vfs_cluster.c revision 10551
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 1.19 1995/09/03 20:32:52 dyson Exp $ 37 */ 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/proc.h> 42#include <sys/buf.h> 43#include <sys/vnode.h> 44#include <sys/mount.h> 45#include <sys/malloc.h> 46#include <sys/resourcevar.h> 47#include <sys/vmmeter.h> 48#include <miscfs/specfs/specdev.h> 49#include <vm/vm.h> 50#include <vm/vm_object.h> 51#include <vm/vm_page.h> 52 53#ifdef DEBUG 54#include <vm/vm.h> 55#include <sys/sysctl.h> 56int doreallocblks = 0; 57struct ctldebug debug13 = {"doreallocblks", &doreallocblks}; 58 59#else 60/* XXX for cluster_write */ 61#define doreallocblks 0 62#endif 63 64/* 65 * Local declarations 66 */ 67static struct buf *cluster_rbuild __P((struct vnode *, u_quad_t, 68 daddr_t, daddr_t, long, int)); 69struct cluster_save *cluster_collectbufs __P((struct vnode *, struct buf *)); 70 71int totreads; 72int totreadblocks; 73extern vm_page_t bogus_page; 74 75#ifdef DIAGNOSTIC 76/* 77 * Set to 1 if reads of block zero should cause readahead to be done. 78 * Set to 0 treats a read of block zero as a non-sequential read. 79 * 80 * Setting to one assumes that most reads of block zero of files are due to 81 * sequential passes over the files (e.g. cat, sum) where additional blocks 82 * will soon be needed. Setting to zero assumes that the majority are 83 * surgical strikes to get particular info (e.g. size, file) where readahead 84 * blocks will not be used and, in fact, push out other potentially useful 85 * blocks from the cache. The former seems intuitive, but some quick tests 86 * showed that the latter performed better from a system-wide point of view. 87 */ 88 int doclusterraz = 0; 89 90#define ISSEQREAD(vp, blk) \ 91 (((blk) != 0 || doclusterraz) && \ 92 ((blk) == (vp)->v_lastr + 1 || (blk) == (vp)->v_lastr)) 93#else 94#define ISSEQREAD(vp, blk) \ 95 (/* (blk) != 0 && */ ((blk) == (vp)->v_lastr + 1 || (blk) == (vp)->v_lastr)) 96#endif 97 98/* 99 * allow for three entire read-aheads... The system will 100 * adjust downwards rapidly if needed... 101 */ 102#define RA_MULTIPLE_FAST 2 103#define RA_MULTIPLE_SLOW 3 104#define RA_SHIFTDOWN 1 /* approx lg2(RA_MULTIPLE) */ 105/* 106 * This replaces bread. If this is a bread at the beginning of a file and 107 * lastr is 0, we assume this is the first read and we'll read up to two 108 * blocks if they are sequential. After that, we'll do regular read ahead 109 * in clustered chunks. 110 * bp is the block requested. 111 * rbp is the read-ahead block. 112 * If either is NULL, then you don't have to do the I/O. 113 */ 114int 115cluster_read(vp, filesize, lblkno, size, cred, bpp) 116 struct vnode *vp; 117 u_quad_t filesize; 118 daddr_t lblkno; 119 long size; 120 struct ucred *cred; 121 struct buf **bpp; 122{ 123 struct buf *bp, *rbp; 124 daddr_t blkno, rablkno, origlblkno; 125 long flags; 126 int error, num_ra, alreadyincore; 127 int i; 128 int seq; 129 130 error = 0; 131 /* 132 * get the requested block 133 */ 134 origlblkno = lblkno; 135 *bpp = bp = getblk(vp, lblkno, size, 0, 0); 136 seq = ISSEQREAD(vp, lblkno); 137 /* 138 * if it is in the cache, then check to see if the reads have been 139 * sequential. If they have, then try some read-ahead, otherwise 140 * back-off on prospective read-aheads. 141 */ 142 if (bp->b_flags & B_CACHE) { 143 if (!seq) { 144 vp->v_maxra = bp->b_lblkno + bp->b_bcount / size; 145 vp->v_ralen >>= RA_SHIFTDOWN; 146 return 0; 147 } else if( vp->v_maxra > lblkno) { 148 if ( (vp->v_maxra + (vp->v_ralen / RA_MULTIPLE_SLOW)) >= (lblkno + vp->v_ralen)) { 149 if ((vp->v_ralen + 1) < RA_MULTIPLE_FAST*(MAXPHYS / size)) 150 ++vp->v_ralen; 151 return 0; 152 } 153 lblkno = vp->v_maxra; 154 } else { 155 lblkno += 1; 156 } 157 bp = NULL; 158 } else { 159 /* 160 * if it isn't in the cache, then get a chunk from disk if 161 * sequential, otherwise just get the block. 162 */ 163 bp->b_flags |= B_READ; 164 lblkno += 1; 165 curproc->p_stats->p_ru.ru_inblock++; /* XXX */ 166 vp->v_ralen = 0; 167 } 168 /* 169 * assume no read-ahead 170 */ 171 alreadyincore = 1; 172 rablkno = lblkno; 173 174 /* 175 * if we have been doing sequential I/O, then do some read-ahead 176 */ 177 if (seq) { 178 179 /* 180 * bump ralen a bit... 181 */ 182 if ((vp->v_ralen + 1) < RA_MULTIPLE_SLOW*(MAXPHYS / size)) 183 ++vp->v_ralen; 184 /* 185 * this code makes sure that the stuff that we have read-ahead 186 * is still in the cache. If it isn't, we have been reading 187 * ahead too much, and we need to back-off, otherwise we might 188 * try to read more. 189 */ 190 for (i = 0; i < vp->v_ralen; i++) { 191 rablkno = lblkno + i; 192 alreadyincore = (int) incore(vp, rablkno); 193 if (!alreadyincore) { 194 if (inmem(vp, rablkno)) { 195 struct buf *bpt; 196 if (vp->v_maxra < rablkno) 197 vp->v_maxra = rablkno + 1; 198 continue; 199 } 200 if (rablkno < vp->v_maxra) { 201 vp->v_maxra = rablkno; 202 vp->v_ralen >>= RA_SHIFTDOWN; 203 alreadyincore = 1; 204 } 205 break; 206 } else if (vp->v_maxra < rablkno) { 207 vp->v_maxra = rablkno + 1; 208 } 209 } 210 } 211 /* 212 * we now build the read-ahead buffer if it is desirable. 213 */ 214 rbp = NULL; 215 if (!alreadyincore && 216 (rablkno + 1) * size <= filesize && 217 !(error = VOP_BMAP(vp, rablkno, NULL, &blkno, &num_ra, NULL)) && 218 blkno != -1) { 219 if (num_ra > vp->v_ralen) 220 num_ra = vp->v_ralen; 221 222 if (num_ra) { 223 rbp = cluster_rbuild(vp, filesize, rablkno, blkno, size, 224 num_ra + 1); 225 } else { 226 rbp = getblk(vp, rablkno, size, 0, 0); 227 rbp->b_flags |= B_READ | B_ASYNC; 228 rbp->b_blkno = blkno; 229 } 230 } 231 232 /* 233 * handle the synchronous read 234 */ 235 if (bp) { 236 if (bp->b_flags & (B_DONE | B_DELWRI)) 237 panic("cluster_read: DONE bp"); 238 else { 239 vfs_busy_pages(bp, 0); 240 error = VOP_STRATEGY(bp); 241 vp->v_maxra = bp->b_lblkno + bp->b_bcount / size; 242 totreads++; 243 totreadblocks += bp->b_bcount / size; 244 curproc->p_stats->p_ru.ru_inblock++; 245 } 246 } 247 /* 248 * and if we have read-aheads, do them too 249 */ 250 if (rbp) { 251 vp->v_maxra = rbp->b_lblkno + rbp->b_bcount / size; 252 if (error || (rbp->b_flags & B_CACHE)) { 253 rbp->b_flags &= ~(B_ASYNC | B_READ); 254 brelse(rbp); 255 } else { 256 if ((rbp->b_flags & B_CLUSTER) == 0) 257 vfs_busy_pages(rbp, 0); 258 (void) VOP_STRATEGY(rbp); 259 totreads++; 260 totreadblocks += rbp->b_bcount / size; 261 curproc->p_stats->p_ru.ru_inblock++; 262 } 263 } 264 if (bp && ((bp->b_flags & B_ASYNC) == 0)) 265 return (biowait(bp)); 266 return (error); 267} 268 269/* 270 * If blocks are contiguous on disk, use this to provide clustered 271 * read ahead. We will read as many blocks as possible sequentially 272 * and then parcel them up into logical blocks in the buffer hash table. 273 */ 274static struct buf * 275cluster_rbuild(vp, filesize, lbn, blkno, size, run) 276 struct vnode *vp; 277 u_quad_t filesize; 278 daddr_t lbn; 279 daddr_t blkno; 280 long size; 281 int run; 282{ 283 struct cluster_save *b_save; 284 struct buf *bp, *tbp; 285 daddr_t bn; 286 int i, inc, j; 287 288#ifdef DIAGNOSTIC 289 if (size != vp->v_mount->mnt_stat.f_iosize) 290 panic("cluster_rbuild: size %d != filesize %d\n", 291 size, vp->v_mount->mnt_stat.f_iosize); 292#endif 293 if (size * (lbn + run + 1) > filesize) 294 --run; 295 296 tbp = getblk(vp, lbn, size, 0, 0); 297 if (tbp->b_flags & B_CACHE) 298 return tbp; 299 300 tbp->b_blkno = blkno; 301 tbp->b_flags |= B_ASYNC | B_READ; 302 if( ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) 303 return tbp; 304 305 bp = trypbuf(); 306 if (bp == 0) 307 return tbp; 308 309 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 310 bp->b_flags = B_ASYNC | B_READ | B_CALL | B_BUSY | B_CLUSTER | B_VMIO; 311 bp->b_iodone = cluster_callback; 312 bp->b_blkno = blkno; 313 bp->b_lblkno = lbn; 314 pbgetvp(vp, bp); 315 316 b_save = malloc(sizeof(struct buf *) * run + sizeof(struct cluster_save), 317 M_SEGMENT, M_WAITOK); 318 b_save->bs_nchildren = 0; 319 b_save->bs_children = (struct buf **) (b_save + 1); 320 bp->b_saveaddr = b_save; 321 322 bp->b_bcount = 0; 323 bp->b_bufsize = 0; 324 bp->b_npages = 0; 325 326 inc = btodb(size); 327 for (bn = blkno, i = 0; i < run; ++i, bn += inc) { 328 if (i != 0) { 329 if ((bp->b_npages * PAGE_SIZE) + size > MAXPHYS) 330 break; 331 if (incore(vp, lbn + i)) 332 break; 333 tbp = getblk(vp, lbn + i, size, 0, 0); 334 335 if ((tbp->b_flags & B_CACHE) || 336 (tbp->b_flags & B_VMIO) == 0) { 337 brelse(tbp); 338 break; 339 } 340 341 for (j=0;j<tbp->b_npages;j++) { 342 if (tbp->b_pages[j]->valid) { 343 break; 344 } 345 } 346 347 if (j != tbp->b_npages) { 348 brelse(tbp); 349 break; 350 } 351 352 tbp->b_flags |= B_READ | B_ASYNC; 353 if( tbp->b_blkno == tbp->b_lblkno) { 354 tbp->b_blkno = bn; 355 } else if (tbp->b_blkno != bn) { 356 brelse(tbp); 357 break; 358 } 359 } 360 ++b_save->bs_nchildren; 361 b_save->bs_children[i] = tbp; 362 for (j = 0; j < tbp->b_npages; j += 1) { 363 vm_page_t m; 364 m = tbp->b_pages[j]; 365 ++m->busy; 366 ++m->object->paging_in_progress; 367 if (m->valid == VM_PAGE_BITS_ALL) { 368 m = bogus_page; 369 } 370 if ((bp->b_npages == 0) || 371 (bp->b_pages[bp->b_npages - 1] != m)) { 372 bp->b_pages[bp->b_npages] = m; 373 bp->b_npages++; 374 } 375 } 376 bp->b_bcount += tbp->b_bcount; 377 bp->b_bufsize += tbp->b_bufsize; 378 } 379 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 380 (vm_page_t *)bp->b_pages, bp->b_npages); 381 return (bp); 382} 383 384/* 385 * Cleanup after a clustered read or write. 386 * This is complicated by the fact that any of the buffers might have 387 * extra memory (if there were no empty buffer headers at allocbuf time) 388 * that we will need to shift around. 389 */ 390void 391cluster_callback(bp) 392 struct buf *bp; 393{ 394 struct cluster_save *b_save; 395 struct buf **bpp, *tbp; 396 caddr_t cp; 397 int error = 0; 398 399 /* 400 * Must propogate errors to all the components. 401 */ 402 if (bp->b_flags & B_ERROR) 403 error = bp->b_error; 404 405 b_save = (struct cluster_save *) (bp->b_saveaddr); 406 pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); 407 /* 408 * Move memory from the large cluster buffer into the component 409 * buffers and mark IO as done on these. 410 */ 411 for (bpp = b_save->bs_children; b_save->bs_nchildren--; ++bpp) { 412 tbp = *bpp; 413 if (error) { 414 tbp->b_flags |= B_ERROR; 415 tbp->b_error = error; 416 } 417 biodone(tbp); 418 } 419 free(b_save, M_SEGMENT); 420 relpbuf(bp); 421} 422 423/* 424 * Do clustered write for FFS. 425 * 426 * Three cases: 427 * 1. Write is not sequential (write asynchronously) 428 * Write is sequential: 429 * 2. beginning of cluster - begin cluster 430 * 3. middle of a cluster - add to cluster 431 * 4. end of a cluster - asynchronously write cluster 432 */ 433void 434cluster_write(bp, filesize) 435 struct buf *bp; 436 u_quad_t filesize; 437{ 438 struct vnode *vp; 439 daddr_t lbn; 440 int maxclen, cursize; 441 int lblocksize; 442 443 vp = bp->b_vp; 444 lblocksize = vp->v_mount->mnt_stat.f_iosize; 445 lbn = bp->b_lblkno; 446 447 /* Initialize vnode to beginning of file. */ 448 if (lbn == 0) 449 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; 450 451 if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || 452 (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { 453 maxclen = MAXPHYS / lblocksize - 1; 454 if (vp->v_clen != 0) { 455 /* 456 * Next block is not sequential. 457 * 458 * If we are not writing at end of file, the process 459 * seeked to another point in the file since its last 460 * write, or we have reached our maximum cluster size, 461 * then push the previous cluster. Otherwise try 462 * reallocating to make it sequential. 463 */ 464 cursize = vp->v_lastw - vp->v_cstart + 1; 465 if (!doreallocblks || 466 (lbn + 1) * lblocksize != filesize || 467 lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { 468 cluster_wbuild(vp, NULL, lblocksize, 469 vp->v_cstart, cursize, lbn); 470 } else { 471 struct buf **bpp, **endbp; 472 struct cluster_save *buflist; 473 474 buflist = cluster_collectbufs(vp, bp); 475 endbp = &buflist->bs_children 476 [buflist->bs_nchildren - 1]; 477 if (VOP_REALLOCBLKS(vp, buflist)) { 478 /* 479 * Failed, push the previous cluster. 480 */ 481 for (bpp = buflist->bs_children; 482 bpp < endbp; bpp++) 483 brelse(*bpp); 484 free(buflist, M_SEGMENT); 485 cluster_wbuild(vp, NULL, lblocksize, 486 vp->v_cstart, cursize, lbn); 487 } else { 488 /* 489 * Succeeded, keep building cluster. 490 */ 491 for (bpp = buflist->bs_children; 492 bpp <= endbp; bpp++) 493 bdwrite(*bpp); 494 free(buflist, M_SEGMENT); 495 vp->v_lastw = lbn; 496 vp->v_lasta = bp->b_blkno; 497 return; 498 } 499 } 500 } 501 /* 502 * Consider beginning a cluster. If at end of file, make 503 * cluster as large as possible, otherwise find size of 504 * existing cluster. 505 */ 506 if ((lbn + 1) * lblocksize != filesize && 507 (bp->b_blkno == bp->b_lblkno) && 508 (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || 509 bp->b_blkno == -1)) { 510 bawrite(bp); 511 vp->v_clen = 0; 512 vp->v_lasta = bp->b_blkno; 513 vp->v_cstart = lbn + 1; 514 vp->v_lastw = lbn; 515 return; 516 } 517 vp->v_clen = maxclen; 518 if (maxclen == 0) { /* I/O not contiguous */ 519 vp->v_cstart = lbn + 1; 520 bawrite(bp); 521 } else { /* Wait for rest of cluster */ 522 vp->v_cstart = lbn; 523 bdwrite(bp); 524 } 525 } else if (lbn == vp->v_cstart + vp->v_clen) { 526 /* 527 * At end of cluster, write it out. 528 */ 529 cluster_wbuild(vp, bp, bp->b_bcount, vp->v_cstart, 530 vp->v_clen + 1, lbn); 531 vp->v_clen = 0; 532 vp->v_cstart = lbn + 1; 533 } else 534 /* 535 * In the middle of a cluster, so just delay the I/O for now. 536 */ 537 bdwrite(bp); 538 vp->v_lastw = lbn; 539 vp->v_lasta = bp->b_blkno; 540} 541 542 543/* 544 * This is an awful lot like cluster_rbuild...wish they could be combined. 545 * The last lbn argument is the current block on which I/O is being 546 * performed. Check to see that it doesn't fall in the middle of 547 * the current block (if last_bp == NULL). 548 */ 549void 550cluster_wbuild(vp, last_bp, size, start_lbn, len, lbn) 551 struct vnode *vp; 552 struct buf *last_bp; 553 long size; 554 daddr_t start_lbn; 555 int len; 556 daddr_t lbn; 557{ 558 struct cluster_save *b_save; 559 struct buf *bp, *tbp, *pb; 560 caddr_t cp; 561 int i, j, s; 562 563#ifdef DIAGNOSTIC 564 if (size != vp->v_mount->mnt_stat.f_iosize) 565 panic("cluster_wbuild: size %d != filesize %d\n", 566 size, vp->v_mount->mnt_stat.f_iosize); 567#endif 568redo: 569 if( (lbn != -1) || (last_bp == 0)) { 570 while ((!(tbp = incore(vp, start_lbn)) || (tbp->b_flags & B_BUSY) 571 || (start_lbn == lbn)) && len) { 572 ++start_lbn; 573 --len; 574 } 575 576 pb = trypbuf(); 577 /* Get more memory for current buffer */ 578 if (len <= 1 || pb == NULL) { 579 if (pb != NULL) 580 relpbuf(pb); 581 if (last_bp) { 582 bawrite(last_bp); 583 } else if (len) { 584 bp = getblk(vp, start_lbn, size, 0, 0); 585 bawrite(bp); 586 } 587 return; 588 } 589 tbp = getblk(vp, start_lbn, size, 0, 0); 590 } else { 591 tbp = last_bp; 592 if( tbp->b_flags & B_BUSY) { 593 printf("vfs_cluster: warning: buffer already busy\n"); 594 } 595 tbp->b_flags |= B_BUSY; 596 last_bp = 0; 597 pb = trypbuf(); 598 if (pb == NULL) { 599 bawrite(tbp); 600 return; 601 } 602 } 603 604 if (!(tbp->b_flags & B_DELWRI)) { 605 relpbuf(pb); 606 ++start_lbn; 607 --len; 608 brelse(tbp); 609 goto redo; 610 } 611 /* 612 * Extra memory in the buffer, punt on this buffer. XXX we could 613 * handle this in most cases, but we would have to push the extra 614 * memory down to after our max possible cluster size and then 615 * potentially pull it back up if the cluster was terminated 616 * prematurely--too much hassle. 617 */ 618 if (tbp->b_bcount != tbp->b_bufsize) { 619 relpbuf(pb); 620 ++start_lbn; 621 --len; 622 bawrite(tbp); 623 goto redo; 624 } 625 bp = pb; 626 b_save = malloc(sizeof(struct buf *) * (len + 1) + sizeof(struct cluster_save), 627 M_SEGMENT, M_WAITOK); 628 b_save->bs_nchildren = 0; 629 b_save->bs_children = (struct buf **) (b_save + 1); 630 bp->b_saveaddr = b_save; 631 bp->b_bcount = 0; 632 bp->b_bufsize = 0; 633 bp->b_npages = 0; 634 635 if (tbp->b_flags & B_VMIO) 636 bp->b_flags |= B_VMIO; 637 638 bp->b_blkno = tbp->b_blkno; 639 bp->b_lblkno = tbp->b_lblkno; 640 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 641 bp->b_flags |= B_CALL | B_BUSY | B_CLUSTER; 642 bp->b_iodone = cluster_callback; 643 pbgetvp(vp, bp); 644 645 for (i = 0; i < len; ++i, ++start_lbn) { 646 if (i != 0) { 647 /* 648 * Block is not in core or the non-sequential block 649 * ending our cluster was part of the cluster (in 650 * which case we don't want to write it twice). 651 */ 652 if (!(tbp = incore(vp, start_lbn)) || 653 (last_bp == NULL && start_lbn == lbn)) 654 break; 655 656 if ((tbp->b_flags & (B_INVAL | B_CLUSTEROK)) != B_CLUSTEROK) 657 break; 658 659 if ((tbp->b_npages + bp->b_npages) > (MAXPHYS / PAGE_SIZE)) 660 break; 661 662 if ( (tbp->b_blkno != tbp->b_lblkno) && 663 ((bp->b_blkno + btodb(size) * i) != tbp->b_blkno)) 664 break; 665 666 /* 667 * Get the desired block buffer (unless it is the 668 * final sequential block whose buffer was passed in 669 * explictly as last_bp). 670 */ 671 if (last_bp == NULL || start_lbn != lbn) { 672 if( tbp->b_flags & B_BUSY) 673 break; 674 tbp = getblk(vp, start_lbn, size, 0, 0); 675 if (!(tbp->b_flags & B_DELWRI) || 676 ((tbp->b_flags & B_VMIO) != (bp->b_flags & B_VMIO))) { 677 brelse(tbp); 678 break; 679 } 680 } else 681 tbp = last_bp; 682 } 683 for (j = 0; j < tbp->b_npages; j += 1) { 684 vm_page_t m; 685 m = tbp->b_pages[j]; 686 ++m->busy; 687 ++m->object->paging_in_progress; 688 if ((bp->b_npages == 0) || 689 (bp->b_pages[bp->b_npages - 1] != m)) { 690 bp->b_pages[bp->b_npages] = m; 691 bp->b_npages++; 692 } 693 } 694 bp->b_bcount += size; 695 bp->b_bufsize += size; 696 697 tbp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI); 698 tbp->b_flags |= B_ASYNC; 699 s = splbio(); 700 reassignbuf(tbp, tbp->b_vp); /* put on clean list */ 701 ++tbp->b_vp->v_numoutput; 702 splx(s); 703 b_save->bs_children[i] = tbp; 704 } 705 b_save->bs_nchildren = i; 706 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 707 (vm_page_t *) bp->b_pages, bp->b_npages); 708 bawrite(bp); 709 710 if (i < len) { 711 len -= i; 712 goto redo; 713 } 714} 715 716/* 717 * Collect together all the buffers in a cluster. 718 * Plus add one additional buffer. 719 */ 720struct cluster_save * 721cluster_collectbufs(vp, last_bp) 722 struct vnode *vp; 723 struct buf *last_bp; 724{ 725 struct cluster_save *buflist; 726 daddr_t lbn; 727 int i, len; 728 729 len = vp->v_lastw - vp->v_cstart + 1; 730 buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), 731 M_SEGMENT, M_WAITOK); 732 buflist->bs_nchildren = 0; 733 buflist->bs_children = (struct buf **) (buflist + 1); 734 for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) 735 (void) bread(vp, lbn, last_bp->b_bcount, NOCRED, 736 &buflist->bs_children[i]); 737 buflist->bs_children[i] = last_bp; 738 buflist->bs_nchildren = i + 1; 739 return (buflist); 740} 741