VICARKeywordHandler::Ingest() broken for End Of file Labels after 2.0.2 update

[swalter75]

There were potential buffer overflows in pre-2.0 versions of gdal, which have been corrected in [32027] and later changesets. The correction was to truncate the amount of bytes read to that of the buffer. This resulted in non-functional (but secure) code.

Here we introduce another buffer variable with the correct size to read the full EOL label.

There are also other minor corrections in the patch, as well as a new label entry to let the user know that the driver is handling the #5941 bug correctly.

[Even Rouault]

pszEOLHeader is not freed. and the return value of the VSIMalloc for pszChunkEOL is not tested against NULL

[swalter75]

Changed patch according to comment #2

[Even Rouault]

I've a bit modified your patch to incorporate trunk changes, and also merged it into trunk. As this wasn't completely trivial, I'd appreciate if you could test that both trunk and 2.0 branches are OK. I've also added a "EOLabelSize > 100 * 1024 * 1024" test to bail out, to avoid potential issues with corrupted/hostile datasets.

trunk r33668, branches/2.0 r33669 "VICAR: fix regression regarding end-of-file labels (patch by Sebastian Walter, #6399)"

[swalter75]

I've tested both trunk and 2.0 branch against my full data repository of vicar files - there seem to be no problems. Does it make sense to change the dirver sources also for the gdal 1.11.x branch? Sebastian

[Even Rouault]

Replying to swalter75:

Does it make sense to change the dirver sources also for the gdal 1.11.x branch?

The VICAR driver was new to GDAL 2.0 ;-)