Created attachment 114718 [details]
Proof of Concept

Hello,


I found an issue that may allow to maliciously deploy file at arbitrary location by opening Okular archives.

When opening an Okular archive, "content.xml" is parsed in order to extract the document name (field "DocumentFileName"). If this file is also present in the archive, it is temporarily saved inside "/tmp" under a random name with "okular_XXXXXX.extension" form and is finally opened by the viewer.

The issue comes from the "unpackDocumentArchive(...)" of "core/document.cpp":

1: To create the temporary name of the document, this function search extract the extension of the "DocumentFileName" using the following line:

const int dotPos = documentFileName.indexOf( QLatin1Char('.') );
if ( dotPos != -1 )
    archiveData->document.setFileTemplate(QDir::tempPath() + QLatin1String("/okular_XXXXXX") + documentFileName.mid(dotPos));

In case of a "DocumentFileName" named "this.is.a.test", the temporary name will have the form "/tmp/okular_XXXXXX.is.a.test".
If "DocumentFileName" was "aaa.aaa/../../testXXXXXX.txt", the temporary name would become the path "/tmp/okular_XXXXXX.aaa/../../testXXXXXX.txt".



2: Before generating this name, the function "unpackDocumentArchive(...)" check if the document exists using the following code:

const KArchiveEntry * docEntry = mainDir->entry( documentFileName );
if ( !docEntry || !docEntry->isFile() )
    return nullptr;

Archives are not supposed to contain files with relative paths but it is possible to manually craft them. As the "KArchiveDirectory" class used to parse archives do not raise any exceptions in such case, it is possible to control the path of the temporary file (with exception of the six random characters).



3: Such temporary files are supposed to be removed when closing Okular which limit the impact of this bug. However, it may be possible to forge malicious archives containing documents that would cause a crash of Okular during opening (for example a document with very large pictures to ran out of memory) in order to avoid the deletion phase. As PDF may be "polyglot" ( https://en.wikipedia.org/wiki/Polyglot_(computing) ), the document could also contain a configuration file or an executable to deploy on the victim workstation.



Correcting this issue may be done by searching the last dot and not the first one on the line :
const int dotPos = documentFileName.indexOf( QLatin1Char('.') );

To illustrate this issue, I have joined an Okular archive file which (when opened with "root" permission) create a polyglot (PDF/script) file named "payloadXXXXX.pdf" in the "/root/" folder.

I'm apologizing for the mistakes and am available if you have any question.


Best regards,

Joran HERVE