vnode_pager.c revision 12914
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 1.56 1995/12/14 09:55:14 phk 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/kernel.h> 57#include <sys/proc.h> 58#include <sys/malloc.h> 59#include <sys/vnode.h> 60#include <sys/uio.h> 61#include <sys/mount.h> 62#include <sys/buf.h> 63#include <sys/vmmeter.h> 64 65#include <vm/vm.h> 66#include <vm/vm_param.h> 67#include <vm/vm_prot.h> 68#include <vm/vm_object.h> 69#include <vm/vm_page.h> 70#include <vm/vm_pager.h> 71#include <vm/vnode_pager.h> 72#include <vm/vm_extern.h> 73 74static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address, 75 int *run)); 76static void vnode_pager_iodone __P((struct buf *bp)); 77static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m)); 78static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m)); 79static void vnode_pager_dealloc __P((vm_object_t)); 80static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 81static int vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *)); 82static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 83 84struct pagerops vnodepagerops = { 85 NULL, 86 vnode_pager_alloc, 87 vnode_pager_dealloc, 88 vnode_pager_getpages, 89 vnode_pager_putpages, 90 vnode_pager_haspage, 91 NULL 92}; 93 94static int vnode_pager_leaf_getpages __P((vm_object_t object, vm_page_t *m, 95 int count, int reqpage)); 96static int vnode_pager_leaf_putpages __P((vm_object_t object, vm_page_t *m, 97 int count, boolean_t sync, 98 int *rtvals)); 99 100/* 101 * Allocate (or lookup) pager for a vnode. 102 * Handle is a vnode pointer. 103 */ 104vm_object_t 105vnode_pager_alloc(handle, size, prot, offset) 106 void *handle; 107 vm_size_t size; 108 vm_prot_t prot; 109 vm_ooffset_t offset; 110{ 111 vm_object_t object; 112 struct vnode *vp; 113 114 /* 115 * Pageout to vnode, no can do yet. 116 */ 117 if (handle == NULL) 118 return (NULL); 119 120 vp = (struct vnode *) handle; 121 122 /* 123 * Prevent race condition when allocating the object. This 124 * can happen with NFS vnodes since the nfsnode isn't locked. 125 */ 126 while (vp->v_flag & VOLOCK) { 127 vp->v_flag |= VOWANT; 128 tsleep(vp, PVM, "vnpobj", 0); 129 } 130 vp->v_flag |= VOLOCK; 131 132 /* 133 * If the object is being terminated, wait for it to 134 * go away. 135 */ 136 while (((object = vp->v_object) != NULL) && (object->flags & OBJ_DEAD)) { 137 tsleep(object, PVM, "vadead", 0); 138 } 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 = OBJ_CANPERSIST; 146 147 /* 148 * Hold a reference to the vnode and initialize object data. 149 */ 150 VREF(vp); 151 object->un_pager.vnp.vnp_size = (vm_ooffset_t) size * PAGE_SIZE; 152 153 object->handle = handle; 154 vp->v_object = object; 155 } else { 156 /* 157 * vm_object_reference() will remove the object from the cache if 158 * found and gain a reference to the object. 159 */ 160 vm_object_reference(object); 161 } 162 163 if (vp->v_type == VREG) 164 vp->v_flag |= VVMIO; 165 166 vp->v_flag &= ~VOLOCK; 167 if (vp->v_flag & VOWANT) { 168 vp->v_flag &= ~VOWANT; 169 wakeup(vp); 170 } 171 return (object); 172} 173 174static void 175vnode_pager_dealloc(object) 176 vm_object_t object; 177{ 178 register struct vnode *vp = object->handle; 179 180 if (vp == NULL) 181 panic("vnode_pager_dealloc: pager already dealloced"); 182 183 if (object->paging_in_progress) { 184 int s = splbio(); 185 while (object->paging_in_progress) { 186 object->flags |= OBJ_PIPWNT; 187 tsleep(object, PVM, "vnpdea", 0); 188 } 189 splx(s); 190 } 191 192 object->handle = NULL; 193 194 vp->v_object = NULL; 195 vp->v_flag &= ~(VTEXT | VVMIO); 196 vp->v_flag |= VAGE; 197 vrele(vp); 198} 199 200static boolean_t 201vnode_pager_haspage(object, pindex, before, after) 202 vm_object_t object; 203 vm_pindex_t pindex; 204 int *before; 205 int *after; 206{ 207 struct vnode *vp = object->handle; 208 daddr_t bn; 209 int err; 210 daddr_t reqblock; 211 int poff; 212 int bsize; 213 int pagesperblock, blocksperpage; 214 215 /* 216 * If filesystem no longer mounted or offset beyond end of file we do 217 * not have the page. 218 */ 219 if ((vp->v_mount == NULL) || 220 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 221 return FALSE; 222 223 bsize = vp->v_mount->mnt_stat.f_iosize; 224 pagesperblock = bsize / PAGE_SIZE; 225 blocksperpage = 0; 226 if (pagesperblock > 0) { 227 reqblock = pindex / pagesperblock; 228 } else { 229 blocksperpage = (PAGE_SIZE / bsize); 230 reqblock = pindex * blocksperpage; 231 } 232 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 233 after, before); 234 if (err) 235 return TRUE; 236 if ( bn == -1) 237 return FALSE; 238 if (pagesperblock > 0) { 239 poff = pindex - (reqblock * pagesperblock); 240 if (before) { 241 *before *= pagesperblock; 242 *before += poff; 243 } 244 if (after) { 245 int numafter; 246 *after *= pagesperblock; 247 numafter = pagesperblock - (poff + 1); 248 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) { 249 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex))); 250 } 251 *after += numafter; 252 } 253 } else { 254 if (before) { 255 *before /= blocksperpage; 256 } 257 258 if (after) { 259 *after /= blocksperpage; 260 } 261 } 262 return TRUE; 263} 264 265/* 266 * Lets the VM system know about a change in size for a file. 267 * We adjust our own internal size and flush any cached pages in 268 * the associated object that are affected by the size change. 269 * 270 * Note: this routine may be invoked as a result of a pager put 271 * operation (possibly at object termination time), so we must be careful. 272 */ 273void 274vnode_pager_setsize(vp, nsize) 275 struct vnode *vp; 276 vm_ooffset_t nsize; 277{ 278 vm_object_t object = vp->v_object; 279 280 if (object == NULL) 281 return; 282 283 /* 284 * Hasn't changed size 285 */ 286 if (nsize == object->un_pager.vnp.vnp_size) 287 return; 288 289 /* 290 * File has shrunk. Toss any cached pages beyond the new EOF. 291 */ 292 if (nsize < object->un_pager.vnp.vnp_size) { 293 vm_ooffset_t nsizerounded; 294 nsizerounded = IDX_TO_OFF(OFF_TO_IDX(nsize + PAGE_SIZE - 1)); 295 if (nsizerounded < object->un_pager.vnp.vnp_size) { 296 vm_object_page_remove(object, 297 OFF_TO_IDX(nsize + PAGE_SIZE - 1), 298 OFF_TO_IDX(object->un_pager.vnp.vnp_size), 299 FALSE); 300 } 301 /* 302 * this gets rid of garbage at the end of a page that is now 303 * only partially backed by the vnode... 304 */ 305 if (nsize & PAGE_MASK) { 306 vm_offset_t kva; 307 vm_page_t m; 308 309 m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 310 if (m) { 311 kva = vm_pager_map_page(m); 312 bzero((caddr_t) kva + (nsize & PAGE_MASK), 313 (int) (round_page(nsize) - nsize)); 314 vm_pager_unmap_page(kva); 315 } 316 } 317 } 318 object->un_pager.vnp.vnp_size = nsize; 319 object->size = OFF_TO_IDX(nsize + PAGE_SIZE - 1); 320} 321 322void 323vnode_pager_umount(mp) 324 register struct mount *mp; 325{ 326 struct vnode *vp, *nvp; 327 328loop: 329 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 330 /* 331 * Vnode can be reclaimed by getnewvnode() while we 332 * traverse the list. 333 */ 334 if (vp->v_mount != mp) 335 goto loop; 336 337 /* 338 * Save the next pointer now since uncaching may terminate the 339 * object and render vnode invalid 340 */ 341 nvp = vp->v_mntvnodes.le_next; 342 343 if (vp->v_object != NULL) { 344 VOP_LOCK(vp); 345 vnode_pager_uncache(vp); 346 VOP_UNLOCK(vp); 347 } 348 } 349} 350 351/* 352 * Remove vnode associated object from the object cache. 353 * This routine must be called with the vnode locked. 354 * 355 * XXX unlock the vnode. 356 * We must do this since uncaching the object may result in its 357 * destruction which may initiate paging activity which may necessitate 358 * re-locking the vnode. 359 */ 360void 361vnode_pager_uncache(vp) 362 struct vnode *vp; 363{ 364 vm_object_t object; 365 366 /* 367 * Not a mapped vnode 368 */ 369 object = vp->v_object; 370 if (object == NULL) 371 return; 372 373 vm_object_reference(object); 374 VOP_UNLOCK(vp); 375 pager_cache(object, FALSE); 376 VOP_LOCK(vp); 377 return; 378} 379 380 381void 382vnode_pager_freepage(m) 383 vm_page_t m; 384{ 385 PAGE_WAKEUP(m); 386 vm_page_free(m); 387} 388 389/* 390 * calculate the linear (byte) disk address of specified virtual 391 * file address 392 */ 393static vm_offset_t 394vnode_pager_addr(vp, address, run) 395 struct vnode *vp; 396 vm_ooffset_t address; 397 int *run; 398{ 399 int rtaddress; 400 int bsize; 401 daddr_t block; 402 struct vnode *rtvp; 403 int err; 404 daddr_t vblock; 405 int voffset; 406 407 if ((int) address < 0) 408 return -1; 409 410 if (vp->v_mount == NULL) 411 return -1; 412 413 bsize = vp->v_mount->mnt_stat.f_iosize; 414 vblock = address / bsize; 415 voffset = address % bsize; 416 417 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 418 419 if (err || (block == -1)) 420 rtaddress = -1; 421 else { 422 rtaddress = block + voffset / DEV_BSIZE; 423 if( run) { 424 *run += 1; 425 *run *= bsize/PAGE_SIZE; 426 *run -= voffset/PAGE_SIZE; 427 } 428 } 429 430 return rtaddress; 431} 432 433/* 434 * interrupt routine for I/O completion 435 */ 436static void 437vnode_pager_iodone(bp) 438 struct buf *bp; 439{ 440 bp->b_flags |= B_DONE; 441 wakeup(bp); 442} 443 444/* 445 * small block file system vnode pager input 446 */ 447static int 448vnode_pager_input_smlfs(object, m) 449 vm_object_t object; 450 vm_page_t m; 451{ 452 int i; 453 int s; 454 struct vnode *dp, *vp; 455 struct buf *bp; 456 vm_offset_t kva; 457 int fileaddr; 458 vm_offset_t bsize; 459 int error = 0; 460 461 vp = object->handle; 462 if (vp->v_mount == NULL) 463 return VM_PAGER_BAD; 464 465 bsize = vp->v_mount->mnt_stat.f_iosize; 466 467 468 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 469 470 kva = vm_pager_map_page(m); 471 472 for (i = 0; i < PAGE_SIZE / bsize; i++) { 473 474 if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid)) 475 continue; 476 477 fileaddr = vnode_pager_addr(vp, 478 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0); 479 if (fileaddr != -1) { 480 bp = getpbuf(); 481 482 /* build a minimal buffer header */ 483 bp->b_flags = B_BUSY | B_READ | B_CALL; 484 bp->b_iodone = vnode_pager_iodone; 485 bp->b_proc = curproc; 486 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 487 if (bp->b_rcred != NOCRED) 488 crhold(bp->b_rcred); 489 if (bp->b_wcred != NOCRED) 490 crhold(bp->b_wcred); 491 bp->b_un.b_addr = (caddr_t) kva + i * bsize; 492 bp->b_blkno = fileaddr; 493 pbgetvp(dp, bp); 494 bp->b_bcount = bsize; 495 bp->b_bufsize = bsize; 496 497 /* do the input */ 498 VOP_STRATEGY(bp); 499 500 /* we definitely need to be at splbio here */ 501 502 s = splbio(); 503 while ((bp->b_flags & B_DONE) == 0) { 504 tsleep(bp, PVM, "vnsrd", 0); 505 } 506 splx(s); 507 if ((bp->b_flags & B_ERROR) != 0) 508 error = EIO; 509 510 /* 511 * free the buffer header back to the swap buffer pool 512 */ 513 relpbuf(bp); 514 if (error) 515 break; 516 517 vm_page_set_validclean(m, (i * bsize) & (PAGE_SIZE-1), bsize); 518 } else { 519 vm_page_set_validclean(m, (i * bsize) & (PAGE_SIZE-1), bsize); 520 bzero((caddr_t) kva + i * bsize, bsize); 521 } 522 } 523 vm_pager_unmap_page(kva); 524 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 525 m->flags &= ~PG_ZERO; 526 if (error) { 527 return VM_PAGER_ERROR; 528 } 529 return VM_PAGER_OK; 530 531} 532 533 534/* 535 * old style vnode pager output routine 536 */ 537static int 538vnode_pager_input_old(object, m) 539 vm_object_t object; 540 vm_page_t m; 541{ 542 struct uio auio; 543 struct iovec aiov; 544 int error; 545 int size; 546 vm_offset_t kva; 547 548 error = 0; 549 550 /* 551 * Return failure if beyond current EOF 552 */ 553 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 554 return VM_PAGER_BAD; 555 } else { 556 size = PAGE_SIZE; 557 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 558 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 559 560 /* 561 * Allocate a kernel virtual address and initialize so that 562 * we can use VOP_READ/WRITE routines. 563 */ 564 kva = vm_pager_map_page(m); 565 566 aiov.iov_base = (caddr_t) kva; 567 aiov.iov_len = size; 568 auio.uio_iov = &aiov; 569 auio.uio_iovcnt = 1; 570 auio.uio_offset = IDX_TO_OFF(m->pindex); 571 auio.uio_segflg = UIO_SYSSPACE; 572 auio.uio_rw = UIO_READ; 573 auio.uio_resid = size; 574 auio.uio_procp = (struct proc *) 0; 575 576 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred); 577 if (!error) { 578 register int count = size - auio.uio_resid; 579 580 if (count == 0) 581 error = EINVAL; 582 else if (count != PAGE_SIZE) 583 bzero((caddr_t) kva + count, PAGE_SIZE - count); 584 } 585 vm_pager_unmap_page(kva); 586 } 587 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 588 m->dirty = 0; 589 m->flags &= ~PG_ZERO; 590 return error ? VM_PAGER_ERROR : VM_PAGER_OK; 591} 592 593/* 594 * generic vnode pager input routine 595 */ 596 597static int 598vnode_pager_getpages(object, m, count, reqpage) 599 vm_object_t object; 600 vm_page_t *m; 601 int count; 602 int reqpage; 603{ 604 int rtval; 605 struct vnode *vp; 606 vp = object->handle; 607 rtval = VOP_GETPAGES(vp, m, count*PAGE_SIZE, reqpage, 0); 608 if (rtval == EOPNOTSUPP) 609 return vnode_pager_leaf_getpages(object, m, count, reqpage); 610 else 611 return rtval; 612} 613 614static int 615vnode_pager_leaf_getpages(object, m, count, reqpage) 616 vm_object_t object; 617 vm_page_t *m; 618 int count; 619 int reqpage; 620{ 621 vm_offset_t kva; 622 off_t foff; 623 int i, size, bsize, first, firstaddr; 624 struct vnode *dp, *vp; 625 int runpg; 626 int runend; 627 struct buf *bp; 628 int s; 629 int error = 0; 630 631 vp = object->handle; 632 if (vp->v_mount == NULL) 633 return VM_PAGER_BAD; 634 635 bsize = vp->v_mount->mnt_stat.f_iosize; 636 637 /* get the UNDERLYING device for the file with VOP_BMAP() */ 638 639 /* 640 * originally, we did not check for an error return value -- assuming 641 * an fs always has a bmap entry point -- that assumption is wrong!!! 642 */ 643 foff = IDX_TO_OFF(m[reqpage]->pindex); 644 645 /* 646 * if we can't bmap, use old VOP code 647 */ 648 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 649 for (i = 0; i < count; i++) { 650 if (i != reqpage) { 651 vnode_pager_freepage(m[i]); 652 } 653 } 654 cnt.v_vnodein++; 655 cnt.v_vnodepgsin++; 656 return vnode_pager_input_old(object, m[reqpage]); 657 658 /* 659 * if the blocksize is smaller than a page size, then use 660 * special small filesystem code. NFS sometimes has a small 661 * blocksize, but it can handle large reads itself. 662 */ 663 } else if ((PAGE_SIZE / bsize) > 1 && 664 (vp->v_mount->mnt_stat.f_type != MOUNT_NFS)) { 665 666 for (i = 0; i < count; i++) { 667 if (i != reqpage) { 668 vnode_pager_freepage(m[i]); 669 } 670 } 671 cnt.v_vnodein++; 672 cnt.v_vnodepgsin++; 673 return vnode_pager_input_smlfs(object, m[reqpage]); 674 } 675 /* 676 * if ANY DEV_BSIZE blocks are valid on a large filesystem block 677 * then, the entire page is valid -- 678 */ 679 if (m[reqpage]->valid) { 680 m[reqpage]->valid = VM_PAGE_BITS_ALL; 681 for (i = 0; i < count; i++) { 682 if (i != reqpage) 683 vnode_pager_freepage(m[i]); 684 } 685 return VM_PAGER_OK; 686 } 687 688 /* 689 * here on direct device I/O 690 */ 691 692 firstaddr = -1; 693 /* 694 * calculate the run that includes the required page 695 */ 696 for(first = 0, i = 0; i < count; i = runend) { 697 firstaddr = vnode_pager_addr(vp, 698 IDX_TO_OFF(m[i]->pindex), &runpg); 699 if (firstaddr == -1) { 700 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 701 panic("vnode_pager_putpages: unexpected missing page: firstaddr: %d, foff: %ld, vnp_size: %d", 702 firstaddr, foff, object->un_pager.vnp.vnp_size); 703 } 704 vnode_pager_freepage(m[i]); 705 runend = i + 1; 706 first = runend; 707 continue; 708 } 709 runend = i + runpg; 710 if (runend <= reqpage) { 711 int j; 712 for (j = i; j < runend; j++) { 713 vnode_pager_freepage(m[j]); 714 } 715 } else { 716 if (runpg < (count - first)) { 717 for (i = first + runpg; i < count; i++) 718 vnode_pager_freepage(m[i]); 719 count = first + runpg; 720 } 721 break; 722 } 723 first = runend; 724 } 725 726 /* 727 * the first and last page have been calculated now, move input pages 728 * to be zero based... 729 */ 730 if (first != 0) { 731 for (i = first; i < count; i++) { 732 m[i - first] = m[i]; 733 } 734 count -= first; 735 reqpage -= first; 736 } 737 738 /* 739 * calculate the file virtual address for the transfer 740 */ 741 foff = IDX_TO_OFF(m[0]->pindex); 742 743 /* 744 * calculate the size of the transfer 745 */ 746 size = count * PAGE_SIZE; 747 if ((foff + size) > object->un_pager.vnp.vnp_size) 748 size = object->un_pager.vnp.vnp_size - foff; 749 750 /* 751 * round up physical size for real devices 752 */ 753 if (dp->v_type == VBLK || dp->v_type == VCHR) 754 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 755 756 bp = getpbuf(); 757 kva = (vm_offset_t) bp->b_data; 758 759 /* 760 * and map the pages to be read into the kva 761 */ 762 pmap_qenter(kva, m, count); 763 764 /* build a minimal buffer header */ 765 bp->b_flags = B_BUSY | B_READ | B_CALL; 766 bp->b_iodone = vnode_pager_iodone; 767 /* B_PHYS is not set, but it is nice to fill this in */ 768 bp->b_proc = curproc; 769 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 770 if (bp->b_rcred != NOCRED) 771 crhold(bp->b_rcred); 772 if (bp->b_wcred != NOCRED) 773 crhold(bp->b_wcred); 774 bp->b_blkno = firstaddr; 775 pbgetvp(dp, bp); 776 bp->b_bcount = size; 777 bp->b_bufsize = size; 778 779 cnt.v_vnodein++; 780 cnt.v_vnodepgsin += count; 781 782 /* do the input */ 783 VOP_STRATEGY(bp); 784 785 s = splbio(); 786 /* we definitely need to be at splbio here */ 787 788 while ((bp->b_flags & B_DONE) == 0) { 789 tsleep(bp, PVM, "vnread", 0); 790 } 791 splx(s); 792 if ((bp->b_flags & B_ERROR) != 0) 793 error = EIO; 794 795 if (!error) { 796 if (size != count * PAGE_SIZE) 797 bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 798 } 799 pmap_qremove(kva, count); 800 801 /* 802 * free the buffer header back to the swap buffer pool 803 */ 804 relpbuf(bp); 805 806 for (i = 0; i < count; i++) { 807 pmap_clear_modify(VM_PAGE_TO_PHYS(m[i])); 808 m[i]->dirty = 0; 809 m[i]->valid = VM_PAGE_BITS_ALL; 810 m[i]->flags &= ~PG_ZERO; 811 if (i != reqpage) { 812 813 /* 814 * whether or not to leave the page activated is up in 815 * the air, but we should put the page on a page queue 816 * somewhere. (it already is in the object). Result: 817 * It appears that emperical results show that 818 * deactivating pages is best. 819 */ 820 821 /* 822 * just in case someone was asking for this page we 823 * now tell them that it is ok to use 824 */ 825 if (!error) { 826 vm_page_deactivate(m[i]); 827 PAGE_WAKEUP(m[i]); 828 } else { 829 vnode_pager_freepage(m[i]); 830 } 831 } 832 } 833 if (error) { 834 printf("vnode_pager_getpages: I/O read error\n"); 835 } 836 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 837} 838 839static int 840vnode_pager_putpages(object, m, count, sync, rtvals) 841 vm_object_t object; 842 vm_page_t *m; 843 int count; 844 boolean_t sync; 845 int *rtvals; 846{ 847 int rtval; 848 struct vnode *vp; 849 vp = object->handle; 850 rtval = VOP_PUTPAGES(vp, m, count*PAGE_SIZE, sync, rtvals, 0); 851 if (rtval == EOPNOTSUPP) 852 return vnode_pager_leaf_putpages(object, m, count, sync, rtvals); 853 else 854 return rtval; 855} 856 857/* 858 * generic vnode pager output routine 859 */ 860static int 861vnode_pager_leaf_putpages(object, m, count, sync, rtvals) 862 vm_object_t object; 863 vm_page_t *m; 864 int count; 865 boolean_t sync; 866 int *rtvals; 867{ 868 int i; 869 870 struct vnode *vp; 871 int maxsize, ncount; 872 vm_ooffset_t poffset; 873 struct uio auio; 874 struct iovec aiov; 875 int error; 876 877 vp = object->handle;; 878 for (i = 0; i < count; i++) 879 rtvals[i] = VM_PAGER_AGAIN; 880 881 if ((int) m[0]->pindex < 0) { 882 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", m[0]->pindex, m[0]->dirty); 883 rtvals[0] = VM_PAGER_BAD; 884 return VM_PAGER_BAD; 885 } 886 887 maxsize = count * PAGE_SIZE; 888 ncount = count; 889 890 poffset = IDX_TO_OFF(m[0]->pindex); 891 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 892 if (object->un_pager.vnp.vnp_size > poffset) 893 maxsize = object->un_pager.vnp.vnp_size - poffset; 894 else 895 maxsize = 0; 896 ncount = (maxsize + PAGE_SIZE - 1) / PAGE_SIZE; 897 if (ncount < count) { 898 for (i = ncount; i < count; i++) { 899 rtvals[i] = VM_PAGER_BAD; 900 } 901#ifdef BOGUS 902 if (ncount == 0) { 903 printf("vnode_pager_putpages: write past end of file: %d, %lu\n", 904 poffset, 905 (unsigned long) object->un_pager.vnp.vnp_size); 906 return rtvals[0]; 907 } 908#endif 909 } 910 } 911 912 for (i = 0; i < count; i++) { 913 m[i]->busy++; 914 m[i]->flags &= ~PG_BUSY; 915 } 916 917 aiov.iov_base = (caddr_t) 0; 918 aiov.iov_len = maxsize; 919 auio.uio_iov = &aiov; 920 auio.uio_iovcnt = 1; 921 auio.uio_offset = poffset; 922 auio.uio_segflg = UIO_NOCOPY; 923 auio.uio_rw = UIO_WRITE; 924 auio.uio_resid = maxsize; 925 auio.uio_procp = (struct proc *) 0; 926 error = VOP_WRITE(vp, &auio, IO_VMIO|(sync?IO_SYNC:0), curproc->p_ucred); 927 cnt.v_vnodeout++; 928 cnt.v_vnodepgsout += ncount; 929 930 if (error) { 931 printf("vnode_pager_putpages: I/O error %d\n", error); 932 } 933 if (auio.uio_resid) { 934 printf("vnode_pager_putpages: residual I/O %d at %ld\n", 935 auio.uio_resid, m[0]->pindex); 936 } 937 for (i = 0; i < count; i++) { 938 m[i]->busy--; 939 if (i < ncount) { 940 rtvals[i] = VM_PAGER_OK; 941 } 942 if ((m[i]->busy == 0) && (m[i]->flags & PG_WANTED)) 943 wakeup(m[i]); 944 } 945 return rtvals[0]; 946} 947 948struct vnode * 949vnode_pager_lock(object) 950 vm_object_t object; 951{ 952 for (; object != NULL; object = object->backing_object) { 953 if (object->type != OBJT_VNODE) 954 continue; 955 956 VOP_LOCK(object->handle); 957 return object->handle; 958 } 959 return NULL; 960} 961