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 1.102 1999/01/21 08:29:12 dillon 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 void 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 int vnode_pbuf_freecnt = -1; /* start out unlimited */ 92 93 94 /* 95 * Allocate (or lookup) pager for a vnode. 96 * Handle is a vnode pointer. 97 */ 98 vm_object_t 99 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 100 vm_ooffset_t offset) 101 { 102 vm_object_t object; 103 struct vnode *vp; 104 105 /* 106 * Pageout to vnode, no can do yet. 107 */ 108 if (handle == NULL) 109 return (NULL); 110 111 /* 112 * XXX hack - This initialization should be put somewhere else. 113 */ 114 if (vnode_pbuf_freecnt < 0) { 115 vnode_pbuf_freecnt = nswbuf / 2 + 1; 116 } 117 118 vp = (struct vnode *) handle; 119 120 /* 121 * Prevent race condition when allocating the object. This 122 * can happen with NFS vnodes since the nfsnode isn't locked. 123 */ 124 while (vp->v_flag & VOLOCK) { 125 vp->v_flag |= VOWANT; 126 tsleep(vp, PVM, "vnpobj", 0); 127 } 128 vp->v_flag |= VOLOCK; 129 130 /* 131 * If the object is being terminated, wait for it to 132 * go away. 133 */ 134 while (((object = vp->v_object) != NULL) && 135 (object->flags & OBJ_DEAD)) { 136 tsleep(object, PVM, "vadead", 0); 137 } 138 139 if (vp->v_usecount == 0) 140 panic("vnode_pager_alloc: no vnode reference"); 141 142 if (object == NULL) { 143 /* 144 * And an object of the appropriate size 145 */ 146 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); 147 object->flags = 0; 148 149 object->un_pager.vnp.vnp_size = size; 150 151 object->handle = handle; 152 vp->v_object = object; 153 vp->v_usecount++; 154 } else { 155 object->ref_count++; 156 vp->v_usecount++; 157 } 158 159 vp->v_flag &= ~VOLOCK; 160 if (vp->v_flag & VOWANT) { 161 vp->v_flag &= ~VOWANT; 162 wakeup(vp); 163 } 164 return (object); 165 } 166 167 static void 168 vnode_pager_dealloc(object) 169 vm_object_t object; 170 { 171 register struct vnode *vp = object->handle; 172 173 if (vp == NULL) 174 panic("vnode_pager_dealloc: pager already dealloced"); 175 176 vm_object_pip_wait(object, "vnpdea"); 177 178 object->handle = NULL; 179 object->type = OBJT_DEAD; 180 vp->v_object = NULL; 181 vp->v_flag &= ~(VTEXT | VOBJBUF); 182 } 183 184 static boolean_t 185 vnode_pager_haspage(object, pindex, before, after) 186 vm_object_t object; 187 vm_pindex_t pindex; 188 int *before; 189 int *after; 190 { 191 struct vnode *vp = object->handle; 192 daddr_t bn; 193 int err; 194 daddr_t reqblock; 195 int poff; 196 int bsize; 197 int pagesperblock, blocksperpage; 198 199 if ((vp == NULL) || (vp->v_flag & VDOOMED)) 200 return FALSE; 201 202 /* 203 * If filesystem no longer mounted or offset beyond end of file we do 204 * not have the page. 205 */ 206 if ((vp->v_mount == NULL) || 207 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)) 208 return FALSE; 209 210 bsize = vp->v_mount->mnt_stat.f_iosize; 211 pagesperblock = bsize / PAGE_SIZE; 212 blocksperpage = 0; 213 if (pagesperblock > 0) { 214 reqblock = pindex / pagesperblock; 215 } else { 216 blocksperpage = (PAGE_SIZE / bsize); 217 reqblock = pindex * blocksperpage; 218 } 219 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn, 220 after, before); 221 if (err) 222 return TRUE; 223 if ( bn == -1) 224 return FALSE; 225 if (pagesperblock > 0) { 226 poff = pindex - (reqblock * pagesperblock); 227 if (before) { 228 *before *= pagesperblock; 229 *before += poff; 230 } 231 if (after) { 232 int numafter; 233 *after *= pagesperblock; 234 numafter = pagesperblock - (poff + 1); 235 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) { 236 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex))); 237 } 238 *after += numafter; 239 } 240 } else { 241 if (before) { 242 *before /= blocksperpage; 243 } 244 245 if (after) { 246 *after /= blocksperpage; 247 } 248 } 249 return TRUE; 250 } 251 252 /* 253 * Lets the VM system know about a change in size for a file. 254 * We adjust our own internal size and flush any cached pages in 255 * the associated object that are affected by the size change. 256 * 257 * Note: this routine may be invoked as a result of a pager put 258 * operation (possibly at object termination time), so we must be careful. 259 */ 260 void 261 vnode_pager_setsize(vp, nsize) 262 struct vnode *vp; 263 vm_ooffset_t nsize; 264 { 265 vm_pindex_t nobjsize; 266 vm_object_t object = vp->v_object; 267 268 if (object == NULL) 269 return; 270 271 /* 272 * Hasn't changed size 273 */ 274 if (nsize == object->un_pager.vnp.vnp_size) 275 return; 276 277 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); 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_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size); 284 if (nobjsize < object->size) { 285 vm_object_page_remove(object, nobjsize, object->size, 286 FALSE); 287 } 288 /* 289 * this gets rid of garbage at the end of a page that is now 290 * only partially backed by the vnode... 291 */ 292 if (nsize & PAGE_MASK) { 293 vm_offset_t kva; 294 vm_page_t m; 295 296 m = vm_page_lookup(object, OFF_TO_IDX(nsize)); 297 if (m) { 298 kva = vm_pager_map_page(m); 299 bzero((caddr_t) kva + (nsize & PAGE_MASK), 300 (int) (round_page(nsize) - nsize)); 301 vm_pager_unmap_page(kva); 302 } 303 } 304 } 305 object->un_pager.vnp.vnp_size = nsize; 306 object->size = nobjsize; 307 } 308 309 void 310 vnode_pager_freepage(m) 311 vm_page_t m; 312 { 313 vm_page_free(m); 314 } 315 316 /* 317 * calculate the linear (byte) disk address of specified virtual 318 * file address 319 */ 320 static vm_offset_t 321 vnode_pager_addr(vp, address, run) 322 struct vnode *vp; 323 vm_ooffset_t address; 324 int *run; 325 { 326 int rtaddress; 327 int bsize; 328 daddr_t block; 329 struct vnode *rtvp; 330 int err; 331 daddr_t vblock; 332 int voffset; 333 334 if ((int) address < 0) 335 return -1; 336 337 if (vp->v_mount == NULL) 338 return -1; 339 340 bsize = vp->v_mount->mnt_stat.f_iosize; 341 vblock = address / bsize; 342 voffset = address % bsize; 343 344 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL); 345 346 if (err || (block == -1)) 347 rtaddress = -1; 348 else { 349 rtaddress = block + voffset / DEV_BSIZE; 350 if( run) { 351 *run += 1; 352 *run *= bsize/PAGE_SIZE; 353 *run -= voffset/PAGE_SIZE; 354 } 355 } 356 357 return rtaddress; 358 } 359 360 /* 361 * interrupt routine for I/O completion 362 */ 363 static void 364 vnode_pager_iodone(bp) 365 struct buf *bp; 366 { 367 bp->b_flags |= B_DONE; 368 wakeup(bp); 369 } 370 371 /* 372 * small block file system vnode pager input 373 */ 374 static int 375 vnode_pager_input_smlfs(object, m) 376 vm_object_t object; 377 vm_page_t m; 378 { 379 int i; 380 int s; 381 struct vnode *dp, *vp; 382 struct buf *bp; 383 vm_offset_t kva; 384 int fileaddr; 385 vm_offset_t bsize; 386 int error = 0; 387 388 vp = object->handle; 389 if (vp->v_mount == NULL) 390 return VM_PAGER_BAD; 391 392 bsize = vp->v_mount->mnt_stat.f_iosize; 393 394 395 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL); 396 397 kva = vm_pager_map_page(m); 398 399 for (i = 0; i < PAGE_SIZE / bsize; i++) { 400 401 if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid)) 402 continue; 403 404 fileaddr = vnode_pager_addr(vp, 405 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0); 406 if (fileaddr != -1) { 407 bp = getpbuf(&vnode_pbuf_freecnt); 408 409 /* build a minimal buffer header */ 410 bp->b_flags = B_BUSY | B_READ | B_CALL; 411 bp->b_iodone = vnode_pager_iodone; 412 bp->b_proc = curproc; 413 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 414 if (bp->b_rcred != NOCRED) 415 crhold(bp->b_rcred); 416 if (bp->b_wcred != NOCRED) 417 crhold(bp->b_wcred); 418 bp->b_data = (caddr_t) kva + i * bsize; 419 bp->b_blkno = fileaddr; 420 pbgetvp(dp, bp); 421 bp->b_bcount = bsize; 422 bp->b_bufsize = bsize; 423 424 /* do the input */ 425 VOP_STRATEGY(bp->b_vp, bp); 426 427 /* we definitely need to be at splvm here */ 428 429 s = splvm(); 430 while ((bp->b_flags & B_DONE) == 0) { 431 tsleep(bp, PVM, "vnsrd", 0); 432 } 433 splx(s); 434 if ((bp->b_flags & B_ERROR) != 0) 435 error = EIO; 436 437 /* 438 * free the buffer header back to the swap buffer pool 439 */ 440 relpbuf(bp, &vnode_pbuf_freecnt); 441 if (error) 442 break; 443 444 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 445 } else { 446 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize); 447 bzero((caddr_t) kva + i * bsize, bsize); 448 } 449 } 450 vm_pager_unmap_page(kva); 451 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 452 vm_page_flag_clear(m, PG_ZERO); 453 if (error) { 454 return VM_PAGER_ERROR; 455 } 456 return VM_PAGER_OK; 457 458 } 459 460 461 /* 462 * old style vnode pager output routine 463 */ 464 static int 465 vnode_pager_input_old(object, m) 466 vm_object_t object; 467 vm_page_t m; 468 { 469 struct uio auio; 470 struct iovec aiov; 471 int error; 472 int size; 473 vm_offset_t kva; 474 475 error = 0; 476 477 /* 478 * Return failure if beyond current EOF 479 */ 480 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { 481 return VM_PAGER_BAD; 482 } else { 483 size = PAGE_SIZE; 484 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) 485 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); 486 487 /* 488 * Allocate a kernel virtual address and initialize so that 489 * we can use VOP_READ/WRITE routines. 490 */ 491 kva = vm_pager_map_page(m); 492 493 aiov.iov_base = (caddr_t) kva; 494 aiov.iov_len = size; 495 auio.uio_iov = &aiov; 496 auio.uio_iovcnt = 1; 497 auio.uio_offset = IDX_TO_OFF(m->pindex); 498 auio.uio_segflg = UIO_SYSSPACE; 499 auio.uio_rw = UIO_READ; 500 auio.uio_resid = size; 501 auio.uio_procp = curproc; 502 503 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred); 504 if (!error) { 505 register int count = size - auio.uio_resid; 506 507 if (count == 0) 508 error = EINVAL; 509 else if (count != PAGE_SIZE) 510 bzero((caddr_t) kva + count, PAGE_SIZE - count); 511 } 512 vm_pager_unmap_page(kva); 513 } 514 pmap_clear_modify(VM_PAGE_TO_PHYS(m)); 515 m->dirty = 0; 516 vm_page_flag_clear(m, PG_ZERO); 517 if (!error) 518 m->valid = VM_PAGE_BITS_ALL; 519 return error ? VM_PAGER_ERROR : VM_PAGER_OK; 520 } 521 522 /* 523 * generic vnode pager input routine 524 */ 525 526 /* 527 * EOPNOTSUPP is no longer legal. For local media VFS's that do not 528 * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to 529 * vnode_pager_generic_getpages() to implement the previous behaviour. 530 * 531 * All other FS's should use the bypass to get to the local media 532 * backing vp's VOP_GETPAGES. 533 */ 534 static int 535 vnode_pager_getpages(object, m, count, reqpage) 536 vm_object_t object; 537 vm_page_t *m; 538 int count; 539 int reqpage; 540 { 541 int rtval; 542 struct vnode *vp; 543 int bytes = count * PAGE_SIZE; 544 545 vp = object->handle; 546 /* 547 * XXX temporary diagnostic message to help track stale FS code, 548 * Returning EOPNOTSUPP from here may make things unhappy. 549 */ 550 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0); 551 if (rtval == EOPNOTSUPP) { 552 printf("vnode_pager: *** WARNING *** stale FS getpages\n"); 553 rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage); 554 } 555 return rtval; 556 } 557 558 559 /* 560 * This is now called from local media FS's to operate against their 561 * own vnodes if they fail to implement VOP_GETPAGES. 562 */ 563 int 564 vnode_pager_generic_getpages(vp, m, bytecount, reqpage) 565 struct vnode *vp; 566 vm_page_t *m; 567 int bytecount; 568 int reqpage; 569 { 570 vm_object_t object; 571 vm_offset_t kva; 572 off_t foff, tfoff, nextoff; 573 int i, size, bsize, first, firstaddr; 574 struct vnode *dp; 575 int runpg; 576 int runend; 577 struct buf *bp; 578 int s; 579 int count; 580 int error = 0; 581 582 object = vp->v_object; 583 count = bytecount / PAGE_SIZE; 584 585 if (vp->v_mount == NULL) 586 return VM_PAGER_BAD; 587 588 bsize = vp->v_mount->mnt_stat.f_iosize; 589 590 /* get the UNDERLYING device for the file with VOP_BMAP() */ 591 592 /* 593 * originally, we did not check for an error return value -- assuming 594 * an fs always has a bmap entry point -- that assumption is wrong!!! 595 */ 596 foff = IDX_TO_OFF(m[reqpage]->pindex); 597 598 /* 599 * if we can't bmap, use old VOP code 600 */ 601 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) { 602 for (i = 0; i < count; i++) { 603 if (i != reqpage) { 604 vnode_pager_freepage(m[i]); 605 } 606 } 607 cnt.v_vnodein++; 608 cnt.v_vnodepgsin++; 609 return vnode_pager_input_old(object, m[reqpage]); 610 611 /* 612 * if the blocksize is smaller than a page size, then use 613 * special small filesystem code. NFS sometimes has a small 614 * blocksize, but it can handle large reads itself. 615 */ 616 } else if ((PAGE_SIZE / bsize) > 1 && 617 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) { 618 for (i = 0; i < count; i++) { 619 if (i != reqpage) { 620 vnode_pager_freepage(m[i]); 621 } 622 } 623 cnt.v_vnodein++; 624 cnt.v_vnodepgsin++; 625 return vnode_pager_input_smlfs(object, m[reqpage]); 626 } 627 /* 628 * if ANY DEV_BSIZE blocks are valid on a large filesystem block 629 * then, the entire page is valid -- 630 * XXX no it isn't 631 */ 632 633 if (m[reqpage]->valid != VM_PAGE_BITS_ALL) 634 m[reqpage]->valid = 0; 635 636 if (m[reqpage]->valid) { 637 m[reqpage]->valid = VM_PAGE_BITS_ALL; 638 for (i = 0; i < count; i++) { 639 if (i != reqpage) 640 vnode_pager_freepage(m[i]); 641 } 642 return VM_PAGER_OK; 643 } 644 645 /* 646 * here on direct device I/O 647 */ 648 649 firstaddr = -1; 650 /* 651 * calculate the run that includes the required page 652 */ 653 for(first = 0, i = 0; i < count; i = runend) { 654 firstaddr = vnode_pager_addr(vp, 655 IDX_TO_OFF(m[i]->pindex), &runpg); 656 if (firstaddr == -1) { 657 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) { 658 /* XXX no %qd in kernel. */ 659 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx", 660 firstaddr, (u_long)(foff >> 32), 661 (u_long)(u_int32_t)foff, 662 (u_long)(u_int32_t) 663 (object->un_pager.vnp.vnp_size >> 32), 664 (u_long)(u_int32_t) 665 object->un_pager.vnp.vnp_size); 666 } 667 vnode_pager_freepage(m[i]); 668 runend = i + 1; 669 first = runend; 670 continue; 671 } 672 runend = i + runpg; 673 if (runend <= reqpage) { 674 int j; 675 for (j = i; j < runend; j++) { 676 vnode_pager_freepage(m[j]); 677 } 678 } else { 679 if (runpg < (count - first)) { 680 for (i = first + runpg; i < count; i++) 681 vnode_pager_freepage(m[i]); 682 count = first + runpg; 683 } 684 break; 685 } 686 first = runend; 687 } 688 689 /* 690 * the first and last page have been calculated now, move input pages 691 * to be zero based... 692 */ 693 if (first != 0) { 694 for (i = first; i < count; i++) { 695 m[i - first] = m[i]; 696 } 697 count -= first; 698 reqpage -= first; 699 } 700 701 /* 702 * calculate the file virtual address for the transfer 703 */ 704 foff = IDX_TO_OFF(m[0]->pindex); 705 706 /* 707 * calculate the size of the transfer 708 */ 709 size = count * PAGE_SIZE; 710 if ((foff + size) > object->un_pager.vnp.vnp_size) 711 size = object->un_pager.vnp.vnp_size - foff; 712 713 /* 714 * round up physical size for real devices 715 */ 716 if (dp->v_type == VBLK || dp->v_type == VCHR) 717 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 718 719 bp = getpbuf(&vnode_pbuf_freecnt); 720 kva = (vm_offset_t) bp->b_data; 721 722 /* 723 * and map the pages to be read into the kva 724 */ 725 pmap_qenter(kva, m, count); 726 727 /* build a minimal buffer header */ 728 bp->b_flags = B_BUSY | B_READ | B_CALL; 729 bp->b_iodone = vnode_pager_iodone; 730 /* B_PHYS is not set, but it is nice to fill this in */ 731 bp->b_proc = curproc; 732 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred; 733 if (bp->b_rcred != NOCRED) 734 crhold(bp->b_rcred); 735 if (bp->b_wcred != NOCRED) 736 crhold(bp->b_wcred); 737 bp->b_blkno = firstaddr; 738 pbgetvp(dp, bp); 739 bp->b_bcount = size; 740 bp->b_bufsize = size; 741 742 cnt.v_vnodein++; 743 cnt.v_vnodepgsin += count; 744 745 /* do the input */ 746 VOP_STRATEGY(bp->b_vp, bp); 747 748 s = splvm(); 749 /* we definitely need to be at splvm here */ 750 751 while ((bp->b_flags & B_DONE) == 0) { 752 tsleep(bp, PVM, "vnread", 0); 753 } 754 splx(s); 755 if ((bp->b_flags & B_ERROR) != 0) 756 error = EIO; 757 758 if (!error) { 759 if (size != count * PAGE_SIZE) 760 bzero((caddr_t) kva + size, PAGE_SIZE * count - size); 761 } 762 pmap_qremove(kva, count); 763 764 /* 765 * free the buffer header back to the swap buffer pool 766 */ 767 relpbuf(bp, &vnode_pbuf_freecnt); 768 769 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) { 770 vm_page_t mt; 771 772 nextoff = tfoff + PAGE_SIZE; 773 mt = m[i]; 774 775 if (nextoff <= size) { 776 mt->valid = VM_PAGE_BITS_ALL; 777 mt->dirty = 0; 778 pmap_clear_modify(VM_PAGE_TO_PHYS(mt)); 779 } else { 780 int nvalid = ((size + DEV_BSIZE - 1) - tfoff) & ~(DEV_BSIZE - 1); 781 vm_page_set_validclean(mt, 0, nvalid); 782 } 783 784 vm_page_flag_clear(mt, PG_ZERO); 785 if (i != reqpage) { 786 787 /* 788 * whether or not to leave the page activated is up in 789 * the air, but we should put the page on a page queue 790 * somewhere. (it already is in the object). Result: 791 * It appears that emperical results show that 792 * deactivating pages is best. 793 */ 794 795 /* 796 * just in case someone was asking for this page we 797 * now tell them that it is ok to use 798 */ 799 if (!error) { 800 if (mt->flags & PG_WANTED) 801 vm_page_activate(mt); 802 else 803 vm_page_deactivate(mt); 804 vm_page_wakeup(mt); 805 } else { 806 vnode_pager_freepage(mt); 807 } 808 } 809 } 810 if (error) { 811 printf("vnode_pager_getpages: I/O read error\n"); 812 } 813 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 814 } 815 816 /* 817 * EOPNOTSUPP is no longer legal. For local media VFS's that do not 818 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to 819 * vnode_pager_generic_putpages() to implement the previous behaviour. 820 * 821 * All other FS's should use the bypass to get to the local media 822 * backing vp's VOP_PUTPAGES. 823 */ 824 static void 825 vnode_pager_putpages(object, m, count, sync, rtvals) 826 vm_object_t object; 827 vm_page_t *m; 828 int count; 829 boolean_t sync; 830 int *rtvals; 831 { 832 int rtval; 833 struct vnode *vp; 834 int bytes = count * PAGE_SIZE; 835 836 vp = object->handle; 837 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0); 838 if (rtval == EOPNOTSUPP) { 839 printf("vnode_pager: *** WARNING *** stale FS putpages\n"); 840 rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals); 841 } 842 } 843 844 845 /* 846 * This is now called from local media FS's to operate against their 847 * own vnodes if they fail to implement VOP_GETPAGES. 848 */ 849 int 850 vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) 851 struct vnode *vp; 852 vm_page_t *m; 853 int bytecount; 854 int flags; 855 int *rtvals; 856 { 857 int i; 858 vm_object_t object; 859 int count; 860 861 int maxsize, ncount; 862 vm_ooffset_t poffset; 863 struct uio auio; 864 struct iovec aiov; 865 int error; 866 int ioflags; 867 868 object = vp->v_object; 869 count = bytecount / PAGE_SIZE; 870 871 for (i = 0; i < count; i++) 872 rtvals[i] = VM_PAGER_AGAIN; 873 874 if ((int) m[0]->pindex < 0) { 875 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n", 876 m[0]->pindex, m[0]->dirty); 877 rtvals[0] = VM_PAGER_BAD; 878 return VM_PAGER_BAD; 879 } 880 881 maxsize = count * PAGE_SIZE; 882 ncount = count; 883 884 poffset = IDX_TO_OFF(m[0]->pindex); 885 if (maxsize + poffset > object->un_pager.vnp.vnp_size) { 886 if (object->un_pager.vnp.vnp_size > poffset) 887 maxsize = object->un_pager.vnp.vnp_size - poffset; 888 else 889 maxsize = 0; 890 ncount = btoc(maxsize); 891 if (ncount < count) { 892 for (i = ncount; i < count; i++) { 893 rtvals[i] = VM_PAGER_BAD; 894 } 895 } 896 } 897 898 ioflags = IO_VMIO; 899 ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: 0; 900 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; 901 902 aiov.iov_base = (caddr_t) 0; 903 aiov.iov_len = maxsize; 904 auio.uio_iov = &aiov; 905 auio.uio_iovcnt = 1; 906 auio.uio_offset = poffset; 907 auio.uio_segflg = UIO_NOCOPY; 908 auio.uio_rw = UIO_WRITE; 909 auio.uio_resid = maxsize; 910 auio.uio_procp = (struct proc *) 0; 911 error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred); 912 cnt.v_vnodeout++; 913 cnt.v_vnodepgsout += ncount; 914 915 if (error) { 916 printf("vnode_pager_putpages: I/O error %d\n", error); 917 } 918 if (auio.uio_resid) { 919 printf("vnode_pager_putpages: residual I/O %d at %lu\n", 920 auio.uio_resid, (u_long)m[0]->pindex); 921 } 922 for (i = 0; i < ncount; i++) { 923 rtvals[i] = VM_PAGER_OK; 924 } 925 return rtvals[0]; 926 } 927 928 struct vnode * 929 vnode_pager_lock(object) 930 vm_object_t object; 931 { 932 struct proc *p = curproc; /* XXX */ 933 934 for (; object != NULL; object = object->backing_object) { 935 if (object->type != OBJT_VNODE) 936 continue; 937 if (object->flags & OBJ_DEAD) 938 return NULL; 939 940 while (vget(object->handle, 941 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) { 942 if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE)) 943 return NULL; 944 printf("vnode_pager_lock: retrying\n"); 945 } 946 return object->handle; 947 } 948 return NULL; 949 } 950