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 1.80 1998/01/17 09:17:02 dyson 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 <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/proc.h> 57 #include <sys/vnode.h> 58 #include <sys/mount.h> 59 #include <sys/buf.h> 60 #include <sys/vmmeter.h> 61 62 #include <vm/vm.h> 63 #include <vm/vm_prot.h> 64 #include <vm/vm_object.h> 65 #include <vm/vm_page.h> 66 #include <vm/vm_pager.h> 67 #include <vm/vm_map.h> 68 #include <vm/vnode_pager.h> 69 #include <vm/vm_extern.h> 70 71 static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address, 72 int *run)); 73 static void vnode_pager_iodone __P((struct buf *bp)); 74 static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m)); 75 static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m)); 76 static void vnode_pager_dealloc __P((vm_object_t)); 77 static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 78 static int vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *)); 79 static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 80 81 struct pagerops vnodepagerops = { 82 NULL, 83 vnode_pager_alloc, 84 vnode_pager_dealloc, 85 vnode_pager_getpages, 86 vnode_pager_putpages, 87 vnode_pager_haspage, 88 NULL 89 }; 90 91 static int vnode_pager_leaf_getpages __P((vm_object_t object, vm_page_t *m, 92 int count, int reqpage)); 93 static int vnode_pager_leaf_putpages __P((vm_object_t object, vm_page_t *m, 94 int count, boolean_t sync, 95 int *rtvals)); 96 97 /* 98 * Allocate (or lookup) pager for a vnode. 99 * Handle is a vnode pointer. 100 */ 101 vm_object_t 102 vnode_pager_alloc(void *handle, vm_size_t size, vm_prot_t prot, 103 vm_ooffset_t offset) 104 { 105 vm_object_t object; 106 struct vnode *vp; 107 108 /* 109 * Pageout to vnode, no can do yet. 110 */ 111 if (handle == NULL) 112 return (NULL); 113 114 vp = (struct vnode *) handle; 115 116 /* 117 * Prevent race condition when allocating the object. This 118 * can happen with NFS vnodes since the nfsnode isn't locked. 119 */ 120 while (vp->v_flag & VOLOCK) { 121 vp->v_flag |= VOWANT; 122 tsleep(vp, PVM, "vnpobj", 0); 123 } 124 vp->v_flag |= VOLOCK; 125 126 /* 127 * If the object is being terminated, wait for it to 128 * go away. 129 */ 130 while (((object = vp->v_object) != NULL) && 131 (object->flags & OBJ_DEAD)) { 132 tsleep(object, PVM, "vadead", 0); 133 } 134 135 if (vp->v_usecount == 0) 136 panic("vnode_pager_alloc: no vnode reference"); 137 138 if (object == NULL) { 139 /* 140 * And an object of the appropriate size 141 */ 142 object = vm_object_allocate(OBJT_VNODE, size); 143 object->flags = 0; 144 145 object->un_pager.vnp.vnp_size = (vm_ooffset_t) size * PAGE_SIZE; 146 147 object->handle = handle; 148 vp->v_object = object; 149 vp->v_usecount++; 150 } else { 151 object->ref_count++; 152 vp->v_usecount++; 153 } 154 155 vp->v_flag &= ~VOLOCK; 156 if (vp->v_flag & VOWANT) { 157 vp->v_flag &= ~VOWANT; 158 wakeup(vp); 159 } 160 return (object); 161 } 162 163 static void 164 vnode_pager_dealloc(object) 165 vm_object_t object; 166 { 167 register struct vnode *vp = object->handle; 168 169 if (vp == NULL) 170 panic("vnode_pager_dealloc: pager already dealloced"); 171 172 if (object->paging_in_progress) { 173 int s = splbio(); 174 while (object->paging_in_progress) { 175 object->flags |= OBJ_PIPWNT; 176 tsleep(object, PVM, "vnpdea", 0); 177 } 178 splx(s); 179 } 180 181 object->handle = NULL; 182 object->type = OBJT_DEFAULT; 183 vp->v_object = NULL; 184 vp->v_flag &= ~(VTEXT|VOBJBUF); 185 } 186 187 static boolean_t 188 vnode_pager_haspage(object, pindex, before, after) 189 vm_object_t object; 190 vm_pindex_t pindex; 191 int *before; 192 int *after; 193 { 194 struct vnode *vp = object->handle; 195 daddr_t bn; 196 int err; 197 daddr_t reqblock; 198 int poff; 199 int bsize; 200 int pagesperblock, blocksperpage; 201 202 if ((vp == NULL) || (vp->v_flag & VDOOMED)) 203 return FALSE; 204 205 /* 206 * If filesystem no longer mounted or offset beyond end of file we do 207 * not have the page. 208 */ 209 if ((vp->v_mount == NULL) || 210 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 211 return FALSE; 212 213 bsize = vp->v_mount->mnt_stat.f_iosize; 214 pagesperblock = bsize / PAGE_SIZE; 215 blocksperpage = 0; 216 if (pagesperblock > 0) { 217 reqblock = pindex / pagesperblock; 218 } else { 219 blocksperpage = (PAGE_SIZE / bsize); 220 reqblock = pindex * blocksperpage; 221 } 222 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 223 after, before); 224 if (err) 225 return TRUE; 226 if ( bn == -1) 227 return FALSE; 228 if (pagesperblock > 0) { 229 poff = pindex - (reqblock * pagesperblock); 230 if (before) { 231 *before *= pagesperblock; 232 *before += poff; 233 } 234 if (after) { 235 int numafter; 236 *after *= pagesperblock; 237 numafter = pagesperblock - (poff + 1); 238 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) { 239 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex))); 240 } 241 *after += numafter; 242 } 243 } else { 244 if (before) { 245 *before /= blocksperpage; 246 } 247 248 if (after) { 249 *after /= blocksperpage; 250 } 251 } 252 return TRUE; 253 } 254 255 /* 256 * Lets the VM system know about a change in size for a file. 257 * We adjust our own internal size and flush any cached pages in 258 * the associated object that are affected by the size change. 259 * 260 * Note: this routine may be invoked as a result of a pager put 261 * operation (possibly at object termination time), so we must be careful. 262 */ 263 void 264 vnode_pager_setsize(vp, nsize) 265 struct vnode *vp; 266 vm_ooffset_t nsize; 267 { 268 vm_object_t object = vp->v_object; 269 270 if (object == NULL) 271 return; 272 273 /* 274 * Hasn't changed size 275 */ 276 if (nsize == object->un_pager.vnp.vnp_size) 277 return; 278 279 /* 280 * File has shrunk. Toss any cached pages beyond the new EOF. 281 */ 282 if (nsize < object->un_pager.vnp.vnp_size) { 283 vm_ooffset_t nsizerounded; 284 nsizerounded = IDX_TO_OFF(OFF_TO_IDX(nsize + PAGE_MASK)); 285 if (nsizerounded < object->un_pager.vnp.vnp_size) { 286 vm_pindex_t st, end; 287 st = OFF_TO_IDX(nsize + PAGE_MASK); 288 end = OFF_TO_IDX(object->un_pager.vnp.vnp_size); 289 290 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size); 291 vm_object_page_remove(object, st, end, FALSE); 292 } 293 /* 294 * this gets rid of garbage at the end of a page that is now 295 * only partially backed by the vnode... 296 */ 297 if (nsize & PAGE_MASK) { 298 vm_offset_t kva; 299 vm_page_t m; 300 301 m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 302 if (m) { 303 kva = vm_pager_map_page(m); 304 bzero((caddr_t) kva + (nsize & PAGE_MASK), 305 (int) (round_page(nsize) - nsize)); 306 vm_pager_unmap_page(kva); 307 } 308 } 309 } 310 object->un_pager.vnp.vnp_size = nsize; 311 object->size = OFF_TO_IDX(nsize + PAGE_MASK); 312 } 313 314 void 315 vnode_pager_freepage(m) 316 vm_page_t m; 317 { 318 vm_page_free(m); 319 } 320 321 /* 322 * calculate the linear (byte) disk address of specified virtual 323 * file address 324 */ 325 static vm_offset_t 326 vnode_pager_addr(vp, address, run) 327 struct vnode *vp; 328 vm_ooffset_t address; 329 int *run; 330 { 331 int rtaddress; 332 int bsize; 333 daddr_t block; 334 struct vnode *rtvp; 335 int err; 336 daddr_t vblock; 337 int voffset; 338 339 if ((int) address < 0) 340 return -1; 341 342 if (vp->v_mount == NULL) 343 return -1; 344 345 bsize = vp->v_mount->mnt_stat.f_iosize; 346 vblock = address / bsize; 347 voffset = address % bsize; 348 349 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 350 351 if (err || (block == -1)) 352 rtaddress = -1; 353 else { 354 rtaddress = block + voffset / DEV_BSIZE; 355 if( run) { 356 *run += 1; 357 *run *= bsize/PAGE_SIZE; 358 *run -= voffset/PAGE_SIZE; 359 } 360 } 361 362 return rtaddress; 363 } 364 365 /* 366 * interrupt routine for I/O completion 367 */ 368 static void 369 vnode_pager_iodone(bp) 370 struct buf *bp; 371 { 372 bp->b_flags |= B_DONE; 373 wakeup(bp); 374 } 375 376 /* 377 * small block file system vnode pager input 378 */ 379 static int 380 vnode_pager_input_smlfs(object, m) 381 vm_object_t object; 382 vm_page_t m; 383 { 384 int i; 385 int s; 386 struct vnode *dp, *vp; 387 struct buf *bp; 388 vm_offset_t kva; 389 int fileaddr; 390 vm_offset_t bsize; 391 int error = 0; 392 393 vp = object->handle; 394 if (vp->v_mount == NULL) 395 return VM_PAGER_BAD; 396 397 bsize = vp->v_mount->mnt_stat.f_iosize; 398 399 400 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 401 402 kva = vm_pager_map_page(m); 403 404 for (i = 0; i < PAGE_SIZE / bsize; i++) { 405 406 if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid)) 407 continue; 408 409 fileaddr = vnode_pager_addr(vp, 410 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0); 411 if (fileaddr != -1) { 412 bp = getpbuf(); 413 414 /* build a minimal buffer header */ 415 bp->b_flags = B_BUSY | B_READ | B_CALL; 416 bp->b_iodone = vnode_pager_iodone; 417 bp->b_proc = curproc; 418 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 419 if (bp->b_rcred != NOCRED) 420 crhold(bp->b_rcred); 421 if (bp->b_wcred != NOCRED) 422 crhold(bp->b_wcred); 423 bp->b_data = (caddr_t) kva + i * bsize; 424 bp->b_blkno = fileaddr; 425 pbgetvp(dp, bp); 426 bp->b_bcount = bsize; 427 bp->b_bufsize = bsize; 428 429 /* do the input */ 430 VOP_STRATEGY(bp); 431 432 /* we definitely need to be at splbio here */ 433 434 s = splbio(); 435 while ((bp->b_flags & B_DONE) == 0) { 436 tsleep(bp, PVM, "vnsrd", 0); 437 } 438 splx(s); 439 if ((bp->b_flags & B_ERROR) != 0) 440 error = EIO; 441 442 /* 443 * free the buffer header back to the swap buffer pool 444 */ 445 relpbuf(bp); 446 if (error) 447 break; 448 449 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 450 } else { 451 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 452 bzero((caddr_t) kva + i * bsize, bsize); 453 } 454 } 455 vm_pager_unmap_page(kva); 456 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 457 m->flags &= ~PG_ZERO; 458 if (error) { 459 return VM_PAGER_ERROR; 460 } 461 return VM_PAGER_OK; 462 463 } 464 465 466 /* 467 * old style vnode pager output routine 468 */ 469 static int 470 vnode_pager_input_old(object, m) 471 vm_object_t object; 472 vm_page_t m; 473 { 474 struct uio auio; 475 struct iovec aiov; 476 int error; 477 int size; 478 vm_offset_t kva; 479 480 error = 0; 481 482 /* 483 * Return failure if beyond current EOF 484 */ 485 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 486 return VM_PAGER_BAD; 487 } else { 488 size = PAGE_SIZE; 489 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 490 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 491 492 /* 493 * Allocate a kernel virtual address and initialize so that 494 * we can use VOP_READ/WRITE routines. 495 */ 496 kva = vm_pager_map_page(m); 497 498 aiov.iov_base = (caddr_t) kva; 499 aiov.iov_len = size; 500 auio.uio_iov = &aiov; 501 auio.uio_iovcnt = 1; 502 auio.uio_offset = IDX_TO_OFF(m->pindex); 503 auio.uio_segflg = UIO_SYSSPACE; 504 auio.uio_rw = UIO_READ; 505 auio.uio_resid = size; 506 auio.uio_procp = (struct proc *) 0; 507 508 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred); 509 if (!error) { 510 register int count = size - auio.uio_resid; 511 512 if (count == 0) 513 error = EINVAL; 514 else if (count != PAGE_SIZE) 515 bzero((caddr_t) kva + count, PAGE_SIZE - count); 516 } 517 vm_pager_unmap_page(kva); 518 } 519 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 520 m->dirty = 0; 521 m->flags &= ~PG_ZERO; 522 return error ? VM_PAGER_ERROR : VM_PAGER_OK; 523 } 524 525 /* 526 * generic vnode pager input routine 527 */ 528 529 static int 530 vnode_pager_getpages(object, m, count, reqpage) 531 vm_object_t object; 532 vm_page_t *m; 533 int count; 534 int reqpage; 535 { 536 int rtval; 537 struct vnode *vp; 538 539 vp = object->handle; 540 rtval = VOP_GETPAGES(vp, m, count*PAGE_SIZE, reqpage, 0); 541 if (rtval == EOPNOTSUPP) 542 return vnode_pager_leaf_getpages(object, m, count, reqpage); 543 else 544 return rtval; 545 } 546 547 static int 548 vnode_pager_leaf_getpages(object, m, count, reqpage) 549 vm_object_t object; 550 vm_page_t *m; 551 int count; 552 int reqpage; 553 { 554 vm_offset_t kva; 555 off_t foff; 556 int i, size, bsize, first, firstaddr; 557 struct vnode *dp, *vp; 558 int runpg; 559 int runend; 560 struct buf *bp; 561 int s; 562 int error = 0; 563 564 vp = object->handle; 565 if (vp->v_mount == NULL) 566 return VM_PAGER_BAD; 567 568 bsize = vp->v_mount->mnt_stat.f_iosize; 569 570 /* get the UNDERLYING device for the file with VOP_BMAP() */ 571 572 /* 573 * originally, we did not check for an error return value -- assuming 574 * an fs always has a bmap entry point -- that assumption is wrong!!! 575 */ 576 foff = IDX_TO_OFF(m[reqpage]->pindex); 577 578 /* 579 * if we can't bmap, use old VOP code 580 */ 581 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 582 for (i = 0; i < count; i++) { 583 if (i != reqpage) { 584 vnode_pager_freepage(m[i]); 585 } 586 } 587 cnt.v_vnodein++; 588 cnt.v_vnodepgsin++; 589 return vnode_pager_input_old(object, m[reqpage]); 590 591 /* 592 * if the blocksize is smaller than a page size, then use 593 * special small filesystem code. NFS sometimes has a small 594 * blocksize, but it can handle large reads itself. 595 */ 596 } else if ((PAGE_SIZE / bsize) > 1 && 597 (vp->v_mount->mnt_stat.f_type != MOUNT_NFS)) { 598 599 for (i = 0; i < count; i++) { 600 if (i != reqpage) { 601 vnode_pager_freepage(m[i]); 602 } 603 } 604 cnt.v_vnodein++; 605 cnt.v_vnodepgsin++; 606 return vnode_pager_input_smlfs(object, m[reqpage]); 607 } 608 /* 609 * if ANY DEV_BSIZE blocks are valid on a large filesystem block 610 * then, the entire page is valid -- 611 * XXX no it isn't 612 */ 613 614 if (m[reqpage]->valid != VM_PAGE_BITS_ALL) 615 m[reqpage]->valid = 0; 616 617 if (m[reqpage]->valid) { 618 m[reqpage]->valid = VM_PAGE_BITS_ALL; 619 for (i = 0; i < count; i++) { 620 if (i != reqpage) 621 vnode_pager_freepage(m[i]); 622 } 623 return VM_PAGER_OK; 624 } 625 626 /* 627 * here on direct device I/O 628 */ 629 630 firstaddr = -1; 631 /* 632 * calculate the run that includes the required page 633 */ 634 for(first = 0, i = 0; i < count; i = runend) { 635 firstaddr = vnode_pager_addr(vp, 636 IDX_TO_OFF(m[i]->pindex), &runpg); 637 if (firstaddr == -1) { 638 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 639 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: %ld, vnp_size: %d", 640 firstaddr, foff, object->un_pager.vnp.vnp_size); 641 } 642 vnode_pager_freepage(m[i]); 643 runend = i + 1; 644 first = runend; 645 continue; 646 } 647 runend = i + runpg; 648 if (runend <= reqpage) { 649 int j; 650 for (j = i; j < runend; j++) { 651 vnode_pager_freepage(m[j]); 652 } 653 } else { 654 if (runpg < (count - first)) { 655 for (i = first + runpg; i < count; i++) 656 vnode_pager_freepage(m[i]); 657 count = first + runpg; 658 } 659 break; 660 } 661 first = runend; 662 } 663 664 /* 665 * the first and last page have been calculated now, move input pages 666 * to be zero based... 667 */ 668 if (first != 0) { 669 for (i = first; i < count; i++) { 670 m[i - first] = m[i]; 671 } 672 count -= first; 673 reqpage -= first; 674 } 675 676 /* 677 * calculate the file virtual address for the transfer 678 */ 679 foff = IDX_TO_OFF(m[0]->pindex); 680 681 /* 682 * calculate the size of the transfer 683 */ 684 size = count * PAGE_SIZE; 685 if ((foff + size) > object->un_pager.vnp.vnp_size) 686 size = object->un_pager.vnp.vnp_size - foff; 687 688 /* 689 * round up physical size for real devices 690 */ 691 if (dp->v_type == VBLK || dp->v_type == VCHR) 692 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 693 694 bp = getpbuf(); 695 kva = (vm_offset_t) bp->b_data; 696 697 /* 698 * and map the pages to be read into the kva 699 */ 700 pmap_qenter(kva, m, count); 701 702 /* build a minimal buffer header */ 703 bp->b_flags = B_BUSY | B_READ | B_CALL; 704 bp->b_iodone = vnode_pager_iodone; 705 /* B_PHYS is not set, but it is nice to fill this in */ 706 bp->b_proc = curproc; 707 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 708 if (bp->b_rcred != NOCRED) 709 crhold(bp->b_rcred); 710 if (bp->b_wcred != NOCRED) 711 crhold(bp->b_wcred); 712 bp->b_blkno = firstaddr; 713 pbgetvp(dp, bp); 714 bp->b_bcount = size; 715 bp->b_bufsize = size; 716 717 cnt.v_vnodein++; 718 cnt.v_vnodepgsin += count; 719 720 /* do the input */ 721 VOP_STRATEGY(bp); 722 723 s = splbio(); 724 /* we definitely need to be at splbio here */ 725 726 while ((bp->b_flags & B_DONE) == 0) { 727 tsleep(bp, PVM, "vnread", 0); 728 } 729 splx(s); 730 if ((bp->b_flags & B_ERROR) != 0) 731 error = EIO; 732 733 if (!error) { 734 if (size != count * PAGE_SIZE) 735 bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 736 } 737 pmap_qremove(kva, count); 738 739 /* 740 * free the buffer header back to the swap buffer pool 741 */ 742 relpbuf(bp); 743 744 for (i = 0; i < count; i++) { 745 pmap_clear_modify(VM_PAGE_TO_PHYS(m[i])); 746 m[i]->dirty = 0; 747 m[i]->valid = VM_PAGE_BITS_ALL; 748 m[i]->flags &= ~PG_ZERO; 749 if (i != reqpage) { 750 751 /* 752 * whether or not to leave the page activated is up in 753 * the air, but we should put the page on a page queue 754 * somewhere. (it already is in the object). Result: 755 * It appears that emperical results show that 756 * deactivating pages is best. 757 */ 758 759 /* 760 * just in case someone was asking for this page we 761 * now tell them that it is ok to use 762 */ 763 if (!error) { 764 vm_page_deactivate(m[i]); 765 PAGE_WAKEUP(m[i]); 766 } else { 767 vnode_pager_freepage(m[i]); 768 } 769 } 770 } 771 if (error) { 772 printf("vnode_pager_getpages: I/O read error\n"); 773 } 774 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 775 } 776 777 static int 778 vnode_pager_putpages(object, m, count, sync, rtvals) 779 vm_object_t object; 780 vm_page_t *m; 781 int count; 782 boolean_t sync; 783 int *rtvals; 784 { 785 int rtval; 786 struct vnode *vp; 787 788 vp = object->handle; 789 rtval = VOP_PUTPAGES(vp, m, count*PAGE_SIZE, sync, rtvals, 0); 790 if (rtval == EOPNOTSUPP) 791 return vnode_pager_leaf_putpages(object, m, count, sync, rtvals); 792 else 793 return rtval; 794 } 795 796 /* 797 * generic vnode pager output routine 798 */ 799 static int 800 vnode_pager_leaf_putpages(object, m, count, sync, rtvals) 801 vm_object_t object; 802 vm_page_t *m; 803 int count; 804 boolean_t sync; 805 int *rtvals; 806 { 807 int i; 808 809 struct vnode *vp; 810 int maxsize, ncount; 811 vm_ooffset_t poffset; 812 struct uio auio; 813 struct iovec aiov; 814 int error; 815 816 vp = object->handle;; 817 for (i = 0; i < count; i++) 818 rtvals[i] = VM_PAGER_AGAIN; 819 820 if ((int) m[0]->pindex < 0) { 821 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", m[0]->pindex, m[0]->dirty); 822 rtvals[0] = VM_PAGER_BAD; 823 return VM_PAGER_BAD; 824 } 825 826 maxsize = count * PAGE_SIZE; 827 ncount = count; 828 829 poffset = IDX_TO_OFF(m[0]->pindex); 830 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 831 if (object->un_pager.vnp.vnp_size > poffset) 832 maxsize = object->un_pager.vnp.vnp_size - poffset; 833 else 834 maxsize = 0; 835 ncount = btoc(maxsize); 836 if (ncount < count) { 837 for (i = ncount; i < count; i++) { 838 rtvals[i] = VM_PAGER_BAD; 839 } 840 #ifdef BOGUS 841 if (ncount == 0) { 842 printf("vnode_pager_putpages: write past end of file: %d, %lu\n", 843 poffset, 844 (unsigned long) object->un_pager.vnp.vnp_size); 845 return rtvals[0]; 846 } 847 #endif 848 } 849 } 850 851 for (i = 0; i < count; i++) { 852 m[i]->busy++; 853 m[i]->flags &= ~PG_BUSY; 854 } 855 856 aiov.iov_base = (caddr_t) 0; 857 aiov.iov_len = maxsize; 858 auio.uio_iov = &aiov; 859 auio.uio_iovcnt = 1; 860 auio.uio_offset = poffset; 861 auio.uio_segflg = UIO_NOCOPY; 862 auio.uio_rw = UIO_WRITE; 863 auio.uio_resid = maxsize; 864 auio.uio_procp = (struct proc *) 0; 865 error = VOP_WRITE(vp, &auio, IO_VMIO|(sync?IO_SYNC:0), curproc->p_ucred); 866 cnt.v_vnodeout++; 867 cnt.v_vnodepgsout += ncount; 868 869 if (error) { 870 printf("vnode_pager_putpages: I/O error %d\n", error); 871 } 872 if (auio.uio_resid) { 873 printf("vnode_pager_putpages: residual I/O %d at %ld\n", 874 auio.uio_resid, m[0]->pindex); 875 } 876 for (i = 0; i < count; i++) { 877 m[i]->busy--; 878 if (i < ncount) { 879 rtvals[i] = VM_PAGER_OK; 880 } 881 if ((m[i]->busy == 0) && (m[i]->flags & PG_WANTED)) 882 wakeup(m[i]); 883 } 884 return rtvals[0]; 885 } 886 887 struct vnode * 888 vnode_pager_lock(object) 889 vm_object_t object; 890 { 891 struct proc *p = curproc; /* XXX */ 892 893 for (; object != NULL; object = object->backing_object) { 894 if (object->type != OBJT_VNODE) 895 continue; 896 if (object->flags & OBJ_DEAD) 897 return NULL; 898 899 while (vget(object->handle, 900 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) { 901 printf("vnode_pager_lock: retrying\n"); 902 } 903 return object->handle; 904 } 905 return NULL; 906 } 907