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 1.56 1998/03/07 21:35:28 dyson Exp $ 37 */ 38 39 #include "opt_debug_cluster.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/proc.h> 44 #include <sys/buf.h> 45 #include <sys/vnode.h> 46 #include <sys/mount.h> 47 #include <sys/resourcevar.h> 48 #include <vm/vm.h> 49 #include <vm/vm_prot.h> 50 #include <vm/vm_object.h> 51 #include <vm/vm_page.h> 52 53 #if defined(CLUSTERDEBUG) 54 #include <sys/sysctl.h> 55 #include <sys/kernel.h> 56 static int rcluster= 0; 57 SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0, ""); 58 #endif 59 60 #ifdef notyet_block_reallocation_enabled 61 static struct cluster_save * 62 cluster_collectbufs __P((struct vnode *vp, struct buf *last_bp)); 63 #endif 64 static struct buf * 65 cluster_rbuild __P((struct vnode *vp, u_quad_t filesize, daddr_t lbn, 66 daddr_t blkno, long size, int run, struct buf *fbp)); 67 68 extern vm_page_t bogus_page; 69 70 /* 71 * Maximum number of blocks for read-ahead. 72 */ 73 #define MAXRA 32 74 75 /* 76 * This replaces bread. 77 */ 78 int 79 cluster_read(vp, filesize, lblkno, size, cred, totread, seqcount, bpp) 80 struct vnode *vp; 81 u_quad_t filesize; 82 daddr_t lblkno; 83 long size; 84 struct ucred *cred; 85 long totread; 86 int seqcount; 87 struct buf **bpp; 88 { 89 struct buf *bp, *rbp, *reqbp; 90 daddr_t blkno, origblkno; 91 int error, num_ra; 92 int i; 93 int maxra, racluster; 94 long origtotread; 95 96 error = 0; 97 if (vp->v_maxio == 0) 98 vp->v_maxio = DFLTPHYS; 99 100 /* 101 * Try to limit the amount of read-ahead by a few 102 * ad-hoc parameters. This needs work!!! 103 */ 104 racluster = vp->v_maxio/size; 105 maxra = 2 * racluster + (totread / size); 106 if (maxra > MAXRA) 107 maxra = MAXRA; 108 if (maxra > nbuf/8) 109 maxra = nbuf/8; 110 111 /* 112 * get the requested block 113 */ 114 *bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0); 115 origblkno = lblkno; 116 origtotread = totread; 117 118 /* 119 * if it is in the cache, then check to see if the reads have been 120 * sequential. If they have, then try some read-ahead, otherwise 121 * back-off on prospective read-aheads. 122 */ 123 if (bp->b_flags & B_CACHE) { 124 if (!seqcount) { 125 return 0; 126 } else if ((bp->b_flags & B_RAM) == 0) { 127 return 0; 128 } else { 129 int s; 130 struct buf *tbp; 131 bp->b_flags &= ~B_RAM; 132 /* 133 * We do the spl here so that there is no window 134 * between the incore and the b_usecount increment 135 * below. We opt to keep the spl out of the loop 136 * for efficiency. 137 */ 138 s = splbio(); 139 for(i=1;i<maxra;i++) { 140 141 if (!(tbp = incore(vp, lblkno+i))) { 142 break; 143 } 144 145 /* 146 * Set another read-ahead mark so we know to check 147 * again. 148 */ 149 if (((i % racluster) == (racluster - 1)) || 150 (i == (maxra - 1))) 151 tbp->b_flags |= B_RAM; 152 153 if ((tbp->b_usecount < 5) && 154 ((tbp->b_flags & B_BUSY) == 0) && 155 (tbp->b_qindex == QUEUE_LRU)) { 156 TAILQ_REMOVE(&bufqueues[QUEUE_LRU], tbp, b_freelist); 157 TAILQ_INSERT_TAIL(&bufqueues[QUEUE_LRU], tbp, b_freelist); 158 } 159 } 160 splx(s); 161 if (i >= maxra) { 162 return 0; 163 } 164 lblkno += i; 165 } 166 reqbp = bp = NULL; 167 } else { 168 u_quad_t firstread; 169 firstread = (u_quad_t) lblkno * size; 170 if (firstread + totread > filesize) 171 totread = filesize - firstread; 172 if (totread > size) { 173 int nblks = 0; 174 int ncontigafter; 175 while (totread > 0) { 176 nblks++; 177 totread -= size; 178 } 179 if (nblks == 1) 180 goto single_block_read; 181 if (nblks > racluster) 182 nblks = racluster; 183 184 error = VOP_BMAP(vp, lblkno, NULL, 185 &blkno, &ncontigafter, NULL); 186 if (error) 187 goto single_block_read; 188 if (blkno == -1) 189 goto single_block_read; 190 if (ncontigafter == 0) 191 goto single_block_read; 192 if (ncontigafter + 1 < nblks) 193 nblks = ncontigafter + 1; 194 195 bp = cluster_rbuild(vp, filesize, lblkno, 196 blkno, size, nblks, bp); 197 lblkno += nblks; 198 } else { 199 single_block_read: 200 /* 201 * if it isn't in the cache, then get a chunk from 202 * disk if sequential, otherwise just get the block. 203 */ 204 bp->b_flags |= B_READ | B_RAM; 205 lblkno += 1; 206 } 207 } 208 209 /* 210 * if we have been doing sequential I/O, then do some read-ahead 211 */ 212 rbp = NULL; 213 if (seqcount && (lblkno < (origblkno + seqcount))) { 214 /* 215 * we now build the read-ahead buffer if it is desirable. 216 */ 217 if (((u_quad_t)(lblkno + 1) * size) <= filesize && 218 !(error = VOP_BMAP(vp, lblkno, NULL, &blkno, &num_ra, NULL)) && 219 blkno != -1) { 220 int nblksread; 221 int ntoread = num_ra + 1; 222 nblksread = (origtotread + size - 1) / size; 223 if (seqcount < nblksread) 224 seqcount = nblksread; 225 if (seqcount < ntoread) 226 ntoread = seqcount; 227 if (num_ra) { 228 rbp = cluster_rbuild(vp, filesize, lblkno, 229 blkno, size, ntoread, NULL); 230 } else { 231 rbp = getblk(vp, lblkno, size, 0, 0); 232 rbp->b_flags |= B_READ | B_ASYNC | B_RAM; 233 rbp->b_blkno = blkno; 234 } 235 } 236 } 237 238 /* 239 * handle the synchronous read 240 */ 241 if (bp) { 242 if (bp->b_flags & (B_DONE | B_DELWRI)) { 243 panic("cluster_read: DONE bp"); 244 } else { 245 #if defined(CLUSTERDEBUG) 246 if (rcluster) 247 printf("S(%d,%d,%d) ", 248 bp->b_lblkno, bp->b_bcount, seqcount); 249 #endif 250 if ((bp->b_flags & B_CLUSTER) == 0) 251 vfs_busy_pages(bp, 0); 252 error = VOP_STRATEGY(bp); 253 curproc->p_stats->p_ru.ru_inblock++; 254 } 255 } 256 /* 257 * and if we have read-aheads, do them too 258 */ 259 if (rbp) { 260 if (error) { 261 rbp->b_flags &= ~(B_ASYNC | B_READ); 262 brelse(rbp); 263 } else if (rbp->b_flags & B_CACHE) { 264 rbp->b_flags &= ~(B_ASYNC | B_READ); 265 bqrelse(rbp); 266 } else { 267 #if defined(CLUSTERDEBUG) 268 if (rcluster) { 269 if (bp) 270 printf("A+(%d,%d,%d,%d) ", 271 rbp->b_lblkno, rbp->b_bcount, 272 rbp->b_lblkno - origblkno, 273 seqcount); 274 else 275 printf("A(%d,%d,%d,%d) ", 276 rbp->b_lblkno, rbp->b_bcount, 277 rbp->b_lblkno - origblkno, 278 seqcount); 279 } 280 #endif 281 282 if ((rbp->b_flags & B_CLUSTER) == 0) 283 vfs_busy_pages(rbp, 0); 284 (void) VOP_STRATEGY(rbp); 285 curproc->p_stats->p_ru.ru_inblock++; 286 } 287 } 288 if (reqbp) 289 return (biowait(reqbp)); 290 else 291 return (error); 292 } 293 294 /* 295 * If blocks are contiguous on disk, use this to provide clustered 296 * read ahead. We will read as many blocks as possible sequentially 297 * and then parcel them up into logical blocks in the buffer hash table. 298 */ 299 static struct buf * 300 cluster_rbuild(vp, filesize, lbn, blkno, size, run, fbp) 301 struct vnode *vp; 302 u_quad_t filesize; 303 daddr_t lbn; 304 daddr_t blkno; 305 long size; 306 int run; 307 struct buf *fbp; 308 { 309 struct buf *bp, *tbp; 310 daddr_t bn; 311 int i, inc, j; 312 313 #ifdef DIAGNOSTIC 314 if (size != vp->v_mount->mnt_stat.f_iosize) 315 panic("cluster_rbuild: size %d != filesize %d\n", 316 size, vp->v_mount->mnt_stat.f_iosize); 317 #endif 318 /* 319 * avoid a division 320 */ 321 while ((u_quad_t) size * (lbn + run) > filesize) { 322 --run; 323 } 324 325 if (fbp) { 326 tbp = fbp; 327 tbp->b_flags |= B_READ; 328 } else { 329 tbp = getblk(vp, lbn, size, 0, 0); 330 if (tbp->b_flags & B_CACHE) 331 return tbp; 332 tbp->b_flags |= B_ASYNC | B_READ | B_RAM; 333 } 334 335 tbp->b_blkno = blkno; 336 if( (tbp->b_flags & B_MALLOC) || 337 ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) 338 return tbp; 339 340 bp = trypbuf(); 341 if (bp == 0) 342 return tbp; 343 344 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 345 bp->b_flags = B_ASYNC | B_READ | B_CALL | B_BUSY | B_CLUSTER | B_VMIO; 346 bp->b_iodone = cluster_callback; 347 bp->b_blkno = blkno; 348 bp->b_lblkno = lbn; 349 pbgetvp(vp, bp); 350 351 TAILQ_INIT(&bp->b_cluster.cluster_head); 352 353 bp->b_bcount = 0; 354 bp->b_bufsize = 0; 355 bp->b_npages = 0; 356 357 if (vp->v_maxio == 0) 358 vp->v_maxio = DFLTPHYS; 359 inc = btodb(size); 360 for (bn = blkno, i = 0; i < run; ++i, bn += inc) { 361 if (i != 0) { 362 if ((bp->b_npages * PAGE_SIZE) + 363 round_page(size) > vp->v_maxio) 364 break; 365 366 if (incore(vp, lbn + i)) 367 break; 368 369 tbp = getblk(vp, lbn + i, size, 0, 0); 370 371 if ((tbp->b_flags & B_CACHE) || 372 (tbp->b_flags & B_VMIO) == 0) { 373 bqrelse(tbp); 374 break; 375 } 376 377 for (j=0;j<tbp->b_npages;j++) { 378 if (tbp->b_pages[j]->valid) { 379 break; 380 } 381 } 382 383 if (j != tbp->b_npages) { 384 /* 385 * force buffer to be re-constituted later 386 */ 387 tbp->b_flags |= B_RELBUF; 388 brelse(tbp); 389 break; 390 } 391 392 if ((fbp && (i == 1)) || (i == (run - 1))) 393 tbp->b_flags |= B_RAM; 394 tbp->b_flags |= B_READ | B_ASYNC; 395 if (tbp->b_blkno == tbp->b_lblkno) { 396 tbp->b_blkno = bn; 397 } else if (tbp->b_blkno != bn) { 398 brelse(tbp); 399 break; 400 } 401 } 402 /* check for latent dependencies to be handled */ 403 if ((LIST_FIRST(&tbp->b_dep)) != NULL && bioops.io_start) 404 (*bioops.io_start)(tbp); 405 TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, 406 tbp, b_cluster.cluster_entry); 407 for (j = 0; j < tbp->b_npages; j += 1) { 408 vm_page_t m; 409 m = tbp->b_pages[j]; 410 ++m->busy; 411 ++m->object->paging_in_progress; 412 if ((bp->b_npages == 0) || 413 (bp->b_pages[bp->b_npages-1] != m)) { 414 bp->b_pages[bp->b_npages] = m; 415 bp->b_npages++; 416 } 417 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) 418 tbp->b_pages[j] = bogus_page; 419 } 420 bp->b_bcount += tbp->b_bcount; 421 bp->b_bufsize += tbp->b_bufsize; 422 } 423 424 for(j=0;j<bp->b_npages;j++) { 425 if ((bp->b_pages[j]->valid & VM_PAGE_BITS_ALL) == 426 VM_PAGE_BITS_ALL) 427 bp->b_pages[j] = bogus_page; 428 } 429 if (bp->b_bufsize > bp->b_kvasize) 430 panic("cluster_rbuild: b_bufsize(%d) > b_kvasize(%d)\n", 431 bp->b_bufsize, bp->b_kvasize); 432 bp->b_kvasize = bp->b_bufsize; 433 434 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 435 (vm_page_t *)bp->b_pages, bp->b_npages); 436 return (bp); 437 } 438 439 /* 440 * Cleanup after a clustered read or write. 441 * This is complicated by the fact that any of the buffers might have 442 * extra memory (if there were no empty buffer headers at allocbuf time) 443 * that we will need to shift around. 444 */ 445 void 446 cluster_callback(bp) 447 struct buf *bp; 448 { 449 struct buf *nbp, *tbp; 450 int error = 0; 451 452 /* 453 * Must propogate errors to all the components. 454 */ 455 if (bp->b_flags & B_ERROR) 456 error = bp->b_error; 457 458 pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); 459 /* 460 * Move memory from the large cluster buffer into the component 461 * buffers and mark IO as done on these. 462 */ 463 for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head); 464 tbp; tbp = nbp) { 465 nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry); 466 if (error) { 467 tbp->b_flags |= B_ERROR; 468 tbp->b_error = error; 469 } else 470 tbp->b_dirtyoff = tbp->b_dirtyend = 0; 471 biodone(tbp); 472 } 473 relpbuf(bp); 474 } 475 476 /* 477 * Do clustered write for FFS. 478 * 479 * Three cases: 480 * 1. Write is not sequential (write asynchronously) 481 * Write is sequential: 482 * 2. beginning of cluster - begin cluster 483 * 3. middle of a cluster - add to cluster 484 * 4. end of a cluster - asynchronously write cluster 485 */ 486 void 487 cluster_write(bp, filesize) 488 struct buf *bp; 489 u_quad_t filesize; 490 { 491 struct vnode *vp; 492 daddr_t lbn; 493 int maxclen, cursize; 494 int lblocksize; 495 int async; 496 497 vp = bp->b_vp; 498 if (vp->v_maxio == 0) 499 vp->v_maxio = DFLTPHYS; 500 if (vp->v_type == VREG) { 501 async = vp->v_mount->mnt_flag & MNT_ASYNC; 502 lblocksize = vp->v_mount->mnt_stat.f_iosize; 503 } else { 504 async = 0; 505 lblocksize = bp->b_bufsize; 506 } 507 lbn = bp->b_lblkno; 508 509 /* Initialize vnode to beginning of file. */ 510 if (lbn == 0) 511 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; 512 513 if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || 514 (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { 515 maxclen = vp->v_maxio / lblocksize - 1; 516 if (vp->v_clen != 0) { 517 /* 518 * Next block is not sequential. 519 * 520 * If we are not writing at end of file, the process 521 * seeked to another point in the file since its last 522 * write, or we have reached our maximum cluster size, 523 * then push the previous cluster. Otherwise try 524 * reallocating to make it sequential. 525 */ 526 cursize = vp->v_lastw - vp->v_cstart + 1; 527 #ifndef notyet_block_reallocation_enabled 528 if (((u_quad_t)(lbn + 1) * lblocksize) != filesize || 529 lbn != vp->v_lastw + 1 || 530 vp->v_clen <= cursize) { 531 if (!async) 532 cluster_wbuild(vp, lblocksize, 533 vp->v_cstart, cursize); 534 } 535 #else 536 if ((lbn + 1) * lblocksize != filesize || 537 lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { 538 if (!async) 539 cluster_wbuild(vp, lblocksize, 540 vp->v_cstart, cursize); 541 } else { 542 struct buf **bpp, **endbp; 543 struct cluster_save *buflist; 544 545 buflist = cluster_collectbufs(vp, bp); 546 endbp = &buflist->bs_children 547 [buflist->bs_nchildren - 1]; 548 if (VOP_REALLOCBLKS(vp, buflist)) { 549 /* 550 * Failed, push the previous cluster. 551 */ 552 for (bpp = buflist->bs_children; 553 bpp < endbp; bpp++) 554 brelse(*bpp); 555 free(buflist, M_SEGMENT); 556 cluster_wbuild(vp, lblocksize, 557 vp->v_cstart, cursize); 558 } else { 559 /* 560 * Succeeded, keep building cluster. 561 */ 562 for (bpp = buflist->bs_children; 563 bpp <= endbp; bpp++) 564 bdwrite(*bpp); 565 free(buflist, M_SEGMENT); 566 vp->v_lastw = lbn; 567 vp->v_lasta = bp->b_blkno; 568 return; 569 } 570 } 571 #endif /* notyet_block_reallocation_enabled */ 572 } 573 /* 574 * Consider beginning a cluster. If at end of file, make 575 * cluster as large as possible, otherwise find size of 576 * existing cluster. 577 */ 578 if ((vp->v_type == VREG) && 579 ((u_quad_t) (lbn + 1) * lblocksize) != filesize && 580 (bp->b_blkno == bp->b_lblkno) && 581 (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || 582 bp->b_blkno == -1)) { 583 bawrite(bp); 584 vp->v_clen = 0; 585 vp->v_lasta = bp->b_blkno; 586 vp->v_cstart = lbn + 1; 587 vp->v_lastw = lbn; 588 return; 589 } 590 vp->v_clen = maxclen; 591 if (!async && maxclen == 0) { /* I/O not contiguous */ 592 vp->v_cstart = lbn + 1; 593 bawrite(bp); 594 } else { /* Wait for rest of cluster */ 595 vp->v_cstart = lbn; 596 bdwrite(bp); 597 } 598 } else if (lbn == vp->v_cstart + vp->v_clen) { 599 /* 600 * At end of cluster, write it out. 601 */ 602 bdwrite(bp); 603 cluster_wbuild(vp, lblocksize, vp->v_cstart, vp->v_clen + 1); 604 vp->v_clen = 0; 605 vp->v_cstart = lbn + 1; 606 } else 607 /* 608 * In the middle of a cluster, so just delay the I/O for now. 609 */ 610 bdwrite(bp); 611 vp->v_lastw = lbn; 612 vp->v_lasta = bp->b_blkno; 613 } 614 615 616 /* 617 * This is an awful lot like cluster_rbuild...wish they could be combined. 618 * The last lbn argument is the current block on which I/O is being 619 * performed. Check to see that it doesn't fall in the middle of 620 * the current block (if last_bp == NULL). 621 */ 622 int 623 cluster_wbuild(vp, size, start_lbn, len) 624 struct vnode *vp; 625 long size; 626 daddr_t start_lbn; 627 int len; 628 { 629 struct buf *bp, *tbp; 630 int i, j, s; 631 int totalwritten = 0; 632 int dbsize = btodb(size); 633 while (len > 0) { 634 s = splbio(); 635 if (((tbp = gbincore(vp, start_lbn)) == NULL) || 636 ((tbp->b_flags & (B_INVAL|B_BUSY|B_DELWRI)) != B_DELWRI)) { 637 ++start_lbn; 638 --len; 639 splx(s); 640 continue; 641 } 642 bremfree(tbp); 643 tbp->b_flags |= B_BUSY; 644 tbp->b_flags &= ~B_DONE; 645 splx(s); 646 647 /* 648 * Extra memory in the buffer, punt on this buffer. XXX we could 649 * handle this in most cases, but we would have to push the extra 650 * memory down to after our max possible cluster size and then 651 * potentially pull it back up if the cluster was terminated 652 * prematurely--too much hassle. 653 */ 654 if (((tbp->b_flags & (B_CLUSTEROK|B_MALLOC)) != B_CLUSTEROK) || 655 (tbp->b_bcount != tbp->b_bufsize) || 656 (tbp->b_bcount != size) || 657 len == 1) { 658 totalwritten += tbp->b_bufsize; 659 bawrite(tbp); 660 ++start_lbn; 661 --len; 662 continue; 663 } 664 665 bp = trypbuf(); 666 if (bp == NULL) { 667 totalwritten += tbp->b_bufsize; 668 bawrite(tbp); 669 ++start_lbn; 670 --len; 671 continue; 672 } 673 674 TAILQ_INIT(&bp->b_cluster.cluster_head); 675 bp->b_bcount = 0; 676 bp->b_bufsize = 0; 677 bp->b_npages = 0; 678 if (tbp->b_wcred != NOCRED) { 679 bp->b_wcred = tbp->b_wcred; 680 crhold(bp->b_wcred); 681 } 682 683 bp->b_blkno = tbp->b_blkno; 684 bp->b_lblkno = tbp->b_lblkno; 685 (vm_offset_t) bp->b_data |= ((vm_offset_t) tbp->b_data) & PAGE_MASK; 686 bp->b_flags |= B_CALL | B_BUSY | B_CLUSTER | 687 (tbp->b_flags & (B_VMIO|B_NEEDCOMMIT)); 688 bp->b_iodone = cluster_callback; 689 pbgetvp(vp, bp); 690 for (i = 0; i < len; ++i, ++start_lbn) { 691 if (i != 0) { 692 s = splbio(); 693 if ((tbp = gbincore(vp, start_lbn)) == NULL) { 694 splx(s); 695 break; 696 } 697 698 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)))) { 699 splx(s); 700 break; 701 } 702 703 if (tbp->b_wcred != bp->b_wcred) { 704 splx(s); 705 break; 706 } 707 708 if ((tbp->b_bcount != size) || 709 ((bp->b_blkno + dbsize * i) != tbp->b_blkno) || 710 ((tbp->b_npages + bp->b_npages) > (vp->v_maxio / PAGE_SIZE))) { 711 splx(s); 712 break; 713 } 714 bremfree(tbp); 715 tbp->b_flags |= B_BUSY; 716 tbp->b_flags &= ~B_DONE; 717 splx(s); 718 } 719 /* check for latent dependencies to be handled */ 720 if ((LIST_FIRST(&tbp->b_dep)) != NULL && 721 bioops.io_start) 722 (*bioops.io_start)(tbp); 723 if (tbp->b_flags & B_VMIO) { 724 vm_page_t m; 725 726 if (i != 0) { 727 for (j = 0; j < tbp->b_npages; j += 1) { 728 m = tbp->b_pages[j]; 729 if (m->flags & PG_BUSY) 730 goto finishcluster; 731 } 732 } 733 734 for (j = 0; j < tbp->b_npages; j += 1) { 735 m = tbp->b_pages[j]; 736 ++m->busy; 737 ++m->object->paging_in_progress; 738 if ((bp->b_npages == 0) || 739 (bp->b_pages[bp->b_npages - 1] != m)) { 740 bp->b_pages[bp->b_npages] = m; 741 bp->b_npages++; 742 } 743 } 744 } 745 bp->b_bcount += size; 746 bp->b_bufsize += size; 747 748 --numdirtybuffers; 749 tbp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI); 750 tbp->b_flags |= B_ASYNC; 751 s = splbio(); 752 reassignbuf(tbp, tbp->b_vp); /* put on clean list */ 753 ++tbp->b_vp->v_numoutput; 754 splx(s); 755 TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, 756 tbp, b_cluster.cluster_entry); 757 } 758 finishcluster: 759 pmap_qenter(trunc_page((vm_offset_t) bp->b_data), 760 (vm_page_t *) bp->b_pages, bp->b_npages); 761 if (bp->b_bufsize > bp->b_kvasize) 762 panic("cluster_wbuild: b_bufsize(%d) > b_kvasize(%d)\n", 763 bp->b_bufsize, bp->b_kvasize); 764 bp->b_kvasize = bp->b_bufsize; 765 totalwritten += bp->b_bufsize; 766 bp->b_dirtyoff = 0; 767 bp->b_dirtyend = bp->b_bufsize; 768 bawrite(bp); 769 770 len -= i; 771 } 772 return totalwritten; 773 } 774 775 #ifdef notyet_block_reallocation_enabled 776 /* 777 * Collect together all the buffers in a cluster. 778 * Plus add one additional buffer. 779 */ 780 static struct cluster_save * 781 cluster_collectbufs(vp, last_bp) 782 struct vnode *vp; 783 struct buf *last_bp; 784 { 785 struct cluster_save *buflist; 786 daddr_t lbn; 787 int i, len; 788 789 len = vp->v_lastw - vp->v_cstart + 1; 790 buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), 791 M_SEGMENT, M_WAITOK); 792 buflist->bs_nchildren = 0; 793 buflist->bs_children = (struct buf **) (buflist + 1); 794 for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) 795 (void) bread(vp, lbn, last_bp->b_bcount, NOCRED, 796 &buflist->bs_children[i]); 797 buflist->bs_children[i] = last_bp; 798 buflist->bs_nchildren = i + 1; 799 return (buflist); 800 } 801 #endif /* notyet_block_reallocation_enabled */ 802