Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 531b1094 authored by Latchesar Ionkov's avatar Latchesar Ionkov Committed by Linus Torvalds
Browse files

[PATCH] v9fs: zero copy implementation 



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

Signed-off-by: default avatarLatchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@ericvh.myip.org>
Signed-off-by: default avatarAndrew Morton <akpm@osdl.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@osdl.org>
parent d8da097a

fs/9p/9p.c

+176 −126
Original line number Diff line number Diff line 
/*

 *  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);



	return err;

	if (ret)

		dprintk(DEBUG_ERROR, "failed fid %d err %d\n", fid, ret);



	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", oldtag);



	dprintk(DEBUG_9P, "oldtag %d\n", tag);

	msg.id = TFLUSH;

	msg.params.tflush.oldtag = tag;

	return v9fs_mux_rpc(v9ses->mux, &msg, NULL);

	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);



	return v9fs_mux_rpc(v9ses->mux, &msg, fcall);

	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 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);

	}

	int ret;

	struct v9fs_fcall *tc, *rc;



	if (fcall)

		*fcall = rc;

	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);



	return errorno;

		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;



	dprintk(DEBUG_9P, "fid %d offset 0x%llx count 0x%x\n", fid,

		(unsigned long long)offset, count);

	dump_data(data, count);



	msg.id = TWRITE;

	msg.params.twrite.fid = fid;

	msg.params.twrite.offset = offset;

	msg.params.twrite.count = count;

	msg.params.twrite.data = data;

	int ret;

	struct v9fs_fcall *tc, *rc;



	errorno = v9fs_mux_rpc(v9ses->mux, &msg, &rc);

	dprintk(DEBUG_9P, "fid %d offset 0x%llux count 0x%x\n", fid,

		(long long unsigned) offset, count);



	if (!errorno)

		errorno = rc->params.rwrite.count;

	ret = -ENOMEM;

	tc = v9fs_create_twrite(fid, offset, count, data);

	if (!IS_ERR(tc)) {

		ret = v9fs_mux_rpc(v9ses->mux, tc, &rc);



	if (fcall)

		*fcall = rc;

		if (!ret)

			ret = rc->params.rwrite.count;

		if (rcp)

			*rcp = rc;

		else

			kfree(rc);



	return errorno;

		kfree(tc);

	} else

		ret = PTR_ERR(tc);



	return ret;

}

fs/9p/9p.h

+54 −21
Original line number Diff line number Diff line
@@ -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;
@@ -115,6 +122,29 @@ struct v9fs_qid { 


/* Plan 9 file metadata (stat) structure */

struct v9fs_stat {

	u16 size;

	u16 type;

	u32 dev;

	struct v9fs_qid qid;

	u32 mode;

	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;
@@ -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);

fs/9p/Makefile

+5 −5
Original line number Diff line number Diff line 
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

fs/9p/conv.c

+518 −377

File changed.

Preview size limit exceeded, changes collapsed.

fs/9p/conv.h

+22 −6
Original line number Diff line number Diff line 
/*

 * linux/fs/9p/conv.h

 *

 * 9P protocol conversion definitions

 * 9P protocol conversion 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>

 *
@@ -25,11 +26,26 @@ 
 */



int v9fs_deserialize_stat(void *buf, u32 buflen, struct v9fs_stat *stat,

	u32 statlen, int extended);

int v9fs_serialize_fcall(struct v9fs_fcall *tcall, void *buf, u32 buflen,

	int extended);

int v9fs_deserialize_fcall(void *buf, u32 buflen, struct v9fs_fcall *rcall,

	int rcalllen, int extended);

	int extended);



void v9fs_set_tag(struct v9fs_fcall *fc, u16 tag);



/* this one is actually in error.c right now */

int v9fs_errstr2errno(char *errstr);
 
struct v9fs_fcall *v9fs_create_tversion(u32 msize, char *version);

struct v9fs_fcall *v9fs_create_tauth(u32 afid, char *uname, char *aname);

struct v9fs_fcall *v9fs_create_tattach(u32 fid, u32 afid, char *uname,

	char *aname);

struct v9fs_fcall *v9fs_create_tflush(u16 oldtag);

struct v9fs_fcall *v9fs_create_twalk(u32 fid, u32 newfid, u16 nwname,

	char **wnames);

struct v9fs_fcall *v9fs_create_topen(u32 fid, u8 mode);

struct v9fs_fcall *v9fs_create_tcreate(u32 fid, char *name, u32 perm, u8 mode);

struct v9fs_fcall *v9fs_create_tread(u32 fid, u64 offset, u32 count);

struct v9fs_fcall *v9fs_create_twrite(u32 fid, u64 offset, u32 count,

	const char __user *data);

struct v9fs_fcall *v9fs_create_tclunk(u32 fid);

struct v9fs_fcall *v9fs_create_tremove(u32 fid);

struct v9fs_fcall *v9fs_create_tstat(u32 fid);

struct v9fs_fcall *v9fs_create_twstat(u32 fid, struct v9fs_wstat *wstat,

	int extended);