1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41 * 42 * Permission to use, copy, modify and distribute this software and 43 * its documentation is hereby granted, provided that both the copyright 44 * notice and this permission notice appear in all copies of the 45 * software, derivative works or modified versions, and any portions 46 * thereof, and that both notices appear in supporting documentation. 47 * 48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51 * 52 * Carnegie Mellon requests users of this software to return to 53 * 54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55 * School of Computer Science 56 * Carnegie Mellon University 57 * Pittsburgh PA 15213-3890 58 * 59 * any improvements or extensions that they make and grant Carnegie the 60 * rights to redistribute these changes. 61 */ 62 63 /* 64 * Virtual memory mapping module. 65 */ 66 67 #include <sys/cdefs.h> 68 __FBSDID("$FreeBSD$"); 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/elf.h> 73 #include <sys/kernel.h> 74 #include <sys/ktr.h> 75 #include <sys/lock.h> 76 #include <sys/mutex.h> 77 #include <sys/proc.h> 78 #include <sys/vmmeter.h> 79 #include <sys/mman.h> 80 #include <sys/vnode.h> 81 #include <sys/racct.h> 82 #include <sys/resourcevar.h> 83 #include <sys/rwlock.h> 84 #include <sys/file.h> 85 #include <sys/sysctl.h> 86 #include <sys/sysent.h> 87 #include <sys/shm.h> 88 89 #include <vm/vm.h> 90 #include <vm/vm_param.h> 91 #include <vm/pmap.h> 92 #include <vm/vm_map.h> 93 #include <vm/vm_page.h> 94 #include <vm/vm_pageout.h> 95 #include <vm/vm_object.h> 96 #include <vm/vm_pager.h> 97 #include <vm/vm_kern.h> 98 #include <vm/vm_extern.h> 99 #include <vm/vnode_pager.h> 100 #include <vm/swap_pager.h> 101 #include <vm/uma.h> 102 103 /* 104 * Virtual memory maps provide for the mapping, protection, 105 * and sharing of virtual memory objects. In addition, 106 * this module provides for an efficient virtual copy of 107 * memory from one map to another. 108 * 109 * Synchronization is required prior to most operations. 110 * 111 * Maps consist of an ordered doubly-linked list of simple 112 * entries; a self-adjusting binary search tree of these 113 * entries is used to speed up lookups. 114 * 115 * Since portions of maps are specified by start/end addresses, 116 * which may not align with existing map entries, all 117 * routines merely "clip" entries to these start/end values. 118 * [That is, an entry is split into two, bordering at a 119 * start or end value.] Note that these clippings may not 120 * always be necessary (as the two resulting entries are then 121 * not changed); however, the clipping is done for convenience. 122 * 123 * As mentioned above, virtual copy operations are performed 124 * by copying VM object references from one map to 125 * another, and then marking both regions as copy-on-write. 126 */ 127 128 static struct mtx map_sleep_mtx; 129 static uma_zone_t mapentzone; 130 static uma_zone_t kmapentzone; 131 static uma_zone_t mapzone; 132 static uma_zone_t vmspace_zone; 133 static int vmspace_zinit(void *mem, int size, int flags); 134 static int vm_map_zinit(void *mem, int ize, int flags); 135 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 136 vm_offset_t max); 137 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 138 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 139 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry); 140 static int vm_map_growstack(vm_map_t map, vm_offset_t addr, 141 vm_map_entry_t gap_entry); 142 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 143 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags); 144 #ifdef INVARIANTS 145 static void vm_map_zdtor(void *mem, int size, void *arg); 146 static void vmspace_zdtor(void *mem, int size, void *arg); 147 #endif 148 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, 149 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, 150 int cow); 151 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 152 vm_offset_t failed_addr); 153 154 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 155 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 156 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 157 158 /* 159 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 160 * stable. 161 */ 162 #define PROC_VMSPACE_LOCK(p) do { } while (0) 163 #define PROC_VMSPACE_UNLOCK(p) do { } while (0) 164 165 /* 166 * VM_MAP_RANGE_CHECK: [ internal use only ] 167 * 168 * Asserts that the starting and ending region 169 * addresses fall within the valid range of the map. 170 */ 171 #define VM_MAP_RANGE_CHECK(map, start, end) \ 172 { \ 173 if (start < vm_map_min(map)) \ 174 start = vm_map_min(map); \ 175 if (end > vm_map_max(map)) \ 176 end = vm_map_max(map); \ 177 if (start > end) \ 178 start = end; \ 179 } 180 181 /* 182 * vm_map_startup: 183 * 184 * Initialize the vm_map module. Must be called before 185 * any other vm_map routines. 186 * 187 * Map and entry structures are allocated from the general 188 * purpose memory pool with some exceptions: 189 * 190 * - The kernel map and kmem submap are allocated statically. 191 * - Kernel map entries are allocated out of a static pool. 192 * 193 * These restrictions are necessary since malloc() uses the 194 * maps and requires map entries. 195 */ 196 197 void 198 vm_map_startup(void) 199 { 200 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 201 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 202 #ifdef INVARIANTS 203 vm_map_zdtor, 204 #else 205 NULL, 206 #endif 207 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 208 uma_prealloc(mapzone, MAX_KMAP); 209 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 210 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 211 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 212 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 214 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 215 #ifdef INVARIANTS 216 vmspace_zdtor, 217 #else 218 NULL, 219 #endif 220 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 221 } 222 223 static int 224 vmspace_zinit(void *mem, int size, int flags) 225 { 226 struct vmspace *vm; 227 228 vm = (struct vmspace *)mem; 229 230 vm->vm_map.pmap = NULL; 231 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 232 PMAP_LOCK_INIT(vmspace_pmap(vm)); 233 return (0); 234 } 235 236 static int 237 vm_map_zinit(void *mem, int size, int flags) 238 { 239 vm_map_t map; 240 241 map = (vm_map_t)mem; 242 memset(map, 0, sizeof(*map)); 243 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 244 sx_init(&map->lock, "vm map (user)"); 245 return (0); 246 } 247 248 #ifdef INVARIANTS 249 static void 250 vmspace_zdtor(void *mem, int size, void *arg) 251 { 252 struct vmspace *vm; 253 254 vm = (struct vmspace *)mem; 255 256 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 257 } 258 static void 259 vm_map_zdtor(void *mem, int size, void *arg) 260 { 261 vm_map_t map; 262 263 map = (vm_map_t)mem; 264 KASSERT(map->nentries == 0, 265 ("map %p nentries == %d on free.", 266 map, map->nentries)); 267 KASSERT(map->size == 0, 268 ("map %p size == %lu on free.", 269 map, (unsigned long)map->size)); 270 } 271 #endif /* INVARIANTS */ 272 273 /* 274 * Allocate a vmspace structure, including a vm_map and pmap, 275 * and initialize those structures. The refcnt is set to 1. 276 * 277 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 278 */ 279 struct vmspace * 280 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 281 { 282 struct vmspace *vm; 283 284 vm = uma_zalloc(vmspace_zone, M_WAITOK); 285 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 286 if (!pinit(vmspace_pmap(vm))) { 287 uma_zfree(vmspace_zone, vm); 288 return (NULL); 289 } 290 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 291 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 292 vm->vm_refcnt = 1; 293 vm->vm_shm = NULL; 294 vm->vm_swrss = 0; 295 vm->vm_tsize = 0; 296 vm->vm_dsize = 0; 297 vm->vm_ssize = 0; 298 vm->vm_taddr = 0; 299 vm->vm_daddr = 0; 300 vm->vm_maxsaddr = 0; 301 return (vm); 302 } 303 304 #ifdef RACCT 305 static void 306 vmspace_container_reset(struct proc *p) 307 { 308 309 PROC_LOCK(p); 310 racct_set(p, RACCT_DATA, 0); 311 racct_set(p, RACCT_STACK, 0); 312 racct_set(p, RACCT_RSS, 0); 313 racct_set(p, RACCT_MEMLOCK, 0); 314 racct_set(p, RACCT_VMEM, 0); 315 PROC_UNLOCK(p); 316 } 317 #endif 318 319 static inline void 320 vmspace_dofree(struct vmspace *vm) 321 { 322 323 CTR1(KTR_VM, "vmspace_free: %p", vm); 324 325 /* 326 * Make sure any SysV shm is freed, it might not have been in 327 * exit1(). 328 */ 329 shmexit(vm); 330 331 /* 332 * Lock the map, to wait out all other references to it. 333 * Delete all of the mappings and pages they hold, then call 334 * the pmap module to reclaim anything left. 335 */ 336 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map), 337 vm_map_max(&vm->vm_map)); 338 339 pmap_release(vmspace_pmap(vm)); 340 vm->vm_map.pmap = NULL; 341 uma_zfree(vmspace_zone, vm); 342 } 343 344 void 345 vmspace_free(struct vmspace *vm) 346 { 347 348 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 349 "vmspace_free() called"); 350 351 if (vm->vm_refcnt == 0) 352 panic("vmspace_free: attempt to free already freed vmspace"); 353 354 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 355 vmspace_dofree(vm); 356 } 357 358 void 359 vmspace_exitfree(struct proc *p) 360 { 361 struct vmspace *vm; 362 363 PROC_VMSPACE_LOCK(p); 364 vm = p->p_vmspace; 365 p->p_vmspace = NULL; 366 PROC_VMSPACE_UNLOCK(p); 367 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 368 vmspace_free(vm); 369 } 370 371 void 372 vmspace_exit(struct thread *td) 373 { 374 int refcnt; 375 struct vmspace *vm; 376 struct proc *p; 377 378 /* 379 * Release user portion of address space. 380 * This releases references to vnodes, 381 * which could cause I/O if the file has been unlinked. 382 * Need to do this early enough that we can still sleep. 383 * 384 * The last exiting process to reach this point releases as 385 * much of the environment as it can. vmspace_dofree() is the 386 * slower fallback in case another process had a temporary 387 * reference to the vmspace. 388 */ 389 390 p = td->td_proc; 391 vm = p->p_vmspace; 392 atomic_add_int(&vmspace0.vm_refcnt, 1); 393 refcnt = vm->vm_refcnt; 394 do { 395 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 396 /* Switch now since other proc might free vmspace */ 397 PROC_VMSPACE_LOCK(p); 398 p->p_vmspace = &vmspace0; 399 PROC_VMSPACE_UNLOCK(p); 400 pmap_activate(td); 401 } 402 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1)); 403 if (refcnt == 1) { 404 if (p->p_vmspace != vm) { 405 /* vmspace not yet freed, switch back */ 406 PROC_VMSPACE_LOCK(p); 407 p->p_vmspace = vm; 408 PROC_VMSPACE_UNLOCK(p); 409 pmap_activate(td); 410 } 411 pmap_remove_pages(vmspace_pmap(vm)); 412 /* Switch now since this proc will free vmspace */ 413 PROC_VMSPACE_LOCK(p); 414 p->p_vmspace = &vmspace0; 415 PROC_VMSPACE_UNLOCK(p); 416 pmap_activate(td); 417 vmspace_dofree(vm); 418 } 419 #ifdef RACCT 420 if (racct_enable) 421 vmspace_container_reset(p); 422 #endif 423 } 424 425 /* Acquire reference to vmspace owned by another process. */ 426 427 struct vmspace * 428 vmspace_acquire_ref(struct proc *p) 429 { 430 struct vmspace *vm; 431 int refcnt; 432 433 PROC_VMSPACE_LOCK(p); 434 vm = p->p_vmspace; 435 if (vm == NULL) { 436 PROC_VMSPACE_UNLOCK(p); 437 return (NULL); 438 } 439 refcnt = vm->vm_refcnt; 440 do { 441 if (refcnt <= 0) { /* Avoid 0->1 transition */ 442 PROC_VMSPACE_UNLOCK(p); 443 return (NULL); 444 } 445 } while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1)); 446 if (vm != p->p_vmspace) { 447 PROC_VMSPACE_UNLOCK(p); 448 vmspace_free(vm); 449 return (NULL); 450 } 451 PROC_VMSPACE_UNLOCK(p); 452 return (vm); 453 } 454 455 /* 456 * Switch between vmspaces in an AIO kernel process. 457 * 458 * The new vmspace is either the vmspace of a user process obtained 459 * from an active AIO request or the initial vmspace of the AIO kernel 460 * process (when it is idling). Because user processes will block to 461 * drain any active AIO requests before proceeding in exit() or 462 * execve(), the reference count for vmspaces from AIO requests can 463 * never be 0. Similarly, AIO kernel processes hold an extra 464 * reference on their initial vmspace for the life of the process. As 465 * a result, the 'newvm' vmspace always has a non-zero reference 466 * count. This permits an additional reference on 'newvm' to be 467 * acquired via a simple atomic increment rather than the loop in 468 * vmspace_acquire_ref() above. 469 */ 470 void 471 vmspace_switch_aio(struct vmspace *newvm) 472 { 473 struct vmspace *oldvm; 474 475 /* XXX: Need some way to assert that this is an aio daemon. */ 476 477 KASSERT(newvm->vm_refcnt > 0, 478 ("vmspace_switch_aio: newvm unreferenced")); 479 480 oldvm = curproc->p_vmspace; 481 if (oldvm == newvm) 482 return; 483 484 /* 485 * Point to the new address space and refer to it. 486 */ 487 curproc->p_vmspace = newvm; 488 atomic_add_int(&newvm->vm_refcnt, 1); 489 490 /* Activate the new mapping. */ 491 pmap_activate(curthread); 492 493 vmspace_free(oldvm); 494 } 495 496 void 497 _vm_map_lock(vm_map_t map, const char *file, int line) 498 { 499 500 if (map->system_map) 501 mtx_lock_flags_(&map->system_mtx, 0, file, line); 502 else 503 sx_xlock_(&map->lock, file, line); 504 map->timestamp++; 505 } 506 507 void 508 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add) 509 { 510 vm_object_t object; 511 struct vnode *vp; 512 bool vp_held; 513 514 if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0) 515 return; 516 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 517 ("Submap with execs")); 518 object = entry->object.vm_object; 519 KASSERT(object != NULL, ("No object for text, entry %p", entry)); 520 if ((object->flags & OBJ_ANON) != 0) 521 object = object->handle; 522 else 523 KASSERT(object->backing_object == NULL, 524 ("non-anon object %p shadows", object)); 525 KASSERT(object != NULL, ("No content object for text, entry %p obj %p", 526 entry, entry->object.vm_object)); 527 528 /* 529 * Mostly, we do not lock the backing object. It is 530 * referenced by the entry we are processing, so it cannot go 531 * away. 532 */ 533 vp = NULL; 534 vp_held = false; 535 if (object->type == OBJT_DEAD) { 536 /* 537 * For OBJT_DEAD objects, v_writecount was handled in 538 * vnode_pager_dealloc(). 539 */ 540 } else if (object->type == OBJT_VNODE) { 541 vp = object->handle; 542 } else if (object->type == OBJT_SWAP) { 543 KASSERT((object->flags & OBJ_TMPFS_NODE) != 0, 544 ("vm_map_entry_set_vnode_text: swap and !TMPFS " 545 "entry %p, object %p, add %d", entry, object, add)); 546 /* 547 * Tmpfs VREG node, which was reclaimed, has 548 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS. In 549 * this case there is no v_writecount to adjust. 550 */ 551 VM_OBJECT_RLOCK(object); 552 if ((object->flags & OBJ_TMPFS) != 0) { 553 vp = object->un_pager.swp.swp_tmpfs; 554 if (vp != NULL) { 555 vhold(vp); 556 vp_held = true; 557 } 558 } 559 VM_OBJECT_RUNLOCK(object); 560 } else { 561 KASSERT(0, 562 ("vm_map_entry_set_vnode_text: wrong object type, " 563 "entry %p, object %p, add %d", entry, object, add)); 564 } 565 if (vp != NULL) { 566 if (add) { 567 VOP_SET_TEXT_CHECKED(vp); 568 } else { 569 vn_lock(vp, LK_SHARED | LK_RETRY); 570 VOP_UNSET_TEXT_CHECKED(vp); 571 VOP_UNLOCK(vp, 0); 572 } 573 if (vp_held) 574 vdrop(vp); 575 } 576 } 577 578 /* 579 * Use a different name for this vm_map_entry field when it's use 580 * is not consistent with its use as part of an ordered search tree. 581 */ 582 #define defer_next right 583 584 static void 585 vm_map_process_deferred(void) 586 { 587 struct thread *td; 588 vm_map_entry_t entry, next; 589 vm_object_t object; 590 591 td = curthread; 592 entry = td->td_map_def_user; 593 td->td_map_def_user = NULL; 594 while (entry != NULL) { 595 next = entry->defer_next; 596 MPASS((entry->eflags & (MAP_ENTRY_WRITECNT | 597 MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT | 598 MAP_ENTRY_VN_EXEC)); 599 if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) { 600 /* 601 * Decrement the object's writemappings and 602 * possibly the vnode's v_writecount. 603 */ 604 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 605 ("Submap with writecount")); 606 object = entry->object.vm_object; 607 KASSERT(object != NULL, ("No object for writecount")); 608 vm_pager_release_writecount(object, entry->start, 609 entry->end); 610 } 611 vm_map_entry_set_vnode_text(entry, false); 612 vm_map_entry_deallocate(entry, FALSE); 613 entry = next; 614 } 615 } 616 617 #ifdef INVARIANTS 618 static void 619 _vm_map_assert_locked(vm_map_t map, const char *file, int line) 620 { 621 622 if (map->system_map) 623 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 624 else 625 sx_assert_(&map->lock, SA_XLOCKED, file, line); 626 } 627 628 #define VM_MAP_ASSERT_LOCKED(map) \ 629 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 630 631 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL }; 632 #ifdef DIAGNOSTIC 633 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK; 634 #else 635 static int enable_vmmap_check = VMMAP_CHECK_NONE; 636 #endif 637 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN, 638 &enable_vmmap_check, 0, "Enable vm map consistency checking"); 639 640 static void _vm_map_assert_consistent(vm_map_t map, int check); 641 642 #define VM_MAP_ASSERT_CONSISTENT(map) \ 643 _vm_map_assert_consistent(map, VMMAP_CHECK_ALL) 644 #ifdef DIAGNOSTIC 645 #define VM_MAP_UNLOCK_CONSISTENT(map) do { \ 646 if (map->nupdates > map->nentries) { \ 647 _vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK); \ 648 map->nupdates = 0; \ 649 } \ 650 } while (0) 651 #else 652 #define VM_MAP_UNLOCK_CONSISTENT(map) 653 #endif 654 #else 655 #define VM_MAP_ASSERT_LOCKED(map) 656 #define VM_MAP_ASSERT_CONSISTENT(map) 657 #define VM_MAP_UNLOCK_CONSISTENT(map) 658 #endif /* INVARIANTS */ 659 660 void 661 _vm_map_unlock(vm_map_t map, const char *file, int line) 662 { 663 664 VM_MAP_UNLOCK_CONSISTENT(map); 665 if (map->system_map) 666 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 667 else { 668 sx_xunlock_(&map->lock, file, line); 669 vm_map_process_deferred(); 670 } 671 } 672 673 void 674 _vm_map_lock_read(vm_map_t map, const char *file, int line) 675 { 676 677 if (map->system_map) 678 mtx_lock_flags_(&map->system_mtx, 0, file, line); 679 else 680 sx_slock_(&map->lock, file, line); 681 } 682 683 void 684 _vm_map_unlock_read(vm_map_t map, const char *file, int line) 685 { 686 687 if (map->system_map) 688 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 689 else { 690 sx_sunlock_(&map->lock, file, line); 691 vm_map_process_deferred(); 692 } 693 } 694 695 int 696 _vm_map_trylock(vm_map_t map, const char *file, int line) 697 { 698 int error; 699 700 error = map->system_map ? 701 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 702 !sx_try_xlock_(&map->lock, file, line); 703 if (error == 0) 704 map->timestamp++; 705 return (error == 0); 706 } 707 708 int 709 _vm_map_trylock_read(vm_map_t map, const char *file, int line) 710 { 711 int error; 712 713 error = map->system_map ? 714 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 715 !sx_try_slock_(&map->lock, file, line); 716 return (error == 0); 717 } 718 719 /* 720 * _vm_map_lock_upgrade: [ internal use only ] 721 * 722 * Tries to upgrade a read (shared) lock on the specified map to a write 723 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 724 * non-zero value if the upgrade fails. If the upgrade fails, the map is 725 * returned without a read or write lock held. 726 * 727 * Requires that the map be read locked. 728 */ 729 int 730 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 731 { 732 unsigned int last_timestamp; 733 734 if (map->system_map) { 735 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 736 } else { 737 if (!sx_try_upgrade_(&map->lock, file, line)) { 738 last_timestamp = map->timestamp; 739 sx_sunlock_(&map->lock, file, line); 740 vm_map_process_deferred(); 741 /* 742 * If the map's timestamp does not change while the 743 * map is unlocked, then the upgrade succeeds. 744 */ 745 sx_xlock_(&map->lock, file, line); 746 if (last_timestamp != map->timestamp) { 747 sx_xunlock_(&map->lock, file, line); 748 return (1); 749 } 750 } 751 } 752 map->timestamp++; 753 return (0); 754 } 755 756 void 757 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 758 { 759 760 if (map->system_map) { 761 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 762 } else { 763 VM_MAP_UNLOCK_CONSISTENT(map); 764 sx_downgrade_(&map->lock, file, line); 765 } 766 } 767 768 /* 769 * vm_map_locked: 770 * 771 * Returns a non-zero value if the caller holds a write (exclusive) lock 772 * on the specified map and the value "0" otherwise. 773 */ 774 int 775 vm_map_locked(vm_map_t map) 776 { 777 778 if (map->system_map) 779 return (mtx_owned(&map->system_mtx)); 780 else 781 return (sx_xlocked(&map->lock)); 782 } 783 784 /* 785 * _vm_map_unlock_and_wait: 786 * 787 * Atomically releases the lock on the specified map and puts the calling 788 * thread to sleep. The calling thread will remain asleep until either 789 * vm_map_wakeup() is performed on the map or the specified timeout is 790 * exceeded. 791 * 792 * WARNING! This function does not perform deferred deallocations of 793 * objects and map entries. Therefore, the calling thread is expected to 794 * reacquire the map lock after reawakening and later perform an ordinary 795 * unlock operation, such as vm_map_unlock(), before completing its 796 * operation on the map. 797 */ 798 int 799 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 800 { 801 802 VM_MAP_UNLOCK_CONSISTENT(map); 803 mtx_lock(&map_sleep_mtx); 804 if (map->system_map) 805 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 806 else 807 sx_xunlock_(&map->lock, file, line); 808 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 809 timo)); 810 } 811 812 /* 813 * vm_map_wakeup: 814 * 815 * Awaken any threads that have slept on the map using 816 * vm_map_unlock_and_wait(). 817 */ 818 void 819 vm_map_wakeup(vm_map_t map) 820 { 821 822 /* 823 * Acquire and release map_sleep_mtx to prevent a wakeup() 824 * from being performed (and lost) between the map unlock 825 * and the msleep() in _vm_map_unlock_and_wait(). 826 */ 827 mtx_lock(&map_sleep_mtx); 828 mtx_unlock(&map_sleep_mtx); 829 wakeup(&map->root); 830 } 831 832 void 833 vm_map_busy(vm_map_t map) 834 { 835 836 VM_MAP_ASSERT_LOCKED(map); 837 map->busy++; 838 } 839 840 void 841 vm_map_unbusy(vm_map_t map) 842 { 843 844 VM_MAP_ASSERT_LOCKED(map); 845 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 846 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 847 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 848 wakeup(&map->busy); 849 } 850 } 851 852 void 853 vm_map_wait_busy(vm_map_t map) 854 { 855 856 VM_MAP_ASSERT_LOCKED(map); 857 while (map->busy) { 858 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 859 if (map->system_map) 860 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 861 else 862 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 863 } 864 map->timestamp++; 865 } 866 867 long 868 vmspace_resident_count(struct vmspace *vmspace) 869 { 870 return pmap_resident_count(vmspace_pmap(vmspace)); 871 } 872 873 /* 874 * vm_map_create: 875 * 876 * Creates and returns a new empty VM map with 877 * the given physical map structure, and having 878 * the given lower and upper address bounds. 879 */ 880 vm_map_t 881 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 882 { 883 vm_map_t result; 884 885 result = uma_zalloc(mapzone, M_WAITOK); 886 CTR1(KTR_VM, "vm_map_create: %p", result); 887 _vm_map_init(result, pmap, min, max); 888 return (result); 889 } 890 891 /* 892 * Initialize an existing vm_map structure 893 * such as that in the vmspace structure. 894 */ 895 static void 896 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 897 { 898 899 map->header.eflags = MAP_ENTRY_HEADER; 900 map->needs_wakeup = FALSE; 901 map->system_map = 0; 902 map->pmap = pmap; 903 map->header.end = min; 904 map->header.start = max; 905 map->flags = 0; 906 map->header.left = map->header.right = &map->header; 907 map->root = NULL; 908 map->timestamp = 0; 909 map->busy = 0; 910 map->anon_loc = 0; 911 #ifdef DIAGNOSTIC 912 map->nupdates = 0; 913 #endif 914 } 915 916 void 917 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 918 { 919 920 _vm_map_init(map, pmap, min, max); 921 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 922 sx_init(&map->lock, "user map"); 923 } 924 925 /* 926 * vm_map_entry_dispose: [ internal use only ] 927 * 928 * Inverse of vm_map_entry_create. 929 */ 930 static void 931 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 932 { 933 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 934 } 935 936 /* 937 * vm_map_entry_create: [ internal use only ] 938 * 939 * Allocates a VM map entry for insertion. 940 * No entry fields are filled in. 941 */ 942 static vm_map_entry_t 943 vm_map_entry_create(vm_map_t map) 944 { 945 vm_map_entry_t new_entry; 946 947 if (map->system_map) 948 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 949 else 950 new_entry = uma_zalloc(mapentzone, M_WAITOK); 951 if (new_entry == NULL) 952 panic("vm_map_entry_create: kernel resources exhausted"); 953 return (new_entry); 954 } 955 956 /* 957 * vm_map_entry_set_behavior: 958 * 959 * Set the expected access behavior, either normal, random, or 960 * sequential. 961 */ 962 static inline void 963 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 964 { 965 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 966 (behavior & MAP_ENTRY_BEHAV_MASK); 967 } 968 969 /* 970 * vm_map_entry_max_free_{left,right}: 971 * 972 * Compute the size of the largest free gap between two entries, 973 * one the root of a tree and the other the ancestor of that root 974 * that is the least or greatest ancestor found on the search path. 975 */ 976 static inline vm_size_t 977 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor) 978 { 979 980 return (root->left != left_ancestor ? 981 root->left->max_free : root->start - left_ancestor->end); 982 } 983 984 static inline vm_size_t 985 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor) 986 { 987 988 return (root->right != right_ancestor ? 989 root->right->max_free : right_ancestor->start - root->end); 990 } 991 992 /* 993 * vm_map_entry_{pred,succ}: 994 * 995 * Find the {predecessor, successor} of the entry by taking one step 996 * in the appropriate direction and backtracking as much as necessary. 997 * vm_map_entry_succ is defined in vm_map.h. 998 */ 999 static inline vm_map_entry_t 1000 vm_map_entry_pred(vm_map_entry_t entry) 1001 { 1002 vm_map_entry_t prior; 1003 1004 prior = entry->left; 1005 if (prior->right->start < entry->start) { 1006 do 1007 prior = prior->right; 1008 while (prior->right != entry); 1009 } 1010 return (prior); 1011 } 1012 1013 static inline vm_size_t 1014 vm_size_max(vm_size_t a, vm_size_t b) 1015 { 1016 1017 return (a > b ? a : b); 1018 } 1019 1020 #define SPLAY_LEFT_STEP(root, y, llist, rlist, test) do { \ 1021 vm_map_entry_t z; \ 1022 vm_size_t max_free; \ 1023 \ 1024 /* \ 1025 * Infer root->right->max_free == root->max_free when \ 1026 * y->max_free < root->max_free || root->max_free == 0. \ 1027 * Otherwise, look right to find it. \ 1028 */ \ 1029 y = root->left; \ 1030 max_free = root->max_free; \ 1031 KASSERT(max_free >= vm_map_entry_max_free_right(root, rlist), \ 1032 ("%s: max_free invariant fails", __func__)); \ 1033 if (y == llist ? max_free > 0 : max_free - 1 < y->max_free) \ 1034 max_free = vm_map_entry_max_free_right(root, rlist); \ 1035 if (y != llist && (test)) { \ 1036 /* Rotate right and make y root. */ \ 1037 z = y->right; \ 1038 if (z != root) { \ 1039 root->left = z; \ 1040 y->right = root; \ 1041 if (max_free < y->max_free) \ 1042 root->max_free = max_free = \ 1043 vm_size_max(max_free, z->max_free); \ 1044 } else if (max_free < y->max_free) \ 1045 root->max_free = max_free = \ 1046 vm_size_max(max_free, root->start - y->end);\ 1047 root = y; \ 1048 y = root->left; \ 1049 } \ 1050 /* Copy right->max_free. Put root on rlist. */ \ 1051 root->max_free = max_free; \ 1052 KASSERT(max_free == vm_map_entry_max_free_right(root, rlist), \ 1053 ("%s: max_free not copied from right", __func__)); \ 1054 root->left = rlist; \ 1055 rlist = root; \ 1056 root = y != llist ? y : NULL; \ 1057 } while (0) 1058 1059 #define SPLAY_RIGHT_STEP(root, y, llist, rlist, test) do { \ 1060 vm_map_entry_t z; \ 1061 vm_size_t max_free; \ 1062 \ 1063 /* \ 1064 * Infer root->left->max_free == root->max_free when \ 1065 * y->max_free < root->max_free || root->max_free == 0. \ 1066 * Otherwise, look left to find it. \ 1067 */ \ 1068 y = root->right; \ 1069 max_free = root->max_free; \ 1070 KASSERT(max_free >= vm_map_entry_max_free_left(root, llist), \ 1071 ("%s: max_free invariant fails", __func__)); \ 1072 if (y == rlist ? max_free > 0 : max_free - 1 < y->max_free) \ 1073 max_free = vm_map_entry_max_free_left(root, llist); \ 1074 if (y != rlist && (test)) { \ 1075 /* Rotate left and make y root. */ \ 1076 z = y->left; \ 1077 if (z != root) { \ 1078 root->right = z; \ 1079 y->left = root; \ 1080 if (max_free < y->max_free) \ 1081 root->max_free = max_free = \ 1082 vm_size_max(max_free, z->max_free); \ 1083 } else if (max_free < y->max_free) \ 1084 root->max_free = max_free = \ 1085 vm_size_max(max_free, y->start - root->end);\ 1086 root = y; \ 1087 y = root->right; \ 1088 } \ 1089 /* Copy left->max_free. Put root on llist. */ \ 1090 root->max_free = max_free; \ 1091 KASSERT(max_free == vm_map_entry_max_free_left(root, llist), \ 1092 ("%s: max_free not copied from left", __func__)); \ 1093 root->right = llist; \ 1094 llist = root; \ 1095 root = y != rlist ? y : NULL; \ 1096 } while (0) 1097 1098 /* 1099 * Walk down the tree until we find addr or a gap where addr would go, breaking 1100 * off left and right subtrees of nodes less than, or greater than addr. Treat 1101 * subtrees with root->max_free < length as empty trees. llist and rlist are 1102 * the two sides in reverse order (bottom-up), with llist linked by the right 1103 * pointer and rlist linked by the left pointer in the vm_map_entry, and both 1104 * lists terminated by &map->header. This function, and the subsequent call to 1105 * vm_map_splay_merge_{left,right,pred,succ}, rely on the start and end address 1106 * values in &map->header. 1107 */ 1108 static __always_inline vm_map_entry_t 1109 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length, 1110 vm_map_entry_t *llist, vm_map_entry_t *rlist) 1111 { 1112 vm_map_entry_t left, right, root, y; 1113 1114 left = right = &map->header; 1115 root = map->root; 1116 while (root != NULL && root->max_free >= length) { 1117 KASSERT(left->end <= root->start && 1118 root->end <= right->start, 1119 ("%s: root not within tree bounds", __func__)); 1120 if (addr < root->start) { 1121 SPLAY_LEFT_STEP(root, y, left, right, 1122 y->max_free >= length && addr < y->start); 1123 } else if (addr >= root->end) { 1124 SPLAY_RIGHT_STEP(root, y, left, right, 1125 y->max_free >= length && addr >= y->end); 1126 } else 1127 break; 1128 } 1129 *llist = left; 1130 *rlist = right; 1131 return (root); 1132 } 1133 1134 static __always_inline void 1135 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *rlist) 1136 { 1137 vm_map_entry_t hi, right, y; 1138 1139 right = *rlist; 1140 hi = root->right == right ? NULL : root->right; 1141 if (hi == NULL) 1142 return; 1143 do 1144 SPLAY_LEFT_STEP(hi, y, root, right, true); 1145 while (hi != NULL); 1146 *rlist = right; 1147 } 1148 1149 static __always_inline void 1150 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *llist) 1151 { 1152 vm_map_entry_t left, lo, y; 1153 1154 left = *llist; 1155 lo = root->left == left ? NULL : root->left; 1156 if (lo == NULL) 1157 return; 1158 do 1159 SPLAY_RIGHT_STEP(lo, y, left, root, true); 1160 while (lo != NULL); 1161 *llist = left; 1162 } 1163 1164 static inline void 1165 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b) 1166 { 1167 vm_map_entry_t tmp; 1168 1169 tmp = *b; 1170 *b = *a; 1171 *a = tmp; 1172 } 1173 1174 /* 1175 * Walk back up the two spines, flip the pointers and set max_free. The 1176 * subtrees of the root go at the bottom of llist and rlist. 1177 */ 1178 static vm_size_t 1179 vm_map_splay_merge_left_walk(vm_map_entry_t header, vm_map_entry_t root, 1180 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t llist) 1181 { 1182 do { 1183 /* 1184 * The max_free values of the children of llist are in 1185 * llist->max_free and max_free. Update with the 1186 * max value. 1187 */ 1188 llist->max_free = max_free = 1189 vm_size_max(llist->max_free, max_free); 1190 vm_map_entry_swap(&llist->right, &tail); 1191 vm_map_entry_swap(&tail, &llist); 1192 } while (llist != header); 1193 root->left = tail; 1194 return (max_free); 1195 } 1196 1197 /* 1198 * When llist is known to be the predecessor of root. 1199 */ 1200 static inline vm_size_t 1201 vm_map_splay_merge_pred(vm_map_entry_t header, vm_map_entry_t root, 1202 vm_map_entry_t llist) 1203 { 1204 vm_size_t max_free; 1205 1206 max_free = root->start - llist->end; 1207 if (llist != header) { 1208 max_free = vm_map_splay_merge_left_walk(header, root, 1209 root, max_free, llist); 1210 } else { 1211 root->left = header; 1212 header->right = root; 1213 } 1214 return (max_free); 1215 } 1216 1217 /* 1218 * When llist may or may not be the predecessor of root. 1219 */ 1220 static inline vm_size_t 1221 vm_map_splay_merge_left(vm_map_entry_t header, vm_map_entry_t root, 1222 vm_map_entry_t llist) 1223 { 1224 vm_size_t max_free; 1225 1226 max_free = vm_map_entry_max_free_left(root, llist); 1227 if (llist != header) { 1228 max_free = vm_map_splay_merge_left_walk(header, root, 1229 root->left == llist ? root : root->left, 1230 max_free, llist); 1231 } 1232 return (max_free); 1233 } 1234 1235 static vm_size_t 1236 vm_map_splay_merge_right_walk(vm_map_entry_t header, vm_map_entry_t root, 1237 vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t rlist) 1238 { 1239 do { 1240 /* 1241 * The max_free values of the children of rlist are in 1242 * rlist->max_free and max_free. Update with the 1243 * max value. 1244 */ 1245 rlist->max_free = max_free = 1246 vm_size_max(rlist->max_free, max_free); 1247 vm_map_entry_swap(&rlist->left, &tail); 1248 vm_map_entry_swap(&tail, &rlist); 1249 } while (rlist != header); 1250 root->right = tail; 1251 return (max_free); 1252 } 1253 1254 /* 1255 * When rlist is known to be the succecessor of root. 1256 */ 1257 static inline vm_size_t 1258 vm_map_splay_merge_succ(vm_map_entry_t header, vm_map_entry_t root, 1259 vm_map_entry_t rlist) 1260 { 1261 vm_size_t max_free; 1262 1263 max_free = rlist->start - root->end; 1264 if (rlist != header) { 1265 max_free = vm_map_splay_merge_right_walk(header, root, 1266 root, max_free, rlist); 1267 } else { 1268 root->right = header; 1269 header->left = root; 1270 } 1271 return (max_free); 1272 } 1273 1274 /* 1275 * When rlist may or may not be the succecessor of root. 1276 */ 1277 static inline vm_size_t 1278 vm_map_splay_merge_right(vm_map_entry_t header, vm_map_entry_t root, 1279 vm_map_entry_t rlist) 1280 { 1281 vm_size_t max_free; 1282 1283 max_free = vm_map_entry_max_free_right(root, rlist); 1284 if (rlist != header) { 1285 max_free = vm_map_splay_merge_right_walk(header, root, 1286 root->right == rlist ? root : root->right, 1287 max_free, rlist); 1288 } 1289 return (max_free); 1290 } 1291 1292 /* 1293 * vm_map_splay: 1294 * 1295 * The Sleator and Tarjan top-down splay algorithm with the 1296 * following variation. Max_free must be computed bottom-up, so 1297 * on the downward pass, maintain the left and right spines in 1298 * reverse order. Then, make a second pass up each side to fix 1299 * the pointers and compute max_free. The time bound is O(log n) 1300 * amortized. 1301 * 1302 * The tree is threaded, which means that there are no null pointers. 1303 * When a node has no left child, its left pointer points to its 1304 * predecessor, which the last ancestor on the search path from the root 1305 * where the search branched right. Likewise, when a node has no right 1306 * child, its right pointer points to its successor. The map header node 1307 * is the predecessor of the first map entry, and the successor of the 1308 * last. 1309 * 1310 * The new root is the vm_map_entry containing "addr", or else an 1311 * adjacent entry (lower if possible) if addr is not in the tree. 1312 * 1313 * The map must be locked, and leaves it so. 1314 * 1315 * Returns: the new root. 1316 */ 1317 static vm_map_entry_t 1318 vm_map_splay(vm_map_t map, vm_offset_t addr) 1319 { 1320 vm_map_entry_t header, llist, rlist, root; 1321 vm_size_t max_free_left, max_free_right; 1322 1323 header = &map->header; 1324 root = vm_map_splay_split(map, addr, 0, &llist, &rlist); 1325 if (root != NULL) { 1326 max_free_left = vm_map_splay_merge_left(header, root, llist); 1327 max_free_right = vm_map_splay_merge_right(header, root, rlist); 1328 } else if (llist != header) { 1329 /* 1330 * Recover the greatest node in the left 1331 * subtree and make it the root. 1332 */ 1333 root = llist; 1334 llist = root->right; 1335 max_free_left = vm_map_splay_merge_left(header, root, llist); 1336 max_free_right = vm_map_splay_merge_succ(header, root, rlist); 1337 } else if (rlist != header) { 1338 /* 1339 * Recover the least node in the right 1340 * subtree and make it the root. 1341 */ 1342 root = rlist; 1343 rlist = root->left; 1344 max_free_left = vm_map_splay_merge_pred(header, root, llist); 1345 max_free_right = vm_map_splay_merge_right(header, root, rlist); 1346 } else { 1347 /* There is no root. */ 1348 return (NULL); 1349 } 1350 root->max_free = vm_size_max(max_free_left, max_free_right); 1351 map->root = root; 1352 VM_MAP_ASSERT_CONSISTENT(map); 1353 return (root); 1354 } 1355 1356 /* 1357 * vm_map_entry_{un,}link: 1358 * 1359 * Insert/remove entries from maps. 1360 */ 1361 static void 1362 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry) 1363 { 1364 vm_map_entry_t header, llist, rlist, root; 1365 1366 CTR3(KTR_VM, 1367 "vm_map_entry_link: map %p, nentries %d, entry %p", map, 1368 map->nentries, entry); 1369 VM_MAP_ASSERT_LOCKED(map); 1370 map->nentries++; 1371 header = &map->header; 1372 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist); 1373 KASSERT(root == NULL, 1374 ("vm_map_entry_link: link object already mapped")); 1375 root = entry; 1376 root->max_free = vm_size_max( 1377 vm_map_splay_merge_pred(header, root, llist), 1378 vm_map_splay_merge_succ(header, root, rlist)); 1379 map->root = root; 1380 VM_MAP_ASSERT_CONSISTENT(map); 1381 } 1382 1383 enum unlink_merge_type { 1384 UNLINK_MERGE_NONE, 1385 UNLINK_MERGE_NEXT 1386 }; 1387 1388 static void 1389 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry, 1390 enum unlink_merge_type op) 1391 { 1392 vm_map_entry_t header, llist, rlist, root; 1393 vm_size_t max_free_left, max_free_right; 1394 1395 VM_MAP_ASSERT_LOCKED(map); 1396 header = &map->header; 1397 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist); 1398 KASSERT(root != NULL, 1399 ("vm_map_entry_unlink: unlink object not mapped")); 1400 1401 vm_map_splay_findprev(root, &llist); 1402 vm_map_splay_findnext(root, &rlist); 1403 if (op == UNLINK_MERGE_NEXT) { 1404 rlist->start = root->start; 1405 rlist->offset = root->offset; 1406 } 1407 if (llist != header) { 1408 root = llist; 1409 llist = root->right; 1410 max_free_left = vm_map_splay_merge_left(header, root, llist); 1411 max_free_right = vm_map_splay_merge_succ(header, root, rlist); 1412 } else if (rlist != header) { 1413 root = rlist; 1414 rlist = root->left; 1415 max_free_left = vm_map_splay_merge_pred(header, root, llist); 1416 max_free_right = vm_map_splay_merge_right(header, root, rlist); 1417 } else { 1418 header->left = header->right = header; 1419 root = NULL; 1420 } 1421 if (root != NULL) 1422 root->max_free = vm_size_max(max_free_left, max_free_right); 1423 map->root = root; 1424 VM_MAP_ASSERT_CONSISTENT(map); 1425 map->nentries--; 1426 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1427 map->nentries, entry); 1428 } 1429 1430 /* 1431 * vm_map_entry_resize: 1432 * 1433 * Resize a vm_map_entry, recompute the amount of free space that 1434 * follows it and propagate that value up the tree. 1435 * 1436 * The map must be locked, and leaves it so. 1437 */ 1438 static void 1439 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount) 1440 { 1441 vm_map_entry_t header, llist, rlist, root; 1442 1443 VM_MAP_ASSERT_LOCKED(map); 1444 header = &map->header; 1445 root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist); 1446 KASSERT(root != NULL, ("%s: resize object not mapped", __func__)); 1447 vm_map_splay_findnext(root, &rlist); 1448 entry->end += grow_amount; 1449 root->max_free = vm_size_max( 1450 vm_map_splay_merge_left(header, root, llist), 1451 vm_map_splay_merge_succ(header, root, rlist)); 1452 map->root = root; 1453 VM_MAP_ASSERT_CONSISTENT(map); 1454 CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p", 1455 __func__, map, map->nentries, entry); 1456 } 1457 1458 /* 1459 * vm_map_lookup_entry: [ internal use only ] 1460 * 1461 * Finds the map entry containing (or 1462 * immediately preceding) the specified address 1463 * in the given map; the entry is returned 1464 * in the "entry" parameter. The boolean 1465 * result indicates whether the address is 1466 * actually contained in the map. 1467 */ 1468 boolean_t 1469 vm_map_lookup_entry( 1470 vm_map_t map, 1471 vm_offset_t address, 1472 vm_map_entry_t *entry) /* OUT */ 1473 { 1474 vm_map_entry_t cur, header, lbound, ubound; 1475 boolean_t locked; 1476 1477 /* 1478 * If the map is empty, then the map entry immediately preceding 1479 * "address" is the map's header. 1480 */ 1481 header = &map->header; 1482 cur = map->root; 1483 if (cur == NULL) { 1484 *entry = header; 1485 return (FALSE); 1486 } 1487 if (address >= cur->start && cur->end > address) { 1488 *entry = cur; 1489 return (TRUE); 1490 } 1491 if ((locked = vm_map_locked(map)) || 1492 sx_try_upgrade(&map->lock)) { 1493 /* 1494 * Splay requires a write lock on the map. However, it only 1495 * restructures the binary search tree; it does not otherwise 1496 * change the map. Thus, the map's timestamp need not change 1497 * on a temporary upgrade. 1498 */ 1499 cur = vm_map_splay(map, address); 1500 if (!locked) { 1501 VM_MAP_UNLOCK_CONSISTENT(map); 1502 sx_downgrade(&map->lock); 1503 } 1504 1505 /* 1506 * If "address" is contained within a map entry, the new root 1507 * is that map entry. Otherwise, the new root is a map entry 1508 * immediately before or after "address". 1509 */ 1510 if (address < cur->start) { 1511 *entry = header; 1512 return (FALSE); 1513 } 1514 *entry = cur; 1515 return (address < cur->end); 1516 } 1517 /* 1518 * Since the map is only locked for read access, perform a 1519 * standard binary search tree lookup for "address". 1520 */ 1521 lbound = ubound = header; 1522 for (;;) { 1523 if (address < cur->start) { 1524 ubound = cur; 1525 cur = cur->left; 1526 if (cur == lbound) 1527 break; 1528 } else if (cur->end <= address) { 1529 lbound = cur; 1530 cur = cur->right; 1531 if (cur == ubound) 1532 break; 1533 } else { 1534 *entry = cur; 1535 return (TRUE); 1536 } 1537 } 1538 *entry = lbound; 1539 return (FALSE); 1540 } 1541 1542 /* 1543 * vm_map_insert: 1544 * 1545 * Inserts the given whole VM object into the target 1546 * map at the specified address range. The object's 1547 * size should match that of the address range. 1548 * 1549 * Requires that the map be locked, and leaves it so. 1550 * 1551 * If object is non-NULL, ref count must be bumped by caller 1552 * prior to making call to account for the new entry. 1553 */ 1554 int 1555 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1556 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow) 1557 { 1558 vm_map_entry_t new_entry, next_entry, prev_entry; 1559 struct ucred *cred; 1560 vm_eflags_t protoeflags; 1561 vm_inherit_t inheritance; 1562 1563 VM_MAP_ASSERT_LOCKED(map); 1564 KASSERT(object != kernel_object || 1565 (cow & MAP_COPY_ON_WRITE) == 0, 1566 ("vm_map_insert: kernel object and COW")); 1567 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0, 1568 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1569 KASSERT((prot & ~max) == 0, 1570 ("prot %#x is not subset of max_prot %#x", prot, max)); 1571 1572 /* 1573 * Check that the start and end points are not bogus. 1574 */ 1575 if (start < vm_map_min(map) || end > vm_map_max(map) || 1576 start >= end) 1577 return (KERN_INVALID_ADDRESS); 1578 1579 /* 1580 * Find the entry prior to the proposed starting address; if it's part 1581 * of an existing entry, this range is bogus. 1582 */ 1583 if (vm_map_lookup_entry(map, start, &prev_entry)) 1584 return (KERN_NO_SPACE); 1585 1586 /* 1587 * Assert that the next entry doesn't overlap the end point. 1588 */ 1589 next_entry = vm_map_entry_succ(prev_entry); 1590 if (next_entry->start < end) 1591 return (KERN_NO_SPACE); 1592 1593 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL || 1594 max != VM_PROT_NONE)) 1595 return (KERN_INVALID_ARGUMENT); 1596 1597 protoeflags = 0; 1598 if (cow & MAP_COPY_ON_WRITE) 1599 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; 1600 if (cow & MAP_NOFAULT) 1601 protoeflags |= MAP_ENTRY_NOFAULT; 1602 if (cow & MAP_DISABLE_SYNCER) 1603 protoeflags |= MAP_ENTRY_NOSYNC; 1604 if (cow & MAP_DISABLE_COREDUMP) 1605 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1606 if (cow & MAP_STACK_GROWS_DOWN) 1607 protoeflags |= MAP_ENTRY_GROWS_DOWN; 1608 if (cow & MAP_STACK_GROWS_UP) 1609 protoeflags |= MAP_ENTRY_GROWS_UP; 1610 if (cow & MAP_WRITECOUNT) 1611 protoeflags |= MAP_ENTRY_WRITECNT; 1612 if (cow & MAP_VN_EXEC) 1613 protoeflags |= MAP_ENTRY_VN_EXEC; 1614 if ((cow & MAP_CREATE_GUARD) != 0) 1615 protoeflags |= MAP_ENTRY_GUARD; 1616 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0) 1617 protoeflags |= MAP_ENTRY_STACK_GAP_DN; 1618 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0) 1619 protoeflags |= MAP_ENTRY_STACK_GAP_UP; 1620 if (cow & MAP_INHERIT_SHARE) 1621 inheritance = VM_INHERIT_SHARE; 1622 else 1623 inheritance = VM_INHERIT_DEFAULT; 1624 1625 cred = NULL; 1626 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0) 1627 goto charged; 1628 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1629 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1630 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1631 return (KERN_RESOURCE_SHORTAGE); 1632 KASSERT(object == NULL || 1633 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 || 1634 object->cred == NULL, 1635 ("overcommit: vm_map_insert o %p", object)); 1636 cred = curthread->td_ucred; 1637 } 1638 1639 charged: 1640 /* Expand the kernel pmap, if necessary. */ 1641 if (map == kernel_map && end > kernel_vm_end) 1642 pmap_growkernel(end); 1643 if (object != NULL) { 1644 /* 1645 * OBJ_ONEMAPPING must be cleared unless this mapping 1646 * is trivially proven to be the only mapping for any 1647 * of the object's pages. (Object granularity 1648 * reference counting is insufficient to recognize 1649 * aliases with precision.) 1650 */ 1651 if ((object->flags & OBJ_ANON) != 0) { 1652 VM_OBJECT_WLOCK(object); 1653 if (object->ref_count > 1 || object->shadow_count != 0) 1654 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1655 VM_OBJECT_WUNLOCK(object); 1656 } 1657 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) == 1658 protoeflags && 1659 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP | 1660 MAP_VN_EXEC)) == 0 && 1661 prev_entry->end == start && (prev_entry->cred == cred || 1662 (prev_entry->object.vm_object != NULL && 1663 prev_entry->object.vm_object->cred == cred)) && 1664 vm_object_coalesce(prev_entry->object.vm_object, 1665 prev_entry->offset, 1666 (vm_size_t)(prev_entry->end - prev_entry->start), 1667 (vm_size_t)(end - prev_entry->end), cred != NULL && 1668 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) { 1669 /* 1670 * We were able to extend the object. Determine if we 1671 * can extend the previous map entry to include the 1672 * new range as well. 1673 */ 1674 if (prev_entry->inheritance == inheritance && 1675 prev_entry->protection == prot && 1676 prev_entry->max_protection == max && 1677 prev_entry->wired_count == 0) { 1678 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) == 1679 0, ("prev_entry %p has incoherent wiring", 1680 prev_entry)); 1681 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0) 1682 map->size += end - prev_entry->end; 1683 vm_map_entry_resize(map, prev_entry, 1684 end - prev_entry->end); 1685 vm_map_try_merge_entries(map, prev_entry, next_entry); 1686 return (KERN_SUCCESS); 1687 } 1688 1689 /* 1690 * If we can extend the object but cannot extend the 1691 * map entry, we have to create a new map entry. We 1692 * must bump the ref count on the extended object to 1693 * account for it. object may be NULL. 1694 */ 1695 object = prev_entry->object.vm_object; 1696 offset = prev_entry->offset + 1697 (prev_entry->end - prev_entry->start); 1698 vm_object_reference(object); 1699 if (cred != NULL && object != NULL && object->cred != NULL && 1700 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1701 /* Object already accounts for this uid. */ 1702 cred = NULL; 1703 } 1704 } 1705 if (cred != NULL) 1706 crhold(cred); 1707 1708 /* 1709 * Create a new entry 1710 */ 1711 new_entry = vm_map_entry_create(map); 1712 new_entry->start = start; 1713 new_entry->end = end; 1714 new_entry->cred = NULL; 1715 1716 new_entry->eflags = protoeflags; 1717 new_entry->object.vm_object = object; 1718 new_entry->offset = offset; 1719 1720 new_entry->inheritance = inheritance; 1721 new_entry->protection = prot; 1722 new_entry->max_protection = max; 1723 new_entry->wired_count = 0; 1724 new_entry->wiring_thread = NULL; 1725 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1726 new_entry->next_read = start; 1727 1728 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1729 ("overcommit: vm_map_insert leaks vm_map %p", new_entry)); 1730 new_entry->cred = cred; 1731 1732 /* 1733 * Insert the new entry into the list 1734 */ 1735 vm_map_entry_link(map, new_entry); 1736 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0) 1737 map->size += new_entry->end - new_entry->start; 1738 1739 /* 1740 * Try to coalesce the new entry with both the previous and next 1741 * entries in the list. Previously, we only attempted to coalesce 1742 * with the previous entry when object is NULL. Here, we handle the 1743 * other cases, which are less common. 1744 */ 1745 vm_map_try_merge_entries(map, prev_entry, new_entry); 1746 vm_map_try_merge_entries(map, new_entry, next_entry); 1747 1748 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) { 1749 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset), 1750 end - start, cow & MAP_PREFAULT_PARTIAL); 1751 } 1752 1753 return (KERN_SUCCESS); 1754 } 1755 1756 /* 1757 * vm_map_findspace: 1758 * 1759 * Find the first fit (lowest VM address) for "length" free bytes 1760 * beginning at address >= start in the given map. 1761 * 1762 * In a vm_map_entry, "max_free" is the maximum amount of 1763 * contiguous free space between an entry in its subtree and a 1764 * neighbor of that entry. This allows finding a free region in 1765 * one path down the tree, so O(log n) amortized with splay 1766 * trees. 1767 * 1768 * The map must be locked, and leaves it so. 1769 * 1770 * Returns: starting address if sufficient space, 1771 * vm_map_max(map)-length+1 if insufficient space. 1772 */ 1773 vm_offset_t 1774 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length) 1775 { 1776 vm_map_entry_t header, llist, rlist, root, y; 1777 vm_size_t left_length, max_free_left, max_free_right; 1778 vm_offset_t gap_end; 1779 1780 /* 1781 * Request must fit within min/max VM address and must avoid 1782 * address wrap. 1783 */ 1784 start = MAX(start, vm_map_min(map)); 1785 if (start >= vm_map_max(map) || length > vm_map_max(map) - start) 1786 return (vm_map_max(map) - length + 1); 1787 1788 /* Empty tree means wide open address space. */ 1789 if (map->root == NULL) 1790 return (start); 1791 1792 /* 1793 * After splay_split, if start is within an entry, push it to the start 1794 * of the following gap. If rlist is at the end of the gap containing 1795 * start, save the end of that gap in gap_end to see if the gap is big 1796 * enough; otherwise set gap_end to start skip gap-checking and move 1797 * directly to a search of the right subtree. 1798 */ 1799 header = &map->header; 1800 root = vm_map_splay_split(map, start, length, &llist, &rlist); 1801 gap_end = rlist->start; 1802 if (root != NULL) { 1803 start = root->end; 1804 if (root->right != rlist) 1805 gap_end = start; 1806 max_free_left = vm_map_splay_merge_left(header, root, llist); 1807 max_free_right = vm_map_splay_merge_right(header, root, rlist); 1808 } else if (rlist != header) { 1809 root = rlist; 1810 rlist = root->left; 1811 max_free_left = vm_map_splay_merge_pred(header, root, llist); 1812 max_free_right = vm_map_splay_merge_right(header, root, rlist); 1813 } else { 1814 root = llist; 1815 llist = root->right; 1816 max_free_left = vm_map_splay_merge_left(header, root, llist); 1817 max_free_right = vm_map_splay_merge_succ(header, root, rlist); 1818 } 1819 root->max_free = vm_size_max(max_free_left, max_free_right); 1820 map->root = root; 1821 VM_MAP_ASSERT_CONSISTENT(map); 1822 if (length <= gap_end - start) 1823 return (start); 1824 1825 /* With max_free, can immediately tell if no solution. */ 1826 if (root->right == header || length > root->right->max_free) 1827 return (vm_map_max(map) - length + 1); 1828 1829 /* 1830 * Splay for the least large-enough gap in the right subtree. 1831 */ 1832 llist = rlist = header; 1833 for (left_length = 0;; 1834 left_length = vm_map_entry_max_free_left(root, llist)) { 1835 if (length <= left_length) 1836 SPLAY_LEFT_STEP(root, y, llist, rlist, 1837 length <= vm_map_entry_max_free_left(y, llist)); 1838 else 1839 SPLAY_RIGHT_STEP(root, y, llist, rlist, 1840 length > vm_map_entry_max_free_left(y, root)); 1841 if (root == NULL) 1842 break; 1843 } 1844 root = llist; 1845 llist = root->right; 1846 max_free_left = vm_map_splay_merge_left(header, root, llist); 1847 if (rlist == header) { 1848 root->max_free = vm_size_max(max_free_left, 1849 vm_map_splay_merge_succ(header, root, rlist)); 1850 } else { 1851 y = rlist; 1852 rlist = y->left; 1853 y->max_free = vm_size_max( 1854 vm_map_splay_merge_pred(root, y, root), 1855 vm_map_splay_merge_right(header, y, rlist)); 1856 root->max_free = vm_size_max(max_free_left, y->max_free); 1857 } 1858 map->root = root; 1859 VM_MAP_ASSERT_CONSISTENT(map); 1860 return (root->end); 1861 } 1862 1863 int 1864 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1865 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1866 vm_prot_t max, int cow) 1867 { 1868 vm_offset_t end; 1869 int result; 1870 1871 end = start + length; 1872 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1873 object == NULL, 1874 ("vm_map_fixed: non-NULL backing object for stack")); 1875 vm_map_lock(map); 1876 VM_MAP_RANGE_CHECK(map, start, end); 1877 if ((cow & MAP_CHECK_EXCL) == 0) 1878 vm_map_delete(map, start, end); 1879 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1880 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1881 prot, max, cow); 1882 } else { 1883 result = vm_map_insert(map, object, offset, start, end, 1884 prot, max, cow); 1885 } 1886 vm_map_unlock(map); 1887 return (result); 1888 } 1889 1890 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10}; 1891 static const int aslr_pages_rnd_32[2] = {0x100, 0x4}; 1892 1893 static int cluster_anon = 1; 1894 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW, 1895 &cluster_anon, 0, 1896 "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always"); 1897 1898 static bool 1899 clustering_anon_allowed(vm_offset_t addr) 1900 { 1901 1902 switch (cluster_anon) { 1903 case 0: 1904 return (false); 1905 case 1: 1906 return (addr == 0); 1907 case 2: 1908 default: 1909 return (true); 1910 } 1911 } 1912 1913 static long aslr_restarts; 1914 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD, 1915 &aslr_restarts, 0, 1916 "Number of aslr failures"); 1917 1918 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 1919 1920 /* 1921 * Searches for the specified amount of free space in the given map with the 1922 * specified alignment. Performs an address-ordered, first-fit search from 1923 * the given address "*addr", with an optional upper bound "max_addr". If the 1924 * parameter "alignment" is zero, then the alignment is computed from the 1925 * given (object, offset) pair so as to enable the greatest possible use of 1926 * superpage mappings. Returns KERN_SUCCESS and the address of the free space 1927 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE. 1928 * 1929 * The map must be locked. Initially, there must be at least "length" bytes 1930 * of free space at the given address. 1931 */ 1932 static int 1933 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1934 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr, 1935 vm_offset_t alignment) 1936 { 1937 vm_offset_t aligned_addr, free_addr; 1938 1939 VM_MAP_ASSERT_LOCKED(map); 1940 free_addr = *addr; 1941 KASSERT(free_addr == vm_map_findspace(map, free_addr, length), 1942 ("caller failed to provide space %#jx at address %p", 1943 (uintmax_t)length, (void *)free_addr)); 1944 for (;;) { 1945 /* 1946 * At the start of every iteration, the free space at address 1947 * "*addr" is at least "length" bytes. 1948 */ 1949 if (alignment == 0) 1950 pmap_align_superpage(object, offset, addr, length); 1951 else if ((*addr & (alignment - 1)) != 0) { 1952 *addr &= ~(alignment - 1); 1953 *addr += alignment; 1954 } 1955 aligned_addr = *addr; 1956 if (aligned_addr == free_addr) { 1957 /* 1958 * Alignment did not change "*addr", so "*addr" must 1959 * still provide sufficient free space. 1960 */ 1961 return (KERN_SUCCESS); 1962 } 1963 1964 /* 1965 * Test for address wrap on "*addr". A wrapped "*addr" could 1966 * be a valid address, in which case vm_map_findspace() cannot 1967 * be relied upon to fail. 1968 */ 1969 if (aligned_addr < free_addr) 1970 return (KERN_NO_SPACE); 1971 *addr = vm_map_findspace(map, aligned_addr, length); 1972 if (*addr + length > vm_map_max(map) || 1973 (max_addr != 0 && *addr + length > max_addr)) 1974 return (KERN_NO_SPACE); 1975 free_addr = *addr; 1976 if (free_addr == aligned_addr) { 1977 /* 1978 * If a successful call to vm_map_findspace() did not 1979 * change "*addr", then "*addr" must still be aligned 1980 * and provide sufficient free space. 1981 */ 1982 return (KERN_SUCCESS); 1983 } 1984 } 1985 } 1986 1987 /* 1988 * vm_map_find finds an unallocated region in the target address 1989 * map with the given length. The search is defined to be 1990 * first-fit from the specified address; the region found is 1991 * returned in the same parameter. 1992 * 1993 * If object is non-NULL, ref count must be bumped by caller 1994 * prior to making call to account for the new entry. 1995 */ 1996 int 1997 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1998 vm_offset_t *addr, /* IN/OUT */ 1999 vm_size_t length, vm_offset_t max_addr, int find_space, 2000 vm_prot_t prot, vm_prot_t max, int cow) 2001 { 2002 vm_offset_t alignment, curr_min_addr, min_addr; 2003 int gap, pidx, rv, try; 2004 bool cluster, en_aslr, update_anon; 2005 2006 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 2007 object == NULL, 2008 ("vm_map_find: non-NULL backing object for stack")); 2009 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE && 2010 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)); 2011 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 2012 (object->flags & OBJ_COLORED) == 0)) 2013 find_space = VMFS_ANY_SPACE; 2014 if (find_space >> 8 != 0) { 2015 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 2016 alignment = (vm_offset_t)1 << (find_space >> 8); 2017 } else 2018 alignment = 0; 2019 en_aslr = (map->flags & MAP_ASLR) != 0; 2020 update_anon = cluster = clustering_anon_allowed(*addr) && 2021 (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 && 2022 find_space != VMFS_NO_SPACE && object == NULL && 2023 (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP | 2024 MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE; 2025 curr_min_addr = min_addr = *addr; 2026 if (en_aslr && min_addr == 0 && !cluster && 2027 find_space != VMFS_NO_SPACE && 2028 (map->flags & MAP_ASLR_IGNSTART) != 0) 2029 curr_min_addr = min_addr = vm_map_min(map); 2030 try = 0; 2031 vm_map_lock(map); 2032 if (cluster) { 2033 curr_min_addr = map->anon_loc; 2034 if (curr_min_addr == 0) 2035 cluster = false; 2036 } 2037 if (find_space != VMFS_NO_SPACE) { 2038 KASSERT(find_space == VMFS_ANY_SPACE || 2039 find_space == VMFS_OPTIMAL_SPACE || 2040 find_space == VMFS_SUPER_SPACE || 2041 alignment != 0, ("unexpected VMFS flag")); 2042 again: 2043 /* 2044 * When creating an anonymous mapping, try clustering 2045 * with an existing anonymous mapping first. 2046 * 2047 * We make up to two attempts to find address space 2048 * for a given find_space value. The first attempt may 2049 * apply randomization or may cluster with an existing 2050 * anonymous mapping. If this first attempt fails, 2051 * perform a first-fit search of the available address 2052 * space. 2053 * 2054 * If all tries failed, and find_space is 2055 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE. 2056 * Again enable clustering and randomization. 2057 */ 2058 try++; 2059 MPASS(try <= 2); 2060 2061 if (try == 2) { 2062 /* 2063 * Second try: we failed either to find a 2064 * suitable region for randomizing the 2065 * allocation, or to cluster with an existing 2066 * mapping. Retry with free run. 2067 */ 2068 curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ? 2069 vm_map_min(map) : min_addr; 2070 atomic_add_long(&aslr_restarts, 1); 2071 } 2072 2073 if (try == 1 && en_aslr && !cluster) { 2074 /* 2075 * Find space for allocation, including 2076 * gap needed for later randomization. 2077 */ 2078 pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 && 2079 (find_space == VMFS_SUPER_SPACE || find_space == 2080 VMFS_OPTIMAL_SPACE) ? 1 : 0; 2081 gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR && 2082 (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ? 2083 aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx]; 2084 *addr = vm_map_findspace(map, curr_min_addr, 2085 length + gap * pagesizes[pidx]); 2086 if (*addr + length + gap * pagesizes[pidx] > 2087 vm_map_max(map)) 2088 goto again; 2089 /* And randomize the start address. */ 2090 *addr += (arc4random() % gap) * pagesizes[pidx]; 2091 if (max_addr != 0 && *addr + length > max_addr) 2092 goto again; 2093 } else { 2094 *addr = vm_map_findspace(map, curr_min_addr, length); 2095 if (*addr + length > vm_map_max(map) || 2096 (max_addr != 0 && *addr + length > max_addr)) { 2097 if (cluster) { 2098 cluster = false; 2099 MPASS(try == 1); 2100 goto again; 2101 } 2102 rv = KERN_NO_SPACE; 2103 goto done; 2104 } 2105 } 2106 2107 if (find_space != VMFS_ANY_SPACE && 2108 (rv = vm_map_alignspace(map, object, offset, addr, length, 2109 max_addr, alignment)) != KERN_SUCCESS) { 2110 if (find_space == VMFS_OPTIMAL_SPACE) { 2111 find_space = VMFS_ANY_SPACE; 2112 curr_min_addr = min_addr; 2113 cluster = update_anon; 2114 try = 0; 2115 goto again; 2116 } 2117 goto done; 2118 } 2119 } else if ((cow & MAP_REMAP) != 0) { 2120 if (*addr < vm_map_min(map) || 2121 *addr + length > vm_map_max(map) || 2122 *addr + length <= length) { 2123 rv = KERN_INVALID_ADDRESS; 2124 goto done; 2125 } 2126 vm_map_delete(map, *addr, *addr + length); 2127 } 2128 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 2129 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot, 2130 max, cow); 2131 } else { 2132 rv = vm_map_insert(map, object, offset, *addr, *addr + length, 2133 prot, max, cow); 2134 } 2135 if (rv == KERN_SUCCESS && update_anon) 2136 map->anon_loc = *addr + length; 2137 done: 2138 vm_map_unlock(map); 2139 return (rv); 2140 } 2141 2142 /* 2143 * vm_map_find_min() is a variant of vm_map_find() that takes an 2144 * additional parameter (min_addr) and treats the given address 2145 * (*addr) differently. Specifically, it treats *addr as a hint 2146 * and not as the minimum address where the mapping is created. 2147 * 2148 * This function works in two phases. First, it tries to 2149 * allocate above the hint. If that fails and the hint is 2150 * greater than min_addr, it performs a second pass, replacing 2151 * the hint with min_addr as the minimum address for the 2152 * allocation. 2153 */ 2154 int 2155 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 2156 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr, 2157 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max, 2158 int cow) 2159 { 2160 vm_offset_t hint; 2161 int rv; 2162 2163 hint = *addr; 2164 for (;;) { 2165 rv = vm_map_find(map, object, offset, addr, length, max_addr, 2166 find_space, prot, max, cow); 2167 if (rv == KERN_SUCCESS || min_addr >= hint) 2168 return (rv); 2169 *addr = hint = min_addr; 2170 } 2171 } 2172 2173 /* 2174 * A map entry with any of the following flags set must not be merged with 2175 * another entry. 2176 */ 2177 #define MAP_ENTRY_NOMERGE_MASK (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \ 2178 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC) 2179 2180 static bool 2181 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry) 2182 { 2183 2184 KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 || 2185 (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0, 2186 ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable", 2187 prev, entry)); 2188 return (prev->end == entry->start && 2189 prev->object.vm_object == entry->object.vm_object && 2190 (prev->object.vm_object == NULL || 2191 prev->offset + (prev->end - prev->start) == entry->offset) && 2192 prev->eflags == entry->eflags && 2193 prev->protection == entry->protection && 2194 prev->max_protection == entry->max_protection && 2195 prev->inheritance == entry->inheritance && 2196 prev->wired_count == entry->wired_count && 2197 prev->cred == entry->cred); 2198 } 2199 2200 static void 2201 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry) 2202 { 2203 2204 /* 2205 * If the backing object is a vnode object, vm_object_deallocate() 2206 * calls vrele(). However, vrele() does not lock the vnode because 2207 * the vnode has additional references. Thus, the map lock can be 2208 * kept without causing a lock-order reversal with the vnode lock. 2209 * 2210 * Since we count the number of virtual page mappings in 2211 * object->un_pager.vnp.writemappings, the writemappings value 2212 * should not be adjusted when the entry is disposed of. 2213 */ 2214 if (entry->object.vm_object != NULL) 2215 vm_object_deallocate(entry->object.vm_object); 2216 if (entry->cred != NULL) 2217 crfree(entry->cred); 2218 vm_map_entry_dispose(map, entry); 2219 } 2220 2221 /* 2222 * vm_map_try_merge_entries: 2223 * 2224 * Compare the given map entry to its predecessor, and merge its precessor 2225 * into it if possible. The entry remains valid, and may be extended. 2226 * The predecessor may be deleted. 2227 * 2228 * The map must be locked. 2229 */ 2230 void 2231 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev_entry, 2232 vm_map_entry_t entry) 2233 { 2234 2235 VM_MAP_ASSERT_LOCKED(map); 2236 if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 && 2237 vm_map_mergeable_neighbors(prev_entry, entry)) { 2238 vm_map_entry_unlink(map, prev_entry, UNLINK_MERGE_NEXT); 2239 vm_map_merged_neighbor_dispose(map, prev_entry); 2240 } 2241 } 2242 2243 /* 2244 * vm_map_entry_back: 2245 * 2246 * Allocate an object to back a map entry. 2247 */ 2248 static inline void 2249 vm_map_entry_back(vm_map_entry_t entry) 2250 { 2251 vm_object_t object; 2252 2253 KASSERT(entry->object.vm_object == NULL, 2254 ("map entry %p has backing object", entry)); 2255 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 2256 ("map entry %p is a submap", entry)); 2257 object = vm_object_allocate_anon(atop(entry->end - entry->start), NULL, 2258 entry->cred, entry->end - entry->start); 2259 entry->object.vm_object = object; 2260 entry->offset = 0; 2261 entry->cred = NULL; 2262 } 2263 2264 /* 2265 * vm_map_entry_charge_object 2266 * 2267 * If there is no object backing this entry, create one. Otherwise, if 2268 * the entry has cred, give it to the backing object. 2269 */ 2270 static inline void 2271 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry) 2272 { 2273 2274 VM_MAP_ASSERT_LOCKED(map); 2275 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 2276 ("map entry %p is a submap", entry)); 2277 if (entry->object.vm_object == NULL && !map->system_map && 2278 (entry->eflags & MAP_ENTRY_GUARD) == 0) 2279 vm_map_entry_back(entry); 2280 else if (entry->object.vm_object != NULL && 2281 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 2282 entry->cred != NULL) { 2283 VM_OBJECT_WLOCK(entry->object.vm_object); 2284 KASSERT(entry->object.vm_object->cred == NULL, 2285 ("OVERCOMMIT: %s: both cred e %p", __func__, entry)); 2286 entry->object.vm_object->cred = entry->cred; 2287 entry->object.vm_object->charge = entry->end - entry->start; 2288 VM_OBJECT_WUNLOCK(entry->object.vm_object); 2289 entry->cred = NULL; 2290 } 2291 } 2292 2293 /* 2294 * vm_map_clip_start: [ internal use only ] 2295 * 2296 * Asserts that the given entry begins at or after 2297 * the specified address; if necessary, 2298 * it splits the entry into two. 2299 */ 2300 #define vm_map_clip_start(map, entry, startaddr) \ 2301 { \ 2302 if (startaddr > entry->start) \ 2303 _vm_map_clip_start(map, entry, startaddr); \ 2304 } 2305 2306 /* 2307 * This routine is called only when it is known that 2308 * the entry must be split. 2309 */ 2310 static void 2311 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 2312 { 2313 vm_map_entry_t new_entry; 2314 2315 VM_MAP_ASSERT_LOCKED(map); 2316 KASSERT(entry->end > start && entry->start < start, 2317 ("_vm_map_clip_start: invalid clip of entry %p", entry)); 2318 2319 /* 2320 * Create a backing object now, if none exists, so that more individual 2321 * objects won't be created after the map entry is split. 2322 */ 2323 vm_map_entry_charge_object(map, entry); 2324 2325 /* Clone the entry. */ 2326 new_entry = vm_map_entry_create(map); 2327 *new_entry = *entry; 2328 2329 /* 2330 * Split off the front portion. Insert the new entry BEFORE this one, 2331 * so that this entry has the specified starting address. 2332 */ 2333 new_entry->end = start; 2334 entry->offset += (start - entry->start); 2335 entry->start = start; 2336 if (new_entry->cred != NULL) 2337 crhold(entry->cred); 2338 2339 vm_map_entry_link(map, new_entry); 2340 2341 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 2342 vm_object_reference(new_entry->object.vm_object); 2343 vm_map_entry_set_vnode_text(new_entry, true); 2344 /* 2345 * The object->un_pager.vnp.writemappings for the 2346 * object of MAP_ENTRY_WRITECNT type entry shall be 2347 * kept as is here. The virtual pages are 2348 * re-distributed among the clipped entries, so the sum is 2349 * left the same. 2350 */ 2351 } 2352 } 2353 2354 /* 2355 * vm_map_clip_end: [ internal use only ] 2356 * 2357 * Asserts that the given entry ends at or before 2358 * the specified address; if necessary, 2359 * it splits the entry into two. 2360 */ 2361 #define vm_map_clip_end(map, entry, endaddr) \ 2362 { \ 2363 if ((endaddr) < (entry->end)) \ 2364 _vm_map_clip_end((map), (entry), (endaddr)); \ 2365 } 2366 2367 /* 2368 * This routine is called only when it is known that 2369 * the entry must be split. 2370 */ 2371 static void 2372 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 2373 { 2374 vm_map_entry_t new_entry; 2375 2376 VM_MAP_ASSERT_LOCKED(map); 2377 KASSERT(entry->start < end && entry->end > end, 2378 ("_vm_map_clip_end: invalid clip of entry %p", entry)); 2379 2380 /* 2381 * Create a backing object now, if none exists, so that more individual 2382 * objects won't be created after the map entry is split. 2383 */ 2384 vm_map_entry_charge_object(map, entry); 2385 2386 /* Clone the entry. */ 2387 new_entry = vm_map_entry_create(map); 2388 *new_entry = *entry; 2389 2390 /* 2391 * Split off the back portion. Insert the new entry AFTER this one, 2392 * so that this entry has the specified ending address. 2393 */ 2394 new_entry->start = entry->end = end; 2395 new_entry->offset += (end - entry->start); 2396 if (new_entry->cred != NULL) 2397 crhold(entry->cred); 2398 2399 vm_map_entry_link(map, new_entry); 2400 2401 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 2402 vm_object_reference(new_entry->object.vm_object); 2403 vm_map_entry_set_vnode_text(new_entry, true); 2404 } 2405 } 2406 2407 /* 2408 * vm_map_submap: [ kernel use only ] 2409 * 2410 * Mark the given range as handled by a subordinate map. 2411 * 2412 * This range must have been created with vm_map_find, 2413 * and no other operations may have been performed on this 2414 * range prior to calling vm_map_submap. 2415 * 2416 * Only a limited number of operations can be performed 2417 * within this rage after calling vm_map_submap: 2418 * vm_fault 2419 * [Don't try vm_map_copy!] 2420 * 2421 * To remove a submapping, one must first remove the 2422 * range from the superior map, and then destroy the 2423 * submap (if desired). [Better yet, don't try it.] 2424 */ 2425 int 2426 vm_map_submap( 2427 vm_map_t map, 2428 vm_offset_t start, 2429 vm_offset_t end, 2430 vm_map_t submap) 2431 { 2432 vm_map_entry_t entry; 2433 int result; 2434 2435 result = KERN_INVALID_ARGUMENT; 2436 2437 vm_map_lock(submap); 2438 submap->flags |= MAP_IS_SUB_MAP; 2439 vm_map_unlock(submap); 2440 2441 vm_map_lock(map); 2442 2443 VM_MAP_RANGE_CHECK(map, start, end); 2444 2445 if (vm_map_lookup_entry(map, start, &entry)) { 2446 vm_map_clip_start(map, entry, start); 2447 } else 2448 entry = vm_map_entry_succ(entry); 2449 2450 vm_map_clip_end(map, entry, end); 2451 2452 if ((entry->start == start) && (entry->end == end) && 2453 ((entry->eflags & MAP_ENTRY_COW) == 0) && 2454 (entry->object.vm_object == NULL)) { 2455 entry->object.sub_map = submap; 2456 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 2457 result = KERN_SUCCESS; 2458 } 2459 vm_map_unlock(map); 2460 2461 if (result != KERN_SUCCESS) { 2462 vm_map_lock(submap); 2463 submap->flags &= ~MAP_IS_SUB_MAP; 2464 vm_map_unlock(submap); 2465 } 2466 return (result); 2467 } 2468 2469 /* 2470 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified 2471 */ 2472 #define MAX_INIT_PT 96 2473 2474 /* 2475 * vm_map_pmap_enter: 2476 * 2477 * Preload the specified map's pmap with mappings to the specified 2478 * object's memory-resident pages. No further physical pages are 2479 * allocated, and no further virtual pages are retrieved from secondary 2480 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 2481 * limited number of page mappings are created at the low-end of the 2482 * specified address range. (For this purpose, a superpage mapping 2483 * counts as one page mapping.) Otherwise, all resident pages within 2484 * the specified address range are mapped. 2485 */ 2486 static void 2487 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 2488 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 2489 { 2490 vm_offset_t start; 2491 vm_page_t p, p_start; 2492 vm_pindex_t mask, psize, threshold, tmpidx; 2493 2494 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 2495 return; 2496 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 2497 VM_OBJECT_WLOCK(object); 2498 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 2499 pmap_object_init_pt(map->pmap, addr, object, pindex, 2500 size); 2501 VM_OBJECT_WUNLOCK(object); 2502 return; 2503 } 2504 VM_OBJECT_LOCK_DOWNGRADE(object); 2505 } else 2506 VM_OBJECT_RLOCK(object); 2507 2508 psize = atop(size); 2509 if (psize + pindex > object->size) { 2510 if (pindex >= object->size) { 2511 VM_OBJECT_RUNLOCK(object); 2512 return; 2513 } 2514 psize = object->size - pindex; 2515 } 2516 2517 start = 0; 2518 p_start = NULL; 2519 threshold = MAX_INIT_PT; 2520 2521 p = vm_page_find_least(object, pindex); 2522 /* 2523 * Assert: the variable p is either (1) the page with the 2524 * least pindex greater than or equal to the parameter pindex 2525 * or (2) NULL. 2526 */ 2527 for (; 2528 p != NULL && (tmpidx = p->pindex - pindex) < psize; 2529 p = TAILQ_NEXT(p, listq)) { 2530 /* 2531 * don't allow an madvise to blow away our really 2532 * free pages allocating pv entries. 2533 */ 2534 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 2535 vm_page_count_severe()) || 2536 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 2537 tmpidx >= threshold)) { 2538 psize = tmpidx; 2539 break; 2540 } 2541 if (vm_page_all_valid(p)) { 2542 if (p_start == NULL) { 2543 start = addr + ptoa(tmpidx); 2544 p_start = p; 2545 } 2546 /* Jump ahead if a superpage mapping is possible. */ 2547 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 2548 (pagesizes[p->psind] - 1)) == 0) { 2549 mask = atop(pagesizes[p->psind]) - 1; 2550 if (tmpidx + mask < psize && 2551 vm_page_ps_test(p, PS_ALL_VALID, NULL)) { 2552 p += mask; 2553 threshold += mask; 2554 } 2555 } 2556 } else if (p_start != NULL) { 2557 pmap_enter_object(map->pmap, start, addr + 2558 ptoa(tmpidx), p_start, prot); 2559 p_start = NULL; 2560 } 2561 } 2562 if (p_start != NULL) 2563 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 2564 p_start, prot); 2565 VM_OBJECT_RUNLOCK(object); 2566 } 2567 2568 /* 2569 * vm_map_protect: 2570 * 2571 * Sets the protection of the specified address 2572 * region in the target map. If "set_max" is 2573 * specified, the maximum protection is to be set; 2574 * otherwise, only the current protection is affected. 2575 */ 2576 int 2577 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 2578 vm_prot_t new_prot, boolean_t set_max) 2579 { 2580 vm_map_entry_t entry, first_entry, in_tran, prev_entry; 2581 vm_object_t obj; 2582 struct ucred *cred; 2583 vm_prot_t old_prot; 2584 int rv; 2585 2586 if (start == end) 2587 return (KERN_SUCCESS); 2588 2589 again: 2590 in_tran = NULL; 2591 vm_map_lock(map); 2592 2593 /* 2594 * Ensure that we are not concurrently wiring pages. vm_map_wire() may 2595 * need to fault pages into the map and will drop the map lock while 2596 * doing so, and the VM object may end up in an inconsistent state if we 2597 * update the protection on the map entry in between faults. 2598 */ 2599 vm_map_wait_busy(map); 2600 2601 VM_MAP_RANGE_CHECK(map, start, end); 2602 2603 if (!vm_map_lookup_entry(map, start, &first_entry)) 2604 first_entry = vm_map_entry_succ(first_entry); 2605 2606 /* 2607 * Make a first pass to check for protection violations. 2608 */ 2609 for (entry = first_entry; entry->start < end; 2610 entry = vm_map_entry_succ(entry)) { 2611 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) 2612 continue; 2613 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) { 2614 vm_map_unlock(map); 2615 return (KERN_INVALID_ARGUMENT); 2616 } 2617 if ((new_prot & entry->max_protection) != new_prot) { 2618 vm_map_unlock(map); 2619 return (KERN_PROTECTION_FAILURE); 2620 } 2621 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0) 2622 in_tran = entry; 2623 } 2624 2625 /* 2626 * Postpone the operation until all in-transition map entries have 2627 * stabilized. An in-transition entry might already have its pages 2628 * wired and wired_count incremented, but not yet have its 2629 * MAP_ENTRY_USER_WIRED flag set. In which case, we would fail to call 2630 * vm_fault_copy_entry() in the final loop below. 2631 */ 2632 if (in_tran != NULL) { 2633 in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2634 vm_map_unlock_and_wait(map, 0); 2635 goto again; 2636 } 2637 2638 /* 2639 * Before changing the protections, try to reserve swap space for any 2640 * private (i.e., copy-on-write) mappings that are transitioning from 2641 * read-only to read/write access. If a reservation fails, break out 2642 * of this loop early and let the next loop simplify the entries, since 2643 * some may now be mergeable. 2644 */ 2645 rv = KERN_SUCCESS; 2646 vm_map_clip_start(map, first_entry, start); 2647 for (entry = first_entry; entry->start < end; 2648 entry = vm_map_entry_succ(entry)) { 2649 vm_map_clip_end(map, entry, end); 2650 2651 if (set_max || 2652 ((new_prot & ~entry->protection) & VM_PROT_WRITE) == 0 || 2653 ENTRY_CHARGED(entry) || 2654 (entry->eflags & MAP_ENTRY_GUARD) != 0) { 2655 continue; 2656 } 2657 2658 cred = curthread->td_ucred; 2659 obj = entry->object.vm_object; 2660 2661 if (obj == NULL || 2662 (entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0) { 2663 if (!swap_reserve(entry->end - entry->start)) { 2664 rv = KERN_RESOURCE_SHORTAGE; 2665 end = entry->end; 2666 break; 2667 } 2668 crhold(cred); 2669 entry->cred = cred; 2670 continue; 2671 } 2672 2673 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) 2674 continue; 2675 VM_OBJECT_WLOCK(obj); 2676 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 2677 VM_OBJECT_WUNLOCK(obj); 2678 continue; 2679 } 2680 2681 /* 2682 * Charge for the whole object allocation now, since 2683 * we cannot distinguish between non-charged and 2684 * charged clipped mapping of the same object later. 2685 */ 2686 KASSERT(obj->charge == 0, 2687 ("vm_map_protect: object %p overcharged (entry %p)", 2688 obj, entry)); 2689 if (!swap_reserve(ptoa(obj->size))) { 2690 VM_OBJECT_WUNLOCK(obj); 2691 rv = KERN_RESOURCE_SHORTAGE; 2692 end = entry->end; 2693 break; 2694 } 2695 2696 crhold(cred); 2697 obj->cred = cred; 2698 obj->charge = ptoa(obj->size); 2699 VM_OBJECT_WUNLOCK(obj); 2700 } 2701 2702 /* 2703 * If enough swap space was available, go back and fix up protections. 2704 * Otherwise, just simplify entries, since some may have been modified. 2705 * [Note that clipping is not necessary the second time.] 2706 */ 2707 for (prev_entry = vm_map_entry_pred(first_entry), entry = first_entry; 2708 entry->start < end; 2709 vm_map_try_merge_entries(map, prev_entry, entry), 2710 prev_entry = entry, entry = vm_map_entry_succ(entry)) { 2711 if (rv != KERN_SUCCESS || 2712 (entry->eflags & MAP_ENTRY_GUARD) != 0) 2713 continue; 2714 2715 old_prot = entry->protection; 2716 2717 if (set_max) 2718 entry->protection = 2719 (entry->max_protection = new_prot) & 2720 old_prot; 2721 else 2722 entry->protection = new_prot; 2723 2724 /* 2725 * For user wired map entries, the normal lazy evaluation of 2726 * write access upgrades through soft page faults is 2727 * undesirable. Instead, immediately copy any pages that are 2728 * copy-on-write and enable write access in the physical map. 2729 */ 2730 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2731 (entry->protection & VM_PROT_WRITE) != 0 && 2732 (old_prot & VM_PROT_WRITE) == 0) 2733 vm_fault_copy_entry(map, map, entry, entry, NULL); 2734 2735 /* 2736 * When restricting access, update the physical map. Worry 2737 * about copy-on-write here. 2738 */ 2739 if ((old_prot & ~entry->protection) != 0) { 2740 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2741 VM_PROT_ALL) 2742 pmap_protect(map->pmap, entry->start, 2743 entry->end, 2744 entry->protection & MASK(entry)); 2745 #undef MASK 2746 } 2747 } 2748 vm_map_try_merge_entries(map, prev_entry, entry); 2749 vm_map_unlock(map); 2750 return (rv); 2751 } 2752 2753 /* 2754 * vm_map_madvise: 2755 * 2756 * This routine traverses a processes map handling the madvise 2757 * system call. Advisories are classified as either those effecting 2758 * the vm_map_entry structure, or those effecting the underlying 2759 * objects. 2760 */ 2761 int 2762 vm_map_madvise( 2763 vm_map_t map, 2764 vm_offset_t start, 2765 vm_offset_t end, 2766 int behav) 2767 { 2768 vm_map_entry_t entry, prev_entry; 2769 bool modify_map; 2770 2771 /* 2772 * Some madvise calls directly modify the vm_map_entry, in which case 2773 * we need to use an exclusive lock on the map and we need to perform 2774 * various clipping operations. Otherwise we only need a read-lock 2775 * on the map. 2776 */ 2777 switch(behav) { 2778 case MADV_NORMAL: 2779 case MADV_SEQUENTIAL: 2780 case MADV_RANDOM: 2781 case MADV_NOSYNC: 2782 case MADV_AUTOSYNC: 2783 case MADV_NOCORE: 2784 case MADV_CORE: 2785 if (start == end) 2786 return (0); 2787 modify_map = true; 2788 vm_map_lock(map); 2789 break; 2790 case MADV_WILLNEED: 2791 case MADV_DONTNEED: 2792 case MADV_FREE: 2793 if (start == end) 2794 return (0); 2795 modify_map = false; 2796 vm_map_lock_read(map); 2797 break; 2798 default: 2799 return (EINVAL); 2800 } 2801 2802 /* 2803 * Locate starting entry and clip if necessary. 2804 */ 2805 VM_MAP_RANGE_CHECK(map, start, end); 2806 2807 if (vm_map_lookup_entry(map, start, &entry)) { 2808 if (modify_map) 2809 vm_map_clip_start(map, entry, start); 2810 prev_entry = vm_map_entry_pred(entry); 2811 } else { 2812 prev_entry = entry; 2813 entry = vm_map_entry_succ(entry); 2814 } 2815 2816 if (modify_map) { 2817 /* 2818 * madvise behaviors that are implemented in the vm_map_entry. 2819 * 2820 * We clip the vm_map_entry so that behavioral changes are 2821 * limited to the specified address range. 2822 */ 2823 for (; entry->start < end; 2824 prev_entry = entry, entry = vm_map_entry_succ(entry)) { 2825 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 2826 continue; 2827 2828 vm_map_clip_end(map, entry, end); 2829 2830 switch (behav) { 2831 case MADV_NORMAL: 2832 vm_map_entry_set_behavior(entry, 2833 MAP_ENTRY_BEHAV_NORMAL); 2834 break; 2835 case MADV_SEQUENTIAL: 2836 vm_map_entry_set_behavior(entry, 2837 MAP_ENTRY_BEHAV_SEQUENTIAL); 2838 break; 2839 case MADV_RANDOM: 2840 vm_map_entry_set_behavior(entry, 2841 MAP_ENTRY_BEHAV_RANDOM); 2842 break; 2843 case MADV_NOSYNC: 2844 entry->eflags |= MAP_ENTRY_NOSYNC; 2845 break; 2846 case MADV_AUTOSYNC: 2847 entry->eflags &= ~MAP_ENTRY_NOSYNC; 2848 break; 2849 case MADV_NOCORE: 2850 entry->eflags |= MAP_ENTRY_NOCOREDUMP; 2851 break; 2852 case MADV_CORE: 2853 entry->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2854 break; 2855 default: 2856 break; 2857 } 2858 vm_map_try_merge_entries(map, prev_entry, entry); 2859 } 2860 vm_map_try_merge_entries(map, prev_entry, entry); 2861 vm_map_unlock(map); 2862 } else { 2863 vm_pindex_t pstart, pend; 2864 2865 /* 2866 * madvise behaviors that are implemented in the underlying 2867 * vm_object. 2868 * 2869 * Since we don't clip the vm_map_entry, we have to clip 2870 * the vm_object pindex and count. 2871 */ 2872 for (; entry->start < end; 2873 entry = vm_map_entry_succ(entry)) { 2874 vm_offset_t useEnd, useStart; 2875 2876 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 2877 continue; 2878 2879 /* 2880 * MADV_FREE would otherwise rewind time to 2881 * the creation of the shadow object. Because 2882 * we hold the VM map read-locked, neither the 2883 * entry's object nor the presence of a 2884 * backing object can change. 2885 */ 2886 if (behav == MADV_FREE && 2887 entry->object.vm_object != NULL && 2888 entry->object.vm_object->backing_object != NULL) 2889 continue; 2890 2891 pstart = OFF_TO_IDX(entry->offset); 2892 pend = pstart + atop(entry->end - entry->start); 2893 useStart = entry->start; 2894 useEnd = entry->end; 2895 2896 if (entry->start < start) { 2897 pstart += atop(start - entry->start); 2898 useStart = start; 2899 } 2900 if (entry->end > end) { 2901 pend -= atop(entry->end - end); 2902 useEnd = end; 2903 } 2904 2905 if (pstart >= pend) 2906 continue; 2907 2908 /* 2909 * Perform the pmap_advise() before clearing 2910 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2911 * concurrent pmap operation, such as pmap_remove(), 2912 * could clear a reference in the pmap and set 2913 * PGA_REFERENCED on the page before the pmap_advise() 2914 * had completed. Consequently, the page would appear 2915 * referenced based upon an old reference that 2916 * occurred before this pmap_advise() ran. 2917 */ 2918 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2919 pmap_advise(map->pmap, useStart, useEnd, 2920 behav); 2921 2922 vm_object_madvise(entry->object.vm_object, pstart, 2923 pend, behav); 2924 2925 /* 2926 * Pre-populate paging structures in the 2927 * WILLNEED case. For wired entries, the 2928 * paging structures are already populated. 2929 */ 2930 if (behav == MADV_WILLNEED && 2931 entry->wired_count == 0) { 2932 vm_map_pmap_enter(map, 2933 useStart, 2934 entry->protection, 2935 entry->object.vm_object, 2936 pstart, 2937 ptoa(pend - pstart), 2938 MAP_PREFAULT_MADVISE 2939 ); 2940 } 2941 } 2942 vm_map_unlock_read(map); 2943 } 2944 return (0); 2945 } 2946 2947 2948 /* 2949 * vm_map_inherit: 2950 * 2951 * Sets the inheritance of the specified address 2952 * range in the target map. Inheritance 2953 * affects how the map will be shared with 2954 * child maps at the time of vmspace_fork. 2955 */ 2956 int 2957 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2958 vm_inherit_t new_inheritance) 2959 { 2960 vm_map_entry_t entry, prev_entry; 2961 2962 switch (new_inheritance) { 2963 case VM_INHERIT_NONE: 2964 case VM_INHERIT_COPY: 2965 case VM_INHERIT_SHARE: 2966 case VM_INHERIT_ZERO: 2967 break; 2968 default: 2969 return (KERN_INVALID_ARGUMENT); 2970 } 2971 if (start == end) 2972 return (KERN_SUCCESS); 2973 vm_map_lock(map); 2974 VM_MAP_RANGE_CHECK(map, start, end); 2975 if (vm_map_lookup_entry(map, start, &prev_entry)) { 2976 entry = prev_entry; 2977 vm_map_clip_start(map, entry, start); 2978 prev_entry = vm_map_entry_pred(entry); 2979 } else 2980 entry = vm_map_entry_succ(prev_entry); 2981 for (; entry->start < end; 2982 prev_entry = entry, entry = vm_map_entry_succ(entry)) { 2983 vm_map_clip_end(map, entry, end); 2984 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 || 2985 new_inheritance != VM_INHERIT_ZERO) 2986 entry->inheritance = new_inheritance; 2987 vm_map_try_merge_entries(map, prev_entry, entry); 2988 } 2989 vm_map_try_merge_entries(map, prev_entry, entry); 2990 vm_map_unlock(map); 2991 return (KERN_SUCCESS); 2992 } 2993 2994 /* 2995 * vm_map_entry_in_transition: 2996 * 2997 * Release the map lock, and sleep until the entry is no longer in 2998 * transition. Awake and acquire the map lock. If the map changed while 2999 * another held the lock, lookup a possibly-changed entry at or after the 3000 * 'start' position of the old entry. 3001 */ 3002 static vm_map_entry_t 3003 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start, 3004 vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry) 3005 { 3006 vm_map_entry_t entry; 3007 vm_offset_t start; 3008 u_int last_timestamp; 3009 3010 VM_MAP_ASSERT_LOCKED(map); 3011 KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 3012 ("not in-tranition map entry %p", in_entry)); 3013 /* 3014 * We have not yet clipped the entry. 3015 */ 3016 start = MAX(in_start, in_entry->start); 3017 in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 3018 last_timestamp = map->timestamp; 3019 if (vm_map_unlock_and_wait(map, 0)) { 3020 /* 3021 * Allow interruption of user wiring/unwiring? 3022 */ 3023 } 3024 vm_map_lock(map); 3025 if (last_timestamp + 1 == map->timestamp) 3026 return (in_entry); 3027 3028 /* 3029 * Look again for the entry because the map was modified while it was 3030 * unlocked. Specifically, the entry may have been clipped, merged, or 3031 * deleted. 3032 */ 3033 if (!vm_map_lookup_entry(map, start, &entry)) { 3034 if (!holes_ok) { 3035 *io_end = start; 3036 return (NULL); 3037 } 3038 entry = vm_map_entry_succ(entry); 3039 } 3040 return (entry); 3041 } 3042 3043 /* 3044 * vm_map_unwire: 3045 * 3046 * Implements both kernel and user unwiring. 3047 */ 3048 int 3049 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 3050 int flags) 3051 { 3052 vm_map_entry_t entry, first_entry, next_entry, prev_entry; 3053 int rv; 3054 bool holes_ok, need_wakeup, user_unwire; 3055 3056 if (start == end) 3057 return (KERN_SUCCESS); 3058 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0; 3059 user_unwire = (flags & VM_MAP_WIRE_USER) != 0; 3060 vm_map_lock(map); 3061 VM_MAP_RANGE_CHECK(map, start, end); 3062 if (!vm_map_lookup_entry(map, start, &first_entry)) { 3063 if (holes_ok) 3064 first_entry = vm_map_entry_succ(first_entry); 3065 else { 3066 vm_map_unlock(map); 3067 return (KERN_INVALID_ADDRESS); 3068 } 3069 } 3070 rv = KERN_SUCCESS; 3071 for (entry = first_entry; entry->start < end; entry = next_entry) { 3072 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 3073 /* 3074 * We have not yet clipped the entry. 3075 */ 3076 next_entry = vm_map_entry_in_transition(map, start, 3077 &end, holes_ok, entry); 3078 if (next_entry == NULL) { 3079 if (entry == first_entry) { 3080 vm_map_unlock(map); 3081 return (KERN_INVALID_ADDRESS); 3082 } 3083 rv = KERN_INVALID_ADDRESS; 3084 break; 3085 } 3086 first_entry = (entry == first_entry) ? 3087 next_entry : NULL; 3088 continue; 3089 } 3090 vm_map_clip_start(map, entry, start); 3091 vm_map_clip_end(map, entry, end); 3092 /* 3093 * Mark the entry in case the map lock is released. (See 3094 * above.) 3095 */ 3096 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 3097 entry->wiring_thread == NULL, 3098 ("owned map entry %p", entry)); 3099 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 3100 entry->wiring_thread = curthread; 3101 next_entry = vm_map_entry_succ(entry); 3102 /* 3103 * Check the map for holes in the specified region. 3104 * If holes_ok, skip this check. 3105 */ 3106 if (!holes_ok && 3107 entry->end < end && next_entry->start > entry->end) { 3108 end = entry->end; 3109 rv = KERN_INVALID_ADDRESS; 3110 break; 3111 } 3112 /* 3113 * If system unwiring, require that the entry is system wired. 3114 */ 3115 if (!user_unwire && 3116 vm_map_entry_system_wired_count(entry) == 0) { 3117 end = entry->end; 3118 rv = KERN_INVALID_ARGUMENT; 3119 break; 3120 } 3121 } 3122 need_wakeup = false; 3123 if (first_entry == NULL && 3124 !vm_map_lookup_entry(map, start, &first_entry)) { 3125 KASSERT(holes_ok, ("vm_map_unwire: lookup failed")); 3126 prev_entry = first_entry; 3127 entry = vm_map_entry_succ(first_entry); 3128 } else { 3129 prev_entry = vm_map_entry_pred(first_entry); 3130 entry = first_entry; 3131 } 3132 for (; entry->start < end; 3133 prev_entry = entry, entry = vm_map_entry_succ(entry)) { 3134 /* 3135 * If holes_ok was specified, an empty 3136 * space in the unwired region could have been mapped 3137 * while the map lock was dropped for draining 3138 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 3139 * could be simultaneously wiring this new mapping 3140 * entry. Detect these cases and skip any entries 3141 * marked as in transition by us. 3142 */ 3143 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 3144 entry->wiring_thread != curthread) { 3145 KASSERT(holes_ok, 3146 ("vm_map_unwire: !HOLESOK and new/changed entry")); 3147 continue; 3148 } 3149 3150 if (rv == KERN_SUCCESS && (!user_unwire || 3151 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 3152 if (entry->wired_count == 1) 3153 vm_map_entry_unwire(map, entry); 3154 else 3155 entry->wired_count--; 3156 if (user_unwire) 3157 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 3158 } 3159 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 3160 ("vm_map_unwire: in-transition flag missing %p", entry)); 3161 KASSERT(entry->wiring_thread == curthread, 3162 ("vm_map_unwire: alien wire %p", entry)); 3163 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 3164 entry->wiring_thread = NULL; 3165 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 3166 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 3167 need_wakeup = true; 3168 } 3169 vm_map_try_merge_entries(map, prev_entry, entry); 3170 } 3171 vm_map_try_merge_entries(map, prev_entry, entry); 3172 vm_map_unlock(map); 3173 if (need_wakeup) 3174 vm_map_wakeup(map); 3175 return (rv); 3176 } 3177 3178 static void 3179 vm_map_wire_user_count_sub(u_long npages) 3180 { 3181 3182 atomic_subtract_long(&vm_user_wire_count, npages); 3183 } 3184 3185 static bool 3186 vm_map_wire_user_count_add(u_long npages) 3187 { 3188 u_long wired; 3189 3190 wired = vm_user_wire_count; 3191 do { 3192 if (npages + wired > vm_page_max_user_wired) 3193 return (false); 3194 } while (!atomic_fcmpset_long(&vm_user_wire_count, &wired, 3195 npages + wired)); 3196 3197 return (true); 3198 } 3199 3200 /* 3201 * vm_map_wire_entry_failure: 3202 * 3203 * Handle a wiring failure on the given entry. 3204 * 3205 * The map should be locked. 3206 */ 3207 static void 3208 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 3209 vm_offset_t failed_addr) 3210 { 3211 3212 VM_MAP_ASSERT_LOCKED(map); 3213 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && 3214 entry->wired_count == 1, 3215 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); 3216 KASSERT(failed_addr < entry->end, 3217 ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); 3218 3219 /* 3220 * If any pages at the start of this entry were successfully wired, 3221 * then unwire them. 3222 */ 3223 if (failed_addr > entry->start) { 3224 pmap_unwire(map->pmap, entry->start, failed_addr); 3225 vm_object_unwire(entry->object.vm_object, entry->offset, 3226 failed_addr - entry->start, PQ_ACTIVE); 3227 } 3228 3229 /* 3230 * Assign an out-of-range value to represent the failure to wire this 3231 * entry. 3232 */ 3233 entry->wired_count = -1; 3234 } 3235 3236 int 3237 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags) 3238 { 3239 int rv; 3240 3241 vm_map_lock(map); 3242 rv = vm_map_wire_locked(map, start, end, flags); 3243 vm_map_unlock(map); 3244 return (rv); 3245 } 3246 3247 3248 /* 3249 * vm_map_wire_locked: 3250 * 3251 * Implements both kernel and user wiring. Returns with the map locked, 3252 * the map lock may be dropped. 3253 */ 3254 int 3255 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags) 3256 { 3257 vm_map_entry_t entry, first_entry, next_entry, prev_entry; 3258 vm_offset_t faddr, saved_end, saved_start; 3259 u_long npages; 3260 u_int last_timestamp; 3261 int rv; 3262 bool holes_ok, need_wakeup, user_wire; 3263 vm_prot_t prot; 3264 3265 VM_MAP_ASSERT_LOCKED(map); 3266 3267 if (start == end) 3268 return (KERN_SUCCESS); 3269 prot = 0; 3270 if (flags & VM_MAP_WIRE_WRITE) 3271 prot |= VM_PROT_WRITE; 3272 holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0; 3273 user_wire = (flags & VM_MAP_WIRE_USER) != 0; 3274 VM_MAP_RANGE_CHECK(map, start, end); 3275 if (!vm_map_lookup_entry(map, start, &first_entry)) { 3276 if (holes_ok) 3277 first_entry = vm_map_entry_succ(first_entry); 3278 else 3279 return (KERN_INVALID_ADDRESS); 3280 } 3281 for (entry = first_entry; entry->start < end; entry = next_entry) { 3282 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 3283 /* 3284 * We have not yet clipped the entry. 3285 */ 3286 next_entry = vm_map_entry_in_transition(map, start, 3287 &end, holes_ok, entry); 3288 if (next_entry == NULL) { 3289 if (entry == first_entry) 3290 return (KERN_INVALID_ADDRESS); 3291 rv = KERN_INVALID_ADDRESS; 3292 goto done; 3293 } 3294 first_entry = (entry == first_entry) ? 3295 next_entry : NULL; 3296 continue; 3297 } 3298 vm_map_clip_start(map, entry, start); 3299 vm_map_clip_end(map, entry, end); 3300 /* 3301 * Mark the entry in case the map lock is released. (See 3302 * above.) 3303 */ 3304 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 3305 entry->wiring_thread == NULL, 3306 ("owned map entry %p", entry)); 3307 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 3308 entry->wiring_thread = curthread; 3309 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 3310 || (entry->protection & prot) != prot) { 3311 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 3312 if (!holes_ok) { 3313 end = entry->end; 3314 rv = KERN_INVALID_ADDRESS; 3315 goto done; 3316 } 3317 } else if (entry->wired_count == 0) { 3318 entry->wired_count++; 3319 3320 npages = atop(entry->end - entry->start); 3321 if (user_wire && !vm_map_wire_user_count_add(npages)) { 3322 vm_map_wire_entry_failure(map, entry, 3323 entry->start); 3324 end = entry->end; 3325 rv = KERN_RESOURCE_SHORTAGE; 3326 goto done; 3327 } 3328 3329 /* 3330 * Release the map lock, relying on the in-transition 3331 * mark. Mark the map busy for fork. 3332 */ 3333 saved_start = entry->start; 3334 saved_end = entry->end; 3335 last_timestamp = map->timestamp; 3336 vm_map_busy(map); 3337 vm_map_unlock(map); 3338 3339 faddr = saved_start; 3340 do { 3341 /* 3342 * Simulate a fault to get the page and enter 3343 * it into the physical map. 3344 */ 3345 if ((rv = vm_fault(map, faddr, 3346 VM_PROT_NONE, VM_FAULT_WIRE, NULL)) != 3347 KERN_SUCCESS) 3348 break; 3349 } while ((faddr += PAGE_SIZE) < saved_end); 3350 vm_map_lock(map); 3351 vm_map_unbusy(map); 3352 if (last_timestamp + 1 != map->timestamp) { 3353 /* 3354 * Look again for the entry because the map was 3355 * modified while it was unlocked. The entry 3356 * may have been clipped, but NOT merged or 3357 * deleted. 3358 */ 3359 if (!vm_map_lookup_entry(map, saved_start, 3360 &next_entry)) 3361 KASSERT(false, 3362 ("vm_map_wire: lookup failed")); 3363 first_entry = (entry == first_entry) ? 3364 next_entry : NULL; 3365 for (entry = next_entry; entry->end < saved_end; 3366 entry = vm_map_entry_succ(entry)) { 3367 /* 3368 * In case of failure, handle entries 3369 * that were not fully wired here; 3370 * fully wired entries are handled 3371 * later. 3372 */ 3373 if (rv != KERN_SUCCESS && 3374 faddr < entry->end) 3375 vm_map_wire_entry_failure(map, 3376 entry, faddr); 3377 } 3378 } 3379 if (rv != KERN_SUCCESS) { 3380 vm_map_wire_entry_failure(map, entry, faddr); 3381 if (user_wire) 3382 vm_map_wire_user_count_sub(npages); 3383 end = entry->end; 3384 goto done; 3385 } 3386 } else if (!user_wire || 3387 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 3388 entry->wired_count++; 3389 } 3390 /* 3391 * Check the map for holes in the specified region. 3392 * If holes_ok was specified, skip this check. 3393 */ 3394 next_entry = vm_map_entry_succ(entry); 3395 if (!holes_ok && 3396 entry->end < end && next_entry->start > entry->end) { 3397 end = entry->end; 3398 rv = KERN_INVALID_ADDRESS; 3399 goto done; 3400 } 3401 } 3402 rv = KERN_SUCCESS; 3403 done: 3404 need_wakeup = false; 3405 if (first_entry == NULL && 3406 !vm_map_lookup_entry(map, start, &first_entry)) { 3407 KASSERT(holes_ok, ("vm_map_wire: lookup failed")); 3408 prev_entry = first_entry; 3409 entry = vm_map_entry_succ(first_entry); 3410 } else { 3411 prev_entry = vm_map_entry_pred(first_entry); 3412 entry = first_entry; 3413 } 3414 for (; entry->start < end; 3415 prev_entry = entry, entry = vm_map_entry_succ(entry)) { 3416 /* 3417 * If holes_ok was specified, an empty 3418 * space in the unwired region could have been mapped 3419 * while the map lock was dropped for faulting in the 3420 * pages or draining MAP_ENTRY_IN_TRANSITION. 3421 * Moreover, another thread could be simultaneously 3422 * wiring this new mapping entry. Detect these cases 3423 * and skip any entries marked as in transition not by us. 3424 */ 3425 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 3426 entry->wiring_thread != curthread) { 3427 KASSERT(holes_ok, 3428 ("vm_map_wire: !HOLESOK and new/changed entry")); 3429 continue; 3430 } 3431 3432 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) { 3433 /* do nothing */ 3434 } else if (rv == KERN_SUCCESS) { 3435 if (user_wire) 3436 entry->eflags |= MAP_ENTRY_USER_WIRED; 3437 } else if (entry->wired_count == -1) { 3438 /* 3439 * Wiring failed on this entry. Thus, unwiring is 3440 * unnecessary. 3441 */ 3442 entry->wired_count = 0; 3443 } else if (!user_wire || 3444 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 3445 /* 3446 * Undo the wiring. Wiring succeeded on this entry 3447 * but failed on a later entry. 3448 */ 3449 if (entry->wired_count == 1) { 3450 vm_map_entry_unwire(map, entry); 3451 if (user_wire) 3452 vm_map_wire_user_count_sub( 3453 atop(entry->end - entry->start)); 3454 } else 3455 entry->wired_count--; 3456 } 3457 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 3458 ("vm_map_wire: in-transition flag missing %p", entry)); 3459 KASSERT(entry->wiring_thread == curthread, 3460 ("vm_map_wire: alien wire %p", entry)); 3461 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 3462 MAP_ENTRY_WIRE_SKIPPED); 3463 entry->wiring_thread = NULL; 3464 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 3465 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 3466 need_wakeup = true; 3467 } 3468 vm_map_try_merge_entries(map, prev_entry, entry); 3469 } 3470 vm_map_try_merge_entries(map, prev_entry, entry); 3471 if (need_wakeup) 3472 vm_map_wakeup(map); 3473 return (rv); 3474 } 3475 3476 /* 3477 * vm_map_sync 3478 * 3479 * Push any dirty cached pages in the address range to their pager. 3480 * If syncio is TRUE, dirty pages are written synchronously. 3481 * If invalidate is TRUE, any cached pages are freed as well. 3482 * 3483 * If the size of the region from start to end is zero, we are 3484 * supposed to flush all modified pages within the region containing 3485 * start. Unfortunately, a region can be split or coalesced with 3486 * neighboring regions, making it difficult to determine what the 3487 * original region was. Therefore, we approximate this requirement by 3488 * flushing the current region containing start. 3489 * 3490 * Returns an error if any part of the specified range is not mapped. 3491 */ 3492 int 3493 vm_map_sync( 3494 vm_map_t map, 3495 vm_offset_t start, 3496 vm_offset_t end, 3497 boolean_t syncio, 3498 boolean_t invalidate) 3499 { 3500 vm_map_entry_t entry, first_entry, next_entry; 3501 vm_size_t size; 3502 vm_object_t object; 3503 vm_ooffset_t offset; 3504 unsigned int last_timestamp; 3505 boolean_t failed; 3506 3507 vm_map_lock_read(map); 3508 VM_MAP_RANGE_CHECK(map, start, end); 3509 if (!vm_map_lookup_entry(map, start, &first_entry)) { 3510 vm_map_unlock_read(map); 3511 return (KERN_INVALID_ADDRESS); 3512 } else if (start == end) { 3513 start = first_entry->start; 3514 end = first_entry->end; 3515 } 3516 /* 3517 * Make a first pass to check for user-wired memory and holes. 3518 */ 3519 for (entry = first_entry; entry->start < end; entry = next_entry) { 3520 if (invalidate && 3521 (entry->eflags & MAP_ENTRY_USER_WIRED) != 0) { 3522 vm_map_unlock_read(map); 3523 return (KERN_INVALID_ARGUMENT); 3524 } 3525 next_entry = vm_map_entry_succ(entry); 3526 if (end > entry->end && 3527 entry->end != next_entry->start) { 3528 vm_map_unlock_read(map); 3529 return (KERN_INVALID_ADDRESS); 3530 } 3531 } 3532 3533 if (invalidate) 3534 pmap_remove(map->pmap, start, end); 3535 failed = FALSE; 3536 3537 /* 3538 * Make a second pass, cleaning/uncaching pages from the indicated 3539 * objects as we go. 3540 */ 3541 for (entry = first_entry; entry->start < end;) { 3542 offset = entry->offset + (start - entry->start); 3543 size = (end <= entry->end ? end : entry->end) - start; 3544 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) { 3545 vm_map_t smap; 3546 vm_map_entry_t tentry; 3547 vm_size_t tsize; 3548 3549 smap = entry->object.sub_map; 3550 vm_map_lock_read(smap); 3551 (void) vm_map_lookup_entry(smap, offset, &tentry); 3552 tsize = tentry->end - offset; 3553 if (tsize < size) 3554 size = tsize; 3555 object = tentry->object.vm_object; 3556 offset = tentry->offset + (offset - tentry->start); 3557 vm_map_unlock_read(smap); 3558 } else { 3559 object = entry->object.vm_object; 3560 } 3561 vm_object_reference(object); 3562 last_timestamp = map->timestamp; 3563 vm_map_unlock_read(map); 3564 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 3565 failed = TRUE; 3566 start += size; 3567 vm_object_deallocate(object); 3568 vm_map_lock_read(map); 3569 if (last_timestamp == map->timestamp || 3570 !vm_map_lookup_entry(map, start, &entry)) 3571 entry = vm_map_entry_succ(entry); 3572 } 3573 3574 vm_map_unlock_read(map); 3575 return (failed ? KERN_FAILURE : KERN_SUCCESS); 3576 } 3577 3578 /* 3579 * vm_map_entry_unwire: [ internal use only ] 3580 * 3581 * Make the region specified by this entry pageable. 3582 * 3583 * The map in question should be locked. 3584 * [This is the reason for this routine's existence.] 3585 */ 3586 static void 3587 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 3588 { 3589 vm_size_t size; 3590 3591 VM_MAP_ASSERT_LOCKED(map); 3592 KASSERT(entry->wired_count > 0, 3593 ("vm_map_entry_unwire: entry %p isn't wired", entry)); 3594 3595 size = entry->end - entry->start; 3596 if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0) 3597 vm_map_wire_user_count_sub(atop(size)); 3598 pmap_unwire(map->pmap, entry->start, entry->end); 3599 vm_object_unwire(entry->object.vm_object, entry->offset, size, 3600 PQ_ACTIVE); 3601 entry->wired_count = 0; 3602 } 3603 3604 static void 3605 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 3606 { 3607 3608 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 3609 vm_object_deallocate(entry->object.vm_object); 3610 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 3611 } 3612 3613 /* 3614 * vm_map_entry_delete: [ internal use only ] 3615 * 3616 * Deallocate the given entry from the target map. 3617 */ 3618 static void 3619 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 3620 { 3621 vm_object_t object; 3622 vm_pindex_t offidxstart, offidxend, count, size1; 3623 vm_size_t size; 3624 3625 vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE); 3626 object = entry->object.vm_object; 3627 3628 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) { 3629 MPASS(entry->cred == NULL); 3630 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0); 3631 MPASS(object == NULL); 3632 vm_map_entry_deallocate(entry, map->system_map); 3633 return; 3634 } 3635 3636 size = entry->end - entry->start; 3637 map->size -= size; 3638 3639 if (entry->cred != NULL) { 3640 swap_release_by_cred(size, entry->cred); 3641 crfree(entry->cred); 3642 } 3643 3644 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || object == NULL) { 3645 entry->object.vm_object = NULL; 3646 } else if ((object->flags & OBJ_ANON) != 0 || 3647 object == kernel_object) { 3648 KASSERT(entry->cred == NULL || object->cred == NULL || 3649 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 3650 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 3651 count = atop(size); 3652 offidxstart = OFF_TO_IDX(entry->offset); 3653 offidxend = offidxstart + count; 3654 VM_OBJECT_WLOCK(object); 3655 if (object->ref_count != 1 && 3656 ((object->flags & OBJ_ONEMAPPING) != 0 || 3657 object == kernel_object)) { 3658 vm_object_collapse(object); 3659 3660 /* 3661 * The option OBJPR_NOTMAPPED can be passed here 3662 * because vm_map_delete() already performed 3663 * pmap_remove() on the only mapping to this range 3664 * of pages. 3665 */ 3666 vm_object_page_remove(object, offidxstart, offidxend, 3667 OBJPR_NOTMAPPED); 3668 if (object->type == OBJT_SWAP) 3669 swap_pager_freespace(object, offidxstart, 3670 count); 3671 if (offidxend >= object->size && 3672 offidxstart < object->size) { 3673 size1 = object->size; 3674 object->size = offidxstart; 3675 if (object->cred != NULL) { 3676 size1 -= object->size; 3677 KASSERT(object->charge >= ptoa(size1), 3678 ("object %p charge < 0", object)); 3679 swap_release_by_cred(ptoa(size1), 3680 object->cred); 3681 object->charge -= ptoa(size1); 3682 } 3683 } 3684 } 3685 VM_OBJECT_WUNLOCK(object); 3686 } 3687 if (map->system_map) 3688 vm_map_entry_deallocate(entry, TRUE); 3689 else { 3690 entry->defer_next = curthread->td_map_def_user; 3691 curthread->td_map_def_user = entry; 3692 } 3693 } 3694 3695 /* 3696 * vm_map_delete: [ internal use only ] 3697 * 3698 * Deallocates the given address range from the target 3699 * map. 3700 */ 3701 int 3702 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 3703 { 3704 vm_map_entry_t entry; 3705 vm_map_entry_t first_entry; 3706 3707 VM_MAP_ASSERT_LOCKED(map); 3708 if (start == end) 3709 return (KERN_SUCCESS); 3710 3711 /* 3712 * Find the start of the region, and clip it 3713 */ 3714 if (!vm_map_lookup_entry(map, start, &first_entry)) 3715 entry = vm_map_entry_succ(first_entry); 3716 else { 3717 entry = first_entry; 3718 vm_map_clip_start(map, entry, start); 3719 } 3720 3721 /* 3722 * Step through all entries in this region 3723 */ 3724 while (entry->start < end) { 3725 vm_map_entry_t next; 3726 3727 /* 3728 * Wait for wiring or unwiring of an entry to complete. 3729 * Also wait for any system wirings to disappear on 3730 * user maps. 3731 */ 3732 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 3733 (vm_map_pmap(map) != kernel_pmap && 3734 vm_map_entry_system_wired_count(entry) != 0)) { 3735 unsigned int last_timestamp; 3736 vm_offset_t saved_start; 3737 vm_map_entry_t tmp_entry; 3738 3739 saved_start = entry->start; 3740 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 3741 last_timestamp = map->timestamp; 3742 (void) vm_map_unlock_and_wait(map, 0); 3743 vm_map_lock(map); 3744 if (last_timestamp + 1 != map->timestamp) { 3745 /* 3746 * Look again for the entry because the map was 3747 * modified while it was unlocked. 3748 * Specifically, the entry may have been 3749 * clipped, merged, or deleted. 3750 */ 3751 if (!vm_map_lookup_entry(map, saved_start, 3752 &tmp_entry)) 3753 entry = vm_map_entry_succ(tmp_entry); 3754 else { 3755 entry = tmp_entry; 3756 vm_map_clip_start(map, entry, 3757 saved_start); 3758 } 3759 } 3760 continue; 3761 } 3762 vm_map_clip_end(map, entry, end); 3763 3764 next = vm_map_entry_succ(entry); 3765 3766 /* 3767 * Unwire before removing addresses from the pmap; otherwise, 3768 * unwiring will put the entries back in the pmap. 3769 */ 3770 if (entry->wired_count != 0) 3771 vm_map_entry_unwire(map, entry); 3772 3773 /* 3774 * Remove mappings for the pages, but only if the 3775 * mappings could exist. For instance, it does not 3776 * make sense to call pmap_remove() for guard entries. 3777 */ 3778 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || 3779 entry->object.vm_object != NULL) 3780 pmap_remove(map->pmap, entry->start, entry->end); 3781 3782 if (entry->end == map->anon_loc) 3783 map->anon_loc = entry->start; 3784 3785 /* 3786 * Delete the entry only after removing all pmap 3787 * entries pointing to its pages. (Otherwise, its 3788 * page frames may be reallocated, and any modify bits 3789 * will be set in the wrong object!) 3790 */ 3791 vm_map_entry_delete(map, entry); 3792 entry = next; 3793 } 3794 return (KERN_SUCCESS); 3795 } 3796 3797 /* 3798 * vm_map_remove: 3799 * 3800 * Remove the given address range from the target map. 3801 * This is the exported form of vm_map_delete. 3802 */ 3803 int 3804 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 3805 { 3806 int result; 3807 3808 vm_map_lock(map); 3809 VM_MAP_RANGE_CHECK(map, start, end); 3810 result = vm_map_delete(map, start, end); 3811 vm_map_unlock(map); 3812 return (result); 3813 } 3814 3815 /* 3816 * vm_map_check_protection: 3817 * 3818 * Assert that the target map allows the specified privilege on the 3819 * entire address region given. The entire region must be allocated. 3820 * 3821 * WARNING! This code does not and should not check whether the 3822 * contents of the region is accessible. For example a smaller file 3823 * might be mapped into a larger address space. 3824 * 3825 * NOTE! This code is also called by munmap(). 3826 * 3827 * The map must be locked. A read lock is sufficient. 3828 */ 3829 boolean_t 3830 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3831 vm_prot_t protection) 3832 { 3833 vm_map_entry_t entry; 3834 vm_map_entry_t tmp_entry; 3835 3836 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3837 return (FALSE); 3838 entry = tmp_entry; 3839 3840 while (start < end) { 3841 /* 3842 * No holes allowed! 3843 */ 3844 if (start < entry->start) 3845 return (FALSE); 3846 /* 3847 * Check protection associated with entry. 3848 */ 3849 if ((entry->protection & protection) != protection) 3850 return (FALSE); 3851 /* go to next entry */ 3852 start = entry->end; 3853 entry = vm_map_entry_succ(entry); 3854 } 3855 return (TRUE); 3856 } 3857 3858 3859 /* 3860 * 3861 * vm_map_copy_swap_object: 3862 * 3863 * Copies a swap-backed object from an existing map entry to a 3864 * new one. Carries forward the swap charge. May change the 3865 * src object on return. 3866 */ 3867 static void 3868 vm_map_copy_swap_object(vm_map_entry_t src_entry, vm_map_entry_t dst_entry, 3869 vm_offset_t size, vm_ooffset_t *fork_charge) 3870 { 3871 vm_object_t src_object; 3872 struct ucred *cred; 3873 int charged; 3874 3875 src_object = src_entry->object.vm_object; 3876 VM_OBJECT_WLOCK(src_object); 3877 charged = ENTRY_CHARGED(src_entry); 3878 vm_object_collapse(src_object); 3879 if ((src_object->flags & OBJ_ONEMAPPING) != 0) { 3880 vm_object_split(src_entry); 3881 src_object = src_entry->object.vm_object; 3882 } 3883 vm_object_reference_locked(src_object); 3884 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3885 if (src_entry->cred != NULL && 3886 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3887 KASSERT(src_object->cred == NULL, 3888 ("OVERCOMMIT: vm_map_copy_anon_entry: cred %p", 3889 src_object)); 3890 src_object->cred = src_entry->cred; 3891 src_object->charge = size; 3892 } 3893 VM_OBJECT_WUNLOCK(src_object); 3894 dst_entry->object.vm_object = src_object; 3895 if (charged) { 3896 cred = curthread->td_ucred; 3897 crhold(cred); 3898 dst_entry->cred = cred; 3899 *fork_charge += size; 3900 if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3901 crhold(cred); 3902 src_entry->cred = cred; 3903 *fork_charge += size; 3904 } 3905 } 3906 } 3907 3908 /* 3909 * vm_map_copy_entry: 3910 * 3911 * Copies the contents of the source entry to the destination 3912 * entry. The entries *must* be aligned properly. 3913 */ 3914 static void 3915 vm_map_copy_entry( 3916 vm_map_t src_map, 3917 vm_map_t dst_map, 3918 vm_map_entry_t src_entry, 3919 vm_map_entry_t dst_entry, 3920 vm_ooffset_t *fork_charge) 3921 { 3922 vm_object_t src_object; 3923 vm_map_entry_t fake_entry; 3924 vm_offset_t size; 3925 3926 VM_MAP_ASSERT_LOCKED(dst_map); 3927 3928 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3929 return; 3930 3931 if (src_entry->wired_count == 0 || 3932 (src_entry->protection & VM_PROT_WRITE) == 0) { 3933 /* 3934 * If the source entry is marked needs_copy, it is already 3935 * write-protected. 3936 */ 3937 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3938 (src_entry->protection & VM_PROT_WRITE) != 0) { 3939 pmap_protect(src_map->pmap, 3940 src_entry->start, 3941 src_entry->end, 3942 src_entry->protection & ~VM_PROT_WRITE); 3943 } 3944 3945 /* 3946 * Make a copy of the object. 3947 */ 3948 size = src_entry->end - src_entry->start; 3949 if ((src_object = src_entry->object.vm_object) != NULL) { 3950 if (src_object->type == OBJT_DEFAULT || 3951 src_object->type == OBJT_SWAP) { 3952 vm_map_copy_swap_object(src_entry, dst_entry, 3953 size, fork_charge); 3954 /* May have split/collapsed, reload obj. */ 3955 src_object = src_entry->object.vm_object; 3956 } else { 3957 vm_object_reference(src_object); 3958 dst_entry->object.vm_object = src_object; 3959 } 3960 src_entry->eflags |= MAP_ENTRY_COW | 3961 MAP_ENTRY_NEEDS_COPY; 3962 dst_entry->eflags |= MAP_ENTRY_COW | 3963 MAP_ENTRY_NEEDS_COPY; 3964 dst_entry->offset = src_entry->offset; 3965 if (src_entry->eflags & MAP_ENTRY_WRITECNT) { 3966 /* 3967 * MAP_ENTRY_WRITECNT cannot 3968 * indicate write reference from 3969 * src_entry, since the entry is 3970 * marked as needs copy. Allocate a 3971 * fake entry that is used to 3972 * decrement object->un_pager writecount 3973 * at the appropriate time. Attach 3974 * fake_entry to the deferred list. 3975 */ 3976 fake_entry = vm_map_entry_create(dst_map); 3977 fake_entry->eflags = MAP_ENTRY_WRITECNT; 3978 src_entry->eflags &= ~MAP_ENTRY_WRITECNT; 3979 vm_object_reference(src_object); 3980 fake_entry->object.vm_object = src_object; 3981 fake_entry->start = src_entry->start; 3982 fake_entry->end = src_entry->end; 3983 fake_entry->defer_next = 3984 curthread->td_map_def_user; 3985 curthread->td_map_def_user = fake_entry; 3986 } 3987 3988 pmap_copy(dst_map->pmap, src_map->pmap, 3989 dst_entry->start, dst_entry->end - dst_entry->start, 3990 src_entry->start); 3991 } else { 3992 dst_entry->object.vm_object = NULL; 3993 dst_entry->offset = 0; 3994 if (src_entry->cred != NULL) { 3995 dst_entry->cred = curthread->td_ucred; 3996 crhold(dst_entry->cred); 3997 *fork_charge += size; 3998 } 3999 } 4000 } else { 4001 /* 4002 * We don't want to make writeable wired pages copy-on-write. 4003 * Immediately copy these pages into the new map by simulating 4004 * page faults. The new pages are pageable. 4005 */ 4006 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 4007 fork_charge); 4008 } 4009 } 4010 4011 /* 4012 * vmspace_map_entry_forked: 4013 * Update the newly-forked vmspace each time a map entry is inherited 4014 * or copied. The values for vm_dsize and vm_tsize are approximate 4015 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 4016 */ 4017 static void 4018 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 4019 vm_map_entry_t entry) 4020 { 4021 vm_size_t entrysize; 4022 vm_offset_t newend; 4023 4024 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) 4025 return; 4026 entrysize = entry->end - entry->start; 4027 vm2->vm_map.size += entrysize; 4028 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 4029 vm2->vm_ssize += btoc(entrysize); 4030 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 4031 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 4032 newend = MIN(entry->end, 4033 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 4034 vm2->vm_dsize += btoc(newend - entry->start); 4035 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 4036 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 4037 newend = MIN(entry->end, 4038 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 4039 vm2->vm_tsize += btoc(newend - entry->start); 4040 } 4041 } 4042 4043 /* 4044 * vmspace_fork: 4045 * Create a new process vmspace structure and vm_map 4046 * based on those of an existing process. The new map 4047 * is based on the old map, according to the inheritance 4048 * values on the regions in that map. 4049 * 4050 * XXX It might be worth coalescing the entries added to the new vmspace. 4051 * 4052 * The source map must not be locked. 4053 */ 4054 struct vmspace * 4055 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 4056 { 4057 struct vmspace *vm2; 4058 vm_map_t new_map, old_map; 4059 vm_map_entry_t new_entry, old_entry; 4060 vm_object_t object; 4061 int error, locked; 4062 vm_inherit_t inh; 4063 4064 old_map = &vm1->vm_map; 4065 /* Copy immutable fields of vm1 to vm2. */ 4066 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), 4067 pmap_pinit); 4068 if (vm2 == NULL) 4069 return (NULL); 4070 4071 vm2->vm_taddr = vm1->vm_taddr; 4072 vm2->vm_daddr = vm1->vm_daddr; 4073 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 4074 vm_map_lock(old_map); 4075 if (old_map->busy) 4076 vm_map_wait_busy(old_map); 4077 new_map = &vm2->vm_map; 4078 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 4079 KASSERT(locked, ("vmspace_fork: lock failed")); 4080 4081 error = pmap_vmspace_copy(new_map->pmap, old_map->pmap); 4082 if (error != 0) { 4083 sx_xunlock(&old_map->lock); 4084 sx_xunlock(&new_map->lock); 4085 vm_map_process_deferred(); 4086 vmspace_free(vm2); 4087 return (NULL); 4088 } 4089 4090 new_map->anon_loc = old_map->anon_loc; 4091 4092 VM_MAP_ENTRY_FOREACH(old_entry, old_map) { 4093 if ((old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 4094 panic("vm_map_fork: encountered a submap"); 4095 4096 inh = old_entry->inheritance; 4097 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 && 4098 inh != VM_INHERIT_NONE) 4099 inh = VM_INHERIT_COPY; 4100 4101 switch (inh) { 4102 case VM_INHERIT_NONE: 4103 break; 4104 4105 case VM_INHERIT_SHARE: 4106 /* 4107 * Clone the entry, creating the shared object if 4108 * necessary. 4109 */ 4110 object = old_entry->object.vm_object; 4111 if (object == NULL) { 4112 vm_map_entry_back(old_entry); 4113 object = old_entry->object.vm_object; 4114 } 4115 4116 /* 4117 * Add the reference before calling vm_object_shadow 4118 * to insure that a shadow object is created. 4119 */ 4120 vm_object_reference(object); 4121 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4122 vm_object_shadow(&old_entry->object.vm_object, 4123 &old_entry->offset, 4124 old_entry->end - old_entry->start, 4125 old_entry->cred, 4126 /* Transfer the second reference too. */ 4127 true); 4128 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4129 old_entry->cred = NULL; 4130 vm_object_reference( 4131 old_entry->object.vm_object); 4132 4133 /* 4134 * As in vm_map_merged_neighbor_dispose(), 4135 * the vnode lock will not be acquired in 4136 * this call to vm_object_deallocate(). 4137 */ 4138 vm_object_deallocate(object); 4139 object = old_entry->object.vm_object; 4140 } else { 4141 VM_OBJECT_WLOCK(object); 4142 vm_object_clear_flag(object, OBJ_ONEMAPPING); 4143 if (old_entry->cred != NULL) { 4144 KASSERT(object->cred == NULL, 4145 ("vmspace_fork both cred")); 4146 object->cred = old_entry->cred; 4147 object->charge = old_entry->end - 4148 old_entry->start; 4149 old_entry->cred = NULL; 4150 } 4151 4152 /* 4153 * Assert the correct state of the vnode 4154 * v_writecount while the object is locked, to 4155 * not relock it later for the assertion 4156 * correctness. 4157 */ 4158 if (old_entry->eflags & MAP_ENTRY_WRITECNT && 4159 object->type == OBJT_VNODE) { 4160 KASSERT(((struct vnode *)object-> 4161 handle)->v_writecount > 0, 4162 ("vmspace_fork: v_writecount %p", 4163 object)); 4164 KASSERT(object->un_pager.vnp. 4165 writemappings > 0, 4166 ("vmspace_fork: vnp.writecount %p", 4167 object)); 4168 } 4169 VM_OBJECT_WUNLOCK(object); 4170 } 4171 4172 /* 4173 * Clone the entry, referencing the shared object. 4174 */ 4175 new_entry = vm_map_entry_create(new_map); 4176 *new_entry = *old_entry; 4177 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 4178 MAP_ENTRY_IN_TRANSITION); 4179 new_entry->wiring_thread = NULL; 4180 new_entry->wired_count = 0; 4181 if (new_entry->eflags & MAP_ENTRY_WRITECNT) { 4182 vm_pager_update_writecount(object, 4183 new_entry->start, new_entry->end); 4184 } 4185 vm_map_entry_set_vnode_text(new_entry, true); 4186 4187 /* 4188 * Insert the entry into the new map -- we know we're 4189 * inserting at the end of the new map. 4190 */ 4191 vm_map_entry_link(new_map, new_entry); 4192 vmspace_map_entry_forked(vm1, vm2, new_entry); 4193 4194 /* 4195 * Update the physical map 4196 */ 4197 pmap_copy(new_map->pmap, old_map->pmap, 4198 new_entry->start, 4199 (old_entry->end - old_entry->start), 4200 old_entry->start); 4201 break; 4202 4203 case VM_INHERIT_COPY: 4204 /* 4205 * Clone the entry and link into the map. 4206 */ 4207 new_entry = vm_map_entry_create(new_map); 4208 *new_entry = *old_entry; 4209 /* 4210 * Copied entry is COW over the old object. 4211 */ 4212 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 4213 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT); 4214 new_entry->wiring_thread = NULL; 4215 new_entry->wired_count = 0; 4216 new_entry->object.vm_object = NULL; 4217 new_entry->cred = NULL; 4218 vm_map_entry_link(new_map, new_entry); 4219 vmspace_map_entry_forked(vm1, vm2, new_entry); 4220 vm_map_copy_entry(old_map, new_map, old_entry, 4221 new_entry, fork_charge); 4222 vm_map_entry_set_vnode_text(new_entry, true); 4223 break; 4224 4225 case VM_INHERIT_ZERO: 4226 /* 4227 * Create a new anonymous mapping entry modelled from 4228 * the old one. 4229 */ 4230 new_entry = vm_map_entry_create(new_map); 4231 memset(new_entry, 0, sizeof(*new_entry)); 4232 4233 new_entry->start = old_entry->start; 4234 new_entry->end = old_entry->end; 4235 new_entry->eflags = old_entry->eflags & 4236 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION | 4237 MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC); 4238 new_entry->protection = old_entry->protection; 4239 new_entry->max_protection = old_entry->max_protection; 4240 new_entry->inheritance = VM_INHERIT_ZERO; 4241 4242 vm_map_entry_link(new_map, new_entry); 4243 vmspace_map_entry_forked(vm1, vm2, new_entry); 4244 4245 new_entry->cred = curthread->td_ucred; 4246 crhold(new_entry->cred); 4247 *fork_charge += (new_entry->end - new_entry->start); 4248 4249 break; 4250 } 4251 } 4252 /* 4253 * Use inlined vm_map_unlock() to postpone handling the deferred 4254 * map entries, which cannot be done until both old_map and 4255 * new_map locks are released. 4256 */ 4257 sx_xunlock(&old_map->lock); 4258 sx_xunlock(&new_map->lock); 4259 vm_map_process_deferred(); 4260 4261 return (vm2); 4262 } 4263 4264 /* 4265 * Create a process's stack for exec_new_vmspace(). This function is never 4266 * asked to wire the newly created stack. 4267 */ 4268 int 4269 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 4270 vm_prot_t prot, vm_prot_t max, int cow) 4271 { 4272 vm_size_t growsize, init_ssize; 4273 rlim_t vmemlim; 4274 int rv; 4275 4276 MPASS((map->flags & MAP_WIREFUTURE) == 0); 4277 growsize = sgrowsiz; 4278 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 4279 vm_map_lock(map); 4280 vmemlim = lim_cur(curthread, RLIMIT_VMEM); 4281 /* If we would blow our VMEM resource limit, no go */ 4282 if (map->size + init_ssize > vmemlim) { 4283 rv = KERN_NO_SPACE; 4284 goto out; 4285 } 4286 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 4287 max, cow); 4288 out: 4289 vm_map_unlock(map); 4290 return (rv); 4291 } 4292 4293 static int stack_guard_page = 1; 4294 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN, 4295 &stack_guard_page, 0, 4296 "Specifies the number of guard pages for a stack that grows"); 4297 4298 static int 4299 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 4300 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 4301 { 4302 vm_map_entry_t new_entry, prev_entry; 4303 vm_offset_t bot, gap_bot, gap_top, top; 4304 vm_size_t init_ssize, sgp; 4305 int orient, rv; 4306 4307 /* 4308 * The stack orientation is piggybacked with the cow argument. 4309 * Extract it into orient and mask the cow argument so that we 4310 * don't pass it around further. 4311 */ 4312 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP); 4313 KASSERT(orient != 0, ("No stack grow direction")); 4314 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP), 4315 ("bi-dir stack")); 4316 4317 if (addrbos < vm_map_min(map) || 4318 addrbos + max_ssize > vm_map_max(map) || 4319 addrbos + max_ssize <= addrbos) 4320 return (KERN_INVALID_ADDRESS); 4321 sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 || 4322 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 : 4323 (vm_size_t)stack_guard_page * PAGE_SIZE; 4324 if (sgp >= max_ssize) 4325 return (KERN_INVALID_ARGUMENT); 4326 4327 init_ssize = growsize; 4328 if (max_ssize < init_ssize + sgp) 4329 init_ssize = max_ssize - sgp; 4330 4331 /* If addr is already mapped, no go */ 4332 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 4333 return (KERN_NO_SPACE); 4334 4335 /* 4336 * If we can't accommodate max_ssize in the current mapping, no go. 4337 */ 4338 if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize) 4339 return (KERN_NO_SPACE); 4340 4341 /* 4342 * We initially map a stack of only init_ssize. We will grow as 4343 * needed later. Depending on the orientation of the stack (i.e. 4344 * the grow direction) we either map at the top of the range, the 4345 * bottom of the range or in the middle. 4346 * 4347 * Note: we would normally expect prot and max to be VM_PROT_ALL, 4348 * and cow to be 0. Possibly we should eliminate these as input 4349 * parameters, and just pass these values here in the insert call. 4350 */ 4351 if (orient == MAP_STACK_GROWS_DOWN) { 4352 bot = addrbos + max_ssize - init_ssize; 4353 top = bot + init_ssize; 4354 gap_bot = addrbos; 4355 gap_top = bot; 4356 } else /* if (orient == MAP_STACK_GROWS_UP) */ { 4357 bot = addrbos; 4358 top = bot + init_ssize; 4359 gap_bot = top; 4360 gap_top = addrbos + max_ssize; 4361 } 4362 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 4363 if (rv != KERN_SUCCESS) 4364 return (rv); 4365 new_entry = vm_map_entry_succ(prev_entry); 4366 KASSERT(new_entry->end == top || new_entry->start == bot, 4367 ("Bad entry start/end for new stack entry")); 4368 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 || 4369 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, 4370 ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 4371 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 || 4372 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0, 4373 ("new entry lacks MAP_ENTRY_GROWS_UP")); 4374 if (gap_bot == gap_top) 4375 return (KERN_SUCCESS); 4376 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE, 4377 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ? 4378 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP)); 4379 if (rv == KERN_SUCCESS) { 4380 /* 4381 * Gap can never successfully handle a fault, so 4382 * read-ahead logic is never used for it. Re-use 4383 * next_read of the gap entry to store 4384 * stack_guard_page for vm_map_growstack(). 4385 */ 4386 if (orient == MAP_STACK_GROWS_DOWN) 4387 vm_map_entry_pred(new_entry)->next_read = sgp; 4388 else 4389 vm_map_entry_succ(new_entry)->next_read = sgp; 4390 } else { 4391 (void)vm_map_delete(map, bot, top); 4392 } 4393 return (rv); 4394 } 4395 4396 /* 4397 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we 4398 * successfully grow the stack. 4399 */ 4400 static int 4401 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry) 4402 { 4403 vm_map_entry_t stack_entry; 4404 struct proc *p; 4405 struct vmspace *vm; 4406 struct ucred *cred; 4407 vm_offset_t gap_end, gap_start, grow_start; 4408 vm_size_t grow_amount, guard, max_grow; 4409 rlim_t lmemlim, stacklim, vmemlim; 4410 int rv, rv1; 4411 bool gap_deleted, grow_down, is_procstack; 4412 #ifdef notyet 4413 uint64_t limit; 4414 #endif 4415 #ifdef RACCT 4416 int error; 4417 #endif 4418 4419 p = curproc; 4420 vm = p->p_vmspace; 4421 4422 /* 4423 * Disallow stack growth when the access is performed by a 4424 * debugger or AIO daemon. The reason is that the wrong 4425 * resource limits are applied. 4426 */ 4427 if (p != initproc && (map != &p->p_vmspace->vm_map || 4428 p->p_textvp == NULL)) 4429 return (KERN_FAILURE); 4430 4431 MPASS(!map->system_map); 4432 4433 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK); 4434 stacklim = lim_cur(curthread, RLIMIT_STACK); 4435 vmemlim = lim_cur(curthread, RLIMIT_VMEM); 4436 retry: 4437 /* If addr is not in a hole for a stack grow area, no need to grow. */ 4438 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry)) 4439 return (KERN_FAILURE); 4440 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0) 4441 return (KERN_SUCCESS); 4442 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) { 4443 stack_entry = vm_map_entry_succ(gap_entry); 4444 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 || 4445 stack_entry->start != gap_entry->end) 4446 return (KERN_FAILURE); 4447 grow_amount = round_page(stack_entry->start - addr); 4448 grow_down = true; 4449 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) { 4450 stack_entry = vm_map_entry_pred(gap_entry); 4451 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 || 4452 stack_entry->end != gap_entry->start) 4453 return (KERN_FAILURE); 4454 grow_amount = round_page(addr + 1 - stack_entry->end); 4455 grow_down = false; 4456 } else { 4457 return (KERN_FAILURE); 4458 } 4459 guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 || 4460 (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 : 4461 gap_entry->next_read; 4462 max_grow = gap_entry->end - gap_entry->start; 4463 if (guard > max_grow) 4464 return (KERN_NO_SPACE); 4465 max_grow -= guard; 4466 if (grow_amount > max_grow) 4467 return (KERN_NO_SPACE); 4468 4469 /* 4470 * If this is the main process stack, see if we're over the stack 4471 * limit. 4472 */ 4473 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr && 4474 addr < (vm_offset_t)p->p_sysent->sv_usrstack; 4475 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) 4476 return (KERN_NO_SPACE); 4477 4478 #ifdef RACCT 4479 if (racct_enable) { 4480 PROC_LOCK(p); 4481 if (is_procstack && racct_set(p, RACCT_STACK, 4482 ctob(vm->vm_ssize) + grow_amount)) { 4483 PROC_UNLOCK(p); 4484 return (KERN_NO_SPACE); 4485 } 4486 PROC_UNLOCK(p); 4487 } 4488 #endif 4489 4490 grow_amount = roundup(grow_amount, sgrowsiz); 4491 if (grow_amount > max_grow) 4492 grow_amount = max_grow; 4493 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 4494 grow_amount = trunc_page((vm_size_t)stacklim) - 4495 ctob(vm->vm_ssize); 4496 } 4497 4498 #ifdef notyet 4499 PROC_LOCK(p); 4500 limit = racct_get_available(p, RACCT_STACK); 4501 PROC_UNLOCK(p); 4502 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 4503 grow_amount = limit - ctob(vm->vm_ssize); 4504 #endif 4505 4506 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) { 4507 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 4508 rv = KERN_NO_SPACE; 4509 goto out; 4510 } 4511 #ifdef RACCT 4512 if (racct_enable) { 4513 PROC_LOCK(p); 4514 if (racct_set(p, RACCT_MEMLOCK, 4515 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 4516 PROC_UNLOCK(p); 4517 rv = KERN_NO_SPACE; 4518 goto out; 4519 } 4520 PROC_UNLOCK(p); 4521 } 4522 #endif 4523 } 4524 4525 /* If we would blow our VMEM resource limit, no go */ 4526 if (map->size + grow_amount > vmemlim) { 4527 rv = KERN_NO_SPACE; 4528 goto out; 4529 } 4530 #ifdef RACCT 4531 if (racct_enable) { 4532 PROC_LOCK(p); 4533 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 4534 PROC_UNLOCK(p); 4535 rv = KERN_NO_SPACE; 4536 goto out; 4537 } 4538 PROC_UNLOCK(p); 4539 } 4540 #endif 4541 4542 if (vm_map_lock_upgrade(map)) { 4543 gap_entry = NULL; 4544 vm_map_lock_read(map); 4545 goto retry; 4546 } 4547 4548 if (grow_down) { 4549 grow_start = gap_entry->end - grow_amount; 4550 if (gap_entry->start + grow_amount == gap_entry->end) { 4551 gap_start = gap_entry->start; 4552 gap_end = gap_entry->end; 4553 vm_map_entry_delete(map, gap_entry); 4554 gap_deleted = true; 4555 } else { 4556 MPASS(gap_entry->start < gap_entry->end - grow_amount); 4557 vm_map_entry_resize(map, gap_entry, -grow_amount); 4558 gap_deleted = false; 4559 } 4560 rv = vm_map_insert(map, NULL, 0, grow_start, 4561 grow_start + grow_amount, 4562 stack_entry->protection, stack_entry->max_protection, 4563 MAP_STACK_GROWS_DOWN); 4564 if (rv != KERN_SUCCESS) { 4565 if (gap_deleted) { 4566 rv1 = vm_map_insert(map, NULL, 0, gap_start, 4567 gap_end, VM_PROT_NONE, VM_PROT_NONE, 4568 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN); 4569 MPASS(rv1 == KERN_SUCCESS); 4570 } else 4571 vm_map_entry_resize(map, gap_entry, 4572 grow_amount); 4573 } 4574 } else { 4575 grow_start = stack_entry->end; 4576 cred = stack_entry->cred; 4577 if (cred == NULL && stack_entry->object.vm_object != NULL) 4578 cred = stack_entry->object.vm_object->cred; 4579 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 4580 rv = KERN_NO_SPACE; 4581 /* Grow the underlying object if applicable. */ 4582 else if (stack_entry->object.vm_object == NULL || 4583 vm_object_coalesce(stack_entry->object.vm_object, 4584 stack_entry->offset, 4585 (vm_size_t)(stack_entry->end - stack_entry->start), 4586 grow_amount, cred != NULL)) { 4587 if (gap_entry->start + grow_amount == gap_entry->end) { 4588 vm_map_entry_delete(map, gap_entry); 4589 vm_map_entry_resize(map, stack_entry, 4590 grow_amount); 4591 } else { 4592 gap_entry->start += grow_amount; 4593 stack_entry->end += grow_amount; 4594 } 4595 map->size += grow_amount; 4596 rv = KERN_SUCCESS; 4597 } else 4598 rv = KERN_FAILURE; 4599 } 4600 if (rv == KERN_SUCCESS && is_procstack) 4601 vm->vm_ssize += btoc(grow_amount); 4602 4603 /* 4604 * Heed the MAP_WIREFUTURE flag if it was set for this process. 4605 */ 4606 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) { 4607 rv = vm_map_wire_locked(map, grow_start, 4608 grow_start + grow_amount, 4609 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 4610 } 4611 vm_map_lock_downgrade(map); 4612 4613 out: 4614 #ifdef RACCT 4615 if (racct_enable && rv != KERN_SUCCESS) { 4616 PROC_LOCK(p); 4617 error = racct_set(p, RACCT_VMEM, map->size); 4618 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 4619 if (!old_mlock) { 4620 error = racct_set(p, RACCT_MEMLOCK, 4621 ptoa(pmap_wired_count(map->pmap))); 4622 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 4623 } 4624 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 4625 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 4626 PROC_UNLOCK(p); 4627 } 4628 #endif 4629 4630 return (rv); 4631 } 4632 4633 /* 4634 * Unshare the specified VM space for exec. If other processes are 4635 * mapped to it, then create a new one. The new vmspace is null. 4636 */ 4637 int 4638 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 4639 { 4640 struct vmspace *oldvmspace = p->p_vmspace; 4641 struct vmspace *newvmspace; 4642 4643 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 4644 ("vmspace_exec recursed")); 4645 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit); 4646 if (newvmspace == NULL) 4647 return (ENOMEM); 4648 newvmspace->vm_swrss = oldvmspace->vm_swrss; 4649 /* 4650 * This code is written like this for prototype purposes. The 4651 * goal is to avoid running down the vmspace here, but let the 4652 * other process's that are still using the vmspace to finally 4653 * run it down. Even though there is little or no chance of blocking 4654 * here, it is a good idea to keep this form for future mods. 4655 */ 4656 PROC_VMSPACE_LOCK(p); 4657 p->p_vmspace = newvmspace; 4658 PROC_VMSPACE_UNLOCK(p); 4659 if (p == curthread->td_proc) 4660 pmap_activate(curthread); 4661 curthread->td_pflags |= TDP_EXECVMSPC; 4662 return (0); 4663 } 4664 4665 /* 4666 * Unshare the specified VM space for forcing COW. This 4667 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 4668 */ 4669 int 4670 vmspace_unshare(struct proc *p) 4671 { 4672 struct vmspace *oldvmspace = p->p_vmspace; 4673 struct vmspace *newvmspace; 4674 vm_ooffset_t fork_charge; 4675 4676 if (oldvmspace->vm_refcnt == 1) 4677 return (0); 4678 fork_charge = 0; 4679 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 4680 if (newvmspace == NULL) 4681 return (ENOMEM); 4682 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 4683 vmspace_free(newvmspace); 4684 return (ENOMEM); 4685 } 4686 PROC_VMSPACE_LOCK(p); 4687 p->p_vmspace = newvmspace; 4688 PROC_VMSPACE_UNLOCK(p); 4689 if (p == curthread->td_proc) 4690 pmap_activate(curthread); 4691 vmspace_free(oldvmspace); 4692 return (0); 4693 } 4694 4695 /* 4696 * vm_map_lookup: 4697 * 4698 * Finds the VM object, offset, and 4699 * protection for a given virtual address in the 4700 * specified map, assuming a page fault of the 4701 * type specified. 4702 * 4703 * Leaves the map in question locked for read; return 4704 * values are guaranteed until a vm_map_lookup_done 4705 * call is performed. Note that the map argument 4706 * is in/out; the returned map must be used in 4707 * the call to vm_map_lookup_done. 4708 * 4709 * A handle (out_entry) is returned for use in 4710 * vm_map_lookup_done, to make that fast. 4711 * 4712 * If a lookup is requested with "write protection" 4713 * specified, the map may be changed to perform virtual 4714 * copying operations, although the data referenced will 4715 * remain the same. 4716 */ 4717 int 4718 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 4719 vm_offset_t vaddr, 4720 vm_prot_t fault_typea, 4721 vm_map_entry_t *out_entry, /* OUT */ 4722 vm_object_t *object, /* OUT */ 4723 vm_pindex_t *pindex, /* OUT */ 4724 vm_prot_t *out_prot, /* OUT */ 4725 boolean_t *wired) /* OUT */ 4726 { 4727 vm_map_entry_t entry; 4728 vm_map_t map = *var_map; 4729 vm_prot_t prot; 4730 vm_prot_t fault_type; 4731 vm_object_t eobject; 4732 vm_size_t size; 4733 struct ucred *cred; 4734 4735 RetryLookup: 4736 4737 vm_map_lock_read(map); 4738 4739 RetryLookupLocked: 4740 /* 4741 * Lookup the faulting address. 4742 */ 4743 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 4744 vm_map_unlock_read(map); 4745 return (KERN_INVALID_ADDRESS); 4746 } 4747 4748 entry = *out_entry; 4749 4750 /* 4751 * Handle submaps. 4752 */ 4753 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4754 vm_map_t old_map = map; 4755 4756 *var_map = map = entry->object.sub_map; 4757 vm_map_unlock_read(old_map); 4758 goto RetryLookup; 4759 } 4760 4761 /* 4762 * Check whether this task is allowed to have this page. 4763 */ 4764 prot = entry->protection; 4765 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) { 4766 fault_typea &= ~VM_PROT_FAULT_LOOKUP; 4767 if (prot == VM_PROT_NONE && map != kernel_map && 4768 (entry->eflags & MAP_ENTRY_GUARD) != 0 && 4769 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN | 4770 MAP_ENTRY_STACK_GAP_UP)) != 0 && 4771 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS) 4772 goto RetryLookupLocked; 4773 } 4774 fault_type = fault_typea & VM_PROT_ALL; 4775 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 4776 vm_map_unlock_read(map); 4777 return (KERN_PROTECTION_FAILURE); 4778 } 4779 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags & 4780 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) != 4781 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY), 4782 ("entry %p flags %x", entry, entry->eflags)); 4783 if ((fault_typea & VM_PROT_COPY) != 0 && 4784 (entry->max_protection & VM_PROT_WRITE) == 0 && 4785 (entry->eflags & MAP_ENTRY_COW) == 0) { 4786 vm_map_unlock_read(map); 4787 return (KERN_PROTECTION_FAILURE); 4788 } 4789 4790 /* 4791 * If this page is not pageable, we have to get it for all possible 4792 * accesses. 4793 */ 4794 *wired = (entry->wired_count != 0); 4795 if (*wired) 4796 fault_type = entry->protection; 4797 size = entry->end - entry->start; 4798 4799 /* 4800 * If the entry was copy-on-write, we either ... 4801 */ 4802 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4803 /* 4804 * If we want to write the page, we may as well handle that 4805 * now since we've got the map locked. 4806 * 4807 * If we don't need to write the page, we just demote the 4808 * permissions allowed. 4809 */ 4810 if ((fault_type & VM_PROT_WRITE) != 0 || 4811 (fault_typea & VM_PROT_COPY) != 0) { 4812 /* 4813 * Make a new object, and place it in the object 4814 * chain. Note that no new references have appeared 4815 * -- one just moved from the map to the new 4816 * object. 4817 */ 4818 if (vm_map_lock_upgrade(map)) 4819 goto RetryLookup; 4820 4821 if (entry->cred == NULL) { 4822 /* 4823 * The debugger owner is charged for 4824 * the memory. 4825 */ 4826 cred = curthread->td_ucred; 4827 crhold(cred); 4828 if (!swap_reserve_by_cred(size, cred)) { 4829 crfree(cred); 4830 vm_map_unlock(map); 4831 return (KERN_RESOURCE_SHORTAGE); 4832 } 4833 entry->cred = cred; 4834 } 4835 eobject = entry->object.vm_object; 4836 vm_object_shadow(&entry->object.vm_object, 4837 &entry->offset, size, entry->cred, false); 4838 if (eobject == entry->object.vm_object) { 4839 /* 4840 * The object was not shadowed. 4841 */ 4842 swap_release_by_cred(size, entry->cred); 4843 crfree(entry->cred); 4844 } 4845 entry->cred = NULL; 4846 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4847 4848 vm_map_lock_downgrade(map); 4849 } else { 4850 /* 4851 * We're attempting to read a copy-on-write page -- 4852 * don't allow writes. 4853 */ 4854 prot &= ~VM_PROT_WRITE; 4855 } 4856 } 4857 4858 /* 4859 * Create an object if necessary. 4860 */ 4861 if (entry->object.vm_object == NULL && !map->system_map) { 4862 if (vm_map_lock_upgrade(map)) 4863 goto RetryLookup; 4864 entry->object.vm_object = vm_object_allocate_anon(atop(size), 4865 NULL, entry->cred, entry->cred != NULL ? size : 0); 4866 entry->offset = 0; 4867 entry->cred = NULL; 4868 vm_map_lock_downgrade(map); 4869 } 4870 4871 /* 4872 * Return the object/offset from this entry. If the entry was 4873 * copy-on-write or empty, it has been fixed up. 4874 */ 4875 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4876 *object = entry->object.vm_object; 4877 4878 *out_prot = prot; 4879 return (KERN_SUCCESS); 4880 } 4881 4882 /* 4883 * vm_map_lookup_locked: 4884 * 4885 * Lookup the faulting address. A version of vm_map_lookup that returns 4886 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4887 */ 4888 int 4889 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4890 vm_offset_t vaddr, 4891 vm_prot_t fault_typea, 4892 vm_map_entry_t *out_entry, /* OUT */ 4893 vm_object_t *object, /* OUT */ 4894 vm_pindex_t *pindex, /* OUT */ 4895 vm_prot_t *out_prot, /* OUT */ 4896 boolean_t *wired) /* OUT */ 4897 { 4898 vm_map_entry_t entry; 4899 vm_map_t map = *var_map; 4900 vm_prot_t prot; 4901 vm_prot_t fault_type = fault_typea; 4902 4903 /* 4904 * Lookup the faulting address. 4905 */ 4906 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4907 return (KERN_INVALID_ADDRESS); 4908 4909 entry = *out_entry; 4910 4911 /* 4912 * Fail if the entry refers to a submap. 4913 */ 4914 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4915 return (KERN_FAILURE); 4916 4917 /* 4918 * Check whether this task is allowed to have this page. 4919 */ 4920 prot = entry->protection; 4921 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4922 if ((fault_type & prot) != fault_type) 4923 return (KERN_PROTECTION_FAILURE); 4924 4925 /* 4926 * If this page is not pageable, we have to get it for all possible 4927 * accesses. 4928 */ 4929 *wired = (entry->wired_count != 0); 4930 if (*wired) 4931 fault_type = entry->protection; 4932 4933 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4934 /* 4935 * Fail if the entry was copy-on-write for a write fault. 4936 */ 4937 if (fault_type & VM_PROT_WRITE) 4938 return (KERN_FAILURE); 4939 /* 4940 * We're attempting to read a copy-on-write page -- 4941 * don't allow writes. 4942 */ 4943 prot &= ~VM_PROT_WRITE; 4944 } 4945 4946 /* 4947 * Fail if an object should be created. 4948 */ 4949 if (entry->object.vm_object == NULL && !map->system_map) 4950 return (KERN_FAILURE); 4951 4952 /* 4953 * Return the object/offset from this entry. If the entry was 4954 * copy-on-write or empty, it has been fixed up. 4955 */ 4956 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); 4957 *object = entry->object.vm_object; 4958 4959 *out_prot = prot; 4960 return (KERN_SUCCESS); 4961 } 4962 4963 /* 4964 * vm_map_lookup_done: 4965 * 4966 * Releases locks acquired by a vm_map_lookup 4967 * (according to the handle returned by that lookup). 4968 */ 4969 void 4970 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4971 { 4972 /* 4973 * Unlock the main-level map 4974 */ 4975 vm_map_unlock_read(map); 4976 } 4977 4978 vm_offset_t 4979 vm_map_max_KBI(const struct vm_map *map) 4980 { 4981 4982 return (vm_map_max(map)); 4983 } 4984 4985 vm_offset_t 4986 vm_map_min_KBI(const struct vm_map *map) 4987 { 4988 4989 return (vm_map_min(map)); 4990 } 4991 4992 pmap_t 4993 vm_map_pmap_KBI(vm_map_t map) 4994 { 4995 4996 return (map->pmap); 4997 } 4998 4999 #ifdef INVARIANTS 5000 static void 5001 _vm_map_assert_consistent(vm_map_t map, int check) 5002 { 5003 vm_map_entry_t entry, prev; 5004 vm_map_entry_t cur, header, lbound, ubound; 5005 vm_size_t max_left, max_right; 5006 5007 #ifdef DIAGNOSTIC 5008 ++map->nupdates; 5009 #endif 5010 if (enable_vmmap_check != check) 5011 return; 5012 5013 header = prev = &map->header; 5014 VM_MAP_ENTRY_FOREACH(entry, map) { 5015 KASSERT(prev->end <= entry->start, 5016 ("map %p prev->end = %jx, start = %jx", map, 5017 (uintmax_t)prev->end, (uintmax_t)entry->start)); 5018 KASSERT(entry->start < entry->end, 5019 ("map %p start = %jx, end = %jx", map, 5020 (uintmax_t)entry->start, (uintmax_t)entry->end)); 5021 KASSERT(entry->left == header || 5022 entry->left->start < entry->start, 5023 ("map %p left->start = %jx, start = %jx", map, 5024 (uintmax_t)entry->left->start, (uintmax_t)entry->start)); 5025 KASSERT(entry->right == header || 5026 entry->start < entry->right->start, 5027 ("map %p start = %jx, right->start = %jx", map, 5028 (uintmax_t)entry->start, (uintmax_t)entry->right->start)); 5029 cur = map->root; 5030 lbound = ubound = header; 5031 for (;;) { 5032 if (entry->start < cur->start) { 5033 ubound = cur; 5034 cur = cur->left; 5035 KASSERT(cur != lbound, 5036 ("map %p cannot find %jx", 5037 map, (uintmax_t)entry->start)); 5038 } else if (cur->end <= entry->start) { 5039 lbound = cur; 5040 cur = cur->right; 5041 KASSERT(cur != ubound, 5042 ("map %p cannot find %jx", 5043 map, (uintmax_t)entry->start)); 5044 } else { 5045 KASSERT(cur == entry, 5046 ("map %p cannot find %jx", 5047 map, (uintmax_t)entry->start)); 5048 break; 5049 } 5050 } 5051 max_left = vm_map_entry_max_free_left(entry, lbound); 5052 max_right = vm_map_entry_max_free_right(entry, ubound); 5053 KASSERT(entry->max_free == vm_size_max(max_left, max_right), 5054 ("map %p max = %jx, max_left = %jx, max_right = %jx", map, 5055 (uintmax_t)entry->max_free, 5056 (uintmax_t)max_left, (uintmax_t)max_right)); 5057 prev = entry; 5058 } 5059 KASSERT(prev->end <= entry->start, 5060 ("map %p prev->end = %jx, start = %jx", map, 5061 (uintmax_t)prev->end, (uintmax_t)entry->start)); 5062 } 5063 #endif 5064 5065 #include "opt_ddb.h" 5066 #ifdef DDB 5067 #include <sys/kernel.h> 5068 5069 #include <ddb/ddb.h> 5070 5071 static void 5072 vm_map_print(vm_map_t map) 5073 { 5074 vm_map_entry_t entry, prev; 5075 5076 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 5077 (void *)map, 5078 (void *)map->pmap, map->nentries, map->timestamp); 5079 5080 db_indent += 2; 5081 prev = &map->header; 5082 VM_MAP_ENTRY_FOREACH(entry, map) { 5083 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n", 5084 (void *)entry, (void *)entry->start, (void *)entry->end, 5085 entry->eflags); 5086 { 5087 static char *inheritance_name[4] = 5088 {"share", "copy", "none", "donate_copy"}; 5089 5090 db_iprintf(" prot=%x/%x/%s", 5091 entry->protection, 5092 entry->max_protection, 5093 inheritance_name[(int)(unsigned char) 5094 entry->inheritance]); 5095 if (entry->wired_count != 0) 5096 db_printf(", wired"); 5097 } 5098 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 5099 db_printf(", share=%p, offset=0x%jx\n", 5100 (void *)entry->object.sub_map, 5101 (uintmax_t)entry->offset); 5102 if (prev == &map->header || 5103 prev->object.sub_map != 5104 entry->object.sub_map) { 5105 db_indent += 2; 5106 vm_map_print((vm_map_t)entry->object.sub_map); 5107 db_indent -= 2; 5108 } 5109 } else { 5110 if (entry->cred != NULL) 5111 db_printf(", ruid %d", entry->cred->cr_ruid); 5112 db_printf(", object=%p, offset=0x%jx", 5113 (void *)entry->object.vm_object, 5114 (uintmax_t)entry->offset); 5115 if (entry->object.vm_object && entry->object.vm_object->cred) 5116 db_printf(", obj ruid %d charge %jx", 5117 entry->object.vm_object->cred->cr_ruid, 5118 (uintmax_t)entry->object.vm_object->charge); 5119 if (entry->eflags & MAP_ENTRY_COW) 5120 db_printf(", copy (%s)", 5121 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 5122 db_printf("\n"); 5123 5124 if (prev == &map->header || 5125 prev->object.vm_object != 5126 entry->object.vm_object) { 5127 db_indent += 2; 5128 vm_object_print((db_expr_t)(intptr_t) 5129 entry->object.vm_object, 5130 0, 0, (char *)0); 5131 db_indent -= 2; 5132 } 5133 } 5134 prev = entry; 5135 } 5136 db_indent -= 2; 5137 } 5138 5139 DB_SHOW_COMMAND(map, map) 5140 { 5141 5142 if (!have_addr) { 5143 db_printf("usage: show map <addr>\n"); 5144 return; 5145 } 5146 vm_map_print((vm_map_t)addr); 5147 } 5148 5149 DB_SHOW_COMMAND(procvm, procvm) 5150 { 5151 struct proc *p; 5152 5153 if (have_addr) { 5154 p = db_lookup_proc(addr); 5155 } else { 5156 p = curproc; 5157 } 5158 5159 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 5160 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 5161 (void *)vmspace_pmap(p->p_vmspace)); 5162 5163 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 5164 } 5165 5166 #endif /* DDB */ 5167