vfs_cluster.c revision 33108
1/*- 2 * Copyright (c) 1993 3 * The Regents of the University of California. All rights reserved. 4 * Modifications/enhancements: 5 * Copyright (c) 1995 John S. Dyson. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)vfs_cluster.c 8.7 (Berkeley) 2/13/94 36 * $Id: vfs_cluster.c,v 1.53 1998/01/31 11:56:01 dyson Exp $ 37 */ 38 39#include "opt_debug_cluster.h" 40#include "opt_diagnostic.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/proc.h> 45#include <sys/buf.h> 46#include <sys/vnode.h> 47#include <sys/mount.h> 48#include <sys/resourcevar.h> 49#include <vm/vm.h> 50#include <vm/vm_prot.h> 51#include <vm/vm_object.h> 52#include <vm/vm_page.h> 53 54#if defined(CLUSTERDEBUG) 55#include <sys/sysctl.h> 56#include <sys/kernel.h> 57static int rcluster= 0; 58SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0, ""); 59#endif 60 61#ifdef notyet_block_reallocation_enabled 62static struct cluster_save * 63 cluster_collectbufs __P((struct vnode *vp, struct buf *last_bp)); 64#endif 65static struct buf * 66 cluster_rbuild __P((struct vnode *vp, u_quad_t filesize, daddr_t lbn, 67 daddr_t blkno, long size, int run, struct buf *fbp)); 68 69extern vm_page_t bogus_page; 70 71/* 72 * Maximum number of blocks for read-ahead. 73 */ 74#define MAXRA 32 75 76/* 77 * This replaces bread. 78 */ 79int 80cluster_read(vp, filesize, lblkno, size, cred, totread, seqcount, bpp) 81 struct vnode *vp; 82 u_quad_t filesize; 83 daddr_t lblkno; 84 long size; 85 struct ucred *cred; 86 long totread; 87 int seqcount; 88 struct buf **bpp; 89{ 90 struct buf *bp, *rbp, *reqbp; 91 daddr_t blkno, origblkno; 92 int error, num_ra; 93 int i; 94 int maxra, racluster; 95 long origtotread; 96 97 error = 0; 98 if (vp->v_maxio == 0) 99 vp->v_maxio = DFLTPHYS; 100 101 /* 102 * Try to limit the amount of read-ahead by a few 103 * ad-hoc parameters. This needs work!!! 104 */ 105 racluster = vp->v_maxio/size; 106 maxra = 2 * racluster + (totread / size); 107 if (maxra > MAXRA) 108 maxra = MAXRA; 109 if (maxra > nbuf/8) 110 maxra = nbuf/8; 111 112 /* 113 * get the requested block 114 */ 115 *bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0); 116 origblkno = lblkno; 117 origtotread = totread; 118 119 /* 120 * if it is in the cache, then check to see if the reads have been 121 * sequential. If they have, then try some read-ahead, otherwise 122 * back-off on prospective read-aheads. 123 */ 124 if (bp->b_flags & B_CACHE) { 125 if (!seqcount) { 126 return 0; 127 } else if ((bp->b_flags & B_RAM) == 0) { 128 return 0; 129 } else { 130 int s; 131 struct buf *tbp; 132 bp->b_flags &= ~B_RAM; 133 /* 134 * We do the spl here so that there is no window 135 * between the incore and the b_usecount increment 136 * below. We opt to keep the spl out of the loop 137 * for efficiency. 138 */ 139 s = splbio(); 140 for(i=1;i<maxra;i++) { 141 142 if (!(tbp = incore(vp, lblkno+i))) { 143 break; 144 } 145 146 /* 147 * Set another read-ahead mark so we know to check 148 * again. 149 */ 150 if (((i % racluster) == (racluster - 1)) || 151 (i == (maxra - 1))) 152 tbp->b_flags |= B_RAM; 153 154#if 0 155 if (tbp->b_usecount == 0) { 156 /* 157 * Make sure that the soon-to-be used readaheads 158 * are still there. The getblk/bqrelse pair will 159 * boost the priority of the buffer. 160 */ 161 tbp = getblk(vp, lblkno+i, size, 0, 0); 162 bqrelse(tbp); 163 } 164#endif 165 } 166 splx(s); 167 if (i >= maxra) { 168 return 0; 169 } 170 lblkno += i; 171 } 172 reqbp = bp = NULL; 173 } else { 174 u_quad_t firstread; 175 firstread = (u_quad_t) lblkno * size; 176 if (firstread + totread > filesize) 177 totread = filesize - firstread; 178 if (totread > size) { 179 int nblks = 0; 180 int ncontigafter; 181 while (totread > 0) { 182 nblks++; 183 totread -= size; 184 } 185 if (nblks == 1) 186 goto single_block_read; 187 if (nblks > racluster) 188 nblks = racluster; 189 190 error = VOP_BMAP(vp, lblkno, NULL, 191 &blkno, &ncontigafter, NULL); 192 if (error) 193 goto single_block_read; 194 if (blkno == -1) 195 goto single_block_read; 196 if (ncontigafter == 0) 197 goto single_block_read; 198 if (ncontigafter + 1 < nblks) 199 nblks = ncontigafter + 1; 200 201 bp = cluster_rbuild(vp, filesize, lblkno, 202 blkno, size, nblks, bp); 203 lblkno += nblks; 204 } else { 205single_block_read: 206 /* 207 * if it isn't in the cache, then get a chunk from 208 * disk if sequential, otherwise just get the block. 209 */ 210 bp->b_flags |= B_READ | B_RAM; 211 lblkno += 1; 212 } 213 } 214 215 /* 216 * if we have been doing sequential I/O, then do some read-ahead 217 */ 218 rbp = NULL; 219 /* if (seqcount && (lblkno < (origblkno + maxra))) { */ 220 if (seqcount && (lblkno < (origblkno + seqcount))) { 221 /* 222 * we now build the read-ahead buffer if it is desirable. 223 */ 224 if (((u_quad_t)(lblkno + 1) * size) <= filesize && 225 !(error = VOP_BMAP(vp, lblkno, NULL, &blkno, &num_ra, NULL)) && 226 blkno != -1) { 227 int nblksread; 228 int ntoread = num_ra + 1; 229 nblksread = (origtotread + size - 1) / size; 230 if (seqcount < nblksread) 231 seqcount = nblksread; 232 if (seqcount < ntoread) 233 ntoread = seqcount; 234 if (num_ra) { 235 rbp = cluster_rbuild(vp, filesize, lblkno, 236 blkno, size, ntoread, NULL); 237 } else { 238 rbp = getblk(vp, lblkno, size, 0, 0); 239 rbp->b_flags |= B_READ | B_ASYNC | B_RAM; 240 rbp->b_blkno = blkno; 241 } 242 } 243 } 244 245 /* 246 * handle the synchronous read 247 */ 248 if (bp) { 249 if (bp->b_flags & (B_DONE | B_DELWRI)) { 250 panic("cluster_read: DONE bp"); 251 } else { 252#if defined(CLUSTERDEBUG) 253 if (rcluster) 254 printf("S(%d,%d,%d) ", 255 bp->b_lblkno, bp->b_bcount, seqcount); 256#endif 257 if ((bp->b_flags & B_CLUSTER) == 0) 258 vfs_busy_pages(bp, 0); 259 error = VOP_STRATEGY(bp); 260 curproc->p_stats->p_ru.ru_inblock++; 261 } 262 } 263 /* 264 * and if we have read-aheads, do them too 265 */ 266 if (rbp) { 267 if (error) { 268 rbp->b_flags &= ~(B_ASYNC | B_READ); 269 brelse(rbp); 270 } else if (rbp->b_flags & B_CACHE) { 271 rbp->b_flags &= ~(B_ASYNC | B_READ); 272 bqrelse(rbp); 273 } else { 274#if defined(CLUSTERDEBUG) 275 if (rcluster) { 276 if (bp) 277 printf("A+(%d,%d,%d,%d) ", 278 rbp->b_lblkno, rbp->b_bcount, 279 rbp->b_lblkno - origblkno, 280 seqcount); 281 else 282 printf("A(%d,%d,%d,%d) ", 283 rbp->b_lblkno, rbp->b_bcount, 284 rbp->b_lblkno - origblkno, 285 seqcount); 286 } 287#endif 288 289 if ((rbp->b_flags & B_CLUSTER) == 0) 290 vfs_busy_pages(rbp, 0); 291 (void) VOP_STRATEGY(rbp); 292 curproc->p_stats->p_ru.ru_inblock++; 293 } 294 } 295 if (reqbp) 296 return (biowait(reqbp)); 297 else 298 return (error); 299} 300 301/* 302 * If blocks are contiguous on disk, use this to provide clustered 303 * read ahead. We will read as many blocks as possible sequentially 304 * and then parcel them up into logical blocks in the buffer hash table. 305 */ 306static struct buf * 307cluster_rbuild(vp, filesize, lbn, blkno, size, run, fbp) 308 struct vnode *vp; 309 u_quad_t filesize; 310 daddr_t lbn; 311 daddr_t blkno; 312 long size; 313 int run; 314 struct buf *fbp; 315{ 316 struct buf *bp, *tbp; 317 daddr_t bn; 318 int i, inc, j; 319 320#ifdef DIAGNOSTIC 321 if (size != vp->v_mount->mnt_stat.f_iosize) 322 panic("cluster_rbuild: size %d != filesize %d\n", 323 size, vp->v_mount->mnt_stat.f_iosize); 324#endif 325 /* 326 * avoid a division 327 */ 328 while ((u_quad_t) size * (lbn + run) > filesize) { 329 --run; 330 } 331 332 if (fbp) { 333 tbp = fbp; 334 tbp->b_flags |= B_READ; 335 } else { 336 tbp = getblk(vp, lbn, size, 0, 0); 337 if (tbp->b_flags & B_CACHE) 338 return tbp; 339 tbp->b_flags |= B_ASYNC | B_READ | B_RAM; 340 } 341 342 tbp->b_blkno = blkno; 343 if( (tbp->b_flags & B_MALLOC) || 344 ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) 345 return tbp; 346 347 bp = trypbuf(); 348 if (bp == 0) 349 return tbp; 350 351 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 352 bp->b_flags = B_ASYNC | B_READ | B_CALL | B_BUSY | B_CLUSTER | B_VMIO; 353 bp->b_iodone = cluster_callback; 354 bp->b_blkno = blkno; 355 bp->b_lblkno = lbn; 356 pbgetvp(vp, bp); 357 358 TAILQ_INIT(&bp->b_cluster.cluster_head); 359 360 bp->b_bcount = 0; 361 bp->b_bufsize = 0; 362 bp->b_npages = 0; 363 364 if (vp->v_maxio == 0) 365 vp->v_maxio = DFLTPHYS; 366 inc = btodb(size); 367 for (bn = blkno, i = 0; i < run; ++i, bn += inc) { 368 if (i != 0) { 369 if ((bp->b_npages * PAGE_SIZE) + 370 round_page(size) > vp->v_maxio) 371 break; 372 373 if (incore(vp, lbn + i)) 374 break; 375 376 tbp = getblk(vp, lbn + i, size, 0, 0); 377 378 if ((tbp->b_flags & B_CACHE) || 379 (tbp->b_flags & B_VMIO) == 0) { 380 bqrelse(tbp); 381 break; 382 } 383 384 for (j=0;j<tbp->b_npages;j++) { 385 if (tbp->b_pages[j]->valid) { 386 break; 387 } 388 } 389 390 if (j != tbp->b_npages) { 391 /* 392 * force buffer to be re-constituted later 393 */ 394 tbp->b_flags |= B_RELBUF; 395 brelse(tbp); 396 break; 397 } 398 399 if ((fbp && (i == 1)) || (i == (run - 1))) 400 tbp->b_flags |= B_RAM; 401 tbp->b_flags |= B_READ | B_ASYNC; 402 if (tbp->b_blkno == tbp->b_lblkno) { 403 tbp->b_blkno = bn; 404 } else if (tbp->b_blkno != bn) { 405 brelse(tbp); 406 break; 407 } 408 } 409 TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, 410 tbp, b_cluster.cluster_entry); 411 for (j = 0; j < tbp->b_npages; j += 1) { 412 vm_page_t m; 413 m = tbp->b_pages[j]; 414 ++m->busy; 415 ++m->object->paging_in_progress; 416 if ((bp->b_npages == 0) || 417 (bp->b_pages[bp->b_npages-1] != m)) { 418 bp->b_pages[bp->b_npages] = m; 419 bp->b_npages++; 420 } 421 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) 422 tbp->b_pages[j] = bogus_page; 423 } 424 bp->b_bcount += tbp->b_bcount; 425 bp->b_bufsize += tbp->b_bufsize; 426 } 427 428 for(j=0;j<bp->b_npages;j++) { 429 if ((bp->b_pages[j]->valid & VM_PAGE_BITS_ALL) == 430 VM_PAGE_BITS_ALL) 431 bp->b_pages[j] = bogus_page; 432 } 433 if (bp->b_bufsize > bp->b_kvasize) 434 panic("cluster_rbuild: b_bufsize(%d) > b_kvasize(%d)\n", 435 bp->b_bufsize, bp->b_kvasize); 436 bp->b_kvasize = bp->b_bufsize; 437 438 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 439 (vm_page_t *)bp->b_pages, bp->b_npages); 440 return (bp); 441} 442 443/* 444 * Cleanup after a clustered read or write. 445 * This is complicated by the fact that any of the buffers might have 446 * extra memory (if there were no empty buffer headers at allocbuf time) 447 * that we will need to shift around. 448 */ 449void 450cluster_callback(bp) 451 struct buf *bp; 452{ 453 struct buf *nbp, *tbp; 454 int error = 0; 455 456 /* 457 * Must propogate errors to all the components. 458 */ 459 if (bp->b_flags & B_ERROR) 460 error = bp->b_error; 461 462 pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); 463 /* 464 * Move memory from the large cluster buffer into the component 465 * buffers and mark IO as done on these. 466 */ 467 for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head); 468 tbp; tbp = nbp) { 469 nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry); 470 if (error) { 471 tbp->b_flags |= B_ERROR; 472 tbp->b_error = error; 473 } else 474 tbp->b_dirtyoff = tbp->b_dirtyend = 0; 475 biodone(tbp); 476 } 477 relpbuf(bp); 478} 479 480/* 481 * Do clustered write for FFS. 482 * 483 * Three cases: 484 * 1. Write is not sequential (write asynchronously) 485 * Write is sequential: 486 * 2. beginning of cluster - begin cluster 487 * 3. middle of a cluster - add to cluster 488 * 4. end of a cluster - asynchronously write cluster 489 */ 490void 491cluster_write(bp, filesize) 492 struct buf *bp; 493 u_quad_t filesize; 494{ 495 struct vnode *vp; 496 daddr_t lbn; 497 int maxclen, cursize; 498 int lblocksize; 499 int async; 500 501 vp = bp->b_vp; 502 if (vp->v_maxio == 0) 503 vp->v_maxio = DFLTPHYS; 504 if (vp->v_type == VREG) { 505 async = vp->v_mount->mnt_flag & MNT_ASYNC; 506 lblocksize = vp->v_mount->mnt_stat.f_iosize; 507 } else { 508 async = 0; 509 lblocksize = bp->b_bufsize; 510 } 511 lbn = bp->b_lblkno; 512 513 /* Initialize vnode to beginning of file. */ 514 if (lbn == 0) 515 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; 516 517 if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || 518 (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { 519 maxclen = vp->v_maxio / lblocksize - 1; 520 if (vp->v_clen != 0) { 521 /* 522 * Next block is not sequential. 523 * 524 * If we are not writing at end of file, the process 525 * seeked to another point in the file since its last 526 * write, or we have reached our maximum cluster size, 527 * then push the previous cluster. Otherwise try 528 * reallocating to make it sequential. 529 */ 530 cursize = vp->v_lastw - vp->v_cstart + 1; 531#ifndef notyet_block_reallocation_enabled 532 if (((u_quad_t)(lbn + 1) * lblocksize) != filesize || 533 lbn != vp->v_lastw + 1 || 534 vp->v_clen <= cursize) { 535 if (!async) 536 cluster_wbuild(vp, lblocksize, 537 vp->v_cstart, cursize); 538 } 539#else 540 if ((lbn + 1) * lblocksize != filesize || 541 lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { 542 if (!async) 543 cluster_wbuild(vp, lblocksize, 544 vp->v_cstart, cursize); 545 } else { 546 struct buf **bpp, **endbp; 547 struct cluster_save *buflist; 548 549 buflist = cluster_collectbufs(vp, bp); 550 endbp = &buflist->bs_children 551 [buflist->bs_nchildren - 1]; 552 if (VOP_REALLOCBLKS(vp, buflist)) { 553 /* 554 * Failed, push the previous cluster. 555 */ 556 for (bpp = buflist->bs_children; 557 bpp < endbp; bpp++) 558 brelse(*bpp); 559 free(buflist, M_SEGMENT); 560 cluster_wbuild(vp, lblocksize, 561 vp->v_cstart, cursize); 562 } else { 563 /* 564 * Succeeded, keep building cluster. 565 */ 566 for (bpp = buflist->bs_children; 567 bpp <= endbp; bpp++) 568 bdwrite(*bpp); 569 free(buflist, M_SEGMENT); 570 vp->v_lastw = lbn; 571 vp->v_lasta = bp->b_blkno; 572 return; 573 } 574 } 575#endif /* notyet_block_reallocation_enabled */ 576 } 577 /* 578 * Consider beginning a cluster. If at end of file, make 579 * cluster as large as possible, otherwise find size of 580 * existing cluster. 581 */ 582 if ((vp->v_type == VREG) && 583 ((u_quad_t) (lbn + 1) * lblocksize) != filesize && 584 (bp->b_blkno == bp->b_lblkno) && 585 (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || 586 bp->b_blkno == -1)) { 587 bawrite(bp); 588 vp->v_clen = 0; 589 vp->v_lasta = bp->b_blkno; 590 vp->v_cstart = lbn + 1; 591 vp->v_lastw = lbn; 592 return; 593 } 594 vp->v_clen = maxclen; 595 if (!async && maxclen == 0) { /* I/O not contiguous */ 596 vp->v_cstart = lbn + 1; 597 bawrite(bp); 598 } else { /* Wait for rest of cluster */ 599 vp->v_cstart = lbn; 600 bdwrite(bp); 601 } 602 } else if (lbn == vp->v_cstart + vp->v_clen) { 603 /* 604 * At end of cluster, write it out. 605 */ 606 bdwrite(bp); 607 cluster_wbuild(vp, lblocksize, vp->v_cstart, vp->v_clen + 1); 608 vp->v_clen = 0; 609 vp->v_cstart = lbn + 1; 610 } else 611 /* 612 * In the middle of a cluster, so just delay the I/O for now. 613 */ 614 bdwrite(bp); 615 vp->v_lastw = lbn; 616 vp->v_lasta = bp->b_blkno; 617} 618 619 620/* 621 * This is an awful lot like cluster_rbuild...wish they could be combined. 622 * The last lbn argument is the current block on which I/O is being 623 * performed. Check to see that it doesn't fall in the middle of 624 * the current block (if last_bp == NULL). 625 */ 626int 627cluster_wbuild(vp, size, start_lbn, len) 628 struct vnode *vp; 629 long size; 630 daddr_t start_lbn; 631 int len; 632{ 633 struct buf *bp, *tbp; 634 int i, j, s; 635 int totalwritten = 0; 636 int dbsize = btodb(size); 637 while (len > 0) { 638 s = splbio(); 639 if (((tbp = gbincore(vp, start_lbn)) == NULL) || 640 ((tbp->b_flags & (B_INVAL|B_BUSY|B_DELWRI)) != B_DELWRI)) { 641 ++start_lbn; 642 --len; 643 splx(s); 644 continue; 645 } 646 bremfree(tbp); 647 tbp->b_flags |= B_BUSY; 648 tbp->b_flags &= ~B_DONE; 649 splx(s); 650 651 /* 652 * Extra memory in the buffer, punt on this buffer. XXX we could 653 * handle this in most cases, but we would have to push the extra 654 * memory down to after our max possible cluster size and then 655 * potentially pull it back up if the cluster was terminated 656 * prematurely--too much hassle. 657 */ 658 if (((tbp->b_flags & (B_CLUSTEROK|B_MALLOC)) != B_CLUSTEROK) || 659 (tbp->b_bcount != tbp->b_bufsize) || 660 (tbp->b_bcount != size) || 661 len == 1) { 662 totalwritten += tbp->b_bufsize; 663 bawrite(tbp); 664 ++start_lbn; 665 --len; 666 continue; 667 } 668 669 bp = trypbuf(); 670 if (bp == NULL) { 671 totalwritten += tbp->b_bufsize; 672 bawrite(tbp); 673 ++start_lbn; 674 --len; 675 continue; 676 } 677 678 TAILQ_INIT(&bp->b_cluster.cluster_head); 679 bp->b_bcount = 0; 680 bp->b_bufsize = 0; 681 bp->b_npages = 0; 682 if (tbp->b_wcred != NOCRED) { 683 bp->b_wcred = tbp->b_wcred; 684 crhold(bp->b_wcred); 685 } 686 687 bp->b_blkno = tbp->b_blkno; 688 bp->b_lblkno = tbp->b_lblkno; 689 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 690 bp->b_flags |= B_CALL | B_BUSY | B_CLUSTER | 691 (tbp->b_flags & (B_VMIO|B_NEEDCOMMIT)); 692 bp->b_iodone = cluster_callback; 693 pbgetvp(vp, bp); 694 695 for (i = 0; i < len; ++i, ++start_lbn) { 696 if (i != 0) { 697 s = splbio(); 698 if ((tbp = gbincore(vp, start_lbn)) == NULL) { 699 splx(s); 700 break; 701 } 702 703 if ((tbp->b_flags & (B_VMIO|B_CLUSTEROK|B_INVAL|B_BUSY|B_DELWRI|B_NEEDCOMMIT)) != (B_DELWRI|B_CLUSTEROK|(bp->b_flags & (B_VMIO|B_NEEDCOMMIT)))) { 704 splx(s); 705 break; 706 } 707 708 if (tbp->b_wcred != bp->b_wcred) { 709 splx(s); 710 break; 711 } 712 713 if ((tbp->b_bcount != size) || 714 ((bp->b_blkno + dbsize * i) != tbp->b_blkno) || 715 ((tbp->b_npages + bp->b_npages) > (vp->v_maxio / PAGE_SIZE))) { 716 splx(s); 717 break; 718 } 719 bremfree(tbp); 720 tbp->b_flags |= B_BUSY; 721 tbp->b_flags &= ~B_DONE; 722 splx(s); 723 } 724 725 if (tbp->b_flags & B_VMIO) { 726 vm_page_t m; 727 728 if (i != 0) { 729 for (j = 0; j < tbp->b_npages; j += 1) { 730 m = tbp->b_pages[j]; 731 if (m->flags & PG_BUSY) 732 goto finishcluster; 733 } 734 } 735 736 for (j = 0; j < tbp->b_npages; j += 1) { 737 m = tbp->b_pages[j]; 738 ++m->busy; 739 ++m->object->paging_in_progress; 740 if ((bp->b_npages == 0) || 741 (bp->b_pages[bp->b_npages - 1] != m)) { 742 bp->b_pages[bp->b_npages] = m; 743 bp->b_npages++; 744 } 745 } 746 } 747 bp->b_bcount += size; 748 bp->b_bufsize += size; 749 750 --numdirtybuffers; 751 tbp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI); 752 tbp->b_flags |= B_ASYNC; 753 s = splbio(); 754 reassignbuf(tbp, tbp->b_vp); /* put on clean list */ 755 ++tbp->b_vp->v_numoutput; 756 splx(s); 757 TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, 758 tbp, b_cluster.cluster_entry); 759 } 760 finishcluster: 761 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 762 (vm_page_t *) bp->b_pages, bp->b_npages); 763 if (bp->b_bufsize > bp->b_kvasize) 764 panic("cluster_wbuild: b_bufsize(%d) > b_kvasize(%d)\n", 765 bp->b_bufsize, bp->b_kvasize); 766 bp->b_kvasize = bp->b_bufsize; 767 totalwritten += bp->b_bufsize; 768 bp->b_dirtyoff = 0; 769 bp->b_dirtyend = bp->b_bufsize; 770 bawrite(bp); 771 772 len -= i; 773 } 774 return totalwritten; 775} 776 777#ifdef notyet_block_reallocation_enabled 778/* 779 * Collect together all the buffers in a cluster. 780 * Plus add one additional buffer. 781 */ 782static struct cluster_save * 783cluster_collectbufs(vp, last_bp) 784 struct vnode *vp; 785 struct buf *last_bp; 786{ 787 struct cluster_save *buflist; 788 daddr_t lbn; 789 int i, len; 790 791 len = vp->v_lastw - vp->v_cstart + 1; 792 buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), 793 M_SEGMENT, M_WAITOK); 794 buflist->bs_nchildren = 0; 795 buflist->bs_children = (struct buf **) (buflist + 1); 796 for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) 797 (void) bread(vp, lbn, last_bp->b_bcount, NOCRED, 798 &buflist->bs_children[i]); 799 buflist->bs_children[i] = last_bp; 800 buflist->bs_nchildren = i + 1; 801 return (buflist); 802} 803#endif /* notyet_block_reallocation_enabled */ 804