import/export-tar: support xattrs, ACLs and such via PAX format

[enkore]

PAX/POSIX format basically means one gets to put arbitrary key/value parts into file metadata in a backwards-compatible fashion.

GNU tar supports a proprietary extension for xattrs, apparently invented by Jörg Schilling: Just put them into the SCHILY.xattr.<xattr>key. (Source: GNU tar 1.29 source, file src/xheader.c, xhdr_tab)

https://www.gnu.org/software/tar/manual/html_node/Extended-File-Attributes.html

[ThomasWaldmann]

related:

• https://mgorny.pl/articles/portability-of-tar-features.html
• https://github.com/libarchive/libarchive/pull/691/files

[ThomasWaldmann]

update from #5830: PAX format support is now basically present.

[ThomasWaldmann]

Borg contributors could help with this, implementing this does not need too much knowledge of borg internals.

How this could be improved:

• implement SCHILY-style xattrs in import-tar (search for "PAX" in the borg source)
• do roundtrip testing, e.g. GNU tar c -> borg import-tar -> borg extract
• implement SCHILY-style xattrs in export-tar
• do roundtrip testing, e.g. GNU tar c -> borg import-tar -> borg export-tar -> GNU tar x
• automate the roundtrip testing in our testsuite (we already use tar elsewhere in the tests)
• after xattrs are done and merged into master, work on ACLs.

[ThomasWaldmann]

Note: I didn't find clear docs about how the xattr keys and values need to be encoded. There is not just the "simple ascii text" case, but theoretically there could be also binary values or keys with "problematic" chars.

Thus, solving this likely needs some digging into gnu tar or star C sources to find out how exactly this needs to get done.

[ThomasWaldmann]

Found something: https://github.com/python/cpython/blob/3.9/Lib/tarfile.py#L1176

That tarfile code expects:

• keys to be str and encodable to utf-8
• values to be str and encodable to utf-8 or using "surrogateescape"

[ThomasWaldmann]

xattrs are done, #8467.

ACLs still todo.