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)); 2014cf9f5d8SKonstantin Belousov KASSERT(atomic_load_int(&object->shadow_count) == 0, 2028355f576SJeff Roberson ("object %p shadow_count = %d", 2034cf9f5d8SKonstantin Belousov object, atomic_load_int(&object->shadow_count))); 204e735691bSJohn Baldwin KASSERT(object->type == OBJT_DEAD, 205e735691bSJohn Baldwin ("object %p has non-dead type %d", 206e735691bSJohn Baldwin object, object->type)); 207e1979b45SMark Johnston KASSERT(object->charge == 0 && object->cred == NULL, 208e1979b45SMark Johnston ("object %p has non-zero charge %ju (%p)", 209e1979b45SMark Johnston object, (uintmax_t)object->charge, object->cred)); 2108355f576SJeff Roberson } 2118355f576SJeff Roberson #endif 2128355f576SJeff Roberson 213b23f72e9SBrian Feldman static int 214b23f72e9SBrian Feldman vm_object_zinit(void *mem, int size, int flags) 2158355f576SJeff Roberson { 2168355f576SJeff Roberson vm_object_t object; 2178355f576SJeff Roberson 2188355f576SJeff Roberson object = (vm_object_t)mem; 219777a36c5SAlan Cox rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW); 2208355f576SJeff Roberson 2218355f576SJeff Roberson /* These are true for any object that has been freed */ 222e735691bSJohn Baldwin object->type = OBJT_DEAD; 223cd1241fbSKonstantin Belousov vm_radix_init(&object->rtree); 22451df5321SJeff Roberson refcount_init(&object->ref_count, 0); 225c99d0c58SMark Johnston blockcount_init(&object->paging_in_progress); 226c99d0c58SMark Johnston blockcount_init(&object->busy); 2278355f576SJeff Roberson object->resident_page_count = 0; 2284cf9f5d8SKonstantin Belousov atomic_store_int(&object->shadow_count, 0); 229f425ab8eSKonstantin Belousov object->flags = OBJ_DEAD; 230e735691bSJohn Baldwin 231e735691bSJohn Baldwin mtx_lock(&vm_object_list_mtx); 232e735691bSJohn Baldwin TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); 233e735691bSJohn Baldwin mtx_unlock(&vm_object_list_mtx); 234b23f72e9SBrian Feldman return (0); 2358355f576SJeff Roberson } 236df8bae1dSRodney W. Grimes 237a4915c21SAttilio Rao static void 23863967687SJeff Roberson _vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags, 23967388836SKonstantin Belousov vm_object_t object, void *handle) 240df8bae1dSRodney W. Grimes { 2410cddd8f0SMatthew Dillon 242df8bae1dSRodney W. Grimes TAILQ_INIT(&object->memq); 2431c500307SAlan Cox LIST_INIT(&object->shadow_head); 244a1f6d91cSDavid Greenman 24524a1cce3SDavid Greenman object->type = type; 2464b8365d7SKonstantin Belousov object->flags = flags; 2475d32157dSMark Johnston if ((flags & OBJ_SWAP) != 0) { 248f425ab8eSKonstantin Belousov pctrie_init(&object->un_pager.swp.swp_blks); 2495d32157dSMark Johnston object->un_pager.swp.writemappings = 0; 2505d32157dSMark Johnston } 251f425ab8eSKonstantin Belousov 252f425ab8eSKonstantin Belousov /* 253f425ab8eSKonstantin Belousov * Ensure that swap_pager_swapoff() iteration over object_list 254f425ab8eSKonstantin Belousov * sees up to date type and pctrie head if it observed 255f425ab8eSKonstantin Belousov * non-dead object. 256f425ab8eSKonstantin Belousov */ 257f425ab8eSKonstantin Belousov atomic_thread_fence_rel(); 258f425ab8eSKonstantin Belousov 25963967687SJeff Roberson object->pg_color = 0; 260df8bae1dSRodney W. Grimes object->size = size; 2614c29d2deSMark Johnston object->domain.dr_policy = NULL; 262b881da26SAlan Cox object->generation = 1; 26367d0e293SJeff Roberson object->cleangeneration = 1; 26451df5321SJeff Roberson refcount_init(&object->ref_count, 1); 2653153e878SAlan Cox object->memattr = VM_MEMATTR_DEFAULT; 266ef694c1aSEdward Tomasz Napierala object->cred = NULL; 2673364c323SKonstantin Belousov object->charge = 0; 26867388836SKonstantin Belousov object->handle = handle; 26924a1cce3SDavid Greenman object->backing_object = NULL; 270a316d390SJohn Dyson object->backing_object_offset = (vm_ooffset_t) 0; 271f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 272f8a47341SAlan Cox LIST_INIT(&object->rvq); 273f8a47341SAlan Cox #endif 2741bdbd705SKonstantin Belousov umtx_shm_object_init(object); 275df8bae1dSRodney W. Grimes } 276df8bae1dSRodney W. Grimes 277df8bae1dSRodney W. Grimes /* 27826f9a767SRodney W. Grimes * vm_object_init: 27926f9a767SRodney W. Grimes * 28026f9a767SRodney W. Grimes * Initialize the VM objects module. 28126f9a767SRodney W. Grimes */ 28226f9a767SRodney W. Grimes void 2831b40f8c0SMatthew Dillon vm_object_init(void) 28426f9a767SRodney W. Grimes { 28526f9a767SRodney W. Grimes TAILQ_INIT(&vm_object_list); 2866008862bSJohn Baldwin mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); 2870217125fSDavid Greenman 28889f6b863SAttilio Rao rw_init(&kernel_object->lock, "kernel vm object"); 289d1780e8dSKonstantin Belousov _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - 29067388836SKonstantin Belousov VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL); 291f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 292f8a47341SAlan Cox kernel_object->flags |= OBJ_COLORED; 293f8a47341SAlan Cox kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); 294f8a47341SAlan Cox #endif 295a720b31cSKonstantin Belousov kernel_object->un_pager.phys.ops = &default_phys_pg_ops; 29626f9a767SRodney W. Grimes 2978dbca793STor Egge /* 2988dbca793STor Egge * The lock portion of struct vm_object must be type stable due 2998dbca793STor Egge * to vm_pageout_fallback_object_lock locking a vm object 3008dbca793STor Egge * without holding any references to it. 301ffae7ea9SKonstantin Belousov * 302ffae7ea9SKonstantin Belousov * paging_in_progress is valid always. Lockless references to 303ffae7ea9SKonstantin Belousov * the objects may acquire pip and then check OBJ_DEAD. 3048dbca793STor Egge */ 3058355f576SJeff Roberson obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, 3068355f576SJeff Roberson #ifdef INVARIANTS 3078355f576SJeff Roberson vm_object_zdtor, 3088355f576SJeff Roberson #else 3098355f576SJeff Roberson NULL, 3108355f576SJeff Roberson #endif 3115df87b21SJeff Roberson vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 312774d251dSAttilio Rao 313cd1241fbSKonstantin Belousov vm_radix_zinit(); 31499448ed1SJohn Dyson } 31599448ed1SJohn Dyson 31699448ed1SJohn Dyson void 3171b40f8c0SMatthew Dillon vm_object_clear_flag(vm_object_t object, u_short bits) 3181b40f8c0SMatthew Dillon { 3195440b5a9SAlan Cox 32089f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 321b06805adSJake Burkholder object->flags &= ~bits; 3221b40f8c0SMatthew Dillon } 3231b40f8c0SMatthew Dillon 3243153e878SAlan Cox /* 3253153e878SAlan Cox * Sets the default memory attribute for the specified object. Pages 3263153e878SAlan Cox * that are allocated to this object are by default assigned this memory 3273153e878SAlan Cox * attribute. 3283153e878SAlan Cox * 3293153e878SAlan Cox * Presently, this function must be called before any pages are allocated 3303153e878SAlan Cox * to the object. In the future, this requirement may be relaxed for 3313153e878SAlan Cox * "default" and "swap" objects. 3323153e878SAlan Cox */ 3333153e878SAlan Cox int 3343153e878SAlan Cox vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) 3353153e878SAlan Cox { 3363153e878SAlan Cox 33789f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 3383e7a11caSKonstantin Belousov 3393e7a11caSKonstantin Belousov if (object->type == OBJT_DEAD) 3403e7a11caSKonstantin Belousov return (KERN_INVALID_ARGUMENT); 3413153e878SAlan Cox if (!TAILQ_EMPTY(&object->memq)) 3423153e878SAlan Cox return (KERN_FAILURE); 3433e7a11caSKonstantin Belousov 3443153e878SAlan Cox object->memattr = memattr; 3453153e878SAlan Cox return (KERN_SUCCESS); 3463153e878SAlan Cox } 3473153e878SAlan Cox 3481b40f8c0SMatthew Dillon void 3491b40f8c0SMatthew Dillon vm_object_pip_add(vm_object_t object, short i) 3501b40f8c0SMatthew Dillon { 351f279b88dSAlan Cox 352c99d0c58SMark Johnston if (i > 0) 353c99d0c58SMark Johnston blockcount_acquire(&object->paging_in_progress, i); 3541b40f8c0SMatthew Dillon } 3551b40f8c0SMatthew Dillon 3561b40f8c0SMatthew Dillon void 3571b40f8c0SMatthew Dillon vm_object_pip_wakeup(vm_object_t object) 3581b40f8c0SMatthew Dillon { 359f279b88dSAlan Cox 360c99d0c58SMark Johnston vm_object_pip_wakeupn(object, 1); 3611b40f8c0SMatthew Dillon } 3621b40f8c0SMatthew Dillon 3631b40f8c0SMatthew Dillon void 3641b40f8c0SMatthew Dillon vm_object_pip_wakeupn(vm_object_t object, short i) 3651b40f8c0SMatthew Dillon { 366d647a0edSAlan Cox 367c99d0c58SMark Johnston if (i > 0) 368c99d0c58SMark Johnston blockcount_release(&object->paging_in_progress, i); 3691b40f8c0SMatthew Dillon } 3701b40f8c0SMatthew Dillon 37198087a06SJeff Roberson /* 372c99d0c58SMark Johnston * Atomically drop the object lock and wait for pip to drain. This protects 373c99d0c58SMark Johnston * from sleep/wakeup races due to identity changes. The lock is not re-acquired 374c99d0c58SMark Johnston * on return. 37598087a06SJeff Roberson */ 37698087a06SJeff Roberson static void 377eaa17d42SRyan Libby vm_object_pip_sleep(vm_object_t object, const char *waitid) 37898087a06SJeff Roberson { 37998087a06SJeff Roberson 380c99d0c58SMark Johnston (void)blockcount_sleep(&object->paging_in_progress, &object->lock, 381c99d0c58SMark Johnston waitid, PVM | PDROP); 38298087a06SJeff Roberson } 38398087a06SJeff Roberson 3841b40f8c0SMatthew Dillon void 385eaa17d42SRyan Libby vm_object_pip_wait(vm_object_t object, const char *waitid) 3861b40f8c0SMatthew Dillon { 3871ca58953SAlan Cox 38889f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 389cf27e0d1SJeff Roberson 390c99d0c58SMark Johnston blockcount_wait(&object->paging_in_progress, &object->lock, waitid, 391c99d0c58SMark Johnston PVM); 3921b40f8c0SMatthew Dillon } 3931b40f8c0SMatthew Dillon 394cf27e0d1SJeff Roberson void 395eaa17d42SRyan Libby vm_object_pip_wait_unlocked(vm_object_t object, const char *waitid) 396cf27e0d1SJeff Roberson { 397cf27e0d1SJeff Roberson 398cf27e0d1SJeff Roberson VM_OBJECT_ASSERT_UNLOCKED(object); 399cf27e0d1SJeff Roberson 400c99d0c58SMark Johnston blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM); 401cf27e0d1SJeff Roberson } 402cf27e0d1SJeff Roberson 40326f9a767SRodney W. Grimes /* 40426f9a767SRodney W. Grimes * vm_object_allocate: 40526f9a767SRodney W. Grimes * 40626f9a767SRodney W. Grimes * Returns a new object with the given size. 40726f9a767SRodney W. Grimes */ 40826f9a767SRodney W. Grimes vm_object_t 4096395da54SIan Dowse vm_object_allocate(objtype_t type, vm_pindex_t size) 41026f9a767SRodney W. Grimes { 41190688d13SAlan Cox vm_object_t object; 41263967687SJeff Roberson u_short flags; 41363967687SJeff Roberson 41463967687SJeff Roberson switch (type) { 41563967687SJeff Roberson case OBJT_DEAD: 41663967687SJeff Roberson panic("vm_object_allocate: can't create OBJT_DEAD"); 4174b8365d7SKonstantin Belousov case OBJT_SWAP: 4184b8365d7SKonstantin Belousov flags = OBJ_COLORED | OBJ_SWAP; 4194b8365d7SKonstantin Belousov break; 42063967687SJeff Roberson case OBJT_DEVICE: 42163967687SJeff Roberson case OBJT_SG: 42263967687SJeff Roberson flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; 42363967687SJeff Roberson break; 42463967687SJeff Roberson case OBJT_MGTDEVICE: 42563967687SJeff Roberson flags = OBJ_FICTITIOUS; 42663967687SJeff Roberson break; 42763967687SJeff Roberson case OBJT_PHYS: 42863967687SJeff Roberson flags = OBJ_UNMANAGED; 42963967687SJeff Roberson break; 43063967687SJeff Roberson case OBJT_VNODE: 43163967687SJeff Roberson flags = 0; 43263967687SJeff Roberson break; 43363967687SJeff Roberson default: 434b730fd30SKonstantin Belousov panic("vm_object_allocate: type %d is undefined or dynamic", 435b730fd30SKonstantin Belousov type); 43663967687SJeff Roberson } 43763967687SJeff Roberson object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 43867388836SKonstantin Belousov _vm_object_allocate(type, size, flags, object, NULL); 43963967687SJeff Roberson 44063967687SJeff Roberson return (object); 44163967687SJeff Roberson } 44263967687SJeff Roberson 443b730fd30SKonstantin Belousov vm_object_t 444b730fd30SKonstantin Belousov vm_object_allocate_dyn(objtype_t dyntype, vm_pindex_t size, u_short flags) 445b730fd30SKonstantin Belousov { 446b730fd30SKonstantin Belousov vm_object_t object; 447b730fd30SKonstantin Belousov 448b730fd30SKonstantin Belousov MPASS(dyntype >= OBJT_FIRST_DYN /* && dyntype < nitems(pagertab) */); 449b730fd30SKonstantin Belousov object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); 450b730fd30SKonstantin Belousov _vm_object_allocate(dyntype, size, flags, object, NULL); 451b730fd30SKonstantin Belousov 452b730fd30SKonstantin Belousov return (object); 453b730fd30SKonstantin Belousov } 454b730fd30SKonstantin Belousov 45563967687SJeff Roberson /* 45663967687SJeff Roberson * vm_object_allocate_anon: 45763967687SJeff Roberson * 45863967687SJeff Roberson * Returns a new default object of the given size and marked as 45963967687SJeff Roberson * anonymous memory for special split/collapse handling. Color 46063967687SJeff Roberson * to be initialized by the caller. 46163967687SJeff Roberson */ 46263967687SJeff Roberson vm_object_t 46367388836SKonstantin Belousov vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object, 46467388836SKonstantin Belousov struct ucred *cred, vm_size_t charge) 46563967687SJeff Roberson { 46667388836SKonstantin Belousov vm_object_t handle, object; 46790688d13SAlan Cox 46867388836SKonstantin Belousov if (backing_object == NULL) 46967388836SKonstantin Belousov handle = NULL; 47067388836SKonstantin Belousov else if ((backing_object->flags & OBJ_ANON) != 0) 47167388836SKonstantin Belousov handle = backing_object->handle; 47267388836SKonstantin Belousov else 47367388836SKonstantin Belousov handle = backing_object; 47467388836SKonstantin Belousov object = uma_zalloc(obj_zone, M_WAITOK); 4755d32157dSMark Johnston _vm_object_allocate(OBJT_SWAP, size, 4765d32157dSMark Johnston OBJ_ANON | OBJ_ONEMAPPING | OBJ_SWAP, object, handle); 47767388836SKonstantin Belousov object->cred = cred; 47867388836SKonstantin Belousov object->charge = cred != NULL ? charge : 0; 47990688d13SAlan Cox return (object); 48026f9a767SRodney W. Grimes } 48126f9a767SRodney W. Grimes 48298087a06SJeff Roberson static void 48398087a06SJeff Roberson vm_object_reference_vnode(vm_object_t object) 484df8bae1dSRodney W. Grimes { 485a67d5408SJeff Roberson u_int old; 486a67d5408SJeff Roberson 487a67d5408SJeff Roberson /* 48898087a06SJeff Roberson * vnode objects need the lock for the first reference 48998087a06SJeff Roberson * to serialize with vnode_object_deallocate(). 490a67d5408SJeff Roberson */ 49198087a06SJeff Roberson if (!refcount_acquire_if_gt(&object->ref_count, 0)) { 49251df5321SJeff Roberson VM_OBJECT_RLOCK(object); 493a67d5408SJeff Roberson old = refcount_acquire(&object->ref_count); 4941a0c234eSJeff Roberson if (object->type == OBJT_VNODE && old == 0) 4951a0c234eSJeff Roberson vref(object->handle); 49651df5321SJeff Roberson VM_OBJECT_RUNLOCK(object); 49795e5e988SJohn Dyson } 498a67d5408SJeff Roberson } 49995e5e988SJohn Dyson 50023955314SAlfred Perlstein /* 50198087a06SJeff Roberson * vm_object_reference: 50298087a06SJeff Roberson * 50398087a06SJeff Roberson * Acquires a reference to the given object. 50498087a06SJeff Roberson */ 50598087a06SJeff Roberson void 50698087a06SJeff Roberson vm_object_reference(vm_object_t object) 50798087a06SJeff Roberson { 50898087a06SJeff Roberson 50998087a06SJeff Roberson if (object == NULL) 51098087a06SJeff Roberson return; 51198087a06SJeff Roberson 51298087a06SJeff Roberson if (object->type == OBJT_VNODE) 51398087a06SJeff Roberson vm_object_reference_vnode(object); 51498087a06SJeff Roberson else 51598087a06SJeff Roberson refcount_acquire(&object->ref_count); 51698087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 51798087a06SJeff Roberson ("vm_object_reference: Referenced dead object.")); 51898087a06SJeff Roberson } 51998087a06SJeff Roberson 52098087a06SJeff Roberson /* 521b921a12bSAlan Cox * vm_object_reference_locked: 522b921a12bSAlan Cox * 523b921a12bSAlan Cox * Gets another reference to the given object. 524b921a12bSAlan Cox * 525b921a12bSAlan Cox * The object must be locked. 526b921a12bSAlan Cox */ 527b921a12bSAlan Cox void 528b921a12bSAlan Cox vm_object_reference_locked(vm_object_t object) 529b921a12bSAlan Cox { 530a67d5408SJeff Roberson u_int old; 531b921a12bSAlan Cox 53251df5321SJeff Roberson VM_OBJECT_ASSERT_LOCKED(object); 533a67d5408SJeff Roberson old = refcount_acquire(&object->ref_count); 5341a0c234eSJeff Roberson if (object->type == OBJT_VNODE && old == 0) 5351a0c234eSJeff Roberson vref(object->handle); 53698087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 53798087a06SJeff Roberson ("vm_object_reference: Referenced dead object.")); 538b921a12bSAlan Cox } 539b921a12bSAlan Cox 540b921a12bSAlan Cox /* 5419d5abbddSJens Schweikhardt * Handle deallocating an object of type OBJT_VNODE. 54223955314SAlfred Perlstein */ 54302dd8331SAlan Cox static void 54498087a06SJeff Roberson vm_object_deallocate_vnode(vm_object_t object) 54595e5e988SJohn Dyson { 54695e5e988SJohn Dyson struct vnode *vp = (struct vnode *) object->handle; 54726c4e983SJeff Roberson bool last; 548219cbf59SEivind Eklund 5495526d2d9SEivind Eklund KASSERT(object->type == OBJT_VNODE, 55098087a06SJeff Roberson ("vm_object_deallocate_vnode: not a vnode object")); 55198087a06SJeff Roberson KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp")); 55295e5e988SJohn Dyson 55326c4e983SJeff Roberson /* Object lock to protect handle lookup. */ 55426c4e983SJeff Roberson last = refcount_release(&object->ref_count); 55526c4e983SJeff Roberson VM_OBJECT_RUNLOCK(object); 55626c4e983SJeff Roberson 55726c4e983SJeff Roberson if (!last) 55826c4e983SJeff Roberson return; 55926c4e983SJeff Roberson 560a67d5408SJeff Roberson if (!umtx_shm_vnobj_persistent) 5611bdbd705SKonstantin Belousov umtx_shm_object_terminated(object); 5621bdbd705SKonstantin Belousov 56303fa5b34SKonstantin Belousov /* vrele may need the vnode lock. */ 56447221757SJohn Dyson vrele(vp); 56586769ac0SKonstantin Belousov } 566df8bae1dSRodney W. Grimes 56798087a06SJeff Roberson /* 56898087a06SJeff Roberson * We dropped a reference on an object and discovered that it had a 56998087a06SJeff Roberson * single remaining shadow. This is a sibling of the reference we 57098087a06SJeff Roberson * dropped. Attempt to collapse the sibling and backing object. 57198087a06SJeff Roberson */ 57298087a06SJeff Roberson static vm_object_t 57398087a06SJeff Roberson vm_object_deallocate_anon(vm_object_t backing_object) 57498087a06SJeff Roberson { 57598087a06SJeff Roberson vm_object_t object; 57698087a06SJeff Roberson 57798087a06SJeff Roberson /* Fetch the final shadow. */ 57898087a06SJeff Roberson object = LIST_FIRST(&backing_object->shadow_head); 5794cf9f5d8SKonstantin Belousov KASSERT(object != NULL && 5804cf9f5d8SKonstantin Belousov atomic_load_int(&backing_object->shadow_count) == 1, 58198087a06SJeff Roberson ("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d", 5824cf9f5d8SKonstantin Belousov backing_object->ref_count, 5834cf9f5d8SKonstantin Belousov atomic_load_int(&backing_object->shadow_count))); 5844b8365d7SKonstantin Belousov KASSERT((object->flags & OBJ_ANON) != 0, 58598087a06SJeff Roberson ("invalid shadow object %p", object)); 58698087a06SJeff Roberson 58798087a06SJeff Roberson if (!VM_OBJECT_TRYWLOCK(object)) { 58898087a06SJeff Roberson /* 58998087a06SJeff Roberson * Prevent object from disappearing since we do not have a 59098087a06SJeff Roberson * reference. 59198087a06SJeff Roberson */ 59298087a06SJeff Roberson vm_object_pip_add(object, 1); 59398087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 59498087a06SJeff Roberson VM_OBJECT_WLOCK(object); 59598087a06SJeff Roberson vm_object_pip_wakeup(object); 59698087a06SJeff Roberson } else 59798087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 59898087a06SJeff Roberson 59998087a06SJeff Roberson /* 60098087a06SJeff Roberson * Check for a collapse/terminate race with the last reference holder. 60198087a06SJeff Roberson */ 60298087a06SJeff Roberson if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 || 60398087a06SJeff Roberson !refcount_acquire_if_not_zero(&object->ref_count)) { 60498087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 60598087a06SJeff Roberson return (NULL); 60698087a06SJeff Roberson } 60798087a06SJeff Roberson backing_object = object->backing_object; 60898087a06SJeff Roberson if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0) 60998087a06SJeff Roberson vm_object_collapse(object); 61098087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 61198087a06SJeff Roberson 61298087a06SJeff Roberson return (object); 61398087a06SJeff Roberson } 61498087a06SJeff Roberson 615df8bae1dSRodney W. Grimes /* 616df8bae1dSRodney W. Grimes * vm_object_deallocate: 617df8bae1dSRodney W. Grimes * 618df8bae1dSRodney W. Grimes * Release a reference to the specified object, 619df8bae1dSRodney W. Grimes * gained either through a vm_object_allocate 620df8bae1dSRodney W. Grimes * or a vm_object_reference call. When all references 621df8bae1dSRodney W. Grimes * are gone, storage associated with this object 622df8bae1dSRodney W. Grimes * may be relinquished. 623df8bae1dSRodney W. Grimes * 624df8bae1dSRodney W. Grimes * No object may be locked. 625df8bae1dSRodney W. Grimes */ 62626f9a767SRodney W. Grimes void 6271b40f8c0SMatthew Dillon vm_object_deallocate(vm_object_t object) 628df8bae1dSRodney W. Grimes { 62998087a06SJeff Roberson vm_object_t temp; 63026c4e983SJeff Roberson bool released; 631df8bae1dSRodney W. Grimes 632df8bae1dSRodney W. Grimes while (object != NULL) { 63351df5321SJeff Roberson /* 63451df5321SJeff Roberson * If the reference count goes to 0 we start calling 63551df5321SJeff Roberson * vm_object_terminate() on the object chain. A ref count 63651df5321SJeff Roberson * of 1 may be a special case depending on the shadow count 63751df5321SJeff Roberson * being 0 or 1. These cases require a write lock on the 63851df5321SJeff Roberson * object. 63951df5321SJeff Roberson */ 64063967687SJeff Roberson if ((object->flags & OBJ_ANON) == 0) 64163967687SJeff Roberson released = refcount_release_if_gt(&object->ref_count, 1); 64263967687SJeff Roberson else 64351df5321SJeff Roberson released = refcount_release_if_gt(&object->ref_count, 2); 64451df5321SJeff Roberson if (released) 64551df5321SJeff Roberson return; 64651df5321SJeff Roberson 647a67d5408SJeff Roberson if (object->type == OBJT_VNODE) { 64826c4e983SJeff Roberson VM_OBJECT_RLOCK(object); 64926c4e983SJeff Roberson if (object->type == OBJT_VNODE) { 65098087a06SJeff Roberson vm_object_deallocate_vnode(object); 651a67d5408SJeff Roberson return; 652a67d5408SJeff Roberson } 65326c4e983SJeff Roberson VM_OBJECT_RUNLOCK(object); 65426c4e983SJeff Roberson } 65526c4e983SJeff Roberson 65626c4e983SJeff Roberson VM_OBJECT_WLOCK(object); 65726c4e983SJeff Roberson KASSERT(object->ref_count > 0, 65826c4e983SJeff Roberson ("vm_object_deallocate: object deallocated too many times: %d", 65926c4e983SJeff Roberson object->type)); 66026c4e983SJeff Roberson 661b72b0115SAlan Cox /* 66298087a06SJeff Roberson * If this is not the final reference to an anonymous 66398087a06SJeff Roberson * object we may need to collapse the shadow chain. 664b72b0115SAlan Cox */ 66598087a06SJeff Roberson if (!refcount_release(&object->ref_count)) { 66698087a06SJeff Roberson if (object->ref_count > 1 || 6674cf9f5d8SKonstantin Belousov atomic_load_int(&object->shadow_count) == 0) { 66898087a06SJeff Roberson if ((object->flags & OBJ_ANON) != 0 && 66998087a06SJeff Roberson object->ref_count == 1) 67098087a06SJeff Roberson vm_object_set_flag(object, 67198087a06SJeff Roberson OBJ_ONEMAPPING); 67289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 67323b186d3SAlan Cox return; 67495e5e988SJohn Dyson } 67598087a06SJeff Roberson 67698087a06SJeff Roberson /* Handle collapsing last ref on anonymous objects. */ 67798087a06SJeff Roberson object = vm_object_deallocate_anon(object); 67898087a06SJeff Roberson continue; 67998087a06SJeff Roberson } 68098087a06SJeff Roberson 68198087a06SJeff Roberson /* 68298087a06SJeff Roberson * Handle the final reference to an object. We restart 68398087a06SJeff Roberson * the loop with the backing object to avoid recursion. 68498087a06SJeff Roberson */ 6851bdbd705SKonstantin Belousov umtx_shm_object_terminated(object); 68624a1cce3SDavid Greenman temp = object->backing_object; 687c9917419SAlan Cox if (temp != NULL) { 6880cb2610eSMark Johnston KASSERT(object->type == OBJT_SWAP, 6894bace8e7SKonstantin Belousov ("shadowed tmpfs v_object 2 %p", object)); 69051b867e5SJeff Roberson vm_object_backing_remove(object); 691de5f6a77SJohn Dyson } 69298087a06SJeff Roberson 69398087a06SJeff Roberson KASSERT((object->flags & OBJ_DEAD) == 0, 69498087a06SJeff Roberson ("vm_object_deallocate: Terminating dead object.")); 695783a68aaSKonstantin Belousov vm_object_set_flag(object, OBJ_DEAD); 696df8bae1dSRodney W. Grimes vm_object_terminate(object); 697df8bae1dSRodney W. Grimes object = temp; 698df8bae1dSRodney W. Grimes } 699df8bae1dSRodney W. Grimes } 700df8bae1dSRodney W. Grimes 7012ac78f0eSStephan Uphoff void 7022ac78f0eSStephan Uphoff vm_object_destroy(vm_object_t object) 7032ac78f0eSStephan Uphoff { 7042ac78f0eSStephan Uphoff uma_zfree(obj_zone, object); 7052ac78f0eSStephan Uphoff } 7062ac78f0eSStephan Uphoff 70751b867e5SJeff Roberson static void 7084cf9f5d8SKonstantin Belousov vm_object_sub_shadow(vm_object_t object) 7094cf9f5d8SKonstantin Belousov { 7104cf9f5d8SKonstantin Belousov KASSERT(object->shadow_count >= 1, 7114cf9f5d8SKonstantin Belousov ("object %p sub_shadow count zero", object)); 7124cf9f5d8SKonstantin Belousov atomic_subtract_int(&object->shadow_count, 1); 7134cf9f5d8SKonstantin Belousov } 7144cf9f5d8SKonstantin Belousov 7154cf9f5d8SKonstantin Belousov static void 71651b867e5SJeff Roberson vm_object_backing_remove_locked(vm_object_t object) 71751b867e5SJeff Roberson { 71851b867e5SJeff Roberson vm_object_t backing_object; 71951b867e5SJeff Roberson 72051b867e5SJeff Roberson backing_object = object->backing_object; 72151b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 72251b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(backing_object); 72351b867e5SJeff Roberson 72498087a06SJeff Roberson KASSERT((object->flags & OBJ_COLLAPSING) == 0, 72598087a06SJeff Roberson ("vm_object_backing_remove: Removing collapsing object.")); 72698087a06SJeff Roberson 7274cf9f5d8SKonstantin Belousov vm_object_sub_shadow(backing_object); 72851b867e5SJeff Roberson if ((object->flags & OBJ_SHADOWLIST) != 0) { 72951b867e5SJeff Roberson LIST_REMOVE(object, shadow_list); 730630f633fSMark Johnston vm_object_clear_flag(object, OBJ_SHADOWLIST); 73151b867e5SJeff Roberson } 73251b867e5SJeff Roberson object->backing_object = NULL; 73351b867e5SJeff Roberson } 73451b867e5SJeff Roberson 73551b867e5SJeff Roberson static void 73651b867e5SJeff Roberson vm_object_backing_remove(vm_object_t object) 73751b867e5SJeff Roberson { 73851b867e5SJeff Roberson vm_object_t backing_object; 73951b867e5SJeff Roberson 74051b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 74151b867e5SJeff Roberson 74251b867e5SJeff Roberson backing_object = object->backing_object; 7434cf9f5d8SKonstantin Belousov if ((object->flags & OBJ_SHADOWLIST) != 0) { 74451b867e5SJeff Roberson VM_OBJECT_WLOCK(backing_object); 74551b867e5SJeff Roberson vm_object_backing_remove_locked(object); 74651b867e5SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 7474cf9f5d8SKonstantin Belousov } else { 74851b867e5SJeff Roberson object->backing_object = NULL; 7494cf9f5d8SKonstantin Belousov vm_object_sub_shadow(backing_object); 7504cf9f5d8SKonstantin Belousov } 75151b867e5SJeff Roberson } 75251b867e5SJeff Roberson 75351b867e5SJeff Roberson static void 75451b867e5SJeff Roberson vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object) 75551b867e5SJeff Roberson { 75651b867e5SJeff Roberson 75751b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 75851b867e5SJeff Roberson 7594cf9f5d8SKonstantin Belousov atomic_add_int(&backing_object->shadow_count, 1); 76051b867e5SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 76151b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(backing_object); 76251b867e5SJeff Roberson LIST_INSERT_HEAD(&backing_object->shadow_head, object, 76351b867e5SJeff Roberson shadow_list); 764630f633fSMark Johnston vm_object_set_flag(object, OBJ_SHADOWLIST); 76551b867e5SJeff Roberson } 76651b867e5SJeff Roberson object->backing_object = backing_object; 76751b867e5SJeff Roberson } 76851b867e5SJeff Roberson 76951b867e5SJeff Roberson static void 77051b867e5SJeff Roberson vm_object_backing_insert(vm_object_t object, vm_object_t backing_object) 77151b867e5SJeff Roberson { 77251b867e5SJeff Roberson 77351b867e5SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 77451b867e5SJeff Roberson 77551b867e5SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 77651b867e5SJeff Roberson VM_OBJECT_WLOCK(backing_object); 77751b867e5SJeff Roberson vm_object_backing_insert_locked(object, backing_object); 77851b867e5SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 7794cf9f5d8SKonstantin Belousov } else { 78051b867e5SJeff Roberson object->backing_object = backing_object; 7814cf9f5d8SKonstantin Belousov atomic_add_int(&backing_object->shadow_count, 1); 7824cf9f5d8SKonstantin Belousov } 78351b867e5SJeff Roberson } 78451b867e5SJeff Roberson 78598087a06SJeff Roberson /* 78698087a06SJeff Roberson * Insert an object into a backing_object's shadow list with an additional 78798087a06SJeff Roberson * reference to the backing_object added. 78898087a06SJeff Roberson */ 78998087a06SJeff Roberson static void 79098087a06SJeff Roberson vm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object) 79198087a06SJeff Roberson { 79298087a06SJeff Roberson 79398087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 79498087a06SJeff Roberson 79598087a06SJeff Roberson if ((backing_object->flags & OBJ_ANON) != 0) { 79698087a06SJeff Roberson VM_OBJECT_WLOCK(backing_object); 79798087a06SJeff Roberson KASSERT((backing_object->flags & OBJ_DEAD) == 0, 79898087a06SJeff Roberson ("shadowing dead anonymous object")); 79998087a06SJeff Roberson vm_object_reference_locked(backing_object); 80098087a06SJeff Roberson vm_object_backing_insert_locked(object, backing_object); 80198087a06SJeff Roberson vm_object_clear_flag(backing_object, OBJ_ONEMAPPING); 80298087a06SJeff Roberson VM_OBJECT_WUNLOCK(backing_object); 80398087a06SJeff Roberson } else { 80498087a06SJeff Roberson vm_object_reference(backing_object); 8054cf9f5d8SKonstantin Belousov atomic_add_int(&backing_object->shadow_count, 1); 80698087a06SJeff Roberson object->backing_object = backing_object; 80798087a06SJeff Roberson } 80898087a06SJeff Roberson } 80998087a06SJeff Roberson 81098087a06SJeff Roberson /* 81198087a06SJeff Roberson * Transfer a backing reference from backing_object to object. 81298087a06SJeff Roberson */ 81398087a06SJeff Roberson static void 81498087a06SJeff Roberson vm_object_backing_transfer(vm_object_t object, vm_object_t backing_object) 81598087a06SJeff Roberson { 81698087a06SJeff Roberson vm_object_t new_backing_object; 81798087a06SJeff Roberson 81898087a06SJeff Roberson /* 81998087a06SJeff Roberson * Note that the reference to backing_object->backing_object 82098087a06SJeff Roberson * moves from within backing_object to within object. 82198087a06SJeff Roberson */ 82298087a06SJeff Roberson vm_object_backing_remove_locked(object); 82398087a06SJeff Roberson new_backing_object = backing_object->backing_object; 82498087a06SJeff Roberson if (new_backing_object == NULL) 82598087a06SJeff Roberson return; 82698087a06SJeff Roberson if ((new_backing_object->flags & OBJ_ANON) != 0) { 82798087a06SJeff Roberson VM_OBJECT_WLOCK(new_backing_object); 82898087a06SJeff Roberson vm_object_backing_remove_locked(backing_object); 82998087a06SJeff Roberson vm_object_backing_insert_locked(object, new_backing_object); 83098087a06SJeff Roberson VM_OBJECT_WUNLOCK(new_backing_object); 83198087a06SJeff Roberson } else { 8324cf9f5d8SKonstantin Belousov /* 8334cf9f5d8SKonstantin Belousov * shadow_count for new_backing_object is left 8344cf9f5d8SKonstantin Belousov * unchanged, its reference provided by backing_object 8354cf9f5d8SKonstantin Belousov * is replaced by object. 8364cf9f5d8SKonstantin Belousov */ 83798087a06SJeff Roberson object->backing_object = new_backing_object; 83898087a06SJeff Roberson backing_object->backing_object = NULL; 83998087a06SJeff Roberson } 84098087a06SJeff Roberson } 84198087a06SJeff Roberson 84298087a06SJeff Roberson /* 84398087a06SJeff Roberson * Wait for a concurrent collapse to settle. 84498087a06SJeff Roberson */ 84598087a06SJeff Roberson static void 84698087a06SJeff Roberson vm_object_collapse_wait(vm_object_t object) 84798087a06SJeff Roberson { 84898087a06SJeff Roberson 84998087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 85098087a06SJeff Roberson 85198087a06SJeff Roberson while ((object->flags & OBJ_COLLAPSING) != 0) { 85298087a06SJeff Roberson vm_object_pip_wait(object, "vmcolwait"); 85398087a06SJeff Roberson counter_u64_add(object_collapse_waits, 1); 85498087a06SJeff Roberson } 85598087a06SJeff Roberson } 85698087a06SJeff Roberson 85798087a06SJeff Roberson /* 85898087a06SJeff Roberson * Waits for a backing object to clear a pending collapse and returns 85998087a06SJeff Roberson * it locked if it is an ANON object. 86098087a06SJeff Roberson */ 86198087a06SJeff Roberson static vm_object_t 86298087a06SJeff Roberson vm_object_backing_collapse_wait(vm_object_t object) 86398087a06SJeff Roberson { 86498087a06SJeff Roberson vm_object_t backing_object; 86598087a06SJeff Roberson 86698087a06SJeff Roberson VM_OBJECT_ASSERT_WLOCKED(object); 86798087a06SJeff Roberson 86898087a06SJeff Roberson for (;;) { 86998087a06SJeff Roberson backing_object = object->backing_object; 87098087a06SJeff Roberson if (backing_object == NULL || 87198087a06SJeff Roberson (backing_object->flags & OBJ_ANON) == 0) 87298087a06SJeff Roberson return (NULL); 87398087a06SJeff Roberson VM_OBJECT_WLOCK(backing_object); 87498087a06SJeff Roberson if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0) 87598087a06SJeff Roberson break; 876cdd02f43SMark Johnston VM_OBJECT_WUNLOCK(object); 87798087a06SJeff Roberson vm_object_pip_sleep(backing_object, "vmbckwait"); 87898087a06SJeff Roberson counter_u64_add(object_collapse_waits, 1); 87998087a06SJeff Roberson VM_OBJECT_WLOCK(object); 88098087a06SJeff Roberson } 88198087a06SJeff Roberson return (backing_object); 88298087a06SJeff Roberson } 88351b867e5SJeff Roberson 8842ac78f0eSStephan Uphoff /* 8857bbdb843SRuslan Bukin * vm_object_terminate_pages removes any remaining pageable pages 8867bbdb843SRuslan Bukin * from the object and resets the object to an empty state. 8877bbdb843SRuslan Bukin */ 8887bbdb843SRuslan Bukin static void 8897bbdb843SRuslan Bukin vm_object_terminate_pages(vm_object_t object) 8907bbdb843SRuslan Bukin { 8917bbdb843SRuslan Bukin vm_page_t p, p_next; 8927bbdb843SRuslan Bukin 8937bbdb843SRuslan Bukin VM_OBJECT_ASSERT_WLOCKED(object); 8947bbdb843SRuslan Bukin 8957bbdb843SRuslan Bukin /* 8967bbdb843SRuslan Bukin * Free any remaining pageable pages. This also removes them from the 8977bbdb843SRuslan Bukin * paging queues. However, don't free wired pages, just remove them 8987bbdb843SRuslan Bukin * from the object. Rather than incrementally removing each page from 8997bbdb843SRuslan Bukin * the object, the page and object are reset to any empty state. 9007bbdb843SRuslan Bukin */ 9017bbdb843SRuslan Bukin TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { 9027bbdb843SRuslan Bukin vm_page_assert_unbusied(p); 903fee2a2faSMark Johnston KASSERT(p->object == object && 904fee2a2faSMark Johnston (p->ref_count & VPRC_OBJREF) != 0, 905fee2a2faSMark Johnston ("vm_object_terminate_pages: page %p is inconsistent", p)); 906fee2a2faSMark Johnston 9077bbdb843SRuslan Bukin p->object = NULL; 908fee2a2faSMark Johnston if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) { 9095cd29d0fSMark Johnston VM_CNT_INC(v_pfree); 9105cd29d0fSMark Johnston vm_page_free(p); 9114074d642SAlan Cox } 912fee2a2faSMark Johnston } 9132fcd1ff6SKonstantin Belousov 9147bbdb843SRuslan Bukin /* 9157bbdb843SRuslan Bukin * If the object contained any pages, then reset it to an empty state. 9167bbdb843SRuslan Bukin * None of the object's fields, including "resident_page_count", were 9177bbdb843SRuslan Bukin * modified by the preceding loop. 9187bbdb843SRuslan Bukin */ 9197bbdb843SRuslan Bukin if (object->resident_page_count != 0) { 9207bbdb843SRuslan Bukin vm_radix_reclaim_allnodes(&object->rtree); 9217bbdb843SRuslan Bukin TAILQ_INIT(&object->memq); 9227bbdb843SRuslan Bukin object->resident_page_count = 0; 9237bbdb843SRuslan Bukin if (object->type == OBJT_VNODE) 9247bbdb843SRuslan Bukin vdrop(object->handle); 9257bbdb843SRuslan Bukin } 9267bbdb843SRuslan Bukin } 9277bbdb843SRuslan Bukin 9287bbdb843SRuslan Bukin /* 929df8bae1dSRodney W. Grimes * vm_object_terminate actually destroys the specified object, freeing 930df8bae1dSRodney W. Grimes * up all previously used resources. 931df8bae1dSRodney W. Grimes * 932df8bae1dSRodney W. Grimes * The object must be locked. 9331c7c3c6aSMatthew Dillon * This routine may block. 934df8bae1dSRodney W. Grimes */ 93595e5e988SJohn Dyson void 9361b40f8c0SMatthew Dillon vm_object_terminate(vm_object_t object) 937df8bae1dSRodney W. Grimes { 93898087a06SJeff Roberson 93989f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 940783a68aaSKonstantin Belousov KASSERT((object->flags & OBJ_DEAD) != 0, 941783a68aaSKonstantin Belousov ("terminating non-dead obj %p", object)); 94298087a06SJeff Roberson KASSERT((object->flags & OBJ_COLLAPSING) == 0, 94398087a06SJeff Roberson ("terminating collapsing obj %p", object)); 94498087a06SJeff Roberson KASSERT(object->backing_object == NULL, 94598087a06SJeff Roberson ("terminating shadow obj %p", object)); 946bef608bdSJohn Dyson 947cf27e0d1SJeff Roberson /* 948ffae7ea9SKonstantin Belousov * Wait for the pageout daemon and other current users to be 949ffae7ea9SKonstantin Belousov * done with the object. Note that new paging_in_progress 950ffae7ea9SKonstantin Belousov * users can come after this wait, but they must check 951ffae7ea9SKonstantin Belousov * OBJ_DEAD flag set (without unlocking the object), and avoid 952ffae7ea9SKonstantin Belousov * the object being terminated. 953cf27e0d1SJeff Roberson */ 954cf27e0d1SJeff Roberson vm_object_pip_wait(object, "objtrm"); 955cf27e0d1SJeff Roberson 956971dd342SAlfred Perlstein KASSERT(object->ref_count == 0, 957971dd342SAlfred Perlstein ("vm_object_terminate: object with references, ref_count=%d", 958971dd342SAlfred Perlstein object->ref_count)); 959996c772fSJohn Dyson 9607bbdb843SRuslan Bukin if ((object->flags & OBJ_PG_DTOR) == 0) 9617bbdb843SRuslan Bukin vm_object_terminate_pages(object); 962bef608bdSJohn Dyson 963f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 964f8a47341SAlan Cox if (__predict_false(!LIST_EMPTY(&object->rvq))) 965f8a47341SAlan Cox vm_reserv_break_all(object); 966f8a47341SAlan Cox #endif 9677bfda801SAlan Cox 9680cb2610eSMark Johnston KASSERT(object->cred == NULL || (object->flags & OBJ_SWAP) != 0, 969e735691bSJohn Baldwin ("%s: non-swap obj %p has cred", __func__, object)); 970e735691bSJohn Baldwin 9712d8acc0fSJohn Dyson /* 9729fcfb650SDavid Greenman * Let the pager know object is dead. 9739fcfb650SDavid Greenman */ 9749fcfb650SDavid Greenman vm_pager_deallocate(object); 97589f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 9769fcfb650SDavid Greenman 9772ac78f0eSStephan Uphoff vm_object_destroy(object); 97847221757SJohn Dyson } 979df8bae1dSRodney W. Grimes 980edf93b25SAlan Cox /* 981edf93b25SAlan Cox * Make the page read-only so that we can clear the object flags. However, if 982edf93b25SAlan Cox * this is a nosync mmap then the object is likely to stay dirty so do not 983edf93b25SAlan Cox * mess with the page and do not clear the object flags. Returns TRUE if the 984edf93b25SAlan Cox * page should be flushed, and FALSE otherwise. 985edf93b25SAlan Cox */ 9863280870dSKonstantin Belousov static boolean_t 98767d0e293SJeff Roberson vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean) 9883280870dSKonstantin Belousov { 9893280870dSKonstantin Belousov 990fff5403fSJeff Roberson vm_page_assert_busied(p); 991fff5403fSJeff Roberson 9923280870dSKonstantin Belousov /* 9933280870dSKonstantin Belousov * If we have been asked to skip nosync pages and this is a 9943280870dSKonstantin Belousov * nosync page, skip it. Note that the object flags were not 9953280870dSKonstantin Belousov * cleared in this case so we do not have to set them. 9963280870dSKonstantin Belousov */ 9975cff1f4dSMark Johnston if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) { 99867d0e293SJeff Roberson *allclean = FALSE; 9993280870dSKonstantin Belousov return (FALSE); 10003280870dSKonstantin Belousov } else { 10013280870dSKonstantin Belousov pmap_remove_write(p); 10023280870dSKonstantin Belousov return (p->dirty != 0); 10033280870dSKonstantin Belousov } 10043280870dSKonstantin Belousov } 10053280870dSKonstantin Belousov 1006df8bae1dSRodney W. Grimes /* 1007df8bae1dSRodney W. Grimes * vm_object_page_clean 1008df8bae1dSRodney W. Grimes * 10094f79d873SMatthew Dillon * Clean all dirty pages in the specified range of object. Leaves page 10104f79d873SMatthew Dillon * on whatever queue it is currently on. If NOSYNC is set then do not 1011fff5403fSJeff Roberson * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC), 10124f79d873SMatthew Dillon * leaving the object dirty. 101326f9a767SRodney W. Grimes * 10148d34a3bfSKonstantin Belousov * For swap objects backing tmpfs regular files, do not flush anything, 10158d34a3bfSKonstantin Belousov * but remove write protection on the mapped pages to update mtime through 10168d34a3bfSKonstantin Belousov * mmaped writes. 10178d34a3bfSKonstantin Belousov * 101843b7990eSMatthew Dillon * When stuffing pages asynchronously, allow clustering. XXX we need a 101943b7990eSMatthew Dillon * synchronous clustering mode implementation. 102043b7990eSMatthew Dillon * 102126f9a767SRodney W. Grimes * Odd semantics: if start == end, we clean everything. 102226f9a767SRodney W. Grimes * 102326f9a767SRodney W. Grimes * The object must be locked. 1024126d6082SKonstantin Belousov * 1025126d6082SKonstantin Belousov * Returns FALSE if some page from the range was not written, as 1026126d6082SKonstantin Belousov * reported by the pager, and TRUE otherwise. 102726f9a767SRodney W. Grimes */ 1028126d6082SKonstantin Belousov boolean_t 102917f3095dSAlan Cox vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, 1030e239bb97SKonstantin Belousov int flags) 1031f6b04d2bSDavid Greenman { 1032e239bb97SKonstantin Belousov vm_page_t np, p; 103317f3095dSAlan Cox vm_pindex_t pi, tend, tstart; 1034126d6082SKonstantin Belousov int curgeneration, n, pagerflags; 103567d0e293SJeff Roberson boolean_t eio, res, allclean; 1036f6b04d2bSDavid Greenman 103789f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 1038e5f299ffSKonstantin Belousov 10398d34a3bfSKonstantin Belousov if (!vm_object_mightbedirty(object) || object->resident_page_count == 0) 1040126d6082SKonstantin Belousov return (TRUE); 1041f6b04d2bSDavid Greenman 1042e239bb97SKonstantin Belousov pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? 1043e239bb97SKonstantin Belousov VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; 1044e239bb97SKonstantin Belousov pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; 1045e239bb97SKonstantin Belousov 104617f3095dSAlan Cox tstart = OFF_TO_IDX(start); 104717f3095dSAlan Cox tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); 104867d0e293SJeff Roberson allclean = tstart == 0 && tend >= object->size; 1049126d6082SKonstantin Belousov res = TRUE; 1050f6b04d2bSDavid Greenman 1051bd7e5f99SJohn Dyson rescan: 10522d8acc0fSJohn Dyson curgeneration = object->generation; 10532d8acc0fSJohn Dyson 105417f3095dSAlan Cox for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { 1055bd7e5f99SJohn Dyson pi = p->pindex; 1056e239bb97SKonstantin Belousov if (pi >= tend) 1057e239bb97SKonstantin Belousov break; 1058e239bb97SKonstantin Belousov np = TAILQ_NEXT(p, listq); 10590012f373SJeff Roberson if (vm_page_none_valid(p)) 1060aef922f5SJohn Dyson continue; 106163e97555SJeff Roberson if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) { 106267d0e293SJeff Roberson if (object->generation != curgeneration && 106367d0e293SJeff Roberson (flags & OBJPC_SYNC) != 0) 1064e239bb97SKonstantin Belousov goto rescan; 1065780636b7SKonstantin Belousov np = vm_page_find_least(object, pi); 1066780636b7SKonstantin Belousov continue; 1067f6b04d2bSDavid Greenman } 106867d0e293SJeff Roberson if (!vm_object_page_remove_write(p, flags, &allclean)) { 106963e97555SJeff Roberson vm_page_xunbusy(p); 1070bd7e5f99SJohn Dyson continue; 107163e97555SJeff Roberson } 10728d34a3bfSKonstantin Belousov if (object->type == OBJT_VNODE) { 10733280870dSKonstantin Belousov n = vm_object_page_collect_flush(object, p, pagerflags, 107467d0e293SJeff Roberson flags, &allclean, &eio); 1075126d6082SKonstantin Belousov if (eio) { 1076126d6082SKonstantin Belousov res = FALSE; 107767d0e293SJeff Roberson allclean = FALSE; 1078126d6082SKonstantin Belousov } 107967d0e293SJeff Roberson if (object->generation != curgeneration && 108067d0e293SJeff Roberson (flags & OBJPC_SYNC) != 0) 1081b9b7a4beSMatthew Dillon goto rescan; 1082031ec8c1SKonstantin Belousov 1083031ec8c1SKonstantin Belousov /* 1084031ec8c1SKonstantin Belousov * If the VOP_PUTPAGES() did a truncated write, so 1085031ec8c1SKonstantin Belousov * that even the first page of the run is not fully 1086031ec8c1SKonstantin Belousov * written, vm_pageout_flush() returns 0 as the run 1087031ec8c1SKonstantin Belousov * length. Since the condition that caused truncated 1088031ec8c1SKonstantin Belousov * write may be permanent, e.g. exhausted free space, 1089031ec8c1SKonstantin Belousov * accepting n == 0 would cause an infinite loop. 1090031ec8c1SKonstantin Belousov * 1091031ec8c1SKonstantin Belousov * Forwarding the iterator leaves the unwritten page 1092031ec8c1SKonstantin Belousov * behind, but there is not much we can do there if 1093031ec8c1SKonstantin Belousov * filesystem refuses to write it. 1094031ec8c1SKonstantin Belousov */ 1095126d6082SKonstantin Belousov if (n == 0) { 1096031ec8c1SKonstantin Belousov n = 1; 109767d0e293SJeff Roberson allclean = FALSE; 1098126d6082SKonstantin Belousov } 10998d34a3bfSKonstantin Belousov } else { 11008d34a3bfSKonstantin Belousov n = 1; 11018d34a3bfSKonstantin Belousov vm_page_xunbusy(p); 11028d34a3bfSKonstantin Belousov } 1103e239bb97SKonstantin Belousov np = vm_page_find_least(object, pi + n); 1104b9b7a4beSMatthew Dillon } 1105b9b7a4beSMatthew Dillon #if 0 1106e239bb97SKonstantin Belousov VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); 1107b9b7a4beSMatthew Dillon #endif 1108b9b7a4beSMatthew Dillon 11098d34a3bfSKonstantin Belousov /* 11108d34a3bfSKonstantin Belousov * Leave updating cleangeneration for tmpfs objects to tmpfs 11118d34a3bfSKonstantin Belousov * scan. It needs to update mtime, which happens for other 11128d34a3bfSKonstantin Belousov * filesystems during page writeouts. 11138d34a3bfSKonstantin Belousov */ 11148d34a3bfSKonstantin Belousov if (allclean && object->type == OBJT_VNODE) 111567d0e293SJeff Roberson object->cleangeneration = curgeneration; 1116126d6082SKonstantin Belousov return (res); 1117b9b7a4beSMatthew Dillon } 1118b9b7a4beSMatthew Dillon 1119b9b7a4beSMatthew Dillon static int 11203280870dSKonstantin Belousov vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, 112167d0e293SJeff Roberson int flags, boolean_t *allclean, boolean_t *eio) 1122b9b7a4beSMatthew Dillon { 11233157c503SKonstantin Belousov vm_page_t ma[vm_pageout_page_count], p_first, tp; 11243157c503SKonstantin Belousov int count, i, mreq, runlen; 1125b9b7a4beSMatthew Dillon 11267bec141bSKip Macy vm_page_lock_assert(p, MA_NOTOWNED); 112763e97555SJeff Roberson vm_page_assert_xbusied(p); 112889f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 11293157c503SKonstantin Belousov 11303157c503SKonstantin Belousov count = 1; 11313157c503SKonstantin Belousov mreq = 0; 11323157c503SKonstantin Belousov 11333157c503SKonstantin Belousov for (tp = p; count < vm_pageout_page_count; count++) { 11343157c503SKonstantin Belousov tp = vm_page_next(tp); 113563e97555SJeff Roberson if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1136bd7e5f99SJohn Dyson break; 113767d0e293SJeff Roberson if (!vm_object_page_remove_write(tp, flags, allclean)) { 113863e97555SJeff Roberson vm_page_xunbusy(tp); 1139bd7e5f99SJohn Dyson break; 1140bd7e5f99SJohn Dyson } 114163e97555SJeff Roberson } 1142aef922f5SJohn Dyson 11433157c503SKonstantin Belousov for (p_first = p; count < vm_pageout_page_count; count++) { 11443157c503SKonstantin Belousov tp = vm_page_prev(p_first); 114563e97555SJeff Roberson if (tp == NULL || vm_page_tryxbusy(tp) == 0) 1146bd7e5f99SJohn Dyson break; 114767d0e293SJeff Roberson if (!vm_object_page_remove_write(tp, flags, allclean)) { 114863e97555SJeff Roberson vm_page_xunbusy(tp); 1149bd7e5f99SJohn Dyson break; 115063e97555SJeff Roberson } 11513157c503SKonstantin Belousov p_first = tp; 11523157c503SKonstantin Belousov mreq++; 1153bd7e5f99SJohn Dyson } 1154bd7e5f99SJohn Dyson 11553157c503SKonstantin Belousov for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) 11563157c503SKonstantin Belousov ma[i] = tp; 1157cf2819ccSJohn Dyson 1158126d6082SKonstantin Belousov vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); 11591e8a675cSKonstantin Belousov return (runlen); 116026f9a767SRodney W. Grimes } 1161df8bae1dSRodney W. Grimes 11621efb74fbSJohn Dyson /* 1163950f8459SAlan Cox * Note that there is absolutely no sense in writing out 1164950f8459SAlan Cox * anonymous objects, so we track down the vnode object 1165950f8459SAlan Cox * to write out. 1166950f8459SAlan Cox * We invalidate (remove) all pages from the address space 1167950f8459SAlan Cox * for semantic correctness. 1168950f8459SAlan Cox * 11696bbee8e2SAlan Cox * If the backing object is a device object with unmanaged pages, then any 11706bbee8e2SAlan Cox * mappings to the specified range of pages must be removed before this 11716bbee8e2SAlan Cox * function is called. 11726bbee8e2SAlan Cox * 1173950f8459SAlan Cox * Note: certain anonymous maps, such as MAP_NOSYNC maps, 1174950f8459SAlan Cox * may start out with a NULL object. 1175950f8459SAlan Cox */ 1176126d6082SKonstantin Belousov boolean_t 1177950f8459SAlan Cox vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, 1178950f8459SAlan Cox boolean_t syncio, boolean_t invalidate) 1179950f8459SAlan Cox { 1180950f8459SAlan Cox vm_object_t backing_object; 1181950f8459SAlan Cox struct vnode *vp; 11823b582b4eSTor Egge struct mount *mp; 1183126d6082SKonstantin Belousov int error, flags, fsync_after; 1184126d6082SKonstantin Belousov boolean_t res; 1185950f8459SAlan Cox 1186950f8459SAlan Cox if (object == NULL) 1187126d6082SKonstantin Belousov return (TRUE); 1188126d6082SKonstantin Belousov res = TRUE; 1189126d6082SKonstantin Belousov error = 0; 119089f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1191950f8459SAlan Cox while ((backing_object = object->backing_object) != NULL) { 119289f6b863SAttilio Rao VM_OBJECT_WLOCK(backing_object); 119356e0670fSAlan Cox offset += object->backing_object_offset; 119489f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1195950f8459SAlan Cox object = backing_object; 1196950f8459SAlan Cox if (object->size < OFF_TO_IDX(offset + size)) 1197950f8459SAlan Cox size = IDX_TO_OFF(object->size) - offset; 1198950f8459SAlan Cox } 1199950f8459SAlan Cox /* 1200950f8459SAlan Cox * Flush pages if writing is allowed, invalidate them 1201950f8459SAlan Cox * if invalidation requested. Pages undergoing I/O 1202950f8459SAlan Cox * will be ignored by vm_object_page_remove(). 1203950f8459SAlan Cox * 1204950f8459SAlan Cox * We cannot lock the vnode and then wait for paging 1205950f8459SAlan Cox * to complete without deadlocking against vm_fault. 1206950f8459SAlan Cox * Instead we simply call vm_object_page_remove() and 1207950f8459SAlan Cox * allow it to block internally on a page-by-page 1208950f8459SAlan Cox * basis when it encounters pages undergoing async 1209950f8459SAlan Cox * I/O. 1210950f8459SAlan Cox */ 1211950f8459SAlan Cox if (object->type == OBJT_VNODE && 121267d0e293SJeff Roberson vm_object_mightbedirty(object) != 0 && 12135bf94937SKonstantin Belousov ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { 121489f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 12153b582b4eSTor Egge (void)vn_start_write(vp, &mp, V_WAIT); 1216cb05b60aSAttilio Rao vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 121775ff604aSKonstantin Belousov if (syncio && !invalidate && offset == 0 && 1218d1780e8dSKonstantin Belousov atop(size) == object->size) { 121975ff604aSKonstantin Belousov /* 122075ff604aSKonstantin Belousov * If syncing the whole mapping of the file, 122175ff604aSKonstantin Belousov * it is faster to schedule all the writes in 122275ff604aSKonstantin Belousov * async mode, also allowing the clustering, 122375ff604aSKonstantin Belousov * and then wait for i/o to complete. 122475ff604aSKonstantin Belousov */ 122575ff604aSKonstantin Belousov flags = 0; 122675ff604aSKonstantin Belousov fsync_after = TRUE; 122775ff604aSKonstantin Belousov } else { 1228950f8459SAlan Cox flags = (syncio || invalidate) ? OBJPC_SYNC : 0; 122975ff604aSKonstantin Belousov flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; 123075ff604aSKonstantin Belousov fsync_after = FALSE; 123175ff604aSKonstantin Belousov } 123289f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1233126d6082SKonstantin Belousov res = vm_object_page_clean(object, offset, offset + size, 1234126d6082SKonstantin Belousov flags); 123589f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1236*6189672eSKonstantin Belousov if (fsync_after) { 1237*6189672eSKonstantin Belousov for (;;) { 1238126d6082SKonstantin Belousov error = VOP_FSYNC(vp, MNT_WAIT, curthread); 1239*6189672eSKonstantin Belousov if (error != ERELOOKUP) 1240*6189672eSKonstantin Belousov break; 1241*6189672eSKonstantin Belousov 1242*6189672eSKonstantin Belousov /* 1243*6189672eSKonstantin Belousov * Allow SU/bufdaemon to handle more 1244*6189672eSKonstantin Belousov * dependencies in the meantime. 1245*6189672eSKonstantin Belousov */ 1246*6189672eSKonstantin Belousov VOP_UNLOCK(vp); 1247*6189672eSKonstantin Belousov vn_finished_write(mp); 1248*6189672eSKonstantin Belousov 1249*6189672eSKonstantin Belousov (void)vn_start_write(vp, &mp, V_WAIT); 1250*6189672eSKonstantin Belousov vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1251*6189672eSKonstantin Belousov } 1252*6189672eSKonstantin Belousov } 1253b249ce48SMateusz Guzik VOP_UNLOCK(vp); 12543b582b4eSTor Egge vn_finished_write(mp); 1255126d6082SKonstantin Belousov if (error != 0) 1256126d6082SKonstantin Belousov res = FALSE; 125789f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 1258950f8459SAlan Cox } 1259950f8459SAlan Cox if ((object->type == OBJT_VNODE || 1260950f8459SAlan Cox object->type == OBJT_DEVICE) && invalidate) { 12616bbee8e2SAlan Cox if (object->type == OBJT_DEVICE) 12626bbee8e2SAlan Cox /* 12636bbee8e2SAlan Cox * The option OBJPR_NOTMAPPED must be passed here 12646bbee8e2SAlan Cox * because vm_object_page_remove() cannot remove 12656bbee8e2SAlan Cox * unmanaged mappings. 12666bbee8e2SAlan Cox */ 12676bbee8e2SAlan Cox flags = OBJPR_NOTMAPPED; 12686bbee8e2SAlan Cox else if (old_msync) 12696195b24aSKonstantin Belousov flags = 0; 12706bbee8e2SAlan Cox else 12716195b24aSKonstantin Belousov flags = OBJPR_CLEANONLY; 12726bbee8e2SAlan Cox vm_object_page_remove(object, OFF_TO_IDX(offset), 12736bbee8e2SAlan Cox OFF_TO_IDX(offset + size + PAGE_MASK), flags); 1274950f8459SAlan Cox } 127589f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1276126d6082SKonstantin Belousov return (res); 1277950f8459SAlan Cox } 1278950f8459SAlan Cox 1279950f8459SAlan Cox /* 1280aa3650eaSMark Johnston * Determine whether the given advice can be applied to the object. Advice is 1281aa3650eaSMark Johnston * not applied to unmanaged pages since they never belong to page queues, and 1282aa3650eaSMark Johnston * since MADV_FREE is destructive, it can apply only to anonymous pages that 1283aa3650eaSMark Johnston * have been mapped at most once. 1284aa3650eaSMark Johnston */ 1285aa3650eaSMark Johnston static bool 1286aa3650eaSMark Johnston vm_object_advice_applies(vm_object_t object, int advice) 1287aa3650eaSMark Johnston { 1288aa3650eaSMark Johnston 1289aa3650eaSMark Johnston if ((object->flags & OBJ_UNMANAGED) != 0) 1290aa3650eaSMark Johnston return (false); 1291aa3650eaSMark Johnston if (advice != MADV_FREE) 1292aa3650eaSMark Johnston return (true); 129363967687SJeff Roberson return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) == 129463967687SJeff Roberson (OBJ_ONEMAPPING | OBJ_ANON)); 1295aa3650eaSMark Johnston } 1296aa3650eaSMark Johnston 1297aa3650eaSMark Johnston static void 1298aa3650eaSMark Johnston vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex, 1299aa3650eaSMark Johnston vm_size_t size) 1300aa3650eaSMark Johnston { 1301aa3650eaSMark Johnston 13021390a5cbSKonstantin Belousov if (advice == MADV_FREE) 13031390a5cbSKonstantin Belousov vm_pager_freespace(object, pindex, size); 1304aa3650eaSMark Johnston } 1305aa3650eaSMark Johnston 1306aa3650eaSMark Johnston /* 1307867a482dSJohn Dyson * vm_object_madvise: 1308867a482dSJohn Dyson * 1309867a482dSJohn Dyson * Implements the madvise function at the object/page level. 13101c7c3c6aSMatthew Dillon * 1311193b9358SAlan Cox * MADV_WILLNEED (any object) 1312193b9358SAlan Cox * 1313193b9358SAlan Cox * Activate the specified pages if they are resident. 1314193b9358SAlan Cox * 1315193b9358SAlan Cox * MADV_DONTNEED (any object) 1316193b9358SAlan Cox * 1317193b9358SAlan Cox * Deactivate the specified pages if they are resident. 1318193b9358SAlan Cox * 13190cb2610eSMark Johnston * MADV_FREE (OBJT_SWAP objects, OBJ_ONEMAPPING only) 1320193b9358SAlan Cox * 1321193b9358SAlan Cox * Deactivate and clean the specified pages if they are 1322193b9358SAlan Cox * resident. This permits the process to reuse the pages 1323193b9358SAlan Cox * without faulting or the kernel to reclaim the pages 1324193b9358SAlan Cox * without I/O. 1325867a482dSJohn Dyson */ 1326867a482dSJohn Dyson void 132792a59946SJohn Baldwin vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, 1328c2655a40SMark Johnston int advice) 1329867a482dSJohn Dyson { 133092a59946SJohn Baldwin vm_pindex_t tpindex; 133134567de7SAlan Cox vm_object_t backing_object, tobject; 1332aa3650eaSMark Johnston vm_page_t m, tm; 1333867a482dSJohn Dyson 1334867a482dSJohn Dyson if (object == NULL) 1335867a482dSJohn Dyson return; 1336c2655a40SMark Johnston 13376e20a165SJohn Dyson relookup: 1338aa3650eaSMark Johnston VM_OBJECT_WLOCK(object); 1339aa3650eaSMark Johnston if (!vm_object_advice_applies(object, advice)) { 1340aa3650eaSMark Johnston VM_OBJECT_WUNLOCK(object); 1341aa3650eaSMark Johnston return; 13426e20a165SJohn Dyson } 1343aa3650eaSMark Johnston for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) { 1344aa3650eaSMark Johnston tobject = object; 1345c2655a40SMark Johnston 13461ce137beSMatthew Dillon /* 1347aa3650eaSMark Johnston * If the next page isn't resident in the top-level object, we 1348aa3650eaSMark Johnston * need to search the shadow chain. When applying MADV_FREE, we 1349aa3650eaSMark Johnston * take care to release any swap space used to store 1350aa3650eaSMark Johnston * non-resident pages. 1351aa3650eaSMark Johnston */ 1352aa3650eaSMark Johnston if (m == NULL || pindex < m->pindex) { 1353aa3650eaSMark Johnston /* 1354aa3650eaSMark Johnston * Optimize a common case: if the top-level object has 1355aa3650eaSMark Johnston * no backing object, we can skip over the non-resident 1356aa3650eaSMark Johnston * range in constant time. 13571ce137beSMatthew Dillon */ 1358c2655a40SMark Johnston if (object->backing_object == NULL) { 1359c2655a40SMark Johnston tpindex = (m != NULL && m->pindex < end) ? 1360c2655a40SMark Johnston m->pindex : end; 1361aa3650eaSMark Johnston vm_object_madvise_freespace(object, advice, 1362aa3650eaSMark Johnston pindex, tpindex - pindex); 1363c2655a40SMark Johnston if ((pindex = tpindex) == end) 1364c2655a40SMark Johnston break; 1365aa3650eaSMark Johnston goto next_page; 1366aa3650eaSMark Johnston } 1367aa3650eaSMark Johnston 1368aa3650eaSMark Johnston tpindex = pindex; 1369aa3650eaSMark Johnston do { 1370aa3650eaSMark Johnston vm_object_madvise_freespace(tobject, advice, 1371aa3650eaSMark Johnston tpindex, 1); 13721ce137beSMatthew Dillon /* 1373aa3650eaSMark Johnston * Prepare to search the next object in the 1374aa3650eaSMark Johnston * chain. 13751ce137beSMatthew Dillon */ 137634567de7SAlan Cox backing_object = tobject->backing_object; 137734567de7SAlan Cox if (backing_object == NULL) 1378aa3650eaSMark Johnston goto next_pindex; 137989f6b863SAttilio Rao VM_OBJECT_WLOCK(backing_object); 1380aa3650eaSMark Johnston tpindex += 1381aa3650eaSMark Johnston OFF_TO_IDX(tobject->backing_object_offset); 13829b98b796SAlan Cox if (tobject != object) 138389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(tobject); 138434567de7SAlan Cox tobject = backing_object; 1385aa3650eaSMark Johnston if (!vm_object_advice_applies(tobject, advice)) 1386aa3650eaSMark Johnston goto next_pindex; 1387aa3650eaSMark Johnston } while ((tm = vm_page_lookup(tobject, tpindex)) == 1388aa3650eaSMark Johnston NULL); 1389aa3650eaSMark Johnston } else { 1390aa3650eaSMark Johnston next_page: 1391aa3650eaSMark Johnston tm = m; 1392aa3650eaSMark Johnston m = TAILQ_NEXT(m, listq); 1393c2655a40SMark Johnston } 1394c2655a40SMark Johnston 1395867a482dSJohn Dyson /* 13960012f373SJeff Roberson * If the page is not in a normal state, skip it. The page 13970012f373SJeff Roberson * can not be invalidated while the object lock is held. 1398867a482dSJohn Dyson */ 13990012f373SJeff Roberson if (!vm_page_all_valid(tm) || vm_page_wired(tm)) 1400aa3650eaSMark Johnston goto next_pindex; 1401aa3650eaSMark Johnston KASSERT((tm->flags & PG_FICTITIOUS) == 0, 1402aa3650eaSMark Johnston ("vm_object_madvise: page %p is fictitious", tm)); 1403aa3650eaSMark Johnston KASSERT((tm->oflags & VPO_UNMANAGED) == 0, 1404aa3650eaSMark Johnston ("vm_object_madvise: page %p is not managed", tm)); 140563e97555SJeff Roberson if (vm_page_tryxbusy(tm) == 0) { 1406aa3650eaSMark Johnston if (object != tobject) 1407aa3650eaSMark Johnston VM_OBJECT_WUNLOCK(object); 1408c2655a40SMark Johnston if (advice == MADV_WILLNEED) { 1409b11b56b5SAlan Cox /* 1410b11b56b5SAlan Cox * Reference the page before unlocking and 1411b11b56b5SAlan Cox * sleeping so that the page daemon is less 1412b11b56b5SAlan Cox * likely to reclaim it. 1413b11b56b5SAlan Cox */ 1414aa3650eaSMark Johnston vm_page_aflag_set(tm, PGA_REFERENCED); 1415567e51e1SAlan Cox } 141687b64663SMark Johnston if (!vm_page_busy_sleep(tm, "madvpo", 0)) 141787b64663SMark Johnston VM_OBJECT_WUNLOCK(tobject); 14186e20a165SJohn Dyson goto relookup; 141934567de7SAlan Cox } 1420aa3650eaSMark Johnston vm_page_advise(tm, advice); 142163e97555SJeff Roberson vm_page_xunbusy(tm); 1422aa3650eaSMark Johnston vm_object_madvise_freespace(tobject, advice, tm->pindex, 1); 1423aa3650eaSMark Johnston next_pindex: 14249b98b796SAlan Cox if (tobject != object) 142589f6b863SAttilio Rao VM_OBJECT_WUNLOCK(tobject); 1426867a482dSJohn Dyson } 142789f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1428867a482dSJohn Dyson } 1429867a482dSJohn Dyson 1430867a482dSJohn Dyson /* 1431df8bae1dSRodney W. Grimes * vm_object_shadow: 1432df8bae1dSRodney W. Grimes * 1433df8bae1dSRodney W. Grimes * Create a new object which is backed by the 1434df8bae1dSRodney W. Grimes * specified existing object range. The source 1435df8bae1dSRodney W. Grimes * object reference is deallocated. 1436df8bae1dSRodney W. Grimes * 1437df8bae1dSRodney W. Grimes * The new object and offset into that object 1438df8bae1dSRodney W. Grimes * are returned in the source parameters. 1439df8bae1dSRodney W. Grimes */ 144026f9a767SRodney W. Grimes void 144167388836SKonstantin Belousov vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length, 144267388836SKonstantin Belousov struct ucred *cred, bool shared) 1443df8bae1dSRodney W. Grimes { 1444d031cff1SMatthew Dillon vm_object_t source; 1445d031cff1SMatthew Dillon vm_object_t result; 1446df8bae1dSRodney W. Grimes 1447df8bae1dSRodney W. Grimes source = *object; 1448df8bae1dSRodney W. Grimes 1449df8bae1dSRodney W. Grimes /* 14509a2f6362SAlan Cox * Don't create the new object if the old object isn't shared. 145163967687SJeff Roberson * 145263967687SJeff Roberson * If we hold the only reference we can guarantee that it won't 145363967687SJeff Roberson * increase while we have the map locked. Otherwise the race is 145463967687SJeff Roberson * harmless and we will end up with an extra shadow object that 145563967687SJeff Roberson * will be collapsed later. 14569a2f6362SAlan Cox */ 145763967687SJeff Roberson if (source != NULL && source->ref_count == 1 && 145832362449SKonstantin Belousov (source->flags & OBJ_ANON) != 0) 14599a2f6362SAlan Cox return; 14609a2f6362SAlan Cox 14619a2f6362SAlan Cox /* 1462570a2f4aSAlan Cox * Allocate a new object with the given length. 1463df8bae1dSRodney W. Grimes */ 146467388836SKonstantin Belousov result = vm_object_allocate_anon(atop(length), source, cred, length); 1465df8bae1dSRodney W. Grimes 1466df8bae1dSRodney W. Grimes /* 146751b867e5SJeff Roberson * Store the offset into the source object, and fix up the offset into 146851b867e5SJeff Roberson * the new object. 146951b867e5SJeff Roberson */ 147051b867e5SJeff Roberson result->backing_object_offset = *offset; 147151b867e5SJeff Roberson 147267388836SKonstantin Belousov if (shared || source != NULL) { 147367388836SKonstantin Belousov VM_OBJECT_WLOCK(result); 147467388836SKonstantin Belousov 147551b867e5SJeff Roberson /* 147667388836SKonstantin Belousov * The new object shadows the source object, adding a 147767388836SKonstantin Belousov * reference to it. Our caller changes his reference 147867388836SKonstantin Belousov * to point to the new object, removing a reference to 147967388836SKonstantin Belousov * the source object. Net result: no change of 148067388836SKonstantin Belousov * reference count, unless the caller needs to add one 148167388836SKonstantin Belousov * more reference due to forking a shared map entry. 148267388836SKonstantin Belousov */ 148367388836SKonstantin Belousov if (shared) { 148467388836SKonstantin Belousov vm_object_reference_locked(result); 148567388836SKonstantin Belousov vm_object_clear_flag(result, OBJ_ONEMAPPING); 148667388836SKonstantin Belousov } 148767388836SKonstantin Belousov 148867388836SKonstantin Belousov /* 148967388836SKonstantin Belousov * Try to optimize the result object's page color when 149067388836SKonstantin Belousov * shadowing in order to maintain page coloring 149167388836SKonstantin Belousov * consistency in the combined shadowed object. 1492df8bae1dSRodney W. Grimes */ 1493570a2f4aSAlan Cox if (source != NULL) { 149451b867e5SJeff Roberson vm_object_backing_insert(result, source); 14953f289c3fSJeff Roberson result->domain = source->domain; 1496f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 1497630f633fSMark Johnston vm_object_set_flag(result, 1498630f633fSMark Johnston (source->flags & OBJ_COLORED)); 149967388836SKonstantin Belousov result->pg_color = (source->pg_color + 150067388836SKonstantin Belousov OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 150167388836SKonstantin Belousov 1)) - 1); 1502f8a47341SAlan Cox #endif 150367388836SKonstantin Belousov } 150451b867e5SJeff Roberson VM_OBJECT_WUNLOCK(result); 1505de5f6a77SJohn Dyson } 1506df8bae1dSRodney W. Grimes 1507df8bae1dSRodney W. Grimes /* 1508df8bae1dSRodney W. Grimes * Return the new things 1509df8bae1dSRodney W. Grimes */ 1510df8bae1dSRodney W. Grimes *offset = 0; 1511df8bae1dSRodney W. Grimes *object = result; 1512df8bae1dSRodney W. Grimes } 1513df8bae1dSRodney W. Grimes 1514c5aaa06dSAlan Cox /* 1515c5aaa06dSAlan Cox * vm_object_split: 1516c5aaa06dSAlan Cox * 1517c5aaa06dSAlan Cox * Split the pages in a map entry into a new object. This affords 1518c5aaa06dSAlan Cox * easier removal of unused pages, and keeps object inheritance from 1519c5aaa06dSAlan Cox * being a negative impact on memory usage. 1520c5aaa06dSAlan Cox */ 1521c5aaa06dSAlan Cox void 1522c5aaa06dSAlan Cox vm_object_split(vm_map_entry_t entry) 1523c5aaa06dSAlan Cox { 1524fff19e0eSMark Johnston vm_page_t m, m_next; 152598087a06SJeff Roberson vm_object_t orig_object, new_object, backing_object; 152673000556SAlan Cox vm_pindex_t idx, offidxstart; 152773000556SAlan Cox vm_size_t size; 1528c5aaa06dSAlan Cox 1529c5aaa06dSAlan Cox orig_object = entry->object.vm_object; 153098087a06SJeff Roberson KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0, 153198087a06SJeff Roberson ("vm_object_split: Splitting object with multiple mappings.")); 153263967687SJeff Roberson if ((orig_object->flags & OBJ_ANON) == 0) 1533c5aaa06dSAlan Cox return; 1534c5aaa06dSAlan Cox if (orig_object->ref_count <= 1) 1535c5aaa06dSAlan Cox return; 153689f6b863SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 1537c5aaa06dSAlan Cox 15384da9f125SAlan Cox offidxstart = OFF_TO_IDX(entry->offset); 153995442adfSAlan Cox size = atop(entry->end - entry->start); 1540c5aaa06dSAlan Cox 154167388836SKonstantin Belousov new_object = vm_object_allocate_anon(size, orig_object, 154267388836SKonstantin Belousov orig_object->cred, ptoa(size)); 1543c5aaa06dSAlan Cox 1544c5474b8fSAlan Cox /* 154598087a06SJeff Roberson * We must wait for the orig_object to complete any in-progress 154698087a06SJeff Roberson * collapse so that the swap blocks are stable below. The 154798087a06SJeff Roberson * additional reference on backing_object by new object will 154898087a06SJeff Roberson * prevent further collapse operations until split completes. 154998087a06SJeff Roberson */ 155098087a06SJeff Roberson VM_OBJECT_WLOCK(orig_object); 155198087a06SJeff Roberson vm_object_collapse_wait(orig_object); 155298087a06SJeff Roberson 155398087a06SJeff Roberson /* 1554c5474b8fSAlan Cox * At this point, the new object is still private, so the order in 1555c5474b8fSAlan Cox * which the original and new objects are locked does not matter. 1556c5474b8fSAlan Cox */ 155789f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 15583f289c3fSJeff Roberson new_object->domain = orig_object->domain; 155998087a06SJeff Roberson backing_object = orig_object->backing_object; 156098087a06SJeff Roberson if (backing_object != NULL) { 156198087a06SJeff Roberson vm_object_backing_insert_ref(new_object, backing_object); 1562c5aaa06dSAlan Cox new_object->backing_object_offset = 15634da9f125SAlan Cox orig_object->backing_object_offset + entry->offset; 1564c5aaa06dSAlan Cox } 1565ef694c1aSEdward Tomasz Napierala if (orig_object->cred != NULL) { 1566ef694c1aSEdward Tomasz Napierala crhold(orig_object->cred); 15673364c323SKonstantin Belousov KASSERT(orig_object->charge >= ptoa(size), 15683364c323SKonstantin Belousov ("orig_object->charge < 0")); 15693364c323SKonstantin Belousov orig_object->charge -= ptoa(size); 15703364c323SKonstantin Belousov } 157198087a06SJeff Roberson 157298087a06SJeff Roberson /* 157398087a06SJeff Roberson * Mark the split operation so that swap_pager_getpages() knows 157498087a06SJeff Roberson * that the object is in transition. 157598087a06SJeff Roberson */ 157698087a06SJeff Roberson vm_object_set_flag(orig_object, OBJ_SPLIT); 1577aec9e7d8SMark Johnston #ifdef INVARIANTS 1578aec9e7d8SMark Johnston idx = 0; 1579aec9e7d8SMark Johnston #endif 1580c5aaa06dSAlan Cox retry: 1581b382c10aSKonstantin Belousov m = vm_page_find_least(orig_object, offidxstart); 1582aec9e7d8SMark Johnston KASSERT(m == NULL || idx <= m->pindex - offidxstart, 1583aec9e7d8SMark Johnston ("%s: object %p was repopulated", __func__, orig_object)); 158473000556SAlan Cox for (; m != NULL && (idx = m->pindex - offidxstart) < size; 158573000556SAlan Cox m = m_next) { 158673000556SAlan Cox m_next = TAILQ_NEXT(m, listq); 1587c5aaa06dSAlan Cox 1588c5aaa06dSAlan Cox /* 1589c5aaa06dSAlan Cox * We must wait for pending I/O to complete before we can 1590c5aaa06dSAlan Cox * rename the page. 1591c5aaa06dSAlan Cox * 1592c5aaa06dSAlan Cox * We do not have to VM_PROT_NONE the page as mappings should 1593c5aaa06dSAlan Cox * not be changed by this operation. 1594c5aaa06dSAlan Cox */ 159563e97555SJeff Roberson if (vm_page_tryxbusy(m) == 0) { 159689f6b863SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 159787b64663SMark Johnston if (vm_page_busy_sleep(m, "spltwt", 0)) 159887b64663SMark Johnston VM_OBJECT_WLOCK(orig_object); 159989f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 1600c5aaa06dSAlan Cox goto retry; 1601de33beddSAlan Cox } 1602e946b949SAttilio Rao 16034bf95d00SJeff Roberson /* 16044bf95d00SJeff Roberson * The page was left invalid. Likely placed there by 16054bf95d00SJeff Roberson * an incomplete fault. Just remove and ignore. 16064bf95d00SJeff Roberson */ 16074bf95d00SJeff Roberson if (vm_page_none_valid(m)) { 16084bf95d00SJeff Roberson if (vm_page_remove(m)) 16094bf95d00SJeff Roberson vm_page_free(m); 16104bf95d00SJeff Roberson continue; 16114bf95d00SJeff Roberson } 16124bf95d00SJeff Roberson 16133453bca8SAlan Cox /* vm_page_rename() will dirty the page. */ 1614e946b949SAttilio Rao if (vm_page_rename(m, new_object, idx)) { 161563e97555SJeff Roberson vm_page_xunbusy(m); 1616e946b949SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 1617e946b949SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 16188d6fbbb8SJeff Roberson vm_radix_wait(); 1619e946b949SAttilio Rao VM_OBJECT_WLOCK(orig_object); 1620e946b949SAttilio Rao VM_OBJECT_WLOCK(new_object); 1621e946b949SAttilio Rao goto retry; 1622e946b949SAttilio Rao } 162363e97555SJeff Roberson 1624b5f359b7SAlan Cox #if VM_NRESERVLEVEL > 0 1625b5f359b7SAlan Cox /* 1626b5f359b7SAlan Cox * If some of the reservation's allocated pages remain with 1627b5f359b7SAlan Cox * the original object, then transferring the reservation to 1628b5f359b7SAlan Cox * the new object is neither particularly beneficial nor 1629b5f359b7SAlan Cox * particularly harmful as compared to leaving the reservation 1630b5f359b7SAlan Cox * with the original object. If, however, all of the 1631b5f359b7SAlan Cox * reservation's allocated pages are transferred to the new 1632b5f359b7SAlan Cox * object, then transferring the reservation is typically 1633b5f359b7SAlan Cox * beneficial. Determining which of these two cases applies 1634b5f359b7SAlan Cox * would be more costly than unconditionally renaming the 1635b5f359b7SAlan Cox * reservation. 1636b5f359b7SAlan Cox */ 1637b5f359b7SAlan Cox vm_reserv_rename(m, new_object, orig_object, offidxstart); 1638b5f359b7SAlan Cox #endif 1639c5aaa06dSAlan Cox } 1640fff19e0eSMark Johnston 1641c5aaa06dSAlan Cox /* 1642c7c8dd7eSAlan Cox * swap_pager_copy() can sleep, in which case the orig_object's 1643c7c8dd7eSAlan Cox * and new_object's locks are released and reacquired. 1644c5aaa06dSAlan Cox */ 1645c5aaa06dSAlan Cox swap_pager_copy(orig_object, new_object, offidxstart, 0); 1646fff19e0eSMark Johnston 1647fff19e0eSMark Johnston TAILQ_FOREACH(m, &new_object->memq, listq) 1648fff19e0eSMark Johnston vm_page_xunbusy(m); 1649fff19e0eSMark Johnston 165098087a06SJeff Roberson vm_object_clear_flag(orig_object, OBJ_SPLIT); 165189f6b863SAttilio Rao VM_OBJECT_WUNLOCK(orig_object); 165289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(new_object); 1653c5aaa06dSAlan Cox entry->object.vm_object = new_object; 1654c5aaa06dSAlan Cox entry->offset = 0LL; 1655c5aaa06dSAlan Cox vm_object_deallocate(orig_object); 165689f6b863SAttilio Rao VM_OBJECT_WLOCK(new_object); 1657c5aaa06dSAlan Cox } 1658c5aaa06dSAlan Cox 165999a1570aSKonstantin Belousov static vm_page_t 166098087a06SJeff Roberson vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p) 166199a1570aSKonstantin Belousov { 166299a1570aSKonstantin Belousov vm_object_t backing_object; 166399a1570aSKonstantin Belousov 166499a1570aSKonstantin Belousov VM_OBJECT_ASSERT_WLOCKED(object); 166599a1570aSKonstantin Belousov backing_object = object->backing_object; 166699a1570aSKonstantin Belousov VM_OBJECT_ASSERT_WLOCKED(backing_object); 166799a1570aSKonstantin Belousov 166899a1570aSKonstantin Belousov KASSERT(p == NULL || p->object == object || p->object == backing_object, 166999a1570aSKonstantin Belousov ("invalid ownership %p %p %p", p, object, backing_object)); 16708d6fbbb8SJeff Roberson /* The page is only NULL when rename fails. */ 16714cdea4a8SJeff Roberson if (p == NULL) { 16726a14746cSRyan Libby VM_OBJECT_WUNLOCK(object); 16736a14746cSRyan Libby VM_OBJECT_WUNLOCK(backing_object); 16748d6fbbb8SJeff Roberson vm_radix_wait(); 167599a1570aSKonstantin Belousov VM_OBJECT_WLOCK(object); 167687b64663SMark Johnston } else if (p->object == object) { 167787b64663SMark Johnston VM_OBJECT_WUNLOCK(backing_object); 167887b64663SMark Johnston if (vm_page_busy_sleep(p, "vmocol", 0)) 167987b64663SMark Johnston VM_OBJECT_WLOCK(object); 168087b64663SMark Johnston } else { 168187b64663SMark Johnston VM_OBJECT_WUNLOCK(object); 168287b64663SMark Johnston if (!vm_page_busy_sleep(p, "vmocol", 0)) 168387b64663SMark Johnston VM_OBJECT_WUNLOCK(backing_object); 168487b64663SMark Johnston VM_OBJECT_WLOCK(object); 168587b64663SMark Johnston } 168699a1570aSKonstantin Belousov VM_OBJECT_WLOCK(backing_object); 168799a1570aSKonstantin Belousov return (TAILQ_FIRST(&backing_object->memq)); 168899a1570aSKonstantin Belousov } 168999a1570aSKonstantin Belousov 169099a1570aSKonstantin Belousov static bool 16914cc8daf7SConrad Meyer vm_object_scan_all_shadowed(vm_object_t object) 16924cc8daf7SConrad Meyer { 16934cc8daf7SConrad Meyer vm_object_t backing_object; 16944cc8daf7SConrad Meyer vm_page_t p, pp; 169577d6fd97SKonstantin Belousov vm_pindex_t backing_offset_index, new_pindex, pi, ps; 16964cc8daf7SConrad Meyer 16974cc8daf7SConrad Meyer VM_OBJECT_ASSERT_WLOCKED(object); 16984cc8daf7SConrad Meyer VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 16994cc8daf7SConrad Meyer 17004cc8daf7SConrad Meyer backing_object = object->backing_object; 17014cc8daf7SConrad Meyer 170263967687SJeff Roberson if ((backing_object->flags & OBJ_ANON) == 0) 17034cc8daf7SConrad Meyer return (false); 17044cc8daf7SConrad Meyer 170577d6fd97SKonstantin Belousov pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 170677d6fd97SKonstantin Belousov p = vm_page_find_least(backing_object, pi); 170777d6fd97SKonstantin Belousov ps = swap_pager_find_least(backing_object, pi); 17084cc8daf7SConrad Meyer 17094cc8daf7SConrad Meyer /* 171077d6fd97SKonstantin Belousov * Only check pages inside the parent object's range and 171177d6fd97SKonstantin Belousov * inside the parent object's mapping of the backing object. 17124cc8daf7SConrad Meyer */ 171377d6fd97SKonstantin Belousov for (;; pi++) { 171477d6fd97SKonstantin Belousov if (p != NULL && p->pindex < pi) 171577d6fd97SKonstantin Belousov p = TAILQ_NEXT(p, listq); 171677d6fd97SKonstantin Belousov if (ps < pi) 171777d6fd97SKonstantin Belousov ps = swap_pager_find_least(backing_object, pi); 171877d6fd97SKonstantin Belousov if (p == NULL && ps >= backing_object->size) 171977d6fd97SKonstantin Belousov break; 172077d6fd97SKonstantin Belousov else if (p == NULL) 172177d6fd97SKonstantin Belousov pi = ps; 172277d6fd97SKonstantin Belousov else 172377d6fd97SKonstantin Belousov pi = MIN(p->pindex, ps); 172477d6fd97SKonstantin Belousov 172577d6fd97SKonstantin Belousov new_pindex = pi - backing_offset_index; 172677d6fd97SKonstantin Belousov if (new_pindex >= object->size) 172777d6fd97SKonstantin Belousov break; 17284cc8daf7SConrad Meyer 1729cd0047f3SKonstantin Belousov if (p != NULL) { 17304cc8daf7SConrad Meyer /* 1731cd0047f3SKonstantin Belousov * If the backing object page is busy a 1732cd0047f3SKonstantin Belousov * grandparent or older page may still be 1733cd0047f3SKonstantin Belousov * undergoing CoW. It is not safe to collapse 1734cd0047f3SKonstantin Belousov * the backing object until it is quiesced. 173558447749SJeff Roberson */ 1736cd0047f3SKonstantin Belousov if (vm_page_tryxbusy(p) == 0) 173758447749SJeff Roberson return (false); 173858447749SJeff Roberson 173958447749SJeff Roberson /* 1740cd0047f3SKonstantin Belousov * We raced with the fault handler that left 1741cd0047f3SKonstantin Belousov * newly allocated invalid page on the object 1742cd0047f3SKonstantin Belousov * queue and retried. 1743cd0047f3SKonstantin Belousov */ 1744cd0047f3SKonstantin Belousov if (!vm_page_all_valid(p)) 1745cd0047f3SKonstantin Belousov goto unbusy_ret; 1746cd0047f3SKonstantin Belousov } 1747cd0047f3SKonstantin Belousov 1748cd0047f3SKonstantin Belousov /* 17494cc8daf7SConrad Meyer * See if the parent has the page or if the parent's object 17504cc8daf7SConrad Meyer * pager has the page. If the parent has the page but the page 17514cc8daf7SConrad Meyer * is not valid, the parent's object pager must have the page. 17524cc8daf7SConrad Meyer * 17534cc8daf7SConrad Meyer * If this fails, the parent does not completely shadow the 17544cc8daf7SConrad Meyer * object and we might as well give up now. 17554cc8daf7SConrad Meyer */ 17564cc8daf7SConrad Meyer pp = vm_page_lookup(object, new_pindex); 1757cd0047f3SKonstantin Belousov 17580012f373SJeff Roberson /* 1759cd0047f3SKonstantin Belousov * The valid check here is stable due to object lock 1760cd0047f3SKonstantin Belousov * being required to clear valid and initiate paging. 1761cd0047f3SKonstantin Belousov * Busy of p disallows fault handler to validate pp. 17620012f373SJeff Roberson */ 17630012f373SJeff Roberson if ((pp == NULL || vm_page_none_valid(pp)) && 17644cc8daf7SConrad Meyer !vm_pager_has_page(object, new_pindex, NULL, NULL)) 1765cd0047f3SKonstantin Belousov goto unbusy_ret; 1766cd0047f3SKonstantin Belousov if (p != NULL) 1767cd0047f3SKonstantin Belousov vm_page_xunbusy(p); 17684cc8daf7SConrad Meyer } 17694cc8daf7SConrad Meyer return (true); 1770cd0047f3SKonstantin Belousov 1771cd0047f3SKonstantin Belousov unbusy_ret: 1772cd0047f3SKonstantin Belousov if (p != NULL) 1773cd0047f3SKonstantin Belousov vm_page_xunbusy(p); 1774cd0047f3SKonstantin Belousov return (false); 17754cc8daf7SConrad Meyer } 17764cc8daf7SConrad Meyer 177798087a06SJeff Roberson static void 177898087a06SJeff Roberson vm_object_collapse_scan(vm_object_t object) 17792ad1a3f7SMatthew Dillon { 17802ad1a3f7SMatthew Dillon vm_object_t backing_object; 178199a1570aSKonstantin Belousov vm_page_t next, p, pp; 178299a1570aSKonstantin Belousov vm_pindex_t backing_offset_index, new_pindex; 17832ad1a3f7SMatthew Dillon 178489f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 178589f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object->backing_object); 17862ad1a3f7SMatthew Dillon 17872ad1a3f7SMatthew Dillon backing_object = object->backing_object; 17882ad1a3f7SMatthew Dillon backing_offset_index = OFF_TO_IDX(object->backing_object_offset); 17892ad1a3f7SMatthew Dillon 17902ad1a3f7SMatthew Dillon /* 17912ad1a3f7SMatthew Dillon * Our scan 17922ad1a3f7SMatthew Dillon */ 17934cc8daf7SConrad Meyer for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) { 179499a1570aSKonstantin Belousov next = TAILQ_NEXT(p, listq); 179599a1570aSKonstantin Belousov new_pindex = p->pindex - backing_offset_index; 17962ad1a3f7SMatthew Dillon 17972ad1a3f7SMatthew Dillon /* 17982ad1a3f7SMatthew Dillon * Check for busy page 17992ad1a3f7SMatthew Dillon */ 180063e97555SJeff Roberson if (vm_page_tryxbusy(p) == 0) { 180198087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, p); 18022ad1a3f7SMatthew Dillon continue; 18032ad1a3f7SMatthew Dillon } 18042ad1a3f7SMatthew Dillon 180598087a06SJeff Roberson KASSERT(object->backing_object == backing_object, 180698087a06SJeff Roberson ("vm_object_collapse_scan: backing object mismatch %p != %p", 180798087a06SJeff Roberson object->backing_object, backing_object)); 180899a1570aSKonstantin Belousov KASSERT(p->object == backing_object, 180998087a06SJeff Roberson ("vm_object_collapse_scan: object mismatch %p != %p", 181098087a06SJeff Roberson p->object, backing_object)); 18112ad1a3f7SMatthew Dillon 181299a1570aSKonstantin Belousov if (p->pindex < backing_offset_index || 181399a1570aSKonstantin Belousov new_pindex >= object->size) { 18141390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, p->pindex, 1); 1815e946b949SAttilio Rao 1816f6d89838SAlan Cox KASSERT(!pmap_page_is_mapped(p), 1817f6d89838SAlan Cox ("freeing mapped page %p", p)); 18180fd977b3SMark Johnston if (vm_page_remove(p)) 18192ad1a3f7SMatthew Dillon vm_page_free(p); 18202ad1a3f7SMatthew Dillon continue; 18212ad1a3f7SMatthew Dillon } 18222ad1a3f7SMatthew Dillon 1823cd0047f3SKonstantin Belousov if (!vm_page_all_valid(p)) { 1824cd0047f3SKonstantin Belousov KASSERT(!pmap_page_is_mapped(p), 1825cd0047f3SKonstantin Belousov ("freeing mapped page %p", p)); 1826cd0047f3SKonstantin Belousov if (vm_page_remove(p)) 1827cd0047f3SKonstantin Belousov vm_page_free(p); 1828cd0047f3SKonstantin Belousov continue; 1829cd0047f3SKonstantin Belousov } 1830cd0047f3SKonstantin Belousov 18312ad1a3f7SMatthew Dillon pp = vm_page_lookup(object, new_pindex); 183263e97555SJeff Roberson if (pp != NULL && vm_page_tryxbusy(pp) == 0) { 183363e97555SJeff Roberson vm_page_xunbusy(p); 1834e18cc7bfSMax Laier /* 18354cc8daf7SConrad Meyer * The page in the parent is busy and possibly not 18364cc8daf7SConrad Meyer * (yet) valid. Until its state is finalized by the 18374cc8daf7SConrad Meyer * busy bit owner, we can't tell whether it shadows the 183898087a06SJeff Roberson * original page. 1839e18cc7bfSMax Laier */ 184098087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, pp); 1841e18cc7bfSMax Laier continue; 1842e18cc7bfSMax Laier } 184399a1570aSKonstantin Belousov 18444bf95d00SJeff Roberson if (pp != NULL && vm_page_none_valid(pp)) { 18454bf95d00SJeff Roberson /* 18464bf95d00SJeff Roberson * The page was invalid in the parent. Likely placed 18474bf95d00SJeff Roberson * there by an incomplete fault. Just remove and 18484bf95d00SJeff Roberson * ignore. p can replace it. 18494bf95d00SJeff Roberson */ 18504bf95d00SJeff Roberson if (vm_page_remove(pp)) 18514bf95d00SJeff Roberson vm_page_free(pp); 18524bf95d00SJeff Roberson pp = NULL; 18534bf95d00SJeff Roberson } 185499a1570aSKonstantin Belousov 18554cc8daf7SConrad Meyer if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL, 18564cc8daf7SConrad Meyer NULL)) { 185799a1570aSKonstantin Belousov /* 18584cc8daf7SConrad Meyer * The page already exists in the parent OR swap exists 18594cc8daf7SConrad Meyer * for this location in the parent. Leave the parent's 18604cc8daf7SConrad Meyer * page alone. Destroy the original page from the 18614cc8daf7SConrad Meyer * backing object. 186299a1570aSKonstantin Belousov */ 18631390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, p->pindex, 1); 1864f6d89838SAlan Cox KASSERT(!pmap_page_is_mapped(p), 1865f6d89838SAlan Cox ("freeing mapped page %p", p)); 18660fd977b3SMark Johnston if (vm_page_remove(p)) 18672ad1a3f7SMatthew Dillon vm_page_free(p); 186863e97555SJeff Roberson if (pp != NULL) 186963e97555SJeff Roberson vm_page_xunbusy(pp); 18702ad1a3f7SMatthew Dillon continue; 18712ad1a3f7SMatthew Dillon } 18722ad1a3f7SMatthew Dillon 1873e946b949SAttilio Rao /* 18744cc8daf7SConrad Meyer * Page does not exist in parent, rename the page from the 18754cc8daf7SConrad Meyer * backing object to the main object. 1876e946b949SAttilio Rao * 18774cc8daf7SConrad Meyer * If the page was mapped to a process, it can remain mapped 18783453bca8SAlan Cox * through the rename. vm_page_rename() will dirty the page. 1879e946b949SAttilio Rao */ 1880e946b949SAttilio Rao if (vm_page_rename(p, object, new_pindex)) { 188163e97555SJeff Roberson vm_page_xunbusy(p); 188298087a06SJeff Roberson next = vm_object_collapse_scan_wait(object, NULL); 1883e946b949SAttilio Rao continue; 1884e946b949SAttilio Rao } 188514a5dc17SAttilio Rao 188614a5dc17SAttilio Rao /* Use the old pindex to free the right page. */ 18871390a5cbSKonstantin Belousov vm_pager_freespace(backing_object, new_pindex + 18881390a5cbSKonstantin Belousov backing_offset_index, 1); 1889e946b949SAttilio Rao 1890f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 1891f8a47341SAlan Cox /* 1892f8a47341SAlan Cox * Rename the reservation. 1893f8a47341SAlan Cox */ 1894f8a47341SAlan Cox vm_reserv_rename(p, object, backing_object, 1895f8a47341SAlan Cox backing_offset_index); 1896f8a47341SAlan Cox #endif 18978da1c098SJeff Roberson vm_page_xunbusy(p); 18982ad1a3f7SMatthew Dillon } 18992fe6e4d7SDavid Greenman return; 19002fe6e4d7SDavid Greenman } 19012fe6e4d7SDavid Greenman 1902df8bae1dSRodney W. Grimes /* 1903df8bae1dSRodney W. Grimes * vm_object_collapse: 1904df8bae1dSRodney W. Grimes * 1905df8bae1dSRodney W. Grimes * Collapse an object with the object backing it. 1906df8bae1dSRodney W. Grimes * Pages in the backing object are moved into the 1907df8bae1dSRodney W. Grimes * parent, and the backing object is deallocated. 1908df8bae1dSRodney W. Grimes */ 190926f9a767SRodney W. Grimes void 19101b40f8c0SMatthew Dillon vm_object_collapse(vm_object_t object) 1911df8bae1dSRodney W. Grimes { 191298f139daSKonstantin Belousov vm_object_t backing_object, new_backing_object; 191398f139daSKonstantin Belousov 191489f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 191523955314SAlfred Perlstein 1916df8bae1dSRodney W. Grimes while (TRUE) { 191798087a06SJeff Roberson KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON, 191898087a06SJeff Roberson ("collapsing invalid object")); 191998087a06SJeff Roberson 1920df8bae1dSRodney W. Grimes /* 192198087a06SJeff Roberson * Wait for the backing_object to finish any pending 192298087a06SJeff Roberson * collapse so that the caller sees the shortest possible 192398087a06SJeff Roberson * shadow chain. 1924df8bae1dSRodney W. Grimes */ 192598087a06SJeff Roberson backing_object = vm_object_backing_collapse_wait(object); 192698087a06SJeff Roberson if (backing_object == NULL) 192798087a06SJeff Roberson return; 192898087a06SJeff Roberson 192998087a06SJeff Roberson KASSERT(object->ref_count > 0 && 19304cf9f5d8SKonstantin Belousov object->ref_count > atomic_load_int(&object->shadow_count), 193198087a06SJeff Roberson ("collapse with invalid ref %d or shadow %d count.", 19324cf9f5d8SKonstantin Belousov object->ref_count, atomic_load_int(&object->shadow_count))); 193398087a06SJeff Roberson KASSERT((backing_object->flags & 193498087a06SJeff Roberson (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 193598087a06SJeff Roberson ("vm_object_collapse: Backing object already collapsing.")); 193698087a06SJeff Roberson KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0, 193798087a06SJeff Roberson ("vm_object_collapse: object is already collapsing.")); 1938df8bae1dSRodney W. Grimes 1939f919ebdeSDavid Greenman /* 194098087a06SJeff Roberson * We know that we can either collapse the backing object if 194198087a06SJeff Roberson * the parent is the only reference to it, or (perhaps) have 19422ad1a3f7SMatthew Dillon * the parent bypass the object if the parent happens to shadow 19432ad1a3f7SMatthew Dillon * all the resident pages in the entire backing object. 1944df8bae1dSRodney W. Grimes */ 1945df8bae1dSRodney W. Grimes if (backing_object->ref_count == 1) { 19464cf9f5d8SKonstantin Belousov KASSERT(atomic_load_int(&backing_object->shadow_count) 19474cf9f5d8SKonstantin Belousov == 1, 194898087a06SJeff Roberson ("vm_object_collapse: shadow_count: %d", 19494cf9f5d8SKonstantin Belousov atomic_load_int(&backing_object->shadow_count))); 1950aa9bc3b1SKonstantin Belousov vm_object_pip_add(object, 1); 195198087a06SJeff Roberson vm_object_set_flag(object, OBJ_COLLAPSING); 1952aa9bc3b1SKonstantin Belousov vm_object_pip_add(backing_object, 1); 195398087a06SJeff Roberson vm_object_set_flag(backing_object, OBJ_DEAD); 1954aa9bc3b1SKonstantin Belousov 1955df8bae1dSRodney W. Grimes /* 19562ad1a3f7SMatthew Dillon * If there is exactly one reference to the backing 19572ad1a3f7SMatthew Dillon * object, we can collapse it into the parent. 1958df8bae1dSRodney W. Grimes */ 195998087a06SJeff Roberson vm_object_collapse_scan(object); 1960df8bae1dSRodney W. Grimes 1961f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 1962f8a47341SAlan Cox /* 1963f8a47341SAlan Cox * Break any reservations from backing_object. 1964f8a47341SAlan Cox */ 1965f8a47341SAlan Cox if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) 1966f8a47341SAlan Cox vm_reserv_break_all(backing_object); 1967f8a47341SAlan Cox #endif 1968f8a47341SAlan Cox 1969df8bae1dSRodney W. Grimes /* 1970df8bae1dSRodney W. Grimes * Move the pager from backing_object to object. 1971fff19e0eSMark Johnston * 1972fff19e0eSMark Johnston * swap_pager_copy() can sleep, in which case the 1973fff19e0eSMark Johnston * backing_object's and object's locks are released and 1974fff19e0eSMark Johnston * reacquired. 1975df8bae1dSRodney W. Grimes */ 1976fff19e0eSMark Johnston swap_pager_copy(backing_object, object, 19771c7c3c6aSMatthew Dillon OFF_TO_IDX(object->backing_object_offset), TRUE); 197898087a06SJeff Roberson 1979df8bae1dSRodney W. Grimes /* 1980df8bae1dSRodney W. Grimes * Object now shadows whatever backing_object did. 1981df8bae1dSRodney W. Grimes */ 198298087a06SJeff Roberson vm_object_clear_flag(object, OBJ_COLLAPSING); 198398087a06SJeff Roberson vm_object_backing_transfer(object, backing_object); 19842ad1a3f7SMatthew Dillon object->backing_object_offset += 19852ad1a3f7SMatthew Dillon backing_object->backing_object_offset; 198698087a06SJeff Roberson VM_OBJECT_WUNLOCK(object); 198798087a06SJeff Roberson vm_object_pip_wakeup(object); 19882ad1a3f7SMatthew Dillon 1989df8bae1dSRodney W. Grimes /* 1990df8bae1dSRodney W. Grimes * Discard backing_object. 1991df8bae1dSRodney W. Grimes * 19920d94caffSDavid Greenman * Since the backing object has no pages, no pager left, 19930d94caffSDavid Greenman * and no object references within it, all that is 19940d94caffSDavid Greenman * necessary is to dispose of it. 1995df8bae1dSRodney W. Grimes */ 19969b4d473aSKonstantin Belousov KASSERT(backing_object->ref_count == 1, ( 19979b4d473aSKonstantin Belousov "backing_object %p was somehow re-referenced during collapse!", 19989b4d473aSKonstantin Belousov backing_object)); 1999aa9bc3b1SKonstantin Belousov vm_object_pip_wakeup(backing_object); 200098087a06SJeff Roberson (void)refcount_release(&backing_object->ref_count); 200198087a06SJeff Roberson vm_object_terminate(backing_object); 200211542376SAlan Cox counter_u64_add(object_collapses, 1); 200398087a06SJeff Roberson VM_OBJECT_WLOCK(object); 20040d94caffSDavid Greenman } else { 2005df8bae1dSRodney W. Grimes /* 20062ad1a3f7SMatthew Dillon * If we do not entirely shadow the backing object, 20072ad1a3f7SMatthew Dillon * there is nothing we can do so we give up. 200898087a06SJeff Roberson * 200998087a06SJeff Roberson * The object lock and backing_object lock must not 201098087a06SJeff Roberson * be dropped during this sequence. 2011df8bae1dSRodney W. Grimes */ 201258447749SJeff Roberson if (!vm_object_scan_all_shadowed(object)) { 201389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(backing_object); 20142ad1a3f7SMatthew Dillon break; 201524a1cce3SDavid Greenman } 2016df8bae1dSRodney W. Grimes 2017df8bae1dSRodney W. Grimes /* 20180d94caffSDavid Greenman * Make the parent shadow the next object in the 20190d94caffSDavid Greenman * chain. Deallocating backing_object will not remove 20200d94caffSDavid Greenman * it, since its reference count is at least 2. 2021df8bae1dSRodney W. Grimes */ 202251b867e5SJeff Roberson vm_object_backing_remove_locked(object); 202395e5e988SJohn Dyson new_backing_object = backing_object->backing_object; 202451b867e5SJeff Roberson if (new_backing_object != NULL) { 202598087a06SJeff Roberson vm_object_backing_insert_ref(object, 202651b867e5SJeff Roberson new_backing_object); 202795e5e988SJohn Dyson object->backing_object_offset += 202895e5e988SJohn Dyson backing_object->backing_object_offset; 2029de5f6a77SJohn Dyson } 2030df8bae1dSRodney W. Grimes 2031df8bae1dSRodney W. Grimes /* 20320d94caffSDavid Greenman * Drop the reference count on backing_object. Since 203322ec553fSAlan Cox * its ref_count was at least 2, it will not vanish. 2034df8bae1dSRodney W. Grimes */ 203598087a06SJeff Roberson (void)refcount_release(&backing_object->ref_count); 203698087a06SJeff Roberson KASSERT(backing_object->ref_count >= 1, ( 203798087a06SJeff Roberson "backing_object %p was somehow dereferenced during collapse!", 203898087a06SJeff Roberson backing_object)); 203989f6b863SAttilio Rao VM_OBJECT_WUNLOCK(backing_object); 204011542376SAlan Cox counter_u64_add(object_bypasses, 1); 2041df8bae1dSRodney W. Grimes } 2042df8bae1dSRodney W. Grimes 2043df8bae1dSRodney W. Grimes /* 2044df8bae1dSRodney W. Grimes * Try again with this object's new backing object. 2045df8bae1dSRodney W. Grimes */ 2046df8bae1dSRodney W. Grimes } 2047df8bae1dSRodney W. Grimes } 2048df8bae1dSRodney W. Grimes 2049df8bae1dSRodney W. Grimes /* 2050bff99f0dSAlan Cox * vm_object_page_remove: 2051df8bae1dSRodney W. Grimes * 205268855966SAlan Cox * For the given object, either frees or invalidates each of the 20536bbee8e2SAlan Cox * specified pages. In general, a page is freed. However, if a page is 20546bbee8e2SAlan Cox * wired for any reason other than the existence of a managed, wired 20556bbee8e2SAlan Cox * mapping, then it may be invalidated but not removed from the object. 20566bbee8e2SAlan Cox * Pages are specified by the given range ["start", "end") and the option 20576bbee8e2SAlan Cox * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range 20586bbee8e2SAlan Cox * extends from "start" to the end of the object. If the option 20596bbee8e2SAlan Cox * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the 20606bbee8e2SAlan Cox * specified range are affected. If the option OBJPR_NOTMAPPED is 20616bbee8e2SAlan Cox * specified, then the pages within the specified range must have no 20626bbee8e2SAlan Cox * mappings. Otherwise, if this option is not specified, any mappings to 20636bbee8e2SAlan Cox * the specified pages are removed before the pages are freed or 20646bbee8e2SAlan Cox * invalidated. 206568855966SAlan Cox * 20666bbee8e2SAlan Cox * In general, this operation should only be performed on objects that 20676bbee8e2SAlan Cox * contain managed pages. There are, however, two exceptions. First, it 20686bbee8e2SAlan Cox * is performed on the kernel and kmem objects by vm_map_entry_delete(). 20696bbee8e2SAlan Cox * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- 20706bbee8e2SAlan Cox * backed pages. In both of these cases, the option OBJPR_CLEANONLY must 20716bbee8e2SAlan Cox * not be specified and the option OBJPR_NOTMAPPED must be specified. 2072df8bae1dSRodney W. Grimes * 2073df8bae1dSRodney W. Grimes * The object must be locked. 2074df8bae1dSRodney W. Grimes */ 207526f9a767SRodney W. Grimes void 2076ecde4b32SAlan Cox vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, 20776bbee8e2SAlan Cox int options) 2078df8bae1dSRodney W. Grimes { 2079d031cff1SMatthew Dillon vm_page_t p, next; 2080df8bae1dSRodney W. Grimes 208189f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 208228634820SAlan Cox KASSERT((object->flags & OBJ_UNMANAGED) == 0 || 20836bbee8e2SAlan Cox (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, 20846bbee8e2SAlan Cox ("vm_object_page_remove: illegal options for object %p", object)); 2085ecde4b32SAlan Cox if (object->resident_page_count == 0) 20867667839aSAlan Cox return; 2087d474eaaaSDoug Rabson vm_object_pip_add(object, 1); 208826f9a767SRodney W. Grimes again: 2089b382c10aSKonstantin Belousov p = vm_page_find_least(object, start); 20902965a453SKip Macy 209175741c04SAlan Cox /* 20926bbee8e2SAlan Cox * Here, the variable "p" is either (1) the page with the least pindex 20936bbee8e2SAlan Cox * greater than or equal to the parameter "start" or (2) NULL. 209475741c04SAlan Cox */ 20956bbee8e2SAlan Cox for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2096b18bfc3dSJohn Dyson next = TAILQ_NEXT(p, listq); 209775741c04SAlan Cox 209859677d3cSAlan Cox /* 2099d28af1abSMark Johnston * Skip invalid pages if asked to do so. Try to avoid acquiring 2100d28af1abSMark Johnston * the busy lock, as some consumers rely on this to avoid 2101d28af1abSMark Johnston * deadlocks. 2102d28af1abSMark Johnston * 2103d28af1abSMark Johnston * A thread may concurrently transition the page from invalid to 2104d28af1abSMark Johnston * valid using only the busy lock, so the result of this check 2105d28af1abSMark Johnston * is immediately stale. It is up to consumers to handle this, 2106d28af1abSMark Johnston * for instance by ensuring that all invalid->valid transitions 2107d28af1abSMark Johnston * happen with a mutex held, as may be possible for a 2108d28af1abSMark Johnston * filesystem. 2109d28af1abSMark Johnston */ 2110d28af1abSMark Johnston if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) 2111d28af1abSMark Johnston continue; 2112d28af1abSMark Johnston 2113d28af1abSMark Johnston /* 21146bbee8e2SAlan Cox * If the page is wired for any reason besides the existence 21156bbee8e2SAlan Cox * of managed, wired mappings, then it cannot be freed. For 21166bbee8e2SAlan Cox * example, fictitious pages, which represent device memory, 21176bbee8e2SAlan Cox * are inherently wired and cannot be freed. They can, 21186bbee8e2SAlan Cox * however, be invalidated if the option OBJPR_CLEANONLY is 21196bbee8e2SAlan Cox * not specified. 212059677d3cSAlan Cox */ 212163e97555SJeff Roberson if (vm_page_tryxbusy(p) == 0) { 212287b64663SMark Johnston if (vm_page_busy_sleep(p, "vmopar", 0)) 212387b64663SMark Johnston VM_OBJECT_WLOCK(object); 2124fee2a2faSMark Johnston goto again; 2125fee2a2faSMark Johnston } 2126d28af1abSMark Johnston if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) { 2127d28af1abSMark Johnston vm_page_xunbusy(p); 2128d28af1abSMark Johnston continue; 2129d28af1abSMark Johnston } 2130d842aa51SMark Johnston if (vm_page_wired(p)) { 2131fee2a2faSMark Johnston wired: 2132cf060942SAlan Cox if ((options & OBJPR_NOTMAPPED) == 0 && 2133cf060942SAlan Cox object->ref_count != 0) 21344fec79beSAlan Cox pmap_remove_all(p); 21356bbee8e2SAlan Cox if ((options & OBJPR_CLEANONLY) == 0) { 21360012f373SJeff Roberson vm_page_invalid(p); 2137a28042d1SAlan Cox vm_page_undirty(p); 2138a28042d1SAlan Cox } 213963e97555SJeff Roberson vm_page_xunbusy(p); 214093c5d3a4SKonstantin Belousov continue; 21410d94caffSDavid Greenman } 214268855966SAlan Cox KASSERT((p->flags & PG_FICTITIOUS) == 0, 214368855966SAlan Cox ("vm_object_page_remove: page %p is fictitious", p)); 21440012f373SJeff Roberson if ((options & OBJPR_CLEANONLY) != 0 && 21450012f373SJeff Roberson !vm_page_none_valid(p)) { 2146cf060942SAlan Cox if ((options & OBJPR_NOTMAPPED) == 0 && 2147fee2a2faSMark Johnston object->ref_count != 0 && 2148fee2a2faSMark Johnston !vm_page_try_remove_write(p)) 2149fee2a2faSMark Johnston goto wired; 215063e97555SJeff Roberson if (p->dirty != 0) { 215163e97555SJeff Roberson vm_page_xunbusy(p); 215293c5d3a4SKonstantin Belousov continue; 21532965a453SKip Macy } 215463e97555SJeff Roberson } 2155fee2a2faSMark Johnston if ((options & OBJPR_NOTMAPPED) == 0 && 2156fee2a2faSMark Johnston object->ref_count != 0 && !vm_page_try_remove_all(p)) 2157fee2a2faSMark Johnston goto wired; 21585cd29d0fSMark Johnston vm_page_free(p); 21592965a453SKip Macy } 2160f919ebdeSDavid Greenman vm_object_pip_wakeup(object); 216184242cf6SMark Johnston 21621390a5cbSKonstantin Belousov vm_pager_freespace(object, start, (end == 0 ? object->size : end) - 21631390a5cbSKonstantin Belousov start); 2164c0503609SDavid Greenman } 2165df8bae1dSRodney W. Grimes 2166df8bae1dSRodney W. Grimes /* 21673138cd36SMark Johnston * vm_object_page_noreuse: 2168936c09acSJohn Baldwin * 21693138cd36SMark Johnston * For the given object, attempt to move the specified pages to 21703138cd36SMark Johnston * the head of the inactive queue. This bypasses regular LRU 21713138cd36SMark Johnston * operation and allows the pages to be reused quickly under memory 21723138cd36SMark Johnston * pressure. If a page is wired for any reason, then it will not 21733138cd36SMark Johnston * be queued. Pages are specified by the range ["start", "end"). 21743138cd36SMark Johnston * As a special case, if "end" is zero, then the range extends from 21753138cd36SMark Johnston * "start" to the end of the object. 2176936c09acSJohn Baldwin * 2177936c09acSJohn Baldwin * This operation should only be performed on objects that 217828634820SAlan Cox * contain non-fictitious, managed pages. 2179936c09acSJohn Baldwin * 2180936c09acSJohn Baldwin * The object must be locked. 2181936c09acSJohn Baldwin */ 2182936c09acSJohn Baldwin void 21833138cd36SMark Johnston vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2184936c09acSJohn Baldwin { 2185936c09acSJohn Baldwin vm_page_t p, next; 2186936c09acSJohn Baldwin 218752d1addaSAlan Cox VM_OBJECT_ASSERT_LOCKED(object); 218828634820SAlan Cox KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, 21893138cd36SMark Johnston ("vm_object_page_noreuse: illegal object %p", object)); 2190936c09acSJohn Baldwin if (object->resident_page_count == 0) 2191936c09acSJohn Baldwin return; 2192936c09acSJohn Baldwin p = vm_page_find_least(object, start); 2193936c09acSJohn Baldwin 2194936c09acSJohn Baldwin /* 2195936c09acSJohn Baldwin * Here, the variable "p" is either (1) the page with the least pindex 2196936c09acSJohn Baldwin * greater than or equal to the parameter "start" or (2) NULL. 2197936c09acSJohn Baldwin */ 2198936c09acSJohn Baldwin for (; p != NULL && (p->pindex < end || end == 0); p = next) { 2199936c09acSJohn Baldwin next = TAILQ_NEXT(p, listq); 22003138cd36SMark Johnston vm_page_deactivate_noreuse(p); 2201936c09acSJohn Baldwin } 2202936c09acSJohn Baldwin } 2203936c09acSJohn Baldwin 2204936c09acSJohn Baldwin /* 2205387aabc5SAlan Cox * Populate the specified range of the object with valid pages. Returns 2206387aabc5SAlan Cox * TRUE if the range is successfully populated and FALSE otherwise. 2207387aabc5SAlan Cox * 2208387aabc5SAlan Cox * Note: This function should be optimized to pass a larger array of 2209387aabc5SAlan Cox * pages to vm_pager_get_pages() before it is applied to a non- 2210387aabc5SAlan Cox * OBJT_DEVICE object. 2211387aabc5SAlan Cox * 2212387aabc5SAlan Cox * The object must be locked. 2213387aabc5SAlan Cox */ 2214387aabc5SAlan Cox boolean_t 2215387aabc5SAlan Cox vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) 2216387aabc5SAlan Cox { 2217093c7f39SGleb Smirnoff vm_page_t m; 2218387aabc5SAlan Cox vm_pindex_t pindex; 2219387aabc5SAlan Cox int rv; 2220387aabc5SAlan Cox 222189f6b863SAttilio Rao VM_OBJECT_ASSERT_WLOCKED(object); 2222387aabc5SAlan Cox for (pindex = start; pindex < end; pindex++) { 2223c7575748SJeff Roberson rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL); 2224c7575748SJeff Roberson if (rv != VM_PAGER_OK) 2225387aabc5SAlan Cox break; 2226c7575748SJeff Roberson 2227387aabc5SAlan Cox /* 2228387aabc5SAlan Cox * Keep "m" busy because a subsequent iteration may unlock 2229387aabc5SAlan Cox * the object. 2230387aabc5SAlan Cox */ 2231387aabc5SAlan Cox } 2232387aabc5SAlan Cox if (pindex > start) { 2233387aabc5SAlan Cox m = vm_page_lookup(object, start); 2234387aabc5SAlan Cox while (m != NULL && m->pindex < pindex) { 2235c7aebda8SAttilio Rao vm_page_xunbusy(m); 2236387aabc5SAlan Cox m = TAILQ_NEXT(m, listq); 2237387aabc5SAlan Cox } 2238387aabc5SAlan Cox } 2239387aabc5SAlan Cox return (pindex == end); 2240387aabc5SAlan Cox } 2241387aabc5SAlan Cox 2242387aabc5SAlan Cox /* 2243df8bae1dSRodney W. Grimes * Routine: vm_object_coalesce 2244df8bae1dSRodney W. Grimes * Function: Coalesces two objects backing up adjoining 2245df8bae1dSRodney W. Grimes * regions of memory into a single object. 2246df8bae1dSRodney W. Grimes * 2247df8bae1dSRodney W. Grimes * returns TRUE if objects were combined. 2248df8bae1dSRodney W. Grimes * 2249df8bae1dSRodney W. Grimes * NOTE: Only works at the moment if the second object is NULL - 2250df8bae1dSRodney W. Grimes * if it's not, which object do we lock first? 2251df8bae1dSRodney W. Grimes * 2252df8bae1dSRodney W. Grimes * Parameters: 2253df8bae1dSRodney W. Grimes * prev_object First object to coalesce 2254df8bae1dSRodney W. Grimes * prev_offset Offset into prev_object 2255df8bae1dSRodney W. Grimes * prev_size Size of reference to prev_object 225657a21abaSAlan Cox * next_size Size of reference to the second object 22573364c323SKonstantin Belousov * reserved Indicator that extension region has 22583364c323SKonstantin Belousov * swap accounted for 2259df8bae1dSRodney W. Grimes * 2260df8bae1dSRodney W. Grimes * Conditions: 2261df8bae1dSRodney W. Grimes * The object must *not* be locked. 2262df8bae1dSRodney W. Grimes */ 22630d94caffSDavid Greenman boolean_t 226457a21abaSAlan Cox vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, 22653364c323SKonstantin Belousov vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) 2266df8bae1dSRodney W. Grimes { 2267ea41812fSAlan Cox vm_pindex_t next_pindex; 2268df8bae1dSRodney W. Grimes 226900e1854aSAlan Cox if (prev_object == NULL) 2270df8bae1dSRodney W. Grimes return (TRUE); 227163967687SJeff Roberson if ((prev_object->flags & OBJ_ANON) == 0) 227230dcfc09SJohn Dyson return (FALSE); 227330dcfc09SJohn Dyson 227463967687SJeff Roberson VM_OBJECT_WLOCK(prev_object); 2275df8bae1dSRodney W. Grimes /* 227698087a06SJeff Roberson * Try to collapse the object first. 2277df8bae1dSRodney W. Grimes */ 2278df8bae1dSRodney W. Grimes vm_object_collapse(prev_object); 2279df8bae1dSRodney W. Grimes 2280df8bae1dSRodney W. Grimes /* 22810d94caffSDavid Greenman * Can't coalesce if: . more than one reference . paged out . shadows 22820d94caffSDavid Greenman * another object . has a copy elsewhere (any of which mean that the 22830d94caffSDavid Greenman * pages not mapped to prev_entry may be in use anyway) 2284df8bae1dSRodney W. Grimes */ 22858cc7e047SJohn Dyson if (prev_object->backing_object != NULL) { 228689f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 2287df8bae1dSRodney W. Grimes return (FALSE); 2288df8bae1dSRodney W. Grimes } 2289a316d390SJohn Dyson 2290a316d390SJohn Dyson prev_size >>= PAGE_SHIFT; 2291a316d390SJohn Dyson next_size >>= PAGE_SHIFT; 229257a21abaSAlan Cox next_pindex = OFF_TO_IDX(prev_offset) + prev_size; 22938cc7e047SJohn Dyson 22940e48e068SMark Johnston if (prev_object->ref_count > 1 && 22950e48e068SMark Johnston prev_object->size != next_pindex && 22960e48e068SMark Johnston (prev_object->flags & OBJ_ONEMAPPING) == 0) { 229789f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 22988cc7e047SJohn Dyson return (FALSE); 22998cc7e047SJohn Dyson } 23008cc7e047SJohn Dyson 2301df8bae1dSRodney W. Grimes /* 23023364c323SKonstantin Belousov * Account for the charge. 23033364c323SKonstantin Belousov */ 2304ef694c1aSEdward Tomasz Napierala if (prev_object->cred != NULL) { 23053364c323SKonstantin Belousov /* 23063364c323SKonstantin Belousov * If prev_object was charged, then this mapping, 2307763df3ecSPedro F. Giffuni * although not charged now, may become writable 2308ef694c1aSEdward Tomasz Napierala * later. Non-NULL cred in the object would prevent 23093364c323SKonstantin Belousov * swap reservation during enabling of the write 23103364c323SKonstantin Belousov * access, so reserve swap now. Failed reservation 23113364c323SKonstantin Belousov * cause allocation of the separate object for the map 23123364c323SKonstantin Belousov * entry, and swap reservation for this entry is 23133364c323SKonstantin Belousov * managed in appropriate time. 23143364c323SKonstantin Belousov */ 2315ef694c1aSEdward Tomasz Napierala if (!reserved && !swap_reserve_by_cred(ptoa(next_size), 2316ef694c1aSEdward Tomasz Napierala prev_object->cred)) { 23179f790a17SKonstantin Belousov VM_OBJECT_WUNLOCK(prev_object); 23183364c323SKonstantin Belousov return (FALSE); 23193364c323SKonstantin Belousov } 23203364c323SKonstantin Belousov prev_object->charge += ptoa(next_size); 23213364c323SKonstantin Belousov } 23223364c323SKonstantin Belousov 23233364c323SKonstantin Belousov /* 23240d94caffSDavid Greenman * Remove any pages that may still be in the object from a previous 23250d94caffSDavid Greenman * deallocation. 2326df8bae1dSRodney W. Grimes */ 2327ea41812fSAlan Cox if (next_pindex < prev_object->size) { 23286bbee8e2SAlan Cox vm_object_page_remove(prev_object, next_pindex, next_pindex + 23296bbee8e2SAlan Cox next_size, 0); 23303364c323SKonstantin Belousov #if 0 2331ef694c1aSEdward Tomasz Napierala if (prev_object->cred != NULL) { 23323364c323SKonstantin Belousov KASSERT(prev_object->charge >= 23333364c323SKonstantin Belousov ptoa(prev_object->size - next_pindex), 23343364c323SKonstantin Belousov ("object %p overcharged 1 %jx %jx", prev_object, 23353364c323SKonstantin Belousov (uintmax_t)next_pindex, (uintmax_t)next_size)); 23363364c323SKonstantin Belousov prev_object->charge -= ptoa(prev_object->size - 23373364c323SKonstantin Belousov next_pindex); 23383364c323SKonstantin Belousov } 23393364c323SKonstantin Belousov #endif 2340ea41812fSAlan Cox } 2341df8bae1dSRodney W. Grimes 2342df8bae1dSRodney W. Grimes /* 2343df8bae1dSRodney W. Grimes * Extend the object if necessary. 2344df8bae1dSRodney W. Grimes */ 2345ea41812fSAlan Cox if (next_pindex + next_size > prev_object->size) 2346ea41812fSAlan Cox prev_object->size = next_pindex + next_size; 2347df8bae1dSRodney W. Grimes 234889f6b863SAttilio Rao VM_OBJECT_WUNLOCK(prev_object); 2349df8bae1dSRodney W. Grimes return (TRUE); 2350df8bae1dSRodney W. Grimes } 2351df8bae1dSRodney W. Grimes 23527a5a6352SMatthew Dillon void 2353180bcaa4SKonstantin Belousov vm_object_set_writeable_dirty_(vm_object_t object) 23547a5a6352SMatthew Dillon { 235567d0e293SJeff Roberson atomic_add_int(&object->generation, 1); 23567a5a6352SMatthew Dillon } 23577a5a6352SMatthew Dillon 2358c23c555bSKonstantin Belousov bool 2359c23c555bSKonstantin Belousov vm_object_mightbedirty_(vm_object_t object) 2360c23c555bSKonstantin Belousov { 2361c23c555bSKonstantin Belousov return (object->generation != object->cleangeneration); 2362c23c555bSKonstantin Belousov } 2363c23c555bSKonstantin Belousov 236403462509SAlan Cox /* 236503462509SAlan Cox * vm_object_unwire: 236603462509SAlan Cox * 236703462509SAlan Cox * For each page offset within the specified range of the given object, 236803462509SAlan Cox * find the highest-level page in the shadow chain and unwire it. A page 236903462509SAlan Cox * must exist at every page offset, and the highest-level page must be 237003462509SAlan Cox * wired. 237103462509SAlan Cox */ 237203462509SAlan Cox void 237303462509SAlan Cox vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, 237403462509SAlan Cox uint8_t queue) 237503462509SAlan Cox { 237620e4afbfSKonstantin Belousov vm_object_t tobject, t1object; 237703462509SAlan Cox vm_page_t m, tm; 237803462509SAlan Cox vm_pindex_t end_pindex, pindex, tpindex; 237903462509SAlan Cox int depth, locked_depth; 238003462509SAlan Cox 238103462509SAlan Cox KASSERT((offset & PAGE_MASK) == 0, 238203462509SAlan Cox ("vm_object_unwire: offset is not page aligned")); 238303462509SAlan Cox KASSERT((length & PAGE_MASK) == 0, 238403462509SAlan Cox ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); 238503462509SAlan Cox /* The wired count of a fictitious page never changes. */ 238603462509SAlan Cox if ((object->flags & OBJ_FICTITIOUS) != 0) 238703462509SAlan Cox return; 238803462509SAlan Cox pindex = OFF_TO_IDX(offset); 238903462509SAlan Cox end_pindex = pindex + atop(length); 239020e4afbfSKonstantin Belousov again: 239103462509SAlan Cox locked_depth = 1; 239203462509SAlan Cox VM_OBJECT_RLOCK(object); 239303462509SAlan Cox m = vm_page_find_least(object, pindex); 239403462509SAlan Cox while (pindex < end_pindex) { 239503462509SAlan Cox if (m == NULL || pindex < m->pindex) { 239603462509SAlan Cox /* 239703462509SAlan Cox * The first object in the shadow chain doesn't 239803462509SAlan Cox * contain a page at the current index. Therefore, 239903462509SAlan Cox * the page must exist in a backing object. 240003462509SAlan Cox */ 240103462509SAlan Cox tobject = object; 240203462509SAlan Cox tpindex = pindex; 240303462509SAlan Cox depth = 0; 240403462509SAlan Cox do { 240503462509SAlan Cox tpindex += 240603462509SAlan Cox OFF_TO_IDX(tobject->backing_object_offset); 240703462509SAlan Cox tobject = tobject->backing_object; 240803462509SAlan Cox KASSERT(tobject != NULL, 240903462509SAlan Cox ("vm_object_unwire: missing page")); 241003462509SAlan Cox if ((tobject->flags & OBJ_FICTITIOUS) != 0) 241103462509SAlan Cox goto next_page; 241203462509SAlan Cox depth++; 241303462509SAlan Cox if (depth == locked_depth) { 241403462509SAlan Cox locked_depth++; 241503462509SAlan Cox VM_OBJECT_RLOCK(tobject); 241603462509SAlan Cox } 241703462509SAlan Cox } while ((tm = vm_page_lookup(tobject, tpindex)) == 241803462509SAlan Cox NULL); 241903462509SAlan Cox } else { 242003462509SAlan Cox tm = m; 242103462509SAlan Cox m = TAILQ_NEXT(m, listq); 242203462509SAlan Cox } 242363e97555SJeff Roberson if (vm_page_trysbusy(tm) == 0) { 242487e93ea6SMark Johnston for (tobject = object; locked_depth >= 1; 242520e4afbfSKonstantin Belousov locked_depth--) { 242620e4afbfSKonstantin Belousov t1object = tobject->backing_object; 242787e93ea6SMark Johnston if (tm->object != tobject) 242820e4afbfSKonstantin Belousov VM_OBJECT_RUNLOCK(tobject); 242920e4afbfSKonstantin Belousov tobject = t1object; 243020e4afbfSKonstantin Belousov } 243187b64663SMark Johnston tobject = tm->object; 243287b64663SMark Johnston if (!vm_page_busy_sleep(tm, "unwbo", 243387b64663SMark Johnston VM_ALLOC_IGN_SBUSY)) 243487b64663SMark Johnston VM_OBJECT_RUNLOCK(tobject); 243520e4afbfSKonstantin Belousov goto again; 243620e4afbfSKonstantin Belousov } 243703462509SAlan Cox vm_page_unwire(tm, queue); 243863e97555SJeff Roberson vm_page_sunbusy(tm); 243903462509SAlan Cox next_page: 244003462509SAlan Cox pindex++; 244103462509SAlan Cox } 244203462509SAlan Cox /* Release the accumulated object locks. */ 244320e4afbfSKonstantin Belousov for (tobject = object; locked_depth >= 1; locked_depth--) { 244420e4afbfSKonstantin Belousov t1object = tobject->backing_object; 244520e4afbfSKonstantin Belousov VM_OBJECT_RUNLOCK(tobject); 244620e4afbfSKonstantin Belousov tobject = t1object; 244703462509SAlan Cox } 244803462509SAlan Cox } 244903462509SAlan Cox 24500951bd36SEric van Gyzen /* 24510951bd36SEric van Gyzen * Return the vnode for the given object, or NULL if none exists. 24520951bd36SEric van Gyzen * For tmpfs objects, the function may return NULL if there is 24530951bd36SEric van Gyzen * no vnode allocated at the time of the call. 24540951bd36SEric van Gyzen */ 245563e4c6cdSEric van Gyzen struct vnode * 245663e4c6cdSEric van Gyzen vm_object_vnode(vm_object_t object) 245763e4c6cdSEric van Gyzen { 24580951bd36SEric van Gyzen struct vnode *vp; 245963e4c6cdSEric van Gyzen 246063e4c6cdSEric van Gyzen VM_OBJECT_ASSERT_LOCKED(object); 2461a7c198a2SKonstantin Belousov vm_pager_getvp(object, &vp, NULL); 24620951bd36SEric van Gyzen return (vp); 246363e4c6cdSEric van Gyzen } 246463e4c6cdSEric van Gyzen 2465205be21dSJeff Roberson /* 2466205be21dSJeff Roberson * Busy the vm object. This prevents new pages belonging to the object from 2467205be21dSJeff Roberson * becoming busy. Existing pages persist as busy. Callers are responsible 2468205be21dSJeff Roberson * for checking page state before proceeding. 2469205be21dSJeff Roberson */ 2470205be21dSJeff Roberson void 2471205be21dSJeff Roberson vm_object_busy(vm_object_t obj) 2472205be21dSJeff Roberson { 2473205be21dSJeff Roberson 2474205be21dSJeff Roberson VM_OBJECT_ASSERT_LOCKED(obj); 2475205be21dSJeff Roberson 2476c99d0c58SMark Johnston blockcount_acquire(&obj->busy, 1); 2477205be21dSJeff Roberson /* The fence is required to order loads of page busy. */ 2478205be21dSJeff Roberson atomic_thread_fence_acq_rel(); 2479205be21dSJeff Roberson } 2480205be21dSJeff Roberson 2481205be21dSJeff Roberson void 2482205be21dSJeff Roberson vm_object_unbusy(vm_object_t obj) 2483205be21dSJeff Roberson { 2484205be21dSJeff Roberson 2485c99d0c58SMark Johnston blockcount_release(&obj->busy, 1); 2486205be21dSJeff Roberson } 2487205be21dSJeff Roberson 2488205be21dSJeff Roberson void 2489205be21dSJeff Roberson vm_object_busy_wait(vm_object_t obj, const char *wmesg) 2490205be21dSJeff Roberson { 2491205be21dSJeff Roberson 2492205be21dSJeff Roberson VM_OBJECT_ASSERT_UNLOCKED(obj); 2493205be21dSJeff Roberson 2494c99d0c58SMark Johnston (void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM); 2495205be21dSJeff Roberson } 2496205be21dSJeff Roberson 2497cd37afd8SRick Macklem /* 2498cd37afd8SRick Macklem * This function aims to determine if the object is mapped, 2499cd37afd8SRick Macklem * specifically, if it is referenced by a vm_map_entry. Because 2500cd37afd8SRick Macklem * objects occasionally acquire transient references that do not 2501cd37afd8SRick Macklem * represent a mapping, the method used here is inexact. However, it 2502cd37afd8SRick Macklem * has very low overhead and is good enough for the advisory 2503cd37afd8SRick Macklem * vm.vmtotal sysctl. 2504cd37afd8SRick Macklem */ 2505cd37afd8SRick Macklem bool 2506cd37afd8SRick Macklem vm_object_is_active(vm_object_t obj) 2507cd37afd8SRick Macklem { 2508cd37afd8SRick Macklem 25094cf9f5d8SKonstantin Belousov return (obj->ref_count > atomic_load_int(&obj->shadow_count)); 2510cd37afd8SRick Macklem } 2511cd37afd8SRick Macklem 2512ff87ae35SJohn Baldwin static int 251342812cccSKonstantin Belousov vm_object_list_handler(struct sysctl_req *req, bool swap_only) 2514ff87ae35SJohn Baldwin { 25150ecee546SKonstantin Belousov struct kinfo_vmobject *kvo; 2516ff87ae35SJohn Baldwin char *fullpath, *freepath; 2517ff87ae35SJohn Baldwin struct vnode *vp; 2518ff87ae35SJohn Baldwin struct vattr va; 2519ff87ae35SJohn Baldwin vm_object_t obj; 2520ff87ae35SJohn Baldwin vm_page_t m; 2521ecfbddf0SKonstantin Belousov u_long sp; 2522ff87ae35SJohn Baldwin int count, error; 2523ff87ae35SJohn Baldwin 2524ff87ae35SJohn Baldwin if (req->oldptr == NULL) { 2525ff87ae35SJohn Baldwin /* 2526ff87ae35SJohn Baldwin * If an old buffer has not been provided, generate an 2527ff87ae35SJohn Baldwin * estimate of the space needed for a subsequent call. 2528ff87ae35SJohn Baldwin */ 2529ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2530ff87ae35SJohn Baldwin count = 0; 2531ff87ae35SJohn Baldwin TAILQ_FOREACH(obj, &vm_object_list, object_list) { 2532ff87ae35SJohn Baldwin if (obj->type == OBJT_DEAD) 2533ff87ae35SJohn Baldwin continue; 2534ff87ae35SJohn Baldwin count++; 2535ff87ae35SJohn Baldwin } 2536ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 2537ff87ae35SJohn Baldwin return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * 2538ff87ae35SJohn Baldwin count * 11 / 10)); 2539ff87ae35SJohn Baldwin } 2540ff87ae35SJohn Baldwin 2541b050ee6cSMark Johnston kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK | M_ZERO); 2542ff87ae35SJohn Baldwin error = 0; 2543ff87ae35SJohn Baldwin 2544ff87ae35SJohn Baldwin /* 2545ff87ae35SJohn Baldwin * VM objects are type stable and are never removed from the 2546ff87ae35SJohn Baldwin * list once added. This allows us to safely read obj->object_list 2547ff87ae35SJohn Baldwin * after reacquiring the VM object lock. 2548ff87ae35SJohn Baldwin */ 2549ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2550ff87ae35SJohn Baldwin TAILQ_FOREACH(obj, &vm_object_list, object_list) { 255142812cccSKonstantin Belousov if (obj->type == OBJT_DEAD || 255242812cccSKonstantin Belousov (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) 2553ff87ae35SJohn Baldwin continue; 2554ff87ae35SJohn Baldwin VM_OBJECT_RLOCK(obj); 255542812cccSKonstantin Belousov if (obj->type == OBJT_DEAD || 255642812cccSKonstantin Belousov (swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) { 2557ff87ae35SJohn Baldwin VM_OBJECT_RUNLOCK(obj); 2558ff87ae35SJohn Baldwin continue; 2559ff87ae35SJohn Baldwin } 2560ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 25610ecee546SKonstantin Belousov kvo->kvo_size = ptoa(obj->size); 25620ecee546SKonstantin Belousov kvo->kvo_resident = obj->resident_page_count; 25630ecee546SKonstantin Belousov kvo->kvo_ref_count = obj->ref_count; 25644cf9f5d8SKonstantin Belousov kvo->kvo_shadow_count = atomic_load_int(&obj->shadow_count); 25650ecee546SKonstantin Belousov kvo->kvo_memattr = obj->memattr; 25660ecee546SKonstantin Belousov kvo->kvo_active = 0; 25670ecee546SKonstantin Belousov kvo->kvo_inactive = 0; 25687738118eSKonstantin Belousov if (!swap_only) { 2569ff87ae35SJohn Baldwin TAILQ_FOREACH(m, &obj->memq, listq) { 2570ff87ae35SJohn Baldwin /* 2571ff87ae35SJohn Baldwin * A page may belong to the object but be 2572ff87ae35SJohn Baldwin * dequeued and set to PQ_NONE while the 2573ff87ae35SJohn Baldwin * object lock is not held. This makes the 2574ff87ae35SJohn Baldwin * reads of m->queue below racy, and we do not 2575ff87ae35SJohn Baldwin * count pages set to PQ_NONE. However, this 2576ff87ae35SJohn Baldwin * sysctl is only meant to give an 2577ff87ae35SJohn Baldwin * approximation of the system anyway. 2578ff87ae35SJohn Baldwin */ 25795cff1f4dSMark Johnston if (m->a.queue == PQ_ACTIVE) 25800ecee546SKonstantin Belousov kvo->kvo_active++; 25815cff1f4dSMark Johnston else if (m->a.queue == PQ_INACTIVE) 25820ecee546SKonstantin Belousov kvo->kvo_inactive++; 2583ff87ae35SJohn Baldwin } 25847738118eSKonstantin Belousov } 2585ff87ae35SJohn Baldwin 25860ecee546SKonstantin Belousov kvo->kvo_vn_fileid = 0; 25870ecee546SKonstantin Belousov kvo->kvo_vn_fsid = 0; 25880ecee546SKonstantin Belousov kvo->kvo_vn_fsid_freebsd11 = 0; 2589ff87ae35SJohn Baldwin freepath = NULL; 2590ff87ae35SJohn Baldwin fullpath = ""; 25917738118eSKonstantin Belousov vp = NULL; 25927738118eSKonstantin Belousov kvo->kvo_type = vm_object_kvme_type(obj, swap_only ? NULL : &vp); 2593ecfbddf0SKonstantin Belousov if (vp != NULL) { 2594ff87ae35SJohn Baldwin vref(vp); 2595ecfbddf0SKonstantin Belousov } else if ((obj->flags & OBJ_ANON) != 0) { 25967f3c78fbSMark Johnston MPASS(kvo->kvo_type == KVME_TYPE_SWAP); 2597ecfbddf0SKonstantin Belousov kvo->kvo_me = (uintptr_t)obj; 2598ecfbddf0SKonstantin Belousov /* tmpfs objs are reported as vnodes */ 2599ecfbddf0SKonstantin Belousov kvo->kvo_backing_obj = (uintptr_t)obj->backing_object; 2600ecfbddf0SKonstantin Belousov sp = swap_pager_swapped_pages(obj); 2601ecfbddf0SKonstantin Belousov kvo->kvo_swapped = sp > UINT32_MAX ? UINT32_MAX : sp; 2602ecfbddf0SKonstantin Belousov } 2603ff87ae35SJohn Baldwin VM_OBJECT_RUNLOCK(obj); 2604ff87ae35SJohn Baldwin if (vp != NULL) { 2605feabaaf9SMateusz Guzik vn_fullpath(vp, &fullpath, &freepath); 2606ff87ae35SJohn Baldwin vn_lock(vp, LK_SHARED | LK_RETRY); 2607ff87ae35SJohn Baldwin if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { 26080ecee546SKonstantin Belousov kvo->kvo_vn_fileid = va.va_fileid; 26090ecee546SKonstantin Belousov kvo->kvo_vn_fsid = va.va_fsid; 26100ecee546SKonstantin Belousov kvo->kvo_vn_fsid_freebsd11 = va.va_fsid; 261169921123SKonstantin Belousov /* truncate */ 2612ff87ae35SJohn Baldwin } 2613ff87ae35SJohn Baldwin vput(vp); 2614ff87ae35SJohn Baldwin } 2615ff87ae35SJohn Baldwin 26160ecee546SKonstantin Belousov strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path)); 2617ff87ae35SJohn Baldwin if (freepath != NULL) 2618ff87ae35SJohn Baldwin free(freepath, M_TEMP); 2619ff87ae35SJohn Baldwin 2620ff87ae35SJohn Baldwin /* Pack record size down */ 26210ecee546SKonstantin Belousov kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) 26220ecee546SKonstantin Belousov + strlen(kvo->kvo_path) + 1; 26230ecee546SKonstantin Belousov kvo->kvo_structsize = roundup(kvo->kvo_structsize, 2624ff87ae35SJohn Baldwin sizeof(uint64_t)); 26250ecee546SKonstantin Belousov error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize); 2626350fc36bSKonstantin Belousov maybe_yield(); 2627ff87ae35SJohn Baldwin mtx_lock(&vm_object_list_mtx); 2628ff87ae35SJohn Baldwin if (error) 2629ff87ae35SJohn Baldwin break; 2630ff87ae35SJohn Baldwin } 2631ff87ae35SJohn Baldwin mtx_unlock(&vm_object_list_mtx); 26320ecee546SKonstantin Belousov free(kvo, M_TEMP); 2633ff87ae35SJohn Baldwin return (error); 2634ff87ae35SJohn Baldwin } 26351b610624SKonstantin Belousov 26361b610624SKonstantin Belousov static int 26371b610624SKonstantin Belousov sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) 26381b610624SKonstantin Belousov { 263942812cccSKonstantin Belousov return (vm_object_list_handler(req, false)); 26401b610624SKonstantin Belousov } 26411b610624SKonstantin Belousov 2642ff87ae35SJohn Baldwin SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | 2643ff87ae35SJohn Baldwin CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", 2644ff87ae35SJohn Baldwin "List of VM objects"); 2645ff87ae35SJohn Baldwin 264642812cccSKonstantin Belousov static int 264742812cccSKonstantin Belousov sysctl_vm_object_list_swap(SYSCTL_HANDLER_ARGS) 264842812cccSKonstantin Belousov { 264942812cccSKonstantin Belousov return (vm_object_list_handler(req, true)); 265042812cccSKonstantin Belousov } 265142812cccSKonstantin Belousov 265242812cccSKonstantin Belousov /* 265342812cccSKonstantin Belousov * This sysctl returns list of the anonymous or swap objects. Intent 265442812cccSKonstantin Belousov * is to provide stripped optimized list useful to analyze swap use. 265542812cccSKonstantin Belousov * Since technically non-swap (default) objects participate in the 265642812cccSKonstantin Belousov * shadow chains, and are converted to swap type as needed by swap 265742812cccSKonstantin Belousov * pager, we must report them. 265842812cccSKonstantin Belousov */ 265942812cccSKonstantin Belousov SYSCTL_PROC(_vm, OID_AUTO, swap_objects, 266042812cccSKonstantin Belousov CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, 266142812cccSKonstantin Belousov sysctl_vm_object_list_swap, "S,kinfo_vmobject", 266242812cccSKonstantin Belousov "List of swap VM objects"); 266342812cccSKonstantin Belousov 2664c7c34a24SBruce Evans #include "opt_ddb.h" 2665c3cb3e12SDavid Greenman #ifdef DDB 2666c7c34a24SBruce Evans #include <sys/kernel.h> 2667c7c34a24SBruce Evans 2668ce9edcf5SPoul-Henning Kamp #include <sys/cons.h> 2669c7c34a24SBruce Evans 2670c7c34a24SBruce Evans #include <ddb/ddb.h> 2671c7c34a24SBruce Evans 2672cac597e4SBruce Evans static int 26731b40f8c0SMatthew Dillon _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) 2674a1f6d91cSDavid Greenman { 2675a1f6d91cSDavid Greenman vm_map_t tmpm; 2676a1f6d91cSDavid Greenman vm_map_entry_t tmpe; 2677a1f6d91cSDavid Greenman vm_object_t obj; 2678a1f6d91cSDavid Greenman 2679a1f6d91cSDavid Greenman if (map == 0) 2680a1f6d91cSDavid Greenman return 0; 2681a1f6d91cSDavid Greenman 2682a1f6d91cSDavid Greenman if (entry == 0) { 26832288078cSDoug Moore VM_MAP_ENTRY_FOREACH(tmpe, map) { 2684a1f6d91cSDavid Greenman if (_vm_object_in_map(map, object, tmpe)) { 2685a1f6d91cSDavid Greenman return 1; 2686a1f6d91cSDavid Greenman } 2687a1f6d91cSDavid Greenman } 26889fdfe602SMatthew Dillon } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { 26899fdfe602SMatthew Dillon tmpm = entry->object.sub_map; 26902288078cSDoug Moore VM_MAP_ENTRY_FOREACH(tmpe, tmpm) { 2691a1f6d91cSDavid Greenman if (_vm_object_in_map(tmpm, object, tmpe)) { 2692a1f6d91cSDavid Greenman return 1; 2693a1f6d91cSDavid Greenman } 2694a1f6d91cSDavid Greenman } 26958aef1712SMatthew Dillon } else if ((obj = entry->object.vm_object) != NULL) { 269624a1cce3SDavid Greenman for (; obj; obj = obj->backing_object) 2697a1f6d91cSDavid Greenman if (obj == object) { 2698a1f6d91cSDavid Greenman return 1; 2699a1f6d91cSDavid Greenman } 2700a1f6d91cSDavid Greenman } 2701a1f6d91cSDavid Greenman return 0; 2702a1f6d91cSDavid Greenman } 2703a1f6d91cSDavid Greenman 2704cac597e4SBruce Evans static int 27051b40f8c0SMatthew Dillon vm_object_in_map(vm_object_t object) 2706a1f6d91cSDavid Greenman { 2707a1f6d91cSDavid Greenman struct proc *p; 27081005a129SJohn Baldwin 270960517fd1SJohn Baldwin /* sx_slock(&allproc_lock); */ 2710f67af5c9SXin LI FOREACH_PROC_IN_SYSTEM(p) { 2711a1f6d91cSDavid Greenman if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) 2712a1f6d91cSDavid Greenman continue; 2713553629ebSJake Burkholder if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { 271460517fd1SJohn Baldwin /* sx_sunlock(&allproc_lock); */ 2715a1f6d91cSDavid Greenman return 1; 2716a1f6d91cSDavid Greenman } 2717553629ebSJake Burkholder } 271860517fd1SJohn Baldwin /* sx_sunlock(&allproc_lock); */ 2719a1f6d91cSDavid Greenman if (_vm_object_in_map(kernel_map, object, 0)) 2720a1f6d91cSDavid Greenman return 1; 2721a1f6d91cSDavid Greenman return 0; 2722a1f6d91cSDavid Greenman } 2723a1f6d91cSDavid Greenman 2724c84c5e00SMitchell Horne DB_SHOW_COMMAND_FLAGS(vmochk, vm_object_check, DB_CMD_MEMSAFE) 2725f708ef1bSPoul-Henning Kamp { 2726a1f6d91cSDavid Greenman vm_object_t object; 2727a1f6d91cSDavid Greenman 2728a1f6d91cSDavid Greenman /* 2729a1f6d91cSDavid Greenman * make sure that internal objs are in a map somewhere 2730a1f6d91cSDavid Greenman * and none have zero ref counts. 2731a1f6d91cSDavid Greenman */ 2732cc64b484SAlfred Perlstein TAILQ_FOREACH(object, &vm_object_list, object_list) { 273332362449SKonstantin Belousov if ((object->flags & OBJ_ANON) != 0) { 2734a1f6d91cSDavid Greenman if (object->ref_count == 0) { 27353efc015bSPeter Wemm db_printf("vmochk: internal obj has zero ref count: %ld\n", 27363efc015bSPeter Wemm (long)object->size); 2737a1f6d91cSDavid Greenman } 2738a1f6d91cSDavid Greenman if (!vm_object_in_map(object)) { 2739fc62ef1fSBruce Evans db_printf( 2740fc62ef1fSBruce Evans "vmochk: internal obj is not in a map: " 2741fc62ef1fSBruce Evans "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", 2742fc62ef1fSBruce Evans object->ref_count, (u_long)object->size, 2743fc62ef1fSBruce Evans (u_long)object->size, 2744fc62ef1fSBruce Evans (void *)object->backing_object); 2745a1f6d91cSDavid Greenman } 2746a1f6d91cSDavid Greenman } 27478cc8c586SEric van Gyzen if (db_pager_quit) 27488cc8c586SEric van Gyzen return; 2749a1f6d91cSDavid Greenman } 2750a1f6d91cSDavid Greenman } 2751a1f6d91cSDavid Greenman 275226f9a767SRodney W. Grimes /* 2753df8bae1dSRodney W. Grimes * vm_object_print: [ debug ] 2754df8bae1dSRodney W. Grimes */ 2755c7c34a24SBruce Evans DB_SHOW_COMMAND(object, vm_object_print_static) 2756df8bae1dSRodney W. Grimes { 2757c7c34a24SBruce Evans /* XXX convert args. */ 2758c7c34a24SBruce Evans vm_object_t object = (vm_object_t)addr; 2759c7c34a24SBruce Evans boolean_t full = have_addr; 2760c7c34a24SBruce Evans 2761d031cff1SMatthew Dillon vm_page_t p; 2762df8bae1dSRodney W. Grimes 2763c7c34a24SBruce Evans /* XXX count is an (unused) arg. Avoid shadowing it. */ 2764c7c34a24SBruce Evans #define count was_count 2765c7c34a24SBruce Evans 2766d031cff1SMatthew Dillon int count; 2767df8bae1dSRodney W. Grimes 2768df8bae1dSRodney W. Grimes if (object == NULL) 2769df8bae1dSRodney W. Grimes return; 2770df8bae1dSRodney W. Grimes 2771eb95adefSBruce Evans db_iprintf( 2772ef694c1aSEdward Tomasz Napierala "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", 2773e47cd172SMaxime Henrion object, (int)object->type, (uintmax_t)object->size, 27743364c323SKonstantin Belousov object->resident_page_count, object->ref_count, object->flags, 2775ef694c1aSEdward Tomasz Napierala object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); 2776e47cd172SMaxime Henrion db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", 27774cf9f5d8SKonstantin Belousov atomic_load_int(&object->shadow_count), 2778eb95adefSBruce Evans object->backing_object ? object->backing_object->ref_count : 0, 2779e47cd172SMaxime Henrion object->backing_object, (uintmax_t)object->backing_object_offset); 2780df8bae1dSRodney W. Grimes 2781df8bae1dSRodney W. Grimes if (!full) 2782df8bae1dSRodney W. Grimes return; 2783df8bae1dSRodney W. Grimes 2784c7c34a24SBruce Evans db_indent += 2; 2785df8bae1dSRodney W. Grimes count = 0; 2786fc2ffbe6SPoul-Henning Kamp TAILQ_FOREACH(p, &object->memq, listq) { 2787df8bae1dSRodney W. Grimes if (count == 0) 2788c7c34a24SBruce Evans db_iprintf("memory:="); 2789df8bae1dSRodney W. Grimes else if (count == 6) { 2790c7c34a24SBruce Evans db_printf("\n"); 2791c7c34a24SBruce Evans db_iprintf(" ..."); 2792df8bae1dSRodney W. Grimes count = 0; 2793df8bae1dSRodney W. Grimes } else 2794c7c34a24SBruce Evans db_printf(","); 2795df8bae1dSRodney W. Grimes count++; 2796df8bae1dSRodney W. Grimes 2797e47cd172SMaxime Henrion db_printf("(off=0x%jx,page=0x%jx)", 2798e47cd172SMaxime Henrion (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); 27998cc8c586SEric van Gyzen 28008cc8c586SEric van Gyzen if (db_pager_quit) 28018cc8c586SEric van Gyzen break; 2802df8bae1dSRodney W. Grimes } 2803df8bae1dSRodney W. Grimes if (count != 0) 2804c7c34a24SBruce Evans db_printf("\n"); 2805c7c34a24SBruce Evans db_indent -= 2; 2806df8bae1dSRodney W. Grimes } 28075070c7f8SJohn Dyson 2808c7c34a24SBruce Evans /* XXX. */ 2809c7c34a24SBruce Evans #undef count 2810c7c34a24SBruce Evans 2811c7c34a24SBruce Evans /* XXX need this non-static entry for calling from vm_map_print. */ 28125070c7f8SJohn Dyson void 28131b40f8c0SMatthew Dillon vm_object_print( 28141b40f8c0SMatthew Dillon /* db_expr_t */ long addr, 28151b40f8c0SMatthew Dillon boolean_t have_addr, 28161b40f8c0SMatthew Dillon /* db_expr_t */ long count, 28171b40f8c0SMatthew Dillon char *modif) 2818c7c34a24SBruce Evans { 2819c7c34a24SBruce Evans vm_object_print_static(addr, have_addr, count, modif); 2820c7c34a24SBruce Evans } 2821c7c34a24SBruce Evans 2822c84c5e00SMitchell Horne DB_SHOW_COMMAND_FLAGS(vmopag, vm_object_print_pages, DB_CMD_MEMSAFE) 28235070c7f8SJohn Dyson { 28245070c7f8SJohn Dyson vm_object_t object; 2825bb2ac86fSKonstantin Belousov vm_pindex_t fidx; 2826bb2ac86fSKonstantin Belousov vm_paddr_t pa; 2827bb2ac86fSKonstantin Belousov vm_page_t m, prev_m; 282818048b6eSStephen J. Kiernan int rcount; 2829cc64b484SAlfred Perlstein 2830cc64b484SAlfred Perlstein TAILQ_FOREACH(object, &vm_object_list, object_list) { 2831fc62ef1fSBruce Evans db_printf("new object: %p\n", (void *)object); 283218048b6eSStephen J. Kiernan if (db_pager_quit) 28335070c7f8SJohn Dyson return; 283418048b6eSStephen J. Kiernan 28355070c7f8SJohn Dyson rcount = 0; 28365070c7f8SJohn Dyson fidx = 0; 2837bb2ac86fSKonstantin Belousov pa = -1; 2838bb2ac86fSKonstantin Belousov TAILQ_FOREACH(m, &object->memq, listq) { 2839bb2ac86fSKonstantin Belousov if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && 2840bb2ac86fSKonstantin Belousov prev_m->pindex + 1 != m->pindex) { 28415070c7f8SJohn Dyson if (rcount) { 28423efc015bSPeter Wemm db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28433efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 284418048b6eSStephen J. Kiernan if (db_pager_quit) 28455070c7f8SJohn Dyson return; 28465070c7f8SJohn Dyson rcount = 0; 28475070c7f8SJohn Dyson } 28485070c7f8SJohn Dyson } 28495070c7f8SJohn Dyson if (rcount && 28505070c7f8SJohn Dyson (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { 28515070c7f8SJohn Dyson ++rcount; 28525070c7f8SJohn Dyson continue; 28535070c7f8SJohn Dyson } 28545070c7f8SJohn Dyson if (rcount) { 28552446e4f0SAlan Cox db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28563efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 285718048b6eSStephen J. Kiernan if (db_pager_quit) 28585070c7f8SJohn Dyson return; 28595070c7f8SJohn Dyson } 2860bb2ac86fSKonstantin Belousov fidx = m->pindex; 28615070c7f8SJohn Dyson pa = VM_PAGE_TO_PHYS(m); 28625070c7f8SJohn Dyson rcount = 1; 28635070c7f8SJohn Dyson } 28645070c7f8SJohn Dyson if (rcount) { 28653efc015bSPeter Wemm db_printf(" index(%ld)run(%d)pa(0x%lx)\n", 28663efc015bSPeter Wemm (long)fidx, rcount, (long)pa); 286718048b6eSStephen J. Kiernan if (db_pager_quit) 28685070c7f8SJohn Dyson return; 28695070c7f8SJohn Dyson } 28705070c7f8SJohn Dyson } 28715070c7f8SJohn Dyson } 2872c3cb3e12SDavid Greenman #endif /* DDB */ 2873