1 /* $NetBSD: uvm_map.c,v 1.404 2023/02/27 16:24:45 riastradh 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 1.404 2023/02/27 16:24:45 riastradh 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 108 static struct kern_history_ent maphistbuf[UVMHIST_MAPHIST_SIZE]; 109 UVMHIST_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) \ 122 struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \ 123 "uvmmap", #name); \ 124 EVCNT_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) */ 129 130 UVMMAP_EVCNT_DEFINE(ubackmerge) 131 UVMMAP_EVCNT_DEFINE(uforwmerge) 132 UVMMAP_EVCNT_DEFINE(ubimerge) 133 UVMMAP_EVCNT_DEFINE(unomerge) 134 UVMMAP_EVCNT_DEFINE(kbackmerge) 135 UVMMAP_EVCNT_DEFINE(kforwmerge) 136 UVMMAP_EVCNT_DEFINE(kbimerge) 137 UVMMAP_EVCNT_DEFINE(knomerge) 138 UVMMAP_EVCNT_DEFINE(map_call) 139 UVMMAP_EVCNT_DEFINE(mlk_call) 140 UVMMAP_EVCNT_DEFINE(mlk_hint) 141 UVMMAP_EVCNT_DEFINE(mlk_tree) 142 UVMMAP_EVCNT_DEFINE(mlk_treeloop) 143 144 const char vmmapbsy[] = "vmmapbsy"; 145 146 /* 147 * cache for vmspace structures. 148 */ 149 150 static struct pool_cache uvm_vmspace_cache; 151 152 /* 153 * cache for dynamically-allocated map entries. 154 */ 155 156 static struct pool_cache uvm_map_entry_cache; 157 158 #ifdef PMAP_GROWKERNEL 159 /* 160 * This global represents the end of the kernel virtual address 161 * space. If we want to exceed this, we must grow the kernel 162 * virtual address space dynamically. 163 * 164 * Note, this variable is locked by kernel_map's lock. 165 */ 166 vaddr_t uvm_maxkaddr; 167 #endif 168 169 #ifndef __USER_VA0_IS_SAFE 170 #ifndef __USER_VA0_DISABLE_DEFAULT 171 #define __USER_VA0_DISABLE_DEFAULT 1 172 #endif 173 #ifdef USER_VA0_DISABLE_DEFAULT /* kernel config option overrides */ 174 #undef __USER_VA0_DISABLE_DEFAULT 175 #define __USER_VA0_DISABLE_DEFAULT USER_VA0_DISABLE_DEFAULT 176 #endif 177 int user_va0_disable = __USER_VA0_DISABLE_DEFAULT; 178 #endif 179 180 /* 181 * macros 182 */ 183 184 /* 185 * uvm_map_align_va: round down or up virtual address 186 */ 187 static __inline void 188 uvm_map_align_va(vaddr_t *vap, vsize_t align, int topdown) 189 { 190 191 KASSERT(powerof2(align)); 192 193 if (align != 0 && (*vap & (align - 1)) != 0) { 194 if (topdown) 195 *vap = rounddown2(*vap, align); 196 else 197 *vap = roundup2(*vap, align); 198 } 199 } 200 201 /* 202 * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging 203 */ 204 extern struct vm_map *pager_map; 205 206 #define UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \ 207 prot, maxprot, inh, adv, wire) \ 208 ((ent)->etype == (type) && \ 209 (((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE)) == 0 && \ 210 (ent)->object.uvm_obj == (uobj) && \ 211 (ent)->protection == (prot) && \ 212 (ent)->max_protection == (maxprot) && \ 213 (ent)->inheritance == (inh) && \ 214 (ent)->advice == (adv) && \ 215 (ent)->wired_count == (wire)) 216 217 /* 218 * uvm_map_entry_link: insert entry into a map 219 * 220 * => map must be locked 221 */ 222 #define uvm_map_entry_link(map, after_where, entry) do { \ 223 uvm_mapent_check(entry); \ 224 (map)->nentries++; \ 225 (entry)->prev = (after_where); \ 226 (entry)->next = (after_where)->next; \ 227 (entry)->prev->next = (entry); \ 228 (entry)->next->prev = (entry); \ 229 uvm_rb_insert((map), (entry)); \ 230 } while (/*CONSTCOND*/ 0) 231 232 /* 233 * uvm_map_entry_unlink: remove entry from a map 234 * 235 * => map must be locked 236 */ 237 #define uvm_map_entry_unlink(map, entry) do { \ 238 KASSERT((entry) != (map)->first_free); \ 239 KASSERT((entry) != (map)->hint); \ 240 uvm_mapent_check(entry); \ 241 (map)->nentries--; \ 242 (entry)->next->prev = (entry)->prev; \ 243 (entry)->prev->next = (entry)->next; \ 244 uvm_rb_remove((map), (entry)); \ 245 } while (/*CONSTCOND*/ 0) 246 247 /* 248 * SAVE_HINT: saves the specified entry as the hint for future lookups. 249 * 250 * => map need not be locked. 251 */ 252 #define SAVE_HINT(map, check, value) do { \ 253 if ((map)->hint == (check)) \ 254 (map)->hint = (value); \ 255 } while (/*CONSTCOND*/ 0) 256 257 /* 258 * clear_hints: ensure that hints don't point to the entry. 259 * 260 * => map must be write-locked. 261 */ 262 static void 263 clear_hints(struct vm_map *map, struct vm_map_entry *ent) 264 { 265 266 SAVE_HINT(map, ent, ent->prev); 267 if (map->first_free == ent) { 268 map->first_free = ent->prev; 269 } 270 } 271 272 /* 273 * VM_MAP_RANGE_CHECK: check and correct range 274 * 275 * => map must at least be read locked 276 */ 277 278 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 279 if (start < vm_map_min(map)) \ 280 start = vm_map_min(map); \ 281 if (end > vm_map_max(map)) \ 282 end = vm_map_max(map); \ 283 if (start > end) \ 284 start = end; \ 285 } while (/*CONSTCOND*/ 0) 286 287 /* 288 * local prototypes 289 */ 290 291 static struct vm_map_entry * 292 uvm_mapent_alloc(struct vm_map *, int); 293 static void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *); 294 static void uvm_mapent_free(struct vm_map_entry *); 295 #if defined(DEBUG) 296 static void _uvm_mapent_check(const struct vm_map_entry *, int); 297 #define uvm_mapent_check(map) _uvm_mapent_check(map, __LINE__) 298 #else /* defined(DEBUG) */ 299 #define uvm_mapent_check(e) /* nothing */ 300 #endif /* defined(DEBUG) */ 301 302 static void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *); 303 static void uvm_map_reference_amap(struct vm_map_entry *, int); 304 static int uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int, 305 int, struct vm_map_entry *); 306 static void uvm_map_unreference_amap(struct vm_map_entry *, int); 307 308 int _uvm_map_sanity(struct vm_map *); 309 int _uvm_tree_sanity(struct vm_map *); 310 static vsize_t uvm_rb_maxgap(const struct vm_map_entry *); 311 312 #define ROOT_ENTRY(map) ((struct vm_map_entry *)(map)->rb_tree.rbt_root) 313 #define LEFT_ENTRY(entry) ((struct vm_map_entry *)(entry)->rb_node.rb_left) 314 #define RIGHT_ENTRY(entry) ((struct vm_map_entry *)(entry)->rb_node.rb_right) 315 #define PARENT_ENTRY(map, entry) \ 316 (ROOT_ENTRY(map) == (entry) \ 317 ? NULL : (struct vm_map_entry *)RB_FATHER(&(entry)->rb_node)) 318 319 /* 320 * These get filled in if/when SYSVSHM shared memory code is loaded 321 * 322 * We do this with function pointers rather the #ifdef SYSVSHM so the 323 * SYSVSHM code can be loaded and unloaded 324 */ 325 void (*uvm_shmexit)(struct vmspace *) = NULL; 326 void (*uvm_shmfork)(struct vmspace *, struct vmspace *) = NULL; 327 328 static int 329 uvm_map_compare_nodes(void *ctx, const void *nparent, const void *nkey) 330 { 331 const struct vm_map_entry *eparent = nparent; 332 const struct vm_map_entry *ekey = nkey; 333 334 KASSERT(eparent->start < ekey->start || eparent->start >= ekey->end); 335 KASSERT(ekey->start < eparent->start || ekey->start >= eparent->end); 336 337 if (eparent->start < ekey->start) 338 return -1; 339 if (eparent->end >= ekey->start) 340 return 1; 341 return 0; 342 } 343 344 static int 345 uvm_map_compare_key(void *ctx, const void *nparent, const void *vkey) 346 { 347 const struct vm_map_entry *eparent = nparent; 348 const vaddr_t va = *(const vaddr_t *) vkey; 349 350 if (eparent->start < va) 351 return -1; 352 if (eparent->end >= va) 353 return 1; 354 return 0; 355 } 356 357 static const rb_tree_ops_t uvm_map_tree_ops = { 358 .rbto_compare_nodes = uvm_map_compare_nodes, 359 .rbto_compare_key = uvm_map_compare_key, 360 .rbto_node_offset = offsetof(struct vm_map_entry, rb_node), 361 .rbto_context = NULL 362 }; 363 364 /* 365 * uvm_rb_gap: return the gap size between our entry and next entry. 366 */ 367 static inline vsize_t 368 uvm_rb_gap(const struct vm_map_entry *entry) 369 { 370 371 KASSERT(entry->next != NULL); 372 return entry->next->start - entry->end; 373 } 374 375 static vsize_t 376 uvm_rb_maxgap(const struct vm_map_entry *entry) 377 { 378 struct vm_map_entry *child; 379 vsize_t maxgap = entry->gap; 380 381 /* 382 * We need maxgap to be the largest gap of us or any of our 383 * descendents. Since each of our children's maxgap is the 384 * cached value of their largest gap of themselves or their 385 * descendents, we can just use that value and avoid recursing 386 * down the tree to calculate it. 387 */ 388 if ((child = LEFT_ENTRY(entry)) != NULL && maxgap < child->maxgap) 389 maxgap = child->maxgap; 390 391 if ((child = RIGHT_ENTRY(entry)) != NULL && maxgap < child->maxgap) 392 maxgap = child->maxgap; 393 394 return maxgap; 395 } 396 397 static void 398 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry) 399 { 400 struct vm_map_entry *parent; 401 402 KASSERT(entry->gap == uvm_rb_gap(entry)); 403 entry->maxgap = uvm_rb_maxgap(entry); 404 405 while ((parent = PARENT_ENTRY(map, entry)) != NULL) { 406 struct vm_map_entry *brother; 407 vsize_t maxgap = parent->gap; 408 unsigned int which; 409 410 KDASSERT(parent->gap == uvm_rb_gap(parent)); 411 if (maxgap < entry->maxgap) 412 maxgap = entry->maxgap; 413 /* 414 * Since we work towards the root, we know entry's maxgap 415 * value is OK, but its brothers may now be out-of-date due 416 * to rebalancing. So refresh it. 417 */ 418 which = RB_POSITION(&entry->rb_node) ^ RB_DIR_OTHER; 419 brother = (struct vm_map_entry *)parent->rb_node.rb_nodes[which]; 420 if (brother != NULL) { 421 KDASSERT(brother->gap == uvm_rb_gap(brother)); 422 brother->maxgap = uvm_rb_maxgap(brother); 423 if (maxgap < brother->maxgap) 424 maxgap = brother->maxgap; 425 } 426 427 parent->maxgap = maxgap; 428 entry = parent; 429 } 430 } 431 432 static void 433 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry) 434 { 435 struct vm_map_entry *ret __diagused; 436 437 entry->gap = entry->maxgap = uvm_rb_gap(entry); 438 if (entry->prev != &map->header) 439 entry->prev->gap = uvm_rb_gap(entry->prev); 440 441 ret = rb_tree_insert_node(&map->rb_tree, entry); 442 KASSERTMSG(ret == entry, 443 "uvm_rb_insert: map %p: duplicate entry %p", map, ret); 444 445 /* 446 * If the previous entry is not our immediate left child, then it's an 447 * ancestor and will be fixed up on the way to the root. We don't 448 * have to check entry->prev against &map->header since &map->header 449 * will never be in the tree. 450 */ 451 uvm_rb_fixup(map, 452 LEFT_ENTRY(entry) == entry->prev ? entry->prev : entry); 453 } 454 455 static void 456 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry) 457 { 458 struct vm_map_entry *prev_parent = NULL, *next_parent = NULL; 459 460 /* 461 * If we are removing an interior node, then an adjacent node will 462 * be used to replace its position in the tree. Therefore we will 463 * need to fixup the tree starting at the parent of the replacement 464 * node. So record their parents for later use. 465 */ 466 if (entry->prev != &map->header) 467 prev_parent = PARENT_ENTRY(map, entry->prev); 468 if (entry->next != &map->header) 469 next_parent = PARENT_ENTRY(map, entry->next); 470 471 rb_tree_remove_node(&map->rb_tree, entry); 472 473 /* 474 * If the previous node has a new parent, fixup the tree starting 475 * at the previous node's old parent. 476 */ 477 if (entry->prev != &map->header) { 478 /* 479 * Update the previous entry's gap due to our absence. 480 */ 481 entry->prev->gap = uvm_rb_gap(entry->prev); 482 uvm_rb_fixup(map, entry->prev); 483 if (prev_parent != NULL 484 && prev_parent != entry 485 && prev_parent != PARENT_ENTRY(map, entry->prev)) 486 uvm_rb_fixup(map, prev_parent); 487 } 488 489 /* 490 * If the next node has a new parent, fixup the tree starting 491 * at the next node's old parent. 492 */ 493 if (entry->next != &map->header) { 494 uvm_rb_fixup(map, entry->next); 495 if (next_parent != NULL 496 && next_parent != entry 497 && next_parent != PARENT_ENTRY(map, entry->next)) 498 uvm_rb_fixup(map, next_parent); 499 } 500 } 501 502 #if defined(DEBUG) 503 int uvm_debug_check_map = 0; 504 int uvm_debug_check_rbtree = 0; 505 #define uvm_map_check(map, name) \ 506 _uvm_map_check((map), (name), __FILE__, __LINE__) 507 static void 508 _uvm_map_check(struct vm_map *map, const char *name, 509 const char *file, int line) 510 { 511 512 if ((uvm_debug_check_map && _uvm_map_sanity(map)) || 513 (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) { 514 panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)", 515 name, map, file, line); 516 } 517 } 518 #else /* defined(DEBUG) */ 519 #define uvm_map_check(map, name) /* nothing */ 520 #endif /* defined(DEBUG) */ 521 522 #if defined(DEBUG) || defined(DDB) 523 int 524 _uvm_map_sanity(struct vm_map *map) 525 { 526 bool first_free_found = false; 527 bool hint_found = false; 528 const struct vm_map_entry *e; 529 struct vm_map_entry *hint = map->hint; 530 531 e = &map->header; 532 for (;;) { 533 if (map->first_free == e) { 534 first_free_found = true; 535 } else if (!first_free_found && e->next->start > e->end) { 536 printf("first_free %p should be %p\n", 537 map->first_free, e); 538 return -1; 539 } 540 if (hint == e) { 541 hint_found = true; 542 } 543 544 e = e->next; 545 if (e == &map->header) { 546 break; 547 } 548 } 549 if (!first_free_found) { 550 printf("stale first_free\n"); 551 return -1; 552 } 553 if (!hint_found) { 554 printf("stale hint\n"); 555 return -1; 556 } 557 return 0; 558 } 559 560 int 561 _uvm_tree_sanity(struct vm_map *map) 562 { 563 struct vm_map_entry *tmp, *trtmp; 564 int n = 0, i = 1; 565 566 for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) { 567 if (tmp->gap != uvm_rb_gap(tmp)) { 568 printf("%d/%d gap %#lx != %#lx %s\n", 569 n + 1, map->nentries, 570 (ulong)tmp->gap, (ulong)uvm_rb_gap(tmp), 571 tmp->next == &map->header ? "(last)" : ""); 572 goto error; 573 } 574 /* 575 * If any entries are out of order, tmp->gap will be unsigned 576 * and will likely exceed the size of the map. 577 */ 578 if (tmp->gap >= vm_map_max(map) - vm_map_min(map)) { 579 printf("too large gap %zu\n", (size_t)tmp->gap); 580 goto error; 581 } 582 n++; 583 } 584 585 if (n != map->nentries) { 586 printf("nentries: %d vs %d\n", n, map->nentries); 587 goto error; 588 } 589 590 trtmp = NULL; 591 for (tmp = map->header.next; tmp != &map->header; tmp = tmp->next) { 592 if (tmp->maxgap != uvm_rb_maxgap(tmp)) { 593 printf("maxgap %#lx != %#lx\n", 594 (ulong)tmp->maxgap, 595 (ulong)uvm_rb_maxgap(tmp)); 596 goto error; 597 } 598 if (trtmp != NULL && trtmp->start >= tmp->start) { 599 printf("corrupt: 0x%"PRIxVADDR"x >= 0x%"PRIxVADDR"x\n", 600 trtmp->start, tmp->start); 601 goto error; 602 } 603 604 trtmp = tmp; 605 } 606 607 for (tmp = map->header.next; tmp != &map->header; 608 tmp = tmp->next, i++) { 609 trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_LEFT); 610 if (trtmp == NULL) 611 trtmp = &map->header; 612 if (tmp->prev != trtmp) { 613 printf("lookup: %d: %p->prev=%p: %p\n", 614 i, tmp, tmp->prev, trtmp); 615 goto error; 616 } 617 trtmp = rb_tree_iterate(&map->rb_tree, tmp, RB_DIR_RIGHT); 618 if (trtmp == NULL) 619 trtmp = &map->header; 620 if (tmp->next != trtmp) { 621 printf("lookup: %d: %p->next=%p: %p\n", 622 i, tmp, tmp->next, trtmp); 623 goto error; 624 } 625 trtmp = rb_tree_find_node(&map->rb_tree, &tmp->start); 626 if (trtmp != tmp) { 627 printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp, 628 PARENT_ENTRY(map, tmp)); 629 goto error; 630 } 631 } 632 633 return (0); 634 error: 635 return (-1); 636 } 637 #endif /* defined(DEBUG) || defined(DDB) */ 638 639 /* 640 * vm_map_lock: acquire an exclusive (write) lock on a map. 641 * 642 * => The locking protocol provides for guaranteed upgrade from shared -> 643 * exclusive by whichever thread currently has the map marked busy. 644 * See "LOCKING PROTOCOL NOTES" in uvm_map.h. This is horrible; among 645 * other problems, it defeats any fairness guarantees provided by RW 646 * locks. 647 */ 648 649 void 650 vm_map_lock(struct vm_map *map) 651 { 652 653 for (;;) { 654 rw_enter(&map->lock, RW_WRITER); 655 if (map->busy == NULL || map->busy == curlwp) { 656 break; 657 } 658 mutex_enter(&map->misc_lock); 659 rw_exit(&map->lock); 660 if (map->busy != NULL) { 661 cv_wait(&map->cv, &map->misc_lock); 662 } 663 mutex_exit(&map->misc_lock); 664 } 665 map->timestamp++; 666 } 667 668 /* 669 * vm_map_lock_try: try to lock a map, failing if it is already locked. 670 */ 671 672 bool 673 vm_map_lock_try(struct vm_map *map) 674 { 675 676 if (!rw_tryenter(&map->lock, RW_WRITER)) { 677 return false; 678 } 679 if (map->busy != NULL) { 680 rw_exit(&map->lock); 681 return false; 682 } 683 map->timestamp++; 684 return true; 685 } 686 687 /* 688 * vm_map_unlock: release an exclusive lock on a map. 689 */ 690 691 void 692 vm_map_unlock(struct vm_map *map) 693 { 694 695 KASSERT(rw_write_held(&map->lock)); 696 KASSERT(map->busy == NULL || map->busy == curlwp); 697 rw_exit(&map->lock); 698 } 699 700 /* 701 * vm_map_unbusy: mark the map as unbusy, and wake any waiters that 702 * want an exclusive lock. 703 */ 704 705 void 706 vm_map_unbusy(struct vm_map *map) 707 { 708 709 KASSERT(map->busy == curlwp); 710 711 /* 712 * Safe to clear 'busy' and 'waiters' with only a read lock held: 713 * 714 * o they can only be set with a write lock held 715 * o writers are blocked out with a read or write hold 716 * o at any time, only one thread owns the set of values 717 */ 718 mutex_enter(&map->misc_lock); 719 map->busy = NULL; 720 cv_broadcast(&map->cv); 721 mutex_exit(&map->misc_lock); 722 } 723 724 /* 725 * vm_map_lock_read: acquire a shared (read) lock on a map. 726 */ 727 728 void 729 vm_map_lock_read(struct vm_map *map) 730 { 731 732 rw_enter(&map->lock, RW_READER); 733 } 734 735 /* 736 * vm_map_unlock_read: release a shared lock on a map. 737 */ 738 739 void 740 vm_map_unlock_read(struct vm_map *map) 741 { 742 743 rw_exit(&map->lock); 744 } 745 746 /* 747 * vm_map_busy: mark a map as busy. 748 * 749 * => the caller must hold the map write locked 750 */ 751 752 void 753 vm_map_busy(struct vm_map *map) 754 { 755 756 KASSERT(rw_write_held(&map->lock)); 757 KASSERT(map->busy == NULL); 758 759 map->busy = curlwp; 760 } 761 762 /* 763 * vm_map_locked_p: return true if the map is write locked. 764 * 765 * => only for debug purposes like KASSERTs. 766 * => should not be used to verify that a map is not locked. 767 */ 768 769 bool 770 vm_map_locked_p(struct vm_map *map) 771 { 772 773 return rw_write_held(&map->lock); 774 } 775 776 /* 777 * uvm_mapent_alloc: allocate a map entry 778 */ 779 780 static struct vm_map_entry * 781 uvm_mapent_alloc(struct vm_map *map, int flags) 782 { 783 struct vm_map_entry *me; 784 int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK; 785 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 786 787 me = pool_cache_get(&uvm_map_entry_cache, pflags); 788 if (__predict_false(me == NULL)) { 789 return NULL; 790 } 791 me->flags = 0; 792 793 UVMHIST_LOG(maphist, "<- new entry=%#jx [kentry=%jd]", (uintptr_t)me, 794 (map == kernel_map), 0, 0); 795 return me; 796 } 797 798 /* 799 * uvm_mapent_free: free map entry 800 */ 801 802 static void 803 uvm_mapent_free(struct vm_map_entry *me) 804 { 805 UVMHIST_FUNC(__func__); 806 UVMHIST_CALLARGS(maphist,"<- freeing map entry=%#jx [flags=%#jx]", 807 (uintptr_t)me, me->flags, 0, 0); 808 pool_cache_put(&uvm_map_entry_cache, me); 809 } 810 811 /* 812 * uvm_mapent_copy: copy a map entry, preserving flags 813 */ 814 815 static inline void 816 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 817 { 818 819 memcpy(dst, src, sizeof(*dst)); 820 dst->flags = 0; 821 } 822 823 #if defined(DEBUG) 824 static void 825 _uvm_mapent_check(const struct vm_map_entry *entry, int line) 826 { 827 828 if (entry->start >= entry->end) { 829 goto bad; 830 } 831 if (UVM_ET_ISOBJ(entry)) { 832 if (entry->object.uvm_obj == NULL) { 833 goto bad; 834 } 835 } else if (UVM_ET_ISSUBMAP(entry)) { 836 if (entry->object.sub_map == NULL) { 837 goto bad; 838 } 839 } else { 840 if (entry->object.uvm_obj != NULL || 841 entry->object.sub_map != NULL) { 842 goto bad; 843 } 844 } 845 if (!UVM_ET_ISOBJ(entry)) { 846 if (entry->offset != 0) { 847 goto bad; 848 } 849 } 850 851 return; 852 853 bad: 854 panic("%s: bad entry %p, line %d", __func__, entry, line); 855 } 856 #endif /* defined(DEBUG) */ 857 858 /* 859 * uvm_map_entry_unwire: unwire a map entry 860 * 861 * => map should be locked by caller 862 */ 863 864 static inline void 865 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry) 866 { 867 868 entry->wired_count = 0; 869 uvm_fault_unwire_locked(map, entry->start, entry->end); 870 } 871 872 873 /* 874 * wrapper for calling amap_ref() 875 */ 876 static inline void 877 uvm_map_reference_amap(struct vm_map_entry *entry, int flags) 878 { 879 880 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 881 (entry->end - entry->start) >> PAGE_SHIFT, flags); 882 } 883 884 885 /* 886 * wrapper for calling amap_unref() 887 */ 888 static inline void 889 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags) 890 { 891 892 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 893 (entry->end - entry->start) >> PAGE_SHIFT, flags); 894 } 895 896 897 /* 898 * uvm_map_init: init mapping system at boot time. 899 */ 900 901 void 902 uvm_map_init(void) 903 { 904 /* 905 * first, init logging system. 906 */ 907 908 UVMHIST_FUNC(__func__); 909 UVMHIST_LINK_STATIC(maphist); 910 UVMHIST_LINK_STATIC(pdhist); 911 UVMHIST_CALLED(maphist); 912 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 913 914 /* 915 * initialize the global lock for kernel map entry. 916 */ 917 918 mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM); 919 } 920 921 /* 922 * uvm_map_init_caches: init mapping system caches. 923 */ 924 void 925 uvm_map_init_caches(void) 926 { 927 /* 928 * initialize caches. 929 */ 930 931 pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry), 932 coherency_unit, 0, PR_LARGECACHE, "vmmpepl", NULL, IPL_NONE, NULL, 933 NULL, NULL); 934 pool_cache_bootstrap(&uvm_vmspace_cache, sizeof(struct vmspace), 935 0, 0, 0, "vmsppl", NULL, IPL_NONE, NULL, NULL, NULL); 936 } 937 938 /* 939 * clippers 940 */ 941 942 /* 943 * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy. 944 */ 945 946 static void 947 uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2, 948 vaddr_t splitat) 949 { 950 vaddr_t adj; 951 952 KASSERT(entry1->start < splitat); 953 KASSERT(splitat < entry1->end); 954 955 adj = splitat - entry1->start; 956 entry1->end = entry2->start = splitat; 957 958 if (entry1->aref.ar_amap) { 959 amap_splitref(&entry1->aref, &entry2->aref, adj); 960 } 961 if (UVM_ET_ISSUBMAP(entry1)) { 962 /* ... unlikely to happen, but play it safe */ 963 uvm_map_reference(entry1->object.sub_map); 964 } else if (UVM_ET_ISOBJ(entry1)) { 965 KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */ 966 entry2->offset += adj; 967 if (entry1->object.uvm_obj->pgops && 968 entry1->object.uvm_obj->pgops->pgo_reference) 969 entry1->object.uvm_obj->pgops->pgo_reference( 970 entry1->object.uvm_obj); 971 } 972 } 973 974 /* 975 * uvm_map_clip_start: ensure that the entry begins at or after 976 * the starting address, if it doesn't we split the entry. 977 * 978 * => caller should use UVM_MAP_CLIP_START macro rather than calling 979 * this directly 980 * => map must be locked by caller 981 */ 982 983 void 984 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, 985 vaddr_t start) 986 { 987 struct vm_map_entry *new_entry; 988 989 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 990 991 uvm_map_check(map, "clip_start entry"); 992 uvm_mapent_check(entry); 993 994 /* 995 * Split off the front portion. note that we must insert the new 996 * entry BEFORE this one, so that this entry has the specified 997 * starting address. 998 */ 999 new_entry = uvm_mapent_alloc(map, 0); 1000 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 1001 uvm_mapent_splitadj(new_entry, entry, start); 1002 uvm_map_entry_link(map, entry->prev, new_entry); 1003 1004 uvm_map_check(map, "clip_start leave"); 1005 } 1006 1007 /* 1008 * uvm_map_clip_end: ensure that the entry ends at or before 1009 * the ending address, if it does't we split the reference 1010 * 1011 * => caller should use UVM_MAP_CLIP_END macro rather than calling 1012 * this directly 1013 * => map must be locked by caller 1014 */ 1015 1016 void 1017 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end) 1018 { 1019 struct vm_map_entry *new_entry; 1020 1021 uvm_map_check(map, "clip_end entry"); 1022 uvm_mapent_check(entry); 1023 1024 /* 1025 * Create a new entry and insert it 1026 * AFTER the specified entry 1027 */ 1028 new_entry = uvm_mapent_alloc(map, 0); 1029 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 1030 uvm_mapent_splitadj(entry, new_entry, end); 1031 uvm_map_entry_link(map, entry, new_entry); 1032 1033 uvm_map_check(map, "clip_end leave"); 1034 } 1035 1036 /* 1037 * M A P - m a i n e n t r y p o i n t 1038 */ 1039 /* 1040 * uvm_map: establish a valid mapping in a map 1041 * 1042 * => assume startp is page aligned. 1043 * => assume size is a multiple of PAGE_SIZE. 1044 * => assume sys_mmap provides enough of a "hint" to have us skip 1045 * over text/data/bss area. 1046 * => map must be unlocked (we will lock it) 1047 * => <uobj,uoffset> value meanings (4 cases): 1048 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 1049 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 1050 * [3] <uobj,uoffset> == normal mapping 1051 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 1052 * 1053 * case [4] is for kernel mappings where we don't know the offset until 1054 * we've found a virtual address. note that kernel object offsets are 1055 * always relative to vm_map_min(kernel_map). 1056 * 1057 * => if `align' is non-zero, we align the virtual address to the specified 1058 * alignment. 1059 * this is provided as a mechanism for large pages. 1060 * 1061 * => XXXCDC: need way to map in external amap? 1062 */ 1063 1064 int 1065 uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size, 1066 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags) 1067 { 1068 struct uvm_map_args args; 1069 struct vm_map_entry *new_entry; 1070 int error; 1071 1072 KASSERT((size & PAGE_MASK) == 0); 1073 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1074 1075 /* 1076 * for pager_map, allocate the new entry first to avoid sleeping 1077 * for memory while we have the map locked. 1078 */ 1079 1080 new_entry = NULL; 1081 if (map == pager_map) { 1082 new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT)); 1083 if (__predict_false(new_entry == NULL)) 1084 return ENOMEM; 1085 } 1086 if (map == pager_map) 1087 flags |= UVM_FLAG_NOMERGE; 1088 1089 error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align, 1090 flags, &args); 1091 if (!error) { 1092 error = uvm_map_enter(map, &args, new_entry); 1093 *startp = args.uma_start; 1094 } else if (new_entry) { 1095 uvm_mapent_free(new_entry); 1096 } 1097 1098 #if defined(DEBUG) 1099 if (!error && VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) { 1100 uvm_km_check_empty(map, *startp, *startp + size); 1101 } 1102 #endif /* defined(DEBUG) */ 1103 1104 return error; 1105 } 1106 1107 /* 1108 * uvm_map_prepare: 1109 * 1110 * called with map unlocked. 1111 * on success, returns the map locked. 1112 */ 1113 1114 int 1115 uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size, 1116 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags, 1117 struct uvm_map_args *args) 1118 { 1119 struct vm_map_entry *prev_entry; 1120 vm_prot_t prot = UVM_PROTECTION(flags); 1121 vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1122 1123 UVMHIST_FUNC(__func__); 1124 UVMHIST_CALLARGS(maphist, "(map=%#jx, start=%#jx, size=%jx, flags=%#jx)", 1125 (uintptr_t)map, start, size, flags); 1126 UVMHIST_LOG(maphist, " uobj/offset %#jx/%jd", (uintptr_t)uobj, 1127 uoffset,0,0); 1128 1129 /* 1130 * detect a popular device driver bug. 1131 */ 1132 1133 KASSERT(doing_shutdown || curlwp != NULL); 1134 1135 /* 1136 * zero-sized mapping doesn't make any sense. 1137 */ 1138 KASSERT(size > 0); 1139 1140 KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0); 1141 1142 uvm_map_check(map, "map entry"); 1143 1144 /* 1145 * check sanity of protection code 1146 */ 1147 1148 if ((prot & maxprot) != prot) { 1149 UVMHIST_LOG(maphist, "<- prot. failure: prot=%#jx, max=%#jx", 1150 prot, maxprot,0,0); 1151 return EACCES; 1152 } 1153 1154 /* 1155 * figure out where to put new VM range 1156 */ 1157 retry: 1158 if (vm_map_lock_try(map) == false) { 1159 if ((flags & UVM_FLAG_TRYLOCK) != 0) { 1160 return EAGAIN; 1161 } 1162 vm_map_lock(map); /* could sleep here */ 1163 } 1164 if (flags & UVM_FLAG_UNMAP) { 1165 KASSERT(flags & UVM_FLAG_FIXED); 1166 KASSERT((flags & UVM_FLAG_NOWAIT) == 0); 1167 1168 /* 1169 * Set prev_entry to what it will need to be after any existing 1170 * entries are removed later in uvm_map_enter(). 1171 */ 1172 1173 if (uvm_map_lookup_entry(map, start, &prev_entry)) { 1174 if (start == prev_entry->start) 1175 prev_entry = prev_entry->prev; 1176 else 1177 UVM_MAP_CLIP_END(map, prev_entry, start); 1178 SAVE_HINT(map, map->hint, prev_entry); 1179 } 1180 } else { 1181 prev_entry = uvm_map_findspace(map, start, size, &start, 1182 uobj, uoffset, align, flags); 1183 } 1184 if (prev_entry == NULL) { 1185 unsigned int timestamp; 1186 1187 timestamp = map->timestamp; 1188 UVMHIST_LOG(maphist,"waiting va timestamp=%#jx", 1189 timestamp,0,0,0); 1190 map->flags |= VM_MAP_WANTVA; 1191 vm_map_unlock(map); 1192 1193 /* 1194 * try to reclaim kva and wait until someone does unmap. 1195 * fragile locking here, so we awaken every second to 1196 * recheck the condition. 1197 */ 1198 1199 mutex_enter(&map->misc_lock); 1200 while ((map->flags & VM_MAP_WANTVA) != 0 && 1201 map->timestamp == timestamp) { 1202 if ((flags & UVM_FLAG_WAITVA) == 0) { 1203 mutex_exit(&map->misc_lock); 1204 UVMHIST_LOG(maphist, 1205 "<- uvm_map_findspace failed!", 0,0,0,0); 1206 return ENOMEM; 1207 } else { 1208 cv_timedwait(&map->cv, &map->misc_lock, hz); 1209 } 1210 } 1211 mutex_exit(&map->misc_lock); 1212 goto retry; 1213 } 1214 1215 #ifdef PMAP_GROWKERNEL 1216 /* 1217 * If the kernel pmap can't map the requested space, 1218 * then allocate more resources for it. 1219 */ 1220 if (map == kernel_map && uvm_maxkaddr < (start + size)) 1221 uvm_maxkaddr = pmap_growkernel(start + size); 1222 #endif 1223 1224 UVMMAP_EVCNT_INCR(map_call); 1225 1226 /* 1227 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 1228 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 1229 * either case we want to zero it before storing it in the map entry 1230 * (because it looks strange and confusing when debugging...) 1231 * 1232 * if uobj is not null 1233 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 1234 * and we do not need to change uoffset. 1235 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 1236 * now (based on the starting address of the map). this case is 1237 * for kernel object mappings where we don't know the offset until 1238 * the virtual address is found (with uvm_map_findspace). the 1239 * offset is the distance we are from the start of the map. 1240 */ 1241 1242 if (uobj == NULL) { 1243 uoffset = 0; 1244 } else { 1245 if (uoffset == UVM_UNKNOWN_OFFSET) { 1246 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 1247 uoffset = start - vm_map_min(kernel_map); 1248 } 1249 } 1250 1251 args->uma_flags = flags; 1252 args->uma_prev = prev_entry; 1253 args->uma_start = start; 1254 args->uma_size = size; 1255 args->uma_uobj = uobj; 1256 args->uma_uoffset = uoffset; 1257 1258 UVMHIST_LOG(maphist, "<- done!", 0,0,0,0); 1259 return 0; 1260 } 1261 1262 /* 1263 * uvm_map_enter: 1264 * 1265 * called with map locked. 1266 * unlock the map before returning. 1267 */ 1268 1269 int 1270 uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args, 1271 struct vm_map_entry *new_entry) 1272 { 1273 struct vm_map_entry *prev_entry = args->uma_prev; 1274 struct vm_map_entry *dead = NULL, *dead_entries = NULL; 1275 1276 const uvm_flag_t flags = args->uma_flags; 1277 const vm_prot_t prot = UVM_PROTECTION(flags); 1278 const vm_prot_t maxprot = UVM_MAXPROTECTION(flags); 1279 const vm_inherit_t inherit = UVM_INHERIT(flags); 1280 const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ? 1281 AMAP_EXTEND_NOWAIT : 0; 1282 const int advice = UVM_ADVICE(flags); 1283 1284 vaddr_t start = args->uma_start; 1285 vsize_t size = args->uma_size; 1286 struct uvm_object *uobj = args->uma_uobj; 1287 voff_t uoffset = args->uma_uoffset; 1288 1289 const int kmap = (vm_map_pmap(map) == pmap_kernel()); 1290 int merged = 0; 1291 int error; 1292 int newetype; 1293 1294 UVMHIST_FUNC(__func__); 1295 UVMHIST_CALLARGS(maphist, "(map=%#jx, start=%#jx, size=%ju, flags=%#jx)", 1296 (uintptr_t)map, start, size, flags); 1297 UVMHIST_LOG(maphist, " uobj/offset %#jx/%jd", (uintptr_t)uobj, 1298 uoffset,0,0); 1299 1300 KASSERT(map->hint == prev_entry); /* bimerge case assumes this */ 1301 KASSERT(vm_map_locked_p(map)); 1302 KASSERT((flags & (UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP)) != 1303 (UVM_FLAG_NOWAIT | UVM_FLAG_UNMAP)); 1304 1305 if (uobj) 1306 newetype = UVM_ET_OBJ; 1307 else 1308 newetype = 0; 1309 1310 if (flags & UVM_FLAG_COPYONW) { 1311 newetype |= UVM_ET_COPYONWRITE; 1312 if ((flags & UVM_FLAG_OVERLAY) == 0) 1313 newetype |= UVM_ET_NEEDSCOPY; 1314 } 1315 1316 /* 1317 * For mappings with unmap, remove any old entries now. Adding the new 1318 * entry cannot fail because that can only happen if UVM_FLAG_NOWAIT 1319 * is set, and we do not support nowait and unmap together. 1320 */ 1321 1322 if (flags & UVM_FLAG_UNMAP) { 1323 KASSERT(flags & UVM_FLAG_FIXED); 1324 uvm_unmap_remove(map, start, start + size, &dead_entries, 0); 1325 #ifdef DEBUG 1326 struct vm_map_entry *tmp_entry __diagused; 1327 bool rv __diagused; 1328 1329 rv = uvm_map_lookup_entry(map, start, &tmp_entry); 1330 KASSERT(!rv); 1331 KASSERTMSG(prev_entry == tmp_entry, 1332 "args %p prev_entry %p tmp_entry %p", 1333 args, prev_entry, tmp_entry); 1334 #endif 1335 SAVE_HINT(map, map->hint, prev_entry); 1336 } 1337 1338 /* 1339 * try and insert in map by extending previous entry, if possible. 1340 * XXX: we don't try and pull back the next entry. might be useful 1341 * for a stack, but we are currently allocating our stack in advance. 1342 */ 1343 1344 if (flags & UVM_FLAG_NOMERGE) 1345 goto nomerge; 1346 1347 if (prev_entry->end == start && 1348 prev_entry != &map->header && 1349 UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, 0, 1350 prot, maxprot, inherit, advice, 0)) { 1351 1352 if (uobj && prev_entry->offset + 1353 (prev_entry->end - prev_entry->start) != uoffset) 1354 goto forwardmerge; 1355 1356 /* 1357 * can't extend a shared amap. note: no need to lock amap to 1358 * look at refs since we don't care about its exact value. 1359 * if it is one (i.e. we have only reference) it will stay there 1360 */ 1361 1362 if (prev_entry->aref.ar_amap && 1363 amap_refs(prev_entry->aref.ar_amap) != 1) { 1364 goto forwardmerge; 1365 } 1366 1367 if (prev_entry->aref.ar_amap) { 1368 error = amap_extend(prev_entry, size, 1369 amapwaitflag | AMAP_EXTEND_FORWARDS); 1370 if (error) 1371 goto nomerge; 1372 } 1373 1374 if (kmap) { 1375 UVMMAP_EVCNT_INCR(kbackmerge); 1376 } else { 1377 UVMMAP_EVCNT_INCR(ubackmerge); 1378 } 1379 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 1380 1381 /* 1382 * drop our reference to uobj since we are extending a reference 1383 * that we already have (the ref count can not drop to zero). 1384 */ 1385 1386 if (uobj && uobj->pgops->pgo_detach) 1387 uobj->pgops->pgo_detach(uobj); 1388 1389 /* 1390 * Now that we've merged the entries, note that we've grown 1391 * and our gap has shrunk. Then fix the tree. 1392 */ 1393 prev_entry->end += size; 1394 prev_entry->gap -= size; 1395 uvm_rb_fixup(map, prev_entry); 1396 1397 uvm_map_check(map, "map backmerged"); 1398 1399 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 1400 merged++; 1401 } 1402 1403 forwardmerge: 1404 if (prev_entry->next->start == (start + size) && 1405 prev_entry->next != &map->header && 1406 UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, 0, 1407 prot, maxprot, inherit, advice, 0)) { 1408 1409 if (uobj && prev_entry->next->offset != uoffset + size) 1410 goto nomerge; 1411 1412 /* 1413 * can't extend a shared amap. note: no need to lock amap to 1414 * look at refs since we don't care about its exact value. 1415 * if it is one (i.e. we have only reference) it will stay there. 1416 * 1417 * note that we also can't merge two amaps, so if we 1418 * merged with the previous entry which has an amap, 1419 * and the next entry also has an amap, we give up. 1420 * 1421 * Interesting cases: 1422 * amap, new, amap -> give up second merge (single fwd extend) 1423 * amap, new, none -> double forward extend (extend again here) 1424 * none, new, amap -> double backward extend (done here) 1425 * uobj, new, amap -> single backward extend (done here) 1426 * 1427 * XXX should we attempt to deal with someone refilling 1428 * the deallocated region between two entries that are 1429 * backed by the same amap (ie, arefs is 2, "prev" and 1430 * "next" refer to it, and adding this allocation will 1431 * close the hole, thus restoring arefs to 1 and 1432 * deallocating the "next" vm_map_entry)? -- @@@ 1433 */ 1434 1435 if (prev_entry->next->aref.ar_amap && 1436 (amap_refs(prev_entry->next->aref.ar_amap) != 1 || 1437 (merged && prev_entry->aref.ar_amap))) { 1438 goto nomerge; 1439 } 1440 1441 if (merged) { 1442 /* 1443 * Try to extend the amap of the previous entry to 1444 * cover the next entry as well. If it doesn't work 1445 * just skip on, don't actually give up, since we've 1446 * already completed the back merge. 1447 */ 1448 if (prev_entry->aref.ar_amap) { 1449 if (amap_extend(prev_entry, 1450 prev_entry->next->end - 1451 prev_entry->next->start, 1452 amapwaitflag | AMAP_EXTEND_FORWARDS)) 1453 goto nomerge; 1454 } 1455 1456 /* 1457 * Try to extend the amap of the *next* entry 1458 * back to cover the new allocation *and* the 1459 * previous entry as well (the previous merge 1460 * didn't have an amap already otherwise we 1461 * wouldn't be checking here for an amap). If 1462 * it doesn't work just skip on, again, don't 1463 * actually give up, since we've already 1464 * completed the back merge. 1465 */ 1466 else if (prev_entry->next->aref.ar_amap) { 1467 if (amap_extend(prev_entry->next, 1468 prev_entry->end - 1469 prev_entry->start, 1470 amapwaitflag | AMAP_EXTEND_BACKWARDS)) 1471 goto nomerge; 1472 } 1473 } else { 1474 /* 1475 * Pull the next entry's amap backwards to cover this 1476 * new allocation. 1477 */ 1478 if (prev_entry->next->aref.ar_amap) { 1479 error = amap_extend(prev_entry->next, size, 1480 amapwaitflag | AMAP_EXTEND_BACKWARDS); 1481 if (error) 1482 goto nomerge; 1483 } 1484 } 1485 1486 if (merged) { 1487 if (kmap) { 1488 UVMMAP_EVCNT_DECR(kbackmerge); 1489 UVMMAP_EVCNT_INCR(kbimerge); 1490 } else { 1491 UVMMAP_EVCNT_DECR(ubackmerge); 1492 UVMMAP_EVCNT_INCR(ubimerge); 1493 } 1494 } else { 1495 if (kmap) { 1496 UVMMAP_EVCNT_INCR(kforwmerge); 1497 } else { 1498 UVMMAP_EVCNT_INCR(uforwmerge); 1499 } 1500 } 1501 UVMHIST_LOG(maphist," starting forward merge", 0, 0, 0, 0); 1502 1503 /* 1504 * drop our reference to uobj since we are extending a reference 1505 * that we already have (the ref count can not drop to zero). 1506 */ 1507 if (uobj && uobj->pgops->pgo_detach) 1508 uobj->pgops->pgo_detach(uobj); 1509 1510 if (merged) { 1511 dead = prev_entry->next; 1512 prev_entry->end = dead->end; 1513 uvm_map_entry_unlink(map, dead); 1514 if (dead->aref.ar_amap != NULL) { 1515 prev_entry->aref = dead->aref; 1516 dead->aref.ar_amap = NULL; 1517 } 1518 } else { 1519 prev_entry->next->start -= size; 1520 if (prev_entry != &map->header) { 1521 prev_entry->gap -= size; 1522 KASSERT(prev_entry->gap == uvm_rb_gap(prev_entry)); 1523 uvm_rb_fixup(map, prev_entry); 1524 } 1525 if (uobj) 1526 prev_entry->next->offset = uoffset; 1527 } 1528 1529 uvm_map_check(map, "map forwardmerged"); 1530 1531 UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0); 1532 merged++; 1533 } 1534 1535 nomerge: 1536 if (!merged) { 1537 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 1538 if (kmap) { 1539 UVMMAP_EVCNT_INCR(knomerge); 1540 } else { 1541 UVMMAP_EVCNT_INCR(unomerge); 1542 } 1543 1544 /* 1545 * allocate new entry and link it in. 1546 */ 1547 1548 if (new_entry == NULL) { 1549 new_entry = uvm_mapent_alloc(map, 1550 (flags & UVM_FLAG_NOWAIT)); 1551 if (__predict_false(new_entry == NULL)) { 1552 error = ENOMEM; 1553 goto done; 1554 } 1555 } 1556 new_entry->start = start; 1557 new_entry->end = new_entry->start + size; 1558 new_entry->object.uvm_obj = uobj; 1559 new_entry->offset = uoffset; 1560 1561 new_entry->etype = newetype; 1562 1563 if (flags & UVM_FLAG_NOMERGE) { 1564 new_entry->flags |= UVM_MAP_NOMERGE; 1565 } 1566 1567 new_entry->protection = prot; 1568 new_entry->max_protection = maxprot; 1569 new_entry->inheritance = inherit; 1570 new_entry->wired_count = 0; 1571 new_entry->advice = advice; 1572 if (flags & UVM_FLAG_OVERLAY) { 1573 1574 /* 1575 * to_add: for BSS we overallocate a little since we 1576 * are likely to extend 1577 */ 1578 1579 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 1580 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 1581 struct vm_amap *amap = amap_alloc(size, to_add, 1582 (flags & UVM_FLAG_NOWAIT)); 1583 if (__predict_false(amap == NULL)) { 1584 error = ENOMEM; 1585 goto done; 1586 } 1587 new_entry->aref.ar_pageoff = 0; 1588 new_entry->aref.ar_amap = amap; 1589 } else { 1590 new_entry->aref.ar_pageoff = 0; 1591 new_entry->aref.ar_amap = NULL; 1592 } 1593 uvm_map_entry_link(map, prev_entry, new_entry); 1594 1595 /* 1596 * Update the free space hint 1597 */ 1598 1599 if ((map->first_free == prev_entry) && 1600 (prev_entry->end >= new_entry->start)) 1601 map->first_free = new_entry; 1602 1603 new_entry = NULL; 1604 } 1605 1606 map->size += size; 1607 1608 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1609 1610 error = 0; 1611 1612 done: 1613 vm_map_unlock(map); 1614 1615 if (new_entry) { 1616 uvm_mapent_free(new_entry); 1617 } 1618 if (dead) { 1619 KDASSERT(merged); 1620 uvm_mapent_free(dead); 1621 } 1622 if (dead_entries) 1623 uvm_unmap_detach(dead_entries, 0); 1624 1625 return error; 1626 } 1627 1628 /* 1629 * uvm_map_lookup_entry_bytree: lookup an entry in tree 1630 */ 1631 1632 static inline bool 1633 uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address, 1634 struct vm_map_entry **entry /* OUT */) 1635 { 1636 struct vm_map_entry *prev = &map->header; 1637 struct vm_map_entry *cur = ROOT_ENTRY(map); 1638 1639 while (cur) { 1640 UVMMAP_EVCNT_INCR(mlk_treeloop); 1641 if (address >= cur->start) { 1642 if (address < cur->end) { 1643 *entry = cur; 1644 return true; 1645 } 1646 prev = cur; 1647 cur = RIGHT_ENTRY(cur); 1648 } else 1649 cur = LEFT_ENTRY(cur); 1650 } 1651 *entry = prev; 1652 return false; 1653 } 1654 1655 /* 1656 * uvm_map_lookup_entry: find map entry at or before an address 1657 * 1658 * => map must at least be read-locked by caller 1659 * => entry is returned in "entry" 1660 * => return value is true if address is in the returned entry 1661 */ 1662 1663 bool 1664 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1665 struct vm_map_entry **entry /* OUT */) 1666 { 1667 struct vm_map_entry *cur; 1668 UVMHIST_FUNC(__func__); 1669 UVMHIST_CALLARGS(maphist,"(map=%#jx,addr=%#jx,ent=%#jx)", 1670 (uintptr_t)map, address, (uintptr_t)entry, 0); 1671 1672 /* 1673 * make a quick check to see if we are already looking at 1674 * the entry we want (which is usually the case). note also 1675 * that we don't need to save the hint here... it is the 1676 * same hint (unless we are at the header, in which case the 1677 * hint didn't buy us anything anyway). 1678 */ 1679 1680 cur = map->hint; 1681 UVMMAP_EVCNT_INCR(mlk_call); 1682 if (cur != &map->header && 1683 address >= cur->start && cur->end > address) { 1684 UVMMAP_EVCNT_INCR(mlk_hint); 1685 *entry = cur; 1686 UVMHIST_LOG(maphist,"<- got it via hint (%#jx)", 1687 (uintptr_t)cur, 0, 0, 0); 1688 uvm_mapent_check(*entry); 1689 return (true); 1690 } 1691 uvm_map_check(map, __func__); 1692 1693 /* 1694 * lookup in the tree. 1695 */ 1696 1697 UVMMAP_EVCNT_INCR(mlk_tree); 1698 if (__predict_true(uvm_map_lookup_entry_bytree(map, address, entry))) { 1699 SAVE_HINT(map, map->hint, *entry); 1700 UVMHIST_LOG(maphist,"<- search got it (%#jx)", 1701 (uintptr_t)cur, 0, 0, 0); 1702 KDASSERT((*entry)->start <= address); 1703 KDASSERT(address < (*entry)->end); 1704 uvm_mapent_check(*entry); 1705 return (true); 1706 } 1707 1708 SAVE_HINT(map, map->hint, *entry); 1709 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1710 KDASSERT((*entry) == &map->header || (*entry)->end <= address); 1711 KDASSERT((*entry)->next == &map->header || 1712 address < (*entry)->next->start); 1713 return (false); 1714 } 1715 1716 /* 1717 * See if the range between start and start + length fits in the gap 1718 * entry->next->start and entry->end. Returns 1 if fits, 0 if doesn't 1719 * fit, and -1 address wraps around. 1720 */ 1721 static int 1722 uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset, 1723 vsize_t align, int flags, int topdown, struct vm_map_entry *entry) 1724 { 1725 vaddr_t end; 1726 1727 #ifdef PMAP_PREFER 1728 /* 1729 * push start address forward as needed to avoid VAC alias problems. 1730 * we only do this if a valid offset is specified. 1731 */ 1732 1733 if (uoffset != UVM_UNKNOWN_OFFSET) 1734 PMAP_PREFER(uoffset, start, length, topdown); 1735 #endif 1736 if ((flags & UVM_FLAG_COLORMATCH) != 0) { 1737 KASSERT(align < uvmexp.ncolors); 1738 if (uvmexp.ncolors > 1) { 1739 const u_int colormask = uvmexp.colormask; 1740 const u_int colorsize = colormask + 1; 1741 vaddr_t hint = atop(*start); 1742 const u_int color = hint & colormask; 1743 if (color != align) { 1744 hint -= color; /* adjust to color boundary */ 1745 KASSERT((hint & colormask) == 0); 1746 if (topdown) { 1747 if (align > color) 1748 hint -= colorsize; 1749 } else { 1750 if (align < color) 1751 hint += colorsize; 1752 } 1753 *start = ptoa(hint + align); /* adjust to color */ 1754 } 1755 } 1756 } else { 1757 KASSERT(powerof2(align)); 1758 uvm_map_align_va(start, align, topdown); 1759 /* 1760 * XXX Should we PMAP_PREFER() here again? 1761 * eh...i think we're okay 1762 */ 1763 } 1764 1765 /* 1766 * Find the end of the proposed new region. Be sure we didn't 1767 * wrap around the address; if so, we lose. Otherwise, if the 1768 * proposed new region fits before the next entry, we win. 1769 */ 1770 1771 end = *start + length; 1772 if (end < *start) 1773 return (-1); 1774 1775 if (entry->next->start >= end && *start >= entry->end) 1776 return (1); 1777 1778 return (0); 1779 } 1780 1781 static void 1782 uvm_findspace_invariants(struct vm_map *map, vaddr_t orig_hint, vaddr_t length, 1783 struct uvm_object *uobj, voff_t uoffset, vsize_t align, int flags, 1784 vaddr_t hint, struct vm_map_entry *entry, int line) 1785 { 1786 const int topdown = map->flags & VM_MAP_TOPDOWN; 1787 1788 KASSERTMSG( topdown || hint >= orig_hint, 1789 "map=%p hint=%#"PRIxVADDR" orig_hint=%#"PRIxVADDR 1790 " length=%#"PRIxVSIZE" uobj=%p uoffset=%#llx align=%"PRIxVSIZE 1791 " flags=%#x entry=%p (uvm_map_findspace line %d)", 1792 map, hint, orig_hint, 1793 length, uobj, (unsigned long long)uoffset, align, 1794 flags, entry, line); 1795 KASSERTMSG(!topdown || hint <= orig_hint, 1796 "map=%p hint=%#"PRIxVADDR" orig_hint=%#"PRIxVADDR 1797 " length=%#"PRIxVSIZE" uobj=%p uoffset=%#llx align=%"PRIxVSIZE 1798 " flags=%#x entry=%p (uvm_map_findspace line %d)", 1799 map, hint, orig_hint, 1800 length, uobj, (unsigned long long)uoffset, align, 1801 flags, entry, line); 1802 } 1803 1804 /* 1805 * uvm_map_findspace: find "length" sized space in "map". 1806 * 1807 * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is 1808 * set in "flags" (in which case we insist on using "hint"). 1809 * => "result" is VA returned 1810 * => uobj/uoffset are to be used to handle VAC alignment, if required 1811 * => if "align" is non-zero, we attempt to align to that value. 1812 * => caller must at least have read-locked map 1813 * => returns NULL on failure, or pointer to prev. map entry if success 1814 * => note this is a cross between the old vm_map_findspace and vm_map_find 1815 */ 1816 1817 struct vm_map_entry * 1818 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length, 1819 vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset, 1820 vsize_t align, int flags) 1821 { 1822 #define INVARIANTS() \ 1823 uvm_findspace_invariants(map, orig_hint, length, uobj, uoffset, align,\ 1824 flags, hint, entry, __LINE__) 1825 struct vm_map_entry *entry = NULL; 1826 struct vm_map_entry *child, *prev, *tmp; 1827 vaddr_t orig_hint __diagused; 1828 const int topdown = map->flags & VM_MAP_TOPDOWN; 1829 int avail; 1830 UVMHIST_FUNC(__func__); 1831 UVMHIST_CALLARGS(maphist, "(map=%#jx, hint=%#jx, len=%ju, flags=%#jx...", 1832 (uintptr_t)map, hint, length, flags); 1833 UVMHIST_LOG(maphist, " uobj=%#jx, uoffset=%#jx, align=%#jx)", 1834 (uintptr_t)uobj, uoffset, align, 0); 1835 1836 KASSERT((flags & UVM_FLAG_COLORMATCH) != 0 || powerof2(align)); 1837 KASSERT((flags & UVM_FLAG_COLORMATCH) == 0 || align < uvmexp.ncolors); 1838 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1839 1840 uvm_map_check(map, "map_findspace entry"); 1841 1842 /* 1843 * Clamp the hint to the VM map's min/max address, and remmeber 1844 * the clamped original hint. Remember the original hint, 1845 * clamped to the min/max address. If we are aligning, then we 1846 * may have to try again with no alignment constraint if we 1847 * fail the first time. 1848 * 1849 * We use the original hint to verify later that the search has 1850 * been monotonic -- that is, nonincreasing or nondecreasing, 1851 * according to topdown or !topdown respectively. But the 1852 * clamping is not monotonic. 1853 */ 1854 if (hint < vm_map_min(map)) { /* check ranges ... */ 1855 if (flags & UVM_FLAG_FIXED) { 1856 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1857 return (NULL); 1858 } 1859 hint = vm_map_min(map); 1860 } 1861 if (hint > vm_map_max(map)) { 1862 UVMHIST_LOG(maphist,"<- VA %#jx > range [%#jx->%#jx]", 1863 hint, vm_map_min(map), vm_map_max(map), 0); 1864 return (NULL); 1865 } 1866 orig_hint = hint; 1867 INVARIANTS(); 1868 1869 UVMHIST_LOG(maphist,"<- VA %#jx vs range [%#jx->%#jx]", 1870 hint, vm_map_min(map), vm_map_max(map), 0); 1871 1872 /* 1873 * hint may not be aligned properly; we need round up or down it 1874 * before proceeding further. 1875 */ 1876 if ((flags & UVM_FLAG_COLORMATCH) == 0) { 1877 uvm_map_align_va(&hint, align, topdown); 1878 INVARIANTS(); 1879 } 1880 1881 UVMHIST_LOG(maphist,"<- VA %#jx vs range [%#jx->%#jx]", 1882 hint, vm_map_min(map), vm_map_max(map), 0); 1883 /* 1884 * Look for the first possible address; if there's already 1885 * something at this address, we have to start after it. 1886 */ 1887 1888 /* 1889 * @@@: there are four, no, eight cases to consider. 1890 * 1891 * 0: found, fixed, bottom up -> fail 1892 * 1: found, fixed, top down -> fail 1893 * 2: found, not fixed, bottom up -> start after entry->end, 1894 * loop up 1895 * 3: found, not fixed, top down -> start before entry->start, 1896 * loop down 1897 * 4: not found, fixed, bottom up -> check entry->next->start, fail 1898 * 5: not found, fixed, top down -> check entry->next->start, fail 1899 * 6: not found, not fixed, bottom up -> check entry->next->start, 1900 * loop up 1901 * 7: not found, not fixed, top down -> check entry->next->start, 1902 * loop down 1903 * 1904 * as you can see, it reduces to roughly five cases, and that 1905 * adding top down mapping only adds one unique case (without 1906 * it, there would be four cases). 1907 */ 1908 1909 if ((flags & UVM_FLAG_FIXED) == 0 && 1910 hint == (topdown ? vm_map_max(map) : vm_map_min(map))) { 1911 /* 1912 * The uvm_map_findspace algorithm is monotonic -- for 1913 * topdown VM it starts with a high hint and returns a 1914 * lower free address; for !topdown VM it starts with a 1915 * low hint and returns a higher free address. As an 1916 * optimization, start with the first (highest for 1917 * topdown, lowest for !topdown) free address. 1918 * 1919 * XXX This `optimization' probably doesn't actually do 1920 * much in practice unless userland explicitly passes 1921 * the VM map's minimum or maximum address, which 1922 * varies from machine to machine (VM_MAX/MIN_ADDRESS, 1923 * e.g. 0x7fbfdfeff000 on amd64 but 0xfffffffff000 on 1924 * aarch64) and may vary according to other factors 1925 * like sysctl vm.user_va0_disable. In particular, if 1926 * the user specifies 0 as a hint to mmap, then mmap 1927 * will choose a default address which is usually _not_ 1928 * VM_MAX/MIN_ADDRESS but something else instead like 1929 * VM_MAX_ADDRESS - stack size - guard page overhead, 1930 * in which case this branch is never hit. 1931 * 1932 * In fact, this branch appears to have been broken for 1933 * two decades between when topdown was introduced in 1934 * ~2003 and when it was adapted to handle the topdown 1935 * case without violating the monotonicity assertion in 1936 * 2022. Maybe Someone^TM should either ditch the 1937 * optimization or find a better way to do it. 1938 */ 1939 entry = map->first_free; 1940 } else { 1941 if (uvm_map_lookup_entry(map, hint, &entry)) { 1942 /* "hint" address already in use ... */ 1943 if (flags & UVM_FLAG_FIXED) { 1944 UVMHIST_LOG(maphist, "<- fixed & VA in use", 1945 0, 0, 0, 0); 1946 return (NULL); 1947 } 1948 if (topdown) 1949 /* Start from lower gap. */ 1950 entry = entry->prev; 1951 } else if (flags & UVM_FLAG_FIXED) { 1952 if (entry->next->start >= hint + length && 1953 hint + length > hint) 1954 goto found; 1955 1956 /* "hint" address is gap but too small */ 1957 UVMHIST_LOG(maphist, "<- fixed mapping failed", 1958 0, 0, 0, 0); 1959 return (NULL); /* only one shot at it ... */ 1960 } else { 1961 /* 1962 * See if given hint fits in this gap. 1963 */ 1964 avail = uvm_map_space_avail(&hint, length, 1965 uoffset, align, flags, topdown, entry); 1966 INVARIANTS(); 1967 switch (avail) { 1968 case 1: 1969 goto found; 1970 case -1: 1971 goto wraparound; 1972 } 1973 1974 if (topdown) { 1975 /* 1976 * Still there is a chance to fit 1977 * if hint > entry->end. 1978 */ 1979 } else { 1980 /* Start from higher gap. */ 1981 entry = entry->next; 1982 if (entry == &map->header) 1983 goto notfound; 1984 goto nextgap; 1985 } 1986 } 1987 } 1988 1989 /* 1990 * Note that all UVM_FLAGS_FIXED case is already handled. 1991 */ 1992 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 1993 1994 /* Try to find the space in the red-black tree */ 1995 1996 /* Check slot before any entry */ 1997 if (topdown) { 1998 KASSERTMSG(entry->next->start >= vm_map_min(map), 1999 "map=%p entry=%p entry->next=%p" 2000 " entry->next->start=0x%"PRIxVADDR" min=0x%"PRIxVADDR, 2001 map, entry, entry->next, 2002 entry->next->start, vm_map_min(map)); 2003 if (length > entry->next->start - vm_map_min(map)) 2004 hint = vm_map_min(map); /* XXX goto wraparound? */ 2005 else 2006 hint = entry->next->start - length; 2007 KASSERT(hint >= vm_map_min(map)); 2008 } else { 2009 hint = entry->end; 2010 } 2011 INVARIANTS(); 2012 avail = uvm_map_space_avail(&hint, length, uoffset, align, flags, 2013 topdown, entry); 2014 INVARIANTS(); 2015 switch (avail) { 2016 case 1: 2017 goto found; 2018 case -1: 2019 goto wraparound; 2020 } 2021 2022 nextgap: 2023 KDASSERT((flags & UVM_FLAG_FIXED) == 0); 2024 /* If there is not enough space in the whole tree, we fail */ 2025 tmp = ROOT_ENTRY(map); 2026 if (tmp == NULL || tmp->maxgap < length) 2027 goto notfound; 2028 2029 prev = NULL; /* previous candidate */ 2030 2031 /* Find an entry close to hint that has enough space */ 2032 for (; tmp;) { 2033 KASSERT(tmp->next->start == tmp->end + tmp->gap); 2034 if (topdown) { 2035 if (tmp->next->start < hint + length && 2036 (prev == NULL || tmp->end > prev->end)) { 2037 if (tmp->gap >= length) 2038 prev = tmp; 2039 else if ((child = LEFT_ENTRY(tmp)) != NULL 2040 && child->maxgap >= length) 2041 prev = tmp; 2042 } 2043 } else { 2044 if (tmp->end >= hint && 2045 (prev == NULL || tmp->end < prev->end)) { 2046 if (tmp->gap >= length) 2047 prev = tmp; 2048 else if ((child = RIGHT_ENTRY(tmp)) != NULL 2049 && child->maxgap >= length) 2050 prev = tmp; 2051 } 2052 } 2053 if (tmp->next->start < hint + length) 2054 child = RIGHT_ENTRY(tmp); 2055 else if (tmp->end > hint) 2056 child = LEFT_ENTRY(tmp); 2057 else { 2058 if (tmp->gap >= length) 2059 break; 2060 if (topdown) 2061 child = LEFT_ENTRY(tmp); 2062 else 2063 child = RIGHT_ENTRY(tmp); 2064 } 2065 if (child == NULL || child->maxgap < length) 2066 break; 2067 tmp = child; 2068 } 2069 2070 if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) { 2071 /* 2072 * Check if the entry that we found satifies the 2073 * space requirement 2074 */ 2075 if (topdown) { 2076 if (hint > tmp->next->start - length) 2077 hint = tmp->next->start - length; 2078 } else { 2079 if (hint < tmp->end) 2080 hint = tmp->end; 2081 } 2082 INVARIANTS(); 2083 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2084 flags, topdown, tmp); 2085 INVARIANTS(); 2086 switch (avail) { 2087 case 1: 2088 entry = tmp; 2089 goto found; 2090 case -1: 2091 goto wraparound; 2092 } 2093 if (tmp->gap >= length) 2094 goto listsearch; 2095 } 2096 if (prev == NULL) 2097 goto notfound; 2098 2099 if (topdown) { 2100 KASSERT(orig_hint >= prev->next->start - length || 2101 prev->next->start - length > prev->next->start); 2102 hint = prev->next->start - length; 2103 } else { 2104 KASSERT(orig_hint <= prev->end); 2105 hint = prev->end; 2106 } 2107 INVARIANTS(); 2108 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2109 flags, topdown, prev); 2110 INVARIANTS(); 2111 switch (avail) { 2112 case 1: 2113 entry = prev; 2114 goto found; 2115 case -1: 2116 goto wraparound; 2117 } 2118 if (prev->gap >= length) 2119 goto listsearch; 2120 2121 if (topdown) 2122 tmp = LEFT_ENTRY(prev); 2123 else 2124 tmp = RIGHT_ENTRY(prev); 2125 for (;;) { 2126 KASSERT(tmp); 2127 KASSERTMSG(tmp->maxgap >= length, 2128 "tmp->maxgap=0x%"PRIxVSIZE" length=0x%"PRIxVSIZE, 2129 tmp->maxgap, length); 2130 if (topdown) 2131 child = RIGHT_ENTRY(tmp); 2132 else 2133 child = LEFT_ENTRY(tmp); 2134 if (child && child->maxgap >= length) { 2135 tmp = child; 2136 continue; 2137 } 2138 if (tmp->gap >= length) 2139 break; 2140 if (topdown) 2141 tmp = LEFT_ENTRY(tmp); 2142 else 2143 tmp = RIGHT_ENTRY(tmp); 2144 } 2145 2146 if (topdown) { 2147 KASSERT(orig_hint >= tmp->next->start - length || 2148 tmp->next->start - length > tmp->next->start); 2149 hint = tmp->next->start - length; 2150 } else { 2151 KASSERT(orig_hint <= tmp->end); 2152 hint = tmp->end; 2153 } 2154 INVARIANTS(); 2155 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2156 flags, topdown, tmp); 2157 INVARIANTS(); 2158 switch (avail) { 2159 case 1: 2160 entry = tmp; 2161 goto found; 2162 case -1: 2163 goto wraparound; 2164 } 2165 2166 /* 2167 * The tree fails to find an entry because of offset or alignment 2168 * restrictions. Search the list instead. 2169 */ 2170 listsearch: 2171 /* 2172 * Look through the rest of the map, trying to fit a new region in 2173 * the gap between existing regions, or after the very last region. 2174 * note: entry->end = base VA of current gap, 2175 * entry->next->start = VA of end of current gap 2176 */ 2177 2178 INVARIANTS(); 2179 for (;;) { 2180 /* Update hint for current gap. */ 2181 hint = topdown ? entry->next->start - length : entry->end; 2182 INVARIANTS(); 2183 2184 /* See if it fits. */ 2185 avail = uvm_map_space_avail(&hint, length, uoffset, align, 2186 flags, topdown, entry); 2187 INVARIANTS(); 2188 switch (avail) { 2189 case 1: 2190 goto found; 2191 case -1: 2192 goto wraparound; 2193 } 2194 2195 /* Advance to next/previous gap */ 2196 if (topdown) { 2197 if (entry == &map->header) { 2198 UVMHIST_LOG(maphist, "<- failed (off start)", 2199 0,0,0,0); 2200 goto notfound; 2201 } 2202 entry = entry->prev; 2203 } else { 2204 entry = entry->next; 2205 if (entry == &map->header) { 2206 UVMHIST_LOG(maphist, "<- failed (off end)", 2207 0,0,0,0); 2208 goto notfound; 2209 } 2210 } 2211 } 2212 2213 found: 2214 SAVE_HINT(map, map->hint, entry); 2215 *result = hint; 2216 UVMHIST_LOG(maphist,"<- got it! (result=%#jx)", hint, 0,0,0); 2217 INVARIANTS(); 2218 KASSERT(entry->end <= hint); 2219 KASSERT(hint + length <= entry->next->start); 2220 return (entry); 2221 2222 wraparound: 2223 UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0); 2224 2225 return (NULL); 2226 2227 notfound: 2228 UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0); 2229 2230 return (NULL); 2231 #undef INVARIANTS 2232 } 2233 2234 /* 2235 * U N M A P - m a i n h e l p e r f u n c t i o n s 2236 */ 2237 2238 /* 2239 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 2240 * 2241 * => caller must check alignment and size 2242 * => map must be locked by caller 2243 * => we return a list of map entries that we've remove from the map 2244 * in "entry_list" 2245 */ 2246 2247 void 2248 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 2249 struct vm_map_entry **entry_list /* OUT */, int flags) 2250 { 2251 struct vm_map_entry *entry, *first_entry, *next; 2252 vaddr_t len; 2253 UVMHIST_FUNC(__func__); 2254 UVMHIST_CALLARGS(maphist,"(map=%#jx, start=%#jx, end=%#jx)", 2255 (uintptr_t)map, start, end, 0); 2256 VM_MAP_RANGE_CHECK(map, start, end); 2257 2258 uvm_map_check(map, "unmap_remove entry"); 2259 2260 /* 2261 * find first entry 2262 */ 2263 2264 if (uvm_map_lookup_entry(map, start, &first_entry) == true) { 2265 /* clip and go... */ 2266 entry = first_entry; 2267 UVM_MAP_CLIP_START(map, entry, start); 2268 /* critical! prevents stale hint */ 2269 SAVE_HINT(map, entry, entry->prev); 2270 } else { 2271 entry = first_entry->next; 2272 } 2273 2274 /* 2275 * save the free space hint 2276 */ 2277 2278 if (map->first_free != &map->header && map->first_free->start >= start) 2279 map->first_free = entry->prev; 2280 2281 /* 2282 * note: we now re-use first_entry for a different task. we remove 2283 * a number of map entries from the map and save them in a linked 2284 * list headed by "first_entry". once we remove them from the map 2285 * the caller should unlock the map and drop the references to the 2286 * backing objects [c.f. uvm_unmap_detach]. the object is to 2287 * separate unmapping from reference dropping. why? 2288 * [1] the map has to be locked for unmapping 2289 * [2] the map need not be locked for reference dropping 2290 * [3] dropping references may trigger pager I/O, and if we hit 2291 * a pager that does synchronous I/O we may have to wait for it. 2292 * [4] we would like all waiting for I/O to occur with maps unlocked 2293 * so that we don't block other threads. 2294 */ 2295 2296 first_entry = NULL; 2297 *entry_list = NULL; 2298 2299 /* 2300 * break up the area into map entry sized regions and unmap. note 2301 * that all mappings have to be removed before we can even consider 2302 * dropping references to amaps or VM objects (otherwise we could end 2303 * up with a mapping to a page on the free list which would be very bad) 2304 */ 2305 2306 while ((entry != &map->header) && (entry->start < end)) { 2307 KASSERT((entry->flags & UVM_MAP_STATIC) == 0); 2308 2309 UVM_MAP_CLIP_END(map, entry, end); 2310 next = entry->next; 2311 len = entry->end - entry->start; 2312 2313 /* 2314 * unwire before removing addresses from the pmap; otherwise 2315 * unwiring will put the entries back into the pmap (XXX). 2316 */ 2317 2318 if (VM_MAPENT_ISWIRED(entry)) { 2319 uvm_map_entry_unwire(map, entry); 2320 } 2321 if (flags & UVM_FLAG_VAONLY) { 2322 2323 /* nothing */ 2324 2325 } else if ((map->flags & VM_MAP_PAGEABLE) == 0) { 2326 2327 /* 2328 * if the map is non-pageable, any pages mapped there 2329 * must be wired and entered with pmap_kenter_pa(), 2330 * and we should free any such pages immediately. 2331 * this is mostly used for kmem_map. 2332 */ 2333 KASSERT(vm_map_pmap(map) == pmap_kernel()); 2334 2335 uvm_km_pgremove_intrsafe(map, entry->start, entry->end); 2336 } else if (UVM_ET_ISOBJ(entry) && 2337 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 2338 panic("%s: kernel object %p %p\n", 2339 __func__, map, entry); 2340 } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) { 2341 /* 2342 * remove mappings the standard way. lock object 2343 * and/or amap to ensure vm_page state does not 2344 * change while in pmap_remove(). 2345 */ 2346 2347 #ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 2348 uvm_map_lock_entry(entry, RW_WRITER); 2349 #else 2350 uvm_map_lock_entry(entry, RW_READER); 2351 #endif 2352 pmap_remove(map->pmap, entry->start, entry->end); 2353 2354 /* 2355 * note: if map is dying, leave pmap_update() for 2356 * later. if the map is to be reused (exec) then 2357 * pmap_update() will be called. if the map is 2358 * being disposed of (exit) then pmap_destroy() 2359 * will be called. 2360 */ 2361 2362 if ((map->flags & VM_MAP_DYING) == 0) { 2363 pmap_update(vm_map_pmap(map)); 2364 } else { 2365 KASSERT(vm_map_pmap(map) != pmap_kernel()); 2366 } 2367 2368 uvm_map_unlock_entry(entry); 2369 } 2370 2371 #if defined(UVMDEBUG) 2372 /* 2373 * check if there's remaining mapping, 2374 * which is a bug in caller. 2375 */ 2376 2377 vaddr_t va; 2378 for (va = entry->start; va < entry->end; 2379 va += PAGE_SIZE) { 2380 if (pmap_extract(vm_map_pmap(map), va, NULL)) { 2381 panic("%s: %#"PRIxVADDR" has mapping", 2382 __func__, va); 2383 } 2384 } 2385 2386 if (VM_MAP_IS_KERNEL(map) && (flags & UVM_FLAG_NOWAIT) == 0) { 2387 uvm_km_check_empty(map, entry->start, 2388 entry->end); 2389 } 2390 #endif /* defined(UVMDEBUG) */ 2391 2392 /* 2393 * remove entry from map and put it on our list of entries 2394 * that we've nuked. then go to next entry. 2395 */ 2396 2397 UVMHIST_LOG(maphist, " removed map entry %#jx", 2398 (uintptr_t)entry, 0, 0, 0); 2399 2400 /* critical! prevents stale hint */ 2401 SAVE_HINT(map, entry, entry->prev); 2402 2403 uvm_map_entry_unlink(map, entry); 2404 KASSERT(map->size >= len); 2405 map->size -= len; 2406 entry->prev = NULL; 2407 entry->next = first_entry; 2408 first_entry = entry; 2409 entry = next; 2410 } 2411 2412 uvm_map_check(map, "unmap_remove leave"); 2413 2414 /* 2415 * now we've cleaned up the map and are ready for the caller to drop 2416 * references to the mapped objects. 2417 */ 2418 2419 *entry_list = first_entry; 2420 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 2421 2422 if (map->flags & VM_MAP_WANTVA) { 2423 mutex_enter(&map->misc_lock); 2424 map->flags &= ~VM_MAP_WANTVA; 2425 cv_broadcast(&map->cv); 2426 mutex_exit(&map->misc_lock); 2427 } 2428 } 2429 2430 /* 2431 * uvm_unmap_detach: drop references in a chain of map entries 2432 * 2433 * => we will free the map entries as we traverse the list. 2434 */ 2435 2436 void 2437 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags) 2438 { 2439 struct vm_map_entry *next_entry; 2440 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 2441 2442 while (first_entry) { 2443 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 2444 UVMHIST_LOG(maphist, 2445 " detach %#jx: amap=%#jx, obj=%#jx, submap?=%jd", 2446 (uintptr_t)first_entry, 2447 (uintptr_t)first_entry->aref.ar_amap, 2448 (uintptr_t)first_entry->object.uvm_obj, 2449 UVM_ET_ISSUBMAP(first_entry)); 2450 2451 /* 2452 * drop reference to amap, if we've got one 2453 */ 2454 2455 if (first_entry->aref.ar_amap) 2456 uvm_map_unreference_amap(first_entry, flags); 2457 2458 /* 2459 * drop reference to our backing object, if we've got one 2460 */ 2461 2462 KASSERT(!UVM_ET_ISSUBMAP(first_entry)); 2463 if (UVM_ET_ISOBJ(first_entry) && 2464 first_entry->object.uvm_obj->pgops->pgo_detach) { 2465 (*first_entry->object.uvm_obj->pgops->pgo_detach) 2466 (first_entry->object.uvm_obj); 2467 } 2468 next_entry = first_entry->next; 2469 uvm_mapent_free(first_entry); 2470 first_entry = next_entry; 2471 } 2472 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 2473 } 2474 2475 /* 2476 * E X T R A C T I O N F U N C T I O N S 2477 */ 2478 2479 /* 2480 * uvm_map_reserve: reserve space in a vm_map for future use. 2481 * 2482 * => we reserve space in a map by putting a dummy map entry in the 2483 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 2484 * => map should be unlocked (we will write lock it) 2485 * => we return true if we were able to reserve space 2486 * => XXXCDC: should be inline? 2487 */ 2488 2489 int 2490 uvm_map_reserve(struct vm_map *map, vsize_t size, 2491 vaddr_t offset /* hint for pmap_prefer */, 2492 vsize_t align /* alignment */, 2493 vaddr_t *raddr /* IN:hint, OUT: reserved VA */, 2494 uvm_flag_t flags /* UVM_FLAG_FIXED or UVM_FLAG_COLORMATCH or 0 */) 2495 { 2496 UVMHIST_FUNC(__func__); 2497 UVMHIST_CALLARGS(maphist, "(map=%#jx, size=%#jx, offset=%#jx, addr=%#jx)", 2498 (uintptr_t)map, size, offset, (uintptr_t)raddr); 2499 2500 size = round_page(size); 2501 2502 /* 2503 * reserve some virtual space. 2504 */ 2505 2506 if (uvm_map(map, raddr, size, NULL, offset, align, 2507 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 2508 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) { 2509 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 2510 return (false); 2511 } 2512 2513 UVMHIST_LOG(maphist, "<- done (*raddr=%#jx)", *raddr,0,0,0); 2514 return (true); 2515 } 2516 2517 /* 2518 * uvm_map_replace: replace a reserved (blank) area of memory with 2519 * real mappings. 2520 * 2521 * => caller must WRITE-LOCK the map 2522 * => we return true if replacement was a success 2523 * => we expect the newents chain to have nnewents entrys on it and 2524 * we expect newents->prev to point to the last entry on the list 2525 * => note newents is allowed to be NULL 2526 */ 2527 2528 static int 2529 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end, 2530 struct vm_map_entry *newents, int nnewents, vsize_t nsize, 2531 struct vm_map_entry **oldentryp) 2532 { 2533 struct vm_map_entry *oldent, *last; 2534 2535 uvm_map_check(map, "map_replace entry"); 2536 2537 /* 2538 * first find the blank map entry at the specified address 2539 */ 2540 2541 if (!uvm_map_lookup_entry(map, start, &oldent)) { 2542 return (false); 2543 } 2544 2545 /* 2546 * check to make sure we have a proper blank entry 2547 */ 2548 2549 if (end < oldent->end) { 2550 UVM_MAP_CLIP_END(map, oldent, end); 2551 } 2552 if (oldent->start != start || oldent->end != end || 2553 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 2554 return (false); 2555 } 2556 2557 #ifdef DIAGNOSTIC 2558 2559 /* 2560 * sanity check the newents chain 2561 */ 2562 2563 { 2564 struct vm_map_entry *tmpent = newents; 2565 int nent = 0; 2566 vsize_t sz = 0; 2567 vaddr_t cur = start; 2568 2569 while (tmpent) { 2570 nent++; 2571 sz += tmpent->end - tmpent->start; 2572 if (tmpent->start < cur) 2573 panic("uvm_map_replace1"); 2574 if (tmpent->start >= tmpent->end || tmpent->end > end) { 2575 panic("uvm_map_replace2: " 2576 "tmpent->start=%#"PRIxVADDR 2577 ", tmpent->end=%#"PRIxVADDR 2578 ", end=%#"PRIxVADDR, 2579 tmpent->start, tmpent->end, end); 2580 } 2581 cur = tmpent->end; 2582 if (tmpent->next) { 2583 if (tmpent->next->prev != tmpent) 2584 panic("uvm_map_replace3"); 2585 } else { 2586 if (newents->prev != tmpent) 2587 panic("uvm_map_replace4"); 2588 } 2589 tmpent = tmpent->next; 2590 } 2591 if (nent != nnewents) 2592 panic("uvm_map_replace5"); 2593 if (sz != nsize) 2594 panic("uvm_map_replace6"); 2595 } 2596 #endif 2597 2598 /* 2599 * map entry is a valid blank! replace it. (this does all the 2600 * work of map entry link/unlink...). 2601 */ 2602 2603 if (newents) { 2604 last = newents->prev; 2605 2606 /* critical: flush stale hints out of map */ 2607 SAVE_HINT(map, map->hint, newents); 2608 if (map->first_free == oldent) 2609 map->first_free = last; 2610 2611 last->next = oldent->next; 2612 last->next->prev = last; 2613 2614 /* Fix RB tree */ 2615 uvm_rb_remove(map, oldent); 2616 2617 newents->prev = oldent->prev; 2618 newents->prev->next = newents; 2619 map->nentries = map->nentries + (nnewents - 1); 2620 2621 /* Fixup the RB tree */ 2622 { 2623 int i; 2624 struct vm_map_entry *tmp; 2625 2626 tmp = newents; 2627 for (i = 0; i < nnewents && tmp; i++) { 2628 uvm_rb_insert(map, tmp); 2629 tmp = tmp->next; 2630 } 2631 } 2632 } else { 2633 /* NULL list of new entries: just remove the old one */ 2634 clear_hints(map, oldent); 2635 uvm_map_entry_unlink(map, oldent); 2636 } 2637 map->size -= end - start - nsize; 2638 2639 uvm_map_check(map, "map_replace leave"); 2640 2641 /* 2642 * now we can free the old blank entry and return. 2643 */ 2644 2645 *oldentryp = oldent; 2646 return (true); 2647 } 2648 2649 /* 2650 * uvm_map_extract: extract a mapping from a map and put it somewhere 2651 * (maybe removing the old mapping) 2652 * 2653 * => maps should be unlocked (we will write lock them) 2654 * => returns 0 on success, error code otherwise 2655 * => start must be page aligned 2656 * => len must be page sized 2657 * => flags: 2658 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 2659 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 2660 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 2661 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 2662 * UVM_EXTRACT_PROT_ALL: set prot to UVM_PROT_ALL as we go 2663 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 2664 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 2665 * be used from within the kernel in a kernel level map <<< 2666 */ 2667 2668 int 2669 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 2670 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags) 2671 { 2672 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge; 2673 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry, 2674 *deadentry, *oldentry; 2675 struct vm_map_entry *resentry = NULL; /* a dummy reservation entry */ 2676 vsize_t elen __unused; 2677 int nchain, error, copy_ok; 2678 vsize_t nsize; 2679 UVMHIST_FUNC(__func__); 2680 UVMHIST_CALLARGS(maphist,"(srcmap=%#jx,start=%#jx, len=%#jx", 2681 (uintptr_t)srcmap, start, len, 0); 2682 UVMHIST_LOG(maphist," ...,dstmap=%#jx, flags=%#jx)", 2683 (uintptr_t)dstmap, flags, 0, 0); 2684 2685 /* 2686 * step 0: sanity check: start must be on a page boundary, length 2687 * must be page sized. can't ask for CONTIG/QREF if you asked for 2688 * REMOVE. 2689 */ 2690 2691 KASSERTMSG((start & PAGE_MASK) == 0, "start=0x%"PRIxVADDR, start); 2692 KASSERTMSG((len & PAGE_MASK) == 0, "len=0x%"PRIxVADDR, len); 2693 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 2694 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 2695 2696 /* 2697 * step 1: reserve space in the target map for the extracted area 2698 */ 2699 2700 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 2701 dstaddr = vm_map_min(dstmap); 2702 if (!uvm_map_reserve(dstmap, len, start, 2703 atop(start) & uvmexp.colormask, &dstaddr, 2704 UVM_FLAG_COLORMATCH)) 2705 return (ENOMEM); 2706 KASSERT((atop(start ^ dstaddr) & uvmexp.colormask) == 0); 2707 *dstaddrp = dstaddr; /* pass address back to caller */ 2708 UVMHIST_LOG(maphist, " dstaddr=%#jx", dstaddr,0,0,0); 2709 } else { 2710 dstaddr = *dstaddrp; 2711 } 2712 2713 /* 2714 * step 2: setup for the extraction process loop by init'ing the 2715 * map entry chain, locking src map, and looking up the first useful 2716 * entry in the map. 2717 */ 2718 2719 end = start + len; 2720 newend = dstaddr + len; 2721 chain = endchain = NULL; 2722 nchain = 0; 2723 nsize = 0; 2724 vm_map_lock(srcmap); 2725 2726 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 2727 2728 /* "start" is within an entry */ 2729 if (flags & UVM_EXTRACT_QREF) { 2730 2731 /* 2732 * for quick references we don't clip the entry, so 2733 * the entry may map space "before" the starting 2734 * virtual address... this is the "fudge" factor 2735 * (which can be non-zero only the first time 2736 * through the "while" loop in step 3). 2737 */ 2738 2739 fudge = start - entry->start; 2740 } else { 2741 2742 /* 2743 * normal reference: we clip the map to fit (thus 2744 * fudge is zero) 2745 */ 2746 2747 UVM_MAP_CLIP_START(srcmap, entry, start); 2748 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 2749 fudge = 0; 2750 } 2751 } else { 2752 2753 /* "start" is not within an entry ... skip to next entry */ 2754 if (flags & UVM_EXTRACT_CONTIG) { 2755 error = EINVAL; 2756 goto bad; /* definite hole here ... */ 2757 } 2758 2759 entry = entry->next; 2760 fudge = 0; 2761 } 2762 2763 /* save values from srcmap for step 6 */ 2764 orig_entry = entry; 2765 orig_fudge = fudge; 2766 2767 /* 2768 * step 3: now start looping through the map entries, extracting 2769 * as we go. 2770 */ 2771 2772 while (entry->start < end && entry != &srcmap->header) { 2773 2774 /* if we are not doing a quick reference, clip it */ 2775 if ((flags & UVM_EXTRACT_QREF) == 0) 2776 UVM_MAP_CLIP_END(srcmap, entry, end); 2777 2778 /* clear needs_copy (allow chunking) */ 2779 if (UVM_ET_ISNEEDSCOPY(entry)) { 2780 amap_copy(srcmap, entry, 2781 AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end); 2782 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 2783 error = ENOMEM; 2784 goto bad; 2785 } 2786 2787 /* amap_copy could clip (during chunk)! update fudge */ 2788 if (fudge) { 2789 fudge = start - entry->start; 2790 orig_fudge = fudge; 2791 } 2792 } 2793 2794 /* calculate the offset of this from "start" */ 2795 oldoffset = (entry->start + fudge) - start; 2796 2797 /* allocate a new map entry */ 2798 newentry = uvm_mapent_alloc(dstmap, 0); 2799 if (newentry == NULL) { 2800 error = ENOMEM; 2801 goto bad; 2802 } 2803 2804 /* set up new map entry */ 2805 newentry->next = NULL; 2806 newentry->prev = endchain; 2807 newentry->start = dstaddr + oldoffset; 2808 newentry->end = 2809 newentry->start + (entry->end - (entry->start + fudge)); 2810 if (newentry->end > newend || newentry->end < newentry->start) 2811 newentry->end = newend; 2812 newentry->object.uvm_obj = entry->object.uvm_obj; 2813 if (newentry->object.uvm_obj) { 2814 if (newentry->object.uvm_obj->pgops->pgo_reference) 2815 newentry->object.uvm_obj->pgops-> 2816 pgo_reference(newentry->object.uvm_obj); 2817 newentry->offset = entry->offset + fudge; 2818 } else { 2819 newentry->offset = 0; 2820 } 2821 newentry->etype = entry->etype; 2822 if (flags & UVM_EXTRACT_PROT_ALL) { 2823 newentry->protection = newentry->max_protection = 2824 UVM_PROT_ALL; 2825 } else { 2826 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 2827 entry->max_protection : entry->protection; 2828 newentry->max_protection = entry->max_protection; 2829 } 2830 newentry->inheritance = entry->inheritance; 2831 newentry->wired_count = 0; 2832 newentry->aref.ar_amap = entry->aref.ar_amap; 2833 if (newentry->aref.ar_amap) { 2834 newentry->aref.ar_pageoff = 2835 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 2836 uvm_map_reference_amap(newentry, AMAP_SHARED | 2837 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 2838 } else { 2839 newentry->aref.ar_pageoff = 0; 2840 } 2841 newentry->advice = entry->advice; 2842 if ((flags & UVM_EXTRACT_QREF) != 0) { 2843 newentry->flags |= UVM_MAP_NOMERGE; 2844 } 2845 2846 /* now link it on the chain */ 2847 nchain++; 2848 nsize += newentry->end - newentry->start; 2849 if (endchain == NULL) { 2850 chain = endchain = newentry; 2851 } else { 2852 endchain->next = newentry; 2853 endchain = newentry; 2854 } 2855 2856 /* end of 'while' loop! */ 2857 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 2858 (entry->next == &srcmap->header || 2859 entry->next->start != entry->end)) { 2860 error = EINVAL; 2861 goto bad; 2862 } 2863 entry = entry->next; 2864 fudge = 0; 2865 } 2866 2867 /* 2868 * step 4: close off chain (in format expected by uvm_map_replace) 2869 */ 2870 2871 if (chain) 2872 chain->prev = endchain; 2873 2874 /* 2875 * step 5: attempt to lock the dest map so we can pmap_copy. 2876 * note usage of copy_ok: 2877 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 2878 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 2879 */ 2880 2881 if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) { 2882 copy_ok = 1; 2883 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2884 nchain, nsize, &resentry)) { 2885 if (srcmap != dstmap) 2886 vm_map_unlock(dstmap); 2887 error = EIO; 2888 goto bad; 2889 } 2890 } else { 2891 copy_ok = 0; 2892 /* replace defered until step 7 */ 2893 } 2894 2895 /* 2896 * step 6: traverse the srcmap a second time to do the following: 2897 * - if we got a lock on the dstmap do pmap_copy 2898 * - if UVM_EXTRACT_REMOVE remove the entries 2899 * we make use of orig_entry and orig_fudge (saved in step 2) 2900 */ 2901 2902 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 2903 2904 /* purge possible stale hints from srcmap */ 2905 if (flags & UVM_EXTRACT_REMOVE) { 2906 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 2907 if (srcmap->first_free != &srcmap->header && 2908 srcmap->first_free->start >= start) 2909 srcmap->first_free = orig_entry->prev; 2910 } 2911 2912 entry = orig_entry; 2913 fudge = orig_fudge; 2914 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 2915 2916 while (entry->start < end && entry != &srcmap->header) { 2917 if (copy_ok) { 2918 oldoffset = (entry->start + fudge) - start; 2919 elen = MIN(end, entry->end) - 2920 (entry->start + fudge); 2921 pmap_copy(dstmap->pmap, srcmap->pmap, 2922 dstaddr + oldoffset, elen, 2923 entry->start + fudge); 2924 } 2925 2926 /* we advance "entry" in the following if statement */ 2927 if (flags & UVM_EXTRACT_REMOVE) { 2928 #ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 2929 uvm_map_lock_entry(entry, RW_WRITER); 2930 #else 2931 uvm_map_lock_entry(entry, RW_READER); 2932 #endif 2933 pmap_remove(srcmap->pmap, entry->start, 2934 entry->end); 2935 uvm_map_unlock_entry(entry); 2936 oldentry = entry; /* save entry */ 2937 entry = entry->next; /* advance */ 2938 uvm_map_entry_unlink(srcmap, oldentry); 2939 /* add to dead list */ 2940 oldentry->next = deadentry; 2941 deadentry = oldentry; 2942 } else { 2943 entry = entry->next; /* advance */ 2944 } 2945 2946 /* end of 'while' loop */ 2947 fudge = 0; 2948 } 2949 pmap_update(srcmap->pmap); 2950 2951 /* 2952 * unlock dstmap. we will dispose of deadentry in 2953 * step 7 if needed 2954 */ 2955 2956 if (copy_ok && srcmap != dstmap) 2957 vm_map_unlock(dstmap); 2958 2959 } else { 2960 deadentry = NULL; 2961 } 2962 2963 /* 2964 * step 7: we are done with the source map, unlock. if copy_ok 2965 * is 0 then we have not replaced the dummy mapping in dstmap yet 2966 * and we need to do so now. 2967 */ 2968 2969 vm_map_unlock(srcmap); 2970 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 2971 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 2972 2973 /* now do the replacement if we didn't do it in step 5 */ 2974 if (copy_ok == 0) { 2975 vm_map_lock(dstmap); 2976 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2977 nchain, nsize, &resentry); 2978 vm_map_unlock(dstmap); 2979 2980 if (error == false) { 2981 error = EIO; 2982 goto bad2; 2983 } 2984 } 2985 2986 if (resentry != NULL) 2987 uvm_mapent_free(resentry); 2988 2989 return (0); 2990 2991 /* 2992 * bad: failure recovery 2993 */ 2994 bad: 2995 vm_map_unlock(srcmap); 2996 bad2: /* src already unlocked */ 2997 if (chain) 2998 uvm_unmap_detach(chain, 2999 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 3000 3001 if (resentry != NULL) 3002 uvm_mapent_free(resentry); 3003 3004 if ((flags & UVM_EXTRACT_RESERVED) == 0) { 3005 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 3006 } 3007 return (error); 3008 } 3009 3010 /* end of extraction functions */ 3011 3012 /* 3013 * uvm_map_submap: punch down part of a map into a submap 3014 * 3015 * => only the kernel_map is allowed to be submapped 3016 * => the purpose of submapping is to break up the locking granularity 3017 * of a larger map 3018 * => the range specified must have been mapped previously with a uvm_map() 3019 * call [with uobj==NULL] to create a blank map entry in the main map. 3020 * [And it had better still be blank!] 3021 * => maps which contain submaps should never be copied or forked. 3022 * => to remove a submap, use uvm_unmap() on the main map 3023 * and then uvm_map_deallocate() the submap. 3024 * => main map must be unlocked. 3025 * => submap must have been init'd and have a zero reference count. 3026 * [need not be locked as we don't actually reference it] 3027 */ 3028 3029 int 3030 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 3031 struct vm_map *submap) 3032 { 3033 struct vm_map_entry *entry; 3034 int error; 3035 3036 vm_map_lock(map); 3037 VM_MAP_RANGE_CHECK(map, start, end); 3038 3039 if (uvm_map_lookup_entry(map, start, &entry)) { 3040 UVM_MAP_CLIP_START(map, entry, start); 3041 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 3042 } else { 3043 entry = NULL; 3044 } 3045 3046 if (entry != NULL && 3047 entry->start == start && entry->end == end && 3048 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 3049 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 3050 entry->etype |= UVM_ET_SUBMAP; 3051 entry->object.sub_map = submap; 3052 entry->offset = 0; 3053 uvm_map_reference(submap); 3054 error = 0; 3055 } else { 3056 error = EINVAL; 3057 } 3058 vm_map_unlock(map); 3059 3060 return error; 3061 } 3062 3063 /* 3064 * uvm_map_protect_user: change map protection on behalf of the user. 3065 * Enforces PAX settings as necessary. 3066 */ 3067 int 3068 uvm_map_protect_user(struct lwp *l, vaddr_t start, vaddr_t end, 3069 vm_prot_t new_prot) 3070 { 3071 int error; 3072 3073 if ((error = PAX_MPROTECT_VALIDATE(l, new_prot))) 3074 return error; 3075 3076 return uvm_map_protect(&l->l_proc->p_vmspace->vm_map, start, end, 3077 new_prot, false); 3078 } 3079 3080 3081 /* 3082 * uvm_map_protect: change map protection 3083 * 3084 * => set_max means set max_protection. 3085 * => map must be unlocked. 3086 */ 3087 3088 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 3089 ~VM_PROT_WRITE : VM_PROT_ALL) 3090 3091 int 3092 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 3093 vm_prot_t new_prot, bool set_max) 3094 { 3095 struct vm_map_entry *current, *entry; 3096 int error = 0; 3097 UVMHIST_FUNC(__func__); 3098 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_prot=%#jx)", 3099 (uintptr_t)map, start, end, new_prot); 3100 3101 vm_map_lock(map); 3102 VM_MAP_RANGE_CHECK(map, start, end); 3103 if (uvm_map_lookup_entry(map, start, &entry)) { 3104 UVM_MAP_CLIP_START(map, entry, start); 3105 } else { 3106 entry = entry->next; 3107 } 3108 3109 /* 3110 * make a first pass to check for protection violations. 3111 */ 3112 3113 current = entry; 3114 while ((current != &map->header) && (current->start < end)) { 3115 if (UVM_ET_ISSUBMAP(current)) { 3116 error = EINVAL; 3117 goto out; 3118 } 3119 if ((new_prot & current->max_protection) != new_prot) { 3120 error = EACCES; 3121 goto out; 3122 } 3123 /* 3124 * Don't allow VM_PROT_EXECUTE to be set on entries that 3125 * point to vnodes that are associated with a NOEXEC file 3126 * system. 3127 */ 3128 if (UVM_ET_ISOBJ(current) && 3129 UVM_OBJ_IS_VNODE(current->object.uvm_obj)) { 3130 struct vnode *vp = 3131 (struct vnode *) current->object.uvm_obj; 3132 3133 if ((new_prot & VM_PROT_EXECUTE) != 0 && 3134 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) { 3135 error = EACCES; 3136 goto out; 3137 } 3138 } 3139 3140 current = current->next; 3141 } 3142 3143 /* go back and fix up protections (no need to clip this time). */ 3144 3145 current = entry; 3146 while ((current != &map->header) && (current->start < end)) { 3147 vm_prot_t old_prot; 3148 3149 UVM_MAP_CLIP_END(map, current, end); 3150 old_prot = current->protection; 3151 if (set_max) 3152 current->protection = 3153 (current->max_protection = new_prot) & old_prot; 3154 else 3155 current->protection = new_prot; 3156 3157 /* 3158 * update physical map if necessary. worry about copy-on-write 3159 * here -- CHECK THIS XXX 3160 */ 3161 3162 if (current->protection != old_prot) { 3163 /* update pmap! */ 3164 #ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 3165 uvm_map_lock_entry(current, RW_WRITER); 3166 #else 3167 uvm_map_lock_entry(current, RW_READER); 3168 #endif 3169 pmap_protect(map->pmap, current->start, current->end, 3170 current->protection & MASK(current)); 3171 uvm_map_unlock_entry(current); 3172 3173 /* 3174 * If this entry points at a vnode, and the 3175 * protection includes VM_PROT_EXECUTE, mark 3176 * the vnode as VEXECMAP. 3177 */ 3178 if (UVM_ET_ISOBJ(current)) { 3179 struct uvm_object *uobj = 3180 current->object.uvm_obj; 3181 3182 if (UVM_OBJ_IS_VNODE(uobj) && 3183 (current->protection & VM_PROT_EXECUTE)) { 3184 vn_markexec((struct vnode *) uobj); 3185 } 3186 } 3187 } 3188 3189 /* 3190 * If the map is configured to lock any future mappings, 3191 * wire this entry now if the old protection was VM_PROT_NONE 3192 * and the new protection is not VM_PROT_NONE. 3193 */ 3194 3195 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 3196 VM_MAPENT_ISWIRED(current) == 0 && 3197 old_prot == VM_PROT_NONE && 3198 new_prot != VM_PROT_NONE) { 3199 3200 /* 3201 * We must call pmap_update() here because the 3202 * pmap_protect() call above might have removed some 3203 * pmap entries and uvm_map_pageable() might create 3204 * some new pmap entries that rely on the prior 3205 * removals being completely finished. 3206 */ 3207 3208 pmap_update(map->pmap); 3209 3210 if (uvm_map_pageable(map, current->start, 3211 current->end, false, 3212 UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 3213 3214 /* 3215 * If locking the entry fails, remember the 3216 * error if it's the first one. Note we 3217 * still continue setting the protection in 3218 * the map, but will return the error 3219 * condition regardless. 3220 * 3221 * XXX Ignore what the actual error is, 3222 * XXX just call it a resource shortage 3223 * XXX so that it doesn't get confused 3224 * XXX what uvm_map_protect() itself would 3225 * XXX normally return. 3226 */ 3227 3228 error = ENOMEM; 3229 } 3230 } 3231 current = current->next; 3232 } 3233 pmap_update(map->pmap); 3234 3235 out: 3236 vm_map_unlock(map); 3237 3238 UVMHIST_LOG(maphist, "<- done, error=%jd",error,0,0,0); 3239 return error; 3240 } 3241 3242 #undef MASK 3243 3244 /* 3245 * uvm_map_inherit: set inheritance code for range of addrs in map. 3246 * 3247 * => map must be unlocked 3248 * => note that the inherit code is used during a "fork". see fork 3249 * code for details. 3250 */ 3251 3252 int 3253 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 3254 vm_inherit_t new_inheritance) 3255 { 3256 struct vm_map_entry *entry, *temp_entry; 3257 UVMHIST_FUNC(__func__); 3258 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_inh=%#jx)", 3259 (uintptr_t)map, start, end, new_inheritance); 3260 3261 switch (new_inheritance) { 3262 case MAP_INHERIT_NONE: 3263 case MAP_INHERIT_COPY: 3264 case MAP_INHERIT_SHARE: 3265 case MAP_INHERIT_ZERO: 3266 break; 3267 default: 3268 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3269 return EINVAL; 3270 } 3271 3272 vm_map_lock(map); 3273 VM_MAP_RANGE_CHECK(map, start, end); 3274 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3275 entry = temp_entry; 3276 UVM_MAP_CLIP_START(map, entry, start); 3277 } else { 3278 entry = temp_entry->next; 3279 } 3280 while ((entry != &map->header) && (entry->start < end)) { 3281 UVM_MAP_CLIP_END(map, entry, end); 3282 entry->inheritance = new_inheritance; 3283 entry = entry->next; 3284 } 3285 vm_map_unlock(map); 3286 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3287 return 0; 3288 } 3289 3290 /* 3291 * uvm_map_advice: set advice code for range of addrs in map. 3292 * 3293 * => map must be unlocked 3294 */ 3295 3296 int 3297 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 3298 { 3299 struct vm_map_entry *entry, *temp_entry; 3300 UVMHIST_FUNC(__func__); 3301 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_adv=%#jx)", 3302 (uintptr_t)map, start, end, new_advice); 3303 3304 vm_map_lock(map); 3305 VM_MAP_RANGE_CHECK(map, start, end); 3306 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 3307 entry = temp_entry; 3308 UVM_MAP_CLIP_START(map, entry, start); 3309 } else { 3310 entry = temp_entry->next; 3311 } 3312 3313 /* 3314 * XXXJRT: disallow holes? 3315 */ 3316 3317 while ((entry != &map->header) && (entry->start < end)) { 3318 UVM_MAP_CLIP_END(map, entry, end); 3319 3320 switch (new_advice) { 3321 case MADV_NORMAL: 3322 case MADV_RANDOM: 3323 case MADV_SEQUENTIAL: 3324 /* nothing special here */ 3325 break; 3326 3327 default: 3328 vm_map_unlock(map); 3329 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 3330 return EINVAL; 3331 } 3332 entry->advice = new_advice; 3333 entry = entry->next; 3334 } 3335 3336 vm_map_unlock(map); 3337 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3338 return 0; 3339 } 3340 3341 /* 3342 * uvm_map_willneed: apply MADV_WILLNEED 3343 */ 3344 3345 int 3346 uvm_map_willneed(struct vm_map *map, vaddr_t start, vaddr_t end) 3347 { 3348 struct vm_map_entry *entry; 3349 UVMHIST_FUNC(__func__); 3350 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx)", 3351 (uintptr_t)map, start, end, 0); 3352 3353 vm_map_lock_read(map); 3354 VM_MAP_RANGE_CHECK(map, start, end); 3355 if (!uvm_map_lookup_entry(map, start, &entry)) { 3356 entry = entry->next; 3357 } 3358 while (entry->start < end) { 3359 struct vm_amap * const amap = entry->aref.ar_amap; 3360 struct uvm_object * const uobj = entry->object.uvm_obj; 3361 3362 KASSERT(entry != &map->header); 3363 KASSERT(start < entry->end); 3364 /* 3365 * For now, we handle only the easy but commonly-requested case. 3366 * ie. start prefetching of backing uobj pages. 3367 * 3368 * XXX It might be useful to pmap_enter() the already-in-core 3369 * pages by inventing a "weak" mode for uvm_fault() which would 3370 * only do the PGO_LOCKED pgo_get(). 3371 */ 3372 if (UVM_ET_ISOBJ(entry) && amap == NULL && uobj != NULL) { 3373 off_t offset; 3374 off_t size; 3375 3376 offset = entry->offset; 3377 if (start < entry->start) { 3378 offset += entry->start - start; 3379 } 3380 size = entry->offset + (entry->end - entry->start); 3381 if (entry->end < end) { 3382 size -= end - entry->end; 3383 } 3384 uvm_readahead(uobj, offset, size); 3385 } 3386 entry = entry->next; 3387 } 3388 vm_map_unlock_read(map); 3389 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 3390 return 0; 3391 } 3392 3393 /* 3394 * uvm_map_pageable: sets the pageability of a range in a map. 3395 * 3396 * => wires map entries. should not be used for transient page locking. 3397 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 3398 * => regions specified as not pageable require lock-down (wired) memory 3399 * and page tables. 3400 * => map must never be read-locked 3401 * => if islocked is true, map is already write-locked 3402 * => we always unlock the map, since we must downgrade to a read-lock 3403 * to call uvm_fault_wire() 3404 * => XXXCDC: check this and try and clean it up. 3405 */ 3406 3407 int 3408 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 3409 bool new_pageable, int lockflags) 3410 { 3411 struct vm_map_entry *entry, *start_entry, *failed_entry; 3412 int rv; 3413 #ifdef DIAGNOSTIC 3414 u_int timestamp_save; 3415 #endif 3416 UVMHIST_FUNC(__func__); 3417 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,new_pageable=%ju)", 3418 (uintptr_t)map, start, end, new_pageable); 3419 KASSERT(map->flags & VM_MAP_PAGEABLE); 3420 3421 if ((lockflags & UVM_LK_ENTER) == 0) 3422 vm_map_lock(map); 3423 VM_MAP_RANGE_CHECK(map, start, end); 3424 3425 /* 3426 * only one pageability change may take place at one time, since 3427 * uvm_fault_wire assumes it will be called only once for each 3428 * wiring/unwiring. therefore, we have to make sure we're actually 3429 * changing the pageability for the entire region. we do so before 3430 * making any changes. 3431 */ 3432 3433 if (uvm_map_lookup_entry(map, start, &start_entry) == false) { 3434 if ((lockflags & UVM_LK_EXIT) == 0) 3435 vm_map_unlock(map); 3436 3437 UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0); 3438 return EFAULT; 3439 } 3440 entry = start_entry; 3441 3442 if (start == end) { /* nothing required */ 3443 if ((lockflags & UVM_LK_EXIT) == 0) 3444 vm_map_unlock(map); 3445 3446 UVMHIST_LOG(maphist,"<- done (nothing)",0,0,0,0); 3447 return 0; 3448 } 3449 3450 /* 3451 * handle wiring and unwiring separately. 3452 */ 3453 3454 if (new_pageable) { /* unwire */ 3455 UVM_MAP_CLIP_START(map, entry, start); 3456 3457 /* 3458 * unwiring. first ensure that the range to be unwired is 3459 * really wired down and that there are no holes. 3460 */ 3461 3462 while ((entry != &map->header) && (entry->start < end)) { 3463 if (entry->wired_count == 0 || 3464 (entry->end < end && 3465 (entry->next == &map->header || 3466 entry->next->start > entry->end))) { 3467 if ((lockflags & UVM_LK_EXIT) == 0) 3468 vm_map_unlock(map); 3469 UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0); 3470 return EINVAL; 3471 } 3472 entry = entry->next; 3473 } 3474 3475 /* 3476 * POSIX 1003.1b - a single munlock call unlocks a region, 3477 * regardless of the number of mlock calls made on that 3478 * region. 3479 */ 3480 3481 entry = start_entry; 3482 while ((entry != &map->header) && (entry->start < end)) { 3483 UVM_MAP_CLIP_END(map, entry, end); 3484 if (VM_MAPENT_ISWIRED(entry)) 3485 uvm_map_entry_unwire(map, entry); 3486 entry = entry->next; 3487 } 3488 if ((lockflags & UVM_LK_EXIT) == 0) 3489 vm_map_unlock(map); 3490 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3491 return 0; 3492 } 3493 3494 /* 3495 * wire case: in two passes [XXXCDC: ugly block of code here] 3496 * 3497 * 1: holding the write lock, we create any anonymous maps that need 3498 * to be created. then we clip each map entry to the region to 3499 * be wired and increment its wiring count. 3500 * 3501 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 3502 * in the pages for any newly wired area (wired_count == 1). 3503 * 3504 * downgrading to a read lock for uvm_fault_wire avoids a possible 3505 * deadlock with another thread that may have faulted on one of 3506 * the pages to be wired (it would mark the page busy, blocking 3507 * us, then in turn block on the map lock that we hold). because 3508 * of problems in the recursive lock package, we cannot upgrade 3509 * to a write lock in vm_map_lookup. thus, any actions that 3510 * require the write lock must be done beforehand. because we 3511 * keep the read lock on the map, the copy-on-write status of the 3512 * entries we modify here cannot change. 3513 */ 3514 3515 while ((entry != &map->header) && (entry->start < end)) { 3516 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3517 3518 /* 3519 * perform actions of vm_map_lookup that need the 3520 * write lock on the map: create an anonymous map 3521 * for a copy-on-write region, or an anonymous map 3522 * for a zero-fill region. (XXXCDC: submap case 3523 * ok?) 3524 */ 3525 3526 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3527 if (UVM_ET_ISNEEDSCOPY(entry) && 3528 ((entry->max_protection & VM_PROT_WRITE) || 3529 (entry->object.uvm_obj == NULL))) { 3530 amap_copy(map, entry, 0, start, end); 3531 /* XXXCDC: wait OK? */ 3532 } 3533 } 3534 } 3535 UVM_MAP_CLIP_START(map, entry, start); 3536 UVM_MAP_CLIP_END(map, entry, end); 3537 entry->wired_count++; 3538 3539 /* 3540 * Check for holes 3541 */ 3542 3543 if (entry->protection == VM_PROT_NONE || 3544 (entry->end < end && 3545 (entry->next == &map->header || 3546 entry->next->start > entry->end))) { 3547 3548 /* 3549 * found one. amap creation actions do not need to 3550 * be undone, but the wired counts need to be restored. 3551 */ 3552 3553 while (entry != &map->header && entry->end > start) { 3554 entry->wired_count--; 3555 entry = entry->prev; 3556 } 3557 if ((lockflags & UVM_LK_EXIT) == 0) 3558 vm_map_unlock(map); 3559 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 3560 return EINVAL; 3561 } 3562 entry = entry->next; 3563 } 3564 3565 /* 3566 * Pass 2. 3567 */ 3568 3569 #ifdef DIAGNOSTIC 3570 timestamp_save = map->timestamp; 3571 #endif 3572 vm_map_busy(map); 3573 vm_map_unlock(map); 3574 3575 rv = 0; 3576 entry = start_entry; 3577 while (entry != &map->header && entry->start < end) { 3578 if (entry->wired_count == 1) { 3579 rv = uvm_fault_wire(map, entry->start, entry->end, 3580 entry->max_protection, 1); 3581 if (rv) { 3582 3583 /* 3584 * wiring failed. break out of the loop. 3585 * we'll clean up the map below, once we 3586 * have a write lock again. 3587 */ 3588 3589 break; 3590 } 3591 } 3592 entry = entry->next; 3593 } 3594 3595 if (rv) { /* failed? */ 3596 3597 /* 3598 * Get back to an exclusive (write) lock. 3599 */ 3600 3601 vm_map_lock(map); 3602 vm_map_unbusy(map); 3603 3604 #ifdef DIAGNOSTIC 3605 if (timestamp_save + 1 != map->timestamp) 3606 panic("uvm_map_pageable: stale map"); 3607 #endif 3608 3609 /* 3610 * first drop the wiring count on all the entries 3611 * which haven't actually been wired yet. 3612 */ 3613 3614 failed_entry = entry; 3615 while (entry != &map->header && entry->start < end) { 3616 entry->wired_count--; 3617 entry = entry->next; 3618 } 3619 3620 /* 3621 * now, unwire all the entries that were successfully 3622 * wired above. 3623 */ 3624 3625 entry = start_entry; 3626 while (entry != failed_entry) { 3627 entry->wired_count--; 3628 if (VM_MAPENT_ISWIRED(entry) == 0) 3629 uvm_map_entry_unwire(map, entry); 3630 entry = entry->next; 3631 } 3632 if ((lockflags & UVM_LK_EXIT) == 0) 3633 vm_map_unlock(map); 3634 UVMHIST_LOG(maphist, "<- done (RV=%jd)", rv,0,0,0); 3635 return (rv); 3636 } 3637 3638 if ((lockflags & UVM_LK_EXIT) == 0) { 3639 vm_map_unbusy(map); 3640 } else { 3641 3642 /* 3643 * Get back to an exclusive (write) lock. 3644 */ 3645 3646 vm_map_lock(map); 3647 vm_map_unbusy(map); 3648 } 3649 3650 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3651 return 0; 3652 } 3653 3654 /* 3655 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 3656 * all mapped regions. 3657 * 3658 * => map must not be locked. 3659 * => if no flags are specified, all regions are unwired. 3660 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 3661 */ 3662 3663 int 3664 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 3665 { 3666 struct vm_map_entry *entry, *failed_entry; 3667 vsize_t size; 3668 int rv; 3669 #ifdef DIAGNOSTIC 3670 u_int timestamp_save; 3671 #endif 3672 UVMHIST_FUNC(__func__); 3673 UVMHIST_CALLARGS(maphist,"(map=%#jx,flags=%#jx)", (uintptr_t)map, flags, 3674 0, 0); 3675 3676 KASSERT(map->flags & VM_MAP_PAGEABLE); 3677 3678 vm_map_lock(map); 3679 3680 /* 3681 * handle wiring and unwiring separately. 3682 */ 3683 3684 if (flags == 0) { /* unwire */ 3685 3686 /* 3687 * POSIX 1003.1b -- munlockall unlocks all regions, 3688 * regardless of how many times mlockall has been called. 3689 */ 3690 3691 for (entry = map->header.next; entry != &map->header; 3692 entry = entry->next) { 3693 if (VM_MAPENT_ISWIRED(entry)) 3694 uvm_map_entry_unwire(map, entry); 3695 } 3696 map->flags &= ~VM_MAP_WIREFUTURE; 3697 vm_map_unlock(map); 3698 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 3699 return 0; 3700 } 3701 3702 if (flags & MCL_FUTURE) { 3703 3704 /* 3705 * must wire all future mappings; remember this. 3706 */ 3707 3708 map->flags |= VM_MAP_WIREFUTURE; 3709 } 3710 3711 if ((flags & MCL_CURRENT) == 0) { 3712 3713 /* 3714 * no more work to do! 3715 */ 3716 3717 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 3718 vm_map_unlock(map); 3719 return 0; 3720 } 3721 3722 /* 3723 * wire case: in three passes [XXXCDC: ugly block of code here] 3724 * 3725 * 1: holding the write lock, count all pages mapped by non-wired 3726 * entries. if this would cause us to go over our limit, we fail. 3727 * 3728 * 2: still holding the write lock, we create any anonymous maps that 3729 * need to be created. then we increment its wiring count. 3730 * 3731 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 3732 * in the pages for any newly wired area (wired_count == 1). 3733 * 3734 * downgrading to a read lock for uvm_fault_wire avoids a possible 3735 * deadlock with another thread that may have faulted on one of 3736 * the pages to be wired (it would mark the page busy, blocking 3737 * us, then in turn block on the map lock that we hold). because 3738 * of problems in the recursive lock package, we cannot upgrade 3739 * to a write lock in vm_map_lookup. thus, any actions that 3740 * require the write lock must be done beforehand. because we 3741 * keep the read lock on the map, the copy-on-write status of the 3742 * entries we modify here cannot change. 3743 */ 3744 3745 for (size = 0, entry = map->header.next; entry != &map->header; 3746 entry = entry->next) { 3747 if (entry->protection != VM_PROT_NONE && 3748 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3749 size += entry->end - entry->start; 3750 } 3751 } 3752 3753 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 3754 vm_map_unlock(map); 3755 return ENOMEM; 3756 } 3757 3758 if (limit != 0 && 3759 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 3760 vm_map_unlock(map); 3761 return ENOMEM; 3762 } 3763 3764 /* 3765 * Pass 2. 3766 */ 3767 3768 for (entry = map->header.next; entry != &map->header; 3769 entry = entry->next) { 3770 if (entry->protection == VM_PROT_NONE) 3771 continue; 3772 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 3773 3774 /* 3775 * perform actions of vm_map_lookup that need the 3776 * write lock on the map: create an anonymous map 3777 * for a copy-on-write region, or an anonymous map 3778 * for a zero-fill region. (XXXCDC: submap case 3779 * ok?) 3780 */ 3781 3782 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 3783 if (UVM_ET_ISNEEDSCOPY(entry) && 3784 ((entry->max_protection & VM_PROT_WRITE) || 3785 (entry->object.uvm_obj == NULL))) { 3786 amap_copy(map, entry, 0, entry->start, 3787 entry->end); 3788 /* XXXCDC: wait OK? */ 3789 } 3790 } 3791 } 3792 entry->wired_count++; 3793 } 3794 3795 /* 3796 * Pass 3. 3797 */ 3798 3799 #ifdef DIAGNOSTIC 3800 timestamp_save = map->timestamp; 3801 #endif 3802 vm_map_busy(map); 3803 vm_map_unlock(map); 3804 3805 rv = 0; 3806 for (entry = map->header.next; entry != &map->header; 3807 entry = entry->next) { 3808 if (entry->wired_count == 1) { 3809 rv = uvm_fault_wire(map, entry->start, entry->end, 3810 entry->max_protection, 1); 3811 if (rv) { 3812 3813 /* 3814 * wiring failed. break out of the loop. 3815 * we'll clean up the map below, once we 3816 * have a write lock again. 3817 */ 3818 3819 break; 3820 } 3821 } 3822 } 3823 3824 if (rv) { 3825 3826 /* 3827 * Get back an exclusive (write) lock. 3828 */ 3829 3830 vm_map_lock(map); 3831 vm_map_unbusy(map); 3832 3833 #ifdef DIAGNOSTIC 3834 if (timestamp_save + 1 != map->timestamp) 3835 panic("uvm_map_pageable_all: stale map"); 3836 #endif 3837 3838 /* 3839 * first drop the wiring count on all the entries 3840 * which haven't actually been wired yet. 3841 * 3842 * Skip VM_PROT_NONE entries like we did above. 3843 */ 3844 3845 failed_entry = entry; 3846 for (/* nothing */; entry != &map->header; 3847 entry = entry->next) { 3848 if (entry->protection == VM_PROT_NONE) 3849 continue; 3850 entry->wired_count--; 3851 } 3852 3853 /* 3854 * now, unwire all the entries that were successfully 3855 * wired above. 3856 * 3857 * Skip VM_PROT_NONE entries like we did above. 3858 */ 3859 3860 for (entry = map->header.next; entry != failed_entry; 3861 entry = entry->next) { 3862 if (entry->protection == VM_PROT_NONE) 3863 continue; 3864 entry->wired_count--; 3865 if (VM_MAPENT_ISWIRED(entry)) 3866 uvm_map_entry_unwire(map, entry); 3867 } 3868 vm_map_unlock(map); 3869 UVMHIST_LOG(maphist,"<- done (RV=%jd)", rv,0,0,0); 3870 return (rv); 3871 } 3872 3873 vm_map_unbusy(map); 3874 3875 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 3876 return 0; 3877 } 3878 3879 /* 3880 * uvm_map_clean: clean out a map range 3881 * 3882 * => valid flags: 3883 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 3884 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 3885 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 3886 * if (flags & PGO_FREE): any cached pages are freed after clean 3887 * => returns an error if any part of the specified range isn't mapped 3888 * => never a need to flush amap layer since the anonymous memory has 3889 * no permanent home, but may deactivate pages there 3890 * => called from sys_msync() and sys_madvise() 3891 * => caller must not write-lock map (read OK). 3892 * => we may sleep while cleaning if SYNCIO [with map read-locked] 3893 */ 3894 3895 int 3896 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 3897 { 3898 struct vm_map_entry *current, *entry; 3899 struct uvm_object *uobj; 3900 struct vm_amap *amap; 3901 struct vm_anon *anon; 3902 struct vm_page *pg; 3903 vaddr_t offset; 3904 vsize_t size; 3905 voff_t uoff; 3906 int error, refs; 3907 UVMHIST_FUNC(__func__); 3908 UVMHIST_CALLARGS(maphist,"(map=%#jx,start=%#jx,end=%#jx,flags=%#jx)", 3909 (uintptr_t)map, start, end, flags); 3910 3911 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 3912 (PGO_FREE|PGO_DEACTIVATE)); 3913 3914 vm_map_lock_read(map); 3915 VM_MAP_RANGE_CHECK(map, start, end); 3916 if (uvm_map_lookup_entry(map, start, &entry) == false) { 3917 vm_map_unlock_read(map); 3918 return EFAULT; 3919 } 3920 3921 /* 3922 * Make a first pass to check for holes and wiring problems. 3923 */ 3924 3925 for (current = entry; current->start < end; current = current->next) { 3926 if (UVM_ET_ISSUBMAP(current)) { 3927 vm_map_unlock_read(map); 3928 return EINVAL; 3929 } 3930 if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) { 3931 vm_map_unlock_read(map); 3932 return EBUSY; 3933 } 3934 if (end <= current->end) { 3935 break; 3936 } 3937 if (current->end != current->next->start) { 3938 vm_map_unlock_read(map); 3939 return EFAULT; 3940 } 3941 } 3942 3943 error = 0; 3944 for (current = entry; start < end; current = current->next) { 3945 amap = current->aref.ar_amap; /* upper layer */ 3946 uobj = current->object.uvm_obj; /* lower layer */ 3947 KASSERT(start >= current->start); 3948 3949 /* 3950 * No amap cleaning necessary if: 3951 * 3952 * (1) There's no amap. 3953 * 3954 * (2) We're not deactivating or freeing pages. 3955 */ 3956 3957 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 3958 goto flush_object; 3959 3960 offset = start - current->start; 3961 size = MIN(end, current->end) - start; 3962 3963 amap_lock(amap, RW_WRITER); 3964 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 3965 anon = amap_lookup(¤t->aref, offset); 3966 if (anon == NULL) 3967 continue; 3968 3969 KASSERT(anon->an_lock == amap->am_lock); 3970 pg = anon->an_page; 3971 if (pg == NULL) { 3972 continue; 3973 } 3974 if (pg->flags & PG_BUSY) { 3975 continue; 3976 } 3977 3978 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 3979 3980 /* 3981 * In these first 3 cases, we just deactivate the page. 3982 */ 3983 3984 case PGO_CLEANIT|PGO_FREE: 3985 case PGO_CLEANIT|PGO_DEACTIVATE: 3986 case PGO_DEACTIVATE: 3987 deactivate_it: 3988 /* 3989 * skip the page if it's loaned or wired, 3990 * since it shouldn't be on a paging queue 3991 * at all in these cases. 3992 */ 3993 3994 if (pg->loan_count != 0 || 3995 pg->wire_count != 0) { 3996 continue; 3997 } 3998 KASSERT(pg->uanon == anon); 3999 uvm_pagelock(pg); 4000 uvm_pagedeactivate(pg); 4001 uvm_pageunlock(pg); 4002 continue; 4003 4004 case PGO_FREE: 4005 4006 /* 4007 * If there are multiple references to 4008 * the amap, just deactivate the page. 4009 */ 4010 4011 if (amap_refs(amap) > 1) 4012 goto deactivate_it; 4013 4014 /* skip the page if it's wired */ 4015 if (pg->wire_count != 0) { 4016 continue; 4017 } 4018 amap_unadd(¤t->aref, offset); 4019 refs = --anon->an_ref; 4020 if (refs == 0) { 4021 uvm_anfree(anon); 4022 } 4023 continue; 4024 } 4025 } 4026 amap_unlock(amap); 4027 4028 flush_object: 4029 /* 4030 * flush pages if we've got a valid backing object. 4031 * note that we must always clean object pages before 4032 * freeing them since otherwise we could reveal stale 4033 * data from files. 4034 */ 4035 4036 uoff = current->offset + (start - current->start); 4037 size = MIN(end, current->end) - start; 4038 if (uobj != NULL) { 4039 rw_enter(uobj->vmobjlock, RW_WRITER); 4040 if (uobj->pgops->pgo_put != NULL) 4041 error = (uobj->pgops->pgo_put)(uobj, uoff, 4042 uoff + size, flags | PGO_CLEANIT); 4043 else 4044 error = 0; 4045 } 4046 start += size; 4047 } 4048 vm_map_unlock_read(map); 4049 return (error); 4050 } 4051 4052 4053 /* 4054 * uvm_map_checkprot: check protection in map 4055 * 4056 * => must allow specified protection in a fully allocated region. 4057 * => map must be read or write locked by caller. 4058 */ 4059 4060 bool 4061 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 4062 vm_prot_t protection) 4063 { 4064 struct vm_map_entry *entry; 4065 struct vm_map_entry *tmp_entry; 4066 4067 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 4068 return (false); 4069 } 4070 entry = tmp_entry; 4071 while (start < end) { 4072 if (entry == &map->header) { 4073 return (false); 4074 } 4075 4076 /* 4077 * no holes allowed 4078 */ 4079 4080 if (start < entry->start) { 4081 return (false); 4082 } 4083 4084 /* 4085 * check protection associated with entry 4086 */ 4087 4088 if ((entry->protection & protection) != protection) { 4089 return (false); 4090 } 4091 start = entry->end; 4092 entry = entry->next; 4093 } 4094 return (true); 4095 } 4096 4097 /* 4098 * uvmspace_alloc: allocate a vmspace structure. 4099 * 4100 * - structure includes vm_map and pmap 4101 * - XXX: no locking on this structure 4102 * - refcnt set to 1, rest must be init'd by caller 4103 */ 4104 struct vmspace * 4105 uvmspace_alloc(vaddr_t vmin, vaddr_t vmax, bool topdown) 4106 { 4107 struct vmspace *vm; 4108 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4109 4110 vm = pool_cache_get(&uvm_vmspace_cache, PR_WAITOK); 4111 uvmspace_init(vm, NULL, vmin, vmax, topdown); 4112 UVMHIST_LOG(maphist,"<- done (vm=%#jx)", (uintptr_t)vm, 0, 0, 0); 4113 return (vm); 4114 } 4115 4116 /* 4117 * uvmspace_init: initialize a vmspace structure. 4118 * 4119 * - XXX: no locking on this structure 4120 * - refcnt set to 1, rest must be init'd by caller 4121 */ 4122 void 4123 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, 4124 vaddr_t vmax, bool topdown) 4125 { 4126 UVMHIST_FUNC(__func__); 4127 UVMHIST_CALLARGS(maphist, "(vm=%#jx, pmap=%#jx, vmin=%#jx, vmax=%#jx", 4128 (uintptr_t)vm, (uintptr_t)pmap, vmin, vmax); 4129 UVMHIST_LOG(maphist, " topdown=%ju)", topdown, 0, 0, 0); 4130 4131 memset(vm, 0, sizeof(*vm)); 4132 uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE 4133 | (topdown ? VM_MAP_TOPDOWN : 0) 4134 ); 4135 if (pmap) 4136 pmap_reference(pmap); 4137 else 4138 pmap = pmap_create(); 4139 vm->vm_map.pmap = pmap; 4140 vm->vm_refcnt = 1; 4141 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4142 } 4143 4144 /* 4145 * uvmspace_share: share a vmspace between two processes 4146 * 4147 * - used for vfork, threads(?) 4148 */ 4149 4150 void 4151 uvmspace_share(struct proc *p1, struct proc *p2) 4152 { 4153 4154 uvmspace_addref(p1->p_vmspace); 4155 p2->p_vmspace = p1->p_vmspace; 4156 } 4157 4158 #if 0 4159 4160 /* 4161 * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace 4162 * 4163 * - XXX: no locking on vmspace 4164 */ 4165 4166 void 4167 uvmspace_unshare(struct lwp *l) 4168 { 4169 struct proc *p = l->l_proc; 4170 struct vmspace *nvm, *ovm = p->p_vmspace; 4171 4172 if (ovm->vm_refcnt == 1) 4173 /* nothing to do: vmspace isn't shared in the first place */ 4174 return; 4175 4176 /* make a new vmspace, still holding old one */ 4177 nvm = uvmspace_fork(ovm); 4178 4179 kpreempt_disable(); 4180 pmap_deactivate(l); /* unbind old vmspace */ 4181 p->p_vmspace = nvm; 4182 pmap_activate(l); /* switch to new vmspace */ 4183 kpreempt_enable(); 4184 4185 uvmspace_free(ovm); /* drop reference to old vmspace */ 4186 } 4187 4188 #endif 4189 4190 4191 /* 4192 * uvmspace_spawn: a new process has been spawned and needs a vmspace 4193 */ 4194 4195 void 4196 uvmspace_spawn(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown) 4197 { 4198 struct proc *p = l->l_proc; 4199 struct vmspace *nvm; 4200 4201 #ifdef __HAVE_CPU_VMSPACE_EXEC 4202 cpu_vmspace_exec(l, start, end); 4203 #endif 4204 4205 nvm = uvmspace_alloc(start, end, topdown); 4206 kpreempt_disable(); 4207 p->p_vmspace = nvm; 4208 pmap_activate(l); 4209 kpreempt_enable(); 4210 } 4211 4212 /* 4213 * uvmspace_exec: the process wants to exec a new program 4214 */ 4215 4216 void 4217 uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end, bool topdown) 4218 { 4219 struct proc *p = l->l_proc; 4220 struct vmspace *nvm, *ovm = p->p_vmspace; 4221 struct vm_map *map; 4222 int flags; 4223 4224 KASSERT(ovm != NULL); 4225 #ifdef __HAVE_CPU_VMSPACE_EXEC 4226 cpu_vmspace_exec(l, start, end); 4227 #endif 4228 4229 map = &ovm->vm_map; 4230 /* 4231 * see if more than one process is using this vmspace... 4232 */ 4233 4234 if (ovm->vm_refcnt == 1 4235 && topdown == ((ovm->vm_map.flags & VM_MAP_TOPDOWN) != 0)) { 4236 4237 /* 4238 * if p is the only process using its vmspace then we can safely 4239 * recycle that vmspace for the program that is being exec'd. 4240 * But only if TOPDOWN matches the requested value for the new 4241 * vm space! 4242 */ 4243 4244 /* 4245 * SYSV SHM semantics require us to kill all segments on an exec 4246 */ 4247 if (uvm_shmexit && ovm->vm_shm) 4248 (*uvm_shmexit)(ovm); 4249 4250 /* 4251 * POSIX 1003.1b -- "lock future mappings" is revoked 4252 * when a process execs another program image. 4253 */ 4254 4255 map->flags &= ~VM_MAP_WIREFUTURE; 4256 4257 /* 4258 * now unmap the old program. 4259 * 4260 * XXX set VM_MAP_DYING for the duration, so pmap_update() 4261 * is not called until the pmap has been totally cleared out 4262 * after pmap_remove_all(), or it can confuse some pmap 4263 * implementations. it would be nice to handle this by 4264 * deferring the pmap_update() while it is known the address 4265 * space is not visible to any user LWP other than curlwp, 4266 * but there isn't an elegant way of inferring that right 4267 * now. 4268 */ 4269 4270 flags = pmap_remove_all(map->pmap) ? UVM_FLAG_VAONLY : 0; 4271 map->flags |= VM_MAP_DYING; 4272 uvm_unmap1(map, vm_map_min(map), vm_map_max(map), flags); 4273 map->flags &= ~VM_MAP_DYING; 4274 pmap_update(map->pmap); 4275 KASSERT(map->header.prev == &map->header); 4276 KASSERT(map->nentries == 0); 4277 4278 /* 4279 * resize the map 4280 */ 4281 4282 vm_map_setmin(map, start); 4283 vm_map_setmax(map, end); 4284 } else { 4285 4286 /* 4287 * p's vmspace is being shared, so we can't reuse it for p since 4288 * it is still being used for others. allocate a new vmspace 4289 * for p 4290 */ 4291 4292 nvm = uvmspace_alloc(start, end, topdown); 4293 4294 /* 4295 * install new vmspace and drop our ref to the old one. 4296 */ 4297 4298 kpreempt_disable(); 4299 pmap_deactivate(l); 4300 p->p_vmspace = nvm; 4301 pmap_activate(l); 4302 kpreempt_enable(); 4303 4304 uvmspace_free(ovm); 4305 } 4306 } 4307 4308 /* 4309 * uvmspace_addref: add a reference to a vmspace. 4310 */ 4311 4312 void 4313 uvmspace_addref(struct vmspace *vm) 4314 { 4315 4316 KASSERT((vm->vm_map.flags & VM_MAP_DYING) == 0); 4317 KASSERT(vm->vm_refcnt > 0); 4318 atomic_inc_uint(&vm->vm_refcnt); 4319 } 4320 4321 /* 4322 * uvmspace_free: free a vmspace data structure 4323 */ 4324 4325 void 4326 uvmspace_free(struct vmspace *vm) 4327 { 4328 struct vm_map_entry *dead_entries; 4329 struct vm_map *map = &vm->vm_map; 4330 int flags; 4331 4332 UVMHIST_FUNC(__func__); 4333 UVMHIST_CALLARGS(maphist,"(vm=%#jx) ref=%jd", (uintptr_t)vm, 4334 vm->vm_refcnt, 0, 0); 4335 4336 membar_release(); 4337 if (atomic_dec_uint_nv(&vm->vm_refcnt) > 0) 4338 return; 4339 membar_acquire(); 4340 4341 /* 4342 * at this point, there should be no other references to the map. 4343 * delete all of the mappings, then destroy the pmap. 4344 */ 4345 4346 map->flags |= VM_MAP_DYING; 4347 flags = pmap_remove_all(map->pmap) ? UVM_FLAG_VAONLY : 0; 4348 4349 /* Get rid of any SYSV shared memory segments. */ 4350 if (uvm_shmexit && vm->vm_shm != NULL) 4351 (*uvm_shmexit)(vm); 4352 4353 if (map->nentries) { 4354 uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map), 4355 &dead_entries, flags); 4356 if (dead_entries != NULL) 4357 uvm_unmap_detach(dead_entries, 0); 4358 } 4359 KASSERT(map->nentries == 0); 4360 KASSERT(map->size == 0); 4361 4362 mutex_destroy(&map->misc_lock); 4363 rw_destroy(&map->lock); 4364 cv_destroy(&map->cv); 4365 pmap_destroy(map->pmap); 4366 pool_cache_put(&uvm_vmspace_cache, vm); 4367 } 4368 4369 static struct vm_map_entry * 4370 uvm_mapent_clone(struct vm_map *new_map, struct vm_map_entry *old_entry, 4371 int flags) 4372 { 4373 struct vm_map_entry *new_entry; 4374 4375 new_entry = uvm_mapent_alloc(new_map, 0); 4376 /* old_entry -> new_entry */ 4377 uvm_mapent_copy(old_entry, new_entry); 4378 4379 /* new pmap has nothing wired in it */ 4380 new_entry->wired_count = 0; 4381 4382 /* 4383 * gain reference to object backing the map (can't 4384 * be a submap, already checked this case). 4385 */ 4386 4387 if (new_entry->aref.ar_amap) 4388 uvm_map_reference_amap(new_entry, flags); 4389 4390 if (new_entry->object.uvm_obj && 4391 new_entry->object.uvm_obj->pgops->pgo_reference) 4392 new_entry->object.uvm_obj->pgops->pgo_reference( 4393 new_entry->object.uvm_obj); 4394 4395 /* insert entry at end of new_map's entry list */ 4396 uvm_map_entry_link(new_map, new_map->header.prev, 4397 new_entry); 4398 4399 return new_entry; 4400 } 4401 4402 /* 4403 * share the mapping: this means we want the old and 4404 * new entries to share amaps and backing objects. 4405 */ 4406 static void 4407 uvm_mapent_forkshared(struct vm_map *new_map, struct vm_map *old_map, 4408 struct vm_map_entry *old_entry) 4409 { 4410 /* 4411 * if the old_entry needs a new amap (due to prev fork) 4412 * then we need to allocate it now so that we have 4413 * something we own to share with the new_entry. [in 4414 * other words, we need to clear needs_copy] 4415 */ 4416 4417 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 4418 /* get our own amap, clears needs_copy */ 4419 amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK, 4420 0, 0); 4421 /* XXXCDC: WAITOK??? */ 4422 } 4423 4424 uvm_mapent_clone(new_map, old_entry, AMAP_SHARED); 4425 } 4426 4427 4428 static void 4429 uvm_mapent_forkcopy(struct vm_map *new_map, struct vm_map *old_map, 4430 struct vm_map_entry *old_entry) 4431 { 4432 struct vm_map_entry *new_entry; 4433 4434 /* 4435 * copy-on-write the mapping (using mmap's 4436 * MAP_PRIVATE semantics) 4437 * 4438 * allocate new_entry, adjust reference counts. 4439 * (note that new references are read-only). 4440 */ 4441 4442 new_entry = uvm_mapent_clone(new_map, old_entry, 0); 4443 4444 new_entry->etype |= 4445 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 4446 4447 /* 4448 * the new entry will need an amap. it will either 4449 * need to be copied from the old entry or created 4450 * from scratch (if the old entry does not have an 4451 * amap). can we defer this process until later 4452 * (by setting "needs_copy") or do we need to copy 4453 * the amap now? 4454 * 4455 * we must copy the amap now if any of the following 4456 * conditions hold: 4457 * 1. the old entry has an amap and that amap is 4458 * being shared. this means that the old (parent) 4459 * process is sharing the amap with another 4460 * process. if we do not clear needs_copy here 4461 * we will end up in a situation where both the 4462 * parent and child process are referring to the 4463 * same amap with "needs_copy" set. if the 4464 * parent write-faults, the fault routine will 4465 * clear "needs_copy" in the parent by allocating 4466 * a new amap. this is wrong because the 4467 * parent is supposed to be sharing the old amap 4468 * and the new amap will break that. 4469 * 4470 * 2. if the old entry has an amap and a non-zero 4471 * wire count then we are going to have to call 4472 * amap_cow_now to avoid page faults in the 4473 * parent process. since amap_cow_now requires 4474 * "needs_copy" to be clear we might as well 4475 * clear it here as well. 4476 * 4477 */ 4478 4479 if (old_entry->aref.ar_amap != NULL) { 4480 if ((amap_flags(old_entry->aref.ar_amap) & AMAP_SHARED) != 0 || 4481 VM_MAPENT_ISWIRED(old_entry)) { 4482 4483 amap_copy(new_map, new_entry, 4484 AMAP_COPY_NOCHUNK, 0, 0); 4485 /* XXXCDC: M_WAITOK ... ok? */ 4486 } 4487 } 4488 4489 /* 4490 * if the parent's entry is wired down, then the 4491 * parent process does not want page faults on 4492 * access to that memory. this means that we 4493 * cannot do copy-on-write because we can't write 4494 * protect the old entry. in this case we 4495 * resolve all copy-on-write faults now, using 4496 * amap_cow_now. note that we have already 4497 * allocated any needed amap (above). 4498 */ 4499 4500 if (VM_MAPENT_ISWIRED(old_entry)) { 4501 4502 /* 4503 * resolve all copy-on-write faults now 4504 * (note that there is nothing to do if 4505 * the old mapping does not have an amap). 4506 */ 4507 if (old_entry->aref.ar_amap) 4508 amap_cow_now(new_map, new_entry); 4509 4510 } else { 4511 /* 4512 * setup mappings to trigger copy-on-write faults 4513 * we must write-protect the parent if it has 4514 * an amap and it is not already "needs_copy"... 4515 * if it is already "needs_copy" then the parent 4516 * has already been write-protected by a previous 4517 * fork operation. 4518 */ 4519 if (old_entry->aref.ar_amap && 4520 !UVM_ET_ISNEEDSCOPY(old_entry)) { 4521 if (old_entry->max_protection & VM_PROT_WRITE) { 4522 #ifdef __HAVE_UNLOCKED_PMAP /* XXX temporary */ 4523 uvm_map_lock_entry(old_entry, RW_WRITER); 4524 #else 4525 uvm_map_lock_entry(old_entry, RW_READER); 4526 #endif 4527 pmap_protect(old_map->pmap, 4528 old_entry->start, old_entry->end, 4529 old_entry->protection & ~VM_PROT_WRITE); 4530 uvm_map_unlock_entry(old_entry); 4531 } 4532 old_entry->etype |= UVM_ET_NEEDSCOPY; 4533 } 4534 } 4535 } 4536 4537 /* 4538 * zero the mapping: the new entry will be zero initialized 4539 */ 4540 static void 4541 uvm_mapent_forkzero(struct vm_map *new_map, struct vm_map *old_map, 4542 struct vm_map_entry *old_entry) 4543 { 4544 struct vm_map_entry *new_entry; 4545 4546 new_entry = uvm_mapent_clone(new_map, old_entry, 0); 4547 4548 new_entry->etype |= 4549 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 4550 4551 if (new_entry->aref.ar_amap) { 4552 uvm_map_unreference_amap(new_entry, 0); 4553 new_entry->aref.ar_pageoff = 0; 4554 new_entry->aref.ar_amap = NULL; 4555 } 4556 4557 if (UVM_ET_ISOBJ(new_entry)) { 4558 if (new_entry->object.uvm_obj->pgops->pgo_detach) 4559 new_entry->object.uvm_obj->pgops->pgo_detach( 4560 new_entry->object.uvm_obj); 4561 new_entry->object.uvm_obj = NULL; 4562 new_entry->offset = 0; 4563 new_entry->etype &= ~UVM_ET_OBJ; 4564 } 4565 } 4566 4567 /* 4568 * F O R K - m a i n e n t r y p o i n t 4569 */ 4570 /* 4571 * uvmspace_fork: fork a process' main map 4572 * 4573 * => create a new vmspace for child process from parent. 4574 * => parent's map must not be locked. 4575 */ 4576 4577 struct vmspace * 4578 uvmspace_fork(struct vmspace *vm1) 4579 { 4580 struct vmspace *vm2; 4581 struct vm_map *old_map = &vm1->vm_map; 4582 struct vm_map *new_map; 4583 struct vm_map_entry *old_entry; 4584 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4585 4586 vm_map_lock(old_map); 4587 4588 vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), 4589 vm1->vm_map.flags & VM_MAP_TOPDOWN); 4590 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 4591 (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy); 4592 new_map = &vm2->vm_map; /* XXX */ 4593 4594 old_entry = old_map->header.next; 4595 new_map->size = old_map->size; 4596 4597 /* 4598 * go entry-by-entry 4599 */ 4600 4601 while (old_entry != &old_map->header) { 4602 4603 /* 4604 * first, some sanity checks on the old entry 4605 */ 4606 4607 KASSERT(!UVM_ET_ISSUBMAP(old_entry)); 4608 KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || 4609 !UVM_ET_ISNEEDSCOPY(old_entry)); 4610 4611 switch (old_entry->inheritance) { 4612 case MAP_INHERIT_NONE: 4613 /* 4614 * drop the mapping, modify size 4615 */ 4616 new_map->size -= old_entry->end - old_entry->start; 4617 break; 4618 4619 case MAP_INHERIT_SHARE: 4620 uvm_mapent_forkshared(new_map, old_map, old_entry); 4621 break; 4622 4623 case MAP_INHERIT_COPY: 4624 uvm_mapent_forkcopy(new_map, old_map, old_entry); 4625 break; 4626 4627 case MAP_INHERIT_ZERO: 4628 uvm_mapent_forkzero(new_map, old_map, old_entry); 4629 break; 4630 default: 4631 KASSERT(0); 4632 break; 4633 } 4634 old_entry = old_entry->next; 4635 } 4636 4637 pmap_update(old_map->pmap); 4638 vm_map_unlock(old_map); 4639 4640 if (uvm_shmfork && vm1->vm_shm) 4641 (*uvm_shmfork)(vm1, vm2); 4642 4643 #ifdef PMAP_FORK 4644 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 4645 #endif 4646 4647 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 4648 return (vm2); 4649 } 4650 4651 4652 /* 4653 * uvm_mapent_trymerge: try to merge an entry with its neighbors. 4654 * 4655 * => called with map locked. 4656 * => return non zero if successfully merged. 4657 */ 4658 4659 int 4660 uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags) 4661 { 4662 struct uvm_object *uobj; 4663 struct vm_map_entry *next; 4664 struct vm_map_entry *prev; 4665 vsize_t size; 4666 int merged = 0; 4667 bool copying; 4668 int newetype; 4669 4670 if (entry->aref.ar_amap != NULL) { 4671 return 0; 4672 } 4673 if ((entry->flags & UVM_MAP_NOMERGE) != 0) { 4674 return 0; 4675 } 4676 4677 uobj = entry->object.uvm_obj; 4678 size = entry->end - entry->start; 4679 copying = (flags & UVM_MERGE_COPYING) != 0; 4680 newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype; 4681 4682 next = entry->next; 4683 if (next != &map->header && 4684 next->start == entry->end && 4685 ((copying && next->aref.ar_amap != NULL && 4686 amap_refs(next->aref.ar_amap) == 1) || 4687 (!copying && next->aref.ar_amap == NULL)) && 4688 UVM_ET_ISCOMPATIBLE(next, newetype, 4689 uobj, entry->flags, entry->protection, 4690 entry->max_protection, entry->inheritance, entry->advice, 4691 entry->wired_count) && 4692 (uobj == NULL || entry->offset + size == next->offset)) { 4693 int error; 4694 4695 if (copying) { 4696 error = amap_extend(next, size, 4697 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS); 4698 } else { 4699 error = 0; 4700 } 4701 if (error == 0) { 4702 if (uobj) { 4703 if (uobj->pgops->pgo_detach) { 4704 uobj->pgops->pgo_detach(uobj); 4705 } 4706 } 4707 4708 entry->end = next->end; 4709 clear_hints(map, next); 4710 uvm_map_entry_unlink(map, next); 4711 if (copying) { 4712 entry->aref = next->aref; 4713 entry->etype &= ~UVM_ET_NEEDSCOPY; 4714 } 4715 uvm_map_check(map, "trymerge forwardmerge"); 4716 uvm_mapent_free(next); 4717 merged++; 4718 } 4719 } 4720 4721 prev = entry->prev; 4722 if (prev != &map->header && 4723 prev->end == entry->start && 4724 ((copying && !merged && prev->aref.ar_amap != NULL && 4725 amap_refs(prev->aref.ar_amap) == 1) || 4726 (!copying && prev->aref.ar_amap == NULL)) && 4727 UVM_ET_ISCOMPATIBLE(prev, newetype, 4728 uobj, entry->flags, entry->protection, 4729 entry->max_protection, entry->inheritance, entry->advice, 4730 entry->wired_count) && 4731 (uobj == NULL || 4732 prev->offset + prev->end - prev->start == entry->offset)) { 4733 int error; 4734 4735 if (copying) { 4736 error = amap_extend(prev, size, 4737 AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS); 4738 } else { 4739 error = 0; 4740 } 4741 if (error == 0) { 4742 if (uobj) { 4743 if (uobj->pgops->pgo_detach) { 4744 uobj->pgops->pgo_detach(uobj); 4745 } 4746 entry->offset = prev->offset; 4747 } 4748 4749 entry->start = prev->start; 4750 clear_hints(map, prev); 4751 uvm_map_entry_unlink(map, prev); 4752 if (copying) { 4753 entry->aref = prev->aref; 4754 entry->etype &= ~UVM_ET_NEEDSCOPY; 4755 } 4756 uvm_map_check(map, "trymerge backmerge"); 4757 uvm_mapent_free(prev); 4758 merged++; 4759 } 4760 } 4761 4762 return merged; 4763 } 4764 4765 /* 4766 * uvm_map_setup: init map 4767 * 4768 * => map must not be in service yet. 4769 */ 4770 4771 void 4772 uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags) 4773 { 4774 4775 rb_tree_init(&map->rb_tree, &uvm_map_tree_ops); 4776 map->header.next = map->header.prev = &map->header; 4777 map->nentries = 0; 4778 map->size = 0; 4779 map->ref_count = 1; 4780 vm_map_setmin(map, vmin); 4781 vm_map_setmax(map, vmax); 4782 map->flags = flags; 4783 map->first_free = &map->header; 4784 map->hint = &map->header; 4785 map->timestamp = 0; 4786 map->busy = NULL; 4787 4788 rw_init(&map->lock); 4789 cv_init(&map->cv, "vm_map"); 4790 mutex_init(&map->misc_lock, MUTEX_DRIVER, IPL_NONE); 4791 } 4792 4793 /* 4794 * U N M A P - m a i n e n t r y p o i n t 4795 */ 4796 4797 /* 4798 * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop") 4799 * 4800 * => caller must check alignment and size 4801 * => map must be unlocked (we will lock it) 4802 * => flags is UVM_FLAG_QUANTUM or 0. 4803 */ 4804 4805 void 4806 uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 4807 { 4808 struct vm_map_entry *dead_entries; 4809 UVMHIST_FUNC(__func__); 4810 UVMHIST_CALLARGS(maphist, " (map=%#jx, start=%#jx, end=%#jx)", 4811 (uintptr_t)map, start, end, 0); 4812 4813 KASSERTMSG(start < end, 4814 "%s: map %p: start %#jx < end %#jx", __func__, map, 4815 (uintmax_t)start, (uintmax_t)end); 4816 if (map == kernel_map) { 4817 LOCKDEBUG_MEM_CHECK((void *)start, end - start); 4818 } 4819 4820 /* 4821 * work now done by helper functions. wipe the pmap's and then 4822 * detach from the dead entries... 4823 */ 4824 vm_map_lock(map); 4825 uvm_unmap_remove(map, start, end, &dead_entries, flags); 4826 vm_map_unlock(map); 4827 4828 if (dead_entries != NULL) 4829 uvm_unmap_detach(dead_entries, 0); 4830 4831 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 4832 } 4833 4834 4835 /* 4836 * uvm_map_reference: add reference to a map 4837 * 4838 * => map need not be locked 4839 */ 4840 4841 void 4842 uvm_map_reference(struct vm_map *map) 4843 { 4844 4845 atomic_inc_uint(&map->ref_count); 4846 } 4847 4848 void 4849 uvm_map_lock_entry(struct vm_map_entry *entry, krw_t op) 4850 { 4851 4852 if (entry->aref.ar_amap != NULL) { 4853 amap_lock(entry->aref.ar_amap, op); 4854 } 4855 if (UVM_ET_ISOBJ(entry)) { 4856 rw_enter(entry->object.uvm_obj->vmobjlock, op); 4857 } 4858 } 4859 4860 void 4861 uvm_map_unlock_entry(struct vm_map_entry *entry) 4862 { 4863 4864 if (UVM_ET_ISOBJ(entry)) { 4865 rw_exit(entry->object.uvm_obj->vmobjlock); 4866 } 4867 if (entry->aref.ar_amap != NULL) { 4868 amap_unlock(entry->aref.ar_amap); 4869 } 4870 } 4871 4872 #define UVM_VOADDR_TYPE_MASK 0x3UL 4873 #define UVM_VOADDR_TYPE_UOBJ 0x1UL 4874 #define UVM_VOADDR_TYPE_ANON 0x2UL 4875 #define UVM_VOADDR_OBJECT_MASK ~UVM_VOADDR_TYPE_MASK 4876 4877 #define UVM_VOADDR_GET_TYPE(voa) \ 4878 ((voa)->object & UVM_VOADDR_TYPE_MASK) 4879 #define UVM_VOADDR_GET_OBJECT(voa) \ 4880 ((voa)->object & UVM_VOADDR_OBJECT_MASK) 4881 #define UVM_VOADDR_SET_OBJECT(voa, obj, type) \ 4882 do { \ 4883 KASSERT(((uintptr_t)(obj) & UVM_VOADDR_TYPE_MASK) == 0); \ 4884 (voa)->object = ((uintptr_t)(obj)) | (type); \ 4885 } while (/*CONSTCOND*/0) 4886 4887 #define UVM_VOADDR_GET_UOBJ(voa) \ 4888 ((struct uvm_object *)UVM_VOADDR_GET_OBJECT(voa)) 4889 #define UVM_VOADDR_SET_UOBJ(voa, uobj) \ 4890 UVM_VOADDR_SET_OBJECT(voa, uobj, UVM_VOADDR_TYPE_UOBJ) 4891 4892 #define UVM_VOADDR_GET_ANON(voa) \ 4893 ((struct vm_anon *)UVM_VOADDR_GET_OBJECT(voa)) 4894 #define UVM_VOADDR_SET_ANON(voa, anon) \ 4895 UVM_VOADDR_SET_OBJECT(voa, anon, UVM_VOADDR_TYPE_ANON) 4896 4897 /* 4898 * uvm_voaddr_acquire: returns the virtual object address corresponding 4899 * to the specified virtual address. 4900 * 4901 * => resolves COW so the true page identity is tracked. 4902 * 4903 * => acquires a reference on the page's owner (uvm_object or vm_anon) 4904 */ 4905 bool 4906 uvm_voaddr_acquire(struct vm_map * const map, vaddr_t const va, 4907 struct uvm_voaddr * const voaddr) 4908 { 4909 struct vm_map_entry *entry; 4910 struct vm_anon *anon = NULL; 4911 bool result = false; 4912 bool exclusive = false; 4913 void (*unlock_fn)(struct vm_map *); 4914 4915 UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist); 4916 UVMHIST_LOG(maphist,"(map=%#jx,va=%#jx)", (uintptr_t)map, va, 0, 0); 4917 4918 const vaddr_t start = trunc_page(va); 4919 const vaddr_t end = round_page(va+1); 4920 4921 lookup_again: 4922 if (__predict_false(exclusive)) { 4923 vm_map_lock(map); 4924 unlock_fn = vm_map_unlock; 4925 } else { 4926 vm_map_lock_read(map); 4927 unlock_fn = vm_map_unlock_read; 4928 } 4929 4930 if (__predict_false(!uvm_map_lookup_entry(map, start, &entry))) { 4931 unlock_fn(map); 4932 UVMHIST_LOG(maphist,"<- done (no entry)",0,0,0,0); 4933 return false; 4934 } 4935 4936 if (__predict_false(entry->protection == VM_PROT_NONE)) { 4937 unlock_fn(map); 4938 UVMHIST_LOG(maphist,"<- done (PROT_NONE)",0,0,0,0); 4939 return false; 4940 } 4941 4942 /* 4943 * We have a fast path for the common case of "no COW resolution 4944 * needed" whereby we have taken a read lock on the map and if 4945 * we don't encounter any need to create a vm_anon then great! 4946 * But if we do, we loop around again, instead taking an exclusive 4947 * lock so that we can perform the fault. 4948 * 4949 * In the event that we have to resolve the fault, we do nearly the 4950 * same work as uvm_map_pageable() does: 4951 * 4952 * 1: holding the write lock, we create any anonymous maps that need 4953 * to be created. however, we do NOT need to clip the map entries 4954 * in this case. 4955 * 4956 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 4957 * in the page (assuming the entry is not already wired). this 4958 * is done because we need the vm_anon to be present. 4959 */ 4960 if (__predict_true(!VM_MAPENT_ISWIRED(entry))) { 4961 4962 bool need_fault = false; 4963 4964 /* 4965 * perform the action of vm_map_lookup that need the 4966 * write lock on the map: create an anonymous map for 4967 * a copy-on-write region, or an anonymous map for 4968 * a zero-fill region. 4969 */ 4970 if (__predict_false(UVM_ET_ISSUBMAP(entry))) { 4971 unlock_fn(map); 4972 UVMHIST_LOG(maphist,"<- done (submap)",0,0,0,0); 4973 return false; 4974 } 4975 if (__predict_false(UVM_ET_ISNEEDSCOPY(entry) && 4976 ((entry->max_protection & VM_PROT_WRITE) || 4977 (entry->object.uvm_obj == NULL)))) { 4978 if (!exclusive) { 4979 /* need to take the slow path */ 4980 KASSERT(unlock_fn == vm_map_unlock_read); 4981 vm_map_unlock_read(map); 4982 exclusive = true; 4983 goto lookup_again; 4984 } 4985 need_fault = true; 4986 amap_copy(map, entry, 0, start, end); 4987 /* XXXCDC: wait OK? */ 4988 } 4989 4990 /* 4991 * do a quick check to see if the fault has already 4992 * been resolved to the upper layer. 4993 */ 4994 if (__predict_true(entry->aref.ar_amap != NULL && 4995 need_fault == false)) { 4996 amap_lock(entry->aref.ar_amap, RW_WRITER); 4997 anon = amap_lookup(&entry->aref, start - entry->start); 4998 if (__predict_true(anon != NULL)) { 4999 /* amap unlocked below */ 5000 goto found_anon; 5001 } 5002 amap_unlock(entry->aref.ar_amap); 5003 need_fault = true; 5004 } 5005 5006 /* 5007 * we predict this test as false because if we reach 5008 * this point, then we are likely dealing with a 5009 * shared memory region backed by a uvm_object, in 5010 * which case a fault to create the vm_anon is not 5011 * necessary. 5012 */ 5013 if (__predict_false(need_fault)) { 5014 if (exclusive) { 5015 vm_map_busy(map); 5016 vm_map_unlock(map); 5017 unlock_fn = vm_map_unbusy; 5018 } 5019 5020 if (uvm_fault_wire(map, start, end, 5021 entry->max_protection, 1)) { 5022 /* wiring failed */ 5023 unlock_fn(map); 5024 UVMHIST_LOG(maphist,"<- done (wire failed)", 5025 0,0,0,0); 5026 return false; 5027 } 5028 5029 /* 5030 * now that we have resolved the fault, we can unwire 5031 * the page. 5032 */ 5033 if (exclusive) { 5034 vm_map_lock(map); 5035 vm_map_unbusy(map); 5036 unlock_fn = vm_map_unlock; 5037 } 5038 5039 uvm_fault_unwire_locked(map, start, end); 5040 } 5041 } 5042 5043 /* check the upper layer */ 5044 if (entry->aref.ar_amap) { 5045 amap_lock(entry->aref.ar_amap, RW_WRITER); 5046 anon = amap_lookup(&entry->aref, start - entry->start); 5047 if (anon) { 5048 found_anon: KASSERT(anon->an_lock == entry->aref.ar_amap->am_lock); 5049 anon->an_ref++; 5050 rw_obj_hold(anon->an_lock); 5051 KASSERT(anon->an_ref != 0); 5052 UVM_VOADDR_SET_ANON(voaddr, anon); 5053 voaddr->offset = va & PAGE_MASK; 5054 result = true; 5055 } 5056 amap_unlock(entry->aref.ar_amap); 5057 } 5058 5059 /* check the lower layer */ 5060 if (!result && UVM_ET_ISOBJ(entry)) { 5061 struct uvm_object *uobj = entry->object.uvm_obj; 5062 5063 KASSERT(uobj != NULL); 5064 (*uobj->pgops->pgo_reference)(uobj); 5065 UVM_VOADDR_SET_UOBJ(voaddr, uobj); 5066 voaddr->offset = entry->offset + (va - entry->start); 5067 result = true; 5068 } 5069 5070 unlock_fn(map); 5071 5072 if (result) { 5073 UVMHIST_LOG(maphist, 5074 "<- done OK (type=%jd,owner=%#jx,offset=%#jx)", 5075 UVM_VOADDR_GET_TYPE(voaddr), 5076 UVM_VOADDR_GET_OBJECT(voaddr), 5077 voaddr->offset, 0); 5078 } else { 5079 UVMHIST_LOG(maphist,"<- done (failed)",0,0,0,0); 5080 } 5081 5082 return result; 5083 } 5084 5085 /* 5086 * uvm_voaddr_release: release the references held by the 5087 * vitual object address. 5088 */ 5089 void 5090 uvm_voaddr_release(struct uvm_voaddr * const voaddr) 5091 { 5092 5093 switch (UVM_VOADDR_GET_TYPE(voaddr)) { 5094 case UVM_VOADDR_TYPE_UOBJ: { 5095 struct uvm_object * const uobj = UVM_VOADDR_GET_UOBJ(voaddr); 5096 5097 KASSERT(uobj != NULL); 5098 KASSERT(uobj->pgops->pgo_detach != NULL); 5099 (*uobj->pgops->pgo_detach)(uobj); 5100 break; 5101 } 5102 case UVM_VOADDR_TYPE_ANON: { 5103 struct vm_anon * const anon = UVM_VOADDR_GET_ANON(voaddr); 5104 krwlock_t *lock; 5105 5106 KASSERT(anon != NULL); 5107 rw_enter((lock = anon->an_lock), RW_WRITER); 5108 KASSERT(anon->an_ref > 0); 5109 if (--anon->an_ref == 0) { 5110 uvm_anfree(anon); 5111 } 5112 rw_exit(lock); 5113 rw_obj_free(lock); 5114 break; 5115 } 5116 default: 5117 panic("uvm_voaddr_release: bad type"); 5118 } 5119 memset(voaddr, 0, sizeof(*voaddr)); 5120 } 5121 5122 /* 5123 * uvm_voaddr_compare: compare two uvm_voaddr objects. 5124 * 5125 * => memcmp() semantics 5126 */ 5127 int 5128 uvm_voaddr_compare(const struct uvm_voaddr * const voaddr1, 5129 const struct uvm_voaddr * const voaddr2) 5130 { 5131 const uintptr_t type1 = UVM_VOADDR_GET_TYPE(voaddr1); 5132 const uintptr_t type2 = UVM_VOADDR_GET_TYPE(voaddr2); 5133 5134 KASSERT(type1 == UVM_VOADDR_TYPE_UOBJ || 5135 type1 == UVM_VOADDR_TYPE_ANON); 5136 5137 KASSERT(type2 == UVM_VOADDR_TYPE_UOBJ || 5138 type2 == UVM_VOADDR_TYPE_ANON); 5139 5140 if (type1 < type2) 5141 return -1; 5142 if (type1 > type2) 5143 return 1; 5144 5145 const uintptr_t addr1 = UVM_VOADDR_GET_OBJECT(voaddr1); 5146 const uintptr_t addr2 = UVM_VOADDR_GET_OBJECT(voaddr2); 5147 5148 if (addr1 < addr2) 5149 return -1; 5150 if (addr1 > addr2) 5151 return 1; 5152 5153 if (voaddr1->offset < voaddr2->offset) 5154 return -1; 5155 if (voaddr1->offset > voaddr2->offset) 5156 return 1; 5157 5158 return 0; 5159 } 5160 5161 #if defined(DDB) || defined(DEBUGPRINT) 5162 5163 /* 5164 * uvm_map_printit: actually prints the map 5165 */ 5166 5167 void 5168 uvm_map_printit(struct vm_map *map, bool full, 5169 void (*pr)(const char *, ...)) 5170 { 5171 struct vm_map_entry *entry; 5172 5173 (*pr)("MAP %p: [%#lx->%#lx]\n", map, vm_map_min(map), 5174 vm_map_max(map)); 5175 (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=%#x\n", 5176 map->nentries, map->size, map->ref_count, map->timestamp, 5177 map->flags); 5178 (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap, 5179 pmap_resident_count(map->pmap), pmap_wired_count(map->pmap)); 5180 if (!full) 5181 return; 5182 for (entry = map->header.next; entry != &map->header; 5183 entry = entry->next) { 5184 (*pr)(" - %p: %#lx->%#lx: obj=%p/%#llx, amap=%p/%d\n", 5185 entry, entry->start, entry->end, entry->object.uvm_obj, 5186 (long long)entry->offset, entry->aref.ar_amap, 5187 entry->aref.ar_pageoff); 5188 (*pr)( 5189 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 5190 "wc=%d, adv=%d%s\n", 5191 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 5192 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 5193 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 5194 entry->protection, entry->max_protection, 5195 entry->inheritance, entry->wired_count, entry->advice, 5196 entry == map->first_free ? " (first_free)" : ""); 5197 } 5198 } 5199 5200 void 5201 uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...)) 5202 { 5203 struct vm_map *map; 5204 5205 for (map = kernel_map;;) { 5206 struct vm_map_entry *entry; 5207 5208 if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) { 5209 break; 5210 } 5211 (*pr)("%p is %p+%zu from VMMAP %p\n", 5212 (void *)addr, (void *)entry->start, 5213 (size_t)(addr - (uintptr_t)entry->start), map); 5214 if (!UVM_ET_ISSUBMAP(entry)) { 5215 break; 5216 } 5217 map = entry->object.sub_map; 5218 } 5219 } 5220 5221 #endif /* DDB || DEBUGPRINT */ 5222 5223 #ifndef __USER_VA0_IS_SAFE 5224 static int 5225 sysctl_user_va0_disable(SYSCTLFN_ARGS) 5226 { 5227 struct sysctlnode node; 5228 int t, error; 5229 5230 node = *rnode; 5231 node.sysctl_data = &t; 5232 t = user_va0_disable; 5233 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 5234 if (error || newp == NULL) 5235 return (error); 5236 5237 if (!t && user_va0_disable && 5238 kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MAP_VA_ZERO, 0, 5239 NULL, NULL, NULL)) 5240 return EPERM; 5241 5242 user_va0_disable = !!t; 5243 return 0; 5244 } 5245 #endif 5246 5247 static int 5248 fill_vmentry(struct lwp *l, struct proc *p, struct kinfo_vmentry *kve, 5249 struct vm_map *m, struct vm_map_entry *e) 5250 { 5251 #ifndef _RUMPKERNEL 5252 int error; 5253 5254 memset(kve, 0, sizeof(*kve)); 5255 KASSERT(e != NULL); 5256 if (UVM_ET_ISOBJ(e)) { 5257 struct uvm_object *uobj = e->object.uvm_obj; 5258 KASSERT(uobj != NULL); 5259 kve->kve_ref_count = uobj->uo_refs; 5260 kve->kve_count = uobj->uo_npages; 5261 if (UVM_OBJ_IS_VNODE(uobj)) { 5262 struct vattr va; 5263 struct vnode *vp = (struct vnode *)uobj; 5264 vn_lock(vp, LK_SHARED | LK_RETRY); 5265 error = VOP_GETATTR(vp, &va, l->l_cred); 5266 VOP_UNLOCK(vp); 5267 kve->kve_type = KVME_TYPE_VNODE; 5268 if (error == 0) { 5269 kve->kve_vn_size = vp->v_size; 5270 kve->kve_vn_type = (int)vp->v_type; 5271 kve->kve_vn_mode = va.va_mode; 5272 kve->kve_vn_rdev = va.va_rdev; 5273 kve->kve_vn_fileid = va.va_fileid; 5274 kve->kve_vn_fsid = va.va_fsid; 5275 error = vnode_to_path(kve->kve_path, 5276 sizeof(kve->kve_path) / 2, vp, l, p); 5277 } 5278 } else if (UVM_OBJ_IS_KERN_OBJECT(uobj)) { 5279 kve->kve_type = KVME_TYPE_KERN; 5280 } else if (UVM_OBJ_IS_DEVICE(uobj)) { 5281 kve->kve_type = KVME_TYPE_DEVICE; 5282 } else if (UVM_OBJ_IS_AOBJ(uobj)) { 5283 kve->kve_type = KVME_TYPE_ANON; 5284 } else { 5285 kve->kve_type = KVME_TYPE_OBJECT; 5286 } 5287 } else if (UVM_ET_ISSUBMAP(e)) { 5288 struct vm_map *map = e->object.sub_map; 5289 KASSERT(map != NULL); 5290 kve->kve_ref_count = map->ref_count; 5291 kve->kve_count = map->nentries; 5292 kve->kve_type = KVME_TYPE_SUBMAP; 5293 } else 5294 kve->kve_type = KVME_TYPE_UNKNOWN; 5295 5296 kve->kve_start = e->start; 5297 kve->kve_end = e->end; 5298 kve->kve_offset = e->offset; 5299 kve->kve_wired_count = e->wired_count; 5300 kve->kve_inheritance = e->inheritance; 5301 kve->kve_attributes = 0; /* unused */ 5302 kve->kve_advice = e->advice; 5303 #define PROT(p) (((p) & VM_PROT_READ) ? KVME_PROT_READ : 0) | \ 5304 (((p) & VM_PROT_WRITE) ? KVME_PROT_WRITE : 0) | \ 5305 (((p) & VM_PROT_EXECUTE) ? KVME_PROT_EXEC : 0) 5306 kve->kve_protection = PROT(e->protection); 5307 kve->kve_max_protection = PROT(e->max_protection); 5308 kve->kve_flags |= (e->etype & UVM_ET_COPYONWRITE) 5309 ? KVME_FLAG_COW : 0; 5310 kve->kve_flags |= (e->etype & UVM_ET_NEEDSCOPY) 5311 ? KVME_FLAG_NEEDS_COPY : 0; 5312 kve->kve_flags |= (m->flags & VM_MAP_TOPDOWN) 5313 ? KVME_FLAG_GROWS_DOWN : KVME_FLAG_GROWS_UP; 5314 kve->kve_flags |= (m->flags & VM_MAP_PAGEABLE) 5315 ? KVME_FLAG_PAGEABLE : 0; 5316 #endif 5317 return 0; 5318 } 5319 5320 static int 5321 fill_vmentries(struct lwp *l, pid_t pid, u_int elem_size, void *oldp, 5322 size_t *oldlenp) 5323 { 5324 int error; 5325 struct proc *p; 5326 struct kinfo_vmentry *vme; 5327 struct vmspace *vm; 5328 struct vm_map *map; 5329 struct vm_map_entry *entry; 5330 char *dp; 5331 size_t count, vmesize; 5332 5333 if (elem_size == 0 || elem_size > 2 * sizeof(*vme)) 5334 return EINVAL; 5335 5336 if (oldp) { 5337 if (*oldlenp > 10UL * 1024UL * 1024UL) 5338 return E2BIG; 5339 count = *oldlenp / elem_size; 5340 if (count == 0) 5341 return ENOMEM; 5342 vmesize = count * sizeof(*vme); 5343 } else 5344 vmesize = 0; 5345 5346 if ((error = proc_find_locked(l, &p, pid)) != 0) 5347 return error; 5348 5349 vme = NULL; 5350 count = 0; 5351 5352 if ((error = proc_vmspace_getref(p, &vm)) != 0) 5353 goto out; 5354 5355 map = &vm->vm_map; 5356 vm_map_lock_read(map); 5357 5358 dp = oldp; 5359 if (oldp) 5360 vme = kmem_alloc(vmesize, KM_SLEEP); 5361 for (entry = map->header.next; entry != &map->header; 5362 entry = entry->next) { 5363 if (oldp && (dp - (char *)oldp) < vmesize) { 5364 error = fill_vmentry(l, p, &vme[count], map, entry); 5365 if (error) 5366 goto out; 5367 dp += elem_size; 5368 } 5369 count++; 5370 } 5371 vm_map_unlock_read(map); 5372 uvmspace_free(vm); 5373 5374 out: 5375 if (pid != -1) 5376 mutex_exit(p->p_lock); 5377 if (error == 0) { 5378 const u_int esize = uimin(sizeof(*vme), elem_size); 5379 dp = oldp; 5380 for (size_t i = 0; i < count; i++) { 5381 if (oldp && (dp - (char *)oldp) < vmesize) { 5382 error = sysctl_copyout(l, &vme[i], dp, esize); 5383 if (error) 5384 break; 5385 dp += elem_size; 5386 } else 5387 break; 5388 } 5389 count *= elem_size; 5390 if (oldp != NULL && *oldlenp < count) 5391 error = ENOSPC; 5392 *oldlenp = count; 5393 } 5394 if (vme) 5395 kmem_free(vme, vmesize); 5396 return error; 5397 } 5398 5399 static int 5400 sysctl_vmproc(SYSCTLFN_ARGS) 5401 { 5402 int error; 5403 5404 if (namelen == 1 && name[0] == CTL_QUERY) 5405 return (sysctl_query(SYSCTLFN_CALL(rnode))); 5406 5407 if (namelen == 0) 5408 return EINVAL; 5409 5410 switch (name[0]) { 5411 case VM_PROC_MAP: 5412 if (namelen != 3) 5413 return EINVAL; 5414 sysctl_unlock(); 5415 error = fill_vmentries(l, name[1], name[2], oldp, oldlenp); 5416 sysctl_relock(); 5417 return error; 5418 default: 5419 return EINVAL; 5420 } 5421 } 5422 5423 SYSCTL_SETUP(sysctl_uvmmap_setup, "sysctl uvmmap setup") 5424 { 5425 5426 sysctl_createv(clog, 0, NULL, NULL, 5427 CTLFLAG_PERMANENT, 5428 CTLTYPE_STRUCT, "proc", 5429 SYSCTL_DESCR("Process vm information"), 5430 sysctl_vmproc, 0, NULL, 0, 5431 CTL_VM, VM_PROC, CTL_EOL); 5432 #ifndef __USER_VA0_IS_SAFE 5433 sysctl_createv(clog, 0, NULL, NULL, 5434 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 5435 CTLTYPE_INT, "user_va0_disable", 5436 SYSCTL_DESCR("Disable VA 0"), 5437 sysctl_user_va0_disable, 0, &user_va0_disable, 0, 5438 CTL_VM, CTL_CREATE, CTL_EOL); 5439 #endif 5440 } 5441