160727d8bSWarner Losh /*- 2796df753SPedro F. Giffuni * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 351369649SPedro F. Giffuni * 4df8bae1dSRodney W. Grimes * Copyright (c) 1991, 1993 5df8bae1dSRodney W. Grimes * The Regents of the University of California. All rights reserved. 6df8bae1dSRodney W. Grimes * 7df8bae1dSRodney W. Grimes * This code is derived from software contributed to Berkeley by 8df8bae1dSRodney W. Grimes * The Mach Operating System project at Carnegie-Mellon University. 9df8bae1dSRodney W. Grimes * 10df8bae1dSRodney W. Grimes * Redistribution and use in source and binary forms, with or without 11df8bae1dSRodney W. Grimes * modification, are permitted provided that the following conditions 12df8bae1dSRodney W. Grimes * are met: 13df8bae1dSRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 14df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer. 15df8bae1dSRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 16df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 17df8bae1dSRodney W. Grimes * documentation and/or other materials provided with the distribution. 18fbbd9655SWarner Losh * 3. Neither the name of the University nor the names of its contributors 19df8bae1dSRodney W. Grimes * may be used to endorse or promote products derived from this software 20df8bae1dSRodney W. Grimes * without specific prior written permission. 21df8bae1dSRodney W. Grimes * 22df8bae1dSRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23df8bae1dSRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24df8bae1dSRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25df8bae1dSRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26df8bae1dSRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27df8bae1dSRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28df8bae1dSRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29df8bae1dSRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30df8bae1dSRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31df8bae1dSRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32df8bae1dSRodney W. Grimes * SUCH DAMAGE. 33df8bae1dSRodney W. Grimes * 343c4dd356SDavid Greenman * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 35df8bae1dSRodney W. Grimes * 36df8bae1dSRodney W. Grimes * 37df8bae1dSRodney W. Grimes * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38df8bae1dSRodney W. Grimes * All rights reserved. 39df8bae1dSRodney W. Grimes * 40df8bae1dSRodney W. Grimes * Authors: Avadis Tevanian, Jr., Michael Wayne Young 41df8bae1dSRodney W. Grimes * 42df8bae1dSRodney W. Grimes * Permission to use, copy, modify and distribute this software and 43df8bae1dSRodney W. Grimes * its documentation is hereby granted, provided that both the copyright 44df8bae1dSRodney W. Grimes * notice and this permission notice appear in all copies of the 45df8bae1dSRodney W. Grimes * software, derivative works or modified versions, and any portions 46df8bae1dSRodney W. Grimes * thereof, and that both notices appear in supporting documentation. 47df8bae1dSRodney W. Grimes * 48df8bae1dSRodney W. Grimes * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 49df8bae1dSRodney W. Grimes * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 50df8bae1dSRodney W. Grimes * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 51df8bae1dSRodney W. Grimes * 52df8bae1dSRodney W. Grimes * Carnegie Mellon requests users of this software to return to 53df8bae1dSRodney W. Grimes * 54df8bae1dSRodney W. Grimes * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 55df8bae1dSRodney W. Grimes * School of Computer Science 56df8bae1dSRodney W. Grimes * Carnegie Mellon University 57df8bae1dSRodney W. Grimes * Pittsburgh PA 15213-3890 58df8bae1dSRodney W. Grimes * 59df8bae1dSRodney W. Grimes * any improvements or extensions that they make and grant Carnegie the 60df8bae1dSRodney W. Grimes * rights to redistribute these changes. 61df8bae1dSRodney W. Grimes */ 62df8bae1dSRodney W. Grimes 63df8bae1dSRodney W. Grimes /* 64df8bae1dSRodney W. Grimes * Virtual memory object module. 65df8bae1dSRodney W. Grimes */ 66df8bae1dSRodney W. Grimes 67874651b1SDavid E. O'Brien #include <sys/cdefs.h> 68874651b1SDavid E. O'Brien __FBSDID("$FreeBSD$"); 69874651b1SDavid E. O'Brien 70f8a47341SAlan Cox #include "opt_vm.h" 71f8a47341SAlan Cox 72df8bae1dSRodney W. Grimes #include <sys/param.h> 73df8bae1dSRodney W. Grimes #include <sys/systm.h> 74c99d0c58SMark Johnston #include <sys/blockcount.h> 753f289c3fSJeff Roberson #include <sys/cpuset.h> 76ecfbddf0SKonstantin Belousov #include <sys/limits.h> 77fb919e4dSMark Murray #include <sys/lock.h> 78867a482dSJohn Dyson #include <sys/mman.h> 79cf2819ccSJohn Dyson #include <sys/mount.h> 80b9b7a4beSMatthew Dillon #include <sys/kernel.h> 81f425ab8eSKonstantin Belousov #include <sys/pctrie.h> 82b9b7a4beSMatthew Dillon #include <sys/sysctl.h> 831b367556SJason Evans #include <sys/mutex.h> 84fb919e4dSMark Murray #include <sys/proc.h> /* for curproc, pageproc */ 85cf27e0d1SJeff Roberson #include <sys/refcount.h> 86fb919e4dSMark Murray #include <sys/socket.h> 873364c323SKonstantin Belousov #include <sys/resourcevar.h> 88205be21dSJeff Roberson #include <sys/refcount.h> 8989f6b863SAttilio Rao #include <sys/rwlock.h> 90ff87ae35SJohn Baldwin #include <sys/user.h> 91fb919e4dSMark Murray #include <sys/vnode.h> 92fb919e4dSMark Murray #include <sys/vmmeter.h> 931005a129SJohn Baldwin #include <sys/sx.h> 94df8bae1dSRodney W. Grimes 95df8bae1dSRodney W. Grimes #include <vm/vm.h> 96efeaf95aSDavid Greenman #include <vm/vm_param.h> 97efeaf95aSDavid Greenman #include <vm/pmap.h> 98efeaf95aSDavid Greenman #include <vm/vm_map.h> 99efeaf95aSDavid Greenman #include <vm/vm_object.h> 100df8bae1dSRodney W. Grimes #include <vm/vm_page.h> 10126f9a767SRodney W. Grimes #include <vm/vm_pageout.h> 1020d94caffSDavid Greenman #include <vm/vm_pager.h> 103e2068d0bSJeff Roberson #include <vm/vm_phys.h> 104e2068d0bSJeff Roberson #include <vm/vm_pagequeue.h> 10505f0fdd2SPoul-Henning Kamp #include <vm/swap_pager.h> 106a1f6d91cSDavid Greenman #include <vm/vm_kern.h> 107efeaf95aSDavid Greenman #include <vm/vm_extern.h> 108774d251dSAttilio Rao #include <vm/vm_radix.h> 109f8a47341SAlan Cox #include <vm/vm_reserv.h> 110670d17b5SJeff Roberson #include <vm/uma.h> 11126f9a767SRodney W. Grimes 112c53f7aceSDag-Erling Smørgrav static int old_msync; 113c53f7aceSDag-Erling Smørgrav SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, 114c53f7aceSDag-Erling Smørgrav "Use old (insecure) msync behavior"); 115c53f7aceSDag-Erling Smørgrav 116757216f3SKonstantin Belousov static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, 11767d0e293SJeff Roberson int pagerflags, int flags, boolean_t *allclean, 118126d6082SKonstantin Belousov boolean_t *eio); 1193280870dSKonstantin Belousov static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, 12067d0e293SJeff Roberson boolean_t *allclean); 12151b867e5SJeff Roberson static void vm_object_backing_remove(vm_object_t object); 122f6b04d2bSDavid Greenman 123df8bae1dSRodney W. Grimes /* 124df8bae1dSRodney W. Grimes * Virtual memory objects maintain the actual data 125df8bae1dSRodney W. Grimes * associated with allocated virtual memory. A given 126df8bae1dSRodney W. Grimes * page of memory exists within exactly one object. 127df8bae1dSRodney W. Grimes * 128df8bae1dSRodney W. Grimes * An object is only deallocated when all "references" 129df8bae1dSRodney W. Grimes * are given up. Only one "reference" to a given 130df8bae1dSRodney W. Grimes * region of an object should be writeable. 131df8bae1dSRodney W. Grimes * 132df8bae1dSRodney W. Grimes * Associated with each object is a list of all resident 133df8bae1dSRodney W. Grimes * memory pages belonging to that object; this list is 134df8bae1dSRodney W. Grimes * maintained by the "vm_page" module, and locked by the object's 135df8bae1dSRodney W. Grimes * lock. 136df8bae1dSRodney W. Grimes * 137df8bae1dSRodney W. Grimes * Each object also records a "pager" routine which is 138df8bae1dSRodney W. Grimes * used to retrieve (and store) pages to the proper backing 139df8bae1dSRodney W. Grimes * storage. In addition, objects may be backed by other 140df8bae1dSRodney W. Grimes * objects from which they were virtual-copied. 141df8bae1dSRodney W. Grimes * 142df8bae1dSRodney W. Grimes * The only items within the object structure which are 143df8bae1dSRodney W. Grimes * modified after time of creation are: 144df8bae1dSRodney W. Grimes * reference count locked by object's lock 145df8bae1dSRodney W. Grimes * pager routine locked by object's lock 146df8bae1dSRodney W. Grimes * 147df8bae1dSRodney W. Grimes */ 148df8bae1dSRodney W. Grimes 14928f8db14SBruce Evans struct object_q vm_object_list; 150a5698387SAlan Cox struct mtx vm_object_list_mtx; /* lock for object list and count */ 151cccf11b8SAlan Cox 152cccf11b8SAlan Cox struct vm_object kernel_object_store; 153df8bae1dSRodney W. Grimes 1547029da5cSPawel Biernacki static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 1556472ac3dSEd Schouten "VM object stats"); 156604c2bbcSAlan Cox 157d869a17eSMark Johnston static COUNTER_U64_DEFINE_EARLY(object_collapses); 15811542376SAlan Cox SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, 15911542376SAlan Cox &object_collapses, 16011542376SAlan Cox "VM object collapses"); 161604c2bbcSAlan Cox 162d869a17eSMark Johnston static COUNTER_U64_DEFINE_EARLY(object_bypasses); 16311542376SAlan Cox SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, 16411542376SAlan Cox &object_bypasses, 16511542376SAlan Cox "VM object bypasses"); 16611542376SAlan Cox 167d869a17eSMark Johnston static COUNTER_U64_DEFINE_EARLY(object_collapse_waits); 16898087a06SJeff Roberson SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapse_waits, CTLFLAG_RD, 16998087a06SJeff Roberson &object_collapse_waits, 17098087a06SJeff Roberson "Number of sleeps for collapse"); 17198087a06SJeff Roberson 172670d17b5SJeff Roberson static uma_zone_t obj_zone; 1738355f576SJeff Roberson 174b23f72e9SBrian Feldman static int vm_object_zinit(void *mem, int size, int flags); 1758355f576SJeff Roberson 1768355f576SJeff Roberson #ifdef INVARIANTS 1778355f576SJeff Roberson static void vm_object_zdtor(void *mem, int size, void *arg); 1788355f576SJeff Roberson 1798355f576SJeff Roberson static void 1808355f576SJeff Roberson vm_object_zdtor(void *mem, int size, void *arg) 1818355f576SJeff Roberson { 1828355f576SJeff Roberson vm_object_t object; 1838355f576SJeff Roberson 1848355f576SJeff Roberson object = (vm_object_t)mem; 185e735691bSJohn Baldwin KASSERT(object->ref_count == 0, 186e735691bSJohn Baldwin ("object %p ref_count = %d", object, object->ref_count)); 18743186e53SAlan Cox KASSERT(TAILQ_EMPTY(&object->memq), 188198da1b2SAttilio Rao ("object %p has resident pages in its memq", object)); 189774d251dSAttilio Rao KASSERT(vm_radix_is_empty(&object->rtree), 190774d251dSAttilio Rao ("object %p has resident pages in its trie", object)); 191f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 192f8a47341SAlan Cox KASSERT(LIST_EMPTY(&object->rvq), 193f8a47341SAlan Cox ("object %p has reservations", 194f8a47341SAlan Cox object)); 195f8a47341SAlan Cox #endif 196c99d0c58SMark Johnston KASSERT(!vm_object_busied(object), 197c99d0c58SMark Johnston ("object %p busy = %d", object, blockcount_read(&object->busy))); 1988355f576SJeff Roberson KASSERT(object->resident_page_count == 0, 1998355f576SJeff Roberson ("object %p resident_page_count = %d", 2008355f576SJeff Roberson object, object->resident_page_count)); 2018355f576SJeff Roberson KASSERT(object->shadow_count == 0, 2028355f576SJeff Roberson ("object %p shadow_count = %d", 2038355f576SJeff Roberson object, object->shadow_count)); 204e735691bSJohn Baldwin KASSERT(object->type == OBJT_DEAD, 205e735691bSJohn Baldwin ("object %p has non-dead type %d", 206e735691bSJohn Baldwin object, object->type)); 2078355f576SJeff Roberson } 2088355f576SJeff Roberson #endif 2098355f576SJeff Roberson 210b23f72e9SBrian Feldman static int 211b23f72e9SBrian Feldman vm_object_zinit(void *mem, int size, int flags) 2128355f576SJeff Roberson { 2138355f576SJeff Roberson vm_object_t object; 2148355f576SJeff Roberson 2158355f576SJeff Roberson object = (vm_object_t)mem; 216777a36c5SAlan Cox rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW); 2178355f576SJeff Roberson 2188355f576SJeff Roberson /* These are true for any object that has been freed */ 219e735691bSJohn Baldwin object->type = OBJT_DEAD; 220cd1241fbSKonstantin Belousov vm_radix_init(&object->rtree); 22151df5321SJeff Roberson refcount_init(&object->ref_count, 0); 222c99d0c58SMark Johnston blockcount_init(&object->paging_in_progress); 223c99d0c58SMark Johnston blockcount_init(&object->busy); 2248355f576SJeff Roberson object->resident_page_count = 0; 2258355f576SJeff Roberson object->shadow_count = 0; 226f425ab8eSKonstantin Belousov object->flags = OBJ_DEAD; 227e735691bSJohn Baldwin 228e735691bSJohn Baldwin mtx_lock(&vm_object_list_mtx); 229e735691bSJohn Baldwin TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); 230e735691bSJohn Baldwin mtx_unlock(&vm_object_list_mtx); 231b23f72e9SBrian Feldman return (0); 2328355f576SJeff Roberson } 233df8bae1dSRodney W. Grimes 234a4915c21SAttilio Rao static void 23563967687SJeff Roberson _vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags, 23667388836SKonstantin Belousov vm_object_t object, void *handle) 237df8bae1dSRodney W. Grimes { 2380cddd8f0SMatthew Dillon 239df8bae1dSRodney W. Grimes TAILQ_INIT(&object->memq); 2401c500307SAlan Cox LIST_INIT(&object->shadow_head); 241a1f6d91cSDavid Greenman 24224a1cce3SDavid Greenman object->type = type; 2434b8365d7SKonstantin Belousov object->flags = flags; 2444b8365d7SKonstantin Belousov if ((flags & OBJ_SWAP) != 0) 245f425ab8eSKonstantin Belousov pctrie_init(&object->un_pager.swp.swp_blks); 246f425ab8eSKonstantin Belousov 247f425ab8eSKonstantin Belousov /* 248f425ab8eSKonstantin Belousov * Ensure that swap_pager_swapoff() iteration over object_list 249f425ab8eSKonstantin Belousov * sees up to date type and pctrie head if it observed 250f425ab8eSKonstantin Belousov * non-dead object. 251f425ab8eSKonstantin Belousov */ 252f425ab8eSKonstantin Belousov atomic_thread_fence_rel(); 253f425ab8eSKonstantin Belousov 25463967687SJeff Roberson object->pg_color = 0; 255df8bae1dSRodney W. Grimes object->size = size; 2564c29d2deSMark Johnston object->domain.dr_policy = NULL; 257b881da26SAlan Cox object->generation = 1; 25867d0e293SJeff Roberson object->cleangeneration = 1; 25951df5321SJeff Roberson refcount_init(&object->ref_count, 1); 2603153e878SAlan Cox object->memattr = VM_MEMATTR_DEFAULT; 261ef694c1aSEdward Tomasz Napierala object->cred = NULL; 2623364c323SKonstantin Belousov object->charge = 0; 26367388836SKonstantin Belousov object->handle = handle; 26424a1cce3SDavid Greenman object->backing_object = NULL; 265a316d390SJohn Dyson object->backing_object_offset = (vm_ooffset_t) 0; 266f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 267f8a47341SAlan Cox LIST_INIT(&object->rvq); 268f8a47341SAlan Cox #endif 2691bdbd705SKonstantin Belousov umtx_shm_object_init(object); 270df8bae1dSRodney W. Grimes } 271df8bae1dSRodney W. Grimes 272df8bae1dSRodney W. Grimes /* 27326f9a767SRodney W. Grimes * vm_object_init: 27426f9a767SRodney W. Grimes * 27526f9a767SRodney W. Grimes * Initialize the VM objects module. 27626f9a767SRodney W. Grimes */ 27726f9a767SRodney W. Grimes void 2781b40f8c0SMatthew Dillon vm_object_init(void) 27926f9a767SRodney W. Grimes { 28026f9a767SRodney W. Grimes TAILQ_INIT(&vm_object_list); 2816008862bSJohn Baldwin mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); 2820217125fSDavid Greenman 28389f6b863SAttilio Rao rw_init(&kernel_object->lock, "kernel vm object"); 284d1780e8dSKonstantin Belousov _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - 28567388836SKonstantin Belousov VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL); 286f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 287f8a47341SAlan Cox kernel_object->flags |= OBJ_COLORED; 288f8a47341SAlan Cox kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); 289f8a47341SAlan Cox #endif 290a720b31cSKonstantin Belousov kernel_object->un_pager.phys.ops = &default_phys_pg_ops; 29126f9a767SRodney W. Grimes 2928dbca793STor Egge /* 2938dbca793STor Egge * The lock portion of struct vm_object must be type stable due 2948dbca793STor Egge * to vm_pageout_fallback_object_lock locking a vm object 2958dbca793STor Egge * without holding any references to it. 296ffae7ea9SKonstantin Belousov * 297ffae7ea9SKonstantin Belousov * paging_in_progress is valid always. Lockless references to 298ffae7ea9SKonstantin Belousov * the objects may acquire pip and then check OBJ_DEAD. 2998dbca793STor Egge */ 3008355f576SJeff Roberson obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, 3018355f576SJeff Roberson #ifdef INVARIANTS 3028355f576SJeff Roberson vm_object_zdtor, 3038355f576SJeff Roberson #else 3048355f576SJeff Roberson NULL, 3058355f576SJeff Roberson #endif 3065df87b21SJeff Roberson vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 307774d251dSAttilio Rao 308cd1241fbSKonstantin Belousov vm_radix_zinit(); 30999448ed1SJohn Dyson } 31099448ed1SJohn Dyson 31199448ed1SJohn Dyson void 3121b40f8c0SMatthew Dillon vm_object_clear_flag(vm_object_t object, u_short bits) 3131b40f8c0SMatthew Dillon { 3145440b5a9SAlan Cox 31589f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 316b06805adSJake Burkholder object->flags &= ~bits; 3171b40f8c0SMatthew Dillon } 3181b40f8c0SMatthew Dillon 3193153e878SAlan Cox /* 3203153e878SAlan Cox * Sets the default memory attribute for the specified object. Pages 3213153e878SAlan Cox * that are allocated to this object are by default assigned this memory 3223153e878SAlan Cox * attribute. 3233153e878SAlan Cox * 3243153e878SAlan Cox * Presently, this function must be called before any pages are allocated 3253153e878SAlan Cox * to the object. In the future, this requirement may be relaxed for 3263153e878SAlan Cox * "default" and "swap" objects. 3273153e878SAlan Cox */ 3283153e878SAlan Cox int 3293153e878SAlan Cox vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) 3303153e878SAlan Cox { 3313153e878SAlan Cox 33289f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 3333e7a11caSKonstantin Belousov 3343e7a11caSKonstantin Belousov if (object->type == OBJT_DEAD) 3353e7a11caSKonstantin Belousov return (KERN_INVALID_ARGUMENT); 3363153e878SAlan Cox if (!TAILQ_EMPTY(&object->memq)) 3373153e878SAlan Cox return (KERN_FAILURE); 3383e7a11caSKonstantin Belousov 3393153e878SAlan Cox object->memattr = memattr; 3403153e878SAlan Cox return (KERN_SUCCESS); 3413153e878SAlan Cox } 3423153e878SAlan Cox 3431b40f8c0SMatthew Dillon void 3441b40f8c0SMatthew Dillon vm_object_pip_add(vm_object_t object, short i) 3451b40f8c0SMatthew Dillon { 346f279b88dSAlan Cox 347c99d0c58SMark Johnston if (i > 0) 348c99d0c58SMark Johnston blockcount_acquire(&object->paging_in_progress, i); 3491b40f8c0SMatthew Dillon } 3501b40f8c0SMatthew Dillon 3511b40f8c0SMatthew Dillon void 3521b40f8c0SMatthew Dillon vm_object_pip_wakeup(vm_object_t object) 3531b40f8c0SMatthew Dillon { 354f279b88dSAlan Cox 355c99d0c58SMark Johnston vm_object_pip_wakeupn(object, 1); 3561b40f8c0SMatthew Dillon } 3571b40f8c0SMatthew Dillon 3581b40f8c0SMatthew Dillon void 3591b40f8c0SMatthew Dillon vm_object_pip_wakeupn(vm_object_t object, short i) 3601b40f8c0SMatthew Dillon { 361d647a0edSAlan Cox 362c99d0c58SMark Johnston if (i > 0) 363c99d0c58SMark Johnston blockcount_release(&object->paging_in_progress, i); 3641b40f8c0SMatthew Dillon } 3651b40f8c0SMatthew Dillon 36698087a06SJeff Roberson /* 367c99d0c58SMark Johnston * Atomically drop the object lock and wait for pip to drain. This protects 368c99d0c58SMark Johnston * from sleep/wakeup races due to identity changes. The lock is not re-acquired 369c99d0c58SMark Johnston * on return. 37098087a06SJeff Roberson */ 37198087a06SJeff Roberson static void 372eaa17d42SRyan Libby vm_object_pip_sleep(vm_object_t object, const char *waitid) 37398087a06SJeff Roberson { 37498087a06SJeff Roberson 375c99d0c58SMark Johnston (void)blockcount_sleep(&object->paging_in_progress, &object->lock, 376c99d0c58SMark Johnston waitid, PVM | PDROP); 37798087a06SJeff Roberson } 37898087a06SJeff Roberson 3791b40f8c0SMatthew Dillon void 380eaa17d42SRyan Libby vm_object_pip_wait(vm_object_t object, const char *waitid) 3811b40f8c0SMatthew Dillon { 3821ca58953SAlan Cox 38389f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 384cf27e0d1SJeff Roberson 385c99d0c58SMark Johnston blockcount_wait(&object->paging_in_progress, &object->lock, waitid, 386c99d0c58SMark Johnston PVM); 3871b40f8c0SMatthew Dillon } 3881b40f8c0SMatthew Dillon 389cf27e0d1SJeff Roberson void 390eaa17d42SRyan Libby vm_object_pip_wait_unlocked(vm_object_t object, const char *waitid) 391cf27e0d1SJeff Roberson { 392cf27e0d1SJeff Roberson 393cf27e0d1SJeff Roberson VM_OBJECT_ASSERT_UNLOCKED(object); 394cf27e0d1SJeff Roberson 395c99d0c58SMark Johnston blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM); 396cf27e0d1SJeff Roberson } 397cf27e0d1SJeff Roberson 39826f9a767SRodney W. Grimes /* 39926f9a767SRodney W. Grimes * vm_object_allocate: 40026f9a767SRodney W. Grimes * 40126f9a767SRodney W. Grimes * Returns a new object with the given size. 40226f9a767SRodney W. Grimes */ 40326f9a767SRodney W. Grimes vm_object_t 4046395da54SIan Dowse vm_object_allocate(objtype_t type, vm_pindex_t size) 40526f9a767SRodney W. Grimes { 40690688d13SAlan Cox vm_object_t object; 40763967687SJeff Roberson u_short flags; 40863967687SJeff Roberson 40963967687SJeff Roberson switch (type) { 41063967687SJeff Roberson case OBJT_DEAD: 41163967687SJeff Roberson panic("vm_object_allocate: can't create OBJT_DEAD"); 41263967687SJeff Roberson case OBJT_DEFAULT: 41363967687SJeff Roberson flags = OBJ_COLORED; 41463967687SJeff Roberson break; 4154b8365d7SKonstantin Belousov case OBJT_SWAP: 4164b8365d7SKonstantin Belousov flags = OBJ_COLORED | OBJ_SWAP; 4174b8365d7SKonstantin Belousov break; 41863967687SJeff Roberson case OBJT_DEVICE: 41963967687SJeff Roberson case OBJT_SG: 42063967687SJeff Roberson flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; 42163967687SJeff Roberson break; 42263967687SJeff Roberson case OBJT_MGTDEVICE: 42363967687SJeff Roberson flags = OBJ_FICTITIOUS; 42463967687SJeff Roberson break; 42563967687SJeff Roberson case OBJT_PHYS: 42663967687SJeff Roberson flags = OBJ_UNMANAGED; 42763967687SJeff Roberson break; 42863967687SJeff Roberson case OBJT_VNODE: 42963967687SJeff Roberson flags = 0; 43063967687SJeff Roberson break; 43163967687SJeff Roberson default: 432b730fd30SKonstantin Belousov panic("vm_object_allocate: type %d is undefined or dynamic", 433b730fd30SKonstantin Belousov type); 43463967687SJeff Roberson } 43563967687SJeff Roberson object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 43667388836SKonstantin Belousov _vm_object_allocate(type, size, flags, object, NULL); 43763967687SJeff Roberson 43863967687SJeff Roberson return (object); 43963967687SJeff Roberson } 44063967687SJeff Roberson 441b730fd30SKonstantin Belousov vm_object_t 442b730fd30SKonstantin Belousov vm_object_allocate_dyn(objtype_t dyntype, vm_pindex_t size, u_short flags) 443b730fd30SKonstantin Belousov { 444b730fd30SKonstantin Belousov vm_object_t object; 445b730fd30SKonstantin Belousov 446b730fd30SKonstantin Belousov MPASS(dyntype >= OBJT_FIRST_DYN /* && dyntype < nitems(pagertab) */); 447b730fd30SKonstantin Belousov object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 448b730fd30SKonstantin Belousov _vm_object_allocate(dyntype, size, flags, object, NULL); 449b730fd30SKonstantin Belousov 450b730fd30SKonstantin Belousov return (object); 451b730fd30SKonstantin Belousov } 452b730fd30SKonstantin Belousov 45363967687SJeff Roberson /* 45463967687SJeff Roberson * vm_object_allocate_anon: 45563967687SJeff Roberson * 45663967687SJeff Roberson * Returns a new default object of the given size and marked as 45763967687SJeff Roberson * anonymous memory for special split/collapse handling. Color 45863967687SJeff Roberson * to be initialized by the caller. 45963967687SJeff Roberson */ 46063967687SJeff Roberson vm_object_t 46167388836SKonstantin Belousov vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object, 46267388836SKonstantin Belousov struct ucred *cred, vm_size_t charge) 46363967687SJeff Roberson { 46467388836SKonstantin Belousov vm_object_t handle, object; 46590688d13SAlan Cox 46667388836SKonstantin Belousov if (backing_object == NULL) 46767388836SKonstantin Belousov handle = NULL; 46867388836SKonstantin Belousov else if ((backing_object->flags & OBJ_ANON) != 0) 46967388836SKonstantin Belousov handle = backing_object->handle; 47067388836SKonstantin Belousov else 47167388836SKonstantin Belousov handle = backing_object; 47267388836SKonstantin Belousov object = uma_zalloc(obj_zone, M_WAITOK); 47363967687SJeff Roberson _vm_object_allocate(OBJT_DEFAULT, size, OBJ_ANON | OBJ_ONEMAPPING, 47467388836SKonstantin Belousov object, handle); 47567388836SKonstantin Belousov object->cred = cred; 47667388836SKonstantin Belousov object->charge = cred != NULL ? charge : 0; 47790688d13SAlan Cox return (object); 47826f9a767SRodney W. Grimes } 47926f9a767SRodney W. Grimes 48098087a06SJeff Roberson static void 48198087a06SJeff Roberson vm_object_reference_vnode(vm_object_t object) 482df8bae1dSRodney W. Grimes { 483a67d5408SJeff Roberson u_int old; 484a67d5408SJeff Roberson 485a67d5408SJeff Roberson /* 48698087a06SJeff Roberson * vnode objects need the lock for the first reference 48798087a06SJeff Roberson * to serialize with vnode_object_deallocate(). 488a67d5408SJeff Roberson */ 48998087a06SJeff Roberson if (!refcount_acquire_if_gt(&object->ref_count, 0)) { 49051df5321SJeff Roberson VM_OBJECT_RLOCK(object); 491a67d5408SJeff Roberson old = refcount_acquire(&object->ref_count); 4921a0c234eSJeff Roberson if (object->type == OBJT_VNODE && old == 0) 4931a0c234eSJeff Roberson vref(object->handle); 49451df5321SJeff Roberson VM_OBJECT_RUNLOCK(object); 49595e5e988SJohn Dyson } 496a67d5408SJeff Roberson } 49795e5e988SJohn Dyson 49823955314SAlfred Perlstein /* 49998087a06SJeff Roberson * vm_object_reference: 50098087a06SJeff Roberson * 50198087a06SJeff Roberson * Acquires a reference to the given object. 50298087a06SJeff Roberson */ 50398087a06SJeff Roberson void 50498087a06SJeff Roberson vm_object_reference(vm_object_t object) 50598087a06SJeff Roberson { 50698087a06SJeff Roberson 50798087a06SJeff Roberson if (object == NULL) 50898087a06SJeff Roberson return; 50998087a06SJeff Roberson 51098087a06SJeff Roberson if (object->type == OBJT_VNODE) 51198087a06SJeff Roberson vm_object_reference_vnode(object); 51298087a06SJeff Roberson else 51398087a06SJeff Roberson refcount_acquire(&object->ref_count); 51498087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 51598087a06SJeff Roberson ("vm_object_reference: Referenced dead object.")); 51698087a06SJeff Roberson } 51798087a06SJeff Roberson 51898087a06SJeff Roberson /* 519b921a12bSAlan Cox * vm_object_reference_locked: 520b921a12bSAlan Cox * 521b921a12bSAlan Cox * Gets another reference to the given object. 522b921a12bSAlan Cox * 523b921a12bSAlan Cox * The object must be locked. 524b921a12bSAlan Cox */ 525b921a12bSAlan Cox void 526b921a12bSAlan Cox vm_object_reference_locked(vm_object_t object) 527b921a12bSAlan Cox { 528a67d5408SJeff Roberson u_int old; 529b921a12bSAlan Cox 53051df5321SJeff Roberson VM_OBJECT_ASSERT_LOCKED(object); 531a67d5408SJeff Roberson old = refcount_acquire(&object->ref_count); 5321a0c234eSJeff Roberson if (object->type == OBJT_VNODE && old == 0) 5331a0c234eSJeff Roberson vref(object->handle); 53498087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 53598087a06SJeff Roberson ("vm_object_reference: Referenced dead object.")); 536b921a12bSAlan Cox } 537b921a12bSAlan Cox 538b921a12bSAlan Cox /* 5399d5abbddSJens Schweikhardt * Handle deallocating an object of type OBJT_VNODE. 54023955314SAlfred Perlstein */ 54102dd8331SAlan Cox static void 54298087a06SJeff Roberson vm_object_deallocate_vnode(vm_object_t object) 54395e5e988SJohn Dyson { 54495e5e988SJohn Dyson struct vnode *vp = (struct vnode *) object->handle; 54526c4e983SJeff Roberson bool last; 546219cbf59SEivind Eklund 5475526d2d9SEivind Eklund KASSERT(object->type == OBJT_VNODE, 54898087a06SJeff Roberson ("vm_object_deallocate_vnode: not a vnode object")); 54998087a06SJeff Roberson KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp")); 55095e5e988SJohn Dyson 55126c4e983SJeff Roberson /* Object lock to protect handle lookup. */ 55226c4e983SJeff Roberson last = refcount_release(&object->ref_count); 55326c4e983SJeff Roberson VM_OBJECT_RUNLOCK(object); 55426c4e983SJeff Roberson 55526c4e983SJeff Roberson if (!last) 55626c4e983SJeff Roberson return; 55726c4e983SJeff Roberson 558a67d5408SJeff Roberson if (!umtx_shm_vnobj_persistent) 5591bdbd705SKonstantin Belousov umtx_shm_object_terminated(object); 5601bdbd705SKonstantin Belousov 56103fa5b34SKonstantin Belousov /* vrele may need the vnode lock. */ 56247221757SJohn Dyson vrele(vp); 56386769ac0SKonstantin Belousov } 564df8bae1dSRodney W. Grimes 56598087a06SJeff Roberson /* 56698087a06SJeff Roberson * We dropped a reference on an object and discovered that it had a 56798087a06SJeff Roberson * single remaining shadow. This is a sibling of the reference we 56898087a06SJeff Roberson * dropped. Attempt to collapse the sibling and backing object. 56998087a06SJeff Roberson */ 57098087a06SJeff Roberson static vm_object_t 57198087a06SJeff Roberson vm_object_deallocate_anon(vm_object_t backing_object) 57298087a06SJeff Roberson { 57398087a06SJeff Roberson vm_object_t object; 57498087a06SJeff Roberson 57598087a06SJeff Roberson /* Fetch the final shadow. */ 57698087a06SJeff Roberson object = LIST_FIRST(&backing_object->shadow_head); 57798087a06SJeff Roberson KASSERT(object != NULL && backing_object->shadow_count == 1, 57898087a06SJeff Roberson ("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d", 57998087a06SJeff Roberson backing_object->ref_count, backing_object->shadow_count)); 5804b8365d7SKonstantin Belousov KASSERT((object->flags & OBJ_ANON) != 0, 58198087a06SJeff Roberson ("invalid shadow object %p", object)); 58298087a06SJeff Roberson 58398087a06SJeff Roberson if (!VM_OBJECT_TRYWLOCK(object)) { 58498087a06SJeff Roberson /* 58598087a06SJeff Roberson * Prevent object from disappearing since we do not have a 58698087a06SJeff Roberson * reference. 58798087a06SJeff Roberson */ 58898087a06SJeff Roberson vm_object_pip_add(object, 1); 58998087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 59098087a06SJeff Roberson VM_OBJECT_WLOCK(object); 59198087a06SJeff Roberson vm_object_pip_wakeup(object); 59298087a06SJeff Roberson } else 59398087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 59498087a06SJeff Roberson 59598087a06SJeff Roberson /* 59698087a06SJeff Roberson * Check for a collapse/terminate race with the last reference holder. 59798087a06SJeff Roberson */ 59898087a06SJeff Roberson if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 || 59998087a06SJeff Roberson !refcount_acquire_if_not_zero(&object->ref_count)) { 60098087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 60198087a06SJeff Roberson return (NULL); 60298087a06SJeff Roberson } 60398087a06SJeff Roberson backing_object = object->backing_object; 60498087a06SJeff Roberson if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0) 60598087a06SJeff Roberson vm_object_collapse(object); 60698087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 60798087a06SJeff Roberson 60898087a06SJeff Roberson return (object); 60998087a06SJeff Roberson } 61098087a06SJeff Roberson 611df8bae1dSRodney W. Grimes /* 612df8bae1dSRodney W. Grimes * vm_object_deallocate: 613df8bae1dSRodney W. Grimes * 614df8bae1dSRodney W. Grimes * Release a reference to the specified object, 615df8bae1dSRodney W. Grimes * gained either through a vm_object_allocate 616df8bae1dSRodney W. Grimes * or a vm_object_reference call. When all references 617df8bae1dSRodney W. Grimes * are gone, storage associated with this object 618df8bae1dSRodney W. Grimes * may be relinquished. 619df8bae1dSRodney W. Grimes * 620df8bae1dSRodney W. Grimes * No object may be locked. 621df8bae1dSRodney W. Grimes */ 62226f9a767SRodney W. Grimes void 6231b40f8c0SMatthew Dillon vm_object_deallocate(vm_object_t object) 624df8bae1dSRodney W. Grimes { 62598087a06SJeff Roberson vm_object_t temp; 62626c4e983SJeff Roberson bool released; 627df8bae1dSRodney W. Grimes 628df8bae1dSRodney W. Grimes while (object != NULL) { 62951df5321SJeff Roberson /* 63051df5321SJeff Roberson * If the reference count goes to 0 we start calling 63151df5321SJeff Roberson * vm_object_terminate() on the object chain. A ref count 63251df5321SJeff Roberson * of 1 may be a special case depending on the shadow count 63351df5321SJeff Roberson * being 0 or 1. These cases require a write lock on the 63451df5321SJeff Roberson * object. 63551df5321SJeff Roberson */ 63663967687SJeff Roberson if ((object->flags & OBJ_ANON) == 0) 63763967687SJeff Roberson released = refcount_release_if_gt(&object->ref_count, 1); 63863967687SJeff Roberson else 63951df5321SJeff Roberson released = refcount_release_if_gt(&object->ref_count, 2); 64051df5321SJeff Roberson if (released) 64151df5321SJeff Roberson return; 64251df5321SJeff Roberson 643a67d5408SJeff Roberson if (object->type == OBJT_VNODE) { 64426c4e983SJeff Roberson VM_OBJECT_RLOCK(object); 64526c4e983SJeff Roberson if (object->type == OBJT_VNODE) { 64698087a06SJeff Roberson vm_object_deallocate_vnode(object); 647a67d5408SJeff Roberson return; 648a67d5408SJeff Roberson } 64926c4e983SJeff Roberson VM_OBJECT_RUNLOCK(object); 65026c4e983SJeff Roberson } 65126c4e983SJeff Roberson 65226c4e983SJeff Roberson VM_OBJECT_WLOCK(object); 65326c4e983SJeff Roberson KASSERT(object->ref_count > 0, 65426c4e983SJeff Roberson ("vm_object_deallocate: object deallocated too many times: %d", 65526c4e983SJeff Roberson object->type)); 65626c4e983SJeff Roberson 657b72b0115SAlan Cox /* 65898087a06SJeff Roberson * If this is not the final reference to an anonymous 65998087a06SJeff Roberson * object we may need to collapse the shadow chain. 660b72b0115SAlan Cox */ 66198087a06SJeff Roberson if (!refcount_release(&object->ref_count)) { 66298087a06SJeff Roberson if (object->ref_count > 1 || 66398087a06SJeff Roberson object->shadow_count == 0) { 66498087a06SJeff Roberson if ((object->flags & OBJ_ANON) != 0 && 66598087a06SJeff Roberson object->ref_count == 1) 66698087a06SJeff Roberson vm_object_set_flag(object, 66798087a06SJeff Roberson OBJ_ONEMAPPING); 66889f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 66923b186d3SAlan Cox return; 67095e5e988SJohn Dyson } 67198087a06SJeff Roberson 67298087a06SJeff Roberson /* Handle collapsing last ref on anonymous objects. */ 67398087a06SJeff Roberson object = vm_object_deallocate_anon(object); 67498087a06SJeff Roberson continue; 67598087a06SJeff Roberson } 67698087a06SJeff Roberson 67798087a06SJeff Roberson /* 67898087a06SJeff Roberson * Handle the final reference to an object. We restart 67998087a06SJeff Roberson * the loop with the backing object to avoid recursion. 68098087a06SJeff Roberson */ 6811bdbd705SKonstantin Belousov umtx_shm_object_terminated(object); 68224a1cce3SDavid Greenman temp = object->backing_object; 683c9917419SAlan Cox if (temp != NULL) { 6847079449bSKonstantin Belousov KASSERT(object->type == OBJT_DEFAULT || 6857079449bSKonstantin Belousov object->type == OBJT_SWAP, 6864bace8e7SKonstantin Belousov ("shadowed tmpfs v_object 2 %p", object)); 68751b867e5SJeff Roberson vm_object_backing_remove(object); 688de5f6a77SJohn Dyson } 68998087a06SJeff Roberson 69098087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 69198087a06SJeff Roberson ("vm_object_deallocate: Terminating dead object.")); 692783a68aaSKonstantin Belousov vm_object_set_flag(object, OBJ_DEAD); 693df8bae1dSRodney W. Grimes vm_object_terminate(object); 694df8bae1dSRodney W. Grimes object = temp; 695df8bae1dSRodney W. Grimes } 696df8bae1dSRodney W. Grimes } 697df8bae1dSRodney W. Grimes 698df8bae1dSRodney W. Grimes /* 6992ac78f0eSStephan Uphoff * vm_object_destroy removes the object from the global object list 7002ac78f0eSStephan Uphoff * and frees the space for the object. 7012ac78f0eSStephan Uphoff */ 7022ac78f0eSStephan Uphoff void 7032ac78f0eSStephan Uphoff vm_object_destroy(vm_object_t object) 7042ac78f0eSStephan Uphoff { 7052ac78f0eSStephan Uphoff 7062ac78f0eSStephan Uphoff /* 7073364c323SKonstantin Belousov * Release the allocation charge. 7083364c323SKonstantin Belousov */ 709ef694c1aSEdward Tomasz Napierala if (object->cred != NULL) { 710ef694c1aSEdward Tomasz Napierala swap_release_by_cred(object->charge, object->cred); 7113364c323SKonstantin Belousov object->charge = 0; 712ef694c1aSEdward Tomasz Napierala crfree(object->cred); 713ef694c1aSEdward Tomasz Napierala object->cred = NULL; 7143364c323SKonstantin Belousov } 7153364c323SKonstantin Belousov 7163364c323SKonstantin Belousov /* 7172ac78f0eSStephan Uphoff * Free the space for the object. 7182ac78f0eSStephan Uphoff */ 7192ac78f0eSStephan Uphoff uma_zfree(obj_zone, object); 7202ac78f0eSStephan Uphoff } 7212ac78f0eSStephan Uphoff 72251b867e5SJeff Roberson static void 72351b867e5SJeff Roberson vm_object_backing_remove_locked(vm_object_t object) 72451b867e5SJeff Roberson { 72551b867e5SJeff Roberson vm_object_t backing_object; 72651b867e5SJeff Roberson 72751b867e5SJeff Roberson backing_object = object->backing_object; 72851b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 72951b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(backing_object); 73051b867e5SJeff Roberson 73198087a06SJeff Roberson KASSERT((object->flags & OBJ_COLLAPSING) == 0, 73298087a06SJeff Roberson ("vm_object_backing_remove: Removing collapsing object.")); 73398087a06SJeff Roberson 73451b867e5SJeff Roberson if ((object->flags & OBJ_SHADOWLIST) != 0) { 73551b867e5SJeff Roberson LIST_REMOVE(object, shadow_list); 73651b867e5SJeff Roberson backing_object->shadow_count--; 73751b867e5SJeff Roberson object->flags &= ~OBJ_SHADOWLIST; 73851b867e5SJeff Roberson } 73951b867e5SJeff Roberson object->backing_object = NULL; 74051b867e5SJeff Roberson } 74151b867e5SJeff Roberson 74251b867e5SJeff Roberson static void 74351b867e5SJeff Roberson vm_object_backing_remove(vm_object_t object) 74451b867e5SJeff Roberson { 74551b867e5SJeff Roberson vm_object_t backing_object; 74651b867e5SJeff Roberson 74751b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 74851b867e5SJeff Roberson 74951b867e5SJeff Roberson if ((object->flags & OBJ_SHADOWLIST) != 0) { 75051b867e5SJeff Roberson backing_object = object->backing_object; 75151b867e5SJeff Roberson VM_OBJECT_WLOCK(backing_object); 75251b867e5SJeff Roberson vm_object_backing_remove_locked(object); 75351b867e5SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 75451b867e5SJeff Roberson } else 75551b867e5SJeff Roberson object->backing_object = NULL; 75651b867e5SJeff Roberson } 75751b867e5SJeff Roberson 75851b867e5SJeff Roberson static void 75951b867e5SJeff Roberson vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object) 76051b867e5SJeff Roberson { 76151b867e5SJeff Roberson 76251b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 76351b867e5SJeff Roberson 76451b867e5SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 76551b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(backing_object); 76651b867e5SJeff Roberson LIST_INSERT_HEAD(&backing_object->shadow_head, object, 76751b867e5SJeff Roberson shadow_list); 76851b867e5SJeff Roberson backing_object->shadow_count++; 76951b867e5SJeff Roberson object->flags |= OBJ_SHADOWLIST; 77051b867e5SJeff Roberson } 77151b867e5SJeff Roberson object->backing_object = backing_object; 77251b867e5SJeff Roberson } 77351b867e5SJeff Roberson 77451b867e5SJeff Roberson static void 77551b867e5SJeff Roberson vm_object_backing_insert(vm_object_t object, vm_object_t backing_object) 77651b867e5SJeff Roberson { 77751b867e5SJeff Roberson 77851b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 77951b867e5SJeff Roberson 78051b867e5SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 78151b867e5SJeff Roberson VM_OBJECT_WLOCK(backing_object); 78251b867e5SJeff Roberson vm_object_backing_insert_locked(object, backing_object); 78351b867e5SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 78451b867e5SJeff Roberson } else 78551b867e5SJeff Roberson object->backing_object = backing_object; 78651b867e5SJeff Roberson } 78751b867e5SJeff Roberson 78898087a06SJeff Roberson /* 78998087a06SJeff Roberson * Insert an object into a backing_object's shadow list with an additional 79098087a06SJeff Roberson * reference to the backing_object added. 79198087a06SJeff Roberson */ 79298087a06SJeff Roberson static void 79398087a06SJeff Roberson vm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object) 79498087a06SJeff Roberson { 79598087a06SJeff Roberson 79698087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 79798087a06SJeff Roberson 79898087a06SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 79998087a06SJeff Roberson VM_OBJECT_WLOCK(backing_object); 80098087a06SJeff Roberson KASSERT((backing_object->flags & OBJ_DEAD) == 0, 80198087a06SJeff Roberson ("shadowing dead anonymous object")); 80298087a06SJeff Roberson vm_object_reference_locked(backing_object); 80398087a06SJeff Roberson vm_object_backing_insert_locked(object, backing_object); 80498087a06SJeff Roberson vm_object_clear_flag(backing_object, OBJ_ONEMAPPING); 80598087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 80698087a06SJeff Roberson } else { 80798087a06SJeff Roberson vm_object_reference(backing_object); 80898087a06SJeff Roberson object->backing_object = backing_object; 80998087a06SJeff Roberson } 81098087a06SJeff Roberson } 81198087a06SJeff Roberson 81298087a06SJeff Roberson /* 81398087a06SJeff Roberson * Transfer a backing reference from backing_object to object. 81498087a06SJeff Roberson */ 81598087a06SJeff Roberson static void 81698087a06SJeff Roberson vm_object_backing_transfer(vm_object_t object, vm_object_t backing_object) 81798087a06SJeff Roberson { 81898087a06SJeff Roberson vm_object_t new_backing_object; 81998087a06SJeff Roberson 82098087a06SJeff Roberson /* 82198087a06SJeff Roberson * Note that the reference to backing_object->backing_object 82298087a06SJeff Roberson * moves from within backing_object to within object. 82398087a06SJeff Roberson */ 82498087a06SJeff Roberson vm_object_backing_remove_locked(object); 82598087a06SJeff Roberson new_backing_object = backing_object->backing_object; 82698087a06SJeff Roberson if (new_backing_object == NULL) 82798087a06SJeff Roberson return; 82898087a06SJeff Roberson if ((new_backing_object->flags & OBJ_ANON) != 0) { 82998087a06SJeff Roberson VM_OBJECT_WLOCK(new_backing_object); 83098087a06SJeff Roberson vm_object_backing_remove_locked(backing_object); 83198087a06SJeff Roberson vm_object_backing_insert_locked(object, new_backing_object); 83298087a06SJeff Roberson VM_OBJECT_WUNLOCK(new_backing_object); 83398087a06SJeff Roberson } else { 83498087a06SJeff Roberson object->backing_object = new_backing_object; 83598087a06SJeff Roberson backing_object->backing_object = NULL; 83698087a06SJeff Roberson } 83798087a06SJeff Roberson } 83898087a06SJeff Roberson 83998087a06SJeff Roberson /* 84098087a06SJeff Roberson * Wait for a concurrent collapse to settle. 84198087a06SJeff Roberson */ 84298087a06SJeff Roberson static void 84398087a06SJeff Roberson vm_object_collapse_wait(vm_object_t object) 84498087a06SJeff Roberson { 84598087a06SJeff Roberson 84698087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 84798087a06SJeff Roberson 84898087a06SJeff Roberson while ((object->flags & OBJ_COLLAPSING) != 0) { 84998087a06SJeff Roberson vm_object_pip_wait(object, "vmcolwait"); 85098087a06SJeff Roberson counter_u64_add(object_collapse_waits, 1); 85198087a06SJeff Roberson } 85298087a06SJeff Roberson } 85398087a06SJeff Roberson 85498087a06SJeff Roberson /* 85598087a06SJeff Roberson * Waits for a backing object to clear a pending collapse and returns 85698087a06SJeff Roberson * it locked if it is an ANON object. 85798087a06SJeff Roberson */ 85898087a06SJeff Roberson static vm_object_t 85998087a06SJeff Roberson vm_object_backing_collapse_wait(vm_object_t object) 86098087a06SJeff Roberson { 86198087a06SJeff Roberson vm_object_t backing_object; 86298087a06SJeff Roberson 86398087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 86498087a06SJeff Roberson 86598087a06SJeff Roberson for (;;) { 86698087a06SJeff Roberson backing_object = object->backing_object; 86798087a06SJeff Roberson if (backing_object == NULL || 86898087a06SJeff Roberson (backing_object->flags & OBJ_ANON) == 0) 86998087a06SJeff Roberson return (NULL); 87098087a06SJeff Roberson VM_OBJECT_WLOCK(backing_object); 87198087a06SJeff Roberson if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0) 87298087a06SJeff Roberson break; 873cdd02f43SMark Johnston VM_OBJECT_WUNLOCK(object); 87498087a06SJeff Roberson vm_object_pip_sleep(backing_object, "vmbckwait"); 87598087a06SJeff Roberson counter_u64_add(object_collapse_waits, 1); 87698087a06SJeff Roberson VM_OBJECT_WLOCK(object); 87798087a06SJeff Roberson } 87898087a06SJeff Roberson return (backing_object); 87998087a06SJeff Roberson } 88051b867e5SJeff Roberson 8812ac78f0eSStephan Uphoff /* 8827bbdb843SRuslan Bukin * vm_object_terminate_pages removes any remaining pageable pages 8837bbdb843SRuslan Bukin * from the object and resets the object to an empty state. 8847bbdb843SRuslan Bukin */ 8857bbdb843SRuslan Bukin static void 8867bbdb843SRuslan Bukin vm_object_terminate_pages(vm_object_t object) 8877bbdb843SRuslan Bukin { 8887bbdb843SRuslan Bukin vm_page_t p, p_next; 8897bbdb843SRuslan Bukin 8907bbdb843SRuslan Bukin VM_OBJECT_ASSERT_WLOCKED(object); 8917bbdb843SRuslan Bukin 8927bbdb843SRuslan Bukin /* 8937bbdb843SRuslan Bukin * Free any remaining pageable pages. This also removes them from the 8947bbdb843SRuslan Bukin * paging queues. However, don't free wired pages, just remove them 8957bbdb843SRuslan Bukin * from the object. Rather than incrementally removing each page from 8967bbdb843SRuslan Bukin * the object, the page and object are reset to any empty state. 8977bbdb843SRuslan Bukin */ 8987bbdb843SRuslan Bukin TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { 8997bbdb843SRuslan Bukin vm_page_assert_unbusied(p); 900fee2a2faSMark Johnston KASSERT(p->object == object && 901fee2a2faSMark Johnston (p->ref_count & VPRC_OBJREF) != 0, 902fee2a2faSMark Johnston ("vm_object_terminate_pages: page %p is inconsistent", p)); 903fee2a2faSMark Johnston 9047bbdb843SRuslan Bukin p->object = NULL; 905fee2a2faSMark Johnston if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) { 9065cd29d0fSMark Johnston VM_CNT_INC(v_pfree); 9075cd29d0fSMark Johnston vm_page_free(p); 9084074d642SAlan Cox } 909fee2a2faSMark Johnston } 9102fcd1ff6SKonstantin Belousov 9117bbdb843SRuslan Bukin /* 9127bbdb843SRuslan Bukin * If the object contained any pages, then reset it to an empty state. 9137bbdb843SRuslan Bukin * None of the object's fields, including "resident_page_count", were 9147bbdb843SRuslan Bukin * modified by the preceding loop. 9157bbdb843SRuslan Bukin */ 9167bbdb843SRuslan Bukin if (object->resident_page_count != 0) { 9177bbdb843SRuslan Bukin vm_radix_reclaim_allnodes(&object->rtree); 9187bbdb843SRuslan Bukin TAILQ_INIT(&object->memq); 9197bbdb843SRuslan Bukin object->resident_page_count = 0; 9207bbdb843SRuslan Bukin if (object->type == OBJT_VNODE) 9217bbdb843SRuslan Bukin vdrop(object->handle); 9227bbdb843SRuslan Bukin } 9237bbdb843SRuslan Bukin } 9247bbdb843SRuslan Bukin 9257bbdb843SRuslan Bukin /* 926df8bae1dSRodney W. Grimes * vm_object_terminate actually destroys the specified object, freeing 927df8bae1dSRodney W. Grimes * up all previously used resources. 928df8bae1dSRodney W. Grimes * 929df8bae1dSRodney W. Grimes * The object must be locked. 9301c7c3c6aSMatthew Dillon * This routine may block. 931df8bae1dSRodney W. Grimes */ 93295e5e988SJohn Dyson void 9331b40f8c0SMatthew Dillon vm_object_terminate(vm_object_t object) 934df8bae1dSRodney W. Grimes { 93598087a06SJeff Roberson 93689f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 937783a68aaSKonstantin Belousov KASSERT((object->flags & OBJ_DEAD) != 0, 938783a68aaSKonstantin Belousov ("terminating non-dead obj %p", object)); 93998087a06SJeff Roberson KASSERT((object->flags & OBJ_COLLAPSING) == 0, 94098087a06SJeff Roberson ("terminating collapsing obj %p", object)); 94198087a06SJeff Roberson KASSERT(object->backing_object == NULL, 94298087a06SJeff Roberson ("terminating shadow obj %p", object)); 943bef608bdSJohn Dyson 944cf27e0d1SJeff Roberson /* 945ffae7ea9SKonstantin Belousov * Wait for the pageout daemon and other current users to be 946ffae7ea9SKonstantin Belousov * done with the object. Note that new paging_in_progress 947ffae7ea9SKonstantin Belousov * users can come after this wait, but they must check 948ffae7ea9SKonstantin Belousov * OBJ_DEAD flag set (without unlocking the object), and avoid 949ffae7ea9SKonstantin Belousov * the object being terminated. 950cf27e0d1SJeff Roberson */ 951cf27e0d1SJeff Roberson vm_object_pip_wait(object, "objtrm"); 952cf27e0d1SJeff Roberson 953971dd342SAlfred Perlstein KASSERT(object->ref_count == 0, 954971dd342SAlfred Perlstein ("vm_object_terminate: object with references, ref_count=%d", 955971dd342SAlfred Perlstein object->ref_count)); 956996c772fSJohn Dyson 9577bbdb843SRuslan Bukin if ((object->flags & OBJ_PG_DTOR) == 0) 9587bbdb843SRuslan Bukin vm_object_terminate_pages(object); 959bef608bdSJohn Dyson 960f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 961f8a47341SAlan Cox if (__predict_false(!LIST_EMPTY(&object->rvq))) 962f8a47341SAlan Cox vm_reserv_break_all(object); 963f8a47341SAlan Cox #endif 9647bfda801SAlan Cox 965e735691bSJohn Baldwin KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT || 9667079449bSKonstantin Belousov (object->flags & OBJ_SWAP) != 0, 967e735691bSJohn Baldwin ("%s: non-swap obj %p has cred", __func__, object)); 968e735691bSJohn Baldwin 9692d8acc0fSJohn Dyson /* 9709fcfb650SDavid Greenman * Let the pager know object is dead. 9719fcfb650SDavid Greenman */ 9729fcfb650SDavid Greenman vm_pager_deallocate(object); 97389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 9749fcfb650SDavid Greenman 9752ac78f0eSStephan Uphoff vm_object_destroy(object); 97647221757SJohn Dyson } 977df8bae1dSRodney W. Grimes 978edf93b25SAlan Cox /* 979edf93b25SAlan Cox * Make the page read-only so that we can clear the object flags. However, if 980edf93b25SAlan Cox * this is a nosync mmap then the object is likely to stay dirty so do not 981edf93b25SAlan Cox * mess with the page and do not clear the object flags. Returns TRUE if the 982edf93b25SAlan Cox * page should be flushed, and FALSE otherwise. 983edf93b25SAlan Cox */ 9843280870dSKonstantin Belousov static boolean_t 98567d0e293SJeff Roberson vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean) 9863280870dSKonstantin Belousov { 9873280870dSKonstantin Belousov 988fff5403fSJeff Roberson vm_page_assert_busied(p); 989fff5403fSJeff Roberson 9903280870dSKonstantin Belousov /* 9913280870dSKonstantin Belousov * If we have been asked to skip nosync pages and this is a 9923280870dSKonstantin Belousov * nosync page, skip it. Note that the object flags were not 9933280870dSKonstantin Belousov * cleared in this case so we do not have to set them. 9943280870dSKonstantin Belousov */ 9955cff1f4dSMark Johnston if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) { 99667d0e293SJeff Roberson *allclean = FALSE; 9973280870dSKonstantin Belousov return (FALSE); 9983280870dSKonstantin Belousov } else { 9993280870dSKonstantin Belousov pmap_remove_write(p); 10003280870dSKonstantin Belousov return (p->dirty != 0); 10013280870dSKonstantin Belousov } 10023280870dSKonstantin Belousov } 10033280870dSKonstantin Belousov 1004df8bae1dSRodney W. Grimes /* 1005df8bae1dSRodney W. Grimes * vm_object_page_clean 1006df8bae1dSRodney W. Grimes * 10074f79d873SMatthew Dillon * Clean all dirty pages in the specified range of object. Leaves page 10084f79d873SMatthew Dillon * on whatever queue it is currently on. If NOSYNC is set then do not 1009fff5403fSJeff Roberson * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC), 10104f79d873SMatthew Dillon * leaving the object dirty. 101126f9a767SRodney W. Grimes * 10128d34a3bfSKonstantin Belousov * For swap objects backing tmpfs regular files, do not flush anything, 10138d34a3bfSKonstantin Belousov * but remove write protection on the mapped pages to update mtime through 10148d34a3bfSKonstantin Belousov * mmaped writes. 10158d34a3bfSKonstantin Belousov * 101643b7990eSMatthew Dillon * When stuffing pages asynchronously, allow clustering. XXX we need a 101743b7990eSMatthew Dillon * synchronous clustering mode implementation. 101843b7990eSMatthew Dillon * 101926f9a767SRodney W. Grimes * Odd semantics: if start == end, we clean everything. 102026f9a767SRodney W. Grimes * 102126f9a767SRodney W. Grimes * The object must be locked. 1022126d6082SKonstantin Belousov * 1023126d6082SKonstantin Belousov * Returns FALSE if some page from the range was not written, as 1024126d6082SKonstantin Belousov * reported by the pager, and TRUE otherwise. 102526f9a767SRodney W. Grimes */ 1026126d6082SKonstantin Belousov boolean_t 102717f3095dSAlan Cox vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, 1028e239bb97SKonstantin Belousov int flags) 1029f6b04d2bSDavid Greenman { 1030e239bb97SKonstantin Belousov vm_page_t np, p; 103117f3095dSAlan Cox vm_pindex_t pi, tend, tstart; 1032126d6082SKonstantin Belousov int curgeneration, n, pagerflags; 103367d0e293SJeff Roberson boolean_t eio, res, allclean; 1034f6b04d2bSDavid Greenman 103589f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 1036e5f299ffSKonstantin Belousov 10378d34a3bfSKonstantin Belousov if (!vm_object_mightbedirty(object) || object->resident_page_count == 0) 1038126d6082SKonstantin Belousov return (TRUE); 1039f6b04d2bSDavid Greenman 1040e239bb97SKonstantin Belousov pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? 1041e239bb97SKonstantin Belousov VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; 1042e239bb97SKonstantin Belousov pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; 1043e239bb97SKonstantin Belousov 104417f3095dSAlan Cox tstart = OFF_TO_IDX(start); 104517f3095dSAlan Cox tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); 104667d0e293SJeff Roberson allclean = tstart == 0 && tend >= object->size; 1047126d6082SKonstantin Belousov res = TRUE; 1048f6b04d2bSDavid Greenman 1049bd7e5f99SJohn Dyson rescan: 10502d8acc0fSJohn Dyson curgeneration = object->generation; 10512d8acc0fSJohn Dyson 105217f3095dSAlan Cox for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { 1053bd7e5f99SJohn Dyson pi = p->pindex; 1054e239bb97SKonstantin Belousov if (pi >= tend) 1055e239bb97SKonstantin Belousov break; 1056e239bb97SKonstantin Belousov np = TAILQ_NEXT(p, listq); 10570012f373SJeff Roberson if (vm_page_none_valid(p)) 1058aef922f5SJohn Dyson continue; 105963e97555SJeff Roberson if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) { 106067d0e293SJeff Roberson if (object->generation != curgeneration && 106167d0e293SJeff Roberson (flags & OBJPC_SYNC) != 0) 1062e239bb97SKonstantin Belousov goto rescan; 1063780636b7SKonstantin Belousov np = vm_page_find_least(object, pi); 1064780636b7SKonstantin Belousov continue; 1065f6b04d2bSDavid Greenman } 106667d0e293SJeff Roberson if (!vm_object_page_remove_write(p, flags, &allclean)) { 106763e97555SJeff Roberson vm_page_xunbusy(p); 1068bd7e5f99SJohn Dyson continue; 106963e97555SJeff Roberson } 10708d34a3bfSKonstantin Belousov if (object->type == OBJT_VNODE) { 10713280870dSKonstantin Belousov n = vm_object_page_collect_flush(object, p, pagerflags, 107267d0e293SJeff Roberson flags, &allclean, &eio); 1073126d6082SKonstantin Belousov if (eio) { 1074126d6082SKonstantin Belousov res = FALSE; 107567d0e293SJeff Roberson allclean = FALSE; 1076126d6082SKonstantin Belousov } 107767d0e293SJeff Roberson if (object->generation != curgeneration && 107867d0e293SJeff Roberson (flags & OBJPC_SYNC) != 0) 1079b9b7a4beSMatthew Dillon goto rescan; 1080031ec8c1SKonstantin Belousov 1081031ec8c1SKonstantin Belousov /* 1082031ec8c1SKonstantin Belousov * If the VOP_PUTPAGES() did a truncated write, so 1083031ec8c1SKonstantin Belousov * that even the first page of the run is not fully 1084031ec8c1SKonstantin Belousov * written, vm_pageout_flush() returns 0 as the run 1085031ec8c1SKonstantin Belousov * length. Since the condition that caused truncated 1086031ec8c1SKonstantin Belousov * write may be permanent, e.g. exhausted free space, 1087031ec8c1SKonstantin Belousov * accepting n == 0 would cause an infinite loop. 1088031ec8c1SKonstantin Belousov * 1089031ec8c1SKonstantin Belousov * Forwarding the iterator leaves the unwritten page 1090031ec8c1SKonstantin Belousov * behind, but there is not much we can do there if 1091031ec8c1SKonstantin Belousov * filesystem refuses to write it. 1092031ec8c1SKonstantin Belousov */ 1093126d6082SKonstantin Belousov if (n == 0) { 1094031ec8c1SKonstantin Belousov n = 1; 109567d0e293SJeff Roberson allclean = FALSE; 1096126d6082SKonstantin Belousov } 10978d34a3bfSKonstantin Belousov } else { 10988d34a3bfSKonstantin Belousov n = 1; 10998d34a3bfSKonstantin Belousov vm_page_xunbusy(p); 11008d34a3bfSKonstantin Belousov } 1101e239bb97SKonstantin Belousov np = vm_page_find_least(object, pi + n); 1102b9b7a4beSMatthew Dillon } 1103b9b7a4beSMatthew Dillon #if 0 1104e239bb97SKonstantin Belousov VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); 1105b9b7a4beSMatthew Dillon #endif 1106b9b7a4beSMatthew Dillon 11078d34a3bfSKonstantin Belousov /* 11088d34a3bfSKonstantin Belousov * Leave updating cleangeneration for tmpfs objects to tmpfs 11098d34a3bfSKonstantin Belousov * scan. It needs to update mtime, which happens for other 11108d34a3bfSKonstantin Belousov * filesystems during page writeouts. 11118d34a3bfSKonstantin Belousov */ 11128d34a3bfSKonstantin Belousov if (allclean && object->type == OBJT_VNODE) 111367d0e293SJeff Roberson object->cleangeneration = curgeneration; 1114126d6082SKonstantin Belousov return (res); 1115b9b7a4beSMatthew Dillon } 1116b9b7a4beSMatthew Dillon 1117b9b7a4beSMatthew Dillon static int 11183280870dSKonstantin Belousov vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, 111967d0e293SJeff Roberson int flags, boolean_t *allclean, boolean_t *eio) 1120b9b7a4beSMatthew Dillon { 11213157c503SKonstantin Belousov vm_page_t ma[vm_pageout_page_count], p_first, tp; 11223157c503SKonstantin Belousov int count, i, mreq, runlen; 1123b9b7a4beSMatthew Dillon 11247bec141bSKip Macy vm_page_lock_assert(p, MA_NOTOWNED); 112563e97555SJeff Roberson vm_page_assert_xbusied(p); 112689f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 11273157c503SKonstantin Belousov 11283157c503SKonstantin Belousov count = 1; 11293157c503SKonstantin Belousov mreq = 0; 11303157c503SKonstantin Belousov 11313157c503SKonstantin Belousov for (tp = p; count < vm_pageout_page_count; count++) { 11323157c503SKonstantin Belousov tp = vm_page_next(tp); 113363e97555SJeff Roberson if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1134bd7e5f99SJohn Dyson break; 113567d0e293SJeff Roberson if (!vm_object_page_remove_write(tp, flags, allclean)) { 113663e97555SJeff Roberson vm_page_xunbusy(tp); 1137bd7e5f99SJohn Dyson break; 1138bd7e5f99SJohn Dyson } 113963e97555SJeff Roberson } 1140aef922f5SJohn Dyson 11413157c503SKonstantin Belousov for (p_first = p; count < vm_pageout_page_count; count++) { 11423157c503SKonstantin Belousov tp = vm_page_prev(p_first); 114363e97555SJeff Roberson if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1144bd7e5f99SJohn Dyson break; 114567d0e293SJeff Roberson if (!vm_object_page_remove_write(tp, flags, allclean)) { 114663e97555SJeff Roberson vm_page_xunbusy(tp); 1147bd7e5f99SJohn Dyson break; 114863e97555SJeff Roberson } 11493157c503SKonstantin Belousov p_first = tp; 11503157c503SKonstantin Belousov mreq++; 1151bd7e5f99SJohn Dyson } 1152bd7e5f99SJohn Dyson 11533157c503SKonstantin Belousov for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) 11543157c503SKonstantin Belousov ma[i] = tp; 1155cf2819ccSJohn Dyson 1156126d6082SKonstantin Belousov vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); 11571e8a675cSKonstantin Belousov return (runlen); 115826f9a767SRodney W. Grimes } 1159df8bae1dSRodney W. Grimes 11601efb74fbSJohn Dyson /* 1161950f8459SAlan Cox * Note that there is absolutely no sense in writing out 1162950f8459SAlan Cox * anonymous objects, so we track down the vnode object 1163950f8459SAlan Cox * to write out. 1164950f8459SAlan Cox * We invalidate (remove) all pages from the address space 1165950f8459SAlan Cox * for semantic correctness. 1166950f8459SAlan Cox * 11676bbee8e2SAlan Cox * If the backing object is a device object with unmanaged pages, then any 11686bbee8e2SAlan Cox * mappings to the specified range of pages must be removed before this 11696bbee8e2SAlan Cox * function is called. 11706bbee8e2SAlan Cox * 1171950f8459SAlan Cox * Note: certain anonymous maps, such as MAP_NOSYNC maps, 1172950f8459SAlan Cox * may start out with a NULL object. 1173950f8459SAlan Cox */ 1174126d6082SKonstantin Belousov boolean_t 1175950f8459SAlan Cox vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, 1176950f8459SAlan Cox boolean_t syncio, boolean_t invalidate) 1177950f8459SAlan Cox { 1178950f8459SAlan Cox vm_object_t backing_object; 1179950f8459SAlan Cox struct vnode *vp; 11803b582b4eSTor Egge struct mount *mp; 1181126d6082SKonstantin Belousov int error, flags, fsync_after; 1182126d6082SKonstantin Belousov boolean_t res; 1183950f8459SAlan Cox 1184950f8459SAlan Cox if (object == NULL) 1185126d6082SKonstantin Belousov return (TRUE); 1186126d6082SKonstantin Belousov res = TRUE; 1187126d6082SKonstantin Belousov error = 0; 118889f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1189950f8459SAlan Cox while ((backing_object = object->backing_object) != NULL) { 119089f6b863SAttilio Rao VM_OBJECT_WLOCK(backing_object); 119156e0670fSAlan Cox offset += object->backing_object_offset; 119289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1193950f8459SAlan Cox object = backing_object; 1194950f8459SAlan Cox if (object->size < OFF_TO_IDX(offset + size)) 1195950f8459SAlan Cox size = IDX_TO_OFF(object->size) - offset; 1196950f8459SAlan Cox } 1197950f8459SAlan Cox /* 1198950f8459SAlan Cox * Flush pages if writing is allowed, invalidate them 1199950f8459SAlan Cox * if invalidation requested. Pages undergoing I/O 1200950f8459SAlan Cox * will be ignored by vm_object_page_remove(). 1201950f8459SAlan Cox * 1202950f8459SAlan Cox * We cannot lock the vnode and then wait for paging 1203950f8459SAlan Cox * to complete without deadlocking against vm_fault. 1204950f8459SAlan Cox * Instead we simply call vm_object_page_remove() and 1205950f8459SAlan Cox * allow it to block internally on a page-by-page 1206950f8459SAlan Cox * basis when it encounters pages undergoing async 1207950f8459SAlan Cox * I/O. 1208950f8459SAlan Cox */ 1209950f8459SAlan Cox if (object->type == OBJT_VNODE && 121067d0e293SJeff Roberson vm_object_mightbedirty(object) != 0 && 12115bf94937SKonstantin Belousov ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { 121289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 12133b582b4eSTor Egge (void) vn_start_write(vp, &mp, V_WAIT); 1214cb05b60aSAttilio Rao vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 121575ff604aSKonstantin Belousov if (syncio && !invalidate && offset == 0 && 1216d1780e8dSKonstantin Belousov atop(size) == object->size) { 121775ff604aSKonstantin Belousov /* 121875ff604aSKonstantin Belousov * If syncing the whole mapping of the file, 121975ff604aSKonstantin Belousov * it is faster to schedule all the writes in 122075ff604aSKonstantin Belousov * async mode, also allowing the clustering, 122175ff604aSKonstantin Belousov * and then wait for i/o to complete. 122275ff604aSKonstantin Belousov */ 122375ff604aSKonstantin Belousov flags = 0; 122475ff604aSKonstantin Belousov fsync_after = TRUE; 122575ff604aSKonstantin Belousov } else { 1226950f8459SAlan Cox flags = (syncio || invalidate) ? OBJPC_SYNC : 0; 122775ff604aSKonstantin Belousov flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; 122875ff604aSKonstantin Belousov fsync_after = FALSE; 122975ff604aSKonstantin Belousov } 123089f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1231126d6082SKonstantin Belousov res = vm_object_page_clean(object, offset, offset + size, 1232126d6082SKonstantin Belousov flags); 123389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 123475ff604aSKonstantin Belousov if (fsync_after) 1235126d6082SKonstantin Belousov error = VOP_FSYNC(vp, MNT_WAIT, curthread); 1236b249ce48SMateusz Guzik VOP_UNLOCK(vp); 12373b582b4eSTor Egge vn_finished_write(mp); 1238126d6082SKonstantin Belousov if (error != 0) 1239126d6082SKonstantin Belousov res = FALSE; 124089f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1241950f8459SAlan Cox } 1242950f8459SAlan Cox if ((object->type == OBJT_VNODE || 1243950f8459SAlan Cox object->type == OBJT_DEVICE) && invalidate) { 12446bbee8e2SAlan Cox if (object->type == OBJT_DEVICE) 12456bbee8e2SAlan Cox /* 12466bbee8e2SAlan Cox * The option OBJPR_NOTMAPPED must be passed here 12476bbee8e2SAlan Cox * because vm_object_page_remove() cannot remove 12486bbee8e2SAlan Cox * unmanaged mappings. 12496bbee8e2SAlan Cox */ 12506bbee8e2SAlan Cox flags = OBJPR_NOTMAPPED; 12516bbee8e2SAlan Cox else if (old_msync) 12526195b24aSKonstantin Belousov flags = 0; 12536bbee8e2SAlan Cox else 12546195b24aSKonstantin Belousov flags = OBJPR_CLEANONLY; 12556bbee8e2SAlan Cox vm_object_page_remove(object, OFF_TO_IDX(offset), 12566bbee8e2SAlan Cox OFF_TO_IDX(offset + size + PAGE_MASK), flags); 1257950f8459SAlan Cox } 125889f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1259126d6082SKonstantin Belousov return (res); 1260950f8459SAlan Cox } 1261950f8459SAlan Cox 1262950f8459SAlan Cox /* 1263aa3650eaSMark Johnston * Determine whether the given advice can be applied to the object. Advice is 1264aa3650eaSMark Johnston * not applied to unmanaged pages since they never belong to page queues, and 1265aa3650eaSMark Johnston * since MADV_FREE is destructive, it can apply only to anonymous pages that 1266aa3650eaSMark Johnston * have been mapped at most once. 1267aa3650eaSMark Johnston */ 1268aa3650eaSMark Johnston static bool 1269aa3650eaSMark Johnston vm_object_advice_applies(vm_object_t object, int advice) 1270aa3650eaSMark Johnston { 1271aa3650eaSMark Johnston 1272aa3650eaSMark Johnston if ((object->flags & OBJ_UNMANAGED) != 0) 1273aa3650eaSMark Johnston return (false); 1274aa3650eaSMark Johnston if (advice != MADV_FREE) 1275aa3650eaSMark Johnston return (true); 127663967687SJeff Roberson return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) == 127763967687SJeff Roberson (OBJ_ONEMAPPING | OBJ_ANON)); 1278aa3650eaSMark Johnston } 1279aa3650eaSMark Johnston 1280aa3650eaSMark Johnston static void 1281aa3650eaSMark Johnston vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex, 1282aa3650eaSMark Johnston vm_size_t size) 1283aa3650eaSMark Johnston { 1284aa3650eaSMark Johnston 12851390a5cbSKonstantin Belousov if (advice == MADV_FREE) 12861390a5cbSKonstantin Belousov vm_pager_freespace(object, pindex, size); 1287aa3650eaSMark Johnston } 1288aa3650eaSMark Johnston 1289aa3650eaSMark Johnston /* 1290867a482dSJohn Dyson * vm_object_madvise: 1291867a482dSJohn Dyson * 1292867a482dSJohn Dyson * Implements the madvise function at the object/page level. 12931c7c3c6aSMatthew Dillon * 1294193b9358SAlan Cox * MADV_WILLNEED (any object) 1295193b9358SAlan Cox * 1296193b9358SAlan Cox * Activate the specified pages if they are resident. 1297193b9358SAlan Cox * 1298193b9358SAlan Cox * MADV_DONTNEED (any object) 1299193b9358SAlan Cox * 1300193b9358SAlan Cox * Deactivate the specified pages if they are resident. 1301193b9358SAlan Cox * 1302193b9358SAlan Cox * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, 1303193b9358SAlan Cox * OBJ_ONEMAPPING only) 1304193b9358SAlan Cox * 1305193b9358SAlan Cox * Deactivate and clean the specified pages if they are 1306193b9358SAlan Cox * resident. This permits the process to reuse the pages 1307193b9358SAlan Cox * without faulting or the kernel to reclaim the pages 1308193b9358SAlan Cox * without I/O. 1309867a482dSJohn Dyson */ 1310867a482dSJohn Dyson void 131192a59946SJohn Baldwin vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, 1312c2655a40SMark Johnston int advice) 1313867a482dSJohn Dyson { 131492a59946SJohn Baldwin vm_pindex_t tpindex; 131534567de7SAlan Cox vm_object_t backing_object, tobject; 1316aa3650eaSMark Johnston vm_page_t m, tm; 1317867a482dSJohn Dyson 1318867a482dSJohn Dyson if (object == NULL) 1319867a482dSJohn Dyson return; 1320c2655a40SMark Johnston 13216e20a165SJohn Dyson relookup: 1322aa3650eaSMark Johnston VM_OBJECT_WLOCK(object); 1323aa3650eaSMark Johnston if (!vm_object_advice_applies(object, advice)) { 1324aa3650eaSMark Johnston VM_OBJECT_WUNLOCK(object); 1325aa3650eaSMark Johnston return; 13266e20a165SJohn Dyson } 1327aa3650eaSMark Johnston for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) { 1328aa3650eaSMark Johnston tobject = object; 1329c2655a40SMark Johnston 13301ce137beSMatthew Dillon /* 1331aa3650eaSMark Johnston * If the next page isn't resident in the top-level object, we 1332aa3650eaSMark Johnston * need to search the shadow chain. When applying MADV_FREE, we 1333aa3650eaSMark Johnston * take care to release any swap space used to store 1334aa3650eaSMark Johnston * non-resident pages. 1335aa3650eaSMark Johnston */ 1336aa3650eaSMark Johnston if (m == NULL || pindex < m->pindex) { 1337aa3650eaSMark Johnston /* 1338aa3650eaSMark Johnston * Optimize a common case: if the top-level object has 1339aa3650eaSMark Johnston * no backing object, we can skip over the non-resident 1340aa3650eaSMark Johnston * range in constant time. 13411ce137beSMatthew Dillon */ 1342c2655a40SMark Johnston if (object->backing_object == NULL) { 1343c2655a40SMark Johnston tpindex = (m != NULL && m->pindex < end) ? 1344c2655a40SMark Johnston m->pindex : end; 1345aa3650eaSMark Johnston vm_object_madvise_freespace(object, advice, 1346aa3650eaSMark Johnston pindex, tpindex - pindex); 1347c2655a40SMark Johnston if ((pindex = tpindex) == end) 1348c2655a40SMark Johnston break; 1349aa3650eaSMark Johnston goto next_page; 1350aa3650eaSMark Johnston } 1351aa3650eaSMark Johnston 1352aa3650eaSMark Johnston tpindex = pindex; 1353aa3650eaSMark Johnston do { 1354aa3650eaSMark Johnston vm_object_madvise_freespace(tobject, advice, 1355aa3650eaSMark Johnston tpindex, 1); 13561ce137beSMatthew Dillon /* 1357aa3650eaSMark Johnston * Prepare to search the next object in the 1358aa3650eaSMark Johnston * chain. 13591ce137beSMatthew Dillon */ 136034567de7SAlan Cox backing_object = tobject->backing_object; 136134567de7SAlan Cox if (backing_object == NULL) 1362aa3650eaSMark Johnston goto next_pindex; 136389f6b863SAttilio Rao VM_OBJECT_WLOCK(backing_object); 1364aa3650eaSMark Johnston tpindex += 1365aa3650eaSMark Johnston OFF_TO_IDX(tobject->backing_object_offset); 13669b98b796SAlan Cox if (tobject != object) 136789f6b863SAttilio Rao VM_OBJECT_WUNLOCK(tobject); 136834567de7SAlan Cox tobject = backing_object; 1369aa3650eaSMark Johnston if (!vm_object_advice_applies(tobject, advice)) 1370aa3650eaSMark Johnston goto next_pindex; 1371aa3650eaSMark Johnston } while ((tm = vm_page_lookup(tobject, tpindex)) == 1372aa3650eaSMark Johnston NULL); 1373aa3650eaSMark Johnston } else { 1374aa3650eaSMark Johnston next_page: 1375aa3650eaSMark Johnston tm = m; 1376aa3650eaSMark Johnston m = TAILQ_NEXT(m, listq); 1377c2655a40SMark Johnston } 1378c2655a40SMark Johnston 1379867a482dSJohn Dyson /* 13800012f373SJeff Roberson * If the page is not in a normal state, skip it. The page 13810012f373SJeff Roberson * can not be invalidated while the object lock is held. 1382867a482dSJohn Dyson */ 13830012f373SJeff Roberson if (!vm_page_all_valid(tm) || vm_page_wired(tm)) 1384aa3650eaSMark Johnston goto next_pindex; 1385aa3650eaSMark Johnston KASSERT((tm->flags & PG_FICTITIOUS) == 0, 1386aa3650eaSMark Johnston ("vm_object_madvise: page %p is fictitious", tm)); 1387aa3650eaSMark Johnston KASSERT((tm->oflags & VPO_UNMANAGED) == 0, 1388aa3650eaSMark Johnston ("vm_object_madvise: page %p is not managed", tm)); 138963e97555SJeff Roberson if (vm_page_tryxbusy(tm) == 0) { 1390aa3650eaSMark Johnston if (object != tobject) 1391aa3650eaSMark Johnston VM_OBJECT_WUNLOCK(object); 1392c2655a40SMark Johnston if (advice == MADV_WILLNEED) { 1393b11b56b5SAlan Cox /* 1394b11b56b5SAlan Cox * Reference the page before unlocking and 1395b11b56b5SAlan Cox * sleeping so that the page daemon is less 1396b11b56b5SAlan Cox * likely to reclaim it. 1397b11b56b5SAlan Cox */ 1398aa3650eaSMark Johnston vm_page_aflag_set(tm, PGA_REFERENCED); 1399567e51e1SAlan Cox } 140087b64663SMark Johnston if (!vm_page_busy_sleep(tm, "madvpo", 0)) 140187b64663SMark Johnston VM_OBJECT_WUNLOCK(tobject); 14026e20a165SJohn Dyson goto relookup; 140334567de7SAlan Cox } 1404aa3650eaSMark Johnston vm_page_advise(tm, advice); 140563e97555SJeff Roberson vm_page_xunbusy(tm); 1406aa3650eaSMark Johnston vm_object_madvise_freespace(tobject, advice, tm->pindex, 1); 1407aa3650eaSMark Johnston next_pindex: 14089b98b796SAlan Cox if (tobject != object) 140989f6b863SAttilio Rao VM_OBJECT_WUNLOCK(tobject); 1410867a482dSJohn Dyson } 141189f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1412867a482dSJohn Dyson } 1413867a482dSJohn Dyson 1414867a482dSJohn Dyson /* 1415df8bae1dSRodney W. Grimes * vm_object_shadow: 1416df8bae1dSRodney W. Grimes * 1417df8bae1dSRodney W. Grimes * Create a new object which is backed by the 1418df8bae1dSRodney W. Grimes * specified existing object range. The source 1419df8bae1dSRodney W. Grimes * object reference is deallocated. 1420df8bae1dSRodney W. Grimes * 1421df8bae1dSRodney W. Grimes * The new object and offset into that object 1422df8bae1dSRodney W. Grimes * are returned in the source parameters. 1423df8bae1dSRodney W. Grimes */ 142426f9a767SRodney W. Grimes void 142567388836SKonstantin Belousov vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length, 142667388836SKonstantin Belousov struct ucred *cred, bool shared) 1427df8bae1dSRodney W. Grimes { 1428d031cff1SMatthew Dillon vm_object_t source; 1429d031cff1SMatthew Dillon vm_object_t result; 1430df8bae1dSRodney W. Grimes 1431df8bae1dSRodney W. Grimes source = *object; 1432df8bae1dSRodney W. Grimes 1433df8bae1dSRodney W. Grimes /* 14349a2f6362SAlan Cox * Don't create the new object if the old object isn't shared. 143563967687SJeff Roberson * 143663967687SJeff Roberson * If we hold the only reference we can guarantee that it won't 143763967687SJeff Roberson * increase while we have the map locked. Otherwise the race is 143863967687SJeff Roberson * harmless and we will end up with an extra shadow object that 143963967687SJeff Roberson * will be collapsed later. 14409a2f6362SAlan Cox */ 144163967687SJeff Roberson if (source != NULL && source->ref_count == 1 && 144232362449SKonstantin Belousov (source->flags & OBJ_ANON) != 0) 14439a2f6362SAlan Cox return; 14449a2f6362SAlan Cox 14459a2f6362SAlan Cox /* 1446570a2f4aSAlan Cox * Allocate a new object with the given length. 1447df8bae1dSRodney W. Grimes */ 144867388836SKonstantin Belousov result = vm_object_allocate_anon(atop(length), source, cred, length); 1449df8bae1dSRodney W. Grimes 1450df8bae1dSRodney W. Grimes /* 145151b867e5SJeff Roberson * Store the offset into the source object, and fix up the offset into 145251b867e5SJeff Roberson * the new object. 145351b867e5SJeff Roberson */ 145451b867e5SJeff Roberson result->backing_object_offset = *offset; 145551b867e5SJeff Roberson 145667388836SKonstantin Belousov if (shared || source != NULL) { 145767388836SKonstantin Belousov VM_OBJECT_WLOCK(result); 145867388836SKonstantin Belousov 145951b867e5SJeff Roberson /* 146067388836SKonstantin Belousov * The new object shadows the source object, adding a 146167388836SKonstantin Belousov * reference to it. Our caller changes his reference 146267388836SKonstantin Belousov * to point to the new object, removing a reference to 146367388836SKonstantin Belousov * the source object. Net result: no change of 146467388836SKonstantin Belousov * reference count, unless the caller needs to add one 146567388836SKonstantin Belousov * more reference due to forking a shared map entry. 146667388836SKonstantin Belousov */ 146767388836SKonstantin Belousov if (shared) { 146867388836SKonstantin Belousov vm_object_reference_locked(result); 146967388836SKonstantin Belousov vm_object_clear_flag(result, OBJ_ONEMAPPING); 147067388836SKonstantin Belousov } 147167388836SKonstantin Belousov 147267388836SKonstantin Belousov /* 147367388836SKonstantin Belousov * Try to optimize the result object's page color when 147467388836SKonstantin Belousov * shadowing in order to maintain page coloring 147567388836SKonstantin Belousov * consistency in the combined shadowed object. 1476df8bae1dSRodney W. Grimes */ 1477570a2f4aSAlan Cox if (source != NULL) { 147851b867e5SJeff Roberson vm_object_backing_insert(result, source); 14793f289c3fSJeff Roberson result->domain = source->domain; 1480f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 14817b54b1a9SAlan Cox result->flags |= source->flags & OBJ_COLORED; 148267388836SKonstantin Belousov result->pg_color = (source->pg_color + 148367388836SKonstantin Belousov OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 148467388836SKonstantin Belousov 1)) - 1); 1485f8a47341SAlan Cox #endif 148667388836SKonstantin Belousov } 148751b867e5SJeff Roberson VM_OBJECT_WUNLOCK(result); 1488de5f6a77SJohn Dyson } 1489df8bae1dSRodney W. Grimes 1490df8bae1dSRodney W. Grimes /* 1491df8bae1dSRodney W. Grimes * Return the new things 1492df8bae1dSRodney W. Grimes */ 1493df8bae1dSRodney W. Grimes *offset = 0; 1494df8bae1dSRodney W. Grimes *object = result; 1495df8bae1dSRodney W. Grimes } 1496df8bae1dSRodney W. Grimes 1497c5aaa06dSAlan Cox /* 1498c5aaa06dSAlan Cox * vm_object_split: 1499c5aaa06dSAlan Cox * 1500c5aaa06dSAlan Cox * Split the pages in a map entry into a new object. This affords 1501c5aaa06dSAlan Cox * easier removal of unused pages, and keeps object inheritance from 1502c5aaa06dSAlan Cox * being a negative impact on memory usage. 1503c5aaa06dSAlan Cox */ 1504c5aaa06dSAlan Cox void 1505c5aaa06dSAlan Cox vm_object_split(vm_map_entry_t entry) 1506c5aaa06dSAlan Cox { 1507aec9e7d8SMark Johnston vm_page_t m, m_busy, m_next; 150898087a06SJeff Roberson vm_object_t orig_object, new_object, backing_object; 150973000556SAlan Cox vm_pindex_t idx, offidxstart; 151073000556SAlan Cox vm_size_t size; 1511c5aaa06dSAlan Cox 1512c5aaa06dSAlan Cox orig_object = entry->object.vm_object; 151398087a06SJeff Roberson KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0, 151498087a06SJeff Roberson ("vm_object_split: Splitting object with multiple mappings.")); 151563967687SJeff Roberson if ((orig_object->flags & OBJ_ANON) == 0) 1516c5aaa06dSAlan Cox return; 1517c5aaa06dSAlan Cox if (orig_object->ref_count <= 1) 1518c5aaa06dSAlan Cox return; 151989f6b863SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 1520c5aaa06dSAlan Cox 15214da9f125SAlan Cox offidxstart = OFF_TO_IDX(entry->offset); 152295442adfSAlan Cox size = atop(entry->end - entry->start); 1523c5aaa06dSAlan Cox 15244da9f125SAlan Cox /* 15254da9f125SAlan Cox * If swap_pager_copy() is later called, it will convert new_object 15264da9f125SAlan Cox * into a swap object. 15274da9f125SAlan Cox */ 152867388836SKonstantin Belousov new_object = vm_object_allocate_anon(size, orig_object, 152967388836SKonstantin Belousov orig_object->cred, ptoa(size)); 1530c5aaa06dSAlan Cox 1531c5474b8fSAlan Cox /* 153298087a06SJeff Roberson * We must wait for the orig_object to complete any in-progress 153398087a06SJeff Roberson * collapse so that the swap blocks are stable below. The 153498087a06SJeff Roberson * additional reference on backing_object by new object will 153598087a06SJeff Roberson * prevent further collapse operations until split completes. 153698087a06SJeff Roberson */ 153798087a06SJeff Roberson VM_OBJECT_WLOCK(orig_object); 153898087a06SJeff Roberson vm_object_collapse_wait(orig_object); 153998087a06SJeff Roberson 154098087a06SJeff Roberson /* 1541c5474b8fSAlan Cox * At this point, the new object is still private, so the order in 1542c5474b8fSAlan Cox * which the original and new objects are locked does not matter. 1543c5474b8fSAlan Cox */ 154489f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 15453f289c3fSJeff Roberson new_object->domain = orig_object->domain; 154698087a06SJeff Roberson backing_object = orig_object->backing_object; 154798087a06SJeff Roberson if (backing_object != NULL) { 154898087a06SJeff Roberson vm_object_backing_insert_ref(new_object, backing_object); 1549c5aaa06dSAlan Cox new_object->backing_object_offset = 15504da9f125SAlan Cox orig_object->backing_object_offset + entry->offset; 1551c5aaa06dSAlan Cox } 1552ef694c1aSEdward Tomasz Napierala if (orig_object->cred != NULL) { 1553ef694c1aSEdward Tomasz Napierala crhold(orig_object->cred); 15543364c323SKonstantin Belousov KASSERT(orig_object->charge >= ptoa(size), 15553364c323SKonstantin Belousov ("orig_object->charge < 0")); 15563364c323SKonstantin Belousov orig_object->charge -= ptoa(size); 15573364c323SKonstantin Belousov } 155898087a06SJeff Roberson 155998087a06SJeff Roberson /* 156098087a06SJeff Roberson * Mark the split operation so that swap_pager_getpages() knows 156198087a06SJeff Roberson * that the object is in transition. 156298087a06SJeff Roberson */ 156398087a06SJeff Roberson vm_object_set_flag(orig_object, OBJ_SPLIT); 1564aec9e7d8SMark Johnston m_busy = NULL; 1565aec9e7d8SMark Johnston #ifdef INVARIANTS 1566aec9e7d8SMark Johnston idx = 0; 1567aec9e7d8SMark Johnston #endif 1568c5aaa06dSAlan Cox retry: 1569b382c10aSKonstantin Belousov m = vm_page_find_least(orig_object, offidxstart); 1570aec9e7d8SMark Johnston KASSERT(m == NULL || idx <= m->pindex - offidxstart, 1571aec9e7d8SMark Johnston ("%s: object %p was repopulated", __func__, orig_object)); 157273000556SAlan Cox for (; m != NULL && (idx = m->pindex - offidxstart) < size; 157373000556SAlan Cox m = m_next) { 157473000556SAlan Cox m_next = TAILQ_NEXT(m, listq); 1575c5aaa06dSAlan Cox 1576c5aaa06dSAlan Cox /* 1577c5aaa06dSAlan Cox * We must wait for pending I/O to complete before we can 1578c5aaa06dSAlan Cox * rename the page. 1579c5aaa06dSAlan Cox * 1580c5aaa06dSAlan Cox * We do not have to VM_PROT_NONE the page as mappings should 1581c5aaa06dSAlan Cox * not be changed by this operation. 1582c5aaa06dSAlan Cox */ 158363e97555SJeff Roberson if (vm_page_tryxbusy(m) == 0) { 158489f6b863SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 158587b64663SMark Johnston if (vm_page_busy_sleep(m, "spltwt", 0)) 158687b64663SMark Johnston VM_OBJECT_WLOCK(orig_object); 158789f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 1588c5aaa06dSAlan Cox goto retry; 1589de33beddSAlan Cox } 1590e946b949SAttilio Rao 15914bf95d00SJeff Roberson /* 15924bf95d00SJeff Roberson * The page was left invalid. Likely placed there by 15934bf95d00SJeff Roberson * an incomplete fault. Just remove and ignore. 15944bf95d00SJeff Roberson */ 15954bf95d00SJeff Roberson if (vm_page_none_valid(m)) { 15964bf95d00SJeff Roberson if (vm_page_remove(m)) 15974bf95d00SJeff Roberson vm_page_free(m); 15984bf95d00SJeff Roberson continue; 15994bf95d00SJeff Roberson } 16004bf95d00SJeff Roberson 16013453bca8SAlan Cox /* vm_page_rename() will dirty the page. */ 1602e946b949SAttilio Rao if (vm_page_rename(m, new_object, idx)) { 160363e97555SJeff Roberson vm_page_xunbusy(m); 1604e946b949SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 1605e946b949SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 16068d6fbbb8SJeff Roberson vm_radix_wait(); 1607e946b949SAttilio Rao VM_OBJECT_WLOCK(orig_object); 1608e946b949SAttilio Rao VM_OBJECT_WLOCK(new_object); 1609e946b949SAttilio Rao goto retry; 1610e946b949SAttilio Rao } 161163e97555SJeff Roberson 1612b5f359b7SAlan Cox #if VM_NRESERVLEVEL > 0 1613b5f359b7SAlan Cox /* 1614b5f359b7SAlan Cox * If some of the reservation's allocated pages remain with 1615b5f359b7SAlan Cox * the original object, then transferring the reservation to 1616b5f359b7SAlan Cox * the new object is neither particularly beneficial nor 1617b5f359b7SAlan Cox * particularly harmful as compared to leaving the reservation 1618b5f359b7SAlan Cox * with the original object. If, however, all of the 1619b5f359b7SAlan Cox * reservation's allocated pages are transferred to the new 1620b5f359b7SAlan Cox * object, then transferring the reservation is typically 1621b5f359b7SAlan Cox * beneficial. Determining which of these two cases applies 1622b5f359b7SAlan Cox * would be more costly than unconditionally renaming the 1623b5f359b7SAlan Cox * reservation. 1624b5f359b7SAlan Cox */ 1625b5f359b7SAlan Cox vm_reserv_rename(m, new_object, orig_object, offidxstart); 1626b5f359b7SAlan Cox #endif 1627aec9e7d8SMark Johnston 1628aec9e7d8SMark Johnston /* 1629aec9e7d8SMark Johnston * orig_object's type may change while sleeping, so keep track 1630aec9e7d8SMark Johnston * of the beginning of the busied range. 1631aec9e7d8SMark Johnston */ 16328da1c098SJeff Roberson if (orig_object->type != OBJT_SWAP) 16338da1c098SJeff Roberson vm_page_xunbusy(m); 1634aec9e7d8SMark Johnston else if (m_busy == NULL) 1635aec9e7d8SMark Johnston m_busy = m; 1636c5aaa06dSAlan Cox } 16374b8365d7SKonstantin Belousov if ((orig_object->flags & OBJ_SWAP) != 0) { 1638c5aaa06dSAlan Cox /* 1639c7c8dd7eSAlan Cox * swap_pager_copy() can sleep, in which case the orig_object's 1640c7c8dd7eSAlan Cox * and new_object's locks are released and reacquired. 1641c5aaa06dSAlan Cox */ 1642c5aaa06dSAlan Cox swap_pager_copy(orig_object, new_object, offidxstart, 0); 1643aec9e7d8SMark Johnston if (m_busy != NULL) 1644aec9e7d8SMark Johnston TAILQ_FOREACH_FROM(m_busy, &new_object->memq, listq) 1645aec9e7d8SMark Johnston vm_page_xunbusy(m_busy); 1646c5aaa06dSAlan Cox } 164798087a06SJeff Roberson vm_object_clear_flag(orig_object, OBJ_SPLIT); 164889f6b863SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 164989f6b863SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 1650c5aaa06dSAlan Cox entry->object.vm_object = new_object; 1651c5aaa06dSAlan Cox entry->offset = 0LL; 1652c5aaa06dSAlan Cox vm_object_deallocate(orig_object); 165389f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 1654c5aaa06dSAlan Cox } 1655c5aaa06dSAlan Cox 165699a1570aSKonstantin Belousov static vm_page_t 165798087a06SJeff Roberson vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p) 165899a1570aSKonstantin Belousov { 165999a1570aSKonstantin Belousov vm_object_t backing_object; 166099a1570aSKonstantin Belousov 166199a1570aSKonstantin Belousov VM_OBJECT_ASSERT_WLOCKED(object); 166299a1570aSKonstantin Belousov backing_object = object->backing_object; 166399a1570aSKonstantin Belousov VM_OBJECT_ASSERT_WLOCKED(backing_object); 166499a1570aSKonstantin Belousov 166599a1570aSKonstantin Belousov KASSERT(p == NULL || p->object == object || p->object == backing_object, 166699a1570aSKonstantin Belousov ("invalid ownership %p %p %p", p, object, backing_object)); 16678d6fbbb8SJeff Roberson /* The page is only NULL when rename fails. */ 16684cdea4a8SJeff Roberson if (p == NULL) { 16696a14746cSRyan Libby VM_OBJECT_WUNLOCK(object); 16706a14746cSRyan Libby VM_OBJECT_WUNLOCK(backing_object); 16718d6fbbb8SJeff Roberson vm_radix_wait(); 167299a1570aSKonstantin Belousov VM_OBJECT_WLOCK(object); 167387b64663SMark Johnston } else if (p->object == object) { 167487b64663SMark Johnston VM_OBJECT_WUNLOCK(backing_object); 167587b64663SMark Johnston if (vm_page_busy_sleep(p, "vmocol", 0)) 167687b64663SMark Johnston VM_OBJECT_WLOCK(object); 167787b64663SMark Johnston } else { 167887b64663SMark Johnston VM_OBJECT_WUNLOCK(object); 167987b64663SMark Johnston if (!vm_page_busy_sleep(p, "vmocol", 0)) 168087b64663SMark Johnston VM_OBJECT_WUNLOCK(backing_object); 168187b64663SMark Johnston VM_OBJECT_WLOCK(object); 168287b64663SMark Johnston } 168399a1570aSKonstantin Belousov VM_OBJECT_WLOCK(backing_object); 168499a1570aSKonstantin Belousov return (TAILQ_FIRST(&backing_object->memq)); 168599a1570aSKonstantin Belousov } 168699a1570aSKonstantin Belousov 168799a1570aSKonstantin Belousov static bool 16884cc8daf7SConrad Meyer vm_object_scan_all_shadowed(vm_object_t object) 16894cc8daf7SConrad Meyer { 16904cc8daf7SConrad Meyer vm_object_t backing_object; 16914cc8daf7SConrad Meyer vm_page_t p, pp; 169277d6fd97SKonstantin Belousov vm_pindex_t backing_offset_index, new_pindex, pi, ps; 16934cc8daf7SConrad Meyer 16944cc8daf7SConrad Meyer VM_OBJECT_ASSERT_WLOCKED(object); 16954cc8daf7SConrad Meyer VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 16964cc8daf7SConrad Meyer 16974cc8daf7SConrad Meyer backing_object = object->backing_object; 16984cc8daf7SConrad Meyer 169963967687SJeff Roberson if ((backing_object->flags & OBJ_ANON) == 0) 17004cc8daf7SConrad Meyer return (false); 17014cc8daf7SConrad Meyer 170277d6fd97SKonstantin Belousov pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 170377d6fd97SKonstantin Belousov p = vm_page_find_least(backing_object, pi); 170477d6fd97SKonstantin Belousov ps = swap_pager_find_least(backing_object, pi); 17054cc8daf7SConrad Meyer 17064cc8daf7SConrad Meyer /* 170777d6fd97SKonstantin Belousov * Only check pages inside the parent object's range and 170877d6fd97SKonstantin Belousov * inside the parent object's mapping of the backing object. 17094cc8daf7SConrad Meyer */ 171077d6fd97SKonstantin Belousov for (;; pi++) { 171177d6fd97SKonstantin Belousov if (p != NULL && p->pindex < pi) 171277d6fd97SKonstantin Belousov p = TAILQ_NEXT(p, listq); 171377d6fd97SKonstantin Belousov if (ps < pi) 171477d6fd97SKonstantin Belousov ps = swap_pager_find_least(backing_object, pi); 171577d6fd97SKonstantin Belousov if (p == NULL && ps >= backing_object->size) 171677d6fd97SKonstantin Belousov break; 171777d6fd97SKonstantin Belousov else if (p == NULL) 171877d6fd97SKonstantin Belousov pi = ps; 171977d6fd97SKonstantin Belousov else 172077d6fd97SKonstantin Belousov pi = MIN(p->pindex, ps); 172177d6fd97SKonstantin Belousov 172277d6fd97SKonstantin Belousov new_pindex = pi - backing_offset_index; 172377d6fd97SKonstantin Belousov if (new_pindex >= object->size) 172477d6fd97SKonstantin Belousov break; 17254cc8daf7SConrad Meyer 1726cd0047f3SKonstantin Belousov if (p != NULL) { 17274cc8daf7SConrad Meyer /* 1728cd0047f3SKonstantin Belousov * If the backing object page is busy a 1729cd0047f3SKonstantin Belousov * grandparent or older page may still be 1730cd0047f3SKonstantin Belousov * undergoing CoW. It is not safe to collapse 1731cd0047f3SKonstantin Belousov * the backing object until it is quiesced. 173258447749SJeff Roberson */ 1733cd0047f3SKonstantin Belousov if (vm_page_tryxbusy(p) == 0) 173458447749SJeff Roberson return (false); 173558447749SJeff Roberson 173658447749SJeff Roberson /* 1737cd0047f3SKonstantin Belousov * We raced with the fault handler that left 1738cd0047f3SKonstantin Belousov * newly allocated invalid page on the object 1739cd0047f3SKonstantin Belousov * queue and retried. 1740cd0047f3SKonstantin Belousov */ 1741cd0047f3SKonstantin Belousov if (!vm_page_all_valid(p)) 1742cd0047f3SKonstantin Belousov goto unbusy_ret; 1743cd0047f3SKonstantin Belousov } 1744cd0047f3SKonstantin Belousov 1745cd0047f3SKonstantin Belousov /* 17464cc8daf7SConrad Meyer * See if the parent has the page or if the parent's object 17474cc8daf7SConrad Meyer * pager has the page. If the parent has the page but the page 17484cc8daf7SConrad Meyer * is not valid, the parent's object pager must have the page. 17494cc8daf7SConrad Meyer * 17504cc8daf7SConrad Meyer * If this fails, the parent does not completely shadow the 17514cc8daf7SConrad Meyer * object and we might as well give up now. 17524cc8daf7SConrad Meyer */ 17534cc8daf7SConrad Meyer pp = vm_page_lookup(object, new_pindex); 1754cd0047f3SKonstantin Belousov 17550012f373SJeff Roberson /* 1756cd0047f3SKonstantin Belousov * The valid check here is stable due to object lock 1757cd0047f3SKonstantin Belousov * being required to clear valid and initiate paging. 1758cd0047f3SKonstantin Belousov * Busy of p disallows fault handler to validate pp. 17590012f373SJeff Roberson */ 17600012f373SJeff Roberson if ((pp == NULL || vm_page_none_valid(pp)) && 17614cc8daf7SConrad Meyer !vm_pager_has_page(object, new_pindex, NULL, NULL)) 1762cd0047f3SKonstantin Belousov goto unbusy_ret; 1763cd0047f3SKonstantin Belousov if (p != NULL) 1764cd0047f3SKonstantin Belousov vm_page_xunbusy(p); 17654cc8daf7SConrad Meyer } 17664cc8daf7SConrad Meyer return (true); 1767cd0047f3SKonstantin Belousov 1768cd0047f3SKonstantin Belousov unbusy_ret: 1769cd0047f3SKonstantin Belousov if (p != NULL) 1770cd0047f3SKonstantin Belousov vm_page_xunbusy(p); 1771cd0047f3SKonstantin Belousov return (false); 17724cc8daf7SConrad Meyer } 17734cc8daf7SConrad Meyer 177498087a06SJeff Roberson static void 177598087a06SJeff Roberson vm_object_collapse_scan(vm_object_t object) 17762ad1a3f7SMatthew Dillon { 17772ad1a3f7SMatthew Dillon vm_object_t backing_object; 177899a1570aSKonstantin Belousov vm_page_t next, p, pp; 177999a1570aSKonstantin Belousov vm_pindex_t backing_offset_index, new_pindex; 17802ad1a3f7SMatthew Dillon 178189f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 178289f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 17832ad1a3f7SMatthew Dillon 17842ad1a3f7SMatthew Dillon backing_object = object->backing_object; 17852ad1a3f7SMatthew Dillon backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 17862ad1a3f7SMatthew Dillon 17872ad1a3f7SMatthew Dillon /* 17882ad1a3f7SMatthew Dillon * Our scan 17892ad1a3f7SMatthew Dillon */ 17904cc8daf7SConrad Meyer for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) { 179199a1570aSKonstantin Belousov next = TAILQ_NEXT(p, listq); 179299a1570aSKonstantin Belousov new_pindex = p->pindex - backing_offset_index; 17932ad1a3f7SMatthew Dillon 17942ad1a3f7SMatthew Dillon /* 17952ad1a3f7SMatthew Dillon * Check for busy page 17962ad1a3f7SMatthew Dillon */ 179763e97555SJeff Roberson if (vm_page_tryxbusy(p) == 0) { 179898087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, p); 17992ad1a3f7SMatthew Dillon continue; 18002ad1a3f7SMatthew Dillon } 18012ad1a3f7SMatthew Dillon 180298087a06SJeff Roberson KASSERT(object->backing_object == backing_object, 180398087a06SJeff Roberson ("vm_object_collapse_scan: backing object mismatch %p != %p", 180498087a06SJeff Roberson object->backing_object, backing_object)); 180599a1570aSKonstantin Belousov KASSERT(p->object == backing_object, 180698087a06SJeff Roberson ("vm_object_collapse_scan: object mismatch %p != %p", 180798087a06SJeff Roberson p->object, backing_object)); 18082ad1a3f7SMatthew Dillon 180999a1570aSKonstantin Belousov if (p->pindex < backing_offset_index || 181099a1570aSKonstantin Belousov new_pindex >= object->size) { 18111390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, p->pindex, 1); 1812e946b949SAttilio Rao 1813f6d89838SAlan Cox KASSERT(!pmap_page_is_mapped(p), 1814f6d89838SAlan Cox ("freeing mapped page %p", p)); 18150fd977b3SMark Johnston if (vm_page_remove(p)) 18162ad1a3f7SMatthew Dillon vm_page_free(p); 18172ad1a3f7SMatthew Dillon continue; 18182ad1a3f7SMatthew Dillon } 18192ad1a3f7SMatthew Dillon 1820cd0047f3SKonstantin Belousov if (!vm_page_all_valid(p)) { 1821cd0047f3SKonstantin Belousov KASSERT(!pmap_page_is_mapped(p), 1822cd0047f3SKonstantin Belousov ("freeing mapped page %p", p)); 1823cd0047f3SKonstantin Belousov if (vm_page_remove(p)) 1824cd0047f3SKonstantin Belousov vm_page_free(p); 1825cd0047f3SKonstantin Belousov continue; 1826cd0047f3SKonstantin Belousov } 1827cd0047f3SKonstantin Belousov 18282ad1a3f7SMatthew Dillon pp = vm_page_lookup(object, new_pindex); 182963e97555SJeff Roberson if (pp != NULL && vm_page_tryxbusy(pp) == 0) { 183063e97555SJeff Roberson vm_page_xunbusy(p); 1831e18cc7bfSMax Laier /* 18324cc8daf7SConrad Meyer * The page in the parent is busy and possibly not 18334cc8daf7SConrad Meyer * (yet) valid. Until its state is finalized by the 18344cc8daf7SConrad Meyer * busy bit owner, we can't tell whether it shadows the 183598087a06SJeff Roberson * original page. 1836e18cc7bfSMax Laier */ 183798087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, pp); 1838e18cc7bfSMax Laier continue; 1839e18cc7bfSMax Laier } 184099a1570aSKonstantin Belousov 18414bf95d00SJeff Roberson if (pp != NULL && vm_page_none_valid(pp)) { 18424bf95d00SJeff Roberson /* 18434bf95d00SJeff Roberson * The page was invalid in the parent. Likely placed 18444bf95d00SJeff Roberson * there by an incomplete fault. Just remove and 18454bf95d00SJeff Roberson * ignore. p can replace it. 18464bf95d00SJeff Roberson */ 18474bf95d00SJeff Roberson if (vm_page_remove(pp)) 18484bf95d00SJeff Roberson vm_page_free(pp); 18494bf95d00SJeff Roberson pp = NULL; 18504bf95d00SJeff Roberson } 185199a1570aSKonstantin Belousov 18524cc8daf7SConrad Meyer if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL, 18534cc8daf7SConrad Meyer NULL)) { 185499a1570aSKonstantin Belousov /* 18554cc8daf7SConrad Meyer * The page already exists in the parent OR swap exists 18564cc8daf7SConrad Meyer * for this location in the parent. Leave the parent's 18574cc8daf7SConrad Meyer * page alone. Destroy the original page from the 18584cc8daf7SConrad Meyer * backing object. 185999a1570aSKonstantin Belousov */ 18601390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, p->pindex, 1); 1861f6d89838SAlan Cox KASSERT(!pmap_page_is_mapped(p), 1862f6d89838SAlan Cox ("freeing mapped page %p", p)); 18630fd977b3SMark Johnston if (vm_page_remove(p)) 18642ad1a3f7SMatthew Dillon vm_page_free(p); 186563e97555SJeff Roberson if (pp != NULL) 186663e97555SJeff Roberson vm_page_xunbusy(pp); 18672ad1a3f7SMatthew Dillon continue; 18682ad1a3f7SMatthew Dillon } 18692ad1a3f7SMatthew Dillon 1870e946b949SAttilio Rao /* 18714cc8daf7SConrad Meyer * Page does not exist in parent, rename the page from the 18724cc8daf7SConrad Meyer * backing object to the main object. 1873e946b949SAttilio Rao * 18744cc8daf7SConrad Meyer * If the page was mapped to a process, it can remain mapped 18753453bca8SAlan Cox * through the rename. vm_page_rename() will dirty the page. 1876e946b949SAttilio Rao */ 1877e946b949SAttilio Rao if (vm_page_rename(p, object, new_pindex)) { 187863e97555SJeff Roberson vm_page_xunbusy(p); 187998087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, NULL); 1880e946b949SAttilio Rao continue; 1881e946b949SAttilio Rao } 188214a5dc17SAttilio Rao 188314a5dc17SAttilio Rao /* Use the old pindex to free the right page. */ 18841390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, new_pindex + 18851390a5cbSKonstantin Belousov backing_offset_index, 1); 1886e946b949SAttilio Rao 1887f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 1888f8a47341SAlan Cox /* 1889f8a47341SAlan Cox * Rename the reservation. 1890f8a47341SAlan Cox */ 1891f8a47341SAlan Cox vm_reserv_rename(p, object, backing_object, 1892f8a47341SAlan Cox backing_offset_index); 1893f8a47341SAlan Cox #endif 18948da1c098SJeff Roberson vm_page_xunbusy(p); 18952ad1a3f7SMatthew Dillon } 18962fe6e4d7SDavid Greenman return; 18972fe6e4d7SDavid Greenman } 18982fe6e4d7SDavid Greenman 1899df8bae1dSRodney W. Grimes /* 1900df8bae1dSRodney W. Grimes * vm_object_collapse: 1901df8bae1dSRodney W. Grimes * 1902df8bae1dSRodney W. Grimes * Collapse an object with the object backing it. 1903df8bae1dSRodney W. Grimes * Pages in the backing object are moved into the 1904df8bae1dSRodney W. Grimes * parent, and the backing object is deallocated. 1905df8bae1dSRodney W. Grimes */ 190626f9a767SRodney W. Grimes void 19071b40f8c0SMatthew Dillon vm_object_collapse(vm_object_t object) 1908df8bae1dSRodney W. Grimes { 190998f139daSKonstantin Belousov vm_object_t backing_object, new_backing_object; 191098f139daSKonstantin Belousov 191189f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 191223955314SAlfred Perlstein 1913df8bae1dSRodney W. Grimes while (TRUE) { 191498087a06SJeff Roberson KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON, 191598087a06SJeff Roberson ("collapsing invalid object")); 191698087a06SJeff Roberson 1917df8bae1dSRodney W. Grimes /* 191898087a06SJeff Roberson * Wait for the backing_object to finish any pending 191998087a06SJeff Roberson * collapse so that the caller sees the shortest possible 192098087a06SJeff Roberson * shadow chain. 1921df8bae1dSRodney W. Grimes */ 192298087a06SJeff Roberson backing_object = vm_object_backing_collapse_wait(object); 192398087a06SJeff Roberson if (backing_object == NULL) 192498087a06SJeff Roberson return; 192598087a06SJeff Roberson 192698087a06SJeff Roberson KASSERT(object->ref_count > 0 && 192798087a06SJeff Roberson object->ref_count > object->shadow_count, 192898087a06SJeff Roberson ("collapse with invalid ref %d or shadow %d count.", 192998087a06SJeff Roberson object->ref_count, object->shadow_count)); 193098087a06SJeff Roberson KASSERT((backing_object->flags & 193198087a06SJeff Roberson (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 193298087a06SJeff Roberson ("vm_object_collapse: Backing object already collapsing.")); 193398087a06SJeff Roberson KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 193498087a06SJeff Roberson ("vm_object_collapse: object is already collapsing.")); 1935df8bae1dSRodney W. Grimes 1936f919ebdeSDavid Greenman /* 193798087a06SJeff Roberson * We know that we can either collapse the backing object if 193898087a06SJeff Roberson * the parent is the only reference to it, or (perhaps) have 19392ad1a3f7SMatthew Dillon * the parent bypass the object if the parent happens to shadow 19402ad1a3f7SMatthew Dillon * all the resident pages in the entire backing object. 1941df8bae1dSRodney W. Grimes */ 1942df8bae1dSRodney W. Grimes if (backing_object->ref_count == 1) { 194398087a06SJeff Roberson KASSERT(backing_object->shadow_count == 1, 194498087a06SJeff Roberson ("vm_object_collapse: shadow_count: %d", 194598087a06SJeff Roberson backing_object->shadow_count)); 1946aa9bc3b1SKonstantin Belousov vm_object_pip_add(object, 1); 194798087a06SJeff Roberson vm_object_set_flag(object, OBJ_COLLAPSING); 1948aa9bc3b1SKonstantin Belousov vm_object_pip_add(backing_object, 1); 194998087a06SJeff Roberson vm_object_set_flag(backing_object, OBJ_DEAD); 1950aa9bc3b1SKonstantin Belousov 1951df8bae1dSRodney W. Grimes /* 19522ad1a3f7SMatthew Dillon * If there is exactly one reference to the backing 19532ad1a3f7SMatthew Dillon * object, we can collapse it into the parent. 1954df8bae1dSRodney W. Grimes */ 195598087a06SJeff Roberson vm_object_collapse_scan(object); 1956df8bae1dSRodney W. Grimes 1957f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 1958f8a47341SAlan Cox /* 1959f8a47341SAlan Cox * Break any reservations from backing_object. 1960f8a47341SAlan Cox */ 1961f8a47341SAlan Cox if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) 1962f8a47341SAlan Cox vm_reserv_break_all(backing_object); 1963f8a47341SAlan Cox #endif 1964f8a47341SAlan Cox 1965df8bae1dSRodney W. Grimes /* 1966df8bae1dSRodney W. Grimes * Move the pager from backing_object to object. 1967df8bae1dSRodney W. Grimes */ 19684b8365d7SKonstantin Belousov if ((backing_object->flags & OBJ_SWAP) != 0) { 196924a1cce3SDavid Greenman /* 1970c7c8dd7eSAlan Cox * swap_pager_copy() can sleep, in which case 1971c7c8dd7eSAlan Cox * the backing_object's and object's locks are 1972c7c8dd7eSAlan Cox * released and reacquired. 1973571a1e92SAttilio Rao * Since swap_pager_copy() is being asked to 197498087a06SJeff Roberson * destroy backing_object, it will change the 197598087a06SJeff Roberson * type to OBJT_DEFAULT. 197624a1cce3SDavid Greenman */ 19771c7c3c6aSMatthew Dillon swap_pager_copy( 19781c7c3c6aSMatthew Dillon backing_object, 19791c7c3c6aSMatthew Dillon object, 19801c7c3c6aSMatthew Dillon OFF_TO_IDX(object->backing_object_offset), TRUE); 1981c0503609SDavid Greenman } 198298087a06SJeff Roberson 1983df8bae1dSRodney W. Grimes /* 1984df8bae1dSRodney W. Grimes * Object now shadows whatever backing_object did. 1985df8bae1dSRodney W. Grimes */ 198698087a06SJeff Roberson vm_object_clear_flag(object, OBJ_COLLAPSING); 198798087a06SJeff Roberson vm_object_backing_transfer(object, backing_object); 19882ad1a3f7SMatthew Dillon object->backing_object_offset += 19892ad1a3f7SMatthew Dillon backing_object->backing_object_offset; 199098087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 199198087a06SJeff Roberson vm_object_pip_wakeup(object); 19922ad1a3f7SMatthew Dillon 1993df8bae1dSRodney W. Grimes /* 1994df8bae1dSRodney W. Grimes * Discard backing_object. 1995df8bae1dSRodney W. Grimes * 19960d94caffSDavid Greenman * Since the backing object has no pages, no pager left, 19970d94caffSDavid Greenman * and no object references within it, all that is 19980d94caffSDavid Greenman * necessary is to dispose of it. 1999df8bae1dSRodney W. Grimes */ 20009b4d473aSKonstantin Belousov KASSERT(backing_object->ref_count == 1, ( 20019b4d473aSKonstantin Belousov "backing_object %p was somehow re-referenced during collapse!", 20029b4d473aSKonstantin Belousov backing_object)); 2003aa9bc3b1SKonstantin Belousov vm_object_pip_wakeup(backing_object); 200498087a06SJeff Roberson (void)refcount_release(&backing_object->ref_count); 200598087a06SJeff Roberson vm_object_terminate(backing_object); 200611542376SAlan Cox counter_u64_add(object_collapses, 1); 200798087a06SJeff Roberson VM_OBJECT_WLOCK(object); 20080d94caffSDavid Greenman } else { 2009df8bae1dSRodney W. Grimes /* 20102ad1a3f7SMatthew Dillon * If we do not entirely shadow the backing object, 20112ad1a3f7SMatthew Dillon * there is nothing we can do so we give up. 201298087a06SJeff Roberson * 201398087a06SJeff Roberson * The object lock and backing_object lock must not 201498087a06SJeff Roberson * be dropped during this sequence. 2015df8bae1dSRodney W. Grimes */ 201658447749SJeff Roberson if (!vm_object_scan_all_shadowed(object)) { 201789f6b863SAttilio Rao VM_OBJECT_WUNLOCK(backing_object); 20182ad1a3f7SMatthew Dillon break; 201924a1cce3SDavid Greenman } 2020df8bae1dSRodney W. Grimes 2021df8bae1dSRodney W. Grimes /* 20220d94caffSDavid Greenman * Make the parent shadow the next object in the 20230d94caffSDavid Greenman * chain. Deallocating backing_object will not remove 20240d94caffSDavid Greenman * it, since its reference count is at least 2. 2025df8bae1dSRodney W. Grimes */ 202651b867e5SJeff Roberson vm_object_backing_remove_locked(object); 202795e5e988SJohn Dyson new_backing_object = backing_object->backing_object; 202851b867e5SJeff Roberson if (new_backing_object != NULL) { 202998087a06SJeff Roberson vm_object_backing_insert_ref(object, 203051b867e5SJeff Roberson new_backing_object); 203195e5e988SJohn Dyson object->backing_object_offset += 203295e5e988SJohn Dyson backing_object->backing_object_offset; 2033de5f6a77SJohn Dyson } 2034df8bae1dSRodney W. Grimes 2035df8bae1dSRodney W. Grimes /* 20360d94caffSDavid Greenman * Drop the reference count on backing_object. Since 203722ec553fSAlan Cox * its ref_count was at least 2, it will not vanish. 2038df8bae1dSRodney W. Grimes */ 203998087a06SJeff Roberson (void)refcount_release(&backing_object->ref_count); 204098087a06SJeff Roberson KASSERT(backing_object->ref_count >= 1, ( 204198087a06SJeff Roberson "backing_object %p was somehow dereferenced during collapse!", 204298087a06SJeff Roberson backing_object)); 204389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(backing_object); 204411542376SAlan Cox counter_u64_add(object_bypasses, 1); 2045df8bae1dSRodney W. Grimes } 2046df8bae1dSRodney W. Grimes 2047df8bae1dSRodney W. Grimes /* 2048df8bae1dSRodney W. Grimes * Try again with this object's new backing object. 2049df8bae1dSRodney W. Grimes */ 2050df8bae1dSRodney W. Grimes } 2051df8bae1dSRodney W. Grimes } 2052df8bae1dSRodney W. Grimes 2053df8bae1dSRodney W. Grimes /* 2054bff99f0dSAlan Cox * vm_object_page_remove: 2055df8bae1dSRodney W. Grimes * 205668855966SAlan Cox * For the given object, either frees or invalidates each of the 20576bbee8e2SAlan Cox * specified pages. In general, a page is freed. However, if a page is 20586bbee8e2SAlan Cox * wired for any reason other than the existence of a managed, wired 20596bbee8e2SAlan Cox * mapping, then it may be invalidated but not removed from the object. 20606bbee8e2SAlan Cox * Pages are specified by the given range ["start", "end") and the option 20616bbee8e2SAlan Cox * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range 20626bbee8e2SAlan Cox * extends from "start" to the end of the object. If the option 20636bbee8e2SAlan Cox * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the 20646bbee8e2SAlan Cox * specified range are affected. If the option OBJPR_NOTMAPPED is 20656bbee8e2SAlan Cox * specified, then the pages within the specified range must have no 20666bbee8e2SAlan Cox * mappings. Otherwise, if this option is not specified, any mappings to 20676bbee8e2SAlan Cox * the specified pages are removed before the pages are freed or 20686bbee8e2SAlan Cox * invalidated. 206968855966SAlan Cox * 20706bbee8e2SAlan Cox * In general, this operation should only be performed on objects that 20716bbee8e2SAlan Cox * contain managed pages. There are, however, two exceptions. First, it 20726bbee8e2SAlan Cox * is performed on the kernel and kmem objects by vm_map_entry_delete(). 20736bbee8e2SAlan Cox * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- 20746bbee8e2SAlan Cox * backed pages. In both of these cases, the option OBJPR_CLEANONLY must 20756bbee8e2SAlan Cox * not be specified and the option OBJPR_NOTMAPPED must be specified. 2076df8bae1dSRodney W. Grimes * 2077df8bae1dSRodney W. Grimes * The object must be locked. 2078df8bae1dSRodney W. Grimes */ 207926f9a767SRodney W. Grimes void 2080ecde4b32SAlan Cox vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 20816bbee8e2SAlan Cox int options) 2082df8bae1dSRodney W. Grimes { 2083d031cff1SMatthew Dillon vm_page_t p, next; 2084df8bae1dSRodney W. Grimes 208589f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 208628634820SAlan Cox KASSERT((object->flags & OBJ_UNMANAGED) == 0 || 20876bbee8e2SAlan Cox (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, 20886bbee8e2SAlan Cox ("vm_object_page_remove: illegal options for object %p", object)); 2089ecde4b32SAlan Cox if (object->resident_page_count == 0) 20907667839aSAlan Cox return; 2091d474eaaaSDoug Rabson vm_object_pip_add(object, 1); 209226f9a767SRodney W. Grimes again: 2093b382c10aSKonstantin Belousov p = vm_page_find_least(object, start); 20942965a453SKip Macy 209575741c04SAlan Cox /* 20966bbee8e2SAlan Cox * Here, the variable "p" is either (1) the page with the least pindex 20976bbee8e2SAlan Cox * greater than or equal to the parameter "start" or (2) NULL. 209875741c04SAlan Cox */ 20996bbee8e2SAlan Cox for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2100b18bfc3dSJohn Dyson next = TAILQ_NEXT(p, listq); 210175741c04SAlan Cox 210259677d3cSAlan Cox /* 2103d28af1abSMark Johnston * Skip invalid pages if asked to do so. Try to avoid acquiring 2104d28af1abSMark Johnston * the busy lock, as some consumers rely on this to avoid 2105d28af1abSMark Johnston * deadlocks. 2106d28af1abSMark Johnston * 2107d28af1abSMark Johnston * A thread may concurrently transition the page from invalid to 2108d28af1abSMark Johnston * valid using only the busy lock, so the result of this check 2109d28af1abSMark Johnston * is immediately stale. It is up to consumers to handle this, 2110d28af1abSMark Johnston * for instance by ensuring that all invalid->valid transitions 2111d28af1abSMark Johnston * happen with a mutex held, as may be possible for a 2112d28af1abSMark Johnston * filesystem. 2113d28af1abSMark Johnston */ 2114d28af1abSMark Johnston if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) 2115d28af1abSMark Johnston continue; 2116d28af1abSMark Johnston 2117d28af1abSMark Johnston /* 21186bbee8e2SAlan Cox * If the page is wired for any reason besides the existence 21196bbee8e2SAlan Cox * of managed, wired mappings, then it cannot be freed. For 21206bbee8e2SAlan Cox * example, fictitious pages, which represent device memory, 21216bbee8e2SAlan Cox * are inherently wired and cannot be freed. They can, 21226bbee8e2SAlan Cox * however, be invalidated if the option OBJPR_CLEANONLY is 21236bbee8e2SAlan Cox * not specified. 212459677d3cSAlan Cox */ 212563e97555SJeff Roberson if (vm_page_tryxbusy(p) == 0) { 212687b64663SMark Johnston if (vm_page_busy_sleep(p, "vmopar", 0)) 212787b64663SMark Johnston VM_OBJECT_WLOCK(object); 2128fee2a2faSMark Johnston goto again; 2129fee2a2faSMark Johnston } 2130d28af1abSMark Johnston if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) { 2131d28af1abSMark Johnston vm_page_xunbusy(p); 2132d28af1abSMark Johnston continue; 2133d28af1abSMark Johnston } 2134d842aa51SMark Johnston if (vm_page_wired(p)) { 2135fee2a2faSMark Johnston wired: 2136cf060942SAlan Cox if ((options & OBJPR_NOTMAPPED) == 0 && 2137cf060942SAlan Cox object->ref_count != 0) 21384fec79beSAlan Cox pmap_remove_all(p); 21396bbee8e2SAlan Cox if ((options & OBJPR_CLEANONLY) == 0) { 21400012f373SJeff Roberson vm_page_invalid(p); 2141a28042d1SAlan Cox vm_page_undirty(p); 2142a28042d1SAlan Cox } 214363e97555SJeff Roberson vm_page_xunbusy(p); 214493c5d3a4SKonstantin Belousov continue; 21450d94caffSDavid Greenman } 214668855966SAlan Cox KASSERT((p->flags & PG_FICTITIOUS) == 0, 214768855966SAlan Cox ("vm_object_page_remove: page %p is fictitious", p)); 21480012f373SJeff Roberson if ((options & OBJPR_CLEANONLY) != 0 && 21490012f373SJeff Roberson !vm_page_none_valid(p)) { 2150cf060942SAlan Cox if ((options & OBJPR_NOTMAPPED) == 0 && 2151fee2a2faSMark Johnston object->ref_count != 0 && 2152fee2a2faSMark Johnston !vm_page_try_remove_write(p)) 2153fee2a2faSMark Johnston goto wired; 215463e97555SJeff Roberson if (p->dirty != 0) { 215563e97555SJeff Roberson vm_page_xunbusy(p); 215693c5d3a4SKonstantin Belousov continue; 21572965a453SKip Macy } 215863e97555SJeff Roberson } 2159fee2a2faSMark Johnston if ((options & OBJPR_NOTMAPPED) == 0 && 2160fee2a2faSMark Johnston object->ref_count != 0 && !vm_page_try_remove_all(p)) 2161fee2a2faSMark Johnston goto wired; 21625cd29d0fSMark Johnston vm_page_free(p); 21632965a453SKip Macy } 2164f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 216584242cf6SMark Johnston 21661390a5cbSKonstantin Belousov vm_pager_freespace(object, start, (end == 0 ? object->size : end) - 21671390a5cbSKonstantin Belousov start); 2168c0503609SDavid Greenman } 2169df8bae1dSRodney W. Grimes 2170df8bae1dSRodney W. Grimes /* 21713138cd36SMark Johnston * vm_object_page_noreuse: 2172936c09acSJohn Baldwin * 21733138cd36SMark Johnston * For the given object, attempt to move the specified pages to 21743138cd36SMark Johnston * the head of the inactive queue. This bypasses regular LRU 21753138cd36SMark Johnston * operation and allows the pages to be reused quickly under memory 21763138cd36SMark Johnston * pressure. If a page is wired for any reason, then it will not 21773138cd36SMark Johnston * be queued. Pages are specified by the range ["start", "end"). 21783138cd36SMark Johnston * As a special case, if "end" is zero, then the range extends from 21793138cd36SMark Johnston * "start" to the end of the object. 2180936c09acSJohn Baldwin * 2181936c09acSJohn Baldwin * This operation should only be performed on objects that 218228634820SAlan Cox * contain non-fictitious, managed pages. 2183936c09acSJohn Baldwin * 2184936c09acSJohn Baldwin * The object must be locked. 2185936c09acSJohn Baldwin */ 2186936c09acSJohn Baldwin void 21873138cd36SMark Johnston vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2188936c09acSJohn Baldwin { 2189936c09acSJohn Baldwin vm_page_t p, next; 2190936c09acSJohn Baldwin 219152d1addaSAlan Cox VM_OBJECT_ASSERT_LOCKED(object); 219228634820SAlan Cox KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, 21933138cd36SMark Johnston ("vm_object_page_noreuse: illegal object %p", object)); 2194936c09acSJohn Baldwin if (object->resident_page_count == 0) 2195936c09acSJohn Baldwin return; 2196936c09acSJohn Baldwin p = vm_page_find_least(object, start); 2197936c09acSJohn Baldwin 2198936c09acSJohn Baldwin /* 2199936c09acSJohn Baldwin * Here, the variable "p" is either (1) the page with the least pindex 2200936c09acSJohn Baldwin * greater than or equal to the parameter "start" or (2) NULL. 2201936c09acSJohn Baldwin */ 2202936c09acSJohn Baldwin for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2203936c09acSJohn Baldwin next = TAILQ_NEXT(p, listq); 22043138cd36SMark Johnston vm_page_deactivate_noreuse(p); 2205936c09acSJohn Baldwin } 2206936c09acSJohn Baldwin } 2207936c09acSJohn Baldwin 2208936c09acSJohn Baldwin /* 2209387aabc5SAlan Cox * Populate the specified range of the object with valid pages. Returns 2210387aabc5SAlan Cox * TRUE if the range is successfully populated and FALSE otherwise. 2211387aabc5SAlan Cox * 2212387aabc5SAlan Cox * Note: This function should be optimized to pass a larger array of 2213387aabc5SAlan Cox * pages to vm_pager_get_pages() before it is applied to a non- 2214387aabc5SAlan Cox * OBJT_DEVICE object. 2215387aabc5SAlan Cox * 2216387aabc5SAlan Cox * The object must be locked. 2217387aabc5SAlan Cox */ 2218387aabc5SAlan Cox boolean_t 2219387aabc5SAlan Cox vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2220387aabc5SAlan Cox { 2221093c7f39SGleb Smirnoff vm_page_t m; 2222387aabc5SAlan Cox vm_pindex_t pindex; 2223387aabc5SAlan Cox int rv; 2224387aabc5SAlan Cox 222589f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 2226387aabc5SAlan Cox for (pindex = start; pindex < end; pindex++) { 2227c7575748SJeff Roberson rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL); 2228c7575748SJeff Roberson if (rv != VM_PAGER_OK) 2229387aabc5SAlan Cox break; 2230c7575748SJeff Roberson 2231387aabc5SAlan Cox /* 2232387aabc5SAlan Cox * Keep "m" busy because a subsequent iteration may unlock 2233387aabc5SAlan Cox * the object. 2234387aabc5SAlan Cox */ 2235387aabc5SAlan Cox } 2236387aabc5SAlan Cox if (pindex > start) { 2237387aabc5SAlan Cox m = vm_page_lookup(object, start); 2238387aabc5SAlan Cox while (m != NULL && m->pindex < pindex) { 2239c7aebda8SAttilio Rao vm_page_xunbusy(m); 2240387aabc5SAlan Cox m = TAILQ_NEXT(m, listq); 2241387aabc5SAlan Cox } 2242387aabc5SAlan Cox } 2243387aabc5SAlan Cox return (pindex == end); 2244387aabc5SAlan Cox } 2245387aabc5SAlan Cox 2246387aabc5SAlan Cox /* 2247df8bae1dSRodney W. Grimes * Routine: vm_object_coalesce 2248df8bae1dSRodney W. Grimes * Function: Coalesces two objects backing up adjoining 2249df8bae1dSRodney W. Grimes * regions of memory into a single object. 2250df8bae1dSRodney W. Grimes * 2251df8bae1dSRodney W. Grimes * returns TRUE if objects were combined. 2252df8bae1dSRodney W. Grimes * 2253df8bae1dSRodney W. Grimes * NOTE: Only works at the moment if the second object is NULL - 2254df8bae1dSRodney W. Grimes * if it's not, which object do we lock first? 2255df8bae1dSRodney W. Grimes * 2256df8bae1dSRodney W. Grimes * Parameters: 2257df8bae1dSRodney W. Grimes * prev_object First object to coalesce 2258df8bae1dSRodney W. Grimes * prev_offset Offset into prev_object 2259df8bae1dSRodney W. Grimes * prev_size Size of reference to prev_object 226057a21abaSAlan Cox * next_size Size of reference to the second object 22613364c323SKonstantin Belousov * reserved Indicator that extension region has 22623364c323SKonstantin Belousov * swap accounted for 2263df8bae1dSRodney W. Grimes * 2264df8bae1dSRodney W. Grimes * Conditions: 2265df8bae1dSRodney W. Grimes * The object must *not* be locked. 2266df8bae1dSRodney W. Grimes */ 22670d94caffSDavid Greenman boolean_t 226857a21abaSAlan Cox vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, 22693364c323SKonstantin Belousov vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) 2270df8bae1dSRodney W. Grimes { 2271ea41812fSAlan Cox vm_pindex_t next_pindex; 2272df8bae1dSRodney W. Grimes 227300e1854aSAlan Cox if (prev_object == NULL) 2274df8bae1dSRodney W. Grimes return (TRUE); 227563967687SJeff Roberson if ((prev_object->flags & OBJ_ANON) == 0) 227630dcfc09SJohn Dyson return (FALSE); 227730dcfc09SJohn Dyson 227863967687SJeff Roberson VM_OBJECT_WLOCK(prev_object); 2279df8bae1dSRodney W. Grimes /* 228098087a06SJeff Roberson * Try to collapse the object first. 2281df8bae1dSRodney W. Grimes */ 2282df8bae1dSRodney W. Grimes vm_object_collapse(prev_object); 2283df8bae1dSRodney W. Grimes 2284df8bae1dSRodney W. Grimes /* 22850d94caffSDavid Greenman * Can't coalesce if: . more than one reference . paged out . shadows 22860d94caffSDavid Greenman * another object . has a copy elsewhere (any of which mean that the 22870d94caffSDavid Greenman * pages not mapped to prev_entry may be in use anyway) 2288df8bae1dSRodney W. Grimes */ 22898cc7e047SJohn Dyson if (prev_object->backing_object != NULL) { 229089f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 2291df8bae1dSRodney W. Grimes return (FALSE); 2292df8bae1dSRodney W. Grimes } 2293a316d390SJohn Dyson 2294a316d390SJohn Dyson prev_size >>= PAGE_SHIFT; 2295a316d390SJohn Dyson next_size >>= PAGE_SHIFT; 229657a21abaSAlan Cox next_pindex = OFF_TO_IDX(prev_offset) + prev_size; 22978cc7e047SJohn Dyson 22980e48e068SMark Johnston if (prev_object->ref_count > 1 && 22990e48e068SMark Johnston prev_object->size != next_pindex && 23000e48e068SMark Johnston (prev_object->flags & OBJ_ONEMAPPING) == 0) { 230189f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 23028cc7e047SJohn Dyson return (FALSE); 23038cc7e047SJohn Dyson } 23048cc7e047SJohn Dyson 2305df8bae1dSRodney W. Grimes /* 23063364c323SKonstantin Belousov * Account for the charge. 23073364c323SKonstantin Belousov */ 2308ef694c1aSEdward Tomasz Napierala if (prev_object->cred != NULL) { 23093364c323SKonstantin Belousov /* 23103364c323SKonstantin Belousov * If prev_object was charged, then this mapping, 2311763df3ecSPedro F. Giffuni * although not charged now, may become writable 2312ef694c1aSEdward Tomasz Napierala * later. Non-NULL cred in the object would prevent 23133364c323SKonstantin Belousov * swap reservation during enabling of the write 23143364c323SKonstantin Belousov * access, so reserve swap now. Failed reservation 23153364c323SKonstantin Belousov * cause allocation of the separate object for the map 23163364c323SKonstantin Belousov * entry, and swap reservation for this entry is 23173364c323SKonstantin Belousov * managed in appropriate time. 23183364c323SKonstantin Belousov */ 2319ef694c1aSEdward Tomasz Napierala if (!reserved && !swap_reserve_by_cred(ptoa(next_size), 2320ef694c1aSEdward Tomasz Napierala prev_object->cred)) { 23219f790a17SKonstantin Belousov VM_OBJECT_WUNLOCK(prev_object); 23223364c323SKonstantin Belousov return (FALSE); 23233364c323SKonstantin Belousov } 23243364c323SKonstantin Belousov prev_object->charge += ptoa(next_size); 23253364c323SKonstantin Belousov } 23263364c323SKonstantin Belousov 23273364c323SKonstantin Belousov /* 23280d94caffSDavid Greenman * Remove any pages that may still be in the object from a previous 23290d94caffSDavid Greenman * deallocation. 2330df8bae1dSRodney W. Grimes */ 2331ea41812fSAlan Cox if (next_pindex < prev_object->size) { 23326bbee8e2SAlan Cox vm_object_page_remove(prev_object, next_pindex, next_pindex + 23336bbee8e2SAlan Cox next_size, 0); 23343364c323SKonstantin Belousov #if 0 2335ef694c1aSEdward Tomasz Napierala if (prev_object->cred != NULL) { 23363364c323SKonstantin Belousov KASSERT(prev_object->charge >= 23373364c323SKonstantin Belousov ptoa(prev_object->size - next_pindex), 23383364c323SKonstantin Belousov ("object %p overcharged 1 %jx %jx", prev_object, 23393364c323SKonstantin Belousov (uintmax_t)next_pindex, (uintmax_t)next_size)); 23403364c323SKonstantin Belousov prev_object->charge -= ptoa(prev_object->size - 23413364c323SKonstantin Belousov next_pindex); 23423364c323SKonstantin Belousov } 23433364c323SKonstantin Belousov #endif 2344ea41812fSAlan Cox } 2345df8bae1dSRodney W. Grimes 2346df8bae1dSRodney W. Grimes /* 2347df8bae1dSRodney W. Grimes * Extend the object if necessary. 2348df8bae1dSRodney W. Grimes */ 2349ea41812fSAlan Cox if (next_pindex + next_size > prev_object->size) 2350ea41812fSAlan Cox prev_object->size = next_pindex + next_size; 2351df8bae1dSRodney W. Grimes 235289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 2353df8bae1dSRodney W. Grimes return (TRUE); 2354df8bae1dSRodney W. Grimes } 2355df8bae1dSRodney W. Grimes 23567a5a6352SMatthew Dillon void 2357180bcaa4SKonstantin Belousov vm_object_set_writeable_dirty_(vm_object_t object) 23587a5a6352SMatthew Dillon { 235967d0e293SJeff Roberson atomic_add_int(&object->generation, 1); 23607a5a6352SMatthew Dillon } 23617a5a6352SMatthew Dillon 2362c23c555bSKonstantin Belousov bool 2363c23c555bSKonstantin Belousov vm_object_mightbedirty_(vm_object_t object) 2364c23c555bSKonstantin Belousov { 2365c23c555bSKonstantin Belousov return (object->generation != object->cleangeneration); 2366c23c555bSKonstantin Belousov } 2367c23c555bSKonstantin Belousov 236803462509SAlan Cox /* 236903462509SAlan Cox * vm_object_unwire: 237003462509SAlan Cox * 237103462509SAlan Cox * For each page offset within the specified range of the given object, 237203462509SAlan Cox * find the highest-level page in the shadow chain and unwire it. A page 237303462509SAlan Cox * must exist at every page offset, and the highest-level page must be 237403462509SAlan Cox * wired. 237503462509SAlan Cox */ 237603462509SAlan Cox void 237703462509SAlan Cox vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, 237803462509SAlan Cox uint8_t queue) 237903462509SAlan Cox { 238020e4afbfSKonstantin Belousov vm_object_t tobject, t1object; 238103462509SAlan Cox vm_page_t m, tm; 238203462509SAlan Cox vm_pindex_t end_pindex, pindex, tpindex; 238303462509SAlan Cox int depth, locked_depth; 238403462509SAlan Cox 238503462509SAlan Cox KASSERT((offset & PAGE_MASK) == 0, 238603462509SAlan Cox ("vm_object_unwire: offset is not page aligned")); 238703462509SAlan Cox KASSERT((length & PAGE_MASK) == 0, 238803462509SAlan Cox ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); 238903462509SAlan Cox /* The wired count of a fictitious page never changes. */ 239003462509SAlan Cox if ((object->flags & OBJ_FICTITIOUS) != 0) 239103462509SAlan Cox return; 239203462509SAlan Cox pindex = OFF_TO_IDX(offset); 239303462509SAlan Cox end_pindex = pindex + atop(length); 239420e4afbfSKonstantin Belousov again: 239503462509SAlan Cox locked_depth = 1; 239603462509SAlan Cox VM_OBJECT_RLOCK(object); 239703462509SAlan Cox m = vm_page_find_least(object, pindex); 239803462509SAlan Cox while (pindex < end_pindex) { 239903462509SAlan Cox if (m == NULL || pindex < m->pindex) { 240003462509SAlan Cox /* 240103462509SAlan Cox * The first object in the shadow chain doesn't 240203462509SAlan Cox * contain a page at the current index. Therefore, 240303462509SAlan Cox * the page must exist in a backing object. 240403462509SAlan Cox */ 240503462509SAlan Cox tobject = object; 240603462509SAlan Cox tpindex = pindex; 240703462509SAlan Cox depth = 0; 240803462509SAlan Cox do { 240903462509SAlan Cox tpindex += 241003462509SAlan Cox OFF_TO_IDX(tobject->backing_object_offset); 241103462509SAlan Cox tobject = tobject->backing_object; 241203462509SAlan Cox KASSERT(tobject != NULL, 241303462509SAlan Cox ("vm_object_unwire: missing page")); 241403462509SAlan Cox if ((tobject->flags & OBJ_FICTITIOUS) != 0) 241503462509SAlan Cox goto next_page; 241603462509SAlan Cox depth++; 241703462509SAlan Cox if (depth == locked_depth) { 241803462509SAlan Cox locked_depth++; 241903462509SAlan Cox VM_OBJECT_RLOCK(tobject); 242003462509SAlan Cox } 242103462509SAlan Cox } while ((tm = vm_page_lookup(tobject, tpindex)) == 242203462509SAlan Cox NULL); 242303462509SAlan Cox } else { 242403462509SAlan Cox tm = m; 242503462509SAlan Cox m = TAILQ_NEXT(m, listq); 242603462509SAlan Cox } 242763e97555SJeff Roberson if (vm_page_trysbusy(tm) == 0) { 242887e93ea6SMark Johnston for (tobject = object; locked_depth >= 1; 242920e4afbfSKonstantin Belousov locked_depth--) { 243020e4afbfSKonstantin Belousov t1object = tobject->backing_object; 243187e93ea6SMark Johnston if (tm->object != tobject) 243220e4afbfSKonstantin Belousov VM_OBJECT_RUNLOCK(tobject); 243320e4afbfSKonstantin Belousov tobject = t1object; 243420e4afbfSKonstantin Belousov } 243587b64663SMark Johnston tobject = tm->object; 243687b64663SMark Johnston if (!vm_page_busy_sleep(tm, "unwbo", 243787b64663SMark Johnston VM_ALLOC_IGN_SBUSY)) 243887b64663SMark Johnston VM_OBJECT_RUNLOCK(tobject); 243920e4afbfSKonstantin Belousov goto again; 244020e4afbfSKonstantin Belousov } 244103462509SAlan Cox vm_page_unwire(tm, queue); 244263e97555SJeff Roberson vm_page_sunbusy(tm); 244303462509SAlan Cox next_page: 244403462509SAlan Cox pindex++; 244503462509SAlan Cox } 244603462509SAlan Cox /* Release the accumulated object locks. */ 244720e4afbfSKonstantin Belousov for (tobject = object; locked_depth >= 1; locked_depth--) { 244820e4afbfSKonstantin Belousov t1object = tobject->backing_object; 244920e4afbfSKonstantin Belousov VM_OBJECT_RUNLOCK(tobject); 245020e4afbfSKonstantin Belousov tobject = t1object; 245103462509SAlan Cox } 245203462509SAlan Cox } 245303462509SAlan Cox 24540951bd36SEric van Gyzen /* 24550951bd36SEric van Gyzen * Return the vnode for the given object, or NULL if none exists. 24560951bd36SEric van Gyzen * For tmpfs objects, the function may return NULL if there is 24570951bd36SEric van Gyzen * no vnode allocated at the time of the call. 24580951bd36SEric van Gyzen */ 245963e4c6cdSEric van Gyzen struct vnode * 246063e4c6cdSEric van Gyzen vm_object_vnode(vm_object_t object) 246163e4c6cdSEric van Gyzen { 24620951bd36SEric van Gyzen struct vnode *vp; 246363e4c6cdSEric van Gyzen 246463e4c6cdSEric van Gyzen VM_OBJECT_ASSERT_LOCKED(object); 2465a7c198a2SKonstantin Belousov vm_pager_getvp(object, &vp, NULL); 24660951bd36SEric van Gyzen return (vp); 246763e4c6cdSEric van Gyzen } 246863e4c6cdSEric van Gyzen 2469205be21dSJeff Roberson /* 2470205be21dSJeff Roberson * Busy the vm object. This prevents new pages belonging to the object from 2471205be21dSJeff Roberson * becoming busy. Existing pages persist as busy. Callers are responsible 2472205be21dSJeff Roberson * for checking page state before proceeding. 2473205be21dSJeff Roberson */ 2474205be21dSJeff Roberson void 2475205be21dSJeff Roberson vm_object_busy(vm_object_t obj) 2476205be21dSJeff Roberson { 2477205be21dSJeff Roberson 2478205be21dSJeff Roberson VM_OBJECT_ASSERT_LOCKED(obj); 2479205be21dSJeff Roberson 2480c99d0c58SMark Johnston blockcount_acquire(&obj->busy, 1); 2481205be21dSJeff Roberson /* The fence is required to order loads of page busy. */ 2482205be21dSJeff Roberson atomic_thread_fence_acq_rel(); 2483205be21dSJeff Roberson } 2484205be21dSJeff Roberson 2485205be21dSJeff Roberson void 2486205be21dSJeff Roberson vm_object_unbusy(vm_object_t obj) 2487205be21dSJeff Roberson { 2488205be21dSJeff Roberson 2489c99d0c58SMark Johnston blockcount_release(&obj->busy, 1); 2490205be21dSJeff Roberson } 2491205be21dSJeff Roberson 2492205be21dSJeff Roberson void 2493205be21dSJeff Roberson vm_object_busy_wait(vm_object_t obj, const char *wmesg) 2494205be21dSJeff Roberson { 2495205be21dSJeff Roberson 2496205be21dSJeff Roberson VM_OBJECT_ASSERT_UNLOCKED(obj); 2497205be21dSJeff Roberson 2498c99d0c58SMark Johnston (void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM); 2499205be21dSJeff Roberson } 2500205be21dSJeff Roberson 2501cd37afd8SRick Macklem /* 2502cd37afd8SRick Macklem * This function aims to determine if the object is mapped, 2503cd37afd8SRick Macklem * specifically, if it is referenced by a vm_map_entry. Because 2504cd37afd8SRick Macklem * objects occasionally acquire transient references that do not 2505cd37afd8SRick Macklem * represent a mapping, the method used here is inexact. However, it 2506cd37afd8SRick Macklem * has very low overhead and is good enough for the advisory 2507cd37afd8SRick Macklem * vm.vmtotal sysctl. 2508cd37afd8SRick Macklem */ 2509cd37afd8SRick Macklem bool 2510cd37afd8SRick Macklem vm_object_is_active(vm_object_t obj) 2511cd37afd8SRick Macklem { 2512cd37afd8SRick Macklem 2513cd37afd8SRick Macklem return (obj->ref_count > obj->shadow_count); 2514cd37afd8SRick Macklem } 2515cd37afd8SRick Macklem 2516ff87ae35SJohn Baldwin static int 251742812cccSKonstantin Belousov vm_object_list_handler(struct sysctl_req *req, bool swap_only) 2518ff87ae35SJohn Baldwin { 25190ecee546SKonstantin Belousov struct kinfo_vmobject *kvo; 2520ff87ae35SJohn Baldwin char *fullpath, *freepath; 2521ff87ae35SJohn Baldwin struct vnode *vp; 2522ff87ae35SJohn Baldwin struct vattr va; 2523ff87ae35SJohn Baldwin vm_object_t obj; 2524ff87ae35SJohn Baldwin vm_page_t m; 2525ecfbddf0SKonstantin Belousov u_long sp; 2526ff87ae35SJohn Baldwin int count, error; 2527ff87ae35SJohn Baldwin 2528ff87ae35SJohn Baldwin if (req->oldptr == NULL) { 2529ff87ae35SJohn Baldwin /* 2530ff87ae35SJohn Baldwin * If an old buffer has not been provided, generate an 2531ff87ae35SJohn Baldwin * estimate of the space needed for a subsequent call. 2532ff87ae35SJohn Baldwin */ 2533ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2534ff87ae35SJohn Baldwin count = 0; 2535ff87ae35SJohn Baldwin TAILQ_FOREACH(obj, &vm_object_list, object_list) { 2536ff87ae35SJohn Baldwin if (obj->type == OBJT_DEAD) 2537ff87ae35SJohn Baldwin continue; 2538ff87ae35SJohn Baldwin count++; 2539ff87ae35SJohn Baldwin } 2540ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 2541ff87ae35SJohn Baldwin return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * 2542ff87ae35SJohn Baldwin count * 11 / 10)); 2543ff87ae35SJohn Baldwin } 2544ff87ae35SJohn Baldwin 25450ecee546SKonstantin Belousov kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK); 2546ff87ae35SJohn Baldwin error = 0; 2547ff87ae35SJohn Baldwin 2548ff87ae35SJohn Baldwin /* 2549ff87ae35SJohn Baldwin * VM objects are type stable and are never removed from the 2550ff87ae35SJohn Baldwin * list once added. This allows us to safely read obj->object_list 2551ff87ae35SJohn Baldwin * after reacquiring the VM object lock. 2552ff87ae35SJohn Baldwin */ 2553ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2554ff87ae35SJohn Baldwin TAILQ_FOREACH(obj, &vm_object_list, object_list) { 255542812cccSKonstantin Belousov if (obj->type == OBJT_DEAD || 255642812cccSKonstantin Belousov (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) 2557ff87ae35SJohn Baldwin continue; 2558ff87ae35SJohn Baldwin VM_OBJECT_RLOCK(obj); 255942812cccSKonstantin Belousov if (obj->type == OBJT_DEAD || 256042812cccSKonstantin Belousov (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) { 2561ff87ae35SJohn Baldwin VM_OBJECT_RUNLOCK(obj); 2562ff87ae35SJohn Baldwin continue; 2563ff87ae35SJohn Baldwin } 2564ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 25650ecee546SKonstantin Belousov kvo->kvo_size = ptoa(obj->size); 25660ecee546SKonstantin Belousov kvo->kvo_resident = obj->resident_page_count; 25670ecee546SKonstantin Belousov kvo->kvo_ref_count = obj->ref_count; 25680ecee546SKonstantin Belousov kvo->kvo_shadow_count = obj->shadow_count; 25690ecee546SKonstantin Belousov kvo->kvo_memattr = obj->memattr; 25700ecee546SKonstantin Belousov kvo->kvo_active = 0; 25710ecee546SKonstantin Belousov kvo->kvo_inactive = 0; 25727738118eSKonstantin Belousov if (!swap_only) { 2573ff87ae35SJohn Baldwin TAILQ_FOREACH(m, &obj->memq, listq) { 2574ff87ae35SJohn Baldwin /* 2575ff87ae35SJohn Baldwin * A page may belong to the object but be 2576ff87ae35SJohn Baldwin * dequeued and set to PQ_NONE while the 2577ff87ae35SJohn Baldwin * object lock is not held. This makes the 2578ff87ae35SJohn Baldwin * reads of m->queue below racy, and we do not 2579ff87ae35SJohn Baldwin * count pages set to PQ_NONE. However, this 2580ff87ae35SJohn Baldwin * sysctl is only meant to give an 2581ff87ae35SJohn Baldwin * approximation of the system anyway. 2582ff87ae35SJohn Baldwin */ 25835cff1f4dSMark Johnston if (m->a.queue == PQ_ACTIVE) 25840ecee546SKonstantin Belousov kvo->kvo_active++; 25855cff1f4dSMark Johnston else if (m->a.queue == PQ_INACTIVE) 25860ecee546SKonstantin Belousov kvo->kvo_inactive++; 2587ff87ae35SJohn Baldwin } 25887738118eSKonstantin Belousov } 2589ff87ae35SJohn Baldwin 25900ecee546SKonstantin Belousov kvo->kvo_vn_fileid = 0; 25910ecee546SKonstantin Belousov kvo->kvo_vn_fsid = 0; 25920ecee546SKonstantin Belousov kvo->kvo_vn_fsid_freebsd11 = 0; 2593ff87ae35SJohn Baldwin freepath = NULL; 2594ff87ae35SJohn Baldwin fullpath = ""; 25957738118eSKonstantin Belousov vp = NULL; 25967738118eSKonstantin Belousov kvo->kvo_type = vm_object_kvme_type(obj, swap_only ? NULL : &vp); 2597ecfbddf0SKonstantin Belousov if (vp != NULL) { 2598ff87ae35SJohn Baldwin vref(vp); 2599ecfbddf0SKonstantin Belousov } else if ((obj->flags & OBJ_ANON) != 0) { 2600ecfbddf0SKonstantin Belousov MPASS(kvo->kvo_type == KVME_TYPE_DEFAULT || 2601ecfbddf0SKonstantin Belousov kvo->kvo_type == KVME_TYPE_SWAP); 2602ecfbddf0SKonstantin Belousov kvo->kvo_me = (uintptr_t)obj; 2603ecfbddf0SKonstantin Belousov /* tmpfs objs are reported as vnodes */ 2604ecfbddf0SKonstantin Belousov kvo->kvo_backing_obj = (uintptr_t)obj->backing_object; 2605ecfbddf0SKonstantin Belousov sp = swap_pager_swapped_pages(obj); 2606ecfbddf0SKonstantin Belousov kvo->kvo_swapped = sp > UINT32_MAX ? UINT32_MAX : sp; 2607ecfbddf0SKonstantin Belousov } 2608ff87ae35SJohn Baldwin VM_OBJECT_RUNLOCK(obj); 2609ff87ae35SJohn Baldwin if (vp != NULL) { 2610feabaaf9SMateusz Guzik vn_fullpath(vp, &fullpath, &freepath); 2611ff87ae35SJohn Baldwin vn_lock(vp, LK_SHARED | LK_RETRY); 2612ff87ae35SJohn Baldwin if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { 26130ecee546SKonstantin Belousov kvo->kvo_vn_fileid = va.va_fileid; 26140ecee546SKonstantin Belousov kvo->kvo_vn_fsid = va.va_fsid; 26150ecee546SKonstantin Belousov kvo->kvo_vn_fsid_freebsd11 = va.va_fsid; 261669921123SKonstantin Belousov /* truncate */ 2617ff87ae35SJohn Baldwin } 2618ff87ae35SJohn Baldwin vput(vp); 2619ff87ae35SJohn Baldwin } 2620ff87ae35SJohn Baldwin 26210ecee546SKonstantin Belousov strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path)); 2622ff87ae35SJohn Baldwin if (freepath != NULL) 2623ff87ae35SJohn Baldwin free(freepath, M_TEMP); 2624ff87ae35SJohn Baldwin 2625ff87ae35SJohn Baldwin /* Pack record size down */ 26260ecee546SKonstantin Belousov kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) 26270ecee546SKonstantin Belousov + strlen(kvo->kvo_path) + 1; 26280ecee546SKonstantin Belousov kvo->kvo_structsize = roundup(kvo->kvo_structsize, 2629ff87ae35SJohn Baldwin sizeof(uint64_t)); 26300ecee546SKonstantin Belousov error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize); 2631350fc36bSKonstantin Belousov maybe_yield(); 2632ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2633ff87ae35SJohn Baldwin if (error) 2634ff87ae35SJohn Baldwin break; 2635ff87ae35SJohn Baldwin } 2636ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 26370ecee546SKonstantin Belousov free(kvo, M_TEMP); 2638ff87ae35SJohn Baldwin return (error); 2639ff87ae35SJohn Baldwin } 26401b610624SKonstantin Belousov 26411b610624SKonstantin Belousov static int 26421b610624SKonstantin Belousov sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) 26431b610624SKonstantin Belousov { 264442812cccSKonstantin Belousov return (vm_object_list_handler(req, false)); 26451b610624SKonstantin Belousov } 26461b610624SKonstantin Belousov 2647ff87ae35SJohn Baldwin SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | 2648ff87ae35SJohn Baldwin CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", 2649ff87ae35SJohn Baldwin "List of VM objects"); 2650ff87ae35SJohn Baldwin 265142812cccSKonstantin Belousov static int 265242812cccSKonstantin Belousov sysctl_vm_object_list_swap(SYSCTL_HANDLER_ARGS) 265342812cccSKonstantin Belousov { 265442812cccSKonstantin Belousov return (vm_object_list_handler(req, true)); 265542812cccSKonstantin Belousov } 265642812cccSKonstantin Belousov 265742812cccSKonstantin Belousov /* 265842812cccSKonstantin Belousov * This sysctl returns list of the anonymous or swap objects. Intent 265942812cccSKonstantin Belousov * is to provide stripped optimized list useful to analyze swap use. 266042812cccSKonstantin Belousov * Since technically non-swap (default) objects participate in the 266142812cccSKonstantin Belousov * shadow chains, and are converted to swap type as needed by swap 266242812cccSKonstantin Belousov * pager, we must report them. 266342812cccSKonstantin Belousov */ 266442812cccSKonstantin Belousov SYSCTL_PROC(_vm, OID_AUTO, swap_objects, 266542812cccSKonstantin Belousov CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, 266642812cccSKonstantin Belousov sysctl_vm_object_list_swap, "S,kinfo_vmobject", 266742812cccSKonstantin Belousov "List of swap VM objects"); 266842812cccSKonstantin Belousov 2669c7c34a24SBruce Evans #include "opt_ddb.h" 2670c3cb3e12SDavid Greenman #ifdef DDB 2671c7c34a24SBruce Evans #include <sys/kernel.h> 2672c7c34a24SBruce Evans 2673ce9edcf5SPoul-Henning Kamp #include <sys/cons.h> 2674c7c34a24SBruce Evans 2675c7c34a24SBruce Evans #include <ddb/ddb.h> 2676c7c34a24SBruce Evans 2677cac597e4SBruce Evans static int 26781b40f8c0SMatthew Dillon _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) 2679a1f6d91cSDavid Greenman { 2680a1f6d91cSDavid Greenman vm_map_t tmpm; 2681a1f6d91cSDavid Greenman vm_map_entry_t tmpe; 2682a1f6d91cSDavid Greenman vm_object_t obj; 2683a1f6d91cSDavid Greenman 2684a1f6d91cSDavid Greenman if (map == 0) 2685a1f6d91cSDavid Greenman return 0; 2686a1f6d91cSDavid Greenman 2687a1f6d91cSDavid Greenman if (entry == 0) { 26882288078cSDoug Moore VM_MAP_ENTRY_FOREACH(tmpe, map) { 2689a1f6d91cSDavid Greenman if (_vm_object_in_map(map, object, tmpe)) { 2690a1f6d91cSDavid Greenman return 1; 2691a1f6d91cSDavid Greenman } 2692a1f6d91cSDavid Greenman } 26939fdfe602SMatthew Dillon } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 26949fdfe602SMatthew Dillon tmpm = entry->object.sub_map; 26952288078cSDoug Moore VM_MAP_ENTRY_FOREACH(tmpe, tmpm) { 2696a1f6d91cSDavid Greenman if (_vm_object_in_map(tmpm, object, tmpe)) { 2697a1f6d91cSDavid Greenman return 1; 2698a1f6d91cSDavid Greenman } 2699a1f6d91cSDavid Greenman } 27008aef1712SMatthew Dillon } else if ((obj = entry->object.vm_object) != NULL) { 270124a1cce3SDavid Greenman for (; obj; obj = obj->backing_object) 2702a1f6d91cSDavid Greenman if (obj == object) { 2703a1f6d91cSDavid Greenman return 1; 2704a1f6d91cSDavid Greenman } 2705a1f6d91cSDavid Greenman } 2706a1f6d91cSDavid Greenman return 0; 2707a1f6d91cSDavid Greenman } 2708a1f6d91cSDavid Greenman 2709cac597e4SBruce Evans static int 27101b40f8c0SMatthew Dillon vm_object_in_map(vm_object_t object) 2711a1f6d91cSDavid Greenman { 2712a1f6d91cSDavid Greenman struct proc *p; 27131005a129SJohn Baldwin 271460517fd1SJohn Baldwin /* sx_slock(&allproc_lock); */ 2715f67af5c9SXin LI FOREACH_PROC_IN_SYSTEM(p) { 2716a1f6d91cSDavid Greenman if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 2717a1f6d91cSDavid Greenman continue; 2718553629ebSJake Burkholder if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 271960517fd1SJohn Baldwin /* sx_sunlock(&allproc_lock); */ 2720a1f6d91cSDavid Greenman return 1; 2721a1f6d91cSDavid Greenman } 2722553629ebSJake Burkholder } 272360517fd1SJohn Baldwin /* sx_sunlock(&allproc_lock); */ 2724a1f6d91cSDavid Greenman if (_vm_object_in_map(kernel_map, object, 0)) 2725a1f6d91cSDavid Greenman return 1; 2726a1f6d91cSDavid Greenman return 0; 2727a1f6d91cSDavid Greenman } 2728a1f6d91cSDavid Greenman 2729c7c34a24SBruce Evans DB_SHOW_COMMAND(vmochk, vm_object_check) 2730f708ef1bSPoul-Henning Kamp { 2731a1f6d91cSDavid Greenman vm_object_t object; 2732a1f6d91cSDavid Greenman 2733a1f6d91cSDavid Greenman /* 2734a1f6d91cSDavid Greenman * make sure that internal objs are in a map somewhere 2735a1f6d91cSDavid Greenman * and none have zero ref counts. 2736a1f6d91cSDavid Greenman */ 2737cc64b484SAlfred Perlstein TAILQ_FOREACH(object, &vm_object_list, object_list) { 273832362449SKonstantin Belousov if ((object->flags & OBJ_ANON) != 0) { 2739a1f6d91cSDavid Greenman if (object->ref_count == 0) { 27403efc015bSPeter Wemm db_printf("vmochk: internal obj has zero ref count: %ld\n", 27413efc015bSPeter Wemm (long)object->size); 2742a1f6d91cSDavid Greenman } 2743a1f6d91cSDavid Greenman if (!vm_object_in_map(object)) { 2744fc62ef1fSBruce Evans db_printf( 2745fc62ef1fSBruce Evans "vmochk: internal obj is not in a map: " 2746fc62ef1fSBruce Evans "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 2747fc62ef1fSBruce Evans object->ref_count, (u_long)object->size, 2748fc62ef1fSBruce Evans (u_long)object->size, 2749fc62ef1fSBruce Evans (void *)object->backing_object); 2750a1f6d91cSDavid Greenman } 2751a1f6d91cSDavid Greenman } 27528cc8c586SEric van Gyzen if (db_pager_quit) 27538cc8c586SEric van Gyzen return; 2754a1f6d91cSDavid Greenman } 2755a1f6d91cSDavid Greenman } 2756a1f6d91cSDavid Greenman 275726f9a767SRodney W. Grimes /* 2758df8bae1dSRodney W. Grimes * vm_object_print: [ debug ] 2759df8bae1dSRodney W. Grimes */ 2760c7c34a24SBruce Evans DB_SHOW_COMMAND(object, vm_object_print_static) 2761df8bae1dSRodney W. Grimes { 2762c7c34a24SBruce Evans /* XXX convert args. */ 2763c7c34a24SBruce Evans vm_object_t object = (vm_object_t)addr; 2764c7c34a24SBruce Evans boolean_t full = have_addr; 2765c7c34a24SBruce Evans 2766d031cff1SMatthew Dillon vm_page_t p; 2767df8bae1dSRodney W. Grimes 2768c7c34a24SBruce Evans /* XXX count is an (unused) arg. Avoid shadowing it. */ 2769c7c34a24SBruce Evans #define count was_count 2770c7c34a24SBruce Evans 2771d031cff1SMatthew Dillon int count; 2772df8bae1dSRodney W. Grimes 2773df8bae1dSRodney W. Grimes if (object == NULL) 2774df8bae1dSRodney W. Grimes return; 2775df8bae1dSRodney W. Grimes 2776eb95adefSBruce Evans db_iprintf( 2777ef694c1aSEdward Tomasz Napierala "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", 2778e47cd172SMaxime Henrion object, (int)object->type, (uintmax_t)object->size, 27793364c323SKonstantin Belousov object->resident_page_count, object->ref_count, object->flags, 2780ef694c1aSEdward Tomasz Napierala object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); 2781e47cd172SMaxime Henrion db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", 27821c7c3c6aSMatthew Dillon object->shadow_count, 2783eb95adefSBruce Evans object->backing_object ? object->backing_object->ref_count : 0, 2784e47cd172SMaxime Henrion object->backing_object, (uintmax_t)object->backing_object_offset); 2785df8bae1dSRodney W. Grimes 2786df8bae1dSRodney W. Grimes if (!full) 2787df8bae1dSRodney W. Grimes return; 2788df8bae1dSRodney W. Grimes 2789c7c34a24SBruce Evans db_indent += 2; 2790df8bae1dSRodney W. Grimes count = 0; 2791fc2ffbe6SPoul-Henning Kamp TAILQ_FOREACH(p, &object->memq, listq) { 2792df8bae1dSRodney W. Grimes if (count == 0) 2793c7c34a24SBruce Evans db_iprintf("memory:="); 2794df8bae1dSRodney W. Grimes else if (count == 6) { 2795c7c34a24SBruce Evans db_printf("\n"); 2796c7c34a24SBruce Evans db_iprintf(" ..."); 2797df8bae1dSRodney W. Grimes count = 0; 2798df8bae1dSRodney W. Grimes } else 2799c7c34a24SBruce Evans db_printf(","); 2800df8bae1dSRodney W. Grimes count++; 2801df8bae1dSRodney W. Grimes 2802e47cd172SMaxime Henrion db_printf("(off=0x%jx,page=0x%jx)", 2803e47cd172SMaxime Henrion (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); 28048cc8c586SEric van Gyzen 28058cc8c586SEric van Gyzen if (db_pager_quit) 28068cc8c586SEric van Gyzen break; 2807df8bae1dSRodney W. Grimes } 2808df8bae1dSRodney W. Grimes if (count != 0) 2809c7c34a24SBruce Evans db_printf("\n"); 2810c7c34a24SBruce Evans db_indent -= 2; 2811df8bae1dSRodney W. Grimes } 28125070c7f8SJohn Dyson 2813c7c34a24SBruce Evans /* XXX. */ 2814c7c34a24SBruce Evans #undef count 2815c7c34a24SBruce Evans 2816c7c34a24SBruce Evans /* XXX need this non-static entry for calling from vm_map_print. */ 28175070c7f8SJohn Dyson void 28181b40f8c0SMatthew Dillon vm_object_print( 28191b40f8c0SMatthew Dillon /* db_expr_t */ long addr, 28201b40f8c0SMatthew Dillon boolean_t have_addr, 28211b40f8c0SMatthew Dillon /* db_expr_t */ long count, 28221b40f8c0SMatthew Dillon char *modif) 2823c7c34a24SBruce Evans { 2824c7c34a24SBruce Evans vm_object_print_static(addr, have_addr, count, modif); 2825c7c34a24SBruce Evans } 2826c7c34a24SBruce Evans 2827c7c34a24SBruce Evans DB_SHOW_COMMAND(vmopag, vm_object_print_pages) 28285070c7f8SJohn Dyson { 28295070c7f8SJohn Dyson vm_object_t object; 2830bb2ac86fSKonstantin Belousov vm_pindex_t fidx; 2831bb2ac86fSKonstantin Belousov vm_paddr_t pa; 2832bb2ac86fSKonstantin Belousov vm_page_t m, prev_m; 2833*18048b6eSStephen J. Kiernan int rcount; 2834cc64b484SAlfred Perlstein 2835cc64b484SAlfred Perlstein TAILQ_FOREACH(object, &vm_object_list, object_list) { 2836fc62ef1fSBruce Evans db_printf("new object: %p\n", (void *)object); 2837*18048b6eSStephen J. Kiernan if (db_pager_quit) 28385070c7f8SJohn Dyson return; 2839*18048b6eSStephen J. Kiernan 28405070c7f8SJohn Dyson rcount = 0; 28415070c7f8SJohn Dyson fidx = 0; 2842bb2ac86fSKonstantin Belousov pa = -1; 2843bb2ac86fSKonstantin Belousov TAILQ_FOREACH(m, &object->memq, listq) { 2844bb2ac86fSKonstantin Belousov if (m->pindex > 128) 2845bb2ac86fSKonstantin Belousov break; 2846bb2ac86fSKonstantin Belousov if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && 2847bb2ac86fSKonstantin Belousov prev_m->pindex + 1 != m->pindex) { 28485070c7f8SJohn Dyson if (rcount) { 28493efc015bSPeter Wemm db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28503efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 2851*18048b6eSStephen J. Kiernan if (db_pager_quit) 28525070c7f8SJohn Dyson return; 28535070c7f8SJohn Dyson rcount = 0; 28545070c7f8SJohn Dyson } 28555070c7f8SJohn Dyson } 28565070c7f8SJohn Dyson if (rcount && 28575070c7f8SJohn Dyson (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 28585070c7f8SJohn Dyson ++rcount; 28595070c7f8SJohn Dyson continue; 28605070c7f8SJohn Dyson } 28615070c7f8SJohn Dyson if (rcount) { 28622446e4f0SAlan Cox db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28633efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 2864*18048b6eSStephen J. Kiernan if (db_pager_quit) 28655070c7f8SJohn Dyson return; 28665070c7f8SJohn Dyson } 2867bb2ac86fSKonstantin Belousov fidx = m->pindex; 28685070c7f8SJohn Dyson pa = VM_PAGE_TO_PHYS(m); 28695070c7f8SJohn Dyson rcount = 1; 28705070c7f8SJohn Dyson } 28715070c7f8SJohn Dyson if (rcount) { 28723efc015bSPeter Wemm db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28733efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 2874*18048b6eSStephen J. Kiernan if (db_pager_quit) 28755070c7f8SJohn Dyson return; 28765070c7f8SJohn Dyson } 28775070c7f8SJohn Dyson } 28785070c7f8SJohn Dyson } 2879c3cb3e12SDavid Greenman #endif /* DDB */ 2880