feat(snapshots): add support for xattr

[miekg]

This adds support for extended attributes on Unix - although I only tested it manually on Linux.

This should include all the needed bits, but lacks proper tests - I only did a manual tests, referenced in one of the commits, included here again:

% ./kopia repository create filesystem --path=/tmp/kopia
Enter password to create new repository:
Re-enter password for verification:

% pwd
/home/miek/src/github.com/miekg/tmp
% getfacl testfile1
# file: testfile1
# owner: miek
# group: miek
user::rw-
user:root:rw-
group::rw-
group:root:rw-
group:miek:rw-
mask::rw-
other::r--

% ./kopia --log-level=debug snapshot create /home/miek/src/github.com/miekg/tmp/

% ./kopia snapshot list
....
  2024-02-12 19:23:57 CET k0e335d9c23c5a20f78e358d1bd286865 2.1 MB drwxrwxr-x files:5 dirs:1 (latest-1..2,hourly-1,daily-1,weekly-1,monthly-1,annual-1)

% ./kopia restore k0e335d9c23c5a20f78e358d1bd286865 /tmp/XXXX
Restoring to local filesystem (/tmp/XXXX) with parallelism=8...
Processed 6 (2.1 MB) of 5 (2.1 MB).
Restored 5 files, 1 directories and 0 symbolic links (2.1 MB).

% cd /tmp/XXXX
% l
go.mod  go.sum  testfile1  tmp*  xattr.go
% getfacl testfile1
# file: testfile1
# owner: miek
# group: miek
user::rw-
user:root:rw-
group::rw-
group:root:rw-
group:miek:rw-
mask::rw-
other::r--

[codecov]

Codecov Report

Attention: 51 lines in your changes are missing coverage. Please review.

Comparison is base (cb455c6) 75.86% compared to head (c4de289) 77.00%.
Report is 24 commits behind head on master.

Files	Patch %	Lines
snapshot/restore/local_fs_output_unix.go	0.00%	18 Missing ⚠️
fs/localfs/local_fs_nonwindows.go	39.28%	16 Missing and 1 partial ⚠️
fs/entry.go	69.56%	7 Missing ⚠️
snapshot/restore/local_fs_output.go	50.00%	2 Missing and 1 partial ⚠️
snapshot/snapshotfs/all_sources.go	0.00%	2 Missing ⚠️
snapshot/snapshotfs/source_directories.go	0.00%	2 Missing ⚠️
snapshot/snapshotfs/source_snapshots.go	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3643      +/-   ##
==========================================
+ Coverage   75.86%   77.00%   +1.13%     
==========================================
  Files         470      470              
  Lines       37301    28537    -8764     
==========================================
- Hits        28299    21975    -6324     
+ Misses       7071     4624    -2447     
- Partials     1931     1938       +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jkowalski]

this is promising but needs a lot more work:

• first and foremost we need a design/strategy for supporting other non-core attributes which are currently unsupported, such as Windows ACLs, additional stat_t fields, etc., perhaps 'attributes` needs to be a map of maps for extensibility, with first-level entry being the type of an attribute ("xattr", "winacl", something like that). Perhaps have a strongly-typed struct?

• we obviously need to get it to compile on all environments and address all linter issues (make lint-all is a good check)

• we need comprehensive tests that pass on all supported

• we will need a way to enable/disable support for xattr for both backup and restore based on policies because fetching them is expensive and/or folks may not want to restore them. I think the default should be to not include xattrs

• we will likely need some kind of filtering/selection so that only certain types of attributes are backed up and some can be ignored

[miekg]

thanks for your comments

we obviously need to get it to compile on all environments and address all linter issues (make lint-all is a good check)

yep

we need comprehensive tests that pass on all supported

yep, and adding support for openbsd and freebsd

we will need a way to enable/disable support for xattr for both backup and restore based on policies because fetching them is expensive and/or folks may not want to restore them. I think the default should be to not include xattrs

It this with your proposed first point, or is this already expensive in it's current form, where it uploads a nil map if there are no attributes?

first and foremost we need a design/strategy for supporting other non-core attributes which are currently unsupported, such as Windows ACLs, additional stat_t fields, etc., perhaps 'attributes` needs to be a map of maps for extensibility, with first-level entry being the type of an attribute ("xattr", "winacl", something like that). Perhaps have a strongly-typed struct?

can do

we will likely need some kind of filtering/selection so that only certain types of attributes are backed up and some can be ignored

based on the first map key? I.e. like --no-axttr woudl exclude those, but leave, say, winacl? Or more detailed on specific axttr fields?