X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/af04ce661ff8505e342753def3016843d663d7cb..c02aa1e03cd871f2719c7b4d094f14662b274093:/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 8adaf7d56d..a7028450d4 100644
--- a/usr/src/sys/kern/vfs_bio.c
+++ b/usr/src/sys/kern/vfs_bio.c
@@ -1,55 +1,82 @@
-/*	vfs_bio.c	4.47	83/06/14	*/
-
-#include "../machine/pte.h"
+/*
+ * 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_bio.c	7.17 (Berkeley) %G%
+ */
 
-#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 "param.h"
+#include "user.h"
+#include "buf.h"
+#include "vnode.h"
+#include "mount.h"
+#include "trace.h"
+#include "ucred.h"
 
 /*
  * Read in (if necessary) the block and return a buffer pointer.
  */
-struct buf *
-bread(dev, blkno, size)
-	dev_t dev;
+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;
 
 	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);
+#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;
 	if (bp->b_bcount > bp->b_bufsize)
 		panic("bread");
-	(*bdevsw[major(dev)].d_strategy)(bp);
-	trace(TR_BREADMISS, dev, blkno);
+	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 */
-	biowait(bp);
-	return(bp);
+	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, size, rablkno, rabsize)
-	dev_t dev;
+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;
 
@@ -59,34 +86,48 @@ breada(dev, blkno, size, rablkno, rabsize)
 	 * 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) {
+	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;
 			if (bp->b_bcount > bp->b_bufsize)
 				panic("breada");
-			(*bdevsw[major(dev)].d_strategy)(bp);
-			trace(TR_BREADMISS, dev, blkno);
+			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, dev, blkno);
+			trace(TR_BREADHIT, pack(vp->v_mount->m_fsid[0], size),
+			    blkno);
 	}
 
 	/*
 	 * 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) {
+	if (rablkno && !incore(vp, rablkno)) {
+		rabp = getblk(vp, rablkno, rabsize);
+#endif SECSIZE
+		if (rabp->b_flags & (B_DONE|B_DELWRI)) {
 			brelse(rabp);
-			trace(TR_BREADHITRA, dev, blkno);
+			trace(TR_BREADHITRA,
+			    pack(vp->v_mount->m_fsid[0], rabsize), rablkno);
 		} 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);
+			if (rabp->b_rcred == NOCRED && cred != NOCRED) {
+				crhold(cred);
+				rabp->b_rcred = cred;
+			}
+			VOP_STRATEGY(rabp);
+			trace(TR_BREADMISSRA,
+			    pack(vp->v_mount->m_fsid[0], rabsize), rablkno);
 			u.u_ru.ru_inblock++;		/* pay in advance */
 		}
 	}
@@ -97,9 +138,11 @@ breada(dev, blkno, size, rablkno, rabsize)
 	 * above, and just wait for it.
 	 */
 	if (bp == NULL)
-		return (bread(dev, blkno, size));
-	biowait(bp);
-	return (bp);
+#ifdef SECSIZE
+		return (bread(dev, blkno, size, secsize));
+#else SECSIZE
+		return (bread(vp, blkno, size, cred, bpp));
+	return (biowait(bp));
 }
 
 /*
@@ -109,16 +152,18 @@ breada(dev, blkno, size, rablkno, rabsize)
 bwrite(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);
 	if ((flag&B_DELWRI) == 0)
 		u.u_ru.ru_oublock++;		/* noone paid yet */
-	trace(TR_BWRITE, bp->b_dev, bp->b_blkno);
+	trace(TR_BWRITE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bcount), bp->b_lblkno);
 	if (bp->b_bcount > bp->b_bufsize)
 		panic("bwrite");
-	(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
+	VOP_STRATEGY(bp);
 
 	/*
 	 * If the write was synchronous, then await i/o completion.
@@ -126,10 +171,13 @@ bwrite(bp)
 	 * the q of blocks awaiting i/o completion status.
 	 */
 	if ((flag&B_ASYNC) == 0) {
-		biowait(bp);
+		error = biowait(bp);
 		brelse(bp);
-	} else if (flag & B_DELWRI)
+	} else if (flag & B_DELWRI) {
 		bp->b_flags |= B_AGE;
+		error = 0;
+	}
+	return (error);
 }
 
 /*
@@ -143,14 +191,15 @@ bwrite(bp)
 bdwrite(bp)
 	register struct buf *bp;
 {
-	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)
+	/*
+	 * If this is a tape drive, the write must be initiated.
+	 */
+	if (bdevsw[major(bp->b_dev)].d_flags & B_TAPE)
 		bawrite(bp);
-	else {
+	} else {
 		bp->b_flags |= B_DELWRI | B_DONE;
 		brelse(bp);
 	}
@@ -164,7 +213,7 @@ bawrite(bp)
 {
 
 	bp->b_flags |= B_ASYNC;
-	bwrite(bp);
+	(void) bwrite(bp);
 }
 
 /*
@@ -176,9 +225,11 @@ brelse(bp)
 	register struct buf *flist;
 	register s;
 
+	trace(TR_BRELSE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bufsize), bp->b_lblkno);
 	/*
-	 * If someone's waiting for the buffer, or
-	 * is waiting for a buffer wake 'em up.
+	 * If a process is waiting for the buffer, or
+	 * is waiting for a free buffer, awaken it.
 	 */
 	if (bp->b_flags&B_WANTED)
 		wakeup((caddr_t)bp);
@@ -186,16 +237,26 @@ brelse(bp)
 		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 */
+	/*
+	 * Retry I/O for locked buffers rather than invalidating them.
+	 */
+	if ((bp->b_flags & B_ERROR) && (bp->b_flags & B_LOCKED))
+		bp->b_flags &= ~B_ERROR;
 
+	/*
+	 * Disassociate buffers that are no longer valid.
+	 */
+	if (bp->b_flags & (B_NOCACHE|B_ERROR))
+		bp->b_flags |= B_INVAL;
+	if ((bp->b_bufsize <= 0) || (bp->b_flags & (B_ERROR|B_INVAL))) {
+		if (bp->b_vp)
+			brelvp(bp);
+		bp->b_flags &= ~B_DELWRI;
+	}
 	/*
 	 * Stick the buffer back on a free list.
 	 */
-	s = spl6();
+	s = splbio();
 	if (bp->b_bufsize <= 0) {
 		/* block has no buffer ... put at front of unused buffer list */
 		flist = &bfreelist[BQ_EMPTY];
@@ -213,7 +274,7 @@ brelse(bp)
 			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);
 }
 
@@ -221,30 +282,39 @@ brelse(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;
+incore(vp, blkno)
+	struct vnode *vp;
 	daddr_t blkno;
 {
 	register struct buf *bp;
 	register struct buf *dp;
 
-	dp = BUFHASH(dev, blkno);
+	dp = BUFHASH(vp, blkno);
 	for (bp = dp->b_forw; bp != dp; bp = bp->b_forw)
-		if (bp->b_blkno == blkno && bp->b_dev == dev &&
+		if (bp->b_lblkno == blkno && bp->b_vp == vp &&
 		    (bp->b_flags & B_INVAL) == 0)
 			return (1);
 	return (0);
 }
 
-struct buf *
-baddr(dev, blkno, size)
-	dev_t dev;
+/*
+ * Return a block if it is in memory.
+ */
+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, size));
+	if (incore(vp, blkno))
+		return (bread(vp, blkno, size, cred, bpp));
+	*bpp = 0;
+#endif SECSIZE
 	return (0);
 }
 
@@ -258,28 +328,48 @@ baddr(dev, blkno, size)
  * want to lower the ipl back to 0.
  */
 struct buf *
-getblk(dev, blkno, size)
-	dev_t dev;
+#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;
 	int s;
 
-	if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT))	/* XXX */
-		blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1);
+	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
+	 *
+	 * Make it a panic to see if it ever really happens. mkm 11/89
+	 */
+	if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-DEV_BSHIFT)) {
+		panic("getblk: blkno too big");
+		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(dev, blkno);
+	dp = BUFHASH(vp, blkno);
 loop:
 	for (bp = dp->b_forw; bp != dp; bp = bp->b_forw) {
-		if (bp->b_blkno != blkno || bp->b_dev != dev ||
+		if (bp->b_lblkno != blkno || bp->b_vp != vp ||
 		    bp->b_flags&B_INVAL)
 			continue;
-		s = spl6();
+		s = splbio();
 		if (bp->b_flags&B_BUSY) {
 			bp->b_flags |= B_WANTED;
 			sleep((caddr_t)bp, PRIBIO+1);
@@ -288,23 +378,29 @@ loop:
 		}
 		splx(s);
 		notavail(bp);
-		if (brealloc(bp, size) == 0)
+		if (bp->b_bcount != size) {
+			printf("getblk: stray size");
+			bp->b_flags |= B_INVAL;
+			bwrite(bp);
 			goto loop;
+		}
 		bp->b_flags |= B_CACHE;
-		return(bp);
+		return (bp);
 	}
-	if (major(dev) >= nblkdev)
-		panic("blkdev");
 	bp = getnewbuf();
 	bfree(bp);
 	bremhash(bp);
-	binshash(bp, dp);
-	bp->b_dev = dev;
+	bgetvp(vp, bp);
+	bp->b_lblkno = blkno;
+#ifdef SECSIZE
+	bp->b_blksize = secsize;
+#endif SECSIZE
 	bp->b_blkno = blkno;
 	bp->b_error = 0;
-	if (brealloc(bp, size) == 0)
-		goto loop;
-	return(bp);
+	bp->b_resid = 0;
+	binshash(bp, dp);
+	brealloc(bp, size);
+	return (bp);
 }
 
 /*
@@ -317,23 +413,25 @@ geteblk(size)
 {
 	register struct buf *bp, *flist;
 
-loop:
+	if (size > MAXBSIZE)
+		panic("geteblk: size too big");
 	bp = getnewbuf();
 	bp->b_flags |= B_INVAL;
 	bfree(bp);
 	bremhash(bp);
 	flist = &bfreelist[BQ_AGE];
-	binshash(bp, flist);
-	bp->b_dev = (dev_t)NODEV;
+#ifdef SECSIZE
+	bp->b_blksize = DEV_BSIZE;
+#endif SECSIZE
 	bp->b_error = 0;
-	if (brealloc(bp, size) == 0)
-		goto loop;
-	return(bp);
+	bp->b_resid = 0;
+	binshash(bp, flist);
+	brealloc(bp, size);
+	return (bp);
 }
 
 /*
  * Allocate space associated with a buffer.
- * If can't get space, buffer is released
  */
 brealloc(bp, size)
 	register struct buf *bp;
@@ -344,61 +442,9 @@ brealloc(bp, size)
 	struct buf *dp;
 	int s;
 
-	/*
-	 * First need to make sure that all overlaping previous I/O
-	 * is dispatched with.
-	 */
 	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));
-
-	/*
-	 * 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;
-	last = start + btodb(size) - 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 + btodb(ep->b_bcount) <= 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;
+	allocbuf(bp, size);
 }
 
 /*
@@ -410,15 +456,12 @@ struct buf *
 getnewbuf()
 {
 	register struct buf *bp, *dp;
+	register struct ucred *cred;
 	int s;
 
 loop:
-	s = spl6();
-#ifndef sun
+	s = splbio();
 	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 */
@@ -431,11 +474,23 @@ loop:
 	bp = dp->av_forw;
 	notavail(bp);
 	if (bp->b_flags & B_DELWRI) {
-		bp->b_flags |= B_ASYNC;
-		bwrite(bp);
+		(void) bawrite(bp);
 		goto loop;
 	}
-	trace(TR_BRELSE, bp->b_dev, bp->b_blkno);
+	trace(TR_BRELSE,
+	    pack(bp->b_vp->v_mount->m_fsid[0], bp->b_bufsize), bp->b_lblkno);
+	if (bp->b_vp)
+		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);
 }
@@ -449,12 +504,19 @@ biowait(bp)
 {
 	int s;
 
-	s = spl6();
-	while ((bp->b_flags&B_DONE)==0)
+	s = splbio();
+	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);
+	/*
+	 * 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);
 }
 
 /*
@@ -470,6 +532,8 @@ biodone(bp)
 	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 (bp->b_flags & B_CALL) {
 		bp->b_flags &= ~B_CALL;
 		(*bp->b_iodone)(bp);
@@ -484,113 +548,287 @@ biodone(bp)
 }
 
 /*
- * Insure that no part of a specified block is in an incore 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_lblkno).
+#endif SECSIZE
+#ifdef SECSIZE
+ * "size" is given in device blocks (the units of b_blkno).
+#endif SECSIZE
  */
-blkflush(dev, blkno, size)
-	dev_t dev;
+blkflush(vp, blkno, size)
+	struct vnode *vp;
 	daddr_t blkno;
+#ifdef SECSIZE
+	int size;
+#else SECSIZE
 	long size;
+#endif SECSIZE
 {
 	register struct buf *ep;
 	struct buf *dp;
-	daddr_t start, last;
+	daddr_t curblk, nextblk, ecurblk, lastblk;
+	int s, error, allerrors = 0;
+
+	/*
+	 * Iterate through each possible hash chain.
+	 */
+	lastblk = blkno + btodb(size) - 1;
+	for (curblk = blkno; curblk <= lastblk; curblk = nextblk) {
+#if RND & (RND-1)
+		nextblk = ((curblk / RND) + 1) * RND;
+#else
+		nextblk = ((curblk & ~(RND-1)) + RND);
+#endif
+		ecurblk = nextblk > lastblk ? lastblk : nextblk - 1;
+		dp = BUFHASH(vp, curblk);
+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_lblkno > ecurblk ||
+			    ep->b_lblkno + btodb(ep->b_bcount) <= curblk)
+				continue;
+			s = splbio();
+			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);
+				if (error = bwrite(ep))
+					allerrors = error;
+				goto loop;
+			}
+			splx(s);
+		}
+	}
+	return (allerrors);
+}
+
+/*
+ * 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;
+	struct vnode *nvp;
+
+loop:
+	for (vp = mountp->m_mounth; vp; vp = nvp) {
+		nvp = vp->v_mountf;
+		if (vget(vp))
+			goto loop;
+		vflushbuf(vp, flags);
+		vput(vp);
+	}
+}
+
+/*
+ * Flush all dirty buffers associated with a vnode.
+ */
+vflushbuf(vp, flags)
+	register struct vnode *vp;
+	int flags;
+{
+	register struct buf *bp;
+	struct buf *nbp;
 	int s;
 
-	start = blkno;
-	last = start + btodb(size) - 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))
+	s = splbio();
+	for (bp = vp->v_blockh; bp; bp = nbp) {
+		nbp = bp->b_blockf;
+		if ((bp->b_flags & B_BUSY))
 			continue;
-		/* look for overlap */
-		if (ep->b_bcount == 0 || ep->b_blkno > last ||
-		    ep->b_blkno + btodb(ep->b_bcount) <= start)
+		if ((bp->b_flags & B_DELWRI) == 0)
 			continue;
-		s = spl6();
-		if (ep->b_flags&B_BUSY) {
-			ep->b_flags |= B_WANTED;
-			sleep((caddr_t)ep, PRIBIO+1);
+		splx(s);
+		notavail(bp);
+		(void) bawrite(bp);
+		goto loop;
+	}
+	splx(s);
+	if ((flags & B_SYNC) == 0)
+		return;
+wloop:
+	s = splbio();
+	for (bp = vp->v_blockh; bp; bp = nbp) {
+		nbp = bp->b_blockf;
+		if (bp->b_flags & B_BUSY) {
+			bp->b_flags |= B_WANTED;
+			sleep((caddr_t)bp, PRIBIO+1);
 			splx(s);
-			goto loop;
+			goto wloop;
 		}
-		if (ep->b_flags & B_DELWRI) {
+		if ((bp->b_flags & B_DELWRI)) {
 			splx(s);
-			notavail(ep);
-			bwrite(ep);
 			goto loop;
 		}
-		splx(s);
 	}
+	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
+ *
+ * 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.
  */
-bflush(dev)
-	dev_t dev;
+mntinvalbuf(mountp)
+	struct mount *mountp;
+{
+	register struct vnode *vp;
+	struct vnode *nvp;
+	int dirty = 0;
+
+loop:
+	for (vp = mountp->m_mounth; vp; vp = nvp) {
+		nvp = vp->v_mountf;
+		if (vget(vp))
+			goto loop;
+		dirty += vinvalbuf(vp, 1);
+		vput(vp);
+	}
+	return (dirty);
+}
+
+/*
+ * Flush out and invalidate all buffers associated with a vnode.
+ * Called with the underlying object locked.
+ */
+vinvalbuf(vp, save)
+	register struct vnode *vp;
+	int save;
 {
 	register struct buf *bp;
-	register struct buf *flist;
-	int s;
+	struct buf *nbp;
+	int s, dirty = 0;
 
 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);
+	for (bp = vp->v_blockh; 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);
 			goto loop;
 		}
+		splx(s);
+		notavail(bp);
+		if (save) {
+			if (bp->b_flags & B_DELWRI) {
+				dirty++;
+				(void) bwrite(bp);
+				goto loop;
+			}
+		}
+		bp->b_flags |= B_INVAL;
+		brelse(bp);
 	}
-	splx(s);
+	if (vp->v_blockh != 0)
+		panic("vinvalbuf: flush failed");
+	return (dirty);
 }
 
 /*
- * 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.
+ * Associate a buffer with a vnode.
  */
-geterror(bp)
+bgetvp(vp, bp)
+	register struct vnode *vp;
 	register struct buf *bp;
 {
-	int error = 0;
 
-	if (bp->b_flags&B_ERROR)
-		if ((error = bp->b_error)==0)
-			return (EIO);
-	return (error);
+	if (bp->b_vp)
+		panic("bgetvp: not free");
+	VREF(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_blockh) {
+		bp->b_blockf = vp->v_blockh;
+		bp->b_blockb = &vp->v_blockh;
+		vp->v_blockh->b_blockb = &bp->b_blockf;
+		vp->v_blockh = bp;
+	} else {
+		vp->v_blockh = bp;
+		bp->b_blockb = &vp->v_blockh;
+		bp->b_blockf = NULL;
+	}
 }
 
 /*
- * 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
+ * Disassociate a buffer from a vnode.
  */
-binval(dev)
-	dev_t dev;
+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;
+	vrele(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 bufhd *hp;
-#define dp ((struct buf *)hp)
+	register struct vnode *newvp;
+{
+	register struct buf *bq;
 
-	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;
+	/*
+	 * 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;
+	}
+	/*
+	 * Insert onto list for new vnode.
+	 */
+	if (newvp->v_blockh) {
+		bp->b_blockf = newvp->v_blockh;
+		bp->b_blockb = &newvp->v_blockh;
+		newvp->v_blockh->b_blockb = &bp->b_blockf;
+		newvp->v_blockh = bp;
+	} else {
+		newvp->v_blockh = bp;
+		bp->b_blockb = &newvp->v_blockh;
+		bp->b_blockf = NULL;
+	}
 }