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 void 152 vm_map_startup(void) 153 { 154 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 155 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 156 #ifdef INVARIANTS 157 vm_map_zdtor, 158 #else 159 NULL, 160 #endif 161 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 162 uma_prealloc(mapzone, MAX_KMAP); 163 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 164 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 165 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 166 uma_prealloc(kmapentzone, MAX_KMAPENT); 167 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 168 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 169 uma_prealloc(mapentzone, MAX_MAPENT); 170 } 171 172 static void 173 vmspace_zfini(void *mem, int size) 174 { 175 struct vmspace *vm; 176 177 vm = (struct vmspace *)mem; 178 pmap_release(vmspace_pmap(vm)); 179 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map)); 180 } 181 182 static int 183 vmspace_zinit(void *mem, int size, int flags) 184 { 185 struct vmspace *vm; 186 187 vm = (struct vmspace *)mem; 188 189 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 190 pmap_pinit(vmspace_pmap(vm)); 191 return (0); 192 } 193 194 static void 195 vm_map_zfini(void *mem, int size) 196 { 197 vm_map_t map; 198 199 map = (vm_map_t)mem; 200 mtx_destroy(&map->system_mtx); 201 sx_destroy(&map->lock); 202 } 203 204 static int 205 vm_map_zinit(void *mem, int size, int flags) 206 { 207 vm_map_t map; 208 209 map = (vm_map_t)mem; 210 map->nentries = 0; 211 map->size = 0; 212 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 213 sx_init(&map->lock, "user map"); 214 return (0); 215 } 216 217 #ifdef INVARIANTS 218 static void 219 vmspace_zdtor(void *mem, int size, void *arg) 220 { 221 struct vmspace *vm; 222 223 vm = (struct vmspace *)mem; 224 225 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 226 } 227 static void 228 vm_map_zdtor(void *mem, int size, void *arg) 229 { 230 vm_map_t map; 231 232 map = (vm_map_t)mem; 233 KASSERT(map->nentries == 0, 234 ("map %p nentries == %d on free.", 235 map, map->nentries)); 236 KASSERT(map->size == 0, 237 ("map %p size == %lu on free.", 238 map, (unsigned long)map->size)); 239 } 240 #endif /* INVARIANTS */ 241 242 /* 243 * Allocate a vmspace structure, including a vm_map and pmap, 244 * and initialize those structures. The refcnt is set to 1. 245 */ 246 struct vmspace * 247 vmspace_alloc(min, max) 248 vm_offset_t min, max; 249 { 250 struct vmspace *vm; 251 252 vm = uma_zalloc(vmspace_zone, M_WAITOK); 253 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 254 _vm_map_init(&vm->vm_map, min, max); 255 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ 256 vm->vm_refcnt = 1; 257 vm->vm_shm = NULL; 258 vm->vm_swrss = 0; 259 vm->vm_tsize = 0; 260 vm->vm_dsize = 0; 261 vm->vm_ssize = 0; 262 vm->vm_taddr = 0; 263 vm->vm_daddr = 0; 264 vm->vm_maxsaddr = 0; 265 vm->vm_exitingcnt = 0; 266 return (vm); 267 } 268 269 void 270 vm_init2(void) 271 { 272 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count, 273 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 + 274 maxproc * 2 + maxfiles); 275 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 276 #ifdef INVARIANTS 277 vmspace_zdtor, 278 #else 279 NULL, 280 #endif 281 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 282 pmap_init2(); 283 } 284 285 static __inline void 286 vmspace_dofree(struct vmspace *vm) 287 { 288 CTR1(KTR_VM, "vmspace_free: %p", vm); 289 290 /* 291 * Make sure any SysV shm is freed, it might not have been in 292 * exit1(). 293 */ 294 shmexit(vm); 295 296 /* 297 * Lock the map, to wait out all other references to it. 298 * Delete all of the mappings and pages they hold, then call 299 * the pmap module to reclaim anything left. 300 */ 301 vm_map_lock(&vm->vm_map); 302 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset, 303 vm->vm_map.max_offset); 304 vm_map_unlock(&vm->vm_map); 305 306 uma_zfree(vmspace_zone, vm); 307 } 308 309 void 310 vmspace_free(struct vmspace *vm) 311 { 312 int refcnt; 313 314 if (vm->vm_refcnt == 0) 315 panic("vmspace_free: attempt to free already freed vmspace"); 316 317 do 318 refcnt = vm->vm_refcnt; 319 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 320 if (refcnt == 1 && vm->vm_exitingcnt == 0) 321 vmspace_dofree(vm); 322 } 323 324 void 325 vmspace_exitfree(struct proc *p) 326 { 327 struct vmspace *vm; 328 int exitingcnt; 329 330 vm = p->p_vmspace; 331 p->p_vmspace = NULL; 332 333 /* 334 * cleanup by parent process wait()ing on exiting child. vm_refcnt 335 * may not be 0 (e.g. fork() and child exits without exec()ing). 336 * exitingcnt may increment above 0 and drop back down to zero 337 * several times while vm_refcnt is held non-zero. vm_refcnt 338 * may also increment above 0 and drop back down to zero several 339 * times while vm_exitingcnt is held non-zero. 340 * 341 * The last wait on the exiting child's vmspace will clean up 342 * the remainder of the vmspace. 343 */ 344 do 345 exitingcnt = vm->vm_exitingcnt; 346 while (!atomic_cmpset_int(&vm->vm_exitingcnt, exitingcnt, 347 exitingcnt - 1)); 348 if (vm->vm_refcnt == 0 && exitingcnt == 1) 349 vmspace_dofree(vm); 350 } 351 352 void 353 _vm_map_lock(vm_map_t map, const char *file, int line) 354 { 355 356 if (map->system_map) 357 _mtx_lock_flags(&map->system_mtx, 0, file, line); 358 else 359 _sx_xlock(&map->lock, file, line); 360 map->timestamp++; 361 } 362 363 void 364 _vm_map_unlock(vm_map_t map, const char *file, int line) 365 { 366 367 if (map->system_map) 368 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 369 else 370 _sx_xunlock(&map->lock, file, line); 371 } 372 373 void 374 _vm_map_lock_read(vm_map_t map, const char *file, int line) 375 { 376 377 if (map->system_map) 378 _mtx_lock_flags(&map->system_mtx, 0, file, line); 379 else 380 _sx_xlock(&map->lock, file, line); 381 } 382 383 void 384 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 385 { 386 387 if (map->system_map) 388 _mtx_unlock_flags(&map->system_mtx, 0, file, line); 389 else 390 _sx_xunlock(&map->lock, file, line); 391 } 392 393 int 394 _vm_map_trylock(vm_map_t map, const char *file, int line) 395 { 396 int error; 397 398 error = map->system_map ? 399 !_mtx_trylock(&map->system_mtx, 0, file, line) : 400 !_sx_try_xlock(&map->lock, file, line); 401 if (error == 0) 402 map->timestamp++; 403 return (error == 0); 404 } 405 406 int 407 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 408 { 409 int error; 410 411 error = map->system_map ? 412 !_mtx_trylock(&map->system_mtx, 0, file, line) : 413 !_sx_try_xlock(&map->lock, file, line); 414 return (error == 0); 415 } 416 417 int 418 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 419 { 420 421 #ifdef INVARIANTS 422 if (map->system_map) { 423 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 424 } else 425 _sx_assert(&map->lock, SX_XLOCKED, file, line); 426 #endif 427 map->timestamp++; 428 return (0); 429 } 430 431 void 432 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 433 { 434 435 #ifdef INVARIANTS 436 if (map->system_map) { 437 _mtx_assert(&map->system_mtx, MA_OWNED, file, line); 438 } else 439 _sx_assert(&map->lock, SX_XLOCKED, file, line); 440 #endif 441 } 442 443 /* 444 * vm_map_unlock_and_wait: 445 */ 446 int 447 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait) 448 { 449 450 mtx_lock(&map_sleep_mtx); 451 vm_map_unlock(map); 452 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0)); 453 } 454 455 /* 456 * vm_map_wakeup: 457 */ 458 void 459 vm_map_wakeup(vm_map_t map) 460 { 461 462 /* 463 * Acquire and release map_sleep_mtx to prevent a wakeup() 464 * from being performed (and lost) between the vm_map_unlock() 465 * and the msleep() in vm_map_unlock_and_wait(). 466 */ 467 mtx_lock(&map_sleep_mtx); 468 mtx_unlock(&map_sleep_mtx); 469 wakeup(&map->root); 470 } 471 472 long 473 vmspace_resident_count(struct vmspace *vmspace) 474 { 475 return pmap_resident_count(vmspace_pmap(vmspace)); 476 } 477 478 long 479 vmspace_wired_count(struct vmspace *vmspace) 480 { 481 return pmap_wired_count(vmspace_pmap(vmspace)); 482 } 483 484 /* 485 * vm_map_create: 486 * 487 * Creates and returns a new empty VM map with 488 * the given physical map structure, and having 489 * the given lower and upper address bounds. 490 */ 491 vm_map_t 492 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 493 { 494 vm_map_t result; 495 496 result = uma_zalloc(mapzone, M_WAITOK); 497 CTR1(KTR_VM, "vm_map_create: %p", result); 498 _vm_map_init(result, min, max); 499 result->pmap = pmap; 500 return (result); 501 } 502 503 /* 504 * Initialize an existing vm_map structure 505 * such as that in the vmspace structure. 506 * The pmap is set elsewhere. 507 */ 508 static void 509 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 510 { 511 512 map->header.next = map->header.prev = &map->header; 513 map->needs_wakeup = FALSE; 514 map->system_map = 0; 515 map->min_offset = min; 516 map->max_offset = max; 517 map->flags = 0; 518 map->root = NULL; 519 map->timestamp = 0; 520 } 521 522 void 523 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max) 524 { 525 _vm_map_init(map, min, max); 526 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 527 sx_init(&map->lock, "user map"); 528 } 529 530 /* 531 * vm_map_entry_dispose: [ internal use only ] 532 * 533 * Inverse of vm_map_entry_create. 534 */ 535 static void 536 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 537 { 538 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 539 } 540 541 /* 542 * vm_map_entry_create: [ internal use only ] 543 * 544 * Allocates a VM map entry for insertion. 545 * No entry fields are filled in. 546 */ 547 static vm_map_entry_t 548 vm_map_entry_create(vm_map_t map) 549 { 550 vm_map_entry_t new_entry; 551 552 if (map->system_map) 553 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 554 else 555 new_entry = uma_zalloc(mapentzone, M_WAITOK); 556 if (new_entry == NULL) 557 panic("vm_map_entry_create: kernel resources exhausted"); 558 return (new_entry); 559 } 560 561 /* 562 * vm_map_entry_set_behavior: 563 * 564 * Set the expected access behavior, either normal, random, or 565 * sequential. 566 */ 567 static __inline void 568 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 569 { 570 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 571 (behavior & MAP_ENTRY_BEHAV_MASK); 572 } 573 574 /* 575 * vm_map_entry_set_max_free: 576 * 577 * Set the max_free field in a vm_map_entry. 578 */ 579 static __inline void 580 vm_map_entry_set_max_free(vm_map_entry_t entry) 581 { 582 583 entry->max_free = entry->adj_free; 584 if (entry->left != NULL && entry->left->max_free > entry->max_free) 585 entry->max_free = entry->left->max_free; 586 if (entry->right != NULL && entry->right->max_free > entry->max_free) 587 entry->max_free = entry->right->max_free; 588 } 589 590 /* 591 * vm_map_entry_splay: 592 * 593 * The Sleator and Tarjan top-down splay algorithm with the 594 * following variation. Max_free must be computed bottom-up, so 595 * on the downward pass, maintain the left and right spines in 596 * reverse order. Then, make a second pass up each side to fix 597 * the pointers and compute max_free. The time bound is O(log n) 598 * amortized. 599 * 600 * The new root is the vm_map_entry containing "addr", or else an 601 * adjacent entry (lower or higher) if addr is not in the tree. 602 * 603 * The map must be locked, and leaves it so. 604 * 605 * Returns: the new root. 606 */ 607 static vm_map_entry_t 608 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 609 { 610 vm_map_entry_t llist, rlist; 611 vm_map_entry_t ltree, rtree; 612 vm_map_entry_t y; 613 614 /* Special case of empty tree. */ 615 if (root == NULL) 616 return (root); 617 618 /* 619 * Pass One: Splay down the tree until we find addr or a NULL 620 * pointer where addr would go. llist and rlist are the two 621 * sides in reverse order (bottom-up), with llist linked by 622 * the right pointer and rlist linked by the left pointer in 623 * the vm_map_entry. Wait until Pass Two to set max_free on 624 * the two spines. 625 */ 626 llist = NULL; 627 rlist = NULL; 628 for (;;) { 629 /* root is never NULL in here. */ 630 if (addr < root->start) { 631 y = root->left; 632 if (y == NULL) 633 break; 634 if (addr < y->start && y->left != NULL) { 635 /* Rotate right and put y on rlist. */ 636 root->left = y->right; 637 y->right = root; 638 vm_map_entry_set_max_free(root); 639 root = y->left; 640 y->left = rlist; 641 rlist = y; 642 } else { 643 /* Put root on rlist. */ 644 root->left = rlist; 645 rlist = root; 646 root = y; 647 } 648 } else { 649 y = root->right; 650 if (addr < root->end || y == NULL) 651 break; 652 if (addr >= y->end && y->right != NULL) { 653 /* Rotate left and put y on llist. */ 654 root->right = y->left; 655 y->left = root; 656 vm_map_entry_set_max_free(root); 657 root = y->right; 658 y->right = llist; 659 llist = y; 660 } else { 661 /* Put root on llist. */ 662 root->right = llist; 663 llist = root; 664 root = y; 665 } 666 } 667 } 668 669 /* 670 * Pass Two: Walk back up the two spines, flip the pointers 671 * and set max_free. The subtrees of the root go at the 672 * bottom of llist and rlist. 673 */ 674 ltree = root->left; 675 while (llist != NULL) { 676 y = llist->right; 677 llist->right = ltree; 678 vm_map_entry_set_max_free(llist); 679 ltree = llist; 680 llist = y; 681 } 682 rtree = root->right; 683 while (rlist != NULL) { 684 y = rlist->left; 685 rlist->left = rtree; 686 vm_map_entry_set_max_free(rlist); 687 rtree = rlist; 688 rlist = y; 689 } 690 691 /* 692 * Final assembly: add ltree and rtree as subtrees of root. 693 */ 694 root->left = ltree; 695 root->right = rtree; 696 vm_map_entry_set_max_free(root); 697 698 return (root); 699 } 700 701 /* 702 * vm_map_entry_{un,}link: 703 * 704 * Insert/remove entries from maps. 705 */ 706 static void 707 vm_map_entry_link(vm_map_t map, 708 vm_map_entry_t after_where, 709 vm_map_entry_t entry) 710 { 711 712 CTR4(KTR_VM, 713 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 714 map->nentries, entry, after_where); 715 map->nentries++; 716 entry->prev = after_where; 717 entry->next = after_where->next; 718 entry->next->prev = entry; 719 after_where->next = entry; 720 721 if (after_where != &map->header) { 722 if (after_where != map->root) 723 vm_map_entry_splay(after_where->start, map->root); 724 entry->right = after_where->right; 725 entry->left = after_where; 726 after_where->right = NULL; 727 after_where->adj_free = entry->start - after_where->end; 728 vm_map_entry_set_max_free(after_where); 729 } else { 730 entry->right = map->root; 731 entry->left = NULL; 732 } 733 entry->adj_free = (entry->next == &map->header ? map->max_offset : 734 entry->next->start) - entry->end; 735 vm_map_entry_set_max_free(entry); 736 map->root = entry; 737 } 738 739 static void 740 vm_map_entry_unlink(vm_map_t map, 741 vm_map_entry_t entry) 742 { 743 vm_map_entry_t next, prev, root; 744 745 if (entry != map->root) 746 vm_map_entry_splay(entry->start, map->root); 747 if (entry->left == NULL) 748 root = entry->right; 749 else { 750 root = vm_map_entry_splay(entry->start, entry->left); 751 root->right = entry->right; 752 root->adj_free = (entry->next == &map->header ? map->max_offset : 753 entry->next->start) - root->end; 754 vm_map_entry_set_max_free(root); 755 } 756 map->root = root; 757 758 prev = entry->prev; 759 next = entry->next; 760 next->prev = prev; 761 prev->next = next; 762 map->nentries--; 763 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 764 map->nentries, entry); 765 } 766 767 /* 768 * vm_map_entry_resize_free: 769 * 770 * Recompute the amount of free space following a vm_map_entry 771 * and propagate that value up the tree. Call this function after 772 * resizing a map entry in-place, that is, without a call to 773 * vm_map_entry_link() or _unlink(). 774 * 775 * The map must be locked, and leaves it so. 776 */ 777 static void 778 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 779 { 780 781 /* 782 * Using splay trees without parent pointers, propagating 783 * max_free up the tree is done by moving the entry to the 784 * root and making the change there. 785 */ 786 if (entry != map->root) 787 map->root = vm_map_entry_splay(entry->start, map->root); 788 789 entry->adj_free = (entry->next == &map->header ? map->max_offset : 790 entry->next->start) - entry->end; 791 vm_map_entry_set_max_free(entry); 792 } 793 794 /* 795 * vm_map_lookup_entry: [ internal use only ] 796 * 797 * Finds the map entry containing (or 798 * immediately preceding) the specified address 799 * in the given map; the entry is returned 800 * in the "entry" parameter. The boolean 801 * result indicates whether the address is 802 * actually contained in the map. 803 */ 804 boolean_t 805 vm_map_lookup_entry( 806 vm_map_t map, 807 vm_offset_t address, 808 vm_map_entry_t *entry) /* OUT */ 809 { 810 vm_map_entry_t cur; 811 812 cur = vm_map_entry_splay(address, map->root); 813 if (cur == NULL) 814 *entry = &map->header; 815 else { 816 map->root = cur; 817 818 if (address >= cur->start) { 819 *entry = cur; 820 if (cur->end > address) 821 return (TRUE); 822 } else 823 *entry = cur->prev; 824 } 825 return (FALSE); 826 } 827 828 /* 829 * vm_map_insert: 830 * 831 * Inserts the given whole VM object into the target 832 * map at the specified address range. The object's 833 * size should match that of the address range. 834 * 835 * Requires that the map be locked, and leaves it so. 836 * 837 * If object is non-NULL, ref count must be bumped by caller 838 * prior to making call to account for the new entry. 839 */ 840 int 841 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 842 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, 843 int cow) 844 { 845 vm_map_entry_t new_entry; 846 vm_map_entry_t prev_entry; 847 vm_map_entry_t temp_entry; 848 vm_eflags_t protoeflags; 849 850 /* 851 * Check that the start and end points are not bogus. 852 */ 853 if ((start < map->min_offset) || (end > map->max_offset) || 854 (start >= end)) 855 return (KERN_INVALID_ADDRESS); 856 857 /* 858 * Find the entry prior to the proposed starting address; if it's part 859 * of an existing entry, this range is bogus. 860 */ 861 if (vm_map_lookup_entry(map, start, &temp_entry)) 862 return (KERN_NO_SPACE); 863 864 prev_entry = temp_entry; 865 866 /* 867 * Assert that the next entry doesn't overlap the end point. 868 */ 869 if ((prev_entry->next != &map->header) && 870 (prev_entry->next->start < end)) 871 return (KERN_NO_SPACE); 872 873 protoeflags = 0; 874 875 if (cow & MAP_COPY_ON_WRITE) 876 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; 877 878 if (cow & MAP_NOFAULT) { 879 protoeflags |= MAP_ENTRY_NOFAULT; 880 881 KASSERT(object == NULL, 882 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 883 } 884 if (cow & MAP_DISABLE_SYNCER) 885 protoeflags |= MAP_ENTRY_NOSYNC; 886 if (cow & MAP_DISABLE_COREDUMP) 887 protoeflags |= MAP_ENTRY_NOCOREDUMP; 888 889 if (object != NULL) { 890 /* 891 * OBJ_ONEMAPPING must be cleared unless this mapping 892 * is trivially proven to be the only mapping for any 893 * of the object's pages. (Object granularity 894 * reference counting is insufficient to recognize 895 * aliases with precision.) 896 */ 897 VM_OBJECT_LOCK(object); 898 if (object->ref_count > 1 || object->shadow_count != 0) 899 vm_object_clear_flag(object, OBJ_ONEMAPPING); 900 VM_OBJECT_UNLOCK(object); 901 } 902 else if ((prev_entry != &map->header) && 903 (prev_entry->eflags == protoeflags) && 904 (prev_entry->end == start) && 905 (prev_entry->wired_count == 0) && 906 ((prev_entry->object.vm_object == NULL) || 907 vm_object_coalesce(prev_entry->object.vm_object, 908 prev_entry->offset, 909 (vm_size_t)(prev_entry->end - prev_entry->start), 910 (vm_size_t)(end - prev_entry->end)))) { 911 /* 912 * We were able to extend the object. Determine if we 913 * can extend the previous map entry to include the 914 * new range as well. 915 */ 916 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && 917 (prev_entry->protection == prot) && 918 (prev_entry->max_protection == max)) { 919 map->size += (end - prev_entry->end); 920 prev_entry->end = end; 921 vm_map_entry_resize_free(map, prev_entry); 922 vm_map_simplify_entry(map, prev_entry); 923 return (KERN_SUCCESS); 924 } 925 926 /* 927 * If we can extend the object but cannot extend the 928 * map entry, we have to create a new map entry. We 929 * must bump the ref count on the extended object to 930 * account for it. object may be NULL. 931 */ 932 object = prev_entry->object.vm_object; 933 offset = prev_entry->offset + 934 (prev_entry->end - prev_entry->start); 935 vm_object_reference(object); 936 } 937 938 /* 939 * NOTE: if conditionals fail, object can be NULL here. This occurs 940 * in things like the buffer map where we manage kva but do not manage 941 * backing objects. 942 */ 943 944 /* 945 * Create a new entry 946 */ 947 new_entry = vm_map_entry_create(map); 948 new_entry->start = start; 949 new_entry->end = end; 950 951 new_entry->eflags = protoeflags; 952 new_entry->object.vm_object = object; 953 new_entry->offset = offset; 954 new_entry->avail_ssize = 0; 955 956 new_entry->inheritance = VM_INHERIT_DEFAULT; 957 new_entry->protection = prot; 958 new_entry->max_protection = max; 959 new_entry->wired_count = 0; 960 961 /* 962 * Insert the new entry into the list 963 */ 964 vm_map_entry_link(map, prev_entry, new_entry); 965 map->size += new_entry->end - new_entry->start; 966 967 #if 0 968 /* 969 * Temporarily removed to avoid MAP_STACK panic, due to 970 * MAP_STACK being a huge hack. Will be added back in 971 * when MAP_STACK (and the user stack mapping) is fixed. 972 */ 973 /* 974 * It may be possible to simplify the entry 975 */ 976 vm_map_simplify_entry(map, new_entry); 977 #endif 978 979 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { 980 vm_map_pmap_enter(map, start, prot, 981 object, OFF_TO_IDX(offset), end - start, 982 cow & MAP_PREFAULT_PARTIAL); 983 } 984 985 return (KERN_SUCCESS); 986 } 987 988 /* 989 * vm_map_findspace: 990 * 991 * Find the first fit (lowest VM address) for "length" free bytes 992 * beginning at address >= start in the given map. 993 * 994 * In a vm_map_entry, "adj_free" is the amount of free space 995 * adjacent (higher address) to this entry, and "max_free" is the 996 * maximum amount of contiguous free space in its subtree. This 997 * allows finding a free region in one path down the tree, so 998 * O(log n) amortized with splay trees. 999 * 1000 * The map must be locked, and leaves it so. 1001 * 1002 * Returns: 0 on success, and starting address in *addr, 1003 * 1 if insufficient space. 1004 */ 1005 int 1006 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1007 vm_offset_t *addr) /* OUT */ 1008 { 1009 vm_map_entry_t entry; 1010 vm_offset_t end, st; 1011 1012 /* 1013 * Request must fit within min/max VM address and must avoid 1014 * address wrap. 1015 */ 1016 if (start < map->min_offset) 1017 start = map->min_offset; 1018 if (start + length > map->max_offset || start + length < start) 1019 return (1); 1020 1021 /* Empty tree means wide open address space. */ 1022 if (map->root == NULL) { 1023 *addr = start; 1024 goto found; 1025 } 1026 1027 /* 1028 * After splay, if start comes before root node, then there 1029 * must be a gap from start to the root. 1030 */ 1031 map->root = vm_map_entry_splay(start, map->root); 1032 if (start + length <= map->root->start) { 1033 *addr = start; 1034 goto found; 1035 } 1036 1037 /* 1038 * Root is the last node that might begin its gap before 1039 * start, and this is the last comparison where address 1040 * wrap might be a problem. 1041 */ 1042 st = (start > map->root->end) ? start : map->root->end; 1043 if (length <= map->root->end + map->root->adj_free - st) { 1044 *addr = st; 1045 goto found; 1046 } 1047 1048 /* With max_free, can immediately tell if no solution. */ 1049 entry = map->root->right; 1050 if (entry == NULL || length > entry->max_free) 1051 return (1); 1052 1053 /* 1054 * Search the right subtree in the order: left subtree, root, 1055 * right subtree (first fit). The previous splay implies that 1056 * all regions in the right subtree have addresses > start. 1057 */ 1058 while (entry != NULL) { 1059 if (entry->left != NULL && entry->left->max_free >= length) 1060 entry = entry->left; 1061 else if (entry->adj_free >= length) { 1062 *addr = entry->end; 1063 goto found; 1064 } else 1065 entry = entry->right; 1066 } 1067 1068 /* Can't get here, so panic if we do. */ 1069 panic("vm_map_findspace: max_free corrupt"); 1070 1071 found: 1072 /* Expand the kernel pmap, if necessary. */ 1073 if (map == kernel_map) { 1074 end = round_page(*addr + length); 1075 if (end > kernel_vm_end) 1076 pmap_growkernel(end); 1077 } 1078 return (0); 1079 } 1080 1081 /* 1082 * vm_map_find finds an unallocated region in the target address 1083 * map with the given length. The search is defined to be 1084 * first-fit from the specified address; the region found is 1085 * returned in the same parameter. 1086 * 1087 * If object is non-NULL, ref count must be bumped by caller 1088 * prior to making call to account for the new entry. 1089 */ 1090 int 1091 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1092 vm_offset_t *addr, /* IN/OUT */ 1093 vm_size_t length, boolean_t find_space, vm_prot_t prot, 1094 vm_prot_t max, int cow) 1095 { 1096 vm_offset_t start; 1097 int result; 1098 1099 start = *addr; 1100 vm_map_lock(map); 1101 if (find_space) { 1102 if (vm_map_findspace(map, start, length, addr)) { 1103 vm_map_unlock(map); 1104 return (KERN_NO_SPACE); 1105 } 1106 start = *addr; 1107 } 1108 result = vm_map_insert(map, object, offset, 1109 start, start + length, prot, max, cow); 1110 vm_map_unlock(map); 1111 return (result); 1112 } 1113 1114 /* 1115 * vm_map_simplify_entry: 1116 * 1117 * Simplify the given map entry by merging with either neighbor. This 1118 * routine also has the ability to merge with both neighbors. 1119 * 1120 * The map must be locked. 1121 * 1122 * This routine guarentees that the passed entry remains valid (though 1123 * possibly extended). When merging, this routine may delete one or 1124 * both neighbors. 1125 */ 1126 void 1127 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1128 { 1129 vm_map_entry_t next, prev; 1130 vm_size_t prevsize, esize; 1131 1132 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) 1133 return; 1134 1135 prev = entry->prev; 1136 if (prev != &map->header) { 1137 prevsize = prev->end - prev->start; 1138 if ( (prev->end == entry->start) && 1139 (prev->object.vm_object == entry->object.vm_object) && 1140 (!prev->object.vm_object || 1141 (prev->offset + prevsize == entry->offset)) && 1142 (prev->eflags == entry->eflags) && 1143 (prev->protection == entry->protection) && 1144 (prev->max_protection == entry->max_protection) && 1145 (prev->inheritance == entry->inheritance) && 1146 (prev->wired_count == entry->wired_count)) { 1147 vm_map_entry_unlink(map, prev); 1148 entry->start = prev->start; 1149 entry->offset = prev->offset; 1150 if (entry->prev != &map->header) 1151 vm_map_entry_resize_free(map, entry->prev); 1152 if (prev->object.vm_object) 1153 vm_object_deallocate(prev->object.vm_object); 1154 vm_map_entry_dispose(map, prev); 1155 } 1156 } 1157 1158 next = entry->next; 1159 if (next != &map->header) { 1160 esize = entry->end - entry->start; 1161 if ((entry->end == next->start) && 1162 (next->object.vm_object == entry->object.vm_object) && 1163 (!entry->object.vm_object || 1164 (entry->offset + esize == next->offset)) && 1165 (next->eflags == entry->eflags) && 1166 (next->protection == entry->protection) && 1167 (next->max_protection == entry->max_protection) && 1168 (next->inheritance == entry->inheritance) && 1169 (next->wired_count == entry->wired_count)) { 1170 vm_map_entry_unlink(map, next); 1171 entry->end = next->end; 1172 vm_map_entry_resize_free(map, entry); 1173 if (next->object.vm_object) 1174 vm_object_deallocate(next->object.vm_object); 1175 vm_map_entry_dispose(map, next); 1176 } 1177 } 1178 } 1179 /* 1180 * vm_map_clip_start: [ internal use only ] 1181 * 1182 * Asserts that the given entry begins at or after 1183 * the specified address; if necessary, 1184 * it splits the entry into two. 1185 */ 1186 #define vm_map_clip_start(map, entry, startaddr) \ 1187 { \ 1188 if (startaddr > entry->start) \ 1189 _vm_map_clip_start(map, entry, startaddr); \ 1190 } 1191 1192 /* 1193 * This routine is called only when it is known that 1194 * the entry must be split. 1195 */ 1196 static void 1197 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1198 { 1199 vm_map_entry_t new_entry; 1200 1201 /* 1202 * Split off the front portion -- note that we must insert the new 1203 * entry BEFORE this one, so that this entry has the specified 1204 * starting address. 1205 */ 1206 vm_map_simplify_entry(map, entry); 1207 1208 /* 1209 * If there is no object backing this entry, we might as well create 1210 * one now. If we defer it, an object can get created after the map 1211 * is clipped, and individual objects will be created for the split-up 1212 * map. This is a bit of a hack, but is also about the best place to 1213 * put this improvement. 1214 */ 1215 if (entry->object.vm_object == NULL && !map->system_map) { 1216 vm_object_t object; 1217 object = vm_object_allocate(OBJT_DEFAULT, 1218 atop(entry->end - entry->start)); 1219 entry->object.vm_object = object; 1220 entry->offset = 0; 1221 } 1222 1223 new_entry = vm_map_entry_create(map); 1224 *new_entry = *entry; 1225 1226 new_entry->end = start; 1227 entry->offset += (start - entry->start); 1228 entry->start = start; 1229 1230 vm_map_entry_link(map, entry->prev, new_entry); 1231 1232 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1233 vm_object_reference(new_entry->object.vm_object); 1234 } 1235 } 1236 1237 /* 1238 * vm_map_clip_end: [ internal use only ] 1239 * 1240 * Asserts that the given entry ends at or before 1241 * the specified address; if necessary, 1242 * it splits the entry into two. 1243 */ 1244 #define vm_map_clip_end(map, entry, endaddr) \ 1245 { \ 1246 if ((endaddr) < (entry->end)) \ 1247 _vm_map_clip_end((map), (entry), (endaddr)); \ 1248 } 1249 1250 /* 1251 * This routine is called only when it is known that 1252 * the entry must be split. 1253 */ 1254 static void 1255 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1256 { 1257 vm_map_entry_t new_entry; 1258 1259 /* 1260 * If there is no object backing this entry, we might as well create 1261 * one now. If we defer it, an object can get created after the map 1262 * is clipped, and individual objects will be created for the split-up 1263 * map. This is a bit of a hack, but is also about the best place to 1264 * put this improvement. 1265 */ 1266 if (entry->object.vm_object == NULL && !map->system_map) { 1267 vm_object_t object; 1268 object = vm_object_allocate(OBJT_DEFAULT, 1269 atop(entry->end - entry->start)); 1270 entry->object.vm_object = object; 1271 entry->offset = 0; 1272 } 1273 1274 /* 1275 * Create a new entry and insert it AFTER the specified entry 1276 */ 1277 new_entry = vm_map_entry_create(map); 1278 *new_entry = *entry; 1279 1280 new_entry->start = entry->end = end; 1281 new_entry->offset += (end - entry->start); 1282 1283 vm_map_entry_link(map, entry, new_entry); 1284 1285 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1286 vm_object_reference(new_entry->object.vm_object); 1287 } 1288 } 1289 1290 /* 1291 * VM_MAP_RANGE_CHECK: [ internal use only ] 1292 * 1293 * Asserts that the starting and ending region 1294 * addresses fall within the valid range of the map. 1295 */ 1296 #define VM_MAP_RANGE_CHECK(map, start, end) \ 1297 { \ 1298 if (start < vm_map_min(map)) \ 1299 start = vm_map_min(map); \ 1300 if (end > vm_map_max(map)) \ 1301 end = vm_map_max(map); \ 1302 if (start > end) \ 1303 start = end; \ 1304 } 1305 1306 /* 1307 * vm_map_submap: [ kernel use only ] 1308 * 1309 * Mark the given range as handled by a subordinate map. 1310 * 1311 * This range must have been created with vm_map_find, 1312 * and no other operations may have been performed on this 1313 * range prior to calling vm_map_submap. 1314 * 1315 * Only a limited number of operations can be performed 1316 * within this rage after calling vm_map_submap: 1317 * vm_fault 1318 * [Don't try vm_map_copy!] 1319 * 1320 * To remove a submapping, one must first remove the 1321 * range from the superior map, and then destroy the 1322 * submap (if desired). [Better yet, don't try it.] 1323 */ 1324 int 1325 vm_map_submap( 1326 vm_map_t map, 1327 vm_offset_t start, 1328 vm_offset_t end, 1329 vm_map_t submap) 1330 { 1331 vm_map_entry_t entry; 1332 int result = KERN_INVALID_ARGUMENT; 1333 1334 vm_map_lock(map); 1335 1336 VM_MAP_RANGE_CHECK(map, start, end); 1337 1338 if (vm_map_lookup_entry(map, start, &entry)) { 1339 vm_map_clip_start(map, entry, start); 1340 } else 1341 entry = entry->next; 1342 1343 vm_map_clip_end(map, entry, end); 1344 1345 if ((entry->start == start) && (entry->end == end) && 1346 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1347 (entry->object.vm_object == NULL)) { 1348 entry->object.sub_map = submap; 1349 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1350 result = KERN_SUCCESS; 1351 } 1352 vm_map_unlock(map); 1353 1354 return (result); 1355 } 1356 1357 /* 1358 * The maximum number of pages to map 1359 */ 1360 #define MAX_INIT_PT 96 1361 1362 /* 1363 * vm_map_pmap_enter: 1364 * 1365 * Preload read-only mappings for the given object into the specified 1366 * map. This eliminates the soft faults on process startup and 1367 * immediately after an mmap(2). 1368 */ 1369 void 1370 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1371 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1372 { 1373 vm_offset_t tmpidx; 1374 int psize; 1375 vm_page_t p, mpte; 1376 boolean_t are_queues_locked; 1377 1378 if ((prot & VM_PROT_READ) == 0 || object == NULL) 1379 return; 1380 VM_OBJECT_LOCK(object); 1381 if (object->type == OBJT_DEVICE) { 1382 pmap_object_init_pt(map->pmap, addr, object, pindex, size); 1383 goto unlock_return; 1384 } 1385 1386 psize = atop(size); 1387 1388 if (object->type != OBJT_VNODE || 1389 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) && 1390 (object->resident_page_count > MAX_INIT_PT))) { 1391 goto unlock_return; 1392 } 1393 1394 if (psize + pindex > object->size) { 1395 if (object->size < pindex) 1396 goto unlock_return; 1397 psize = object->size - pindex; 1398 } 1399 1400 are_queues_locked = FALSE; 1401 mpte = NULL; 1402 1403 if ((p = TAILQ_FIRST(&object->memq)) != NULL) { 1404 if (p->pindex < pindex) { 1405 p = vm_page_splay(pindex, object->root); 1406 if ((object->root = p)->pindex < pindex) 1407 p = TAILQ_NEXT(p, listq); 1408 } 1409 } 1410 /* 1411 * Assert: the variable p is either (1) the page with the 1412 * least pindex greater than or equal to the parameter pindex 1413 * or (2) NULL. 1414 */ 1415 for (; 1416 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1417 p = TAILQ_NEXT(p, listq)) { 1418 /* 1419 * don't allow an madvise to blow away our really 1420 * free pages allocating pv entries. 1421 */ 1422 if ((flags & MAP_PREFAULT_MADVISE) && 1423 cnt.v_free_count < cnt.v_free_reserved) { 1424 break; 1425 } 1426 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 1427 (p->busy == 0) && 1428 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) { 1429 if (!are_queues_locked) { 1430 are_queues_locked = TRUE; 1431 vm_page_lock_queues(); 1432 } 1433 if ((p->queue - p->pc) == PQ_CACHE) 1434 vm_page_deactivate(p); 1435 mpte = pmap_enter_quick(map->pmap, 1436 addr + ptoa(tmpidx), p, mpte); 1437 } 1438 } 1439 if (are_queues_locked) 1440 vm_page_unlock_queues(); 1441 unlock_return: 1442 VM_OBJECT_UNLOCK(object); 1443 } 1444 1445 /* 1446 * vm_map_protect: 1447 * 1448 * Sets the protection of the specified address 1449 * region in the target map. If "set_max" is 1450 * specified, the maximum protection is to be set; 1451 * otherwise, only the current protection is affected. 1452 */ 1453 int 1454 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 1455 vm_prot_t new_prot, boolean_t set_max) 1456 { 1457 vm_map_entry_t current; 1458 vm_map_entry_t entry; 1459 1460 vm_map_lock(map); 1461 1462 VM_MAP_RANGE_CHECK(map, start, end); 1463 1464 if (vm_map_lookup_entry(map, start, &entry)) { 1465 vm_map_clip_start(map, entry, start); 1466 } else { 1467 entry = entry->next; 1468 } 1469 1470 /* 1471 * Make a first pass to check for protection violations. 1472 */ 1473 current = entry; 1474 while ((current != &map->header) && (current->start < end)) { 1475 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 1476 vm_map_unlock(map); 1477 return (KERN_INVALID_ARGUMENT); 1478 } 1479 if ((new_prot & current->max_protection) != new_prot) { 1480 vm_map_unlock(map); 1481 return (KERN_PROTECTION_FAILURE); 1482 } 1483 current = current->next; 1484 } 1485 1486 /* 1487 * Go back and fix up protections. [Note that clipping is not 1488 * necessary the second time.] 1489 */ 1490 current = entry; 1491 while ((current != &map->header) && (current->start < end)) { 1492 vm_prot_t old_prot; 1493 1494 vm_map_clip_end(map, current, end); 1495 1496 old_prot = current->protection; 1497 if (set_max) 1498 current->protection = 1499 (current->max_protection = new_prot) & 1500 old_prot; 1501 else 1502 current->protection = new_prot; 1503 1504 /* 1505 * Update physical map if necessary. Worry about copy-on-write 1506 * here -- CHECK THIS XXX 1507 */ 1508 if (current->protection != old_prot) { 1509 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 1510 VM_PROT_ALL) 1511 pmap_protect(map->pmap, current->start, 1512 current->end, 1513 current->protection & MASK(current)); 1514 #undef MASK 1515 } 1516 vm_map_simplify_entry(map, current); 1517 current = current->next; 1518 } 1519 vm_map_unlock(map); 1520 return (KERN_SUCCESS); 1521 } 1522 1523 /* 1524 * vm_map_madvise: 1525 * 1526 * This routine traverses a processes map handling the madvise 1527 * system call. Advisories are classified as either those effecting 1528 * the vm_map_entry structure, or those effecting the underlying 1529 * objects. 1530 */ 1531 int 1532 vm_map_madvise( 1533 vm_map_t map, 1534 vm_offset_t start, 1535 vm_offset_t end, 1536 int behav) 1537 { 1538 vm_map_entry_t current, entry; 1539 int modify_map = 0; 1540 1541 /* 1542 * Some madvise calls directly modify the vm_map_entry, in which case 1543 * we need to use an exclusive lock on the map and we need to perform 1544 * various clipping operations. Otherwise we only need a read-lock 1545 * on the map. 1546 */ 1547 switch(behav) { 1548 case MADV_NORMAL: 1549 case MADV_SEQUENTIAL: 1550 case MADV_RANDOM: 1551 case MADV_NOSYNC: 1552 case MADV_AUTOSYNC: 1553 case MADV_NOCORE: 1554 case MADV_CORE: 1555 modify_map = 1; 1556 vm_map_lock(map); 1557 break; 1558 case MADV_WILLNEED: 1559 case MADV_DONTNEED: 1560 case MADV_FREE: 1561 vm_map_lock_read(map); 1562 break; 1563 default: 1564 return (KERN_INVALID_ARGUMENT); 1565 } 1566 1567 /* 1568 * Locate starting entry and clip if necessary. 1569 */ 1570 VM_MAP_RANGE_CHECK(map, start, end); 1571 1572 if (vm_map_lookup_entry(map, start, &entry)) { 1573 if (modify_map) 1574 vm_map_clip_start(map, entry, start); 1575 } else { 1576 entry = entry->next; 1577 } 1578 1579 if (modify_map) { 1580 /* 1581 * madvise behaviors that are implemented in the vm_map_entry. 1582 * 1583 * We clip the vm_map_entry so that behavioral changes are 1584 * limited to the specified address range. 1585 */ 1586 for (current = entry; 1587 (current != &map->header) && (current->start < end); 1588 current = current->next 1589 ) { 1590 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1591 continue; 1592 1593 vm_map_clip_end(map, current, end); 1594 1595 switch (behav) { 1596 case MADV_NORMAL: 1597 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 1598 break; 1599 case MADV_SEQUENTIAL: 1600 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 1601 break; 1602 case MADV_RANDOM: 1603 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 1604 break; 1605 case MADV_NOSYNC: 1606 current->eflags |= MAP_ENTRY_NOSYNC; 1607 break; 1608 case MADV_AUTOSYNC: 1609 current->eflags &= ~MAP_ENTRY_NOSYNC; 1610 break; 1611 case MADV_NOCORE: 1612 current->eflags |= MAP_ENTRY_NOCOREDUMP; 1613 break; 1614 case MADV_CORE: 1615 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 1616 break; 1617 default: 1618 break; 1619 } 1620 vm_map_simplify_entry(map, current); 1621 } 1622 vm_map_unlock(map); 1623 } else { 1624 vm_pindex_t pindex; 1625 int count; 1626 1627 /* 1628 * madvise behaviors that are implemented in the underlying 1629 * vm_object. 1630 * 1631 * Since we don't clip the vm_map_entry, we have to clip 1632 * the vm_object pindex and count. 1633 */ 1634 for (current = entry; 1635 (current != &map->header) && (current->start < end); 1636 current = current->next 1637 ) { 1638 vm_offset_t useStart; 1639 1640 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 1641 continue; 1642 1643 pindex = OFF_TO_IDX(current->offset); 1644 count = atop(current->end - current->start); 1645 useStart = current->start; 1646 1647 if (current->start < start) { 1648 pindex += atop(start - current->start); 1649 count -= atop(start - current->start); 1650 useStart = start; 1651 } 1652 if (current->end > end) 1653 count -= atop(current->end - end); 1654 1655 if (count <= 0) 1656 continue; 1657 1658 vm_object_madvise(current->object.vm_object, 1659 pindex, count, behav); 1660 if (behav == MADV_WILLNEED) { 1661 vm_map_pmap_enter(map, 1662 useStart, 1663 current->protection, 1664 current->object.vm_object, 1665 pindex, 1666 (count << PAGE_SHIFT), 1667 MAP_PREFAULT_MADVISE 1668 ); 1669 } 1670 } 1671 vm_map_unlock_read(map); 1672 } 1673 return (0); 1674 } 1675 1676 1677 /* 1678 * vm_map_inherit: 1679 * 1680 * Sets the inheritance of the specified address 1681 * range in the target map. Inheritance 1682 * affects how the map will be shared with 1683 * child maps at the time of vm_map_fork. 1684 */ 1685 int 1686 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 1687 vm_inherit_t new_inheritance) 1688 { 1689 vm_map_entry_t entry; 1690 vm_map_entry_t temp_entry; 1691 1692 switch (new_inheritance) { 1693 case VM_INHERIT_NONE: 1694 case VM_INHERIT_COPY: 1695 case VM_INHERIT_SHARE: 1696 break; 1697 default: 1698 return (KERN_INVALID_ARGUMENT); 1699 } 1700 vm_map_lock(map); 1701 VM_MAP_RANGE_CHECK(map, start, end); 1702 if (vm_map_lookup_entry(map, start, &temp_entry)) { 1703 entry = temp_entry; 1704 vm_map_clip_start(map, entry, start); 1705 } else 1706 entry = temp_entry->next; 1707 while ((entry != &map->header) && (entry->start < end)) { 1708 vm_map_clip_end(map, entry, end); 1709 entry->inheritance = new_inheritance; 1710 vm_map_simplify_entry(map, entry); 1711 entry = entry->next; 1712 } 1713 vm_map_unlock(map); 1714 return (KERN_SUCCESS); 1715 } 1716 1717 /* 1718 * vm_map_unwire: 1719 * 1720 * Implements both kernel and user unwiring. 1721 */ 1722 int 1723 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1724 int flags) 1725 { 1726 vm_map_entry_t entry, first_entry, tmp_entry; 1727 vm_offset_t saved_start; 1728 unsigned int last_timestamp; 1729 int rv; 1730 boolean_t need_wakeup, result, user_unwire; 1731 1732 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1733 vm_map_lock(map); 1734 VM_MAP_RANGE_CHECK(map, start, end); 1735 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1736 if (flags & VM_MAP_WIRE_HOLESOK) 1737 first_entry = first_entry->next; 1738 else { 1739 vm_map_unlock(map); 1740 return (KERN_INVALID_ADDRESS); 1741 } 1742 } 1743 last_timestamp = map->timestamp; 1744 entry = first_entry; 1745 while (entry != &map->header && entry->start < end) { 1746 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1747 /* 1748 * We have not yet clipped the entry. 1749 */ 1750 saved_start = (start >= entry->start) ? start : 1751 entry->start; 1752 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1753 if (vm_map_unlock_and_wait(map, user_unwire)) { 1754 /* 1755 * Allow interruption of user unwiring? 1756 */ 1757 } 1758 vm_map_lock(map); 1759 if (last_timestamp+1 != map->timestamp) { 1760 /* 1761 * Look again for the entry because the map was 1762 * modified while it was unlocked. 1763 * Specifically, the entry may have been 1764 * clipped, merged, or deleted. 1765 */ 1766 if (!vm_map_lookup_entry(map, saved_start, 1767 &tmp_entry)) { 1768 if (flags & VM_MAP_WIRE_HOLESOK) 1769 tmp_entry = tmp_entry->next; 1770 else { 1771 if (saved_start == start) { 1772 /* 1773 * First_entry has been deleted. 1774 */ 1775 vm_map_unlock(map); 1776 return (KERN_INVALID_ADDRESS); 1777 } 1778 end = saved_start; 1779 rv = KERN_INVALID_ADDRESS; 1780 goto done; 1781 } 1782 } 1783 if (entry == first_entry) 1784 first_entry = tmp_entry; 1785 else 1786 first_entry = NULL; 1787 entry = tmp_entry; 1788 } 1789 last_timestamp = map->timestamp; 1790 continue; 1791 } 1792 vm_map_clip_start(map, entry, start); 1793 vm_map_clip_end(map, entry, end); 1794 /* 1795 * Mark the entry in case the map lock is released. (See 1796 * above.) 1797 */ 1798 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1799 /* 1800 * Check the map for holes in the specified region. 1801 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 1802 */ 1803 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 1804 (entry->end < end && (entry->next == &map->header || 1805 entry->next->start > entry->end))) { 1806 end = entry->end; 1807 rv = KERN_INVALID_ADDRESS; 1808 goto done; 1809 } 1810 /* 1811 * If system unwiring, require that the entry is system wired. 1812 */ 1813 if (!user_unwire && 1814 vm_map_entry_system_wired_count(entry) == 0) { 1815 end = entry->end; 1816 rv = KERN_INVALID_ARGUMENT; 1817 goto done; 1818 } 1819 entry = entry->next; 1820 } 1821 rv = KERN_SUCCESS; 1822 done: 1823 need_wakeup = FALSE; 1824 if (first_entry == NULL) { 1825 result = vm_map_lookup_entry(map, start, &first_entry); 1826 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 1827 first_entry = first_entry->next; 1828 else 1829 KASSERT(result, ("vm_map_unwire: lookup failed")); 1830 } 1831 entry = first_entry; 1832 while (entry != &map->header && entry->start < end) { 1833 if (rv == KERN_SUCCESS && (!user_unwire || 1834 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 1835 if (user_unwire) 1836 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1837 entry->wired_count--; 1838 if (entry->wired_count == 0) { 1839 /* 1840 * Retain the map lock. 1841 */ 1842 vm_fault_unwire(map, entry->start, entry->end, 1843 entry->object.vm_object != NULL && 1844 entry->object.vm_object->type == OBJT_DEVICE); 1845 } 1846 } 1847 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 1848 ("vm_map_unwire: in-transition flag missing")); 1849 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 1850 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 1851 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 1852 need_wakeup = TRUE; 1853 } 1854 vm_map_simplify_entry(map, entry); 1855 entry = entry->next; 1856 } 1857 vm_map_unlock(map); 1858 if (need_wakeup) 1859 vm_map_wakeup(map); 1860 return (rv); 1861 } 1862 1863 /* 1864 * vm_map_wire: 1865 * 1866 * Implements both kernel and user wiring. 1867 */ 1868 int 1869 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 1870 int flags) 1871 { 1872 vm_map_entry_t entry, first_entry, tmp_entry; 1873 vm_offset_t saved_end, saved_start; 1874 unsigned int last_timestamp; 1875 int rv; 1876 boolean_t fictitious, need_wakeup, result, user_wire; 1877 1878 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 1879 vm_map_lock(map); 1880 VM_MAP_RANGE_CHECK(map, start, end); 1881 if (!vm_map_lookup_entry(map, start, &first_entry)) { 1882 if (flags & VM_MAP_WIRE_HOLESOK) 1883 first_entry = first_entry->next; 1884 else { 1885 vm_map_unlock(map); 1886 return (KERN_INVALID_ADDRESS); 1887 } 1888 } 1889 last_timestamp = map->timestamp; 1890 entry = first_entry; 1891 while (entry != &map->header && entry->start < end) { 1892 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1893 /* 1894 * We have not yet clipped the entry. 1895 */ 1896 saved_start = (start >= entry->start) ? start : 1897 entry->start; 1898 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1899 if (vm_map_unlock_and_wait(map, user_wire)) { 1900 /* 1901 * Allow interruption of user wiring? 1902 */ 1903 } 1904 vm_map_lock(map); 1905 if (last_timestamp + 1 != map->timestamp) { 1906 /* 1907 * Look again for the entry because the map was 1908 * modified while it was unlocked. 1909 * Specifically, the entry may have been 1910 * clipped, merged, or deleted. 1911 */ 1912 if (!vm_map_lookup_entry(map, saved_start, 1913 &tmp_entry)) { 1914 if (flags & VM_MAP_WIRE_HOLESOK) 1915 tmp_entry = tmp_entry->next; 1916 else { 1917 if (saved_start == start) { 1918 /* 1919 * first_entry has been deleted. 1920 */ 1921 vm_map_unlock(map); 1922 return (KERN_INVALID_ADDRESS); 1923 } 1924 end = saved_start; 1925 rv = KERN_INVALID_ADDRESS; 1926 goto done; 1927 } 1928 } 1929 if (entry == first_entry) 1930 first_entry = tmp_entry; 1931 else 1932 first_entry = NULL; 1933 entry = tmp_entry; 1934 } 1935 last_timestamp = map->timestamp; 1936 continue; 1937 } 1938 vm_map_clip_start(map, entry, start); 1939 vm_map_clip_end(map, entry, end); 1940 /* 1941 * Mark the entry in case the map lock is released. (See 1942 * above.) 1943 */ 1944 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 1945 /* 1946 * 1947 */ 1948 if (entry->wired_count == 0) { 1949 entry->wired_count++; 1950 saved_start = entry->start; 1951 saved_end = entry->end; 1952 fictitious = entry->object.vm_object != NULL && 1953 entry->object.vm_object->type == OBJT_DEVICE; 1954 /* 1955 * Release the map lock, relying on the in-transition 1956 * mark. 1957 */ 1958 vm_map_unlock(map); 1959 rv = vm_fault_wire(map, saved_start, saved_end, 1960 user_wire, fictitious); 1961 vm_map_lock(map); 1962 if (last_timestamp + 1 != map->timestamp) { 1963 /* 1964 * Look again for the entry because the map was 1965 * modified while it was unlocked. The entry 1966 * may have been clipped, but NOT merged or 1967 * deleted. 1968 */ 1969 result = vm_map_lookup_entry(map, saved_start, 1970 &tmp_entry); 1971 KASSERT(result, ("vm_map_wire: lookup failed")); 1972 if (entry == first_entry) 1973 first_entry = tmp_entry; 1974 else 1975 first_entry = NULL; 1976 entry = tmp_entry; 1977 while (entry->end < saved_end) { 1978 if (rv != KERN_SUCCESS) { 1979 KASSERT(entry->wired_count == 1, 1980 ("vm_map_wire: bad count")); 1981 entry->wired_count = -1; 1982 } 1983 entry = entry->next; 1984 } 1985 } 1986 last_timestamp = map->timestamp; 1987 if (rv != KERN_SUCCESS) { 1988 KASSERT(entry->wired_count == 1, 1989 ("vm_map_wire: bad count")); 1990 /* 1991 * Assign an out-of-range value to represent 1992 * the failure to wire this entry. 1993 */ 1994 entry->wired_count = -1; 1995 end = entry->end; 1996 goto done; 1997 } 1998 } else if (!user_wire || 1999 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2000 entry->wired_count++; 2001 } 2002 /* 2003 * Check the map for holes in the specified region. 2004 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2005 */ 2006 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2007 (entry->end < end && (entry->next == &map->header || 2008 entry->next->start > entry->end))) { 2009 end = entry->end; 2010 rv = KERN_INVALID_ADDRESS; 2011 goto done; 2012 } 2013 entry = entry->next; 2014 } 2015 rv = KERN_SUCCESS; 2016 done: 2017 need_wakeup = FALSE; 2018 if (first_entry == NULL) { 2019 result = vm_map_lookup_entry(map, start, &first_entry); 2020 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2021 first_entry = first_entry->next; 2022 else 2023 KASSERT(result, ("vm_map_wire: lookup failed")); 2024 } 2025 entry = first_entry; 2026 while (entry != &map->header && entry->start < end) { 2027 if (rv == KERN_SUCCESS) { 2028 if (user_wire) 2029 entry->eflags |= MAP_ENTRY_USER_WIRED; 2030 } else if (entry->wired_count == -1) { 2031 /* 2032 * Wiring failed on this entry. Thus, unwiring is 2033 * unnecessary. 2034 */ 2035 entry->wired_count = 0; 2036 } else { 2037 if (!user_wire || 2038 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 2039 entry->wired_count--; 2040 if (entry->wired_count == 0) { 2041 /* 2042 * Retain the map lock. 2043 */ 2044 vm_fault_unwire(map, entry->start, entry->end, 2045 entry->object.vm_object != NULL && 2046 entry->object.vm_object->type == OBJT_DEVICE); 2047 } 2048 } 2049 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, 2050 ("vm_map_wire: in-transition flag missing")); 2051 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2052 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2053 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2054 need_wakeup = TRUE; 2055 } 2056 vm_map_simplify_entry(map, entry); 2057 entry = entry->next; 2058 } 2059 vm_map_unlock(map); 2060 if (need_wakeup) 2061 vm_map_wakeup(map); 2062 return (rv); 2063 } 2064 2065 /* 2066 * vm_map_sync 2067 * 2068 * Push any dirty cached pages in the address range to their pager. 2069 * If syncio is TRUE, dirty pages are written synchronously. 2070 * If invalidate is TRUE, any cached pages are freed as well. 2071 * 2072 * If the size of the region from start to end is zero, we are 2073 * supposed to flush all modified pages within the region containing 2074 * start. Unfortunately, a region can be split or coalesced with 2075 * neighboring regions, making it difficult to determine what the 2076 * original region was. Therefore, we approximate this requirement by 2077 * flushing the current region containing start. 2078 * 2079 * Returns an error if any part of the specified range is not mapped. 2080 */ 2081 int 2082 vm_map_sync( 2083 vm_map_t map, 2084 vm_offset_t start, 2085 vm_offset_t end, 2086 boolean_t syncio, 2087 boolean_t invalidate) 2088 { 2089 vm_map_entry_t current; 2090 vm_map_entry_t entry; 2091 vm_size_t size; 2092 vm_object_t object; 2093 vm_ooffset_t offset; 2094 2095 vm_map_lock_read(map); 2096 VM_MAP_RANGE_CHECK(map, start, end); 2097 if (!vm_map_lookup_entry(map, start, &entry)) { 2098 vm_map_unlock_read(map); 2099 return (KERN_INVALID_ADDRESS); 2100 } else if (start == end) { 2101 start = entry->start; 2102 end = entry->end; 2103 } 2104 /* 2105 * Make a first pass to check for user-wired memory and holes. 2106 */ 2107 for (current = entry; current->start < end; current = current->next) { 2108 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2109 vm_map_unlock_read(map); 2110 return (KERN_INVALID_ARGUMENT); 2111 } 2112 if (end > current->end && 2113 (current->next == &map->header || 2114 current->end != current->next->start)) { 2115 vm_map_unlock_read(map); 2116 return (KERN_INVALID_ADDRESS); 2117 } 2118 } 2119 2120 if (invalidate) { 2121 VM_LOCK_GIANT(); 2122 pmap_remove(map->pmap, start, end); 2123 VM_UNLOCK_GIANT(); 2124 } 2125 /* 2126 * Make a second pass, cleaning/uncaching pages from the indicated 2127 * objects as we go. 2128 */ 2129 for (current = entry; current->start < end; current = current->next) { 2130 offset = current->offset + (start - current->start); 2131 size = (end <= current->end ? end : current->end) - start; 2132 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2133 vm_map_t smap; 2134 vm_map_entry_t tentry; 2135 vm_size_t tsize; 2136 2137 smap = current->object.sub_map; 2138 vm_map_lock_read(smap); 2139 (void) vm_map_lookup_entry(smap, offset, &tentry); 2140 tsize = tentry->end - offset; 2141 if (tsize < size) 2142 size = tsize; 2143 object = tentry->object.vm_object; 2144 offset = tentry->offset + (offset - tentry->start); 2145 vm_map_unlock_read(smap); 2146 } else { 2147 object = current->object.vm_object; 2148 } 2149 vm_object_sync(object, offset, size, syncio, invalidate); 2150 start += size; 2151 } 2152 2153 vm_map_unlock_read(map); 2154 return (KERN_SUCCESS); 2155 } 2156 2157 /* 2158 * vm_map_entry_unwire: [ internal use only ] 2159 * 2160 * Make the region specified by this entry pageable. 2161 * 2162 * The map in question should be locked. 2163 * [This is the reason for this routine's existence.] 2164 */ 2165 static void 2166 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2167 { 2168 vm_fault_unwire(map, entry->start, entry->end, 2169 entry->object.vm_object != NULL && 2170 entry->object.vm_object->type == OBJT_DEVICE); 2171 entry->wired_count = 0; 2172 } 2173 2174 /* 2175 * vm_map_entry_delete: [ internal use only ] 2176 * 2177 * Deallocate the given entry from the target map. 2178 */ 2179 static void 2180 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2181 { 2182 vm_object_t object; 2183 vm_pindex_t offidxstart, offidxend, count; 2184 2185 vm_map_entry_unlink(map, entry); 2186 map->size -= entry->end - entry->start; 2187 2188 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2189 (object = entry->object.vm_object) != NULL) { 2190 count = OFF_TO_IDX(entry->end - entry->start); 2191 offidxstart = OFF_TO_IDX(entry->offset); 2192 offidxend = offidxstart + count; 2193 VM_OBJECT_LOCK(object); 2194 if (object->ref_count != 1 && 2195 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2196 object == kernel_object || object == kmem_object) && 2197 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { 2198 vm_object_collapse(object); 2199 vm_object_page_remove(object, offidxstart, offidxend, FALSE); 2200 if (object->type == OBJT_SWAP) 2201 swap_pager_freespace(object, offidxstart, count); 2202 if (offidxend >= object->size && 2203 offidxstart < object->size) 2204 object->size = offidxstart; 2205 } 2206 VM_OBJECT_UNLOCK(object); 2207 vm_object_deallocate(object); 2208 } 2209 2210 vm_map_entry_dispose(map, entry); 2211 } 2212 2213 /* 2214 * vm_map_delete: [ internal use only ] 2215 * 2216 * Deallocates the given address range from the target 2217 * map. 2218 */ 2219 int 2220 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 2221 { 2222 vm_map_entry_t entry; 2223 vm_map_entry_t first_entry; 2224 2225 /* 2226 * Find the start of the region, and clip it 2227 */ 2228 if (!vm_map_lookup_entry(map, start, &first_entry)) 2229 entry = first_entry->next; 2230 else { 2231 entry = first_entry; 2232 vm_map_clip_start(map, entry, start); 2233 } 2234 2235 /* 2236 * Step through all entries in this region 2237 */ 2238 while ((entry != &map->header) && (entry->start < end)) { 2239 vm_map_entry_t next; 2240 2241 /* 2242 * Wait for wiring or unwiring of an entry to complete. 2243 * Also wait for any system wirings to disappear on 2244 * user maps. 2245 */ 2246 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 2247 (vm_map_pmap(map) != kernel_pmap && 2248 vm_map_entry_system_wired_count(entry) != 0)) { 2249 unsigned int last_timestamp; 2250 vm_offset_t saved_start; 2251 vm_map_entry_t tmp_entry; 2252 2253 saved_start = entry->start; 2254 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2255 last_timestamp = map->timestamp; 2256 (void) vm_map_unlock_and_wait(map, FALSE); 2257 vm_map_lock(map); 2258 if (last_timestamp + 1 != map->timestamp) { 2259 /* 2260 * Look again for the entry because the map was 2261 * modified while it was unlocked. 2262 * Specifically, the entry may have been 2263 * clipped, merged, or deleted. 2264 */ 2265 if (!vm_map_lookup_entry(map, saved_start, 2266 &tmp_entry)) 2267 entry = tmp_entry->next; 2268 else { 2269 entry = tmp_entry; 2270 vm_map_clip_start(map, entry, 2271 saved_start); 2272 } 2273 } 2274 continue; 2275 } 2276 vm_map_clip_end(map, entry, end); 2277 2278 next = entry->next; 2279 2280 /* 2281 * Unwire before removing addresses from the pmap; otherwise, 2282 * unwiring will put the entries back in the pmap. 2283 */ 2284 if (entry->wired_count != 0) { 2285 vm_map_entry_unwire(map, entry); 2286 } 2287 2288 if (!map->system_map) 2289 VM_LOCK_GIANT(); 2290 pmap_remove(map->pmap, entry->start, entry->end); 2291 if (!map->system_map) 2292 VM_UNLOCK_GIANT(); 2293 2294 /* 2295 * Delete the entry (which may delete the object) only after 2296 * removing all pmap entries pointing to its pages. 2297 * (Otherwise, its page frames may be reallocated, and any 2298 * modify bits will be set in the wrong object!) 2299 */ 2300 vm_map_entry_delete(map, entry); 2301 entry = next; 2302 } 2303 return (KERN_SUCCESS); 2304 } 2305 2306 /* 2307 * vm_map_remove: 2308 * 2309 * Remove the given address range from the target map. 2310 * This is the exported form of vm_map_delete. 2311 */ 2312 int 2313 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 2314 { 2315 int result; 2316 2317 vm_map_lock(map); 2318 VM_MAP_RANGE_CHECK(map, start, end); 2319 result = vm_map_delete(map, start, end); 2320 vm_map_unlock(map); 2321 return (result); 2322 } 2323 2324 /* 2325 * vm_map_check_protection: 2326 * 2327 * Assert that the target map allows the specified privilege on the 2328 * entire address region given. The entire region must be allocated. 2329 * 2330 * WARNING! This code does not and should not check whether the 2331 * contents of the region is accessible. For example a smaller file 2332 * might be mapped into a larger address space. 2333 * 2334 * NOTE! This code is also called by munmap(). 2335 * 2336 * The map must be locked. A read lock is sufficient. 2337 */ 2338 boolean_t 2339 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 2340 vm_prot_t protection) 2341 { 2342 vm_map_entry_t entry; 2343 vm_map_entry_t tmp_entry; 2344 2345 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 2346 return (FALSE); 2347 entry = tmp_entry; 2348 2349 while (start < end) { 2350 if (entry == &map->header) 2351 return (FALSE); 2352 /* 2353 * No holes allowed! 2354 */ 2355 if (start < entry->start) 2356 return (FALSE); 2357 /* 2358 * Check protection associated with entry. 2359 */ 2360 if ((entry->protection & protection) != protection) 2361 return (FALSE); 2362 /* go to next entry */ 2363 start = entry->end; 2364 entry = entry->next; 2365 } 2366 return (TRUE); 2367 } 2368 2369 /* 2370 * vm_map_copy_entry: 2371 * 2372 * Copies the contents of the source entry to the destination 2373 * entry. The entries *must* be aligned properly. 2374 */ 2375 static void 2376 vm_map_copy_entry( 2377 vm_map_t src_map, 2378 vm_map_t dst_map, 2379 vm_map_entry_t src_entry, 2380 vm_map_entry_t dst_entry) 2381 { 2382 vm_object_t src_object; 2383 2384 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 2385 return; 2386 2387 if (src_entry->wired_count == 0) { 2388 2389 /* 2390 * If the source entry is marked needs_copy, it is already 2391 * write-protected. 2392 */ 2393 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { 2394 pmap_protect(src_map->pmap, 2395 src_entry->start, 2396 src_entry->end, 2397 src_entry->protection & ~VM_PROT_WRITE); 2398 } 2399 2400 /* 2401 * Make a copy of the object. 2402 */ 2403 if ((src_object = src_entry->object.vm_object) != NULL) { 2404 VM_OBJECT_LOCK(src_object); 2405 if ((src_object->handle == NULL) && 2406 (src_object->type == OBJT_DEFAULT || 2407 src_object->type == OBJT_SWAP)) { 2408 vm_object_collapse(src_object); 2409 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 2410 vm_object_split(src_entry); 2411 src_object = src_entry->object.vm_object; 2412 } 2413 } 2414 vm_object_reference_locked(src_object); 2415 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 2416 VM_OBJECT_UNLOCK(src_object); 2417 dst_entry->object.vm_object = src_object; 2418 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2419 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); 2420 dst_entry->offset = src_entry->offset; 2421 } else { 2422 dst_entry->object.vm_object = NULL; 2423 dst_entry->offset = 0; 2424 } 2425 2426 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, 2427 dst_entry->end - dst_entry->start, src_entry->start); 2428 } else { 2429 /* 2430 * Of course, wired down pages can't be set copy-on-write. 2431 * Cause wired pages to be copied into the new map by 2432 * simulating faults (the new pages are pageable) 2433 */ 2434 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); 2435 } 2436 } 2437 2438 /* 2439 * vmspace_map_entry_forked: 2440 * Update the newly-forked vmspace each time a map entry is inherited 2441 * or copied. The values for vm_dsize and vm_tsize are approximate 2442 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 2443 */ 2444 static void 2445 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 2446 vm_map_entry_t entry) 2447 { 2448 vm_size_t entrysize; 2449 vm_offset_t newend; 2450 2451 entrysize = entry->end - entry->start; 2452 vm2->vm_map.size += entrysize; 2453 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 2454 vm2->vm_ssize += btoc(entrysize); 2455 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 2456 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 2457 newend = MIN(entry->end, 2458 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 2459 vm2->vm_dsize += btoc(newend - entry->start); 2460 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 2461 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 2462 newend = MIN(entry->end, 2463 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 2464 vm2->vm_tsize += btoc(newend - entry->start); 2465 } 2466 } 2467 2468 /* 2469 * vmspace_fork: 2470 * Create a new process vmspace structure and vm_map 2471 * based on those of an existing process. The new map 2472 * is based on the old map, according to the inheritance 2473 * values on the regions in that map. 2474 * 2475 * XXX It might be worth coalescing the entries added to the new vmspace. 2476 * 2477 * The source map must not be locked. 2478 */ 2479 struct vmspace * 2480 vmspace_fork(struct vmspace *vm1) 2481 { 2482 struct vmspace *vm2; 2483 vm_map_t old_map = &vm1->vm_map; 2484 vm_map_t new_map; 2485 vm_map_entry_t old_entry; 2486 vm_map_entry_t new_entry; 2487 vm_object_t object; 2488 2489 vm_map_lock(old_map); 2490 2491 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); 2492 vm2->vm_taddr = vm1->vm_taddr; 2493 vm2->vm_daddr = vm1->vm_daddr; 2494 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 2495 new_map = &vm2->vm_map; /* XXX */ 2496 new_map->timestamp = 1; 2497 2498 /* Do not inherit the MAP_WIREFUTURE property. */ 2499 if ((new_map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE) 2500 new_map->flags &= ~MAP_WIREFUTURE; 2501 2502 old_entry = old_map->header.next; 2503 2504 while (old_entry != &old_map->header) { 2505 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 2506 panic("vm_map_fork: encountered a submap"); 2507 2508 switch (old_entry->inheritance) { 2509 case VM_INHERIT_NONE: 2510 break; 2511 2512 case VM_INHERIT_SHARE: 2513 /* 2514 * Clone the entry, creating the shared object if necessary. 2515 */ 2516 object = old_entry->object.vm_object; 2517 if (object == NULL) { 2518 object = vm_object_allocate(OBJT_DEFAULT, 2519 atop(old_entry->end - old_entry->start)); 2520 old_entry->object.vm_object = object; 2521 old_entry->offset = (vm_offset_t) 0; 2522 } 2523 2524 /* 2525 * Add the reference before calling vm_object_shadow 2526 * to insure that a shadow object is created. 2527 */ 2528 vm_object_reference(object); 2529 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 2530 vm_object_shadow(&old_entry->object.vm_object, 2531 &old_entry->offset, 2532 atop(old_entry->end - old_entry->start)); 2533 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 2534 /* Transfer the second reference too. */ 2535 vm_object_reference( 2536 old_entry->object.vm_object); 2537 vm_object_deallocate(object); 2538 object = old_entry->object.vm_object; 2539 } 2540 VM_OBJECT_LOCK(object); 2541 vm_object_clear_flag(object, OBJ_ONEMAPPING); 2542 VM_OBJECT_UNLOCK(object); 2543 2544 /* 2545 * Clone the entry, referencing the shared object. 2546 */ 2547 new_entry = vm_map_entry_create(new_map); 2548 *new_entry = *old_entry; 2549 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2550 new_entry->wired_count = 0; 2551 2552 /* 2553 * Insert the entry into the new map -- we know we're 2554 * inserting at the end of the new map. 2555 */ 2556 vm_map_entry_link(new_map, new_map->header.prev, 2557 new_entry); 2558 vmspace_map_entry_forked(vm1, vm2, new_entry); 2559 2560 /* 2561 * Update the physical map 2562 */ 2563 pmap_copy(new_map->pmap, old_map->pmap, 2564 new_entry->start, 2565 (old_entry->end - old_entry->start), 2566 old_entry->start); 2567 break; 2568 2569 case VM_INHERIT_COPY: 2570 /* 2571 * Clone the entry and link into the map. 2572 */ 2573 new_entry = vm_map_entry_create(new_map); 2574 *new_entry = *old_entry; 2575 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2576 new_entry->wired_count = 0; 2577 new_entry->object.vm_object = NULL; 2578 vm_map_entry_link(new_map, new_map->header.prev, 2579 new_entry); 2580 vmspace_map_entry_forked(vm1, vm2, new_entry); 2581 vm_map_copy_entry(old_map, new_map, old_entry, 2582 new_entry); 2583 break; 2584 } 2585 old_entry = old_entry->next; 2586 } 2587 2588 vm_map_unlock(old_map); 2589 2590 return (vm2); 2591 } 2592 2593 int 2594 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 2595 vm_prot_t prot, vm_prot_t max, int cow) 2596 { 2597 vm_map_entry_t new_entry, prev_entry; 2598 vm_offset_t bot, top; 2599 vm_size_t init_ssize; 2600 int orient, rv; 2601 rlim_t vmemlim; 2602 2603 /* 2604 * The stack orientation is piggybacked with the cow argument. 2605 * Extract it into orient and mask the cow argument so that we 2606 * don't pass it around further. 2607 * NOTE: We explicitly allow bi-directional stacks. 2608 */ 2609 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); 2610 cow &= ~orient; 2611 KASSERT(orient != 0, ("No stack grow direction")); 2612 2613 if (addrbos < vm_map_min(map) || addrbos > map->max_offset) 2614 return (KERN_NO_SPACE); 2615 2616 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz; 2617 2618 PROC_LOCK(curthread->td_proc); 2619 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM); 2620 PROC_UNLOCK(curthread->td_proc); 2621 2622 vm_map_lock(map); 2623 2624 /* If addr is already mapped, no go */ 2625 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { 2626 vm_map_unlock(map); 2627 return (KERN_NO_SPACE); 2628 } 2629 2630 /* If we would blow our VMEM resource limit, no go */ 2631 if (map->size + init_ssize > vmemlim) { 2632 vm_map_unlock(map); 2633 return (KERN_NO_SPACE); 2634 } 2635 2636 /* 2637 * If we can't accomodate max_ssize in the current mapping, no go. 2638 * However, we need to be aware that subsequent user mappings might 2639 * map into the space we have reserved for stack, and currently this 2640 * space is not protected. 2641 * 2642 * Hopefully we will at least detect this condition when we try to 2643 * grow the stack. 2644 */ 2645 if ((prev_entry->next != &map->header) && 2646 (prev_entry->next->start < addrbos + max_ssize)) { 2647 vm_map_unlock(map); 2648 return (KERN_NO_SPACE); 2649 } 2650 2651 /* 2652 * We initially map a stack of only init_ssize. We will grow as 2653 * needed later. Depending on the orientation of the stack (i.e. 2654 * the grow direction) we either map at the top of the range, the 2655 * bottom of the range or in the middle. 2656 * 2657 * Note: we would normally expect prot and max to be VM_PROT_ALL, 2658 * and cow to be 0. Possibly we should eliminate these as input 2659 * parameters, and just pass these values here in the insert call. 2660 */ 2661 if (orient == MAP_STACK_GROWS_DOWN) 2662 bot = addrbos + max_ssize - init_ssize; 2663 else if (orient == MAP_STACK_GROWS_UP) 2664 bot = addrbos; 2665 else 2666 bot = round_page(addrbos + max_ssize/2 - init_ssize/2); 2667 top = bot + init_ssize; 2668 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 2669 2670 /* Now set the avail_ssize amount. */ 2671 if (rv == KERN_SUCCESS) { 2672 if (prev_entry != &map->header) 2673 vm_map_clip_end(map, prev_entry, bot); 2674 new_entry = prev_entry->next; 2675 if (new_entry->end != top || new_entry->start != bot) 2676 panic("Bad entry start/end for new stack entry"); 2677 2678 new_entry->avail_ssize = max_ssize - init_ssize; 2679 if (orient & MAP_STACK_GROWS_DOWN) 2680 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2681 if (orient & MAP_STACK_GROWS_UP) 2682 new_entry->eflags |= MAP_ENTRY_GROWS_UP; 2683 } 2684 2685 vm_map_unlock(map); 2686 return (rv); 2687 } 2688 2689 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the 2690 * desired address is already mapped, or if we successfully grow 2691 * the stack. Also returns KERN_SUCCESS if addr is outside the 2692 * stack range (this is strange, but preserves compatibility with 2693 * the grow function in vm_machdep.c). 2694 */ 2695 int 2696 vm_map_growstack(struct proc *p, vm_offset_t addr) 2697 { 2698 vm_map_entry_t next_entry, prev_entry; 2699 vm_map_entry_t new_entry, stack_entry; 2700 struct vmspace *vm = p->p_vmspace; 2701 vm_map_t map = &vm->vm_map; 2702 vm_offset_t end; 2703 size_t grow_amount, max_grow; 2704 rlim_t stacklim, vmemlim; 2705 int is_procstack, rv; 2706 2707 Retry: 2708 PROC_LOCK(p); 2709 stacklim = lim_cur(p, RLIMIT_STACK); 2710 vmemlim = lim_cur(p, RLIMIT_VMEM); 2711 PROC_UNLOCK(p); 2712 2713 vm_map_lock_read(map); 2714 2715 /* If addr is already in the entry range, no need to grow.*/ 2716 if (vm_map_lookup_entry(map, addr, &prev_entry)) { 2717 vm_map_unlock_read(map); 2718 return (KERN_SUCCESS); 2719 } 2720 2721 next_entry = prev_entry->next; 2722 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { 2723 /* 2724 * This entry does not grow upwards. Since the address lies 2725 * beyond this entry, the next entry (if one exists) has to 2726 * be a downward growable entry. The entry list header is 2727 * never a growable entry, so it suffices to check the flags. 2728 */ 2729 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { 2730 vm_map_unlock_read(map); 2731 return (KERN_SUCCESS); 2732 } 2733 stack_entry = next_entry; 2734 } else { 2735 /* 2736 * This entry grows upward. If the next entry does not at 2737 * least grow downwards, this is the entry we need to grow. 2738 * otherwise we have two possible choices and we have to 2739 * select one. 2740 */ 2741 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { 2742 /* 2743 * We have two choices; grow the entry closest to 2744 * the address to minimize the amount of growth. 2745 */ 2746 if (addr - prev_entry->end <= next_entry->start - addr) 2747 stack_entry = prev_entry; 2748 else 2749 stack_entry = next_entry; 2750 } else 2751 stack_entry = prev_entry; 2752 } 2753 2754 if (stack_entry == next_entry) { 2755 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); 2756 KASSERT(addr < stack_entry->start, ("foo")); 2757 end = (prev_entry != &map->header) ? prev_entry->end : 2758 stack_entry->start - stack_entry->avail_ssize; 2759 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); 2760 max_grow = stack_entry->start - end; 2761 } else { 2762 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); 2763 KASSERT(addr >= stack_entry->end, ("foo")); 2764 end = (next_entry != &map->header) ? next_entry->start : 2765 stack_entry->end + stack_entry->avail_ssize; 2766 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); 2767 max_grow = end - stack_entry->end; 2768 } 2769 2770 if (grow_amount > stack_entry->avail_ssize) { 2771 vm_map_unlock_read(map); 2772 return (KERN_NO_SPACE); 2773 } 2774 2775 /* 2776 * If there is no longer enough space between the entries nogo, and 2777 * adjust the available space. Note: this should only happen if the 2778 * user has mapped into the stack area after the stack was created, 2779 * and is probably an error. 2780 * 2781 * This also effectively destroys any guard page the user might have 2782 * intended by limiting the stack size. 2783 */ 2784 if (grow_amount > max_grow) { 2785 if (vm_map_lock_upgrade(map)) 2786 goto Retry; 2787 2788 stack_entry->avail_ssize = max_grow; 2789 2790 vm_map_unlock(map); 2791 return (KERN_NO_SPACE); 2792 } 2793 2794 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; 2795 2796 /* 2797 * If this is the main process stack, see if we're over the stack 2798 * limit. 2799 */ 2800 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2801 vm_map_unlock_read(map); 2802 return (KERN_NO_SPACE); 2803 } 2804 2805 /* Round up the grow amount modulo SGROWSIZ */ 2806 grow_amount = roundup (grow_amount, sgrowsiz); 2807 if (grow_amount > stack_entry->avail_ssize) 2808 grow_amount = stack_entry->avail_ssize; 2809 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 2810 grow_amount = stacklim - ctob(vm->vm_ssize); 2811 } 2812 2813 /* If we would blow our VMEM resource limit, no go */ 2814 if (map->size + grow_amount > vmemlim) { 2815 vm_map_unlock_read(map); 2816 return (KERN_NO_SPACE); 2817 } 2818 2819 if (vm_map_lock_upgrade(map)) 2820 goto Retry; 2821 2822 if (stack_entry == next_entry) { 2823 /* 2824 * Growing downward. 2825 */ 2826 /* Get the preliminary new entry start value */ 2827 addr = stack_entry->start - grow_amount; 2828 2829 /* 2830 * If this puts us into the previous entry, cut back our 2831 * growth to the available space. Also, see the note above. 2832 */ 2833 if (addr < end) { 2834 stack_entry->avail_ssize = max_grow; 2835 addr = end; 2836 } 2837 2838 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, 2839 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0); 2840 2841 /* Adjust the available stack space by the amount we grew. */ 2842 if (rv == KERN_SUCCESS) { 2843 if (prev_entry != &map->header) 2844 vm_map_clip_end(map, prev_entry, addr); 2845 new_entry = prev_entry->next; 2846 KASSERT(new_entry == stack_entry->prev, ("foo")); 2847 KASSERT(new_entry->end == stack_entry->start, ("foo")); 2848 KASSERT(new_entry->start == addr, ("foo")); 2849 grow_amount = new_entry->end - new_entry->start; 2850 new_entry->avail_ssize = stack_entry->avail_ssize - 2851 grow_amount; 2852 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; 2853 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN; 2854 } 2855 } else { 2856 /* 2857 * Growing upward. 2858 */ 2859 addr = stack_entry->end + grow_amount; 2860 2861 /* 2862 * If this puts us into the next entry, cut back our growth 2863 * to the available space. Also, see the note above. 2864 */ 2865 if (addr > end) { 2866 stack_entry->avail_ssize = end - stack_entry->end; 2867 addr = end; 2868 } 2869 2870 grow_amount = addr - stack_entry->end; 2871 2872 /* Grow the underlying object if applicable. */ 2873 if (stack_entry->object.vm_object == NULL || 2874 vm_object_coalesce(stack_entry->object.vm_object, 2875 stack_entry->offset, 2876 (vm_size_t)(stack_entry->end - stack_entry->start), 2877 (vm_size_t)grow_amount)) { 2878 map->size += (addr - stack_entry->end); 2879 /* Update the current entry. */ 2880 stack_entry->end = addr; 2881 stack_entry->avail_ssize -= grow_amount; 2882 vm_map_entry_resize_free(map, stack_entry); 2883 rv = KERN_SUCCESS; 2884 2885 if (next_entry != &map->header) 2886 vm_map_clip_start(map, next_entry, addr); 2887 } else 2888 rv = KERN_FAILURE; 2889 } 2890 2891 if (rv == KERN_SUCCESS && is_procstack) 2892 vm->vm_ssize += btoc(grow_amount); 2893 2894 vm_map_unlock(map); 2895 2896 /* 2897 * Heed the MAP_WIREFUTURE flag if it was set for this process. 2898 */ 2899 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { 2900 vm_map_wire(map, 2901 (stack_entry == next_entry) ? addr : addr - grow_amount, 2902 (stack_entry == next_entry) ? stack_entry->start : addr, 2903 (p->p_flag & P_SYSTEM) 2904 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES 2905 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); 2906 } 2907 2908 return (rv); 2909 } 2910 2911 /* 2912 * Unshare the specified VM space for exec. If other processes are 2913 * mapped to it, then create a new one. The new vmspace is null. 2914 */ 2915 void 2916 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 2917 { 2918 struct vmspace *oldvmspace = p->p_vmspace; 2919 struct vmspace *newvmspace; 2920 2921 GIANT_REQUIRED; 2922 newvmspace = vmspace_alloc(minuser, maxuser); 2923 newvmspace->vm_swrss = oldvmspace->vm_swrss; 2924 /* 2925 * This code is written like this for prototype purposes. The 2926 * goal is to avoid running down the vmspace here, but let the 2927 * other process's that are still using the vmspace to finally 2928 * run it down. Even though there is little or no chance of blocking 2929 * here, it is a good idea to keep this form for future mods. 2930 */ 2931 p->p_vmspace = newvmspace; 2932 if (p == curthread->td_proc) /* XXXKSE ? */ 2933 pmap_activate(curthread); 2934 vmspace_free(oldvmspace); 2935 } 2936 2937 /* 2938 * Unshare the specified VM space for forcing COW. This 2939 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 2940 */ 2941 void 2942 vmspace_unshare(struct proc *p) 2943 { 2944 struct vmspace *oldvmspace = p->p_vmspace; 2945 struct vmspace *newvmspace; 2946 2947 if (oldvmspace->vm_refcnt == 1) 2948 return; 2949 newvmspace = vmspace_fork(oldvmspace); 2950 p->p_vmspace = newvmspace; 2951 if (p == curthread->td_proc) /* XXXKSE ? */ 2952 pmap_activate(curthread); 2953 vmspace_free(oldvmspace); 2954 } 2955 2956 /* 2957 * vm_map_lookup: 2958 * 2959 * Finds the VM object, offset, and 2960 * protection for a given virtual address in the 2961 * specified map, assuming a page fault of the 2962 * type specified. 2963 * 2964 * Leaves the map in question locked for read; return 2965 * values are guaranteed until a vm_map_lookup_done 2966 * call is performed. Note that the map argument 2967 * is in/out; the returned map must be used in 2968 * the call to vm_map_lookup_done. 2969 * 2970 * A handle (out_entry) is returned for use in 2971 * vm_map_lookup_done, to make that fast. 2972 * 2973 * If a lookup is requested with "write protection" 2974 * specified, the map may be changed to perform virtual 2975 * copying operations, although the data referenced will 2976 * remain the same. 2977 */ 2978 int 2979 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 2980 vm_offset_t vaddr, 2981 vm_prot_t fault_typea, 2982 vm_map_entry_t *out_entry, /* OUT */ 2983 vm_object_t *object, /* OUT */ 2984 vm_pindex_t *pindex, /* OUT */ 2985 vm_prot_t *out_prot, /* OUT */ 2986 boolean_t *wired) /* OUT */ 2987 { 2988 vm_map_entry_t entry; 2989 vm_map_t map = *var_map; 2990 vm_prot_t prot; 2991 vm_prot_t fault_type = fault_typea; 2992 2993 RetryLookup:; 2994 /* 2995 * Lookup the faulting address. 2996 */ 2997 2998 vm_map_lock_read(map); 2999 #define RETURN(why) \ 3000 { \ 3001 vm_map_unlock_read(map); \ 3002 return (why); \ 3003 } 3004 3005 /* 3006 * If the map has an interesting hint, try it before calling full 3007 * blown lookup routine. 3008 */ 3009 entry = map->root; 3010 *out_entry = entry; 3011 if (entry == NULL || 3012 (vaddr < entry->start) || (vaddr >= entry->end)) { 3013 /* 3014 * Entry was either not a valid hint, or the vaddr was not 3015 * contained in the entry, so do a full lookup. 3016 */ 3017 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3018 RETURN(KERN_INVALID_ADDRESS); 3019 3020 entry = *out_entry; 3021 } 3022 3023 /* 3024 * Handle submaps. 3025 */ 3026 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3027 vm_map_t old_map = map; 3028 3029 *var_map = map = entry->object.sub_map; 3030 vm_map_unlock_read(old_map); 3031 goto RetryLookup; 3032 } 3033 3034 /* 3035 * Check whether this task is allowed to have this page. 3036 * Note the special case for MAP_ENTRY_COW 3037 * pages with an override. This is to implement a forced 3038 * COW for debuggers. 3039 */ 3040 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3041 prot = entry->max_protection; 3042 else 3043 prot = entry->protection; 3044 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); 3045 if ((fault_type & prot) != fault_type) { 3046 RETURN(KERN_PROTECTION_FAILURE); 3047 } 3048 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3049 (entry->eflags & MAP_ENTRY_COW) && 3050 (fault_type & VM_PROT_WRITE) && 3051 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { 3052 RETURN(KERN_PROTECTION_FAILURE); 3053 } 3054 3055 /* 3056 * If this page is not pageable, we have to get it for all possible 3057 * accesses. 3058 */ 3059 *wired = (entry->wired_count != 0); 3060 if (*wired) 3061 prot = fault_type = entry->protection; 3062 3063 /* 3064 * If the entry was copy-on-write, we either ... 3065 */ 3066 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3067 /* 3068 * If we want to write the page, we may as well handle that 3069 * now since we've got the map locked. 3070 * 3071 * If we don't need to write the page, we just demote the 3072 * permissions allowed. 3073 */ 3074 if (fault_type & VM_PROT_WRITE) { 3075 /* 3076 * Make a new object, and place it in the object 3077 * chain. Note that no new references have appeared 3078 * -- one just moved from the map to the new 3079 * object. 3080 */ 3081 if (vm_map_lock_upgrade(map)) 3082 goto RetryLookup; 3083 3084 vm_object_shadow( 3085 &entry->object.vm_object, 3086 &entry->offset, 3087 atop(entry->end - entry->start)); 3088 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3089 3090 vm_map_lock_downgrade(map); 3091 } else { 3092 /* 3093 * We're attempting to read a copy-on-write page -- 3094 * don't allow writes. 3095 */ 3096 prot &= ~VM_PROT_WRITE; 3097 } 3098 } 3099 3100 /* 3101 * Create an object if necessary. 3102 */ 3103 if (entry->object.vm_object == NULL && 3104 !map->system_map) { 3105 if (vm_map_lock_upgrade(map)) 3106 goto RetryLookup; 3107 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 3108 atop(entry->end - entry->start)); 3109 entry->offset = 0; 3110 vm_map_lock_downgrade(map); 3111 } 3112 3113 /* 3114 * Return the object/offset from this entry. If the entry was 3115 * copy-on-write or empty, it has been fixed up. 3116 */ 3117 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3118 *object = entry->object.vm_object; 3119 3120 *out_prot = prot; 3121 return (KERN_SUCCESS); 3122 3123 #undef RETURN 3124 } 3125 3126 /* 3127 * vm_map_lookup_locked: 3128 * 3129 * Lookup the faulting address. A version of vm_map_lookup that returns 3130 * KERN_FAILURE instead of blocking on map lock or memory allocation. 3131 */ 3132 int 3133 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 3134 vm_offset_t vaddr, 3135 vm_prot_t fault_typea, 3136 vm_map_entry_t *out_entry, /* OUT */ 3137 vm_object_t *object, /* OUT */ 3138 vm_pindex_t *pindex, /* OUT */ 3139 vm_prot_t *out_prot, /* OUT */ 3140 boolean_t *wired) /* OUT */ 3141 { 3142 vm_map_entry_t entry; 3143 vm_map_t map = *var_map; 3144 vm_prot_t prot; 3145 vm_prot_t fault_type = fault_typea; 3146 3147 /* 3148 * If the map has an interesting hint, try it before calling full 3149 * blown lookup routine. 3150 */ 3151 entry = map->root; 3152 *out_entry = entry; 3153 if (entry == NULL || 3154 (vaddr < entry->start) || (vaddr >= entry->end)) { 3155 /* 3156 * Entry was either not a valid hint, or the vaddr was not 3157 * contained in the entry, so do a full lookup. 3158 */ 3159 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 3160 return (KERN_INVALID_ADDRESS); 3161 3162 entry = *out_entry; 3163 } 3164 3165 /* 3166 * Fail if the entry refers to a submap. 3167 */ 3168 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3169 return (KERN_FAILURE); 3170 3171 /* 3172 * Check whether this task is allowed to have this page. 3173 * Note the special case for MAP_ENTRY_COW 3174 * pages with an override. This is to implement a forced 3175 * COW for debuggers. 3176 */ 3177 if (fault_type & VM_PROT_OVERRIDE_WRITE) 3178 prot = entry->max_protection; 3179 else 3180 prot = entry->protection; 3181 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 3182 if ((fault_type & prot) != fault_type) 3183 return (KERN_PROTECTION_FAILURE); 3184 if ((entry->eflags & MAP_ENTRY_USER_WIRED) && 3185 (entry->eflags & MAP_ENTRY_COW) && 3186 (fault_type & VM_PROT_WRITE) && 3187 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) 3188 return (KERN_PROTECTION_FAILURE); 3189 3190 /* 3191 * If this page is not pageable, we have to get it for all possible 3192 * accesses. 3193 */ 3194 *wired = (entry->wired_count != 0); 3195 if (*wired) 3196 prot = fault_type = entry->protection; 3197 3198 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3199 /* 3200 * Fail if the entry was copy-on-write for a write fault. 3201 */ 3202 if (fault_type & VM_PROT_WRITE) 3203 return (KERN_FAILURE); 3204 /* 3205 * We're attempting to read a copy-on-write page -- 3206 * don't allow writes. 3207 */ 3208 prot &= ~VM_PROT_WRITE; 3209 } 3210 3211 /* 3212 * Fail if an object should be created. 3213 */ 3214 if (entry->object.vm_object == NULL && !map->system_map) 3215 return (KERN_FAILURE); 3216 3217 /* 3218 * Return the object/offset from this entry. If the entry was 3219 * copy-on-write or empty, it has been fixed up. 3220 */ 3221 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 3222 *object = entry->object.vm_object; 3223 3224 *out_prot = prot; 3225 return (KERN_SUCCESS); 3226 } 3227 3228 /* 3229 * vm_map_lookup_done: 3230 * 3231 * Releases locks acquired by a vm_map_lookup 3232 * (according to the handle returned by that lookup). 3233 */ 3234 void 3235 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 3236 { 3237 /* 3238 * Unlock the main-level map 3239 */ 3240 vm_map_unlock_read(map); 3241 } 3242 3243 #include "opt_ddb.h" 3244 #ifdef DDB 3245 #include <sys/kernel.h> 3246 3247 #include <ddb/ddb.h> 3248 3249 /* 3250 * vm_map_print: [ debug ] 3251 */ 3252 DB_SHOW_COMMAND(map, vm_map_print) 3253 { 3254 static int nlines; 3255 /* XXX convert args. */ 3256 vm_map_t map = (vm_map_t)addr; 3257 boolean_t full = have_addr; 3258 3259 vm_map_entry_t entry; 3260 3261 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 3262 (void *)map, 3263 (void *)map->pmap, map->nentries, map->timestamp); 3264 nlines++; 3265 3266 if (!full && db_indent) 3267 return; 3268 3269 db_indent += 2; 3270 for (entry = map->header.next; entry != &map->header; 3271 entry = entry->next) { 3272 db_iprintf("map entry %p: start=%p, end=%p\n", 3273 (void *)entry, (void *)entry->start, (void *)entry->end); 3274 nlines++; 3275 { 3276 static char *inheritance_name[4] = 3277 {"share", "copy", "none", "donate_copy"}; 3278 3279 db_iprintf(" prot=%x/%x/%s", 3280 entry->protection, 3281 entry->max_protection, 3282 inheritance_name[(int)(unsigned char)entry->inheritance]); 3283 if (entry->wired_count != 0) 3284 db_printf(", wired"); 3285 } 3286 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 3287 db_printf(", share=%p, offset=0x%jx\n", 3288 (void *)entry->object.sub_map, 3289 (uintmax_t)entry->offset); 3290 nlines++; 3291 if ((entry->prev == &map->header) || 3292 (entry->prev->object.sub_map != 3293 entry->object.sub_map)) { 3294 db_indent += 2; 3295 vm_map_print((db_expr_t)(intptr_t) 3296 entry->object.sub_map, 3297 full, 0, (char *)0); 3298 db_indent -= 2; 3299 } 3300 } else { 3301 db_printf(", object=%p, offset=0x%jx", 3302 (void *)entry->object.vm_object, 3303 (uintmax_t)entry->offset); 3304 if (entry->eflags & MAP_ENTRY_COW) 3305 db_printf(", copy (%s)", 3306 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 3307 db_printf("\n"); 3308 nlines++; 3309 3310 if ((entry->prev == &map->header) || 3311 (entry->prev->object.vm_object != 3312 entry->object.vm_object)) { 3313 db_indent += 2; 3314 vm_object_print((db_expr_t)(intptr_t) 3315 entry->object.vm_object, 3316 full, 0, (char *)0); 3317 nlines += 4; 3318 db_indent -= 2; 3319 } 3320 } 3321 } 3322 db_indent -= 2; 3323 if (db_indent == 0) 3324 nlines = 0; 3325 } 3326 3327 3328 DB_SHOW_COMMAND(procvm, procvm) 3329 { 3330 struct proc *p; 3331 3332 if (have_addr) { 3333 p = (struct proc *) addr; 3334 } else { 3335 p = curproc; 3336 } 3337 3338 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 3339 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 3340 (void *)vmspace_pmap(p->p_vmspace)); 3341 3342 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); 3343 } 3344 3345 #endif /* DDB */ 3346