Reading gzip file with very long filename or comment takes long time

The gzip file can contain filename and comment which are written like null terminated sequences of bytes. GzipFile ignores filename and comment (only calculates their checksum if needed), but simply searching for the terminating null byte, while reading byte-by-byte, takes time. On my computer, with fast CPU and SSD, reading a gzip file containing 1 GiB filename or comment will take over 5 minutes. This is not a security issue per se, because to trigger it, attacker need to send a large file at first place, but this is not fine.

This issue was discovered during discussion in #149945. The original proposed solution for that issue imposed a limit on the size of filename and comment. While the limit on filename is reasonable (but it can depend on platform?), we cannot be sure that there are no uses cases for large comments.

The following PR uses reading by chunks of growing size. It reads a 1 GiB header in fractions of second.

Linked PRs

• gh-150145

Start by reviewing GzipFile's handling of the null-terminated filename and comment fields, then compare the proposed chunked-reading approach in PR #150145. Done means reading headers with very long fields no longer spends excessive time scanning byte by byte.