commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jeremy Gustie (JIRA)" <>
Subject [jira] [Commented] (COMPRESS-376) decompressConcatenated improvement
Date Tue, 20 Dec 2016 02:57:58 GMT


Jeremy Gustie commented on COMPRESS-376:

Agree, definitely more sophisticated...

I have been trying to isolate what it is that made the archive we found fail, my best guess
is that the last entry needs to cross a boundary occurring at {{length - 512}} bytes of the
file. I was able to produce an 8K file that only allows me to read the final entry if I do
not buffer between the compressor and archiving streams by doing this:

head -c 7680 </dev/urandom > test1.rnd
head -c 10 </dev/urandom > test2.rnd
tar cz test1.rnd test2.rnd | cat - /dev/zero | head -c 8192 > COMPRESS-376.tar.gz

I haven't had much of a chance to figure out exactly why this is, but hopefully this at least
gives you an archive that can fail that test.

> decompressConcatenated improvement
> ----------------------------------
>                 Key: COMPRESS-376
>                 URL:
>             Project: Commons Compress
>          Issue Type: Improvement
>          Components: Compressors
>            Reporter: Jeremy Gustie
> First the problem I am seeing: in general I am always setting {{decompressConcatenated}}
to {{true}}, most of the time this works fine. However, it seems like some versions of Python
tarfile will pad a compressed TAR file with null bytes. The null bytes are recognized as garbage,
causing decompression to fail. Unfortunately this failure occurs while filling a buffer for
data used to read the final entry in the TAR file causing {{TarArchiveInputStream.getNextEntry}}
to fail before the last entry can be returned.
> There are a couple of potential solutions I can see:
> 1. The easiest thing to do we be to special case the null padding and just terminate
without failing (in the {{GzipCompressorInputStream.init}} method, this amounts to adding
a check for {{magic0 == 0 && (magic1 == 0 || magic1 == -1)}} and returning {{false}}).
Perhaps draining the underlying stream to ensure that the remaining bytes are all null could
reduce the likelihood of a false positive recognizing the padding.
> 2. Change {{decompressConcatenated}} to a tri-state value (maybe add an extra {{ignoreGarbage}}
flag) to suppress the failure; basically concatenated streams would be decompressed only if
the appropriate magic is found. This has API impact but completely preserves backwards compatibility.
> 3. Finally, deferring the failure to the next read attempt may also be a viable solution
that nearly preserves backwards compatibility. As I mentioned before, the "Garbage after..."
error occurs while reading the final entry in a TAR file: if the current read (which contains
all of the final data from the compression stream) were allowed to complete normally, the
downstream consumer might also complete normally; the next attempt to read (the garbage past
the end of the compression stream) would be the read that fails with the "Garbage after..."
error. This gives the downstream code the best opportunity to both process the full compression
stream and receive the unexpected garbage failure.
> I was mostly looking at the {{GzipCompressorInputStream}}, I suspect similar changes
would be needed in the other decompress-concatenated compressor streams.

This message was sent by Atlassian JIRA

View raw message