[PATCH] v9fs: zero copy implementation

Performance enhancement reducing the number of copies in the data and
stat paths.

Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@ericvh.myip.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

18 files changed, 1083 insertions, 1047 deletions

diff --git a/fs/9p/9p.c b/fs/9p/9p.c
index a3a1ac610723..dc3ce44ec836 100644
--- a/fs/9p/9p.c
+++ b/fs/9p/9p.c
@@ -1,8 +1,9 @@
 /*
  *  linux/fs/9p/9p.c
  *
- *  This file contains functions 9P2000 functions
+ *  This file contains functions to perform synchronous 9P calls
  *
+ *  Copyright (C) 2004 by Latchesar Ionkov <lucho@ionkov.net>
  *  Copyright (C) 2004 by Eric Van Hensbergen <ericvh@gmail.com>
  *  Copyright (C) 2002 by Ron Minnich <rminnich@lanl.gov>
  *
@@ -33,6 +34,7 @@
 #include "debug.h"
 #include "v9fs.h"
 #include "9p.h"
+#include "conv.h"
 #include "mux.h"
 
 /**
@@ -46,17 +48,21 @@
 
 int
 v9fs_t_version(struct v9fs_session_info *v9ses, u32 msize,
-	       char *version, struct v9fs_fcall **fcall)
+	       char *version, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
 	dprintk(DEBUG_9P, "msize: %d version: %s\n", msize, version);
-	msg.id = TVERSION;
-	msg.tag = ~0;
-	msg.params.tversion.msize = msize;
-	msg.params.tversion.version = version;
+	tc = v9fs_create_tversion(msize, version);
 
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -72,19 +78,23 @@ v9fs_t_version(struct v9fs_session_info *v9ses, u32 msize,
 
 int
 v9fs_t_attach(struct v9fs_session_info *v9ses, char *uname, char *aname,
-	      u32 fid, u32 afid, struct v9fs_fcall **fcall)
+	      u32 fid, u32 afid, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall* tc;
 
 	dprintk(DEBUG_9P, "uname '%s' aname '%s' fid %d afid %d\n", uname,
 		aname, fid, afid);
-	msg.id = TATTACH;
-	msg.params.tattach.fid = fid;
-	msg.params.tattach.afid = afid;
-	msg.params.tattach.uname = uname;
-	msg.params.tattach.aname = aname;
 
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	ret = -ENOMEM;
+	tc = v9fs_create_tattach(fid, afid, uname, aname);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 static void v9fs_t_clunk_cb(void *a, struct v9fs_fcall *tc,
@@ -117,24 +127,28 @@ static void v9fs_t_clunk_cb(void *a, struct v9fs_fcall *tc,
  * @fcall: pointer to response fcall pointer
  *
  */
+
 int
 v9fs_t_clunk(struct v9fs_session_info *v9ses, u32 fid)
 {
-	int err;
+	int ret;
 	struct v9fs_fcall *tc, *rc;
 
-	tc = kmalloc(sizeof(struct v9fs_fcall), GFP_KERNEL);
-
 	dprintk(DEBUG_9P, "fid %d\n", fid);
-	tc->id = TCLUNK;
-	tc->params.tclunk.fid = fid;
 
-	err = v9fs_mux_rpc(v9ses->mux, tc, &rc);
-	if (err >= 0) {
-		v9fs_t_clunk_cb(v9ses, tc, rc, 0);
-	}
+	ret = -ENOMEM;
+	rc = NULL;
+	tc = v9fs_create_tclunk(fid);
+	if (!IS_ERR(tc))
+		ret = v9fs_mux_rpc(v9ses->mux, tc, &rc);
+	else
+		ret = PTR_ERR(tc);
+
+	if (ret)
+		dprintk(DEBUG_ERROR, "failed fid %d err %d\n", fid, ret);
 
-	return err;
+	v9fs_t_clunk_cb(v9ses, tc, rc, ret);
+	return ret;
 }
 
 /**
@@ -144,14 +158,22 @@ v9fs_t_clunk(struct v9fs_session_info *v9ses, u32 fid)
  *
  */
 
-int v9fs_t_flush(struct v9fs_session_info *v9ses, u16 tag)
+int v9fs_t_flush(struct v9fs_session_info *v9ses, u16 oldtag)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
-	dprintk(DEBUG_9P, "oldtag %d\n", tag);
-	msg.id = TFLUSH;
-	msg.params.tflush.oldtag = tag;
-	return v9fs_mux_rpc(v9ses->mux, &msg, NULL);
+	dprintk(DEBUG_9P, "oldtag %d\n", oldtag);
+
+	ret = -ENOMEM;
+	tc = v9fs_create_tflush(oldtag);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, NULL);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -163,17 +185,22 @@ int v9fs_t_flush(struct v9fs_session_info *v9ses, u16 tag)
  */
 
 int
-v9fs_t_stat(struct v9fs_session_info *v9ses, u32 fid, struct v9fs_fcall **fcall)
+v9fs_t_stat(struct v9fs_session_info *v9ses, u32 fid, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
 	dprintk(DEBUG_9P, "fid %d\n", fid);
-	if (fcall)
-		*fcall = NULL;
 
-	msg.id = TSTAT;
-	msg.params.tstat.fid = fid;
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	ret = -ENOMEM;
+	tc = v9fs_create_tstat(fid);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -187,16 +214,22 @@ v9fs_t_stat(struct v9fs_session_info *v9ses, u32 fid, struct v9fs_fcall **fcall)
 
 int
 v9fs_t_wstat(struct v9fs_session_info *v9ses, u32 fid,
-	     struct v9fs_stat *stat, struct v9fs_fcall **fcall)
+	     struct v9fs_wstat *wstat, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
-	dprintk(DEBUG_9P, "fid %d length %d\n", fid, (int)stat->length);
-	msg.id = TWSTAT;
-	msg.params.twstat.fid = fid;
-	msg.params.twstat.stat = stat;
+	dprintk(DEBUG_9P, "fid %d\n", fid);
+
+	ret = -ENOMEM;
+	tc = v9fs_create_twstat(fid, wstat, v9ses->extended);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
 
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	return ret;
 }
 
 /**
@@ -213,23 +246,28 @@ v9fs_t_wstat(struct v9fs_session_info *v9ses, u32 fid,
 
 int
 v9fs_t_walk(struct v9fs_session_info *v9ses, u32 fid, u32 newfid,
-	    char *name, struct v9fs_fcall **fcall)
+	    char *name, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
+	int nwname;
 
 	dprintk(DEBUG_9P, "fid %d newfid %d wname '%s'\n", fid, newfid, name);
-	msg.id = TWALK;
-	msg.params.twalk.fid = fid;
-	msg.params.twalk.newfid = newfid;
-
-	if (name) {
-		msg.params.twalk.nwname = 1;
-		msg.params.twalk.wnames = &name;
-	} else {
-		msg.params.twalk.nwname = 0;
-	}
-
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+
+	if (name)
+		nwname = 1;
+	else
+		nwname = 0;
+
+	ret = -ENOMEM;
+	tc = v9fs_create_twalk(fid, newfid, nwname, &name);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -244,19 +282,22 @@ v9fs_t_walk(struct v9fs_session_info *v9ses, u32 fid, u32 newfid,
 
 int
 v9fs_t_open(struct v9fs_session_info *v9ses, u32 fid, u8 mode,
-	    struct v9fs_fcall **fcall)
+	    struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
-	int errorno = -1;
+	int ret;
+	struct v9fs_fcall *tc;
 
 	dprintk(DEBUG_9P, "fid %d mode %d\n", fid, mode);
-	msg.id = TOPEN;
-	msg.params.topen.fid = fid;
-	msg.params.topen.mode = mode;
 
-	errorno = v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	ret = -ENOMEM;
+	tc = v9fs_create_topen(fid, mode);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
 
-	return errorno;
+	return ret;
 }
 
 /**
@@ -269,14 +310,22 @@ v9fs_t_open(struct v9fs_session_info *v9ses, u32 fid, u8 mode,
 
 int
 v9fs_t_remove(struct v9fs_session_info *v9ses, u32 fid,
-	      struct v9fs_fcall **fcall)
+	      struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
 	dprintk(DEBUG_9P, "fid %d\n", fid);
-	msg.id = TREMOVE;
-	msg.params.tremove.fid = fid;
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+
+	ret = -ENOMEM;
+	tc = v9fs_create_tremove(fid);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -292,20 +341,23 @@ v9fs_t_remove(struct v9fs_session_info *v9ses, u32 fid,
 
 int
 v9fs_t_create(struct v9fs_session_info *v9ses, u32 fid, char *name,
-	      u32 perm, u8 mode, struct v9fs_fcall **fcall)
+	      u32 perm, u8 mode, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
+	int ret;
+	struct v9fs_fcall *tc;
 
 	dprintk(DEBUG_9P, "fid %d name '%s' perm %x mode %d\n",
 		fid, name, perm, mode);
 
-	msg.id = TCREATE;
-	msg.params.tcreate.fid = fid;
-	msg.params.tcreate.name = name;
-	msg.params.tcreate.perm = perm;
-	msg.params.tcreate.mode = mode;
+	ret = -ENOMEM;
+	tc = v9fs_create_tcreate(fid, name, perm, mode);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, rcp);
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
 
-	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);
+	return ret;
 }
 
 /**
@@ -320,31 +372,30 @@ v9fs_t_create(struct v9fs_session_info *v9ses, u32 fid, char *name,
 
 int
 v9fs_t_read(struct v9fs_session_info *v9ses, u32 fid, u64 offset,
-	    u32 count, struct v9fs_fcall **fcall)
+	    u32 count, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
-	struct v9fs_fcall *rc = NULL;
-	long errorno = -1;
-
-	dprintk(DEBUG_9P, "fid %d offset 0x%lx count 0x%x\n", fid,
-		(long unsigned int)offset, count);
-	msg.id = TREAD;
-	msg.params.tread.fid = fid;
-	msg.params.tread.offset = offset;
-	msg.params.tread.count = count;
-	errorno = v9fs_mux_rpc(v9ses->mux, &msg, &rc);
-
-	if (!errorno) {
-		errorno = rc->params.rread.count;
-		dump_data(rc->params.rread.data, rc->params.rread.count);
-	}
-
-	if (fcall)
-		*fcall = rc;
-	else
-		kfree(rc);
+	int ret;
+	struct v9fs_fcall *tc, *rc;
 
-	return errorno;
+	dprintk(DEBUG_9P, "fid %d offset 0x%llux count 0x%x\n", fid,
+		(long long unsigned) offset, count);
+
+	ret = -ENOMEM;
+	tc = v9fs_create_tread(fid, offset, count);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, &rc);
+		if (!ret)
+			ret = rc->params.rread.count;
+		if (rcp)
+			*rcp = rc;
+		else
+			kfree(rc);
+
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
+
+	return ret;
 }
 
 /**
@@ -358,32 +409,31 @@ v9fs_t_read(struct v9fs_session_info *v9ses, u32 fid, u64 offset,
  */
 
 int
-v9fs_t_write(struct v9fs_session_info *v9ses, u32 fid,
-	     u64 offset, u32 count, void *data, struct v9fs_fcall **fcall)
+v9fs_t_write(struct v9fs_session_info *v9ses, u32 fid, u64 offset, u32 count,
+	const char __user *data, struct v9fs_fcall **rcp)
 {
-	struct v9fs_fcall msg;
-	struct v9fs_fcall *rc = NULL;
-	long errorno = -1;
+	int ret;
+	struct v9fs_fcall *tc, *rc;
 
-	dprintk(DEBUG_9P, "fid %d offset 0x%llx count 0x%x\n", fid,
-		(unsigned long long)offset, count);
-	dump_data(data, count);
+	dprintk(DEBUG_9P, "fid %d offset 0x%llux count 0x%x\n", fid,
+		(long long unsigned) offset, count);
 
-	msg.id = TWRITE;
-	msg.params.twrite.fid = fid;
-	msg.params.twrite.offset = offset;
-	msg.params.twrite.count = count;
-	msg.params.twrite.data = data;
+	ret = -ENOMEM;
+	tc = v9fs_create_twrite(fid, offset, count, data);
+	if (!IS_ERR(tc)) {
+		ret = v9fs_mux_rpc(v9ses->mux, tc, &rc);
 
-	errorno = v9fs_mux_rpc(v9ses->mux, &msg, &rc);
+		if (!ret)
+			ret = rc->params.rwrite.count;
+		if (rcp)
+			*rcp = rc;
+		else
+			kfree(rc);
 
-	if (!errorno)
-		errorno = rc->params.rwrite.count;
+		kfree(tc);
+	} else
+		ret = PTR_ERR(tc);
 
-	if (fcall)
-		*fcall = rc;
-	else
-		kfree(rc);
-
-	return errorno;
+	return ret;
 }
+
diff --git a/fs/9p/9p.h b/fs/9p/9p.h
index 6355392786e2..007ff639777d 100644
--- a/fs/9p/9p.h
+++ b/fs/9p/9p.h
@@ -3,6 +3,7 @@
  *
  * 9P protocol definitions.
  *
+ *  Copyright (C) 2005 by Latchesar Ionkov <lucho@ionkov.net>
  *  Copyright (C) 2004 by Eric Van Hensbergen <ericvh@gmail.com>
  *  Copyright (C) 2002 by Ron Minnich <rminnich@lanl.gov>
  *
@@ -102,10 +103,16 @@ enum {
 
 #define V9FS_NOTAG	(u16)(~0)
 #define V9FS_NOFID	(u32)(~0)
+#define V9FS_MAXWELEM	16
 
 /* ample room for Twrite/Rread header (iounit) */
 #define V9FS_IOHDRSZ	24
 
+struct v9fs_str {
+	u16 len;
+	char *str;
+};
+
 /* qids are the unique ID for a file (like an inode */
 struct v9fs_qid {
 	u8 type;
@@ -123,6 +130,29 @@ struct v9fs_stat {
 	u32 atime;
 	u32 mtime;
 	u64 length;
+	struct v9fs_str name;
+	struct v9fs_str uid;
+	struct v9fs_str gid;
+	struct v9fs_str muid;
+	struct v9fs_str extension;	/* 9p2000.u extensions */
+	u32 n_uid;		/* 9p2000.u extensions */
+	u32 n_gid;		/* 9p2000.u extensions */
+	u32 n_muid;		/* 9p2000.u extensions */
+};
+
+/* file metadata (stat) structure used to create Twstat message
+   The is similar to v9fs_stat, but the strings don't point to
+   the same memory block and should be freed separately
+*/
+struct v9fs_wstat {
+	u16 size;
+	u16 type;
+	u32 dev;
+	struct v9fs_qid qid;
+	u32 mode;
+	u32 atime;
+	u32 mtime;
+	u64 length;
 	char *name;
 	char *uid;
 	char *gid;
@@ -131,25 +161,24 @@ struct v9fs_stat {
 	u32 n_uid;		/* 9p2000.u extensions */
 	u32 n_gid;		/* 9p2000.u extensions */
 	u32 n_muid;		/* 9p2000.u extensions */
-	char data[0];
 };
 
 /* Structures for Protocol Operations */
 
 struct Tversion {
 	u32 msize;
-	char *version;
+	struct v9fs_str version;
 };
 
 struct Rversion {
 	u32 msize;
-	char *version;
+	struct v9fs_str version;
 };
 
 struct Tauth {
 	u32 afid;
-	char *uname;
-	char *aname;
+	struct v9fs_str uname;
+	struct v9fs_str aname;
 };
 
 struct Rauth {
@@ -157,12 +186,12 @@ struct Rauth {
 };
 
 struct Rerror {
-	char *error;
+	struct v9fs_str error;
 	u32 errno;		/* 9p2000.u extension */
 };
 
 struct Tflush {
-	u32 oldtag;
+	u16 oldtag;
 };
 
 struct Rflush {
@@ -171,8 +200,8 @@ struct Rflush {
 struct Tattach {
 	u32 fid;
 	u32 afid;
-	char *uname;
-	char *aname;
+	struct v9fs_str uname;
+	struct v9fs_str aname;
 };
 
 struct Rattach {
@@ -182,13 +211,13 @@ struct Rattach {
 struct Twalk {
 	u32 fid;
 	u32 newfid;
-	u32 nwname;
-	char **wnames;
+	u16 nwname;
+	struct v9fs_str wnames[16];
 };
 
 struct Rwalk {
-	u32 nwqid;
-	struct v9fs_qid *wqids;
+	u16 nwqid;
+	struct v9fs_qid wqids[16];
 };
 
 struct Topen {
@@ -203,7 +232,7 @@ struct Ropen {
 
 struct Tcreate {
 	u32 fid;
-	char *name;
+	struct v9fs_str name;
 	u32 perm;
 	u8 mode;
 };
@@ -254,12 +283,12 @@ struct Tstat {
 };
 
 struct Rstat {
-	struct v9fs_stat *stat;
+	struct v9fs_stat stat;
 };
 
 struct Twstat {
 	u32 fid;
-	struct v9fs_stat *stat;
+	struct v9fs_stat stat;
 };
 
 struct Rwstat {
@@ -274,6 +303,7 @@ struct v9fs_fcall {
 	u32 size;
 	u8 id;
 	u16 tag;
+	void *sdata;
 
 	union {
 		struct Tversion tversion;
@@ -306,10 +336,12 @@ struct v9fs_fcall {
 	} params;
 };
 
-#define V9FS_FCALLHDRSZ (sizeof(struct v9fs_fcall) + \
-	sizeof(struct v9fs_stat) + 16*sizeof(struct v9fs_qid) + 16)
+#define PRINT_FCALL_ERROR(s, fcall) dprintk(DEBUG_ERROR, "%s: %.*s\n", s, \
+	fcall?fcall->params.rerror.error.len:0, \
+	fcall?fcall->params.rerror.error.str:"");
 
-#define FCALL_ERROR(fcall) (fcall ? fcall->params.rerror.error : "")
+char *v9fs_str_copy(char *buf, int buflen, struct v9fs_str *str);
+int v9fs_str_compare(char *buf, struct v9fs_str *str);
 
 int v9fs_t_version(struct v9fs_session_info *v9ses, u32 msize,
 		   char *version, struct v9fs_fcall **rcall);
@@ -325,7 +357,7 @@ int v9fs_t_stat(struct v9fs_session_info *v9ses, u32 fid,
 		struct v9fs_fcall **rcall);
 
 int v9fs_t_wstat(struct v9fs_session_info *v9ses, u32 fid,
-		 struct v9fs_stat *stat, struct v9fs_fcall **rcall);
+		 struct v9fs_wstat *wstat, struct v9fs_fcall **rcall);
 
 int v9fs_t_walk(struct v9fs_session_info *v9ses, u32 fid, u32 newfid,
 		char *name, struct v9fs_fcall **rcall);
@@ -343,4 +375,5 @@ int v9fs_t_read(struct v9fs_session_info *v9ses, u32 fid,
 		u64 offset, u32 count, struct v9fs_fcall **rcall);
 
 int v9fs_t_write(struct v9fs_session_info *v9ses, u32 fid, u64 offset,
-		 u32 count, void *data, struct v9fs_fcall **rcall);
+		 u32 count, const char __user * data,
+		 struct v9fs_fcall **rcall);
diff --git a/fs/9p/Makefile b/fs/9p/Makefile
index e4e4ffe5a7dc..3d023089707e 100644
--- a/fs/9p/Makefile
+++ b/fs/9p/Makefile
@@ -1,17 +1,17 @@
 obj-$(CONFIG_9P_FS) := 9p2000.o
 
 9p2000-objs := \
+	trans_fd.o \
+	trans_sock.o \
+	mux.o \
+	9p.o \
+	conv.o \
 	vfs_super.o \
 	vfs_inode.o \
 	vfs_file.o \
 	vfs_dir.o \
 	vfs_dentry.o \
 	error.o \
-	mux.o \
-	trans_fd.o \
-	trans_sock.o \
-	9p.o \
-	conv.o \
 	v9fs.o \
 	fid.o
 
diff --git a/fs/9p/conv.c b/fs/9p/conv.c
index 1b9b15dfeaf0..f62434d435b3 100644
--- a/fs/9p/conv.c
+++ b/fs/9p/conv.c
@@ -30,7 +30,7 @@
 #include <linux/errno.h>
 #include <linux/fs.h>
 #include <linux/idr.h>
-
+#include <asm/uaccess.h>
 #include "debug.h"
 #include "v9fs.h"
 #include "9p.h"
@@ -45,6 +45,37 @@ struct cbuf {
 	unsigned char *ep;
 };
 
+char *v9fs_str_copy(char *buf, int buflen, struct v9fs_str *str)
+{
+	int n;
+
+	if (buflen < str->len)
+		n = buflen;
+	else
+		n = str->len;
+
+	memmove(buf, str->str, n - 1);
+
+	return buf;
+}
+
+int v9fs_str_compare(char *buf, struct v9fs_str *str)
+{
+	int n, ret;
+
+	ret = strncmp(buf, str->str, str->len);
+
+	if (!ret) {
+		n = strlen(buf);
+		if (n < str->len)
+			ret = -1;
+		else if (n > str->len)
+			ret = 1;
+	}
+
+	return ret;
+}
+
 static inline void buf_init(struct cbuf *buf, void *data, int datalen)
 {
 	buf->sp = buf->p = data;
@@ -58,12 +89,12 @@ static inline int buf_check_overflow(struct cbuf *buf)
 
 static inline int buf_check_size(struct cbuf *buf, int len)
 {
-	if (buf->p+len > buf->ep) {
-		if (buf->p < buf->ep) {
-			eprintk(KERN_ERR, "buffer overflow\n");
-			buf->p = buf->ep + 1;
-			return 0;
-		}
+	if (buf->p + len > buf->ep && buf->p < buf->ep) {
+		eprintk(KERN_ERR, "buffer overflow: want %d has %d\n",
+			len, (int)(buf->ep - buf->p));
+		dump_stack();
+		buf->p = buf->ep + 1;
+		return 0;
 	}
 
 	return 1;
@@ -127,14 +158,6 @@ static inline void buf_put_string(struct cbuf *buf, const char *s)
 	buf_put_stringn(buf, s, strlen(s));
 }
 
-static inline void buf_put_data(struct cbuf *buf, void *data, u32 datalen)
-{
-	if (buf_check_size(buf, datalen)) {
-		memcpy(buf->p, data, datalen);
-		buf->p += datalen;
-	}
-}
-
 static inline u8 buf_get_int8(struct cbuf *buf)
 {
 	u8 ret = 0;
@@ -183,85 +206,37 @@ static inline u64 buf_get_int64(struct cbuf *buf)
 	return ret;
 }
 
-static inline int
-buf_get_string(struct cbuf *buf, char *data, unsigned int datalen)
-{
-	u16 len = 0;
-
-	len = buf_get_int16(buf);
-	if (!buf_check_overflow(buf) && buf_check_size(buf, len) && len+1>datalen) {
-		memcpy(data, buf->p, len);
-		data[len] = 0;
-		buf->p += len;
-		len++;
-	}
-
-	return len;
-}
-
-static inline char *buf_get_stringb(struct cbuf *buf, struct cbuf *sbuf)
-{
-	char *ret;
-	u16 len;
-
-	ret = NULL;
-	len = buf_get_int16(buf);
-
-	if (!buf_check_overflow(buf) && buf_check_size(buf, len) &&
-		buf_check_size(sbuf, len + 1)) {
-
-		memcpy(sbuf->p, buf->p, len);
-		sbuf->p[len] = 0;
-		ret = sbuf->p;
-		buf->p += len;
-		sbuf->p += len + 1;
-	}
-
-	return ret;
-}
-
-static inline int buf_get_data(struct cbuf *buf, void *data, int datalen)
+static inline void buf_get_str(struct cbuf *buf, struct v9fs_str *vstr)
 {
-	int ret = 0;
-
-	if (buf_check_size(buf, datalen)) {
-		memcpy(data, buf->p, datalen);
-		buf->p += datalen;
-		ret = datalen;
+	vstr->len = buf_get_int16(buf);
+	if (!buf_check_overflow(buf) && buf_check_size(buf, vstr->len)) {
+		vstr->str = buf->p;
+		buf->p += vstr->len;
+	} else {
+		vstr->len = 0;
+		vstr->str = NULL;
 	}
-
-	return ret;
 }
 
-static inline void *buf_get_datab(struct cbuf *buf, struct cbuf *dbuf,
-				  int datalen)
+static inline void buf_get_qid(struct cbuf *bufp, struct v9fs_qid *qid)
 {
-	char *ret = NULL;
-	int n = 0;
-
-	if (buf_check_size(dbuf, datalen)) {
-		n = buf_get_data(buf, dbuf->p, datalen);
-		if (n > 0) {
-			ret = dbuf->p;
-			dbuf->p += n;
-		}
-	}
-
-	return ret;
+	qid->type = buf_get_int8(bufp);
+	qid->version = buf_get_int32(bufp);
+	qid->path = buf_get_int64(bufp);
 }
 
 /**
- * v9fs_size_stat - calculate the size of a variable length stat struct
+ * v9fs_size_wstat - calculate the size of a variable length stat struct
  * @stat: metadata (stat) structure
  * @extended: non-zero if 9P2000.u
  *
  */
 
-static int v9fs_size_stat(struct v9fs_stat *stat, int extended)
+static int v9fs_size_wstat(struct v9fs_wstat *wstat, int extended)
 {
 	int size = 0;
 
-	if (stat == NULL) {
+	if (wstat == NULL) {
 		eprintk(KERN_ERR, "v9fs_size_stat: got a NULL stat pointer\n");
 		return 0;
 	}
@@ -278,81 +253,38 @@ static int v9fs_size_stat(struct v9fs_stat *stat, int extended)
 	    8 +			/* length[8] */
 	    8;			/* minimum sum of string lengths */
 
-	if (stat->name)
-		size += strlen(stat->name);
-	if (stat->uid)
-		size += strlen(stat->uid);
-	if (stat->gid)
-		size += strlen(stat->gid);
-	if (stat->muid)
-		size += strlen(stat->muid);
+	if (wstat->name)
+		size += strlen(wstat->name);
+	if (wstat->uid)
+		size += strlen(wstat->uid);
+	if (wstat->gid)
+		size += strlen(wstat->gid);
+	if (wstat->muid)
+		size += strlen(wstat->muid);
 
 	if (extended) {
 		size += 4 +	/* n_uid[4] */
 		    4 +		/* n_gid[4] */
 		    4 +		/* n_muid[4] */
 		    2;		/* string length of extension[4] */
-		if (stat->extension)
-			size += strlen(stat->extension);
+		if (wstat->extension)
+			size += strlen(wstat->extension);
 	}
 
 	return size;
 }
 
 /**
- * serialize_stat - safely format a stat structure for transmission
- * @stat: metadata (stat) structure
- * @bufp: buffer to serialize structure into
- * @extended: non-zero if 9P2000.u
- *
- */
-
-static int
-serialize_stat(struct v9fs_stat *stat, struct cbuf *bufp, int extended)
-{
-	buf_put_int16(bufp, stat->size);
-	buf_put_int16(bufp, stat->type);
-	buf_put_int32(bufp, stat->dev);
-	buf_put_int8(bufp, stat->qid.type);
-	buf_put_int32(bufp, stat->qid.version);
-	buf_put_int64(bufp, stat->qid.path);
-	buf_put_int32(bufp, stat->mode);
-	buf_put_int32(bufp, stat->atime);
-	buf_put_int32(bufp, stat->mtime);
-	buf_put_int64(bufp, stat->length);
-
-	buf_put_string(bufp, stat->name);
-	buf_put_string(bufp, stat->uid);
-	buf_put_string(bufp, stat->gid);
-	buf_put_string(bufp, stat->muid);
-
-	if (extended) {
-		buf_put_string(bufp, stat->extension);
-		buf_put_int32(bufp, stat->n_uid);
-		buf_put_int32(bufp, stat->n_gid);
-		buf_put_int32(bufp, stat->n_muid);
-	}
-
-	if (buf_check_overflow(bufp))
-		return 0;
-
-	return stat->size;
-}
-
-/**
- * deserialize_stat - safely decode a recieved metadata (stat) structure
+ * buf_get_stat - safely decode a recieved metadata (stat) structure
  * @bufp: buffer to deserialize
  * @stat: metadata (stat) structure
- * @dbufp: buffer to deserialize variable strings into
  * @extended: non-zero if 9P2000.u
  *
  */
 
-static inline int
-deserialize_stat(struct cbuf *bufp, struct v9fs_stat *stat,
-		 struct cbuf *dbufp, int extended)
+static inline void
+buf_get_stat(struct cbuf *bufp, struct v9fs_stat *stat, int extended)
 {
-
 	stat->size = buf_get_int16(bufp);
 	stat->type = buf_get_int16(bufp);
 	stat->dev = buf_get_int32(bufp);
@@ -363,45 +295,17 @@ deserialize_stat(struct cbuf *bufp, struct v9fs_stat *stat,
 	stat->atime = buf_get_int32(bufp);
 	stat->mtime = buf_get_int32(bufp);
 	stat->length = buf_get_int64(bufp);
-	stat->name = buf_get_stringb(bufp, dbufp);
-	stat->uid = buf_get_stringb(bufp, dbufp);
-	stat->gid = buf_get_stringb(bufp, dbufp);
-	stat->muid = buf_get_stringb(bufp, dbufp);
+	buf_get_str(bufp, &stat->name);
+	buf_get_str(bufp, &stat->uid);
+	buf_get_str(bufp, &stat->gid);
+	buf_get_str(bufp, &stat->muid);
 
 	if (extended) {
-		stat->extension = buf_get_stringb(bufp, dbufp);
+		buf_get_str(bufp, &stat->extension);
 		stat->n_uid = buf_get_int32(bufp);
 		stat->n_gid = buf_get_int32(bufp);
 		stat->n_muid = buf_get_int32(bufp);
 	}
-
-	if (buf_check_overflow(bufp) || buf_check_overflow(dbufp))
-		return 0;
-
-	return stat->size + 2;
-}
-
-/**
- * deserialize_statb - wrapper for decoding a received metadata structure
- * @bufp: buffer to deserialize
- * @dbufp: buffer to deserialize variable strings into
- * @extended: non-zero if 9P2000.u
- *
- */
-
-static inline struct v9fs_stat *deserialize_statb(struct cbuf *bufp,
-						  struct cbuf *dbufp,
-						  int extended)
-{
-	struct v9fs_stat *ret = buf_alloc(dbufp, sizeof(struct v9fs_stat));
-
-	if (ret) {
-		int n = deserialize_stat(bufp, ret, dbufp, extended);
-		if (n <= 0)
-			return NULL;
-	}
-
-	return ret;
 }
 
 /**
@@ -409,194 +313,27 @@ static inline struct v9fs_stat *deserialize_statb(struct cbuf *bufp,
  * @buf: buffer to deserialize
  * @buflen: length of received buffer
  * @stat: metadata structure to decode into
- * @statlen: length of destination metadata structure
  * @extended: non-zero if 9P2000.u
  *
+ * Note: stat will point to the buf region.
  */
 
-int v9fs_deserialize_stat(void *buf, u32 buflen, struct v9fs_stat *stat,
-			  u32 statlen, int extended)
+int
+v9fs_deserialize_stat(void *buf, u32 buflen, struct v9fs_stat *stat,
+		int extended)
 {
 	struct cbuf buffer;
 	struct cbuf *bufp = &buffer;
-	struct cbuf dbuffer;
-	struct cbuf *dbufp = &dbuffer;
+	unsigned char *p;
 
 	buf_init(bufp, buf, buflen);
-	buf_init(dbufp, (char *)stat + sizeof(struct v9fs_stat),
-		 statlen - sizeof(struct v9fs_stat));
-
-	return deserialize_stat(bufp, stat, dbufp, extended);
-}
-
-static inline int v9fs_size_fcall(struct v9fs_fcall *fcall, int extended)
-{
-	int size = 4 + 1 + 2;	/* size[4] msg[1] tag[2] */
-	int i = 0;
+	p = bufp->p;
+	buf_get_stat(bufp, stat, extended);
 
-	switch (fcall->id) {
-	default:
-		eprintk(KERN_ERR, "bad msg type %d\n", fcall->id);
+	if (buf_check_overflow(bufp))
 		return 0;
-	case TVERSION:		/* msize[4] version[s] */
-		size += 4 + 2 + strlen(fcall->params.tversion.version);
-		break;
-	case TAUTH:		/* afid[4] uname[s] aname[s] */
-		size += 4 + 2 + strlen(fcall->params.tauth.uname) +
-		    2 + strlen(fcall->params.tauth.aname);
-		break;
-	case TFLUSH:		/* oldtag[2] */
-		size += 2;
-		break;
-	case TATTACH:		/* fid[4] afid[4] uname[s] aname[s] */
-		size += 4 + 4 + 2 + strlen(fcall->params.tattach.uname) +
-		    2 + strlen(fcall->params.tattach.aname);
-		break;
-	case TWALK:		/* fid[4] newfid[4] nwname[2] nwname*(wname[s]) */
-		size += 4 + 4 + 2;
-		/* now compute total for the array of names */
-		for (i = 0; i < fcall->params.twalk.nwname; i++)
-			size += 2 + strlen(fcall->params.twalk.wnames[i]);
-		break;
-	case TOPEN:		/* fid[4] mode[1] */
-		size += 4 + 1;
-		break;
-	case TCREATE:		/* fid[4] name[s] perm[4] mode[1] */
-		size += 4 + 2 + strlen(fcall->params.tcreate.name) + 4 + 1;
-		break;
-	case TREAD:		/* fid[4] offset[8] count[4] */
-		size += 4 + 8 + 4;
-		break;
-	case TWRITE:		/* fid[4] offset[8] count[4] data[count] */
-		size += 4 + 8 + 4 + fcall->params.twrite.count;
-		break;
-	case TCLUNK:		/* fid[4] */
-		size += 4;
-		break;
-	case TREMOVE:		/* fid[4] */
-		size += 4;
-		break;
-	case TSTAT:		/* fid[4] */
-		size += 4;
-		break;
-	case TWSTAT:		/* fid[4] stat[n] */
-		fcall->params.twstat.stat->size =