vnode_pager.c revision 116167
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 116167 2003-06-10 20:28:41Z 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 95104094Sphkstatic void 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 123103923Sjeff ASSERT_VOP_LOCKED(vp, "vnode_pager_alloc"); 124103923Sjeff 12598604Salc mtx_lock(&Giant); 1261541Srgrimes /* 1279411Sdg * Prevent race condition when allocating the object. This 1289411Sdg * can happen with NFS vnodes since the nfsnode isn't locked. 1291541Srgrimes */ 130101308Sjeff VI_LOCK(vp); 131101308Sjeff while (vp->v_iflag & VI_OLOCK) { 132101308Sjeff vp->v_iflag |= VI_OWANT; 133101308Sjeff msleep(vp, VI_MTX(vp), PVM, "vnpobj", 0); 1349411Sdg } 135101308Sjeff vp->v_iflag |= VI_OLOCK; 136101308Sjeff VI_UNLOCK(vp); 1379411Sdg 1389411Sdg /* 1399411Sdg * If the object is being terminated, wait for it to 1409411Sdg * go away. 1419411Sdg */ 142114074Salc while ((object = vp->v_object) != NULL) { 143114074Salc VM_OBJECT_LOCK(object); 144114074Salc if ((object->flags & OBJ_DEAD) == 0) 145114074Salc break; 146114074Salc msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0); 1479507Sdg } 1485455Sdg 14932071Sdyson if (vp->v_usecount == 0) 15032071Sdyson panic("vnode_pager_alloc: no vnode reference"); 15132071Sdyson 1529507Sdg if (object == NULL) { 1531541Srgrimes /* 1541541Srgrimes * And an object of the appropriate size 1551541Srgrimes */ 15640286Sdg object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); 1571827Sdg 15840286Sdg object->un_pager.vnp.vnp_size = size; 1591549Srgrimes 1609507Sdg object->handle = handle; 1619507Sdg vp->v_object = object; 1621541Srgrimes } else { 16332286Sdyson object->ref_count++; 164114074Salc VM_OBJECT_UNLOCK(object); 1651541Srgrimes } 166101308Sjeff VI_LOCK(vp); 16798604Salc vp->v_usecount++; 168101308Sjeff vp->v_iflag &= ~VI_OLOCK; 169101308Sjeff if (vp->v_iflag & VI_OWANT) { 170101308Sjeff vp->v_iflag &= ~VI_OWANT; 1719411Sdg wakeup(vp); 1729411Sdg } 173101308Sjeff VI_UNLOCK(vp); 17498604Salc mtx_unlock(&Giant); 1759507Sdg return (object); 1761541Srgrimes} 1771541Srgrimes 178114774Salc/* 179114774Salc * The object must be locked. 180114774Salc */ 18112820Sphkstatic void 1829507Sdgvnode_pager_dealloc(object) 1839507Sdg vm_object_t object; 1841541Srgrimes{ 18579242Sdillon struct vnode *vp = object->handle; 1861541Srgrimes 1879507Sdg if (vp == NULL) 1889507Sdg panic("vnode_pager_dealloc: pager already dealloced"); 1899507Sdg 190114774Salc VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 19133817Sdyson vm_object_pip_wait(object, "vnpdea"); 1921541Srgrimes 1939507Sdg object->handle = NULL; 19433109Sdyson object->type = OBJT_DEAD; 195101308Sjeff ASSERT_VOP_LOCKED(vp, "vnode_pager_dealloc"); 1969507Sdg vp->v_object = NULL; 197101308Sjeff vp->v_vflag &= ~(VV_TEXT | VV_OBJBUF); 1981549Srgrimes} 1991541Srgrimes 20012820Sphkstatic boolean_t 20112767Sdysonvnode_pager_haspage(object, pindex, before, after) 2029507Sdg vm_object_t object; 20312767Sdyson vm_pindex_t pindex; 2049507Sdg int *before; 2059507Sdg int *after; 2061541Srgrimes{ 2079507Sdg struct vnode *vp = object->handle; 20896572Sphk daddr_t bn; 20912423Sphk int err; 21010556Sdyson daddr_t reqblock; 21111701Sdyson int poff; 21211701Sdyson int bsize; 21312914Sdyson int pagesperblock, blocksperpage; 2141541Srgrimes 21579224Sdillon GIANT_REQUIRED; 21651340Sdillon /* 21751340Sdillon * If no vp or vp is doomed or marked transparent to VM, we do not 21851340Sdillon * have the page. 21951340Sdillon */ 220101308Sjeff if (vp == NULL) 22132585Sdyson return FALSE; 22232585Sdyson 223103923Sjeff VI_LOCK(vp); 224103923Sjeff if (vp->v_iflag & VI_DOOMED) { 225103923Sjeff VI_UNLOCK(vp); 226101308Sjeff return FALSE; 227103923Sjeff } 228103923Sjeff VI_UNLOCK(vp); 2291541Srgrimes /* 2305455Sdg * If filesystem no longer mounted or offset beyond end of file we do 2315455Sdg * not have the page. 2321541Srgrimes */ 23312767Sdyson if ((vp->v_mount == NULL) || 23481140Sjhb (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 2354797Sdg return FALSE; 2361541Srgrimes 23711576Sdg bsize = vp->v_mount->mnt_stat.f_iosize; 23810556Sdyson pagesperblock = bsize / PAGE_SIZE; 23912914Sdyson blocksperpage = 0; 24012914Sdyson if (pagesperblock > 0) { 24112914Sdyson reqblock = pindex / pagesperblock; 24212914Sdyson } else { 24312914Sdyson blocksperpage = (PAGE_SIZE / bsize); 24412914Sdyson reqblock = pindex * blocksperpage; 24512914Sdyson } 24610556Sdyson err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 24710556Sdyson after, before); 2488876Srgrimes if (err) 2499507Sdg return TRUE; 25092029Seivind if (bn == -1) 25110576Sdyson return FALSE; 25212914Sdyson if (pagesperblock > 0) { 25312914Sdyson poff = pindex - (reqblock * pagesperblock); 25412914Sdyson if (before) { 25512914Sdyson *before *= pagesperblock; 25612914Sdyson *before += poff; 25710669Sdyson } 25812914Sdyson if (after) { 25912914Sdyson int numafter; 26012914Sdyson *after *= pagesperblock; 26112914Sdyson numafter = pagesperblock - (poff + 1); 26299211Srobert if (IDX_TO_OFF(pindex + numafter) > 26399211Srobert object->un_pager.vnp.vnp_size) { 26499211Srobert numafter = 26599211Srobert OFF_TO_IDX(object->un_pager.vnp.vnp_size) - 26699211Srobert pindex; 26712914Sdyson } 26812914Sdyson *after += numafter; 26912914Sdyson } 27012914Sdyson } else { 27112914Sdyson if (before) { 27212914Sdyson *before /= blocksperpage; 27312914Sdyson } 27412914Sdyson 27512914Sdyson if (after) { 27612914Sdyson *after /= blocksperpage; 27712914Sdyson } 27810556Sdyson } 27910576Sdyson return TRUE; 2801541Srgrimes} 2811541Srgrimes 2821541Srgrimes/* 2831541Srgrimes * Lets the VM system know about a change in size for a file. 2849507Sdg * We adjust our own internal size and flush any cached pages in 2851541Srgrimes * the associated object that are affected by the size change. 2861541Srgrimes * 2871541Srgrimes * Note: this routine may be invoked as a result of a pager put 2881541Srgrimes * operation (possibly at object termination time), so we must be careful. 2891541Srgrimes */ 2901541Srgrimesvoid 2911541Srgrimesvnode_pager_setsize(vp, nsize) 2921541Srgrimes struct vnode *vp; 29312767Sdyson vm_ooffset_t nsize; 2941541Srgrimes{ 295116167Salc vm_object_t object; 296116167Salc vm_page_t m; 29738542Sluoqi vm_pindex_t nobjsize; 2981541Srgrimes 299116167Salc if ((object = vp->v_object) == NULL) 3001541Srgrimes return; 301116167Salc VM_OBJECT_LOCK(object); 302116167Salc if (nsize == object->un_pager.vnp.vnp_size) { 303116167Salc /* 304116167Salc * Hasn't changed size 305116167Salc */ 306116167Salc VM_OBJECT_UNLOCK(object); 3073374Sdg return; 308116167Salc } 30938542Sluoqi nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); 3109507Sdg if (nsize < object->un_pager.vnp.vnp_size) { 311116167Salc /* 312116167Salc * File has shrunk. Toss any cached pages beyond the new EOF. 313116167Salc */ 314116167Salc if (nobjsize < object->size) 31538542Sluoqi vm_object_page_remove(object, nobjsize, object->size, 316116167Salc FALSE); 3171827Sdg /* 3181827Sdg * this gets rid of garbage at the end of a page that is now 31987834Sdillon * only partially backed by the vnode. 32087834Sdillon * 32187834Sdillon * XXX for some reason (I don't know yet), if we take a 32287834Sdillon * completely invalid page and mark it partially valid 32387834Sdillon * it can screw up NFS reads, so we don't allow the case. 3241827Sdg */ 325116167Salc if ((nsize & PAGE_MASK) && 326116167Salc (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL) { 327116167Salc vm_page_lock_queues(); 328116167Salc if (m->valid) { 32970374Sdillon int base = (int)nsize & PAGE_MASK; 33070374Sdillon int size = PAGE_SIZE - base; 33170374Sdillon 33270374Sdillon /* 33370374Sdillon * Clear out partial-page garbage in case 33470374Sdillon * the page has been mapped. 33570374Sdillon */ 336102382Salc pmap_zero_page_area(m, base, size); 33770374Sdillon 33870374Sdillon /* 33987834Sdillon * XXX work around SMP data integrity race 34087834Sdillon * by unmapping the page from user processes. 34187834Sdillon * The garbage we just cleared may be mapped 34287834Sdillon * to a user process running on another cpu 34387834Sdillon * and this code is not running through normal 34487834Sdillon * I/O channels which handle SMP issues for 34587834Sdillon * us, so unmap page to synchronize all cpus. 34687834Sdillon * 34787834Sdillon * XXX should vm_pager_unmap_page() have 34887834Sdillon * dealt with this? 34987834Sdillon */ 350106981Salc pmap_remove_all(m); 35187834Sdillon 35287834Sdillon /* 35370374Sdillon * Clear out partial-page dirty bits. This 35470374Sdillon * has the side effect of setting the valid 35570374Sdillon * bits, but that is ok. There are a bunch 35670374Sdillon * of places in the VM system where we expected 35770374Sdillon * m->dirty == VM_PAGE_BITS_ALL. The file EOF 35870374Sdillon * case is one of them. If the page is still 35970374Sdillon * partially dirty, make it fully dirty. 36087834Sdillon * 36187834Sdillon * note that we do not clear out the valid 36287834Sdillon * bits. This would prevent bogus_page 36387834Sdillon * replacement from working properly. 36470374Sdillon */ 36570374Sdillon vm_page_set_validclean(m, base, size); 36670374Sdillon if (m->dirty != 0) 36770374Sdillon m->dirty = VM_PAGE_BITS_ALL; 3681827Sdg } 369116167Salc vm_page_unlock_queues(); 3701827Sdg } 3711541Srgrimes } 37212767Sdyson object->un_pager.vnp.vnp_size = nsize; 37338542Sluoqi object->size = nobjsize; 374116167Salc VM_OBJECT_UNLOCK(object); 3751541Srgrimes} 3761541Srgrimes 3771549Srgrimes/* 3781549Srgrimes * calculate the linear (byte) disk address of specified virtual 3791549Srgrimes * file address 3801549Srgrimes */ 38112820Sphkstatic vm_offset_t 3826151Sdgvnode_pager_addr(vp, address, run) 3831549Srgrimes struct vnode *vp; 38412767Sdyson vm_ooffset_t address; 3856151Sdg int *run; 3861549Srgrimes{ 3875455Sdg int rtaddress; 3885455Sdg int bsize; 38996572Sphk daddr_t block; 3901549Srgrimes struct vnode *rtvp; 3915455Sdg int err; 39212767Sdyson daddr_t vblock; 39312767Sdyson int voffset; 3941549Srgrimes 39579224Sdillon GIANT_REQUIRED; 3965455Sdg if ((int) address < 0) 3975455Sdg return -1; 3985455Sdg 39911701Sdyson if (vp->v_mount == NULL) 40011701Sdyson return -1; 40111701Sdyson 4021549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 4031549Srgrimes vblock = address / bsize; 4041549Srgrimes voffset = address % bsize; 4051549Srgrimes 40610551Sdyson err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 4071549Srgrimes 4086151Sdg if (err || (block == -1)) 4091549Srgrimes rtaddress = -1; 4106151Sdg else { 4116626Sdg rtaddress = block + voffset / DEV_BSIZE; 41292029Seivind if (run) { 4136151Sdg *run += 1; 4146151Sdg *run *= bsize/PAGE_SIZE; 4156151Sdg *run -= voffset/PAGE_SIZE; 4166151Sdg } 4176151Sdg } 4181549Srgrimes 4191549Srgrimes return rtaddress; 4201549Srgrimes} 4211549Srgrimes 4221549Srgrimes/* 4231549Srgrimes * interrupt routine for I/O completion 4241549Srgrimes */ 42512820Sphkstatic void 4261549Srgrimesvnode_pager_iodone(bp) 4271549Srgrimes struct buf *bp; 4281549Srgrimes{ 4291549Srgrimes bp->b_flags |= B_DONE; 4309507Sdg wakeup(bp); 4311549Srgrimes} 4321549Srgrimes 4331549Srgrimes/* 43496755Strhodes * small block filesystem vnode pager input 4351549Srgrimes */ 43612820Sphkstatic int 4379507Sdgvnode_pager_input_smlfs(object, m) 4389507Sdg vm_object_t object; 4391549Srgrimes vm_page_t m; 4401549Srgrimes{ 4415455Sdg int i; 4425455Sdg int s; 4431549Srgrimes struct vnode *dp, *vp; 4441549Srgrimes struct buf *bp; 4451549Srgrimes vm_offset_t kva; 4465455Sdg int fileaddr; 4471549Srgrimes vm_offset_t bsize; 4485455Sdg int error = 0; 4491549Srgrimes 45079224Sdillon GIANT_REQUIRED; 45179224Sdillon 4529507Sdg vp = object->handle; 45311701Sdyson if (vp->v_mount == NULL) 45411701Sdyson return VM_PAGER_BAD; 45511701Sdyson 4561549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 4571549Srgrimes 45810551Sdyson VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 4591549Srgrimes 4601549Srgrimes kva = vm_pager_map_page(m); 4611549Srgrimes 4621827Sdg for (i = 0; i < PAGE_SIZE / bsize; i++) { 46386092Sdillon vm_ooffset_t address; 4641827Sdg 46545561Sdt if (vm_page_bits(i * bsize, bsize) & m->valid) 4665455Sdg continue; 4671549Srgrimes 46886092Sdillon address = IDX_TO_OFF(m->pindex) + i * bsize; 46986092Sdillon if (address >= object->un_pager.vnp.vnp_size) { 47086092Sdillon fileaddr = -1; 47186092Sdillon } else { 47286092Sdillon fileaddr = vnode_pager_addr(vp, address, NULL); 47386092Sdillon } 4741827Sdg if (fileaddr != -1) { 47542957Sdillon bp = getpbuf(&vnode_pbuf_freecnt); 4761549Srgrimes 4771827Sdg /* build a minimal buffer header */ 47858345Sphk bp->b_iocmd = BIO_READ; 4791549Srgrimes bp->b_iodone = vnode_pager_iodone; 48084827Sjhb KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 48184827Sjhb KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 48291406Sjhb bp->b_rcred = crhold(curthread->td_ucred); 48391406Sjhb bp->b_wcred = crhold(curthread->td_ucred); 48431493Sphk bp->b_data = (caddr_t) kva + i * bsize; 4856626Sdg bp->b_blkno = fileaddr; 4865455Sdg pbgetvp(dp, bp); 4871549Srgrimes bp->b_bcount = bsize; 4881549Srgrimes bp->b_bufsize = bsize; 48970374Sdillon bp->b_runningbufspace = bp->b_bufsize; 49070374Sdillon runningbufspace += bp->b_runningbufspace; 4911827Sdg 4921827Sdg /* do the input */ 493108723Sphk VOP_SPECSTRATEGY(bp->b_vp, bp); 4941549Srgrimes 49533758Sdyson /* we definitely need to be at splvm here */ 4961549Srgrimes 49733758Sdyson s = splvm(); 4981549Srgrimes while ((bp->b_flags & B_DONE) == 0) { 4999356Sdg tsleep(bp, PVM, "vnsrd", 0); 5001549Srgrimes } 5011549Srgrimes splx(s); 50258934Sphk if ((bp->b_ioflags & BIO_ERROR) != 0) 5031549Srgrimes error = EIO; 5041549Srgrimes 5051827Sdg /* 5061827Sdg * free the buffer header back to the swap buffer pool 5071827Sdg */ 50842957Sdillon relpbuf(bp, &vnode_pbuf_freecnt); 5091827Sdg if (error) 5101549Srgrimes break; 5115455Sdg 512107189Salc vm_page_lock_queues(); 51315583Sphk vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 514107189Salc vm_page_unlock_queues(); 5151549Srgrimes } else { 516107189Salc vm_page_lock_queues(); 51715583Sphk vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 518107189Salc vm_page_unlock_queues(); 5191549Srgrimes bzero((caddr_t) kva + i * bsize, bsize); 5201549Srgrimes } 5211549Srgrimes } 5221549Srgrimes vm_pager_unmap_page(kva); 523107347Salc vm_page_lock_queues(); 52460755Speter pmap_clear_modify(m); 52538799Sdfr vm_page_flag_clear(m, PG_ZERO); 526107347Salc vm_page_unlock_queues(); 5271827Sdg if (error) { 5284207Sdg return VM_PAGER_ERROR; 5291549Srgrimes } 5301549Srgrimes return VM_PAGER_OK; 5311549Srgrimes 5321549Srgrimes} 5331549Srgrimes 5341549Srgrimes 5351549Srgrimes/* 5361549Srgrimes * old style vnode pager output routine 5371549Srgrimes */ 53812820Sphkstatic int 5399507Sdgvnode_pager_input_old(object, m) 5409507Sdg vm_object_t object; 5411549Srgrimes vm_page_t m; 5421549Srgrimes{ 5431541Srgrimes struct uio auio; 5441541Srgrimes struct iovec aiov; 5455455Sdg int error; 5465455Sdg int size; 5471549Srgrimes vm_offset_t kva; 54877398Sjhb struct vnode *vp; 5491549Srgrimes 55079224Sdillon GIANT_REQUIRED; 5511549Srgrimes error = 0; 5521827Sdg 5531549Srgrimes /* 5541549Srgrimes * Return failure if beyond current EOF 5551549Srgrimes */ 55612767Sdyson if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 5571549Srgrimes return VM_PAGER_BAD; 5581549Srgrimes } else { 5591549Srgrimes size = PAGE_SIZE; 56012767Sdyson if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 56112767Sdyson size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 5627178Sdg 5635455Sdg /* 5645455Sdg * Allocate a kernel virtual address and initialize so that 5655455Sdg * we can use VOP_READ/WRITE routines. 5665455Sdg */ 5671549Srgrimes kva = vm_pager_map_page(m); 5687178Sdg 56977398Sjhb vp = object->handle; 5701827Sdg aiov.iov_base = (caddr_t) kva; 5711549Srgrimes aiov.iov_len = size; 5721549Srgrimes auio.uio_iov = &aiov; 5731549Srgrimes auio.uio_iovcnt = 1; 57412767Sdyson auio.uio_offset = IDX_TO_OFF(m->pindex); 5751549Srgrimes auio.uio_segflg = UIO_SYSSPACE; 5761549Srgrimes auio.uio_rw = UIO_READ; 5771549Srgrimes auio.uio_resid = size; 57883366Sjulian auio.uio_td = curthread; 5791549Srgrimes 58091406Sjhb error = VOP_READ(vp, &auio, 0, curthread->td_ucred); 5811549Srgrimes if (!error) { 58279242Sdillon int count = size - auio.uio_resid; 5831549Srgrimes 5841549Srgrimes if (count == 0) 5851549Srgrimes error = EINVAL; 5861549Srgrimes else if (count != PAGE_SIZE) 5871827Sdg bzero((caddr_t) kva + count, PAGE_SIZE - count); 5881549Srgrimes } 5891549Srgrimes vm_pager_unmap_page(kva); 5901549Srgrimes } 591107347Salc vm_page_lock_queues(); 59260755Speter pmap_clear_modify(m); 59349945Salc vm_page_undirty(m); 59438799Sdfr vm_page_flag_clear(m, PG_ZERO); 59539739Srvb if (!error) 59639739Srvb m->valid = VM_PAGE_BITS_ALL; 597107347Salc vm_page_unlock_queues(); 5984207Sdg return error ? VM_PAGER_ERROR : VM_PAGER_OK; 5991549Srgrimes} 6001549Srgrimes 6011549Srgrimes/* 6021549Srgrimes * generic vnode pager input routine 6031549Srgrimes */ 60410556Sdyson 60533847Smsmith/* 60676827Salfred * Local media VFS's that do not implement their own VOP_GETPAGES 60799211Srobert * should have their VOP_GETPAGES call to vnode_pager_generic_getpages() 60899211Srobert * to implement the previous behaviour. 60933847Smsmith * 61033847Smsmith * All other FS's should use the bypass to get to the local media 61133847Smsmith * backing vp's VOP_GETPAGES. 61233847Smsmith */ 61312820Sphkstatic int 6149507Sdgvnode_pager_getpages(object, m, count, reqpage) 6159507Sdg vm_object_t object; 6161549Srgrimes vm_page_t *m; 6179507Sdg int count; 6189507Sdg int reqpage; 6191549Srgrimes{ 62010556Sdyson int rtval; 62110556Sdyson struct vnode *vp; 62234403Smsmith int bytes = count * PAGE_SIZE; 62332286Sdyson 62479224Sdillon GIANT_REQUIRED; 62510556Sdyson vp = object->handle; 62634403Smsmith rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0); 62776827Salfred KASSERT(rtval != EOPNOTSUPP, 62876827Salfred ("vnode_pager: FS getpages not implemented\n")); 62933847Smsmith return rtval; 63010556Sdyson} 63110556Sdyson 63233847Smsmith/* 63333847Smsmith * This is now called from local media FS's to operate against their 63433847Smsmith * own vnodes if they fail to implement VOP_GETPAGES. 63533847Smsmith */ 63633847Smsmithint 63733847Smsmithvnode_pager_generic_getpages(vp, m, bytecount, reqpage) 63833847Smsmith struct vnode *vp; 63910556Sdyson vm_page_t *m; 64033847Smsmith int bytecount; 64110556Sdyson int reqpage; 64210556Sdyson{ 64333847Smsmith vm_object_t object; 64412767Sdyson vm_offset_t kva; 64534206Sdyson off_t foff, tfoff, nextoff; 646100832Salc int i, j, size, bsize, first, firstaddr; 64733847Smsmith struct vnode *dp; 6486151Sdg int runpg; 6496151Sdg int runend; 6507178Sdg struct buf *bp; 6515455Sdg int s; 65233847Smsmith int count; 6535455Sdg int error = 0; 6541549Srgrimes 65579224Sdillon GIANT_REQUIRED; 65633847Smsmith object = vp->v_object; 65733847Smsmith count = bytecount / PAGE_SIZE; 65833847Smsmith 65911701Sdyson if (vp->v_mount == NULL) 66011701Sdyson return VM_PAGER_BAD; 66111701Sdyson 6621549Srgrimes bsize = vp->v_mount->mnt_stat.f_iosize; 6631549Srgrimes 6641549Srgrimes /* get the UNDERLYING device for the file with VOP_BMAP() */ 6651827Sdg 6661549Srgrimes /* 6671827Sdg * originally, we did not check for an error return value -- assuming 6681827Sdg * an fs always has a bmap entry point -- that assumption is wrong!!! 6691549Srgrimes */ 67012767Sdyson foff = IDX_TO_OFF(m[reqpage]->pindex); 6711827Sdg 6721549Srgrimes /* 6731887Sdg * if we can't bmap, use old VOP code 6741549Srgrimes */ 67510551Sdyson if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 676100832Salc vm_page_lock_queues(); 677100832Salc for (i = 0; i < count; i++) 678100832Salc if (i != reqpage) 67975692Salfred vm_page_free(m[i]); 680100832Salc vm_page_unlock_queues(); 6813612Sdg cnt.v_vnodein++; 6823612Sdg cnt.v_vnodepgsin++; 6839507Sdg return vnode_pager_input_old(object, m[reqpage]); 6841549Srgrimes 6851827Sdg /* 6861827Sdg * if the blocksize is smaller than a page size, then use 6871827Sdg * special small filesystem code. NFS sometimes has a small 6881827Sdg * blocksize, but it can handle large reads itself. 6891827Sdg */ 6901827Sdg } else if ((PAGE_SIZE / bsize) > 1 && 69138866Sbde (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) { 692100832Salc vm_page_lock_queues(); 693100832Salc for (i = 0; i < count; i++) 694100832Salc if (i != reqpage) 69575692Salfred vm_page_free(m[i]); 696100832Salc vm_page_unlock_queues(); 6973612Sdg cnt.v_vnodein++; 6983612Sdg cnt.v_vnodepgsin++; 6999507Sdg return vnode_pager_input_smlfs(object, m[reqpage]); 7001549Srgrimes } 70145347Sjulian 7021549Srgrimes /* 70345347Sjulian * If we have a completely valid page available to us, we can 70445347Sjulian * clean up and return. Otherwise we have to re-read the 70545347Sjulian * media. 7061549Srgrimes */ 70745347Sjulian if (m[reqpage]->valid == VM_PAGE_BITS_ALL) { 708100832Salc vm_page_lock_queues(); 709100832Salc for (i = 0; i < count; i++) 7105455Sdg if (i != reqpage) 71175692Salfred vm_page_free(m[i]); 712100832Salc vm_page_unlock_queues(); 7135455Sdg return VM_PAGER_OK; 7141549Srgrimes } 71545347Sjulian m[reqpage]->valid = 0; 7167178Sdg 7175455Sdg /* 7185455Sdg * here on direct device I/O 7195455Sdg */ 72092029Seivind firstaddr = -1; 7211549Srgrimes 7221549Srgrimes /* 7236151Sdg * calculate the run that includes the required page 7241549Srgrimes */ 72592029Seivind for (first = 0, i = 0; i < count; i = runend) { 72612767Sdyson firstaddr = vnode_pager_addr(vp, 72712767Sdyson IDX_TO_OFF(m[i]->pindex), &runpg); 7286151Sdg if (firstaddr == -1) { 7299507Sdg if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 730106603Smux panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx", 731106603Smux firstaddr, (uintmax_t)(foff >> 32), 732106603Smux (uintmax_t)foff, 733106603Smux (uintmax_t) 734106603Smux (object->un_pager.vnp.vnp_size >> 32), 735106603Smux (uintmax_t)object->un_pager.vnp.vnp_size); 7366151Sdg } 737100832Salc vm_page_lock_queues(); 73875692Salfred vm_page_free(m[i]); 739100832Salc vm_page_unlock_queues(); 7406151Sdg runend = i + 1; 7416151Sdg first = runend; 7426151Sdg continue; 7431549Srgrimes } 7446151Sdg runend = i + runpg; 7459507Sdg if (runend <= reqpage) { 746100832Salc vm_page_lock_queues(); 747100832Salc for (j = i; j < runend; j++) 74875692Salfred vm_page_free(m[j]); 749100832Salc vm_page_unlock_queues(); 7501549Srgrimes } else { 7519507Sdg if (runpg < (count - first)) { 752100832Salc vm_page_lock_queues(); 7539507Sdg for (i = first + runpg; i < count; i++) 75475692Salfred vm_page_free(m[i]); 755100832Salc vm_page_unlock_queues(); 7566151Sdg count = first + runpg; 7576151Sdg } 7586151Sdg break; 7591549Srgrimes } 7606151Sdg first = runend; 7611549Srgrimes } 7621549Srgrimes 7631549Srgrimes /* 7641827Sdg * the first and last page have been calculated now, move input pages 7651827Sdg * to be zero based... 7661549Srgrimes */ 7671549Srgrimes if (first != 0) { 7681549Srgrimes for (i = first; i < count; i++) { 7691549Srgrimes m[i - first] = m[i]; 7701549Srgrimes } 7711549Srgrimes count -= first; 7721549Srgrimes reqpage -= first; 7731549Srgrimes } 7746151Sdg 7751549Srgrimes /* 7761549Srgrimes * calculate the file virtual address for the transfer 7771549Srgrimes */ 77812767Sdyson foff = IDX_TO_OFF(m[0]->pindex); 7791827Sdg 7801549Srgrimes /* 7811549Srgrimes * calculate the size of the transfer 7821549Srgrimes */ 7831549Srgrimes size = count * PAGE_SIZE; 7849507Sdg if ((foff + size) > object->un_pager.vnp.vnp_size) 7859507Sdg size = object->un_pager.vnp.vnp_size - foff; 7861549Srgrimes 7871549Srgrimes /* 78851340Sdillon * round up physical size for real devices. 7891549Srgrimes */ 79051340Sdillon if (dp->v_type == VBLK || dp->v_type == VCHR) { 79151340Sdillon int secmask = dp->v_rdev->si_bsize_phys - 1; 79251340Sdillon KASSERT(secmask < PAGE_SIZE, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask + 1)); 79351340Sdillon size = (size + secmask) & ~secmask; 79451340Sdillon } 7951549Srgrimes 79642957Sdillon bp = getpbuf(&vnode_pbuf_freecnt); 7975455Sdg kva = (vm_offset_t) bp->b_data; 7981887Sdg 7991549Srgrimes /* 8001549Srgrimes * and map the pages to be read into the kva 8011549Srgrimes */ 8021887Sdg pmap_qenter(kva, m, count); 8031549Srgrimes 8041549Srgrimes /* build a minimal buffer header */ 80558345Sphk bp->b_iocmd = BIO_READ; 8061549Srgrimes bp->b_iodone = vnode_pager_iodone; 8071549Srgrimes /* B_PHYS is not set, but it is nice to fill this in */ 80884827Sjhb KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 80984827Sjhb KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 81091406Sjhb bp->b_rcred = crhold(curthread->td_ucred); 81191406Sjhb bp->b_wcred = crhold(curthread->td_ucred); 8126626Sdg bp->b_blkno = firstaddr; 8135455Sdg pbgetvp(dp, bp); 8141549Srgrimes bp->b_bcount = size; 8151549Srgrimes bp->b_bufsize = size; 81670374Sdillon bp->b_runningbufspace = bp->b_bufsize; 81770374Sdillon runningbufspace += bp->b_runningbufspace; 8181549Srgrimes 8193612Sdg cnt.v_vnodein++; 8203612Sdg cnt.v_vnodepgsin += count; 8213612Sdg 8221549Srgrimes /* do the input */ 823109198Sphk if (dp->v_type == VCHR) 824109198Sphk VOP_SPECSTRATEGY(bp->b_vp, bp); 825109198Sphk else 826109198Sphk VOP_STRATEGY(bp->b_vp, bp); 8273612Sdg 82833758Sdyson s = splvm(); 82933758Sdyson /* we definitely need to be at splvm here */ 8301549Srgrimes 8311549Srgrimes while ((bp->b_flags & B_DONE) == 0) { 8329356Sdg tsleep(bp, PVM, "vnread", 0); 8331549Srgrimes } 8341549Srgrimes splx(s); 83558934Sphk if ((bp->b_ioflags & BIO_ERROR) != 0) 8361549Srgrimes error = EIO; 8371549Srgrimes 8381549Srgrimes if (!error) { 8391549Srgrimes if (size != count * PAGE_SIZE) 8401827Sdg bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 8411549Srgrimes } 8425455Sdg pmap_qremove(kva, count); 8431549Srgrimes 8441549Srgrimes /* 8451549Srgrimes * free the buffer header back to the swap buffer pool 8461549Srgrimes */ 84742957Sdillon relpbuf(bp, &vnode_pbuf_freecnt); 8481549Srgrimes 849100736Salc vm_page_lock_queues(); 85034206Sdyson for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { 85134206Sdyson vm_page_t mt; 85234206Sdyson 85334206Sdyson nextoff = tfoff + PAGE_SIZE; 85434206Sdyson mt = m[i]; 85534206Sdyson 85647239Sdt if (nextoff <= object->un_pager.vnp.vnp_size) { 85745347Sjulian /* 85845347Sjulian * Read filled up entire page. 85945347Sjulian */ 86034206Sdyson mt->valid = VM_PAGE_BITS_ALL; 86149945Salc vm_page_undirty(mt); /* should be an assert? XXX */ 86260755Speter pmap_clear_modify(mt); 86334206Sdyson } else { 86445347Sjulian /* 86545347Sjulian * Read did not fill up entire page. Since this 86645347Sjulian * is getpages, the page may be mapped, so we have 86745347Sjulian * to zero the invalid portions of the page even 86845347Sjulian * though we aren't setting them valid. 86945347Sjulian * 87045347Sjulian * Currently we do not set the entire page valid, 87145347Sjulian * we just try to clear the piece that we couldn't 87245347Sjulian * read. 87345347Sjulian */ 87447239Sdt vm_page_set_validclean(mt, 0, 87547239Sdt object->un_pager.vnp.vnp_size - tfoff); 87646349Salc /* handled by vm_fault now */ 87746349Salc /* vm_page_zero_invalid(mt, FALSE); */ 87834206Sdyson } 87934206Sdyson 88038799Sdfr vm_page_flag_clear(mt, PG_ZERO); 8811549Srgrimes if (i != reqpage) { 8821827Sdg 8831549Srgrimes /* 8841827Sdg * whether or not to leave the page activated is up in 8851827Sdg * the air, but we should put the page on a page queue 8861827Sdg * somewhere. (it already is in the object). Result: 88758634Scharnier * It appears that empirical results show that 8881827Sdg * deactivating pages is best. 8891549Srgrimes */ 8901827Sdg 8911549Srgrimes /* 8921827Sdg * just in case someone was asking for this page we 8931827Sdg * now tell them that it is ok to use 8941549Srgrimes */ 8951549Srgrimes if (!error) { 89634206Sdyson if (mt->flags & PG_WANTED) 89734206Sdyson vm_page_activate(mt); 89833109Sdyson else 89934206Sdyson vm_page_deactivate(mt); 90038799Sdfr vm_page_wakeup(mt); 9011549Srgrimes } else { 90275692Salfred vm_page_free(mt); 9031549Srgrimes } 9041549Srgrimes } 9051549Srgrimes } 906100736Salc vm_page_unlock_queues(); 9071549Srgrimes if (error) { 9089507Sdg printf("vnode_pager_getpages: I/O read error\n"); 9091549Srgrimes } 9104207Sdg return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 9111549Srgrimes} 9121549Srgrimes 91333847Smsmith/* 91433847Smsmith * EOPNOTSUPP is no longer legal. For local media VFS's that do not 91533847Smsmith * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to 91633847Smsmith * vnode_pager_generic_putpages() to implement the previous behaviour. 91733847Smsmith * 91833847Smsmith * All other FS's should use the bypass to get to the local media 91933847Smsmith * backing vp's VOP_PUTPAGES. 92033847Smsmith */ 92143129Sdillonstatic void 92210556Sdysonvnode_pager_putpages(object, m, count, sync, rtvals) 92310556Sdyson vm_object_t object; 92410556Sdyson vm_page_t *m; 92510556Sdyson int count; 92610556Sdyson boolean_t sync; 92710556Sdyson int *rtvals; 92810556Sdyson{ 92910556Sdyson int rtval; 93010556Sdyson struct vnode *vp; 93162976Smckusick struct mount *mp; 93234403Smsmith int bytes = count * PAGE_SIZE; 93318973Sdyson 93479224Sdillon GIANT_REQUIRED; 93544321Salc /* 93644321Salc * Force synchronous operation if we are extremely low on memory 93744321Salc * to prevent a low-memory deadlock. VOP operations often need to 93844321Salc * allocate more memory to initiate the I/O ( i.e. do a BMAP 93944321Salc * operation ). The swapper handles the case by limiting the amount 94044321Salc * of asynchronous I/O, but that sort of solution doesn't scale well 94144321Salc * for the vnode pager without a lot of work. 94244321Salc * 94344321Salc * Also, the backing vnode's iodone routine may not wake the pageout 94444321Salc * daemon up. This should be probably be addressed XXX. 94544321Salc */ 94644321Salc 94744321Salc if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min) 94844321Salc sync |= OBJPC_SYNC; 94944321Salc 95044321Salc /* 95144321Salc * Call device-specific putpages function 95244321Salc */ 95310556Sdyson vp = object->handle; 95462976Smckusick if (vp->v_type != VREG) 95562976Smckusick mp = NULL; 95662976Smckusick (void)vn_start_write(vp, &mp, V_WAIT); 95734403Smsmith rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0); 95876827Salfred KASSERT(rtval != EOPNOTSUPP, 95976827Salfred ("vnode_pager: stale FS putpages\n")); 96062976Smckusick vn_finished_write(mp); 96110556Sdyson} 96210556Sdyson 96333847Smsmith 9641549Srgrimes/* 96533847Smsmith * This is now called from local media FS's to operate against their 96645057Seivind * own vnodes if they fail to implement VOP_PUTPAGES. 96770374Sdillon * 96870374Sdillon * This is typically called indirectly via the pageout daemon and 96970374Sdillon * clustering has already typically occured, so in general we ask the 97070374Sdillon * underlying filesystem to write the data out asynchronously rather 97170374Sdillon * then delayed. 9721549Srgrimes */ 97333847Smsmithint 97434206Sdysonvnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) 97533847Smsmith struct vnode *vp; 9761549Srgrimes vm_page_t *m; 97733847Smsmith int bytecount; 97834206Sdyson int flags; 9795455Sdg int *rtvals; 9801549Srgrimes{ 9817695Sdg int i; 98233847Smsmith vm_object_t object; 98333847Smsmith int count; 9841549Srgrimes 9857695Sdg int maxsize, ncount; 98612767Sdyson vm_ooffset_t poffset; 9877695Sdg struct uio auio; 9887695Sdg struct iovec aiov; 9897695Sdg int error; 99034206Sdyson int ioflags; 9911549Srgrimes 99279224Sdillon GIANT_REQUIRED; 99333847Smsmith object = vp->v_object; 99433847Smsmith count = bytecount / PAGE_SIZE; 99533847Smsmith 9961827Sdg for (i = 0; i < count; i++) 9971549Srgrimes rtvals[i] = VM_PAGER_AGAIN; 9981549Srgrimes 99912767Sdyson if ((int) m[0]->pindex < 0) { 100048409Speter printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n", 100148409Speter (long)m[0]->pindex, m[0]->dirty); 10027695Sdg rtvals[0] = VM_PAGER_BAD; 10037695Sdg return VM_PAGER_BAD; 10045455Sdg } 10057178Sdg 10067695Sdg maxsize = count * PAGE_SIZE; 10077695Sdg ncount = count; 10081549Srgrimes 100912767Sdyson poffset = IDX_TO_OFF(m[0]->pindex); 101084854Sdillon 101184854Sdillon /* 101284854Sdillon * If the page-aligned write is larger then the actual file we 101384854Sdillon * have to invalidate pages occuring beyond the file EOF. However, 101484854Sdillon * there is an edge case where a file may not be page-aligned where 101584854Sdillon * the last page is partially invalid. In this case the filesystem 101684854Sdillon * may not properly clear the dirty bits for the entire page (which 101784854Sdillon * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). 101884854Sdillon * With the page locked we are free to fix-up the dirty bits here. 101987834Sdillon * 102087834Sdillon * We do not under any circumstances truncate the valid bits, as 102187834Sdillon * this will screw up bogus page replacement. 102284854Sdillon */ 102312767Sdyson if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 102484854Sdillon if (object->un_pager.vnp.vnp_size > poffset) { 102584854Sdillon int pgoff; 102684854Sdillon 102712767Sdyson maxsize = object->un_pager.vnp.vnp_size - poffset; 102884854Sdillon ncount = btoc(maxsize); 102984854Sdillon if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { 103084854Sdillon vm_page_clear_dirty(m[ncount - 1], pgoff, 103184854Sdillon PAGE_SIZE - pgoff); 103284854Sdillon } 103384854Sdillon } else { 10348585Sdg maxsize = 0; 103584854Sdillon ncount = 0; 103684854Sdillon } 10378585Sdg if (ncount < count) { 10388585Sdg for (i = ncount; i < count; i++) { 10397695Sdg rtvals[i] = VM_PAGER_BAD; 10401549Srgrimes } 10411549Srgrimes } 10421541Srgrimes } 10437695Sdg 104470374Sdillon /* 104570374Sdillon * pageouts are already clustered, use IO_ASYNC t o force a bawrite() 104670374Sdillon * rather then a bdwrite() to prevent paging I/O from saturating 1047108358Sdillon * the buffer cache. Dummy-up the sequential heuristic to cause 1048108358Sdillon * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set, 1049108358Sdillon * the system decides how to cluster. 105070374Sdillon */ 105134206Sdyson ioflags = IO_VMIO; 1052108358Sdillon if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) 1053108358Sdillon ioflags |= IO_SYNC; 1054108358Sdillon else if ((flags & VM_PAGER_CLUSTER_OK) == 0) 1055108358Sdillon ioflags |= IO_ASYNC; 105634206Sdyson ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; 1057108358Sdillon ioflags |= IO_SEQMAX << IO_SEQSHIFT; 10581827Sdg 10597695Sdg aiov.iov_base = (caddr_t) 0; 10607695Sdg aiov.iov_len = maxsize; 10617695Sdg auio.uio_iov = &aiov; 10627695Sdg auio.uio_iovcnt = 1; 106312767Sdyson auio.uio_offset = poffset; 10647695Sdg auio.uio_segflg = UIO_NOCOPY; 10657695Sdg auio.uio_rw = UIO_WRITE; 10667695Sdg auio.uio_resid = maxsize; 106783366Sjulian auio.uio_td = (struct thread *) 0; 106891406Sjhb error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); 10693612Sdg cnt.v_vnodeout++; 10707695Sdg cnt.v_vnodepgsout += ncount; 10713612Sdg 10728585Sdg if (error) { 10739507Sdg printf("vnode_pager_putpages: I/O error %d\n", error); 10747695Sdg } 10758585Sdg if (auio.uio_resid) { 107637555Sbde printf("vnode_pager_putpages: residual I/O %d at %lu\n", 107737555Sbde auio.uio_resid, (u_long)m[0]->pindex); 10787695Sdg } 107933936Sdyson for (i = 0; i < ncount; i++) { 108033936Sdyson rtvals[i] = VM_PAGER_OK; 10817695Sdg } 10827695Sdg return rtvals[0]; 10837695Sdg} 10841549Srgrimes 10857695Sdgstruct vnode * 10869507Sdgvnode_pager_lock(object) 10879507Sdg vm_object_t object; 10889507Sdg{ 108983366Sjulian struct thread *td = curthread; /* XXX */ 109022521Sdyson 109179224Sdillon GIANT_REQUIRED; 109279224Sdillon 10939507Sdg for (; object != NULL; object = object->backing_object) { 10949507Sdg if (object->type != OBJT_VNODE) 10957695Sdg continue; 109677094Sjhb if (object->flags & OBJ_DEAD) { 109732585Sdyson return NULL; 109877094Sjhb } 10991549Srgrimes 110077094Sjhb /* XXX; If object->handle can change, we need to cache it. */ 110132585Sdyson while (vget(object->handle, 110283366Sjulian LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, td)){ 110334611Sdyson if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE)) 110434611Sdyson return NULL; 110532585Sdyson printf("vnode_pager_lock: retrying\n"); 110632585Sdyson } 11079507Sdg return object->handle; 11081549Srgrimes } 11099507Sdg return NULL; 11107695Sdg} 1111