X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/fe8987fbae68c824eca164260c258f3146538522..a937f8567ba9375553a85507f18042a6faaffaab:/usr/src/sys/kern/vfs_cluster.c

diff --git a/usr/src/sys/kern/vfs_cluster.c b/usr/src/sys/kern/vfs_cluster.c
index 1a3d565039..362ad50350 100644
--- a/usr/src/sys/kern/vfs_cluster.c
+++ b/usr/src/sys/kern/vfs_cluster.c
@@ -1,162 +1,147 @@
-/*	vfs_cluster.c	3.3	%H%	*/
-
-#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/pte.h"
-#include "../h/vm.h"
-
-/*
- * The following several routines allocate and free
- * buffers with various side effects.  In general the
- * arguments to an allocate routine are a device and
- * a block number, and the value is a pointer to
- * to the buffer header; the buffer is marked "busy"
- * so that no one else can touch it.  If the block was
- * already in core, no I/O need be done; if it is
- * already busy, the process waits until it becomes free.
- * The following routines allocate a buffer:
- *	getblk
- *	bread
- *	breada
- *	baddr	(if it is incore)
- * Eventually the buffer must be released, possibly with the
- * side effect of writing it out, by using one of
- *	bwrite
- *	bdwrite
- *	bawrite
- *	brelse
- */
-
-#define	BUFHSZ	63
-#define	BUFHASH(blkno)	(blkno % BUFHSZ)
-short	bufhash[BUFHSZ];
-
 /*
- * Initialize hash links for buffers.
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms are permitted
+ * provided that the above copyright notice and this paragraph are
+ * duplicated in all such forms and that any documentation,
+ * advertising materials, and other materials related to such
+ * distribution and use acknowledge that the software was developed
+ * by the University of California, Berkeley.  The name of the
+ * University may not be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ *	@(#)vfs_cluster.c	7.10 (Berkeley) %G%
  */
-bhinit()
-{
-	register int i;
-
-	for (i = 0; i < BUFHSZ; i++)
-		bufhash[i] = -1;
-}
 
-/* #define	DISKMON	1 */
-
-#ifdef	DISKMON
-struct {
-	int	nbuf;
-	long	nread;
-	long	nreada;
-	long	ncache;
-	long	nwrite;
-	long	bufcount[NBUF];
-} io_info;
-#endif
-
-/*
- * Swap IO headers -
- * They contain the necessary information for the swap I/O.
- * At any given time, a swap header can be in three
- * different lists. When free it is in the free list, 
- * when allocated and the I/O queued, it is on the swap 
- * device list, and finally, if the operation was a dirty 
- * page push, when the I/O completes, it is inserted 
- * in a list of cleaned pages to be processed by the pageout daemon.
- */
-struct	buf swbuf[NSWBUF];
-short	swsize[NSWBUF];		/* CAN WE JUST USE B_BCOUNT? */
-int	swpf[NSWBUF];
-
-
-#ifdef	FASTVAX
-#define	notavail(bp) \
-{ \
-	int s = spl6(); \
-	(bp)->av_back->av_forw = (bp)->av_forw; \
-	(bp)->av_forw->av_back = (bp)->av_back; \
-	(bp)->b_flags |= B_BUSY; \
-	splx(s); \
-}
-#endif
+#include "param.h"
+#include "user.h"
+#include "buf.h"
+#include "vnode.h"
+#include "trace.h"
+#include "ucred.h"
 
 /*
  * Read in (if necessary) the block and return a buffer pointer.
  */
-struct buf *
-bread(dev, blkno)
-dev_t dev;
-daddr_t blkno;
+bread(vp, blkno, size, cred, bpp)
+	struct vnode *vp;
+	daddr_t blkno;
+	int size;
+	struct ucred *cred;
+	struct buf **bpp;
+#ifdef SECSIZE
+	long secsize;
+#endif SECSIZE
 {
 	register struct buf *bp;
 
-	bp = getblk(dev, blkno);
-	if (bp->b_flags&B_DONE) {
-#ifdef	DISKMON
-		io_info.ncache++;
-#endif
-		return(bp);
+	if (size == 0)
+		panic("bread: size 0");
+#ifdef SECSIZE
+	bp = getblk(dev, blkno, size, secsize);
+#else SECSIZE
+	*bpp = bp = getblk(vp, blkno, size);
+#endif SECSIZE
+	if (bp->b_flags&(B_DONE|B_DELWRI)) {
+		trace(TR_BREADHIT, pack(vp->v_mount->m_fsid[0], size), blkno);
+		return (0);
 	}
 	bp->b_flags |= B_READ;
-	bp->b_bcount = BSIZE;
-	(*bdevsw[major(dev)].d_strategy)(bp);
-#ifdef	DISKMON
-	io_info.nread++;
-#endif
-	u.u_vm.vm_inblk++;		/* pay for read */
-	iowait(bp);
-	return(bp);
+	if (bp->b_bcount > bp->b_bufsize)
+		panic("bread");
+	if (bp->b_rcred == NOCRED && cred != NOCRED) {
+		crhold(cred);
+		bp->b_rcred = cred;
+	}
+	VOP_STRATEGY(bp);
+	trace(TR_BREADMISS, pack(vp->v_mount->m_fsid[0], size), blkno);
+	u.u_ru.ru_inblock++;		/* pay for read */
+	return (biowait(bp));
 }
 
 /*
  * Read in the block, like bread, but also start I/O on the
  * read-ahead block (which is not allocated to the caller)
  */
-struct buf *
-breada(dev, blkno, rablkno)
-dev_t dev;
-daddr_t blkno, rablkno;
+breada(vp, blkno, size, rablkno, rabsize, cred, bpp)
+	struct vnode *vp;
+	daddr_t blkno; int size;
+#ifdef SECSIZE
+	long secsize;
+#endif SECSIZE
+	daddr_t rablkno; int rabsize;
+	struct ucred *cred;
+	struct buf **bpp;
 {
 	register struct buf *bp, *rabp;
 
 	bp = NULL;
-	if (!incore(dev, blkno)) {
-		bp = getblk(dev, blkno);
-		if ((bp->b_flags&B_DONE) == 0) {
+	/*
+	 * If the block isn't in core, then allocate
+	 * a buffer and initiate i/o (getblk checks
+	 * for a cache hit).
+	 */
+	if (!incore(vp, blkno)) {
+		*bpp = bp = getblk(vp, blkno, size);
+#endif SECSIZE
+		if ((bp->b_flags&(B_DONE|B_DELWRI)) == 0) {
 			bp->b_flags |= B_READ;
-			bp->b_bcount = BSIZE;
-			(*bdevsw[major(dev)].d_strategy)(bp);
-#ifdef	DISKMON
-			io_info.nread++;
-#endif
-			u.u_vm.vm_inblk++;		/* pay for read */
-		}
+			if (bp->b_bcount > bp->b_bufsize)
+				panic("breada");
+			if (bp->b_rcred == NOCRED && cred != NOCRED) {
+				crhold(cred);
+				bp->b_rcred = cred;
+			}
+			VOP_STRATEGY(bp);
+			trace(TR_BREADMISS, pack(vp->v_mount->m_fsid[0], size),
+			    blkno);
+			u.u_ru.ru_inblock++;		/* pay for read */
+		} else
+			trace(TR_BREADHIT, pack(vp->v_mount->m_fsid[0], size),
+			    blkno);
 	}
-	if (rablkno && !incore(dev, rablkno)) {
-		rabp = getblk(dev, rablkno);
-		if (rabp->b_flags & B_DONE)
+
+	/*
+	 * If there's a read-ahead block, start i/o
+	 * on it also (as above).
+	 */
+	if (rablkno && !incore(vp, rablkno)) {
+		rabp = getblk(vp, rablkno, rabsize);
+#endif SECSIZE
+		if (rabp->b_flags & (B_DONE|B_DELWRI)) {
 			brelse(rabp);
-		else {
+			trace(TR_BREADHITRA,
+			    pack(vp->v_mount->m_fsid[0], rabsize), blkno);
+		} else {
 			rabp->b_flags |= B_READ|B_ASYNC;
-			rabp->b_bcount = BSIZE;
-			(*bdevsw[major(dev)].d_strategy)(rabp);
-#ifdef	DISKMON
-			io_info.nreada++;
-#endif
-			u.u_vm.vm_inblk++;		/* pay in advance */
+			if (rabp->b_bcount > rabp->b_bufsize)
+				panic("breadrabp");
+			if (bp->b_rcred == NOCRED && cred != NOCRED) {
+				crhold(cred);
+				bp->b_rcred = cred;
+			}
+			VOP_STRATEGY(rabp);
+			trace(TR_BREADMISSRA,
+			    pack(vp->v_mount->m_fsid[0], rabsize), rablock);
+			u.u_ru.ru_inblock++;		/* pay in advance */
 		}
 	}
-	if(bp == NULL)
-		return(bread(dev, blkno));
-	iowait(bp);
-	return(bp);
+
+	/*
+	 * 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)
+#ifdef SECSIZE
+		return (bread(dev, blkno, size, secsize));
+#else SECSIZE
+		return (bread(vp, blkno, size, cred, bpp));
+	return (biowait(bp));
 }
 
 /*
@@ -164,26 +149,34 @@ daddr_t blkno, rablkno;
  * Then release the buffer.
  */
 bwrite(bp)
-register struct buf *bp;
+	register struct buf *bp;
 {
-	register flag;
+	register int flag;
+	int error;
 
 	flag = bp->b_flags;
-	bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI | B_AGE);
-	bp->b_bcount = BSIZE;
-#ifdef	DISKMON
-	io_info.nwrite++;
-#endif
+	bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI);
 	if ((flag&B_DELWRI) == 0)
-		u.u_vm.vm_oublk++;		/* noone paid yet */
-	(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
+		u.u_ru.ru_oublock++;		/* noone paid yet */
+	trace(TR_BWRITE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bcount), bp->b_blkno);
+	if (bp->b_bcount > bp->b_bufsize)
+		panic("bwrite");
+	VOP_STRATEGY(bp);
+
+	/*
+	 * 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.
+	 */
 	if ((flag&B_ASYNC) == 0) {
-		iowait(bp);
+		error = biowait(bp);
 		brelse(bp);
-	} else if (flag & B_DELWRI)
+	} else if (flag & B_DELWRI) {
 		bp->b_flags |= B_AGE;
-	else
-		geterror(bp);
+		error = 0;
+	}
+	return (error);
 }
 
 /*
@@ -195,66 +188,92 @@ register struct buf *bp;
  * in the same order as requested.
  */
 bdwrite(bp)
-register struct buf *bp;
+	register struct buf *bp;
 {
-	register struct buf *dp;
 
 	if ((bp->b_flags&B_DELWRI) == 0)
-		u.u_vm.vm_oublk++;		/* noone paid yet */
-	dp = bdevsw[major(bp->b_dev)].d_tab;
-	if(dp->b_flags & B_TAPE)
+		u.u_ru.ru_oublock++;		/* noone paid yet */
+#ifdef notdef
+	/*
+	 * This does not work for buffers associated with
+	 * vnodes that are remote - they have no dev.
+	 * Besides, we don't use bio with tapes, so rather
+	 * than develop a fix, we just ifdef this out for now.
+	 */
+	if (bdevsw[major(bp->b_dev)].d_flags & B_TAPE)
 		bawrite(bp);
 	else {
 		bp->b_flags |= B_DELWRI | B_DONE;
 		brelse(bp);
 	}
+#endif
+	bp->b_flags |= B_DELWRI | B_DONE;
+	brelse(bp);
 }
 
 /*
  * Release the buffer, start I/O on it, but don't wait for completion.
  */
 bawrite(bp)
-register struct buf *bp;
+	register struct buf *bp;
 {
 
 	bp->b_flags |= B_ASYNC;
-	bwrite(bp);
+	(void) bwrite(bp);
 }
 
 /*
- * release the buffer, with no I/O implied.
+ * Release the buffer, with no I/O implied.
  */
 brelse(bp)
-register struct buf *bp;
+	register struct buf *bp;
 {
-	register struct buf **backp;
+	register struct buf *flist;
 	register s;
 
+	trace(TR_BRELSE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bufsize), bp->b_blkno);
+	/*
+	 * If someone's waiting for the buffer, or
+	 * is waiting for a buffer wake 'em up.
+	 */
 	if (bp->b_flags&B_WANTED)
 		wakeup((caddr_t)bp);
-	if (bfreelist.b_flags&B_WANTED) {
-		bfreelist.b_flags &= ~B_WANTED;
-		wakeup((caddr_t)&bfreelist);
+	if (bfreelist[0].b_flags&B_WANTED) {
+		bfreelist[0].b_flags &= ~B_WANTED;
+		wakeup((caddr_t)bfreelist);
 	}
-	if ((bp->b_flags&B_ERROR) && bp->b_dev != NODEV) {
-		bunhash(bp);
-		bp->b_dev = NODEV;  /* no assoc. on error */
+	if (bp->b_flags & B_NOCACHE) {
+		bp->b_flags |= B_INVAL;
 	}
-	s = spl6();
-	if(bp->b_flags & (B_AGE|B_ERROR)) {
-		backp = &bfreelist.av_forw;
-		(*backp)->av_back = bp;
-		bp->av_forw = *backp;
-		*backp = bp;
-		bp->av_back = &bfreelist;
+	if (bp->b_flags&B_ERROR)
+		if (bp->b_flags & B_LOCKED)
+			bp->b_flags &= ~B_ERROR;	/* try again later */
+		else
+			brelvp(bp); 	 		/* no assoc */
+
+	/*
+	 * 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 */
+		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 {
-		backp = &bfreelist.av_back;
-		(*backp)->av_forw = bp;
-		bp->av_back = *backp;
-		*backp = bp;
-		bp->av_forw = &bfreelist;
+		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);
+	bp->b_flags &= ~(B_WANTED|B_BUSY|B_ASYNC|B_AGE|B_NOCACHE);
 	splx(s);
 }
 
@@ -262,28 +281,36 @@ register struct buf *bp;
  * 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;
+incore(vp, blkno)
+	struct vnode *vp;
+	daddr_t blkno;
 {
 	register struct buf *bp;
-	register int dblkno = fsbtodb(blkno);
+	register struct buf *dp;
 
-	for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1];
-	    bp = &buf[bp->b_hlink])
-		if (bp->b_blkno == dblkno && bp->b_dev == dev)
+	dp = BUFHASH(vp, blkno);
+	for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
+		if (bp->b_blkno == blkno && bp->b_vp == vp &&
+		    (bp->b_flags & B_INVAL) == 0)
 			return (1);
 	return (0);
 }
 
-struct buf *
-baddr(dev, blkno)
-dev_t dev;
-daddr_t blkno;
+baddr(vp, blkno, size, cred, bpp)
+	struct vnode *vp;
+	daddr_t blkno;
+	int size;
+	struct ucred *cred;
+	struct buf **bpp;
+#ifdef SECSIZE
+	long secsize;
+#endif SECSIZE
 {
 
-	if (incore(dev, blkno))
-		return (bread(dev, blkno));
+	if (incore(vp, blkno))
+		return (bread(vp, blkno, size, cred, bpp));
+	*bpp = 0;
+#endif SECSIZE
 	return (0);
 }
 
@@ -291,93 +318,97 @@ daddr_t blkno;
  * 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.
+ *
+ * If we find the buffer, but it is dirty (marked DELWRI) and
+ * its size is changing, we must write it out first. When the
+ * buffer is shrinking, the write is done by brealloc to avoid
+ * losing the unwritten data. When the buffer is growing, the
+ * write is done by getblk, so that bread will not read stale
+ * disk data over the modified data in the buffer.
+ *
+ * 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)
-dev_t dev;
-daddr_t blkno;
+#ifdef SECSIZE
+getblk(dev, blkno, size, secsize)
+#else SECSIZE
+getblk(vp, blkno, size)
+	register struct vnode *vp;
+	daddr_t blkno;
+	int size;
+#ifdef SECSIZE
+	long secsize;
+#endif SECSIZE
 {
-	register struct buf *bp, *dp, *ep;
-	register int i, x;
-	register int dblkno = fsbtodb(blkno);
-
-    loop:
-	VOID spl0();
-	for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1];
-	    bp = &buf[bp->b_hlink]) {
-		if (bp->b_blkno != dblkno || bp->b_dev != dev)
+	register struct buf *bp, *dp;
+	int s;
+
+	if (size > MAXBSIZE)
+		panic("getblk: size too big");
+	/*
+	 * To prevent overflow of 32-bit ints when converting block
+	 * numbers to byte offsets, blknos > 2^32 / DEV_BSIZE are set
+	 * to the maximum number that can be converted to a byte offset
+	 * without overflow. This is historic code; what bug it fixed,
+	 * or whether it is still a reasonable thing to do is open to
+	 * dispute. mkm 9/85
+	 */
+	if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-DEV_BSHIFT))
+		blkno = 1 << ((sizeof(int)*NBBY-DEV_BSHIFT) + 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.
+	 */
+	dp = BUFHASH(vp, blkno);
+loop:
+	for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
+		if (bp->b_blkno != blkno || bp->b_vp != vp ||
+		    bp->b_flags&B_INVAL)
 			continue;
-		VOID spl6();
+		s = splbio();
 		if (bp->b_flags&B_BUSY) {
 			bp->b_flags |= B_WANTED;
 			sleep((caddr_t)bp, PRIBIO+1);
+			splx(s);
 			goto loop;
 		}
-		VOID spl0();
-#ifdef	DISKMON
-		i = 0;
-		dp = bp->av_forw;
-		while (dp != &bfreelist) {
-			i++;
-			dp = dp->av_forw;
-		}
-		if (i<NBUF)
-			io_info.bufcount[i]++;
-#endif
+		splx(s);
 		notavail(bp);
+		if (bp->b_bcount != size) {
+			if (bp->b_bcount < size && (bp->b_flags&B_DELWRI)) {
+				bp->b_flags &= ~B_ASYNC;
+				(void) bwrite(bp);
+				goto loop;
+			}
+			if (brealloc(bp, size) == 0)
+				goto loop;
+		}
+		if (bp->b_bcount != size && brealloc(bp, size) == 0)
+			goto loop;
 		bp->b_flags |= B_CACHE;
-		return(bp);
-	}
-	if (major(dev) >= nblkdev)
-		panic("blkdev");
-	dp = bdevsw[major(dev)].d_tab;
-	if (dp == NULL)
-		panic("devtab");
-	VOID spl6();
-	if (bfreelist.av_forw == &bfreelist) {
-		bfreelist.b_flags |= B_WANTED;
-		sleep((caddr_t)&bfreelist, PRIBIO+1);
-		goto loop;
+		return (bp);
 	}
-	spl0();
-	bp = bfreelist.av_forw;
-	notavail(bp);
-	if (bp->b_flags & B_DELWRI) {
-		bp->b_flags |= B_ASYNC;
-		bwrite(bp);
+	bp = getnewbuf();
+	bfree(bp);
+	bremhash(bp);
+	if (bp->b_vp)
+		brelvp(bp);
+	VREF(vp);
+	bp->b_vp = vp;
+	bp->b_dev = vp->v_rdev;
+#ifdef SECSIZE
+	bp->b_blksize = secsize;
+#endif SECSIZE
+	bp->b_blkno = blkno;
+	bp->b_error = 0;
+	bp->b_resid = 0;
+	binshash(bp, dp);
+	if (brealloc(bp, size) == 0)
 		goto loop;
-	}
-	if (bp->b_dev == NODEV)
-		goto done;
-	/* INLINE EXPANSION OF bunhash(bp) */
-	i = BUFHASH(dbtofsb(bp->b_blkno));
-	x = bp - buf;
-	if (bufhash[i] == x) {
-		bufhash[i] = bp->b_hlink;
-	} else {
-		for (ep = &buf[bufhash[i]]; ep != &buf[-1];
-		    ep = &buf[ep->b_hlink])
-			if (ep->b_hlink == x) {
-				ep->b_hlink = bp->b_hlink;
-				goto done;
-			}
-		panic("getblk");
-	}
-done:
-	/* END INLINE EXPANSION */
-	bp->b_flags = B_BUSY;
-	bp->b_back->b_forw = bp->b_forw;
-	bp->b_forw->b_back = bp->b_back;
-	bp->b_forw = dp->b_forw;
-	bp->b_back = dp;
-	dp->b_forw->b_back = bp;
-	dp->b_forw = bp;
-	bp->b_dev = dev;
-	bp->b_blkno = dblkno;
-	i = BUFHASH(blkno);
-	bp->b_hlink = bufhash[i];
-	bufhash[i] = bp - buf;
-	return(bp);
+	return (bp);
 }
 
 /*
@@ -385,122 +416,202 @@ done:
  * not assigned to any particular device
  */
 struct buf *
-geteblk()
+geteblk(size)
+	int size;
 {
-	register struct buf *bp, *dp, *ep;
-	register int i, x;
+	register struct buf *bp, *flist;
 
+	if (size > MAXBSIZE)
+		panic("geteblk: size too big");
 loop:
-	VOID spl6();
-	while (bfreelist.av_forw == &bfreelist) {
-		bfreelist.b_flags |= B_WANTED;
-		sleep((caddr_t)&bfreelist, PRIBIO+1);
-	}
-	VOID spl0();
-	dp = &bfreelist;
-	bp = bfreelist.av_forw;
-	notavail(bp);
-	if (bp->b_flags & B_DELWRI) {
-		bp->b_flags |= B_ASYNC;
-		bwrite(bp);
+	bp = getnewbuf();
+	bp->b_flags |= B_INVAL;
+	bfree(bp);
+	bremhash(bp);
+	flist = &bfreelist[BQ_AGE];
+	brelvp(bp);
+#ifdef SECSIZE
+	bp->b_blksize = DEV_BSIZE;
+#endif SECSIZE
+	bp->b_error = 0;
+	bp->b_resid = 0;
+	binshash(bp, flist);
+	if (brealloc(bp, size) == 0)
 		goto loop;
-	}
-	if (bp->b_dev != NODEV)
-		bunhash(bp);
-	bp->b_flags = B_BUSY;
-	bp->b_back->b_forw = bp->b_forw;
-	bp->b_forw->b_back = bp->b_back;
-	bp->b_forw = dp->b_forw;
-	bp->b_back = dp;
-	dp->b_forw->b_back = bp;
-	dp->b_forw = bp;
-	bp->b_dev = (dev_t)NODEV;
-	bp->b_hlink = -1;
-	return(bp);
+	return (bp);
 }
 
-bunhash(bp)
+/*
+ * Allocate space associated with a buffer.
+ * If can't get space, buffer is released
+ */
+brealloc(bp, size)
 	register struct buf *bp;
+	int size;
 {
+	daddr_t start, last;
 	register struct buf *ep;
-	register int i, x;
+	struct buf *dp;
+	int s;
 
-	if (bp->b_dev == NODEV)
-		return;
-	i = BUFHASH(dbtofsb(bp->b_blkno));
-	x = bp - buf;
-	if (bufhash[i] == x) {
-		bufhash[i] = bp->b_hlink;
-		return;
+	/*
+	 * First need to make sure that all overlapping previous I/O
+	 * is dispatched with.
+	 */
+	if (size == bp->b_bcount)
+		return (1);
+	if (size < bp->b_bcount) { 
+		if (bp->b_flags & B_DELWRI) {
+			(void) bwrite(bp);
+			return (0);
+		}
+		if (bp->b_flags & B_LOCKED)
+			panic("brealloc");
+		return (allocbuf(bp, size));
 	}
-	for (ep = &buf[bufhash[i]]; ep != &buf[-1];
-	    ep = &buf[ep->b_hlink])
-		if (ep->b_hlink == x) {
-			ep->b_hlink = bp->b_hlink;
-			return;
+	bp->b_flags &= ~B_DONE;
+	if (bp->b_vp == (struct vnode *)0)
+		return (allocbuf(bp, size));
+
+	trace(TR_BREALLOC,
+	    pack(bp->b_vp->v_mount->m_fsid[0], size), bp->b_blkno);
+	/*
+	 * 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.
+	 */
+	start = bp->b_blkno;
+#ifdef SECSIZE
+	last = start + size/bp->b_blksize - 1;
+#else SECSIZE
+	last = start + btodb(size) - 1;
+#endif SECSIZE
+	dp = BUFHASH(bp->b_vp, bp->b_blkno);
+loop:
+	for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
+		if (ep == bp || ep->b_vp != bp->b_vp ||
+		    (ep->b_flags & B_INVAL))
+			continue;
+		/* look for overlap */
+		if (ep->b_bcount == 0 || ep->b_blkno > last ||
+#ifdef SECSIZE
+		    ep->b_blkno + ep->b_bcount/ep->b_blksize <= start)
+#else SECSIZE
+		    ep->b_blkno + btodb(ep->b_bcount) <= start)
+#endif SECSIZE
+			continue;
+		s = splbio();
+		if (ep->b_flags&B_BUSY) {
+			ep->b_flags |= B_WANTED;
+			sleep((caddr_t)ep, PRIBIO+1);
+			splx(s);
+			goto loop;
 		}
-	panic("bunhash");
+		splx(s);
+		notavail(ep);
+		if (ep->b_flags & B_DELWRI) {
+			(void) bwrite(ep);
+			goto loop;
+		}
+		ep->b_flags |= B_INVAL;
+		brelse(ep);
+	}
+	return (allocbuf(bp, size));
 }
 
 /*
- * Wait for I/O completion on the buffer; return errors
- * to the user.
+ * Find a buffer which is available for use.
+ * Select something from a free list.
+ * Preference is to AGE list, then LRU list.
  */
-iowait(bp)
-register struct buf *bp;
+struct buf *
+getnewbuf()
 {
+	register struct buf *bp, *dp;
+	register struct ucred *cred;
+	int s;
 
-	VOID spl6();
-	while ((bp->b_flags&B_DONE)==0)
-		sleep((caddr_t)bp, PRIBIO);
-	VOID spl0();
-	geterror(bp);
+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);
+		splx(s);
+		goto loop;
+	}
+	splx(s);
+	bp = dp->av_forw;
+	notavail(bp);
+	if (bp->b_flags & B_DELWRI) {
+		(void) bawrite(bp);
+		goto loop;
+	}
+	trace(TR_BRELSE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bufsize), bp->b_blkno);
+	brelvp(bp);
+	if (bp->b_rcred != NOCRED) {
+		cred = bp->b_rcred;
+		bp->b_rcred = NOCRED;
+		crfree(cred);
+	}
+	if (bp->b_wcred != NOCRED) {
+		cred = bp->b_wcred;
+		bp->b_wcred = NOCRED;
+		crfree(cred);
+	}
+	bp->b_flags = B_BUSY;
+	return (bp);
 }
 
-#ifndef FASTVAX
 /*
- * Unlink a buffer from the available list and mark it busy.
- * (internal interface)
+ * Wait for I/O completion on the buffer; return errors
+ * to the user.
  */
-notavail(bp)
-register struct buf *bp;
+biowait(bp)
+	register struct buf *bp;
 {
-	register s;
+	int s;
 
-	s = spl6();
-	bp->av_back->av_forw = bp->av_forw;
-	bp->av_forw->av_back = bp->av_back;
-	bp->b_flags |= B_BUSY;
+	s = splbio();
+	while ((bp->b_flags & B_DONE) == 0)
+		sleep((caddr_t)bp, PRIBIO);
 	splx(s);
+	/*
+	 * 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.
+	 */
+	if ((bp->b_flags & B_ERROR) == 0)
+		return (0);
+	if (bp->b_error)
+		return (bp->b_error);
+	return (EIO);
 }
-#endif
 
 /*
- * 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.
+ * Mark I/O complete on a buffer.
+ * If someone should be called, e.g. the pageout
+ * daemon, do so.  Otherwise, wake up anyone
+ * waiting for it.
  */
-iodone(bp)
-register struct buf *bp;
+biodone(bp)
+	register struct buf *bp;
 {
-	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();
-		cnt.v_pgout++;
-		bp->av_forw = bclnlist;
-		bp->b_bcount = swsize[bp - swbuf];
-		bp->b_pfcent = swpf[bp - swbuf];
-		bclnlist = bp;
-		if (bswlist.b_flags & B_WANTED)
-			wakeup((caddr_t)&proc[2]);
-		splx(s);
+	if ((bp->b_flags & B_READ) == 0)
+		bp->b_dirtyoff = bp->b_dirtyend = 0;
+	if (bp->b_flags & B_CALL) {
+		bp->b_flags &= ~B_CALL;
+		(*bp->b_iodone)(bp);
+		return;
 	}
 	if (bp->b_flags&B_ASYNC)
 		brelse(bp);
@@ -511,247 +622,147 @@ register struct buf *bp;
 }
 
 /*
- * Zero the core associated with a buffer.
+ * Ensure that no part of a specified block is in an incore buffer.
+#ifdef SECSIZE
+ * "size" is given in device blocks (the units of b_blkno).
+#endif SECSIZE
+#ifdef SECSIZE
+ * "size" is given in device blocks (the units of b_blkno).
+#endif SECSIZE
  */
-clrbuf(bp)
-struct buf *bp;
+blkflush(vp, blkno, size)
+	struct vnode *vp;
+	daddr_t blkno;
+#ifdef SECSIZE
+	int size;
+#else SECSIZE
+	long size;
+#endif SECSIZE
 {
-	register *p;
-	register c;
-
-	p = bp->b_un.b_words;
-	c = BSIZE/sizeof(int);
-	do
-		*p++ = 0;
-	while (--c);
-	bp->b_resid = 0;
-}
-
-/*
- * swap I/O -
- *
- * If the flag indicates a dirty page push initiated
- * by the pageout daemon, we map the page into the i th
- * virtual page of process 2 (the daemon itself) where i is
- * the index of the swap header that has been allocated.
- * We simply initialize the header and queue the I/O but
- * do not wait for completion. When the I/O completes,
- * iodone() will link the header to a list of cleaned
- * pages to be processed by the pageout daemon.
- */
-swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
-	struct proc *p;
-	swblk_t dblkno;
-	caddr_t addr;
-	int flag, nbytes;
-	dev_t dev;
-	unsigned pfcent;
-{
-	register struct buf *bp;
-	register int c;
-	int p2dp;
-	register struct pte *dpte, *vpte;
-
-	VOID spl6();
-	while (bswlist.av_forw == NULL) {
-		bswlist.b_flags |= B_WANTED;
-		sleep((caddr_t)&bswlist, PSWP+1);
-	}
-	bp = bswlist.av_forw;
-	bswlist.av_forw = bp->av_forw;
-	VOID spl0();
-
-	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
-	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
-		if (rdflg == B_READ)
-			sum.v_pswpin += btoc(nbytes);
-		else
-			sum.v_pswpout += btoc(nbytes);
-	bp->b_proc = p;
-	if (flag & B_DIRTY) {
-		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
-		dpte = dptopte(&proc[2], p2dp);
-		vpte = vtopte(p, btop(addr));
-		for (c = 0; c < nbytes; c += NBPG) {
-			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
-				panic("swap bad pte");
-			*dpte++ = *vpte++;
-		}
-		bp->b_un.b_addr = (caddr_t)ctob(p2dp);
-	} else
-		bp->b_un.b_addr = addr;
-	while (nbytes > 0) {
-		c = imin(ctob(120), nbytes);
-		bp->b_bcount = c;
-		bp->b_blkno = dblkno;
-		bp->b_dev = dev;
-		if (dev == swapdev)
-			bp->b_blkno += swplo;
-		(*bdevsw[major(dev)].d_strategy)(bp);
-		if (flag & B_DIRTY) {
-			if (c < nbytes)
-				panic("big push");
-			swsize[bp - swbuf] = nbytes;
-			swpf[bp - swbuf] = pfcent;
-			return;
+	register struct buf *ep;
+	struct buf *dp;
+	daddr_t start, last;
+	int s, error, allerrors = 0;
+
+	start = blkno;
+#ifdef SECSIZE
+	last = start + size - 1;
+#else SECSIZE
+	last = start + btodb(size) - 1;
+#endif SECSIZE
+	dp = BUFHASH(vp, blkno);
+loop:
+	for (ep = dp->b_forw; ep != dp; ep = ep->b_forw) {
+		if (ep->b_vp != vp || (ep->b_flags & B_INVAL))
+			continue;
+		/* look for overlap */
+		if (ep->b_bcount == 0 || ep->b_blkno > last ||
+#ifdef SECSIZE
+		    ep->b_blkno + ep->b_bcount / ep->b_blksize <= start)
+#else SECSIZE
+		    ep->b_blkno + btodb(ep->b_bcount) <= start)
+#endif SECSIZE
+			continue;
+		s = splbio();
+		if (ep->b_flags&B_BUSY) {
+			ep->b_flags |= B_WANTED;
+			sleep((caddr_t)ep, PRIBIO+1);
+			splx(s);
+			goto loop;
 		}
-		VOID spl6();
-		while((bp->b_flags&B_DONE)==0)
-			sleep((caddr_t)bp, PSWP);
-		VOID spl0();
-		bp->b_un.b_addr += c;
-		bp->b_flags &= ~B_DONE;
-		if (bp->b_flags & B_ERROR) {
-			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
-				panic("hard IO err in swap");
-			swkill(p, (char *)0);
+		if (ep->b_flags & B_DELWRI) {
+			splx(s);
+			notavail(ep);
+			if (error = bwrite(ep))
+				allerrors = error;
+			goto loop;
 		}
-		nbytes -= c;
-		dblkno += btoc(c);
-	}
-	VOID spl6();
-	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
-	bp->av_forw = bswlist.av_forw;
-	bswlist.av_forw = bp;
-	if (bswlist.b_flags & B_WANTED) {
-		bswlist.b_flags &= ~B_WANTED;
-		wakeup((caddr_t)&bswlist);
-		wakeup((caddr_t)&proc[2]);
+		splx(s);
 	}
-	VOID spl0();
+	return (allerrors);
 }
 
 /*
- * If rout == 0 then killed on swap error, else
- * rout is the name of the routine where we ran out of
- * swap space.
+ * Make sure all write-behind blocks associated
+ * with mount point are flushed out (from sync).
  */
-swkill(p, rout)
-	struct proc *p;
-	char *rout;
+bflush(mountp)
+	struct mount *mountp;
 {
+	register struct buf *bp;
+	register struct buf *flist;
+	int s;
 
-	printf("%d: ", p->p_pid);
-	if (rout)
-		printf("out of swap space in %s\n", rout);
-	else
-		printf("killed on swap error\n");
-	/*
-	 * To be sure no looping (e.g. in vmsched trying to
-	 * swap out) mark process locked in core (as though
-	 * done by user) after killing it so noone will try
-	 * to swap it out.
-	 */
-	psignal(p, SIGKIL);
-	p->p_flag |= SULOCK;
+loop:
+	s = splbio();
+	for (flist = bfreelist; flist < &bfreelist[BQ_EMPTY]; flist++) {
+		for (bp = flist->av_forw; bp != flist; bp = bp->av_forw) {
+			if ((bp->b_flags & B_BUSY))
+				continue;
+			if ((bp->b_flags & B_DELWRI) == 0)
+				continue;
+			if (bp->b_vp && bp->b_vp->v_mount == mountp) {
+				notavail(bp);
+				(void) bawrite(bp);
+				splx(s);
+				goto loop;
+			}
+		}
+	}
+	splx(s);
 }
 
 /*
- * make sure all write-behind blocks
- * on dev (or NODEV for all)
- * are flushed out.
- * (from umount and update)
+ * Invalidate in core blocks belonging to closed or umounted filesystem
+ *
+ * We walk through the buffer pool and invalidate any buffers for the
+ * indicated mount point. 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. We return the count of dirty buffers when
+ * we are finished.
  */
-bflush(dev)
-dev_t dev;
+binval(mountp)
+	struct mount *mountp;
 {
 	register struct buf *bp;
+	register struct bufhd *hp;
+	int s, dirty = 0;
+#define dp ((struct buf *)hp)
 
 loop:
-	VOID spl6();
-	for (bp = bfreelist.av_forw; bp != &bfreelist; bp = bp->av_forw) {
-		if (bp->b_flags&B_DELWRI && (dev == NODEV||dev==bp->b_dev)) {
-			bp->b_flags |= B_ASYNC;
+	s = splbio();
+	for (hp = bufhash; hp < &bufhash[BUFHSZ]; hp++) {
+		for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
+			if (bp->b_vp == NULL || bp->b_vp->v_mount != mountp)
+				continue;
+			if (bp->b_flags & B_BUSY) {
+				bp->b_flags |= B_WANTED;
+				sleep((caddr_t)bp, PRIBIO+1);
+				splx(s);
+				goto loop;
+			}
 			notavail(bp);
-			bwrite(bp);
-			goto loop;
+			if (bp->b_flags & B_DELWRI) {
+				(void) bawrite(bp);
+				dirty++;
+				continue;
+			}
+			bp->b_flags |= B_INVAL;
+			brelvp(bp);
+			brelse(bp);
 		}
 	}
-	VOID spl0();
+	return (dirty);
 }
 
-/*
- * Raw I/O. The arguments are
- *	The strategy routine for the device
- *	A buffer, which will always be a special buffer
- *	  header owned exclusively by the device for this purpose
- *	The device number
- *	Read/write flag
- * Essentially all the work is computing physical addresses and
- * validating them.
- * If the user has the proper access privilidges, the process is
- * marked 'delayed unlock' and the pages involved in the I/O are
- * faulted and locked. After the completion of the I/O, the above pages
- * are unlocked.
- */
-physio(strat, bp, dev, rw, mincnt)
-int (*strat)(); 
-register struct buf *bp;
-unsigned (*mincnt)();
+brelvp(bp)
+	struct buf *bp;
 {
-	register int c;
-	char *a;
+	struct vnode *vp;
 
-	if (useracc(u.u_base,u.u_count,rw==B_READ?B_WRITE:B_READ) == NULL) {
-		u.u_error = EFAULT;
+	if (bp->b_vp == (struct vnode *) 0)
 		return;
-	}
-	VOID spl6();
-	while (bp->b_flags&B_BUSY) {
-		bp->b_flags |= B_WANTED;
-		sleep((caddr_t)bp, PRIBIO+1);
-	}
-	bp->b_error = 0;
-	bp->b_proc = u.u_procp;
-	bp->b_un.b_addr = u.u_base;
-	while (u.u_count != 0 && bp->b_error==0) {
-		bp->b_flags = B_BUSY | B_PHYS | rw;
-		bp->b_dev = dev;
-		bp->b_blkno = u.u_offset >> PGSHIFT;
-		bp->b_bcount = u.u_count;
-		(*mincnt)(bp);
-		c = bp->b_bcount;
-		u.u_procp->p_flag |= SPHYSIO;
-		vslock(a = bp->b_un.b_addr, c);
-		(*strat)(bp);
-		VOID spl6();
-		while ((bp->b_flags&B_DONE) == 0)
-			sleep((caddr_t)bp, PRIBIO);
-		vsunlock(a, c, rw);
-		u.u_procp->p_flag &= ~SPHYSIO;
-		if (bp->b_flags&B_WANTED)
-			wakeup((caddr_t)bp);
-		VOID spl0();
-		bp->b_un.b_addr += c;
-		u.u_count -= c;
-		u.u_offset += c;
-	}
-	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
-	u.u_count = bp->b_resid;
-	geterror(bp);
-}
-
-/*ARGSUSED*/
-unsigned
-minphys(bp)
-struct buf *bp;
-{
-
-	if (bp->b_bcount > 60 * 1024)
-		bp->b_bcount = 60 * 1024;
-}
-
-/*
- * 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;
-{
-
-	if (bp->b_flags&B_ERROR)
-		if ((u.u_error = bp->b_error)==0)
-			u.u_error = EIO;
+	vp = bp->b_vp;
+	bp->b_vp = (struct vnode *) 0;
+	vrele(vp);
 }