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 1.64 1996/09/10 05:28:23 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/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 74 static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address, 75 int *run)); 76 static void vnode_pager_iodone __P((struct buf *bp)); 77 static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m)); 78 static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m)); 79 static void vnode_pager_dealloc __P((vm_object_t)); 80 static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int)); 81 static int vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *)); 82 static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *)); 83 84 struct 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 94 static int vnode_pager_leaf_getpages __P((vm_object_t object, vm_page_t *m, 95 int count, int reqpage)); 96 static 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 */ 104 vm_object_t 105 vnode_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) && 137 (object->flags & OBJ_DEAD)) { 138 tsleep(object, PVM, "vadead", 0); 139 } 140 141 if (object == NULL) { 142 /* 143 * And an object of the appropriate size 144 */ 145 object = vm_object_allocate(OBJT_VNODE, size); 146 if (vp->v_type == VREG) 147 object->flags = OBJ_CANPERSIST; 148 else 149 object->flags = 0; 150 151 /* 152 * Hold a reference to the vnode and initialize object data. 153 */ 154 vp->v_usecount++; 155 object->un_pager.vnp.vnp_size = (vm_ooffset_t) size * PAGE_SIZE; 156 157 object->handle = handle; 158 vp->v_object = object; 159 } else { 160 /* 161 * vm_object_reference() will remove the object from the cache if 162 * found and gain a reference to the object. 163 */ 164 vm_object_reference(object); 165 } 166 167 if (vp->v_type == VREG) 168 vp->v_flag |= VVMIO; 169 170 vp->v_flag &= ~VOLOCK; 171 if (vp->v_flag & VOWANT) { 172 vp->v_flag &= ~VOWANT; 173 wakeup(vp); 174 } 175 return (object); 176 } 177 178 static void 179 vnode_pager_dealloc(object) 180 vm_object_t object; 181 { 182 register struct vnode *vp = object->handle; 183 184 if (vp == NULL) 185 panic("vnode_pager_dealloc: pager already dealloced"); 186 187 if (object->paging_in_progress) { 188 int s = splbio(); 189 while (object->paging_in_progress) { 190 object->flags |= OBJ_PIPWNT; 191 tsleep(object, PVM, "vnpdea", 0); 192 } 193 splx(s); 194 } 195 196 object->handle = NULL; 197 198 vp->v_object = NULL; 199 vp->v_flag &= ~(VTEXT | VVMIO); 200 vp->v_flag |= VAGE; 201 vrele(vp); 202 } 203 204 static boolean_t 205 vnode_pager_haspage(object, pindex, before, after) 206 vm_object_t object; 207 vm_pindex_t pindex; 208 int *before; 209 int *after; 210 { 211 struct vnode *vp = object->handle; 212 daddr_t bn; 213 int err; 214 daddr_t reqblock; 215 int poff; 216 int bsize; 217 int pagesperblock, blocksperpage; 218 219 /* 220 * If filesystem no longer mounted or offset beyond end of file we do 221 * not have the page. 222 */ 223 if ((vp->v_mount == NULL) || 224 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 225 return FALSE; 226 227 bsize = vp->v_mount->mnt_stat.f_iosize; 228 pagesperblock = bsize / PAGE_SIZE; 229 blocksperpage = 0; 230 if (pagesperblock > 0) { 231 reqblock = pindex / pagesperblock; 232 } else { 233 blocksperpage = (PAGE_SIZE / bsize); 234 reqblock = pindex * blocksperpage; 235 } 236 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 237 after, before); 238 if (err) 239 return TRUE; 240 if ( bn == -1) 241 return FALSE; 242 if (pagesperblock > 0) { 243 poff = pindex - (reqblock * pagesperblock); 244 if (before) { 245 *before *= pagesperblock; 246 *before += poff; 247 } 248 if (after) { 249 int numafter; 250 *after *= pagesperblock; 251 numafter = pagesperblock - (poff + 1); 252 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) { 253 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex))); 254 } 255 *after += numafter; 256 } 257 } else { 258 if (before) { 259 *before /= blocksperpage; 260 } 261 262 if (after) { 263 *after /= blocksperpage; 264 } 265 } 266 return TRUE; 267 } 268 269 /* 270 * Lets the VM system know about a change in size for a file. 271 * We adjust our own internal size and flush any cached pages in 272 * the associated object that are affected by the size change. 273 * 274 * Note: this routine may be invoked as a result of a pager put 275 * operation (possibly at object termination time), so we must be careful. 276 */ 277 void 278 vnode_pager_setsize(vp, nsize) 279 struct vnode *vp; 280 vm_ooffset_t nsize; 281 { 282 vm_object_t object = vp->v_object; 283 284 if (object == NULL) 285 return; 286 287 /* 288 * Hasn't changed size 289 */ 290 if (nsize == object->un_pager.vnp.vnp_size) 291 return; 292 293 /* 294 * File has shrunk. Toss any cached pages beyond the new EOF. 295 */ 296 if (nsize < object->un_pager.vnp.vnp_size) { 297 vm_ooffset_t nsizerounded; 298 nsizerounded = IDX_TO_OFF(OFF_TO_IDX(nsize + PAGE_MASK)); 299 if (nsizerounded < object->un_pager.vnp.vnp_size) { 300 vm_object_page_remove(object, 301 OFF_TO_IDX(nsize + PAGE_MASK), 302 OFF_TO_IDX(object->un_pager.vnp.vnp_size), 303 FALSE); 304 } 305 /* 306 * this gets rid of garbage at the end of a page that is now 307 * only partially backed by the vnode... 308 */ 309 if (nsize & PAGE_MASK) { 310 vm_offset_t kva; 311 vm_page_t m; 312 313 m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 314 if (m) { 315 kva = vm_pager_map_page(m); 316 bzero((caddr_t) kva + (nsize & PAGE_MASK), 317 (int) (round_page(nsize) - nsize)); 318 vm_pager_unmap_page(kva); 319 } 320 } 321 } 322 object->un_pager.vnp.vnp_size = nsize; 323 object->size = OFF_TO_IDX(nsize + PAGE_MASK); 324 } 325 326 void 327 vnode_pager_umount(mp) 328 register struct mount *mp; 329 { 330 struct vnode *vp, *nvp; 331 332 loop: 333 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 334 /* 335 * Vnode can be reclaimed by getnewvnode() while we 336 * traverse the list. 337 */ 338 if (vp->v_mount != mp) 339 goto loop; 340 341 /* 342 * Save the next pointer now since uncaching may terminate the 343 * object and render vnode invalid 344 */ 345 nvp = vp->v_mntvnodes.le_next; 346 347 if (vp->v_object != NULL) { 348 VOP_LOCK(vp); 349 vnode_pager_uncache(vp); 350 VOP_UNLOCK(vp); 351 } 352 } 353 } 354 355 /* 356 * Remove vnode associated object from the object cache. 357 * This routine must be called with the vnode locked. 358 * 359 * XXX unlock the vnode. 360 * We must do this since uncaching the object may result in its 361 * destruction which may initiate paging activity which may necessitate 362 * re-locking the vnode. 363 */ 364 void 365 vnode_pager_uncache(vp) 366 struct vnode *vp; 367 { 368 vm_object_t object; 369 370 /* 371 * Not a mapped vnode 372 */ 373 object = vp->v_object; 374 if (object == NULL) 375 return; 376 377 vm_object_reference(object); 378 379 /* 380 * XXX We really should handle locking on 381 * VBLK devices... 382 */ 383 if (vp->v_type != VBLK) 384 VOP_UNLOCK(vp); 385 pager_cache(object, FALSE); 386 if (vp->v_type != VBLK) 387 VOP_LOCK(vp); 388 return; 389 } 390 391 392 void 393 vnode_pager_freepage(m) 394 vm_page_t m; 395 { 396 PAGE_WAKEUP(m); 397 vm_page_free(m); 398 } 399 400 /* 401 * calculate the linear (byte) disk address of specified virtual 402 * file address 403 */ 404 static vm_offset_t 405 vnode_pager_addr(vp, address, run) 406 struct vnode *vp; 407 vm_ooffset_t address; 408 int *run; 409 { 410 int rtaddress; 411 int bsize; 412 daddr_t block; 413 struct vnode *rtvp; 414 int err; 415 daddr_t vblock; 416 int voffset; 417 418 if ((int) address < 0) 419 return -1; 420 421 if (vp->v_mount == NULL) 422 return -1; 423 424 bsize = vp->v_mount->mnt_stat.f_iosize; 425 vblock = address / bsize; 426 voffset = address % bsize; 427 428 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 429 430 if (err || (block == -1)) 431 rtaddress = -1; 432 else { 433 rtaddress = block + voffset / DEV_BSIZE; 434 if( run) { 435 *run += 1; 436 *run *= bsize/PAGE_SIZE; 437 *run -= voffset/PAGE_SIZE; 438 } 439 } 440 441 return rtaddress; 442 } 443 444 /* 445 * interrupt routine for I/O completion 446 */ 447 static void 448 vnode_pager_iodone(bp) 449 struct buf *bp; 450 { 451 bp->b_flags |= B_DONE; 452 wakeup(bp); 453 } 454 455 /* 456 * small block file system vnode pager input 457 */ 458 static int 459 vnode_pager_input_smlfs(object, m) 460 vm_object_t object; 461 vm_page_t m; 462 { 463 int i; 464 int s; 465 struct vnode *dp, *vp; 466 struct buf *bp; 467 vm_offset_t kva; 468 int fileaddr; 469 vm_offset_t bsize; 470 int error = 0; 471 472 vp = object->handle; 473 if (vp->v_mount == NULL) 474 return VM_PAGER_BAD; 475 476 bsize = vp->v_mount->mnt_stat.f_iosize; 477 478 479 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 480 481 kva = vm_pager_map_page(m); 482 483 for (i = 0; i < PAGE_SIZE / bsize; i++) { 484 485 if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid)) 486 continue; 487 488 fileaddr = vnode_pager_addr(vp, 489 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0); 490 if (fileaddr != -1) { 491 bp = getpbuf(); 492 493 /* build a minimal buffer header */ 494 bp->b_flags = B_BUSY | B_READ | B_CALL; 495 bp->b_iodone = vnode_pager_iodone; 496 bp->b_proc = curproc; 497 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 498 if (bp->b_rcred != NOCRED) 499 crhold(bp->b_rcred); 500 if (bp->b_wcred != NOCRED) 501 crhold(bp->b_wcred); 502 bp->b_un.b_addr = (caddr_t) kva + i * bsize; 503 bp->b_blkno = fileaddr; 504 pbgetvp(dp, bp); 505 bp->b_bcount = bsize; 506 bp->b_bufsize = bsize; 507 508 /* do the input */ 509 VOP_STRATEGY(bp); 510 511 /* we definitely need to be at splbio here */ 512 513 s = splbio(); 514 while ((bp->b_flags & B_DONE) == 0) { 515 tsleep(bp, PVM, "vnsrd", 0); 516 } 517 splx(s); 518 if ((bp->b_flags & B_ERROR) != 0) 519 error = EIO; 520 521 /* 522 * free the buffer header back to the swap buffer pool 523 */ 524 relpbuf(bp); 525 if (error) 526 break; 527 528 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 529 } else { 530 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 531 bzero((caddr_t) kva + i * bsize, bsize); 532 } 533 } 534 vm_pager_unmap_page(kva); 535 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 536 m->flags &= ~PG_ZERO; 537 if (error) { 538 return VM_PAGER_ERROR; 539 } 540 return VM_PAGER_OK; 541 542 } 543 544 545 /* 546 * old style vnode pager output routine 547 */ 548 static int 549 vnode_pager_input_old(object, m) 550 vm_object_t object; 551 vm_page_t m; 552 { 553 struct uio auio; 554 struct iovec aiov; 555 int error; 556 int size; 557 vm_offset_t kva; 558 559 error = 0; 560 561 /* 562 * Return failure if beyond current EOF 563 */ 564 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 565 return VM_PAGER_BAD; 566 } else { 567 size = PAGE_SIZE; 568 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 569 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 570 571 /* 572 * Allocate a kernel virtual address and initialize so that 573 * we can use VOP_READ/WRITE routines. 574 */ 575 kva = vm_pager_map_page(m); 576 577 aiov.iov_base = (caddr_t) kva; 578 aiov.iov_len = size; 579 auio.uio_iov = &aiov; 580 auio.uio_iovcnt = 1; 581 auio.uio_offset = IDX_TO_OFF(m->pindex); 582 auio.uio_segflg = UIO_SYSSPACE; 583 auio.uio_rw = UIO_READ; 584 auio.uio_resid = size; 585 auio.uio_procp = (struct proc *) 0; 586 587 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred); 588 if (!error) { 589 register int count = size - auio.uio_resid; 590 591 if (count == 0) 592 error = EINVAL; 593 else if (count != PAGE_SIZE) 594 bzero((caddr_t) kva + count, PAGE_SIZE - count); 595 } 596 vm_pager_unmap_page(kva); 597 } 598 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 599 m->dirty = 0; 600 m->flags &= ~PG_ZERO; 601 return error ? VM_PAGER_ERROR : VM_PAGER_OK; 602 } 603 604 /* 605 * generic vnode pager input routine 606 */ 607 608 static int 609 vnode_pager_getpages(object, m, count, reqpage) 610 vm_object_t object; 611 vm_page_t *m; 612 int count; 613 int reqpage; 614 { 615 int rtval; 616 struct vnode *vp; 617 if (object->flags & OBJ_VNODE_GONE) 618 return VM_PAGER_ERROR; 619 vp = object->handle; 620 rtval = VOP_GETPAGES(vp, m, count*PAGE_SIZE, reqpage, 0); 621 if (rtval == EOPNOTSUPP) 622 return vnode_pager_leaf_getpages(object, m, count, reqpage); 623 else 624 return rtval; 625 } 626 627 static int 628 vnode_pager_leaf_getpages(object, m, count, reqpage) 629 vm_object_t object; 630 vm_page_t *m; 631 int count; 632 int reqpage; 633 { 634 vm_offset_t kva; 635 off_t foff; 636 int i, size, bsize, first, firstaddr; 637 struct vnode *dp, *vp; 638 int runpg; 639 int runend; 640 struct buf *bp; 641 int s; 642 int error = 0; 643 644 vp = object->handle; 645 if (vp->v_mount == NULL) 646 return VM_PAGER_BAD; 647 648 bsize = vp->v_mount->mnt_stat.f_iosize; 649 650 /* get the UNDERLYING device for the file with VOP_BMAP() */ 651 652 /* 653 * originally, we did not check for an error return value -- assuming 654 * an fs always has a bmap entry point -- that assumption is wrong!!! 655 */ 656 foff = IDX_TO_OFF(m[reqpage]->pindex); 657 658 /* 659 * if we can't bmap, use old VOP code 660 */ 661 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 662 for (i = 0; i < count; i++) { 663 if (i != reqpage) { 664 vnode_pager_freepage(m[i]); 665 } 666 } 667 cnt.v_vnodein++; 668 cnt.v_vnodepgsin++; 669 return vnode_pager_input_old(object, m[reqpage]); 670 671 /* 672 * if the blocksize is smaller than a page size, then use 673 * special small filesystem code. NFS sometimes has a small 674 * blocksize, but it can handle large reads itself. 675 */ 676 } else if ((PAGE_SIZE / bsize) > 1 && 677 (vp->v_mount->mnt_stat.f_type != MOUNT_NFS)) { 678 679 for (i = 0; i < count; i++) { 680 if (i != reqpage) { 681 vnode_pager_freepage(m[i]); 682 } 683 } 684 cnt.v_vnodein++; 685 cnt.v_vnodepgsin++; 686 return vnode_pager_input_smlfs(object, m[reqpage]); 687 } 688 /* 689 * if ANY DEV_BSIZE blocks are valid on a large filesystem block 690 * then, the entire page is valid -- 691 */ 692 if (m[reqpage]->valid) { 693 m[reqpage]->valid = VM_PAGE_BITS_ALL; 694 for (i = 0; i < count; i++) { 695 if (i != reqpage) 696 vnode_pager_freepage(m[i]); 697 } 698 return VM_PAGER_OK; 699 } 700 701 /* 702 * here on direct device I/O 703 */ 704 705 firstaddr = -1; 706 /* 707 * calculate the run that includes the required page 708 */ 709 for(first = 0, i = 0; i < count; i = runend) { 710 firstaddr = vnode_pager_addr(vp, 711 IDX_TO_OFF(m[i]->pindex), &runpg); 712 if (firstaddr == -1) { 713 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 714 panic("vnode_pager_putpages: unexpected missing page: firstaddr: %d, foff: %ld, vnp_size: %d", 715 firstaddr, foff, object->un_pager.vnp.vnp_size); 716 } 717 vnode_pager_freepage(m[i]); 718 runend = i + 1; 719 first = runend; 720 continue; 721 } 722 runend = i + runpg; 723 if (runend <= reqpage) { 724 int j; 725 for (j = i; j < runend; j++) { 726 vnode_pager_freepage(m[j]); 727 } 728 } else { 729 if (runpg < (count - first)) { 730 for (i = first + runpg; i < count; i++) 731 vnode_pager_freepage(m[i]); 732 count = first + runpg; 733 } 734 break; 735 } 736 first = runend; 737 } 738 739 /* 740 * the first and last page have been calculated now, move input pages 741 * to be zero based... 742 */ 743 if (first != 0) { 744 for (i = first; i < count; i++) { 745 m[i - first] = m[i]; 746 } 747 count -= first; 748 reqpage -= first; 749 } 750 751 /* 752 * calculate the file virtual address for the transfer 753 */ 754 foff = IDX_TO_OFF(m[0]->pindex); 755 756 /* 757 * calculate the size of the transfer 758 */ 759 size = count * PAGE_SIZE; 760 if ((foff + size) > object->un_pager.vnp.vnp_size) 761 size = object->un_pager.vnp.vnp_size - foff; 762 763 /* 764 * round up physical size for real devices 765 */ 766 if (dp->v_type == VBLK || dp->v_type == VCHR) 767 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 768 769 bp = getpbuf(); 770 kva = (vm_offset_t) bp->b_data; 771 772 /* 773 * and map the pages to be read into the kva 774 */ 775 pmap_qenter(kva, m, count); 776 777 /* build a minimal buffer header */ 778 bp->b_flags = B_BUSY | B_READ | B_CALL; 779 bp->b_iodone = vnode_pager_iodone; 780 /* B_PHYS is not set, but it is nice to fill this in */ 781 bp->b_proc = curproc; 782 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 783 if (bp->b_rcred != NOCRED) 784 crhold(bp->b_rcred); 785 if (bp->b_wcred != NOCRED) 786 crhold(bp->b_wcred); 787 bp->b_blkno = firstaddr; 788 pbgetvp(dp, bp); 789 bp->b_bcount = size; 790 bp->b_bufsize = size; 791 792 cnt.v_vnodein++; 793 cnt.v_vnodepgsin += count; 794 795 /* do the input */ 796 VOP_STRATEGY(bp); 797 798 s = splbio(); 799 /* we definitely need to be at splbio here */ 800 801 while ((bp->b_flags & B_DONE) == 0) { 802 tsleep(bp, PVM, "vnread", 0); 803 } 804 splx(s); 805 if ((bp->b_flags & B_ERROR) != 0) 806 error = EIO; 807 808 if (!error) { 809 if (size != count * PAGE_SIZE) 810 bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 811 } 812 pmap_qremove(kva, count); 813 814 /* 815 * free the buffer header back to the swap buffer pool 816 */ 817 relpbuf(bp); 818 819 for (i = 0; i < count; i++) { 820 pmap_clear_modify(VM_PAGE_TO_PHYS(m[i])); 821 m[i]->dirty = 0; 822 m[i]->valid = VM_PAGE_BITS_ALL; 823 m[i]->flags &= ~PG_ZERO; 824 if (i != reqpage) { 825 826 /* 827 * whether or not to leave the page activated is up in 828 * the air, but we should put the page on a page queue 829 * somewhere. (it already is in the object). Result: 830 * It appears that emperical results show that 831 * deactivating pages is best. 832 */ 833 834 /* 835 * just in case someone was asking for this page we 836 * now tell them that it is ok to use 837 */ 838 if (!error) { 839 vm_page_deactivate(m[i]); 840 PAGE_WAKEUP(m[i]); 841 } else { 842 vnode_pager_freepage(m[i]); 843 } 844 } 845 } 846 if (error) { 847 printf("vnode_pager_getpages: I/O read error\n"); 848 } 849 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 850 } 851 852 static int 853 vnode_pager_putpages(object, m, count, sync, rtvals) 854 vm_object_t object; 855 vm_page_t *m; 856 int count; 857 boolean_t sync; 858 int *rtvals; 859 { 860 int rtval; 861 struct vnode *vp; 862 863 if (object->flags & OBJ_VNODE_GONE) 864 return VM_PAGER_ERROR; 865 866 vp = object->handle; 867 rtval = VOP_PUTPAGES(vp, m, count*PAGE_SIZE, sync, rtvals, 0); 868 if (rtval == EOPNOTSUPP) 869 return vnode_pager_leaf_putpages(object, m, count, sync, rtvals); 870 else 871 return rtval; 872 } 873 874 /* 875 * generic vnode pager output routine 876 */ 877 static int 878 vnode_pager_leaf_putpages(object, m, count, sync, rtvals) 879 vm_object_t object; 880 vm_page_t *m; 881 int count; 882 boolean_t sync; 883 int *rtvals; 884 { 885 int i; 886 887 struct vnode *vp; 888 int maxsize, ncount; 889 vm_ooffset_t poffset; 890 struct uio auio; 891 struct iovec aiov; 892 int error; 893 894 vp = object->handle;; 895 for (i = 0; i < count; i++) 896 rtvals[i] = VM_PAGER_AGAIN; 897 898 if ((int) m[0]->pindex < 0) { 899 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", m[0]->pindex, m[0]->dirty); 900 rtvals[0] = VM_PAGER_BAD; 901 return VM_PAGER_BAD; 902 } 903 904 maxsize = count * PAGE_SIZE; 905 ncount = count; 906 907 poffset = IDX_TO_OFF(m[0]->pindex); 908 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 909 if (object->un_pager.vnp.vnp_size > poffset) 910 maxsize = object->un_pager.vnp.vnp_size - poffset; 911 else 912 maxsize = 0; 913 ncount = btoc(maxsize); 914 if (ncount < count) { 915 for (i = ncount; i < count; i++) { 916 rtvals[i] = VM_PAGER_BAD; 917 } 918 #ifdef BOGUS 919 if (ncount == 0) { 920 printf("vnode_pager_putpages: write past end of file: %d, %lu\n", 921 poffset, 922 (unsigned long) object->un_pager.vnp.vnp_size); 923 return rtvals[0]; 924 } 925 #endif 926 } 927 } 928 929 for (i = 0; i < count; i++) { 930 m[i]->busy++; 931 m[i]->flags &= ~PG_BUSY; 932 } 933 934 aiov.iov_base = (caddr_t) 0; 935 aiov.iov_len = maxsize; 936 auio.uio_iov = &aiov; 937 auio.uio_iovcnt = 1; 938 auio.uio_offset = poffset; 939 auio.uio_segflg = UIO_NOCOPY; 940 auio.uio_rw = UIO_WRITE; 941 auio.uio_resid = maxsize; 942 auio.uio_procp = (struct proc *) 0; 943 error = VOP_WRITE(vp, &auio, IO_VMIO|(sync?IO_SYNC:0), curproc->p_ucred); 944 cnt.v_vnodeout++; 945 cnt.v_vnodepgsout += ncount; 946 947 if (error) { 948 printf("vnode_pager_putpages: I/O error %d\n", error); 949 } 950 if (auio.uio_resid) { 951 printf("vnode_pager_putpages: residual I/O %d at %ld\n", 952 auio.uio_resid, m[0]->pindex); 953 } 954 for (i = 0; i < count; i++) { 955 m[i]->busy--; 956 if (i < ncount) { 957 rtvals[i] = VM_PAGER_OK; 958 } 959 if ((m[i]->busy == 0) && (m[i]->flags & PG_WANTED)) 960 wakeup(m[i]); 961 } 962 return rtvals[0]; 963 } 964 965 struct vnode * 966 vnode_pager_lock(object) 967 vm_object_t object; 968 { 969 for (; object != NULL; object = object->backing_object) { 970 if (object->type != OBJT_VNODE) 971 continue; 972 973 VOP_LOCK(object->handle); 974 return object->handle; 975 } 976 return NULL; 977 } 978