1193326Sed/*- 2193326Sed * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3193326Sed * 4193326Sed * Copyright (c) 1991, 1993 5193326Sed * The Regents of the University of California. All rights reserved. 6193326Sed * 7193326Sed * This code is derived from software contributed to Berkeley by 8193326Sed * The Mach Operating System project at Carnegie-Mellon University. 9193326Sed * 10193326Sed * Redistribution and use in source and binary forms, with or without 11193326Sed * modification, are permitted provided that the following conditions 12193326Sed * are met: 13193326Sed * 1. Redistributions of source code must retain the above copyright 14193326Sed * notice, this list of conditions and the following disclaimer. 15193326Sed * 2. Redistributions in binary form must reproduce the above copyright 16252723Sdim * notice, this list of conditions and the following disclaimer in the 17252723Sdim * documentation and/or other materials provided with the distribution. 18252723Sdim * 3. Neither the name of the University nor the names of its contributors 19193326Sed * may be used to endorse or promote products derived from this software 20193326Sed * without specific prior written permission. 21252723Sdim * 22226890Sdim * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23252723Sdim * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24193326Sed * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25193326Sed * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26226890Sdim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27226890Sdim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28226890Sdim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29226890Sdim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30226890Sdim * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31226890Sdim * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32226890Sdim * SUCH DAMAGE. 33226890Sdim * 34226890Sdim * 35226890Sdim * Copyright (c) 1987, 1990 Carnegie-Mellon University. 36226890Sdim * All rights reserved. 37226890Sdim * 38226890Sdim * Authors: Avadis Tevanian, Jr., Michael Wayne Young 39226890Sdim * 40226890Sdim * Permission to use, copy, modify and distribute this software and 41226890Sdim * its documentation is hereby granted, provided that both the copyright 42226890Sdim * notice and this permission notice appear in all copies of the 43226890Sdim * software, derivative works or modified versions, and any portions 44226890Sdim * thereof, and that both notices appear in supporting documentation. 45245431Sdim * 46245431Sdim * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47245431Sdim * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48245431Sdim * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49245431Sdim * 50245431Sdim * Carnegie Mellon requests users of this software to return to 51245431Sdim * 52245431Sdim * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53245431Sdim * School of Computer Science 54245431Sdim * Carnegie Mellon University 55245431Sdim * Pittsburgh PA 15213-3890 56245431Sdim * 57245431Sdim * any improvements or extensions that they make and grant Carnegie the 58245431Sdim * rights to redistribute these changes. 59245431Sdim */ 60245431Sdim 61245431Sdim/* 62245431Sdim * Virtual memory object module. 63245431Sdim */ 64245431Sdim 65245431Sdim#include "opt_vm.h" 66245431Sdim 67245431Sdim#include <sys/systm.h> 68245431Sdim#include <sys/blockcount.h> 69245431Sdim#include <sys/cpuset.h> 70245431Sdim#include <sys/jail.h> 71245431Sdim#include <sys/limits.h> 72245431Sdim#include <sys/lock.h> 73245431Sdim#include <sys/mman.h> 74245431Sdim#include <sys/mount.h> 75193326Sed#include <sys/kernel.h> 76193326Sed#include <sys/mutex.h> 77245431Sdim#include <sys/pctrie.h> 78245431Sdim#include <sys/proc.h> 79193326Sed#include <sys/refcount.h> 80226890Sdim#include <sys/sx.h> 81245431Sdim#include <sys/sysctl.h> 82245431Sdim#include <sys/resourcevar.h> 83245431Sdim#include <sys/refcount.h> 84245431Sdim#include <sys/rwlock.h> 85193326Sed#include <sys/user.h> 86193326Sed#include <sys/vnode.h> 87193326Sed#include <sys/vmmeter.h> 88193326Sed 89193326Sed#include <vm/vm.h> 90193326Sed#include <vm/vm_param.h> 91193326Sed#include <vm/pmap.h> 92193326Sed#include <vm/vm_map.h> 93193326Sed#include <vm/vm_object.h> 94198092Srdivacky#include <vm/vm_page.h> 95193326Sed#include <vm/vm_pageout.h> 96193326Sed#include <vm/vm_pager.h> 97193326Sed#include <vm/vm_phys.h> 98193326Sed#include <vm/vm_pagequeue.h> 99193326Sed#include <vm/swap_pager.h> 100193326Sed#include <vm/vm_kern.h> 101193326Sed#include <vm/vm_extern.h> 102193326Sed#include <vm/vm_radix.h> 103193326Sed#include <vm/vm_reserv.h> 104218893Sdim#include <vm/uma.h> 105245431Sdim 106245431Sdimstatic int old_msync; 107193326SedSYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, 108193326Sed "Use old (insecure) msync behavior"); 109194179Sed 110218893Sdimstatic int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, 111245431Sdim int pagerflags, int flags, boolean_t *allclean, 112245431Sdim boolean_t *eio); 113194179Sedstatic boolean_t vm_object_page_remove_write(vm_page_t p, int flags, 114194179Sed boolean_t *allclean); 115193326Sedstatic void vm_object_backing_remove(vm_object_t object); 116193326Sed 117193326Sed/* 118193326Sed * Virtual memory objects maintain the actual data 119193326Sed * associated with allocated virtual memory. A given 120193326Sed * page of memory exists within exactly one object. 121193326Sed * 122193326Sed * An object is only deallocated when all "references" 123193326Sed * are given up. Only one "reference" to a given 124193326Sed * region of an object should be writeable. 125193326Sed * 126193326Sed * Associated with each object is a list of all resident 127193326Sed * memory pages belonging to that object; this list is 128193326Sed * maintained by the "vm_page" module, and locked by the object's 129193326Sed * lock. 130193326Sed * 131193326Sed * Each object also records a "pager" routine which is 132252723Sdim * used to retrieve (and store) pages to the proper backing 133218893Sdim * storage. In addition, objects may be backed by other 134245431Sdim * objects from which they were virtual-copied. 135252723Sdim * 136193326Sed * The only items within the object structure which are 137193326Sed * modified after time of creation are: 138193326Sed * reference count locked by object's lock 139198092Srdivacky * pager routine locked by object's lock 140193326Sed * 141193326Sed */ 142193326Sed 143252723Sdimstruct object_q vm_object_list; 144193326Sedstruct mtx vm_object_list_mtx; /* lock for object list and count */ 145193326Sed 146193326Sedstruct vm_object kernel_object_store; 147193326Sed 148193326Sedstatic SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 149193326Sed "VM object stats"); 150193326Sed 151193326Sedstatic COUNTER_U64_DEFINE_EARLY(object_collapses); 152193326SedSYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, 153193326Sed &object_collapses, 154193326Sed "VM object collapses"); 155193326Sed 156198092Srdivackystatic COUNTER_U64_DEFINE_EARLY(object_bypasses); 157193326SedSYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, 158218893Sdim &object_bypasses, 159245431Sdim "VM object bypasses"); 160263509Sdim 161193326Sedstatic COUNTER_U64_DEFINE_EARLY(object_collapse_waits); 162193326SedSYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapse_waits, CTLFLAG_RD, 163193326Sed &object_collapse_waits, 164193326Sed "Number of sleeps for collapse"); 165193326Sed 166193326Sedstatic uma_zone_t obj_zone; 167193326Sed 168193326Sedstatic int vm_object_zinit(void *mem, int size, int flags); 169193326Sed 170193326Sed#ifdef INVARIANTS 171193326Sedstatic void vm_object_zdtor(void *mem, int size, void *arg); 172193326Sed 173193326Sedstatic void 174193326Sedvm_object_zdtor(void *mem, int size, void *arg) 175193326Sed{ 176193326Sed vm_object_t object; 177193326Sed 178198092Srdivacky object = (vm_object_t)mem; 179193326Sed KASSERT(object->ref_count == 0, 180193326Sed ("object %p ref_count = %d", object, object->ref_count)); 181218893Sdim KASSERT(TAILQ_EMPTY(&object->memq), 182245431Sdim ("object %p has resident pages in its memq", object)); 183263509Sdim KASSERT(vm_radix_is_empty(&object->rtree), 184193326Sed ("object %p has resident pages in its trie", object)); 185193326Sed#if VM_NRESERVLEVEL > 0 186193326Sed KASSERT(LIST_EMPTY(&object->rvq), 187193326Sed ("object %p has reservations", 188198092Srdivacky object)); 189193326Sed#endif 190193326Sed KASSERT(!vm_object_busied(object), 191193326Sed ("object %p busy = %d", object, blockcount_read(&object->busy))); 192218893Sdim KASSERT(object->resident_page_count == 0, 193245431Sdim ("object %p resident_page_count = %d", 194245431Sdim object, object->resident_page_count)); 195245431Sdim KASSERT(atomic_load_int(&object->shadow_count) == 0, 196193326Sed ("object %p shadow_count = %d", 197193326Sed object, atomic_load_int(&object->shadow_count))); 198218893Sdim KASSERT(object->type == OBJT_DEAD, 199208600Srdivacky ("object %p has non-dead type %d", 200245431Sdim object, object->type)); 201252723Sdim KASSERT(object->charge == 0 && object->cred == NULL, 202245431Sdim ("object %p has non-zero charge %ju (%p)", 203245431Sdim object, (uintmax_t)object->charge, object->cred)); 204245431Sdim} 205193326Sed#endif 206245431Sdim 207245431Sdimstatic int 208245431Sdimvm_object_zinit(void *mem, int size, int flags) 209193326Sed{ 210193326Sed vm_object_t object; 211198092Srdivacky 212193326Sed object = (vm_object_t)mem; 213193326Sed rw_init_flags(&object->lock, "vmobject", RW_DUPOK | RW_NEW); 214193326Sed 215193326Sed /* These are true for any object that has been freed */ 216218893Sdim object->type = OBJT_DEAD; 217235633Sdim vm_radix_init(&object->rtree); 218235633Sdim refcount_init(&object->ref_count, 0); 219218893Sdim blockcount_init(&object->paging_in_progress); 220226890Sdim blockcount_init(&object->busy); 221235633Sdim object->resident_page_count = 0; 222235633Sdim atomic_store_int(&object->shadow_count, 0); 223235633Sdim object->flags = OBJ_DEAD; 224198092Srdivacky 225193326Sed mtx_lock(&vm_object_list_mtx); 226193326Sed TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); 227193326Sed mtx_unlock(&vm_object_list_mtx); 228252723Sdim return (0); 229218893Sdim} 230245431Sdim 231252723Sdimstatic void 232218893Sdim_vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags, 233193326Sed vm_object_t object, void *handle) 234218893Sdim{ 235212904Sdim 236218893Sdim TAILQ_INIT(&object->memq); 237252723Sdim LIST_INIT(&object->shadow_head); 238193326Sed 239193326Sed object->type = type; 240193326Sed object->flags = flags; 241193326Sed if ((flags & OBJ_SWAP) != 0) { 242193326Sed pctrie_init(&object->un_pager.swp.swp_blks); 243218893Sdim object->un_pager.swp.writemappings = 0; 244245431Sdim } 245245431Sdim 246245431Sdim /* 247218893Sdim * Ensure that swap_pager_swapoff() iteration over object_list 248193326Sed * sees up to date type and pctrie head if it observed 249235633Sdim * non-dead object. 250235633Sdim */ 251235633Sdim atomic_thread_fence_rel(); 252235633Sdim 253218893Sdim object->pg_color = 0; 254245431Sdim object->size = size; 255245431Sdim object->domain.dr_policy = NULL; 256218893Sdim object->generation = 1; 257218893Sdim object->cleangeneration = 1; 258235633Sdim refcount_init(&object->ref_count, 1); 259235633Sdim object->memattr = VM_MEMATTR_DEFAULT; 260235633Sdim object->cred = NULL; 261235633Sdim object->charge = 0; 262235633Sdim object->handle = handle; 263252723Sdim object->backing_object = NULL; 264235633Sdim object->backing_object_offset = (vm_ooffset_t) 0; 265235633Sdim#if VM_NRESERVLEVEL > 0 266235633Sdim LIST_INIT(&object->rvq); 267245431Sdim#endif 268245431Sdim umtx_shm_object_init(object); 269245431Sdim} 270245431Sdim 271235633Sdim/* 272245431Sdim * vm_object_init: 273245431Sdim * 274245431Sdim * Initialize the VM objects module. 275235633Sdim */ 276235633Sdimvoid 277235633Sdimvm_object_init(void) 278235633Sdim{ 279235633Sdim TAILQ_INIT(&vm_object_list); 280245431Sdim mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); 281245431Sdim 282252723Sdim rw_init(&kernel_object->lock, "kernel vm object"); 283245431Sdim vm_radix_init(&kernel_object->rtree); 284218893Sdim _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - 285218893Sdim VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL); 286218893Sdim#if VM_NRESERVLEVEL > 0 287245431Sdim kernel_object->flags |= OBJ_COLORED; 288245431Sdim kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); 289245431Sdim#endif 290245431Sdim kernel_object->un_pager.phys.ops = &default_phys_pg_ops; 291245431Sdim 292245431Sdim /* 293245431Sdim * The lock portion of struct vm_object must be type stable due 294245431Sdim * to vm_pageout_fallback_object_lock locking a vm object 295245431Sdim * without holding any references to it. 296245431Sdim * 297245431Sdim * paging_in_progress is valid always. Lockless references to 298245431Sdim * the objects may acquire pip and then check OBJ_DEAD. 299245431Sdim */ 300245431Sdim obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, 301245431Sdim#ifdef INVARIANTS 302245431Sdim vm_object_zdtor, 303245431Sdim#else 304245431Sdim NULL, 305245431Sdim#endif 306245431Sdim vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 307245431Sdim 308245431Sdim vm_radix_zinit(); 309245431Sdim} 310245431Sdim 311245431Sdimvoid 312245431Sdimvm_object_clear_flag(vm_object_t object, u_short bits) 313245431Sdim{ 314245431Sdim 315245431Sdim VM_OBJECT_ASSERT_WLOCKED(object); 316245431Sdim object->flags &= ~bits; 317245431Sdim} 318245431Sdim 319245431Sdim/* 320218893Sdim * Sets the default memory attribute for the specified object. Pages 321218893Sdim * that are allocated to this object are by default assigned this memory 322218893Sdim * attribute. 323218893Sdim * 324235633Sdim * Presently, this function must be called before any pages are allocated 325235633Sdim * to the object. In the future, this requirement may be relaxed for 326218893Sdim * "default" and "swap" objects. 327226890Sdim */ 328226890Sdimint 329218893Sdimvm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) 330218893Sdim{ 331218893Sdim 332218893Sdim VM_OBJECT_ASSERT_WLOCKED(object); 333235633Sdim 334235633Sdim if (object->type == OBJT_DEAD) 335245431Sdim return (KERN_INVALID_ARGUMENT); 336193326Sed if (!TAILQ_EMPTY(&object->memq)) 337193326Sed return (KERN_FAILURE); 338218893Sdim 339263509Sdim object->memattr = memattr; 340226890Sdim return (KERN_SUCCESS); 341218893Sdim} 342226890Sdim 343226890Sdimvoid 344218893Sdimvm_object_pip_add(vm_object_t object, short i) 345226890Sdim{ 346235633Sdim 347235633Sdim if (i > 0) 348235633Sdim blockcount_acquire(&object->paging_in_progress, i); 349235633Sdim} 350235633Sdim 351226890Sdimvoid 352226890Sdimvm_object_pip_wakeup(vm_object_t object) 353226890Sdim{ 354226890Sdim 355235633Sdim vm_object_pip_wakeupn(object, 1); 356235633Sdim} 357235633Sdim 358235633Sdimvoid 359245431Sdimvm_object_pip_wakeupn(vm_object_t object, short i) 360235633Sdim{ 361235633Sdim 362218893Sdim if (i > 0) 363218893Sdim blockcount_release(&object->paging_in_progress, i); 364218893Sdim} 365235633Sdim 366218893Sdim/* 367235633Sdim * Atomically drop the object lock and wait for pip to drain. This protects 368235633Sdim * from sleep/wakeup races due to identity changes. The lock is not re-acquired 369235633Sdim * on return. 370212904Sdim */ 371212904Sdimstatic void 372226890Sdimvm_object_pip_sleep(vm_object_t object, const char *waitid) 373226890Sdim{ 374226890Sdim 375193326Sed (void)blockcount_sleep(&object->paging_in_progress, &object->lock, 376193326Sed waitid, PVM | PDROP); 377193326Sed} 378198092Srdivacky 379193326Sedvoid 380193326Sedvm_object_pip_wait(vm_object_t object, const char *waitid) 381193326Sed{ 382193326Sed 383193326Sed VM_OBJECT_ASSERT_WLOCKED(object); 384193326Sed 385193326Sed blockcount_wait(&object->paging_in_progress, &object->lock, waitid, 386193326Sed PVM); 387193326Sed} 388193326Sed 389193326Sedvoid 390198092Srdivackyvm_object_pip_wait_unlocked(vm_object_t object, const char *waitid) 391193326Sed{ 392193326Sed 393198092Srdivacky VM_OBJECT_ASSERT_UNLOCKED(object); 394193326Sed 395193326Sed blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM); 396198092Srdivacky} 397193326Sed 398193326Sed/* 399193326Sed * vm_object_allocate: 400193326Sed * 401193326Sed * Returns a new object with the given size. 402245431Sdim */ 403245431Sdimvm_object_t 404245431Sdimvm_object_allocate(objtype_t type, vm_pindex_t size) 405245431Sdim{ 406193326Sed vm_object_t object; 407193326Sed u_short flags; 408193326Sed 409193326Sed switch (type) { 410193326Sed case OBJT_DEAD: 411193326Sed panic("vm_object_allocate: can't create OBJT_DEAD"); 412193326Sed case OBJT_SWAP: 413193326Sed flags = OBJ_COLORED | OBJ_SWAP; 414193326Sed break; 415193326Sed case OBJT_DEVICE: 416263509Sdim case OBJT_SG: 417235633Sdim flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; 418235633Sdim break; 419235633Sdim case OBJT_MGTDEVICE: 420235633Sdim flags = OBJ_FICTITIOUS; 421263509Sdim break; 422198092Srdivacky case OBJT_PHYS: 423193326Sed flags = OBJ_UNMANAGED; 424193326Sed break; 425198092Srdivacky case OBJT_VNODE: 426193326Sed flags = 0; 427193326Sed break; 428198092Srdivacky default: 429193326Sed panic("vm_object_allocate: type %d is undefined or dynamic", 430193326Sed type); 431193326Sed } 432193326Sed object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 433263509Sdim _vm_object_allocate(type, size, flags, object, NULL); 434263509Sdim 435263509Sdim return (object); 436263509Sdim} 437193326Sed 438193326Sedvm_object_t 439193326Sedvm_object_allocate_dyn(objtype_t dyntype, vm_pindex_t size, u_short flags) 440193326Sed{ 441193326Sed vm_object_t object; 442193326Sed 443193326Sed MPASS(dyntype >= OBJT_FIRST_DYN /* && dyntype < nitems(pagertab) */); 444193326Sed object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 445193326Sed _vm_object_allocate(dyntype, size, flags, object, NULL); 446193326Sed 447193326Sed return (object); 448193326Sed} 449193326Sed 450263509Sdim/* 451263509Sdim * vm_object_allocate_anon: 452263509Sdim * 453193326Sed * Returns a new default object of the given size and marked as 454193326Sed * anonymous memory for special split/collapse handling. Color 455193326Sed * to be initialized by the caller. 456193326Sed */ 457198092Srdivackyvm_object_t 458193326Sedvm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object, 459193326Sed struct ucred *cred, vm_size_t charge) 460193326Sed{ 461193326Sed vm_object_t handle, object; 462193326Sed 463245431Sdim if (backing_object == NULL) 464245431Sdim handle = NULL; 465245431Sdim else if ((backing_object->flags & OBJ_ANON) != 0) 466245431Sdim handle = backing_object->handle; 467198092Srdivacky else 468193326Sed handle = backing_object; 469193326Sed object = uma_zalloc(obj_zone, M_WAITOK); 470193326Sed _vm_object_allocate(OBJT_SWAP, size, 471193326Sed OBJ_ANON | OBJ_ONEMAPPING | OBJ_SWAP, object, handle); 472252723Sdim object->cred = cred; 473198092Srdivacky object->charge = cred != NULL ? charge : 0; 474245431Sdim return (object); 475193326Sed} 476193326Sed 477235633Sdimstatic void 478193326Sedvm_object_reference_vnode(vm_object_t object) 479193326Sed{ 480193326Sed u_int old; 481193326Sed 482193326Sed /* 483198092Srdivacky * vnode objects need the lock for the first reference 484193326Sed * to serialize with vnode_object_deallocate(). 485198092Srdivacky */ 486193326Sed if (!refcount_acquire_if_gt(&object->ref_count, 0)) { 487193326Sed VM_OBJECT_RLOCK(object); 488193326Sed old = refcount_acquire(&object->ref_count); 489193326Sed if (object->type == OBJT_VNODE && old == 0) 490193326Sed vref(object->handle); 491193326Sed VM_OBJECT_RUNLOCK(object); 492193326Sed } 493193326Sed} 494193326Sed 495252723Sdim/* 496245431Sdim * vm_object_reference: 497226890Sdim * 498193326Sed * Acquires a reference to the given object. 499193326Sed */ 500193326Sedvoid 501263509Sdimvm_object_reference(vm_object_t object) 502193326Sed{ 503193326Sed 504263509Sdim if (object == NULL) 505193326Sed return; 506198092Srdivacky 507193326Sed if (object->type == OBJT_VNODE) 508263509Sdim vm_object_reference_vnode(object); 509193326Sed else 510193326Sed refcount_acquire(&object->ref_count); 511193326Sed KASSERT((object->flags & OBJ_DEAD) == 0, 512193326Sed ("vm_object_reference: Referenced dead object.")); 513263509Sdim} 514193326Sed 515193326Sed/* 516193326Sed * vm_object_reference_locked: 517193326Sed * 518245431Sdim * Gets another reference to the given object. 519193326Sed * 520193326Sed * The object must be locked. 521193326Sed */ 522193326Sedvoid 523193326Sedvm_object_reference_locked(vm_object_t object) 524193326Sed{ 525193326Sed u_int old; 526193326Sed 527263509Sdim VM_OBJECT_ASSERT_LOCKED(object); 528198092Srdivacky old = refcount_acquire(&object->ref_count); 529193326Sed if (object->type == OBJT_VNODE && old == 0) 530193326Sed vref(object->handle); 531193326Sed KASSERT((object->flags & OBJ_DEAD) == 0, 532263509Sdim ("vm_object_reference: Referenced dead object.")); 533198092Srdivacky} 534193326Sed 535193326Sed/* 536193326Sed * Handle deallocating an object of type OBJT_VNODE. 537193326Sed */ 538193326Sedstatic void 539193326Sedvm_object_deallocate_vnode(vm_object_t object) 540193326Sed{ 541193326Sed struct vnode *vp = (struct vnode *) object->handle; 542193326Sed bool last; 543193326Sed 544193326Sed KASSERT(object->type == OBJT_VNODE, 545193326Sed ("vm_object_deallocate_vnode: not a vnode object")); 546193326Sed KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp")); 547193326Sed 548193326Sed /* Object lock to protect handle lookup. */ 549193326Sed last = refcount_release(&object->ref_count); 550193326Sed VM_OBJECT_RUNLOCK(object); 551193326Sed 552193326Sed if (!last) 553193326Sed return; 554198092Srdivacky 555193326Sed if (!umtx_shm_vnobj_persistent) 556193326Sed umtx_shm_object_terminated(object); 557193326Sed 558193326Sed /* vrele may need the vnode lock. */ 559193326Sed vrele(vp); 560193326Sed} 561193326Sed 562193326Sed/* 563193326Sed * We dropped a reference on an object and discovered that it had a 564193326Sed * single remaining shadow. This is a sibling of the reference we 565218893Sdim * dropped. Attempt to collapse the sibling and backing object. 566235633Sdim */ 567202879Srdivackystatic vm_object_t 568199990Srdivackyvm_object_deallocate_anon(vm_object_t backing_object) 569193326Sed{ 570198092Srdivacky vm_object_t object; 571198092Srdivacky 572198092Srdivacky /* Fetch the final shadow. */ 573198092Srdivacky object = LIST_FIRST(&backing_object->shadow_head); 574198092Srdivacky KASSERT(object != NULL && 575199990Srdivacky atomic_load_int(&backing_object->shadow_count) == 1, 576198092Srdivacky ("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d", 577199990Srdivacky backing_object->ref_count, 578218893Sdim atomic_load_int(&backing_object->shadow_count))); 579218893Sdim KASSERT((object->flags & OBJ_ANON) != 0, 580218893Sdim ("invalid shadow object %p", object)); 581218893Sdim 582199990Srdivacky if (!VM_OBJECT_TRYWLOCK(object)) { 583199990Srdivacky /* 584218893Sdim * Prevent object from disappearing since we do not have a 585218893Sdim * reference. 586218893Sdim */ 587218893Sdim vm_object_pip_add(object, 1); 588198092Srdivacky VM_OBJECT_WUNLOCK(backing_object); 589199990Srdivacky VM_OBJECT_WLOCK(object); 590198092Srdivacky vm_object_pip_wakeup(object); 591199990Srdivacky } else 592218893Sdim VM_OBJECT_WUNLOCK(backing_object); 593218893Sdim 594218893Sdim /* 595218893Sdim * Check for a collapse/terminate race with the last reference holder. 596218893Sdim */ 597199990Srdivacky if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 || 598198092Srdivacky !refcount_acquire_if_not_zero(&object->ref_count)) { 599199990Srdivacky VM_OBJECT_WUNLOCK(object); 600218893Sdim return (NULL); 601218893Sdim } 602218893Sdim backing_object = object->backing_object; 603218893Sdim if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0) 604218893Sdim vm_object_collapse(object); 605199990Srdivacky VM_OBJECT_WUNLOCK(object); 606198092Srdivacky 607198092Srdivacky return (object); 608198092Srdivacky} 609198092Srdivacky 610193326Sed/* 611193326Sed * vm_object_deallocate: 612263509Sdim * 613263509Sdim * Release a reference to the specified object, 614263509Sdim * gained either through a vm_object_allocate 615193326Sed * or a vm_object_reference call. When all references 616193326Sed * are gone, storage associated with this object 617193326Sed * may be relinquished. 618193326Sed * 619193326Sed * No object may be locked. 620263509Sdim */ 621193326Sedvoid 622193326Sedvm_object_deallocate(vm_object_t object) 623235633Sdim{ 624193326Sed vm_object_t temp; 625193326Sed bool released; 626198092Srdivacky 627193326Sed while (object != NULL) { 628263509Sdim /* 629263509Sdim * If the reference count goes to 0 we start calling 630263509Sdim * vm_object_terminate() on the object chain. A ref count 631263509Sdim * of 1 may be a special case depending on the shadow count 632263509Sdim * being 0 or 1. These cases require a write lock on the 633263509Sdim * object. 634263509Sdim */ 635263509Sdim if ((object->flags & OBJ_ANON) == 0) 636263509Sdim released = refcount_release_if_gt(&object->ref_count, 1); 637263509Sdim else 638263509Sdim released = refcount_release_if_gt(&object->ref_count, 2); 639235633Sdim if (released) 640235633Sdim return; 641235633Sdim 642235633Sdim if (object->type == OBJT_VNODE) { 643235633Sdim VM_OBJECT_RLOCK(object); 644245431Sdim if (object->type == OBJT_VNODE) { 645193326Sed vm_object_deallocate_vnode(object); 646193326Sed return; 647226890Sdim } 648193326Sed VM_OBJECT_RUNLOCK(object); 649193326Sed } 650193326Sed 651263509Sdim VM_OBJECT_WLOCK(object); 652263509Sdim KASSERT(object->ref_count > 0, 653263509Sdim ("vm_object_deallocate: object deallocated too many times: %d", 654263509Sdim object->type)); 655263509Sdim 656263509Sdim /* 657193326Sed * If this is not the final reference to an anonymous 658193326Sed * object we may need to collapse the shadow chain. 659263509Sdim */ 660263509Sdim if (!refcount_release(&object->ref_count)) { 661263509Sdim if (object->ref_count > 1 || 662263509Sdim atomic_load_int(&object->shadow_count) == 0) { 663263509Sdim if ((object->flags & OBJ_ANON) != 0 && 664263509Sdim object->ref_count == 1) 665263509Sdim vm_object_set_flag(object, 666263509Sdim OBJ_ONEMAPPING); 667263509Sdim VM_OBJECT_WUNLOCK(object); 668263509Sdim return; 669263509Sdim } 670263509Sdim 671263509Sdim /* Handle collapsing last ref on anonymous objects. */ 672263509Sdim object = vm_object_deallocate_anon(object); 673263509Sdim continue; 674263509Sdim } 675263509Sdim 676263509Sdim /* 677263509Sdim * Handle the final reference to an object. We restart 678263509Sdim * the loop with the backing object to avoid recursion. 679263509Sdim */ 680263509Sdim umtx_shm_object_terminated(object); 681263509Sdim temp = object->backing_object; 682263509Sdim if (temp != NULL) { 683263509Sdim KASSERT(object->type == OBJT_SWAP, 684263509Sdim ("shadowed tmpfs v_object 2 %p", object)); 685263509Sdim vm_object_backing_remove(object); 686263509Sdim } 687263509Sdim 688263509Sdim KASSERT((object->flags & OBJ_DEAD) == 0, 689263509Sdim ("vm_object_deallocate: Terminating dead object.")); 690263509Sdim vm_object_set_flag(object, OBJ_DEAD); 691263509Sdim vm_object_terminate(object); 692263509Sdim object = temp; 693263509Sdim } 694263509Sdim} 695263509Sdim 696263509Sdimvoid 697263509Sdimvm_object_destroy(vm_object_t object) 698263509Sdim{ 699263509Sdim uma_zfree(obj_zone, object); 700193326Sed} 701193326Sed 702193326Sedstatic void 703198092Srdivackyvm_object_sub_shadow(vm_object_t object) 704193326Sed{ 705193326Sed KASSERT(object->shadow_count >= 1, 706193326Sed ("object %p sub_shadow count zero", object)); 707193326Sed atomic_subtract_int(&object->shadow_count, 1); 708198092Srdivacky} 709263509Sdim 710263509Sdimstatic void 711263509Sdimvm_object_backing_remove_locked(vm_object_t object) 712193326Sed{ 713263509Sdim vm_object_t backing_object; 714193326Sed 715193326Sed backing_object = object->backing_object; 716193326Sed VM_OBJECT_ASSERT_WLOCKED(object); 717193326Sed VM_OBJECT_ASSERT_WLOCKED(backing_object); 718235633Sdim 719193326Sed KASSERT((object->flags & OBJ_COLLAPSING) == 0, 720193326Sed ("vm_object_backing_remove: Removing collapsing object.")); 721193326Sed 722193326Sed vm_object_sub_shadow(backing_object); 723193326Sed if ((object->flags & OBJ_SHADOWLIST) != 0) { 724235633Sdim LIST_REMOVE(object, shadow_list); 725193326Sed vm_object_clear_flag(object, OBJ_SHADOWLIST); 726235633Sdim } 727193326Sed object->backing_object = NULL; 728235633Sdim} 729235633Sdim 730245431Sdimstatic void 731235633Sdimvm_object_backing_remove(vm_object_t object) 732235633Sdim{ 733235633Sdim vm_object_t backing_object; 734235633Sdim 735235633Sdim VM_OBJECT_ASSERT_WLOCKED(object); 736193326Sed 737198092Srdivacky backing_object = object->backing_object; 738226890Sdim if ((object->flags & OBJ_SHADOWLIST) != 0) { 739193326Sed VM_OBJECT_WLOCK(backing_object); 740193326Sed vm_object_backing_remove_locked(object); 741193326Sed VM_OBJECT_WUNLOCK(backing_object); 742193326Sed } else { 743193326Sed object->backing_object = NULL; 744193326Sed vm_object_sub_shadow(backing_object); 745193326Sed } 746193326Sed} 747193326Sed 748193326Sedstatic void 749193326Sedvm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object) 750193326Sed{ 751198092Srdivacky 752235633Sdim VM_OBJECT_ASSERT_WLOCKED(object); 753193326Sed 754193326Sed atomic_add_int(&backing_object->shadow_count, 1); 755193326Sed if ((backing_object->flags & OBJ_ANON) != 0) { 756193326Sed VM_OBJECT_ASSERT_WLOCKED(backing_object); 757193326Sed LIST_INSERT_HEAD(&backing_object->shadow_head, object, 758193326Sed shadow_list); 759193326Sed vm_object_set_flag(object, OBJ_SHADOWLIST); 760193326Sed } 761198092Srdivacky object->backing_object = backing_object; 762193326Sed} 763263509Sdim 764252723Sdimstatic void 765252723Sdimvm_object_backing_insert(vm_object_t object, vm_object_t backing_object) 766252723Sdim{ 767252723Sdim 768252723Sdim VM_OBJECT_ASSERT_WLOCKED(object); 769252723Sdim 770252723Sdim if ((backing_object->flags & OBJ_ANON) != 0) { 771193326Sed VM_OBJECT_WLOCK(backing_object); 772193326Sed vm_object_backing_insert_locked(object, backing_object); 773193326Sed VM_OBJECT_WUNLOCK(backing_object); 774193326Sed } else { 775193326Sed object->backing_object = backing_object; 776193326Sed atomic_add_int(&backing_object->shadow_count, 1); 777252723Sdim } 778193326Sed} 779226890Sdim 780193326Sed/* 781193326Sed * Insert an object into a backing_object's shadow list with an additional 782193326Sed * reference to the backing_object added. 783193326Sed */ 784193326Sedstatic void 785198092Srdivackyvm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object) 786193326Sed{ 787193326Sed 788198092Srdivacky VM_OBJECT_ASSERT_WLOCKED(object); 789193326Sed 790193326Sed if ((backing_object->flags & OBJ_ANON) != 0) { 791193326Sed VM_OBJECT_WLOCK(backing_object); 792193326Sed KASSERT((backing_object->flags & OBJ_DEAD) == 0, 793193326Sed ("shadowing dead anonymous object")); 794198092Srdivacky vm_object_reference_locked(backing_object); 795193326Sed vm_object_backing_insert_locked(object, backing_object); 796193326Sed vm_object_clear_flag(backing_object, OBJ_ONEMAPPING); 797252723Sdim VM_OBJECT_WUNLOCK(backing_object); 798193326Sed } else { 799193326Sed vm_object_reference(backing_object); 800193326Sed atomic_add_int(&backing_object->shadow_count, 1); 801193326Sed object->backing_object = backing_object; 802193326Sed } 803193326Sed} 804198092Srdivacky 805193326Sed/* 806193326Sed * Transfer a backing reference from backing_object to object. 807252723Sdim */ 808193326Sedstatic void 809226890Sdimvm_object_backing_transfer(vm_object_t object, vm_object_t backing_object) 810193326Sed{ 811193326Sed vm_object_t new_backing_object; 812193326Sed 813198092Srdivacky /* 814193326Sed * Note that the reference to backing_object->backing_object 815193326Sed * moves from within backing_object to within object. 816193326Sed */ 817193326Sed vm_object_backing_remove_locked(object); 818193326Sed new_backing_object = backing_object->backing_object; 819193326Sed if (new_backing_object == NULL) 820193326Sed return; 821193326Sed if ((new_backing_object->flags & OBJ_ANON) != 0) { 822193326Sed VM_OBJECT_WLOCK(new_backing_object); 823193326Sed vm_object_backing_remove_locked(backing_object); 824193326Sed vm_object_backing_insert_locked(object, new_backing_object); 825193326Sed VM_OBJECT_WUNLOCK(new_backing_object); 826193326Sed } else { 827193326Sed /* 828193326Sed * shadow_count for new_backing_object is left 829193326Sed * unchanged, its reference provided by backing_object 830198092Srdivacky * is replaced by object. 831193326Sed */ 832193326Sed object->backing_object = new_backing_object; 833193326Sed backing_object->backing_object = NULL; 834193326Sed } 835193326Sed} 836193326Sed 837193326Sed/* 838245431Sdim * Wait for a concurrent collapse to settle. 839245431Sdim */ 840245431Sdimstatic void 841245431Sdimvm_object_collapse_wait(vm_object_t object) 842245431Sdim{ 843245431Sdim 844245431Sdim VM_OBJECT_ASSERT_WLOCKED(object); 845245431Sdim 846245431Sdim while ((object->flags & OBJ_COLLAPSING) != 0) { 847245431Sdim vm_object_pip_wait(object, "vmcolwait"); 848245431Sdim counter_u64_add(object_collapse_waits, 1); 849245431Sdim } 850245431Sdim} 851245431Sdim 852193326Sed/* 853193326Sed * Waits for a backing object to clear a pending collapse and returns 854193326Sed * it locked if it is an ANON object. 855193326Sed */ 856193326Sedstatic vm_object_t 857193326Sedvm_object_backing_collapse_wait(vm_object_t object) 858193326Sed{ 859193326Sed vm_object_t backing_object; 860193326Sed 861193326Sed VM_OBJECT_ASSERT_WLOCKED(object); 862193326Sed 863245431Sdim for (;;) { 864245431Sdim backing_object = object->backing_object; 865193326Sed if (backing_object == NULL || 866245431Sdim (backing_object->flags & OBJ_ANON) == 0) 867263509Sdim return (NULL); 868263509Sdim VM_OBJECT_WLOCK(backing_object); 869193326Sed if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0) 870193326Sed break; 871193326Sed VM_OBJECT_WUNLOCK(object); 872193326Sed vm_object_pip_sleep(backing_object, "vmbckwait"); 873193326Sed counter_u64_add(object_collapse_waits, 1); 874193326Sed VM_OBJECT_WLOCK(object); 875193326Sed } 876193326Sed return (backing_object); 877245431Sdim} 878193326Sed 879193326Sed/* 880193326Sed * vm_object_terminate_pages removes any remaining pageable pages 881193326Sed * from the object and resets the object to an empty state. 882198092Srdivacky */ 883193326Sedstatic void 884245431Sdimvm_object_terminate_pages(vm_object_t object) 885263509Sdim{ 886263509Sdim vm_page_t p, p_next; 887263509Sdim 888263509Sdim VM_OBJECT_ASSERT_WLOCKED(object); 889263509Sdim 890252723Sdim /* 891198092Srdivacky * Free any remaining pageable pages. This also removes them from the 892193326Sed * paging queues. However, don't free wired pages, just remove them 893193326Sed * from the object. Rather than incrementally removing each page from 894198092Srdivacky * the object, the page and object are reset to any empty state. 895193326Sed */ 896198092Srdivacky TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { 897193326Sed vm_page_assert_unbusied(p); 898193326Sed KASSERT(p->object == object && 899193326Sed (p->ref_count & VPRC_OBJREF) != 0, 900193326Sed ("vm_object_terminate_pages: page %p is inconsistent", p)); 901193326Sed 902193326Sed p->object = NULL; 903193326Sed if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) { 904193326Sed VM_CNT_INC(v_pfree); 905193326Sed vm_page_free(p); 906193326Sed } 907193326Sed } 908193326Sed 909193326Sed /* 910193326Sed * If the object contained any pages, then reset it to an empty state. 911193326Sed * None of the object's fields, including "resident_page_count", were 912193326Sed * modified by the preceding loop. 913201361Srdivacky */ 914201361Srdivacky if (object->resident_page_count != 0) { 915193326Sed vm_radix_reclaim_allnodes(&object->rtree); 916198092Srdivacky TAILQ_INIT(&object->memq); 917198092Srdivacky object->resident_page_count = 0; 918263509Sdim if (object->type == OBJT_VNODE) 919263509Sdim vdrop(object->handle); 920263509Sdim } 921263509Sdim} 922263509Sdim 923263509Sdim/* 924263509Sdim * vm_object_terminate actually destroys the specified object, freeing 925263509Sdim * up all previously used resources. 926263509Sdim * 927263509Sdim * The object must be locked. 928263509Sdim * This routine may block. 929193326Sed */ 930193326Sedvoid 931193326Sedvm_object_terminate(vm_object_t object) 932245431Sdim{ 933235633Sdim 934235633Sdim VM_OBJECT_ASSERT_WLOCKED(object); 935235633Sdim KASSERT((object->flags & OBJ_DEAD) != 0, 936235633Sdim ("terminating non-dead obj %p", object)); 937235633Sdim KASSERT((object->flags & OBJ_COLLAPSING) == 0, 938226890Sdim ("terminating collapsing obj %p", object)); 939226890Sdim KASSERT(object->backing_object == NULL, 940226890Sdim ("terminating shadow obj %p", object)); 941226890Sdim 942226890Sdim /* 943226890Sdim * Wait for the pageout daemon and other current users to be 944226890Sdim * done with the object. Note that new paging_in_progress 945226890Sdim * users can come after this wait, but they must check 946226890Sdim * OBJ_DEAD flag set (without unlocking the object), and avoid 947226890Sdim * the object being terminated. 948226890Sdim */ 949226890Sdim vm_object_pip_wait(object, "objtrm"); 950235633Sdim 951226890Sdim KASSERT(object->ref_count == 0, 952226890Sdim ("vm_object_terminate: object with references, ref_count=%d", 953226890Sdim object->ref_count)); 954226890Sdim 955226890Sdim if ((object->flags & OBJ_PG_DTOR) == 0) 956226890Sdim vm_object_terminate_pages(object); 957226890Sdim 958226890Sdim#if VM_NRESERVLEVEL > 0 959226890Sdim if (__predict_false(!LIST_EMPTY(&object->rvq))) 960226890Sdim vm_reserv_break_all(object); 961226890Sdim#endif 962226890Sdim 963235633Sdim KASSERT(object->cred == NULL || (object->flags & OBJ_SWAP) != 0, 964226890Sdim ("%s: non-swap obj %p has cred", __func__, object)); 965226890Sdim 966226890Sdim /* 967226890Sdim * Let the pager know object is dead. 968245431Sdim */ 969226890Sdim vm_pager_deallocate(object); 970193326Sed VM_OBJECT_WUNLOCK(object); 971226890Sdim 972226890Sdim vm_object_destroy(object); 973235633Sdim} 974193326Sed 975198092Srdivacky/* 976226890Sdim * Make the page read-only so that we can clear the object flags. However, if 977198092Srdivacky * this is a nosync mmap then the object is likely to stay dirty so do not 978235633Sdim * mess with the page and do not clear the object flags. Returns TRUE if the 979235633Sdim * page should be flushed, and FALSE otherwise. 980235633Sdim */ 981235633Sdimstatic boolean_t 982226890Sdimvm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean) 983226890Sdim{ 984226890Sdim 985193326Sed vm_page_assert_busied(p); 986193326Sed 987193326Sed /* 988193326Sed * If we have been asked to skip nosync pages and this is a 989235633Sdim * nosync page, skip it. Note that the object flags were not 990235633Sdim * cleared in this case so we do not have to set them. 991235633Sdim */ 992235633Sdim if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) { 993235633Sdim *allclean = FALSE; 994235633Sdim return (FALSE); 995235633Sdim } else { 996235633Sdim pmap_remove_write(p); 997235633Sdim return (p->dirty != 0); 998235633Sdim } 999235633Sdim} 1000235633Sdim 1001235633Sdim/* 1002235633Sdim * vm_object_page_clean 1003198092Srdivacky * 1004221345Sdim * Clean all dirty pages in the specified range of object. Leaves page 1005193326Sed * on whatever queue it is currently on. If NOSYNC is set then do not 1006193326Sed * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC), 1007193326Sed * leaving the object dirty. 1008193326Sed * 1009193326Sed * For swap objects backing tmpfs regular files, do not flush anything, 1010193326Sed * but remove write protection on the mapped pages to update mtime through 1011193326Sed * mmaped writes. 1012193326Sed * 1013193326Sed * When stuffing pages asynchronously, allow clustering. XXX we need a 1014263509Sdim * synchronous clustering mode implementation. 1015263509Sdim * 1016235633Sdim * Odd semantics: if start == end, we clean everything. 1017235633Sdim * 1018198092Srdivacky * The object must be locked. 1019235633Sdim * 1020235633Sdim * Returns FALSE if some page from the range was not written, as 1021235633Sdim * reported by the pager, and TRUE otherwise. 1022235633Sdim */ 1023235633Sdimboolean_t 1024263509Sdimvm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, 1025263509Sdim int flags) 1026235633Sdim{ 1027235633Sdim vm_page_t np, p; 1028235633Sdim vm_pindex_t pi, tend, tstart; 1029235633Sdim int curgeneration, n, pagerflags; 1030235633Sdim boolean_t eio, res, allclean; 1031235633Sdim 1032235633Sdim VM_OBJECT_ASSERT_WLOCKED(object); 1033218893Sdim 1034263509Sdim if (!vm_object_mightbedirty(object) || object->resident_page_count == 0) 1035235633Sdim return (TRUE); 1036235633Sdim 1037235633Sdim pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? 1038235633Sdim VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; 1039235633Sdim pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; 1040235633Sdim 1041235633Sdim tstart = OFF_TO_IDX(start); 1042235633Sdim tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); 1043235633Sdim allclean = tstart == 0 && tend >= object->size; 1044235633Sdim res = TRUE; 1045263509Sdim 1046263509Sdimrescan: 1047263509Sdim curgeneration = object->generation; 1048263509Sdim 1049263509Sdim for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { 1050263509Sdim pi = p->pindex; 1051235633Sdim if (pi >= tend) 1052235633Sdim break; 1053235633Sdim np = TAILQ_NEXT(p, listq); 1054235633Sdim if (vm_page_none_valid(p)) 1055235633Sdim continue; 1056235633Sdim if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) { 1057235633Sdim if (object->generation != curgeneration && 1058235633Sdim (flags & OBJPC_SYNC) != 0) 1059235633Sdim goto rescan; 1060263509Sdim np = vm_page_find_least(object, pi); 1061235633Sdim continue; 1062235633Sdim } 1063235633Sdim if (!vm_object_page_remove_write(p, flags, &allclean)) { 1064218893Sdim vm_page_xunbusy(p); 1065235633Sdim continue; 1066263509Sdim } 1067235633Sdim if (object->type == OBJT_VNODE) { 1068235633Sdim n = vm_object_page_collect_flush(object, p, pagerflags, 1069235633Sdim flags, &allclean, &eio); 1070235633Sdim if (eio) { 1071226890Sdim res = FALSE; 1072218893Sdim allclean = FALSE; 1073235633Sdim } 1074235633Sdim if (object->generation != curgeneration && 1075218893Sdim (flags & OBJPC_SYNC) != 0) 1076263509Sdim goto rescan; 1077235633Sdim 1078235633Sdim /* 1079235633Sdim * If the VOP_PUTPAGES() did a truncated write, so 1080235633Sdim * that even the first page of the run is not fully 1081263509Sdim * written, vm_pageout_flush() returns 0 as the run 1082235633Sdim * length. Since the condition that caused truncated 1083235633Sdim * write may be permanent, e.g. exhausted free space, 1084235633Sdim * accepting n == 0 would cause an infinite loop. 1085245431Sdim * 1086235633Sdim * Forwarding the iterator leaves the unwritten page 1087193326Sed * behind, but there is not much we can do there if 1088235633Sdim * filesystem refuses to write it. 1089235633Sdim */ 1090193326Sed if (n == 0) { 1091235633Sdim n = 1; 1092235633Sdim allclean = FALSE; 1093245431Sdim } 1094245431Sdim } else { 1095245431Sdim n = 1; 1096235633Sdim vm_page_xunbusy(p); 1097193326Sed } 1098193326Sed np = vm_page_find_least(object, pi + n); 1099263509Sdim } 1100235633Sdim#if 0 1101193326Sed VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); 1102235633Sdim#endif 1103226890Sdim 1104235633Sdim /* 1105235633Sdim * Leave updating cleangeneration for tmpfs objects to tmpfs 1106235633Sdim * scan. It needs to update mtime, which happens for other 1107193326Sed * filesystems during page writeouts. 1108193326Sed */ 1109235633Sdim if (allclean && object->type == OBJT_VNODE) 1110193326Sed object->cleangeneration = curgeneration; 1111263509Sdim return (res); 1112198092Srdivacky} 1113263509Sdim 1114193326Sedstatic int 1115235633Sdimvm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, 1116235633Sdim int flags, boolean_t *allclean, boolean_t *eio) 1117235633Sdim{ 1118235633Sdim vm_page_t ma[vm_pageout_page_count], p_first, tp; 1119235633Sdim int count, i, mreq, runlen; 1120235633Sdim 1121235633Sdim vm_page_lock_assert(p, MA_NOTOWNED); 1122263509Sdim vm_page_assert_xbusied(p); 1123235633Sdim VM_OBJECT_ASSERT_WLOCKED(object); 1124235633Sdim 1125235633Sdim count = 1; 1126235633Sdim mreq = 0; 1127235633Sdim 1128235633Sdim for (tp = p; count < vm_pageout_page_count; count++) { 1129235633Sdim tp = vm_page_next(tp); 1130235633Sdim if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1131235633Sdim break; 1132235633Sdim if (!vm_object_page_remove_write(tp, flags, allclean)) { 1133235633Sdim vm_page_xunbusy(tp); 1134263509Sdim break; 1135235633Sdim } 1136235633Sdim } 1137193326Sed 1138193326Sed for (p_first = p; count < vm_pageout_page_count; count++) { 1139198092Srdivacky tp = vm_page_prev(p_first); 1140193326Sed if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1141193326Sed break; 1142193326Sed if (!vm_object_page_remove_write(tp, flags, allclean)) { 1143193326Sed vm_page_xunbusy(tp); 1144226890Sdim break; 1145235633Sdim } 1146193326Sed p_first = tp; 1147193326Sed mreq++; 1148193326Sed } 1149245431Sdim 1150226890Sdim for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) 1151226890Sdim ma[i] = tp; 1152226890Sdim 1153226890Sdim vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); 1154226890Sdim return (runlen); 1155226890Sdim} 1156226890Sdim 1157226890Sdim/* 1158193326Sed * Note that there is absolutely no sense in writing out 1159193326Sed * anonymous objects, so we track down the vnode object 1160193326Sed * to write out. 1161193326Sed * We invalidate (remove) all pages from the address space 1162226890Sdim * for semantic correctness. 1163226890Sdim * 1164226890Sdim * If the backing object is a device object with unmanaged pages, then any 1165193326Sed * mappings to the specified range of pages must be removed before this 1166226890Sdim * function is called. 1167226890Sdim * 1168226890Sdim * Note: certain anonymous maps, such as MAP_NOSYNC maps, 1169226890Sdim * may start out with a NULL object. 1170226890Sdim */ 1171226890Sdimboolean_t 1172226890Sdimvm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, 1173226890Sdim boolean_t syncio, boolean_t invalidate) 1174226890Sdim{ 1175226890Sdim vm_object_t backing_object; 1176226890Sdim struct vnode *vp; 1177226890Sdim struct mount *mp; 1178226890Sdim int error, flags, fsync_after; 1179226890Sdim boolean_t res; 1180226890Sdim 1181226890Sdim if (object == NULL) 1182226890Sdim return (TRUE); 1183226890Sdim res = TRUE; 1184226890Sdim error = 0; 1185226890Sdim VM_OBJECT_WLOCK(object); 1186226890Sdim while ((backing_object = object->backing_object) != NULL) { 1187226890Sdim VM_OBJECT_WLOCK(backing_object); 1188226890Sdim offset += object->backing_object_offset; 1189226890Sdim VM_OBJECT_WUNLOCK(object); 1190226890Sdim object = backing_object; 1191226890Sdim if (object->size < OFF_TO_IDX(offset + size)) 1192226890Sdim size = IDX_TO_OFF(object->size) - offset; 1193226890Sdim } 1194226890Sdim /* 1195226890Sdim * Flush pages if writing is allowed, invalidate them 1196193326Sed * if invalidation requested. Pages undergoing I/O 1197193326Sed * will be ignored by vm_object_page_remove(). 1198193326Sed * 1199193326Sed * We cannot lock the vnode and then wait for paging 1200193326Sed * to complete without deadlocking against vm_fault. 1201245431Sdim * Instead we simply call vm_object_page_remove() and 1202193326Sed * allow it to block internally on a page-by-page 1203193326Sed * basis when it encounters pages undergoing async 1204193326Sed * I/O. 1205218893Sdim */ 1206235633Sdim if (object->type == OBJT_VNODE && 1207223017Sdim vm_object_mightbedirty(object) != 0 && 1208226890Sdim ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { 1209226890Sdim VM_OBJECT_WUNLOCK(object); 1210218893Sdim (void)vn_start_write(vp, &mp, V_WAIT); 1211218893Sdim vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1212218893Sdim if (syncio && !invalidate && offset == 0 && 1213218893Sdim atop(size) == object->size) { 1214223017Sdim /* 1215223017Sdim * If syncing the whole mapping of the file, 1216245431Sdim * it is faster to schedule all the writes in 1217245431Sdim * async mode, also allowing the clustering, 1218223017Sdim * and then wait for i/o to complete. 1219193326Sed */ 1220193326Sed flags = 0; 1221193326Sed fsync_after = TRUE; 1222193326Sed } else { 1223223017Sdim flags = (syncio || invalidate) ? OBJPC_SYNC : 0; 1224193326Sed flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; 1225223017Sdim fsync_after = FALSE; 1226193326Sed } 1227226890Sdim VM_OBJECT_WLOCK(object); 1228198092Srdivacky res = vm_object_page_clean(object, offset, offset + size, 1229193326Sed flags); 1230193326Sed VM_OBJECT_WUNLOCK(object); 1231193326Sed if (fsync_after) { 1232193326Sed for (;;) { 1233193326Sed error = VOP_FSYNC(vp, MNT_WAIT, curthread); 1234245431Sdim if (error != ERELOOKUP) 1235245431Sdim break; 1236223017Sdim 1237193326Sed /* 1238193326Sed * Allow SU/bufdaemon to handle more 1239223017Sdim * dependencies in the meantime. 1240193326Sed */ 1241198092Srdivacky VOP_UNLOCK(vp); 1242193326Sed vn_finished_write(mp); 1243198092Srdivacky 1244193326Sed (void)vn_start_write(vp, &mp, V_WAIT); 1245198092Srdivacky vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1246226890Sdim } 1247226890Sdim } 1248226890Sdim VOP_UNLOCK(vp); 1249226890Sdim vn_finished_write(mp); 1250226890Sdim if (error != 0) 1251226890Sdim res = FALSE; 1252226890Sdim VM_OBJECT_WLOCK(object); 1253245431Sdim } 1254226890Sdim if ((object->type == OBJT_VNODE || 1255226890Sdim object->type == OBJT_DEVICE) && invalidate) { 1256226890Sdim if (object->type == OBJT_DEVICE) 1257226890Sdim /* 1258193326Sed * The option OBJPR_NOTMAPPED must be passed here 1259193326Sed * because vm_object_page_remove() cannot remove 1260193326Sed * unmanaged mappings. 1261193326Sed */ 1262198092Srdivacky flags = OBJPR_NOTMAPPED; 1263193326Sed else if (old_msync) 1264198092Srdivacky flags = 0; 1265226890Sdim else 1266226890Sdim flags = OBJPR_CLEANONLY; 1267226890Sdim vm_object_page_remove(object, OFF_TO_IDX(offset), 1268226890Sdim OFF_TO_IDX(offset + size + PAGE_MASK), flags); 1269198092Srdivacky } 1270193326Sed VM_OBJECT_WUNLOCK(object); 1271193326Sed return (res); 1272226890Sdim} 1273198092Srdivacky 1274193326Sed/* 1275193326Sed * Determine whether the given advice can be applied to the object. Advice is 1276198092Srdivacky * not applied to unmanaged pages since they never belong to page queues, and 1277193326Sed * since MADV_FREE is destructive, it can apply only to anonymous pages that 1278235633Sdim * have been mapped at most once. 1279193326Sed */ 1280198092Srdivackystatic bool 1281193326Sedvm_object_advice_applies(vm_object_t object, int advice) 1282193326Sed{ 1283193326Sed 1284198092Srdivacky if ((object->flags & OBJ_UNMANAGED) != 0) 1285193326Sed return (false); 1286198092Srdivacky if (advice != MADV_FREE) 1287235633Sdim return (true); 1288235633Sdim return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) == 1289193326Sed (OBJ_ONEMAPPING | OBJ_ANON)); 1290193326Sed} 1291193326Sed 1292193326Sedstatic void 1293193326Sedvm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex, 1294205219Srdivacky vm_size_t size) 1295218893Sdim{ 1296218893Sdim 1297218893Sdim if (advice == MADV_FREE) 1298245431Sdim vm_pager_freespace(object, pindex, size); 1299245431Sdim} 1300205219Srdivacky 1301235633Sdim/* 1302235633Sdim * vm_object_madvise: 1303235633Sdim * 1304235633Sdim * Implements the madvise function at the object/page level. 1305235633Sdim * 1306235633Sdim * MADV_WILLNEED (any object) 1307235633Sdim * 1308235633Sdim * Activate the specified pages if they are resident. 1309235633Sdim * 1310235633Sdim * MADV_DONTNEED (any object) 1311235633Sdim * 1312235633Sdim * Deactivate the specified pages if they are resident. 1313235633Sdim * 1314235633Sdim * MADV_FREE (OBJT_SWAP objects, OBJ_ONEMAPPING only) 1315235633Sdim * 1316235633Sdim * Deactivate and clean the specified pages if they are 1317235633Sdim * resident. This permits the process to reuse the pages 1318235633Sdim * without faulting or the kernel to reclaim the pages 1319235633Sdim * without I/O. 1320235633Sdim */ 1321235633Sdimvoid 1322235633Sdimvm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, 1323235633Sdim int advice) 1324235633Sdim{ 1325235633Sdim vm_pindex_t tpindex; 1326235633Sdim vm_object_t backing_object, tobject; 1327235633Sdim vm_page_t m, tm; 1328235633Sdim 1329235633Sdim if (object == NULL) 1330235633Sdim return; 1331235633Sdim 1332235633Sdimrelookup: 1333235633Sdim VM_OBJECT_WLOCK(object); 1334235633Sdim if (!vm_object_advice_applies(object, advice)) { 1335235633Sdim VM_OBJECT_WUNLOCK(object); 1336235633Sdim return; 1337193326Sed } 1338198092Srdivacky for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) { 1339226890Sdim tobject = object; 1340226890Sdim 1341193326Sed /* 1342226890Sdim * If the next page isn't resident in the top-level object, we 1343226890Sdim * need to search the shadow chain. When applying MADV_FREE, we 1344226890Sdim * take care to release any swap space used to store 1345212904Sdim * non-resident pages. 1346198092Srdivacky */ 1347226890Sdim if (m == NULL || pindex < m->pindex) { 1348226890Sdim /* 1349226890Sdim * Optimize a common case: if the top-level object has 1350198092Srdivacky * no backing object, we can skip over the non-resident 1351226890Sdim * range in constant time. 1352226890Sdim */ 1353193326Sed if (object->backing_object == NULL) { 1354226890Sdim tpindex = (m != NULL && m->pindex < end) ? 1355198092Srdivacky m->pindex : end; 1356235633Sdim vm_object_madvise_freespace(object, advice, 1357235633Sdim pindex, tpindex - pindex); 1358235633Sdim if ((pindex = tpindex) == end) 1359226890Sdim break; 1360226890Sdim goto next_page; 1361245431Sdim } 1362245431Sdim 1363245431Sdim tpindex = pindex; 1364226890Sdim do { 1365245431Sdim vm_object_madvise_freespace(tobject, advice, 1366245431Sdim tpindex, 1); 1367245431Sdim /* 1368245431Sdim * Prepare to search the next object in the 1369245431Sdim * chain. 1370226890Sdim */ 1371226890Sdim backing_object = tobject->backing_object; 1372226890Sdim if (backing_object == NULL) 1373226890Sdim goto next_pindex; 1374226890Sdim VM_OBJECT_WLOCK(backing_object); 1375226890Sdim tpindex += 1376226890Sdim OFF_TO_IDX(tobject->backing_object_offset); 1377226890Sdim if (tobject != object) 1378226890Sdim VM_OBJECT_WUNLOCK(tobject); 1379193326Sed tobject = backing_object; 1380226890Sdim if (!vm_object_advice_applies(tobject, advice)) 1381226890Sdim goto next_pindex; 1382226890Sdim } while ((tm = vm_page_lookup(tobject, tpindex)) == 1383226890Sdim NULL); 1384198092Srdivacky } else { 1385226890Sdimnext_page: 1386226890Sdim tm = m; 1387226890Sdim m = TAILQ_NEXT(m, listq); 1388226890Sdim } 1389226890Sdim 1390226890Sdim /* 1391226890Sdim * If the page is not in a normal state, skip it. The page 1392226890Sdim * can not be invalidated while the object lock is held. 1393226890Sdim */ 1394245431Sdim if (!vm_page_all_valid(tm) || vm_page_wired(tm)) 1395245431Sdim goto next_pindex; 1396245431Sdim KASSERT((tm->flags & PG_FICTITIOUS) == 0, 1397226890Sdim ("vm_object_madvise: page %p is fictitious", tm)); 1398193326Sed KASSERT((tm->oflags & VPO_UNMANAGED) == 0, 1399226890Sdim ("vm_object_madvise: page %p is not managed", tm)); 1400226890Sdim if (vm_page_tryxbusy(tm) == 0) { 1401235633Sdim if (object != tobject) 1402235633Sdim VM_OBJECT_WUNLOCK(object); 1403235633Sdim if (advice == MADV_WILLNEED) { 1404226890Sdim /* 1405226890Sdim * Reference the page before unlocking and 1406226890Sdim * sleeping so that the page daemon is less 1407226890Sdim * likely to reclaim it. 1408226890Sdim */ 1409245431Sdim vm_page_aflag_set(tm, PGA_REFERENCED); 1410226890Sdim } 1411245431Sdim if (!vm_page_busy_sleep(tm, "madvpo", 0)) 1412198092Srdivacky VM_OBJECT_WUNLOCK(tobject); 1413235633Sdim goto relookup; 1414235633Sdim } 1415235633Sdim vm_page_advise(tm, advice); 1416235633Sdim vm_page_xunbusy(tm); 1417235633Sdim vm_object_madvise_freespace(tobject, advice, tm->pindex, 1); 1418235633Sdimnext_pindex: 1419235633Sdim if (tobject != object) 1420235633Sdim VM_OBJECT_WUNLOCK(tobject); 1421235633Sdim } 1422235633Sdim VM_OBJECT_WUNLOCK(object); 1423235633Sdim} 1424235633Sdim 1425235633Sdim/* 1426235633Sdim * vm_object_shadow: 1427235633Sdim * 1428235633Sdim * Create a new object which is backed by the 1429193326Sed * specified existing object range. The source 1430193326Sed * object reference is deallocated. 1431193326Sed * 1432198092Srdivacky * The new object and offset into that object 1433193326Sed * are returned in the source parameters. 1434235633Sdim */ 1435235633Sdimvoid 1436235633Sdimvm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length, 1437235633Sdim struct ucred *cred, bool shared) 1438235633Sdim{ 1439235633Sdim vm_object_t source; 1440235633Sdim vm_object_t result; 1441235633Sdim 1442235633Sdim source = *object; 1443235633Sdim 1444235633Sdim /* 1445235633Sdim * Don't create the new object if the old object isn't shared. 1446235633Sdim * 1447235633Sdim * If we hold the only reference we can guarantee that it won't 1448193326Sed * increase while we have the map locked. Otherwise the race is 1449193326Sed * harmless and we will end up with an extra shadow object that 1450218893Sdim * will be collapsed later. 1451245431Sdim */ 1452245431Sdim if (source != NULL && source->ref_count == 1 && 1453218893Sdim (source->flags & OBJ_ANON) != 0) 1454218893Sdim return; 1455218893Sdim 1456226890Sdim /* 1457193326Sed * Allocate a new object with the given length. 1458193326Sed */ 1459218893Sdim result = vm_object_allocate_anon(atop(length), source, cred, length); 1460212904Sdim 1461218893Sdim /* 1462245431Sdim * Store the offset into the source object, and fix up the offset into 1463212904Sdim * the new object. 1464245431Sdim */ 1465218893Sdim result->backing_object_offset = *offset; 1466212904Sdim 1467218893Sdim if (shared || source != NULL) { 1468212904Sdim VM_OBJECT_WLOCK(result); 1469212904Sdim 1470193326Sed /* 1471193326Sed * The new object shadows the source object, adding a 1472193326Sed * reference to it. Our caller changes his reference 1473245431Sdim * to point to the new object, removing a reference to 1474245431Sdim * the source object. Net result: no change of 1475245431Sdim * reference count, unless the caller needs to add one 1476245431Sdim * more reference due to forking a shared map entry. 1477245431Sdim */ 1478245431Sdim if (shared) { 1479245431Sdim vm_object_reference_locked(result); 1480245431Sdim vm_object_clear_flag(result, OBJ_ONEMAPPING); 1481193326Sed } 1482245431Sdim 1483245431Sdim /* 1484226890Sdim * Try to optimize the result object's page color when 1485245431Sdim * shadowing in order to maintain page coloring 1486245431Sdim * consistency in the combined shadowed object. 1487245431Sdim */ 1488245431Sdim if (source != NULL) { 1489245431Sdim vm_object_backing_insert(result, source); 1490245431Sdim result->domain = source->domain; 1491226890Sdim#if VM_NRESERVLEVEL > 0 1492235633Sdim vm_object_set_flag(result, 1493235633Sdim (source->flags & OBJ_COLORED)); 1494235633Sdim result->pg_color = (source->pg_color + 1495235633Sdim OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 1496245431Sdim 1)) - 1); 1497245431Sdim#endif 1498245431Sdim } 1499245431Sdim VM_OBJECT_WUNLOCK(result); 1500245431Sdim } 1501245431Sdim 1502245431Sdim /* 1503245431Sdim * Return the new things 1504245431Sdim */ 1505245431Sdim *offset = 0; 1506245431Sdim *object = result; 1507245431Sdim} 1508245431Sdim 1509245431Sdim/* 1510245431Sdim * vm_object_split: 1511245431Sdim * 1512245431Sdim * Split the pages in a map entry into a new object. This affords 1513245431Sdim * easier removal of unused pages, and keeps object inheritance from 1514245431Sdim * being a negative impact on memory usage. 1515245431Sdim */ 1516245431Sdimvoid 1517245431Sdimvm_object_split(vm_map_entry_t entry) 1518245431Sdim{ 1519245431Sdim vm_page_t m, m_next; 1520245431Sdim vm_object_t orig_object, new_object, backing_object; 1521245431Sdim vm_pindex_t idx, offidxstart; 1522245431Sdim vm_size_t size; 1523245431Sdim 1524245431Sdim orig_object = entry->object.vm_object; 1525245431Sdim KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0, 1526245431Sdim ("vm_object_split: Splitting object with multiple mappings.")); 1527245431Sdim if ((orig_object->flags & OBJ_ANON) == 0) 1528235633Sdim return; 1529235633Sdim if (orig_object->ref_count <= 1) 1530226890Sdim return; 1531245431Sdim VM_OBJECT_WUNLOCK(orig_object); 1532245431Sdim 1533245431Sdim offidxstart = OFF_TO_IDX(entry->offset); 1534245431Sdim size = atop(entry->end - entry->start); 1535245431Sdim 1536245431Sdim new_object = vm_object_allocate_anon(size, orig_object, 1537193326Sed orig_object->cred, ptoa(size)); 1538193326Sed 1539193326Sed /* 1540193326Sed * We must wait for the orig_object to complete any in-progress 1541218893Sdim * collapse so that the swap blocks are stable below. The 1542193326Sed * additional reference on backing_object by new object will 1543235633Sdim * prevent further collapse operations until split completes. 1544193326Sed */ 1545198092Srdivacky VM_OBJECT_WLOCK(orig_object); 1546205219Srdivacky vm_object_collapse_wait(orig_object); 1547193326Sed 1548218893Sdim /* 1549218893Sdim * At this point, the new object is still private, so the order in 1550218893Sdim * which the original and new objects are locked does not matter. 1551205219Srdivacky */ 1552193326Sed VM_OBJECT_WLOCK(new_object); 1553245431Sdim new_object->domain = orig_object->domain; 1554245431Sdim backing_object = orig_object->backing_object; 1555245431Sdim if (backing_object != NULL) { 1556245431Sdim vm_object_backing_insert_ref(new_object, backing_object); 1557245431Sdim new_object->backing_object_offset = 1558245431Sdim orig_object->backing_object_offset + entry->offset; 1559245431Sdim } 1560226890Sdim if (orig_object->cred != NULL) { 1561226890Sdim crhold(orig_object->cred); 1562193326Sed KASSERT(orig_object->charge >= ptoa(size), 1563245431Sdim ("orig_object->charge < 0")); 1564245431Sdim orig_object->charge -= ptoa(size); 1565245431Sdim } 1566245431Sdim 1567245431Sdim /* 1568245431Sdim * Mark the split operation so that swap_pager_getpages() knows 1569245431Sdim * that the object is in transition. 1570245431Sdim */ 1571245431Sdim vm_object_set_flag(orig_object, OBJ_SPLIT); 1572245431Sdim#ifdef INVARIANTS 1573245431Sdim idx = 0; 1574245431Sdim#endif 1575245431Sdimretry: 1576198092Srdivacky m = vm_page_find_least(orig_object, offidxstart); 1577245431Sdim KASSERT(m == NULL || idx <= m->pindex - offidxstart, 1578193326Sed ("%s: object %p was repopulated", __func__, orig_object)); 1579193326Sed for (; m != NULL && (idx = m->pindex - offidxstart) < size; 1580198092Srdivacky m = m_next) { 1581193326Sed m_next = TAILQ_NEXT(m, listq); 1582193326Sed 1583193326Sed /* 1584198092Srdivacky * We must wait for pending I/O to complete before we can 1585193326Sed * rename the page. 1586193326Sed * 1587193326Sed * We do not have to VM_PROT_NONE the page as mappings should 1588193326Sed * not be changed by this operation. 1589193326Sed */ 1590193326Sed if (vm_page_tryxbusy(m) == 0) { 1591198092Srdivacky VM_OBJECT_WUNLOCK(new_object); 1592193326Sed if (vm_page_busy_sleep(m, "spltwt", 0)) 1593193326Sed VM_OBJECT_WLOCK(orig_object); 1594245431Sdim VM_OBJECT_WLOCK(new_object); 1595245431Sdim goto retry; 1596245431Sdim } 1597245431Sdim 1598245431Sdim /* 1599245431Sdim * The page was left invalid. Likely placed there by 1600245431Sdim * an incomplete fault. Just remove and ignore. 1601245431Sdim */ 1602245431Sdim if (vm_page_none_valid(m)) { 1603245431Sdim if (vm_page_remove(m)) 1604245431Sdim vm_page_free(m); 1605245431Sdim continue; 1606245431Sdim } 1607245431Sdim 1608245431Sdim /* vm_page_rename() will dirty the page. */ 1609245431Sdim if (vm_page_rename(m, new_object, idx)) { 1610193326Sed vm_page_xunbusy(m); 1611193326Sed VM_OBJECT_WUNLOCK(new_object); 1612198092Srdivacky VM_OBJECT_WUNLOCK(orig_object); 1613193326Sed vm_radix_wait(); 1614193326Sed VM_OBJECT_WLOCK(orig_object); 1615 VM_OBJECT_WLOCK(new_object); 1616 goto retry; 1617 } 1618 1619#if VM_NRESERVLEVEL > 0 1620 /* 1621 * If some of the reservation's allocated pages remain with 1622 * the original object, then transferring the reservation to 1623 * the new object is neither particularly beneficial nor 1624 * particularly harmful as compared to leaving the reservation 1625 * with the original object. If, however, all of the 1626 * reservation's allocated pages are transferred to the new 1627 * object, then transferring the reservation is typically 1628 * beneficial. Determining which of these two cases applies 1629 * would be more costly than unconditionally renaming the 1630 * reservation. 1631 */ 1632 vm_reserv_rename(m, new_object, orig_object, offidxstart); 1633#endif 1634 } 1635 1636 /* 1637 * swap_pager_copy() can sleep, in which case the orig_object's 1638 * and new_object's locks are released and reacquired. 1639 */ 1640 swap_pager_copy(orig_object, new_object, offidxstart, 0); 1641 1642 TAILQ_FOREACH(m, &new_object->memq, listq) 1643 vm_page_xunbusy(m); 1644 1645 vm_object_clear_flag(orig_object, OBJ_SPLIT); 1646 VM_OBJECT_WUNLOCK(orig_object); 1647 VM_OBJECT_WUNLOCK(new_object); 1648 entry->object.vm_object = new_object; 1649 entry->offset = 0LL; 1650 vm_object_deallocate(orig_object); 1651 VM_OBJECT_WLOCK(new_object); 1652} 1653 1654static vm_page_t 1655vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p) 1656{ 1657 vm_object_t backing_object; 1658 1659 VM_OBJECT_ASSERT_WLOCKED(object); 1660 backing_object = object->backing_object; 1661 VM_OBJECT_ASSERT_WLOCKED(backing_object); 1662 1663 KASSERT(p == NULL || p->object == object || p->object == backing_object, 1664 ("invalid ownership %p %p %p", p, object, backing_object)); 1665 /* The page is only NULL when rename fails. */ 1666 if (p == NULL) { 1667 VM_OBJECT_WUNLOCK(object); 1668 VM_OBJECT_WUNLOCK(backing_object); 1669 vm_radix_wait(); 1670 VM_OBJECT_WLOCK(object); 1671 } else if (p->object == object) { 1672 VM_OBJECT_WUNLOCK(backing_object); 1673 if (vm_page_busy_sleep(p, "vmocol", 0)) 1674 VM_OBJECT_WLOCK(object); 1675 } else { 1676 VM_OBJECT_WUNLOCK(object); 1677 if (!vm_page_busy_sleep(p, "vmocol", 0)) 1678 VM_OBJECT_WUNLOCK(backing_object); 1679 VM_OBJECT_WLOCK(object); 1680 } 1681 VM_OBJECT_WLOCK(backing_object); 1682 return (TAILQ_FIRST(&backing_object->memq)); 1683} 1684 1685static bool 1686vm_object_scan_all_shadowed(vm_object_t object) 1687{ 1688 vm_object_t backing_object; 1689 vm_page_t p, pp; 1690 vm_pindex_t backing_offset_index, new_pindex, pi, ps; 1691 1692 VM_OBJECT_ASSERT_WLOCKED(object); 1693 VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 1694 1695 backing_object = object->backing_object; 1696 1697 if ((backing_object->flags & OBJ_ANON) == 0) 1698 return (false); 1699 1700 pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 1701 p = vm_page_find_least(backing_object, pi); 1702 ps = swap_pager_find_least(backing_object, pi); 1703 1704 /* 1705 * Only check pages inside the parent object's range and 1706 * inside the parent object's mapping of the backing object. 1707 */ 1708 for (;; pi++) { 1709 if (p != NULL && p->pindex < pi) 1710 p = TAILQ_NEXT(p, listq); 1711 if (ps < pi) 1712 ps = swap_pager_find_least(backing_object, pi); 1713 if (p == NULL && ps >= backing_object->size) 1714 break; 1715 else if (p == NULL) 1716 pi = ps; 1717 else 1718 pi = MIN(p->pindex, ps); 1719 1720 new_pindex = pi - backing_offset_index; 1721 if (new_pindex >= object->size) 1722 break; 1723 1724 if (p != NULL) { 1725 /* 1726 * If the backing object page is busy a 1727 * grandparent or older page may still be 1728 * undergoing CoW. It is not safe to collapse 1729 * the backing object until it is quiesced. 1730 */ 1731 if (vm_page_tryxbusy(p) == 0) 1732 return (false); 1733 1734 /* 1735 * We raced with the fault handler that left 1736 * newly allocated invalid page on the object 1737 * queue and retried. 1738 */ 1739 if (!vm_page_all_valid(p)) 1740 goto unbusy_ret; 1741 } 1742 1743 /* 1744 * See if the parent has the page or if the parent's object 1745 * pager has the page. If the parent has the page but the page 1746 * is not valid, the parent's object pager must have the page. 1747 * 1748 * If this fails, the parent does not completely shadow the 1749 * object and we might as well give up now. 1750 */ 1751 pp = vm_page_lookup(object, new_pindex); 1752 1753 /* 1754 * The valid check here is stable due to object lock 1755 * being required to clear valid and initiate paging. 1756 * Busy of p disallows fault handler to validate pp. 1757 */ 1758 if ((pp == NULL || vm_page_none_valid(pp)) && 1759 !vm_pager_has_page(object, new_pindex, NULL, NULL)) 1760 goto unbusy_ret; 1761 if (p != NULL) 1762 vm_page_xunbusy(p); 1763 } 1764 return (true); 1765 1766unbusy_ret: 1767 if (p != NULL) 1768 vm_page_xunbusy(p); 1769 return (false); 1770} 1771 1772static void 1773vm_object_collapse_scan(vm_object_t object) 1774{ 1775 vm_object_t backing_object; 1776 vm_page_t next, p, pp; 1777 vm_pindex_t backing_offset_index, new_pindex; 1778 1779 VM_OBJECT_ASSERT_WLOCKED(object); 1780 VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 1781 1782 backing_object = object->backing_object; 1783 backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 1784 1785 /* 1786 * Our scan 1787 */ 1788 for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) { 1789 next = TAILQ_NEXT(p, listq); 1790 new_pindex = p->pindex - backing_offset_index; 1791 1792 /* 1793 * Check for busy page 1794 */ 1795 if (vm_page_tryxbusy(p) == 0) { 1796 next = vm_object_collapse_scan_wait(object, p); 1797 continue; 1798 } 1799 1800 KASSERT(object->backing_object == backing_object, 1801 ("vm_object_collapse_scan: backing object mismatch %p != %p", 1802 object->backing_object, backing_object)); 1803 KASSERT(p->object == backing_object, 1804 ("vm_object_collapse_scan: object mismatch %p != %p", 1805 p->object, backing_object)); 1806 1807 if (p->pindex < backing_offset_index || 1808 new_pindex >= object->size) { 1809 vm_pager_freespace(backing_object, p->pindex, 1); 1810 1811 KASSERT(!pmap_page_is_mapped(p), 1812 ("freeing mapped page %p", p)); 1813 if (vm_page_remove(p)) 1814 vm_page_free(p); 1815 continue; 1816 } 1817 1818 if (!vm_page_all_valid(p)) { 1819 KASSERT(!pmap_page_is_mapped(p), 1820 ("freeing mapped page %p", p)); 1821 if (vm_page_remove(p)) 1822 vm_page_free(p); 1823 continue; 1824 } 1825 1826 pp = vm_page_lookup(object, new_pindex); 1827 if (pp != NULL && vm_page_tryxbusy(pp) == 0) { 1828 vm_page_xunbusy(p); 1829 /* 1830 * The page in the parent is busy and possibly not 1831 * (yet) valid. Until its state is finalized by the 1832 * busy bit owner, we can't tell whether it shadows the 1833 * original page. 1834 */ 1835 next = vm_object_collapse_scan_wait(object, pp); 1836 continue; 1837 } 1838 1839 if (pp != NULL && vm_page_none_valid(pp)) { 1840 /* 1841 * The page was invalid in the parent. Likely placed 1842 * there by an incomplete fault. Just remove and 1843 * ignore. p can replace it. 1844 */ 1845 if (vm_page_remove(pp)) 1846 vm_page_free(pp); 1847 pp = NULL; 1848 } 1849 1850 if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL, 1851 NULL)) { 1852 /* 1853 * The page already exists in the parent OR swap exists 1854 * for this location in the parent. Leave the parent's 1855 * page alone. Destroy the original page from the 1856 * backing object. 1857 */ 1858 vm_pager_freespace(backing_object, p->pindex, 1); 1859 KASSERT(!pmap_page_is_mapped(p), 1860 ("freeing mapped page %p", p)); 1861 if (vm_page_remove(p)) 1862 vm_page_free(p); 1863 if (pp != NULL) 1864 vm_page_xunbusy(pp); 1865 continue; 1866 } 1867 1868 /* 1869 * Page does not exist in parent, rename the page from the 1870 * backing object to the main object. 1871 * 1872 * If the page was mapped to a process, it can remain mapped 1873 * through the rename. vm_page_rename() will dirty the page. 1874 */ 1875 if (vm_page_rename(p, object, new_pindex)) { 1876 vm_page_xunbusy(p); 1877 next = vm_object_collapse_scan_wait(object, NULL); 1878 continue; 1879 } 1880 1881 /* Use the old pindex to free the right page. */ 1882 vm_pager_freespace(backing_object, new_pindex + 1883 backing_offset_index, 1); 1884 1885#if VM_NRESERVLEVEL > 0 1886 /* 1887 * Rename the reservation. 1888 */ 1889 vm_reserv_rename(p, object, backing_object, 1890 backing_offset_index); 1891#endif 1892 vm_page_xunbusy(p); 1893 } 1894 return; 1895} 1896 1897/* 1898 * vm_object_collapse: 1899 * 1900 * Collapse an object with the object backing it. 1901 * Pages in the backing object are moved into the 1902 * parent, and the backing object is deallocated. 1903 */ 1904void 1905vm_object_collapse(vm_object_t object) 1906{ 1907 vm_object_t backing_object, new_backing_object; 1908 1909 VM_OBJECT_ASSERT_WLOCKED(object); 1910 1911 while (TRUE) { 1912 KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON, 1913 ("collapsing invalid object")); 1914 1915 /* 1916 * Wait for the backing_object to finish any pending 1917 * collapse so that the caller sees the shortest possible 1918 * shadow chain. 1919 */ 1920 backing_object = vm_object_backing_collapse_wait(object); 1921 if (backing_object == NULL) 1922 return; 1923 1924 KASSERT(object->ref_count > 0 && 1925 object->ref_count > atomic_load_int(&object->shadow_count), 1926 ("collapse with invalid ref %d or shadow %d count.", 1927 object->ref_count, atomic_load_int(&object->shadow_count))); 1928 KASSERT((backing_object->flags & 1929 (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 1930 ("vm_object_collapse: Backing object already collapsing.")); 1931 KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 1932 ("vm_object_collapse: object is already collapsing.")); 1933 1934 /* 1935 * We know that we can either collapse the backing object if 1936 * the parent is the only reference to it, or (perhaps) have 1937 * the parent bypass the object if the parent happens to shadow 1938 * all the resident pages in the entire backing object. 1939 */ 1940 if (backing_object->ref_count == 1) { 1941 KASSERT(atomic_load_int(&backing_object->shadow_count) 1942 == 1, 1943 ("vm_object_collapse: shadow_count: %d", 1944 atomic_load_int(&backing_object->shadow_count))); 1945 vm_object_pip_add(object, 1); 1946 vm_object_set_flag(object, OBJ_COLLAPSING); 1947 vm_object_pip_add(backing_object, 1); 1948 vm_object_set_flag(backing_object, OBJ_DEAD); 1949 1950 /* 1951 * If there is exactly one reference to the backing 1952 * object, we can collapse it into the parent. 1953 */ 1954 vm_object_collapse_scan(object); 1955 1956 /* 1957 * Move the pager from backing_object to object. 1958 * 1959 * swap_pager_copy() can sleep, in which case the 1960 * backing_object's and object's locks are released and 1961 * reacquired. 1962 */ 1963 swap_pager_copy(backing_object, object, 1964 OFF_TO_IDX(object->backing_object_offset), TRUE); 1965 1966 /* 1967 * Object now shadows whatever backing_object did. 1968 */ 1969 vm_object_clear_flag(object, OBJ_COLLAPSING); 1970 vm_object_backing_transfer(object, backing_object); 1971 object->backing_object_offset += 1972 backing_object->backing_object_offset; 1973 VM_OBJECT_WUNLOCK(object); 1974 vm_object_pip_wakeup(object); 1975 1976 /* 1977 * Discard backing_object. 1978 * 1979 * Since the backing object has no pages, no pager left, 1980 * and no object references within it, all that is 1981 * necessary is to dispose of it. 1982 */ 1983 KASSERT(backing_object->ref_count == 1, ( 1984"backing_object %p was somehow re-referenced during collapse!", 1985 backing_object)); 1986 vm_object_pip_wakeup(backing_object); 1987 (void)refcount_release(&backing_object->ref_count); 1988 umtx_shm_object_terminated(backing_object); 1989 vm_object_terminate(backing_object); 1990 counter_u64_add(object_collapses, 1); 1991 VM_OBJECT_WLOCK(object); 1992 } else { 1993 /* 1994 * If we do not entirely shadow the backing object, 1995 * there is nothing we can do so we give up. 1996 * 1997 * The object lock and backing_object lock must not 1998 * be dropped during this sequence. 1999 */ 2000 if (!vm_object_scan_all_shadowed(object)) { 2001 VM_OBJECT_WUNLOCK(backing_object); 2002 break; 2003 } 2004 2005 /* 2006 * Make the parent shadow the next object in the 2007 * chain. Deallocating backing_object will not remove 2008 * it, since its reference count is at least 2. 2009 */ 2010 vm_object_backing_remove_locked(object); 2011 new_backing_object = backing_object->backing_object; 2012 if (new_backing_object != NULL) { 2013 vm_object_backing_insert_ref(object, 2014 new_backing_object); 2015 object->backing_object_offset += 2016 backing_object->backing_object_offset; 2017 } 2018 2019 /* 2020 * Drop the reference count on backing_object. Since 2021 * its ref_count was at least 2, it will not vanish. 2022 */ 2023 (void)refcount_release(&backing_object->ref_count); 2024 KASSERT(backing_object->ref_count >= 1, ( 2025"backing_object %p was somehow dereferenced during collapse!", 2026 backing_object)); 2027 VM_OBJECT_WUNLOCK(backing_object); 2028 counter_u64_add(object_bypasses, 1); 2029 } 2030 2031 /* 2032 * Try again with this object's new backing object. 2033 */ 2034 } 2035} 2036 2037/* 2038 * vm_object_page_remove: 2039 * 2040 * For the given object, either frees or invalidates each of the 2041 * specified pages. In general, a page is freed. However, if a page is 2042 * wired for any reason other than the existence of a managed, wired 2043 * mapping, then it may be invalidated but not removed from the object. 2044 * Pages are specified by the given range ["start", "end") and the option 2045 * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range 2046 * extends from "start" to the end of the object. If the option 2047 * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the 2048 * specified range are affected. If the option OBJPR_NOTMAPPED is 2049 * specified, then the pages within the specified range must have no 2050 * mappings. Otherwise, if this option is not specified, any mappings to 2051 * the specified pages are removed before the pages are freed or 2052 * invalidated. 2053 * 2054 * In general, this operation should only be performed on objects that 2055 * contain managed pages. There are, however, two exceptions. First, it 2056 * is performed on the kernel and kmem objects by vm_map_entry_delete(). 2057 * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- 2058 * backed pages. In both of these cases, the option OBJPR_CLEANONLY must 2059 * not be specified and the option OBJPR_NOTMAPPED must be specified. 2060 * 2061 * The object must be locked. 2062 */ 2063void 2064vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 2065 int options) 2066{ 2067 vm_page_t p, next; 2068 2069 VM_OBJECT_ASSERT_WLOCKED(object); 2070 KASSERT((object->flags & OBJ_UNMANAGED) == 0 || 2071 (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, 2072 ("vm_object_page_remove: illegal options for object %p", object)); 2073 if (object->resident_page_count == 0) 2074 return; 2075 vm_object_pip_add(object, 1); 2076again: 2077 p = vm_page_find_least(object, start); 2078 2079 /* 2080 * Here, the variable "p" is either (1) the page with the least pindex 2081 * greater than or equal to the parameter "start" or (2) NULL. 2082 */ 2083 for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2084 next = TAILQ_NEXT(p, listq); 2085 2086 /* 2087 * Skip invalid pages if asked to do so. Try to avoid acquiring 2088 * the busy lock, as some consumers rely on this to avoid 2089 * deadlocks. 2090 * 2091 * A thread may concurrently transition the page from invalid to 2092 * valid using only the busy lock, so the result of this check 2093 * is immediately stale. It is up to consumers to handle this, 2094 * for instance by ensuring that all invalid->valid transitions 2095 * happen with a mutex held, as may be possible for a 2096 * filesystem. 2097 */ 2098 if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) 2099 continue; 2100 2101 /* 2102 * If the page is wired for any reason besides the existence 2103 * of managed, wired mappings, then it cannot be freed. For 2104 * example, fictitious pages, which represent device memory, 2105 * are inherently wired and cannot be freed. They can, 2106 * however, be invalidated if the option OBJPR_CLEANONLY is 2107 * not specified. 2108 */ 2109 if (vm_page_tryxbusy(p) == 0) { 2110 if (vm_page_busy_sleep(p, "vmopar", 0)) 2111 VM_OBJECT_WLOCK(object); 2112 goto again; 2113 } 2114 if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) { 2115 vm_page_xunbusy(p); 2116 continue; 2117 } 2118 if (vm_page_wired(p)) { 2119wired: 2120 if ((options & OBJPR_NOTMAPPED) == 0 && 2121 object->ref_count != 0) 2122 pmap_remove_all(p); 2123 if ((options & OBJPR_CLEANONLY) == 0) { 2124 vm_page_invalid(p); 2125 vm_page_undirty(p); 2126 } 2127 vm_page_xunbusy(p); 2128 continue; 2129 } 2130 KASSERT((p->flags & PG_FICTITIOUS) == 0, 2131 ("vm_object_page_remove: page %p is fictitious", p)); 2132 if ((options & OBJPR_CLEANONLY) != 0 && 2133 !vm_page_none_valid(p)) { 2134 if ((options & OBJPR_NOTMAPPED) == 0 && 2135 object->ref_count != 0 && 2136 !vm_page_try_remove_write(p)) 2137 goto wired; 2138 if (p->dirty != 0) { 2139 vm_page_xunbusy(p); 2140 continue; 2141 } 2142 } 2143 if ((options & OBJPR_NOTMAPPED) == 0 && 2144 object->ref_count != 0 && !vm_page_try_remove_all(p)) 2145 goto wired; 2146 vm_page_free(p); 2147 } 2148 vm_object_pip_wakeup(object); 2149 2150 vm_pager_freespace(object, start, (end == 0 ? object->size : end) - 2151 start); 2152} 2153 2154/* 2155 * vm_object_page_noreuse: 2156 * 2157 * For the given object, attempt to move the specified pages to 2158 * the head of the inactive queue. This bypasses regular LRU 2159 * operation and allows the pages to be reused quickly under memory 2160 * pressure. If a page is wired for any reason, then it will not 2161 * be queued. Pages are specified by the range ["start", "end"). 2162 * As a special case, if "end" is zero, then the range extends from 2163 * "start" to the end of the object. 2164 * 2165 * This operation should only be performed on objects that 2166 * contain non-fictitious, managed pages. 2167 * 2168 * The object must be locked. 2169 */ 2170void 2171vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2172{ 2173 vm_page_t p, next; 2174 2175 VM_OBJECT_ASSERT_LOCKED(object); 2176 KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, 2177 ("vm_object_page_noreuse: illegal object %p", object)); 2178 if (object->resident_page_count == 0) 2179 return; 2180 p = vm_page_find_least(object, start); 2181 2182 /* 2183 * Here, the variable "p" is either (1) the page with the least pindex 2184 * greater than or equal to the parameter "start" or (2) NULL. 2185 */ 2186 for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2187 next = TAILQ_NEXT(p, listq); 2188 vm_page_deactivate_noreuse(p); 2189 } 2190} 2191 2192/* 2193 * Populate the specified range of the object with valid pages. Returns 2194 * TRUE if the range is successfully populated and FALSE otherwise. 2195 * 2196 * Note: This function should be optimized to pass a larger array of 2197 * pages to vm_pager_get_pages() before it is applied to a non- 2198 * OBJT_DEVICE object. 2199 * 2200 * The object must be locked. 2201 */ 2202boolean_t 2203vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2204{ 2205 vm_page_t m; 2206 vm_pindex_t pindex; 2207 int rv; 2208 2209 VM_OBJECT_ASSERT_WLOCKED(object); 2210 for (pindex = start; pindex < end; pindex++) { 2211 rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL); 2212 if (rv != VM_PAGER_OK) 2213 break; 2214 2215 /* 2216 * Keep "m" busy because a subsequent iteration may unlock 2217 * the object. 2218 */ 2219 } 2220 if (pindex > start) { 2221 m = vm_page_lookup(object, start); 2222 while (m != NULL && m->pindex < pindex) { 2223 vm_page_xunbusy(m); 2224 m = TAILQ_NEXT(m, listq); 2225 } 2226 } 2227 return (pindex == end); 2228} 2229 2230/* 2231 * Routine: vm_object_coalesce 2232 * Function: Coalesces two objects backing up adjoining 2233 * regions of memory into a single object. 2234 * 2235 * returns TRUE if objects were combined. 2236 * 2237 * NOTE: Only works at the moment if the second object is NULL - 2238 * if it's not, which object do we lock first? 2239 * 2240 * Parameters: 2241 * prev_object First object to coalesce 2242 * prev_offset Offset into prev_object 2243 * prev_size Size of reference to prev_object 2244 * next_size Size of reference to the second object 2245 * reserved Indicator that extension region has 2246 * swap accounted for 2247 * 2248 * Conditions: 2249 * The object must *not* be locked. 2250 */ 2251boolean_t 2252vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, 2253 vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) 2254{ 2255 vm_pindex_t next_pindex; 2256 2257 if (prev_object == NULL) 2258 return (TRUE); 2259 if ((prev_object->flags & OBJ_ANON) == 0) 2260 return (FALSE); 2261 2262 VM_OBJECT_WLOCK(prev_object); 2263 /* 2264 * Try to collapse the object first. 2265 */ 2266 vm_object_collapse(prev_object); 2267 2268 /* 2269 * Can't coalesce if: . more than one reference . paged out . shadows 2270 * another object . has a copy elsewhere (any of which mean that the 2271 * pages not mapped to prev_entry may be in use anyway) 2272 */ 2273 if (prev_object->backing_object != NULL) { 2274 VM_OBJECT_WUNLOCK(prev_object); 2275 return (FALSE); 2276 } 2277 2278 prev_size >>= PAGE_SHIFT; 2279 next_size >>= PAGE_SHIFT; 2280 next_pindex = OFF_TO_IDX(prev_offset) + prev_size; 2281 2282 if (prev_object->ref_count > 1 && 2283 prev_object->size != next_pindex && 2284 (prev_object->flags & OBJ_ONEMAPPING) == 0) { 2285 VM_OBJECT_WUNLOCK(prev_object); 2286 return (FALSE); 2287 } 2288 2289 /* 2290 * Account for the charge. 2291 */ 2292 if (prev_object->cred != NULL) { 2293 /* 2294 * If prev_object was charged, then this mapping, 2295 * although not charged now, may become writable 2296 * later. Non-NULL cred in the object would prevent 2297 * swap reservation during enabling of the write 2298 * access, so reserve swap now. Failed reservation 2299 * cause allocation of the separate object for the map 2300 * entry, and swap reservation for this entry is 2301 * managed in appropriate time. 2302 */ 2303 if (!reserved && !swap_reserve_by_cred(ptoa(next_size), 2304 prev_object->cred)) { 2305 VM_OBJECT_WUNLOCK(prev_object); 2306 return (FALSE); 2307 } 2308 prev_object->charge += ptoa(next_size); 2309 } 2310 2311 /* 2312 * Remove any pages that may still be in the object from a previous 2313 * deallocation. 2314 */ 2315 if (next_pindex < prev_object->size) { 2316 vm_object_page_remove(prev_object, next_pindex, next_pindex + 2317 next_size, 0); 2318#if 0 2319 if (prev_object->cred != NULL) { 2320 KASSERT(prev_object->charge >= 2321 ptoa(prev_object->size - next_pindex), 2322 ("object %p overcharged 1 %jx %jx", prev_object, 2323 (uintmax_t)next_pindex, (uintmax_t)next_size)); 2324 prev_object->charge -= ptoa(prev_object->size - 2325 next_pindex); 2326 } 2327#endif 2328 } 2329 2330 /* 2331 * Extend the object if necessary. 2332 */ 2333 if (next_pindex + next_size > prev_object->size) 2334 prev_object->size = next_pindex + next_size; 2335 2336 VM_OBJECT_WUNLOCK(prev_object); 2337 return (TRUE); 2338} 2339 2340void 2341vm_object_set_writeable_dirty_(vm_object_t object) 2342{ 2343 atomic_add_int(&object->generation, 1); 2344} 2345 2346bool 2347vm_object_mightbedirty_(vm_object_t object) 2348{ 2349 return (object->generation != object->cleangeneration); 2350} 2351 2352/* 2353 * vm_object_unwire: 2354 * 2355 * For each page offset within the specified range of the given object, 2356 * find the highest-level page in the shadow chain and unwire it. A page 2357 * must exist at every page offset, and the highest-level page must be 2358 * wired. 2359 */ 2360void 2361vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, 2362 uint8_t queue) 2363{ 2364 vm_object_t tobject, t1object; 2365 vm_page_t m, tm; 2366 vm_pindex_t end_pindex, pindex, tpindex; 2367 int depth, locked_depth; 2368 2369 KASSERT((offset & PAGE_MASK) == 0, 2370 ("vm_object_unwire: offset is not page aligned")); 2371 KASSERT((length & PAGE_MASK) == 0, 2372 ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); 2373 /* The wired count of a fictitious page never changes. */ 2374 if ((object->flags & OBJ_FICTITIOUS) != 0) 2375 return; 2376 pindex = OFF_TO_IDX(offset); 2377 end_pindex = pindex + atop(length); 2378again: 2379 locked_depth = 1; 2380 VM_OBJECT_RLOCK(object); 2381 m = vm_page_find_least(object, pindex); 2382 while (pindex < end_pindex) { 2383 if (m == NULL || pindex < m->pindex) { 2384 /* 2385 * The first object in the shadow chain doesn't 2386 * contain a page at the current index. Therefore, 2387 * the page must exist in a backing object. 2388 */ 2389 tobject = object; 2390 tpindex = pindex; 2391 depth = 0; 2392 do { 2393 tpindex += 2394 OFF_TO_IDX(tobject->backing_object_offset); 2395 tobject = tobject->backing_object; 2396 KASSERT(tobject != NULL, 2397 ("vm_object_unwire: missing page")); 2398 if ((tobject->flags & OBJ_FICTITIOUS) != 0) 2399 goto next_page; 2400 depth++; 2401 if (depth == locked_depth) { 2402 locked_depth++; 2403 VM_OBJECT_RLOCK(tobject); 2404 } 2405 } while ((tm = vm_page_lookup(tobject, tpindex)) == 2406 NULL); 2407 } else { 2408 tm = m; 2409 m = TAILQ_NEXT(m, listq); 2410 } 2411 if (vm_page_trysbusy(tm) == 0) { 2412 for (tobject = object; locked_depth >= 1; 2413 locked_depth--) { 2414 t1object = tobject->backing_object; 2415 if (tm->object != tobject) 2416 VM_OBJECT_RUNLOCK(tobject); 2417 tobject = t1object; 2418 } 2419 tobject = tm->object; 2420 if (!vm_page_busy_sleep(tm, "unwbo", 2421 VM_ALLOC_IGN_SBUSY)) 2422 VM_OBJECT_RUNLOCK(tobject); 2423 goto again; 2424 } 2425 vm_page_unwire(tm, queue); 2426 vm_page_sunbusy(tm); 2427next_page: 2428 pindex++; 2429 } 2430 /* Release the accumulated object locks. */ 2431 for (tobject = object; locked_depth >= 1; locked_depth--) { 2432 t1object = tobject->backing_object; 2433 VM_OBJECT_RUNLOCK(tobject); 2434 tobject = t1object; 2435 } 2436} 2437 2438/* 2439 * Return the vnode for the given object, or NULL if none exists. 2440 * For tmpfs objects, the function may return NULL if there is 2441 * no vnode allocated at the time of the call. 2442 */ 2443struct vnode * 2444vm_object_vnode(vm_object_t object) 2445{ 2446 struct vnode *vp; 2447 2448 VM_OBJECT_ASSERT_LOCKED(object); 2449 vm_pager_getvp(object, &vp, NULL); 2450 return (vp); 2451} 2452 2453/* 2454 * Busy the vm object. This prevents new pages belonging to the object from 2455 * becoming busy. Existing pages persist as busy. Callers are responsible 2456 * for checking page state before proceeding. 2457 */ 2458void 2459vm_object_busy(vm_object_t obj) 2460{ 2461 2462 VM_OBJECT_ASSERT_LOCKED(obj); 2463 2464 blockcount_acquire(&obj->busy, 1); 2465 /* The fence is required to order loads of page busy. */ 2466 atomic_thread_fence_acq_rel(); 2467} 2468 2469void 2470vm_object_unbusy(vm_object_t obj) 2471{ 2472 2473 blockcount_release(&obj->busy, 1); 2474} 2475 2476void 2477vm_object_busy_wait(vm_object_t obj, const char *wmesg) 2478{ 2479 2480 VM_OBJECT_ASSERT_UNLOCKED(obj); 2481 2482 (void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM); 2483} 2484 2485/* 2486 * This function aims to determine if the object is mapped, 2487 * specifically, if it is referenced by a vm_map_entry. Because 2488 * objects occasionally acquire transient references that do not 2489 * represent a mapping, the method used here is inexact. However, it 2490 * has very low overhead and is good enough for the advisory 2491 * vm.vmtotal sysctl. 2492 */ 2493bool 2494vm_object_is_active(vm_object_t obj) 2495{ 2496 2497 return (obj->ref_count > atomic_load_int(&obj->shadow_count)); 2498} 2499 2500static int 2501vm_object_list_handler(struct sysctl_req *req, bool swap_only) 2502{ 2503 struct kinfo_vmobject *kvo; 2504 char *fullpath, *freepath; 2505 struct vnode *vp; 2506 struct vattr va; 2507 vm_object_t obj; 2508 vm_page_t m; 2509 u_long sp; 2510 int count, error; 2511 bool want_path; 2512 2513 if (req->oldptr == NULL) { 2514 /* 2515 * If an old buffer has not been provided, generate an 2516 * estimate of the space needed for a subsequent call. 2517 */ 2518 mtx_lock(&vm_object_list_mtx); 2519 count = 0; 2520 TAILQ_FOREACH(obj, &vm_object_list, object_list) { 2521 if (obj->type == OBJT_DEAD) 2522 continue; 2523 count++; 2524 } 2525 mtx_unlock(&vm_object_list_mtx); 2526 return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * 2527 count * 11 / 10)); 2528 } 2529 2530 want_path = !(swap_only || jailed(curthread->td_ucred)); 2531 kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK | M_ZERO); 2532 error = 0; 2533 2534 /* 2535 * VM objects are type stable and are never removed from the 2536 * list once added. This allows us to safely read obj->object_list 2537 * after reacquiring the VM object lock. 2538 */ 2539 mtx_lock(&vm_object_list_mtx); 2540 TAILQ_FOREACH(obj, &vm_object_list, object_list) { 2541 if (obj->type == OBJT_DEAD || 2542 (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) 2543 continue; 2544 VM_OBJECT_RLOCK(obj); 2545 if (obj->type == OBJT_DEAD || 2546 (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) { 2547 VM_OBJECT_RUNLOCK(obj); 2548 continue; 2549 } 2550 mtx_unlock(&vm_object_list_mtx); 2551 kvo->kvo_size = ptoa(obj->size); 2552 kvo->kvo_resident = obj->resident_page_count; 2553 kvo->kvo_ref_count = obj->ref_count; 2554 kvo->kvo_shadow_count = atomic_load_int(&obj->shadow_count); 2555 kvo->kvo_memattr = obj->memattr; 2556 kvo->kvo_active = 0; 2557 kvo->kvo_inactive = 0; 2558 if (!swap_only) { 2559 TAILQ_FOREACH(m, &obj->memq, listq) { 2560 /* 2561 * A page may belong to the object but be 2562 * dequeued and set to PQ_NONE while the 2563 * object lock is not held. This makes the 2564 * reads of m->queue below racy, and we do not 2565 * count pages set to PQ_NONE. However, this 2566 * sysctl is only meant to give an 2567 * approximation of the system anyway. 2568 */ 2569 if (m->a.queue == PQ_ACTIVE) 2570 kvo->kvo_active++; 2571 else if (m->a.queue == PQ_INACTIVE) 2572 kvo->kvo_inactive++; 2573 } 2574 } 2575 2576 kvo->kvo_vn_fileid = 0; 2577 kvo->kvo_vn_fsid = 0; 2578 kvo->kvo_vn_fsid_freebsd11 = 0; 2579 freepath = NULL; 2580 fullpath = ""; 2581 vp = NULL; 2582 kvo->kvo_type = vm_object_kvme_type(obj, want_path ? &vp : 2583 NULL); 2584 if (vp != NULL) { 2585 vref(vp); 2586 } else if ((obj->flags & OBJ_ANON) != 0) { 2587 MPASS(kvo->kvo_type == KVME_TYPE_SWAP); 2588 kvo->kvo_me = (uintptr_t)obj; 2589 /* tmpfs objs are reported as vnodes */ 2590 kvo->kvo_backing_obj = (uintptr_t)obj->backing_object; 2591 sp = swap_pager_swapped_pages(obj); 2592 kvo->kvo_swapped = sp > UINT32_MAX ? UINT32_MAX : sp; 2593 } 2594 VM_OBJECT_RUNLOCK(obj); 2595 if (vp != NULL) { 2596 vn_fullpath(vp, &fullpath, &freepath); 2597 vn_lock(vp, LK_SHARED | LK_RETRY); 2598 if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { 2599 kvo->kvo_vn_fileid = va.va_fileid; 2600 kvo->kvo_vn_fsid = va.va_fsid; 2601 kvo->kvo_vn_fsid_freebsd11 = va.va_fsid; 2602 /* truncate */ 2603 } 2604 vput(vp); 2605 } 2606 2607 strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path)); 2608 free(freepath, M_TEMP); 2609 2610 /* Pack record size down */ 2611 kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) 2612 + strlen(kvo->kvo_path) + 1; 2613 kvo->kvo_structsize = roundup(kvo->kvo_structsize, 2614 sizeof(uint64_t)); 2615 error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize); 2616 maybe_yield(); 2617 mtx_lock(&vm_object_list_mtx); 2618 if (error) 2619 break; 2620 } 2621 mtx_unlock(&vm_object_list_mtx); 2622 free(kvo, M_TEMP); 2623 return (error); 2624} 2625 2626static int 2627sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) 2628{ 2629 return (vm_object_list_handler(req, false)); 2630} 2631 2632SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | 2633 CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", 2634 "List of VM objects"); 2635 2636static int 2637sysctl_vm_object_list_swap(SYSCTL_HANDLER_ARGS) 2638{ 2639 return (vm_object_list_handler(req, true)); 2640} 2641 2642/* 2643 * This sysctl returns list of the anonymous or swap objects. Intent 2644 * is to provide stripped optimized list useful to analyze swap use. 2645 * Since technically non-swap (default) objects participate in the 2646 * shadow chains, and are converted to swap type as needed by swap 2647 * pager, we must report them. 2648 */ 2649SYSCTL_PROC(_vm, OID_AUTO, swap_objects, 2650 CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, 2651 sysctl_vm_object_list_swap, "S,kinfo_vmobject", 2652 "List of swap VM objects"); 2653 2654#include "opt_ddb.h" 2655#ifdef DDB 2656#include <sys/kernel.h> 2657 2658#include <sys/cons.h> 2659 2660#include <ddb/ddb.h> 2661 2662static int 2663_vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) 2664{ 2665 vm_map_t tmpm; 2666 vm_map_entry_t tmpe; 2667 vm_object_t obj; 2668 2669 if (map == 0) 2670 return 0; 2671 2672 if (entry == 0) { 2673 VM_MAP_ENTRY_FOREACH(tmpe, map) { 2674 if (_vm_object_in_map(map, object, tmpe)) { 2675 return 1; 2676 } 2677 } 2678 } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 2679 tmpm = entry->object.sub_map; 2680 VM_MAP_ENTRY_FOREACH(tmpe, tmpm) { 2681 if (_vm_object_in_map(tmpm, object, tmpe)) { 2682 return 1; 2683 } 2684 } 2685 } else if ((obj = entry->object.vm_object) != NULL) { 2686 for (; obj; obj = obj->backing_object) 2687 if (obj == object) { 2688 return 1; 2689 } 2690 } 2691 return 0; 2692} 2693 2694static int 2695vm_object_in_map(vm_object_t object) 2696{ 2697 struct proc *p; 2698 2699 /* sx_slock(&allproc_lock); */ 2700 FOREACH_PROC_IN_SYSTEM(p) { 2701 if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 2702 continue; 2703 if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 2704 /* sx_sunlock(&allproc_lock); */ 2705 return 1; 2706 } 2707 } 2708 /* sx_sunlock(&allproc_lock); */ 2709 if (_vm_object_in_map(kernel_map, object, 0)) 2710 return 1; 2711 return 0; 2712} 2713 2714DB_SHOW_COMMAND_FLAGS(vmochk, vm_object_check, DB_CMD_MEMSAFE) 2715{ 2716 vm_object_t object; 2717 2718 /* 2719 * make sure that internal objs are in a map somewhere 2720 * and none have zero ref counts. 2721 */ 2722 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2723 if ((object->flags & OBJ_ANON) != 0) { 2724 if (object->ref_count == 0) { 2725 db_printf("vmochk: internal obj has zero ref count: %ld\n", 2726 (long)object->size); 2727 } 2728 if (!vm_object_in_map(object)) { 2729 db_printf( 2730 "vmochk: internal obj is not in a map: " 2731 "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 2732 object->ref_count, (u_long)object->size, 2733 (u_long)object->size, 2734 (void *)object->backing_object); 2735 } 2736 } 2737 if (db_pager_quit) 2738 return; 2739 } 2740} 2741 2742/* 2743 * vm_object_print: [ debug ] 2744 */ 2745DB_SHOW_COMMAND(object, vm_object_print_static) 2746{ 2747 /* XXX convert args. */ 2748 vm_object_t object = (vm_object_t)addr; 2749 boolean_t full = have_addr; 2750 2751 vm_page_t p; 2752 2753 /* XXX count is an (unused) arg. Avoid shadowing it. */ 2754#define count was_count 2755 2756 int count; 2757 2758 if (object == NULL) 2759 return; 2760 2761 db_iprintf( 2762 "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", 2763 object, (int)object->type, (uintmax_t)object->size, 2764 object->resident_page_count, object->ref_count, object->flags, 2765 object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); 2766 db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", 2767 atomic_load_int(&object->shadow_count), 2768 object->backing_object ? object->backing_object->ref_count : 0, 2769 object->backing_object, (uintmax_t)object->backing_object_offset); 2770 2771 if (!full) 2772 return; 2773 2774 db_indent += 2; 2775 count = 0; 2776 TAILQ_FOREACH(p, &object->memq, listq) { 2777 if (count == 0) 2778 db_iprintf("memory:="); 2779 else if (count == 6) { 2780 db_printf("\n"); 2781 db_iprintf(" ..."); 2782 count = 0; 2783 } else 2784 db_printf(","); 2785 count++; 2786 2787 db_printf("(off=0x%jx,page=0x%jx)", 2788 (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); 2789 2790 if (db_pager_quit) 2791 break; 2792 } 2793 if (count != 0) 2794 db_printf("\n"); 2795 db_indent -= 2; 2796} 2797 2798/* XXX. */ 2799#undef count 2800 2801/* XXX need this non-static entry for calling from vm_map_print. */ 2802void 2803vm_object_print( 2804 /* db_expr_t */ long addr, 2805 boolean_t have_addr, 2806 /* db_expr_t */ long count, 2807 char *modif) 2808{ 2809 vm_object_print_static(addr, have_addr, count, modif); 2810} 2811 2812DB_SHOW_COMMAND_FLAGS(vmopag, vm_object_print_pages, DB_CMD_MEMSAFE) 2813{ 2814 vm_object_t object; 2815 vm_pindex_t fidx; 2816 vm_paddr_t pa; 2817 vm_page_t m, prev_m; 2818 int rcount; 2819 2820 TAILQ_FOREACH(object, &vm_object_list, object_list) { 2821 db_printf("new object: %p\n", (void *)object); 2822 if (db_pager_quit) 2823 return; 2824 2825 rcount = 0; 2826 fidx = 0; 2827 pa = -1; 2828 TAILQ_FOREACH(m, &object->memq, listq) { 2829 if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && 2830 prev_m->pindex + 1 != m->pindex) { 2831 if (rcount) { 2832 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2833 (long)fidx, rcount, (long)pa); 2834 if (db_pager_quit) 2835 return; 2836 rcount = 0; 2837 } 2838 } 2839 if (rcount && 2840 (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 2841 ++rcount; 2842 continue; 2843 } 2844 if (rcount) { 2845 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2846 (long)fidx, rcount, (long)pa); 2847 if (db_pager_quit) 2848 return; 2849 } 2850 fidx = m->pindex; 2851 pa = VM_PAGE_TO_PHYS(m); 2852 rcount = 1; 2853 } 2854 if (rcount) { 2855 db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 2856 (long)fidx, rcount, (long)pa); 2857 if (db_pager_quit) 2858 return; 2859 } 2860 } 2861} 2862#endif /* DDB */ 2863