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