BSD 4_4_Lite1 release
[unix-history] / usr / src / sys / kern / vfs_bio.c
index 30bacb7..ec5c962 100644 (file)
-/*     vfs_bio.c       4.43    83/05/01        */
-
-#include "../machine/pte.h"
+/*-
+ * Copyright (c) 1982, 1986, 1989, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ * (c) UNIX System Laboratories, Inc.
+ * All or some portions of this file are derived from material licensed
+ * to the University of California by American Telephone and Telegraph
+ * Co. or Unix System Laboratories, Inc. and are reproduced herein with
+ * the permission of UNIX System Laboratories, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     from: @(#)vfs_bio.c     8.6 (Berkeley) 1/11/94
+ */
 
 
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/buf.h"
-#include "../h/conf.h"
-#include "../h/proc.h"
-#include "../h/seg.h"
-#include "../h/vm.h"
-#include "../h/trace.h"
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/buf.h>
+#include <sys/vnode.h>
+#include <sys/mount.h>
+#include <sys/trace.h>
+#include <sys/malloc.h>
+#include <sys/resourcevar.h>
 
 /*
 
 /*
- * Read in (if necessary) the block and return a buffer pointer.
+ * Definitions for the buffer hash lists.
  */
  */
-struct buf *
-bread(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       int size;
-{
-       register struct buf *bp;
+#define        BUFHASH(dvp, lbn)       \
+       (&bufhashtbl[((int)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
+LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
+u_long bufhash;
 
 
-       if (size == 0)
-               panic("bread: size 0");
-       bp = getblk(dev, blkno, size);
-       if (bp->b_flags&B_DONE) {
-               trace(TR_BREADHIT, dev, blkno);
-               return(bp);
-       }
-       bp->b_flags |= B_READ;
-       if (bp->b_bcount > bp->b_bufsize)
-               panic("bread");
-       (*bdevsw[major(dev)].d_strategy)(bp);
-       trace(TR_BREADMISS, dev, blkno);
-       u.u_ru.ru_inblock++;            /* pay for read */
-       biowait(bp);
-       return(bp);
-}
+/*
+ * Insq/Remq for the buffer hash lists.
+ */
+#define        binshash(bp, dp)        LIST_INSERT_HEAD(dp, bp, b_hash)
+#define        bremhash(bp)            LIST_REMOVE(bp, b_hash)
 
 /*
 
 /*
- * Read in the block, like bread, but also start I/O on the
- * read-ahead block (which is not allocated to the caller)
+ * Definitions for the buffer free lists.
  */
  */
-struct buf *
-breada(dev, blkno, size, rablkno, rabsize)
-       dev_t dev;
-       daddr_t blkno; int size;
-       daddr_t rablkno; int rabsize;
-{
-       register struct buf *bp, *rabp;
+#define        BQUEUES         4               /* number of free buffer queues */
 
 
-       bp = NULL;
-       /*
-        * If the block isn't in core, then allocate
-        * a buffer and initiate i/o (getblk checks
-        * for a cache hit).
-        */
-       if (!incore(dev, blkno)) {
-               bp = getblk(dev, blkno, size);
-               if ((bp->b_flags&B_DONE) == 0) {
-                       bp->b_flags |= B_READ;
-                       if (bp->b_bcount > bp->b_bufsize)
-                               panic("breada");
-                       (*bdevsw[major(dev)].d_strategy)(bp);
-                       trace(TR_BREADMISS, dev, blkno);
-                       u.u_ru.ru_inblock++;            /* pay for read */
-               } else
-                       trace(TR_BREADHIT, dev, blkno);
-       }
+#define        BQ_LOCKED       0               /* super-blocks &c */
+#define        BQ_LRU          1               /* lru, useful buffers */
+#define        BQ_AGE          2               /* rubbish */
+#define        BQ_EMPTY        3               /* buffer headers with no memory */
 
 
-       /*
-        * If there's a read-ahead block, start i/o
-        * on it also (as above).
-        */
-       if (rablkno && !incore(dev, rablkno)) {
-               rabp = getblk(dev, rablkno, rabsize);
-               if (rabp->b_flags & B_DONE) {
-                       brelse(rabp);
-                       trace(TR_BREADHITRA, dev, blkno);
-               } else {
-                       rabp->b_flags |= B_READ|B_ASYNC;
-                       if (rabp->b_bcount > rabp->b_bufsize)
-                               panic("breadrabp");
-                       (*bdevsw[major(dev)].d_strategy)(rabp);
-                       trace(TR_BREADMISSRA, dev, rablock);
-                       u.u_ru.ru_inblock++;            /* pay in advance */
-               }
-       }
-
-       /*
-        * If block was in core, let bread get it.
-        * If block wasn't in core, then the read was started
-        * above, and just wait for it.
-        */
-       if (bp == NULL)
-               return (bread(dev, blkno, size));
-       biowait(bp);
-       return (bp);
-}
+TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
+int needbuffer;
 
 /*
 
 /*
- * Write the buffer, waiting for completion.
- * Then release the buffer.
+ * Insq/Remq for the buffer free lists.
  */
  */
-bwrite(bp)
-       register struct buf *bp;
-{
-       register flag;
+#define        binsheadfree(bp, dp)    TAILQ_INSERT_HEAD(dp, bp, b_freelist)
+#define        binstailfree(bp, dp)    TAILQ_INSERT_TAIL(dp, bp, b_freelist)
 
 
-       flag = bp->b_flags;
-       bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
-       if ((flag&B_DELWRI) == 0)
-               u.u_ru.ru_oublock++;            /* noone paid yet */
-       trace(TR_BWRITE, bp->b_dev, bp->b_blkno);
-       if (bp->b_bcount > bp->b_bufsize)
-               panic("bwrite");
-       (*bdevsw[major(bp->b_dev)].d_strategy)(bp);
+void
+bremfree(bp)
+       struct buf *bp;
+{
+       struct bqueues *dp = NULL;
 
        /*
 
        /*
-        * If the write was synchronous, then await i/o completion.
-        * If the write was "delayed", then we put the buffer on
-        * the q of blocks awaiting i/o completion status.
+        * We only calculate the head of the freelist when removing
+        * the last element of the list as that is the only time that
+        * it is needed (e.g. to reset the tail pointer).
+        *
+        * NB: This makes an assumption about how tailq's are implemented.
         */
         */
-       if ((flag&B_ASYNC) == 0) {
-               biowait(bp);
-               brelse(bp);
-       } else if (flag & B_DELWRI)
-               bp->b_flags |= B_AGE;
+       if (bp->b_freelist.tqe_next == NULL) {
+               for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
+                       if (dp->tqh_last == &bp->b_freelist.tqe_next)
+                               break;
+               if (dp == &bufqueues[BQUEUES])
+                       panic("bremfree: lost tail");
+       }
+       TAILQ_REMOVE(dp, bp, b_freelist);
 }
 
 /*
 }
 
 /*
- * Release the buffer, marking it so that if it is grabbed
- * for another purpose it will be written out before being
- * given up (e.g. when writing a partial block where it is
- * assumed that another write for the same block will soon follow).
- * This can't be done for magtape, since writes must be done
- * in the same order as requested.
+ * Initialize buffers and hash links for buffers.
  */
  */
-bdwrite(bp)
-       register struct buf *bp;
+void
+bufinit()
 {
 {
-       register int flags;
-
-       if ((bp->b_flags&B_DELWRI) == 0)
-               u.u_ru.ru_oublock++;            /* noone paid yet */
-       flags = bdevsw[major(bp->b_dev)].d_flags;
-       if(flags & B_TAPE)
-               bawrite(bp);
-       else {
-               bp->b_flags |= B_DELWRI | B_DONE;
-               brelse(bp);
+       register struct buf *bp;
+       struct bqueues *dp;
+       register int i;
+       int base, residual;
+
+       for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
+               TAILQ_INIT(dp);
+       bufhashtbl = hashinit(nbuf, M_CACHE, &bufhash);
+       base = bufpages / nbuf;
+       residual = bufpages % nbuf;
+       for (i = 0; i < nbuf; i++) {
+               bp = &buf[i];
+               bzero((char *)bp, sizeof *bp);
+               bp->b_dev = NODEV;
+               bp->b_rcred = NOCRED;
+               bp->b_wcred = NOCRED;
+               bp->b_vnbufs.le_next = NOLIST;
+               bp->b_data = buffers + i * MAXBSIZE;
+               if (i < residual)
+                       bp->b_bufsize = (base + 1) * CLBYTES;
+               else
+                       bp->b_bufsize = base * CLBYTES;
+               bp->b_flags = B_INVAL;
+               dp = bp->b_bufsize ? &bufqueues[BQ_AGE] : &bufqueues[BQ_EMPTY];
+               binsheadfree(bp, dp);
+               binshash(bp, &invalhash);
        }
 }
 
        }
 }
 
-/*
- * Release the buffer, start I/O on it, but don't wait for completion.
- */
-bawrite(bp)
-       register struct buf *bp;
+bread(a1, a2, a3, a4, a5)
+       struct vnode *a1;
+       daddr_t a2;
+       int a3;
+       struct ucred *a4;
+       struct buf **a5;
 {
 
 {
 
-       bp->b_flags |= B_ASYNC;
-       bwrite(bp);
+       /*
+        * Body deleted.
+        */
+       return (EIO);
 }
 
 }
 
-/*
- * Release the buffer, with no I/O implied.
- */
-brelse(bp)
-       register struct buf *bp;
+breadn(a1, a2, a3, a4, a5, a6, a7, a8)
+       struct vnode *a1;
+       daddr_t a2; int a3;
+       daddr_t a4[]; int a5[];
+       int a6;
+       struct ucred *a7;
+       struct buf **a8;
 {
 {
-       register struct buf *flist;
-       register s;
 
        /*
 
        /*
-        * If someone's waiting for the buffer, or
-        * is waiting for a buffer wake 'em up.
+        * Body deleted.
         */
         */
-       if (bp->b_flags&B_WANTED)
-               wakeup((caddr_t)bp);
-       if (bfreelist[0].b_flags&B_WANTED) {
-               bfreelist[0].b_flags &= ~B_WANTED;
-               wakeup((caddr_t)bfreelist);
-       }
-       if (bp->b_flags&B_ERROR)
-               if (bp->b_flags & B_LOCKED)
-                       bp->b_flags &= ~B_ERROR;        /* try again later */
-               else
-                       bp->b_dev = NODEV;              /* no assoc */
+       return (EIO);
+}
+
+bwrite(a1)
+       struct buf *a1;
+{
 
        /*
 
        /*
-        * Stick the buffer back on a free list.
+        * Body deleted.
         */
         */
-       s = spl6();
-       if (bp->b_bufsize <= 0) {
-               /* block has no buffer ... put at front of unused buffer list */
-               flist = &bfreelist[BQ_EMPTY];
-               binsheadfree(bp, flist);
-       } else if (bp->b_flags & (B_ERROR|B_INVAL)) {
-               /* block has no info ... put at front of most free list */
-               flist = &bfreelist[BQ_AGE];
-               binsheadfree(bp, flist);
-       } else {
-               if (bp->b_flags & B_LOCKED)
-                       flist = &bfreelist[BQ_LOCKED];
-               else if (bp->b_flags & B_AGE)
-                       flist = &bfreelist[BQ_AGE];
-               else
-                       flist = &bfreelist[BQ_LRU];
-               binstailfree(bp, flist);
-       }
-       bp->b_flags &= ~(B_WANTED|B_BUSY|B_ASYNC|B_AGE);
-       splx(s);
+       return (EIO);
 }
 
 }
 
-/*
- * See if the block is associated with some buffer
- * (mainly to avoid getting hung up on a wait in breada)
- */
-incore(dev, blkno)
-       dev_t dev;
-       daddr_t blkno;
+int
+vn_bwrite(ap)
+       struct vop_bwrite_args *ap;
 {
 {
-       register struct buf *bp;
-       register struct buf *dp;
-
-       dp = BUFHASH(dev, blkno);
-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-               if (bp->b_blkno == blkno && bp->b_dev == dev &&
-                   (bp->b_flags & B_INVAL) == 0)
-                       return (1);
-       return (0);
+       return (bwrite(ap->a_bp));
 }
 
 }
 
-struct buf *
-baddr(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       int size;
+bdwrite(a1)
+       struct buf *a1;
 {
 
 {
 
-       if (incore(dev, blkno))
-               return (bread(dev, blkno, size));
-       return (0);
+       /*
+        * Body deleted.
+        */
+       return;
 }
 
 }
 
-/*
- * Assign a buffer for the given block.  If the appropriate
- * block is already associated, return it; otherwise search
- * for the oldest non-busy buffer and reassign it.
- *
- * We use splx here because this routine may be called
- * on the interrupt stack during a dump, and we don't
- * want to lower the ipl back to 0.
- */
-struct buf *
-getblk(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       int size;
+bawrite(a1)
+       struct buf *a1;
 {
 {
-       register struct buf *bp, *dp;
-       int s;
 
 
-       if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT)) /* XXX */
-               blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1);
        /*
        /*
-        * Search the cache for the block.  If we hit, but
-        * the buffer is in use for i/o, then we wait until
-        * the i/o has completed.
+        * Body deleted.
         */
         */
-       dp = BUFHASH(dev, blkno);
-loop:
-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
-               if (bp->b_blkno != blkno || bp->b_dev != dev ||
-                   bp->b_flags&B_INVAL)
-                       continue;
-               s = spl6();
-               if (bp->b_flags&B_BUSY) {
-                       bp->b_flags |= B_WANTED;
-                       sleep((caddr_t)bp, PRIBIO+1);
-                       splx(s);
-                       goto loop;
-               }
-               splx(s);
-               notavail(bp);
-               if (brealloc(bp, size) == 0)
-                       goto loop;
-               bp->b_flags |= B_CACHE;
-               return(bp);
-       }
-       if (major(dev) >= nblkdev)
-               panic("blkdev");
-       bp = getnewbuf();
-       bfree(bp);
-       bremhash(bp);
-       binshash(bp, dp);
-       bp->b_dev = dev;
-       bp->b_blkno = blkno;
-       bp->b_error = 0;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
+       return;
 }
 
 }
 
-/*
- * get an empty block,
- * not assigned to any particular device
- */
-struct buf *
-geteblk(size)
-       int size;
+brelse(a1)
+       struct buf *a1;
 {
 {
-       register struct buf *bp, *flist;
-
-loop:
-       bp = getnewbuf();
-       bp->b_flags |= B_INVAL;
-       bfree(bp);
-       bremhash(bp);
-       flist = &bfreelist[BQ_AGE];
-       binshash(bp, flist);
-       bp->b_dev = (dev_t)NODEV;
-       bp->b_error = 0;
-       if (brealloc(bp, size) == 0)
-               goto loop;
-       return(bp);
+
+       /*
+        * Body deleted.
+        */
+       return;
 }
 
 }
 
-/*
- * Allocate space associated with a buffer.
- * If can't get space, buffer is released
- */
-brealloc(bp, size)
-       register struct buf *bp;
-       int size;
+struct buf *
+incore(a1, a2)
+       struct vnode *a1;
+       daddr_t a2;
 {
 {
-       daddr_t start, last;
-       register struct buf *ep;
-       struct buf *dp;
-       int s;
 
        /*
 
        /*
-        * First need to make sure that all overlaping previous I/O
-        * is dispatched with.
+        * Body deleted.
         */
         */
-       if (size == bp->b_bcount)
-               return (1);
-       if (size < bp->b_bcount) { 
-               if (bp->b_flags & B_DELWRI) {
-                       bwrite(bp);
-                       return (0);
-               }
-               if (bp->b_flags & B_LOCKED)
-                       panic("brealloc");
-               return (allocbuf(bp, size));
-       }
-       bp->b_flags &= ~B_DONE;
-       if (bp->b_dev == NODEV)
-               return (allocbuf(bp, size));
+       return (0);
+}
+
+struct buf *
+getblk(a1, a2, a3, a4, a5)
+       struct vnode *a1;
+       daddr_t a2;
+       int a3, a4, a5;
+{
 
        /*
 
        /*
-        * Search cache for any buffers that overlap the one that we
-        * are trying to allocate. Overlapping buffers must be marked
-        * invalid, after being written out if they are dirty. (indicated
-        * by B_DELWRI) A disk block must be mapped by at most one buffer
-        * at any point in time. Care must be taken to avoid deadlocking
-        * when two buffer are trying to get the same set of disk blocks.
+        * Body deleted.
         */
         */
-       start = bp->b_blkno;
-       last = start + (size / DEV_BSIZE) - 1;
-       dp = BUFHASH(bp->b_dev, bp->b_blkno);
-loop:
-       for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
-               if (ep == bp || ep->b_dev != bp->b_dev || (ep->b_flags&B_INVAL))
-                       continue;
-               /* look for overlap */
-               if (ep->b_bcount == 0 || ep->b_blkno > last ||
-                   ep->b_blkno + (ep->b_bcount / DEV_BSIZE) <= start)
-                       continue;
-               s = spl6();
-               if (ep->b_flags&B_BUSY) {
-                       ep->b_flags |= B_WANTED;
-                       sleep((caddr_t)ep, PRIBIO+1);
-                       splx(s);
-                       goto loop;
-               }
-               splx(s);
-               notavail(ep);
-               if (ep->b_flags & B_DELWRI) {
-                       bwrite(ep);
-                       goto loop;
-               }
-               ep->b_flags |= B_INVAL;
-               brelse(ep);
-       }
-       return (allocbuf(bp, size));
+       return ((struct buf *)0);
 }
 
 }
 
-/*
- * Expand or contract the actual memory allocated to a buffer.
- * If no memory is available, release buffer and take error exit
- */
-allocbuf(tp, size)
-       register struct buf *tp;
-       int size;
+struct buf *
+geteblk(a1)
+       int a1;
 {
 {
-       register struct buf *bp, *ep;
-       int sizealloc, take;
-#ifdef sun
-       register char *a;
-       int osize;
-#endif
-
-#ifndef sun
-       sizealloc = roundup(size, CLBYTES);
-#else
-       sizealloc = roundup(size, BUFALLOCSIZE);
-#endif
-       /*
-        * Buffer size does not change
-        */
-       if (sizealloc == tp->b_bufsize)
-               goto out;
-#ifndef sun
-       /*
-        * Buffer size is shrinking.
-        * Place excess space in a buffer header taken from the
-        * BQ_EMPTY buffer list and placed on the "most free" list.
-        * If no extra buffer headers are available, leave the
-        * extra space in the present buffer.
-        */
-       if (sizealloc < tp->b_bufsize) {
-               ep = bfreelist[BQ_EMPTY].av_forw;
-               if (ep == &bfreelist[BQ_EMPTY])
-                       goto out;
-               notavail(ep);
-               pagemove(tp->b_un.b_addr + sizealloc, ep->b_un.b_addr,
-                   (int)tp->b_bufsize - sizealloc);
-               ep->b_bufsize = tp->b_bufsize - sizealloc;
-               tp->b_bufsize = sizealloc;
-               ep->b_flags |= B_INVAL;
-               ep->b_bcount = 0;
-               brelse(ep);
-               goto out;
-       }
-       /*
-        * More buffer space is needed. Get it out of buffers on
-        * the "most free" list, placing the empty headers on the
-        * BQ_EMPTY buffer header list.
-        */
-       while (tp->b_bufsize < sizealloc) {
-               take = sizealloc - tp->b_bufsize;
-               bp = getnewbuf();
-               if (take >= bp->b_bufsize)
-                       take = bp->b_bufsize;
-               pagemove(&bp->b_un.b_addr[bp->b_bufsize - take],
-                   &tp->b_un.b_addr[tp->b_bufsize], take);
-               tp->b_bufsize += take;
-               bp->b_bufsize = bp->b_bufsize - take;
-               if (bp->b_bcount > bp->b_bufsize)
-                       bp->b_bcount = bp->b_bufsize;
-               if (bp->b_bufsize <= 0) {
-                       bremhash(bp);
-                       binshash(bp, &bfreelist[BQ_EMPTY]);
-                       bp->b_dev = (dev_t)NODEV;
-                       bp->b_error = 0;
-                       bp->b_flags |= B_INVAL;
-               }
-               brelse(bp);
-       }
-#else
-       /*
-        * Buffer size is shrinking
-        * Just put the tail end back in the map
-        */
-       if (sizealloc < tp->b_bufsize) {
-               rmfree(buffermap, (long)(tp->b_bufsize - sizealloc),
-                       (long)(tp->b_un.b_addr + sizealloc));
-               tp->b_bufsize = sizealloc;
-               goto out;
-       }
+
        /*
        /*
-        * Buffer is being expanded or created
-        * If being expanded, attempt to get contiguous
-        * section, otherwise get a new chunk and copy.
-        * If no space, free up a buffer on the AGE list
-        * and try again.
+        * Body deleted.
         */
         */
-       do {
-               if ((osize = tp->b_bufsize)) {
-                       a = (char *)rmget(buffermap, (long)(sizealloc-osize),
-                               (long)(tp->b_un.b_addr + osize));
-                       if (a == 0) {
-                               a = (char *)rmalloc(buffermap, (long)sizealloc);
-                               if (a != 0) {
-                                       bcopy(tp->b_un.b_addr, a, osize);
-                                       rmfree(buffermap, (long)osize,
-                                               (long)tp->b_un.b_addr);
-                                       tp->b_un.b_addr = a;
-                               }
-                       }
-               } else {
-                       a = (char *)rmalloc(buffermap, (long)sizealloc);
-                       if (a != 0)
-                               tp->b_un.b_addr = a;
-               }
-       } while (a == 0 && bfreemem());
-       if (a == 0) {
-               brelse(tp);
-               return (0);
-       }
-       tp->b_bufsize = sizealloc;
-#endif
-out:
-       tp->b_bcount = size;
-       return (1);
+       return ((struct buf *)0);
 }
 
 }
 
-/*
- * Release space associated with a buffer.
- */
-bfree(bp)
-       struct buf *bp;
+allocbuf(a1, a2)
+       struct buf *a1;
+       int a2;
 {
 {
-#ifdef sun
-       if (bp->b_bufsize) {
-               rmfree(buffermap, (long)bp->b_bufsize, (long)bp->b_un.b_addr);
-               bp->b_bufsize = 0;
-       }
-#endif
-       bp->b_bcount = 0;
-}
 
 
-#ifdef sun
-/*
- * Attempt to free up buffer space by flushing
- * something in the free list.
- * Don't wait for something, that could cause deadlocks
- * We start with BQ_AGE because we know BQ_EMPTY take no memory.
- */
-bfreemem()
-{
-       register struct buf *bp, *dp;
-       int s;
-
-loop:
-       s = spl6();
-       for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
-               if (dp->av_forw != dp)
-                       break;
-       splx(s);
-       if (dp == bfreelist) {          /* no free blocks */
-               return (0);
-       }
-       bp = dp->av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-       trace(TR_BRELSE, bp->b_dev, bp->b_blkno);
-       bp->b_flags = B_BUSY | B_INVAL;
-       bfree(bp);
-       bremhash(bp);
-       binshash(bp, &bfreelist[BQ_EMPTY]);
-       bp->b_dev = (dev_t)NODEV;
-       bp->b_error = 0;
-       brelse(bp);
-       return (1);
+       /*
+        * Body deleted.
+        */
+       return (0);
 }
 }
-#endif
 
 
-/*
- * Find a buffer which is available for use.
- * Select something from a free list.
- * Preference is to AGE list, then LRU list.
- */
 struct buf *
 struct buf *
-getnewbuf()
+getnewbuf(a1, a2)
+       int a1, a2;
 {
 {
-       register struct buf *bp, *dp;
-       int s;
-
-loop:
-       s = spl6();
-#ifndef sun
-       for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
-#else
-       for (dp = &bfreelist[BQ_EMPTY]; dp > bfreelist; dp--)
-#endif
-               if (dp->av_forw != dp)
-                       break;
-       if (dp == bfreelist) {          /* no free blocks */
-               dp->b_flags |= B_WANTED;
-               sleep((caddr_t)dp, PRIBIO+1);
-               splx(s);
-               goto loop;
-       }
-       splx(s);
-       bp = dp->av_forw;
-       notavail(bp);
-       if (bp->b_flags & B_DELWRI) {
-               bp->b_flags |= B_ASYNC;
-               bwrite(bp);
-               goto loop;
-       }
-       trace(TR_BRELSE, bp->b_dev, bp->b_blkno);
-       bp->b_flags = B_BUSY;
-       return (bp);
-}
 
 
-/*
- * Wait for I/O completion on the buffer; return errors
- * to the user.
- */
-biowait(bp)
-       register struct buf *bp;
-{
-       int s;
-
-       s = spl6();
-       while ((bp->b_flags&B_DONE)==0)
-               sleep((caddr_t)bp, PRIBIO);
-       splx(s);
-       if (u.u_error == 0)                     /* XXX */
-               u.u_error = geterror(bp);
+       /*
+        * Body deleted.
+        */
+       return ((struct buf *)0);
 }
 
 }
 
-/*
- * Mark I/O complete on a buffer. If the header
- * indicates a dirty page push completion, the
- * header is inserted into the ``cleaned'' list
- * to be processed by the pageout daemon. Otherwise
- * release it if I/O is asynchronous, and wake 
- * up anyone waiting for it.
- */
-biodone(bp)
-       register struct buf *bp;
+biowait(a1)
+       struct buf *a1;
 {
 {
-       register int s;
-
-       if (bp->b_flags & B_DONE)
-               panic("dup biodone");
-       bp->b_flags |= B_DONE;
-       if (bp->b_flags & B_DIRTY) {
-               if (bp->b_flags & B_ERROR)
-                       panic("IO err in push");
-               s = spl6();
-               bp->av_forw = bclnlist;
-               bp->b_bcount = swsize[bp - swbuf];
-               bp->b_pfcent = swpf[bp - swbuf];
-               cnt.v_pgout++;
-               cnt.v_pgpgout += bp->b_bcount / NBPG;
-               bclnlist = bp;
-               if (bswlist.b_flags & B_WANTED)
-                       wakeup((caddr_t)&proc[2]);
-               splx(s);
-               return;
-       }
-       if (bp->b_flags & B_CALL) {
-               bp->b_flags &= ~B_CALL;
-               (*bp->b_iodone)(bp);
-               return;
-       }
-       if (bp->b_flags&B_ASYNC)
-               brelse(bp);
-       else {
-               bp->b_flags &= ~B_WANTED;
-               wakeup((caddr_t)bp);
-       }
-}
 
 
-/*
- * Insure that no part of a specified block is in an incore buffer.
- */
-blkflush(dev, blkno, size)
-       dev_t dev;
-       daddr_t blkno;
-       long size;
-{
-       register struct buf *ep;
-       struct buf *dp;
-       daddr_t start, last;
-       int s;
-
-       start = blkno;
-       last = start + (size / DEV_BSIZE) - 1;
-       dp = BUFHASH(dev, blkno);
-loop:
-       for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
-               if (ep->b_dev != dev || (ep->b_flags&B_INVAL))
-                       continue;
-               /* look for overlap */
-               if (ep->b_bcount == 0 || ep->b_blkno > last ||
-                   ep->b_blkno + (ep->b_bcount / DEV_BSIZE) <= start)
-                       continue;
-               s = spl6();
-               if (ep->b_flags&B_BUSY) {
-                       ep->b_flags |= B_WANTED;
-                       sleep((caddr_t)ep, PRIBIO+1);
-                       splx(s);
-                       goto loop;
-               }
-               if (ep->b_flags & B_DELWRI) {
-                       splx(s);
-                       notavail(ep);
-                       bwrite(ep);
-                       goto loop;
-               }
-               splx(s);
-       }
+       /*
+        * Body deleted.
+        */
+       return (EIO);
 }
 
 }
 
-/*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
- * (and temporarily pagein)
- */
-bflush(dev)
-       dev_t dev;
+void
+biodone(a1)
+       struct buf *a1;
 {
 {
-       register struct buf *bp;
-       register struct buf *flist;
-       int s;
-
-loop:
-       s = spl6();
-       for (flist = bfreelist; flist < &bfreelist[BQ_EMPTY]; flist++)
-       for (bp = flist->av_forw; bp != flist; bp = bp->av_forw) {
-               if ((bp->b_flags & B_DELWRI) == 0)
-                       continue;
-               if (dev == NODEV || dev == bp->b_dev) {
-                       bp->b_flags |= B_ASYNC;
-                       notavail(bp);
-                       bwrite(bp);
-                       goto loop;
-               }
-       }
-       splx(s);
+
+       /*
+        * Body deleted.
+        */
+       return;
 }
 
 }
 
-/*
- * Pick up the device's error number and pass it to the user;
- * if there is an error but the number is 0 set a generalized
- * code.  Actually the latter is always true because devices
- * don't yet return specific errors.
- */
-geterror(bp)
-       register struct buf *bp;
+int
+count_lock_queue()
 {
 {
-       int error = 0;
 
 
-       if (bp->b_flags&B_ERROR)
-               if ((error = bp->b_error)==0)
-                       return (EIO);
-       return (error);
+       /*
+        * Body deleted.
+        */
+       return (0);
 }
 
 }
 
+#ifdef DIAGNOSTIC
 /*
 /*
- * Invalidate in core blocks belonging to closed or umounted filesystem
- *
- * This is not nicely done at all - the buffer ought to be removed from the
- * hash chains & have its dev/blkno fields clobbered, but unfortunately we
- * can't do that here, as it is quite possible that the block is still
- * being used for i/o. Eventually, all disc drivers should be forced to
- * have a close routine, which ought ensure that the queue is empty, then
- * properly flush the queues. Until that happy day, this suffices for
- * correctness.                                                ... kre
+ * Print out statistics on the current allocation of the buffer pool.
+ * Can be enabled to print out on every ``sync'' by setting "syncprt"
+ * in vfs_syscalls.c using sysctl.
  */
  */
-binval(dev)
-       dev_t dev;
+void
+vfs_bufstats()
 {
 {
+       int s, i, j, count;
        register struct buf *bp;
        register struct buf *bp;
-       register struct bufhd *hp;
-#define dp ((struct buf *)hp)
-
-       for (hp = bufhash; hp < &bufhash[BUFHSZ]; hp++)
-               for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-                       if (bp->b_dev == dev)
-                               bp->b_flags |= B_INVAL;
+       register struct bqueues *dp;
+       int counts[MAXBSIZE/CLBYTES+1];
+       static char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE", "EMPTY" };
+
+       for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) {
+               count = 0;
+               for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
+                       counts[j] = 0;
+               s = splbio();
+               for (bp = dp->tqh_first; bp; bp = bp->b_freelist.tqe_next) {
+                       counts[bp->b_bufsize/CLBYTES]++;
+                       count++;
+               }
+               splx(s);
+               printf("%s: total-%d", bname[i], count);
+               for (j = 0; j <= MAXBSIZE/CLBYTES; j++)
+                       if (counts[j] != 0)
+                               printf(", %d-%d", j * CLBYTES, counts[j]);
+               printf("\n");
+       }
 }
 }
+#endif /* DIAGNOSTIC */