vnode_pager.c revision 33109
1/* 2 * Copyright (c) 1990 University of Utah. 3 * Copyright (c) 1991 The Regents of the University of California. 4 * All rights reserved. 5 * Copyright (c) 1993, 1994 John S. Dyson 6 * Copyright (c) 1995, David Greenman 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the Systems Programming Group of the University of Utah Computer 10 * Science Department. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by the University of 23 * California, Berkeley and its contributors. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 * 40 * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 41 * $Id: vnode_pager.c,v 1.82 1998/02/04 22:34:03 eivind Exp $ 42 */ 43 44/* 45 * Page to/from files (vnodes). 46 */ 47 48/* 49 * TODO: 50 * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will 51 * greatly re-simplify the vnode_pager. 52 */ 53 54#include "opt_diagnostic.h" 55 56#include <sys/param.h> 57#include <sys/systm.h> 58#include <sys/proc.h> 59#include <sys/vnode.h> 60#include <sys/mount.h> 61#include <sys/buf.h> 62#include <sys/vmmeter.h> 63 64#include <vm/vm.h> 65#include <vm/vm_prot.h> 66#include <vm/vm_object.h> 67#include <vm/vm_page.h> 68#include <vm/vm_pager.h> 69#include <vm/vm_map.h> 70#include <vm/vnode_pager.h> 71#include <vm/vm_extern.h> 72 73static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address, 74 int *run)); 75static void vnode_pager_iodone __P((struct buf *bp)); 76static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m)); 77static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m)); 78static void vnode_pager_dealloc __P((vm_object_t)); 79static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 80static int vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *)); 81static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 82 83struct pagerops vnodepagerops = { 84 NULL, 85 vnode_pager_alloc, 86 vnode_pager_dealloc, 87 vnode_pager_getpages, 88 vnode_pager_putpages, 89 vnode_pager_haspage, 90 NULL 91}; 92 93static int vnode_pager_leaf_getpages __P((vm_object_t object, vm_page_t *m, 94 int count, int reqpage)); 95static int vnode_pager_leaf_putpages __P((vm_object_t object, vm_page_t *m, 96 int count, boolean_t sync, 97 int *rtvals)); 98 99/* 100 * Allocate (or lookup) pager for a vnode. 101 * Handle is a vnode pointer. 102 */ 103vm_object_t 104vnode_pager_alloc(void *handle, vm_size_t size, vm_prot_t prot, 105 vm_ooffset_t offset) 106{ 107 vm_object_t object; 108 struct vnode *vp; 109 110 /* 111 * Pageout to vnode, no can do yet. 112 */ 113 if (handle == NULL) 114 return (NULL); 115 116 vp = (struct vnode *) handle; 117 118 /* 119 * Prevent race condition when allocating the object. This 120 * can happen with NFS vnodes since the nfsnode isn't locked. 121 */ 122 while (vp->v_flag & VOLOCK) { 123 vp->v_flag |= VOWANT; 124 tsleep(vp, PVM, "vnpobj", 0); 125 } 126 vp->v_flag |= VOLOCK; 127 128 /* 129 * If the object is being terminated, wait for it to 130 * go away. 131 */ 132 while (((object = vp->v_object) != NULL) && 133 (object->flags & OBJ_DEAD)) { 134 tsleep(object, PVM, "vadead", 0); 135 } 136 137 if (vp->v_usecount == 0) 138 panic("vnode_pager_alloc: no vnode reference"); 139 140 if (object == NULL) { 141 /* 142 * And an object of the appropriate size 143 */ 144 object = vm_object_allocate(OBJT_VNODE, size); 145 object->flags = 0; 146 147 object->un_pager.vnp.vnp_size = (vm_ooffset_t) size * PAGE_SIZE; 148 149 object->handle = handle; 150 vp->v_object = object; 151 vp->v_usecount++; 152 } else { 153 object->ref_count++; 154 vp->v_usecount++; 155 } 156 157 vp->v_flag &= ~VOLOCK; 158 if (vp->v_flag & VOWANT) { 159 vp->v_flag &= ~VOWANT; 160 wakeup(vp); 161 } 162 return (object); 163} 164 165static void 166vnode_pager_dealloc(object) 167 vm_object_t object; 168{ 169 register struct vnode *vp = object->handle; 170 171 if (vp == NULL) 172 panic("vnode_pager_dealloc: pager already dealloced"); 173 174 if (object->paging_in_progress) { 175 int s = splbio(); 176 while (object->paging_in_progress) { 177 object->flags |= OBJ_PIPWNT; 178 tsleep(object, PVM, "vnpdea", 0); 179 } 180 splx(s); 181 } 182 183 object->handle = NULL; 184 object->type = OBJT_DEAD; 185 vp->v_object = NULL; 186 vp->v_flag &= ~(VTEXT | VOBJBUF); 187} 188 189static boolean_t 190vnode_pager_haspage(object, pindex, before, after) 191 vm_object_t object; 192 vm_pindex_t pindex; 193 int *before; 194 int *after; 195{ 196 struct vnode *vp = object->handle; 197 daddr_t bn; 198 int err; 199 daddr_t reqblock; 200 int poff; 201 int bsize; 202 int pagesperblock, blocksperpage; 203 204 if ((vp == NULL) || (vp->v_flag & VDOOMED)) 205 return FALSE; 206 207 /* 208 * If filesystem no longer mounted or offset beyond end of file we do 209 * not have the page. 210 */ 211 if ((vp->v_mount == NULL) || 212 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 213 return FALSE; 214 215 bsize = vp->v_mount->mnt_stat.f_iosize; 216 pagesperblock = bsize / PAGE_SIZE; 217 blocksperpage = 0; 218 if (pagesperblock > 0) { 219 reqblock = pindex / pagesperblock; 220 } else { 221 blocksperpage = (PAGE_SIZE / bsize); 222 reqblock = pindex * blocksperpage; 223 } 224 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 225 after, before); 226 if (err) 227 return TRUE; 228 if ( bn == -1) 229 return FALSE; 230 if (pagesperblock > 0) { 231 poff = pindex - (reqblock * pagesperblock); 232 if (before) { 233 *before *= pagesperblock; 234 *before += poff; 235 } 236 if (after) { 237 int numafter; 238 *after *= pagesperblock; 239 numafter = pagesperblock - (poff + 1); 240 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) { 241 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex))); 242 } 243 *after += numafter; 244 } 245 } else { 246 if (before) { 247 *before /= blocksperpage; 248 } 249 250 if (after) { 251 *after /= blocksperpage; 252 } 253 } 254 return TRUE; 255} 256 257/* 258 * Lets the VM system know about a change in size for a file. 259 * We adjust our own internal size and flush any cached pages in 260 * the associated object that are affected by the size change. 261 * 262 * Note: this routine may be invoked as a result of a pager put 263 * operation (possibly at object termination time), so we must be careful. 264 */ 265void 266vnode_pager_setsize(vp, nsize) 267 struct vnode *vp; 268 vm_ooffset_t nsize; 269{ 270 vm_object_t object = vp->v_object; 271 272 if (object == NULL) 273 return; 274 275 /* 276 * Hasn't changed size 277 */ 278 if (nsize == object->un_pager.vnp.vnp_size) 279 return; 280 281 /* 282 * File has shrunk. Toss any cached pages beyond the new EOF. 283 */ 284 if (nsize < object->un_pager.vnp.vnp_size) { 285 vm_ooffset_t nsizerounded; 286 nsizerounded = IDX_TO_OFF(OFF_TO_IDX(nsize + PAGE_MASK)); 287 if (nsizerounded < object->un_pager.vnp.vnp_size) { 288 vm_pindex_t st, end; 289 st = OFF_TO_IDX(nsize + PAGE_MASK); 290 end = OFF_TO_IDX(object->un_pager.vnp.vnp_size); 291 292 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size); 293 vm_object_page_remove(object, st, end, FALSE); 294 } 295 /* 296 * this gets rid of garbage at the end of a page that is now 297 * only partially backed by the vnode... 298 */ 299 if (nsize & PAGE_MASK) { 300 vm_offset_t kva; 301 vm_page_t m; 302 303 m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 304 if (m) { 305 kva = vm_pager_map_page(m); 306 bzero((caddr_t) kva + (nsize & PAGE_MASK), 307 (int) (round_page(nsize) - nsize)); 308 vm_pager_unmap_page(kva); 309 } 310 } 311 } 312 object->un_pager.vnp.vnp_size = nsize; 313 object->size = OFF_TO_IDX(nsize + PAGE_MASK); 314} 315 316void 317vnode_pager_freepage(m) 318 vm_page_t m; 319{ 320 vm_page_free(m); 321} 322 323/* 324 * calculate the linear (byte) disk address of specified virtual 325 * file address 326 */ 327static vm_offset_t 328vnode_pager_addr(vp, address, run) 329 struct vnode *vp; 330 vm_ooffset_t address; 331 int *run; 332{ 333 int rtaddress; 334 int bsize; 335 daddr_t block; 336 struct vnode *rtvp; 337 int err; 338 daddr_t vblock; 339 int voffset; 340 341 if ((int) address < 0) 342 return -1; 343 344 if (vp->v_mount == NULL) 345 return -1; 346 347 bsize = vp->v_mount->mnt_stat.f_iosize; 348 vblock = address / bsize; 349 voffset = address % bsize; 350 351 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 352 353 if (err || (block == -1)) 354 rtaddress = -1; 355 else { 356 rtaddress = block + voffset / DEV_BSIZE; 357 if( run) { 358 *run += 1; 359 *run *= bsize/PAGE_SIZE; 360 *run -= voffset/PAGE_SIZE; 361 } 362 } 363 364 return rtaddress; 365} 366 367/* 368 * interrupt routine for I/O completion 369 */ 370static void 371vnode_pager_iodone(bp) 372 struct buf *bp; 373{ 374 bp->b_flags |= B_DONE; 375 wakeup(bp); 376} 377 378/* 379 * small block file system vnode pager input 380 */ 381static int 382vnode_pager_input_smlfs(object, m) 383 vm_object_t object; 384 vm_page_t m; 385{ 386 int i; 387 int s; 388 struct vnode *dp, *vp; 389 struct buf *bp; 390 vm_offset_t kva; 391 int fileaddr; 392 vm_offset_t bsize; 393 int error = 0; 394 395 vp = object->handle; 396 if (vp->v_mount == NULL) 397 return VM_PAGER_BAD; 398 399 bsize = vp->v_mount->mnt_stat.f_iosize; 400 401 402 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 403 404 kva = vm_pager_map_page(m); 405 406 for (i = 0; i < PAGE_SIZE / bsize; i++) { 407 408 if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid)) 409 continue; 410 411 fileaddr = vnode_pager_addr(vp, 412 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0); 413 if (fileaddr != -1) { 414 bp = getpbuf(); 415 416 /* build a minimal buffer header */ 417 bp->b_flags = B_BUSY | B_READ | B_CALL; 418 bp->b_iodone = vnode_pager_iodone; 419 bp->b_proc = curproc; 420 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 421 if (bp->b_rcred != NOCRED) 422 crhold(bp->b_rcred); 423 if (bp->b_wcred != NOCRED) 424 crhold(bp->b_wcred); 425 bp->b_data = (caddr_t) kva + i * bsize; 426 bp->b_blkno = fileaddr; 427 pbgetvp(dp, bp); 428 bp->b_bcount = bsize; 429 bp->b_bufsize = bsize; 430 431 /* do the input */ 432 VOP_STRATEGY(bp); 433 434 /* we definitely need to be at splbio here */ 435 436 s = splbio(); 437 while ((bp->b_flags & B_DONE) == 0) { 438 tsleep(bp, PVM, "vnsrd", 0); 439 } 440 splx(s); 441 if ((bp->b_flags & B_ERROR) != 0) 442 error = EIO; 443 444 /* 445 * free the buffer header back to the swap buffer pool 446 */ 447 relpbuf(bp); 448 if (error) 449 break; 450 451 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 452 } else { 453 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 454 bzero((caddr_t) kva + i * bsize, bsize); 455 } 456 } 457 vm_pager_unmap_page(kva); 458 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 459 m->flags &= ~PG_ZERO; 460 if (error) { 461 return VM_PAGER_ERROR; 462 } 463 return VM_PAGER_OK; 464 465} 466 467 468/* 469 * old style vnode pager output routine 470 */ 471static int 472vnode_pager_input_old(object, m) 473 vm_object_t object; 474 vm_page_t m; 475{ 476 struct uio auio; 477 struct iovec aiov; 478 int error; 479 int size; 480 vm_offset_t kva; 481 482 error = 0; 483 484 /* 485 * Return failure if beyond current EOF 486 */ 487 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 488 return VM_PAGER_BAD; 489 } else { 490 size = PAGE_SIZE; 491 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 492 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 493 494 /* 495 * Allocate a kernel virtual address and initialize so that 496 * we can use VOP_READ/WRITE routines. 497 */ 498 kva = vm_pager_map_page(m); 499 500 aiov.iov_base = (caddr_t) kva; 501 aiov.iov_len = size; 502 auio.uio_iov = &aiov; 503 auio.uio_iovcnt = 1; 504 auio.uio_offset = IDX_TO_OFF(m->pindex); 505 auio.uio_segflg = UIO_SYSSPACE; 506 auio.uio_rw = UIO_READ; 507 auio.uio_resid = size; 508 auio.uio_procp = (struct proc *) 0; 509 510 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred); 511 if (!error) { 512 register int count = size - auio.uio_resid; 513 514 if (count == 0) 515 error = EINVAL; 516 else if (count != PAGE_SIZE) 517 bzero((caddr_t) kva + count, PAGE_SIZE - count); 518 } 519 vm_pager_unmap_page(kva); 520 } 521 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 522 m->dirty = 0; 523 m->flags &= ~PG_ZERO; 524 return error ? VM_PAGER_ERROR : VM_PAGER_OK; 525} 526 527/* 528 * generic vnode pager input routine 529 */ 530 531static int 532vnode_pager_getpages(object, m, count, reqpage) 533 vm_object_t object; 534 vm_page_t *m; 535 int count; 536 int reqpage; 537{ 538 int rtval; 539 struct vnode *vp; 540 541 vp = object->handle; 542 rtval = VOP_GETPAGES(vp, m, count*PAGE_SIZE, reqpage, 0); 543 if (rtval == EOPNOTSUPP) 544 return vnode_pager_leaf_getpages(object, m, count, reqpage); 545 else 546 return rtval; 547} 548 549static int 550vnode_pager_leaf_getpages(object, m, count, reqpage) 551 vm_object_t object; 552 vm_page_t *m; 553 int count; 554 int reqpage; 555{ 556 vm_offset_t kva; 557 off_t foff; 558 int i, size, bsize, first, firstaddr; 559 struct vnode *dp, *vp; 560 int runpg; 561 int runend; 562 struct buf *bp; 563 int s; 564 int error = 0; 565 566 vp = object->handle; 567 if (vp->v_mount == NULL) 568 return VM_PAGER_BAD; 569 570 bsize = vp->v_mount->mnt_stat.f_iosize; 571 572 /* get the UNDERLYING device for the file with VOP_BMAP() */ 573 574 /* 575 * originally, we did not check for an error return value -- assuming 576 * an fs always has a bmap entry point -- that assumption is wrong!!! 577 */ 578 foff = IDX_TO_OFF(m[reqpage]->pindex); 579 580 /* 581 * if we can't bmap, use old VOP code 582 */ 583 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 584 for (i = 0; i < count; i++) { 585 if (i != reqpage) { 586 vnode_pager_freepage(m[i]); 587 } 588 } 589 cnt.v_vnodein++; 590 cnt.v_vnodepgsin++; 591 return vnode_pager_input_old(object, m[reqpage]); 592 593 /* 594 * if the blocksize is smaller than a page size, then use 595 * special small filesystem code. NFS sometimes has a small 596 * blocksize, but it can handle large reads itself. 597 */ 598 } else if ((PAGE_SIZE / bsize) > 1 && 599 (vp->v_mount->mnt_stat.f_type != MOUNT_NFS)) { 600 601 for (i = 0; i < count; i++) { 602 if (i != reqpage) { 603 vnode_pager_freepage(m[i]); 604 } 605 } 606 cnt.v_vnodein++; 607 cnt.v_vnodepgsin++; 608 return vnode_pager_input_smlfs(object, m[reqpage]); 609 } 610 /* 611 * if ANY DEV_BSIZE blocks are valid on a large filesystem block 612 * then, the entire page is valid -- 613 * XXX no it isn't 614 */ 615 616 if (m[reqpage]->valid != VM_PAGE_BITS_ALL) 617 m[reqpage]->valid = 0; 618 619 if (m[reqpage]->valid) { 620 m[reqpage]->valid = VM_PAGE_BITS_ALL; 621 for (i = 0; i < count; i++) { 622 if (i != reqpage) 623 vnode_pager_freepage(m[i]); 624 } 625 return VM_PAGER_OK; 626 } 627 628 /* 629 * here on direct device I/O 630 */ 631 632 firstaddr = -1; 633 /* 634 * calculate the run that includes the required page 635 */ 636 for(first = 0, i = 0; i < count; i = runend) { 637 firstaddr = vnode_pager_addr(vp, 638 IDX_TO_OFF(m[i]->pindex), &runpg); 639 if (firstaddr == -1) { 640 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 641 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: %ld, vnp_size: %d", 642 firstaddr, foff, object->un_pager.vnp.vnp_size); 643 } 644 vnode_pager_freepage(m[i]); 645 runend = i + 1; 646 first = runend; 647 continue; 648 } 649 runend = i + runpg; 650 if (runend <= reqpage) { 651 int j; 652 for (j = i; j < runend; j++) { 653 vnode_pager_freepage(m[j]); 654 } 655 } else { 656 if (runpg < (count - first)) { 657 for (i = first + runpg; i < count; i++) 658 vnode_pager_freepage(m[i]); 659 count = first + runpg; 660 } 661 break; 662 } 663 first = runend; 664 } 665 666 /* 667 * the first and last page have been calculated now, move input pages 668 * to be zero based... 669 */ 670 if (first != 0) { 671 for (i = first; i < count; i++) { 672 m[i - first] = m[i]; 673 } 674 count -= first; 675 reqpage -= first; 676 } 677 678 /* 679 * calculate the file virtual address for the transfer 680 */ 681 foff = IDX_TO_OFF(m[0]->pindex); 682 683 /* 684 * calculate the size of the transfer 685 */ 686 size = count * PAGE_SIZE; 687 if ((foff + size) > object->un_pager.vnp.vnp_size) 688 size = object->un_pager.vnp.vnp_size - foff; 689 690 /* 691 * round up physical size for real devices 692 */ 693 if (dp->v_type == VBLK || dp->v_type == VCHR) 694 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 695 696 bp = getpbuf(); 697 kva = (vm_offset_t) bp->b_data; 698 699 /* 700 * and map the pages to be read into the kva 701 */ 702 pmap_qenter(kva, m, count); 703 704 /* build a minimal buffer header */ 705 bp->b_flags = B_BUSY | B_READ | B_CALL; 706 bp->b_iodone = vnode_pager_iodone; 707 /* B_PHYS is not set, but it is nice to fill this in */ 708 bp->b_proc = curproc; 709 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 710 if (bp->b_rcred != NOCRED) 711 crhold(bp->b_rcred); 712 if (bp->b_wcred != NOCRED) 713 crhold(bp->b_wcred); 714 bp->b_blkno = firstaddr; 715 pbgetvp(dp, bp); 716 bp->b_bcount = size; 717 bp->b_bufsize = size; 718 719 cnt.v_vnodein++; 720 cnt.v_vnodepgsin += count; 721 722 /* do the input */ 723 VOP_STRATEGY(bp); 724 725 s = splbio(); 726 /* we definitely need to be at splbio here */ 727 728 while ((bp->b_flags & B_DONE) == 0) { 729 tsleep(bp, PVM, "vnread", 0); 730 } 731 splx(s); 732 if ((bp->b_flags & B_ERROR) != 0) 733 error = EIO; 734 735 if (!error) { 736 if (size != count * PAGE_SIZE) 737 bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 738 } 739 pmap_qremove(kva, count); 740 741 /* 742 * free the buffer header back to the swap buffer pool 743 */ 744 relpbuf(bp); 745 746 for (i = 0; i < count; i++) { 747 pmap_clear_modify(VM_PAGE_TO_PHYS(m[i])); 748 m[i]->dirty = 0; 749 m[i]->valid = VM_PAGE_BITS_ALL; 750 m[i]->flags &= ~PG_ZERO; 751 if (i != reqpage) { 752 753 /* 754 * whether or not to leave the page activated is up in 755 * the air, but we should put the page on a page queue 756 * somewhere. (it already is in the object). Result: 757 * It appears that emperical results show that 758 * deactivating pages is best. 759 */ 760 761 /* 762 * just in case someone was asking for this page we 763 * now tell them that it is ok to use 764 */ 765 if (!error) { 766 if (m[i]->flags & PG_WANTED) 767 vm_page_activate(m[i]); 768 else 769 vm_page_deactivate(m[i]); 770 PAGE_WAKEUP(m[i]); 771 } else { 772 vnode_pager_freepage(m[i]); 773 } 774 } 775 } 776 if (error) { 777 printf("vnode_pager_getpages: I/O read error\n"); 778 } 779 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 780} 781 782static int 783vnode_pager_putpages(object, m, count, sync, rtvals) 784 vm_object_t object; 785 vm_page_t *m; 786 int count; 787 boolean_t sync; 788 int *rtvals; 789{ 790 int rtval; 791 struct vnode *vp; 792 793 vp = object->handle; 794 rtval = VOP_PUTPAGES(vp, m, count*PAGE_SIZE, sync, rtvals, 0); 795 if (rtval == EOPNOTSUPP) 796 return vnode_pager_leaf_putpages(object, m, count, sync, rtvals); 797 else 798 return rtval; 799} 800 801/* 802 * generic vnode pager output routine 803 */ 804static int 805vnode_pager_leaf_putpages(object, m, count, sync, rtvals) 806 vm_object_t object; 807 vm_page_t *m; 808 int count; 809 boolean_t sync; 810 int *rtvals; 811{ 812 int i; 813 814 struct vnode *vp; 815 int maxsize, ncount; 816 vm_ooffset_t poffset; 817 struct uio auio; 818 struct iovec aiov; 819 int error; 820 821 vp = object->handle;; 822 for (i = 0; i < count; i++) 823 rtvals[i] = VM_PAGER_AGAIN; 824 825 if ((int) m[0]->pindex < 0) { 826 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", m[0]->pindex, m[0]->dirty); 827 rtvals[0] = VM_PAGER_BAD; 828 return VM_PAGER_BAD; 829 } 830 831 maxsize = count * PAGE_SIZE; 832 ncount = count; 833 834 poffset = IDX_TO_OFF(m[0]->pindex); 835 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 836 if (object->un_pager.vnp.vnp_size > poffset) 837 maxsize = object->un_pager.vnp.vnp_size - poffset; 838 else 839 maxsize = 0; 840 ncount = btoc(maxsize); 841 if (ncount < count) { 842 for (i = ncount; i < count; i++) { 843 rtvals[i] = VM_PAGER_BAD; 844 } 845#ifdef BOGUS 846 if (ncount == 0) { 847 printf("vnode_pager_putpages: write past end of file: %d, %lu\n", 848 poffset, 849 (unsigned long) object->un_pager.vnp.vnp_size); 850 return rtvals[0]; 851 } 852#endif 853 } 854 } 855 856 for (i = 0; i < count; i++) { 857 m[i]->busy++; 858 m[i]->flags &= ~PG_BUSY; 859 } 860 861 aiov.iov_base = (caddr_t) 0; 862 aiov.iov_len = maxsize; 863 auio.uio_iov = &aiov; 864 auio.uio_iovcnt = 1; 865 auio.uio_offset = poffset; 866 auio.uio_segflg = UIO_NOCOPY; 867 auio.uio_rw = UIO_WRITE; 868 auio.uio_resid = maxsize; 869 auio.uio_procp = (struct proc *) 0; 870 error = VOP_WRITE(vp, &auio, IO_VMIO|(sync?IO_SYNC:0), curproc->p_ucred); 871 cnt.v_vnodeout++; 872 cnt.v_vnodepgsout += ncount; 873 874 if (error) { 875 printf("vnode_pager_putpages: I/O error %d\n", error); 876 } 877 if (auio.uio_resid) { 878 printf("vnode_pager_putpages: residual I/O %d at %ld\n", 879 auio.uio_resid, m[0]->pindex); 880 } 881 for (i = 0; i < count; i++) { 882 m[i]->busy--; 883 if (i < ncount) { 884 rtvals[i] = VM_PAGER_OK; 885 } 886 if ((m[i]->busy == 0) && (m[i]->flags & PG_WANTED)) { 887 vm_page_activate(m[i]); 888 wakeup(m[i]); 889 } 890 } 891 return rtvals[0]; 892} 893 894struct vnode * 895vnode_pager_lock(object) 896 vm_object_t object; 897{ 898 struct proc *p = curproc; /* XXX */ 899 900 for (; object != NULL; object = object->backing_object) { 901 if (object->type != OBJT_VNODE) 902 continue; 903 if (object->flags & OBJ_DEAD) 904 return NULL; 905 906 while (vget(object->handle, 907 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) { 908 printf("vnode_pager_lock: retrying\n"); 909 } 910 return object->handle; 911 } 912 return NULL; 913} 914