160727d8bSWarner Losh /*- 2df8bae1dSRodney W. Grimes * Copyright (c) 1990 University of Utah. 326f9a767SRodney W. Grimes * Copyright (c) 1991 The Regents of the University of California. 426f9a767SRodney W. Grimes * All rights reserved. 526f9a767SRodney W. Grimes * Copyright (c) 1993, 1994 John S. Dyson 624a1cce3SDavid Greenman * Copyright (c) 1995, David Greenman 7df8bae1dSRodney W. Grimes * 8df8bae1dSRodney W. Grimes * This code is derived from software contributed to Berkeley by 9df8bae1dSRodney W. Grimes * the Systems Programming Group of the University of Utah Computer 10df8bae1dSRodney W. Grimes * Science Department. 11df8bae1dSRodney W. Grimes * 12df8bae1dSRodney W. Grimes * Redistribution and use in source and binary forms, with or without 13df8bae1dSRodney W. Grimes * modification, are permitted provided that the following conditions 14df8bae1dSRodney W. Grimes * are met: 15df8bae1dSRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 16df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer. 17df8bae1dSRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 18df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 19df8bae1dSRodney W. Grimes * documentation and/or other materials provided with the distribution. 20df8bae1dSRodney W. Grimes * 3. All advertising materials mentioning features or use of this software 215929bcfaSPhilippe Charnier * must display the following acknowledgement: 22df8bae1dSRodney W. Grimes * This product includes software developed by the University of 23df8bae1dSRodney W. Grimes * California, Berkeley and its contributors. 24df8bae1dSRodney W. Grimes * 4. Neither the name of the University nor the names of its contributors 25df8bae1dSRodney W. Grimes * may be used to endorse or promote products derived from this software 26df8bae1dSRodney W. Grimes * without specific prior written permission. 27df8bae1dSRodney W. Grimes * 28df8bae1dSRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29df8bae1dSRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30df8bae1dSRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31df8bae1dSRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32df8bae1dSRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33df8bae1dSRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34df8bae1dSRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35df8bae1dSRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36df8bae1dSRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37df8bae1dSRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38df8bae1dSRodney W. Grimes * SUCH DAMAGE. 39df8bae1dSRodney W. Grimes * 4026f9a767SRodney W. Grimes * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 41df8bae1dSRodney W. Grimes */ 42df8bae1dSRodney W. Grimes 43df8bae1dSRodney W. Grimes /* 44df8bae1dSRodney W. Grimes * Page to/from files (vnodes). 45df8bae1dSRodney W. Grimes */ 46df8bae1dSRodney W. Grimes 4726f9a767SRodney W. Grimes /* 4826f9a767SRodney W. Grimes * TODO: 4924a1cce3SDavid Greenman * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will 50f6b04d2bSDavid Greenman * greatly re-simplify the vnode_pager. 5126f9a767SRodney W. Grimes */ 5226f9a767SRodney W. Grimes 53874651b1SDavid E. O'Brien #include <sys/cdefs.h> 54874651b1SDavid E. O'Brien __FBSDID("$FreeBSD$"); 55874651b1SDavid E. O'Brien 56df8bae1dSRodney W. Grimes #include <sys/param.h> 57df8bae1dSRodney W. Grimes #include <sys/systm.h> 58df8bae1dSRodney W. Grimes #include <sys/proc.h> 59df8bae1dSRodney W. Grimes #include <sys/vnode.h> 60df8bae1dSRodney W. Grimes #include <sys/mount.h> 619626b608SPoul-Henning Kamp #include <sys/bio.h> 6224a1cce3SDavid Greenman #include <sys/buf.h> 63efeaf95aSDavid Greenman #include <sys/vmmeter.h> 64d07a6d3fSPoul-Henning Kamp #include <sys/limits.h> 6524579ca1SMatthew Dillon #include <sys/conf.h> 669e0ddbd0SAlan Cox #include <sys/sf_buf.h> 67df8bae1dSRodney W. Grimes 684f12e0acSSuleiman Souhlal #include <machine/atomic.h> 694f12e0acSSuleiman Souhlal 70df8bae1dSRodney W. Grimes #include <vm/vm.h> 71efeaf95aSDavid Greenman #include <vm/vm_object.h> 72df8bae1dSRodney W. Grimes #include <vm/vm_page.h> 7324a1cce3SDavid Greenman #include <vm/vm_pager.h> 741efb74fbSJohn Dyson #include <vm/vm_map.h> 75df8bae1dSRodney W. Grimes #include <vm/vnode_pager.h> 76efeaf95aSDavid Greenman #include <vm/vm_extern.h> 77df8bae1dSRodney W. Grimes 78bff76343SAlan Cox static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, 79bff76343SAlan Cox daddr_t *rtaddress, int *run); 8011caded3SAlfred Perlstein static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m); 8111caded3SAlfred Perlstein static int vnode_pager_input_old(vm_object_t object, vm_page_t m); 8211caded3SAlfred Perlstein static void vnode_pager_dealloc(vm_object_t); 8311caded3SAlfred Perlstein static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int); 8411caded3SAlfred Perlstein static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); 8511caded3SAlfred Perlstein static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 863364c323SKonstantin Belousov static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 873364c323SKonstantin Belousov vm_ooffset_t, struct ucred *cred); 880b8253a7SBruce Evans 89df8bae1dSRodney W. Grimes struct pagerops vnodepagerops = { 904e658600SPoul-Henning Kamp .pgo_alloc = vnode_pager_alloc, 914e658600SPoul-Henning Kamp .pgo_dealloc = vnode_pager_dealloc, 924e658600SPoul-Henning Kamp .pgo_getpages = vnode_pager_getpages, 934e658600SPoul-Henning Kamp .pgo_putpages = vnode_pager_putpages, 944e658600SPoul-Henning Kamp .pgo_haspage = vnode_pager_haspage, 95df8bae1dSRodney W. Grimes }; 96df8bae1dSRodney W. Grimes 97b62b9b64SJohn Baldwin int vnode_pbuf_freecnt; 981c7c3c6aSMatthew Dillon 99d07a6d3fSPoul-Henning Kamp /* Create the VM system backing object for this vnode */ 100d07a6d3fSPoul-Henning Kamp int 101731959b1SYaroslav Tykhiy vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td) 102d07a6d3fSPoul-Henning Kamp { 103d07a6d3fSPoul-Henning Kamp vm_object_t object; 104d07a6d3fSPoul-Henning Kamp vm_ooffset_t size = isize; 105d07a6d3fSPoul-Henning Kamp struct vattr va; 106d07a6d3fSPoul-Henning Kamp 107d07a6d3fSPoul-Henning Kamp if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE) 108d07a6d3fSPoul-Henning Kamp return (0); 109d07a6d3fSPoul-Henning Kamp 110d07a6d3fSPoul-Henning Kamp while ((object = vp->v_object) != NULL) { 111d07a6d3fSPoul-Henning Kamp VM_OBJECT_LOCK(object); 112d07a6d3fSPoul-Henning Kamp if (!(object->flags & OBJ_DEAD)) { 113d07a6d3fSPoul-Henning Kamp VM_OBJECT_UNLOCK(object); 114d07a6d3fSPoul-Henning Kamp return (0); 115d07a6d3fSPoul-Henning Kamp } 11622db15c0SAttilio Rao VOP_UNLOCK(vp, 0); 117d07a6d3fSPoul-Henning Kamp vm_object_set_flag(object, OBJ_DISCONNECTWNT); 118d07a6d3fSPoul-Henning Kamp msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vodead", 0); 119cb05b60aSAttilio Rao vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 120d07a6d3fSPoul-Henning Kamp } 121d07a6d3fSPoul-Henning Kamp 122d07a6d3fSPoul-Henning Kamp if (size == 0) { 123d07a6d3fSPoul-Henning Kamp if (vn_isdisk(vp, NULL)) { 124d07a6d3fSPoul-Henning Kamp size = IDX_TO_OFF(INT_MAX); 125d07a6d3fSPoul-Henning Kamp } else { 1260359a12eSAttilio Rao if (VOP_GETATTR(vp, &va, td->td_ucred)) 127d07a6d3fSPoul-Henning Kamp return (0); 128d07a6d3fSPoul-Henning Kamp size = va.va_size; 129d07a6d3fSPoul-Henning Kamp } 130d07a6d3fSPoul-Henning Kamp } 131d07a6d3fSPoul-Henning Kamp 1323364c323SKonstantin Belousov object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred); 133d07a6d3fSPoul-Henning Kamp /* 134d07a6d3fSPoul-Henning Kamp * Dereference the reference we just created. This assumes 135d07a6d3fSPoul-Henning Kamp * that the object is associated with the vp. 136d07a6d3fSPoul-Henning Kamp */ 137d07a6d3fSPoul-Henning Kamp VM_OBJECT_LOCK(object); 138d07a6d3fSPoul-Henning Kamp object->ref_count--; 139d07a6d3fSPoul-Henning Kamp VM_OBJECT_UNLOCK(object); 140d07a6d3fSPoul-Henning Kamp vrele(vp); 141d07a6d3fSPoul-Henning Kamp 142d07a6d3fSPoul-Henning Kamp KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object")); 143d07a6d3fSPoul-Henning Kamp 144d07a6d3fSPoul-Henning Kamp return (0); 145d07a6d3fSPoul-Henning Kamp } 146d07a6d3fSPoul-Henning Kamp 1477146d6cbSPoul-Henning Kamp void 1487146d6cbSPoul-Henning Kamp vnode_destroy_vobject(struct vnode *vp) 1497146d6cbSPoul-Henning Kamp { 1507146d6cbSPoul-Henning Kamp struct vm_object *obj; 1517146d6cbSPoul-Henning Kamp 1527146d6cbSPoul-Henning Kamp obj = vp->v_object; 1537146d6cbSPoul-Henning Kamp if (obj == NULL) 1547146d6cbSPoul-Henning Kamp return; 15557fd3d55SPawel Jakub Dawidek ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject"); 1567146d6cbSPoul-Henning Kamp VM_OBJECT_LOCK(obj); 1577146d6cbSPoul-Henning Kamp if (obj->ref_count == 0) { 1587146d6cbSPoul-Henning Kamp /* 1597146d6cbSPoul-Henning Kamp * vclean() may be called twice. The first time 1607146d6cbSPoul-Henning Kamp * removes the primary reference to the object, 1617146d6cbSPoul-Henning Kamp * the second time goes one further and is a 1627146d6cbSPoul-Henning Kamp * special-case to terminate the object. 1637146d6cbSPoul-Henning Kamp * 1647146d6cbSPoul-Henning Kamp * don't double-terminate the object 1657146d6cbSPoul-Henning Kamp */ 1667146d6cbSPoul-Henning Kamp if ((obj->flags & OBJ_DEAD) == 0) 1677146d6cbSPoul-Henning Kamp vm_object_terminate(obj); 1687146d6cbSPoul-Henning Kamp else 1697146d6cbSPoul-Henning Kamp VM_OBJECT_UNLOCK(obj); 1707146d6cbSPoul-Henning Kamp } else { 1717146d6cbSPoul-Henning Kamp /* 1727146d6cbSPoul-Henning Kamp * Woe to the process that tries to page now :-). 1737146d6cbSPoul-Henning Kamp */ 1747146d6cbSPoul-Henning Kamp vm_pager_deallocate(obj); 1757146d6cbSPoul-Henning Kamp VM_OBJECT_UNLOCK(obj); 1767146d6cbSPoul-Henning Kamp } 1776e4b2820SJeff Roberson vp->v_object = NULL; 1787146d6cbSPoul-Henning Kamp } 1797146d6cbSPoul-Henning Kamp 1807146d6cbSPoul-Henning Kamp 181df8bae1dSRodney W. Grimes /* 182df8bae1dSRodney W. Grimes * Allocate (or lookup) pager for a vnode. 183df8bae1dSRodney W. Grimes * Handle is a vnode pointer. 184990ab7adSAlan Cox * 185990ab7adSAlan Cox * MPSAFE 186df8bae1dSRodney W. Grimes */ 18724a1cce3SDavid Greenman vm_object_t 1886cde7a16SDavid Greenman vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 1893364c323SKonstantin Belousov vm_ooffset_t offset, struct ucred *cred) 190df8bae1dSRodney W. Grimes { 19106cb7259SDavid Greenman vm_object_t object; 192df8bae1dSRodney W. Grimes struct vnode *vp; 193df8bae1dSRodney W. Grimes 194df8bae1dSRodney W. Grimes /* 195df8bae1dSRodney W. Grimes * Pageout to vnode, no can do yet. 196df8bae1dSRodney W. Grimes */ 197df8bae1dSRodney W. Grimes if (handle == NULL) 198df8bae1dSRodney W. Grimes return (NULL); 199df8bae1dSRodney W. Grimes 200df8bae1dSRodney W. Grimes vp = (struct vnode *) handle; 20139d38f93SDavid Greenman 20239d38f93SDavid Greenman /* 20339d38f93SDavid Greenman * If the object is being terminated, wait for it to 20439d38f93SDavid Greenman * go away. 20539d38f93SDavid Greenman */ 2062ac78f0eSStephan Uphoff retry: 2071ca58953SAlan Cox while ((object = vp->v_object) != NULL) { 2081ca58953SAlan Cox VM_OBJECT_LOCK(object); 2091ca58953SAlan Cox if ((object->flags & OBJ_DEAD) == 0) 2101ca58953SAlan Cox break; 21119187819SAlan Cox vm_object_set_flag(object, OBJ_DISCONNECTWNT); 2121ca58953SAlan Cox msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0); 21324a1cce3SDavid Greenman } 2140d94caffSDavid Greenman 2152be70f79SJohn Dyson if (vp->v_usecount == 0) 2162be70f79SJohn Dyson panic("vnode_pager_alloc: no vnode reference"); 2172be70f79SJohn Dyson 21824a1cce3SDavid Greenman if (object == NULL) { 219df8bae1dSRodney W. Grimes /* 2202ac78f0eSStephan Uphoff * Add an object of the appropriate size 221df8bae1dSRodney W. Grimes */ 2226cde7a16SDavid Greenman object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); 223bbc0ec52SDavid Greenman 2246cde7a16SDavid Greenman object->un_pager.vnp.vnp_size = size; 22526f9a767SRodney W. Grimes 22624a1cce3SDavid Greenman object->handle = handle; 22711be8415SStephan Uphoff VI_LOCK(vp); 2282ac78f0eSStephan Uphoff if (vp->v_object != NULL) { 2292ac78f0eSStephan Uphoff /* 2302ac78f0eSStephan Uphoff * Object has been created while we were sleeping 2312ac78f0eSStephan Uphoff */ 23211be8415SStephan Uphoff VI_UNLOCK(vp); 2332ac78f0eSStephan Uphoff vm_object_destroy(object); 2342ac78f0eSStephan Uphoff goto retry; 235df8bae1dSRodney W. Grimes } 2362ac78f0eSStephan Uphoff vp->v_object = object; 23711be8415SStephan Uphoff VI_UNLOCK(vp); 23811be8415SStephan Uphoff } else { 2392ac78f0eSStephan Uphoff object->ref_count++; 2402ac78f0eSStephan Uphoff VM_OBJECT_UNLOCK(object); 24111be8415SStephan Uphoff } 2427747c038SJeff Roberson vref(vp); 24324a1cce3SDavid Greenman return (object); 244df8bae1dSRodney W. Grimes } 245df8bae1dSRodney W. Grimes 246658ad5ffSAlan Cox /* 247658ad5ffSAlan Cox * The object must be locked. 248658ad5ffSAlan Cox */ 249f708ef1bSPoul-Henning Kamp static void 25024a1cce3SDavid Greenman vnode_pager_dealloc(object) 2510d94caffSDavid Greenman vm_object_t object; 25224a1cce3SDavid Greenman { 253b9f180d1SKonstantin Belousov struct vnode *vp; 254b9f180d1SKonstantin Belousov int refs; 255df8bae1dSRodney W. Grimes 256b9f180d1SKonstantin Belousov vp = object->handle; 25724a1cce3SDavid Greenman if (vp == NULL) 25824a1cce3SDavid Greenman panic("vnode_pager_dealloc: pager already dealloced"); 25924a1cce3SDavid Greenman 260658ad5ffSAlan Cox VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 26166095752SJohn Dyson vm_object_pip_wait(object, "vnpdea"); 262b9f180d1SKonstantin Belousov refs = object->ref_count; 26324a1cce3SDavid Greenman 26424a1cce3SDavid Greenman object->handle = NULL; 26595461b45SJohn Dyson object->type = OBJT_DEAD; 26619187819SAlan Cox if (object->flags & OBJ_DISCONNECTWNT) { 26719187819SAlan Cox vm_object_clear_flag(object, OBJ_DISCONNECTWNT); 26819187819SAlan Cox wakeup(object); 26919187819SAlan Cox } 27057fd3d55SPawel Jakub Dawidek ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc"); 271aa2cabb9SDavid Greenman vp->v_object = NULL; 27235764be3SPoul-Henning Kamp vp->v_vflag &= ~VV_TEXT; 273b9f180d1SKonstantin Belousov while (refs-- > 0) 274b9f180d1SKonstantin Belousov vunref(vp); 275df8bae1dSRodney W. Grimes } 27626f9a767SRodney W. Grimes 277f708ef1bSPoul-Henning Kamp static boolean_t 278a316d390SJohn Dyson vnode_pager_haspage(object, pindex, before, after) 27924a1cce3SDavid Greenman vm_object_t object; 280a316d390SJohn Dyson vm_pindex_t pindex; 28124a1cce3SDavid Greenman int *before; 28224a1cce3SDavid Greenman int *after; 283df8bae1dSRodney W. Grimes { 28424a1cce3SDavid Greenman struct vnode *vp = object->handle; 28598b0c789SPoul-Henning Kamp daddr_t bn; 2863af76890SPoul-Henning Kamp int err; 287170db9c6SJohn Dyson daddr_t reqblock; 2882c4488fcSJohn Dyson int poff; 2892c4488fcSJohn Dyson int bsize; 290d63596ceSJohn Dyson int pagesperblock, blocksperpage; 291ae51ff11SJeff Roberson int vfslocked; 292df8bae1dSRodney W. Grimes 293f29ba63eSAlan Cox VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 29424579ca1SMatthew Dillon /* 29524579ca1SMatthew Dillon * If no vp or vp is doomed or marked transparent to VM, we do not 29624579ca1SMatthew Dillon * have the page. 29724579ca1SMatthew Dillon */ 298b73f64c4SJeff Roberson if (vp == NULL || vp->v_iflag & VI_DOOMED) 29947221757SJohn Dyson return FALSE; 300df8bae1dSRodney W. Grimes /* 301b73f64c4SJeff Roberson * If the offset is beyond end of file we do 3020d94caffSDavid Greenman * not have the page. 303df8bae1dSRodney W. Grimes */ 304b73f64c4SJeff Roberson if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size) 3054abc71c0SDavid Greenman return FALSE; 306df8bae1dSRodney W. Grimes 307eed2d59bSDavid Greenman bsize = vp->v_mount->mnt_stat.f_iosize; 308170db9c6SJohn Dyson pagesperblock = bsize / PAGE_SIZE; 309d63596ceSJohn Dyson blocksperpage = 0; 310d63596ceSJohn Dyson if (pagesperblock > 0) { 311a316d390SJohn Dyson reqblock = pindex / pagesperblock; 312d63596ceSJohn Dyson } else { 313d63596ceSJohn Dyson blocksperpage = (PAGE_SIZE / bsize); 314d63596ceSJohn Dyson reqblock = pindex * blocksperpage; 315d63596ceSJohn Dyson } 316f29ba63eSAlan Cox VM_OBJECT_UNLOCK(object); 317ae51ff11SJeff Roberson vfslocked = VFS_LOCK_GIANT(vp->v_mount); 318ec794849SPoul-Henning Kamp err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before); 319ae51ff11SJeff Roberson VFS_UNLOCK_GIANT(vfslocked); 320f29ba63eSAlan Cox VM_OBJECT_LOCK(object); 3210d94caffSDavid Greenman if (err) 32224a1cce3SDavid Greenman return TRUE; 3236eab77f2SJohn Dyson if (bn == -1) 324ced399eeSJohn Dyson return FALSE; 325d63596ceSJohn Dyson if (pagesperblock > 0) { 326a316d390SJohn Dyson poff = pindex - (reqblock * pagesperblock); 327170db9c6SJohn Dyson if (before) { 328170db9c6SJohn Dyson *before *= pagesperblock; 329170db9c6SJohn Dyson *before += poff; 330170db9c6SJohn Dyson } 331170db9c6SJohn Dyson if (after) { 332b1fc01b7SJohn Dyson int numafter; 333170db9c6SJohn Dyson *after *= pagesperblock; 334b1fc01b7SJohn Dyson numafter = pagesperblock - (poff + 1); 33547e151ddSRobert Drehmel if (IDX_TO_OFF(pindex + numafter) > 33647e151ddSRobert Drehmel object->un_pager.vnp.vnp_size) { 33747e151ddSRobert Drehmel numafter = 33847e151ddSRobert Drehmel OFF_TO_IDX(object->un_pager.vnp.vnp_size) - 33947e151ddSRobert Drehmel pindex; 340b1fc01b7SJohn Dyson } 341b1fc01b7SJohn Dyson *after += numafter; 342170db9c6SJohn Dyson } 343d63596ceSJohn Dyson } else { 344d63596ceSJohn Dyson if (before) { 345d63596ceSJohn Dyson *before /= blocksperpage; 346d63596ceSJohn Dyson } 347d63596ceSJohn Dyson 348d63596ceSJohn Dyson if (after) { 349d63596ceSJohn Dyson *after /= blocksperpage; 350d63596ceSJohn Dyson } 351d63596ceSJohn Dyson } 352ced399eeSJohn Dyson return TRUE; 353df8bae1dSRodney W. Grimes } 354df8bae1dSRodney W. Grimes 355df8bae1dSRodney W. Grimes /* 356df8bae1dSRodney W. Grimes * Lets the VM system know about a change in size for a file. 35724a1cce3SDavid Greenman * We adjust our own internal size and flush any cached pages in 358df8bae1dSRodney W. Grimes * the associated object that are affected by the size change. 359df8bae1dSRodney W. Grimes * 360df8bae1dSRodney W. Grimes * Note: this routine may be invoked as a result of a pager put 361df8bae1dSRodney W. Grimes * operation (possibly at object termination time), so we must be careful. 362df8bae1dSRodney W. Grimes */ 363df8bae1dSRodney W. Grimes void 364df8bae1dSRodney W. Grimes vnode_pager_setsize(vp, nsize) 365df8bae1dSRodney W. Grimes struct vnode *vp; 366a316d390SJohn Dyson vm_ooffset_t nsize; 367df8bae1dSRodney W. Grimes { 3682a8f9ab5SAlan Cox vm_object_t object; 3692a8f9ab5SAlan Cox vm_page_t m; 370c576d121SLuoqi Chen vm_pindex_t nobjsize; 371df8bae1dSRodney W. Grimes 3722a8f9ab5SAlan Cox if ((object = vp->v_object) == NULL) 373df8bae1dSRodney W. Grimes return; 374cb61d698SKonstantin Belousov /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */ 3752a8f9ab5SAlan Cox VM_OBJECT_LOCK(object); 3762a8f9ab5SAlan Cox if (nsize == object->un_pager.vnp.vnp_size) { 377df8bae1dSRodney W. Grimes /* 378df8bae1dSRodney W. Grimes * Hasn't changed size 379df8bae1dSRodney W. Grimes */ 3802a8f9ab5SAlan Cox VM_OBJECT_UNLOCK(object); 381df8bae1dSRodney W. Grimes return; 3822a8f9ab5SAlan Cox } 383c576d121SLuoqi Chen nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); 3842a8f9ab5SAlan Cox if (nsize < object->un_pager.vnp.vnp_size) { 385df8bae1dSRodney W. Grimes /* 386bbc0ec52SDavid Greenman * File has shrunk. Toss any cached pages beyond the new EOF. 387df8bae1dSRodney W. Grimes */ 3882a8f9ab5SAlan Cox if (nobjsize < object->size) 389c576d121SLuoqi Chen vm_object_page_remove(object, nobjsize, object->size, 3906bbee8e2SAlan Cox 0); 391bbc0ec52SDavid Greenman /* 392bbc0ec52SDavid Greenman * this gets rid of garbage at the end of a page that is now 3933ebeaf59SMatthew Dillon * only partially backed by the vnode. 3943ebeaf59SMatthew Dillon * 3953ebeaf59SMatthew Dillon * XXX for some reason (I don't know yet), if we take a 3963ebeaf59SMatthew Dillon * completely invalid page and mark it partially valid 3973ebeaf59SMatthew Dillon * it can screw up NFS reads, so we don't allow the case. 398bbc0ec52SDavid Greenman */ 3992a8f9ab5SAlan Cox if ((nsize & PAGE_MASK) && 4001b26eb10SAlan Cox (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL && 4011b26eb10SAlan Cox m->valid != 0) { 4022b6b0df7SMatthew Dillon int base = (int)nsize & PAGE_MASK; 4032b6b0df7SMatthew Dillon int size = PAGE_SIZE - base; 4042b6b0df7SMatthew Dillon 4052b6b0df7SMatthew Dillon /* 4062b6b0df7SMatthew Dillon * Clear out partial-page garbage in case 4072b6b0df7SMatthew Dillon * the page has been mapped. 4082b6b0df7SMatthew Dillon */ 409fff6062aSAlan Cox pmap_zero_page_area(m, base, size); 4102b6b0df7SMatthew Dillon 4112b6b0df7SMatthew Dillon /* 4123c33df62SAlan Cox * Update the valid bits to reflect the blocks that 4133c33df62SAlan Cox * have been zeroed. Some of these valid bits may 4143c33df62SAlan Cox * have already been set. 4153c33df62SAlan Cox */ 4163c33df62SAlan Cox vm_page_set_valid(m, base, size); 4173c33df62SAlan Cox 4183c33df62SAlan Cox /* 4193c33df62SAlan Cox * Round "base" to the next block boundary so that the 4203c33df62SAlan Cox * dirty bit for a partially zeroed block is not 4213c33df62SAlan Cox * cleared. 4223c33df62SAlan Cox */ 4233c33df62SAlan Cox base = roundup2(base, DEV_BSIZE); 4243c33df62SAlan Cox 4253c33df62SAlan Cox /* 4263c33df62SAlan Cox * Clear out partial-page dirty bits. 4273ebeaf59SMatthew Dillon * 4283ebeaf59SMatthew Dillon * note that we do not clear out the valid 4293ebeaf59SMatthew Dillon * bits. This would prevent bogus_page 4303ebeaf59SMatthew Dillon * replacement from working properly. 4312b6b0df7SMatthew Dillon */ 4323c33df62SAlan Cox vm_page_clear_dirty(m, base, PAGE_SIZE - base); 4330ab3c7a5SAlan Cox } else if ((nsize & PAGE_MASK) && 4340ab3c7a5SAlan Cox __predict_false(object->cache != NULL)) { 4350ab3c7a5SAlan Cox vm_page_cache_free(object, OFF_TO_IDX(nsize), 4360ab3c7a5SAlan Cox nobjsize); 437bbc0ec52SDavid Greenman } 438bbc0ec52SDavid Greenman } 439a316d390SJohn Dyson object->un_pager.vnp.vnp_size = nsize; 440c576d121SLuoqi Chen object->size = nobjsize; 4412a8f9ab5SAlan Cox VM_OBJECT_UNLOCK(object); 442df8bae1dSRodney W. Grimes } 443df8bae1dSRodney W. Grimes 44426f9a767SRodney W. Grimes /* 44526f9a767SRodney W. Grimes * calculate the linear (byte) disk address of specified virtual 44626f9a767SRodney W. Grimes * file address 44726f9a767SRodney W. Grimes */ 448bff76343SAlan Cox static int 449bff76343SAlan Cox vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress, 450bff76343SAlan Cox int *run) 45126f9a767SRodney W. Grimes { 45226f9a767SRodney W. Grimes int bsize; 45326f9a767SRodney W. Grimes int err; 454a316d390SJohn Dyson daddr_t vblock; 455f3aad9a6SBjoern A. Zeeb daddr_t voffset; 45626f9a767SRodney W. Grimes 4572ad036b6SAlan Cox if (address < 0) 4580d94caffSDavid Greenman return -1; 4590d94caffSDavid Greenman 460b73f64c4SJeff Roberson if (vp->v_iflag & VI_DOOMED) 4612c4488fcSJohn Dyson return -1; 4622c4488fcSJohn Dyson 46326f9a767SRodney W. Grimes bsize = vp->v_mount->mnt_stat.f_iosize; 46426f9a767SRodney W. Grimes vblock = address / bsize; 46526f9a767SRodney W. Grimes voffset = address % bsize; 46626f9a767SRodney W. Grimes 467bff76343SAlan Cox err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL); 468bff76343SAlan Cox if (err == 0) { 469bff76343SAlan Cox if (*rtaddress != -1) 470bff76343SAlan Cox *rtaddress += voffset / DEV_BSIZE; 471efc68ce1SDavid Greenman if (run) { 472efc68ce1SDavid Greenman *run += 1; 473efc68ce1SDavid Greenman *run *= bsize/PAGE_SIZE; 474efc68ce1SDavid Greenman *run -= voffset/PAGE_SIZE; 475efc68ce1SDavid Greenman } 476efc68ce1SDavid Greenman } 47726f9a767SRodney W. Grimes 478bff76343SAlan Cox return (err); 47926f9a767SRodney W. Grimes } 48026f9a767SRodney W. Grimes 48126f9a767SRodney W. Grimes /* 48226f9a767SRodney W. Grimes * small block filesystem vnode pager input 48326f9a767SRodney W. Grimes */ 484f708ef1bSPoul-Henning Kamp static int 48524a1cce3SDavid Greenman vnode_pager_input_smlfs(object, m) 48624a1cce3SDavid Greenman vm_object_t object; 48726f9a767SRodney W. Grimes vm_page_t m; 48826f9a767SRodney W. Grimes { 4899c83534dSPoul-Henning Kamp struct vnode *vp; 4909c83534dSPoul-Henning Kamp struct bufobj *bo; 49126f9a767SRodney W. Grimes struct buf *bp; 4929e0ddbd0SAlan Cox struct sf_buf *sf; 493f3aad9a6SBjoern A. Zeeb daddr_t fileaddr; 49426f9a767SRodney W. Grimes vm_offset_t bsize; 495*561cc9fcSKonstantin Belousov vm_page_bits_t bits; 496*561cc9fcSKonstantin Belousov int error, i; 49726f9a767SRodney W. Grimes 498*561cc9fcSKonstantin Belousov error = 0; 49924a1cce3SDavid Greenman vp = object->handle; 500b73f64c4SJeff Roberson if (vp->v_iflag & VI_DOOMED) 5012c4488fcSJohn Dyson return VM_PAGER_BAD; 5022c4488fcSJohn Dyson 50326f9a767SRodney W. Grimes bsize = vp->v_mount->mnt_stat.f_iosize; 5040bdb7528SDavid Greenman 5059c83534dSPoul-Henning Kamp VOP_BMAP(vp, 0, &bo, 0, NULL, NULL); 50626f9a767SRodney W. Grimes 5079e0ddbd0SAlan Cox sf = sf_buf_alloc(m, 0); 50826f9a767SRodney W. Grimes 50926f9a767SRodney W. Grimes for (i = 0; i < PAGE_SIZE / bsize; i++) { 51033c67741SMatthew Dillon vm_ooffset_t address; 511bbc0ec52SDavid Greenman 5120d53a17bSAlan Cox bits = vm_page_bits(i * bsize, bsize); 5130d53a17bSAlan Cox if (m->valid & bits) 51426f9a767SRodney W. Grimes continue; 51526f9a767SRodney W. Grimes 51633c67741SMatthew Dillon address = IDX_TO_OFF(m->pindex) + i * bsize; 51733c67741SMatthew Dillon if (address >= object->un_pager.vnp.vnp_size) { 51833c67741SMatthew Dillon fileaddr = -1; 51933c67741SMatthew Dillon } else { 520bff76343SAlan Cox error = vnode_pager_addr(vp, address, &fileaddr, NULL); 521bff76343SAlan Cox if (error) 522bff76343SAlan Cox break; 52333c67741SMatthew Dillon } 52426f9a767SRodney W. Grimes if (fileaddr != -1) { 5251c7c3c6aSMatthew Dillon bp = getpbuf(&vnode_pbuf_freecnt); 52626f9a767SRodney W. Grimes 52726f9a767SRodney W. Grimes /* build a minimal buffer header */ 52821144e3bSPoul-Henning Kamp bp->b_iocmd = BIO_READ; 5296a4b5823SPoul-Henning Kamp bp->b_iodone = bdone; 530bd78ceceSJohn Baldwin KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 531bd78ceceSJohn Baldwin KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 532a854ed98SJohn Baldwin bp->b_rcred = crhold(curthread->td_ucred); 533a854ed98SJohn Baldwin bp->b_wcred = crhold(curthread->td_ucred); 5349e0ddbd0SAlan Cox bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize; 535187f0071SDavid Greenman bp->b_blkno = fileaddr; 5369c83534dSPoul-Henning Kamp pbgetbo(bo, bp); 53726f9a767SRodney W. Grimes bp->b_bcount = bsize; 53826f9a767SRodney W. Grimes bp->b_bufsize = bsize; 5392b6b0df7SMatthew Dillon bp->b_runningbufspace = bp->b_bufsize; 5405bd65606SJohn Baldwin atomic_add_long(&runningbufspace, bp->b_runningbufspace); 54126f9a767SRodney W. Grimes 54226f9a767SRodney W. Grimes /* do the input */ 5432c18019fSPoul-Henning Kamp bp->b_iooffset = dbtob(bp->b_blkno); 544b792bebeSPoul-Henning Kamp bstrategy(bp); 54526f9a767SRodney W. Grimes 5466a4b5823SPoul-Henning Kamp bwait(bp, PVM, "vnsrd"); 5476a4b5823SPoul-Henning Kamp 548c244d2deSPoul-Henning Kamp if ((bp->b_ioflags & BIO_ERROR) != 0) 54926f9a767SRodney W. Grimes error = EIO; 55026f9a767SRodney W. Grimes 55126f9a767SRodney W. Grimes /* 55226f9a767SRodney W. Grimes * free the buffer header back to the swap buffer pool 55326f9a767SRodney W. Grimes */ 5549c83534dSPoul-Henning Kamp pbrelbo(bp); 5551c7c3c6aSMatthew Dillon relpbuf(bp, &vnode_pbuf_freecnt); 55626f9a767SRodney W. Grimes if (error) 55726f9a767SRodney W. Grimes break; 5580d53a17bSAlan Cox } else 5599e0ddbd0SAlan Cox bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize); 5600d53a17bSAlan Cox KASSERT((m->dirty & bits) == 0, 5610d53a17bSAlan Cox ("vnode_pager_input_smlfs: page %p is dirty", m)); 5620d53a17bSAlan Cox VM_OBJECT_LOCK(object); 5630d53a17bSAlan Cox m->valid |= bits; 5640d53a17bSAlan Cox VM_OBJECT_UNLOCK(object); 56526f9a767SRodney W. Grimes } 5669e0ddbd0SAlan Cox sf_buf_free(sf); 56726f9a767SRodney W. Grimes if (error) { 568a83c285cSDavid Greenman return VM_PAGER_ERROR; 56926f9a767SRodney W. Grimes } 57026f9a767SRodney W. Grimes return VM_PAGER_OK; 57126f9a767SRodney W. Grimes } 57226f9a767SRodney W. Grimes 57326f9a767SRodney W. Grimes /* 574475e8cc3SPoul-Henning Kamp * old style vnode pager input routine 57526f9a767SRodney W. Grimes */ 576f708ef1bSPoul-Henning Kamp static int 57724a1cce3SDavid Greenman vnode_pager_input_old(object, m) 57824a1cce3SDavid Greenman vm_object_t object; 57926f9a767SRodney W. Grimes vm_page_t m; 58026f9a767SRodney W. Grimes { 581df8bae1dSRodney W. Grimes struct uio auio; 582df8bae1dSRodney W. Grimes struct iovec aiov; 58326f9a767SRodney W. Grimes int error; 58426f9a767SRodney W. Grimes int size; 5859e0ddbd0SAlan Cox struct sf_buf *sf; 586342a1480SJohn Baldwin struct vnode *vp; 587df8bae1dSRodney W. Grimes 58852051abcSAlan Cox VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 58926f9a767SRodney W. Grimes error = 0; 590bbc0ec52SDavid Greenman 591df8bae1dSRodney W. Grimes /* 59226f9a767SRodney W. Grimes * Return failure if beyond current EOF 59326f9a767SRodney W. Grimes */ 594a316d390SJohn Dyson if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 59526f9a767SRodney W. Grimes return VM_PAGER_BAD; 59626f9a767SRodney W. Grimes } else { 59726f9a767SRodney W. Grimes size = PAGE_SIZE; 598a316d390SJohn Dyson if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 599a316d390SJohn Dyson size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 60052051abcSAlan Cox vp = object->handle; 60152051abcSAlan Cox VM_OBJECT_UNLOCK(object); 6020bdb7528SDavid Greenman 60326f9a767SRodney W. Grimes /* 604df8bae1dSRodney W. Grimes * Allocate a kernel virtual address and initialize so that 605df8bae1dSRodney W. Grimes * we can use VOP_READ/WRITE routines. 606df8bae1dSRodney W. Grimes */ 6079e0ddbd0SAlan Cox sf = sf_buf_alloc(m, 0); 6080bdb7528SDavid Greenman 6099e0ddbd0SAlan Cox aiov.iov_base = (caddr_t)sf_buf_kva(sf); 610df8bae1dSRodney W. Grimes aiov.iov_len = size; 611df8bae1dSRodney W. Grimes auio.uio_iov = &aiov; 612df8bae1dSRodney W. Grimes auio.uio_iovcnt = 1; 613a316d390SJohn Dyson auio.uio_offset = IDX_TO_OFF(m->pindex); 614df8bae1dSRodney W. Grimes auio.uio_segflg = UIO_SYSSPACE; 61526f9a767SRodney W. Grimes auio.uio_rw = UIO_READ; 616df8bae1dSRodney W. Grimes auio.uio_resid = size; 617b40ce416SJulian Elischer auio.uio_td = curthread; 61826f9a767SRodney W. Grimes 619a854ed98SJohn Baldwin error = VOP_READ(vp, &auio, 0, curthread->td_ucred); 620df8bae1dSRodney W. Grimes if (!error) { 62154d92145SMatthew Dillon int count = size - auio.uio_resid; 622df8bae1dSRodney W. Grimes 623df8bae1dSRodney W. Grimes if (count == 0) 624df8bae1dSRodney W. Grimes error = EINVAL; 62526f9a767SRodney W. Grimes else if (count != PAGE_SIZE) 6269e0ddbd0SAlan Cox bzero((caddr_t)sf_buf_kva(sf) + count, 6279e0ddbd0SAlan Cox PAGE_SIZE - count); 628df8bae1dSRodney W. Grimes } 6299e0ddbd0SAlan Cox sf_buf_free(sf); 6301b26eb10SAlan Cox 6311b26eb10SAlan Cox VM_OBJECT_LOCK(object); 632df8bae1dSRodney W. Grimes } 6330d53a17bSAlan Cox KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m)); 6346e3a3f38SRobert V. Baron if (!error) 6356e3a3f38SRobert V. Baron m->valid = VM_PAGE_BITS_ALL; 636a83c285cSDavid Greenman return error ? VM_PAGER_ERROR : VM_PAGER_OK; 63726f9a767SRodney W. Grimes } 63826f9a767SRodney W. Grimes 63926f9a767SRodney W. Grimes /* 64026f9a767SRodney W. Grimes * generic vnode pager input routine 64126f9a767SRodney W. Grimes */ 642170db9c6SJohn Dyson 643ce75f2c3SMike Smith /* 64423955314SAlfred Perlstein * Local media VFS's that do not implement their own VOP_GETPAGES 64547e151ddSRobert Drehmel * should have their VOP_GETPAGES call to vnode_pager_generic_getpages() 64647e151ddSRobert Drehmel * to implement the previous behaviour. 647ce75f2c3SMike Smith * 648ce75f2c3SMike Smith * All other FS's should use the bypass to get to the local media 649ce75f2c3SMike Smith * backing vp's VOP_GETPAGES. 650ce75f2c3SMike Smith */ 651f708ef1bSPoul-Henning Kamp static int 65224a1cce3SDavid Greenman vnode_pager_getpages(object, m, count, reqpage) 65326f9a767SRodney W. Grimes vm_object_t object; 65424a1cce3SDavid Greenman vm_page_t *m; 65524a1cce3SDavid Greenman int count; 65624a1cce3SDavid Greenman int reqpage; 65724a1cce3SDavid Greenman { 658170db9c6SJohn Dyson int rtval; 659170db9c6SJohn Dyson struct vnode *vp; 66086ffbd76SMike Smith int bytes = count * PAGE_SIZE; 661ae51ff11SJeff Roberson int vfslocked; 66295e5e988SJohn Dyson 663170db9c6SJohn Dyson vp = object->handle; 6648630c117SAlan Cox VM_OBJECT_UNLOCK(object); 665ae51ff11SJeff Roberson vfslocked = VFS_LOCK_GIANT(vp->v_mount); 66686ffbd76SMike Smith rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0); 66723955314SAlfred Perlstein KASSERT(rtval != EOPNOTSUPP, 66823955314SAlfred Perlstein ("vnode_pager: FS getpages not implemented\n")); 669ae51ff11SJeff Roberson VFS_UNLOCK_GIANT(vfslocked); 6708630c117SAlan Cox VM_OBJECT_LOCK(object); 671170db9c6SJohn Dyson return rtval; 672170db9c6SJohn Dyson } 673170db9c6SJohn Dyson 674ce75f2c3SMike Smith /* 675ce75f2c3SMike Smith * This is now called from local media FS's to operate against their 676ce75f2c3SMike Smith * own vnodes if they fail to implement VOP_GETPAGES. 677ce75f2c3SMike Smith */ 678ce75f2c3SMike Smith int 679ce75f2c3SMike Smith vnode_pager_generic_getpages(vp, m, bytecount, reqpage) 680ce75f2c3SMike Smith struct vnode *vp; 681170db9c6SJohn Dyson vm_page_t *m; 682ce75f2c3SMike Smith int bytecount; 683170db9c6SJohn Dyson int reqpage; 684170db9c6SJohn Dyson { 685ce75f2c3SMike Smith vm_object_t object; 686a316d390SJohn Dyson vm_offset_t kva; 6878f9110f6SJohn Dyson off_t foff, tfoff, nextoff; 688f3aad9a6SBjoern A. Zeeb int i, j, size, bsize, first; 689f4f83da0SAlan Cox daddr_t firstaddr, reqblock; 6909c83534dSPoul-Henning Kamp struct bufobj *bo; 691efc68ce1SDavid Greenman int runpg; 692efc68ce1SDavid Greenman int runend; 6930bdb7528SDavid Greenman struct buf *bp; 694ce75f2c3SMike Smith int count; 6951de11f1aSAlan Cox int error; 69626f9a767SRodney W. Grimes 697ce75f2c3SMike Smith object = vp->v_object; 698ce75f2c3SMike Smith count = bytecount / PAGE_SIZE; 699ce75f2c3SMike Smith 7009c83534dSPoul-Henning Kamp KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, 7019c83534dSPoul-Henning Kamp ("vnode_pager_generic_getpages does not support devices")); 702b73f64c4SJeff Roberson if (vp->v_iflag & VI_DOOMED) 7032c4488fcSJohn Dyson return VM_PAGER_BAD; 7042c4488fcSJohn Dyson 70526f9a767SRodney W. Grimes bsize = vp->v_mount->mnt_stat.f_iosize; 70626f9a767SRodney W. Grimes 70726f9a767SRodney W. Grimes /* get the UNDERLYING device for the file with VOP_BMAP() */ 708bbc0ec52SDavid Greenman 70926f9a767SRodney W. Grimes /* 710bbc0ec52SDavid Greenman * originally, we did not check for an error return value -- assuming 711bbc0ec52SDavid Greenman * an fs always has a bmap entry point -- that assumption is wrong!!! 71226f9a767SRodney W. Grimes */ 713a316d390SJohn Dyson foff = IDX_TO_OFF(m[reqpage]->pindex); 714bbc0ec52SDavid Greenman 71526f9a767SRodney W. Grimes /* 71616f62314SDavid Greenman * if we can't bmap, use old VOP code 71726f9a767SRodney W. Grimes */ 7181de11f1aSAlan Cox error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL); 7191de11f1aSAlan Cox if (error == EOPNOTSUPP) { 72031953be9SAlan Cox VM_OBJECT_LOCK(object); 7212965a453SKip Macy 722e43c2eabSAlan Cox for (i = 0; i < count; i++) 7232965a453SKip Macy if (i != reqpage) { 7242965a453SKip Macy vm_page_lock(m[i]); 725d8d5fa88SAlfred Perlstein vm_page_free(m[i]); 7262965a453SKip Macy vm_page_unlock(m[i]); 7272965a453SKip Macy } 728b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodein); 729b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodepgsin); 73052051abcSAlan Cox error = vnode_pager_input_old(object, m[reqpage]); 73152051abcSAlan Cox VM_OBJECT_UNLOCK(object); 73252051abcSAlan Cox return (error); 7331de11f1aSAlan Cox } else if (error != 0) { 7341de11f1aSAlan Cox VM_OBJECT_LOCK(object); 7351de11f1aSAlan Cox for (i = 0; i < count; i++) 7362965a453SKip Macy if (i != reqpage) { 7372965a453SKip Macy vm_page_lock(m[i]); 7381de11f1aSAlan Cox vm_page_free(m[i]); 7392965a453SKip Macy vm_page_unlock(m[i]); 7402965a453SKip Macy } 7411de11f1aSAlan Cox VM_OBJECT_UNLOCK(object); 7421de11f1aSAlan Cox return (VM_PAGER_ERROR); 743bbc0ec52SDavid Greenman 74426f9a767SRodney W. Grimes /* 74526f9a767SRodney W. Grimes * if the blocksize is smaller than a page size, then use 74626f9a767SRodney W. Grimes * special small filesystem code. NFS sometimes has a small 74726f9a767SRodney W. Grimes * blocksize, but it can handle large reads itself. 74826f9a767SRodney W. Grimes */ 74926f9a767SRodney W. Grimes } else if ((PAGE_SIZE / bsize) > 1 && 750500b04a2SBruce Evans (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) { 75131953be9SAlan Cox VM_OBJECT_LOCK(object); 752e43c2eabSAlan Cox for (i = 0; i < count; i++) 7532965a453SKip Macy if (i != reqpage) { 7542965a453SKip Macy vm_page_lock(m[i]); 755d8d5fa88SAlfred Perlstein vm_page_free(m[i]); 7562965a453SKip Macy vm_page_unlock(m[i]); 7572965a453SKip Macy } 75831953be9SAlan Cox VM_OBJECT_UNLOCK(object); 759b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodein); 760b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodepgsin); 76124a1cce3SDavid Greenman return vnode_pager_input_smlfs(object, m[reqpage]); 76226f9a767SRodney W. Grimes } 7638d17e694SJulian Elischer 76426f9a767SRodney W. Grimes /* 7658d17e694SJulian Elischer * If we have a completely valid page available to us, we can 7668d17e694SJulian Elischer * clean up and return. Otherwise we have to re-read the 7678d17e694SJulian Elischer * media. 7680d94caffSDavid Greenman */ 76931953be9SAlan Cox VM_OBJECT_LOCK(object); 7708b575f6cSAlan Cox if (m[reqpage]->valid == VM_PAGE_BITS_ALL) { 771e43c2eabSAlan Cox for (i = 0; i < count; i++) 7722965a453SKip Macy if (i != reqpage) { 7732965a453SKip Macy vm_page_lock(m[i]); 774d8d5fa88SAlfred Perlstein vm_page_free(m[i]); 7752965a453SKip Macy vm_page_unlock(m[i]); 7762965a453SKip Macy } 77731953be9SAlan Cox VM_OBJECT_UNLOCK(object); 7780d94caffSDavid Greenman return VM_PAGER_OK; 779f4f83da0SAlan Cox } else if (reqblock == -1) { 780f4f83da0SAlan Cox pmap_zero_page(m[reqpage]); 78112aa4fdcSAlan Cox KASSERT(m[reqpage]->dirty == 0, 78212aa4fdcSAlan Cox ("vnode_pager_generic_getpages: page %p is dirty", m)); 783f4f83da0SAlan Cox m[reqpage]->valid = VM_PAGE_BITS_ALL; 784f4f83da0SAlan Cox for (i = 0; i < count; i++) 7852965a453SKip Macy if (i != reqpage) { 7862965a453SKip Macy vm_page_lock(m[i]); 787f4f83da0SAlan Cox vm_page_free(m[i]); 7882965a453SKip Macy vm_page_unlock(m[i]); 7892965a453SKip Macy } 790f4f83da0SAlan Cox VM_OBJECT_UNLOCK(object); 791f4f83da0SAlan Cox return (VM_PAGER_OK); 7920d94caffSDavid Greenman } 7938d17e694SJulian Elischer m[reqpage]->valid = 0; 7948b575f6cSAlan Cox VM_OBJECT_UNLOCK(object); 7950bdb7528SDavid Greenman 7960d94caffSDavid Greenman /* 79726f9a767SRodney W. Grimes * here on direct device I/O 79826f9a767SRodney W. Grimes */ 799efc68ce1SDavid Greenman firstaddr = -1; 800a1287949SEivind Eklund 80126f9a767SRodney W. Grimes /* 802efc68ce1SDavid Greenman * calculate the run that includes the required page 80326f9a767SRodney W. Grimes */ 804efc68ce1SDavid Greenman for (first = 0, i = 0; i < count; i = runend) { 805bff76343SAlan Cox if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr, 806bff76343SAlan Cox &runpg) != 0) { 807bff76343SAlan Cox VM_OBJECT_LOCK(object); 808bff76343SAlan Cox for (; i < count; i++) 8092965a453SKip Macy if (i != reqpage) { 8102965a453SKip Macy vm_page_lock(m[i]); 811bff76343SAlan Cox vm_page_free(m[i]); 8122965a453SKip Macy vm_page_unlock(m[i]); 8132965a453SKip Macy } 814bff76343SAlan Cox VM_OBJECT_UNLOCK(object); 815bff76343SAlan Cox return (VM_PAGER_ERROR); 816bff76343SAlan Cox } 817efc68ce1SDavid Greenman if (firstaddr == -1) { 81831953be9SAlan Cox VM_OBJECT_LOCK(object); 81924a1cce3SDavid Greenman if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 820f3aad9a6SBjoern A. Zeeb panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx", 821f3aad9a6SBjoern A. Zeeb (intmax_t)firstaddr, (uintmax_t)(foff >> 32), 822bf1001faSMaxime Henrion (uintmax_t)foff, 823bf1001faSMaxime Henrion (uintmax_t) 824fc62ef1fSBruce Evans (object->un_pager.vnp.vnp_size >> 32), 825bf1001faSMaxime Henrion (uintmax_t)object->un_pager.vnp.vnp_size); 826efc68ce1SDavid Greenman } 8272965a453SKip Macy vm_page_lock(m[i]); 828d8d5fa88SAlfred Perlstein vm_page_free(m[i]); 8292965a453SKip Macy vm_page_unlock(m[i]); 83031953be9SAlan Cox VM_OBJECT_UNLOCK(object); 831efc68ce1SDavid Greenman runend = i + 1; 832efc68ce1SDavid Greenman first = runend; 833efc68ce1SDavid Greenman continue; 834efc68ce1SDavid Greenman } 835efc68ce1SDavid Greenman runend = i + runpg; 836efc68ce1SDavid Greenman if (runend <= reqpage) { 83731953be9SAlan Cox VM_OBJECT_LOCK(object); 8382965a453SKip Macy for (j = i; j < runend; j++) { 8392965a453SKip Macy vm_page_lock(m[j]); 840d8d5fa88SAlfred Perlstein vm_page_free(m[j]); 8412965a453SKip Macy vm_page_unlock(m[j]); 8422965a453SKip Macy } 84331953be9SAlan Cox VM_OBJECT_UNLOCK(object); 84426f9a767SRodney W. Grimes } else { 845efc68ce1SDavid Greenman if (runpg < (count - first)) { 84631953be9SAlan Cox VM_OBJECT_LOCK(object); 8472965a453SKip Macy for (i = first + runpg; i < count; i++) { 8482965a453SKip Macy vm_page_lock(m[i]); 849d8d5fa88SAlfred Perlstein vm_page_free(m[i]); 8502965a453SKip Macy vm_page_unlock(m[i]); 8512965a453SKip Macy } 85231953be9SAlan Cox VM_OBJECT_UNLOCK(object); 853efc68ce1SDavid Greenman count = first + runpg; 85426f9a767SRodney W. Grimes } 855efc68ce1SDavid Greenman break; 85626f9a767SRodney W. Grimes } 857efc68ce1SDavid Greenman first = runend; 858efc68ce1SDavid Greenman } 85926f9a767SRodney W. Grimes 86026f9a767SRodney W. Grimes /* 861bbc0ec52SDavid Greenman * the first and last page have been calculated now, move input pages 862bbc0ec52SDavid Greenman * to be zero based... 86326f9a767SRodney W. Grimes */ 86426f9a767SRodney W. Grimes if (first != 0) { 8657e2393ffSAlan Cox m += first; 86626f9a767SRodney W. Grimes count -= first; 86726f9a767SRodney W. Grimes reqpage -= first; 86826f9a767SRodney W. Grimes } 869efc68ce1SDavid Greenman 87026f9a767SRodney W. Grimes /* 87126f9a767SRodney W. Grimes * calculate the file virtual address for the transfer 87226f9a767SRodney W. Grimes */ 873a316d390SJohn Dyson foff = IDX_TO_OFF(m[0]->pindex); 87426f9a767SRodney W. Grimes 87526f9a767SRodney W. Grimes /* 87626f9a767SRodney W. Grimes * calculate the size of the transfer 87726f9a767SRodney W. Grimes */ 87826f9a767SRodney W. Grimes size = count * PAGE_SIZE; 8791a31a6c3SPoul-Henning Kamp KASSERT(count > 0, ("zero count")); 88024a1cce3SDavid Greenman if ((foff + size) > object->un_pager.vnp.vnp_size) 88124a1cce3SDavid Greenman size = object->un_pager.vnp.vnp_size - foff; 8821a31a6c3SPoul-Henning Kamp KASSERT(size > 0, ("zero size")); 88326f9a767SRodney W. Grimes 88426f9a767SRodney W. Grimes /* 88524579ca1SMatthew Dillon * round up physical size for real devices. 88626f9a767SRodney W. Grimes */ 8879c83534dSPoul-Henning Kamp if (1) { 8889c83534dSPoul-Henning Kamp int secmask = bo->bo_bsize - 1; 8896229cc50SPoul-Henning Kamp KASSERT(secmask < PAGE_SIZE && secmask > 0, 8906229cc50SPoul-Henning Kamp ("vnode_pager_generic_getpages: sector size %d too large", 8916229cc50SPoul-Henning Kamp secmask + 1)); 89224579ca1SMatthew Dillon size = (size + secmask) & ~secmask; 89324579ca1SMatthew Dillon } 89426f9a767SRodney W. Grimes 8951c7c3c6aSMatthew Dillon bp = getpbuf(&vnode_pbuf_freecnt); 89616f62314SDavid Greenman kva = (vm_offset_t) bp->b_data; 89716f62314SDavid Greenman 89826f9a767SRodney W. Grimes /* 89926f9a767SRodney W. Grimes * and map the pages to be read into the kva 90026f9a767SRodney W. Grimes */ 90116f62314SDavid Greenman pmap_qenter(kva, m, count); 90226f9a767SRodney W. Grimes 90326f9a767SRodney W. Grimes /* build a minimal buffer header */ 90421144e3bSPoul-Henning Kamp bp->b_iocmd = BIO_READ; 9056a4b5823SPoul-Henning Kamp bp->b_iodone = bdone; 906bd78ceceSJohn Baldwin KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 907bd78ceceSJohn Baldwin KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 908a854ed98SJohn Baldwin bp->b_rcred = crhold(curthread->td_ucred); 909a854ed98SJohn Baldwin bp->b_wcred = crhold(curthread->td_ucred); 910187f0071SDavid Greenman bp->b_blkno = firstaddr; 9119c83534dSPoul-Henning Kamp pbgetbo(bo, bp); 91226f9a767SRodney W. Grimes bp->b_bcount = size; 91326f9a767SRodney W. Grimes bp->b_bufsize = size; 9142b6b0df7SMatthew Dillon bp->b_runningbufspace = bp->b_bufsize; 9155bd65606SJohn Baldwin atomic_add_long(&runningbufspace, bp->b_runningbufspace); 91626f9a767SRodney W. Grimes 917b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodein); 918b4b70819SAttilio Rao PCPU_ADD(cnt.v_vnodepgsin, count); 919976e77fcSDavid Greenman 92026f9a767SRodney W. Grimes /* do the input */ 9212c18019fSPoul-Henning Kamp bp->b_iooffset = dbtob(bp->b_blkno); 922b792bebeSPoul-Henning Kamp bstrategy(bp); 923976e77fcSDavid Greenman 9246a4b5823SPoul-Henning Kamp bwait(bp, PVM, "vnread"); 92526f9a767SRodney W. Grimes 926c244d2deSPoul-Henning Kamp if ((bp->b_ioflags & BIO_ERROR) != 0) 92726f9a767SRodney W. Grimes error = EIO; 92826f9a767SRodney W. Grimes 92926f9a767SRodney W. Grimes if (!error) { 93026f9a767SRodney W. Grimes if (size != count * PAGE_SIZE) 93126f9a767SRodney W. Grimes bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 93226f9a767SRodney W. Grimes } 93316f62314SDavid Greenman pmap_qremove(kva, count); 93426f9a767SRodney W. Grimes 93526f9a767SRodney W. Grimes /* 93626f9a767SRodney W. Grimes * free the buffer header back to the swap buffer pool 93726f9a767SRodney W. Grimes */ 9389c83534dSPoul-Henning Kamp pbrelbo(bp); 9391c7c3c6aSMatthew Dillon relpbuf(bp, &vnode_pbuf_freecnt); 94026f9a767SRodney W. Grimes 94131953be9SAlan Cox VM_OBJECT_LOCK(object); 9428f9110f6SJohn Dyson for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { 9438f9110f6SJohn Dyson vm_page_t mt; 9448f9110f6SJohn Dyson 9458f9110f6SJohn Dyson nextoff = tfoff + PAGE_SIZE; 9468f9110f6SJohn Dyson mt = m[i]; 9478f9110f6SJohn Dyson 94854746b67SDmitrij Tejblum if (nextoff <= object->un_pager.vnp.vnp_size) { 9498d17e694SJulian Elischer /* 9508d17e694SJulian Elischer * Read filled up entire page. 9518d17e694SJulian Elischer */ 9528f9110f6SJohn Dyson mt->valid = VM_PAGE_BITS_ALL; 953016a3c93SAlan Cox KASSERT(mt->dirty == 0, 954016a3c93SAlan Cox ("vnode_pager_generic_getpages: page %p is dirty", 955016a3c93SAlan Cox mt)); 956016a3c93SAlan Cox KASSERT(!pmap_page_is_mapped(mt), 957016a3c93SAlan Cox ("vnode_pager_generic_getpages: page %p is mapped", 958016a3c93SAlan Cox mt)); 9598f9110f6SJohn Dyson } else { 9608d17e694SJulian Elischer /* 96142eb4108SAlan Cox * Read did not fill up entire page. 9628d17e694SJulian Elischer * 9638d17e694SJulian Elischer * Currently we do not set the entire page valid, 9648d17e694SJulian Elischer * we just try to clear the piece that we couldn't 9658d17e694SJulian Elischer * read. 9668d17e694SJulian Elischer */ 96742eb4108SAlan Cox vm_page_set_valid(mt, 0, 96854746b67SDmitrij Tejblum object->un_pager.vnp.vnp_size - tfoff); 96942eb4108SAlan Cox KASSERT((mt->dirty & vm_page_bits(0, 97042eb4108SAlan Cox object->un_pager.vnp.vnp_size - tfoff)) == 0, 97142eb4108SAlan Cox ("vnode_pager_generic_getpages: page %p is dirty", 97242eb4108SAlan Cox mt)); 9738f9110f6SJohn Dyson } 9748f9110f6SJohn Dyson 97526f9a767SRodney W. Grimes if (i != reqpage) { 976bbc0ec52SDavid Greenman 97726f9a767SRodney W. Grimes /* 978bbc0ec52SDavid Greenman * whether or not to leave the page activated is up in 979bbc0ec52SDavid Greenman * the air, but we should put the page on a page queue 980bbc0ec52SDavid Greenman * somewhere. (it already is in the object). Result: 981956f3135SPhilippe Charnier * It appears that empirical results show that 982bbc0ec52SDavid Greenman * deactivating pages is best. 98326f9a767SRodney W. Grimes */ 984bbc0ec52SDavid Greenman 98526f9a767SRodney W. Grimes /* 986bbc0ec52SDavid Greenman * just in case someone was asking for this page we 987bbc0ec52SDavid Greenman * now tell them that it is ok to use 98826f9a767SRodney W. Grimes */ 98926f9a767SRodney W. Grimes if (!error) { 99003679e23SAlan Cox if (mt->oflags & VPO_WANTED) { 99103679e23SAlan Cox vm_page_lock(mt); 9928f9110f6SJohn Dyson vm_page_activate(mt); 99303679e23SAlan Cox vm_page_unlock(mt); 99403679e23SAlan Cox } else { 99503679e23SAlan Cox vm_page_lock(mt); 9968f9110f6SJohn Dyson vm_page_deactivate(mt); 99703679e23SAlan Cox vm_page_unlock(mt); 99803679e23SAlan Cox } 999e69763a3SDoug Rabson vm_page_wakeup(mt); 100026f9a767SRodney W. Grimes } else { 100103679e23SAlan Cox vm_page_lock(mt); 1002d8d5fa88SAlfred Perlstein vm_page_free(mt); 10032965a453SKip Macy vm_page_unlock(mt); 10042965a453SKip Macy } 100503679e23SAlan Cox } 100603679e23SAlan Cox } 100731953be9SAlan Cox VM_OBJECT_UNLOCK(object); 100826f9a767SRodney W. Grimes if (error) { 100924a1cce3SDavid Greenman printf("vnode_pager_getpages: I/O read error\n"); 101026f9a767SRodney W. Grimes } 1011a83c285cSDavid Greenman return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 101226f9a767SRodney W. Grimes } 101326f9a767SRodney W. Grimes 1014ce75f2c3SMike Smith /* 1015ce75f2c3SMike Smith * EOPNOTSUPP is no longer legal. For local media VFS's that do not 1016ce75f2c3SMike Smith * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to 1017ce75f2c3SMike Smith * vnode_pager_generic_putpages() to implement the previous behaviour. 1018ce75f2c3SMike Smith * 1019ce75f2c3SMike Smith * All other FS's should use the bypass to get to the local media 1020ce75f2c3SMike Smith * backing vp's VOP_PUTPAGES. 1021ce75f2c3SMike Smith */ 1022e4542174SMatthew Dillon static void 1023170db9c6SJohn Dyson vnode_pager_putpages(object, m, count, sync, rtvals) 1024170db9c6SJohn Dyson vm_object_t object; 1025170db9c6SJohn Dyson vm_page_t *m; 1026170db9c6SJohn Dyson int count; 1027170db9c6SJohn Dyson boolean_t sync; 1028170db9c6SJohn Dyson int *rtvals; 1029170db9c6SJohn Dyson { 1030170db9c6SJohn Dyson int rtval; 1031170db9c6SJohn Dyson struct vnode *vp; 103286ffbd76SMike Smith int bytes = count * PAGE_SIZE; 1033ad980522SJohn Dyson 10340e3cdf2cSAlan Cox /* 10350e3cdf2cSAlan Cox * Force synchronous operation if we are extremely low on memory 10360e3cdf2cSAlan Cox * to prevent a low-memory deadlock. VOP operations often need to 10370e3cdf2cSAlan Cox * allocate more memory to initiate the I/O ( i.e. do a BMAP 10380e3cdf2cSAlan Cox * operation ). The swapper handles the case by limiting the amount 10390e3cdf2cSAlan Cox * of asynchronous I/O, but that sort of solution doesn't scale well 10400e3cdf2cSAlan Cox * for the vnode pager without a lot of work. 10410e3cdf2cSAlan Cox * 10420e3cdf2cSAlan Cox * Also, the backing vnode's iodone routine may not wake the pageout 10430e3cdf2cSAlan Cox * daemon up. This should be probably be addressed XXX. 10440e3cdf2cSAlan Cox */ 10450e3cdf2cSAlan Cox 10462feb50bfSAttilio Rao if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min) 10470e3cdf2cSAlan Cox sync |= OBJPC_SYNC; 10480e3cdf2cSAlan Cox 10490e3cdf2cSAlan Cox /* 10500e3cdf2cSAlan Cox * Call device-specific putpages function 10510e3cdf2cSAlan Cox */ 1052170db9c6SJohn Dyson vp = object->handle; 10532e3b314dSAlan Cox VM_OBJECT_UNLOCK(object); 105486ffbd76SMike Smith rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0); 105523955314SAlfred Perlstein KASSERT(rtval != EOPNOTSUPP, 105623955314SAlfred Perlstein ("vnode_pager: stale FS putpages\n")); 10572e3b314dSAlan Cox VM_OBJECT_LOCK(object); 1058170db9c6SJohn Dyson } 1059170db9c6SJohn Dyson 1060ce75f2c3SMike Smith 106126f9a767SRodney W. Grimes /* 1062ce75f2c3SMike Smith * This is now called from local media FS's to operate against their 10634491ea91SEivind Eklund * own vnodes if they fail to implement VOP_PUTPAGES. 10642b6b0df7SMatthew Dillon * 10652b6b0df7SMatthew Dillon * This is typically called indirectly via the pageout daemon and 10662b6b0df7SMatthew Dillon * clustering has already typically occured, so in general we ask the 10672b6b0df7SMatthew Dillon * underlying filesystem to write the data out asynchronously rather 10682b6b0df7SMatthew Dillon * then delayed. 106926f9a767SRodney W. Grimes */ 1070ce75f2c3SMike Smith int 1071c46b90e9SAlan Cox vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, 1072c46b90e9SAlan Cox int flags, int *rtvals) 107326f9a767SRodney W. Grimes { 1074f6b04d2bSDavid Greenman int i; 1075ce75f2c3SMike Smith vm_object_t object; 1076c46b90e9SAlan Cox vm_page_t m; 1077ce75f2c3SMike Smith int count; 107826f9a767SRodney W. Grimes 1079f6b04d2bSDavid Greenman int maxsize, ncount; 1080a316d390SJohn Dyson vm_ooffset_t poffset; 1081f6b04d2bSDavid Greenman struct uio auio; 1082f6b04d2bSDavid Greenman struct iovec aiov; 1083f6b04d2bSDavid Greenman int error; 10848f9110f6SJohn Dyson int ioflags; 1085dd498befSPaul Saab int ppscheck = 0; 1086dd498befSPaul Saab static struct timeval lastfail; 1087dd498befSPaul Saab static int curfail; 108826f9a767SRodney W. Grimes 1089ce75f2c3SMike Smith object = vp->v_object; 1090ce75f2c3SMike Smith count = bytecount / PAGE_SIZE; 1091ce75f2c3SMike Smith 109226f9a767SRodney W. Grimes for (i = 0; i < count; i++) 1093031ec8c1SKonstantin Belousov rtvals[i] = VM_PAGER_ERROR; 109426f9a767SRodney W. Grimes 1095c46b90e9SAlan Cox if ((int64_t)ma[0]->pindex < 0) { 109623562e4bSMarcel Moolenaar printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n", 1097c46b90e9SAlan Cox (long)ma[0]->pindex, (u_long)ma[0]->dirty); 1098f6b04d2bSDavid Greenman rtvals[0] = VM_PAGER_BAD; 1099f6b04d2bSDavid Greenman return VM_PAGER_BAD; 11000d94caffSDavid Greenman } 11010bdb7528SDavid Greenman 1102f6b04d2bSDavid Greenman maxsize = count * PAGE_SIZE; 1103f6b04d2bSDavid Greenman ncount = count; 110426f9a767SRodney W. Grimes 1105c46b90e9SAlan Cox poffset = IDX_TO_OFF(ma[0]->pindex); 110600a6f47fSMatthew Dillon 110700a6f47fSMatthew Dillon /* 110800a6f47fSMatthew Dillon * If the page-aligned write is larger then the actual file we 110900a6f47fSMatthew Dillon * have to invalidate pages occuring beyond the file EOF. However, 111000a6f47fSMatthew Dillon * there is an edge case where a file may not be page-aligned where 111100a6f47fSMatthew Dillon * the last page is partially invalid. In this case the filesystem 111200a6f47fSMatthew Dillon * may not properly clear the dirty bits for the entire page (which 111300a6f47fSMatthew Dillon * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). 111400a6f47fSMatthew Dillon * With the page locked we are free to fix-up the dirty bits here. 11153ebeaf59SMatthew Dillon * 11163ebeaf59SMatthew Dillon * We do not under any circumstances truncate the valid bits, as 11173ebeaf59SMatthew Dillon * this will screw up bogus page replacement. 111800a6f47fSMatthew Dillon */ 1119c46b90e9SAlan Cox VM_OBJECT_LOCK(object); 1120a316d390SJohn Dyson if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 112100a6f47fSMatthew Dillon if (object->un_pager.vnp.vnp_size > poffset) { 112200a6f47fSMatthew Dillon int pgoff; 112300a6f47fSMatthew Dillon 1124a316d390SJohn Dyson maxsize = object->un_pager.vnp.vnp_size - poffset; 1125aa8de40aSPoul-Henning Kamp ncount = btoc(maxsize); 112600a6f47fSMatthew Dillon if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { 1127c46b90e9SAlan Cox /* 1128c46b90e9SAlan Cox * If the object is locked and the following 1129c46b90e9SAlan Cox * conditions hold, then the page's dirty 1130c46b90e9SAlan Cox * field cannot be concurrently changed by a 1131c46b90e9SAlan Cox * pmap operation. 1132c46b90e9SAlan Cox */ 1133c46b90e9SAlan Cox m = ma[ncount - 1]; 1134c46b90e9SAlan Cox KASSERT(m->busy > 0, 1135c46b90e9SAlan Cox ("vnode_pager_generic_putpages: page %p is not busy", m)); 11363407fefeSKonstantin Belousov KASSERT((m->aflags & PGA_WRITEABLE) == 0, 1137c46b90e9SAlan Cox ("vnode_pager_generic_putpages: page %p is not read-only", m)); 1138c46b90e9SAlan Cox vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 1139c46b90e9SAlan Cox pgoff); 114000a6f47fSMatthew Dillon } 114100a6f47fSMatthew Dillon } else { 114200a6f47fSMatthew Dillon maxsize = 0; 114300a6f47fSMatthew Dillon ncount = 0; 114400a6f47fSMatthew Dillon } 1145f6b04d2bSDavid Greenman if (ncount < count) { 1146f6b04d2bSDavid Greenman for (i = ncount; i < count; i++) { 1147f6b04d2bSDavid Greenman rtvals[i] = VM_PAGER_BAD; 1148f6b04d2bSDavid Greenman } 1149f6b04d2bSDavid Greenman } 1150f6b04d2bSDavid Greenman } 1151c46b90e9SAlan Cox VM_OBJECT_UNLOCK(object); 115226f9a767SRodney W. Grimes 11532b6b0df7SMatthew Dillon /* 11542b6b0df7SMatthew Dillon * pageouts are already clustered, use IO_ASYNC to force a bawrite() 11552b6b0df7SMatthew Dillon * rather then a bdwrite() to prevent paging I/O from saturating 115643b7990eSMatthew Dillon * the buffer cache. Dummy-up the sequential heuristic to cause 115743b7990eSMatthew Dillon * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set, 115843b7990eSMatthew Dillon * the system decides how to cluster. 11592b6b0df7SMatthew Dillon */ 11608f9110f6SJohn Dyson ioflags = IO_VMIO; 116143b7990eSMatthew Dillon if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) 116243b7990eSMatthew Dillon ioflags |= IO_SYNC; 116343b7990eSMatthew Dillon else if ((flags & VM_PAGER_CLUSTER_OK) == 0) 116443b7990eSMatthew Dillon ioflags |= IO_ASYNC; 11658f9110f6SJohn Dyson ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; 116643b7990eSMatthew Dillon ioflags |= IO_SEQMAX << IO_SEQSHIFT; 1167f6b04d2bSDavid Greenman 1168f6b04d2bSDavid Greenman aiov.iov_base = (caddr_t) 0; 1169f6b04d2bSDavid Greenman aiov.iov_len = maxsize; 1170f6b04d2bSDavid Greenman auio.uio_iov = &aiov; 1171f6b04d2bSDavid Greenman auio.uio_iovcnt = 1; 1172a316d390SJohn Dyson auio.uio_offset = poffset; 1173f6b04d2bSDavid Greenman auio.uio_segflg = UIO_NOCOPY; 1174f6b04d2bSDavid Greenman auio.uio_rw = UIO_WRITE; 1175f6b04d2bSDavid Greenman auio.uio_resid = maxsize; 1176b40ce416SJulian Elischer auio.uio_td = (struct thread *) 0; 1177a854ed98SJohn Baldwin error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); 1178b4b70819SAttilio Rao PCPU_INC(cnt.v_vnodeout); 1179b4b70819SAttilio Rao PCPU_ADD(cnt.v_vnodepgsout, ncount); 1180f6b04d2bSDavid Greenman 1181f6b04d2bSDavid Greenman if (error) { 1182dd498befSPaul Saab if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1))) 118324a1cce3SDavid Greenman printf("vnode_pager_putpages: I/O error %d\n", error); 1184f6b04d2bSDavid Greenman } 1185f6b04d2bSDavid Greenman if (auio.uio_resid) { 1186dd498befSPaul Saab if (ppscheck || ppsratecheck(&lastfail, &curfail, 1)) 11879f80ce04SKonstantin Belousov printf("vnode_pager_putpages: residual I/O %zd at %lu\n", 1188c46b90e9SAlan Cox auio.uio_resid, (u_long)ma[0]->pindex); 118926f9a767SRodney W. Grimes } 1190ffc82b0aSJohn Dyson for (i = 0; i < ncount; i++) { 119126f9a767SRodney W. Grimes rtvals[i] = VM_PAGER_OK; 119226f9a767SRodney W. Grimes } 1193f6b04d2bSDavid Greenman return rtvals[0]; 119426f9a767SRodney W. Grimes } 1195031ec8c1SKonstantin Belousov 1196031ec8c1SKonstantin Belousov void 1197031ec8c1SKonstantin Belousov vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written) 1198031ec8c1SKonstantin Belousov { 11999d17da3bSKonstantin Belousov vm_object_t obj; 1200031ec8c1SKonstantin Belousov int i, pos; 1201031ec8c1SKonstantin Belousov 12029d17da3bSKonstantin Belousov if (written == 0) 12039d17da3bSKonstantin Belousov return; 12049d17da3bSKonstantin Belousov obj = ma[0]->object; 12059d17da3bSKonstantin Belousov VM_OBJECT_LOCK(obj); 1206031ec8c1SKonstantin Belousov for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) { 1207031ec8c1SKonstantin Belousov if (pos < trunc_page(written)) { 1208031ec8c1SKonstantin Belousov rtvals[i] = VM_PAGER_OK; 1209031ec8c1SKonstantin Belousov vm_page_undirty(ma[i]); 1210031ec8c1SKonstantin Belousov } else { 1211031ec8c1SKonstantin Belousov /* Partially written page. */ 1212031ec8c1SKonstantin Belousov rtvals[i] = VM_PAGER_AGAIN; 1213031ec8c1SKonstantin Belousov vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK); 1214031ec8c1SKonstantin Belousov } 1215031ec8c1SKonstantin Belousov } 12169d17da3bSKonstantin Belousov VM_OBJECT_UNLOCK(obj); 1217031ec8c1SKonstantin Belousov } 1218