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 1.22 1995/10/09 03:19:49 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> 56 int doreallocblks = 0; 57 struct ctldebug debug13 = {"doreallocblks", &doreallocblks}; 58 59 #else 60 /* XXX for cluster_write */ 61 #define doreallocblks 0 62 #endif 63 64 /* 65 * Local declarations 66 */ 67 static struct buf *cluster_rbuild __P((struct vnode *, u_quad_t, 68 daddr_t, daddr_t, long, int)); 69 struct cluster_save *cluster_collectbufs __P((struct vnode *, struct buf *)); 70 71 int totreads; 72 int totreadblocks; 73 extern 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 */ 114 int 115 cluster_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 int error, num_ra, alreadyincore; 126 int i; 127 int seq; 128 129 error = 0; 130 /* 131 * get the requested block 132 */ 133 origlblkno = lblkno; 134 *bpp = bp = getblk(vp, lblkno, size, 0, 0); 135 seq = ISSEQREAD(vp, lblkno); 136 /* 137 * if it is in the cache, then check to see if the reads have been 138 * sequential. If they have, then try some read-ahead, otherwise 139 * back-off on prospective read-aheads. 140 */ 141 if (bp->b_flags & B_CACHE) { 142 if (!seq) { 143 vp->v_maxra = bp->b_lblkno + bp->b_bcount / size; 144 vp->v_ralen >>= RA_SHIFTDOWN; 145 return 0; 146 } else if( vp->v_maxra > lblkno) { 147 if ( (vp->v_maxra + (vp->v_ralen / RA_MULTIPLE_SLOW)) >= (lblkno + vp->v_ralen)) { 148 if ((vp->v_ralen + 1) < RA_MULTIPLE_FAST*(MAXPHYS / size)) 149 ++vp->v_ralen; 150 return 0; 151 } 152 lblkno = vp->v_maxra; 153 } else { 154 lblkno += 1; 155 } 156 bp = NULL; 157 } else { 158 /* 159 * if it isn't in the cache, then get a chunk from disk if 160 * sequential, otherwise just get the block. 161 */ 162 bp->b_flags |= B_READ; 163 lblkno += 1; 164 curproc->p_stats->p_ru.ru_inblock++; /* XXX */ 165 vp->v_ralen = 0; 166 } 167 /* 168 * assume no read-ahead 169 */ 170 alreadyincore = 1; 171 rablkno = lblkno; 172 173 /* 174 * if we have been doing sequential I/O, then do some read-ahead 175 */ 176 if (seq) { 177 178 /* 179 * bump ralen a bit... 180 */ 181 if ((vp->v_ralen + 1) < RA_MULTIPLE_SLOW*(MAXPHYS / size)) 182 ++vp->v_ralen; 183 /* 184 * this code makes sure that the stuff that we have read-ahead 185 * is still in the cache. If it isn't, we have been reading 186 * ahead too much, and we need to back-off, otherwise we might 187 * try to read more. 188 */ 189 for (i = 0; i < vp->v_ralen; i++) { 190 rablkno = lblkno + i; 191 alreadyincore = (int) incore(vp, rablkno); 192 if (!alreadyincore) { 193 if (inmem(vp, rablkno)) { 194 if (vp->v_maxra < rablkno) 195 vp->v_maxra = rablkno + 1; 196 continue; 197 } 198 if (rablkno < vp->v_maxra) { 199 vp->v_maxra = rablkno; 200 vp->v_ralen >>= RA_SHIFTDOWN; 201 alreadyincore = 1; 202 } 203 break; 204 } else if (vp->v_maxra < rablkno) { 205 vp->v_maxra = rablkno + 1; 206 } 207 } 208 } 209 /* 210 * we now build the read-ahead buffer if it is desirable. 211 */ 212 rbp = NULL; 213 if (!alreadyincore && 214 (rablkno + 1) * size <= filesize && 215 !(error = VOP_BMAP(vp, rablkno, NULL, &blkno, &num_ra, NULL)) && 216 blkno != -1) { 217 if (num_ra > vp->v_ralen) 218 num_ra = vp->v_ralen; 219 220 if (num_ra) { 221 rbp = cluster_rbuild(vp, filesize, rablkno, blkno, size, 222 num_ra + 1); 223 } else { 224 rbp = getblk(vp, rablkno, size, 0, 0); 225 rbp->b_flags |= B_READ | B_ASYNC; 226 rbp->b_blkno = blkno; 227 } 228 } 229 230 /* 231 * handle the synchronous read 232 */ 233 if (bp) { 234 if (bp->b_flags & (B_DONE | B_DELWRI)) 235 panic("cluster_read: DONE bp"); 236 else { 237 vfs_busy_pages(bp, 0); 238 error = VOP_STRATEGY(bp); 239 vp->v_maxra = bp->b_lblkno + bp->b_bcount / size; 240 totreads++; 241 totreadblocks += bp->b_bcount / size; 242 curproc->p_stats->p_ru.ru_inblock++; 243 } 244 } 245 /* 246 * and if we have read-aheads, do them too 247 */ 248 if (rbp) { 249 vp->v_maxra = rbp->b_lblkno + rbp->b_bcount / size; 250 if (error || (rbp->b_flags & B_CACHE)) { 251 rbp->b_flags &= ~(B_ASYNC | B_READ); 252 brelse(rbp); 253 } else { 254 if ((rbp->b_flags & B_CLUSTER) == 0) 255 vfs_busy_pages(rbp, 0); 256 (void) VOP_STRATEGY(rbp); 257 totreads++; 258 totreadblocks += rbp->b_bcount / size; 259 curproc->p_stats->p_ru.ru_inblock++; 260 } 261 } 262 if (bp && ((bp->b_flags & B_ASYNC) == 0)) 263 return (biowait(bp)); 264 return (error); 265 } 266 267 /* 268 * If blocks are contiguous on disk, use this to provide clustered 269 * read ahead. We will read as many blocks as possible sequentially 270 * and then parcel them up into logical blocks in the buffer hash table. 271 */ 272 static struct buf * 273 cluster_rbuild(vp, filesize, lbn, blkno, size, run) 274 struct vnode *vp; 275 u_quad_t filesize; 276 daddr_t lbn; 277 daddr_t blkno; 278 long size; 279 int run; 280 { 281 struct cluster_save *b_save; 282 struct buf *bp, *tbp; 283 daddr_t bn; 284 int i, inc, j; 285 286 #ifdef DIAGNOSTIC 287 if (size != vp->v_mount->mnt_stat.f_iosize) 288 panic("cluster_rbuild: size %d != filesize %d\n", 289 size, vp->v_mount->mnt_stat.f_iosize); 290 #endif 291 if (size * (lbn + run) > filesize) 292 --run; 293 294 tbp = getblk(vp, lbn, size, 0, 0); 295 if (tbp->b_flags & B_CACHE) 296 return tbp; 297 298 tbp->b_blkno = blkno; 299 tbp->b_flags |= B_ASYNC | B_READ; 300 if( ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) 301 return tbp; 302 303 bp = trypbuf(); 304 if (bp == 0) 305 return tbp; 306 307 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 308 bp->b_flags = B_ASYNC | B_READ | B_CALL | B_BUSY | B_CLUSTER | B_VMIO; 309 bp->b_iodone = cluster_callback; 310 bp->b_blkno = blkno; 311 bp->b_lblkno = lbn; 312 pbgetvp(vp, bp); 313 314 b_save = malloc(sizeof(struct buf *) * run + 315 sizeof(struct cluster_save), M_SEGMENT, M_WAITOK); 316 b_save->bs_nchildren = 0; 317 b_save->bs_children = (struct buf **) (b_save + 1); 318 bp->b_saveaddr = b_save; 319 320 bp->b_bcount = 0; 321 bp->b_bufsize = 0; 322 bp->b_npages = 0; 323 324 inc = btodb(size); 325 for (bn = blkno, i = 0; i < run; ++i, bn += inc) { 326 if (i != 0) { 327 if ((bp->b_npages * PAGE_SIZE) + size > MAXPHYS) 328 break; 329 330 if (incore(vp, lbn + i)) 331 break; 332 tbp = getblk(vp, lbn + i, size, 0, 0); 333 334 if ((tbp->b_flags & B_CACHE) || 335 (tbp->b_flags & B_VMIO) == 0) { 336 brelse(tbp); 337 break; 338 } 339 340 for (j=0;j<tbp->b_npages;j++) { 341 if (tbp->b_pages[j]->valid) { 342 break; 343 } 344 } 345 346 if (j != tbp->b_npages) { 347 /* 348 * force buffer to be re-constituted later 349 */ 350 tbp->b_flags |= B_RELBUF; 351 brelse(tbp); 352 break; 353 } 354 355 tbp->b_flags |= B_READ | B_ASYNC; 356 if( tbp->b_blkno == tbp->b_lblkno) { 357 tbp->b_blkno = bn; 358 } else if (tbp->b_blkno != bn) { 359 brelse(tbp); 360 break; 361 } 362 } 363 ++b_save->bs_nchildren; 364 b_save->bs_children[i] = tbp; 365 for (j = 0; j < tbp->b_npages; j += 1) { 366 vm_page_t m; 367 m = tbp->b_pages[j]; 368 ++m->busy; 369 ++m->object->paging_in_progress; 370 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 371 m = bogus_page; 372 } 373 if ((bp->b_npages == 0) || 374 (bp->b_bufsize & PAGE_MASK) == 0) { 375 bp->b_pages[bp->b_npages] = m; 376 bp->b_npages++; 377 } else { 378 if ( tbp->b_npages > 1) { 379 panic("cluster_rbuild: page unaligned filesystems not supported"); 380 } 381 } 382 } 383 bp->b_bcount += tbp->b_bcount; 384 bp->b_bufsize += tbp->b_bufsize; 385 } 386 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 387 (vm_page_t *)bp->b_pages, bp->b_npages); 388 return (bp); 389 } 390 391 /* 392 * Cleanup after a clustered read or write. 393 * This is complicated by the fact that any of the buffers might have 394 * extra memory (if there were no empty buffer headers at allocbuf time) 395 * that we will need to shift around. 396 */ 397 void 398 cluster_callback(bp) 399 struct buf *bp; 400 { 401 struct cluster_save *b_save; 402 struct buf **bpp, *tbp; 403 int error = 0; 404 405 /* 406 * Must propogate errors to all the components. 407 */ 408 if (bp->b_flags & B_ERROR) 409 error = bp->b_error; 410 411 b_save = (struct cluster_save *) (bp->b_saveaddr); 412 pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); 413 /* 414 * Move memory from the large cluster buffer into the component 415 * buffers and mark IO as done on these. 416 */ 417 for (bpp = b_save->bs_children; b_save->bs_nchildren--; ++bpp) { 418 tbp = *bpp; 419 if (error) { 420 tbp->b_flags |= B_ERROR; 421 tbp->b_error = error; 422 } 423 biodone(tbp); 424 } 425 free(b_save, M_SEGMENT); 426 relpbuf(bp); 427 } 428 429 /* 430 * Do clustered write for FFS. 431 * 432 * Three cases: 433 * 1. Write is not sequential (write asynchronously) 434 * Write is sequential: 435 * 2. beginning of cluster - begin cluster 436 * 3. middle of a cluster - add to cluster 437 * 4. end of a cluster - asynchronously write cluster 438 */ 439 void 440 cluster_write(bp, filesize) 441 struct buf *bp; 442 u_quad_t filesize; 443 { 444 struct vnode *vp; 445 daddr_t lbn; 446 int maxclen, cursize; 447 int lblocksize; 448 449 vp = bp->b_vp; 450 lblocksize = vp->v_mount->mnt_stat.f_iosize; 451 lbn = bp->b_lblkno; 452 453 /* Initialize vnode to beginning of file. */ 454 if (lbn == 0) 455 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; 456 457 if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || 458 (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { 459 maxclen = MAXPHYS / lblocksize - 1; 460 if (vp->v_clen != 0) { 461 /* 462 * Next block is not sequential. 463 * 464 * If we are not writing at end of file, the process 465 * seeked to another point in the file since its last 466 * write, or we have reached our maximum cluster size, 467 * then push the previous cluster. Otherwise try 468 * reallocating to make it sequential. 469 */ 470 cursize = vp->v_lastw - vp->v_cstart + 1; 471 if (!doreallocblks || 472 (lbn + 1) * lblocksize != filesize || 473 lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { 474 cluster_wbuild(vp, NULL, lblocksize, 475 vp->v_cstart, cursize, lbn); 476 } else { 477 struct buf **bpp, **endbp; 478 struct cluster_save *buflist; 479 480 buflist = cluster_collectbufs(vp, bp); 481 endbp = &buflist->bs_children 482 [buflist->bs_nchildren - 1]; 483 if (VOP_REALLOCBLKS(vp, buflist)) { 484 /* 485 * Failed, push the previous cluster. 486 */ 487 for (bpp = buflist->bs_children; 488 bpp < endbp; bpp++) 489 brelse(*bpp); 490 free(buflist, M_SEGMENT); 491 cluster_wbuild(vp, NULL, lblocksize, 492 vp->v_cstart, cursize, lbn); 493 } else { 494 /* 495 * Succeeded, keep building cluster. 496 */ 497 for (bpp = buflist->bs_children; 498 bpp <= endbp; bpp++) 499 bdwrite(*bpp); 500 free(buflist, M_SEGMENT); 501 vp->v_lastw = lbn; 502 vp->v_lasta = bp->b_blkno; 503 return; 504 } 505 } 506 } 507 /* 508 * Consider beginning a cluster. If at end of file, make 509 * cluster as large as possible, otherwise find size of 510 * existing cluster. 511 */ 512 if ((lbn + 1) * lblocksize != filesize && 513 (bp->b_blkno == bp->b_lblkno) && 514 (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || 515 bp->b_blkno == -1)) { 516 bawrite(bp); 517 vp->v_clen = 0; 518 vp->v_lasta = bp->b_blkno; 519 vp->v_cstart = lbn + 1; 520 vp->v_lastw = lbn; 521 return; 522 } 523 vp->v_clen = maxclen; 524 if (maxclen == 0) { /* I/O not contiguous */ 525 vp->v_cstart = lbn + 1; 526 bawrite(bp); 527 } else { /* Wait for rest of cluster */ 528 vp->v_cstart = lbn; 529 bdwrite(bp); 530 } 531 } else if (lbn == vp->v_cstart + vp->v_clen) { 532 /* 533 * At end of cluster, write it out. 534 */ 535 cluster_wbuild(vp, bp, bp->b_bcount, vp->v_cstart, 536 vp->v_clen + 1, lbn); 537 vp->v_clen = 0; 538 vp->v_cstart = lbn + 1; 539 } else 540 /* 541 * In the middle of a cluster, so just delay the I/O for now. 542 */ 543 bdwrite(bp); 544 vp->v_lastw = lbn; 545 vp->v_lasta = bp->b_blkno; 546 } 547 548 549 /* 550 * This is an awful lot like cluster_rbuild...wish they could be combined. 551 * The last lbn argument is the current block on which I/O is being 552 * performed. Check to see that it doesn't fall in the middle of 553 * the current block (if last_bp == NULL). 554 */ 555 void 556 cluster_wbuild(vp, last_bp, size, start_lbn, len, lbn) 557 struct vnode *vp; 558 struct buf *last_bp; 559 long size; 560 daddr_t start_lbn; 561 int len; 562 daddr_t lbn; 563 { 564 struct cluster_save *b_save; 565 struct buf *bp, *tbp, *pb; 566 int i, j, s; 567 568 #ifdef DIAGNOSTIC 569 if (size != vp->v_mount->mnt_stat.f_iosize) 570 panic("cluster_wbuild: size %d != filesize %d\n", 571 size, vp->v_mount->mnt_stat.f_iosize); 572 #endif 573 redo: 574 if( (lbn != -1) || (last_bp == 0)) { 575 while ((!(tbp = incore(vp, start_lbn)) || (tbp->b_flags & B_BUSY) 576 || (start_lbn == lbn)) && len) { 577 ++start_lbn; 578 --len; 579 } 580 581 pb = trypbuf(); 582 /* Get more memory for current buffer */ 583 if (len <= 1 || pb == NULL) { 584 if (pb != NULL) 585 relpbuf(pb); 586 if (last_bp) { 587 bawrite(last_bp); 588 } else if (len) { 589 bp = getblk(vp, start_lbn, size, 0, 0); 590 bawrite(bp); 591 } 592 return; 593 } 594 tbp = getblk(vp, start_lbn, size, 0, 0); 595 } else { 596 tbp = last_bp; 597 if( tbp->b_flags & B_BUSY) { 598 printf("vfs_cluster: warning: buffer already busy\n"); 599 } 600 tbp->b_flags |= B_BUSY; 601 last_bp = 0; 602 pb = trypbuf(); 603 if (pb == NULL) { 604 bawrite(tbp); 605 return; 606 } 607 } 608 609 if (!(tbp->b_flags & B_DELWRI)) { 610 relpbuf(pb); 611 ++start_lbn; 612 --len; 613 brelse(tbp); 614 goto redo; 615 } 616 /* 617 * Extra memory in the buffer, punt on this buffer. XXX we could 618 * handle this in most cases, but we would have to push the extra 619 * memory down to after our max possible cluster size and then 620 * potentially pull it back up if the cluster was terminated 621 * prematurely--too much hassle. 622 */ 623 if (((tbp->b_flags & B_VMIO) == 0) || 624 (tbp->b_bcount != tbp->b_bufsize)) { 625 relpbuf(pb); 626 ++start_lbn; 627 --len; 628 bawrite(tbp); 629 goto redo; 630 } 631 bp = pb; 632 b_save = malloc(sizeof(struct buf *) * (len + 1) + sizeof(struct cluster_save), 633 M_SEGMENT, M_WAITOK); 634 b_save->bs_nchildren = 0; 635 b_save->bs_children = (struct buf **) (b_save + 1); 636 bp->b_saveaddr = b_save; 637 bp->b_bcount = 0; 638 bp->b_bufsize = 0; 639 bp->b_npages = 0; 640 641 if (tbp->b_flags & B_VMIO) 642 bp->b_flags |= B_VMIO; 643 644 bp->b_blkno = tbp->b_blkno; 645 bp->b_lblkno = tbp->b_lblkno; 646 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 647 bp->b_flags |= B_CALL | B_BUSY | B_CLUSTER; 648 bp->b_iodone = cluster_callback; 649 pbgetvp(vp, bp); 650 651 for (i = 0; i < len; ++i, ++start_lbn) { 652 if (i != 0) { 653 /* 654 * Block is not in core or the non-sequential block 655 * ending our cluster was part of the cluster (in 656 * which case we don't want to write it twice). 657 */ 658 if (!(tbp = incore(vp, start_lbn)) || 659 (last_bp == NULL && start_lbn == lbn)) 660 break; 661 662 if ((tbp->b_flags & (B_INVAL | B_CLUSTEROK)) != B_CLUSTEROK) 663 break; 664 665 if ((tbp->b_npages + bp->b_npages) > (MAXPHYS / PAGE_SIZE)) 666 break; 667 668 if ( (tbp->b_blkno != tbp->b_lblkno) && 669 ((bp->b_blkno + btodb(size) * i) != tbp->b_blkno)) 670 break; 671 672 /* 673 * Get the desired block buffer (unless it is the 674 * final sequential block whose buffer was passed in 675 * explictly as last_bp). 676 */ 677 if (last_bp == NULL || start_lbn != lbn) { 678 if( tbp->b_flags & B_BUSY) 679 break; 680 tbp = getblk(vp, start_lbn, size, 0, 0); 681 if (!(tbp->b_flags & B_DELWRI) || 682 ((tbp->b_flags & B_VMIO) != (bp->b_flags & B_VMIO))) { 683 brelse(tbp); 684 break; 685 } 686 } else 687 tbp = last_bp; 688 } 689 for (j = 0; j < tbp->b_npages; j += 1) { 690 vm_page_t m; 691 m = tbp->b_pages[j]; 692 ++m->busy; 693 ++m->object->paging_in_progress; 694 if ((bp->b_npages == 0) || 695 (bp->b_pages[bp->b_npages - 1] != m)) { 696 bp->b_pages[bp->b_npages] = m; 697 bp->b_npages++; 698 } 699 } 700 bp->b_bcount += size; 701 bp->b_bufsize += size; 702 703 tbp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI); 704 tbp->b_flags |= B_ASYNC; 705 s = splbio(); 706 reassignbuf(tbp, tbp->b_vp); /* put on clean list */ 707 ++tbp->b_vp->v_numoutput; 708 splx(s); 709 b_save->bs_children[i] = tbp; 710 } 711 b_save->bs_nchildren = i; 712 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 713 (vm_page_t *) bp->b_pages, bp->b_npages); 714 bawrite(bp); 715 716 if (i < len) { 717 len -= i; 718 goto redo; 719 } 720 } 721 722 /* 723 * Collect together all the buffers in a cluster. 724 * Plus add one additional buffer. 725 */ 726 struct cluster_save * 727 cluster_collectbufs(vp, last_bp) 728 struct vnode *vp; 729 struct buf *last_bp; 730 { 731 struct cluster_save *buflist; 732 daddr_t lbn; 733 int i, len; 734 735 len = vp->v_lastw - vp->v_cstart + 1; 736 buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), 737 M_SEGMENT, M_WAITOK); 738 buflist->bs_nchildren = 0; 739 buflist->bs_children = (struct buf **) (buflist + 1); 740 for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) 741 (void) bread(vp, lbn, last_bp->b_bcount, NOCRED, 742 &buflist->bs_children[i]); 743 buflist->bs_children[i] = last_bp; 744 buflist->bs_nchildren = i + 1; 745 return (buflist); 746 } 747