gpg: [don't know]: invalid packet (ctb=2d)
Closed, ResolvedPublic

Assigned To
Authored By
mlongtin
Jan 8 2010, 9:02 PM

Description

I have an issue like bug 997. The whole file gets decrypted, but it finishes
with this:

gpg: [don't know]: invalid packet (ctb=2d)
gpg: [don't know]: invalid packet (ctb=2d)

It's happening with an automated process, and about 1 out of 5 of the files it
encrypts get this issue. The files are encrypted with the -a option.

The commonality is that they are exactly 512 lines of encoded data + the
checksum line, or 32K on the nose. Like this:

...
BH0rOw1S24VKlqKfFrMujgPTp7DZOJzm6JEA2Khx3vm2L9iYL6VgltJZS7sAeuc1

yz2X

-----END PGP MESSAGE-----

If you want, I can send you a few sample of the source file + encrypted files.

I can't seem to replicate the encryption on a different machine, don't know why.

The encrypting software is 1.4.9. I tried decrypting with both 1.4.9 and 1.4.10.

Details

Version
1.4.10

Event Timeline

mlongtin set Version to 1.4.10.
mlongtin added a subscriber: mlongtin.

To tarck this down I would also need the key. Can you create such a broken
message using one of the example keys (e.g. alfa@example.net)? IF so, feel free
to send it to me directly wk@gnupg.org - if the data is kind of sensitive, feeel
free to encrypt. Due to anti-spam meastures it is best to send a gzipped
tarball or any other gzipped data; or send me an URL.

I tracked it down with GPG (hadn't used that in 10 years).

Basically, if the CRC is alone on a line by itself, the function radix64_read in
armor.c returns -1 because of this line:

if( !n && !onlypad )
    rc = -1;

This cause iobuf.c:underflow to deallocate the filter, leaving the file_filter
alone. So next time it reads using the file_filter and gets the '-----END PGP
MESSAGE-----'.

If the CRC is not a line by itself, the armor_filter is not deallocated, but
reads to the end of the file first, which deallocates the file_filter, then says
it doesn't have any more data (rc=-1,len=0).

I am inclined to fix this by putting this check towards the end of radix64_read:

Line 929-941 of armor.c:
#endif

}
if(n==0)
  onlypad=1;
    }
}

if( !n && !onlypad )
    rc = -1;

*retn = n;
return rc;

}

This fixes my issue.

Included is an example of a message that once encrypted causes problems. The
message was encrypted with a public key but not signed.

Thanks for the excellent description.

• werner changed Due Date from Jan 31 2009, 1:00 AM to May 31 2009, 2:00 AM.May 12 2010, 12:04 PM
• werner claimed this task.
• werner removed Due Date.

It turned out to be some work to reproduce the case. The length of the output
depends on the key and other factors. I created a test case which is now used
in armor.test which clearly shows this.

The bug fix was to extend David's, 4 year old fix for a key import problem.
We will release 1.4.11 in a couple of weeks.