1 /* $OpenBSD: uvm_map.c,v 1.238 2018/07/22 14:33:44 kettenis Exp $ */ 2 /* $NetBSD: uvm_map.c,v 1.86 2000/11/27 08:40:03 chs Exp $ */ 3 4 /* 5 * Copyright (c) 2011 Ariane van der Steldt <ariane@openbsd.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 * 19 * 20 * Copyright (c) 1997 Charles D. Cranor and Washington University. 21 * Copyright (c) 1991, 1993, The Regents of the University of California. 22 * 23 * All rights reserved. 24 * 25 * This code is derived from software contributed to Berkeley by 26 * The Mach Operating System project at Carnegie-Mellon University. 27 * 28 * Redistribution and use in source and binary forms, with or without 29 * modification, are permitted provided that the following conditions 30 * are met: 31 * 1. Redistributions of source code must retain the above copyright 32 * notice, this list of conditions and the following disclaimer. 33 * 2. Redistributions in binary form must reproduce the above copyright 34 * notice, this list of conditions and the following disclaimer in the 35 * documentation and/or other materials provided with the distribution. 36 * 3. Neither the name of the University nor the names of its contributors 37 * may be used to endorse or promote products derived from this software 38 * without specific prior written permission. 39 * 40 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 41 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 42 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 43 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 44 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 45 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 46 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 47 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 48 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 49 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 50 * SUCH DAMAGE. 51 * 52 * @(#)vm_map.c 8.3 (Berkeley) 1/12/94 53 * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp 54 * 55 * 56 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 57 * All rights reserved. 58 * 59 * Permission to use, copy, modify and distribute this software and 60 * its documentation is hereby granted, provided that both the copyright 61 * notice and this permission notice appear in all copies of the 62 * software, derivative works or modified versions, and any portions 63 * thereof, and that both notices appear in supporting documentation. 64 * 65 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 66 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 67 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 68 * 69 * Carnegie Mellon requests users of this software to return to 70 * 71 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 72 * School of Computer Science 73 * Carnegie Mellon University 74 * Pittsburgh PA 15213-3890 75 * 76 * any improvements or extensions that they make and grant Carnegie the 77 * rights to redistribute these changes. 78 */ 79 80 /* 81 * uvm_map.c: uvm map operations 82 */ 83 84 /* #define DEBUG */ 85 /* #define VMMAP_DEBUG */ 86 87 #include <sys/param.h> 88 #include <sys/systm.h> 89 #include <sys/mman.h> 90 #include <sys/proc.h> 91 #include <sys/malloc.h> 92 #include <sys/pool.h> 93 #include <sys/sysctl.h> 94 #include <sys/syslog.h> 95 96 #ifdef SYSVSHM 97 #include <sys/shm.h> 98 #endif 99 100 #include <uvm/uvm.h> 101 102 #ifdef DDB 103 #include <uvm/uvm_ddb.h> 104 #endif 105 106 #include <uvm/uvm_addr.h> 107 108 109 vsize_t uvmspace_dused(struct vm_map*, vaddr_t, vaddr_t); 110 int uvm_mapent_isjoinable(struct vm_map*, 111 struct vm_map_entry*, struct vm_map_entry*); 112 struct vm_map_entry *uvm_mapent_merge(struct vm_map*, struct vm_map_entry*, 113 struct vm_map_entry*, struct uvm_map_deadq*); 114 struct vm_map_entry *uvm_mapent_tryjoin(struct vm_map*, 115 struct vm_map_entry*, struct uvm_map_deadq*); 116 struct vm_map_entry *uvm_map_mkentry(struct vm_map*, struct vm_map_entry*, 117 struct vm_map_entry*, vaddr_t, vsize_t, int, 118 struct uvm_map_deadq*, struct vm_map_entry*); 119 struct vm_map_entry *uvm_mapent_alloc(struct vm_map*, int); 120 void uvm_mapent_free(struct vm_map_entry*); 121 void uvm_unmap_kill_entry(struct vm_map*, 122 struct vm_map_entry*); 123 void uvm_unmap_detach_intrsafe(struct uvm_map_deadq *); 124 void uvm_mapent_mkfree(struct vm_map*, 125 struct vm_map_entry*, struct vm_map_entry**, 126 struct uvm_map_deadq*, boolean_t); 127 void uvm_map_pageable_pgon(struct vm_map*, 128 struct vm_map_entry*, struct vm_map_entry*, 129 vaddr_t, vaddr_t); 130 int uvm_map_pageable_wire(struct vm_map*, 131 struct vm_map_entry*, struct vm_map_entry*, 132 vaddr_t, vaddr_t, int); 133 void uvm_map_setup_entries(struct vm_map*); 134 void uvm_map_setup_md(struct vm_map*); 135 void uvm_map_teardown(struct vm_map*); 136 void uvm_map_vmspace_update(struct vm_map*, 137 struct uvm_map_deadq*, int); 138 void uvm_map_kmem_grow(struct vm_map*, 139 struct uvm_map_deadq*, vsize_t, int); 140 void uvm_map_freelist_update_clear(struct vm_map*, 141 struct uvm_map_deadq*); 142 void uvm_map_freelist_update_refill(struct vm_map *, int); 143 void uvm_map_freelist_update(struct vm_map*, 144 struct uvm_map_deadq*, vaddr_t, vaddr_t, 145 vaddr_t, vaddr_t, int); 146 struct vm_map_entry *uvm_map_fix_space(struct vm_map*, struct vm_map_entry*, 147 vaddr_t, vaddr_t, int); 148 int uvm_map_sel_limits(vaddr_t*, vaddr_t*, vsize_t, int, 149 struct vm_map_entry*, vaddr_t, vaddr_t, vaddr_t, 150 int); 151 int uvm_map_findspace(struct vm_map*, 152 struct vm_map_entry**, struct vm_map_entry**, 153 vaddr_t*, vsize_t, vaddr_t, vaddr_t, vm_prot_t, 154 vaddr_t); 155 vsize_t uvm_map_addr_augment_get(struct vm_map_entry*); 156 void uvm_map_addr_augment(struct vm_map_entry*); 157 158 /* 159 * Tree management functions. 160 */ 161 162 static __inline void uvm_mapent_copy(struct vm_map_entry*, 163 struct vm_map_entry*); 164 static inline int uvm_mapentry_addrcmp(const struct vm_map_entry*, 165 const struct vm_map_entry*); 166 void uvm_mapent_free_insert(struct vm_map*, 167 struct uvm_addr_state*, struct vm_map_entry*); 168 void uvm_mapent_free_remove(struct vm_map*, 169 struct uvm_addr_state*, struct vm_map_entry*); 170 void uvm_mapent_addr_insert(struct vm_map*, 171 struct vm_map_entry*); 172 void uvm_mapent_addr_remove(struct vm_map*, 173 struct vm_map_entry*); 174 void uvm_map_splitentry(struct vm_map*, 175 struct vm_map_entry*, struct vm_map_entry*, 176 vaddr_t); 177 vsize_t uvm_map_boundary(struct vm_map*, vaddr_t, vaddr_t); 178 int uvm_mapent_bias(struct vm_map*, struct vm_map_entry*); 179 180 /* 181 * uvm_vmspace_fork helper functions. 182 */ 183 struct vm_map_entry *uvm_mapent_clone(struct vm_map*, vaddr_t, vsize_t, 184 vsize_t, vm_prot_t, vm_prot_t, 185 struct vm_map_entry*, struct uvm_map_deadq*, int, 186 int); 187 struct vm_map_entry *uvm_mapent_share(struct vm_map*, vaddr_t, vsize_t, 188 vsize_t, vm_prot_t, vm_prot_t, struct vm_map*, 189 struct vm_map_entry*, struct uvm_map_deadq*); 190 struct vm_map_entry *uvm_mapent_forkshared(struct vmspace*, struct vm_map*, 191 struct vm_map*, struct vm_map_entry*, 192 struct uvm_map_deadq*); 193 struct vm_map_entry *uvm_mapent_forkcopy(struct vmspace*, struct vm_map*, 194 struct vm_map*, struct vm_map_entry*, 195 struct uvm_map_deadq*); 196 struct vm_map_entry *uvm_mapent_forkzero(struct vmspace*, struct vm_map*, 197 struct vm_map*, struct vm_map_entry*, 198 struct uvm_map_deadq*); 199 200 /* 201 * Tree validation. 202 */ 203 #ifdef VMMAP_DEBUG 204 void uvm_tree_assert(struct vm_map*, int, char*, 205 char*, int); 206 #define UVM_ASSERT(map, cond, file, line) \ 207 uvm_tree_assert((map), (cond), #cond, (file), (line)) 208 void uvm_tree_sanity(struct vm_map*, char*, int); 209 void uvm_tree_size_chk(struct vm_map*, char*, int); 210 void vmspace_validate(struct vm_map*); 211 #else 212 #define uvm_tree_sanity(_map, _file, _line) do {} while (0) 213 #define uvm_tree_size_chk(_map, _file, _line) do {} while (0) 214 #define vmspace_validate(_map) do {} while (0) 215 #endif 216 217 /* 218 * All architectures will have pmap_prefer. 219 */ 220 #ifndef PMAP_PREFER 221 #define PMAP_PREFER_ALIGN() (vaddr_t)PAGE_SIZE 222 #define PMAP_PREFER_OFFSET(off) 0 223 #define PMAP_PREFER(addr, off) (addr) 224 #endif 225 226 227 /* 228 * The kernel map will initially be VM_MAP_KSIZE_INIT bytes. 229 * Every time that gets cramped, we grow by at least VM_MAP_KSIZE_DELTA bytes. 230 * 231 * We attempt to grow by UVM_MAP_KSIZE_ALLOCMUL times the allocation size 232 * each time. 233 */ 234 #define VM_MAP_KSIZE_INIT (512 * (vaddr_t)PAGE_SIZE) 235 #define VM_MAP_KSIZE_DELTA (256 * (vaddr_t)PAGE_SIZE) 236 #define VM_MAP_KSIZE_ALLOCMUL 4 237 /* 238 * When selecting a random free-space block, look at most FSPACE_DELTA blocks 239 * ahead. 240 */ 241 #define FSPACE_DELTA 8 242 /* 243 * Put allocations adjecent to previous allocations when the free-space tree 244 * is larger than FSPACE_COMPACT entries. 245 * 246 * Alignment and PMAP_PREFER may still cause the entry to not be fully 247 * adjecent. Note that this strategy reduces memory fragmentation (by leaving 248 * a large space before or after the allocation). 249 */ 250 #define FSPACE_COMPACT 128 251 /* 252 * Make the address selection skip at most this many bytes from the start of 253 * the free space in which the allocation takes place. 254 * 255 * The main idea behind a randomized address space is that an attacker cannot 256 * know where to target his attack. Therefore, the location of objects must be 257 * as random as possible. However, the goal is not to create the most sparse 258 * map that is possible. 259 * FSPACE_MAXOFF pushes the considered range in bytes down to less insane 260 * sizes, thereby reducing the sparseness. The biggest randomization comes 261 * from fragmentation, i.e. FSPACE_COMPACT. 262 */ 263 #define FSPACE_MAXOFF ((vaddr_t)32 * 1024 * 1024) 264 /* 265 * Allow for small gaps in the overflow areas. 266 * Gap size is in bytes and does not have to be a multiple of page-size. 267 */ 268 #define FSPACE_BIASGAP ((vaddr_t)32 * 1024) 269 270 /* auto-allocate address lower bound */ 271 #define VMMAP_MIN_ADDR PAGE_SIZE 272 273 274 #ifdef DEADBEEF0 275 #define UVMMAP_DEADBEEF ((unsigned long)DEADBEEF0) 276 #else 277 #define UVMMAP_DEADBEEF ((unsigned long)0xdeadd0d0) 278 #endif 279 280 #ifdef DEBUG 281 int uvm_map_printlocks = 0; 282 283 #define LPRINTF(_args) \ 284 do { \ 285 if (uvm_map_printlocks) \ 286 printf _args; \ 287 } while (0) 288 #else 289 #define LPRINTF(_args) do {} while (0) 290 #endif 291 292 static struct mutex uvm_kmapent_mtx; 293 static struct timeval uvm_kmapent_last_warn_time; 294 static struct timeval uvm_kmapent_warn_rate = { 10, 0 }; 295 296 const char vmmapbsy[] = "vmmapbsy"; 297 298 /* 299 * pool for vmspace structures. 300 */ 301 struct pool uvm_vmspace_pool; 302 303 /* 304 * pool for dynamically-allocated map entries. 305 */ 306 struct pool uvm_map_entry_pool; 307 struct pool uvm_map_entry_kmem_pool; 308 309 /* 310 * This global represents the end of the kernel virtual address 311 * space. If we want to exceed this, we must grow the kernel 312 * virtual address space dynamically. 313 * 314 * Note, this variable is locked by kernel_map's lock. 315 */ 316 vaddr_t uvm_maxkaddr; 317 318 /* 319 * Locking predicate. 320 */ 321 #define UVM_MAP_REQ_WRITE(_map) \ 322 do { \ 323 if ((_map)->ref_count > 0) { \ 324 if (((_map)->flags & VM_MAP_INTRSAFE) == 0) \ 325 rw_assert_wrlock(&(_map)->lock); \ 326 else \ 327 MUTEX_ASSERT_LOCKED(&(_map)->mtx); \ 328 } \ 329 } while (0) 330 331 /* 332 * Tree describing entries by address. 333 * 334 * Addresses are unique. 335 * Entries with start == end may only exist if they are the first entry 336 * (sorted by address) within a free-memory tree. 337 */ 338 339 static inline int 340 uvm_mapentry_addrcmp(const struct vm_map_entry *e1, 341 const struct vm_map_entry *e2) 342 { 343 return e1->start < e2->start ? -1 : e1->start > e2->start; 344 } 345 346 /* 347 * Copy mapentry. 348 */ 349 static __inline void 350 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 351 { 352 caddr_t csrc, cdst; 353 size_t sz; 354 355 csrc = (caddr_t)src; 356 cdst = (caddr_t)dst; 357 csrc += offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 358 cdst += offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 359 360 sz = offsetof(struct vm_map_entry, uvm_map_entry_stop_copy) - 361 offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 362 memcpy(cdst, csrc, sz); 363 } 364 365 /* 366 * Handle free-list insertion. 367 */ 368 void 369 uvm_mapent_free_insert(struct vm_map *map, struct uvm_addr_state *uaddr, 370 struct vm_map_entry *entry) 371 { 372 const struct uvm_addr_functions *fun; 373 #ifdef VMMAP_DEBUG 374 vaddr_t min, max, bound; 375 #endif 376 377 #ifdef VMMAP_DEBUG 378 /* 379 * Boundary check. 380 * Boundaries are folded if they go on the same free list. 381 */ 382 min = VMMAP_FREE_START(entry); 383 max = VMMAP_FREE_END(entry); 384 385 while (min < max) { 386 bound = uvm_map_boundary(map, min, max); 387 KASSERT(uvm_map_uaddr(map, min) == uaddr); 388 min = bound; 389 } 390 #endif 391 KDASSERT((entry->fspace & (vaddr_t)PAGE_MASK) == 0); 392 KASSERT((entry->etype & UVM_ET_FREEMAPPED) == 0); 393 394 UVM_MAP_REQ_WRITE(map); 395 396 /* Actual insert: forward to uaddr pointer. */ 397 if (uaddr != NULL) { 398 fun = uaddr->uaddr_functions; 399 KDASSERT(fun != NULL); 400 if (fun->uaddr_free_insert != NULL) 401 (*fun->uaddr_free_insert)(map, uaddr, entry); 402 entry->etype |= UVM_ET_FREEMAPPED; 403 } 404 405 /* Update fspace augmentation. */ 406 uvm_map_addr_augment(entry); 407 } 408 409 /* 410 * Handle free-list removal. 411 */ 412 void 413 uvm_mapent_free_remove(struct vm_map *map, struct uvm_addr_state *uaddr, 414 struct vm_map_entry *entry) 415 { 416 const struct uvm_addr_functions *fun; 417 418 KASSERT((entry->etype & UVM_ET_FREEMAPPED) != 0 || uaddr == NULL); 419 KASSERT(uvm_map_uaddr_e(map, entry) == uaddr); 420 UVM_MAP_REQ_WRITE(map); 421 422 if (uaddr != NULL) { 423 fun = uaddr->uaddr_functions; 424 if (fun->uaddr_free_remove != NULL) 425 (*fun->uaddr_free_remove)(map, uaddr, entry); 426 entry->etype &= ~UVM_ET_FREEMAPPED; 427 } 428 } 429 430 /* 431 * Handle address tree insertion. 432 */ 433 void 434 uvm_mapent_addr_insert(struct vm_map *map, struct vm_map_entry *entry) 435 { 436 struct vm_map_entry *res; 437 438 if (!RBT_CHECK(uvm_map_addr, entry, UVMMAP_DEADBEEF)) 439 panic("uvm_mapent_addr_insert: entry still in addr list"); 440 KDASSERT(entry->start <= entry->end); 441 KDASSERT((entry->start & (vaddr_t)PAGE_MASK) == 0 && 442 (entry->end & (vaddr_t)PAGE_MASK) == 0); 443 444 UVM_MAP_REQ_WRITE(map); 445 res = RBT_INSERT(uvm_map_addr, &map->addr, entry); 446 if (res != NULL) { 447 panic("uvm_mapent_addr_insert: map %p entry %p " 448 "(0x%lx-0x%lx G=0x%lx F=0x%lx) insert collision " 449 "with entry %p (0x%lx-0x%lx G=0x%lx F=0x%lx)", 450 map, entry, 451 entry->start, entry->end, entry->guard, entry->fspace, 452 res, res->start, res->end, res->guard, res->fspace); 453 } 454 } 455 456 /* 457 * Handle address tree removal. 458 */ 459 void 460 uvm_mapent_addr_remove(struct vm_map *map, struct vm_map_entry *entry) 461 { 462 struct vm_map_entry *res; 463 464 UVM_MAP_REQ_WRITE(map); 465 res = RBT_REMOVE(uvm_map_addr, &map->addr, entry); 466 if (res != entry) 467 panic("uvm_mapent_addr_remove"); 468 RBT_POISON(uvm_map_addr, entry, UVMMAP_DEADBEEF); 469 } 470 471 /* 472 * uvm_map_reference: add reference to a map 473 * 474 * XXX check map reference counter lock 475 */ 476 #define uvm_map_reference(_map) \ 477 do { \ 478 map->ref_count++; \ 479 } while (0) 480 481 /* 482 * Calculate the dused delta. 483 */ 484 vsize_t 485 uvmspace_dused(struct vm_map *map, vaddr_t min, vaddr_t max) 486 { 487 struct vmspace *vm; 488 vsize_t sz; 489 vaddr_t lmax; 490 vaddr_t stack_begin, stack_end; /* Position of stack. */ 491 492 KASSERT(map->flags & VM_MAP_ISVMSPACE); 493 vm = (struct vmspace *)map; 494 stack_begin = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 495 stack_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 496 497 sz = 0; 498 while (min != max) { 499 lmax = max; 500 if (min < stack_begin && lmax > stack_begin) 501 lmax = stack_begin; 502 else if (min < stack_end && lmax > stack_end) 503 lmax = stack_end; 504 505 if (min >= stack_begin && min < stack_end) { 506 /* nothing */ 507 } else 508 sz += lmax - min; 509 min = lmax; 510 } 511 512 return sz >> PAGE_SHIFT; 513 } 514 515 /* 516 * Find the entry describing the given address. 517 */ 518 struct vm_map_entry* 519 uvm_map_entrybyaddr(struct uvm_map_addr *atree, vaddr_t addr) 520 { 521 struct vm_map_entry *iter; 522 523 iter = RBT_ROOT(uvm_map_addr, atree); 524 while (iter != NULL) { 525 if (iter->start > addr) 526 iter = RBT_LEFT(uvm_map_addr, iter); 527 else if (VMMAP_FREE_END(iter) <= addr) 528 iter = RBT_RIGHT(uvm_map_addr, iter); 529 else 530 return iter; 531 } 532 return NULL; 533 } 534 535 /* 536 * DEAD_ENTRY_PUSH(struct vm_map_deadq *deadq, struct vm_map_entry *entry) 537 * 538 * Push dead entries into a linked list. 539 * Since the linked list abuses the address tree for storage, the entry 540 * may not be linked in a map. 541 * 542 * *head must be initialized to NULL before the first call to this macro. 543 * uvm_unmap_detach(*head, 0) will remove dead entries. 544 */ 545 static __inline void 546 dead_entry_push(struct uvm_map_deadq *deadq, struct vm_map_entry *entry) 547 { 548 TAILQ_INSERT_TAIL(deadq, entry, dfree.deadq); 549 } 550 #define DEAD_ENTRY_PUSH(_headptr, _entry) \ 551 dead_entry_push((_headptr), (_entry)) 552 553 /* 554 * Helper function for uvm_map_findspace_tree. 555 * 556 * Given allocation constraints and pmap constraints, finds the 557 * lowest and highest address in a range that can be used for the 558 * allocation. 559 * 560 * pmap_align and pmap_off are ignored on non-PMAP_PREFER archs. 561 * 562 * 563 * Big chunk of math with a seasoning of dragons. 564 */ 565 int 566 uvm_map_sel_limits(vaddr_t *min, vaddr_t *max, vsize_t sz, int guardpg, 567 struct vm_map_entry *sel, vaddr_t align, 568 vaddr_t pmap_align, vaddr_t pmap_off, int bias) 569 { 570 vaddr_t sel_min, sel_max; 571 #ifdef PMAP_PREFER 572 vaddr_t pmap_min, pmap_max; 573 #endif /* PMAP_PREFER */ 574 #ifdef DIAGNOSTIC 575 int bad; 576 #endif /* DIAGNOSTIC */ 577 578 sel_min = VMMAP_FREE_START(sel); 579 sel_max = VMMAP_FREE_END(sel) - sz - (guardpg ? PAGE_SIZE : 0); 580 581 #ifdef PMAP_PREFER 582 583 /* 584 * There are two special cases, in which we can satisfy the align 585 * requirement and the pmap_prefer requirement. 586 * - when pmap_off == 0, we always select the largest of the two 587 * - when pmap_off % align == 0 and pmap_align > align, we simply 588 * satisfy the pmap_align requirement and automatically 589 * satisfy the align requirement. 590 */ 591 if (align > PAGE_SIZE && 592 !(pmap_align > align && (pmap_off & (align - 1)) == 0)) { 593 /* 594 * Simple case: only use align. 595 */ 596 sel_min = roundup(sel_min, align); 597 sel_max &= ~(align - 1); 598 599 if (sel_min > sel_max) 600 return ENOMEM; 601 602 /* Correct for bias. */ 603 if (sel_max - sel_min > FSPACE_BIASGAP) { 604 if (bias > 0) { 605 sel_min = sel_max - FSPACE_BIASGAP; 606 sel_min = roundup(sel_min, align); 607 } else if (bias < 0) { 608 sel_max = sel_min + FSPACE_BIASGAP; 609 sel_max &= ~(align - 1); 610 } 611 } 612 } else if (pmap_align != 0) { 613 /* 614 * Special case: satisfy both pmap_prefer and 615 * align argument. 616 */ 617 pmap_max = sel_max & ~(pmap_align - 1); 618 pmap_min = sel_min; 619 if (pmap_max < sel_min) 620 return ENOMEM; 621 622 /* Adjust pmap_min for BIASGAP for top-addr bias. */ 623 if (bias > 0 && pmap_max - pmap_min > FSPACE_BIASGAP) 624 pmap_min = pmap_max - FSPACE_BIASGAP; 625 /* Align pmap_min. */ 626 pmap_min &= ~(pmap_align - 1); 627 if (pmap_min < sel_min) 628 pmap_min += pmap_align; 629 if (pmap_min > pmap_max) 630 return ENOMEM; 631 632 /* Adjust pmap_max for BIASGAP for bottom-addr bias. */ 633 if (bias < 0 && pmap_max - pmap_min > FSPACE_BIASGAP) { 634 pmap_max = (pmap_min + FSPACE_BIASGAP) & 635 ~(pmap_align - 1); 636 } 637 if (pmap_min > pmap_max) 638 return ENOMEM; 639 640 /* Apply pmap prefer offset. */ 641 pmap_max |= pmap_off; 642 if (pmap_max > sel_max) 643 pmap_max -= pmap_align; 644 pmap_min |= pmap_off; 645 if (pmap_min < sel_min) 646 pmap_min += pmap_align; 647 648 /* 649 * Fixup: it's possible that pmap_min and pmap_max 650 * cross eachother. In this case, try to find one 651 * address that is allowed. 652 * (This usually happens in biased case.) 653 */ 654 if (pmap_min > pmap_max) { 655 if (pmap_min < sel_max) 656 pmap_max = pmap_min; 657 else if (pmap_max > sel_min) 658 pmap_min = pmap_max; 659 else 660 return ENOMEM; 661 } 662 663 /* Internal validation. */ 664 KDASSERT(pmap_min <= pmap_max); 665 666 sel_min = pmap_min; 667 sel_max = pmap_max; 668 } else if (bias > 0 && sel_max - sel_min > FSPACE_BIASGAP) 669 sel_min = sel_max - FSPACE_BIASGAP; 670 else if (bias < 0 && sel_max - sel_min > FSPACE_BIASGAP) 671 sel_max = sel_min + FSPACE_BIASGAP; 672 673 #else 674 675 if (align > PAGE_SIZE) { 676 sel_min = roundup(sel_min, align); 677 sel_max &= ~(align - 1); 678 if (sel_min > sel_max) 679 return ENOMEM; 680 681 if (bias != 0 && sel_max - sel_min > FSPACE_BIASGAP) { 682 if (bias > 0) { 683 sel_min = roundup(sel_max - FSPACE_BIASGAP, 684 align); 685 } else { 686 sel_max = (sel_min + FSPACE_BIASGAP) & 687 ~(align - 1); 688 } 689 } 690 } else if (bias > 0 && sel_max - sel_min > FSPACE_BIASGAP) 691 sel_min = sel_max - FSPACE_BIASGAP; 692 else if (bias < 0 && sel_max - sel_min > FSPACE_BIASGAP) 693 sel_max = sel_min + FSPACE_BIASGAP; 694 695 #endif 696 697 if (sel_min > sel_max) 698 return ENOMEM; 699 700 #ifdef DIAGNOSTIC 701 bad = 0; 702 /* Lower boundary check. */ 703 if (sel_min < VMMAP_FREE_START(sel)) { 704 printf("sel_min: 0x%lx, but should be at least 0x%lx\n", 705 sel_min, VMMAP_FREE_START(sel)); 706 bad++; 707 } 708 /* Upper boundary check. */ 709 if (sel_max > VMMAP_FREE_END(sel) - sz - (guardpg ? PAGE_SIZE : 0)) { 710 printf("sel_max: 0x%lx, but should be at most 0x%lx\n", 711 sel_max, 712 VMMAP_FREE_END(sel) - sz - (guardpg ? PAGE_SIZE : 0)); 713 bad++; 714 } 715 /* Lower boundary alignment. */ 716 if (align != 0 && (sel_min & (align - 1)) != 0) { 717 printf("sel_min: 0x%lx, not aligned to 0x%lx\n", 718 sel_min, align); 719 bad++; 720 } 721 /* Upper boundary alignment. */ 722 if (align != 0 && (sel_max & (align - 1)) != 0) { 723 printf("sel_max: 0x%lx, not aligned to 0x%lx\n", 724 sel_max, align); 725 bad++; 726 } 727 /* Lower boundary PMAP_PREFER check. */ 728 if (pmap_align != 0 && align == 0 && 729 (sel_min & (pmap_align - 1)) != pmap_off) { 730 printf("sel_min: 0x%lx, aligned to 0x%lx, expected 0x%lx\n", 731 sel_min, sel_min & (pmap_align - 1), pmap_off); 732 bad++; 733 } 734 /* Upper boundary PMAP_PREFER check. */ 735 if (pmap_align != 0 && align == 0 && 736 (sel_max & (pmap_align - 1)) != pmap_off) { 737 printf("sel_max: 0x%lx, aligned to 0x%lx, expected 0x%lx\n", 738 sel_max, sel_max & (pmap_align - 1), pmap_off); 739 bad++; 740 } 741 742 if (bad) { 743 panic("uvm_map_sel_limits(sz = %lu, guardpg = %c, " 744 "align = 0x%lx, pmap_align = 0x%lx, pmap_off = 0x%lx, " 745 "bias = %d, " 746 "FREE_START(sel) = 0x%lx, FREE_END(sel) = 0x%lx)", 747 sz, (guardpg ? 'T' : 'F'), align, pmap_align, pmap_off, 748 bias, VMMAP_FREE_START(sel), VMMAP_FREE_END(sel)); 749 } 750 #endif /* DIAGNOSTIC */ 751 752 *min = sel_min; 753 *max = sel_max; 754 return 0; 755 } 756 757 /* 758 * Test if memory starting at addr with sz bytes is free. 759 * 760 * Fills in *start_ptr and *end_ptr to be the first and last entry describing 761 * the space. 762 * If called with prefilled *start_ptr and *end_ptr, they are to be correct. 763 */ 764 int 765 uvm_map_isavail(struct vm_map *map, struct uvm_addr_state *uaddr, 766 struct vm_map_entry **start_ptr, struct vm_map_entry **end_ptr, 767 vaddr_t addr, vsize_t sz) 768 { 769 struct uvm_addr_state *free; 770 struct uvm_map_addr *atree; 771 struct vm_map_entry *i, *i_end; 772 773 if (addr + sz < addr) 774 return 0; 775 776 /* 777 * Kernel memory above uvm_maxkaddr is considered unavailable. 778 */ 779 if ((map->flags & VM_MAP_ISVMSPACE) == 0) { 780 if (addr + sz > uvm_maxkaddr) 781 return 0; 782 } 783 784 atree = &map->addr; 785 786 /* 787 * Fill in first, last, so they point at the entries containing the 788 * first and last address of the range. 789 * Note that if they are not NULL, we don't perform the lookup. 790 */ 791 KDASSERT(atree != NULL && start_ptr != NULL && end_ptr != NULL); 792 if (*start_ptr == NULL) { 793 *start_ptr = uvm_map_entrybyaddr(atree, addr); 794 if (*start_ptr == NULL) 795 return 0; 796 } else 797 KASSERT(*start_ptr == uvm_map_entrybyaddr(atree, addr)); 798 if (*end_ptr == NULL) { 799 if (VMMAP_FREE_END(*start_ptr) >= addr + sz) 800 *end_ptr = *start_ptr; 801 else { 802 *end_ptr = uvm_map_entrybyaddr(atree, addr + sz - 1); 803 if (*end_ptr == NULL) 804 return 0; 805 } 806 } else 807 KASSERT(*end_ptr == uvm_map_entrybyaddr(atree, addr + sz - 1)); 808 809 /* Validation. */ 810 KDASSERT(*start_ptr != NULL && *end_ptr != NULL); 811 KDASSERT((*start_ptr)->start <= addr && 812 VMMAP_FREE_END(*start_ptr) > addr && 813 (*end_ptr)->start < addr + sz && 814 VMMAP_FREE_END(*end_ptr) >= addr + sz); 815 816 /* 817 * Check the none of the entries intersects with <addr, addr+sz>. 818 * Also, if the entry belong to uaddr_exe or uaddr_brk_stack, it is 819 * considered unavailable unless called by those allocators. 820 */ 821 i = *start_ptr; 822 i_end = RBT_NEXT(uvm_map_addr, *end_ptr); 823 for (; i != i_end; 824 i = RBT_NEXT(uvm_map_addr, i)) { 825 if (i->start != i->end && i->end > addr) 826 return 0; 827 828 /* 829 * uaddr_exe and uaddr_brk_stack may only be used 830 * by these allocators and the NULL uaddr (i.e. no 831 * uaddr). 832 * Reject if this requirement is not met. 833 */ 834 if (uaddr != NULL) { 835 free = uvm_map_uaddr_e(map, i); 836 837 if (uaddr != free && free != NULL && 838 (free == map->uaddr_exe || 839 free == map->uaddr_brk_stack)) 840 return 0; 841 } 842 } 843 844 return -1; 845 } 846 847 /* 848 * Invoke each address selector until an address is found. 849 * Will not invoke uaddr_exe. 850 */ 851 int 852 uvm_map_findspace(struct vm_map *map, struct vm_map_entry**first, 853 struct vm_map_entry**last, vaddr_t *addr, vsize_t sz, 854 vaddr_t pmap_align, vaddr_t pmap_offset, vm_prot_t prot, vaddr_t hint) 855 { 856 struct uvm_addr_state *uaddr; 857 int i; 858 859 /* 860 * Allocation for sz bytes at any address, 861 * using the addr selectors in order. 862 */ 863 for (i = 0; i < nitems(map->uaddr_any); i++) { 864 uaddr = map->uaddr_any[i]; 865 866 if (uvm_addr_invoke(map, uaddr, first, last, 867 addr, sz, pmap_align, pmap_offset, prot, hint) == 0) 868 return 0; 869 } 870 871 /* Fall back to brk() and stack() address selectors. */ 872 uaddr = map->uaddr_brk_stack; 873 if (uvm_addr_invoke(map, uaddr, first, last, 874 addr, sz, pmap_align, pmap_offset, prot, hint) == 0) 875 return 0; 876 877 return ENOMEM; 878 } 879 880 /* Calculate entry augmentation value. */ 881 vsize_t 882 uvm_map_addr_augment_get(struct vm_map_entry *entry) 883 { 884 vsize_t augment; 885 struct vm_map_entry *left, *right; 886 887 augment = entry->fspace; 888 if ((left = RBT_LEFT(uvm_map_addr, entry)) != NULL) 889 augment = MAX(augment, left->fspace_augment); 890 if ((right = RBT_RIGHT(uvm_map_addr, entry)) != NULL) 891 augment = MAX(augment, right->fspace_augment); 892 return augment; 893 } 894 895 /* 896 * Update augmentation data in entry. 897 */ 898 void 899 uvm_map_addr_augment(struct vm_map_entry *entry) 900 { 901 vsize_t augment; 902 903 while (entry != NULL) { 904 /* Calculate value for augmentation. */ 905 augment = uvm_map_addr_augment_get(entry); 906 907 /* 908 * Descend update. 909 * Once we find an entry that already has the correct value, 910 * stop, since it means all its parents will use the correct 911 * value too. 912 */ 913 if (entry->fspace_augment == augment) 914 return; 915 entry->fspace_augment = augment; 916 entry = RBT_PARENT(uvm_map_addr, entry); 917 } 918 } 919 920 /* 921 * uvm_mapanon: establish a valid mapping in map for an anon 922 * 923 * => *addr and sz must be a multiple of PAGE_SIZE. 924 * => *addr is ignored, except if flags contains UVM_FLAG_FIXED. 925 * => map must be unlocked. 926 * 927 * => align: align vaddr, must be a power-of-2. 928 * Align is only a hint and will be ignored if the alignment fails. 929 */ 930 int 931 uvm_mapanon(struct vm_map *map, vaddr_t *addr, vsize_t sz, 932 vsize_t align, unsigned int flags) 933 { 934 struct vm_map_entry *first, *last, *entry, *new; 935 struct uvm_map_deadq dead; 936 vm_prot_t prot; 937 vm_prot_t maxprot; 938 vm_inherit_t inherit; 939 int advice; 940 int error; 941 vaddr_t pmap_align, pmap_offset; 942 vaddr_t hint; 943 944 KASSERT((map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE); 945 KASSERT(map != kernel_map); 946 KASSERT((map->flags & UVM_FLAG_HOLE) == 0); 947 948 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 949 splassert(IPL_NONE); 950 951 /* 952 * We use pmap_align and pmap_offset as alignment and offset variables. 953 * 954 * Because the align parameter takes precedence over pmap prefer, 955 * the pmap_align will need to be set to align, with pmap_offset = 0, 956 * if pmap_prefer will not align. 957 */ 958 pmap_align = MAX(align, PAGE_SIZE); 959 pmap_offset = 0; 960 961 /* Decode parameters. */ 962 prot = UVM_PROTECTION(flags); 963 maxprot = UVM_MAXPROTECTION(flags); 964 advice = UVM_ADVICE(flags); 965 inherit = UVM_INHERIT(flags); 966 error = 0; 967 hint = trunc_page(*addr); 968 TAILQ_INIT(&dead); 969 KASSERT((sz & (vaddr_t)PAGE_MASK) == 0); 970 KASSERT((align & (align - 1)) == 0); 971 972 /* Check protection. */ 973 if ((prot & maxprot) != prot) 974 return EACCES; 975 976 /* 977 * Before grabbing the lock, allocate a map entry for later 978 * use to ensure we don't wait for memory while holding the 979 * vm_map_lock. 980 */ 981 new = uvm_mapent_alloc(map, flags); 982 if (new == NULL) 983 return(ENOMEM); 984 985 if (flags & UVM_FLAG_TRYLOCK) { 986 if (vm_map_lock_try(map) == FALSE) { 987 error = EFAULT; 988 goto out; 989 } 990 } else 991 vm_map_lock(map); 992 993 first = last = NULL; 994 if (flags & UVM_FLAG_FIXED) { 995 /* 996 * Fixed location. 997 * 998 * Note: we ignore align, pmap_prefer. 999 * Fill in first, last and *addr. 1000 */ 1001 KASSERT((*addr & PAGE_MASK) == 0); 1002 1003 /* Check that the space is available. */ 1004 if (flags & UVM_FLAG_UNMAP) { 1005 if ((flags & UVM_FLAG_STACK) && 1006 !uvm_map_is_stack_remappable(map, *addr, sz)) { 1007 error = EINVAL; 1008 goto unlock; 1009 } 1010 uvm_unmap_remove(map, *addr, *addr + sz, &dead, FALSE, TRUE); 1011 } 1012 if (!uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1013 error = ENOMEM; 1014 goto unlock; 1015 } 1016 } else if (*addr != 0 && (*addr & PAGE_MASK) == 0 && 1017 (align == 0 || (*addr & (align - 1)) == 0) && 1018 uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1019 /* 1020 * Address used as hint. 1021 * 1022 * Note: we enforce the alignment restriction, 1023 * but ignore pmap_prefer. 1024 */ 1025 } else if ((prot & PROT_EXEC) != 0 && map->uaddr_exe != NULL) { 1026 /* Run selection algorithm for executables. */ 1027 error = uvm_addr_invoke(map, map->uaddr_exe, &first, &last, 1028 addr, sz, pmap_align, pmap_offset, prot, hint); 1029 1030 if (error != 0) 1031 goto unlock; 1032 } else { 1033 /* Update freelists from vmspace. */ 1034 uvm_map_vmspace_update(map, &dead, flags); 1035 1036 error = uvm_map_findspace(map, &first, &last, addr, sz, 1037 pmap_align, pmap_offset, prot, hint); 1038 1039 if (error != 0) 1040 goto unlock; 1041 } 1042 1043 /* Double-check if selected address doesn't cause overflow. */ 1044 if (*addr + sz < *addr) { 1045 error = ENOMEM; 1046 goto unlock; 1047 } 1048 1049 /* If we only want a query, return now. */ 1050 if (flags & UVM_FLAG_QUERY) { 1051 error = 0; 1052 goto unlock; 1053 } 1054 1055 /* 1056 * Create new entry. 1057 * first and last may be invalidated after this call. 1058 */ 1059 entry = uvm_map_mkentry(map, first, last, *addr, sz, flags, &dead, 1060 new); 1061 if (entry == NULL) { 1062 error = ENOMEM; 1063 goto unlock; 1064 } 1065 new = NULL; 1066 KDASSERT(entry->start == *addr && entry->end == *addr + sz); 1067 entry->object.uvm_obj = NULL; 1068 entry->offset = 0; 1069 entry->protection = prot; 1070 entry->max_protection = maxprot; 1071 entry->inheritance = inherit; 1072 entry->wired_count = 0; 1073 entry->advice = advice; 1074 if (flags & UVM_FLAG_STACK) { 1075 entry->etype |= UVM_ET_STACK; 1076 if (flags & (UVM_FLAG_FIXED | UVM_FLAG_UNMAP)) 1077 map->serial++; 1078 } 1079 if (flags & UVM_FLAG_COPYONW) { 1080 entry->etype |= UVM_ET_COPYONWRITE; 1081 if ((flags & UVM_FLAG_OVERLAY) == 0) 1082 entry->etype |= UVM_ET_NEEDSCOPY; 1083 } 1084 if (flags & UVM_FLAG_OVERLAY) { 1085 KERNEL_LOCK(); 1086 entry->aref.ar_pageoff = 0; 1087 entry->aref.ar_amap = amap_alloc(sz, M_WAITOK, 0); 1088 KERNEL_UNLOCK(); 1089 } 1090 1091 /* Update map and process statistics. */ 1092 map->size += sz; 1093 ((struct vmspace *)map)->vm_dused += uvmspace_dused(map, *addr, *addr + sz); 1094 1095 unlock: 1096 vm_map_unlock(map); 1097 1098 /* 1099 * Remove dead entries. 1100 * 1101 * Dead entries may be the result of merging. 1102 * uvm_map_mkentry may also create dead entries, when it attempts to 1103 * destroy free-space entries. 1104 */ 1105 uvm_unmap_detach(&dead, 0); 1106 out: 1107 if (new) 1108 uvm_mapent_free(new); 1109 return error; 1110 } 1111 1112 /* 1113 * uvm_map: establish a valid mapping in map 1114 * 1115 * => *addr and sz must be a multiple of PAGE_SIZE. 1116 * => map must be unlocked. 1117 * => <uobj,uoffset> value meanings (4 cases): 1118 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 1119 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 1120 * [3] <uobj,uoffset> == normal mapping 1121 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 1122 * 1123 * case [4] is for kernel mappings where we don't know the offset until 1124 * we've found a virtual address. note that kernel object offsets are 1125 * always relative to vm_map_min(kernel_map). 1126 * 1127 * => align: align vaddr, must be a power-of-2. 1128 * Align is only a hint and will be ignored if the alignment fails. 1129 */ 1130 int 1131 uvm_map(struct vm_map *map, vaddr_t *addr, vsize_t sz, 1132 struct uvm_object *uobj, voff_t uoffset, 1133 vsize_t align, unsigned int flags) 1134 { 1135 struct vm_map_entry *first, *last, *entry, *new; 1136 struct uvm_map_deadq dead; 1137 vm_prot_t prot; 1138 vm_prot_t maxprot; 1139 vm_inherit_t inherit; 1140 int advice; 1141 int error; 1142 vaddr_t pmap_align, pmap_offset; 1143 vaddr_t hint; 1144 1145 if ((map->flags & VM_MAP_INTRSAFE) == 0) 1146 splassert(IPL_NONE); 1147 else 1148 splassert(IPL_VM); 1149 1150 /* 1151 * We use pmap_align and pmap_offset as alignment and offset variables. 1152 * 1153 * Because the align parameter takes precedence over pmap prefer, 1154 * the pmap_align will need to be set to align, with pmap_offset = 0, 1155 * if pmap_prefer will not align. 1156 */ 1157 if (uoffset == UVM_UNKNOWN_OFFSET) { 1158 pmap_align = MAX(align, PAGE_SIZE); 1159 pmap_offset = 0; 1160 } else { 1161 pmap_align = MAX(PMAP_PREFER_ALIGN(), PAGE_SIZE); 1162 pmap_offset = PMAP_PREFER_OFFSET(uoffset); 1163 1164 if (align == 0 || 1165 (align <= pmap_align && (pmap_offset & (align - 1)) == 0)) { 1166 /* pmap_offset satisfies align, no change. */ 1167 } else { 1168 /* Align takes precedence over pmap prefer. */ 1169 pmap_align = align; 1170 pmap_offset = 0; 1171 } 1172 } 1173 1174 /* Decode parameters. */ 1175 prot = UVM_PROTECTION(flags); 1176 maxprot = UVM_MAXPROTECTION(flags); 1177 advice = UVM_ADVICE(flags); 1178 inherit = UVM_INHERIT(flags); 1179 error = 0; 1180 hint = trunc_page(*addr); 1181 TAILQ_INIT(&dead); 1182 KASSERT((sz & (vaddr_t)PAGE_MASK) == 0); 1183 KASSERT((align & (align - 1)) == 0); 1184 1185 /* Holes are incompatible with other types of mappings. */ 1186 if (flags & UVM_FLAG_HOLE) { 1187 KASSERT(uobj == NULL && (flags & UVM_FLAG_FIXED) && 1188 (flags & (UVM_FLAG_OVERLAY | UVM_FLAG_COPYONW)) == 0); 1189 } 1190 1191 /* Unset hint for kernel_map non-fixed allocations. */ 1192 if (!(map->flags & VM_MAP_ISVMSPACE) && !(flags & UVM_FLAG_FIXED)) 1193 hint = 0; 1194 1195 /* Check protection. */ 1196 if ((prot & maxprot) != prot) 1197 return EACCES; 1198 1199 if (map == kernel_map && 1200 (prot & (PROT_WRITE | PROT_EXEC)) == (PROT_WRITE | PROT_EXEC)) 1201 panic("uvm_map: kernel map W^X violation requested"); 1202 1203 /* 1204 * Before grabbing the lock, allocate a map entry for later 1205 * use to ensure we don't wait for memory while holding the 1206 * vm_map_lock. 1207 */ 1208 new = uvm_mapent_alloc(map, flags); 1209 if (new == NULL) 1210 return(ENOMEM); 1211 1212 if (flags & UVM_FLAG_TRYLOCK) { 1213 if (vm_map_lock_try(map) == FALSE) { 1214 error = EFAULT; 1215 goto out; 1216 } 1217 } else { 1218 vm_map_lock(map); 1219 } 1220 1221 first = last = NULL; 1222 if (flags & UVM_FLAG_FIXED) { 1223 /* 1224 * Fixed location. 1225 * 1226 * Note: we ignore align, pmap_prefer. 1227 * Fill in first, last and *addr. 1228 */ 1229 KASSERT((*addr & PAGE_MASK) == 0); 1230 1231 /* 1232 * Grow pmap to include allocated address. 1233 * If the growth fails, the allocation will fail too. 1234 */ 1235 if ((map->flags & VM_MAP_ISVMSPACE) == 0 && 1236 uvm_maxkaddr < (*addr + sz)) { 1237 uvm_map_kmem_grow(map, &dead, 1238 *addr + sz - uvm_maxkaddr, flags); 1239 } 1240 1241 /* Check that the space is available. */ 1242 if (flags & UVM_FLAG_UNMAP) 1243 uvm_unmap_remove(map, *addr, *addr + sz, &dead, FALSE, TRUE); 1244 if (!uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1245 error = ENOMEM; 1246 goto unlock; 1247 } 1248 } else if (*addr != 0 && (*addr & PAGE_MASK) == 0 && 1249 (map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE && 1250 (align == 0 || (*addr & (align - 1)) == 0) && 1251 uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1252 /* 1253 * Address used as hint. 1254 * 1255 * Note: we enforce the alignment restriction, 1256 * but ignore pmap_prefer. 1257 */ 1258 } else if ((prot & PROT_EXEC) != 0 && map->uaddr_exe != NULL) { 1259 /* Run selection algorithm for executables. */ 1260 error = uvm_addr_invoke(map, map->uaddr_exe, &first, &last, 1261 addr, sz, pmap_align, pmap_offset, prot, hint); 1262 1263 /* Grow kernel memory and try again. */ 1264 if (error != 0 && (map->flags & VM_MAP_ISVMSPACE) == 0) { 1265 uvm_map_kmem_grow(map, &dead, sz, flags); 1266 1267 error = uvm_addr_invoke(map, map->uaddr_exe, 1268 &first, &last, addr, sz, 1269 pmap_align, pmap_offset, prot, hint); 1270 } 1271 1272 if (error != 0) 1273 goto unlock; 1274 } else { 1275 /* Update freelists from vmspace. */ 1276 if (map->flags & VM_MAP_ISVMSPACE) 1277 uvm_map_vmspace_update(map, &dead, flags); 1278 1279 error = uvm_map_findspace(map, &first, &last, addr, sz, 1280 pmap_align, pmap_offset, prot, hint); 1281 1282 /* Grow kernel memory and try again. */ 1283 if (error != 0 && (map->flags & VM_MAP_ISVMSPACE) == 0) { 1284 uvm_map_kmem_grow(map, &dead, sz, flags); 1285 1286 error = uvm_map_findspace(map, &first, &last, addr, sz, 1287 pmap_align, pmap_offset, prot, hint); 1288 } 1289 1290 if (error != 0) 1291 goto unlock; 1292 } 1293 1294 /* Double-check if selected address doesn't cause overflow. */ 1295 if (*addr + sz < *addr) { 1296 error = ENOMEM; 1297 goto unlock; 1298 } 1299 1300 KASSERT((map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE || 1301 uvm_maxkaddr >= *addr + sz); 1302 1303 /* If we only want a query, return now. */ 1304 if (flags & UVM_FLAG_QUERY) { 1305 error = 0; 1306 goto unlock; 1307 } 1308 1309 if (uobj == NULL) 1310 uoffset = 0; 1311 else if (uoffset == UVM_UNKNOWN_OFFSET) { 1312 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 1313 uoffset = *addr - vm_map_min(kernel_map); 1314 } 1315 1316 /* 1317 * Create new entry. 1318 * first and last may be invalidated after this call. 1319 */ 1320 entry = uvm_map_mkentry(map, first, last, *addr, sz, flags, &dead, 1321 new); 1322 if (entry == NULL) { 1323 error = ENOMEM; 1324 goto unlock; 1325 } 1326 new = NULL; 1327 KDASSERT(entry->start == *addr && entry->end == *addr + sz); 1328 entry->object.uvm_obj = uobj; 1329 entry->offset = uoffset; 1330 entry->protection = prot; 1331 entry->max_protection = maxprot; 1332 entry->inheritance = inherit; 1333 entry->wired_count = 0; 1334 entry->advice = advice; 1335 if (flags & UVM_FLAG_STACK) { 1336 entry->etype |= UVM_ET_STACK; 1337 if (flags & UVM_FLAG_UNMAP) 1338 map->serial++; 1339 } 1340 if (uobj) 1341 entry->etype |= UVM_ET_OBJ; 1342 else if (flags & UVM_FLAG_HOLE) 1343 entry->etype |= UVM_ET_HOLE; 1344 if (flags & UVM_FLAG_NOFAULT) 1345 entry->etype |= UVM_ET_NOFAULT; 1346 if (flags & UVM_FLAG_COPYONW) { 1347 entry->etype |= UVM_ET_COPYONWRITE; 1348 if ((flags & UVM_FLAG_OVERLAY) == 0) 1349 entry->etype |= UVM_ET_NEEDSCOPY; 1350 } 1351 if (flags & UVM_FLAG_OVERLAY) { 1352 entry->aref.ar_pageoff = 0; 1353 entry->aref.ar_amap = amap_alloc(sz, M_WAITOK, 0); 1354 } 1355 1356 /* Update map and process statistics. */ 1357 if (!(flags & UVM_FLAG_HOLE)) { 1358 map->size += sz; 1359 if ((map->flags & VM_MAP_ISVMSPACE) && uobj == NULL) { 1360 ((struct vmspace *)map)->vm_dused += 1361 uvmspace_dused(map, *addr, *addr + sz); 1362 } 1363 } 1364 1365 /* 1366 * Try to merge entry. 1367 * 1368 * Userland allocations are kept separated most of the time. 1369 * Forego the effort of merging what most of the time can't be merged 1370 * and only try the merge if it concerns a kernel entry. 1371 */ 1372 if ((flags & UVM_FLAG_NOMERGE) == 0 && 1373 (map->flags & VM_MAP_ISVMSPACE) == 0) 1374 uvm_mapent_tryjoin(map, entry, &dead); 1375 1376 unlock: 1377 vm_map_unlock(map); 1378 1379 /* 1380 * Remove dead entries. 1381 * 1382 * Dead entries may be the result of merging. 1383 * uvm_map_mkentry may also create dead entries, when it attempts to 1384 * destroy free-space entries. 1385 */ 1386 if (map->flags & VM_MAP_INTRSAFE) 1387 uvm_unmap_detach_intrsafe(&dead); 1388 else 1389 uvm_unmap_detach(&dead, 0); 1390 out: 1391 if (new) 1392 uvm_mapent_free(new); 1393 return error; 1394 } 1395 1396 /* 1397 * True iff e1 and e2 can be joined together. 1398 */ 1399 int 1400 uvm_mapent_isjoinable(struct vm_map *map, struct vm_map_entry *e1, 1401 struct vm_map_entry *e2) 1402 { 1403 KDASSERT(e1 != NULL && e2 != NULL); 1404 1405 /* Must be the same entry type and not have free memory between. */ 1406 if (e1->etype != e2->etype || e1->end != e2->start) 1407 return 0; 1408 1409 /* Submaps are never joined. */ 1410 if (UVM_ET_ISSUBMAP(e1)) 1411 return 0; 1412 1413 /* Never merge wired memory. */ 1414 if (VM_MAPENT_ISWIRED(e1) || VM_MAPENT_ISWIRED(e2)) 1415 return 0; 1416 1417 /* Protection, inheritance and advice must be equal. */ 1418 if (e1->protection != e2->protection || 1419 e1->max_protection != e2->max_protection || 1420 e1->inheritance != e2->inheritance || 1421 e1->advice != e2->advice) 1422 return 0; 1423 1424 /* If uvm_object: object itself and offsets within object must match. */ 1425 if (UVM_ET_ISOBJ(e1)) { 1426 if (e1->object.uvm_obj != e2->object.uvm_obj) 1427 return 0; 1428 if (e1->offset + (e1->end - e1->start) != e2->offset) 1429 return 0; 1430 } 1431 1432 /* 1433 * Cannot join shared amaps. 1434 * Note: no need to lock amap to look at refs, since we don't care 1435 * about its exact value. 1436 * If it is 1 (i.e. we have the only reference) it will stay there. 1437 */ 1438 if (e1->aref.ar_amap && amap_refs(e1->aref.ar_amap) != 1) 1439 return 0; 1440 if (e2->aref.ar_amap && amap_refs(e2->aref.ar_amap) != 1) 1441 return 0; 1442 1443 /* Apprently, e1 and e2 match. */ 1444 return 1; 1445 } 1446 1447 /* 1448 * Join support function. 1449 * 1450 * Returns the merged entry on succes. 1451 * Returns NULL if the merge failed. 1452 */ 1453 struct vm_map_entry* 1454 uvm_mapent_merge(struct vm_map *map, struct vm_map_entry *e1, 1455 struct vm_map_entry *e2, struct uvm_map_deadq *dead) 1456 { 1457 struct uvm_addr_state *free; 1458 1459 /* 1460 * Merging is not supported for map entries that 1461 * contain an amap in e1. This should never happen 1462 * anyway, because only kernel entries are merged. 1463 * These do not contain amaps. 1464 * e2 contains no real information in its amap, 1465 * so it can be erased immediately. 1466 */ 1467 KASSERT(e1->aref.ar_amap == NULL); 1468 1469 /* 1470 * Don't drop obj reference: 1471 * uvm_unmap_detach will do this for us. 1472 */ 1473 free = uvm_map_uaddr_e(map, e1); 1474 uvm_mapent_free_remove(map, free, e1); 1475 1476 free = uvm_map_uaddr_e(map, e2); 1477 uvm_mapent_free_remove(map, free, e2); 1478 uvm_mapent_addr_remove(map, e2); 1479 e1->end = e2->end; 1480 e1->guard = e2->guard; 1481 e1->fspace = e2->fspace; 1482 uvm_mapent_free_insert(map, free, e1); 1483 1484 DEAD_ENTRY_PUSH(dead, e2); 1485 return e1; 1486 } 1487 1488 /* 1489 * Attempt forward and backward joining of entry. 1490 * 1491 * Returns entry after joins. 1492 * We are guaranteed that the amap of entry is either non-existant or 1493 * has never been used. 1494 */ 1495 struct vm_map_entry* 1496 uvm_mapent_tryjoin(struct vm_map *map, struct vm_map_entry *entry, 1497 struct uvm_map_deadq *dead) 1498 { 1499 struct vm_map_entry *other; 1500 struct vm_map_entry *merged; 1501 1502 /* Merge with previous entry. */ 1503 other = RBT_PREV(uvm_map_addr, entry); 1504 if (other && uvm_mapent_isjoinable(map, other, entry)) { 1505 merged = uvm_mapent_merge(map, other, entry, dead); 1506 if (merged) 1507 entry = merged; 1508 } 1509 1510 /* 1511 * Merge with next entry. 1512 * 1513 * Because amap can only extend forward and the next entry 1514 * probably contains sensible info, only perform forward merging 1515 * in the absence of an amap. 1516 */ 1517 other = RBT_NEXT(uvm_map_addr, entry); 1518 if (other && entry->aref.ar_amap == NULL && 1519 other->aref.ar_amap == NULL && 1520 uvm_mapent_isjoinable(map, entry, other)) { 1521 merged = uvm_mapent_merge(map, entry, other, dead); 1522 if (merged) 1523 entry = merged; 1524 } 1525 1526 return entry; 1527 } 1528 1529 /* 1530 * Kill entries that are no longer in a map. 1531 */ 1532 void 1533 uvm_unmap_detach(struct uvm_map_deadq *deadq, int flags) 1534 { 1535 struct vm_map_entry *entry; 1536 int waitok = flags & UVM_PLA_WAITOK; 1537 1538 if (TAILQ_EMPTY(deadq)) 1539 return; 1540 1541 KERNEL_LOCK(); 1542 while ((entry = TAILQ_FIRST(deadq)) != NULL) { 1543 if (waitok) 1544 uvm_pause(); 1545 /* Drop reference to amap, if we've got one. */ 1546 if (entry->aref.ar_amap) 1547 amap_unref(entry->aref.ar_amap, 1548 entry->aref.ar_pageoff, 1549 atop(entry->end - entry->start), 1550 flags & AMAP_REFALL); 1551 1552 /* Drop reference to our backing object, if we've got one. */ 1553 if (UVM_ET_ISSUBMAP(entry)) { 1554 /* ... unlikely to happen, but play it safe */ 1555 uvm_map_deallocate(entry->object.sub_map); 1556 } else if (UVM_ET_ISOBJ(entry) && 1557 entry->object.uvm_obj->pgops->pgo_detach) { 1558 entry->object.uvm_obj->pgops->pgo_detach( 1559 entry->object.uvm_obj); 1560 } 1561 1562 /* Step to next. */ 1563 TAILQ_REMOVE(deadq, entry, dfree.deadq); 1564 uvm_mapent_free(entry); 1565 } 1566 KERNEL_UNLOCK(); 1567 } 1568 1569 void 1570 uvm_unmap_detach_intrsafe(struct uvm_map_deadq *deadq) 1571 { 1572 struct vm_map_entry *entry; 1573 1574 while ((entry = TAILQ_FIRST(deadq)) != NULL) { 1575 KASSERT(entry->aref.ar_amap == NULL); 1576 KASSERT(!UVM_ET_ISSUBMAP(entry)); 1577 KASSERT(!UVM_ET_ISOBJ(entry)); 1578 TAILQ_REMOVE(deadq, entry, dfree.deadq); 1579 uvm_mapent_free(entry); 1580 } 1581 } 1582 1583 /* 1584 * Create and insert new entry. 1585 * 1586 * Returned entry contains new addresses and is inserted properly in the tree. 1587 * first and last are (probably) no longer valid. 1588 */ 1589 struct vm_map_entry* 1590 uvm_map_mkentry(struct vm_map *map, struct vm_map_entry *first, 1591 struct vm_map_entry *last, vaddr_t addr, vsize_t sz, int flags, 1592 struct uvm_map_deadq *dead, struct vm_map_entry *new) 1593 { 1594 struct vm_map_entry *entry, *prev; 1595 struct uvm_addr_state *free; 1596 vaddr_t min, max; /* free space boundaries for new entry */ 1597 1598 KDASSERT(map != NULL); 1599 KDASSERT(first != NULL); 1600 KDASSERT(last != NULL); 1601 KDASSERT(dead != NULL); 1602 KDASSERT(sz > 0); 1603 KDASSERT(addr + sz > addr); 1604 KDASSERT(first->end <= addr && VMMAP_FREE_END(first) > addr); 1605 KDASSERT(last->start < addr + sz && VMMAP_FREE_END(last) >= addr + sz); 1606 KDASSERT(uvm_map_isavail(map, NULL, &first, &last, addr, sz)); 1607 uvm_tree_sanity(map, __FILE__, __LINE__); 1608 1609 min = addr + sz; 1610 max = VMMAP_FREE_END(last); 1611 1612 /* Initialize new entry. */ 1613 if (new == NULL) 1614 entry = uvm_mapent_alloc(map, flags); 1615 else 1616 entry = new; 1617 if (entry == NULL) 1618 return NULL; 1619 entry->offset = 0; 1620 entry->etype = 0; 1621 entry->wired_count = 0; 1622 entry->aref.ar_pageoff = 0; 1623 entry->aref.ar_amap = NULL; 1624 1625 entry->start = addr; 1626 entry->end = min; 1627 entry->guard = 0; 1628 entry->fspace = 0; 1629 1630 /* Reset free space in first. */ 1631 free = uvm_map_uaddr_e(map, first); 1632 uvm_mapent_free_remove(map, free, first); 1633 first->guard = 0; 1634 first->fspace = 0; 1635 1636 /* 1637 * Remove all entries that are fully replaced. 1638 * We are iterating using last in reverse order. 1639 */ 1640 for (; first != last; last = prev) { 1641 prev = RBT_PREV(uvm_map_addr, last); 1642 1643 KDASSERT(last->start == last->end); 1644 free = uvm_map_uaddr_e(map, last); 1645 uvm_mapent_free_remove(map, free, last); 1646 uvm_mapent_addr_remove(map, last); 1647 DEAD_ENTRY_PUSH(dead, last); 1648 } 1649 /* Remove first if it is entirely inside <addr, addr+sz>. */ 1650 if (first->start == addr) { 1651 uvm_mapent_addr_remove(map, first); 1652 DEAD_ENTRY_PUSH(dead, first); 1653 } else { 1654 uvm_map_fix_space(map, first, VMMAP_FREE_START(first), 1655 addr, flags); 1656 } 1657 1658 /* Finally, link in entry. */ 1659 uvm_mapent_addr_insert(map, entry); 1660 uvm_map_fix_space(map, entry, min, max, flags); 1661 1662 uvm_tree_sanity(map, __FILE__, __LINE__); 1663 return entry; 1664 } 1665 1666 1667 /* 1668 * uvm_mapent_alloc: allocate a map entry 1669 */ 1670 struct vm_map_entry * 1671 uvm_mapent_alloc(struct vm_map *map, int flags) 1672 { 1673 struct vm_map_entry *me, *ne; 1674 int pool_flags; 1675 int i; 1676 1677 pool_flags = PR_WAITOK; 1678 if (flags & UVM_FLAG_TRYLOCK) 1679 pool_flags = PR_NOWAIT; 1680 1681 if (map->flags & VM_MAP_INTRSAFE || cold) { 1682 mtx_enter(&uvm_kmapent_mtx); 1683 if (SLIST_EMPTY(&uvm.kentry_free)) { 1684 ne = km_alloc(PAGE_SIZE, &kv_page, &kp_dirty, 1685 &kd_nowait); 1686 if (ne == NULL) 1687 panic("uvm_mapent_alloc: cannot allocate map " 1688 "entry"); 1689 for (i = 0; i < PAGE_SIZE / sizeof(*ne); i++) { 1690 SLIST_INSERT_HEAD(&uvm.kentry_free, 1691 &ne[i], daddrs.addr_kentry); 1692 } 1693 if (ratecheck(&uvm_kmapent_last_warn_time, 1694 &uvm_kmapent_warn_rate)) 1695 printf("uvm_mapent_alloc: out of static " 1696 "map entries\n"); 1697 } 1698 me = SLIST_FIRST(&uvm.kentry_free); 1699 SLIST_REMOVE_HEAD(&uvm.kentry_free, daddrs.addr_kentry); 1700 uvmexp.kmapent++; 1701 mtx_leave(&uvm_kmapent_mtx); 1702 me->flags = UVM_MAP_STATIC; 1703 } else if (map == kernel_map) { 1704 splassert(IPL_NONE); 1705 me = pool_get(&uvm_map_entry_kmem_pool, pool_flags); 1706 if (me == NULL) 1707 goto out; 1708 me->flags = UVM_MAP_KMEM; 1709 } else { 1710 splassert(IPL_NONE); 1711 me = pool_get(&uvm_map_entry_pool, pool_flags); 1712 if (me == NULL) 1713 goto out; 1714 me->flags = 0; 1715 } 1716 1717 if (me != NULL) { 1718 RBT_POISON(uvm_map_addr, me, UVMMAP_DEADBEEF); 1719 } 1720 1721 out: 1722 return(me); 1723 } 1724 1725 /* 1726 * uvm_mapent_free: free map entry 1727 * 1728 * => XXX: static pool for kernel map? 1729 */ 1730 void 1731 uvm_mapent_free(struct vm_map_entry *me) 1732 { 1733 if (me->flags & UVM_MAP_STATIC) { 1734 mtx_enter(&uvm_kmapent_mtx); 1735 SLIST_INSERT_HEAD(&uvm.kentry_free, me, daddrs.addr_kentry); 1736 uvmexp.kmapent--; 1737 mtx_leave(&uvm_kmapent_mtx); 1738 } else if (me->flags & UVM_MAP_KMEM) { 1739 splassert(IPL_NONE); 1740 pool_put(&uvm_map_entry_kmem_pool, me); 1741 } else { 1742 splassert(IPL_NONE); 1743 pool_put(&uvm_map_entry_pool, me); 1744 } 1745 } 1746 1747 /* 1748 * uvm_map_lookup_entry: find map entry at or before an address. 1749 * 1750 * => map must at least be read-locked by caller 1751 * => entry is returned in "entry" 1752 * => return value is true if address is in the returned entry 1753 * ET_HOLE entries are considered to not contain a mapping, ergo FALSE is 1754 * returned for those mappings. 1755 */ 1756 boolean_t 1757 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1758 struct vm_map_entry **entry) 1759 { 1760 *entry = uvm_map_entrybyaddr(&map->addr, address); 1761 return *entry != NULL && !UVM_ET_ISHOLE(*entry) && 1762 (*entry)->start <= address && (*entry)->end > address; 1763 } 1764 1765 /* 1766 * Inside a vm_map find the sp address and verify MAP_STACK, and also 1767 * remember low and high regions of that of region which is marked 1768 * with MAP_STACK. Return TRUE. 1769 * If sp isn't in a MAP_STACK region return FALSE. 1770 */ 1771 boolean_t 1772 uvm_map_check_stack_range(struct proc *p, vaddr_t sp) 1773 { 1774 vm_map_t map = &p->p_vmspace->vm_map; 1775 vm_map_entry_t entry; 1776 1777 if (sp < map->min_offset || sp >= map->max_offset) 1778 return(FALSE); 1779 1780 /* lock map */ 1781 vm_map_lock_read(map); 1782 1783 /* lookup */ 1784 if (!uvm_map_lookup_entry(map, trunc_page(sp), &entry)) { 1785 vm_map_unlock_read(map); 1786 return(FALSE); 1787 } 1788 1789 if ((entry->etype & UVM_ET_STACK) == 0) { 1790 vm_map_unlock_read(map); 1791 return (FALSE); 1792 } 1793 p->p_spstart = entry->start; 1794 p->p_spend = entry->end; 1795 p->p_spserial = map->serial; 1796 vm_map_unlock_read(map); 1797 return(TRUE); 1798 } 1799 1800 /* 1801 * Check whether the given address range can be converted to a MAP_STACK 1802 * mapping. 1803 * 1804 * Must be called with map locked. 1805 */ 1806 boolean_t 1807 uvm_map_is_stack_remappable(struct vm_map *map, vaddr_t addr, vaddr_t sz) 1808 { 1809 vaddr_t end = addr + sz; 1810 struct vm_map_entry *first, *iter, *prev = NULL; 1811 1812 if (!uvm_map_lookup_entry(map, addr, &first)) { 1813 printf("map stack 0x%lx-0x%lx of map %p failed: no mapping\n", 1814 addr, end, map); 1815 return FALSE; 1816 } 1817 1818 /* 1819 * Check that the address range exists and is contiguous. 1820 */ 1821 for (iter = first; iter != NULL && iter->start < end; 1822 prev = iter, iter = RBT_NEXT(uvm_map_addr, iter)) { 1823 /* 1824 * Make sure that we do not have holes in the range. 1825 */ 1826 #if 0 1827 if (prev != NULL) { 1828 printf("prev->start 0x%lx, prev->end 0x%lx, " 1829 "iter->start 0x%lx, iter->end 0x%lx\n", 1830 prev->start, prev->end, iter->start, iter->end); 1831 } 1832 #endif 1833 1834 if (prev != NULL && prev->end != iter->start) { 1835 printf("map stack 0x%lx-0x%lx of map %p failed: " 1836 "hole in range\n", addr, end, map); 1837 return FALSE; 1838 } 1839 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) { 1840 printf("map stack 0x%lx-0x%lx of map %p failed: " 1841 "hole in range\n", addr, end, map); 1842 return FALSE; 1843 } 1844 } 1845 1846 return TRUE; 1847 } 1848 1849 /* 1850 * Remap the middle-pages of an existing mapping as a stack range. 1851 * If there exists a previous contiguous mapping with the given range 1852 * [addr, addr + sz), with protection PROT_READ|PROT_WRITE, then the 1853 * mapping is dropped, and a new anon mapping is created and marked as 1854 * a stack. 1855 * 1856 * Must be called with map unlocked. 1857 */ 1858 int 1859 uvm_map_remap_as_stack(struct proc *p, vaddr_t addr, vaddr_t sz) 1860 { 1861 vm_map_t map = &p->p_vmspace->vm_map; 1862 vaddr_t start, end; 1863 int error; 1864 int flags = UVM_MAPFLAG(PROT_READ | PROT_WRITE, 1865 PROT_READ | PROT_WRITE | PROT_EXEC, 1866 MAP_INHERIT_COPY, MADV_NORMAL, 1867 UVM_FLAG_STACK | UVM_FLAG_FIXED | UVM_FLAG_UNMAP | 1868 UVM_FLAG_COPYONW); 1869 1870 start = round_page(addr); 1871 end = trunc_page(addr + sz); 1872 #ifdef MACHINE_STACK_GROWS_UP 1873 if (end == addr + sz) 1874 end -= PAGE_SIZE; 1875 #else 1876 if (start == addr) 1877 start += PAGE_SIZE; 1878 #endif 1879 1880 if (start < map->min_offset || end >= map->max_offset || end < start) 1881 return EINVAL; 1882 1883 error = uvm_mapanon(map, &start, end - start, 0, flags); 1884 if (error != 0) 1885 printf("map stack for pid %d failed\n", p->p_p->ps_pid); 1886 1887 return error; 1888 } 1889 1890 /* 1891 * uvm_map_pie: return a random load address for a PIE executable 1892 * properly aligned. 1893 */ 1894 #ifndef VM_PIE_MAX_ADDR 1895 #define VM_PIE_MAX_ADDR (VM_MAXUSER_ADDRESS / 4) 1896 #endif 1897 1898 #ifndef VM_PIE_MIN_ADDR 1899 #define VM_PIE_MIN_ADDR VM_MIN_ADDRESS 1900 #endif 1901 1902 #ifndef VM_PIE_MIN_ALIGN 1903 #define VM_PIE_MIN_ALIGN PAGE_SIZE 1904 #endif 1905 1906 vaddr_t 1907 uvm_map_pie(vaddr_t align) 1908 { 1909 vaddr_t addr, space, min; 1910 1911 align = MAX(align, VM_PIE_MIN_ALIGN); 1912 1913 /* round up to next alignment */ 1914 min = (VM_PIE_MIN_ADDR + align - 1) & ~(align - 1); 1915 1916 if (align >= VM_PIE_MAX_ADDR || min >= VM_PIE_MAX_ADDR) 1917 return (align); 1918 1919 space = (VM_PIE_MAX_ADDR - min) / align; 1920 space = MIN(space, (u_int32_t)-1); 1921 1922 addr = (vaddr_t)arc4random_uniform((u_int32_t)space) * align; 1923 addr += min; 1924 1925 return (addr); 1926 } 1927 1928 void 1929 uvm_unmap(struct vm_map *map, vaddr_t start, vaddr_t end) 1930 { 1931 struct uvm_map_deadq dead; 1932 1933 KASSERT((start & (vaddr_t)PAGE_MASK) == 0 && 1934 (end & (vaddr_t)PAGE_MASK) == 0); 1935 TAILQ_INIT(&dead); 1936 vm_map_lock(map); 1937 uvm_unmap_remove(map, start, end, &dead, FALSE, TRUE); 1938 vm_map_unlock(map); 1939 1940 if (map->flags & VM_MAP_INTRSAFE) 1941 uvm_unmap_detach_intrsafe(&dead); 1942 else 1943 uvm_unmap_detach(&dead, 0); 1944 } 1945 1946 /* 1947 * Mark entry as free. 1948 * 1949 * entry will be put on the dead list. 1950 * The free space will be merged into the previous or a new entry, 1951 * unless markfree is false. 1952 */ 1953 void 1954 uvm_mapent_mkfree(struct vm_map *map, struct vm_map_entry *entry, 1955 struct vm_map_entry **prev_ptr, struct uvm_map_deadq *dead, 1956 boolean_t markfree) 1957 { 1958 struct uvm_addr_state *free; 1959 struct vm_map_entry *prev; 1960 vaddr_t addr; /* Start of freed range. */ 1961 vaddr_t end; /* End of freed range. */ 1962 1963 prev = *prev_ptr; 1964 if (prev == entry) 1965 *prev_ptr = prev = NULL; 1966 1967 if (prev == NULL || 1968 VMMAP_FREE_END(prev) != entry->start) 1969 prev = RBT_PREV(uvm_map_addr, entry); 1970 1971 /* Entry is describing only free memory and has nothing to drain into. */ 1972 if (prev == NULL && entry->start == entry->end && markfree) { 1973 *prev_ptr = entry; 1974 return; 1975 } 1976 1977 addr = entry->start; 1978 end = VMMAP_FREE_END(entry); 1979 free = uvm_map_uaddr_e(map, entry); 1980 uvm_mapent_free_remove(map, free, entry); 1981 uvm_mapent_addr_remove(map, entry); 1982 DEAD_ENTRY_PUSH(dead, entry); 1983 1984 if (markfree) { 1985 if (prev) { 1986 free = uvm_map_uaddr_e(map, prev); 1987 uvm_mapent_free_remove(map, free, prev); 1988 } 1989 *prev_ptr = uvm_map_fix_space(map, prev, addr, end, 0); 1990 } 1991 } 1992 1993 /* 1994 * Unwire and release referenced amap and object from map entry. 1995 */ 1996 void 1997 uvm_unmap_kill_entry(struct vm_map *map, struct vm_map_entry *entry) 1998 { 1999 /* Unwire removed map entry. */ 2000 if (VM_MAPENT_ISWIRED(entry)) { 2001 KERNEL_LOCK(); 2002 entry->wired_count = 0; 2003 uvm_fault_unwire_locked(map, entry->start, entry->end); 2004 KERNEL_UNLOCK(); 2005 } 2006 2007 /* Entry-type specific code. */ 2008 if (UVM_ET_ISHOLE(entry)) { 2009 /* Nothing to be done for holes. */ 2010 } else if (map->flags & VM_MAP_INTRSAFE) { 2011 KASSERT(vm_map_pmap(map) == pmap_kernel()); 2012 uvm_km_pgremove_intrsafe(entry->start, entry->end); 2013 pmap_kremove(entry->start, entry->end - entry->start); 2014 } else if (UVM_ET_ISOBJ(entry) && 2015 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 2016 KASSERT(vm_map_pmap(map) == pmap_kernel()); 2017 /* 2018 * Note: kernel object mappings are currently used in 2019 * two ways: 2020 * [1] "normal" mappings of pages in the kernel object 2021 * [2] uvm_km_valloc'd allocations in which we 2022 * pmap_enter in some non-kernel-object page 2023 * (e.g. vmapbuf). 2024 * 2025 * for case [1], we need to remove the mapping from 2026 * the pmap and then remove the page from the kernel 2027 * object (because, once pages in a kernel object are 2028 * unmapped they are no longer needed, unlike, say, 2029 * a vnode where you might want the data to persist 2030 * until flushed out of a queue). 2031 * 2032 * for case [2], we need to remove the mapping from 2033 * the pmap. there shouldn't be any pages at the 2034 * specified offset in the kernel object [but it 2035 * doesn't hurt to call uvm_km_pgremove just to be 2036 * safe?] 2037 * 2038 * uvm_km_pgremove currently does the following: 2039 * for pages in the kernel object range: 2040 * - drops the swap slot 2041 * - uvm_pagefree the page 2042 * 2043 * note there is version of uvm_km_pgremove() that 2044 * is used for "intrsafe" objects. 2045 */ 2046 /* 2047 * remove mappings from pmap and drop the pages 2048 * from the object. offsets are always relative 2049 * to vm_map_min(kernel_map). 2050 */ 2051 pmap_remove(pmap_kernel(), entry->start, entry->end); 2052 uvm_km_pgremove(entry->object.uvm_obj, 2053 entry->start - vm_map_min(kernel_map), 2054 entry->end - vm_map_min(kernel_map)); 2055 2056 /* 2057 * null out kernel_object reference, we've just 2058 * dropped it 2059 */ 2060 entry->etype &= ~UVM_ET_OBJ; 2061 entry->object.uvm_obj = NULL; /* to be safe */ 2062 } else { 2063 /* remove mappings the standard way. */ 2064 pmap_remove(map->pmap, entry->start, entry->end); 2065 } 2066 } 2067 2068 /* 2069 * Remove all entries from start to end. 2070 * 2071 * If remove_holes, then remove ET_HOLE entries as well. 2072 * If markfree, entry will be properly marked free, otherwise, no replacement 2073 * entry will be put in the tree (corrupting the tree). 2074 */ 2075 void 2076 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 2077 struct uvm_map_deadq *dead, boolean_t remove_holes, 2078 boolean_t markfree) 2079 { 2080 struct vm_map_entry *prev_hint, *next, *entry; 2081 2082 start = MAX(start, map->min_offset); 2083 end = MIN(end, map->max_offset); 2084 if (start >= end) 2085 return; 2086 2087 if ((map->flags & VM_MAP_INTRSAFE) == 0) 2088 splassert(IPL_NONE); 2089 else 2090 splassert(IPL_VM); 2091 2092 /* Find first affected entry. */ 2093 entry = uvm_map_entrybyaddr(&map->addr, start); 2094 KDASSERT(entry != NULL && entry->start <= start); 2095 if (entry->end <= start && markfree) 2096 entry = RBT_NEXT(uvm_map_addr, entry); 2097 else 2098 UVM_MAP_CLIP_START(map, entry, start); 2099 2100 /* 2101 * Iterate entries until we reach end address. 2102 * prev_hint hints where the freed space can be appended to. 2103 */ 2104 prev_hint = NULL; 2105 for (; entry != NULL && entry->start < end; entry = next) { 2106 KDASSERT(entry->start >= start); 2107 if (entry->end > end || !markfree) 2108 UVM_MAP_CLIP_END(map, entry, end); 2109 KDASSERT(entry->start >= start && entry->end <= end); 2110 next = RBT_NEXT(uvm_map_addr, entry); 2111 2112 /* Don't remove holes unless asked to do so. */ 2113 if (UVM_ET_ISHOLE(entry)) { 2114 if (!remove_holes) { 2115 prev_hint = entry; 2116 continue; 2117 } 2118 } 2119 2120 /* A stack has been removed.. */ 2121 if (UVM_ET_ISSTACK(entry) && (map->flags & VM_MAP_ISVMSPACE)) 2122 map->serial++; 2123 2124 /* Kill entry. */ 2125 uvm_unmap_kill_entry(map, entry); 2126 2127 /* Update space usage. */ 2128 if ((map->flags & VM_MAP_ISVMSPACE) && 2129 entry->object.uvm_obj == NULL && 2130 !UVM_ET_ISHOLE(entry)) { 2131 ((struct vmspace *)map)->vm_dused -= 2132 uvmspace_dused(map, entry->start, entry->end); 2133 } 2134 if (!UVM_ET_ISHOLE(entry)) 2135 map->size -= entry->end - entry->start; 2136 2137 /* Actual removal of entry. */ 2138 uvm_mapent_mkfree(map, entry, &prev_hint, dead, markfree); 2139 } 2140 2141 pmap_update(vm_map_pmap(map)); 2142 2143 #ifdef VMMAP_DEBUG 2144 if (markfree) { 2145 for (entry = uvm_map_entrybyaddr(&map->addr, start); 2146 entry != NULL && entry->start < end; 2147 entry = RBT_NEXT(uvm_map_addr, entry)) { 2148 KDASSERT(entry->end <= start || 2149 entry->start == entry->end || 2150 UVM_ET_ISHOLE(entry)); 2151 } 2152 } else { 2153 vaddr_t a; 2154 for (a = start; a < end; a += PAGE_SIZE) 2155 KDASSERT(uvm_map_entrybyaddr(&map->addr, a) == NULL); 2156 } 2157 #endif 2158 } 2159 2160 /* 2161 * Mark all entries from first until end (exclusive) as pageable. 2162 * 2163 * Lock must be exclusive on entry and will not be touched. 2164 */ 2165 void 2166 uvm_map_pageable_pgon(struct vm_map *map, struct vm_map_entry *first, 2167 struct vm_map_entry *end, vaddr_t start_addr, vaddr_t end_addr) 2168 { 2169 struct vm_map_entry *iter; 2170 2171 for (iter = first; iter != end; 2172 iter = RBT_NEXT(uvm_map_addr, iter)) { 2173 KDASSERT(iter->start >= start_addr && iter->end <= end_addr); 2174 if (!VM_MAPENT_ISWIRED(iter) || UVM_ET_ISHOLE(iter)) 2175 continue; 2176 2177 iter->wired_count = 0; 2178 uvm_fault_unwire_locked(map, iter->start, iter->end); 2179 } 2180 } 2181 2182 /* 2183 * Mark all entries from first until end (exclusive) as wired. 2184 * 2185 * Lockflags determines the lock state on return from this function. 2186 * Lock must be exclusive on entry. 2187 */ 2188 int 2189 uvm_map_pageable_wire(struct vm_map *map, struct vm_map_entry *first, 2190 struct vm_map_entry *end, vaddr_t start_addr, vaddr_t end_addr, 2191 int lockflags) 2192 { 2193 struct vm_map_entry *iter; 2194 #ifdef DIAGNOSTIC 2195 unsigned int timestamp_save; 2196 #endif 2197 int error; 2198 2199 /* 2200 * Wire pages in two passes: 2201 * 2202 * 1: holding the write lock, we create any anonymous maps that need 2203 * to be created. then we clip each map entry to the region to 2204 * be wired and increment its wiring count. 2205 * 2206 * 2: we downgrade to a read lock, and call uvm_fault_wire to fault 2207 * in the pages for any newly wired area (wired_count == 1). 2208 * 2209 * downgrading to a read lock for uvm_fault_wire avoids a possible 2210 * deadlock with another thread that may have faulted on one of 2211 * the pages to be wired (it would mark the page busy, blocking 2212 * us, then in turn block on the map lock that we hold). 2213 * because we keep the read lock on the map, the copy-on-write 2214 * status of the entries we modify here cannot change. 2215 */ 2216 for (iter = first; iter != end; 2217 iter = RBT_NEXT(uvm_map_addr, iter)) { 2218 KDASSERT(iter->start >= start_addr && iter->end <= end_addr); 2219 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2220 iter->protection == PROT_NONE) 2221 continue; 2222 2223 /* 2224 * Perform actions of vm_map_lookup that need the write lock. 2225 * - create an anonymous map for copy-on-write 2226 * - anonymous map for zero-fill 2227 * Skip submaps. 2228 */ 2229 if (!VM_MAPENT_ISWIRED(iter) && !UVM_ET_ISSUBMAP(iter) && 2230 UVM_ET_ISNEEDSCOPY(iter) && 2231 ((iter->protection & PROT_WRITE) || 2232 iter->object.uvm_obj == NULL)) { 2233 amap_copy(map, iter, M_WAITOK, 2234 UVM_ET_ISSTACK(iter) ? FALSE : TRUE, 2235 iter->start, iter->end); 2236 } 2237 iter->wired_count++; 2238 } 2239 2240 /* 2241 * Pass 2. 2242 */ 2243 #ifdef DIAGNOSTIC 2244 timestamp_save = map->timestamp; 2245 #endif 2246 vm_map_busy(map); 2247 vm_map_downgrade(map); 2248 2249 error = 0; 2250 for (iter = first; error == 0 && iter != end; 2251 iter = RBT_NEXT(uvm_map_addr, iter)) { 2252 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2253 iter->protection == PROT_NONE) 2254 continue; 2255 2256 error = uvm_fault_wire(map, iter->start, iter->end, 2257 iter->protection); 2258 } 2259 2260 if (error) { 2261 /* 2262 * uvm_fault_wire failure 2263 * 2264 * Reacquire lock and undo our work. 2265 */ 2266 vm_map_upgrade(map); 2267 vm_map_unbusy(map); 2268 #ifdef DIAGNOSTIC 2269 if (timestamp_save != map->timestamp) 2270 panic("uvm_map_pageable_wire: stale map"); 2271 #endif 2272 2273 /* 2274 * first is no longer needed to restart loops. 2275 * Use it as iterator to unmap successful mappings. 2276 */ 2277 for (; first != iter; 2278 first = RBT_NEXT(uvm_map_addr, first)) { 2279 if (UVM_ET_ISHOLE(first) || 2280 first->start == first->end || 2281 first->protection == PROT_NONE) 2282 continue; 2283 2284 first->wired_count--; 2285 if (!VM_MAPENT_ISWIRED(first)) { 2286 uvm_fault_unwire_locked(map, 2287 iter->start, iter->end); 2288 } 2289 } 2290 2291 /* decrease counter in the rest of the entries */ 2292 for (; iter != end; 2293 iter = RBT_NEXT(uvm_map_addr, iter)) { 2294 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2295 iter->protection == PROT_NONE) 2296 continue; 2297 2298 iter->wired_count--; 2299 } 2300 2301 if ((lockflags & UVM_LK_EXIT) == 0) 2302 vm_map_unlock(map); 2303 return error; 2304 } 2305 2306 /* We are currently holding a read lock. */ 2307 if ((lockflags & UVM_LK_EXIT) == 0) { 2308 vm_map_unbusy(map); 2309 vm_map_unlock_read(map); 2310 } else { 2311 vm_map_upgrade(map); 2312 vm_map_unbusy(map); 2313 #ifdef DIAGNOSTIC 2314 if (timestamp_save != map->timestamp) 2315 panic("uvm_map_pageable_wire: stale map"); 2316 #endif 2317 } 2318 return 0; 2319 } 2320 2321 /* 2322 * uvm_map_pageable: set pageability of a range in a map. 2323 * 2324 * Flags: 2325 * UVM_LK_ENTER: map is already locked by caller 2326 * UVM_LK_EXIT: don't unlock map on exit 2327 * 2328 * The full range must be in use (entries may not have fspace != 0). 2329 * UVM_ET_HOLE counts as unmapped. 2330 */ 2331 int 2332 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 2333 boolean_t new_pageable, int lockflags) 2334 { 2335 struct vm_map_entry *first, *last, *tmp; 2336 int error; 2337 2338 start = trunc_page(start); 2339 end = round_page(end); 2340 2341 if (start > end) 2342 return EINVAL; 2343 if (start == end) 2344 return 0; /* nothing to do */ 2345 if (start < map->min_offset) 2346 return EFAULT; /* why? see first XXX below */ 2347 if (end > map->max_offset) 2348 return EINVAL; /* why? see second XXX below */ 2349 2350 KASSERT(map->flags & VM_MAP_PAGEABLE); 2351 if ((lockflags & UVM_LK_ENTER) == 0) 2352 vm_map_lock(map); 2353 2354 /* 2355 * Find first entry. 2356 * 2357 * Initial test on start is different, because of the different 2358 * error returned. Rest is tested further down. 2359 */ 2360 first = uvm_map_entrybyaddr(&map->addr, start); 2361 if (first->end <= start || UVM_ET_ISHOLE(first)) { 2362 /* 2363 * XXX if the first address is not mapped, it is EFAULT? 2364 */ 2365 error = EFAULT; 2366 goto out; 2367 } 2368 2369 /* Check that the range has no holes. */ 2370 for (last = first; last != NULL && last->start < end; 2371 last = RBT_NEXT(uvm_map_addr, last)) { 2372 if (UVM_ET_ISHOLE(last) || 2373 (last->end < end && VMMAP_FREE_END(last) != last->end)) { 2374 /* 2375 * XXX unmapped memory in range, why is it EINVAL 2376 * instead of EFAULT? 2377 */ 2378 error = EINVAL; 2379 goto out; 2380 } 2381 } 2382 2383 /* 2384 * Last ended at the first entry after the range. 2385 * Move back one step. 2386 * 2387 * Note that last may be NULL. 2388 */ 2389 if (last == NULL) { 2390 last = RBT_MAX(uvm_map_addr, &map->addr); 2391 if (last->end < end) { 2392 error = EINVAL; 2393 goto out; 2394 } 2395 } else { 2396 KASSERT(last != first); 2397 last = RBT_PREV(uvm_map_addr, last); 2398 } 2399 2400 /* Wire/unwire pages here. */ 2401 if (new_pageable) { 2402 /* 2403 * Mark pageable. 2404 * entries that are not wired are untouched. 2405 */ 2406 if (VM_MAPENT_ISWIRED(first)) 2407 UVM_MAP_CLIP_START(map, first, start); 2408 /* 2409 * Split last at end. 2410 * Make tmp be the first entry after what is to be touched. 2411 * If last is not wired, don't touch it. 2412 */ 2413 if (VM_MAPENT_ISWIRED(last)) { 2414 UVM_MAP_CLIP_END(map, last, end); 2415 tmp = RBT_NEXT(uvm_map_addr, last); 2416 } else 2417 tmp = last; 2418 2419 uvm_map_pageable_pgon(map, first, tmp, start, end); 2420 error = 0; 2421 2422 out: 2423 if ((lockflags & UVM_LK_EXIT) == 0) 2424 vm_map_unlock(map); 2425 return error; 2426 } else { 2427 /* 2428 * Mark entries wired. 2429 * entries are always touched (because recovery needs this). 2430 */ 2431 if (!VM_MAPENT_ISWIRED(first)) 2432 UVM_MAP_CLIP_START(map, first, start); 2433 /* 2434 * Split last at end. 2435 * Make tmp be the first entry after what is to be touched. 2436 * If last is not wired, don't touch it. 2437 */ 2438 if (!VM_MAPENT_ISWIRED(last)) { 2439 UVM_MAP_CLIP_END(map, last, end); 2440 tmp = RBT_NEXT(uvm_map_addr, last); 2441 } else 2442 tmp = last; 2443 2444 return uvm_map_pageable_wire(map, first, tmp, start, end, 2445 lockflags); 2446 } 2447 } 2448 2449 /* 2450 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2451 * all mapped regions. 2452 * 2453 * Map must not be locked. 2454 * If no flags are specified, all ragions are unwired. 2455 */ 2456 int 2457 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 2458 { 2459 vsize_t size; 2460 struct vm_map_entry *iter; 2461 2462 KASSERT(map->flags & VM_MAP_PAGEABLE); 2463 vm_map_lock(map); 2464 2465 if (flags == 0) { 2466 uvm_map_pageable_pgon(map, RBT_MIN(uvm_map_addr, &map->addr), 2467 NULL, map->min_offset, map->max_offset); 2468 2469 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2470 vm_map_unlock(map); 2471 return 0; 2472 } 2473 2474 if (flags & MCL_FUTURE) 2475 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2476 if (!(flags & MCL_CURRENT)) { 2477 vm_map_unlock(map); 2478 return 0; 2479 } 2480 2481 /* 2482 * Count number of pages in all non-wired entries. 2483 * If the number exceeds the limit, abort. 2484 */ 2485 size = 0; 2486 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2487 if (VM_MAPENT_ISWIRED(iter) || UVM_ET_ISHOLE(iter)) 2488 continue; 2489 2490 size += iter->end - iter->start; 2491 } 2492 2493 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2494 vm_map_unlock(map); 2495 return ENOMEM; 2496 } 2497 2498 /* XXX non-pmap_wired_count case must be handled by caller */ 2499 #ifdef pmap_wired_count 2500 if (limit != 0 && 2501 size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit) { 2502 vm_map_unlock(map); 2503 return ENOMEM; 2504 } 2505 #endif 2506 2507 /* 2508 * uvm_map_pageable_wire will release lcok 2509 */ 2510 return uvm_map_pageable_wire(map, RBT_MIN(uvm_map_addr, &map->addr), 2511 NULL, map->min_offset, map->max_offset, 0); 2512 } 2513 2514 /* 2515 * Initialize map. 2516 * 2517 * Allocates sufficient entries to describe the free memory in the map. 2518 */ 2519 void 2520 uvm_map_setup(struct vm_map *map, vaddr_t min, vaddr_t max, int flags) 2521 { 2522 int i; 2523 2524 KASSERT((min & (vaddr_t)PAGE_MASK) == 0); 2525 KASSERT((max & (vaddr_t)PAGE_MASK) == 0 || 2526 (max & (vaddr_t)PAGE_MASK) == (vaddr_t)PAGE_MASK); 2527 2528 /* 2529 * Update parameters. 2530 * 2531 * This code handles (vaddr_t)-1 and other page mask ending addresses 2532 * properly. 2533 * We lose the top page if the full virtual address space is used. 2534 */ 2535 if (max & (vaddr_t)PAGE_MASK) { 2536 max += 1; 2537 if (max == 0) /* overflow */ 2538 max -= PAGE_SIZE; 2539 } 2540 2541 RBT_INIT(uvm_map_addr, &map->addr); 2542 map->uaddr_exe = NULL; 2543 for (i = 0; i < nitems(map->uaddr_any); ++i) 2544 map->uaddr_any[i] = NULL; 2545 map->uaddr_brk_stack = NULL; 2546 2547 map->size = 0; 2548 map->ref_count = 0; 2549 map->min_offset = min; 2550 map->max_offset = max; 2551 map->b_start = map->b_end = 0; /* Empty brk() area by default. */ 2552 map->s_start = map->s_end = 0; /* Empty stack area by default. */ 2553 map->flags = flags; 2554 map->timestamp = 0; 2555 rw_init_flags(&map->lock, "vmmaplk", RWL_DUPOK); 2556 mtx_init(&map->mtx, IPL_VM); 2557 mtx_init(&map->flags_lock, IPL_VM); 2558 2559 /* Configure the allocators. */ 2560 if (flags & VM_MAP_ISVMSPACE) 2561 uvm_map_setup_md(map); 2562 else 2563 map->uaddr_any[3] = &uaddr_kbootstrap; 2564 2565 /* 2566 * Fill map entries. 2567 * We do not need to write-lock the map here because only the current 2568 * thread sees it right now. Initialize ref_count to 0 above to avoid 2569 * bogus triggering of lock-not-held assertions. 2570 */ 2571 uvm_map_setup_entries(map); 2572 uvm_tree_sanity(map, __FILE__, __LINE__); 2573 map->ref_count = 1; 2574 } 2575 2576 /* 2577 * Destroy the map. 2578 * 2579 * This is the inverse operation to uvm_map_setup. 2580 */ 2581 void 2582 uvm_map_teardown(struct vm_map *map) 2583 { 2584 struct uvm_map_deadq dead_entries; 2585 struct vm_map_entry *entry, *tmp; 2586 #ifdef VMMAP_DEBUG 2587 size_t numq, numt; 2588 #endif 2589 int i; 2590 2591 KERNEL_ASSERT_LOCKED(); 2592 KERNEL_UNLOCK(); 2593 KERNEL_ASSERT_UNLOCKED(); 2594 2595 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 2596 2597 /* Remove address selectors. */ 2598 uvm_addr_destroy(map->uaddr_exe); 2599 map->uaddr_exe = NULL; 2600 for (i = 0; i < nitems(map->uaddr_any); i++) { 2601 uvm_addr_destroy(map->uaddr_any[i]); 2602 map->uaddr_any[i] = NULL; 2603 } 2604 uvm_addr_destroy(map->uaddr_brk_stack); 2605 map->uaddr_brk_stack = NULL; 2606 2607 /* 2608 * Remove entries. 2609 * 2610 * The following is based on graph breadth-first search. 2611 * 2612 * In color terms: 2613 * - the dead_entries set contains all nodes that are reachable 2614 * (i.e. both the black and the grey nodes) 2615 * - any entry not in dead_entries is white 2616 * - any entry that appears in dead_entries before entry, 2617 * is black, the rest is grey. 2618 * The set [entry, end] is also referred to as the wavefront. 2619 * 2620 * Since the tree is always a fully connected graph, the breadth-first 2621 * search guarantees that each vmmap_entry is visited exactly once. 2622 * The vm_map is broken down in linear time. 2623 */ 2624 TAILQ_INIT(&dead_entries); 2625 if ((entry = RBT_ROOT(uvm_map_addr, &map->addr)) != NULL) 2626 DEAD_ENTRY_PUSH(&dead_entries, entry); 2627 while (entry != NULL) { 2628 sched_pause(yield); 2629 uvm_unmap_kill_entry(map, entry); 2630 if ((tmp = RBT_LEFT(uvm_map_addr, entry)) != NULL) 2631 DEAD_ENTRY_PUSH(&dead_entries, tmp); 2632 if ((tmp = RBT_RIGHT(uvm_map_addr, entry)) != NULL) 2633 DEAD_ENTRY_PUSH(&dead_entries, tmp); 2634 /* Update wave-front. */ 2635 entry = TAILQ_NEXT(entry, dfree.deadq); 2636 } 2637 2638 #ifdef VMMAP_DEBUG 2639 numt = numq = 0; 2640 RBT_FOREACH(entry, uvm_map_addr, &map->addr) 2641 numt++; 2642 TAILQ_FOREACH(entry, &dead_entries, dfree.deadq) 2643 numq++; 2644 KASSERT(numt == numq); 2645 #endif 2646 uvm_unmap_detach(&dead_entries, UVM_PLA_WAITOK); 2647 2648 KERNEL_LOCK(); 2649 2650 pmap_destroy(map->pmap); 2651 map->pmap = NULL; 2652 } 2653 2654 /* 2655 * Populate map with free-memory entries. 2656 * 2657 * Map must be initialized and empty. 2658 */ 2659 void 2660 uvm_map_setup_entries(struct vm_map *map) 2661 { 2662 KDASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 2663 2664 uvm_map_fix_space(map, NULL, map->min_offset, map->max_offset, 0); 2665 } 2666 2667 /* 2668 * Split entry at given address. 2669 * 2670 * orig: entry that is to be split. 2671 * next: a newly allocated map entry that is not linked. 2672 * split: address at which the split is done. 2673 */ 2674 void 2675 uvm_map_splitentry(struct vm_map *map, struct vm_map_entry *orig, 2676 struct vm_map_entry *next, vaddr_t split) 2677 { 2678 struct uvm_addr_state *free, *free_before; 2679 vsize_t adj; 2680 2681 if ((split & PAGE_MASK) != 0) { 2682 panic("uvm_map_splitentry: split address 0x%lx " 2683 "not on page boundary!", split); 2684 } 2685 KDASSERT(map != NULL && orig != NULL && next != NULL); 2686 uvm_tree_sanity(map, __FILE__, __LINE__); 2687 KASSERT(orig->start < split && VMMAP_FREE_END(orig) > split); 2688 2689 #ifdef VMMAP_DEBUG 2690 KDASSERT(RBT_FIND(uvm_map_addr, &map->addr, orig) == orig); 2691 KDASSERT(RBT_FIND(uvm_map_addr, &map->addr, next) != next); 2692 #endif /* VMMAP_DEBUG */ 2693 2694 /* 2695 * Free space will change, unlink from free space tree. 2696 */ 2697 free = uvm_map_uaddr_e(map, orig); 2698 uvm_mapent_free_remove(map, free, orig); 2699 2700 adj = split - orig->start; 2701 2702 uvm_mapent_copy(orig, next); 2703 if (split >= orig->end) { 2704 next->etype = 0; 2705 next->offset = 0; 2706 next->wired_count = 0; 2707 next->start = next->end = split; 2708 next->guard = 0; 2709 next->fspace = VMMAP_FREE_END(orig) - split; 2710 next->aref.ar_amap = NULL; 2711 next->aref.ar_pageoff = 0; 2712 orig->guard = MIN(orig->guard, split - orig->end); 2713 orig->fspace = split - VMMAP_FREE_START(orig); 2714 } else { 2715 orig->fspace = 0; 2716 orig->guard = 0; 2717 orig->end = next->start = split; 2718 2719 if (next->aref.ar_amap) { 2720 KERNEL_LOCK(); 2721 amap_splitref(&orig->aref, &next->aref, adj); 2722 KERNEL_UNLOCK(); 2723 } 2724 if (UVM_ET_ISSUBMAP(orig)) { 2725 uvm_map_reference(next->object.sub_map); 2726 next->offset += adj; 2727 } else if (UVM_ET_ISOBJ(orig)) { 2728 if (next->object.uvm_obj->pgops && 2729 next->object.uvm_obj->pgops->pgo_reference) { 2730 KERNEL_LOCK(); 2731 next->object.uvm_obj->pgops->pgo_reference( 2732 next->object.uvm_obj); 2733 KERNEL_UNLOCK(); 2734 } 2735 next->offset += adj; 2736 } 2737 } 2738 2739 /* 2740 * Link next into address tree. 2741 * Link orig and next into free-space tree. 2742 * 2743 * Don't insert 'next' into the addr tree until orig has been linked, 2744 * in case the free-list looks at adjecent entries in the addr tree 2745 * for its decisions. 2746 */ 2747 if (orig->fspace > 0) 2748 free_before = free; 2749 else 2750 free_before = uvm_map_uaddr_e(map, orig); 2751 uvm_mapent_free_insert(map, free_before, orig); 2752 uvm_mapent_addr_insert(map, next); 2753 uvm_mapent_free_insert(map, free, next); 2754 2755 uvm_tree_sanity(map, __FILE__, __LINE__); 2756 } 2757 2758 2759 #ifdef VMMAP_DEBUG 2760 2761 void 2762 uvm_tree_assert(struct vm_map *map, int test, char *test_str, 2763 char *file, int line) 2764 { 2765 char* map_special; 2766 2767 if (test) 2768 return; 2769 2770 if (map == kernel_map) 2771 map_special = " (kernel_map)"; 2772 else if (map == kmem_map) 2773 map_special = " (kmem_map)"; 2774 else 2775 map_special = ""; 2776 panic("uvm_tree_sanity %p%s (%s %d): %s", map, map_special, file, 2777 line, test_str); 2778 } 2779 2780 /* 2781 * Check that map is sane. 2782 */ 2783 void 2784 uvm_tree_sanity(struct vm_map *map, char *file, int line) 2785 { 2786 struct vm_map_entry *iter; 2787 vaddr_t addr; 2788 vaddr_t min, max, bound; /* Bounds checker. */ 2789 struct uvm_addr_state *free; 2790 2791 addr = vm_map_min(map); 2792 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2793 /* 2794 * Valid start, end. 2795 * Catch overflow for end+fspace. 2796 */ 2797 UVM_ASSERT(map, iter->end >= iter->start, file, line); 2798 UVM_ASSERT(map, VMMAP_FREE_END(iter) >= iter->end, file, line); 2799 2800 /* May not be empty. */ 2801 UVM_ASSERT(map, iter->start < VMMAP_FREE_END(iter), 2802 file, line); 2803 2804 /* Addresses for entry must lie within map boundaries. */ 2805 UVM_ASSERT(map, iter->start >= vm_map_min(map) && 2806 VMMAP_FREE_END(iter) <= vm_map_max(map), file, line); 2807 2808 /* Tree may not have gaps. */ 2809 UVM_ASSERT(map, iter->start == addr, file, line); 2810 addr = VMMAP_FREE_END(iter); 2811 2812 /* 2813 * Free space may not cross boundaries, unless the same 2814 * free list is used on both sides of the border. 2815 */ 2816 min = VMMAP_FREE_START(iter); 2817 max = VMMAP_FREE_END(iter); 2818 2819 while (min < max && 2820 (bound = uvm_map_boundary(map, min, max)) != max) { 2821 UVM_ASSERT(map, 2822 uvm_map_uaddr(map, bound - 1) == 2823 uvm_map_uaddr(map, bound), 2824 file, line); 2825 min = bound; 2826 } 2827 2828 free = uvm_map_uaddr_e(map, iter); 2829 if (free) { 2830 UVM_ASSERT(map, (iter->etype & UVM_ET_FREEMAPPED) != 0, 2831 file, line); 2832 } else { 2833 UVM_ASSERT(map, (iter->etype & UVM_ET_FREEMAPPED) == 0, 2834 file, line); 2835 } 2836 } 2837 UVM_ASSERT(map, addr == vm_map_max(map), file, line); 2838 } 2839 2840 void 2841 uvm_tree_size_chk(struct vm_map *map, char *file, int line) 2842 { 2843 struct vm_map_entry *iter; 2844 vsize_t size; 2845 2846 size = 0; 2847 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2848 if (!UVM_ET_ISHOLE(iter)) 2849 size += iter->end - iter->start; 2850 } 2851 2852 if (map->size != size) 2853 printf("map size = 0x%lx, should be 0x%lx\n", map->size, size); 2854 UVM_ASSERT(map, map->size == size, file, line); 2855 2856 vmspace_validate(map); 2857 } 2858 2859 /* 2860 * This function validates the statistics on vmspace. 2861 */ 2862 void 2863 vmspace_validate(struct vm_map *map) 2864 { 2865 struct vmspace *vm; 2866 struct vm_map_entry *iter; 2867 vaddr_t imin, imax; 2868 vaddr_t stack_begin, stack_end; /* Position of stack. */ 2869 vsize_t stack, heap; /* Measured sizes. */ 2870 2871 if (!(map->flags & VM_MAP_ISVMSPACE)) 2872 return; 2873 2874 vm = (struct vmspace *)map; 2875 stack_begin = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 2876 stack_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 2877 2878 stack = heap = 0; 2879 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2880 imin = imax = iter->start; 2881 2882 if (UVM_ET_ISHOLE(iter) || iter->object.uvm_obj != NULL) 2883 continue; 2884 2885 /* 2886 * Update stack, heap. 2887 * Keep in mind that (theoretically) the entries of 2888 * userspace and stack may be joined. 2889 */ 2890 while (imin != iter->end) { 2891 /* 2892 * Set imax to the first boundary crossed between 2893 * imin and stack addresses. 2894 */ 2895 imax = iter->end; 2896 if (imin < stack_begin && imax > stack_begin) 2897 imax = stack_begin; 2898 else if (imin < stack_end && imax > stack_end) 2899 imax = stack_end; 2900 2901 if (imin >= stack_begin && imin < stack_end) 2902 stack += imax - imin; 2903 else 2904 heap += imax - imin; 2905 imin = imax; 2906 } 2907 } 2908 2909 heap >>= PAGE_SHIFT; 2910 if (heap != vm->vm_dused) { 2911 printf("vmspace stack range: 0x%lx-0x%lx\n", 2912 stack_begin, stack_end); 2913 panic("vmspace_validate: vmspace.vm_dused invalid, " 2914 "expected %ld pgs, got %ld pgs in map %p", 2915 heap, vm->vm_dused, 2916 map); 2917 } 2918 } 2919 2920 #endif /* VMMAP_DEBUG */ 2921 2922 /* 2923 * uvm_map_init: init mapping system at boot time. note that we allocate 2924 * and init the static pool of structs vm_map_entry for the kernel here. 2925 */ 2926 void 2927 uvm_map_init(void) 2928 { 2929 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 2930 int lcv; 2931 2932 /* now set up static pool of kernel map entries ... */ 2933 mtx_init(&uvm_kmapent_mtx, IPL_VM); 2934 SLIST_INIT(&uvm.kentry_free); 2935 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 2936 SLIST_INSERT_HEAD(&uvm.kentry_free, 2937 &kernel_map_entry[lcv], daddrs.addr_kentry); 2938 } 2939 2940 /* initialize the map-related pools. */ 2941 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 0, 2942 IPL_NONE, PR_WAITOK, "vmsppl", NULL); 2943 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 2944 IPL_VM, PR_WAITOK, "vmmpepl", NULL); 2945 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 0, 2946 IPL_VM, 0, "vmmpekpl", NULL); 2947 pool_sethiwat(&uvm_map_entry_pool, 8192); 2948 2949 uvm_addr_init(); 2950 } 2951 2952 #if defined(DDB) 2953 2954 /* 2955 * DDB hooks 2956 */ 2957 2958 /* 2959 * uvm_map_printit: actually prints the map 2960 */ 2961 void 2962 uvm_map_printit(struct vm_map *map, boolean_t full, 2963 int (*pr)(const char *, ...)) 2964 { 2965 struct vmspace *vm; 2966 struct vm_map_entry *entry; 2967 struct uvm_addr_state *free; 2968 int in_free, i; 2969 char buf[8]; 2970 2971 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 2972 (*pr)("\tbrk() allocate range: 0x%lx-0x%lx\n", 2973 map->b_start, map->b_end); 2974 (*pr)("\tstack allocate range: 0x%lx-0x%lx\n", 2975 map->s_start, map->s_end); 2976 (*pr)("\tsz=%u, ref=%d, version=%u, flags=0x%x\n", 2977 map->size, map->ref_count, map->timestamp, 2978 map->flags); 2979 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 2980 pmap_resident_count(map->pmap)); 2981 2982 /* struct vmspace handling. */ 2983 if (map->flags & VM_MAP_ISVMSPACE) { 2984 vm = (struct vmspace *)map; 2985 2986 (*pr)("\tvm_refcnt=%d vm_shm=%p vm_rssize=%u vm_swrss=%u\n", 2987 vm->vm_refcnt, vm->vm_shm, vm->vm_rssize, vm->vm_swrss); 2988 (*pr)("\tvm_tsize=%u vm_dsize=%u\n", 2989 vm->vm_tsize, vm->vm_dsize); 2990 (*pr)("\tvm_taddr=%p vm_daddr=%p\n", 2991 vm->vm_taddr, vm->vm_daddr); 2992 (*pr)("\tvm_maxsaddr=%p vm_minsaddr=%p\n", 2993 vm->vm_maxsaddr, vm->vm_minsaddr); 2994 } 2995 2996 if (!full) 2997 goto print_uaddr; 2998 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 2999 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 3000 entry, entry->start, entry->end, entry->object.uvm_obj, 3001 (long long)entry->offset, entry->aref.ar_amap, 3002 entry->aref.ar_pageoff); 3003 (*pr)("\tsubmap=%c, cow=%c, nc=%c, stack=%c, prot(max)=%d/%d, inh=%d, " 3004 "wc=%d, adv=%d\n", 3005 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 3006 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 3007 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 3008 (entry->etype & UVM_ET_STACK) ? 'T' : 'F', 3009 entry->protection, entry->max_protection, 3010 entry->inheritance, entry->wired_count, entry->advice); 3011 3012 free = uvm_map_uaddr_e(map, entry); 3013 in_free = (free != NULL); 3014 (*pr)("\thole=%c, free=%c, guard=0x%lx, " 3015 "free=0x%lx-0x%lx\n", 3016 (entry->etype & UVM_ET_HOLE) ? 'T' : 'F', 3017 in_free ? 'T' : 'F', 3018 entry->guard, 3019 VMMAP_FREE_START(entry), VMMAP_FREE_END(entry)); 3020 (*pr)("\tfspace_augment=%lu\n", entry->fspace_augment); 3021 (*pr)("\tfreemapped=%c, uaddr=%p\n", 3022 (entry->etype & UVM_ET_FREEMAPPED) ? 'T' : 'F', free); 3023 if (free) { 3024 (*pr)("\t\t(0x%lx-0x%lx %s)\n", 3025 free->uaddr_minaddr, free->uaddr_maxaddr, 3026 free->uaddr_functions->uaddr_name); 3027 } 3028 } 3029 3030 print_uaddr: 3031 uvm_addr_print(map->uaddr_exe, "exe", full, pr); 3032 for (i = 0; i < nitems(map->uaddr_any); i++) { 3033 snprintf(&buf[0], sizeof(buf), "any[%d]", i); 3034 uvm_addr_print(map->uaddr_any[i], &buf[0], full, pr); 3035 } 3036 uvm_addr_print(map->uaddr_brk_stack, "brk/stack", full, pr); 3037 } 3038 3039 /* 3040 * uvm_object_printit: actually prints the object 3041 */ 3042 void 3043 uvm_object_printit(uobj, full, pr) 3044 struct uvm_object *uobj; 3045 boolean_t full; 3046 int (*pr)(const char *, ...); 3047 { 3048 struct vm_page *pg; 3049 int cnt = 0; 3050 3051 (*pr)("OBJECT %p: pgops=%p, npages=%d, ", 3052 uobj, uobj->pgops, uobj->uo_npages); 3053 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 3054 (*pr)("refs=<SYSTEM>\n"); 3055 else 3056 (*pr)("refs=%d\n", uobj->uo_refs); 3057 3058 if (!full) { 3059 return; 3060 } 3061 (*pr)(" PAGES <pg,offset>:\n "); 3062 RBT_FOREACH(pg, uvm_objtree, &uobj->memt) { 3063 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 3064 if ((cnt % 3) == 2) { 3065 (*pr)("\n "); 3066 } 3067 cnt++; 3068 } 3069 if ((cnt % 3) != 2) { 3070 (*pr)("\n"); 3071 } 3072 } 3073 3074 /* 3075 * uvm_page_printit: actually print the page 3076 */ 3077 static const char page_flagbits[] = 3078 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5CLEANCHK\6RELEASED\7FAKE\10RDONLY" 3079 "\11ZERO\12DEV\15PAGER1\21FREE\22INACTIVE\23ACTIVE\25ANON\26AOBJ" 3080 "\27ENCRYPT\31PMAP0\32PMAP1\33PMAP2\34PMAP3\35PMAP4\36PMAP5"; 3081 3082 void 3083 uvm_page_printit(pg, full, pr) 3084 struct vm_page *pg; 3085 boolean_t full; 3086 int (*pr)(const char *, ...); 3087 { 3088 struct vm_page *tpg; 3089 struct uvm_object *uobj; 3090 struct pglist *pgl; 3091 3092 (*pr)("PAGE %p:\n", pg); 3093 (*pr)(" flags=%b, vers=%d, wire_count=%d, pa=0x%llx\n", 3094 pg->pg_flags, page_flagbits, pg->pg_version, pg->wire_count, 3095 (long long)pg->phys_addr); 3096 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx\n", 3097 pg->uobject, pg->uanon, (long long)pg->offset); 3098 #if defined(UVM_PAGE_TRKOWN) 3099 if (pg->pg_flags & PG_BUSY) 3100 (*pr)(" owning thread = %d, tag=%s", 3101 pg->owner, pg->owner_tag); 3102 else 3103 (*pr)(" page not busy, no owner"); 3104 #else 3105 (*pr)(" [page ownership tracking disabled]"); 3106 #endif 3107 (*pr)("\tvm_page_md %p\n", &pg->mdpage); 3108 3109 if (!full) 3110 return; 3111 3112 /* cross-verify object/anon */ 3113 if ((pg->pg_flags & PQ_FREE) == 0) { 3114 if (pg->pg_flags & PQ_ANON) { 3115 if (pg->uanon == NULL || pg->uanon->an_page != pg) 3116 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 3117 (pg->uanon) ? pg->uanon->an_page : NULL); 3118 else 3119 (*pr)(" anon backpointer is OK\n"); 3120 } else { 3121 uobj = pg->uobject; 3122 if (uobj) { 3123 (*pr)(" checking object list\n"); 3124 RBT_FOREACH(tpg, uvm_objtree, &uobj->memt) { 3125 if (tpg == pg) { 3126 break; 3127 } 3128 } 3129 if (tpg) 3130 (*pr)(" page found on object list\n"); 3131 else 3132 (*pr)(" >>> PAGE NOT FOUND " 3133 "ON OBJECT LIST! <<<\n"); 3134 } 3135 } 3136 } 3137 3138 /* cross-verify page queue */ 3139 if (pg->pg_flags & PQ_FREE) { 3140 if (uvm_pmr_isfree(pg)) 3141 (*pr)(" page found in uvm_pmemrange\n"); 3142 else 3143 (*pr)(" >>> page not found in uvm_pmemrange <<<\n"); 3144 pgl = NULL; 3145 } else if (pg->pg_flags & PQ_INACTIVE) { 3146 pgl = (pg->pg_flags & PQ_SWAPBACKED) ? 3147 &uvm.page_inactive_swp : &uvm.page_inactive_obj; 3148 } else if (pg->pg_flags & PQ_ACTIVE) { 3149 pgl = &uvm.page_active; 3150 } else { 3151 pgl = NULL; 3152 } 3153 3154 if (pgl) { 3155 (*pr)(" checking pageq list\n"); 3156 TAILQ_FOREACH(tpg, pgl, pageq) { 3157 if (tpg == pg) { 3158 break; 3159 } 3160 } 3161 if (tpg) 3162 (*pr)(" page found on pageq list\n"); 3163 else 3164 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3165 } 3166 } 3167 #endif 3168 3169 /* 3170 * uvm_map_protect: change map protection 3171 * 3172 * => set_max means set max_protection. 3173 * => map must be unlocked. 3174 */ 3175 int 3176 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 3177 vm_prot_t new_prot, boolean_t set_max) 3178 { 3179 struct vm_map_entry *first, *iter; 3180 vm_prot_t old_prot; 3181 vm_prot_t mask; 3182 int error; 3183 3184 if (start > end) 3185 return EINVAL; 3186 start = MAX(start, map->min_offset); 3187 end = MIN(end, map->max_offset); 3188 if (start >= end) 3189 return 0; 3190 3191 error = 0; 3192 vm_map_lock(map); 3193 3194 /* 3195 * Set up first and last. 3196 * - first will contain first entry at or after start. 3197 */ 3198 first = uvm_map_entrybyaddr(&map->addr, start); 3199 KDASSERT(first != NULL); 3200 if (first->end <= start) 3201 first = RBT_NEXT(uvm_map_addr, first); 3202 3203 /* First, check for protection violations. */ 3204 for (iter = first; iter != NULL && iter->start < end; 3205 iter = RBT_NEXT(uvm_map_addr, iter)) { 3206 /* Treat memory holes as free space. */ 3207 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) 3208 continue; 3209 3210 if (UVM_ET_ISSUBMAP(iter)) { 3211 error = EINVAL; 3212 goto out; 3213 } 3214 if ((new_prot & iter->max_protection) != new_prot) { 3215 error = EACCES; 3216 goto out; 3217 } 3218 if (map == kernel_map && 3219 (new_prot & (PROT_WRITE | PROT_EXEC)) == (PROT_WRITE | PROT_EXEC)) 3220 panic("uvm_map_protect: kernel map W^X violation requested"); 3221 } 3222 3223 /* Fix protections. */ 3224 for (iter = first; iter != NULL && iter->start < end; 3225 iter = RBT_NEXT(uvm_map_addr, iter)) { 3226 /* Treat memory holes as free space. */ 3227 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) 3228 continue; 3229 3230 old_prot = iter->protection; 3231 3232 /* 3233 * Skip adapting protection iff old and new protection 3234 * are equal. 3235 */ 3236 if (set_max) { 3237 if (old_prot == (new_prot & old_prot) && 3238 iter->max_protection == new_prot) 3239 continue; 3240 } else { 3241 if (old_prot == new_prot) 3242 continue; 3243 } 3244 3245 UVM_MAP_CLIP_START(map, iter, start); 3246 UVM_MAP_CLIP_END(map, iter, end); 3247 3248 if (set_max) { 3249 iter->max_protection = new_prot; 3250 iter->protection &= new_prot; 3251 } else 3252 iter->protection = new_prot; 3253 3254 /* 3255 * update physical map if necessary. worry about copy-on-write 3256 * here -- CHECK THIS XXX 3257 */ 3258 if (iter->protection != old_prot) { 3259 mask = UVM_ET_ISCOPYONWRITE(iter) ? 3260 ~PROT_WRITE : PROT_MASK; 3261 3262 /* update pmap */ 3263 if ((iter->protection & mask) == PROT_NONE && 3264 VM_MAPENT_ISWIRED(iter)) { 3265 /* 3266 * TODO(ariane) this is stupid. wired_count 3267 * is 0 if not wired, otherwise anything 3268 * larger than 0 (incremented once each time 3269 * wire is called). 3270 * Mostly to be able to undo the damage on 3271 * failure. Not the actually be a wired 3272 * refcounter... 3273 * Originally: iter->wired_count--; 3274 * (don't we have to unwire this in the pmap 3275 * as well?) 3276 */ 3277 iter->wired_count = 0; 3278 } 3279 pmap_protect(map->pmap, iter->start, iter->end, 3280 iter->protection & mask); 3281 } 3282 3283 /* 3284 * If the map is configured to lock any future mappings, 3285 * wire this entry now if the old protection was PROT_NONE 3286 * and the new protection is not PROT_NONE. 3287 */ 3288 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 3289 VM_MAPENT_ISWIRED(iter) == 0 && 3290 old_prot == PROT_NONE && 3291 new_prot != PROT_NONE) { 3292 if (uvm_map_pageable(map, iter->start, iter->end, 3293 FALSE, UVM_LK_ENTER | UVM_LK_EXIT) != 0) { 3294 /* 3295 * If locking the entry fails, remember the 3296 * error if it's the first one. Note we 3297 * still continue setting the protection in 3298 * the map, but it will return the resource 3299 * storage condition regardless. 3300 * 3301 * XXX Ignore what the actual error is, 3302 * XXX just call it a resource shortage 3303 * XXX so that it doesn't get confused 3304 * XXX what uvm_map_protect() itself would 3305 * XXX normally return. 3306 */ 3307 error = ENOMEM; 3308 } 3309 } 3310 } 3311 pmap_update(map->pmap); 3312 3313 out: 3314 vm_map_unlock(map); 3315 return error; 3316 } 3317 3318 /* 3319 * uvmspace_alloc: allocate a vmspace structure. 3320 * 3321 * - structure includes vm_map and pmap 3322 * - XXX: no locking on this structure 3323 * - refcnt set to 1, rest must be init'd by caller 3324 */ 3325 struct vmspace * 3326 uvmspace_alloc(vaddr_t min, vaddr_t max, boolean_t pageable, 3327 boolean_t remove_holes) 3328 { 3329 struct vmspace *vm; 3330 3331 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK | PR_ZERO); 3332 uvmspace_init(vm, NULL, min, max, pageable, remove_holes); 3333 return (vm); 3334 } 3335 3336 /* 3337 * uvmspace_init: initialize a vmspace structure. 3338 * 3339 * - XXX: no locking on this structure 3340 * - refcnt set to 1, rest must be init'd by caller 3341 */ 3342 void 3343 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max, 3344 boolean_t pageable, boolean_t remove_holes) 3345 { 3346 KASSERT(pmap == NULL || pmap == pmap_kernel()); 3347 3348 if (pmap) 3349 pmap_reference(pmap); 3350 else 3351 pmap = pmap_create(); 3352 vm->vm_map.pmap = pmap; 3353 3354 uvm_map_setup(&vm->vm_map, min, max, 3355 (pageable ? VM_MAP_PAGEABLE : 0) | VM_MAP_ISVMSPACE); 3356 3357 vm->vm_refcnt = 1; 3358 3359 if (remove_holes) 3360 pmap_remove_holes(vm); 3361 } 3362 3363 /* 3364 * uvmspace_share: share a vmspace between two processes 3365 * 3366 * - XXX: no locking on vmspace 3367 * - used for vfork 3368 */ 3369 3370 struct vmspace * 3371 uvmspace_share(struct process *pr) 3372 { 3373 struct vmspace *vm = pr->ps_vmspace; 3374 3375 vm->vm_refcnt++; 3376 return vm; 3377 } 3378 3379 /* 3380 * uvmspace_exec: the process wants to exec a new program 3381 * 3382 * - XXX: no locking on vmspace 3383 */ 3384 3385 void 3386 uvmspace_exec(struct proc *p, vaddr_t start, vaddr_t end) 3387 { 3388 struct process *pr = p->p_p; 3389 struct vmspace *nvm, *ovm = pr->ps_vmspace; 3390 struct vm_map *map = &ovm->vm_map; 3391 struct uvm_map_deadq dead_entries; 3392 3393 KASSERT((start & (vaddr_t)PAGE_MASK) == 0); 3394 KASSERT((end & (vaddr_t)PAGE_MASK) == 0 || 3395 (end & (vaddr_t)PAGE_MASK) == (vaddr_t)PAGE_MASK); 3396 3397 pmap_unuse_final(p); /* before stack addresses go away */ 3398 TAILQ_INIT(&dead_entries); 3399 3400 /* see if more than one process is using this vmspace... */ 3401 if (ovm->vm_refcnt == 1) { 3402 /* 3403 * If pr is the only process using its vmspace then 3404 * we can safely recycle that vmspace for the program 3405 * that is being exec'd. 3406 */ 3407 3408 #ifdef SYSVSHM 3409 /* 3410 * SYSV SHM semantics require us to kill all segments on an exec 3411 */ 3412 if (ovm->vm_shm) 3413 shmexit(ovm); 3414 #endif 3415 3416 /* 3417 * POSIX 1003.1b -- "lock future mappings" is revoked 3418 * when a process execs another program image. 3419 */ 3420 vm_map_lock(map); 3421 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 3422 3423 /* 3424 * now unmap the old program 3425 * 3426 * Instead of attempting to keep the map valid, we simply 3427 * nuke all entries and ask uvm_map_setup to reinitialize 3428 * the map to the new boundaries. 3429 * 3430 * uvm_unmap_remove will actually nuke all entries for us 3431 * (as in, not replace them with free-memory entries). 3432 */ 3433 uvm_unmap_remove(map, map->min_offset, map->max_offset, 3434 &dead_entries, TRUE, FALSE); 3435 3436 KDASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 3437 3438 /* Nuke statistics and boundaries. */ 3439 memset(&ovm->vm_startcopy, 0, 3440 (caddr_t) (ovm + 1) - (caddr_t) &ovm->vm_startcopy); 3441 3442 3443 if (end & (vaddr_t)PAGE_MASK) { 3444 end += 1; 3445 if (end == 0) /* overflow */ 3446 end -= PAGE_SIZE; 3447 } 3448 3449 /* Setup new boundaries and populate map with entries. */ 3450 map->min_offset = start; 3451 map->max_offset = end; 3452 uvm_map_setup_entries(map); 3453 vm_map_unlock(map); 3454 3455 /* but keep MMU holes unavailable */ 3456 pmap_remove_holes(ovm); 3457 } else { 3458 /* 3459 * pr's vmspace is being shared, so we can't reuse 3460 * it for pr since it is still being used for others. 3461 * allocate a new vmspace for pr 3462 */ 3463 nvm = uvmspace_alloc(start, end, 3464 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, TRUE); 3465 3466 /* install new vmspace and drop our ref to the old one. */ 3467 pmap_deactivate(p); 3468 p->p_vmspace = pr->ps_vmspace = nvm; 3469 pmap_activate(p); 3470 3471 uvmspace_free(ovm); 3472 } 3473 3474 /* Release dead entries */ 3475 uvm_unmap_detach(&dead_entries, 0); 3476 } 3477 3478 /* 3479 * uvmspace_free: free a vmspace data structure 3480 * 3481 * - XXX: no locking on vmspace 3482 */ 3483 void 3484 uvmspace_free(struct vmspace *vm) 3485 { 3486 if (--vm->vm_refcnt == 0) { 3487 /* 3488 * lock the map, to wait out all other references to it. delete 3489 * all of the mappings and pages they hold, then call the pmap 3490 * module to reclaim anything left. 3491 */ 3492 #ifdef SYSVSHM 3493 /* Get rid of any SYSV shared memory segments. */ 3494 if (vm->vm_shm != NULL) 3495 shmexit(vm); 3496 #endif 3497 3498 uvm_map_teardown(&vm->vm_map); 3499 pool_put(&uvm_vmspace_pool, vm); 3500 } 3501 } 3502 3503 /* 3504 * uvm_share: Map the address range [srcaddr, srcaddr + sz) in 3505 * srcmap to the address range [dstaddr, dstaddr + sz) in 3506 * dstmap. 3507 * 3508 * The whole address range in srcmap must be backed by an object 3509 * (no holes). 3510 * 3511 * If successful, the address ranges share memory and the destination 3512 * address range uses the protection flags in prot. 3513 * 3514 * This routine assumes that sz is a multiple of PAGE_SIZE and 3515 * that dstaddr and srcaddr are page-aligned. 3516 */ 3517 int 3518 uvm_share(struct vm_map *dstmap, vaddr_t dstaddr, vm_prot_t prot, 3519 struct vm_map *srcmap, vaddr_t srcaddr, vsize_t sz) 3520 { 3521 int ret = 0; 3522 vaddr_t unmap_end; 3523 vaddr_t dstva; 3524 vsize_t off, len, n = sz; 3525 struct vm_map_entry *first = NULL, *last = NULL; 3526 struct vm_map_entry *src_entry, *psrc_entry = NULL; 3527 struct uvm_map_deadq dead; 3528 3529 if (srcaddr >= srcmap->max_offset || sz > srcmap->max_offset - srcaddr) 3530 return EINVAL; 3531 3532 TAILQ_INIT(&dead); 3533 vm_map_lock(dstmap); 3534 vm_map_lock_read(srcmap); 3535 3536 if (!uvm_map_isavail(dstmap, NULL, &first, &last, dstaddr, sz)) { 3537 ret = ENOMEM; 3538 goto exit_unlock; 3539 } 3540 if (!uvm_map_lookup_entry(srcmap, srcaddr, &src_entry)) { 3541 ret = EINVAL; 3542 goto exit_unlock; 3543 } 3544 3545 unmap_end = dstaddr; 3546 for (; src_entry != NULL; 3547 psrc_entry = src_entry, 3548 src_entry = RBT_NEXT(uvm_map_addr, src_entry)) { 3549 /* hole in address space, bail out */ 3550 if (psrc_entry != NULL && psrc_entry->end != src_entry->start) 3551 break; 3552 if (src_entry->start >= srcaddr + sz) 3553 break; 3554 3555 if (UVM_ET_ISSUBMAP(src_entry)) 3556 panic("uvm_share: encountered a submap (illegal)"); 3557 if (!UVM_ET_ISCOPYONWRITE(src_entry) && 3558 UVM_ET_ISNEEDSCOPY(src_entry)) 3559 panic("uvm_share: non-copy_on_write map entries " 3560 "marked needs_copy (illegal)"); 3561 3562 dstva = dstaddr; 3563 if (src_entry->start > srcaddr) { 3564 dstva += src_entry->start - srcaddr; 3565 off = 0; 3566 } else 3567 off = srcaddr - src_entry->start; 3568 3569 if (n < src_entry->end - src_entry->start) 3570 len = n; 3571 else 3572 len = src_entry->end - src_entry->start; 3573 n -= len; 3574 3575 if (uvm_mapent_share(dstmap, dstva, len, off, prot, prot, 3576 srcmap, src_entry, &dead) == NULL) 3577 break; 3578 3579 unmap_end = dstva + len; 3580 if (n == 0) 3581 goto exit_unlock; 3582 } 3583 3584 ret = EINVAL; 3585 uvm_unmap_remove(dstmap, dstaddr, unmap_end, &dead, FALSE, TRUE); 3586 3587 exit_unlock: 3588 vm_map_unlock_read(srcmap); 3589 vm_map_unlock(dstmap); 3590 uvm_unmap_detach(&dead, 0); 3591 3592 return ret; 3593 } 3594 3595 /* 3596 * Clone map entry into other map. 3597 * 3598 * Mapping will be placed at dstaddr, for the same length. 3599 * Space must be available. 3600 * Reference counters are incremented. 3601 */ 3602 struct vm_map_entry * 3603 uvm_mapent_clone(struct vm_map *dstmap, vaddr_t dstaddr, vsize_t dstlen, 3604 vsize_t off, vm_prot_t prot, vm_prot_t maxprot, 3605 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead, 3606 int mapent_flags, int amap_share_flags) 3607 { 3608 struct vm_map_entry *new_entry, *first, *last; 3609 3610 KDASSERT(!UVM_ET_ISSUBMAP(old_entry)); 3611 3612 /* Create new entry (linked in on creation). Fill in first, last. */ 3613 first = last = NULL; 3614 if (!uvm_map_isavail(dstmap, NULL, &first, &last, dstaddr, dstlen)) { 3615 panic("uvmspace_fork: no space in map for " 3616 "entry in empty map"); 3617 } 3618 new_entry = uvm_map_mkentry(dstmap, first, last, 3619 dstaddr, dstlen, mapent_flags, dead, NULL); 3620 if (new_entry == NULL) 3621 return NULL; 3622 /* old_entry -> new_entry */ 3623 new_entry->object = old_entry->object; 3624 new_entry->offset = old_entry->offset; 3625 new_entry->aref = old_entry->aref; 3626 new_entry->etype |= old_entry->etype & ~UVM_ET_FREEMAPPED; 3627 new_entry->protection = prot; 3628 new_entry->max_protection = maxprot; 3629 new_entry->inheritance = old_entry->inheritance; 3630 new_entry->advice = old_entry->advice; 3631 3632 /* gain reference to object backing the map (can't be a submap). */ 3633 if (new_entry->aref.ar_amap) { 3634 new_entry->aref.ar_pageoff += off >> PAGE_SHIFT; 3635 amap_ref(new_entry->aref.ar_amap, new_entry->aref.ar_pageoff, 3636 (new_entry->end - new_entry->start) >> PAGE_SHIFT, 3637 amap_share_flags); 3638 } 3639 3640 if (UVM_ET_ISOBJ(new_entry) && 3641 new_entry->object.uvm_obj->pgops->pgo_reference) { 3642 new_entry->offset += off; 3643 new_entry->object.uvm_obj->pgops->pgo_reference 3644 (new_entry->object.uvm_obj); 3645 } 3646 3647 return new_entry; 3648 } 3649 3650 struct vm_map_entry * 3651 uvm_mapent_share(struct vm_map *dstmap, vaddr_t dstaddr, vsize_t dstlen, 3652 vsize_t off, vm_prot_t prot, vm_prot_t maxprot, struct vm_map *old_map, 3653 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3654 { 3655 /* 3656 * If old_entry refers to a copy-on-write region that has not yet been 3657 * written to (needs_copy flag is set), then we need to allocate a new 3658 * amap for old_entry. 3659 * 3660 * If we do not do this, and the process owning old_entry does a copy-on 3661 * write later, old_entry and new_entry will refer to different memory 3662 * regions, and the memory between the processes is no longer shared. 3663 * 3664 * [in other words, we need to clear needs_copy] 3665 */ 3666 3667 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3668 /* get our own amap, clears needs_copy */ 3669 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 3670 0, 0); 3671 /* XXXCDC: WAITOK??? */ 3672 } 3673 3674 return uvm_mapent_clone(dstmap, dstaddr, dstlen, off, 3675 prot, maxprot, old_entry, dead, 0, AMAP_SHARED); 3676 } 3677 3678 /* 3679 * share the mapping: this means we want the old and 3680 * new entries to share amaps and backing objects. 3681 */ 3682 struct vm_map_entry * 3683 uvm_mapent_forkshared(struct vmspace *new_vm, struct vm_map *new_map, 3684 struct vm_map *old_map, 3685 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3686 { 3687 struct vm_map_entry *new_entry; 3688 3689 new_entry = uvm_mapent_share(new_map, old_entry->start, 3690 old_entry->end - old_entry->start, 0, old_entry->protection, 3691 old_entry->max_protection, old_map, old_entry, dead); 3692 3693 /* 3694 * pmap_copy the mappings: this routine is optional 3695 * but if it is there it will reduce the number of 3696 * page faults in the new proc. 3697 */ 3698 if (!UVM_ET_ISHOLE(new_entry)) 3699 pmap_copy(new_map->pmap, old_map->pmap, new_entry->start, 3700 (new_entry->end - new_entry->start), new_entry->start); 3701 3702 return (new_entry); 3703 } 3704 3705 /* 3706 * copy-on-write the mapping (using mmap's 3707 * MAP_PRIVATE semantics) 3708 * 3709 * allocate new_entry, adjust reference counts. 3710 * (note that new references are read-only). 3711 */ 3712 struct vm_map_entry * 3713 uvm_mapent_forkcopy(struct vmspace *new_vm, struct vm_map *new_map, 3714 struct vm_map *old_map, 3715 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3716 { 3717 struct vm_map_entry *new_entry; 3718 boolean_t protect_child; 3719 3720 new_entry = uvm_mapent_clone(new_map, old_entry->start, 3721 old_entry->end - old_entry->start, 0, old_entry->protection, 3722 old_entry->max_protection, old_entry, dead, 0, 0); 3723 3724 new_entry->etype |= 3725 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3726 3727 /* 3728 * the new entry will need an amap. it will either 3729 * need to be copied from the old entry or created 3730 * from scratch (if the old entry does not have an 3731 * amap). can we defer this process until later 3732 * (by setting "needs_copy") or do we need to copy 3733 * the amap now? 3734 * 3735 * we must copy the amap now if any of the following 3736 * conditions hold: 3737 * 1. the old entry has an amap and that amap is 3738 * being shared. this means that the old (parent) 3739 * process is sharing the amap with another 3740 * process. if we do not clear needs_copy here 3741 * we will end up in a situation where both the 3742 * parent and child process are referring to the 3743 * same amap with "needs_copy" set. if the 3744 * parent write-faults, the fault routine will 3745 * clear "needs_copy" in the parent by allocating 3746 * a new amap. this is wrong because the 3747 * parent is supposed to be sharing the old amap 3748 * and the new amap will break that. 3749 * 3750 * 2. if the old entry has an amap and a non-zero 3751 * wire count then we are going to have to call 3752 * amap_cow_now to avoid page faults in the 3753 * parent process. since amap_cow_now requires 3754 * "needs_copy" to be clear we might as well 3755 * clear it here as well. 3756 * 3757 */ 3758 if (old_entry->aref.ar_amap != NULL && 3759 ((amap_flags(old_entry->aref.ar_amap) & 3760 AMAP_SHARED) != 0 || 3761 VM_MAPENT_ISWIRED(old_entry))) { 3762 amap_copy(new_map, new_entry, M_WAITOK, FALSE, 3763 0, 0); 3764 /* XXXCDC: M_WAITOK ... ok? */ 3765 } 3766 3767 /* 3768 * if the parent's entry is wired down, then the 3769 * parent process does not want page faults on 3770 * access to that memory. this means that we 3771 * cannot do copy-on-write because we can't write 3772 * protect the old entry. in this case we 3773 * resolve all copy-on-write faults now, using 3774 * amap_cow_now. note that we have already 3775 * allocated any needed amap (above). 3776 */ 3777 if (VM_MAPENT_ISWIRED(old_entry)) { 3778 /* 3779 * resolve all copy-on-write faults now 3780 * (note that there is nothing to do if 3781 * the old mapping does not have an amap). 3782 * XXX: is it worthwhile to bother with 3783 * pmap_copy in this case? 3784 */ 3785 if (old_entry->aref.ar_amap) 3786 amap_cow_now(new_map, new_entry); 3787 } else { 3788 if (old_entry->aref.ar_amap) { 3789 /* 3790 * setup mappings to trigger copy-on-write faults 3791 * we must write-protect the parent if it has 3792 * an amap and it is not already "needs_copy"... 3793 * if it is already "needs_copy" then the parent 3794 * has already been write-protected by a previous 3795 * fork operation. 3796 * 3797 * if we do not write-protect the parent, then 3798 * we must be sure to write-protect the child 3799 * after the pmap_copy() operation. 3800 * 3801 * XXX: pmap_copy should have some way of telling 3802 * us that it didn't do anything so we can avoid 3803 * calling pmap_protect needlessly. 3804 */ 3805 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3806 if (old_entry->max_protection & PROT_WRITE) { 3807 pmap_protect(old_map->pmap, 3808 old_entry->start, 3809 old_entry->end, 3810 old_entry->protection & 3811 ~PROT_WRITE); 3812 pmap_update(old_map->pmap); 3813 } 3814 old_entry->etype |= UVM_ET_NEEDSCOPY; 3815 } 3816 3817 /* parent must now be write-protected */ 3818 protect_child = FALSE; 3819 } else { 3820 /* 3821 * we only need to protect the child if the 3822 * parent has write access. 3823 */ 3824 if (old_entry->max_protection & PROT_WRITE) 3825 protect_child = TRUE; 3826 else 3827 protect_child = FALSE; 3828 } 3829 /* 3830 * copy the mappings 3831 * XXX: need a way to tell if this does anything 3832 */ 3833 if (!UVM_ET_ISHOLE(new_entry)) 3834 pmap_copy(new_map->pmap, old_map->pmap, 3835 new_entry->start, 3836 (old_entry->end - old_entry->start), 3837 old_entry->start); 3838 3839 /* protect the child's mappings if necessary */ 3840 if (protect_child) { 3841 pmap_protect(new_map->pmap, new_entry->start, 3842 new_entry->end, 3843 new_entry->protection & 3844 ~PROT_WRITE); 3845 } 3846 } 3847 3848 return (new_entry); 3849 } 3850 3851 /* 3852 * zero the mapping: the new entry will be zero initialized 3853 */ 3854 struct vm_map_entry * 3855 uvm_mapent_forkzero(struct vmspace *new_vm, struct vm_map *new_map, 3856 struct vm_map *old_map, 3857 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3858 { 3859 struct vm_map_entry *new_entry; 3860 3861 new_entry = uvm_mapent_clone(new_map, old_entry->start, 3862 old_entry->end - old_entry->start, 0, old_entry->protection, 3863 old_entry->max_protection, old_entry, dead, 0, 0); 3864 3865 new_entry->etype |= 3866 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3867 3868 if (new_entry->aref.ar_amap) { 3869 amap_unref(new_entry->aref.ar_amap, new_entry->aref.ar_pageoff, 3870 atop(new_entry->end - new_entry->start), 0); 3871 new_entry->aref.ar_amap = NULL; 3872 new_entry->aref.ar_pageoff = 0; 3873 } 3874 3875 if (UVM_ET_ISOBJ(new_entry)) { 3876 if (new_entry->object.uvm_obj->pgops->pgo_detach) 3877 new_entry->object.uvm_obj->pgops->pgo_detach( 3878 new_entry->object.uvm_obj); 3879 new_entry->object.uvm_obj = NULL; 3880 new_entry->etype &= ~UVM_ET_OBJ; 3881 } 3882 3883 return (new_entry); 3884 } 3885 3886 /* 3887 * uvmspace_fork: fork a process' main map 3888 * 3889 * => create a new vmspace for child process from parent. 3890 * => parent's map must not be locked. 3891 */ 3892 struct vmspace * 3893 uvmspace_fork(struct process *pr) 3894 { 3895 struct vmspace *vm1 = pr->ps_vmspace; 3896 struct vmspace *vm2; 3897 struct vm_map *old_map = &vm1->vm_map; 3898 struct vm_map *new_map; 3899 struct vm_map_entry *old_entry, *new_entry; 3900 struct uvm_map_deadq dead; 3901 3902 vm_map_lock(old_map); 3903 3904 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset, 3905 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, FALSE); 3906 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 3907 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 3908 vm2->vm_dused = 0; /* Statistic managed by us. */ 3909 new_map = &vm2->vm_map; 3910 vm_map_lock(new_map); 3911 3912 /* go entry-by-entry */ 3913 TAILQ_INIT(&dead); 3914 RBT_FOREACH(old_entry, uvm_map_addr, &old_map->addr) { 3915 if (old_entry->start == old_entry->end) 3916 continue; 3917 3918 /* first, some sanity checks on the old entry */ 3919 if (UVM_ET_ISSUBMAP(old_entry)) { 3920 panic("fork: encountered a submap during fork " 3921 "(illegal)"); 3922 } 3923 3924 if (!UVM_ET_ISCOPYONWRITE(old_entry) && 3925 UVM_ET_ISNEEDSCOPY(old_entry)) { 3926 panic("fork: non-copy_on_write map entry marked " 3927 "needs_copy (illegal)"); 3928 } 3929 3930 /* Apply inheritance. */ 3931 switch (old_entry->inheritance) { 3932 case MAP_INHERIT_SHARE: 3933 new_entry = uvm_mapent_forkshared(vm2, new_map, 3934 old_map, old_entry, &dead); 3935 break; 3936 case MAP_INHERIT_COPY: 3937 new_entry = uvm_mapent_forkcopy(vm2, new_map, 3938 old_map, old_entry, &dead); 3939 break; 3940 case MAP_INHERIT_ZERO: 3941 new_entry = uvm_mapent_forkzero(vm2, new_map, 3942 old_map, old_entry, &dead); 3943 break; 3944 default: 3945 continue; 3946 } 3947 3948 /* Update process statistics. */ 3949 if (!UVM_ET_ISHOLE(new_entry)) 3950 new_map->size += new_entry->end - new_entry->start; 3951 if (!UVM_ET_ISOBJ(new_entry) && !UVM_ET_ISHOLE(new_entry)) { 3952 vm2->vm_dused += uvmspace_dused( 3953 new_map, new_entry->start, new_entry->end); 3954 } 3955 } 3956 3957 vm_map_unlock(old_map); 3958 vm_map_unlock(new_map); 3959 3960 /* 3961 * This can actually happen, if multiple entries described a 3962 * space in which an entry was inherited. 3963 */ 3964 uvm_unmap_detach(&dead, 0); 3965 3966 #ifdef SYSVSHM 3967 if (vm1->vm_shm) 3968 shmfork(vm1, vm2); 3969 #endif 3970 3971 return vm2; 3972 } 3973 3974 /* 3975 * uvm_map_hint: return the beginning of the best area suitable for 3976 * creating a new mapping with "prot" protection. 3977 */ 3978 vaddr_t 3979 uvm_map_hint(struct vmspace *vm, vm_prot_t prot, vaddr_t minaddr, 3980 vaddr_t maxaddr) 3981 { 3982 vaddr_t addr; 3983 vaddr_t spacing; 3984 3985 #ifdef __i386__ 3986 /* 3987 * If executable skip first two pages, otherwise start 3988 * after data + heap region. 3989 */ 3990 if ((prot & PROT_EXEC) != 0 && 3991 (vaddr_t)vm->vm_daddr >= I386_MAX_EXE_ADDR) { 3992 addr = (PAGE_SIZE*2) + 3993 (arc4random() & (I386_MAX_EXE_ADDR / 2 - 1)); 3994 return (round_page(addr)); 3995 } 3996 #endif 3997 3998 #if defined (__LP64__) 3999 spacing = MIN(4UL * 1024 * 1024 * 1024, MAXDSIZ) - 1; 4000 #else 4001 spacing = MIN(1 * 1024 * 1024 * 1024, MAXDSIZ) - 1; 4002 #endif 4003 4004 /* 4005 * Start malloc/mmap after the brk. 4006 */ 4007 addr = (vaddr_t)vm->vm_daddr + BRKSIZ; 4008 addr = MAX(addr, minaddr); 4009 4010 if (addr < maxaddr) { 4011 while (spacing > maxaddr - addr) 4012 spacing >>= 1; 4013 } 4014 addr += arc4random() & spacing; 4015 return (round_page(addr)); 4016 } 4017 4018 /* 4019 * uvm_map_submap: punch down part of a map into a submap 4020 * 4021 * => only the kernel_map is allowed to be submapped 4022 * => the purpose of submapping is to break up the locking granularity 4023 * of a larger map 4024 * => the range specified must have been mapped previously with a uvm_map() 4025 * call [with uobj==NULL] to create a blank map entry in the main map. 4026 * [And it had better still be blank!] 4027 * => maps which contain submaps should never be copied or forked. 4028 * => to remove a submap, use uvm_unmap() on the main map 4029 * and then uvm_map_deallocate() the submap. 4030 * => main map must be unlocked. 4031 * => submap must have been init'd and have a zero reference count. 4032 * [need not be locked as we don't actually reference it] 4033 */ 4034 int 4035 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 4036 struct vm_map *submap) 4037 { 4038 struct vm_map_entry *entry; 4039 int result; 4040 4041 if (start > map->max_offset || end > map->max_offset || 4042 start < map->min_offset || end < map->min_offset) 4043 return EINVAL; 4044 4045 vm_map_lock(map); 4046 4047 if (uvm_map_lookup_entry(map, start, &entry)) { 4048 UVM_MAP_CLIP_START(map, entry, start); 4049 UVM_MAP_CLIP_END(map, entry, end); 4050 } else 4051 entry = NULL; 4052 4053 if (entry != NULL && 4054 entry->start == start && entry->end == end && 4055 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 4056 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 4057 entry->etype |= UVM_ET_SUBMAP; 4058 entry->object.sub_map = submap; 4059 entry->offset = 0; 4060 uvm_map_reference(submap); 4061 result = 0; 4062 } else 4063 result = EINVAL; 4064 4065 vm_map_unlock(map); 4066 return(result); 4067 } 4068 4069 /* 4070 * uvm_map_checkprot: check protection in map 4071 * 4072 * => must allow specific protection in a fully allocated region. 4073 * => map mut be read or write locked by caller. 4074 */ 4075 boolean_t 4076 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 4077 vm_prot_t protection) 4078 { 4079 struct vm_map_entry *entry; 4080 4081 if (start < map->min_offset || end > map->max_offset || start > end) 4082 return FALSE; 4083 if (start == end) 4084 return TRUE; 4085 4086 /* 4087 * Iterate entries. 4088 */ 4089 for (entry = uvm_map_entrybyaddr(&map->addr, start); 4090 entry != NULL && entry->start < end; 4091 entry = RBT_NEXT(uvm_map_addr, entry)) { 4092 /* Fail if a hole is found. */ 4093 if (UVM_ET_ISHOLE(entry) || 4094 (entry->end < end && entry->end != VMMAP_FREE_END(entry))) 4095 return FALSE; 4096 4097 /* Check protection. */ 4098 if ((entry->protection & protection) != protection) 4099 return FALSE; 4100 } 4101 return TRUE; 4102 } 4103 4104 /* 4105 * uvm_map_create: create map 4106 */ 4107 vm_map_t 4108 uvm_map_create(pmap_t pmap, vaddr_t min, vaddr_t max, int flags) 4109 { 4110 vm_map_t map; 4111 4112 map = malloc(sizeof *map, M_VMMAP, M_WAITOK); 4113 map->pmap = pmap; 4114 uvm_map_setup(map, min, max, flags); 4115 return (map); 4116 } 4117 4118 /* 4119 * uvm_map_deallocate: drop reference to a map 4120 * 4121 * => caller must not lock map 4122 * => we will zap map if ref count goes to zero 4123 */ 4124 void 4125 uvm_map_deallocate(vm_map_t map) 4126 { 4127 int c; 4128 struct uvm_map_deadq dead; 4129 4130 c = --map->ref_count; 4131 if (c > 0) { 4132 return; 4133 } 4134 4135 /* 4136 * all references gone. unmap and free. 4137 * 4138 * No lock required: we are only one to access this map. 4139 */ 4140 TAILQ_INIT(&dead); 4141 uvm_tree_sanity(map, __FILE__, __LINE__); 4142 uvm_unmap_remove(map, map->min_offset, map->max_offset, &dead, 4143 TRUE, FALSE); 4144 pmap_destroy(map->pmap); 4145 KASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 4146 free(map, M_VMMAP, sizeof *map); 4147 4148 uvm_unmap_detach(&dead, 0); 4149 } 4150 4151 /* 4152 * uvm_map_inherit: set inheritance code for range of addrs in map. 4153 * 4154 * => map must be unlocked 4155 * => note that the inherit code is used during a "fork". see fork 4156 * code for details. 4157 */ 4158 int 4159 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 4160 vm_inherit_t new_inheritance) 4161 { 4162 struct vm_map_entry *entry; 4163 4164 switch (new_inheritance) { 4165 case MAP_INHERIT_NONE: 4166 case MAP_INHERIT_COPY: 4167 case MAP_INHERIT_SHARE: 4168 case MAP_INHERIT_ZERO: 4169 break; 4170 default: 4171 return (EINVAL); 4172 } 4173 4174 if (start > end) 4175 return EINVAL; 4176 start = MAX(start, map->min_offset); 4177 end = MIN(end, map->max_offset); 4178 if (start >= end) 4179 return 0; 4180 4181 vm_map_lock(map); 4182 4183 entry = uvm_map_entrybyaddr(&map->addr, start); 4184 if (entry->end > start) 4185 UVM_MAP_CLIP_START(map, entry, start); 4186 else 4187 entry = RBT_NEXT(uvm_map_addr, entry); 4188 4189 while (entry != NULL && entry->start < end) { 4190 UVM_MAP_CLIP_END(map, entry, end); 4191 entry->inheritance = new_inheritance; 4192 entry = RBT_NEXT(uvm_map_addr, entry); 4193 } 4194 4195 vm_map_unlock(map); 4196 return (0); 4197 } 4198 4199 /* 4200 * uvm_map_advice: set advice code for range of addrs in map. 4201 * 4202 * => map must be unlocked 4203 */ 4204 int 4205 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 4206 { 4207 struct vm_map_entry *entry; 4208 4209 switch (new_advice) { 4210 case MADV_NORMAL: 4211 case MADV_RANDOM: 4212 case MADV_SEQUENTIAL: 4213 break; 4214 default: 4215 return (EINVAL); 4216 } 4217 4218 if (start > end) 4219 return EINVAL; 4220 start = MAX(start, map->min_offset); 4221 end = MIN(end, map->max_offset); 4222 if (start >= end) 4223 return 0; 4224 4225 vm_map_lock(map); 4226 4227 entry = uvm_map_entrybyaddr(&map->addr, start); 4228 if (entry != NULL && entry->end > start) 4229 UVM_MAP_CLIP_START(map, entry, start); 4230 else if (entry!= NULL) 4231 entry = RBT_NEXT(uvm_map_addr, entry); 4232 4233 /* 4234 * XXXJRT: disallow holes? 4235 */ 4236 while (entry != NULL && entry->start < end) { 4237 UVM_MAP_CLIP_END(map, entry, end); 4238 entry->advice = new_advice; 4239 entry = RBT_NEXT(uvm_map_addr, entry); 4240 } 4241 4242 vm_map_unlock(map); 4243 return (0); 4244 } 4245 4246 /* 4247 * uvm_map_extract: extract a mapping from a map and put it somewhere 4248 * in the kernel_map, setting protection to max_prot. 4249 * 4250 * => map should be unlocked (we will write lock it and kernel_map) 4251 * => returns 0 on success, error code otherwise 4252 * => start must be page aligned 4253 * => len must be page sized 4254 * => flags: 4255 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 4256 * Mappings are QREF's. 4257 */ 4258 int 4259 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 4260 vaddr_t *dstaddrp, int flags) 4261 { 4262 struct uvm_map_deadq dead; 4263 struct vm_map_entry *first, *entry, *newentry, *tmp1, *tmp2; 4264 vaddr_t dstaddr; 4265 vaddr_t end; 4266 vaddr_t cp_start; 4267 vsize_t cp_len, cp_off; 4268 int error; 4269 4270 TAILQ_INIT(&dead); 4271 end = start + len; 4272 4273 /* 4274 * Sanity check on the parameters. 4275 * Also, since the mapping may not contain gaps, error out if the 4276 * mapped area is not in source map. 4277 */ 4278 if ((start & (vaddr_t)PAGE_MASK) != 0 || 4279 (end & (vaddr_t)PAGE_MASK) != 0 || end < start) 4280 return EINVAL; 4281 if (start < srcmap->min_offset || end > srcmap->max_offset) 4282 return EINVAL; 4283 4284 /* Initialize dead entries. Handle len == 0 case. */ 4285 if (len == 0) 4286 return 0; 4287 4288 /* Acquire lock on srcmap. */ 4289 vm_map_lock(srcmap); 4290 4291 /* Lock srcmap, lookup first and last entry in <start,len>. */ 4292 first = uvm_map_entrybyaddr(&srcmap->addr, start); 4293 4294 /* Check that the range is contiguous. */ 4295 for (entry = first; entry != NULL && entry->end < end; 4296 entry = RBT_NEXT(uvm_map_addr, entry)) { 4297 if (VMMAP_FREE_END(entry) != entry->end || 4298 UVM_ET_ISHOLE(entry)) { 4299 error = EINVAL; 4300 goto fail; 4301 } 4302 } 4303 if (entry == NULL || UVM_ET_ISHOLE(entry)) { 4304 error = EINVAL; 4305 goto fail; 4306 } 4307 4308 /* 4309 * Handle need-copy flag. 4310 */ 4311 for (entry = first; entry != NULL && entry->start < end; 4312 entry = RBT_NEXT(uvm_map_addr, entry)) { 4313 if (UVM_ET_ISNEEDSCOPY(entry)) 4314 amap_copy(srcmap, entry, M_NOWAIT, 4315 UVM_ET_ISSTACK(entry) ? FALSE : TRUE, start, end); 4316 if (UVM_ET_ISNEEDSCOPY(entry)) { 4317 /* 4318 * amap_copy failure 4319 */ 4320 error = ENOMEM; 4321 goto fail; 4322 } 4323 } 4324 4325 /* Lock destination map (kernel_map). */ 4326 vm_map_lock(kernel_map); 4327 4328 if (uvm_map_findspace(kernel_map, &tmp1, &tmp2, &dstaddr, len, 4329 MAX(PAGE_SIZE, PMAP_PREFER_ALIGN()), PMAP_PREFER_OFFSET(start), 4330 PROT_NONE, 0) != 0) { 4331 error = ENOMEM; 4332 goto fail2; 4333 } 4334 *dstaddrp = dstaddr; 4335 4336 /* 4337 * We now have srcmap and kernel_map locked. 4338 * dstaddr contains the destination offset in dstmap. 4339 */ 4340 /* step 1: start looping through map entries, performing extraction. */ 4341 for (entry = first; entry != NULL && entry->start < end; 4342 entry = RBT_NEXT(uvm_map_addr, entry)) { 4343 KDASSERT(!UVM_ET_ISNEEDSCOPY(entry)); 4344 if (UVM_ET_ISHOLE(entry)) 4345 continue; 4346 4347 /* Calculate uvm_mapent_clone parameters. */ 4348 cp_start = entry->start; 4349 if (cp_start < start) { 4350 cp_off = start - cp_start; 4351 cp_start = start; 4352 } else 4353 cp_off = 0; 4354 cp_len = MIN(entry->end, end) - cp_start; 4355 4356 newentry = uvm_mapent_clone(kernel_map, 4357 cp_start - start + dstaddr, cp_len, cp_off, 4358 entry->protection, entry->max_protection, 4359 entry, &dead, flags, AMAP_SHARED | AMAP_REFALL); 4360 if (newentry == NULL) { 4361 error = ENOMEM; 4362 goto fail2_unmap; 4363 } 4364 kernel_map->size += cp_len; 4365 if (flags & UVM_EXTRACT_FIXPROT) 4366 newentry->protection = newentry->max_protection; 4367 4368 /* 4369 * Step 2: perform pmap copy. 4370 * (Doing this in the loop saves one RB traversal.) 4371 */ 4372 pmap_copy(kernel_map->pmap, srcmap->pmap, 4373 cp_start - start + dstaddr, cp_len, cp_start); 4374 } 4375 pmap_update(kernel_map->pmap); 4376 4377 error = 0; 4378 4379 /* Unmap copied entries on failure. */ 4380 fail2_unmap: 4381 if (error) { 4382 uvm_unmap_remove(kernel_map, dstaddr, dstaddr + len, &dead, 4383 FALSE, TRUE); 4384 } 4385 4386 /* Release maps, release dead entries. */ 4387 fail2: 4388 vm_map_unlock(kernel_map); 4389 4390 fail: 4391 vm_map_unlock(srcmap); 4392 4393 uvm_unmap_detach(&dead, 0); 4394 4395 return error; 4396 } 4397 4398 /* 4399 * uvm_map_clean: clean out a map range 4400 * 4401 * => valid flags: 4402 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 4403 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 4404 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 4405 * if (flags & PGO_FREE): any cached pages are freed after clean 4406 * => returns an error if any part of the specified range isn't mapped 4407 * => never a need to flush amap layer since the anonymous memory has 4408 * no permanent home, but may deactivate pages there 4409 * => called from sys_msync() and sys_madvise() 4410 * => caller must not write-lock map (read OK). 4411 * => we may sleep while cleaning if SYNCIO [with map read-locked] 4412 */ 4413 4414 int 4415 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 4416 { 4417 struct vm_map_entry *first, *entry; 4418 struct vm_amap *amap; 4419 struct vm_anon *anon; 4420 struct vm_page *pg; 4421 struct uvm_object *uobj; 4422 vaddr_t cp_start, cp_end; 4423 int refs; 4424 int error; 4425 boolean_t rv; 4426 4427 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 4428 (PGO_FREE|PGO_DEACTIVATE)); 4429 4430 if (start > end || start < map->min_offset || end > map->max_offset) 4431 return EINVAL; 4432 4433 vm_map_lock_read(map); 4434 first = uvm_map_entrybyaddr(&map->addr, start); 4435 4436 /* Make a first pass to check for holes. */ 4437 for (entry = first; entry != NULL && entry->start < end; 4438 entry = RBT_NEXT(uvm_map_addr, entry)) { 4439 if (UVM_ET_ISSUBMAP(entry)) { 4440 vm_map_unlock_read(map); 4441 return EINVAL; 4442 } 4443 if (UVM_ET_ISSUBMAP(entry) || 4444 UVM_ET_ISHOLE(entry) || 4445 (entry->end < end && 4446 VMMAP_FREE_END(entry) != entry->end)) { 4447 vm_map_unlock_read(map); 4448 return EFAULT; 4449 } 4450 } 4451 4452 error = 0; 4453 for (entry = first; entry != NULL && entry->start < end; 4454 entry = RBT_NEXT(uvm_map_addr, entry)) { 4455 amap = entry->aref.ar_amap; /* top layer */ 4456 if (UVM_ET_ISOBJ(entry)) 4457 uobj = entry->object.uvm_obj; 4458 else 4459 uobj = NULL; 4460 4461 /* 4462 * No amap cleaning necessary if: 4463 * - there's no amap 4464 * - we're not deactivating or freeing pages. 4465 */ 4466 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 4467 goto flush_object; 4468 4469 cp_start = MAX(entry->start, start); 4470 cp_end = MIN(entry->end, end); 4471 4472 for (; cp_start != cp_end; cp_start += PAGE_SIZE) { 4473 anon = amap_lookup(&entry->aref, 4474 cp_start - entry->start); 4475 if (anon == NULL) 4476 continue; 4477 4478 pg = anon->an_page; 4479 if (pg == NULL) { 4480 continue; 4481 } 4482 KASSERT(pg->pg_flags & PQ_ANON); 4483 4484 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 4485 /* 4486 * XXX In these first 3 cases, we always just 4487 * XXX deactivate the page. We may want to 4488 * XXX handle the different cases more 4489 * XXX specifically, in the future. 4490 */ 4491 case PGO_CLEANIT|PGO_FREE: 4492 case PGO_CLEANIT|PGO_DEACTIVATE: 4493 case PGO_DEACTIVATE: 4494 deactivate_it: 4495 /* skip the page if it's wired */ 4496 if (pg->wire_count != 0) 4497 break; 4498 4499 uvm_lock_pageq(); 4500 4501 KASSERT(pg->uanon == anon); 4502 4503 /* zap all mappings for the page. */ 4504 pmap_page_protect(pg, PROT_NONE); 4505 4506 /* ...and deactivate the page. */ 4507 uvm_pagedeactivate(pg); 4508 4509 uvm_unlock_pageq(); 4510 break; 4511 case PGO_FREE: 4512 /* 4513 * If there are multiple references to 4514 * the amap, just deactivate the page. 4515 */ 4516 if (amap_refs(amap) > 1) 4517 goto deactivate_it; 4518 4519 /* XXX skip the page if it's wired */ 4520 if (pg->wire_count != 0) { 4521 break; 4522 } 4523 amap_unadd(&entry->aref, 4524 cp_start - entry->start); 4525 refs = --anon->an_ref; 4526 if (refs == 0) 4527 uvm_anfree(anon); 4528 break; 4529 default: 4530 panic("uvm_map_clean: weird flags"); 4531 } 4532 } 4533 4534 flush_object: 4535 cp_start = MAX(entry->start, start); 4536 cp_end = MIN(entry->end, end); 4537 4538 /* 4539 * flush pages if we've got a valid backing object. 4540 * 4541 * Don't PGO_FREE if we don't have write permission 4542 * and don't flush if this is a copy-on-write object 4543 * since we can't know our permissions on it. 4544 */ 4545 if (uobj != NULL && 4546 ((flags & PGO_FREE) == 0 || 4547 ((entry->max_protection & PROT_WRITE) != 0 && 4548 (entry->etype & UVM_ET_COPYONWRITE) == 0))) { 4549 rv = uobj->pgops->pgo_flush(uobj, 4550 cp_start - entry->start + entry->offset, 4551 cp_end - entry->start + entry->offset, flags); 4552 4553 if (rv == FALSE) 4554 error = EFAULT; 4555 } 4556 } 4557 4558 vm_map_unlock_read(map); 4559 return error; 4560 } 4561 4562 /* 4563 * UVM_MAP_CLIP_END implementation 4564 */ 4565 void 4566 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t addr) 4567 { 4568 struct vm_map_entry *tmp; 4569 4570 KASSERT(entry->start < addr && VMMAP_FREE_END(entry) > addr); 4571 tmp = uvm_mapent_alloc(map, 0); 4572 4573 /* Invoke splitentry. */ 4574 uvm_map_splitentry(map, entry, tmp, addr); 4575 } 4576 4577 /* 4578 * UVM_MAP_CLIP_START implementation 4579 * 4580 * Clippers are required to not change the pointers to the entry they are 4581 * clipping on. 4582 * Since uvm_map_splitentry turns the original entry into the lowest 4583 * entry (address wise) we do a swap between the new entry and the original 4584 * entry, prior to calling uvm_map_splitentry. 4585 */ 4586 void 4587 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, vaddr_t addr) 4588 { 4589 struct vm_map_entry *tmp; 4590 struct uvm_addr_state *free; 4591 4592 /* Unlink original. */ 4593 free = uvm_map_uaddr_e(map, entry); 4594 uvm_mapent_free_remove(map, free, entry); 4595 uvm_mapent_addr_remove(map, entry); 4596 4597 /* Copy entry. */ 4598 KASSERT(entry->start < addr && VMMAP_FREE_END(entry) > addr); 4599 tmp = uvm_mapent_alloc(map, 0); 4600 uvm_mapent_copy(entry, tmp); 4601 4602 /* Put new entry in place of original entry. */ 4603 uvm_mapent_addr_insert(map, tmp); 4604 uvm_mapent_free_insert(map, free, tmp); 4605 4606 /* Invoke splitentry. */ 4607 uvm_map_splitentry(map, tmp, entry, addr); 4608 } 4609 4610 /* 4611 * Boundary fixer. 4612 */ 4613 static __inline vaddr_t uvm_map_boundfix(vaddr_t, vaddr_t, vaddr_t); 4614 static __inline vaddr_t 4615 uvm_map_boundfix(vaddr_t min, vaddr_t max, vaddr_t bound) 4616 { 4617 return (min < bound && max > bound) ? bound : max; 4618 } 4619 4620 /* 4621 * Choose free list based on address at start of free space. 4622 * 4623 * The uvm_addr_state returned contains addr and is the first of: 4624 * - uaddr_exe 4625 * - uaddr_brk_stack 4626 * - uaddr_any 4627 */ 4628 struct uvm_addr_state* 4629 uvm_map_uaddr(struct vm_map *map, vaddr_t addr) 4630 { 4631 struct uvm_addr_state *uaddr; 4632 int i; 4633 4634 /* Special case the first page, to prevent mmap from returning 0. */ 4635 if (addr < VMMAP_MIN_ADDR) 4636 return NULL; 4637 4638 /* Upper bound for kernel maps at uvm_maxkaddr. */ 4639 if ((map->flags & VM_MAP_ISVMSPACE) == 0) { 4640 if (addr >= uvm_maxkaddr) 4641 return NULL; 4642 } 4643 4644 /* Is the address inside the exe-only map? */ 4645 if (map->uaddr_exe != NULL && addr >= map->uaddr_exe->uaddr_minaddr && 4646 addr < map->uaddr_exe->uaddr_maxaddr) 4647 return map->uaddr_exe; 4648 4649 /* Check if the space falls inside brk/stack area. */ 4650 if ((addr >= map->b_start && addr < map->b_end) || 4651 (addr >= map->s_start && addr < map->s_end)) { 4652 if (map->uaddr_brk_stack != NULL && 4653 addr >= map->uaddr_brk_stack->uaddr_minaddr && 4654 addr < map->uaddr_brk_stack->uaddr_maxaddr) { 4655 return map->uaddr_brk_stack; 4656 } else 4657 return NULL; 4658 } 4659 4660 /* 4661 * Check the other selectors. 4662 * 4663 * These selectors are only marked as the owner, if they have insert 4664 * functions. 4665 */ 4666 for (i = 0; i < nitems(map->uaddr_any); i++) { 4667 uaddr = map->uaddr_any[i]; 4668 if (uaddr == NULL) 4669 continue; 4670 if (uaddr->uaddr_functions->uaddr_free_insert == NULL) 4671 continue; 4672 4673 if (addr >= uaddr->uaddr_minaddr && 4674 addr < uaddr->uaddr_maxaddr) 4675 return uaddr; 4676 } 4677 4678 return NULL; 4679 } 4680 4681 /* 4682 * Choose free list based on address at start of free space. 4683 * 4684 * The uvm_addr_state returned contains addr and is the first of: 4685 * - uaddr_exe 4686 * - uaddr_brk_stack 4687 * - uaddr_any 4688 */ 4689 struct uvm_addr_state* 4690 uvm_map_uaddr_e(struct vm_map *map, struct vm_map_entry *entry) 4691 { 4692 return uvm_map_uaddr(map, VMMAP_FREE_START(entry)); 4693 } 4694 4695 /* 4696 * Returns the first free-memory boundary that is crossed by [min-max]. 4697 */ 4698 vsize_t 4699 uvm_map_boundary(struct vm_map *map, vaddr_t min, vaddr_t max) 4700 { 4701 struct uvm_addr_state *uaddr; 4702 int i; 4703 4704 /* Never return first page. */ 4705 max = uvm_map_boundfix(min, max, VMMAP_MIN_ADDR); 4706 4707 /* Treat the maxkaddr special, if the map is a kernel_map. */ 4708 if ((map->flags & VM_MAP_ISVMSPACE) == 0) 4709 max = uvm_map_boundfix(min, max, uvm_maxkaddr); 4710 4711 /* Check for exe-only boundaries. */ 4712 if (map->uaddr_exe != NULL) { 4713 max = uvm_map_boundfix(min, max, map->uaddr_exe->uaddr_minaddr); 4714 max = uvm_map_boundfix(min, max, map->uaddr_exe->uaddr_maxaddr); 4715 } 4716 4717 /* Check for exe-only boundaries. */ 4718 if (map->uaddr_brk_stack != NULL) { 4719 max = uvm_map_boundfix(min, max, 4720 map->uaddr_brk_stack->uaddr_minaddr); 4721 max = uvm_map_boundfix(min, max, 4722 map->uaddr_brk_stack->uaddr_maxaddr); 4723 } 4724 4725 /* Check other boundaries. */ 4726 for (i = 0; i < nitems(map->uaddr_any); i++) { 4727 uaddr = map->uaddr_any[i]; 4728 if (uaddr != NULL) { 4729 max = uvm_map_boundfix(min, max, uaddr->uaddr_minaddr); 4730 max = uvm_map_boundfix(min, max, uaddr->uaddr_maxaddr); 4731 } 4732 } 4733 4734 /* Boundaries at stack and brk() area. */ 4735 max = uvm_map_boundfix(min, max, map->s_start); 4736 max = uvm_map_boundfix(min, max, map->s_end); 4737 max = uvm_map_boundfix(min, max, map->b_start); 4738 max = uvm_map_boundfix(min, max, map->b_end); 4739 4740 return max; 4741 } 4742 4743 /* 4744 * Update map allocation start and end addresses from proc vmspace. 4745 */ 4746 void 4747 uvm_map_vmspace_update(struct vm_map *map, 4748 struct uvm_map_deadq *dead, int flags) 4749 { 4750 struct vmspace *vm; 4751 vaddr_t b_start, b_end, s_start, s_end; 4752 4753 KASSERT(map->flags & VM_MAP_ISVMSPACE); 4754 KASSERT(offsetof(struct vmspace, vm_map) == 0); 4755 4756 /* 4757 * Derive actual allocation boundaries from vmspace. 4758 */ 4759 vm = (struct vmspace *)map; 4760 b_start = (vaddr_t)vm->vm_daddr; 4761 b_end = b_start + BRKSIZ; 4762 s_start = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 4763 s_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 4764 #ifdef DIAGNOSTIC 4765 if ((b_start & (vaddr_t)PAGE_MASK) != 0 || 4766 (b_end & (vaddr_t)PAGE_MASK) != 0 || 4767 (s_start & (vaddr_t)PAGE_MASK) != 0 || 4768 (s_end & (vaddr_t)PAGE_MASK) != 0) { 4769 panic("uvm_map_vmspace_update: vmspace %p invalid bounds: " 4770 "b=0x%lx-0x%lx s=0x%lx-0x%lx", 4771 vm, b_start, b_end, s_start, s_end); 4772 } 4773 #endif 4774 4775 if (__predict_true(map->b_start == b_start && map->b_end == b_end && 4776 map->s_start == s_start && map->s_end == s_end)) 4777 return; 4778 4779 uvm_map_freelist_update(map, dead, b_start, b_end, 4780 s_start, s_end, flags); 4781 } 4782 4783 /* 4784 * Grow kernel memory. 4785 * 4786 * This function is only called for kernel maps when an allocation fails. 4787 * 4788 * If the map has a gap that is large enough to accommodate alloc_sz, this 4789 * function will make sure map->free will include it. 4790 */ 4791 void 4792 uvm_map_kmem_grow(struct vm_map *map, struct uvm_map_deadq *dead, 4793 vsize_t alloc_sz, int flags) 4794 { 4795 vsize_t sz; 4796 vaddr_t end; 4797 struct vm_map_entry *entry; 4798 4799 /* Kernel memory only. */ 4800 KASSERT((map->flags & VM_MAP_ISVMSPACE) == 0); 4801 /* Destroy free list. */ 4802 uvm_map_freelist_update_clear(map, dead); 4803 4804 /* Include the guard page in the hard minimum requirement of alloc_sz. */ 4805 if (map->flags & VM_MAP_GUARDPAGES) 4806 alloc_sz += PAGE_SIZE; 4807 4808 /* 4809 * Grow by ALLOCMUL * alloc_sz, but at least VM_MAP_KSIZE_DELTA. 4810 * 4811 * Don't handle the case where the multiplication overflows: 4812 * if that happens, the allocation is probably too big anyway. 4813 */ 4814 sz = MAX(VM_MAP_KSIZE_ALLOCMUL * alloc_sz, VM_MAP_KSIZE_DELTA); 4815 4816 /* 4817 * Walk forward until a gap large enough for alloc_sz shows up. 4818 * 4819 * We assume the kernel map has no boundaries. 4820 * uvm_maxkaddr may be zero. 4821 */ 4822 end = MAX(uvm_maxkaddr, map->min_offset); 4823 entry = uvm_map_entrybyaddr(&map->addr, end); 4824 while (entry && entry->fspace < alloc_sz) 4825 entry = RBT_NEXT(uvm_map_addr, entry); 4826 if (entry) { 4827 end = MAX(VMMAP_FREE_START(entry), end); 4828 end += MIN(sz, map->max_offset - end); 4829 } else 4830 end = map->max_offset; 4831 4832 /* Reserve pmap entries. */ 4833 #ifdef PMAP_GROWKERNEL 4834 uvm_maxkaddr = pmap_growkernel(end); 4835 #else 4836 uvm_maxkaddr = MAX(uvm_maxkaddr, end); 4837 #endif 4838 4839 /* Rebuild free list. */ 4840 uvm_map_freelist_update_refill(map, flags); 4841 } 4842 4843 /* 4844 * Freelist update subfunction: unlink all entries from freelists. 4845 */ 4846 void 4847 uvm_map_freelist_update_clear(struct vm_map *map, struct uvm_map_deadq *dead) 4848 { 4849 struct uvm_addr_state *free; 4850 struct vm_map_entry *entry, *prev, *next; 4851 4852 prev = NULL; 4853 for (entry = RBT_MIN(uvm_map_addr, &map->addr); entry != NULL; 4854 entry = next) { 4855 next = RBT_NEXT(uvm_map_addr, entry); 4856 4857 free = uvm_map_uaddr_e(map, entry); 4858 uvm_mapent_free_remove(map, free, entry); 4859 4860 if (prev != NULL && entry->start == entry->end) { 4861 prev->fspace += VMMAP_FREE_END(entry) - entry->end; 4862 uvm_mapent_addr_remove(map, entry); 4863 DEAD_ENTRY_PUSH(dead, entry); 4864 } else 4865 prev = entry; 4866 } 4867 } 4868 4869 /* 4870 * Freelist update subfunction: refill the freelists with entries. 4871 */ 4872 void 4873 uvm_map_freelist_update_refill(struct vm_map *map, int flags) 4874 { 4875 struct vm_map_entry *entry; 4876 vaddr_t min, max; 4877 4878 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 4879 min = VMMAP_FREE_START(entry); 4880 max = VMMAP_FREE_END(entry); 4881 entry->fspace = 0; 4882 4883 entry = uvm_map_fix_space(map, entry, min, max, flags); 4884 } 4885 4886 uvm_tree_sanity(map, __FILE__, __LINE__); 4887 } 4888 4889 /* 4890 * Change {a,b}_{start,end} allocation ranges and associated free lists. 4891 */ 4892 void 4893 uvm_map_freelist_update(struct vm_map *map, struct uvm_map_deadq *dead, 4894 vaddr_t b_start, vaddr_t b_end, vaddr_t s_start, vaddr_t s_end, int flags) 4895 { 4896 KDASSERT(b_end >= b_start && s_end >= s_start); 4897 4898 /* Clear all free lists. */ 4899 uvm_map_freelist_update_clear(map, dead); 4900 4901 /* Apply new bounds. */ 4902 map->b_start = b_start; 4903 map->b_end = b_end; 4904 map->s_start = s_start; 4905 map->s_end = s_end; 4906 4907 /* Refill free lists. */ 4908 uvm_map_freelist_update_refill(map, flags); 4909 } 4910 4911 /* 4912 * Assign a uvm_addr_state to the specified pointer in vm_map. 4913 * 4914 * May sleep. 4915 */ 4916 void 4917 uvm_map_set_uaddr(struct vm_map *map, struct uvm_addr_state **which, 4918 struct uvm_addr_state *newval) 4919 { 4920 struct uvm_map_deadq dead; 4921 4922 /* Pointer which must be in this map. */ 4923 KASSERT(which != NULL); 4924 KASSERT((void*)map <= (void*)(which) && 4925 (void*)(which) < (void*)(map + 1)); 4926 4927 vm_map_lock(map); 4928 TAILQ_INIT(&dead); 4929 uvm_map_freelist_update_clear(map, &dead); 4930 4931 uvm_addr_destroy(*which); 4932 *which = newval; 4933 4934 uvm_map_freelist_update_refill(map, 0); 4935 vm_map_unlock(map); 4936 uvm_unmap_detach(&dead, 0); 4937 } 4938 4939 /* 4940 * Correct space insert. 4941 * 4942 * Entry must not be on any freelist. 4943 */ 4944 struct vm_map_entry* 4945 uvm_map_fix_space(struct vm_map *map, struct vm_map_entry *entry, 4946 vaddr_t min, vaddr_t max, int flags) 4947 { 4948 struct uvm_addr_state *free, *entfree; 4949 vaddr_t lmax; 4950 4951 KASSERT(entry == NULL || (entry->etype & UVM_ET_FREEMAPPED) == 0); 4952 KDASSERT(min <= max); 4953 KDASSERT((entry != NULL && VMMAP_FREE_END(entry) == min) || 4954 min == map->min_offset); 4955 4956 /* 4957 * During the function, entfree will always point at the uaddr state 4958 * for entry. 4959 */ 4960 entfree = (entry == NULL ? NULL : 4961 uvm_map_uaddr_e(map, entry)); 4962 4963 while (min != max) { 4964 /* Claim guard page for entry. */ 4965 if ((map->flags & VM_MAP_GUARDPAGES) && entry != NULL && 4966 VMMAP_FREE_END(entry) == entry->end && 4967 entry->start != entry->end) { 4968 if (max - min == 2 * PAGE_SIZE) { 4969 /* 4970 * If the free-space gap is exactly 2 pages, 4971 * we make the guard 2 pages instead of 1. 4972 * Because in a guarded map, an area needs 4973 * at least 2 pages to allocate from: 4974 * one page for the allocation and one for 4975 * the guard. 4976 */ 4977 entry->guard = 2 * PAGE_SIZE; 4978 min = max; 4979 } else { 4980 entry->guard = PAGE_SIZE; 4981 min += PAGE_SIZE; 4982 } 4983 continue; 4984 } 4985 4986 /* 4987 * Handle the case where entry has a 2-page guard, but the 4988 * space after entry is freed. 4989 */ 4990 if (entry != NULL && entry->fspace == 0 && 4991 entry->guard > PAGE_SIZE) { 4992 entry->guard = PAGE_SIZE; 4993 min = VMMAP_FREE_START(entry); 4994 } 4995 4996 lmax = uvm_map_boundary(map, min, max); 4997 free = uvm_map_uaddr(map, min); 4998 4999 /* 5000 * Entries are merged if they point at the same uvm_free(). 5001 * Exception to that rule: if min == uvm_maxkaddr, a new 5002 * entry is started regardless (otherwise the allocators 5003 * will get confused). 5004 */ 5005 if (entry != NULL && free == entfree && 5006 !((map->flags & VM_MAP_ISVMSPACE) == 0 && 5007 min == uvm_maxkaddr)) { 5008 KDASSERT(VMMAP_FREE_END(entry) == min); 5009 entry->fspace += lmax - min; 5010 } else { 5011 /* 5012 * Commit entry to free list: it'll not be added to 5013 * anymore. 5014 * We'll start a new entry and add to that entry 5015 * instead. 5016 */ 5017 if (entry != NULL) 5018 uvm_mapent_free_insert(map, entfree, entry); 5019 5020 /* New entry for new uaddr. */ 5021 entry = uvm_mapent_alloc(map, flags); 5022 KDASSERT(entry != NULL); 5023 entry->end = entry->start = min; 5024 entry->guard = 0; 5025 entry->fspace = lmax - min; 5026 entry->object.uvm_obj = NULL; 5027 entry->offset = 0; 5028 entry->etype = 0; 5029 entry->protection = entry->max_protection = 0; 5030 entry->inheritance = 0; 5031 entry->wired_count = 0; 5032 entry->advice = 0; 5033 entry->aref.ar_pageoff = 0; 5034 entry->aref.ar_amap = NULL; 5035 uvm_mapent_addr_insert(map, entry); 5036 5037 entfree = free; 5038 } 5039 5040 min = lmax; 5041 } 5042 /* Finally put entry on the uaddr state. */ 5043 if (entry != NULL) 5044 uvm_mapent_free_insert(map, entfree, entry); 5045 5046 return entry; 5047 } 5048 5049 /* 5050 * MQuery style of allocation. 5051 * 5052 * This allocator searches forward until sufficient space is found to map 5053 * the given size. 5054 * 5055 * XXX: factor in offset (via pmap_prefer) and protection? 5056 */ 5057 int 5058 uvm_map_mquery(struct vm_map *map, vaddr_t *addr_p, vsize_t sz, voff_t offset, 5059 int flags) 5060 { 5061 struct vm_map_entry *entry, *last; 5062 vaddr_t addr; 5063 vaddr_t tmp, pmap_align, pmap_offset; 5064 int error; 5065 5066 addr = *addr_p; 5067 vm_map_lock_read(map); 5068 5069 /* Configure pmap prefer. */ 5070 if (offset != UVM_UNKNOWN_OFFSET) { 5071 pmap_align = MAX(PAGE_SIZE, PMAP_PREFER_ALIGN()); 5072 pmap_offset = PMAP_PREFER_OFFSET(offset); 5073 } else { 5074 pmap_align = PAGE_SIZE; 5075 pmap_offset = 0; 5076 } 5077 5078 /* Align address to pmap_prefer unless FLAG_FIXED is set. */ 5079 if (!(flags & UVM_FLAG_FIXED) && offset != UVM_UNKNOWN_OFFSET) { 5080 tmp = (addr & ~(pmap_align - 1)) | pmap_offset; 5081 if (tmp < addr) 5082 tmp += pmap_align; 5083 addr = tmp; 5084 } 5085 5086 /* First, check if the requested range is fully available. */ 5087 entry = uvm_map_entrybyaddr(&map->addr, addr); 5088 last = NULL; 5089 if (uvm_map_isavail(map, NULL, &entry, &last, addr, sz)) { 5090 error = 0; 5091 goto out; 5092 } 5093 if (flags & UVM_FLAG_FIXED) { 5094 error = EINVAL; 5095 goto out; 5096 } 5097 5098 error = ENOMEM; /* Default error from here. */ 5099 5100 /* 5101 * At this point, the memory at <addr, sz> is not available. 5102 * The reasons are: 5103 * [1] it's outside the map, 5104 * [2] it starts in used memory (and therefore needs to move 5105 * toward the first free page in entry), 5106 * [3] it starts in free memory but bumps into used memory. 5107 * 5108 * Note that for case [2], the forward moving is handled by the 5109 * for loop below. 5110 */ 5111 if (entry == NULL) { 5112 /* [1] Outside the map. */ 5113 if (addr >= map->max_offset) 5114 goto out; 5115 else 5116 entry = RBT_MIN(uvm_map_addr, &map->addr); 5117 } else if (VMMAP_FREE_START(entry) <= addr) { 5118 /* [3] Bumped into used memory. */ 5119 entry = RBT_NEXT(uvm_map_addr, entry); 5120 } 5121 5122 /* Test if the next entry is sufficient for the allocation. */ 5123 for (; entry != NULL; 5124 entry = RBT_NEXT(uvm_map_addr, entry)) { 5125 if (entry->fspace == 0) 5126 continue; 5127 addr = VMMAP_FREE_START(entry); 5128 5129 restart: /* Restart address checks on address change. */ 5130 tmp = (addr & ~(pmap_align - 1)) | pmap_offset; 5131 if (tmp < addr) 5132 tmp += pmap_align; 5133 addr = tmp; 5134 if (addr >= VMMAP_FREE_END(entry)) 5135 continue; 5136 5137 /* Skip brk() allocation addresses. */ 5138 if (addr + sz > map->b_start && addr < map->b_end) { 5139 if (VMMAP_FREE_END(entry) > map->b_end) { 5140 addr = map->b_end; 5141 goto restart; 5142 } else 5143 continue; 5144 } 5145 /* Skip stack allocation addresses. */ 5146 if (addr + sz > map->s_start && addr < map->s_end) { 5147 if (VMMAP_FREE_END(entry) > map->s_end) { 5148 addr = map->s_end; 5149 goto restart; 5150 } else 5151 continue; 5152 } 5153 5154 last = NULL; 5155 if (uvm_map_isavail(map, NULL, &entry, &last, addr, sz)) { 5156 error = 0; 5157 goto out; 5158 } 5159 } 5160 5161 out: 5162 vm_map_unlock_read(map); 5163 if (error == 0) 5164 *addr_p = addr; 5165 return error; 5166 } 5167 5168 /* 5169 * Determine allocation bias. 5170 * 5171 * Returns 1 if we should bias to high addresses, -1 for a bias towards low 5172 * addresses, or 0 for no bias. 5173 * The bias mechanism is intended to avoid clashing with brk() and stack 5174 * areas. 5175 */ 5176 int 5177 uvm_mapent_bias(struct vm_map *map, struct vm_map_entry *entry) 5178 { 5179 vaddr_t start, end; 5180 5181 start = VMMAP_FREE_START(entry); 5182 end = VMMAP_FREE_END(entry); 5183 5184 /* Stay at the top of brk() area. */ 5185 if (end >= map->b_start && start < map->b_end) 5186 return 1; 5187 /* Stay at the far end of the stack area. */ 5188 if (end >= map->s_start && start < map->s_end) { 5189 #ifdef MACHINE_STACK_GROWS_UP 5190 return 1; 5191 #else 5192 return -1; 5193 #endif 5194 } 5195 5196 /* No bias, this area is meant for us. */ 5197 return 0; 5198 } 5199 5200 5201 boolean_t 5202 vm_map_lock_try_ln(struct vm_map *map, char *file, int line) 5203 { 5204 boolean_t rv; 5205 5206 if (map->flags & VM_MAP_INTRSAFE) { 5207 rv = _mtx_enter_try(&map->mtx LOCK_FL_ARGS); 5208 } else { 5209 mtx_enter(&map->flags_lock); 5210 if (map->flags & VM_MAP_BUSY) { 5211 mtx_leave(&map->flags_lock); 5212 return (FALSE); 5213 } 5214 mtx_leave(&map->flags_lock); 5215 rv = (_rw_enter(&map->lock, RW_WRITE|RW_NOSLEEP LOCK_FL_ARGS) 5216 == 0); 5217 /* check if the lock is busy and back out if we won the race */ 5218 if (rv) { 5219 mtx_enter(&map->flags_lock); 5220 if (map->flags & VM_MAP_BUSY) { 5221 _rw_exit(&map->lock LOCK_FL_ARGS); 5222 rv = FALSE; 5223 } 5224 mtx_leave(&map->flags_lock); 5225 } 5226 } 5227 5228 if (rv) { 5229 map->timestamp++; 5230 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5231 uvm_tree_sanity(map, file, line); 5232 uvm_tree_size_chk(map, file, line); 5233 } 5234 5235 return (rv); 5236 } 5237 5238 void 5239 vm_map_lock_ln(struct vm_map *map, char *file, int line) 5240 { 5241 if ((map->flags & VM_MAP_INTRSAFE) == 0) { 5242 do { 5243 mtx_enter(&map->flags_lock); 5244 tryagain: 5245 while (map->flags & VM_MAP_BUSY) { 5246 map->flags |= VM_MAP_WANTLOCK; 5247 msleep(&map->flags, &map->flags_lock, 5248 PVM, vmmapbsy, 0); 5249 } 5250 mtx_leave(&map->flags_lock); 5251 } while (_rw_enter(&map->lock, RW_WRITE|RW_SLEEPFAIL 5252 LOCK_FL_ARGS) != 0); 5253 /* check if the lock is busy and back out if we won the race */ 5254 mtx_enter(&map->flags_lock); 5255 if (map->flags & VM_MAP_BUSY) { 5256 _rw_exit(&map->lock LOCK_FL_ARGS); 5257 goto tryagain; 5258 } 5259 mtx_leave(&map->flags_lock); 5260 } else { 5261 _mtx_enter(&map->mtx LOCK_FL_ARGS); 5262 } 5263 5264 map->timestamp++; 5265 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5266 uvm_tree_sanity(map, file, line); 5267 uvm_tree_size_chk(map, file, line); 5268 } 5269 5270 void 5271 vm_map_lock_read_ln(struct vm_map *map, char *file, int line) 5272 { 5273 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5274 _rw_enter_read(&map->lock LOCK_FL_ARGS); 5275 else 5276 _mtx_enter(&map->mtx LOCK_FL_ARGS); 5277 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5278 uvm_tree_sanity(map, file, line); 5279 uvm_tree_size_chk(map, file, line); 5280 } 5281 5282 void 5283 vm_map_unlock_ln(struct vm_map *map, char *file, int line) 5284 { 5285 uvm_tree_sanity(map, file, line); 5286 uvm_tree_size_chk(map, file, line); 5287 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5288 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5289 _rw_exit(&map->lock LOCK_FL_ARGS); 5290 else 5291 _mtx_leave(&map->mtx LOCK_FL_ARGS); 5292 } 5293 5294 void 5295 vm_map_unlock_read_ln(struct vm_map *map, char *file, int line) 5296 { 5297 /* XXX: RO */ uvm_tree_sanity(map, file, line); 5298 /* XXX: RO */ uvm_tree_size_chk(map, file, line); 5299 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5300 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5301 _rw_exit_read(&map->lock LOCK_FL_ARGS); 5302 else 5303 _mtx_leave(&map->mtx LOCK_FL_ARGS); 5304 } 5305 5306 void 5307 vm_map_downgrade_ln(struct vm_map *map, char *file, int line) 5308 { 5309 uvm_tree_sanity(map, file, line); 5310 uvm_tree_size_chk(map, file, line); 5311 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5312 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5313 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5314 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5315 _rw_enter(&map->lock, RW_DOWNGRADE LOCK_FL_ARGS); 5316 } 5317 5318 void 5319 vm_map_upgrade_ln(struct vm_map *map, char *file, int line) 5320 { 5321 /* XXX: RO */ uvm_tree_sanity(map, file, line); 5322 /* XXX: RO */ uvm_tree_size_chk(map, file, line); 5323 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5324 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5325 if ((map->flags & VM_MAP_INTRSAFE) == 0) { 5326 _rw_exit_read(&map->lock LOCK_FL_ARGS); 5327 _rw_enter_write(&map->lock LOCK_FL_ARGS); 5328 } 5329 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5330 uvm_tree_sanity(map, file, line); 5331 } 5332 5333 void 5334 vm_map_busy_ln(struct vm_map *map, char *file, int line) 5335 { 5336 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5337 mtx_enter(&map->flags_lock); 5338 map->flags |= VM_MAP_BUSY; 5339 mtx_leave(&map->flags_lock); 5340 } 5341 5342 void 5343 vm_map_unbusy_ln(struct vm_map *map, char *file, int line) 5344 { 5345 int oflags; 5346 5347 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5348 mtx_enter(&map->flags_lock); 5349 oflags = map->flags; 5350 map->flags &= ~(VM_MAP_BUSY|VM_MAP_WANTLOCK); 5351 mtx_leave(&map->flags_lock); 5352 if (oflags & VM_MAP_WANTLOCK) 5353 wakeup(&map->flags); 5354 } 5355 5356 #ifndef SMALL_KERNEL 5357 int 5358 uvm_map_fill_vmmap(struct vm_map *map, struct kinfo_vmentry *kve, 5359 size_t *lenp) 5360 { 5361 struct vm_map_entry *entry; 5362 vaddr_t start; 5363 int cnt, maxcnt, error = 0; 5364 5365 KASSERT(*lenp > 0); 5366 KASSERT((*lenp % sizeof(*kve)) == 0); 5367 cnt = 0; 5368 maxcnt = *lenp / sizeof(*kve); 5369 KASSERT(maxcnt > 0); 5370 5371 /* 5372 * Return only entries whose address is above the given base 5373 * address. This allows userland to iterate without knowing the 5374 * number of entries beforehand. 5375 */ 5376 start = (vaddr_t)kve[0].kve_start; 5377 5378 vm_map_lock(map); 5379 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 5380 if (cnt == maxcnt) { 5381 error = ENOMEM; 5382 break; 5383 } 5384 if (start != 0 && entry->start < start) 5385 continue; 5386 kve->kve_start = entry->start; 5387 kve->kve_end = entry->end; 5388 kve->kve_guard = entry->guard; 5389 kve->kve_fspace = entry->fspace; 5390 kve->kve_fspace_augment = entry->fspace_augment; 5391 kve->kve_offset = entry->offset; 5392 kve->kve_wired_count = entry->wired_count; 5393 kve->kve_etype = entry->etype; 5394 kve->kve_protection = entry->protection; 5395 kve->kve_max_protection = entry->max_protection; 5396 kve->kve_advice = entry->advice; 5397 kve->kve_inheritance = entry->inheritance; 5398 kve->kve_flags = entry->flags; 5399 kve++; 5400 cnt++; 5401 } 5402 vm_map_unlock(map); 5403 5404 KASSERT(cnt <= maxcnt); 5405 5406 *lenp = sizeof(*kve) * cnt; 5407 return error; 5408 } 5409 #endif 5410 5411 5412 RBT_GENERATE_AUGMENT(uvm_map_addr, vm_map_entry, daddrs.addr_entry, 5413 uvm_mapentry_addrcmp, uvm_map_addr_augment); 5414 5415 5416 /* 5417 * MD code: vmspace allocator setup. 5418 */ 5419 5420 #ifdef __i386__ 5421 void 5422 uvm_map_setup_md(struct vm_map *map) 5423 { 5424 vaddr_t min, max; 5425 5426 min = map->min_offset; 5427 max = map->max_offset; 5428 5429 /* 5430 * Ensure the selectors will not try to manage page 0; 5431 * it's too special. 5432 */ 5433 if (min < VMMAP_MIN_ADDR) 5434 min = VMMAP_MIN_ADDR; 5435 5436 #if 0 /* Cool stuff, not yet */ 5437 /* Executable code is special. */ 5438 map->uaddr_exe = uaddr_rnd_create(min, I386_MAX_EXE_ADDR); 5439 /* Place normal allocations beyond executable mappings. */ 5440 map->uaddr_any[3] = uaddr_pivot_create(2 * I386_MAX_EXE_ADDR, max); 5441 #else /* Crappy stuff, for now */ 5442 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5443 #endif 5444 5445 #ifndef SMALL_KERNEL 5446 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5447 #endif /* !SMALL_KERNEL */ 5448 } 5449 #elif __LP64__ 5450 void 5451 uvm_map_setup_md(struct vm_map *map) 5452 { 5453 vaddr_t min, max; 5454 5455 min = map->min_offset; 5456 max = map->max_offset; 5457 5458 /* 5459 * Ensure the selectors will not try to manage page 0; 5460 * it's too special. 5461 */ 5462 if (min < VMMAP_MIN_ADDR) 5463 min = VMMAP_MIN_ADDR; 5464 5465 #if 0 /* Cool stuff, not yet */ 5466 map->uaddr_any[3] = uaddr_pivot_create(MAX(min, 0x100000000ULL), max); 5467 #else /* Crappy stuff, for now */ 5468 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5469 #endif 5470 5471 #ifndef SMALL_KERNEL 5472 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5473 #endif /* !SMALL_KERNEL */ 5474 } 5475 #else /* non-i386, 32 bit */ 5476 void 5477 uvm_map_setup_md(struct vm_map *map) 5478 { 5479 vaddr_t min, max; 5480 5481 min = map->min_offset; 5482 max = map->max_offset; 5483 5484 /* 5485 * Ensure the selectors will not try to manage page 0; 5486 * it's too special. 5487 */ 5488 if (min < VMMAP_MIN_ADDR) 5489 min = VMMAP_MIN_ADDR; 5490 5491 #if 0 /* Cool stuff, not yet */ 5492 map->uaddr_any[3] = uaddr_pivot_create(min, max); 5493 #else /* Crappy stuff, for now */ 5494 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5495 #endif 5496 5497 #ifndef SMALL_KERNEL 5498 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5499 #endif /* !SMALL_KERNEL */ 5500 } 5501 #endif 5502