1 /* $OpenBSD: uvm_map.c,v 1.109 2009/03/25 20:00:18 oga Exp $ */ 2 /* $NetBSD: uvm_map.c,v 1.86 2000/11/27 08:40:03 chs Exp $ */ 3 4 /* 5 * Copyright (c) 1997 Charles D. Cranor and Washington University. 6 * Copyright (c) 1991, 1993, The Regents of the University of California. 7 * 8 * All rights reserved. 9 * 10 * This code is derived from software contributed to Berkeley by 11 * The Mach Operating System project at Carnegie-Mellon University. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by Charles D. Cranor, 24 * Washington University, the University of California, Berkeley and 25 * its contributors. 26 * 4. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94 43 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp 44 * 45 * 46 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 47 * All rights reserved. 48 * 49 * Permission to use, copy, modify and distribute this software and 50 * its documentation is hereby granted, provided that both the copyright 51 * notice and this permission notice appear in all copies of the 52 * software, derivative works or modified versions, and any portions 53 * thereof, and that both notices appear in supporting documentation. 54 * 55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 58 * 59 * Carnegie Mellon requests users of this software to return to 60 * 61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 62 * School of Computer Science 63 * Carnegie Mellon University 64 * Pittsburgh PA 15213-3890 65 * 66 * any improvements or extensions that they make and grant Carnegie the 67 * rights to redistribute these changes. 68 */ 69 70 /* 71 * uvm_map.c: uvm map operations 72 */ 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/mman.h> 77 #include <sys/proc.h> 78 #include <sys/malloc.h> 79 #include <sys/pool.h> 80 #include <sys/kernel.h> 81 82 #include <dev/rndvar.h> 83 84 #ifdef SYSVSHM 85 #include <sys/shm.h> 86 #endif 87 88 #include <uvm/uvm.h> 89 #undef RB_AUGMENT 90 #define RB_AUGMENT(x) uvm_rb_augment(x) 91 92 #ifdef DDB 93 #include <uvm/uvm_ddb.h> 94 #endif 95 96 static struct timeval uvm_kmapent_last_warn_time; 97 static struct timeval uvm_kmapent_warn_rate = { 10, 0 }; 98 99 struct uvm_cnt uvm_map_call, map_backmerge, map_forwmerge; 100 struct uvm_cnt map_nousermerge; 101 struct uvm_cnt uvm_mlk_call, uvm_mlk_hint; 102 const char vmmapbsy[] = "vmmapbsy"; 103 104 /* 105 * Da history books 106 */ 107 UVMHIST_DECL(maphist); 108 UVMHIST_DECL(pdhist); 109 110 /* 111 * pool for vmspace structures. 112 */ 113 114 struct pool uvm_vmspace_pool; 115 116 /* 117 * pool for dynamically-allocated map entries. 118 */ 119 120 struct pool uvm_map_entry_pool; 121 struct pool uvm_map_entry_kmem_pool; 122 123 #ifdef PMAP_GROWKERNEL 124 /* 125 * This global represents the end of the kernel virtual address 126 * space. If we want to exceed this, we must grow the kernel 127 * virtual address space dynamically. 128 * 129 * Note, this variable is locked by kernel_map's lock. 130 */ 131 vaddr_t uvm_maxkaddr; 132 #endif 133 134 /* 135 * macros 136 */ 137 138 /* 139 * uvm_map_entry_link: insert entry into a map 140 * 141 * => map must be locked 142 */ 143 #define uvm_map_entry_link(map, after_where, entry) do { \ 144 (map)->nentries++; \ 145 (entry)->prev = (after_where); \ 146 (entry)->next = (after_where)->next; \ 147 (entry)->prev->next = (entry); \ 148 (entry)->next->prev = (entry); \ 149 uvm_rb_insert(map, entry); \ 150 } while (0) 151 152 /* 153 * uvm_map_entry_unlink: remove entry from a map 154 * 155 * => map must be locked 156 */ 157 #define uvm_map_entry_unlink(map, entry) do { \ 158 (map)->nentries--; \ 159 (entry)->next->prev = (entry)->prev; \ 160 (entry)->prev->next = (entry)->next; \ 161 uvm_rb_remove(map, entry); \ 162 } while (0) 163 164 /* 165 * SAVE_HINT: saves the specified entry as the hint for future lookups. 166 * 167 * => map need not be locked (protected by hint_lock). 168 */ 169 #define SAVE_HINT(map,check,value) do { \ 170 simple_lock(&(map)->hint_lock); \ 171 if ((map)->hint == (check)) \ 172 (map)->hint = (value); \ 173 simple_unlock(&(map)->hint_lock); \ 174 } while (0) 175 176 /* 177 * VM_MAP_RANGE_CHECK: check and correct range 178 * 179 * => map must at least be read locked 180 */ 181 182 #define VM_MAP_RANGE_CHECK(map, start, end) do { \ 183 if (start < vm_map_min(map)) \ 184 start = vm_map_min(map); \ 185 if (end > vm_map_max(map)) \ 186 end = vm_map_max(map); \ 187 if (start > end) \ 188 start = end; \ 189 } while (0) 190 191 /* 192 * local prototypes 193 */ 194 195 void uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *); 196 void uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *); 197 void uvm_map_reference_amap(struct vm_map_entry *, int); 198 void uvm_map_unreference_amap(struct vm_map_entry *, int); 199 int uvm_map_spacefits(struct vm_map *, vaddr_t *, vsize_t, 200 struct vm_map_entry *, voff_t, vsize_t); 201 202 struct vm_map_entry *uvm_mapent_alloc(struct vm_map *); 203 void uvm_mapent_free(struct vm_map_entry *); 204 205 206 /* 207 * Tree manipulation. 208 */ 209 void uvm_rb_insert(struct vm_map *, struct vm_map_entry *); 210 void uvm_rb_remove(struct vm_map *, struct vm_map_entry *); 211 vsize_t uvm_rb_space(struct vm_map *, struct vm_map_entry *); 212 213 #ifdef DEBUG 214 int _uvm_tree_sanity(struct vm_map *map, const char *name); 215 #endif 216 vsize_t uvm_rb_subtree_space(struct vm_map_entry *); 217 void uvm_rb_fixup(struct vm_map *, struct vm_map_entry *); 218 219 static __inline int 220 uvm_compare(struct vm_map_entry *a, struct vm_map_entry *b) 221 { 222 if (a->start < b->start) 223 return (-1); 224 else if (a->start > b->start) 225 return (1); 226 227 return (0); 228 } 229 230 231 static __inline void 232 uvm_rb_augment(struct vm_map_entry *entry) 233 { 234 entry->space = uvm_rb_subtree_space(entry); 235 } 236 237 RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 238 239 RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare); 240 241 vsize_t 242 uvm_rb_space(struct vm_map *map, struct vm_map_entry *entry) 243 { 244 struct vm_map_entry *next; 245 vaddr_t space; 246 247 if ((next = entry->next) == &map->header) 248 space = map->max_offset - entry->end; 249 else { 250 KASSERT(next); 251 space = next->start - entry->end; 252 } 253 return (space); 254 } 255 256 vsize_t 257 uvm_rb_subtree_space(struct vm_map_entry *entry) 258 { 259 vaddr_t space, tmp; 260 261 space = entry->ownspace; 262 if (RB_LEFT(entry, rb_entry)) { 263 tmp = RB_LEFT(entry, rb_entry)->space; 264 if (tmp > space) 265 space = tmp; 266 } 267 268 if (RB_RIGHT(entry, rb_entry)) { 269 tmp = RB_RIGHT(entry, rb_entry)->space; 270 if (tmp > space) 271 space = tmp; 272 } 273 274 return (space); 275 } 276 277 void 278 uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry) 279 { 280 /* We need to traverse to the very top */ 281 do { 282 entry->ownspace = uvm_rb_space(map, entry); 283 entry->space = uvm_rb_subtree_space(entry); 284 } while ((entry = RB_PARENT(entry, rb_entry)) != NULL); 285 } 286 287 void 288 uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry) 289 { 290 vaddr_t space = uvm_rb_space(map, entry); 291 struct vm_map_entry *tmp; 292 293 entry->ownspace = entry->space = space; 294 tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry); 295 #ifdef DIAGNOSTIC 296 if (tmp != NULL) 297 panic("uvm_rb_insert: duplicate entry?"); 298 #endif 299 uvm_rb_fixup(map, entry); 300 if (entry->prev != &map->header) 301 uvm_rb_fixup(map, entry->prev); 302 } 303 304 void 305 uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry) 306 { 307 struct vm_map_entry *parent; 308 309 parent = RB_PARENT(entry, rb_entry); 310 RB_REMOVE(uvm_tree, &(map)->rbhead, entry); 311 if (entry->prev != &map->header) 312 uvm_rb_fixup(map, entry->prev); 313 if (parent) 314 uvm_rb_fixup(map, parent); 315 } 316 317 #ifdef DEBUG 318 #define uvm_tree_sanity(x,y) _uvm_tree_sanity(x,y) 319 #else 320 #define uvm_tree_sanity(x,y) 321 #endif 322 323 #ifdef DEBUG 324 int 325 _uvm_tree_sanity(struct vm_map *map, const char *name) 326 { 327 struct vm_map_entry *tmp, *trtmp; 328 int n = 0, i = 1; 329 330 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 331 if (tmp->ownspace != uvm_rb_space(map, tmp)) { 332 printf("%s: %d/%d ownspace %x != %x %s\n", 333 name, n + 1, map->nentries, 334 tmp->ownspace, uvm_rb_space(map, tmp), 335 tmp->next == &map->header ? "(last)" : ""); 336 goto error; 337 } 338 } 339 trtmp = NULL; 340 RB_FOREACH(tmp, uvm_tree, &map->rbhead) { 341 if (tmp->space != uvm_rb_subtree_space(tmp)) { 342 printf("%s: space %d != %d\n", 343 name, tmp->space, uvm_rb_subtree_space(tmp)); 344 goto error; 345 } 346 if (trtmp != NULL && trtmp->start >= tmp->start) { 347 printf("%s: corrupt: 0x%lx >= 0x%lx\n", 348 name, trtmp->start, tmp->start); 349 goto error; 350 } 351 n++; 352 353 trtmp = tmp; 354 } 355 356 if (n != map->nentries) { 357 printf("%s: nentries: %d vs %d\n", 358 name, n, map->nentries); 359 goto error; 360 } 361 362 for (tmp = map->header.next; tmp && tmp != &map->header; 363 tmp = tmp->next, i++) { 364 trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp); 365 if (trtmp != tmp) { 366 printf("%s: lookup: %d: %p - %p: %p\n", 367 name, i, tmp, trtmp, 368 RB_PARENT(tmp, rb_entry)); 369 goto error; 370 } 371 } 372 373 return (0); 374 error: 375 #ifdef DDB 376 /* handy breakpoint location for error case */ 377 __asm(".globl treesanity_label\ntreesanity_label:"); 378 #endif 379 return (-1); 380 } 381 #endif 382 383 /* 384 * uvm_mapent_alloc: allocate a map entry 385 */ 386 387 struct vm_map_entry * 388 uvm_mapent_alloc(struct vm_map *map) 389 { 390 struct vm_map_entry *me, *ne; 391 int s, i; 392 int slowdown; 393 UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist); 394 395 if (map->flags & VM_MAP_INTRSAFE || cold) { 396 s = splvm(); 397 simple_lock(&uvm.kentry_lock); 398 me = uvm.kentry_free; 399 if (me == NULL) { 400 ne = uvm_km_getpage(0, &slowdown); 401 if (ne == NULL) 402 panic("uvm_mapent_alloc: cannot allocate map " 403 "entry"); 404 for (i = 0; 405 i < PAGE_SIZE / sizeof(struct vm_map_entry) - 1; 406 i++) 407 ne[i].next = &ne[i + 1]; 408 ne[i].next = NULL; 409 me = ne; 410 if (ratecheck(&uvm_kmapent_last_warn_time, 411 &uvm_kmapent_warn_rate)) 412 printf("uvm_mapent_alloc: out of static " 413 "map entries\n"); 414 } 415 uvm.kentry_free = me->next; 416 uvmexp.kmapent++; 417 simple_unlock(&uvm.kentry_lock); 418 splx(s); 419 me->flags = UVM_MAP_STATIC; 420 } else if (map == kernel_map) { 421 splassert(IPL_NONE); 422 me = pool_get(&uvm_map_entry_kmem_pool, PR_WAITOK); 423 me->flags = UVM_MAP_KMEM; 424 } else { 425 splassert(IPL_NONE); 426 me = pool_get(&uvm_map_entry_pool, PR_WAITOK); 427 me->flags = 0; 428 } 429 430 UVMHIST_LOG(maphist, "<- new entry=%p [kentry=%ld]", me, 431 ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0); 432 return(me); 433 } 434 435 /* 436 * uvm_mapent_free: free map entry 437 * 438 * => XXX: static pool for kernel map? 439 */ 440 441 void 442 uvm_mapent_free(struct vm_map_entry *me) 443 { 444 int s; 445 UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist); 446 447 UVMHIST_LOG(maphist,"<- freeing map entry=%p [flags=%ld]", 448 me, me->flags, 0, 0); 449 if (me->flags & UVM_MAP_STATIC) { 450 s = splvm(); 451 simple_lock(&uvm.kentry_lock); 452 me->next = uvm.kentry_free; 453 uvm.kentry_free = me; 454 uvmexp.kmapent--; 455 simple_unlock(&uvm.kentry_lock); 456 splx(s); 457 } else if (me->flags & UVM_MAP_KMEM) { 458 splassert(IPL_NONE); 459 pool_put(&uvm_map_entry_kmem_pool, me); 460 } else { 461 splassert(IPL_NONE); 462 pool_put(&uvm_map_entry_pool, me); 463 } 464 } 465 466 /* 467 * uvm_mapent_copy: copy a map entry, preserving flags 468 */ 469 470 void 471 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 472 { 473 memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) - 474 ((char *)src)); 475 } 476 477 /* 478 * uvm_map_entry_unwire: unwire a map entry 479 * 480 * => map should be locked by caller 481 */ 482 void 483 uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry) 484 { 485 486 entry->wired_count = 0; 487 uvm_fault_unwire_locked(map, entry->start, entry->end); 488 } 489 490 491 /* 492 * wrapper for calling amap_ref() 493 */ 494 void 495 uvm_map_reference_amap(struct vm_map_entry *entry, int flags) 496 { 497 amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff, 498 (entry->end - entry->start) >> PAGE_SHIFT, flags); 499 } 500 501 502 /* 503 * wrapper for calling amap_unref() 504 */ 505 void 506 uvm_map_unreference_amap(struct vm_map_entry *entry, int flags) 507 { 508 amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff, 509 (entry->end - entry->start) >> PAGE_SHIFT, flags); 510 } 511 512 513 /* 514 * uvm_map_init: init mapping system at boot time. note that we allocate 515 * and init the static pool of structs vm_map_entry for the kernel here. 516 */ 517 518 void 519 uvm_map_init(void) 520 { 521 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 522 #if defined(UVMHIST) 523 static struct uvm_history_ent maphistbuf[100]; 524 static struct uvm_history_ent pdhistbuf[100]; 525 #endif 526 int lcv; 527 528 /* 529 * first, init logging system. 530 */ 531 532 UVMHIST_FUNC("uvm_map_init"); 533 UVMHIST_INIT_STATIC(maphist, maphistbuf); 534 UVMHIST_INIT_STATIC(pdhist, pdhistbuf); 535 UVMHIST_CALLED(maphist); 536 UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0); 537 UVMCNT_INIT(uvm_map_call, UVMCNT_CNT, 0, 538 "# uvm_map() successful calls", 0); 539 UVMCNT_INIT(map_backmerge, UVMCNT_CNT, 0, "# uvm_map() back merges", 0); 540 UVMCNT_INIT(map_forwmerge, UVMCNT_CNT, 0, "# uvm_map() missed forward", 541 0); 542 UVMCNT_INIT(map_nousermerge, UVMCNT_CNT, 0, "# back merges skipped", 0); 543 UVMCNT_INIT(uvm_mlk_call, UVMCNT_CNT, 0, "# map lookup calls", 0); 544 UVMCNT_INIT(uvm_mlk_hint, UVMCNT_CNT, 0, "# map lookup hint hits", 0); 545 546 /* 547 * now set up static pool of kernel map entries ... 548 */ 549 550 simple_lock_init(&uvm.kentry_lock); 551 uvm.kentry_free = NULL; 552 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 553 kernel_map_entry[lcv].next = uvm.kentry_free; 554 uvm.kentry_free = &kernel_map_entry[lcv]; 555 } 556 557 /* 558 * initialize the map-related pools. 559 */ 560 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 561 0, 0, 0, "vmsppl", &pool_allocator_nointr); 562 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 563 0, 0, 0, "vmmpepl", &pool_allocator_nointr); 564 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 565 0, 0, 0, "vmmpekpl", NULL); 566 pool_sethiwat(&uvm_map_entry_pool, 8192); 567 } 568 569 /* 570 * clippers 571 */ 572 573 /* 574 * uvm_map_clip_start: ensure that the entry begins at or after 575 * the starting address, if it doesn't we split the entry. 576 * 577 * => caller should use UVM_MAP_CLIP_START macro rather than calling 578 * this directly 579 * => map must be locked by caller 580 */ 581 582 void 583 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, 584 vaddr_t start) 585 { 586 struct vm_map_entry *new_entry; 587 vaddr_t new_adj; 588 589 /* uvm_map_simplify_entry(map, entry); */ /* XXX */ 590 591 uvm_tree_sanity(map, "clip_start entry"); 592 593 /* 594 * Split off the front portion. note that we must insert the new 595 * entry BEFORE this one, so that this entry has the specified 596 * starting address. 597 */ 598 599 new_entry = uvm_mapent_alloc(map); 600 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 601 602 new_entry->end = start; 603 new_adj = start - new_entry->start; 604 if (entry->object.uvm_obj) 605 entry->offset += new_adj; /* shift start over */ 606 607 /* Does not change order for the RB tree */ 608 entry->start = start; 609 610 if (new_entry->aref.ar_amap) { 611 amap_splitref(&new_entry->aref, &entry->aref, new_adj); 612 } 613 614 uvm_map_entry_link(map, entry->prev, new_entry); 615 616 if (UVM_ET_ISSUBMAP(entry)) { 617 /* ... unlikely to happen, but play it safe */ 618 uvm_map_reference(new_entry->object.sub_map); 619 } else { 620 if (UVM_ET_ISOBJ(entry) && 621 entry->object.uvm_obj->pgops && 622 entry->object.uvm_obj->pgops->pgo_reference) 623 entry->object.uvm_obj->pgops->pgo_reference( 624 entry->object.uvm_obj); 625 } 626 627 uvm_tree_sanity(map, "clip_start leave"); 628 } 629 630 /* 631 * uvm_map_clip_end: ensure that the entry ends at or before 632 * the ending address, if it doesn't we split the reference 633 * 634 * => caller should use UVM_MAP_CLIP_END macro rather than calling 635 * this directly 636 * => map must be locked by caller 637 */ 638 639 void 640 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end) 641 { 642 struct vm_map_entry *new_entry; 643 vaddr_t new_adj; /* #bytes we move start forward */ 644 645 uvm_tree_sanity(map, "clip_end entry"); 646 /* 647 * Create a new entry and insert it 648 * AFTER the specified entry 649 */ 650 651 new_entry = uvm_mapent_alloc(map); 652 uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */ 653 654 new_entry->start = entry->end = end; 655 new_adj = end - entry->start; 656 if (new_entry->object.uvm_obj) 657 new_entry->offset += new_adj; 658 659 if (entry->aref.ar_amap) 660 amap_splitref(&entry->aref, &new_entry->aref, new_adj); 661 662 uvm_rb_fixup(map, entry); 663 664 uvm_map_entry_link(map, entry, new_entry); 665 666 if (UVM_ET_ISSUBMAP(entry)) { 667 /* ... unlikely to happen, but play it safe */ 668 uvm_map_reference(new_entry->object.sub_map); 669 } else { 670 if (UVM_ET_ISOBJ(entry) && 671 entry->object.uvm_obj->pgops && 672 entry->object.uvm_obj->pgops->pgo_reference) 673 entry->object.uvm_obj->pgops->pgo_reference( 674 entry->object.uvm_obj); 675 } 676 uvm_tree_sanity(map, "clip_end leave"); 677 } 678 679 680 /* 681 * M A P - m a i n e n t r y p o i n t 682 */ 683 /* 684 * uvm_map: establish a valid mapping in a map 685 * 686 * => assume startp is page aligned. 687 * => assume size is a multiple of PAGE_SIZE. 688 * => assume sys_mmap provides enough of a "hint" to have us skip 689 * over text/data/bss area. 690 * => map must be unlocked (we will lock it) 691 * => <uobj,uoffset> value meanings (4 cases): 692 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 693 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 694 * [3] <uobj,uoffset> == normal mapping 695 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 696 * 697 * case [4] is for kernel mappings where we don't know the offset until 698 * we've found a virtual address. note that kernel object offsets are 699 * always relative to vm_map_min(kernel_map). 700 * 701 * => if `align' is non-zero, we try to align the virtual address to 702 * the specified alignment. this is only a hint; if we can't 703 * do it, the address will be unaligned. this is provided as 704 * a mechanism for large pages. 705 * 706 * => XXXCDC: need way to map in external amap? 707 */ 708 709 int 710 uvm_map_p(struct vm_map *map, vaddr_t *startp, vsize_t size, 711 struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags, 712 struct proc *p) 713 { 714 struct vm_map_entry *prev_entry, *new_entry; 715 vm_prot_t prot = UVM_PROTECTION(flags), maxprot = 716 UVM_MAXPROTECTION(flags); 717 vm_inherit_t inherit = UVM_INHERIT(flags); 718 int advice = UVM_ADVICE(flags); 719 int error; 720 UVMHIST_FUNC("uvm_map"); 721 UVMHIST_CALLED(maphist); 722 723 UVMHIST_LOG(maphist, "(map=%p, *startp=0x%lx, size=%ld, flags=0x%lx)", 724 map, *startp, size, flags); 725 UVMHIST_LOG(maphist, " uobj/offset %p/%ld", uobj, (u_long)uoffset,0,0); 726 727 uvm_tree_sanity(map, "map entry"); 728 729 if ((map->flags & VM_MAP_INTRSAFE) == 0) 730 splassert(IPL_NONE); 731 else 732 splassert(IPL_VM); 733 734 /* 735 * step 0: sanity check of protection code 736 */ 737 738 if ((prot & maxprot) != prot) { 739 UVMHIST_LOG(maphist, "<- prot. failure: prot=0x%lx, max=0x%lx", 740 prot, maxprot,0,0); 741 return (EACCES); 742 } 743 744 /* 745 * step 1: figure out where to put new VM range 746 */ 747 748 if (vm_map_lock_try(map) == FALSE) { 749 if (flags & UVM_FLAG_TRYLOCK) 750 return (EFAULT); 751 vm_map_lock(map); /* could sleep here */ 752 } 753 if ((prev_entry = uvm_map_findspace(map, *startp, size, startp, 754 uobj, uoffset, align, flags)) == NULL) { 755 UVMHIST_LOG(maphist,"<- uvm_map_findspace failed!",0,0,0,0); 756 vm_map_unlock(map); 757 return (ENOMEM); 758 } 759 760 #ifdef PMAP_GROWKERNEL 761 { 762 /* 763 * If the kernel pmap can't map the requested space, 764 * then allocate more resources for it. 765 */ 766 if (map == kernel_map && uvm_maxkaddr < (*startp + size)) 767 uvm_maxkaddr = pmap_growkernel(*startp + size); 768 } 769 #endif 770 771 UVMCNT_INCR(uvm_map_call); 772 773 /* 774 * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER 775 * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET. in 776 * either case we want to zero it before storing it in the map entry 777 * (because it looks strange and confusing when debugging...) 778 * 779 * if uobj is not null 780 * if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping 781 * and we do not need to change uoffset. 782 * if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset 783 * now (based on the starting address of the map). this case is 784 * for kernel object mappings where we don't know the offset until 785 * the virtual address is found (with uvm_map_findspace). the 786 * offset is the distance we are from the start of the map. 787 */ 788 789 if (uobj == NULL) { 790 uoffset = 0; 791 } else { 792 if (uoffset == UVM_UNKNOWN_OFFSET) { 793 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 794 uoffset = *startp - vm_map_min(kernel_map); 795 } 796 } 797 798 /* 799 * step 2: try and insert in map by extending previous entry, if 800 * possible 801 * XXX: we don't try and pull back the next entry. might be useful 802 * for a stack, but we are currently allocating our stack in advance. 803 */ 804 805 if ((flags & UVM_FLAG_NOMERGE) == 0 && 806 prev_entry->end == *startp && prev_entry != &map->header && 807 prev_entry->object.uvm_obj == uobj) { 808 809 if (uobj && prev_entry->offset + 810 (prev_entry->end - prev_entry->start) != uoffset) 811 goto step3; 812 813 if (UVM_ET_ISSUBMAP(prev_entry)) 814 goto step3; 815 816 if (prev_entry->protection != prot || 817 prev_entry->max_protection != maxprot) 818 goto step3; 819 820 if (prev_entry->inheritance != inherit || 821 prev_entry->advice != advice) 822 goto step3; 823 824 /* wiring status must match (new area is unwired) */ 825 if (VM_MAPENT_ISWIRED(prev_entry)) 826 goto step3; 827 828 /* 829 * can't extend a shared amap. note: no need to lock amap to 830 * look at refs since we don't care about its exact value. 831 * if it is one (i.e. we have only reference) it will stay there 832 */ 833 834 if (prev_entry->aref.ar_amap && 835 amap_refs(prev_entry->aref.ar_amap) != 1) { 836 goto step3; 837 } 838 839 /* 840 * Only merge kernel mappings, but keep track 841 * of how much we skipped. 842 */ 843 if (map != kernel_map && map != kmem_map) { 844 UVMCNT_INCR(map_nousermerge); 845 goto step3; 846 } 847 848 if (prev_entry->aref.ar_amap) { 849 error = amap_extend(prev_entry, size); 850 if (error) 851 goto step3; 852 } 853 854 UVMCNT_INCR(map_backmerge); 855 UVMHIST_LOG(maphist," starting back merge", 0, 0, 0, 0); 856 857 /* 858 * drop our reference to uobj since we are extending a reference 859 * that we already have (the ref count can not drop to zero). 860 */ 861 862 if (uobj && uobj->pgops->pgo_detach) 863 uobj->pgops->pgo_detach(uobj); 864 865 prev_entry->end += size; 866 uvm_rb_fixup(map, prev_entry); 867 map->size += size; 868 if (p && uobj == NULL) 869 p->p_vmspace->vm_dused += atop(size); 870 871 uvm_tree_sanity(map, "map leave 2"); 872 873 UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0); 874 vm_map_unlock(map); 875 return (0); 876 877 } 878 step3: 879 UVMHIST_LOG(maphist," allocating new map entry", 0, 0, 0, 0); 880 881 /* 882 * check for possible forward merge (which we don't do) and count 883 * the number of times we missed a *possible* chance to merge more 884 */ 885 886 if ((flags & UVM_FLAG_NOMERGE) == 0 && 887 prev_entry->next != &map->header && 888 prev_entry->next->start == (*startp + size)) 889 UVMCNT_INCR(map_forwmerge); 890 891 /* 892 * step 3: allocate new entry and link it in 893 */ 894 895 new_entry = uvm_mapent_alloc(map); 896 new_entry->start = *startp; 897 new_entry->end = new_entry->start + size; 898 new_entry->object.uvm_obj = uobj; 899 new_entry->offset = uoffset; 900 901 if (uobj) 902 new_entry->etype = UVM_ET_OBJ; 903 else 904 new_entry->etype = 0; 905 906 if (flags & UVM_FLAG_COPYONW) { 907 new_entry->etype |= UVM_ET_COPYONWRITE; 908 if ((flags & UVM_FLAG_OVERLAY) == 0) 909 new_entry->etype |= UVM_ET_NEEDSCOPY; 910 } 911 if (flags & UVM_FLAG_HOLE) 912 new_entry->etype |= UVM_ET_HOLE; 913 914 new_entry->protection = prot; 915 new_entry->max_protection = maxprot; 916 new_entry->inheritance = inherit; 917 new_entry->wired_count = 0; 918 new_entry->advice = advice; 919 if (flags & UVM_FLAG_OVERLAY) { 920 /* 921 * to_add: for BSS we overallocate a little since we 922 * are likely to extend 923 */ 924 vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ? 925 UVM_AMAP_CHUNK << PAGE_SHIFT : 0; 926 struct vm_amap *amap = amap_alloc(size, to_add, M_WAITOK); 927 new_entry->aref.ar_pageoff = 0; 928 new_entry->aref.ar_amap = amap; 929 } else { 930 new_entry->aref.ar_pageoff = 0; 931 new_entry->aref.ar_amap = NULL; 932 } 933 934 uvm_map_entry_link(map, prev_entry, new_entry); 935 936 map->size += size; 937 if (p && uobj == NULL) 938 p->p_vmspace->vm_dused += atop(size); 939 940 941 /* 942 * Update the free space hint 943 */ 944 945 if ((map->first_free == prev_entry) && 946 (prev_entry->end >= new_entry->start)) 947 map->first_free = new_entry; 948 949 uvm_tree_sanity(map, "map leave"); 950 951 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 952 vm_map_unlock(map); 953 return (0); 954 } 955 956 /* 957 * uvm_map_lookup_entry: find map entry at or before an address 958 * 959 * => map must at least be read-locked by caller 960 * => entry is returned in "entry" 961 * => return value is true if address is in the returned entry 962 */ 963 964 boolean_t 965 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 966 struct vm_map_entry **entry) 967 { 968 struct vm_map_entry *cur; 969 struct vm_map_entry *last; 970 int use_tree = 0; 971 UVMHIST_FUNC("uvm_map_lookup_entry"); 972 UVMHIST_CALLED(maphist); 973 974 UVMHIST_LOG(maphist,"(map=%p,addr=0x%lx,ent=%p)", 975 map, address, entry, 0); 976 977 /* 978 * start looking either from the head of the 979 * list, or from the hint. 980 */ 981 982 simple_lock(&map->hint_lock); 983 cur = map->hint; 984 simple_unlock(&map->hint_lock); 985 986 if (cur == &map->header) 987 cur = cur->next; 988 989 UVMCNT_INCR(uvm_mlk_call); 990 if (address >= cur->start) { 991 /* 992 * go from hint to end of list. 993 * 994 * but first, make a quick check to see if 995 * we are already looking at the entry we 996 * want (which is usually the case). 997 * note also that we don't need to save the hint 998 * here... it is the same hint (unless we are 999 * at the header, in which case the hint didn't 1000 * buy us anything anyway). 1001 */ 1002 last = &map->header; 1003 if ((cur != last) && (cur->end > address)) { 1004 UVMCNT_INCR(uvm_mlk_hint); 1005 *entry = cur; 1006 UVMHIST_LOG(maphist,"<- got it via hint (%p)", 1007 cur, 0, 0, 0); 1008 return (TRUE); 1009 } 1010 1011 if (map->nentries > 30) 1012 use_tree = 1; 1013 } else { 1014 /* 1015 * go from start to hint, *inclusively* 1016 */ 1017 last = cur->next; 1018 cur = map->header.next; 1019 use_tree = 1; 1020 } 1021 1022 uvm_tree_sanity(map, __func__); 1023 1024 if (use_tree) { 1025 struct vm_map_entry *prev = &map->header; 1026 cur = RB_ROOT(&map->rbhead); 1027 1028 /* 1029 * Simple lookup in the tree. Happens when the hint is 1030 * invalid, or nentries reach a threshold. 1031 */ 1032 while (cur) { 1033 if (address >= cur->start) { 1034 if (address < cur->end) { 1035 *entry = cur; 1036 SAVE_HINT(map, map->hint, cur); 1037 return (TRUE); 1038 } 1039 prev = cur; 1040 cur = RB_RIGHT(cur, rb_entry); 1041 } else 1042 cur = RB_LEFT(cur, rb_entry); 1043 } 1044 *entry = prev; 1045 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1046 return (FALSE); 1047 } 1048 1049 /* 1050 * search linearly 1051 */ 1052 1053 while (cur != last) { 1054 if (cur->end > address) { 1055 if (address >= cur->start) { 1056 /* 1057 * save this lookup for future 1058 * hints, and return 1059 */ 1060 1061 *entry = cur; 1062 SAVE_HINT(map, map->hint, cur); 1063 UVMHIST_LOG(maphist,"<- search got it (%p)", 1064 cur, 0, 0, 0); 1065 return (TRUE); 1066 } 1067 break; 1068 } 1069 cur = cur->next; 1070 } 1071 1072 *entry = cur->prev; 1073 SAVE_HINT(map, map->hint, *entry); 1074 UVMHIST_LOG(maphist,"<- failed!",0,0,0,0); 1075 return (FALSE); 1076 } 1077 1078 /* 1079 * Checks if address pointed to by phint fits into the empty 1080 * space before the vm_map_entry after. Takes aligment and 1081 * offset into consideration. 1082 */ 1083 1084 int 1085 uvm_map_spacefits(struct vm_map *map, vaddr_t *phint, vsize_t length, 1086 struct vm_map_entry *after, voff_t uoffset, vsize_t align) 1087 { 1088 vaddr_t hint = *phint; 1089 vaddr_t end; 1090 1091 #ifdef PMAP_PREFER 1092 /* 1093 * push hint forward as needed to avoid VAC alias problems. 1094 * we only do this if a valid offset is specified. 1095 */ 1096 if (uoffset != UVM_UNKNOWN_OFFSET) 1097 PMAP_PREFER(uoffset, &hint); 1098 #endif 1099 if (align != 0) 1100 if ((hint & (align - 1)) != 0) 1101 hint = roundup(hint, align); 1102 *phint = hint; 1103 1104 end = hint + length; 1105 if (end > map->max_offset || end < hint) 1106 return (FALSE); 1107 if (after != NULL && after != &map->header && after->start < end) 1108 return (FALSE); 1109 1110 return (TRUE); 1111 } 1112 1113 /* 1114 * uvm_map_pie: return a random load address for a PIE executable 1115 * properly aligned. 1116 */ 1117 1118 #ifndef VM_PIE_MAX_ADDR 1119 #define VM_PIE_MAX_ADDR (VM_MAXUSER_ADDRESS / 4) 1120 #endif 1121 1122 #ifndef VM_PIE_MIN_ADDR 1123 #define VM_PIE_MIN_ADDR VM_MIN_ADDRESS 1124 #endif 1125 1126 #ifndef VM_PIE_MIN_ALIGN 1127 #define VM_PIE_MIN_ALIGN PAGE_SIZE 1128 #endif 1129 1130 vaddr_t 1131 uvm_map_pie(vaddr_t align) 1132 { 1133 vaddr_t addr, space, min; 1134 1135 align = MAX(align, VM_PIE_MIN_ALIGN); 1136 1137 /* round up to next alignment */ 1138 min = (VM_PIE_MIN_ADDR + align - 1) & ~(align - 1); 1139 1140 if (align >= VM_PIE_MAX_ADDR || min >= VM_PIE_MAX_ADDR) 1141 return (align); 1142 1143 space = (VM_PIE_MAX_ADDR - min) / align; 1144 space = MIN(space, (u_int32_t)-1); 1145 1146 addr = (vaddr_t)arc4random_uniform((u_int32_t)space) * align; 1147 addr += min; 1148 1149 return (addr); 1150 } 1151 1152 /* 1153 * uvm_map_hint: return the beginning of the best area suitable for 1154 * creating a new mapping with "prot" protection. 1155 */ 1156 vaddr_t 1157 uvm_map_hint(struct proc *p, vm_prot_t prot) 1158 { 1159 vaddr_t addr; 1160 1161 #ifdef __i386__ 1162 /* 1163 * If executable skip first two pages, otherwise start 1164 * after data + heap region. 1165 */ 1166 if ((prot & VM_PROT_EXECUTE) && 1167 ((vaddr_t)p->p_vmspace->vm_daddr >= I386_MAX_EXE_ADDR)) { 1168 addr = (PAGE_SIZE*2) + 1169 (arc4random() & (I386_MAX_EXE_ADDR / 2 - 1)); 1170 return (round_page(addr)); 1171 } 1172 #endif 1173 addr = (vaddr_t)p->p_vmspace->vm_daddr + MAXDSIZ; 1174 #if !defined(__vax__) 1175 addr += arc4random() & (MIN((256 * 1024 * 1024), MAXDSIZ) - 1); 1176 #else 1177 /* start malloc/mmap after the brk */ 1178 addr = (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ; 1179 #endif 1180 return (round_page(addr)); 1181 } 1182 1183 /* 1184 * uvm_map_findspace: find "length" sized space in "map". 1185 * 1186 * => "hint" is a hint about where we want it, unless FINDSPACE_FIXED is 1187 * set (in which case we insist on using "hint"). 1188 * => "result" is VA returned 1189 * => uobj/uoffset are to be used to handle VAC alignment, if required 1190 * => if `align' is non-zero, we attempt to align to that value. 1191 * => caller must at least have read-locked map 1192 * => returns NULL on failure, or pointer to prev. map entry if success 1193 * => note this is a cross between the old vm_map_findspace and vm_map_find 1194 */ 1195 1196 struct vm_map_entry * 1197 uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length, 1198 vaddr_t *result, struct uvm_object *uobj, voff_t uoffset, vsize_t align, 1199 int flags) 1200 { 1201 struct vm_map_entry *entry, *next, *tmp; 1202 struct vm_map_entry *child, *prev = NULL; 1203 1204 vaddr_t end, orig_hint; 1205 UVMHIST_FUNC("uvm_map_findspace"); 1206 UVMHIST_CALLED(maphist); 1207 1208 UVMHIST_LOG(maphist, "(map=%p, hint=0x%lx, len=%ld, flags=0x%lx)", 1209 map, hint, length, flags); 1210 KASSERT((align & (align - 1)) == 0); 1211 KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0); 1212 1213 uvm_tree_sanity(map, "map_findspace entry"); 1214 1215 /* 1216 * remember the original hint. if we are aligning, then we 1217 * may have to try again with no alignment constraint if 1218 * we fail the first time. 1219 */ 1220 1221 orig_hint = hint; 1222 if (hint < map->min_offset) { /* check ranges ... */ 1223 if (flags & UVM_FLAG_FIXED) { 1224 UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0); 1225 return(NULL); 1226 } 1227 hint = map->min_offset; 1228 } 1229 if (hint > map->max_offset) { 1230 UVMHIST_LOG(maphist,"<- VA 0x%lx > range [0x%lx->0x%lx]", 1231 hint, map->min_offset, map->max_offset, 0); 1232 return(NULL); 1233 } 1234 1235 /* 1236 * Look for the first possible address; if there's already 1237 * something at this address, we have to start after it. 1238 */ 1239 1240 if ((flags & UVM_FLAG_FIXED) == 0 && hint == map->min_offset) { 1241 if ((entry = map->first_free) != &map->header) 1242 hint = entry->end; 1243 } else { 1244 if (uvm_map_lookup_entry(map, hint, &tmp)) { 1245 /* "hint" address already in use ... */ 1246 if (flags & UVM_FLAG_FIXED) { 1247 UVMHIST_LOG(maphist,"<- fixed & VA in use", 1248 0, 0, 0, 0); 1249 return(NULL); 1250 } 1251 hint = tmp->end; 1252 } 1253 entry = tmp; 1254 } 1255 1256 if (flags & UVM_FLAG_FIXED) { 1257 end = hint + length; 1258 if (end > map->max_offset || end < hint) { 1259 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 1260 goto error; 1261 } 1262 next = entry->next; 1263 if (next == &map->header || next->start >= end) 1264 goto found; 1265 UVMHIST_LOG(maphist,"<- fixed mapping failed", 0,0,0,0); 1266 return(NULL); /* only one shot at it ... */ 1267 } 1268 1269 /* Try to find the space in the red-black tree */ 1270 1271 /* Check slot before any entry */ 1272 if (uvm_map_spacefits(map, &hint, length, entry->next, uoffset, align)) 1273 goto found; 1274 1275 /* If there is not enough space in the whole tree, we fail */ 1276 tmp = RB_ROOT(&map->rbhead); 1277 if (tmp == NULL || tmp->space < length) 1278 goto error; 1279 1280 /* Find an entry close to hint that has enough space */ 1281 for (; tmp;) { 1282 if (tmp->end >= hint && 1283 (prev == NULL || tmp->end < prev->end)) { 1284 if (tmp->ownspace >= length) 1285 prev = tmp; 1286 else if ((child = RB_RIGHT(tmp, rb_entry)) != NULL && 1287 child->space >= length) 1288 prev = tmp; 1289 } 1290 if (tmp->end < hint) 1291 child = RB_RIGHT(tmp, rb_entry); 1292 else if (tmp->end > hint) 1293 child = RB_LEFT(tmp, rb_entry); 1294 else { 1295 if (tmp->ownspace >= length) 1296 break; 1297 child = RB_RIGHT(tmp, rb_entry); 1298 } 1299 if (child == NULL || child->space < length) 1300 break; 1301 tmp = child; 1302 } 1303 1304 if (tmp != NULL && hint < tmp->end + tmp->ownspace) { 1305 /* 1306 * Check if the entry that we found satifies the 1307 * space requirement 1308 */ 1309 if (hint < tmp->end) 1310 hint = tmp->end; 1311 if (uvm_map_spacefits(map, &hint, length, tmp->next, uoffset, 1312 align)) { 1313 entry = tmp; 1314 goto found; 1315 } else if (tmp->ownspace >= length) 1316 goto listsearch; 1317 } 1318 if (prev == NULL) 1319 goto error; 1320 1321 hint = prev->end; 1322 if (uvm_map_spacefits(map, &hint, length, prev->next, uoffset, 1323 align)) { 1324 entry = prev; 1325 goto found; 1326 } else if (prev->ownspace >= length) 1327 goto listsearch; 1328 1329 tmp = RB_RIGHT(prev, rb_entry); 1330 for (;;) { 1331 KASSERT(tmp && tmp->space >= length); 1332 child = RB_LEFT(tmp, rb_entry); 1333 if (child && child->space >= length) { 1334 tmp = child; 1335 continue; 1336 } 1337 if (tmp->ownspace >= length) 1338 break; 1339 tmp = RB_RIGHT(tmp, rb_entry); 1340 } 1341 1342 hint = tmp->end; 1343 if (uvm_map_spacefits(map, &hint, length, tmp->next, uoffset, align)) { 1344 entry = tmp; 1345 goto found; 1346 } 1347 1348 /* 1349 * The tree fails to find an entry because of offset or alignment 1350 * restrictions. Search the list instead. 1351 */ 1352 listsearch: 1353 /* 1354 * Look through the rest of the map, trying to fit a new region in 1355 * the gap between existing regions, or after the very last region. 1356 * note: entry->end = base VA of current gap, 1357 * next->start = VA of end of current gap 1358 */ 1359 for (;; hint = (entry = next)->end) { 1360 /* 1361 * Find the end of the proposed new region. Be sure we didn't 1362 * go beyond the end of the map, or wrap around the address; 1363 * if so, we lose. Otherwise, if this is the last entry, or 1364 * if the proposed new region fits before the next entry, we 1365 * win. 1366 */ 1367 1368 #ifdef PMAP_PREFER 1369 /* 1370 * push hint forward as needed to avoid VAC alias problems. 1371 * we only do this if a valid offset is specified. 1372 */ 1373 if (uoffset != UVM_UNKNOWN_OFFSET) 1374 PMAP_PREFER(uoffset, &hint); 1375 #endif 1376 if (align != 0) { 1377 if ((hint & (align - 1)) != 0) 1378 hint = roundup(hint, align); 1379 /* 1380 * XXX Should we PMAP_PREFER() here again? 1381 */ 1382 } 1383 end = hint + length; 1384 if (end > map->max_offset || end < hint) { 1385 UVMHIST_LOG(maphist,"<- failed (off end)", 0,0,0,0); 1386 goto error; 1387 } 1388 next = entry->next; 1389 if (next == &map->header || next->start >= end) 1390 break; 1391 } 1392 found: 1393 SAVE_HINT(map, map->hint, entry); 1394 *result = hint; 1395 UVMHIST_LOG(maphist,"<- got it! (result=0x%lx)", hint, 0,0,0); 1396 return (entry); 1397 1398 error: 1399 if (align != 0) { 1400 UVMHIST_LOG(maphist, 1401 "calling recursively, no align", 1402 0,0,0,0); 1403 return (uvm_map_findspace(map, orig_hint, 1404 length, result, uobj, uoffset, 0, flags)); 1405 } 1406 return (NULL); 1407 } 1408 1409 /* 1410 * U N M A P - m a i n e n t r y p o i n t 1411 */ 1412 1413 /* 1414 * uvm_unmap: remove mappings from a vm_map (from "start" up to "stop") 1415 * 1416 * => caller must check alignment and size 1417 * => map must be unlocked (we will lock it) 1418 */ 1419 void 1420 uvm_unmap_p(vm_map_t map, vaddr_t start, vaddr_t end, struct proc *p) 1421 { 1422 vm_map_entry_t dead_entries; 1423 UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist); 1424 1425 UVMHIST_LOG(maphist, " (map=%p, start=0x%lx, end=0x%lx)", 1426 map, start, end, 0); 1427 /* 1428 * work now done by helper functions. wipe the pmap's and then 1429 * detach from the dead entries... 1430 */ 1431 vm_map_lock(map); 1432 uvm_unmap_remove(map, start, end, &dead_entries, p); 1433 vm_map_unlock(map); 1434 1435 if (dead_entries != NULL) 1436 uvm_unmap_detach(dead_entries, 0); 1437 1438 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1439 } 1440 1441 1442 /* 1443 * U N M A P - m a i n h e l p e r f u n c t i o n s 1444 */ 1445 1446 /* 1447 * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop") 1448 * 1449 * => caller must check alignment and size 1450 * => map must be locked by caller 1451 * => we return a list of map entries that we've remove from the map 1452 * in "entry_list" 1453 */ 1454 1455 void 1456 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 1457 struct vm_map_entry **entry_list, struct proc *p) 1458 { 1459 struct vm_map_entry *entry, *first_entry, *next; 1460 vaddr_t len; 1461 UVMHIST_FUNC("uvm_unmap_remove"); 1462 UVMHIST_CALLED(maphist); 1463 1464 UVMHIST_LOG(maphist,"(map=%p, start=0x%lx, end=0x%lx)", 1465 map, start, end, 0); 1466 1467 VM_MAP_RANGE_CHECK(map, start, end); 1468 1469 uvm_tree_sanity(map, "unmap_remove entry"); 1470 1471 if ((map->flags & VM_MAP_INTRSAFE) == 0) 1472 splassert(IPL_NONE); 1473 else 1474 splassert(IPL_VM); 1475 1476 /* 1477 * find first entry 1478 */ 1479 if (uvm_map_lookup_entry(map, start, &first_entry) == TRUE) { 1480 /* clip and go... */ 1481 entry = first_entry; 1482 UVM_MAP_CLIP_START(map, entry, start); 1483 /* critical! prevents stale hint */ 1484 SAVE_HINT(map, entry, entry->prev); 1485 1486 } else { 1487 entry = first_entry->next; 1488 } 1489 1490 /* 1491 * Save the free space hint 1492 */ 1493 1494 if (map->first_free->start >= start) 1495 map->first_free = entry->prev; 1496 1497 /* 1498 * note: we now re-use first_entry for a different task. we remove 1499 * a number of map entries from the map and save them in a linked 1500 * list headed by "first_entry". once we remove them from the map 1501 * the caller should unlock the map and drop the references to the 1502 * backing objects [c.f. uvm_unmap_detach]. the object is to 1503 * separate unmapping from reference dropping. why? 1504 * [1] the map has to be locked for unmapping 1505 * [2] the map need not be locked for reference dropping 1506 * [3] dropping references may trigger pager I/O, and if we hit 1507 * a pager that does synchronous I/O we may have to wait for it. 1508 * [4] we would like all waiting for I/O to occur with maps unlocked 1509 * so that we don't block other threads. 1510 */ 1511 first_entry = NULL; 1512 *entry_list = NULL; /* to be safe */ 1513 1514 /* 1515 * break up the area into map entry sized regions and unmap. note 1516 * that all mappings have to be removed before we can even consider 1517 * dropping references to amaps or VM objects (otherwise we could end 1518 * up with a mapping to a page on the free list which would be very bad) 1519 */ 1520 1521 while ((entry != &map->header) && (entry->start < end)) { 1522 1523 UVM_MAP_CLIP_END(map, entry, end); 1524 next = entry->next; 1525 len = entry->end - entry->start; 1526 if (p && entry->object.uvm_obj == NULL) 1527 p->p_vmspace->vm_dused -= atop(len); 1528 1529 /* 1530 * unwire before removing addresses from the pmap; otherwise 1531 * unwiring will put the entries back into the pmap (XXX). 1532 */ 1533 1534 if (VM_MAPENT_ISWIRED(entry)) 1535 uvm_map_entry_unwire(map, entry); 1536 1537 /* 1538 * special case: handle mappings to anonymous kernel objects. 1539 * we want to free these pages right away... 1540 */ 1541 if (UVM_ET_ISHOLE(entry)) { 1542 /* nothing to do! */ 1543 } else if (map->flags & VM_MAP_INTRSAFE) { 1544 uvm_km_pgremove_intrsafe(entry->start, entry->end); 1545 pmap_kremove(entry->start, len); 1546 } else if (UVM_ET_ISOBJ(entry) && 1547 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1548 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1549 1550 /* 1551 * note: kernel object mappings are currently used in 1552 * two ways: 1553 * [1] "normal" mappings of pages in the kernel object 1554 * [2] uvm_km_valloc'd allocations in which we 1555 * pmap_enter in some non-kernel-object page 1556 * (e.g. vmapbuf). 1557 * 1558 * for case [1], we need to remove the mapping from 1559 * the pmap and then remove the page from the kernel 1560 * object (because, once pages in a kernel object are 1561 * unmapped they are no longer needed, unlike, say, 1562 * a vnode where you might want the data to persist 1563 * until flushed out of a queue). 1564 * 1565 * for case [2], we need to remove the mapping from 1566 * the pmap. there shouldn't be any pages at the 1567 * specified offset in the kernel object [but it 1568 * doesn't hurt to call uvm_km_pgremove just to be 1569 * safe?] 1570 * 1571 * uvm_km_pgremove currently does the following: 1572 * for pages in the kernel object in range: 1573 * - drops the swap slot 1574 * - uvm_pagefree the page 1575 * 1576 * note there is version of uvm_km_pgremove() that 1577 * is used for "intrsafe" objects. 1578 */ 1579 1580 /* 1581 * remove mappings from pmap and drop the pages 1582 * from the object. offsets are always relative 1583 * to vm_map_min(kernel_map). 1584 */ 1585 pmap_remove(pmap_kernel(), entry->start, entry->end); 1586 uvm_km_pgremove(entry->object.uvm_obj, 1587 entry->start - vm_map_min(kernel_map), 1588 entry->end - vm_map_min(kernel_map)); 1589 1590 /* 1591 * null out kernel_object reference, we've just 1592 * dropped it 1593 */ 1594 entry->etype &= ~UVM_ET_OBJ; 1595 entry->object.uvm_obj = NULL; /* to be safe */ 1596 1597 } else { 1598 /* 1599 * remove mappings the standard way. 1600 */ 1601 pmap_remove(map->pmap, entry->start, entry->end); 1602 } 1603 1604 /* 1605 * remove entry from map and put it on our list of entries 1606 * that we've nuked. then go do next entry. 1607 */ 1608 UVMHIST_LOG(maphist, " removed map entry %p", entry, 0, 0,0); 1609 1610 /* critical! prevents stale hint */ 1611 SAVE_HINT(map, entry, entry->prev); 1612 1613 uvm_map_entry_unlink(map, entry); 1614 map->size -= len; 1615 entry->next = first_entry; 1616 first_entry = entry; 1617 entry = next; /* next entry, please */ 1618 } 1619 /* if ((map->flags & VM_MAP_DYING) == 0) { */ 1620 pmap_update(vm_map_pmap(map)); 1621 /* } */ 1622 1623 1624 uvm_tree_sanity(map, "unmap_remove leave"); 1625 1626 /* 1627 * now we've cleaned up the map and are ready for the caller to drop 1628 * references to the mapped objects. 1629 */ 1630 1631 *entry_list = first_entry; 1632 UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0); 1633 } 1634 1635 /* 1636 * uvm_unmap_detach: drop references in a chain of map entries 1637 * 1638 * => we will free the map entries as we traverse the list. 1639 */ 1640 1641 void 1642 uvm_unmap_detach(struct vm_map_entry *first_entry, int flags) 1643 { 1644 struct vm_map_entry *next_entry; 1645 UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist); 1646 1647 while (first_entry) { 1648 KASSERT(!VM_MAPENT_ISWIRED(first_entry)); 1649 UVMHIST_LOG(maphist, 1650 " detach 0x%lx: amap=%p, obj=%p, submap?=%ld", 1651 first_entry, first_entry->aref.ar_amap, 1652 first_entry->object.uvm_obj, 1653 UVM_ET_ISSUBMAP(first_entry)); 1654 1655 /* 1656 * drop reference to amap, if we've got one 1657 */ 1658 1659 if (first_entry->aref.ar_amap) 1660 uvm_map_unreference_amap(first_entry, flags); 1661 1662 /* 1663 * drop reference to our backing object, if we've got one 1664 */ 1665 1666 if (UVM_ET_ISSUBMAP(first_entry)) { 1667 /* ... unlikely to happen, but play it safe */ 1668 uvm_map_deallocate(first_entry->object.sub_map); 1669 } else { 1670 if (UVM_ET_ISOBJ(first_entry) && 1671 first_entry->object.uvm_obj->pgops->pgo_detach) 1672 first_entry->object.uvm_obj->pgops-> 1673 pgo_detach(first_entry->object.uvm_obj); 1674 } 1675 1676 next_entry = first_entry->next; 1677 uvm_mapent_free(first_entry); 1678 first_entry = next_entry; 1679 } 1680 UVMHIST_LOG(maphist, "<- done", 0,0,0,0); 1681 } 1682 1683 /* 1684 * E X T R A C T I O N F U N C T I O N S 1685 */ 1686 1687 /* 1688 * uvm_map_reserve: reserve space in a vm_map for future use. 1689 * 1690 * => we reserve space in a map by putting a dummy map entry in the 1691 * map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE) 1692 * => map should be unlocked (we will write lock it) 1693 * => we return true if we were able to reserve space 1694 * => XXXCDC: should be inline? 1695 */ 1696 1697 int 1698 uvm_map_reserve(struct vm_map *map, vsize_t size, vaddr_t offset, 1699 vsize_t align, vaddr_t *raddr) 1700 { 1701 UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist); 1702 1703 UVMHIST_LOG(maphist, "(map=%p, size=0x%lx, offset=0x%lx,addr=0x%lx)", 1704 map,size,offset,raddr); 1705 1706 size = round_page(size); 1707 if (*raddr < vm_map_min(map)) 1708 *raddr = vm_map_min(map); /* hint */ 1709 1710 /* 1711 * reserve some virtual space. 1712 */ 1713 1714 if (uvm_map(map, raddr, size, NULL, offset, 0, 1715 UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE, 1716 UVM_ADV_RANDOM, UVM_FLAG_NOMERGE)) != 0) { 1717 UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0); 1718 return (FALSE); 1719 } 1720 1721 UVMHIST_LOG(maphist, "<- done (*raddr=0x%lx)", *raddr,0,0,0); 1722 return (TRUE); 1723 } 1724 1725 /* 1726 * uvm_map_replace: replace a reserved (blank) area of memory with 1727 * real mappings. 1728 * 1729 * => caller must WRITE-LOCK the map 1730 * => we return TRUE if replacement was a success 1731 * => we expect the newents chain to have nnewents entries on it and 1732 * we expect newents->prev to point to the last entry on the list 1733 * => note newents is allowed to be NULL 1734 */ 1735 1736 int 1737 uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end, 1738 struct vm_map_entry *newents, int nnewents) 1739 { 1740 struct vm_map_entry *oldent, *last; 1741 1742 uvm_tree_sanity(map, "map_replace entry"); 1743 1744 /* 1745 * first find the blank map entry at the specified address 1746 */ 1747 1748 if (!uvm_map_lookup_entry(map, start, &oldent)) { 1749 return(FALSE); 1750 } 1751 1752 /* 1753 * check to make sure we have a proper blank entry 1754 */ 1755 1756 if (oldent->start != start || oldent->end != end || 1757 oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) { 1758 return (FALSE); 1759 } 1760 1761 #ifdef DIAGNOSTIC 1762 /* 1763 * sanity check the newents chain 1764 */ 1765 { 1766 struct vm_map_entry *tmpent = newents; 1767 int nent = 0; 1768 vaddr_t cur = start; 1769 1770 while (tmpent) { 1771 nent++; 1772 if (tmpent->start < cur) 1773 panic("uvm_map_replace1"); 1774 if (tmpent->start > tmpent->end || tmpent->end > end) { 1775 printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n", 1776 tmpent->start, tmpent->end, end); 1777 panic("uvm_map_replace2"); 1778 } 1779 cur = tmpent->end; 1780 if (tmpent->next) { 1781 if (tmpent->next->prev != tmpent) 1782 panic("uvm_map_replace3"); 1783 } else { 1784 if (newents->prev != tmpent) 1785 panic("uvm_map_replace4"); 1786 } 1787 tmpent = tmpent->next; 1788 } 1789 if (nent != nnewents) 1790 panic("uvm_map_replace5"); 1791 } 1792 #endif 1793 1794 /* 1795 * map entry is a valid blank! replace it. (this does all the 1796 * work of map entry link/unlink...). 1797 */ 1798 1799 if (newents) { 1800 last = newents->prev; /* we expect this */ 1801 1802 /* critical: flush stale hints out of map */ 1803 SAVE_HINT(map, map->hint, newents); 1804 if (map->first_free == oldent) 1805 map->first_free = last; 1806 1807 last->next = oldent->next; 1808 last->next->prev = last; 1809 1810 /* Fix RB tree */ 1811 uvm_rb_remove(map, oldent); 1812 1813 newents->prev = oldent->prev; 1814 newents->prev->next = newents; 1815 map->nentries = map->nentries + (nnewents - 1); 1816 1817 /* Fixup the RB tree */ 1818 { 1819 int i; 1820 struct vm_map_entry *tmp; 1821 1822 tmp = newents; 1823 for (i = 0; i < nnewents && tmp; i++) { 1824 uvm_rb_insert(map, tmp); 1825 tmp = tmp->next; 1826 } 1827 } 1828 } else { 1829 1830 /* critical: flush stale hints out of map */ 1831 SAVE_HINT(map, map->hint, oldent->prev); 1832 if (map->first_free == oldent) 1833 map->first_free = oldent->prev; 1834 1835 /* NULL list of new entries: just remove the old one */ 1836 uvm_map_entry_unlink(map, oldent); 1837 } 1838 1839 1840 uvm_tree_sanity(map, "map_replace leave"); 1841 1842 /* 1843 * now we can free the old blank entry, unlock the map and return. 1844 */ 1845 1846 uvm_mapent_free(oldent); 1847 return(TRUE); 1848 } 1849 1850 /* 1851 * uvm_map_extract: extract a mapping from a map and put it somewhere 1852 * (maybe removing the old mapping) 1853 * 1854 * => maps should be unlocked (we will write lock them) 1855 * => returns 0 on success, error code otherwise 1856 * => start must be page aligned 1857 * => len must be page sized 1858 * => flags: 1859 * UVM_EXTRACT_REMOVE: remove mappings from srcmap 1860 * UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only) 1861 * UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs 1862 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 1863 * >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<< 1864 * >>>NOTE: QREF's must be unmapped via the QREF path, thus should only 1865 * be used from within the kernel in a kernel level map <<< 1866 */ 1867 1868 int 1869 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 1870 struct vm_map *dstmap, vaddr_t *dstaddrp, int flags) 1871 { 1872 vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge, 1873 oldstart; 1874 struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry; 1875 struct vm_map_entry *deadentry, *oldentry; 1876 vsize_t elen; 1877 int nchain, error, copy_ok; 1878 UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist); 1879 1880 UVMHIST_LOG(maphist,"(srcmap=%p,start=0x%lx, len=0x%lx", srcmap, start, 1881 len,0); 1882 UVMHIST_LOG(maphist," ...,dstmap=%p, flags=0x%lx)", dstmap,flags,0,0); 1883 1884 uvm_tree_sanity(srcmap, "map_extract src enter"); 1885 uvm_tree_sanity(dstmap, "map_extract dst enter"); 1886 1887 /* 1888 * step 0: sanity check: start must be on a page boundary, length 1889 * must be page sized. can't ask for CONTIG/QREF if you asked for 1890 * REMOVE. 1891 */ 1892 1893 KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0); 1894 KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 || 1895 (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0); 1896 1897 /* 1898 * step 1: reserve space in the target map for the extracted area 1899 */ 1900 1901 dstaddr = vm_map_min(dstmap); 1902 if (uvm_map_reserve(dstmap, len, start, 0, &dstaddr) == FALSE) 1903 return(ENOMEM); 1904 *dstaddrp = dstaddr; /* pass address back to caller */ 1905 UVMHIST_LOG(maphist, " dstaddr=0x%lx", dstaddr,0,0,0); 1906 1907 /* 1908 * step 2: setup for the extraction process loop by init'ing the 1909 * map entry chain, locking src map, and looking up the first useful 1910 * entry in the map. 1911 */ 1912 1913 end = start + len; 1914 newend = dstaddr + len; 1915 chain = endchain = NULL; 1916 nchain = 0; 1917 vm_map_lock(srcmap); 1918 1919 if (uvm_map_lookup_entry(srcmap, start, &entry)) { 1920 1921 /* "start" is within an entry */ 1922 if (flags & UVM_EXTRACT_QREF) { 1923 1924 /* 1925 * for quick references we don't clip the entry, so 1926 * the entry may map space "before" the starting 1927 * virtual address... this is the "fudge" factor 1928 * (which can be non-zero only the first time 1929 * through the "while" loop in step 3). 1930 */ 1931 1932 fudge = start - entry->start; 1933 } else { 1934 1935 /* 1936 * normal reference: we clip the map to fit (thus 1937 * fudge is zero) 1938 */ 1939 1940 UVM_MAP_CLIP_START(srcmap, entry, start); 1941 SAVE_HINT(srcmap, srcmap->hint, entry->prev); 1942 fudge = 0; 1943 } 1944 } else { 1945 1946 /* "start" is not within an entry ... skip to next entry */ 1947 if (flags & UVM_EXTRACT_CONTIG) { 1948 error = EINVAL; 1949 goto bad; /* definite hole here ... */ 1950 } 1951 1952 entry = entry->next; 1953 fudge = 0; 1954 } 1955 1956 /* save values from srcmap for step 6 */ 1957 orig_entry = entry; 1958 orig_fudge = fudge; 1959 1960 /* 1961 * step 3: now start looping through the map entries, extracting 1962 * as we go. 1963 */ 1964 1965 while (entry->start < end && entry != &srcmap->header) { 1966 1967 /* if we are not doing a quick reference, clip it */ 1968 if ((flags & UVM_EXTRACT_QREF) == 0) 1969 UVM_MAP_CLIP_END(srcmap, entry, end); 1970 1971 /* clear needs_copy (allow chunking) */ 1972 if (UVM_ET_ISNEEDSCOPY(entry)) { 1973 if (fudge) 1974 oldstart = entry->start; 1975 else 1976 oldstart = 0; /* XXX: gcc */ 1977 amap_copy(srcmap, entry, M_NOWAIT, TRUE, start, end); 1978 if (UVM_ET_ISNEEDSCOPY(entry)) { /* failed? */ 1979 error = ENOMEM; 1980 goto bad; 1981 } 1982 1983 /* amap_copy could clip (during chunk)! update fudge */ 1984 if (fudge) { 1985 fudge = fudge - (entry->start - oldstart); 1986 orig_fudge = fudge; 1987 } 1988 } 1989 1990 /* calculate the offset of this from "start" */ 1991 oldoffset = (entry->start + fudge) - start; 1992 1993 /* allocate a new map entry */ 1994 newentry = uvm_mapent_alloc(dstmap); 1995 if (newentry == NULL) { 1996 error = ENOMEM; 1997 goto bad; 1998 } 1999 2000 /* set up new map entry */ 2001 newentry->next = NULL; 2002 newentry->prev = endchain; 2003 newentry->start = dstaddr + oldoffset; 2004 newentry->end = 2005 newentry->start + (entry->end - (entry->start + fudge)); 2006 if (newentry->end > newend || newentry->end < newentry->start) 2007 newentry->end = newend; 2008 newentry->object.uvm_obj = entry->object.uvm_obj; 2009 if (newentry->object.uvm_obj) { 2010 if (newentry->object.uvm_obj->pgops->pgo_reference) 2011 newentry->object.uvm_obj->pgops-> 2012 pgo_reference(newentry->object.uvm_obj); 2013 newentry->offset = entry->offset + fudge; 2014 } else { 2015 newentry->offset = 0; 2016 } 2017 newentry->etype = entry->etype; 2018 newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ? 2019 entry->max_protection : entry->protection; 2020 newentry->max_protection = entry->max_protection; 2021 newentry->inheritance = entry->inheritance; 2022 newentry->wired_count = 0; 2023 newentry->aref.ar_amap = entry->aref.ar_amap; 2024 if (newentry->aref.ar_amap) { 2025 newentry->aref.ar_pageoff = 2026 entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT); 2027 uvm_map_reference_amap(newentry, AMAP_SHARED | 2028 ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0)); 2029 } else { 2030 newentry->aref.ar_pageoff = 0; 2031 } 2032 newentry->advice = entry->advice; 2033 2034 /* now link it on the chain */ 2035 nchain++; 2036 if (endchain == NULL) { 2037 chain = endchain = newentry; 2038 } else { 2039 endchain->next = newentry; 2040 endchain = newentry; 2041 } 2042 2043 /* end of 'while' loop! */ 2044 if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end && 2045 (entry->next == &srcmap->header || 2046 entry->next->start != entry->end)) { 2047 error = EINVAL; 2048 goto bad; 2049 } 2050 entry = entry->next; 2051 fudge = 0; 2052 } 2053 2054 /* 2055 * step 4: close off chain (in format expected by uvm_map_replace) 2056 */ 2057 2058 if (chain) 2059 chain->prev = endchain; 2060 2061 /* 2062 * step 5: attempt to lock the dest map so we can pmap_copy. 2063 * note usage of copy_ok: 2064 * 1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5) 2065 * 0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7 2066 */ 2067 2068 if (srcmap == dstmap || vm_map_lock_try(dstmap) == TRUE) { 2069 copy_ok = 1; 2070 if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2071 nchain)) { 2072 if (srcmap != dstmap) 2073 vm_map_unlock(dstmap); 2074 error = EIO; 2075 goto bad; 2076 } 2077 } else { 2078 copy_ok = 0; 2079 /* replace defered until step 7 */ 2080 } 2081 2082 /* 2083 * step 6: traverse the srcmap a second time to do the following: 2084 * - if we got a lock on the dstmap do pmap_copy 2085 * - if UVM_EXTRACT_REMOVE remove the entries 2086 * we make use of orig_entry and orig_fudge (saved in step 2) 2087 */ 2088 2089 if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) { 2090 2091 /* purge possible stale hints from srcmap */ 2092 if (flags & UVM_EXTRACT_REMOVE) { 2093 SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev); 2094 if (srcmap->first_free->start >= start) 2095 srcmap->first_free = orig_entry->prev; 2096 } 2097 2098 entry = orig_entry; 2099 fudge = orig_fudge; 2100 deadentry = NULL; /* for UVM_EXTRACT_REMOVE */ 2101 2102 while (entry->start < end && entry != &srcmap->header) { 2103 if (copy_ok) { 2104 oldoffset = (entry->start + fudge) - start; 2105 elen = MIN(end, entry->end) - 2106 (entry->start + fudge); 2107 pmap_copy(dstmap->pmap, srcmap->pmap, 2108 dstaddr + oldoffset, elen, 2109 entry->start + fudge); 2110 } 2111 2112 /* we advance "entry" in the following if statement */ 2113 if (flags & UVM_EXTRACT_REMOVE) { 2114 pmap_remove(srcmap->pmap, entry->start, 2115 entry->end); 2116 oldentry = entry; /* save entry */ 2117 entry = entry->next; /* advance */ 2118 uvm_map_entry_unlink(srcmap, oldentry); 2119 /* add to dead list */ 2120 oldentry->next = deadentry; 2121 deadentry = oldentry; 2122 } else { 2123 entry = entry->next; /* advance */ 2124 } 2125 2126 /* end of 'while' loop */ 2127 fudge = 0; 2128 } 2129 pmap_update(srcmap->pmap); 2130 2131 /* 2132 * unlock dstmap. we will dispose of deadentry in 2133 * step 7 if needed 2134 */ 2135 2136 if (copy_ok && srcmap != dstmap) 2137 vm_map_unlock(dstmap); 2138 2139 } 2140 else 2141 deadentry = NULL; /* XXX: gcc */ 2142 2143 /* 2144 * step 7: we are done with the source map, unlock. if copy_ok 2145 * is 0 then we have not replaced the dummy mapping in dstmap yet 2146 * and we need to do so now. 2147 */ 2148 2149 vm_map_unlock(srcmap); 2150 if ((flags & UVM_EXTRACT_REMOVE) && deadentry) 2151 uvm_unmap_detach(deadentry, 0); /* dispose of old entries */ 2152 2153 /* now do the replacement if we didn't do it in step 5 */ 2154 if (copy_ok == 0) { 2155 vm_map_lock(dstmap); 2156 error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain, 2157 nchain); 2158 vm_map_unlock(dstmap); 2159 2160 if (error == FALSE) { 2161 error = EIO; 2162 goto bad2; 2163 } 2164 } 2165 2166 uvm_tree_sanity(srcmap, "map_extract src leave"); 2167 uvm_tree_sanity(dstmap, "map_extract dst leave"); 2168 2169 return(0); 2170 2171 /* 2172 * bad: failure recovery 2173 */ 2174 bad: 2175 vm_map_unlock(srcmap); 2176 bad2: /* src already unlocked */ 2177 if (chain) 2178 uvm_unmap_detach(chain, 2179 (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0); 2180 2181 uvm_tree_sanity(srcmap, "map_extract src err leave"); 2182 uvm_tree_sanity(dstmap, "map_extract dst err leave"); 2183 2184 uvm_unmap(dstmap, dstaddr, dstaddr+len); /* ??? */ 2185 return(error); 2186 } 2187 2188 /* end of extraction functions */ 2189 2190 /* 2191 * uvm_map_submap: punch down part of a map into a submap 2192 * 2193 * => only the kernel_map is allowed to be submapped 2194 * => the purpose of submapping is to break up the locking granularity 2195 * of a larger map 2196 * => the range specified must have been mapped previously with a uvm_map() 2197 * call [with uobj==NULL] to create a blank map entry in the main map. 2198 * [And it had better still be blank!] 2199 * => maps which contain submaps should never be copied or forked. 2200 * => to remove a submap, use uvm_unmap() on the main map 2201 * and then uvm_map_deallocate() the submap. 2202 * => main map must be unlocked. 2203 * => submap must have been init'd and have a zero reference count. 2204 * [need not be locked as we don't actually reference it] 2205 */ 2206 2207 int 2208 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 2209 struct vm_map *submap) 2210 { 2211 struct vm_map_entry *entry; 2212 int result; 2213 2214 vm_map_lock(map); 2215 2216 VM_MAP_RANGE_CHECK(map, start, end); 2217 2218 if (uvm_map_lookup_entry(map, start, &entry)) { 2219 UVM_MAP_CLIP_START(map, entry, start); 2220 UVM_MAP_CLIP_END(map, entry, end); /* to be safe */ 2221 } else { 2222 entry = NULL; 2223 } 2224 2225 if (entry != NULL && 2226 entry->start == start && entry->end == end && 2227 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 2228 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 2229 entry->etype |= UVM_ET_SUBMAP; 2230 entry->object.sub_map = submap; 2231 entry->offset = 0; 2232 uvm_map_reference(submap); 2233 result = 0; 2234 } else { 2235 result = EINVAL; 2236 } 2237 vm_map_unlock(map); 2238 return(result); 2239 } 2240 2241 2242 /* 2243 * uvm_map_protect: change map protection 2244 * 2245 * => set_max means set max_protection. 2246 * => map must be unlocked. 2247 */ 2248 2249 #define MASK(entry) (UVM_ET_ISCOPYONWRITE(entry) ? \ 2250 ~VM_PROT_WRITE : VM_PROT_ALL) 2251 #define max(a,b) ((a) > (b) ? (a) : (b)) 2252 2253 int 2254 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 2255 vm_prot_t new_prot, boolean_t set_max) 2256 { 2257 struct vm_map_entry *current, *entry; 2258 int error = 0; 2259 UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist); 2260 UVMHIST_LOG(maphist,"(map=%p,start=0x%lx,end=0x%lx,new_prot=0x%lx)", 2261 map, start, end, new_prot); 2262 2263 vm_map_lock(map); 2264 2265 VM_MAP_RANGE_CHECK(map, start, end); 2266 2267 if (uvm_map_lookup_entry(map, start, &entry)) { 2268 UVM_MAP_CLIP_START(map, entry, start); 2269 } else { 2270 entry = entry->next; 2271 } 2272 2273 /* 2274 * make a first pass to check for protection violations. 2275 */ 2276 2277 current = entry; 2278 while ((current != &map->header) && (current->start < end)) { 2279 if (UVM_ET_ISSUBMAP(current)) { 2280 error = EINVAL; 2281 goto out; 2282 } 2283 if ((new_prot & current->max_protection) != new_prot) { 2284 error = EACCES; 2285 goto out; 2286 } 2287 current = current->next; 2288 } 2289 2290 /* go back and fix up protections (no need to clip this time). */ 2291 2292 current = entry; 2293 2294 while ((current != &map->header) && (current->start < end)) { 2295 vm_prot_t old_prot; 2296 2297 UVM_MAP_CLIP_END(map, current, end); 2298 2299 old_prot = current->protection; 2300 if (set_max) 2301 current->protection = 2302 (current->max_protection = new_prot) & old_prot; 2303 else 2304 current->protection = new_prot; 2305 2306 /* 2307 * update physical map if necessary. worry about copy-on-write 2308 * here -- CHECK THIS XXX 2309 */ 2310 2311 if (current->protection != old_prot) { 2312 /* update pmap! */ 2313 if ((current->protection & MASK(entry)) == PROT_NONE && 2314 VM_MAPENT_ISWIRED(entry)) 2315 current->wired_count--; 2316 pmap_protect(map->pmap, current->start, current->end, 2317 current->protection & MASK(entry)); 2318 } 2319 2320 /* 2321 * If the map is configured to lock any future mappings, 2322 * wire this entry now if the old protection was VM_PROT_NONE 2323 * and the new protection is not VM_PROT_NONE. 2324 */ 2325 2326 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 2327 VM_MAPENT_ISWIRED(entry) == 0 && 2328 old_prot == VM_PROT_NONE && 2329 new_prot != VM_PROT_NONE) { 2330 if (uvm_map_pageable(map, entry->start, entry->end, 2331 FALSE, UVM_LK_ENTER|UVM_LK_EXIT) != 0) { 2332 /* 2333 * If locking the entry fails, remember the 2334 * error if it's the first one. Note we 2335 * still continue setting the protection in 2336 * the map, but will return the resource 2337 * shortage condition regardless. 2338 * 2339 * XXX Ignore what the actual error is, 2340 * XXX just call it a resource shortage 2341 * XXX so that it doesn't get confused 2342 * XXX what uvm_map_protect() itself would 2343 * XXX normally return. 2344 */ 2345 error = ENOMEM; 2346 } 2347 } 2348 2349 current = current->next; 2350 } 2351 pmap_update(map->pmap); 2352 2353 out: 2354 vm_map_unlock(map); 2355 UVMHIST_LOG(maphist, "<- done, rv=%ld",error,0,0,0); 2356 return (error); 2357 } 2358 2359 #undef max 2360 #undef MASK 2361 2362 /* 2363 * uvm_map_inherit: set inheritance code for range of addrs in map. 2364 * 2365 * => map must be unlocked 2366 * => note that the inherit code is used during a "fork". see fork 2367 * code for details. 2368 */ 2369 2370 int 2371 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 2372 vm_inherit_t new_inheritance) 2373 { 2374 struct vm_map_entry *entry, *temp_entry; 2375 UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist); 2376 UVMHIST_LOG(maphist,"(map=%p,start=0x%lx,end=0x%lx,new_inh=0x%lx)", 2377 map, start, end, new_inheritance); 2378 2379 switch (new_inheritance) { 2380 case MAP_INHERIT_NONE: 2381 case MAP_INHERIT_COPY: 2382 case MAP_INHERIT_SHARE: 2383 break; 2384 default: 2385 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2386 return (EINVAL); 2387 } 2388 2389 vm_map_lock(map); 2390 2391 VM_MAP_RANGE_CHECK(map, start, end); 2392 2393 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2394 entry = temp_entry; 2395 UVM_MAP_CLIP_START(map, entry, start); 2396 } else { 2397 entry = temp_entry->next; 2398 } 2399 2400 while ((entry != &map->header) && (entry->start < end)) { 2401 UVM_MAP_CLIP_END(map, entry, end); 2402 entry->inheritance = new_inheritance; 2403 entry = entry->next; 2404 } 2405 2406 vm_map_unlock(map); 2407 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2408 return (0); 2409 } 2410 2411 /* 2412 * uvm_map_advice: set advice code for range of addrs in map. 2413 * 2414 * => map must be unlocked 2415 */ 2416 2417 int 2418 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 2419 { 2420 struct vm_map_entry *entry, *temp_entry; 2421 UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist); 2422 UVMHIST_LOG(maphist,"(map=%p,start=0x%lx,end=0x%lx,new_adv=0x%lx)", 2423 map, start, end, new_advice); 2424 2425 vm_map_lock(map); 2426 VM_MAP_RANGE_CHECK(map, start, end); 2427 if (uvm_map_lookup_entry(map, start, &temp_entry)) { 2428 entry = temp_entry; 2429 UVM_MAP_CLIP_START(map, entry, start); 2430 } else { 2431 entry = temp_entry->next; 2432 } 2433 2434 /* 2435 * XXXJRT: disallow holes? 2436 */ 2437 2438 while ((entry != &map->header) && (entry->start < end)) { 2439 UVM_MAP_CLIP_END(map, entry, end); 2440 2441 switch (new_advice) { 2442 case MADV_NORMAL: 2443 case MADV_RANDOM: 2444 case MADV_SEQUENTIAL: 2445 /* nothing special here */ 2446 break; 2447 2448 default: 2449 vm_map_unlock(map); 2450 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2451 return (EINVAL); 2452 } 2453 entry->advice = new_advice; 2454 entry = entry->next; 2455 } 2456 2457 vm_map_unlock(map); 2458 UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); 2459 return (0); 2460 } 2461 2462 /* 2463 * uvm_map_pageable: sets the pageability of a range in a map. 2464 * 2465 * => wires map entries. should not be used for transient page locking. 2466 * for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()). 2467 * => regions sepcified as not pageable require lock-down (wired) memory 2468 * and page tables. 2469 * => map must never be read-locked 2470 * => if islocked is TRUE, map is already write-locked 2471 * => we always unlock the map, since we must downgrade to a read-lock 2472 * to call uvm_fault_wire() 2473 * => XXXCDC: check this and try and clean it up. 2474 */ 2475 2476 int 2477 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 2478 boolean_t new_pageable, int lockflags) 2479 { 2480 struct vm_map_entry *entry, *start_entry, *failed_entry; 2481 int rv; 2482 #ifdef DIAGNOSTIC 2483 u_int timestamp_save; 2484 #endif 2485 UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist); 2486 UVMHIST_LOG(maphist,"(map=%p,start=0x%lx,end=0x%lx,new_pageable=0x%lx)", 2487 map, start, end, new_pageable); 2488 KASSERT(map->flags & VM_MAP_PAGEABLE); 2489 2490 if ((lockflags & UVM_LK_ENTER) == 0) 2491 vm_map_lock(map); 2492 2493 VM_MAP_RANGE_CHECK(map, start, end); 2494 2495 /* 2496 * only one pageability change may take place at one time, since 2497 * uvm_fault_wire assumes it will be called only once for each 2498 * wiring/unwiring. therefore, we have to make sure we're actually 2499 * changing the pageability for the entire region. we do so before 2500 * making any changes. 2501 */ 2502 2503 if (uvm_map_lookup_entry(map, start, &start_entry) == FALSE) { 2504 if ((lockflags & UVM_LK_EXIT) == 0) 2505 vm_map_unlock(map); 2506 2507 UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0); 2508 return (EFAULT); 2509 } 2510 entry = start_entry; 2511 2512 /* 2513 * handle wiring and unwiring separately. 2514 */ 2515 2516 if (new_pageable) { /* unwire */ 2517 UVM_MAP_CLIP_START(map, entry, start); 2518 2519 /* 2520 * unwiring. first ensure that the range to be unwired is 2521 * really wired down and that there are no holes. 2522 */ 2523 2524 while ((entry != &map->header) && (entry->start < end)) { 2525 if (entry->wired_count == 0 || 2526 (entry->end < end && 2527 (entry->next == &map->header || 2528 entry->next->start > entry->end))) { 2529 if ((lockflags & UVM_LK_EXIT) == 0) 2530 vm_map_unlock(map); 2531 UVMHIST_LOG(maphist, 2532 "<- done (INVALID UNWIRE ARG)",0,0,0,0); 2533 return (EINVAL); 2534 } 2535 entry = entry->next; 2536 } 2537 2538 /* 2539 * POSIX 1003.1b - a single munlock call unlocks a region, 2540 * regardless of the number of mlock calls made on that 2541 * region. 2542 */ 2543 2544 entry = start_entry; 2545 while ((entry != &map->header) && (entry->start < end)) { 2546 UVM_MAP_CLIP_END(map, entry, end); 2547 if (VM_MAPENT_ISWIRED(entry)) 2548 uvm_map_entry_unwire(map, entry); 2549 entry = entry->next; 2550 } 2551 if ((lockflags & UVM_LK_EXIT) == 0) 2552 vm_map_unlock(map); 2553 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2554 return (0); 2555 } 2556 2557 /* 2558 * wire case: in two passes [XXXCDC: ugly block of code here] 2559 * 2560 * 1: holding the write lock, we create any anonymous maps that need 2561 * to be created. then we clip each map entry to the region to 2562 * be wired and increment its wiring count. 2563 * 2564 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2565 * in the pages for any newly wired area (wired_count == 1). 2566 * 2567 * downgrading to a read lock for uvm_fault_wire avoids a possible 2568 * deadlock with another thread that may have faulted on one of 2569 * the pages to be wired (it would mark the page busy, blocking 2570 * us, then in turn block on the map lock that we hold). because 2571 * of problems in the recursive lock package, we cannot upgrade 2572 * to a write lock in vm_map_lookup. thus, any actions that 2573 * require the write lock must be done beforehand. because we 2574 * keep the read lock on the map, the copy-on-write status of the 2575 * entries we modify here cannot change. 2576 */ 2577 2578 while ((entry != &map->header) && (entry->start < end)) { 2579 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2580 2581 /* 2582 * perform actions of vm_map_lookup that need the 2583 * write lock on the map: create an anonymous map 2584 * for a copy-on-write region, or an anonymous map 2585 * for a zero-fill region. (XXXCDC: submap case 2586 * ok?) 2587 */ 2588 2589 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2590 if (UVM_ET_ISNEEDSCOPY(entry) && 2591 ((entry->protection & VM_PROT_WRITE) || 2592 (entry->object.uvm_obj == NULL))) { 2593 amap_copy(map, entry, M_WAITOK, TRUE, 2594 start, end); 2595 /* XXXCDC: wait OK? */ 2596 } 2597 } 2598 } 2599 UVM_MAP_CLIP_START(map, entry, start); 2600 UVM_MAP_CLIP_END(map, entry, end); 2601 entry->wired_count++; 2602 2603 /* 2604 * Check for holes 2605 */ 2606 2607 if (entry->protection == VM_PROT_NONE || 2608 (entry->end < end && 2609 (entry->next == &map->header || 2610 entry->next->start > entry->end))) { 2611 2612 /* 2613 * found one. amap creation actions do not need to 2614 * be undone, but the wired counts need to be restored. 2615 */ 2616 2617 while (entry != &map->header && entry->end > start) { 2618 entry->wired_count--; 2619 entry = entry->prev; 2620 } 2621 if ((lockflags & UVM_LK_EXIT) == 0) 2622 vm_map_unlock(map); 2623 UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0); 2624 return (EINVAL); 2625 } 2626 entry = entry->next; 2627 } 2628 2629 /* 2630 * Pass 2. 2631 */ 2632 2633 #ifdef DIAGNOSTIC 2634 timestamp_save = map->timestamp; 2635 #endif 2636 vm_map_busy(map); 2637 vm_map_downgrade(map); 2638 2639 rv = 0; 2640 entry = start_entry; 2641 while (entry != &map->header && entry->start < end) { 2642 if (entry->wired_count == 1) { 2643 rv = uvm_fault_wire(map, entry->start, entry->end, 2644 entry->protection); 2645 if (rv) { 2646 /* 2647 * wiring failed. break out of the loop. 2648 * we'll clean up the map below, once we 2649 * have a write lock again. 2650 */ 2651 break; 2652 } 2653 } 2654 entry = entry->next; 2655 } 2656 2657 if (rv) { /* failed? */ 2658 2659 /* 2660 * Get back to an exclusive (write) lock. 2661 */ 2662 2663 vm_map_upgrade(map); 2664 vm_map_unbusy(map); 2665 2666 #ifdef DIAGNOSTIC 2667 if (timestamp_save != map->timestamp) 2668 panic("uvm_map_pageable: stale map"); 2669 #endif 2670 2671 /* 2672 * first drop the wiring count on all the entries 2673 * which haven't actually been wired yet. 2674 */ 2675 2676 failed_entry = entry; 2677 while (entry != &map->header && entry->start < end) { 2678 entry->wired_count--; 2679 entry = entry->next; 2680 } 2681 2682 /* 2683 * now, unwire all the entries that were successfully 2684 * wired above. 2685 */ 2686 2687 entry = start_entry; 2688 while (entry != failed_entry) { 2689 entry->wired_count--; 2690 if (VM_MAPENT_ISWIRED(entry) == 0) 2691 uvm_map_entry_unwire(map, entry); 2692 entry = entry->next; 2693 } 2694 if ((lockflags & UVM_LK_EXIT) == 0) 2695 vm_map_unlock(map); 2696 UVMHIST_LOG(maphist, "<- done (RV=%ld)", rv,0,0,0); 2697 return(rv); 2698 } 2699 2700 /* We are holding a read lock here. */ 2701 if ((lockflags & UVM_LK_EXIT) == 0) { 2702 vm_map_unbusy(map); 2703 vm_map_unlock_read(map); 2704 } else { 2705 2706 /* 2707 * Get back to an exclusive (write) lock. 2708 */ 2709 2710 vm_map_upgrade(map); 2711 vm_map_unbusy(map); 2712 } 2713 2714 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2715 return (0); 2716 } 2717 2718 /* 2719 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2720 * all mapped regions. 2721 * 2722 * => map must not be locked. 2723 * => if no flags are specified, all regions are unwired. 2724 * => XXXJRT: has some of the same problems as uvm_map_pageable() above. 2725 */ 2726 2727 int 2728 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 2729 { 2730 struct vm_map_entry *entry, *failed_entry; 2731 vsize_t size; 2732 int error; 2733 #ifdef DIAGNOSTIC 2734 u_int timestamp_save; 2735 #endif 2736 UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist); 2737 UVMHIST_LOG(maphist,"(map=%p,flags=0x%lx)", map, flags, 0, 0); 2738 2739 KASSERT(map->flags & VM_MAP_PAGEABLE); 2740 2741 vm_map_lock(map); 2742 2743 /* 2744 * handle wiring and unwiring separately. 2745 */ 2746 2747 if (flags == 0) { /* unwire */ 2748 /* 2749 * POSIX 1003.1b -- munlockall unlocks all regions, 2750 * regardless of how many times mlockall has been called. 2751 */ 2752 for (entry = map->header.next; entry != &map->header; 2753 entry = entry->next) { 2754 if (VM_MAPENT_ISWIRED(entry)) 2755 uvm_map_entry_unwire(map, entry); 2756 } 2757 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2758 vm_map_unlock(map); 2759 UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0); 2760 return (0); 2761 2762 /* 2763 * end of unwire case! 2764 */ 2765 } 2766 2767 if (flags & MCL_FUTURE) { 2768 /* 2769 * must wire all future mappings; remember this. 2770 */ 2771 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2772 } 2773 2774 if ((flags & MCL_CURRENT) == 0) { 2775 /* 2776 * no more work to do! 2777 */ 2778 UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0); 2779 vm_map_unlock(map); 2780 return (0); 2781 } 2782 2783 /* 2784 * wire case: in three passes [XXXCDC: ugly block of code here] 2785 * 2786 * 1: holding the write lock, count all pages mapped by non-wired 2787 * entries. if this would cause us to go over our limit, we fail. 2788 * 2789 * 2: still holding the write lock, we create any anonymous maps that 2790 * need to be created. then we increment its wiring count. 2791 * 2792 * 3: we downgrade to a read lock, and call uvm_fault_wire to fault 2793 * in the pages for any newly wired area (wired_count == 1). 2794 * 2795 * downgrading to a read lock for uvm_fault_wire avoids a possible 2796 * deadlock with another thread that may have faulted on one of 2797 * the pages to be wired (it would mark the page busy, blocking 2798 * us, then in turn block on the map lock that we hold). because 2799 * of problems in the recursive lock package, we cannot upgrade 2800 * to a write lock in vm_map_lookup. thus, any actions that 2801 * require the write lock must be done beforehand. because we 2802 * keep the read lock on the map, the copy-on-write status of the 2803 * entries we modify here cannot change. 2804 */ 2805 2806 for (size = 0, entry = map->header.next; entry != &map->header; 2807 entry = entry->next) { 2808 if (entry->protection != VM_PROT_NONE && 2809 VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2810 size += entry->end - entry->start; 2811 } 2812 } 2813 2814 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2815 vm_map_unlock(map); 2816 return (ENOMEM); /* XXX overloaded */ 2817 } 2818 2819 /* XXX non-pmap_wired_count case must be handled by caller */ 2820 #ifdef pmap_wired_count 2821 if (limit != 0 && 2822 (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) { 2823 vm_map_unlock(map); 2824 return (ENOMEM); /* XXX overloaded */ 2825 } 2826 #endif 2827 2828 /* 2829 * Pass 2. 2830 */ 2831 2832 for (entry = map->header.next; entry != &map->header; 2833 entry = entry->next) { 2834 if (entry->protection == VM_PROT_NONE) 2835 continue; 2836 if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */ 2837 /* 2838 * perform actions of vm_map_lookup that need the 2839 * write lock on the map: create an anonymous map 2840 * for a copy-on-write region, or an anonymous map 2841 * for a zero-fill region. (XXXCDC: submap case 2842 * ok?) 2843 */ 2844 if (!UVM_ET_ISSUBMAP(entry)) { /* not submap */ 2845 if (UVM_ET_ISNEEDSCOPY(entry) && 2846 ((entry->protection & VM_PROT_WRITE) || 2847 (entry->object.uvm_obj == NULL))) { 2848 amap_copy(map, entry, M_WAITOK, TRUE, 2849 entry->start, entry->end); 2850 /* XXXCDC: wait OK? */ 2851 } 2852 } 2853 } 2854 entry->wired_count++; 2855 } 2856 2857 /* 2858 * Pass 3. 2859 */ 2860 2861 #ifdef DIAGNOSTIC 2862 timestamp_save = map->timestamp; 2863 #endif 2864 vm_map_busy(map); 2865 vm_map_downgrade(map); 2866 2867 for (error = 0, entry = map->header.next; 2868 entry != &map->header && error == 0; 2869 entry = entry->next) { 2870 if (entry->wired_count == 1) { 2871 error = uvm_fault_wire(map, entry->start, entry->end, 2872 entry->protection); 2873 } 2874 } 2875 2876 if (error) { /* failed? */ 2877 /* 2878 * Get back an exclusive (write) lock. 2879 */ 2880 vm_map_upgrade(map); 2881 vm_map_unbusy(map); 2882 2883 #ifdef DIAGNOSTIC 2884 if (timestamp_save != map->timestamp) 2885 panic("uvm_map_pageable_all: stale map"); 2886 #endif 2887 2888 /* 2889 * first drop the wiring count on all the entries 2890 * which haven't actually been wired yet. 2891 * 2892 * Skip VM_PROT_NONE entries like we did above. 2893 */ 2894 failed_entry = entry; 2895 for (/* nothing */; entry != &map->header; 2896 entry = entry->next) { 2897 if (entry->protection == VM_PROT_NONE) 2898 continue; 2899 entry->wired_count--; 2900 } 2901 2902 /* 2903 * now, unwire all the entries that were successfully 2904 * wired above. 2905 * 2906 * Skip VM_PROT_NONE entries like we did above. 2907 */ 2908 for (entry = map->header.next; entry != failed_entry; 2909 entry = entry->next) { 2910 if (entry->protection == VM_PROT_NONE) 2911 continue; 2912 entry->wired_count--; 2913 if (VM_MAPENT_ISWIRED(entry)) 2914 uvm_map_entry_unwire(map, entry); 2915 } 2916 vm_map_unlock(map); 2917 UVMHIST_LOG(maphist,"<- done (RV=%ld)", error,0,0,0); 2918 return (error); 2919 } 2920 2921 /* We are holding a read lock here. */ 2922 vm_map_unbusy(map); 2923 vm_map_unlock_read(map); 2924 2925 UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0); 2926 return (0); 2927 } 2928 2929 /* 2930 * uvm_map_clean: clean out a map range 2931 * 2932 * => valid flags: 2933 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 2934 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 2935 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 2936 * if (flags & PGO_FREE): any cached pages are freed after clean 2937 * => returns an error if any part of the specified range isn't mapped 2938 * => never a need to flush amap layer since the anonymous memory has 2939 * no permanent home, but may deactivate pages there 2940 * => called from sys_msync() and sys_madvise() 2941 * => caller must not write-lock map (read OK). 2942 * => we may sleep while cleaning if SYNCIO [with map read-locked] 2943 */ 2944 2945 int amap_clean_works = 1; /* XXX for now, just in case... */ 2946 2947 int 2948 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 2949 { 2950 struct vm_map_entry *current, *entry; 2951 struct uvm_object *uobj; 2952 struct vm_amap *amap; 2953 struct vm_anon *anon; 2954 struct vm_page *pg; 2955 vaddr_t offset; 2956 vsize_t size; 2957 int rv, error, refs; 2958 UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist); 2959 2960 UVMHIST_LOG(maphist,"(map=%p,start=0x%lx,end=0x%lx,flags=0x%lx)", 2961 map, start, end, flags); 2962 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 2963 (PGO_FREE|PGO_DEACTIVATE)); 2964 2965 vm_map_lock_read(map); 2966 VM_MAP_RANGE_CHECK(map, start, end); 2967 if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { 2968 vm_map_unlock_read(map); 2969 return (EFAULT); 2970 } 2971 2972 /* 2973 * Make a first pass to check for holes. 2974 */ 2975 2976 for (current = entry; current->start < end; current = current->next) { 2977 if (UVM_ET_ISSUBMAP(current)) { 2978 vm_map_unlock_read(map); 2979 return (EINVAL); 2980 } 2981 if (end > current->end && (current->next == &map->header || 2982 current->end != current->next->start)) { 2983 vm_map_unlock_read(map); 2984 return (EFAULT); 2985 } 2986 } 2987 2988 error = 0; 2989 2990 for (current = entry; current->start < end; current = current->next) { 2991 amap = current->aref.ar_amap; /* top layer */ 2992 uobj = current->object.uvm_obj; /* bottom layer */ 2993 KASSERT(start >= current->start); 2994 2995 /* 2996 * No amap cleaning necessary if: 2997 * 2998 * (1) There's no amap. 2999 * 3000 * (2) We're not deactivating or freeing pages. 3001 */ 3002 3003 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 3004 goto flush_object; 3005 3006 /* XXX for now, just in case... */ 3007 if (amap_clean_works == 0) 3008 goto flush_object; 3009 3010 offset = start - current->start; 3011 size = MIN(end, current->end) - start; 3012 for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) { 3013 anon = amap_lookup(¤t->aref, offset); 3014 if (anon == NULL) 3015 continue; 3016 3017 simple_lock(&anon->an_lock); 3018 3019 pg = anon->an_page; 3020 if (pg == NULL) { 3021 simple_unlock(&anon->an_lock); 3022 continue; 3023 } 3024 3025 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 3026 3027 /* 3028 * XXX In these first 3 cases, we always just 3029 * XXX deactivate the page. We may want to 3030 * XXX handle the different cases more 3031 * XXX specifically, in the future. 3032 */ 3033 3034 case PGO_CLEANIT|PGO_FREE: 3035 case PGO_CLEANIT|PGO_DEACTIVATE: 3036 case PGO_DEACTIVATE: 3037 deactivate_it: 3038 /* skip the page if it's loaned or wired */ 3039 if (pg->loan_count != 0 || 3040 pg->wire_count != 0) { 3041 simple_unlock(&anon->an_lock); 3042 continue; 3043 } 3044 3045 uvm_lock_pageq(); 3046 3047 /* 3048 * skip the page if it's not actually owned 3049 * by the anon (may simply be loaned to the 3050 * anon). 3051 */ 3052 3053 if ((pg->pg_flags & PQ_ANON) == 0) { 3054 KASSERT(pg->uobject == NULL); 3055 uvm_unlock_pageq(); 3056 simple_unlock(&anon->an_lock); 3057 continue; 3058 } 3059 KASSERT(pg->uanon == anon); 3060 3061 #ifdef UBC 3062 /* ...and deactivate the page. */ 3063 pmap_clear_reference(pg); 3064 #else 3065 /* zap all mappings for the page. */ 3066 pmap_page_protect(pg, VM_PROT_NONE); 3067 3068 /* ...and deactivate the page. */ 3069 #endif 3070 uvm_pagedeactivate(pg); 3071 3072 uvm_unlock_pageq(); 3073 simple_unlock(&anon->an_lock); 3074 continue; 3075 3076 case PGO_FREE: 3077 3078 /* 3079 * If there are multiple references to 3080 * the amap, just deactivate the page. 3081 */ 3082 3083 if (amap_refs(amap) > 1) 3084 goto deactivate_it; 3085 3086 /* XXX skip the page if it's wired */ 3087 if (pg->wire_count != 0) { 3088 simple_unlock(&anon->an_lock); 3089 continue; 3090 } 3091 amap_unadd(¤t->aref, offset); 3092 refs = --anon->an_ref; 3093 simple_unlock(&anon->an_lock); 3094 if (refs == 0) 3095 uvm_anfree(anon); 3096 continue; 3097 3098 default: 3099 panic("uvm_map_clean: weird flags"); 3100 } 3101 } 3102 3103 flush_object: 3104 /* 3105 * flush pages if we've got a valid backing object. 3106 * 3107 * Don't PGO_FREE if we don't have write permission 3108 * and don't flush if this is a copy-on-write object 3109 * since we can't know our permissions on it. 3110 */ 3111 3112 offset = current->offset + (start - current->start); 3113 size = MIN(end, current->end) - start; 3114 if (uobj != NULL && 3115 ((flags & PGO_FREE) == 0 || 3116 ((entry->max_protection & VM_PROT_WRITE) != 0 && 3117 (entry->etype & UVM_ET_COPYONWRITE) == 0))) { 3118 simple_lock(&uobj->vmobjlock); 3119 rv = uobj->pgops->pgo_flush(uobj, offset, 3120 offset + size, flags); 3121 simple_unlock(&uobj->vmobjlock); 3122 3123 if (rv == FALSE) 3124 error = EFAULT; 3125 } 3126 start += size; 3127 } 3128 vm_map_unlock_read(map); 3129 return (error); 3130 } 3131 3132 3133 /* 3134 * uvm_map_checkprot: check protection in map 3135 * 3136 * => must allow specified protection in a fully allocated region. 3137 * => map must be read or write locked by caller. 3138 */ 3139 3140 boolean_t 3141 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 3142 vm_prot_t protection) 3143 { 3144 struct vm_map_entry *entry; 3145 struct vm_map_entry *tmp_entry; 3146 3147 if (!uvm_map_lookup_entry(map, start, &tmp_entry)) { 3148 return(FALSE); 3149 } 3150 entry = tmp_entry; 3151 while (start < end) { 3152 if (entry == &map->header) { 3153 return(FALSE); 3154 } 3155 3156 /* 3157 * no holes allowed 3158 */ 3159 3160 if (start < entry->start) { 3161 return(FALSE); 3162 } 3163 3164 /* 3165 * check protection associated with entry 3166 */ 3167 3168 if ((entry->protection & protection) != protection) { 3169 return(FALSE); 3170 } 3171 3172 /* go to next entry */ 3173 3174 start = entry->end; 3175 entry = entry->next; 3176 } 3177 return(TRUE); 3178 } 3179 3180 /* 3181 * uvmspace_alloc: allocate a vmspace structure. 3182 * 3183 * - structure includes vm_map and pmap 3184 * - XXX: no locking on this structure 3185 * - refcnt set to 1, rest must be init'd by caller 3186 */ 3187 struct vmspace * 3188 uvmspace_alloc(vaddr_t min, vaddr_t max, boolean_t pageable, 3189 boolean_t remove_holes) 3190 { 3191 struct vmspace *vm; 3192 UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist); 3193 3194 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK); 3195 uvmspace_init(vm, NULL, min, max, pageable, remove_holes); 3196 UVMHIST_LOG(maphist,"<- done (vm=%p)", vm,0,0,0); 3197 return (vm); 3198 } 3199 3200 /* 3201 * uvmspace_init: initialize a vmspace structure. 3202 * 3203 * - XXX: no locking on this structure 3204 * - refcnt set to 1, rest must be init'd by caller 3205 */ 3206 void 3207 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max, 3208 boolean_t pageable, boolean_t remove_holes) 3209 { 3210 UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist); 3211 3212 memset(vm, 0, sizeof(*vm)); 3213 3214 uvm_map_setup(&vm->vm_map, min, max, pageable ? VM_MAP_PAGEABLE : 0); 3215 3216 if (pmap) 3217 pmap_reference(pmap); 3218 else 3219 pmap = pmap_create(); 3220 vm->vm_map.pmap = pmap; 3221 3222 vm->vm_refcnt = 1; 3223 3224 if (remove_holes) 3225 pmap_remove_holes(&vm->vm_map); 3226 3227 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3228 } 3229 3230 /* 3231 * uvmspace_share: share a vmspace between two proceses 3232 * 3233 * - XXX: no locking on vmspace 3234 * - used for vfork, threads(?) 3235 */ 3236 3237 void 3238 uvmspace_share(p1, p2) 3239 struct proc *p1, *p2; 3240 { 3241 p2->p_vmspace = p1->p_vmspace; 3242 p1->p_vmspace->vm_refcnt++; 3243 } 3244 3245 /* 3246 * uvmspace_exec: the process wants to exec a new program 3247 * 3248 * - XXX: no locking on vmspace 3249 */ 3250 3251 void 3252 uvmspace_exec(struct proc *p, vaddr_t start, vaddr_t end) 3253 { 3254 struct vmspace *nvm, *ovm = p->p_vmspace; 3255 struct vm_map *map = &ovm->vm_map; 3256 3257 pmap_unuse_final(p); /* before stack addresses go away */ 3258 3259 /* 3260 * see if more than one process is using this vmspace... 3261 */ 3262 3263 if (ovm->vm_refcnt == 1) { 3264 3265 /* 3266 * if p is the only process using its vmspace then we can safely 3267 * recycle that vmspace for the program that is being exec'd. 3268 */ 3269 3270 #ifdef SYSVSHM 3271 /* 3272 * SYSV SHM semantics require us to kill all segments on an exec 3273 */ 3274 if (ovm->vm_shm) 3275 shmexit(ovm); 3276 #endif 3277 3278 /* 3279 * POSIX 1003.1b -- "lock future mappings" is revoked 3280 * when a process execs another program image. 3281 */ 3282 vm_map_lock(map); 3283 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 3284 vm_map_unlock(map); 3285 3286 /* 3287 * now unmap the old program 3288 */ 3289 uvm_unmap(map, map->min_offset, map->max_offset); 3290 3291 /* 3292 * but keep MMU holes unavailable 3293 */ 3294 pmap_remove_holes(map); 3295 3296 /* 3297 * resize the map 3298 */ 3299 vm_map_lock(map); 3300 map->min_offset = start; 3301 uvm_tree_sanity(map, "resize enter"); 3302 map->max_offset = end; 3303 if (map->header.prev != &map->header) 3304 uvm_rb_fixup(map, map->header.prev); 3305 uvm_tree_sanity(map, "resize leave"); 3306 vm_map_unlock(map); 3307 3308 3309 } else { 3310 3311 /* 3312 * p's vmspace is being shared, so we can't reuse it for p since 3313 * it is still being used for others. allocate a new vmspace 3314 * for p 3315 */ 3316 nvm = uvmspace_alloc(start, end, 3317 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, TRUE); 3318 3319 /* 3320 * install new vmspace and drop our ref to the old one. 3321 */ 3322 3323 pmap_deactivate(p); 3324 p->p_vmspace = nvm; 3325 pmap_activate(p); 3326 3327 uvmspace_free(ovm); 3328 } 3329 } 3330 3331 /* 3332 * uvmspace_free: free a vmspace data structure 3333 * 3334 * - XXX: no locking on vmspace 3335 */ 3336 3337 void 3338 uvmspace_free(struct vmspace *vm) 3339 { 3340 struct vm_map_entry *dead_entries; 3341 UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist); 3342 3343 UVMHIST_LOG(maphist,"(vm=%p) ref=%ld", vm, vm->vm_refcnt,0,0); 3344 if (--vm->vm_refcnt == 0) { 3345 /* 3346 * lock the map, to wait out all other references to it. delete 3347 * all of the mappings and pages they hold, then call the pmap 3348 * module to reclaim anything left. 3349 */ 3350 #ifdef SYSVSHM 3351 /* Get rid of any SYSV shared memory segments. */ 3352 if (vm->vm_shm != NULL) 3353 shmexit(vm); 3354 #endif 3355 vm_map_lock(&vm->vm_map); 3356 if (vm->vm_map.nentries) { 3357 uvm_unmap_remove(&vm->vm_map, 3358 vm->vm_map.min_offset, vm->vm_map.max_offset, 3359 &dead_entries, NULL); 3360 if (dead_entries != NULL) 3361 uvm_unmap_detach(dead_entries, 0); 3362 } 3363 pmap_destroy(vm->vm_map.pmap); 3364 vm->vm_map.pmap = NULL; 3365 pool_put(&uvm_vmspace_pool, vm); 3366 } 3367 UVMHIST_LOG(maphist,"<- done", 0,0,0,0); 3368 } 3369 3370 /* 3371 * uvm_map_create: create map 3372 */ 3373 vm_map_t 3374 uvm_map_create(pmap_t pmap, vaddr_t min, vaddr_t max, int flags) 3375 { 3376 vm_map_t result; 3377 3378 result = malloc(sizeof(struct vm_map), M_VMMAP, M_WAITOK); 3379 uvm_map_setup(result, min, max, flags); 3380 result->pmap = pmap; 3381 return(result); 3382 } 3383 3384 /* 3385 * uvm_map_setup: init map 3386 * 3387 * => map must not be in service yet. 3388 */ 3389 void 3390 uvm_map_setup(vm_map_t map, vaddr_t min, vaddr_t max, int flags) 3391 { 3392 3393 RB_INIT(&map->rbhead); 3394 map->header.next = map->header.prev = &map->header; 3395 map->nentries = 0; 3396 map->size = 0; 3397 map->ref_count = 1; 3398 map->min_offset = min; 3399 map->max_offset = max; 3400 map->flags = flags; 3401 map->first_free = &map->header; 3402 map->hint = &map->header; 3403 map->timestamp = 0; 3404 rw_init(&map->lock, "vmmaplk"); 3405 simple_lock_init(&map->ref_lock); 3406 simple_lock_init(&map->hint_lock); 3407 } 3408 3409 3410 3411 /* 3412 * uvm_map_reference: add reference to a map 3413 * 3414 * => map need not be locked (we use ref_lock). 3415 */ 3416 void 3417 uvm_map_reference(vm_map_t map) 3418 { 3419 simple_lock(&map->ref_lock); 3420 map->ref_count++; 3421 simple_unlock(&map->ref_lock); 3422 } 3423 3424 /* 3425 * uvm_map_deallocate: drop reference to a map 3426 * 3427 * => caller must not lock map 3428 * => we will zap map if ref count goes to zero 3429 */ 3430 void 3431 uvm_map_deallocate(vm_map_t map) 3432 { 3433 int c; 3434 3435 simple_lock(&map->ref_lock); 3436 c = --map->ref_count; 3437 simple_unlock(&map->ref_lock); 3438 if (c > 0) { 3439 return; 3440 } 3441 3442 /* 3443 * all references gone. unmap and free. 3444 */ 3445 3446 uvm_unmap(map, map->min_offset, map->max_offset); 3447 pmap_destroy(map->pmap); 3448 free(map, M_VMMAP); 3449 } 3450 3451 /* 3452 * F O R K - m a i n e n t r y p o i n t 3453 */ 3454 /* 3455 * uvmspace_fork: fork a process' main map 3456 * 3457 * => create a new vmspace for child process from parent. 3458 * => parent's map must not be locked. 3459 */ 3460 3461 struct vmspace * 3462 uvmspace_fork(struct vmspace *vm1) 3463 { 3464 struct vmspace *vm2; 3465 struct vm_map *old_map = &vm1->vm_map; 3466 struct vm_map *new_map; 3467 struct vm_map_entry *old_entry; 3468 struct vm_map_entry *new_entry; 3469 pmap_t new_pmap; 3470 boolean_t protect_child; 3471 UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); 3472 3473 vm_map_lock(old_map); 3474 3475 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset, 3476 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, FALSE); 3477 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 3478 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 3479 new_map = &vm2->vm_map; /* XXX */ 3480 new_pmap = new_map->pmap; 3481 3482 old_entry = old_map->header.next; 3483 3484 /* 3485 * go entry-by-entry 3486 */ 3487 3488 while (old_entry != &old_map->header) { 3489 3490 /* 3491 * first, some sanity checks on the old entry 3492 */ 3493 if (UVM_ET_ISSUBMAP(old_entry)) 3494 panic("fork: encountered a submap during fork (illegal)"); 3495 3496 if (!UVM_ET_ISCOPYONWRITE(old_entry) && 3497 UVM_ET_ISNEEDSCOPY(old_entry)) 3498 panic("fork: non-copy_on_write map entry marked needs_copy (illegal)"); 3499 3500 3501 switch (old_entry->inheritance) { 3502 case MAP_INHERIT_NONE: 3503 /* 3504 * drop the mapping 3505 */ 3506 break; 3507 3508 case MAP_INHERIT_SHARE: 3509 /* 3510 * share the mapping: this means we want the old and 3511 * new entries to share amaps and backing objects. 3512 */ 3513 3514 /* 3515 * if the old_entry needs a new amap (due to prev fork) 3516 * then we need to allocate it now so that we have 3517 * something we own to share with the new_entry. [in 3518 * other words, we need to clear needs_copy] 3519 */ 3520 3521 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3522 /* get our own amap, clears needs_copy */ 3523 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 3524 0, 0); 3525 /* XXXCDC: WAITOK??? */ 3526 } 3527 3528 new_entry = uvm_mapent_alloc(new_map); 3529 /* old_entry -> new_entry */ 3530 uvm_mapent_copy(old_entry, new_entry); 3531 3532 /* new pmap has nothing wired in it */ 3533 new_entry->wired_count = 0; 3534 3535 /* 3536 * gain reference to object backing the map (can't 3537 * be a submap, already checked this case). 3538 */ 3539 if (new_entry->aref.ar_amap) 3540 /* share reference */ 3541 uvm_map_reference_amap(new_entry, AMAP_SHARED); 3542 3543 if (new_entry->object.uvm_obj && 3544 new_entry->object.uvm_obj->pgops->pgo_reference) 3545 new_entry->object.uvm_obj-> 3546 pgops->pgo_reference( 3547 new_entry->object.uvm_obj); 3548 3549 /* insert entry at end of new_map's entry list */ 3550 uvm_map_entry_link(new_map, new_map->header.prev, 3551 new_entry); 3552 3553 /* 3554 * pmap_copy the mappings: this routine is optional 3555 * but if it is there it will reduce the number of 3556 * page faults in the new proc. 3557 */ 3558 3559 pmap_copy(new_pmap, old_map->pmap, new_entry->start, 3560 (old_entry->end - old_entry->start), 3561 old_entry->start); 3562 3563 break; 3564 3565 case MAP_INHERIT_COPY: 3566 3567 /* 3568 * copy-on-write the mapping (using mmap's 3569 * MAP_PRIVATE semantics) 3570 * 3571 * allocate new_entry, adjust reference counts. 3572 * (note that new references are read-only). 3573 */ 3574 3575 new_entry = uvm_mapent_alloc(new_map); 3576 /* old_entry -> new_entry */ 3577 uvm_mapent_copy(old_entry, new_entry); 3578 3579 if (new_entry->aref.ar_amap) 3580 uvm_map_reference_amap(new_entry, 0); 3581 3582 if (new_entry->object.uvm_obj && 3583 new_entry->object.uvm_obj->pgops->pgo_reference) 3584 new_entry->object.uvm_obj->pgops->pgo_reference 3585 (new_entry->object.uvm_obj); 3586 3587 /* new pmap has nothing wired in it */ 3588 new_entry->wired_count = 0; 3589 3590 new_entry->etype |= 3591 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3592 uvm_map_entry_link(new_map, new_map->header.prev, 3593 new_entry); 3594 3595 /* 3596 * the new entry will need an amap. it will either 3597 * need to be copied from the old entry or created 3598 * from scratch (if the old entry does not have an 3599 * amap). can we defer this process until later 3600 * (by setting "needs_copy") or do we need to copy 3601 * the amap now? 3602 * 3603 * we must copy the amap now if any of the following 3604 * conditions hold: 3605 * 1. the old entry has an amap and that amap is 3606 * being shared. this means that the old (parent) 3607 * process is sharing the amap with another 3608 * process. if we do not clear needs_copy here 3609 * we will end up in a situation where both the 3610 * parent and child process are referring to the 3611 * same amap with "needs_copy" set. if the 3612 * parent write-faults, the fault routine will 3613 * clear "needs_copy" in the parent by allocating 3614 * a new amap. this is wrong because the 3615 * parent is supposed to be sharing the old amap 3616 * and the new amap will break that. 3617 * 3618 * 2. if the old entry has an amap and a non-zero 3619 * wire count then we are going to have to call 3620 * amap_cow_now to avoid page faults in the 3621 * parent process. since amap_cow_now requires 3622 * "needs_copy" to be clear we might as well 3623 * clear it here as well. 3624 * 3625 */ 3626 3627 if (old_entry->aref.ar_amap != NULL) { 3628 3629 if ((amap_flags(old_entry->aref.ar_amap) & 3630 AMAP_SHARED) != 0 || 3631 VM_MAPENT_ISWIRED(old_entry)) { 3632 3633 amap_copy(new_map, new_entry, M_WAITOK, FALSE, 3634 0, 0); 3635 /* XXXCDC: M_WAITOK ... ok? */ 3636 } 3637 } 3638 3639 /* 3640 * if the parent's entry is wired down, then the 3641 * parent process does not want page faults on 3642 * access to that memory. this means that we 3643 * cannot do copy-on-write because we can't write 3644 * protect the old entry. in this case we 3645 * resolve all copy-on-write faults now, using 3646 * amap_cow_now. note that we have already 3647 * allocated any needed amap (above). 3648 */ 3649 3650 if (VM_MAPENT_ISWIRED(old_entry)) { 3651 3652 /* 3653 * resolve all copy-on-write faults now 3654 * (note that there is nothing to do if 3655 * the old mapping does not have an amap). 3656 * XXX: is it worthwhile to bother with pmap_copy 3657 * in this case? 3658 */ 3659 if (old_entry->aref.ar_amap) 3660 amap_cow_now(new_map, new_entry); 3661 3662 } else { 3663 3664 /* 3665 * setup mappings to trigger copy-on-write faults 3666 * we must write-protect the parent if it has 3667 * an amap and it is not already "needs_copy"... 3668 * if it is already "needs_copy" then the parent 3669 * has already been write-protected by a previous 3670 * fork operation. 3671 * 3672 * if we do not write-protect the parent, then 3673 * we must be sure to write-protect the child 3674 * after the pmap_copy() operation. 3675 * 3676 * XXX: pmap_copy should have some way of telling 3677 * us that it didn't do anything so we can avoid 3678 * calling pmap_protect needlessly. 3679 */ 3680 3681 if (old_entry->aref.ar_amap) { 3682 3683 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3684 if (old_entry->max_protection & VM_PROT_WRITE) { 3685 pmap_protect(old_map->pmap, 3686 old_entry->start, 3687 old_entry->end, 3688 old_entry->protection & 3689 ~VM_PROT_WRITE); 3690 pmap_update(old_map->pmap); 3691 3692 } 3693 old_entry->etype |= UVM_ET_NEEDSCOPY; 3694 } 3695 3696 /* 3697 * parent must now be write-protected 3698 */ 3699 protect_child = FALSE; 3700 } else { 3701 3702 /* 3703 * we only need to protect the child if the 3704 * parent has write access. 3705 */ 3706 if (old_entry->max_protection & VM_PROT_WRITE) 3707 protect_child = TRUE; 3708 else 3709 protect_child = FALSE; 3710 3711 } 3712 3713 /* 3714 * copy the mappings 3715 * XXX: need a way to tell if this does anything 3716 */ 3717 3718 pmap_copy(new_pmap, old_map->pmap, 3719 new_entry->start, 3720 (old_entry->end - old_entry->start), 3721 old_entry->start); 3722 3723 /* 3724 * protect the child's mappings if necessary 3725 */ 3726 if (protect_child) { 3727 pmap_protect(new_pmap, new_entry->start, 3728 new_entry->end, 3729 new_entry->protection & 3730 ~VM_PROT_WRITE); 3731 } 3732 3733 } 3734 break; 3735 } /* end of switch statement */ 3736 old_entry = old_entry->next; 3737 } 3738 3739 new_map->size = old_map->size; 3740 vm_map_unlock(old_map); 3741 3742 #ifdef SYSVSHM 3743 if (vm1->vm_shm) 3744 shmfork(vm1, vm2); 3745 #endif 3746 3747 #ifdef PMAP_FORK 3748 pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap); 3749 #endif 3750 3751 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 3752 return(vm2); 3753 } 3754 3755 #if defined(DDB) 3756 3757 /* 3758 * DDB hooks 3759 */ 3760 3761 /* 3762 * uvm_map_printit: actually prints the map 3763 */ 3764 3765 void 3766 uvm_map_printit(struct vm_map *map, boolean_t full, 3767 int (*pr)(const char *, ...)) 3768 { 3769 struct vm_map_entry *entry; 3770 3771 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 3772 (*pr)("\t#ent=%d, sz=%u, ref=%d, version=%u, flags=0x%x\n", 3773 map->nentries, map->size, map->ref_count, map->timestamp, 3774 map->flags); 3775 #ifdef pmap_resident_count 3776 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 3777 pmap_resident_count(map->pmap)); 3778 #else 3779 /* XXXCDC: this should be required ... */ 3780 (*pr)("\tpmap=%p(resident=<<NOT SUPPORTED!!!>>)\n", map->pmap); 3781 #endif 3782 if (!full) 3783 return; 3784 for (entry = map->header.next; entry != &map->header; 3785 entry = entry->next) { 3786 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3787 entry, entry->start, entry->end, entry->object.uvm_obj, 3788 (long long)entry->offset, entry->aref.ar_amap, 3789 entry->aref.ar_pageoff); 3790 (*pr)( 3791 "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, " 3792 "wc=%d, adv=%d\n", 3793 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3794 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3795 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3796 entry->protection, entry->max_protection, 3797 entry->inheritance, entry->wired_count, entry->advice); 3798 } 3799 } 3800 3801 /* 3802 * uvm_object_printit: actually prints the object 3803 */ 3804 3805 void 3806 uvm_object_printit(uobj, full, pr) 3807 struct uvm_object *uobj; 3808 boolean_t full; 3809 int (*pr)(const char *, ...); 3810 { 3811 struct vm_page *pg; 3812 int cnt = 0; 3813 3814 (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ", 3815 uobj, uobj->vmobjlock.lock_data, uobj->pgops, uobj->uo_npages); 3816 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3817 (*pr)("refs=<SYSTEM>\n"); 3818 else 3819 (*pr)("refs=%d\n", uobj->uo_refs); 3820 3821 if (!full) { 3822 return; 3823 } 3824 (*pr)(" PAGES <pg,offset>:\n "); 3825 for (pg = TAILQ_FIRST(&uobj->memq); 3826 pg != NULL; 3827 pg = TAILQ_NEXT(pg, listq), cnt++) { 3828 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3829 if ((cnt % 3) == 2) { 3830 (*pr)("\n "); 3831 } 3832 } 3833 if ((cnt % 3) != 2) { 3834 (*pr)("\n"); 3835 } 3836 } 3837 3838 /* 3839 * uvm_page_printit: actually print the page 3840 */ 3841 3842 static const char page_flagbits[] = 3843 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5CLEANCHK\6RELEASED\7FAKE\10RDONLY" 3844 "\11ZERO\15PAGER1\20FREE\21INACTIVE\22ACTIVE\24ENCRYPT\30PMAP0" 3845 "\31PMAP1\32PMAP2\33PMAP3"; 3846 3847 void 3848 uvm_page_printit(pg, full, pr) 3849 struct vm_page *pg; 3850 boolean_t full; 3851 int (*pr)(const char *, ...); 3852 { 3853 struct vm_page *tpg; 3854 struct uvm_object *uobj; 3855 struct pglist *pgl; 3856 3857 (*pr)("PAGE %p:\n", pg); 3858 (*pr)(" flags=%b, vers=%d, wire_count=%d, pa=0x%llx\n", 3859 pg->pg_flags, page_flagbits, pg->pg_version, pg->wire_count, 3860 (long long)pg->phys_addr); 3861 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n", 3862 pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count); 3863 #if defined(UVM_PAGE_TRKOWN) 3864 if (pg->pg_flags & PG_BUSY) 3865 (*pr)(" owning process = %d, tag=%s\n", 3866 pg->owner, pg->owner_tag); 3867 else 3868 (*pr)(" page not busy, no owner\n"); 3869 #else 3870 (*pr)(" [page ownership tracking disabled]\n"); 3871 #endif 3872 3873 if (!full) 3874 return; 3875 3876 /* cross-verify object/anon */ 3877 if ((pg->pg_flags & PQ_FREE) == 0) { 3878 if (pg->pg_flags & PQ_ANON) { 3879 if (pg->uanon == NULL || pg->uanon->an_page != pg) 3880 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3881 (pg->uanon) ? pg->uanon->an_page : NULL); 3882 else 3883 (*pr)(" anon backpointer is OK\n"); 3884 } else { 3885 uobj = pg->uobject; 3886 if (uobj) { 3887 (*pr)(" checking object list\n"); 3888 TAILQ_FOREACH(tpg, &uobj->memq, listq) { 3889 if (tpg == pg) { 3890 break; 3891 } 3892 } 3893 if (tpg) 3894 (*pr)(" page found on object list\n"); 3895 else 3896 (*pr)(" >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n"); 3897 } 3898 } 3899 } 3900 3901 /* cross-verify page queue */ 3902 if (pg->pg_flags & PQ_FREE) { 3903 int fl = uvm_page_lookup_freelist(pg); 3904 pgl = &uvm.page_free[fl].pgfl_queues[((pg)->pg_flags & PG_ZERO) ? 3905 PGFL_ZEROS : PGFL_UNKNOWN]; 3906 } else if (pg->pg_flags & PQ_INACTIVE) { 3907 pgl = (pg->pg_flags & PQ_SWAPBACKED) ? 3908 &uvm.page_inactive_swp : &uvm.page_inactive_obj; 3909 } else if (pg->pg_flags & PQ_ACTIVE) { 3910 pgl = &uvm.page_active; 3911 } else { 3912 pgl = NULL; 3913 } 3914 3915 if (pgl) { 3916 (*pr)(" checking pageq list\n"); 3917 TAILQ_FOREACH(tpg, pgl, pageq) { 3918 if (tpg == pg) { 3919 break; 3920 } 3921 } 3922 if (tpg) 3923 (*pr)(" page found on pageq list\n"); 3924 else 3925 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3926 } 3927 } 3928 #endif 3929