By Daniel Hellerstein
15 May 2000
RxRsync ver 1.03: Rexx procedures for the rsync differencing protocol
Absract:
RxRsync contains several OS/2 classic REXX procedures that implement
the rsync "differencing" protocol. This document describes their use.
---------------------
I) Introduction
The Rsync protocol is a "client/server" differencing protocol that
does not require that both parties have the same copy of a prior
version. Instead, the client sends a synopsis of a prior
version, information which the server can use to create a difference
file (it uses this synopsis in lieu of the actual contents of the
prior version).
Thus, rsync trades off some extra exchange of information (the synopsis
the client sends to the server), in return for removing the
need for both sides of the transaction having identical copies of
the prior version. Of course, the efficiency of rsync is a function
of how close the prior version (on the client side) is to the current
version (on the server side). Nevertheless, in even the worst case
(comparison of two different random files), the penalty is small; wheras
in the best case (slight changes), 10 to 1 reductions in total (both ways)
message size are not uncommon.
The rsync protocol has several steps. Assume a client wants a current
version of a file; and that the client has a prior version of the
file available. Then:
1) A "client" creates a "synopsis" of it's prior version.
2) The client requests the "server" for the current version, and
sends a copy of this synopsis along with this request
3) The server creates a "difference" file by comparing the synopsis
to it's copy of the current version.
4) The server send this difference file back to the client
5) The client combines the difference file with it's copy of the prior
version to create an exact duplicate of the new version.
Note that the server is NOT expected to have a copy of the prior version!
The procedures in RxRsync can be used to implement this chain of events.
About all you need to do is worry about the communication steps (steps
2 and 4).
---------------------
II) Installing RxRsync
First, unzip RXRSYNC.ZIP to an empty temporary directory.
1) Copy RXRsync.DLL to your LIBPATH (say, copy it to x:\OS2\DLL).
You will also need a copy of REXXUTIL, but that's part of
most OS/2 installations.
2) Write a REXX program, and either:
a)include a copy of RxRSYNC.REX (say, put it at the end of your
rexx program file)
b) Load RxRSYNC.RXL into "macrospace", and load a few dlls.
See the notes in section IV below for the details.
3) Within your REXX programs, call the procedures.
Alternatively, you can use the RXSYNC.CMD program -- it's an "all rexx"
implementation of rsync. It's very slow, but it should work on
any REXX system.
---------------------
III) Description of procedures
There are three REXX procedures:
Rsync_Synopsis: creates a synopsis of an old version.
Rsync_Gdiff: uses this synopsis, and the new version, to create
a gdiff-formatted difference file
Rsync_Uungdiff: uses a gdiff-formatted difference file, and the old
version, to build a copy of the new version
Rsync_Synopsis: create a synopsis of an old version
Syntax:
status=Rsync_Synopsis(oldver_file,synopsis_file,comment,quiet,blocksize)
where:
oldver_file: a fully qualified file name (the old version).
synopsis_file: a fully qualified file name (the synopsis file)
comment: an optional comment
quiet: set to 1 to suppress runtime status messages
This is an optional parameter (the default is 0).
blocksize: blocksize to be used when creating synopsis file.
Sizes between 500 and 1000 seem to work best.
This is an optional parameter (the default is 500)
and
status: A status message of the form:
stat multi-word message
where
stat= OK for success
ERROR for failure
Notes:
* Examples of status returned values:
ERROR no such old version
OK 5151 bytes written to C1FILE.RSY
* The synopsis_file will be created in "overwrite mode". If
a file with this name exists, it will be deleted (that is,
synopsis_file is written in "overwrite" mode).
* The comment can be up to 80 characters long. If not specified,
a timestamp is used
Rsync_Gdiff: use a synopsis to create a GDIFF-formatted difference file
Syntax:
status=Rsync_Gdiff(synopsis_file,newver_file,diff_file,quiet)
where:
synopsis_file: a synopsis file
newver_file: a fully qualified file name (the new version)
diff_file: a fully qualified file name (the difference file)
quiet: set to 1 to suppress runtime status messages
This is an optional parameter (the default is 0).
and
status: A status message
The status message is either:
OK md4_value
or
ERROR Error message
The md4_value is a 32 hex character MD4 hash of the newver_file.
See Rsync_unGdiff for an example of how it can be used.
Notes:
* The synopsis_file should exist -- say, due to a prior
call to Rsync_Synopsis; or because a client "sent it to you").
* The newver_file should exist -- it's the "new version" of the
oldver_file
* The diff_file will be created in "overwrite" mode.
rsync_unGdiff: create a duplicate of a new file from a difference file
Syntax:
status=rsync_unGdiff(oldver_file,diff_file,newver_file,amd4,quiet)
where:
oldver_file: a fully qualified file name (the old version)
diff_file: a fully qualified file name (the difference file)
newver_file: a fully qualified file name (the "duplicate" new ver)
amd4: (optional) md4 of the "server's new" version of the file.
quiet:(optional) set to 1 to suppress runtime status messages
The default is 0).
and
status: A status message
Notes:
* the oldver_file must exist and be available
* the diff_file must exist -- say, it's a file that a "server" created
using rsync_ungdiff and sent to you.
* the newver_file will be created in "overwrite" mode.
* the status has the same structure as status in Rsync_Synopsis
* if you specify amd4, and the md4 hash of the newver_file (that
is created) does NOT match amd4, then an error message is
generated (and newver_file is NOT created).
* rsync_unGdiff can be used for ANY "gdiff-formatted" difference
file -- not just "gdiff-formatted" difference files produced by
rsync_Gdiff. That is, if a program produced a GDiff formatted
difference file by comparing oldver (which you have available)
and newver (which you don't have), then you can recreate
this newever by rsync_ungdiff (with oldver_file and this "difference file"
as arguments)
---------------------
IV) The rxRsync.Dll dynamic link library.
Since REXX is very slow at repetitive math, the above rexx procedures
use several procedures in rxRsync.dll.
RxRsyncLoad: Loads the rxRsync procedures.
For example:
if rxfuncquery('rx_md4')=1 then do
call RXFuncAdd 'RXRsyncLoad', 'RXRSYNC', 'RxRsyncLoad'
call RxRsyncLoad
end
if rxfuncquery('rx_md4')=1 then do
return "ERROR could not load RxRsync.DLL"
end
RxRsyncDrop: unload the rxRsync procedures
For example:
call RxRsyncDrop
RX_RSYNC32: Compute a 32 bit rolling checksum of a string
For example:
csum32=rx_rsync32('some kind of string of any length')
csum32 will be an 8 character hex number
RX_ADLER32: Compute an adler-32 checksum
This is the standard adler-32 checksum (which differs from the
"modified adler 32 checksum" used by rsync).
Syntax:
a32=rx_adler32(astring)
where
a32 is an 8 hex character string (possibly padded with 0s on the left)
Example:
a32x=rx_adler32('This is a very long string. Really!')
Alternatively, you can build the checksum in pieces; by sequentially
calling rx_adler32 with substrings. This is useful if you want the checksum
of a large file.
Syntax:
a32=rx_adler32(astring,olda32)
where
olda32 is an 8 hex character string returned by an earlier call to
rx_adler32
Example:
a32a=rx_adler32('This is a very long string.')
a32b=rx_adler32(' Really!',a32a)
Note: a32b will equal a32x
RX_MD4: Compute an MD4 hash of a string or a file
syntax:
amd4=rx_md4(stuff,[fileflag])
If fileflag not specified, or not equal to 1, then stuff is a string.
Otherwise, stuff should be a fully qualified filename.
Examples:
amd4=rx_md4('some kind of string of any length')
fmd4=rx_md4('d:\test\foo.bar',1)
amd4 (or fmd4) will be a 32 hex character md4.
RX_RSYNC32_MD4: Compute a 32 bit rolling checksum, and an md4 hash
For example:
csum32_md4=rx_rsync32_md4('some kind of string of any length')
csum32 will be an 20 characters. The first 4 are the rolling
checksum, characters 5 to 20 are the MD4.
Thus: csum32=c2x(substr(csum_32_md4),1,4)
amd4=c2x(substr(csum_32_md4),5,16)
RX_Rsync_Gdiff: Compute a gdiff-formatted "difference", given a
"current instance and a "synopsis".
status=rx_Rsync_Gdiff(newverfile,synopsis,outfile,use4)
where:
newverfile: a filename, pointing to the "current instance"
synopsis: the string containing the "synopsis" of the "old
instance" (as may be produced by Rsync_Synopsis)
Note: this is a character string, it's NOT a filename!
outfile: a file name, the "difference" file will be written
to this file name (in overwrite mode)
use4: Optional. If set to 1, then only the first 4 characters of
the md4 checksum are used to verify. This is useful
when using the http version of rsync (smaller request
headers, with some small risk of an incorrect "undifferencing",
which may necessitate a re-request).
status is a status message. It can be:
OK md4_value
or
ERROR error message
The md4_value is a 32 hex character md4 hash of newverfile
---------------------
V) Notes and disclaimer
* Ver 1.03 fixed a "sharing violation" (RX_RSYNC_UNGDIFF was neglecting
to close an input file). Thanks to Heiko Scheidt for finding the problem
and the solution!
* RSYNCtst.CMD demonstrates the use of these procedures as text inclusions.
* RSyncts2.CMD demonstrates their use as a macrospace library.
* !!! If you use the "macrospace" version, you MUST be sure to
load several dlls, and to load the macrospace library.
RsyncTs2 contains a simple procedure (LOAD_LIBS) that will do this.
* The Rsync protocol was invented by Andrew Tridgell. For more information,
see http://samba.org.au/rsync/
* A description of the GDIFF format can be found at:
http://www.w3.org/TR/NOTE-gdiff-19970901.html
* Users of the SRE-http web server (http://www.srehttp.org) can use
the sreRsync "pre-reply procedure", and the DoGET.CMD http requester,
as an http implementation of rsync.
* Structure of a synopsis file
Comment -- 80 characters (i.e.; a requested file name)
1 space
Blocksize -- 6 digit integer (i.e; 500)
1 space
#Blocks -- 8 digit character (N)
1 space
md4 -- 32 digit md4
3 spaces
chksum1||md41||..||chksumN||md4N -- chksum and md4 values
(machine integer format,
high order bytes first)
Note: this is subject to change (it may be standardized to
be compatible with unix implementations of rsync)
* Contents of rxRsync.zip
read.me -- a small read.me file
RxRSYNC.RXL -- REXX "macrospace" version of the three REXX procedures
rxrsync.rex -- REXX code version of the three REXX procedures
rxsync.cmd -- An all REXX implementation of Rsync (for demo purposes)
rsyncts2.cmd -- Demo of rsync, using RxRSYNC.DLL
rsynctst.cmd -- Demo of rsync, using (local copy of) rxrsync.rex
rxrsync.doc -- this documentation file
RxRsync.dll -- a Rexx callable dll containing several procedures
dllsrc.zip -- Source code (watcom fortran, and rexx) used to create
RxRsync.dll and RxRsync.rxl
---------------------
Disclaimer:
This is freeware that is to be used at your own risk -- the
author and any potentially affiliated institutions disclaim all
responsibilties for any consequence arising from the use, misuse, or abuse
of this software (or pieces of this software).
You may use this (or subsets of this) program as you see fit,
including for commercial purposes; so long as proper attribution
is made, and so long as such use does not in any way preclude
others from making use of this code.
Contact:
Daniel Hellerstein (danielh@crosslink.net or danielh@econ.ag.gov)