![[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 XNU] / XNU / bsd / vfs|
Return to vfs_bio.c CVS log | Up to [Apple XNU] / XNU / bsd / vfs |
1.1 root 1: /*
2: * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
3: *
4: * @APPLE_LICENSE_HEADER_START@
5: *
6: * The contents of this file constitute Original Code as defined in and
7: * are subject to the Apple Public Source License Version 1.1 (the
8: * "License"). You may not use this file except in compliance with the
9: * License. Please obtain a copy of the License at
10: * http://www.apple.com/publicsource and read it before using this file.
11: *
12: * This Original Code and all software distributed under the License are
13: * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14: * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15: * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16: * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17: * License for the specific language governing rights and limitations
18: * under the License.
19: *
20: * @APPLE_LICENSE_HEADER_END@
21: */
22: /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
23: /*-
24: * Copyright (c) 1994 Christopher G. Demetriou
25: * Copyright (c) 1982, 1986, 1989, 1993
26: * The Regents of the University of California. All rights reserved.
27: * (c) UNIX System Laboratories, Inc.
28: * All or some portions of this file are derived from material licensed
29: * to the University of California by American Telephone and Telegraph
30: * Co. or Unix System Laboratories, Inc. and are reproduced herein with
31: * the permission of UNIX System Laboratories, Inc.
32: *
33: * Redistribution and use in source and binary forms, with or without
34: * modification, are permitted provided that the following conditions
35: * are met:
36: * 1. Redistributions of source code must retain the above copyright
37: * notice, this list of conditions and the following disclaimer.
38: * 2. Redistributions in binary form must reproduce the above copyright
39: * notice, this list of conditions and the following disclaimer in the
40: * documentation and/or other materials provided with the distribution.
41: * 3. All advertising materials mentioning features or use of this software
42: * must display the following acknowledgement:
43: * This product includes software developed by the University of
44: * California, Berkeley and its contributors.
45: * 4. Neither the name of the University nor the names of its contributors
46: * may be used to endorse or promote products derived from this software
47: * without specific prior written permission.
48: *
49: * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50: * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51: * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52: * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53: * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54: * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55: * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57: * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59: * SUCH DAMAGE.
60: *
61: * The NEXTSTEP Software License Agreement specifies the terms
62: * and conditions for redistribution.
63: *
64: * @(#)vfs_bio.c 8.6 (Berkeley) 1/11/94
65: */
66:
67: /*
68: * Some references:
69: * Bach: The Design of the UNIX Operating System (Prentice Hall, 1986)
70: * Leffler, et al.: The Design and Implementation of the 4.3BSD
71: * UNIX Operating System (Addison Welley, 1989)
72: */
73: /*
74: * HISTORY
75: * 17-July-97 Umesh Vaishampayan ([email protected])
76: * Eliminated multiple definition of buffers and buf which are defined in
77: * conf/param.c.
78: * Eliminated multiple definition of nbuf and bufpages which are defined
79: * in machdep/XXX/unix_startup.c
80: *
81: * 11-July-97 Umesh Vaishampayan ([email protected])
82: * Defined global variables for use when tracing is turned on.
83: */
84:
85: #include <mach_nbc.h>
86: #include <sys/param.h>
87: #include <sys/systm.h>
88: #include <sys/proc.h>
89: #include <sys/buf.h>
90: #include <sys/vnode.h>
91: #include <sys/mount.h>
92: #include <sys/trace.h>
93: #include <sys/malloc.h>
94: #include <sys/resourcevar.h>
95: #include <miscfs/specfs/specdev.h>
96:
97: extern void reassignbuf(struct buf *, struct vnode *);
98:
99: extern int nbuf; /* The number of buffer headers */
100: extern int niobuf;
101: extern struct buf *buf; /* The buffer headers. */
102: extern char *buffers; /* The buffer contents. */
103: extern int bufpages; /* Number of memory pages in the buffer pool. */
104: struct buf *swbuf; /* Swap I/O buffer headers. */
105: int nswbuf; /* Number of swap I/O buffer headers. */
106: struct buf bswlist; /* Head of swap I/O buffer headers free list. */
107: struct buf *bclnlist;/* Head of cleaned page list. */
108:
109: #if TRACE
110: struct proc *traceproc;
111: int tracewhich, tracebuf[TRCSIZ];
112: u_int tracex;
113: char traceflags[TR_NFLAGS];
114: #endif /* TRACE */
115:
116: /* Macros to clear/set/test flags. */
117: #define SET(t, f) (t) |= (f)
118: #define CLR(t, f) (t) &= ~(f)
119: #define ISSET(t, f) ((t) & (f))
120:
121: /*
122: * Definitions for the buffer hash lists.
123: */
124: #define BUFHASH(dvp, lbn) \
125: (&bufhashtbl[((long)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
126: LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
127: u_long bufhash;
128:
129: /*
130: * Insq/Remq for the buffer hash lists.
131: */
132: #define binshash(bp, dp) LIST_INSERT_HEAD(dp, bp, b_hash)
133: #define bremhash(bp) LIST_REMOVE(bp, b_hash)
134:
135: /*
136: * Definitions for the buffer free lists.
137: */
138: #define BQUEUES 4 /* number of free buffer queues */
139:
140: #define BQ_LOCKED 0 /* super-blocks &c */
141: #define BQ_LRU 1 /* lru, useful buffers */
142: #define BQ_AGE 2 /* rubbish */
143: #define BQ_EMPTY 3 /* buffer headers with no memory */
144:
145: TAILQ_HEAD(ioqueue, buf) iobufqueue;
146: TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
147: int needbuffer;
148:
149: /*
150: * Insq/Remq for the buffer free lists.
151: */
152: #define binsheadfree(bp, dp) do { \
153: TAILQ_INSERT_HEAD(dp, bp, b_freelist); \
154: (bp)->b_timestamp = time.tv_sec; \
155: } while (0)
156:
157: #define binstailfree(bp, dp) do { \
158: TAILQ_INSERT_TAIL(dp, bp, b_freelist); \
159: (bp)->b_timestamp = time.tv_sec; \
160: } while (0)
161:
162:
163: /* Time in seconds before a buf on a list is considered as a stale buf */
164: #define LRU_IS_STALE 120 /* default value for the LRU */
165: #define AGE_IS_STALE 60 /* default value for the AGE */
166:
167: int lru_is_stale = LRU_IS_STALE;
168: int age_is_stale = AGE_IS_STALE;
169:
170:
171:
172: void
173: bremfree(bp)
174: struct buf *bp;
175: {
176: struct bqueues *dp = NULL;
177:
178: /*
179: * We only calculate the head of the freelist when removing
180: * the last element of the list as that is the only time that
181: * it is needed (e.g. to reset the tail pointer).
182: *
183: * NB: This makes an assumption about how tailq's are implemented.
184: */
185: if (bp->b_freelist.tqe_next == NULL) {
186: for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
187: if (dp->tqh_last == &bp->b_freelist.tqe_next)
188: break;
189: if (dp == &bufqueues[BQUEUES])
190: panic("bremfree: lost tail");
191: }
192: TAILQ_REMOVE(dp, bp, b_freelist);
193: bp->b_timestamp = 0;
194: }
195:
196: /*
197: * Initialize buffers and hash links for buffers.
198: */
199: void
200: bufinit()
201: {
202: register struct buf *bp;
203: struct bqueues *dp;
204: register int i;
205: int base, residual;
206:
207: for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
208: TAILQ_INIT(dp);
209: bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash);
210: base = bufpages / nbuf;
211: residual = bufpages % nbuf;
212: for (i = 0; i < nbuf; i++) {
213: bp = &buf[i];
214: bzero((char *)bp, sizeof *bp);
215: bp->b_dev = NODEV;
216: bp->b_rcred = NOCRED;
217: bp->b_wcred = NOCRED;
218: bp->b_vnbufs.le_next = NOLIST;
219: bp->b_data = buffers + i * MAXBSIZE;
220: if (i < residual)
221: bp->b_bufsize = (base + 1) * CLBYTES;
222: else
223: bp->b_bufsize = base * CLBYTES;
224: bp->b_flags = B_INVAL;
225: dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY];
226: binsheadfree(bp, dp);
227: binshash(bp, &invalhash);
228: }
229: base = (int )(buffers + (i * MAXBSIZE));
230:
231: for (; i < nbuf + niobuf; i++) {
232: bp = &buf[i];
233: bzero((char *)bp, sizeof *bp);
234: bp->b_dev = NODEV;
235: bp->b_rcred = NOCRED;
236: bp->b_wcred = NOCRED;
237: bp->b_vnbufs.le_next = NOLIST;
238: bp->b_data = (char *)base;
239: bp->b_bufsize = 0;
240: bp->b_flags = B_INVAL;
241: binsheadfree(bp, &iobufqueue);
242:
243: base += MAXPHYSIO;
244: }
245: }
246:
247: __inline struct buf *
248: bio_doread(vp, blkno, size, cred, async)
249: struct vnode *vp;
250: daddr_t blkno;
251: int size;
252: struct ucred *cred;
253: int async;
254: {
255: register struct buf *bp;
256: struct proc *p = current_proc();
257:
258: bp = getblk(vp, blkno, size, 0, 0);
259:
260: /*
261: * If buffer does not have data valid, start a read.
262: * Note that if buffer is B_INVAL, getblk() won't return it.
263: * Therefore, it's valid if it's I/O has completed or been delayed.
264: */
265: if (!ISSET(bp->b_flags, (B_DONE | B_DELWRI))) {
266: /* Start I/O for the buffer (keeping credentials). */
267: SET(bp->b_flags, B_READ | async);
268: if (cred != NOCRED && bp->b_rcred == NOCRED) {
269: crhold(cred);
270: bp->b_rcred = cred;
271: }
272: VOP_STRATEGY(bp);
273:
274: trace(TR_BREADMISS, pack(vp, size), blkno);
275:
276: /* Pay for the read. */
277: if (p && p->p_stats)
278: p->p_stats->p_ru.ru_inblock++; /* XXX */
279: } else if (async) {
280: brelse(bp);
281: }
282:
283: trace(TR_BREADHIT, pack(vp, size), blkno);
284:
285: return (bp);
286: }
287:
288: /*
289: * Read a disk block.
290: * This algorithm described in Bach (p.54).
291: */
292: int
293: bread(vp, blkno, size, cred, bpp)
294: struct vnode *vp;
295: daddr_t blkno;
296: int size;
297: struct ucred *cred;
298: struct buf **bpp;
299: {
300: register struct buf *bp;
301:
302: /* Get buffer for block. */
303: bp = *bpp = bio_doread(vp, blkno, size, cred, 0);
304:
305: /* Wait for the read to complete, and return result. */
306: return (biowait(bp));
307: }
308:
309: /*
310: * Read-ahead multiple disk blocks. The first is sync, the rest async.
311: * Trivial modification to the breada algorithm presented in Bach (p.55).
312: */
313: int
314: breadn(vp, blkno, size, rablks, rasizes, nrablks, cred, bpp)
315: struct vnode *vp;
316: daddr_t blkno; int size;
317: daddr_t rablks[]; int rasizes[];
318: int nrablks;
319: struct ucred *cred;
320: struct buf **bpp;
321: {
322: register struct buf *bp;
323: int i;
324:
325: bp = *bpp = bio_doread(vp, blkno, size, cred, 0);
326:
327: /*
328: * For each of the read-ahead blocks, start a read, if necessary.
329: */
330: for (i = 0; i < nrablks; i++) {
331: /* If it's in the cache, just go on to next one. */
332: if (incore(vp, rablks[i]))
333: continue;
334:
335: /* Get a buffer for the read-ahead block */
336: (void) bio_doread(vp, rablks[i], rasizes[i], cred, B_ASYNC);
337: }
338:
339: /* Otherwise, we had to start a read for it; wait until it's valid. */
340: return (biowait(bp));
341: }
342:
343: /*
344: * Read with single-block read-ahead. Defined in Bach (p.55), but
345: * implemented as a call to breadn().
346: * XXX for compatibility with old file systems.
347: */
348: int
349: breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
350: struct vnode *vp;
351: daddr_t blkno; int size;
352: daddr_t rablkno; int rabsize;
353: struct ucred *cred;
354: struct buf **bpp;
355: {
356:
357: return (breadn(vp, blkno, size, &rablkno, &rabsize, 1, cred, bpp));
358: }
359:
360: /*
361: * Block write. Described in Bach (p.56)
362: */
363: int
364: bwrite(bp)
365: struct buf *bp;
366: {
367: int rv, sync, wasdelayed;
368: struct proc *p = current_proc();
369:
370: /* Remember buffer type, to switch on it later. */
371: sync = !ISSET(bp->b_flags, B_ASYNC);
372: wasdelayed = ISSET(bp->b_flags, B_DELWRI);
373: CLR(bp->b_flags, (B_READ | B_DONE | B_ERROR | B_DELWRI));
374:
375: if (!sync) {
376: /*
377: * If not synchronous, pay for the I/O operation and make
378: * sure the buf is on the correct vnode queue. We have
379: * to do this now, because if we don't, the vnode may not
380: * be properly notified that its I/O has completed.
381: */
382: if (wasdelayed)
383: reassignbuf(bp, bp->b_vp);
384: else
385: if (p && p->p_stats)
386: p->p_stats->p_ru.ru_oublock++; /* XXX */
387: }
388:
389: trace(TR_BWRITE, pack(bp->b_vp, bp->b_bcount), bp->b_lblkno);
390:
391: /* Initiate disk write. Make sure the appropriate party is charged. */
392: SET(bp->b_flags, B_WRITEINPROG);
393: bp->b_vp->v_numoutput++;
394: VOP_STRATEGY(bp);
395:
396: if (sync) {
397: /*
398: * If I/O was synchronous, wait for it to complete.
399: */
400: rv = biowait(bp);
401:
402: /*
403: * Pay for the I/O operation, if it's not been paid for, and
404: * make sure it's on the correct vnode queue. (async operatings
405: * were payed for above.)
406: */
407: if (wasdelayed)
408: reassignbuf(bp, bp->b_vp);
409: else
410: if (p && p->p_stats)
411: p->p_stats->p_ru.ru_oublock++; /* XXX */
412:
413: /* Release the buffer. */
414: brelse(bp);
415:
416: return (rv);
417: } else {
418: return (0);
419: }
420: }
421:
422: int
423: vn_bwrite(ap)
424: struct vop_bwrite_args *ap;
425: {
426:
427: return (bwrite(ap->a_bp));
428: }
429:
430: /*
431: * Delayed write.
432: *
433: * The buffer is marked dirty, but is not queued for I/O.
434: * This routine should be used when the buffer is expected
435: * to be modified again soon, typically a small write that
436: * partially fills a buffer.
437: *
438: * NB: magnetic tapes cannot be delayed; they must be
439: * written in the order that the writes are requested.
440: *
441: * Described in Leffler, et al. (pp. 208-213).
442: */
443: void
444: bdwrite(bp)
445: struct buf *bp;
446: {
447: struct proc *p = current_proc();
448:
449: /*
450: * If the block hasn't been seen before:
451: * (1) Mark it as having been seen,
452: * (2) Charge for the write.
453: * (3) Make sure it's on its vnode's correct block list,
454: */
455: if (!ISSET(bp->b_flags, B_DELWRI)) {
456: SET(bp->b_flags, B_DELWRI);
457: if (p && p->p_stats)
458: p->p_stats->p_ru.ru_oublock++; /* XXX */
459: reassignbuf(bp, bp->b_vp);
460: }
461:
462: /* If this is a tape block, write it the block now. */
463: if (ISSET(bp->b_flags, B_TAPE)) {
464: bwrite(bp);
465: return;
466: }
467:
468: /* Otherwise, the "write" is done, so mark and release the buffer. */
469: SET(bp->b_flags, B_DONE);
470: brelse(bp);
471: }
472:
473: /*
474: * Asynchronous block write; just an asynchronous bwrite().
475: */
476: void
477: bawrite(bp)
478: struct buf *bp;
479: {
480:
481: SET(bp->b_flags, B_ASYNC);
482: VOP_BWRITE(bp);
483: }
484:
485: /*
486: * Release a buffer on to the free lists.
487: * Described in Bach (p. 46).
488: */
489: void
490: brelse(bp)
491: struct buf *bp;
492: {
493: struct bqueues *bufq;
494: int s;
495:
496: trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
497:
498: /* Wake up any processes waiting for any buffer to become free. */
499: if (needbuffer) {
500: needbuffer = 0;
501: wakeup(&needbuffer);
502: }
503:
504: /* Wake up any proceeses waiting for _this_ buffer to become free. */
505: if (ISSET(bp->b_flags, B_WANTED)) {
506: CLR(bp->b_flags, B_WANTED);
507: wakeup(bp);
508: }
509:
510: /* Block disk interrupts. */
511: s = splbio();
512:
513: /*
514: * Determine which queue the buffer should be on, then put it there.
515: */
516:
517: /* If it's locked, don't report an error; try again later. */
518: if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR))
519: CLR(bp->b_flags, B_ERROR);
520:
521: /* If it's not cacheable, or an error, mark it invalid. */
522: if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR)))
523: SET(bp->b_flags, B_INVAL);
524:
525: if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) {
526: /*
527: * If it's invalid or empty, dissociate it from its vnode
528: * and put on the head of the appropriate queue.
529: */
530: if (bp->b_vp)
531: brelvp(bp);
532: CLR(bp->b_flags, B_DELWRI);
533: if (bp->b_bufsize <= 0)
534: /* no data */
535: bufq = &bufqueues[BQ_EMPTY];
536: else
537: /* invalid data */
538: bufq = &bufqueues[BQ_AGE];
539: binsheadfree(bp, bufq);
540: } else {
541: /*
542: * It has valid data. Put it on the end of the appropriate
543: * queue, so that it'll stick around for as long as possible.
544: */
545: if (ISSET(bp->b_flags, B_LOCKED))
546: /* locked in core */
547: bufq = &bufqueues[BQ_LOCKED];
548: else if (ISSET(bp->b_flags, B_AGE))
549: /* stale but valid data */
550: bufq = &bufqueues[BQ_AGE];
551: else
552: /* valid data */
553: bufq = &bufqueues[BQ_LRU];
554: binstailfree(bp, bufq);
555: }
556:
557: /* Unlock the buffer. */
558: CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE));
559:
560: /* Allow disk interrupts. */
561: splx(s);
562: }
563:
564: /*
565: * Determine if a block is in the cache.
566: * Just look on what would be its hash chain. If it's there, return
567: * a pointer to it, unless it's marked invalid. If it's marked invalid,
568: * we normally don't return the buffer, unless the caller explicitly
569: * wants us to.
570: */
571: struct buf *
572: incore(vp, blkno)
573: struct vnode *vp;
574: daddr_t blkno;
575: {
576: struct buf *bp;
577:
578: bp = BUFHASH(vp, blkno)->lh_first;
579:
580: /* Search hash chain */
581: for (; bp != NULL; bp = bp->b_hash.le_next) {
582: if (bp->b_lblkno == blkno && bp->b_vp == vp &&
583: !ISSET(bp->b_flags, B_INVAL))
584: return (bp);
585: }
586:
587: return (0);
588: }
589:
590: /*
591: * Get a block of requested size that is associated with
592: * a given vnode and block offset. If it is found in the
593: * block cache, mark it as having been found, make it busy
594: * and return it. Otherwise, return an empty block of the
595: * correct size. It is up to the caller to insure that the
596: * cached blocks be of the correct size.
597: */
598: struct buf *
599: getblk(vp, blkno, size, slpflag, slptimeo)
600: register struct vnode *vp;
601: daddr_t blkno;
602: int size, slpflag, slptimeo;
603: {
604: struct buf *bp;
605: int s, err;
606:
607: start:
608: s = splbio();
609: if (bp = incore(vp, blkno)) { /* XXX NFS VOP_BWRITE foolishness */
610: if (ISSET(bp->b_flags, B_BUSY)) {
611: SET(bp->b_flags, B_WANTED);
612: err = tsleep(bp, slpflag | (PRIBIO + 1), "getblk",
613: slptimeo);
614: splx(s);
615: if (err)
616: return (NULL);
617: goto start;
618: }
619: SET(bp->b_flags, (B_BUSY | B_CACHE));
620: bremfree(bp);
621: splx(s);
622: allocbuf(bp, size);
623: } else {
624: splx(s);
625: if ((bp = getnewbuf(slpflag, slptimeo)) == NULL)
626: goto start;
627: binshash(bp, BUFHASH(vp, blkno));
628: allocbuf(bp, size);
629: bp->b_blkno = bp->b_lblkno = blkno;
630: s = splbio();
631: bgetvp(vp, bp);
632: splx(s);
633: }
634: return (bp);
635: }
636:
637: /*
638: * Get an empty, disassociated buffer of given size.
639: */
640: struct buf *
641: geteblk(size)
642: int size;
643: {
644: struct buf *bp;
645:
646: while ((bp = getnewbuf(0, 0)) == 0)
647: ;
648: SET(bp->b_flags, B_INVAL);
649: binshash(bp, &invalhash);
650: allocbuf(bp, size);
651:
652: return (bp);
653: }
654:
655: /*
656: * Expand or contract the actual memory allocated to a buffer.
657: *
658: * If the buffer shrinks, data is lost, so it's up to the
659: * caller to have written it out *first*; this routine will not
660: * start a write. If the buffer grows, it's the callers
661: * responsibility to fill out the buffer's additional contents.
662: */
663: int
664: allocbuf(bp, size)
665: struct buf *bp;
666: int size;
667: {
668: struct buf *nbp;
669: vm_size_t desired_size;
670: int s;
671:
672: desired_size = roundup(size, CLBYTES);
673: if (desired_size > MAXBSIZE)
674: panic("allocbuf: buffer larger than MAXBSIZE requested");
675:
676: if (bp->b_bufsize == desired_size)
677: goto out;
678:
679: /*
680: * If the buffer is smaller than the desired size, we need to snarf
681: * it from other buffers. Get buffers (via getnewbuf()), and
682: * steal their pages.
683: */
684: while (bp->b_bufsize < desired_size) {
685: int amt;
686:
687: /* find a buffer */
688: while ((nbp = getnewbuf(0, 0)) == NULL)
689: ;
690: SET(nbp->b_flags, B_INVAL);
691: binshash(nbp, &invalhash);
692:
693: /* and steal its pages, up to the amount we need */
694: amt = min(nbp->b_bufsize, (desired_size - bp->b_bufsize));
695: pagemove((nbp->b_data + nbp->b_bufsize - amt),
696: bp->b_data + bp->b_bufsize, amt);
697: bp->b_bufsize += amt;
698: nbp->b_bufsize -= amt;
699:
700: /* reduce transfer count if we stole some data */
701: if (nbp->b_bcount > nbp->b_bufsize)
702: nbp->b_bcount = nbp->b_bufsize;
703:
704: #if DIAGNOSTIC
705: if (nbp->b_bufsize < 0)
706: panic("allocbuf: negative bufsize");
707: #endif
708:
709: brelse(nbp);
710: }
711:
712: /*
713: * If we want a buffer smaller than the current size,
714: * shrink this buffer. Grab a buf head from the EMPTY queue,
715: * move a page onto it, and put it on front of the AGE queue.
716: * If there are no free buffer headers, leave the buffer alone.
717: */
718: if (bp->b_bufsize > desired_size) {
719: s = splbio();
720: if ((nbp = bufqueues[BQ_EMPTY].tqh_first) == NULL) {
721: /* No free buffer head */
722: splx(s);
723: goto out;
724: }
725: bremfree(nbp);
726: SET(nbp->b_flags, B_BUSY);
727: splx(s);
728:
729: /* move the page to it and note this change */
730: pagemove(bp->b_data + desired_size,
731: nbp->b_data, bp->b_bufsize - desired_size);
732: nbp->b_bufsize = bp->b_bufsize - desired_size;
733: bp->b_bufsize = desired_size;
734: nbp->b_bcount = 0;
735: SET(nbp->b_flags, B_INVAL);
736:
737: /* release the newly-filled buffer and leave */
738: brelse(nbp);
739: }
740:
741: out:
742: bp->b_bcount = size;
743: }
744:
745: /*
746: * Find a buffer which is available for use.
747: * Select something from a free list.
748: * Preference is to AGE list, then LRU list.
749: */
750: struct buf *
751: getnewbuf(slpflag, slptimeo)
752: int slpflag, slptimeo;
753: {
754: register struct buf *bp;
755: register struct buf *lru_bp;
756: register struct buf *age_bp;
757: register int age_time, lru_time;
758: int s;
759: struct ucred *cred;
760:
761: start:
762: s = splbio();
763:
764: age_bp = bufqueues[BQ_AGE].tqh_first;
765: lru_bp = bufqueues[BQ_LRU].tqh_first;
766:
767: if (age_bp == NULL && lru_bp == NULL) {
768: /* wait for a free buffer of any kind */
769: needbuffer = 1;
770: tsleep(&needbuffer, slpflag|(PRIBIO+1), "getnewbuf", slptimeo);
771: splx(s);
772: return (0);
773: }
774: if (age_bp == NULL)
775: bp = lru_bp;
776: else if (lru_bp == NULL)
777: bp = age_bp;
778: else {
779: if (((age_time = (time.tv_sec - age_bp->b_timestamp)) < 0) ||
780: ((lru_time = (time.tv_sec - lru_bp->b_timestamp)) < 0)) {
781: /* time was set backwards */
782: bp = age_bp;
783: /*
784: * we should probably re-timestamp eveything in the queues
785: * at this point with the current time
786: */
787: } else {
788: if (lru_time >= lru_is_stale && age_time < age_is_stale)
789: bp = lru_bp;
790: else
791: bp = age_bp;
792: }
793: }
794: bremfree(bp);
795:
796: /* Buffer is no longer on free lists. */
797: SET(bp->b_flags, B_BUSY);
798: splx(s);
799:
800: /* If buffer was a delayed write, start it, and go back to the top. */
801: if (ISSET(bp->b_flags, B_DELWRI)) {
802: bawrite (bp);
803: goto start;
804: }
805:
806: trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
807:
808: /* disassociate us from our vnode, if we had one... */
809: s = splbio();
810: if (bp->b_vp)
811: brelvp(bp);
812: splx(s);
813:
814: /* clear out various other fields */
815: bp->b_flags = B_BUSY;
816: bp->b_dev = NODEV;
817: bp->b_blkno = bp->b_lblkno = 0;
818: bp->b_iodone = 0;
819: bp->b_error = 0;
820: bp->b_resid = 0;
821: bp->b_bcount = 0;
822: bp->b_dirtyoff = bp->b_dirtyend = 0;
823: bp->b_validoff = bp->b_validend = 0;
824:
825: /* nuke any credentials we were holding */
826: cred = bp->b_rcred;
827: if (cred != NOCRED) {
828: bp->b_rcred = NOCRED;
829: crfree(cred);
830: }
831: cred = bp->b_wcred;
832: if (cred != NOCRED) {
833: bp->b_wcred = NOCRED;
834: crfree(cred);
835: }
836:
837: bremhash(bp);
838: return (bp);
839: }
840:
841: /*
842: * Wait for operations on the buffer to complete.
843: * When they do, extract and return the I/O's error value.
844: */
845: int
846: biowait(bp)
847: struct buf *bp;
848: {
849: int s;
850:
851: s = splbio();
852: while (!ISSET(bp->b_flags, B_DONE))
853: tsleep(bp, PRIBIO + 1, "biowait", 0);
854: splx(s);
855:
856: /* check for interruption of I/O (e.g. via NFS), then errors. */
857: if (ISSET(bp->b_flags, B_EINTR)) {
858: CLR(bp->b_flags, B_EINTR);
859: return (EINTR);
860: } else if (ISSET(bp->b_flags, B_ERROR))
861: return (bp->b_error ? bp->b_error : EIO);
862: else
863: return (0);
864: }
865:
866: /*
867: * Mark I/O complete on a buffer.
868: *
869: * If a callback has been requested, e.g. the pageout
870: * daemon, do so. Otherwise, awaken waiting processes.
871: *
872: * [ Leffler, et al., says on p.247:
873: * "This routine wakes up the blocked process, frees the buffer
874: * for an asynchronous write, or, for a request by the pagedaemon
875: * process, invokes a procedure specified in the buffer structure" ]
876: *
877: * In real life, the pagedaemon (or other system processes) wants
878: * to do async stuff to, and doesn't want the buffer brelse()'d.
879: * (for swap pager, that puts swap buffers on the free lists (!!!),
880: * for the vn device, that puts malloc'd buffers on the free lists!)
881: */
882: void
883: biodone(bp)
884: struct buf *bp;
885: {
886: boolean_t funnel_state;
887:
888: funnel_state = thread_set_funneled(TRUE);
889: if (ISSET(bp->b_flags, B_DONE))
890: panic("biodone already");
891: SET(bp->b_flags, B_DONE); /* note that it's done */
892:
893: if (!ISSET(bp->b_flags, B_READ) && !ISSET(bp->b_flags, B_RAW)) /* wake up reader */
894: vwakeup(bp);
895:
896: if (ISSET(bp->b_flags, B_CALL)) { /* if necessary, call out */
897: CLR(bp->b_flags, B_CALL); /* but note callout done */
898: (*bp->b_iodone)(bp);
899: } else if (ISSET(bp->b_flags, B_ASYNC)) /* if async, release it */
900: brelse(bp);
901: else { /* or just wakeup the buffer */
902: CLR(bp->b_flags, B_WANTED);
903: wakeup(bp);
904: }
905: (void) thread_set_funneled(funnel_state);
906: }
907:
908: /*
909: * Return a count of buffers on the "locked" queue.
910: */
911: int
912: count_lock_queue()
913: {
914: register struct buf *bp;
915: register int n = 0;
916:
917: for (bp = bufqueues[BQ_LOCKED].tqh_first; bp;
918: bp = bp->b_freelist.tqe_next)
919: n++;
920: return (n);
921: }
922:
923: #if MACH_NBC
924: #include <ufs/ufs/quota.h>
925: #include <ufs/ufs/inode.h>
926:
927: #define btodevblk(b) ((b) / devBlocksize)
928: void
929: blkflush(struct vnode *vp, daddr_t blkno, vm_size_t size)
930: {
931: register struct buf *ep, *nbp;
932: daddr_t start, last;
933: int s,err;
934: struct inode *ip= VTOI(vp);
935: int devBlocksize=1024;
936:
937: #if 1
938: VOP_DEVBLOCKSIZE(ip->i_devvp, &devBlocksize);
939: #endif
940:
941:
942: start = blkno;
943: last = start + btodb(size, devBlocksize) - 1;
944: loop:
945: for(ep = vp->v_dirtyblkhd.lh_first; ep; ep = nbp) {
946: nbp = ep->b_vnbufs.le_next;
947: if (ep->b_vp != vp || ISSET(ep->b_flags, B_INVAL))
948: continue;
949: /* look for overlap */
950: if (ep->b_bcount == 0 || ep->b_blkno > last ||
951: ep->b_blkno + btodevblk(ep->b_bcount) <= start)
952: continue;
953: s = splbio();
954: if (ISSET(ep->b_flags, B_BUSY)) {
955: SET(ep->b_flags, B_WANTED);
956: err = tsleep(ep, (PRIBIO + 1), "blkflush",
957: 0);
958: splx(s);
959: goto loop;
960: }
961: if(ISSET(ep->b_flags, B_DELWRI)) {
962: bremfree(ep);
963: SET(ep->b_flags, B_BUSY);
964: (void) splx(s);
965: bwrite(ep);
966: goto loop;
967: }
968: (void) splx(s);
969: }
970:
971: }
972: #endif /* MACH_NBC */
973: #if DIAGNOSTIC
974: /*
975: * Print out statistics on the current allocation of the buffer pool.
976: * Can be enabled to print out on every ``sync'' by setting "syncprt"
977: * in vfs_syscalls.c using sysctl.
978: */
979: void
980: vfs_bufstats()
981: {
982: int s, i, j, count;
983: register struct buf *bp;
984: register struct bqueues *dp;
985: int counts[MAXBSIZE/CLBYTES+1];
986: static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" };
987:
988: for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) {
989: count = 0;
990: for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
991: counts[j] = 0;
992: s = splbio();
993: for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) {
994: counts[bp->b_bufsize/CLBYTES]++;
995: count++;
996: }
997: splx(s);
998: printf("%s: total-%d", bname[i], count);
999: for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
1000: if (counts[j] != 0)
1001: printf(", %d-%d", j * CLBYTES, counts[j]);
1002: printf("\n");
1003: }
1004: }
1005: #endif /* DIAGNOSTIC */
1006:
1007:
1008: struct buf *
1009: alloc_io_buf(vp)
1010: struct vnode *vp;
1011: { register struct buf *bp;
1012: int s;
1013:
1014: s = splbio();
1015:
1016: if ((bp = iobufqueue.tqh_first) == NULL) {
1017: splx(s);
1018: return (NULL);
1019: }
1020: TAILQ_REMOVE(&iobufqueue, bp, b_freelist);
1021: bp->b_timestamp = 0;
1022:
1023: /* clear out various fields */
1024: bp->b_flags = (B_BUSY | B_RAW);
1025: bp->b_blkno = bp->b_lblkno = 0;
1026: bp->b_iodone = 0;
1027: bp->b_error = 0;
1028: bp->b_resid = 0;
1029: bp->b_bcount = 0;
1030: bp->b_bufsize = 0;
1031: bp->b_vp = vp;
1032:
1033: if (vp->v_type == VBLK || vp->v_type == VCHR)
1034: bp->b_dev = vp->v_rdev;
1035: else
1036: bp->b_dev = NODEV;
1037: splx(s);
1038:
1039: return (bp);
1040: }
1041:
1042: void
1043: free_io_buf(bp)
1044: struct buf *bp;
1045: {
1046: int s;
1047:
1048: s = splbio();
1049: /*
1050: * put buffer back on the head of the iobufqueue
1051: */
1052: bp->b_vp = NULL;
1053: bp->b_flags = B_INVAL;
1054:
1055: binsheadfree(bp, &iobufqueue);
1056:
1057: splx(s);
1058: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.