![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to vfs_bio.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [Apple Darwin 0.x] / kernel / bsd / vfs|
Return to vfs_bio.c CVS log | Up to [Apple Darwin 0.x] / kernel / bsd / vfs |
1.1 root 1: /*
2: * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
3: *
4: * @APPLE_LICENSE_HEADER_START@
5: *
6: * Portions Copyright (c) 1999 Apple Computer, Inc. All Rights
7: * Reserved. This file contains Original Code and/or Modifications of
8: * Original Code as defined in and that are subject to the Apple Public
9: * Source License Version 1.1 (the "License"). You may not use this file
10: * except in compliance with the License. Please obtain a copy of the
11: * License at http://www.apple.com/publicsource and read it before using
12: * this file.
13: *
14: * The Original Code and all software distributed under the License are
15: * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
16: * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
17: * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
18: * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
19: * License for the specific language governing rights and limitations
20: * under the License.
21: *
22: * @APPLE_LICENSE_HEADER_END@
23: */
24:
25: /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
26: /*-
27: * Copyright (c) 1982, 1986, 1989, 1993
28: * The Regents of the University of California. All rights reserved.
29: * (c) UNIX System Laboratories, Inc.
30: * All or some portions of this file are derived from material licensed
31: * to the University of California by American Telephone and Telegraph
32: * Co. or Unix System Laboratories, Inc. and are reproduced herein with
33: * the permission of UNIX System Laboratories, Inc.
34: *
35: * Redistribution and use in source and binary forms, with or without
36: * modification, are permitted provided that the following conditions
37: * are met:
38: * 1. Redistributions of source code must retain the above copyright
39: * notice, this list of conditions and the following disclaimer.
40: * 2. Redistributions in binary form must reproduce the above copyright
41: * notice, this list of conditions and the following disclaimer in the
42: * documentation and/or other materials provided with the distribution.
43: * 3. All advertising materials mentioning features or use of this software
44: * must display the following acknowledgement:
45: * This product includes software developed by the University of
46: * California, Berkeley and its contributors.
47: * 4. Neither the name of the University nor the names of its contributors
48: * may be used to endorse or promote products derived from this software
49: * without specific prior written permission.
50: *
51: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61: * SUCH DAMAGE.
62: *
63: * The NEXTSTEP Software License Agreement specifies the terms
64: * and conditions for redistribution.
65: *
66: * @(#)vfs_bio.c 8.6 (Berkeley) 1/11/94
67: */
68:
69: /*
70: * Some references:
71: * Bach: The Design of the UNIX Operating System (Prentice Hall, 1986)
72: * Leffler, et al.: The Design and Implementation of the 4.3BSD
73: * UNIX Operating System (Addison Welley, 1989)
74: */
75: /*
76: * HISTORY
77: * 17-July-97 Umesh Vaishampayan ([email protected])
78: * Eliminated multiple definition of buffers and buf which are defined in
79: * conf/param.c.
80: * Eliminated multiple definition of nbuf and bufpages which are defined
81: * in machdep/XXX/unix_startup.c
82: *
83: * 11-July-97 Umesh Vaishampayan ([email protected])
84: * Defined global variables for use when tracing is turned on.
85: */
86:
87: #include <mach_nbc.h>
88: #include <sys/param.h>
89: #include <sys/systm.h>
90: #include <sys/proc.h>
91: #include <sys/buf.h>
92: #include <sys/vnode.h>
93: #include <sys/mount.h>
94: #include <sys/trace.h>
95: #include <sys/malloc.h>
96: #include <sys/resourcevar.h>
97: #include <miscfs/specfs/specdev.h>
98:
99: extern void reassignbuf(struct buf *, struct vnode *);
100:
101: extern int nbuf; /* The number of buffer headers */
102: extern int niobuf;
103: extern struct buf *buf; /* The buffer headers. */
104: extern char *buffers; /* The buffer contents. */
105: extern int bufpages; /* Number of memory pages in the buffer pool. */
106: struct buf *swbuf; /* Swap I/O buffer headers. */
107: int nswbuf; /* Number of swap I/O buffer headers. */
108: struct buf bswlist; /* Head of swap I/O buffer headers free list. */
109: struct buf *bclnlist;/* Head of cleaned page list. */
110:
111: #if TRACE
112: struct proc *traceproc;
113: int tracewhich, tracebuf[TRCSIZ];
114: u_int tracex;
115: char traceflags[TR_NFLAGS];
116: #endif /* TRACE */
117:
118: /* Macros to clear/set/test flags. */
119: #define SET(t, f) (t) |= (f)
120: #define CLR(t, f) (t) &= ~(f)
121: #define ISSET(t, f) ((t) & (f))
122:
123: /*
124: * Definitions for the buffer hash lists.
125: */
126: #define BUFHASH(dvp, lbn) \
127: (&bufhashtbl[((long)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
128: LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
129: u_long bufhash;
130:
131: /*
132: * Insq/Remq for the buffer hash lists.
133: */
134: #define binshash(bp, dp) LIST_INSERT_HEAD(dp, bp, b_hash)
135: #define bremhash(bp) LIST_REMOVE(bp, b_hash)
136:
137: /*
138: * Definitions for the buffer free lists.
139: */
140: #define BQUEUES 4 /* number of free buffer queues */
141:
142: #define BQ_LOCKED 0 /* super-blocks &c */
143: #define BQ_LRU 1 /* lru, useful buffers */
144: #define BQ_AGE 2 /* rubbish */
145: #define BQ_EMPTY 3 /* buffer headers with no memory */
146:
147: TAILQ_HEAD(ioqueue, buf) iobufqueue;
148: TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
149: int needbuffer;
150:
151: /*
152: * Insq/Remq for the buffer free lists.
153: */
154: #define binsheadfree(bp, dp) TAILQ_INSERT_HEAD(dp, bp, b_freelist)
155: #define binstailfree(bp, dp) TAILQ_INSERT_TAIL(dp, bp, b_freelist)
156:
157: void
158: bremfree(bp)
159: struct buf *bp;
160: {
161: struct bqueues *dp = NULL;
162:
163: /*
164: * We only calculate the head of the freelist when removing
165: * the last element of the list as that is the only time that
166: * it is needed (e.g. to reset the tail pointer).
167: *
168: * NB: This makes an assumption about how tailq's are implemented.
169: */
170: if (bp->b_freelist.tqe_next == NULL) {
171: for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
172: if (dp->tqh_last == &bp->b_freelist.tqe_next)
173: break;
174: if (dp == &bufqueues[BQUEUES])
175: panic("bremfree: lost tail");
176: }
177: TAILQ_REMOVE(dp, bp, b_freelist);
178: }
179:
180: /*
181: * Initialize buffers and hash links for buffers.
182: */
183: void
184: bufinit()
185: {
186: register struct buf *bp;
187: struct bqueues *dp;
188: register int i;
189: int base, residual;
190:
191: for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
192: TAILQ_INIT(dp);
193: bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash);
194: base = bufpages / nbuf;
195: residual = bufpages % nbuf;
196: for (i = 0; i < nbuf; i++) {
197: bp = &buf[i];
198: bzero((char *)bp, sizeof *bp);
199: bp->b_dev = NODEV;
200: bp->b_rcred = NOCRED;
201: bp->b_wcred = NOCRED;
202: bp->b_vnbufs.le_next = NOLIST;
203: bp->b_data = buffers + i * MAXBSIZE;
204: if (i < residual)
205: bp->b_bufsize = (base + 1) * CLBYTES;
206: else
207: bp->b_bufsize = base * CLBYTES;
208: bp->b_flags = B_INVAL;
209: dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY];
210: binsheadfree(bp, dp);
211: binshash(bp, &invalhash);
212: }
213: base = (int )(buffers + (i * MAXBSIZE));
214:
215: for (; i < nbuf + niobuf; i++) {
216: bp = &buf[i];
217: bzero((char *)bp, sizeof *bp);
218: bp->b_dev = NODEV;
219: bp->b_rcred = NOCRED;
220: bp->b_wcred = NOCRED;
221: bp->b_vnbufs.le_next = NOLIST;
222: bp->b_data = (char *)base;
223: bp->b_bufsize = 0;
224: bp->b_flags = B_INVAL;
225: binsheadfree(bp, &iobufqueue);
226:
227: base += MAXPHYSIO;
228: }
229: }
230:
231: __inline struct buf *
232: bio_doread(vp, blkno, size, cred, async)
233: struct vnode *vp;
234: daddr_t blkno;
235: int size;
236: struct ucred *cred;
237: int async;
238: {
239: register struct buf *bp;
240: struct proc *p = current_proc();
241:
242: bp = getblk(vp, blkno, size, 0, 0);
243:
244: /*
245: * If buffer does not have data valid, start a read.
246: * Note that if buffer is B_INVAL, getblk() won't return it.
247: * Therefore, it's valid if it's I/O has completed or been delayed.
248: */
249: if (!ISSET(bp->b_flags, (B_DONE | B_DELWRI))) {
250: /* Start I/O for the buffer (keeping credentials). */
251: SET(bp->b_flags, B_READ | async);
252: if (cred != NOCRED && bp->b_rcred == NOCRED) {
253: crhold(cred);
254: bp->b_rcred = cred;
255: }
256: VOP_STRATEGY(bp);
257:
258: trace(TR_BREADMISS, pack(vp, size), blkno);
259:
260: /* Pay for the read. */
261: if (p && p->p_stats)
262: p->p_stats->p_ru.ru_inblock++; /* XXX */
263: } else if (async) {
264: brelse(bp);
265: }
266:
267: trace(TR_BREADHIT, pack(vp, size), blkno);
268:
269: return (bp);
270: }
271:
272: /*
273: * Read a disk block.
274: * This algorithm described in Bach (p.54).
275: */
276: int
277: bread(vp, blkno, size, cred, bpp)
278: struct vnode *vp;
279: daddr_t blkno;
280: int size;
281: struct ucred *cred;
282: struct buf **bpp;
283: {
284: register struct buf *bp;
285:
286: /* Get buffer for block. */
287: bp = *bpp = bio_doread(vp, blkno, size, cred, 0);
288:
289: /* Wait for the read to complete, and return result. */
290: return (biowait(bp));
291: }
292:
293: /*
294: * Read-ahead multiple disk blocks. The first is sync, the rest async.
295: * Trivial modification to the breada algorithm presented in Bach (p.55).
296: */
297: int
298: breadn(vp, blkno, size, rablks, rasizes, nrablks, cred, bpp)
299: struct vnode *vp;
300: daddr_t blkno; int size;
301: daddr_t rablks[]; int rasizes[];
302: int nrablks;
303: struct ucred *cred;
304: struct buf **bpp;
305: {
306: register struct buf *bp;
307: int i;
308:
309: bp = *bpp = bio_doread(vp, blkno, size, cred, 0);
310:
311: /*
312: * For each of the read-ahead blocks, start a read, if necessary.
313: */
314: for (i = 0; i < nrablks; i++) {
315: /* If it's in the cache, just go on to next one. */
316: if (incore(vp, rablks[i]))
317: continue;
318:
319: /* Get a buffer for the read-ahead block */
320: (void) bio_doread(vp, rablks[i], rasizes[i], cred, B_ASYNC);
321: }
322:
323: /* Otherwise, we had to start a read for it; wait until it's valid. */
324: return (biowait(bp));
325: }
326:
327: /*
328: * Read with single-block read-ahead. Defined in Bach (p.55), but
329: * implemented as a call to breadn().
330: * XXX for compatibility with old file systems.
331: */
332: int
333: breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
334: struct vnode *vp;
335: daddr_t blkno; int size;
336: daddr_t rablkno; int rabsize;
337: struct ucred *cred;
338: struct buf **bpp;
339: {
340:
341: return (breadn(vp, blkno, size, &rablkno, &rabsize, 1, cred, bpp));
342: }
343:
344: /*
345: * Block write. Described in Bach (p.56)
346: */
347: int
348: bwrite(bp)
349: struct buf *bp;
350: {
351: int rv, sync, wasdelayed;
352: struct proc *p = current_proc();
353:
354: /* Remember buffer type, to switch on it later. */
355: sync = !ISSET(bp->b_flags, B_ASYNC);
356: wasdelayed = ISSET(bp->b_flags, B_DELWRI);
357: CLR(bp->b_flags, (B_READ | B_DONE | B_ERROR | B_DELWRI));
358:
359: if (!sync) {
360: /*
361: * If not synchronous, pay for the I/O operation and make
362: * sure the buf is on the correct vnode queue. We have
363: * to do this now, because if we don't, the vnode may not
364: * be properly notified that its I/O has completed.
365: */
366: if (wasdelayed)
367: reassignbuf(bp, bp->b_vp);
368: else
369: if (p && p->p_stats)
370: p->p_stats->p_ru.ru_oublock++; /* XXX */
371: }
372:
373: trace(TR_BWRITE, pack(bp->b_vp, bp->b_bcount), bp->b_lblkno);
374:
375: /* Initiate disk write. Make sure the appropriate party is charged. */
376: SET(bp->b_flags, B_WRITEINPROG);
377: bp->b_vp->v_numoutput++;
378: VOP_STRATEGY(bp);
379:
380: if (sync) {
381: /*
382: * If I/O was synchronous, wait for it to complete.
383: */
384: rv = biowait(bp);
385:
386: /*
387: * Pay for the I/O operation, if it's not been paid for, and
388: * make sure it's on the correct vnode queue. (async operatings
389: * were payed for above.)
390: */
391: if (wasdelayed)
392: reassignbuf(bp, bp->b_vp);
393: else
394: if (p && p->p_stats)
395: p->p_stats->p_ru.ru_oublock++; /* XXX */
396:
397: /* Release the buffer. */
398: brelse(bp);
399:
400: return (rv);
401: } else {
402: return (0);
403: }
404: }
405:
406: int
407: vn_bwrite(ap)
408: struct vop_bwrite_args *ap;
409: {
410:
411: return (bwrite(ap->a_bp));
412: }
413:
414: /*
415: * Delayed write.
416: *
417: * The buffer is marked dirty, but is not queued for I/O.
418: * This routine should be used when the buffer is expected
419: * to be modified again soon, typically a small write that
420: * partially fills a buffer.
421: *
422: * NB: magnetic tapes cannot be delayed; they must be
423: * written in the order that the writes are requested.
424: *
425: * Described in Leffler, et al. (pp. 208-213).
426: */
427: void
428: bdwrite(bp)
429: struct buf *bp;
430: {
431: struct proc *p = current_proc();
432:
433: /*
434: * If the block hasn't been seen before:
435: * (1) Mark it as having been seen,
436: * (2) Charge for the write.
437: * (3) Make sure it's on its vnode's correct block list,
438: */
439: if (!ISSET(bp->b_flags, B_DELWRI)) {
440: SET(bp->b_flags, B_DELWRI);
441: if (p && p->p_stats)
442: p->p_stats->p_ru.ru_oublock++; /* XXX */
443: reassignbuf(bp, bp->b_vp);
444: }
445:
446: /* If this is a tape block, write it the block now. */
447: if (ISSET(bp->b_flags, B_TAPE)) {
448: bwrite(bp);
449: return;
450: }
451:
452: /* Otherwise, the "write" is done, so mark and release the buffer. */
453: SET(bp->b_flags, B_DONE);
454: brelse(bp);
455: }
456:
457: /*
458: * Asynchronous block write; just an asynchronous bwrite().
459: */
460: void
461: bawrite(bp)
462: struct buf *bp;
463: {
464:
465: SET(bp->b_flags, B_ASYNC);
466: VOP_BWRITE(bp);
467: }
468:
469: /*
470: * Release a buffer on to the free lists.
471: * Described in Bach (p. 46).
472: */
473: void
474: brelse(bp)
475: struct buf *bp;
476: {
477: struct bqueues *bufq;
478: int s;
479:
480: trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
481:
482: /* Wake up any processes waiting for any buffer to become free. */
483: if (needbuffer) {
484: needbuffer = 0;
485: wakeup(&needbuffer);
486: }
487:
488: /* Wake up any proceeses waiting for _this_ buffer to become free. */
489: if (ISSET(bp->b_flags, B_WANTED)) {
490: CLR(bp->b_flags, B_WANTED);
491: wakeup(bp);
492: }
493:
494: /* Block disk interrupts. */
495: s = splbio();
496:
497: /*
498: * Determine which queue the buffer should be on, then put it there.
499: */
500:
501: /* If it's locked, don't report an error; try again later. */
502: if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR))
503: CLR(bp->b_flags, B_ERROR);
504:
505: /* If it's not cacheable, or an error, mark it invalid. */
506: if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR)))
507: SET(bp->b_flags, B_INVAL);
508:
509: if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) {
510: /*
511: * If it's invalid or empty, dissociate it from its vnode
512: * and put on the head of the appropriate queue.
513: */
514: if (bp->b_vp)
515: brelvp(bp);
516: CLR(bp->b_flags, B_DELWRI);
517: if (bp->b_bufsize <= 0)
518: /* no data */
519: bufq = &bufqueues[BQ_EMPTY];
520: else
521: /* invalid data */
522: bufq = &bufqueues[BQ_AGE];
523: binsheadfree(bp, bufq);
524: } else {
525: /*
526: * It has valid data. Put it on the end of the appropriate
527: * queue, so that it'll stick around for as long as possible.
528: */
529: if (ISSET(bp->b_flags, B_LOCKED))
530: /* locked in core */
531: bufq = &bufqueues[BQ_LOCKED];
532: else if (ISSET(bp->b_flags, B_AGE))
533: /* stale but valid data */
534: bufq = &bufqueues[BQ_AGE];
535: else
536: /* valid data */
537: bufq = &bufqueues[BQ_LRU];
538: binstailfree(bp, bufq);
539: }
540:
541: /* Unlock the buffer. */
542: CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE));
543:
544: /* Allow disk interrupts. */
545: splx(s);
546: }
547:
548: /*
549: * Determine if a block is in the cache.
550: * Just look on what would be its hash chain. If it's there, return
551: * a pointer to it, unless it's marked invalid. If it's marked invalid,
552: * we normally don't return the buffer, unless the caller explicitly
553: * wants us to.
554: */
555: struct buf *
556: incore(vp, blkno)
557: struct vnode *vp;
558: daddr_t blkno;
559: {
560: struct buf *bp;
561:
562: bp = BUFHASH(vp, blkno)->lh_first;
563:
564: /* Search hash chain */
565: for (; bp != NULL; bp = bp->b_hash.le_next) {
566: if (bp->b_lblkno == blkno && bp->b_vp == vp &&
567: !ISSET(bp->b_flags, B_INVAL))
568: return (bp);
569: }
570:
571: return (0);
572: }
573:
574: /*
575: * Get a block of requested size that is associated with
576: * a given vnode and block offset. If it is found in the
577: * block cache, mark it as having been found, make it busy
578: * and return it. Otherwise, return an empty block of the
579: * correct size. It is up to the caller to insure that the
580: * cached blocks be of the correct size.
581: */
582: struct buf *
583: getblk(vp, blkno, size, slpflag, slptimeo)
584: register struct vnode *vp;
585: daddr_t blkno;
586: int size, slpflag, slptimeo;
587: {
588: struct buf *bp;
589: int s, err;
590:
591: start:
592: s = splbio();
593: if (bp = incore(vp, blkno)) { /* XXX NFS VOP_BWRITE foolishness */
594: if (ISSET(bp->b_flags, B_BUSY)) {
595: SET(bp->b_flags, B_WANTED);
596: err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk",
597: slptimeo);
598: splx(s);
599: if (err)
600: return (NULL);
601: goto start;
602: }
603: SET(bp->b_flags, (B_BUSY | B_CACHE));
604: bremfree(bp);
605: splx(s);
606: allocbuf(bp, size);
607: } else {
608: splx(s);
609: if ((bp = getnewbuf(slpflag, slptimeo)) == NULL)
610: goto start;
611: binshash(bp, BUFHASH(vp, blkno));
612: allocbuf(bp, size);
613: bp->b_blkno = bp->b_lblkno = blkno;
614: s = splbio();
615: bgetvp(vp, bp);
616: splx(s);
617: }
618: return (bp);
619: }
620:
621: /*
622: * Get an empty, disassociated buffer of given size.
623: */
624: struct buf *
625: geteblk(size)
626: int size;
627: {
628: struct buf *bp;
629:
630: while ((bp = getnewbuf(0, 0)) == 0)
631: ;
632: SET(bp->b_flags, B_INVAL);
633: binshash(bp, &invalhash);
634: allocbuf(bp, size);
635:
636: return (bp);
637: }
638:
639: /*
640: * Expand or contract the actual memory allocated to a buffer.
641: *
642: * If the buffer shrinks, data is lost, so it's up to the
643: * caller to have written it out *first*; this routine will not
644: * start a write. If the buffer grows, it's the callers
645: * responsibility to fill out the buffer's additional contents.
646: */
647: int
648: allocbuf(bp, size)
649: struct buf *bp;
650: int size;
651: {
652: struct buf *nbp;
653: vm_size_t desired_size;
654: int s;
655:
656: desired_size = roundup(size, CLBYTES);
657: if (desired_size > MAXBSIZE)
658: panic("allocbuf: buffer larger than MAXBSIZE requested");
659:
660: if (bp->b_bufsize == desired_size)
661: goto out;
662:
663: /*
664: * If the buffer is smaller than the desired size, we need to snarf
665: * it from other buffers. Get buffers (via getnewbuf()), and
666: * steal their pages.
667: */
668: while (bp->b_bufsize < desired_size) {
669: int amt;
670:
671: /* find a buffer */
672: while ((nbp = getnewbuf(0, 0)) == NULL)
673: ;
674: SET(nbp->b_flags, B_INVAL);
675: binshash(nbp, &invalhash);
676:
677: /* and steal its pages, up to the amount we need */
678: amt = min(nbp->b_bufsize, (desired_size - bp->b_bufsize));
679: pagemove((nbp->b_data + nbp->b_bufsize - amt),
680: bp->b_data + bp->b_bufsize, amt);
681: bp->b_bufsize += amt;
682: nbp->b_bufsize -= amt;
683:
684: /* reduce transfer count if we stole some data */
685: if (nbp->b_bcount > nbp->b_bufsize)
686: nbp->b_bcount = nbp->b_bufsize;
687:
688: #if DIAGNOSTIC
689: if (nbp->b_bufsize < 0)
690: panic("allocbuf: negative bufsize");
691: #endif
692:
693: brelse(nbp);
694: }
695:
696: /*
697: * If we want a buffer smaller than the current size,
698: * shrink this buffer. Grab a buf head from the EMPTY queue,
699: * move a page onto it, and put it on front of the AGE queue.
700: * If there are no free buffer headers, leave the buffer alone.
701: */
702: if (bp->b_bufsize > desired_size) {
703: s = splbio();
704: if ((nbp = bufqueues[BQ_EMPTY].tqh_first) == NULL) {
705: /* No free buffer head */
706: splx(s);
707: goto out;
708: }
709: bremfree(nbp);
710: SET(nbp->b_flags, B_BUSY);
711: splx(s);
712:
713: /* move the page to it and note this change */
714: pagemove(bp->b_data + desired_size,
715: nbp->b_data, bp->b_bufsize - desired_size);
716: nbp->b_bufsize = bp->b_bufsize - desired_size;
717: bp->b_bufsize = desired_size;
718: nbp->b_bcount = 0;
719: SET(nbp->b_flags, B_INVAL);
720:
721: /* release the newly-filled buffer and leave */
722: brelse(nbp);
723: }
724:
725: out:
726: bp->b_bcount = size;
727: }
728:
729: /*
730: * Find a buffer which is available for use.
731: * Select something from a free list.
732: * Preference is to AGE list, then LRU list.
733: */
734: struct buf *
735: getnewbuf(slpflag, slptimeo)
736: int slpflag, slptimeo;
737: {
738: register struct buf *bp;
739: int s;
740: struct ucred *cred;
741:
742: start:
743: s = splbio();
744: if ((bp = bufqueues[BQ_AGE].tqh_first) != NULL ||
745: (bp = bufqueues[BQ_LRU].tqh_first) != NULL) {
746: bremfree(bp);
747: } else {
748: /* wait for a free buffer of any kind */
749: needbuffer = 1;
750: tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo);
751: splx(s);
752: return (0);
753: }
754:
755: /* Buffer is no longer on free lists. */
756: SET(bp->b_flags, B_BUSY);
757: splx(s);
758:
759: /* If buffer was a delayed write, start it, and go back to the top. */
760: if (ISSET(bp->b_flags, B_DELWRI)) {
761: bawrite (bp);
762: goto start;
763: }
764:
765: trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
766:
767: /* disassociate us from our vnode, if we had one... */
768: s = splbio();
769: if (bp->b_vp)
770: brelvp(bp);
771: splx(s);
772:
773: /* clear out various other fields */
774: bp->b_flags = B_BUSY;
775: bp->b_dev = NODEV;
776: bp->b_blkno = bp->b_lblkno = 0;
777: bp->b_iodone = 0;
778: bp->b_error = 0;
779: bp->b_resid = 0;
780: bp->b_bcount = 0;
781: bp->b_dirtyoff = bp->b_dirtyend = 0;
782: bp->b_validoff = bp->b_validend = 0;
783:
784: /* nuke any credentials we were holding */
785: cred = bp->b_rcred;
786: if (cred != NOCRED) {
787: bp->b_rcred = NOCRED;
788: crfree(cred);
789: }
790: cred = bp->b_wcred;
791: if (cred != NOCRED) {
792: bp->b_wcred = NOCRED;
793: crfree(cred);
794: }
795:
796: bremhash(bp);
797: return (bp);
798: }
799:
800: /*
801: * Wait for operations on the buffer to complete.
802: * When they do, extract and return the I/O's error value.
803: */
804: int
805: biowait(bp)
806: struct buf *bp;
807: {
808: int s;
809:
810: s = splbio();
811: while (!ISSET(bp->b_flags, B_DONE))
812: tsleep(bp, PRIBIO + 1, "biowait", 0);
813: splx(s);
814:
815: /* check for interruption of I/O (e.g. via NFS), then errors. */
816: if (ISSET(bp->b_flags, B_EINTR)) {
817: CLR(bp->b_flags, B_EINTR);
818: return (EINTR);
819: } else if (ISSET(bp->b_flags, B_ERROR))
820: return (bp->b_error ? bp->b_error : EIO);
821: else
822: return (0);
823: }
824:
825: /*
826: * Mark I/O complete on a buffer.
827: *
828: * If a callback has been requested, e.g. the pageout
829: * daemon, do so. Otherwise, awaken waiting processes.
830: *
831: * [ Leffler, et al., says on p.247:
832: * "This routine wakes up the blocked process, frees the buffer
833: * for an asynchronous write, or, for a request by the pagedaemon
834: * process, invokes a procedure specified in the buffer structure" ]
835: *
836: * In real life, the pagedaemon (or other system processes) wants
837: * to do async stuff to, and doesn't want the buffer brelse()'d.
838: * (for swap pager, that puts swap buffers on the free lists (!!!),
839: * for the vn device, that puts malloc'd buffers on the free lists!)
840: */
841: void
842: biodone(bp)
843: struct buf *bp;
844: {
845: if (ISSET(bp->b_flags, B_DONE))
846: panic("biodone already");
847: SET(bp->b_flags, B_DONE); /* note that it's done */
848:
849: if (!ISSET(bp->b_flags, B_READ) && !ISSET(bp->b_flags, B_RAW)) /* wake up reader */
850: vwakeup(bp);
851:
852: if (ISSET(bp->b_flags, B_CALL)) { /* if necessary, call out */
853: CLR(bp->b_flags, B_CALL); /* but note callout done */
854: (*bp->b_iodone)(bp);
855: } else if (ISSET(bp->b_flags, B_ASYNC)) /* if async, release it */
856: brelse(bp);
857: else { /* or just wakeup the buffer */
858: CLR(bp->b_flags, B_WANTED);
859: wakeup(bp);
860: }
861: }
862:
863: /*
864: * Return a count of buffers on the "locked" queue.
865: */
866: int
867: count_lock_queue()
868: {
869: register struct buf *bp;
870: register int n = 0;
871:
872: for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
873: bp = bp->b_freelist.tqe_next)
874: n++;
875: return (n);
876: }
877:
878: #if MACH_NBC
879: #include <ufs/ufs/quota.h>
880: #include <ufs/ufs/inode.h>
881:
882: #define btodevblk(b) ((b) / devBlocksize)
883: void
884: blkflush(struct vnode *vp, daddr_t blkno, vm_size_t size)
885: {
886: register struct buf *ep, *nbp;
887: daddr_t start, last;
888: int s,err;
889: struct inode *ip= VTOI(vp);
890: int devBlocksize=1024;
891:
892: #if 1
893: VOP_DEVBLOCKSIZE(ip->i_devvp, &devBlocksize);
894: #endif
895:
896:
897: start = blkno;
898: last = start + btodb(size, devBlocksize) - 1;
899: loop:
900: for(ep = vp->v_dirtyblkhd.lh_first; ep; ep = nbp) {
901: nbp = ep->b_vnbufs.le_next;
902: if (ep->b_vp != vp || ISSET(ep->b_flags, B_INVAL))
903: continue;
904: /* look for overlap */
905: if (ep->b_bcount == 0 || ep->b_blkno > last ||
906: ep->b_blkno + btodevblk(ep->b_bcount) <= start)
907: continue;
908: s = splbio();
909: if (ISSET(ep->b_flags, B_BUSY)) {
910: SET(ep->b_flags, B_WANTED);
911: err = tsleep(ep, (PRIBIO + 1), "blkflush",
912: 0);
913: splx(s);
914: goto loop;
915: }
916: if(ISSET(ep->b_flags, B_DELWRI)) {
917: bremfree(ep);
918: SET(ep->b_flags, B_BUSY);
919: (void) splx(s);
920: bwrite(ep);
921: goto loop;
922: }
923: (void) splx(s);
924: }
925:
926: }
927: #endif /* MACH_NBC */
928: #if DIAGNOSTIC
929: /*
930: * Print out statistics on the current allocation of the buffer pool.
931: * Can be enabled to print out on every ``sync'' by setting "syncprt"
932: * in vfs_syscalls.c using sysctl.
933: */
934: void
935: vfs_bufstats()
936: {
937: int s, i, j, count;
938: register struct buf *bp;
939: register struct bqueues *dp;
940: int counts[MAXBSIZE/CLBYTES+1];
941: static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" };
942:
943: for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) {
944: count = 0;
945: for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
946: counts[j] = 0;
947: s = splbio();
948: for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) {
949: counts[bp->b_bufsize/CLBYTES]++;
950: count++;
951: }
952: splx(s);
953: printf("%s: total-%d", bname[i], count);
954: for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
955: if (counts[j] != 0)
956: printf(", %d-%d", j * CLBYTES, counts[j]);
957: printf("\n");
958: }
959: }
960: #endif /* DIAGNOSTIC */
961:
962:
963: struct buf *
964: alloc_io_buf(vp)
965: struct vnode *vp;
966: { register struct buf *bp;
967: int s;
968:
969: s = splbio();
970:
971: if ((bp = iobufqueue.tqh_first) == NULL) {
972: splx(s);
973: return (NULL);
974: }
975: TAILQ_REMOVE(&iobufqueue, bp, b_freelist);
976:
977: /* clear out various fields */
978: bp->b_flags = (B_BUSY | B_RAW);
979: bp->b_blkno = bp->b_lblkno = 0;
980: bp->b_iodone = 0;
981: bp->b_error = 0;
982: bp->b_resid = 0;
983: bp->b_bcount = 0;
984: bp->b_bufsize = 0;
985: bp->b_vp = vp;
986:
987: if (vp->v_type == VBLK || vp->v_type == VCHR)
988: bp->b_dev = vp->v_rdev;
989: else
990: bp->b_dev = NODEV;
991: splx(s);
992:
993: return (bp);
994: }
995:
996: void
997: free_io_buf(bp)
998: struct buf *bp;
999: {
1000: int s;
1001:
1002: s = splbio();
1003: /*
1004: * put buffer back on the head of the iobufqueue
1005: */
1006: bp->b_vp = NULL;
1007: bp->b_flags = B_INVAL;
1008:
1009: binsheadfree(bp, &iobufqueue);
1010:
1011: splx(s);
1012: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.