vnode_pager.c revision 119045
150477Speter/* 21817Sdg * Copyright (c) 1990 University of Utah. 31817Sdg * Copyright (c) 1991 The Regents of the University of California. 41541Srgrimes * All rights reserved. 51541Srgrimes * Copyright (c) 1993, 1994 John S. Dyson 61541Srgrimes * Copyright (c) 1995, David Greenman 7160798Sjhb * 81541Srgrimes * This code is derived from software contributed to Berkeley by 9146806Srwatson * the Systems Programming Group of the University of Utah Computer 10146806Srwatson * Science Department. 11146806Srwatson * 12146806Srwatson * Redistribution and use in source and binary forms, with or without 13146806Srwatson * modification, are permitted provided that the following conditions 14194390Sjhb * are met: 15203660Sed * 1. Redistributions of source code must retain the above copyright 16194390Sjhb * notice, this list of conditions and the following disclaimer. 17194390Sjhb * 2. Redistributions in binary form must reproduce the above copyright 1811294Sswallace * notice, this list of conditions and the following disclaimer in the 1910905Sbde * documentation and/or other materials provided with the distribution. 201541Srgrimes * 3. All advertising materials mentioning features or use of this software 2110905Sbde * must display the following acknowledgement: 2210905Sbde * 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 * 2899855Salfred * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29194645Sjhb * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30194833Sjhb * 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 3369449Salfred * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34194383Sjhb * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35160797Sjhb * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36181972Sobrien * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37181972Sobrien * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38183361Sjhb * SUCH DAMAGE. 39181972Sobrien * 40181972Sobrien * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 41181972Sobrien */ 42181972Sobrien 43211838Skib/* 44104747Srwatson * Page to/from files (vnodes). 45104747Srwatson */ 46123408Speter 47123408Speter/* 481541Srgrimes * TODO: 491541Srgrimes * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will 5011294Sswallace * greatly re-simplify the vnode_pager. 5111294Sswallace */ 5211294Sswallace 5311294Sswallace#include <sys/cdefs.h> 541541Srgrimes__FBSDID("$FreeBSD: head/sys/vm/vnode_pager.c 119045 2003-08-17 18:54:23Z phk $"); 551541Srgrimes 561541Srgrimes#include <sys/param.h> 571541Srgrimes#include <sys/systm.h> 581541Srgrimes#include <sys/proc.h> 591541Srgrimes#include <sys/vnode.h> 60160798Sjhb#include <sys/mount.h> 61160798Sjhb#include <sys/bio.h> 62146806Srwatson#include <sys/buf.h> 63160798Sjhb#include <sys/vmmeter.h> 64160798Sjhb#include <sys/conf.h> 65146806Srwatson 66160798Sjhb#include <vm/vm.h> 67146806Srwatson#include <vm/vm_object.h> 68160798Sjhb#include <vm/vm_page.h> 6912216Sbde#include <vm/vm_pager.h> 7012216Sbde#include <vm/vm_map.h> 7112216Sbde#include <vm/vnode_pager.h> 72160798Sjhb#include <vm/vm_extern.h> 73160798Sjhb 74146806Srwatsonstatic void vnode_pager_init(void); 75146806Srwatsonstatic vm_offset_t vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, 76162991Srwatson int *run); 77160798Sjhbstatic void vnode_pager_iodone(struct buf *bp); 78160798Sjhbstatic int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m); 79146806Srwatsonstatic int vnode_pager_input_old(vm_object_t object, vm_page_t m); 80160798Sjhbstatic void vnode_pager_dealloc(vm_object_t); 81160798Sjhbstatic int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int); 82160798Sjhbstatic void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); 83160798Sjhbstatic boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 84160798Sjhb 85160798Sjhbstruct pagerops vnodepagerops = { 86146806Srwatson .pgo_init = vnode_pager_init, 87160798Sjhb .pgo_alloc = vnode_pager_alloc, 88146806Srwatson .pgo_dealloc = vnode_pager_dealloc, 89160798Sjhb .pgo_getpages = vnode_pager_getpages, 90146806Srwatson .pgo_putpages = vnode_pager_putpages, 91160798Sjhb .pgo_haspage = vnode_pager_haspage, 92160798Sjhb}; 93146806Srwatson 9412216Sbdeint vnode_pbuf_freecnt; 95160798Sjhb 96160798Sjhbstatic void 97160798Sjhbvnode_pager_init(void) 98160798Sjhb{ 99160798Sjhb 100146806Srwatson vnode_pbuf_freecnt = nswbuf / 2 + 1; 101160798Sjhb} 102146806Srwatson 103160798Sjhb/* 104146806Srwatson * Allocate (or lookup) pager for a vnode. 105160798Sjhb * Handle is a vnode pointer. 106146806Srwatson * 107146806Srwatson * MPSAFE 108146806Srwatson */ 109160798Sjhbvm_object_t 110146806Srwatsonvnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 111146806Srwatson vm_ooffset_t offset) 112160798Sjhb{ 113146806Srwatson vm_object_t object; 114146806Srwatson struct vnode *vp; 115160798Sjhb 116146806Srwatson /* 117146806Srwatson * Pageout to vnode, no can do yet. 118160798Sjhb */ 119160798Sjhb if (handle == NULL) 120160798Sjhb return (NULL); 121160798Sjhb 122160798Sjhb vp = (struct vnode *) handle; 123160798Sjhb 124160798Sjhb ASSERT_VOP_LOCKED(vp, "vnode_pager_alloc"); 125160798Sjhb 126160798Sjhb mtx_lock(&Giant); 127160798Sjhb /* 128160798Sjhb * Prevent race condition when allocating the object. This 129160798Sjhb * can happen with NFS vnodes since the nfsnode isn't locked. 130146806Srwatson */ 131160798Sjhb VI_LOCK(vp); 132146806Srwatson while (vp->v_iflag & VI_OLOCK) { 133160798Sjhb vp->v_iflag |= VI_OWANT; 134146806Srwatson msleep(vp, VI_MTX(vp), PVM, "vnpobj", 0); 135146806Srwatson } 136160798Sjhb vp->v_iflag |= VI_OLOCK; 137160798Sjhb VI_UNLOCK(vp); 13821776Sbde 13921776Sbde /* 14021776Sbde * If the object is being terminated, wait for it to 141160798Sjhb * go away. 142146806Srwatson */ 143160798Sjhb while ((object = vp->v_object) != NULL) { 144160798Sjhb VM_OBJECT_LOCK(object); 145160798Sjhb if ((object->flags & OBJ_DEAD) == 0) 146162373Srwatson break; 147146806Srwatson msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0); 148160798Sjhb } 149146806Srwatson 150160798Sjhb if (vp->v_usecount == 0) 151160798Sjhb panic("vnode_pager_alloc: no vnode reference"); 152160798Sjhb 153176215Sru if (object == NULL) { 154176215Sru /* 155160798Sjhb * And an object of the appropriate size 156146806Srwatson */ 157160798Sjhb object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); 158146806Srwatson 159160798Sjhb object->un_pager.vnp.vnp_size = size; 160160798Sjhb 161160798Sjhb object->handle = handle; 162146806Srwatson vp->v_object = object; 163146806Srwatson } else { 164162991Srwatson object->ref_count++; 165146806Srwatson VM_OBJECT_UNLOCK(object); 166160798Sjhb } 167146806Srwatson VI_LOCK(vp); 168160798Sjhb vp->v_usecount++; 169146806Srwatson vp->v_iflag &= ~VI_OLOCK; 170146806Srwatson if (vp->v_iflag & VI_OWANT) { 171160798Sjhb vp->v_iflag &= ~VI_OWANT; 172160798Sjhb wakeup(vp); 173160798Sjhb } 174146806Srwatson VI_UNLOCK(vp); 175160798Sjhb mtx_unlock(&Giant); 176146806Srwatson return (object); 177160798Sjhb} 178160798Sjhb 179146806Srwatson/* 180160798Sjhb * The object must be locked. 181146806Srwatson */ 182146806Srwatsonstatic void 183146806Srwatsonvnode_pager_dealloc(object) 184160798Sjhb vm_object_t object; 185146806Srwatson{ 186160798Sjhb struct vnode *vp = object->handle; 187146806Srwatson 188160798Sjhb if (vp == NULL) 189146806Srwatson panic("vnode_pager_dealloc: pager already dealloced"); 190160798Sjhb 191160798Sjhb VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 192160798Sjhb vm_object_pip_wait(object, "vnpdea"); 193146806Srwatson 194160798Sjhb object->handle = NULL; 195160798Sjhb object->type = OBJT_DEAD; 196160798Sjhb ASSERT_VOP_LOCKED(vp, "vnode_pager_dealloc"); 197146806Srwatson vp->v_object = NULL; 198160798Sjhb vp->v_vflag &= ~(VV_TEXT | VV_OBJBUF); 199146806Srwatson} 200146806Srwatson 201160798Sjhbstatic boolean_t 202146806Srwatsonvnode_pager_haspage(object, pindex, before, after) 203146806Srwatson vm_object_t object; 204160798Sjhb vm_pindex_t pindex; 205160798Sjhb int *before; 206146806Srwatson int *after; 207160798Sjhb{ 208123750Speter struct vnode *vp = object->handle; 20912216Sbde daddr_t bn; 210160798Sjhb int err; 211146806Srwatson daddr_t reqblock; 212146806Srwatson int poff; 213160798Sjhb int bsize; 214160798Sjhb int pagesperblock, blocksperpage; 215146806Srwatson 216160798Sjhb VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 217146806Srwatson /* 218160798Sjhb * If no vp or vp is doomed or marked transparent to VM, we do not 219146806Srwatson * have the page. 220194390Sjhb */ 221146806Srwatson if (vp == NULL) 222160798Sjhb return FALSE; 223160798Sjhb 224146806Srwatson VI_LOCK(vp); 225160798Sjhb if (vp->v_iflag & VI_DOOMED) { 226146806Srwatson VI_UNLOCK(vp); 227160798Sjhb return FALSE; 228146806Srwatson } 229160798Sjhb VI_UNLOCK(vp); 230146806Srwatson /* 231160798Sjhb * If filesystem no longer mounted or offset beyond end of file we do 232146806Srwatson * not have the page. 233160798Sjhb */ 234146806Srwatson if ((vp->v_mount == NULL) || 235160798Sjhb (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 236146806Srwatson return FALSE; 237160798Sjhb 238160798Sjhb bsize = vp->v_mount->mnt_stat.f_iosize; 239160798Sjhb pagesperblock = bsize / PAGE_SIZE; 24021776Sbde blocksperpage = 0; 24121776Sbde if (pagesperblock > 0) { 242160798Sjhb reqblock = pindex / pagesperblock; 243146806Srwatson } else { 244160798Sjhb blocksperpage = (PAGE_SIZE / bsize); 245146806Srwatson reqblock = pindex * blocksperpage; 246160798Sjhb } 247146806Srwatson VM_OBJECT_UNLOCK(object); 248146806Srwatson err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before); 249160798Sjhb VM_OBJECT_LOCK(object); 250146806Srwatson if (err) 251160798Sjhb return TRUE; 252146806Srwatson if (bn == -1) 253160798Sjhb return FALSE; 254146806Srwatson if (pagesperblock > 0) { 255146806Srwatson poff = pindex - (reqblock * pagesperblock); 256160798Sjhb if (before) { 257146806Srwatson *before *= pagesperblock; 258160798Sjhb *before += poff; 259146806Srwatson } 260160798Sjhb if (after) { 261146806Srwatson int numafter; 262160798Sjhb *after *= pagesperblock; 263160798Sjhb numafter = pagesperblock - (poff + 1); 264194390Sjhb if (IDX_TO_OFF(pindex + numafter) > 265146806Srwatson object->un_pager.vnp.vnp_size) { 266146806Srwatson numafter = 267146806Srwatson OFF_TO_IDX(object->un_pager.vnp.vnp_size) - 268160798Sjhb pindex; 269160798Sjhb } 270160798Sjhb *after += numafter; 271160798Sjhb } 272160798Sjhb } else { 273160798Sjhb if (before) { 274160798Sjhb *before /= blocksperpage; 275160798Sjhb } 276146806Srwatson 277160798Sjhb if (after) { 278160798Sjhb *after /= blocksperpage; 279146806Srwatson } 280160798Sjhb } 281160798Sjhb return TRUE; 282160798Sjhb} 283146806Srwatson 284146806Srwatson/* 285160798Sjhb * Lets the VM system know about a change in size for a file. 286146806Srwatson * We adjust our own internal size and flush any cached pages in 287160798Sjhb * the associated object that are affected by the size change. 288146806Srwatson * 289160798Sjhb * Note: this routine may be invoked as a result of a pager put 290160798Sjhb * operation (possibly at object termination time), so we must be careful. 291160798Sjhb */ 292146806Srwatsonvoid 293160798Sjhbvnode_pager_setsize(vp, nsize) 294146806Srwatson struct vnode *vp; 295160798Sjhb vm_ooffset_t nsize; 296160798Sjhb{ 297160798Sjhb vm_object_t object; 298146806Srwatson vm_page_t m; 299160798Sjhb vm_pindex_t nobjsize; 300194390Sjhb 301146806Srwatson if ((object = vp->v_object) == NULL) 302146806Srwatson return; 3031541Srgrimes VM_OBJECT_LOCK(object); 3041541Srgrimes if (nsize == object->un_pager.vnp.vnp_size) { 3051541Srgrimes /* 3061541Srgrimes * Hasn't changed size 3071541Srgrimes */ 308146806Srwatson VM_OBJECT_UNLOCK(object); 309146806Srwatson return; 310146806Srwatson } 311177633Sdfr nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); 312177633Sdfr if (nsize < object->un_pager.vnp.vnp_size) { 31330740Sphk /* 314161325Sjhb * File has shrunk. Toss any cached pages beyond the new EOF. 315160798Sjhb */ 316146806Srwatson if (nobjsize < object->size) 317160798Sjhb vm_object_page_remove(object, nobjsize, object->size, 318146806Srwatson FALSE); 319160798Sjhb /* 320146806Srwatson * this gets rid of garbage at the end of a page that is now 321146806Srwatson * only partially backed by the vnode. 322160798Sjhb * 323146806Srwatson * XXX for some reason (I don't know yet), if we take a 324160798Sjhb * completely invalid page and mark it partially valid 325146806Srwatson * it can screw up NFS reads, so we don't allow the case. 326184789Sed */ 327146806Srwatson if ((nsize & PAGE_MASK) && 328184789Sed (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL) { 329146806Srwatson vm_page_lock_queues(); 330184789Sed if (m->valid) { 331161952Srwatson int base = (int)nsize & PAGE_MASK; 332161952Srwatson int size = PAGE_SIZE - base; 333146806Srwatson 334146806Srwatson /* 335146806Srwatson * Clear out partial-page garbage in case 336160798Sjhb * the page has been mapped. 337146806Srwatson */ 338123750Speter pmap_zero_page_area(m, base, size); 339160798Sjhb 340146806Srwatson /* 341123750Speter * XXX work around SMP data integrity race 342160798Sjhb * by unmapping the page from user processes. 343146806Srwatson * The garbage we just cleared may be mapped 344123750Speter * to a user process running on another cpu 345146806Srwatson * and this code is not running through normal 346171209Speter * I/O channels which handle SMP issues for 347146806Srwatson * us, so unmap page to synchronize all cpus. 348171209Speter * 349171209Speter * XXX should vm_pager_unmap_page() have 350146806Srwatson * dealt with this? 351178888Sjulian */ 352161946Srwatson pmap_remove_all(m); 353146806Srwatson 354146806Srwatson /* 355146806Srwatson * Clear out partial-page dirty bits. This 356146806Srwatson * has the side effect of setting the valid 3571541Srgrimes * bits, but that is ok. There are a bunch 35849428Sjkh * of places in the VM system where we expected 359160798Sjhb * m->dirty == VM_PAGE_BITS_ALL. The file EOF 360160798Sjhb * case is one of them. If the page is still 361160798Sjhb * partially dirty, make it fully dirty. 362146806Srwatson * 363146806Srwatson * note that we do not clear out the valid 364146806Srwatson * bits. This would prevent bogus_page 365146806Srwatson * replacement from working properly. 366160798Sjhb */ 367160798Sjhb vm_page_set_validclean(m, base, size); 368160798Sjhb if (m->dirty != 0) 369160798Sjhb m->dirty = VM_PAGE_BITS_ALL; 370160798Sjhb } 371146806Srwatson vm_page_unlock_queues(); 372160798Sjhb } 373146806Srwatson } 374146806Srwatson object->un_pager.vnp.vnp_size = nsize; 375160798Sjhb object->size = nobjsize; 376146806Srwatson VM_OBJECT_UNLOCK(object); 377146806Srwatson} 378160798Sjhb 379146806Srwatson/* 380171209Speter * calculate the linear (byte) disk address of specified virtual 381171209Speter * file address 382171209Speter */ 383183361Sjhbstatic vm_offset_t 384146806Srwatsonvnode_pager_addr(vp, address, run) 385171209Speter struct vnode *vp; 386171209Speter vm_ooffset_t address; 387171209Speter int *run; 388146806Srwatson{ 389171209Speter int rtaddress; 390146806Srwatson int bsize; 391160798Sjhb daddr_t block; 392146806Srwatson int err; 393146806Srwatson daddr_t vblock; 394160798Sjhb int voffset; 395160798Sjhb 396160798Sjhb GIANT_REQUIRED; 397160798Sjhb if ((int) address < 0) 398160798Sjhb return -1; 399146806Srwatson 400160798Sjhb if (vp->v_mount == NULL) 401146806Srwatson return -1; 4022124Sdg 4032124Sdg bsize = vp->v_mount->mnt_stat.f_iosize; 4042124Sdg vblock = address / bsize; 4052124Sdg voffset = address % bsize; 406209579Skib 407209579Skib err = VOP_BMAP(vp, vblock, NULL, &block, run, NULL); 408209579Skib 409209579Skib if (err || (block == -1)) 410209579Skib rtaddress = -1; 411209579Skib else { 412209579Skib rtaddress = block + voffset / DEV_BSIZE; 413209579Skib if (run) { 414209579Skib *run += 1; 415209579Skib *run *= bsize/PAGE_SIZE; 41612864Speter *run -= voffset/PAGE_SIZE; 41712864Speter } 41814215Speter } 419194910Sjhb 420194910Sjhb return rtaddress; 421160798Sjhb} 422146806Srwatson 423160798Sjhb/* 424146806Srwatson * interrupt routine for I/O completion 425146806Srwatson */ 426194910Sjhbstatic void 427194910Sjhbvnode_pager_iodone(bp) 428160798Sjhb struct buf *bp; 429160798Sjhb{ 430146806Srwatson bp->b_flags |= B_DONE; 431160798Sjhb wakeup(bp); 432146806Srwatson} 433160798Sjhb 434146806Srwatson/* 435194910Sjhb * small block filesystem vnode pager input 436194910Sjhb */ 437160798Sjhbstatic int 438160798Sjhbvnode_pager_input_smlfs(object, m) 439146806Srwatson vm_object_t object; 44014219Speter vm_page_t m; 441160798Sjhb{ 442146806Srwatson int i; 443161952Srwatson int s; 444161952Srwatson struct vnode *dp, *vp; 445146806Srwatson struct buf *bp; 446160798Sjhb vm_offset_t kva; 447146806Srwatson int fileaddr; 448160798Sjhb vm_offset_t bsize; 449156134Sdavidxu int error = 0; 450160798Sjhb 451160798Sjhb GIANT_REQUIRED; 452151576Sdavidxu 453151576Sdavidxu vp = object->handle; 454160798Sjhb if (vp->v_mount == NULL) 455151576Sdavidxu return VM_PAGER_BAD; 456160798Sjhb 457160798Sjhb bsize = vp->v_mount->mnt_stat.f_iosize; 458146806Srwatson 459146806Srwatson VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 460146806Srwatson 461146806Srwatson kva = vm_pager_map_page(m); 462146806Srwatson 463146806Srwatson for (i = 0; i < PAGE_SIZE / bsize; i++) { 464146806Srwatson vm_ooffset_t address; 465146806Srwatson 466160798Sjhb if (vm_page_bits(i * bsize, bsize) & m->valid) 467146806Srwatson continue; 46814219Speter 469160798Sjhb address = IDX_TO_OFF(m->pindex) + i * bsize; 470146806Srwatson if (address >= object->un_pager.vnp.vnp_size) { 471160798Sjhb fileaddr = -1; 472160798Sjhb } else { 473146806Srwatson fileaddr = vnode_pager_addr(vp, address, NULL); 474160798Sjhb } 475160798Sjhb if (fileaddr != -1) { 476160798Sjhb bp = getpbuf(&vnode_pbuf_freecnt); 477160798Sjhb 478160798Sjhb /* build a minimal buffer header */ 479151867Sdavidxu bp->b_iocmd = BIO_READ; 480151867Sdavidxu bp->b_iodone = vnode_pager_iodone; 481152845Sdavidxu KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 482152845Sdavidxu KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 483152845Sdavidxu bp->b_rcred = crhold(curthread->td_ucred); 484152845Sdavidxu bp->b_wcred = crhold(curthread->td_ucred); 485152845Sdavidxu bp->b_data = (caddr_t) kva + i * bsize; 486152845Sdavidxu bp->b_blkno = fileaddr; 487146806Srwatson pbgetvp(dp, bp); 488146806Srwatson bp->b_bcount = bsize; 489146806Srwatson bp->b_bufsize = bsize; 490146806Srwatson bp->b_runningbufspace = bp->b_bufsize; 491146806Srwatson runningbufspace += bp->b_runningbufspace; 492146806Srwatson 493146806Srwatson /* do the input */ 494146806Srwatson VOP_SPECSTRATEGY(bp->b_vp, bp); 495160798Sjhb 496146806Srwatson /* we definitely need to be at splvm here */ 497146806Srwatson 498160798Sjhb s = splvm(); 499160798Sjhb while ((bp->b_flags & B_DONE) == 0) { 500146806Srwatson tsleep(bp, PVM, "vnsrd", 0); 501146806Srwatson } 502160798Sjhb splx(s); 503146806Srwatson if ((bp->b_ioflags & BIO_ERROR) != 0) 504160798Sjhb error = EIO; 505146806Srwatson 506160798Sjhb /* 507160798Sjhb * free the buffer header back to the swap buffer pool 508160798Sjhb */ 509146806Srwatson relpbuf(bp, &vnode_pbuf_freecnt); 510146806Srwatson if (error) 511146806Srwatson break; 512146806Srwatson 513146806Srwatson vm_page_lock_queues(); 514146806Srwatson vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 515146806Srwatson vm_page_unlock_queues(); 516146806Srwatson } else { 517147813Sjhb vm_page_lock_queues(); 518161952Srwatson vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 519147813Sjhb vm_page_unlock_queues(); 520161952Srwatson bzero((caddr_t) kva + i * bsize, bsize); 521147813Sjhb } 522146806Srwatson } 523146806Srwatson vm_pager_unmap_page(kva); 524146806Srwatson vm_page_lock_queues(); 525146806Srwatson pmap_clear_modify(m); 526146806Srwatson vm_page_flag_clear(m, PG_ZERO); 527146806Srwatson vm_page_unlock_queues(); 52851138Salfred if (error) { 529160798Sjhb return VM_PAGER_ERROR; 530146806Srwatson } 531146806Srwatson return VM_PAGER_OK; 532160798Sjhb 533146806Srwatson} 534160798Sjhb 535146806Srwatson 53625537Sdfr/* 537160798Sjhb * old style vnode pager output routine 538160798Sjhb */ 539146806Srwatsonstatic int 540160798Sjhbvnode_pager_input_old(object, m) 541160798Sjhb vm_object_t object; 542160798Sjhb vm_page_t m; 543160798Sjhb{ 544160798Sjhb struct uio auio; 545160798Sjhb struct iovec aiov; 546160798Sjhb int error; 547146806Srwatson int size; 548160798Sjhb vm_offset_t kva; 549160798Sjhb struct vnode *vp; 550160798Sjhb 551146806Srwatson GIANT_REQUIRED; 552160798Sjhb error = 0; 553146806Srwatson 554146806Srwatson /* 555160798Sjhb * Return failure if beyond current EOF 556160798Sjhb */ 557146806Srwatson if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 558146806Srwatson return VM_PAGER_BAD; 559160798Sjhb } else { 560146806Srwatson size = PAGE_SIZE; 561160798Sjhb if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 562160798Sjhb size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 563160798Sjhb 564160798Sjhb /* 565151867Sdavidxu * Allocate a kernel virtual address and initialize so that 566151867Sdavidxu * we can use VOP_READ/WRITE routines. 567160798Sjhb */ 568146806Srwatson kva = vm_pager_map_page(m); 569146806Srwatson 570160798Sjhb vp = object->handle; 571160798Sjhb aiov.iov_base = (caddr_t) kva; 572161952Srwatson aiov.iov_len = size; 57334925Sdufault auio.uio_iov = &aiov; 574160798Sjhb auio.uio_iovcnt = 1; 575146806Srwatson auio.uio_offset = IDX_TO_OFF(m->pindex); 576160798Sjhb auio.uio_segflg = UIO_SYSSPACE; 577146806Srwatson auio.uio_rw = UIO_READ; 57834925Sdufault auio.uio_resid = size; 579160798Sjhb auio.uio_td = curthread; 580146806Srwatson 581146806Srwatson error = VOP_READ(vp, &auio, 0, curthread->td_ucred); 582160798Sjhb if (!error) { 58334925Sdufault int count = size - auio.uio_resid; 584160798Sjhb 585160798Sjhb if (count == 0) 586160798Sjhb error = EINVAL; 587160798Sjhb else if (count != PAGE_SIZE) 588146806Srwatson bzero((caddr_t) kva + count, PAGE_SIZE - count); 589160798Sjhb } 590160798Sjhb vm_pager_unmap_page(kva); 591146806Srwatson } 592146806Srwatson vm_page_lock_queues(); 593146806Srwatson pmap_clear_modify(m); 594160798Sjhb vm_page_undirty(m); 595146806Srwatson vm_page_flag_clear(m, PG_ZERO); 596160798Sjhb if (!error) 597211998Skib m->valid = VM_PAGE_BITS_ALL; 598211998Skib vm_page_unlock_queues(); 599211998Skib return error ? VM_PAGER_ERROR : VM_PAGER_OK; 600160798Sjhb} 601146806Srwatson 602160798Sjhb/* 603160798Sjhb * generic vnode pager input routine 604146806Srwatson */ 605146806Srwatson 606160798Sjhb/* 607160798Sjhb * Local media VFS's that do not implement their own VOP_GETPAGES 608146806Srwatson * should have their VOP_GETPAGES call to vnode_pager_generic_getpages() 609160798Sjhb * to implement the previous behaviour. 610146806Srwatson * 611146806Srwatson * All other FS's should use the bypass to get to the local media 612160798Sjhb * backing vp's VOP_GETPAGES. 613146806Srwatson */ 614160798Sjhbstatic int 615146806Srwatsonvnode_pager_getpages(object, m, count, reqpage) 616160798Sjhb vm_object_t object; 617146806Srwatson vm_page_t *m; 618160798Sjhb int count; 619146806Srwatson int reqpage; 620160798Sjhb{ 621146806Srwatson int rtval; 622160798Sjhb struct vnode *vp; 623146806Srwatson int bytes = count * PAGE_SIZE; 624160798Sjhb 625146806Srwatson vp = object->handle; 626160798Sjhb VM_OBJECT_UNLOCK(object); 627146806Srwatson rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0); 628160798Sjhb KASSERT(rtval != EOPNOTSUPP, 629146806Srwatson ("vnode_pager: FS getpages not implemented\n")); 630160798Sjhb VM_OBJECT_LOCK(object); 631146806Srwatson return rtval; 632146806Srwatson} 633160798Sjhb 634160111Swsalamon/* 635160111Swsalamon * This is now called from local media FS's to operate against their 636160111Swsalamon * own vnodes if they fail to implement VOP_GETPAGES. 637160798Sjhb */ 638160111Swsalamonint 639160111Swsalamonvnode_pager_generic_getpages(vp, m, bytecount, reqpage) 640160111Swsalamon struct vnode *vp; 641160798Sjhb vm_page_t *m; 642146806Srwatson int bytecount; 643146806Srwatson int reqpage; 644160798Sjhb{ 645146806Srwatson vm_object_t object; 646146806Srwatson vm_offset_t kva; 647160798Sjhb off_t foff, tfoff, nextoff; 648146806Srwatson int i, j, size, bsize, first, firstaddr; 649160798Sjhb struct vnode *dp; 650146806Srwatson int runpg; 651161952Srwatson int runend; 652160798Sjhb struct buf *bp; 653146806Srwatson int s; 654146806Srwatson int count; 655146806Srwatson int error = 0; 656146806Srwatson 657146806Srwatson GIANT_REQUIRED; 658146806Srwatson object = vp->v_object; 659146806Srwatson count = bytecount / PAGE_SIZE; 660146806Srwatson 661146806Srwatson if (vp->v_mount == NULL) 662183361Sjhb return VM_PAGER_BAD; 663160798Sjhb 664146806Srwatson bsize = vp->v_mount->mnt_stat.f_iosize; 665146806Srwatson 666160798Sjhb /* get the UNDERLYING device for the file with VOP_BMAP() */ 667146806Srwatson 668146806Srwatson /* 669160798Sjhb * originally, we did not check for an error return value -- assuming 670146806Srwatson * an fs always has a bmap entry point -- that assumption is wrong!!! 671146806Srwatson */ 672160798Sjhb foff = IDX_TO_OFF(m[reqpage]->pindex); 673194383Sjhb 674160798Sjhb /* 675211998Skib * if we can't bmap, use old VOP code 676211998Skib */ 677211998Skib if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 678160798Sjhb VM_OBJECT_LOCK(object); 679146806Srwatson vm_page_lock_queues(); 680177091Sjeff for (i = 0; i < count; i++) 681177091Sjeff if (i != reqpage) 682177091Sjeff vm_page_free(m[i]); 683177091Sjeff vm_page_unlock_queues(); 684177091Sjeff VM_OBJECT_UNLOCK(object); 685160798Sjhb cnt.v_vnodein++; 686160798Sjhb cnt.v_vnodepgsin++; 687160798Sjhb return vnode_pager_input_old(object, m[reqpage]); 688146806Srwatson 689160798Sjhb /* 690146806Srwatson * if the blocksize is smaller than a page size, then use 691160798Sjhb * special small filesystem code. NFS sometimes has a small 692146806Srwatson * blocksize, but it can handle large reads itself. 693160798Sjhb */ 694146806Srwatson } else if ((PAGE_SIZE / bsize) > 1 && 695160798Sjhb (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) { 696146806Srwatson VM_OBJECT_LOCK(object); 697160798Sjhb vm_page_lock_queues(); 698160798Sjhb for (i = 0; i < count; i++) 699146806Srwatson if (i != reqpage) 700160798Sjhb vm_page_free(m[i]); 701146806Srwatson vm_page_unlock_queues(); 702146806Srwatson VM_OBJECT_UNLOCK(object); 703160798Sjhb cnt.v_vnodein++; 704146806Srwatson cnt.v_vnodepgsin++; 705160798Sjhb return vnode_pager_input_smlfs(object, m[reqpage]); 706146806Srwatson } 707160798Sjhb 708146806Srwatson /* 709160798Sjhb * If we have a completely valid page available to us, we can 710161952Srwatson * clean up and return. Otherwise we have to re-read the 711146806Srwatson * media. 712146806Srwatson */ 713160798Sjhb if (m[reqpage]->valid == VM_PAGE_BITS_ALL) { 714160798Sjhb VM_OBJECT_LOCK(object); 715160798Sjhb vm_page_lock_queues(); 716160798Sjhb for (i = 0; i < count; i++) 717160798Sjhb if (i != reqpage) 718146806Srwatson vm_page_free(m[i]); 719160798Sjhb vm_page_unlock_queues(); 720146806Srwatson VM_OBJECT_UNLOCK(object); 721146806Srwatson return VM_PAGER_OK; 722160798Sjhb } 723160798Sjhb m[reqpage]->valid = 0; 724160798Sjhb 725160798Sjhb /* 726146806Srwatson * here on direct device I/O 727160798Sjhb */ 728146806Srwatson firstaddr = -1; 729160798Sjhb 730146806Srwatson /* 731160798Sjhb * calculate the run that includes the required page 732160111Swsalamon */ 733160111Swsalamon for (first = 0, i = 0; i < count; i = runend) { 734160111Swsalamon firstaddr = vnode_pager_addr(vp, 735160798Sjhb IDX_TO_OFF(m[i]->pindex), &runpg); 736160111Swsalamon if (firstaddr == -1) { 737160111Swsalamon VM_OBJECT_LOCK(object); 738160111Swsalamon if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 739160798Sjhb panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx", 740160111Swsalamon firstaddr, (uintmax_t)(foff >> 32), 741146806Srwatson (uintmax_t)foff, 742160798Sjhb (uintmax_t) 743146806Srwatson (object->un_pager.vnp.vnp_size >> 32), 744160798Sjhb (uintmax_t)object->un_pager.vnp.vnp_size); 745146806Srwatson } 746146806Srwatson vm_page_lock_queues(); 747160798Sjhb vm_page_free(m[i]); 748146806Srwatson vm_page_unlock_queues(); 749146806Srwatson VM_OBJECT_UNLOCK(object); 750146806Srwatson runend = i + 1; 751146806Srwatson first = runend; 752160798Sjhb continue; 753160798Sjhb } 754146806Srwatson runend = i + runpg; 755160798Sjhb if (runend <= reqpage) { 756146806Srwatson VM_OBJECT_LOCK(object); 757160798Sjhb vm_page_lock_queues(); 758160798Sjhb for (j = i; j < runend; j++) 759146806Srwatson vm_page_free(m[j]); 760160798Sjhb vm_page_unlock_queues(); 761146806Srwatson VM_OBJECT_UNLOCK(object); 762160798Sjhb } else { 763146806Srwatson if (runpg < (count - first)) { 764160798Sjhb VM_OBJECT_LOCK(object); 765146806Srwatson vm_page_lock_queues(); 766160798Sjhb for (i = first + runpg; i < count; i++) 767146806Srwatson vm_page_free(m[i]); 768160798Sjhb vm_page_unlock_queues(); 769146806Srwatson VM_OBJECT_UNLOCK(object); 770160798Sjhb count = first + runpg; 771160798Sjhb } 772160798Sjhb break; 773160798Sjhb } 774160798Sjhb first = runend; 775160798Sjhb } 776160798Sjhb 777146806Srwatson /* 778146806Srwatson * the first and last page have been calculated now, move input pages 779160798Sjhb * to be zero based... 780146806Srwatson */ 781146806Srwatson if (first != 0) { 782160798Sjhb for (i = first; i < count; i++) { 783146806Srwatson m[i - first] = m[i]; 784146806Srwatson } 785177091Sjeff count -= first; 786160798Sjhb reqpage -= first; 787151445Sstefanf } 788160798Sjhb 789146806Srwatson /* 790160798Sjhb * calculate the file virtual address for the transfer 791161952Srwatson */ 792160798Sjhb foff = IDX_TO_OFF(m[0]->pindex); 793146806Srwatson 794160798Sjhb /* 795146806Srwatson * calculate the size of the transfer 796160798Sjhb */ 797160798Sjhb size = count * PAGE_SIZE; 798160798Sjhb if ((foff + size) > object->un_pager.vnp.vnp_size) 799160798Sjhb size = object->un_pager.vnp.vnp_size - foff; 800160798Sjhb 801146806Srwatson /* 802146806Srwatson * round up physical size for real devices. 803160798Sjhb */ 804146806Srwatson if (dp->v_type == VBLK || dp->v_type == VCHR) { 805146806Srwatson int secmask = dp->v_rdev->si_bsize_phys - 1; 806160798Sjhb KASSERT(secmask < PAGE_SIZE, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask + 1)); 807161678Sdavidxu size = (size + secmask) & ~secmask; 808163449Sdavidxu } 809160798Sjhb 810146806Srwatson bp = getpbuf(&vnode_pbuf_freecnt); 811160798Sjhb kva = (vm_offset_t) bp->b_data; 812160798Sjhb 813152845Sdavidxu /* 814160798Sjhb * and map the pages to be read into the kva 815152845Sdavidxu */ 816152845Sdavidxu pmap_qenter(kva, m, count); 817160798Sjhb 818152845Sdavidxu /* build a minimal buffer header */ 819152845Sdavidxu bp->b_iocmd = BIO_READ; 820152845Sdavidxu bp->b_iodone = vnode_pager_iodone; 821160798Sjhb /* B_PHYS is not set, but it is nice to fill this in */ 822152845Sdavidxu KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); 823152845Sdavidxu KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); 824152845Sdavidxu bp->b_rcred = crhold(curthread->td_ucred); 825160798Sjhb bp->b_wcred = crhold(curthread->td_ucred); 826152845Sdavidxu bp->b_blkno = firstaddr; 827160798Sjhb pbgetvp(dp, bp); 828160798Sjhb bp->b_bcount = size; 829160798Sjhb bp->b_bufsize = size; 830160798Sjhb bp->b_runningbufspace = bp->b_bufsize; 831162497Sdavidxu runningbufspace += bp->b_runningbufspace; 832162497Sdavidxu 833162497Sdavidxu cnt.v_vnodein++; 834162497Sdavidxu cnt.v_vnodepgsin += count; 835161367Speter 836161367Speter /* do the input */ 837163953Srrs if (dp->v_type == VCHR) 838163953Srrs VOP_SPECSTRATEGY(bp->b_vp, bp); 839163953Srrs else 840163953Srrs VOP_STRATEGY(bp->b_vp, bp); 841163953Srrs 842163953Srrs s = splvm(); 843163953Srrs /* we definitely need to be at splvm here */ 844163953Srrs 845163953Srrs while ((bp->b_flags & B_DONE) == 0) { 846163953Srrs tsleep(bp, PVM, "vnread", 0); 847171209Speter } 848171209Speter splx(s); 849171209Speter if ((bp->b_ioflags & BIO_ERROR) != 0) 850171209Speter error = EIO; 851171209Speter 852171209Speter if (!error) { 853171209Speter if (size != count * PAGE_SIZE) 854171209Speter bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 855171209Speter } 856171209Speter pmap_qremove(kva, count); 857171859Sdavidxu 858175517Srwatson /* 859175164Sjhb * free the buffer header back to the swap buffer pool 860175517Srwatson */ 861176730Sjeff relpbuf(bp, &vnode_pbuf_freecnt); 862176730Sjeff 863176730Sjeff VM_OBJECT_LOCK(object); 864176730Sjeff vm_page_lock_queues(); 865176730Sjeff for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { 866176730Sjeff vm_page_t mt; 867176730Sjeff 868177597Sru nextoff = tfoff + PAGE_SIZE; 869177597Sru mt = m[i]; 870176730Sjeff 871177597Sru if (nextoff <= object->un_pager.vnp.vnp_size) { 872177597Sru /* 873177788Skib * Read filled up entire page. 874177788Skib */ 875177788Skib mt->valid = VM_PAGE_BITS_ALL; 876177788Skib vm_page_undirty(mt); /* should be an assert? XXX */ 877177788Skib pmap_clear_modify(mt); 878177788Skib } else { 879177788Skib /* 880177788Skib * Read did not fill up entire page. Since this 881177788Skib * is getpages, the page may be mapped, so we have 882177788Skib * to zero the invalid portions of the page even 883177788Skib * though we aren't setting them valid. 884177788Skib * 885177788Skib * Currently we do not set the entire page valid, 886177788Skib * we just try to clear the piece that we couldn't 887177788Skib * read. 888177788Skib */ 889177788Skib vm_page_set_validclean(mt, 0, 890177788Skib object->un_pager.vnp.vnp_size - tfoff); 891177788Skib /* handled by vm_fault now */ 892177788Skib /* vm_page_zero_invalid(mt, FALSE); */ 893177788Skib } 894177788Skib 895177788Skib vm_page_flag_clear(mt, PG_ZERO); 896177788Skib if (i != reqpage) { 897177788Skib 898177788Skib /* 899177788Skib * whether or not to leave the page activated is up in 900177788Skib * the air, but we should put the page on a page queue 901182123Srwatson * somewhere. (it already is in the object). Result: 902184588Sdfr * It appears that empirical results show that 903184588Sdfr * deactivating pages is best. 904191673Sjamie */ 905191673Sjamie 906191673Sjamie /* 907191673Sjamie * just in case someone was asking for this page we 908191673Sjamie * now tell them that it is ok to use 909194262Sjhb */ 910194910Sjhb if (!error) { 911194910Sjhb if (mt->flags & PG_WANTED) 912194910Sjhb vm_page_activate(mt); 913194910Sjhb else 914194910Sjhb vm_page_deactivate(mt); 915194910Sjhb vm_page_wakeup(mt); 916195458Strasz } else { 917224066Sjonathan vm_page_free(mt); 918224066Sjonathan } 919224066Sjonathan } 920219129Srwatson } 921219129Srwatson vm_page_unlock_queues(); 922224987Sjonathan VM_OBJECT_UNLOCK(object); 923224987Sjonathan if (error) { 924224987Sjonathan printf("vnode_pager_getpages: I/O read error\n"); 925224987Sjonathan } 926198508Skib return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 927198508Skib} 928198508Skib 929198508Skib/* 930219304Strasz * EOPNOTSUPP is no longer legal. For local media VFS's that do not 931219304Strasz * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to 932219304Strasz * vnode_pager_generic_putpages() to implement the previous behaviour. 933220163Strasz * 934220163Strasz * All other FS's should use the bypass to get to the local media 935220163Strasz * backing vp's VOP_PUTPAGES. 936220163Strasz */ 937220163Straszstatic void 938220163Straszvnode_pager_putpages(object, m, count, sync, rtvals) 939220163Strasz vm_object_t object; 940220163Strasz vm_page_t *m; 941220163Strasz int count; 942220163Strasz boolean_t sync; 943220163Strasz int *rtvals; 944220163Strasz{ 945220163Strasz int rtval; 946220163Strasz struct vnode *vp; 947220163Strasz struct mount *mp; 948220791Smdf int bytes = count * PAGE_SIZE; 949220791Smdf 950220791Smdf GIANT_REQUIRED; 951105144Speter /* 952123408Speter * Force synchronous operation if we are extremely low on memory 953 * to prevent a low-memory deadlock. VOP operations often need to 954 * allocate more memory to initiate the I/O ( i.e. do a BMAP 955 * operation ). The swapper handles the case by limiting the amount 956 * of asynchronous I/O, but that sort of solution doesn't scale well 957 * for the vnode pager without a lot of work. 958 * 959 * Also, the backing vnode's iodone routine may not wake the pageout 960 * daemon up. This should be probably be addressed XXX. 961 */ 962 963 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min) 964 sync |= OBJPC_SYNC; 965 966 /* 967 * Call device-specific putpages function 968 */ 969 vp = object->handle; 970 if (vp->v_type != VREG) 971 mp = NULL; 972 (void)vn_start_write(vp, &mp, V_WAIT); 973 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0); 974 KASSERT(rtval != EOPNOTSUPP, 975 ("vnode_pager: stale FS putpages\n")); 976 vn_finished_write(mp); 977} 978 979 980/* 981 * This is now called from local media FS's to operate against their 982 * own vnodes if they fail to implement VOP_PUTPAGES. 983 * 984 * This is typically called indirectly via the pageout daemon and 985 * clustering has already typically occured, so in general we ask the 986 * underlying filesystem to write the data out asynchronously rather 987 * then delayed. 988 */ 989int 990vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) 991 struct vnode *vp; 992 vm_page_t *m; 993 int bytecount; 994 int flags; 995 int *rtvals; 996{ 997 int i; 998 vm_object_t object; 999 int count; 1000 1001 int maxsize, ncount; 1002 vm_ooffset_t poffset; 1003 struct uio auio; 1004 struct iovec aiov; 1005 int error; 1006 int ioflags; 1007 1008 GIANT_REQUIRED; 1009 object = vp->v_object; 1010 count = bytecount / PAGE_SIZE; 1011 1012 for (i = 0; i < count; i++) 1013 rtvals[i] = VM_PAGER_AGAIN; 1014 1015 if ((int) m[0]->pindex < 0) { 1016 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n", 1017 (long)m[0]->pindex, m[0]->dirty); 1018 rtvals[0] = VM_PAGER_BAD; 1019 return VM_PAGER_BAD; 1020 } 1021 1022 maxsize = count * PAGE_SIZE; 1023 ncount = count; 1024 1025 poffset = IDX_TO_OFF(m[0]->pindex); 1026 1027 /* 1028 * If the page-aligned write is larger then the actual file we 1029 * have to invalidate pages occuring beyond the file EOF. However, 1030 * there is an edge case where a file may not be page-aligned where 1031 * the last page is partially invalid. In this case the filesystem 1032 * may not properly clear the dirty bits for the entire page (which 1033 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). 1034 * With the page locked we are free to fix-up the dirty bits here. 1035 * 1036 * We do not under any circumstances truncate the valid bits, as 1037 * this will screw up bogus page replacement. 1038 */ 1039 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 1040 if (object->un_pager.vnp.vnp_size > poffset) { 1041 int pgoff; 1042 1043 maxsize = object->un_pager.vnp.vnp_size - poffset; 1044 ncount = btoc(maxsize); 1045 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { 1046 vm_page_clear_dirty(m[ncount - 1], pgoff, 1047 PAGE_SIZE - pgoff); 1048 } 1049 } else { 1050 maxsize = 0; 1051 ncount = 0; 1052 } 1053 if (ncount < count) { 1054 for (i = ncount; i < count; i++) { 1055 rtvals[i] = VM_PAGER_BAD; 1056 } 1057 } 1058 } 1059 1060 /* 1061 * pageouts are already clustered, use IO_ASYNC t o force a bawrite() 1062 * rather then a bdwrite() to prevent paging I/O from saturating 1063 * the buffer cache. Dummy-up the sequential heuristic to cause 1064 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set, 1065 * the system decides how to cluster. 1066 */ 1067 ioflags = IO_VMIO; 1068 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) 1069 ioflags |= IO_SYNC; 1070 else if ((flags & VM_PAGER_CLUSTER_OK) == 0) 1071 ioflags |= IO_ASYNC; 1072 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; 1073 ioflags |= IO_SEQMAX << IO_SEQSHIFT; 1074 1075 aiov.iov_base = (caddr_t) 0; 1076 aiov.iov_len = maxsize; 1077 auio.uio_iov = &aiov; 1078 auio.uio_iovcnt = 1; 1079 auio.uio_offset = poffset; 1080 auio.uio_segflg = UIO_NOCOPY; 1081 auio.uio_rw = UIO_WRITE; 1082 auio.uio_resid = maxsize; 1083 auio.uio_td = (struct thread *) 0; 1084 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); 1085 cnt.v_vnodeout++; 1086 cnt.v_vnodepgsout += ncount; 1087 1088 if (error) { 1089 printf("vnode_pager_putpages: I/O error %d\n", error); 1090 } 1091 if (auio.uio_resid) { 1092 printf("vnode_pager_putpages: residual I/O %d at %lu\n", 1093 auio.uio_resid, (u_long)m[0]->pindex); 1094 } 1095 for (i = 0; i < ncount; i++) { 1096 rtvals[i] = VM_PAGER_OK; 1097 } 1098 return rtvals[0]; 1099} 1100 1101struct vnode * 1102vnode_pager_lock(object) 1103 vm_object_t object; 1104{ 1105 struct thread *td = curthread; /* XXX */ 1106 1107 GIANT_REQUIRED; 1108 1109 for (; object != NULL; object = object->backing_object) { 1110 if (object->type != OBJT_VNODE) 1111 continue; 1112 if (object->flags & OBJ_DEAD) { 1113 return NULL; 1114 } 1115 1116 /* XXX; If object->handle can change, we need to cache it. */ 1117 while (vget(object->handle, 1118 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, td)){ 1119 if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE)) 1120 return NULL; 1121 printf("vnode_pager_lock: retrying\n"); 1122 } 1123 return object->handle; 1124 } 1125 return NULL; 1126} 1127