1 /* $NetBSD: vfs_bio.c,v 1.50 1997/04/09 21:12:10 mycroft 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 #include <sys/conf.h> 61 62 #include <vm/vm.h> 63 64 /* Macros to clear/set/test flags. */ 65 #define SET(t, f) (t) |= (f) 66 #define CLR(t, f) (t) &= ~(f) 67 #define ISSET(t, f) ((t) & (f)) 68 69 /* 70 * Definitions for the buffer hash lists. 71 */ 72 #define BUFHASH(dvp, lbn) \ 73 (&bufhashtbl[((long)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash]) 74 LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash; 75 u_long bufhash; 76 77 /* 78 * Insq/Remq for the buffer hash lists. 79 */ 80 #define binshash(bp, dp) LIST_INSERT_HEAD(dp, bp, b_hash) 81 #define bremhash(bp) LIST_REMOVE(bp, b_hash) 82 83 /* 84 * Definitions for the buffer free lists. 85 */ 86 #define BQUEUES 4 /* number of free buffer queues */ 87 88 #define BQ_LOCKED 0 /* super-blocks &c */ 89 #define BQ_LRU 1 /* lru, useful buffers */ 90 #define BQ_AGE 2 /* rubbish */ 91 #define BQ_EMPTY 3 /* buffer headers with no memory */ 92 93 TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES]; 94 int needbuffer; 95 96 /* 97 * Insq/Remq for the buffer free lists. 98 */ 99 #define binsheadfree(bp, dp) TAILQ_INSERT_HEAD(dp, bp, b_freelist) 100 #define binstailfree(bp, dp) TAILQ_INSERT_TAIL(dp, bp, b_freelist) 101 102 static __inline struct buf *bio_doread __P((struct vnode *, daddr_t, int, 103 struct ucred *, int)); 104 int count_lock_queue __P((void)); 105 106 void 107 bremfree(bp) 108 struct buf *bp; 109 { 110 struct bqueues *dp = NULL; 111 112 /* 113 * We only calculate the head of the freelist when removing 114 * the last element of the list as that is the only time that 115 * it is needed (e.g. to reset the tail pointer). 116 * 117 * NB: This makes an assumption about how tailq's are implemented. 118 */ 119 if (bp->b_freelist.tqe_next == NULL) { 120 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) 121 if (dp->tqh_last == &bp->b_freelist.tqe_next) 122 break; 123 if (dp == &bufqueues[BQUEUES]) 124 panic("bremfree: lost tail"); 125 } 126 TAILQ_REMOVE(dp, bp, b_freelist); 127 } 128 129 /* 130 * Initialize buffers and hash links for buffers. 131 */ 132 void 133 bufinit() 134 { 135 register struct buf *bp; 136 struct bqueues *dp; 137 register int i; 138 int base, residual; 139 140 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) 141 TAILQ_INIT(dp); 142 bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash); 143 base = bufpages / nbuf; 144 residual = bufpages % nbuf; 145 for (i = 0; i < nbuf; i++) { 146 bp = &buf[i]; 147 bzero((char *)bp, sizeof *bp); 148 bp->b_dev = NODEV; 149 bp->b_rcred = NOCRED; 150 bp->b_wcred = NOCRED; 151 bp->b_vnbufs.le_next = NOLIST; 152 bp->b_data = buffers + i * MAXBSIZE; 153 if (i < residual) 154 bp->b_bufsize = (base + 1) * CLBYTES; 155 else 156 bp->b_bufsize = base * CLBYTES; 157 bp->b_flags = B_INVAL; 158 dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY]; 159 binsheadfree(bp, dp); 160 binshash(bp, &invalhash); 161 } 162 } 163 164 static __inline struct buf * 165 bio_doread(vp, blkno, size, cred, async) 166 struct vnode *vp; 167 daddr_t blkno; 168 int size; 169 struct ucred *cred; 170 int async; 171 { 172 register struct buf *bp; 173 struct proc *p = (curproc != NULL ? curproc : &proc0); /* XXX */ 174 175 bp = getblk(vp, blkno, size, 0, 0); 176 177 /* 178 * If buffer does not have data valid, start a read. 179 * Note that if buffer is B_INVAL, getblk() won't return it. 180 * Therefore, it's valid if it's I/O has completed or been delayed. 181 */ 182 if (!ISSET(bp->b_flags, (B_DONE | B_DELWRI))) { 183 /* Start I/O for the buffer (keeping credentials). */ 184 SET(bp->b_flags, B_READ | async); 185 if (cred != NOCRED && bp->b_rcred == NOCRED) { 186 crhold(cred); 187 bp->b_rcred = cred; 188 } 189 VOP_STRATEGY(bp); 190 191 /* Pay for the read. */ 192 p->p_stats->p_ru.ru_inblock++; 193 } else if (async) { 194 brelse(bp); 195 } 196 197 return (bp); 198 } 199 200 /* 201 * Read a disk block. 202 * This algorithm described in Bach (p.54). 203 */ 204 int 205 bread(vp, blkno, size, cred, bpp) 206 struct vnode *vp; 207 daddr_t blkno; 208 int size; 209 struct ucred *cred; 210 struct buf **bpp; 211 { 212 register struct buf *bp; 213 214 /* Get buffer for block. */ 215 bp = *bpp = bio_doread(vp, blkno, size, cred, 0); 216 217 /* Wait for the read to complete, and return result. */ 218 return (biowait(bp)); 219 } 220 221 /* 222 * Read-ahead multiple disk blocks. The first is sync, the rest async. 223 * Trivial modification to the breada algorithm presented in Bach (p.55). 224 */ 225 int 226 breadn(vp, blkno, size, rablks, rasizes, nrablks, cred, bpp) 227 struct vnode *vp; 228 daddr_t blkno; int size; 229 daddr_t rablks[]; int rasizes[]; 230 int nrablks; 231 struct ucred *cred; 232 struct buf **bpp; 233 { 234 register struct buf *bp; 235 int i; 236 237 bp = *bpp = bio_doread(vp, blkno, size, cred, 0); 238 239 /* 240 * For each of the read-ahead blocks, start a read, if necessary. 241 */ 242 for (i = 0; i < nrablks; i++) { 243 /* If it's in the cache, just go on to next one. */ 244 if (incore(vp, rablks[i])) 245 continue; 246 247 /* Get a buffer for the read-ahead block */ 248 (void) bio_doread(vp, rablks[i], rasizes[i], cred, B_ASYNC); 249 } 250 251 /* Otherwise, we had to start a read for it; wait until it's valid. */ 252 return (biowait(bp)); 253 } 254 255 /* 256 * Read with single-block read-ahead. Defined in Bach (p.55), but 257 * implemented as a call to breadn(). 258 * XXX for compatibility with old file systems. 259 */ 260 int 261 breada(vp, blkno, size, rablkno, rabsize, cred, bpp) 262 struct vnode *vp; 263 daddr_t blkno; int size; 264 daddr_t rablkno; int rabsize; 265 struct ucred *cred; 266 struct buf **bpp; 267 { 268 269 return (breadn(vp, blkno, size, &rablkno, &rabsize, 1, cred, bpp)); 270 } 271 272 /* 273 * Block write. Described in Bach (p.56) 274 */ 275 int 276 bwrite(bp) 277 struct buf *bp; 278 { 279 int rv, sync, wasdelayed, s; 280 struct proc *p = (curproc != NULL ? curproc : &proc0); /* XXX */ 281 282 /* 283 * Remember buffer type, to switch on it later. If the write was 284 * synchronous, but the file system was mounted with MNT_ASYNC, 285 * convert it to a delayed write. 286 * XXX note that this relies on delayed tape writes being converted 287 * to async, not sync writes (which is safe, but ugly). 288 */ 289 sync = !ISSET(bp->b_flags, B_ASYNC); 290 if (sync && bp->b_vp && bp->b_vp->v_mount && 291 ISSET(bp->b_vp->v_mount->mnt_flag, MNT_ASYNC)) { 292 bdwrite(bp); 293 return (0); 294 } 295 296 wasdelayed = ISSET(bp->b_flags, B_DELWRI); 297 CLR(bp->b_flags, (B_READ | B_DONE | B_ERROR | B_DELWRI)); 298 299 s = splbio(); 300 301 /* 302 * Pay for the I/O operation and make sure the buf is on the correct 303 * vnode queue. 304 */ 305 if (wasdelayed) 306 reassignbuf(bp, bp->b_vp); 307 else 308 p->p_stats->p_ru.ru_oublock++; 309 310 /* Initiate disk write. Make sure the appropriate party is charged. */ 311 bp->b_vp->v_numoutput++; 312 splx(s); 313 314 SET(bp->b_flags, B_WRITEINPROG); 315 VOP_STRATEGY(bp); 316 317 if (sync) { 318 /* If I/O was synchronous, wait for it to complete. */ 319 rv = biowait(bp); 320 321 /* Release the buffer. */ 322 brelse(bp); 323 324 return (rv); 325 } else { 326 return (0); 327 } 328 } 329 330 int 331 vn_bwrite(v) 332 void *v; 333 { 334 struct vop_bwrite_args *ap = v; 335 336 return (bwrite(ap->a_bp)); 337 } 338 339 /* 340 * Delayed write. 341 * 342 * The buffer is marked dirty, but is not queued for I/O. 343 * This routine should be used when the buffer is expected 344 * to be modified again soon, typically a small write that 345 * partially fills a buffer. 346 * 347 * NB: magnetic tapes cannot be delayed; they must be 348 * written in the order that the writes are requested. 349 * 350 * Described in Leffler, et al. (pp. 208-213). 351 */ 352 void 353 bdwrite(bp) 354 struct buf *bp; 355 { 356 int s; 357 struct proc *p = (curproc != NULL ? curproc : &proc0); /* XXX */ 358 359 /* If this is a tape block, write the block now. */ 360 if (bdevsw[major(bp->b_dev)].d_type == D_TAPE) { 361 bawrite(bp); 362 return; 363 } 364 365 /* 366 * If the block hasn't been seen before: 367 * (1) Mark it as having been seen, 368 * (2) Charge for the write, 369 * (3) Make sure it's on its vnode's correct block list. 370 */ 371 if (!ISSET(bp->b_flags, B_DELWRI)) { 372 SET(bp->b_flags, B_DELWRI); 373 p->p_stats->p_ru.ru_oublock++; 374 s = splbio(); 375 reassignbuf(bp, bp->b_vp); 376 splx(s); 377 } 378 379 /* Otherwise, the "write" is done, so mark and release the buffer. */ 380 CLR(bp->b_flags, B_NEEDCOMMIT); 381 SET(bp->b_flags, B_DONE); 382 brelse(bp); 383 } 384 385 /* 386 * Asynchronous block write; just an asynchronous bwrite(). 387 */ 388 void 389 bawrite(bp) 390 struct buf *bp; 391 { 392 393 SET(bp->b_flags, B_ASYNC); 394 VOP_BWRITE(bp); 395 } 396 397 /* 398 * Release a buffer on to the free lists. 399 * Described in Bach (p. 46). 400 */ 401 void 402 brelse(bp) 403 struct buf *bp; 404 { 405 struct bqueues *bufq; 406 int s; 407 408 /* Wake up any processes waiting for any buffer to become free. */ 409 if (needbuffer) { 410 needbuffer = 0; 411 wakeup(&needbuffer); 412 } 413 414 /* Wake up any proceeses waiting for _this_ buffer to become free. */ 415 if (ISSET(bp->b_flags, B_WANTED)) { 416 CLR(bp->b_flags, B_WANTED); 417 wakeup(bp); 418 } 419 420 /* Block disk interrupts. */ 421 s = splbio(); 422 423 /* 424 * Determine which queue the buffer should be on, then put it there. 425 */ 426 427 /* If it's locked, don't report an error; try again later. */ 428 if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR)) 429 CLR(bp->b_flags, B_ERROR); 430 431 /* If it's not cacheable, or an error, mark it invalid. */ 432 if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR))) 433 SET(bp->b_flags, B_INVAL); 434 435 if (ISSET(bp->b_flags, B_VFLUSH)) { 436 /* 437 * This is a delayed write buffer that was just flushed to 438 * disk. It is still on the LRU queue. If it's become 439 * invalid, then we need to move it to a different queue; 440 * otherwise leave it in its current position. 441 */ 442 CLR(bp->b_flags, B_VFLUSH); 443 if (!ISSET(bp->b_flags, B_ERROR|B_INVAL|B_LOCKED|B_AGE)) 444 goto already_queued; 445 else 446 bremfree(bp); 447 } 448 449 if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) { 450 /* 451 * If it's invalid or empty, dissociate it from its vnode 452 * and put on the head of the appropriate queue. 453 */ 454 if (bp->b_vp) 455 brelvp(bp); 456 CLR(bp->b_flags, B_DELWRI); 457 if (bp->b_bufsize <= 0) 458 /* no data */ 459 bufq = &bufqueues[BQ_EMPTY]; 460 else 461 /* invalid data */ 462 bufq = &bufqueues[BQ_AGE]; 463 binsheadfree(bp, bufq); 464 } else { 465 /* 466 * It has valid data. Put it on the end of the appropriate 467 * queue, so that it'll stick around for as long as possible. 468 */ 469 if (ISSET(bp->b_flags, B_LOCKED)) 470 /* locked in core */ 471 bufq = &bufqueues[BQ_LOCKED]; 472 else if (ISSET(bp->b_flags, B_AGE)) 473 /* stale but valid data */ 474 bufq = &bufqueues[BQ_AGE]; 475 else 476 /* valid data */ 477 bufq = &bufqueues[BQ_LRU]; 478 binstailfree(bp, bufq); 479 } 480 481 already_queued: 482 /* Unlock the buffer. */ 483 CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE)); 484 485 /* Allow disk interrupts. */ 486 splx(s); 487 } 488 489 /* 490 * Determine if a block is in the cache. 491 * Just look on what would be its hash chain. If it's there, return 492 * a pointer to it, unless it's marked invalid. If it's marked invalid, 493 * we normally don't return the buffer, unless the caller explicitly 494 * wants us to. 495 */ 496 struct buf * 497 incore(vp, blkno) 498 struct vnode *vp; 499 daddr_t blkno; 500 { 501 struct buf *bp; 502 503 bp = BUFHASH(vp, blkno)->lh_first; 504 505 /* Search hash chain */ 506 for (; bp != NULL; bp = bp->b_hash.le_next) { 507 if (bp->b_lblkno == blkno && bp->b_vp == vp && 508 !ISSET(bp->b_flags, B_INVAL)) 509 return (bp); 510 } 511 512 return (0); 513 } 514 515 /* 516 * Get a block of requested size that is associated with 517 * a given vnode and block offset. If it is found in the 518 * block cache, mark it as having been found, make it busy 519 * and return it. Otherwise, return an empty block of the 520 * correct size. It is up to the caller to insure that the 521 * cached blocks be of the correct size. 522 */ 523 struct buf * 524 getblk(vp, blkno, size, slpflag, slptimeo) 525 register struct vnode *vp; 526 daddr_t blkno; 527 int size, slpflag, slptimeo; 528 { 529 struct bufhashhdr *bh; 530 struct buf *bp; 531 int s, err; 532 533 /* 534 * XXX 535 * The following is an inlined version of 'incore()', but with 536 * the 'invalid' test moved to after the 'busy' test. It's 537 * necessary because there are some cases in which the NFS 538 * code sets B_INVAL prior to writing data to the server, but 539 * in which the buffers actually contain valid data. In this 540 * case, we can't allow the system to allocate a new buffer for 541 * the block until the write is finished. 542 */ 543 bh = BUFHASH(vp, blkno); 544 start: 545 bp = bh->lh_first; 546 for (; bp != NULL; bp = bp->b_hash.le_next) { 547 if (bp->b_lblkno != blkno || bp->b_vp != vp) 548 continue; 549 550 s = splbio(); 551 if (ISSET(bp->b_flags, B_BUSY)) { 552 SET(bp->b_flags, B_WANTED); 553 err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk", 554 slptimeo); 555 splx(s); 556 if (err) 557 return (NULL); 558 goto start; 559 } 560 561 if (!ISSET(bp->b_flags, B_INVAL)) { 562 SET(bp->b_flags, (B_BUSY | B_CACHE)); 563 bremfree(bp); 564 splx(s); 565 break; 566 } 567 splx(s); 568 } 569 570 if (bp == NULL) { 571 if ((bp = getnewbuf(slpflag, slptimeo)) == NULL) 572 goto start; 573 binshash(bp, bh); 574 bp->b_blkno = bp->b_lblkno = blkno; 575 s = splbio(); 576 bgetvp(vp, bp); 577 splx(s); 578 } 579 allocbuf(bp, size); 580 return (bp); 581 } 582 583 /* 584 * Get an empty, disassociated buffer of given size. 585 */ 586 struct buf * 587 geteblk(size) 588 int size; 589 { 590 struct buf *bp; 591 592 while ((bp = getnewbuf(0, 0)) == 0) 593 ; 594 SET(bp->b_flags, B_INVAL); 595 binshash(bp, &invalhash); 596 allocbuf(bp, size); 597 598 return (bp); 599 } 600 601 /* 602 * Expand or contract the actual memory allocated to a buffer. 603 * 604 * If the buffer shrinks, data is lost, so it's up to the 605 * caller to have written it out *first*; this routine will not 606 * start a write. If the buffer grows, it's the callers 607 * responsibility to fill out the buffer's additional contents. 608 */ 609 void 610 allocbuf(bp, size) 611 struct buf *bp; 612 int size; 613 { 614 struct buf *nbp; 615 vm_size_t desired_size; 616 int s; 617 618 desired_size = roundup(size, CLBYTES); 619 if (desired_size > MAXBSIZE) 620 panic("allocbuf: buffer larger than MAXBSIZE requested"); 621 622 if (bp->b_bufsize == desired_size) 623 goto out; 624 625 /* 626 * If the buffer is smaller than the desired size, we need to snarf 627 * it from other buffers. Get buffers (via getnewbuf()), and 628 * steal their pages. 629 */ 630 while (bp->b_bufsize < desired_size) { 631 int amt; 632 633 /* find a buffer */ 634 while ((nbp = getnewbuf(0, 0)) == NULL) 635 ; 636 SET(nbp->b_flags, B_INVAL); 637 binshash(nbp, &invalhash); 638 639 /* and steal its pages, up to the amount we need */ 640 amt = min(nbp->b_bufsize, (desired_size - bp->b_bufsize)); 641 pagemove((nbp->b_data + nbp->b_bufsize - amt), 642 bp->b_data + bp->b_bufsize, amt); 643 bp->b_bufsize += amt; 644 nbp->b_bufsize -= amt; 645 646 /* reduce transfer count if we stole some data */ 647 if (nbp->b_bcount > nbp->b_bufsize) 648 nbp->b_bcount = nbp->b_bufsize; 649 650 #ifdef DIAGNOSTIC 651 if (nbp->b_bufsize < 0) 652 panic("allocbuf: negative bufsize"); 653 #endif 654 655 brelse(nbp); 656 } 657 658 /* 659 * If we want a buffer smaller than the current size, 660 * shrink this buffer. Grab a buf head from the EMPTY queue, 661 * move a page onto it, and put it on front of the AGE queue. 662 * If there are no free buffer headers, leave the buffer alone. 663 */ 664 if (bp->b_bufsize > desired_size) { 665 s = splbio(); 666 if ((nbp = bufqueues[BQ_EMPTY].tqh_first) == NULL) { 667 /* No free buffer head */ 668 splx(s); 669 goto out; 670 } 671 bremfree(nbp); 672 SET(nbp->b_flags, B_BUSY); 673 splx(s); 674 675 /* move the page to it and note this change */ 676 pagemove(bp->b_data + desired_size, 677 nbp->b_data, bp->b_bufsize - desired_size); 678 nbp->b_bufsize = bp->b_bufsize - desired_size; 679 bp->b_bufsize = desired_size; 680 nbp->b_bcount = 0; 681 SET(nbp->b_flags, B_INVAL); 682 683 /* release the newly-filled buffer and leave */ 684 brelse(nbp); 685 } 686 687 out: 688 bp->b_bcount = size; 689 } 690 691 /* 692 * Find a buffer which is available for use. 693 * Select something from a free list. 694 * Preference is to AGE list, then LRU list. 695 */ 696 struct buf * 697 getnewbuf(slpflag, slptimeo) 698 int slpflag, slptimeo; 699 { 700 register struct buf *bp; 701 int s; 702 703 start: 704 s = splbio(); 705 if ((bp = bufqueues[BQ_AGE].tqh_first) != NULL || 706 (bp = bufqueues[BQ_LRU].tqh_first) != NULL) { 707 bremfree(bp); 708 } else { 709 /* wait for a free buffer of any kind */ 710 needbuffer = 1; 711 tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo); 712 splx(s); 713 return (0); 714 } 715 716 if (ISSET(bp->b_flags, B_VFLUSH)) { 717 /* 718 * This is a delayed write buffer being flushed to disk. Make 719 * sure it gets aged out of the queue when it's finished, and 720 * leave it off the LRU queue. 721 */ 722 CLR(bp->b_flags, B_VFLUSH); 723 SET(bp->b_flags, B_AGE); 724 splx(s); 725 goto start; 726 } 727 728 /* Buffer is no longer on free lists. */ 729 SET(bp->b_flags, B_BUSY); 730 731 /* If buffer was a delayed write, start it, and go back to the top. */ 732 if (ISSET(bp->b_flags, B_DELWRI)) { 733 splx(s); 734 /* 735 * This buffer has gone through the LRU, so make sure it gets 736 * reused ASAP. 737 */ 738 SET(bp->b_flags, B_AGE); 739 bawrite(bp); 740 goto start; 741 } 742 743 /* disassociate us from our vnode, if we had one... */ 744 if (bp->b_vp) 745 brelvp(bp); 746 splx(s); 747 748 /* clear out various other fields */ 749 bp->b_flags = B_BUSY; 750 bp->b_dev = NODEV; 751 bp->b_blkno = bp->b_lblkno = 0; 752 bp->b_iodone = 0; 753 bp->b_error = 0; 754 bp->b_resid = 0; 755 bp->b_bcount = 0; 756 bp->b_dirtyoff = bp->b_dirtyend = 0; 757 bp->b_validoff = bp->b_validend = 0; 758 759 /* nuke any credentials we were holding */ 760 if (bp->b_rcred != NOCRED) { 761 crfree(bp->b_rcred); 762 bp->b_rcred = NOCRED; 763 } 764 if (bp->b_wcred != NOCRED) { 765 crfree(bp->b_wcred); 766 bp->b_wcred = NOCRED; 767 } 768 769 bremhash(bp); 770 return (bp); 771 } 772 773 /* 774 * Wait for operations on the buffer to complete. 775 * When they do, extract and return the I/O's error value. 776 */ 777 int 778 biowait(bp) 779 struct buf *bp; 780 { 781 int s; 782 783 s = splbio(); 784 while (!ISSET(bp->b_flags, B_DONE)) 785 tsleep(bp, PRIBIO + 1, "biowait", 0); 786 splx(s); 787 788 /* check for interruption of I/O (e.g. via NFS), then errors. */ 789 if (ISSET(bp->b_flags, B_EINTR)) { 790 CLR(bp->b_flags, B_EINTR); 791 return (EINTR); 792 } else if (ISSET(bp->b_flags, B_ERROR)) 793 return (bp->b_error ? bp->b_error : EIO); 794 else 795 return (0); 796 } 797 798 /* 799 * Mark I/O complete on a buffer. 800 * 801 * If a callback has been requested, e.g. the pageout 802 * daemon, do so. Otherwise, awaken waiting processes. 803 * 804 * [ Leffler, et al., says on p.247: 805 * "This routine wakes up the blocked process, frees the buffer 806 * for an asynchronous write, or, for a request by the pagedaemon 807 * process, invokes a procedure specified in the buffer structure" ] 808 * 809 * In real life, the pagedaemon (or other system processes) wants 810 * to do async stuff to, and doesn't want the buffer brelse()'d. 811 * (for swap pager, that puts swap buffers on the free lists (!!!), 812 * for the vn device, that puts malloc'd buffers on the free lists!) 813 */ 814 void 815 biodone(bp) 816 struct buf *bp; 817 { 818 if (ISSET(bp->b_flags, B_DONE)) 819 panic("biodone already"); 820 SET(bp->b_flags, B_DONE); /* note that it's done */ 821 822 if (!ISSET(bp->b_flags, B_READ)) /* wake up reader */ 823 vwakeup(bp); 824 825 if (ISSET(bp->b_flags, B_CALL)) { /* if necessary, call out */ 826 CLR(bp->b_flags, B_CALL); /* but note callout done */ 827 (*bp->b_iodone)(bp); 828 } else if (ISSET(bp->b_flags, B_ASYNC)) /* if async, release it */ 829 brelse(bp); 830 else { /* or just wakeup the buffer */ 831 CLR(bp->b_flags, B_WANTED); 832 wakeup(bp); 833 } 834 } 835 836 /* 837 * Return a count of buffers on the "locked" queue. 838 */ 839 int 840 count_lock_queue() 841 { 842 register struct buf *bp; 843 register int n = 0; 844 845 for (bp = bufqueues[BQ_LOCKED].tqh_first; bp; 846 bp = bp->b_freelist.tqe_next) 847 n++; 848 return (n); 849 } 850 851 #ifdef DEBUG 852 /* 853 * Print out statistics on the current allocation of the buffer pool. 854 * Can be enabled to print out on every ``sync'' by setting "syncprt" 855 * in vfs_syscalls.c using sysctl. 856 */ 857 void 858 vfs_bufstats() 859 { 860 int s, i, j, count; 861 register struct buf *bp; 862 register struct bqueues *dp; 863 int counts[MAXBSIZE/CLBYTES+1]; 864 static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" }; 865 866 for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) { 867 count = 0; 868 for (j = 0; j <= MAXBSIZE/CLBYTES; j++) 869 counts[j] = 0; 870 s = splbio(); 871 for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) { 872 counts[bp->b_bufsize/CLBYTES]++; 873 count++; 874 } 875 splx(s); 876 printf("%s: total-%d", bname[i], count); 877 for (j = 0; j <= MAXBSIZE/CLBYTES; j++) 878 if (counts[j] != 0) 879 printf(", %d-%d", j * CLBYTES, counts[j]); 880 printf("\n"); 881 } 882 } 883 #endif /* DEBUG */ 884