2017.07.14 15:48 "[Tiff] Need for still supporting truncated StripByteCount/StripOffsets tag ?", by Even Rouault
currently TIFFFetchStripThing() which is used to read the StripByteCount/StripOffsets tags and corresponding tags for tiling supports reading files where the declared number of values in the tag is lower than the expected number of strips/tiles (*). Missing entries are set to 0. Does that make sense to support those (invalid, since the spec mentions the expected number of values) files? One issue with the current behaviour is that it makes it very easy to cause huge allocation of memory (and the memory is actually pinned since it is memset to zero, so potentially causing OOM killer to run) for very small (in bytes) files.
Digging into history, I see this was added on purpose in libtiff 3.7.0 per https://github.com/vadz/libtiff/commit/d78ce54306e5cdcd916a1e87a2738cdb210062bd
2002-12-17 Frank Warmerdam <firstname.lastname@example.org>
* libtiff/tif_dirread.c: Allow wrong sized arrays in
The ticket mentionning in this commit message (42) or the one in the above quoted corresponding changelog entry are unfortunately no longer accessible, so difficult to know what was the actual use case.
I'd be in favor of removing that capability, or perhaps limiting this up to a not so big number of tiles (let's say 1 million, with a warning stating this is an invalid file. And beyond that, error out with an error message)
(*) A few weeks ago, I fixed the other case, where the number of values is greater than the expected number of striles, so as to just allocate the final number and avoid a uncessary temporary big array.
Spatialys - Geospatial professional services