| 1 | /* |
| 2 | * Copyright (c) 1991, 1993 |
| 3 | * The Regents of the University of California. All rights reserved. |
| 4 | * |
| 5 | * This code is derived from software contributed to Berkeley by |
| 6 | * The Mach Operating System project at Carnegie-Mellon University. |
| 7 | * |
| 8 | * Redistribution and use in source and binary forms, with or without |
| 9 | * modification, are permitted provided that the following conditions |
| 10 | * are met: |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in the |
| 15 | * documentation and/or other materials provided with the distribution. |
| 16 | * 3. All advertising materials mentioning features or use of this software |
| 17 | * must display the following acknowledgement: |
| 18 | * This product includes software developed by the University of |
| 19 | * California, Berkeley and its contributors. |
| 20 | * 4. Neither the name of the University nor the names of its contributors |
| 21 | * may be used to endorse or promote products derived from this software |
| 22 | * without specific prior written permission. |
| 23 | * |
| 24 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 25 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 26 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 27 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 28 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 29 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 30 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 31 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 32 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 33 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 34 | * SUCH DAMAGE. |
| 35 | * |
| 36 | * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94 |
| 37 | * |
| 38 | * |
| 39 | * Copyright (c) 1987, 1990 Carnegie-Mellon University. |
| 40 | * All rights reserved. |
| 41 | * |
| 42 | * Authors: Avadis Tevanian, Jr., Michael Wayne Young |
| 43 | * |
| 44 | * Permission to use, copy, modify and distribute this software and |
| 45 | * its documentation is hereby granted, provided that both the copyright |
| 46 | * notice and this permission notice appear in all copies of the |
| 47 | * software, derivative works or modified versions, and any portions |
| 48 | * thereof, and that both notices appear in supporting documentation. |
| 49 | * |
| 50 | * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" |
| 51 | * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND |
| 52 | * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. |
| 53 | * |
| 54 | * Carnegie Mellon requests users of this software to return to |
| 55 | * |
| 56 | * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU |
| 57 | * School of Computer Science |
| 58 | * Carnegie Mellon University |
| 59 | * Pittsburgh PA 15213-3890 |
| 60 | * |
| 61 | * any improvements or extensions that they make and grant Carnegie the |
| 62 | * rights to redistribute these changes. |
| 63 | * |
| 64 | * $FreeBSD: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $ |
| 65 | * $DragonFly: src/sys/vm/vm_map.c,v 1.56 2007/04/29 18:25:41 dillon Exp $ |
| 66 | */ |
| 67 | |
| 68 | /* |
| 69 | * Virtual memory mapping module. |
| 70 | */ |
| 71 | |
| 72 | #include <sys/param.h> |
| 73 | #include <sys/systm.h> |
| 74 | #include <sys/kernel.h> |
| 75 | #include <sys/proc.h> |
| 76 | #include <sys/lock.h> |
| 77 | #include <sys/vmmeter.h> |
| 78 | #include <sys/mman.h> |
| 79 | #include <sys/vnode.h> |
| 80 | #include <sys/resourcevar.h> |
| 81 | #include <sys/shm.h> |
| 82 | #include <sys/tree.h> |
| 83 | #include <sys/malloc.h> |
| 84 | |
| 85 | #include <vm/vm.h> |
| 86 | #include <vm/vm_param.h> |
| 87 | #include <vm/pmap.h> |
| 88 | #include <vm/vm_map.h> |
| 89 | #include <vm/vm_page.h> |
| 90 | #include <vm/vm_object.h> |
| 91 | #include <vm/vm_pager.h> |
| 92 | #include <vm/vm_kern.h> |
| 93 | #include <vm/vm_extern.h> |
| 94 | #include <vm/swap_pager.h> |
| 95 | #include <vm/vm_zone.h> |
| 96 | |
| 97 | #include <sys/thread2.h> |
| 98 | #include <sys/sysref2.h> |
| 99 | |
| 100 | /* |
| 101 | * Virtual memory maps provide for the mapping, protection, |
| 102 | * and sharing of virtual memory objects. In addition, |
| 103 | * this module provides for an efficient virtual copy of |
| 104 | * memory from one map to another. |
| 105 | * |
| 106 | * Synchronization is required prior to most operations. |
| 107 | * |
| 108 | * Maps consist of an ordered doubly-linked list of simple |
| 109 | * entries; a single hint is used to speed up lookups. |
| 110 | * |
| 111 | * Since portions of maps are specified by start/end addresses, |
| 112 | * which may not align with existing map entries, all |
| 113 | * routines merely "clip" entries to these start/end values. |
| 114 | * [That is, an entry is split into two, bordering at a |
| 115 | * start or end value.] Note that these clippings may not |
| 116 | * always be necessary (as the two resulting entries are then |
| 117 | * not changed); however, the clipping is done for convenience. |
| 118 | * |
| 119 | * As mentioned above, virtual copy operations are performed |
| 120 | * by copying VM object references from one map to |
| 121 | * another, and then marking both regions as copy-on-write. |
| 122 | */ |
| 123 | |
| 124 | static void vmspace_terminate(struct vmspace *vm); |
| 125 | static void vmspace_lock(struct vmspace *vm); |
| 126 | static void vmspace_unlock(struct vmspace *vm); |
| 127 | static void vmspace_dtor(void *obj, void *private); |
| 128 | |
| 129 | MALLOC_DEFINE(M_VMSPACE, "vmspace", "vmspace objcache backingstore"); |
| 130 | |
| 131 | struct sysref_class vmspace_sysref_class = { |
| 132 | .name = "vmspace", |
| 133 | .mtype = M_VMSPACE, |
| 134 | .proto = SYSREF_PROTO_VMSPACE, |
| 135 | .offset = offsetof(struct vmspace, vm_sysref), |
| 136 | .objsize = sizeof(struct vmspace), |
| 137 | .mag_capacity = 32, |
| 138 | .flags = SRC_MANAGEDINIT, |
| 139 | .dtor = vmspace_dtor, |
| 140 | .ops = { |
| 141 | .terminate = (sysref_terminate_func_t)vmspace_terminate, |
| 142 | .lock = (sysref_lock_func_t)vmspace_lock, |
| 143 | .unlock = (sysref_lock_func_t)vmspace_unlock |
| 144 | } |
| 145 | }; |
| 146 | |
| 147 | #define VMEPERCPU 2 |
| 148 | |
| 149 | static struct vm_zone mapentzone_store, mapzone_store; |
| 150 | static vm_zone_t mapentzone, mapzone; |
| 151 | static struct vm_object mapentobj, mapobj; |
| 152 | |
| 153 | static struct vm_map_entry map_entry_init[MAX_MAPENT]; |
| 154 | static struct vm_map_entry cpu_map_entry_init[MAXCPU][VMEPERCPU]; |
| 155 | static struct vm_map map_init[MAX_KMAP]; |
| 156 | |
| 157 | static void vm_map_entry_shadow(vm_map_entry_t entry); |
| 158 | static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *); |
| 159 | static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *); |
| 160 | static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *); |
| 161 | static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *); |
| 162 | static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *); |
| 163 | static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t); |
| 164 | static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t, |
| 165 | vm_map_entry_t); |
| 166 | static void vm_map_split (vm_map_entry_t); |
| 167 | static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int *count, int flags); |
| 168 | |
| 169 | /* |
| 170 | * vm_map_startup: |
| 171 | * |
| 172 | * Initialize the vm_map module. Must be called before |
| 173 | * any other vm_map routines. |
| 174 | * |
| 175 | * Map and entry structures are allocated from the general |
| 176 | * purpose memory pool with some exceptions: |
| 177 | * |
| 178 | * - The kernel map and kmem submap are allocated statically. |
| 179 | * - Kernel map entries are allocated out of a static pool. |
| 180 | * |
| 181 | * These restrictions are necessary since malloc() uses the |
| 182 | * maps and requires map entries. |
| 183 | */ |
| 184 | void |
| 185 | vm_map_startup(void) |
| 186 | { |
| 187 | mapzone = &mapzone_store; |
| 188 | zbootinit(mapzone, "MAP", sizeof (struct vm_map), |
| 189 | map_init, MAX_KMAP); |
| 190 | mapentzone = &mapentzone_store; |
| 191 | zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry), |
| 192 | map_entry_init, MAX_MAPENT); |
| 193 | } |
| 194 | |
| 195 | /* |
| 196 | * vm_init2 - called prior to any vmspace allocations |
| 197 | */ |
| 198 | void |
| 199 | vm_init2(void) |
| 200 | { |
| 201 | zinitna(mapentzone, &mapentobj, NULL, 0, 0, |
| 202 | ZONE_USE_RESERVE | ZONE_SPECIAL, 1); |
| 203 | zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1); |
| 204 | pmap_init2(); |
| 205 | vm_object_init2(); |
| 206 | } |
| 207 | |
| 208 | |
| 209 | /* |
| 210 | * Red black tree functions |
| 211 | */ |
| 212 | static int rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b); |
| 213 | RB_GENERATE(vm_map_rb_tree, vm_map_entry, rb_entry, rb_vm_map_compare); |
| 214 | |
| 215 | /* a->start is address, and the only field has to be initialized */ |
| 216 | static int |
| 217 | rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b) |
| 218 | { |
| 219 | if (a->start < b->start) |
| 220 | return(-1); |
| 221 | else if (a->start > b->start) |
| 222 | return(1); |
| 223 | return(0); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Allocate a vmspace structure, including a vm_map and pmap. |
| 228 | * Initialize numerous fields. While the initial allocation is zerod, |
| 229 | * subsequence reuse from the objcache leaves elements of the structure |
| 230 | * intact (particularly the pmap), so portions must be zerod. |
| 231 | * |
| 232 | * The structure is not considered activated until we call sysref_activate(). |
| 233 | */ |
| 234 | struct vmspace * |
| 235 | vmspace_alloc(vm_offset_t min, vm_offset_t max) |
| 236 | { |
| 237 | struct vmspace *vm; |
| 238 | |
| 239 | vm = sysref_alloc(&vmspace_sysref_class); |
| 240 | bzero(&vm->vm_startcopy, |
| 241 | (char *)&vm->vm_endcopy - (char *)&vm->vm_startcopy); |
| 242 | vm_map_init(&vm->vm_map, min, max, NULL); |
| 243 | pmap_pinit(vmspace_pmap(vm)); /* (some fields reused) */ |
| 244 | vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */ |
| 245 | vm->vm_shm = NULL; |
| 246 | vm->vm_exitingcnt = 0; |
| 247 | cpu_vmspace_alloc(vm); |
| 248 | sysref_activate(&vm->vm_sysref); |
| 249 | return (vm); |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * dtor function - Some elements of the pmap are retained in the |
| 254 | * free-cached vmspaces to improve performance. We have to clean them up |
| 255 | * here before returning the vmspace to the memory pool. |
| 256 | */ |
| 257 | static void |
| 258 | vmspace_dtor(void *obj, void *private) |
| 259 | { |
| 260 | struct vmspace *vm = obj; |
| 261 | |
| 262 | pmap_puninit(vmspace_pmap(vm)); |
| 263 | } |
| 264 | |
| 265 | /* |
| 266 | * Called in two cases: |
| 267 | * |
| 268 | * (1) When the last sysref is dropped, but exitingcnt might still be |
| 269 | * non-zero. |
| 270 | * |
| 271 | * (2) When there are no sysrefs (i.e. refcnt is negative) left and the |
| 272 | * exitingcnt becomes zero |
| 273 | * |
| 274 | * sysref will not scrap the object until we call sysref_put() once more |
| 275 | * after the last ref has been dropped. |
| 276 | */ |
| 277 | static void |
| 278 | vmspace_terminate(struct vmspace *vm) |
| 279 | { |
| 280 | int count; |
| 281 | |
| 282 | /* |
| 283 | * If exitingcnt is non-zero we can't get rid of the entire vmspace |
| 284 | * yet, but we can scrap user memory. |
| 285 | */ |
| 286 | if (vm->vm_exitingcnt) { |
| 287 | shmexit(vm); |
| 288 | pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS, |
| 289 | VM_MAX_USER_ADDRESS); |
| 290 | vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS, |
| 291 | VM_MAX_USER_ADDRESS); |
| 292 | |
| 293 | return; |
| 294 | } |
| 295 | cpu_vmspace_free(vm); |
| 296 | |
| 297 | /* |
| 298 | * Make sure any SysV shm is freed, it might not have in |
| 299 | * exit1() |
| 300 | */ |
| 301 | shmexit(vm); |
| 302 | |
| 303 | KKASSERT(vm->vm_upcalls == NULL); |
| 304 | |
| 305 | /* |
| 306 | * Lock the map, to wait out all other references to it. |
| 307 | * Delete all of the mappings and pages they hold, then call |
| 308 | * the pmap module to reclaim anything left. |
| 309 | */ |
| 310 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 311 | vm_map_lock(&vm->vm_map); |
| 312 | vm_map_delete(&vm->vm_map, vm->vm_map.min_offset, |
| 313 | vm->vm_map.max_offset, &count); |
| 314 | vm_map_unlock(&vm->vm_map); |
| 315 | vm_map_entry_release(count); |
| 316 | |
| 317 | pmap_release(vmspace_pmap(vm)); |
| 318 | sysref_put(&vm->vm_sysref); |
| 319 | } |
| 320 | |
| 321 | static void |
| 322 | vmspace_lock(struct vmspace *vm __unused) |
| 323 | { |
| 324 | } |
| 325 | |
| 326 | static void |
| 327 | vmspace_unlock(struct vmspace *vm __unused) |
| 328 | { |
| 329 | } |
| 330 | |
| 331 | /* |
| 332 | * This is called in the wait*() handling code. The vmspace can be terminated |
| 333 | * after the last wait is finished using it. |
| 334 | */ |
| 335 | void |
| 336 | vmspace_exitfree(struct proc *p) |
| 337 | { |
| 338 | struct vmspace *vm; |
| 339 | |
| 340 | vm = p->p_vmspace; |
| 341 | p->p_vmspace = NULL; |
| 342 | |
| 343 | if (--vm->vm_exitingcnt == 0 && sysref_isinactive(&vm->vm_sysref)) |
| 344 | vmspace_terminate(vm); |
| 345 | } |
| 346 | |
| 347 | /* |
| 348 | * vmspace_swap_count() |
| 349 | * |
| 350 | * Swap useage is determined by taking the proportional swap used by |
| 351 | * VM objects backing the VM map. To make up for fractional losses, |
| 352 | * if the VM object has any swap use at all the associated map entries |
| 353 | * count for at least 1 swap page. |
| 354 | */ |
| 355 | int |
| 356 | vmspace_swap_count(struct vmspace *vmspace) |
| 357 | { |
| 358 | vm_map_t map = &vmspace->vm_map; |
| 359 | vm_map_entry_t cur; |
| 360 | vm_object_t object; |
| 361 | int count = 0; |
| 362 | int n; |
| 363 | |
| 364 | for (cur = map->header.next; cur != &map->header; cur = cur->next) { |
| 365 | switch(cur->maptype) { |
| 366 | case VM_MAPTYPE_NORMAL: |
| 367 | case VM_MAPTYPE_VPAGETABLE: |
| 368 | if ((object = cur->object.vm_object) == NULL) |
| 369 | break; |
| 370 | if (object->swblock_count) { |
| 371 | n = (cur->end - cur->start) / PAGE_SIZE; |
| 372 | count += object->swblock_count * |
| 373 | SWAP_META_PAGES * n / object->size + 1; |
| 374 | } |
| 375 | break; |
| 376 | default: |
| 377 | break; |
| 378 | } |
| 379 | } |
| 380 | return(count); |
| 381 | } |
| 382 | |
| 383 | /* |
| 384 | * vmspace_anonymous_count() |
| 385 | * |
| 386 | * Calculate the approximate number of anonymous pages in use by |
| 387 | * this vmspace. To make up for fractional losses, we count each |
| 388 | * VM object as having at least 1 anonymous page. |
| 389 | */ |
| 390 | int |
| 391 | vmspace_anonymous_count(struct vmspace *vmspace) |
| 392 | { |
| 393 | vm_map_t map = &vmspace->vm_map; |
| 394 | vm_map_entry_t cur; |
| 395 | vm_object_t object; |
| 396 | int count = 0; |
| 397 | |
| 398 | for (cur = map->header.next; cur != &map->header; cur = cur->next) { |
| 399 | switch(cur->maptype) { |
| 400 | case VM_MAPTYPE_NORMAL: |
| 401 | case VM_MAPTYPE_VPAGETABLE: |
| 402 | if ((object = cur->object.vm_object) == NULL) |
| 403 | break; |
| 404 | if (object->type != OBJT_DEFAULT && |
| 405 | object->type != OBJT_SWAP) { |
| 406 | break; |
| 407 | } |
| 408 | count += object->resident_page_count; |
| 409 | break; |
| 410 | default: |
| 411 | break; |
| 412 | } |
| 413 | } |
| 414 | return(count); |
| 415 | } |
| 416 | |
| 417 | |
| 418 | |
| 419 | |
| 420 | /* |
| 421 | * vm_map_create: |
| 422 | * |
| 423 | * Creates and returns a new empty VM map with |
| 424 | * the given physical map structure, and having |
| 425 | * the given lower and upper address bounds. |
| 426 | */ |
| 427 | vm_map_t |
| 428 | vm_map_create(vm_map_t result, pmap_t pmap, vm_offset_t min, vm_offset_t max) |
| 429 | { |
| 430 | if (result == NULL) |
| 431 | result = zalloc(mapzone); |
| 432 | vm_map_init(result, min, max, pmap); |
| 433 | return (result); |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * Initialize an existing vm_map structure |
| 438 | * such as that in the vmspace structure. |
| 439 | * The pmap is set elsewhere. |
| 440 | */ |
| 441 | void |
| 442 | vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max, pmap_t pmap) |
| 443 | { |
| 444 | map->header.next = map->header.prev = &map->header; |
| 445 | RB_INIT(&map->rb_root); |
| 446 | map->nentries = 0; |
| 447 | map->size = 0; |
| 448 | map->system_map = 0; |
| 449 | map->infork = 0; |
| 450 | map->min_offset = min; |
| 451 | map->max_offset = max; |
| 452 | map->pmap = pmap; |
| 453 | map->first_free = &map->header; |
| 454 | map->hint = &map->header; |
| 455 | map->timestamp = 0; |
| 456 | lockinit(&map->lock, "thrd_sleep", 0, 0); |
| 457 | } |
| 458 | |
| 459 | /* |
| 460 | * Shadow the vm_map_entry's object. This typically needs to be done when |
| 461 | * a write fault is taken on an entry which had previously been cloned by |
| 462 | * fork(). The shared object (which might be NULL) must become private so |
| 463 | * we add a shadow layer above it. |
| 464 | * |
| 465 | * Object allocation for anonymous mappings is defered as long as possible. |
| 466 | * When creating a shadow, however, the underlying object must be instantiated |
| 467 | * so it can be shared. |
| 468 | * |
| 469 | * If the map segment is governed by a virtual page table then it is |
| 470 | * possible to address offsets beyond the mapped area. Just allocate |
| 471 | * a maximally sized object for this case. |
| 472 | */ |
| 473 | static |
| 474 | void |
| 475 | vm_map_entry_shadow(vm_map_entry_t entry) |
| 476 | { |
| 477 | if (entry->maptype == VM_MAPTYPE_VPAGETABLE) { |
| 478 | vm_object_shadow(&entry->object.vm_object, &entry->offset, |
| 479 | 0x7FFFFFFF); /* XXX */ |
| 480 | } else { |
| 481 | vm_object_shadow(&entry->object.vm_object, &entry->offset, |
| 482 | atop(entry->end - entry->start)); |
| 483 | } |
| 484 | entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; |
| 485 | } |
| 486 | |
| 487 | /* |
| 488 | * Allocate an object for a vm_map_entry. |
| 489 | * |
| 490 | * Object allocation for anonymous mappings is defered as long as possible. |
| 491 | * This function is called when we can defer no longer, generally when a map |
| 492 | * entry might be split or forked or takes a page fault. |
| 493 | * |
| 494 | * If the map segment is governed by a virtual page table then it is |
| 495 | * possible to address offsets beyond the mapped area. Just allocate |
| 496 | * a maximally sized object for this case. |
| 497 | */ |
| 498 | void |
| 499 | vm_map_entry_allocate_object(vm_map_entry_t entry) |
| 500 | { |
| 501 | vm_object_t obj; |
| 502 | |
| 503 | if (entry->maptype == VM_MAPTYPE_VPAGETABLE) { |
| 504 | obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */ |
| 505 | } else { |
| 506 | obj = vm_object_allocate(OBJT_DEFAULT, |
| 507 | atop(entry->end - entry->start)); |
| 508 | } |
| 509 | entry->object.vm_object = obj; |
| 510 | entry->offset = 0; |
| 511 | } |
| 512 | |
| 513 | /* |
| 514 | * vm_map_entry_reserve_cpu_init: |
| 515 | * |
| 516 | * Set an initial negative count so the first attempt to reserve |
| 517 | * space preloads a bunch of vm_map_entry's for this cpu. Also |
| 518 | * pre-allocate 2 vm_map_entries which will be needed by zalloc() to |
| 519 | * map a new page for vm_map_entry structures. SMP systems are |
| 520 | * particularly sensitive. |
| 521 | * |
| 522 | * This routine is called in early boot so we cannot just call |
| 523 | * vm_map_entry_reserve(). |
| 524 | * |
| 525 | * May be called for a gd other then mycpu, but may only be called |
| 526 | * during early boot. |
| 527 | */ |
| 528 | void |
| 529 | vm_map_entry_reserve_cpu_init(globaldata_t gd) |
| 530 | { |
| 531 | vm_map_entry_t entry; |
| 532 | int i; |
| 533 | |
| 534 | gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2; |
| 535 | entry = &cpu_map_entry_init[gd->gd_cpuid][0]; |
| 536 | for (i = 0; i < VMEPERCPU; ++i, ++entry) { |
| 537 | entry->next = gd->gd_vme_base; |
| 538 | gd->gd_vme_base = entry; |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | /* |
| 543 | * vm_map_entry_reserve: |
| 544 | * |
| 545 | * Reserves vm_map_entry structures so code later on can manipulate |
| 546 | * map_entry structures within a locked map without blocking trying |
| 547 | * to allocate a new vm_map_entry. |
| 548 | */ |
| 549 | int |
| 550 | vm_map_entry_reserve(int count) |
| 551 | { |
| 552 | struct globaldata *gd = mycpu; |
| 553 | vm_map_entry_t entry; |
| 554 | |
| 555 | crit_enter(); |
| 556 | |
| 557 | /* |
| 558 | * Make sure we have enough structures in gd_vme_base to handle |
| 559 | * the reservation request. |
| 560 | */ |
| 561 | while (gd->gd_vme_avail < count) { |
| 562 | entry = zalloc(mapentzone); |
| 563 | entry->next = gd->gd_vme_base; |
| 564 | gd->gd_vme_base = entry; |
| 565 | ++gd->gd_vme_avail; |
| 566 | } |
| 567 | gd->gd_vme_avail -= count; |
| 568 | crit_exit(); |
| 569 | return(count); |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * vm_map_entry_release: |
| 574 | * |
| 575 | * Releases previously reserved vm_map_entry structures that were not |
| 576 | * used. If we have too much junk in our per-cpu cache clean some of |
| 577 | * it out. |
| 578 | */ |
| 579 | void |
| 580 | vm_map_entry_release(int count) |
| 581 | { |
| 582 | struct globaldata *gd = mycpu; |
| 583 | vm_map_entry_t entry; |
| 584 | |
| 585 | crit_enter(); |
| 586 | gd->gd_vme_avail += count; |
| 587 | while (gd->gd_vme_avail > MAP_RESERVE_SLOP) { |
| 588 | entry = gd->gd_vme_base; |
| 589 | KKASSERT(entry != NULL); |
| 590 | gd->gd_vme_base = entry->next; |
| 591 | --gd->gd_vme_avail; |
| 592 | crit_exit(); |
| 593 | zfree(mapentzone, entry); |
| 594 | crit_enter(); |
| 595 | } |
| 596 | crit_exit(); |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * vm_map_entry_kreserve: |
| 601 | * |
| 602 | * Reserve map entry structures for use in kernel_map itself. These |
| 603 | * entries have *ALREADY* been reserved on a per-cpu basis when the map |
| 604 | * was inited. This function is used by zalloc() to avoid a recursion |
| 605 | * when zalloc() itself needs to allocate additional kernel memory. |
| 606 | * |
| 607 | * This function works like the normal reserve but does not load the |
| 608 | * vm_map_entry cache (because that would result in an infinite |
| 609 | * recursion). Note that gd_vme_avail may go negative. This is expected. |
| 610 | * |
| 611 | * Any caller of this function must be sure to renormalize after |
| 612 | * potentially eating entries to ensure that the reserve supply |
| 613 | * remains intact. |
| 614 | */ |
| 615 | int |
| 616 | vm_map_entry_kreserve(int count) |
| 617 | { |
| 618 | struct globaldata *gd = mycpu; |
| 619 | |
| 620 | crit_enter(); |
| 621 | gd->gd_vme_avail -= count; |
| 622 | crit_exit(); |
| 623 | KASSERT(gd->gd_vme_base != NULL, ("no reserved entries left, gd_vme_avail = %d\n", gd->gd_vme_avail)); |
| 624 | return(count); |
| 625 | } |
| 626 | |
| 627 | /* |
| 628 | * vm_map_entry_krelease: |
| 629 | * |
| 630 | * Release previously reserved map entries for kernel_map. We do not |
| 631 | * attempt to clean up like the normal release function as this would |
| 632 | * cause an unnecessary (but probably not fatal) deep procedure call. |
| 633 | */ |
| 634 | void |
| 635 | vm_map_entry_krelease(int count) |
| 636 | { |
| 637 | struct globaldata *gd = mycpu; |
| 638 | |
| 639 | crit_enter(); |
| 640 | gd->gd_vme_avail += count; |
| 641 | crit_exit(); |
| 642 | } |
| 643 | |
| 644 | /* |
| 645 | * vm_map_entry_create: [ internal use only ] |
| 646 | * |
| 647 | * Allocates a VM map entry for insertion. No entry fields are filled |
| 648 | * in. |
| 649 | * |
| 650 | * This routine may be called from an interrupt thread but not a FAST |
| 651 | * interrupt. This routine may recurse the map lock. |
| 652 | */ |
| 653 | static vm_map_entry_t |
| 654 | vm_map_entry_create(vm_map_t map, int *countp) |
| 655 | { |
| 656 | struct globaldata *gd = mycpu; |
| 657 | vm_map_entry_t entry; |
| 658 | |
| 659 | KKASSERT(*countp > 0); |
| 660 | --*countp; |
| 661 | crit_enter(); |
| 662 | entry = gd->gd_vme_base; |
| 663 | KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp)); |
| 664 | gd->gd_vme_base = entry->next; |
| 665 | crit_exit(); |
| 666 | return(entry); |
| 667 | } |
| 668 | |
| 669 | /* |
| 670 | * vm_map_entry_dispose: [ internal use only ] |
| 671 | * |
| 672 | * Dispose of a vm_map_entry that is no longer being referenced. This |
| 673 | * function may be called from an interrupt. |
| 674 | */ |
| 675 | static void |
| 676 | vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp) |
| 677 | { |
| 678 | struct globaldata *gd = mycpu; |
| 679 | |
| 680 | KKASSERT(map->hint != entry); |
| 681 | KKASSERT(map->first_free != entry); |
| 682 | |
| 683 | ++*countp; |
| 684 | crit_enter(); |
| 685 | entry->next = gd->gd_vme_base; |
| 686 | gd->gd_vme_base = entry; |
| 687 | crit_exit(); |
| 688 | } |
| 689 | |
| 690 | |
| 691 | /* |
| 692 | * vm_map_entry_{un,}link: |
| 693 | * |
| 694 | * Insert/remove entries from maps. |
| 695 | */ |
| 696 | static __inline void |
| 697 | vm_map_entry_link(vm_map_t map, |
| 698 | vm_map_entry_t after_where, |
| 699 | vm_map_entry_t entry) |
| 700 | { |
| 701 | map->nentries++; |
| 702 | entry->prev = after_where; |
| 703 | entry->next = after_where->next; |
| 704 | entry->next->prev = entry; |
| 705 | after_where->next = entry; |
| 706 | if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry)) |
| 707 | panic("vm_map_entry_link: dup addr map %p ent %p", map, entry); |
| 708 | } |
| 709 | |
| 710 | static __inline void |
| 711 | vm_map_entry_unlink(vm_map_t map, |
| 712 | vm_map_entry_t entry) |
| 713 | { |
| 714 | vm_map_entry_t prev; |
| 715 | vm_map_entry_t next; |
| 716 | |
| 717 | if (entry->eflags & MAP_ENTRY_IN_TRANSITION) |
| 718 | panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry); |
| 719 | prev = entry->prev; |
| 720 | next = entry->next; |
| 721 | next->prev = prev; |
| 722 | prev->next = next; |
| 723 | vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry); |
| 724 | map->nentries--; |
| 725 | } |
| 726 | |
| 727 | /* |
| 728 | * vm_map_lookup_entry: [ internal use only ] |
| 729 | * |
| 730 | * Finds the map entry containing (or |
| 731 | * immediately preceding) the specified address |
| 732 | * in the given map; the entry is returned |
| 733 | * in the "entry" parameter. The boolean |
| 734 | * result indicates whether the address is |
| 735 | * actually contained in the map. |
| 736 | */ |
| 737 | boolean_t |
| 738 | vm_map_lookup_entry(vm_map_t map, vm_offset_t address, |
| 739 | vm_map_entry_t *entry /* OUT */) |
| 740 | { |
| 741 | vm_map_entry_t tmp; |
| 742 | vm_map_entry_t last; |
| 743 | |
| 744 | #if 0 |
| 745 | /* |
| 746 | * XXX TEMPORARILY DISABLED. For some reason our attempt to revive |
| 747 | * the hint code with the red-black lookup meets with system crashes |
| 748 | * and lockups. We do not yet know why. |
| 749 | * |
| 750 | * It is possible that the problem is related to the setting |
| 751 | * of the hint during map_entry deletion, in the code specified |
| 752 | * at the GGG comment later on in this file. |
| 753 | */ |
| 754 | /* |
| 755 | * Quickly check the cached hint, there's a good chance of a match. |
| 756 | */ |
| 757 | if (map->hint != &map->header) { |
| 758 | tmp = map->hint; |
| 759 | if (address >= tmp->start && address < tmp->end) { |
| 760 | *entry = tmp; |
| 761 | return(TRUE); |
| 762 | } |
| 763 | } |
| 764 | #endif |
| 765 | |
| 766 | /* |
| 767 | * Locate the record from the top of the tree. 'last' tracks the |
| 768 | * closest prior record and is returned if no match is found, which |
| 769 | * in binary tree terms means tracking the most recent right-branch |
| 770 | * taken. If there is no prior record, &map->header is returned. |
| 771 | */ |
| 772 | last = &map->header; |
| 773 | tmp = RB_ROOT(&map->rb_root); |
| 774 | |
| 775 | while (tmp) { |
| 776 | if (address >= tmp->start) { |
| 777 | if (address < tmp->end) { |
| 778 | *entry = tmp; |
| 779 | map->hint = tmp; |
| 780 | return(TRUE); |
| 781 | } |
| 782 | last = tmp; |
| 783 | tmp = RB_RIGHT(tmp, rb_entry); |
| 784 | } else { |
| 785 | tmp = RB_LEFT(tmp, rb_entry); |
| 786 | } |
| 787 | } |
| 788 | *entry = last; |
| 789 | return (FALSE); |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * vm_map_insert: |
| 794 | * |
| 795 | * Inserts the given whole VM object into the target |
| 796 | * map at the specified address range. The object's |
| 797 | * size should match that of the address range. |
| 798 | * |
| 799 | * Requires that the map be locked, and leaves it so. Requires that |
| 800 | * sufficient vm_map_entry structures have been reserved and tracks |
| 801 | * the use via countp. |
| 802 | * |
| 803 | * If object is non-NULL, ref count must be bumped by caller |
| 804 | * prior to making call to account for the new entry. |
| 805 | */ |
| 806 | int |
| 807 | vm_map_insert(vm_map_t map, int *countp, |
| 808 | vm_object_t object, vm_ooffset_t offset, |
| 809 | vm_offset_t start, vm_offset_t end, |
| 810 | vm_maptype_t maptype, |
| 811 | vm_prot_t prot, vm_prot_t max, |
| 812 | int cow) |
| 813 | { |
| 814 | vm_map_entry_t new_entry; |
| 815 | vm_map_entry_t prev_entry; |
| 816 | vm_map_entry_t temp_entry; |
| 817 | vm_eflags_t protoeflags; |
| 818 | |
| 819 | /* |
| 820 | * Check that the start and end points are not bogus. |
| 821 | */ |
| 822 | |
| 823 | if ((start < map->min_offset) || (end > map->max_offset) || |
| 824 | (start >= end)) |
| 825 | return (KERN_INVALID_ADDRESS); |
| 826 | |
| 827 | /* |
| 828 | * Find the entry prior to the proposed starting address; if it's part |
| 829 | * of an existing entry, this range is bogus. |
| 830 | */ |
| 831 | |
| 832 | if (vm_map_lookup_entry(map, start, &temp_entry)) |
| 833 | return (KERN_NO_SPACE); |
| 834 | |
| 835 | prev_entry = temp_entry; |
| 836 | |
| 837 | /* |
| 838 | * Assert that the next entry doesn't overlap the end point. |
| 839 | */ |
| 840 | |
| 841 | if ((prev_entry->next != &map->header) && |
| 842 | (prev_entry->next->start < end)) |
| 843 | return (KERN_NO_SPACE); |
| 844 | |
| 845 | protoeflags = 0; |
| 846 | |
| 847 | if (cow & MAP_COPY_ON_WRITE) |
| 848 | protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY; |
| 849 | |
| 850 | if (cow & MAP_NOFAULT) { |
| 851 | protoeflags |= MAP_ENTRY_NOFAULT; |
| 852 | |
| 853 | KASSERT(object == NULL, |
| 854 | ("vm_map_insert: paradoxical MAP_NOFAULT request")); |
| 855 | } |
| 856 | if (cow & MAP_DISABLE_SYNCER) |
| 857 | protoeflags |= MAP_ENTRY_NOSYNC; |
| 858 | if (cow & MAP_DISABLE_COREDUMP) |
| 859 | protoeflags |= MAP_ENTRY_NOCOREDUMP; |
| 860 | if (cow & MAP_IS_STACK) |
| 861 | protoeflags |= MAP_ENTRY_STACK; |
| 862 | |
| 863 | if (object) { |
| 864 | /* |
| 865 | * When object is non-NULL, it could be shared with another |
| 866 | * process. We have to set or clear OBJ_ONEMAPPING |
| 867 | * appropriately. |
| 868 | */ |
| 869 | if ((object->ref_count > 1) || (object->shadow_count != 0)) { |
| 870 | vm_object_clear_flag(object, OBJ_ONEMAPPING); |
| 871 | } |
| 872 | } |
| 873 | else if ((prev_entry != &map->header) && |
| 874 | (prev_entry->eflags == protoeflags) && |
| 875 | (prev_entry->end == start) && |
| 876 | (prev_entry->wired_count == 0) && |
| 877 | prev_entry->maptype == maptype && |
| 878 | ((prev_entry->object.vm_object == NULL) || |
| 879 | vm_object_coalesce(prev_entry->object.vm_object, |
| 880 | OFF_TO_IDX(prev_entry->offset), |
| 881 | (vm_size_t)(prev_entry->end - prev_entry->start), |
| 882 | (vm_size_t)(end - prev_entry->end)))) { |
| 883 | /* |
| 884 | * We were able to extend the object. Determine if we |
| 885 | * can extend the previous map entry to include the |
| 886 | * new range as well. |
| 887 | */ |
| 888 | if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) && |
| 889 | (prev_entry->protection == prot) && |
| 890 | (prev_entry->max_protection == max)) { |
| 891 | map->size += (end - prev_entry->end); |
| 892 | prev_entry->end = end; |
| 893 | vm_map_simplify_entry(map, prev_entry, countp); |
| 894 | return (KERN_SUCCESS); |
| 895 | } |
| 896 | |
| 897 | /* |
| 898 | * If we can extend the object but cannot extend the |
| 899 | * map entry, we have to create a new map entry. We |
| 900 | * must bump the ref count on the extended object to |
| 901 | * account for it. object may be NULL. |
| 902 | */ |
| 903 | object = prev_entry->object.vm_object; |
| 904 | offset = prev_entry->offset + |
| 905 | (prev_entry->end - prev_entry->start); |
| 906 | vm_object_reference(object); |
| 907 | } |
| 908 | |
| 909 | /* |
| 910 | * NOTE: if conditionals fail, object can be NULL here. This occurs |
| 911 | * in things like the buffer map where we manage kva but do not manage |
| 912 | * backing objects. |
| 913 | */ |
| 914 | |
| 915 | /* |
| 916 | * Create a new entry |
| 917 | */ |
| 918 | |
| 919 | new_entry = vm_map_entry_create(map, countp); |
| 920 | new_entry->start = start; |
| 921 | new_entry->end = end; |
| 922 | |
| 923 | new_entry->maptype = maptype; |
| 924 | new_entry->eflags = protoeflags; |
| 925 | new_entry->object.vm_object = object; |
| 926 | new_entry->offset = offset; |
| 927 | new_entry->aux.master_pde = 0; |
| 928 | |
| 929 | new_entry->inheritance = VM_INHERIT_DEFAULT; |
| 930 | new_entry->protection = prot; |
| 931 | new_entry->max_protection = max; |
| 932 | new_entry->wired_count = 0; |
| 933 | |
| 934 | /* |
| 935 | * Insert the new entry into the list |
| 936 | */ |
| 937 | |
| 938 | vm_map_entry_link(map, prev_entry, new_entry); |
| 939 | map->size += new_entry->end - new_entry->start; |
| 940 | |
| 941 | /* |
| 942 | * Update the free space hint. Entries cannot overlap. |
| 943 | * An exact comparison is needed to avoid matching |
| 944 | * against the map->header. |
| 945 | */ |
| 946 | if ((map->first_free == prev_entry) && |
| 947 | (prev_entry->end == new_entry->start)) { |
| 948 | map->first_free = new_entry; |
| 949 | } |
| 950 | |
| 951 | #if 0 |
| 952 | /* |
| 953 | * Temporarily removed to avoid MAP_STACK panic, due to |
| 954 | * MAP_STACK being a huge hack. Will be added back in |
| 955 | * when MAP_STACK (and the user stack mapping) is fixed. |
| 956 | */ |
| 957 | /* |
| 958 | * It may be possible to simplify the entry |
| 959 | */ |
| 960 | vm_map_simplify_entry(map, new_entry, countp); |
| 961 | #endif |
| 962 | |
| 963 | /* |
| 964 | * Try to pre-populate the page table. Mappings governed by virtual |
| 965 | * page tables cannot be prepopulated without a lot of work, so |
| 966 | * don't try. |
| 967 | */ |
| 968 | if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) && |
| 969 | maptype != VM_MAPTYPE_VPAGETABLE) { |
| 970 | pmap_object_init_pt(map->pmap, start, prot, |
| 971 | object, OFF_TO_IDX(offset), end - start, |
| 972 | cow & MAP_PREFAULT_PARTIAL); |
| 973 | } |
| 974 | |
| 975 | return (KERN_SUCCESS); |
| 976 | } |
| 977 | |
| 978 | /* |
| 979 | * Find sufficient space for `length' bytes in the given map, starting at |
| 980 | * `start'. The map must be locked. Returns 0 on success, 1 on no space. |
| 981 | * |
| 982 | * This function will returned an arbitrarily aligned pointer. If no |
| 983 | * particular alignment is required you should pass align as 1. Note that |
| 984 | * the map may return PAGE_SIZE aligned pointers if all the lengths used in |
| 985 | * the map are a multiple of PAGE_SIZE, even if you pass a smaller align |
| 986 | * argument. |
| 987 | * |
| 988 | * 'align' should be a power of 2 but is not required to be. |
| 989 | */ |
| 990 | int |
| 991 | vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, |
| 992 | vm_size_t align, int flags, vm_offset_t *addr) |
| 993 | { |
| 994 | vm_map_entry_t entry, next; |
| 995 | vm_offset_t end; |
| 996 | vm_offset_t align_mask; |
| 997 | |
| 998 | if (start < map->min_offset) |
| 999 | start = map->min_offset; |
| 1000 | if (start > map->max_offset) |
| 1001 | return (1); |
| 1002 | |
| 1003 | /* |
| 1004 | * If the alignment is not a power of 2 we will have to use |
| 1005 | * a mod/division, set align_mask to a special value. |
| 1006 | */ |
| 1007 | if ((align | (align - 1)) + 1 != (align << 1)) |
| 1008 | align_mask = (vm_offset_t)-1; |
| 1009 | else |
| 1010 | align_mask = align - 1; |
| 1011 | |
| 1012 | retry: |
| 1013 | /* |
| 1014 | * Look for the first possible address; if there's already something |
| 1015 | * at this address, we have to start after it. |
| 1016 | */ |
| 1017 | if (start == map->min_offset) { |
| 1018 | if ((entry = map->first_free) != &map->header) |
| 1019 | start = entry->end; |
| 1020 | } else { |
| 1021 | vm_map_entry_t tmp; |
| 1022 | |
| 1023 | if (vm_map_lookup_entry(map, start, &tmp)) |
| 1024 | start = tmp->end; |
| 1025 | entry = tmp; |
| 1026 | } |
| 1027 | |
| 1028 | /* |
| 1029 | * Look through the rest of the map, trying to fit a new region in the |
| 1030 | * gap between existing regions, or after the very last region. |
| 1031 | */ |
| 1032 | for (;; start = (entry = next)->end) { |
| 1033 | /* |
| 1034 | * Adjust the proposed start by the requested alignment, |
| 1035 | * be sure that we didn't wrap the address. |
| 1036 | */ |
| 1037 | if (align_mask == (vm_offset_t)-1) |
| 1038 | end = ((start + align - 1) / align) * align; |
| 1039 | else |
| 1040 | end = (start + align_mask) & ~align_mask; |
| 1041 | if (end < start) |
| 1042 | return (1); |
| 1043 | start = end; |
| 1044 | /* |
| 1045 | * Find the end of the proposed new region. Be sure we didn't |
| 1046 | * go beyond the end of the map, or wrap around the address. |
| 1047 | * Then check to see if this is the last entry or if the |
| 1048 | * proposed end fits in the gap between this and the next |
| 1049 | * entry. |
| 1050 | */ |
| 1051 | end = start + length; |
| 1052 | if (end > map->max_offset || end < start) |
| 1053 | return (1); |
| 1054 | next = entry->next; |
| 1055 | |
| 1056 | /* |
| 1057 | * If the next entry's start address is beyond the desired |
| 1058 | * end address we may have found a good entry. |
| 1059 | * |
| 1060 | * If the next entry is a stack mapping we do not map into |
| 1061 | * the stack's reserved space. |
| 1062 | * |
| 1063 | * XXX continue to allow mapping into the stack's reserved |
| 1064 | * space if doing a MAP_STACK mapping inside a MAP_STACK |
| 1065 | * mapping, for backwards compatibility. But the caller |
| 1066 | * really should use MAP_STACK | MAP_TRYFIXED if they |
| 1067 | * want to do that. |
| 1068 | */ |
| 1069 | if (next == &map->header) |
| 1070 | break; |
| 1071 | if (next->start >= end) { |
| 1072 | if ((next->eflags & MAP_ENTRY_STACK) == 0) |
| 1073 | break; |
| 1074 | if (flags & MAP_STACK) |
| 1075 | break; |
| 1076 | if (next->start - next->aux.avail_ssize >= end) |
| 1077 | break; |
| 1078 | } |
| 1079 | } |
| 1080 | map->hint = entry; |
| 1081 | if (map == &kernel_map) { |
| 1082 | vm_offset_t ksize; |
| 1083 | if ((ksize = round_page(start + length)) > kernel_vm_end) { |
| 1084 | pmap_growkernel(ksize); |
| 1085 | goto retry; |
| 1086 | } |
| 1087 | } |
| 1088 | *addr = start; |
| 1089 | return (0); |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * vm_map_find finds an unallocated region in the target address |
| 1094 | * map with the given length. The search is defined to be |
| 1095 | * first-fit from the specified address; the region found is |
| 1096 | * returned in the same parameter. |
| 1097 | * |
| 1098 | * If object is non-NULL, ref count must be bumped by caller |
| 1099 | * prior to making call to account for the new entry. |
| 1100 | */ |
| 1101 | int |
| 1102 | vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, |
| 1103 | vm_offset_t *addr, vm_size_t length, vm_size_t align, |
| 1104 | boolean_t fitit, |
| 1105 | vm_maptype_t maptype, |
| 1106 | vm_prot_t prot, vm_prot_t max, |
| 1107 | int cow) |
| 1108 | { |
| 1109 | vm_offset_t start; |
| 1110 | int result; |
| 1111 | int count; |
| 1112 | |
| 1113 | start = *addr; |
| 1114 | |
| 1115 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1116 | vm_map_lock(map); |
| 1117 | if (fitit) { |
| 1118 | if (vm_map_findspace(map, start, length, align, 0, addr)) { |
| 1119 | vm_map_unlock(map); |
| 1120 | vm_map_entry_release(count); |
| 1121 | return (KERN_NO_SPACE); |
| 1122 | } |
| 1123 | start = *addr; |
| 1124 | } |
| 1125 | result = vm_map_insert(map, &count, object, offset, |
| 1126 | start, start + length, |
| 1127 | maptype, |
| 1128 | prot, max, |
| 1129 | cow); |
| 1130 | vm_map_unlock(map); |
| 1131 | vm_map_entry_release(count); |
| 1132 | |
| 1133 | return (result); |
| 1134 | } |
| 1135 | |
| 1136 | /* |
| 1137 | * vm_map_simplify_entry: |
| 1138 | * |
| 1139 | * Simplify the given map entry by merging with either neighbor. This |
| 1140 | * routine also has the ability to merge with both neighbors. |
| 1141 | * |
| 1142 | * The map must be locked. |
| 1143 | * |
| 1144 | * This routine guarentees that the passed entry remains valid (though |
| 1145 | * possibly extended). When merging, this routine may delete one or |
| 1146 | * both neighbors. No action is taken on entries which have their |
| 1147 | * in-transition flag set. |
| 1148 | */ |
| 1149 | void |
| 1150 | vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp) |
| 1151 | { |
| 1152 | vm_map_entry_t next, prev; |
| 1153 | vm_size_t prevsize, esize; |
| 1154 | |
| 1155 | if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { |
| 1156 | ++mycpu->gd_cnt.v_intrans_coll; |
| 1157 | return; |
| 1158 | } |
| 1159 | |
| 1160 | if (entry->maptype == VM_MAPTYPE_SUBMAP) |
| 1161 | return; |
| 1162 | |
| 1163 | prev = entry->prev; |
| 1164 | if (prev != &map->header) { |
| 1165 | prevsize = prev->end - prev->start; |
| 1166 | if ( (prev->end == entry->start) && |
| 1167 | (prev->maptype == entry->maptype) && |
| 1168 | (prev->object.vm_object == entry->object.vm_object) && |
| 1169 | (!prev->object.vm_object || |
| 1170 | (prev->offset + prevsize == entry->offset)) && |
| 1171 | (prev->eflags == entry->eflags) && |
| 1172 | (prev->protection == entry->protection) && |
| 1173 | (prev->max_protection == entry->max_protection) && |
| 1174 | (prev->inheritance == entry->inheritance) && |
| 1175 | (prev->wired_count == entry->wired_count)) { |
| 1176 | if (map->first_free == prev) |
| 1177 | map->first_free = entry; |
| 1178 | if (map->hint == prev) |
| 1179 | map->hint = entry; |
| 1180 | vm_map_entry_unlink(map, prev); |
| 1181 | entry->start = prev->start; |
| 1182 | entry->offset = prev->offset; |
| 1183 | if (prev->object.vm_object) |
| 1184 | vm_object_deallocate(prev->object.vm_object); |
| 1185 | vm_map_entry_dispose(map, prev, countp); |
| 1186 | } |
| 1187 | } |
| 1188 | |
| 1189 | next = entry->next; |
| 1190 | if (next != &map->header) { |
| 1191 | esize = entry->end - entry->start; |
| 1192 | if ((entry->end == next->start) && |
| 1193 | (next->maptype == entry->maptype) && |
| 1194 | (next->object.vm_object == entry->object.vm_object) && |
| 1195 | (!entry->object.vm_object || |
| 1196 | (entry->offset + esize == next->offset)) && |
| 1197 | (next->eflags == entry->eflags) && |
| 1198 | (next->protection == entry->protection) && |
| 1199 | (next->max_protection == entry->max_protection) && |
| 1200 | (next->inheritance == entry->inheritance) && |
| 1201 | (next->wired_count == entry->wired_count)) { |
| 1202 | if (map->first_free == next) |
| 1203 | map->first_free = entry; |
| 1204 | if (map->hint == next) |
| 1205 | map->hint = entry; |
| 1206 | vm_map_entry_unlink(map, next); |
| 1207 | entry->end = next->end; |
| 1208 | if (next->object.vm_object) |
| 1209 | vm_object_deallocate(next->object.vm_object); |
| 1210 | vm_map_entry_dispose(map, next, countp); |
| 1211 | } |
| 1212 | } |
| 1213 | } |
| 1214 | /* |
| 1215 | * vm_map_clip_start: [ internal use only ] |
| 1216 | * |
| 1217 | * Asserts that the given entry begins at or after |
| 1218 | * the specified address; if necessary, |
| 1219 | * it splits the entry into two. |
| 1220 | */ |
| 1221 | #define vm_map_clip_start(map, entry, startaddr, countp) \ |
| 1222 | { \ |
| 1223 | if (startaddr > entry->start) \ |
| 1224 | _vm_map_clip_start(map, entry, startaddr, countp); \ |
| 1225 | } |
| 1226 | |
| 1227 | /* |
| 1228 | * This routine is called only when it is known that |
| 1229 | * the entry must be split. |
| 1230 | */ |
| 1231 | static void |
| 1232 | _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start, int *countp) |
| 1233 | { |
| 1234 | vm_map_entry_t new_entry; |
| 1235 | |
| 1236 | /* |
| 1237 | * Split off the front portion -- note that we must insert the new |
| 1238 | * entry BEFORE this one, so that this entry has the specified |
| 1239 | * starting address. |
| 1240 | */ |
| 1241 | |
| 1242 | vm_map_simplify_entry(map, entry, countp); |
| 1243 | |
| 1244 | /* |
| 1245 | * If there is no object backing this entry, we might as well create |
| 1246 | * one now. If we defer it, an object can get created after the map |
| 1247 | * is clipped, and individual objects will be created for the split-up |
| 1248 | * map. This is a bit of a hack, but is also about the best place to |
| 1249 | * put this improvement. |
| 1250 | */ |
| 1251 | if (entry->object.vm_object == NULL && !map->system_map) { |
| 1252 | vm_map_entry_allocate_object(entry); |
| 1253 | } |
| 1254 | |
| 1255 | new_entry = vm_map_entry_create(map, countp); |
| 1256 | *new_entry = *entry; |
| 1257 | |
| 1258 | new_entry->end = start; |
| 1259 | entry->offset += (start - entry->start); |
| 1260 | entry->start = start; |
| 1261 | |
| 1262 | vm_map_entry_link(map, entry->prev, new_entry); |
| 1263 | |
| 1264 | switch(entry->maptype) { |
| 1265 | case VM_MAPTYPE_NORMAL: |
| 1266 | case VM_MAPTYPE_VPAGETABLE: |
| 1267 | vm_object_reference(new_entry->object.vm_object); |
| 1268 | break; |
| 1269 | default: |
| 1270 | break; |
| 1271 | } |
| 1272 | } |
| 1273 | |
| 1274 | /* |
| 1275 | * vm_map_clip_end: [ internal use only ] |
| 1276 | * |
| 1277 | * Asserts that the given entry ends at or before |
| 1278 | * the specified address; if necessary, |
| 1279 | * it splits the entry into two. |
| 1280 | */ |
| 1281 | |
| 1282 | #define vm_map_clip_end(map, entry, endaddr, countp) \ |
| 1283 | { \ |
| 1284 | if (endaddr < entry->end) \ |
| 1285 | _vm_map_clip_end(map, entry, endaddr, countp); \ |
| 1286 | } |
| 1287 | |
| 1288 | /* |
| 1289 | * This routine is called only when it is known that |
| 1290 | * the entry must be split. |
| 1291 | */ |
| 1292 | static void |
| 1293 | _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end, int *countp) |
| 1294 | { |
| 1295 | vm_map_entry_t new_entry; |
| 1296 | |
| 1297 | /* |
| 1298 | * If there is no object backing this entry, we might as well create |
| 1299 | * one now. If we defer it, an object can get created after the map |
| 1300 | * is clipped, and individual objects will be created for the split-up |
| 1301 | * map. This is a bit of a hack, but is also about the best place to |
| 1302 | * put this improvement. |
| 1303 | */ |
| 1304 | |
| 1305 | if (entry->object.vm_object == NULL && !map->system_map) { |
| 1306 | vm_map_entry_allocate_object(entry); |
| 1307 | } |
| 1308 | |
| 1309 | /* |
| 1310 | * Create a new entry and insert it AFTER the specified entry |
| 1311 | */ |
| 1312 | |
| 1313 | new_entry = vm_map_entry_create(map, countp); |
| 1314 | *new_entry = *entry; |
| 1315 | |
| 1316 | new_entry->start = entry->end = end; |
| 1317 | new_entry->offset += (end - entry->start); |
| 1318 | |
| 1319 | vm_map_entry_link(map, entry, new_entry); |
| 1320 | |
| 1321 | switch(entry->maptype) { |
| 1322 | case VM_MAPTYPE_NORMAL: |
| 1323 | case VM_MAPTYPE_VPAGETABLE: |
| 1324 | vm_object_reference(new_entry->object.vm_object); |
| 1325 | break; |
| 1326 | default: |
| 1327 | break; |
| 1328 | } |
| 1329 | } |
| 1330 | |
| 1331 | /* |
| 1332 | * VM_MAP_RANGE_CHECK: [ internal use only ] |
| 1333 | * |
| 1334 | * Asserts that the starting and ending region |
| 1335 | * addresses fall within the valid range of the map. |
| 1336 | */ |
| 1337 | #define VM_MAP_RANGE_CHECK(map, start, end) \ |
| 1338 | { \ |
| 1339 | if (start < vm_map_min(map)) \ |
| 1340 | start = vm_map_min(map); \ |
| 1341 | if (end > vm_map_max(map)) \ |
| 1342 | end = vm_map_max(map); \ |
| 1343 | if (start > end) \ |
| 1344 | start = end; \ |
| 1345 | } |
| 1346 | |
| 1347 | /* |
| 1348 | * vm_map_transition_wait: [ kernel use only ] |
| 1349 | * |
| 1350 | * Used to block when an in-transition collison occurs. The map |
| 1351 | * is unlocked for the sleep and relocked before the return. |
| 1352 | */ |
| 1353 | static |
| 1354 | void |
| 1355 | vm_map_transition_wait(vm_map_t map) |
| 1356 | { |
| 1357 | vm_map_unlock(map); |
| 1358 | tsleep(map, 0, "vment", 0); |
| 1359 | vm_map_lock(map); |
| 1360 | } |
| 1361 | |
| 1362 | /* |
| 1363 | * CLIP_CHECK_BACK |
| 1364 | * CLIP_CHECK_FWD |
| 1365 | * |
| 1366 | * When we do blocking operations with the map lock held it is |
| 1367 | * possible that a clip might have occured on our in-transit entry, |
| 1368 | * requiring an adjustment to the entry in our loop. These macros |
| 1369 | * help the pageable and clip_range code deal with the case. The |
| 1370 | * conditional costs virtually nothing if no clipping has occured. |
| 1371 | */ |
| 1372 | |
| 1373 | #define CLIP_CHECK_BACK(entry, save_start) \ |
| 1374 | do { \ |
| 1375 | while (entry->start != save_start) { \ |
| 1376 | entry = entry->prev; \ |
| 1377 | KASSERT(entry != &map->header, ("bad entry clip")); \ |
| 1378 | } \ |
| 1379 | } while(0) |
| 1380 | |
| 1381 | #define CLIP_CHECK_FWD(entry, save_end) \ |
| 1382 | do { \ |
| 1383 | while (entry->end != save_end) { \ |
| 1384 | entry = entry->next; \ |
| 1385 | KASSERT(entry != &map->header, ("bad entry clip")); \ |
| 1386 | } \ |
| 1387 | } while(0) |
| 1388 | |
| 1389 | |
| 1390 | /* |
| 1391 | * vm_map_clip_range: [ kernel use only ] |
| 1392 | * |
| 1393 | * Clip the specified range and return the base entry. The |
| 1394 | * range may cover several entries starting at the returned base |
| 1395 | * and the first and last entry in the covering sequence will be |
| 1396 | * properly clipped to the requested start and end address. |
| 1397 | * |
| 1398 | * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES |
| 1399 | * flag. |
| 1400 | * |
| 1401 | * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries |
| 1402 | * covered by the requested range. |
| 1403 | * |
| 1404 | * The map must be exclusively locked on entry and will remain locked |
| 1405 | * on return. If no range exists or the range contains holes and you |
| 1406 | * specified that no holes were allowed, NULL will be returned. This |
| 1407 | * routine may temporarily unlock the map in order avoid a deadlock when |
| 1408 | * sleeping. |
| 1409 | */ |
| 1410 | static |
| 1411 | vm_map_entry_t |
| 1412 | vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 1413 | int *countp, int flags) |
| 1414 | { |
| 1415 | vm_map_entry_t start_entry; |
| 1416 | vm_map_entry_t entry; |
| 1417 | |
| 1418 | /* |
| 1419 | * Locate the entry and effect initial clipping. The in-transition |
| 1420 | * case does not occur very often so do not try to optimize it. |
| 1421 | */ |
| 1422 | again: |
| 1423 | if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) |
| 1424 | return (NULL); |
| 1425 | entry = start_entry; |
| 1426 | if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { |
| 1427 | entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; |
| 1428 | ++mycpu->gd_cnt.v_intrans_coll; |
| 1429 | ++mycpu->gd_cnt.v_intrans_wait; |
| 1430 | vm_map_transition_wait(map); |
| 1431 | /* |
| 1432 | * entry and/or start_entry may have been clipped while |
| 1433 | * we slept, or may have gone away entirely. We have |
| 1434 | * to restart from the lookup. |
| 1435 | */ |
| 1436 | goto again; |
| 1437 | } |
| 1438 | /* |
| 1439 | * Since we hold an exclusive map lock we do not have to restart |
| 1440 | * after clipping, even though clipping may block in zalloc. |
| 1441 | */ |
| 1442 | vm_map_clip_start(map, entry, start, countp); |
| 1443 | vm_map_clip_end(map, entry, end, countp); |
| 1444 | entry->eflags |= MAP_ENTRY_IN_TRANSITION; |
| 1445 | |
| 1446 | /* |
| 1447 | * Scan entries covered by the range. When working on the next |
| 1448 | * entry a restart need only re-loop on the current entry which |
| 1449 | * we have already locked, since 'next' may have changed. Also, |
| 1450 | * even though entry is safe, it may have been clipped so we |
| 1451 | * have to iterate forwards through the clip after sleeping. |
| 1452 | */ |
| 1453 | while (entry->next != &map->header && entry->next->start < end) { |
| 1454 | vm_map_entry_t next = entry->next; |
| 1455 | |
| 1456 | if (flags & MAP_CLIP_NO_HOLES) { |
| 1457 | if (next->start > entry->end) { |
| 1458 | vm_map_unclip_range(map, start_entry, |
| 1459 | start, entry->end, countp, flags); |
| 1460 | return(NULL); |
| 1461 | } |
| 1462 | } |
| 1463 | |
| 1464 | if (next->eflags & MAP_ENTRY_IN_TRANSITION) { |
| 1465 | vm_offset_t save_end = entry->end; |
| 1466 | next->eflags |= MAP_ENTRY_NEEDS_WAKEUP; |
| 1467 | ++mycpu->gd_cnt.v_intrans_coll; |
| 1468 | ++mycpu->gd_cnt.v_intrans_wait; |
| 1469 | vm_map_transition_wait(map); |
| 1470 | |
| 1471 | /* |
| 1472 | * clips might have occured while we blocked. |
| 1473 | */ |
| 1474 | CLIP_CHECK_FWD(entry, save_end); |
| 1475 | CLIP_CHECK_BACK(start_entry, start); |
| 1476 | continue; |
| 1477 | } |
| 1478 | /* |
| 1479 | * No restart necessary even though clip_end may block, we |
| 1480 | * are holding the map lock. |
| 1481 | */ |
| 1482 | vm_map_clip_end(map, next, end, countp); |
| 1483 | next->eflags |= MAP_ENTRY_IN_TRANSITION; |
| 1484 | entry = next; |
| 1485 | } |
| 1486 | if (flags & MAP_CLIP_NO_HOLES) { |
| 1487 | if (entry->end != end) { |
| 1488 | vm_map_unclip_range(map, start_entry, |
| 1489 | start, entry->end, countp, flags); |
| 1490 | return(NULL); |
| 1491 | } |
| 1492 | } |
| 1493 | return(start_entry); |
| 1494 | } |
| 1495 | |
| 1496 | /* |
| 1497 | * vm_map_unclip_range: [ kernel use only ] |
| 1498 | * |
| 1499 | * Undo the effect of vm_map_clip_range(). You should pass the same |
| 1500 | * flags and the same range that you passed to vm_map_clip_range(). |
| 1501 | * This code will clear the in-transition flag on the entries and |
| 1502 | * wake up anyone waiting. This code will also simplify the sequence |
| 1503 | * and attempt to merge it with entries before and after the sequence. |
| 1504 | * |
| 1505 | * The map must be locked on entry and will remain locked on return. |
| 1506 | * |
| 1507 | * Note that you should also pass the start_entry returned by |
| 1508 | * vm_map_clip_range(). However, if you block between the two calls |
| 1509 | * with the map unlocked please be aware that the start_entry may |
| 1510 | * have been clipped and you may need to scan it backwards to find |
| 1511 | * the entry corresponding with the original start address. You are |
| 1512 | * responsible for this, vm_map_unclip_range() expects the correct |
| 1513 | * start_entry to be passed to it and will KASSERT otherwise. |
| 1514 | */ |
| 1515 | static |
| 1516 | void |
| 1517 | vm_map_unclip_range( |
| 1518 | vm_map_t map, |
| 1519 | vm_map_entry_t start_entry, |
| 1520 | vm_offset_t start, |
| 1521 | vm_offset_t end, |
| 1522 | int *countp, |
| 1523 | int flags) |
| 1524 | { |
| 1525 | vm_map_entry_t entry; |
| 1526 | |
| 1527 | entry = start_entry; |
| 1528 | |
| 1529 | KASSERT(entry->start == start, ("unclip_range: illegal base entry")); |
| 1530 | while (entry != &map->header && entry->start < end) { |
| 1531 | KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry)); |
| 1532 | KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped")); |
| 1533 | entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; |
| 1534 | if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { |
| 1535 | entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; |
| 1536 | wakeup(map); |
| 1537 | } |
| 1538 | entry = entry->next; |
| 1539 | } |
| 1540 | |
| 1541 | /* |
| 1542 | * Simplification does not block so there is no restart case. |
| 1543 | */ |
| 1544 | entry = start_entry; |
| 1545 | while (entry != &map->header && entry->start < end) { |
| 1546 | vm_map_simplify_entry(map, entry, countp); |
| 1547 | entry = entry->next; |
| 1548 | } |
| 1549 | } |
| 1550 | |
| 1551 | /* |
| 1552 | * vm_map_submap: [ kernel use only ] |
| 1553 | * |
| 1554 | * Mark the given range as handled by a subordinate map. |
| 1555 | * |
| 1556 | * This range must have been created with vm_map_find, |
| 1557 | * and no other operations may have been performed on this |
| 1558 | * range prior to calling vm_map_submap. |
| 1559 | * |
| 1560 | * Only a limited number of operations can be performed |
| 1561 | * within this rage after calling vm_map_submap: |
| 1562 | * vm_fault |
| 1563 | * [Don't try vm_map_copy!] |
| 1564 | * |
| 1565 | * To remove a submapping, one must first remove the |
| 1566 | * range from the superior map, and then destroy the |
| 1567 | * submap (if desired). [Better yet, don't try it.] |
| 1568 | */ |
| 1569 | int |
| 1570 | vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap) |
| 1571 | { |
| 1572 | vm_map_entry_t entry; |
| 1573 | int result = KERN_INVALID_ARGUMENT; |
| 1574 | int count; |
| 1575 | |
| 1576 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1577 | vm_map_lock(map); |
| 1578 | |
| 1579 | VM_MAP_RANGE_CHECK(map, start, end); |
| 1580 | |
| 1581 | if (vm_map_lookup_entry(map, start, &entry)) { |
| 1582 | vm_map_clip_start(map, entry, start, &count); |
| 1583 | } else { |
| 1584 | entry = entry->next; |
| 1585 | } |
| 1586 | |
| 1587 | vm_map_clip_end(map, entry, end, &count); |
| 1588 | |
| 1589 | if ((entry->start == start) && (entry->end == end) && |
| 1590 | ((entry->eflags & MAP_ENTRY_COW) == 0) && |
| 1591 | (entry->object.vm_object == NULL)) { |
| 1592 | entry->object.sub_map = submap; |
| 1593 | entry->maptype = VM_MAPTYPE_SUBMAP; |
| 1594 | result = KERN_SUCCESS; |
| 1595 | } |
| 1596 | vm_map_unlock(map); |
| 1597 | vm_map_entry_release(count); |
| 1598 | |
| 1599 | return (result); |
| 1600 | } |
| 1601 | |
| 1602 | /* |
| 1603 | * vm_map_protect: |
| 1604 | * |
| 1605 | * Sets the protection of the specified address region in the target map. |
| 1606 | * If "set_max" is specified, the maximum protection is to be set; |
| 1607 | * otherwise, only the current protection is affected. |
| 1608 | * |
| 1609 | * The protection is not applicable to submaps, but is applicable to normal |
| 1610 | * maps and maps governed by virtual page tables. For example, when operating |
| 1611 | * on a virtual page table our protection basically controls how COW occurs |
| 1612 | * on the backing object, whereas the virtual page table abstraction itself |
| 1613 | * is an abstraction for userland. |
| 1614 | */ |
| 1615 | int |
| 1616 | vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 1617 | vm_prot_t new_prot, boolean_t set_max) |
| 1618 | { |
| 1619 | vm_map_entry_t current; |
| 1620 | vm_map_entry_t entry; |
| 1621 | int count; |
| 1622 | |
| 1623 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1624 | vm_map_lock(map); |
| 1625 | |
| 1626 | VM_MAP_RANGE_CHECK(map, start, end); |
| 1627 | |
| 1628 | if (vm_map_lookup_entry(map, start, &entry)) { |
| 1629 | vm_map_clip_start(map, entry, start, &count); |
| 1630 | } else { |
| 1631 | entry = entry->next; |
| 1632 | } |
| 1633 | |
| 1634 | /* |
| 1635 | * Make a first pass to check for protection violations. |
| 1636 | */ |
| 1637 | current = entry; |
| 1638 | while ((current != &map->header) && (current->start < end)) { |
| 1639 | if (current->maptype == VM_MAPTYPE_SUBMAP) { |
| 1640 | vm_map_unlock(map); |
| 1641 | vm_map_entry_release(count); |
| 1642 | return (KERN_INVALID_ARGUMENT); |
| 1643 | } |
| 1644 | if ((new_prot & current->max_protection) != new_prot) { |
| 1645 | vm_map_unlock(map); |
| 1646 | vm_map_entry_release(count); |
| 1647 | return (KERN_PROTECTION_FAILURE); |
| 1648 | } |
| 1649 | current = current->next; |
| 1650 | } |
| 1651 | |
| 1652 | /* |
| 1653 | * Go back and fix up protections. [Note that clipping is not |
| 1654 | * necessary the second time.] |
| 1655 | */ |
| 1656 | current = entry; |
| 1657 | |
| 1658 | while ((current != &map->header) && (current->start < end)) { |
| 1659 | vm_prot_t old_prot; |
| 1660 | |
| 1661 | vm_map_clip_end(map, current, end, &count); |
| 1662 | |
| 1663 | old_prot = current->protection; |
| 1664 | if (set_max) { |
| 1665 | current->protection = |
| 1666 | (current->max_protection = new_prot) & |
| 1667 | old_prot; |
| 1668 | } else { |
| 1669 | current->protection = new_prot; |
| 1670 | } |
| 1671 | |
| 1672 | /* |
| 1673 | * Update physical map if necessary. Worry about copy-on-write |
| 1674 | * here -- CHECK THIS XXX |
| 1675 | */ |
| 1676 | |
| 1677 | if (current->protection != old_prot) { |
| 1678 | #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ |
| 1679 | VM_PROT_ALL) |
| 1680 | |
| 1681 | pmap_protect(map->pmap, current->start, |
| 1682 | current->end, |
| 1683 | current->protection & MASK(current)); |
| 1684 | #undef MASK |
| 1685 | } |
| 1686 | |
| 1687 | vm_map_simplify_entry(map, current, &count); |
| 1688 | |
| 1689 | current = current->next; |
| 1690 | } |
| 1691 | |
| 1692 | vm_map_unlock(map); |
| 1693 | vm_map_entry_release(count); |
| 1694 | return (KERN_SUCCESS); |
| 1695 | } |
| 1696 | |
| 1697 | /* |
| 1698 | * vm_map_madvise: |
| 1699 | * |
| 1700 | * This routine traverses a processes map handling the madvise |
| 1701 | * system call. Advisories are classified as either those effecting |
| 1702 | * the vm_map_entry structure, or those effecting the underlying |
| 1703 | * objects. |
| 1704 | * |
| 1705 | * The <value> argument is used for extended madvise calls. |
| 1706 | */ |
| 1707 | int |
| 1708 | vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 1709 | int behav, off_t value) |
| 1710 | { |
| 1711 | vm_map_entry_t current, entry; |
| 1712 | int modify_map = 0; |
| 1713 | int error = 0; |
| 1714 | int count; |
| 1715 | |
| 1716 | /* |
| 1717 | * Some madvise calls directly modify the vm_map_entry, in which case |
| 1718 | * we need to use an exclusive lock on the map and we need to perform |
| 1719 | * various clipping operations. Otherwise we only need a read-lock |
| 1720 | * on the map. |
| 1721 | */ |
| 1722 | |
| 1723 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1724 | |
| 1725 | switch(behav) { |
| 1726 | case MADV_NORMAL: |
| 1727 | case MADV_SEQUENTIAL: |
| 1728 | case MADV_RANDOM: |
| 1729 | case MADV_NOSYNC: |
| 1730 | case MADV_AUTOSYNC: |
| 1731 | case MADV_NOCORE: |
| 1732 | case MADV_CORE: |
| 1733 | case MADV_SETMAP: |
| 1734 | case MADV_INVAL: |
| 1735 | modify_map = 1; |
| 1736 | vm_map_lock(map); |
| 1737 | break; |
| 1738 | case MADV_WILLNEED: |
| 1739 | case MADV_DONTNEED: |
| 1740 | case MADV_FREE: |
| 1741 | vm_map_lock_read(map); |
| 1742 | break; |
| 1743 | default: |
| 1744 | vm_map_entry_release(count); |
| 1745 | return (EINVAL); |
| 1746 | } |
| 1747 | |
| 1748 | /* |
| 1749 | * Locate starting entry and clip if necessary. |
| 1750 | */ |
| 1751 | |
| 1752 | VM_MAP_RANGE_CHECK(map, start, end); |
| 1753 | |
| 1754 | if (vm_map_lookup_entry(map, start, &entry)) { |
| 1755 | if (modify_map) |
| 1756 | vm_map_clip_start(map, entry, start, &count); |
| 1757 | } else { |
| 1758 | entry = entry->next; |
| 1759 | } |
| 1760 | |
| 1761 | if (modify_map) { |
| 1762 | /* |
| 1763 | * madvise behaviors that are implemented in the vm_map_entry. |
| 1764 | * |
| 1765 | * We clip the vm_map_entry so that behavioral changes are |
| 1766 | * limited to the specified address range. |
| 1767 | */ |
| 1768 | for (current = entry; |
| 1769 | (current != &map->header) && (current->start < end); |
| 1770 | current = current->next |
| 1771 | ) { |
| 1772 | if (current->maptype == VM_MAPTYPE_SUBMAP) |
| 1773 | continue; |
| 1774 | |
| 1775 | vm_map_clip_end(map, current, end, &count); |
| 1776 | |
| 1777 | switch (behav) { |
| 1778 | case MADV_NORMAL: |
| 1779 | vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); |
| 1780 | break; |
| 1781 | case MADV_SEQUENTIAL: |
| 1782 | vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); |
| 1783 | break; |
| 1784 | case MADV_RANDOM: |
| 1785 | vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); |
| 1786 | break; |
| 1787 | case MADV_NOSYNC: |
| 1788 | current->eflags |= MAP_ENTRY_NOSYNC; |
| 1789 | break; |
| 1790 | case MADV_AUTOSYNC: |
| 1791 | current->eflags &= ~MAP_ENTRY_NOSYNC; |
| 1792 | break; |
| 1793 | case MADV_NOCORE: |
| 1794 | current->eflags |= MAP_ENTRY_NOCOREDUMP; |
| 1795 | break; |
| 1796 | case MADV_CORE: |
| 1797 | current->eflags &= ~MAP_ENTRY_NOCOREDUMP; |
| 1798 | break; |
| 1799 | case MADV_INVAL: |
| 1800 | /* |
| 1801 | * Invalidate the related pmap entries, used |
| 1802 | * to flush portions of the real kernel's |
| 1803 | * pmap when the caller has removed or |
| 1804 | * modified existing mappings in a virtual |
| 1805 | * page table. |
| 1806 | */ |
| 1807 | pmap_remove(map->pmap, |
| 1808 | current->start, current->end); |
| 1809 | break; |
| 1810 | case MADV_SETMAP: |
| 1811 | /* |
| 1812 | * Set the page directory page for a map |
| 1813 | * governed by a virtual page table. Mark |
| 1814 | * the entry as being governed by a virtual |
| 1815 | * page table if it is not. |
| 1816 | * |
| 1817 | * XXX the page directory page is stored |
| 1818 | * in the avail_ssize field if the map_entry. |
| 1819 | * |
| 1820 | * XXX the map simplification code does not |
| 1821 | * compare this field so weird things may |
| 1822 | * happen if you do not apply this function |
| 1823 | * to the entire mapping governed by the |
| 1824 | * virtual page table. |
| 1825 | */ |
| 1826 | if (current->maptype != VM_MAPTYPE_VPAGETABLE) { |
| 1827 | error = EINVAL; |
| 1828 | break; |
| 1829 | } |
| 1830 | current->aux.master_pde = value; |
| 1831 | pmap_remove(map->pmap, |
| 1832 | current->start, current->end); |
| 1833 | break; |
| 1834 | default: |
| 1835 | error = EINVAL; |
| 1836 | break; |
| 1837 | } |
| 1838 | vm_map_simplify_entry(map, current, &count); |
| 1839 | } |
| 1840 | vm_map_unlock(map); |
| 1841 | } else { |
| 1842 | vm_pindex_t pindex; |
| 1843 | int count; |
| 1844 | |
| 1845 | /* |
| 1846 | * madvise behaviors that are implemented in the underlying |
| 1847 | * vm_object. |
| 1848 | * |
| 1849 | * Since we don't clip the vm_map_entry, we have to clip |
| 1850 | * the vm_object pindex and count. |
| 1851 | * |
| 1852 | * NOTE! We currently do not support these functions on |
| 1853 | * virtual page tables. |
| 1854 | */ |
| 1855 | for (current = entry; |
| 1856 | (current != &map->header) && (current->start < end); |
| 1857 | current = current->next |
| 1858 | ) { |
| 1859 | vm_offset_t useStart; |
| 1860 | |
| 1861 | if (current->maptype != VM_MAPTYPE_NORMAL) |
| 1862 | continue; |
| 1863 | |
| 1864 | pindex = OFF_TO_IDX(current->offset); |
| 1865 | count = atop(current->end - current->start); |
| 1866 | useStart = current->start; |
| 1867 | |
| 1868 | if (current->start < start) { |
| 1869 | pindex += atop(start - current->start); |
| 1870 | count -= atop(start - current->start); |
| 1871 | useStart = start; |
| 1872 | } |
| 1873 | if (current->end > end) |
| 1874 | count -= atop(current->end - end); |
| 1875 | |
| 1876 | if (count <= 0) |
| 1877 | continue; |
| 1878 | |
| 1879 | vm_object_madvise(current->object.vm_object, |
| 1880 | pindex, count, behav); |
| 1881 | |
| 1882 | /* |
| 1883 | * Try to populate the page table. Mappings governed |
| 1884 | * by virtual page tables cannot be pre-populated |
| 1885 | * without a lot of work so don't try. |
| 1886 | */ |
| 1887 | if (behav == MADV_WILLNEED && |
| 1888 | current->maptype != VM_MAPTYPE_VPAGETABLE) { |
| 1889 | pmap_object_init_pt( |
| 1890 | map->pmap, |
| 1891 | useStart, |
| 1892 | current->protection, |
| 1893 | current->object.vm_object, |
| 1894 | pindex, |
| 1895 | (count << PAGE_SHIFT), |
| 1896 | MAP_PREFAULT_MADVISE |
| 1897 | ); |
| 1898 | } |
| 1899 | } |
| 1900 | vm_map_unlock_read(map); |
| 1901 | } |
| 1902 | vm_map_entry_release(count); |
| 1903 | return(error); |
| 1904 | } |
| 1905 | |
| 1906 | |
| 1907 | /* |
| 1908 | * vm_map_inherit: |
| 1909 | * |
| 1910 | * Sets the inheritance of the specified address |
| 1911 | * range in the target map. Inheritance |
| 1912 | * affects how the map will be shared with |
| 1913 | * child maps at the time of vm_map_fork. |
| 1914 | */ |
| 1915 | int |
| 1916 | vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 1917 | vm_inherit_t new_inheritance) |
| 1918 | { |
| 1919 | vm_map_entry_t entry; |
| 1920 | vm_map_entry_t temp_entry; |
| 1921 | int count; |
| 1922 | |
| 1923 | switch (new_inheritance) { |
| 1924 | case VM_INHERIT_NONE: |
| 1925 | case VM_INHERIT_COPY: |
| 1926 | case VM_INHERIT_SHARE: |
| 1927 | break; |
| 1928 | default: |
| 1929 | return (KERN_INVALID_ARGUMENT); |
| 1930 | } |
| 1931 | |
| 1932 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1933 | vm_map_lock(map); |
| 1934 | |
| 1935 | VM_MAP_RANGE_CHECK(map, start, end); |
| 1936 | |
| 1937 | if (vm_map_lookup_entry(map, start, &temp_entry)) { |
| 1938 | entry = temp_entry; |
| 1939 | vm_map_clip_start(map, entry, start, &count); |
| 1940 | } else |
| 1941 | entry = temp_entry->next; |
| 1942 | |
| 1943 | while ((entry != &map->header) && (entry->start < end)) { |
| 1944 | vm_map_clip_end(map, entry, end, &count); |
| 1945 | |
| 1946 | entry->inheritance = new_inheritance; |
| 1947 | |
| 1948 | vm_map_simplify_entry(map, entry, &count); |
| 1949 | |
| 1950 | entry = entry->next; |
| 1951 | } |
| 1952 | vm_map_unlock(map); |
| 1953 | vm_map_entry_release(count); |
| 1954 | return (KERN_SUCCESS); |
| 1955 | } |
| 1956 | |
| 1957 | /* |
| 1958 | * Implement the semantics of mlock |
| 1959 | */ |
| 1960 | int |
| 1961 | vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, |
| 1962 | boolean_t new_pageable) |
| 1963 | { |
| 1964 | vm_map_entry_t entry; |
| 1965 | vm_map_entry_t start_entry; |
| 1966 | vm_offset_t end; |
| 1967 | int rv = KERN_SUCCESS; |
| 1968 | int count; |
| 1969 | |
| 1970 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 1971 | vm_map_lock(map); |
| 1972 | VM_MAP_RANGE_CHECK(map, start, real_end); |
| 1973 | end = real_end; |
| 1974 | |
| 1975 | start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES); |
| 1976 | if (start_entry == NULL) { |
| 1977 | vm_map_unlock(map); |
| 1978 | vm_map_entry_release(count); |
| 1979 | return (KERN_INVALID_ADDRESS); |
| 1980 | } |
| 1981 | |
| 1982 | if (new_pageable == 0) { |
| 1983 | entry = start_entry; |
| 1984 | while ((entry != &map->header) && (entry->start < end)) { |
| 1985 | vm_offset_t save_start; |
| 1986 | vm_offset_t save_end; |
| 1987 | |
| 1988 | /* |
| 1989 | * Already user wired or hard wired (trivial cases) |
| 1990 | */ |
| 1991 | if (entry->eflags & MAP_ENTRY_USER_WIRED) { |
| 1992 | entry = entry->next; |
| 1993 | continue; |
| 1994 | } |
| 1995 | if (entry->wired_count != 0) { |
| 1996 | entry->wired_count++; |
| 1997 | entry->eflags |= MAP_ENTRY_USER_WIRED; |
| 1998 | entry = entry->next; |
| 1999 | continue; |
| 2000 | } |
| 2001 | |
| 2002 | /* |
| 2003 | * A new wiring requires instantiation of appropriate |
| 2004 | * management structures and the faulting in of the |
| 2005 | * page. |
| 2006 | */ |
| 2007 | if (entry->maptype != VM_MAPTYPE_SUBMAP) { |
| 2008 | int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY; |
| 2009 | if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) { |
| 2010 | vm_map_entry_shadow(entry); |
| 2011 | } else if (entry->object.vm_object == NULL && |
| 2012 | !map->system_map) { |
| 2013 | vm_map_entry_allocate_object(entry); |
| 2014 | } |
| 2015 | } |
| 2016 | entry->wired_count++; |
| 2017 | entry->eflags |= MAP_ENTRY_USER_WIRED; |
| 2018 | |
| 2019 | /* |
| 2020 | * Now fault in the area. Note that vm_fault_wire() |
| 2021 | * may release the map lock temporarily, it will be |
| 2022 | * relocked on return. The in-transition |
| 2023 | * flag protects the entries. |
| 2024 | */ |
| 2025 | save_start = entry->start; |
| 2026 | save_end = entry->end; |
| 2027 | rv = vm_fault_wire(map, entry, TRUE); |
| 2028 | if (rv) { |
| 2029 | CLIP_CHECK_BACK(entry, save_start); |
| 2030 | for (;;) { |
| 2031 | KASSERT(entry->wired_count == 1, ("bad wired_count on entry")); |
| 2032 | entry->eflags &= ~MAP_ENTRY_USER_WIRED; |
| 2033 | entry->wired_count = 0; |
| 2034 | if (entry->end == save_end) |
| 2035 | break; |
| 2036 | entry = entry->next; |
| 2037 | KASSERT(entry != &map->header, ("bad entry clip during backout")); |
| 2038 | } |
| 2039 | end = save_start; /* unwire the rest */ |
| 2040 | break; |
| 2041 | } |
| 2042 | /* |
| 2043 | * note that even though the entry might have been |
| 2044 | * clipped, the USER_WIRED flag we set prevents |
| 2045 | * duplication so we do not have to do a |
| 2046 | * clip check. |
| 2047 | */ |
| 2048 | entry = entry->next; |
| 2049 | } |
| 2050 | |
| 2051 | /* |
| 2052 | * If we failed fall through to the unwiring section to |
| 2053 | * unwire what we had wired so far. 'end' has already |
| 2054 | * been adjusted. |
| 2055 | */ |
| 2056 | if (rv) |
| 2057 | new_pageable = 1; |
| 2058 | |
| 2059 | /* |
| 2060 | * start_entry might have been clipped if we unlocked the |
| 2061 | * map and blocked. No matter how clipped it has gotten |
| 2062 | * there should be a fragment that is on our start boundary. |
| 2063 | */ |
| 2064 | CLIP_CHECK_BACK(start_entry, start); |
| 2065 | } |
| 2066 | |
| 2067 | /* |
| 2068 | * Deal with the unwiring case. |
| 2069 | */ |
| 2070 | if (new_pageable) { |
| 2071 | /* |
| 2072 | * This is the unwiring case. We must first ensure that the |
| 2073 | * range to be unwired is really wired down. We know there |
| 2074 | * are no holes. |
| 2075 | */ |
| 2076 | entry = start_entry; |
| 2077 | while ((entry != &map->header) && (entry->start < end)) { |
| 2078 | if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { |
| 2079 | rv = KERN_INVALID_ARGUMENT; |
| 2080 | goto done; |
| 2081 | } |
| 2082 | KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry)); |
| 2083 | entry = entry->next; |
| 2084 | } |
| 2085 | |
| 2086 | /* |
| 2087 | * Now decrement the wiring count for each region. If a region |
| 2088 | * becomes completely unwired, unwire its physical pages and |
| 2089 | * mappings. |
| 2090 | */ |
| 2091 | /* |
| 2092 | * The map entries are processed in a loop, checking to |
| 2093 | * make sure the entry is wired and asserting it has a wired |
| 2094 | * count. However, another loop was inserted more-or-less in |
| 2095 | * the middle of the unwiring path. This loop picks up the |
| 2096 | * "entry" loop variable from the first loop without first |
| 2097 | * setting it to start_entry. Naturally, the secound loop |
| 2098 | * is never entered and the pages backing the entries are |
| 2099 | * never unwired. This can lead to a leak of wired pages. |
| 2100 | */ |
| 2101 | entry = start_entry; |
| 2102 | while ((entry != &map->header) && (entry->start < end)) { |
| 2103 | KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED, |
| 2104 | ("expected USER_WIRED on entry %p", entry)); |
| 2105 | entry->eflags &= ~MAP_ENTRY_USER_WIRED; |
| 2106 | entry->wired_count--; |
| 2107 | if (entry->wired_count == 0) |
| 2108 | vm_fault_unwire(map, entry); |
| 2109 | entry = entry->next; |
| 2110 | } |
| 2111 | } |
| 2112 | done: |
| 2113 | vm_map_unclip_range(map, start_entry, start, real_end, &count, |
| 2114 | MAP_CLIP_NO_HOLES); |
| 2115 | map->timestamp++; |
| 2116 | vm_map_unlock(map); |
| 2117 | vm_map_entry_release(count); |
| 2118 | return (rv); |
| 2119 | } |
| 2120 | |
| 2121 | /* |
| 2122 | * vm_map_wire: |
| 2123 | * |
| 2124 | * Sets the pageability of the specified address |
| 2125 | * range in the target map. Regions specified |
| 2126 | * as not pageable require locked-down physical |
| 2127 | * memory and physical page maps. |
| 2128 | * |
| 2129 | * The map must not be locked, but a reference |
| 2130 | * must remain to the map throughout the call. |
| 2131 | * |
| 2132 | * This function may be called via the zalloc path and must properly |
| 2133 | * reserve map entries for kernel_map. |
| 2134 | */ |
| 2135 | int |
| 2136 | vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags) |
| 2137 | { |
| 2138 | vm_map_entry_t entry; |
| 2139 | vm_map_entry_t start_entry; |
| 2140 | vm_offset_t end; |
| 2141 | int rv = KERN_SUCCESS; |
| 2142 | int count; |
| 2143 | |
| 2144 | if (kmflags & KM_KRESERVE) |
| 2145 | count = vm_map_entry_kreserve(MAP_RESERVE_COUNT); |
| 2146 | else |
| 2147 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 2148 | vm_map_lock(map); |
| 2149 | VM_MAP_RANGE_CHECK(map, start, real_end); |
| 2150 | end = real_end; |
| 2151 | |
| 2152 | start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES); |
| 2153 | if (start_entry == NULL) { |
| 2154 | vm_map_unlock(map); |
| 2155 | rv = KERN_INVALID_ADDRESS; |
| 2156 | goto failure; |
| 2157 | } |
| 2158 | if ((kmflags & KM_PAGEABLE) == 0) { |
| 2159 | /* |
| 2160 | * Wiring. |
| 2161 | * |
| 2162 | * 1. Holding the write lock, we create any shadow or zero-fill |
| 2163 | * objects that need to be created. Then we clip each map |
| 2164 | * entry to the region to be wired and increment its wiring |
| 2165 | * count. We create objects before clipping the map entries |
| 2166 | * to avoid object proliferation. |
| 2167 | * |
| 2168 | * 2. We downgrade to a read lock, and call vm_fault_wire to |
| 2169 | * fault in the pages for any newly wired area (wired_count is |
| 2170 | * 1). |
| 2171 | * |
| 2172 | * Downgrading to a read lock for vm_fault_wire avoids a |
| 2173 | * possible deadlock with another process that may have faulted |
| 2174 | * on one of the pages to be wired (it would mark the page busy, |
| 2175 | * blocking us, then in turn block on the map lock that we |
| 2176 | * hold). Because of problems in the recursive lock package, |
| 2177 | * we cannot upgrade to a write lock in vm_map_lookup. Thus, |
| 2178 | * any actions that require the write lock must be done |
| 2179 | * beforehand. Because we keep the read lock on the map, the |
| 2180 | * copy-on-write status of the entries we modify here cannot |
| 2181 | * change. |
| 2182 | */ |
| 2183 | |
| 2184 | entry = start_entry; |
| 2185 | while ((entry != &map->header) && (entry->start < end)) { |
| 2186 | /* |
| 2187 | * Trivial case if the entry is already wired |
| 2188 | */ |
| 2189 | if (entry->wired_count) { |
| 2190 | entry->wired_count++; |
| 2191 | entry = entry->next; |
| 2192 | continue; |
| 2193 | } |
| 2194 | |
| 2195 | /* |
| 2196 | * The entry is being newly wired, we have to setup |
| 2197 | * appropriate management structures. A shadow |
| 2198 | * object is required for a copy-on-write region, |
| 2199 | * or a normal object for a zero-fill region. We |
| 2200 | * do not have to do this for entries that point to sub |
| 2201 | * maps because we won't hold the lock on the sub map. |
| 2202 | */ |
| 2203 | if (entry->maptype != VM_MAPTYPE_SUBMAP) { |
| 2204 | int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY; |
| 2205 | if (copyflag && |
| 2206 | ((entry->protection & VM_PROT_WRITE) != 0)) { |
| 2207 | vm_map_entry_shadow(entry); |
| 2208 | } else if (entry->object.vm_object == NULL && |
| 2209 | !map->system_map) { |
| 2210 | vm_map_entry_allocate_object(entry); |
| 2211 | } |
| 2212 | } |
| 2213 | |
| 2214 | entry->wired_count++; |
| 2215 | entry = entry->next; |
| 2216 | } |
| 2217 | |
| 2218 | /* |
| 2219 | * Pass 2. |
| 2220 | */ |
| 2221 | |
| 2222 | /* |
| 2223 | * HACK HACK HACK HACK |
| 2224 | * |
| 2225 | * Unlock the map to avoid deadlocks. The in-transit flag |
| 2226 | * protects us from most changes but note that |
| 2227 | * clipping may still occur. To prevent clipping from |
| 2228 | * occuring after the unlock, except for when we are |
| 2229 | * blocking in vm_fault_wire, we must run in a critical |
| 2230 | * section, otherwise our accesses to entry->start and |
| 2231 | * entry->end could be corrupted. We have to enter the |
| 2232 | * critical section prior to unlocking so start_entry does |
| 2233 | * not change out from under us at the very beginning of the |
| 2234 | * loop. |
| 2235 | * |
| 2236 | * HACK HACK HACK HACK |
| 2237 | */ |
| 2238 | |
| 2239 | crit_enter(); |
| 2240 | |
| 2241 | entry = start_entry; |
| 2242 | while (entry != &map->header && entry->start < end) { |
| 2243 | /* |
| 2244 | * If vm_fault_wire fails for any page we need to undo |
| 2245 | * what has been done. We decrement the wiring count |
| 2246 | * for those pages which have not yet been wired (now) |
| 2247 | * and unwire those that have (later). |
| 2248 | */ |
| 2249 | vm_offset_t save_start = entry->start; |
| 2250 | vm_offset_t save_end = entry->end; |
| 2251 | |
| 2252 | if (entry->wired_count == 1) |
| 2253 | rv = vm_fault_wire(map, entry, FALSE); |
| 2254 | if (rv) { |
| 2255 | CLIP_CHECK_BACK(entry, save_start); |
| 2256 | for (;;) { |
| 2257 | KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly")); |
| 2258 | entry->wired_count = 0; |
| 2259 | if (entry->end == save_end) |
| 2260 | break; |
| 2261 | entry = entry->next; |
| 2262 | KASSERT(entry != &map->header, ("bad entry clip during backout")); |
| 2263 | } |
| 2264 | end = save_start; |
| 2265 | break; |
| 2266 | } |
| 2267 | CLIP_CHECK_FWD(entry, save_end); |
| 2268 | entry = entry->next; |
| 2269 | } |
| 2270 | crit_exit(); |
| 2271 | |
| 2272 | /* |
| 2273 | * If a failure occured undo everything by falling through |
| 2274 | * to the unwiring code. 'end' has already been adjusted |
| 2275 | * appropriately. |
| 2276 | */ |
| 2277 | if (rv) |
| 2278 | kmflags |= KM_PAGEABLE; |
| 2279 | |
| 2280 | /* |
| 2281 | * start_entry is still IN_TRANSITION but may have been |
| 2282 | * clipped since vm_fault_wire() unlocks and relocks the |
| 2283 | * map. No matter how clipped it has gotten there should |
| 2284 | * be a fragment that is on our start boundary. |
| 2285 | */ |
| 2286 | CLIP_CHECK_BACK(start_entry, start); |
| 2287 | } |
| 2288 | |
| 2289 | if (kmflags & KM_PAGEABLE) { |
| 2290 | /* |
| 2291 | * This is the unwiring case. We must first ensure that the |
| 2292 | * range to be unwired is really wired down. We know there |
| 2293 | * are no holes. |
| 2294 | */ |
| 2295 | entry = start_entry; |
| 2296 | while ((entry != &map->header) && (entry->start < end)) { |
| 2297 | if (entry->wired_count == 0) { |
| 2298 | rv = KERN_INVALID_ARGUMENT; |
| 2299 | goto done; |
| 2300 | } |
| 2301 | entry = entry->next; |
| 2302 | } |
| 2303 | |
| 2304 | /* |
| 2305 | * Now decrement the wiring count for each region. If a region |
| 2306 | * becomes completely unwired, unwire its physical pages and |
| 2307 | * mappings. |
| 2308 | */ |
| 2309 | entry = start_entry; |
| 2310 | while ((entry != &map->header) && (entry->start < end)) { |
| 2311 | entry->wired_count--; |
| 2312 | if (entry->wired_count == 0) |
| 2313 | vm_fault_unwire(map, entry); |
| 2314 | entry = entry->next; |
| 2315 | } |
| 2316 | } |
| 2317 | done: |
| 2318 | vm_map_unclip_range(map, start_entry, start, real_end, &count, |
| 2319 | MAP_CLIP_NO_HOLES); |
| 2320 | map->timestamp++; |
| 2321 | vm_map_unlock(map); |
| 2322 | failure: |
| 2323 | if (kmflags & KM_KRESERVE) |
| 2324 | vm_map_entry_krelease(count); |
| 2325 | else |
| 2326 | vm_map_entry_release(count); |
| 2327 | return (rv); |
| 2328 | } |
| 2329 | |
| 2330 | /* |
| 2331 | * vm_map_set_wired_quick() |
| 2332 | * |
| 2333 | * Mark a newly allocated address range as wired but do not fault in |
| 2334 | * the pages. The caller is expected to load the pages into the object. |
| 2335 | * |
| 2336 | * The map must be locked on entry and will remain locked on return. |
| 2337 | */ |
| 2338 | void |
| 2339 | vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size, int *countp) |
| 2340 | { |
| 2341 | vm_map_entry_t scan; |
| 2342 | vm_map_entry_t entry; |
| 2343 | |
| 2344 | entry = vm_map_clip_range(map, addr, addr + size, countp, MAP_CLIP_NO_HOLES); |
| 2345 | for (scan = entry; scan != &map->header && scan->start < addr + size; scan = scan->next) { |
| 2346 | KKASSERT(entry->wired_count == 0); |
| 2347 | entry->wired_count = 1; |
| 2348 | } |
| 2349 | vm_map_unclip_range(map, entry, addr, addr + size, countp, MAP_CLIP_NO_HOLES); |
| 2350 | } |
| 2351 | |
| 2352 | /* |
| 2353 | * vm_map_clean |
| 2354 | * |
| 2355 | * Push any dirty cached pages in the address range to their pager. |
| 2356 | * If syncio is TRUE, dirty pages are written synchronously. |
| 2357 | * If invalidate is TRUE, any cached pages are freed as well. |
| 2358 | * |
| 2359 | * This routine is called by sys_msync() |
| 2360 | * |
| 2361 | * Returns an error if any part of the specified range is not mapped. |
| 2362 | */ |
| 2363 | int |
| 2364 | vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 2365 | boolean_t syncio, boolean_t invalidate) |
| 2366 | { |
| 2367 | vm_map_entry_t current; |
| 2368 | vm_map_entry_t entry; |
| 2369 | vm_size_t size; |
| 2370 | vm_object_t object; |
| 2371 | vm_ooffset_t offset; |
| 2372 | |
| 2373 | vm_map_lock_read(map); |
| 2374 | VM_MAP_RANGE_CHECK(map, start, end); |
| 2375 | if (!vm_map_lookup_entry(map, start, &entry)) { |
| 2376 | vm_map_unlock_read(map); |
| 2377 | return (KERN_INVALID_ADDRESS); |
| 2378 | } |
| 2379 | /* |
| 2380 | * Make a first pass to check for holes. |
| 2381 | */ |
| 2382 | for (current = entry; current->start < end; current = current->next) { |
| 2383 | if (current->maptype == VM_MAPTYPE_SUBMAP) { |
| 2384 | vm_map_unlock_read(map); |
| 2385 | return (KERN_INVALID_ARGUMENT); |
| 2386 | } |
| 2387 | if (end > current->end && |
| 2388 | (current->next == &map->header || |
| 2389 | current->end != current->next->start)) { |
| 2390 | vm_map_unlock_read(map); |
| 2391 | return (KERN_INVALID_ADDRESS); |
| 2392 | } |
| 2393 | } |
| 2394 | |
| 2395 | if (invalidate) |
| 2396 | pmap_remove(vm_map_pmap(map), start, end); |
| 2397 | /* |
| 2398 | * Make a second pass, cleaning/uncaching pages from the indicated |
| 2399 | * objects as we go. |
| 2400 | */ |
| 2401 | for (current = entry; current->start < end; current = current->next) { |
| 2402 | offset = current->offset + (start - current->start); |
| 2403 | size = (end <= current->end ? end : current->end) - start; |
| 2404 | if (current->maptype == VM_MAPTYPE_SUBMAP) { |
| 2405 | vm_map_t smap; |
| 2406 | vm_map_entry_t tentry; |
| 2407 | vm_size_t tsize; |
| 2408 | |
| 2409 | smap = current->object.sub_map; |
| 2410 | vm_map_lock_read(smap); |
| 2411 | vm_map_lookup_entry(smap, offset, &tentry); |
| 2412 | tsize = tentry->end - offset; |
| 2413 | if (tsize < size) |
| 2414 | size = tsize; |
| 2415 | object = tentry->object.vm_object; |
| 2416 | offset = tentry->offset + (offset - tentry->start); |
| 2417 | vm_map_unlock_read(smap); |
| 2418 | } else { |
| 2419 | object = current->object.vm_object; |
| 2420 | } |
| 2421 | /* |
| 2422 | * Note that there is absolutely no sense in writing out |
| 2423 | * anonymous objects, so we track down the vnode object |
| 2424 | * to write out. |
| 2425 | * We invalidate (remove) all pages from the address space |
| 2426 | * anyway, for semantic correctness. |
| 2427 | * |
| 2428 | * note: certain anonymous maps, such as MAP_NOSYNC maps, |
| 2429 | * may start out with a NULL object. |
| 2430 | */ |
| 2431 | while (object && object->backing_object) { |
| 2432 | offset += object->backing_object_offset; |
| 2433 | object = object->backing_object; |
| 2434 | if (object->size < OFF_TO_IDX( offset + size)) |
| 2435 | size = IDX_TO_OFF(object->size) - offset; |
| 2436 | } |
| 2437 | if (object && (object->type == OBJT_VNODE) && |
| 2438 | (current->protection & VM_PROT_WRITE) && |
| 2439 | (object->flags & OBJ_NOMSYNC) == 0) { |
| 2440 | /* |
| 2441 | * Flush pages if writing is allowed, invalidate them |
| 2442 | * if invalidation requested. Pages undergoing I/O |
| 2443 | * will be ignored by vm_object_page_remove(). |
| 2444 | * |
| 2445 | * We cannot lock the vnode and then wait for paging |
| 2446 | * to complete without deadlocking against vm_fault. |
| 2447 | * Instead we simply call vm_object_page_remove() and |
| 2448 | * allow it to block internally on a page-by-page |
| 2449 | * basis when it encounters pages undergoing async |
| 2450 | * I/O. |
| 2451 | */ |
| 2452 | int flags; |
| 2453 | |
| 2454 | vm_object_reference(object); |
| 2455 | vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY); |
| 2456 | flags = (syncio || invalidate) ? OBJPC_SYNC : 0; |
| 2457 | flags |= invalidate ? OBJPC_INVAL : 0; |
| 2458 | |
| 2459 | /* |
| 2460 | * When operating on a virtual page table just |
| 2461 | * flush the whole object. XXX we probably ought |
| 2462 | * to |
| 2463 | */ |
| 2464 | switch(current->maptype) { |
| 2465 | case VM_MAPTYPE_NORMAL: |
| 2466 | vm_object_page_clean(object, |
| 2467 | OFF_TO_IDX(offset), |
| 2468 | OFF_TO_IDX(offset + size + PAGE_MASK), |
| 2469 | flags); |
| 2470 | break; |
| 2471 | case VM_MAPTYPE_VPAGETABLE: |
| 2472 | vm_object_page_clean(object, 0, 0, flags); |
| 2473 | break; |
| 2474 | } |
| 2475 | vn_unlock(((struct vnode *)object->handle)); |
| 2476 | vm_object_deallocate(object); |
| 2477 | } |
| 2478 | if (object && invalidate && |
| 2479 | ((object->type == OBJT_VNODE) || |
| 2480 | (object->type == OBJT_DEVICE))) { |
| 2481 | int clean_only = |
| 2482 | (object->type == OBJT_DEVICE) ? FALSE : TRUE; |
| 2483 | vm_object_reference(object); |
| 2484 | switch(current->maptype) { |
| 2485 | case VM_MAPTYPE_NORMAL: |
| 2486 | vm_object_page_remove(object, |
| 2487 | OFF_TO_IDX(offset), |
| 2488 | OFF_TO_IDX(offset + size + PAGE_MASK), |
| 2489 | clean_only); |
| 2490 | break; |
| 2491 | case VM_MAPTYPE_VPAGETABLE: |
| 2492 | vm_object_page_remove(object, 0, 0, clean_only); |
| 2493 | break; |
| 2494 | } |
| 2495 | vm_object_deallocate(object); |
| 2496 | } |
| 2497 | start += size; |
| 2498 | } |
| 2499 | |
| 2500 | vm_map_unlock_read(map); |
| 2501 | return (KERN_SUCCESS); |
| 2502 | } |
| 2503 | |
| 2504 | /* |
| 2505 | * vm_map_entry_unwire: [ internal use only ] |
| 2506 | * |
| 2507 | * Make the region specified by this entry pageable. |
| 2508 | * |
| 2509 | * The map in question should be locked. |
| 2510 | * [This is the reason for this routine's existence.] |
| 2511 | */ |
| 2512 | static void |
| 2513 | vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) |
| 2514 | { |
| 2515 | entry->eflags &= ~MAP_ENTRY_USER_WIRED; |
| 2516 | entry->wired_count = 0; |
| 2517 | vm_fault_unwire(map, entry); |
| 2518 | } |
| 2519 | |
| 2520 | /* |
| 2521 | * vm_map_entry_delete: [ internal use only ] |
| 2522 | * |
| 2523 | * Deallocate the given entry from the target map. |
| 2524 | */ |
| 2525 | static void |
| 2526 | vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp) |
| 2527 | { |
| 2528 | vm_map_entry_unlink(map, entry); |
| 2529 | map->size -= entry->end - entry->start; |
| 2530 | |
| 2531 | switch(entry->maptype) { |
| 2532 | case VM_MAPTYPE_NORMAL: |
| 2533 | case VM_MAPTYPE_VPAGETABLE: |
| 2534 | vm_object_deallocate(entry->object.vm_object); |
| 2535 | break; |
| 2536 | default: |
| 2537 | break; |
| 2538 | } |
| 2539 | |
| 2540 | vm_map_entry_dispose(map, entry, countp); |
| 2541 | } |
| 2542 | |
| 2543 | /* |
| 2544 | * vm_map_delete: [ internal use only ] |
| 2545 | * |
| 2546 | * Deallocates the given address range from the target |
| 2547 | * map. |
| 2548 | */ |
| 2549 | int |
| 2550 | vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp) |
| 2551 | { |
| 2552 | vm_object_t object; |
| 2553 | vm_map_entry_t entry; |
| 2554 | vm_map_entry_t first_entry; |
| 2555 | |
| 2556 | again: |
| 2557 | /* |
| 2558 | * Find the start of the region, and clip it. Set entry to point |
| 2559 | * at the first record containing the requested address or, if no |
| 2560 | * such record exists, the next record with a greater address. The |
| 2561 | * loop will run from this point until a record beyond the termination |
| 2562 | * address is encountered. |
| 2563 | * |
| 2564 | * map->hint must be adjusted to not point to anything we delete, |
| 2565 | * so set it to the entry prior to the one being deleted. |
| 2566 | * |
| 2567 | * GGG see other GGG comment. |
| 2568 | */ |
| 2569 | if (vm_map_lookup_entry(map, start, &first_entry)) { |
| 2570 | entry = first_entry; |
| 2571 | vm_map_clip_start(map, entry, start, countp); |
| 2572 | map->hint = entry->prev; /* possible problem XXX */ |
| 2573 | } else { |
| 2574 | map->hint = first_entry; /* possible problem XXX */ |
| 2575 | entry = first_entry->next; |
| 2576 | } |
| 2577 | |
| 2578 | /* |
| 2579 | * If a hole opens up prior to the current first_free then |
| 2580 | * adjust first_free. As with map->hint, map->first_free |
| 2581 | * cannot be left set to anything we might delete. |
| 2582 | */ |
| 2583 | if (entry == &map->header) { |
| 2584 | map->first_free = &map->header; |
| 2585 | } else if (map->first_free->start >= start) { |
| 2586 | map->first_free = entry->prev; |
| 2587 | } |
| 2588 | |
| 2589 | /* |
| 2590 | * Step through all entries in this region |
| 2591 | */ |
| 2592 | |
| 2593 | while ((entry != &map->header) && (entry->start < end)) { |
| 2594 | vm_map_entry_t next; |
| 2595 | vm_offset_t s, e; |
| 2596 | vm_pindex_t offidxstart, offidxend, count; |
| 2597 | |
| 2598 | /* |
| 2599 | * If we hit an in-transition entry we have to sleep and |
| 2600 | * retry. It's easier (and not really slower) to just retry |
| 2601 | * since this case occurs so rarely and the hint is already |
| 2602 | * pointing at the right place. We have to reset the |
| 2603 | * start offset so as not to accidently delete an entry |
| 2604 | * another process just created in vacated space. |
| 2605 | */ |
| 2606 | if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { |
| 2607 | entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; |
| 2608 | start = entry->start; |
| 2609 | ++mycpu->gd_cnt.v_intrans_coll; |
| 2610 | ++mycpu->gd_cnt.v_intrans_wait; |
| 2611 | vm_map_transition_wait(map); |
| 2612 | goto again; |
| 2613 | } |
| 2614 | vm_map_clip_end(map, entry, end, countp); |
| 2615 | |
| 2616 | s = entry->start; |
| 2617 | e = entry->end; |
| 2618 | next = entry->next; |
| 2619 | |
| 2620 | offidxstart = OFF_TO_IDX(entry->offset); |
| 2621 | count = OFF_TO_IDX(e - s); |
| 2622 | object = entry->object.vm_object; |
| 2623 | |
| 2624 | /* |
| 2625 | * Unwire before removing addresses from the pmap; otherwise, |
| 2626 | * unwiring will put the entries back in the pmap. |
| 2627 | */ |
| 2628 | if (entry->wired_count != 0) |
| 2629 | vm_map_entry_unwire(map, entry); |
| 2630 | |
| 2631 | offidxend = offidxstart + count; |
| 2632 | |
| 2633 | if (object == &kernel_object) { |
| 2634 | vm_object_page_remove(object, offidxstart, offidxend, FALSE); |
| 2635 | } else { |
| 2636 | pmap_remove(map->pmap, s, e); |
| 2637 | if (object != NULL && |
| 2638 | object->ref_count != 1 && |
| 2639 | (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING && |
| 2640 | (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { |
| 2641 | vm_object_collapse(object); |
| 2642 | vm_object_page_remove(object, offidxstart, offidxend, FALSE); |
| 2643 | if (object->type == OBJT_SWAP) { |
| 2644 | swap_pager_freespace(object, offidxstart, count); |
| 2645 | } |
| 2646 | if (offidxend >= object->size && |
| 2647 | offidxstart < object->size) { |
| 2648 | object->size = offidxstart; |
| 2649 | } |
| 2650 | } |
| 2651 | } |
| 2652 | |
| 2653 | /* |
| 2654 | * Delete the entry (which may delete the object) only after |
| 2655 | * removing all pmap entries pointing to its pages. |
| 2656 | * (Otherwise, its page frames may be reallocated, and any |
| 2657 | * modify bits will be set in the wrong object!) |
| 2658 | */ |
| 2659 | vm_map_entry_delete(map, entry, countp); |
| 2660 | entry = next; |
| 2661 | } |
| 2662 | return (KERN_SUCCESS); |
| 2663 | } |
| 2664 | |
| 2665 | /* |
| 2666 | * vm_map_remove: |
| 2667 | * |
| 2668 | * Remove the given address range from the target map. |
| 2669 | * This is the exported form of vm_map_delete. |
| 2670 | */ |
| 2671 | int |
| 2672 | vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) |
| 2673 | { |
| 2674 | int result; |
| 2675 | int count; |
| 2676 | |
| 2677 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 2678 | vm_map_lock(map); |
| 2679 | VM_MAP_RANGE_CHECK(map, start, end); |
| 2680 | result = vm_map_delete(map, start, end, &count); |
| 2681 | vm_map_unlock(map); |
| 2682 | vm_map_entry_release(count); |
| 2683 | |
| 2684 | return (result); |
| 2685 | } |
| 2686 | |
| 2687 | /* |
| 2688 | * vm_map_check_protection: |
| 2689 | * |
| 2690 | * Assert that the target map allows the specified |
| 2691 | * privilege on the entire address region given. |
| 2692 | * The entire region must be allocated. |
| 2693 | */ |
| 2694 | boolean_t |
| 2695 | vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, |
| 2696 | vm_prot_t protection) |
| 2697 | { |
| 2698 | vm_map_entry_t entry; |
| 2699 | vm_map_entry_t tmp_entry; |
| 2700 | |
| 2701 | if (!vm_map_lookup_entry(map, start, &tmp_entry)) { |
| 2702 | return (FALSE); |
| 2703 | } |
| 2704 | entry = tmp_entry; |
| 2705 | |
| 2706 | while (start < end) { |
| 2707 | if (entry == &map->header) { |
| 2708 | return (FALSE); |
| 2709 | } |
| 2710 | /* |
| 2711 | * No holes allowed! |
| 2712 | */ |
| 2713 | |
| 2714 | if (start < entry->start) { |
| 2715 | return (FALSE); |
| 2716 | } |
| 2717 | /* |
| 2718 | * Check protection associated with entry. |
| 2719 | */ |
| 2720 | |
| 2721 | if ((entry->protection & protection) != protection) { |
| 2722 | return (FALSE); |
| 2723 | } |
| 2724 | /* go to next entry */ |
| 2725 | |
| 2726 | start = entry->end; |
| 2727 | entry = entry->next; |
| 2728 | } |
| 2729 | return (TRUE); |
| 2730 | } |
| 2731 | |
| 2732 | /* |
| 2733 | * Split the pages in a map entry into a new object. This affords |
| 2734 | * easier removal of unused pages, and keeps object inheritance from |
| 2735 | * being a negative impact on memory usage. |
| 2736 | */ |
| 2737 | static void |
| 2738 | vm_map_split(vm_map_entry_t entry) |
| 2739 | { |
| 2740 | vm_page_t m; |
| 2741 | vm_object_t orig_object, new_object, source; |
| 2742 | vm_offset_t s, e; |
| 2743 | vm_pindex_t offidxstart, offidxend, idx; |
| 2744 | vm_size_t size; |
| 2745 | vm_ooffset_t offset; |
| 2746 | |
| 2747 | orig_object = entry->object.vm_object; |
| 2748 | if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP) |
| 2749 | return; |
| 2750 | if (orig_object->ref_count <= 1) |
| 2751 | return; |
| 2752 | |
| 2753 | offset = entry->offset; |
| 2754 | s = entry->start; |
| 2755 | e = entry->end; |
| 2756 | |
| 2757 | offidxstart = OFF_TO_IDX(offset); |
| 2758 | offidxend = offidxstart + OFF_TO_IDX(e - s); |
| 2759 | size = offidxend - offidxstart; |
| 2760 | |
| 2761 | switch(orig_object->type) { |
| 2762 | case OBJT_DEFAULT: |
| 2763 | new_object = default_pager_alloc(NULL, IDX_TO_OFF(size), |
| 2764 | VM_PROT_ALL, 0); |
| 2765 | break; |
| 2766 | case OBJT_SWAP: |
| 2767 | new_object = swap_pager_alloc(NULL, IDX_TO_OFF(size), |
| 2768 | VM_PROT_ALL, 0); |
| 2769 | break; |
| 2770 | default: |
| 2771 | /* not reached */ |
| 2772 | new_object = NULL; |
| 2773 | KKASSERT(0); |
| 2774 | } |
| 2775 | if (new_object == NULL) |
| 2776 | return; |
| 2777 | |
| 2778 | source = orig_object->backing_object; |
| 2779 | if (source != NULL) { |
| 2780 | vm_object_reference(source); /* Referenced by new_object */ |
| 2781 | LIST_INSERT_HEAD(&source->shadow_head, |
| 2782 | new_object, shadow_list); |
| 2783 | vm_object_clear_flag(source, OBJ_ONEMAPPING); |
| 2784 | new_object->backing_object_offset = |
| 2785 | orig_object->backing_object_offset + IDX_TO_OFF(offidxstart); |
| 2786 | new_object->backing_object = source; |
| 2787 | source->shadow_count++; |
| 2788 | source->generation++; |
| 2789 | } |
| 2790 | |
| 2791 | for (idx = 0; idx < size; idx++) { |
| 2792 | vm_page_t m; |
| 2793 | |
| 2794 | /* |
| 2795 | * A critical section is required to avoid a race between |
| 2796 | * the lookup and an interrupt/unbusy/free and our busy |
| 2797 | * check. |
| 2798 | */ |
| 2799 | crit_enter(); |
| 2800 | retry: |
| 2801 | m = vm_page_lookup(orig_object, offidxstart + idx); |
| 2802 | if (m == NULL) { |
| 2803 | crit_exit(); |
| 2804 | continue; |
| 2805 | } |
| 2806 | |
| 2807 | /* |
| 2808 | * We must wait for pending I/O to complete before we can |
| 2809 | * rename the page. |
| 2810 | * |
| 2811 | * We do not have to VM_PROT_NONE the page as mappings should |
| 2812 | * not be changed by this operation. |
| 2813 | */ |
| 2814 | if (vm_page_sleep_busy(m, TRUE, "spltwt")) |
| 2815 | goto retry; |
| 2816 | vm_page_busy(m); |
| 2817 | vm_page_rename(m, new_object, idx); |
| 2818 | /* page automatically made dirty by rename and cache handled */ |
| 2819 | vm_page_busy(m); |
| 2820 | crit_exit(); |
| 2821 | } |
| 2822 | |
| 2823 | if (orig_object->type == OBJT_SWAP) { |
| 2824 | vm_object_pip_add(orig_object, 1); |
| 2825 | /* |
| 2826 | * copy orig_object pages into new_object |
| 2827 | * and destroy unneeded pages in |
| 2828 | * shadow object. |
| 2829 | */ |
| 2830 | swap_pager_copy(orig_object, new_object, offidxstart, 0); |
| 2831 | vm_object_pip_wakeup(orig_object); |
| 2832 | } |
| 2833 | |
| 2834 | /* |
| 2835 | * Wakeup the pages we played with. No spl protection is needed |
| 2836 | * for a simple wakeup. |
| 2837 | */ |
| 2838 | for (idx = 0; idx < size; idx++) { |
| 2839 | m = vm_page_lookup(new_object, idx); |
| 2840 | if (m) |
| 2841 | vm_page_wakeup(m); |
| 2842 | } |
| 2843 | |
| 2844 | entry->object.vm_object = new_object; |
| 2845 | entry->offset = 0LL; |
| 2846 | vm_object_deallocate(orig_object); |
| 2847 | } |
| 2848 | |
| 2849 | /* |
| 2850 | * vm_map_copy_entry: |
| 2851 | * |
| 2852 | * Copies the contents of the source entry to the destination |
| 2853 | * entry. The entries *must* be aligned properly. |
| 2854 | */ |
| 2855 | static void |
| 2856 | vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map, |
| 2857 | vm_map_entry_t src_entry, vm_map_entry_t dst_entry) |
| 2858 | { |
| 2859 | vm_object_t src_object; |
| 2860 | |
| 2861 | if (dst_entry->maptype == VM_MAPTYPE_SUBMAP) |
| 2862 | return; |
| 2863 | if (src_entry->maptype == VM_MAPTYPE_SUBMAP) |
| 2864 | return; |
| 2865 | |
| 2866 | if (src_entry->wired_count == 0) { |
| 2867 | /* |
| 2868 | * If the source entry is marked needs_copy, it is already |
| 2869 | * write-protected. |
| 2870 | */ |
| 2871 | if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) { |
| 2872 | pmap_protect(src_map->pmap, |
| 2873 | src_entry->start, |
| 2874 | src_entry->end, |
| 2875 | src_entry->protection & ~VM_PROT_WRITE); |
| 2876 | } |
| 2877 | |
| 2878 | /* |
| 2879 | * Make a copy of the object. |
| 2880 | */ |
| 2881 | if ((src_object = src_entry->object.vm_object) != NULL) { |
| 2882 | if ((src_object->handle == NULL) && |
| 2883 | (src_object->type == OBJT_DEFAULT || |
| 2884 | src_object->type == OBJT_SWAP)) { |
| 2885 | vm_object_collapse(src_object); |
| 2886 | if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { |
| 2887 | vm_map_split(src_entry); |
| 2888 | src_object = src_entry->object.vm_object; |
| 2889 | } |
| 2890 | } |
| 2891 | |
| 2892 | vm_object_reference(src_object); |
| 2893 | vm_object_clear_flag(src_object, OBJ_ONEMAPPING); |
| 2894 | dst_entry->object.vm_object = src_object; |
| 2895 | src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); |
| 2896 | dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); |
| 2897 | dst_entry->offset = src_entry->offset; |
| 2898 | } else { |
| 2899 | dst_entry->object.vm_object = NULL; |
| 2900 | dst_entry->offset = 0; |
| 2901 | } |
| 2902 | |
| 2903 | pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, |
| 2904 | dst_entry->end - dst_entry->start, src_entry->start); |
| 2905 | } else { |
| 2906 | /* |
| 2907 | * Of course, wired down pages can't be set copy-on-write. |
| 2908 | * Cause wired pages to be copied into the new map by |
| 2909 | * simulating faults (the new pages are pageable) |
| 2910 | */ |
| 2911 | vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry); |
| 2912 | } |
| 2913 | } |
| 2914 | |
| 2915 | /* |
| 2916 | * vmspace_fork: |
| 2917 | * Create a new process vmspace structure and vm_map |
| 2918 | * based on those of an existing process. The new map |
| 2919 | * is based on the old map, according to the inheritance |
| 2920 | * values on the regions in that map. |
| 2921 | * |
| 2922 | * The source map must not be locked. |
| 2923 | */ |
| 2924 | struct vmspace * |
| 2925 | vmspace_fork(struct vmspace *vm1) |
| 2926 | { |
| 2927 | struct vmspace *vm2; |
| 2928 | vm_map_t old_map = &vm1->vm_map; |
| 2929 | vm_map_t new_map; |
| 2930 | vm_map_entry_t old_entry; |
| 2931 | vm_map_entry_t new_entry; |
| 2932 | vm_object_t object; |
| 2933 | int count; |
| 2934 | |
| 2935 | vm_map_lock(old_map); |
| 2936 | old_map->infork = 1; |
| 2937 | |
| 2938 | /* |
| 2939 | * XXX Note: upcalls are not copied. |
| 2940 | */ |
| 2941 | vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset); |
| 2942 | bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy, |
| 2943 | (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy); |
| 2944 | new_map = &vm2->vm_map; /* XXX */ |
| 2945 | new_map->timestamp = 1; |
| 2946 | |
| 2947 | count = 0; |
| 2948 | old_entry = old_map->header.next; |
| 2949 | while (old_entry != &old_map->header) { |
| 2950 | ++count; |
| 2951 | old_entry = old_entry->next; |
| 2952 | } |
| 2953 | |
| 2954 | count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT); |
| 2955 | |
| 2956 | old_entry = old_map->header.next; |
| 2957 | while (old_entry != &old_map->header) { |
| 2958 | if (old_entry->maptype == VM_MAPTYPE_SUBMAP) |
| 2959 | panic("vm_map_fork: encountered a submap"); |
| 2960 | |
| 2961 | switch (old_entry->inheritance) { |
| 2962 | case VM_INHERIT_NONE: |
| 2963 | break; |
| 2964 | |
| 2965 | case VM_INHERIT_SHARE: |
| 2966 | /* |
| 2967 | * Clone the entry, creating the shared object if |
| 2968 | * necessary. |
| 2969 | */ |
| 2970 | object = old_entry->object.vm_object; |
| 2971 | if (object == NULL) { |
| 2972 | vm_map_entry_allocate_object(old_entry); |
| 2973 | object = old_entry->object.vm_object; |
| 2974 | } |
| 2975 | |
| 2976 | /* |
| 2977 | * Add the reference before calling vm_map_entry_shadow |
| 2978 | * to insure that a shadow object is created. |
| 2979 | */ |
| 2980 | vm_object_reference(object); |
| 2981 | if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { |
| 2982 | vm_map_entry_shadow(old_entry); |
| 2983 | /* Transfer the second reference too. */ |
| 2984 | vm_object_reference( |
| 2985 | old_entry->object.vm_object); |
| 2986 | vm_object_deallocate(object); |
| 2987 | object = old_entry->object.vm_object; |
| 2988 | } |
| 2989 | vm_object_clear_flag(object, OBJ_ONEMAPPING); |
| 2990 | |
| 2991 | /* |
| 2992 | * Clone the entry, referencing the shared object. |
| 2993 | */ |
| 2994 | new_entry = vm_map_entry_create(new_map, &count); |
| 2995 | *new_entry = *old_entry; |
| 2996 | new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; |
| 2997 | new_entry->wired_count = 0; |
| 2998 | |
| 2999 | /* |
| 3000 | * Insert the entry into the new map -- we know we're |
| 3001 | * inserting at the end of the new map. |
| 3002 | */ |
| 3003 | |
| 3004 | vm_map_entry_link(new_map, new_map->header.prev, |
| 3005 | new_entry); |
| 3006 | |
| 3007 | /* |
| 3008 | * Update the physical map |
| 3009 | */ |
| 3010 | |
| 3011 | pmap_copy(new_map->pmap, old_map->pmap, |
| 3012 | new_entry->start, |
| 3013 | (old_entry->end - old_entry->start), |
| 3014 | old_entry->start); |
| 3015 | break; |
| 3016 | |
| 3017 | case VM_INHERIT_COPY: |
| 3018 | /* |
| 3019 | * Clone the entry and link into the map. |
| 3020 | */ |
| 3021 | new_entry = vm_map_entry_create(new_map, &count); |
| 3022 | *new_entry = *old_entry; |
| 3023 | new_entry->eflags &= ~MAP_ENTRY_USER_WIRED; |
| 3024 | new_entry->wired_count = 0; |
| 3025 | new_entry->object.vm_object = NULL; |
| 3026 | vm_map_entry_link(new_map, new_map->header.prev, |
| 3027 | new_entry); |
| 3028 | vm_map_copy_entry(old_map, new_map, old_entry, |
| 3029 | new_entry); |
| 3030 | break; |
| 3031 | } |
| 3032 | old_entry = old_entry->next; |
| 3033 | } |
| 3034 | |
| 3035 | new_map->size = old_map->size; |
| 3036 | old_map->infork = 0; |
| 3037 | vm_map_unlock(old_map); |
| 3038 | vm_map_entry_release(count); |
| 3039 | |
| 3040 | return (vm2); |
| 3041 | } |
| 3042 | |
| 3043 | int |
| 3044 | vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, |
| 3045 | int flags, vm_prot_t prot, vm_prot_t max, int cow) |
| 3046 | { |
| 3047 | vm_map_entry_t prev_entry; |
| 3048 | vm_map_entry_t new_stack_entry; |
| 3049 | vm_size_t init_ssize; |
| 3050 | int rv; |
| 3051 | int count; |
| 3052 | vm_offset_t tmpaddr; |
| 3053 | |
| 3054 | cow |= MAP_IS_STACK; |
| 3055 | |
| 3056 | if (max_ssize < sgrowsiz) |
| 3057 | init_ssize = max_ssize; |
| 3058 | else |
| 3059 | init_ssize = sgrowsiz; |
| 3060 | |
| 3061 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 3062 | vm_map_lock(map); |
| 3063 | |
| 3064 | /* |
| 3065 | * Find space for the mapping |
| 3066 | */ |
| 3067 | if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) { |
| 3068 | if (vm_map_findspace(map, addrbos, max_ssize, 1, |
| 3069 | flags, &tmpaddr)) { |
| 3070 | vm_map_unlock(map); |
| 3071 | vm_map_entry_release(count); |
| 3072 | return (KERN_NO_SPACE); |
| 3073 | } |
| 3074 | addrbos = tmpaddr; |
| 3075 | } |
| 3076 | |
| 3077 | /* If addr is already mapped, no go */ |
| 3078 | if (vm_map_lookup_entry(map, addrbos, &prev_entry)) { |
| 3079 | vm_map_unlock(map); |
| 3080 | vm_map_entry_release(count); |
| 3081 | return (KERN_NO_SPACE); |
| 3082 | } |
| 3083 | |
| 3084 | #if 0 |
| 3085 | /* XXX already handled by kern_mmap() */ |
| 3086 | /* If we would blow our VMEM resource limit, no go */ |
| 3087 | if (map->size + init_ssize > |
| 3088 | curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) { |
| 3089 | vm_map_unlock(map); |
| 3090 | vm_map_entry_release(count); |
| 3091 | return (KERN_NO_SPACE); |
| 3092 | } |
| 3093 | #endif |
| 3094 | |
| 3095 | /* |
| 3096 | * If we can't accomodate max_ssize in the current mapping, |
| 3097 | * no go. However, we need to be aware that subsequent user |
| 3098 | * mappings might map into the space we have reserved for |
| 3099 | * stack, and currently this space is not protected. |
| 3100 | * |
| 3101 | * Hopefully we will at least detect this condition |
| 3102 | * when we try to grow the stack. |
| 3103 | */ |
| 3104 | if ((prev_entry->next != &map->header) && |
| 3105 | (prev_entry->next->start < addrbos + max_ssize)) { |
| 3106 | vm_map_unlock(map); |
| 3107 | vm_map_entry_release(count); |
| 3108 | return (KERN_NO_SPACE); |
| 3109 | } |
| 3110 | |
| 3111 | /* |
| 3112 | * We initially map a stack of only init_ssize. We will |
| 3113 | * grow as needed later. Since this is to be a grow |
| 3114 | * down stack, we map at the top of the range. |
| 3115 | * |
| 3116 | * Note: we would normally expect prot and max to be |
| 3117 | * VM_PROT_ALL, and cow to be 0. Possibly we should |
| 3118 | * eliminate these as input parameters, and just |
| 3119 | * pass these values here in the insert call. |
| 3120 | */ |
| 3121 | rv = vm_map_insert(map, &count, |
| 3122 | NULL, 0, addrbos + max_ssize - init_ssize, |
| 3123 | addrbos + max_ssize, |
| 3124 | VM_MAPTYPE_NORMAL, |
| 3125 | prot, max, |
| 3126 | cow); |
| 3127 | |
| 3128 | /* Now set the avail_ssize amount */ |
| 3129 | if (rv == KERN_SUCCESS) { |
| 3130 | if (prev_entry != &map->header) |
| 3131 | vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count); |
| 3132 | new_stack_entry = prev_entry->next; |
| 3133 | if (new_stack_entry->end != addrbos + max_ssize || |
| 3134 | new_stack_entry->start != addrbos + max_ssize - init_ssize) |
| 3135 | panic ("Bad entry start/end for new stack entry"); |
| 3136 | else |
| 3137 | new_stack_entry->aux.avail_ssize = max_ssize - init_ssize; |
| 3138 | } |
| 3139 | |
| 3140 | vm_map_unlock(map); |
| 3141 | vm_map_entry_release(count); |
| 3142 | return (rv); |
| 3143 | } |
| 3144 | |
| 3145 | /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the |
| 3146 | * desired address is already mapped, or if we successfully grow |
| 3147 | * the stack. Also returns KERN_SUCCESS if addr is outside the |
| 3148 | * stack range (this is strange, but preserves compatibility with |
| 3149 | * the grow function in vm_machdep.c). |
| 3150 | */ |
| 3151 | int |
| 3152 | vm_map_growstack (struct proc *p, vm_offset_t addr) |
| 3153 | { |
| 3154 | vm_map_entry_t prev_entry; |
| 3155 | vm_map_entry_t stack_entry; |
| 3156 | vm_map_entry_t new_stack_entry; |
| 3157 | struct vmspace *vm = p->p_vmspace; |
| 3158 | vm_map_t map = &vm->vm_map; |
| 3159 | vm_offset_t end; |
| 3160 | int grow_amount; |
| 3161 | int rv = KERN_SUCCESS; |
| 3162 | int is_procstack; |
| 3163 | int use_read_lock = 1; |
| 3164 | int count; |
| 3165 | |
| 3166 | count = vm_map_entry_reserve(MAP_RESERVE_COUNT); |
| 3167 | Retry: |
| 3168 | if (use_read_lock) |
| 3169 | vm_map_lock_read(map); |
| 3170 | else |
| 3171 | vm_map_lock(map); |
| 3172 | |
| 3173 | /* If addr is already in the entry range, no need to grow.*/ |
| 3174 | if (vm_map_lookup_entry(map, addr, &prev_entry)) |
| 3175 | goto done; |
| 3176 | |
| 3177 | if ((stack_entry = prev_entry->next) == &map->header) |
| 3178 | goto done; |
| 3179 | if (prev_entry == &map->header) |
| 3180 | end = stack_entry->start - stack_entry->aux.avail_ssize; |
| 3181 | else |
| 3182 | end = prev_entry->end; |
| 3183 | |
| 3184 | /* |
| 3185 | * This next test mimics the old grow function in vm_machdep.c. |
| 3186 | * It really doesn't quite make sense, but we do it anyway |
| 3187 | * for compatibility. |
| 3188 | * |
| 3189 | * If not growable stack, return success. This signals the |
| 3190 | * caller to proceed as he would normally with normal vm. |
| 3191 | */ |
| 3192 | if (stack_entry->aux.avail_ssize < 1 || |
| 3193 | addr >= stack_entry->start || |
| 3194 | addr < stack_entry->start - stack_entry->aux.avail_ssize) { |
| 3195 | goto done; |
| 3196 | } |
| 3197 | |
| 3198 | /* Find the minimum grow amount */ |
| 3199 | grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE); |
| 3200 | if (grow_amount > stack_entry->aux.avail_ssize) { |
| 3201 | rv = KERN_NO_SPACE; |
| 3202 | goto done; |
| 3203 | } |
| 3204 | |
| 3205 | /* |
| 3206 | * If there is no longer enough space between the entries |
| 3207 | * nogo, and adjust the available space. Note: this |
| 3208 | * should only happen if the user has mapped into the |
| 3209 | * stack area after the stack was created, and is |
| 3210 | * probably an error. |
| 3211 | * |
| 3212 | * This also effectively destroys any guard page the user |
| 3213 | * might have intended by limiting the stack size. |
| 3214 | */ |
| 3215 | if (grow_amount > stack_entry->start - end) { |
| 3216 | if (use_read_lock && vm_map_lock_upgrade(map)) { |
| 3217 | use_read_lock = 0; |
| 3218 | goto Retry; |
| 3219 | } |
| 3220 | use_read_lock = 0; |
| 3221 | stack_entry->aux.avail_ssize = stack_entry->start - end; |
| 3222 | rv = KERN_NO_SPACE; |
| 3223 | goto done; |
| 3224 | } |
| 3225 | |
| 3226 | is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr; |
| 3227 | |
| 3228 | /* If this is the main process stack, see if we're over the |
| 3229 | * stack limit. |
| 3230 | */ |
| 3231 | if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > |
| 3232 | p->p_rlimit[RLIMIT_STACK].rlim_cur)) { |
| 3233 | rv = KERN_NO_SPACE; |
| 3234 | goto done; |
| 3235 | } |
| 3236 | |
| 3237 | /* Round up the grow amount modulo SGROWSIZ */ |
| 3238 | grow_amount = roundup (grow_amount, sgrowsiz); |
| 3239 | if (grow_amount > stack_entry->aux.avail_ssize) { |
| 3240 | grow_amount = stack_entry->aux.avail_ssize; |
| 3241 | } |
| 3242 | if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > |
| 3243 | p->p_rlimit[RLIMIT_STACK].rlim_cur)) { |
| 3244 | grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur - |
| 3245 | ctob(vm->vm_ssize); |
| 3246 | } |
| 3247 | |
| 3248 | /* If we would blow our VMEM resource limit, no go */ |
| 3249 | if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) { |
| 3250 | rv = KERN_NO_SPACE; |
| 3251 | goto done; |
| 3252 | } |
| 3253 | |
| 3254 | if (use_read_lock && vm_map_lock_upgrade(map)) { |
| 3255 | use_read_lock = 0; |
| 3256 | goto Retry; |
| 3257 | } |
| 3258 | use_read_lock = 0; |
| 3259 | |
| 3260 | /* Get the preliminary new entry start value */ |
| 3261 | addr = stack_entry->start - grow_amount; |
| 3262 | |
| 3263 | /* If this puts us into the previous entry, cut back our growth |
| 3264 | * to the available space. Also, see the note above. |
| 3265 | */ |
| 3266 | if (addr < end) { |
| 3267 | stack_entry->aux.avail_ssize = stack_entry->start - end; |
| 3268 | addr = end; |
| 3269 | } |
| 3270 | |
| 3271 | rv = vm_map_insert(map, &count, |
| 3272 | NULL, 0, addr, stack_entry->start, |
| 3273 | VM_MAPTYPE_NORMAL, |
| 3274 | VM_PROT_ALL, VM_PROT_ALL, |
| 3275 | 0); |
| 3276 | |
| 3277 | /* Adjust the available stack space by the amount we grew. */ |
| 3278 | if (rv == KERN_SUCCESS) { |
| 3279 | if (prev_entry != &map->header) |
| 3280 | vm_map_clip_end(map, prev_entry, addr, &count); |
| 3281 | new_stack_entry = prev_entry->next; |
| 3282 | if (new_stack_entry->end != stack_entry->start || |
| 3283 | new_stack_entry->start != addr) |
| 3284 | panic ("Bad stack grow start/end in new stack entry"); |
| 3285 | else { |
| 3286 | new_stack_entry->aux.avail_ssize = |
| 3287 | stack_entry->aux.avail_ssize - |
| 3288 | (new_stack_entry->end - new_stack_entry->start); |
| 3289 | if (is_procstack) |
| 3290 | vm->vm_ssize += btoc(new_stack_entry->end - |
| 3291 | new_stack_entry->start); |
| 3292 | } |
| 3293 | } |
| 3294 | |
| 3295 | done: |
| 3296 | if (use_read_lock) |
| 3297 | vm_map_unlock_read(map); |
| 3298 | else |
| 3299 | vm_map_unlock(map); |
| 3300 | vm_map_entry_release(count); |
| 3301 | return (rv); |
| 3302 | } |
| 3303 | |
| 3304 | /* |
| 3305 | * Unshare the specified VM space for exec. If other processes are |
| 3306 | * mapped to it, then create a new one. The new vmspace is null. |
| 3307 | */ |
| 3308 | void |
| 3309 | vmspace_exec(struct proc *p, struct vmspace *vmcopy) |
| 3310 | { |
| 3311 | struct vmspace *oldvmspace = p->p_vmspace; |
| 3312 | struct vmspace *newvmspace; |
| 3313 | vm_map_t map = &p->p_vmspace->vm_map; |
| 3314 | |
| 3315 | /* |
| 3316 | * If we are execing a resident vmspace we fork it, otherwise |
| 3317 | * we create a new vmspace. Note that exitingcnt and upcalls |
| 3318 | * are not copied to the new vmspace. |
| 3319 | */ |
| 3320 | if (vmcopy) { |
| 3321 | newvmspace = vmspace_fork(vmcopy); |
| 3322 | } else { |
| 3323 | newvmspace = vmspace_alloc(map->min_offset, map->max_offset); |
| 3324 | bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy, |
| 3325 | (caddr_t)&oldvmspace->vm_endcopy - |
| 3326 | (caddr_t)&oldvmspace->vm_startcopy); |
| 3327 | } |
| 3328 | |
| 3329 | /* |
| 3330 | * Finish initializing the vmspace before assigning it |
| 3331 | * to the process. The vmspace will become the current vmspace |
| 3332 | * if p == curproc. |
| 3333 | */ |
| 3334 | pmap_pinit2(vmspace_pmap(newvmspace)); |
| 3335 | pmap_replacevm(p, newvmspace, 0); |
| 3336 | sysref_put(&oldvmspace->vm_sysref); |
| 3337 | } |
| 3338 | |
| 3339 | /* |
| 3340 | * Unshare the specified VM space for forcing COW. This |
| 3341 | * is called by rfork, for the (RFMEM|RFPROC) == 0 case. |
| 3342 | * |
| 3343 | * The exitingcnt test is not strictly necessary but has been |
| 3344 | * included for code sanity (to make the code a bit more deterministic). |
| 3345 | */ |
| 3346 | |
| 3347 | void |
| 3348 | vmspace_unshare(struct proc *p) |
| 3349 | { |
| 3350 | struct vmspace *oldvmspace = p->p_vmspace; |
| 3351 | struct vmspace *newvmspace; |
| 3352 | |
| 3353 | if (oldvmspace->vm_sysref.refcnt == 1 && oldvmspace->vm_exitingcnt == 0) |
| 3354 | return; |
| 3355 | newvmspace = vmspace_fork(oldvmspace); |
| 3356 | pmap_pinit2(vmspace_pmap(newvmspace)); |
| 3357 | pmap_replacevm(p, newvmspace, 0); |
| 3358 | sysref_put(&oldvmspace->vm_sysref); |
| 3359 | } |
| 3360 | |
| 3361 | /* |
| 3362 | * vm_map_lookup: |
| 3363 | * |
| 3364 | * Finds the VM object, offset, and |
| 3365 | * protection for a given virtual address in the |
| 3366 | * specified map, assuming a page fault of the |
| 3367 | * type specified. |
| 3368 | * |
| 3369 | * Leaves the map in question locked for read; return |
| 3370 | * values are guaranteed until a vm_map_lookup_done |
| 3371 | * call is performed. Note that the map argument |
| 3372 | * is in/out; the returned map must be used in |
| 3373 | * the call to vm_map_lookup_done. |
| 3374 | * |
| 3375 | * A handle (out_entry) is returned for use in |
| 3376 | * vm_map_lookup_done, to make that fast. |
| 3377 | * |
| 3378 | * If a lookup is requested with "write protection" |
| 3379 | * specified, the map may be changed to perform virtual |
| 3380 | * copying operations, although the data referenced will |
| 3381 | * remain the same. |
| 3382 | */ |
| 3383 | int |
| 3384 | vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ |
| 3385 | vm_offset_t vaddr, |
| 3386 | vm_prot_t fault_typea, |
| 3387 | vm_map_entry_t *out_entry, /* OUT */ |
| 3388 | vm_object_t *object, /* OUT */ |
| 3389 | vm_pindex_t *pindex, /* OUT */ |
| 3390 | vm_prot_t *out_prot, /* OUT */ |
| 3391 | boolean_t *wired) /* OUT */ |
| 3392 | { |
| 3393 | vm_map_entry_t entry; |
| 3394 | vm_map_t map = *var_map; |
| 3395 | vm_prot_t prot; |
| 3396 | vm_prot_t fault_type = fault_typea; |
| 3397 | int use_read_lock = 1; |
| 3398 | int rv = KERN_SUCCESS; |
| 3399 | |
| 3400 | RetryLookup: |
| 3401 | if (use_read_lock) |
| 3402 | vm_map_lock_read(map); |
| 3403 | else |
| 3404 | vm_map_lock(map); |
| 3405 | |
| 3406 | /* |
| 3407 | * If the map has an interesting hint, try it before calling full |
| 3408 | * blown lookup routine. |
| 3409 | */ |
| 3410 | entry = map->hint; |
| 3411 | *out_entry = entry; |
| 3412 | |
| 3413 | if ((entry == &map->header) || |
| 3414 | (vaddr < entry->start) || (vaddr >= entry->end)) { |
| 3415 | vm_map_entry_t tmp_entry; |
| 3416 | |
| 3417 | /* |
| 3418 | * Entry was either not a valid hint, or the vaddr was not |
| 3419 | * contained in the entry, so do a full lookup. |
| 3420 | */ |
| 3421 | if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) { |
| 3422 | rv = KERN_INVALID_ADDRESS; |
| 3423 | goto done; |
| 3424 | } |
| 3425 | |
| 3426 | entry = tmp_entry; |
| 3427 | *out_entry = entry; |
| 3428 | } |
| 3429 | |
| 3430 | /* |
| 3431 | * Handle submaps. |
| 3432 | */ |
| 3433 | if (entry->maptype == VM_MAPTYPE_SUBMAP) { |
| 3434 | vm_map_t old_map = map; |
| 3435 | |
| 3436 | *var_map = map = entry->object.sub_map; |
| 3437 | if (use_read_lock) |
| 3438 | vm_map_unlock_read(old_map); |
| 3439 | else |
| 3440 | vm_map_unlock(old_map); |
| 3441 | use_read_lock = 1; |
| 3442 | goto RetryLookup; |
| 3443 | } |
| 3444 | |
| 3445 | /* |
| 3446 | * Check whether this task is allowed to have this page. |
| 3447 | * Note the special case for MAP_ENTRY_COW |
| 3448 | * pages with an override. This is to implement a forced |
| 3449 | * COW for debuggers. |
| 3450 | */ |
| 3451 | |
| 3452 | if (fault_type & VM_PROT_OVERRIDE_WRITE) |
| 3453 | prot = entry->max_protection; |
| 3454 | else |
| 3455 | prot = entry->protection; |
| 3456 | |
| 3457 | fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); |
| 3458 | if ((fault_type & prot) != fault_type) { |
| 3459 | rv = KERN_PROTECTION_FAILURE; |
| 3460 | goto done; |
| 3461 | } |
| 3462 | |
| 3463 | if ((entry->eflags & MAP_ENTRY_USER_WIRED) && |
| 3464 | (entry->eflags & MAP_ENTRY_COW) && |
| 3465 | (fault_type & VM_PROT_WRITE) && |
| 3466 | (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) { |
| 3467 | rv = KERN_PROTECTION_FAILURE; |
| 3468 | goto done; |
| 3469 | } |
| 3470 | |
| 3471 | /* |
| 3472 | * If this page is not pageable, we have to get it for all possible |
| 3473 | * accesses. |
| 3474 | */ |
| 3475 | *wired = (entry->wired_count != 0); |
| 3476 | if (*wired) |
| 3477 | prot = fault_type = entry->protection; |
| 3478 | |
| 3479 | /* |
| 3480 | * Virtual page tables may need to update the accessed (A) bit |
| 3481 | * in a page table entry. Upgrade the fault to a write fault for |
| 3482 | * that case if the map will support it. If the map does not support |
| 3483 | * it the page table entry simply will not be updated. |
| 3484 | */ |
| 3485 | if (entry->maptype == VM_MAPTYPE_VPAGETABLE) { |
| 3486 | if (prot & VM_PROT_WRITE) |
| 3487 | fault_type |= VM_PROT_WRITE; |
| 3488 | } |
| 3489 | |
| 3490 | /* |
| 3491 | * If the entry was copy-on-write, we either ... |
| 3492 | */ |
| 3493 | if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { |
| 3494 | /* |
| 3495 | * If we want to write the page, we may as well handle that |
| 3496 | * now since we've got the map locked. |
| 3497 | * |
| 3498 | * If we don't need to write the page, we just demote the |
| 3499 | * permissions allowed. |
| 3500 | */ |
| 3501 | |
| 3502 | if (fault_type & VM_PROT_WRITE) { |
| 3503 | /* |
| 3504 | * Make a new object, and place it in the object |
| 3505 | * chain. Note that no new references have appeared |
| 3506 | * -- one just moved from the map to the new |
| 3507 | * object. |
| 3508 | */ |
| 3509 | |
| 3510 | if (use_read_lock && vm_map_lock_upgrade(map)) { |
| 3511 | use_read_lock = 0; |
| 3512 | goto RetryLookup; |
| 3513 | } |
| 3514 | use_read_lock = 0; |
| 3515 | |
| 3516 | vm_map_entry_shadow(entry); |
| 3517 | } else { |
| 3518 | /* |
| 3519 | * We're attempting to read a copy-on-write page -- |
| 3520 | * don't allow writes. |
| 3521 | */ |
| 3522 | |
| 3523 | prot &= ~VM_PROT_WRITE; |
| 3524 | } |
| 3525 | } |
| 3526 | |
| 3527 | /* |
| 3528 | * Create an object if necessary. |
| 3529 | */ |
| 3530 | if (entry->object.vm_object == NULL && |
| 3531 | !map->system_map) { |
| 3532 | if (use_read_lock && vm_map_lock_upgrade(map)) { |
| 3533 | use_read_lock = 0; |
| 3534 | goto RetryLookup; |
| 3535 | } |
| 3536 | use_read_lock = 0; |
| 3537 | vm_map_entry_allocate_object(entry); |
| 3538 | } |
| 3539 | |
| 3540 | /* |
| 3541 | * Return the object/offset from this entry. If the entry was |
| 3542 | * copy-on-write or empty, it has been fixed up. |
| 3543 | */ |
| 3544 | |
| 3545 | *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); |
| 3546 | *object = entry->object.vm_object; |
| 3547 | |
| 3548 | /* |
| 3549 | * Return whether this is the only map sharing this data. On |
| 3550 | * success we return with a read lock held on the map. On failure |
| 3551 | * we return with the map unlocked. |
| 3552 | */ |
| 3553 | *out_prot = prot; |
| 3554 | done: |
| 3555 | if (rv == KERN_SUCCESS) { |
| 3556 | if (use_read_lock == 0) |
| 3557 | vm_map_lock_downgrade(map); |
| 3558 | } else if (use_read_lock) { |
| 3559 | vm_map_unlock_read(map); |
| 3560 | } else { |
| 3561 | vm_map_unlock(map); |
| 3562 | } |
| 3563 | return (rv); |
| 3564 | } |
| 3565 | |
| 3566 | /* |
| 3567 | * vm_map_lookup_done: |
| 3568 | * |
| 3569 | * Releases locks acquired by a vm_map_lookup |
| 3570 | * (according to the handle returned by that lookup). |
| 3571 | */ |
| 3572 | |
| 3573 | void |
| 3574 | vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count) |
| 3575 | { |
| 3576 | /* |
| 3577 | * Unlock the main-level map |
| 3578 | */ |
| 3579 | vm_map_unlock_read(map); |
| 3580 | if (count) |
| 3581 | vm_map_entry_release(count); |
| 3582 | } |
| 3583 | |
| 3584 | #include "opt_ddb.h" |
| 3585 | #ifdef DDB |
| 3586 | #include <sys/kernel.h> |
| 3587 | |
| 3588 | #include <ddb/ddb.h> |
| 3589 | |
| 3590 | /* |
| 3591 | * vm_map_print: [ debug ] |
| 3592 | */ |
| 3593 | DB_SHOW_COMMAND(map, vm_map_print) |
| 3594 | { |
| 3595 | static int nlines; |
| 3596 | /* XXX convert args. */ |
| 3597 | vm_map_t map = (vm_map_t)addr; |
| 3598 | boolean_t full = have_addr; |
| 3599 | |
| 3600 | vm_map_entry_t entry; |
| 3601 | |
| 3602 | db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", |
| 3603 | (void *)map, |
| 3604 | (void *)map->pmap, map->nentries, map->timestamp); |
| 3605 | nlines++; |
| 3606 | |
| 3607 | if (!full && db_indent) |
| 3608 | return; |
| 3609 | |
| 3610 | db_indent += 2; |
| 3611 | for (entry = map->header.next; entry != &map->header; |
| 3612 | entry = entry->next) { |
| 3613 | db_iprintf("map entry %p: start=%p, end=%p\n", |
| 3614 | (void *)entry, (void *)entry->start, (void *)entry->end); |
| 3615 | nlines++; |
| 3616 | { |
| 3617 | static char *inheritance_name[4] = |
| 3618 | {"share", "copy", "none", "donate_copy"}; |
| 3619 | |
| 3620 | db_iprintf(" prot=%x/%x/%s", |
| 3621 | entry->protection, |
| 3622 | entry->max_protection, |
| 3623 | inheritance_name[(int)(unsigned char)entry->inheritance]); |
| 3624 | if (entry->wired_count != 0) |
| 3625 | db_printf(", wired"); |
| 3626 | } |
| 3627 | if (entry->maptype == VM_MAPTYPE_SUBMAP) { |
| 3628 | /* XXX no %qd in kernel. Truncate entry->offset. */ |
| 3629 | db_printf(", share=%p, offset=0x%lx\n", |
| 3630 | (void *)entry->object.sub_map, |
| 3631 | (long)entry->offset); |
| 3632 | nlines++; |
| 3633 | if ((entry->prev == &map->header) || |
| 3634 | (entry->prev->object.sub_map != |
| 3635 | entry->object.sub_map)) { |
| 3636 | db_indent += 2; |
| 3637 | vm_map_print((db_expr_t)(intptr_t) |
| 3638 | entry->object.sub_map, |
| 3639 | full, 0, NULL); |
| 3640 | db_indent -= 2; |
| 3641 | } |
| 3642 | } else { |
| 3643 | /* XXX no %qd in kernel. Truncate entry->offset. */ |
| 3644 | db_printf(", object=%p, offset=0x%lx", |
| 3645 | (void *)entry->object.vm_object, |
| 3646 | (long)entry->offset); |
| 3647 | if (entry->eflags & MAP_ENTRY_COW) |
| 3648 | db_printf(", copy (%s)", |
| 3649 | (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); |
| 3650 | db_printf("\n"); |
| 3651 | nlines++; |
| 3652 | |
| 3653 | if ((entry->prev == &map->header) || |
| 3654 | (entry->prev->object.vm_object != |
| 3655 | entry->object.vm_object)) { |
| 3656 | db_indent += 2; |
| 3657 | vm_object_print((db_expr_t)(intptr_t) |
| 3658 | entry->object.vm_object, |
| 3659 | full, 0, NULL); |
| 3660 | nlines += 4; |
| 3661 | db_indent -= 2; |
| 3662 | } |
| 3663 | } |
| 3664 | } |
| 3665 | db_indent -= 2; |
| 3666 | if (db_indent == 0) |
| 3667 | nlines = 0; |
| 3668 | } |
| 3669 | |
| 3670 | |
| 3671 | DB_SHOW_COMMAND(procvm, procvm) |
| 3672 | { |
| 3673 | struct proc *p; |
| 3674 | |
| 3675 | if (have_addr) { |
| 3676 | p = (struct proc *) addr; |
| 3677 | } else { |
| 3678 | p = curproc; |
| 3679 | } |
| 3680 | |
| 3681 | db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", |
| 3682 | (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, |
| 3683 | (void *)vmspace_pmap(p->p_vmspace)); |
| 3684 | |
| 3685 | vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL); |
| 3686 | } |
| 3687 | |
| 3688 | #endif /* DDB */ |