MD5 and SCP/SFTP

[Bart Buitenhuis_1]

I'm using either SFTP or SCP to transfer files from OpenVMS to HPUX. Unfortunately the MD5 checksum on OpenVMS gives us a different result on the HPUX-box. When I FTP the files (in BINAIRY MODE) the checksum gives a good result.
Anybody any ideas how to get this working with SCP/SFTP?

[Willem Grooters]

It might be that encrypting/decrypting might cause a problem. FTP doesn't encrypt and will, in binary mode, send the files as they are. No wonder that MD5 checksum is the same.
Also, I IRC, there are some issues with SCP and/or SFTP on OpenVMS.

Could is be that the endian difference (VMS is little-endian, HP-UX is big-endian) is causing a problem?

Willem

[Bart Buitenhuis_1]

OK, So we want to be really secure with our transfer but the method we use might screw up the file and there is no way we can check this by using a checksum...*sigh*
;-)

[Ian Miller.]

is sftp doing a binary mode transfer?
If not then it may be messing with the file contents.

Binary file transfers with sftp have been fixed in recent ECO kits for hp TCPIP Services for VMS.

[Bojan Nemec]

Bart,

Maybe the diference is because of different file structure. On HPUX you have stream files with records terminated by . On VMS you can have different record formats (fixed,stream,stream_lf,stream_cr,variable,VFC...).

You can see this with DIR /FULL as "Record format:". If yours file is not Stream_LF try to convert the file with:

$ CONVERT/FDL=SYS$INPUT input output
RECORD; FORMAT STREAM_LF


Bojan

[Bart Buitenhuis_1]

Ok,
On the testfile I could do this but when I receive files that are being transferred to another node/application I'm not sure if I can.

I've SCP'd a ascii file to UX and SCP'd it back. The checksum is correct so it's certain that the file landed correctly on UX. Perhaps I'll write my own "checksum" this way.

[Bojan Nemec]

Bart,

What are you using to compute the checksum on the VMS part?

If my previous suggestion works (I am not able to test it right now) could you arrange that the files will be originaly written as stream_lf files.

Bojan

[Bart Buitenhuis_1]

Bojan,
I use MD5 on VMS and on UX

[Bojan Nemec]

Bart,

I have no MD5 on my VMS system. Maybe you are on 8.2 and you use $ CHECKSUM/ALGORITHM=MD5. Or you use md5sum from the GNV distribution. (this seems broken on my system).

I copy the Reference Implementation from the RFC:

http://www.faqs.org/rfcs/rfc1321.html

and try it on a file. The error was exacltly as I predicted in my previous post. If you compute the checksum on two files where one is variable and the other is the same but stream_lf the checksum is different.

If you modify the source (mddriver.c) to open the file in ascii mode "r" and not binary mode "rb" the result is same. It is also the same as my linux system gives me with md5sum.

Bojan

[Bojan Nemec]

Sorry,

I reread my previous post and see that is not clear.

I mean that if you modify the open mode of the file the checksums are equal independent on the record format.

Bojan