correctly init v_ralen and v_maxra

[unix-history] / usr / src / sys / kern / vfs_bio.c

diff --git a/usr/src/sys/kern/vfs_bio.c b/usr/src/sys/kern/vfs_bio.c
index 4b8e9dd..8e527c8 100644 (file)
--- a/usr/src/sys/kern/vfs_bio.c
+++ b/usr/src/sys/kern/vfs_bio.c
@@ -1,20 +1,117 @@
-/*
- * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+/*-
+ * Copyright (c) 1986, 1989, 1993 The Regents of the University of California.

  * All rights reserved.
  *
  * All rights reserved.
  *

+ * This code is derived from software contributed to Berkeley by
+ * Berkeley Software Design Inc.
+ *

  * %sccs.include.redist.c%
  *
  * %sccs.include.redist.c%
  *

- *     @(#)vfs_bio.c   7.33 (Berkeley) %G%
+ *     @(#)vfs_bio.c   8.6 (Berkeley) %G%
+ */
+
+#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>
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
+
+/*
+ * Definitions for the buffer hash lists.
+ */
+#define        BUFHASH(dvp, lbn)       \
+       (&bufhashtbl[((int)(dvp) / sizeof(*(dvp)) + (int)(lbn)) & bufhash])
+LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash;
+u_long bufhash;
+
+/*
+ * 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)
+
+/*
+ * Definitions for the buffer free lists.
+ */
+#define        BQUEUES         4               /* number of free buffer queues */
+
+#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 */
+
+TAILQ_HEAD(bqueues, buf) bufqueues[BQUEUES];
+int needbuffer;
+
+/*
+ * Insq/Remq for the buffer free lists.

  */
  */

+#define        binsheadfree(bp, dp)    TAILQ_INSERT_HEAD(dp, bp, b_freelist)
+#define        binstailfree(bp, dp)    TAILQ_INSERT_TAIL(dp, bp, b_freelist)
+
+void
+bremfree(bp)
+       struct buf *bp;
+{
+       struct bqueues *dp = NULL;
+
+       /*
+        * 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 (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);
+}

 
 

-#include "param.h"
-#include "user.h"
-#include "buf.h"
-#include "vnode.h"
-#include "specdev.h"
-#include "mount.h"
-#include "trace.h"
-#include "ucred.h"
+/*
+ * Initialize buffers and hash links for buffers.
+ */
+void
+bufinit()
+{
+       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);
+       }
+}

 
 /*
  * Find the block in the buffer pool.
 
 /*
  * Find the block in the buffer pool.


@@ -31,6 +128,7 @@ bread(vp, blkno, size, cred, bpp)
        long secsize;
 #endif SECSIZE
 {
        long secsize;
 #endif SECSIZE
 {

+       struct proc *p = curproc;               /* XXX */

        register struct buf *bp;
 
        if (size == 0)
        register struct buf *bp;
 
        if (size == 0)


@@ -38,7 +136,7 @@ bread(vp, blkno, size, cred, bpp)
 #ifdef SECSIZE
        bp = getblk(dev, blkno, size, secsize);
 #else SECSIZE
 #ifdef SECSIZE
        bp = getblk(dev, blkno, size, secsize);
 #else SECSIZE

-       *bpp = bp = getblk(vp, blkno, size);
+       *bpp = bp = getblk(vp, blkno, size, 0, 0);

 #endif SECSIZE
        if (bp->b_flags & (B_DONE | B_DELWRI)) {
                trace(TR_BREADHIT, pack(vp, size), blkno);
 #endif SECSIZE
        if (bp->b_flags & (B_DONE | B_DELWRI)) {
                trace(TR_BREADHIT, pack(vp, size), blkno);


@@ -53,25 +151,28 @@ bread(vp, blkno, size, cred, bpp)
        }
        VOP_STRATEGY(bp);
        trace(TR_BREADMISS, pack(vp, size), blkno);
        }
        VOP_STRATEGY(bp);
        trace(TR_BREADMISS, pack(vp, size), blkno);

-       u.u_ru.ru_inblock++;            /* pay for read */
+       p->p_stats->p_ru.ru_inblock++;          /* pay for read */

        return (biowait(bp));
 }
 
 /*
        return (biowait(bp));
 }
 
 /*

- * Operates like bread, but also starts I/O on the specified
- * read-ahead block.
+ * Operates like bread, but also starts I/O on the N specified
+ * read-ahead blocks.

  */
  */

-breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
+breadn(vp, blkno, size, rablkno, rabsize, num, cred, bpp)

        struct vnode *vp;
        daddr_t blkno; int size;
 #ifdef SECSIZE
        long secsize;
 #endif SECSIZE
        struct vnode *vp;
        daddr_t blkno; int size;
 #ifdef SECSIZE
        long secsize;
 #endif SECSIZE

-       daddr_t rablkno; int rabsize;
+       daddr_t rablkno[]; int rabsize[];
+       int num;

        struct ucred *cred;
        struct buf **bpp;
 {
        struct ucred *cred;
        struct buf **bpp;
 {

+       struct proc *p = curproc;               /* XXX */

        register struct buf *bp, *rabp;
        register struct buf *bp, *rabp;

+       register int i;

 
        bp = NULL;
        /*
 
        bp = NULL;
        /*


@@ -79,32 +180,36 @@ breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
         * allocate a buffer and start I/O.
         */
        if (!incore(vp, blkno)) {
         * allocate a buffer and start I/O.
         */
        if (!incore(vp, blkno)) {

-               *bpp = bp = getblk(vp, blkno, size);
+               *bpp = bp = getblk(vp, blkno, size, 0, 0);

 #endif SECSIZE
                if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) {
                        bp->b_flags |= B_READ;
                        if (bp->b_bcount > bp->b_bufsize)
 #endif SECSIZE
                if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) {
                        bp->b_flags |= B_READ;
                        if (bp->b_bcount > bp->b_bufsize)

-                               panic("breada");
+                               panic("breadn");

                        if (bp->b_rcred == NOCRED && cred != NOCRED) {
                                crhold(cred);
                                bp->b_rcred = cred;
                        }
                        VOP_STRATEGY(bp);
                        trace(TR_BREADMISS, pack(vp, size), blkno);
                        if (bp->b_rcred == NOCRED && cred != NOCRED) {
                                crhold(cred);
                                bp->b_rcred = cred;
                        }
                        VOP_STRATEGY(bp);
                        trace(TR_BREADMISS, pack(vp, size), blkno);

-                       u.u_ru.ru_inblock++;            /* pay for read */
-               } else
+                       p->p_stats->p_ru.ru_inblock++;  /* pay for read */
+               } else {

                        trace(TR_BREADHIT, pack(vp, size), blkno);
                        trace(TR_BREADHIT, pack(vp, size), blkno);

+               }

        }
 
        /*
        }
 
        /*

-        * If there is a read-ahead block, start I/O on it too.
+        * If there's read-ahead block(s), start I/O
+        * on them also (as above).

         */
         */

-       if (!incore(vp, rablkno)) {
-               rabp = getblk(vp, rablkno, rabsize);
+       for (i = 0; i < num; i++) {
+               if (incore(vp, rablkno[i]))
+                       continue;
+               rabp = getblk(vp, rablkno[i], rabsize[i], 0, 0);

 #endif SECSIZE
                if (rabp->b_flags & (B_DONE | B_DELWRI)) {
                        brelse(rabp);
 #endif SECSIZE
                if (rabp->b_flags & (B_DONE | B_DELWRI)) {
                        brelse(rabp);

-                       trace(TR_BREADHITRA, pack(vp, rabsize), rablkno);
+                       trace(TR_BREADHITRA, pack(vp, rabsize[i]), rablkno[i]);

                } else {
                        rabp->b_flags |= B_ASYNC | B_READ;
                        if (rabp->b_bcount > rabp->b_bufsize)
                } else {
                        rabp->b_flags |= B_ASYNC | B_READ;
                        if (rabp->b_bcount > rabp->b_bufsize)


@@ -114,8 +219,8 @@ breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
                                rabp->b_rcred = cred;
                        }
                        VOP_STRATEGY(rabp);
                                rabp->b_rcred = cred;
                        }
                        VOP_STRATEGY(rabp);

-                       trace(TR_BREADMISSRA, pack(vp, rabsize), rablkno);
-                       u.u_ru.ru_inblock++;            /* pay in advance */
+                       trace(TR_BREADMISSRA, pack(vp, rabsize[i]), rablkno[i]);
+                       p->p_stats->p_ru.ru_inblock++;  /* pay in advance */

                }
        }
 
                }
        }
 


@@ -139,20 +244,26 @@ breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
 bwrite(bp)
        register struct buf *bp;
 {
 bwrite(bp)
        register struct buf *bp;
 {

+       struct proc *p = curproc;               /* XXX */

        register int flag;
        register int flag;

-       int s, error;
+       int s, error = 0;

 
 

+       if (bp->b_vp && (bp->b_vp->v_mount->mnt_flag & MNT_ASYNC))
+               bp->b_flags |= B_ASYNC;

        flag = bp->b_flags;
        bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
        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 */
-       else
-               reassignbuf(bp, bp->b_vp);
+       if (flag & B_ASYNC) {
+               if ((flag & B_DELWRI) == 0)
+                       p->p_stats->p_ru.ru_oublock++;  /* no one paid yet */
+               else
+                       reassignbuf(bp, bp->b_vp);
+       }

        trace(TR_BWRITE, pack(bp->b_vp, bp->b_bcount), bp->b_lblkno);
        if (bp->b_bcount > bp->b_bufsize)
                panic("bwrite");
        s = splbio();
        bp->b_vp->v_numoutput++;
        trace(TR_BWRITE, pack(bp->b_vp, bp->b_bcount), bp->b_lblkno);
        if (bp->b_bcount > bp->b_bufsize)
                panic("bwrite");
        s = splbio();
        bp->b_vp->v_numoutput++;

+       bp->b_flags |= B_WRITEINPROG;

        splx(s);
        VOP_STRATEGY(bp);
 
        splx(s);
        VOP_STRATEGY(bp);
 


@@ -163,14 +274,32 @@ bwrite(bp)
         */
        if ((flag & B_ASYNC) == 0) {
                error = biowait(bp);
         */
        if ((flag & B_ASYNC) == 0) {
                error = biowait(bp);

+               if ((flag&B_DELWRI) == 0)
+                       p->p_stats->p_ru.ru_oublock++;  /* no one paid yet */
+               else
+                       reassignbuf(bp, bp->b_vp);
+               if (bp->b_flags & B_EINTR) {
+                       bp->b_flags &= ~B_EINTR;
+                       error = EINTR;
+               }

                brelse(bp);
        } else if (flag & B_DELWRI) {
                brelse(bp);
        } else if (flag & B_DELWRI) {

+               s = splbio();

                bp->b_flags |= B_AGE;
                bp->b_flags |= B_AGE;

-               error = 0;
+               splx(s);

        }
        return (error);
 }
 
        }
        return (error);
 }
 

+int
+vn_bwrite(ap)
+       struct vop_bwrite_args *ap;
+{
+       
+       return (bwrite(ap->a_bp));
+}
+
+

 /*
  * Delayed write.
  *
 /*
  * Delayed write.
  *


@@ -185,11 +314,12 @@ bwrite(bp)
 bdwrite(bp)
        register struct buf *bp;
 {
 bdwrite(bp)
        register struct buf *bp;
 {

+       struct proc *p = curproc;               /* XXX */

 
        if ((bp->b_flags & B_DELWRI) == 0) {
                bp->b_flags |= B_DELWRI;
                reassignbuf(bp, bp->b_vp);
 
        if ((bp->b_flags & B_DELWRI) == 0) {
                bp->b_flags |= B_DELWRI;
                reassignbuf(bp, bp->b_vp);

-               u.u_ru.ru_oublock++;            /* noone paid yet */
+               p->p_stats->p_ru.ru_oublock++;          /* no one paid yet */

        }
        /*
         * If this is a tape drive, the write must be initiated.
        }
        /*
         * If this is a tape drive, the write must be initiated.


@@ -216,7 +346,7 @@ bawrite(bp)
         * after starting the I/O.
         */
        bp->b_flags |= B_ASYNC;
         * after starting the I/O.
         */
        bp->b_flags |= B_ASYNC;

-       (void) bwrite(bp);
+       (void) VOP_BWRITE(bp);

 }
 
 /*
 }
 
 /*


@@ -226,7 +356,7 @@ bawrite(bp)
 brelse(bp)
        register struct buf *bp;
 {
 brelse(bp)
        register struct buf *bp;
 {

-       register struct buf *flist;
+       register struct bqueues *flist;

        int s;
 
        trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
        int s;
 
        trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);


@@ -236,13 +366,14 @@ brelse(bp)
         */
        if (bp->b_flags & B_WANTED)
                wakeup((caddr_t)bp);
         */
        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 (needbuffer) {
+               needbuffer = 0;
+               wakeup((caddr_t)&needbuffer);

        }
        /*
         * Retry I/O for locked buffers rather than invalidating them.
         */
        }
        /*
         * Retry I/O for locked buffers rather than invalidating them.
         */

+       s = splbio();

        if ((bp->b_flags & B_ERROR) && (bp->b_flags & B_LOCKED))
                bp->b_flags &= ~B_ERROR;
        /*
        if ((bp->b_flags & B_ERROR) && (bp->b_flags & B_LOCKED))
                bp->b_flags &= ~B_ERROR;
        /*


@@ -258,22 +389,21 @@ brelse(bp)
        /*
         * Stick the buffer back on a free list.
         */
        /*
         * Stick the buffer back on a free list.
         */

-       s = splbio();

        if (bp->b_bufsize <= 0) {
                /* block has no buffer ... put at front of unused buffer list */
        if (bp->b_bufsize <= 0) {
                /* block has no buffer ... put at front of unused buffer list */

-               flist = &bfreelist[BQ_EMPTY];
+               flist = &bufqueues[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 */
                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];
+               flist = &bufqueues[BQ_AGE];

                binsheadfree(bp, flist);
        } else {
                if (bp->b_flags & B_LOCKED)
                binsheadfree(bp, flist);
        } else {
                if (bp->b_flags & B_LOCKED)

-                       flist = &bfreelist[BQ_LOCKED];
+                       flist = &bufqueues[BQ_LOCKED];

                else if (bp->b_flags & B_AGE)
                else if (bp->b_flags & B_AGE)

-                       flist = &bfreelist[BQ_AGE];
+                       flist = &bufqueues[BQ_AGE];

                else
                else

-                       flist = &bfreelist[BQ_LRU];
+                       flist = &bufqueues[BQ_LRU];

                binstailfree(bp, flist);
        }
        bp->b_flags &= ~(B_WANTED | B_BUSY | B_ASYNC | B_AGE | B_NOCACHE);
                binstailfree(bp, flist);
        }
        bp->b_flags &= ~(B_WANTED | B_BUSY | B_ASYNC | B_AGE | B_NOCACHE);


@@ -283,19 +413,18 @@ brelse(bp)
 /*
  * Check to see if a block is currently memory resident.
  */
 /*
  * Check to see if a block is currently memory resident.
  */

+struct buf *

 incore(vp, blkno)
        struct vnode *vp;
        daddr_t blkno;
 {
        register struct buf *bp;
 incore(vp, blkno)
        struct vnode *vp;
        daddr_t blkno;
 {
        register struct buf *bp;

-       register struct buf *dp;

 
 

-       dp = BUFHASH(vp, blkno);
-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
+       for (bp = BUFHASH(vp, blkno)->lh_first; bp; bp = bp->b_hash.le_next)

                if (bp->b_lblkno == blkno && bp->b_vp == vp &&
                    (bp->b_flags & B_INVAL) == 0)
                if (bp->b_lblkno == blkno && bp->b_vp == vp &&
                    (bp->b_flags & B_INVAL) == 0)

-                       return (1);
-       return (0);
+                       return (bp);
+       return (NULL);

 }
 
 /*
 }
 
 /*


@@ -307,16 +436,17 @@ struct buf *
 #ifdef SECSIZE
 getblk(dev, blkno, size, secsize)
 #else SECSIZE
 #ifdef SECSIZE
 getblk(dev, blkno, size, secsize)
 #else SECSIZE

-getblk(vp, blkno, size)
+getblk(vp, blkno, size, slpflag, slptimeo)

        register struct vnode *vp;
        daddr_t blkno;
        register struct vnode *vp;
        daddr_t blkno;

-       int size;
+       int size, slpflag, slptimeo;

 #ifdef SECSIZE
        long secsize;
 #endif SECSIZE
 {
 #ifdef SECSIZE
        long secsize;
 #endif SECSIZE
 {

-       register struct buf *bp, *dp;
-       int s;
+       register struct buf *bp;
+       struct bufhashhdr *dp;
+       int s, error;

 
        if (size > MAXBSIZE)
                panic("getblk: size too big");
 
        if (size > MAXBSIZE)
                panic("getblk: size too big");


@@ -327,30 +457,52 @@ getblk(vp, blkno, size)
         */
        dp = BUFHASH(vp, blkno);
 loop:
         */
        dp = BUFHASH(vp, blkno);
 loop:

-       for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
-               if (bp->b_lblkno != blkno || bp->b_vp != vp ||
-                   (bp->b_flags & B_INVAL))
+       for (bp = dp->lh_first; bp; bp = bp->b_hash.le_next) {
+               if (bp->b_lblkno != blkno || bp->b_vp != vp)

                        continue;
                s = splbio();
                if (bp->b_flags & B_BUSY) {
                        bp->b_flags |= B_WANTED;
                        continue;
                s = splbio();
                if (bp->b_flags & B_BUSY) {
                        bp->b_flags |= B_WANTED;

-                       sleep((caddr_t)bp, PRIBIO + 1);
+                       error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
+                               "getblk", slptimeo);

                        splx(s);
                        splx(s);

+                       if (error)
+                               return (NULL);

                        goto loop;
                }
                        goto loop;
                }

+               /*
+                * The test for B_INVAL is moved down here, since there
+                * are cases where B_INVAL is set before VOP_BWRITE() is
+                * called and for NFS, the process cannot be allowed to
+                * allocate a new buffer for the same block until the write
+                * back to the server has been completed. (ie. B_BUSY clears)
+                */
+               if (bp->b_flags & B_INVAL) {
+                       splx(s);
+                       continue;
+               }

                bremfree(bp);
                bp->b_flags |= B_BUSY;
                splx(s);
                if (bp->b_bcount != size) {
                        printf("getblk: stray size");
                        bp->b_flags |= B_INVAL;
                bremfree(bp);
                bp->b_flags |= B_BUSY;
                splx(s);
                if (bp->b_bcount != size) {
                        printf("getblk: stray size");
                        bp->b_flags |= B_INVAL;

-                       bwrite(bp);
+                       VOP_BWRITE(bp);

                        goto loop;
                }
                bp->b_flags |= B_CACHE;
                return (bp);
        }
                        goto loop;
                }
                bp->b_flags |= B_CACHE;
                return (bp);
        }

-       bp = getnewbuf();
+       /*
+        * The loop back to the top when getnewbuf() fails is because
+        * stateless filesystems like NFS have no node locks. Thus,
+        * there is a slight chance that more than one process will
+        * try and getnewbuf() for the same block concurrently when
+        * the first sleeps in getnewbuf(). So after a sleep, go back
+        * up to the top to check the hash lists again.
+        */
+       if ((bp = getnewbuf(slpflag, slptimeo)) == 0)
+               goto loop;

        bremhash(bp);
        bgetvp(vp, bp);
        bp->b_bcount = 0;
        bremhash(bp);
        bgetvp(vp, bp);
        bp->b_bcount = 0;


@@ -374,21 +526,21 @@ struct buf *
 geteblk(size)
        int size;
 {
 geteblk(size)
        int size;
 {

-       register struct buf *bp, *flist;
+       register struct buf *bp;

 
        if (size > MAXBSIZE)
                panic("geteblk: size too big");
 
        if (size > MAXBSIZE)
                panic("geteblk: size too big");

-       bp = getnewbuf();
+       while ((bp = getnewbuf(0, 0)) == NULL)
+               /* void */;

        bp->b_flags |= B_INVAL;
        bremhash(bp);
        bp->b_flags |= B_INVAL;
        bremhash(bp);

-       flist = &bfreelist[BQ_AGE];
+       binshash(bp, &invalhash);

        bp->b_bcount = 0;
 #ifdef SECSIZE
        bp->b_blksize = DEV_BSIZE;
 #endif SECSIZE
        bp->b_error = 0;
        bp->b_resid = 0;
        bp->b_bcount = 0;
 #ifdef SECSIZE
        bp->b_blksize = DEV_BSIZE;
 #endif SECSIZE
        bp->b_error = 0;
        bp->b_resid = 0;

-       binshash(bp, flist);

        allocbuf(bp, size);
        return (bp);
 }
        allocbuf(bp, size);
        return (bp);
 }


@@ -418,14 +570,13 @@ allocbuf(tp, size)
         * extra space in the present buffer.
         */
        if (sizealloc < tp->b_bufsize) {
         * extra space in the present buffer.
         */
        if (sizealloc < tp->b_bufsize) {

-               ep = bfreelist[BQ_EMPTY].av_forw;
-               if (ep == &bfreelist[BQ_EMPTY])
+               if ((ep = bufqueues[BQ_EMPTY].tqh_first) == NULL)

                        goto out;
                s = splbio();
                bremfree(ep);
                ep->b_flags |= B_BUSY;
                splx(s);
                        goto out;
                s = splbio();
                bremfree(ep);
                ep->b_flags |= B_BUSY;
                splx(s);

-               pagemove(tp->b_un.b_addr + sizealloc, ep->b_un.b_addr,
+               pagemove((char *)tp->b_data + sizealloc, ep->b_data,

                    (int)tp->b_bufsize - sizealloc);
                ep->b_bufsize = tp->b_bufsize - sizealloc;
                tp->b_bufsize = sizealloc;
                    (int)tp->b_bufsize - sizealloc);
                ep->b_bufsize = tp->b_bufsize - sizealloc;
                tp->b_bufsize = sizealloc;


@@ -441,18 +592,19 @@ allocbuf(tp, size)
         */
        while (tp->b_bufsize < sizealloc) {
                take = sizealloc - tp->b_bufsize;
         */
        while (tp->b_bufsize < sizealloc) {
                take = sizealloc - tp->b_bufsize;

-               bp = getnewbuf();
+               while ((bp = getnewbuf(0, 0)) == NULL)
+                       /* void */;

                if (take >= bp->b_bufsize)
                        take = bp->b_bufsize;
                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);
+               pagemove(&((char *)bp->b_data)[bp->b_bufsize - take],
+                   &((char *)tp->b_data)[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);
                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]);
+                       binshash(bp, &invalhash);

                        bp->b_dev = NODEV;
                        bp->b_error = 0;
                        bp->b_flags |= B_INVAL;
                        bp->b_dev = NODEV;
                        bp->b_error = 0;
                        bp->b_flags |= B_INVAL;


@@ -470,24 +622,47 @@ out:
  * Preference is to AGE list, then LRU list.
  */
 struct buf *
  * Preference is to AGE list, then LRU list.
  */
 struct buf *

-getnewbuf()
+getnewbuf(slpflag, slptimeo)
+       int slpflag, slptimeo;

 {
 {

-       register struct buf *bp, *dp;
+       register struct buf *bp;
+       register struct bqueues *dp;

        register struct ucred *cred;
        int s;
        register struct ucred *cred;
        int s;

+       struct buf *abp;
+       static int losecnt = 0;

 
 loop:
        s = splbio();
 
 loop:
        s = splbio();

-       for (dp = &bfreelist[BQ_AGE]; dp > bfreelist; dp--)
-               if (dp->av_forw != dp)
-                       break;
-       if (dp == bfreelist) {          /* no free blocks */
-               dp->b_flags |= B_WANTED;
-               sleep((caddr_t)dp, PRIBIO + 1);
+       abp = NULL;
+       for (dp = &bufqueues[BQ_AGE]; dp > bufqueues; dp--) {
+               for (bp = dp->qe_next; bp; bp = bp->b_freelist.qe_next) {
+                       if (abp == NULL)
+                               abp = bp;
+                       if ((bp->b_flags & B_DELWRI) &&
+                           bp->b_vp && VOP_ISLOCKED(bp->b_vp))
+                               continue;
+                       goto found;
+               }
+       }
+       if (dp == bufqueues) {          /* no free blocks */
+               if (abp) {
+                       bp = abp;
+                       bp->b_flags |= B_XXX;
+                       if (losecnt++ < 20) {
+                               vprint("skipping blkno check", bp->b_vp);
+                               printf("\tlblkno %d, blkno %d\n",
+                                  bp->b_lblkno, bp->b_blkno);
+                       }
+                       goto found;
+               }
+               needbuffer = 1;
+               (void) tsleep((caddr_t)&needbuffer, slpflag | (PRIBIO + 1),
+                       "getnewbuf", slptimeo);

                splx(s);
                splx(s);

-               goto loop;
+               return (NULL);

        }
        }

-       bp = dp->av_forw;
+found:

        bremfree(bp);
        bp->b_flags |= B_BUSY;
        splx(s);
        bremfree(bp);
        bp->b_flags |= B_BUSY;
        splx(s);


@@ -510,6 +685,7 @@ loop:
        }
        bp->b_flags = B_BUSY;
        bp->b_dirtyoff = bp->b_dirtyend = 0;
        }
        bp->b_flags = B_BUSY;
        bp->b_dirtyoff = bp->b_dirtyend = 0;

+       bp->b_validoff = bp->b_validend = 0;

        return (bp);
 }
 
        return (bp);
 }
 


@@ -542,26 +718,16 @@ biowait(bp)
  * If a callback has been requested, e.g. the pageout
  * daemon, do so. Otherwise, awaken waiting processes.
  */
  * If a callback has been requested, e.g. the pageout
  * daemon, do so. Otherwise, awaken waiting processes.
  */

+void

 biodone(bp)
        register struct buf *bp;
 {
 biodone(bp)
        register struct buf *bp;
 {

-       register struct vnode *vp;

 
        if (bp->b_flags & B_DONE)
                panic("dup biodone");
        bp->b_flags |= B_DONE;
 
        if (bp->b_flags & B_DONE)
                panic("dup biodone");
        bp->b_flags |= B_DONE;

-       if ((bp->b_flags & B_READ) == 0) {
-               bp->b_dirtyoff = bp->b_dirtyend = 0;
-               if (vp = bp->b_vp) {
-                       vp->v_numoutput--;
-                       if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
-                               if (vp->v_numoutput < 0)
-                                       panic("biodone: neg numoutput");
-                               vp->v_flag &= ~VBWAIT;
-                               wakeup((caddr_t)&vp->v_numoutput);
-                       }
-               }
-       }
+       if ((bp->b_flags & B_READ) == 0)
+               vwakeup(bp);

        if (bp->b_flags & B_CALL) {
                bp->b_flags &= ~B_CALL;
                (*bp->b_iodone)(bp);
        if (bp->b_flags & B_CALL) {
                bp->b_flags &= ~B_CALL;
                (*bp->b_iodone)(bp);


@@ -575,251 +741,48 @@ biodone(bp)
        }
 }
 
        }
 }
 

-/*
- * Make sure all write-behind blocks associated
- * with mount point are flushed out (from sync).
- */
-mntflushbuf(mountp, flags)
-       struct mount *mountp;
-       int flags;
-{
-       register struct vnode *vp;
-
-       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
-               panic("mntflushbuf: not busy");
-loop:
-       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
-               if (vget(vp))
-                       goto loop;
-               vflushbuf(vp, flags);
-               vput(vp);
-               if (vp->v_mount != mountp)
-                       goto loop;
-       }
-}
-
-/*
- * Flush all dirty buffers associated with a vnode.
- */
-vflushbuf(vp, flags)
-       register struct vnode *vp;
-       int flags;
+int
+count_lock_queue()

 {
        register struct buf *bp;
 {
        register struct buf *bp;

-       struct buf *nbp;
-       int s;
-
-loop:
-       s = splbio();
-       for (bp = vp->v_dirtyblkhd; bp; bp = nbp) {
-               nbp = bp->b_blockf;
-               if ((bp->b_flags & B_BUSY))
-                       continue;
-               if ((bp->b_flags & B_DELWRI) == 0)
-                       panic("vflushbuf: not dirty");
-               bremfree(bp);
-               bp->b_flags |= B_BUSY;
-               splx(s);
-               /*
-                * Wait for I/O associated with indirect blocks to complete,
-                * since there is no way to quickly wait for them below.
-                * NB: This is really specific to ufs, but is done here
-                * as it is easier and quicker.
-                */
-               if (bp->b_vp == vp || (flags & B_SYNC) == 0) {
-                       (void) bawrite(bp);
-                       s = splbio();
-               } else {
-                       (void) bwrite(bp);
-                       goto loop;
-               }
-       }
-       splx(s);
-       if ((flags & B_SYNC) == 0)
-               return;
-       s = splbio();
-       while (vp->v_numoutput) {
-               vp->v_flag |= VBWAIT;
-               sleep((caddr_t)&vp->v_numoutput, PRIBIO + 1);
-       }
-       splx(s);
-       if (vp->v_dirtyblkhd) {
-               vprint("vflushbuf: dirty", vp);
-               goto loop;
-       }
-}
+       register int ret;

 
 

-/*
- * Invalidate in core blocks belonging to closed or umounted filesystem
- *
- * Go through the list of vnodes associated with the file system;
- * for each vnode invalidate any buffers that it holds. Normally
- * this routine is preceeded by a bflush call, so that on a quiescent
- * filesystem there will be no dirty buffers when we are done. Binval
- * returns the count of dirty buffers when it is finished.
- */
-mntinvalbuf(mountp)
-       struct mount *mountp;
-{
-       register struct vnode *vp;
-       int dirty = 0;
-
-       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
-               panic("mntinvalbuf: not busy");
-loop:
-       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
-               if (vget(vp))
-                       goto loop;
-               dirty += vinvalbuf(vp, 1);
-               vput(vp);
-               if (vp->v_mount != mountp)
-                       goto loop;
-       }
-       return (dirty);
+       for (ret = 0, bp = (struct buf *)bufqueues[BQ_LOCKED].tqh_first;
+           bp; bp = (struct buf *)bp->b_freelist.tqe_next)
+               ++ret;
+       return(ret);

 }
 
 }
 

+#ifdef DIAGNOSTIC

 /*
 /*

- * Flush out and invalidate all buffers associated with a vnode.
- * Called with the underlying object locked.
+ * 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.

  */
  */

-vinvalbuf(vp, save)
-       register struct vnode *vp;
-       int save;
+void
+vfs_bufstats()

 {
 {

+       int s, i, j, count;

        register struct buf *bp;
        register struct buf *bp;

-       struct buf *nbp, *blist;
-       int s, dirty = 0;
-
-       for (;;) {
-               if (blist = vp->v_dirtyblkhd)
-                       /* void */;
-               else if (blist = vp->v_cleanblkhd)
-                       /* void */;
-               else
-                       break;
-               for (bp = blist; bp; bp = nbp) {
-                       nbp = bp->b_blockf;
-                       s = splbio();
-                       if (bp->b_flags & B_BUSY) {
-                               bp->b_flags |= B_WANTED;
-                               sleep((caddr_t)bp, PRIBIO + 1);
-                               splx(s);
-                               break;
-                       }
-                       bremfree(bp);
-                       bp->b_flags |= B_BUSY;
-                       splx(s);
-                       if (save && (bp->b_flags & B_DELWRI)) {
-                               dirty++;
-                               (void) bwrite(bp);
-                               break;
-                       }
-                       if (bp->b_vp != vp)
-                               reassignbuf(bp, bp->b_vp);
-                       else
-                               bp->b_flags |= B_INVAL;
-                       brelse(bp);
+       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++;

                }
                }

-       }
-       if (vp->v_dirtyblkhd || vp->v_cleanblkhd)
-               panic("vinvalbuf: flush failed");
-       return (dirty);
-}
-
-/*
- * Associate a buffer with a vnode.
- */
-bgetvp(vp, bp)
-       register struct vnode *vp;
-       register struct buf *bp;
-{
-
-       if (bp->b_vp)
-               panic("bgetvp: not free");
-       VHOLD(vp);
-       bp->b_vp = vp;
-       if (vp->v_type == VBLK || vp->v_type == VCHR)
-               bp->b_dev = vp->v_rdev;
-       else
-               bp->b_dev = NODEV;
-       /*
-        * Insert onto list for new vnode.
-        */
-       if (vp->v_cleanblkhd) {
-               bp->b_blockf = vp->v_cleanblkhd;
-               bp->b_blockb = &vp->v_cleanblkhd;
-               vp->v_cleanblkhd->b_blockb = &bp->b_blockf;
-               vp->v_cleanblkhd = bp;
-       } else {
-               vp->v_cleanblkhd = bp;
-               bp->b_blockb = &vp->v_cleanblkhd;
-               bp->b_blockf = NULL;
-       }
-}
-
-/*
- * Disassociate a buffer from a vnode.
- */
-brelvp(bp)
-       register struct buf *bp;
-{
-       struct buf *bq;
-       struct vnode *vp;
-
-       if (bp->b_vp == (struct vnode *) 0)
-               panic("brelvp: NULL");
-       /*
-        * Delete from old vnode list, if on one.
-        */
-       if (bp->b_blockb) {
-               if (bq = bp->b_blockf)
-                       bq->b_blockb = bp->b_blockb;
-               *bp->b_blockb = bq;
-               bp->b_blockf = NULL;
-               bp->b_blockb = NULL;
-       }
-       vp = bp->b_vp;
-       bp->b_vp = (struct vnode *) 0;
-       HOLDRELE(vp);
-}
-
-/*
- * Reassign a buffer from one vnode to another.
- * Used to assign file specific control information
- * (indirect blocks) to the vnode to which they belong.
- */
-reassignbuf(bp, newvp)
-       register struct buf *bp;
-       register struct vnode *newvp;
-{
-       register struct buf *bq, **listheadp;
-
-       if (newvp == NULL)
-               panic("reassignbuf: NULL");
-       /*
-        * Delete from old vnode list, if on one.
-        */
-       if (bp->b_blockb) {
-               if (bq = bp->b_blockf)
-                       bq->b_blockb = bp->b_blockb;
-               *bp->b_blockb = bq;
-       }
-       /*
-        * If dirty, put on list of dirty buffers;
-        * otherwise insert onto list of clean buffers.
-        */
-       if (bp->b_flags & B_DELWRI)
-               listheadp = &newvp->v_dirtyblkhd;
-       else
-               listheadp = &newvp->v_cleanblkhd;
-       if (*listheadp) {
-               bp->b_blockf = *listheadp;
-               bp->b_blockb = listheadp;
-               bp->b_blockf->b_blockb = &bp->b_blockf;
-               *listheadp = bp;
-       } else {
-               *listheadp = bp;
-               bp->b_blockb = listheadp;
-               bp->b_blockf = NULL;
+               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 */