vm_map.c revision 343426
1/*- 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * The Mach Operating System project at Carnegie-Mellon University. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61/* 62 * Virtual memory mapping module. 63 */ 64 65#include <sys/cdefs.h> 66__FBSDID("$FreeBSD: stable/11/sys/vm/vm_map.c 343426 2019-01-25 11:46:07Z kib $"); 67 68#include <sys/param.h> 69#include <sys/systm.h> 70#include <sys/kernel.h> 71#include <sys/ktr.h> 72#include <sys/lock.h> 73#include <sys/mutex.h> 74#include <sys/proc.h> 75#include <sys/vmmeter.h> 76#include <sys/mman.h> 77#include <sys/vnode.h> 78#include <sys/racct.h> 79#include <sys/resourcevar.h> 80#include <sys/rwlock.h> 81#include <sys/file.h> 82#include <sys/sysctl.h> 83#include <sys/sysent.h> 84#include <sys/shm.h> 85 86#include <vm/vm.h> 87#include <vm/vm_param.h> 88#include <vm/pmap.h> 89#include <vm/vm_map.h> 90#include <vm/vm_page.h> 91#include <vm/vm_object.h> 92#include <vm/vm_pager.h> 93#include <vm/vm_kern.h> 94#include <vm/vm_extern.h> 95#include <vm/vnode_pager.h> 96#include <vm/swap_pager.h> 97#include <vm/uma.h> 98 99/* 100 * Virtual memory maps provide for the mapping, protection, 101 * and sharing of virtual memory objects. In addition, 102 * this module provides for an efficient virtual copy of 103 * memory from one map to another. 104 * 105 * Synchronization is required prior to most operations. 106 * 107 * Maps consist of an ordered doubly-linked list of simple 108 * entries; a self-adjusting binary search tree of these 109 * entries is used to speed up lookups. 110 * 111 * Since portions of maps are specified by start/end addresses, 112 * which may not align with existing map entries, all 113 * routines merely "clip" entries to these start/end values. 114 * [That is, an entry is split into two, bordering at a 115 * start or end value.] Note that these clippings may not 116 * always be necessary (as the two resulting entries are then 117 * not changed); however, the clipping is done for convenience. 118 * 119 * As mentioned above, virtual copy operations are performed 120 * by copying VM object references from one map to 121 * another, and then marking both regions as copy-on-write. 122 */ 123 124static struct mtx map_sleep_mtx; 125static uma_zone_t mapentzone; 126static uma_zone_t kmapentzone; 127static uma_zone_t mapzone; 128static uma_zone_t vmspace_zone; 129static int vmspace_zinit(void *mem, int size, int flags); 130static int vm_map_zinit(void *mem, int ize, int flags); 131static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, 132 vm_offset_t max); 133static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); 134static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); 135static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry); 136static int vm_map_growstack(vm_map_t map, vm_offset_t addr, 137 vm_map_entry_t gap_entry); 138static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 139 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags); 140#ifdef INVARIANTS 141static void vm_map_zdtor(void *mem, int size, void *arg); 142static void vmspace_zdtor(void *mem, int size, void *arg); 143#endif 144static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, 145 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, 146 int cow); 147static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 148 vm_offset_t failed_addr); 149 150#define ENTRY_CHARGED(e) ((e)->cred != NULL || \ 151 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ 152 !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) 153 154/* 155 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type 156 * stable. 157 */ 158#define PROC_VMSPACE_LOCK(p) do { } while (0) 159#define PROC_VMSPACE_UNLOCK(p) do { } while (0) 160 161/* 162 * VM_MAP_RANGE_CHECK: [ internal use only ] 163 * 164 * Asserts that the starting and ending region 165 * addresses fall within the valid range of the map. 166 */ 167#define VM_MAP_RANGE_CHECK(map, start, end) \ 168 { \ 169 if (start < vm_map_min(map)) \ 170 start = vm_map_min(map); \ 171 if (end > vm_map_max(map)) \ 172 end = vm_map_max(map); \ 173 if (start > end) \ 174 start = end; \ 175 } 176 177/* 178 * vm_map_startup: 179 * 180 * Initialize the vm_map module. Must be called before 181 * any other vm_map routines. 182 * 183 * Map and entry structures are allocated from the general 184 * purpose memory pool with some exceptions: 185 * 186 * - The kernel map and kmem submap are allocated statically. 187 * - Kernel map entries are allocated out of a static pool. 188 * 189 * These restrictions are necessary since malloc() uses the 190 * maps and requires map entries. 191 */ 192 193void 194vm_map_startup(void) 195{ 196 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); 197 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, 198#ifdef INVARIANTS 199 vm_map_zdtor, 200#else 201 NULL, 202#endif 203 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 204 uma_prealloc(mapzone, MAX_KMAP); 205 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), 206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 207 UMA_ZONE_MTXCLASS | UMA_ZONE_VM); 208 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), 209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 210 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, 211#ifdef INVARIANTS 212 vmspace_zdtor, 213#else 214 NULL, 215#endif 216 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 217} 218 219static int 220vmspace_zinit(void *mem, int size, int flags) 221{ 222 struct vmspace *vm; 223 224 vm = (struct vmspace *)mem; 225 226 vm->vm_map.pmap = NULL; 227 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); 228 PMAP_LOCK_INIT(vmspace_pmap(vm)); 229 return (0); 230} 231 232static int 233vm_map_zinit(void *mem, int size, int flags) 234{ 235 vm_map_t map; 236 237 map = (vm_map_t)mem; 238 memset(map, 0, sizeof(*map)); 239 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); 240 sx_init(&map->lock, "vm map (user)"); 241 return (0); 242} 243 244#ifdef INVARIANTS 245static void 246vmspace_zdtor(void *mem, int size, void *arg) 247{ 248 struct vmspace *vm; 249 250 vm = (struct vmspace *)mem; 251 252 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); 253} 254static void 255vm_map_zdtor(void *mem, int size, void *arg) 256{ 257 vm_map_t map; 258 259 map = (vm_map_t)mem; 260 KASSERT(map->nentries == 0, 261 ("map %p nentries == %d on free.", 262 map, map->nentries)); 263 KASSERT(map->size == 0, 264 ("map %p size == %lu on free.", 265 map, (unsigned long)map->size)); 266} 267#endif /* INVARIANTS */ 268 269/* 270 * Allocate a vmspace structure, including a vm_map and pmap, 271 * and initialize those structures. The refcnt is set to 1. 272 * 273 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). 274 */ 275struct vmspace * 276vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) 277{ 278 struct vmspace *vm; 279 280 vm = uma_zalloc(vmspace_zone, M_WAITOK); 281 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); 282 if (!pinit(vmspace_pmap(vm))) { 283 uma_zfree(vmspace_zone, vm); 284 return (NULL); 285 } 286 CTR1(KTR_VM, "vmspace_alloc: %p", vm); 287 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); 288 vm->vm_refcnt = 1; 289 vm->vm_shm = NULL; 290 vm->vm_swrss = 0; 291 vm->vm_tsize = 0; 292 vm->vm_dsize = 0; 293 vm->vm_ssize = 0; 294 vm->vm_taddr = 0; 295 vm->vm_daddr = 0; 296 vm->vm_maxsaddr = 0; 297 return (vm); 298} 299 300#ifdef RACCT 301static void 302vmspace_container_reset(struct proc *p) 303{ 304 305 PROC_LOCK(p); 306 racct_set(p, RACCT_DATA, 0); 307 racct_set(p, RACCT_STACK, 0); 308 racct_set(p, RACCT_RSS, 0); 309 racct_set(p, RACCT_MEMLOCK, 0); 310 racct_set(p, RACCT_VMEM, 0); 311 PROC_UNLOCK(p); 312} 313#endif 314 315static inline void 316vmspace_dofree(struct vmspace *vm) 317{ 318 319 CTR1(KTR_VM, "vmspace_free: %p", vm); 320 321 /* 322 * Make sure any SysV shm is freed, it might not have been in 323 * exit1(). 324 */ 325 shmexit(vm); 326 327 /* 328 * Lock the map, to wait out all other references to it. 329 * Delete all of the mappings and pages they hold, then call 330 * the pmap module to reclaim anything left. 331 */ 332 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map), 333 vm_map_max(&vm->vm_map)); 334 335 pmap_release(vmspace_pmap(vm)); 336 vm->vm_map.pmap = NULL; 337 uma_zfree(vmspace_zone, vm); 338} 339 340void 341vmspace_free(struct vmspace *vm) 342{ 343 344 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, 345 "vmspace_free() called"); 346 347 if (vm->vm_refcnt == 0) 348 panic("vmspace_free: attempt to free already freed vmspace"); 349 350 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) 351 vmspace_dofree(vm); 352} 353 354void 355vmspace_exitfree(struct proc *p) 356{ 357 struct vmspace *vm; 358 359 PROC_VMSPACE_LOCK(p); 360 vm = p->p_vmspace; 361 p->p_vmspace = NULL; 362 PROC_VMSPACE_UNLOCK(p); 363 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); 364 vmspace_free(vm); 365} 366 367void 368vmspace_exit(struct thread *td) 369{ 370 int refcnt; 371 struct vmspace *vm; 372 struct proc *p; 373 374 /* 375 * Release user portion of address space. 376 * This releases references to vnodes, 377 * which could cause I/O if the file has been unlinked. 378 * Need to do this early enough that we can still sleep. 379 * 380 * The last exiting process to reach this point releases as 381 * much of the environment as it can. vmspace_dofree() is the 382 * slower fallback in case another process had a temporary 383 * reference to the vmspace. 384 */ 385 386 p = td->td_proc; 387 vm = p->p_vmspace; 388 atomic_add_int(&vmspace0.vm_refcnt, 1); 389 do { 390 refcnt = vm->vm_refcnt; 391 if (refcnt > 1 && p->p_vmspace != &vmspace0) { 392 /* Switch now since other proc might free vmspace */ 393 PROC_VMSPACE_LOCK(p); 394 p->p_vmspace = &vmspace0; 395 PROC_VMSPACE_UNLOCK(p); 396 pmap_activate(td); 397 } 398 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); 399 if (refcnt == 1) { 400 if (p->p_vmspace != vm) { 401 /* vmspace not yet freed, switch back */ 402 PROC_VMSPACE_LOCK(p); 403 p->p_vmspace = vm; 404 PROC_VMSPACE_UNLOCK(p); 405 pmap_activate(td); 406 } 407 pmap_remove_pages(vmspace_pmap(vm)); 408 /* Switch now since this proc will free vmspace */ 409 PROC_VMSPACE_LOCK(p); 410 p->p_vmspace = &vmspace0; 411 PROC_VMSPACE_UNLOCK(p); 412 pmap_activate(td); 413 vmspace_dofree(vm); 414 } 415#ifdef RACCT 416 if (racct_enable) 417 vmspace_container_reset(p); 418#endif 419} 420 421/* Acquire reference to vmspace owned by another process. */ 422 423struct vmspace * 424vmspace_acquire_ref(struct proc *p) 425{ 426 struct vmspace *vm; 427 int refcnt; 428 429 PROC_VMSPACE_LOCK(p); 430 vm = p->p_vmspace; 431 if (vm == NULL) { 432 PROC_VMSPACE_UNLOCK(p); 433 return (NULL); 434 } 435 do { 436 refcnt = vm->vm_refcnt; 437 if (refcnt <= 0) { /* Avoid 0->1 transition */ 438 PROC_VMSPACE_UNLOCK(p); 439 return (NULL); 440 } 441 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); 442 if (vm != p->p_vmspace) { 443 PROC_VMSPACE_UNLOCK(p); 444 vmspace_free(vm); 445 return (NULL); 446 } 447 PROC_VMSPACE_UNLOCK(p); 448 return (vm); 449} 450 451/* 452 * Switch between vmspaces in an AIO kernel process. 453 * 454 * The AIO kernel processes switch to and from a user process's 455 * vmspace while performing an I/O operation on behalf of a user 456 * process. The new vmspace is either the vmspace of a user process 457 * obtained from an active AIO request or the initial vmspace of the 458 * AIO kernel process (when it is idling). Because user processes 459 * will block to drain any active AIO requests before proceeding in 460 * exit() or execve(), the vmspace reference count for these vmspaces 461 * can never be 0. This allows for a much simpler implementation than 462 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel 463 * processes hold an extra reference on their initial vmspace for the 464 * life of the process so that this guarantee is true for any vmspace 465 * passed as 'newvm'. 466 */ 467void 468vmspace_switch_aio(struct vmspace *newvm) 469{ 470 struct vmspace *oldvm; 471 472 /* XXX: Need some way to assert that this is an aio daemon. */ 473 474 KASSERT(newvm->vm_refcnt > 0, 475 ("vmspace_switch_aio: newvm unreferenced")); 476 477 oldvm = curproc->p_vmspace; 478 if (oldvm == newvm) 479 return; 480 481 /* 482 * Point to the new address space and refer to it. 483 */ 484 curproc->p_vmspace = newvm; 485 atomic_add_int(&newvm->vm_refcnt, 1); 486 487 /* Activate the new mapping. */ 488 pmap_activate(curthread); 489 490 /* Remove the daemon's reference to the old address space. */ 491 KASSERT(oldvm->vm_refcnt > 1, 492 ("vmspace_switch_aio: oldvm dropping last reference")); 493 vmspace_free(oldvm); 494} 495 496void 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 507static void 508vm_map_process_deferred(void) 509{ 510 struct thread *td; 511 vm_map_entry_t entry, next; 512 vm_object_t object; 513 514 td = curthread; 515 entry = td->td_map_def_user; 516 td->td_map_def_user = NULL; 517 while (entry != NULL) { 518 next = entry->next; 519 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { 520 /* 521 * Decrement the object's writemappings and 522 * possibly the vnode's v_writecount. 523 */ 524 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, 525 ("Submap with writecount")); 526 object = entry->object.vm_object; 527 KASSERT(object != NULL, ("No object for writecount")); 528 vnode_pager_release_writecount(object, entry->start, 529 entry->end); 530 } 531 vm_map_entry_deallocate(entry, FALSE); 532 entry = next; 533 } 534} 535 536void 537_vm_map_unlock(vm_map_t map, const char *file, int line) 538{ 539 540 if (map->system_map) 541 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 542 else { 543 sx_xunlock_(&map->lock, file, line); 544 vm_map_process_deferred(); 545 } 546} 547 548void 549_vm_map_lock_read(vm_map_t map, const char *file, int line) 550{ 551 552 if (map->system_map) 553 mtx_lock_flags_(&map->system_mtx, 0, file, line); 554 else 555 sx_slock_(&map->lock, file, line); 556} 557 558void 559_vm_map_unlock_read(vm_map_t map, const char *file, int line) 560{ 561 562 if (map->system_map) 563 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 564 else { 565 sx_sunlock_(&map->lock, file, line); 566 vm_map_process_deferred(); 567 } 568} 569 570int 571_vm_map_trylock(vm_map_t map, const char *file, int line) 572{ 573 int error; 574 575 error = map->system_map ? 576 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 577 !sx_try_xlock_(&map->lock, file, line); 578 if (error == 0) 579 map->timestamp++; 580 return (error == 0); 581} 582 583int 584_vm_map_trylock_read(vm_map_t map, const char *file, int line) 585{ 586 int error; 587 588 error = map->system_map ? 589 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : 590 !sx_try_slock_(&map->lock, file, line); 591 return (error == 0); 592} 593 594/* 595 * _vm_map_lock_upgrade: [ internal use only ] 596 * 597 * Tries to upgrade a read (shared) lock on the specified map to a write 598 * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a 599 * non-zero value if the upgrade fails. If the upgrade fails, the map is 600 * returned without a read or write lock held. 601 * 602 * Requires that the map be read locked. 603 */ 604int 605_vm_map_lock_upgrade(vm_map_t map, const char *file, int line) 606{ 607 unsigned int last_timestamp; 608 609 if (map->system_map) { 610 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 611 } else { 612 if (!sx_try_upgrade_(&map->lock, file, line)) { 613 last_timestamp = map->timestamp; 614 sx_sunlock_(&map->lock, file, line); 615 vm_map_process_deferred(); 616 /* 617 * If the map's timestamp does not change while the 618 * map is unlocked, then the upgrade succeeds. 619 */ 620 sx_xlock_(&map->lock, file, line); 621 if (last_timestamp != map->timestamp) { 622 sx_xunlock_(&map->lock, file, line); 623 return (1); 624 } 625 } 626 } 627 map->timestamp++; 628 return (0); 629} 630 631void 632_vm_map_lock_downgrade(vm_map_t map, const char *file, int line) 633{ 634 635 if (map->system_map) { 636 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 637 } else 638 sx_downgrade_(&map->lock, file, line); 639} 640 641/* 642 * vm_map_locked: 643 * 644 * Returns a non-zero value if the caller holds a write (exclusive) lock 645 * on the specified map and the value "0" otherwise. 646 */ 647int 648vm_map_locked(vm_map_t map) 649{ 650 651 if (map->system_map) 652 return (mtx_owned(&map->system_mtx)); 653 else 654 return (sx_xlocked(&map->lock)); 655} 656 657#ifdef INVARIANTS 658static void 659_vm_map_assert_locked(vm_map_t map, const char *file, int line) 660{ 661 662 if (map->system_map) 663 mtx_assert_(&map->system_mtx, MA_OWNED, file, line); 664 else 665 sx_assert_(&map->lock, SA_XLOCKED, file, line); 666} 667 668#define VM_MAP_ASSERT_LOCKED(map) \ 669 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) 670#else 671#define VM_MAP_ASSERT_LOCKED(map) 672#endif 673 674/* 675 * _vm_map_unlock_and_wait: 676 * 677 * Atomically releases the lock on the specified map and puts the calling 678 * thread to sleep. The calling thread will remain asleep until either 679 * vm_map_wakeup() is performed on the map or the specified timeout is 680 * exceeded. 681 * 682 * WARNING! This function does not perform deferred deallocations of 683 * objects and map entries. Therefore, the calling thread is expected to 684 * reacquire the map lock after reawakening and later perform an ordinary 685 * unlock operation, such as vm_map_unlock(), before completing its 686 * operation on the map. 687 */ 688int 689_vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) 690{ 691 692 mtx_lock(&map_sleep_mtx); 693 if (map->system_map) 694 mtx_unlock_flags_(&map->system_mtx, 0, file, line); 695 else 696 sx_xunlock_(&map->lock, file, line); 697 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 698 timo)); 699} 700 701/* 702 * vm_map_wakeup: 703 * 704 * Awaken any threads that have slept on the map using 705 * vm_map_unlock_and_wait(). 706 */ 707void 708vm_map_wakeup(vm_map_t map) 709{ 710 711 /* 712 * Acquire and release map_sleep_mtx to prevent a wakeup() 713 * from being performed (and lost) between the map unlock 714 * and the msleep() in _vm_map_unlock_and_wait(). 715 */ 716 mtx_lock(&map_sleep_mtx); 717 mtx_unlock(&map_sleep_mtx); 718 wakeup(&map->root); 719} 720 721void 722vm_map_busy(vm_map_t map) 723{ 724 725 VM_MAP_ASSERT_LOCKED(map); 726 map->busy++; 727} 728 729void 730vm_map_unbusy(vm_map_t map) 731{ 732 733 VM_MAP_ASSERT_LOCKED(map); 734 KASSERT(map->busy, ("vm_map_unbusy: not busy")); 735 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { 736 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); 737 wakeup(&map->busy); 738 } 739} 740 741void 742vm_map_wait_busy(vm_map_t map) 743{ 744 745 VM_MAP_ASSERT_LOCKED(map); 746 while (map->busy) { 747 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); 748 if (map->system_map) 749 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); 750 else 751 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); 752 } 753 map->timestamp++; 754} 755 756long 757vmspace_resident_count(struct vmspace *vmspace) 758{ 759 return pmap_resident_count(vmspace_pmap(vmspace)); 760} 761 762/* 763 * vm_map_create: 764 * 765 * Creates and returns a new empty VM map with 766 * the given physical map structure, and having 767 * the given lower and upper address bounds. 768 */ 769vm_map_t 770vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) 771{ 772 vm_map_t result; 773 774 result = uma_zalloc(mapzone, M_WAITOK); 775 CTR1(KTR_VM, "vm_map_create: %p", result); 776 _vm_map_init(result, pmap, min, max); 777 return (result); 778} 779 780/* 781 * Initialize an existing vm_map structure 782 * such as that in the vmspace structure. 783 */ 784static void 785_vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 786{ 787 788 map->header.next = map->header.prev = &map->header; 789 map->needs_wakeup = FALSE; 790 map->system_map = 0; 791 map->pmap = pmap; 792 map->header.end = min; 793 map->header.start = max; 794 map->flags = 0; 795 map->root = NULL; 796 map->timestamp = 0; 797 map->busy = 0; 798} 799 800void 801vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) 802{ 803 804 _vm_map_init(map, pmap, min, max); 805 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); 806 sx_init(&map->lock, "user map"); 807} 808 809/* 810 * vm_map_entry_dispose: [ internal use only ] 811 * 812 * Inverse of vm_map_entry_create. 813 */ 814static void 815vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) 816{ 817 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); 818} 819 820/* 821 * vm_map_entry_create: [ internal use only ] 822 * 823 * Allocates a VM map entry for insertion. 824 * No entry fields are filled in. 825 */ 826static vm_map_entry_t 827vm_map_entry_create(vm_map_t map) 828{ 829 vm_map_entry_t new_entry; 830 831 if (map->system_map) 832 new_entry = uma_zalloc(kmapentzone, M_NOWAIT); 833 else 834 new_entry = uma_zalloc(mapentzone, M_WAITOK); 835 if (new_entry == NULL) 836 panic("vm_map_entry_create: kernel resources exhausted"); 837 return (new_entry); 838} 839 840/* 841 * vm_map_entry_set_behavior: 842 * 843 * Set the expected access behavior, either normal, random, or 844 * sequential. 845 */ 846static inline void 847vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) 848{ 849 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | 850 (behavior & MAP_ENTRY_BEHAV_MASK); 851} 852 853/* 854 * vm_map_entry_set_max_free: 855 * 856 * Set the max_free field in a vm_map_entry. 857 */ 858static inline void 859vm_map_entry_set_max_free(vm_map_entry_t entry) 860{ 861 862 entry->max_free = entry->adj_free; 863 if (entry->left != NULL && entry->left->max_free > entry->max_free) 864 entry->max_free = entry->left->max_free; 865 if (entry->right != NULL && entry->right->max_free > entry->max_free) 866 entry->max_free = entry->right->max_free; 867} 868 869/* 870 * vm_map_entry_splay: 871 * 872 * The Sleator and Tarjan top-down splay algorithm with the 873 * following variation. Max_free must be computed bottom-up, so 874 * on the downward pass, maintain the left and right spines in 875 * reverse order. Then, make a second pass up each side to fix 876 * the pointers and compute max_free. The time bound is O(log n) 877 * amortized. 878 * 879 * The new root is the vm_map_entry containing "addr", or else an 880 * adjacent entry (lower or higher) if addr is not in the tree. 881 * 882 * The map must be locked, and leaves it so. 883 * 884 * Returns: the new root. 885 */ 886static vm_map_entry_t 887vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) 888{ 889 vm_map_entry_t llist, rlist; 890 vm_map_entry_t ltree, rtree; 891 vm_map_entry_t y; 892 893 /* Special case of empty tree. */ 894 if (root == NULL) 895 return (root); 896 897 /* 898 * Pass One: Splay down the tree until we find addr or a NULL 899 * pointer where addr would go. llist and rlist are the two 900 * sides in reverse order (bottom-up), with llist linked by 901 * the right pointer and rlist linked by the left pointer in 902 * the vm_map_entry. Wait until Pass Two to set max_free on 903 * the two spines. 904 */ 905 llist = NULL; 906 rlist = NULL; 907 for (;;) { 908 /* root is never NULL in here. */ 909 if (addr < root->start) { 910 y = root->left; 911 if (y == NULL) 912 break; 913 if (addr < y->start && y->left != NULL) { 914 /* Rotate right and put y on rlist. */ 915 root->left = y->right; 916 y->right = root; 917 vm_map_entry_set_max_free(root); 918 root = y->left; 919 y->left = rlist; 920 rlist = y; 921 } else { 922 /* Put root on rlist. */ 923 root->left = rlist; 924 rlist = root; 925 root = y; 926 } 927 } else if (addr >= root->end) { 928 y = root->right; 929 if (y == NULL) 930 break; 931 if (addr >= y->end && y->right != NULL) { 932 /* Rotate left and put y on llist. */ 933 root->right = y->left; 934 y->left = root; 935 vm_map_entry_set_max_free(root); 936 root = y->right; 937 y->right = llist; 938 llist = y; 939 } else { 940 /* Put root on llist. */ 941 root->right = llist; 942 llist = root; 943 root = y; 944 } 945 } else 946 break; 947 } 948 949 /* 950 * Pass Two: Walk back up the two spines, flip the pointers 951 * and set max_free. The subtrees of the root go at the 952 * bottom of llist and rlist. 953 */ 954 ltree = root->left; 955 while (llist != NULL) { 956 y = llist->right; 957 llist->right = ltree; 958 vm_map_entry_set_max_free(llist); 959 ltree = llist; 960 llist = y; 961 } 962 rtree = root->right; 963 while (rlist != NULL) { 964 y = rlist->left; 965 rlist->left = rtree; 966 vm_map_entry_set_max_free(rlist); 967 rtree = rlist; 968 rlist = y; 969 } 970 971 /* 972 * Final assembly: add ltree and rtree as subtrees of root. 973 */ 974 root->left = ltree; 975 root->right = rtree; 976 vm_map_entry_set_max_free(root); 977 978 return (root); 979} 980 981/* 982 * vm_map_entry_{un,}link: 983 * 984 * Insert/remove entries from maps. 985 */ 986static void 987vm_map_entry_link(vm_map_t map, 988 vm_map_entry_t after_where, 989 vm_map_entry_t entry) 990{ 991 992 CTR4(KTR_VM, 993 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, 994 map->nentries, entry, after_where); 995 VM_MAP_ASSERT_LOCKED(map); 996 KASSERT(after_where->end <= entry->start, 997 ("vm_map_entry_link: prev end %jx new start %jx overlap", 998 (uintmax_t)after_where->end, (uintmax_t)entry->start)); 999 KASSERT(entry->end <= after_where->next->start, 1000 ("vm_map_entry_link: new end %jx next start %jx overlap", 1001 (uintmax_t)entry->end, (uintmax_t)after_where->next->start)); 1002 1003 map->nentries++; 1004 entry->prev = after_where; 1005 entry->next = after_where->next; 1006 entry->next->prev = entry; 1007 after_where->next = entry; 1008 1009 if (after_where != &map->header) { 1010 if (after_where != map->root) 1011 vm_map_entry_splay(after_where->start, map->root); 1012 entry->right = after_where->right; 1013 entry->left = after_where; 1014 after_where->right = NULL; 1015 after_where->adj_free = entry->start - after_where->end; 1016 vm_map_entry_set_max_free(after_where); 1017 } else { 1018 entry->right = map->root; 1019 entry->left = NULL; 1020 } 1021 entry->adj_free = entry->next->start - entry->end; 1022 vm_map_entry_set_max_free(entry); 1023 map->root = entry; 1024} 1025 1026static void 1027vm_map_entry_unlink(vm_map_t map, 1028 vm_map_entry_t entry) 1029{ 1030 vm_map_entry_t next, prev, root; 1031 1032 VM_MAP_ASSERT_LOCKED(map); 1033 if (entry != map->root) 1034 vm_map_entry_splay(entry->start, map->root); 1035 if (entry->left == NULL) 1036 root = entry->right; 1037 else { 1038 root = vm_map_entry_splay(entry->start, entry->left); 1039 root->right = entry->right; 1040 root->adj_free = entry->next->start - root->end; 1041 vm_map_entry_set_max_free(root); 1042 } 1043 map->root = root; 1044 1045 prev = entry->prev; 1046 next = entry->next; 1047 next->prev = prev; 1048 prev->next = next; 1049 map->nentries--; 1050 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, 1051 map->nentries, entry); 1052} 1053 1054/* 1055 * vm_map_entry_resize_free: 1056 * 1057 * Recompute the amount of free space following a vm_map_entry 1058 * and propagate that value up the tree. Call this function after 1059 * resizing a map entry in-place, that is, without a call to 1060 * vm_map_entry_link() or _unlink(). 1061 * 1062 * The map must be locked, and leaves it so. 1063 */ 1064static void 1065vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) 1066{ 1067 1068 /* 1069 * Using splay trees without parent pointers, propagating 1070 * max_free up the tree is done by moving the entry to the 1071 * root and making the change there. 1072 */ 1073 if (entry != map->root) 1074 map->root = vm_map_entry_splay(entry->start, map->root); 1075 1076 entry->adj_free = entry->next->start - entry->end; 1077 vm_map_entry_set_max_free(entry); 1078} 1079 1080/* 1081 * vm_map_lookup_entry: [ internal use only ] 1082 * 1083 * Finds the map entry containing (or 1084 * immediately preceding) the specified address 1085 * in the given map; the entry is returned 1086 * in the "entry" parameter. The boolean 1087 * result indicates whether the address is 1088 * actually contained in the map. 1089 */ 1090boolean_t 1091vm_map_lookup_entry( 1092 vm_map_t map, 1093 vm_offset_t address, 1094 vm_map_entry_t *entry) /* OUT */ 1095{ 1096 vm_map_entry_t cur; 1097 boolean_t locked; 1098 1099 /* 1100 * If the map is empty, then the map entry immediately preceding 1101 * "address" is the map's header. 1102 */ 1103 cur = map->root; 1104 if (cur == NULL) 1105 *entry = &map->header; 1106 else if (address >= cur->start && cur->end > address) { 1107 *entry = cur; 1108 return (TRUE); 1109 } else if ((locked = vm_map_locked(map)) || 1110 sx_try_upgrade(&map->lock)) { 1111 /* 1112 * Splay requires a write lock on the map. However, it only 1113 * restructures the binary search tree; it does not otherwise 1114 * change the map. Thus, the map's timestamp need not change 1115 * on a temporary upgrade. 1116 */ 1117 map->root = cur = vm_map_entry_splay(address, cur); 1118 if (!locked) 1119 sx_downgrade(&map->lock); 1120 1121 /* 1122 * If "address" is contained within a map entry, the new root 1123 * is that map entry. Otherwise, the new root is a map entry 1124 * immediately before or after "address". 1125 */ 1126 if (address >= cur->start) { 1127 *entry = cur; 1128 if (cur->end > address) 1129 return (TRUE); 1130 } else 1131 *entry = cur->prev; 1132 } else 1133 /* 1134 * Since the map is only locked for read access, perform a 1135 * standard binary search tree lookup for "address". 1136 */ 1137 for (;;) { 1138 if (address < cur->start) { 1139 if (cur->left == NULL) { 1140 *entry = cur->prev; 1141 break; 1142 } 1143 cur = cur->left; 1144 } else if (cur->end > address) { 1145 *entry = cur; 1146 return (TRUE); 1147 } else { 1148 if (cur->right == NULL) { 1149 *entry = cur; 1150 break; 1151 } 1152 cur = cur->right; 1153 } 1154 } 1155 return (FALSE); 1156} 1157 1158/* 1159 * vm_map_insert: 1160 * 1161 * Inserts the given whole VM object into the target 1162 * map at the specified address range. The object's 1163 * size should match that of the address range. 1164 * 1165 * Requires that the map be locked, and leaves it so. 1166 * 1167 * If object is non-NULL, ref count must be bumped by caller 1168 * prior to making call to account for the new entry. 1169 */ 1170int 1171vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1172 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow) 1173{ 1174 vm_map_entry_t new_entry, prev_entry, temp_entry; 1175 struct ucred *cred; 1176 vm_eflags_t protoeflags; 1177 vm_inherit_t inheritance; 1178 1179 VM_MAP_ASSERT_LOCKED(map); 1180 KASSERT((object != kmem_object && object != kernel_object) || 1181 (cow & MAP_COPY_ON_WRITE) == 0, 1182 ("vm_map_insert: kmem or kernel object and COW")); 1183 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0, 1184 ("vm_map_insert: paradoxical MAP_NOFAULT request")); 1185 KASSERT((prot & ~max) == 0, 1186 ("prot %#x is not subset of max_prot %#x", prot, max)); 1187 1188 /* 1189 * Check that the start and end points are not bogus. 1190 */ 1191 if (start < vm_map_min(map) || end > vm_map_max(map) || 1192 start >= end) 1193 return (KERN_INVALID_ADDRESS); 1194 1195 /* 1196 * Find the entry prior to the proposed starting address; if it's part 1197 * of an existing entry, this range is bogus. 1198 */ 1199 if (vm_map_lookup_entry(map, start, &temp_entry)) 1200 return (KERN_NO_SPACE); 1201 1202 prev_entry = temp_entry; 1203 1204 /* 1205 * Assert that the next entry doesn't overlap the end point. 1206 */ 1207 if (prev_entry->next->start < end) 1208 return (KERN_NO_SPACE); 1209 1210 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL || 1211 max != VM_PROT_NONE)) 1212 return (KERN_INVALID_ARGUMENT); 1213 1214 protoeflags = 0; 1215 if (cow & MAP_COPY_ON_WRITE) 1216 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; 1217 if (cow & MAP_NOFAULT) 1218 protoeflags |= MAP_ENTRY_NOFAULT; 1219 if (cow & MAP_DISABLE_SYNCER) 1220 protoeflags |= MAP_ENTRY_NOSYNC; 1221 if (cow & MAP_DISABLE_COREDUMP) 1222 protoeflags |= MAP_ENTRY_NOCOREDUMP; 1223 if (cow & MAP_STACK_GROWS_DOWN) 1224 protoeflags |= MAP_ENTRY_GROWS_DOWN; 1225 if (cow & MAP_STACK_GROWS_UP) 1226 protoeflags |= MAP_ENTRY_GROWS_UP; 1227 if (cow & MAP_VN_WRITECOUNT) 1228 protoeflags |= MAP_ENTRY_VN_WRITECNT; 1229 if ((cow & MAP_CREATE_GUARD) != 0) 1230 protoeflags |= MAP_ENTRY_GUARD; 1231 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0) 1232 protoeflags |= MAP_ENTRY_STACK_GAP_DN; 1233 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0) 1234 protoeflags |= MAP_ENTRY_STACK_GAP_UP; 1235 if (cow & MAP_INHERIT_SHARE) 1236 inheritance = VM_INHERIT_SHARE; 1237 else 1238 inheritance = VM_INHERIT_DEFAULT; 1239 1240 cred = NULL; 1241 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0) 1242 goto charged; 1243 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && 1244 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { 1245 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) 1246 return (KERN_RESOURCE_SHORTAGE); 1247 KASSERT(object == NULL || 1248 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 || 1249 object->cred == NULL, 1250 ("overcommit: vm_map_insert o %p", object)); 1251 cred = curthread->td_ucred; 1252 } 1253 1254charged: 1255 /* Expand the kernel pmap, if necessary. */ 1256 if (map == kernel_map && end > kernel_vm_end) 1257 pmap_growkernel(end); 1258 if (object != NULL) { 1259 /* 1260 * OBJ_ONEMAPPING must be cleared unless this mapping 1261 * is trivially proven to be the only mapping for any 1262 * of the object's pages. (Object granularity 1263 * reference counting is insufficient to recognize 1264 * aliases with precision.) 1265 */ 1266 VM_OBJECT_WLOCK(object); 1267 if (object->ref_count > 1 || object->shadow_count != 0) 1268 vm_object_clear_flag(object, OBJ_ONEMAPPING); 1269 VM_OBJECT_WUNLOCK(object); 1270 } else if (prev_entry != &map->header && 1271 prev_entry->eflags == protoeflags && 1272 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 && 1273 prev_entry->end == start && prev_entry->wired_count == 0 && 1274 (prev_entry->cred == cred || 1275 (prev_entry->object.vm_object != NULL && 1276 prev_entry->object.vm_object->cred == cred)) && 1277 vm_object_coalesce(prev_entry->object.vm_object, 1278 prev_entry->offset, 1279 (vm_size_t)(prev_entry->end - prev_entry->start), 1280 (vm_size_t)(end - prev_entry->end), cred != NULL && 1281 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) { 1282 /* 1283 * We were able to extend the object. Determine if we 1284 * can extend the previous map entry to include the 1285 * new range as well. 1286 */ 1287 if (prev_entry->inheritance == inheritance && 1288 prev_entry->protection == prot && 1289 prev_entry->max_protection == max) { 1290 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0) 1291 map->size += end - prev_entry->end; 1292 prev_entry->end = end; 1293 vm_map_entry_resize_free(map, prev_entry); 1294 vm_map_simplify_entry(map, prev_entry); 1295 return (KERN_SUCCESS); 1296 } 1297 1298 /* 1299 * If we can extend the object but cannot extend the 1300 * map entry, we have to create a new map entry. We 1301 * must bump the ref count on the extended object to 1302 * account for it. object may be NULL. 1303 */ 1304 object = prev_entry->object.vm_object; 1305 offset = prev_entry->offset + 1306 (prev_entry->end - prev_entry->start); 1307 vm_object_reference(object); 1308 if (cred != NULL && object != NULL && object->cred != NULL && 1309 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 1310 /* Object already accounts for this uid. */ 1311 cred = NULL; 1312 } 1313 } 1314 if (cred != NULL) 1315 crhold(cred); 1316 1317 /* 1318 * Create a new entry 1319 */ 1320 new_entry = vm_map_entry_create(map); 1321 new_entry->start = start; 1322 new_entry->end = end; 1323 new_entry->cred = NULL; 1324 1325 new_entry->eflags = protoeflags; 1326 new_entry->object.vm_object = object; 1327 new_entry->offset = offset; 1328 1329 new_entry->inheritance = inheritance; 1330 new_entry->protection = prot; 1331 new_entry->max_protection = max; 1332 new_entry->wired_count = 0; 1333 new_entry->wiring_thread = NULL; 1334 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; 1335 new_entry->next_read = start; 1336 1337 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), 1338 ("overcommit: vm_map_insert leaks vm_map %p", new_entry)); 1339 new_entry->cred = cred; 1340 1341 /* 1342 * Insert the new entry into the list 1343 */ 1344 vm_map_entry_link(map, prev_entry, new_entry); 1345 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0) 1346 map->size += new_entry->end - new_entry->start; 1347 1348 /* 1349 * Try to coalesce the new entry with both the previous and next 1350 * entries in the list. Previously, we only attempted to coalesce 1351 * with the previous entry when object is NULL. Here, we handle the 1352 * other cases, which are less common. 1353 */ 1354 vm_map_simplify_entry(map, new_entry); 1355 1356 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) { 1357 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset), 1358 end - start, cow & MAP_PREFAULT_PARTIAL); 1359 } 1360 1361 return (KERN_SUCCESS); 1362} 1363 1364/* 1365 * vm_map_findspace: 1366 * 1367 * Find the first fit (lowest VM address) for "length" free bytes 1368 * beginning at address >= start in the given map. 1369 * 1370 * In a vm_map_entry, "adj_free" is the amount of free space 1371 * adjacent (higher address) to this entry, and "max_free" is the 1372 * maximum amount of contiguous free space in its subtree. This 1373 * allows finding a free region in one path down the tree, so 1374 * O(log n) amortized with splay trees. 1375 * 1376 * The map must be locked, and leaves it so. 1377 * 1378 * Returns: 0 on success, and starting address in *addr, 1379 * 1 if insufficient space. 1380 */ 1381int 1382vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, 1383 vm_offset_t *addr) /* OUT */ 1384{ 1385 vm_map_entry_t entry; 1386 vm_offset_t st; 1387 1388 /* 1389 * Request must fit within min/max VM address and must avoid 1390 * address wrap. 1391 */ 1392 start = MAX(start, vm_map_min(map)); 1393 if (start + length > vm_map_max(map) || start + length < start) 1394 return (1); 1395 1396 /* Empty tree means wide open address space. */ 1397 if (map->root == NULL) { 1398 *addr = start; 1399 return (0); 1400 } 1401 1402 /* 1403 * After splay, if start comes before root node, then there 1404 * must be a gap from start to the root. 1405 */ 1406 map->root = vm_map_entry_splay(start, map->root); 1407 if (start + length <= map->root->start) { 1408 *addr = start; 1409 return (0); 1410 } 1411 1412 /* 1413 * Root is the last node that might begin its gap before 1414 * start, and this is the last comparison where address 1415 * wrap might be a problem. 1416 */ 1417 st = (start > map->root->end) ? start : map->root->end; 1418 if (length <= map->root->end + map->root->adj_free - st) { 1419 *addr = st; 1420 return (0); 1421 } 1422 1423 /* With max_free, can immediately tell if no solution. */ 1424 entry = map->root->right; 1425 if (entry == NULL || length > entry->max_free) 1426 return (1); 1427 1428 /* 1429 * Search the right subtree in the order: left subtree, root, 1430 * right subtree (first fit). The previous splay implies that 1431 * all regions in the right subtree have addresses > start. 1432 */ 1433 while (entry != NULL) { 1434 if (entry->left != NULL && entry->left->max_free >= length) 1435 entry = entry->left; 1436 else if (entry->adj_free >= length) { 1437 *addr = entry->end; 1438 return (0); 1439 } else 1440 entry = entry->right; 1441 } 1442 1443 /* Can't get here, so panic if we do. */ 1444 panic("vm_map_findspace: max_free corrupt"); 1445} 1446 1447int 1448vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1449 vm_offset_t start, vm_size_t length, vm_prot_t prot, 1450 vm_prot_t max, int cow) 1451{ 1452 vm_offset_t end; 1453 int result; 1454 1455 end = start + length; 1456 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1457 object == NULL, 1458 ("vm_map_fixed: non-NULL backing object for stack")); 1459 vm_map_lock(map); 1460 VM_MAP_RANGE_CHECK(map, start, end); 1461 if ((cow & MAP_CHECK_EXCL) == 0) 1462 vm_map_delete(map, start, end); 1463 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1464 result = vm_map_stack_locked(map, start, length, sgrowsiz, 1465 prot, max, cow); 1466 } else { 1467 result = vm_map_insert(map, object, offset, start, end, 1468 prot, max, cow); 1469 } 1470 vm_map_unlock(map); 1471 return (result); 1472} 1473 1474/* 1475 * vm_map_find finds an unallocated region in the target address 1476 * map with the given length. The search is defined to be 1477 * first-fit from the specified address; the region found is 1478 * returned in the same parameter. 1479 * 1480 * If object is non-NULL, ref count must be bumped by caller 1481 * prior to making call to account for the new entry. 1482 */ 1483int 1484vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1485 vm_offset_t *addr, /* IN/OUT */ 1486 vm_size_t length, vm_offset_t max_addr, int find_space, 1487 vm_prot_t prot, vm_prot_t max, int cow) 1488{ 1489 vm_offset_t alignment, initial_addr, start; 1490 int result; 1491 1492 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || 1493 object == NULL, 1494 ("vm_map_find: non-NULL backing object for stack")); 1495 MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE && 1496 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)); 1497 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || 1498 (object->flags & OBJ_COLORED) == 0)) 1499 find_space = VMFS_ANY_SPACE; 1500 if (find_space >> 8 != 0) { 1501 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); 1502 alignment = (vm_offset_t)1 << (find_space >> 8); 1503 } else 1504 alignment = 0; 1505 initial_addr = *addr; 1506again: 1507 start = initial_addr; 1508 vm_map_lock(map); 1509 do { 1510 if (find_space != VMFS_NO_SPACE) { 1511 if (vm_map_findspace(map, start, length, addr) || 1512 (max_addr != 0 && *addr + length > max_addr)) { 1513 vm_map_unlock(map); 1514 if (find_space == VMFS_OPTIMAL_SPACE) { 1515 find_space = VMFS_ANY_SPACE; 1516 goto again; 1517 } 1518 return (KERN_NO_SPACE); 1519 } 1520 switch (find_space) { 1521 case VMFS_SUPER_SPACE: 1522 case VMFS_OPTIMAL_SPACE: 1523 pmap_align_superpage(object, offset, addr, 1524 length); 1525 break; 1526 case VMFS_ANY_SPACE: 1527 break; 1528 default: 1529 if ((*addr & (alignment - 1)) != 0) { 1530 *addr &= ~(alignment - 1); 1531 *addr += alignment; 1532 } 1533 break; 1534 } 1535 1536 start = *addr; 1537 } else if ((cow & MAP_REMAP) != 0) { 1538 if (start < vm_map_min(map) || 1539 start + length > vm_map_max(map) || 1540 start + length <= length) { 1541 result = KERN_INVALID_ADDRESS; 1542 break; 1543 } 1544 vm_map_delete(map, start, start + length); 1545 } 1546 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { 1547 result = vm_map_stack_locked(map, start, length, 1548 sgrowsiz, prot, max, cow); 1549 } else { 1550 result = vm_map_insert(map, object, offset, start, 1551 start + length, prot, max, cow); 1552 } 1553 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && 1554 find_space != VMFS_ANY_SPACE); 1555 vm_map_unlock(map); 1556 return (result); 1557} 1558 1559/* 1560 * vm_map_find_min() is a variant of vm_map_find() that takes an 1561 * additional parameter (min_addr) and treats the given address 1562 * (*addr) differently. Specifically, it treats *addr as a hint 1563 * and not as the minimum address where the mapping is created. 1564 * 1565 * This function works in two phases. First, it tries to 1566 * allocate above the hint. If that fails and the hint is 1567 * greater than min_addr, it performs a second pass, replacing 1568 * the hint with min_addr as the minimum address for the 1569 * allocation. 1570 */ 1571int 1572vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 1573 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr, 1574 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max, 1575 int cow) 1576{ 1577 vm_offset_t hint; 1578 int rv; 1579 1580 hint = *addr; 1581 for (;;) { 1582 rv = vm_map_find(map, object, offset, addr, length, max_addr, 1583 find_space, prot, max, cow); 1584 if (rv == KERN_SUCCESS || min_addr >= hint) 1585 return (rv); 1586 *addr = hint = min_addr; 1587 } 1588} 1589 1590/* 1591 * vm_map_simplify_entry: 1592 * 1593 * Simplify the given map entry by merging with either neighbor. This 1594 * routine also has the ability to merge with both neighbors. 1595 * 1596 * The map must be locked. 1597 * 1598 * This routine guarantees that the passed entry remains valid (though 1599 * possibly extended). When merging, this routine may delete one or 1600 * both neighbors. 1601 */ 1602void 1603vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) 1604{ 1605 vm_map_entry_t next, prev; 1606 vm_size_t prevsize, esize; 1607 1608 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | 1609 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0) 1610 return; 1611 1612 prev = entry->prev; 1613 if (prev != &map->header) { 1614 prevsize = prev->end - prev->start; 1615 if ( (prev->end == entry->start) && 1616 (prev->object.vm_object == entry->object.vm_object) && 1617 (!prev->object.vm_object || 1618 (prev->offset + prevsize == entry->offset)) && 1619 (prev->eflags == entry->eflags) && 1620 (prev->protection == entry->protection) && 1621 (prev->max_protection == entry->max_protection) && 1622 (prev->inheritance == entry->inheritance) && 1623 (prev->wired_count == entry->wired_count) && 1624 (prev->cred == entry->cred)) { 1625 vm_map_entry_unlink(map, prev); 1626 entry->start = prev->start; 1627 entry->offset = prev->offset; 1628 if (entry->prev != &map->header) 1629 vm_map_entry_resize_free(map, entry->prev); 1630 1631 /* 1632 * If the backing object is a vnode object, 1633 * vm_object_deallocate() calls vrele(). 1634 * However, vrele() does not lock the vnode 1635 * because the vnode has additional 1636 * references. Thus, the map lock can be kept 1637 * without causing a lock-order reversal with 1638 * the vnode lock. 1639 * 1640 * Since we count the number of virtual page 1641 * mappings in object->un_pager.vnp.writemappings, 1642 * the writemappings value should not be adjusted 1643 * when the entry is disposed of. 1644 */ 1645 if (prev->object.vm_object) 1646 vm_object_deallocate(prev->object.vm_object); 1647 if (prev->cred != NULL) 1648 crfree(prev->cred); 1649 vm_map_entry_dispose(map, prev); 1650 } 1651 } 1652 1653 next = entry->next; 1654 if (next != &map->header) { 1655 esize = entry->end - entry->start; 1656 if ((entry->end == next->start) && 1657 (next->object.vm_object == entry->object.vm_object) && 1658 (!entry->object.vm_object || 1659 (entry->offset + esize == next->offset)) && 1660 (next->eflags == entry->eflags) && 1661 (next->protection == entry->protection) && 1662 (next->max_protection == entry->max_protection) && 1663 (next->inheritance == entry->inheritance) && 1664 (next->wired_count == entry->wired_count) && 1665 (next->cred == entry->cred)) { 1666 vm_map_entry_unlink(map, next); 1667 entry->end = next->end; 1668 vm_map_entry_resize_free(map, entry); 1669 1670 /* 1671 * See comment above. 1672 */ 1673 if (next->object.vm_object) 1674 vm_object_deallocate(next->object.vm_object); 1675 if (next->cred != NULL) 1676 crfree(next->cred); 1677 vm_map_entry_dispose(map, next); 1678 } 1679 } 1680} 1681/* 1682 * vm_map_clip_start: [ internal use only ] 1683 * 1684 * Asserts that the given entry begins at or after 1685 * the specified address; if necessary, 1686 * it splits the entry into two. 1687 */ 1688#define vm_map_clip_start(map, entry, startaddr) \ 1689{ \ 1690 if (startaddr > entry->start) \ 1691 _vm_map_clip_start(map, entry, startaddr); \ 1692} 1693 1694/* 1695 * This routine is called only when it is known that 1696 * the entry must be split. 1697 */ 1698static void 1699_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) 1700{ 1701 vm_map_entry_t new_entry; 1702 1703 VM_MAP_ASSERT_LOCKED(map); 1704 KASSERT(entry->end > start && entry->start < start, 1705 ("_vm_map_clip_start: invalid clip of entry %p", entry)); 1706 1707 /* 1708 * Split off the front portion -- note that we must insert the new 1709 * entry BEFORE this one, so that this entry has the specified 1710 * starting address. 1711 */ 1712 vm_map_simplify_entry(map, entry); 1713 1714 /* 1715 * If there is no object backing this entry, we might as well create 1716 * one now. If we defer it, an object can get created after the map 1717 * is clipped, and individual objects will be created for the split-up 1718 * map. This is a bit of a hack, but is also about the best place to 1719 * put this improvement. 1720 */ 1721 if (entry->object.vm_object == NULL && !map->system_map && 1722 (entry->eflags & MAP_ENTRY_GUARD) == 0) { 1723 vm_object_t object; 1724 object = vm_object_allocate(OBJT_DEFAULT, 1725 atop(entry->end - entry->start)); 1726 entry->object.vm_object = object; 1727 entry->offset = 0; 1728 if (entry->cred != NULL) { 1729 object->cred = entry->cred; 1730 object->charge = entry->end - entry->start; 1731 entry->cred = NULL; 1732 } 1733 } else if (entry->object.vm_object != NULL && 1734 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1735 entry->cred != NULL) { 1736 VM_OBJECT_WLOCK(entry->object.vm_object); 1737 KASSERT(entry->object.vm_object->cred == NULL, 1738 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); 1739 entry->object.vm_object->cred = entry->cred; 1740 entry->object.vm_object->charge = entry->end - entry->start; 1741 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1742 entry->cred = NULL; 1743 } 1744 1745 new_entry = vm_map_entry_create(map); 1746 *new_entry = *entry; 1747 1748 new_entry->end = start; 1749 entry->offset += (start - entry->start); 1750 entry->start = start; 1751 if (new_entry->cred != NULL) 1752 crhold(entry->cred); 1753 1754 vm_map_entry_link(map, entry->prev, new_entry); 1755 1756 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1757 vm_object_reference(new_entry->object.vm_object); 1758 /* 1759 * The object->un_pager.vnp.writemappings for the 1760 * object of MAP_ENTRY_VN_WRITECNT type entry shall be 1761 * kept as is here. The virtual pages are 1762 * re-distributed among the clipped entries, so the sum is 1763 * left the same. 1764 */ 1765 } 1766} 1767 1768/* 1769 * vm_map_clip_end: [ internal use only ] 1770 * 1771 * Asserts that the given entry ends at or before 1772 * the specified address; if necessary, 1773 * it splits the entry into two. 1774 */ 1775#define vm_map_clip_end(map, entry, endaddr) \ 1776{ \ 1777 if ((endaddr) < (entry->end)) \ 1778 _vm_map_clip_end((map), (entry), (endaddr)); \ 1779} 1780 1781/* 1782 * This routine is called only when it is known that 1783 * the entry must be split. 1784 */ 1785static void 1786_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) 1787{ 1788 vm_map_entry_t new_entry; 1789 1790 VM_MAP_ASSERT_LOCKED(map); 1791 KASSERT(entry->start < end && entry->end > end, 1792 ("_vm_map_clip_end: invalid clip of entry %p", entry)); 1793 1794 /* 1795 * If there is no object backing this entry, we might as well create 1796 * one now. If we defer it, an object can get created after the map 1797 * is clipped, and individual objects will be created for the split-up 1798 * map. This is a bit of a hack, but is also about the best place to 1799 * put this improvement. 1800 */ 1801 if (entry->object.vm_object == NULL && !map->system_map && 1802 (entry->eflags & MAP_ENTRY_GUARD) == 0) { 1803 vm_object_t object; 1804 object = vm_object_allocate(OBJT_DEFAULT, 1805 atop(entry->end - entry->start)); 1806 entry->object.vm_object = object; 1807 entry->offset = 0; 1808 if (entry->cred != NULL) { 1809 object->cred = entry->cred; 1810 object->charge = entry->end - entry->start; 1811 entry->cred = NULL; 1812 } 1813 } else if (entry->object.vm_object != NULL && 1814 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && 1815 entry->cred != NULL) { 1816 VM_OBJECT_WLOCK(entry->object.vm_object); 1817 KASSERT(entry->object.vm_object->cred == NULL, 1818 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); 1819 entry->object.vm_object->cred = entry->cred; 1820 entry->object.vm_object->charge = entry->end - entry->start; 1821 VM_OBJECT_WUNLOCK(entry->object.vm_object); 1822 entry->cred = NULL; 1823 } 1824 1825 /* 1826 * Create a new entry and insert it AFTER the specified entry 1827 */ 1828 new_entry = vm_map_entry_create(map); 1829 *new_entry = *entry; 1830 1831 new_entry->start = entry->end = end; 1832 new_entry->offset += (end - entry->start); 1833 if (new_entry->cred != NULL) 1834 crhold(entry->cred); 1835 1836 vm_map_entry_link(map, entry, new_entry); 1837 1838 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { 1839 vm_object_reference(new_entry->object.vm_object); 1840 } 1841} 1842 1843/* 1844 * vm_map_submap: [ kernel use only ] 1845 * 1846 * Mark the given range as handled by a subordinate map. 1847 * 1848 * This range must have been created with vm_map_find, 1849 * and no other operations may have been performed on this 1850 * range prior to calling vm_map_submap. 1851 * 1852 * Only a limited number of operations can be performed 1853 * within this rage after calling vm_map_submap: 1854 * vm_fault 1855 * [Don't try vm_map_copy!] 1856 * 1857 * To remove a submapping, one must first remove the 1858 * range from the superior map, and then destroy the 1859 * submap (if desired). [Better yet, don't try it.] 1860 */ 1861int 1862vm_map_submap( 1863 vm_map_t map, 1864 vm_offset_t start, 1865 vm_offset_t end, 1866 vm_map_t submap) 1867{ 1868 vm_map_entry_t entry; 1869 int result = KERN_INVALID_ARGUMENT; 1870 1871 vm_map_lock(map); 1872 1873 VM_MAP_RANGE_CHECK(map, start, end); 1874 1875 if (vm_map_lookup_entry(map, start, &entry)) { 1876 vm_map_clip_start(map, entry, start); 1877 } else 1878 entry = entry->next; 1879 1880 vm_map_clip_end(map, entry, end); 1881 1882 if ((entry->start == start) && (entry->end == end) && 1883 ((entry->eflags & MAP_ENTRY_COW) == 0) && 1884 (entry->object.vm_object == NULL)) { 1885 entry->object.sub_map = submap; 1886 entry->eflags |= MAP_ENTRY_IS_SUB_MAP; 1887 result = KERN_SUCCESS; 1888 } 1889 vm_map_unlock(map); 1890 1891 return (result); 1892} 1893 1894/* 1895 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified 1896 */ 1897#define MAX_INIT_PT 96 1898 1899/* 1900 * vm_map_pmap_enter: 1901 * 1902 * Preload the specified map's pmap with mappings to the specified 1903 * object's memory-resident pages. No further physical pages are 1904 * allocated, and no further virtual pages are retrieved from secondary 1905 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a 1906 * limited number of page mappings are created at the low-end of the 1907 * specified address range. (For this purpose, a superpage mapping 1908 * counts as one page mapping.) Otherwise, all resident pages within 1909 * the specified address range are mapped. 1910 */ 1911static void 1912vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, 1913 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) 1914{ 1915 vm_offset_t start; 1916 vm_page_t p, p_start; 1917 vm_pindex_t mask, psize, threshold, tmpidx; 1918 1919 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) 1920 return; 1921 VM_OBJECT_RLOCK(object); 1922 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1923 VM_OBJECT_RUNLOCK(object); 1924 VM_OBJECT_WLOCK(object); 1925 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { 1926 pmap_object_init_pt(map->pmap, addr, object, pindex, 1927 size); 1928 VM_OBJECT_WUNLOCK(object); 1929 return; 1930 } 1931 VM_OBJECT_LOCK_DOWNGRADE(object); 1932 } 1933 1934 psize = atop(size); 1935 if (psize + pindex > object->size) { 1936 if (object->size < pindex) { 1937 VM_OBJECT_RUNLOCK(object); 1938 return; 1939 } 1940 psize = object->size - pindex; 1941 } 1942 1943 start = 0; 1944 p_start = NULL; 1945 threshold = MAX_INIT_PT; 1946 1947 p = vm_page_find_least(object, pindex); 1948 /* 1949 * Assert: the variable p is either (1) the page with the 1950 * least pindex greater than or equal to the parameter pindex 1951 * or (2) NULL. 1952 */ 1953 for (; 1954 p != NULL && (tmpidx = p->pindex - pindex) < psize; 1955 p = TAILQ_NEXT(p, listq)) { 1956 /* 1957 * don't allow an madvise to blow away our really 1958 * free pages allocating pv entries. 1959 */ 1960 if (((flags & MAP_PREFAULT_MADVISE) != 0 && 1961 vm_cnt.v_free_count < vm_cnt.v_free_reserved) || 1962 ((flags & MAP_PREFAULT_PARTIAL) != 0 && 1963 tmpidx >= threshold)) { 1964 psize = tmpidx; 1965 break; 1966 } 1967 if (p->valid == VM_PAGE_BITS_ALL) { 1968 if (p_start == NULL) { 1969 start = addr + ptoa(tmpidx); 1970 p_start = p; 1971 } 1972 /* Jump ahead if a superpage mapping is possible. */ 1973 if (p->psind > 0 && ((addr + ptoa(tmpidx)) & 1974 (pagesizes[p->psind] - 1)) == 0) { 1975 mask = atop(pagesizes[p->psind]) - 1; 1976 if (tmpidx + mask < psize && 1977 vm_page_ps_test(p, PS_ALL_VALID, NULL)) { 1978 p += mask; 1979 threshold += mask; 1980 } 1981 } 1982 } else if (p_start != NULL) { 1983 pmap_enter_object(map->pmap, start, addr + 1984 ptoa(tmpidx), p_start, prot); 1985 p_start = NULL; 1986 } 1987 } 1988 if (p_start != NULL) 1989 pmap_enter_object(map->pmap, start, addr + ptoa(psize), 1990 p_start, prot); 1991 VM_OBJECT_RUNLOCK(object); 1992} 1993 1994/* 1995 * vm_map_protect: 1996 * 1997 * Sets the protection of the specified address 1998 * region in the target map. If "set_max" is 1999 * specified, the maximum protection is to be set; 2000 * otherwise, only the current protection is affected. 2001 */ 2002int 2003vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, 2004 vm_prot_t new_prot, boolean_t set_max) 2005{ 2006 vm_map_entry_t current, entry; 2007 vm_object_t obj; 2008 struct ucred *cred; 2009 vm_prot_t old_prot; 2010 2011 if (start == end) 2012 return (KERN_SUCCESS); 2013 2014 vm_map_lock(map); 2015 2016 /* 2017 * Ensure that we are not concurrently wiring pages. vm_map_wire() may 2018 * need to fault pages into the map and will drop the map lock while 2019 * doing so, and the VM object may end up in an inconsistent state if we 2020 * update the protection on the map entry in between faults. 2021 */ 2022 vm_map_wait_busy(map); 2023 2024 VM_MAP_RANGE_CHECK(map, start, end); 2025 2026 if (vm_map_lookup_entry(map, start, &entry)) { 2027 vm_map_clip_start(map, entry, start); 2028 } else { 2029 entry = entry->next; 2030 } 2031 2032 /* 2033 * Make a first pass to check for protection violations. 2034 */ 2035 for (current = entry; current->start < end; current = current->next) { 2036 if ((current->eflags & MAP_ENTRY_GUARD) != 0) 2037 continue; 2038 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2039 vm_map_unlock(map); 2040 return (KERN_INVALID_ARGUMENT); 2041 } 2042 if ((new_prot & current->max_protection) != new_prot) { 2043 vm_map_unlock(map); 2044 return (KERN_PROTECTION_FAILURE); 2045 } 2046 } 2047 2048 /* 2049 * Do an accounting pass for private read-only mappings that 2050 * now will do cow due to allowed write (e.g. debugger sets 2051 * breakpoint on text segment) 2052 */ 2053 for (current = entry; current->start < end; current = current->next) { 2054 2055 vm_map_clip_end(map, current, end); 2056 2057 if (set_max || 2058 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || 2059 ENTRY_CHARGED(current) || 2060 (current->eflags & MAP_ENTRY_GUARD) != 0) { 2061 continue; 2062 } 2063 2064 cred = curthread->td_ucred; 2065 obj = current->object.vm_object; 2066 2067 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { 2068 if (!swap_reserve(current->end - current->start)) { 2069 vm_map_unlock(map); 2070 return (KERN_RESOURCE_SHORTAGE); 2071 } 2072 crhold(cred); 2073 current->cred = cred; 2074 continue; 2075 } 2076 2077 VM_OBJECT_WLOCK(obj); 2078 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { 2079 VM_OBJECT_WUNLOCK(obj); 2080 continue; 2081 } 2082 2083 /* 2084 * Charge for the whole object allocation now, since 2085 * we cannot distinguish between non-charged and 2086 * charged clipped mapping of the same object later. 2087 */ 2088 KASSERT(obj->charge == 0, 2089 ("vm_map_protect: object %p overcharged (entry %p)", 2090 obj, current)); 2091 if (!swap_reserve(ptoa(obj->size))) { 2092 VM_OBJECT_WUNLOCK(obj); 2093 vm_map_unlock(map); 2094 return (KERN_RESOURCE_SHORTAGE); 2095 } 2096 2097 crhold(cred); 2098 obj->cred = cred; 2099 obj->charge = ptoa(obj->size); 2100 VM_OBJECT_WUNLOCK(obj); 2101 } 2102 2103 /* 2104 * Go back and fix up protections. [Note that clipping is not 2105 * necessary the second time.] 2106 */ 2107 for (current = entry; current->start < end; current = current->next) { 2108 if ((current->eflags & MAP_ENTRY_GUARD) != 0) 2109 continue; 2110 2111 old_prot = current->protection; 2112 2113 if (set_max) 2114 current->protection = 2115 (current->max_protection = new_prot) & 2116 old_prot; 2117 else 2118 current->protection = new_prot; 2119 2120 /* 2121 * For user wired map entries, the normal lazy evaluation of 2122 * write access upgrades through soft page faults is 2123 * undesirable. Instead, immediately copy any pages that are 2124 * copy-on-write and enable write access in the physical map. 2125 */ 2126 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && 2127 (current->protection & VM_PROT_WRITE) != 0 && 2128 (old_prot & VM_PROT_WRITE) == 0) 2129 vm_fault_copy_entry(map, map, current, current, NULL); 2130 2131 /* 2132 * When restricting access, update the physical map. Worry 2133 * about copy-on-write here. 2134 */ 2135 if ((old_prot & ~current->protection) != 0) { 2136#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ 2137 VM_PROT_ALL) 2138 pmap_protect(map->pmap, current->start, 2139 current->end, 2140 current->protection & MASK(current)); 2141#undef MASK 2142 } 2143 vm_map_simplify_entry(map, current); 2144 } 2145 vm_map_unlock(map); 2146 return (KERN_SUCCESS); 2147} 2148 2149/* 2150 * vm_map_madvise: 2151 * 2152 * This routine traverses a processes map handling the madvise 2153 * system call. Advisories are classified as either those effecting 2154 * the vm_map_entry structure, or those effecting the underlying 2155 * objects. 2156 */ 2157int 2158vm_map_madvise( 2159 vm_map_t map, 2160 vm_offset_t start, 2161 vm_offset_t end, 2162 int behav) 2163{ 2164 vm_map_entry_t current, entry; 2165 int modify_map = 0; 2166 2167 /* 2168 * Some madvise calls directly modify the vm_map_entry, in which case 2169 * we need to use an exclusive lock on the map and we need to perform 2170 * various clipping operations. Otherwise we only need a read-lock 2171 * on the map. 2172 */ 2173 switch(behav) { 2174 case MADV_NORMAL: 2175 case MADV_SEQUENTIAL: 2176 case MADV_RANDOM: 2177 case MADV_NOSYNC: 2178 case MADV_AUTOSYNC: 2179 case MADV_NOCORE: 2180 case MADV_CORE: 2181 if (start == end) 2182 return (KERN_SUCCESS); 2183 modify_map = 1; 2184 vm_map_lock(map); 2185 break; 2186 case MADV_WILLNEED: 2187 case MADV_DONTNEED: 2188 case MADV_FREE: 2189 if (start == end) 2190 return (KERN_SUCCESS); 2191 vm_map_lock_read(map); 2192 break; 2193 default: 2194 return (KERN_INVALID_ARGUMENT); 2195 } 2196 2197 /* 2198 * Locate starting entry and clip if necessary. 2199 */ 2200 VM_MAP_RANGE_CHECK(map, start, end); 2201 2202 if (vm_map_lookup_entry(map, start, &entry)) { 2203 if (modify_map) 2204 vm_map_clip_start(map, entry, start); 2205 } else { 2206 entry = entry->next; 2207 } 2208 2209 if (modify_map) { 2210 /* 2211 * madvise behaviors that are implemented in the vm_map_entry. 2212 * 2213 * We clip the vm_map_entry so that behavioral changes are 2214 * limited to the specified address range. 2215 */ 2216 for (current = entry; current->start < end; 2217 current = current->next) { 2218 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2219 continue; 2220 2221 vm_map_clip_end(map, current, end); 2222 2223 switch (behav) { 2224 case MADV_NORMAL: 2225 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); 2226 break; 2227 case MADV_SEQUENTIAL: 2228 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); 2229 break; 2230 case MADV_RANDOM: 2231 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); 2232 break; 2233 case MADV_NOSYNC: 2234 current->eflags |= MAP_ENTRY_NOSYNC; 2235 break; 2236 case MADV_AUTOSYNC: 2237 current->eflags &= ~MAP_ENTRY_NOSYNC; 2238 break; 2239 case MADV_NOCORE: 2240 current->eflags |= MAP_ENTRY_NOCOREDUMP; 2241 break; 2242 case MADV_CORE: 2243 current->eflags &= ~MAP_ENTRY_NOCOREDUMP; 2244 break; 2245 default: 2246 break; 2247 } 2248 vm_map_simplify_entry(map, current); 2249 } 2250 vm_map_unlock(map); 2251 } else { 2252 vm_pindex_t pstart, pend; 2253 2254 /* 2255 * madvise behaviors that are implemented in the underlying 2256 * vm_object. 2257 * 2258 * Since we don't clip the vm_map_entry, we have to clip 2259 * the vm_object pindex and count. 2260 */ 2261 for (current = entry; current->start < end; 2262 current = current->next) { 2263 vm_offset_t useEnd, useStart; 2264 2265 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) 2266 continue; 2267 2268 pstart = OFF_TO_IDX(current->offset); 2269 pend = pstart + atop(current->end - current->start); 2270 useStart = current->start; 2271 useEnd = current->end; 2272 2273 if (current->start < start) { 2274 pstart += atop(start - current->start); 2275 useStart = start; 2276 } 2277 if (current->end > end) { 2278 pend -= atop(current->end - end); 2279 useEnd = end; 2280 } 2281 2282 if (pstart >= pend) 2283 continue; 2284 2285 /* 2286 * Perform the pmap_advise() before clearing 2287 * PGA_REFERENCED in vm_page_advise(). Otherwise, a 2288 * concurrent pmap operation, such as pmap_remove(), 2289 * could clear a reference in the pmap and set 2290 * PGA_REFERENCED on the page before the pmap_advise() 2291 * had completed. Consequently, the page would appear 2292 * referenced based upon an old reference that 2293 * occurred before this pmap_advise() ran. 2294 */ 2295 if (behav == MADV_DONTNEED || behav == MADV_FREE) 2296 pmap_advise(map->pmap, useStart, useEnd, 2297 behav); 2298 2299 vm_object_madvise(current->object.vm_object, pstart, 2300 pend, behav); 2301 2302 /* 2303 * Pre-populate paging structures in the 2304 * WILLNEED case. For wired entries, the 2305 * paging structures are already populated. 2306 */ 2307 if (behav == MADV_WILLNEED && 2308 current->wired_count == 0) { 2309 vm_map_pmap_enter(map, 2310 useStart, 2311 current->protection, 2312 current->object.vm_object, 2313 pstart, 2314 ptoa(pend - pstart), 2315 MAP_PREFAULT_MADVISE 2316 ); 2317 } 2318 } 2319 vm_map_unlock_read(map); 2320 } 2321 return (0); 2322} 2323 2324 2325/* 2326 * vm_map_inherit: 2327 * 2328 * Sets the inheritance of the specified address 2329 * range in the target map. Inheritance 2330 * affects how the map will be shared with 2331 * child maps at the time of vmspace_fork. 2332 */ 2333int 2334vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, 2335 vm_inherit_t new_inheritance) 2336{ 2337 vm_map_entry_t entry; 2338 vm_map_entry_t temp_entry; 2339 2340 switch (new_inheritance) { 2341 case VM_INHERIT_NONE: 2342 case VM_INHERIT_COPY: 2343 case VM_INHERIT_SHARE: 2344 case VM_INHERIT_ZERO: 2345 break; 2346 default: 2347 return (KERN_INVALID_ARGUMENT); 2348 } 2349 if (start == end) 2350 return (KERN_SUCCESS); 2351 vm_map_lock(map); 2352 VM_MAP_RANGE_CHECK(map, start, end); 2353 if (vm_map_lookup_entry(map, start, &temp_entry)) { 2354 entry = temp_entry; 2355 vm_map_clip_start(map, entry, start); 2356 } else 2357 entry = temp_entry->next; 2358 while (entry->start < end) { 2359 vm_map_clip_end(map, entry, end); 2360 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 || 2361 new_inheritance != VM_INHERIT_ZERO) 2362 entry->inheritance = new_inheritance; 2363 vm_map_simplify_entry(map, entry); 2364 entry = entry->next; 2365 } 2366 vm_map_unlock(map); 2367 return (KERN_SUCCESS); 2368} 2369 2370/* 2371 * vm_map_unwire: 2372 * 2373 * Implements both kernel and user unwiring. 2374 */ 2375int 2376vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2377 int flags) 2378{ 2379 vm_map_entry_t entry, first_entry, tmp_entry; 2380 vm_offset_t saved_start; 2381 unsigned int last_timestamp; 2382 int rv; 2383 boolean_t need_wakeup, result, user_unwire; 2384 2385 if (start == end) 2386 return (KERN_SUCCESS); 2387 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2388 vm_map_lock(map); 2389 VM_MAP_RANGE_CHECK(map, start, end); 2390 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2391 if (flags & VM_MAP_WIRE_HOLESOK) 2392 first_entry = first_entry->next; 2393 else { 2394 vm_map_unlock(map); 2395 return (KERN_INVALID_ADDRESS); 2396 } 2397 } 2398 last_timestamp = map->timestamp; 2399 entry = first_entry; 2400 while (entry->start < end) { 2401 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2402 /* 2403 * We have not yet clipped the entry. 2404 */ 2405 saved_start = (start >= entry->start) ? start : 2406 entry->start; 2407 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2408 if (vm_map_unlock_and_wait(map, 0)) { 2409 /* 2410 * Allow interruption of user unwiring? 2411 */ 2412 } 2413 vm_map_lock(map); 2414 if (last_timestamp+1 != map->timestamp) { 2415 /* 2416 * Look again for the entry because the map was 2417 * modified while it was unlocked. 2418 * Specifically, the entry may have been 2419 * clipped, merged, or deleted. 2420 */ 2421 if (!vm_map_lookup_entry(map, saved_start, 2422 &tmp_entry)) { 2423 if (flags & VM_MAP_WIRE_HOLESOK) 2424 tmp_entry = tmp_entry->next; 2425 else { 2426 if (saved_start == start) { 2427 /* 2428 * First_entry has been deleted. 2429 */ 2430 vm_map_unlock(map); 2431 return (KERN_INVALID_ADDRESS); 2432 } 2433 end = saved_start; 2434 rv = KERN_INVALID_ADDRESS; 2435 goto done; 2436 } 2437 } 2438 if (entry == first_entry) 2439 first_entry = tmp_entry; 2440 else 2441 first_entry = NULL; 2442 entry = tmp_entry; 2443 } 2444 last_timestamp = map->timestamp; 2445 continue; 2446 } 2447 vm_map_clip_start(map, entry, start); 2448 vm_map_clip_end(map, entry, end); 2449 /* 2450 * Mark the entry in case the map lock is released. (See 2451 * above.) 2452 */ 2453 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2454 entry->wiring_thread == NULL, 2455 ("owned map entry %p", entry)); 2456 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2457 entry->wiring_thread = curthread; 2458 /* 2459 * Check the map for holes in the specified region. 2460 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2461 */ 2462 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && 2463 (entry->end < end && entry->next->start > entry->end)) { 2464 end = entry->end; 2465 rv = KERN_INVALID_ADDRESS; 2466 goto done; 2467 } 2468 /* 2469 * If system unwiring, require that the entry is system wired. 2470 */ 2471 if (!user_unwire && 2472 vm_map_entry_system_wired_count(entry) == 0) { 2473 end = entry->end; 2474 rv = KERN_INVALID_ARGUMENT; 2475 goto done; 2476 } 2477 entry = entry->next; 2478 } 2479 rv = KERN_SUCCESS; 2480done: 2481 need_wakeup = FALSE; 2482 if (first_entry == NULL) { 2483 result = vm_map_lookup_entry(map, start, &first_entry); 2484 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2485 first_entry = first_entry->next; 2486 else 2487 KASSERT(result, ("vm_map_unwire: lookup failed")); 2488 } 2489 for (entry = first_entry; entry->start < end; entry = entry->next) { 2490 /* 2491 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2492 * space in the unwired region could have been mapped 2493 * while the map lock was dropped for draining 2494 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread 2495 * could be simultaneously wiring this new mapping 2496 * entry. Detect these cases and skip any entries 2497 * marked as in transition by us. 2498 */ 2499 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2500 entry->wiring_thread != curthread) { 2501 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2502 ("vm_map_unwire: !HOLESOK and new/changed entry")); 2503 continue; 2504 } 2505 2506 if (rv == KERN_SUCCESS && (!user_unwire || 2507 (entry->eflags & MAP_ENTRY_USER_WIRED))) { 2508 if (user_unwire) 2509 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 2510 if (entry->wired_count == 1) 2511 vm_map_entry_unwire(map, entry); 2512 else 2513 entry->wired_count--; 2514 } 2515 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2516 ("vm_map_unwire: in-transition flag missing %p", entry)); 2517 KASSERT(entry->wiring_thread == curthread, 2518 ("vm_map_unwire: alien wire %p", entry)); 2519 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; 2520 entry->wiring_thread = NULL; 2521 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2522 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2523 need_wakeup = TRUE; 2524 } 2525 vm_map_simplify_entry(map, entry); 2526 } 2527 vm_map_unlock(map); 2528 if (need_wakeup) 2529 vm_map_wakeup(map); 2530 return (rv); 2531} 2532 2533/* 2534 * vm_map_wire_entry_failure: 2535 * 2536 * Handle a wiring failure on the given entry. 2537 * 2538 * The map should be locked. 2539 */ 2540static void 2541vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, 2542 vm_offset_t failed_addr) 2543{ 2544 2545 VM_MAP_ASSERT_LOCKED(map); 2546 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && 2547 entry->wired_count == 1, 2548 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); 2549 KASSERT(failed_addr < entry->end, 2550 ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); 2551 2552 /* 2553 * If any pages at the start of this entry were successfully wired, 2554 * then unwire them. 2555 */ 2556 if (failed_addr > entry->start) { 2557 pmap_unwire(map->pmap, entry->start, failed_addr); 2558 vm_object_unwire(entry->object.vm_object, entry->offset, 2559 failed_addr - entry->start, PQ_ACTIVE); 2560 } 2561 2562 /* 2563 * Assign an out-of-range value to represent the failure to wire this 2564 * entry. 2565 */ 2566 entry->wired_count = -1; 2567} 2568 2569/* 2570 * vm_map_wire: 2571 * 2572 * Implements both kernel and user wiring. 2573 */ 2574int 2575vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, 2576 int flags) 2577{ 2578 vm_map_entry_t entry, first_entry, tmp_entry; 2579 vm_offset_t faddr, saved_end, saved_start; 2580 unsigned int last_timestamp; 2581 int rv; 2582 boolean_t need_wakeup, result, user_wire; 2583 vm_prot_t prot; 2584 2585 if (start == end) 2586 return (KERN_SUCCESS); 2587 prot = 0; 2588 if (flags & VM_MAP_WIRE_WRITE) 2589 prot |= VM_PROT_WRITE; 2590 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; 2591 vm_map_lock(map); 2592 VM_MAP_RANGE_CHECK(map, start, end); 2593 if (!vm_map_lookup_entry(map, start, &first_entry)) { 2594 if (flags & VM_MAP_WIRE_HOLESOK) 2595 first_entry = first_entry->next; 2596 else { 2597 vm_map_unlock(map); 2598 return (KERN_INVALID_ADDRESS); 2599 } 2600 } 2601 last_timestamp = map->timestamp; 2602 entry = first_entry; 2603 while (entry->start < end) { 2604 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 2605 /* 2606 * We have not yet clipped the entry. 2607 */ 2608 saved_start = (start >= entry->start) ? start : 2609 entry->start; 2610 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 2611 if (vm_map_unlock_and_wait(map, 0)) { 2612 /* 2613 * Allow interruption of user wiring? 2614 */ 2615 } 2616 vm_map_lock(map); 2617 if (last_timestamp + 1 != map->timestamp) { 2618 /* 2619 * Look again for the entry because the map was 2620 * modified while it was unlocked. 2621 * Specifically, the entry may have been 2622 * clipped, merged, or deleted. 2623 */ 2624 if (!vm_map_lookup_entry(map, saved_start, 2625 &tmp_entry)) { 2626 if (flags & VM_MAP_WIRE_HOLESOK) 2627 tmp_entry = tmp_entry->next; 2628 else { 2629 if (saved_start == start) { 2630 /* 2631 * first_entry has been deleted. 2632 */ 2633 vm_map_unlock(map); 2634 return (KERN_INVALID_ADDRESS); 2635 } 2636 end = saved_start; 2637 rv = KERN_INVALID_ADDRESS; 2638 goto done; 2639 } 2640 } 2641 if (entry == first_entry) 2642 first_entry = tmp_entry; 2643 else 2644 first_entry = NULL; 2645 entry = tmp_entry; 2646 } 2647 last_timestamp = map->timestamp; 2648 continue; 2649 } 2650 vm_map_clip_start(map, entry, start); 2651 vm_map_clip_end(map, entry, end); 2652 /* 2653 * Mark the entry in case the map lock is released. (See 2654 * above.) 2655 */ 2656 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && 2657 entry->wiring_thread == NULL, 2658 ("owned map entry %p", entry)); 2659 entry->eflags |= MAP_ENTRY_IN_TRANSITION; 2660 entry->wiring_thread = curthread; 2661 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 2662 || (entry->protection & prot) != prot) { 2663 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; 2664 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { 2665 end = entry->end; 2666 rv = KERN_INVALID_ADDRESS; 2667 goto done; 2668 } 2669 goto next_entry; 2670 } 2671 if (entry->wired_count == 0) { 2672 entry->wired_count++; 2673 saved_start = entry->start; 2674 saved_end = entry->end; 2675 2676 /* 2677 * Release the map lock, relying on the in-transition 2678 * mark. Mark the map busy for fork. 2679 */ 2680 vm_map_busy(map); 2681 vm_map_unlock(map); 2682 2683 faddr = saved_start; 2684 do { 2685 /* 2686 * Simulate a fault to get the page and enter 2687 * it into the physical map. 2688 */ 2689 if ((rv = vm_fault(map, faddr, VM_PROT_NONE, 2690 VM_FAULT_WIRE)) != KERN_SUCCESS) 2691 break; 2692 } while ((faddr += PAGE_SIZE) < saved_end); 2693 vm_map_lock(map); 2694 vm_map_unbusy(map); 2695 if (last_timestamp + 1 != map->timestamp) { 2696 /* 2697 * Look again for the entry because the map was 2698 * modified while it was unlocked. The entry 2699 * may have been clipped, but NOT merged or 2700 * deleted. 2701 */ 2702 result = vm_map_lookup_entry(map, saved_start, 2703 &tmp_entry); 2704 KASSERT(result, ("vm_map_wire: lookup failed")); 2705 if (entry == first_entry) 2706 first_entry = tmp_entry; 2707 else 2708 first_entry = NULL; 2709 entry = tmp_entry; 2710 while (entry->end < saved_end) { 2711 /* 2712 * In case of failure, handle entries 2713 * that were not fully wired here; 2714 * fully wired entries are handled 2715 * later. 2716 */ 2717 if (rv != KERN_SUCCESS && 2718 faddr < entry->end) 2719 vm_map_wire_entry_failure(map, 2720 entry, faddr); 2721 entry = entry->next; 2722 } 2723 } 2724 last_timestamp = map->timestamp; 2725 if (rv != KERN_SUCCESS) { 2726 vm_map_wire_entry_failure(map, entry, faddr); 2727 end = entry->end; 2728 goto done; 2729 } 2730 } else if (!user_wire || 2731 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2732 entry->wired_count++; 2733 } 2734 /* 2735 * Check the map for holes in the specified region. 2736 * If VM_MAP_WIRE_HOLESOK was specified, skip this check. 2737 */ 2738 next_entry: 2739 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 && 2740 entry->end < end && entry->next->start > entry->end) { 2741 end = entry->end; 2742 rv = KERN_INVALID_ADDRESS; 2743 goto done; 2744 } 2745 entry = entry->next; 2746 } 2747 rv = KERN_SUCCESS; 2748done: 2749 need_wakeup = FALSE; 2750 if (first_entry == NULL) { 2751 result = vm_map_lookup_entry(map, start, &first_entry); 2752 if (!result && (flags & VM_MAP_WIRE_HOLESOK)) 2753 first_entry = first_entry->next; 2754 else 2755 KASSERT(result, ("vm_map_wire: lookup failed")); 2756 } 2757 for (entry = first_entry; entry->start < end; entry = entry->next) { 2758 /* 2759 * If VM_MAP_WIRE_HOLESOK was specified, an empty 2760 * space in the unwired region could have been mapped 2761 * while the map lock was dropped for faulting in the 2762 * pages or draining MAP_ENTRY_IN_TRANSITION. 2763 * Moreover, another thread could be simultaneously 2764 * wiring this new mapping entry. Detect these cases 2765 * and skip any entries marked as in transition not by us. 2766 */ 2767 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || 2768 entry->wiring_thread != curthread) { 2769 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, 2770 ("vm_map_wire: !HOLESOK and new/changed entry")); 2771 continue; 2772 } 2773 2774 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) 2775 goto next_entry_done; 2776 2777 if (rv == KERN_SUCCESS) { 2778 if (user_wire) 2779 entry->eflags |= MAP_ENTRY_USER_WIRED; 2780 } else if (entry->wired_count == -1) { 2781 /* 2782 * Wiring failed on this entry. Thus, unwiring is 2783 * unnecessary. 2784 */ 2785 entry->wired_count = 0; 2786 } else if (!user_wire || 2787 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { 2788 /* 2789 * Undo the wiring. Wiring succeeded on this entry 2790 * but failed on a later entry. 2791 */ 2792 if (entry->wired_count == 1) 2793 vm_map_entry_unwire(map, entry); 2794 else 2795 entry->wired_count--; 2796 } 2797 next_entry_done: 2798 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, 2799 ("vm_map_wire: in-transition flag missing %p", entry)); 2800 KASSERT(entry->wiring_thread == curthread, 2801 ("vm_map_wire: alien wire %p", entry)); 2802 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | 2803 MAP_ENTRY_WIRE_SKIPPED); 2804 entry->wiring_thread = NULL; 2805 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { 2806 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; 2807 need_wakeup = TRUE; 2808 } 2809 vm_map_simplify_entry(map, entry); 2810 } 2811 vm_map_unlock(map); 2812 if (need_wakeup) 2813 vm_map_wakeup(map); 2814 return (rv); 2815} 2816 2817/* 2818 * vm_map_sync 2819 * 2820 * Push any dirty cached pages in the address range to their pager. 2821 * If syncio is TRUE, dirty pages are written synchronously. 2822 * If invalidate is TRUE, any cached pages are freed as well. 2823 * 2824 * If the size of the region from start to end is zero, we are 2825 * supposed to flush all modified pages within the region containing 2826 * start. Unfortunately, a region can be split or coalesced with 2827 * neighboring regions, making it difficult to determine what the 2828 * original region was. Therefore, we approximate this requirement by 2829 * flushing the current region containing start. 2830 * 2831 * Returns an error if any part of the specified range is not mapped. 2832 */ 2833int 2834vm_map_sync( 2835 vm_map_t map, 2836 vm_offset_t start, 2837 vm_offset_t end, 2838 boolean_t syncio, 2839 boolean_t invalidate) 2840{ 2841 vm_map_entry_t current; 2842 vm_map_entry_t entry; 2843 vm_size_t size; 2844 vm_object_t object; 2845 vm_ooffset_t offset; 2846 unsigned int last_timestamp; 2847 boolean_t failed; 2848 2849 vm_map_lock_read(map); 2850 VM_MAP_RANGE_CHECK(map, start, end); 2851 if (!vm_map_lookup_entry(map, start, &entry)) { 2852 vm_map_unlock_read(map); 2853 return (KERN_INVALID_ADDRESS); 2854 } else if (start == end) { 2855 start = entry->start; 2856 end = entry->end; 2857 } 2858 /* 2859 * Make a first pass to check for user-wired memory and holes. 2860 */ 2861 for (current = entry; current->start < end; current = current->next) { 2862 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { 2863 vm_map_unlock_read(map); 2864 return (KERN_INVALID_ARGUMENT); 2865 } 2866 if (end > current->end && 2867 current->end != current->next->start) { 2868 vm_map_unlock_read(map); 2869 return (KERN_INVALID_ADDRESS); 2870 } 2871 } 2872 2873 if (invalidate) 2874 pmap_remove(map->pmap, start, end); 2875 failed = FALSE; 2876 2877 /* 2878 * Make a second pass, cleaning/uncaching pages from the indicated 2879 * objects as we go. 2880 */ 2881 for (current = entry; current->start < end;) { 2882 offset = current->offset + (start - current->start); 2883 size = (end <= current->end ? end : current->end) - start; 2884 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { 2885 vm_map_t smap; 2886 vm_map_entry_t tentry; 2887 vm_size_t tsize; 2888 2889 smap = current->object.sub_map; 2890 vm_map_lock_read(smap); 2891 (void) vm_map_lookup_entry(smap, offset, &tentry); 2892 tsize = tentry->end - offset; 2893 if (tsize < size) 2894 size = tsize; 2895 object = tentry->object.vm_object; 2896 offset = tentry->offset + (offset - tentry->start); 2897 vm_map_unlock_read(smap); 2898 } else { 2899 object = current->object.vm_object; 2900 } 2901 vm_object_reference(object); 2902 last_timestamp = map->timestamp; 2903 vm_map_unlock_read(map); 2904 if (!vm_object_sync(object, offset, size, syncio, invalidate)) 2905 failed = TRUE; 2906 start += size; 2907 vm_object_deallocate(object); 2908 vm_map_lock_read(map); 2909 if (last_timestamp == map->timestamp || 2910 !vm_map_lookup_entry(map, start, ¤t)) 2911 current = current->next; 2912 } 2913 2914 vm_map_unlock_read(map); 2915 return (failed ? KERN_FAILURE : KERN_SUCCESS); 2916} 2917 2918/* 2919 * vm_map_entry_unwire: [ internal use only ] 2920 * 2921 * Make the region specified by this entry pageable. 2922 * 2923 * The map in question should be locked. 2924 * [This is the reason for this routine's existence.] 2925 */ 2926static void 2927vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) 2928{ 2929 2930 VM_MAP_ASSERT_LOCKED(map); 2931 KASSERT(entry->wired_count > 0, 2932 ("vm_map_entry_unwire: entry %p isn't wired", entry)); 2933 pmap_unwire(map->pmap, entry->start, entry->end); 2934 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end - 2935 entry->start, PQ_ACTIVE); 2936 entry->wired_count = 0; 2937} 2938 2939static void 2940vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) 2941{ 2942 2943 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) 2944 vm_object_deallocate(entry->object.vm_object); 2945 uma_zfree(system_map ? kmapentzone : mapentzone, entry); 2946} 2947 2948/* 2949 * vm_map_entry_delete: [ internal use only ] 2950 * 2951 * Deallocate the given entry from the target map. 2952 */ 2953static void 2954vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) 2955{ 2956 vm_object_t object; 2957 vm_pindex_t offidxstart, offidxend, count, size1; 2958 vm_size_t size; 2959 2960 vm_map_entry_unlink(map, entry); 2961 object = entry->object.vm_object; 2962 2963 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) { 2964 MPASS(entry->cred == NULL); 2965 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0); 2966 MPASS(object == NULL); 2967 vm_map_entry_deallocate(entry, map->system_map); 2968 return; 2969 } 2970 2971 size = entry->end - entry->start; 2972 map->size -= size; 2973 2974 if (entry->cred != NULL) { 2975 swap_release_by_cred(size, entry->cred); 2976 crfree(entry->cred); 2977 } 2978 2979 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && 2980 (object != NULL)) { 2981 KASSERT(entry->cred == NULL || object->cred == NULL || 2982 (entry->eflags & MAP_ENTRY_NEEDS_COPY), 2983 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); 2984 count = atop(size); 2985 offidxstart = OFF_TO_IDX(entry->offset); 2986 offidxend = offidxstart + count; 2987 VM_OBJECT_WLOCK(object); 2988 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT | 2989 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || 2990 object == kernel_object || object == kmem_object)) { 2991 vm_object_collapse(object); 2992 2993 /* 2994 * The option OBJPR_NOTMAPPED can be passed here 2995 * because vm_map_delete() already performed 2996 * pmap_remove() on the only mapping to this range 2997 * of pages. 2998 */ 2999 vm_object_page_remove(object, offidxstart, offidxend, 3000 OBJPR_NOTMAPPED); 3001 if (object->type == OBJT_SWAP) 3002 swap_pager_freespace(object, offidxstart, 3003 count); 3004 if (offidxend >= object->size && 3005 offidxstart < object->size) { 3006 size1 = object->size; 3007 object->size = offidxstart; 3008 if (object->cred != NULL) { 3009 size1 -= object->size; 3010 KASSERT(object->charge >= ptoa(size1), 3011 ("object %p charge < 0", object)); 3012 swap_release_by_cred(ptoa(size1), 3013 object->cred); 3014 object->charge -= ptoa(size1); 3015 } 3016 } 3017 } 3018 VM_OBJECT_WUNLOCK(object); 3019 } else 3020 entry->object.vm_object = NULL; 3021 if (map->system_map) 3022 vm_map_entry_deallocate(entry, TRUE); 3023 else { 3024 entry->next = curthread->td_map_def_user; 3025 curthread->td_map_def_user = entry; 3026 } 3027} 3028 3029/* 3030 * vm_map_delete: [ internal use only ] 3031 * 3032 * Deallocates the given address range from the target 3033 * map. 3034 */ 3035int 3036vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) 3037{ 3038 vm_map_entry_t entry; 3039 vm_map_entry_t first_entry; 3040 3041 VM_MAP_ASSERT_LOCKED(map); 3042 if (start == end) 3043 return (KERN_SUCCESS); 3044 3045 /* 3046 * Find the start of the region, and clip it 3047 */ 3048 if (!vm_map_lookup_entry(map, start, &first_entry)) 3049 entry = first_entry->next; 3050 else { 3051 entry = first_entry; 3052 vm_map_clip_start(map, entry, start); 3053 } 3054 3055 /* 3056 * Step through all entries in this region 3057 */ 3058 while (entry->start < end) { 3059 vm_map_entry_t next; 3060 3061 /* 3062 * Wait for wiring or unwiring of an entry to complete. 3063 * Also wait for any system wirings to disappear on 3064 * user maps. 3065 */ 3066 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || 3067 (vm_map_pmap(map) != kernel_pmap && 3068 vm_map_entry_system_wired_count(entry) != 0)) { 3069 unsigned int last_timestamp; 3070 vm_offset_t saved_start; 3071 vm_map_entry_t tmp_entry; 3072 3073 saved_start = entry->start; 3074 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 3075 last_timestamp = map->timestamp; 3076 (void) vm_map_unlock_and_wait(map, 0); 3077 vm_map_lock(map); 3078 if (last_timestamp + 1 != map->timestamp) { 3079 /* 3080 * Look again for the entry because the map was 3081 * modified while it was unlocked. 3082 * Specifically, the entry may have been 3083 * clipped, merged, or deleted. 3084 */ 3085 if (!vm_map_lookup_entry(map, saved_start, 3086 &tmp_entry)) 3087 entry = tmp_entry->next; 3088 else { 3089 entry = tmp_entry; 3090 vm_map_clip_start(map, entry, 3091 saved_start); 3092 } 3093 } 3094 continue; 3095 } 3096 vm_map_clip_end(map, entry, end); 3097 3098 next = entry->next; 3099 3100 /* 3101 * Unwire before removing addresses from the pmap; otherwise, 3102 * unwiring will put the entries back in the pmap. 3103 */ 3104 if (entry->wired_count != 0) 3105 vm_map_entry_unwire(map, entry); 3106 3107 /* 3108 * Remove mappings for the pages, but only if the 3109 * mappings could exist. For instance, it does not 3110 * make sense to call pmap_remove() for guard entries. 3111 */ 3112 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || 3113 entry->object.vm_object != NULL) 3114 pmap_remove(map->pmap, entry->start, entry->end); 3115 3116 /* 3117 * Delete the entry only after removing all pmap 3118 * entries pointing to its pages. (Otherwise, its 3119 * page frames may be reallocated, and any modify bits 3120 * will be set in the wrong object!) 3121 */ 3122 vm_map_entry_delete(map, entry); 3123 entry = next; 3124 } 3125 return (KERN_SUCCESS); 3126} 3127 3128/* 3129 * vm_map_remove: 3130 * 3131 * Remove the given address range from the target map. 3132 * This is the exported form of vm_map_delete. 3133 */ 3134int 3135vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) 3136{ 3137 int result; 3138 3139 vm_map_lock(map); 3140 VM_MAP_RANGE_CHECK(map, start, end); 3141 result = vm_map_delete(map, start, end); 3142 vm_map_unlock(map); 3143 return (result); 3144} 3145 3146/* 3147 * vm_map_check_protection: 3148 * 3149 * Assert that the target map allows the specified privilege on the 3150 * entire address region given. The entire region must be allocated. 3151 * 3152 * WARNING! This code does not and should not check whether the 3153 * contents of the region is accessible. For example a smaller file 3154 * might be mapped into a larger address space. 3155 * 3156 * NOTE! This code is also called by munmap(). 3157 * 3158 * The map must be locked. A read lock is sufficient. 3159 */ 3160boolean_t 3161vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, 3162 vm_prot_t protection) 3163{ 3164 vm_map_entry_t entry; 3165 vm_map_entry_t tmp_entry; 3166 3167 if (!vm_map_lookup_entry(map, start, &tmp_entry)) 3168 return (FALSE); 3169 entry = tmp_entry; 3170 3171 while (start < end) { 3172 /* 3173 * No holes allowed! 3174 */ 3175 if (start < entry->start) 3176 return (FALSE); 3177 /* 3178 * Check protection associated with entry. 3179 */ 3180 if ((entry->protection & protection) != protection) 3181 return (FALSE); 3182 /* go to next entry */ 3183 start = entry->end; 3184 entry = entry->next; 3185 } 3186 return (TRUE); 3187} 3188 3189/* 3190 * vm_map_copy_entry: 3191 * 3192 * Copies the contents of the source entry to the destination 3193 * entry. The entries *must* be aligned properly. 3194 */ 3195static void 3196vm_map_copy_entry( 3197 vm_map_t src_map, 3198 vm_map_t dst_map, 3199 vm_map_entry_t src_entry, 3200 vm_map_entry_t dst_entry, 3201 vm_ooffset_t *fork_charge) 3202{ 3203 vm_object_t src_object; 3204 vm_map_entry_t fake_entry; 3205 vm_offset_t size; 3206 struct ucred *cred; 3207 int charged; 3208 3209 VM_MAP_ASSERT_LOCKED(dst_map); 3210 3211 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) 3212 return; 3213 3214 if (src_entry->wired_count == 0 || 3215 (src_entry->protection & VM_PROT_WRITE) == 0) { 3216 /* 3217 * If the source entry is marked needs_copy, it is already 3218 * write-protected. 3219 */ 3220 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && 3221 (src_entry->protection & VM_PROT_WRITE) != 0) { 3222 pmap_protect(src_map->pmap, 3223 src_entry->start, 3224 src_entry->end, 3225 src_entry->protection & ~VM_PROT_WRITE); 3226 } 3227 3228 /* 3229 * Make a copy of the object. 3230 */ 3231 size = src_entry->end - src_entry->start; 3232 if ((src_object = src_entry->object.vm_object) != NULL) { 3233 VM_OBJECT_WLOCK(src_object); 3234 charged = ENTRY_CHARGED(src_entry); 3235 if (src_object->handle == NULL && 3236 (src_object->type == OBJT_DEFAULT || 3237 src_object->type == OBJT_SWAP)) { 3238 vm_object_collapse(src_object); 3239 if ((src_object->flags & (OBJ_NOSPLIT | 3240 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { 3241 vm_object_split(src_entry); 3242 src_object = 3243 src_entry->object.vm_object; 3244 } 3245 } 3246 vm_object_reference_locked(src_object); 3247 vm_object_clear_flag(src_object, OBJ_ONEMAPPING); 3248 if (src_entry->cred != NULL && 3249 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { 3250 KASSERT(src_object->cred == NULL, 3251 ("OVERCOMMIT: vm_map_copy_entry: cred %p", 3252 src_object)); 3253 src_object->cred = src_entry->cred; 3254 src_object->charge = size; 3255 } 3256 VM_OBJECT_WUNLOCK(src_object); 3257 dst_entry->object.vm_object = src_object; 3258 if (charged) { 3259 cred = curthread->td_ucred; 3260 crhold(cred); 3261 dst_entry->cred = cred; 3262 *fork_charge += size; 3263 if (!(src_entry->eflags & 3264 MAP_ENTRY_NEEDS_COPY)) { 3265 crhold(cred); 3266 src_entry->cred = cred; 3267 *fork_charge += size; 3268 } 3269 } 3270 src_entry->eflags |= MAP_ENTRY_COW | 3271 MAP_ENTRY_NEEDS_COPY; 3272 dst_entry->eflags |= MAP_ENTRY_COW | 3273 MAP_ENTRY_NEEDS_COPY; 3274 dst_entry->offset = src_entry->offset; 3275 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3276 /* 3277 * MAP_ENTRY_VN_WRITECNT cannot 3278 * indicate write reference from 3279 * src_entry, since the entry is 3280 * marked as needs copy. Allocate a 3281 * fake entry that is used to 3282 * decrement object->un_pager.vnp.writecount 3283 * at the appropriate time. Attach 3284 * fake_entry to the deferred list. 3285 */ 3286 fake_entry = vm_map_entry_create(dst_map); 3287 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; 3288 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; 3289 vm_object_reference(src_object); 3290 fake_entry->object.vm_object = src_object; 3291 fake_entry->start = src_entry->start; 3292 fake_entry->end = src_entry->end; 3293 fake_entry->next = curthread->td_map_def_user; 3294 curthread->td_map_def_user = fake_entry; 3295 } 3296 3297 pmap_copy(dst_map->pmap, src_map->pmap, 3298 dst_entry->start, dst_entry->end - dst_entry->start, 3299 src_entry->start); 3300 } else { 3301 dst_entry->object.vm_object = NULL; 3302 dst_entry->offset = 0; 3303 if (src_entry->cred != NULL) { 3304 dst_entry->cred = curthread->td_ucred; 3305 crhold(dst_entry->cred); 3306 *fork_charge += size; 3307 } 3308 } 3309 } else { 3310 /* 3311 * We don't want to make writeable wired pages copy-on-write. 3312 * Immediately copy these pages into the new map by simulating 3313 * page faults. The new pages are pageable. 3314 */ 3315 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, 3316 fork_charge); 3317 } 3318} 3319 3320/* 3321 * vmspace_map_entry_forked: 3322 * Update the newly-forked vmspace each time a map entry is inherited 3323 * or copied. The values for vm_dsize and vm_tsize are approximate 3324 * (and mostly-obsolete ideas in the face of mmap(2) et al.) 3325 */ 3326static void 3327vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, 3328 vm_map_entry_t entry) 3329{ 3330 vm_size_t entrysize; 3331 vm_offset_t newend; 3332 3333 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) 3334 return; 3335 entrysize = entry->end - entry->start; 3336 vm2->vm_map.size += entrysize; 3337 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { 3338 vm2->vm_ssize += btoc(entrysize); 3339 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && 3340 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { 3341 newend = MIN(entry->end, 3342 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); 3343 vm2->vm_dsize += btoc(newend - entry->start); 3344 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && 3345 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { 3346 newend = MIN(entry->end, 3347 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); 3348 vm2->vm_tsize += btoc(newend - entry->start); 3349 } 3350} 3351 3352/* 3353 * vmspace_fork: 3354 * Create a new process vmspace structure and vm_map 3355 * based on those of an existing process. The new map 3356 * is based on the old map, according to the inheritance 3357 * values on the regions in that map. 3358 * 3359 * XXX It might be worth coalescing the entries added to the new vmspace. 3360 * 3361 * The source map must not be locked. 3362 */ 3363struct vmspace * 3364vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) 3365{ 3366 struct vmspace *vm2; 3367 vm_map_t new_map, old_map; 3368 vm_map_entry_t new_entry, old_entry; 3369 vm_object_t object; 3370 int locked; 3371 vm_inherit_t inh; 3372 3373 old_map = &vm1->vm_map; 3374 /* Copy immutable fields of vm1 to vm2. */ 3375 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), 3376 pmap_pinit); 3377 if (vm2 == NULL) 3378 return (NULL); 3379 vm2->vm_taddr = vm1->vm_taddr; 3380 vm2->vm_daddr = vm1->vm_daddr; 3381 vm2->vm_maxsaddr = vm1->vm_maxsaddr; 3382 vm_map_lock(old_map); 3383 if (old_map->busy) 3384 vm_map_wait_busy(old_map); 3385 new_map = &vm2->vm_map; 3386 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ 3387 KASSERT(locked, ("vmspace_fork: lock failed")); 3388 3389 old_entry = old_map->header.next; 3390 3391 while (old_entry != &old_map->header) { 3392 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) 3393 panic("vm_map_fork: encountered a submap"); 3394 3395 inh = old_entry->inheritance; 3396 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 && 3397 inh != VM_INHERIT_NONE) 3398 inh = VM_INHERIT_COPY; 3399 3400 switch (inh) { 3401 case VM_INHERIT_NONE: 3402 break; 3403 3404 case VM_INHERIT_SHARE: 3405 /* 3406 * Clone the entry, creating the shared object if necessary. 3407 */ 3408 object = old_entry->object.vm_object; 3409 if (object == NULL) { 3410 object = vm_object_allocate(OBJT_DEFAULT, 3411 atop(old_entry->end - old_entry->start)); 3412 old_entry->object.vm_object = object; 3413 old_entry->offset = 0; 3414 if (old_entry->cred != NULL) { 3415 object->cred = old_entry->cred; 3416 object->charge = old_entry->end - 3417 old_entry->start; 3418 old_entry->cred = NULL; 3419 } 3420 } 3421 3422 /* 3423 * Add the reference before calling vm_object_shadow 3424 * to insure that a shadow object is created. 3425 */ 3426 vm_object_reference(object); 3427 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { 3428 vm_object_shadow(&old_entry->object.vm_object, 3429 &old_entry->offset, 3430 old_entry->end - old_entry->start); 3431 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 3432 /* Transfer the second reference too. */ 3433 vm_object_reference( 3434 old_entry->object.vm_object); 3435 3436 /* 3437 * As in vm_map_simplify_entry(), the 3438 * vnode lock will not be acquired in 3439 * this call to vm_object_deallocate(). 3440 */ 3441 vm_object_deallocate(object); 3442 object = old_entry->object.vm_object; 3443 } 3444 VM_OBJECT_WLOCK(object); 3445 vm_object_clear_flag(object, OBJ_ONEMAPPING); 3446 if (old_entry->cred != NULL) { 3447 KASSERT(object->cred == NULL, ("vmspace_fork both cred")); 3448 object->cred = old_entry->cred; 3449 object->charge = old_entry->end - old_entry->start; 3450 old_entry->cred = NULL; 3451 } 3452 3453 /* 3454 * Assert the correct state of the vnode 3455 * v_writecount while the object is locked, to 3456 * not relock it later for the assertion 3457 * correctness. 3458 */ 3459 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && 3460 object->type == OBJT_VNODE) { 3461 KASSERT(((struct vnode *)object->handle)-> 3462 v_writecount > 0, 3463 ("vmspace_fork: v_writecount %p", object)); 3464 KASSERT(object->un_pager.vnp.writemappings > 0, 3465 ("vmspace_fork: vnp.writecount %p", 3466 object)); 3467 } 3468 VM_OBJECT_WUNLOCK(object); 3469 3470 /* 3471 * Clone the entry, referencing the shared object. 3472 */ 3473 new_entry = vm_map_entry_create(new_map); 3474 *new_entry = *old_entry; 3475 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3476 MAP_ENTRY_IN_TRANSITION); 3477 new_entry->wiring_thread = NULL; 3478 new_entry->wired_count = 0; 3479 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { 3480 vnode_pager_update_writecount(object, 3481 new_entry->start, new_entry->end); 3482 } 3483 3484 /* 3485 * Insert the entry into the new map -- we know we're 3486 * inserting at the end of the new map. 3487 */ 3488 vm_map_entry_link(new_map, new_map->header.prev, 3489 new_entry); 3490 vmspace_map_entry_forked(vm1, vm2, new_entry); 3491 3492 /* 3493 * Update the physical map 3494 */ 3495 pmap_copy(new_map->pmap, old_map->pmap, 3496 new_entry->start, 3497 (old_entry->end - old_entry->start), 3498 old_entry->start); 3499 break; 3500 3501 case VM_INHERIT_COPY: 3502 /* 3503 * Clone the entry and link into the map. 3504 */ 3505 new_entry = vm_map_entry_create(new_map); 3506 *new_entry = *old_entry; 3507 /* 3508 * Copied entry is COW over the old object. 3509 */ 3510 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | 3511 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); 3512 new_entry->wiring_thread = NULL; 3513 new_entry->wired_count = 0; 3514 new_entry->object.vm_object = NULL; 3515 new_entry->cred = NULL; 3516 vm_map_entry_link(new_map, new_map->header.prev, 3517 new_entry); 3518 vmspace_map_entry_forked(vm1, vm2, new_entry); 3519 vm_map_copy_entry(old_map, new_map, old_entry, 3520 new_entry, fork_charge); 3521 break; 3522 3523 case VM_INHERIT_ZERO: 3524 /* 3525 * Create a new anonymous mapping entry modelled from 3526 * the old one. 3527 */ 3528 new_entry = vm_map_entry_create(new_map); 3529 memset(new_entry, 0, sizeof(*new_entry)); 3530 3531 new_entry->start = old_entry->start; 3532 new_entry->end = old_entry->end; 3533 new_entry->eflags = old_entry->eflags & 3534 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION | 3535 MAP_ENTRY_VN_WRITECNT); 3536 new_entry->protection = old_entry->protection; 3537 new_entry->max_protection = old_entry->max_protection; 3538 new_entry->inheritance = VM_INHERIT_ZERO; 3539 3540 vm_map_entry_link(new_map, new_map->header.prev, 3541 new_entry); 3542 vmspace_map_entry_forked(vm1, vm2, new_entry); 3543 3544 new_entry->cred = curthread->td_ucred; 3545 crhold(new_entry->cred); 3546 *fork_charge += (new_entry->end - new_entry->start); 3547 3548 break; 3549 } 3550 old_entry = old_entry->next; 3551 } 3552 /* 3553 * Use inlined vm_map_unlock() to postpone handling the deferred 3554 * map entries, which cannot be done until both old_map and 3555 * new_map locks are released. 3556 */ 3557 sx_xunlock(&old_map->lock); 3558 sx_xunlock(&new_map->lock); 3559 vm_map_process_deferred(); 3560 3561 return (vm2); 3562} 3563 3564/* 3565 * Create a process's stack for exec_new_vmspace(). This function is never 3566 * asked to wire the newly created stack. 3567 */ 3568int 3569vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3570 vm_prot_t prot, vm_prot_t max, int cow) 3571{ 3572 vm_size_t growsize, init_ssize; 3573 rlim_t vmemlim; 3574 int rv; 3575 3576 MPASS((map->flags & MAP_WIREFUTURE) == 0); 3577 growsize = sgrowsiz; 3578 init_ssize = (max_ssize < growsize) ? max_ssize : growsize; 3579 vm_map_lock(map); 3580 vmemlim = lim_cur(curthread, RLIMIT_VMEM); 3581 /* If we would blow our VMEM resource limit, no go */ 3582 if (map->size + init_ssize > vmemlim) { 3583 rv = KERN_NO_SPACE; 3584 goto out; 3585 } 3586 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, 3587 max, cow); 3588out: 3589 vm_map_unlock(map); 3590 return (rv); 3591} 3592 3593static int stack_guard_page = 1; 3594SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN, 3595 &stack_guard_page, 0, 3596 "Specifies the number of guard pages for a stack that grows"); 3597 3598static int 3599vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, 3600 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) 3601{ 3602 vm_map_entry_t new_entry, prev_entry; 3603 vm_offset_t bot, gap_bot, gap_top, top; 3604 vm_size_t init_ssize, sgp; 3605 int orient, rv; 3606 3607 /* 3608 * The stack orientation is piggybacked with the cow argument. 3609 * Extract it into orient and mask the cow argument so that we 3610 * don't pass it around further. 3611 */ 3612 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP); 3613 KASSERT(orient != 0, ("No stack grow direction")); 3614 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP), 3615 ("bi-dir stack")); 3616 3617 if (addrbos < vm_map_min(map) || 3618 addrbos + max_ssize > vm_map_max(map) || 3619 addrbos + max_ssize <= addrbos) 3620 return (KERN_INVALID_ADDRESS); 3621 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE; 3622 if (sgp >= max_ssize) 3623 return (KERN_INVALID_ARGUMENT); 3624 3625 init_ssize = growsize; 3626 if (max_ssize < init_ssize + sgp) 3627 init_ssize = max_ssize - sgp; 3628 3629 /* If addr is already mapped, no go */ 3630 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) 3631 return (KERN_NO_SPACE); 3632 3633 /* 3634 * If we can't accommodate max_ssize in the current mapping, no go. 3635 */ 3636 if (prev_entry->next->start < addrbos + max_ssize) 3637 return (KERN_NO_SPACE); 3638 3639 /* 3640 * We initially map a stack of only init_ssize. We will grow as 3641 * needed later. Depending on the orientation of the stack (i.e. 3642 * the grow direction) we either map at the top of the range, the 3643 * bottom of the range or in the middle. 3644 * 3645 * Note: we would normally expect prot and max to be VM_PROT_ALL, 3646 * and cow to be 0. Possibly we should eliminate these as input 3647 * parameters, and just pass these values here in the insert call. 3648 */ 3649 if (orient == MAP_STACK_GROWS_DOWN) { 3650 bot = addrbos + max_ssize - init_ssize; 3651 top = bot + init_ssize; 3652 gap_bot = addrbos; 3653 gap_top = bot; 3654 } else /* if (orient == MAP_STACK_GROWS_UP) */ { 3655 bot = addrbos; 3656 top = bot + init_ssize; 3657 gap_bot = top; 3658 gap_top = addrbos + max_ssize; 3659 } 3660 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); 3661 if (rv != KERN_SUCCESS) 3662 return (rv); 3663 new_entry = prev_entry->next; 3664 KASSERT(new_entry->end == top || new_entry->start == bot, 3665 ("Bad entry start/end for new stack entry")); 3666 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 || 3667 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, 3668 ("new entry lacks MAP_ENTRY_GROWS_DOWN")); 3669 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 || 3670 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0, 3671 ("new entry lacks MAP_ENTRY_GROWS_UP")); 3672 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE, 3673 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ? 3674 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP)); 3675 if (rv != KERN_SUCCESS) 3676 (void)vm_map_delete(map, bot, top); 3677 return (rv); 3678} 3679 3680/* 3681 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we 3682 * successfully grow the stack. 3683 */ 3684static int 3685vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry) 3686{ 3687 vm_map_entry_t stack_entry; 3688 struct proc *p; 3689 struct vmspace *vm; 3690 struct ucred *cred; 3691 vm_offset_t gap_end, gap_start, grow_start; 3692 size_t grow_amount, guard, max_grow; 3693 rlim_t lmemlim, stacklim, vmemlim; 3694 int rv, rv1; 3695 bool gap_deleted, grow_down, is_procstack; 3696#ifdef notyet 3697 uint64_t limit; 3698#endif 3699#ifdef RACCT 3700 int error; 3701#endif 3702 3703 p = curproc; 3704 vm = p->p_vmspace; 3705 3706 /* 3707 * Disallow stack growth when the access is performed by a 3708 * debugger or AIO daemon. The reason is that the wrong 3709 * resource limits are applied. 3710 */ 3711 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL) 3712 return (KERN_FAILURE); 3713 3714 MPASS(!map->system_map); 3715 3716 guard = stack_guard_page * PAGE_SIZE; 3717 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK); 3718 stacklim = lim_cur(curthread, RLIMIT_STACK); 3719 vmemlim = lim_cur(curthread, RLIMIT_VMEM); 3720retry: 3721 /* If addr is not in a hole for a stack grow area, no need to grow. */ 3722 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry)) 3723 return (KERN_FAILURE); 3724 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0) 3725 return (KERN_SUCCESS); 3726 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) { 3727 stack_entry = gap_entry->next; 3728 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 || 3729 stack_entry->start != gap_entry->end) 3730 return (KERN_FAILURE); 3731 grow_amount = round_page(stack_entry->start - addr); 3732 grow_down = true; 3733 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) { 3734 stack_entry = gap_entry->prev; 3735 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 || 3736 stack_entry->end != gap_entry->start) 3737 return (KERN_FAILURE); 3738 grow_amount = round_page(addr + 1 - stack_entry->end); 3739 grow_down = false; 3740 } else { 3741 return (KERN_FAILURE); 3742 } 3743 max_grow = gap_entry->end - gap_entry->start; 3744 if (guard > max_grow) 3745 return (KERN_NO_SPACE); 3746 max_grow -= guard; 3747 if (grow_amount > max_grow) 3748 return (KERN_NO_SPACE); 3749 3750 /* 3751 * If this is the main process stack, see if we're over the stack 3752 * limit. 3753 */ 3754 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr && 3755 addr < (vm_offset_t)p->p_sysent->sv_usrstack; 3756 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) 3757 return (KERN_NO_SPACE); 3758 3759#ifdef RACCT 3760 if (racct_enable) { 3761 PROC_LOCK(p); 3762 if (is_procstack && racct_set(p, RACCT_STACK, 3763 ctob(vm->vm_ssize) + grow_amount)) { 3764 PROC_UNLOCK(p); 3765 return (KERN_NO_SPACE); 3766 } 3767 PROC_UNLOCK(p); 3768 } 3769#endif 3770 3771 grow_amount = roundup(grow_amount, sgrowsiz); 3772 if (grow_amount > max_grow) 3773 grow_amount = max_grow; 3774 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { 3775 grow_amount = trunc_page((vm_size_t)stacklim) - 3776 ctob(vm->vm_ssize); 3777 } 3778 3779#ifdef notyet 3780 PROC_LOCK(p); 3781 limit = racct_get_available(p, RACCT_STACK); 3782 PROC_UNLOCK(p); 3783 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) 3784 grow_amount = limit - ctob(vm->vm_ssize); 3785#endif 3786 3787 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) { 3788 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { 3789 rv = KERN_NO_SPACE; 3790 goto out; 3791 } 3792#ifdef RACCT 3793 if (racct_enable) { 3794 PROC_LOCK(p); 3795 if (racct_set(p, RACCT_MEMLOCK, 3796 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { 3797 PROC_UNLOCK(p); 3798 rv = KERN_NO_SPACE; 3799 goto out; 3800 } 3801 PROC_UNLOCK(p); 3802 } 3803#endif 3804 } 3805 3806 /* If we would blow our VMEM resource limit, no go */ 3807 if (map->size + grow_amount > vmemlim) { 3808 rv = KERN_NO_SPACE; 3809 goto out; 3810 } 3811#ifdef RACCT 3812 if (racct_enable) { 3813 PROC_LOCK(p); 3814 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { 3815 PROC_UNLOCK(p); 3816 rv = KERN_NO_SPACE; 3817 goto out; 3818 } 3819 PROC_UNLOCK(p); 3820 } 3821#endif 3822 3823 if (vm_map_lock_upgrade(map)) { 3824 gap_entry = NULL; 3825 vm_map_lock_read(map); 3826 goto retry; 3827 } 3828 3829 if (grow_down) { 3830 grow_start = gap_entry->end - grow_amount; 3831 if (gap_entry->start + grow_amount == gap_entry->end) { 3832 gap_start = gap_entry->start; 3833 gap_end = gap_entry->end; 3834 vm_map_entry_delete(map, gap_entry); 3835 gap_deleted = true; 3836 } else { 3837 MPASS(gap_entry->start < gap_entry->end - grow_amount); 3838 gap_entry->end -= grow_amount; 3839 vm_map_entry_resize_free(map, gap_entry); 3840 gap_deleted = false; 3841 } 3842 rv = vm_map_insert(map, NULL, 0, grow_start, 3843 grow_start + grow_amount, 3844 stack_entry->protection, stack_entry->max_protection, 3845 MAP_STACK_GROWS_DOWN); 3846 if (rv != KERN_SUCCESS) { 3847 if (gap_deleted) { 3848 rv1 = vm_map_insert(map, NULL, 0, gap_start, 3849 gap_end, VM_PROT_NONE, VM_PROT_NONE, 3850 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN); 3851 MPASS(rv1 == KERN_SUCCESS); 3852 } else { 3853 gap_entry->end += grow_amount; 3854 vm_map_entry_resize_free(map, gap_entry); 3855 } 3856 } 3857 } else { 3858 grow_start = stack_entry->end; 3859 cred = stack_entry->cred; 3860 if (cred == NULL && stack_entry->object.vm_object != NULL) 3861 cred = stack_entry->object.vm_object->cred; 3862 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) 3863 rv = KERN_NO_SPACE; 3864 /* Grow the underlying object if applicable. */ 3865 else if (stack_entry->object.vm_object == NULL || 3866 vm_object_coalesce(stack_entry->object.vm_object, 3867 stack_entry->offset, 3868 (vm_size_t)(stack_entry->end - stack_entry->start), 3869 (vm_size_t)grow_amount, cred != NULL)) { 3870 if (gap_entry->start + grow_amount == gap_entry->end) 3871 vm_map_entry_delete(map, gap_entry); 3872 else 3873 gap_entry->start += grow_amount; 3874 stack_entry->end += grow_amount; 3875 map->size += grow_amount; 3876 vm_map_entry_resize_free(map, stack_entry); 3877 rv = KERN_SUCCESS; 3878 } else 3879 rv = KERN_FAILURE; 3880 } 3881 if (rv == KERN_SUCCESS && is_procstack) 3882 vm->vm_ssize += btoc(grow_amount); 3883 3884 /* 3885 * Heed the MAP_WIREFUTURE flag if it was set for this process. 3886 */ 3887 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) { 3888 vm_map_unlock(map); 3889 vm_map_wire(map, grow_start, grow_start + grow_amount, 3890 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 3891 vm_map_lock_read(map); 3892 } else 3893 vm_map_lock_downgrade(map); 3894 3895out: 3896#ifdef RACCT 3897 if (racct_enable && rv != KERN_SUCCESS) { 3898 PROC_LOCK(p); 3899 error = racct_set(p, RACCT_VMEM, map->size); 3900 KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); 3901 if (!old_mlock) { 3902 error = racct_set(p, RACCT_MEMLOCK, 3903 ptoa(pmap_wired_count(map->pmap))); 3904 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); 3905 } 3906 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); 3907 KASSERT(error == 0, ("decreasing RACCT_STACK failed")); 3908 PROC_UNLOCK(p); 3909 } 3910#endif 3911 3912 return (rv); 3913} 3914 3915/* 3916 * Unshare the specified VM space for exec. If other processes are 3917 * mapped to it, then create a new one. The new vmspace is null. 3918 */ 3919int 3920vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) 3921{ 3922 struct vmspace *oldvmspace = p->p_vmspace; 3923 struct vmspace *newvmspace; 3924 3925 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, 3926 ("vmspace_exec recursed")); 3927 newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit); 3928 if (newvmspace == NULL) 3929 return (ENOMEM); 3930 newvmspace->vm_swrss = oldvmspace->vm_swrss; 3931 /* 3932 * This code is written like this for prototype purposes. The 3933 * goal is to avoid running down the vmspace here, but let the 3934 * other process's that are still using the vmspace to finally 3935 * run it down. Even though there is little or no chance of blocking 3936 * here, it is a good idea to keep this form for future mods. 3937 */ 3938 PROC_VMSPACE_LOCK(p); 3939 p->p_vmspace = newvmspace; 3940 PROC_VMSPACE_UNLOCK(p); 3941 if (p == curthread->td_proc) 3942 pmap_activate(curthread); 3943 curthread->td_pflags |= TDP_EXECVMSPC; 3944 return (0); 3945} 3946 3947/* 3948 * Unshare the specified VM space for forcing COW. This 3949 * is called by rfork, for the (RFMEM|RFPROC) == 0 case. 3950 */ 3951int 3952vmspace_unshare(struct proc *p) 3953{ 3954 struct vmspace *oldvmspace = p->p_vmspace; 3955 struct vmspace *newvmspace; 3956 vm_ooffset_t fork_charge; 3957 3958 if (oldvmspace->vm_refcnt == 1) 3959 return (0); 3960 fork_charge = 0; 3961 newvmspace = vmspace_fork(oldvmspace, &fork_charge); 3962 if (newvmspace == NULL) 3963 return (ENOMEM); 3964 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { 3965 vmspace_free(newvmspace); 3966 return (ENOMEM); 3967 } 3968 PROC_VMSPACE_LOCK(p); 3969 p->p_vmspace = newvmspace; 3970 PROC_VMSPACE_UNLOCK(p); 3971 if (p == curthread->td_proc) 3972 pmap_activate(curthread); 3973 vmspace_free(oldvmspace); 3974 return (0); 3975} 3976 3977/* 3978 * vm_map_lookup: 3979 * 3980 * Finds the VM object, offset, and 3981 * protection for a given virtual address in the 3982 * specified map, assuming a page fault of the 3983 * type specified. 3984 * 3985 * Leaves the map in question locked for read; return 3986 * values are guaranteed until a vm_map_lookup_done 3987 * call is performed. Note that the map argument 3988 * is in/out; the returned map must be used in 3989 * the call to vm_map_lookup_done. 3990 * 3991 * A handle (out_entry) is returned for use in 3992 * vm_map_lookup_done, to make that fast. 3993 * 3994 * If a lookup is requested with "write protection" 3995 * specified, the map may be changed to perform virtual 3996 * copying operations, although the data referenced will 3997 * remain the same. 3998 */ 3999int 4000vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ 4001 vm_offset_t vaddr, 4002 vm_prot_t fault_typea, 4003 vm_map_entry_t *out_entry, /* OUT */ 4004 vm_object_t *object, /* OUT */ 4005 vm_pindex_t *pindex, /* OUT */ 4006 vm_prot_t *out_prot, /* OUT */ 4007 boolean_t *wired) /* OUT */ 4008{ 4009 vm_map_entry_t entry; 4010 vm_map_t map = *var_map; 4011 vm_prot_t prot; 4012 vm_prot_t fault_type = fault_typea; 4013 vm_object_t eobject; 4014 vm_size_t size; 4015 struct ucred *cred; 4016 4017RetryLookup: 4018 4019 vm_map_lock_read(map); 4020 4021RetryLookupLocked: 4022 /* 4023 * Lookup the faulting address. 4024 */ 4025 if (!vm_map_lookup_entry(map, vaddr, out_entry)) { 4026 vm_map_unlock_read(map); 4027 return (KERN_INVALID_ADDRESS); 4028 } 4029 4030 entry = *out_entry; 4031 4032 /* 4033 * Handle submaps. 4034 */ 4035 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4036 vm_map_t old_map = map; 4037 4038 *var_map = map = entry->object.sub_map; 4039 vm_map_unlock_read(old_map); 4040 goto RetryLookup; 4041 } 4042 4043 /* 4044 * Check whether this task is allowed to have this page. 4045 */ 4046 prot = entry->protection; 4047 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) { 4048 fault_typea &= ~VM_PROT_FAULT_LOOKUP; 4049 if (prot == VM_PROT_NONE && map != kernel_map && 4050 (entry->eflags & MAP_ENTRY_GUARD) != 0 && 4051 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN | 4052 MAP_ENTRY_STACK_GAP_UP)) != 0 && 4053 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS) 4054 goto RetryLookupLocked; 4055 } 4056 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4057 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { 4058 vm_map_unlock_read(map); 4059 return (KERN_PROTECTION_FAILURE); 4060 } 4061 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags & 4062 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) != 4063 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY), 4064 ("entry %p flags %x", entry, entry->eflags)); 4065 if ((fault_typea & VM_PROT_COPY) != 0 && 4066 (entry->max_protection & VM_PROT_WRITE) == 0 && 4067 (entry->eflags & MAP_ENTRY_COW) == 0) { 4068 vm_map_unlock_read(map); 4069 return (KERN_PROTECTION_FAILURE); 4070 } 4071 4072 /* 4073 * If this page is not pageable, we have to get it for all possible 4074 * accesses. 4075 */ 4076 *wired = (entry->wired_count != 0); 4077 if (*wired) 4078 fault_type = entry->protection; 4079 size = entry->end - entry->start; 4080 /* 4081 * If the entry was copy-on-write, we either ... 4082 */ 4083 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4084 /* 4085 * If we want to write the page, we may as well handle that 4086 * now since we've got the map locked. 4087 * 4088 * If we don't need to write the page, we just demote the 4089 * permissions allowed. 4090 */ 4091 if ((fault_type & VM_PROT_WRITE) != 0 || 4092 (fault_typea & VM_PROT_COPY) != 0) { 4093 /* 4094 * Make a new object, and place it in the object 4095 * chain. Note that no new references have appeared 4096 * -- one just moved from the map to the new 4097 * object. 4098 */ 4099 if (vm_map_lock_upgrade(map)) 4100 goto RetryLookup; 4101 4102 if (entry->cred == NULL) { 4103 /* 4104 * The debugger owner is charged for 4105 * the memory. 4106 */ 4107 cred = curthread->td_ucred; 4108 crhold(cred); 4109 if (!swap_reserve_by_cred(size, cred)) { 4110 crfree(cred); 4111 vm_map_unlock(map); 4112 return (KERN_RESOURCE_SHORTAGE); 4113 } 4114 entry->cred = cred; 4115 } 4116 vm_object_shadow(&entry->object.vm_object, 4117 &entry->offset, size); 4118 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; 4119 eobject = entry->object.vm_object; 4120 if (eobject->cred != NULL) { 4121 /* 4122 * The object was not shadowed. 4123 */ 4124 swap_release_by_cred(size, entry->cred); 4125 crfree(entry->cred); 4126 entry->cred = NULL; 4127 } else if (entry->cred != NULL) { 4128 VM_OBJECT_WLOCK(eobject); 4129 eobject->cred = entry->cred; 4130 eobject->charge = size; 4131 VM_OBJECT_WUNLOCK(eobject); 4132 entry->cred = NULL; 4133 } 4134 4135 vm_map_lock_downgrade(map); 4136 } else { 4137 /* 4138 * We're attempting to read a copy-on-write page -- 4139 * don't allow writes. 4140 */ 4141 prot &= ~VM_PROT_WRITE; 4142 } 4143 } 4144 4145 /* 4146 * Create an object if necessary. 4147 */ 4148 if (entry->object.vm_object == NULL && 4149 !map->system_map) { 4150 if (vm_map_lock_upgrade(map)) 4151 goto RetryLookup; 4152 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, 4153 atop(size)); 4154 entry->offset = 0; 4155 if (entry->cred != NULL) { 4156 VM_OBJECT_WLOCK(entry->object.vm_object); 4157 entry->object.vm_object->cred = entry->cred; 4158 entry->object.vm_object->charge = size; 4159 VM_OBJECT_WUNLOCK(entry->object.vm_object); 4160 entry->cred = NULL; 4161 } 4162 vm_map_lock_downgrade(map); 4163 } 4164 4165 /* 4166 * Return the object/offset from this entry. If the entry was 4167 * copy-on-write or empty, it has been fixed up. 4168 */ 4169 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset); 4170 *object = entry->object.vm_object; 4171 4172 *out_prot = prot; 4173 return (KERN_SUCCESS); 4174} 4175 4176/* 4177 * vm_map_lookup_locked: 4178 * 4179 * Lookup the faulting address. A version of vm_map_lookup that returns 4180 * KERN_FAILURE instead of blocking on map lock or memory allocation. 4181 */ 4182int 4183vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ 4184 vm_offset_t vaddr, 4185 vm_prot_t fault_typea, 4186 vm_map_entry_t *out_entry, /* OUT */ 4187 vm_object_t *object, /* OUT */ 4188 vm_pindex_t *pindex, /* OUT */ 4189 vm_prot_t *out_prot, /* OUT */ 4190 boolean_t *wired) /* OUT */ 4191{ 4192 vm_map_entry_t entry; 4193 vm_map_t map = *var_map; 4194 vm_prot_t prot; 4195 vm_prot_t fault_type = fault_typea; 4196 4197 /* 4198 * Lookup the faulting address. 4199 */ 4200 if (!vm_map_lookup_entry(map, vaddr, out_entry)) 4201 return (KERN_INVALID_ADDRESS); 4202 4203 entry = *out_entry; 4204 4205 /* 4206 * Fail if the entry refers to a submap. 4207 */ 4208 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) 4209 return (KERN_FAILURE); 4210 4211 /* 4212 * Check whether this task is allowed to have this page. 4213 */ 4214 prot = entry->protection; 4215 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; 4216 if ((fault_type & prot) != fault_type) 4217 return (KERN_PROTECTION_FAILURE); 4218 4219 /* 4220 * If this page is not pageable, we have to get it for all possible 4221 * accesses. 4222 */ 4223 *wired = (entry->wired_count != 0); 4224 if (*wired) 4225 fault_type = entry->protection; 4226 4227 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { 4228 /* 4229 * Fail if the entry was copy-on-write for a write fault. 4230 */ 4231 if (fault_type & VM_PROT_WRITE) 4232 return (KERN_FAILURE); 4233 /* 4234 * We're attempting to read a copy-on-write page -- 4235 * don't allow writes. 4236 */ 4237 prot &= ~VM_PROT_WRITE; 4238 } 4239 4240 /* 4241 * Fail if an object should be created. 4242 */ 4243 if (entry->object.vm_object == NULL && !map->system_map) 4244 return (KERN_FAILURE); 4245 4246 /* 4247 * Return the object/offset from this entry. If the entry was 4248 * copy-on-write or empty, it has been fixed up. 4249 */ 4250 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset); 4251 *object = entry->object.vm_object; 4252 4253 *out_prot = prot; 4254 return (KERN_SUCCESS); 4255} 4256 4257/* 4258 * vm_map_lookup_done: 4259 * 4260 * Releases locks acquired by a vm_map_lookup 4261 * (according to the handle returned by that lookup). 4262 */ 4263void 4264vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) 4265{ 4266 /* 4267 * Unlock the main-level map 4268 */ 4269 vm_map_unlock_read(map); 4270} 4271 4272vm_offset_t 4273vm_map_max_KBI(const struct vm_map *map) 4274{ 4275 4276 return (vm_map_max(map)); 4277} 4278 4279vm_offset_t 4280vm_map_min_KBI(const struct vm_map *map) 4281{ 4282 4283 return (vm_map_min(map)); 4284} 4285 4286pmap_t 4287vm_map_pmap_KBI(vm_map_t map) 4288{ 4289 4290 return (map->pmap); 4291} 4292 4293#include "opt_ddb.h" 4294#ifdef DDB 4295#include <sys/kernel.h> 4296 4297#include <ddb/ddb.h> 4298 4299static void 4300vm_map_print(vm_map_t map) 4301{ 4302 vm_map_entry_t entry; 4303 4304 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", 4305 (void *)map, 4306 (void *)map->pmap, map->nentries, map->timestamp); 4307 4308 db_indent += 2; 4309 for (entry = map->header.next; entry != &map->header; 4310 entry = entry->next) { 4311 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n", 4312 (void *)entry, (void *)entry->start, (void *)entry->end, 4313 entry->eflags); 4314 { 4315 static char *inheritance_name[4] = 4316 {"share", "copy", "none", "donate_copy"}; 4317 4318 db_iprintf(" prot=%x/%x/%s", 4319 entry->protection, 4320 entry->max_protection, 4321 inheritance_name[(int)(unsigned char)entry->inheritance]); 4322 if (entry->wired_count != 0) 4323 db_printf(", wired"); 4324 } 4325 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 4326 db_printf(", share=%p, offset=0x%jx\n", 4327 (void *)entry->object.sub_map, 4328 (uintmax_t)entry->offset); 4329 if ((entry->prev == &map->header) || 4330 (entry->prev->object.sub_map != 4331 entry->object.sub_map)) { 4332 db_indent += 2; 4333 vm_map_print((vm_map_t)entry->object.sub_map); 4334 db_indent -= 2; 4335 } 4336 } else { 4337 if (entry->cred != NULL) 4338 db_printf(", ruid %d", entry->cred->cr_ruid); 4339 db_printf(", object=%p, offset=0x%jx", 4340 (void *)entry->object.vm_object, 4341 (uintmax_t)entry->offset); 4342 if (entry->object.vm_object && entry->object.vm_object->cred) 4343 db_printf(", obj ruid %d charge %jx", 4344 entry->object.vm_object->cred->cr_ruid, 4345 (uintmax_t)entry->object.vm_object->charge); 4346 if (entry->eflags & MAP_ENTRY_COW) 4347 db_printf(", copy (%s)", 4348 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); 4349 db_printf("\n"); 4350 4351 if ((entry->prev == &map->header) || 4352 (entry->prev->object.vm_object != 4353 entry->object.vm_object)) { 4354 db_indent += 2; 4355 vm_object_print((db_expr_t)(intptr_t) 4356 entry->object.vm_object, 4357 0, 0, (char *)0); 4358 db_indent -= 2; 4359 } 4360 } 4361 } 4362 db_indent -= 2; 4363} 4364 4365DB_SHOW_COMMAND(map, map) 4366{ 4367 4368 if (!have_addr) { 4369 db_printf("usage: show map <addr>\n"); 4370 return; 4371 } 4372 vm_map_print((vm_map_t)addr); 4373} 4374 4375DB_SHOW_COMMAND(procvm, procvm) 4376{ 4377 struct proc *p; 4378 4379 if (have_addr) { 4380 p = db_lookup_proc(addr); 4381 } else { 4382 p = curproc; 4383 } 4384 4385 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", 4386 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, 4387 (void *)vmspace_pmap(p->p_vmspace)); 4388 4389 vm_map_print((vm_map_t)&p->p_vmspace->vm_map); 4390} 4391 4392#endif /* DDB */ 4393