Import sources from 2011-03-30 iso image

15 files changed, 2570 insertions, 0 deletions

diff --git a/sys/src/cmd/cfs/bcache.c b/sys/src/cmd/cfs/bcache.c
new file mode 100755
index 000000000..becd675ea
--- /dev/null
+++ b/sys/src/cmd/cfs/bcache.c
@@ -0,0 +1,197 @@
+#include <u.h>
+#include <libc.h>
+#include "cformat.h"
+#include "lru.h"
+#include "bcache.h"
+
+int
+bcinit(Bcache *bc, int f, int bsize)
+{
+	Bbuf *b;
+
+	/*
+	 *  allocate space for all buffers
+	 *  point all buffers into outer space
+	 */
+	bc->dfirst = 0;
+	bc->bsize = bsize;
+	bc->f = f;
+	lruinit(bc);
+	for(b = bc->bb; b < &bc->bb[Nbcache]; b++){
+		b->inuse = 0;
+		b->next = 0;
+		b->dirty = 0;
+		if(b->data == 0)
+			b->data = (char *)malloc(bc->bsize);
+		if(b->data == 0)
+			return -1;
+		lruadd(bc, b);
+	}
+
+	return 0;
+}
+
+/*
+ *  Find a buffer for block b.  If it's dirty, write it out.
+ */
+Bbuf *
+bcfind(Bcache *bc, ulong bno)
+{
+	Bbuf *b;
+
+	if(bno == Notabno)
+		error("bcfind: Notabno");
+	bno &= ~Indbno;
+
+	/*
+	 *  if we already have a buffer for this bno, use it
+	 */
+	for(b = bc->bb; b < &bc->bb[Nbcache]; b++)
+		if(b->inuse && b->bno==bno)
+			goto out;
+
+	/*
+	 *  get least recently used block
+	 */
+	b = (Bbuf*)bc->lnext;
+out:
+	/*
+	 *  if dirty, write it out
+	 */
+	if(b->dirty)
+		if(bcwrite(bc, b) < 0)
+			warning("writing dirty page");
+	lruref(bc, b);
+	return b;
+}
+
+/*
+ *  allocate a buffer block for a block.  it's guaranteed to be there till
+ *  the next Nbcache bcread's.
+ */
+Bbuf *
+bcalloc(Bcache *bc, ulong bno)
+{
+	Bbuf *b;
+
+	b = bcfind(bc, bno);
+	bno &= ~Indbno;
+	b->bno = bno;
+	b->inuse = 1;
+	return b;
+}
+
+/*
+ *  read a block into a buffer cache.  it's guaranteed to be there till
+ *  the next Nbcache bcread's.
+ */
+Bbuf *
+bcread(Bcache *bc, ulong bno)
+{
+	Bbuf *b;
+
+	b = bcfind(bc, bno);
+	bno &= ~Indbno;
+	if(b->bno!=bno || !b->inuse)
+		/*
+		 *  read in the one we really want
+		 */
+		if(bread(bc, bno, b->data) < 0){
+			b->inuse = 0;
+			return 0;
+		}
+	b->bno = bno;
+	b->inuse = 1;
+	return b;
+}
+
+/*
+ *  mark a page dirty, if it's already dirty force a write
+ *
+ *	N.B: ordering is important.
+ */
+void
+bcmark(Bcache *bc, Bbuf *b)
+{
+	lruref(bc, b);
+
+	if(b->dirty){
+		bcwrite(bc, b);
+		return;
+	}
+
+	b->dirty = 1;
+	if(bc->dfirst)
+		bc->dlast->next = b;
+	else
+		bc->dfirst = b;
+	bc->dlast = b;
+}
+
+/*
+ *  write out a page (and all preceding dirty ones)
+ */
+int
+bcwrite(Bcache *bc, Bbuf *b)
+{
+	Bbuf *nb;
+
+	/*
+	 *  write out all preceding pages
+	 */
+	while(nb = bc->dfirst){
+		if(bwrite(bc, nb->bno, nb->data) < 0)
+			return -1;
+		nb->dirty = 0;
+		bc->dfirst = nb->next;
+		nb->next = 0;
+		if(nb == b)
+			return 0;
+	}
+
+	/*
+	 *  write out this page
+	 */
+	if(bwrite(bc, b->bno, b->data) < 0)
+		return -1;
+	b->dirty = 0;
+	b->next = 0;
+	return 0;
+}
+
+/*
+ *  write out all dirty pages (in order)
+ */
+int
+bcsync(Bcache *bc)
+{
+	if(bc->dfirst)
+		return bcwrite(bc, bc->dlast);
+	return 0;
+}
+
+/*
+ *  read a block from disk
+ */
+int
+bread(Bcache *bc, ulong bno, void *buf)
+{
+	uvlong x = (uvlong)bno * bc->bsize;
+
+	if(pread(bc->f, buf, bc->bsize, x) != bc->bsize)
+		return -1;
+	return 0;
+}
+
+/*
+ *  write a block to disk
+ */
+int
+bwrite(Bcache *bc, ulong bno, void *buf)
+{
+	uvlong x = (uvlong)bno * bc->bsize;
+
+	if(pwrite(bc->f, buf, bc->bsize, x) != bc->bsize)
+		return -1;
+	return 0;
+}
diff --git a/sys/src/cmd/cfs/bcache.h b/sys/src/cmd/cfs/bcache.h
new file mode 100755
index 000000000..0029ab524
--- /dev/null
+++ b/sys/src/cmd/cfs/bcache.h
@@ -0,0 +1,45 @@
+typedef struct Bbuf	Bbuf;
+typedef struct Bcache	Bcache;
+
+enum
+{
+	Nbcache=	32,		/* number of blocks kept in pool */
+};
+
+/*
+ *  block cache descriptor
+ */
+struct Bbuf
+{
+	Lru;				/* must be first in struct */
+	ulong	bno;
+	int	inuse;
+	Bbuf	*next;			/* next in dirty list */
+	int	dirty;
+	char	*data;
+};
+
+/*
+ *  the buffer cache
+ */
+struct Bcache
+{
+	Lru;
+	int	bsize;			/* block size in bytes */
+	int	f;			/* fd to disk */
+	Bbuf	*dfirst;		/* dirty list */
+	Bbuf	*dlast;
+	Bbuf	bb[Nbcache];
+};
+
+int	bcinit(Bcache*, int, int);
+Bbuf*	bcalloc(Bcache*, ulong);
+Bbuf*	bcread(Bcache*, ulong);
+void	bcmark(Bcache*, Bbuf*);
+int	bcwrite(Bcache*, Bbuf*);
+int	bcsync(Bcache*);
+int	bread(Bcache*, ulong, void*);
+int	bwrite(Bcache*, ulong, void*);
+int	bref(Bcache*, Bbuf*);
+void	error(char*, ...);
+void	warning(char*);
diff --git a/sys/src/cmd/cfs/cformat.h b/sys/src/cmd/cfs/cformat.h
new file mode 100755
index 000000000..b69e3ee8b
--- /dev/null
+++ b/sys/src/cmd/cfs/cformat.h
@@ -0,0 +1,75 @@
+/*
+ *  format of cache on disk
+ */
+typedef struct Dptr	Dptr;
+typedef struct Dahdr	Dahdr;
+typedef struct Dalloc	Dalloc;
+typedef struct Fphdr	Fphdr;
+typedef struct Fptr	Fptr;
+typedef struct Inode	Inode;
+typedef struct Dihdr	Dihdr;
+typedef struct Dinode	Dinode;
+
+enum
+{
+	Amagic= 	0xbebeefed,	/* allocation block magic */
+	Imagic=		0xbadc00ce,	/* inode block magic */
+	BtoUL=		8*sizeof(ulong),/* bits in a ulong */
+	CACHENAMELEN=	128
+};
+#define	Indbno		0x80000000	/* indirect block */
+#define	Notabno		0xFFFFFFFF	/* not a block number */
+
+/*
+ *  Allocation blocks at the begining of the disk.  There are
+ *  enough of these blocks to supply 1 bit for each block on the
+ *  disk;
+ */
+struct Dahdr
+{
+	ulong	magic;
+	ulong	bsize;		/* logical block size */
+	char	name[CACHENAMELEN];
+	short	nab;		/* number of allocation blocks */
+};
+struct Dalloc
+{
+	Dahdr;
+	ulong	bits[1];
+};
+
+/*
+ *  A pointer to disk data
+ */
+struct Dptr
+{
+	ulong	fbno;		/* file block number */
+	ulong	bno;		/* disk block number */
+	ushort	start;		/* offset into block of valid data */
+	ushort	end;		/* offset into block after valid data */
+};
+
+/*
+ *  A file descriptor.
+ */
+struct Inode
+{
+	Qid	qid;
+	vlong	length;
+	Dptr	ptr;		/* pointer page */	
+	char	inuse;
+};
+
+/*
+ *  inode blocks (after allocation blocks)
+ */
+struct Dihdr
+{
+	ulong	magic;
+	ulong	nino;		/* number of inodes */
+};
+struct Dinode
+{
+	Dihdr;
+	Inode	inode[1];
+};
diff --git a/sys/src/cmd/cfs/cfs.c b/sys/src/cmd/cfs/cfs.c
new file mode 100755
index 000000000..8bedeeb88
--- /dev/null
+++ b/sys/src/cmd/cfs/cfs.c
@@ -0,0 +1,921 @@
+#include <u.h>
+#include <libc.h>
+#include <auth.h>
+#include <fcall.h>
+
+#include "cformat.h"
+#include "lru.h"
+#include "bcache.h"
+#include "disk.h"
+#include "inode.h"
+#include "file.h"
+#include "stats.h"
+
+enum
+{
+	Nfid=		10240,
+};
+
+/* maximum length of a file */
+enum { MAXLEN = ~0ULL >> 1 };
+
+typedef struct Mfile Mfile;
+typedef struct Ram Ram;
+typedef struct P9fs P9fs;
+
+struct Mfile
+{
+	Qid	qid;
+	char	busy;
+};
+
+Mfile	mfile[Nfid];
+Icache	ic;
+int	debug, statson, noauth, openserver;
+
+struct P9fs
+{
+	int	fd[2];
+	Fcall	rhdr;
+	Fcall	thdr;
+	long	len;
+	char	*name;
+};
+
+P9fs	c;	/* client conversation */
+P9fs	s;	/* server conversation */
+
+struct Cfsstat  cfsstat, cfsprev;
+char	statbuf[2048];
+int	statlen;
+
+#define	MAXFDATA	8192	/* i/o size for read/write */
+
+int		messagesize = MAXFDATA+IOHDRSZ;
+
+uchar	datasnd[MAXFDATA + IOHDRSZ];
+uchar	datarcv[MAXFDATA + IOHDRSZ];
+
+Qid	rootqid;
+Qid	ctlqid = {0x5555555555555555LL, 0, 0};
+
+void	rversion(void);
+void	rauth(Mfile*);
+void	rflush(void);
+void	rattach(Mfile*);
+void	rwalk(Mfile*);
+void	ropen(Mfile*);
+void	rcreate(Mfile*);
+void	rread(Mfile*);
+void	rwrite(Mfile*);
+void	rclunk(Mfile*);
+void	rremove(Mfile*);
+void	rstat(Mfile*);
+void	rwstat(Mfile*);
+void	error(char*, ...);
+void	warning(char*);
+void	mountinit(char*, char*);
+void	io(void);
+void	sendreply(char*);
+void	sendmsg(P9fs*, Fcall*);
+void	rcvmsg(P9fs*, Fcall*);
+int	delegate(void);
+int	askserver(void);
+void	cachesetup(int, char*, char*);
+int	ctltest(Mfile*);
+void	genstats(void);
+
+char *mname[]={
+	[Tversion]		"Tversion",
+	[Tauth]	"Tauth",
+	[Tflush]	"Tflush",
+	[Tattach]	"Tattach",
+	[Twalk]		"Twalk",
+	[Topen]		"Topen",
+	[Tcreate]	"Tcreate",
+	[Tclunk]	"Tclunk",
+	[Tread]		"Tread",
+	[Twrite]	"Twrite",
+	[Tremove]	"Tremove",
+	[Tstat]		"Tstat",
+	[Twstat]	"Twstat",
+	[Rversion]	"Rversion",
+	[Rauth]	"Rauth",
+	[Rerror]	"Rerror",
+	[Rflush]	"Rflush",
+	[Rattach]	"Rattach",
+	[Rwalk]		"Rwalk",
+	[Ropen]		"Ropen",
+	[Rcreate]	"Rcreate",
+	[Rclunk]	"Rclunk",
+	[Rread]		"Rread",
+	[Rwrite]	"Rwrite",
+	[Rremove]	"Rremove",
+	[Rstat]		"Rstat",
+	[Rwstat]	"Rwstat",
+			0,
+};
+
+void
+usage(void)
+{
+	fprint(2, "usage:\tcfs -s [-dknrS] [-f partition]\n");
+	fprint(2, "\tcfs [-a netaddr | -F srv] [-dknrS] [-f partition] [mntpt]\n");
+	exits("usage");
+}
+
+void
+main(int argc, char *argv[])
+{
+	int std, format, chkid;
+	char *part, *server, *mtpt;
+	NetConnInfo *snci;
+
+	std = 0;
+	format = 0;
+	chkid = 1;
+	part = "/dev/sdC0/cache";
+	server = "tcp!fs";
+	mtpt = "/tmp";
+
+	ARGBEGIN{
+	case 'a':
+		server = EARGF(usage());
+		break;
+	case 'd':
+		debug = 1;
+		break;
+	case 'f':
+		part = EARGF(usage());
+		break;
+	case 'F':
+		server = EARGF(usage());
+		openserver = 1;
+		break;
+	case 'k':
+		chkid = 0;
+		break;
+	case 'n':
+		noauth = 1;
+		break;
+	case 'r':
+		format = 1;
+		break;
+	case 'S':
+		statson = 1;
+		break;
+	case 's':
+		std = 1;
+		break;
+	default:
+		usage();
+	}ARGEND
+	if(argc && *argv)
+		mtpt = *argv;
+
+	if(debug)
+		fmtinstall('F', fcallfmt);
+
+	c.name = "client";
+	s.name = "server";
+	if(std){
+		c.fd[0] = c.fd[1] = 1;
+		s.fd[0] = s.fd[1] = 0;
+	}else
+		mountinit(server, mtpt);
+
+	if(chkid){
+		if((snci = getnetconninfo(nil, s.fd[0])) == nil)
+			/* Failed to lookup information; format */
+			cachesetup(1, nil, part);
+		else
+			/* Do partition check */
+			cachesetup(0, snci->raddr, part);
+	}else
+		/* Obey -f w/o regard to cache vs. remote server */
+		cachesetup(format, nil, part);
+
+	switch(fork()){
+	case 0:
+		io();
+		exits("");
+	case -1:
+		error("fork");
+	default:
+		exits("");
+	}
+}
+
+void
+cachesetup(int format, char *name, char *partition)
+{
+	int f;
+	int secsize;
+	int inodes;
+	int blocksize;
+
+	secsize = 512;
+	inodes = 1024;
+	blocksize = 4*1024;
+
+	f = open(partition, ORDWR);
+	if(f < 0)
+		error("opening partition");
+
+	if(format || iinit(&ic, f, secsize, name) < 0){
+		/*
+		 * If we need to format and don't have a name, fall
+		 * back to our old behavior of using "bootes"
+		 */
+		name = (name == nil? "bootes": name);
+		if(iformat(&ic, f, inodes, name, blocksize, secsize) < 0)
+			error("formatting failed");
+	}
+}
+
+void
+mountinit(char *server, char *mountpoint)
+{
+	int err;
+	int p[2];
+
+	/*
+	 *  grab a channel and call up the file server
+	 */
+	if (openserver)
+		s.fd[0] = open(server, ORDWR);
+	else
+		s.fd[0] = dial(netmkaddr(server, 0, "9fs"), 0, 0, 0);
+	if(s.fd[0] < 0)
+		error("opening data: %r");
+	s.fd[1] = s.fd[0];
+
+	/*
+ 	 *  mount onto name space
+	 */
+	if(pipe(p) < 0)
+		error("pipe failed");
+	switch(fork()){
+	case 0:
+		break;
+	default:
+		if (noauth)
+			err = mount(p[1], -1, mountpoint, MREPL|MCREATE, "");
+		else
+			err = amount(p[1], mountpoint, MREPL|MCREATE, "");
+		if (err < 0)
+			error("mount failed: %r");
+		exits(0);
+	case -1:
+		error("fork failed\n");
+/*BUG: no wait!*/
+	}
+	c.fd[0] = c.fd[1] = p[0];
+}
+
+void
+io(void)
+{
+	int type;
+	Mfile *mf;
+    loop:
+	rcvmsg(&c, &c.thdr);
+
+	type = c.thdr.type;
+
+	if(statson){
+		cfsstat.cm[type].n++;
+		cfsstat.cm[type].s = nsec();
+	}
+	mf = &mfile[c.thdr.fid];
+	switch(type){
+	default:
+		error("type");
+		break;
+	case Tversion:
+		rversion();
+		break;
+	case Tauth:
+		mf = &mfile[c.thdr.afid];
+		rauth(mf);
+		break;
+	case Tflush:
+		rflush();
+		break;
+	case Tattach:
+		rattach(mf);
+		break;
+	case Twalk:
+		rwalk(mf);
+		break;
+	case Topen:
+		ropen(mf);
+		break;
+	case Tcreate:
+		rcreate(mf);
+		break;
+	case Tread:
+		rread(mf);
+		break;
+	case Twrite:
+		rwrite(mf);
+		break;
+	case Tclunk:
+		rclunk(mf);
+		break;
+	case Tremove:
+		rremove(mf);
+		break;
+	case Tstat:
+		rstat(mf);
+		break;
+	case Twstat:
+		rwstat(mf);
+		break;
+	}
+	if(statson){
+		cfsstat.cm[type].t += nsec() -cfsstat.cm[type].s;
+	}
+	goto loop;
+}
+
+void
+rversion(void)
+{
+	if(messagesize > c.thdr.msize)
+		messagesize = c.thdr.msize;
+	c.thdr.msize = messagesize;	/* set downstream size */
+	delegate();
+}
+
+void
+rauth(Mfile *mf)
+{
+	if(mf->busy)
+		error("auth to used channel");
+
+	if(delegate() == 0){
+		mf->qid = s.rhdr.aqid;
+		mf->busy = 1;
+	}
+}
+
+void
+rflush(void)		/* synchronous so easy */
+{
+	sendreply(0);
+}
+
+void
+rattach(Mfile *mf)
+{
+	if(delegate() == 0){
+		mf->qid = s.rhdr.qid;
+		mf->busy = 1;
+		if (statson == 1){
+			statson++;
+			rootqid = mf->qid;
+		}
+	}
+}
+
+void
+rwalk(Mfile *mf)
+{
+	Mfile *nmf;
+
+	nmf = nil;
+	if(statson
+	  && mf->qid.type == rootqid.type && mf->qid.path == rootqid.path
+	  && c.thdr.nwname == 1 && strcmp(c.thdr.wname[0], "cfsctl") == 0){
+		/* This is the ctl file */
+		nmf = &mfile[c.thdr.newfid];
+		if(c.thdr.newfid != c.thdr.fid && nmf->busy)
+			error("clone to used channel");
+		nmf = &mfile[c.thdr.newfid];
+		nmf->qid = ctlqid;
+		nmf->busy = 1;
+		c.rhdr.nwqid = 1;
+		c.rhdr.wqid[0] = ctlqid;
+		sendreply(0);
+		return;
+	}
+	if(c.thdr.newfid != c.thdr.fid){
+		if(c.thdr.newfid >= Nfid)
+			error("clone nfid out of range");
+		nmf = &mfile[c.thdr.newfid];
+		if(nmf->busy)
+			error("clone to used channel");
+		nmf = &mfile[c.thdr.newfid];
+		nmf->qid = mf->qid;
+		nmf->busy = 1;
+		mf = nmf; /* Walk mf */
+	}
+
+	if(delegate() < 0){	/* complete failure */
+		if(nmf)
+			nmf->busy = 0;
+		return;
+	}
+
+	if(s.rhdr.nwqid == c.thdr.nwname){	/* complete success */
+		if(s.rhdr.nwqid > 0)
+			mf->qid = s.rhdr.wqid[s.rhdr.nwqid-1];
+		return;
+	}
+
+	/* partial success; release fid */
+	if(nmf)
+		nmf->busy = 0;
+}
+
+void
+ropen(Mfile *mf)
+{
+	if(statson && ctltest(mf)){
+		/* Opening ctl file */
+		if(c.thdr.mode != OREAD){
+			sendreply("does not exist");
+			return;
+		}
+		c.rhdr.qid = ctlqid;
+		c.rhdr.iounit = 0;
+		sendreply(0);
+		genstats();
+		return;
+	}
+	if(delegate() == 0){
+		mf->qid = s.rhdr.qid;
+		if(c.thdr.mode & OTRUNC)
+			iget(&ic, mf->qid);
+	}
+}
+
+void
+rcreate(Mfile *mf)
+{
+	if(statson && ctltest(mf)){
+		sendreply("exists");
+		return;
+	}
+	if(delegate() == 0){
+		mf->qid = s.rhdr.qid;
+		mf->qid.vers++;
+	}
+}
+
+void
+rclunk(Mfile *mf)
+{
+	if(!mf->busy){
+		sendreply(0);
+		return;
+	}
+	mf->busy = 0;
+	delegate();
+}
+
+void
+rremove(Mfile *mf)
+{
+	if(statson && ctltest(mf)){
+		sendreply("not removed");
+		return;
+	}
+	mf->busy = 0;
+	delegate();
+}
+
+void
+rread(Mfile *mf)
+{
+	int cnt, done;
+	long n;
+	vlong off, first;
+	char *cp;
+	char data[MAXFDATA];
+	Ibuf *b;
+
+	off = c.thdr.offset;
+	first = off;
+	cnt = c.thdr.count;
+
+	if(statson && ctltest(mf)){
+		if(cnt > statlen-off)
+			c.rhdr.count = statlen-off;
+		else
+			c.rhdr.count = cnt;
+		if((int)c.rhdr.count < 0){
+			sendreply("eof");
+			return;
+		}
+		c.rhdr.data = statbuf + off;
+		sendreply(0);
+		return;
+	}
+	if(mf->qid.type & (QTDIR|QTAUTH)){
+		delegate();
+		if (statson) {
+			cfsstat.ndirread++;
+			if(c.rhdr.count > 0){
+				cfsstat.bytesread += c.rhdr.count;
+				cfsstat.bytesfromdirs += c.rhdr.count;
+			}
+		}
+		return;
+	}
+
+	b = iget(&ic, mf->qid);
+	if(b == 0){
+		DPRINT(2, "delegating read\n");
+		delegate();
+		if (statson){
+			cfsstat.ndelegateread++;
+			if(c.rhdr.count > 0){
+				cfsstat.bytesread += c.rhdr.count;
+				cfsstat.bytesfromserver += c.rhdr.count;
+			}
+		}
+		return;
+	}
+
+	cp = data;
+	done = 0;
+	while(cnt>0 && !done){
+		if(off >= b->inode.length){
+			DPRINT(2, "offset %lld greater than length %lld\n",
+				off, b->inode.length);
+			break;
+		}
+		n = fread(&ic, b, cp, off, cnt);
+		if(n <= 0){
+			n = -n;
+			if(n==0 || n>cnt)
+				n = cnt;
+			DPRINT(2,
+			 "fetch %ld bytes of data from server at offset %lld\n",
+				n, off);
+			s.thdr.type = c.thdr.type;
+			s.thdr.fid = c.thdr.fid;
+			s.thdr.tag = c.thdr.tag;
+			s.thdr.offset = off;
+			s.thdr.count = n;
+			if(statson)
+				cfsstat.ndelegateread++;
+			if(askserver() < 0){
+				sendreply(s.rhdr.ename);
+				return;
+			}
+			if(s.rhdr.count != n)
+				done = 1;
+			n = s.rhdr.count;
+			if(n == 0){
+				/* end of file */
+				if(b->inode.length > off){
+					DPRINT(2, "file %llud.%ld, length %lld\n",
+						b->inode.qid.path,
+						b->inode.qid.vers, off);
+					b->inode.length = off;
+				}
+				break;
+			}
+			memmove(cp, s.rhdr.data, n);
+			fwrite(&ic, b, cp, off, n);
+			if (statson){
+				cfsstat.bytestocache += n;
+				cfsstat.bytesfromserver += n;
+			}
+		}else{
+			DPRINT(2, "fetched %ld bytes from cache\n", n);
+			if(statson)
+				cfsstat.bytesfromcache += n;
+		}
+		cnt -= n;
+		off += n;
+		cp += n;
+	}
+	c.rhdr.data = data;
+	c.rhdr.count = off - first;
+	if(statson)
+		cfsstat.bytesread += c.rhdr.count;
+	sendreply(0);
+}
+
+void
+rwrite(Mfile *mf)
+{
+	Ibuf *b;
+	char buf[MAXFDATA];
+
+	if(statson && ctltest(mf)){
+		sendreply("read only");
+		return;
+	}
+	if(mf->qid.type & (QTDIR|QTAUTH)){
+		delegate();
+		if(statson && c.rhdr.count > 0)
+			cfsstat.byteswritten += c.rhdr.count;
+		return;
+	}
+
+	memmove(buf, c.thdr.data, c.thdr.count);
+	if(delegate() < 0)
+		return;
+
+	if(s.rhdr.count > 0)
+		cfsstat.byteswritten += s.rhdr.count;
+	/* don't modify our cache for append-only data; always read from server*/
+	if(mf->qid.type & QTAPPEND)
+		return;
+	b = iget(&ic, mf->qid);
+	if(b == 0)
+		return;
+	if (b->inode.length < c.thdr.offset + s.rhdr.count)
+		b->inode.length = c.thdr.offset + s.rhdr.count;
+	mf->qid.vers++;
+	if (s.rhdr.count != c.thdr.count)
+		syslog(0, "cfslog", "rhdr.count %ud, thdr.count %ud\n",
+			s.rhdr.count, c.thdr.count);
+	if(fwrite(&ic, b, buf, c.thdr.offset, s.rhdr.count) == s.rhdr.count){
+		iinc(&ic, b);
+		if(statson)
+			cfsstat.bytestocache += s.rhdr.count;
+	}
+}
+
+void
+rstat(Mfile *mf)
+{
+	Dir d;
+
+	if(statson && ctltest(mf)){
+		genstats();
+		d.qid = ctlqid;
+		d.mode = 0444;
+		d.length = statlen;	/* would be nice to do better */
+		d.name = "cfsctl";
+		d.uid = "none";
+		d.gid = "none";
+		d.muid = "none";
+		d.atime = time(nil);
+		d.mtime = d.atime;
+		c.rhdr.nstat = convD2M(&d, c.rhdr.stat,
+			sizeof c.rhdr - (c.rhdr.stat - (uchar*)&c.rhdr));
+		sendreply(0);
+		return;
+	}
+	if(delegate() == 0){
+		Ibuf *b;
+
+		convM2D(s.rhdr.stat, s.rhdr.nstat , &d, nil);
+		mf->qid = d.qid;
+		b = iget(&ic, mf->qid);
+		if(b)
+			b->inode.length = d.length;
+	}
+}
+
+void
+rwstat(Mfile *mf)
+{
+	Ibuf *b;
+
+	if(statson && ctltest(mf)){
+		sendreply("read only");
+		return;
+	}
+	delegate();
+	if(b = iget(&ic, mf->qid))
+		b->inode.length = MAXLEN;
+}
+
+void
+error(char *fmt, ...)
+{
+	va_list arg;
+	static char buf[2048];
+
+	va_start(arg, fmt);
+	vseprint(buf, buf+sizeof(buf), fmt, arg);
+	va_end(arg);
+	fprint(2, "%s: %s\n", argv0, buf);
+	exits("error");
+}
+
+void
+warning(char *s)
+{
+	fprint(2, "cfs: %s: %r\n", s);
+}
+
+/*
+ *  send a reply to the client
+ */
+void
+sendreply(char *err)
+{
+
+	if(err){
+		c.rhdr.type = Rerror;
+		c.rhdr.ename = err;
+	}else{
+		c.rhdr.type = c.thdr.type+1;
+		c.rhdr.fid = c.thdr.fid;
+	}
+	c.rhdr.tag = c.thdr.tag;
+	sendmsg(&c, &c.rhdr);
+}
+
+/*
+ *  send a request to the server, get the reply, and send that to
+ *  the client
+ */
+int
+delegate(void)
+{
+	int type;
+
+	type = c.thdr.type;
+	if(statson){
+		cfsstat.sm[type].n++;
+		cfsstat.sm[type].s = nsec();
+	}
+
+	sendmsg(&s, &c.thdr);
+	rcvmsg(&s, &s.rhdr);
+
+	if(statson)
+		cfsstat.sm[type].t += nsec() - cfsstat.sm[type].s;
+
+	sendmsg(&c, &s.rhdr);
+	return c.thdr.type+1 == s.rhdr.type ? 0 : -1;
+}
+
+/*
+ *  send a request to the server and get a reply
+ */
+int
+askserver(void)
+{
+	int type;
+
+	s.thdr.tag = c.thdr.tag;
+
+	type = s.thdr.type;
+	if(statson){
+		cfsstat.sm[type].n++;
+		cfsstat.sm[type].s = nsec();
+	}
+
+	sendmsg(&s, &s.thdr);
+	rcvmsg(&s, &s.rhdr);
+
+	if(statson)
+		cfsstat.sm[type].t += nsec() - cfsstat.sm[type].s;
+
+	return s.thdr.type+1 == s.rhdr.type ? 0 : -1;
+}
+
+/*
+ *  send/receive messages with logging
+ */
+void
+sendmsg(P9fs *p, Fcall *f)
+{
+	DPRINT(2, "->%s: %F\n", p->name, f);
+
+	p->len = convS2M(f, datasnd, messagesize);
+	if(p->len <= 0)
+		error("convS2M");
+	if(write(p->fd[1], datasnd, p->len)!=p->len)
+		error("sendmsg");
+}
+
+void
+dump(uchar *p, int len)
+{
+	fprint(2, "%d bytes", len);
+	while(len-- > 0)
+		fprint(2, " %.2ux", *p++);
+	fprint(2, "\n");
+}
+
+void
+rcvmsg(P9fs *p, Fcall *f)
+{
+	int olen, rlen;
+	char buf[128];
+
+	olen = p->len;
+	p->len = read9pmsg(p->fd[0], datarcv, sizeof(datarcv));
+	if(p->len <= 0){
+		snprint(buf, sizeof buf, "read9pmsg(%d)->%ld: %r",
+			p->fd[0], p->len);
+		error(buf);
+	}
+
+	if((rlen = convM2S(datarcv, p->len, f)) != p->len)
+		error("rcvmsg format error, expected length %d, got %d",
+			rlen, p->len);
+	if(f->fid >= Nfid){
+		fprint(2, "<-%s: %d %s on %d\n", p->name, f->type,
+			mname[f->type]? mname[f->type]: "mystery", f->fid);
+		dump((uchar*)datasnd, olen);
+		dump((uchar*)datarcv, p->len);
+		error("rcvmsg fid out of range");
+	}
+	DPRINT(2, "<-%s: %F\n", p->name, f);
+}
+
+int
+ctltest(Mfile *mf)
+{
+	return mf->busy && mf->qid.type == ctlqid.type &&
+		mf->qid.path == ctlqid.path;
+}
+
+void
+genstats(void)
+{
+	int i;
+	char *p;
+
+	p = statbuf;
+
+	p += snprint(p, sizeof statbuf+statbuf-p,
+		"        Client                          Server\n");
+	p += snprint(p, sizeof statbuf+statbuf-p,
+	    "   #calls     Δ  ms/call    Δ      #calls     Δ  ms/call    Δ\n");
+	for (i = 0; i < nelem(cfsstat.cm); i++)
+		if(cfsstat.cm[i].n || cfsstat.sm[i].n) {
+			p += snprint(p, sizeof statbuf+statbuf-p,
+				"%7lud %7lud ", cfsstat.cm[i].n,
+				cfsstat.cm[i].n - cfsprev.cm[i].n);
+			if (cfsstat.cm[i].n)
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"%7.3f ", 0.000001*cfsstat.cm[i].t/
+					cfsstat.cm[i].n);
+			else
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"        ");
+			if(cfsstat.cm[i].n - cfsprev.cm[i].n)
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"%7.3f ", 0.000001*
+					(cfsstat.cm[i].t - cfsprev.cm[i].t)/
+					(cfsstat.cm[i].n - cfsprev.cm[i].n));
+			else
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"        ");
+			p += snprint(p, sizeof statbuf+statbuf-p,
+				"%7lud %7lud ", cfsstat.sm[i].n,
+				cfsstat.sm[i].n - cfsprev.sm[i].n);
+			if (cfsstat.sm[i].n)
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"%7.3f ", 0.000001*cfsstat.sm[i].t/
+					cfsstat.sm[i].n);
+			else
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"        ");
+			if(cfsstat.sm[i].n - cfsprev.sm[i].n)
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"%7.3f ", 0.000001*
+					(cfsstat.sm[i].t - cfsprev.sm[i].t)/
+					(cfsstat.sm[i].n - cfsprev.sm[i].n));
+			else
+				p += snprint(p, sizeof statbuf+statbuf-p,
+					"        ");
+			p += snprint(p, sizeof statbuf+statbuf-p, "%s\n",
+				mname[i]);
+		}
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7lud %7lud ndirread\n",
+		cfsstat.ndirread, cfsstat.ndirread - cfsprev.ndirread);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7lud %7lud ndelegateread\n",
+		cfsstat.ndelegateread, cfsstat.ndelegateread -
+		cfsprev.ndelegateread);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7lud %7lud ninsert\n",
+		cfsstat.ninsert, cfsstat.ninsert - cfsprev.ninsert);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7lud %7lud ndelete\n",
+		cfsstat.ndelete, cfsstat.ndelete - cfsprev.ndelete);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7lud %7lud nupdate\n",
+		cfsstat.nupdate, cfsstat.nupdate - cfsprev.nupdate);
+
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud bytesread\n",
+		cfsstat.bytesread, cfsstat.bytesread - cfsprev.bytesread);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud byteswritten\n",
+		cfsstat.byteswritten, cfsstat.byteswritten -
+		cfsprev.byteswritten);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud bytesfromserver\n",
+		cfsstat.bytesfromserver, cfsstat.bytesfromserver -
+		cfsprev.bytesfromserver);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud bytesfromdirs\n",
+		cfsstat.bytesfromdirs, cfsstat.bytesfromdirs -
+		cfsprev.bytesfromdirs);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud bytesfromcache\n",
+		cfsstat.bytesfromcache, cfsstat.bytesfromcache -
+		cfsprev.bytesfromcache);
+	p += snprint(p, sizeof statbuf+statbuf-p, "%7llud %7llud bytestocache\n",
+		cfsstat.bytestocache, cfsstat.bytestocache -
+		cfsprev.bytestocache);
+	statlen = p - statbuf;
+	cfsprev = cfsstat;
+}
diff --git a/sys/src/cmd/cfs/disk.c b/sys/src/cmd/cfs/disk.c
new file mode 100755
index 000000000..2a39271eb
--- /dev/null
+++ b/sys/src/cmd/cfs/disk.c
@@ -0,0 +1,361 @@
+#include <u.h>
+#include <libc.h>
+#include "cformat.h"
+#include "lru.h"
+#include "bcache.h"
+#include "disk.h"
+
+int	icformat(Disk*, ulong);
+
+/*
+ *  read in the disk structures,  return -1 if the format
+ *  is inconsistent.
+ */
+int
+dinit(Disk *d, int f, int psize, char *expname)
+{
+	ulong	i;
+	uvlong	length;
+	char	buf[1024];
+	Bbuf	*b;
+	Dalloc	*ba;
+	Dir	*dir;
+
+	/*
+	 *  get disk size
+	 */
+	dir = dirfstat(f);
+	if(dir == nil){
+		perror("dinit: stat");
+		return -1;
+	}
+	length = dir->length;
+	free(dir);
+
+	/*
+	 *  read first physical block to get logical block size, # of inodes,
+	 *  and # of allocation blocks
+	 */
+	if(seek(f, 0, 0) < 0){
+		perror("dinit: seek");
+		return -1;
+	}
+	if(read(f, buf, sizeof(buf)) != sizeof(buf)){
+		perror("dinit: read");
+		return -1;
+	}
+	ba = (Dalloc*)buf;
+	if(ba->bsize <= 0){
+		fprint(2, "dinit: bsize 0x%lux<= 0\n", ba->bsize);
+		return -1;
+	}
+	if((ba->bsize % psize) != 0){
+		fprint(2, "dinit: logical bsize (%lud) not multiple of physical (%ud)\n",
+			ba->bsize, psize);
+		return -1;
+	}
+	d->bsize = ba->bsize;
+	d->nb = length/d->bsize;
+	d->b2b = (d->bsize - sizeof(Dahdr))*8;
+	d->nab = (d->nb+d->b2b-1)/d->b2b;
+	d->p2b = d->bsize/sizeof(Dptr);
+	strncpy(d->name, ba->name, sizeof d->name);
+
+	if (expname != nil && strncmp(d->name, expname, sizeof d->name) != 0) {
+		/* Mismatch with recorded name; fail here to force a format */
+		fprint(2, "cfs: name mismatch\n");
+		return -1;
+	}
+
+	/*
+	 *  check allocation blocks for consistency
+	 */
+	if(bcinit(d, f, d->bsize) < 0){
+		fprint(2, "dinit: couldn't init block cache\n");
+		return -1;
+	}
+	for(i = 0; i < d->nab; i++){
+		b = bcread(d, i);
+		if(b == 0){
+			perror("dinit: read");
+			return -1;
+		}
+		ba = (Dalloc*)b->data;
+		if(ba->magic != Amagic){
+			fprint(2, "dinit: bad magic in alloc block %uld\n", i);
+			return -1;
+		}
+		if(d->bsize != ba->bsize){
+			fprint(2, "dinit: bad bsize in alloc block %uld\n", i);
+			return -1;
+		}
+		if(d->nab != ba->nab){
+			fprint(2, "dinit: bad nab in alloc block %uld\n", i);
+			return -1;
+		}
+		if(strncmp(d->name, ba->name, sizeof(d->name))){
+			fprint(2, "dinit: bad name in alloc block %uld\n", i);
+			return -1;
+		}
+	}
+	return 0;
+}
+
+/*
+ *  format the disk as a cache
+ */
+int
+dformat(Disk *d, int f, char *name, ulong bsize, ulong psize)
+{
+	int	i;
+	uvlong	length;
+	Bbuf	*b;
+	Dalloc	*ba;
+	Dir	*dir;
+	Dptr	dptr;
+
+	fprint(2, "formatting disk\n");
+
+	/*
+	 *  calculate basic numbers
+	 */
+	dir = dirfstat(f);
+	if(dir == nil)
+		return -1;
+	length = dir->length;
+	d->bsize = bsize;
+	if((d->bsize % psize) != 0){
+		fprint(2, "cfs: logical bsize not multiple of physical\n");
+		return -1;
+	}
+	d->nb = length/d->bsize;
+	d->b2b = (d->bsize - sizeof(Dahdr))*8;
+	d->nab = (d->nb+d->b2b-1)/d->b2b;
+	d->p2b = d->bsize/sizeof(Dptr);
+
+	/*
+	 *  init allocation blocks
+	 */
+	if(bcinit(d, f, d->bsize) < 0)
+		return -1;
+	for(i = 0; i < d->nab; i++){
+		b = bcalloc(d, i);
+		if(b == 0){
+			perror("cfs: bcalloc");
+			return -1;
+		}
+		memset(b->data, 0, d->bsize);
+		ba = (Dalloc*)b->data;
+		ba->magic = Amagic;
+		ba->bsize = d->bsize;
+		ba->nab = d->nab;
+		strncpy(ba->name, name, sizeof(ba->name));
+		bcmark(d, b);
+	}
+
+	/*
+	 *  allocate allocation blocks
+	 */
+	for(i = 0; i < d->nab; i++)
+		if(dalloc(d, &dptr) == Notabno){
+			fprint(2, "can't allocate allocation blocks\n");
+			return -1;
+		}
+
+	return bcsync(d);
+}
+
+/*
+ *  allocate a block from a bit vector page
+ *
+ *  a return value of Notabno means no blocks left
+ */
+static ulong
+_balloc(Dalloc *ba, ulong max)
+{
+	int len;	/* number of valid words */
+	ulong i;	/* bit position in long */
+	ulong m;	/* 1<<i */
+	ulong v;	/* old value of long */
+	ulong *p, *e;
+
+	/*
+	 *  find a word with a 0 bit
+	 */
+	len = (max+BtoUL-1)/BtoUL;
+	for(p = ba->bits, e = p + len; p < e; p++)
+		if(*p != 0xFFFFFFFF)
+			break;
+	if(p == e)
+		return Notabno;
+
+	/*
+	 *  find the first 0 bit
+	 */
+	v = *p;
+	for(m = 1, i = 0; i < BtoUL; i++, m <<= 1)
+		if((m|v) != v)
+			break;
+
+	/*
+	 *  calculate block number
+	 */
+	i += (p - ba->bits)*BtoUL;
+	if(i >= max)
+		return Notabno;
+
+	/*
+	 *  set bit to 1
+	 */
+	*p = v | m;
+	return i;
+}
+
+/*
+ *  allocate a block
+ *
+ *  return Notabno if none left
+ */
+ulong
+dalloc(Disk *d, Dptr *p)
+{
+	ulong	bno, max, rv;
+	Bbuf	*b;
+	Dalloc	*ba;
+
+	max = d->nb;
+	for(bno = 0; bno < d->nab; bno++){
+		b = bcread(d, bno);
+		ba = (Dalloc*)b->data;
+		rv = _balloc(ba, max > d->b2b ? d->b2b : max);
+		if(rv != Notabno){
+			rv = bno*d->b2b + rv;
+			if(p){
+				p->start = p->end = 0;
+				p->bno = rv;
+			}
+			bcmark(d, b);
+			return rv;
+		}
+		max -= d->b2b;
+	}
+	if(p)
+		p->bno = Notabno;
+	return Notabno;
+}
+
+/*
+ *  allocate a block of pointers
+ */
+ulong
+dpalloc(Disk *d, Dptr *p)
+{
+	Bbuf *b;
+	Dptr *sp, *ep;
+
+	if(dalloc(d, p) == Notabno)
+		return Notabno;
+
+	/*
+	 *  allocate the page and invalidate all the
+	 *  pointers
+	 */
+	b = bcalloc(d, p->bno);
+	if(b == 0)
+		return -1;
+	sp = (Dptr*)b->data;
+	for(ep = sp + d->p2b; sp < ep; sp++){
+		sp->bno = Notabno;
+		sp->start = sp->end = 0;
+	}
+	p->bno |= Indbno;
+	p->start = 0;
+	p->end = d->bsize;
+
+	/*
+	 *  mark the page as dirty
+	 */
+	bcmark(d, b);
+	return 0;
+}
+
+/*
+ *  free a block
+ */
+int
+_bfree(Disk *d, ulong i)
+{
+	ulong bno, m;
+	ulong *p;
+	Bbuf *b;
+	Dalloc *ba;
+
+	/*
+	 *  get correct allocation block
+	 */
+	bno = i/d->b2b;
+	if(bno >= d->nab)
+		return -1;
+	b = bcread(d, bno);
+	if(b == 0)
+		return -1;
+	ba = (Dalloc*)b->data;
+
+	/*
+	 *  change bit
+	 */
+	i -= bno*d->b2b;
+	p = ba->bits + (i/BtoUL);
+	m = 1<<(i%BtoUL);
+	*p &= ~m;
+	bcmark(d, b);
+
+	return 0;
+}
+
+/*
+ *  free a block (or blocks)
+ */
+int
+dfree(Disk *d, Dptr *dp)
+{
+	ulong bno;
+	Dptr *sp, *ep;
+	Bbuf *b;
+
+	bno = dp->bno;
+	dp->bno = Notabno;
+
+	/*
+	 *  nothing to free
+	 */
+	if(bno == Notabno)
+		return 0;
+
+	/*
+	 *  direct pointer
+	 */
+	if((bno & Indbno) == 0)
+		return _bfree(d, bno);
+
+	/*
+	 *  first indirect page
+	 */
+	bno &= ~Indbno;
+	_bfree(d, bno);
+
+	/*
+	 *  then all the pages it points to
+	 *
+	 *	DANGER: this algorithm may fail if there are more
+	 *		allocation blocks than block buffers
+	 */
+	b = bcread(d, bno);
+	if(b == 0)
+		return -1;
+	sp = (Dptr*)b->data;
+	for(ep = sp + d->p2b; sp < ep; sp++)
+		if(dfree(d, sp) < 0)
+			return -1;
+	return 0;
+}
diff --git a/sys/src/cmd/cfs/disk.h b/sys/src/cmd/cfs/disk.h
new file mode 100755
index 000000000..a085a82ec
--- /dev/null
+++ b/sys/src/cmd/cfs/disk.h
@@ -0,0 +1,24 @@
+typedef struct Disk	Disk;
+
+/*
+ *  Reference to the disk
+ */
+struct Disk
+{
+	Bcache;
+	ulong	nb;	/* number of blocks */
+	ulong	nab;	/* number of allocation blocks */
+	int	b2b;	/* allocation bits to a block */
+	int	p2b;	/* Dptr's per page */
+	char	name[CACHENAMELEN];
+};
+
+int	dinit(Disk*, int, int, char*);
+int	dformat(Disk*, int, char*, ulong, ulong);
+ulong	dalloc(Disk*, Dptr*);
+ulong	dpalloc(Disk*, Dptr*);
+int	dfree(Disk*, Dptr*);
+
+extern int debug;
+
+#define DPRINT if(debug)fprint
diff --git a/sys/src/cmd/cfs/file.c b/sys/src/cmd/cfs/file.c
new file mode 100755
index 000000000..457a434e6
--- /dev/null
+++ b/sys/src/cmd/cfs/file.c
@@ -0,0 +1,298 @@
+#include <u.h>
+#include <libc.h>
+#include "cformat.h"
+#include "lru.h"
+#include "bcache.h"
+#include "disk.h"
+#include "inode.h"
+#include "file.h"
+
+/*
+ *  merge data with that which already exists in a block
+ *
+ *  we allow only one range per block, always use the new
+ *  data if the ranges don't overlap.
+ */
+void
+fmerge(Dptr *p, char *to, char *from, int start, int len)
+{
+	int end;
+
+	end = start + len;
+	memmove(to+start, from, end-start);
+
+	/*
+	 *  if ranges do not overlap...
+	 */
+	if(start>p->end || p->start>end){
+		/*
+		 *  just use the new data
+		 */
+		p->start = start;
+		p->end = end;
+	} else {
+		/*
+		 *  merge ranges
+		 */
+		if(start < p->start)
+			p->start = start;
+		if(end > p->end)
+			p->end = end;
+	}
+
+}
+
+/*
+ *  write a block (or less) of data onto a disk, follow it with any necessary
+ *  pointer writes.
+ *
+ *	N.B. ordering is everything
+ */
+int
+fbwrite(Icache *ic, Ibuf *b, char *a, ulong off, int len)
+{
+	int wrinode;
+	ulong fbno;
+	Bbuf *dbb;	/* data block */
+	Bbuf *ibb;	/* indirect block */
+	Dptr *p;
+	Dptr t;
+
+	fbno = off / ic->bsize;
+	p = &b->inode.ptr;
+	ibb = 0;
+	wrinode = 0;
+
+	/*
+	 *  are there any pages for this inode?
+	 */
+	if(p->bno == Notabno){
+		wrinode = 1;
+		goto dowrite;
+	}
+	
+	/*
+ 	 *  is it an indirect block?
+	 */
+	if(p->bno & Indbno){
+		ibb = bcread(ic, p->bno);
+		if(ibb == 0)
+			return -1;
+		p = (Dptr*)ibb->data;
+		p += fbno % ic->p2b;
+		goto dowrite;
+	}
+
+	/*
+ 	 *  is it the wrong direct block?
+	 */
+	if((p->fbno%ic->p2b) != (fbno%ic->p2b)){
+		/*
+		 *  yes, make an indirect block
+		 */
+		t = *p;
+		dpalloc(ic, p);
+		if(p->bno == Notabno){
+			*p = t;
+			return -1;
+		}
+		ibb = bcalloc(ic, p->bno);
+		if(ibb == 0){
+			*p = t;
+			return -1;
+		}
+		p = (Dptr*)ibb->data;
+		p += t.fbno % ic->p2b;
+		*p = t;
+		p = (Dptr*)ibb->data;
+		p += fbno % ic->p2b;
+	}
+	wrinode = 1;
+
+dowrite:
+	/*
+	 *  get the data block into the block cache
+	 */
+	if(p->bno == Notabno){
+		/*
+		 *  create a new block
+		 */
+		dalloc(ic, p);
+		if(p->bno == Notabno)
+			return -1;		/* no blocks left (maybe) */
+		dbb = bcalloc(ic, p->bno);
+	} else {
+		/*
+		 *  use what's there
+		 */
+		dbb = bcread(ic, p->bno);
+	}
+	if(dbb == 0)
+		return -1;
+
+	/*
+	 *  merge in the new data
+	 */
+	if(p->fbno != fbno){
+		p->start = p->end = 0;
+		p->fbno = fbno;
+	}
+	fmerge(p, dbb->data, a, off % ic->bsize, len);
+
+	/*
+	 *  write changed blocks back in the
+	 *  correct order
+	 */
+	bcmark(ic, dbb);
+	if(ibb)
+		bcmark(ic, ibb);
+	if(wrinode)
+		if(iwrite(ic, b) < 0)
+			return -1;
+	return len;
+}
+
+/*
+ *  write `n' bytes to the cache
+ *
+ *  return number of bytes written
+ */
+long
+fwrite(Icache *ic, Ibuf *b, char *a, ulong off, long n)
+{
+	int len;
+	long sofar;
+
+	for(sofar = 0; sofar < n; sofar += len){
+		len = ic->bsize - ((off+sofar)%ic->bsize);
+		if(len > n - sofar)
+			len = n - sofar;
+		if(fbwrite(ic, b, a+sofar, off+sofar, len) < 0)
+			return sofar;
+	}
+	return sofar;
+}
+
+/*
+ *  get a pointer to the next valid data at or after `off'
+ */
+Dptr *
+fpget(Icache *ic, Ibuf *b, ulong off)
+{
+	ulong fbno;
+	long doff;
+	Bbuf *ibb;	/* indirect block */
+	Dptr *p, *p0, *pf;
+
+	fbno = off / ic->bsize;
+	p = &b->inode.ptr;
+
+	/*
+	 *  are there any pages for this inode?
+	 */
+	if(p->bno == Notabno)
+		return 0;
+	
+	/*
+ 	 *  if it's a direct block, life is easy?
+	 */
+	if(!(p->bno & Indbno)){
+		/*
+		 *  a direct block, return p if it's at least past what we want
+		 */
+		if(p->fbno > fbno)
+			return p;
+		if(p->fbno < fbno)
+			return 0;
+		doff = off % ic->bsize;
+		if(doff>=p->start && doff<p->end)
+			return p;
+		else
+			return 0;
+	}
+
+	/*
+	 *  read the indirect block
+	 */
+	ibb = bcread(ic, p->bno);
+	if(ibb == 0)
+		return 0;
+
+	/*
+	 *  find the next valid pointer
+	 */
+	p0 = (Dptr*)ibb->data;
+	pf = p0 + (fbno % ic->p2b);
+	if(pf->bno!=Notabno && pf->fbno==fbno){
+		doff = off % ic->bsize;
+		if(doff<pf->end)
+			return pf;
+	}
+	for(p = pf+1; p < p0 + ic->p2b; p++){
+		fbno++;
+		if(p->fbno==fbno && p->bno!=Notabno && p->start<p->end)
+			return p;
+	}
+	for(p = p0; p < pf; p++){
+		fbno++;
+		if(p->fbno==fbno && p->bno!=Notabno && p->start<p->end)
+			return p;
+	}
+	return 0;
+}
+
+/*
+ *  read `n' bytes from the cache.
+ *
+ *  if we hit a gap and we've read something,
+ *  return number of bytes read so far.
+ *
+ *  if we start with a gap, return minus the number of bytes
+ *  to the next data.
+ *
+ *  if there are no bytes cached, return 0.
+ */
+long
+fread(Icache *ic, Ibuf *b, char *a, ulong off, long n)
+{
+	int len, start;
+	long sofar, gap;
+	Dptr *p;
+	Bbuf *bb;
+
+	for(sofar = 0; sofar < n; sofar += len, off += len){
+		/*
+		 *  get pointer to next data
+		 */
+		len = n - sofar;
+		p = fpget(ic, b, off);
+
+		/*
+		 *  if no more data, return what we have so far
+		 */
+		if(p == 0)
+			return sofar;
+
+		/*
+		 *  if there's a gap, return the size of the gap
+		 */
+		gap = (ic->bsize*p->fbno + p->start) - off;
+		if(gap>0)
+			if(sofar == 0)
+				return -gap;
+			else
+				return sofar;
+
+		/*
+		 *  return what we have
+		 */
+		bb = bcread(ic, p->bno);
+		if(bb == 0)
+			return sofar;
+		start = p->start - gap;
+		if(p->end - start < len)
+			len = p->end - start;
+		memmove(a + sofar, bb->data + start, len);
+	}
+	return sofar;
+}
diff --git a/sys/src/cmd/cfs/file.h b/sys/src/cmd/cfs/file.h
new file mode 100755
index 000000000..244bca930
--- /dev/null
+++ b/sys/src/cmd/cfs/file.h
@@ -0,0 +1,5 @@
+void	fmerge(Dptr*, char*, char*, int, int);
+int	fbwrite(Icache*, Ibuf*, char*, ulong, int);
+long	fwrite(Icache*, Ibuf*, char*, ulong, long);
+Dptr*	fpget(Icache*, Ibuf*, ulong);
+long	fread(Icache*, Ibuf*, char*, ulong, long);
diff --git a/sys/src/cmd/cfs/inode.c b/sys/src/cmd/cfs/inode.c
new file mode 100755
index 000000000..8b72a2742
--- /dev/null
+++ b/sys/src/cmd/cfs/inode.c
@@ -0,0 +1,408 @@
+#include <u.h>
+#include <libc.h>
+#include "cformat.h"
+#include "lru.h"
+#include "bcache.h"
+#include "disk.h"
+#include "inode.h"
+#include "stats.h"
+
+/*
+ *  read the inode blocks and make sure they
+ *  haven't been trashed.
+ *
+ *  make the in-core table of qid to inode mappings.
+ *	N.B. this is just an array. we need a linear search to find
+ *	     a particular inode. this could be done faster.
+ *
+ *  nab is the first inode block.
+ */
+int
+iinit(Icache *ic, int f, int psize, char* name)
+{
+	Ibuf *b;
+	Imap *m;
+	ulong ino;
+	Bbuf *bb;
+	Dinode *bi;
+
+	/*
+	 *  get basic sizes and allocation info from disk
+	 */
+	if(dinit(ic, f, psize, name) < 0)
+		return -1;
+
+	/*
+	 *  read first inode block to get number of inodes
+	 */
+	bb = bcread(ic, ic->nab);
+	if(bb == 0){
+		fprint(2, "iinit: can't read disk\n");
+		return -1;
+	}
+	bi = (Dinode*)bb->data;
+	if(bi->nino==0 || bi->nino>2048){
+		fprint(2, "iinit: bad nino\n");
+		return -1;
+	}
+	ic->nino = bi->nino;
+
+	/*
+	 *  set up sizing constants
+	 */
+	ic->i2b = (ic->bsize - sizeof(Dihdr))/sizeof(Inode);
+	ic->nib = (ic->nino + ic->i2b - 1)/ic->i2b;
+
+	/*
+	 *  allocate the in-core qid/inode map, build it's lru
+	 */
+	if(ic->map)
+		free(ic->map);
+	ic->map = malloc(sizeof(Imap)*ic->nino);
+	if(ic->map == 0){
+		fprint(2, "iinit: can't alloc map\n");
+		return -1;
+	}
+	lruinit(&ic->mlru);
+	for(m = ic->map; m < &ic->map[ic->nino]; m++){
+		m->inuse = 0;
+		m->b = 0;
+		lruadd(&ic->mlru, m);
+	}
+
+	/*
+	 *  mark all cache buffers as empty, put them on the lru list
+	 */
+	lruinit(&ic->blru);
+	for(b = ic->ib; b < &ic->ib[Nicache]; b++){
+		b->inuse = 0;
+		lruadd(&ic->blru, b);
+	}
+
+	/*
+	 *  Read all inodes and
+	 *  build the in-core qid/inode map
+	 */
+	for(ino = 0; ino < ic->nino; ino++){
+		b = iread(ic, ino);
+		if(b == 0){
+			fprint(2, "iinit: can't read inode %ld\n", ino);
+			return -1;
+		}
+		if(b->inode.inuse){
+			m = &ic->map[ino];
+			m->inuse = 1;
+			m->qid = b->inode.qid;
+			lruref(&ic->mlru, m);
+		}
+	}
+	return 0;
+}
+
+/*
+ *  format the inode blocks
+ */
+int
+iformat(Icache *ic, int f, ulong nino, char *name, int bsize, int psize)
+{
+	int nib;
+	ulong bno, i2b, i;
+	Bbuf *bb;
+	Dinode *bi;
+
+	/*
+	 *  first format disk allocation
+	 */
+	if(dformat(ic, f, name, bsize, psize) < 0)
+		return -1;
+
+	fprint(2, "formatting inodes\n");
+
+	i2b = (bsize - sizeof(Dihdr))/sizeof(Inode);
+	nib = (nino + i2b - 1)/i2b;
+
+	for(bno = ic->nab; bno < ic->nab + nib; bno++){
+		if(dalloc(ic, 0) == Notabno){
+			fprint(2, "iformat: balloc failed\n");
+			return -1;
+		}
+		bb = bcalloc(ic, bno);
+		if(bb == 0){
+			fprint(2, "iformat: bcalloc failed\n");
+			return -1;
+		}
+		bi = (Dinode*)bb->data;
+		bi->magic = Imagic;
+		bi->nino = nino;
+		for(i = 0; i < i2b; i++)
+			bi->inode[i].inuse = 0;
+		bcmark(ic, bb);
+	}
+
+	bcsync(ic);
+
+	return iinit(ic, f, psize, name);
+}
+
+/*
+ *  allocate a cache buffer, use least recently used
+ */
+Ibuf*
+ialloc(Icache *ic, ulong ino)
+{
+	Imap *m;
+	Ibuf *b;
+
+	b = (Ibuf*)ic->blru.lnext;
+	if(b->inuse)
+		ic->map[b->ino].b = 0;
+	b->ino = ino;
+	b->inuse = 1;
+	m = &ic->map[ino];
+	m->b = b;
+	return b;
+}
+
+/*
+ *  free a cache buffer
+ */
+void
+ifree(Icache *ic, Ibuf *b)
+{
+	b->inuse = 0;
+	if(b->inuse)
+		ic->map[b->ino].b = 0;
+	lruderef(&ic->blru, b);
+}
+
+/*
+ *  get an inode into the cache.  if no inode exists for this qid, create one
+ *  from an unused qid/inode map.
+ */
+Ibuf *
+iget(Icache *ic, Qid qid)
+{
+	Imap *m, *me;
+	Ibuf *b;
+
+	/*
+	 *  find map entry with same qid.path
+	 */
+	for(m = ic->map, me = &ic->map[ic->nino]; m < me; m++)
+		if(m->inuse && m->qid.path==qid.path){
+			if(m->qid.vers != qid.vers){
+				/*
+				 *  our info is old, forget it
+				 */
+				DPRINT(2, "updating old file %llud.%lud\n",
+					qid.path, qid.vers);
+				m->qid = qid;
+				iupdate(ic, m - ic->map, qid);
+			}
+			break;
+		}
+
+	/*
+ 	 *  if an already existing inode, just get it
+	 */
+	if(m != me)
+		return iread(ic, m - ic->map);
+
+	/*
+	 *  create a new inode, throw out the least recently used inode
+	 *  if necessary
+	 */
+	m = (Imap*)ic->mlru.lnext;
+	if(m->inuse){
+		DPRINT(2, "superceding file %llud.%ld by %llud.%ld\n",
+			m->qid.path, m->qid.vers, qid.path, qid.vers);
+		if(iremove(ic, m - ic->map) < 0)
+			return 0;
+	}
+
+	if(statson)
+		cfsstat.ninsert++;
+	/*
+	 *  init inode and write to disk
+	 */
+	DPRINT(2, "new file %llud.%ld ino %ld\n",
+		qid.path, qid.vers, m - ic->map);
+	b = ialloc(ic, m - ic->map);
+	b->inode.inuse = m->inuse = 1;
+	b->inode.qid = qid;
+	b->inode.length = 0x7fffffffffffffffLL;
+	m->qid = qid;
+	b->inode.ptr.bno = Notabno;
+	iwrite(ic, b);
+	return b;
+}
+
+/*
+ *  read an inode into the cache
+ *
+ *  ASSUMPTION: the inode is valid
+ */
+Ibuf*
+iread(Icache *ic, ulong ino)
+{
+	Ibuf *b;
+	Imap *m;
+	ulong bno;
+	Bbuf *bb;
+	Dinode *bi;
+
+	/*
+	 *  first see if we already have it in a cache entry
+	 */
+	m = &ic->map[ino];
+	if(m->inuse && m->b){
+		b = m->b;
+		goto out;
+	}
+
+	/*
+	 *  read it
+	 */
+	b = ialloc(ic, ino);
+	bno = ic->nab + ino/ic->i2b;
+	bb = bcread(ic, bno);
+	if(bb == 0){
+		ifree(ic, b);
+		return 0;
+	}
+	bi = (Dinode*)bb->data;
+	b->inode = bi->inode[ino % ic->i2b];
+
+	/*
+	 *  consistency check
+	 */
+	if(bi->nino!=ic->nino || bi->magic!=Imagic){
+		fprint(2, "iread: inconsistent inode block\n");
+		ifree(ic, b);
+		return 0;
+	}
+out:
+	b->inuse = 1;
+	m->b = b;
+	if(b->inode.inuse)
+		lruref(&ic->mlru, m);
+	lruref(&ic->blru, b);
+	return b;
+}
+
+/*
+ *  write an inode back to disk
+ */
+int
+iwrite(Icache *ic, Ibuf *b)
+{
+	ulong bno;
+	Bbuf *bb;
+	Dinode *bi;
+
+	bno = ic->nab + b->ino/ic->i2b;
+	bb = bcread(ic, bno);
+	if(bb == 0)
+		return 0;
+	bi = (Dinode*)bb->data;
+	bi->inode[b->ino % ic->i2b] = b->inode;
+	bcmark(ic, bb);
+	lruref(&ic->mlru, &ic->map[b->ino]);
+	lruref(&ic->blru, b);
+	return 0;
+}
+
+/*
+ *  Forget what we know about an inode without removing it
+ *
+ *	N.B: ordering of iwrite and dfree is important
+ */
+int
+iupdate(Icache *ic, ulong ino, Qid qid)
+{
+	Ibuf *b;
+	Imap *m;
+	Dptr d;
+
+	if(statson)
+		cfsstat.nupdate++;
+	b = iread(ic, ino);
+	if(b == 0)
+		return -1;
+
+	/*
+	 *  update inode and map
+	 */
+	b->inode.qid = qid;
+	b->inode.length = 0x7fffffffffffffffLL;	/* Set to maximum */
+	m = &ic->map[ino];
+	m->qid = qid;
+
+	/*
+	 *  the free is not done if the write fails!
+	 *  this is important
+	 */
+	d = b->inode.ptr;
+	b->inode.ptr.bno = Notabno;
+	if(iwrite(ic, b) < 0)
+		return -1;
+	dfree(ic, &d);
+	return 0;
+}
+
+/*
+ *  remove an inode
+ *
+ *	N.B: ordering of iwrite and dfree is important
+ */
+int
+iremove(Icache *ic, ulong ino)
+{
+	Ibuf *b;
+	Imap *m;
+
+	if(statson)
+		cfsstat.ndelete++;
+	m = &ic->map[ino];
+
+	/*
+	 *  read in inode
+	 */
+	b = iread(ic, ino);
+	if(b == 0)
+		return -1;
+
+	/*
+	 *  mark it unused on disk
+	 */
+	b->inode.inuse = 0;
+	if(iwrite(ic, b) < 0)
+		return -1;
+
+	/*
+	 *  throw out it's data pages
+	 */
+	dfree(ic, &b->inode.ptr);
+
+	/*
+	 *  free the inode buffer
+	 */
+	ifree(ic, b);
+
+	/*
+	 *  make map entry least recently used
+	 */
+	lruderef(&ic->mlru, m);
+	return 0;
+}
+
+/*
+ *  increment our version number
+ */
+void
+iinc(Icache *ic, Ibuf *b)
+{
+	b->inode.qid.vers++;
+	ic->map[b->ino].qid = b->inode.qid;
+	iwrite(ic, b);
+}
diff --git a/sys/src/cmd/cfs/inode.h b/sys/src/cmd/cfs/inode.h
new file mode 100755
index 000000000..22fa16047
--- /dev/null
+++ b/sys/src/cmd/cfs/inode.h
@@ -0,0 +1,60 @@
+typedef struct Ibuf	Ibuf;
+typedef struct Imap	Imap;
+typedef struct Icache	Icache;
+
+enum
+{
+	Nicache=	64,		/* number of inodes kept in pool */
+};
+
+/*
+ *  a cached inode buffer
+ */
+struct Ibuf
+{
+	Lru;			/* must be first in structure */
+	int	inuse;		/* non-0 if in use */
+	ulong	ino;		/* index into inode table */
+	Inode	inode;		/* the inode contents */
+};
+
+/*
+ *  in-core qid to inode mapping
+ */
+struct Imap
+{
+	Lru;			/* must be first in structure */
+	Qid	qid;
+	Ibuf	*b;		/* cache buffer */
+	int	inuse;		/* non-0 if in use */
+};
+
+/*
+ *  the inode cache
+ */
+struct Icache
+{
+	Disk;
+
+	int	nino;		/* number of inodes */
+	ulong	ib0;		/* first inode block */
+	int	nib;		/* number of inode blocks */
+	int	i2b;		/* inodes to a block */
+
+	Ibuf	ib[Nicache];	/* inode buffers */
+	Lru	blru;
+
+	Imap	*map;		/* inode to qid mapping */
+	Lru	mlru;
+};
+
+Ibuf*	ialloc(Icache*, ulong);
+Ibuf*	iget(Icache*, Qid);
+Ibuf*	iread(Icache*, ulong);
+int	iformat(Icache*, int, ulong, char*, int, int);
+int	iinit(Icache*, int, int, char*);
+int	iremove(Icache*, ulong);
+int	iupdate(Icache*, ulong, Qid);
+int	iwrite(Icache*, Ibuf*);
+void	ifree(Icache*, Ibuf*);
+void	iinc(Icache*, Ibuf*);
diff --git a/sys/src/cmd/cfs/lru.c b/sys/src/cmd/cfs/lru.c
new file mode 100755
index 000000000..55f6c6e8b
--- /dev/null
+++ b/sys/src/cmd/cfs/lru.c
@@ -0,0 +1,75 @@
+/*
+ *  lru lists are circular with a list head
+ *  pointing to the start and end of the list
+ */
+#include <u.h>
+#include "lru.h"
+
+/*
+ *  Create an lru chain of buffers
+ */
+void
+lruinit(Lru *h)
+{
+	h->lprev = h->lnext = h;
+}
+
+/*
+ *  Add a member to an lru chain
+ */
+void
+lruadd(Lru *h, Lru *m)
+{
+	h->lprev->lnext = m;
+	m->lprev = h->lprev;
+	h->lprev = m;
+	m->lnext = h;
+}
+
+/*
+ *  Move to end of lru list
+ */
+void
+lruref(Lru *h, Lru *m)
+{
+	if(h->lprev == m)
+		return;		/* alread at end of list */
+
+	/*
+	 *  remove from list
+	 */
+	m->lprev->lnext = m->lnext;
+	m->lnext->lprev = m->lprev;
+
+	/*
+	 *  add in at end
+	 */
+	h->lprev->lnext = m;
+	m->lprev = h->lprev;
+	h->lprev = m;
+	m->lnext = h;
+}
+
+/*
+ *  Move to head of lru list
+ */
+void
+lruderef(Lru *h, Lru *m)
+{
+	if(h->lnext == m)
+		return;		/* alread at head of list */
+
+	/*
+	 *  remove from list
+	 */
+	m->lprev->lnext = m->lnext;
+	m->lnext->lprev = m->lprev;
+
+	/*
+	 *  add in at head
+	 */
+	h->lnext->lprev = m;
+	m->lnext = h->lnext;
+	h->lnext = m;
+	m->lprev = h;
+}
diff --git a/sys/src/cmd/cfs/lru.h b/sys/src/cmd/cfs/lru.h
new file mode 100755
index 000000000..58dbe6d28
--- /dev/null
+++ b/sys/src/cmd/cfs/lru.h
@@ -0,0 +1,13 @@
+typedef	struct Lruhead	Lruhead;
+typedef struct Lru	Lru;
+
+struct Lru
+{
+	Lru	*lprev;
+	Lru	*lnext;
+};
+
+void	lruinit(Lru*);
+void	lruadd(Lru*, Lru*);
+void	lruref(Lru*, Lru*);
+void	lruderef(Lru*, Lru*);
diff --git a/sys/src/cmd/cfs/mkfile b/sys/src/cmd/cfs/mkfile
new file mode 100755
index 000000000..35375d0dd
--- /dev/null
+++ b/sys/src/cmd/cfs/mkfile
@@ -0,0 +1,41 @@
+</$objtype/mkfile
+
+TARG=cfs
+COMMONOBJ=\
+	disk.$O\
+	bcache.$O\
+	inode.$O\
+	lru.$O\
+	file.$O\
+
+OFILES=cfs.$O $COMMONOBJ
+
+HFILES=\
+	bcache.h\
+	cformat.h\
+	disk.h\
+	file.h\
+	inode.h\
+	lru.h\
+	stats.h\
+
+CFILES=${OFILES:%.$O=%.c}
+
+UPDATE=mkfile\
+	/386/bin/$TARG\
+	$HFILES\
+	$CFILES\
+
+BIN=/$objtype/bin
+</sys/src/cmd/mkone
+
+cfs.$O:		cformat.h inode.h bcache.h disk.h lru.h file.h
+file.$O:	cformat.h inode.h bcache.h disk.h lru.h file.h
+disk.$O:	cformat.h inode.h bcache.h disk.h lru.h
+bcache.$O:	cformat.h bcache.h lru.h
+main.$O:	cformat.h bcache.h disk.h lru.h
+inode.$O:	cformat.h inode.h bcache.h disk.h lru.h
+lru.$O:		lru.h
+
+clean:V:
+	rm -f *.[$OS] [$OS].* cfs cfsck
diff --git a/sys/src/cmd/cfs/notes b/sys/src/cmd/cfs/notes
new file mode 100755
index 000000000..f7618fb41
--- /dev/null
+++ b/sys/src/cmd/cfs/notes
@@ -0,0 +1,21 @@
+- virtual directories, reading a directory wipes out the
+ virtual directory. virtual property in the inode.
+
+- inode contains the qid, inuse, pointers, and virtual
+ property.  one pointer each for 3 levels.
+
+? reserve version == -1 to mean a virtual file? makes
+ some things easier.
+
+- first inode is always the root.
+
+- adding any virtual entry to a directory makes the
+ whole directory virtual.
+
+? how do we keep cfs from swapping?
+
+? how do we resolve access to missing parts when running
+ remotely?
+
+? what are default properties of files (ones we haven't
+ stat'd)?
diff --git a/sys/src/cmd/cfs/stats.h b/sys/src/cmd/cfs/stats.h
new file mode 100755
index 000000000..4f28a0b9e
--- /dev/null
+++ b/sys/src/cmd/cfs/stats.h
@@ -0,0 +1,26 @@
+struct Cfsmsg {
+	ulong	n;			/* number of messages (of some type) */
+	vlong	t;			/* time spent in these messages */
+	vlong	s;			/* start time of last call */
+};
+
+struct Cfsstat {
+	struct Cfsmsg cm[128];		/* client messages */
+	struct Cfsmsg sm[128];		/* server messages */
+
+	ulong ndirread;			/* # of directory read ops */
+	ulong ndelegateread;		/* # of read ops delegated */
+	ulong ninsert;			/* # of cache insert ops */
+	ulong ndelete;			/* # of cache delete ops */
+	ulong nupdate;			/* # of cache update ops */
+
+	uvlong bytesread;		/* # of bytes read by client */
+	uvlong byteswritten;		/* # of bytes written by client */
+	uvlong bytesfromserver;		/* # of bytes read from server */
+	uvlong bytesfromdirs;		/* # of directory bytes read from server */
+	uvlong bytesfromcache;		/* # of bytes read from cache */
+	uvlong bytestocache;		/* # of bytes written to cache */
+};
+
+extern struct Cfsstat cfsstat, cfsprev;
+extern int statson;