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 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 33 * 34 * 35 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36 * All rights reserved. 37 * 38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61 /* 62 * Virtual memory object module. 63 */ 64 65 #include <sys/cdefs.h> 66 __FBSDID("$FreeBSD$"); 67 68 #include "opt_vm.h" 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/lock.h> 73 #include <sys/mman.h> 74 #include <sys/mount.h> 75 #include <sys/kernel.h> 76 #include <sys/sysctl.h> 77 #include <sys/mutex.h> 78 #include <sys/proc.h> /* for curproc, pageproc */ 79 #include <sys/socket.h> 80 #include <sys/resourcevar.h> 81 #include <sys/vnode.h> 82 #include <sys/vmmeter.h> 83 #include <sys/sx.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_object.h> 90 #include <vm/vm_page.h> 91 #include <vm/vm_pageout.h> 92 #include <vm/vm_pager.h> 93 #include <vm/swap_pager.h> 94 #include <vm/vm_kern.h> 95 #include <vm/vm_extern.h> 96 #include <vm/vm_reserv.h> 97 #include <vm/uma.h> 98 99 static int old_msync; 100 SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, 101 "Use old (insecure) msync behavior"); 102 103 static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, 104 int pagerflags, int flags, int *clearobjflags); 105 static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, 106 int *clearobjflags); 107 static void vm_object_qcollapse(vm_object_t object); 108 static void vm_object_vndeallocate(vm_object_t object); 109 110 /* 111 * Virtual memory objects maintain the actual data 112 * associated with allocated virtual memory. A given 113 * page of memory exists within exactly one object. 114 * 115 * An object is only deallocated when all "references" 116 * are given up. Only one "reference" to a given 117 * region of an object should be writeable. 118 * 119 * Associated with each object is a list of all resident 120 * memory pages belonging to that object; this list is 121 * maintained by the "vm_page" module, and locked by the object's 122 * lock. 123 * 124 * Each object also records a "pager" routine which is 125 * used to retrieve (and store) pages to the proper backing 126 * storage. In addition, objects may be backed by other 127 * objects from which they were virtual-copied. 128 * 129 * The only items within the object structure which are 130 * modified after time of creation are: 131 * reference count locked by object's lock 132 * pager routine locked by object's lock 133 * 134 */ 135 136 struct object_q vm_object_list; 137 struct mtx vm_object_list_mtx; /* lock for object list and count */ 138 139 struct vm_object kernel_object_store; 140 struct vm_object kmem_object_store; 141 142 SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats"); 143 144 static long object_collapses; 145 SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, 146 &object_collapses, 0, "VM object collapses"); 147 148 static long object_bypasses; 149 SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, 150 &object_bypasses, 0, "VM object bypasses"); 151 152 static uma_zone_t obj_zone; 153 154 static int vm_object_zinit(void *mem, int size, int flags); 155 156 #ifdef INVARIANTS 157 static void vm_object_zdtor(void *mem, int size, void *arg); 158 159 static void 160 vm_object_zdtor(void *mem, int size, void *arg) 161 { 162 vm_object_t object; 163 164 object = (vm_object_t)mem; 165 KASSERT(TAILQ_EMPTY(&object->memq), 166 ("object %p has resident pages", 167 object)); 168 #if VM_NRESERVLEVEL > 0 169 KASSERT(LIST_EMPTY(&object->rvq), 170 ("object %p has reservations", 171 object)); 172 #endif 173 KASSERT(object->cache == NULL, 174 ("object %p has cached pages", 175 object)); 176 KASSERT(object->paging_in_progress == 0, 177 ("object %p paging_in_progress = %d", 178 object, object->paging_in_progress)); 179 KASSERT(object->resident_page_count == 0, 180 ("object %p resident_page_count = %d", 181 object, object->resident_page_count)); 182 KASSERT(object->shadow_count == 0, 183 ("object %p shadow_count = %d", 184 object, object->shadow_count)); 185 } 186 #endif 187 188 static int 189 vm_object_zinit(void *mem, int size, int flags) 190 { 191 vm_object_t object; 192 193 object = (vm_object_t)mem; 194 bzero(&object->mtx, sizeof(object->mtx)); 195 VM_OBJECT_LOCK_INIT(object, "standard object"); 196 197 /* These are true for any object that has been freed */ 198 object->paging_in_progress = 0; 199 object->resident_page_count = 0; 200 object->shadow_count = 0; 201 return (0); 202 } 203 204 void 205 _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object) 206 { 207 208 TAILQ_INIT(&object->memq); 209 LIST_INIT(&object->shadow_head); 210 211 object->root = NULL; 212 object->type = type; 213 object->size = size; 214 object->generation = 1; 215 object->ref_count = 1; 216 object->memattr = VM_MEMATTR_DEFAULT; 217 object->flags = 0; 218 object->cred = NULL; 219 object->charge = 0; 220 if ((object->type == OBJT_DEFAULT) || (object->type == OBJT_SWAP)) 221 object->flags = OBJ_ONEMAPPING; 222 object->pg_color = 0; 223 object->handle = NULL; 224 object->backing_object = NULL; 225 object->backing_object_offset = (vm_ooffset_t) 0; 226 #if VM_NRESERVLEVEL > 0 227 LIST_INIT(&object->rvq); 228 #endif 229 object->cache = NULL; 230 231 mtx_lock(&vm_object_list_mtx); 232 TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); 233 mtx_unlock(&vm_object_list_mtx); 234 } 235 236 /* 237 * vm_object_init: 238 * 239 * Initialize the VM objects module. 240 */ 241 void 242 vm_object_init(void) 243 { 244 TAILQ_INIT(&vm_object_list); 245 mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); 246 247 VM_OBJECT_LOCK_INIT(kernel_object, "kernel object"); 248 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), 249 kernel_object); 250 #if VM_NRESERVLEVEL > 0 251 kernel_object->flags |= OBJ_COLORED; 252 kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); 253 #endif 254 255 VM_OBJECT_LOCK_INIT(kmem_object, "kmem object"); 256 _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), 257 kmem_object); 258 #if VM_NRESERVLEVEL > 0 259 kmem_object->flags |= OBJ_COLORED; 260 kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); 261 #endif 262 263 /* 264 * The lock portion of struct vm_object must be type stable due 265 * to vm_pageout_fallback_object_lock locking a vm object 266 * without holding any references to it. 267 */ 268 obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, 269 #ifdef INVARIANTS 270 vm_object_zdtor, 271 #else 272 NULL, 273 #endif 274 vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE); 275 } 276 277 void 278 vm_object_clear_flag(vm_object_t object, u_short bits) 279 { 280 281 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 282 object->flags &= ~bits; 283 } 284 285 /* 286 * Sets the default memory attribute for the specified object. Pages 287 * that are allocated to this object are by default assigned this memory 288 * attribute. 289 * 290 * Presently, this function must be called before any pages are allocated 291 * to the object. In the future, this requirement may be relaxed for 292 * "default" and "swap" objects. 293 */ 294 int 295 vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) 296 { 297 298 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 299 switch (object->type) { 300 case OBJT_DEFAULT: 301 case OBJT_DEVICE: 302 case OBJT_PHYS: 303 case OBJT_SG: 304 case OBJT_SWAP: 305 case OBJT_VNODE: 306 if (!TAILQ_EMPTY(&object->memq)) 307 return (KERN_FAILURE); 308 break; 309 case OBJT_DEAD: 310 return (KERN_INVALID_ARGUMENT); 311 } 312 object->memattr = memattr; 313 return (KERN_SUCCESS); 314 } 315 316 void 317 vm_object_pip_add(vm_object_t object, short i) 318 { 319 320 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 321 object->paging_in_progress += i; 322 } 323 324 void 325 vm_object_pip_subtract(vm_object_t object, short i) 326 { 327 328 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 329 object->paging_in_progress -= i; 330 } 331 332 void 333 vm_object_pip_wakeup(vm_object_t object) 334 { 335 336 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 337 object->paging_in_progress--; 338 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { 339 vm_object_clear_flag(object, OBJ_PIPWNT); 340 wakeup(object); 341 } 342 } 343 344 void 345 vm_object_pip_wakeupn(vm_object_t object, short i) 346 { 347 348 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 349 if (i) 350 object->paging_in_progress -= i; 351 if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { 352 vm_object_clear_flag(object, OBJ_PIPWNT); 353 wakeup(object); 354 } 355 } 356 357 void 358 vm_object_pip_wait(vm_object_t object, char *waitid) 359 { 360 361 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 362 while (object->paging_in_progress) { 363 object->flags |= OBJ_PIPWNT; 364 msleep(object, VM_OBJECT_MTX(object), PVM, waitid, 0); 365 } 366 } 367 368 /* 369 * vm_object_allocate: 370 * 371 * Returns a new object with the given size. 372 */ 373 vm_object_t 374 vm_object_allocate(objtype_t type, vm_pindex_t size) 375 { 376 vm_object_t object; 377 378 object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 379 _vm_object_allocate(type, size, object); 380 return (object); 381 } 382 383 384 /* 385 * vm_object_reference: 386 * 387 * Gets another reference to the given object. Note: OBJ_DEAD 388 * objects can be referenced during final cleaning. 389 */ 390 void 391 vm_object_reference(vm_object_t object) 392 { 393 if (object == NULL) 394 return; 395 VM_OBJECT_LOCK(object); 396 vm_object_reference_locked(object); 397 VM_OBJECT_UNLOCK(object); 398 } 399 400 /* 401 * vm_object_reference_locked: 402 * 403 * Gets another reference to the given object. 404 * 405 * The object must be locked. 406 */ 407 void 408 vm_object_reference_locked(vm_object_t object) 409 { 410 struct vnode *vp; 411 412 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 413 object->ref_count++; 414 if (object->type == OBJT_VNODE) { 415 vp = object->handle; 416 vref(vp); 417 } 418 } 419 420 /* 421 * Handle deallocating an object of type OBJT_VNODE. 422 */ 423 static void 424 vm_object_vndeallocate(vm_object_t object) 425 { 426 struct vnode *vp = (struct vnode *) object->handle; 427 428 VFS_ASSERT_GIANT(vp->v_mount); 429 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 430 KASSERT(object->type == OBJT_VNODE, 431 ("vm_object_vndeallocate: not a vnode object")); 432 KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp")); 433 #ifdef INVARIANTS 434 if (object->ref_count == 0) { 435 vprint("vm_object_vndeallocate", vp); 436 panic("vm_object_vndeallocate: bad object reference count"); 437 } 438 #endif 439 440 object->ref_count--; 441 if (object->ref_count == 0) { 442 mp_fixme("Unlocked vflag access."); 443 vp->v_vflag &= ~VV_TEXT; 444 } 445 VM_OBJECT_UNLOCK(object); 446 /* 447 * vrele may need a vop lock 448 */ 449 vrele(vp); 450 } 451 452 /* 453 * vm_object_deallocate: 454 * 455 * Release a reference to the specified object, 456 * gained either through a vm_object_allocate 457 * or a vm_object_reference call. When all references 458 * are gone, storage associated with this object 459 * may be relinquished. 460 * 461 * No object may be locked. 462 */ 463 void 464 vm_object_deallocate(vm_object_t object) 465 { 466 vm_object_t temp; 467 468 while (object != NULL) { 469 int vfslocked; 470 471 vfslocked = 0; 472 restart: 473 VM_OBJECT_LOCK(object); 474 if (object->type == OBJT_VNODE) { 475 struct vnode *vp = (struct vnode *) object->handle; 476 477 /* 478 * Conditionally acquire Giant for a vnode-backed 479 * object. We have to be careful since the type of 480 * a vnode object can change while the object is 481 * unlocked. 482 */ 483 if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) { 484 vfslocked = 1; 485 if (!mtx_trylock(&Giant)) { 486 VM_OBJECT_UNLOCK(object); 487 mtx_lock(&Giant); 488 goto restart; 489 } 490 } 491 vm_object_vndeallocate(object); 492 VFS_UNLOCK_GIANT(vfslocked); 493 return; 494 } else 495 /* 496 * This is to handle the case that the object 497 * changed type while we dropped its lock to 498 * obtain Giant. 499 */ 500 VFS_UNLOCK_GIANT(vfslocked); 501 502 KASSERT(object->ref_count != 0, 503 ("vm_object_deallocate: object deallocated too many times: %d", object->type)); 504 505 /* 506 * If the reference count goes to 0 we start calling 507 * vm_object_terminate() on the object chain. 508 * A ref count of 1 may be a special case depending on the 509 * shadow count being 0 or 1. 510 */ 511 object->ref_count--; 512 if (object->ref_count > 1) { 513 VM_OBJECT_UNLOCK(object); 514 return; 515 } else if (object->ref_count == 1) { 516 if (object->shadow_count == 0 && 517 object->handle == NULL && 518 (object->type == OBJT_DEFAULT || 519 object->type == OBJT_SWAP)) { 520 vm_object_set_flag(object, OBJ_ONEMAPPING); 521 } else if ((object->shadow_count == 1) && 522 (object->handle == NULL) && 523 (object->type == OBJT_DEFAULT || 524 object->type == OBJT_SWAP)) { 525 vm_object_t robject; 526 527 robject = LIST_FIRST(&object->shadow_head); 528 KASSERT(robject != NULL, 529 ("vm_object_deallocate: ref_count: %d, shadow_count: %d", 530 object->ref_count, 531 object->shadow_count)); 532 if (!VM_OBJECT_TRYLOCK(robject)) { 533 /* 534 * Avoid a potential deadlock. 535 */ 536 object->ref_count++; 537 VM_OBJECT_UNLOCK(object); 538 /* 539 * More likely than not the thread 540 * holding robject's lock has lower 541 * priority than the current thread. 542 * Let the lower priority thread run. 543 */ 544 pause("vmo_de", 1); 545 continue; 546 } 547 /* 548 * Collapse object into its shadow unless its 549 * shadow is dead. In that case, object will 550 * be deallocated by the thread that is 551 * deallocating its shadow. 552 */ 553 if ((robject->flags & OBJ_DEAD) == 0 && 554 (robject->handle == NULL) && 555 (robject->type == OBJT_DEFAULT || 556 robject->type == OBJT_SWAP)) { 557 558 robject->ref_count++; 559 retry: 560 if (robject->paging_in_progress) { 561 VM_OBJECT_UNLOCK(object); 562 vm_object_pip_wait(robject, 563 "objde1"); 564 temp = robject->backing_object; 565 if (object == temp) { 566 VM_OBJECT_LOCK(object); 567 goto retry; 568 } 569 } else if (object->paging_in_progress) { 570 VM_OBJECT_UNLOCK(robject); 571 object->flags |= OBJ_PIPWNT; 572 msleep(object, 573 VM_OBJECT_MTX(object), 574 PDROP | PVM, "objde2", 0); 575 VM_OBJECT_LOCK(robject); 576 temp = robject->backing_object; 577 if (object == temp) { 578 VM_OBJECT_LOCK(object); 579 goto retry; 580 } 581 } else 582 VM_OBJECT_UNLOCK(object); 583 584 if (robject->ref_count == 1) { 585 robject->ref_count--; 586 object = robject; 587 goto doterm; 588 } 589 object = robject; 590 vm_object_collapse(object); 591 VM_OBJECT_UNLOCK(object); 592 continue; 593 } 594 VM_OBJECT_UNLOCK(robject); 595 } 596 VM_OBJECT_UNLOCK(object); 597 return; 598 } 599 doterm: 600 temp = object->backing_object; 601 if (temp != NULL) { 602 VM_OBJECT_LOCK(temp); 603 LIST_REMOVE(object, shadow_list); 604 temp->shadow_count--; 605 VM_OBJECT_UNLOCK(temp); 606 object->backing_object = NULL; 607 } 608 /* 609 * Don't double-terminate, we could be in a termination 610 * recursion due to the terminate having to sync data 611 * to disk. 612 */ 613 if ((object->flags & OBJ_DEAD) == 0) 614 vm_object_terminate(object); 615 else 616 VM_OBJECT_UNLOCK(object); 617 object = temp; 618 } 619 } 620 621 /* 622 * vm_object_destroy removes the object from the global object list 623 * and frees the space for the object. 624 */ 625 void 626 vm_object_destroy(vm_object_t object) 627 { 628 629 /* 630 * Remove the object from the global object list. 631 */ 632 mtx_lock(&vm_object_list_mtx); 633 TAILQ_REMOVE(&vm_object_list, object, object_list); 634 mtx_unlock(&vm_object_list_mtx); 635 636 /* 637 * Release the allocation charge. 638 */ 639 if (object->cred != NULL) { 640 KASSERT(object->type == OBJT_DEFAULT || 641 object->type == OBJT_SWAP, 642 ("vm_object_terminate: non-swap obj %p has cred", 643 object)); 644 swap_release_by_cred(object->charge, object->cred); 645 object->charge = 0; 646 crfree(object->cred); 647 object->cred = NULL; 648 } 649 650 /* 651 * Free the space for the object. 652 */ 653 uma_zfree(obj_zone, object); 654 } 655 656 /* 657 * vm_object_terminate actually destroys the specified object, freeing 658 * up all previously used resources. 659 * 660 * The object must be locked. 661 * This routine may block. 662 */ 663 void 664 vm_object_terminate(vm_object_t object) 665 { 666 vm_page_t p, p_next; 667 668 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 669 670 /* 671 * Make sure no one uses us. 672 */ 673 vm_object_set_flag(object, OBJ_DEAD); 674 675 /* 676 * wait for the pageout daemon to be done with the object 677 */ 678 vm_object_pip_wait(object, "objtrm"); 679 680 KASSERT(!object->paging_in_progress, 681 ("vm_object_terminate: pageout in progress")); 682 683 /* 684 * Clean and free the pages, as appropriate. All references to the 685 * object are gone, so we don't need to lock it. 686 */ 687 if (object->type == OBJT_VNODE) { 688 struct vnode *vp = (struct vnode *)object->handle; 689 690 /* 691 * Clean pages and flush buffers. 692 */ 693 vm_object_page_clean(object, 0, 0, OBJPC_SYNC); 694 VM_OBJECT_UNLOCK(object); 695 696 vinvalbuf(vp, V_SAVE, 0, 0); 697 698 VM_OBJECT_LOCK(object); 699 } 700 701 KASSERT(object->ref_count == 0, 702 ("vm_object_terminate: object with references, ref_count=%d", 703 object->ref_count)); 704 705 /* 706 * Free any remaining pageable pages. This also removes them from the 707 * paging queues. However, don't free wired pages, just remove them 708 * from the object. Rather than incrementally removing each page from 709 * the object, the page and object are reset to any empty state. 710 */ 711 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { 712 KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0, 713 ("vm_object_terminate: freeing busy page %p", p)); 714 vm_page_lock(p); 715 /* 716 * Optimize the page's removal from the object by resetting 717 * its "object" field. Specifically, if the page is not 718 * wired, then the effect of this assignment is that 719 * vm_page_free()'s call to vm_page_remove() will return 720 * immediately without modifying the page or the object. 721 */ 722 p->object = NULL; 723 if (p->wire_count == 0) { 724 vm_page_free(p); 725 PCPU_INC(cnt.v_pfree); 726 } 727 vm_page_unlock(p); 728 } 729 /* 730 * If the object contained any pages, then reset it to an empty state. 731 * None of the object's fields, including "resident_page_count", were 732 * modified by the preceding loop. 733 */ 734 if (object->resident_page_count != 0) { 735 object->root = NULL; 736 TAILQ_INIT(&object->memq); 737 object->resident_page_count = 0; 738 if (object->type == OBJT_VNODE) 739 vdrop(object->handle); 740 } 741 742 #if VM_NRESERVLEVEL > 0 743 if (__predict_false(!LIST_EMPTY(&object->rvq))) 744 vm_reserv_break_all(object); 745 #endif 746 if (__predict_false(object->cache != NULL)) 747 vm_page_cache_free(object, 0, 0); 748 749 /* 750 * Let the pager know object is dead. 751 */ 752 vm_pager_deallocate(object); 753 VM_OBJECT_UNLOCK(object); 754 755 vm_object_destroy(object); 756 } 757 758 /* 759 * Make the page read-only so that we can clear the object flags. However, if 760 * this is a nosync mmap then the object is likely to stay dirty so do not 761 * mess with the page and do not clear the object flags. Returns TRUE if the 762 * page should be flushed, and FALSE otherwise. 763 */ 764 static boolean_t 765 vm_object_page_remove_write(vm_page_t p, int flags, int *clearobjflags) 766 { 767 768 /* 769 * If we have been asked to skip nosync pages and this is a 770 * nosync page, skip it. Note that the object flags were not 771 * cleared in this case so we do not have to set them. 772 */ 773 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) { 774 *clearobjflags = 0; 775 return (FALSE); 776 } else { 777 pmap_remove_write(p); 778 return (p->dirty != 0); 779 } 780 } 781 782 /* 783 * vm_object_page_clean 784 * 785 * Clean all dirty pages in the specified range of object. Leaves page 786 * on whatever queue it is currently on. If NOSYNC is set then do not 787 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC), 788 * leaving the object dirty. 789 * 790 * When stuffing pages asynchronously, allow clustering. XXX we need a 791 * synchronous clustering mode implementation. 792 * 793 * Odd semantics: if start == end, we clean everything. 794 * 795 * The object must be locked. 796 */ 797 void 798 vm_object_page_clean(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 799 int flags) 800 { 801 vm_page_t np, p; 802 vm_pindex_t pi, tend; 803 int clearobjflags, curgeneration, n, pagerflags; 804 805 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED); 806 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 807 KASSERT(object->type == OBJT_VNODE, ("Not a vnode object")); 808 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 || 809 object->resident_page_count == 0) 810 return; 811 812 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? 813 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; 814 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; 815 816 tend = (end == 0) ? object->size : end; 817 clearobjflags = start == 0 && tend == object->size; 818 819 rescan: 820 curgeneration = object->generation; 821 822 for (p = vm_page_find_least(object, start); p != NULL; p = np) { 823 pi = p->pindex; 824 if (pi >= tend) 825 break; 826 np = TAILQ_NEXT(p, listq); 827 if (p->valid == 0) 828 continue; 829 if (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) { 830 if (object->generation != curgeneration) 831 goto rescan; 832 np = vm_page_find_least(object, pi); 833 continue; 834 } 835 if (!vm_object_page_remove_write(p, flags, &clearobjflags)) 836 continue; 837 838 n = vm_object_page_collect_flush(object, p, pagerflags, 839 flags, &clearobjflags); 840 if (object->generation != curgeneration) 841 goto rescan; 842 np = vm_page_find_least(object, pi + n); 843 } 844 #if 0 845 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); 846 #endif 847 848 if (clearobjflags) 849 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY); 850 } 851 852 static int 853 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, 854 int flags, int *clearobjflags) 855 { 856 vm_page_t ma[vm_pageout_page_count], p_first, tp; 857 int count, i, mreq, runlen; 858 859 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED); 860 vm_page_lock_assert(p, MA_NOTOWNED); 861 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 862 863 count = 1; 864 mreq = 0; 865 866 for (tp = p; count < vm_pageout_page_count; count++) { 867 tp = vm_page_next(tp); 868 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0) 869 break; 870 if (!vm_object_page_remove_write(tp, flags, clearobjflags)) 871 break; 872 } 873 874 for (p_first = p; count < vm_pageout_page_count; count++) { 875 tp = vm_page_prev(p_first); 876 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0) 877 break; 878 if (!vm_object_page_remove_write(tp, flags, clearobjflags)) 879 break; 880 p_first = tp; 881 mreq++; 882 } 883 884 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) 885 ma[i] = tp; 886 887 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen); 888 return (runlen); 889 } 890 891 /* 892 * Note that there is absolutely no sense in writing out 893 * anonymous objects, so we track down the vnode object 894 * to write out. 895 * We invalidate (remove) all pages from the address space 896 * for semantic correctness. 897 * 898 * Note: certain anonymous maps, such as MAP_NOSYNC maps, 899 * may start out with a NULL object. 900 */ 901 void 902 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, 903 boolean_t syncio, boolean_t invalidate) 904 { 905 vm_object_t backing_object; 906 struct vnode *vp; 907 struct mount *mp; 908 int flags; 909 910 if (object == NULL) 911 return; 912 VM_OBJECT_LOCK(object); 913 while ((backing_object = object->backing_object) != NULL) { 914 VM_OBJECT_LOCK(backing_object); 915 offset += object->backing_object_offset; 916 VM_OBJECT_UNLOCK(object); 917 object = backing_object; 918 if (object->size < OFF_TO_IDX(offset + size)) 919 size = IDX_TO_OFF(object->size) - offset; 920 } 921 /* 922 * Flush pages if writing is allowed, invalidate them 923 * if invalidation requested. Pages undergoing I/O 924 * will be ignored by vm_object_page_remove(). 925 * 926 * We cannot lock the vnode and then wait for paging 927 * to complete without deadlocking against vm_fault. 928 * Instead we simply call vm_object_page_remove() and 929 * allow it to block internally on a page-by-page 930 * basis when it encounters pages undergoing async 931 * I/O. 932 */ 933 if (object->type == OBJT_VNODE && 934 (object->flags & OBJ_MIGHTBEDIRTY) != 0) { 935 int vfslocked; 936 vp = object->handle; 937 VM_OBJECT_UNLOCK(object); 938 (void) vn_start_write(vp, &mp, V_WAIT); 939 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 940 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 941 flags = (syncio || invalidate) ? OBJPC_SYNC : 0; 942 flags |= invalidate ? OBJPC_INVAL : 0; 943 VM_OBJECT_LOCK(object); 944 vm_object_page_clean(object, 945 OFF_TO_IDX(offset), 946 OFF_TO_IDX(offset + size + PAGE_MASK), 947 flags); 948 VM_OBJECT_UNLOCK(object); 949 VOP_UNLOCK(vp, 0); 950 VFS_UNLOCK_GIANT(vfslocked); 951 vn_finished_write(mp); 952 VM_OBJECT_LOCK(object); 953 } 954 if ((object->type == OBJT_VNODE || 955 object->type == OBJT_DEVICE) && invalidate) { 956 boolean_t purge; 957 purge = old_msync || (object->type == OBJT_DEVICE); 958 vm_object_page_remove(object, 959 OFF_TO_IDX(offset), 960 OFF_TO_IDX(offset + size + PAGE_MASK), 961 purge ? FALSE : TRUE); 962 } 963 VM_OBJECT_UNLOCK(object); 964 } 965 966 /* 967 * vm_object_madvise: 968 * 969 * Implements the madvise function at the object/page level. 970 * 971 * MADV_WILLNEED (any object) 972 * 973 * Activate the specified pages if they are resident. 974 * 975 * MADV_DONTNEED (any object) 976 * 977 * Deactivate the specified pages if they are resident. 978 * 979 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, 980 * OBJ_ONEMAPPING only) 981 * 982 * Deactivate and clean the specified pages if they are 983 * resident. This permits the process to reuse the pages 984 * without faulting or the kernel to reclaim the pages 985 * without I/O. 986 */ 987 void 988 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise) 989 { 990 vm_pindex_t end, tpindex; 991 vm_object_t backing_object, tobject; 992 vm_page_t m; 993 994 if (object == NULL) 995 return; 996 VM_OBJECT_LOCK(object); 997 end = pindex + count; 998 /* 999 * Locate and adjust resident pages 1000 */ 1001 for (; pindex < end; pindex += 1) { 1002 relookup: 1003 tobject = object; 1004 tpindex = pindex; 1005 shadowlookup: 1006 /* 1007 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages 1008 * and those pages must be OBJ_ONEMAPPING. 1009 */ 1010 if (advise == MADV_FREE) { 1011 if ((tobject->type != OBJT_DEFAULT && 1012 tobject->type != OBJT_SWAP) || 1013 (tobject->flags & OBJ_ONEMAPPING) == 0) { 1014 goto unlock_tobject; 1015 } 1016 } else if (tobject->type == OBJT_PHYS) 1017 goto unlock_tobject; 1018 m = vm_page_lookup(tobject, tpindex); 1019 if (m == NULL && advise == MADV_WILLNEED) { 1020 /* 1021 * If the page is cached, reactivate it. 1022 */ 1023 m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED | 1024 VM_ALLOC_NOBUSY); 1025 } 1026 if (m == NULL) { 1027 /* 1028 * There may be swap even if there is no backing page 1029 */ 1030 if (advise == MADV_FREE && tobject->type == OBJT_SWAP) 1031 swap_pager_freespace(tobject, tpindex, 1); 1032 /* 1033 * next object 1034 */ 1035 backing_object = tobject->backing_object; 1036 if (backing_object == NULL) 1037 goto unlock_tobject; 1038 VM_OBJECT_LOCK(backing_object); 1039 tpindex += OFF_TO_IDX(tobject->backing_object_offset); 1040 if (tobject != object) 1041 VM_OBJECT_UNLOCK(tobject); 1042 tobject = backing_object; 1043 goto shadowlookup; 1044 } else if (m->valid != VM_PAGE_BITS_ALL) 1045 goto unlock_tobject; 1046 /* 1047 * If the page is not in a normal state, skip it. 1048 */ 1049 vm_page_lock(m); 1050 if (m->hold_count != 0 || m->wire_count != 0) { 1051 vm_page_unlock(m); 1052 goto unlock_tobject; 1053 } 1054 KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0, 1055 ("vm_object_madvise: page %p is not managed", m)); 1056 if ((m->oflags & VPO_BUSY) || m->busy) { 1057 if (advise == MADV_WILLNEED) { 1058 /* 1059 * Reference the page before unlocking and 1060 * sleeping so that the page daemon is less 1061 * likely to reclaim it. 1062 */ 1063 vm_page_lock_queues(); 1064 vm_page_flag_set(m, PG_REFERENCED); 1065 vm_page_unlock_queues(); 1066 } 1067 vm_page_unlock(m); 1068 if (object != tobject) 1069 VM_OBJECT_UNLOCK(object); 1070 m->oflags |= VPO_WANTED; 1071 msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo", 1072 0); 1073 VM_OBJECT_LOCK(object); 1074 goto relookup; 1075 } 1076 if (advise == MADV_WILLNEED) { 1077 vm_page_activate(m); 1078 } else if (advise == MADV_DONTNEED) { 1079 vm_page_dontneed(m); 1080 } else if (advise == MADV_FREE) { 1081 /* 1082 * Mark the page clean. This will allow the page 1083 * to be freed up by the system. However, such pages 1084 * are often reused quickly by malloc()/free() 1085 * so we do not do anything that would cause 1086 * a page fault if we can help it. 1087 * 1088 * Specifically, we do not try to actually free 1089 * the page now nor do we try to put it in the 1090 * cache (which would cause a page fault on reuse). 1091 * 1092 * But we do make the page is freeable as we 1093 * can without actually taking the step of unmapping 1094 * it. 1095 */ 1096 pmap_clear_modify(m); 1097 m->dirty = 0; 1098 m->act_count = 0; 1099 vm_page_dontneed(m); 1100 } 1101 vm_page_unlock(m); 1102 if (advise == MADV_FREE && tobject->type == OBJT_SWAP) 1103 swap_pager_freespace(tobject, tpindex, 1); 1104 unlock_tobject: 1105 if (tobject != object) 1106 VM_OBJECT_UNLOCK(tobject); 1107 } 1108 VM_OBJECT_UNLOCK(object); 1109 } 1110 1111 /* 1112 * vm_object_shadow: 1113 * 1114 * Create a new object which is backed by the 1115 * specified existing object range. The source 1116 * object reference is deallocated. 1117 * 1118 * The new object and offset into that object 1119 * are returned in the source parameters. 1120 */ 1121 void 1122 vm_object_shadow( 1123 vm_object_t *object, /* IN/OUT */ 1124 vm_ooffset_t *offset, /* IN/OUT */ 1125 vm_size_t length) 1126 { 1127 vm_object_t source; 1128 vm_object_t result; 1129 1130 source = *object; 1131 1132 /* 1133 * Don't create the new object if the old object isn't shared. 1134 */ 1135 if (source != NULL) { 1136 VM_OBJECT_LOCK(source); 1137 if (source->ref_count == 1 && 1138 source->handle == NULL && 1139 (source->type == OBJT_DEFAULT || 1140 source->type == OBJT_SWAP)) { 1141 VM_OBJECT_UNLOCK(source); 1142 return; 1143 } 1144 VM_OBJECT_UNLOCK(source); 1145 } 1146 1147 /* 1148 * Allocate a new object with the given length. 1149 */ 1150 result = vm_object_allocate(OBJT_DEFAULT, length); 1151 1152 /* 1153 * The new object shadows the source object, adding a reference to it. 1154 * Our caller changes his reference to point to the new object, 1155 * removing a reference to the source object. Net result: no change 1156 * of reference count. 1157 * 1158 * Try to optimize the result object's page color when shadowing 1159 * in order to maintain page coloring consistency in the combined 1160 * shadowed object. 1161 */ 1162 result->backing_object = source; 1163 /* 1164 * Store the offset into the source object, and fix up the offset into 1165 * the new object. 1166 */ 1167 result->backing_object_offset = *offset; 1168 if (source != NULL) { 1169 VM_OBJECT_LOCK(source); 1170 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list); 1171 source->shadow_count++; 1172 #if VM_NRESERVLEVEL > 0 1173 result->flags |= source->flags & OBJ_COLORED; 1174 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & 1175 ((1 << (VM_NFREEORDER - 1)) - 1); 1176 #endif 1177 VM_OBJECT_UNLOCK(source); 1178 } 1179 1180 1181 /* 1182 * Return the new things 1183 */ 1184 *offset = 0; 1185 *object = result; 1186 } 1187 1188 /* 1189 * vm_object_split: 1190 * 1191 * Split the pages in a map entry into a new object. This affords 1192 * easier removal of unused pages, and keeps object inheritance from 1193 * being a negative impact on memory usage. 1194 */ 1195 void 1196 vm_object_split(vm_map_entry_t entry) 1197 { 1198 vm_page_t m, m_next; 1199 vm_object_t orig_object, new_object, source; 1200 vm_pindex_t idx, offidxstart; 1201 vm_size_t size; 1202 1203 orig_object = entry->object.vm_object; 1204 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP) 1205 return; 1206 if (orig_object->ref_count <= 1) 1207 return; 1208 VM_OBJECT_UNLOCK(orig_object); 1209 1210 offidxstart = OFF_TO_IDX(entry->offset); 1211 size = atop(entry->end - entry->start); 1212 1213 /* 1214 * If swap_pager_copy() is later called, it will convert new_object 1215 * into a swap object. 1216 */ 1217 new_object = vm_object_allocate(OBJT_DEFAULT, size); 1218 1219 /* 1220 * At this point, the new object is still private, so the order in 1221 * which the original and new objects are locked does not matter. 1222 */ 1223 VM_OBJECT_LOCK(new_object); 1224 VM_OBJECT_LOCK(orig_object); 1225 source = orig_object->backing_object; 1226 if (source != NULL) { 1227 VM_OBJECT_LOCK(source); 1228 if ((source->flags & OBJ_DEAD) != 0) { 1229 VM_OBJECT_UNLOCK(source); 1230 VM_OBJECT_UNLOCK(orig_object); 1231 VM_OBJECT_UNLOCK(new_object); 1232 vm_object_deallocate(new_object); 1233 VM_OBJECT_LOCK(orig_object); 1234 return; 1235 } 1236 LIST_INSERT_HEAD(&source->shadow_head, 1237 new_object, shadow_list); 1238 source->shadow_count++; 1239 vm_object_reference_locked(source); /* for new_object */ 1240 vm_object_clear_flag(source, OBJ_ONEMAPPING); 1241 VM_OBJECT_UNLOCK(source); 1242 new_object->backing_object_offset = 1243 orig_object->backing_object_offset + entry->offset; 1244 new_object->backing_object = source; 1245 } 1246 if (orig_object->cred != NULL) { 1247 new_object->cred = orig_object->cred; 1248 crhold(orig_object->cred); 1249 new_object->charge = ptoa(size); 1250 KASSERT(orig_object->charge >= ptoa(size), 1251 ("orig_object->charge < 0")); 1252 orig_object->charge -= ptoa(size); 1253 } 1254 retry: 1255 m = vm_page_find_least(orig_object, offidxstart); 1256 for (; m != NULL && (idx = m->pindex - offidxstart) < size; 1257 m = m_next) { 1258 m_next = TAILQ_NEXT(m, listq); 1259 1260 /* 1261 * We must wait for pending I/O to complete before we can 1262 * rename the page. 1263 * 1264 * We do not have to VM_PROT_NONE the page as mappings should 1265 * not be changed by this operation. 1266 */ 1267 if ((m->oflags & VPO_BUSY) || m->busy) { 1268 VM_OBJECT_UNLOCK(new_object); 1269 m->oflags |= VPO_WANTED; 1270 msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0); 1271 VM_OBJECT_LOCK(new_object); 1272 goto retry; 1273 } 1274 vm_page_lock(m); 1275 vm_page_rename(m, new_object, idx); 1276 vm_page_unlock(m); 1277 /* page automatically made dirty by rename and cache handled */ 1278 vm_page_busy(m); 1279 } 1280 if (orig_object->type == OBJT_SWAP) { 1281 /* 1282 * swap_pager_copy() can sleep, in which case the orig_object's 1283 * and new_object's locks are released and reacquired. 1284 */ 1285 swap_pager_copy(orig_object, new_object, offidxstart, 0); 1286 1287 /* 1288 * Transfer any cached pages from orig_object to new_object. 1289 */ 1290 if (__predict_false(orig_object->cache != NULL)) 1291 vm_page_cache_transfer(orig_object, offidxstart, 1292 new_object); 1293 } 1294 VM_OBJECT_UNLOCK(orig_object); 1295 TAILQ_FOREACH(m, &new_object->memq, listq) 1296 vm_page_wakeup(m); 1297 VM_OBJECT_UNLOCK(new_object); 1298 entry->object.vm_object = new_object; 1299 entry->offset = 0LL; 1300 vm_object_deallocate(orig_object); 1301 VM_OBJECT_LOCK(new_object); 1302 } 1303 1304 #define OBSC_TEST_ALL_SHADOWED 0x0001 1305 #define OBSC_COLLAPSE_NOWAIT 0x0002 1306 #define OBSC_COLLAPSE_WAIT 0x0004 1307 1308 static int 1309 vm_object_backing_scan(vm_object_t object, int op) 1310 { 1311 int r = 1; 1312 vm_page_t p; 1313 vm_object_t backing_object; 1314 vm_pindex_t backing_offset_index; 1315 1316 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1317 VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED); 1318 1319 backing_object = object->backing_object; 1320 backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 1321 1322 /* 1323 * Initial conditions 1324 */ 1325 if (op & OBSC_TEST_ALL_SHADOWED) { 1326 /* 1327 * We do not want to have to test for the existence of cache 1328 * or swap pages in the backing object. XXX but with the 1329 * new swapper this would be pretty easy to do. 1330 * 1331 * XXX what about anonymous MAP_SHARED memory that hasn't 1332 * been ZFOD faulted yet? If we do not test for this, the 1333 * shadow test may succeed! XXX 1334 */ 1335 if (backing_object->type != OBJT_DEFAULT) { 1336 return (0); 1337 } 1338 } 1339 if (op & OBSC_COLLAPSE_WAIT) { 1340 vm_object_set_flag(backing_object, OBJ_DEAD); 1341 } 1342 1343 /* 1344 * Our scan 1345 */ 1346 p = TAILQ_FIRST(&backing_object->memq); 1347 while (p) { 1348 vm_page_t next = TAILQ_NEXT(p, listq); 1349 vm_pindex_t new_pindex = p->pindex - backing_offset_index; 1350 1351 if (op & OBSC_TEST_ALL_SHADOWED) { 1352 vm_page_t pp; 1353 1354 /* 1355 * Ignore pages outside the parent object's range 1356 * and outside the parent object's mapping of the 1357 * backing object. 1358 * 1359 * note that we do not busy the backing object's 1360 * page. 1361 */ 1362 if ( 1363 p->pindex < backing_offset_index || 1364 new_pindex >= object->size 1365 ) { 1366 p = next; 1367 continue; 1368 } 1369 1370 /* 1371 * See if the parent has the page or if the parent's 1372 * object pager has the page. If the parent has the 1373 * page but the page is not valid, the parent's 1374 * object pager must have the page. 1375 * 1376 * If this fails, the parent does not completely shadow 1377 * the object and we might as well give up now. 1378 */ 1379 1380 pp = vm_page_lookup(object, new_pindex); 1381 if ( 1382 (pp == NULL || pp->valid == 0) && 1383 !vm_pager_has_page(object, new_pindex, NULL, NULL) 1384 ) { 1385 r = 0; 1386 break; 1387 } 1388 } 1389 1390 /* 1391 * Check for busy page 1392 */ 1393 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) { 1394 vm_page_t pp; 1395 1396 if (op & OBSC_COLLAPSE_NOWAIT) { 1397 if ((p->oflags & VPO_BUSY) || 1398 !p->valid || 1399 p->busy) { 1400 p = next; 1401 continue; 1402 } 1403 } else if (op & OBSC_COLLAPSE_WAIT) { 1404 if ((p->oflags & VPO_BUSY) || p->busy) { 1405 VM_OBJECT_UNLOCK(object); 1406 p->oflags |= VPO_WANTED; 1407 msleep(p, VM_OBJECT_MTX(backing_object), 1408 PDROP | PVM, "vmocol", 0); 1409 VM_OBJECT_LOCK(object); 1410 VM_OBJECT_LOCK(backing_object); 1411 /* 1412 * If we slept, anything could have 1413 * happened. Since the object is 1414 * marked dead, the backing offset 1415 * should not have changed so we 1416 * just restart our scan. 1417 */ 1418 p = TAILQ_FIRST(&backing_object->memq); 1419 continue; 1420 } 1421 } 1422 1423 KASSERT( 1424 p->object == backing_object, 1425 ("vm_object_backing_scan: object mismatch") 1426 ); 1427 1428 /* 1429 * Destroy any associated swap 1430 */ 1431 if (backing_object->type == OBJT_SWAP) { 1432 swap_pager_freespace( 1433 backing_object, 1434 p->pindex, 1435 1 1436 ); 1437 } 1438 1439 if ( 1440 p->pindex < backing_offset_index || 1441 new_pindex >= object->size 1442 ) { 1443 /* 1444 * Page is out of the parent object's range, we 1445 * can simply destroy it. 1446 */ 1447 vm_page_lock(p); 1448 KASSERT(!pmap_page_is_mapped(p), 1449 ("freeing mapped page %p", p)); 1450 if (p->wire_count == 0) 1451 vm_page_free(p); 1452 else 1453 vm_page_remove(p); 1454 vm_page_unlock(p); 1455 p = next; 1456 continue; 1457 } 1458 1459 pp = vm_page_lookup(object, new_pindex); 1460 if ( 1461 pp != NULL || 1462 vm_pager_has_page(object, new_pindex, NULL, NULL) 1463 ) { 1464 /* 1465 * page already exists in parent OR swap exists 1466 * for this location in the parent. Destroy 1467 * the original page from the backing object. 1468 * 1469 * Leave the parent's page alone 1470 */ 1471 vm_page_lock(p); 1472 KASSERT(!pmap_page_is_mapped(p), 1473 ("freeing mapped page %p", p)); 1474 if (p->wire_count == 0) 1475 vm_page_free(p); 1476 else 1477 vm_page_remove(p); 1478 vm_page_unlock(p); 1479 p = next; 1480 continue; 1481 } 1482 1483 #if VM_NRESERVLEVEL > 0 1484 /* 1485 * Rename the reservation. 1486 */ 1487 vm_reserv_rename(p, object, backing_object, 1488 backing_offset_index); 1489 #endif 1490 1491 /* 1492 * Page does not exist in parent, rename the 1493 * page from the backing object to the main object. 1494 * 1495 * If the page was mapped to a process, it can remain 1496 * mapped through the rename. 1497 */ 1498 vm_page_lock(p); 1499 vm_page_rename(p, object, new_pindex); 1500 vm_page_unlock(p); 1501 /* page automatically made dirty by rename */ 1502 } 1503 p = next; 1504 } 1505 return (r); 1506 } 1507 1508 1509 /* 1510 * this version of collapse allows the operation to occur earlier and 1511 * when paging_in_progress is true for an object... This is not a complete 1512 * operation, but should plug 99.9% of the rest of the leaks. 1513 */ 1514 static void 1515 vm_object_qcollapse(vm_object_t object) 1516 { 1517 vm_object_t backing_object = object->backing_object; 1518 1519 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1520 VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED); 1521 1522 if (backing_object->ref_count != 1) 1523 return; 1524 1525 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT); 1526 } 1527 1528 /* 1529 * vm_object_collapse: 1530 * 1531 * Collapse an object with the object backing it. 1532 * Pages in the backing object are moved into the 1533 * parent, and the backing object is deallocated. 1534 */ 1535 void 1536 vm_object_collapse(vm_object_t object) 1537 { 1538 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1539 1540 while (TRUE) { 1541 vm_object_t backing_object; 1542 1543 /* 1544 * Verify that the conditions are right for collapse: 1545 * 1546 * The object exists and the backing object exists. 1547 */ 1548 if ((backing_object = object->backing_object) == NULL) 1549 break; 1550 1551 /* 1552 * we check the backing object first, because it is most likely 1553 * not collapsable. 1554 */ 1555 VM_OBJECT_LOCK(backing_object); 1556 if (backing_object->handle != NULL || 1557 (backing_object->type != OBJT_DEFAULT && 1558 backing_object->type != OBJT_SWAP) || 1559 (backing_object->flags & OBJ_DEAD) || 1560 object->handle != NULL || 1561 (object->type != OBJT_DEFAULT && 1562 object->type != OBJT_SWAP) || 1563 (object->flags & OBJ_DEAD)) { 1564 VM_OBJECT_UNLOCK(backing_object); 1565 break; 1566 } 1567 1568 if ( 1569 object->paging_in_progress != 0 || 1570 backing_object->paging_in_progress != 0 1571 ) { 1572 vm_object_qcollapse(object); 1573 VM_OBJECT_UNLOCK(backing_object); 1574 break; 1575 } 1576 /* 1577 * We know that we can either collapse the backing object (if 1578 * the parent is the only reference to it) or (perhaps) have 1579 * the parent bypass the object if the parent happens to shadow 1580 * all the resident pages in the entire backing object. 1581 * 1582 * This is ignoring pager-backed pages such as swap pages. 1583 * vm_object_backing_scan fails the shadowing test in this 1584 * case. 1585 */ 1586 if (backing_object->ref_count == 1) { 1587 /* 1588 * If there is exactly one reference to the backing 1589 * object, we can collapse it into the parent. 1590 */ 1591 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT); 1592 1593 #if VM_NRESERVLEVEL > 0 1594 /* 1595 * Break any reservations from backing_object. 1596 */ 1597 if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) 1598 vm_reserv_break_all(backing_object); 1599 #endif 1600 1601 /* 1602 * Move the pager from backing_object to object. 1603 */ 1604 if (backing_object->type == OBJT_SWAP) { 1605 /* 1606 * swap_pager_copy() can sleep, in which case 1607 * the backing_object's and object's locks are 1608 * released and reacquired. 1609 */ 1610 swap_pager_copy( 1611 backing_object, 1612 object, 1613 OFF_TO_IDX(object->backing_object_offset), TRUE); 1614 1615 /* 1616 * Free any cached pages from backing_object. 1617 */ 1618 if (__predict_false(backing_object->cache != NULL)) 1619 vm_page_cache_free(backing_object, 0, 0); 1620 } 1621 /* 1622 * Object now shadows whatever backing_object did. 1623 * Note that the reference to 1624 * backing_object->backing_object moves from within 1625 * backing_object to within object. 1626 */ 1627 LIST_REMOVE(object, shadow_list); 1628 backing_object->shadow_count--; 1629 if (backing_object->backing_object) { 1630 VM_OBJECT_LOCK(backing_object->backing_object); 1631 LIST_REMOVE(backing_object, shadow_list); 1632 LIST_INSERT_HEAD( 1633 &backing_object->backing_object->shadow_head, 1634 object, shadow_list); 1635 /* 1636 * The shadow_count has not changed. 1637 */ 1638 VM_OBJECT_UNLOCK(backing_object->backing_object); 1639 } 1640 object->backing_object = backing_object->backing_object; 1641 object->backing_object_offset += 1642 backing_object->backing_object_offset; 1643 1644 /* 1645 * Discard backing_object. 1646 * 1647 * Since the backing object has no pages, no pager left, 1648 * and no object references within it, all that is 1649 * necessary is to dispose of it. 1650 */ 1651 KASSERT(backing_object->ref_count == 1, ( 1652 "backing_object %p was somehow re-referenced during collapse!", 1653 backing_object)); 1654 VM_OBJECT_UNLOCK(backing_object); 1655 vm_object_destroy(backing_object); 1656 1657 object_collapses++; 1658 } else { 1659 vm_object_t new_backing_object; 1660 1661 /* 1662 * If we do not entirely shadow the backing object, 1663 * there is nothing we can do so we give up. 1664 */ 1665 if (object->resident_page_count != object->size && 1666 vm_object_backing_scan(object, 1667 OBSC_TEST_ALL_SHADOWED) == 0) { 1668 VM_OBJECT_UNLOCK(backing_object); 1669 break; 1670 } 1671 1672 /* 1673 * Make the parent shadow the next object in the 1674 * chain. Deallocating backing_object will not remove 1675 * it, since its reference count is at least 2. 1676 */ 1677 LIST_REMOVE(object, shadow_list); 1678 backing_object->shadow_count--; 1679 1680 new_backing_object = backing_object->backing_object; 1681 if ((object->backing_object = new_backing_object) != NULL) { 1682 VM_OBJECT_LOCK(new_backing_object); 1683 LIST_INSERT_HEAD( 1684 &new_backing_object->shadow_head, 1685 object, 1686 shadow_list 1687 ); 1688 new_backing_object->shadow_count++; 1689 vm_object_reference_locked(new_backing_object); 1690 VM_OBJECT_UNLOCK(new_backing_object); 1691 object->backing_object_offset += 1692 backing_object->backing_object_offset; 1693 } 1694 1695 /* 1696 * Drop the reference count on backing_object. Since 1697 * its ref_count was at least 2, it will not vanish. 1698 */ 1699 backing_object->ref_count--; 1700 VM_OBJECT_UNLOCK(backing_object); 1701 object_bypasses++; 1702 } 1703 1704 /* 1705 * Try again with this object's new backing object. 1706 */ 1707 } 1708 } 1709 1710 /* 1711 * vm_object_page_remove: 1712 * 1713 * For the given object, either frees or invalidates each of the 1714 * specified pages. In general, a page is freed. However, if a 1715 * page is wired for any reason other than the existence of a 1716 * managed, wired mapping, then it may be invalidated but not 1717 * removed from the object. Pages are specified by the given 1718 * range ["start", "end") and Boolean "clean_only". As a 1719 * special case, if "end" is zero, then the range extends from 1720 * "start" to the end of the object. If "clean_only" is TRUE, 1721 * then only the non-dirty pages within the specified range are 1722 * affected. 1723 * 1724 * In general, this operation should only be performed on objects 1725 * that contain managed pages. There are two exceptions. First, 1726 * it may be performed on the kernel and kmem objects. Second, 1727 * it may be used by msync(..., MS_INVALIDATE) to invalidate 1728 * device-backed pages. In both of these cases, "clean_only" 1729 * must be FALSE. 1730 * 1731 * The object must be locked. 1732 */ 1733 void 1734 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 1735 boolean_t clean_only) 1736 { 1737 vm_page_t p, next; 1738 int wirings; 1739 1740 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1741 if (object->resident_page_count == 0) 1742 goto skipmemq; 1743 1744 /* 1745 * Since physically-backed objects do not use managed pages, we can't 1746 * remove pages from the object (we must instead remove the page 1747 * references, and then destroy the object). 1748 */ 1749 KASSERT(object->type != OBJT_PHYS || object == kernel_object || 1750 object == kmem_object, 1751 ("attempt to remove pages from a physical object")); 1752 1753 vm_object_pip_add(object, 1); 1754 again: 1755 p = vm_page_find_least(object, start); 1756 1757 /* 1758 * Assert: the variable p is either (1) the page with the 1759 * least pindex greater than or equal to the parameter pindex 1760 * or (2) NULL. 1761 */ 1762 for (; 1763 p != NULL && (p->pindex < end || end == 0); 1764 p = next) { 1765 next = TAILQ_NEXT(p, listq); 1766 1767 /* 1768 * If the page is wired for any reason besides the 1769 * existence of managed, wired mappings, then it cannot 1770 * be freed. For example, fictitious pages, which 1771 * represent device memory, are inherently wired and 1772 * cannot be freed. They can, however, be invalidated 1773 * if "clean_only" is FALSE. 1774 */ 1775 vm_page_lock(p); 1776 if ((wirings = p->wire_count) != 0 && 1777 (wirings = pmap_page_wired_mappings(p)) != p->wire_count) { 1778 /* Fictitious pages do not have managed mappings. */ 1779 if ((p->flags & PG_FICTITIOUS) == 0) 1780 pmap_remove_all(p); 1781 /* Account for removal of managed, wired mappings. */ 1782 p->wire_count -= wirings; 1783 if (!clean_only) { 1784 p->valid = 0; 1785 vm_page_undirty(p); 1786 } 1787 vm_page_unlock(p); 1788 continue; 1789 } 1790 if (vm_page_sleep_if_busy(p, TRUE, "vmopar")) 1791 goto again; 1792 KASSERT((p->flags & PG_FICTITIOUS) == 0, 1793 ("vm_object_page_remove: page %p is fictitious", p)); 1794 if (clean_only && p->valid) { 1795 pmap_remove_write(p); 1796 if (p->dirty) { 1797 vm_page_unlock(p); 1798 continue; 1799 } 1800 } 1801 pmap_remove_all(p); 1802 /* Account for removal of managed, wired mappings. */ 1803 if (wirings != 0) 1804 p->wire_count -= wirings; 1805 vm_page_free(p); 1806 vm_page_unlock(p); 1807 } 1808 vm_object_pip_wakeup(object); 1809 skipmemq: 1810 if (__predict_false(object->cache != NULL)) 1811 vm_page_cache_free(object, start, end); 1812 } 1813 1814 /* 1815 * Populate the specified range of the object with valid pages. Returns 1816 * TRUE if the range is successfully populated and FALSE otherwise. 1817 * 1818 * Note: This function should be optimized to pass a larger array of 1819 * pages to vm_pager_get_pages() before it is applied to a non- 1820 * OBJT_DEVICE object. 1821 * 1822 * The object must be locked. 1823 */ 1824 boolean_t 1825 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 1826 { 1827 vm_page_t m, ma[1]; 1828 vm_pindex_t pindex; 1829 int rv; 1830 1831 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1832 for (pindex = start; pindex < end; pindex++) { 1833 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | 1834 VM_ALLOC_RETRY); 1835 if (m->valid != VM_PAGE_BITS_ALL) { 1836 ma[0] = m; 1837 rv = vm_pager_get_pages(object, ma, 1, 0); 1838 m = vm_page_lookup(object, pindex); 1839 if (m == NULL) 1840 break; 1841 if (rv != VM_PAGER_OK) { 1842 vm_page_lock(m); 1843 vm_page_free(m); 1844 vm_page_unlock(m); 1845 break; 1846 } 1847 } 1848 /* 1849 * Keep "m" busy because a subsequent iteration may unlock 1850 * the object. 1851 */ 1852 } 1853 if (pindex > start) { 1854 m = vm_page_lookup(object, start); 1855 while (m != NULL && m->pindex < pindex) { 1856 vm_page_wakeup(m); 1857 m = TAILQ_NEXT(m, listq); 1858 } 1859 } 1860 return (pindex == end); 1861 } 1862 1863 /* 1864 * Routine: vm_object_coalesce 1865 * Function: Coalesces two objects backing up adjoining 1866 * regions of memory into a single object. 1867 * 1868 * returns TRUE if objects were combined. 1869 * 1870 * NOTE: Only works at the moment if the second object is NULL - 1871 * if it's not, which object do we lock first? 1872 * 1873 * Parameters: 1874 * prev_object First object to coalesce 1875 * prev_offset Offset into prev_object 1876 * prev_size Size of reference to prev_object 1877 * next_size Size of reference to the second object 1878 * reserved Indicator that extension region has 1879 * swap accounted for 1880 * 1881 * Conditions: 1882 * The object must *not* be locked. 1883 */ 1884 boolean_t 1885 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, 1886 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) 1887 { 1888 vm_pindex_t next_pindex; 1889 1890 if (prev_object == NULL) 1891 return (TRUE); 1892 VM_OBJECT_LOCK(prev_object); 1893 if (prev_object->type != OBJT_DEFAULT && 1894 prev_object->type != OBJT_SWAP) { 1895 VM_OBJECT_UNLOCK(prev_object); 1896 return (FALSE); 1897 } 1898 1899 /* 1900 * Try to collapse the object first 1901 */ 1902 vm_object_collapse(prev_object); 1903 1904 /* 1905 * Can't coalesce if: . more than one reference . paged out . shadows 1906 * another object . has a copy elsewhere (any of which mean that the 1907 * pages not mapped to prev_entry may be in use anyway) 1908 */ 1909 if (prev_object->backing_object != NULL) { 1910 VM_OBJECT_UNLOCK(prev_object); 1911 return (FALSE); 1912 } 1913 1914 prev_size >>= PAGE_SHIFT; 1915 next_size >>= PAGE_SHIFT; 1916 next_pindex = OFF_TO_IDX(prev_offset) + prev_size; 1917 1918 if ((prev_object->ref_count > 1) && 1919 (prev_object->size != next_pindex)) { 1920 VM_OBJECT_UNLOCK(prev_object); 1921 return (FALSE); 1922 } 1923 1924 /* 1925 * Account for the charge. 1926 */ 1927 if (prev_object->cred != NULL) { 1928 1929 /* 1930 * If prev_object was charged, then this mapping, 1931 * althought not charged now, may become writable 1932 * later. Non-NULL cred in the object would prevent 1933 * swap reservation during enabling of the write 1934 * access, so reserve swap now. Failed reservation 1935 * cause allocation of the separate object for the map 1936 * entry, and swap reservation for this entry is 1937 * managed in appropriate time. 1938 */ 1939 if (!reserved && !swap_reserve_by_cred(ptoa(next_size), 1940 prev_object->cred)) { 1941 return (FALSE); 1942 } 1943 prev_object->charge += ptoa(next_size); 1944 } 1945 1946 /* 1947 * Remove any pages that may still be in the object from a previous 1948 * deallocation. 1949 */ 1950 if (next_pindex < prev_object->size) { 1951 vm_object_page_remove(prev_object, 1952 next_pindex, 1953 next_pindex + next_size, FALSE); 1954 if (prev_object->type == OBJT_SWAP) 1955 swap_pager_freespace(prev_object, 1956 next_pindex, next_size); 1957 #if 0 1958 if (prev_object->cred != NULL) { 1959 KASSERT(prev_object->charge >= 1960 ptoa(prev_object->size - next_pindex), 1961 ("object %p overcharged 1 %jx %jx", prev_object, 1962 (uintmax_t)next_pindex, (uintmax_t)next_size)); 1963 prev_object->charge -= ptoa(prev_object->size - 1964 next_pindex); 1965 } 1966 #endif 1967 } 1968 1969 /* 1970 * Extend the object if necessary. 1971 */ 1972 if (next_pindex + next_size > prev_object->size) 1973 prev_object->size = next_pindex + next_size; 1974 1975 VM_OBJECT_UNLOCK(prev_object); 1976 return (TRUE); 1977 } 1978 1979 void 1980 vm_object_set_writeable_dirty(vm_object_t object) 1981 { 1982 1983 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1984 if (object->type != OBJT_VNODE) 1985 return; 1986 object->generation++; 1987 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) 1988 return; 1989 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY); 1990 } 1991 1992 #include "opt_ddb.h" 1993 #ifdef DDB 1994 #include <sys/kernel.h> 1995 1996 #include <sys/cons.h> 1997 1998 #include <ddb/ddb.h> 1999 2000 static int 2001 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) 2002 { 2003 vm_map_t tmpm; 2004 vm_map_entry_t tmpe; 2005 vm_object_t obj; 2006 int entcount; 2007 2008 if (map == 0) 2009 return 0; 2010 2011 if (entry == 0) { 2012 tmpe = map->header.next; 2013 entcount = map->nentries; 2014 while (entcount-- && (tmpe != &map->header)) { 2015 if (_vm_object_in_map(map, object, tmpe)) { 2016 return 1; 2017 } 2018 tmpe = tmpe->next; 2019 } 2020 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 2021 tmpm = entry->object.sub_map; 2022 tmpe = tmpm->header.next; 2023 entcount = tmpm->nentries; 2024 while (entcount-- && tmpe != &tmpm->header) { 2025 if (_vm_object_in_map(tmpm, object, tmpe)) { 2026 return 1; 2027 } 2028 tmpe = tmpe->next; 2029 } 2030 } else if ((obj = entry->object.vm_object) != NULL) { 2031 for (; obj; obj = obj->backing_object) 2032 if (obj == object) { 2033 return 1; 2034 } 2035 } 2036 return 0; 2037 } 2038 2039 static int 2040 vm_object_in_map(vm_object_t object) 2041 { 2042 struct proc *p; 2043 2044 /* sx_slock(&allproc_lock); */ 2045 FOREACH_PROC_IN_SYSTEM(p) { 2046 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 2047 continue; 2048 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 2049 /* sx_sunlock(&allproc_lock); */ 2050 return 1; 2051 } 2052 } 2053 /* sx_sunlock(&allproc_lock); */ 2054 if (_vm_object_in_map(kernel_map, object, 0)) 2055 return 1; 2056 if (_vm_object_in_map(kmem_map, object, 0)) 2057 return 1; 2058 if (_vm_object_in_map(pager_map, object, 0)) 2059 return 1; 2060 if (_vm_object_in_map(buffer_map, object, 0)) 2061 return 1; 2062 return 0; 2063 } 2064 2065 DB_SHOW_COMMAND(vmochk, vm_object_check) 2066 { 2067 vm_object_t object; 2068 2069 /* 2070 * make sure that internal objs are in a map somewhere 2071 * and none have zero ref counts. 2072 */ 2073 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2074 if (object->handle == NULL && 2075 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { 2076 if (object->ref_count == 0) { 2077 db_printf("vmochk: internal obj has zero ref count: %ld\n", 2078 (long)object->size); 2079 } 2080 if (!vm_object_in_map(object)) { 2081 db_printf( 2082 "vmochk: internal obj is not in a map: " 2083 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 2084 object->ref_count, (u_long)object->size, 2085 (u_long)object->size, 2086 (void *)object->backing_object); 2087 } 2088 } 2089 } 2090 } 2091 2092 /* 2093 * vm_object_print: [ debug ] 2094 */ 2095 DB_SHOW_COMMAND(object, vm_object_print_static) 2096 { 2097 /* XXX convert args. */ 2098 vm_object_t object = (vm_object_t)addr; 2099 boolean_t full = have_addr; 2100 2101 vm_page_t p; 2102 2103 /* XXX count is an (unused) arg. Avoid shadowing it. */ 2104 #define count was_count 2105 2106 int count; 2107 2108 if (object == NULL) 2109 return; 2110 2111 db_iprintf( 2112 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", 2113 object, (int)object->type, (uintmax_t)object->size, 2114 object->resident_page_count, object->ref_count, object->flags, 2115 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); 2116 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", 2117 object->shadow_count, 2118 object->backing_object ? object->backing_object->ref_count : 0, 2119 object->backing_object, (uintmax_t)object->backing_object_offset); 2120 2121 if (!full) 2122 return; 2123 2124 db_indent += 2; 2125 count = 0; 2126 TAILQ_FOREACH(p, &object->memq, listq) { 2127 if (count == 0) 2128 db_iprintf("memory:="); 2129 else if (count == 6) { 2130 db_printf("\n"); 2131 db_iprintf(" ..."); 2132 count = 0; 2133 } else 2134 db_printf(","); 2135 count++; 2136 2137 db_printf("(off=0x%jx,page=0x%jx)", 2138 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); 2139 } 2140 if (count != 0) 2141 db_printf("\n"); 2142 db_indent -= 2; 2143 } 2144 2145 /* XXX. */ 2146 #undef count 2147 2148 /* XXX need this non-static entry for calling from vm_map_print. */ 2149 void 2150 vm_object_print( 2151 /* db_expr_t */ long addr, 2152 boolean_t have_addr, 2153 /* db_expr_t */ long count, 2154 char *modif) 2155 { 2156 vm_object_print_static(addr, have_addr, count, modif); 2157 } 2158 2159 DB_SHOW_COMMAND(vmopag, vm_object_print_pages) 2160 { 2161 vm_object_t object; 2162 vm_pindex_t fidx; 2163 vm_paddr_t pa; 2164 vm_page_t m, prev_m; 2165 int rcount, nl, c; 2166 2167 nl = 0; 2168 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2169 db_printf("new object: %p\n", (void *)object); 2170 if (nl > 18) { 2171 c = cngetc(); 2172 if (c != ' ') 2173 return; 2174 nl = 0; 2175 } 2176 nl++; 2177 rcount = 0; 2178 fidx = 0; 2179 pa = -1; 2180 TAILQ_FOREACH(m, &object->memq, listq) { 2181 if (m->pindex > 128) 2182 break; 2183 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && 2184 prev_m->pindex + 1 != m->pindex) { 2185 if (rcount) { 2186 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2187 (long)fidx, rcount, (long)pa); 2188 if (nl > 18) { 2189 c = cngetc(); 2190 if (c != ' ') 2191 return; 2192 nl = 0; 2193 } 2194 nl++; 2195 rcount = 0; 2196 } 2197 } 2198 if (rcount && 2199 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 2200 ++rcount; 2201 continue; 2202 } 2203 if (rcount) { 2204 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2205 (long)fidx, rcount, (long)pa); 2206 if (nl > 18) { 2207 c = cngetc(); 2208 if (c != ' ') 2209 return; 2210 nl = 0; 2211 } 2212 nl++; 2213 } 2214 fidx = m->pindex; 2215 pa = VM_PAGE_TO_PHYS(m); 2216 rcount = 1; 2217 } 2218 if (rcount) { 2219 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2220 (long)fidx, rcount, (long)pa); 2221 if (nl > 18) { 2222 c = cngetc(); 2223 if (c != ' ') 2224 return; 2225 nl = 0; 2226 } 2227 nl++; 2228 } 2229 } 2230 } 2231 #endif /* DDB */ 2232