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 if (object->ref_count > 1) { 441 object->ref_count--; 442 VM_OBJECT_UNLOCK(object); 443 /* vrele may need the vnode lock. */ 444 vrele(vp); 445 } else { 446 vhold(vp); 447 VM_OBJECT_UNLOCK(object); 448 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 449 vdrop(vp); 450 VM_OBJECT_LOCK(object); 451 object->ref_count--; 452 if (object->type == OBJT_DEAD) { 453 VM_OBJECT_UNLOCK(object); 454 VOP_UNLOCK(vp, 0); 455 } else { 456 if (object->ref_count == 0) 457 vp->v_vflag &= ~VV_TEXT; 458 VM_OBJECT_UNLOCK(object); 459 vput(vp); 460 } 461 } 462 } 463 464 /* 465 * vm_object_deallocate: 466 * 467 * Release a reference to the specified object, 468 * gained either through a vm_object_allocate 469 * or a vm_object_reference call. When all references 470 * are gone, storage associated with this object 471 * may be relinquished. 472 * 473 * No object may be locked. 474 */ 475 void 476 vm_object_deallocate(vm_object_t object) 477 { 478 vm_object_t temp; 479 480 while (object != NULL) { 481 int vfslocked; 482 483 vfslocked = 0; 484 restart: 485 VM_OBJECT_LOCK(object); 486 if (object->type == OBJT_VNODE) { 487 struct vnode *vp = (struct vnode *) object->handle; 488 489 /* 490 * Conditionally acquire Giant for a vnode-backed 491 * object. We have to be careful since the type of 492 * a vnode object can change while the object is 493 * unlocked. 494 */ 495 if (VFS_NEEDSGIANT(vp->v_mount) && !vfslocked) { 496 vfslocked = 1; 497 if (!mtx_trylock(&Giant)) { 498 VM_OBJECT_UNLOCK(object); 499 mtx_lock(&Giant); 500 goto restart; 501 } 502 } 503 vm_object_vndeallocate(object); 504 VFS_UNLOCK_GIANT(vfslocked); 505 return; 506 } else 507 /* 508 * This is to handle the case that the object 509 * changed type while we dropped its lock to 510 * obtain Giant. 511 */ 512 VFS_UNLOCK_GIANT(vfslocked); 513 514 KASSERT(object->ref_count != 0, 515 ("vm_object_deallocate: object deallocated too many times: %d", object->type)); 516 517 /* 518 * If the reference count goes to 0 we start calling 519 * vm_object_terminate() on the object chain. 520 * A ref count of 1 may be a special case depending on the 521 * shadow count being 0 or 1. 522 */ 523 object->ref_count--; 524 if (object->ref_count > 1) { 525 VM_OBJECT_UNLOCK(object); 526 return; 527 } else if (object->ref_count == 1) { 528 if (object->shadow_count == 0 && 529 object->handle == NULL && 530 (object->type == OBJT_DEFAULT || 531 object->type == OBJT_SWAP)) { 532 vm_object_set_flag(object, OBJ_ONEMAPPING); 533 } else if ((object->shadow_count == 1) && 534 (object->handle == NULL) && 535 (object->type == OBJT_DEFAULT || 536 object->type == OBJT_SWAP)) { 537 vm_object_t robject; 538 539 robject = LIST_FIRST(&object->shadow_head); 540 KASSERT(robject != NULL, 541 ("vm_object_deallocate: ref_count: %d, shadow_count: %d", 542 object->ref_count, 543 object->shadow_count)); 544 if (!VM_OBJECT_TRYLOCK(robject)) { 545 /* 546 * Avoid a potential deadlock. 547 */ 548 object->ref_count++; 549 VM_OBJECT_UNLOCK(object); 550 /* 551 * More likely than not the thread 552 * holding robject's lock has lower 553 * priority than the current thread. 554 * Let the lower priority thread run. 555 */ 556 pause("vmo_de", 1); 557 continue; 558 } 559 /* 560 * Collapse object into its shadow unless its 561 * shadow is dead. In that case, object will 562 * be deallocated by the thread that is 563 * deallocating its shadow. 564 */ 565 if ((robject->flags & OBJ_DEAD) == 0 && 566 (robject->handle == NULL) && 567 (robject->type == OBJT_DEFAULT || 568 robject->type == OBJT_SWAP)) { 569 570 robject->ref_count++; 571 retry: 572 if (robject->paging_in_progress) { 573 VM_OBJECT_UNLOCK(object); 574 vm_object_pip_wait(robject, 575 "objde1"); 576 temp = robject->backing_object; 577 if (object == temp) { 578 VM_OBJECT_LOCK(object); 579 goto retry; 580 } 581 } else if (object->paging_in_progress) { 582 VM_OBJECT_UNLOCK(robject); 583 object->flags |= OBJ_PIPWNT; 584 msleep(object, 585 VM_OBJECT_MTX(object), 586 PDROP | PVM, "objde2", 0); 587 VM_OBJECT_LOCK(robject); 588 temp = robject->backing_object; 589 if (object == temp) { 590 VM_OBJECT_LOCK(object); 591 goto retry; 592 } 593 } else 594 VM_OBJECT_UNLOCK(object); 595 596 if (robject->ref_count == 1) { 597 robject->ref_count--; 598 object = robject; 599 goto doterm; 600 } 601 object = robject; 602 vm_object_collapse(object); 603 VM_OBJECT_UNLOCK(object); 604 continue; 605 } 606 VM_OBJECT_UNLOCK(robject); 607 } 608 VM_OBJECT_UNLOCK(object); 609 return; 610 } 611 doterm: 612 temp = object->backing_object; 613 if (temp != NULL) { 614 VM_OBJECT_LOCK(temp); 615 LIST_REMOVE(object, shadow_list); 616 temp->shadow_count--; 617 VM_OBJECT_UNLOCK(temp); 618 object->backing_object = NULL; 619 } 620 /* 621 * Don't double-terminate, we could be in a termination 622 * recursion due to the terminate having to sync data 623 * to disk. 624 */ 625 if ((object->flags & OBJ_DEAD) == 0) 626 vm_object_terminate(object); 627 else 628 VM_OBJECT_UNLOCK(object); 629 object = temp; 630 } 631 } 632 633 /* 634 * vm_object_destroy removes the object from the global object list 635 * and frees the space for the object. 636 */ 637 void 638 vm_object_destroy(vm_object_t object) 639 { 640 641 /* 642 * Remove the object from the global object list. 643 */ 644 mtx_lock(&vm_object_list_mtx); 645 TAILQ_REMOVE(&vm_object_list, object, object_list); 646 mtx_unlock(&vm_object_list_mtx); 647 648 /* 649 * Release the allocation charge. 650 */ 651 if (object->cred != NULL) { 652 KASSERT(object->type == OBJT_DEFAULT || 653 object->type == OBJT_SWAP, 654 ("vm_object_terminate: non-swap obj %p has cred", 655 object)); 656 swap_release_by_cred(object->charge, object->cred); 657 object->charge = 0; 658 crfree(object->cred); 659 object->cred = NULL; 660 } 661 662 /* 663 * Free the space for the object. 664 */ 665 uma_zfree(obj_zone, object); 666 } 667 668 /* 669 * vm_object_terminate actually destroys the specified object, freeing 670 * up all previously used resources. 671 * 672 * The object must be locked. 673 * This routine may block. 674 */ 675 void 676 vm_object_terminate(vm_object_t object) 677 { 678 vm_page_t p, p_next; 679 680 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 681 682 /* 683 * Make sure no one uses us. 684 */ 685 vm_object_set_flag(object, OBJ_DEAD); 686 687 /* 688 * wait for the pageout daemon to be done with the object 689 */ 690 vm_object_pip_wait(object, "objtrm"); 691 692 KASSERT(!object->paging_in_progress, 693 ("vm_object_terminate: pageout in progress")); 694 695 /* 696 * Clean and free the pages, as appropriate. All references to the 697 * object are gone, so we don't need to lock it. 698 */ 699 if (object->type == OBJT_VNODE) { 700 struct vnode *vp = (struct vnode *)object->handle; 701 702 /* 703 * Clean pages and flush buffers. 704 */ 705 vm_object_page_clean(object, 0, 0, OBJPC_SYNC); 706 VM_OBJECT_UNLOCK(object); 707 708 vinvalbuf(vp, V_SAVE, 0, 0); 709 710 VM_OBJECT_LOCK(object); 711 } 712 713 KASSERT(object->ref_count == 0, 714 ("vm_object_terminate: object with references, ref_count=%d", 715 object->ref_count)); 716 717 /* 718 * Free any remaining pageable pages. This also removes them from the 719 * paging queues. However, don't free wired pages, just remove them 720 * from the object. Rather than incrementally removing each page from 721 * the object, the page and object are reset to any empty state. 722 */ 723 TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { 724 KASSERT(!p->busy && (p->oflags & VPO_BUSY) == 0, 725 ("vm_object_terminate: freeing busy page %p", p)); 726 vm_page_lock(p); 727 /* 728 * Optimize the page's removal from the object by resetting 729 * its "object" field. Specifically, if the page is not 730 * wired, then the effect of this assignment is that 731 * vm_page_free()'s call to vm_page_remove() will return 732 * immediately without modifying the page or the object. 733 */ 734 p->object = NULL; 735 if (p->wire_count == 0) { 736 vm_page_free(p); 737 PCPU_INC(cnt.v_pfree); 738 } 739 vm_page_unlock(p); 740 } 741 /* 742 * If the object contained any pages, then reset it to an empty state. 743 * None of the object's fields, including "resident_page_count", were 744 * modified by the preceding loop. 745 */ 746 if (object->resident_page_count != 0) { 747 object->root = NULL; 748 TAILQ_INIT(&object->memq); 749 object->resident_page_count = 0; 750 if (object->type == OBJT_VNODE) 751 vdrop(object->handle); 752 } 753 754 #if VM_NRESERVLEVEL > 0 755 if (__predict_false(!LIST_EMPTY(&object->rvq))) 756 vm_reserv_break_all(object); 757 #endif 758 if (__predict_false(object->cache != NULL)) 759 vm_page_cache_free(object, 0, 0); 760 761 /* 762 * Let the pager know object is dead. 763 */ 764 vm_pager_deallocate(object); 765 VM_OBJECT_UNLOCK(object); 766 767 vm_object_destroy(object); 768 } 769 770 /* 771 * Make the page read-only so that we can clear the object flags. However, if 772 * this is a nosync mmap then the object is likely to stay dirty so do not 773 * mess with the page and do not clear the object flags. Returns TRUE if the 774 * page should be flushed, and FALSE otherwise. 775 */ 776 static boolean_t 777 vm_object_page_remove_write(vm_page_t p, int flags, int *clearobjflags) 778 { 779 780 /* 781 * If we have been asked to skip nosync pages and this is a 782 * nosync page, skip it. Note that the object flags were not 783 * cleared in this case so we do not have to set them. 784 */ 785 if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) { 786 *clearobjflags = 0; 787 return (FALSE); 788 } else { 789 pmap_remove_write(p); 790 return (p->dirty != 0); 791 } 792 } 793 794 /* 795 * vm_object_page_clean 796 * 797 * Clean all dirty pages in the specified range of object. Leaves page 798 * on whatever queue it is currently on. If NOSYNC is set then do not 799 * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC), 800 * leaving the object dirty. 801 * 802 * When stuffing pages asynchronously, allow clustering. XXX we need a 803 * synchronous clustering mode implementation. 804 * 805 * Odd semantics: if start == end, we clean everything. 806 * 807 * The object must be locked. 808 */ 809 void 810 vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, 811 int flags) 812 { 813 vm_page_t np, p; 814 vm_pindex_t pi, tend, tstart; 815 int clearobjflags, curgeneration, n, pagerflags; 816 817 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED); 818 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 819 KASSERT(object->type == OBJT_VNODE, ("Not a vnode object")); 820 if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 || 821 object->resident_page_count == 0) 822 return; 823 824 pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? 825 VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; 826 pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; 827 828 tstart = OFF_TO_IDX(start); 829 tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); 830 clearobjflags = tstart == 0 && tend >= object->size; 831 832 rescan: 833 curgeneration = object->generation; 834 835 for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { 836 pi = p->pindex; 837 if (pi >= tend) 838 break; 839 np = TAILQ_NEXT(p, listq); 840 if (p->valid == 0) 841 continue; 842 if (vm_page_sleep_if_busy(p, TRUE, "vpcwai")) { 843 if (object->generation != curgeneration) 844 goto rescan; 845 np = vm_page_find_least(object, pi); 846 continue; 847 } 848 if (!vm_object_page_remove_write(p, flags, &clearobjflags)) 849 continue; 850 851 n = vm_object_page_collect_flush(object, p, pagerflags, 852 flags, &clearobjflags); 853 if (object->generation != curgeneration) 854 goto rescan; 855 np = vm_page_find_least(object, pi + n); 856 } 857 #if 0 858 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); 859 #endif 860 861 if (clearobjflags) 862 vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY); 863 } 864 865 static int 866 vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, 867 int flags, int *clearobjflags) 868 { 869 vm_page_t ma[vm_pageout_page_count], p_first, tp; 870 int count, i, mreq, runlen; 871 872 mtx_assert(&vm_page_queue_mtx, MA_NOTOWNED); 873 vm_page_lock_assert(p, MA_NOTOWNED); 874 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 875 876 count = 1; 877 mreq = 0; 878 879 for (tp = p; count < vm_pageout_page_count; count++) { 880 tp = vm_page_next(tp); 881 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0) 882 break; 883 if (!vm_object_page_remove_write(tp, flags, clearobjflags)) 884 break; 885 } 886 887 for (p_first = p; count < vm_pageout_page_count; count++) { 888 tp = vm_page_prev(p_first); 889 if (tp == NULL || tp->busy != 0 || (tp->oflags & VPO_BUSY) != 0) 890 break; 891 if (!vm_object_page_remove_write(tp, flags, clearobjflags)) 892 break; 893 p_first = tp; 894 mreq++; 895 } 896 897 for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) 898 ma[i] = tp; 899 900 vm_pageout_flush(ma, count, pagerflags, mreq, &runlen); 901 return (runlen); 902 } 903 904 /* 905 * Note that there is absolutely no sense in writing out 906 * anonymous objects, so we track down the vnode object 907 * to write out. 908 * We invalidate (remove) all pages from the address space 909 * for semantic correctness. 910 * 911 * Note: certain anonymous maps, such as MAP_NOSYNC maps, 912 * may start out with a NULL object. 913 */ 914 void 915 vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, 916 boolean_t syncio, boolean_t invalidate) 917 { 918 vm_object_t backing_object; 919 struct vnode *vp; 920 struct mount *mp; 921 int flags; 922 923 if (object == NULL) 924 return; 925 VM_OBJECT_LOCK(object); 926 while ((backing_object = object->backing_object) != NULL) { 927 VM_OBJECT_LOCK(backing_object); 928 offset += object->backing_object_offset; 929 VM_OBJECT_UNLOCK(object); 930 object = backing_object; 931 if (object->size < OFF_TO_IDX(offset + size)) 932 size = IDX_TO_OFF(object->size) - offset; 933 } 934 /* 935 * Flush pages if writing is allowed, invalidate them 936 * if invalidation requested. Pages undergoing I/O 937 * will be ignored by vm_object_page_remove(). 938 * 939 * We cannot lock the vnode and then wait for paging 940 * to complete without deadlocking against vm_fault. 941 * Instead we simply call vm_object_page_remove() and 942 * allow it to block internally on a page-by-page 943 * basis when it encounters pages undergoing async 944 * I/O. 945 */ 946 if (object->type == OBJT_VNODE && 947 (object->flags & OBJ_MIGHTBEDIRTY) != 0) { 948 int vfslocked; 949 vp = object->handle; 950 VM_OBJECT_UNLOCK(object); 951 (void) vn_start_write(vp, &mp, V_WAIT); 952 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 953 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 954 flags = (syncio || invalidate) ? OBJPC_SYNC : 0; 955 flags |= invalidate ? OBJPC_INVAL : 0; 956 VM_OBJECT_LOCK(object); 957 vm_object_page_clean(object, offset, offset + size, flags); 958 VM_OBJECT_UNLOCK(object); 959 VOP_UNLOCK(vp, 0); 960 VFS_UNLOCK_GIANT(vfslocked); 961 vn_finished_write(mp); 962 VM_OBJECT_LOCK(object); 963 } 964 if ((object->type == OBJT_VNODE || 965 object->type == OBJT_DEVICE) && invalidate) { 966 boolean_t purge; 967 purge = old_msync || (object->type == OBJT_DEVICE); 968 vm_object_page_remove(object, 969 OFF_TO_IDX(offset), 970 OFF_TO_IDX(offset + size + PAGE_MASK), 971 purge ? FALSE : TRUE); 972 } 973 VM_OBJECT_UNLOCK(object); 974 } 975 976 /* 977 * vm_object_madvise: 978 * 979 * Implements the madvise function at the object/page level. 980 * 981 * MADV_WILLNEED (any object) 982 * 983 * Activate the specified pages if they are resident. 984 * 985 * MADV_DONTNEED (any object) 986 * 987 * Deactivate the specified pages if they are resident. 988 * 989 * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, 990 * OBJ_ONEMAPPING only) 991 * 992 * Deactivate and clean the specified pages if they are 993 * resident. This permits the process to reuse the pages 994 * without faulting or the kernel to reclaim the pages 995 * without I/O. 996 */ 997 void 998 vm_object_madvise(vm_object_t object, vm_pindex_t pindex, int count, int advise) 999 { 1000 vm_pindex_t end, tpindex; 1001 vm_object_t backing_object, tobject; 1002 vm_page_t m; 1003 1004 if (object == NULL) 1005 return; 1006 VM_OBJECT_LOCK(object); 1007 end = pindex + count; 1008 /* 1009 * Locate and adjust resident pages 1010 */ 1011 for (; pindex < end; pindex += 1) { 1012 relookup: 1013 tobject = object; 1014 tpindex = pindex; 1015 shadowlookup: 1016 /* 1017 * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages 1018 * and those pages must be OBJ_ONEMAPPING. 1019 */ 1020 if (advise == MADV_FREE) { 1021 if ((tobject->type != OBJT_DEFAULT && 1022 tobject->type != OBJT_SWAP) || 1023 (tobject->flags & OBJ_ONEMAPPING) == 0) { 1024 goto unlock_tobject; 1025 } 1026 } else if (tobject->type == OBJT_PHYS) 1027 goto unlock_tobject; 1028 m = vm_page_lookup(tobject, tpindex); 1029 if (m == NULL && advise == MADV_WILLNEED) { 1030 /* 1031 * If the page is cached, reactivate it. 1032 */ 1033 m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED | 1034 VM_ALLOC_NOBUSY); 1035 } 1036 if (m == NULL) { 1037 /* 1038 * There may be swap even if there is no backing page 1039 */ 1040 if (advise == MADV_FREE && tobject->type == OBJT_SWAP) 1041 swap_pager_freespace(tobject, tpindex, 1); 1042 /* 1043 * next object 1044 */ 1045 backing_object = tobject->backing_object; 1046 if (backing_object == NULL) 1047 goto unlock_tobject; 1048 VM_OBJECT_LOCK(backing_object); 1049 tpindex += OFF_TO_IDX(tobject->backing_object_offset); 1050 if (tobject != object) 1051 VM_OBJECT_UNLOCK(tobject); 1052 tobject = backing_object; 1053 goto shadowlookup; 1054 } else if (m->valid != VM_PAGE_BITS_ALL) 1055 goto unlock_tobject; 1056 /* 1057 * If the page is not in a normal state, skip it. 1058 */ 1059 vm_page_lock(m); 1060 if (m->hold_count != 0 || m->wire_count != 0) { 1061 vm_page_unlock(m); 1062 goto unlock_tobject; 1063 } 1064 KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0, 1065 ("vm_object_madvise: page %p is not managed", m)); 1066 if ((m->oflags & VPO_BUSY) || m->busy) { 1067 if (advise == MADV_WILLNEED) { 1068 /* 1069 * Reference the page before unlocking and 1070 * sleeping so that the page daemon is less 1071 * likely to reclaim it. 1072 */ 1073 vm_page_lock_queues(); 1074 vm_page_flag_set(m, PG_REFERENCED); 1075 vm_page_unlock_queues(); 1076 } 1077 vm_page_unlock(m); 1078 if (object != tobject) 1079 VM_OBJECT_UNLOCK(object); 1080 m->oflags |= VPO_WANTED; 1081 msleep(m, VM_OBJECT_MTX(tobject), PDROP | PVM, "madvpo", 1082 0); 1083 VM_OBJECT_LOCK(object); 1084 goto relookup; 1085 } 1086 if (advise == MADV_WILLNEED) { 1087 vm_page_activate(m); 1088 } else if (advise == MADV_DONTNEED) { 1089 vm_page_dontneed(m); 1090 } else if (advise == MADV_FREE) { 1091 /* 1092 * Mark the page clean. This will allow the page 1093 * to be freed up by the system. However, such pages 1094 * are often reused quickly by malloc()/free() 1095 * so we do not do anything that would cause 1096 * a page fault if we can help it. 1097 * 1098 * Specifically, we do not try to actually free 1099 * the page now nor do we try to put it in the 1100 * cache (which would cause a page fault on reuse). 1101 * 1102 * But we do make the page is freeable as we 1103 * can without actually taking the step of unmapping 1104 * it. 1105 */ 1106 pmap_clear_modify(m); 1107 m->dirty = 0; 1108 m->act_count = 0; 1109 vm_page_dontneed(m); 1110 } 1111 vm_page_unlock(m); 1112 if (advise == MADV_FREE && tobject->type == OBJT_SWAP) 1113 swap_pager_freespace(tobject, tpindex, 1); 1114 unlock_tobject: 1115 if (tobject != object) 1116 VM_OBJECT_UNLOCK(tobject); 1117 } 1118 VM_OBJECT_UNLOCK(object); 1119 } 1120 1121 /* 1122 * vm_object_shadow: 1123 * 1124 * Create a new object which is backed by the 1125 * specified existing object range. The source 1126 * object reference is deallocated. 1127 * 1128 * The new object and offset into that object 1129 * are returned in the source parameters. 1130 */ 1131 void 1132 vm_object_shadow( 1133 vm_object_t *object, /* IN/OUT */ 1134 vm_ooffset_t *offset, /* IN/OUT */ 1135 vm_size_t length) 1136 { 1137 vm_object_t source; 1138 vm_object_t result; 1139 1140 source = *object; 1141 1142 /* 1143 * Don't create the new object if the old object isn't shared. 1144 */ 1145 if (source != NULL) { 1146 VM_OBJECT_LOCK(source); 1147 if (source->ref_count == 1 && 1148 source->handle == NULL && 1149 (source->type == OBJT_DEFAULT || 1150 source->type == OBJT_SWAP)) { 1151 VM_OBJECT_UNLOCK(source); 1152 return; 1153 } 1154 VM_OBJECT_UNLOCK(source); 1155 } 1156 1157 /* 1158 * Allocate a new object with the given length. 1159 */ 1160 result = vm_object_allocate(OBJT_DEFAULT, atop(length)); 1161 1162 /* 1163 * The new object shadows the source object, adding a reference to it. 1164 * Our caller changes his reference to point to the new object, 1165 * removing a reference to the source object. Net result: no change 1166 * of reference count. 1167 * 1168 * Try to optimize the result object's page color when shadowing 1169 * in order to maintain page coloring consistency in the combined 1170 * shadowed object. 1171 */ 1172 result->backing_object = source; 1173 /* 1174 * Store the offset into the source object, and fix up the offset into 1175 * the new object. 1176 */ 1177 result->backing_object_offset = *offset; 1178 if (source != NULL) { 1179 VM_OBJECT_LOCK(source); 1180 LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list); 1181 source->shadow_count++; 1182 #if VM_NRESERVLEVEL > 0 1183 result->flags |= source->flags & OBJ_COLORED; 1184 result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & 1185 ((1 << (VM_NFREEORDER - 1)) - 1); 1186 #endif 1187 VM_OBJECT_UNLOCK(source); 1188 } 1189 1190 1191 /* 1192 * Return the new things 1193 */ 1194 *offset = 0; 1195 *object = result; 1196 } 1197 1198 /* 1199 * vm_object_split: 1200 * 1201 * Split the pages in a map entry into a new object. This affords 1202 * easier removal of unused pages, and keeps object inheritance from 1203 * being a negative impact on memory usage. 1204 */ 1205 void 1206 vm_object_split(vm_map_entry_t entry) 1207 { 1208 vm_page_t m, m_next; 1209 vm_object_t orig_object, new_object, source; 1210 vm_pindex_t idx, offidxstart; 1211 vm_size_t size; 1212 1213 orig_object = entry->object.vm_object; 1214 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP) 1215 return; 1216 if (orig_object->ref_count <= 1) 1217 return; 1218 VM_OBJECT_UNLOCK(orig_object); 1219 1220 offidxstart = OFF_TO_IDX(entry->offset); 1221 size = atop(entry->end - entry->start); 1222 1223 /* 1224 * If swap_pager_copy() is later called, it will convert new_object 1225 * into a swap object. 1226 */ 1227 new_object = vm_object_allocate(OBJT_DEFAULT, size); 1228 1229 /* 1230 * At this point, the new object is still private, so the order in 1231 * which the original and new objects are locked does not matter. 1232 */ 1233 VM_OBJECT_LOCK(new_object); 1234 VM_OBJECT_LOCK(orig_object); 1235 source = orig_object->backing_object; 1236 if (source != NULL) { 1237 VM_OBJECT_LOCK(source); 1238 if ((source->flags & OBJ_DEAD) != 0) { 1239 VM_OBJECT_UNLOCK(source); 1240 VM_OBJECT_UNLOCK(orig_object); 1241 VM_OBJECT_UNLOCK(new_object); 1242 vm_object_deallocate(new_object); 1243 VM_OBJECT_LOCK(orig_object); 1244 return; 1245 } 1246 LIST_INSERT_HEAD(&source->shadow_head, 1247 new_object, shadow_list); 1248 source->shadow_count++; 1249 vm_object_reference_locked(source); /* for new_object */ 1250 vm_object_clear_flag(source, OBJ_ONEMAPPING); 1251 VM_OBJECT_UNLOCK(source); 1252 new_object->backing_object_offset = 1253 orig_object->backing_object_offset + entry->offset; 1254 new_object->backing_object = source; 1255 } 1256 if (orig_object->cred != NULL) { 1257 new_object->cred = orig_object->cred; 1258 crhold(orig_object->cred); 1259 new_object->charge = ptoa(size); 1260 KASSERT(orig_object->charge >= ptoa(size), 1261 ("orig_object->charge < 0")); 1262 orig_object->charge -= ptoa(size); 1263 } 1264 retry: 1265 m = vm_page_find_least(orig_object, offidxstart); 1266 for (; m != NULL && (idx = m->pindex - offidxstart) < size; 1267 m = m_next) { 1268 m_next = TAILQ_NEXT(m, listq); 1269 1270 /* 1271 * We must wait for pending I/O to complete before we can 1272 * rename the page. 1273 * 1274 * We do not have to VM_PROT_NONE the page as mappings should 1275 * not be changed by this operation. 1276 */ 1277 if ((m->oflags & VPO_BUSY) || m->busy) { 1278 VM_OBJECT_UNLOCK(new_object); 1279 m->oflags |= VPO_WANTED; 1280 msleep(m, VM_OBJECT_MTX(orig_object), PVM, "spltwt", 0); 1281 VM_OBJECT_LOCK(new_object); 1282 goto retry; 1283 } 1284 vm_page_lock(m); 1285 vm_page_rename(m, new_object, idx); 1286 vm_page_unlock(m); 1287 /* page automatically made dirty by rename and cache handled */ 1288 vm_page_busy(m); 1289 } 1290 if (orig_object->type == OBJT_SWAP) { 1291 /* 1292 * swap_pager_copy() can sleep, in which case the orig_object's 1293 * and new_object's locks are released and reacquired. 1294 */ 1295 swap_pager_copy(orig_object, new_object, offidxstart, 0); 1296 1297 /* 1298 * Transfer any cached pages from orig_object to new_object. 1299 */ 1300 if (__predict_false(orig_object->cache != NULL)) 1301 vm_page_cache_transfer(orig_object, offidxstart, 1302 new_object); 1303 } 1304 VM_OBJECT_UNLOCK(orig_object); 1305 TAILQ_FOREACH(m, &new_object->memq, listq) 1306 vm_page_wakeup(m); 1307 VM_OBJECT_UNLOCK(new_object); 1308 entry->object.vm_object = new_object; 1309 entry->offset = 0LL; 1310 vm_object_deallocate(orig_object); 1311 VM_OBJECT_LOCK(new_object); 1312 } 1313 1314 #define OBSC_TEST_ALL_SHADOWED 0x0001 1315 #define OBSC_COLLAPSE_NOWAIT 0x0002 1316 #define OBSC_COLLAPSE_WAIT 0x0004 1317 1318 static int 1319 vm_object_backing_scan(vm_object_t object, int op) 1320 { 1321 int r = 1; 1322 vm_page_t p; 1323 vm_object_t backing_object; 1324 vm_pindex_t backing_offset_index; 1325 1326 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1327 VM_OBJECT_LOCK_ASSERT(object->backing_object, MA_OWNED); 1328 1329 backing_object = object->backing_object; 1330 backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 1331 1332 /* 1333 * Initial conditions 1334 */ 1335 if (op & OBSC_TEST_ALL_SHADOWED) { 1336 /* 1337 * We do not want to have to test for the existence of cache 1338 * or swap pages in the backing object. XXX but with the 1339 * new swapper this would be pretty easy to do. 1340 * 1341 * XXX what about anonymous MAP_SHARED memory that hasn't 1342 * been ZFOD faulted yet? If we do not test for this, the 1343 * shadow test may succeed! XXX 1344 */ 1345 if (backing_object->type != OBJT_DEFAULT) { 1346 return (0); 1347 } 1348 } 1349 if (op & OBSC_COLLAPSE_WAIT) { 1350 vm_object_set_flag(backing_object, OBJ_DEAD); 1351 } 1352 1353 /* 1354 * Our scan 1355 */ 1356 p = TAILQ_FIRST(&backing_object->memq); 1357 while (p) { 1358 vm_page_t next = TAILQ_NEXT(p, listq); 1359 vm_pindex_t new_pindex = p->pindex - backing_offset_index; 1360 1361 if (op & OBSC_TEST_ALL_SHADOWED) { 1362 vm_page_t pp; 1363 1364 /* 1365 * Ignore pages outside the parent object's range 1366 * and outside the parent object's mapping of the 1367 * backing object. 1368 * 1369 * note that we do not busy the backing object's 1370 * page. 1371 */ 1372 if ( 1373 p->pindex < backing_offset_index || 1374 new_pindex >= object->size 1375 ) { 1376 p = next; 1377 continue; 1378 } 1379 1380 /* 1381 * See if the parent has the page or if the parent's 1382 * object pager has the page. If the parent has the 1383 * page but the page is not valid, the parent's 1384 * object pager must have the page. 1385 * 1386 * If this fails, the parent does not completely shadow 1387 * the object and we might as well give up now. 1388 */ 1389 1390 pp = vm_page_lookup(object, new_pindex); 1391 if ( 1392 (pp == NULL || pp->valid == 0) && 1393 !vm_pager_has_page(object, new_pindex, NULL, NULL) 1394 ) { 1395 r = 0; 1396 break; 1397 } 1398 } 1399 1400 /* 1401 * Check for busy page 1402 */ 1403 if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) { 1404 vm_page_t pp; 1405 1406 if (op & OBSC_COLLAPSE_NOWAIT) { 1407 if ((p->oflags & VPO_BUSY) || 1408 !p->valid || 1409 p->busy) { 1410 p = next; 1411 continue; 1412 } 1413 } else if (op & OBSC_COLLAPSE_WAIT) { 1414 if ((p->oflags & VPO_BUSY) || p->busy) { 1415 VM_OBJECT_UNLOCK(object); 1416 p->oflags |= VPO_WANTED; 1417 msleep(p, VM_OBJECT_MTX(backing_object), 1418 PDROP | PVM, "vmocol", 0); 1419 VM_OBJECT_LOCK(object); 1420 VM_OBJECT_LOCK(backing_object); 1421 /* 1422 * If we slept, anything could have 1423 * happened. Since the object is 1424 * marked dead, the backing offset 1425 * should not have changed so we 1426 * just restart our scan. 1427 */ 1428 p = TAILQ_FIRST(&backing_object->memq); 1429 continue; 1430 } 1431 } 1432 1433 KASSERT( 1434 p->object == backing_object, 1435 ("vm_object_backing_scan: object mismatch") 1436 ); 1437 1438 /* 1439 * Destroy any associated swap 1440 */ 1441 if (backing_object->type == OBJT_SWAP) { 1442 swap_pager_freespace( 1443 backing_object, 1444 p->pindex, 1445 1 1446 ); 1447 } 1448 1449 if ( 1450 p->pindex < backing_offset_index || 1451 new_pindex >= object->size 1452 ) { 1453 /* 1454 * Page is out of the parent object's range, we 1455 * can simply destroy it. 1456 */ 1457 vm_page_lock(p); 1458 KASSERT(!pmap_page_is_mapped(p), 1459 ("freeing mapped page %p", p)); 1460 if (p->wire_count == 0) 1461 vm_page_free(p); 1462 else 1463 vm_page_remove(p); 1464 vm_page_unlock(p); 1465 p = next; 1466 continue; 1467 } 1468 1469 pp = vm_page_lookup(object, new_pindex); 1470 if ( 1471 (op & OBSC_COLLAPSE_NOWAIT) != 0 && 1472 (pp != NULL && pp->valid == 0) 1473 ) { 1474 /* 1475 * The page in the parent is not (yet) valid. 1476 * We don't know anything about the state of 1477 * the original page. It might be mapped, 1478 * so we must avoid the next if here. 1479 * 1480 * This is due to a race in vm_fault() where 1481 * we must unbusy the original (backing_obj) 1482 * page before we can (re)lock the parent. 1483 * Hence we can get here. 1484 */ 1485 p = next; 1486 continue; 1487 } 1488 if ( 1489 pp != NULL || 1490 vm_pager_has_page(object, new_pindex, NULL, NULL) 1491 ) { 1492 /* 1493 * page already exists in parent OR swap exists 1494 * for this location in the parent. Destroy 1495 * the original page from the backing object. 1496 * 1497 * Leave the parent's page alone 1498 */ 1499 vm_page_lock(p); 1500 KASSERT(!pmap_page_is_mapped(p), 1501 ("freeing mapped page %p", p)); 1502 if (p->wire_count == 0) 1503 vm_page_free(p); 1504 else 1505 vm_page_remove(p); 1506 vm_page_unlock(p); 1507 p = next; 1508 continue; 1509 } 1510 1511 #if VM_NRESERVLEVEL > 0 1512 /* 1513 * Rename the reservation. 1514 */ 1515 vm_reserv_rename(p, object, backing_object, 1516 backing_offset_index); 1517 #endif 1518 1519 /* 1520 * Page does not exist in parent, rename the 1521 * page from the backing object to the main object. 1522 * 1523 * If the page was mapped to a process, it can remain 1524 * mapped through the rename. 1525 */ 1526 vm_page_lock(p); 1527 vm_page_rename(p, object, new_pindex); 1528 vm_page_unlock(p); 1529 /* page automatically made dirty by rename */ 1530 } 1531 p = next; 1532 } 1533 return (r); 1534 } 1535 1536 1537 /* 1538 * this version of collapse allows the operation to occur earlier and 1539 * when paging_in_progress is true for an object... This is not a complete 1540 * operation, but should plug 99.9% of the rest of the leaks. 1541 */ 1542 static void 1543 vm_object_qcollapse(vm_object_t object) 1544 { 1545 vm_object_t backing_object = object->backing_object; 1546 1547 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1548 VM_OBJECT_LOCK_ASSERT(backing_object, MA_OWNED); 1549 1550 if (backing_object->ref_count != 1) 1551 return; 1552 1553 vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT); 1554 } 1555 1556 /* 1557 * vm_object_collapse: 1558 * 1559 * Collapse an object with the object backing it. 1560 * Pages in the backing object are moved into the 1561 * parent, and the backing object is deallocated. 1562 */ 1563 void 1564 vm_object_collapse(vm_object_t object) 1565 { 1566 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1567 1568 while (TRUE) { 1569 vm_object_t backing_object; 1570 1571 /* 1572 * Verify that the conditions are right for collapse: 1573 * 1574 * The object exists and the backing object exists. 1575 */ 1576 if ((backing_object = object->backing_object) == NULL) 1577 break; 1578 1579 /* 1580 * we check the backing object first, because it is most likely 1581 * not collapsable. 1582 */ 1583 VM_OBJECT_LOCK(backing_object); 1584 if (backing_object->handle != NULL || 1585 (backing_object->type != OBJT_DEFAULT && 1586 backing_object->type != OBJT_SWAP) || 1587 (backing_object->flags & OBJ_DEAD) || 1588 object->handle != NULL || 1589 (object->type != OBJT_DEFAULT && 1590 object->type != OBJT_SWAP) || 1591 (object->flags & OBJ_DEAD)) { 1592 VM_OBJECT_UNLOCK(backing_object); 1593 break; 1594 } 1595 1596 if ( 1597 object->paging_in_progress != 0 || 1598 backing_object->paging_in_progress != 0 1599 ) { 1600 vm_object_qcollapse(object); 1601 VM_OBJECT_UNLOCK(backing_object); 1602 break; 1603 } 1604 /* 1605 * We know that we can either collapse the backing object (if 1606 * the parent is the only reference to it) or (perhaps) have 1607 * the parent bypass the object if the parent happens to shadow 1608 * all the resident pages in the entire backing object. 1609 * 1610 * This is ignoring pager-backed pages such as swap pages. 1611 * vm_object_backing_scan fails the shadowing test in this 1612 * case. 1613 */ 1614 if (backing_object->ref_count == 1) { 1615 /* 1616 * If there is exactly one reference to the backing 1617 * object, we can collapse it into the parent. 1618 */ 1619 vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT); 1620 1621 #if VM_NRESERVLEVEL > 0 1622 /* 1623 * Break any reservations from backing_object. 1624 */ 1625 if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) 1626 vm_reserv_break_all(backing_object); 1627 #endif 1628 1629 /* 1630 * Move the pager from backing_object to object. 1631 */ 1632 if (backing_object->type == OBJT_SWAP) { 1633 /* 1634 * swap_pager_copy() can sleep, in which case 1635 * the backing_object's and object's locks are 1636 * released and reacquired. 1637 */ 1638 swap_pager_copy( 1639 backing_object, 1640 object, 1641 OFF_TO_IDX(object->backing_object_offset), TRUE); 1642 1643 /* 1644 * Free any cached pages from backing_object. 1645 */ 1646 if (__predict_false(backing_object->cache != NULL)) 1647 vm_page_cache_free(backing_object, 0, 0); 1648 } 1649 /* 1650 * Object now shadows whatever backing_object did. 1651 * Note that the reference to 1652 * backing_object->backing_object moves from within 1653 * backing_object to within object. 1654 */ 1655 LIST_REMOVE(object, shadow_list); 1656 backing_object->shadow_count--; 1657 if (backing_object->backing_object) { 1658 VM_OBJECT_LOCK(backing_object->backing_object); 1659 LIST_REMOVE(backing_object, shadow_list); 1660 LIST_INSERT_HEAD( 1661 &backing_object->backing_object->shadow_head, 1662 object, shadow_list); 1663 /* 1664 * The shadow_count has not changed. 1665 */ 1666 VM_OBJECT_UNLOCK(backing_object->backing_object); 1667 } 1668 object->backing_object = backing_object->backing_object; 1669 object->backing_object_offset += 1670 backing_object->backing_object_offset; 1671 1672 /* 1673 * Discard backing_object. 1674 * 1675 * Since the backing object has no pages, no pager left, 1676 * and no object references within it, all that is 1677 * necessary is to dispose of it. 1678 */ 1679 KASSERT(backing_object->ref_count == 1, ( 1680 "backing_object %p was somehow re-referenced during collapse!", 1681 backing_object)); 1682 VM_OBJECT_UNLOCK(backing_object); 1683 vm_object_destroy(backing_object); 1684 1685 object_collapses++; 1686 } else { 1687 vm_object_t new_backing_object; 1688 1689 /* 1690 * If we do not entirely shadow the backing object, 1691 * there is nothing we can do so we give up. 1692 */ 1693 if (object->resident_page_count != object->size && 1694 vm_object_backing_scan(object, 1695 OBSC_TEST_ALL_SHADOWED) == 0) { 1696 VM_OBJECT_UNLOCK(backing_object); 1697 break; 1698 } 1699 1700 /* 1701 * Make the parent shadow the next object in the 1702 * chain. Deallocating backing_object will not remove 1703 * it, since its reference count is at least 2. 1704 */ 1705 LIST_REMOVE(object, shadow_list); 1706 backing_object->shadow_count--; 1707 1708 new_backing_object = backing_object->backing_object; 1709 if ((object->backing_object = new_backing_object) != NULL) { 1710 VM_OBJECT_LOCK(new_backing_object); 1711 LIST_INSERT_HEAD( 1712 &new_backing_object->shadow_head, 1713 object, 1714 shadow_list 1715 ); 1716 new_backing_object->shadow_count++; 1717 vm_object_reference_locked(new_backing_object); 1718 VM_OBJECT_UNLOCK(new_backing_object); 1719 object->backing_object_offset += 1720 backing_object->backing_object_offset; 1721 } 1722 1723 /* 1724 * Drop the reference count on backing_object. Since 1725 * its ref_count was at least 2, it will not vanish. 1726 */ 1727 backing_object->ref_count--; 1728 VM_OBJECT_UNLOCK(backing_object); 1729 object_bypasses++; 1730 } 1731 1732 /* 1733 * Try again with this object's new backing object. 1734 */ 1735 } 1736 } 1737 1738 /* 1739 * vm_object_page_remove: 1740 * 1741 * For the given object, either frees or invalidates each of the 1742 * specified pages. In general, a page is freed. However, if a 1743 * page is wired for any reason other than the existence of a 1744 * managed, wired mapping, then it may be invalidated but not 1745 * removed from the object. Pages are specified by the given 1746 * range ["start", "end") and Boolean "clean_only". As a 1747 * special case, if "end" is zero, then the range extends from 1748 * "start" to the end of the object. If "clean_only" is TRUE, 1749 * then only the non-dirty pages within the specified range are 1750 * affected. 1751 * 1752 * In general, this operation should only be performed on objects 1753 * that contain managed pages. There are two exceptions. First, 1754 * it may be performed on the kernel and kmem objects. Second, 1755 * it may be used by msync(..., MS_INVALIDATE) to invalidate 1756 * device-backed pages. In both of these cases, "clean_only" 1757 * must be FALSE. 1758 * 1759 * The object must be locked. 1760 */ 1761 void 1762 vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 1763 boolean_t clean_only) 1764 { 1765 vm_page_t p, next; 1766 int wirings; 1767 1768 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1769 if (object->resident_page_count == 0) 1770 goto skipmemq; 1771 1772 /* 1773 * Since physically-backed objects do not use managed pages, we can't 1774 * remove pages from the object (we must instead remove the page 1775 * references, and then destroy the object). 1776 */ 1777 KASSERT(object->type != OBJT_PHYS || object == kernel_object || 1778 object == kmem_object, 1779 ("attempt to remove pages from a physical object")); 1780 1781 vm_object_pip_add(object, 1); 1782 again: 1783 p = vm_page_find_least(object, start); 1784 1785 /* 1786 * Assert: the variable p is either (1) the page with the 1787 * least pindex greater than or equal to the parameter pindex 1788 * or (2) NULL. 1789 */ 1790 for (; 1791 p != NULL && (p->pindex < end || end == 0); 1792 p = next) { 1793 next = TAILQ_NEXT(p, listq); 1794 1795 /* 1796 * If the page is wired for any reason besides the 1797 * existence of managed, wired mappings, then it cannot 1798 * be freed. For example, fictitious pages, which 1799 * represent device memory, are inherently wired and 1800 * cannot be freed. They can, however, be invalidated 1801 * if "clean_only" is FALSE. 1802 */ 1803 vm_page_lock(p); 1804 if ((wirings = p->wire_count) != 0 && 1805 (wirings = pmap_page_wired_mappings(p)) != p->wire_count) { 1806 /* Fictitious pages do not have managed mappings. */ 1807 if ((p->flags & PG_FICTITIOUS) == 0) 1808 pmap_remove_all(p); 1809 /* Account for removal of managed, wired mappings. */ 1810 p->wire_count -= wirings; 1811 if (!clean_only) { 1812 p->valid = 0; 1813 vm_page_undirty(p); 1814 } 1815 vm_page_unlock(p); 1816 continue; 1817 } 1818 if (vm_page_sleep_if_busy(p, TRUE, "vmopar")) 1819 goto again; 1820 KASSERT((p->flags & PG_FICTITIOUS) == 0, 1821 ("vm_object_page_remove: page %p is fictitious", p)); 1822 if (clean_only && p->valid) { 1823 pmap_remove_write(p); 1824 if (p->dirty) { 1825 vm_page_unlock(p); 1826 continue; 1827 } 1828 } 1829 pmap_remove_all(p); 1830 /* Account for removal of managed, wired mappings. */ 1831 if (wirings != 0) 1832 p->wire_count -= wirings; 1833 vm_page_free(p); 1834 vm_page_unlock(p); 1835 } 1836 vm_object_pip_wakeup(object); 1837 skipmemq: 1838 if (__predict_false(object->cache != NULL)) 1839 vm_page_cache_free(object, start, end); 1840 } 1841 1842 /* 1843 * Populate the specified range of the object with valid pages. Returns 1844 * TRUE if the range is successfully populated and FALSE otherwise. 1845 * 1846 * Note: This function should be optimized to pass a larger array of 1847 * pages to vm_pager_get_pages() before it is applied to a non- 1848 * OBJT_DEVICE object. 1849 * 1850 * The object must be locked. 1851 */ 1852 boolean_t 1853 vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 1854 { 1855 vm_page_t m, ma[1]; 1856 vm_pindex_t pindex; 1857 int rv; 1858 1859 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 1860 for (pindex = start; pindex < end; pindex++) { 1861 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | 1862 VM_ALLOC_RETRY); 1863 if (m->valid != VM_PAGE_BITS_ALL) { 1864 ma[0] = m; 1865 rv = vm_pager_get_pages(object, ma, 1, 0); 1866 m = vm_page_lookup(object, pindex); 1867 if (m == NULL) 1868 break; 1869 if (rv != VM_PAGER_OK) { 1870 vm_page_lock(m); 1871 vm_page_free(m); 1872 vm_page_unlock(m); 1873 break; 1874 } 1875 } 1876 /* 1877 * Keep "m" busy because a subsequent iteration may unlock 1878 * the object. 1879 */ 1880 } 1881 if (pindex > start) { 1882 m = vm_page_lookup(object, start); 1883 while (m != NULL && m->pindex < pindex) { 1884 vm_page_wakeup(m); 1885 m = TAILQ_NEXT(m, listq); 1886 } 1887 } 1888 return (pindex == end); 1889 } 1890 1891 /* 1892 * Routine: vm_object_coalesce 1893 * Function: Coalesces two objects backing up adjoining 1894 * regions of memory into a single object. 1895 * 1896 * returns TRUE if objects were combined. 1897 * 1898 * NOTE: Only works at the moment if the second object is NULL - 1899 * if it's not, which object do we lock first? 1900 * 1901 * Parameters: 1902 * prev_object First object to coalesce 1903 * prev_offset Offset into prev_object 1904 * prev_size Size of reference to prev_object 1905 * next_size Size of reference to the second object 1906 * reserved Indicator that extension region has 1907 * swap accounted for 1908 * 1909 * Conditions: 1910 * The object must *not* be locked. 1911 */ 1912 boolean_t 1913 vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, 1914 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) 1915 { 1916 vm_pindex_t next_pindex; 1917 1918 if (prev_object == NULL) 1919 return (TRUE); 1920 VM_OBJECT_LOCK(prev_object); 1921 if (prev_object->type != OBJT_DEFAULT && 1922 prev_object->type != OBJT_SWAP) { 1923 VM_OBJECT_UNLOCK(prev_object); 1924 return (FALSE); 1925 } 1926 1927 /* 1928 * Try to collapse the object first 1929 */ 1930 vm_object_collapse(prev_object); 1931 1932 /* 1933 * Can't coalesce if: . more than one reference . paged out . shadows 1934 * another object . has a copy elsewhere (any of which mean that the 1935 * pages not mapped to prev_entry may be in use anyway) 1936 */ 1937 if (prev_object->backing_object != NULL) { 1938 VM_OBJECT_UNLOCK(prev_object); 1939 return (FALSE); 1940 } 1941 1942 prev_size >>= PAGE_SHIFT; 1943 next_size >>= PAGE_SHIFT; 1944 next_pindex = OFF_TO_IDX(prev_offset) + prev_size; 1945 1946 if ((prev_object->ref_count > 1) && 1947 (prev_object->size != next_pindex)) { 1948 VM_OBJECT_UNLOCK(prev_object); 1949 return (FALSE); 1950 } 1951 1952 /* 1953 * Account for the charge. 1954 */ 1955 if (prev_object->cred != NULL) { 1956 1957 /* 1958 * If prev_object was charged, then this mapping, 1959 * althought not charged now, may become writable 1960 * later. Non-NULL cred in the object would prevent 1961 * swap reservation during enabling of the write 1962 * access, so reserve swap now. Failed reservation 1963 * cause allocation of the separate object for the map 1964 * entry, and swap reservation for this entry is 1965 * managed in appropriate time. 1966 */ 1967 if (!reserved && !swap_reserve_by_cred(ptoa(next_size), 1968 prev_object->cred)) { 1969 return (FALSE); 1970 } 1971 prev_object->charge += ptoa(next_size); 1972 } 1973 1974 /* 1975 * Remove any pages that may still be in the object from a previous 1976 * deallocation. 1977 */ 1978 if (next_pindex < prev_object->size) { 1979 vm_object_page_remove(prev_object, 1980 next_pindex, 1981 next_pindex + next_size, FALSE); 1982 if (prev_object->type == OBJT_SWAP) 1983 swap_pager_freespace(prev_object, 1984 next_pindex, next_size); 1985 #if 0 1986 if (prev_object->cred != NULL) { 1987 KASSERT(prev_object->charge >= 1988 ptoa(prev_object->size - next_pindex), 1989 ("object %p overcharged 1 %jx %jx", prev_object, 1990 (uintmax_t)next_pindex, (uintmax_t)next_size)); 1991 prev_object->charge -= ptoa(prev_object->size - 1992 next_pindex); 1993 } 1994 #endif 1995 } 1996 1997 /* 1998 * Extend the object if necessary. 1999 */ 2000 if (next_pindex + next_size > prev_object->size) 2001 prev_object->size = next_pindex + next_size; 2002 2003 VM_OBJECT_UNLOCK(prev_object); 2004 return (TRUE); 2005 } 2006 2007 void 2008 vm_object_set_writeable_dirty(vm_object_t object) 2009 { 2010 2011 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 2012 if (object->type != OBJT_VNODE) 2013 return; 2014 object->generation++; 2015 if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) 2016 return; 2017 vm_object_set_flag(object, OBJ_MIGHTBEDIRTY); 2018 } 2019 2020 #include "opt_ddb.h" 2021 #ifdef DDB 2022 #include <sys/kernel.h> 2023 2024 #include <sys/cons.h> 2025 2026 #include <ddb/ddb.h> 2027 2028 static int 2029 _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) 2030 { 2031 vm_map_t tmpm; 2032 vm_map_entry_t tmpe; 2033 vm_object_t obj; 2034 int entcount; 2035 2036 if (map == 0) 2037 return 0; 2038 2039 if (entry == 0) { 2040 tmpe = map->header.next; 2041 entcount = map->nentries; 2042 while (entcount-- && (tmpe != &map->header)) { 2043 if (_vm_object_in_map(map, object, tmpe)) { 2044 return 1; 2045 } 2046 tmpe = tmpe->next; 2047 } 2048 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 2049 tmpm = entry->object.sub_map; 2050 tmpe = tmpm->header.next; 2051 entcount = tmpm->nentries; 2052 while (entcount-- && tmpe != &tmpm->header) { 2053 if (_vm_object_in_map(tmpm, object, tmpe)) { 2054 return 1; 2055 } 2056 tmpe = tmpe->next; 2057 } 2058 } else if ((obj = entry->object.vm_object) != NULL) { 2059 for (; obj; obj = obj->backing_object) 2060 if (obj == object) { 2061 return 1; 2062 } 2063 } 2064 return 0; 2065 } 2066 2067 static int 2068 vm_object_in_map(vm_object_t object) 2069 { 2070 struct proc *p; 2071 2072 /* sx_slock(&allproc_lock); */ 2073 FOREACH_PROC_IN_SYSTEM(p) { 2074 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 2075 continue; 2076 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 2077 /* sx_sunlock(&allproc_lock); */ 2078 return 1; 2079 } 2080 } 2081 /* sx_sunlock(&allproc_lock); */ 2082 if (_vm_object_in_map(kernel_map, object, 0)) 2083 return 1; 2084 if (_vm_object_in_map(kmem_map, object, 0)) 2085 return 1; 2086 if (_vm_object_in_map(pager_map, object, 0)) 2087 return 1; 2088 if (_vm_object_in_map(buffer_map, object, 0)) 2089 return 1; 2090 return 0; 2091 } 2092 2093 DB_SHOW_COMMAND(vmochk, vm_object_check) 2094 { 2095 vm_object_t object; 2096 2097 /* 2098 * make sure that internal objs are in a map somewhere 2099 * and none have zero ref counts. 2100 */ 2101 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2102 if (object->handle == NULL && 2103 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { 2104 if (object->ref_count == 0) { 2105 db_printf("vmochk: internal obj has zero ref count: %ld\n", 2106 (long)object->size); 2107 } 2108 if (!vm_object_in_map(object)) { 2109 db_printf( 2110 "vmochk: internal obj is not in a map: " 2111 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 2112 object->ref_count, (u_long)object->size, 2113 (u_long)object->size, 2114 (void *)object->backing_object); 2115 } 2116 } 2117 } 2118 } 2119 2120 /* 2121 * vm_object_print: [ debug ] 2122 */ 2123 DB_SHOW_COMMAND(object, vm_object_print_static) 2124 { 2125 /* XXX convert args. */ 2126 vm_object_t object = (vm_object_t)addr; 2127 boolean_t full = have_addr; 2128 2129 vm_page_t p; 2130 2131 /* XXX count is an (unused) arg. Avoid shadowing it. */ 2132 #define count was_count 2133 2134 int count; 2135 2136 if (object == NULL) 2137 return; 2138 2139 db_iprintf( 2140 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", 2141 object, (int)object->type, (uintmax_t)object->size, 2142 object->resident_page_count, object->ref_count, object->flags, 2143 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); 2144 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", 2145 object->shadow_count, 2146 object->backing_object ? object->backing_object->ref_count : 0, 2147 object->backing_object, (uintmax_t)object->backing_object_offset); 2148 2149 if (!full) 2150 return; 2151 2152 db_indent += 2; 2153 count = 0; 2154 TAILQ_FOREACH(p, &object->memq, listq) { 2155 if (count == 0) 2156 db_iprintf("memory:="); 2157 else if (count == 6) { 2158 db_printf("\n"); 2159 db_iprintf(" ..."); 2160 count = 0; 2161 } else 2162 db_printf(","); 2163 count++; 2164 2165 db_printf("(off=0x%jx,page=0x%jx)", 2166 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); 2167 } 2168 if (count != 0) 2169 db_printf("\n"); 2170 db_indent -= 2; 2171 } 2172 2173 /* XXX. */ 2174 #undef count 2175 2176 /* XXX need this non-static entry for calling from vm_map_print. */ 2177 void 2178 vm_object_print( 2179 /* db_expr_t */ long addr, 2180 boolean_t have_addr, 2181 /* db_expr_t */ long count, 2182 char *modif) 2183 { 2184 vm_object_print_static(addr, have_addr, count, modif); 2185 } 2186 2187 DB_SHOW_COMMAND(vmopag, vm_object_print_pages) 2188 { 2189 vm_object_t object; 2190 vm_pindex_t fidx; 2191 vm_paddr_t pa; 2192 vm_page_t m, prev_m; 2193 int rcount, nl, c; 2194 2195 nl = 0; 2196 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2197 db_printf("new object: %p\n", (void *)object); 2198 if (nl > 18) { 2199 c = cngetc(); 2200 if (c != ' ') 2201 return; 2202 nl = 0; 2203 } 2204 nl++; 2205 rcount = 0; 2206 fidx = 0; 2207 pa = -1; 2208 TAILQ_FOREACH(m, &object->memq, listq) { 2209 if (m->pindex > 128) 2210 break; 2211 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && 2212 prev_m->pindex + 1 != m->pindex) { 2213 if (rcount) { 2214 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2215 (long)fidx, rcount, (long)pa); 2216 if (nl > 18) { 2217 c = cngetc(); 2218 if (c != ' ') 2219 return; 2220 nl = 0; 2221 } 2222 nl++; 2223 rcount = 0; 2224 } 2225 } 2226 if (rcount && 2227 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 2228 ++rcount; 2229 continue; 2230 } 2231 if (rcount) { 2232 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2233 (long)fidx, rcount, (long)pa); 2234 if (nl > 18) { 2235 c = cngetc(); 2236 if (c != ' ') 2237 return; 2238 nl = 0; 2239 } 2240 nl++; 2241 } 2242 fidx = m->pindex; 2243 pa = VM_PAGE_TO_PHYS(m); 2244 rcount = 1; 2245 } 2246 if (rcount) { 2247 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2248 (long)fidx, rcount, (long)pa); 2249 if (nl > 18) { 2250 c = cngetc(); 2251 if (c != ' ') 2252 return; 2253 nl = 0; 2254 } 2255 nl++; 2256 } 2257 } 2258 } 2259 #endif /* DDB */ 2260