vfs_bio.c revision 1.31
1/* $NetBSD: vfs_bio.c,v 1.31 1994/07/03 07:57:32 cgd Exp $ */ 2 3/*- 4 * Copyright (c) 1994 Christopher G. Demetriou 5 * Copyright (c) 1982, 1986, 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the University of 24 * California, Berkeley and its contributors. 25 * 4. Neither the name of the University nor the names of its contributors 26 * may be used to endorse or promote products derived from this software 27 * without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 39 * SUCH DAMAGE. 40 * 41 * @(#)vfs_bio.c 8.6 (Berkeley) 1/11/94 42 */ 43 44/* 45 * Some references: 46 * Bach: The Design of the UNIX Operating System (Prentice Hall, 1986) 47 * Leffler, et al.: The Design and Implementation of the 4.3BSD 48 * UNIX Operating System (Addison Welley, 1989) 49 */ 50 51#include <sys/param.h> 52#include <sys/systm.h> 53#include <sys/proc.h> 54#include <sys/buf.h> 55#include <sys/vnode.h> 56#include <sys/mount.h> 57#include <sys/trace.h> 58#include <sys/malloc.h> 59#include <sys/resourcevar.h> 60 61/* Macros to clear/set/test flags. */ 62#define SET(t, f) (t) |= (f) 63#define CLR(t, f) (t) &= ~(f) 64#define ISSET(t, f) ((t) & (f)) 65 66/* 67 * Definitions for the buffer hash lists. 68 */ 69#define BUFHASH(dvp, lbn) \ 70 (&bufhashtbl[((int)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash]) 71LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash; 72u_long bufhash; 73 74/* 75 * Insq/Remq for the buffer hash lists. 76 */ 77#define binshash(bp, dp) LIST_INSERT_HEAD(dp, bp, b_hash) 78#define bremhash(bp) LIST_REMOVE(bp, b_hash) 79 80/* 81 * Definitions for the buffer free lists. 82 */ 83#define BQUEUES 4 /* number of free buffer queues */ 84 85#define BQ_LOCKED 0 /* super-blocks &c */ 86#define BQ_LRU 1 /* lru, useful buffers */ 87#define BQ_AGE 2 /* rubbish */ 88#define BQ_EMPTY 3 /* buffer headers with no memory */ 89 90TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES]; 91int needbuffer; 92 93/* 94 * Insq/Remq for the buffer free lists. 95 */ 96#define binsheadfree(bp, dp) TAILQ_INSERT_HEAD(dp, bp, b_freelist) 97#define binstailfree(bp, dp) TAILQ_INSERT_TAIL(dp, bp, b_freelist) 98 99void 100bremfree(bp) 101 struct buf *bp; 102{ 103 struct bqueues *dp = NULL; 104 105 /* 106 * We only calculate the head of the freelist when removing 107 * the last element of the list as that is the only time that 108 * it is needed (e.g. to reset the tail pointer). 109 * 110 * NB: This makes an assumption about how tailq's are implemented. 111 */ 112 if (bp->b_freelist.tqe_next == NULL) { 113 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) 114 if (dp->tqh_last == &bp->b_freelist.tqe_next) 115 break; 116 if (dp == &bufqueues[BQUEUES]) 117 panic("bremfree: lost tail"); 118 } 119 TAILQ_REMOVE(dp, bp, b_freelist); 120} 121 122/* 123 * Initialize buffers and hash links for buffers. 124 */ 125void 126bufinit() 127{ 128 register struct buf *bp; 129 struct bqueues *dp; 130 register int i; 131 int base, residual; 132 133 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) 134 TAILQ_INIT(dp); 135 bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash); 136 base = bufpages / nbuf; 137 residual = bufpages % nbuf; 138 for (i = 0; i < nbuf; i++) { 139 bp = &buf[i]; 140 bzero((char *)bp, sizeof *bp); 141 bp->b_dev = NODEV; 142 bp->b_rcred = NOCRED; 143 bp->b_wcred = NOCRED; 144 bp->b_vnbufs.le_next = NOLIST; 145 bp->b_data = buffers + i * MAXBSIZE; 146 if (i < residual) 147 bp->b_bufsize = (base + 1) * CLBYTES; 148 else 149 bp->b_bufsize = base * CLBYTES; 150 bp->b_flags = B_INVAL; 151 dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY]; 152 binsheadfree(bp, dp); 153 binshash(bp, &invalhash); 154 } 155} 156 157/* 158 * Read a disk block. 159 * This algorithm described in Bach (p.54). 160 */ 161bread(vp, blkno, size, cred, bpp) 162 struct vnode *vp; 163 daddr_t blkno; 164 int size; 165 struct ucred *cred; 166 struct buf **bpp; 167{ 168 register struct buf *bp; 169 170 /* Get buffer for block. */ 171 bp = *bpp = getblk(vp, blkno, size, 0, 0); 172 173 /* 174 * If buffer data valid, return it. 175 * Note that if buffer is B_INVAL, getblk() won't return it. 176 * Therefore, it's valid if it's I/O has completed or been delayed. 177 */ 178 if (ISSET(bp->b_flags, (B_DONE | B_DELWRI))) 179 return (0); 180 181 /* Start some I/O for the buffer (keeping credentials, if needed). */ 182 SET(bp->b_flags, B_READ); 183 if (cred != NOCRED && bp->b_rcred == NOCRED) { 184 crhold(cred); 185 bp->b_rcred = cred; 186 } 187 VOP_STRATEGY(bp); 188 189 /* Pay for the read. */ 190 curproc->p_stats->p_ru.ru_inblock++; /* XXX */ 191 192 /* Wait for the read to complete, and return result. */ 193 return (biowait(bp)); 194} 195 196/* 197 * Read-ahead multiple disk blocks. The first is sync, the rest async. 198 * Trivial modification to the breada algorithm presented in Bach (p.55). 199 */ 200breadn(vp, blkno, size, rablks, rasizes, nrablks, cred, bpp) 201 struct vnode *vp; 202 daddr_t blkno; int size; 203 daddr_t rablks[]; int rasizes[]; 204 int nrablks; 205 struct ucred *cred; 206 struct buf **bpp; 207{ 208 struct buf *bp, *rabp; 209 int i; 210 211 bp = NULL; /* We don't have a buffer yet. */ 212 213 /* If first block not in cache, get buffer for it and read it in. */ 214 if (!incore(vp, blkno)) { 215 bp = *bpp = getblk(vp, blkno, size, 0, 0); 216 217 /* 218 * If buffer data not valid, we have to read it in. 219 * If it is valid, just hold on to the buffer pointer. 220 */ 221 if (!ISSET(bp->b_flags, (B_DONE | B_DELWRI))) { 222 /* Start I/O for the buffer (keeping credentials). */ 223 SET(bp->b_flags, B_READ); 224 if (cred != NOCRED && bp->b_rcred == NOCRED) { 225 crhold(cred); 226 bp->b_rcred = cred; 227 } 228 VOP_STRATEGY(bp); 229 230 /* Pay for the read. */ 231 curproc->p_stats->p_ru.ru_inblock++; /* XXX */ 232 } 233 } 234 235 /* 236 * For each of the read-ahead blocks, start a read, if necessary. 237 */ 238 for (i = 0; i < nrablks; i++) { 239 /* If it's in the cache, just go on to next one. */ 240 if (incore(vp, rablks[i])) 241 continue; 242 243 /* Get a buffer for the read-ahead block */ 244 rabp = getblk(vp, rablks[i], rasizes[i], 0, 0); 245 246 /* 247 * If buffer data valid, just release the buffer back into 248 * the cache. If it's not valid, we have to read it in. 249 */ 250 if (ISSET(rabp->b_flags, (B_DONE | B_DELWRI))) 251 brelse(rabp); 252 else { 253 /* Start I/O for the buffer (keeping credentials). */ 254 SET(rabp->b_flags, (B_READ | B_ASYNC)); 255 if (cred != NOCRED && rabp->b_rcred == NOCRED) { 256 crhold(cred); 257 rabp->b_rcred = cred; 258 } 259 VOP_STRATEGY(rabp); 260 261 /* Pay for the read. */ 262 curproc->p_stats->p_ru.ru_inblock++; /* XXX */ 263 } 264 } 265 266 /* 267 * If first block was originally in the cache (i.e. we *still* don't 268 * have buffer), use bread to get and return it. 269 */ 270 if (bp == NULL) 271 return (bread(vp, blkno, size, cred, bpp)); 272 273 /* Otherwise, we had to start a read for it; wait until it's valid. */ 274 return (biowait(bp)); 275} 276 277/* 278 * Read with single-block read-ahead. Defined in Bach (p.55), but 279 * implemented as a call to breadn(). 280 * XXX for compatibility with old file systems. 281 */ 282breada(vp, blkno, size, rablkno, rabsize, cred, bpp) 283 struct vnode *vp; 284 daddr_t blkno; int size; 285 daddr_t rablkno; int rabsize; 286 struct ucred *cred; 287 struct buf **bpp; 288{ 289 return (breadn(vp, blkno, size, &rablkno, &rabsize, 1, cred, bpp)); 290} 291 292/* 293 * Block write. Described in Bach (p.56) 294 */ 295bwrite(bp) 296 struct buf *bp; 297{ 298 int rv, s, sync, wasdelayed; 299 300 rv = 0; 301 302 /* Remember buffer type, to switch on it later. */ 303 sync = !ISSET(bp->b_flags, B_ASYNC); 304 wasdelayed = ISSET(bp->b_flags, B_DELWRI); 305 CLR(bp->b_flags, (B_READ | B_DONE | B_ERROR | B_DELWRI)); 306 307 /* Initiate disk write. Make sure the appropriate party is charged. */ 308 SET(bp->b_flags, B_WRITEINPROG); 309 bp->b_vp->v_numoutput++; 310 VOP_STRATEGY(bp); 311 312 /* 313 * If I/O was synchronous, wait for it to complete. 314 */ 315 if (sync) 316 rv = biowait(bp); 317 318 /* 319 * Pay for the I/O operation, if it's not been paid for, and 320 * make sure it's on the correct vnode queue. 321 */ 322 if (wasdelayed) 323 reassignbuf(bp, bp->b_vp); 324 else 325 curproc->p_stats->p_ru.ru_oublock++; 326 327 /* Release the buffer, or, if async, make sure it gets reused ASAP. */ 328 if (sync) 329 brelse(bp); 330 else if (wasdelayed) { 331 s = splbio(); 332 SET(bp->b_flags, B_AGE); 333 splx(s); 334 } 335 return (rv); 336} 337 338int 339vn_bwrite(ap) 340 struct vop_bwrite_args *ap; 341{ 342 return (bwrite(ap->a_bp)); 343} 344 345/* 346 * Delayed write. 347 * 348 * The buffer is marked dirty, but is not queued for I/O. 349 * This routine should be used when the buffer is expected 350 * to be modified again soon, typically a small write that 351 * partially fills a buffer. 352 * 353 * NB: magnetic tapes cannot be delayed; they must be 354 * written in the order that the writes are requested. 355 * 356 * Described in Leffler, et al. (pp. 208-213). 357 */ 358void 359bdwrite(bp) 360 struct buf *bp; 361{ 362 363 /* 364 * If the block hasn't been seen before: 365 * (1) Mark it as having been seen, 366 * (2) Charge for the write. 367 * (3) Make sure it's on its vnode's correct block list, 368 */ 369 if (!ISSET(bp->b_flags, B_DELWRI)) { 370 SET(bp->b_flags, B_DELWRI); 371 curproc->p_stats->p_ru.ru_oublock++; /* XXX */ 372 reassignbuf(bp, bp->b_vp); 373 } 374 375 /* If this is a tape block, write it the block now. */ 376 if (ISSET(bp->b_flags, B_TAPE)) { 377 bwrite(bp); 378 return; 379 } 380 381 /* Otherwise, the "write" is done, so mark and release the buffer. */ 382 SET(bp->b_flags, B_DONE); 383 brelse(bp); 384} 385 386/* 387 * Asynchronous block write; just an asynchronous bwrite(). 388 */ 389void 390bawrite(bp) 391 struct buf *bp; 392{ 393 394 SET(bp->b_flags, B_ASYNC); 395 VOP_BWRITE(bp); 396} 397 398/* 399 * Release a buffer on to the free lists. 400 * Described in Bach (p. 46). 401 */ 402void 403brelse(bp) 404 struct buf *bp; 405{ 406 struct bqueues *bufq; 407 int s; 408 409 /* Wake up any processes waiting for any buffer to become free. */ 410 if (needbuffer) { 411 needbuffer = 0; 412 wakeup(&needbuffer); 413 } 414 415 /* Wake up any proceeses waiting for _this_ buffer to become free. */ 416 if (ISSET(bp->b_flags, B_WANTED)) { 417 CLR(bp->b_flags, B_WANTED); 418 wakeup(bp); 419 } 420 421 /* Block disk interrupts. */ 422 s = splbio(); 423 424 /* 425 * Determine which queue the buffer should be on, then put it there. 426 */ 427 428 /* If it's locked, don't report an error; try again later. */ 429 if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR)) 430 CLR(bp->b_flags, B_ERROR); 431 432 /* If it's not cacheable, or an error, mark it invalid. */ 433 if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR))) 434 SET(bp->b_flags, B_INVAL); 435 436 if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) { 437 /* 438 * If it's invalid or empty, dissociate it from its vnode 439 * and put on the head of the appropriate queue. 440 */ 441 if (bp->b_vp) 442 brelvp(bp); 443 CLR(bp->b_flags, B_DELWRI); 444 if (bp->b_bufsize <= 0) 445 /* no data */ 446 bufq = &bufqueues[BQ_EMPTY]; 447 else 448 /* invalid data */ 449 bufq = &bufqueues[BQ_AGE]; 450 binsheadfree(bp, bufq); 451 } else { 452 /* 453 * It has valid data. Put it on the end of the appropriate 454 * queue, so that it'll stick around for as long as possible. 455 */ 456 if (ISSET(bp->b_flags, B_LOCKED)) 457 /* locked in core */ 458 bufq = &bufqueues[BQ_LOCKED]; 459 else if (ISSET(bp->b_flags, B_AGE)) 460 /* stale but valid data */ 461 bufq = &bufqueues[BQ_AGE]; 462 else 463 /* valid data */ 464 bufq = &bufqueues[BQ_LRU]; 465 binstailfree(bp, bufq); 466 } 467 468 /* Unlock the buffer. */ 469 CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE)); 470 471 /* Allow disk interrupts. */ 472 splx(s); 473} 474 475/* 476 * Determine if a block is in the cache. 477 * Just look on what would be its hash chain. If it's there, return 478 * a pointer to it, unless it's marked invalid. If it's marked invalid, 479 * we normally don't return the buffer, unless the caller explicitly 480 * wants us to. 481 */ 482struct buf * 483incore(vp, blkno) 484 struct vnode *vp; 485 daddr_t blkno; 486{ 487 struct buf *bp; 488 489 bp = BUFHASH(vp, blkno)->lh_first; 490 491 /* Search hash chain */ 492 for (; bp != NULL; bp = bp->b_hash.le_next) { 493 if (bp->b_lblkno == blkno && bp->b_vp == vp && 494 !ISSET(bp->b_flags, B_INVAL)) 495 return (bp); 496 } 497 498 return (0); 499} 500 501/* 502 * Get a block of requested size that is associated with 503 * a given vnode and block offset. If it is found in the 504 * block cache, mark it as having been found, make it busy 505 * and return it. Otherwise, return an empty block of the 506 * correct size. It is up to the caller to insure that the 507 * cached blocks be of the correct size. 508 */ 509struct buf * 510getblk(vp, blkno, size, slpflag, slptimeo) 511 register struct vnode *vp; 512 daddr_t blkno; 513 int size, slpflag, slptimeo; 514{ 515 struct buf *bp; 516 int s, err; 517 518start: 519 s = splbio(); 520 if (bp = incore(vp, blkno)) { /* XXX NFS VOP_BWRITE foolishness */ 521 if (ISSET(bp->b_flags, B_BUSY)) { 522 SET(bp->b_flags, B_WANTED); 523 err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk", 524 slptimeo); 525 splx(s); 526 if (err) 527 return (NULL); 528 goto start; 529 } 530 SET(bp->b_flags, (B_BUSY | B_CACHE)); 531 bremfree(bp); 532 splx(s); 533 allocbuf(bp, size); 534 } else { 535 splx(s); 536 if ((bp = getnewbuf(slpflag, slptimeo)) == NULL) 537 goto start; 538 allocbuf(bp, size); 539 bp->b_blkno = bp->b_lblkno = blkno; 540 s = splbio(); 541 bgetvp(vp, bp); 542 splx(s); 543 bremhash(bp); 544 binshash(bp, BUFHASH(vp, blkno)); 545 } 546 return (bp); 547} 548 549/* 550 * Get an empty, disassociated buffer of given size. 551 */ 552struct buf * 553geteblk(size) 554 int size; 555{ 556 struct buf *bp; 557 558 while ((bp = getnewbuf(0, 0)) == 0) 559 ; 560 SET(bp->b_flags, B_INVAL); 561 bremhash(bp); 562 binshash(bp, &invalhash); 563 allocbuf(bp, size); 564 bp->b_bcount = 0; 565 bp->b_error = 0; 566 bp->b_resid = 0; 567 568 return (bp); 569} 570 571/* 572 * Expand or contract the actual memory allocated to a buffer. 573 * 574 * If the buffer shrinks, data is lost, so it's up to the 575 * caller to have written it out *first*; this routine will not 576 * start a write. If the buffer grows, it's the callers 577 * responsibility to fill out the buffer's additional contents. 578 */ 579allocbuf(bp, size) 580 struct buf *bp; 581 int size; 582{ 583 struct buf *nbp; 584 vm_size_t desired_size; 585 int s; 586 587 desired_size = roundup(size, CLBYTES); 588 if (desired_size > MAXBSIZE) 589 panic("allocbuf: buffer larger than MAXBSIZE requested"); 590 591 if (bp->b_bufsize == desired_size) 592 goto out; 593 594 /* 595 * If the buffer is smaller than the desired size, we need to snarf 596 * it from other buffers. Get buffers (via getnewbuf()), and 597 * steal their pages. 598 */ 599 while (bp->b_bufsize < desired_size) { 600 int amt; 601 602 /* find a buffer */ 603 while ((nbp = getnewbuf(0, 0)) == NULL) 604 ; 605 606 /* and steal its pages, up to the amount we need */ 607 amt = min(nbp->b_bufsize, (desired_size - bp->b_bufsize)); 608 pagemove((nbp->b_data + nbp->b_bufsize - amt), 609 bp->b_data + bp->b_bufsize, amt); 610 bp->b_bufsize += amt; 611 nbp->b_bufsize -= amt; 612 613 /* reduce transfer count if we stole some data */ 614 if (nbp->b_bcount > nbp->b_bufsize) 615 nbp->b_bcount = nbp->b_bufsize; 616 617#ifdef DIAGNOSTIC 618 if (nbp->b_bufsize < 0) 619 panic("allocbuf: negative bufsize"); 620#endif 621 if (nbp->b_bufsize == 0) { 622 bremhash(nbp); 623 binshash(nbp, &invalhash); 624 SET(nbp->b_flags, B_INVAL); 625 nbp->b_error = 0; 626 nbp->b_dev = NODEV; 627 } 628 brelse(nbp); 629 } 630 631 /* 632 * If we want a buffer smaller than the current size, 633 * shrink this buffer. Grab a buf head from the EMPTY queue, 634 * move a page onto it, and put it on front of the AGE queue. 635 * If there are no free buffer headers, leave the buffer alone. 636 */ 637 if (bp->b_bufsize > desired_size) { 638 s = splbio(); 639 if ((nbp = bufqueues[BQ_EMPTY].tqh_first) == NULL) { 640 /* No free buffer head */ 641 splx(s); 642 goto out; 643 } 644 bremfree(nbp); 645 SET(nbp->b_flags, B_BUSY); 646 splx(s); 647 648 /* move the page to it and note this change */ 649 pagemove(bp->b_data + desired_size, 650 nbp->b_data, bp->b_bufsize - desired_size); 651 nbp->b_bufsize = bp->b_bufsize - desired_size; 652 bp->b_bufsize = desired_size; 653 nbp->b_bcount = 0; 654 SET(nbp->b_flags, B_INVAL); 655 656 /* release the newly-filled buffer and leave */ 657 brelse(nbp); 658 } 659 660out: 661 bp->b_bcount = size; 662} 663 664/* 665 * Find a buffer which is available for use. 666 * Select something from a free list. 667 * Preference is to AGE list, then LRU list. 668 */ 669struct buf * 670getnewbuf(slpflag, slptimeo) 671 int slpflag, slptimeo; 672{ 673 register struct buf *bp; 674 int s; 675 676start: 677 s = splbio(); 678 if ((bp = bufqueues[BQ_AGE].tqh_first) != NULL || 679 (bp = bufqueues[BQ_LRU].tqh_first) != NULL) { 680 bremfree(bp); 681 } else { 682 /* wait for a free buffer of any kind */ 683 needbuffer = 1; 684 tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo); 685 splx(s); 686 return (0); 687 } 688 689 /* Buffer is no longer on free lists. */ 690 SET(bp->b_flags, B_BUSY); 691 splx(s); 692 693 /* If buffer was a delayed write, start it, and go back to the top. */ 694 if (ISSET(bp->b_flags, B_DELWRI)) { 695 bawrite (bp); 696 goto start; 697 } 698 699 /* disassociate us from our vnode, if we had one... */ 700 s = splbio(); 701 if (bp->b_vp) 702 brelvp(bp); 703 splx(s); 704 705 /* clear out various other fields */ 706 bp->b_flags = B_BUSY; 707 bp->b_dev = NODEV; 708 bp->b_blkno = bp->b_lblkno = 0; 709 bp->b_iodone = 0; 710 bp->b_error = 0; 711 bp->b_resid = 0; 712 bp->b_bcount = 0; 713 bp->b_dirtyoff = bp->b_dirtyend = 0; 714 bp->b_validoff = bp->b_validend = 0; 715 716 /* nuke any credentials we were holding */ 717 if (bp->b_rcred != NOCRED) { 718 crfree(bp->b_rcred); 719 bp->b_rcred = NOCRED; 720 } 721 if (bp->b_wcred != NOCRED) { 722 crfree(bp->b_wcred); 723 bp->b_wcred = NOCRED; 724 } 725 726 return (bp); 727} 728 729/* 730 * Wait for operations on the buffer to complete. 731 * When they do, extract and return the I/O's error value. 732 */ 733int 734biowait(bp) 735 struct buf *bp; 736{ 737 int s; 738 739 s = splbio(); 740 while (!ISSET(bp->b_flags, B_DONE)) 741 tsleep(bp, PRIBIO + 1, "biowait", 0); 742 splx(s); 743 744 /* check for interruption of I/O (e.g. via NFS), then errors. */ 745 if (ISSET(bp->b_flags, B_EINTR)) { 746 CLR(bp->b_flags, B_EINTR); 747 return (EINTR); 748 } else if (ISSET(bp->b_flags, B_ERROR)) 749 return (bp->b_error ? bp->b_error : EIO); 750 else 751 return (0); 752} 753 754/* 755 * Mark I/O complete on a buffer. 756 * 757 * If a callback has been requested, e.g. the pageout 758 * daemon, do so. Otherwise, awaken waiting processes. 759 * 760 * [ Leffler, et al., says on p.247: 761 * "This routine wakes up the blocked process, frees the buffer 762 * for an asynchronous write, or, for a request by the pagedaemon 763 * process, invokes a procedure specified in the buffer structure" ] 764 * 765 * In real life, the pagedaemon (or other system processes) wants 766 * to do async stuff to, and doesn't want the buffer brelse()'d. 767 * (for swap pager, that puts swap buffers on the free lists (!!!), 768 * for the vn device, that puts malloc'd buffers on the free lists!) 769 */ 770void 771biodone(bp) 772 struct buf *bp; 773{ 774 if (ISSET(bp->b_flags, B_DONE)) 775 panic("biodone already"); 776 SET(bp->b_flags, B_DONE); /* note that it's done */ 777 778 if (!ISSET(bp->b_flags, B_READ)) /* wake up reader */ 779 vwakeup(bp); 780 781 if (ISSET(bp->b_flags, B_CALL)) { /* if necessary, call out */ 782 CLR(bp->b_flags, B_CALL); /* but note callout done */ 783 (*bp->b_iodone)(bp); 784 } else if (ISSET(bp->b_flags, B_ASYNC)) /* if async, release it */ 785 brelse(bp); 786 else { /* or just wakeup the buffer */ 787 CLR(bp->b_flags, B_WANTED); 788 wakeup(bp); 789 } 790} 791 792/* 793 * Return a count of buffers on the "locked" queue. 794 */ 795int 796count_lock_queue() 797{ 798 register struct buf *bp; 799 register int n = 0; 800 801 for (bp = bufqueues[BQ_LOCKED].tqh_first; bp; 802 bp = bp->b_freelist.tqe_next) 803 n++; 804 return (n); 805} 806 807#ifdef DIAGNOSTIC 808/* 809 * Print out statistics on the current allocation of the buffer pool. 810 * Can be enabled to print out on every ``sync'' by setting "syncprt" 811 * in vfs_syscalls.c using sysctl. 812 */ 813void 814vfs_bufstats() 815{ 816 int s, i, j, count; 817 register struct buf *bp; 818 register struct bqueues *dp; 819 int counts[MAXBSIZE/CLBYTES+1]; 820 static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" }; 821 822 for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) { 823 count = 0; 824 for (j = 0; j <= MAXBSIZE/CLBYTES; j++) 825 counts[j] = 0; 826 s = splbio(); 827 for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) { 828 counts[bp->b_bufsize/CLBYTES]++; 829 count++; 830 } 831 splx(s); 832 printf("%s: total-%d", bname[i], count); 833 for (j = 0; j <= MAXBSIZE/CLBYTES; j++) 834 if (counts[j] != 0) 835 printf(", %d-%d", j * CLBYTES, counts[j]); 836 printf("\n"); 837 } 838} 839#endif /* DIAGNOSTIC */ 840