1 /* $OpenBSD: uvm_map.c,v 1.337 2024/12/27 12:04:40 mpi 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/acct.h> 90 #include <sys/mman.h> 91 #include <sys/proc.h> 92 #include <sys/malloc.h> 93 #include <sys/pool.h> 94 #include <sys/sysctl.h> 95 #include <sys/signalvar.h> 96 #include <sys/syslog.h> 97 #include <sys/user.h> 98 #include <sys/tracepoint.h> 99 100 #ifdef SYSVSHM 101 #include <sys/shm.h> 102 #endif 103 104 #include <uvm/uvm.h> 105 106 #ifdef DDB 107 #include <uvm/uvm_ddb.h> 108 #endif 109 110 #include <uvm/uvm_addr.h> 111 112 113 vsize_t uvmspace_dused(struct vm_map*, vaddr_t, vaddr_t); 114 int uvm_mapent_isjoinable(struct vm_map*, 115 struct vm_map_entry*, struct vm_map_entry*); 116 struct vm_map_entry *uvm_mapent_merge(struct vm_map*, struct vm_map_entry*, 117 struct vm_map_entry*, struct uvm_map_deadq*); 118 struct vm_map_entry *uvm_mapent_tryjoin(struct vm_map*, 119 struct vm_map_entry*, struct uvm_map_deadq*); 120 struct vm_map_entry *uvm_map_mkentry(struct vm_map*, struct vm_map_entry*, 121 struct vm_map_entry*, vaddr_t, vsize_t, int, 122 struct uvm_map_deadq*, struct vm_map_entry*); 123 struct vm_map_entry *uvm_mapent_alloc(struct vm_map*, int); 124 void uvm_mapent_free(struct vm_map_entry*); 125 void uvm_unmap_kill_entry(struct vm_map*, 126 struct vm_map_entry*); 127 void uvm_unmap_kill_entry_withlock(struct vm_map *, 128 struct vm_map_entry *, int); 129 void uvm_unmap_detach_intrsafe(struct uvm_map_deadq *); 130 void uvm_mapent_mkfree(struct vm_map*, 131 struct vm_map_entry*, struct vm_map_entry**, 132 struct uvm_map_deadq*, boolean_t); 133 void uvm_map_pageable_pgon(struct vm_map*, 134 struct vm_map_entry*, struct vm_map_entry*, 135 vaddr_t, vaddr_t); 136 int uvm_map_pageable_wire(struct vm_map*, 137 struct vm_map_entry*, struct vm_map_entry*, 138 vaddr_t, vaddr_t, int); 139 void uvm_map_setup_entries(struct vm_map*); 140 void uvm_map_setup_md(struct vm_map*); 141 void uvm_map_teardown(struct vm_map*); 142 void uvm_map_vmspace_update(struct vm_map*, 143 struct uvm_map_deadq*, int); 144 void uvm_map_kmem_grow(struct vm_map*, 145 struct uvm_map_deadq*, vsize_t, int); 146 void uvm_map_freelist_update_clear(struct vm_map*, 147 struct uvm_map_deadq*); 148 void uvm_map_freelist_update_refill(struct vm_map *, int); 149 void uvm_map_freelist_update(struct vm_map*, 150 struct uvm_map_deadq*, vaddr_t, vaddr_t, 151 vaddr_t, vaddr_t, int); 152 struct vm_map_entry *uvm_map_fix_space(struct vm_map*, struct vm_map_entry*, 153 vaddr_t, vaddr_t, int); 154 int uvm_map_findspace(struct vm_map*, 155 struct vm_map_entry**, struct vm_map_entry**, 156 vaddr_t*, vsize_t, vaddr_t, vaddr_t, vm_prot_t, 157 vaddr_t); 158 vsize_t uvm_map_addr_augment_get(struct vm_map_entry*); 159 void uvm_map_addr_augment(struct vm_map_entry*); 160 161 int uvm_map_inentry_recheck(u_long, vaddr_t, 162 struct p_inentry *); 163 boolean_t uvm_map_inentry_fix(struct proc *, struct p_inentry *, 164 vaddr_t, int (*)(vm_map_entry_t), u_long); 165 /* 166 * Tree management functions. 167 */ 168 169 static inline void uvm_mapent_copy(struct vm_map_entry*, 170 struct vm_map_entry*); 171 static inline int uvm_mapentry_addrcmp(const struct vm_map_entry*, 172 const struct vm_map_entry*); 173 void uvm_mapent_free_insert(struct vm_map*, 174 struct uvm_addr_state*, struct vm_map_entry*); 175 void uvm_mapent_free_remove(struct vm_map*, 176 struct uvm_addr_state*, struct vm_map_entry*); 177 void uvm_mapent_addr_insert(struct vm_map*, 178 struct vm_map_entry*); 179 void uvm_mapent_addr_remove(struct vm_map*, 180 struct vm_map_entry*); 181 void uvm_map_splitentry(struct vm_map*, 182 struct vm_map_entry*, struct vm_map_entry*, 183 vaddr_t); 184 vsize_t uvm_map_boundary(struct vm_map*, vaddr_t, vaddr_t); 185 186 /* 187 * uvm_vmspace_fork helper functions. 188 */ 189 struct vm_map_entry *uvm_mapent_clone(struct vm_map*, vaddr_t, vsize_t, 190 vsize_t, vm_prot_t, vm_prot_t, 191 struct vm_map_entry*, struct uvm_map_deadq*, int, 192 int); 193 struct vm_map_entry *uvm_mapent_share(struct vm_map*, vaddr_t, vsize_t, 194 vsize_t, vm_prot_t, vm_prot_t, struct vm_map*, 195 struct vm_map_entry*, struct uvm_map_deadq*); 196 struct vm_map_entry *uvm_mapent_forkshared(struct vmspace*, struct vm_map*, 197 struct vm_map*, struct vm_map_entry*, 198 struct uvm_map_deadq*); 199 struct vm_map_entry *uvm_mapent_forkcopy(struct vmspace*, struct vm_map*, 200 struct vm_map*, struct vm_map_entry*, 201 struct uvm_map_deadq*); 202 struct vm_map_entry *uvm_mapent_forkzero(struct vmspace*, struct vm_map*, 203 struct vm_map*, struct vm_map_entry*, 204 struct uvm_map_deadq*); 205 206 /* 207 * Tree validation. 208 */ 209 #ifdef VMMAP_DEBUG 210 void uvm_tree_assert(struct vm_map*, int, char*, 211 char*, int); 212 #define UVM_ASSERT(map, cond, file, line) \ 213 uvm_tree_assert((map), (cond), #cond, (file), (line)) 214 void uvm_tree_sanity(struct vm_map*, char*, int); 215 void uvm_tree_size_chk(struct vm_map*, char*, int); 216 void vmspace_validate(struct vm_map*); 217 #else 218 #define uvm_tree_sanity(_map, _file, _line) do {} while (0) 219 #define uvm_tree_size_chk(_map, _file, _line) do {} while (0) 220 #define vmspace_validate(_map) do {} while (0) 221 #endif 222 223 /* 224 * The kernel map will initially be VM_MAP_KSIZE_INIT bytes. 225 * Every time that gets cramped, we grow by at least VM_MAP_KSIZE_DELTA bytes. 226 * 227 * We attempt to grow by UVM_MAP_KSIZE_ALLOCMUL times the allocation size 228 * each time. 229 */ 230 #define VM_MAP_KSIZE_INIT (512 * (vaddr_t)PAGE_SIZE) 231 #define VM_MAP_KSIZE_DELTA (256 * (vaddr_t)PAGE_SIZE) 232 #define VM_MAP_KSIZE_ALLOCMUL 4 233 234 /* auto-allocate address lower bound */ 235 #define VMMAP_MIN_ADDR PAGE_SIZE 236 237 238 #ifdef DEADBEEF0 239 #define UVMMAP_DEADBEEF ((unsigned long)DEADBEEF0) 240 #else 241 #define UVMMAP_DEADBEEF ((unsigned long)0xdeadd0d0) 242 #endif 243 244 #ifdef DEBUG 245 int uvm_map_printlocks = 0; 246 247 #define LPRINTF(_args) \ 248 do { \ 249 if (uvm_map_printlocks) \ 250 printf _args; \ 251 } while (0) 252 #else 253 #define LPRINTF(_args) do {} while (0) 254 #endif 255 256 static struct mutex uvm_kmapent_mtx; 257 static struct timeval uvm_kmapent_last_warn_time; 258 static struct timeval uvm_kmapent_warn_rate = { 10, 0 }; 259 260 const char vmmapbsy[] = "vmmapbsy"; 261 262 /* 263 * pool for vmspace structures. 264 */ 265 struct pool uvm_vmspace_pool; 266 267 /* 268 * pool for dynamically-allocated map entries. 269 */ 270 struct pool uvm_map_entry_pool; 271 struct pool uvm_map_entry_kmem_pool; 272 273 /* 274 * This global represents the end of the kernel virtual address 275 * space. If we want to exceed this, we must grow the kernel 276 * virtual address space dynamically. 277 * 278 * Note, this variable is locked by kernel_map's lock. 279 */ 280 vaddr_t uvm_maxkaddr; 281 282 /* 283 * Locking predicate. 284 */ 285 #define UVM_MAP_REQ_WRITE(_map) \ 286 do { \ 287 if ((_map)->ref_count > 0) { \ 288 if (((_map)->flags & VM_MAP_INTRSAFE) == 0) \ 289 rw_assert_wrlock(&(_map)->lock); \ 290 else \ 291 MUTEX_ASSERT_LOCKED(&(_map)->mtx); \ 292 } \ 293 } while (0) 294 295 #define vm_map_modflags(map, set, clear) \ 296 do { \ 297 mtx_enter(&(map)->flags_lock); \ 298 (map)->flags = ((map)->flags | (set)) & ~(clear); \ 299 mtx_leave(&(map)->flags_lock); \ 300 } while (0) 301 302 303 /* 304 * Tree describing entries by address. 305 * 306 * Addresses are unique. 307 * Entries with start == end may only exist if they are the first entry 308 * (sorted by address) within a free-memory tree. 309 */ 310 311 static inline int 312 uvm_mapentry_addrcmp(const struct vm_map_entry *e1, 313 const struct vm_map_entry *e2) 314 { 315 return e1->start < e2->start ? -1 : e1->start > e2->start; 316 } 317 318 /* 319 * Copy mapentry. 320 */ 321 static inline void 322 uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst) 323 { 324 caddr_t csrc, cdst; 325 size_t sz; 326 327 csrc = (caddr_t)src; 328 cdst = (caddr_t)dst; 329 csrc += offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 330 cdst += offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 331 332 sz = offsetof(struct vm_map_entry, uvm_map_entry_stop_copy) - 333 offsetof(struct vm_map_entry, uvm_map_entry_start_copy); 334 memcpy(cdst, csrc, sz); 335 } 336 337 /* 338 * Handle free-list insertion. 339 */ 340 void 341 uvm_mapent_free_insert(struct vm_map *map, struct uvm_addr_state *uaddr, 342 struct vm_map_entry *entry) 343 { 344 const struct uvm_addr_functions *fun; 345 #ifdef VMMAP_DEBUG 346 vaddr_t min, max, bound; 347 #endif 348 349 #ifdef VMMAP_DEBUG 350 /* 351 * Boundary check. 352 * Boundaries are folded if they go on the same free list. 353 */ 354 min = VMMAP_FREE_START(entry); 355 max = VMMAP_FREE_END(entry); 356 357 while (min < max) { 358 bound = uvm_map_boundary(map, min, max); 359 KASSERT(uvm_map_uaddr(map, min) == uaddr); 360 min = bound; 361 } 362 #endif 363 KDASSERT((entry->fspace & (vaddr_t)PAGE_MASK) == 0); 364 KASSERT((entry->etype & UVM_ET_FREEMAPPED) == 0); 365 366 UVM_MAP_REQ_WRITE(map); 367 368 /* Actual insert: forward to uaddr pointer. */ 369 if (uaddr != NULL) { 370 fun = uaddr->uaddr_functions; 371 KDASSERT(fun != NULL); 372 if (fun->uaddr_free_insert != NULL) 373 (*fun->uaddr_free_insert)(map, uaddr, entry); 374 entry->etype |= UVM_ET_FREEMAPPED; 375 } 376 377 /* Update fspace augmentation. */ 378 uvm_map_addr_augment(entry); 379 } 380 381 /* 382 * Handle free-list removal. 383 */ 384 void 385 uvm_mapent_free_remove(struct vm_map *map, struct uvm_addr_state *uaddr, 386 struct vm_map_entry *entry) 387 { 388 const struct uvm_addr_functions *fun; 389 390 KASSERT((entry->etype & UVM_ET_FREEMAPPED) != 0 || uaddr == NULL); 391 KASSERT(uvm_map_uaddr_e(map, entry) == uaddr); 392 UVM_MAP_REQ_WRITE(map); 393 394 if (uaddr != NULL) { 395 fun = uaddr->uaddr_functions; 396 if (fun->uaddr_free_remove != NULL) 397 (*fun->uaddr_free_remove)(map, uaddr, entry); 398 entry->etype &= ~UVM_ET_FREEMAPPED; 399 } 400 } 401 402 /* 403 * Handle address tree insertion. 404 */ 405 void 406 uvm_mapent_addr_insert(struct vm_map *map, struct vm_map_entry *entry) 407 { 408 struct vm_map_entry *res; 409 410 if (!RBT_CHECK(uvm_map_addr, entry, UVMMAP_DEADBEEF)) 411 panic("uvm_mapent_addr_insert: entry still in addr list"); 412 KDASSERT(entry->start <= entry->end); 413 KDASSERT((entry->start & (vaddr_t)PAGE_MASK) == 0 && 414 (entry->end & (vaddr_t)PAGE_MASK) == 0); 415 416 TRACEPOINT(uvm, map_insert, 417 entry->start, entry->end, entry->protection, NULL); 418 419 UVM_MAP_REQ_WRITE(map); 420 res = RBT_INSERT(uvm_map_addr, &map->addr, entry); 421 if (res != NULL) { 422 panic("uvm_mapent_addr_insert: map %p entry %p " 423 "(0x%lx-0x%lx G=0x%lx F=0x%lx) insert collision " 424 "with entry %p (0x%lx-0x%lx G=0x%lx F=0x%lx)", 425 map, entry, 426 entry->start, entry->end, entry->guard, entry->fspace, 427 res, res->start, res->end, res->guard, res->fspace); 428 } 429 } 430 431 /* 432 * Handle address tree removal. 433 */ 434 void 435 uvm_mapent_addr_remove(struct vm_map *map, struct vm_map_entry *entry) 436 { 437 struct vm_map_entry *res; 438 439 TRACEPOINT(uvm, map_remove, 440 entry->start, entry->end, entry->protection, NULL); 441 442 UVM_MAP_REQ_WRITE(map); 443 res = RBT_REMOVE(uvm_map_addr, &map->addr, entry); 444 if (res != entry) 445 panic("uvm_mapent_addr_remove"); 446 RBT_POISON(uvm_map_addr, entry, UVMMAP_DEADBEEF); 447 } 448 449 /* 450 * uvm_map_reference: add reference to a map 451 * 452 * => map need not be locked 453 */ 454 void 455 uvm_map_reference(struct vm_map *map) 456 { 457 atomic_inc_int(&map->ref_count); 458 } 459 460 void 461 uvm_map_lock_entry(struct vm_map_entry *entry) 462 { 463 if (entry->aref.ar_amap != NULL) { 464 amap_lock(entry->aref.ar_amap, RW_WRITE); 465 } 466 if (UVM_ET_ISOBJ(entry)) { 467 rw_enter(entry->object.uvm_obj->vmobjlock, RW_WRITE); 468 } 469 } 470 471 void 472 uvm_map_unlock_entry(struct vm_map_entry *entry) 473 { 474 if (UVM_ET_ISOBJ(entry)) { 475 rw_exit(entry->object.uvm_obj->vmobjlock); 476 } 477 if (entry->aref.ar_amap != NULL) { 478 amap_unlock(entry->aref.ar_amap); 479 } 480 } 481 482 /* 483 * Calculate the dused delta. 484 */ 485 vsize_t 486 uvmspace_dused(struct vm_map *map, vaddr_t min, vaddr_t max) 487 { 488 struct vmspace *vm; 489 vsize_t sz; 490 vaddr_t lmax; 491 vaddr_t stack_begin, stack_end; /* Position of stack. */ 492 493 KASSERT(map->flags & VM_MAP_ISVMSPACE); 494 vm_map_assert_anylock(map); 495 496 vm = (struct vmspace *)map; 497 stack_begin = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 498 stack_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 499 500 sz = 0; 501 while (min != max) { 502 lmax = max; 503 if (min < stack_begin && lmax > stack_begin) 504 lmax = stack_begin; 505 else if (min < stack_end && lmax > stack_end) 506 lmax = stack_end; 507 508 if (min >= stack_begin && min < stack_end) { 509 /* nothing */ 510 } else 511 sz += lmax - min; 512 min = lmax; 513 } 514 515 return sz >> PAGE_SHIFT; 516 } 517 518 /* 519 * Find the entry describing the given address. 520 */ 521 struct vm_map_entry* 522 uvm_map_entrybyaddr(struct uvm_map_addr *atree, vaddr_t addr) 523 { 524 struct vm_map_entry *iter; 525 526 iter = RBT_ROOT(uvm_map_addr, atree); 527 while (iter != NULL) { 528 if (iter->start > addr) 529 iter = RBT_LEFT(uvm_map_addr, iter); 530 else if (VMMAP_FREE_END(iter) <= addr) 531 iter = RBT_RIGHT(uvm_map_addr, iter); 532 else 533 return iter; 534 } 535 return NULL; 536 } 537 538 /* 539 * DEAD_ENTRY_PUSH(struct vm_map_deadq *deadq, struct vm_map_entry *entry) 540 * 541 * Push dead entries into a linked list. 542 * Since the linked list abuses the address tree for storage, the entry 543 * may not be linked in a map. 544 * 545 * *head must be initialized to NULL before the first call to this macro. 546 * uvm_unmap_detach(*head, 0) will remove dead entries. 547 */ 548 static inline void 549 dead_entry_push(struct uvm_map_deadq *deadq, struct vm_map_entry *entry) 550 { 551 TAILQ_INSERT_TAIL(deadq, entry, dfree.deadq); 552 } 553 #define DEAD_ENTRY_PUSH(_headptr, _entry) \ 554 dead_entry_push((_headptr), (_entry)) 555 556 /* 557 * Test if memory starting at addr with sz bytes is free. 558 * 559 * Fills in *start_ptr and *end_ptr to be the first and last entry describing 560 * the space. 561 * If called with prefilled *start_ptr and *end_ptr, they are to be correct. 562 */ 563 int 564 uvm_map_isavail(struct vm_map *map, struct uvm_addr_state *uaddr, 565 struct vm_map_entry **start_ptr, struct vm_map_entry **end_ptr, 566 vaddr_t addr, vsize_t sz) 567 { 568 struct uvm_addr_state *free; 569 struct uvm_map_addr *atree; 570 struct vm_map_entry *i, *i_end; 571 572 if (addr + sz < addr) 573 return 0; 574 575 vm_map_assert_anylock(map); 576 577 /* 578 * Kernel memory above uvm_maxkaddr is considered unavailable. 579 */ 580 if ((map->flags & VM_MAP_ISVMSPACE) == 0) { 581 if (addr + sz > uvm_maxkaddr) 582 return 0; 583 } 584 585 atree = &map->addr; 586 587 /* 588 * Fill in first, last, so they point at the entries containing the 589 * first and last address of the range. 590 * Note that if they are not NULL, we don't perform the lookup. 591 */ 592 KDASSERT(atree != NULL && start_ptr != NULL && end_ptr != NULL); 593 if (*start_ptr == NULL) { 594 *start_ptr = uvm_map_entrybyaddr(atree, addr); 595 if (*start_ptr == NULL) 596 return 0; 597 } else 598 KASSERT(*start_ptr == uvm_map_entrybyaddr(atree, addr)); 599 if (*end_ptr == NULL) { 600 if (VMMAP_FREE_END(*start_ptr) >= addr + sz) 601 *end_ptr = *start_ptr; 602 else { 603 *end_ptr = uvm_map_entrybyaddr(atree, addr + sz - 1); 604 if (*end_ptr == NULL) 605 return 0; 606 } 607 } else 608 KASSERT(*end_ptr == uvm_map_entrybyaddr(atree, addr + sz - 1)); 609 610 /* Validation. */ 611 KDASSERT(*start_ptr != NULL && *end_ptr != NULL); 612 KDASSERT((*start_ptr)->start <= addr && 613 VMMAP_FREE_END(*start_ptr) > addr && 614 (*end_ptr)->start < addr + sz && 615 VMMAP_FREE_END(*end_ptr) >= addr + sz); 616 617 /* 618 * Check the none of the entries intersects with <addr, addr+sz>. 619 * Also, if the entry belong to uaddr_exe or uaddr_brk_stack, it is 620 * considered unavailable unless called by those allocators. 621 */ 622 i = *start_ptr; 623 i_end = RBT_NEXT(uvm_map_addr, *end_ptr); 624 for (; i != i_end; 625 i = RBT_NEXT(uvm_map_addr, i)) { 626 if (i->start != i->end && i->end > addr) 627 return 0; 628 629 /* 630 * uaddr_exe and uaddr_brk_stack may only be used 631 * by these allocators and the NULL uaddr (i.e. no 632 * uaddr). 633 * Reject if this requirement is not met. 634 */ 635 if (uaddr != NULL) { 636 free = uvm_map_uaddr_e(map, i); 637 638 if (uaddr != free && free != NULL && 639 (free == map->uaddr_exe || 640 free == map->uaddr_brk_stack)) 641 return 0; 642 } 643 } 644 645 return -1; 646 } 647 648 /* 649 * Invoke each address selector until an address is found. 650 * Will not invoke uaddr_exe. 651 */ 652 int 653 uvm_map_findspace(struct vm_map *map, struct vm_map_entry**first, 654 struct vm_map_entry**last, vaddr_t *addr, vsize_t sz, 655 vaddr_t pmap_align, vaddr_t pmap_offset, vm_prot_t prot, vaddr_t hint) 656 { 657 struct uvm_addr_state *uaddr; 658 int i; 659 660 /* 661 * Allocation for sz bytes at any address, 662 * using the addr selectors in order. 663 */ 664 for (i = 0; i < nitems(map->uaddr_any); i++) { 665 uaddr = map->uaddr_any[i]; 666 667 if (uvm_addr_invoke(map, uaddr, first, last, 668 addr, sz, pmap_align, pmap_offset, prot, hint) == 0) 669 return 0; 670 } 671 672 /* Fall back to brk() and stack() address selectors. */ 673 uaddr = map->uaddr_brk_stack; 674 if (uvm_addr_invoke(map, uaddr, first, last, 675 addr, sz, pmap_align, pmap_offset, prot, hint) == 0) 676 return 0; 677 678 return ENOMEM; 679 } 680 681 /* Calculate entry augmentation value. */ 682 vsize_t 683 uvm_map_addr_augment_get(struct vm_map_entry *entry) 684 { 685 vsize_t augment; 686 struct vm_map_entry *left, *right; 687 688 augment = entry->fspace; 689 if ((left = RBT_LEFT(uvm_map_addr, entry)) != NULL) 690 augment = MAX(augment, left->fspace_augment); 691 if ((right = RBT_RIGHT(uvm_map_addr, entry)) != NULL) 692 augment = MAX(augment, right->fspace_augment); 693 return augment; 694 } 695 696 /* 697 * Update augmentation data in entry. 698 */ 699 void 700 uvm_map_addr_augment(struct vm_map_entry *entry) 701 { 702 vsize_t augment; 703 704 while (entry != NULL) { 705 /* Calculate value for augmentation. */ 706 augment = uvm_map_addr_augment_get(entry); 707 708 /* 709 * Descend update. 710 * Once we find an entry that already has the correct value, 711 * stop, since it means all its parents will use the correct 712 * value too. 713 */ 714 if (entry->fspace_augment == augment) 715 return; 716 entry->fspace_augment = augment; 717 entry = RBT_PARENT(uvm_map_addr, entry); 718 } 719 } 720 721 /* 722 * uvm_mapanon: establish a valid mapping in map for an anon 723 * 724 * => *addr and sz must be a multiple of PAGE_SIZE. 725 * => *addr is ignored, except if flags contains UVM_FLAG_FIXED. 726 * => map must be unlocked. 727 * 728 * => align: align vaddr, must be a power-of-2. 729 * Align is only a hint and will be ignored if the alignment fails. 730 */ 731 int 732 uvm_mapanon(struct vm_map *map, vaddr_t *addr, vsize_t sz, 733 vsize_t align, unsigned int flags) 734 { 735 struct vm_map_entry *first, *last, *entry, *new; 736 struct uvm_map_deadq dead; 737 vm_prot_t prot; 738 vm_prot_t maxprot; 739 vm_inherit_t inherit; 740 int advice; 741 int error; 742 vaddr_t pmap_align, pmap_offset; 743 vaddr_t hint; 744 745 KASSERT((map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE); 746 KASSERT(map != kernel_map); 747 KASSERT((map->flags & UVM_FLAG_HOLE) == 0); 748 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 749 splassert(IPL_NONE); 750 KASSERT((flags & UVM_FLAG_TRYLOCK) == 0); 751 752 /* 753 * We use pmap_align and pmap_offset as alignment and offset variables. 754 * 755 * Because the align parameter takes precedence over pmap prefer, 756 * the pmap_align will need to be set to align, with pmap_offset = 0, 757 * if pmap_prefer will not align. 758 */ 759 pmap_align = MAX(align, PAGE_SIZE); 760 pmap_offset = 0; 761 762 /* Decode parameters. */ 763 prot = UVM_PROTECTION(flags); 764 maxprot = UVM_MAXPROTECTION(flags); 765 advice = UVM_ADVICE(flags); 766 inherit = UVM_INHERIT(flags); 767 error = 0; 768 hint = trunc_page(*addr); 769 TAILQ_INIT(&dead); 770 KASSERT((sz & (vaddr_t)PAGE_MASK) == 0); 771 KASSERT((align & (align - 1)) == 0); 772 773 /* Check protection. */ 774 if ((prot & maxprot) != prot) 775 return EACCES; 776 777 /* 778 * Before grabbing the lock, allocate a map entry for later 779 * use to ensure we don't wait for memory while holding the 780 * vm_map_lock. 781 */ 782 new = uvm_mapent_alloc(map, flags); 783 if (new == NULL) 784 return ENOMEM; 785 786 vm_map_lock(map); 787 first = last = NULL; 788 if (flags & UVM_FLAG_FIXED) { 789 /* 790 * Fixed location. 791 * 792 * Note: we ignore align, pmap_prefer. 793 * Fill in first, last and *addr. 794 */ 795 KASSERT((*addr & PAGE_MASK) == 0); 796 797 /* Check that the space is available. */ 798 if (flags & UVM_FLAG_UNMAP) { 799 if ((flags & UVM_FLAG_STACK) && 800 !uvm_map_is_stack_remappable(map, *addr, sz, 801 (flags & UVM_FLAG_SIGALTSTACK))) { 802 error = EINVAL; 803 goto unlock; 804 } 805 if (uvm_unmap_remove(map, *addr, *addr + sz, &dead, 806 FALSE, TRUE, 807 (flags & UVM_FLAG_SIGALTSTACK) ? FALSE : TRUE) != 0) { 808 error = EPERM; /* immutable entries found */ 809 goto unlock; 810 } 811 } 812 if (!uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 813 error = ENOMEM; 814 goto unlock; 815 } 816 } else if (*addr != 0 && (*addr & PAGE_MASK) == 0 && 817 (align == 0 || (*addr & (align - 1)) == 0) && 818 uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 819 /* 820 * Address used as hint. 821 * 822 * Note: we enforce the alignment restriction, 823 * but ignore pmap_prefer. 824 */ 825 } else if ((prot & PROT_EXEC) != 0 && map->uaddr_exe != NULL) { 826 /* Run selection algorithm for executables. */ 827 error = uvm_addr_invoke(map, map->uaddr_exe, &first, &last, 828 addr, sz, pmap_align, pmap_offset, prot, hint); 829 830 if (error != 0) 831 goto unlock; 832 } else { 833 /* Update freelists from vmspace. */ 834 uvm_map_vmspace_update(map, &dead, flags); 835 836 error = uvm_map_findspace(map, &first, &last, addr, sz, 837 pmap_align, pmap_offset, prot, hint); 838 839 if (error != 0) 840 goto unlock; 841 } 842 843 /* Double-check if selected address doesn't cause overflow. */ 844 if (*addr + sz < *addr) { 845 error = ENOMEM; 846 goto unlock; 847 } 848 849 /* If we only want a query, return now. */ 850 if (flags & UVM_FLAG_QUERY) { 851 error = 0; 852 goto unlock; 853 } 854 855 /* 856 * Create new entry. 857 * first and last may be invalidated after this call. 858 */ 859 entry = uvm_map_mkentry(map, first, last, *addr, sz, flags, &dead, 860 new); 861 if (entry == NULL) { 862 error = ENOMEM; 863 goto unlock; 864 } 865 new = NULL; 866 KDASSERT(entry->start == *addr && entry->end == *addr + sz); 867 entry->object.uvm_obj = NULL; 868 entry->offset = 0; 869 entry->protection = prot; 870 entry->max_protection = maxprot; 871 entry->inheritance = inherit; 872 entry->wired_count = 0; 873 entry->advice = advice; 874 if (flags & UVM_FLAG_STACK) { 875 entry->etype |= UVM_ET_STACK; 876 if (flags & (UVM_FLAG_FIXED | UVM_FLAG_UNMAP)) 877 map->sserial++; 878 } 879 if (flags & UVM_FLAG_COPYONW) { 880 entry->etype |= UVM_ET_COPYONWRITE; 881 if ((flags & UVM_FLAG_OVERLAY) == 0) 882 entry->etype |= UVM_ET_NEEDSCOPY; 883 } 884 if (flags & UVM_FLAG_CONCEAL) 885 entry->etype |= UVM_ET_CONCEAL; 886 if (flags & UVM_FLAG_OVERLAY) { 887 entry->aref.ar_pageoff = 0; 888 entry->aref.ar_amap = amap_alloc(sz, M_WAITOK, 0); 889 } 890 891 /* Update map and process statistics. */ 892 map->size += sz; 893 if (prot != PROT_NONE) { 894 ((struct vmspace *)map)->vm_dused += 895 uvmspace_dused(map, *addr, *addr + sz); 896 } 897 898 unlock: 899 vm_map_unlock(map); 900 901 /* 902 * Remove dead entries. 903 * 904 * Dead entries may be the result of merging. 905 * uvm_map_mkentry may also create dead entries, when it attempts to 906 * destroy free-space entries. 907 */ 908 uvm_unmap_detach(&dead, 0); 909 910 if (new) 911 uvm_mapent_free(new); 912 return error; 913 } 914 915 /* 916 * uvm_map: establish a valid mapping in map 917 * 918 * => *addr and sz must be a multiple of PAGE_SIZE. 919 * => map must be unlocked. 920 * => <uobj,uoffset> value meanings (4 cases): 921 * [1] <NULL,uoffset> == uoffset is a hint for PMAP_PREFER 922 * [2] <NULL,UVM_UNKNOWN_OFFSET> == don't PMAP_PREFER 923 * [3] <uobj,uoffset> == normal mapping 924 * [4] <uobj,UVM_UNKNOWN_OFFSET> == uvm_map finds offset based on VA 925 * 926 * case [4] is for kernel mappings where we don't know the offset until 927 * we've found a virtual address. note that kernel object offsets are 928 * always relative to vm_map_min(kernel_map). 929 * 930 * => align: align vaddr, must be a power-of-2. 931 * Align is only a hint and will be ignored if the alignment fails. 932 */ 933 int 934 uvm_map(struct vm_map *map, vaddr_t *addr, vsize_t sz, 935 struct uvm_object *uobj, voff_t uoffset, 936 vsize_t align, unsigned int flags) 937 { 938 struct vm_map_entry *first, *last, *entry, *new; 939 struct uvm_map_deadq dead; 940 vm_prot_t prot; 941 vm_prot_t maxprot; 942 vm_inherit_t inherit; 943 int advice; 944 int error; 945 vaddr_t pmap_align, pmap_offset; 946 vaddr_t hint; 947 948 if ((map->flags & VM_MAP_INTRSAFE) == 0) 949 splassert(IPL_NONE); 950 else 951 splassert(IPL_VM); 952 953 /* 954 * We use pmap_align and pmap_offset as alignment and offset variables. 955 * 956 * Because the align parameter takes precedence over pmap prefer, 957 * the pmap_align will need to be set to align, with pmap_offset = 0, 958 * if pmap_prefer will not align. 959 */ 960 if (uoffset == UVM_UNKNOWN_OFFSET) { 961 pmap_align = MAX(align, PAGE_SIZE); 962 pmap_offset = 0; 963 } else { 964 pmap_align = MAX(PMAP_PREFER_ALIGN(), PAGE_SIZE); 965 pmap_offset = PMAP_PREFER_OFFSET(uoffset); 966 967 if (align == 0 || 968 (align <= pmap_align && (pmap_offset & (align - 1)) == 0)) { 969 /* pmap_offset satisfies align, no change. */ 970 } else { 971 /* Align takes precedence over pmap prefer. */ 972 pmap_align = align; 973 pmap_offset = 0; 974 } 975 } 976 977 /* Decode parameters. */ 978 prot = UVM_PROTECTION(flags); 979 maxprot = UVM_MAXPROTECTION(flags); 980 advice = UVM_ADVICE(flags); 981 inherit = UVM_INHERIT(flags); 982 error = 0; 983 hint = trunc_page(*addr); 984 TAILQ_INIT(&dead); 985 KASSERT((sz & (vaddr_t)PAGE_MASK) == 0); 986 KASSERT((align & (align - 1)) == 0); 987 988 /* Holes are incompatible with other types of mappings. */ 989 if (flags & UVM_FLAG_HOLE) { 990 KASSERT(uobj == NULL && (flags & UVM_FLAG_FIXED) && 991 (flags & (UVM_FLAG_OVERLAY | UVM_FLAG_COPYONW)) == 0); 992 } 993 994 /* Unset hint for kernel_map non-fixed allocations. */ 995 if (!(map->flags & VM_MAP_ISVMSPACE) && !(flags & UVM_FLAG_FIXED)) 996 hint = 0; 997 998 /* Check protection. */ 999 if ((prot & maxprot) != prot) 1000 return EACCES; 1001 1002 if (map == kernel_map && 1003 (prot & (PROT_WRITE | PROT_EXEC)) == (PROT_WRITE | PROT_EXEC)) 1004 panic("uvm_map: kernel map W^X violation requested"); 1005 1006 /* 1007 * Before grabbing the lock, allocate a map entry for later 1008 * use to ensure we don't wait for memory while holding the 1009 * vm_map_lock. 1010 */ 1011 new = uvm_mapent_alloc(map, flags); 1012 if (new == NULL) 1013 return ENOMEM; 1014 1015 if (flags & UVM_FLAG_TRYLOCK) { 1016 if (vm_map_lock_try(map) == FALSE) { 1017 error = EFAULT; 1018 goto out; 1019 } 1020 } else { 1021 vm_map_lock(map); 1022 } 1023 1024 first = last = NULL; 1025 if (flags & UVM_FLAG_FIXED) { 1026 /* 1027 * Fixed location. 1028 * 1029 * Note: we ignore align, pmap_prefer. 1030 * Fill in first, last and *addr. 1031 */ 1032 KASSERT((*addr & PAGE_MASK) == 0); 1033 1034 /* 1035 * Grow pmap to include allocated address. 1036 * If the growth fails, the allocation will fail too. 1037 */ 1038 if ((map->flags & VM_MAP_ISVMSPACE) == 0 && 1039 uvm_maxkaddr < (*addr + sz)) { 1040 uvm_map_kmem_grow(map, &dead, 1041 *addr + sz - uvm_maxkaddr, flags); 1042 } 1043 1044 /* Check that the space is available. */ 1045 if (flags & UVM_FLAG_UNMAP) { 1046 if (uvm_unmap_remove(map, *addr, *addr + sz, &dead, 1047 FALSE, TRUE, TRUE) != 0) { 1048 error = EPERM; /* immutable entries found */ 1049 goto unlock; 1050 } 1051 } 1052 if (!uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1053 error = ENOMEM; 1054 goto unlock; 1055 } 1056 } else if (*addr != 0 && (*addr & PAGE_MASK) == 0 && 1057 (map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE && 1058 (align == 0 || (*addr & (align - 1)) == 0) && 1059 uvm_map_isavail(map, NULL, &first, &last, *addr, sz)) { 1060 /* 1061 * Address used as hint. 1062 * 1063 * Note: we enforce the alignment restriction, 1064 * but ignore pmap_prefer. 1065 */ 1066 } else if ((prot & PROT_EXEC) != 0 && map->uaddr_exe != NULL) { 1067 /* Run selection algorithm for executables. */ 1068 error = uvm_addr_invoke(map, map->uaddr_exe, &first, &last, 1069 addr, sz, pmap_align, pmap_offset, prot, hint); 1070 1071 /* Grow kernel memory and try again. */ 1072 if (error != 0 && (map->flags & VM_MAP_ISVMSPACE) == 0) { 1073 uvm_map_kmem_grow(map, &dead, sz, flags); 1074 1075 error = uvm_addr_invoke(map, map->uaddr_exe, 1076 &first, &last, addr, sz, 1077 pmap_align, pmap_offset, prot, hint); 1078 } 1079 1080 if (error != 0) 1081 goto unlock; 1082 } else { 1083 /* Update freelists from vmspace. */ 1084 if (map->flags & VM_MAP_ISVMSPACE) 1085 uvm_map_vmspace_update(map, &dead, flags); 1086 1087 error = uvm_map_findspace(map, &first, &last, addr, sz, 1088 pmap_align, pmap_offset, prot, hint); 1089 1090 /* Grow kernel memory and try again. */ 1091 if (error != 0 && (map->flags & VM_MAP_ISVMSPACE) == 0) { 1092 uvm_map_kmem_grow(map, &dead, sz, flags); 1093 1094 error = uvm_map_findspace(map, &first, &last, addr, sz, 1095 pmap_align, pmap_offset, prot, hint); 1096 } 1097 1098 if (error != 0) 1099 goto unlock; 1100 } 1101 1102 /* Double-check if selected address doesn't cause overflow. */ 1103 if (*addr + sz < *addr) { 1104 error = ENOMEM; 1105 goto unlock; 1106 } 1107 1108 KASSERT((map->flags & VM_MAP_ISVMSPACE) == VM_MAP_ISVMSPACE || 1109 uvm_maxkaddr >= *addr + sz); 1110 1111 /* If we only want a query, return now. */ 1112 if (flags & UVM_FLAG_QUERY) { 1113 error = 0; 1114 goto unlock; 1115 } 1116 1117 if (uobj == NULL) 1118 uoffset = 0; 1119 else if (uoffset == UVM_UNKNOWN_OFFSET) { 1120 KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj)); 1121 uoffset = *addr - vm_map_min(kernel_map); 1122 } 1123 1124 /* 1125 * Create new entry. 1126 * first and last may be invalidated after this call. 1127 */ 1128 entry = uvm_map_mkentry(map, first, last, *addr, sz, flags, &dead, 1129 new); 1130 if (entry == NULL) { 1131 error = ENOMEM; 1132 goto unlock; 1133 } 1134 new = NULL; 1135 KDASSERT(entry->start == *addr && entry->end == *addr + sz); 1136 entry->object.uvm_obj = uobj; 1137 entry->offset = uoffset; 1138 entry->protection = prot; 1139 entry->max_protection = maxprot; 1140 entry->inheritance = inherit; 1141 entry->wired_count = 0; 1142 entry->advice = advice; 1143 if (flags & UVM_FLAG_STACK) { 1144 entry->etype |= UVM_ET_STACK; 1145 if (flags & UVM_FLAG_UNMAP) 1146 map->sserial++; 1147 } 1148 if (uobj) 1149 entry->etype |= UVM_ET_OBJ; 1150 else if (flags & UVM_FLAG_HOLE) 1151 entry->etype |= UVM_ET_HOLE; 1152 if (flags & UVM_FLAG_NOFAULT) 1153 entry->etype |= UVM_ET_NOFAULT; 1154 if (flags & UVM_FLAG_WC) 1155 entry->etype |= UVM_ET_WC; 1156 if (flags & UVM_FLAG_COPYONW) { 1157 entry->etype |= UVM_ET_COPYONWRITE; 1158 if ((flags & UVM_FLAG_OVERLAY) == 0) 1159 entry->etype |= UVM_ET_NEEDSCOPY; 1160 } 1161 if (flags & UVM_FLAG_CONCEAL) 1162 entry->etype |= UVM_ET_CONCEAL; 1163 if (flags & UVM_FLAG_OVERLAY) { 1164 entry->aref.ar_pageoff = 0; 1165 entry->aref.ar_amap = amap_alloc(sz, M_WAITOK, 0); 1166 } 1167 1168 /* Update map and process statistics. */ 1169 if (!(flags & UVM_FLAG_HOLE)) { 1170 map->size += sz; 1171 if ((map->flags & VM_MAP_ISVMSPACE) && uobj == NULL && 1172 prot != PROT_NONE) { 1173 ((struct vmspace *)map)->vm_dused += 1174 uvmspace_dused(map, *addr, *addr + sz); 1175 } 1176 } 1177 1178 /* 1179 * Try to merge entry. 1180 * 1181 * Userland allocations are kept separated most of the time. 1182 * Forego the effort of merging what most of the time can't be merged 1183 * and only try the merge if it concerns a kernel entry. 1184 */ 1185 if ((flags & UVM_FLAG_NOMERGE) == 0 && 1186 (map->flags & VM_MAP_ISVMSPACE) == 0) 1187 uvm_mapent_tryjoin(map, entry, &dead); 1188 1189 unlock: 1190 vm_map_unlock(map); 1191 1192 /* 1193 * Remove dead entries. 1194 * 1195 * Dead entries may be the result of merging. 1196 * uvm_map_mkentry may also create dead entries, when it attempts to 1197 * destroy free-space entries. 1198 */ 1199 if (map->flags & VM_MAP_INTRSAFE) 1200 uvm_unmap_detach_intrsafe(&dead); 1201 else 1202 uvm_unmap_detach(&dead, 0); 1203 out: 1204 if (new) 1205 uvm_mapent_free(new); 1206 return error; 1207 } 1208 1209 /* 1210 * True iff e1 and e2 can be joined together. 1211 */ 1212 int 1213 uvm_mapent_isjoinable(struct vm_map *map, struct vm_map_entry *e1, 1214 struct vm_map_entry *e2) 1215 { 1216 KDASSERT(e1 != NULL && e2 != NULL); 1217 1218 /* Must be the same entry type and not have free memory between. */ 1219 if (e1->etype != e2->etype || e1->end != e2->start) 1220 return 0; 1221 1222 /* Submaps are never joined. */ 1223 if (UVM_ET_ISSUBMAP(e1)) 1224 return 0; 1225 1226 /* Never merge wired memory. */ 1227 if (VM_MAPENT_ISWIRED(e1) || VM_MAPENT_ISWIRED(e2)) 1228 return 0; 1229 1230 /* Protection, inheritance and advice must be equal. */ 1231 if (e1->protection != e2->protection || 1232 e1->max_protection != e2->max_protection || 1233 e1->inheritance != e2->inheritance || 1234 e1->advice != e2->advice) 1235 return 0; 1236 1237 /* If uvm_object: object itself and offsets within object must match. */ 1238 if (UVM_ET_ISOBJ(e1)) { 1239 if (e1->object.uvm_obj != e2->object.uvm_obj) 1240 return 0; 1241 if (e1->offset + (e1->end - e1->start) != e2->offset) 1242 return 0; 1243 } 1244 1245 /* 1246 * Cannot join shared amaps. 1247 * Note: no need to lock amap to look at refs, since we don't care 1248 * about its exact value. 1249 * If it is 1 (i.e. we have the only reference) it will stay there. 1250 */ 1251 if (e1->aref.ar_amap && amap_refs(e1->aref.ar_amap) != 1) 1252 return 0; 1253 if (e2->aref.ar_amap && amap_refs(e2->aref.ar_amap) != 1) 1254 return 0; 1255 1256 /* Apparently, e1 and e2 match. */ 1257 return 1; 1258 } 1259 1260 /* 1261 * Join support function. 1262 * 1263 * Returns the merged entry on success. 1264 * Returns NULL if the merge failed. 1265 */ 1266 struct vm_map_entry* 1267 uvm_mapent_merge(struct vm_map *map, struct vm_map_entry *e1, 1268 struct vm_map_entry *e2, struct uvm_map_deadq *dead) 1269 { 1270 struct uvm_addr_state *free; 1271 1272 /* 1273 * Merging is not supported for map entries that 1274 * contain an amap in e1. This should never happen 1275 * anyway, because only kernel entries are merged. 1276 * These do not contain amaps. 1277 * e2 contains no real information in its amap, 1278 * so it can be erased immediately. 1279 */ 1280 KASSERT(e1->aref.ar_amap == NULL); 1281 1282 /* 1283 * Don't drop obj reference: 1284 * uvm_unmap_detach will do this for us. 1285 */ 1286 free = uvm_map_uaddr_e(map, e1); 1287 uvm_mapent_free_remove(map, free, e1); 1288 1289 free = uvm_map_uaddr_e(map, e2); 1290 uvm_mapent_free_remove(map, free, e2); 1291 uvm_mapent_addr_remove(map, e2); 1292 e1->end = e2->end; 1293 e1->guard = e2->guard; 1294 e1->fspace = e2->fspace; 1295 uvm_mapent_free_insert(map, free, e1); 1296 1297 DEAD_ENTRY_PUSH(dead, e2); 1298 return e1; 1299 } 1300 1301 /* 1302 * Attempt forward and backward joining of entry. 1303 * 1304 * Returns entry after joins. 1305 * We are guaranteed that the amap of entry is either non-existent or 1306 * has never been used. 1307 */ 1308 struct vm_map_entry* 1309 uvm_mapent_tryjoin(struct vm_map *map, struct vm_map_entry *entry, 1310 struct uvm_map_deadq *dead) 1311 { 1312 struct vm_map_entry *other; 1313 struct vm_map_entry *merged; 1314 1315 /* Merge with previous entry. */ 1316 other = RBT_PREV(uvm_map_addr, entry); 1317 if (other && uvm_mapent_isjoinable(map, other, entry)) { 1318 merged = uvm_mapent_merge(map, other, entry, dead); 1319 if (merged) 1320 entry = merged; 1321 } 1322 1323 /* 1324 * Merge with next entry. 1325 * 1326 * Because amap can only extend forward and the next entry 1327 * probably contains sensible info, only perform forward merging 1328 * in the absence of an amap. 1329 */ 1330 other = RBT_NEXT(uvm_map_addr, entry); 1331 if (other && entry->aref.ar_amap == NULL && 1332 other->aref.ar_amap == NULL && 1333 uvm_mapent_isjoinable(map, entry, other)) { 1334 merged = uvm_mapent_merge(map, entry, other, dead); 1335 if (merged) 1336 entry = merged; 1337 } 1338 1339 return entry; 1340 } 1341 1342 /* 1343 * Kill entries that are no longer in a map. 1344 */ 1345 void 1346 uvm_unmap_detach(struct uvm_map_deadq *deadq, int flags) 1347 { 1348 struct vm_map_entry *entry, *tmp; 1349 1350 TAILQ_FOREACH_SAFE(entry, deadq, dfree.deadq, tmp) { 1351 /* Drop reference to amap, if we've got one. */ 1352 if (entry->aref.ar_amap) 1353 amap_unref(entry->aref.ar_amap, 1354 entry->aref.ar_pageoff, 1355 atop(entry->end - entry->start), 1356 flags & AMAP_REFALL); 1357 1358 /* Drop reference to our backing object, if we've got one. */ 1359 if (UVM_ET_ISSUBMAP(entry)) { 1360 /* ... unlikely to happen, but play it safe */ 1361 uvm_map_deallocate(entry->object.sub_map); 1362 } else if (UVM_ET_ISOBJ(entry) && 1363 entry->object.uvm_obj->pgops->pgo_detach) { 1364 entry->object.uvm_obj->pgops->pgo_detach( 1365 entry->object.uvm_obj); 1366 } 1367 1368 TAILQ_REMOVE(deadq, entry, dfree.deadq); 1369 uvm_mapent_free(entry); 1370 } 1371 } 1372 1373 void 1374 uvm_unmap_detach_intrsafe(struct uvm_map_deadq *deadq) 1375 { 1376 struct vm_map_entry *entry; 1377 1378 while ((entry = TAILQ_FIRST(deadq)) != NULL) { 1379 KASSERT(entry->aref.ar_amap == NULL); 1380 KASSERT(!UVM_ET_ISSUBMAP(entry)); 1381 KASSERT(!UVM_ET_ISOBJ(entry)); 1382 TAILQ_REMOVE(deadq, entry, dfree.deadq); 1383 uvm_mapent_free(entry); 1384 } 1385 } 1386 1387 /* 1388 * Create and insert new entry. 1389 * 1390 * Returned entry contains new addresses and is inserted properly in the tree. 1391 * first and last are (probably) no longer valid. 1392 */ 1393 struct vm_map_entry* 1394 uvm_map_mkentry(struct vm_map *map, struct vm_map_entry *first, 1395 struct vm_map_entry *last, vaddr_t addr, vsize_t sz, int flags, 1396 struct uvm_map_deadq *dead, struct vm_map_entry *new) 1397 { 1398 struct vm_map_entry *entry, *prev; 1399 struct uvm_addr_state *free; 1400 vaddr_t min, max; /* free space boundaries for new entry */ 1401 1402 KDASSERT(map != NULL); 1403 KDASSERT(first != NULL); 1404 KDASSERT(last != NULL); 1405 KDASSERT(dead != NULL); 1406 KDASSERT(sz > 0); 1407 KDASSERT(addr + sz > addr); 1408 KDASSERT(first->end <= addr && VMMAP_FREE_END(first) > addr); 1409 KDASSERT(last->start < addr + sz && VMMAP_FREE_END(last) >= addr + sz); 1410 KDASSERT(uvm_map_isavail(map, NULL, &first, &last, addr, sz)); 1411 uvm_tree_sanity(map, __FILE__, __LINE__); 1412 1413 min = addr + sz; 1414 max = VMMAP_FREE_END(last); 1415 1416 /* Initialize new entry. */ 1417 if (new == NULL) 1418 entry = uvm_mapent_alloc(map, flags); 1419 else 1420 entry = new; 1421 if (entry == NULL) 1422 return NULL; 1423 entry->offset = 0; 1424 entry->etype = 0; 1425 entry->wired_count = 0; 1426 entry->aref.ar_pageoff = 0; 1427 entry->aref.ar_amap = NULL; 1428 1429 entry->start = addr; 1430 entry->end = min; 1431 entry->guard = 0; 1432 entry->fspace = 0; 1433 1434 vm_map_assert_wrlock(map); 1435 1436 /* Reset free space in first. */ 1437 free = uvm_map_uaddr_e(map, first); 1438 uvm_mapent_free_remove(map, free, first); 1439 first->guard = 0; 1440 first->fspace = 0; 1441 1442 /* 1443 * Remove all entries that are fully replaced. 1444 * We are iterating using last in reverse order. 1445 */ 1446 for (; first != last; last = prev) { 1447 prev = RBT_PREV(uvm_map_addr, last); 1448 1449 KDASSERT(last->start == last->end); 1450 free = uvm_map_uaddr_e(map, last); 1451 uvm_mapent_free_remove(map, free, last); 1452 uvm_mapent_addr_remove(map, last); 1453 DEAD_ENTRY_PUSH(dead, last); 1454 } 1455 /* Remove first if it is entirely inside <addr, addr+sz>. */ 1456 if (first->start == addr) { 1457 uvm_mapent_addr_remove(map, first); 1458 DEAD_ENTRY_PUSH(dead, first); 1459 } else { 1460 uvm_map_fix_space(map, first, VMMAP_FREE_START(first), 1461 addr, flags); 1462 } 1463 1464 /* Finally, link in entry. */ 1465 uvm_mapent_addr_insert(map, entry); 1466 uvm_map_fix_space(map, entry, min, max, flags); 1467 1468 uvm_tree_sanity(map, __FILE__, __LINE__); 1469 return entry; 1470 } 1471 1472 1473 /* 1474 * uvm_mapent_alloc: allocate a map entry 1475 */ 1476 struct vm_map_entry * 1477 uvm_mapent_alloc(struct vm_map *map, int flags) 1478 { 1479 struct vm_map_entry *me, *ne; 1480 int pool_flags; 1481 int i; 1482 1483 pool_flags = PR_WAITOK; 1484 if (flags & UVM_FLAG_TRYLOCK) 1485 pool_flags = PR_NOWAIT; 1486 1487 if (map->flags & VM_MAP_INTRSAFE || cold) { 1488 mtx_enter(&uvm_kmapent_mtx); 1489 if (SLIST_EMPTY(&uvm.kentry_free)) { 1490 ne = km_alloc(PAGE_SIZE, &kv_page, &kp_dirty, 1491 &kd_nowait); 1492 if (ne == NULL) 1493 panic("uvm_mapent_alloc: cannot allocate map " 1494 "entry"); 1495 for (i = 0; i < PAGE_SIZE / sizeof(*ne); i++) { 1496 SLIST_INSERT_HEAD(&uvm.kentry_free, 1497 &ne[i], daddrs.addr_kentry); 1498 } 1499 if (ratecheck(&uvm_kmapent_last_warn_time, 1500 &uvm_kmapent_warn_rate)) 1501 printf("uvm_mapent_alloc: out of static " 1502 "map entries\n"); 1503 } 1504 me = SLIST_FIRST(&uvm.kentry_free); 1505 SLIST_REMOVE_HEAD(&uvm.kentry_free, daddrs.addr_kentry); 1506 uvmexp.kmapent++; 1507 mtx_leave(&uvm_kmapent_mtx); 1508 me->flags = UVM_MAP_STATIC; 1509 } else if (map == kernel_map) { 1510 splassert(IPL_NONE); 1511 me = pool_get(&uvm_map_entry_kmem_pool, pool_flags); 1512 if (me == NULL) 1513 goto out; 1514 me->flags = UVM_MAP_KMEM; 1515 } else { 1516 splassert(IPL_NONE); 1517 me = pool_get(&uvm_map_entry_pool, pool_flags); 1518 if (me == NULL) 1519 goto out; 1520 me->flags = 0; 1521 } 1522 1523 RBT_POISON(uvm_map_addr, me, UVMMAP_DEADBEEF); 1524 out: 1525 return me; 1526 } 1527 1528 /* 1529 * uvm_mapent_free: free map entry 1530 * 1531 * => XXX: static pool for kernel map? 1532 */ 1533 void 1534 uvm_mapent_free(struct vm_map_entry *me) 1535 { 1536 if (me->flags & UVM_MAP_STATIC) { 1537 mtx_enter(&uvm_kmapent_mtx); 1538 SLIST_INSERT_HEAD(&uvm.kentry_free, me, daddrs.addr_kentry); 1539 uvmexp.kmapent--; 1540 mtx_leave(&uvm_kmapent_mtx); 1541 } else if (me->flags & UVM_MAP_KMEM) { 1542 splassert(IPL_NONE); 1543 pool_put(&uvm_map_entry_kmem_pool, me); 1544 } else { 1545 splassert(IPL_NONE); 1546 pool_put(&uvm_map_entry_pool, me); 1547 } 1548 } 1549 1550 /* 1551 * uvm_map_lookup_entry: find map entry at or before an address. 1552 * 1553 * => map must at least be read-locked by caller 1554 * => entry is returned in "entry" 1555 * => return value is true if address is in the returned entry 1556 * ET_HOLE entries are considered to not contain a mapping, ergo FALSE is 1557 * returned for those mappings. 1558 */ 1559 boolean_t 1560 uvm_map_lookup_entry(struct vm_map *map, vaddr_t address, 1561 struct vm_map_entry **entry) 1562 { 1563 vm_map_assert_anylock(map); 1564 1565 *entry = uvm_map_entrybyaddr(&map->addr, address); 1566 return *entry != NULL && !UVM_ET_ISHOLE(*entry) && 1567 (*entry)->start <= address && (*entry)->end > address; 1568 } 1569 1570 /* 1571 * Stack must be in a MAP_STACK entry. PROT_NONE indicates stack not yet 1572 * grown -- then uvm_map_check_region_range() should not cache the entry 1573 * because growth won't be seen. 1574 */ 1575 int 1576 uvm_map_inentry_sp(vm_map_entry_t entry) 1577 { 1578 if ((entry->etype & UVM_ET_STACK) == 0) { 1579 if (entry->protection == PROT_NONE) 1580 return (-1); /* don't update range */ 1581 return (0); 1582 } 1583 return (1); 1584 } 1585 1586 int 1587 uvm_map_inentry_recheck(u_long serial, vaddr_t addr, struct p_inentry *ie) 1588 { 1589 return (serial != ie->ie_serial || ie->ie_start == 0 || 1590 addr < ie->ie_start || addr >= ie->ie_end); 1591 } 1592 1593 /* 1594 * Inside a vm_map find the reg address and verify it via function. 1595 * Remember low and high addresses of region if valid and return TRUE, 1596 * else return FALSE. 1597 */ 1598 boolean_t 1599 uvm_map_inentry_fix(struct proc *p, struct p_inentry *ie, vaddr_t addr, 1600 int (*fn)(vm_map_entry_t), u_long serial) 1601 { 1602 vm_map_t map = &p->p_vmspace->vm_map; 1603 vm_map_entry_t entry; 1604 int ret; 1605 1606 if (addr < map->min_offset || addr >= map->max_offset) 1607 return (FALSE); 1608 1609 /* lock map */ 1610 vm_map_lock_read(map); 1611 1612 /* lookup */ 1613 if (!uvm_map_lookup_entry(map, trunc_page(addr), &entry)) { 1614 vm_map_unlock_read(map); 1615 return (FALSE); 1616 } 1617 1618 ret = (*fn)(entry); 1619 if (ret == 0) { 1620 vm_map_unlock_read(map); 1621 return (FALSE); 1622 } else if (ret == 1) { 1623 ie->ie_start = entry->start; 1624 ie->ie_end = entry->end; 1625 ie->ie_serial = serial; 1626 } else { 1627 /* do not update, re-check later */ 1628 } 1629 vm_map_unlock_read(map); 1630 return (TRUE); 1631 } 1632 1633 boolean_t 1634 uvm_map_inentry(struct proc *p, struct p_inentry *ie, vaddr_t addr, 1635 const char *fmt, int (*fn)(vm_map_entry_t), u_long serial) 1636 { 1637 union sigval sv; 1638 boolean_t ok = TRUE; 1639 1640 if (uvm_map_inentry_recheck(serial, addr, ie)) { 1641 ok = uvm_map_inentry_fix(p, ie, addr, fn, serial); 1642 if (!ok) { 1643 KERNEL_LOCK(); 1644 uprintf(fmt, p->p_p->ps_comm, p->p_p->ps_pid, p->p_tid, 1645 addr, ie->ie_start, ie->ie_end-1); 1646 p->p_p->ps_acflag |= AMAP; 1647 sv.sival_ptr = (void *)PROC_PC(p); 1648 trapsignal(p, SIGSEGV, 0, SEGV_ACCERR, sv); 1649 KERNEL_UNLOCK(); 1650 } 1651 } 1652 return (ok); 1653 } 1654 1655 /* 1656 * Check whether the given address range can be converted to a MAP_STACK 1657 * mapping. 1658 * 1659 * Must be called with map locked. 1660 */ 1661 boolean_t 1662 uvm_map_is_stack_remappable(struct vm_map *map, vaddr_t addr, vaddr_t sz, 1663 int sigaltstack_check) 1664 { 1665 vaddr_t end = addr + sz; 1666 struct vm_map_entry *first, *iter, *prev = NULL; 1667 1668 vm_map_assert_anylock(map); 1669 1670 if (!uvm_map_lookup_entry(map, addr, &first)) 1671 return FALSE; 1672 1673 /* 1674 * Check that the address range exists and is contiguous. 1675 */ 1676 for (iter = first; iter != NULL && iter->start < end; 1677 prev = iter, iter = RBT_NEXT(uvm_map_addr, iter)) { 1678 /* 1679 * Make sure that we do not have holes in the range. 1680 */ 1681 #if 0 1682 if (prev != NULL) { 1683 printf("prev->start 0x%lx, prev->end 0x%lx, " 1684 "iter->start 0x%lx, iter->end 0x%lx\n", 1685 prev->start, prev->end, iter->start, iter->end); 1686 } 1687 #endif 1688 1689 if (prev != NULL && prev->end != iter->start) 1690 return FALSE; 1691 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) 1692 return FALSE; 1693 if (sigaltstack_check) { 1694 if (iter->protection != (PROT_READ | PROT_WRITE)) 1695 return FALSE; 1696 } 1697 } 1698 1699 return TRUE; 1700 } 1701 1702 /* 1703 * Remap the middle-pages of an existing mapping as a stack range. 1704 * If there exists a previous contiguous mapping with the given range 1705 * [addr, addr + sz), with protection PROT_READ|PROT_WRITE, then the 1706 * mapping is dropped, and a new anon mapping is created and marked as 1707 * a stack. 1708 * 1709 * Must be called with map unlocked. 1710 */ 1711 int 1712 uvm_map_remap_as_stack(struct proc *p, vaddr_t addr, vaddr_t sz) 1713 { 1714 vm_map_t map = &p->p_vmspace->vm_map; 1715 vaddr_t start, end; 1716 int flags = UVM_MAPFLAG(PROT_READ | PROT_WRITE, 1717 PROT_READ | PROT_WRITE | PROT_EXEC, 1718 MAP_INHERIT_COPY, MADV_NORMAL, 1719 UVM_FLAG_STACK | UVM_FLAG_FIXED | UVM_FLAG_UNMAP | 1720 UVM_FLAG_COPYONW | UVM_FLAG_SIGALTSTACK); 1721 1722 start = round_page(addr); 1723 end = trunc_page(addr + sz); 1724 #ifdef MACHINE_STACK_GROWS_UP 1725 if (end == addr + sz) 1726 end -= PAGE_SIZE; 1727 #else 1728 if (start == addr) 1729 start += PAGE_SIZE; 1730 #endif 1731 1732 if (start < map->min_offset || end >= map->max_offset || end < start) 1733 return EINVAL; 1734 1735 /* 1736 * UVM_FLAG_SIGALTSTACK indicates that immutable may be bypassed, 1737 * but the range is checked that it is contiguous, is not a syscall 1738 * mapping, and protection RW. Then, a new mapping (all zero) is 1739 * placed upon the region, which prevents an attacker from pivoting 1740 * into pre-placed MAP_STACK space. 1741 */ 1742 return uvm_mapanon(map, &start, end - start, 0, flags); 1743 } 1744 1745 /* 1746 * uvm_map_pie: return a random load address for a PIE executable 1747 * properly aligned. 1748 */ 1749 #ifndef VM_PIE_MAX_ADDR 1750 #define VM_PIE_MAX_ADDR (VM_MAXUSER_ADDRESS / 4) 1751 #endif 1752 1753 #ifndef VM_PIE_MIN_ADDR 1754 #define VM_PIE_MIN_ADDR VM_MIN_ADDRESS 1755 #endif 1756 1757 #ifndef VM_PIE_MIN_ALIGN 1758 #define VM_PIE_MIN_ALIGN PAGE_SIZE 1759 #endif 1760 1761 vaddr_t 1762 uvm_map_pie(vaddr_t align) 1763 { 1764 vaddr_t addr, space, min; 1765 1766 align = MAX(align, VM_PIE_MIN_ALIGN); 1767 1768 /* round up to next alignment */ 1769 min = (VM_PIE_MIN_ADDR + align - 1) & ~(align - 1); 1770 1771 if (align >= VM_PIE_MAX_ADDR || min >= VM_PIE_MAX_ADDR) 1772 return (align); 1773 1774 space = (VM_PIE_MAX_ADDR - min) / align; 1775 space = MIN(space, (u_int32_t)-1); 1776 1777 addr = (vaddr_t)arc4random_uniform((u_int32_t)space) * align; 1778 addr += min; 1779 1780 return (addr); 1781 } 1782 1783 void 1784 uvm_unmap(struct vm_map *map, vaddr_t start, vaddr_t end) 1785 { 1786 struct uvm_map_deadq dead; 1787 1788 KASSERT((start & (vaddr_t)PAGE_MASK) == 0 && 1789 (end & (vaddr_t)PAGE_MASK) == 0); 1790 TAILQ_INIT(&dead); 1791 vm_map_lock(map); 1792 uvm_unmap_remove(map, start, end, &dead, FALSE, TRUE, FALSE); 1793 vm_map_unlock(map); 1794 1795 if (map->flags & VM_MAP_INTRSAFE) 1796 uvm_unmap_detach_intrsafe(&dead); 1797 else 1798 uvm_unmap_detach(&dead, 0); 1799 } 1800 1801 /* 1802 * Mark entry as free. 1803 * 1804 * entry will be put on the dead list. 1805 * The free space will be merged into the previous or a new entry, 1806 * unless markfree is false. 1807 */ 1808 void 1809 uvm_mapent_mkfree(struct vm_map *map, struct vm_map_entry *entry, 1810 struct vm_map_entry **prev_ptr, struct uvm_map_deadq *dead, 1811 boolean_t markfree) 1812 { 1813 struct uvm_addr_state *free; 1814 struct vm_map_entry *prev; 1815 vaddr_t addr; /* Start of freed range. */ 1816 vaddr_t end; /* End of freed range. */ 1817 1818 UVM_MAP_REQ_WRITE(map); 1819 1820 prev = *prev_ptr; 1821 if (prev == entry) 1822 *prev_ptr = prev = NULL; 1823 1824 if (prev == NULL || 1825 VMMAP_FREE_END(prev) != entry->start) 1826 prev = RBT_PREV(uvm_map_addr, entry); 1827 1828 /* Entry is describing only free memory and has nothing to drain into. */ 1829 if (prev == NULL && entry->start == entry->end && markfree) { 1830 *prev_ptr = entry; 1831 return; 1832 } 1833 1834 addr = entry->start; 1835 end = VMMAP_FREE_END(entry); 1836 free = uvm_map_uaddr_e(map, entry); 1837 uvm_mapent_free_remove(map, free, entry); 1838 uvm_mapent_addr_remove(map, entry); 1839 DEAD_ENTRY_PUSH(dead, entry); 1840 1841 if (markfree) { 1842 if (prev) { 1843 free = uvm_map_uaddr_e(map, prev); 1844 uvm_mapent_free_remove(map, free, prev); 1845 } 1846 *prev_ptr = uvm_map_fix_space(map, prev, addr, end, 0); 1847 } 1848 } 1849 1850 /* 1851 * Unwire and release referenced amap and object from map entry. 1852 */ 1853 void 1854 uvm_unmap_kill_entry_withlock(struct vm_map *map, struct vm_map_entry *entry, 1855 int needlock) 1856 { 1857 /* Unwire removed map entry. */ 1858 if (VM_MAPENT_ISWIRED(entry)) { 1859 entry->wired_count = 0; 1860 uvm_fault_unwire_locked(map, entry->start, entry->end); 1861 } 1862 1863 if (needlock) 1864 uvm_map_lock_entry(entry); 1865 1866 /* Entry-type specific code. */ 1867 if (UVM_ET_ISHOLE(entry)) { 1868 /* Nothing to be done for holes. */ 1869 } else if (map->flags & VM_MAP_INTRSAFE) { 1870 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1871 1872 uvm_km_pgremove_intrsafe(entry->start, entry->end); 1873 } else if (UVM_ET_ISOBJ(entry) && 1874 UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) { 1875 KASSERT(vm_map_pmap(map) == pmap_kernel()); 1876 /* 1877 * Note: kernel object mappings are currently used in 1878 * two ways: 1879 * [1] "normal" mappings of pages in the kernel object 1880 * [2] uvm_km_valloc'd allocations in which we 1881 * pmap_enter in some non-kernel-object page 1882 * (e.g. vmapbuf). 1883 * 1884 * for case [1], we need to remove the mapping from 1885 * the pmap and then remove the page from the kernel 1886 * object (because, once pages in a kernel object are 1887 * unmapped they are no longer needed, unlike, say, 1888 * a vnode where you might want the data to persist 1889 * until flushed out of a queue). 1890 * 1891 * for case [2], we need to remove the mapping from 1892 * the pmap. there shouldn't be any pages at the 1893 * specified offset in the kernel object [but it 1894 * doesn't hurt to call uvm_km_pgremove just to be 1895 * safe?] 1896 * 1897 * uvm_km_pgremove currently does the following: 1898 * for pages in the kernel object range: 1899 * - drops the swap slot 1900 * - uvm_pagefree the page 1901 * 1902 * note there is version of uvm_km_pgremove() that 1903 * is used for "intrsafe" objects. 1904 */ 1905 /* 1906 * remove mappings from pmap and drop the pages 1907 * from the object. offsets are always relative 1908 * to vm_map_min(kernel_map). 1909 */ 1910 uvm_km_pgremove(entry->object.uvm_obj, entry->start, 1911 entry->end); 1912 } else { 1913 /* remove mappings the standard way. */ 1914 pmap_remove(map->pmap, entry->start, entry->end); 1915 } 1916 1917 if (needlock) 1918 uvm_map_unlock_entry(entry); 1919 } 1920 1921 void 1922 uvm_unmap_kill_entry(struct vm_map *map, struct vm_map_entry *entry) 1923 { 1924 uvm_unmap_kill_entry_withlock(map, entry, 0); 1925 } 1926 1927 /* 1928 * Remove all entries from start to end. 1929 * 1930 * If remove_holes, then remove ET_HOLE entries as well. 1931 * If markfree, entry will be properly marked free, otherwise, no replacement 1932 * entry will be put in the tree (corrupting the tree). 1933 */ 1934 int 1935 uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end, 1936 struct uvm_map_deadq *dead, boolean_t remove_holes, 1937 boolean_t markfree, boolean_t checkimmutable) 1938 { 1939 struct vm_map_entry *prev_hint, *next, *entry; 1940 1941 start = MAX(start, map->min_offset); 1942 end = MIN(end, map->max_offset); 1943 if (start >= end) 1944 return 0; 1945 1946 vm_map_assert_wrlock(map); 1947 1948 /* Find first affected entry. */ 1949 entry = uvm_map_entrybyaddr(&map->addr, start); 1950 KDASSERT(entry != NULL && entry->start <= start); 1951 1952 if (checkimmutable) { 1953 struct vm_map_entry *entry1 = entry; 1954 1955 /* Refuse to unmap if any entries are immutable */ 1956 if (entry1->end <= start) 1957 entry1 = RBT_NEXT(uvm_map_addr, entry1); 1958 for (; entry1 != NULL && entry1->start < end; entry1 = next) { 1959 KDASSERT(entry1->start >= start); 1960 next = RBT_NEXT(uvm_map_addr, entry1); 1961 /* Treat memory holes as free space. */ 1962 if (entry1->start == entry1->end || UVM_ET_ISHOLE(entry1)) 1963 continue; 1964 if (entry1->etype & UVM_ET_IMMUTABLE) 1965 return EPERM; 1966 } 1967 } 1968 1969 if (entry->end <= start && markfree) 1970 entry = RBT_NEXT(uvm_map_addr, entry); 1971 else 1972 UVM_MAP_CLIP_START(map, entry, start); 1973 1974 /* 1975 * Iterate entries until we reach end address. 1976 * prev_hint hints where the freed space can be appended to. 1977 */ 1978 prev_hint = NULL; 1979 for (; entry != NULL && entry->start < end; entry = next) { 1980 KDASSERT(entry->start >= start); 1981 if (entry->end > end || !markfree) 1982 UVM_MAP_CLIP_END(map, entry, end); 1983 KDASSERT(entry->start >= start && entry->end <= end); 1984 next = RBT_NEXT(uvm_map_addr, entry); 1985 1986 /* Don't remove holes unless asked to do so. */ 1987 if (UVM_ET_ISHOLE(entry)) { 1988 if (!remove_holes) { 1989 prev_hint = entry; 1990 continue; 1991 } 1992 } 1993 1994 /* A stack has been removed.. */ 1995 if (UVM_ET_ISSTACK(entry) && (map->flags & VM_MAP_ISVMSPACE)) 1996 map->sserial++; 1997 1998 /* Kill entry. */ 1999 uvm_unmap_kill_entry_withlock(map, entry, 1); 2000 2001 /* Update space usage. */ 2002 if ((map->flags & VM_MAP_ISVMSPACE) && 2003 entry->object.uvm_obj == NULL && 2004 entry->protection != PROT_NONE && 2005 !UVM_ET_ISHOLE(entry)) { 2006 ((struct vmspace *)map)->vm_dused -= 2007 uvmspace_dused(map, entry->start, entry->end); 2008 } 2009 if (!UVM_ET_ISHOLE(entry)) 2010 map->size -= entry->end - entry->start; 2011 2012 /* Actual removal of entry. */ 2013 uvm_mapent_mkfree(map, entry, &prev_hint, dead, markfree); 2014 } 2015 2016 pmap_update(vm_map_pmap(map)); 2017 2018 #ifdef VMMAP_DEBUG 2019 if (markfree) { 2020 for (entry = uvm_map_entrybyaddr(&map->addr, start); 2021 entry != NULL && entry->start < end; 2022 entry = RBT_NEXT(uvm_map_addr, entry)) { 2023 KDASSERT(entry->end <= start || 2024 entry->start == entry->end || 2025 UVM_ET_ISHOLE(entry)); 2026 } 2027 } else { 2028 vaddr_t a; 2029 for (a = start; a < end; a += PAGE_SIZE) 2030 KDASSERT(uvm_map_entrybyaddr(&map->addr, a) == NULL); 2031 } 2032 #endif 2033 return 0; 2034 } 2035 2036 /* 2037 * Mark all entries from first until end (exclusive) as pageable. 2038 * 2039 * Lock must be exclusive on entry and will not be touched. 2040 */ 2041 void 2042 uvm_map_pageable_pgon(struct vm_map *map, struct vm_map_entry *first, 2043 struct vm_map_entry *end, vaddr_t start_addr, vaddr_t end_addr) 2044 { 2045 struct vm_map_entry *iter; 2046 2047 for (iter = first; iter != end; 2048 iter = RBT_NEXT(uvm_map_addr, iter)) { 2049 KDASSERT(iter->start >= start_addr && iter->end <= end_addr); 2050 if (!VM_MAPENT_ISWIRED(iter) || UVM_ET_ISHOLE(iter)) 2051 continue; 2052 2053 iter->wired_count = 0; 2054 uvm_fault_unwire_locked(map, iter->start, iter->end); 2055 } 2056 } 2057 2058 /* 2059 * Mark all entries from first until end (exclusive) as wired. 2060 * 2061 * Lockflags determines the lock state on return from this function. 2062 * Lock must be exclusive on entry. 2063 */ 2064 int 2065 uvm_map_pageable_wire(struct vm_map *map, struct vm_map_entry *first, 2066 struct vm_map_entry *end, vaddr_t start_addr, vaddr_t end_addr, 2067 int lockflags) 2068 { 2069 struct vm_map_entry *iter; 2070 #ifdef DIAGNOSTIC 2071 unsigned int timestamp_save; 2072 #endif 2073 int error; 2074 2075 /* 2076 * Wire pages in two passes: 2077 * 2078 * 1: holding the write lock, we create any anonymous maps that need 2079 * to be created. then we clip each map entry to the region to 2080 * be wired and increment its wiring count. 2081 * 2082 * 2: we mark the map busy, unlock it and call uvm_fault_wire to fault 2083 * in the pages for any newly wired area (wired_count == 1). 2084 */ 2085 for (iter = first; iter != end; 2086 iter = RBT_NEXT(uvm_map_addr, iter)) { 2087 KDASSERT(iter->start >= start_addr && iter->end <= end_addr); 2088 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2089 iter->protection == PROT_NONE) 2090 continue; 2091 2092 /* 2093 * Perform actions of vm_map_lookup that need the write lock. 2094 * - create an anonymous map for copy-on-write 2095 * - anonymous map for zero-fill 2096 * Skip submaps. 2097 */ 2098 if (!VM_MAPENT_ISWIRED(iter) && !UVM_ET_ISSUBMAP(iter) && 2099 UVM_ET_ISNEEDSCOPY(iter) && 2100 ((iter->protection & PROT_WRITE) || 2101 iter->object.uvm_obj == NULL)) { 2102 amap_copy(map, iter, M_WAITOK, 2103 UVM_ET_ISSTACK(iter) ? FALSE : TRUE, 2104 iter->start, iter->end); 2105 } 2106 iter->wired_count++; 2107 } 2108 2109 /* 2110 * Pass 2. 2111 */ 2112 #ifdef DIAGNOSTIC 2113 timestamp_save = map->timestamp; 2114 #endif 2115 vm_map_busy(map); 2116 vm_map_unlock(map); 2117 2118 error = 0; 2119 for (iter = first; error == 0 && iter != end; 2120 iter = RBT_NEXT(uvm_map_addr, iter)) { 2121 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2122 iter->protection == PROT_NONE) 2123 continue; 2124 2125 error = uvm_fault_wire(map, iter->start, iter->end, 2126 iter->protection); 2127 } 2128 2129 vm_map_lock(map); 2130 vm_map_unbusy(map); 2131 2132 if (error) { 2133 #ifdef DIAGNOSTIC 2134 if (timestamp_save != map->timestamp) 2135 panic("uvm_map_pageable_wire: stale map"); 2136 #endif 2137 2138 /* 2139 * first is no longer needed to restart loops. 2140 * Use it as iterator to unmap successful mappings. 2141 */ 2142 for (; first != iter; 2143 first = RBT_NEXT(uvm_map_addr, first)) { 2144 if (UVM_ET_ISHOLE(first) || 2145 first->start == first->end || 2146 first->protection == PROT_NONE) 2147 continue; 2148 2149 first->wired_count--; 2150 if (!VM_MAPENT_ISWIRED(first)) { 2151 uvm_fault_unwire_locked(map, 2152 first->start, first->end); 2153 } 2154 } 2155 2156 /* decrease counter in the rest of the entries */ 2157 for (; iter != end; 2158 iter = RBT_NEXT(uvm_map_addr, iter)) { 2159 if (UVM_ET_ISHOLE(iter) || iter->start == iter->end || 2160 iter->protection == PROT_NONE) 2161 continue; 2162 2163 iter->wired_count--; 2164 } 2165 2166 if ((lockflags & UVM_LK_EXIT) == 0) 2167 vm_map_unlock(map); 2168 return error; 2169 } 2170 2171 2172 if ((lockflags & UVM_LK_EXIT) == 0) { 2173 vm_map_unlock(map); 2174 } else { 2175 #ifdef DIAGNOSTIC 2176 if (timestamp_save != map->timestamp) 2177 panic("uvm_map_pageable_wire: stale map"); 2178 #endif 2179 } 2180 return 0; 2181 } 2182 2183 /* 2184 * uvm_map_pageable: set pageability of a range in a map. 2185 * 2186 * Flags: 2187 * UVM_LK_ENTER: map is already locked by caller 2188 * UVM_LK_EXIT: don't unlock map on exit 2189 * 2190 * The full range must be in use (entries may not have fspace != 0). 2191 * UVM_ET_HOLE counts as unmapped. 2192 */ 2193 int 2194 uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end, 2195 boolean_t new_pageable, int lockflags) 2196 { 2197 struct vm_map_entry *first, *last, *tmp; 2198 int error; 2199 2200 start = trunc_page(start); 2201 end = round_page(end); 2202 2203 if (start > end) 2204 return EINVAL; 2205 if (start == end) 2206 return 0; /* nothing to do */ 2207 if (start < map->min_offset) 2208 return EFAULT; /* why? see first XXX below */ 2209 if (end > map->max_offset) 2210 return EINVAL; /* why? see second XXX below */ 2211 2212 KASSERT(map->flags & VM_MAP_PAGEABLE); 2213 if ((lockflags & UVM_LK_ENTER) == 0) 2214 vm_map_lock(map); 2215 2216 /* 2217 * Find first entry. 2218 * 2219 * Initial test on start is different, because of the different 2220 * error returned. Rest is tested further down. 2221 */ 2222 first = uvm_map_entrybyaddr(&map->addr, start); 2223 if (first->end <= start || UVM_ET_ISHOLE(first)) { 2224 /* 2225 * XXX if the first address is not mapped, it is EFAULT? 2226 */ 2227 error = EFAULT; 2228 goto out; 2229 } 2230 2231 /* Check that the range has no holes. */ 2232 for (last = first; last != NULL && last->start < end; 2233 last = RBT_NEXT(uvm_map_addr, last)) { 2234 if (UVM_ET_ISHOLE(last) || 2235 (last->end < end && VMMAP_FREE_END(last) != last->end)) { 2236 /* 2237 * XXX unmapped memory in range, why is it EINVAL 2238 * instead of EFAULT? 2239 */ 2240 error = EINVAL; 2241 goto out; 2242 } 2243 } 2244 2245 /* 2246 * Last ended at the first entry after the range. 2247 * Move back one step. 2248 * 2249 * Note that last may be NULL. 2250 */ 2251 if (last == NULL) { 2252 last = RBT_MAX(uvm_map_addr, &map->addr); 2253 if (last->end < end) { 2254 error = EINVAL; 2255 goto out; 2256 } 2257 } else { 2258 KASSERT(last != first); 2259 last = RBT_PREV(uvm_map_addr, last); 2260 } 2261 2262 /* Wire/unwire pages here. */ 2263 if (new_pageable) { 2264 /* 2265 * Mark pageable. 2266 * entries that are not wired are untouched. 2267 */ 2268 if (VM_MAPENT_ISWIRED(first)) 2269 UVM_MAP_CLIP_START(map, first, start); 2270 /* 2271 * Split last at end. 2272 * Make tmp be the first entry after what is to be touched. 2273 * If last is not wired, don't touch it. 2274 */ 2275 if (VM_MAPENT_ISWIRED(last)) { 2276 UVM_MAP_CLIP_END(map, last, end); 2277 tmp = RBT_NEXT(uvm_map_addr, last); 2278 } else 2279 tmp = last; 2280 2281 uvm_map_pageable_pgon(map, first, tmp, start, end); 2282 error = 0; 2283 2284 out: 2285 if ((lockflags & UVM_LK_EXIT) == 0) 2286 vm_map_unlock(map); 2287 return error; 2288 } else { 2289 /* 2290 * Mark entries wired. 2291 * entries are always touched (because recovery needs this). 2292 */ 2293 if (!VM_MAPENT_ISWIRED(first)) 2294 UVM_MAP_CLIP_START(map, first, start); 2295 /* 2296 * Split last at end. 2297 * Make tmp be the first entry after what is to be touched. 2298 * If last is not wired, don't touch it. 2299 */ 2300 if (!VM_MAPENT_ISWIRED(last)) { 2301 UVM_MAP_CLIP_END(map, last, end); 2302 tmp = RBT_NEXT(uvm_map_addr, last); 2303 } else 2304 tmp = last; 2305 2306 return uvm_map_pageable_wire(map, first, tmp, start, end, 2307 lockflags); 2308 } 2309 } 2310 2311 /* 2312 * uvm_map_pageable_all: special case of uvm_map_pageable - affects 2313 * all mapped regions. 2314 * 2315 * Map must not be locked. 2316 * If no flags are specified, all regions are unwired. 2317 */ 2318 int 2319 uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit) 2320 { 2321 vsize_t size; 2322 struct vm_map_entry *iter; 2323 2324 KASSERT(map->flags & VM_MAP_PAGEABLE); 2325 vm_map_lock(map); 2326 2327 if (flags == 0) { 2328 uvm_map_pageable_pgon(map, RBT_MIN(uvm_map_addr, &map->addr), 2329 NULL, map->min_offset, map->max_offset); 2330 2331 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE); 2332 vm_map_unlock(map); 2333 return 0; 2334 } 2335 2336 if (flags & MCL_FUTURE) 2337 vm_map_modflags(map, VM_MAP_WIREFUTURE, 0); 2338 if (!(flags & MCL_CURRENT)) { 2339 vm_map_unlock(map); 2340 return 0; 2341 } 2342 2343 /* 2344 * Count number of pages in all non-wired entries. 2345 * If the number exceeds the limit, abort. 2346 */ 2347 size = 0; 2348 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2349 if (VM_MAPENT_ISWIRED(iter) || UVM_ET_ISHOLE(iter)) 2350 continue; 2351 2352 size += iter->end - iter->start; 2353 } 2354 2355 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) { 2356 vm_map_unlock(map); 2357 return ENOMEM; 2358 } 2359 2360 /* XXX non-pmap_wired_count case must be handled by caller */ 2361 #ifdef pmap_wired_count 2362 if (limit != 0 && 2363 size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit) { 2364 vm_map_unlock(map); 2365 return ENOMEM; 2366 } 2367 #endif 2368 2369 /* 2370 * uvm_map_pageable_wire will release lock 2371 */ 2372 return uvm_map_pageable_wire(map, RBT_MIN(uvm_map_addr, &map->addr), 2373 NULL, map->min_offset, map->max_offset, 0); 2374 } 2375 2376 /* 2377 * Initialize map. 2378 * 2379 * Allocates sufficient entries to describe the free memory in the map. 2380 */ 2381 void 2382 uvm_map_setup(struct vm_map *map, pmap_t pmap, vaddr_t min, vaddr_t max, 2383 int flags) 2384 { 2385 int i; 2386 2387 KASSERT((min & (vaddr_t)PAGE_MASK) == 0); 2388 KASSERT((max & (vaddr_t)PAGE_MASK) == 0 || 2389 (max & (vaddr_t)PAGE_MASK) == (vaddr_t)PAGE_MASK); 2390 2391 /* 2392 * Update parameters. 2393 * 2394 * This code handles (vaddr_t)-1 and other page mask ending addresses 2395 * properly. 2396 * We lose the top page if the full virtual address space is used. 2397 */ 2398 if (max & (vaddr_t)PAGE_MASK) { 2399 max += 1; 2400 if (max == 0) /* overflow */ 2401 max -= PAGE_SIZE; 2402 } 2403 2404 RBT_INIT(uvm_map_addr, &map->addr); 2405 map->uaddr_exe = NULL; 2406 for (i = 0; i < nitems(map->uaddr_any); ++i) 2407 map->uaddr_any[i] = NULL; 2408 map->uaddr_brk_stack = NULL; 2409 2410 map->pmap = pmap; 2411 map->size = 0; 2412 map->ref_count = 0; 2413 map->min_offset = min; 2414 map->max_offset = max; 2415 map->b_start = map->b_end = 0; /* Empty brk() area by default. */ 2416 map->s_start = map->s_end = 0; /* Empty stack area by default. */ 2417 map->flags = flags; 2418 map->timestamp = 0; 2419 map->busy = NULL; 2420 if (flags & VM_MAP_ISVMSPACE) 2421 rw_init_flags(&map->lock, "vmmaplk", RWL_DUPOK); 2422 else 2423 rw_init(&map->lock, "kmmaplk"); 2424 mtx_init(&map->mtx, IPL_VM); 2425 mtx_init(&map->flags_lock, IPL_VM); 2426 2427 /* Configure the allocators. */ 2428 if (flags & VM_MAP_ISVMSPACE) 2429 uvm_map_setup_md(map); 2430 else 2431 map->uaddr_any[3] = &uaddr_kbootstrap; 2432 2433 /* 2434 * Fill map entries. 2435 * We do not need to write-lock the map here because only the current 2436 * thread sees it right now. Initialize ref_count to 0 above to avoid 2437 * bogus triggering of lock-not-held assertions. 2438 */ 2439 uvm_map_setup_entries(map); 2440 uvm_tree_sanity(map, __FILE__, __LINE__); 2441 map->ref_count = 1; 2442 } 2443 2444 /* 2445 * Destroy the map. 2446 * 2447 * This is the inverse operation to uvm_map_setup. 2448 */ 2449 void 2450 uvm_map_teardown(struct vm_map *map) 2451 { 2452 struct uvm_map_deadq dead_entries; 2453 struct vm_map_entry *entry, *tmp; 2454 #ifdef VMMAP_DEBUG 2455 size_t numq, numt; 2456 #endif 2457 int i; 2458 2459 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 2460 2461 vm_map_lock(map); 2462 2463 /* Remove address selectors. */ 2464 uvm_addr_destroy(map->uaddr_exe); 2465 map->uaddr_exe = NULL; 2466 for (i = 0; i < nitems(map->uaddr_any); i++) { 2467 uvm_addr_destroy(map->uaddr_any[i]); 2468 map->uaddr_any[i] = NULL; 2469 } 2470 uvm_addr_destroy(map->uaddr_brk_stack); 2471 map->uaddr_brk_stack = NULL; 2472 2473 /* 2474 * Remove entries. 2475 * 2476 * The following is based on graph breadth-first search. 2477 * 2478 * In color terms: 2479 * - the dead_entries set contains all nodes that are reachable 2480 * (i.e. both the black and the grey nodes) 2481 * - any entry not in dead_entries is white 2482 * - any entry that appears in dead_entries before entry, 2483 * is black, the rest is grey. 2484 * The set [entry, end] is also referred to as the wavefront. 2485 * 2486 * Since the tree is always a fully connected graph, the breadth-first 2487 * search guarantees that each vmmap_entry is visited exactly once. 2488 * The vm_map is broken down in linear time. 2489 */ 2490 TAILQ_INIT(&dead_entries); 2491 if ((entry = RBT_ROOT(uvm_map_addr, &map->addr)) != NULL) 2492 DEAD_ENTRY_PUSH(&dead_entries, entry); 2493 while (entry != NULL) { 2494 sched_pause(yield); 2495 uvm_unmap_kill_entry(map, entry); 2496 if ((tmp = RBT_LEFT(uvm_map_addr, entry)) != NULL) 2497 DEAD_ENTRY_PUSH(&dead_entries, tmp); 2498 if ((tmp = RBT_RIGHT(uvm_map_addr, entry)) != NULL) 2499 DEAD_ENTRY_PUSH(&dead_entries, tmp); 2500 /* Update wave-front. */ 2501 entry = TAILQ_NEXT(entry, dfree.deadq); 2502 } 2503 2504 vm_map_unlock(map); 2505 2506 #ifdef VMMAP_DEBUG 2507 numt = numq = 0; 2508 RBT_FOREACH(entry, uvm_map_addr, &map->addr) 2509 numt++; 2510 TAILQ_FOREACH(entry, &dead_entries, dfree.deadq) 2511 numq++; 2512 KASSERT(numt == numq); 2513 #endif 2514 uvm_unmap_detach(&dead_entries, 0); 2515 2516 pmap_destroy(map->pmap); 2517 map->pmap = NULL; 2518 } 2519 2520 /* 2521 * Populate map with free-memory entries. 2522 * 2523 * Map must be initialized and empty. 2524 */ 2525 void 2526 uvm_map_setup_entries(struct vm_map *map) 2527 { 2528 KDASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 2529 2530 uvm_map_fix_space(map, NULL, map->min_offset, map->max_offset, 0); 2531 } 2532 2533 /* 2534 * Split entry at given address. 2535 * 2536 * orig: entry that is to be split. 2537 * next: a newly allocated map entry that is not linked. 2538 * split: address at which the split is done. 2539 */ 2540 void 2541 uvm_map_splitentry(struct vm_map *map, struct vm_map_entry *orig, 2542 struct vm_map_entry *next, vaddr_t split) 2543 { 2544 struct uvm_addr_state *free, *free_before; 2545 vsize_t adj; 2546 2547 if ((split & PAGE_MASK) != 0) { 2548 panic("uvm_map_splitentry: split address 0x%lx " 2549 "not on page boundary!", split); 2550 } 2551 KDASSERT(map != NULL && orig != NULL && next != NULL); 2552 uvm_tree_sanity(map, __FILE__, __LINE__); 2553 KASSERT(orig->start < split && VMMAP_FREE_END(orig) > split); 2554 2555 #ifdef VMMAP_DEBUG 2556 KDASSERT(RBT_FIND(uvm_map_addr, &map->addr, orig) == orig); 2557 KDASSERT(RBT_FIND(uvm_map_addr, &map->addr, next) != next); 2558 #endif /* VMMAP_DEBUG */ 2559 2560 /* 2561 * Free space will change, unlink from free space tree. 2562 */ 2563 free = uvm_map_uaddr_e(map, orig); 2564 uvm_mapent_free_remove(map, free, orig); 2565 2566 adj = split - orig->start; 2567 2568 uvm_mapent_copy(orig, next); 2569 if (split >= orig->end) { 2570 next->etype = 0; 2571 next->offset = 0; 2572 next->wired_count = 0; 2573 next->start = next->end = split; 2574 next->guard = 0; 2575 next->fspace = VMMAP_FREE_END(orig) - split; 2576 next->aref.ar_amap = NULL; 2577 next->aref.ar_pageoff = 0; 2578 orig->guard = MIN(orig->guard, split - orig->end); 2579 orig->fspace = split - VMMAP_FREE_START(orig); 2580 } else { 2581 orig->fspace = 0; 2582 orig->guard = 0; 2583 orig->end = next->start = split; 2584 2585 if (next->aref.ar_amap) { 2586 amap_splitref(&orig->aref, &next->aref, adj); 2587 } 2588 if (UVM_ET_ISSUBMAP(orig)) { 2589 uvm_map_reference(next->object.sub_map); 2590 next->offset += adj; 2591 } else if (UVM_ET_ISOBJ(orig)) { 2592 if (next->object.uvm_obj->pgops && 2593 next->object.uvm_obj->pgops->pgo_reference) { 2594 KERNEL_LOCK(); 2595 next->object.uvm_obj->pgops->pgo_reference( 2596 next->object.uvm_obj); 2597 KERNEL_UNLOCK(); 2598 } 2599 next->offset += adj; 2600 } 2601 } 2602 2603 /* 2604 * Link next into address tree. 2605 * Link orig and next into free-space tree. 2606 * 2607 * Don't insert 'next' into the addr tree until orig has been linked, 2608 * in case the free-list looks at adjacent entries in the addr tree 2609 * for its decisions. 2610 */ 2611 if (orig->fspace > 0) 2612 free_before = free; 2613 else 2614 free_before = uvm_map_uaddr_e(map, orig); 2615 uvm_mapent_free_insert(map, free_before, orig); 2616 uvm_mapent_addr_insert(map, next); 2617 uvm_mapent_free_insert(map, free, next); 2618 2619 uvm_tree_sanity(map, __FILE__, __LINE__); 2620 } 2621 2622 2623 #ifdef VMMAP_DEBUG 2624 2625 void 2626 uvm_tree_assert(struct vm_map *map, int test, char *test_str, 2627 char *file, int line) 2628 { 2629 char* map_special; 2630 2631 if (test) 2632 return; 2633 2634 if (map == kernel_map) 2635 map_special = " (kernel_map)"; 2636 else if (map == kmem_map) 2637 map_special = " (kmem_map)"; 2638 else 2639 map_special = ""; 2640 panic("uvm_tree_sanity %p%s (%s %d): %s", map, map_special, file, 2641 line, test_str); 2642 } 2643 2644 /* 2645 * Check that map is sane. 2646 */ 2647 void 2648 uvm_tree_sanity(struct vm_map *map, char *file, int line) 2649 { 2650 struct vm_map_entry *iter; 2651 vaddr_t addr; 2652 vaddr_t min, max, bound; /* Bounds checker. */ 2653 struct uvm_addr_state *free; 2654 2655 addr = vm_map_min(map); 2656 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2657 /* 2658 * Valid start, end. 2659 * Catch overflow for end+fspace. 2660 */ 2661 UVM_ASSERT(map, iter->end >= iter->start, file, line); 2662 UVM_ASSERT(map, VMMAP_FREE_END(iter) >= iter->end, file, line); 2663 2664 /* May not be empty. */ 2665 UVM_ASSERT(map, iter->start < VMMAP_FREE_END(iter), 2666 file, line); 2667 2668 /* Addresses for entry must lie within map boundaries. */ 2669 UVM_ASSERT(map, iter->start >= vm_map_min(map) && 2670 VMMAP_FREE_END(iter) <= vm_map_max(map), file, line); 2671 2672 /* Tree may not have gaps. */ 2673 UVM_ASSERT(map, iter->start == addr, file, line); 2674 addr = VMMAP_FREE_END(iter); 2675 2676 /* 2677 * Free space may not cross boundaries, unless the same 2678 * free list is used on both sides of the border. 2679 */ 2680 min = VMMAP_FREE_START(iter); 2681 max = VMMAP_FREE_END(iter); 2682 2683 while (min < max && 2684 (bound = uvm_map_boundary(map, min, max)) != max) { 2685 UVM_ASSERT(map, 2686 uvm_map_uaddr(map, bound - 1) == 2687 uvm_map_uaddr(map, bound), 2688 file, line); 2689 min = bound; 2690 } 2691 2692 free = uvm_map_uaddr_e(map, iter); 2693 if (free) { 2694 UVM_ASSERT(map, (iter->etype & UVM_ET_FREEMAPPED) != 0, 2695 file, line); 2696 } else { 2697 UVM_ASSERT(map, (iter->etype & UVM_ET_FREEMAPPED) == 0, 2698 file, line); 2699 } 2700 } 2701 UVM_ASSERT(map, addr == vm_map_max(map), file, line); 2702 } 2703 2704 void 2705 uvm_tree_size_chk(struct vm_map *map, char *file, int line) 2706 { 2707 struct vm_map_entry *iter; 2708 vsize_t size; 2709 2710 size = 0; 2711 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2712 if (!UVM_ET_ISHOLE(iter)) 2713 size += iter->end - iter->start; 2714 } 2715 2716 if (map->size != size) 2717 printf("map size = 0x%lx, should be 0x%lx\n", map->size, size); 2718 UVM_ASSERT(map, map->size == size, file, line); 2719 2720 vmspace_validate(map); 2721 } 2722 2723 /* 2724 * This function validates the statistics on vmspace. 2725 */ 2726 void 2727 vmspace_validate(struct vm_map *map) 2728 { 2729 struct vmspace *vm; 2730 struct vm_map_entry *iter; 2731 vaddr_t imin, imax; 2732 vaddr_t stack_begin, stack_end; /* Position of stack. */ 2733 vsize_t stack, heap; /* Measured sizes. */ 2734 2735 if (!(map->flags & VM_MAP_ISVMSPACE)) 2736 return; 2737 2738 vm = (struct vmspace *)map; 2739 stack_begin = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 2740 stack_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 2741 2742 stack = heap = 0; 2743 RBT_FOREACH(iter, uvm_map_addr, &map->addr) { 2744 imin = imax = iter->start; 2745 2746 if (UVM_ET_ISHOLE(iter) || iter->object.uvm_obj != NULL || 2747 iter->protection != PROT_NONE) 2748 continue; 2749 2750 /* 2751 * Update stack, heap. 2752 * Keep in mind that (theoretically) the entries of 2753 * userspace and stack may be joined. 2754 */ 2755 while (imin != iter->end) { 2756 /* 2757 * Set imax to the first boundary crossed between 2758 * imin and stack addresses. 2759 */ 2760 imax = iter->end; 2761 if (imin < stack_begin && imax > stack_begin) 2762 imax = stack_begin; 2763 else if (imin < stack_end && imax > stack_end) 2764 imax = stack_end; 2765 2766 if (imin >= stack_begin && imin < stack_end) 2767 stack += imax - imin; 2768 else 2769 heap += imax - imin; 2770 imin = imax; 2771 } 2772 } 2773 2774 heap >>= PAGE_SHIFT; 2775 if (heap != vm->vm_dused) { 2776 printf("vmspace stack range: 0x%lx-0x%lx\n", 2777 stack_begin, stack_end); 2778 panic("vmspace_validate: vmspace.vm_dused invalid, " 2779 "expected %ld pgs, got %d pgs in map %p", 2780 heap, vm->vm_dused, 2781 map); 2782 } 2783 } 2784 2785 #endif /* VMMAP_DEBUG */ 2786 2787 /* 2788 * uvm_map_init: init mapping system at boot time. note that we allocate 2789 * and init the static pool of structs vm_map_entry for the kernel here. 2790 */ 2791 void 2792 uvm_map_init(void) 2793 { 2794 static struct vm_map_entry kernel_map_entry[MAX_KMAPENT]; 2795 int lcv; 2796 2797 /* now set up static pool of kernel map entries ... */ 2798 mtx_init(&uvm_kmapent_mtx, IPL_VM); 2799 SLIST_INIT(&uvm.kentry_free); 2800 for (lcv = 0 ; lcv < MAX_KMAPENT ; lcv++) { 2801 SLIST_INSERT_HEAD(&uvm.kentry_free, 2802 &kernel_map_entry[lcv], daddrs.addr_kentry); 2803 } 2804 2805 /* initialize the map-related pools. */ 2806 pool_init(&uvm_vmspace_pool, sizeof(struct vmspace), 0, 2807 IPL_NONE, PR_WAITOK, "vmsppl", NULL); 2808 pool_init(&uvm_map_entry_pool, sizeof(struct vm_map_entry), 0, 2809 IPL_VM, PR_WAITOK, "vmmpepl", NULL); 2810 pool_init(&uvm_map_entry_kmem_pool, sizeof(struct vm_map_entry), 0, 2811 IPL_VM, 0, "vmmpekpl", NULL); 2812 pool_sethiwat(&uvm_map_entry_pool, 8192); 2813 2814 uvm_addr_init(); 2815 } 2816 2817 #if defined(DDB) 2818 2819 /* 2820 * DDB hooks 2821 */ 2822 2823 /* 2824 * uvm_map_printit: actually prints the map 2825 */ 2826 void 2827 uvm_map_printit(struct vm_map *map, boolean_t full, 2828 int (*pr)(const char *, ...)) 2829 { 2830 struct vmspace *vm; 2831 struct vm_map_entry *entry; 2832 struct uvm_addr_state *free; 2833 int in_free, i; 2834 char buf[8]; 2835 2836 (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, map->min_offset,map->max_offset); 2837 (*pr)("\tbrk() allocate range: 0x%lx-0x%lx\n", 2838 map->b_start, map->b_end); 2839 (*pr)("\tstack allocate range: 0x%lx-0x%lx\n", 2840 map->s_start, map->s_end); 2841 (*pr)("\tsz=%u, ref=%d, version=%u, flags=0x%x\n", 2842 map->size, map->ref_count, map->timestamp, 2843 map->flags); 2844 (*pr)("\tpmap=%p(resident=%d)\n", map->pmap, 2845 pmap_resident_count(map->pmap)); 2846 2847 /* struct vmspace handling. */ 2848 if (map->flags & VM_MAP_ISVMSPACE) { 2849 vm = (struct vmspace *)map; 2850 2851 (*pr)("\tvm_refcnt=%d vm_shm=%p vm_rssize=%u vm_swrss=%u\n", 2852 vm->vm_refcnt, vm->vm_shm, vm->vm_rssize, vm->vm_swrss); 2853 (*pr)("\tvm_tsize=%u vm_dsize=%u\n", 2854 vm->vm_tsize, vm->vm_dsize); 2855 (*pr)("\tvm_taddr=%p vm_daddr=%p\n", 2856 vm->vm_taddr, vm->vm_daddr); 2857 (*pr)("\tvm_maxsaddr=%p vm_minsaddr=%p\n", 2858 vm->vm_maxsaddr, vm->vm_minsaddr); 2859 } 2860 2861 if (!full) 2862 goto print_uaddr; 2863 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 2864 (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n", 2865 entry, entry->start, entry->end, entry->object.uvm_obj, 2866 (long long)entry->offset, entry->aref.ar_amap, 2867 entry->aref.ar_pageoff); 2868 (*pr)("\tsubmap=%c, cow=%c, nc=%c, stack=%c, " 2869 "prot(max)=%d/%d, inh=%d, " 2870 "wc=%d, adv=%d\n", 2871 (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F', 2872 (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F', 2873 (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F', 2874 (entry->etype & UVM_ET_STACK) ? 'T' : 'F', 2875 entry->protection, entry->max_protection, 2876 entry->inheritance, entry->wired_count, entry->advice); 2877 2878 free = uvm_map_uaddr_e(map, entry); 2879 in_free = (free != NULL); 2880 (*pr)("\thole=%c, free=%c, guard=0x%lx, " 2881 "free=0x%lx-0x%lx\n", 2882 (entry->etype & UVM_ET_HOLE) ? 'T' : 'F', 2883 in_free ? 'T' : 'F', 2884 entry->guard, 2885 VMMAP_FREE_START(entry), VMMAP_FREE_END(entry)); 2886 (*pr)("\tfspace_augment=%lu\n", entry->fspace_augment); 2887 (*pr)("\tfreemapped=%c, uaddr=%p\n", 2888 (entry->etype & UVM_ET_FREEMAPPED) ? 'T' : 'F', free); 2889 if (free) { 2890 (*pr)("\t\t(0x%lx-0x%lx %s)\n", 2891 free->uaddr_minaddr, free->uaddr_maxaddr, 2892 free->uaddr_functions->uaddr_name); 2893 } 2894 } 2895 2896 print_uaddr: 2897 uvm_addr_print(map->uaddr_exe, "exe", full, pr); 2898 for (i = 0; i < nitems(map->uaddr_any); i++) { 2899 snprintf(&buf[0], sizeof(buf), "any[%d]", i); 2900 uvm_addr_print(map->uaddr_any[i], &buf[0], full, pr); 2901 } 2902 uvm_addr_print(map->uaddr_brk_stack, "brk/stack", full, pr); 2903 } 2904 2905 /* 2906 * uvm_object_printit: actually prints the object 2907 */ 2908 void 2909 uvm_object_printit(struct uvm_object *uobj, boolean_t full, 2910 int (*pr)(const char *, ...)) 2911 { 2912 struct vm_page *pg; 2913 int cnt = 0; 2914 2915 (*pr)("OBJECT %p: pgops=%p, npages=%d, ", 2916 uobj, uobj->pgops, uobj->uo_npages); 2917 if (UVM_OBJ_IS_KERN_OBJECT(uobj)) 2918 (*pr)("refs=<SYSTEM>\n"); 2919 else 2920 (*pr)("refs=%d\n", uobj->uo_refs); 2921 2922 if (!full) { 2923 return; 2924 } 2925 (*pr)(" PAGES <pg,offset>:\n "); 2926 RBT_FOREACH(pg, uvm_objtree, &uobj->memt) { 2927 (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset); 2928 if ((cnt % 3) == 2) { 2929 (*pr)("\n "); 2930 } 2931 cnt++; 2932 } 2933 if ((cnt % 3) != 2) { 2934 (*pr)("\n"); 2935 } 2936 } 2937 2938 /* 2939 * uvm_page_printit: actually print the page 2940 */ 2941 static const char page_flagbits[] = 2942 "\20\1BUSY\2WANTED\3TABLED\4CLEAN\5CLEANCHK\6RELEASED\7FAKE\10RDONLY" 2943 "\11ZERO\12DEV\15PAGER1\21FREE\22INACTIVE\23ACTIVE\25ANON\26AOBJ" 2944 "\27ENCRYPT\31PMAP0\32PMAP1\33PMAP2\34PMAP3\35PMAP4\36PMAP5"; 2945 2946 void 2947 uvm_page_printit(struct vm_page *pg, boolean_t full, 2948 int (*pr)(const char *, ...)) 2949 { 2950 struct vm_page *tpg; 2951 struct uvm_object *uobj; 2952 struct pglist *pgl; 2953 2954 (*pr)("PAGE %p:\n", pg); 2955 (*pr)(" flags=%b, vers=%d, wire_count=%d, pa=0x%llx\n", 2956 pg->pg_flags, page_flagbits, pg->pg_version, pg->wire_count, 2957 (long long)pg->phys_addr); 2958 (*pr)(" uobject=%p, uanon=%p, offset=0x%llx\n", 2959 pg->uobject, pg->uanon, (long long)pg->offset); 2960 #if defined(UVM_PAGE_TRKOWN) 2961 if (pg->pg_flags & PG_BUSY) 2962 (*pr)(" owning thread = %d, tag=%s", 2963 pg->owner, pg->owner_tag); 2964 else 2965 (*pr)(" page not busy, no owner"); 2966 #else 2967 (*pr)(" [page ownership tracking disabled]"); 2968 #endif 2969 (*pr)("\tvm_page_md %p\n", &pg->mdpage); 2970 2971 if (!full) 2972 return; 2973 2974 /* cross-verify object/anon */ 2975 if ((pg->pg_flags & PQ_FREE) == 0) { 2976 if (pg->pg_flags & PQ_ANON) { 2977 if (pg->uanon == NULL || pg->uanon->an_page != pg) 2978 (*pr)(" >>> ANON DOES NOT POINT HERE <<< (%p)\n", 2979 (pg->uanon) ? pg->uanon->an_page : NULL); 2980 else 2981 (*pr)(" anon backpointer is OK\n"); 2982 } else { 2983 uobj = pg->uobject; 2984 if (uobj) { 2985 (*pr)(" checking object list\n"); 2986 RBT_FOREACH(tpg, uvm_objtree, &uobj->memt) { 2987 if (tpg == pg) { 2988 break; 2989 } 2990 } 2991 if (tpg) 2992 (*pr)(" page found on object list\n"); 2993 else 2994 (*pr)(" >>> PAGE NOT FOUND " 2995 "ON OBJECT LIST! <<<\n"); 2996 } 2997 } 2998 } 2999 3000 /* cross-verify page queue */ 3001 if (pg->pg_flags & PQ_FREE) { 3002 if (uvm_pmr_isfree(pg)) 3003 (*pr)(" page found in uvm_pmemrange\n"); 3004 else 3005 (*pr)(" >>> page not found in uvm_pmemrange <<<\n"); 3006 pgl = NULL; 3007 } else if (pg->pg_flags & PQ_INACTIVE) { 3008 pgl = &uvm.page_inactive; 3009 } else if (pg->pg_flags & PQ_ACTIVE) { 3010 pgl = &uvm.page_active; 3011 } else { 3012 pgl = NULL; 3013 } 3014 3015 if (pgl) { 3016 (*pr)(" checking pageq list\n"); 3017 TAILQ_FOREACH(tpg, pgl, pageq) { 3018 if (tpg == pg) { 3019 break; 3020 } 3021 } 3022 if (tpg) 3023 (*pr)(" page found on pageq list\n"); 3024 else 3025 (*pr)(" >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n"); 3026 } 3027 } 3028 #endif 3029 3030 /* 3031 * uvm_map_protect: change map protection 3032 * 3033 * => set_max means set max_protection. 3034 * => map must be unlocked. 3035 */ 3036 int 3037 uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end, 3038 vm_prot_t new_prot, int etype, boolean_t set_max, boolean_t checkimmutable) 3039 { 3040 struct vm_map_entry *first, *iter; 3041 vm_prot_t old_prot; 3042 vm_prot_t mask; 3043 vsize_t dused; 3044 int error; 3045 3046 KASSERT((etype & ~UVM_ET_STACK) == 0); /* only UVM_ET_STACK allowed */ 3047 3048 if (start > end) 3049 return EINVAL; 3050 start = MAX(start, map->min_offset); 3051 end = MIN(end, map->max_offset); 3052 if (start >= end) 3053 return 0; 3054 3055 dused = 0; 3056 error = 0; 3057 vm_map_lock(map); 3058 3059 /* 3060 * Set up first and last. 3061 * - first will contain first entry at or after start. 3062 */ 3063 first = uvm_map_entrybyaddr(&map->addr, start); 3064 KDASSERT(first != NULL); 3065 if (first->end <= start) 3066 first = RBT_NEXT(uvm_map_addr, first); 3067 3068 /* First, check for protection violations. */ 3069 for (iter = first; iter != NULL && iter->start < end; 3070 iter = RBT_NEXT(uvm_map_addr, iter)) { 3071 /* Treat memory holes as free space. */ 3072 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) 3073 continue; 3074 3075 if (checkimmutable && (iter->etype & UVM_ET_IMMUTABLE)) { 3076 error = EPERM; 3077 goto out; 3078 } 3079 old_prot = iter->protection; 3080 if (old_prot == PROT_NONE && new_prot != old_prot) { 3081 dused += uvmspace_dused( 3082 map, MAX(start, iter->start), MIN(end, iter->end)); 3083 } 3084 3085 if (UVM_ET_ISSUBMAP(iter)) { 3086 error = EINVAL; 3087 goto out; 3088 } 3089 if ((new_prot & iter->max_protection) != new_prot) { 3090 error = EACCES; 3091 goto out; 3092 } 3093 if (map == kernel_map && 3094 (new_prot & (PROT_WRITE | PROT_EXEC)) == (PROT_WRITE | PROT_EXEC)) 3095 panic("uvm_map_protect: kernel map W^X violation requested"); 3096 } 3097 3098 /* Check limits. */ 3099 if (dused > 0 && (map->flags & VM_MAP_ISVMSPACE)) { 3100 vsize_t limit = lim_cur(RLIMIT_DATA); 3101 dused = ptoa(dused); 3102 if (limit < dused || 3103 limit - dused < ptoa(((struct vmspace *)map)->vm_dused)) { 3104 error = ENOMEM; 3105 goto out; 3106 } 3107 } 3108 3109 /* only apply UVM_ET_STACK on a mapping changing to RW */ 3110 if (etype && new_prot != (PROT_READ|PROT_WRITE)) 3111 etype = 0; 3112 3113 /* Fix protections. */ 3114 for (iter = first; iter != NULL && iter->start < end; 3115 iter = RBT_NEXT(uvm_map_addr, iter)) { 3116 /* Treat memory holes as free space. */ 3117 if (iter->start == iter->end || UVM_ET_ISHOLE(iter)) 3118 continue; 3119 3120 old_prot = iter->protection; 3121 3122 /* 3123 * Skip adapting protection iff old and new protection 3124 * are equal. 3125 */ 3126 if (set_max) { 3127 if (old_prot == (new_prot & old_prot) && 3128 iter->max_protection == new_prot) 3129 continue; 3130 } else { 3131 if (old_prot == new_prot) 3132 continue; 3133 } 3134 3135 UVM_MAP_CLIP_START(map, iter, start); 3136 UVM_MAP_CLIP_END(map, iter, end); 3137 3138 if (set_max) { 3139 iter->max_protection = new_prot; 3140 iter->protection &= new_prot; 3141 } else 3142 iter->protection = new_prot; 3143 iter->etype |= etype; /* potentially add UVM_ET_STACK */ 3144 3145 /* 3146 * update physical map if necessary. worry about copy-on-write 3147 * here -- CHECK THIS XXX 3148 */ 3149 if (iter->protection != old_prot) { 3150 mask = UVM_ET_ISCOPYONWRITE(iter) ? 3151 ~PROT_WRITE : PROT_MASK; 3152 3153 if (map->flags & VM_MAP_ISVMSPACE) { 3154 if (old_prot == PROT_NONE) { 3155 ((struct vmspace *)map)->vm_dused += 3156 uvmspace_dused(map, iter->start, 3157 iter->end); 3158 } 3159 if (iter->protection == PROT_NONE) { 3160 ((struct vmspace *)map)->vm_dused -= 3161 uvmspace_dused(map, iter->start, 3162 iter->end); 3163 } 3164 } 3165 3166 /* update pmap */ 3167 if ((iter->protection & mask) == PROT_NONE && 3168 VM_MAPENT_ISWIRED(iter)) { 3169 /* 3170 * TODO(ariane) this is stupid. wired_count 3171 * is 0 if not wired, otherwise anything 3172 * larger than 0 (incremented once each time 3173 * wire is called). 3174 * Mostly to be able to undo the damage on 3175 * failure. Not the actually be a wired 3176 * refcounter... 3177 * Originally: iter->wired_count--; 3178 * (don't we have to unwire this in the pmap 3179 * as well?) 3180 */ 3181 iter->wired_count = 0; 3182 } 3183 uvm_map_lock_entry(iter); 3184 pmap_protect(map->pmap, iter->start, iter->end, 3185 iter->protection & mask); 3186 uvm_map_unlock_entry(iter); 3187 } 3188 3189 /* 3190 * If the map is configured to lock any future mappings, 3191 * wire this entry now if the old protection was PROT_NONE 3192 * and the new protection is not PROT_NONE. 3193 */ 3194 if ((map->flags & VM_MAP_WIREFUTURE) != 0 && 3195 VM_MAPENT_ISWIRED(iter) == 0 && 3196 old_prot == PROT_NONE && 3197 new_prot != PROT_NONE) { 3198 if (uvm_map_pageable(map, iter->start, iter->end, 3199 FALSE, UVM_LK_ENTER | UVM_LK_EXIT) != 0) { 3200 /* 3201 * If locking the entry fails, remember the 3202 * error if it's the first one. Note we 3203 * still continue setting the protection in 3204 * the map, but it will return the resource 3205 * storage condition regardless. 3206 * 3207 * XXX Ignore what the actual error is, 3208 * XXX just call it a resource shortage 3209 * XXX so that it doesn't get confused 3210 * XXX what uvm_map_protect() itself would 3211 * XXX normally return. 3212 */ 3213 error = ENOMEM; 3214 } 3215 } 3216 } 3217 pmap_update(map->pmap); 3218 3219 out: 3220 if (etype & UVM_ET_STACK) 3221 map->sserial++; 3222 vm_map_unlock(map); 3223 return error; 3224 } 3225 3226 /* 3227 * uvmspace_alloc: allocate a vmspace structure. 3228 * 3229 * - structure includes vm_map and pmap 3230 * - XXX: no locking on this structure 3231 * - refcnt set to 1, rest must be init'd by caller 3232 */ 3233 struct vmspace * 3234 uvmspace_alloc(vaddr_t min, vaddr_t max, boolean_t pageable, 3235 boolean_t remove_holes) 3236 { 3237 struct vmspace *vm; 3238 3239 vm = pool_get(&uvm_vmspace_pool, PR_WAITOK | PR_ZERO); 3240 uvmspace_init(vm, NULL, min, max, pageable, remove_holes); 3241 return (vm); 3242 } 3243 3244 /* 3245 * uvmspace_init: initialize a vmspace structure. 3246 * 3247 * - XXX: no locking on this structure 3248 * - refcnt set to 1, rest must be init'd by caller 3249 */ 3250 void 3251 uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t min, vaddr_t max, 3252 boolean_t pageable, boolean_t remove_holes) 3253 { 3254 KASSERT(pmap == NULL || pmap == pmap_kernel()); 3255 3256 if (pmap) 3257 pmap_reference(pmap); 3258 else 3259 pmap = pmap_create(); 3260 3261 uvm_map_setup(&vm->vm_map, pmap, min, max, 3262 (pageable ? VM_MAP_PAGEABLE : 0) | VM_MAP_ISVMSPACE); 3263 3264 vm->vm_refcnt = 1; 3265 3266 if (remove_holes) 3267 pmap_remove_holes(vm); 3268 } 3269 3270 /* 3271 * uvmspace_share: share a vmspace between two processes 3272 * 3273 * - used for vfork 3274 */ 3275 3276 struct vmspace * 3277 uvmspace_share(struct process *pr) 3278 { 3279 struct vmspace *vm = pr->ps_vmspace; 3280 3281 uvmspace_addref(vm); 3282 return vm; 3283 } 3284 3285 /* 3286 * uvmspace_exec: the process wants to exec a new program 3287 * 3288 * - XXX: no locking on vmspace 3289 */ 3290 3291 void 3292 uvmspace_exec(struct proc *p, vaddr_t start, vaddr_t end) 3293 { 3294 struct process *pr = p->p_p; 3295 struct vmspace *nvm, *ovm = pr->ps_vmspace; 3296 struct vm_map *map = &ovm->vm_map; 3297 struct uvm_map_deadq dead_entries; 3298 3299 KASSERT((start & (vaddr_t)PAGE_MASK) == 0); 3300 KASSERT((end & (vaddr_t)PAGE_MASK) == 0 || 3301 (end & (vaddr_t)PAGE_MASK) == (vaddr_t)PAGE_MASK); 3302 3303 pmap_unuse_final(p); /* before stack addresses go away */ 3304 TAILQ_INIT(&dead_entries); 3305 3306 /* see if more than one process is using this vmspace... */ 3307 if (ovm->vm_refcnt == 1) { 3308 /* 3309 * If pr is the only process using its vmspace then 3310 * we can safely recycle that vmspace for the program 3311 * that is being exec'd. 3312 */ 3313 3314 #ifdef SYSVSHM 3315 /* 3316 * SYSV SHM semantics require us to kill all segments on an exec 3317 */ 3318 if (ovm->vm_shm) 3319 shmexit(ovm); 3320 #endif 3321 3322 /* 3323 * POSIX 1003.1b -- "lock future mappings" is revoked 3324 * when a process execs another program image. 3325 */ 3326 vm_map_lock(map); 3327 vm_map_modflags(map, 0, VM_MAP_WIREFUTURE | 3328 VM_MAP_PINSYSCALL_ONCE); 3329 3330 /* 3331 * now unmap the old program 3332 * 3333 * Instead of attempting to keep the map valid, we simply 3334 * nuke all entries and ask uvm_map_setup to reinitialize 3335 * the map to the new boundaries. 3336 * 3337 * uvm_unmap_remove will actually nuke all entries for us 3338 * (as in, not replace them with free-memory entries). 3339 */ 3340 uvm_unmap_remove(map, map->min_offset, map->max_offset, 3341 &dead_entries, TRUE, FALSE, FALSE); 3342 3343 KDASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 3344 3345 /* Nuke statistics and boundaries. */ 3346 memset(&ovm->vm_startcopy, 0, 3347 (caddr_t) (ovm + 1) - (caddr_t) &ovm->vm_startcopy); 3348 3349 3350 if (end & (vaddr_t)PAGE_MASK) { 3351 end += 1; 3352 if (end == 0) /* overflow */ 3353 end -= PAGE_SIZE; 3354 } 3355 3356 /* Setup new boundaries and populate map with entries. */ 3357 map->min_offset = start; 3358 map->max_offset = end; 3359 uvm_map_setup_entries(map); 3360 vm_map_unlock(map); 3361 3362 /* but keep MMU holes unavailable */ 3363 pmap_remove_holes(ovm); 3364 } else { 3365 /* 3366 * pr's vmspace is being shared, so we can't reuse 3367 * it for pr since it is still being used for others. 3368 * allocate a new vmspace for pr 3369 */ 3370 nvm = uvmspace_alloc(start, end, 3371 (map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, TRUE); 3372 3373 /* install new vmspace and drop our ref to the old one. */ 3374 pmap_deactivate(p); 3375 p->p_vmspace = pr->ps_vmspace = nvm; 3376 pmap_activate(p); 3377 3378 uvmspace_free(ovm); 3379 } 3380 #ifdef PMAP_CHECK_COPYIN 3381 p->p_vmspace->vm_map.check_copyin_count = 0; /* disable checks */ 3382 #endif 3383 3384 /* Release dead entries */ 3385 uvm_unmap_detach(&dead_entries, 0); 3386 } 3387 3388 /* 3389 * uvmspace_addref: add a reference to a vmspace. 3390 */ 3391 void 3392 uvmspace_addref(struct vmspace *vm) 3393 { 3394 KASSERT(vm->vm_refcnt > 0); 3395 atomic_inc_int(&vm->vm_refcnt); 3396 } 3397 3398 /* 3399 * uvmspace_free: free a vmspace data structure 3400 */ 3401 void 3402 uvmspace_free(struct vmspace *vm) 3403 { 3404 if (atomic_dec_int_nv(&vm->vm_refcnt) == 0) { 3405 /* 3406 * lock the map, to wait out all other references to it. delete 3407 * all of the mappings and pages they hold, then call the pmap 3408 * module to reclaim anything left. 3409 */ 3410 #ifdef SYSVSHM 3411 /* Get rid of any SYSV shared memory segments. */ 3412 if (vm->vm_shm != NULL) { 3413 KERNEL_LOCK(); 3414 shmexit(vm); 3415 KERNEL_UNLOCK(); 3416 } 3417 #endif 3418 3419 uvm_map_teardown(&vm->vm_map); 3420 pool_put(&uvm_vmspace_pool, vm); 3421 } 3422 } 3423 3424 /* 3425 * uvm_share: Map the address range [srcaddr, srcaddr + sz) in 3426 * srcmap to the address range [dstaddr, dstaddr + sz) in 3427 * dstmap. 3428 * 3429 * The whole address range in srcmap must be backed by an object 3430 * (no holes). 3431 * 3432 * If successful, the address ranges share memory and the destination 3433 * address range uses the protection flags in prot. 3434 * 3435 * This routine assumes that sz is a multiple of PAGE_SIZE and 3436 * that dstaddr and srcaddr are page-aligned. 3437 */ 3438 int 3439 uvm_share(struct vm_map *dstmap, vaddr_t dstaddr, vm_prot_t prot, 3440 struct vm_map *srcmap, vaddr_t srcaddr, vsize_t sz) 3441 { 3442 int ret = 0; 3443 vaddr_t unmap_end; 3444 vaddr_t dstva; 3445 vsize_t s_off, len, n = sz, remain; 3446 struct vm_map_entry *first = NULL, *last = NULL; 3447 struct vm_map_entry *src_entry, *psrc_entry = NULL; 3448 struct uvm_map_deadq dead; 3449 3450 if (srcaddr >= srcmap->max_offset || sz > srcmap->max_offset - srcaddr) 3451 return EINVAL; 3452 3453 TAILQ_INIT(&dead); 3454 vm_map_lock(dstmap); 3455 vm_map_lock_read(srcmap); 3456 3457 if (!uvm_map_isavail(dstmap, NULL, &first, &last, dstaddr, sz)) { 3458 ret = ENOMEM; 3459 goto exit_unlock; 3460 } 3461 if (!uvm_map_lookup_entry(srcmap, srcaddr, &src_entry)) { 3462 ret = EINVAL; 3463 goto exit_unlock; 3464 } 3465 3466 dstva = dstaddr; 3467 unmap_end = dstaddr; 3468 for (; src_entry != NULL; 3469 psrc_entry = src_entry, 3470 src_entry = RBT_NEXT(uvm_map_addr, src_entry)) { 3471 /* hole in address space, bail out */ 3472 if (psrc_entry != NULL && psrc_entry->end != src_entry->start) 3473 break; 3474 if (src_entry->start >= srcaddr + sz) 3475 break; 3476 3477 if (UVM_ET_ISSUBMAP(src_entry)) 3478 panic("uvm_share: encountered a submap (illegal)"); 3479 if (!UVM_ET_ISCOPYONWRITE(src_entry) && 3480 UVM_ET_ISNEEDSCOPY(src_entry)) 3481 panic("uvm_share: non-copy_on_write map entries " 3482 "marked needs_copy (illegal)"); 3483 3484 /* 3485 * srcaddr > map entry start? means we are in the middle of a 3486 * map, so we calculate the offset to use in the source map. 3487 */ 3488 if (srcaddr > src_entry->start) 3489 s_off = srcaddr - src_entry->start; 3490 else if (srcaddr == src_entry->start) 3491 s_off = 0; 3492 else 3493 panic("uvm_share: map entry start > srcaddr"); 3494 3495 remain = src_entry->end - src_entry->start - s_off; 3496 3497 /* Determine how many bytes to share in this pass */ 3498 if (n < remain) 3499 len = n; 3500 else 3501 len = remain; 3502 3503 if (uvm_mapent_share(dstmap, dstva, len, s_off, prot, prot, 3504 srcmap, src_entry, &dead) == NULL) 3505 break; 3506 3507 n -= len; 3508 dstva += len; 3509 srcaddr += len; 3510 unmap_end = dstva + len; 3511 if (n == 0) 3512 goto exit_unlock; 3513 } 3514 3515 ret = EINVAL; 3516 uvm_unmap_remove(dstmap, dstaddr, unmap_end, &dead, FALSE, TRUE, FALSE); 3517 3518 exit_unlock: 3519 vm_map_unlock_read(srcmap); 3520 vm_map_unlock(dstmap); 3521 uvm_unmap_detach(&dead, 0); 3522 3523 return ret; 3524 } 3525 3526 /* 3527 * Clone map entry into other map. 3528 * 3529 * Mapping will be placed at dstaddr, for the same length. 3530 * Space must be available. 3531 * Reference counters are incremented. 3532 */ 3533 struct vm_map_entry * 3534 uvm_mapent_clone(struct vm_map *dstmap, vaddr_t dstaddr, vsize_t dstlen, 3535 vsize_t off, vm_prot_t prot, vm_prot_t maxprot, 3536 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead, 3537 int mapent_flags, int amap_share_flags) 3538 { 3539 struct vm_map_entry *new_entry, *first, *last; 3540 3541 KDASSERT(!UVM_ET_ISSUBMAP(old_entry)); 3542 3543 /* Create new entry (linked in on creation). Fill in first, last. */ 3544 first = last = NULL; 3545 if (!uvm_map_isavail(dstmap, NULL, &first, &last, dstaddr, dstlen)) { 3546 panic("uvm_mapent_clone: no space in map for " 3547 "entry in empty map"); 3548 } 3549 new_entry = uvm_map_mkentry(dstmap, first, last, 3550 dstaddr, dstlen, mapent_flags, dead, NULL); 3551 if (new_entry == NULL) 3552 return NULL; 3553 /* old_entry -> new_entry */ 3554 new_entry->object = old_entry->object; 3555 new_entry->offset = old_entry->offset; 3556 new_entry->aref = old_entry->aref; 3557 new_entry->etype |= old_entry->etype & ~UVM_ET_FREEMAPPED; 3558 new_entry->protection = prot; 3559 new_entry->max_protection = maxprot; 3560 new_entry->inheritance = old_entry->inheritance; 3561 new_entry->advice = old_entry->advice; 3562 3563 /* gain reference to object backing the map (can't be a submap). */ 3564 if (new_entry->aref.ar_amap) { 3565 new_entry->aref.ar_pageoff += off >> PAGE_SHIFT; 3566 amap_ref(new_entry->aref.ar_amap, new_entry->aref.ar_pageoff, 3567 (new_entry->end - new_entry->start) >> PAGE_SHIFT, 3568 amap_share_flags); 3569 } 3570 3571 if (UVM_ET_ISOBJ(new_entry) && 3572 new_entry->object.uvm_obj->pgops->pgo_reference) { 3573 new_entry->offset += off; 3574 new_entry->object.uvm_obj->pgops->pgo_reference 3575 (new_entry->object.uvm_obj); 3576 } 3577 3578 return new_entry; 3579 } 3580 3581 struct vm_map_entry * 3582 uvm_mapent_share(struct vm_map *dstmap, vaddr_t dstaddr, vsize_t dstlen, 3583 vsize_t off, vm_prot_t prot, vm_prot_t maxprot, struct vm_map *old_map, 3584 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3585 { 3586 /* 3587 * If old_entry refers to a copy-on-write region that has not yet been 3588 * written to (needs_copy flag is set), then we need to allocate a new 3589 * amap for old_entry. 3590 * 3591 * If we do not do this, and the process owning old_entry does a copy-on 3592 * write later, old_entry and new_entry will refer to different memory 3593 * regions, and the memory between the processes is no longer shared. 3594 * 3595 * [in other words, we need to clear needs_copy] 3596 */ 3597 3598 if (UVM_ET_ISNEEDSCOPY(old_entry)) { 3599 /* get our own amap, clears needs_copy */ 3600 amap_copy(old_map, old_entry, M_WAITOK, FALSE, 0, 0); 3601 /* XXXCDC: WAITOK??? */ 3602 } 3603 3604 return uvm_mapent_clone(dstmap, dstaddr, dstlen, off, 3605 prot, maxprot, old_entry, dead, 0, AMAP_SHARED); 3606 } 3607 3608 /* 3609 * share the mapping: this means we want the old and 3610 * new entries to share amaps and backing objects. 3611 */ 3612 struct vm_map_entry * 3613 uvm_mapent_forkshared(struct vmspace *new_vm, struct vm_map *new_map, 3614 struct vm_map *old_map, 3615 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3616 { 3617 struct vm_map_entry *new_entry; 3618 3619 new_entry = uvm_mapent_share(new_map, old_entry->start, 3620 old_entry->end - old_entry->start, 0, old_entry->protection, 3621 old_entry->max_protection, old_map, old_entry, dead); 3622 3623 return (new_entry); 3624 } 3625 3626 /* 3627 * copy-on-write the mapping (using mmap's 3628 * MAP_PRIVATE semantics) 3629 * 3630 * allocate new_entry, adjust reference counts. 3631 * (note that new references are read-only). 3632 */ 3633 struct vm_map_entry * 3634 uvm_mapent_forkcopy(struct vmspace *new_vm, struct vm_map *new_map, 3635 struct vm_map *old_map, 3636 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3637 { 3638 struct vm_map_entry *new_entry; 3639 boolean_t protect_child; 3640 3641 new_entry = uvm_mapent_clone(new_map, old_entry->start, 3642 old_entry->end - old_entry->start, 0, old_entry->protection, 3643 old_entry->max_protection, old_entry, dead, 0, 0); 3644 3645 new_entry->etype |= 3646 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3647 3648 /* 3649 * the new entry will need an amap. it will either 3650 * need to be copied from the old entry or created 3651 * from scratch (if the old entry does not have an 3652 * amap). can we defer this process until later 3653 * (by setting "needs_copy") or do we need to copy 3654 * the amap now? 3655 * 3656 * we must copy the amap now if any of the following 3657 * conditions hold: 3658 * 1. the old entry has an amap and that amap is 3659 * being shared. this means that the old (parent) 3660 * process is sharing the amap with another 3661 * process. if we do not clear needs_copy here 3662 * we will end up in a situation where both the 3663 * parent and child process are referring to the 3664 * same amap with "needs_copy" set. if the 3665 * parent write-faults, the fault routine will 3666 * clear "needs_copy" in the parent by allocating 3667 * a new amap. this is wrong because the 3668 * parent is supposed to be sharing the old amap 3669 * and the new amap will break that. 3670 * 3671 * 2. if the old entry has an amap and a non-zero 3672 * wire count then we are going to have to call 3673 * amap_cow_now to avoid page faults in the 3674 * parent process. since amap_cow_now requires 3675 * "needs_copy" to be clear we might as well 3676 * clear it here as well. 3677 * 3678 */ 3679 if (old_entry->aref.ar_amap != NULL && 3680 ((amap_flags(old_entry->aref.ar_amap) & 3681 AMAP_SHARED) != 0 || 3682 VM_MAPENT_ISWIRED(old_entry))) { 3683 amap_copy(new_map, new_entry, M_WAITOK, FALSE, 3684 0, 0); 3685 /* XXXCDC: M_WAITOK ... ok? */ 3686 } 3687 3688 /* 3689 * if the parent's entry is wired down, then the 3690 * parent process does not want page faults on 3691 * access to that memory. this means that we 3692 * cannot do copy-on-write because we can't write 3693 * protect the old entry. in this case we 3694 * resolve all copy-on-write faults now, using 3695 * amap_cow_now. note that we have already 3696 * allocated any needed amap (above). 3697 */ 3698 if (VM_MAPENT_ISWIRED(old_entry)) { 3699 /* 3700 * resolve all copy-on-write faults now 3701 * (note that there is nothing to do if 3702 * the old mapping does not have an amap). 3703 */ 3704 if (old_entry->aref.ar_amap) 3705 amap_cow_now(new_map, new_entry); 3706 } else { 3707 if (old_entry->aref.ar_amap) { 3708 /* 3709 * setup mappings to trigger copy-on-write faults 3710 * we must write-protect the parent if it has 3711 * an amap and it is not already "needs_copy"... 3712 * if it is already "needs_copy" then the parent 3713 * has already been write-protected by a previous 3714 * fork operation. 3715 * 3716 * if we do not write-protect the parent, then 3717 * we must be sure to write-protect the child. 3718 */ 3719 if (!UVM_ET_ISNEEDSCOPY(old_entry)) { 3720 if (old_entry->max_protection & PROT_WRITE) { 3721 uvm_map_lock_entry(old_entry); 3722 pmap_protect(old_map->pmap, 3723 old_entry->start, 3724 old_entry->end, 3725 old_entry->protection & 3726 ~PROT_WRITE); 3727 uvm_map_unlock_entry(old_entry); 3728 pmap_update(old_map->pmap); 3729 } 3730 old_entry->etype |= UVM_ET_NEEDSCOPY; 3731 } 3732 3733 /* parent must now be write-protected */ 3734 protect_child = FALSE; 3735 } else { 3736 /* 3737 * we only need to protect the child if the 3738 * parent has write access. 3739 */ 3740 if (old_entry->max_protection & PROT_WRITE) 3741 protect_child = TRUE; 3742 else 3743 protect_child = FALSE; 3744 } 3745 3746 /* protect the child's mappings if necessary */ 3747 if (protect_child) { 3748 pmap_protect(new_map->pmap, new_entry->start, 3749 new_entry->end, 3750 new_entry->protection & 3751 ~PROT_WRITE); 3752 } 3753 } 3754 3755 return (new_entry); 3756 } 3757 3758 /* 3759 * zero the mapping: the new entry will be zero initialized 3760 */ 3761 struct vm_map_entry * 3762 uvm_mapent_forkzero(struct vmspace *new_vm, struct vm_map *new_map, 3763 struct vm_map *old_map, 3764 struct vm_map_entry *old_entry, struct uvm_map_deadq *dead) 3765 { 3766 struct vm_map_entry *new_entry; 3767 3768 new_entry = uvm_mapent_clone(new_map, old_entry->start, 3769 old_entry->end - old_entry->start, 0, old_entry->protection, 3770 old_entry->max_protection, old_entry, dead, 0, 0); 3771 3772 new_entry->etype |= 3773 (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY); 3774 3775 if (new_entry->aref.ar_amap) { 3776 amap_unref(new_entry->aref.ar_amap, new_entry->aref.ar_pageoff, 3777 atop(new_entry->end - new_entry->start), 0); 3778 new_entry->aref.ar_amap = NULL; 3779 new_entry->aref.ar_pageoff = 0; 3780 } 3781 3782 if (UVM_ET_ISOBJ(new_entry)) { 3783 if (new_entry->object.uvm_obj->pgops->pgo_detach) 3784 new_entry->object.uvm_obj->pgops->pgo_detach( 3785 new_entry->object.uvm_obj); 3786 new_entry->object.uvm_obj = NULL; 3787 new_entry->etype &= ~UVM_ET_OBJ; 3788 } 3789 3790 return (new_entry); 3791 } 3792 3793 /* 3794 * uvmspace_fork: fork a process' main map 3795 * 3796 * => create a new vmspace for child process from parent. 3797 * => parent's map must not be locked. 3798 */ 3799 struct vmspace * 3800 uvmspace_fork(struct process *pr) 3801 { 3802 struct vmspace *vm1 = pr->ps_vmspace; 3803 struct vmspace *vm2; 3804 struct vm_map *old_map = &vm1->vm_map; 3805 struct vm_map *new_map; 3806 struct vm_map_entry *old_entry, *new_entry; 3807 struct uvm_map_deadq dead; 3808 3809 vm_map_lock(old_map); 3810 3811 vm2 = uvmspace_alloc(old_map->min_offset, old_map->max_offset, 3812 (old_map->flags & VM_MAP_PAGEABLE) ? TRUE : FALSE, FALSE); 3813 memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy, 3814 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy); 3815 vm2->vm_dused = 0; /* Statistic managed by us. */ 3816 new_map = &vm2->vm_map; 3817 vm_map_lock(new_map); 3818 3819 /* go entry-by-entry */ 3820 TAILQ_INIT(&dead); 3821 RBT_FOREACH(old_entry, uvm_map_addr, &old_map->addr) { 3822 if (old_entry->start == old_entry->end) 3823 continue; 3824 3825 /* first, some sanity checks on the old entry */ 3826 if (UVM_ET_ISSUBMAP(old_entry)) { 3827 panic("fork: encountered a submap during fork " 3828 "(illegal)"); 3829 } 3830 3831 if (!UVM_ET_ISCOPYONWRITE(old_entry) && 3832 UVM_ET_ISNEEDSCOPY(old_entry)) { 3833 panic("fork: non-copy_on_write map entry marked " 3834 "needs_copy (illegal)"); 3835 } 3836 3837 /* Apply inheritance. */ 3838 switch (old_entry->inheritance) { 3839 case MAP_INHERIT_SHARE: 3840 new_entry = uvm_mapent_forkshared(vm2, new_map, 3841 old_map, old_entry, &dead); 3842 break; 3843 case MAP_INHERIT_COPY: 3844 new_entry = uvm_mapent_forkcopy(vm2, new_map, 3845 old_map, old_entry, &dead); 3846 break; 3847 case MAP_INHERIT_ZERO: 3848 new_entry = uvm_mapent_forkzero(vm2, new_map, 3849 old_map, old_entry, &dead); 3850 break; 3851 default: 3852 continue; 3853 } 3854 3855 /* Update process statistics. */ 3856 if (!UVM_ET_ISHOLE(new_entry)) 3857 new_map->size += new_entry->end - new_entry->start; 3858 if (!UVM_ET_ISOBJ(new_entry) && !UVM_ET_ISHOLE(new_entry) && 3859 new_entry->protection != PROT_NONE) { 3860 vm2->vm_dused += uvmspace_dused( 3861 new_map, new_entry->start, new_entry->end); 3862 } 3863 } 3864 new_map->flags |= old_map->flags & VM_MAP_PINSYSCALL_ONCE; 3865 #ifdef PMAP_CHECK_COPYIN 3866 if (PMAP_CHECK_COPYIN) { 3867 memcpy(&new_map->check_copyin, &old_map->check_copyin, 3868 sizeof(new_map->check_copyin)); 3869 membar_producer(); 3870 new_map->check_copyin_count = old_map->check_copyin_count; 3871 } 3872 #endif 3873 3874 vm_map_unlock(old_map); 3875 vm_map_unlock(new_map); 3876 3877 /* 3878 * This can actually happen, if multiple entries described a 3879 * space in which an entry was inherited. 3880 */ 3881 uvm_unmap_detach(&dead, 0); 3882 3883 #ifdef SYSVSHM 3884 if (vm1->vm_shm) 3885 shmfork(vm1, vm2); 3886 #endif 3887 3888 return vm2; 3889 } 3890 3891 /* 3892 * uvm_map_hint: return the beginning of the best area suitable for 3893 * creating a new mapping with "prot" protection. 3894 */ 3895 vaddr_t 3896 uvm_map_hint(struct vmspace *vm, vm_prot_t prot, vaddr_t minaddr, 3897 vaddr_t maxaddr) 3898 { 3899 vaddr_t addr; 3900 vaddr_t spacing; 3901 3902 #ifdef __i386__ 3903 /* 3904 * If executable skip first two pages, otherwise start 3905 * after data + heap region. 3906 */ 3907 if ((prot & PROT_EXEC) != 0 && 3908 (vaddr_t)vm->vm_daddr >= I386_MAX_EXE_ADDR) { 3909 addr = (PAGE_SIZE*2) + 3910 (arc4random() & (I386_MAX_EXE_ADDR / 2 - 1)); 3911 return (round_page(addr)); 3912 } 3913 #endif 3914 3915 #if defined (__LP64__) 3916 spacing = MIN(4UL * 1024 * 1024 * 1024, MAXDSIZ) - 1; 3917 #else 3918 spacing = MIN(1 * 1024 * 1024 * 1024, MAXDSIZ) - 1; 3919 #endif 3920 3921 /* 3922 * Start malloc/mmap after the brk. 3923 */ 3924 addr = (vaddr_t)vm->vm_daddr + BRKSIZ; 3925 addr = MAX(addr, minaddr); 3926 3927 if (addr < maxaddr) { 3928 while (spacing > maxaddr - addr) 3929 spacing >>= 1; 3930 } 3931 addr += arc4random() & spacing; 3932 return (round_page(addr)); 3933 } 3934 3935 /* 3936 * uvm_map_submap: punch down part of a map into a submap 3937 * 3938 * => only the kernel_map is allowed to be submapped 3939 * => the purpose of submapping is to break up the locking granularity 3940 * of a larger map 3941 * => the range specified must have been mapped previously with a uvm_map() 3942 * call [with uobj==NULL] to create a blank map entry in the main map. 3943 * [And it had better still be blank!] 3944 * => maps which contain submaps should never be copied or forked. 3945 * => to remove a submap, use uvm_unmap() on the main map 3946 * and then uvm_map_deallocate() the submap. 3947 * => main map must be unlocked. 3948 * => submap must have been init'd and have a zero reference count. 3949 * [need not be locked as we don't actually reference it] 3950 */ 3951 int 3952 uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end, 3953 struct vm_map *submap) 3954 { 3955 struct vm_map_entry *entry; 3956 int result; 3957 3958 if (start > map->max_offset || end > map->max_offset || 3959 start < map->min_offset || end < map->min_offset) 3960 return EINVAL; 3961 3962 vm_map_lock(map); 3963 3964 if (uvm_map_lookup_entry(map, start, &entry)) { 3965 UVM_MAP_CLIP_START(map, entry, start); 3966 UVM_MAP_CLIP_END(map, entry, end); 3967 } else 3968 entry = NULL; 3969 3970 if (entry != NULL && 3971 entry->start == start && entry->end == end && 3972 entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL && 3973 !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) { 3974 entry->etype |= UVM_ET_SUBMAP; 3975 entry->object.sub_map = submap; 3976 entry->offset = 0; 3977 uvm_map_reference(submap); 3978 result = 0; 3979 } else 3980 result = EINVAL; 3981 3982 vm_map_unlock(map); 3983 return result; 3984 } 3985 3986 /* 3987 * uvm_map_checkprot: check protection in map 3988 * 3989 * => must allow specific protection in a fully allocated region. 3990 * => map must be read or write locked by caller. 3991 */ 3992 boolean_t 3993 uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end, 3994 vm_prot_t protection) 3995 { 3996 struct vm_map_entry *entry; 3997 3998 vm_map_assert_anylock(map); 3999 4000 if (start < map->min_offset || end > map->max_offset || start > end) 4001 return FALSE; 4002 if (start == end) 4003 return TRUE; 4004 4005 /* 4006 * Iterate entries. 4007 */ 4008 for (entry = uvm_map_entrybyaddr(&map->addr, start); 4009 entry != NULL && entry->start < end; 4010 entry = RBT_NEXT(uvm_map_addr, entry)) { 4011 /* Fail if a hole is found. */ 4012 if (UVM_ET_ISHOLE(entry) || 4013 (entry->end < end && entry->end != VMMAP_FREE_END(entry))) 4014 return FALSE; 4015 4016 /* Check protection. */ 4017 if ((entry->protection & protection) != protection) 4018 return FALSE; 4019 } 4020 return TRUE; 4021 } 4022 4023 /* 4024 * uvm_map_create: create map 4025 */ 4026 vm_map_t 4027 uvm_map_create(pmap_t pmap, vaddr_t min, vaddr_t max, int flags) 4028 { 4029 vm_map_t map; 4030 4031 map = malloc(sizeof *map, M_VMMAP, M_WAITOK); 4032 uvm_map_setup(map, pmap, min, max, flags); 4033 return (map); 4034 } 4035 4036 /* 4037 * uvm_map_deallocate: drop reference to a map 4038 * 4039 * => caller must not lock map 4040 * => we will zap map if ref count goes to zero 4041 */ 4042 void 4043 uvm_map_deallocate(vm_map_t map) 4044 { 4045 int c; 4046 struct uvm_map_deadq dead; 4047 4048 c = atomic_dec_int_nv(&map->ref_count); 4049 if (c > 0) { 4050 return; 4051 } 4052 4053 /* 4054 * all references gone. unmap and free. 4055 * 4056 * No lock required: we are only one to access this map. 4057 */ 4058 TAILQ_INIT(&dead); 4059 uvm_tree_sanity(map, __FILE__, __LINE__); 4060 vm_map_lock(map); 4061 uvm_unmap_remove(map, map->min_offset, map->max_offset, &dead, 4062 TRUE, FALSE, FALSE); 4063 vm_map_unlock(map); 4064 pmap_destroy(map->pmap); 4065 KASSERT(RBT_EMPTY(uvm_map_addr, &map->addr)); 4066 free(map, M_VMMAP, sizeof *map); 4067 4068 uvm_unmap_detach(&dead, 0); 4069 } 4070 4071 /* 4072 * uvm_map_inherit: set inheritance code for range of addrs in map. 4073 * 4074 * => map must be unlocked 4075 * => note that the inherit code is used during a "fork". see fork 4076 * code for details. 4077 */ 4078 int 4079 uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end, 4080 vm_inherit_t new_inheritance) 4081 { 4082 struct vm_map_entry *entry, *entry1; 4083 int error = EPERM; 4084 4085 switch (new_inheritance) { 4086 case MAP_INHERIT_NONE: 4087 case MAP_INHERIT_COPY: 4088 case MAP_INHERIT_SHARE: 4089 case MAP_INHERIT_ZERO: 4090 break; 4091 default: 4092 return (EINVAL); 4093 } 4094 4095 if (start > end) 4096 return EINVAL; 4097 start = MAX(start, map->min_offset); 4098 end = MIN(end, map->max_offset); 4099 if (start >= end) 4100 return 0; 4101 4102 vm_map_lock(map); 4103 4104 entry = uvm_map_entrybyaddr(&map->addr, start); 4105 if (entry->end > start) 4106 UVM_MAP_CLIP_START(map, entry, start); 4107 else 4108 entry = RBT_NEXT(uvm_map_addr, entry); 4109 4110 /* First check for illegal operations */ 4111 entry1 = entry; 4112 while (entry1 != NULL && entry1->start < end) { 4113 if (entry1->etype & UVM_ET_IMMUTABLE) 4114 goto out; 4115 if (new_inheritance == MAP_INHERIT_ZERO && 4116 (entry1->protection & PROT_WRITE) == 0) 4117 goto out; 4118 entry1 = RBT_NEXT(uvm_map_addr, entry1); 4119 } 4120 4121 while (entry != NULL && entry->start < end) { 4122 UVM_MAP_CLIP_END(map, entry, end); 4123 entry->inheritance = new_inheritance; 4124 entry = RBT_NEXT(uvm_map_addr, entry); 4125 } 4126 4127 error = 0; 4128 out: 4129 vm_map_unlock(map); 4130 return (error); 4131 } 4132 4133 #ifdef PMAP_CHECK_COPYIN 4134 static void inline 4135 check_copyin_add(struct vm_map *map, vaddr_t start, vaddr_t end) 4136 { 4137 if (PMAP_CHECK_COPYIN == 0 || 4138 map->check_copyin_count >= UVM_MAP_CHECK_COPYIN_MAX) 4139 return; 4140 vm_map_assert_wrlock(map); 4141 map->check_copyin[map->check_copyin_count].start = start; 4142 map->check_copyin[map->check_copyin_count].end = end; 4143 membar_producer(); 4144 map->check_copyin_count++; 4145 } 4146 4147 /* 4148 * uvm_map_check_copyin_add: remember regions which are X-only for copyin(), 4149 * copyinstr(), uiomove(), and others 4150 * 4151 * => map must be unlocked 4152 */ 4153 int 4154 uvm_map_check_copyin_add(struct vm_map *map, vaddr_t start, vaddr_t end) 4155 { 4156 if (start > end) 4157 return EINVAL; 4158 start = MAX(start, map->min_offset); 4159 end = MIN(end, map->max_offset); 4160 if (start >= end) 4161 return 0; 4162 vm_map_lock(map); 4163 check_copyin_add(map, start, end); 4164 vm_map_unlock(map); 4165 return (0); 4166 } 4167 #endif /* PMAP_CHECK_COPYIN */ 4168 4169 /* 4170 * uvm_map_immutable: block mapping/mprotect for range of addrs in map. 4171 * 4172 * => map must be unlocked 4173 */ 4174 int 4175 uvm_map_immutable(struct vm_map *map, vaddr_t start, vaddr_t end, int imut) 4176 { 4177 struct vm_map_entry *entry, *entry1; 4178 int error = EPERM; 4179 4180 if (start > end) 4181 return EINVAL; 4182 start = MAX(start, map->min_offset); 4183 end = MIN(end, map->max_offset); 4184 if (start >= end) 4185 return 0; 4186 4187 vm_map_lock(map); 4188 4189 entry = uvm_map_entrybyaddr(&map->addr, start); 4190 if (entry->end > start) 4191 UVM_MAP_CLIP_START(map, entry, start); 4192 else 4193 entry = RBT_NEXT(uvm_map_addr, entry); 4194 4195 /* First check for illegal operations */ 4196 entry1 = entry; 4197 while (entry1 != NULL && entry1->start < end) { 4198 if (entry1->inheritance == MAP_INHERIT_ZERO) 4199 goto out; 4200 entry1 = RBT_NEXT(uvm_map_addr, entry1); 4201 } 4202 4203 while (entry != NULL && entry->start < end) { 4204 UVM_MAP_CLIP_END(map, entry, end); 4205 if (imut) 4206 entry->etype |= UVM_ET_IMMUTABLE; 4207 else 4208 entry->etype &= ~UVM_ET_IMMUTABLE; 4209 entry = RBT_NEXT(uvm_map_addr, entry); 4210 } 4211 error = 0; 4212 out: 4213 vm_map_unlock(map); 4214 return (0); 4215 } 4216 4217 /* 4218 * uvm_map_advice: set advice code for range of addrs in map. 4219 * 4220 * => map must be unlocked 4221 */ 4222 int 4223 uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice) 4224 { 4225 struct vm_map_entry *entry; 4226 4227 switch (new_advice) { 4228 case MADV_NORMAL: 4229 case MADV_RANDOM: 4230 case MADV_SEQUENTIAL: 4231 break; 4232 default: 4233 return (EINVAL); 4234 } 4235 4236 if (start > end) 4237 return EINVAL; 4238 start = MAX(start, map->min_offset); 4239 end = MIN(end, map->max_offset); 4240 if (start >= end) 4241 return 0; 4242 4243 vm_map_lock(map); 4244 4245 entry = uvm_map_entrybyaddr(&map->addr, start); 4246 if (entry != NULL && entry->end > start) 4247 UVM_MAP_CLIP_START(map, entry, start); 4248 else if (entry!= NULL) 4249 entry = RBT_NEXT(uvm_map_addr, entry); 4250 4251 /* 4252 * XXXJRT: disallow holes? 4253 */ 4254 while (entry != NULL && entry->start < end) { 4255 UVM_MAP_CLIP_END(map, entry, end); 4256 entry->advice = new_advice; 4257 entry = RBT_NEXT(uvm_map_addr, entry); 4258 } 4259 4260 vm_map_unlock(map); 4261 return (0); 4262 } 4263 4264 /* 4265 * uvm_map_extract: extract a mapping from a map and put it somewhere 4266 * in the kernel_map, setting protection to max_prot. 4267 * 4268 * => map should be unlocked (we will write lock it and kernel_map) 4269 * => returns 0 on success, error code otherwise 4270 * => start must be page aligned 4271 * => len must be page sized 4272 * => flags: 4273 * UVM_EXTRACT_FIXPROT: set prot to maxprot as we go 4274 * Mappings are QREF's. 4275 */ 4276 int 4277 uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len, 4278 vaddr_t *dstaddrp, int flags) 4279 { 4280 struct uvm_map_deadq dead; 4281 struct vm_map_entry *first, *entry, *newentry, *tmp1, *tmp2; 4282 vaddr_t dstaddr; 4283 vaddr_t end; 4284 vaddr_t cp_start; 4285 vsize_t cp_len, cp_off; 4286 int error; 4287 4288 TAILQ_INIT(&dead); 4289 end = start + len; 4290 4291 /* 4292 * Sanity check on the parameters. 4293 * Also, since the mapping may not contain gaps, error out if the 4294 * mapped area is not in source map. 4295 */ 4296 if ((start & (vaddr_t)PAGE_MASK) != 0 || 4297 (end & (vaddr_t)PAGE_MASK) != 0 || end < start) 4298 return EINVAL; 4299 if (start < srcmap->min_offset || end > srcmap->max_offset) 4300 return EINVAL; 4301 4302 /* Initialize dead entries. Handle len == 0 case. */ 4303 if (len == 0) 4304 return 0; 4305 4306 /* Acquire lock on srcmap. */ 4307 vm_map_lock(srcmap); 4308 4309 /* Lock srcmap, lookup first and last entry in <start,len>. */ 4310 first = uvm_map_entrybyaddr(&srcmap->addr, start); 4311 4312 /* Check that the range is contiguous. */ 4313 for (entry = first; entry != NULL && entry->end < end; 4314 entry = RBT_NEXT(uvm_map_addr, entry)) { 4315 if (VMMAP_FREE_END(entry) != entry->end || 4316 UVM_ET_ISHOLE(entry)) { 4317 error = EINVAL; 4318 goto fail; 4319 } 4320 } 4321 if (entry == NULL || UVM_ET_ISHOLE(entry)) { 4322 error = EINVAL; 4323 goto fail; 4324 } 4325 4326 /* 4327 * Handle need-copy flag. 4328 */ 4329 for (entry = first; entry != NULL && entry->start < end; 4330 entry = RBT_NEXT(uvm_map_addr, entry)) { 4331 if (UVM_ET_ISNEEDSCOPY(entry)) 4332 amap_copy(srcmap, entry, M_NOWAIT, 4333 UVM_ET_ISSTACK(entry) ? FALSE : TRUE, start, end); 4334 if (UVM_ET_ISNEEDSCOPY(entry)) { 4335 /* 4336 * amap_copy failure 4337 */ 4338 error = ENOMEM; 4339 goto fail; 4340 } 4341 } 4342 4343 /* Lock destination map (kernel_map). */ 4344 vm_map_lock(kernel_map); 4345 4346 if (uvm_map_findspace(kernel_map, &tmp1, &tmp2, &dstaddr, len, 4347 MAX(PAGE_SIZE, PMAP_PREFER_ALIGN()), PMAP_PREFER_OFFSET(start), 4348 PROT_NONE, 0) != 0) { 4349 error = ENOMEM; 4350 goto fail2; 4351 } 4352 *dstaddrp = dstaddr; 4353 4354 /* 4355 * We now have srcmap and kernel_map locked. 4356 * dstaddr contains the destination offset in dstmap. 4357 */ 4358 /* step 1: start looping through map entries, performing extraction. */ 4359 for (entry = first; entry != NULL && entry->start < end; 4360 entry = RBT_NEXT(uvm_map_addr, entry)) { 4361 KDASSERT(!UVM_ET_ISNEEDSCOPY(entry)); 4362 if (UVM_ET_ISHOLE(entry)) 4363 continue; 4364 4365 /* Calculate uvm_mapent_clone parameters. */ 4366 cp_start = entry->start; 4367 if (cp_start < start) { 4368 cp_off = start - cp_start; 4369 cp_start = start; 4370 } else 4371 cp_off = 0; 4372 cp_len = MIN(entry->end, end) - cp_start; 4373 4374 newentry = uvm_mapent_clone(kernel_map, 4375 cp_start - start + dstaddr, cp_len, cp_off, 4376 entry->protection, entry->max_protection, 4377 entry, &dead, flags, AMAP_SHARED | AMAP_REFALL); 4378 if (newentry == NULL) { 4379 error = ENOMEM; 4380 goto fail2_unmap; 4381 } 4382 kernel_map->size += cp_len; 4383 4384 /* Figure out the best protection */ 4385 if ((flags & UVM_EXTRACT_FIXPROT) && 4386 newentry->protection != PROT_NONE) 4387 newentry->protection = newentry->max_protection; 4388 newentry->protection &= ~PROT_EXEC; 4389 } 4390 pmap_update(kernel_map->pmap); 4391 4392 error = 0; 4393 4394 /* Unmap copied entries on failure. */ 4395 fail2_unmap: 4396 if (error) { 4397 uvm_unmap_remove(kernel_map, dstaddr, dstaddr + len, &dead, 4398 FALSE, TRUE, FALSE); 4399 } 4400 4401 /* Release maps, release dead entries. */ 4402 fail2: 4403 vm_map_unlock(kernel_map); 4404 4405 fail: 4406 vm_map_unlock(srcmap); 4407 4408 uvm_unmap_detach(&dead, 0); 4409 4410 return error; 4411 } 4412 4413 /* 4414 * uvm_map_clean: clean out a map range 4415 * 4416 * => valid flags: 4417 * if (flags & PGO_CLEANIT): dirty pages are cleaned first 4418 * if (flags & PGO_SYNCIO): dirty pages are written synchronously 4419 * if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean 4420 * if (flags & PGO_FREE): any cached pages are freed after clean 4421 * => returns an error if any part of the specified range isn't mapped 4422 * => never a need to flush amap layer since the anonymous memory has 4423 * no permanent home, but may deactivate pages there 4424 * => called from sys_msync() and sys_madvise() 4425 * => caller must not have map locked 4426 */ 4427 4428 int 4429 uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags) 4430 { 4431 struct vm_map_entry *first, *entry; 4432 struct vm_amap *amap; 4433 struct vm_anon *anon; 4434 struct vm_page *pg; 4435 struct uvm_object *uobj; 4436 vaddr_t cp_start, cp_end; 4437 int refs, imut = 0; 4438 int error; 4439 boolean_t rv; 4440 4441 KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) != 4442 (PGO_FREE|PGO_DEACTIVATE)); 4443 4444 if (start > end || start < map->min_offset || end > map->max_offset) 4445 return EINVAL; 4446 4447 vm_map_lock(map); 4448 first = uvm_map_entrybyaddr(&map->addr, start); 4449 4450 /* Make a first pass to check for various conditions. */ 4451 for (entry = first; entry != NULL && entry->start < end; 4452 entry = RBT_NEXT(uvm_map_addr, entry)) { 4453 if (entry->etype & UVM_ET_IMMUTABLE) 4454 imut = 1; 4455 if (UVM_ET_ISSUBMAP(entry)) { 4456 vm_map_unlock(map); 4457 return EINVAL; 4458 } 4459 if (UVM_ET_ISSUBMAP(entry) || 4460 UVM_ET_ISHOLE(entry) || 4461 (entry->end < end && 4462 VMMAP_FREE_END(entry) != entry->end)) { 4463 vm_map_unlock(map); 4464 return EFAULT; 4465 } 4466 } 4467 4468 vm_map_busy(map); 4469 vm_map_unlock(map); 4470 error = 0; 4471 for (entry = first; entry != NULL && entry->start < end; 4472 entry = RBT_NEXT(uvm_map_addr, entry)) { 4473 amap = entry->aref.ar_amap; /* top layer */ 4474 if (UVM_ET_ISOBJ(entry)) 4475 uobj = entry->object.uvm_obj; 4476 else 4477 uobj = NULL; 4478 4479 /* 4480 * No amap cleaning necessary if: 4481 * - there's no amap 4482 * - we're not deactivating or freeing pages. 4483 */ 4484 if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0) 4485 goto flush_object; 4486 4487 if (imut) { 4488 vm_map_unbusy(map); 4489 return EPERM; 4490 } 4491 4492 cp_start = MAX(entry->start, start); 4493 cp_end = MIN(entry->end, end); 4494 4495 amap_lock(amap, RW_WRITE); 4496 for (; cp_start != cp_end; cp_start += PAGE_SIZE) { 4497 anon = amap_lookup(&entry->aref, 4498 cp_start - entry->start); 4499 if (anon == NULL) 4500 continue; 4501 4502 KASSERT(anon->an_lock == amap->am_lock); 4503 pg = anon->an_page; 4504 if (pg == NULL) { 4505 continue; 4506 } 4507 KASSERT(pg->pg_flags & PQ_ANON); 4508 4509 switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) { 4510 /* 4511 * XXX In these first 3 cases, we always just 4512 * XXX deactivate the page. We may want to 4513 * XXX handle the different cases more 4514 * XXX specifically, in the future. 4515 */ 4516 case PGO_CLEANIT|PGO_FREE: 4517 case PGO_CLEANIT|PGO_DEACTIVATE: 4518 case PGO_DEACTIVATE: 4519 deactivate_it: 4520 /* skip the page if it's wired */ 4521 if (pg->wire_count != 0) 4522 break; 4523 4524 uvm_lock_pageq(); 4525 4526 KASSERT(pg->uanon == anon); 4527 uvm_pagedeactivate(pg); 4528 4529 uvm_unlock_pageq(); 4530 break; 4531 case PGO_FREE: 4532 /* 4533 * If there are multiple references to 4534 * the amap, just deactivate the page. 4535 */ 4536 if (amap_refs(amap) > 1) 4537 goto deactivate_it; 4538 4539 /* XXX skip the page if it's wired */ 4540 if (pg->wire_count != 0) { 4541 break; 4542 } 4543 amap_unadd(&entry->aref, 4544 cp_start - entry->start); 4545 refs = --anon->an_ref; 4546 if (refs == 0) 4547 uvm_anfree(anon); 4548 break; 4549 default: 4550 panic("uvm_map_clean: weird flags"); 4551 } 4552 } 4553 amap_unlock(amap); 4554 4555 flush_object: 4556 cp_start = MAX(entry->start, start); 4557 cp_end = MIN(entry->end, end); 4558 4559 /* 4560 * flush pages if we've got a valid backing object. 4561 * 4562 * Don't PGO_FREE if we don't have write permission 4563 * and don't flush if this is a copy-on-write object 4564 * since we can't know our permissions on it. 4565 */ 4566 if (uobj != NULL && 4567 ((flags & PGO_FREE) == 0 || 4568 ((entry->max_protection & PROT_WRITE) != 0 && 4569 (entry->etype & UVM_ET_COPYONWRITE) == 0))) { 4570 rw_enter(uobj->vmobjlock, RW_WRITE); 4571 rv = uobj->pgops->pgo_flush(uobj, 4572 cp_start - entry->start + entry->offset, 4573 cp_end - entry->start + entry->offset, flags); 4574 rw_exit(uobj->vmobjlock); 4575 4576 if (rv == FALSE) 4577 error = EFAULT; 4578 } 4579 } 4580 4581 vm_map_unbusy(map); 4582 return error; 4583 } 4584 4585 /* 4586 * UVM_MAP_CLIP_END implementation 4587 */ 4588 void 4589 uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t addr) 4590 { 4591 struct vm_map_entry *tmp; 4592 4593 KASSERT(entry->start < addr && VMMAP_FREE_END(entry) > addr); 4594 tmp = uvm_mapent_alloc(map, 0); 4595 4596 /* Invoke splitentry. */ 4597 uvm_map_splitentry(map, entry, tmp, addr); 4598 } 4599 4600 /* 4601 * UVM_MAP_CLIP_START implementation 4602 * 4603 * Clippers are required to not change the pointers to the entry they are 4604 * clipping on. 4605 * Since uvm_map_splitentry turns the original entry into the lowest 4606 * entry (address wise) we do a swap between the new entry and the original 4607 * entry, prior to calling uvm_map_splitentry. 4608 */ 4609 void 4610 uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry, vaddr_t addr) 4611 { 4612 struct vm_map_entry *tmp; 4613 struct uvm_addr_state *free; 4614 4615 /* Unlink original. */ 4616 free = uvm_map_uaddr_e(map, entry); 4617 uvm_mapent_free_remove(map, free, entry); 4618 uvm_mapent_addr_remove(map, entry); 4619 4620 /* Copy entry. */ 4621 KASSERT(entry->start < addr && VMMAP_FREE_END(entry) > addr); 4622 tmp = uvm_mapent_alloc(map, 0); 4623 uvm_mapent_copy(entry, tmp); 4624 4625 /* Put new entry in place of original entry. */ 4626 uvm_mapent_addr_insert(map, tmp); 4627 uvm_mapent_free_insert(map, free, tmp); 4628 4629 /* Invoke splitentry. */ 4630 uvm_map_splitentry(map, tmp, entry, addr); 4631 } 4632 4633 /* 4634 * Boundary fixer. 4635 */ 4636 static inline vaddr_t uvm_map_boundfix(vaddr_t, vaddr_t, vaddr_t); 4637 static inline vaddr_t 4638 uvm_map_boundfix(vaddr_t min, vaddr_t max, vaddr_t bound) 4639 { 4640 return (min < bound && max > bound) ? bound : max; 4641 } 4642 4643 /* 4644 * Choose free list based on address at start of free space. 4645 * 4646 * The uvm_addr_state returned contains addr and is the first of: 4647 * - uaddr_exe 4648 * - uaddr_brk_stack 4649 * - uaddr_any 4650 */ 4651 struct uvm_addr_state* 4652 uvm_map_uaddr(struct vm_map *map, vaddr_t addr) 4653 { 4654 struct uvm_addr_state *uaddr; 4655 int i; 4656 4657 /* Special case the first page, to prevent mmap from returning 0. */ 4658 if (addr < VMMAP_MIN_ADDR) 4659 return NULL; 4660 4661 /* Upper bound for kernel maps at uvm_maxkaddr. */ 4662 if ((map->flags & VM_MAP_ISVMSPACE) == 0) { 4663 if (addr >= uvm_maxkaddr) 4664 return NULL; 4665 } 4666 4667 /* Is the address inside the exe-only map? */ 4668 if (map->uaddr_exe != NULL && addr >= map->uaddr_exe->uaddr_minaddr && 4669 addr < map->uaddr_exe->uaddr_maxaddr) 4670 return map->uaddr_exe; 4671 4672 /* Check if the space falls inside brk/stack area. */ 4673 if ((addr >= map->b_start && addr < map->b_end) || 4674 (addr >= map->s_start && addr < map->s_end)) { 4675 if (map->uaddr_brk_stack != NULL && 4676 addr >= map->uaddr_brk_stack->uaddr_minaddr && 4677 addr < map->uaddr_brk_stack->uaddr_maxaddr) { 4678 return map->uaddr_brk_stack; 4679 } else 4680 return NULL; 4681 } 4682 4683 /* 4684 * Check the other selectors. 4685 * 4686 * These selectors are only marked as the owner, if they have insert 4687 * functions. 4688 */ 4689 for (i = 0; i < nitems(map->uaddr_any); i++) { 4690 uaddr = map->uaddr_any[i]; 4691 if (uaddr == NULL) 4692 continue; 4693 if (uaddr->uaddr_functions->uaddr_free_insert == NULL) 4694 continue; 4695 4696 if (addr >= uaddr->uaddr_minaddr && 4697 addr < uaddr->uaddr_maxaddr) 4698 return uaddr; 4699 } 4700 4701 return NULL; 4702 } 4703 4704 /* 4705 * Choose free list based on address at start of free space. 4706 * 4707 * The uvm_addr_state returned contains addr and is the first of: 4708 * - uaddr_exe 4709 * - uaddr_brk_stack 4710 * - uaddr_any 4711 */ 4712 struct uvm_addr_state* 4713 uvm_map_uaddr_e(struct vm_map *map, struct vm_map_entry *entry) 4714 { 4715 return uvm_map_uaddr(map, VMMAP_FREE_START(entry)); 4716 } 4717 4718 /* 4719 * Returns the first free-memory boundary that is crossed by [min-max]. 4720 */ 4721 vsize_t 4722 uvm_map_boundary(struct vm_map *map, vaddr_t min, vaddr_t max) 4723 { 4724 struct uvm_addr_state *uaddr; 4725 int i; 4726 4727 /* Never return first page. */ 4728 max = uvm_map_boundfix(min, max, VMMAP_MIN_ADDR); 4729 4730 /* Treat the maxkaddr special, if the map is a kernel_map. */ 4731 if ((map->flags & VM_MAP_ISVMSPACE) == 0) 4732 max = uvm_map_boundfix(min, max, uvm_maxkaddr); 4733 4734 /* Check for exe-only boundaries. */ 4735 if (map->uaddr_exe != NULL) { 4736 max = uvm_map_boundfix(min, max, map->uaddr_exe->uaddr_minaddr); 4737 max = uvm_map_boundfix(min, max, map->uaddr_exe->uaddr_maxaddr); 4738 } 4739 4740 /* Check for exe-only boundaries. */ 4741 if (map->uaddr_brk_stack != NULL) { 4742 max = uvm_map_boundfix(min, max, 4743 map->uaddr_brk_stack->uaddr_minaddr); 4744 max = uvm_map_boundfix(min, max, 4745 map->uaddr_brk_stack->uaddr_maxaddr); 4746 } 4747 4748 /* Check other boundaries. */ 4749 for (i = 0; i < nitems(map->uaddr_any); i++) { 4750 uaddr = map->uaddr_any[i]; 4751 if (uaddr != NULL) { 4752 max = uvm_map_boundfix(min, max, uaddr->uaddr_minaddr); 4753 max = uvm_map_boundfix(min, max, uaddr->uaddr_maxaddr); 4754 } 4755 } 4756 4757 /* Boundaries at stack and brk() area. */ 4758 max = uvm_map_boundfix(min, max, map->s_start); 4759 max = uvm_map_boundfix(min, max, map->s_end); 4760 max = uvm_map_boundfix(min, max, map->b_start); 4761 max = uvm_map_boundfix(min, max, map->b_end); 4762 4763 return max; 4764 } 4765 4766 /* 4767 * Update map allocation start and end addresses from proc vmspace. 4768 */ 4769 void 4770 uvm_map_vmspace_update(struct vm_map *map, 4771 struct uvm_map_deadq *dead, int flags) 4772 { 4773 struct vmspace *vm; 4774 vaddr_t b_start, b_end, s_start, s_end; 4775 4776 KASSERT(map->flags & VM_MAP_ISVMSPACE); 4777 KASSERT(offsetof(struct vmspace, vm_map) == 0); 4778 4779 /* 4780 * Derive actual allocation boundaries from vmspace. 4781 */ 4782 vm = (struct vmspace *)map; 4783 b_start = (vaddr_t)vm->vm_daddr; 4784 b_end = b_start + BRKSIZ; 4785 s_start = MIN((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 4786 s_end = MAX((vaddr_t)vm->vm_maxsaddr, (vaddr_t)vm->vm_minsaddr); 4787 #ifdef DIAGNOSTIC 4788 if ((b_start & (vaddr_t)PAGE_MASK) != 0 || 4789 (b_end & (vaddr_t)PAGE_MASK) != 0 || 4790 (s_start & (vaddr_t)PAGE_MASK) != 0 || 4791 (s_end & (vaddr_t)PAGE_MASK) != 0) { 4792 panic("uvm_map_vmspace_update: vmspace %p invalid bounds: " 4793 "b=0x%lx-0x%lx s=0x%lx-0x%lx", 4794 vm, b_start, b_end, s_start, s_end); 4795 } 4796 #endif 4797 4798 if (__predict_true(map->b_start == b_start && map->b_end == b_end && 4799 map->s_start == s_start && map->s_end == s_end)) 4800 return; 4801 4802 uvm_map_freelist_update(map, dead, b_start, b_end, 4803 s_start, s_end, flags); 4804 } 4805 4806 /* 4807 * Grow kernel memory. 4808 * 4809 * This function is only called for kernel maps when an allocation fails. 4810 * 4811 * If the map has a gap that is large enough to accommodate alloc_sz, this 4812 * function will make sure map->free will include it. 4813 */ 4814 void 4815 uvm_map_kmem_grow(struct vm_map *map, struct uvm_map_deadq *dead, 4816 vsize_t alloc_sz, int flags) 4817 { 4818 vsize_t sz; 4819 vaddr_t end; 4820 struct vm_map_entry *entry; 4821 4822 /* Kernel memory only. */ 4823 KASSERT((map->flags & VM_MAP_ISVMSPACE) == 0); 4824 /* Destroy free list. */ 4825 uvm_map_freelist_update_clear(map, dead); 4826 4827 /* Include the guard page in the hard minimum requirement of alloc_sz. */ 4828 if (map->flags & VM_MAP_GUARDPAGES) 4829 alloc_sz += PAGE_SIZE; 4830 4831 /* 4832 * Grow by ALLOCMUL * alloc_sz, but at least VM_MAP_KSIZE_DELTA. 4833 * 4834 * Don't handle the case where the multiplication overflows: 4835 * if that happens, the allocation is probably too big anyway. 4836 */ 4837 sz = MAX(VM_MAP_KSIZE_ALLOCMUL * alloc_sz, VM_MAP_KSIZE_DELTA); 4838 4839 /* 4840 * Walk forward until a gap large enough for alloc_sz shows up. 4841 * 4842 * We assume the kernel map has no boundaries. 4843 * uvm_maxkaddr may be zero. 4844 */ 4845 end = MAX(uvm_maxkaddr, map->min_offset); 4846 entry = uvm_map_entrybyaddr(&map->addr, end); 4847 while (entry && entry->fspace < alloc_sz) 4848 entry = RBT_NEXT(uvm_map_addr, entry); 4849 if (entry) { 4850 end = MAX(VMMAP_FREE_START(entry), end); 4851 end += MIN(sz, map->max_offset - end); 4852 } else 4853 end = map->max_offset; 4854 4855 /* Reserve pmap entries. */ 4856 #ifdef PMAP_GROWKERNEL 4857 uvm_maxkaddr = pmap_growkernel(end); 4858 #else 4859 uvm_maxkaddr = MAX(uvm_maxkaddr, end); 4860 #endif 4861 4862 /* Rebuild free list. */ 4863 uvm_map_freelist_update_refill(map, flags); 4864 } 4865 4866 /* 4867 * Freelist update subfunction: unlink all entries from freelists. 4868 */ 4869 void 4870 uvm_map_freelist_update_clear(struct vm_map *map, struct uvm_map_deadq *dead) 4871 { 4872 struct uvm_addr_state *free; 4873 struct vm_map_entry *entry, *prev, *next; 4874 4875 prev = NULL; 4876 for (entry = RBT_MIN(uvm_map_addr, &map->addr); entry != NULL; 4877 entry = next) { 4878 next = RBT_NEXT(uvm_map_addr, entry); 4879 4880 free = uvm_map_uaddr_e(map, entry); 4881 uvm_mapent_free_remove(map, free, entry); 4882 4883 if (prev != NULL && entry->start == entry->end) { 4884 prev->fspace += VMMAP_FREE_END(entry) - entry->end; 4885 uvm_mapent_addr_remove(map, entry); 4886 DEAD_ENTRY_PUSH(dead, entry); 4887 } else 4888 prev = entry; 4889 } 4890 } 4891 4892 /* 4893 * Freelist update subfunction: refill the freelists with entries. 4894 */ 4895 void 4896 uvm_map_freelist_update_refill(struct vm_map *map, int flags) 4897 { 4898 struct vm_map_entry *entry; 4899 vaddr_t min, max; 4900 4901 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 4902 min = VMMAP_FREE_START(entry); 4903 max = VMMAP_FREE_END(entry); 4904 entry->fspace = 0; 4905 4906 entry = uvm_map_fix_space(map, entry, min, max, flags); 4907 } 4908 4909 uvm_tree_sanity(map, __FILE__, __LINE__); 4910 } 4911 4912 /* 4913 * Change {a,b}_{start,end} allocation ranges and associated free lists. 4914 */ 4915 void 4916 uvm_map_freelist_update(struct vm_map *map, struct uvm_map_deadq *dead, 4917 vaddr_t b_start, vaddr_t b_end, vaddr_t s_start, vaddr_t s_end, int flags) 4918 { 4919 KDASSERT(b_end >= b_start && s_end >= s_start); 4920 vm_map_assert_wrlock(map); 4921 4922 /* Clear all free lists. */ 4923 uvm_map_freelist_update_clear(map, dead); 4924 4925 /* Apply new bounds. */ 4926 map->b_start = b_start; 4927 map->b_end = b_end; 4928 map->s_start = s_start; 4929 map->s_end = s_end; 4930 4931 /* Refill free lists. */ 4932 uvm_map_freelist_update_refill(map, flags); 4933 } 4934 4935 /* 4936 * Assign a uvm_addr_state to the specified pointer in vm_map. 4937 * 4938 * May sleep. 4939 */ 4940 void 4941 uvm_map_set_uaddr(struct vm_map *map, struct uvm_addr_state **which, 4942 struct uvm_addr_state *newval) 4943 { 4944 struct uvm_map_deadq dead; 4945 4946 /* Pointer which must be in this map. */ 4947 KASSERT(which != NULL); 4948 KASSERT((void*)map <= (void*)(which) && 4949 (void*)(which) < (void*)(map + 1)); 4950 4951 vm_map_lock(map); 4952 TAILQ_INIT(&dead); 4953 uvm_map_freelist_update_clear(map, &dead); 4954 4955 uvm_addr_destroy(*which); 4956 *which = newval; 4957 4958 uvm_map_freelist_update_refill(map, 0); 4959 vm_map_unlock(map); 4960 uvm_unmap_detach(&dead, 0); 4961 } 4962 4963 /* 4964 * Correct space insert. 4965 * 4966 * Entry must not be on any freelist. 4967 */ 4968 struct vm_map_entry* 4969 uvm_map_fix_space(struct vm_map *map, struct vm_map_entry *entry, 4970 vaddr_t min, vaddr_t max, int flags) 4971 { 4972 struct uvm_addr_state *free, *entfree; 4973 vaddr_t lmax; 4974 4975 KASSERT(entry == NULL || (entry->etype & UVM_ET_FREEMAPPED) == 0); 4976 KDASSERT(min <= max); 4977 KDASSERT((entry != NULL && VMMAP_FREE_END(entry) == min) || 4978 min == map->min_offset); 4979 4980 UVM_MAP_REQ_WRITE(map); 4981 4982 /* 4983 * During the function, entfree will always point at the uaddr state 4984 * for entry. 4985 */ 4986 entfree = (entry == NULL ? NULL : 4987 uvm_map_uaddr_e(map, entry)); 4988 4989 while (min != max) { 4990 /* Claim guard page for entry. */ 4991 if ((map->flags & VM_MAP_GUARDPAGES) && entry != NULL && 4992 VMMAP_FREE_END(entry) == entry->end && 4993 entry->start != entry->end) { 4994 if (max - min == 2 * PAGE_SIZE) { 4995 /* 4996 * If the free-space gap is exactly 2 pages, 4997 * we make the guard 2 pages instead of 1. 4998 * Because in a guarded map, an area needs 4999 * at least 2 pages to allocate from: 5000 * one page for the allocation and one for 5001 * the guard. 5002 */ 5003 entry->guard = 2 * PAGE_SIZE; 5004 min = max; 5005 } else { 5006 entry->guard = PAGE_SIZE; 5007 min += PAGE_SIZE; 5008 } 5009 continue; 5010 } 5011 5012 /* 5013 * Handle the case where entry has a 2-page guard, but the 5014 * space after entry is freed. 5015 */ 5016 if (entry != NULL && entry->fspace == 0 && 5017 entry->guard > PAGE_SIZE) { 5018 entry->guard = PAGE_SIZE; 5019 min = VMMAP_FREE_START(entry); 5020 } 5021 5022 lmax = uvm_map_boundary(map, min, max); 5023 free = uvm_map_uaddr(map, min); 5024 5025 /* 5026 * Entries are merged if they point at the same uvm_free(). 5027 * Exception to that rule: if min == uvm_maxkaddr, a new 5028 * entry is started regardless (otherwise the allocators 5029 * will get confused). 5030 */ 5031 if (entry != NULL && free == entfree && 5032 !((map->flags & VM_MAP_ISVMSPACE) == 0 && 5033 min == uvm_maxkaddr)) { 5034 KDASSERT(VMMAP_FREE_END(entry) == min); 5035 entry->fspace += lmax - min; 5036 } else { 5037 /* 5038 * Commit entry to free list: it'll not be added to 5039 * anymore. 5040 * We'll start a new entry and add to that entry 5041 * instead. 5042 */ 5043 if (entry != NULL) 5044 uvm_mapent_free_insert(map, entfree, entry); 5045 5046 /* New entry for new uaddr. */ 5047 entry = uvm_mapent_alloc(map, flags); 5048 KDASSERT(entry != NULL); 5049 entry->end = entry->start = min; 5050 entry->guard = 0; 5051 entry->fspace = lmax - min; 5052 entry->object.uvm_obj = NULL; 5053 entry->offset = 0; 5054 entry->etype = 0; 5055 entry->protection = entry->max_protection = 0; 5056 entry->inheritance = 0; 5057 entry->wired_count = 0; 5058 entry->advice = 0; 5059 entry->aref.ar_pageoff = 0; 5060 entry->aref.ar_amap = NULL; 5061 uvm_mapent_addr_insert(map, entry); 5062 5063 entfree = free; 5064 } 5065 5066 min = lmax; 5067 } 5068 /* Finally put entry on the uaddr state. */ 5069 if (entry != NULL) 5070 uvm_mapent_free_insert(map, entfree, entry); 5071 5072 return entry; 5073 } 5074 5075 /* 5076 * MQuery style of allocation. 5077 * 5078 * This allocator searches forward until sufficient space is found to map 5079 * the given size. 5080 * 5081 * XXX: factor in offset (via pmap_prefer) and protection? 5082 */ 5083 int 5084 uvm_map_mquery(struct vm_map *map, vaddr_t *addr_p, vsize_t sz, voff_t offset, 5085 int flags) 5086 { 5087 struct vm_map_entry *entry, *last; 5088 vaddr_t addr; 5089 vaddr_t tmp, pmap_align, pmap_offset; 5090 int error; 5091 5092 addr = *addr_p; 5093 vm_map_lock_read(map); 5094 5095 /* Configure pmap prefer. */ 5096 if (offset != UVM_UNKNOWN_OFFSET) { 5097 pmap_align = MAX(PAGE_SIZE, PMAP_PREFER_ALIGN()); 5098 pmap_offset = PMAP_PREFER_OFFSET(offset); 5099 } else { 5100 pmap_align = PAGE_SIZE; 5101 pmap_offset = 0; 5102 } 5103 5104 /* Align address to pmap_prefer unless FLAG_FIXED is set. */ 5105 if (!(flags & UVM_FLAG_FIXED) && offset != UVM_UNKNOWN_OFFSET) { 5106 tmp = (addr & ~(pmap_align - 1)) | pmap_offset; 5107 if (tmp < addr) 5108 tmp += pmap_align; 5109 addr = tmp; 5110 } 5111 5112 /* First, check if the requested range is fully available. */ 5113 entry = uvm_map_entrybyaddr(&map->addr, addr); 5114 last = NULL; 5115 if (uvm_map_isavail(map, NULL, &entry, &last, addr, sz)) { 5116 error = 0; 5117 goto out; 5118 } 5119 if (flags & UVM_FLAG_FIXED) { 5120 error = EINVAL; 5121 goto out; 5122 } 5123 5124 error = ENOMEM; /* Default error from here. */ 5125 5126 /* 5127 * At this point, the memory at <addr, sz> is not available. 5128 * The reasons are: 5129 * [1] it's outside the map, 5130 * [2] it starts in used memory (and therefore needs to move 5131 * toward the first free page in entry), 5132 * [3] it starts in free memory but bumps into used memory. 5133 * 5134 * Note that for case [2], the forward moving is handled by the 5135 * for loop below. 5136 */ 5137 if (entry == NULL) { 5138 /* [1] Outside the map. */ 5139 if (addr >= map->max_offset) 5140 goto out; 5141 else 5142 entry = RBT_MIN(uvm_map_addr, &map->addr); 5143 } else if (VMMAP_FREE_START(entry) <= addr) { 5144 /* [3] Bumped into used memory. */ 5145 entry = RBT_NEXT(uvm_map_addr, entry); 5146 } 5147 5148 /* Test if the next entry is sufficient for the allocation. */ 5149 for (; entry != NULL; 5150 entry = RBT_NEXT(uvm_map_addr, entry)) { 5151 if (entry->fspace == 0) 5152 continue; 5153 addr = VMMAP_FREE_START(entry); 5154 5155 restart: /* Restart address checks on address change. */ 5156 tmp = (addr & ~(pmap_align - 1)) | pmap_offset; 5157 if (tmp < addr) 5158 tmp += pmap_align; 5159 addr = tmp; 5160 if (addr >= VMMAP_FREE_END(entry)) 5161 continue; 5162 5163 /* Skip brk() allocation addresses. */ 5164 if (addr + sz > map->b_start && addr < map->b_end) { 5165 if (VMMAP_FREE_END(entry) > map->b_end) { 5166 addr = map->b_end; 5167 goto restart; 5168 } else 5169 continue; 5170 } 5171 /* Skip stack allocation addresses. */ 5172 if (addr + sz > map->s_start && addr < map->s_end) { 5173 if (VMMAP_FREE_END(entry) > map->s_end) { 5174 addr = map->s_end; 5175 goto restart; 5176 } else 5177 continue; 5178 } 5179 5180 last = NULL; 5181 if (uvm_map_isavail(map, NULL, &entry, &last, addr, sz)) { 5182 error = 0; 5183 goto out; 5184 } 5185 } 5186 5187 out: 5188 vm_map_unlock_read(map); 5189 if (error == 0) 5190 *addr_p = addr; 5191 return error; 5192 } 5193 5194 boolean_t 5195 vm_map_lock_try_ln(struct vm_map *map, char *file, int line) 5196 { 5197 int rv; 5198 5199 if (map->flags & VM_MAP_INTRSAFE) { 5200 if (!mtx_enter_try(&map->mtx)) 5201 return FALSE; 5202 } else { 5203 struct proc *busy; 5204 5205 mtx_enter(&map->flags_lock); 5206 busy = map->busy; 5207 mtx_leave(&map->flags_lock); 5208 if (busy != NULL && busy != curproc) 5209 return FALSE; 5210 5211 rv = rw_enter(&map->lock, RW_WRITE|RW_NOSLEEP); 5212 if (rv != 0) 5213 return FALSE; 5214 5215 /* to be sure, to be sure */ 5216 mtx_enter(&map->flags_lock); 5217 busy = map->busy; 5218 mtx_leave(&map->flags_lock); 5219 if (busy != NULL && busy != curproc) { 5220 rw_exit(&map->lock); 5221 return FALSE; 5222 } 5223 } 5224 5225 map->timestamp++; 5226 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5227 uvm_tree_sanity(map, file, line); 5228 uvm_tree_size_chk(map, file, line); 5229 5230 return TRUE; 5231 } 5232 5233 void 5234 vm_map_lock_ln(struct vm_map *map, char *file, int line) 5235 { 5236 if ((map->flags & VM_MAP_INTRSAFE) == 0) { 5237 mtx_enter(&map->flags_lock); 5238 for (;;) { 5239 while (map->busy != NULL && map->busy != curproc) { 5240 map->nbusy++; 5241 msleep_nsec(&map->busy, &map->flags_lock, 5242 PVM, vmmapbsy, INFSLP); 5243 map->nbusy--; 5244 } 5245 mtx_leave(&map->flags_lock); 5246 5247 rw_enter_write(&map->lock); 5248 5249 /* to be sure, to be sure */ 5250 mtx_enter(&map->flags_lock); 5251 if (map->busy != NULL && map->busy != curproc) { 5252 /* go around again */ 5253 rw_exit_write(&map->lock); 5254 } else { 5255 /* we won */ 5256 break; 5257 } 5258 } 5259 mtx_leave(&map->flags_lock); 5260 } else { 5261 mtx_enter(&map->mtx); 5262 } 5263 5264 if (map->busy != curproc) { 5265 KASSERT(map->busy == NULL); 5266 map->timestamp++; 5267 } 5268 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5269 uvm_tree_sanity(map, file, line); 5270 uvm_tree_size_chk(map, file, line); 5271 } 5272 5273 void 5274 vm_map_lock_read_ln(struct vm_map *map, char *file, int line) 5275 { 5276 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5277 rw_enter_read(&map->lock); 5278 else 5279 mtx_enter(&map->mtx); 5280 LPRINTF(("map lock: %p (at %s %d)\n", map, file, line)); 5281 uvm_tree_sanity(map, file, line); 5282 uvm_tree_size_chk(map, file, line); 5283 } 5284 5285 void 5286 vm_map_unlock_ln(struct vm_map *map, char *file, int line) 5287 { 5288 KASSERT(map->busy == NULL || map->busy == curproc); 5289 uvm_tree_sanity(map, file, line); 5290 uvm_tree_size_chk(map, file, line); 5291 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5292 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5293 rw_exit(&map->lock); 5294 else 5295 mtx_leave(&map->mtx); 5296 } 5297 5298 void 5299 vm_map_unlock_read_ln(struct vm_map *map, char *file, int line) 5300 { 5301 /* XXX: RO */ uvm_tree_sanity(map, file, line); 5302 /* XXX: RO */ uvm_tree_size_chk(map, file, line); 5303 LPRINTF(("map unlock: %p (at %s %d)\n", map, file, line)); 5304 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5305 rw_exit_read(&map->lock); 5306 else 5307 mtx_leave(&map->mtx); 5308 } 5309 5310 void 5311 vm_map_busy_ln(struct vm_map *map, char *file, int line) 5312 { 5313 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5314 KASSERT(rw_write_held(&map->lock)); 5315 KASSERT(map->busy == NULL); 5316 5317 mtx_enter(&map->flags_lock); 5318 map->busy = curproc; 5319 mtx_leave(&map->flags_lock); 5320 } 5321 5322 void 5323 vm_map_unbusy_ln(struct vm_map *map, char *file, int line) 5324 { 5325 unsigned int nbusy; 5326 5327 KASSERT((map->flags & VM_MAP_INTRSAFE) == 0); 5328 KASSERT(map->busy == curproc); 5329 5330 mtx_enter(&map->flags_lock); 5331 nbusy = map->nbusy; 5332 map->busy = NULL; 5333 mtx_leave(&map->flags_lock); 5334 5335 if (nbusy > 0) 5336 wakeup(&map->busy); 5337 } 5338 5339 void 5340 vm_map_assert_anylock_ln(struct vm_map *map, char *file, int line) 5341 { 5342 LPRINTF(("map assert read or write locked: %p (at %s %d)\n", map, file, line)); 5343 if ((map->flags & VM_MAP_INTRSAFE) == 0) 5344 rw_assert_anylock(&map->lock); 5345 else 5346 MUTEX_ASSERT_LOCKED(&map->mtx); 5347 } 5348 5349 void 5350 vm_map_assert_wrlock_ln(struct vm_map *map, char *file, int line) 5351 { 5352 LPRINTF(("map assert write locked: %p (at %s %d)\n", map, file, line)); 5353 if ((map->flags & VM_MAP_INTRSAFE) == 0) { 5354 splassert(IPL_NONE); 5355 rw_assert_wrlock(&map->lock); 5356 } else 5357 MUTEX_ASSERT_LOCKED(&map->mtx); 5358 } 5359 5360 #ifndef SMALL_KERNEL 5361 int 5362 uvm_map_fill_vmmap(struct vm_map *map, struct kinfo_vmentry *kve, 5363 size_t *lenp) 5364 { 5365 struct vm_map_entry *entry; 5366 vaddr_t start; 5367 int cnt, maxcnt, error = 0; 5368 5369 KASSERT(*lenp > 0); 5370 KASSERT((*lenp % sizeof(*kve)) == 0); 5371 cnt = 0; 5372 maxcnt = *lenp / sizeof(*kve); 5373 KASSERT(maxcnt > 0); 5374 5375 /* 5376 * Return only entries whose address is above the given base 5377 * address. This allows userland to iterate without knowing the 5378 * number of entries beforehand. 5379 */ 5380 start = (vaddr_t)kve[0].kve_start; 5381 5382 vm_map_lock(map); 5383 RBT_FOREACH(entry, uvm_map_addr, &map->addr) { 5384 if (cnt == maxcnt) { 5385 error = ENOMEM; 5386 break; 5387 } 5388 if (start != 0 && entry->start < start) 5389 continue; 5390 kve->kve_start = entry->start; 5391 kve->kve_end = entry->end; 5392 kve->kve_guard = entry->guard; 5393 kve->kve_fspace = entry->fspace; 5394 kve->kve_fspace_augment = entry->fspace_augment; 5395 kve->kve_offset = entry->offset; 5396 kve->kve_wired_count = entry->wired_count; 5397 kve->kve_etype = entry->etype; 5398 kve->kve_protection = entry->protection; 5399 kve->kve_max_protection = entry->max_protection; 5400 kve->kve_advice = entry->advice; 5401 kve->kve_inheritance = entry->inheritance; 5402 kve->kve_flags = entry->flags; 5403 kve++; 5404 cnt++; 5405 } 5406 vm_map_unlock(map); 5407 5408 KASSERT(cnt <= maxcnt); 5409 5410 *lenp = sizeof(*kve) * cnt; 5411 return error; 5412 } 5413 #endif 5414 5415 5416 RBT_GENERATE_AUGMENT(uvm_map_addr, vm_map_entry, daddrs.addr_entry, 5417 uvm_mapentry_addrcmp, uvm_map_addr_augment); 5418 5419 5420 /* 5421 * MD code: vmspace allocator setup. 5422 */ 5423 5424 #ifdef __i386__ 5425 void 5426 uvm_map_setup_md(struct vm_map *map) 5427 { 5428 vaddr_t min, max; 5429 5430 min = map->min_offset; 5431 max = map->max_offset; 5432 5433 /* 5434 * Ensure the selectors will not try to manage page 0; 5435 * it's too special. 5436 */ 5437 if (min < VMMAP_MIN_ADDR) 5438 min = VMMAP_MIN_ADDR; 5439 5440 #if 0 /* Cool stuff, not yet */ 5441 /* Executable code is special. */ 5442 map->uaddr_exe = uaddr_rnd_create(min, I386_MAX_EXE_ADDR); 5443 /* Place normal allocations beyond executable mappings. */ 5444 map->uaddr_any[3] = uaddr_pivot_create(2 * I386_MAX_EXE_ADDR, max); 5445 #else /* Crappy stuff, for now */ 5446 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5447 #endif 5448 5449 #ifndef SMALL_KERNEL 5450 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5451 #endif /* !SMALL_KERNEL */ 5452 } 5453 #elif __LP64__ 5454 void 5455 uvm_map_setup_md(struct vm_map *map) 5456 { 5457 vaddr_t min, max; 5458 5459 min = map->min_offset; 5460 max = map->max_offset; 5461 5462 /* 5463 * Ensure the selectors will not try to manage page 0; 5464 * it's too special. 5465 */ 5466 if (min < VMMAP_MIN_ADDR) 5467 min = VMMAP_MIN_ADDR; 5468 5469 #if 0 /* Cool stuff, not yet */ 5470 map->uaddr_any[3] = uaddr_pivot_create(MAX(min, 0x100000000ULL), max); 5471 #else /* Crappy stuff, for now */ 5472 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5473 #endif 5474 5475 #ifndef SMALL_KERNEL 5476 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5477 #endif /* !SMALL_KERNEL */ 5478 } 5479 #else /* non-i386, 32 bit */ 5480 void 5481 uvm_map_setup_md(struct vm_map *map) 5482 { 5483 vaddr_t min, max; 5484 5485 min = map->min_offset; 5486 max = map->max_offset; 5487 5488 /* 5489 * Ensure the selectors will not try to manage page 0; 5490 * it's too special. 5491 */ 5492 if (min < VMMAP_MIN_ADDR) 5493 min = VMMAP_MIN_ADDR; 5494 5495 #if 0 /* Cool stuff, not yet */ 5496 map->uaddr_any[3] = uaddr_pivot_create(min, max); 5497 #else /* Crappy stuff, for now */ 5498 map->uaddr_any[0] = uaddr_rnd_create(min, max); 5499 #endif 5500 5501 #ifndef SMALL_KERNEL 5502 map->uaddr_brk_stack = uaddr_stack_brk_create(min, max); 5503 #endif /* !SMALL_KERNEL */ 5504 } 5505 #endif 5506