[
https://issues.apache.org/jira/browse/TIKA-2446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16503280#comment-16503280
]
Tim Allison edited comment on TIKA-2446 at 6/6/18 2:01 PM:
-----------------------------------------------------------
We can prevent an OOM in our code by changing how we open the package:
{noformat}
ZipEntrySource zipEntrySource = new ZipFileZipEntrySource(new
java.util.zip.ZipFile(stream.getFile()));
OPCPackage pkg = OPCPackage.open(zipEntrySource);{noformat}
Java's ZipFile throws an IOException during detection which is swallowed, and
then commons-compress throws an EOFException when the parser tries to parse the
file.
I agree that we should try to fix this at the POI level, though. In 3.17 at
least there's a check that the length of the stream is < Integer.MAX_VALUE,
which, as noted above is 2GB. Although I've done it before it feels hacky to
limit this to, say 1 GB. I wonder if POI should write out to a new temp file,
rather than trying to buffer the defective ZIP into memory? Maybe not
pre-allocate the byte[]...given, as pointed out above, that by this point we
know we're dealing with a corrupted file?
was (Author: talli...@mitre.org):
We can prevent an OOM in our code by changing how we open the package:
{noformat}
ZipEntrySource zipEntrySource = new ZipFileZipEntrySource(new
java.util.zip.ZipFile(stream.getFile()));
OPCPackage pkg = OPCPackage.open(zipEntrySource);{noformat}
Java's ZipFile throws an IOException during detection which is swallowed, and
then commons-compress throws an EOFException when the parser tries to parse the
file.
I agree that we should try to fix this at the POI level, though. In 3.17 at
least there's a check that the length of the stream is < Integer.MAX_VALUE,
which, as noted above is 2GB. Although I've done it before it feels hacky to
limit this to, say 1 GB. I wonder if POI should write out to a new temp file,
rather than trying to buffer the defective ZIP into memory?
> Tainted Zip file can provoke OOM errors
> ---------------------------------------
>
> Key: TIKA-2446
> URL: https://issues.apache.org/jira/browse/TIKA-2446
> Project: Tika
> Issue Type: Bug
> Affects Versions: 1.16
> Reporter: Thorsten Schäfer
> Priority: Major
> Attachments: corrupt_zip.zip
>
>
> Hi,
> using Tika 1.16 with embedded POI 3.17-beta1 we experienced an OutOfMemory
> error on a Zip file. The suspicious code is in the constructor of
> FakeZipEntry in line 125. Here a ByteArrayOutputStream of up to 2 GiB in size
> is opened which will most probably lead to an OutOfMemory. The entry size in
> the zip file can be easily faked by an attacker.
> The code path to FakeZipEntry will be used only if the native
> java.util.zip.ZipFile implementation already failed to open the (possibly
> corrupted) Zip. Possibly a more fine grained error analysis could be done in
> ZipPackage.
> I have attached a tweaked zip file that will provoke this error.
> {code:java}
> public FakeZipEntry(ZipEntry entry, InputStream inp) throws IOException {
> super(entry.getName());
>
> // Grab the de-compressed contents for later
> ByteArrayOutputStream baos;
> long entrySize = entry.getSize();
> if (entrySize !=-1) {
> if (entrySize>=Integer.MAX_VALUE) {
> throw new IOException("ZIP entry size is too large");
> }
> baos = new ByteArrayOutputStream((int) entrySize);
> } else {
> baos = new ByteArrayOutputStream();
> }
> {code}
> Kinds,
> Thorsten
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
