vnode_pager.c revision 100832
11541Srgrimes/* 21541Srgrimes * Copyright (c) 1990 University of Utah. 31549Srgrimes * Copyright (c) 1991 The Regents of the University of California. 41549Srgrimes * All rights reserved. 59507Sdg * Copyright (c) 1993, 1994 John S. Dyson 69507Sdg * Copyright (c) 1995, David Greenman 71541Srgrimes * 81541Srgrimes * This code is derived from software contributed to Berkeley by 91541Srgrimes * the Systems Programming Group of the University of Utah Computer 101541Srgrimes * Science Department. 111541Srgrimes * 121541Srgrimes * Redistribution and use in source and binary forms, with or without 131541Srgrimes * modification, are permitted provided that the following conditions 141541Srgrimes * are met: 151541Srgrimes * 1. Redistributions of source code must retain the above copyright 161541Srgrimes * notice, this list of conditions and the following disclaimer. 171541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 181541Srgrimes * notice, this list of conditions and the following disclaimer in the 191541Srgrimes * documentation and/or other materials provided with the distribution. 201541Srgrimes * 3. All advertising materials mentioning features or use of this software 2158705Scharnier * must display the following acknowledgement: 221541Srgrimes * This product includes software developed by the University of 231541Srgrimes * California, Berkeley and its contributors. 241541Srgrimes * 4. Neither the name of the University nor the names of its contributors 251541Srgrimes * may be used to endorse or promote products derived from this software 261541Srgrimes * without specific prior written permission. 271541Srgrimes * 281541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 291541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 301541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 311541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 321541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 331541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 341541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 351541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 361541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 371541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 381541Srgrimes * SUCH DAMAGE. 391541Srgrimes * 401549Srgrimes * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 4150477Speter * $FreeBSD: head/sys/vm/vnode_pager.c 100832 2002-07-28 20:13:48Z alc $ 421541Srgrimes */ 431541Srgrimes 441541Srgrimes/* 451541Srgrimes * Page to/from files (vnodes). 461541Srgrimes */ 471541Srgrimes 481549Srgrimes/* 491549Srgrimes * TODO: 509507Sdg * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will 517695Sdg * greatly re-simplify the vnode_pager. 521549Srgrimes */ 531549Srgrimes 541541Srgrimes#include <sys/param.h> 551541Srgrimes#include <sys/systm.h> 561541Srgrimes#include <sys/proc.h> 571541Srgrimes#include <sys/vnode.h> 581541Srgrimes#include <sys/mount.h> 5960041Sphk#include <sys/bio.h> 609507Sdg#include <sys/buf.h> 6112662Sdg#include <sys/vmmeter.h> 6251340Sdillon#include <sys/conf.h> 631541Srgrimes 641541Srgrimes#include <vm/vm.h> 6512662Sdg#include <vm/vm_object.h> 661541Srgrimes#include <vm/vm_page.h> 679507Sdg#include <vm/vm_pager.h> 6831853Sdyson#include <vm/vm_map.h> 691541Srgrimes#include <vm/vnode_pager.h> 7012662Sdg#include <vm/vm_extern.h> 711541Srgrimes 7292727Salfredstatic void vnode_pager_init(void); 7392727Salfredstatic vm_offset_t vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, 7492727Salfred int *run); 7592727Salfredstatic void vnode_pager_iodone(struct buf *bp); 7692727Salfredstatic int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m); 7792727Salfredstatic int vnode_pager_input_old(vm_object_t object, vm_page_t m); 7892727Salfredstatic void vnode_pager_dealloc(vm_object_t); 7992727Salfredstatic int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int); 8092727Salfredstatic void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); 8192727Salfredstatic boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 8211943Sbde 831541Srgrimesstruct pagerops vnodepagerops = { 8479127Sjhb vnode_pager_init, 851541Srgrimes vnode_pager_alloc, 861541Srgrimes vnode_pager_dealloc, 879507Sdg vnode_pager_getpages, 889507Sdg vnode_pager_putpages, 899507Sdg vnode_pager_haspage, 909507Sdg NULL 911541Srgrimes}; 921541Srgrimes 9379127Sjhbint vnode_pbuf_freecnt; 9410556Sdyson 9579127Sjhbvoid 9679127Sjhbvnode_pager_init(void) 9779127Sjhb{ 9842957Sdillon 9979127Sjhb vnode_pbuf_freecnt = nswbuf / 2 + 1; 10079127Sjhb} 10179127Sjhb 1021541Srgrimes/* 1031541Srgrimes * Allocate (or lookup) pager for a vnode. 1041541Srgrimes * Handle is a vnode pointer. 10598604Salc * 10698604Salc * MPSAFE 1071541Srgrimes */ 1089507Sdgvm_object_t 10940286Sdgvnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 11028751Sbde vm_ooffset_t offset) 1111541Srgrimes{ 1129456Sdg vm_object_t object; 1131541Srgrimes struct vnode *vp; 1141541Srgrimes 1151541Srgrimes /* 1161541Srgrimes * Pageout to vnode, no can do yet. 1171541Srgrimes */ 1181541Srgrimes if (handle == NULL) 1191827Sdg return (NULL); 1201541Srgrimes 1219411Sdg vp = (struct vnode *) handle; 1229411Sdg 12398604Salc mtx_lock(&Giant); 1241541Srgrimes /* 1259411Sdg * Prevent race condition when allocating the object. This 1269411Sdg * can happen with NFS vnodes since the nfsnode isn't locked. 1271541Srgrimes */ 1289411Sdg while (vp->v_flag & VOLOCK) { 1299411Sdg vp->v_flag |= VOWANT; 1309411Sdg tsleep(vp, PVM, "vnpobj", 0); 1319411Sdg } 1329411Sdg vp->v_flag |= VOLOCK; 1339411Sdg 1349411Sdg /* 1359411Sdg * If the object is being terminated, wait for it to 1369411Sdg * go away. 1379411Sdg */ 13813490Sdyson while (((object = vp->v_object) != NULL) && 13913490Sdyson (object->flags & OBJ_DEAD)) { 14079224Sdillon tsleep(object, PVM, "vadead", 0); 1419507Sdg } 1425455Sdg 14332071Sdyson if (vp->v_usecount == 0) 14432071Sdyson panic("vnode_pager_alloc: no vnode reference"); 14532071Sdyson 1469507Sdg if (object == NULL) { 1471541Srgrimes /* 1481541Srgrimes * And an object of the appropriate size 1491541Srgrimes */ 15040286Sdg object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); 1511827Sdg 15240286Sdg object->un_pager.vnp.vnp_size = size; 1531549Srgrimes 1549507Sdg object->handle = handle; 1559507Sdg vp->v_object = object; 1561541Srgrimes } else { 15732286Sdyson object->ref_count++; 1581541Srgrimes } 15998604Salc vp->v_usecount++; 1609411Sdg vp->v_flag &= ~VOLOCK; 1619411Sdg if (vp->v_flag & VOWANT) { 1629411Sdg vp->v_flag &= ~VOWANT; 1639411Sdg wakeup(vp); 1649411Sdg } 16598604Salc mtx_unlock(&Giant); 1669507Sdg return (object); 1671541Srgrimes} 1681541Srgrimes 16912820Sphkstatic void 1709507Sdgvnode_pager_dealloc(object) 1719507Sdg vm_object_t object; 1721541Srgrimes{ 17379242Sdillon struct vnode *vp = object->handle; 1741541Srgrimes 17579224Sdillon GIANT_REQUIRED; 1769507Sdg if (vp == NULL) 1779507Sdg panic("vnode_pager_dealloc: pager already dealloced"); 1789507Sdg 17933817Sdyson vm_object_pip_wait(object, "vnpdea"); 1801541Srgrimes 1819507Sdg object->handle = NULL; 18233109Sdyson object->type = OBJT_DEAD; 1839507Sdg vp->v_object = NULL; 18433109Sdyson vp->v_flag &= ~(VTEXT | VOBJBUF); 1851549Srgrimes} 1861541Srgrimes 18712820Sphkstatic boolean_t 18812767Sdysonvnode_pager_haspage(object, pindex, before, after) 1899507Sdg vm_object_t object; 19012767Sdyson vm_pindex_t pindex; 1919507Sdg int *before; 1929507Sdg int *after; 1931541Srgrimes{ 1949507Sdg struct vnode *vp = object->handle; 19596572Sphk daddr_t bn; 19612423Sphk int err; 19710556Sdyson daddr_t reqblock; 19811701Sdyson int poff; 19911701Sdyson int bsize; 20012914Sdyson int pagesperblock, blocksperpage; 2011541Srgrimes 20279224Sdillon GIANT_REQUIRED; 20351340Sdillon /* 20451340Sdillon * If no vp or vp is doomed or marked transparent to VM, we do not 20551340Sdillon * have the page. 20651340Sdillon */ 20732585Sdyson if ((vp == NULL) || (vp->v_flag & VDOOMED)) 20832585Sdyson return FALSE; 20932585Sdyson 2101541Srgrimes /* 2115455Sdg * If filesystem no longer mounted or offset beyond end of file we do 2125455Sdg * not have the page. 2131541Srgrimes */ 21412767Sdyson if ((vp->v_mount == NULL) || 21581140Sjhb (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 2164797Sdg return FALSE; 2171541Srgrimes 21811576Sdg bsize = vp->v_mount->mnt_stat.f_iosize; 21910556Sdyson pagesperblock = bsize / PAGE_SIZE; 22012914Sdyson blocksperpage = 0; 22112914Sdyson if (pagesperblock > 0) { 22212914Sdyson reqblock = pindex / pagesperblock; 22312914Sdyson } else { 22412914Sdyson blocksperpage = (PAGE_SIZE / bsize); 22512914Sdyson reqblock = pindex * blocksperpage; 22612914Sdyson } 22710556Sdyson err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 22810556Sdyson after, before); 2298876Srgrimes if (err) 2309507Sdg return TRUE; 23192029Seivind if (bn == -1) 23210576Sdyson return FALSE; 23312914Sdyson if (pagesperblock > 0) { 23412914Sdyson poff = pindex - (reqblock * pagesperblock); 23512914Sdyson if (before) { 23612914Sdyson *before *= pagesperblock; 23712914Sdyson *before += poff; 23810669Sdyson } 23912914Sdyson if (after) { 24012914Sdyson int numafter; 24112914Sdyson *after *= pagesperblock; 24212914Sdyson numafter = pagesperblock - (poff + 1); 24399211Srobert if (IDX_TO_OFF(pindex + numafter) > 24499211Srobert object->un_pager.vnp.vnp_size) { 24599211Srobert numafter = 24699211Srobert OFF_TO_IDX(object->un_pager.vnp.vnp_size) - 24799211Srobert pindex; 24812914Sdyson } 24912914Sdyson *after += numafter; 25012914Sdyson } 25112914Sdyson } else { 25212914Sdyson if (before) { 25312914Sdyson *before /= blocksperpage; 25412914Sdyson } 25512914Sdyson 25612914Sdyson if (after) { 25712914Sdyson *after /= blocksperpage; 25812914Sdyson } 25910556Sdyson } 26010576Sdyson return TRUE; 2611541Srgrimes} 2621541Srgrimes 2631541Srgrimes/* 2641541Srgrimes * Lets the VM system know about a change in size for a file. 2659507Sdg * We adjust our own internal size and flush any cached pages in 2661541Srgrimes * the associated object that are affected by the size change. 2671541Srgrimes * 2681541Srgrimes * Note: this routine may be invoked as a result of a pager put 2691541Srgrimes * operation (possibly at object termination time), so we must be careful. 2701541Srgrimes */ 2711541Srgrimesvoid 2721541Srgrimesvnode_pager_setsize(vp, nsize) 2731541Srgrimes struct vnode *vp; 27412767Sdyson vm_ooffset_t nsize; 2751541Srgrimes{ 27638542Sluoqi vm_pindex_t nobjsize; 2779507Sdg vm_object_t object = vp->v_object; 2781541Srgrimes 27979224Sdillon GIANT_REQUIRED; 28079224Sdillon 2819507Sdg if (object == NULL) 2821541Srgrimes return; 2831827Sdg 2841541Srgrimes /* 2851541Srgrimes * Hasn't changed size 2861541Srgrimes */ 2879507Sdg if (nsize == object->un_pager.vnp.vnp_size) 2883374Sdg return; 2891827Sdg 29038542Sluoqi nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); 29138542Sluoqi 2921541Srgrimes /* 2931827Sdg * File has shrunk. Toss any cached pages beyond the new EOF. 2941541Srgrimes */ 2959507Sdg if (nsize < object->un_pager.vnp.vnp_size) { 29696095Salc#ifdef ENABLE_VFS_IOOPT 29738542Sluoqi vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size); 29896095Salc#endif 29938542Sluoqi if (nobjsize < object->size) { 30038542Sluoqi vm_object_page_remove(object, nobjsize, object->size, 30138542Sluoqi FALSE); 3025455Sdg } 3031827Sdg /* 3041827Sdg * this gets rid of garbage at the end of a page that is now 30587834Sdillon * only partially backed by the vnode. 30687834Sdillon * 30787834Sdillon * XXX for some reason (I don't know yet), if we take a 30887834Sdillon * completely invalid page and mark it partially valid 30987834Sdillon * it can screw up NFS reads, so we don't allow the case. 3101827Sdg */ 3111827Sdg if (nsize & PAGE_MASK) { 3121827Sdg vm_page_t m; 3131827Sdg 31412767Sdyson m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 31587834Sdillon if (m && m->valid) { 31670374Sdillon int base = (int)nsize & PAGE_MASK; 31770374Sdillon int size = PAGE_SIZE - base; 31870374Sdillon 31970374Sdillon /* 32070374Sdillon * Clear out partial-page garbage in case 32170374Sdillon * the page has been mapped. 32270374Sdillon */ 32395598Speter vm_page_zero_fill_area(m, base, size); 32470374Sdillon 32570374Sdillon /* 32687834Sdillon * XXX work around SMP data integrity race 32787834Sdillon * by unmapping the page from user processes. 32887834Sdillon * The garbage we just cleared may be mapped 32987834Sdillon * to a user process running on another cpu 33087834Sdillon * and this code is not running through normal 33187834Sdillon * I/O channels which handle SMP issues for 33287834Sdillon * us, so unmap page to synchronize all cpus. 33387834Sdillon * 33487834Sdillon * XXX should vm_pager_unmap_page() have 33587834Sdillon * dealt with this? 33687834Sdillon */ 33787834Sdillon vm_page_protect(m, VM_PROT_NONE); 33887834Sdillon 33987834Sdillon /* 34070374Sdillon * Clear out partial-page dirty bits. This 34170374Sdillon * has the side effect of setting the valid 34270374Sdillon * bits, but that is ok. There are a bunch 34370374Sdillon * of places in the VM system where we expected 34470374Sdillon * m->dirty == VM_PAGE_BITS_ALL. The file EOF 34570374Sdillon * case is one of them. If the page is still 34670374Sdillon * partially dirty, make it fully dirty. 34787834Sdillon * 34887834Sdillon * note that we do not clear out the valid 34987834Sdillon * bits. This would prevent bogus_page 35087834Sdillon * replacement from working properly. 35170374Sdillon */ 35270374Sdillon vm_page_set_validclean(m, base, size); 35370374Sdillon if (m->dirty != 0) 35470374Sdillon m->dirty = VM_PAGE_BITS_ALL; 3551827Sdg } 3561827Sdg } 3571541Srgrimes } 35812767Sdyson object->un_pager.vnp.vnp_size = nsize; 35938542Sluoqi object->size = nobjsize; 3601541Srgrimes} 3611541Srgrimes 3621549Srgrimes/* 3631549Srgrimes * calculate the linear (byte) disk address of specified virtual 3641549Srgrimes * file address 3651549Srgrimes */ 36612820Sphkstatic vm_offset_t 3676151Sdgvnode_pager_addr(vp, address, run) 3681549Srgrimes struct vnode *vp; 36912767Sdyson vm_ooffset_t address; 3706151Sdg int *run; 3711549Srgrimes{ 3725455Sdg int rtaddress; 3735455Sdg int bsize; 37496572Sphk daddr_t block; 3751549Srgrimes struct vnode *rtvp; 3765455Sdg int err; 37712767Sdyson daddr_t vblock; 37812767Sdyson int voffset; 3791549Srgrimes 38079224Sdillon GIANT_REQUIRED; 3815455Sdg if ((int) address < 0) 3825455Sdg return -1; 3835455Sdg 38411701Sdyson if (vp->v_mount == NULL) 38511701Sdyson return -1; 38611701Sdyson 3871549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 3881549Srgrimes vblock = address / bsize; 3891549Srgrimes voffset = address % bsize; 3901549Srgrimes 39110551Sdyson err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 3921549Srgrimes 3936151Sdg if (err || (block == -1)) 3941549Srgrimes rtaddress = -1; 3956151Sdg else { 3966626Sdg rtaddress = block + voffset / DEV_BSIZE; 39792029Seivind if (run) { 3986151Sdg *run += 1; 3996151Sdg *run *= bsize/PAGE_SIZE; 4006151Sdg *run -= voffset/PAGE_SIZE; 4016151Sdg } 4026151Sdg } 4031549Srgrimes 4041549Srgrimes return rtaddress; 4051549Srgrimes} 4061549Srgrimes 4071549Srgrimes/* 4081549Srgrimes * interrupt routine for I/O completion 4091549Srgrimes */ 41012820Sphkstatic void 4111549Srgrimesvnode_pager_iodone(bp) 4121549Srgrimes struct buf *bp; 4131549Srgrimes{ 4141549Srgrimes bp->b_flags |= B_DONE; 4159507Sdg wakeup(bp); 4161549Srgrimes} 4171549Srgrimes 4181549Srgrimes/* 41996755Strhodes * small block filesystem vnode pager input 4201549Srgrimes */ 42112820Sphkstatic int 4229507Sdgvnode_pager_input_smlfs(object, m) 4239507Sdg vm_object_t object; 4241549Srgrimes vm_page_t m; 4251549Srgrimes{ 4265455Sdg int i; 4275455Sdg int s; 4281549Srgrimes struct vnode *dp, *vp; 4291549Srgrimes struct buf *bp; 4301549Srgrimes vm_offset_t kva; 4315455Sdg int fileaddr; 4321549Srgrimes vm_offset_t bsize; 4335455Sdg int error = 0; 4341549Srgrimes 43579224Sdillon GIANT_REQUIRED; 43679224Sdillon 4379507Sdg vp = object->handle; 43811701Sdyson if (vp->v_mount == NULL) 43911701Sdyson return VM_PAGER_BAD; 44011701Sdyson 4411549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 4421549Srgrimes 44310551Sdyson VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 4441549Srgrimes 4451549Srgrimes kva = vm_pager_map_page(m); 4461549Srgrimes 4471827Sdg for (i = 0; i < PAGE_SIZE / bsize; i++) { 44886092Sdillon vm_ooffset_t address; 4491827Sdg 45045561Sdt if (vm_page_bits(i * bsize, bsize) & m->valid) 4515455Sdg continue; 4521549Srgrimes 45386092Sdillon address = IDX_TO_OFF(m->pindex) + i * bsize; 45486092Sdillon if (address >= object->un_pager.vnp.vnp_size) { 45586092Sdillon fileaddr = -1; 45686092Sdillon } else { 45786092Sdillon fileaddr = vnode_pager_addr(vp, address, NULL); 45886092Sdillon } 4591827Sdg if (fileaddr != -1) { 46042957Sdillon bp = getpbuf(&vnode_pbuf_freecnt); 4611549Srgrimes 4621827Sdg /* build a minimal buffer header */ 46358345Sphk bp->b_iocmd = BIO_READ; 4641549Srgrimes bp->b_iodone = vnode_pager_iodone; 46584827Sjhb KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 46684827Sjhb KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 46791406Sjhb bp->b_rcred = crhold(curthread->td_ucred); 46891406Sjhb bp->b_wcred = crhold(curthread->td_ucred); 46931493Sphk bp->b_data = (caddr_t) kva + i * bsize; 4706626Sdg bp->b_blkno = fileaddr; 4715455Sdg pbgetvp(dp, bp); 4721549Srgrimes bp->b_bcount = bsize; 4731549Srgrimes bp->b_bufsize = bsize; 47470374Sdillon bp->b_runningbufspace = bp->b_bufsize; 47570374Sdillon runningbufspace += bp->b_runningbufspace; 4761827Sdg 4771827Sdg /* do the input */ 47858349Sphk BUF_STRATEGY(bp); 4791549Srgrimes 48033758Sdyson /* we definitely need to be at splvm here */ 4811549Srgrimes 48233758Sdyson s = splvm(); 4831549Srgrimes while ((bp->b_flags & B_DONE) == 0) { 4849356Sdg tsleep(bp, PVM, "vnsrd", 0); 4851549Srgrimes } 4861549Srgrimes splx(s); 48758934Sphk if ((bp->b_ioflags & BIO_ERROR) != 0) 4881549Srgrimes error = EIO; 4891549Srgrimes 4901827Sdg /* 4911827Sdg * free the buffer header back to the swap buffer pool 4921827Sdg */ 49342957Sdillon relpbuf(bp, &vnode_pbuf_freecnt); 4941827Sdg if (error) 4951549Srgrimes break; 4965455Sdg 49715583Sphk vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 4981549Srgrimes } else { 49915583Sphk vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 5001549Srgrimes bzero((caddr_t) kva + i * bsize, bsize); 5011549Srgrimes } 5021549Srgrimes } 5031549Srgrimes vm_pager_unmap_page(kva); 50460755Speter pmap_clear_modify(m); 50538799Sdfr vm_page_flag_clear(m, PG_ZERO); 5061827Sdg if (error) { 5074207Sdg return VM_PAGER_ERROR; 5081549Srgrimes } 5091549Srgrimes return VM_PAGER_OK; 5101549Srgrimes 5111549Srgrimes} 5121549Srgrimes 5131549Srgrimes 5141549Srgrimes/* 5151549Srgrimes * old style vnode pager output routine 5161549Srgrimes */ 51712820Sphkstatic int 5189507Sdgvnode_pager_input_old(object, m) 5199507Sdg vm_object_t object; 5201549Srgrimes vm_page_t m; 5211549Srgrimes{ 5221541Srgrimes struct uio auio; 5231541Srgrimes struct iovec aiov; 5245455Sdg int error; 5255455Sdg int size; 5261549Srgrimes vm_offset_t kva; 52777398Sjhb struct vnode *vp; 5281549Srgrimes 52979224Sdillon GIANT_REQUIRED; 5301549Srgrimes error = 0; 5311827Sdg 5321549Srgrimes /* 5331549Srgrimes * Return failure if beyond current EOF 5341549Srgrimes */ 53512767Sdyson if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 5361549Srgrimes return VM_PAGER_BAD; 5371549Srgrimes } else { 5381549Srgrimes size = PAGE_SIZE; 53912767Sdyson if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 54012767Sdyson size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 5417178Sdg 5425455Sdg /* 5435455Sdg * Allocate a kernel virtual address and initialize so that 5445455Sdg * we can use VOP_READ/WRITE routines. 5455455Sdg */ 5461549Srgrimes kva = vm_pager_map_page(m); 5477178Sdg 54877398Sjhb vp = object->handle; 5491827Sdg aiov.iov_base = (caddr_t) kva; 5501549Srgrimes aiov.iov_len = size; 5511549Srgrimes auio.uio_iov = &aiov; 5521549Srgrimes auio.uio_iovcnt = 1; 55312767Sdyson auio.uio_offset = IDX_TO_OFF(m->pindex); 5541549Srgrimes auio.uio_segflg = UIO_SYSSPACE; 5551549Srgrimes auio.uio_rw = UIO_READ; 5561549Srgrimes auio.uio_resid = size; 55783366Sjulian auio.uio_td = curthread; 5581549Srgrimes 55991406Sjhb error = VOP_READ(vp, &auio, 0, curthread->td_ucred); 5601549Srgrimes if (!error) { 56179242Sdillon int count = size - auio.uio_resid; 5621549Srgrimes 5631549Srgrimes if (count == 0) 5641549Srgrimes error = EINVAL; 5651549Srgrimes else if (count != PAGE_SIZE) 5661827Sdg bzero((caddr_t) kva + count, PAGE_SIZE - count); 5671549Srgrimes } 5681549Srgrimes vm_pager_unmap_page(kva); 5691549Srgrimes } 57060755Speter pmap_clear_modify(m); 57149945Salc vm_page_undirty(m); 57238799Sdfr vm_page_flag_clear(m, PG_ZERO); 57339739Srvb if (!error) 57439739Srvb m->valid = VM_PAGE_BITS_ALL; 5754207Sdg return error ? VM_PAGER_ERROR : VM_PAGER_OK; 5761549Srgrimes} 5771549Srgrimes 5781549Srgrimes/* 5791549Srgrimes * generic vnode pager input routine 5801549Srgrimes */ 58110556Sdyson 58233847Smsmith/* 58376827Salfred * Local media VFS's that do not implement their own VOP_GETPAGES 58499211Srobert * should have their VOP_GETPAGES call to vnode_pager_generic_getpages() 58599211Srobert * to implement the previous behaviour. 58633847Smsmith * 58733847Smsmith * All other FS's should use the bypass to get to the local media 58833847Smsmith * backing vp's VOP_GETPAGES. 58933847Smsmith */ 59012820Sphkstatic int 5919507Sdgvnode_pager_getpages(object, m, count, reqpage) 5929507Sdg vm_object_t object; 5931549Srgrimes vm_page_t *m; 5949507Sdg int count; 5959507Sdg int reqpage; 5961549Srgrimes{ 59710556Sdyson int rtval; 59810556Sdyson struct vnode *vp; 59934403Smsmith int bytes = count * PAGE_SIZE; 60032286Sdyson 60179224Sdillon GIANT_REQUIRED; 60210556Sdyson vp = object->handle; 60334403Smsmith rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0); 60476827Salfred KASSERT(rtval != EOPNOTSUPP, 60576827Salfred ("vnode_pager: FS getpages not implemented\n")); 60633847Smsmith return rtval; 60710556Sdyson} 60810556Sdyson 60933847Smsmith/* 61033847Smsmith * This is now called from local media FS's to operate against their 61133847Smsmith * own vnodes if they fail to implement VOP_GETPAGES. 61233847Smsmith */ 61333847Smsmithint 61433847Smsmithvnode_pager_generic_getpages(vp, m, bytecount, reqpage) 61533847Smsmith struct vnode *vp; 61610556Sdyson vm_page_t *m; 61733847Smsmith int bytecount; 61810556Sdyson int reqpage; 61910556Sdyson{ 62033847Smsmith vm_object_t object; 62112767Sdyson vm_offset_t kva; 62234206Sdyson off_t foff, tfoff, nextoff; 623100832Salc int i, j, size, bsize, first, firstaddr; 62433847Smsmith struct vnode *dp; 6256151Sdg int runpg; 6266151Sdg int runend; 6277178Sdg struct buf *bp; 6285455Sdg int s; 62933847Smsmith int count; 6305455Sdg int error = 0; 6311549Srgrimes 63279224Sdillon GIANT_REQUIRED; 63333847Smsmith object = vp->v_object; 63433847Smsmith count = bytecount / PAGE_SIZE; 63533847Smsmith 63611701Sdyson if (vp->v_mount == NULL) 63711701Sdyson return VM_PAGER_BAD; 63811701Sdyson 6391549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 6401549Srgrimes 6411549Srgrimes /* get the UNDERLYING device for the file with VOP_BMAP() */ 6421827Sdg 6431549Srgrimes /* 6441827Sdg * originally, we did not check for an error return value -- assuming 6451827Sdg * an fs always has a bmap entry point -- that assumption is wrong!!! 6461549Srgrimes */ 64712767Sdyson foff = IDX_TO_OFF(m[reqpage]->pindex); 6481827Sdg 6491549Srgrimes /* 6501887Sdg * if we can't bmap, use old VOP code 6511549Srgrimes */ 65210551Sdyson if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 653100832Salc vm_page_lock_queues(); 654100832Salc for (i = 0; i < count; i++) 655100832Salc if (i != reqpage) 65675692Salfred vm_page_free(m[i]); 657100832Salc vm_page_unlock_queues(); 6583612Sdg cnt.v_vnodein++; 6593612Sdg cnt.v_vnodepgsin++; 6609507Sdg return vnode_pager_input_old(object, m[reqpage]); 6611549Srgrimes 6621827Sdg /* 6631827Sdg * if the blocksize is smaller than a page size, then use 6641827Sdg * special small filesystem code. NFS sometimes has a small 6651827Sdg * blocksize, but it can handle large reads itself. 6661827Sdg */ 6671827Sdg } else if ((PAGE_SIZE / bsize) > 1 && 66838866Sbde (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) { 669100832Salc vm_page_lock_queues(); 670100832Salc for (i = 0; i < count; i++) 671100832Salc if (i != reqpage) 67275692Salfred vm_page_free(m[i]); 673100832Salc vm_page_unlock_queues(); 6743612Sdg cnt.v_vnodein++; 6753612Sdg cnt.v_vnodepgsin++; 6769507Sdg return vnode_pager_input_smlfs(object, m[reqpage]); 6771549Srgrimes } 67845347Sjulian 6791549Srgrimes /* 68045347Sjulian * If we have a completely valid page available to us, we can 68145347Sjulian * clean up and return. Otherwise we have to re-read the 68245347Sjulian * media. 6831549Srgrimes */ 68445347Sjulian if (m[reqpage]->valid == VM_PAGE_BITS_ALL) { 685100832Salc vm_page_lock_queues(); 686100832Salc for (i = 0; i < count; i++) 6875455Sdg if (i != reqpage) 68875692Salfred vm_page_free(m[i]); 689100832Salc vm_page_unlock_queues(); 6905455Sdg return VM_PAGER_OK; 6911549Srgrimes } 69245347Sjulian m[reqpage]->valid = 0; 6937178Sdg 6945455Sdg /* 6955455Sdg * here on direct device I/O 6965455Sdg */ 69792029Seivind firstaddr = -1; 6981549Srgrimes 6991549Srgrimes /* 7006151Sdg * calculate the run that includes the required page 7011549Srgrimes */ 70292029Seivind for (first = 0, i = 0; i < count; i = runend) { 70312767Sdyson firstaddr = vnode_pager_addr(vp, 70412767Sdyson IDX_TO_OFF(m[i]->pindex), &runpg); 7056151Sdg if (firstaddr == -1) { 7069507Sdg if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 70737562Sbde /* XXX no %qd in kernel. */ 70837562Sbde panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx", 70937562Sbde firstaddr, (u_long)(foff >> 32), 71037562Sbde (u_long)(u_int32_t)foff, 71137562Sbde (u_long)(u_int32_t) 71237562Sbde (object->un_pager.vnp.vnp_size >> 32), 71337562Sbde (u_long)(u_int32_t) 71437562Sbde object->un_pager.vnp.vnp_size); 7156151Sdg } 716100832Salc vm_page_lock_queues(); 71775692Salfred vm_page_free(m[i]); 718100832Salc vm_page_unlock_queues(); 7196151Sdg runend = i + 1; 7206151Sdg first = runend; 7216151Sdg continue; 7221549Srgrimes } 7236151Sdg runend = i + runpg; 7249507Sdg if (runend <= reqpage) { 725100832Salc vm_page_lock_queues(); 726100832Salc for (j = i; j < runend; j++) 72775692Salfred vm_page_free(m[j]); 728100832Salc vm_page_unlock_queues(); 7291549Srgrimes } else { 7309507Sdg if (runpg < (count - first)) { 731100832Salc vm_page_lock_queues(); 7329507Sdg for (i = first + runpg; i < count; i++) 73375692Salfred vm_page_free(m[i]); 734100832Salc vm_page_unlock_queues(); 7356151Sdg count = first + runpg; 7366151Sdg } 7376151Sdg break; 7381549Srgrimes } 7396151Sdg first = runend; 7401549Srgrimes } 7411549Srgrimes 7421549Srgrimes /* 7431827Sdg * the first and last page have been calculated now, move input pages 7441827Sdg * to be zero based... 7451549Srgrimes */ 7461549Srgrimes if (first != 0) { 7471549Srgrimes for (i = first; i < count; i++) { 7481549Srgrimes m[i - first] = m[i]; 7491549Srgrimes } 7501549Srgrimes count -= first; 7511549Srgrimes reqpage -= first; 7521549Srgrimes } 7536151Sdg 7541549Srgrimes /* 7551549Srgrimes * calculate the file virtual address for the transfer 7561549Srgrimes */ 75712767Sdyson foff = IDX_TO_OFF(m[0]->pindex); 7581827Sdg 7591549Srgrimes /* 7601549Srgrimes * calculate the size of the transfer 7611549Srgrimes */ 7621549Srgrimes size = count * PAGE_SIZE; 7639507Sdg if ((foff + size) > object->un_pager.vnp.vnp_size) 7649507Sdg size = object->un_pager.vnp.vnp_size - foff; 7651549Srgrimes 7661549Srgrimes /* 76751340Sdillon * round up physical size for real devices. 7681549Srgrimes */ 76951340Sdillon if (dp->v_type == VBLK || dp->v_type == VCHR) { 77051340Sdillon int secmask = dp->v_rdev->si_bsize_phys - 1; 77151340Sdillon KASSERT(secmask < PAGE_SIZE, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask + 1)); 77251340Sdillon size = (size + secmask) & ~secmask; 77351340Sdillon } 7741549Srgrimes 77542957Sdillon bp = getpbuf(&vnode_pbuf_freecnt); 7765455Sdg kva = (vm_offset_t) bp->b_data; 7771887Sdg 7781549Srgrimes /* 7791549Srgrimes * and map the pages to be read into the kva 7801549Srgrimes */ 7811887Sdg pmap_qenter(kva, m, count); 7821549Srgrimes 7831549Srgrimes /* build a minimal buffer header */ 78458345Sphk bp->b_iocmd = BIO_READ; 7851549Srgrimes bp->b_iodone = vnode_pager_iodone; 7861549Srgrimes /* B_PHYS is not set, but it is nice to fill this in */ 78784827Sjhb KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 78884827Sjhb KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 78991406Sjhb bp->b_rcred = crhold(curthread->td_ucred); 79091406Sjhb bp->b_wcred = crhold(curthread->td_ucred); 7916626Sdg bp->b_blkno = firstaddr; 7925455Sdg pbgetvp(dp, bp); 7931549Srgrimes bp->b_bcount = size; 7941549Srgrimes bp->b_bufsize = size; 79570374Sdillon bp->b_runningbufspace = bp->b_bufsize; 79670374Sdillon runningbufspace += bp->b_runningbufspace; 7971549Srgrimes 7983612Sdg cnt.v_vnodein++; 7993612Sdg cnt.v_vnodepgsin += count; 8003612Sdg 8011549Srgrimes /* do the input */ 80258349Sphk BUF_STRATEGY(bp); 8033612Sdg 80433758Sdyson s = splvm(); 80533758Sdyson /* we definitely need to be at splvm here */ 8061549Srgrimes 8071549Srgrimes while ((bp->b_flags & B_DONE) == 0) { 8089356Sdg tsleep(bp, PVM, "vnread", 0); 8091549Srgrimes } 8101549Srgrimes splx(s); 81158934Sphk if ((bp->b_ioflags & BIO_ERROR) != 0) 8121549Srgrimes error = EIO; 8131549Srgrimes 8141549Srgrimes if (!error) { 8151549Srgrimes if (size != count * PAGE_SIZE) 8161827Sdg bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 8171549Srgrimes } 8185455Sdg pmap_qremove(kva, count); 8191549Srgrimes 8201549Srgrimes /* 8211549Srgrimes * free the buffer header back to the swap buffer pool 8221549Srgrimes */ 82342957Sdillon relpbuf(bp, &vnode_pbuf_freecnt); 8241549Srgrimes 825100736Salc vm_page_lock_queues(); 82634206Sdyson for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { 82734206Sdyson vm_page_t mt; 82834206Sdyson 82934206Sdyson nextoff = tfoff + PAGE_SIZE; 83034206Sdyson mt = m[i]; 83134206Sdyson 83247239Sdt if (nextoff <= object->un_pager.vnp.vnp_size) { 83345347Sjulian /* 83445347Sjulian * Read filled up entire page. 83545347Sjulian */ 83634206Sdyson mt->valid = VM_PAGE_BITS_ALL; 83749945Salc vm_page_undirty(mt); /* should be an assert? XXX */ 83860755Speter pmap_clear_modify(mt); 83934206Sdyson } else { 84045347Sjulian /* 84145347Sjulian * Read did not fill up entire page. Since this 84245347Sjulian * is getpages, the page may be mapped, so we have 84345347Sjulian * to zero the invalid portions of the page even 84445347Sjulian * though we aren't setting them valid. 84545347Sjulian * 84645347Sjulian * Currently we do not set the entire page valid, 84745347Sjulian * we just try to clear the piece that we couldn't 84845347Sjulian * read. 84945347Sjulian */ 85047239Sdt vm_page_set_validclean(mt, 0, 85147239Sdt object->un_pager.vnp.vnp_size - tfoff); 85246349Salc /* handled by vm_fault now */ 85346349Salc /* vm_page_zero_invalid(mt, FALSE); */ 85434206Sdyson } 85534206Sdyson 85638799Sdfr vm_page_flag_clear(mt, PG_ZERO); 8571549Srgrimes if (i != reqpage) { 8581827Sdg 8591549Srgrimes /* 8601827Sdg * whether or not to leave the page activated is up in 8611827Sdg * the air, but we should put the page on a page queue 8621827Sdg * somewhere. (it already is in the object). Result: 86358634Scharnier * It appears that empirical results show that 8641827Sdg * deactivating pages is best. 8651549Srgrimes */ 8661827Sdg 8671549Srgrimes /* 8681827Sdg * just in case someone was asking for this page we 8691827Sdg * now tell them that it is ok to use 8701549Srgrimes */ 8711549Srgrimes if (!error) { 87234206Sdyson if (mt->flags & PG_WANTED) 87334206Sdyson vm_page_activate(mt); 87433109Sdyson else 87534206Sdyson vm_page_deactivate(mt); 87638799Sdfr vm_page_wakeup(mt); 8771549Srgrimes } else { 87875692Salfred vm_page_free(mt); 8791549Srgrimes } 8801549Srgrimes } 8811549Srgrimes } 882100736Salc vm_page_unlock_queues(); 8831549Srgrimes if (error) { 8849507Sdg printf("vnode_pager_getpages: I/O read error\n"); 8851549Srgrimes } 8864207Sdg return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 8871549Srgrimes} 8881549Srgrimes 88933847Smsmith/* 89033847Smsmith * EOPNOTSUPP is no longer legal. For local media VFS's that do not 89133847Smsmith * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to 89233847Smsmith * vnode_pager_generic_putpages() to implement the previous behaviour. 89333847Smsmith * 89433847Smsmith * All other FS's should use the bypass to get to the local media 89533847Smsmith * backing vp's VOP_PUTPAGES. 89633847Smsmith */ 89743129Sdillonstatic void 89810556Sdysonvnode_pager_putpages(object, m, count, sync, rtvals) 89910556Sdyson vm_object_t object; 90010556Sdyson vm_page_t *m; 90110556Sdyson int count; 90210556Sdyson boolean_t sync; 90310556Sdyson int *rtvals; 90410556Sdyson{ 90510556Sdyson int rtval; 90610556Sdyson struct vnode *vp; 90762976Smckusick struct mount *mp; 90834403Smsmith int bytes = count * PAGE_SIZE; 90918973Sdyson 91079224Sdillon GIANT_REQUIRED; 91144321Salc /* 91244321Salc * Force synchronous operation if we are extremely low on memory 91344321Salc * to prevent a low-memory deadlock. VOP operations often need to 91444321Salc * allocate more memory to initiate the I/O ( i.e. do a BMAP 91544321Salc * operation ). The swapper handles the case by limiting the amount 91644321Salc * of asynchronous I/O, but that sort of solution doesn't scale well 91744321Salc * for the vnode pager without a lot of work. 91844321Salc * 91944321Salc * Also, the backing vnode's iodone routine may not wake the pageout 92044321Salc * daemon up. This should be probably be addressed XXX. 92144321Salc */ 92244321Salc 92344321Salc if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min) 92444321Salc sync |= OBJPC_SYNC; 92544321Salc 92644321Salc /* 92744321Salc * Call device-specific putpages function 92844321Salc */ 92910556Sdyson vp = object->handle; 93062976Smckusick if (vp->v_type != VREG) 93162976Smckusick mp = NULL; 93262976Smckusick (void)vn_start_write(vp, &mp, V_WAIT); 93334403Smsmith rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0); 93476827Salfred KASSERT(rtval != EOPNOTSUPP, 93576827Salfred ("vnode_pager: stale FS putpages\n")); 93662976Smckusick vn_finished_write(mp); 93710556Sdyson} 93810556Sdyson 93933847Smsmith 9401549Srgrimes/* 94133847Smsmith * This is now called from local media FS's to operate against their 94245057Seivind * own vnodes if they fail to implement VOP_PUTPAGES. 94370374Sdillon * 94470374Sdillon * This is typically called indirectly via the pageout daemon and 94570374Sdillon * clustering has already typically occured, so in general we ask the 94670374Sdillon * underlying filesystem to write the data out asynchronously rather 94770374Sdillon * then delayed. 9481549Srgrimes */ 94933847Smsmithint 95034206Sdysonvnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) 95133847Smsmith struct vnode *vp; 9521549Srgrimes vm_page_t *m; 95333847Smsmith int bytecount; 95434206Sdyson int flags; 9555455Sdg int *rtvals; 9561549Srgrimes{ 9577695Sdg int i; 95833847Smsmith vm_object_t object; 95933847Smsmith int count; 9601549Srgrimes 9617695Sdg int maxsize, ncount; 96212767Sdyson vm_ooffset_t poffset; 9637695Sdg struct uio auio; 9647695Sdg struct iovec aiov; 9657695Sdg int error; 96634206Sdyson int ioflags; 9671549Srgrimes 96879224Sdillon GIANT_REQUIRED; 96933847Smsmith object = vp->v_object; 97033847Smsmith count = bytecount / PAGE_SIZE; 97133847Smsmith 9721827Sdg for (i = 0; i < count; i++) 9731549Srgrimes rtvals[i] = VM_PAGER_AGAIN; 9741549Srgrimes 97512767Sdyson if ((int) m[0]->pindex < 0) { 97648409Speter printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n", 97748409Speter (long)m[0]->pindex, m[0]->dirty); 9787695Sdg rtvals[0] = VM_PAGER_BAD; 9797695Sdg return VM_PAGER_BAD; 9805455Sdg } 9817178Sdg 9827695Sdg maxsize = count * PAGE_SIZE; 9837695Sdg ncount = count; 9841549Srgrimes 98512767Sdyson poffset = IDX_TO_OFF(m[0]->pindex); 98684854Sdillon 98784854Sdillon /* 98884854Sdillon * If the page-aligned write is larger then the actual file we 98984854Sdillon * have to invalidate pages occuring beyond the file EOF. However, 99084854Sdillon * there is an edge case where a file may not be page-aligned where 99184854Sdillon * the last page is partially invalid. In this case the filesystem 99284854Sdillon * may not properly clear the dirty bits for the entire page (which 99384854Sdillon * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). 99484854Sdillon * With the page locked we are free to fix-up the dirty bits here. 99587834Sdillon * 99687834Sdillon * We do not under any circumstances truncate the valid bits, as 99787834Sdillon * this will screw up bogus page replacement. 99884854Sdillon */ 99912767Sdyson if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 100084854Sdillon if (object->un_pager.vnp.vnp_size > poffset) { 100184854Sdillon int pgoff; 100284854Sdillon 100312767Sdyson maxsize = object->un_pager.vnp.vnp_size - poffset; 100484854Sdillon ncount = btoc(maxsize); 100584854Sdillon if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { 100684854Sdillon vm_page_clear_dirty(m[ncount - 1], pgoff, 100784854Sdillon PAGE_SIZE - pgoff); 100884854Sdillon } 100984854Sdillon } else { 10108585Sdg maxsize = 0; 101184854Sdillon ncount = 0; 101284854Sdillon } 10138585Sdg if (ncount < count) { 10148585Sdg for (i = ncount; i < count; i++) { 10157695Sdg rtvals[i] = VM_PAGER_BAD; 10161549Srgrimes } 10171549Srgrimes } 10181541Srgrimes } 10197695Sdg 102070374Sdillon /* 102170374Sdillon * pageouts are already clustered, use IO_ASYNC t o force a bawrite() 102270374Sdillon * rather then a bdwrite() to prevent paging I/O from saturating 102370374Sdillon * the buffer cache. 102470374Sdillon */ 102534206Sdyson ioflags = IO_VMIO; 102670374Sdillon ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: IO_ASYNC; 102734206Sdyson ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; 10281827Sdg 10297695Sdg aiov.iov_base = (caddr_t) 0; 10307695Sdg aiov.iov_len = maxsize; 10317695Sdg auio.uio_iov = &aiov; 10327695Sdg auio.uio_iovcnt = 1; 103312767Sdyson auio.uio_offset = poffset; 10347695Sdg auio.uio_segflg = UIO_NOCOPY; 10357695Sdg auio.uio_rw = UIO_WRITE; 10367695Sdg auio.uio_resid = maxsize; 103783366Sjulian auio.uio_td = (struct thread *) 0; 103891406Sjhb error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); 10393612Sdg cnt.v_vnodeout++; 10407695Sdg cnt.v_vnodepgsout += ncount; 10413612Sdg 10428585Sdg if (error) { 10439507Sdg printf("vnode_pager_putpages: I/O error %d\n", error); 10447695Sdg } 10458585Sdg if (auio.uio_resid) { 104637555Sbde printf("vnode_pager_putpages: residual I/O %d at %lu\n", 104737555Sbde auio.uio_resid, (u_long)m[0]->pindex); 10487695Sdg } 104933936Sdyson for (i = 0; i < ncount; i++) { 105033936Sdyson rtvals[i] = VM_PAGER_OK; 10517695Sdg } 10527695Sdg return rtvals[0]; 10537695Sdg} 10541549Srgrimes 10557695Sdgstruct vnode * 10569507Sdgvnode_pager_lock(object) 10579507Sdg vm_object_t object; 10589507Sdg{ 105983366Sjulian struct thread *td = curthread; /* XXX */ 106022521Sdyson 106179224Sdillon GIANT_REQUIRED; 106279224Sdillon 10639507Sdg for (; object != NULL; object = object->backing_object) { 10649507Sdg if (object->type != OBJT_VNODE) 10657695Sdg continue; 106677094Sjhb if (object->flags & OBJ_DEAD) { 106732585Sdyson return NULL; 106877094Sjhb } 10691549Srgrimes 107077094Sjhb /* XXX; If object->handle can change, we need to cache it. */ 107132585Sdyson while (vget(object->handle, 107283366Sjulian LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, td)){ 107334611Sdyson if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE)) 107434611Sdyson return NULL; 107532585Sdyson printf("vnode_pager_lock: retrying\n"); 107632585Sdyson } 10779507Sdg return object->handle; 10781549Srgrimes } 10799507Sdg return NULL; 10807695Sdg} 1081