1/* $NetBSD: uvm_map.c,v 1.411 2024/02/09 22:08:38 andvar Exp $ */ 2 3/* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993, The Regents of the University of California. 6 * 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * The Mach Operating System project at Carnegie-Mellon University. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94 37 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp 38 * 39 * 40 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 41 * All rights reserved. 42 * 43 * Permission to use, copy, modify and distribute this software and 44 * its documentation is hereby granted, provided that both the copyright 45 * notice and this permission notice appear in all copies of the 46 * software, derivative works or modified versions, and any portions 47 * thereof, and that both notices appear in supporting documentation. 48 * 49 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 50 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 51 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 52 * 53 * Carnegie Mellon requests users of this software to return to 54 * 55 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 56 * School of Computer Science 57 * Carnegie Mellon University 58 * Pittsburgh PA 15213-3890 59 * 60 * any improvements or extensions that they make and grant Carnegie the 61 * rights to redistribute these changes. 62 */ 63 64/* 65 * uvm_map.c: uvm map operations 66 */ 67 68#include <sys/cdefs.h> 69__KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.411 2024/02/09 22:08:38 andvar Exp $"); 70 71#include "opt_ddb.h" 72#include "opt_pax.h" 73#include "opt_uvmhist.h" 74#include "opt_uvm.h" 75#include "opt_sysv.h" 76 77#include <sys/param.h> 78#include <sys/systm.h> 79#include <sys/mman.h> 80#include <sys/proc.h> 81#include <sys/pool.h> 82#include <sys/kernel.h> 83#include <sys/mount.h> 84#include <sys/pax.h> 85#include <sys/vnode.h> 86#include <sys/filedesc.h> 87#include <sys/lockdebug.h> 88#include <sys/atomic.h> 89#include <sys/sysctl.h> 90#ifndef __USER_VA0_IS_SAFE 91#include <sys/kauth.h> 92#include "opt_user_va0_disable_default.h" 93#endif 94 95#include <sys/shm.h> 96 97#include <uvm/uvm.h> 98#include <uvm/uvm_readahead.h> 99 100#if defined(DDB) || defined(DEBUGPRINT) 101#include <uvm/uvm_ddb.h> 102#endif 103 104#ifdef UVMHIST 105#ifndef UVMHIST_MAPHIST_SIZE 106#define UVMHIST_MAPHIST_SIZE 100 107#endif 108static struct kern_history_ent maphistbuf[UVMHIST_MAPHIST_SIZE]; 109UVMHIST_DEFINE(maphist) = UVMHIST_INITIALIZER(maphist, maphistbuf); 110#endif 111 112#if !defined(UVMMAP_COUNTERS) 113 114#define UVMMAP_EVCNT_DEFINE(name) /* nothing */ 115#define UVMMAP_EVCNT_INCR(ev) /* nothing */ 116#define UVMMAP_EVCNT_DECR(ev) /* nothing */ 117 118#else /* defined(UVMMAP_NOCOUNTERS) */ 119 120#include <sys/evcnt.h> 121#define UVMMAP_EVCNT_DEFINE(name) \ 122struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \ 123 "uvmmap", #name); \ 124EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name); 125#define UVMMAP_EVCNT_INCR(ev) uvmmap_evcnt_##ev.ev_count++ 126#define UVMMAP_EVCNT_DECR(ev) uvmmap_evcnt_##ev.ev_count-- 127 128#endif /* defined(UVMMAP_NOCOUNTERS) */ 129 130UVMMAP_EVCNT_DEFINE(ubackmerge) 131UVMMAP_EVCNT_DEFINE(uforwmerge) 132UVMMAP_EVCNT_DEFINE(ubimerge) 133UVMMAP_EVCNT_DEFINE(unomerge) 134UVMMAP_EVCNT_DEFINE(kbackmerge) 135UVMMAP_EVCNT_DEFINE(kforwmerge) 136UVMMAP_EVCNT_DEFINE(kbimerge) 137UVMMAP_EVCNT_DEFINE(knomerge) 138UVMMAP_EVCNT_DEFINE(map_call) 139UVMMAP_EVCNT_DEFINE(mlk_call) 140UVMMAP_EVCNT_DEFINE(mlk_hint) 141UVMMAP_EVCNT_DEFINE(mlk_tree) 142UVMMAP_EVCNT_DEFINE(mlk_treeloop) 143 144const char vmmapbsy[] = "vmmapbsy"; 145 146/* 147 * cache for dynamically-allocated map entries. 148 */ 149 150static struct pool_cache uvm_map_entry_cache; 151 152#ifdef PMAP_GROWKERNEL 153/* 154 * This global represents the end of the kernel virtual address 155 * space. If we want to exceed this, we must grow the kernel 156 * virtual address space dynamically. 157 * 158 * Note, this variable is locked by kernel_map's lock. 159 */ 160vaddr_t uvm_maxkaddr; 161#endif 162 163#ifndef __USER_VA0_IS_SAFE 164#ifndef __USER_VA0_DISABLE_DEFAULT 165#define __USER_VA0_DISABLE_DEFAULT 1 166#endif 167#ifdef USER_VA0_DISABLE_DEFAULT /* kernel config option overrides */ 168#undef __USER_VA0_DISABLE_DEFAULT 169#define __USER_VA0_DISABLE_DEFAULT USER_VA0_DISABLE_DEFAULT 170#endif 171int user_va0_disable = __USER_VA0_DISABLE_DEFAULT; 172#endif 173 174/* 175 * macros 176 */ 177 178/* 179 * uvm_map_align_va: round down or up virtual address 180 */ 181static __inline void 182uvm_map_align_va(vaddr_t *vap, vsize_t align, int topdown) 183{ 184 185 KASSERT(powerof2(align)); 186 187 if (align != 0 && (*vap & (align - 1)) != 0) { 188 if (topdown) 189 *vap = rounddown2(*vap, align); 190 else 191 *vap = roundup2(*vap, align); 192 } 193} 194 195/* 196 * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging 197 */ 198extern struct vm_map *pager_map; 199 200#define UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \ 201 prot, maxprot, inh, adv, wire) \ 202 ((ent)->etype == (type) && \ 203 (((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE)) == 0 && \ 204 (ent)->object.uvm_obj == (uobj) && \ 205 (ent)->protection == (prot) && \ 206 (ent)->max_protection == (maxprot) && \ 207 (ent)->inheritance == (inh) && \ 208 (ent)->advice == (adv) && \ 209 (ent)->wired_count == (wire)) 210 211/* 212 * uvm_map_entry_link: insert entry into a map 213 * 214 * => map must be locked 215 */ 216#define uvm_map_entry_link(map, after_where, entry) do { \ 217 uvm_mapent_check(entry); \ 218 (map)->nentries++; \ 219 (entry)->prev = (after_where); \ 220 (entry)->next = (after_where)->next; \ 221 (entry)->prev->next = (entry); \ 222 (entry)->next->prev = (entry); \ 223 uvm_rb_insert((map), (entry)); \ 224} while (/*CONSTCOND*/ 0) 225 226/* 227 * uvm_map_entry_unlink: remove entry from a map 228 * 229 * => map must be locked 230 */ 231#define uvm_map_entry_unlink(map, entry) do { \ 232 KASSERT((entry) != (map)->first_free); \ 233 KASSERT((entry) != (map)->hint); \ 234 uvm_mapent_check(entry); \ 235 (map)->nentries--; \ 236 (entry)->next->prev = (entry)->prev; \ 237 (entry)->prev->next = (entry)->next; \ 238 uvm_rb_remove((map), (entry)); \ 239} while (/*CONSTCOND*/ 0) 240 241/* 242 * SAVE_HINT: saves the specified entry as the hint for future lookups. 243 * 244 * => map need not be locked. 245 */ 246#define SAVE_HINT(map, check, value) do { \ 247 if ((map)->hint == (check)) \ 248 (map)->hint = (value); \ 249} while (/*CONSTCOND*/ 0) 250 251/* 252 * clear_hints: ensure that hints don't point to the entry. 253 * 254 * => map must be write-locked. 255 */ 256static void 257clear_hints(struct vm_map *map, struct vm_map_entry *ent) 258{ 259 260 SAVE_HINT(map, ent, ent->prev); 261 if (map->first_free == ent) { 262 map->first_free = ent->prev; 263 } 264} 265 266/* 267 * VM_MAP_RANGE_CHECK: check and correct range 268 * 269 * => map must at least be read locked 270 */ 271 272#define VM_MAP_RANGE_CHECK(map, start, end) do { \ 273 if (start < vm_map_min(map)) \ 274 start = vm_map_min(map); \ 275 if (end > vm_map_max(map)) \ 276 end = vm_map_max(map); \ 277 if (start > end) \ 278 start = end; \ 279} while (/*CONSTCOND*/ 0) 280 281/* 282 * local prototypes 283 */ 284 285static struct vm_map_entry * 286 uvm_mapent_alloc(struct vm_map *, int); 287static void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *); 288static void uvm_mapent_free(struct vm_map_entry *); 289#if defined(DEBUG) 290static void _uvm_mapent_check(const struct vm_map_entry *, int); 291#define uvm_mapent_check(map) _uvm_mapent_check(map, __LINE__) 292#else /* defined(DEBUG) */ 293#define uvm_mapent_check(e) /* nothing */ 294#endif /* defined(DEBUG) */ 295 296static void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *); 297static void uvm_map_reference_amap(struct vm_map_entry *, int); 298static int uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int, 299 int, struct vm_map_entry *); 300static void uvm_map_unreference_amap(struct vm_map_entry *, int); 301 302int _uvm_map_sanity(struct vm_map *); 303int _uvm_tree_sanity(struct vm_map *); 304static vsize_t uvm_rb_maxgap(const struct vm_map_entry *); 305 306#define ROOT_ENTRY(map) ((struct vm_map_entry *)(map)->rb_tree.rbt_root) 307#define LEFT_ENTRY(entry) ((struct vm_map_entry *)(entry)->rb_node.rb_left) 308#define RIGHT_ENTRY(entry) ((struct vm_map_entry *)(entry)->rb_node.rb_right) 309#define PARENT_ENTRY(map, entry) \ 310 (ROOT_ENTRY(map) == (entry) \ 311 ? NULL : (struct vm_map_entry *)RB_FATHER(&(entry)->rb_node)) 312 313/* 314 * These get filled in if/when SYSVSHM shared memory code is loaded 315 * 316 * We do this with function pointers rather the #ifdef SYSVSHM so the 317 * SYSVSHM code can be loaded and unloaded 318 */ 319void (*uvm_shmexit)(struct vmspace *) = NULL; 320void (*uvm_shmfork)(struct vmspace *, struct vmspace *) = NULL; 321 322static int 323uvm_map_compare_nodes(void *ctx, const void *nparent, const void *nkey) 324{ 325 const struct vm_map_entry *eparent = nparent; 326 const struct vm_map_entry *ekey = nkey; 327 328 KASSERT(eparent->start < ekey->start || eparent->start >= ekey->end); 329 KASSERT(ekey->start < eparent->start || ekey->start >= eparent->end); 330 331 if (eparent->start < ekey->start) 332 return -1; 333 if (eparent->end >= ekey->start) 334 return 1; 335 return 0; 336} 337 338static int 339uvm_map_compare_key(void *ctx, const void *nparent, const void *vkey) 340{ 341 const struct vm_map_entry *eparent = nparent; 342 const vaddr_t va = *(const vaddr_t *) vkey; 343 344 if (eparent->start < va) 345 return -1; 346 if (eparent->end >= va) 347 return 1; 348 return 0; 349} 350 351static const rb_tree_ops_t uvm_map_tree_ops = { 352 .rbto_compare_nodes = uvm_map_compare_nodes, 353 .rbto_compare_key = uvm_map_compare_key, 354 .rbto_node_offset = offsetof(struct vm_map_entry, rb_node), 355 .rbto_context = NULL 356}; 357 358/* 359 * uvm_rb_gap: return the gap size between our entry and next entry. 360 */ 361static inline vsize_t 362uvm_rb_gap(const struct vm_map_entry *entry) 363{ 364 365 KASSERT(entry->next != NULL); 366 return entry->next->start - entry->end; 367} 368 369static vsize_t 370uvm_rb_maxgap(const struct vm_map_entry *entry) 371{ 372 struct vm_map_entry *child; 373 vsize_t maxgap = entry->gap; 374 375 /* 376 * We need maxgap to be the largest gap of us or any of our 377 * descendents. Since each of our children's maxgap is the 378 * cached value of their largest gap of themselves or their 379 * descendents, we can just use that value and avoid recursing 380 * down the tree to calculate it. 381 */ 382 if ((child = LEFT_ENTRY(entry)) != NULL && maxgap < child->maxgap) 383 maxgap = child->maxgap; 384 385 if ((child = RIGHT_ENTRY(entry)) != NULL && maxgap < child->maxgap) 386 maxgap = child->maxgap; 387 388 return maxgap; 389} 390 391static void 392uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry) 393{ 394 struct vm_map_entry *parent; 395 396 KASSERT(entry->gap == uvm_rb_gap(entry)); 397 entry->maxgap = uvm_rb_maxgap(entry); 398 399 while ((parent = PARENT_ENTRY(map, entry)) != NULL) { 400 struct vm_map_entry *brother; 401 vsize_t maxgap = parent->gap; 402 unsigned int which; 403 404 KDASSERT(parent->gap == uvm_rb_gap(parent)); 405 if (maxgap < entry->maxgap) 406 maxgap = entry->maxgap; 407 /* 408 * Since we work towards the root, we know entry's maxgap 409 * value is OK, but its brothers may now be out-of-date due 410 * to rebalancing. So refresh it. 411 */ 412 which = RB_POSITION(&entry->rb_node) ^ RB_DIR_OTHER; 413 brother = (struct vm_map_entry *)parent->rb_node.rb_nodes[which]; 414 if (brother != NULL) { 415 KDASSERT(brother->gap == uvm_rb_gap(brother)); 416 brother->maxgap = uvm_rb_maxgap(brother); 417 if (maxgap < brother->maxgap) 418 maxgap = brother->maxgap; 419 } 420 421 parent->maxgap = maxgap; 422 entry = parent; 423 } 424} 425 426static void 427uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry) 428{ 429 struct vm_map_entry *ret __diagused; 430 431 entry->gap = entry->maxgap = uvm_rb_gap(entry); 432 if (entry->prev != &map->header) 433 entry->prev->gap = uvm_rb_gap(entry->prev); 434 435 ret = rb_tree_insert_node(&map->rb_tree, entry); 436 KASSERTMSG(ret == entry, 437 "uvm_rb_insert: map %p: duplicate entry %p", map, ret); 438 439 /* 440 * If the previous entry is not our immediate left child, then it's an 441 * ancestor and will be fixed up on the way to the root. We don't 442 * have to check entry->prev against &map->header since &map->header 443 * will never be in the tree. 444 */ 445 uvm_rb_fixup(map, 446 LEFT_ENTRY(entry) == entry->prev ? entry->prev : entry); 447} 448 449static void 450uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry) 451{ 452 struct vm_map_entry *prev_parent = NULL, *next_parent = NULL; 453 454 /* 455 * If we are removing an interior node, then an adjacent node will 456 * be used to replace its position in the tree. Therefore we will 457 * need to fixup the tree starting at the parent of the replacement 458 * node. So record their parents for later use. 459 */ 460 if (entry->prev != &map->header) 461 prev_parent = PARENT_ENTRY(map, entry->prev); 462 if (entry->next != &map->header) 463 next_parent = PARENT_ENTRY(map, entry->next); 464 465 rb_tree_remove_node(&map->rb_tree, entry); 466 467 /* 468 * If the previous node has a new parent, fixup the tree starting 469 * at the previous node's old parent. 470 */ 471 if (entry->prev != &map->header) { 472 /* 473 * Update the previous entry's gap due to our absence. 474 */ 475 entry->prev->gap = uvm_rb_gap(entry->prev); 476 uvm_rb_fixup(map, entry->prev); 477 if (prev_parent != NULL 478 && prev_parent != entry 479 && prev_parent != PARENT_ENTRY(map, entry->prev)) 480 uvm_rb_fixup(map, prev_parent); 481 } 482 483 /* 484 * If the next node has a new parent, fixup the tree starting 485 * at the next node's old parent. 486 */ 487 if (entry->next != &map->header) { 488 uvm_rb_fixup(map, entry->next); 489 if (next_parent != NULL 490 && next_parent != entry 491 && next_parent != PARENT_ENTRY(map, entry->next)) 492 uvm_rb_fixup(map, next_parent); 493 } 494} 495 496#if defined(DEBUG) 497int uvm_debug_check_map = 0; 498int uvm_debug_check_rbtree = 0; 499#define uvm_map_check(map, name) \ 500 _uvm_map_check((map), (name), __FILE__, __LINE__) 501static void 502_uvm_map_check(struct vm_map *map, const char *name, 503 const char *file, int line) 504{ 505 506 if ((uvm_debug_check_map && _uvm_map_sanity(map)) || 507 (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) { 508 panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)", 509 name, map, file, line); 510 } 511} 512#else /* defined(DEBUG) */ 513#define uvm_map_check(map, name) /* nothing */ 514#endif /* defined(DEBUG) */ 515 516#if defined(DEBUG) || defined(DDB) 517int 518_uvm_map_sanity(struct vm_map *map) 519{ 520 bool first_free_found = false; 521 bool hint_found = false; 522 const struct vm_map_entry *e; 523 struct vm_map_entry *hint = map->hint; 524 525 e = &map->header; 526 for (;;) { 527 if (map->first_free == e) { 528 first_free_found = true; 529 } else if (!first_free_found && e->next->start > e->end) { 530 printf("first_free %p should be %p\n", 531 map->first_free, e); 532 return -1; 533 } 534 if (hint == e) { 535 hint_found = true; 536 } 537 538 e = e->next; 539 if (e == &map->header) { 540 break; 541 } 542 } 543 if (!first_free_found) { 544 printf("stale first_free\n"); 545 return -1; 546 } 547 if (!hint_found) { 548 printf("stale hint\n"); 549 return -1; 550 } 551 return 0; 552} 553 554int 555_uvm_tree_sanity(struct vm_map *map) 556{ 557 struct vm_map_entry *tmp, *trtmp; 558 int n = 0, i = 1; 559 560 for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) { 561 if (tmp->gap != uvm_rb_gap(tmp)) { 562 printf("%d/%d gap %#lx != %#lx %s\n", 563 n + 1, map->nentries, 564 (ulong)tmp->gap, (ulong)uvm_rb_gap(tmp), 565 tmp->next == &map->header ? "(last)" : ""); 566 goto error; 567 } 568 /* 569 * If any entries are out of order, tmp->gap will be unsigned 570 * and will likely exceed the size of the map. 571 */ 572 if (tmp->gap >= vm_map_max(map) - vm_map_min(map)) { 573 printf("too large gap %zu\n", (size_t)tmp->gap); 574 goto error; 575 } 576 n++; 577 } 578 579 if (n != map->nentries) { 580 printf("nentries: %d vs %d\n", n, map->nentries); 581 goto error; 582 } 583 584 trtmp = NULL; 585 for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) { 586 if (tmp->maxgap != uvm_rb_maxgap(tmp)) { 587 printf("maxgap %#lx != %#lx\n", 588 (ulong)tmp->maxgap, 589 (ulong)uvm_rb_maxgap(tmp)); 590 goto error; 591 } 592 if (trtmp != NULL && trtmp->start >= tmp->start) { 593 printf("corrupt: 0x%"PRIxVADDR"x >= 0x%"PRIxVADDR"x\n", 594 trtmp->start, tmp->start); 595 goto error; 596 } 597 598 trtmp = tmp; 599 } 600 601 for (tmp = map->header.next; tmp != &map->header; 602 tmp = tmp->next, i++) { 603 trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_LEFT); 604 if (trtmp == NULL) 605 trtmp = &map->header; 606 if (tmp->prev != trtmp) { 607 printf("lookup: %d: %p->prev=%p: %p\n", 608 i, tmp, tmp->prev, trtmp); 609 goto error; 610 } 611 trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_RIGHT); 612 if (trtmp == NULL) 613 trtmp = &map->header; 614 if (tmp->next != trtmp) { 615 printf("lookup: %d: %p->next=%p: %p\n", 616 i, tmp, tmp->next, trtmp); 617 goto error; 618 } 619 trtmp = rb_tree_find_node(&map->rb_tree, &tmp->start); 620 if (trtmp != tmp) { 621 printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp, 622 PARENT_ENTRY(map, tmp)); 623 goto error; 624 } 625 } 626 627 return (0); 628 error: 629 return (-1); 630} 631#endif /* defined(DEBUG) || defined(DDB) */ 632 633/* 634 * vm_map_lock: acquire an exclusive (write) lock on a map. 635 * 636 * => The locking protocol provides for guaranteed upgrade from shared -> 637 * exclusive by whichever thread currently has the map marked busy. 638 * See "LOCKING PROTOCOL NOTES" in uvm_map.h. This is horrible; among 639 * other problems, it defeats any fairness guarantees provided by RW 640 * locks. 641 */ 642 643void 644vm_map_lock(struct vm_map *map) 645{ 646 647 for (;;) { 648 rw_enter(&map->lock, RW_WRITER); 649 if (map->busy == NULL || map->busy == curlwp) { 650 break; 651 } 652 mutex_enter(&map->misc_lock); 653 rw_exit(&map->lock); 654 if (map->busy != NULL) { 655 cv_wait(&map->cv, &map->misc_lock); 656 } 657 mutex_exit(&map->misc_lock); 658 } 659 map->timestamp++; 660} 661 662/* 663 * vm_map_lock_try: try to lock a map, failing if it is already locked. 664 */ 665 666bool 667vm_map_lock_try(struct vm_map *map) 668{ 669 670 if (!rw_tryenter(&map->lock, RW_WRITER)) { 671 return false; 672 } 673 if (map->busy != NULL) { 674 rw_exit(&map->lock); 675 return false; 676 } 677 map->timestamp++; 678 return true; 679} 680 681/* 682 * vm_map_unlock: release an exclusive lock on a map. 683 */ 684 685void 686vm_map_unlock(struct vm_map *map) 687{ 688 689 KASSERT(rw_write_held(&map->lock)); 690 KASSERT(map->busy == NULL || map->busy == curlwp); 691 rw_exit(&map->lock); 692} 693 694/* 695 * vm_map_unbusy: mark the map as unbusy, and wake any waiters that 696 * want an exclusive lock. 697 */ 698 699void 700vm_map_unbusy(struct vm_map *map) 701{ 702 703 KASSERT(map->busy == curlwp); 704 705 /* 706 * Safe to clear 'busy' and 'waiters' with only a read lock held: 707 * 708 * o they can only be set with a write lock held 709 * o writers are blocked out with a read or write hold 710 * o at any time, only one thread owns the set of values 711 */ 712 mutex_enter(&map->misc_lock); 713 map->busy = NULL; 714 cv_broadcast(&map->cv); 715 mutex_exit(&map->misc_lock); 716} 717 718/* 719 * vm_map_lock_read: acquire a shared (read) lock on a map. 720 */ 721 722void 723vm_map_lock_read(struct vm_map *map) 724{ 725 726 rw_enter(&map->lock, RW_READER); 727} 728 729/* 730 * vm_map_unlock_read: release a shared lock on a map. 731 */ 732 733void 734vm_map_unlock_read(struct vm_map *map) 735{ 736 737 rw_exit(&map->lock); 738} 739 740/* 741 * vm_map_busy: mark a map as busy. 742 * 743 * => the caller must hold the map write locked 744 */ 745 746void 747vm_map_busy(struct vm_map *map) 748{ 749 750 KASSERT(rw_write_held(&map->lock)); 751 KASSERT(map->busy == NULL); 752 753 map->busy = curlwp; 754} 755 756/* 757 * vm_map_locked_p: return true if the map is write locked. 758 * 759 * => only for debug purposes like KASSERTs. 760 * => should not be used to verify that a map is not locked. 761 */ 762 763bool 764vm_map_locked_p(struct vm_map *map) 765{ 766 767 return rw_write_held(&map->lock); 768} 769 770/* 771 * uvm_mapent_alloc: allocate a map entry 772 */ 773 774static struct vm_map_entry * 775uvm_mapent_alloc(struct vm_map *map, int flags) 776{ 777 struct vm_map_entry *me; 778 int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK; 779 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 780 781 me = pool_cache_get(&uvm_map_entry_cache, pflags); 782 if (__predict_false(me == NULL)) { 783 return NULL; 784 } 785 me->flags = 0; 786 787 UVMHIST_LOG(maphist, "<- new entry=%#jx [kentry=%jd]", (uintptr_t)me, 788 (map == kernel_map), 0, 0); 789 return me; 790} 791 792/* 793 * uvm_mapent_free: free map entry 794 */ 795 796static void 797uvm_mapent_free(struct vm_map_entry *me) 798{ 799 UVMHIST_FUNC(__func__); 800 UVMHIST_CALLARGS(maphist,"<- freeing map entry=%#jx [flags=%#jx]", 801 (uintptr_t)me, me->flags, 0, 0); 802 pool_cache_put(&uvm_map_entry_cache, me); 803} 804 805/* 806 * uvm_mapent_copy: copy a map entry, preserving flags 807 */ 808 809static inline void 810uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 811{ 812 813 memcpy(dst, src, sizeof(*dst)); 814 dst->flags = 0; 815} 816 817#if defined(DEBUG) 818static void 819_uvm_mapent_check(const struct vm_map_entry *entry, int line) 820{ 821 822 if (entry->start >= entry->end) { 823 goto bad; 824 } 825 if (UVM_ET_ISOBJ(entry)) { 826 if (entry->object.uvm_obj == NULL) { 827 goto bad; 828 } 829 } else if (UVM_ET_ISSUBMAP(entry)) { 830 if (entry->object.sub_map == NULL) { 831 goto bad; 832 } 833 } else { 834 if (entry->object.uvm_obj != NULL || 835 entry->object.sub_map != NULL) { 836 goto bad; 837 } 838 } 839 if (!UVM_ET_ISOBJ(entry)) { 840 if (entry->offset != 0) { 841 goto bad; 842 } 843 } 844 845 return; 846 847bad: 848 panic("%s: bad entry %p, line %d", __func__, entry, line); 849} 850#endif /* defined(DEBUG) */ 851 852/* 853 * uvm_map_entry_unwire: unwire a map entry 854 * 855 * => map should be locked by caller 856 */ 857 858static inline void 859uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry) 860{ 861 862 entry->wired_count = 0; 863 uvm_fault_unwire_locked(map, entry->start, entry->end); 864} 865 866 867/* 868 * wrapper for calling amap_ref() 869 */ 870static inline void 871uvm_map_reference_amap(struct vm_map_entry *entry, int flags) 872{ 873 874 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 875 (entry->end - entry->start) >> PAGE_SHIFT, flags); 876} 877 878 879/* 880 * wrapper for calling amap_unref() 881 */ 882static inline void 883uvm_map_unreference_amap(struct vm_map_entry *entry, int flags) 884{ 885 886 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 887 (entry->end - entry->start) >> PAGE_SHIFT, flags); 888} 889 890 891/* 892 * uvm_map_init: init mapping system at boot time. 893 */ 894 895void 896uvm_map_init(void) 897{ 898 /* 899 * first, init logging system. 900 */ 901 902 UVMHIST_FUNC(__func__); 903 UVMHIST_LINK_STATIC(maphist); 904 UVMHIST_LINK_STATIC(pdhist); 905 UVMHIST_CALLED(maphist); 906 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 907 908 /* 909 * initialize the global lock for kernel map entry. 910 */ 911 912 mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM); 913} 914 915/* 916 * uvm_map_init_caches: init mapping system caches. 917 */ 918void 919uvm_map_init_caches(void) 920{ 921 /* 922 * initialize caches. 923 */ 924 925 pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry), 926 coherency_unit, 0, PR_LARGECACHE, "vmmpepl", NULL, IPL_NONE, NULL, 927 NULL, NULL); 928} 929 930/* 931 * clippers 932 */ 933 934/* 935 * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy. 936 */ 937 938static void 939uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2, 940 vaddr_t splitat) 941{ 942 vaddr_t adj; 943 944 KASSERT(entry1->start < splitat); 945 KASSERT(splitat < entry1->end); 946 947 adj = splitat - entry1->start; 948 entry1->end = entry2->start = splitat; 949 950 if (entry1->aref.ar_amap) { 951 amap_splitref(&entry1->aref, &entry2->aref, adj); 952 } 953 if (UVM_ET_ISSUBMAP(entry1)) { 954 /* ... unlikely to happen, but play it safe */ 955 uvm_map_reference(entry1->object.sub_map); 956 } else if (UVM_ET_ISOBJ(entry1)) { 957 KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */ 958 entry2->offset += adj; 959 if (entry1->object.uvm_obj->pgops && 960 entry1->object.uvm_obj->pgops->pgo_reference) 961 entry1->object.uvm_obj->pgops->pgo_reference( 962 entry1->object.uvm_obj); 963 } 964} 965 966/* 967 * uvm_map_clip_start: ensure that the entry begins at or after 968 * the starting address, if it doesn't we split the entry. 969 * 970 * => caller should use UVM_MAP_CLIP_START macro rather than calling 971 * this directly 972 * => map must be locked by caller 973 */ 974 975void 976uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, 977 vaddr_t start) 978{ 979 struct vm_map_entry *new_entry; 980 981 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 982 983 uvm_map_check(map, "clip_start entry"); 984 uvm_mapent_check(entry); 985 986 /* 987 * Split off the front portion. note that we must insert the new 988 * entry BEFORE this one, so that this entry has the specified 989 * starting address. 990 */ 991 new_entry = uvm_mapent_alloc(map, 0); 992 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 993 uvm_mapent_splitadj(new_entry, entry, start); 994 uvm_map_entry_link(map, entry->prev, new_entry); 995 996 uvm_map_check(map, "clip_start leave"); 997} 998 999/* 1000 * uvm_map_clip_end: ensure that the entry ends at or before 1001 * the ending address, if it does't we split the reference 1002 * 1003 * => caller should use UVM_MAP_CLIP_END macro rather than calling 1004 * this directly 1005 * => map must be locked by caller 1006 */ 1007 1008void 1009uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end) 1010{ 1011 struct vm_map_entry *new_entry; 1012 1013 uvm_map_check(map, "clip_end entry"); 1014 uvm_mapent_check(entry); 1015 1016 /* 1017 * Create a new entry and insert it 1018 * AFTER the specified entry 1019 */ 1020 new_entry = uvm_mapent_alloc(map, 0); 1021 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 1022 uvm_mapent_splitadj(entry, new_entry, end); 1023 uvm_map_entry_link(map, entry, new_entry); 1024 1025 uvm_map_check(map, "clip_end leave"); 1026} 1027 1028/* 1029 * M A P - m a i n e n t r y p o i n t 1030 */ 1031/* 1032 * uvm_map: establish a valid mapping in a map 1033 * 1034 * => assume startp is page aligned. 1035 * => assume size is a multiple of PAGE_SIZE. 1036 * => assume sys_mmap provides enough of a "hint" to have us skip 1037 * over text/data/bss area. 1038 * => map must be unlocked (we will lock it) 1039 * => <uobj,uoffset> value meanings (4 cases): 1040 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 1041 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 1042 * [3] <uobj,uoffset> == normal mapping 1043 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 1044 * 1045 * case [4] is for kernel mappings where we don't know the offset until 1046 * we've found a virtual address. note that kernel object offsets are 1047 * always relative to vm_map_min(kernel_map). 1048 * 1049 * => if `align' is non-zero, we align the virtual address to the specified 1050 * alignment. 1051 * this is provided as a mechanism for large pages. 1052 * 1053 * => XXXCDC: need way to map in external amap? 1054 */ 1055 1056int 1057uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size, 1058 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags) 1059{ 1060 struct uvm_map_args args; 1061 struct vm_map_entry *new_entry; 1062 int error; 1063 1064 KASSERT((size & PAGE_MASK) == 0); 1065 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1066 1067 /* 1068 * for pager_map, allocate the new entry first to avoid sleeping 1069 * for memory while we have the map locked. 1070 */ 1071 1072 new_entry = NULL; 1073 if (map == pager_map) { 1074 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT)); 1075 if (__predict_false(new_entry == NULL)) 1076 return ENOMEM; 1077 } 1078 if (map == pager_map) 1079 flags |= UVM_FLAG_NOMERGE; 1080 1081 error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align, 1082 flags, &args); 1083 if (!error) { 1084 error = uvm_map_enter(map, &args, new_entry); 1085 *startp = args.uma_start; 1086 } else if (new_entry) { 1087 uvm_mapent_free(new_entry); 1088 } 1089 1090#if defined(DEBUG) 1091 if (!error && VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) { 1092 uvm_km_check_empty(map, *startp, *startp + size); 1093 } 1094#endif /* defined(DEBUG) */ 1095 1096 return error; 1097} 1098 1099/* 1100 * uvm_map_prepare: 1101 * 1102 * called with map unlocked. 1103 * on success, returns the map locked. 1104 */ 1105 1106int 1107uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size, 1108 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags, 1109 struct uvm_map_args *args) 1110{ 1111 struct vm_map_entry *prev_entry; 1112 vm_prot_t prot = UVM_PROTECTION(flags); 1113 vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1114 1115 UVMHIST_FUNC(__func__); 1116 UVMHIST_CALLARGS(maphist, "(map=%#jx, start=%#jx, size=%jx, flags=%#jx)", 1117 (uintptr_t)map, start, size, flags); 1118 UVMHIST_LOG(maphist, " uobj/offset %#jx/%jd", (uintptr_t)uobj, 1119 uoffset,0,0); 1120 1121 /* 1122 * detect a popular device driver bug. 1123 */ 1124 1125 KASSERT(doing_shutdown || curlwp != NULL); 1126 1127 /* 1128 * zero-sized mapping doesn't make any sense. 1129 */ 1130 KASSERT(size > 0); 1131 1132 KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0); 1133 1134 uvm_map_check(map, "map entry"); 1135 1136 /* 1137 * check sanity of protection code 1138 */ 1139 1140 if ((prot & maxprot) != prot) { 1141 UVMHIST_LOG(maphist, "<- prot. failure: prot=%#jx, max=%#jx", 1142 prot, maxprot,0,0); 1143 return EACCES; 1144 } 1145 1146 /* 1147 * figure out where to put new VM range 1148 */ 1149retry: 1150 if (vm_map_lock_try(map) == false) { 1151 if ((flags & UVM_FLAG_TRYLOCK) != 0) { 1152 return EAGAIN; 1153 } 1154 vm_map_lock(map); /* could sleep here */ 1155 } 1156 if (flags & UVM_FLAG_UNMAP) { 1157 KASSERT(flags & UVM_FLAG_FIXED); 1158 KASSERT((flags & UVM_FLAG_NOWAIT) == 0); 1159 1160 /* 1161 * Set prev_entry to what it will need to be after any existing 1162 * entries are removed later in uvm_map_enter(). 1163 */ 1164 1165 if (uvm_map_lookup_entry(map, start, &prev_entry)) { 1166 if (start == prev_entry->start) 1167 prev_entry = prev_entry->prev; 1168 else 1169 UVM_MAP_CLIP_END(map, prev_entry, start); 1170 SAVE_HINT(map, map->hint, prev_entry); 1171 } 1172 } else { 1173 prev_entry = uvm_map_findspace(map, start, size, &start, 1174 uobj, uoffset, align, flags); 1175 } 1176 if (prev_entry == NULL) { 1177 unsigned int timestamp; 1178 1179 timestamp = map->timestamp; 1180 UVMHIST_LOG(maphist,"waiting va timestamp=%#jx", 1181 timestamp,0,0,0); 1182 map->flags |= VM_MAP_WANTVA; 1183 vm_map_unlock(map); 1184 1185 /* 1186 * try to reclaim kva and wait until someone does unmap. 1187 * fragile locking here, so we awaken every second to 1188 * recheck the condition. 1189 */ 1190 1191 mutex_enter(&map->misc_lock); 1192 while ((map->flags & VM_MAP_WANTVA) != 0 && 1193 map->timestamp == timestamp) { 1194 if ((flags & UVM_FLAG_WAITVA) == 0) { 1195 mutex_exit(&map->misc_lock); 1196 UVMHIST_LOG(maphist, 1197 "<- uvm_map_findspace failed!", 0,0,0,0); 1198 return ENOMEM; 1199 } else { 1200 cv_timedwait(&map->cv, &map->misc_lock, hz); 1201 } 1202 } 1203 mutex_exit(&map->misc_lock); 1204 goto retry; 1205 } 1206 1207#ifdef PMAP_GROWKERNEL 1208 /* 1209 * If the kernel pmap can't map the requested space, 1210 * then allocate more resources for it. 1211 */ 1212 if (map == kernel_map && uvm_maxkaddr < (start + size)) 1213 uvm_maxkaddr = pmap_growkernel(start + size); 1214#endif 1215 1216 UVMMAP_EVCNT_INCR(map_call); 1217 1218 /* 1219 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 1220 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 1221 * either case we want to zero it before storing it in the map entry 1222 * (because it looks strange and confusing when debugging...) 1223 * 1224 * if uobj is not null 1225 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 1226 * and we do not need to change uoffset. 1227 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 1228 * now (based on the starting address of the map). this case is 1229 * for kernel object mappings where we don't know the offset until 1230 * the virtual address is found (with uvm_map_findspace). the 1231 * offset is the distance we are from the start of the map. 1232 */ 1233 1234 if (uobj == NULL) { 1235 uoffset = 0; 1236 } else { 1237 if (uoffset == UVM_UNKNOWN_OFFSET) { 1238 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 1239 uoffset = start - vm_map_min(kernel_map); 1240 } 1241 } 1242 1243 args->uma_flags = flags; 1244 args->uma_prev = prev_entry; 1245 args->uma_start = start; 1246 args->uma_size = size; 1247 args->uma_uobj = uobj; 1248 args->uma_uoffset = uoffset; 1249 1250 UVMHIST_LOG(maphist, "<- done!", 0,0,0,0); 1251 return 0; 1252} 1253 1254/* 1255 * uvm_map_enter: 1256 * 1257 * called with map locked. 1258 * unlock the map before returning. 1259 */ 1260 1261int 1262uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args, 1263 struct vm_map_entry *new_entry) 1264{ 1265 struct vm_map_entry *prev_entry = args->uma_prev; 1266 struct vm_map_entry *dead = NULL, *dead_entries = NULL; 1267 1268 const uvm_flag_t flags = args->uma_flags; 1269 const vm_prot_t prot = UVM_PROTECTION(flags); 1270 const vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1271 const vm_inherit_t inherit = UVM_INHERIT(flags); 1272 const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ? 1273 AMAP_EXTEND_NOWAIT : 0; 1274 const int advice = UVM_ADVICE(flags); 1275 1276 vaddr_t start = args->uma_start; 1277 vsize_t size = args->uma_size; 1278 struct uvm_object *uobj = args->uma_uobj; 1279 voff_t uoffset = args->uma_uoffset; 1280 1281 const int kmap = (vm_map_pmap(map) == pmap_kernel()); 1282 int merged = 0; 1283 int error; 1284 int newetype; 1285 1286 UVMHIST_FUNC(__func__); 1287 UVMHIST_CALLARGS(maphist, "(map=%#jx, start=%#jx, size=%ju, flags=%#jx)", 1288 (uintptr_t)map, start, size, flags); 1289 UVMHIST_LOG(maphist, " uobj/offset %#jx/%jd", (uintptr_t)uobj, 1290 uoffset,0,0); 1291 1292 KASSERT(map->hint == prev_entry); /* bimerge case assumes this */ 1293 KASSERT(vm_map_locked_p(map)); 1294 KASSERT((flags & (UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP)) != 1295 (UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP)); 1296 1297 if (uobj) 1298 newetype = UVM_ET_OBJ; 1299 else 1300 newetype = 0; 1301 1302 if (flags & UVM_FLAG_COPYONW) { 1303 newetype |= UVM_ET_COPYONWRITE; 1304 if ((flags & UVM_FLAG_OVERLAY) == 0) 1305 newetype |= UVM_ET_NEEDSCOPY; 1306 } 1307 1308 /* 1309 * For mappings with unmap, remove any old entries now. Adding the new 1310 * entry cannot fail because that can only happen if UVM_FLAG_NOWAIT 1311 * is set, and we do not support nowait and unmap together. 1312 */ 1313 1314 if (flags & UVM_FLAG_UNMAP) { 1315 KASSERT(flags & UVM_FLAG_FIXED); 1316 uvm_unmap_remove(map, start, start + size, &dead_entries, 0); 1317#ifdef DEBUG 1318 struct vm_map_entry *tmp_entry __diagused; 1319 bool rv __diagused; 1320 1321 rv = uvm_map_lookup_entry(map, start, &tmp_entry); 1322 KASSERT(!rv); 1323 KASSERTMSG(prev_entry == tmp_entry, 1324 "args %p prev_entry %p tmp_entry %p", 1325 args, prev_entry, tmp_entry); 1326#endif 1327 SAVE_HINT(map, map->hint, prev_entry); 1328 } 1329 1330 /* 1331 * try and insert in map by extending previous entry, if possible. 1332 * XXX: we don't try and pull back the next entry. might be useful 1333 * for a stack, but we are currently allocating our stack in advance. 1334 */ 1335 1336 if (flags & UVM_FLAG_NOMERGE) 1337 goto nomerge; 1338 1339 if (prev_entry->end == start && 1340 prev_entry != &map->header && 1341 UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, 0, 1342 prot, maxprot, inherit, advice, 0)) { 1343 1344 if (uobj && prev_entry->offset + 1345 (prev_entry->end - prev_entry->start) != uoffset) 1346 goto forwardmerge; 1347 1348 /* 1349 * can't extend a shared amap. note: no need to lock amap to 1350 * look at refs since we don't care about its exact value. 1351 * if it is one (i.e. we have only reference) it will stay there 1352 */ 1353 1354 if (prev_entry->aref.ar_amap && 1355 amap_refs(prev_entry->aref.ar_amap) != 1) { 1356 goto forwardmerge; 1357 } 1358 1359 if (prev_entry->aref.ar_amap) { 1360 error = amap_extend(prev_entry, size, 1361 amapwaitflag | AMAP_EXTEND_FORWARDS); 1362 if (error) 1363 goto nomerge; 1364 } 1365 1366 if (kmap) { 1367 UVMMAP_EVCNT_INCR(kbackmerge); 1368 } else { 1369 UVMMAP_EVCNT_INCR(ubackmerge); 1370 } 1371 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 1372 1373 /* 1374 * drop our reference to uobj since we are extending a reference 1375 * that we already have (the ref count can not drop to zero). 1376 */ 1377 1378 if (uobj && uobj->pgops->pgo_detach) 1379 uobj->pgops->pgo_detach(uobj); 1380 1381 /* 1382 * Now that we've merged the entries, note that we've grown 1383 * and our gap has shrunk. Then fix the tree. 1384 */ 1385 prev_entry->end += size; 1386 prev_entry->gap -= size; 1387 uvm_rb_fixup(map, prev_entry); 1388 1389 uvm_map_check(map, "map backmerged"); 1390 1391 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 1392 merged++; 1393 } 1394 1395forwardmerge: 1396 if (prev_entry->next->start == (start + size) && 1397 prev_entry->next != &map->header && 1398 UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, 0, 1399 prot, maxprot, inherit, advice, 0)) { 1400 1401 if (uobj && prev_entry->next->offset != uoffset + size) 1402 goto nomerge; 1403 1404 /* 1405 * can't extend a shared amap. note: no need to lock amap to 1406 * look at refs since we don't care about its exact value. 1407 * if it is one (i.e. we have only reference) it will stay there. 1408 * 1409 * note that we also can't merge two amaps, so if we 1410 * merged with the previous entry which has an amap, 1411 * and the next entry also has an amap, we give up. 1412 * 1413 * Interesting cases: 1414 * amap, new, amap -> give up second merge (single fwd extend) 1415 * amap, new, none -> double forward extend (extend again here) 1416 * none, new, amap -> double backward extend (done here) 1417 * uobj, new, amap -> single backward extend (done here) 1418 * 1419 * XXX should we attempt to deal with someone refilling 1420 * the deallocated region between two entries that are 1421 * backed by the same amap (ie, arefs is 2, "prev" and 1422 * "next" refer to it, and adding this allocation will 1423 * close the hole, thus restoring arefs to 1 and 1424 * deallocating the "next" vm_map_entry)? -- @@@ 1425 */ 1426 1427 if (prev_entry->next->aref.ar_amap && 1428 (amap_refs(prev_entry->next->aref.ar_amap) != 1 || 1429 (merged && prev_entry->aref.ar_amap))) { 1430 goto nomerge; 1431 } 1432 1433 if (merged) { 1434 /* 1435 * Try to extend the amap of the previous entry to 1436 * cover the next entry as well. If it doesn't work 1437 * just skip on, don't actually give up, since we've 1438 * already completed the back merge. 1439 */ 1440 if (prev_entry->aref.ar_amap) { 1441 if (amap_extend(prev_entry, 1442 prev_entry->next->end - 1443 prev_entry->next->start, 1444 amapwaitflag | AMAP_EXTEND_FORWARDS)) 1445 goto nomerge; 1446 } 1447 1448 /* 1449 * Try to extend the amap of the *next* entry 1450 * back to cover the new allocation *and* the 1451 * previous entry as well (the previous merge 1452 * didn't have an amap already otherwise we 1453 * wouldn't be checking here for an amap). If 1454 * it doesn't work just skip on, again, don't 1455 * actually give up, since we've already 1456 * completed the back merge. 1457 */ 1458 else if (prev_entry->next->aref.ar_amap) { 1459 if (amap_extend(prev_entry->next, 1460 prev_entry->end - 1461 prev_entry->start, 1462 amapwaitflag | AMAP_EXTEND_BACKWARDS)) 1463 goto nomerge; 1464 } 1465 } else { 1466 /* 1467 * Pull the next entry's amap backwards to cover this 1468 * new allocation. 1469 */ 1470 if (prev_entry->next->aref.ar_amap) { 1471 error = amap_extend(prev_entry->next, size, 1472 amapwaitflag | AMAP_EXTEND_BACKWARDS); 1473 if (error) 1474 goto nomerge; 1475 } 1476 } 1477 1478 if (merged) { 1479 if (kmap) { 1480 UVMMAP_EVCNT_DECR(kbackmerge); 1481 UVMMAP_EVCNT_INCR(kbimerge); 1482 } else { 1483 UVMMAP_EVCNT_DECR(ubackmerge); 1484 UVMMAP_EVCNT_INCR(ubimerge); 1485 } 1486 } else { 1487 if (kmap) { 1488 UVMMAP_EVCNT_INCR(kforwmerge); 1489 } else { 1490 UVMMAP_EVCNT_INCR(uforwmerge); 1491 } 1492 } 1493 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0); 1494 1495 /* 1496 * drop our reference to uobj since we are extending a reference 1497 * that we already have (the ref count can not drop to zero). 1498 */ 1499 if (uobj && uobj->pgops->pgo_detach) 1500 uobj->pgops->pgo_detach(uobj); 1501 1502 if (merged) { 1503 dead = prev_entry->next; 1504 prev_entry->end = dead->end; 1505 uvm_map_entry_unlink(map, dead); 1506 if (dead->aref.ar_amap != NULL) { 1507 prev_entry->aref = dead->aref; 1508 dead->aref.ar_amap = NULL; 1509 } 1510 } else { 1511 prev_entry->next->start -= size; 1512 if (prev_entry != &map->header) { 1513 prev_entry->gap -= size; 1514 KASSERT(prev_entry->gap == uvm_rb_gap(prev_entry)); 1515 uvm_rb_fixup(map, prev_entry); 1516 } 1517 if (uobj) 1518 prev_entry->next->offset = uoffset; 1519 } 1520 1521 uvm_map_check(map, "map forwardmerged"); 1522 1523 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0); 1524 merged++; 1525 } 1526 1527nomerge: 1528 if (!merged) { 1529 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 1530 if (kmap) { 1531 UVMMAP_EVCNT_INCR(knomerge); 1532 } else { 1533 UVMMAP_EVCNT_INCR(unomerge); 1534 } 1535 1536 /* 1537 * allocate new entry and link it in. 1538 */ 1539 1540 if (new_entry == NULL) { 1541 new_entry = uvm_mapent_alloc(map, 1542 (flags & UVM_FLAG_NOWAIT)); 1543 if (__predict_false(new_entry == NULL)) { 1544 error = ENOMEM; 1545 goto done; 1546 } 1547 } 1548 new_entry->start = start; 1549 new_entry->end = new_entry->start + size; 1550 new_entry->object.uvm_obj = uobj; 1551 new_entry->offset = uoffset; 1552 1553 new_entry->etype = newetype; 1554 1555 if (flags & UVM_FLAG_NOMERGE) { 1556 new_entry->flags |= UVM_MAP_NOMERGE; 1557 } 1558 1559 new_entry->protection = prot; 1560 new_entry->max_protection = maxprot; 1561 new_entry->inheritance = inherit; 1562 new_entry->wired_count = 0; 1563 new_entry->advice = advice; 1564 if (flags & UVM_FLAG_OVERLAY) { 1565 1566 /* 1567 * to_add: for BSS we overallocate a little since we 1568 * are likely to extend 1569 */ 1570 1571 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 1572 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 1573 struct vm_amap *amap = amap_alloc(size, to_add, 1574 (flags & UVM_FLAG_NOWAIT)); 1575 if (__predict_false(amap == NULL)) { 1576 error = ENOMEM; 1577 goto done; 1578 } 1579 new_entry->aref.ar_pageoff = 0; 1580 new_entry->aref.ar_amap = amap; 1581 } else { 1582 new_entry->aref.ar_pageoff = 0; 1583 new_entry->aref.ar_amap = NULL; 1584 } 1585 uvm_map_entry_link(map, prev_entry, new_entry); 1586 1587 /* 1588 * Update the free space hint 1589 */ 1590 1591 if ((map->first_free == prev_entry) && 1592 (prev_entry->end >= new_entry->start)) 1593 map->first_free = new_entry; 1594 1595 new_entry = NULL; 1596 } 1597 1598 map->size += size; 1599 1600 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1601 1602 error = 0; 1603 1604done: 1605 vm_map_unlock(map); 1606 1607 if (new_entry) { 1608 uvm_mapent_free(new_entry); 1609 } 1610 if (dead) { 1611 KDASSERT(merged); 1612 uvm_mapent_free(dead); 1613 } 1614 if (dead_entries) 1615 uvm_unmap_detach(dead_entries, 0); 1616 1617 return error; 1618} 1619 1620/* 1621 * uvm_map_lookup_entry_bytree: lookup an entry in tree 1622 */ 1623 1624static inline bool 1625uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address, 1626 struct vm_map_entry **entry /* OUT */) 1627{ 1628 struct vm_map_entry *prev = &map->header; 1629 struct vm_map_entry *cur = ROOT_ENTRY(map); 1630 1631 while (cur) { 1632 UVMMAP_EVCNT_INCR(mlk_treeloop); 1633 if (address >= cur->start) { 1634 if (address < cur->end) { 1635 *entry = cur; 1636 return true; 1637 } 1638 prev = cur; 1639 cur = RIGHT_ENTRY(cur); 1640 } else 1641 cur = LEFT_ENTRY(cur); 1642 } 1643 *entry = prev; 1644 return false; 1645} 1646 1647/* 1648 * uvm_map_lookup_entry: find map entry at or before an address 1649 * 1650 * => map must at least be read-locked by caller 1651 * => entry is returned in "entry" 1652 * => return value is true if address is in the returned entry 1653 */ 1654 1655bool 1656uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1657 struct vm_map_entry **entry /* OUT */) 1658{ 1659 struct vm_map_entry *cur; 1660 UVMHIST_FUNC(__func__); 1661 UVMHIST_CALLARGS(maphist,"(map=%#jx,addr=%#jx,ent=%#jx)", 1662 (uintptr_t)map, address, (uintptr_t)entry, 0); 1663 1664 /* 1665 * make a quick check to see if we are already looking at 1666 * the entry we want (which is usually the case). note also 1667 * that we don't need to save the hint here... it is the 1668 * same hint (unless we are at the header, in which case the 1669 * hint didn't buy us anything anyway). 1670 */ 1671 1672 cur = map->hint; 1673 UVMMAP_EVCNT_INCR(mlk_call); 1674 if (cur != &map->header && 1675 address >= cur->start && cur->end > address) { 1676 UVMMAP_EVCNT_INCR(mlk_hint); 1677 *entry = cur; 1678 UVMHIST_LOG(maphist,"<- got it via hint (%#jx)", 1679 (uintptr_t)cur, 0, 0, 0); 1680 uvm_mapent_check(*entry); 1681 return (true); 1682 } 1683 uvm_map_check(map, __func__); 1684 1685 /* 1686 * lookup in the tree. 1687 */ 1688 1689 UVMMAP_EVCNT_INCR(mlk_tree); 1690 if (__predict_true(uvm_map_lookup_entry_bytree(map, address, entry))) { 1691 SAVE_HINT(map, map->hint, *entry); 1692 UVMHIST_LOG(maphist,"<- search got it (%#jx)", 1693 (uintptr_t)cur, 0, 0, 0); 1694 KDASSERT((*entry)->start <= address); 1695 KDASSERT(address < (*entry)->end); 1696 uvm_mapent_check(*entry); 1697 return (true); 1698 } 1699 1700 SAVE_HINT(map, map->hint, *entry); 1701 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1702 KDASSERT((*entry) == &map->header || (*entry)->end <= address); 1703 KDASSERT((*entry)->next == &map->header || 1704 address < (*entry)->next->start); 1705 return (false); 1706} 1707 1708/* 1709 * See if the range between start and start + length fits in the gap 1710 * entry->next->start and entry->end. Returns 1 if fits, 0 if doesn't 1711 * fit, and -1 address wraps around. 1712 */ 1713static int 1714uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset, 1715 vsize_t align, int flags, int topdown, struct vm_map_entry *entry) 1716{ 1717 vaddr_t end; 1718 1719#ifdef PMAP_PREFER 1720 /* 1721 * push start address forward as needed to avoid VAC alias problems. 1722 * we only do this if a valid offset is specified. 1723 */ 1724 1725 if (uoffset != UVM_UNKNOWN_OFFSET) 1726 PMAP_PREFER(uoffset, start, length, topdown); 1727#endif 1728 if ((flags & UVM_FLAG_COLORMATCH) != 0) { 1729 KASSERT(align < uvmexp.ncolors); 1730 if (uvmexp.ncolors > 1) { 1731 const u_int colormask = uvmexp.colormask; 1732 const u_int colorsize = colormask + 1; 1733 vaddr_t hint = atop(*start); 1734 const u_int color = hint & colormask; 1735 if (color != align) { 1736 hint -= color; /* adjust to color boundary */ 1737 KASSERT((hint & colormask) == 0); 1738 if (topdown) { 1739 if (align > color) 1740 hint -= colorsize; 1741 } else { 1742 if (align < color) 1743 hint += colorsize; 1744 } 1745 *start = ptoa(hint + align); /* adjust to color */ 1746 } 1747 } 1748 } else { 1749 KASSERT(powerof2(align)); 1750 uvm_map_align_va(start, align, topdown); 1751 /* 1752 * XXX Should we PMAP_PREFER() here again? 1753 * eh...i think we're okay 1754 */ 1755 } 1756 1757 /* 1758 * Find the end of the proposed new region. Be sure we didn't 1759 * wrap around the address; if so, we lose. Otherwise, if the 1760 * proposed new region fits before the next entry, we win. 1761 */ 1762 1763 end = *start + length; 1764 if (end < *start) 1765 return (-1); 1766 1767 if (entry->next->start >= end && *start >= entry->end) 1768 return (1); 1769 1770 return (0); 1771} 1772 1773static void 1774uvm_findspace_invariants(struct vm_map *map, vaddr_t orig_hint, vaddr_t length, 1775 struct uvm_object *uobj, voff_t uoffset, vsize_t align, int flags, 1776 vaddr_t hint, struct vm_map_entry *entry, int line) 1777{ 1778 const int topdown = map->flags & VM_MAP_TOPDOWN; 1779 1780 KASSERTMSG( topdown || hint >= orig_hint, 1781 "map=%p hint=%#"PRIxVADDR" orig_hint=%#"PRIxVADDR 1782 " length=%#"PRIxVSIZE" uobj=%p uoffset=%#llx align=%"PRIxVSIZE 1783 " flags=%#x entry=%p (uvm_map_findspace line %d)", 1784 map, hint, orig_hint, 1785 length, uobj, (unsigned long long)uoffset, align, 1786 flags, entry, line); 1787#ifndef __sh3__ /* XXXRO: kern/51254 */ 1788 KASSERTMSG(!topdown || hint <= orig_hint, 1789#else 1790 if (__predict_false(!(!topdown || hint <= orig_hint))) 1791 printf( 1792#endif 1793 "map=%p hint=%#"PRIxVADDR" orig_hint=%#"PRIxVADDR 1794 " length=%#"PRIxVSIZE" uobj=%p uoffset=%#llx align=%"PRIxVSIZE 1795 " flags=%#x entry=%p (uvm_map_findspace line %d)", 1796 map, hint, orig_hint, 1797 length, uobj, (unsigned long long)uoffset, align, 1798 flags, entry, line); 1799} 1800 1801/* 1802 * uvm_map_findspace: find "length" sized space in "map". 1803 * 1804 * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is 1805 * set in "flags" (in which case we insist on using "hint"). 1806 * => "result" is VA returned 1807 * => uobj/uoffset are to be used to handle VAC alignment, if required 1808 * => if "align" is non-zero, we attempt to align to that value. 1809 * => caller must at least have read-locked map 1810 * => returns NULL on failure, or pointer to prev. map entry if success 1811 * => note this is a cross between the old vm_map_findspace and vm_map_find 1812 */ 1813 1814struct vm_map_entry * 1815uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length, 1816 vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset, 1817 vsize_t align, int flags) 1818{ 1819#define INVARIANTS() \ 1820 uvm_findspace_invariants(map, orig_hint, length, uobj, uoffset, align,\ 1821 flags, hint, entry, __LINE__) 1822 struct vm_map_entry *entry = NULL; 1823 struct vm_map_entry *child, *prev, *tmp; 1824 vaddr_t orig_hint __diagused; 1825 const int topdown = map->flags & VM_MAP_TOPDOWN; 1826 int avail; 1827 UVMHIST_FUNC(__func__); 1828 UVMHIST_CALLARGS(maphist, "(map=%#jx, hint=%#jx, len=%ju, flags=%#jx...", 1829 (uintptr_t)map, hint, length, flags); 1830 UVMHIST_LOG(maphist, " uobj=%#jx, uoffset=%#jx, align=%#jx)", 1831 (uintptr_t)uobj, uoffset, align, 0); 1832 1833 KASSERT((flags & UVM_FLAG_COLORMATCH) != 0 || powerof2(align)); 1834 KASSERT((flags & UVM_FLAG_COLORMATCH) == 0 || align < uvmexp.ncolors); 1835 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1836 1837 uvm_map_check(map, "map_findspace entry"); 1838 1839 /* 1840 * Clamp the hint to the VM map's min/max address, and remmeber 1841 * the clamped original hint. Remember the original hint, 1842 * clamped to the min/max address. If we are aligning, then we 1843 * may have to try again with no alignment constraint if we 1844 * fail the first time. 1845 * 1846 * We use the original hint to verify later that the search has 1847 * been monotonic -- that is, nonincreasing or nondecreasing, 1848 * according to topdown or !topdown respectively. But the 1849 * clamping is not monotonic. 1850 */ 1851 if (hint < vm_map_min(map)) { /* check ranges ... */ 1852 if (flags & UVM_FLAG_FIXED) { 1853 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1854 return (NULL); 1855 } 1856 hint = vm_map_min(map); 1857 } 1858 if (hint > vm_map_max(map)) { 1859 UVMHIST_LOG(maphist,"<- VA %#jx > range [%#jx->%#jx]", 1860 hint, vm_map_min(map), vm_map_max(map), 0); 1861 return (NULL); 1862 } 1863 orig_hint = hint; 1864 INVARIANTS(); 1865 1866 UVMHIST_LOG(maphist,"<- VA %#jx vs range [%#jx->%#jx]", 1867 hint, vm_map_min(map), vm_map_max(map), 0); 1868 1869 /* 1870 * hint may not be aligned properly; we need round up or down it 1871 * before proceeding further. 1872 */ 1873 if ((flags & UVM_FLAG_COLORMATCH) == 0) { 1874 uvm_map_align_va(&hint, align, topdown); 1875 INVARIANTS(); 1876 } 1877 1878 UVMHIST_LOG(maphist,"<- VA %#jx vs range [%#jx->%#jx]", 1879 hint, vm_map_min(map), vm_map_max(map), 0); 1880 /* 1881 * Look for the first possible address; if there's already 1882 * something at this address, we have to start after it. 1883 */ 1884 1885 /* 1886 * @@@: there are four, no, eight cases to consider. 1887 * 1888 * 0: found, fixed, bottom up -> fail 1889 * 1: found, fixed, top down -> fail 1890 * 2: found, not fixed, bottom up -> start after entry->end, 1891 * loop up 1892 * 3: found, not fixed, top down -> start before entry->start, 1893 * loop down 1894 * 4: not found, fixed, bottom up -> check entry->next->start, fail 1895 * 5: not found, fixed, top down -> check entry->next->start, fail 1896 * 6: not found, not fixed, bottom up -> check entry->next->start, 1897 * loop up 1898 * 7: not found, not fixed, top down -> check entry->next->start, 1899 * loop down 1900 * 1901 * as you can see, it reduces to roughly five cases, and that 1902 * adding top down mapping only adds one unique case (without 1903 * it, there would be four cases). 1904 */ 1905 1906 if ((flags & UVM_FLAG_FIXED) == 0 && 1907 hint == (topdown ? vm_map_max(map) : vm_map_min(map))) { 1908 /* 1909 * The uvm_map_findspace algorithm is monotonic -- for 1910 * topdown VM it starts with a high hint and returns a 1911 * lower free address; for !topdown VM it starts with a 1912 * low hint and returns a higher free address. As an 1913 * optimization, start with the first (highest for 1914 * topdown, lowest for !topdown) free address. 1915 * 1916 * XXX This `optimization' probably doesn't actually do 1917 * much in practice unless userland explicitly passes 1918 * the VM map's minimum or maximum address, which 1919 * varies from machine to machine (VM_MAX/MIN_ADDRESS, 1920 * e.g. 0x7fbfdfeff000 on amd64 but 0xfffffffff000 on 1921 * aarch64) and may vary according to other factors 1922 * like sysctl vm.user_va0_disable. In particular, if 1923 * the user specifies 0 as a hint to mmap, then mmap 1924 * will choose a default address which is usually _not_ 1925 * VM_MAX/MIN_ADDRESS but something else instead like 1926 * VM_MAX_ADDRESS - stack size - guard page overhead, 1927 * in which case this branch is never hit. 1928 * 1929 * In fact, this branch appears to have been broken for 1930 * two decades between when topdown was introduced in 1931 * ~2003 and when it was adapted to handle the topdown 1932 * case without violating the monotonicity assertion in 1933 * 2022. Maybe Someone^TM should either ditch the 1934 * optimization or find a better way to do it. 1935 */ 1936 entry = map->first_free; 1937 } else { 1938 if (uvm_map_lookup_entry(map, hint, &entry)) { 1939 /* "hint" address already in use ... */ 1940 if (flags & UVM_FLAG_FIXED) { 1941 UVMHIST_LOG(maphist, "<- fixed & VA in use", 1942 0, 0, 0, 0); 1943 return (NULL); 1944 } 1945 if (topdown) 1946 /* Start from lower gap. */ 1947 entry = entry->prev; 1948 } else if (flags & UVM_FLAG_FIXED) { 1949 if (entry->next->start >= hint + length && 1950 hint + length > hint) 1951 goto found; 1952 1953 /* "hint" address is gap but too small */ 1954 UVMHIST_LOG(maphist, "<- fixed mapping failed", 1955 0, 0, 0, 0); 1956 return (NULL); /* only one shot at it ... */ 1957 } else { 1958 /* 1959 * See if given hint fits in this gap. 1960 */ 1961 avail = uvm_map_space_avail(&hint, length, 1962 uoffset, align, flags, topdown, entry); 1963 INVARIANTS(); 1964 switch (avail) { 1965 case 1: 1966 goto found; 1967 case -1: 1968 goto wraparound; 1969 } 1970 1971 if (topdown) { 1972 /* 1973 * Still there is a chance to fit 1974 * if hint > entry->end. 1975 */ 1976 } else { 1977 /* Start from higher gap. */ 1978 entry = entry->next; 1979 if (entry == &map->header) 1980 goto notfound; 1981 goto nextgap; 1982 } 1983 } 1984 } 1985 1986 /* 1987 * Note that all UVM_FLAGS_FIXED case is already handled. 1988 */ 1989 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1990 1991 /* Try to find the space in the red-black tree */ 1992 1993 /* Check slot before any entry */ 1994 if (topdown) { 1995 KASSERTMSG(entry->next->start >= vm_map_min(map), 1996 "map=%p entry=%p entry->next=%p" 1997 " entry->next->start=0x%"PRIxVADDR" min=0x%"PRIxVADDR, 1998 map, entry, entry->next, 1999 entry->next->start, vm_map_min(map)); 2000 if (length > entry->next->start - vm_map_min(map)) 2001 hint = vm_map_min(map); /* XXX goto wraparound? */ 2002 else 2003 hint = entry->next->start - length; 2004 KASSERT(hint >= vm_map_min(map)); 2005 } else { 2006 hint = entry->end; 2007 } 2008 INVARIANTS(); 2009 avail = uvm_map_space_avail(&hint, length, uoffset, align, flags, 2010 topdown, entry); 2011 INVARIANTS(); 2012 switch (avail) { 2013 case 1: 2014 goto found; 2015 case -1: 2016 goto wraparound; 2017 } 2018 2019nextgap: 2020 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 2021 /* If there is not enough space in the whole tree, we fail */ 2022 tmp = ROOT_ENTRY(map); 2023 if (tmp == NULL || tmp->maxgap < length) 2024 goto notfound; 2025 2026 prev = NULL; /* previous candidate */ 2027 2028 /* Find an entry close to hint that has enough space */ 2029 for (; tmp;) { 2030 KASSERT(tmp->next->start == tmp->end + tmp->gap); 2031 if (topdown) { 2032 if (tmp->next->start < hint + length && 2033 (prev == NULL || tmp->end > prev->end)) { 2034 if (tmp->gap >= length) 2035 prev = tmp; 2036 else if ((child = LEFT_ENTRY(tmp)) != NULL 2037 && child->maxgap >= length) 2038 prev = tmp; 2039 } 2040 } else { 2041 if (tmp->end >= hint && 2042 (prev == NULL || tmp->end < prev->end)) { 2043 if (tmp->gap >= length) 2044 prev = tmp; 2045 else if ((child = RIGHT_ENTRY(tmp)) != NULL 2046 && child->maxgap >= length) 2047 prev = tmp; 2048 } 2049 } 2050 if (tmp->next->start < hint + length) 2051 child = RIGHT_ENTRY(tmp); 2052 else if (tmp->end > hint) 2053 child = LEFT_ENTRY(tmp); 2054 else { 2055 if (tmp->gap >= length) 2056 break; 2057 if (topdown) 2058 child = LEFT_ENTRY(tmp); 2059 else 2060 child = RIGHT_ENTRY(tmp); 2061 } 2062 if (child == NULL || child->maxgap < length) 2063 break; 2064 tmp = child; 2065 } 2066 2067 if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) { 2068 /* 2069 * Check if the entry that we found satifies the 2070 * space requirement 2071 */ 2072 if (topdown) { 2073 if (hint > tmp->next->start - length) 2074 hint = tmp->next->start - length; 2075 } else { 2076 if (hint < tmp->end) 2077 hint = tmp->end; 2078 } 2079 INVARIANTS(); 2080 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2081 flags, topdown, tmp); 2082 INVARIANTS(); 2083 switch (avail) { 2084 case 1: 2085 entry = tmp; 2086 goto found; 2087 case -1: 2088 goto wraparound; 2089 } 2090 if (tmp->gap >= length) 2091 goto listsearch; 2092 } 2093 if (prev == NULL) 2094 goto notfound; 2095 2096 if (topdown) { 2097 KASSERT(orig_hint >= prev->next->start - length || 2098 prev->next->start - length > prev->next->start); 2099 hint = prev->next->start - length; 2100 } else { 2101 KASSERT(orig_hint <= prev->end); 2102 hint = prev->end; 2103 } 2104 INVARIANTS(); 2105 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2106 flags, topdown, prev); 2107 INVARIANTS(); 2108 switch (avail) { 2109 case 1: 2110 entry = prev; 2111 goto found; 2112 case -1: 2113 goto wraparound; 2114 } 2115 if (prev->gap >= length) 2116 goto listsearch; 2117 2118 if (topdown) 2119 tmp = LEFT_ENTRY(prev); 2120 else 2121 tmp = RIGHT_ENTRY(prev); 2122 for (;;) { 2123 KASSERT(tmp); 2124 KASSERTMSG(tmp->maxgap >= length, 2125 "tmp->maxgap=0x%"PRIxVSIZE" length=0x%"PRIxVSIZE, 2126 tmp->maxgap, length); 2127 if (topdown) 2128 child = RIGHT_ENTRY(tmp); 2129 else 2130 child = LEFT_ENTRY(tmp); 2131 if (child && child->maxgap >= length) { 2132 tmp = child; 2133 continue; 2134 } 2135 if (tmp->gap >= length) 2136 break; 2137 if (topdown) 2138 tmp = LEFT_ENTRY(tmp); 2139 else 2140 tmp = RIGHT_ENTRY(tmp); 2141 } 2142 2143 if (topdown) { 2144 KASSERT(orig_hint >= tmp->next->start - length || 2145 tmp->next->start - length > tmp->next->start); 2146 hint = tmp->next->start - length; 2147 } else { 2148 KASSERT(orig_hint <= tmp->end); 2149 hint = tmp->end; 2150 } 2151 INVARIANTS(); 2152 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2153 flags, topdown, tmp); 2154 INVARIANTS(); 2155 switch (avail) { 2156 case 1: 2157 entry = tmp; 2158 goto found; 2159 case -1: 2160 goto wraparound; 2161 } 2162 2163 /* 2164 * The tree fails to find an entry because of offset or alignment 2165 * restrictions. Search the list instead. 2166 */ 2167 listsearch: 2168 /* 2169 * Look through the rest of the map, trying to fit a new region in 2170 * the gap between existing regions, or after the very last region. 2171 * note: entry->end = base VA of current gap, 2172 * entry->next->start = VA of end of current gap 2173 */ 2174 2175 INVARIANTS(); 2176 for (;;) { 2177 /* Update hint for current gap. */ 2178 hint = topdown ? entry->next->start - length : entry->end; 2179 INVARIANTS(); 2180 2181 /* See if it fits. */ 2182 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2183 flags, topdown, entry); 2184 INVARIANTS(); 2185 switch (avail) { 2186 case 1: 2187 goto found; 2188 case -1: 2189 goto wraparound; 2190 } 2191 2192 /* Advance to next/previous gap */ 2193 if (topdown) { 2194 if (entry == &map->header) { 2195 UVMHIST_LOG(maphist, "<- failed (off start)", 2196 0,0,0,0); 2197 goto notfound; 2198 } 2199 entry = entry->prev; 2200 } else { 2201 entry = entry->next; 2202 if (entry == &map->header) { 2203 UVMHIST_LOG(maphist, "<- failed (off end)", 2204 0,0,0,0); 2205 goto notfound; 2206 } 2207 } 2208 } 2209 2210 found: 2211 SAVE_HINT(map, map->hint, entry); 2212 *result = hint; 2213 UVMHIST_LOG(maphist,"<- got it! (result=%#jx)", hint, 0,0,0); 2214 INVARIANTS(); 2215 KASSERT(entry->end <= hint); 2216 KASSERT(hint + length <= entry->next->start); 2217 return (entry); 2218 2219 wraparound: 2220 UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0); 2221 2222 return (NULL); 2223 2224 notfound: 2225 UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0); 2226 2227 return (NULL); 2228#undef INVARIANTS 2229} 2230 2231/* 2232 * U N M A P - m a i n h e l p e r f u n c t i o n s 2233 */ 2234 2235/* 2236 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 2237 * 2238 * => caller must check alignment and size 2239 * => map must be locked by caller 2240 * => we return a list of map entries that we've remove from the map 2241 * in "entry_list" 2242 */ 2243 2244void 2245uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 2246 struct vm_map_entry **entry_list /* OUT */, int flags) 2247{ 2248 struct vm_map_entry *entry, *first_entry, *next; 2249 vaddr_t len; 2250 UVMHIST_FUNC(__func__); 2251 UVMHIST_CALLARGS(maphist,"(map=%#jx, start=%#jx, end=%#jx)", 2252 (uintptr_t)map, start, end, 0); 2253 VM_MAP_RANGE_CHECK(map, start, end); 2254 2255 uvm_map_check(map, "unmap_remove entry"); 2256 2257 /* 2258 * find first entry 2259 */ 2260 2261 if (uvm_map_lookup_entry(map, start, &first_entry) == true) { 2262 /* clip and go... */ 2263 entry = first_entry; 2264 UVM_MAP_CLIP_START(map, entry, start); 2265 /* critical! prevents stale hint */ 2266 SAVE_HINT(map, entry, entry->prev); 2267 } else { 2268 entry = first_entry->next; 2269 } 2270 2271 /* 2272 * save the free space hint 2273 */ 2274 2275 if (map->first_free != &map->header && map->first_free->start >= start) 2276 map->first_free = entry->prev; 2277 2278 /* 2279 * note: we now re-use first_entry for a different task. we remove 2280 * a number of map entries from the map and save them in a linked 2281 * list headed by "first_entry". once we remove them from the map 2282 * the caller should unlock the map and drop the references to the 2283 * backing objects [c.f. uvm_unmap_detach]. the object is to 2284 * separate unmapping from reference dropping. why? 2285 * [1] the map has to be locked for unmapping 2286 * [2] the map need not be locked for reference dropping 2287 * [3] dropping references may trigger pager I/O, and if we hit 2288 * a pager that does synchronous I/O we may have to wait for it. 2289 * [4] we would like all waiting for I/O to occur with maps unlocked 2290 * so that we don't block other threads. 2291 */ 2292 2293 first_entry = NULL; 2294 *entry_list = NULL; 2295 2296 /* 2297 * break up the area into map entry sized regions and unmap. note 2298 * that all mappings have to be removed before we can even consider 2299 * dropping references to amaps or VM objects (otherwise we could end 2300 * up with a mapping to a page on the free list which would be very bad) 2301 */ 2302 2303 while ((entry != &map->header) && (entry->start < end)) { 2304 KASSERT((entry->flags & UVM_MAP_STATIC) == 0); 2305 2306 UVM_MAP_CLIP_END(map, entry, end); 2307 next = entry->next; 2308 len = entry->end - entry->start; 2309 2310 /* 2311 * unwire before removing addresses from the pmap; otherwise 2312 * unwiring will put the entries back into the pmap (XXX). 2313 */ 2314 2315 if (VM_MAPENT_ISWIRED(entry)) { 2316 uvm_map_entry_unwire(map, entry); 2317 } 2318 if (flags & UVM_FLAG_VAONLY) { 2319 2320 /* nothing */ 2321 2322 } else if ((map->flags & VM_MAP_PAGEABLE) == 0) { 2323 2324 /* 2325 * if the map is non-pageable, any pages mapped there 2326 * must be wired and entered with pmap_kenter_pa(), 2327 * and we should free any such pages immediately. 2328 * this is mostly used for kmem_map. 2329 */ 2330 KASSERT(vm_map_pmap(map) == pmap_kernel()); 2331 2332 uvm_km_pgremove_intrsafe(map, entry->start, entry->end); 2333 } else if (UVM_ET_ISOBJ(entry) && 2334 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 2335 panic("%s: kernel object %p %p\n", 2336 __func__, map, entry); 2337 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 2338 /* 2339 * remove mappings the standard way. lock object 2340 * and/or amap to ensure vm_page state does not 2341 * change while in pmap_remove(). 2342 */ 2343 2344#ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 2345 uvm_map_lock_entry(entry, RW_WRITER); 2346#else 2347 uvm_map_lock_entry(entry, RW_READER); 2348#endif 2349 pmap_remove(map->pmap, entry->start, entry->end); 2350 2351 /* 2352 * note: if map is dying, leave pmap_update() for 2353 * later. if the map is to be reused (exec) then 2354 * pmap_update() will be called. if the map is 2355 * being disposed of (exit) then pmap_destroy() 2356 * will be called. 2357 */ 2358 2359 if ((map->flags & VM_MAP_DYING) == 0) { 2360 pmap_update(vm_map_pmap(map)); 2361 } else { 2362 KASSERT(vm_map_pmap(map) != pmap_kernel()); 2363 } 2364 2365 uvm_map_unlock_entry(entry); 2366 } 2367 2368#if defined(UVMDEBUG) 2369 /* 2370 * check if there's remaining mapping, 2371 * which is a bug in caller. 2372 */ 2373 2374 vaddr_t va; 2375 for (va = entry->start; va < entry->end; 2376 va += PAGE_SIZE) { 2377 if (pmap_extract(vm_map_pmap(map), va, NULL)) { 2378 panic("%s: %#"PRIxVADDR" has mapping", 2379 __func__, va); 2380 } 2381 } 2382 2383 if (VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) { 2384 uvm_km_check_empty(map, entry->start, entry->end); 2385 } 2386#endif /* defined(UVMDEBUG) */ 2387 2388 /* 2389 * remove entry from map and put it on our list of entries 2390 * that we've nuked. then go to next entry. 2391 */ 2392 2393 UVMHIST_LOG(maphist, " removed map entry %#jx", 2394 (uintptr_t)entry, 0, 0, 0); 2395 2396 /* critical! prevents stale hint */ 2397 SAVE_HINT(map, entry, entry->prev); 2398 2399 uvm_map_entry_unlink(map, entry); 2400 KASSERT(map->size >= len); 2401 map->size -= len; 2402 entry->prev = NULL; 2403 entry->next = first_entry; 2404 first_entry = entry; 2405 entry = next; 2406 } 2407 2408 uvm_map_check(map, "unmap_remove leave"); 2409 2410 /* 2411 * now we've cleaned up the map and are ready for the caller to drop 2412 * references to the mapped objects. 2413 */ 2414 2415 *entry_list = first_entry; 2416 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 2417 2418 if (map->flags & VM_MAP_WANTVA) { 2419 mutex_enter(&map->misc_lock); 2420 map->flags &= ~VM_MAP_WANTVA; 2421 cv_broadcast(&map->cv); 2422 mutex_exit(&map->misc_lock); 2423 } 2424} 2425 2426/* 2427 * uvm_unmap_detach: drop references in a chain of map entries 2428 * 2429 * => we will free the map entries as we traverse the list. 2430 */ 2431 2432void 2433uvm_unmap_detach(struct vm_map_entry *first_entry, int flags) 2434{ 2435 struct vm_map_entry *next_entry; 2436 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 2437 2438 while (first_entry) { 2439 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 2440 UVMHIST_LOG(maphist, 2441 " detach %#jx: amap=%#jx, obj=%#jx, submap?=%jd", 2442 (uintptr_t)first_entry, 2443 (uintptr_t)first_entry->aref.ar_amap, 2444 (uintptr_t)first_entry->object.uvm_obj, 2445 UVM_ET_ISSUBMAP(first_entry)); 2446 2447 /* 2448 * drop reference to amap, if we've got one 2449 */ 2450 2451 if (first_entry->aref.ar_amap) 2452 uvm_map_unreference_amap(first_entry, flags); 2453 2454 /* 2455 * drop reference to our backing object, if we've got one 2456 */ 2457 2458 KASSERT(!UVM_ET_ISSUBMAP(first_entry)); 2459 if (UVM_ET_ISOBJ(first_entry) && 2460 first_entry->object.uvm_obj->pgops->pgo_detach) { 2461 (*first_entry->object.uvm_obj->pgops->pgo_detach) 2462 (first_entry->object.uvm_obj); 2463 } 2464 next_entry = first_entry->next; 2465 uvm_mapent_free(first_entry); 2466 first_entry = next_entry; 2467 } 2468 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 2469} 2470 2471/* 2472 * E X T R A C T I O N F U N C T I O N S 2473 */ 2474 2475/* 2476 * uvm_map_reserve: reserve space in a vm_map for future use. 2477 * 2478 * => we reserve space in a map by putting a dummy map entry in the 2479 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 2480 * => map should be unlocked (we will write lock it) 2481 * => we return true if we were able to reserve space 2482 * => XXXCDC: should be inline? 2483 */ 2484 2485int 2486uvm_map_reserve(struct vm_map *map, vsize_t size, 2487 vaddr_t offset /* hint for pmap_prefer */, 2488 vsize_t align /* alignment */, 2489 vaddr_t *raddr /* IN:hint, OUT: reserved VA */, 2490 uvm_flag_t flags /* UVM_FLAG_FIXED or UVM_FLAG_COLORMATCH or 0 */) 2491{ 2492 UVMHIST_FUNC(__func__); 2493 UVMHIST_CALLARGS(maphist, "(map=%#jx, size=%#jx, offset=%#jx, addr=%#jx)", 2494 (uintptr_t)map, size, offset, (uintptr_t)raddr); 2495 2496 size = round_page(size); 2497 2498 /* 2499 * reserve some virtual space. 2500 */ 2501 2502 if (uvm_map(map, raddr, size, NULL, offset, align, 2503 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 2504 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) { 2505 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 2506 return (false); 2507 } 2508 2509 UVMHIST_LOG(maphist, "<- done (*raddr=%#jx)", *raddr,0,0,0); 2510 return (true); 2511} 2512 2513/* 2514 * uvm_map_replace: replace a reserved (blank) area of memory with 2515 * real mappings. 2516 * 2517 * => caller must WRITE-LOCK the map 2518 * => we return true if replacement was a success 2519 * => we expect the newents chain to have nnewents entrys on it and 2520 * we expect newents->prev to point to the last entry on the list 2521 * => note newents is allowed to be NULL 2522 */ 2523 2524static int 2525uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end, 2526 struct vm_map_entry *newents, int nnewents, vsize_t nsize, 2527 struct vm_map_entry **oldentryp) 2528{ 2529 struct vm_map_entry *oldent, *last; 2530 2531 uvm_map_check(map, "map_replace entry"); 2532 2533 /* 2534 * first find the blank map entry at the specified address 2535 */ 2536 2537 if (!uvm_map_lookup_entry(map, start, &oldent)) { 2538 return (false); 2539 } 2540 2541 /* 2542 * check to make sure we have a proper blank entry 2543 */ 2544 2545 if (end < oldent->end) { 2546 UVM_MAP_CLIP_END(map, oldent, end); 2547 } 2548 if (oldent->start != start || oldent->end != end || 2549 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 2550 return (false); 2551 } 2552 2553#ifdef DIAGNOSTIC 2554 2555 /* 2556 * sanity check the newents chain 2557 */ 2558 2559 { 2560 struct vm_map_entry *tmpent = newents; 2561 int nent = 0; 2562 vsize_t sz = 0; 2563 vaddr_t cur = start; 2564 2565 while (tmpent) { 2566 nent++; 2567 sz += tmpent->end - tmpent->start; 2568 if (tmpent->start < cur) 2569 panic("uvm_map_replace1"); 2570 if (tmpent->start >= tmpent->end || tmpent->end > end) { 2571 panic("uvm_map_replace2: " 2572 "tmpent->start=%#"PRIxVADDR 2573 ", tmpent->end=%#"PRIxVADDR 2574 ", end=%#"PRIxVADDR, 2575 tmpent->start, tmpent->end, end); 2576 } 2577 cur = tmpent->end; 2578 if (tmpent->next) { 2579 if (tmpent->next->prev != tmpent) 2580 panic("uvm_map_replace3"); 2581 } else { 2582 if (newents->prev != tmpent) 2583 panic("uvm_map_replace4"); 2584 } 2585 tmpent = tmpent->next; 2586 } 2587 if (nent != nnewents) 2588 panic("uvm_map_replace5"); 2589 if (sz != nsize) 2590 panic("uvm_map_replace6"); 2591 } 2592#endif 2593 2594 /* 2595 * map entry is a valid blank! replace it. (this does all the 2596 * work of map entry link/unlink...). 2597 */ 2598 2599 if (newents) { 2600 last = newents->prev; 2601 2602 /* critical: flush stale hints out of map */ 2603 SAVE_HINT(map, map->hint, newents); 2604 if (map->first_free == oldent) 2605 map->first_free = last; 2606 2607 last->next = oldent->next; 2608 last->next->prev = last; 2609 2610 /* Fix RB tree */ 2611 uvm_rb_remove(map, oldent); 2612 2613 newents->prev = oldent->prev; 2614 newents->prev->next = newents; 2615 map->nentries = map->nentries + (nnewents - 1); 2616 2617 /* Fixup the RB tree */ 2618 { 2619 int i; 2620 struct vm_map_entry *tmp; 2621 2622 tmp = newents; 2623 for (i = 0; i < nnewents && tmp; i++) { 2624 uvm_rb_insert(map, tmp); 2625 tmp = tmp->next; 2626 } 2627 } 2628 } else { 2629 /* NULL list of new entries: just remove the old one */ 2630 clear_hints(map, oldent); 2631 uvm_map_entry_unlink(map, oldent); 2632 } 2633 map->size -= end - start - nsize; 2634 2635 uvm_map_check(map, "map_replace leave"); 2636 2637 /* 2638 * now we can free the old blank entry and return. 2639 */ 2640 2641 *oldentryp = oldent; 2642 return (true); 2643} 2644 2645/* 2646 * uvm_map_extract: extract a mapping from a map and put it somewhere 2647 * (maybe removing the old mapping) 2648 * 2649 * => maps should be unlocked (we will write lock them) 2650 * => returns 0 on success, error code otherwise 2651 * => start must be page aligned 2652 * => len must be page sized 2653 * => flags: 2654 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 2655 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 2656 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 2657 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 2658 * UVM_EXTRACT_PROT_ALL: set prot to UVM_PROT_ALL as we go 2659 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 2660 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 2661 * be used from within the kernel in a kernel level map <<< 2662 */ 2663 2664int 2665uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 2666 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags) 2667{ 2668 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge; 2669 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 2670 *deadentry, *oldentry; 2671 struct vm_map_entry *resentry = NULL; /* a dummy reservation entry */ 2672 vsize_t elen __unused; 2673 int nchain, error, copy_ok; 2674 vsize_t nsize; 2675 UVMHIST_FUNC(__func__); 2676 UVMHIST_CALLARGS(maphist,"(srcmap=%#jx,start=%#jx, len=%#jx", 2677 (uintptr_t)srcmap, start, len, 0); 2678 UVMHIST_LOG(maphist," ...,dstmap=%#jx, flags=%#jx)", 2679 (uintptr_t)dstmap, flags, 0, 0); 2680 2681 /* 2682 * step 0: sanity check: start must be on a page boundary, length 2683 * must be page sized. can't ask for CONTIG/QREF if you asked for 2684 * REMOVE. 2685 */ 2686 2687 KASSERTMSG((start & PAGE_MASK) == 0, "start=0x%"PRIxVADDR, start); 2688 KASSERTMSG((len & PAGE_MASK) == 0, "len=0x%"PRIxVADDR, len); 2689 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 2690 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 2691 2692 /* 2693 * step 1: reserve space in the target map for the extracted area 2694 */ 2695 2696 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 2697 dstaddr = vm_map_min(dstmap); 2698 if (!uvm_map_reserve(dstmap, len, start, 2699 atop(start) & uvmexp.colormask, &dstaddr, 2700 UVM_FLAG_COLORMATCH)) 2701 return (ENOMEM); 2702 KASSERT((atop(start ^ dstaddr) & uvmexp.colormask) == 0); 2703 *dstaddrp = dstaddr; /* pass address back to caller */ 2704 UVMHIST_LOG(maphist, " dstaddr=%#jx", dstaddr,0,0,0); 2705 } else { 2706 dstaddr = *dstaddrp; 2707 } 2708 2709 /* 2710 * step 2: setup for the extraction process loop by init'ing the 2711 * map entry chain, locking src map, and looking up the first useful 2712 * entry in the map. 2713 */ 2714 2715 end = start + len; 2716 newend = dstaddr + len; 2717 chain = endchain = NULL; 2718 nchain = 0; 2719 nsize = 0; 2720 vm_map_lock(srcmap); 2721 2722 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 2723 2724 /* "start" is within an entry */ 2725 if (flags & UVM_EXTRACT_QREF) { 2726 2727 /* 2728 * for quick references we don't clip the entry, so 2729 * the entry may map space "before" the starting 2730 * virtual address... this is the "fudge" factor 2731 * (which can be non-zero only the first time 2732 * through the "while" loop in step 3). 2733 */ 2734 2735 fudge = start - entry->start; 2736 } else { 2737 2738 /* 2739 * normal reference: we clip the map to fit (thus 2740 * fudge is zero) 2741 */ 2742 2743 UVM_MAP_CLIP_START(srcmap, entry, start); 2744 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 2745 fudge = 0; 2746 } 2747 } else { 2748 2749 /* "start" is not within an entry ... skip to next entry */ 2750 if (flags & UVM_EXTRACT_CONTIG) { 2751 error = EINVAL; 2752 goto bad; /* definite hole here ... */ 2753 } 2754 2755 entry = entry->next; 2756 fudge = 0; 2757 } 2758 2759 /* save values from srcmap for step 6 */ 2760 orig_entry = entry; 2761 orig_fudge = fudge; 2762 2763 /* 2764 * step 3: now start looping through the map entries, extracting 2765 * as we go. 2766 */ 2767 2768 while (entry->start < end && entry != &srcmap->header) { 2769 2770 /* if we are not doing a quick reference, clip it */ 2771 if ((flags & UVM_EXTRACT_QREF) == 0) 2772 UVM_MAP_CLIP_END(srcmap, entry, end); 2773 2774 /* clear needs_copy (allow chunking) */ 2775 if (UVM_ET_ISNEEDSCOPY(entry)) { 2776 amap_copy(srcmap, entry, 2777 AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end); 2778 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 2779 error = ENOMEM; 2780 goto bad; 2781 } 2782 2783 /* amap_copy could clip (during chunk)! update fudge */ 2784 if (fudge) { 2785 fudge = start - entry->start; 2786 orig_fudge = fudge; 2787 } 2788 } 2789 2790 /* calculate the offset of this from "start" */ 2791 oldoffset = (entry->start + fudge) - start; 2792 2793 /* allocate a new map entry */ 2794 newentry = uvm_mapent_alloc(dstmap, 0); 2795 if (newentry == NULL) { 2796 error = ENOMEM; 2797 goto bad; 2798 } 2799 2800 /* set up new map entry */ 2801 newentry->next = NULL; 2802 newentry->prev = endchain; 2803 newentry->start = dstaddr + oldoffset; 2804 newentry->end = 2805 newentry->start + (entry->end - (entry->start + fudge)); 2806 if (newentry->end > newend || newentry->end < newentry->start) 2807 newentry->end = newend; 2808 newentry->object.uvm_obj = entry->object.uvm_obj; 2809 if (newentry->object.uvm_obj) { 2810 if (newentry->object.uvm_obj->pgops->pgo_reference) 2811 newentry->object.uvm_obj->pgops-> 2812 pgo_reference(newentry->object.uvm_obj); 2813 newentry->offset = entry->offset + fudge; 2814 } else { 2815 newentry->offset = 0; 2816 } 2817 newentry->etype = entry->etype; 2818 if (flags & UVM_EXTRACT_PROT_ALL) { 2819 newentry->protection = newentry->max_protection = 2820 UVM_PROT_ALL; 2821 } else { 2822 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 2823 entry->max_protection : entry->protection; 2824 newentry->max_protection = entry->max_protection; 2825 } 2826 newentry->inheritance = entry->inheritance; 2827 newentry->wired_count = 0; 2828 newentry->aref.ar_amap = entry->aref.ar_amap; 2829 if (newentry->aref.ar_amap) { 2830 newentry->aref.ar_pageoff = 2831 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 2832 uvm_map_reference_amap(newentry, AMAP_SHARED | 2833 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 2834 } else { 2835 newentry->aref.ar_pageoff = 0; 2836 } 2837 newentry->advice = entry->advice; 2838 if ((flags & UVM_EXTRACT_QREF) != 0) { 2839 newentry->flags |= UVM_MAP_NOMERGE; 2840 } 2841 2842 /* now link it on the chain */ 2843 nchain++; 2844 nsize += newentry->end - newentry->start; 2845 if (endchain == NULL) { 2846 chain = endchain = newentry; 2847 } else { 2848 endchain->next = newentry; 2849 endchain = newentry; 2850 } 2851 2852 /* end of 'while' loop! */ 2853 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 2854 (entry->next == &srcmap->header || 2855 entry->next->start != entry->end)) { 2856 error = EINVAL; 2857 goto bad; 2858 } 2859 entry = entry->next; 2860 fudge = 0; 2861 } 2862 2863 /* 2864 * step 4: close off chain (in format expected by uvm_map_replace) 2865 */ 2866 2867 if (chain) 2868 chain->prev = endchain; 2869 2870 /* 2871 * step 5: attempt to lock the dest map so we can pmap_copy. 2872 * note usage of copy_ok: 2873 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 2874 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 2875 */ 2876 2877 if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) { 2878 copy_ok = 1; 2879 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2880 nchain, nsize, &resentry)) { 2881 if (srcmap != dstmap) 2882 vm_map_unlock(dstmap); 2883 error = EIO; 2884 goto bad; 2885 } 2886 } else { 2887 copy_ok = 0; 2888 /* replace deferred until step 7 */ 2889 } 2890 2891 /* 2892 * step 6: traverse the srcmap a second time to do the following: 2893 * - if we got a lock on the dstmap do pmap_copy 2894 * - if UVM_EXTRACT_REMOVE remove the entries 2895 * we make use of orig_entry and orig_fudge (saved in step 2) 2896 */ 2897 2898 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 2899 2900 /* purge possible stale hints from srcmap */ 2901 if (flags & UVM_EXTRACT_REMOVE) { 2902 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 2903 if (srcmap->first_free != &srcmap->header && 2904 srcmap->first_free->start >= start) 2905 srcmap->first_free = orig_entry->prev; 2906 } 2907 2908 entry = orig_entry; 2909 fudge = orig_fudge; 2910 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 2911 2912 while (entry->start < end && entry != &srcmap->header) { 2913 if (copy_ok) { 2914 oldoffset = (entry->start + fudge) - start; 2915 elen = MIN(end, entry->end) - 2916 (entry->start + fudge); 2917 pmap_copy(dstmap->pmap, srcmap->pmap, 2918 dstaddr + oldoffset, elen, 2919 entry->start + fudge); 2920 } 2921 2922 /* we advance "entry" in the following if statement */ 2923 if (flags & UVM_EXTRACT_REMOVE) { 2924#ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 2925 uvm_map_lock_entry(entry, RW_WRITER); 2926#else 2927 uvm_map_lock_entry(entry, RW_READER); 2928#endif 2929 pmap_remove(srcmap->pmap, entry->start, 2930 entry->end); 2931 uvm_map_unlock_entry(entry); 2932 oldentry = entry; /* save entry */ 2933 entry = entry->next; /* advance */ 2934 uvm_map_entry_unlink(srcmap, oldentry); 2935 /* add to dead list */ 2936 oldentry->next = deadentry; 2937 deadentry = oldentry; 2938 } else { 2939 entry = entry->next; /* advance */ 2940 } 2941 2942 /* end of 'while' loop */ 2943 fudge = 0; 2944 } 2945 pmap_update(srcmap->pmap); 2946 2947 /* 2948 * unlock dstmap. we will dispose of deadentry in 2949 * step 7 if needed 2950 */ 2951 2952 if (copy_ok && srcmap != dstmap) 2953 vm_map_unlock(dstmap); 2954 2955 } else { 2956 deadentry = NULL; 2957 } 2958 2959 /* 2960 * step 7: we are done with the source map, unlock. if copy_ok 2961 * is 0 then we have not replaced the dummy mapping in dstmap yet 2962 * and we need to do so now. 2963 */ 2964 2965 vm_map_unlock(srcmap); 2966 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 2967 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 2968 2969 /* now do the replacement if we didn't do it in step 5 */ 2970 if (copy_ok == 0) { 2971 vm_map_lock(dstmap); 2972 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2973 nchain, nsize, &resentry); 2974 vm_map_unlock(dstmap); 2975 2976 if (error == false) { 2977 error = EIO; 2978 goto bad2; 2979 } 2980 } 2981 2982 if (resentry != NULL) 2983 uvm_mapent_free(resentry); 2984 2985 return (0); 2986 2987 /* 2988 * bad: failure recovery 2989 */ 2990bad: 2991 vm_map_unlock(srcmap); 2992bad2: /* src already unlocked */ 2993 if (chain) 2994 uvm_unmap_detach(chain, 2995 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 2996 2997 if (resentry != NULL) 2998 uvm_mapent_free(resentry); 2999 3000 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 3001 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 3002 } 3003 return (error); 3004} 3005 3006/* end of extraction functions */ 3007 3008/* 3009 * uvm_map_submap: punch down part of a map into a submap 3010 * 3011 * => only the kernel_map is allowed to be submapped 3012 * => the purpose of submapping is to break up the locking granularity 3013 * of a larger map 3014 * => the range specified must have been mapped previously with a uvm_map() 3015 * call [with uobj==NULL] to create a blank map entry in the main map. 3016 * [And it had better still be blank!] 3017 * => maps which contain submaps should never be copied or forked. 3018 * => to remove a submap, use uvm_unmap() on the main map 3019 * and then uvm_map_deallocate() the submap. 3020 * => main map must be unlocked. 3021 * => submap must have been init'd and have a zero reference count. 3022 * [need not be locked as we don't actually reference it] 3023 */ 3024 3025int 3026uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 3027 struct vm_map *submap) 3028{ 3029 struct vm_map_entry *entry; 3030 int error; 3031 3032 vm_map_lock(map); 3033 VM_MAP_RANGE_CHECK(map, start, end); 3034 3035 if (uvm_map_lookup_entry(map, start, &entry)) { 3036 UVM_MAP_CLIP_START(map, entry, start); 3037 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 3038 } else { 3039 entry = NULL; 3040 } 3041 3042 if (entry != NULL && 3043 entry->start == start && entry->end == end && 3044 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 3045 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 3046 entry->etype |= UVM_ET_SUBMAP; 3047 entry->object.sub_map = submap; 3048 entry->offset = 0; 3049 uvm_map_reference(submap); 3050 error = 0; 3051 } else { 3052 error = EINVAL; 3053 } 3054 vm_map_unlock(map); 3055 3056 return error; 3057} 3058 3059/* 3060 * uvm_map_protect_user: change map protection on behalf of the user. 3061 * Enforces PAX settings as necessary. 3062 */ 3063int 3064uvm_map_protect_user(struct lwp *l, vaddr_t start, vaddr_t end, 3065 vm_prot_t new_prot) 3066{ 3067 int error; 3068 3069 if ((error = PAX_MPROTECT_VALIDATE(l, new_prot))) 3070 return error; 3071 3072 return uvm_map_protect(&l->l_proc->p_vmspace->vm_map, start, end, 3073 new_prot, false); 3074} 3075 3076 3077/* 3078 * uvm_map_protect: change map protection 3079 * 3080 * => set_max means set max_protection. 3081 * => map must be unlocked. 3082 */ 3083 3084#define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 3085 ~VM_PROT_WRITE : VM_PROT_ALL) 3086 3087int 3088uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 3089 vm_prot_t new_prot, bool set_max) 3090{ 3091 struct vm_map_entry *current, *entry; 3092 int error = 0; 3093 UVMHIST_FUNC(__func__); 3094 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_prot=%#jx)", 3095 (uintptr_t)map, start, end, new_prot); 3096 3097 vm_map_lock(map); 3098 VM_MAP_RANGE_CHECK(map, start, end); 3099 if (uvm_map_lookup_entry(map, start, &entry)) { 3100 UVM_MAP_CLIP_START(map, entry, start); 3101 } else { 3102 entry = entry->next; 3103 } 3104 3105 /* 3106 * make a first pass to check for protection violations. 3107 */ 3108 3109 current = entry; 3110 while ((current != &map->header) && (current->start < end)) { 3111 if (UVM_ET_ISSUBMAP(current)) { 3112 error = EINVAL; 3113 goto out; 3114 } 3115 if ((new_prot & current->max_protection) != new_prot) { 3116 error = EACCES; 3117 goto out; 3118 } 3119 /* 3120 * Don't allow VM_PROT_EXECUTE to be set on entries that 3121 * point to vnodes that are associated with a NOEXEC file 3122 * system. 3123 */ 3124 if (UVM_ET_ISOBJ(current) && 3125 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 3126 struct vnode *vp = 3127 (struct vnode *) current->object.uvm_obj; 3128 3129 if ((new_prot & VM_PROT_EXECUTE) != 0 && 3130 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 3131 error = EACCES; 3132 goto out; 3133 } 3134 } 3135 3136 current = current->next; 3137 } 3138 3139 /* go back and fix up protections (no need to clip this time). */ 3140 3141 current = entry; 3142 while ((current != &map->header) && (current->start < end)) { 3143 vm_prot_t old_prot; 3144 3145 UVM_MAP_CLIP_END(map, current, end); 3146 old_prot = current->protection; 3147 if (set_max) 3148 current->protection = 3149 (current->max_protection = new_prot) & old_prot; 3150 else 3151 current->protection = new_prot; 3152 3153 /* 3154 * update physical map if necessary. worry about copy-on-write 3155 * here -- CHECK THIS XXX 3156 */ 3157 3158 if (current->protection != old_prot) { 3159 /* update pmap! */ 3160#ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 3161 uvm_map_lock_entry(current, RW_WRITER); 3162#else 3163 uvm_map_lock_entry(current, RW_READER); 3164#endif 3165 pmap_protect(map->pmap, current->start, current->end, 3166 current->protection & MASK(current)); 3167 uvm_map_unlock_entry(current); 3168 3169 /* 3170 * If this entry points at a vnode, and the 3171 * protection includes VM_PROT_EXECUTE, mark 3172 * the vnode as VEXECMAP. 3173 */ 3174 if (UVM_ET_ISOBJ(current)) { 3175 struct uvm_object *uobj = 3176 current->object.uvm_obj; 3177 3178 if (UVM_OBJ_IS_VNODE(uobj) && 3179 (current->protection & VM_PROT_EXECUTE)) { 3180 vn_markexec((struct vnode *) uobj); 3181 } 3182 } 3183 } 3184 3185 /* 3186 * If the map is configured to lock any future mappings, 3187 * wire this entry now if the old protection was VM_PROT_NONE 3188 * and the new protection is not VM_PROT_NONE. 3189 */ 3190 3191 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 3192 VM_MAPENT_ISWIRED(current) == 0 && 3193 old_prot == VM_PROT_NONE && 3194 new_prot != VM_PROT_NONE) { 3195 3196 /* 3197 * We must call pmap_update() here because the 3198 * pmap_protect() call above might have removed some 3199 * pmap entries and uvm_map_pageable() might create 3200 * some new pmap entries that rely on the prior 3201 * removals being completely finished. 3202 */ 3203 3204 pmap_update(map->pmap); 3205 3206 if (uvm_map_pageable(map, current->start, 3207 current->end, false, 3208 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 3209 3210 /* 3211 * If locking the entry fails, remember the 3212 * error if it's the first one. Note we 3213 * still continue setting the protection in 3214 * the map, but will return the error 3215 * condition regardless. 3216 * 3217 * XXX Ignore what the actual error is, 3218 * XXX just call it a resource shortage 3219 * XXX so that it doesn't get confused 3220 * XXX what uvm_map_protect() itself would 3221 * XXX normally return. 3222 */ 3223 3224 error = ENOMEM; 3225 } 3226 } 3227 current = current->next; 3228 } 3229 pmap_update(map->pmap); 3230 3231 out: 3232 vm_map_unlock(map); 3233 3234 UVMHIST_LOG(maphist, "<- done, error=%jd",error,0,0,0); 3235 return error; 3236} 3237 3238#undef MASK 3239 3240/* 3241 * uvm_map_inherit: set inheritance code for range of addrs in map. 3242 * 3243 * => map must be unlocked 3244 * => note that the inherit code is used during a "fork". see fork 3245 * code for details. 3246 */ 3247 3248int 3249uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 3250 vm_inherit_t new_inheritance) 3251{ 3252 struct vm_map_entry *entry, *temp_entry; 3253 UVMHIST_FUNC(__func__); 3254 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_inh=%#jx)", 3255 (uintptr_t)map, start, end, new_inheritance); 3256 3257 switch (new_inheritance) { 3258 case MAP_INHERIT_NONE: 3259 case MAP_INHERIT_COPY: 3260 case MAP_INHERIT_SHARE: 3261 case MAP_INHERIT_ZERO: 3262 break; 3263 default: 3264 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3265 return EINVAL; 3266 } 3267 3268 vm_map_lock(map); 3269 VM_MAP_RANGE_CHECK(map, start, end); 3270 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3271 entry = temp_entry; 3272 UVM_MAP_CLIP_START(map, entry, start); 3273 } else { 3274 entry = temp_entry->next; 3275 } 3276 while ((entry != &map->header) && (entry->start < end)) { 3277 UVM_MAP_CLIP_END(map, entry, end); 3278 entry->inheritance = new_inheritance; 3279 entry = entry->next; 3280 } 3281 vm_map_unlock(map); 3282 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3283 return 0; 3284} 3285 3286/* 3287 * uvm_map_advice: set advice code for range of addrs in map. 3288 * 3289 * => map must be unlocked 3290 */ 3291 3292int 3293uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 3294{ 3295 struct vm_map_entry *entry, *temp_entry; 3296 UVMHIST_FUNC(__func__); 3297 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_adv=%#jx)", 3298 (uintptr_t)map, start, end, new_advice); 3299 3300 vm_map_lock(map); 3301 VM_MAP_RANGE_CHECK(map, start, end); 3302 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3303 entry = temp_entry; 3304 UVM_MAP_CLIP_START(map, entry, start); 3305 } else { 3306 entry = temp_entry->next; 3307 } 3308 3309 /* 3310 * XXXJRT: disallow holes? 3311 */ 3312 3313 while ((entry != &map->header) && (entry->start < end)) { 3314 UVM_MAP_CLIP_END(map, entry, end); 3315 3316 switch (new_advice) { 3317 case MADV_NORMAL: 3318 case MADV_RANDOM: 3319 case MADV_SEQUENTIAL: 3320 /* nothing special here */ 3321 break; 3322 3323 default: 3324 vm_map_unlock(map); 3325 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3326 return EINVAL; 3327 } 3328 entry->advice = new_advice; 3329 entry = entry->next; 3330 } 3331 3332 vm_map_unlock(map); 3333 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3334 return 0; 3335} 3336 3337/* 3338 * uvm_map_willneed: apply MADV_WILLNEED 3339 */ 3340 3341int 3342uvm_map_willneed(struct vm_map *map, vaddr_t start, vaddr_t end) 3343{ 3344 struct vm_map_entry *entry; 3345 UVMHIST_FUNC(__func__); 3346 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx)", 3347 (uintptr_t)map, start, end, 0); 3348 3349 vm_map_lock_read(map); 3350 VM_MAP_RANGE_CHECK(map, start, end); 3351 if (!uvm_map_lookup_entry(map, start, &entry)) { 3352 entry = entry->next; 3353 } 3354 while (entry->start < end) { 3355 struct vm_amap * const amap = entry->aref.ar_amap; 3356 struct uvm_object * const uobj = entry->object.uvm_obj; 3357 3358 KASSERT(entry != &map->header); 3359 KASSERT(start < entry->end); 3360 /* 3361 * For now, we handle only the easy but commonly-requested case. 3362 * ie. start prefetching of backing uobj pages. 3363 * 3364 * XXX It might be useful to pmap_enter() the already-in-core 3365 * pages by inventing a "weak" mode for uvm_fault() which would 3366 * only do the PGO_LOCKED pgo_get(). 3367 */ 3368 if (UVM_ET_ISOBJ(entry) && amap == NULL && uobj != NULL) { 3369 off_t offset; 3370 off_t size; 3371 3372 offset = entry->offset; 3373 if (start < entry->start) { 3374 offset += entry->start - start; 3375 } 3376 size = entry->offset + (entry->end - entry->start); 3377 if (entry->end < end) { 3378 size -= end - entry->end; 3379 } 3380 uvm_readahead(uobj, offset, size); 3381 } 3382 entry = entry->next; 3383 } 3384 vm_map_unlock_read(map); 3385 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3386 return 0; 3387} 3388 3389/* 3390 * uvm_map_pageable: sets the pageability of a range in a map. 3391 * 3392 * => wires map entries. should not be used for transient page locking. 3393 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 3394 * => regions specified as not pageable require lock-down (wired) memory 3395 * and page tables. 3396 * => map must never be read-locked 3397 * => if islocked is true, map is already write-locked 3398 * => we always unlock the map, since we must downgrade to a read-lock 3399 * to call uvm_fault_wire() 3400 * => XXXCDC: check this and try and clean it up. 3401 */ 3402 3403int 3404uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 3405 bool new_pageable, int lockflags) 3406{ 3407 struct vm_map_entry *entry, *start_entry, *failed_entry; 3408 int rv; 3409#ifdef DIAGNOSTIC 3410 u_int timestamp_save; 3411#endif 3412 UVMHIST_FUNC(__func__); 3413 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_pageable=%ju)", 3414 (uintptr_t)map, start, end, new_pageable); 3415 KASSERT(map->flags & VM_MAP_PAGEABLE); 3416 3417 if ((lockflags & UVM_LK_ENTER) == 0) 3418 vm_map_lock(map); 3419 VM_MAP_RANGE_CHECK(map, start, end); 3420 3421 /* 3422 * only one pageability change may take place at one time, since 3423 * uvm_fault_wire assumes it will be called only once for each 3424 * wiring/unwiring. therefore, we have to make sure we're actually 3425 * changing the pageability for the entire region. we do so before 3426 * making any changes. 3427 */ 3428 3429 if (uvm_map_lookup_entry(map, start, &start_entry) == false) { 3430 if ((lockflags & UVM_LK_EXIT) == 0) 3431 vm_map_unlock(map); 3432 3433 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 3434 return EFAULT; 3435 } 3436 entry = start_entry; 3437 3438 if (start == end) { /* nothing required */ 3439 if ((lockflags & UVM_LK_EXIT) == 0) 3440 vm_map_unlock(map); 3441 3442 UVMHIST_LOG(maphist,"<- done (nothing)",0,0,0,0); 3443 return 0; 3444 } 3445 3446 /* 3447 * handle wiring and unwiring separately. 3448 */ 3449 3450 if (new_pageable) { /* unwire */ 3451 UVM_MAP_CLIP_START(map, entry, start); 3452 3453 /* 3454 * unwiring. first ensure that the range to be unwired is 3455 * really wired down and that there are no holes. 3456 */ 3457 3458 while ((entry != &map->header) && (entry->start < end)) { 3459 if (entry->wired_count == 0 || 3460 (entry->end < end && 3461 (entry->next == &map->header || 3462 entry->next->start > entry->end))) { 3463 if ((lockflags & UVM_LK_EXIT) == 0) 3464 vm_map_unlock(map); 3465 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 3466 return EINVAL; 3467 } 3468 entry = entry->next; 3469 } 3470 3471 /* 3472 * POSIX 1003.1b - a single munlock call unlocks a region, 3473 * regardless of the number of mlock calls made on that 3474 * region. 3475 */ 3476 3477 entry = start_entry; 3478 while ((entry != &map->header) && (entry->start < end)) { 3479 UVM_MAP_CLIP_END(map, entry, end); 3480 if (VM_MAPENT_ISWIRED(entry)) 3481 uvm_map_entry_unwire(map, entry); 3482 entry = entry->next; 3483 } 3484 if ((lockflags & UVM_LK_EXIT) == 0) 3485 vm_map_unlock(map); 3486 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3487 return 0; 3488 } 3489 3490 /* 3491 * wire case: in two passes [XXXCDC: ugly block of code here] 3492 * 3493 * 1: holding the write lock, we create any anonymous maps that need 3494 * to be created. then we clip each map entry to the region to 3495 * be wired and increment its wiring count. 3496 * 3497 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 3498 * in the pages for any newly wired area (wired_count == 1). 3499 * 3500 * downgrading to a read lock for uvm_fault_wire avoids a possible 3501 * deadlock with another thread that may have faulted on one of 3502 * the pages to be wired (it would mark the page busy, blocking 3503 * us, then in turn block on the map lock that we hold). because 3504 * of problems in the recursive lock package, we cannot upgrade 3505 * to a write lock in vm_map_lookup. thus, any actions that 3506 * require the write lock must be done beforehand. because we 3507 * keep the read lock on the map, the copy-on-write status of the 3508 * entries we modify here cannot change. 3509 */ 3510 3511 while ((entry != &map->header) && (entry->start < end)) { 3512 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3513 3514 /* 3515 * perform actions of vm_map_lookup that need the 3516 * write lock on the map: create an anonymous map 3517 * for a copy-on-write region, or an anonymous map 3518 * for a zero-fill region. (XXXCDC: submap case 3519 * ok?) 3520 */ 3521 3522 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3523 if (UVM_ET_ISNEEDSCOPY(entry) && 3524 ((entry->max_protection & VM_PROT_WRITE) || 3525 (entry->object.uvm_obj == NULL))) { 3526 amap_copy(map, entry, 0, start, end); 3527 /* XXXCDC: wait OK? */ 3528 } 3529 } 3530 } 3531 UVM_MAP_CLIP_START(map, entry, start); 3532 UVM_MAP_CLIP_END(map, entry, end); 3533 entry->wired_count++; 3534 3535 /* 3536 * Check for holes 3537 */ 3538 3539 if (entry->protection == VM_PROT_NONE || 3540 (entry->end < end && 3541 (entry->next == &map->header || 3542 entry->next->start > entry->end))) { 3543 3544 /* 3545 * found one. amap creation actions do not need to 3546 * be undone, but the wired counts need to be restored. 3547 */ 3548 3549 while (entry != &map->header && entry->end > start) { 3550 entry->wired_count--; 3551 entry = entry->prev; 3552 } 3553 if ((lockflags & UVM_LK_EXIT) == 0) 3554 vm_map_unlock(map); 3555 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 3556 return EINVAL; 3557 } 3558 entry = entry->next; 3559 } 3560 3561 /* 3562 * Pass 2. 3563 */ 3564 3565#ifdef DIAGNOSTIC 3566 timestamp_save = map->timestamp; 3567#endif 3568 vm_map_busy(map); 3569 vm_map_unlock(map); 3570 3571 rv = 0; 3572 entry = start_entry; 3573 while (entry != &map->header && entry->start < end) { 3574 if (entry->wired_count == 1) { 3575 rv = uvm_fault_wire(map, entry->start, entry->end, 3576 entry->max_protection, 1); 3577 if (rv) { 3578 3579 /* 3580 * wiring failed. break out of the loop. 3581 * we'll clean up the map below, once we 3582 * have a write lock again. 3583 */ 3584 3585 break; 3586 } 3587 } 3588 entry = entry->next; 3589 } 3590 3591 if (rv) { /* failed? */ 3592 3593 /* 3594 * Get back to an exclusive (write) lock. 3595 */ 3596 3597 vm_map_lock(map); 3598 vm_map_unbusy(map); 3599 3600#ifdef DIAGNOSTIC 3601 if (timestamp_save + 1 != map->timestamp) 3602 panic("uvm_map_pageable: stale map"); 3603#endif 3604 3605 /* 3606 * first drop the wiring count on all the entries 3607 * which haven't actually been wired yet. 3608 */ 3609 3610 failed_entry = entry; 3611 while (entry != &map->header && entry->start < end) { 3612 entry->wired_count--; 3613 entry = entry->next; 3614 } 3615 3616 /* 3617 * now, unwire all the entries that were successfully 3618 * wired above. 3619 */ 3620 3621 entry = start_entry; 3622 while (entry != failed_entry) { 3623 entry->wired_count--; 3624 if (VM_MAPENT_ISWIRED(entry) == 0) 3625 uvm_map_entry_unwire(map, entry); 3626 entry = entry->next; 3627 } 3628 if ((lockflags & UVM_LK_EXIT) == 0) 3629 vm_map_unlock(map); 3630 UVMHIST_LOG(maphist, "<- done (RV=%jd)", rv,0,0,0); 3631 return (rv); 3632 } 3633 3634 if ((lockflags & UVM_LK_EXIT) == 0) { 3635 vm_map_unbusy(map); 3636 } else { 3637 3638 /* 3639 * Get back to an exclusive (write) lock. 3640 */ 3641 3642 vm_map_lock(map); 3643 vm_map_unbusy(map); 3644 } 3645 3646 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3647 return 0; 3648} 3649 3650/* 3651 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 3652 * all mapped regions. 3653 * 3654 * => map must not be locked. 3655 * => if no flags are specified, all regions are unwired. 3656 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 3657 */ 3658 3659int 3660uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 3661{ 3662 struct vm_map_entry *entry, *failed_entry; 3663 vsize_t size; 3664 int rv; 3665#ifdef DIAGNOSTIC 3666 u_int timestamp_save; 3667#endif 3668 UVMHIST_FUNC(__func__); 3669 UVMHIST_CALLARGS(maphist,"(map=%#jx,flags=%#jx)", (uintptr_t)map, flags, 3670 0, 0); 3671 3672 KASSERT(map->flags & VM_MAP_PAGEABLE); 3673 3674 vm_map_lock(map); 3675 3676 /* 3677 * handle wiring and unwiring separately. 3678 */ 3679 3680 if (flags == 0) { /* unwire */ 3681 3682 /* 3683 * POSIX 1003.1b -- munlockall unlocks all regions, 3684 * regardless of how many times mlockall has been called. 3685 */ 3686 3687 for (entry = map->header.next; entry != &map->header; 3688 entry = entry->next) { 3689 if (VM_MAPENT_ISWIRED(entry)) 3690 uvm_map_entry_unwire(map, entry); 3691 } 3692 map->flags &= ~VM_MAP_WIREFUTURE; 3693 vm_map_unlock(map); 3694 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3695 return 0; 3696 } 3697 3698 if (flags & MCL_FUTURE) { 3699 3700 /* 3701 * must wire all future mappings; remember this. 3702 */ 3703 3704 map->flags |= VM_MAP_WIREFUTURE; 3705 } 3706 3707 if ((flags & MCL_CURRENT) == 0) { 3708 3709 /* 3710 * no more work to do! 3711 */ 3712 3713 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 3714 vm_map_unlock(map); 3715 return 0; 3716 } 3717 3718 /* 3719 * wire case: in three passes [XXXCDC: ugly block of code here] 3720 * 3721 * 1: holding the write lock, count all pages mapped by non-wired 3722 * entries. if this would cause us to go over our limit, we fail. 3723 * 3724 * 2: still holding the write lock, we create any anonymous maps that 3725 * need to be created. then we increment its wiring count. 3726 * 3727 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 3728 * in the pages for any newly wired area (wired_count == 1). 3729 * 3730 * downgrading to a read lock for uvm_fault_wire avoids a possible 3731 * deadlock with another thread that may have faulted on one of 3732 * the pages to be wired (it would mark the page busy, blocking 3733 * us, then in turn block on the map lock that we hold). because 3734 * of problems in the recursive lock package, we cannot upgrade 3735 * to a write lock in vm_map_lookup. thus, any actions that 3736 * require the write lock must be done beforehand. because we 3737 * keep the read lock on the map, the copy-on-write status of the 3738 * entries we modify here cannot change. 3739 */ 3740 3741 for (size = 0, entry = map->header.next; entry != &map->header; 3742 entry = entry->next) { 3743 if (entry->protection != VM_PROT_NONE && 3744 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3745 size += entry->end - entry->start; 3746 } 3747 } 3748 3749 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 3750 vm_map_unlock(map); 3751 return ENOMEM; 3752 } 3753 3754 if (limit != 0 && 3755 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 3756 vm_map_unlock(map); 3757 return ENOMEM; 3758 } 3759 3760 /* 3761 * Pass 2. 3762 */ 3763 3764 for (entry = map->header.next; entry != &map->header; 3765 entry = entry->next) { 3766 if (entry->protection == VM_PROT_NONE) 3767 continue; 3768 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3769 3770 /* 3771 * perform actions of vm_map_lookup that need the 3772 * write lock on the map: create an anonymous map 3773 * for a copy-on-write region, or an anonymous map 3774 * for a zero-fill region. (XXXCDC: submap case 3775 * ok?) 3776 */ 3777 3778 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3779 if (UVM_ET_ISNEEDSCOPY(entry) && 3780 ((entry->max_protection & VM_PROT_WRITE) || 3781 (entry->object.uvm_obj == NULL))) { 3782 amap_copy(map, entry, 0, entry->start, 3783 entry->end); 3784 /* XXXCDC: wait OK? */ 3785 } 3786 } 3787 } 3788 entry->wired_count++; 3789 } 3790 3791 /* 3792 * Pass 3. 3793 */ 3794 3795#ifdef DIAGNOSTIC 3796 timestamp_save = map->timestamp; 3797#endif 3798 vm_map_busy(map); 3799 vm_map_unlock(map); 3800 3801 rv = 0; 3802 for (entry = map->header.next; entry != &map->header; 3803 entry = entry->next) { 3804 if (entry->wired_count == 1) { 3805 rv = uvm_fault_wire(map, entry->start, entry->end, 3806 entry->max_protection, 1); 3807 if (rv) { 3808 3809 /* 3810 * wiring failed. break out of the loop. 3811 * we'll clean up the map below, once we 3812 * have a write lock again. 3813 */ 3814 3815 break; 3816 } 3817 } 3818 } 3819 3820 if (rv) { 3821 3822 /* 3823 * Get back an exclusive (write) lock. 3824 */ 3825 3826 vm_map_lock(map); 3827 vm_map_unbusy(map); 3828 3829#ifdef DIAGNOSTIC 3830 if (timestamp_save + 1 != map->timestamp) 3831 panic("uvm_map_pageable_all: stale map"); 3832#endif 3833 3834 /* 3835 * first drop the wiring count on all the entries 3836 * which haven't actually been wired yet. 3837 * 3838 * Skip VM_PROT_NONE entries like we did above. 3839 */ 3840 3841 failed_entry = entry; 3842 for (/* nothing */; entry != &map->header; 3843 entry = entry->next) { 3844 if (entry->protection == VM_PROT_NONE) 3845 continue; 3846 entry->wired_count--; 3847 } 3848 3849 /* 3850 * now, unwire all the entries that were successfully 3851 * wired above. 3852 * 3853 * Skip VM_PROT_NONE entries like we did above. 3854 */ 3855 3856 for (entry = map->header.next; entry != failed_entry; 3857 entry = entry->next) { 3858 if (entry->protection == VM_PROT_NONE) 3859 continue; 3860 entry->wired_count--; 3861 if (VM_MAPENT_ISWIRED(entry)) 3862 uvm_map_entry_unwire(map, entry); 3863 } 3864 vm_map_unlock(map); 3865 UVMHIST_LOG(maphist,"<- done (RV=%jd)", rv,0,0,0); 3866 return (rv); 3867 } 3868 3869 vm_map_unbusy(map); 3870 3871 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3872 return 0; 3873} 3874 3875/* 3876 * uvm_map_clean: clean out a map range 3877 * 3878 * => valid flags: 3879 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 3880 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 3881 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 3882 * if (flags & PGO_FREE): any cached pages are freed after clean 3883 * => returns an error if any part of the specified range isn't mapped 3884 * => never a need to flush amap layer since the anonymous memory has 3885 * no permanent home, but may deactivate pages there 3886 * => called from sys_msync() and sys_madvise() 3887 * => caller must not have map locked 3888 */ 3889 3890int 3891uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 3892{ 3893 struct vm_map_entry *current, *entry; 3894 struct uvm_object *uobj; 3895 struct vm_amap *amap; 3896 struct vm_anon *anon; 3897 struct vm_page *pg; 3898 vaddr_t offset; 3899 vsize_t size; 3900 voff_t uoff; 3901 int error, refs; 3902 UVMHIST_FUNC(__func__); 3903 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,flags=%#jx)", 3904 (uintptr_t)map, start, end, flags); 3905 3906 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 3907 (PGO_FREE|PGO_DEACTIVATE)); 3908 3909 vm_map_lock(map); 3910 VM_MAP_RANGE_CHECK(map, start, end); 3911 if (!uvm_map_lookup_entry(map, start, &entry)) { 3912 vm_map_unlock(map); 3913 return EFAULT; 3914 } 3915 3916 /* 3917 * Make a first pass to check for holes and wiring problems. 3918 */ 3919 3920 for (current = entry; current->start < end; current = current->next) { 3921 if (UVM_ET_ISSUBMAP(current)) { 3922 vm_map_unlock(map); 3923 return EINVAL; 3924 } 3925 if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) { 3926 vm_map_unlock(map); 3927 return EBUSY; 3928 } 3929 if (end <= current->end) { 3930 break; 3931 } 3932 if (current->end != current->next->start) { 3933 vm_map_unlock(map); 3934 return EFAULT; 3935 } 3936 } 3937 3938 vm_map_busy(map); 3939 vm_map_unlock(map); 3940 error = 0; 3941 for (current = entry; start < end; current = current->next) { 3942 amap = current->aref.ar_amap; /* upper layer */ 3943 uobj = current->object.uvm_obj; /* lower layer */ 3944 KASSERT(start >= current->start); 3945 3946 /* 3947 * No amap cleaning necessary if: 3948 * 3949 * (1) There's no amap. 3950 * 3951 * (2) We're not deactivating or freeing pages. 3952 */ 3953 3954 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 3955 goto flush_object; 3956 3957 offset = start - current->start; 3958 size = MIN(end, current->end) - start; 3959 3960 amap_lock(amap, RW_WRITER); 3961 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 3962 anon = amap_lookup(¤t->aref, offset); 3963 if (anon == NULL) 3964 continue; 3965 3966 KASSERT(anon->an_lock == amap->am_lock); 3967 pg = anon->an_page; 3968 if (pg == NULL) { 3969 continue; 3970 } 3971 if (pg->flags & PG_BUSY) { 3972 continue; 3973 } 3974 3975 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 3976 3977 /* 3978 * In these first 3 cases, we just deactivate the page. 3979 */ 3980 3981 case PGO_CLEANIT|PGO_FREE: 3982 case PGO_CLEANIT|PGO_DEACTIVATE: 3983 case PGO_DEACTIVATE: 3984 deactivate_it: 3985 /* 3986 * skip the page if it's loaned or wired, 3987 * since it shouldn't be on a paging queue 3988 * at all in these cases. 3989 */ 3990 3991 if (pg->loan_count != 0 || 3992 pg->wire_count != 0) { 3993 continue; 3994 } 3995 KASSERT(pg->uanon == anon); 3996 uvm_pagelock(pg); 3997 uvm_pagedeactivate(pg); 3998 uvm_pageunlock(pg); 3999 continue; 4000 4001 case PGO_FREE: 4002 4003 /* 4004 * If there are multiple references to 4005 * the amap, just deactivate the page. 4006 */ 4007 4008 if (amap_refs(amap) > 1) 4009 goto deactivate_it; 4010 4011 /* skip the page if it's wired */ 4012 if (pg->wire_count != 0) { 4013 continue; 4014 } 4015 amap_unadd(¤t->aref, offset); 4016 refs = --anon->an_ref; 4017 if (refs == 0) { 4018 uvm_anfree(anon); 4019 } 4020 continue; 4021 } 4022 } 4023 amap_unlock(amap); 4024 4025 flush_object: 4026 /* 4027 * flush pages if we've got a valid backing object. 4028 * note that we must always clean object pages before 4029 * freeing them since otherwise we could reveal stale 4030 * data from files. 4031 */ 4032 4033 uoff = current->offset + (start - current->start); 4034 size = MIN(end, current->end) - start; 4035 if (uobj != NULL) { 4036 rw_enter(uobj->vmobjlock, RW_WRITER); 4037 if (uobj->pgops->pgo_put != NULL) 4038 error = (uobj->pgops->pgo_put)(uobj, uoff, 4039 uoff + size, flags | PGO_CLEANIT); 4040 else 4041 error = 0; 4042 } 4043 start += size; 4044 } 4045 vm_map_unbusy(map); 4046 return error; 4047} 4048 4049 4050/* 4051 * uvm_map_checkprot: check protection in map 4052 * 4053 * => must allow specified protection in a fully allocated region. 4054 * => map must be read or write locked by caller. 4055 */ 4056 4057bool 4058uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 4059 vm_prot_t protection) 4060{ 4061 struct vm_map_entry *entry; 4062 struct vm_map_entry *tmp_entry; 4063 4064 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 4065 return (false); 4066 } 4067 entry = tmp_entry; 4068 while (start < end) { 4069 if (entry == &map->header) { 4070 return (false); 4071 } 4072 4073 /* 4074 * no holes allowed 4075 */ 4076 4077 if (start < entry->start) { 4078 return (false); 4079 } 4080 4081 /* 4082 * check protection associated with entry 4083 */ 4084 4085 if ((entry->protection & protection) != protection) { 4086 return (false); 4087 } 4088 start = entry->end; 4089 entry = entry->next; 4090 } 4091 return (true); 4092} 4093 4094/* 4095 * uvmspace_alloc: allocate a vmspace structure. 4096 * 4097 * - structure includes vm_map and pmap 4098 * - XXX: no locking on this structure 4099 * - refcnt set to 1, rest must be init'd by caller 4100 */ 4101struct vmspace * 4102uvmspace_alloc(vaddr_t vmin, vaddr_t vmax, bool topdown) 4103{ 4104 struct vmspace *vm; 4105 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4106 4107 vm = kmem_alloc(sizeof(*vm), KM_SLEEP); 4108 uvmspace_init(vm, NULL, vmin, vmax, topdown); 4109 UVMHIST_LOG(maphist,"<- done (vm=%#jx)", (uintptr_t)vm, 0, 0, 0); 4110 return (vm); 4111} 4112 4113/* 4114 * uvmspace_init: initialize a vmspace structure. 4115 * 4116 * - XXX: no locking on this structure 4117 * - refcnt set to 1, rest must be init'd by caller 4118 */ 4119void 4120uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, 4121 vaddr_t vmax, bool topdown) 4122{ 4123 UVMHIST_FUNC(__func__); 4124 UVMHIST_CALLARGS(maphist, "(vm=%#jx, pmap=%#jx, vmin=%#jx, vmax=%#jx", 4125 (uintptr_t)vm, (uintptr_t)pmap, vmin, vmax); 4126 UVMHIST_LOG(maphist, " topdown=%ju)", topdown, 0, 0, 0); 4127 4128 memset(vm, 0, sizeof(*vm)); 4129 uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE 4130 | (topdown ? VM_MAP_TOPDOWN : 0) 4131 ); 4132 if (pmap) 4133 pmap_reference(pmap); 4134 else 4135 pmap = pmap_create(); 4136 vm->vm_map.pmap = pmap; 4137 vm->vm_refcnt = 1; 4138 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4139} 4140 4141/* 4142 * uvmspace_share: share a vmspace between two processes 4143 * 4144 * - used for vfork, threads(?) 4145 */ 4146 4147void 4148uvmspace_share(struct proc *p1, struct proc *p2) 4149{ 4150 4151 uvmspace_addref(p1->p_vmspace); 4152 p2->p_vmspace = p1->p_vmspace; 4153} 4154 4155#if 0 4156 4157/* 4158 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 4159 * 4160 * - XXX: no locking on vmspace 4161 */ 4162 4163void 4164uvmspace_unshare(struct lwp *l) 4165{ 4166 struct proc *p = l->l_proc; 4167 struct vmspace *nvm, *ovm = p->p_vmspace; 4168 4169 if (ovm->vm_refcnt == 1) 4170 /* nothing to do: vmspace isn't shared in the first place */ 4171 return; 4172 4173 /* make a new vmspace, still holding old one */ 4174 nvm = uvmspace_fork(ovm); 4175 4176 kpreempt_disable(); 4177 pmap_deactivate(l); /* unbind old vmspace */ 4178 p->p_vmspace = nvm; 4179 pmap_activate(l); /* switch to new vmspace */ 4180 kpreempt_enable(); 4181 4182 uvmspace_free(ovm); /* drop reference to old vmspace */ 4183} 4184 4185#endif 4186 4187 4188/* 4189 * uvmspace_spawn: a new process has been spawned and needs a vmspace 4190 */ 4191 4192void 4193uvmspace_spawn(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown) 4194{ 4195 struct proc *p = l->l_proc; 4196 struct vmspace *nvm; 4197 4198#ifdef __HAVE_CPU_VMSPACE_EXEC 4199 cpu_vmspace_exec(l, start, end); 4200#endif 4201 4202 nvm = uvmspace_alloc(start, end, topdown); 4203 kpreempt_disable(); 4204 p->p_vmspace = nvm; 4205 pmap_activate(l); 4206 kpreempt_enable(); 4207} 4208 4209/* 4210 * uvmspace_exec: the process wants to exec a new program 4211 */ 4212 4213void 4214uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown) 4215{ 4216 struct proc *p = l->l_proc; 4217 struct vmspace *nvm, *ovm = p->p_vmspace; 4218 struct vm_map *map; 4219 int flags; 4220 4221 KASSERT(ovm != NULL); 4222#ifdef __HAVE_CPU_VMSPACE_EXEC 4223 cpu_vmspace_exec(l, start, end); 4224#endif 4225 4226 map = &ovm->vm_map; 4227 /* 4228 * see if more than one process is using this vmspace... 4229 */ 4230 4231 if (ovm->vm_refcnt == 1 4232 && topdown == ((ovm->vm_map.flags & VM_MAP_TOPDOWN) != 0)) { 4233 4234 /* 4235 * if p is the only process using its vmspace then we can safely 4236 * recycle that vmspace for the program that is being exec'd. 4237 * But only if TOPDOWN matches the requested value for the new 4238 * vm space! 4239 */ 4240 4241 /* 4242 * SYSV SHM semantics require us to kill all segments on an exec 4243 */ 4244 if (uvm_shmexit && ovm->vm_shm) 4245 (*uvm_shmexit)(ovm); 4246 4247 /* 4248 * POSIX 1003.1b -- "lock future mappings" is revoked 4249 * when a process execs another program image. 4250 */ 4251 4252 map->flags &= ~VM_MAP_WIREFUTURE; 4253 4254 /* 4255 * now unmap the old program. 4256 * 4257 * XXX set VM_MAP_DYING for the duration, so pmap_update() 4258 * is not called until the pmap has been totally cleared out 4259 * after pmap_remove_all(), or it can confuse some pmap 4260 * implementations. it would be nice to handle this by 4261 * deferring the pmap_update() while it is known the address 4262 * space is not visible to any user LWP other than curlwp, 4263 * but there isn't an elegant way of inferring that right 4264 * now. 4265 */ 4266 4267 flags = pmap_remove_all(map->pmap) ? UVM_FLAG_VAONLY : 0; 4268 map->flags |= VM_MAP_DYING; 4269 uvm_unmap1(map, vm_map_min(map), vm_map_max(map), flags); 4270 map->flags &= ~VM_MAP_DYING; 4271 pmap_update(map->pmap); 4272 KASSERT(map->header.prev == &map->header); 4273 KASSERT(map->nentries == 0); 4274 4275 /* 4276 * resize the map 4277 */ 4278 4279 vm_map_setmin(map, start); 4280 vm_map_setmax(map, end); 4281 } else { 4282 4283 /* 4284 * p's vmspace is being shared, so we can't reuse it for p since 4285 * it is still being used for others. allocate a new vmspace 4286 * for p 4287 */ 4288 4289 nvm = uvmspace_alloc(start, end, topdown); 4290 4291 /* 4292 * install new vmspace and drop our ref to the old one. 4293 */ 4294 4295 kpreempt_disable(); 4296 pmap_deactivate(l); 4297 p->p_vmspace = nvm; 4298 pmap_activate(l); 4299 kpreempt_enable(); 4300 4301 uvmspace_free(ovm); 4302 } 4303} 4304 4305/* 4306 * uvmspace_addref: add a reference to a vmspace. 4307 */ 4308 4309void 4310uvmspace_addref(struct vmspace *vm) 4311{ 4312 4313 KASSERT((vm->vm_map.flags & VM_MAP_DYING) == 0); 4314 KASSERT(vm->vm_refcnt > 0); 4315 atomic_inc_uint(&vm->vm_refcnt); 4316} 4317 4318/* 4319 * uvmspace_free: free a vmspace data structure 4320 */ 4321 4322void 4323uvmspace_free(struct vmspace *vm) 4324{ 4325 struct vm_map_entry *dead_entries; 4326 struct vm_map *map = &vm->vm_map; 4327 int flags; 4328 4329 UVMHIST_FUNC(__func__); 4330 UVMHIST_CALLARGS(maphist,"(vm=%#jx) ref=%jd", (uintptr_t)vm, 4331 vm->vm_refcnt, 0, 0); 4332 4333 membar_release(); 4334 if (atomic_dec_uint_nv(&vm->vm_refcnt) > 0) 4335 return; 4336 membar_acquire(); 4337 4338 /* 4339 * at this point, there should be no other references to the map. 4340 * delete all of the mappings, then destroy the pmap. 4341 */ 4342 4343 map->flags |= VM_MAP_DYING; 4344 flags = pmap_remove_all(map->pmap) ? UVM_FLAG_VAONLY : 0; 4345 4346 /* Get rid of any SYSV shared memory segments. */ 4347 if (uvm_shmexit && vm->vm_shm != NULL) 4348 (*uvm_shmexit)(vm); 4349 4350 if (map->nentries) { 4351 uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map), 4352 &dead_entries, flags); 4353 if (dead_entries != NULL) 4354 uvm_unmap_detach(dead_entries, 0); 4355 } 4356 KASSERT(map->nentries == 0); 4357 KASSERT(map->size == 0); 4358 4359 mutex_destroy(&map->misc_lock); 4360 rw_destroy(&map->lock); 4361 cv_destroy(&map->cv); 4362 pmap_destroy(map->pmap); 4363 kmem_free(vm, sizeof(*vm)); 4364} 4365 4366static struct vm_map_entry * 4367uvm_mapent_clone(struct vm_map *new_map, struct vm_map_entry *old_entry, 4368 int flags) 4369{ 4370 struct vm_map_entry *new_entry; 4371 4372 new_entry = uvm_mapent_alloc(new_map, 0); 4373 /* old_entry -> new_entry */ 4374 uvm_mapent_copy(old_entry, new_entry); 4375 4376 /* new pmap has nothing wired in it */ 4377 new_entry->wired_count = 0; 4378 4379 /* 4380 * gain reference to object backing the map (can't 4381 * be a submap, already checked this case). 4382 */ 4383 4384 if (new_entry->aref.ar_amap) 4385 uvm_map_reference_amap(new_entry, flags); 4386 4387 if (new_entry->object.uvm_obj && 4388 new_entry->object.uvm_obj->pgops->pgo_reference) 4389 new_entry->object.uvm_obj->pgops->pgo_reference( 4390 new_entry->object.uvm_obj); 4391 4392 /* insert entry at end of new_map's entry list */ 4393 uvm_map_entry_link(new_map, new_map->header.prev, 4394 new_entry); 4395 4396 return new_entry; 4397} 4398 4399/* 4400 * share the mapping: this means we want the old and 4401 * new entries to share amaps and backing objects. 4402 */ 4403static void 4404uvm_mapent_forkshared(struct vm_map *new_map, struct vm_map *old_map, 4405 struct vm_map_entry *old_entry) 4406{ 4407 /* 4408 * if the old_entry needs a new amap (due to prev fork) 4409 * then we need to allocate it now so that we have 4410 * something we own to share with the new_entry. [in 4411 * other words, we need to clear needs_copy] 4412 */ 4413 4414 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 4415 /* get our own amap, clears needs_copy */ 4416 amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK, 4417 0, 0); 4418 /* XXXCDC: WAITOK??? */ 4419 } 4420 4421 uvm_mapent_clone(new_map, old_entry, AMAP_SHARED); 4422} 4423 4424 4425static void 4426uvm_mapent_forkcopy(struct vm_map *new_map, struct vm_map *old_map, 4427 struct vm_map_entry *old_entry) 4428{ 4429 struct vm_map_entry *new_entry; 4430 4431 /* 4432 * copy-on-write the mapping (using mmap's 4433 * MAP_PRIVATE semantics) 4434 * 4435 * allocate new_entry, adjust reference counts. 4436 * (note that new references are read-only). 4437 */ 4438 4439 new_entry = uvm_mapent_clone(new_map, old_entry, 0); 4440 4441 new_entry->etype |= 4442 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 4443 4444 /* 4445 * the new entry will need an amap. it will either 4446 * need to be copied from the old entry or created 4447 * from scratch (if the old entry does not have an 4448 * amap). can we defer this process until later 4449 * (by setting "needs_copy") or do we need to copy 4450 * the amap now? 4451 * 4452 * we must copy the amap now if any of the following 4453 * conditions hold: 4454 * 1. the old entry has an amap and that amap is 4455 * being shared. this means that the old (parent) 4456 * process is sharing the amap with another 4457 * process. if we do not clear needs_copy here 4458 * we will end up in a situation where both the 4459 * parent and child process are referring to the 4460 * same amap with "needs_copy" set. if the 4461 * parent write-faults, the fault routine will 4462 * clear "needs_copy" in the parent by allocating 4463 * a new amap. this is wrong because the 4464 * parent is supposed to be sharing the old amap 4465 * and the new amap will break that. 4466 * 4467 * 2. if the old entry has an amap and a non-zero 4468 * wire count then we are going to have to call 4469 * amap_cow_now to avoid page faults in the 4470 * parent process. since amap_cow_now requires 4471 * "needs_copy" to be clear we might as well 4472 * clear it here as well. 4473 * 4474 */ 4475 4476 if (old_entry->aref.ar_amap != NULL) { 4477 if ((amap_flags(old_entry->aref.ar_amap) & AMAP_SHARED) != 0 || 4478 VM_MAPENT_ISWIRED(old_entry)) { 4479 4480 amap_copy(new_map, new_entry, 4481 AMAP_COPY_NOCHUNK, 0, 0); 4482 /* XXXCDC: M_WAITOK ... ok? */ 4483 } 4484 } 4485 4486 /* 4487 * if the parent's entry is wired down, then the 4488 * parent process does not want page faults on 4489 * access to that memory. this means that we 4490 * cannot do copy-on-write because we can't write 4491 * protect the old entry. in this case we 4492 * resolve all copy-on-write faults now, using 4493 * amap_cow_now. note that we have already 4494 * allocated any needed amap (above). 4495 */ 4496 4497 if (VM_MAPENT_ISWIRED(old_entry)) { 4498 4499 /* 4500 * resolve all copy-on-write faults now 4501 * (note that there is nothing to do if 4502 * the old mapping does not have an amap). 4503 */ 4504 if (old_entry->aref.ar_amap) 4505 amap_cow_now(new_map, new_entry); 4506 4507 } else { 4508 /* 4509 * setup mappings to trigger copy-on-write faults 4510 * we must write-protect the parent if it has 4511 * an amap and it is not already "needs_copy"... 4512 * if it is already "needs_copy" then the parent 4513 * has already been write-protected by a previous 4514 * fork operation. 4515 */ 4516 if (old_entry->aref.ar_amap && 4517 !UVM_ET_ISNEEDSCOPY(old_entry)) { 4518 if (old_entry->max_protection & VM_PROT_WRITE) { 4519#ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 4520 uvm_map_lock_entry(old_entry, RW_WRITER); 4521#else 4522 uvm_map_lock_entry(old_entry, RW_READER); 4523#endif 4524 pmap_protect(old_map->pmap, 4525 old_entry->start, old_entry->end, 4526 old_entry->protection & ~VM_PROT_WRITE); 4527 uvm_map_unlock_entry(old_entry); 4528 } 4529 old_entry->etype |= UVM_ET_NEEDSCOPY; 4530 } 4531 } 4532} 4533 4534/* 4535 * zero the mapping: the new entry will be zero initialized 4536 */ 4537static void 4538uvm_mapent_forkzero(struct vm_map *new_map, struct vm_map *old_map, 4539 struct vm_map_entry *old_entry) 4540{ 4541 struct vm_map_entry *new_entry; 4542 4543 new_entry = uvm_mapent_clone(new_map, old_entry, 0); 4544 4545 new_entry->etype |= 4546 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 4547 4548 if (new_entry->aref.ar_amap) { 4549 uvm_map_unreference_amap(new_entry, 0); 4550 new_entry->aref.ar_pageoff = 0; 4551 new_entry->aref.ar_amap = NULL; 4552 } 4553 4554 if (UVM_ET_ISOBJ(new_entry)) { 4555 if (new_entry->object.uvm_obj->pgops->pgo_detach) 4556 new_entry->object.uvm_obj->pgops->pgo_detach( 4557 new_entry->object.uvm_obj); 4558 new_entry->object.uvm_obj = NULL; 4559 new_entry->offset = 0; 4560 new_entry->etype &= ~UVM_ET_OBJ; 4561 } 4562} 4563 4564/* 4565 * F O R K - m a i n e n t r y p o i n t 4566 */ 4567/* 4568 * uvmspace_fork: fork a process' main map 4569 * 4570 * => create a new vmspace for child process from parent. 4571 * => parent's map must not be locked. 4572 */ 4573 4574struct vmspace * 4575uvmspace_fork(struct vmspace *vm1) 4576{ 4577 struct vmspace *vm2; 4578 struct vm_map *old_map = &vm1->vm_map; 4579 struct vm_map *new_map; 4580 struct vm_map_entry *old_entry; 4581 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4582 4583 vm_map_lock(old_map); 4584 4585 vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), 4586 vm1->vm_map.flags & VM_MAP_TOPDOWN); 4587 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 4588 (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy); 4589 new_map = &vm2->vm_map; /* XXX */ 4590 4591 old_entry = old_map->header.next; 4592 new_map->size = old_map->size; 4593 4594 /* 4595 * go entry-by-entry 4596 */ 4597 4598 while (old_entry != &old_map->header) { 4599 4600 /* 4601 * first, some sanity checks on the old entry 4602 */ 4603 4604 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 4605 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 4606 !UVM_ET_ISNEEDSCOPY(old_entry)); 4607 4608 switch (old_entry->inheritance) { 4609 case MAP_INHERIT_NONE: 4610 /* 4611 * drop the mapping, modify size 4612 */ 4613 new_map->size -= old_entry->end - old_entry->start; 4614 break; 4615 4616 case MAP_INHERIT_SHARE: 4617 uvm_mapent_forkshared(new_map, old_map, old_entry); 4618 break; 4619 4620 case MAP_INHERIT_COPY: 4621 uvm_mapent_forkcopy(new_map, old_map, old_entry); 4622 break; 4623 4624 case MAP_INHERIT_ZERO: 4625 uvm_mapent_forkzero(new_map, old_map, old_entry); 4626 break; 4627 default: 4628 KASSERT(0); 4629 break; 4630 } 4631 old_entry = old_entry->next; 4632 } 4633 4634 pmap_update(old_map->pmap); 4635 vm_map_unlock(old_map); 4636 4637 if (uvm_shmfork && vm1->vm_shm) 4638 (*uvm_shmfork)(vm1, vm2); 4639 4640#ifdef PMAP_FORK 4641 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 4642#endif 4643 4644 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4645 return (vm2); 4646} 4647 4648 4649/* 4650 * uvm_mapent_trymerge: try to merge an entry with its neighbors. 4651 * 4652 * => called with map locked. 4653 * => return non zero if successfully merged. 4654 */ 4655 4656int 4657uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags) 4658{ 4659 struct uvm_object *uobj; 4660 struct vm_map_entry *next; 4661 struct vm_map_entry *prev; 4662 vsize_t size; 4663 int merged = 0; 4664 bool copying; 4665 int newetype; 4666 4667 if (entry->aref.ar_amap != NULL) { 4668 return 0; 4669 } 4670 if ((entry->flags & UVM_MAP_NOMERGE) != 0) { 4671 return 0; 4672 } 4673 4674 uobj = entry->object.uvm_obj; 4675 size = entry->end - entry->start; 4676 copying = (flags & UVM_MERGE_COPYING) != 0; 4677 newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype; 4678 4679 next = entry->next; 4680 if (next != &map->header && 4681 next->start == entry->end && 4682 ((copying && next->aref.ar_amap != NULL && 4683 amap_refs(next->aref.ar_amap) == 1) || 4684 (!copying && next->aref.ar_amap == NULL)) && 4685 UVM_ET_ISCOMPATIBLE(next, newetype, 4686 uobj, entry->flags, entry->protection, 4687 entry->max_protection, entry->inheritance, entry->advice, 4688 entry->wired_count) && 4689 (uobj == NULL || entry->offset + size == next->offset)) { 4690 int error; 4691 4692 if (copying) { 4693 error = amap_extend(next, size, 4694 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS); 4695 } else { 4696 error = 0; 4697 } 4698 if (error == 0) { 4699 if (uobj) { 4700 if (uobj->pgops->pgo_detach) { 4701 uobj->pgops->pgo_detach(uobj); 4702 } 4703 } 4704 4705 entry->end = next->end; 4706 clear_hints(map, next); 4707 uvm_map_entry_unlink(map, next); 4708 if (copying) { 4709 entry->aref = next->aref; 4710 entry->etype &= ~UVM_ET_NEEDSCOPY; 4711 } 4712 uvm_map_check(map, "trymerge forwardmerge"); 4713 uvm_mapent_free(next); 4714 merged++; 4715 } 4716 } 4717 4718 prev = entry->prev; 4719 if (prev != &map->header && 4720 prev->end == entry->start && 4721 ((copying && !merged && prev->aref.ar_amap != NULL && 4722 amap_refs(prev->aref.ar_amap) == 1) || 4723 (!copying && prev->aref.ar_amap == NULL)) && 4724 UVM_ET_ISCOMPATIBLE(prev, newetype, 4725 uobj, entry->flags, entry->protection, 4726 entry->max_protection, entry->inheritance, entry->advice, 4727 entry->wired_count) && 4728 (uobj == NULL || 4729 prev->offset + prev->end - prev->start == entry->offset)) { 4730 int error; 4731 4732 if (copying) { 4733 error = amap_extend(prev, size, 4734 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS); 4735 } else { 4736 error = 0; 4737 } 4738 if (error == 0) { 4739 if (uobj) { 4740 if (uobj->pgops->pgo_detach) { 4741 uobj->pgops->pgo_detach(uobj); 4742 } 4743 entry->offset = prev->offset; 4744 } 4745 4746 entry->start = prev->start; 4747 clear_hints(map, prev); 4748 uvm_map_entry_unlink(map, prev); 4749 if (copying) { 4750 entry->aref = prev->aref; 4751 entry->etype &= ~UVM_ET_NEEDSCOPY; 4752 } 4753 uvm_map_check(map, "trymerge backmerge"); 4754 uvm_mapent_free(prev); 4755 merged++; 4756 } 4757 } 4758 4759 return merged; 4760} 4761 4762/* 4763 * uvm_map_setup: init map 4764 * 4765 * => map must not be in service yet. 4766 */ 4767 4768void 4769uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags) 4770{ 4771 4772 rb_tree_init(&map->rb_tree, &uvm_map_tree_ops); 4773 map->header.next = map->header.prev = &map->header; 4774 map->nentries = 0; 4775 map->size = 0; 4776 map->ref_count = 1; 4777 vm_map_setmin(map, vmin); 4778 vm_map_setmax(map, vmax); 4779 map->flags = flags; 4780 map->first_free = &map->header; 4781 map->hint = &map->header; 4782 map->timestamp = 0; 4783 map->busy = NULL; 4784 4785 rw_init(&map->lock); 4786 cv_init(&map->cv, "vm_map"); 4787 mutex_init(&map->misc_lock, MUTEX_DRIVER, IPL_NONE); 4788} 4789 4790/* 4791 * U N M A P - m a i n e n t r y p o i n t 4792 */ 4793 4794/* 4795 * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop") 4796 * 4797 * => caller must check alignment and size 4798 * => map must be unlocked (we will lock it) 4799 * => flags is UVM_FLAG_QUANTUM or 0. 4800 */ 4801 4802void 4803uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 4804{ 4805 struct vm_map_entry *dead_entries; 4806 UVMHIST_FUNC(__func__); 4807 UVMHIST_CALLARGS(maphist, " (map=%#jx, start=%#jx, end=%#jx)", 4808 (uintptr_t)map, start, end, 0); 4809 4810 KASSERTMSG(start < end, 4811 "%s: map %p: start %#jx < end %#jx", __func__, map, 4812 (uintmax_t)start, (uintmax_t)end); 4813 if (map == kernel_map) { 4814 LOCKDEBUG_MEM_CHECK((void *)start, end - start); 4815 } 4816 4817 /* 4818 * work now done by helper functions. wipe the pmap's and then 4819 * detach from the dead entries... 4820 */ 4821 vm_map_lock(map); 4822 uvm_unmap_remove(map, start, end, &dead_entries, flags); 4823 vm_map_unlock(map); 4824 4825 if (dead_entries != NULL) 4826 uvm_unmap_detach(dead_entries, 0); 4827 4828 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 4829} 4830 4831 4832/* 4833 * uvm_map_reference: add reference to a map 4834 * 4835 * => map need not be locked 4836 */ 4837 4838void 4839uvm_map_reference(struct vm_map *map) 4840{ 4841 4842 atomic_inc_uint(&map->ref_count); 4843} 4844 4845void 4846uvm_map_lock_entry(struct vm_map_entry *entry, krw_t op) 4847{ 4848 4849 if (entry->aref.ar_amap != NULL) { 4850 amap_lock(entry->aref.ar_amap, op); 4851 } 4852 if (UVM_ET_ISOBJ(entry)) { 4853 rw_enter(entry->object.uvm_obj->vmobjlock, op); 4854 } 4855} 4856 4857void 4858uvm_map_unlock_entry(struct vm_map_entry *entry) 4859{ 4860 4861 if (UVM_ET_ISOBJ(entry)) { 4862 rw_exit(entry->object.uvm_obj->vmobjlock); 4863 } 4864 if (entry->aref.ar_amap != NULL) { 4865 amap_unlock(entry->aref.ar_amap); 4866 } 4867} 4868 4869#define UVM_VOADDR_TYPE_MASK 0x3UL 4870#define UVM_VOADDR_TYPE_UOBJ 0x1UL 4871#define UVM_VOADDR_TYPE_ANON 0x2UL 4872#define UVM_VOADDR_OBJECT_MASK ~UVM_VOADDR_TYPE_MASK 4873 4874#define UVM_VOADDR_GET_TYPE(voa) \ 4875 ((voa)->object & UVM_VOADDR_TYPE_MASK) 4876#define UVM_VOADDR_GET_OBJECT(voa) \ 4877 ((voa)->object & UVM_VOADDR_OBJECT_MASK) 4878#define UVM_VOADDR_SET_OBJECT(voa, obj, type) \ 4879do { \ 4880 KASSERT(((uintptr_t)(obj) & UVM_VOADDR_TYPE_MASK) == 0); \ 4881 (voa)->object = ((uintptr_t)(obj)) | (type); \ 4882} while (/*CONSTCOND*/0) 4883 4884#define UVM_VOADDR_GET_UOBJ(voa) \ 4885 ((struct uvm_object *)UVM_VOADDR_GET_OBJECT(voa)) 4886#define UVM_VOADDR_SET_UOBJ(voa, uobj) \ 4887 UVM_VOADDR_SET_OBJECT(voa, uobj, UVM_VOADDR_TYPE_UOBJ) 4888 4889#define UVM_VOADDR_GET_ANON(voa) \ 4890 ((struct vm_anon *)UVM_VOADDR_GET_OBJECT(voa)) 4891#define UVM_VOADDR_SET_ANON(voa, anon) \ 4892 UVM_VOADDR_SET_OBJECT(voa, anon, UVM_VOADDR_TYPE_ANON) 4893 4894/* 4895 * uvm_voaddr_acquire: returns the virtual object address corresponding 4896 * to the specified virtual address. 4897 * 4898 * => resolves COW so the true page identity is tracked. 4899 * 4900 * => acquires a reference on the page's owner (uvm_object or vm_anon) 4901 */ 4902bool 4903uvm_voaddr_acquire(struct vm_map * const map, vaddr_t const va, 4904 struct uvm_voaddr * const voaddr) 4905{ 4906 struct vm_map_entry *entry; 4907 struct vm_anon *anon = NULL; 4908 bool result = false; 4909 bool exclusive = false; 4910 void (*unlock_fn)(struct vm_map *); 4911 4912 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4913 UVMHIST_LOG(maphist,"(map=%#jx,va=%#jx)", (uintptr_t)map, va, 0, 0); 4914 4915 const vaddr_t start = trunc_page(va); 4916 const vaddr_t end = round_page(va+1); 4917 4918 lookup_again: 4919 if (__predict_false(exclusive)) { 4920 vm_map_lock(map); 4921 unlock_fn = vm_map_unlock; 4922 } else { 4923 vm_map_lock_read(map); 4924 unlock_fn = vm_map_unlock_read; 4925 } 4926 4927 if (__predict_false(!uvm_map_lookup_entry(map, start, &entry))) { 4928 unlock_fn(map); 4929 UVMHIST_LOG(maphist,"<- done (no entry)",0,0,0,0); 4930 return false; 4931 } 4932 4933 if (__predict_false(entry->protection == VM_PROT_NONE)) { 4934 unlock_fn(map); 4935 UVMHIST_LOG(maphist,"<- done (PROT_NONE)",0,0,0,0); 4936 return false; 4937 } 4938 4939 /* 4940 * We have a fast path for the common case of "no COW resolution 4941 * needed" whereby we have taken a read lock on the map and if 4942 * we don't encounter any need to create a vm_anon then great! 4943 * But if we do, we loop around again, instead taking an exclusive 4944 * lock so that we can perform the fault. 4945 * 4946 * In the event that we have to resolve the fault, we do nearly the 4947 * same work as uvm_map_pageable() does: 4948 * 4949 * 1: holding the write lock, we create any anonymous maps that need 4950 * to be created. however, we do NOT need to clip the map entries 4951 * in this case. 4952 * 4953 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 4954 * in the page (assuming the entry is not already wired). this 4955 * is done because we need the vm_anon to be present. 4956 */ 4957 if (__predict_true(!VM_MAPENT_ISWIRED(entry))) { 4958 4959 bool need_fault = false; 4960 4961 /* 4962 * perform the action of vm_map_lookup that need the 4963 * write lock on the map: create an anonymous map for 4964 * a copy-on-write region, or an anonymous map for 4965 * a zero-fill region. 4966 */ 4967 if (__predict_false(UVM_ET_ISSUBMAP(entry))) { 4968 unlock_fn(map); 4969 UVMHIST_LOG(maphist,"<- done (submap)",0,0,0,0); 4970 return false; 4971 } 4972 if (__predict_false(UVM_ET_ISNEEDSCOPY(entry) && 4973 ((entry->max_protection & VM_PROT_WRITE) || 4974 (entry->object.uvm_obj == NULL)))) { 4975 if (!exclusive) { 4976 /* need to take the slow path */ 4977 KASSERT(unlock_fn == vm_map_unlock_read); 4978 vm_map_unlock_read(map); 4979 exclusive = true; 4980 goto lookup_again; 4981 } 4982 need_fault = true; 4983 amap_copy(map, entry, 0, start, end); 4984 /* XXXCDC: wait OK? */ 4985 } 4986 4987 /* 4988 * do a quick check to see if the fault has already 4989 * been resolved to the upper layer. 4990 */ 4991 if (__predict_true(entry->aref.ar_amap != NULL && 4992 need_fault == false)) { 4993 amap_lock(entry->aref.ar_amap, RW_WRITER); 4994 anon = amap_lookup(&entry->aref, start - entry->start); 4995 if (__predict_true(anon != NULL)) { 4996 /* amap unlocked below */ 4997 goto found_anon; 4998 } 4999 amap_unlock(entry->aref.ar_amap); 5000 need_fault = true; 5001 } 5002 5003 /* 5004 * we predict this test as false because if we reach 5005 * this point, then we are likely dealing with a 5006 * shared memory region backed by a uvm_object, in 5007 * which case a fault to create the vm_anon is not 5008 * necessary. 5009 */ 5010 if (__predict_false(need_fault)) { 5011 if (exclusive) { 5012 vm_map_busy(map); 5013 vm_map_unlock(map); 5014 unlock_fn = vm_map_unbusy; 5015 } 5016 5017 if (uvm_fault_wire(map, start, end, 5018 entry->max_protection, 1)) { 5019 /* wiring failed */ 5020 unlock_fn(map); 5021 UVMHIST_LOG(maphist,"<- done (wire failed)", 5022 0,0,0,0); 5023 return false; 5024 } 5025 5026 /* 5027 * now that we have resolved the fault, we can unwire 5028 * the page. 5029 */ 5030 if (exclusive) { 5031 vm_map_lock(map); 5032 vm_map_unbusy(map); 5033 unlock_fn = vm_map_unlock; 5034 } 5035 5036 uvm_fault_unwire_locked(map, start, end); 5037 } 5038 } 5039 5040 /* check the upper layer */ 5041 if (entry->aref.ar_amap) { 5042 amap_lock(entry->aref.ar_amap, RW_WRITER); 5043 anon = amap_lookup(&entry->aref, start - entry->start); 5044 if (anon) { 5045 found_anon: KASSERT(anon->an_lock == entry->aref.ar_amap->am_lock); 5046 anon->an_ref++; 5047 rw_obj_hold(anon->an_lock); 5048 KASSERT(anon->an_ref != 0); 5049 UVM_VOADDR_SET_ANON(voaddr, anon); 5050 voaddr->offset = va & PAGE_MASK; 5051 result = true; 5052 } 5053 amap_unlock(entry->aref.ar_amap); 5054 } 5055 5056 /* check the lower layer */ 5057 if (!result && UVM_ET_ISOBJ(entry)) { 5058 struct uvm_object *uobj = entry->object.uvm_obj; 5059 5060 KASSERT(uobj != NULL); 5061 (*uobj->pgops->pgo_reference)(uobj); 5062 UVM_VOADDR_SET_UOBJ(voaddr, uobj); 5063 voaddr->offset = entry->offset + (va - entry->start); 5064 result = true; 5065 } 5066 5067 unlock_fn(map); 5068 5069 if (result) { 5070 UVMHIST_LOG(maphist, 5071 "<- done OK (type=%jd,owner=%#jx,offset=%#jx)", 5072 UVM_VOADDR_GET_TYPE(voaddr), 5073 UVM_VOADDR_GET_OBJECT(voaddr), 5074 voaddr->offset, 0); 5075 } else { 5076 UVMHIST_LOG(maphist,"<- done (failed)",0,0,0,0); 5077 } 5078 5079 return result; 5080} 5081 5082/* 5083 * uvm_voaddr_release: release the references held by the 5084 * vitual object address. 5085 */ 5086void 5087uvm_voaddr_release(struct uvm_voaddr * const voaddr) 5088{ 5089 5090 switch (UVM_VOADDR_GET_TYPE(voaddr)) { 5091 case UVM_VOADDR_TYPE_UOBJ: { 5092 struct uvm_object * const uobj = UVM_VOADDR_GET_UOBJ(voaddr); 5093 5094 KASSERT(uobj != NULL); 5095 KASSERT(uobj->pgops->pgo_detach != NULL); 5096 (*uobj->pgops->pgo_detach)(uobj); 5097 break; 5098 } 5099 case UVM_VOADDR_TYPE_ANON: { 5100 struct vm_anon * const anon = UVM_VOADDR_GET_ANON(voaddr); 5101 krwlock_t *lock; 5102 5103 KASSERT(anon != NULL); 5104 rw_enter((lock = anon->an_lock), RW_WRITER); 5105 KASSERT(anon->an_ref > 0); 5106 if (--anon->an_ref == 0) { 5107 uvm_anfree(anon); 5108 } 5109 rw_exit(lock); 5110 rw_obj_free(lock); 5111 break; 5112 } 5113 default: 5114 panic("uvm_voaddr_release: bad type"); 5115 } 5116 memset(voaddr, 0, sizeof(*voaddr)); 5117} 5118 5119/* 5120 * uvm_voaddr_compare: compare two uvm_voaddr objects. 5121 * 5122 * => memcmp() semantics 5123 */ 5124int 5125uvm_voaddr_compare(const struct uvm_voaddr * const voaddr1, 5126 const struct uvm_voaddr * const voaddr2) 5127{ 5128 const uintptr_t type1 = UVM_VOADDR_GET_TYPE(voaddr1); 5129 const uintptr_t type2 = UVM_VOADDR_GET_TYPE(voaddr2); 5130 5131 KASSERT(type1 == UVM_VOADDR_TYPE_UOBJ || 5132 type1 == UVM_VOADDR_TYPE_ANON); 5133 5134 KASSERT(type2 == UVM_VOADDR_TYPE_UOBJ || 5135 type2 == UVM_VOADDR_TYPE_ANON); 5136 5137 if (type1 < type2) 5138 return -1; 5139 if (type1 > type2) 5140 return 1; 5141 5142 const uintptr_t addr1 = UVM_VOADDR_GET_OBJECT(voaddr1); 5143 const uintptr_t addr2 = UVM_VOADDR_GET_OBJECT(voaddr2); 5144 5145 if (addr1 < addr2) 5146 return -1; 5147 if (addr1 > addr2) 5148 return 1; 5149 5150 if (voaddr1->offset < voaddr2->offset) 5151 return -1; 5152 if (voaddr1->offset > voaddr2->offset) 5153 return 1; 5154 5155 return 0; 5156} 5157 5158#if defined(DDB) || defined(DEBUGPRINT) 5159 5160/* 5161 * uvm_map_printit: actually prints the map 5162 */ 5163 5164void 5165uvm_map_printit(struct vm_map *map, bool full, 5166 void (*pr)(const char *, ...)) 5167{ 5168 struct vm_map_entry *entry; 5169 5170 (*pr)("MAP %p: [%#lx->%#lx]\n", map, vm_map_min(map), 5171 vm_map_max(map)); 5172 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=%#x\n", 5173 map->nentries, map->size, map->ref_count, map->timestamp, 5174 map->flags); 5175 (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap, 5176 pmap_resident_count(map->pmap), pmap_wired_count(map->pmap)); 5177 if (!full) 5178 return; 5179 for (entry = map->header.next; entry != &map->header; 5180 entry = entry->next) { 5181 (*pr)(" - %p: %#lx->%#lx: obj=%p/%#llx, amap=%p/%d\n", 5182 entry, entry->start, entry->end, entry->object.uvm_obj, 5183 (long long)entry->offset, entry->aref.ar_amap, 5184 entry->aref.ar_pageoff); 5185 (*pr)( 5186 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 5187 "wc=%d, adv=%d%s\n", 5188 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 5189 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 5190 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 5191 entry->protection, entry->max_protection, 5192 entry->inheritance, entry->wired_count, entry->advice, 5193 entry == map->first_free ? " (first_free)" : ""); 5194 } 5195} 5196 5197void 5198uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...)) 5199{ 5200 struct vm_map *map; 5201 5202 for (map = kernel_map;;) { 5203 struct vm_map_entry *entry; 5204 5205 if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) { 5206 break; 5207 } 5208 (*pr)("%p is %p+%zu from VMMAP %p\n", 5209 (void *)addr, (void *)entry->start, 5210 (size_t)(addr - (uintptr_t)entry->start), map); 5211 if (!UVM_ET_ISSUBMAP(entry)) { 5212 break; 5213 } 5214 map = entry->object.sub_map; 5215 } 5216} 5217 5218#endif /* DDB || DEBUGPRINT */ 5219 5220#ifndef __USER_VA0_IS_SAFE 5221static int 5222sysctl_user_va0_disable(SYSCTLFN_ARGS) 5223{ 5224 struct sysctlnode node; 5225 int t, error; 5226 5227 node = *rnode; 5228 node.sysctl_data = &t; 5229 t = user_va0_disable; 5230 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 5231 if (error || newp == NULL) 5232 return (error); 5233 5234 if (!t && user_va0_disable && 5235 kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MAP_VA_ZERO, 0, 5236 NULL, NULL, NULL)) 5237 return EPERM; 5238 5239 user_va0_disable = !!t; 5240 return 0; 5241} 5242#endif 5243 5244static int 5245fill_vmentry(struct lwp *l, struct proc *p, struct kinfo_vmentry *kve, 5246 struct vm_map *m, struct vm_map_entry *e) 5247{ 5248#ifndef _RUMPKERNEL 5249 int error; 5250 5251 memset(kve, 0, sizeof(*kve)); 5252 KASSERT(e != NULL); 5253 if (UVM_ET_ISOBJ(e)) { 5254 struct uvm_object *uobj = e->object.uvm_obj; 5255 KASSERT(uobj != NULL); 5256 kve->kve_ref_count = uobj->uo_refs; 5257 kve->kve_count = uobj->uo_npages; 5258 if (UVM_OBJ_IS_VNODE(uobj)) { 5259 struct vattr va; 5260 struct vnode *vp = (struct vnode *)uobj; 5261 vn_lock(vp, LK_SHARED | LK_RETRY); 5262 error = VOP_GETATTR(vp, &va, l->l_cred); 5263 VOP_UNLOCK(vp); 5264 kve->kve_type = KVME_TYPE_VNODE; 5265 if (error == 0) { 5266 kve->kve_vn_size = vp->v_size; 5267 kve->kve_vn_type = (int)vp->v_type; 5268 kve->kve_vn_mode = va.va_mode; 5269 kve->kve_vn_rdev = va.va_rdev; 5270 kve->kve_vn_fileid = va.va_fileid; 5271 kve->kve_vn_fsid = va.va_fsid; 5272 error = vnode_to_path(kve->kve_path, 5273 sizeof(kve->kve_path) / 2, vp, l, p); 5274 } 5275 } else if (UVM_OBJ_IS_KERN_OBJECT(uobj)) { 5276 kve->kve_type = KVME_TYPE_KERN; 5277 } else if (UVM_OBJ_IS_DEVICE(uobj)) { 5278 kve->kve_type = KVME_TYPE_DEVICE; 5279 } else if (UVM_OBJ_IS_AOBJ(uobj)) { 5280 kve->kve_type = KVME_TYPE_ANON; 5281 } else { 5282 kve->kve_type = KVME_TYPE_OBJECT; 5283 } 5284 } else if (UVM_ET_ISSUBMAP(e)) { 5285 struct vm_map *map = e->object.sub_map; 5286 KASSERT(map != NULL); 5287 kve->kve_ref_count = map->ref_count; 5288 kve->kve_count = map->nentries; 5289 kve->kve_type = KVME_TYPE_SUBMAP; 5290 } else 5291 kve->kve_type = KVME_TYPE_UNKNOWN; 5292 5293 kve->kve_start = e->start; 5294 kve->kve_end = e->end; 5295 kve->kve_offset = e->offset; 5296 kve->kve_wired_count = e->wired_count; 5297 kve->kve_inheritance = e->inheritance; 5298 kve->kve_attributes = 0; /* unused */ 5299 kve->kve_advice = e->advice; 5300#define PROT(p) (((p) & VM_PROT_READ) ? KVME_PROT_READ : 0) | \ 5301 (((p) & VM_PROT_WRITE) ? KVME_PROT_WRITE : 0) | \ 5302 (((p) & VM_PROT_EXECUTE) ? KVME_PROT_EXEC : 0) 5303 kve->kve_protection = PROT(e->protection); 5304 kve->kve_max_protection = PROT(e->max_protection); 5305 kve->kve_flags |= (e->etype & UVM_ET_COPYONWRITE) 5306 ? KVME_FLAG_COW : 0; 5307 kve->kve_flags |= (e->etype & UVM_ET_NEEDSCOPY) 5308 ? KVME_FLAG_NEEDS_COPY : 0; 5309 kve->kve_flags |= (m->flags & VM_MAP_TOPDOWN) 5310 ? KVME_FLAG_GROWS_DOWN : KVME_FLAG_GROWS_UP; 5311 kve->kve_flags |= (m->flags & VM_MAP_PAGEABLE) 5312 ? KVME_FLAG_PAGEABLE : 0; 5313#endif 5314 return 0; 5315} 5316 5317static int 5318fill_vmentries(struct lwp *l, pid_t pid, u_int elem_size, void *oldp, 5319 size_t *oldlenp) 5320{ 5321 int error; 5322 struct proc *p; 5323 struct kinfo_vmentry *vme; 5324 struct vmspace *vm; 5325 struct vm_map *map; 5326 struct vm_map_entry *entry; 5327 char *dp; 5328 size_t count, vmesize; 5329 5330 if (elem_size == 0 || elem_size > 2 * sizeof(*vme)) 5331 return EINVAL; 5332 5333 if (oldp) { 5334 if (*oldlenp > 10UL * 1024UL * 1024UL) 5335 return E2BIG; 5336 count = *oldlenp / elem_size; 5337 if (count == 0) 5338 return ENOMEM; 5339 vmesize = count * sizeof(*vme); 5340 } else 5341 vmesize = 0; 5342 5343 if ((error = proc_find_locked(l, &p, pid)) != 0) 5344 return error; 5345 5346 vme = NULL; 5347 count = 0; 5348 5349 if ((error = proc_vmspace_getref(p, &vm)) != 0) 5350 goto out; 5351 5352 map = &vm->vm_map; 5353 vm_map_lock_read(map); 5354 5355 dp = oldp; 5356 if (oldp) 5357 vme = kmem_alloc(vmesize, KM_SLEEP); 5358 for (entry = map->header.next; entry != &map->header; 5359 entry = entry->next) { 5360 if (oldp && (dp - (char *)oldp) < vmesize) { 5361 error = fill_vmentry(l, p, &vme[count], map, entry); 5362 if (error) 5363 goto out; 5364 dp += elem_size; 5365 } 5366 count++; 5367 } 5368 vm_map_unlock_read(map); 5369 uvmspace_free(vm); 5370 5371out: 5372 if (pid != -1) 5373 mutex_exit(p->p_lock); 5374 if (error == 0) { 5375 const u_int esize = uimin(sizeof(*vme), elem_size); 5376 dp = oldp; 5377 for (size_t i = 0; i < count; i++) { 5378 if (oldp && (dp - (char *)oldp) < vmesize) { 5379 error = sysctl_copyout(l, &vme[i], dp, esize); 5380 if (error) 5381 break; 5382 dp += elem_size; 5383 } else 5384 break; 5385 } 5386 count *= elem_size; 5387 if (oldp != NULL && *oldlenp < count) 5388 error = ENOSPC; 5389 *oldlenp = count; 5390 } 5391 if (vme) 5392 kmem_free(vme, vmesize); 5393 return error; 5394} 5395 5396static int 5397sysctl_vmproc(SYSCTLFN_ARGS) 5398{ 5399 int error; 5400 5401 if (namelen == 1 && name[0] == CTL_QUERY) 5402 return (sysctl_query(SYSCTLFN_CALL(rnode))); 5403 5404 if (namelen == 0) 5405 return EINVAL; 5406 5407 switch (name[0]) { 5408 case VM_PROC_MAP: 5409 if (namelen != 3) 5410 return EINVAL; 5411 sysctl_unlock(); 5412 error = fill_vmentries(l, name[1], name[2], oldp, oldlenp); 5413 sysctl_relock(); 5414 return error; 5415 default: 5416 return EINVAL; 5417 } 5418} 5419 5420SYSCTL_SETUP(sysctl_uvmmap_setup, "sysctl uvmmap setup") 5421{ 5422 5423 sysctl_createv(clog, 0, NULL, NULL, 5424 CTLFLAG_PERMANENT, 5425 CTLTYPE_STRUCT, "proc", 5426 SYSCTL_DESCR("Process vm information"), 5427 sysctl_vmproc, 0, NULL, 0, 5428 CTL_VM, VM_PROC, CTL_EOL); 5429#ifndef __USER_VA0_IS_SAFE 5430 sysctl_createv(clog, 0, NULL, NULL, 5431 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 5432 CTLTYPE_INT, "user_va0_disable", 5433 SYSCTL_DESCR("Disable VA 0"), 5434 sysctl_user_va0_disable, 0, &user_va0_disable, 0, 5435 CTL_VM, CTL_CREATE, CTL_EOL); 5436#endif 5437} 5438