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 (void)_sx_xlock(&map->lock, 0, 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 (void)_sx_xlock(&map->lock, 0, 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's resident pages into 1433 * the given map. This eliminates the soft faults on process startup and 1434 * immediately after an mmap(2). Unless the given flags include 1435 * MAP_PREFAULT_MADVISE, cached pages are not reactivated and mapped. 1436 */ 1437 void 1438 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1439 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1440 { 1441 vm_offset_t start; 1442 vm_page_t p, p_start; 1443 vm_pindex_t psize, tmpidx; 1444 boolean_t are_queues_locked; 1445 1446 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1447 return; 1448 VM_OBJECT_LOCK(object); 1449 if (object->type == OBJT_DEVICE) { 1450 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1451 goto unlock_return; 1452 } 1453 1454 psize = atop(size); 1455 1456 if (object->type != OBJT_VNODE || 1457 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 1458 (object->resident_page_count > MAX_INIT_PT))) { 1459 goto unlock_return; 1460 } 1461 1462 if (psize + pindex > object->size) { 1463 if (object->size < pindex) 1464 goto unlock_return; 1465 psize = object->size - pindex; 1466 } 1467 1468 are_queues_locked = FALSE; 1469 start = 0; 1470 p_start = NULL; 1471 1472 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1473 if (p->pindex < pindex) { 1474 p = vm_page_splay(pindex, object->root); 1475 if ((object->root = p)->pindex < pindex) 1476 p = TAILQ_NEXT(p, listq); 1477 } 1478 } 1479 /* 1480 * Assert: the variable p is either (1) the page with the 1481 * least pindex greater than or equal to the parameter pindex 1482 * or (2) NULL. 1483 */ 1484 for (; 1485 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1486 p = TAILQ_NEXT(p, listq)) { 1487 /* 1488 * don't allow an madvise to blow away our really 1489 * free pages allocating pv entries. 1490 */ 1491 if ((flags & MAP_PREFAULT_MADVISE) && 1492 cnt.v_free_count < cnt.v_free_reserved) { 1493 psize = tmpidx; 1494 break; 1495 } 1496 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 1497 (p->busy == 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 if ((flags & MAP_PREFAULT_MADVISE) != 0) 1508 vm_page_deactivate(p); 1509 else if (p_start != NULL) { 1510 pmap_enter_object(map->pmap, start, addr + 1511 ptoa(tmpidx), p_start, prot); 1512 p_start = NULL; 1513 } 1514 } 1515 } else if (p_start != NULL) { 1516 pmap_enter_object(map->pmap, start, addr + 1517 ptoa(tmpidx), p_start, prot); 1518 p_start = NULL; 1519 } 1520 } 1521 if (p_start != NULL) 1522 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1523 p_start, prot); 1524 if (are_queues_locked) 1525 vm_page_unlock_queues(); 1526 unlock_return: 1527 VM_OBJECT_UNLOCK(object); 1528 } 1529 1530 /* 1531 * vm_map_protect: 1532 * 1533 * Sets the protection of the specified address 1534 * region in the target map. If "set_max" is 1535 * specified, the maximum protection is to be set; 1536 * otherwise, only the current protection is affected. 1537 */ 1538 int 1539 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1540 vm_prot_t new_prot, boolean_t set_max) 1541 { 1542 vm_map_entry_t current; 1543 vm_map_entry_t entry; 1544 1545 vm_map_lock(map); 1546 1547 VM_MAP_RANGE_CHECK(map, start, end); 1548 1549 if (vm_map_lookup_entry(map, start, &entry)) { 1550 vm_map_clip_start(map, entry, start); 1551 } else { 1552 entry = entry->next; 1553 } 1554 1555 /* 1556 * Make a first pass to check for protection violations. 1557 */ 1558 current = entry; 1559 while ((current != &map->header) && (current->start < end)) { 1560 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1561 vm_map_unlock(map); 1562 return (KERN_INVALID_ARGUMENT); 1563 } 1564 if ((new_prot & current->max_protection) != new_prot) { 1565 vm_map_unlock(map); 1566 return (KERN_PROTECTION_FAILURE); 1567 } 1568 current = current->next; 1569 } 1570 1571 /* 1572 * Go back and fix up protections. [Note that clipping is not 1573 * necessary the second time.] 1574 */ 1575 current = entry; 1576 while ((current != &map->header) && (current->start < end)) { 1577 vm_prot_t old_prot; 1578 1579 vm_map_clip_end(map, current, end); 1580 1581 old_prot = current->protection; 1582 if (set_max) 1583 current->protection = 1584 (current->max_protection = new_prot) & 1585 old_prot; 1586 else 1587 current->protection = new_prot; 1588 1589 /* 1590 * Update physical map if necessary. Worry about copy-on-write 1591 * here -- CHECK THIS XXX 1592 */ 1593 if (current->protection != old_prot) { 1594 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1595 VM_PROT_ALL) 1596 pmap_protect(map->pmap, current->start, 1597 current->end, 1598 current->protection & MASK(current)); 1599 #undef MASK 1600 } 1601 vm_map_simplify_entry(map, current); 1602 current = current->next; 1603 } 1604 vm_map_unlock(map); 1605 return (KERN_SUCCESS); 1606 } 1607 1608 /* 1609 * vm_map_madvise: 1610 * 1611 * This routine traverses a processes map handling the madvise 1612 * system call. Advisories are classified as either those effecting 1613 * the vm_map_entry structure, or those effecting the underlying 1614 * objects. 1615 */ 1616 int 1617 vm_map_madvise( 1618 vm_map_t map, 1619 vm_offset_t start, 1620 vm_offset_t end, 1621 int behav) 1622 { 1623 vm_map_entry_t current, entry; 1624 int modify_map = 0; 1625 1626 /* 1627 * Some madvise calls directly modify the vm_map_entry, in which case 1628 * we need to use an exclusive lock on the map and we need to perform 1629 * various clipping operations. Otherwise we only need a read-lock 1630 * on the map. 1631 */ 1632 switch(behav) { 1633 case MADV_NORMAL: 1634 case MADV_SEQUENTIAL: 1635 case MADV_RANDOM: 1636 case MADV_NOSYNC: 1637 case MADV_AUTOSYNC: 1638 case MADV_NOCORE: 1639 case MADV_CORE: 1640 modify_map = 1; 1641 vm_map_lock(map); 1642 break; 1643 case MADV_WILLNEED: 1644 case MADV_DONTNEED: 1645 case MADV_FREE: 1646 vm_map_lock_read(map); 1647 break; 1648 default: 1649 return (KERN_INVALID_ARGUMENT); 1650 } 1651 1652 /* 1653 * Locate starting entry and clip if necessary. 1654 */ 1655 VM_MAP_RANGE_CHECK(map, start, end); 1656 1657 if (vm_map_lookup_entry(map, start, &entry)) { 1658 if (modify_map) 1659 vm_map_clip_start(map, entry, start); 1660 } else { 1661 entry = entry->next; 1662 } 1663 1664 if (modify_map) { 1665 /* 1666 * madvise behaviors that are implemented in the vm_map_entry. 1667 * 1668 * We clip the vm_map_entry so that behavioral changes are 1669 * limited to the specified address range. 1670 */ 1671 for (current = entry; 1672 (current != &map->header) && (current->start < end); 1673 current = current->next 1674 ) { 1675 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1676 continue; 1677 1678 vm_map_clip_end(map, current, end); 1679 1680 switch (behav) { 1681 case MADV_NORMAL: 1682 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 1683 break; 1684 case MADV_SEQUENTIAL: 1685 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 1686 break; 1687 case MADV_RANDOM: 1688 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 1689 break; 1690 case MADV_NOSYNC: 1691 current->eflags |= MAP_ENTRY_NOSYNC; 1692 break; 1693 case MADV_AUTOSYNC: 1694 current->eflags &= ~MAP_ENTRY_NOSYNC; 1695 break; 1696 case MADV_NOCORE: 1697 current->eflags |= MAP_ENTRY_NOCOREDUMP; 1698 break; 1699 case MADV_CORE: 1700 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 1701 break; 1702 default: 1703 break; 1704 } 1705 vm_map_simplify_entry(map, current); 1706 } 1707 vm_map_unlock(map); 1708 } else { 1709 vm_pindex_t pindex; 1710 int count; 1711 1712 /* 1713 * madvise behaviors that are implemented in the underlying 1714 * vm_object. 1715 * 1716 * Since we don't clip the vm_map_entry, we have to clip 1717 * the vm_object pindex and count. 1718 */ 1719 for (current = entry; 1720 (current != &map->header) && (current->start < end); 1721 current = current->next 1722 ) { 1723 vm_offset_t useStart; 1724 1725 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1726 continue; 1727 1728 pindex = OFF_TO_IDX(current->offset); 1729 count = atop(current->end - current->start); 1730 useStart = current->start; 1731 1732 if (current->start < start) { 1733 pindex += atop(start - current->start); 1734 count -= atop(start - current->start); 1735 useStart = start; 1736 } 1737 if (current->end > end) 1738 count -= atop(current->end - end); 1739 1740 if (count <= 0) 1741 continue; 1742 1743 vm_object_madvise(current->object.vm_object, 1744 pindex, count, behav); 1745 if (behav == MADV_WILLNEED) { 1746 vm_map_pmap_enter(map, 1747 useStart, 1748 current->protection, 1749 current->object.vm_object, 1750 pindex, 1751 (count << PAGE_SHIFT), 1752 MAP_PREFAULT_MADVISE 1753 ); 1754 } 1755 } 1756 vm_map_unlock_read(map); 1757 } 1758 return (0); 1759 } 1760 1761 1762 /* 1763 * vm_map_inherit: 1764 * 1765 * Sets the inheritance of the specified address 1766 * range in the target map. Inheritance 1767 * affects how the map will be shared with 1768 * child maps at the time of vm_map_fork. 1769 */ 1770 int 1771 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 1772 vm_inherit_t new_inheritance) 1773 { 1774 vm_map_entry_t entry; 1775 vm_map_entry_t temp_entry; 1776 1777 switch (new_inheritance) { 1778 case VM_INHERIT_NONE: 1779 case VM_INHERIT_COPY: 1780 case VM_INHERIT_SHARE: 1781 break; 1782 default: 1783 return (KERN_INVALID_ARGUMENT); 1784 } 1785 vm_map_lock(map); 1786 VM_MAP_RANGE_CHECK(map, start, end); 1787 if (vm_map_lookup_entry(map, start, &temp_entry)) { 1788 entry = temp_entry; 1789 vm_map_clip_start(map, entry, start); 1790 } else 1791 entry = temp_entry->next; 1792 while ((entry != &map->header) && (entry->start < end)) { 1793 vm_map_clip_end(map, entry, end); 1794 entry->inheritance = new_inheritance; 1795 vm_map_simplify_entry(map, entry); 1796 entry = entry->next; 1797 } 1798 vm_map_unlock(map); 1799 return (KERN_SUCCESS); 1800 } 1801 1802 /* 1803 * vm_map_unwire: 1804 * 1805 * Implements both kernel and user unwiring. 1806 */ 1807 int 1808 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1809 int flags) 1810 { 1811 vm_map_entry_t entry, first_entry, tmp_entry; 1812 vm_offset_t saved_start; 1813 unsigned int last_timestamp; 1814 int rv; 1815 boolean_t need_wakeup, result, user_unwire; 1816 1817 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1818 vm_map_lock(map); 1819 VM_MAP_RANGE_CHECK(map, start, end); 1820 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1821 if (flags & VM_MAP_WIRE_HOLESOK) 1822 first_entry = first_entry->next; 1823 else { 1824 vm_map_unlock(map); 1825 return (KERN_INVALID_ADDRESS); 1826 } 1827 } 1828 last_timestamp = map->timestamp; 1829 entry = first_entry; 1830 while (entry != &map->header && entry->start < end) { 1831 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1832 /* 1833 * We have not yet clipped the entry. 1834 */ 1835 saved_start = (start >= entry->start) ? start : 1836 entry->start; 1837 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1838 if (vm_map_unlock_and_wait(map, user_unwire)) { 1839 /* 1840 * Allow interruption of user unwiring? 1841 */ 1842 } 1843 vm_map_lock(map); 1844 if (last_timestamp+1 != map->timestamp) { 1845 /* 1846 * Look again for the entry because the map was 1847 * modified while it was unlocked. 1848 * Specifically, the entry may have been 1849 * clipped, merged, or deleted. 1850 */ 1851 if (!vm_map_lookup_entry(map, saved_start, 1852 &tmp_entry)) { 1853 if (flags & VM_MAP_WIRE_HOLESOK) 1854 tmp_entry = tmp_entry->next; 1855 else { 1856 if (saved_start == start) { 1857 /* 1858 * First_entry has been deleted. 1859 */ 1860 vm_map_unlock(map); 1861 return (KERN_INVALID_ADDRESS); 1862 } 1863 end = saved_start; 1864 rv = KERN_INVALID_ADDRESS; 1865 goto done; 1866 } 1867 } 1868 if (entry == first_entry) 1869 first_entry = tmp_entry; 1870 else 1871 first_entry = NULL; 1872 entry = tmp_entry; 1873 } 1874 last_timestamp = map->timestamp; 1875 continue; 1876 } 1877 vm_map_clip_start(map, entry, start); 1878 vm_map_clip_end(map, entry, end); 1879 /* 1880 * Mark the entry in case the map lock is released. (See 1881 * above.) 1882 */ 1883 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1884 /* 1885 * Check the map for holes in the specified region. 1886 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 1887 */ 1888 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 1889 (entry->end < end && (entry->next == &map->header || 1890 entry->next->start > entry->end))) { 1891 end = entry->end; 1892 rv = KERN_INVALID_ADDRESS; 1893 goto done; 1894 } 1895 /* 1896 * If system unwiring, require that the entry is system wired. 1897 */ 1898 if (!user_unwire && 1899 vm_map_entry_system_wired_count(entry) == 0) { 1900 end = entry->end; 1901 rv = KERN_INVALID_ARGUMENT; 1902 goto done; 1903 } 1904 entry = entry->next; 1905 } 1906 rv = KERN_SUCCESS; 1907 done: 1908 need_wakeup = FALSE; 1909 if (first_entry == NULL) { 1910 result = vm_map_lookup_entry(map, start, &first_entry); 1911 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 1912 first_entry = first_entry->next; 1913 else 1914 KASSERT(result, ("vm_map_unwire: lookup failed")); 1915 } 1916 entry = first_entry; 1917 while (entry != &map->header && entry->start < end) { 1918 if (rv == KERN_SUCCESS && (!user_unwire || 1919 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 1920 if (user_unwire) 1921 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1922 entry->wired_count--; 1923 if (entry->wired_count == 0) { 1924 /* 1925 * Retain the map lock. 1926 */ 1927 vm_fault_unwire(map, entry->start, entry->end, 1928 entry->object.vm_object != NULL && 1929 entry->object.vm_object->type == OBJT_DEVICE); 1930 } 1931 } 1932 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 1933 ("vm_map_unwire: in-transition flag missing")); 1934 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 1935 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 1936 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 1937 need_wakeup = TRUE; 1938 } 1939 vm_map_simplify_entry(map, entry); 1940 entry = entry->next; 1941 } 1942 vm_map_unlock(map); 1943 if (need_wakeup) 1944 vm_map_wakeup(map); 1945 return (rv); 1946 } 1947 1948 /* 1949 * vm_map_wire: 1950 * 1951 * Implements both kernel and user wiring. 1952 */ 1953 int 1954 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1955 int flags) 1956 { 1957 vm_map_entry_t entry, first_entry, tmp_entry; 1958 vm_offset_t saved_end, saved_start; 1959 unsigned int last_timestamp; 1960 int rv; 1961 boolean_t fictitious, need_wakeup, result, user_wire; 1962 1963 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1964 vm_map_lock(map); 1965 VM_MAP_RANGE_CHECK(map, start, end); 1966 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1967 if (flags & VM_MAP_WIRE_HOLESOK) 1968 first_entry = first_entry->next; 1969 else { 1970 vm_map_unlock(map); 1971 return (KERN_INVALID_ADDRESS); 1972 } 1973 } 1974 last_timestamp = map->timestamp; 1975 entry = first_entry; 1976 while (entry != &map->header && entry->start < end) { 1977 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1978 /* 1979 * We have not yet clipped the entry. 1980 */ 1981 saved_start = (start >= entry->start) ? start : 1982 entry->start; 1983 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1984 if (vm_map_unlock_and_wait(map, user_wire)) { 1985 /* 1986 * Allow interruption of user wiring? 1987 */ 1988 } 1989 vm_map_lock(map); 1990 if (last_timestamp + 1 != map->timestamp) { 1991 /* 1992 * Look again for the entry because the map was 1993 * modified while it was unlocked. 1994 * Specifically, the entry may have been 1995 * clipped, merged, or deleted. 1996 */ 1997 if (!vm_map_lookup_entry(map, saved_start, 1998 &tmp_entry)) { 1999 if (flags & VM_MAP_WIRE_HOLESOK) 2000 tmp_entry = tmp_entry->next; 2001 else { 2002 if (saved_start == start) { 2003 /* 2004 * first_entry has been deleted. 2005 */ 2006 vm_map_unlock(map); 2007 return (KERN_INVALID_ADDRESS); 2008 } 2009 end = saved_start; 2010 rv = KERN_INVALID_ADDRESS; 2011 goto done; 2012 } 2013 } 2014 if (entry == first_entry) 2015 first_entry = tmp_entry; 2016 else 2017 first_entry = NULL; 2018 entry = tmp_entry; 2019 } 2020 last_timestamp = map->timestamp; 2021 continue; 2022 } 2023 vm_map_clip_start(map, entry, start); 2024 vm_map_clip_end(map, entry, end); 2025 /* 2026 * Mark the entry in case the map lock is released. (See 2027 * above.) 2028 */ 2029 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2030 /* 2031 * 2032 */ 2033 if (entry->wired_count == 0) { 2034 entry->wired_count++; 2035 saved_start = entry->start; 2036 saved_end = entry->end; 2037 fictitious = entry->object.vm_object != NULL && 2038 entry->object.vm_object->type == OBJT_DEVICE; 2039 /* 2040 * Release the map lock, relying on the in-transition 2041 * mark. 2042 */ 2043 vm_map_unlock(map); 2044 rv = vm_fault_wire(map, saved_start, saved_end, 2045 user_wire, fictitious); 2046 vm_map_lock(map); 2047 if (last_timestamp + 1 != map->timestamp) { 2048 /* 2049 * Look again for the entry because the map was 2050 * modified while it was unlocked. The entry 2051 * may have been clipped, but NOT merged or 2052 * deleted. 2053 */ 2054 result = vm_map_lookup_entry(map, saved_start, 2055 &tmp_entry); 2056 KASSERT(result, ("vm_map_wire: lookup failed")); 2057 if (entry == first_entry) 2058 first_entry = tmp_entry; 2059 else 2060 first_entry = NULL; 2061 entry = tmp_entry; 2062 while (entry->end < saved_end) { 2063 if (rv != KERN_SUCCESS) { 2064 KASSERT(entry->wired_count == 1, 2065 ("vm_map_wire: bad count")); 2066 entry->wired_count = -1; 2067 } 2068 entry = entry->next; 2069 } 2070 } 2071 last_timestamp = map->timestamp; 2072 if (rv != KERN_SUCCESS) { 2073 KASSERT(entry->wired_count == 1, 2074 ("vm_map_wire: bad count")); 2075 /* 2076 * Assign an out-of-range value to represent 2077 * the failure to wire this entry. 2078 */ 2079 entry->wired_count = -1; 2080 end = entry->end; 2081 goto done; 2082 } 2083 } else if (!user_wire || 2084 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2085 entry->wired_count++; 2086 } 2087 /* 2088 * Check the map for holes in the specified region. 2089 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2090 */ 2091 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2092 (entry->end < end && (entry->next == &map->header || 2093 entry->next->start > entry->end))) { 2094 end = entry->end; 2095 rv = KERN_INVALID_ADDRESS; 2096 goto done; 2097 } 2098 entry = entry->next; 2099 } 2100 rv = KERN_SUCCESS; 2101 done: 2102 need_wakeup = FALSE; 2103 if (first_entry == NULL) { 2104 result = vm_map_lookup_entry(map, start, &first_entry); 2105 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2106 first_entry = first_entry->next; 2107 else 2108 KASSERT(result, ("vm_map_wire: lookup failed")); 2109 } 2110 entry = first_entry; 2111 while (entry != &map->header && entry->start < end) { 2112 if (rv == KERN_SUCCESS) { 2113 if (user_wire) 2114 entry->eflags |= MAP_ENTRY_USER_WIRED; 2115 } else if (entry->wired_count == -1) { 2116 /* 2117 * Wiring failed on this entry. Thus, unwiring is 2118 * unnecessary. 2119 */ 2120 entry->wired_count = 0; 2121 } else { 2122 if (!user_wire || 2123 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2124 entry->wired_count--; 2125 if (entry->wired_count == 0) { 2126 /* 2127 * Retain the map lock. 2128 */ 2129 vm_fault_unwire(map, entry->start, entry->end, 2130 entry->object.vm_object != NULL && 2131 entry->object.vm_object->type == OBJT_DEVICE); 2132 } 2133 } 2134 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2135 ("vm_map_wire: in-transition flag missing")); 2136 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2137 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2138 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2139 need_wakeup = TRUE; 2140 } 2141 vm_map_simplify_entry(map, entry); 2142 entry = entry->next; 2143 } 2144 vm_map_unlock(map); 2145 if (need_wakeup) 2146 vm_map_wakeup(map); 2147 return (rv); 2148 } 2149 2150 /* 2151 * vm_map_sync 2152 * 2153 * Push any dirty cached pages in the address range to their pager. 2154 * If syncio is TRUE, dirty pages are written synchronously. 2155 * If invalidate is TRUE, any cached pages are freed as well. 2156 * 2157 * If the size of the region from start to end is zero, we are 2158 * supposed to flush all modified pages within the region containing 2159 * start. Unfortunately, a region can be split or coalesced with 2160 * neighboring regions, making it difficult to determine what the 2161 * original region was. Therefore, we approximate this requirement by 2162 * flushing the current region containing start. 2163 * 2164 * Returns an error if any part of the specified range is not mapped. 2165 */ 2166 int 2167 vm_map_sync( 2168 vm_map_t map, 2169 vm_offset_t start, 2170 vm_offset_t end, 2171 boolean_t syncio, 2172 boolean_t invalidate) 2173 { 2174 vm_map_entry_t current; 2175 vm_map_entry_t entry; 2176 vm_size_t size; 2177 vm_object_t object; 2178 vm_ooffset_t offset; 2179 2180 vm_map_lock_read(map); 2181 VM_MAP_RANGE_CHECK(map, start, end); 2182 if (!vm_map_lookup_entry(map, start, &entry)) { 2183 vm_map_unlock_read(map); 2184 return (KERN_INVALID_ADDRESS); 2185 } else if (start == end) { 2186 start = entry->start; 2187 end = entry->end; 2188 } 2189 /* 2190 * Make a first pass to check for user-wired memory and holes. 2191 */ 2192 for (current = entry; current->start < end; current = current->next) { 2193 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2194 vm_map_unlock_read(map); 2195 return (KERN_INVALID_ARGUMENT); 2196 } 2197 if (end > current->end && 2198 (current->next == &map->header || 2199 current->end != current->next->start)) { 2200 vm_map_unlock_read(map); 2201 return (KERN_INVALID_ADDRESS); 2202 } 2203 } 2204 2205 if (invalidate) 2206 pmap_remove(map->pmap, start, end); 2207 2208 /* 2209 * Make a second pass, cleaning/uncaching pages from the indicated 2210 * objects as we go. 2211 */ 2212 for (current = entry; current->start < end; current = current->next) { 2213 offset = current->offset + (start - current->start); 2214 size = (end <= current->end ? end : current->end) - start; 2215 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2216 vm_map_t smap; 2217 vm_map_entry_t tentry; 2218 vm_size_t tsize; 2219 2220 smap = current->object.sub_map; 2221 vm_map_lock_read(smap); 2222 (void) vm_map_lookup_entry(smap, offset, &tentry); 2223 tsize = tentry->end - offset; 2224 if (tsize < size) 2225 size = tsize; 2226 object = tentry->object.vm_object; 2227 offset = tentry->offset + (offset - tentry->start); 2228 vm_map_unlock_read(smap); 2229 } else { 2230 object = current->object.vm_object; 2231 } 2232 vm_object_sync(object, offset, size, syncio, invalidate); 2233 start += size; 2234 } 2235 2236 vm_map_unlock_read(map); 2237 return (KERN_SUCCESS); 2238 } 2239 2240 /* 2241 * vm_map_entry_unwire: [ internal use only ] 2242 * 2243 * Make the region specified by this entry pageable. 2244 * 2245 * The map in question should be locked. 2246 * [This is the reason for this routine's existence.] 2247 */ 2248 static void 2249 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2250 { 2251 vm_fault_unwire(map, entry->start, entry->end, 2252 entry->object.vm_object != NULL && 2253 entry->object.vm_object->type == OBJT_DEVICE); 2254 entry->wired_count = 0; 2255 } 2256 2257 /* 2258 * vm_map_entry_delete: [ internal use only ] 2259 * 2260 * Deallocate the given entry from the target map. 2261 */ 2262 static void 2263 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2264 { 2265 vm_object_t object; 2266 vm_pindex_t offidxstart, offidxend, count; 2267 2268 vm_map_entry_unlink(map, entry); 2269 map->size -= entry->end - entry->start; 2270 2271 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2272 (object = entry->object.vm_object) != NULL) { 2273 count = OFF_TO_IDX(entry->end - entry->start); 2274 offidxstart = OFF_TO_IDX(entry->offset); 2275 offidxend = offidxstart + count; 2276 VM_OBJECT_LOCK(object); 2277 if (object->ref_count != 1 && 2278 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2279 object == kernel_object || object == kmem_object)) { 2280 vm_object_collapse(object); 2281 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2282 if (object->type == OBJT_SWAP) 2283 swap_pager_freespace(object, offidxstart, count); 2284 if (offidxend >= object->size && 2285 offidxstart < object->size) 2286 object->size = offidxstart; 2287 } 2288 VM_OBJECT_UNLOCK(object); 2289 vm_object_deallocate(object); 2290 } 2291 2292 vm_map_entry_dispose(map, entry); 2293 } 2294 2295 /* 2296 * vm_map_delete: [ internal use only ] 2297 * 2298 * Deallocates the given address range from the target 2299 * map. 2300 */ 2301 int 2302 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2303 { 2304 vm_map_entry_t entry; 2305 vm_map_entry_t first_entry; 2306 2307 /* 2308 * Find the start of the region, and clip it 2309 */ 2310 if (!vm_map_lookup_entry(map, start, &first_entry)) 2311 entry = first_entry->next; 2312 else { 2313 entry = first_entry; 2314 vm_map_clip_start(map, entry, start); 2315 } 2316 2317 /* 2318 * Step through all entries in this region 2319 */ 2320 while ((entry != &map->header) && (entry->start < end)) { 2321 vm_map_entry_t next; 2322 2323 /* 2324 * Wait for wiring or unwiring of an entry to complete. 2325 * Also wait for any system wirings to disappear on 2326 * user maps. 2327 */ 2328 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2329 (vm_map_pmap(map) != kernel_pmap && 2330 vm_map_entry_system_wired_count(entry) != 0)) { 2331 unsigned int last_timestamp; 2332 vm_offset_t saved_start; 2333 vm_map_entry_t tmp_entry; 2334 2335 saved_start = entry->start; 2336 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2337 last_timestamp = map->timestamp; 2338 (void) vm_map_unlock_and_wait(map, FALSE); 2339 vm_map_lock(map); 2340 if (last_timestamp + 1 != map->timestamp) { 2341 /* 2342 * Look again for the entry because the map was 2343 * modified while it was unlocked. 2344 * Specifically, the entry may have been 2345 * clipped, merged, or deleted. 2346 */ 2347 if (!vm_map_lookup_entry(map, saved_start, 2348 &tmp_entry)) 2349 entry = tmp_entry->next; 2350 else { 2351 entry = tmp_entry; 2352 vm_map_clip_start(map, entry, 2353 saved_start); 2354 } 2355 } 2356 continue; 2357 } 2358 vm_map_clip_end(map, entry, end); 2359 2360 next = entry->next; 2361 2362 /* 2363 * Unwire before removing addresses from the pmap; otherwise, 2364 * unwiring will put the entries back in the pmap. 2365 */ 2366 if (entry->wired_count != 0) { 2367 vm_map_entry_unwire(map, entry); 2368 } 2369 2370 pmap_remove(map->pmap, entry->start, entry->end); 2371 2372 /* 2373 * Delete the entry (which may delete the object) only after 2374 * removing all pmap entries pointing to its pages. 2375 * (Otherwise, its page frames may be reallocated, and any 2376 * modify bits will be set in the wrong object!) 2377 */ 2378 vm_map_entry_delete(map, entry); 2379 entry = next; 2380 } 2381 return (KERN_SUCCESS); 2382 } 2383 2384 /* 2385 * vm_map_remove: 2386 * 2387 * Remove the given address range from the target map. 2388 * This is the exported form of vm_map_delete. 2389 */ 2390 int 2391 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2392 { 2393 int result; 2394 2395 vm_map_lock(map); 2396 VM_MAP_RANGE_CHECK(map, start, end); 2397 result = vm_map_delete(map, start, end); 2398 vm_map_unlock(map); 2399 return (result); 2400 } 2401 2402 /* 2403 * vm_map_check_protection: 2404 * 2405 * Assert that the target map allows the specified privilege on the 2406 * entire address region given. The entire region must be allocated. 2407 * 2408 * WARNING! This code does not and should not check whether the 2409 * contents of the region is accessible. For example a smaller file 2410 * might be mapped into a larger address space. 2411 * 2412 * NOTE! This code is also called by munmap(). 2413 * 2414 * The map must be locked. A read lock is sufficient. 2415 */ 2416 boolean_t 2417 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2418 vm_prot_t protection) 2419 { 2420 vm_map_entry_t entry; 2421 vm_map_entry_t tmp_entry; 2422 2423 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2424 return (FALSE); 2425 entry = tmp_entry; 2426 2427 while (start < end) { 2428 if (entry == &map->header) 2429 return (FALSE); 2430 /* 2431 * No holes allowed! 2432 */ 2433 if (start < entry->start) 2434 return (FALSE); 2435 /* 2436 * Check protection associated with entry. 2437 */ 2438 if ((entry->protection & protection) != protection) 2439 return (FALSE); 2440 /* go to next entry */ 2441 start = entry->end; 2442 entry = entry->next; 2443 } 2444 return (TRUE); 2445 } 2446 2447 /* 2448 * vm_map_copy_entry: 2449 * 2450 * Copies the contents of the source entry to the destination 2451 * entry. The entries *must* be aligned properly. 2452 */ 2453 static void 2454 vm_map_copy_entry( 2455 vm_map_t src_map, 2456 vm_map_t dst_map, 2457 vm_map_entry_t src_entry, 2458 vm_map_entry_t dst_entry) 2459 { 2460 vm_object_t src_object; 2461 2462 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2463 return; 2464 2465 if (src_entry->wired_count == 0) { 2466 2467 /* 2468 * If the source entry is marked needs_copy, it is already 2469 * write-protected. 2470 */ 2471 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2472 pmap_protect(src_map->pmap, 2473 src_entry->start, 2474 src_entry->end, 2475 src_entry->protection & ~VM_PROT_WRITE); 2476 } 2477 2478 /* 2479 * Make a copy of the object. 2480 */ 2481 if ((src_object = src_entry->object.vm_object) != NULL) { 2482 VM_OBJECT_LOCK(src_object); 2483 if ((src_object->handle == NULL) && 2484 (src_object->type == OBJT_DEFAULT || 2485 src_object->type == OBJT_SWAP)) { 2486 vm_object_collapse(src_object); 2487 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2488 vm_object_split(src_entry); 2489 src_object = src_entry->object.vm_object; 2490 } 2491 } 2492 vm_object_reference_locked(src_object); 2493 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2494 VM_OBJECT_UNLOCK(src_object); 2495 dst_entry->object.vm_object = src_object; 2496 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2497 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2498 dst_entry->offset = src_entry->offset; 2499 } else { 2500 dst_entry->object.vm_object = NULL; 2501 dst_entry->offset = 0; 2502 } 2503 2504 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2505 dst_entry->end - dst_entry->start, src_entry->start); 2506 } else { 2507 /* 2508 * Of course, wired down pages can't be set copy-on-write. 2509 * Cause wired pages to be copied into the new map by 2510 * simulating faults (the new pages are pageable) 2511 */ 2512 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2513 } 2514 } 2515 2516 /* 2517 * vmspace_map_entry_forked: 2518 * Update the newly-forked vmspace each time a map entry is inherited 2519 * or copied. The values for vm_dsize and vm_tsize are approximate 2520 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2521 */ 2522 static void 2523 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2524 vm_map_entry_t entry) 2525 { 2526 vm_size_t entrysize; 2527 vm_offset_t newend; 2528 2529 entrysize = entry->end - entry->start; 2530 vm2->vm_map.size += entrysize; 2531 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2532 vm2->vm_ssize += btoc(entrysize); 2533 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2534 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2535 newend = MIN(entry->end, 2536 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2537 vm2->vm_dsize += btoc(newend - entry->start); 2538 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2539 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2540 newend = MIN(entry->end, 2541 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2542 vm2->vm_tsize += btoc(newend - entry->start); 2543 } 2544 } 2545 2546 /* 2547 * vmspace_fork: 2548 * Create a new process vmspace structure and vm_map 2549 * based on those of an existing process. The new map 2550 * is based on the old map, according to the inheritance 2551 * values on the regions in that map. 2552 * 2553 * XXX It might be worth coalescing the entries added to the new vmspace. 2554 * 2555 * The source map must not be locked. 2556 */ 2557 struct vmspace * 2558 vmspace_fork(struct vmspace *vm1) 2559 { 2560 struct vmspace *vm2; 2561 vm_map_t old_map = &vm1->vm_map; 2562 vm_map_t new_map; 2563 vm_map_entry_t old_entry; 2564 vm_map_entry_t new_entry; 2565 vm_object_t object; 2566 2567 vm_map_lock(old_map); 2568 2569 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2570 vm2->vm_taddr = vm1->vm_taddr; 2571 vm2->vm_daddr = vm1->vm_daddr; 2572 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2573 new_map = &vm2->vm_map; /* XXX */ 2574 new_map->timestamp = 1; 2575 2576 old_entry = old_map->header.next; 2577 2578 while (old_entry != &old_map->header) { 2579 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2580 panic("vm_map_fork: encountered a submap"); 2581 2582 switch (old_entry->inheritance) { 2583 case VM_INHERIT_NONE: 2584 break; 2585 2586 case VM_INHERIT_SHARE: 2587 /* 2588 * Clone the entry, creating the shared object if necessary. 2589 */ 2590 object = old_entry->object.vm_object; 2591 if (object == NULL) { 2592 object = vm_object_allocate(OBJT_DEFAULT, 2593 atop(old_entry->end - old_entry->start)); 2594 old_entry->object.vm_object = object; 2595 old_entry->offset = 0; 2596 } 2597 2598 /* 2599 * Add the reference before calling vm_object_shadow 2600 * to insure that a shadow object is created. 2601 */ 2602 vm_object_reference(object); 2603 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 2604 vm_object_shadow(&old_entry->object.vm_object, 2605 &old_entry->offset, 2606 atop(old_entry->end - old_entry->start)); 2607 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 2608 /* Transfer the second reference too. */ 2609 vm_object_reference( 2610 old_entry->object.vm_object); 2611 vm_object_deallocate(object); 2612 object = old_entry->object.vm_object; 2613 } 2614 VM_OBJECT_LOCK(object); 2615 vm_object_clear_flag(object, OBJ_ONEMAPPING); 2616 VM_OBJECT_UNLOCK(object); 2617 2618 /* 2619 * Clone the entry, referencing the shared object. 2620 */ 2621 new_entry = vm_map_entry_create(new_map); 2622 *new_entry = *old_entry; 2623 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2624 new_entry->wired_count = 0; 2625 2626 /* 2627 * Insert the entry into the new map -- we know we're 2628 * inserting at the end of the new map. 2629 */ 2630 vm_map_entry_link(new_map, new_map->header.prev, 2631 new_entry); 2632 vmspace_map_entry_forked(vm1, vm2, new_entry); 2633 2634 /* 2635 * Update the physical map 2636 */ 2637 pmap_copy(new_map->pmap, old_map->pmap, 2638 new_entry->start, 2639 (old_entry->end - old_entry->start), 2640 old_entry->start); 2641 break; 2642 2643 case VM_INHERIT_COPY: 2644 /* 2645 * Clone the entry and link into the map. 2646 */ 2647 new_entry = vm_map_entry_create(new_map); 2648 *new_entry = *old_entry; 2649 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2650 new_entry->wired_count = 0; 2651 new_entry->object.vm_object = NULL; 2652 vm_map_entry_link(new_map, new_map->header.prev, 2653 new_entry); 2654 vmspace_map_entry_forked(vm1, vm2, new_entry); 2655 vm_map_copy_entry(old_map, new_map, old_entry, 2656 new_entry); 2657 break; 2658 } 2659 old_entry = old_entry->next; 2660 } 2661 2662 vm_map_unlock(old_map); 2663 2664 return (vm2); 2665 } 2666 2667 int 2668 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 2669 vm_prot_t prot, vm_prot_t max, int cow) 2670 { 2671 vm_map_entry_t new_entry, prev_entry; 2672 vm_offset_t bot, top; 2673 vm_size_t init_ssize; 2674 int orient, rv; 2675 rlim_t vmemlim; 2676 2677 /* 2678 * The stack orientation is piggybacked with the cow argument. 2679 * Extract it into orient and mask the cow argument so that we 2680 * don't pass it around further. 2681 * NOTE: We explicitly allow bi-directional stacks. 2682 */ 2683 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 2684 cow &= ~orient; 2685 KASSERT(orient != 0, ("No stack grow direction")); 2686 2687 if (addrbos < vm_map_min(map) || addrbos > map->max_offset) 2688 return (KERN_NO_SPACE); 2689 2690 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 2691 2692 PROC_LOCK(curthread->td_proc); 2693 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 2694 PROC_UNLOCK(curthread->td_proc); 2695 2696 vm_map_lock(map); 2697 2698 /* If addr is already mapped, no go */ 2699 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 2700 vm_map_unlock(map); 2701 return (KERN_NO_SPACE); 2702 } 2703 2704 /* If we would blow our VMEM resource limit, no go */ 2705 if (map->size + init_ssize > vmemlim) { 2706 vm_map_unlock(map); 2707 return (KERN_NO_SPACE); 2708 } 2709 2710 /* 2711 * If we can't accomodate max_ssize in the current mapping, no go. 2712 * However, we need to be aware that subsequent user mappings might 2713 * map into the space we have reserved for stack, and currently this 2714 * space is not protected. 2715 * 2716 * Hopefully we will at least detect this condition when we try to 2717 * grow the stack. 2718 */ 2719 if ((prev_entry->next != &map->header) && 2720 (prev_entry->next->start < addrbos + max_ssize)) { 2721 vm_map_unlock(map); 2722 return (KERN_NO_SPACE); 2723 } 2724 2725 /* 2726 * We initially map a stack of only init_ssize. We will grow as 2727 * needed later. Depending on the orientation of the stack (i.e. 2728 * the grow direction) we either map at the top of the range, the 2729 * bottom of the range or in the middle. 2730 * 2731 * Note: we would normally expect prot and max to be VM_PROT_ALL, 2732 * and cow to be 0. Possibly we should eliminate these as input 2733 * parameters, and just pass these values here in the insert call. 2734 */ 2735 if (orient == MAP_STACK_GROWS_DOWN) 2736 bot = addrbos + max_ssize - init_ssize; 2737 else if (orient == MAP_STACK_GROWS_UP) 2738 bot = addrbos; 2739 else 2740 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 2741 top = bot + init_ssize; 2742 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 2743 2744 /* Now set the avail_ssize amount. */ 2745 if (rv == KERN_SUCCESS) { 2746 if (prev_entry != &map->header) 2747 vm_map_clip_end(map, prev_entry, bot); 2748 new_entry = prev_entry->next; 2749 if (new_entry->end != top || new_entry->start != bot) 2750 panic("Bad entry start/end for new stack entry"); 2751 2752 new_entry->avail_ssize = max_ssize - init_ssize; 2753 if (orient & MAP_STACK_GROWS_DOWN) 2754 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2755 if (orient & MAP_STACK_GROWS_UP) 2756 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 2757 } 2758 2759 vm_map_unlock(map); 2760 return (rv); 2761 } 2762 2763 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 2764 * desired address is already mapped, or if we successfully grow 2765 * the stack. Also returns KERN_SUCCESS if addr is outside the 2766 * stack range (this is strange, but preserves compatibility with 2767 * the grow function in vm_machdep.c). 2768 */ 2769 int 2770 vm_map_growstack(struct proc *p, vm_offset_t addr) 2771 { 2772 vm_map_entry_t next_entry, prev_entry; 2773 vm_map_entry_t new_entry, stack_entry; 2774 struct vmspace *vm = p->p_vmspace; 2775 vm_map_t map = &vm->vm_map; 2776 vm_offset_t end; 2777 size_t grow_amount, max_grow; 2778 rlim_t stacklim, vmemlim; 2779 int is_procstack, rv; 2780 2781 Retry: 2782 PROC_LOCK(p); 2783 stacklim = lim_cur(p, RLIMIT_STACK); 2784 vmemlim = lim_cur(p, RLIMIT_VMEM); 2785 PROC_UNLOCK(p); 2786 2787 vm_map_lock_read(map); 2788 2789 /* If addr is already in the entry range, no need to grow.*/ 2790 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 2791 vm_map_unlock_read(map); 2792 return (KERN_SUCCESS); 2793 } 2794 2795 next_entry = prev_entry->next; 2796 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 2797 /* 2798 * This entry does not grow upwards. Since the address lies 2799 * beyond this entry, the next entry (if one exists) has to 2800 * be a downward growable entry. The entry list header is 2801 * never a growable entry, so it suffices to check the flags. 2802 */ 2803 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 2804 vm_map_unlock_read(map); 2805 return (KERN_SUCCESS); 2806 } 2807 stack_entry = next_entry; 2808 } else { 2809 /* 2810 * This entry grows upward. If the next entry does not at 2811 * least grow downwards, this is the entry we need to grow. 2812 * otherwise we have two possible choices and we have to 2813 * select one. 2814 */ 2815 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 2816 /* 2817 * We have two choices; grow the entry closest to 2818 * the address to minimize the amount of growth. 2819 */ 2820 if (addr - prev_entry->end <= next_entry->start - addr) 2821 stack_entry = prev_entry; 2822 else 2823 stack_entry = next_entry; 2824 } else 2825 stack_entry = prev_entry; 2826 } 2827 2828 if (stack_entry == next_entry) { 2829 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 2830 KASSERT(addr < stack_entry->start, ("foo")); 2831 end = (prev_entry != &map->header) ? prev_entry->end : 2832 stack_entry->start - stack_entry->avail_ssize; 2833 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 2834 max_grow = stack_entry->start - end; 2835 } else { 2836 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 2837 KASSERT(addr >= stack_entry->end, ("foo")); 2838 end = (next_entry != &map->header) ? next_entry->start : 2839 stack_entry->end + stack_entry->avail_ssize; 2840 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 2841 max_grow = end - stack_entry->end; 2842 } 2843 2844 if (grow_amount > stack_entry->avail_ssize) { 2845 vm_map_unlock_read(map); 2846 return (KERN_NO_SPACE); 2847 } 2848 2849 /* 2850 * If there is no longer enough space between the entries nogo, and 2851 * adjust the available space. Note: this should only happen if the 2852 * user has mapped into the stack area after the stack was created, 2853 * and is probably an error. 2854 * 2855 * This also effectively destroys any guard page the user might have 2856 * intended by limiting the stack size. 2857 */ 2858 if (grow_amount > max_grow) { 2859 if (vm_map_lock_upgrade(map)) 2860 goto Retry; 2861 2862 stack_entry->avail_ssize = max_grow; 2863 2864 vm_map_unlock(map); 2865 return (KERN_NO_SPACE); 2866 } 2867 2868 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 2869 2870 /* 2871 * If this is the main process stack, see if we're over the stack 2872 * limit. 2873 */ 2874 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2875 vm_map_unlock_read(map); 2876 return (KERN_NO_SPACE); 2877 } 2878 2879 /* Round up the grow amount modulo SGROWSIZ */ 2880 grow_amount = roundup (grow_amount, sgrowsiz); 2881 if (grow_amount > stack_entry->avail_ssize) 2882 grow_amount = stack_entry->avail_ssize; 2883 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2884 grow_amount = stacklim - ctob(vm->vm_ssize); 2885 } 2886 2887 /* If we would blow our VMEM resource limit, no go */ 2888 if (map->size + grow_amount > vmemlim) { 2889 vm_map_unlock_read(map); 2890 return (KERN_NO_SPACE); 2891 } 2892 2893 if (vm_map_lock_upgrade(map)) 2894 goto Retry; 2895 2896 if (stack_entry == next_entry) { 2897 /* 2898 * Growing downward. 2899 */ 2900 /* Get the preliminary new entry start value */ 2901 addr = stack_entry->start - grow_amount; 2902 2903 /* 2904 * If this puts us into the previous entry, cut back our 2905 * growth to the available space. Also, see the note above. 2906 */ 2907 if (addr < end) { 2908 stack_entry->avail_ssize = max_grow; 2909 addr = end; 2910 } 2911 2912 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 2913 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 2914 2915 /* Adjust the available stack space by the amount we grew. */ 2916 if (rv == KERN_SUCCESS) { 2917 if (prev_entry != &map->header) 2918 vm_map_clip_end(map, prev_entry, addr); 2919 new_entry = prev_entry->next; 2920 KASSERT(new_entry == stack_entry->prev, ("foo")); 2921 KASSERT(new_entry->end == stack_entry->start, ("foo")); 2922 KASSERT(new_entry->start == addr, ("foo")); 2923 grow_amount = new_entry->end - new_entry->start; 2924 new_entry->avail_ssize = stack_entry->avail_ssize - 2925 grow_amount; 2926 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 2927 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2928 } 2929 } else { 2930 /* 2931 * Growing upward. 2932 */ 2933 addr = stack_entry->end + grow_amount; 2934 2935 /* 2936 * If this puts us into the next entry, cut back our growth 2937 * to the available space. Also, see the note above. 2938 */ 2939 if (addr > end) { 2940 stack_entry->avail_ssize = end - stack_entry->end; 2941 addr = end; 2942 } 2943 2944 grow_amount = addr - stack_entry->end; 2945 2946 /* Grow the underlying object if applicable. */ 2947 if (stack_entry->object.vm_object == NULL || 2948 vm_object_coalesce(stack_entry->object.vm_object, 2949 stack_entry->offset, 2950 (vm_size_t)(stack_entry->end - stack_entry->start), 2951 (vm_size_t)grow_amount)) { 2952 map->size += (addr - stack_entry->end); 2953 /* Update the current entry. */ 2954 stack_entry->end = addr; 2955 stack_entry->avail_ssize -= grow_amount; 2956 vm_map_entry_resize_free(map, stack_entry); 2957 rv = KERN_SUCCESS; 2958 2959 if (next_entry != &map->header) 2960 vm_map_clip_start(map, next_entry, addr); 2961 } else 2962 rv = KERN_FAILURE; 2963 } 2964 2965 if (rv == KERN_SUCCESS && is_procstack) 2966 vm->vm_ssize += btoc(grow_amount); 2967 2968 vm_map_unlock(map); 2969 2970 /* 2971 * Heed the MAP_WIREFUTURE flag if it was set for this process. 2972 */ 2973 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 2974 vm_map_wire(map, 2975 (stack_entry == next_entry) ? addr : addr - grow_amount, 2976 (stack_entry == next_entry) ? stack_entry->start : addr, 2977 (p->p_flag & P_SYSTEM) 2978 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 2979 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 2980 } 2981 2982 return (rv); 2983 } 2984 2985 /* 2986 * Unshare the specified VM space for exec. If other processes are 2987 * mapped to it, then create a new one. The new vmspace is null. 2988 */ 2989 void 2990 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 2991 { 2992 struct vmspace *oldvmspace = p->p_vmspace; 2993 struct vmspace *newvmspace; 2994 2995 newvmspace = vmspace_alloc(minuser, maxuser); 2996 newvmspace->vm_swrss = oldvmspace->vm_swrss; 2997 /* 2998 * This code is written like this for prototype purposes. The 2999 * goal is to avoid running down the vmspace here, but let the 3000 * other process's that are still using the vmspace to finally 3001 * run it down. Even though there is little or no chance of blocking 3002 * here, it is a good idea to keep this form for future mods. 3003 */ 3004 PROC_VMSPACE_LOCK(p); 3005 p->p_vmspace = newvmspace; 3006 PROC_VMSPACE_UNLOCK(p); 3007 if (p == curthread->td_proc) /* XXXKSE ? */ 3008 pmap_activate(curthread); 3009 vmspace_free(oldvmspace); 3010 } 3011 3012 /* 3013 * Unshare the specified VM space for forcing COW. This 3014 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3015 */ 3016 void 3017 vmspace_unshare(struct proc *p) 3018 { 3019 struct vmspace *oldvmspace = p->p_vmspace; 3020 struct vmspace *newvmspace; 3021 3022 if (oldvmspace->vm_refcnt == 1) 3023 return; 3024 newvmspace = vmspace_fork(oldvmspace); 3025 PROC_VMSPACE_LOCK(p); 3026 p->p_vmspace = newvmspace; 3027 PROC_VMSPACE_UNLOCK(p); 3028 if (p == curthread->td_proc) /* XXXKSE ? */ 3029 pmap_activate(curthread); 3030 vmspace_free(oldvmspace); 3031 } 3032 3033 /* 3034 * vm_map_lookup: 3035 * 3036 * Finds the VM object, offset, and 3037 * protection for a given virtual address in the 3038 * specified map, assuming a page fault of the 3039 * type specified. 3040 * 3041 * Leaves the map in question locked for read; return 3042 * values are guaranteed until a vm_map_lookup_done 3043 * call is performed. Note that the map argument 3044 * is in/out; the returned map must be used in 3045 * the call to vm_map_lookup_done. 3046 * 3047 * A handle (out_entry) is returned for use in 3048 * vm_map_lookup_done, to make that fast. 3049 * 3050 * If a lookup is requested with "write protection" 3051 * specified, the map may be changed to perform virtual 3052 * copying operations, although the data referenced will 3053 * remain the same. 3054 */ 3055 int 3056 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 3057 vm_offset_t vaddr, 3058 vm_prot_t fault_typea, 3059 vm_map_entry_t *out_entry, /* OUT */ 3060 vm_object_t *object, /* OUT */ 3061 vm_pindex_t *pindex, /* OUT */ 3062 vm_prot_t *out_prot, /* OUT */ 3063 boolean_t *wired) /* OUT */ 3064 { 3065 vm_map_entry_t entry; 3066 vm_map_t map = *var_map; 3067 vm_prot_t prot; 3068 vm_prot_t fault_type = fault_typea; 3069 3070 RetryLookup:; 3071 /* 3072 * Lookup the faulting address. 3073 */ 3074 3075 vm_map_lock_read(map); 3076 #define RETURN(why) \ 3077 { \ 3078 vm_map_unlock_read(map); \ 3079 return (why); \ 3080 } 3081 3082 /* 3083 * If the map has an interesting hint, try it before calling full 3084 * blown lookup routine. 3085 */ 3086 entry = map->root; 3087 *out_entry = entry; 3088 if (entry == NULL || 3089 (vaddr < entry->start) || (vaddr >= entry->end)) { 3090 /* 3091 * Entry was either not a valid hint, or the vaddr was not 3092 * contained in the entry, so do a full lookup. 3093 */ 3094 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3095 RETURN(KERN_INVALID_ADDRESS); 3096 3097 entry = *out_entry; 3098 } 3099 3100 /* 3101 * Handle submaps. 3102 */ 3103 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3104 vm_map_t old_map = map; 3105 3106 *var_map = map = entry->object.sub_map; 3107 vm_map_unlock_read(old_map); 3108 goto RetryLookup; 3109 } 3110 3111 /* 3112 * Check whether this task is allowed to have this page. 3113 * Note the special case for MAP_ENTRY_COW 3114 * pages with an override. This is to implement a forced 3115 * COW for debuggers. 3116 */ 3117 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3118 prot = entry->max_protection; 3119 else 3120 prot = entry->protection; 3121 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3122 if ((fault_type & prot) != fault_type) { 3123 RETURN(KERN_PROTECTION_FAILURE); 3124 } 3125 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3126 (entry->eflags & MAP_ENTRY_COW) && 3127 (fault_type & VM_PROT_WRITE) && 3128 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3129 RETURN(KERN_PROTECTION_FAILURE); 3130 } 3131 3132 /* 3133 * If this page is not pageable, we have to get it for all possible 3134 * accesses. 3135 */ 3136 *wired = (entry->wired_count != 0); 3137 if (*wired) 3138 prot = fault_type = entry->protection; 3139 3140 /* 3141 * If the entry was copy-on-write, we either ... 3142 */ 3143 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3144 /* 3145 * If we want to write the page, we may as well handle that 3146 * now since we've got the map locked. 3147 * 3148 * If we don't need to write the page, we just demote the 3149 * permissions allowed. 3150 */ 3151 if (fault_type & VM_PROT_WRITE) { 3152 /* 3153 * Make a new object, and place it in the object 3154 * chain. Note that no new references have appeared 3155 * -- one just moved from the map to the new 3156 * object. 3157 */ 3158 if (vm_map_lock_upgrade(map)) 3159 goto RetryLookup; 3160 3161 vm_object_shadow( 3162 &entry->object.vm_object, 3163 &entry->offset, 3164 atop(entry->end - entry->start)); 3165 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3166 3167 vm_map_lock_downgrade(map); 3168 } else { 3169 /* 3170 * We're attempting to read a copy-on-write page -- 3171 * don't allow writes. 3172 */ 3173 prot &= ~VM_PROT_WRITE; 3174 } 3175 } 3176 3177 /* 3178 * Create an object if necessary. 3179 */ 3180 if (entry->object.vm_object == NULL && 3181 !map->system_map) { 3182 if (vm_map_lock_upgrade(map)) 3183 goto RetryLookup; 3184 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3185 atop(entry->end - entry->start)); 3186 entry->offset = 0; 3187 vm_map_lock_downgrade(map); 3188 } 3189 3190 /* 3191 * Return the object/offset from this entry. If the entry was 3192 * copy-on-write or empty, it has been fixed up. 3193 */ 3194 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3195 *object = entry->object.vm_object; 3196 3197 *out_prot = prot; 3198 return (KERN_SUCCESS); 3199 3200 #undef RETURN 3201 } 3202 3203 /* 3204 * vm_map_lookup_locked: 3205 * 3206 * Lookup the faulting address. A version of vm_map_lookup that returns 3207 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3208 */ 3209 int 3210 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3211 vm_offset_t vaddr, 3212 vm_prot_t fault_typea, 3213 vm_map_entry_t *out_entry, /* OUT */ 3214 vm_object_t *object, /* OUT */ 3215 vm_pindex_t *pindex, /* OUT */ 3216 vm_prot_t *out_prot, /* OUT */ 3217 boolean_t *wired) /* OUT */ 3218 { 3219 vm_map_entry_t entry; 3220 vm_map_t map = *var_map; 3221 vm_prot_t prot; 3222 vm_prot_t fault_type = fault_typea; 3223 3224 /* 3225 * If the map has an interesting hint, try it before calling full 3226 * blown lookup routine. 3227 */ 3228 entry = map->root; 3229 *out_entry = entry; 3230 if (entry == NULL || 3231 (vaddr < entry->start) || (vaddr >= entry->end)) { 3232 /* 3233 * Entry was either not a valid hint, or the vaddr was not 3234 * contained in the entry, so do a full lookup. 3235 */ 3236 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3237 return (KERN_INVALID_ADDRESS); 3238 3239 entry = *out_entry; 3240 } 3241 3242 /* 3243 * Fail if the entry refers to a submap. 3244 */ 3245 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3246 return (KERN_FAILURE); 3247 3248 /* 3249 * Check whether this task is allowed to have this page. 3250 * Note the special case for MAP_ENTRY_COW 3251 * pages with an override. This is to implement a forced 3252 * COW for debuggers. 3253 */ 3254 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3255 prot = entry->max_protection; 3256 else 3257 prot = entry->protection; 3258 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3259 if ((fault_type & prot) != fault_type) 3260 return (KERN_PROTECTION_FAILURE); 3261 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3262 (entry->eflags & MAP_ENTRY_COW) && 3263 (fault_type & VM_PROT_WRITE) && 3264 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3265 return (KERN_PROTECTION_FAILURE); 3266 3267 /* 3268 * If this page is not pageable, we have to get it for all possible 3269 * accesses. 3270 */ 3271 *wired = (entry->wired_count != 0); 3272 if (*wired) 3273 prot = fault_type = entry->protection; 3274 3275 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3276 /* 3277 * Fail if the entry was copy-on-write for a write fault. 3278 */ 3279 if (fault_type & VM_PROT_WRITE) 3280 return (KERN_FAILURE); 3281 /* 3282 * We're attempting to read a copy-on-write page -- 3283 * don't allow writes. 3284 */ 3285 prot &= ~VM_PROT_WRITE; 3286 } 3287 3288 /* 3289 * Fail if an object should be created. 3290 */ 3291 if (entry->object.vm_object == NULL && !map->system_map) 3292 return (KERN_FAILURE); 3293 3294 /* 3295 * Return the object/offset from this entry. If the entry was 3296 * copy-on-write or empty, it has been fixed up. 3297 */ 3298 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3299 *object = entry->object.vm_object; 3300 3301 *out_prot = prot; 3302 return (KERN_SUCCESS); 3303 } 3304 3305 /* 3306 * vm_map_lookup_done: 3307 * 3308 * Releases locks acquired by a vm_map_lookup 3309 * (according to the handle returned by that lookup). 3310 */ 3311 void 3312 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3313 { 3314 /* 3315 * Unlock the main-level map 3316 */ 3317 vm_map_unlock_read(map); 3318 } 3319 3320 #include "opt_ddb.h" 3321 #ifdef DDB 3322 #include <sys/kernel.h> 3323 3324 #include <ddb/ddb.h> 3325 3326 /* 3327 * vm_map_print: [ debug ] 3328 */ 3329 DB_SHOW_COMMAND(map, vm_map_print) 3330 { 3331 static int nlines; 3332 /* XXX convert args. */ 3333 vm_map_t map = (vm_map_t)addr; 3334 boolean_t full = have_addr; 3335 3336 vm_map_entry_t entry; 3337 3338 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3339 (void *)map, 3340 (void *)map->pmap, map->nentries, map->timestamp); 3341 nlines++; 3342 3343 if (!full && db_indent) 3344 return; 3345 3346 db_indent += 2; 3347 for (entry = map->header.next; entry != &map->header; 3348 entry = entry->next) { 3349 db_iprintf("map entry %p: start=%p, end=%p\n", 3350 (void *)entry, (void *)entry->start, (void *)entry->end); 3351 nlines++; 3352 { 3353 static char *inheritance_name[4] = 3354 {"share", "copy", "none", "donate_copy"}; 3355 3356 db_iprintf(" prot=%x/%x/%s", 3357 entry->protection, 3358 entry->max_protection, 3359 inheritance_name[(int)(unsigned char)entry->inheritance]); 3360 if (entry->wired_count != 0) 3361 db_printf(", wired"); 3362 } 3363 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3364 db_printf(", share=%p, offset=0x%jx\n", 3365 (void *)entry->object.sub_map, 3366 (uintmax_t)entry->offset); 3367 nlines++; 3368 if ((entry->prev == &map->header) || 3369 (entry->prev->object.sub_map != 3370 entry->object.sub_map)) { 3371 db_indent += 2; 3372 vm_map_print((db_expr_t)(intptr_t) 3373 entry->object.sub_map, 3374 full, 0, (char *)0); 3375 db_indent -= 2; 3376 } 3377 } else { 3378 db_printf(", object=%p, offset=0x%jx", 3379 (void *)entry->object.vm_object, 3380 (uintmax_t)entry->offset); 3381 if (entry->eflags & MAP_ENTRY_COW) 3382 db_printf(", copy (%s)", 3383 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3384 db_printf("\n"); 3385 nlines++; 3386 3387 if ((entry->prev == &map->header) || 3388 (entry->prev->object.vm_object != 3389 entry->object.vm_object)) { 3390 db_indent += 2; 3391 vm_object_print((db_expr_t)(intptr_t) 3392 entry->object.vm_object, 3393 full, 0, (char *)0); 3394 nlines += 4; 3395 db_indent -= 2; 3396 } 3397 } 3398 } 3399 db_indent -= 2; 3400 if (db_indent == 0) 3401 nlines = 0; 3402 } 3403 3404 3405 DB_SHOW_COMMAND(procvm, procvm) 3406 { 3407 struct proc *p; 3408 3409 if (have_addr) { 3410 p = (struct proc *) addr; 3411 } else { 3412 p = curproc; 3413 } 3414 3415 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3416 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3417 (void *)vmspace_pmap(p->p_vmspace)); 3418 3419 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3420 } 3421 3422 #endif /* DDB */ 3423