old_map); 2561 2562 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2563 vm2->vm_taddr = vm1->vm_taddr; 2564 vm2->vm_daddr = vm1->vm_daddr; 2565 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2566 new_map = &vm2->vm_map; /* XXX */ 2567 new_map->timestamp = 1; 2568 2569 /* Do not inherit the MAP_WIREFUTURE property. */ 2570 if ((new_map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE) 2571 new_map->flags &= ~MAP_WIREFUTURE; 2572 2573 old_entry = old_map->header.next; 2574 2575 while (old_entry != &old_map->header) { 2576 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2577 panic("vm_map_fork: encountered a submap"); 2578 2579 switch (old_entry->inheritance) { 2580 case VM_INHERIT_NONE: 2581 break; 2582 2583 case VM_INHERIT_SHARE: 2584 /* 2585 * Clone the entry, creating the shared object if necessary. 2586 */ 2587 object = old_entry->object.vm_object; 2588 if (object == NULL) { 2589 object = vm_object_allocate(OBJT_DEFAULT, 2590 atop(old_entry->end - old_entry->start)); 2591 old_entry->object.vm_object = object; 2592 old_entry->offset = 0; 2593 } 2594 2595 /* 2596 * Add the reference before calling vm_object_shadow 2597 * to insure that a shadow object is created. 2598 */ 2599 vm_object_reference(object); 2600 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 2601 vm_object_shadow(&old_entry->object.vm_object, 2602 &old_entry->offset, 2603 atop(old_entry->end - old_entry->start)); 2604 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 2605 /* Transfer the second reference too. */ 2606 vm_object_reference( 2607 old_entry->object.vm_object); 2608 vm_object_deallocate(object); 2609 object = old_entry->object.vm_object; 2610 } 2611 VM_OBJECT_LOCK(object); 2612 vm_object_clear_flag(object, OBJ_ONEMAPPING); 2613 VM_OBJECT_UNLOCK(object); 2614 2615 /* 2616 * Clone the entry, referencing the shared object. 2617 */ 2618 new_entry = vm_map_entry_create(new_map); 2619 *new_entry = *old_entry; 2620 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2621 new_entry->wired_count = 0; 2622 2623 /* 2624 * Insert the entry into the new map -- we know we're 2625 * inserting at the end of the new map. 2626 */ 2627 vm_map_entry_link(new_map, new_map->header.prev, 2628 new_entry); 2629 vmspace_map_entry_forked(vm1, vm2, new_entry); 2630 2631 /* 2632 * Update the physical map 2633 */ 2634 pmap_copy(new_map->pmap, old_map->pmap, 2635 new_entry->start, 2636 (old_entry->end - old_entry->start), 2637 old_entry->start); 2638 break; 2639 2640 case VM_INHERIT_COPY: 2641 /* 2642 * Clone the entry and link into the map. 2643 */ 2644 new_entry = vm_map_entry_create(new_map); 2645 *new_entry = *old_entry; 2646 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2647 new_entry->wired_count = 0; 2648 new_entry->object.vm_object = NULL; 2649 vm_map_entry_link(new_map, new_map->header.prev, 2650 new_entry); 2651 vmspace_map_entry_forked(vm1, vm2, new_entry); 2652 vm_map_copy_entry(old_map, new_map, old_entry, 2653 new_entry); 2654 break; 2655 } 2656 old_entry = old_entry->next; 2657 } 2658 2659 vm_map_unlock(old_map); 2660 2661 return (vm2); 2662 } 2663 2664 int 2665 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 2666 vm_prot_t prot, vm_prot_t max, int cow) 2667 { 2668 vm_map_entry_t new_entry, prev_entry; 2669 vm_offset_t bot, top; 2670 vm_size_t init_ssize; 2671 int orient, rv; 2672 rlim_t vmemlim; 2673 2674 /* 2675 * The stack orientation is piggybacked with the cow argument. 2676 * Extract it into orient and mask the cow argument so that we 2677 * don't pass it around further. 2678 * NOTE: We explicitly allow bi-directional stacks. 2679 */ 2680 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 2681 cow &= ~orient; 2682 KASSERT(orient != 0, ("No stack grow direction")); 2683 2684 if (addrbos < vm_map_min(map) || addrbos > map->max_offset) 2685 return (KERN_NO_SPACE); 2686 2687 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 2688 2689 PROC_LOCK(curthread->td_proc); 2690 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 2691 PROC_UNLOCK(curthread->td_proc); 2692 2693 vm_map_lock(map); 2694 2695 /* If addr is already mapped, no go */ 2696 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 2697 vm_map_unlock(map); 2698 return (KERN_NO_SPACE); 2699 } 2700 2701 /* If we would blow our VMEM resource limit, no go */ 2702 if (map->size + init_ssize > vmemlim) { 2703 vm_map_unlock(map); 2704 return (KERN_NO_SPACE); 2705 } 2706 2707 /* 2708 * If we can't accomodate max_ssize in the current mapping, no go. 2709 * However, we need to be aware that subsequent user mappings might 2710 * map into the space we have reserved for stack, and currently this 2711 * space is not protected. 2712 * 2713 * Hopefully we will at least detect this condition when we try to 2714 * grow the stack. 2715 */ 2716 if ((prev_entry->next != &map->header) && 2717 (prev_entry->next->start < addrbos + max_ssize)) { 2718 vm_map_unlock(map); 2719 return (KERN_NO_SPACE); 2720 } 2721 2722 /* 2723 * We initially map a stack of only init_ssize. We will grow as 2724 * needed later. Depending on the orientation of the stack (i.e. 2725 * the grow direction) we either map at the top of the range, the 2726 * bottom of the range or in the middle. 2727 * 2728 * Note: we would normally expect prot and max to be VM_PROT_ALL, 2729 * and cow to be 0. Possibly we should eliminate these as input 2730 * parameters, and just pass these values here in the insert call. 2731 */ 2732 if (orient == MAP_STACK_GROWS_DOWN) 2733 bot = addrbos + max_ssize - init_ssize; 2734 else if (orient == MAP_STACK_GROWS_UP) 2735 bot = addrbos; 2736 else 2737 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 2738 top = bot + init_ssize; 2739 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 2740 2741 /* Now set the avail_ssize amount. */ 2742 if (rv == KERN_SUCCESS) { 2743 if (prev_entry != &map->header) 2744 vm_map_clip_end(map, prev_entry, bot); 2745 new_entry = prev_entry->next; 2746 if (new_entry->end != top || new_entry->start != bot) 2747 panic("Bad entry start/end for new stack entry"); 2748 2749 new_entry->avail_ssize = max_ssize - init_ssize; 2750 if (orient & MAP_STACK_GROWS_DOWN) 2751 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2752 if (orient & MAP_STACK_GROWS_UP) 2753 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 2754 } 2755 2756 vm_map_unlock(map); 2757 return (rv); 2758 } 2759 2760 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 2761 * desired address is already mapped, or if we successfully grow 2762 * the stack. Also returns KERN_SUCCESS if addr is outside the 2763 * stack range (this is strange, but preserves compatibility with 2764 * the grow function in vm_machdep.c). 2765 */ 2766 int 2767 vm_map_growstack(struct proc *p, vm_offset_t addr) 2768 { 2769 vm_map_entry_t next_entry, prev_entry; 2770 vm_map_entry_t new_entry, stack_entry; 2771 struct vmspace *vm = p->p_vmspace; 2772 vm_map_t map = &vm->vm_map; 2773 vm_offset_t end; 2774 size_t grow_amount, max_grow; 2775 rlim_t stacklim, vmemlim; 2776 int is_procstack, rv; 2777 2778 Retry: 2779 PROC_LOCK(p); 2780 stacklim = lim_cur(p, RLIMIT_STACK); 2781 vmemlim = lim_cur(p, RLIMIT_VMEM); 2782 PROC_UNLOCK(p); 2783 2784 vm_map_lock_read(map); 2785 2786 /* If addr is already in the entry range, no need to grow.*/ 2787 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 2788 vm_map_unlock_read(map); 2789 return (KERN_SUCCESS); 2790 } 2791 2792 next_entry = prev_entry->next; 2793 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 2794 /* 2795 * This entry does not grow upwards. Since the address lies 2796 * beyond this entry, the next entry (if one exists) has to 2797 * be a downward growable entry. The entry list header is 2798 * never a growable entry, so it suffices to check the flags. 2799 */ 2800 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 2801 vm_map_unlock_read(map); 2802 return (KERN_SUCCESS); 2803 } 2804 stack_entry = next_entry; 2805 } else { 2806 /* 2807 * This entry grows upward. If the next entry does not at 2808 * least grow downwards, this is the entry we need to grow. 2809 * otherwise we have two possible choices and we have to 2810 * select one. 2811 */ 2812 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 2813 /* 2814 * We have two choices; grow the entry closest to 2815 * the address to minimize the amount of growth. 2816 */ 2817 if (addr - prev_entry->end <= next_entry->start - addr) 2818 stack_entry = prev_entry; 2819 else 2820 stack_entry = next_entry; 2821 } else 2822 stack_entry = prev_entry; 2823 } 2824 2825 if (stack_entry == next_entry) { 2826 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 2827 KASSERT(addr < stack_entry->start, ("foo")); 2828 end = (prev_entry != &map->header) ? prev_entry->end : 2829 stack_entry->start - stack_entry->avail_ssize; 2830 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 2831 max_grow = stack_entry->start - end; 2832 } else { 2833 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 2834 KASSERT(addr >= stack_entry->end, ("foo")); 2835 end = (next_entry != &map->header) ? next_entry->start : 2836 stack_entry->end + stack_entry->avail_ssize; 2837 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 2838 max_grow = end - stack_entry->end; 2839 } 2840 2841 if (grow_amount > stack_entry->avail_ssize) { 2842 vm_map_unlock_read(map); 2843 return (KERN_NO_SPACE); 2844 } 2845 2846 /* 2847 * If there is no longer enough space between the entries nogo, and 2848 * adjust the available space. Note: this should only happen if the 2849 * user has mapped into the stack area after the stack was created, 2850 * and is probably an error. 2851 * 2852 * This also effectively destroys any guard page the user might have 2853 * intended by limiting the stack size. 2854 */ 2855 if (grow_amount > max_grow) { 2856 if (vm_map_lock_upgrade(map)) 2857 goto Retry; 2858 2859 stack_entry->avail_ssize = max_grow; 2860 2861 vm_map_unlock(map); 2862 return (KERN_NO_SPACE); 2863 } 2864 2865 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 2866 2867 /* 2868 * If this is the main process stack, see if we're over the stack 2869 * limit. 2870 */ 2871 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2872 vm_map_unlock_read(map); 2873 return (KERN_NO_SPACE); 2874 } 2875 2876 /* Round up the grow amount modulo SGROWSIZ */ 2877 grow_amount = roundup (grow_amount, sgrowsiz); 2878 if (grow_amount > stack_entry->avail_ssize) 2879 grow_amount = stack_entry->avail_ssize; 2880 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2881 grow_amount = stacklim - ctob(vm->vm_ssize); 2882 } 2883 2884 /* If we would blow our VMEM resource limit, no go */ 2885 if (map->size + grow_amount > vmemlim) { 2886 vm_map_unlock_read(map); 2887 return (KERN_NO_SPACE); 2888 } 2889 2890 if (vm_map_lock_upgrade(map)) 2891 goto Retry; 2892 2893 if (stack_entry == next_entry) { 2894 /* 2895 * Growing downward. 2896 */ 2897 /* Get the preliminary new entry start value */ 2898 addr = stack_entry->start - grow_amount; 2899 2900 /* 2901 * If this puts us into the previous entry, cut back our 2902 * growth to the available space. Also, see the note above. 2903 */ 2904 if (addr < end) { 2905 stack_entry->avail_ssize = max_grow; 2906 addr = end; 2907 } 2908 2909 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 2910 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 2911 2912 /* Adjust the available stack space by the amount we grew. */ 2913 if (rv == KERN_SUCCESS) { 2914 if (prev_entry != &map->header) 2915 vm_map_clip_end(map, prev_entry, addr); 2916 new_entry = prev_entry->next; 2917 KASSERT(new_entry == stack_entry->prev, ("foo")); 2918 KASSERT(new_entry->end == stack_entry->start, ("foo")); 2919 KASSERT(new_entry->start == addr, ("foo")); 2920 grow_amount = new_entry->end - new_entry->start; 2921 new_entry->avail_ssize = stack_entry->avail_ssize - 2922 grow_amount; 2923 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 2924 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2925 } 2926 } else { 2927 /* 2928 * Growing upward. 2929 */ 2930 addr = stack_entry->end + grow_amount; 2931 2932 /* 2933 * If this puts us into the next entry, cut back our growth 2934 * to the available space. Also, see the note above. 2935 */ 2936 if (addr > end) { 2937 stack_entry->avail_ssize = end - stack_entry->end; 2938 addr = end; 2939 } 2940 2941 grow_amount = addr - stack_entry->end; 2942 2943 /* Grow the underlying object if applicable. */ 2944 if (stack_entry->object.vm_object == NULL || 2945 vm_object_coalesce(stack_entry->object.vm_object, 2946 stack_entry->offset, 2947 (vm_size_t)(stack_entry->end - stack_entry->start), 2948 (vm_size_t)grow_amount)) { 2949 map->size += (addr - stack_entry->end); 2950 /* Update the current entry. */ 2951 stack_entry->end = addr; 2952 stack_entry->avail_ssize -= grow_amount; 2953 vm_map_entry_resize_free(map, stack_entry); 2954 rv = KERN_SUCCESS; 2955 2956 if (next_entry != &map->header) 2957 vm_map_clip_start(map, next_entry, addr); 2958 } else 2959 rv = KERN_FAILURE; 2960 } 2961 2962 if (rv == KERN_SUCCESS && is_procstack) 2963 vm->vm_ssize += btoc(grow_amount); 2964 2965 vm_map_unlock(map); 2966 2967 /* 2968 * Heed the MAP_WIREFUTURE flag if it was set for this process. 2969 */ 2970 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 2971 vm_map_wire(map, 2972 (stack_entry == next_entry) ? addr : addr - grow_amount, 2973 (stack_entry == next_entry) ? stack_entry->start : addr, 2974 (p->p_flag & P_SYSTEM) 2975 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 2976 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 2977 } 2978 2979 return (rv); 2980 } 2981 2982 /* 2983 * Unshare the specified VM space for exec. If other processes are 2984 * mapped to it, then create a new one. The new vmspace is null. 2985 */ 2986 void 2987 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 2988 { 2989 struct vmspace *oldvmspace = p->p_vmspace; 2990 struct vmspace *newvmspace; 2991 2992 newvmspace = vmspace_alloc(minuser, maxuser); 2993 newvmspace->vm_swrss = oldvmspace->vm_swrss; 2994 /* 2995 * This code is written like this for prototype purposes. The 2996 * goal is to avoid running down the vmspace here, but let the 2997 * other process's that are still using the vmspace to finally 2998 * run it down. Even though there is little or no chance of blocking 2999 * here, it is a good idea to keep this form for future mods. 3000 */ 3001 PROC_VMSPACE_LOCK(p); 3002 p->p_vmspace = newvmspace; 3003 PROC_VMSPACE_UNLOCK(p); 3004 if (p == curthread->td_proc) /* XXXKSE ? */ 3005 pmap_activate(curthread); 3006 vmspace_free(oldvmspace); 3007 } 3008 3009 /* 3010 * Unshare the specified VM space for forcing COW. This 3011 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3012 */ 3013 void 3014 vmspace_unshare(struct proc *p) 3015 { 3016 struct vmspace *oldvmspace = p->p_vmspace; 3017 struct vmspace *newvmspace; 3018 3019 if (oldvmspace->vm_refcnt == 1) 3020 return; 3021 newvmspace = vmspace_fork(oldvmspace); 3022 PROC_VMSPACE_LOCK(p); 3023 p->p_vmspace = newvmspace; 3024 PROC_VMSPACE_UNLOCK(p); 3025 if (p == curthread->td_proc) /* XXXKSE ? */ 3026 pmap_activate(curthread); 3027 vmspace_free(oldvmspace); 3028 } 3029 3030 /* 3031 * vm_map_lookup: 3032 * 3033 * Finds the VM object, offset, and 3034 * protection for a given virtual address in the 3035 * specified map, assuming a page fault of the 3036 * type specified. 3037 * 3038 * Leaves the map in question locked for read; return 3039 * values are guaranteed until a vm_map_lookup_done 3040 * call is performed. Note that the map argument 3041 * is in/out; the returned map must be used in 3042 * the call to vm_map_lookup_done. 3043 * 3044 * A handle (out_entry) is returned for use in 3045 * vm_map_lookup_done, to make that fast. 3046 * 3047 * If a lookup is requested with "write protection" 3048 * specified, the map may be changed to perform virtual 3049 * copying operations, although the data referenced will 3050 * remain the same. 3051 */ 3052 int 3053 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3054 vm_offset_t vaddr, 3055 vm_prot_t fault_typea, 3056 vm_map_entry_t *out_entry, /* OUT */ 3057 vm_object_t *object, /* OUT */ 3058 vm_pindex_t *pindex, /* OUT */ 3059 vm_prot_t *out_prot, /* OUT */ 3060 boolean_t *wired) /* OUT */ 3061 { 3062 vm_map_entry_t entry; 3063 vm_map_t map = *var_map; 3064 vm_prot_t prot; 3065 vm_prot_t fault_type = fault_typea; 3066 3067 RetryLookup:; 3068 /* 3069 * Lookup the faulting address. 3070 */ 3071 3072 vm_map_lock_read(map); 3073 #define RETURN(why) \ 3074 { \ 3075 vm_map_unlock_read(map); \ 3076 return (why); \ 3077 } 3078 3079 /* 3080 * If the map has an interesting hint, try it before calling full 3081 * blown lookup routine. 3082 */ 3083 entry = map->root; 3084 *out_entry = entry; 3085 if (entry == NULL || 3086 (vaddr < entry->start) || (vaddr >= entry->end)) { 3087 /* 3088 * Entry was either not a valid hint, or the vaddr was not 3089 * contained in the entry, so do a full lookup. 3090 */ 3091 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3092 RETURN(KERN_INVALID_ADDRESS); 3093 3094 entry = *out_entry; 3095 } 3096 3097 /* 3098 * Handle submaps. 3099 */ 3100 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3101 vm_map_t old_map = map; 3102 3103 *var_map = map = entry->object.sub_map; 3104 vm_map_unlock_read(old_map); 3105 goto RetryLookup; 3106 } 3107 3108 /* 3109 * Check whether this task is allowed to have this page. 3110 * Note the special case for MAP_ENTRY_COW 3111 * pages with an override. This is to implement a forced 3112 * COW for debuggers. 3113 */ 3114 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3115 prot = entry->max_protection; 3116 else 3117 prot = entry->protection; 3118 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3119 if ((fault_type & prot) != fault_type) { 3120 RETURN(KERN_PROTECTION_FAILURE); 3121 } 3122 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3123 (entry->eflags & MAP_ENTRY_COW) && 3124 (fault_type & VM_PROT_WRITE) && 3125 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3126 RETURN(KERN_PROTECTION_FAILURE); 3127 } 3128 3129 /* 3130 * If this page is not pageable, we have to get it for all possible 3131 * accesses. 3132 */ 3133 *wired = (entry->wired_count != 0); 3134 if (*wired) 3135 prot = fault_type = entry->protection; 3136 3137 /* 3138 * If the entry was copy-on-write, we either ... 3139 */ 3140 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3141 /* 3142 * If we want to write the page, we may as well handle that 3143 * now since we've got the map locked. 3144 * 3145 * If we don't need to write the page, we just demote the 3146 * permissions allowed. 3147 */ 3148 if (fault_type & VM_PROT_WRITE) { 3149 /* 3150 * Make a new object, and place it in the object 3151 * chain. Note that no new references have appeared 3152 * -- one just moved from the map to the new 3153 * object. 3154 */ 3155 if (vm_map_lock_upgrade(map)) 3156 goto RetryLookup; 3157 3158 vm_object_shadow( 3159 &entry->object.vm_object, 3160 &entry->offset, 3161 atop(entry->end - entry->start)); 3162 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3163 3164 vm_map_lock_downgrade(map); 3165 } else { 3166 /* 3167 * We're attempting to read a copy-on-write page -- 3168 * don't allow writes. 3169 */ 3170 prot &= ~VM_PROT_WRITE; 3171 } 3172 } 3173 3174 /* 3175 * Create an object if necessary. 3176 */ 3177 if (entry->object.vm_object == NULL && 3178 !map->system_map) { 3179 if (vm_map_lock_upgrade(map)) 3180 goto RetryLookup; 3181 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3182 atop(entry->end - entry->start)); 3183 entry->offset = 0; 3184 vm_map_lock_downgrade(map); 3185 } 3186 3187 /* 3188 * Return the object/offset from this entry. If the entry was 3189 * copy-on-write or empty, it has been fixed up. 3190 */ 3191 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3192 *object = entry->object.vm_object; 3193 3194 *out_prot = prot; 3195 return (KERN_SUCCESS); 3196 3197 #undef RETURN 3198 } 3199 3200 /* 3201 * vm_map_lookup_locked: 3202 * 3203 * Lookup the faulting address. A version of vm_map_lookup that returns 3204 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3205 */ 3206 int 3207 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3208 vm_offset_t vaddr, 3209 vm_prot_t fault_typea, 3210 vm_map_entry_t *out_entry, /* OUT */ 3211 vm_object_t *object, /* OUT */ 3212 vm_pindex_t *pindex, /* OUT */ 3213 vm_prot_t *out_prot, /* OUT */ 3214 boolean_t *wired) /* OUT */ 3215 { 3216 vm_map_entry_t entry; 3217 vm_map_t map = *var_map; 3218 vm_prot_t prot; 3219 vm_prot_t fault_type = fault_typea; 3220 3221 /* 3222 * If the map has an interesting hint, try it before calling full 3223 * blown lookup routine. 3224 */ 3225 entry = map->root; 3226 *out_entry = entry; 3227 if (entry == NULL || 3228 (vaddr < entry->start) || (vaddr >= entry->end)) { 3229 /* 3230 * Entry was either not a valid hint, or the vaddr was not 3231 * contained in the entry, so do a full lookup. 3232 */ 3233 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3234 return (KERN_INVALID_ADDRESS); 3235 3236 entry = *out_entry; 3237 } 3238 3239 /* 3240 * Fail if the entry refers to a submap. 3241 */ 3242 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3243 return (KERN_FAILURE); 3244 3245 /* 3246 * Check whether this task is allowed to have this page. 3247 * Note the special case for MAP_ENTRY_COW 3248 * pages with an override. This is to implement a forced 3249 * COW for debuggers. 3250 */ 3251 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3252 prot = entry->max_protection; 3253 else 3254 prot = entry->protection; 3255 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3256 if ((fault_type & prot) != fault_type) 3257 return (KERN_PROTECTION_FAILURE); 3258 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3259 (entry->eflags & MAP_ENTRY_COW) && 3260 (fault_type & VM_PROT_WRITE) && 3261 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3262 return (KERN_PROTECTION_FAILURE); 3263 3264 /* 3265 * If this page is not pageable, we have to get it for all possible 3266 * accesses. 3267 */ 3268 *wired = (entry->wired_count != 0); 3269 if (*wired) 3270 prot = fault_type = entry->protection; 3271 3272 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3273 /* 3274 * Fail if the entry was copy-on-write for a write fault. 3275 */ 3276 if (fault_type & VM_PROT_WRITE) 3277 return (KERN_FAILURE); 3278 /* 3279 * We're attempting to read a copy-on-write page -- 3280 * don't allow writes. 3281 */ 3282 prot &= ~VM_PROT_WRITE; 3283 } 3284 3285 /* 3286 * Fail if an object should be created. 3287 */ 3288 if (entry->object.vm_object == NULL && !map->system_map) 3289 return (KERN_FAILURE); 3290 3291 /* 3292 * Return the object/offset from this entry. If the entry was 3293 * copy-on-write or empty, it has been fixed up. 3294 */ 3295 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3296 *object = entry->object.vm_object; 3297 3298 *out_prot = prot; 3299 return (KERN_SUCCESS); 3300 } 3301 3302 /* 3303 * vm_map_lookup_done: 3304 * 3305 * Releases locks acquired by a vm_map_lookup 3306 * (according to the handle returned by that lookup). 3307 */ 3308 void 3309 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3310 { 3311 /* 3312 * Unlock the main-level map 3313 */ 3314 vm_map_unlock_read(map); 3315 } 3316 3317 #include "opt_ddb.h" 3318 #ifdef DDB 3319 #include <sys/kernel.h> 3320 3321 #include <ddb/ddb.h> 3322 3323 /* 3324 * vm_map_print: [ debug ] 3325 */ 3326 DB_SHOW_COMMAND(map, vm_map_print) 3327 { 3328 static int nlines; 3329 /* XXX convert args. */ 3330 vm_map_t map = (vm_map_t)addr; 3331 boolean_t full = have_addr; 3332 3333 vm_map_entry_t entry; 3334 3335 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3336 (void *)map, 3337 (void *)map->pmap, map->nentries, map->timestamp); 3338 nlines++; 3339 3340 if (!full && db_indent) 3341 return; 3342 3343 db_indent += 2; 3344 for (entry = map->header.next; entry != &map->header; 3345 entry = entry->next) { 3346 db_iprintf("map entry %p: start=%p, end=%p\n", 3347 (void *)entry, (void *)entry->start, (void *)entry->end); 3348 nlines++; 3349 { 3350 static char *inheritance_name[4] = 3351 {"share", "copy", "none", "donate_copy"}; 3352 3353 db_iprintf(" prot=%x/%x/%s", 3354 entry->protection, 3355 entry->max_protection, 3356 inheritance_name[(int)(unsigned char)entry->inheritance]); 3357 if (entry->wired_count != 0) 3358 db_printf(", wired"); 3359 } 3360 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3361 db_printf(", share=%p, offset=0x%jx\n", 3362 (void *)entry->object.sub_map, 3363 (uintmax_t)entry->offset); 3364 nlines++; 3365 if ((entry->prev == &map->header) || 3366 (entry->prev->object.sub_map != 3367 entry->object.sub_map)) { 3368 db_indent += 2; 3369 vm_map_print((db_expr_t)(intptr_t) 3370 entry->object.sub_map, 3371 full, 0, (char *)0); 3372 db_indent -= 2; 3373 } 3374 } else { 3375 db_printf(", object=%p, offset=0x%jx", 3376 (void *)entry->object.vm_object, 3377 (uintmax_t)entry->offset); 3378 if (entry->eflags & MAP_ENTRY_COW) 3379 db_printf(", copy (%s)", 3380 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3381 db_printf("\n"); 3382 nlines++; 3383 3384 if ((entry->prev == &map->header) || 3385 (entry->prev->object.vm_object != 3386 entry->object.vm_object)) { 3387 db_indent += 2; 3388 vm_object_print((db_expr_t)(intptr_t) 3389 entry->object.vm_object, 3390 full, 0, (char *)0); 3391 nlines += 4; 3392 db_indent -= 2; 3393 } 3394 } 3395 } 3396 db_indent -= 2; 3397 if (db_indent == 0) 3398 nlines = 0; 3399 } 3400 3401 3402 DB_SHOW_COMMAND(procvm, procvm) 3403 { 3404 struct proc *p; 3405 3406 if (have_addr) { 3407 p = (struct proc *) addr; 3408 } else { 3409 p = curproc; 3410 } 3411 3412 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3413 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3414 (void *)vmspace_pmap(p->p_vmspace)); 3415 3416 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3417 } 3418 3419 #endif /* DDB */ 3420